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>
59 #include "coalesced_mmio.h"
63 #define CREATE_TRACE_POINTS
64 #include <trace/events/kvm.h>
66 /* Worst case buffer size needed for holding an integer. */
67 #define ITOA_MAX_LEN 12
69 MODULE_AUTHOR("Qumranet");
70 MODULE_LICENSE("GPL");
72 /* Architectures should define their poll value according to the halt latency */
73 unsigned int halt_poll_ns = KVM_HALT_POLL_NS_DEFAULT;
74 module_param(halt_poll_ns, uint, 0644);
75 EXPORT_SYMBOL_GPL(halt_poll_ns);
77 /* Default doubles per-vcpu halt_poll_ns. */
78 unsigned int halt_poll_ns_grow = 2;
79 module_param(halt_poll_ns_grow, uint, 0644);
80 EXPORT_SYMBOL_GPL(halt_poll_ns_grow);
82 /* The start value to grow halt_poll_ns from */
83 unsigned int halt_poll_ns_grow_start = 10000; /* 10us */
84 module_param(halt_poll_ns_grow_start, uint, 0644);
85 EXPORT_SYMBOL_GPL(halt_poll_ns_grow_start);
87 /* Default resets per-vcpu halt_poll_ns . */
88 unsigned int halt_poll_ns_shrink;
89 module_param(halt_poll_ns_shrink, uint, 0644);
90 EXPORT_SYMBOL_GPL(halt_poll_ns_shrink);
95 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
98 DEFINE_MUTEX(kvm_lock);
99 static DEFINE_RAW_SPINLOCK(kvm_count_lock);
102 static cpumask_var_t cpus_hardware_enabled;
103 static int kvm_usage_count;
104 static atomic_t hardware_enable_failed;
106 static struct kmem_cache *kvm_vcpu_cache;
108 static __read_mostly struct preempt_ops kvm_preempt_ops;
109 static DEFINE_PER_CPU(struct kvm_vcpu *, kvm_running_vcpu);
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 #define KVM_EVENT_CREATE_VM 0
152 #define KVM_EVENT_DESTROY_VM 1
153 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm);
154 static unsigned long long kvm_createvm_count;
155 static unsigned long long kvm_active_vms;
157 __weak void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
158 unsigned long start, unsigned long end)
162 bool kvm_is_zone_device_pfn(kvm_pfn_t pfn)
165 * The metadata used by is_zone_device_page() to determine whether or
166 * not a page is ZONE_DEVICE is guaranteed to be valid if and only if
167 * the device has been pinned, e.g. by get_user_pages(). WARN if the
168 * page_count() is zero to help detect bad usage of this helper.
170 if (!pfn_valid(pfn) || WARN_ON_ONCE(!page_count(pfn_to_page(pfn))))
173 return is_zone_device_page(pfn_to_page(pfn));
176 bool kvm_is_reserved_pfn(kvm_pfn_t pfn)
179 * ZONE_DEVICE pages currently set PG_reserved, but from a refcounting
180 * perspective they are "normal" pages, albeit with slightly different
184 return PageReserved(pfn_to_page(pfn)) &&
186 !kvm_is_zone_device_pfn(pfn);
191 bool kvm_is_transparent_hugepage(kvm_pfn_t pfn)
193 struct page *page = pfn_to_page(pfn);
195 if (!PageTransCompoundMap(page))
198 return is_transparent_hugepage(compound_head(page));
202 * Switches to specified vcpu, until a matching vcpu_put()
204 void vcpu_load(struct kvm_vcpu *vcpu)
208 __this_cpu_write(kvm_running_vcpu, vcpu);
209 preempt_notifier_register(&vcpu->preempt_notifier);
210 kvm_arch_vcpu_load(vcpu, cpu);
213 EXPORT_SYMBOL_GPL(vcpu_load);
215 void vcpu_put(struct kvm_vcpu *vcpu)
218 kvm_arch_vcpu_put(vcpu);
219 preempt_notifier_unregister(&vcpu->preempt_notifier);
220 __this_cpu_write(kvm_running_vcpu, NULL);
223 EXPORT_SYMBOL_GPL(vcpu_put);
225 /* TODO: merge with kvm_arch_vcpu_should_kick */
226 static bool kvm_request_needs_ipi(struct kvm_vcpu *vcpu, unsigned req)
228 int mode = kvm_vcpu_exiting_guest_mode(vcpu);
231 * We need to wait for the VCPU to reenable interrupts and get out of
232 * READING_SHADOW_PAGE_TABLES mode.
234 if (req & KVM_REQUEST_WAIT)
235 return mode != OUTSIDE_GUEST_MODE;
238 * Need to kick a running VCPU, but otherwise there is nothing to do.
240 return mode == IN_GUEST_MODE;
243 static void ack_flush(void *_completed)
247 static inline bool kvm_kick_many_cpus(const struct cpumask *cpus, bool wait)
250 cpus = cpu_online_mask;
252 if (cpumask_empty(cpus))
255 smp_call_function_many(cpus, ack_flush, NULL, wait);
259 bool kvm_make_vcpus_request_mask(struct kvm *kvm, unsigned int req,
260 struct kvm_vcpu *except,
261 unsigned long *vcpu_bitmap, cpumask_var_t tmp)
264 struct kvm_vcpu *vcpu;
269 kvm_for_each_vcpu(i, vcpu, kvm) {
270 if ((vcpu_bitmap && !test_bit(i, vcpu_bitmap)) ||
274 kvm_make_request(req, vcpu);
277 if (!(req & KVM_REQUEST_NO_WAKEUP) && kvm_vcpu_wake_up(vcpu))
280 if (tmp != NULL && cpu != -1 && cpu != me &&
281 kvm_request_needs_ipi(vcpu, req))
282 __cpumask_set_cpu(cpu, tmp);
285 called = kvm_kick_many_cpus(tmp, !!(req & KVM_REQUEST_WAIT));
291 bool kvm_make_all_cpus_request_except(struct kvm *kvm, unsigned int req,
292 struct kvm_vcpu *except)
297 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
299 called = kvm_make_vcpus_request_mask(kvm, req, except, NULL, cpus);
301 free_cpumask_var(cpus);
305 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
307 return kvm_make_all_cpus_request_except(kvm, req, NULL);
310 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
311 void kvm_flush_remote_tlbs(struct kvm *kvm)
314 * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in
315 * kvm_make_all_cpus_request.
317 long dirty_count = smp_load_acquire(&kvm->tlbs_dirty);
320 * We want to publish modifications to the page tables before reading
321 * mode. Pairs with a memory barrier in arch-specific code.
322 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
323 * and smp_mb in walk_shadow_page_lockless_begin/end.
324 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
326 * There is already an smp_mb__after_atomic() before
327 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
330 if (!kvm_arch_flush_remote_tlb(kvm)
331 || kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
332 ++kvm->stat.remote_tlb_flush;
333 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
335 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
338 void kvm_reload_remote_mmus(struct kvm *kvm)
340 kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
343 static void kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
345 mutex_init(&vcpu->mutex);
350 rcuwait_init(&vcpu->wait);
351 kvm_async_pf_vcpu_init(vcpu);
354 INIT_LIST_HEAD(&vcpu->blocked_vcpu_list);
356 kvm_vcpu_set_in_spin_loop(vcpu, false);
357 kvm_vcpu_set_dy_eligible(vcpu, false);
358 vcpu->preempted = false;
360 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
363 void kvm_vcpu_destroy(struct kvm_vcpu *vcpu)
365 kvm_arch_vcpu_destroy(vcpu);
368 * No need for rcu_read_lock as VCPU_RUN is the only place that changes
369 * the vcpu->pid pointer, and at destruction time all file descriptors
372 put_pid(rcu_dereference_protected(vcpu->pid, 1));
374 free_page((unsigned long)vcpu->run);
375 kmem_cache_free(kvm_vcpu_cache, vcpu);
377 EXPORT_SYMBOL_GPL(kvm_vcpu_destroy);
379 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
380 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
382 return container_of(mn, struct kvm, mmu_notifier);
385 static void kvm_mmu_notifier_invalidate_range(struct mmu_notifier *mn,
386 struct mm_struct *mm,
387 unsigned long start, unsigned long end)
389 struct kvm *kvm = mmu_notifier_to_kvm(mn);
392 idx = srcu_read_lock(&kvm->srcu);
393 kvm_arch_mmu_notifier_invalidate_range(kvm, start, end);
394 srcu_read_unlock(&kvm->srcu, idx);
397 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
398 struct mm_struct *mm,
399 unsigned long address,
402 struct kvm *kvm = mmu_notifier_to_kvm(mn);
405 idx = srcu_read_lock(&kvm->srcu);
406 spin_lock(&kvm->mmu_lock);
407 kvm->mmu_notifier_seq++;
409 if (kvm_set_spte_hva(kvm, address, pte))
410 kvm_flush_remote_tlbs(kvm);
412 spin_unlock(&kvm->mmu_lock);
413 srcu_read_unlock(&kvm->srcu, idx);
416 static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
417 const struct mmu_notifier_range *range)
419 struct kvm *kvm = mmu_notifier_to_kvm(mn);
420 int need_tlb_flush = 0, idx;
422 idx = srcu_read_lock(&kvm->srcu);
423 spin_lock(&kvm->mmu_lock);
425 * The count increase must become visible at unlock time as no
426 * spte can be established without taking the mmu_lock and
427 * count is also read inside the mmu_lock critical section.
429 kvm->mmu_notifier_count++;
430 need_tlb_flush = kvm_unmap_hva_range(kvm, range->start, range->end);
431 need_tlb_flush |= kvm->tlbs_dirty;
432 /* we've to flush the tlb before the pages can be freed */
434 kvm_flush_remote_tlbs(kvm);
436 spin_unlock(&kvm->mmu_lock);
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 = kvm_mmu_notifier_invalidate_range,
546 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
547 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
548 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
549 .clear_young = kvm_mmu_notifier_clear_young,
550 .test_young = kvm_mmu_notifier_test_young,
551 .change_pte = kvm_mmu_notifier_change_pte,
552 .release = kvm_mmu_notifier_release,
555 static int kvm_init_mmu_notifier(struct kvm *kvm)
557 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
558 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
561 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
563 static int kvm_init_mmu_notifier(struct kvm *kvm)
568 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
570 static struct kvm_memslots *kvm_alloc_memslots(void)
573 struct kvm_memslots *slots;
575 slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL_ACCOUNT);
579 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
580 slots->id_to_index[i] = -1;
585 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
587 if (!memslot->dirty_bitmap)
590 kvfree(memslot->dirty_bitmap);
591 memslot->dirty_bitmap = NULL;
594 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
596 kvm_destroy_dirty_bitmap(slot);
598 kvm_arch_free_memslot(kvm, slot);
604 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
606 struct kvm_memory_slot *memslot;
611 kvm_for_each_memslot(memslot, slots)
612 kvm_free_memslot(kvm, memslot);
617 static void kvm_destroy_vm_debugfs(struct kvm *kvm)
621 if (!kvm->debugfs_dentry)
624 debugfs_remove_recursive(kvm->debugfs_dentry);
626 if (kvm->debugfs_stat_data) {
627 for (i = 0; i < kvm_debugfs_num_entries; i++)
628 kfree(kvm->debugfs_stat_data[i]);
629 kfree(kvm->debugfs_stat_data);
633 static int kvm_create_vm_debugfs(struct kvm *kvm, int fd)
635 char dir_name[ITOA_MAX_LEN * 2];
636 struct kvm_stat_data *stat_data;
637 struct kvm_stats_debugfs_item *p;
639 if (!debugfs_initialized())
642 snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd);
643 kvm->debugfs_dentry = debugfs_create_dir(dir_name, kvm_debugfs_dir);
645 kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries,
646 sizeof(*kvm->debugfs_stat_data),
648 if (!kvm->debugfs_stat_data)
651 for (p = debugfs_entries; p->name; p++) {
652 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT);
656 stat_data->kvm = kvm;
657 stat_data->dbgfs_item = p;
658 kvm->debugfs_stat_data[p - debugfs_entries] = stat_data;
659 debugfs_create_file(p->name, KVM_DBGFS_GET_MODE(p),
660 kvm->debugfs_dentry, stat_data,
667 * Called after the VM is otherwise initialized, but just before adding it to
670 int __weak kvm_arch_post_init_vm(struct kvm *kvm)
676 * Called just after removing the VM from the vm_list, but before doing any
679 void __weak kvm_arch_pre_destroy_vm(struct kvm *kvm)
683 static struct kvm *kvm_create_vm(unsigned long type)
685 struct kvm *kvm = kvm_arch_alloc_vm();
690 return ERR_PTR(-ENOMEM);
692 spin_lock_init(&kvm->mmu_lock);
694 kvm->mm = current->mm;
695 kvm_eventfd_init(kvm);
696 mutex_init(&kvm->lock);
697 mutex_init(&kvm->irq_lock);
698 mutex_init(&kvm->slots_lock);
699 INIT_LIST_HEAD(&kvm->devices);
701 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
703 if (init_srcu_struct(&kvm->srcu))
704 goto out_err_no_srcu;
705 if (init_srcu_struct(&kvm->irq_srcu))
706 goto out_err_no_irq_srcu;
708 refcount_set(&kvm->users_count, 1);
709 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
710 struct kvm_memslots *slots = kvm_alloc_memslots();
713 goto out_err_no_arch_destroy_vm;
714 /* Generations must be different for each address space. */
715 slots->generation = i;
716 rcu_assign_pointer(kvm->memslots[i], slots);
719 for (i = 0; i < KVM_NR_BUSES; i++) {
720 rcu_assign_pointer(kvm->buses[i],
721 kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL_ACCOUNT));
723 goto out_err_no_arch_destroy_vm;
726 kvm->max_halt_poll_ns = halt_poll_ns;
728 r = kvm_arch_init_vm(kvm, type);
730 goto out_err_no_arch_destroy_vm;
732 r = hardware_enable_all();
734 goto out_err_no_disable;
736 #ifdef CONFIG_HAVE_KVM_IRQFD
737 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
740 r = kvm_init_mmu_notifier(kvm);
742 goto out_err_no_mmu_notifier;
744 r = kvm_arch_post_init_vm(kvm);
748 mutex_lock(&kvm_lock);
749 list_add(&kvm->vm_list, &vm_list);
750 mutex_unlock(&kvm_lock);
752 preempt_notifier_inc();
757 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
758 if (kvm->mmu_notifier.ops)
759 mmu_notifier_unregister(&kvm->mmu_notifier, current->mm);
761 out_err_no_mmu_notifier:
762 hardware_disable_all();
764 kvm_arch_destroy_vm(kvm);
765 out_err_no_arch_destroy_vm:
766 WARN_ON_ONCE(!refcount_dec_and_test(&kvm->users_count));
767 for (i = 0; i < KVM_NR_BUSES; i++)
768 kfree(kvm_get_bus(kvm, i));
769 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
770 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
771 cleanup_srcu_struct(&kvm->irq_srcu);
773 cleanup_srcu_struct(&kvm->srcu);
775 kvm_arch_free_vm(kvm);
780 static void kvm_destroy_devices(struct kvm *kvm)
782 struct kvm_device *dev, *tmp;
785 * We do not need to take the kvm->lock here, because nobody else
786 * has a reference to the struct kvm at this point and therefore
787 * cannot access the devices list anyhow.
