1 // SPDX-License-Identifier: GPL-2.0-only
3 * Kernel-based Virtual Machine driver for Linux
5 * This module enables machines with Intel VT-x extensions to run virtual
6 * machines without emulation or binary translation.
8 * Copyright (C) 2006 Qumranet, Inc.
9 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
12 * Avi Kivity <avi@qumranet.com>
13 * Yaniv Kamay <yaniv@qumranet.com>
16 #include <kvm/iodev.h>
18 #include <linux/kvm_host.h>
19 #include <linux/kvm.h>
20 #include <linux/module.h>
21 #include <linux/errno.h>
22 #include <linux/percpu.h>
24 #include <linux/miscdevice.h>
25 #include <linux/vmalloc.h>
26 #include <linux/reboot.h>
27 #include <linux/debugfs.h>
28 #include <linux/highmem.h>
29 #include <linux/file.h>
30 #include <linux/syscore_ops.h>
31 #include <linux/cpu.h>
32 #include <linux/sched/signal.h>
33 #include <linux/sched/mm.h>
34 #include <linux/sched/stat.h>
35 #include <linux/cpumask.h>
36 #include <linux/smp.h>
37 #include <linux/anon_inodes.h>
38 #include <linux/profile.h>
39 #include <linux/kvm_para.h>
40 #include <linux/pagemap.h>
41 #include <linux/mman.h>
42 #include <linux/swap.h>
43 #include <linux/bitops.h>
44 #include <linux/spinlock.h>
45 #include <linux/compat.h>
46 #include <linux/srcu.h>
47 #include <linux/hugetlb.h>
48 #include <linux/slab.h>
49 #include <linux/sort.h>
50 #include <linux/bsearch.h>
52 #include <linux/lockdep.h>
53 #include <linux/kthread.h>
55 #include <asm/processor.h>
56 #include <asm/ioctl.h>
57 #include <linux/uaccess.h>
58 #include <asm/pgtable.h>
60 #include "coalesced_mmio.h"
64 #define CREATE_TRACE_POINTS
65 #include <trace/events/kvm.h>
67 /* Worst case buffer size needed for holding an integer. */
68 #define ITOA_MAX_LEN 12
70 MODULE_AUTHOR("Qumranet");
71 MODULE_LICENSE("GPL");
73 /* Architectures should define their poll value according to the halt latency */
74 unsigned int halt_poll_ns = KVM_HALT_POLL_NS_DEFAULT;
75 module_param(halt_poll_ns, uint, 0644);
76 EXPORT_SYMBOL_GPL(halt_poll_ns);
78 /* Default doubles per-vcpu halt_poll_ns. */
79 unsigned int halt_poll_ns_grow = 2;
80 module_param(halt_poll_ns_grow, uint, 0644);
81 EXPORT_SYMBOL_GPL(halt_poll_ns_grow);
83 /* The start value to grow halt_poll_ns from */
84 unsigned int halt_poll_ns_grow_start = 10000; /* 10us */
85 module_param(halt_poll_ns_grow_start, uint, 0644);
86 EXPORT_SYMBOL_GPL(halt_poll_ns_grow_start);
88 /* Default resets per-vcpu halt_poll_ns . */
89 unsigned int halt_poll_ns_shrink;
90 module_param(halt_poll_ns_shrink, uint, 0644);
91 EXPORT_SYMBOL_GPL(halt_poll_ns_shrink);
96 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
99 DEFINE_MUTEX(kvm_lock);
100 static DEFINE_RAW_SPINLOCK(kvm_count_lock);
103 static cpumask_var_t cpus_hardware_enabled;
104 static int kvm_usage_count;
105 static atomic_t hardware_enable_failed;
107 static struct kmem_cache *kvm_vcpu_cache;
109 static __read_mostly struct preempt_ops kvm_preempt_ops;
110 static DEFINE_PER_CPU(struct kvm_vcpu *, kvm_running_vcpu);
112 struct dentry *kvm_debugfs_dir;
113 EXPORT_SYMBOL_GPL(kvm_debugfs_dir);
115 static int kvm_debugfs_num_entries;
116 static const struct file_operations stat_fops_per_vm;
118 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
120 #ifdef CONFIG_KVM_COMPAT
121 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
123 #define KVM_COMPAT(c) .compat_ioctl = (c)
126 * For architectures that don't implement a compat infrastructure,
127 * adopt a double line of defense:
128 * - Prevent a compat task from opening /dev/kvm
129 * - If the open has been done by a 64bit task, and the KVM fd
130 * passed to a compat task, let the ioctls fail.
132 static long kvm_no_compat_ioctl(struct file *file, unsigned int ioctl,
133 unsigned long arg) { return -EINVAL; }
135 static int kvm_no_compat_open(struct inode *inode, struct file *file)
137 return is_compat_task() ? -ENODEV : 0;
139 #define KVM_COMPAT(c) .compat_ioctl = kvm_no_compat_ioctl, \
140 .open = kvm_no_compat_open
142 static int hardware_enable_all(void);
143 static void hardware_disable_all(void);
145 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
147 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot, gfn_t gfn);
149 __visible bool kvm_rebooting;
150 EXPORT_SYMBOL_GPL(kvm_rebooting);
152 static bool largepages_enabled = true;
154 #define KVM_EVENT_CREATE_VM 0
155 #define KVM_EVENT_DESTROY_VM 1
156 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm);
157 static unsigned long long kvm_createvm_count;
158 static unsigned long long kvm_active_vms;
160 __weak int kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
161 unsigned long start, unsigned long end, bool blockable)
166 bool kvm_is_zone_device_pfn(kvm_pfn_t pfn)
169 * The metadata used by is_zone_device_page() to determine whether or
170 * not a page is ZONE_DEVICE is guaranteed to be valid if and only if
171 * the device has been pinned, e.g. by get_user_pages(). WARN if the
172 * page_count() is zero to help detect bad usage of this helper.
174 if (!pfn_valid(pfn) || WARN_ON_ONCE(!page_count(pfn_to_page(pfn))))
177 return is_zone_device_page(pfn_to_page(pfn));
180 bool kvm_is_reserved_pfn(kvm_pfn_t pfn)
183 * ZONE_DEVICE pages currently set PG_reserved, but from a refcounting
184 * perspective they are "normal" pages, albeit with slightly different
188 return PageReserved(pfn_to_page(pfn)) &&
189 !kvm_is_zone_device_pfn(pfn);
194 bool kvm_is_transparent_hugepage(kvm_pfn_t pfn)
196 struct page *page = pfn_to_page(pfn);
198 if (!PageTransCompoundMap(page))
201 return is_transparent_hugepage(compound_head(page));
205 * Switches to specified vcpu, until a matching vcpu_put()
207 void vcpu_load(struct kvm_vcpu *vcpu)
211 __this_cpu_write(kvm_running_vcpu, vcpu);
212 preempt_notifier_register(&vcpu->preempt_notifier);
213 kvm_arch_vcpu_load(vcpu, cpu);
216 EXPORT_SYMBOL_GPL(vcpu_load);
218 void vcpu_put(struct kvm_vcpu *vcpu)
221 kvm_arch_vcpu_put(vcpu);
222 preempt_notifier_unregister(&vcpu->preempt_notifier);
223 __this_cpu_write(kvm_running_vcpu, NULL);
226 EXPORT_SYMBOL_GPL(vcpu_put);
228 /* TODO: merge with kvm_arch_vcpu_should_kick */
229 static bool kvm_request_needs_ipi(struct kvm_vcpu *vcpu, unsigned req)
231 int mode = kvm_vcpu_exiting_guest_mode(vcpu);
234 * We need to wait for the VCPU to reenable interrupts and get out of
235 * READING_SHADOW_PAGE_TABLES mode.
237 if (req & KVM_REQUEST_WAIT)
238 return mode != OUTSIDE_GUEST_MODE;
241 * Need to kick a running VCPU, but otherwise there is nothing to do.
243 return mode == IN_GUEST_MODE;
246 static void ack_flush(void *_completed)
250 static inline bool kvm_kick_many_cpus(const struct cpumask *cpus, bool wait)
253 cpus = cpu_online_mask;
255 if (cpumask_empty(cpus))
258 smp_call_function_many(cpus, ack_flush, NULL, wait);
262 bool kvm_make_vcpus_request_mask(struct kvm *kvm, unsigned int req,
263 unsigned long *vcpu_bitmap, cpumask_var_t tmp)
266 struct kvm_vcpu *vcpu;
271 kvm_for_each_vcpu(i, vcpu, kvm) {
272 if (vcpu_bitmap && !test_bit(i, vcpu_bitmap))
275 kvm_make_request(req, vcpu);
278 if (!(req & KVM_REQUEST_NO_WAKEUP) && kvm_vcpu_wake_up(vcpu))
281 if (tmp != NULL && cpu != -1 && cpu != me &&
282 kvm_request_needs_ipi(vcpu, req))
283 __cpumask_set_cpu(cpu, tmp);
286 called = kvm_kick_many_cpus(tmp, !!(req & KVM_REQUEST_WAIT));
292 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
297 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
299 called = kvm_make_vcpus_request_mask(kvm, req, NULL, cpus);
301 free_cpumask_var(cpus);
305 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
306 void kvm_flush_remote_tlbs(struct kvm *kvm)
309 * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in
310 * kvm_make_all_cpus_request.
312 long dirty_count = smp_load_acquire(&kvm->tlbs_dirty);
315 * We want to publish modifications to the page tables before reading
316 * mode. Pairs with a memory barrier in arch-specific code.
317 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
318 * and smp_mb in walk_shadow_page_lockless_begin/end.
319 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
321 * There is already an smp_mb__after_atomic() before
322 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
325 if (!kvm_arch_flush_remote_tlb(kvm)
326 || kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
327 ++kvm->stat.remote_tlb_flush;
328 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
330 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
333 void kvm_reload_remote_mmus(struct kvm *kvm)
335 kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
338 static void kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
340 mutex_init(&vcpu->mutex);
345 init_swait_queue_head(&vcpu->wq);
346 kvm_async_pf_vcpu_init(vcpu);
349 INIT_LIST_HEAD(&vcpu->blocked_vcpu_list);
351 kvm_vcpu_set_in_spin_loop(vcpu, false);
352 kvm_vcpu_set_dy_eligible(vcpu, false);
353 vcpu->preempted = false;
355 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
358 void kvm_vcpu_destroy(struct kvm_vcpu *vcpu)
360 kvm_arch_vcpu_destroy(vcpu);
363 * No need for rcu_read_lock as VCPU_RUN is the only place that changes
364 * the vcpu->pid pointer, and at destruction time all file descriptors
367 put_pid(rcu_dereference_protected(vcpu->pid, 1));
369 free_page((unsigned long)vcpu->run);
370 kmem_cache_free(kvm_vcpu_cache, vcpu);
372 EXPORT_SYMBOL_GPL(kvm_vcpu_destroy);
374 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
375 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
377 return container_of(mn, struct kvm, mmu_notifier);
380 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
381 struct mm_struct *mm,
382 unsigned long address,
385 struct kvm *kvm = mmu_notifier_to_kvm(mn);
388 idx = srcu_read_lock(&kvm->srcu);
389 spin_lock(&kvm->mmu_lock);
390 kvm->mmu_notifier_seq++;
392 if (kvm_set_spte_hva(kvm, address, pte))
393 kvm_flush_remote_tlbs(kvm);
395 spin_unlock(&kvm->mmu_lock);
396 srcu_read_unlock(&kvm->srcu, idx);
399 static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
400 const struct mmu_notifier_range *range)
402 struct kvm *kvm = mmu_notifier_to_kvm(mn);
403 int need_tlb_flush = 0, idx;
406 idx = srcu_read_lock(&kvm->srcu);
407 spin_lock(&kvm->mmu_lock);
409 * The count increase must become visible at unlock time as no
410 * spte can be established without taking the mmu_lock and
411 * count is also read inside the mmu_lock critical section.
413 kvm->mmu_notifier_count++;
414 need_tlb_flush = kvm_unmap_hva_range(kvm, range->start, range->end);
415 need_tlb_flush |= kvm->tlbs_dirty;
416 /* we've to flush the tlb before the pages can be freed */
418 kvm_flush_remote_tlbs(kvm);
420 spin_unlock(&kvm->mmu_lock);
422 ret = kvm_arch_mmu_notifier_invalidate_range(kvm, range->start,
424 mmu_notifier_range_blockable(range));
426 srcu_read_unlock(&kvm->srcu, idx);
431 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
432 const struct mmu_notifier_range *range)
434 struct kvm *kvm = mmu_notifier_to_kvm(mn);
436 spin_lock(&kvm->mmu_lock);
438 * This sequence increase will notify the kvm page fault that
439 * the page that is going to be mapped in the spte could have
442 kvm->mmu_notifier_seq++;
445 * The above sequence increase must be visible before the
446 * below count decrease, which is ensured by the smp_wmb above
447 * in conjunction with the smp_rmb in mmu_notifier_retry().