789 list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
790 list_del(&dev->vm_node);
791 dev->ops->destroy(dev);
795 static void kvm_destroy_vm(struct kvm *kvm)
798 struct mm_struct *mm = kvm->mm;
800 kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm);
801 kvm_destroy_vm_debugfs(kvm);
802 kvm_arch_sync_events(kvm);
803 mutex_lock(&kvm_lock);
804 list_del(&kvm->vm_list);
805 mutex_unlock(&kvm_lock);
806 kvm_arch_pre_destroy_vm(kvm);
808 kvm_free_irq_routing(kvm);
809 for (i = 0; i < KVM_NR_BUSES; i++) {
810 struct kvm_io_bus *bus = kvm_get_bus(kvm, i);
813 kvm_io_bus_destroy(bus);
814 kvm->buses[i] = NULL;
816 kvm_coalesced_mmio_free(kvm);
817 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
818 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
820 kvm_arch_flush_shadow_all(kvm);
822 kvm_arch_destroy_vm(kvm);
823 kvm_destroy_devices(kvm);
824 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
825 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
826 cleanup_srcu_struct(&kvm->irq_srcu);
827 cleanup_srcu_struct(&kvm->srcu);
828 kvm_arch_free_vm(kvm);
829 preempt_notifier_dec();
830 hardware_disable_all();
834 void kvm_get_kvm(struct kvm *kvm)
836 refcount_inc(&kvm->users_count);
838 EXPORT_SYMBOL_GPL(kvm_get_kvm);
840 void kvm_put_kvm(struct kvm *kvm)
842 if (refcount_dec_and_test(&kvm->users_count))
845 EXPORT_SYMBOL_GPL(kvm_put_kvm);
848 * Used to put a reference that was taken on behalf of an object associated
849 * with a user-visible file descriptor, e.g. a vcpu or device, if installation
850 * of the new file descriptor fails and the reference cannot be transferred to
851 * its final owner. In such cases, the caller is still actively using @kvm and
852 * will fail miserably if the refcount unexpectedly hits zero.
854 void kvm_put_kvm_no_destroy(struct kvm *kvm)
856 WARN_ON(refcount_dec_and_test(&kvm->users_count));
858 EXPORT_SYMBOL_GPL(kvm_put_kvm_no_destroy);
860 static int kvm_vm_release(struct inode *inode, struct file *filp)
862 struct kvm *kvm = filp->private_data;
864 kvm_irqfd_release(kvm);
871 * Allocation size is twice as large as the actual dirty bitmap size.
872 * See kvm_vm_ioctl_get_dirty_log() why this is needed.
874 static int kvm_alloc_dirty_bitmap(struct kvm_memory_slot *memslot)
876 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
878 memslot->dirty_bitmap = kvzalloc(dirty_bytes, GFP_KERNEL_ACCOUNT);
879 if (!memslot->dirty_bitmap)
886 * Delete a memslot by decrementing the number of used slots and shifting all
887 * other entries in the array forward one spot.
889 static inline void kvm_memslot_delete(struct kvm_memslots *slots,
890 struct kvm_memory_slot *memslot)
892 struct kvm_memory_slot *mslots = slots->memslots;
895 if (WARN_ON(slots->id_to_index[memslot->id] == -1))
900 if (atomic_read(&slots->lru_slot) >= slots->used_slots)
901 atomic_set(&slots->lru_slot, 0);
903 for (i = slots->id_to_index[memslot->id]; i < slots->used_slots; i++) {
904 mslots[i] = mslots[i + 1];
905 slots->id_to_index[mslots[i].id] = i;
907 mslots[i] = *memslot;
908 slots->id_to_index[memslot->id] = -1;
912 * "Insert" a new memslot by incrementing the number of used slots. Returns
913 * the new slot's initial index into the memslots array.
915 static inline int kvm_memslot_insert_back(struct kvm_memslots *slots)
917 return slots->used_slots++;
921 * Move a changed memslot backwards in the array by shifting existing slots
922 * with a higher GFN toward the front of the array. Note, the changed memslot
923 * itself is not preserved in the array, i.e. not swapped at this time, only
924 * its new index into the array is tracked. Returns the changed memslot's
925 * current index into the memslots array.
927 static inline int kvm_memslot_move_backward(struct kvm_memslots *slots,
928 struct kvm_memory_slot *memslot)
930 struct kvm_memory_slot *mslots = slots->memslots;
933 if (WARN_ON_ONCE(slots->id_to_index[memslot->id] == -1) ||
934 WARN_ON_ONCE(!slots->used_slots))
938 * Move the target memslot backward in the array by shifting existing
939 * memslots with a higher GFN (than the target memslot) towards the
940 * front of the array.
942 for (i = slots->id_to_index[memslot->id]; i < slots->used_slots - 1; i++) {
943 if (memslot->base_gfn > mslots[i + 1].base_gfn)
946 WARN_ON_ONCE(memslot->base_gfn == mslots[i + 1].base_gfn);
948 /* Shift the next memslot forward one and update its index. */
949 mslots[i] = mslots[i + 1];
950 slots->id_to_index[mslots[i].id] = i;
956 * Move a changed memslot forwards in the array by shifting existing slots with
957 * a lower GFN toward the back of the array. Note, the changed memslot itself
958 * is not preserved in the array, i.e. not swapped at this time, only its new
959 * index into the array is tracked. Returns the changed memslot's final index
960 * into the memslots array.
962 static inline int kvm_memslot_move_forward(struct kvm_memslots *slots,
963 struct kvm_memory_slot *memslot,
966 struct kvm_memory_slot *mslots = slots->memslots;
969 for (i = start; i > 0; i--) {
970 if (memslot->base_gfn < mslots[i - 1].base_gfn)
973 WARN_ON_ONCE(memslot->base_gfn == mslots[i - 1].base_gfn);
975 /* Shift the next memslot back one and update its index. */
976 mslots[i] = mslots[i - 1];
977 slots->id_to_index[mslots[i].id] = i;
983 * Re-sort memslots based on their GFN to account for an added, deleted, or
984 * moved memslot. Sorting memslots by GFN allows using a binary search during
987 * IMPORTANT: Slots are sorted from highest GFN to lowest GFN! I.e. the entry
988 * at memslots[0] has the highest GFN.
990 * The sorting algorithm takes advantage of having initially sorted memslots
991 * and knowing the position of the changed memslot. Sorting is also optimized
992 * by not swapping the updated memslot and instead only shifting other memslots
993 * and tracking the new index for the update memslot. Only once its final
994 * index is known is the updated memslot copied into its position in the array.
996 * - When deleting a memslot, the deleted memslot simply needs to be moved to
997 * the end of the array.
999 * - When creating a memslot, the algorithm "inserts" the new memslot at the
1000 * end of the array and then it forward to its correct location.
1002 * - When moving a memslot, the algorithm first moves the updated memslot
1003 * backward to handle the scenario where the memslot's GFN was changed to a
1004 * lower value. update_memslots() then falls through and runs the same flow
1005 * as creating a memslot to move the memslot forward to handle the scenario
1006 * where its GFN was changed to a higher value.
1008 * Note, slots are sorted from highest->lowest instead of lowest->highest for
1009 * historical reasons. Originally, invalid memslots where denoted by having
1010 * GFN=0, thus sorting from highest->lowest naturally sorted invalid memslots
1011 * to the end of the array. The current algorithm uses dedicated logic to
1012 * delete a memslot and thus does not rely on invalid memslots having GFN=0.
1014 * The other historical motiviation for highest->lowest was to improve the
1015 * performance of memslot lookup. KVM originally used a linear search starting
1016 * at memslots[0]. On x86, the largest memslot usually has one of the highest,
1017 * if not *the* highest, GFN, as the bulk of the guest's RAM is located in a
1018 * single memslot above the 4gb boundary. As the largest memslot is also the
1019 * most likely to be referenced, sorting it to the front of the array was
1020 * advantageous. The current binary search starts from the middle of the array
1021 * and uses an LRU pointer to improve performance for all memslots and GFNs.
1023 static void update_memslots(struct kvm_memslots *slots,
1024 struct kvm_memory_slot *memslot,
1025 enum kvm_mr_change change)
1029 if (change == KVM_MR_DELETE) {
1030 kvm_memslot_delete(slots, memslot);
1032 if (change == KVM_MR_CREATE)
1033 i = kvm_memslot_insert_back(slots);
1035 i = kvm_memslot_move_backward(slots, memslot);
1036 i = kvm_memslot_move_forward(slots, memslot, i);
1039 * Copy the memslot to its new position in memslots and update
1040 * its index accordingly.
1042 slots->memslots[i] = *memslot;
1043 slots->id_to_index[memslot->id] = i;
1047 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
1049 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
1051 #ifdef __KVM_HAVE_READONLY_MEM
1052 valid_flags |= KVM_MEM_READONLY;
1055 if (mem->flags & ~valid_flags)
1061 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
1062 int as_id, struct kvm_memslots *slots)
1064 struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id);
1065 u64 gen = old_memslots->generation;
1067 WARN_ON(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS);
1068 slots->generation = gen | KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
1070 rcu_assign_pointer(kvm->memslots[as_id], slots);
1071 synchronize_srcu_expedited(&kvm->srcu);
1074 * Increment the new memslot generation a second time, dropping the
1075 * update in-progress flag and incrementing the generation based on
1076 * the number of address spaces. This provides a unique and easily
1077 * identifiable generation number while the memslots are in flux.
1079 gen = slots->generation & ~KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
1082 * Generations must be unique even across address spaces. We do not need
1083 * a global counter for that, instead the generation space is evenly split
1084 * across address spaces. For example, with two address spaces, address
1085 * space 0 will use generations 0, 2, 4, ... while address space 1 will
1086 * use generations 1, 3, 5, ...
1088 gen += KVM_ADDRESS_SPACE_NUM;
1090 kvm_arch_memslots_updated(kvm, gen);
1092 slots->generation = gen;
1094 return old_memslots;
1098 * Note, at a minimum, the current number of used slots must be allocated, even
1099 * when deleting a memslot, as we need a complete duplicate of the memslots for
1100 * use when invalidating a memslot prior to deleting/moving the memslot.
1102 static struct kvm_memslots *kvm_dup_memslots(struct kvm_memslots *old,
1103 enum kvm_mr_change change)
1105 struct kvm_memslots *slots;
1106 size_t old_size, new_size;
1108 old_size = sizeof(struct kvm_memslots) +
1109 (sizeof(struct kvm_memory_slot) * old->used_slots);
1111 if (change == KVM_MR_CREATE)
1112 new_size = old_size + sizeof(struct kvm_memory_slot);
1114 new_size = old_size;
1116 slots = kvzalloc(new_size, GFP_KERNEL_ACCOUNT);
1118 memcpy(slots, old, old_size);
1123 static int kvm_set_memslot(struct kvm *kvm,
1124 const struct kvm_userspace_memory_region *mem,
1125 struct kvm_memory_slot *old,
1126 struct kvm_memory_slot *new, int as_id,
1127 enum kvm_mr_change change)
1129 struct kvm_memory_slot *slot;
1130 struct kvm_memslots *slots;
1133 slots = kvm_dup_memslots(__kvm_memslots(kvm, as_id), change);
1137 if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) {
1139 * Note, the INVALID flag needs to be in the appropriate entry
1140 * in the freshly allocated memslots, not in @old or @new.
1142 slot = id_to_memslot(slots, old->id);
1143 slot->flags |= KVM_MEMSLOT_INVALID;
1146 * We can re-use the old memslots, the only difference from the
1147 * newly installed memslots is the invalid flag, which will get
1148 * dropped by update_memslots anyway. We'll also revert to the
1149 * old memslots if preparing the new memory region fails.
1151 slots = install_new_memslots(kvm, as_id, slots);
1153 /* From this point no new shadow pages pointing to a deleted,
1154 * or moved, memslot will be created.
1156 * validation of sp->gfn happens in:
1157 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1158 * - kvm_is_visible_gfn (mmu_check_root)
1160 kvm_arch_flush_shadow_memslot(kvm, slot);
1163 r = kvm_arch_prepare_memory_region(kvm, new, mem, change);
1167 update_memslots(slots, new, change);
1168 slots = install_new_memslots(kvm, as_id, slots);
1170 kvm_arch_commit_memory_region(kvm, mem, old, new, change);
1176 if (change == KVM_MR_DELETE || change == KVM_MR_MOVE)
1177 slots = install_new_memslots(kvm, as_id, slots);
1182 static int kvm_delete_memslot(struct kvm *kvm,
1183 const struct kvm_userspace_memory_region *mem,
1184 struct kvm_memory_slot *old, int as_id)
1186 struct kvm_memory_slot new;
1192 memset(&new, 0, sizeof(new));
1195 r = kvm_set_memslot(kvm, mem, old, &new, as_id, KVM_MR_DELETE);
1199 kvm_free_memslot(kvm, old);
1204 * Allocate some memory and give it an address in the guest physical address
1207 * Discontiguous memory is allowed, mostly for framebuffers.
1209 * Must be called holding kvm->slots_lock for write.
1211 int __kvm_set_memory_region(struct kvm *kvm,
1212 const struct kvm_userspace_memory_region *mem)
1214 struct kvm_memory_slot old, new;
1215 struct kvm_memory_slot *tmp;
1216 enum kvm_mr_change change;
1220 r = check_memory_region_flags(mem);
1224 as_id = mem->slot >> 16;
1225 id = (u16)mem->slot;
1227 /* General sanity checks */
1228 if (mem->memory_size & (PAGE_SIZE - 1))
1230 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
1232 /* We can read the guest memory with __xxx_user() later on. */
1233 if ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
1234 !access_ok((void __user *)(unsigned long)mem->userspace_addr,
1237 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
1239 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
1243 * Make a full copy of the old memslot, the pointer will become stale
1244 * when the memslots are re-sorted by update_memslots(), and the old
1245 * memslot needs to be referenced after calling update_memslots(), e.g.
1246 * to free its resources and for arch specific behavior.
1248 tmp = id_to_memslot(__kvm_memslots(kvm, as_id), id);
1253 memset(&old, 0, sizeof(old));
1257 if (!mem->memory_size)
1258 return kvm_delete_memslot(kvm, mem, &old, as_id);
1261 new.base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
1262 new.npages = mem->memory_size >> PAGE_SHIFT;
1263 new.flags = mem->flags;
1264 new.userspace_addr = mem->userspace_addr;
1266 if (new.npages > KVM_MEM_MAX_NR_PAGES)
1270 change = KVM_MR_CREATE;
1271 new.dirty_bitmap = NULL;
1272 memset(&new.arch, 0, sizeof(new.arch));
1273 } else { /* Modify an existing slot. */
1274 if ((new.userspace_addr != old.userspace_addr) ||
1275 (new.npages != old.npages) ||
1276 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
1279 if (new.base_gfn != old.base_gfn)
1280 change = KVM_MR_MOVE;
1281 else if (new.flags != old.flags)
1282 change = KVM_MR_FLAGS_ONLY;
1283 else /* Nothing to change. */
1286 /* Copy dirty_bitmap and arch from the current memslot. */
1287 new.dirty_bitmap = old.dirty_bitmap;
1288 memcpy(&new.arch, &old.arch, sizeof(new.arch));
1291 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
1292 /* Check for overlaps */
1293 kvm_for_each_memslot(tmp, __kvm_memslots(kvm, as_id)) {
1296 if (!((new.base_gfn + new.npages <= tmp->base_gfn) ||
1297 (new.base_gfn >= tmp->base_gfn + tmp->npages)))
1302 /* Allocate/free page dirty bitmap as needed */
1303 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
1304 new.dirty_bitmap = NULL;
1305 else if (!new.dirty_bitmap) {
1306 r = kvm_alloc_dirty_bitmap(&new);
1310 if (kvm_dirty_log_manual_protect_and_init_set(kvm))
1311 bitmap_set(new.dirty_bitmap, 0, new.npages);
1314 r = kvm_set_memslot(kvm, mem, &old, &new, as_id, change);
1318 if (old.dirty_bitmap && !new.dirty_bitmap)
1319 kvm_destroy_dirty_bitmap(&old);
1323 if (new.dirty_bitmap && !old.dirty_bitmap)
1324 kvm_destroy_dirty_bitmap(&new);
1327 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1329 int kvm_set_memory_region(struct kvm *kvm,
1330 const struct kvm_userspace_memory_region *mem)
1334 mutex_lock(&kvm->slots_lock);
1335 r = __kvm_set_memory_region(kvm, mem);
1336 mutex_unlock(&kvm->slots_lock);
1339 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1341 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1342 struct kvm_userspace_memory_region *mem)
1344 if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1347 return kvm_set_memory_region(kvm, mem);
1350 #ifndef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1352 * kvm_get_dirty_log - get a snapshot of dirty pages
1353 * @kvm: pointer to kvm instance
1354 * @log: slot id and address to which we copy the log
1355 * @is_dirty: set to '1' if any dirty pages were found
1356 * @memslot: set to the associated memslot, always valid on success
1358 int kvm_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log,
1359 int *is_dirty, struct kvm_memory_slot **memslot)
1361 struct kvm_memslots *slots;
1364 unsigned long any = 0;
1369 as_id = log->slot >> 16;
1370 id = (u16)log->slot;
1371 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1374 slots = __kvm_memslots(kvm, as_id);
1375 *memslot = id_to_memslot(slots, id);
1376 if (!(*memslot) || !(*memslot)->dirty_bitmap)
1379 kvm_arch_sync_dirty_log(kvm, *memslot);
1381 n = kvm_dirty_bitmap_bytes(*memslot);
1383 for (i = 0; !any && i < n/sizeof(long); ++i)
1384 any = (*memslot)->dirty_bitmap[i];
1386 if (copy_to_user(log->dirty_bitmap, (*memslot)->dirty_bitmap, n))
1393 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1395 #else /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
1397 * kvm_get_dirty_log_protect - get a snapshot of dirty pages
1398 * and reenable dirty page tracking for the corresponding pages.
1399 * @kvm: pointer to kvm instance
1400 * @log: slot id and address to which we copy the log
1402 * We need to keep it in mind that VCPU threads can write to the bitmap
1403 * concurrently. So, to avoid losing track of dirty pages we keep the
1406 * 1. Take a snapshot of the bit and clear it if needed.
1407 * 2. Write protect the corresponding page.
1408 * 3. Copy the snapshot to the userspace.
1409 * 4. Upon return caller flushes TLB's if needed.
1411 * Between 2 and 4, the guest may write to the page using the remaining TLB
1412 * entry. This is not a problem because the page is reported dirty using
1413 * the snapshot taken before and step 4 ensures that writes done after
1414 * exiting to userspace will be logged for the next call.
1417 static int kvm_get_dirty_log_protect(struct kvm *kvm, struct kvm_dirty_log *log)
1419 struct kvm_memslots *slots;
1420 struct kvm_memory_slot *memslot;
1423 unsigned long *dirty_bitmap;
1424 unsigned long *dirty_bitmap_buffer;
1427 as_id = log->slot >> 16;
1428 id = (u16)log->slot;
1429 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1432 slots = __kvm_memslots(kvm, as_id);
1433 memslot = id_to_memslot(slots, id);
1434 if (!memslot || !memslot->dirty_bitmap)
1437 dirty_bitmap = memslot->dirty_bitmap;
1439 kvm_arch_sync_dirty_log(kvm, memslot);
1441 n = kvm_dirty_bitmap_bytes(memslot);
1443 if (kvm->manual_dirty_log_protect) {
1445 * Unlike kvm_get_dirty_log, we always return false in *flush,
1446 * because no flush is needed until KVM_CLEAR_DIRTY_LOG. There
1447 * is some code duplication between this function and
1448 * kvm_get_dirty_log, but hopefully all architecture
1449 * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log
1450 * can be eliminated.
1452 dirty_bitmap_buffer = dirty_bitmap;
1454 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
1455 memset(dirty_bitmap_buffer, 0, n);
1457 spin_lock(&kvm->mmu_lock);
1458 for (i = 0; i < n / sizeof(long); i++) {
1462 if (!dirty_bitmap[i])
1466 mask = xchg(&dirty_bitmap[i], 0);
1467 dirty_bitmap_buffer[i] = mask;
1469 offset = i * BITS_PER_LONG;
1470 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1473 spin_unlock(&kvm->mmu_lock);
1477 kvm_arch_flush_remote_tlbs_memslot(kvm, memslot);
1479 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1486 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
1487 * @kvm: kvm instance
1488 * @log: slot id and address to which we copy the log
1490 * Steps 1-4 below provide general overview of dirty page logging. See
1491 * kvm_get_dirty_log_protect() function description for additional details.
1493 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
1494 * always flush the TLB (step 4) even if previous step failed and the dirty
1495 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
1496 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
1497 * writes will be marked dirty for next log read.
1499 * 1. Take a snapshot of the bit and clear it if needed.
1500 * 2. Write protect the corresponding page.
1501 * 3. Copy the snapshot to the userspace.
1502 * 4. Flush TLB's if needed.
1504 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1505 struct kvm_dirty_log *log)
1509 mutex_lock(&kvm->slots_lock);
1511 r = kvm_get_dirty_log_protect(kvm, log);
1513 mutex_unlock(&kvm->slots_lock);
1518 * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap
1519 * and reenable dirty page tracking for the corresponding pages.
1520 * @kvm: pointer to kvm instance
1521 * @log: slot id and address from which to fetch the bitmap of dirty pages
1523 static int kvm_clear_dirty_log_protect(struct kvm *kvm,
1524 struct kvm_clear_dirty_log *log)
1526 struct kvm_memslots *slots;
1527 struct kvm_memory_slot *memslot;
1531 unsigned long *dirty_bitmap;
1532 unsigned long *dirty_bitmap_buffer;
1535 as_id = log->slot >> 16;
1536 id = (u16)log->slot;
1537 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1540 if (log->first_page & 63)
1543 slots = __kvm_memslots(kvm, as_id);
1544 memslot = id_to_memslot(slots, id);
1545 if (!memslot || !memslot->dirty_bitmap)
1548 dirty_bitmap = memslot->dirty_bitmap;
1550 n = ALIGN(log->num_pages, BITS_PER_LONG) / 8;
1552 if (log->first_page > memslot->npages ||
1553 log->num_pages > memslot->npages - log->first_page ||
1554 (log->num_pages < memslot->npages - log->first_page && (log->num_pages & 63)))
1557 kvm_arch_sync_dirty_log(kvm, memslot);
1560 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
1561 if (copy_from_user(dirty_bitmap_buffer, log->dirty_bitmap, n))
1564 spin_lock(&kvm->mmu_lock);
1565 for (offset = log->first_page, i = offset / BITS_PER_LONG,
1566 n = DIV_ROUND_UP(log->num_pages, BITS_PER_LONG); n--;
1567 i++, offset += BITS_PER_LONG) {
1568 unsigned long mask = *dirty_bitmap_buffer++;
1569 atomic_long_t *p = (atomic_long_t *) &dirty_bitmap[i];
1573 mask &= atomic_long_fetch_andnot(mask, p);
1576 * mask contains the bits that really have been cleared. This
1577 * never includes any bits beyond the length of the memslot (if
1578 * the length is not aligned to 64 pages), therefore it is not
1579 * a problem if userspace sets them in log->dirty_bitmap.
1583 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1587 spin_unlock(&kvm->mmu_lock);
1590 kvm_arch_flush_remote_tlbs_memslot(kvm, memslot);
1595 static int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm,
1596 struct kvm_clear_dirty_log *log)
1600 mutex_lock(&kvm->slots_lock);
1602 r = kvm_clear_dirty_log_protect(kvm, log);
1604 mutex_unlock(&kvm->slots_lock);
1607 #endif /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
1609 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1611 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1613 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1615 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
1617 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn);
1619 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_memslot);
1621 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1623 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1625 return kvm_is_visible_memslot(memslot);
1627 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1629 unsigned long kvm_host_page_size(struct kvm_vcpu *vcpu, gfn_t gfn)
1631 struct vm_area_struct *vma;
1632 unsigned long addr, size;
1636 addr = kvm_vcpu_gfn_to_hva_prot(vcpu, gfn, NULL);
1637 if (kvm_is_error_hva(addr))
1640 mmap_read_lock(current->mm);
1641 vma = find_vma(current->mm, addr);
1645 size = vma_kernel_pagesize(vma);
1648 mmap_read_unlock(current->mm);
1653 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1655 return slot->flags & KVM_MEM_READONLY;
1658 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1659 gfn_t *nr_pages, bool write)
1661 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1662 return KVM_HVA_ERR_BAD;
1664 if (memslot_is_readonly(slot) && write)
1665 return KVM_HVA_ERR_RO_BAD;
1668 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1670 return __gfn_to_hva_memslot(slot, gfn);
1673 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1676 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1679 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1682 return gfn_to_hva_many(slot, gfn, NULL);
1684 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1686 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1688 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1690 EXPORT_SYMBOL_GPL(gfn_to_hva);
1692 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
1694 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
1696 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
1699 * Return the hva of a @gfn and the R/W attribute if possible.