449 kvm->mmu_notifier_count--;
450 spin_unlock(&kvm->mmu_lock);
452 BUG_ON(kvm->mmu_notifier_count < 0);
455 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
456 struct mm_struct *mm,
460 struct kvm *kvm = mmu_notifier_to_kvm(mn);
463 idx = srcu_read_lock(&kvm->srcu);
464 spin_lock(&kvm->mmu_lock);
466 young = kvm_age_hva(kvm, start, end);
468 kvm_flush_remote_tlbs(kvm);
470 spin_unlock(&kvm->mmu_lock);
471 srcu_read_unlock(&kvm->srcu, idx);
476 static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
477 struct mm_struct *mm,
481 struct kvm *kvm = mmu_notifier_to_kvm(mn);
484 idx = srcu_read_lock(&kvm->srcu);
485 spin_lock(&kvm->mmu_lock);
487 * Even though we do not flush TLB, this will still adversely
488 * affect performance on pre-Haswell Intel EPT, where there is
489 * no EPT Access Bit to clear so that we have to tear down EPT
490 * tables instead. If we find this unacceptable, we can always
491 * add a parameter to kvm_age_hva so that it effectively doesn't
492 * do anything on clear_young.
494 * Also note that currently we never issue secondary TLB flushes
495 * from clear_young, leaving this job up to the regular system
496 * cadence. If we find this inaccurate, we might come up with a
497 * more sophisticated heuristic later.
499 young = kvm_age_hva(kvm, start, end);
500 spin_unlock(&kvm->mmu_lock);
501 srcu_read_unlock(&kvm->srcu, idx);
506 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
507 struct mm_struct *mm,
508 unsigned long address)
510 struct kvm *kvm = mmu_notifier_to_kvm(mn);
513 idx = srcu_read_lock(&kvm->srcu);
514 spin_lock(&kvm->mmu_lock);
515 young = kvm_test_age_hva(kvm, address);
516 spin_unlock(&kvm->mmu_lock);
517 srcu_read_unlock(&kvm->srcu, idx);
522 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
523 struct mm_struct *mm)
525 struct kvm *kvm = mmu_notifier_to_kvm(mn);
528 idx = srcu_read_lock(&kvm->srcu);
529 kvm_arch_flush_shadow_all(kvm);
530 srcu_read_unlock(&kvm->srcu, idx);
533 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
534 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
535 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
536 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
537 .clear_young = kvm_mmu_notifier_clear_young,
538 .test_young = kvm_mmu_notifier_test_young,
539 .change_pte = kvm_mmu_notifier_change_pte,
540 .release = kvm_mmu_notifier_release,
543 static int kvm_init_mmu_notifier(struct kvm *kvm)
545 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
546 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
549 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
551 static int kvm_init_mmu_notifier(struct kvm *kvm)
556 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
558 static struct kvm_memslots *kvm_alloc_memslots(void)
561 struct kvm_memslots *slots;
563 slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL_ACCOUNT);
567 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
568 slots->id_to_index[i] = slots->memslots[i].id = i;
573 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
575 if (!memslot->dirty_bitmap)
578 kvfree(memslot->dirty_bitmap);
579 memslot->dirty_bitmap = NULL;
583 * Free any memory in @free but not in @dont.
585 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
586 struct kvm_memory_slot *dont)
588 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
589 kvm_destroy_dirty_bitmap(free);
591 kvm_arch_free_memslot(kvm, free, dont);
596 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
598 struct kvm_memory_slot *memslot;
603 kvm_for_each_memslot(memslot, slots)
604 kvm_free_memslot(kvm, memslot, NULL);
609 static void kvm_destroy_vm_debugfs(struct kvm *kvm)
613 if (!kvm->debugfs_dentry)
616 debugfs_remove_recursive(kvm->debugfs_dentry);
618 if (kvm->debugfs_stat_data) {
619 for (i = 0; i < kvm_debugfs_num_entries; i++)
620 kfree(kvm->debugfs_stat_data[i]);
621 kfree(kvm->debugfs_stat_data);
625 static int kvm_create_vm_debugfs(struct kvm *kvm, int fd)
627 char dir_name[ITOA_MAX_LEN * 2];
628 struct kvm_stat_data *stat_data;
629 struct kvm_stats_debugfs_item *p;
631 if (!debugfs_initialized())
634 snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd);
635 kvm->debugfs_dentry = debugfs_create_dir(dir_name, kvm_debugfs_dir);
637 kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries,
638 sizeof(*kvm->debugfs_stat_data),
640 if (!kvm->debugfs_stat_data)
643 for (p = debugfs_entries; p->name; p++) {
644 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT);
648 stat_data->kvm = kvm;
649 stat_data->dbgfs_item = p;
650 kvm->debugfs_stat_data[p - debugfs_entries] = stat_data;
651 debugfs_create_file(p->name, KVM_DBGFS_GET_MODE(p),
652 kvm->debugfs_dentry, stat_data,
659 * Called after the VM is otherwise initialized, but just before adding it to
662 int __weak kvm_arch_post_init_vm(struct kvm *kvm)
668 * Called just after removing the VM from the vm_list, but before doing any
671 void __weak kvm_arch_pre_destroy_vm(struct kvm *kvm)
675 static struct kvm *kvm_create_vm(unsigned long type)
677 struct kvm *kvm = kvm_arch_alloc_vm();
682 return ERR_PTR(-ENOMEM);
684 spin_lock_init(&kvm->mmu_lock);
686 kvm->mm = current->mm;
687 kvm_eventfd_init(kvm);
688 mutex_init(&kvm->lock);
689 mutex_init(&kvm->irq_lock);
690 mutex_init(&kvm->slots_lock);
691 INIT_LIST_HEAD(&kvm->devices);
693 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
695 if (init_srcu_struct(&kvm->srcu))
696 goto out_err_no_srcu;
697 if (init_srcu_struct(&kvm->irq_srcu))
698 goto out_err_no_irq_srcu;
700 refcount_set(&kvm->users_count, 1);
701 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
702 struct kvm_memslots *slots = kvm_alloc_memslots();
705 goto out_err_no_arch_destroy_vm;
706 /* Generations must be different for each address space. */
707 slots->generation = i;
708 rcu_assign_pointer(kvm->memslots[i], slots);
711 for (i = 0; i < KVM_NR_BUSES; i++) {
712 rcu_assign_pointer(kvm->buses[i],
713 kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL_ACCOUNT));
715 goto out_err_no_arch_destroy_vm;
718 r = kvm_arch_init_vm(kvm, type);
720 goto out_err_no_arch_destroy_vm;
722 r = hardware_enable_all();
724 goto out_err_no_disable;
726 #ifdef CONFIG_HAVE_KVM_IRQFD
727 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
730 r = kvm_init_mmu_notifier(kvm);
732 goto out_err_no_mmu_notifier;
734 r = kvm_arch_post_init_vm(kvm);
738 mutex_lock(&kvm_lock);
739 list_add(&kvm->vm_list, &vm_list);
740 mutex_unlock(&kvm_lock);
742 preempt_notifier_inc();
747 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
748 if (kvm->mmu_notifier.ops)
749 mmu_notifier_unregister(&kvm->mmu_notifier, current->mm);
751 out_err_no_mmu_notifier:
752 hardware_disable_all();
754 kvm_arch_destroy_vm(kvm);
755 out_err_no_arch_destroy_vm:
756 WARN_ON_ONCE(!refcount_dec_and_test(&kvm->users_count));
757 for (i = 0; i < KVM_NR_BUSES; i++)
758 kfree(kvm_get_bus(kvm, i));
759 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
760 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
761 cleanup_srcu_struct(&kvm->irq_srcu);
763 cleanup_srcu_struct(&kvm->srcu);
765 kvm_arch_free_vm(kvm);
770 static void kvm_destroy_devices(struct kvm *kvm)
772 struct kvm_device *dev, *tmp;
775 * We do not need to take the kvm->lock here, because nobody else
776 * has a reference to the struct kvm at this point and therefore
777 * cannot access the devices list anyhow.
779 list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
780 list_del(&dev->vm_node);
781 dev->ops->destroy(dev);
785 static void kvm_destroy_vm(struct kvm *kvm)
788 struct mm_struct *mm = kvm->mm;
790 kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm);
791 kvm_destroy_vm_debugfs(kvm);
792 kvm_arch_sync_events(kvm);
793 mutex_lock(&kvm_lock);
794 list_del(&kvm->vm_list);
795 mutex_unlock(&kvm_lock);
796 kvm_arch_pre_destroy_vm(kvm);
798 kvm_free_irq_routing(kvm);
799 for (i = 0; i < KVM_NR_BUSES; i++) {
800 struct kvm_io_bus *bus = kvm_get_bus(kvm, i);
803 kvm_io_bus_destroy(bus);
804 kvm->buses[i] = NULL;
806 kvm_coalesced_mmio_free(kvm);
807 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
808 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
810 kvm_arch_flush_shadow_all(kvm);
812 kvm_arch_destroy_vm(kvm);
813 kvm_destroy_devices(kvm);
814 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
815 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
816 cleanup_srcu_struct(&kvm->irq_srcu);
817 cleanup_srcu_struct(&kvm->srcu);
818 kvm_arch_free_vm(kvm);
819 preempt_notifier_dec();
820 hardware_disable_all();
824 void kvm_get_kvm(struct kvm *kvm)
826 refcount_inc(&kvm->users_count);
828 EXPORT_SYMBOL_GPL(kvm_get_kvm);
830 void kvm_put_kvm(struct kvm *kvm)
832 if (refcount_dec_and_test(&kvm->users_count))
835 EXPORT_SYMBOL_GPL(kvm_put_kvm);
838 * Used to put a reference that was taken on behalf of an object associated
839 * with a user-visible file descriptor, e.g. a vcpu or device, if installation
840 * of the new file descriptor fails and the reference cannot be transferred to
841 * its final owner. In such cases, the caller is still actively using @kvm and
842 * will fail miserably if the refcount unexpectedly hits zero.
844 void kvm_put_kvm_no_destroy(struct kvm *kvm)
846 WARN_ON(refcount_dec_and_test(&kvm->users_count));
848 EXPORT_SYMBOL_GPL(kvm_put_kvm_no_destroy);
850 static int kvm_vm_release(struct inode *inode, struct file *filp)
852 struct kvm *kvm = filp->private_data;
854 kvm_irqfd_release(kvm);
861 * Allocation size is twice as large as the actual dirty bitmap size.
862 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
864 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
866 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
868 memslot->dirty_bitmap = kvzalloc(dirty_bytes, GFP_KERNEL_ACCOUNT);
869 if (!memslot->dirty_bitmap)
876 * Insert memslot and re-sort memslots based on their GFN,
877 * so binary search could be used to lookup GFN.
878 * Sorting algorithm takes advantage of having initially
879 * sorted array and known changed memslot position.
881 static void update_memslots(struct kvm_memslots *slots,
882 struct kvm_memory_slot *new,
883 enum kvm_mr_change change)
886 int i = slots->id_to_index[id];
887 struct kvm_memory_slot *mslots = slots->memslots;
889 WARN_ON(mslots[i].id != id);
893 WARN_ON(mslots[i].npages || !new->npages);
897 WARN_ON(new->npages || !mslots[i].npages);
903 while (i < KVM_MEM_SLOTS_NUM - 1 &&
904 new->base_gfn <= mslots[i + 1].base_gfn) {
905 if (!mslots[i + 1].npages)
907 mslots[i] = mslots[i + 1];
908 slots->id_to_index[mslots[i].id] = i;
913 * The ">=" is needed when creating a slot with base_gfn == 0,
914 * so that it moves before all those with base_gfn == npages == 0.
916 * On the other hand, if new->npages is zero, the above loop has
917 * already left i pointing to the beginning of the empty part of
918 * mslots, and the ">=" would move the hole backwards in this
919 * case---which is wrong. So skip the loop when deleting a slot.
923 new->base_gfn >= mslots[i - 1].base_gfn) {
924 mslots[i] = mslots[i - 1];
925 slots->id_to_index[mslots[i].id] = i;
929 WARN_ON_ONCE(i != slots->used_slots);
932 slots->id_to_index[mslots[i].id] = i;
935 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
937 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
939 #ifdef __KVM_HAVE_READONLY_MEM
940 valid_flags |= KVM_MEM_READONLY;
943 if (mem->flags & ~valid_flags)
949 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
950 int as_id, struct kvm_memslots *slots)
952 struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id);
953 u64 gen = old_memslots->generation;
955 WARN_ON(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS);
956 slots->generation = gen | KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
958 rcu_assign_pointer(kvm->memslots[as_id], slots);
959 synchronize_srcu_expedited(&kvm->srcu);
962 * Increment the new memslot generation a second time, dropping the
963 * update in-progress flag and incrementing the generation based on
964 * the number of address spaces. This provides a unique and easily
965 * identifiable generation number while the memslots are in flux.