1701 * @slot: the kvm_memory_slot which contains @gfn
1702 * @gfn: the gfn to be translated
1703 * @writable: used to return the read/write attribute of the @slot if the hva
1704 * is valid and @writable is not NULL
1706 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1707 gfn_t gfn, bool *writable)
1709 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1711 if (!kvm_is_error_hva(hva) && writable)
1712 *writable = !memslot_is_readonly(slot);
1717 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1719 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1721 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1724 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
1726 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1728 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1731 static inline int check_user_page_hwpoison(unsigned long addr)
1733 int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
1735 rc = get_user_pages(addr, 1, flags, NULL, NULL);
1736 return rc == -EHWPOISON;
1740 * The fast path to get the writable pfn which will be stored in @pfn,
1741 * true indicates success, otherwise false is returned. It's also the
1742 * only part that runs if we can in atomic context.
1744 static bool hva_to_pfn_fast(unsigned long addr, bool write_fault,
1745 bool *writable, kvm_pfn_t *pfn)
1747 struct page *page[1];
1750 * Fast pin a writable pfn only if it is a write fault request
1751 * or the caller allows to map a writable pfn for a read fault
1754 if (!(write_fault || writable))
1757 if (get_user_page_fast_only(addr, FOLL_WRITE, page)) {
1758 *pfn = page_to_pfn(page[0]);
1769 * The slow path to get the pfn of the specified host virtual address,
1770 * 1 indicates success, -errno is returned if error is detected.
1772 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1773 bool *writable, kvm_pfn_t *pfn)
1775 unsigned int flags = FOLL_HWPOISON;
1782 *writable = write_fault;
1785 flags |= FOLL_WRITE;
1787 flags |= FOLL_NOWAIT;
1789 npages = get_user_pages_unlocked(addr, 1, &page, flags);
1793 /* map read fault as writable if possible */
1794 if (unlikely(!write_fault) && writable) {
1797 if (get_user_page_fast_only(addr, FOLL_WRITE, &wpage)) {
1803 *pfn = page_to_pfn(page);
1807 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1809 if (unlikely(!(vma->vm_flags & VM_READ)))
1812 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1818 static int hva_to_pfn_remapped(struct vm_area_struct *vma,
1819 unsigned long addr, bool *async,
1820 bool write_fault, bool *writable,
1826 r = follow_pfn(vma, addr, &pfn);
1829 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
1830 * not call the fault handler, so do it here.
1832 bool unlocked = false;
1833 r = fixup_user_fault(current, current->mm, addr,
1834 (write_fault ? FAULT_FLAG_WRITE : 0),
1841 r = follow_pfn(vma, addr, &pfn);
1851 * Get a reference here because callers of *hva_to_pfn* and
1852 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
1853 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
1854 * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will
1855 * simply do nothing for reserved pfns.
1857 * Whoever called remap_pfn_range is also going to call e.g.
1858 * unmap_mapping_range before the underlying pages are freed,
1859 * causing a call to our MMU notifier.
1868 * Pin guest page in memory and return its pfn.
1869 * @addr: host virtual address which maps memory to the guest
1870 * @atomic: whether this function can sleep
1871 * @async: whether this function need to wait IO complete if the
1872 * host page is not in the memory
1873 * @write_fault: whether we should get a writable host page
1874 * @writable: whether it allows to map a writable host page for !@write_fault
1876 * The function will map a writable host page for these two cases:
1877 * 1): @write_fault = true
1878 * 2): @write_fault = false && @writable, @writable will tell the caller
1879 * whether the mapping is writable.
1881 static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1882 bool write_fault, bool *writable)
1884 struct vm_area_struct *vma;
1888 /* we can do it either atomically or asynchronously, not both */
1889 BUG_ON(atomic && async);
1891 if (hva_to_pfn_fast(addr, write_fault, writable, &pfn))
1895 return KVM_PFN_ERR_FAULT;
1897 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1901 mmap_read_lock(current->mm);
1902 if (npages == -EHWPOISON ||
1903 (!async && check_user_page_hwpoison(addr))) {
1904 pfn = KVM_PFN_ERR_HWPOISON;
1909 vma = find_vma_intersection(current->mm, addr, addr + 1);
1912 pfn = KVM_PFN_ERR_FAULT;
1913 else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
1914 r = hva_to_pfn_remapped(vma, addr, async, write_fault, writable, &pfn);
1918 pfn = KVM_PFN_ERR_FAULT;
1920 if (async && vma_is_valid(vma, write_fault))
1922 pfn = KVM_PFN_ERR_FAULT;
1925 mmap_read_unlock(current->mm);
1929 kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn,
1930 bool atomic, bool *async, bool write_fault,
1933 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1935 if (addr == KVM_HVA_ERR_RO_BAD) {
1938 return KVM_PFN_ERR_RO_FAULT;
1941 if (kvm_is_error_hva(addr)) {
1944 return KVM_PFN_NOSLOT;
1947 /* Do not map writable pfn in the readonly memslot. */
1948 if (writable && memslot_is_readonly(slot)) {
1953 return hva_to_pfn(addr, atomic, async, write_fault,
1956 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
1958 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1961 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
1962 write_fault, writable);
1964 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1966 kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1968 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1970 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
1972 kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1974 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1976 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1978 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
1980 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1982 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
1984 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1986 return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
1988 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1990 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
1992 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1994 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
1996 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1997 struct page **pages, int nr_pages)
2002 addr = gfn_to_hva_many(slot, gfn, &entry);
2003 if (kvm_is_error_hva(addr))
2006 if (entry < nr_pages)
2009 return get_user_pages_fast_only(addr, nr_pages, FOLL_WRITE, pages);
2011 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
2013 static struct page *kvm_pfn_to_page(kvm_pfn_t pfn)
2015 if (is_error_noslot_pfn(pfn))
2016 return KVM_ERR_PTR_BAD_PAGE;
2018 if (kvm_is_reserved_pfn(pfn)) {
2020 return KVM_ERR_PTR_BAD_PAGE;
2023 return pfn_to_page(pfn);
2026 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
2030 pfn = gfn_to_pfn(kvm, gfn);
2032 return kvm_pfn_to_page(pfn);
2034 EXPORT_SYMBOL_GPL(gfn_to_page);
2036 void kvm_release_pfn(kvm_pfn_t pfn, bool dirty, struct gfn_to_pfn_cache *cache)
2042 cache->pfn = cache->gfn = 0;
2045 kvm_release_pfn_dirty(pfn);
2047 kvm_release_pfn_clean(pfn);
2050 static void kvm_cache_gfn_to_pfn(struct kvm_memory_slot *slot, gfn_t gfn,
2051 struct gfn_to_pfn_cache *cache, u64 gen)
2053 kvm_release_pfn(cache->pfn, cache->dirty, cache);
2055 cache->pfn = gfn_to_pfn_memslot(slot, gfn);
2057 cache->dirty = false;
2058 cache->generation = gen;
2061 static int __kvm_map_gfn(struct kvm_memslots *slots, gfn_t gfn,
2062 struct kvm_host_map *map,
2063 struct gfn_to_pfn_cache *cache,
2068 struct page *page = KVM_UNMAPPED_PAGE;
2069 struct kvm_memory_slot *slot = __gfn_to_memslot(slots, gfn);
2070 u64 gen = slots->generation;
2076 if (!cache->pfn || cache->gfn != gfn ||
2077 cache->generation != gen) {
2080 kvm_cache_gfn_to_pfn(slot, gfn, cache, gen);
2086 pfn = gfn_to_pfn_memslot(slot, gfn);
2088 if (is_error_noslot_pfn(pfn))
2091 if (pfn_valid(pfn)) {
2092 page = pfn_to_page(pfn);
2094 hva = kmap_atomic(page);
2097 #ifdef CONFIG_HAS_IOMEM
2098 } else if (!atomic) {
2099 hva = memremap(pfn_to_hpa(pfn), PAGE_SIZE, MEMREMAP_WB);
2116 int kvm_map_gfn(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map,
2117 struct gfn_to_pfn_cache *cache, bool atomic)
2119 return __kvm_map_gfn(kvm_memslots(vcpu->kvm), gfn, map,
2122 EXPORT_SYMBOL_GPL(kvm_map_gfn);
2124 int kvm_vcpu_map(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map)
2126 return __kvm_map_gfn(kvm_vcpu_memslots(vcpu), gfn, map,
2129 EXPORT_SYMBOL_GPL(kvm_vcpu_map);
2131 static void __kvm_unmap_gfn(struct kvm_memory_slot *memslot,
2132 struct kvm_host_map *map,
2133 struct gfn_to_pfn_cache *cache,
2134 bool dirty, bool atomic)
2142 if (map->page != KVM_UNMAPPED_PAGE) {
2144 kunmap_atomic(map->hva);
2148 #ifdef CONFIG_HAS_IOMEM
2152 WARN_ONCE(1, "Unexpected unmapping in atomic context");
2156 mark_page_dirty_in_slot(memslot, map->gfn);
2159 cache->dirty |= dirty;
2161 kvm_release_pfn(map->pfn, dirty, NULL);
2167 int kvm_unmap_gfn(struct kvm_vcpu *vcpu, struct kvm_host_map *map,
2168 struct gfn_to_pfn_cache *cache, bool dirty, bool atomic)
2170 __kvm_unmap_gfn(gfn_to_memslot(vcpu->kvm, map->gfn), map,
2171 cache, dirty, atomic);
2174 EXPORT_SYMBOL_GPL(kvm_unmap_gfn);
2176 void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map, bool dirty)
2178 __kvm_unmap_gfn(kvm_vcpu_gfn_to_memslot(vcpu, map->gfn), map, NULL,
2181 EXPORT_SYMBOL_GPL(kvm_vcpu_unmap);
2183 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
2187 pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
2189 return kvm_pfn_to_page(pfn);
2191 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
2193 void kvm_release_page_clean(struct page *page)
2195 WARN_ON(is_error_page(page));
2197 kvm_release_pfn_clean(page_to_pfn(page));
2199 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
2201 void kvm_release_pfn_clean(kvm_pfn_t pfn)
2203 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
2204 put_page(pfn_to_page(pfn));
2206 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
2208 void kvm_release_page_dirty(struct page *page)
2210 WARN_ON(is_error_page(page));
2212 kvm_release_pfn_dirty(page_to_pfn(page));
2214 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
2216 void kvm_release_pfn_dirty(kvm_pfn_t pfn)
2218 kvm_set_pfn_dirty(pfn);
2219 kvm_release_pfn_clean(pfn);
2221 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
2223 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
2225 if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn))
2226 SetPageDirty(pfn_to_page(pfn));
2228 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
2230 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
2232 if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn))
2233 mark_page_accessed(pfn_to_page(pfn));
2235 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
2237 void kvm_get_pfn(kvm_pfn_t pfn)
2239 if (!kvm_is_reserved_pfn(pfn))
2240 get_page(pfn_to_page(pfn));
2242 EXPORT_SYMBOL_GPL(kvm_get_pfn);
2244 static int next_segment(unsigned long len, int offset)
2246 if (len > PAGE_SIZE - offset)
2247 return PAGE_SIZE - offset;
2252 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
2253 void *data, int offset, int len)
2258 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
2259 if (kvm_is_error_hva(addr))
2261 r = __copy_from_user(data, (void __user *)addr + offset, len);
2267 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
2270 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2272 return __kvm_read_guest_page(slot, gfn, data, offset, len);
2274 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
2276 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
2277 int offset, int len)
2279 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2281 return __kvm_read_guest_page(slot, gfn, data, offset, len);
2283 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
2285 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
2287 gfn_t gfn = gpa >> PAGE_SHIFT;
2289 int offset = offset_in_page(gpa);
2292 while ((seg = next_segment(len, offset)) != 0) {
2293 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
2303 EXPORT_SYMBOL_GPL(kvm_read_guest);
2305 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
2307 gfn_t gfn = gpa >> PAGE_SHIFT;
2309 int offset = offset_in_page(gpa);
2312 while ((seg = next_segment(len, offset)) != 0) {
2313 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
2323 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
2325 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
2326 void *data, int offset, unsigned long len)
2331 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
2332 if (kvm_is_error_hva(addr))
2334 pagefault_disable();
2335 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
2342 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
2343 void *data, unsigned long len)
2345 gfn_t gfn = gpa >> PAGE_SHIFT;
2346 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2347 int offset = offset_in_page(gpa);
2349 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
2351 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
2353 static int __kvm_write_guest_page(struct kvm_memory_slot *memslot, gfn_t gfn,
2354 const void *data, int offset, int len)
2359 addr = gfn_to_hva_memslot(memslot, gfn);
2360 if (kvm_is_error_hva(addr))
2362 r = __copy_to_user((void __user *)addr + offset, data, len);
2365 mark_page_dirty_in_slot(memslot, gfn);
2369 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
2370 const void *data, int offset, int len)
2372 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2374 return __kvm_write_guest_page(slot, gfn, data, offset, len);
2376 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
2378 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
2379 const void *data, int offset, int len)
2381 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2383 return __kvm_write_guest_page(slot, gfn, data, offset, len);
2385 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
2387 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
2390 gfn_t gfn = gpa >> PAGE_SHIFT;
2392 int offset = offset_in_page(gpa);
2395 while ((seg = next_segment(len, offset)) != 0) {
2396 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
2406 EXPORT_SYMBOL_GPL(kvm_write_guest);
2408 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
2411 gfn_t gfn = gpa >> PAGE_SHIFT;
2413 int offset = offset_in_page(gpa);
2416 while ((seg = next_segment(len, offset)) != 0) {
2417 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
2427 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
2429 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
2430 struct gfn_to_hva_cache *ghc,
2431 gpa_t gpa, unsigned long len)
2433 int offset = offset_in_page(gpa);
2434 gfn_t start_gfn = gpa >> PAGE_SHIFT;
2435 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
2436 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
2437 gfn_t nr_pages_avail;
2439 /* Update ghc->generation before performing any error checks. */
2440 ghc->generation = slots->generation;
2442 if (start_gfn > end_gfn) {
2443 ghc->hva = KVM_HVA_ERR_BAD;
2448 * If the requested region crosses two memslots, we still
2449 * verify that the entire region is valid here.
2451 for ( ; start_gfn <= end_gfn; start_gfn += nr_pages_avail) {
2452 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
2453 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
2455 if (kvm_is_error_hva(ghc->hva))
2459 /* Use the slow path for cross page reads and writes. */
2460 if (nr_pages_needed == 1)
2463 ghc->memslot = NULL;
2470 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2471 gpa_t gpa, unsigned long len)
2473 struct kvm_memslots *slots = kvm_memslots(kvm);
2474 return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len);
2476 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
2478 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2479 void *data, unsigned int offset,
2482 struct kvm_memslots *slots = kvm_memslots(kvm);
2484 gpa_t gpa = ghc->gpa + offset;
2486 BUG_ON(len + offset > ghc->len);
2488 if (slots->generation != ghc->generation) {
2489 if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len))
2493 if (kvm_is_error_hva(ghc->hva))
2496 if (unlikely(!ghc->memslot))
2497 return kvm_write_guest(kvm, gpa, data, len);
2499 r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
2502 mark_page_dirty_in_slot(ghc->memslot, gpa >> PAGE_SHIFT);
2506 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached);
2508 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2509 void *data, unsigned long len)
2511 return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len);
2513 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
2515 int kvm_read_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2516 void *data, unsigned int offset,
2519 struct kvm_memslots *slots = kvm_memslots(kvm);
2521 gpa_t gpa = ghc->gpa + offset;
2523 BUG_ON(len + offset > ghc->len);
2525 if (slots->generation != ghc->generation) {
2526 if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len))
2530 if (kvm_is_error_hva(ghc->hva))
2533 if (unlikely(!ghc->memslot))
2534 return kvm_read_guest(kvm, gpa, data, len);
2536 r = __copy_from_user(data, (void __user *)ghc->hva + offset, len);
2542 EXPORT_SYMBOL_GPL(kvm_read_guest_offset_cached);
2544 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2545 void *data, unsigned long len)
2547 return kvm_read_guest_offset_cached(kvm, ghc, data, 0, len);
2549 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
2551 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
2553 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
2555 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
2557 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
2559 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
2561 gfn_t gfn = gpa >> PAGE_SHIFT;
2563 int offset = offset_in_page(gpa);
2566 while ((seg = next_segment(len, offset)) != 0) {
2567 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
2576 EXPORT_SYMBOL_GPL(kvm_clear_guest);
2578 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot,
2581 if (memslot && memslot->dirty_bitmap) {
2582 unsigned long rel_gfn = gfn - memslot->base_gfn;
2584 set_bit_le(rel_gfn, memslot->dirty_bitmap);
2588 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
2590 struct kvm_memory_slot *memslot;
2592 memslot = gfn_to_memslot(kvm, gfn);
2593 mark_page_dirty_in_slot(memslot, gfn);
2595 EXPORT_SYMBOL_GPL(mark_page_dirty);
2597 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
2599 struct kvm_memory_slot *memslot;
2601 memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2602 mark_page_dirty_in_slot(memslot, gfn);
2604 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
2606 void kvm_sigset_activate(struct kvm_vcpu *vcpu)
2608 if (!vcpu->sigset_active)
2612 * This does a lockless modification of ->real_blocked, which is fine
2613 * because, only current can change ->real_blocked and all readers of
2614 * ->real_blocked don't care as long ->real_blocked is always a subset
2617 sigprocmask(SIG_SETMASK, &vcpu->sigset, ¤t->real_blocked);
2620 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu)
2622 if (!vcpu->sigset_active)
2625 sigprocmask(SIG_SETMASK, ¤t->real_blocked, NULL);
2626 sigemptyset(¤t->real_blocked);
2629 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
2631 unsigned int old, val, grow, grow_start;
2633 old = val = vcpu->halt_poll_ns;
2634 grow_start = READ_ONCE(halt_poll_ns_grow_start);
2635 grow = READ_ONCE(halt_poll_ns_grow);
2640 if (val < grow_start)
2643 if (val > halt_poll_ns)
2646 vcpu->halt_poll_ns = val;
2648 trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
2651 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
2653 unsigned int old, val, shrink;
2655 old = val = vcpu->halt_poll_ns;
2656 shrink = READ_ONCE(halt_poll_ns_shrink);
2662 vcpu->halt_poll_ns = val;
2663 trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
2666 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
2669 int idx = srcu_read_lock(&vcpu->kvm->srcu);
2671 if (kvm_arch_vcpu_runnable(vcpu)) {
2672 kvm_make_request(KVM_REQ_UNHALT, vcpu);
2675 if (kvm_cpu_has_pending_timer(vcpu))
2677 if (signal_pending(current))
2682 srcu_read_unlock(&vcpu->kvm->srcu, idx);
2687 update_halt_poll_stats(struct kvm_vcpu *vcpu, u64 poll_ns, bool waited)
2690 vcpu->stat.halt_poll_fail_ns += poll_ns;
2692 vcpu->stat.halt_poll_success_ns += poll_ns;
2696 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2698 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
2700 ktime_t start, cur, poll_end;
2701 bool waited = false;
2704 kvm_arch_vcpu_blocking(vcpu);
2706 start = cur = poll_end = ktime_get();
2707 if (vcpu->halt_poll_ns && !kvm_arch_no_poll(vcpu)) {
2708 ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns);
2710 ++vcpu->stat.halt_attempted_poll;
2713 * This sets KVM_REQ_UNHALT if an interrupt
2716 if (kvm_vcpu_check_block(vcpu) < 0) {
2717 ++vcpu->stat.halt_successful_poll;
2718 if (!vcpu_valid_wakeup(vcpu))
2719 ++vcpu->stat.halt_poll_invalid;
2722 poll_end = cur = ktime_get();
2723 } while (single_task_running() && ktime_before(cur, stop));
2726 prepare_to_rcuwait(&vcpu->wait);
2728 set_current_state(TASK_INTERRUPTIBLE);
2730 if (kvm_vcpu_check_block(vcpu) < 0)
2736 finish_rcuwait(&vcpu->wait);
2739 kvm_arch_vcpu_unblocking(vcpu);
2740 block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
2742 update_halt_poll_stats(
2743 vcpu, ktime_to_ns(ktime_sub(poll_end, start)), waited);
2745 if (!kvm_arch_no_poll(vcpu)) {
2746 if (!vcpu_valid_wakeup(vcpu)) {
2747 shrink_halt_poll_ns(vcpu);
2748 } else if (vcpu->kvm->max_halt_poll_ns) {
2749 if (block_ns <= vcpu->halt_poll_ns)
2751 /* we had a long block, shrink polling */
2752 else if (vcpu->halt_poll_ns &&
2753 block_ns > vcpu->kvm->max_halt_poll_ns)
2754 shrink_halt_poll_ns(vcpu);
2755 /* we had a short halt and our poll time is too small */
2756 else if (vcpu->halt_poll_ns < vcpu->kvm->max_halt_poll_ns &&
2757 block_ns < vcpu->kvm->max_halt_poll_ns)
2758 grow_halt_poll_ns(vcpu);
2760 vcpu->halt_poll_ns = 0;
2764 trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu));
2765 kvm_arch_vcpu_block_finish(vcpu);
2767 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
2769 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
2771 struct rcuwait *waitp;
2773 waitp = kvm_arch_vcpu_get_wait(vcpu);
2774 if (rcuwait_wake_up(waitp)) {
2775 WRITE_ONCE(vcpu->ready, true);
2776 ++vcpu->stat.halt_wakeup;
2782 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
2786 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2788 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
2791 int cpu = vcpu->cpu;
2793 if (kvm_vcpu_wake_up(vcpu))
2797 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
2798 if (kvm_arch_vcpu_should_kick(vcpu))
2799 smp_send_reschedule(cpu);
2802 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
2803 #endif /* !CONFIG_S390 */
2805 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
2808 struct task_struct *task = NULL;
2812 pid = rcu_dereference(target->pid);
2814 task = get_pid_task(pid, PIDTYPE_PID);
2818 ret = yield_to(task, 1);
2819 put_task_struct(task);
2823 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
2826 * Helper that checks whether a VCPU is eligible for directed yield.
2827 * Most eligible candidate to yield is decided by following heuristics:
2829 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2830 * (preempted lock holder), indicated by @in_spin_loop.
2831 * Set at the beginning and cleared at the end of interception/PLE handler.
2833 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2834 * chance last time (mostly it has become eligible now since we have probably
2835 * yielded to lockholder in last iteration. This is done by toggling
2836 * @dy_eligible each time a VCPU checked for eligibility.)
2838 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2839 * to preempted lock-holder could result in wrong VCPU selection and CPU
2840 * burning. Giving priority for a potential lock-holder increases lock
2843 * Since algorithm is based on heuristics, accessing another VCPU data without
2844 * locking does not harm. It may result in trying to yield to same VCPU, fail
2845 * and continue with next VCPU and so on.
2847 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
2849 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2852 eligible = !vcpu->spin_loop.in_spin_loop ||
2853 vcpu->spin_loop.dy_eligible;
2855 if (vcpu->spin_loop.in_spin_loop)
2856 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
2865 * Unlike kvm_arch_vcpu_runnable, this function is called outside
2866 * a vcpu_load/vcpu_put pair. However, for most architectures
2867 * kvm_arch_vcpu_runnable does not require vcpu_load.
2869 bool __weak kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
2871 return kvm_arch_vcpu_runnable(vcpu);
2874 static bool vcpu_dy_runnable(struct kvm_vcpu *vcpu)
2876 if (kvm_arch_dy_runnable(vcpu))
2879 #ifdef CONFIG_KVM_ASYNC_PF
2880 if (!list_empty_careful(&vcpu->async_pf.done))
2887 void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode)
2889 struct kvm *kvm = me->kvm;
2890 struct kvm_vcpu *vcpu;
2891 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
2897 kvm_vcpu_set_in_spin_loop(me, true);
2899 * We boost the priority of a VCPU that is runnable but not
2900 * currently running, because it got preempted by something
2901 * else and called schedule in __vcpu_run. Hopefully that
2902 * VCPU is holding the lock that we need and will release it.
2903 * We approximate round-robin by starting at the last boosted VCPU.
2905 for (pass = 0; pass < 2 && !yielded && try; pass++) {
2906 kvm_for_each_vcpu(i, vcpu, kvm) {
2907 if (!pass && i <= last_boosted_vcpu) {
2908 i = last_boosted_vcpu;
2910 } else if (pass && i > last_boosted_vcpu)
2912 if (!READ_ONCE(vcpu->ready))
2916 if (rcuwait_active(&vcpu->wait) &&
2917 !vcpu_dy_runnable(vcpu))
2919 if (READ_ONCE(vcpu->preempted) && yield_to_kernel_mode &&
2920 !kvm_arch_vcpu_in_kernel(vcpu))
2922 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
2925 yielded = kvm_vcpu_yield_to(vcpu);
2927 kvm->last_boosted_vcpu = i;
2929 } else if (yielded < 0) {
2936 kvm_vcpu_set_in_spin_loop(me, false);
2938 /* Ensure vcpu is not eligible during next spinloop */
2939 kvm_vcpu_set_dy_eligible(me, false);
2941 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
2943 static vm_fault_t kvm_vcpu_fault(struct vm_fault *vmf)
2945 struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data;
2948 if (vmf->pgoff == 0)
2949 page = virt_to_page(vcpu->run);
2951 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
2952 page = virt_to_page(vcpu->arch.pio_data);
2954 #ifdef CONFIG_KVM_MMIO
2955 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
2956 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
2959 return kvm_arch_vcpu_fault(vcpu, vmf);
2965 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
2966 .fault = kvm_vcpu_fault,
2969 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2971 vma->vm_ops = &kvm_vcpu_vm_ops;
2975 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2977 struct kvm_vcpu *vcpu = filp->private_data;
2979 kvm_put_kvm(vcpu->kvm);
2983 static struct file_operations kvm_vcpu_fops = {
2984 .release = kvm_vcpu_release,
2985 .unlocked_ioctl = kvm_vcpu_ioctl,
2986 .mmap = kvm_vcpu_mmap,
2987 .llseek = noop_llseek,
2988 KVM_COMPAT(kvm_vcpu_compat_ioctl),
2992 * Allocates an inode for the vcpu.