967 gen = slots->generation & ~KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
970 * Generations must be unique even across address spaces. We do not need
971 * a global counter for that, instead the generation space is evenly split
972 * across address spaces. For example, with two address spaces, address
973 * space 0 will use generations 0, 2, 4, ... while address space 1 will
974 * use generations 1, 3, 5, ...
976 gen += KVM_ADDRESS_SPACE_NUM;
978 kvm_arch_memslots_updated(kvm, gen);
980 slots->generation = gen;
986 * Allocate some memory and give it an address in the guest physical address
989 * Discontiguous memory is allowed, mostly for framebuffers.
991 * Must be called holding kvm->slots_lock for write.
993 int __kvm_set_memory_region(struct kvm *kvm,
994 const struct kvm_userspace_memory_region *mem)
998 unsigned long npages;
999 struct kvm_memory_slot *slot;
1000 struct kvm_memory_slot old, new;
1001 struct kvm_memslots *slots = NULL, *old_memslots;
1003 enum kvm_mr_change change;
1005 r = check_memory_region_flags(mem);
1010 as_id = mem->slot >> 16;
1011 id = (u16)mem->slot;
1013 /* General sanity checks */
1014 if (mem->memory_size & (PAGE_SIZE - 1))
1016 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
1018 /* We can read the guest memory with __xxx_user() later on. */
1019 if ((id < KVM_USER_MEM_SLOTS) &&
1020 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
1021 !access_ok((void __user *)(unsigned long)mem->userspace_addr,
1024 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
1026 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
1029 slot = id_to_memslot(__kvm_memslots(kvm, as_id), id);
1030 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
1031 npages = mem->memory_size >> PAGE_SHIFT;
1033 if (npages > KVM_MEM_MAX_NR_PAGES)
1039 new.base_gfn = base_gfn;
1040 new.npages = npages;
1041 new.flags = mem->flags;
1045 change = KVM_MR_CREATE;
1046 else { /* Modify an existing slot. */
1047 if ((mem->userspace_addr != old.userspace_addr) ||
1048 (npages != old.npages) ||
1049 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
1052 if (base_gfn != old.base_gfn)
1053 change = KVM_MR_MOVE;
1054 else if (new.flags != old.flags)
1055 change = KVM_MR_FLAGS_ONLY;
1056 else { /* Nothing to change. */
1065 change = KVM_MR_DELETE;
1070 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
1071 /* Check for overlaps */
1073 kvm_for_each_memslot(slot, __kvm_memslots(kvm, as_id)) {
1076 if (!((base_gfn + npages <= slot->base_gfn) ||
1077 (base_gfn >= slot->base_gfn + slot->npages)))
1082 /* Free page dirty bitmap if unneeded */
1083 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
1084 new.dirty_bitmap = NULL;
1087 if (change == KVM_MR_CREATE) {
1088 new.userspace_addr = mem->userspace_addr;
1090 if (kvm_arch_create_memslot(kvm, &new, npages))
1094 /* Allocate page dirty bitmap if needed */
1095 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
1096 if (kvm_create_dirty_bitmap(&new) < 0)
1100 slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL_ACCOUNT);
1103 memcpy(slots, __kvm_memslots(kvm, as_id), sizeof(struct kvm_memslots));
1105 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
1106 slot = id_to_memslot(slots, id);
1107 slot->flags |= KVM_MEMSLOT_INVALID;
1109 old_memslots = install_new_memslots(kvm, as_id, slots);
1111 /* From this point no new shadow pages pointing to a deleted,
1112 * or moved, memslot will be created.
1114 * validation of sp->gfn happens in:
1115 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1116 * - kvm_is_visible_gfn (mmu_check_root)
1118 kvm_arch_flush_shadow_memslot(kvm, slot);
1121 * We can re-use the old_memslots from above, the only difference
1122 * from the currently installed memslots is the invalid flag. This
1123 * will get overwritten by update_memslots anyway.
1125 slots = old_memslots;
1128 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
1132 /* actual memory is freed via old in kvm_free_memslot below */
1133 if (change == KVM_MR_DELETE) {
1134 new.dirty_bitmap = NULL;
1135 memset(&new.arch, 0, sizeof(new.arch));
1138 update_memslots(slots, &new, change);
1139 old_memslots = install_new_memslots(kvm, as_id, slots);
1141 kvm_arch_commit_memory_region(kvm, mem, &old, &new, change);
1143 kvm_free_memslot(kvm, &old, &new);
1144 kvfree(old_memslots);
1150 kvm_free_memslot(kvm, &new, &old);
1154 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1156 int kvm_set_memory_region(struct kvm *kvm,
1157 const struct kvm_userspace_memory_region *mem)
1161 mutex_lock(&kvm->slots_lock);
1162 r = __kvm_set_memory_region(kvm, mem);
1163 mutex_unlock(&kvm->slots_lock);
1166 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1168 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1169 struct kvm_userspace_memory_region *mem)
1171 if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1174 return kvm_set_memory_region(kvm, mem);
1177 int kvm_get_dirty_log(struct kvm *kvm,
1178 struct kvm_dirty_log *log, int *is_dirty)
1180 struct kvm_memslots *slots;
1181 struct kvm_memory_slot *memslot;
1184 unsigned long any = 0;
1186 as_id = log->slot >> 16;
1187 id = (u16)log->slot;
1188 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1191 slots = __kvm_memslots(kvm, as_id);
1192 memslot = id_to_memslot(slots, id);
1193 if (!memslot->dirty_bitmap)
1196 n = kvm_dirty_bitmap_bytes(memslot);
1198 for (i = 0; !any && i < n/sizeof(long); ++i)
1199 any = memslot->dirty_bitmap[i];
1201 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1208 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1210 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1212 * kvm_get_dirty_log_protect - get a snapshot of dirty pages
1213 * and reenable dirty page tracking for the corresponding pages.
1214 * @kvm: pointer to kvm instance
1215 * @log: slot id and address to which we copy the log
1216 * @flush: true if TLB flush is needed by caller
1218 * We need to keep it in mind that VCPU threads can write to the bitmap
1219 * concurrently. So, to avoid losing track of dirty pages we keep the
1222 * 1. Take a snapshot of the bit and clear it if needed.
1223 * 2. Write protect the corresponding page.
1224 * 3. Copy the snapshot to the userspace.
1225 * 4. Upon return caller flushes TLB's if needed.
1227 * Between 2 and 4, the guest may write to the page using the remaining TLB
1228 * entry. This is not a problem because the page is reported dirty using
1229 * the snapshot taken before and step 4 ensures that writes done after
1230 * exiting to userspace will be logged for the next call.
1233 int kvm_get_dirty_log_protect(struct kvm *kvm,
1234 struct kvm_dirty_log *log, bool *flush)
1236 struct kvm_memslots *slots;
1237 struct kvm_memory_slot *memslot;
1240 unsigned long *dirty_bitmap;
1241 unsigned long *dirty_bitmap_buffer;
1243 as_id = log->slot >> 16;
1244 id = (u16)log->slot;
1245 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1248 slots = __kvm_memslots(kvm, as_id);
1249 memslot = id_to_memslot(slots, id);
1251 dirty_bitmap = memslot->dirty_bitmap;
1255 n = kvm_dirty_bitmap_bytes(memslot);
1257 if (kvm->manual_dirty_log_protect) {
1259 * Unlike kvm_get_dirty_log, we always return false in *flush,
1260 * because no flush is needed until KVM_CLEAR_DIRTY_LOG. There
1261 * is some code duplication between this function and
1262 * kvm_get_dirty_log, but hopefully all architecture
1263 * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log
1264 * can be eliminated.
1266 dirty_bitmap_buffer = dirty_bitmap;
1268 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
1269 memset(dirty_bitmap_buffer, 0, n);
1271 spin_lock(&kvm->mmu_lock);
1272 for (i = 0; i < n / sizeof(long); i++) {
1276 if (!dirty_bitmap[i])
1280 mask = xchg(&dirty_bitmap[i], 0);
1281 dirty_bitmap_buffer[i] = mask;
1283 offset = i * BITS_PER_LONG;
1284 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1287 spin_unlock(&kvm->mmu_lock);
1290 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1294 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect);
1297 * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap
1298 * and reenable dirty page tracking for the corresponding pages.
1299 * @kvm: pointer to kvm instance
1300 * @log: slot id and address from which to fetch the bitmap of dirty pages
1301 * @flush: true if TLB flush is needed by caller
1303 int kvm_clear_dirty_log_protect(struct kvm *kvm,
1304 struct kvm_clear_dirty_log *log, bool *flush)
1306 struct kvm_memslots *slots;
1307 struct kvm_memory_slot *memslot;
1311 unsigned long *dirty_bitmap;
1312 unsigned long *dirty_bitmap_buffer;
1314 as_id = log->slot >> 16;
1315 id = (u16)log->slot;
1316 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1319 if (log->first_page & 63)
1322 slots = __kvm_memslots(kvm, as_id);
1323 memslot = id_to_memslot(slots, id);
1325 dirty_bitmap = memslot->dirty_bitmap;
1329 n = ALIGN(log->num_pages, BITS_PER_LONG) / 8;
1331 if (log->first_page > memslot->npages ||
1332 log->num_pages > memslot->npages - log->first_page ||
1333 (log->num_pages < memslot->npages - log->first_page && (log->num_pages & 63)))
1337 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
1338 if (copy_from_user(dirty_bitmap_buffer, log->dirty_bitmap, n))
1341 spin_lock(&kvm->mmu_lock);
1342 for (offset = log->first_page, i = offset / BITS_PER_LONG,
1343 n = DIV_ROUND_UP(log->num_pages, BITS_PER_LONG); n--;
1344 i++, offset += BITS_PER_LONG) {
1345 unsigned long mask = *dirty_bitmap_buffer++;
1346 atomic_long_t *p = (atomic_long_t *) &dirty_bitmap[i];
1350 mask &= atomic_long_fetch_andnot(mask, p);
1353 * mask contains the bits that really have been cleared. This
1354 * never includes any bits beyond the length of the memslot (if
1355 * the length is not aligned to 64 pages), therefore it is not
1356 * a problem if userspace sets them in log->dirty_bitmap.
1360 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1364 spin_unlock(&kvm->mmu_lock);
1368 EXPORT_SYMBOL_GPL(kvm_clear_dirty_log_protect);
1371 bool kvm_largepages_enabled(void)
1373 return largepages_enabled;
1376 void kvm_disable_largepages(void)
1378 largepages_enabled = false;
1380 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1382 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1384 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1386 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1388 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
1390 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn);
1393 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1395 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1397 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1398 memslot->flags & KVM_MEMSLOT_INVALID)
1403 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1405 unsigned long kvm_host_page_size(struct kvm_vcpu *vcpu, gfn_t gfn)
1407 struct vm_area_struct *vma;
1408 unsigned long addr, size;
1412 addr = kvm_vcpu_gfn_to_hva_prot(vcpu, gfn, NULL);
1413 if (kvm_is_error_hva(addr))
1416 down_read(¤t->mm->mmap_sem);
1417 vma = find_vma(current->mm, addr);
1421 size = vma_kernel_pagesize(vma);
1424 up_read(¤t->mm->mmap_sem);
1429 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1431 return slot->flags & KVM_MEM_READONLY;
1434 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1435 gfn_t *nr_pages, bool write)
1437 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1438 return KVM_HVA_ERR_BAD;
1440 if (memslot_is_readonly(slot) && write)
1441 return KVM_HVA_ERR_RO_BAD;
1444 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1446 return __gfn_to_hva_memslot(slot, gfn);
1449 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1452 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1455 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1458 return gfn_to_hva_many(slot, gfn, NULL);
1460 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1462 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1464 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1466 EXPORT_SYMBOL_GPL(gfn_to_hva);
1468 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
1470 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
1472 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
1475 * Return the hva of a @gfn and the R/W attribute if possible.
1477 * @slot: the kvm_memory_slot which contains @gfn
1478 * @gfn: the gfn to be translated
1479 * @writable: used to return the read/write attribute of the @slot if the hva
1480 * is valid and @writable is not NULL
1482 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1483 gfn_t gfn, bool *writable)
1485 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1487 if (!kvm_is_error_hva(hva) && writable)
1488 *writable = !memslot_is_readonly(slot);
1493 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1495 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1497 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1500 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
1502 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1504 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1507 static inline int check_user_page_hwpoison(unsigned long addr)
1509 int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
1511 rc = get_user_pages(addr, 1, flags, NULL, NULL);
1512 return rc == -EHWPOISON;
1516 * The fast path to get the writable pfn which will be stored in @pfn,
1517 * true indicates success, otherwise false is returned. It's also the
1518 * only part that runs if we can in atomic context.