2994 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2996 char name[8 + 1 + ITOA_MAX_LEN + 1];
2998 snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id);
2999 return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
3002 static void kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
3004 #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS
3005 struct dentry *debugfs_dentry;
3006 char dir_name[ITOA_MAX_LEN * 2];
3008 if (!debugfs_initialized())
3011 snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
3012 debugfs_dentry = debugfs_create_dir(dir_name,
3013 vcpu->kvm->debugfs_dentry);
3015 kvm_arch_create_vcpu_debugfs(vcpu, debugfs_dentry);
3020 * Creates some virtual cpus. Good luck creating more than one.
3022 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
3025 struct kvm_vcpu *vcpu;
3028 if (id >= KVM_MAX_VCPU_ID)
3031 mutex_lock(&kvm->lock);
3032 if (kvm->created_vcpus == KVM_MAX_VCPUS) {
3033 mutex_unlock(&kvm->lock);
3037 kvm->created_vcpus++;
3038 mutex_unlock(&kvm->lock);
3040 r = kvm_arch_vcpu_precreate(kvm, id);
3042 goto vcpu_decrement;
3044 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
3047 goto vcpu_decrement;
3050 BUILD_BUG_ON(sizeof(struct kvm_run) > PAGE_SIZE);
3051 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
3056 vcpu->run = page_address(page);
3058 kvm_vcpu_init(vcpu, kvm, id);
3060 r = kvm_arch_vcpu_create(vcpu);
3062 goto vcpu_free_run_page;
3064 mutex_lock(&kvm->lock);
3065 if (kvm_get_vcpu_by_id(kvm, id)) {
3067 goto unlock_vcpu_destroy;
3070 vcpu->vcpu_idx = atomic_read(&kvm->online_vcpus);
3071 BUG_ON(kvm->vcpus[vcpu->vcpu_idx]);
3073 /* Now it's all set up, let userspace reach it */
3075 r = create_vcpu_fd(vcpu);
3077 kvm_put_kvm_no_destroy(kvm);
3078 goto unlock_vcpu_destroy;
3081 kvm->vcpus[vcpu->vcpu_idx] = vcpu;
3084 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
3085 * before kvm->online_vcpu's incremented value.
3088 atomic_inc(&kvm->online_vcpus);
3090 mutex_unlock(&kvm->lock);
3091 kvm_arch_vcpu_postcreate(vcpu);
3092 kvm_create_vcpu_debugfs(vcpu);
3095 unlock_vcpu_destroy:
3096 mutex_unlock(&kvm->lock);
3097 kvm_arch_vcpu_destroy(vcpu);
3099 free_page((unsigned long)vcpu->run);
3101 kmem_cache_free(kvm_vcpu_cache, vcpu);
3103 mutex_lock(&kvm->lock);
3104 kvm->created_vcpus--;
3105 mutex_unlock(&kvm->lock);
3109 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
3112 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
3113 vcpu->sigset_active = 1;
3114 vcpu->sigset = *sigset;
3116 vcpu->sigset_active = 0;
3120 static long kvm_vcpu_ioctl(struct file *filp,
3121 unsigned int ioctl, unsigned long arg)
3123 struct kvm_vcpu *vcpu = filp->private_data;
3124 void __user *argp = (void __user *)arg;
3126 struct kvm_fpu *fpu = NULL;
3127 struct kvm_sregs *kvm_sregs = NULL;
3129 if (vcpu->kvm->mm != current->mm)
3132 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
3136 * Some architectures have vcpu ioctls that are asynchronous to vcpu
3137 * execution; mutex_lock() would break them.
3139 r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg);
3140 if (r != -ENOIOCTLCMD)
3143 if (mutex_lock_killable(&vcpu->mutex))
3151 oldpid = rcu_access_pointer(vcpu->pid);
3152 if (unlikely(oldpid != task_pid(current))) {
3153 /* The thread running this VCPU changed. */
3156 r = kvm_arch_vcpu_run_pid_change(vcpu);
3160 newpid = get_task_pid(current, PIDTYPE_PID);
3161 rcu_assign_pointer(vcpu->pid, newpid);
3166 r = kvm_arch_vcpu_ioctl_run(vcpu);
3167 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
3170 case KVM_GET_REGS: {
3171 struct kvm_regs *kvm_regs;
3174 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL_ACCOUNT);
3177 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
3181 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
3188 case KVM_SET_REGS: {
3189 struct kvm_regs *kvm_regs;
3191 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
3192 if (IS_ERR(kvm_regs)) {
3193 r = PTR_ERR(kvm_regs);
3196 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
3200 case KVM_GET_SREGS: {
3201 kvm_sregs = kzalloc(sizeof(struct kvm_sregs),
3202 GFP_KERNEL_ACCOUNT);
3206 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
3210 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
3215 case KVM_SET_SREGS: {
3216 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
3217 if (IS_ERR(kvm_sregs)) {
3218 r = PTR_ERR(kvm_sregs);
3222 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
3225 case KVM_GET_MP_STATE: {
3226 struct kvm_mp_state mp_state;
3228 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
3232 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
3237 case KVM_SET_MP_STATE: {
3238 struct kvm_mp_state mp_state;
3241 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
3243 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
3246 case KVM_TRANSLATE: {
3247 struct kvm_translation tr;
3250 if (copy_from_user(&tr, argp, sizeof(tr)))
3252 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
3256 if (copy_to_user(argp, &tr, sizeof(tr)))
3261 case KVM_SET_GUEST_DEBUG: {
3262 struct kvm_guest_debug dbg;
3265 if (copy_from_user(&dbg, argp, sizeof(dbg)))
3267 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
3270 case KVM_SET_SIGNAL_MASK: {
3271 struct kvm_signal_mask __user *sigmask_arg = argp;
3272 struct kvm_signal_mask kvm_sigmask;
3273 sigset_t sigset, *p;
3278 if (copy_from_user(&kvm_sigmask, argp,
3279 sizeof(kvm_sigmask)))
3282 if (kvm_sigmask.len != sizeof(sigset))
3285 if (copy_from_user(&sigset, sigmask_arg->sigset,
3290 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
3294 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL_ACCOUNT);
3298 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
3302 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
3308 fpu = memdup_user(argp, sizeof(*fpu));
3314 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
3318 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
3321 mutex_unlock(&vcpu->mutex);
3327 #ifdef CONFIG_KVM_COMPAT
3328 static long kvm_vcpu_compat_ioctl(struct file *filp,
3329 unsigned int ioctl, unsigned long arg)
3331 struct kvm_vcpu *vcpu = filp->private_data;
3332 void __user *argp = compat_ptr(arg);
3335 if (vcpu->kvm->mm != current->mm)
3339 case KVM_SET_SIGNAL_MASK: {
3340 struct kvm_signal_mask __user *sigmask_arg = argp;
3341 struct kvm_signal_mask kvm_sigmask;
3346 if (copy_from_user(&kvm_sigmask, argp,
3347 sizeof(kvm_sigmask)))
3350 if (kvm_sigmask.len != sizeof(compat_sigset_t))
3353 if (get_compat_sigset(&sigset, (void *)sigmask_arg->sigset))
3355 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
3357 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
3361 r = kvm_vcpu_ioctl(filp, ioctl, arg);
3369 static int kvm_device_mmap(struct file *filp, struct vm_area_struct *vma)
3371 struct kvm_device *dev = filp->private_data;
3374 return dev->ops->mmap(dev, vma);
3379 static int kvm_device_ioctl_attr(struct kvm_device *dev,
3380 int (*accessor)(struct kvm_device *dev,
3381 struct kvm_device_attr *attr),
3384 struct kvm_device_attr attr;
3389 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
3392 return accessor(dev, &attr);
3395 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
3398 struct kvm_device *dev = filp->private_data;
3400 if (dev->kvm->mm != current->mm)
3404 case KVM_SET_DEVICE_ATTR:
3405 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
3406 case KVM_GET_DEVICE_ATTR:
3407 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
3408 case KVM_HAS_DEVICE_ATTR:
3409 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
3411 if (dev->ops->ioctl)
3412 return dev->ops->ioctl(dev, ioctl, arg);
3418 static int kvm_device_release(struct inode *inode, struct file *filp)
3420 struct kvm_device *dev = filp->private_data;
3421 struct kvm *kvm = dev->kvm;
3423 if (dev->ops->release) {
3424 mutex_lock(&kvm->lock);
3425 list_del(&dev->vm_node);
3426 dev->ops->release(dev);
3427 mutex_unlock(&kvm->lock);
3434 static const struct file_operations kvm_device_fops = {
3435 .unlocked_ioctl = kvm_device_ioctl,
3436 .release = kvm_device_release,
3437 KVM_COMPAT(kvm_device_ioctl),
3438 .mmap = kvm_device_mmap,
3441 struct kvm_device *kvm_device_from_filp(struct file *filp)
3443 if (filp->f_op != &kvm_device_fops)
3446 return filp->private_data;
3449 static const struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
3450 #ifdef CONFIG_KVM_MPIC
3451 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
3452 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
3456 int kvm_register_device_ops(const struct kvm_device_ops *ops, u32 type)
3458 if (type >= ARRAY_SIZE(kvm_device_ops_table))
3461 if (kvm_device_ops_table[type] != NULL)
3464 kvm_device_ops_table[type] = ops;
3468 void kvm_unregister_device_ops(u32 type)
3470 if (kvm_device_ops_table[type] != NULL)
3471 kvm_device_ops_table[type] = NULL;
3474 static int kvm_ioctl_create_device(struct kvm *kvm,
3475 struct kvm_create_device *cd)
3477 const struct kvm_device_ops *ops = NULL;
3478 struct kvm_device *dev;
3479 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
3483 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
3486 type = array_index_nospec(cd->type, ARRAY_SIZE(kvm_device_ops_table));
3487 ops = kvm_device_ops_table[type];
3494 dev = kzalloc(sizeof(*dev), GFP_KERNEL_ACCOUNT);
3501 mutex_lock(&kvm->lock);
3502 ret = ops->create(dev, type);
3504 mutex_unlock(&kvm->lock);
3508 list_add(&dev->vm_node, &kvm->devices);
3509 mutex_unlock(&kvm->lock);
3515 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
3517 kvm_put_kvm_no_destroy(kvm);
3518 mutex_lock(&kvm->lock);
3519 list_del(&dev->vm_node);
3520 mutex_unlock(&kvm->lock);
3529 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
3532 case KVM_CAP_USER_MEMORY:
3533 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
3534 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
3535 case KVM_CAP_INTERNAL_ERROR_DATA:
3536 #ifdef CONFIG_HAVE_KVM_MSI
3537 case KVM_CAP_SIGNAL_MSI:
3539 #ifdef CONFIG_HAVE_KVM_IRQFD
3541 case KVM_CAP_IRQFD_RESAMPLE:
3543 case KVM_CAP_IOEVENTFD_ANY_LENGTH:
3544 case KVM_CAP_CHECK_EXTENSION_VM:
3545 case KVM_CAP_ENABLE_CAP_VM:
3546 case KVM_CAP_HALT_POLL:
3548 #ifdef CONFIG_KVM_MMIO
3549 case KVM_CAP_COALESCED_MMIO:
3550 return KVM_COALESCED_MMIO_PAGE_OFFSET;
3551 case KVM_CAP_COALESCED_PIO:
3554 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3555 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2:
3556 return KVM_DIRTY_LOG_MANUAL_CAPS;
3558 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3559 case KVM_CAP_IRQ_ROUTING:
3560 return KVM_MAX_IRQ_ROUTES;
3562 #if KVM_ADDRESS_SPACE_NUM > 1
3563 case KVM_CAP_MULTI_ADDRESS_SPACE:
3564 return KVM_ADDRESS_SPACE_NUM;
3566 case KVM_CAP_NR_MEMSLOTS:
3567 return KVM_USER_MEM_SLOTS;
3571 return kvm_vm_ioctl_check_extension(kvm, arg);
3574 int __attribute__((weak)) kvm_vm_ioctl_enable_cap(struct kvm *kvm,
3575 struct kvm_enable_cap *cap)
3580 static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm,
3581 struct kvm_enable_cap *cap)
3584 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3585 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2: {
3586 u64 allowed_options = KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE;
3588 if (cap->args[0] & KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE)
3589 allowed_options = KVM_DIRTY_LOG_MANUAL_CAPS;
3591 if (cap->flags || (cap->args[0] & ~allowed_options))
3593 kvm->manual_dirty_log_protect = cap->args[0];
3597 case KVM_CAP_HALT_POLL: {
3598 if (cap->flags || cap->args[0] != (unsigned int)cap->args[0])
3601 kvm->max_halt_poll_ns = cap->args[0];
3605 return kvm_vm_ioctl_enable_cap(kvm, cap);
3609 static long kvm_vm_ioctl(struct file *filp,
3610 unsigned int ioctl, unsigned long arg)
3612 struct kvm *kvm = filp->private_data;
3613 void __user *argp = (void __user *)arg;
3616 if (kvm->mm != current->mm)
3619 case KVM_CREATE_VCPU:
3620 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
3622 case KVM_ENABLE_CAP: {
3623 struct kvm_enable_cap cap;
3626 if (copy_from_user(&cap, argp, sizeof(cap)))
3628 r = kvm_vm_ioctl_enable_cap_generic(kvm, &cap);
3631 case KVM_SET_USER_MEMORY_REGION: {
3632 struct kvm_userspace_memory_region kvm_userspace_mem;
3635 if (copy_from_user(&kvm_userspace_mem, argp,
3636 sizeof(kvm_userspace_mem)))
3639 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
3642 case KVM_GET_DIRTY_LOG: {
3643 struct kvm_dirty_log log;
3646 if (copy_from_user(&log, argp, sizeof(log)))
3648 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3651 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3652 case KVM_CLEAR_DIRTY_LOG: {
3653 struct kvm_clear_dirty_log log;
3656 if (copy_from_user(&log, argp, sizeof(log)))
3658 r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
3662 #ifdef CONFIG_KVM_MMIO
3663 case KVM_REGISTER_COALESCED_MMIO: {
3664 struct kvm_coalesced_mmio_zone zone;
3667 if (copy_from_user(&zone, argp, sizeof(zone)))
3669 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
3672 case KVM_UNREGISTER_COALESCED_MMIO: {
3673 struct kvm_coalesced_mmio_zone zone;
3676 if (copy_from_user(&zone, argp, sizeof(zone)))
3678 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
3683 struct kvm_irqfd data;
3686 if (copy_from_user(&data, argp, sizeof(data)))
3688 r = kvm_irqfd(kvm, &data);
3691 case KVM_IOEVENTFD: {
3692 struct kvm_ioeventfd data;
3695 if (copy_from_user(&data, argp, sizeof(data)))
3697 r = kvm_ioeventfd(kvm, &data);
3700 #ifdef CONFIG_HAVE_KVM_MSI
3701 case KVM_SIGNAL_MSI: {
3705 if (copy_from_user(&msi, argp, sizeof(msi)))
3707 r = kvm_send_userspace_msi(kvm, &msi);
3711 #ifdef __KVM_HAVE_IRQ_LINE
3712 case KVM_IRQ_LINE_STATUS:
3713 case KVM_IRQ_LINE: {
3714 struct kvm_irq_level irq_event;
3717 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
3720 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
3721 ioctl == KVM_IRQ_LINE_STATUS);
3726 if (ioctl == KVM_IRQ_LINE_STATUS) {
3727 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
3735 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3736 case KVM_SET_GSI_ROUTING: {
3737 struct kvm_irq_routing routing;
3738 struct kvm_irq_routing __user *urouting;
3739 struct kvm_irq_routing_entry *entries = NULL;
3742 if (copy_from_user(&routing, argp, sizeof(routing)))
3745 if (!kvm_arch_can_set_irq_routing(kvm))
3747 if (routing.nr > KVM_MAX_IRQ_ROUTES)
3753 entries = vmemdup_user(urouting->entries,
3754 array_size(sizeof(*entries),
3756 if (IS_ERR(entries)) {
3757 r = PTR_ERR(entries);
3761 r = kvm_set_irq_routing(kvm, entries, routing.nr,
3766 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
3767 case KVM_CREATE_DEVICE: {
3768 struct kvm_create_device cd;
3771 if (copy_from_user(&cd, argp, sizeof(cd)))
3774 r = kvm_ioctl_create_device(kvm, &cd);
3779 if (copy_to_user(argp, &cd, sizeof(cd)))
3785 case KVM_CHECK_EXTENSION:
3786 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
3789 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
3795 #ifdef CONFIG_KVM_COMPAT
3796 struct compat_kvm_dirty_log {
3800 compat_uptr_t dirty_bitmap; /* one bit per page */
3805 static long kvm_vm_compat_ioctl(struct file *filp,
3806 unsigned int ioctl, unsigned long arg)
3808 struct kvm *kvm = filp->private_data;
3811 if (kvm->mm != current->mm)
3814 case KVM_GET_DIRTY_LOG: {
3815 struct compat_kvm_dirty_log compat_log;
3816 struct kvm_dirty_log log;
3818 if (copy_from_user(&compat_log, (void __user *)arg,
3819 sizeof(compat_log)))
3821 log.slot = compat_log.slot;
3822 log.padding1 = compat_log.padding1;
3823 log.padding2 = compat_log.padding2;
3824 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
3826 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3830 r = kvm_vm_ioctl(filp, ioctl, arg);
3836 static struct file_operations kvm_vm_fops = {
3837 .release = kvm_vm_release,
3838 .unlocked_ioctl = kvm_vm_ioctl,
3839 .llseek = noop_llseek,
3840 KVM_COMPAT(kvm_vm_compat_ioctl),
3843 static int kvm_dev_ioctl_create_vm(unsigned long type)
3849 kvm = kvm_create_vm(type);
3851 return PTR_ERR(kvm);
3852 #ifdef CONFIG_KVM_MMIO
3853 r = kvm_coalesced_mmio_init(kvm);
3857 r = get_unused_fd_flags(O_CLOEXEC);
3861 file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
3869 * Don't call kvm_put_kvm anymore at this point; file->f_op is
3870 * already set, with ->release() being kvm_vm_release(). In error
3871 * cases it will be called by the final fput(file) and will take
3872 * care of doing kvm_put_kvm(kvm).
3874 if (kvm_create_vm_debugfs(kvm, r) < 0) {
3879 kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm);
3881 fd_install(r, file);
3889 static long kvm_dev_ioctl(struct file *filp,
3890 unsigned int ioctl, unsigned long arg)
3895 case KVM_GET_API_VERSION:
3898 r = KVM_API_VERSION;
3901 r = kvm_dev_ioctl_create_vm(arg);
3903 case KVM_CHECK_EXTENSION:
3904 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
3906 case KVM_GET_VCPU_MMAP_SIZE:
3909 r = PAGE_SIZE; /* struct kvm_run */
3911 r += PAGE_SIZE; /* pio data page */
3913 #ifdef CONFIG_KVM_MMIO
3914 r += PAGE_SIZE; /* coalesced mmio ring page */
3917 case KVM_TRACE_ENABLE:
3918 case KVM_TRACE_PAUSE:
3919 case KVM_TRACE_DISABLE:
3923 return kvm_arch_dev_ioctl(filp, ioctl, arg);
3929 static struct file_operations kvm_chardev_ops = {
3930 .unlocked_ioctl = kvm_dev_ioctl,
3931 .llseek = noop_llseek,
3932 KVM_COMPAT(kvm_dev_ioctl),
3935 static struct miscdevice kvm_dev = {
3941 static void hardware_enable_nolock(void *junk)
3943 int cpu = raw_smp_processor_id();
3946 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
3949 cpumask_set_cpu(cpu, cpus_hardware_enabled);
3951 r = kvm_arch_hardware_enable();
3954 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3955 atomic_inc(&hardware_enable_failed);
3956 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
3960 static int kvm_starting_cpu(unsigned int cpu)
3962 raw_spin_lock(&kvm_count_lock);
3963 if (kvm_usage_count)
3964 hardware_enable_nolock(NULL);
3965 raw_spin_unlock(&kvm_count_lock);
3969 static void hardware_disable_nolock(void *junk)
3971 int cpu = raw_smp_processor_id();
3973 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
3975 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3976 kvm_arch_hardware_disable();
3979 static int kvm_dying_cpu(unsigned int cpu)
3981 raw_spin_lock(&kvm_count_lock);
3982 if (kvm_usage_count)
3983 hardware_disable_nolock(NULL);
3984 raw_spin_unlock(&kvm_count_lock);
3988 static void hardware_disable_all_nolock(void)
3990 BUG_ON(!kvm_usage_count);
3993 if (!kvm_usage_count)
3994 on_each_cpu(hardware_disable_nolock, NULL, 1);
3997 static void hardware_disable_all(void)
3999 raw_spin_lock(&kvm_count_lock);
4000 hardware_disable_all_nolock();
4001 raw_spin_unlock(&kvm_count_lock);
4004 static int hardware_enable_all(void)
4008 raw_spin_lock(&kvm_count_lock);
4011 if (kvm_usage_count == 1) {
4012 atomic_set(&hardware_enable_failed, 0);
4013 on_each_cpu(hardware_enable_nolock, NULL, 1);
4015 if (atomic_read(&hardware_enable_failed)) {
4016 hardware_disable_all_nolock();
4021 raw_spin_unlock(&kvm_count_lock);
4026 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
4030 * Some (well, at least mine) BIOSes hang on reboot if
4033 * And Intel TXT required VMX off for all cpu when system shutdown.
4035 pr_info("kvm: exiting hardware virtualization\n");
4036 kvm_rebooting = true;
4037 on_each_cpu(hardware_disable_nolock, NULL, 1);
4041 static struct notifier_block kvm_reboot_notifier = {
4042 .notifier_call = kvm_reboot,
4046 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
4050 for (i = 0; i < bus->dev_count; i++) {
4051 struct kvm_io_device *pos = bus->range[i].dev;
4053 kvm_iodevice_destructor(pos);
4058 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
4059 const struct kvm_io_range *r2)
4061 gpa_t addr1 = r1->addr;
4062 gpa_t addr2 = r2->addr;
4067 /* If r2->len == 0, match the exact address. If r2->len != 0,
4068 * accept any overlapping write. Any order is acceptable for
4069 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
4070 * we process all of them.
4083 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
4085 return kvm_io_bus_cmp(p1, p2);
4088 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
4089 gpa_t addr, int len)
4091 struct kvm_io_range *range, key;
4094 key = (struct kvm_io_range) {
4099 range = bsearch(&key, bus->range, bus->dev_count,
4100 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
4104 off = range - bus->range;
4106 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
4112 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
4113 struct kvm_io_range *range, const void *val)
4117 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
4121 while (idx < bus->dev_count &&
4122 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
4123 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
4132 /* kvm_io_bus_write - called under kvm->slots_lock */
4133 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
4134 int len, const void *val)
4136 struct kvm_io_bus *bus;
4137 struct kvm_io_range range;
4140 range = (struct kvm_io_range) {
4145 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
4148 r = __kvm_io_bus_write(vcpu, bus, &range, val);
4149 return r < 0 ? r : 0;
4151 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
4153 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
4154 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
4155 gpa_t addr, int len, const void *val, long cookie)
4157 struct kvm_io_bus *bus;
4158 struct kvm_io_range range;
4160 range = (struct kvm_io_range) {
4165 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
4169 /* First try the device referenced by cookie. */
4170 if ((cookie >= 0) && (cookie < bus->dev_count) &&
4171 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
4172 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
4177 * cookie contained garbage; fall back to search and return the
4178 * correct cookie value.
4180 return __kvm_io_bus_write(vcpu, bus, &range, val);
4183 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
4184 struct kvm_io_range *range, void *val)
4188 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
4192 while (idx < bus->dev_count &&
4193 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
4194 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
4203 /* kvm_io_bus_read - called under kvm->slots_lock */
4204 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
4207 struct kvm_io_bus *bus;
4208 struct kvm_io_range range;
4211 range = (struct kvm_io_range) {
4216 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
4219 r = __kvm_io_bus_read(vcpu, bus, &range, val);
4220 return r < 0 ? r : 0;
4223 /* Caller must hold slots_lock. */
4224 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
4225 int len, struct kvm_io_device *dev)
4228 struct kvm_io_bus *new_bus, *bus;
4229 struct kvm_io_range range;
4231 bus = kvm_get_bus(kvm, bus_idx);
4235 /* exclude ioeventfd which is limited by maximum fd */
4236 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
4239 new_bus = kmalloc(struct_size(bus, range, bus->dev_count + 1),
4240 GFP_KERNEL_ACCOUNT);
4244 range = (struct kvm_io_range) {
4250 for (i = 0; i < bus->dev_count; i++)
4251 if (kvm_io_bus_cmp(&bus->range[i], &range) > 0)
4254 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
4255 new_bus->dev_count++;
4256 new_bus->range[i] = range;
4257 memcpy(new_bus->range + i + 1, bus->range + i,
4258 (bus->dev_count - i) * sizeof(struct kvm_io_range));
4259 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
4260 synchronize_srcu_expedited(&kvm->srcu);
4266 /* Caller must hold slots_lock. */
4267 void kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
4268 struct kvm_io_device *dev)
4271 struct kvm_io_bus *new_bus, *bus;
4273 bus = kvm_get_bus(kvm, bus_idx);
4277 for (i = 0; i < bus->dev_count; i++)
4278 if (bus->range[i].dev == dev) {
4282 if (i == bus->dev_count)
4285 new_bus = kmalloc(struct_size(bus, range, bus->dev_count - 1),
4286 GFP_KERNEL_ACCOUNT);
4288 pr_err("kvm: failed to shrink bus, removing it completely\n");
4292 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
4293 new_bus->dev_count--;
4294 memcpy(new_bus->range + i, bus->range + i + 1,
4295 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
4298 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
4299 synchronize_srcu_expedited(&kvm->srcu);
4304 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
4307 struct kvm_io_bus *bus;
4308 int dev_idx, srcu_idx;
4309 struct kvm_io_device *iodev = NULL;
4311 srcu_idx = srcu_read_lock(&kvm->srcu);
4313 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
4317 dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
4321 iodev = bus->range[dev_idx].dev;
4324 srcu_read_unlock(&kvm->srcu, srcu_idx);
4328 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
4330 static int kvm_debugfs_open(struct inode *inode, struct file *file,
4331 int (*get)(void *, u64 *), int (*set)(void *, u64),
4334 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
4337 /* The debugfs files are a reference to the kvm struct which
4338 * is still valid when kvm_destroy_vm is called.
4339 * To avoid the race between open and the removal of the debugfs
4340 * directory we test against the users count.