1520 static bool hva_to_pfn_fast(unsigned long addr, bool write_fault,
1521 bool *writable, kvm_pfn_t *pfn)
1523 struct page *page[1];
1527 * Fast pin a writable pfn only if it is a write fault request
1528 * or the caller allows to map a writable pfn for a read fault
1531 if (!(write_fault || writable))
1534 npages = __get_user_pages_fast(addr, 1, 1, page);
1536 *pfn = page_to_pfn(page[0]);
1547 * The slow path to get the pfn of the specified host virtual address,
1548 * 1 indicates success, -errno is returned if error is detected.
1550 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1551 bool *writable, kvm_pfn_t *pfn)
1553 unsigned int flags = FOLL_HWPOISON;
1560 *writable = write_fault;
1563 flags |= FOLL_WRITE;
1565 flags |= FOLL_NOWAIT;
1567 npages = get_user_pages_unlocked(addr, 1, &page, flags);
1571 /* map read fault as writable if possible */
1572 if (unlikely(!write_fault) && writable) {
1575 if (__get_user_pages_fast(addr, 1, 1, &wpage) == 1) {
1581 *pfn = page_to_pfn(page);
1585 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1587 if (unlikely(!(vma->vm_flags & VM_READ)))
1590 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1596 static int hva_to_pfn_remapped(struct vm_area_struct *vma,
1597 unsigned long addr, bool *async,
1598 bool write_fault, bool *writable,
1604 r = follow_pfn(vma, addr, &pfn);
1607 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
1608 * not call the fault handler, so do it here.
1610 bool unlocked = false;
1611 r = fixup_user_fault(current, current->mm, addr,
1612 (write_fault ? FAULT_FLAG_WRITE : 0),
1619 r = follow_pfn(vma, addr, &pfn);
1629 * Get a reference here because callers of *hva_to_pfn* and
1630 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
1631 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
1632 * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will
1633 * simply do nothing for reserved pfns.
1635 * Whoever called remap_pfn_range is also going to call e.g.
1636 * unmap_mapping_range before the underlying pages are freed,
1637 * causing a call to our MMU notifier.
1646 * Pin guest page in memory and return its pfn.
1647 * @addr: host virtual address which maps memory to the guest
1648 * @atomic: whether this function can sleep
1649 * @async: whether this function need to wait IO complete if the
1650 * host page is not in the memory
1651 * @write_fault: whether we should get a writable host page
1652 * @writable: whether it allows to map a writable host page for !@write_fault
1654 * The function will map a writable host page for these two cases:
1655 * 1): @write_fault = true
1656 * 2): @write_fault = false && @writable, @writable will tell the caller
1657 * whether the mapping is writable.
1659 static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1660 bool write_fault, bool *writable)
1662 struct vm_area_struct *vma;
1666 /* we can do it either atomically or asynchronously, not both */
1667 BUG_ON(atomic && async);
1669 if (hva_to_pfn_fast(addr, write_fault, writable, &pfn))
1673 return KVM_PFN_ERR_FAULT;
1675 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1679 down_read(¤t->mm->mmap_sem);
1680 if (npages == -EHWPOISON ||
1681 (!async && check_user_page_hwpoison(addr))) {
1682 pfn = KVM_PFN_ERR_HWPOISON;
1687 vma = find_vma_intersection(current->mm, addr, addr + 1);
1690 pfn = KVM_PFN_ERR_FAULT;
1691 else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
1692 r = hva_to_pfn_remapped(vma, addr, async, write_fault, writable, &pfn);
1696 pfn = KVM_PFN_ERR_FAULT;
1698 if (async && vma_is_valid(vma, write_fault))
1700 pfn = KVM_PFN_ERR_FAULT;
1703 up_read(¤t->mm->mmap_sem);
1707 kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn,
1708 bool atomic, bool *async, bool write_fault,
1711 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1713 if (addr == KVM_HVA_ERR_RO_BAD) {
1716 return KVM_PFN_ERR_RO_FAULT;
1719 if (kvm_is_error_hva(addr)) {
1722 return KVM_PFN_NOSLOT;
1725 /* Do not map writable pfn in the readonly memslot. */
1726 if (writable && memslot_is_readonly(slot)) {
1731 return hva_to_pfn(addr, atomic, async, write_fault,
1734 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
1736 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1739 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
1740 write_fault, writable);
1742 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1744 kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1746 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1748 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
1750 kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1752 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1754 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1756 kvm_pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1758 return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm, gfn), gfn);
1760 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1762 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
1764 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1766 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
1768 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1770 return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
1772 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1774 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
1776 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1778 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
1780 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1781 struct page **pages, int nr_pages)
1786 addr = gfn_to_hva_many(slot, gfn, &entry);
1787 if (kvm_is_error_hva(addr))
1790 if (entry < nr_pages)
1793 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1795 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1797 static struct page *kvm_pfn_to_page(kvm_pfn_t pfn)
1799 if (is_error_noslot_pfn(pfn))
1800 return KVM_ERR_PTR_BAD_PAGE;
1802 if (kvm_is_reserved_pfn(pfn)) {
1804 return KVM_ERR_PTR_BAD_PAGE;
1807 return pfn_to_page(pfn);
1810 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1814 pfn = gfn_to_pfn(kvm, gfn);
1816 return kvm_pfn_to_page(pfn);
1818 EXPORT_SYMBOL_GPL(gfn_to_page);
1820 void kvm_release_pfn(kvm_pfn_t pfn, bool dirty, struct gfn_to_pfn_cache *cache)
1826 cache->pfn = cache->gfn = 0;
1829 kvm_release_pfn_dirty(pfn);
1831 kvm_release_pfn_clean(pfn);
1834 static void kvm_cache_gfn_to_pfn(struct kvm_memory_slot *slot, gfn_t gfn,
1835 struct gfn_to_pfn_cache *cache, u64 gen)
1837 kvm_release_pfn(cache->pfn, cache->dirty, cache);
1839 cache->pfn = gfn_to_pfn_memslot(slot, gfn);
1841 cache->dirty = false;
1842 cache->generation = gen;
1845 static int __kvm_map_gfn(struct kvm_memslots *slots, gfn_t gfn,
1846 struct kvm_host_map *map,
1847 struct gfn_to_pfn_cache *cache,
1852 struct page *page = KVM_UNMAPPED_PAGE;
1853 struct kvm_memory_slot *slot = __gfn_to_memslot(slots, gfn);
1854 u64 gen = slots->generation;
1860 if (!cache->pfn || cache->gfn != gfn ||
1861 cache->generation != gen) {
1864 kvm_cache_gfn_to_pfn(slot, gfn, cache, gen);
1870 pfn = gfn_to_pfn_memslot(slot, gfn);
1872 if (is_error_noslot_pfn(pfn))
1875 if (pfn_valid(pfn)) {
1876 page = pfn_to_page(pfn);
1878 hva = kmap_atomic(page);
1881 #ifdef CONFIG_HAS_IOMEM
1882 } else if (!atomic) {
1883 hva = memremap(pfn_to_hpa(pfn), PAGE_SIZE, MEMREMAP_WB);
1900 int kvm_map_gfn(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map,
1901 struct gfn_to_pfn_cache *cache, bool atomic)
1903 return __kvm_map_gfn(kvm_memslots(vcpu->kvm), gfn, map,
1906 EXPORT_SYMBOL_GPL(kvm_map_gfn);
1908 int kvm_vcpu_map(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map)
1910 return __kvm_map_gfn(kvm_vcpu_memslots(vcpu), gfn, map,
1913 EXPORT_SYMBOL_GPL(kvm_vcpu_map);
1915 static void __kvm_unmap_gfn(struct kvm_memory_slot *memslot,
1916 struct kvm_host_map *map,
1917 struct gfn_to_pfn_cache *cache,
1918 bool dirty, bool atomic)
1926 if (map->page != KVM_UNMAPPED_PAGE) {
1928 kunmap_atomic(map->hva);
1932 #ifdef CONFIG_HAS_IOMEM
1936 WARN_ONCE(1, "Unexpected unmapping in atomic context");
1940 mark_page_dirty_in_slot(memslot, map->gfn);
1943 cache->dirty |= dirty;
1945 kvm_release_pfn(map->pfn, dirty, NULL);
1951 int kvm_unmap_gfn(struct kvm_vcpu *vcpu, struct kvm_host_map *map,
1952 struct gfn_to_pfn_cache *cache, bool dirty, bool atomic)
1954 __kvm_unmap_gfn(gfn_to_memslot(vcpu->kvm, map->gfn), map,
1955 cache, dirty, atomic);
1958 EXPORT_SYMBOL_GPL(kvm_unmap_gfn);
1960 void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map, bool dirty)
1962 __kvm_unmap_gfn(kvm_vcpu_gfn_to_memslot(vcpu, map->gfn), map, NULL,
1965 EXPORT_SYMBOL_GPL(kvm_vcpu_unmap);
1967 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
1971 pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
1973 return kvm_pfn_to_page(pfn);
1975 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
1977 void kvm_release_page_clean(struct page *page)
1979 WARN_ON(is_error_page(page));
1981 kvm_release_pfn_clean(page_to_pfn(page));
1983 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1985 void kvm_release_pfn_clean(kvm_pfn_t pfn)
1987 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
1988 put_page(pfn_to_page(pfn));
1990 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1992 void kvm_release_page_dirty(struct page *page)
1994 WARN_ON(is_error_page(page));
1996 kvm_release_pfn_dirty(page_to_pfn(page));
1998 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
2000 void kvm_release_pfn_dirty(kvm_pfn_t pfn)
2002 kvm_set_pfn_dirty(pfn);
2003 kvm_release_pfn_clean(pfn);
2005 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
2007 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
2009 if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn))
2010 SetPageDirty(pfn_to_page(pfn));
2012 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
2014 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
2016 if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn))
2017 mark_page_accessed(pfn_to_page(pfn));
2019 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
2021 void kvm_get_pfn(kvm_pfn_t pfn)
2023 if (!kvm_is_reserved_pfn(pfn))
2024 get_page(pfn_to_page(pfn));
2026 EXPORT_SYMBOL_GPL(kvm_get_pfn);
2028 static int next_segment(unsigned long len, int offset)
2030 if (len > PAGE_SIZE - offset)
2031 return PAGE_SIZE - offset;
2036 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
2037 void *data, int offset, int len)
2042 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
2043 if (kvm_is_error_hva(addr))
2045 r = __copy_from_user(data, (void __user *)addr + offset, len);
2051 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
2054 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2056 return __kvm_read_guest_page(slot, gfn, data, offset, len);
2058 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
2060 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
2061 int offset, int len)
2063 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2065 return __kvm_read_guest_page(slot, gfn, data, offset, len);
2067 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
2069 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
2071 gfn_t gfn = gpa >> PAGE_SHIFT;
2073 int offset = offset_in_page(gpa);
2076 while ((seg = next_segment(len, offset)) != 0) {
2077 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
2087 EXPORT_SYMBOL_GPL(kvm_read_guest);
2089 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
2091 gfn_t gfn = gpa >> PAGE_SHIFT;
2093 int offset = offset_in_page(gpa);
2096 while ((seg = next_segment(len, offset)) != 0) {
2097 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
2107 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
2109 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
2110 void *data, int offset, unsigned long len)
2115 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
2116 if (kvm_is_error_hva(addr))
2118 pagefault_disable();
2119 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
2126 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
2127 void *data, unsigned long len)
2129 gfn_t gfn = gpa >> PAGE_SHIFT;
2130 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2131 int offset = offset_in_page(gpa);
2133 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
2135 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
2137 static int __kvm_write_guest_page(struct kvm_memory_slot *memslot, gfn_t gfn,
2138 const void *data, int offset, int len)
2143 addr = gfn_to_hva_memslot(memslot, gfn);
2144 if (kvm_is_error_hva(addr))
2146 r = __copy_to_user((void __user *)addr + offset, data, len);
2149 mark_page_dirty_in_slot(memslot, gfn);
2153 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
2154 const void *data, int offset, int len)
2156 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2158 return __kvm_write_guest_page(slot, gfn, data, offset, len);
2160 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
2162 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
2163 const void *data, int offset, int len)
2165 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2167 return __kvm_write_guest_page(slot, gfn, data, offset, len);
2169 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
2171 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
2174 gfn_t gfn = gpa >> PAGE_SHIFT;
2176 int offset = offset_in_page(gpa);
2179 while ((seg = next_segment(len, offset)) != 0) {
2180 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
2190 EXPORT_SYMBOL_GPL(kvm_write_guest);
2192 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
2195 gfn_t gfn = gpa >> PAGE_SHIFT;
2197 int offset = offset_in_page(gpa);
2200 while ((seg = next_segment(len, offset)) != 0) {
2201 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
2211 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
2213 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
2214 struct gfn_to_hva_cache *ghc,
2215 gpa_t gpa, unsigned long len)
2217 int offset = offset_in_page(gpa);
2218 gfn_t start_gfn = gpa >> PAGE_SHIFT;
2219 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
2220 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
2221 gfn_t nr_pages_avail;
2223 /* Update ghc->generation before performing any error checks. */
2224 ghc->generation = slots->generation;
2226 if (start_gfn > end_gfn) {
2227 ghc->hva = KVM_HVA_ERR_BAD;
2232 * If the requested region crosses two memslots, we still
2233 * verify that the entire region is valid here.
2235 for ( ; start_gfn <= end_gfn; start_gfn += nr_pages_avail) {
2236 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
2237 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
2239 if (kvm_is_error_hva(ghc->hva))
2243 /* Use the slow path for cross page reads and writes. */
2244 if (nr_pages_needed == 1)
2247 ghc->memslot = NULL;
2254 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2255 gpa_t gpa, unsigned long len)
2257 struct kvm_memslots *slots = kvm_memslots(kvm);
2258 return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len);
2260 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
2262 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2263 void *data, unsigned int offset,
2266 struct kvm_memslots *slots = kvm_memslots(kvm);
2268 gpa_t gpa = ghc->gpa + offset;
2270 BUG_ON(len + offset > ghc->len);
2272 if (slots->generation != ghc->generation) {
2273 if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len))
2277 if (kvm_is_error_hva(ghc->hva))
2280 if (unlikely(!ghc->memslot))
2281 return kvm_write_guest(kvm, gpa, data, len);
2283 r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
2286 mark_page_dirty_in_slot(ghc->memslot, gpa >> PAGE_SHIFT);
2290 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached);
2292 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2293 void *data, unsigned long len)
2295 return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len);
2297 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
2299 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2300 void *data, unsigned long len)
2302 struct kvm_memslots *slots = kvm_memslots(kvm);
2305 BUG_ON(len > ghc->len);
2307 if (slots->generation != ghc->generation) {
2308 if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len))
2312 if (kvm_is_error_hva(ghc->hva))
2315 if (unlikely(!ghc->memslot))
2316 return kvm_read_guest(kvm, ghc->gpa, data, len);
2318 r = __copy_from_user(data, (void __user *)ghc->hva, len);
2324 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
2326 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
2328 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
2330 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
2332 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
2334 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
2336 gfn_t gfn = gpa >> PAGE_SHIFT;
2338 int offset = offset_in_page(gpa);
2341 while ((seg = next_segment(len, offset)) != 0) {
2342 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
2351 EXPORT_SYMBOL_GPL(kvm_clear_guest);
2353 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot,
2356 if (memslot && memslot->dirty_bitmap) {
2357 unsigned long rel_gfn = gfn - memslot->base_gfn;
2359 set_bit_le(rel_gfn, memslot->dirty_bitmap);
2363 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
2365 struct kvm_memory_slot *memslot;
2367 memslot = gfn_to_memslot(kvm, gfn);
2368 mark_page_dirty_in_slot(memslot, gfn);
2370 EXPORT_SYMBOL_GPL(mark_page_dirty);
2372 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
2374 struct kvm_memory_slot *memslot;
2376 memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2377 mark_page_dirty_in_slot(memslot, gfn);
2379 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
2381 void kvm_sigset_activate(struct kvm_vcpu *vcpu)
2383 if (!vcpu->sigset_active)
2387 * This does a lockless modification of ->real_blocked, which is fine
2388 * because, only current can change ->real_blocked and all readers of
2389 * ->real_blocked don't care as long ->real_blocked is always a subset
2392 sigprocmask(SIG_SETMASK, &vcpu->sigset, ¤t->real_blocked);
2395 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu)
2397 if (!vcpu->sigset_active)
2400 sigprocmask(SIG_SETMASK, ¤t->real_blocked, NULL);
2401 sigemptyset(¤t->real_blocked);
2404 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
2406 unsigned int old, val, grow, grow_start;
2408 old = val = vcpu->halt_poll_ns;
2409 grow_start = READ_ONCE(halt_poll_ns_grow_start);
2410 grow = READ_ONCE(halt_poll_ns_grow);
2415 if (val < grow_start)
2418 if (val > halt_poll_ns)
2421 vcpu->halt_poll_ns = val;
2423 trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
2426 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
2428 unsigned int old, val, shrink;
2430 old = val = vcpu->halt_poll_ns;
2431 shrink = READ_ONCE(halt_poll_ns_shrink);
2437 vcpu->halt_poll_ns = val;
2438 trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
2441 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
2444 int idx = srcu_read_lock(&vcpu->kvm->srcu);
2446 if (kvm_arch_vcpu_runnable(vcpu)) {
2447 kvm_make_request(KVM_REQ_UNHALT, vcpu);
2450 if (kvm_cpu_has_pending_timer(vcpu))
2452 if (signal_pending(current))
2457 srcu_read_unlock(&vcpu->kvm->srcu, idx);
2462 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2464 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
2467 DECLARE_SWAITQUEUE(wait);
2468 bool waited = false;
2471 kvm_arch_vcpu_blocking(vcpu);
2473 start = cur = ktime_get();
2474 if (vcpu->halt_poll_ns && !kvm_arch_no_poll(vcpu)) {
2475 ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns);
2477 ++vcpu->stat.halt_attempted_poll;
2480 * This sets KVM_REQ_UNHALT if an interrupt
2483 if (kvm_vcpu_check_block(vcpu) < 0) {
2484 ++vcpu->stat.halt_successful_poll;
2485 if (!vcpu_valid_wakeup(vcpu))
2486 ++vcpu->stat.halt_poll_invalid;
2490 } while (single_task_running() && ktime_before(cur, stop));
2494 prepare_to_swait_exclusive(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
2496 if (kvm_vcpu_check_block(vcpu) < 0)
2503 finish_swait(&vcpu->wq, &wait);
2506 kvm_arch_vcpu_unblocking(vcpu);
2507 block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
2509 if (!kvm_arch_no_poll(vcpu)) {
2510 if (!vcpu_valid_wakeup(vcpu)) {
2511 shrink_halt_poll_ns(vcpu);
2512 } else if (halt_poll_ns) {
2513 if (block_ns <= vcpu->halt_poll_ns)
2515 /* we had a long block, shrink polling */
2516 else if (vcpu->halt_poll_ns && block_ns > halt_poll_ns)
2517 shrink_halt_poll_ns(vcpu);
2518 /* we had a short halt and our poll time is too small */
2519 else if (vcpu->halt_poll_ns < halt_poll_ns &&
2520 block_ns < halt_poll_ns)
2521 grow_halt_poll_ns(vcpu);
2523 vcpu->halt_poll_ns = 0;
2527 trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu));
2528 kvm_arch_vcpu_block_finish(vcpu);
2530 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
2532 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
2534 struct swait_queue_head *wqp;
2536 wqp = kvm_arch_vcpu_wq(vcpu);
2537 if (swq_has_sleeper(wqp)) {
2539 WRITE_ONCE(vcpu->ready, true);
2540 ++vcpu->stat.halt_wakeup;
2546 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
2550 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2552 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
2555 int cpu = vcpu->cpu;
2557 if (kvm_vcpu_wake_up(vcpu))
2561 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
2562 if (kvm_arch_vcpu_should_kick(vcpu))
2563 smp_send_reschedule(cpu);
2566 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
2567 #endif /* !CONFIG_S390 */
2569 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
2572 struct task_struct *task = NULL;
2576 pid = rcu_dereference(target->pid);
2578 task = get_pid_task(pid, PIDTYPE_PID);
2582 ret = yield_to(task, 1);
2583 put_task_struct(task);
2587 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
2590 * Helper that checks whether a VCPU is eligible for directed yield.
2591 * Most eligible candidate to yield is decided by following heuristics:
2593 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2594 * (preempted lock holder), indicated by @in_spin_loop.
2595 * Set at the beiginning and cleared at the end of interception/PLE handler.
2597 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2598 * chance last time (mostly it has become eligible now since we have probably
2599 * yielded to lockholder in last iteration. This is done by toggling
2600 * @dy_eligible each time a VCPU checked for eligibility.)
2602 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2603 * to preempted lock-holder could result in wrong VCPU selection and CPU
2604 * burning. Giving priority for a potential lock-holder increases lock
2607 * Since algorithm is based on heuristics, accessing another VCPU data without
2608 * locking does not harm. It may result in trying to yield to same VCPU, fail
2609 * and continue with next VCPU and so on.
2611 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
2613 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2616 eligible = !vcpu->spin_loop.in_spin_loop ||
2617 vcpu->spin_loop.dy_eligible;
2619 if (vcpu->spin_loop.in_spin_loop)
2620 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
2629 * Unlike kvm_arch_vcpu_runnable, this function is called outside
2630 * a vcpu_load/vcpu_put pair. However, for most architectures
2631 * kvm_arch_vcpu_runnable does not require vcpu_load.
2633 bool __weak kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
2635 return kvm_arch_vcpu_runnable(vcpu);
2638 static bool vcpu_dy_runnable(struct kvm_vcpu *vcpu)
2640 if (kvm_arch_dy_runnable(vcpu))
2643 #ifdef CONFIG_KVM_ASYNC_PF
2644 if (!list_empty_careful(&vcpu->async_pf.done))
2651 void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode)
2653 struct kvm *kvm = me->kvm;
2654 struct kvm_vcpu *vcpu;
2655 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
2661 kvm_vcpu_set_in_spin_loop(me, true);
2663 * We boost the priority of a VCPU that is runnable but not
2664 * currently running, because it got preempted by something
2665 * else and called schedule in __vcpu_run. Hopefully that
2666 * VCPU is holding the lock that we need and will release it.
2667 * We approximate round-robin by starting at the last boosted VCPU.
2669 for (pass = 0; pass < 2 && !yielded && try; pass++) {
2670 kvm_for_each_vcpu(i, vcpu, kvm) {
2671 if (!pass && i <= last_boosted_vcpu) {
2672 i = last_boosted_vcpu;
2674 } else if (pass && i > last_boosted_vcpu)
2676 if (!READ_ONCE(vcpu->ready))
2680 if (swait_active(&vcpu->wq) && !vcpu_dy_runnable(vcpu))
2682 if (READ_ONCE(vcpu->preempted) && yield_to_kernel_mode &&
2683 !kvm_arch_vcpu_in_kernel(vcpu))
2685 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
2688 yielded = kvm_vcpu_yield_to(vcpu);
2690 kvm->last_boosted_vcpu = i;
2692 } else if (yielded < 0) {
2699 kvm_vcpu_set_in_spin_loop(me, false);
2701 /* Ensure vcpu is not eligible during next spinloop */
2702 kvm_vcpu_set_dy_eligible(me, false);
2704 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
2706 static vm_fault_t kvm_vcpu_fault(struct vm_fault *vmf)
2708 struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data;
2711 if (vmf->pgoff == 0)
2712 page = virt_to_page(vcpu->run);
2714 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
2715 page = virt_to_page(vcpu->arch.pio_data);
2717 #ifdef CONFIG_KVM_MMIO
2718 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
2719 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
2722 return kvm_arch_vcpu_fault(vcpu, vmf);
2728 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
2729 .fault = kvm_vcpu_fault,
2732 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2734 vma->vm_ops = &kvm_vcpu_vm_ops;
2738 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2740 struct kvm_vcpu *vcpu = filp->private_data;
2742 debugfs_remove_recursive(vcpu->debugfs_dentry);
2743 kvm_put_kvm(vcpu->kvm);
2747 static struct file_operations kvm_vcpu_fops = {
2748 .release = kvm_vcpu_release,
2749 .unlocked_ioctl = kvm_vcpu_ioctl,
2750 .mmap = kvm_vcpu_mmap,
2751 .llseek = noop_llseek,
2752 KVM_COMPAT(kvm_vcpu_compat_ioctl),
2756 * Allocates an inode for the vcpu.
2758 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2760 char name[8 + 1 + ITOA_MAX_LEN + 1];
2762 snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id);
2763 return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
2766 static void kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
2768 #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS
2769 char dir_name[ITOA_MAX_LEN * 2];
2771 if (!debugfs_initialized())
2774 snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
2775 vcpu->debugfs_dentry = debugfs_create_dir(dir_name,
2776 vcpu->kvm->debugfs_dentry);
2778 kvm_arch_create_vcpu_debugfs(vcpu);
2783 * Creates some virtual cpus. Good luck creating more than one.