4342 if (!refcount_inc_not_zero(&stat_data->kvm->users_count))
4345 if (simple_attr_open(inode, file, get,
4346 KVM_DBGFS_GET_MODE(stat_data->dbgfs_item) & 0222
4349 kvm_put_kvm(stat_data->kvm);
4356 static int kvm_debugfs_release(struct inode *inode, struct file *file)
4358 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
4361 simple_attr_release(inode, file);
4362 kvm_put_kvm(stat_data->kvm);
4367 static int kvm_get_stat_per_vm(struct kvm *kvm, size_t offset, u64 *val)
4369 *val = *(ulong *)((void *)kvm + offset);
4374 static int kvm_clear_stat_per_vm(struct kvm *kvm, size_t offset)
4376 *(ulong *)((void *)kvm + offset) = 0;
4381 static int kvm_get_stat_per_vcpu(struct kvm *kvm, size_t offset, u64 *val)
4384 struct kvm_vcpu *vcpu;
4388 kvm_for_each_vcpu(i, vcpu, kvm)
4389 *val += *(u64 *)((void *)vcpu + offset);
4394 static int kvm_clear_stat_per_vcpu(struct kvm *kvm, size_t offset)
4397 struct kvm_vcpu *vcpu;
4399 kvm_for_each_vcpu(i, vcpu, kvm)
4400 *(u64 *)((void *)vcpu + offset) = 0;
4405 static int kvm_stat_data_get(void *data, u64 *val)
4408 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
4410 switch (stat_data->dbgfs_item->kind) {
4412 r = kvm_get_stat_per_vm(stat_data->kvm,
4413 stat_data->dbgfs_item->offset, val);
4416 r = kvm_get_stat_per_vcpu(stat_data->kvm,
4417 stat_data->dbgfs_item->offset, val);
4424 static int kvm_stat_data_clear(void *data, u64 val)
4427 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
4432 switch (stat_data->dbgfs_item->kind) {
4434 r = kvm_clear_stat_per_vm(stat_data->kvm,
4435 stat_data->dbgfs_item->offset);
4438 r = kvm_clear_stat_per_vcpu(stat_data->kvm,
4439 stat_data->dbgfs_item->offset);
4446 static int kvm_stat_data_open(struct inode *inode, struct file *file)
4448 __simple_attr_check_format("%llu\n", 0ull);
4449 return kvm_debugfs_open(inode, file, kvm_stat_data_get,
4450 kvm_stat_data_clear, "%llu\n");
4453 static const struct file_operations stat_fops_per_vm = {
4454 .owner = THIS_MODULE,
4455 .open = kvm_stat_data_open,
4456 .release = kvm_debugfs_release,
4457 .read = simple_attr_read,
4458 .write = simple_attr_write,
4459 .llseek = no_llseek,
4462 static int vm_stat_get(void *_offset, u64 *val)
4464 unsigned offset = (long)_offset;
4469 mutex_lock(&kvm_lock);
4470 list_for_each_entry(kvm, &vm_list, vm_list) {
4471 kvm_get_stat_per_vm(kvm, offset, &tmp_val);
4474 mutex_unlock(&kvm_lock);
4478 static int vm_stat_clear(void *_offset, u64 val)
4480 unsigned offset = (long)_offset;
4486 mutex_lock(&kvm_lock);
4487 list_for_each_entry(kvm, &vm_list, vm_list) {
4488 kvm_clear_stat_per_vm(kvm, offset);
4490 mutex_unlock(&kvm_lock);
4495 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n");
4497 static int vcpu_stat_get(void *_offset, u64 *val)
4499 unsigned offset = (long)_offset;
4504 mutex_lock(&kvm_lock);
4505 list_for_each_entry(kvm, &vm_list, vm_list) {
4506 kvm_get_stat_per_vcpu(kvm, offset, &tmp_val);
4509 mutex_unlock(&kvm_lock);
4513 static int vcpu_stat_clear(void *_offset, u64 val)
4515 unsigned offset = (long)_offset;
4521 mutex_lock(&kvm_lock);
4522 list_for_each_entry(kvm, &vm_list, vm_list) {
4523 kvm_clear_stat_per_vcpu(kvm, offset);
4525 mutex_unlock(&kvm_lock);
4530 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear,
4533 static const struct file_operations *stat_fops[] = {
4534 [KVM_STAT_VCPU] = &vcpu_stat_fops,
4535 [KVM_STAT_VM] = &vm_stat_fops,
4538 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm)
4540 struct kobj_uevent_env *env;
4541 unsigned long long created, active;
4543 if (!kvm_dev.this_device || !kvm)
4546 mutex_lock(&kvm_lock);
4547 if (type == KVM_EVENT_CREATE_VM) {
4548 kvm_createvm_count++;
4550 } else if (type == KVM_EVENT_DESTROY_VM) {
4553 created = kvm_createvm_count;
4554 active = kvm_active_vms;
4555 mutex_unlock(&kvm_lock);
4557 env = kzalloc(sizeof(*env), GFP_KERNEL_ACCOUNT);
4561 add_uevent_var(env, "CREATED=%llu", created);
4562 add_uevent_var(env, "COUNT=%llu", active);
4564 if (type == KVM_EVENT_CREATE_VM) {
4565 add_uevent_var(env, "EVENT=create");
4566 kvm->userspace_pid = task_pid_nr(current);
4567 } else if (type == KVM_EVENT_DESTROY_VM) {
4568 add_uevent_var(env, "EVENT=destroy");
4570 add_uevent_var(env, "PID=%d", kvm->userspace_pid);
4572 if (!IS_ERR_OR_NULL(kvm->debugfs_dentry)) {
4573 char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL_ACCOUNT);
4576 tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX);
4578 add_uevent_var(env, "STATS_PATH=%s", tmp);
4582 /* no need for checks, since we are adding at most only 5 keys */
4583 env->envp[env->envp_idx++] = NULL;
4584 kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp);
4588 static void kvm_init_debug(void)
4590 struct kvm_stats_debugfs_item *p;
4592 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
4594 kvm_debugfs_num_entries = 0;
4595 for (p = debugfs_entries; p->name; ++p, kvm_debugfs_num_entries++) {
4596 debugfs_create_file(p->name, KVM_DBGFS_GET_MODE(p),
4597 kvm_debugfs_dir, (void *)(long)p->offset,
4598 stat_fops[p->kind]);
4602 static int kvm_suspend(void)
4604 if (kvm_usage_count)
4605 hardware_disable_nolock(NULL);
4609 static void kvm_resume(void)
4611 if (kvm_usage_count) {
4612 #ifdef CONFIG_LOCKDEP
4613 WARN_ON(lockdep_is_held(&kvm_count_lock));
4615 hardware_enable_nolock(NULL);
4619 static struct syscore_ops kvm_syscore_ops = {
4620 .suspend = kvm_suspend,
4621 .resume = kvm_resume,
4625 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
4627 return container_of(pn, struct kvm_vcpu, preempt_notifier);
4630 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
4632 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
4634 WRITE_ONCE(vcpu->preempted, false);
4635 WRITE_ONCE(vcpu->ready, false);
4637 __this_cpu_write(kvm_running_vcpu, vcpu);
4638 kvm_arch_sched_in(vcpu, cpu);
4639 kvm_arch_vcpu_load(vcpu, cpu);
4642 static void kvm_sched_out(struct preempt_notifier *pn,
4643 struct task_struct *next)
4645 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
4647 if (current->state == TASK_RUNNING) {
4648 WRITE_ONCE(vcpu->preempted, true);
4649 WRITE_ONCE(vcpu->ready, true);
4651 kvm_arch_vcpu_put(vcpu);
4652 __this_cpu_write(kvm_running_vcpu, NULL);
4656 * kvm_get_running_vcpu - get the vcpu running on the current CPU.
4658 * We can disable preemption locally around accessing the per-CPU variable,
4659 * and use the resolved vcpu pointer after enabling preemption again,
4660 * because even if the current thread is migrated to another CPU, reading
4661 * the per-CPU value later will give us the same value as we update the
4662 * per-CPU variable in the preempt notifier handlers.
4664 struct kvm_vcpu *kvm_get_running_vcpu(void)
4666 struct kvm_vcpu *vcpu;
4669 vcpu = __this_cpu_read(kvm_running_vcpu);
4674 EXPORT_SYMBOL_GPL(kvm_get_running_vcpu);
4677 * kvm_get_running_vcpus - get the per-CPU array of currently running vcpus.
4679 struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void)
4681 return &kvm_running_vcpu;
4684 struct kvm_cpu_compat_check {
4689 static void check_processor_compat(void *data)
4691 struct kvm_cpu_compat_check *c = data;
4693 *c->ret = kvm_arch_check_processor_compat(c->opaque);
4696 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
4697 struct module *module)
4699 struct kvm_cpu_compat_check c;
4703 r = kvm_arch_init(opaque);
4708 * kvm_arch_init makes sure there's at most one caller
4709 * for architectures that support multiple implementations,
4710 * like intel and amd on x86.
4711 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
4712 * conflicts in case kvm is already setup for another implementation.
4714 r = kvm_irqfd_init();
4718 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
4723 r = kvm_arch_hardware_setup(opaque);
4729 for_each_online_cpu(cpu) {
4730 smp_call_function_single(cpu, check_processor_compat, &c, 1);
4735 r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting",
4736 kvm_starting_cpu, kvm_dying_cpu);
4739 register_reboot_notifier(&kvm_reboot_notifier);
4741 /* A kmem cache lets us meet the alignment requirements of fx_save. */
4743 vcpu_align = __alignof__(struct kvm_vcpu);
4745 kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align,
4747 offsetof(struct kvm_vcpu, arch),
4748 sizeof_field(struct kvm_vcpu, arch),
4750 if (!kvm_vcpu_cache) {
4755 r = kvm_async_pf_init();
4759 kvm_chardev_ops.owner = module;
4760 kvm_vm_fops.owner = module;
4761 kvm_vcpu_fops.owner = module;
4763 r = misc_register(&kvm_dev);
4765 pr_err("kvm: misc device register failed\n");
4769 register_syscore_ops(&kvm_syscore_ops);
4771 kvm_preempt_ops.sched_in = kvm_sched_in;
4772 kvm_preempt_ops.sched_out = kvm_sched_out;
4776 r = kvm_vfio_ops_init();
4782 kvm_async_pf_deinit();
4784 kmem_cache_destroy(kvm_vcpu_cache);
4786 unregister_reboot_notifier(&kvm_reboot_notifier);
4787 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4789 kvm_arch_hardware_unsetup();
4791 free_cpumask_var(cpus_hardware_enabled);
4799 EXPORT_SYMBOL_GPL(kvm_init);
4803 debugfs_remove_recursive(kvm_debugfs_dir);
4804 misc_deregister(&kvm_dev);
4805 kmem_cache_destroy(kvm_vcpu_cache);
4806 kvm_async_pf_deinit();
4807 unregister_syscore_ops(&kvm_syscore_ops);
4808 unregister_reboot_notifier(&kvm_reboot_notifier);
4809 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4810 on_each_cpu(hardware_disable_nolock, NULL, 1);
4811 kvm_arch_hardware_unsetup();
4814 free_cpumask_var(cpus_hardware_enabled);
4815 kvm_vfio_ops_exit();
4817 EXPORT_SYMBOL_GPL(kvm_exit);
4819 struct kvm_vm_worker_thread_context {
4821 struct task_struct *parent;
4822 struct completion init_done;
4823 kvm_vm_thread_fn_t thread_fn;
4828 static int kvm_vm_worker_thread(void *context)
4831 * The init_context is allocated on the stack of the parent thread, so
4832 * we have to locally copy anything that is needed beyond initialization
4834 struct kvm_vm_worker_thread_context *init_context = context;
4835 struct kvm *kvm = init_context->kvm;
4836 kvm_vm_thread_fn_t thread_fn = init_context->thread_fn;
4837 uintptr_t data = init_context->data;
4840 err = kthread_park(current);
4841 /* kthread_park(current) is never supposed to return an error */
4846 err = cgroup_attach_task_all(init_context->parent, current);
4848 kvm_err("%s: cgroup_attach_task_all failed with err %d\n",
4853 set_user_nice(current, task_nice(init_context->parent));
4856 init_context->err = err;
4857 complete(&init_context->init_done);
4858 init_context = NULL;
4863 /* Wait to be woken up by the spawner before proceeding. */
4866 if (!kthread_should_stop())
4867 err = thread_fn(kvm, data);
4872 int kvm_vm_create_worker_thread(struct kvm *kvm, kvm_vm_thread_fn_t thread_fn,
4873 uintptr_t data, const char *name,
4874 struct task_struct **thread_ptr)
4876 struct kvm_vm_worker_thread_context init_context = {};
4877 struct task_struct *thread;
4880 init_context.kvm = kvm;
4881 init_context.parent = current;
4882 init_context.thread_fn = thread_fn;
4883 init_context.data = data;
4884 init_completion(&init_context.init_done);
4886 thread = kthread_run(kvm_vm_worker_thread, &init_context,
4887 "%s-%d", name, task_pid_nr(current));
4889 return PTR_ERR(thread);
4891 /* kthread_run is never supposed to return NULL */
4892 WARN_ON(thread == NULL);
4894 wait_for_completion(&init_context.init_done);
4896 if (!init_context.err)
4897 *thread_ptr = thread;
4899 return init_context.err;