2785 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
2788 struct kvm_vcpu *vcpu;
2791 if (id >= KVM_MAX_VCPU_ID)
2794 mutex_lock(&kvm->lock);
2795 if (kvm->created_vcpus == KVM_MAX_VCPUS) {
2796 mutex_unlock(&kvm->lock);
2800 kvm->created_vcpus++;
2801 mutex_unlock(&kvm->lock);
2803 r = kvm_arch_vcpu_precreate(kvm, id);
2805 goto vcpu_decrement;
2807 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
2810 goto vcpu_decrement;
2813 BUILD_BUG_ON(sizeof(struct kvm_run) > PAGE_SIZE);
2814 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2819 vcpu->run = page_address(page);
2821 kvm_vcpu_init(vcpu, kvm, id);
2823 r = kvm_arch_vcpu_create(vcpu);
2825 goto vcpu_free_run_page;
2827 kvm_create_vcpu_debugfs(vcpu);
2829 mutex_lock(&kvm->lock);
2830 if (kvm_get_vcpu_by_id(kvm, id)) {
2832 goto unlock_vcpu_destroy;
2835 vcpu->vcpu_idx = atomic_read(&kvm->online_vcpus);
2836 BUG_ON(kvm->vcpus[vcpu->vcpu_idx]);
2838 /* Now it's all set up, let userspace reach it */
2840 r = create_vcpu_fd(vcpu);
2842 kvm_put_kvm_no_destroy(kvm);
2843 goto unlock_vcpu_destroy;
2846 kvm->vcpus[vcpu->vcpu_idx] = vcpu;
2849 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
2850 * before kvm->online_vcpu's incremented value.
2853 atomic_inc(&kvm->online_vcpus);
2855 mutex_unlock(&kvm->lock);
2856 kvm_arch_vcpu_postcreate(vcpu);
2859 unlock_vcpu_destroy:
2860 mutex_unlock(&kvm->lock);
2861 debugfs_remove_recursive(vcpu->debugfs_dentry);
2862 kvm_arch_vcpu_destroy(vcpu);
2864 free_page((unsigned long)vcpu->run);
2866 kmem_cache_free(kvm_vcpu_cache, vcpu);
2868 mutex_lock(&kvm->lock);
2869 kvm->created_vcpus--;
2870 mutex_unlock(&kvm->lock);
2874 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2877 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2878 vcpu->sigset_active = 1;
2879 vcpu->sigset = *sigset;
2881 vcpu->sigset_active = 0;
2885 static long kvm_vcpu_ioctl(struct file *filp,
2886 unsigned int ioctl, unsigned long arg)
2888 struct kvm_vcpu *vcpu = filp->private_data;
2889 void __user *argp = (void __user *)arg;
2891 struct kvm_fpu *fpu = NULL;
2892 struct kvm_sregs *kvm_sregs = NULL;
2894 if (vcpu->kvm->mm != current->mm)
2897 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2901 * Some architectures have vcpu ioctls that are asynchronous to vcpu
2902 * execution; mutex_lock() would break them.
2904 r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg);
2905 if (r != -ENOIOCTLCMD)
2908 if (mutex_lock_killable(&vcpu->mutex))
2916 oldpid = rcu_access_pointer(vcpu->pid);
2917 if (unlikely(oldpid != task_pid(current))) {
2918 /* The thread running this VCPU changed. */
2921 r = kvm_arch_vcpu_run_pid_change(vcpu);
2925 newpid = get_task_pid(current, PIDTYPE_PID);
2926 rcu_assign_pointer(vcpu->pid, newpid);
2931 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2932 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2935 case KVM_GET_REGS: {
2936 struct kvm_regs *kvm_regs;
2939 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL_ACCOUNT);
2942 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2946 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2953 case KVM_SET_REGS: {
2954 struct kvm_regs *kvm_regs;
2957 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2958 if (IS_ERR(kvm_regs)) {
2959 r = PTR_ERR(kvm_regs);
2962 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2966 case KVM_GET_SREGS: {
2967 kvm_sregs = kzalloc(sizeof(struct kvm_sregs),
2968 GFP_KERNEL_ACCOUNT);
2972 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2976 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2981 case KVM_SET_SREGS: {
2982 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2983 if (IS_ERR(kvm_sregs)) {
2984 r = PTR_ERR(kvm_sregs);
2988 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2991 case KVM_GET_MP_STATE: {
2992 struct kvm_mp_state mp_state;
2994 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2998 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
3003 case KVM_SET_MP_STATE: {
3004 struct kvm_mp_state mp_state;
3007 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
3009 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
3012 case KVM_TRANSLATE: {
3013 struct kvm_translation tr;
3016 if (copy_from_user(&tr, argp, sizeof(tr)))
3018 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
3022 if (copy_to_user(argp, &tr, sizeof(tr)))
3027 case KVM_SET_GUEST_DEBUG: {
3028 struct kvm_guest_debug dbg;
3031 if (copy_from_user(&dbg, argp, sizeof(dbg)))
3033 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
3036 case KVM_SET_SIGNAL_MASK: {
3037 struct kvm_signal_mask __user *sigmask_arg = argp;
3038 struct kvm_signal_mask kvm_sigmask;
3039 sigset_t sigset, *p;
3044 if (copy_from_user(&kvm_sigmask, argp,
3045 sizeof(kvm_sigmask)))
3048 if (kvm_sigmask.len != sizeof(sigset))
3051 if (copy_from_user(&sigset, sigmask_arg->sigset,
3056 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
3060 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL_ACCOUNT);
3064 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
3068 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
3074 fpu = memdup_user(argp, sizeof(*fpu));
3080 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
3084 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
3087 mutex_unlock(&vcpu->mutex);
3093 #ifdef CONFIG_KVM_COMPAT
3094 static long kvm_vcpu_compat_ioctl(struct file *filp,
3095 unsigned int ioctl, unsigned long arg)
3097 struct kvm_vcpu *vcpu = filp->private_data;
3098 void __user *argp = compat_ptr(arg);
3101 if (vcpu->kvm->mm != current->mm)
3105 case KVM_SET_SIGNAL_MASK: {
3106 struct kvm_signal_mask __user *sigmask_arg = argp;
3107 struct kvm_signal_mask kvm_sigmask;
3112 if (copy_from_user(&kvm_sigmask, argp,
3113 sizeof(kvm_sigmask)))
3116 if (kvm_sigmask.len != sizeof(compat_sigset_t))
3119 if (get_compat_sigset(&sigset, (void *)sigmask_arg->sigset))
3121 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
3123 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
3127 r = kvm_vcpu_ioctl(filp, ioctl, arg);
3135 static int kvm_device_mmap(struct file *filp, struct vm_area_struct *vma)
3137 struct kvm_device *dev = filp->private_data;
3140 return dev->ops->mmap(dev, vma);
3145 static int kvm_device_ioctl_attr(struct kvm_device *dev,
3146 int (*accessor)(struct kvm_device *dev,
3147 struct kvm_device_attr *attr),
3150 struct kvm_device_attr attr;
3155 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
3158 return accessor(dev, &attr);
3161 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
3164 struct kvm_device *dev = filp->private_data;
3166 if (dev->kvm->mm != current->mm)
3170 case KVM_SET_DEVICE_ATTR:
3171 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
3172 case KVM_GET_DEVICE_ATTR:
3173 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
3174 case KVM_HAS_DEVICE_ATTR:
3175 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
3177 if (dev->ops->ioctl)
3178 return dev->ops->ioctl(dev, ioctl, arg);
3184 static int kvm_device_release(struct inode *inode, struct file *filp)
3186 struct kvm_device *dev = filp->private_data;
3187 struct kvm *kvm = dev->kvm;
3189 if (dev->ops->release) {
3190 mutex_lock(&kvm->lock);
3191 list_del(&dev->vm_node);
3192 dev->ops->release(dev);
3193 mutex_unlock(&kvm->lock);
3200 static const struct file_operations kvm_device_fops = {
3201 .unlocked_ioctl = kvm_device_ioctl,
3202 .release = kvm_device_release,
3203 KVM_COMPAT(kvm_device_ioctl),
3204 .mmap = kvm_device_mmap,
3207 struct kvm_device *kvm_device_from_filp(struct file *filp)
3209 if (filp->f_op != &kvm_device_fops)
3212 return filp->private_data;
3215 static const struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
3216 #ifdef CONFIG_KVM_MPIC
3217 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
3218 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
3222 int kvm_register_device_ops(const struct kvm_device_ops *ops, u32 type)
3224 if (type >= ARRAY_SIZE(kvm_device_ops_table))
3227 if (kvm_device_ops_table[type] != NULL)
3230 kvm_device_ops_table[type] = ops;
3234 void kvm_unregister_device_ops(u32 type)
3236 if (kvm_device_ops_table[type] != NULL)
3237 kvm_device_ops_table[type] = NULL;
3240 static int kvm_ioctl_create_device(struct kvm *kvm,
3241 struct kvm_create_device *cd)
3243 const struct kvm_device_ops *ops = NULL;
3244 struct kvm_device *dev;
3245 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
3249 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
3252 type = array_index_nospec(cd->type, ARRAY_SIZE(kvm_device_ops_table));
3253 ops = kvm_device_ops_table[type];
3260 dev = kzalloc(sizeof(*dev), GFP_KERNEL_ACCOUNT);
3267 mutex_lock(&kvm->lock);
3268 ret = ops->create(dev, type);
3270 mutex_unlock(&kvm->lock);
3274 list_add(&dev->vm_node, &kvm->devices);
3275 mutex_unlock(&kvm->lock);
3281 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
3283 kvm_put_kvm_no_destroy(kvm);
3284 mutex_lock(&kvm->lock);
3285 list_del(&dev->vm_node);
3286 mutex_unlock(&kvm->lock);
3295 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
3298 case KVM_CAP_USER_MEMORY:
3299 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
3300 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
3301 case KVM_CAP_INTERNAL_ERROR_DATA:
3302 #ifdef CONFIG_HAVE_KVM_MSI
3303 case KVM_CAP_SIGNAL_MSI:
3305 #ifdef CONFIG_HAVE_KVM_IRQFD
3307 case KVM_CAP_IRQFD_RESAMPLE:
3309 case KVM_CAP_IOEVENTFD_ANY_LENGTH:
3310 case KVM_CAP_CHECK_EXTENSION_VM:
3311 case KVM_CAP_ENABLE_CAP_VM:
3312 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3313 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2:
3316 #ifdef CONFIG_KVM_MMIO
3317 case KVM_CAP_COALESCED_MMIO:
3318 return KVM_COALESCED_MMIO_PAGE_OFFSET;
3319 case KVM_CAP_COALESCED_PIO:
3322 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3323 case KVM_CAP_IRQ_ROUTING:
3324 return KVM_MAX_IRQ_ROUTES;
3326 #if KVM_ADDRESS_SPACE_NUM > 1
3327 case KVM_CAP_MULTI_ADDRESS_SPACE:
3328 return KVM_ADDRESS_SPACE_NUM;
3330 case KVM_CAP_NR_MEMSLOTS:
3331 return KVM_USER_MEM_SLOTS;
3335 return kvm_vm_ioctl_check_extension(kvm, arg);
3338 int __attribute__((weak)) kvm_vm_ioctl_enable_cap(struct kvm *kvm,
3339 struct kvm_enable_cap *cap)
3344 static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm,
3345 struct kvm_enable_cap *cap)
3348 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3349 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2:
3350 if (cap->flags || (cap->args[0] & ~1))
3352 kvm->manual_dirty_log_protect = cap->args[0];
3356 return kvm_vm_ioctl_enable_cap(kvm, cap);
3360 static long kvm_vm_ioctl(struct file *filp,
3361 unsigned int ioctl, unsigned long arg)
3363 struct kvm *kvm = filp->private_data;
3364 void __user *argp = (void __user *)arg;
3367 if (kvm->mm != current->mm)
3370 case KVM_CREATE_VCPU:
3371 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
3373 case KVM_ENABLE_CAP: {
3374 struct kvm_enable_cap cap;
3377 if (copy_from_user(&cap, argp, sizeof(cap)))
3379 r = kvm_vm_ioctl_enable_cap_generic(kvm, &cap);
3382 case KVM_SET_USER_MEMORY_REGION: {
3383 struct kvm_userspace_memory_region kvm_userspace_mem;
3386 if (copy_from_user(&kvm_userspace_mem, argp,
3387 sizeof(kvm_userspace_mem)))
3390 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
3393 case KVM_GET_DIRTY_LOG: {
3394 struct kvm_dirty_log log;
3397 if (copy_from_user(&log, argp, sizeof(log)))
3399 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3402 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3403 case KVM_CLEAR_DIRTY_LOG: {
3404 struct kvm_clear_dirty_log log;
3407 if (copy_from_user(&log, argp, sizeof(log)))
3409 r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
3413 #ifdef CONFIG_KVM_MMIO
3414 case KVM_REGISTER_COALESCED_MMIO: {
3415 struct kvm_coalesced_mmio_zone zone;
3418 if (copy_from_user(&zone, argp, sizeof(zone)))
3420 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
3423 case KVM_UNREGISTER_COALESCED_MMIO: {
3424 struct kvm_coalesced_mmio_zone zone;
3427 if (copy_from_user(&zone, argp, sizeof(zone)))
3429 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
3434 struct kvm_irqfd data;
3437 if (copy_from_user(&data, argp, sizeof(data)))
3439 r = kvm_irqfd(kvm, &data);
3442 case KVM_IOEVENTFD: {
3443 struct kvm_ioeventfd data;
3446 if (copy_from_user(&data, argp, sizeof(data)))
3448 r = kvm_ioeventfd(kvm, &data);
3451 #ifdef CONFIG_HAVE_KVM_MSI
3452 case KVM_SIGNAL_MSI: {
3456 if (copy_from_user(&msi, argp, sizeof(msi)))
3458 r = kvm_send_userspace_msi(kvm, &msi);
3462 #ifdef __KVM_HAVE_IRQ_LINE
3463 case KVM_IRQ_LINE_STATUS:
3464 case KVM_IRQ_LINE: {
3465 struct kvm_irq_level irq_event;
3468 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
3471 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
3472 ioctl == KVM_IRQ_LINE_STATUS);
3477 if (ioctl == KVM_IRQ_LINE_STATUS) {
3478 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
3486 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3487 case KVM_SET_GSI_ROUTING: {
3488 struct kvm_irq_routing routing;
3489 struct kvm_irq_routing __user *urouting;
3490 struct kvm_irq_routing_entry *entries = NULL;
3493 if (copy_from_user(&routing, argp, sizeof(routing)))
3496 if (!kvm_arch_can_set_irq_routing(kvm))
3498 if (routing.nr > KVM_MAX_IRQ_ROUTES)
3504 entries = vmalloc(array_size(sizeof(*entries),
3510 if (copy_from_user(entries, urouting->entries,
3511 routing.nr * sizeof(*entries)))
3512 goto out_free_irq_routing;
3514 r = kvm_set_irq_routing(kvm, entries, routing.nr,
3516 out_free_irq_routing:
3520 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
3521 case KVM_CREATE_DEVICE: {
3522 struct kvm_create_device cd;
3525 if (copy_from_user(&cd, argp, sizeof(cd)))
3528 r = kvm_ioctl_create_device(kvm, &cd);
3533 if (copy_to_user(argp, &cd, sizeof(cd)))
3539 case KVM_CHECK_EXTENSION:
3540 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
3543 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
3549 #ifdef CONFIG_KVM_COMPAT
3550 struct compat_kvm_dirty_log {
3554 compat_uptr_t dirty_bitmap; /* one bit per page */
3559 static long kvm_vm_compat_ioctl(struct file *filp,
3560 unsigned int ioctl, unsigned long arg)
3562 struct kvm *kvm = filp->private_data;
3565 if (kvm->mm != current->mm)
3568 case KVM_GET_DIRTY_LOG: {
3569 struct compat_kvm_dirty_log compat_log;
3570 struct kvm_dirty_log log;
3572 if (copy_from_user(&compat_log, (void __user *)arg,
3573 sizeof(compat_log)))
3575 log.slot = compat_log.slot;
3576 log.padding1 = compat_log.padding1;
3577 log.padding2 = compat_log.padding2;
3578 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
3580 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3584 r = kvm_vm_ioctl(filp, ioctl, arg);
3590 static struct file_operations kvm_vm_fops = {
3591 .release = kvm_vm_release,
3592 .unlocked_ioctl = kvm_vm_ioctl,
3593 .llseek = noop_llseek,
3594 KVM_COMPAT(kvm_vm_compat_ioctl),
3597 static int kvm_dev_ioctl_create_vm(unsigned long type)
3603 kvm = kvm_create_vm(type);
3605 return PTR_ERR(kvm);
3606 #ifdef CONFIG_KVM_MMIO
3607 r = kvm_coalesced_mmio_init(kvm);
3611 r = get_unused_fd_flags(O_CLOEXEC);
3615 file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
3623 * Don't call kvm_put_kvm anymore at this point; file->f_op is
3624 * already set, with ->release() being kvm_vm_release(). In error
3625 * cases it will be called by the final fput(file) and will take
3626 * care of doing kvm_put_kvm(kvm).
3628 if (kvm_create_vm_debugfs(kvm, r) < 0) {
3633 kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm);
3635 fd_install(r, file);
3643 static long kvm_dev_ioctl(struct file *filp,
3644 unsigned int ioctl, unsigned long arg)
3649 case KVM_GET_API_VERSION:
3652 r = KVM_API_VERSION;
3655 r = kvm_dev_ioctl_create_vm(arg);
3657 case KVM_CHECK_EXTENSION:
3658 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
3660 case KVM_GET_VCPU_MMAP_SIZE:
3663 r = PAGE_SIZE; /* struct kvm_run */
3665 r += PAGE_SIZE; /* pio data page */
3667 #ifdef CONFIG_KVM_MMIO
3668 r += PAGE_SIZE; /* coalesced mmio ring page */
3671 case KVM_TRACE_ENABLE:
3672 case KVM_TRACE_PAUSE:
3673 case KVM_TRACE_DISABLE:
3677 return kvm_arch_dev_ioctl(filp, ioctl, arg);
3683 static struct file_operations kvm_chardev_ops = {
3684 .unlocked_ioctl = kvm_dev_ioctl,
3685 .llseek = noop_llseek,
3686 KVM_COMPAT(kvm_dev_ioctl),
3689 static struct miscdevice kvm_dev = {
3695 static void hardware_enable_nolock(void *junk)
3697 int cpu = raw_smp_processor_id();
3700 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
3703 cpumask_set_cpu(cpu, cpus_hardware_enabled);
3705 r = kvm_arch_hardware_enable();
3708 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3709 atomic_inc(&hardware_enable_failed);
3710 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
3714 static int kvm_starting_cpu(unsigned int cpu)
3716 raw_spin_lock(&kvm_count_lock);
3717 if (kvm_usage_count)
3718 hardware_enable_nolock(NULL);
3719 raw_spin_unlock(&kvm_count_lock);
3723 static void hardware_disable_nolock(void *junk)
3725 int cpu = raw_smp_processor_id();
3727 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
3729 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3730 kvm_arch_hardware_disable();
3733 static int kvm_dying_cpu(unsigned int cpu)
3735 raw_spin_lock(&kvm_count_lock);
3736 if (kvm_usage_count)
3737 hardware_disable_nolock(NULL);
3738 raw_spin_unlock(&kvm_count_lock);
3742 static void hardware_disable_all_nolock(void)
3744 BUG_ON(!kvm_usage_count);
3747 if (!kvm_usage_count)
3748 on_each_cpu(hardware_disable_nolock, NULL, 1);
3751 static void hardware_disable_all(void)
3753 raw_spin_lock(&kvm_count_lock);
3754 hardware_disable_all_nolock();
3755 raw_spin_unlock(&kvm_count_lock);
3758 static int hardware_enable_all(void)
3762 raw_spin_lock(&kvm_count_lock);
3765 if (kvm_usage_count == 1) {
3766 atomic_set(&hardware_enable_failed, 0);
3767 on_each_cpu(hardware_enable_nolock, NULL, 1);
3769 if (atomic_read(&hardware_enable_failed)) {
3770 hardware_disable_all_nolock();
3775 raw_spin_unlock(&kvm_count_lock);
3780 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3784 * Some (well, at least mine) BIOSes hang on reboot if
3787 * And Intel TXT required VMX off for all cpu when system shutdown.
3789 pr_info("kvm: exiting hardware virtualization\n");
3790 kvm_rebooting = true;
3791 on_each_cpu(hardware_disable_nolock, NULL, 1);
3795 static struct notifier_block kvm_reboot_notifier = {
3796 .notifier_call = kvm_reboot,
3800 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3804 for (i = 0; i < bus->dev_count; i++) {
3805 struct kvm_io_device *pos = bus->range[i].dev;
3807 kvm_iodevice_destructor(pos);
3812 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
3813 const struct kvm_io_range *r2)
3815 gpa_t addr1 = r1->addr;
3816 gpa_t addr2 = r2->addr;
3821 /* If r2->len == 0, match the exact address. If r2->len != 0,
3822 * accept any overlapping write. Any order is acceptable for
3823 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
3824 * we process all of them.
3837 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
3839 return kvm_io_bus_cmp(p1, p2);
3842 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
3843 gpa_t addr, int len)
3845 struct kvm_io_range *range, key;
3848 key = (struct kvm_io_range) {
3853 range = bsearch(&key, bus->range, bus->dev_count,
3854 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
3858 off = range - bus->range;
3860 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
3866 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3867 struct kvm_io_range *range, const void *val)
3871 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3875 while (idx < bus->dev_count &&
3876 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3877 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
3886 /* kvm_io_bus_write - called under kvm->slots_lock */
3887 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3888 int len, const void *val)
3890 struct kvm_io_bus *bus;
3891 struct kvm_io_range range;
3894 range = (struct kvm_io_range) {
3899 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3902 r = __kvm_io_bus_write(vcpu, bus, &range, val);
3903 return r < 0 ? r : 0;
3905 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
3907 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3908 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
3909 gpa_t addr, int len, const void *val, long cookie)
3911 struct kvm_io_bus *bus;
3912 struct kvm_io_range range;
3914 range = (struct kvm_io_range) {
3919 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3923 /* First try the device referenced by cookie. */
3924 if ((cookie >= 0) && (cookie < bus->dev_count) &&
3925 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
3926 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
3931 * cookie contained garbage; fall back to search and return the
3932 * correct cookie value.
3934 return __kvm_io_bus_write(vcpu, bus, &range, val);
3937 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3938 struct kvm_io_range *range, void *val)
3942 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3946 while (idx < bus->dev_count &&
3947 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3948 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
3957 /* kvm_io_bus_read - called under kvm->slots_lock */
3958 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3961 struct kvm_io_bus *bus;
3962 struct kvm_io_range range;
3965 range = (struct kvm_io_range) {
3970 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3973 r = __kvm_io_bus_read(vcpu, bus, &range, val);
3974 return r < 0 ? r : 0;
3977 /* Caller must hold slots_lock. */
3978 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3979 int len, struct kvm_io_device *dev)
3982 struct kvm_io_bus *new_bus, *bus;
3983 struct kvm_io_range range;
3985 bus = kvm_get_bus(kvm, bus_idx);
3989 /* exclude ioeventfd which is limited by maximum fd */
3990 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3993 new_bus = kmalloc(struct_size(bus, range, bus->dev_count + 1),
3994 GFP_KERNEL_ACCOUNT);
3998 range = (struct kvm_io_range) {
4004 for (i = 0; i < bus->dev_count; i++)
4005 if (kvm_io_bus_cmp(&bus->range[i], &range) > 0)
4008 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
4009 new_bus->dev_count++;
4010 new_bus->range[i] = range;
4011 memcpy(new_bus->range + i + 1, bus->range + i,
4012 (bus->dev_count - i) * sizeof(struct kvm_io_range));
4013 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
4014 synchronize_srcu_expedited(&kvm->srcu);
4020 /* Caller must hold slots_lock. */
4021 void kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
4022 struct kvm_io_device *dev)
4025 struct kvm_io_bus *new_bus, *bus;
4027 bus = kvm_get_bus(kvm, bus_idx);
4031 for (i = 0; i < bus->dev_count; i++)
4032 if (bus->range[i].dev == dev) {
4036 if (i == bus->dev_count)
4039 new_bus = kmalloc(struct_size(bus, range, bus->dev_count - 1),
4040 GFP_KERNEL_ACCOUNT);
4042 pr_err("kvm: failed to shrink bus, removing it completely\n");
4046 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
4047 new_bus->dev_count--;
4048 memcpy(new_bus->range + i, bus->range + i + 1,
4049 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
4052 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
4053 synchronize_srcu_expedited(&kvm->srcu);
4058 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
4061 struct kvm_io_bus *bus;
4062 int dev_idx, srcu_idx;
4063 struct kvm_io_device *iodev = NULL;
4065 srcu_idx = srcu_read_lock(&kvm->srcu);
4067 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
4071 dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
4075 iodev = bus->range[dev_idx].dev;
4078 srcu_read_unlock(&kvm->srcu, srcu_idx);
4082 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
4084 static int kvm_debugfs_open(struct inode *inode, struct file *file,
4085 int (*get)(void *, u64 *), int (*set)(void *, u64),
4088 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
4091 /* The debugfs files are a reference to the kvm struct which
4092 * is still valid when kvm_destroy_vm is called.
4093 * To avoid the race between open and the removal of the debugfs
4094 * directory we test against the users count.
4096 if (!refcount_inc_not_zero(&stat_data->kvm->users_count))
4099 if (simple_attr_open(inode, file, get,
4100 KVM_DBGFS_GET_MODE(stat_data->dbgfs_item) & 0222
4103 kvm_put_kvm(stat_data->kvm);
4110 static int kvm_debugfs_release(struct inode *inode, struct file *file)
4112 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
4115 simple_attr_release(inode, file);
4116 kvm_put_kvm(stat_data->kvm);
4121 static int kvm_get_stat_per_vm(struct kvm *kvm, size_t offset, u64 *val)
4123 *val = *(ulong *)((void *)kvm + offset);
4128 static int kvm_clear_stat_per_vm(struct kvm *kvm, size_t offset)
4130 *(ulong *)((void *)kvm + offset) = 0;
4135 static int kvm_get_stat_per_vcpu(struct kvm *kvm, size_t offset, u64 *val)
4138 struct kvm_vcpu *vcpu;
4142 kvm_for_each_vcpu(i, vcpu, kvm)
4143 *val += *(u64 *)((void *)vcpu + offset);
4148 static int kvm_clear_stat_per_vcpu(struct kvm *kvm, size_t offset)
4151 struct kvm_vcpu *vcpu;
4153 kvm_for_each_vcpu(i, vcpu, kvm)
4154 *(u64 *)((void *)vcpu + offset) = 0;
4159 static int kvm_stat_data_get(void *data, u64 *val)
4162 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
4164 switch (stat_data->dbgfs_item->kind) {
4166 r = kvm_get_stat_per_vm(stat_data->kvm,
4167 stat_data->dbgfs_item->offset, val);
4170 r = kvm_get_stat_per_vcpu(stat_data->kvm,
4171 stat_data->dbgfs_item->offset, val);
4178 static int kvm_stat_data_clear(void *data, u64 val)
4181 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
4186 switch (stat_data->dbgfs_item->kind) {
4188 r = kvm_clear_stat_per_vm(stat_data->kvm,
4189 stat_data->dbgfs_item->offset);
4192 r = kvm_clear_stat_per_vcpu(stat_data->kvm,
4193 stat_data->dbgfs_item->offset);
4200 static int kvm_stat_data_open(struct inode *inode, struct file *file)
4202 __simple_attr_check_format("%llu\n", 0ull);
4203 return kvm_debugfs_open(inode, file, kvm_stat_data_get,
4204 kvm_stat_data_clear, "%llu\n");
4207 static const struct file_operations stat_fops_per_vm = {
4208 .owner = THIS_MODULE,
4209 .open = kvm_stat_data_open,
4210 .release = kvm_debugfs_release,
4211 .read = simple_attr_read,
4212 .write = simple_attr_write,
4213 .llseek = no_llseek,
4216 static int vm_stat_get(void *_offset, u64 *val)
4218 unsigned offset = (long)_offset;
4223 mutex_lock(&kvm_lock);
4224 list_for_each_entry(kvm, &vm_list, vm_list) {
4225 kvm_get_stat_per_vm(kvm, offset, &tmp_val);
4228 mutex_unlock(&kvm_lock);
4232 static int vm_stat_clear(void *_offset, u64 val)
4234 unsigned offset = (long)_offset;
4240 mutex_lock(&kvm_lock);
4241 list_for_each_entry(kvm, &vm_list, vm_list) {
4242 kvm_clear_stat_per_vm(kvm, offset);
4244 mutex_unlock(&kvm_lock);
4249 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n");
4251 static int vcpu_stat_get(void *_offset, u64 *val)
4253 unsigned offset = (long)_offset;
4258 mutex_lock(&kvm_lock);
4259 list_for_each_entry(kvm, &vm_list, vm_list) {
4260 kvm_get_stat_per_vcpu(kvm, offset, &tmp_val);
4263 mutex_unlock(&kvm_lock);
4267 static int vcpu_stat_clear(void *_offset, u64 val)
4269 unsigned offset = (long)_offset;
4275 mutex_lock(&kvm_lock);
4276 list_for_each_entry(kvm, &vm_list, vm_list) {
4277 kvm_clear_stat_per_vcpu(kvm, offset);
4279 mutex_unlock(&kvm_lock);
4284 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear,
4287 static const struct file_operations *stat_fops[] = {
4288 [KVM_STAT_VCPU] = &vcpu_stat_fops,
4289 [KVM_STAT_VM] = &vm_stat_fops,
4292 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm)
4294 struct kobj_uevent_env *env;
4295 unsigned long long created, active;
4297 if (!kvm_dev.this_device || !kvm)
4300 mutex_lock(&kvm_lock);
4301 if (type == KVM_EVENT_CREATE_VM) {
4302 kvm_createvm_count++;
4304 } else if (type == KVM_EVENT_DESTROY_VM) {
4307 created = kvm_createvm_count;
4308 active = kvm_active_vms;
4309 mutex_unlock(&kvm_lock);
4311 env = kzalloc(sizeof(*env), GFP_KERNEL_ACCOUNT);
4315 add_uevent_var(env, "CREATED=%llu", created);
4316 add_uevent_var(env, "COUNT=%llu", active);
4318 if (type == KVM_EVENT_CREATE_VM) {
4319 add_uevent_var(env, "EVENT=create");
4320 kvm->userspace_pid = task_pid_nr(current);
4321 } else if (type == KVM_EVENT_DESTROY_VM) {
4322 add_uevent_var(env, "EVENT=destroy");
4324 add_uevent_var(env, "PID=%d", kvm->userspace_pid);
4326 if (!IS_ERR_OR_NULL(kvm->debugfs_dentry)) {
4327 char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL_ACCOUNT);
4330 tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX);
4332 add_uevent_var(env, "STATS_PATH=%s", tmp);
4336 /* no need for checks, since we are adding at most only 5 keys */
4337 env->envp[env->envp_idx++] = NULL;
4338 kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp);
4342 static void kvm_init_debug(void)
4344 struct kvm_stats_debugfs_item *p;
4346 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
4348 kvm_debugfs_num_entries = 0;
4349 for (p = debugfs_entries; p->name; ++p, kvm_debugfs_num_entries++) {
4350 debugfs_create_file(p->name, KVM_DBGFS_GET_MODE(p),
4351 kvm_debugfs_dir, (void *)(long)p->offset,
4352 stat_fops[p->kind]);
4356 static int kvm_suspend(void)
4358 if (kvm_usage_count)
4359 hardware_disable_nolock(NULL);
4363 static void kvm_resume(void)
4365 if (kvm_usage_count) {
4366 #ifdef CONFIG_LOCKDEP
4367 WARN_ON(lockdep_is_held(&kvm_count_lock));
4369 hardware_enable_nolock(NULL);
4373 static struct syscore_ops kvm_syscore_ops = {
4374 .suspend = kvm_suspend,
4375 .resume = kvm_resume,
4379 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
4381 return container_of(pn, struct kvm_vcpu, preempt_notifier);
4384 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
4386 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
4388 WRITE_ONCE(vcpu->preempted, false);
4389 WRITE_ONCE(vcpu->ready, false);
4391 __this_cpu_write(kvm_running_vcpu, vcpu);
4392 kvm_arch_sched_in(vcpu, cpu);
4393 kvm_arch_vcpu_load(vcpu, cpu);
4396 static void kvm_sched_out(struct preempt_notifier *pn,
4397 struct task_struct *next)
4399 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
4401 if (current->state == TASK_RUNNING) {
4402 WRITE_ONCE(vcpu->preempted, true);
4403 WRITE_ONCE(vcpu->ready, true);
4405 kvm_arch_vcpu_put(vcpu);
4406 __this_cpu_write(kvm_running_vcpu, NULL);
4410 * kvm_get_running_vcpu - get the vcpu running on the current CPU.
4411 * Thanks to preempt notifiers, this can also be called from
4412 * preemptible context.
4414 struct kvm_vcpu *kvm_get_running_vcpu(void)
4416 return __this_cpu_read(kvm_running_vcpu);
4420 * kvm_get_running_vcpus - get the per-CPU array of currently running vcpus.
4422 struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void)
4424 return &kvm_running_vcpu;
4427 static void check_processor_compat(void *rtn)
4429 *(int *)rtn = kvm_arch_check_processor_compat();
4432 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
4433 struct module *module)
4438 r = kvm_arch_init(opaque);
4443 * kvm_arch_init makes sure there's at most one caller
4444 * for architectures that support multiple implementations,
4445 * like intel and amd on x86.
4446 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
4447 * conflicts in case kvm is already setup for another implementation.
4449 r = kvm_irqfd_init();
4453 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
4458 r = kvm_arch_hardware_setup();
4462 for_each_online_cpu(cpu) {
4463 smp_call_function_single(cpu, check_processor_compat, &r, 1);
4468 r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting",
4469 kvm_starting_cpu, kvm_dying_cpu);
4472 register_reboot_notifier(&kvm_reboot_notifier);
4474 /* A kmem cache lets us meet the alignment requirements of fx_save. */
4476 vcpu_align = __alignof__(struct kvm_vcpu);
4478 kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align,
4480 offsetof(struct kvm_vcpu, arch),
4481 sizeof_field(struct kvm_vcpu, arch),
4483 if (!kvm_vcpu_cache) {
4488 r = kvm_async_pf_init();
4492 kvm_chardev_ops.owner = module;
4493 kvm_vm_fops.owner = module;
4494 kvm_vcpu_fops.owner = module;
4496 r = misc_register(&kvm_dev);
4498 pr_err("kvm: misc device register failed\n");
4502 register_syscore_ops(&kvm_syscore_ops);
4504 kvm_preempt_ops.sched_in = kvm_sched_in;
4505 kvm_preempt_ops.sched_out = kvm_sched_out;
4509 r = kvm_vfio_ops_init();
4515 kvm_async_pf_deinit();
4517 kmem_cache_destroy(kvm_vcpu_cache);
4519 unregister_reboot_notifier(&kvm_reboot_notifier);
4520 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4522 kvm_arch_hardware_unsetup();
4524 free_cpumask_var(cpus_hardware_enabled);
4532 EXPORT_SYMBOL_GPL(kvm_init);
4536 debugfs_remove_recursive(kvm_debugfs_dir);
4537 misc_deregister(&kvm_dev);
4538 kmem_cache_destroy(kvm_vcpu_cache);
4539 kvm_async_pf_deinit();
4540 unregister_syscore_ops(&kvm_syscore_ops);
4541 unregister_reboot_notifier(&kvm_reboot_notifier);
4542 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4543 on_each_cpu(hardware_disable_nolock, NULL, 1);
4544 kvm_arch_hardware_unsetup();
4547 free_cpumask_var(cpus_hardware_enabled);
4548 kvm_vfio_ops_exit();
4550 EXPORT_SYMBOL_GPL(kvm_exit);
4552 struct kvm_vm_worker_thread_context {
4554 struct task_struct *parent;
4555 struct completion init_done;
4556 kvm_vm_thread_fn_t thread_fn;
4561 static int kvm_vm_worker_thread(void *context)
4564 * The init_context is allocated on the stack of the parent thread, so
4565 * we have to locally copy anything that is needed beyond initialization
4567 struct kvm_vm_worker_thread_context *init_context = context;
4568 struct kvm *kvm = init_context->kvm;
4569 kvm_vm_thread_fn_t thread_fn = init_context->thread_fn;
4570 uintptr_t data = init_context->data;
4573 err = kthread_park(current);
4574 /* kthread_park(current) is never supposed to return an error */
4579 err = cgroup_attach_task_all(init_context->parent, current);
4581 kvm_err("%s: cgroup_attach_task_all failed with err %d\n",
4586 set_user_nice(current, task_nice(init_context->parent));
4589 init_context->err = err;
4590 complete(&init_context->init_done);
4591 init_context = NULL;
4596 /* Wait to be woken up by the spawner before proceeding. */
4599 if (!kthread_should_stop())
4600 err = thread_fn(kvm, data);
4605 int kvm_vm_create_worker_thread(struct kvm *kvm, kvm_vm_thread_fn_t thread_fn,
4606 uintptr_t data, const char *name,
4607 struct task_struct **thread_ptr)
4609 struct kvm_vm_worker_thread_context init_context = {};
4610 struct task_struct *thread;
4613 init_context.kvm = kvm;
4614 init_context.parent = current;
4615 init_context.thread_fn = thread_fn;
4616 init_context.data = data;
4617 init_completion(&init_context.init_done);
4619 thread = kthread_run(kvm_vm_worker_thread, &init_context,
4620 "%s-%d", name, task_pid_nr(current));
4622 return PTR_ERR(thread);
4624 /* kthread_run is never supposed to return NULL */
4625 WARN_ON(thread == NULL);
4627 wait_for_completion(&init_context.init_done);
4629 if (!init_context.err)
4630 *thread_ptr = thread;
4632 return init_context.err;