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)) &&
190 !kvm_is_zone_device_pfn(pfn);
195 bool kvm_is_transparent_hugepage(kvm_pfn_t pfn)
197 struct page *page = pfn_to_page(pfn);
199 if (!PageTransCompoundMap(page))
202 return is_transparent_hugepage(compound_head(page));
206 * Switches to specified vcpu, until a matching vcpu_put()
208 void vcpu_load(struct kvm_vcpu *vcpu)
212 __this_cpu_write(kvm_running_vcpu, vcpu);
213 preempt_notifier_register(&vcpu->preempt_notifier);
214 kvm_arch_vcpu_load(vcpu, cpu);
217 EXPORT_SYMBOL_GPL(vcpu_load);
219 void vcpu_put(struct kvm_vcpu *vcpu)
222 kvm_arch_vcpu_put(vcpu);
223 preempt_notifier_unregister(&vcpu->preempt_notifier);
224 __this_cpu_write(kvm_running_vcpu, NULL);
227 EXPORT_SYMBOL_GPL(vcpu_put);
229 /* TODO: merge with kvm_arch_vcpu_should_kick */
230 static bool kvm_request_needs_ipi(struct kvm_vcpu *vcpu, unsigned req)
232 int mode = kvm_vcpu_exiting_guest_mode(vcpu);
235 * We need to wait for the VCPU to reenable interrupts and get out of
236 * READING_SHADOW_PAGE_TABLES mode.
238 if (req & KVM_REQUEST_WAIT)
239 return mode != OUTSIDE_GUEST_MODE;
242 * Need to kick a running VCPU, but otherwise there is nothing to do.
244 return mode == IN_GUEST_MODE;
247 static void ack_flush(void *_completed)
251 static inline bool kvm_kick_many_cpus(const struct cpumask *cpus, bool wait)
254 cpus = cpu_online_mask;
256 if (cpumask_empty(cpus))
259 smp_call_function_many(cpus, ack_flush, NULL, wait);
263 bool kvm_make_vcpus_request_mask(struct kvm *kvm, unsigned int req,
264 unsigned long *vcpu_bitmap, cpumask_var_t tmp)
267 struct kvm_vcpu *vcpu;
272 kvm_for_each_vcpu(i, vcpu, kvm) {
273 if (vcpu_bitmap && !test_bit(i, vcpu_bitmap))
276 kvm_make_request(req, vcpu);
279 if (!(req & KVM_REQUEST_NO_WAKEUP) && kvm_vcpu_wake_up(vcpu))
282 if (tmp != NULL && cpu != -1 && cpu != me &&
283 kvm_request_needs_ipi(vcpu, req))
284 __cpumask_set_cpu(cpu, tmp);
287 called = kvm_kick_many_cpus(tmp, !!(req & KVM_REQUEST_WAIT));
293 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
298 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
300 called = kvm_make_vcpus_request_mask(kvm, req, NULL, cpus);
302 free_cpumask_var(cpus);
306 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
307 void kvm_flush_remote_tlbs(struct kvm *kvm)
310 * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in
311 * kvm_make_all_cpus_request.
313 long dirty_count = smp_load_acquire(&kvm->tlbs_dirty);
316 * We want to publish modifications to the page tables before reading
317 * mode. Pairs with a memory barrier in arch-specific code.
318 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
319 * and smp_mb in walk_shadow_page_lockless_begin/end.
320 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
322 * There is already an smp_mb__after_atomic() before
323 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
326 if (!kvm_arch_flush_remote_tlb(kvm)
327 || kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
328 ++kvm->stat.remote_tlb_flush;
329 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
331 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
334 void kvm_reload_remote_mmus(struct kvm *kvm)
336 kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
339 static void kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
341 mutex_init(&vcpu->mutex);
346 init_swait_queue_head(&vcpu->wq);
347 kvm_async_pf_vcpu_init(vcpu);
350 INIT_LIST_HEAD(&vcpu->blocked_vcpu_list);
352 kvm_vcpu_set_in_spin_loop(vcpu, false);
353 kvm_vcpu_set_dy_eligible(vcpu, false);
354 vcpu->preempted = false;
356 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
359 void kvm_vcpu_destroy(struct kvm_vcpu *vcpu)
361 kvm_arch_vcpu_destroy(vcpu);
364 * No need for rcu_read_lock as VCPU_RUN is the only place that changes
365 * the vcpu->pid pointer, and at destruction time all file descriptors
368 put_pid(rcu_dereference_protected(vcpu->pid, 1));
370 free_page((unsigned long)vcpu->run);
371 kmem_cache_free(kvm_vcpu_cache, vcpu);
373 EXPORT_SYMBOL_GPL(kvm_vcpu_destroy);
375 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
376 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
378 return container_of(mn, struct kvm, mmu_notifier);
381 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
382 struct mm_struct *mm,
383 unsigned long address,
386 struct kvm *kvm = mmu_notifier_to_kvm(mn);
389 idx = srcu_read_lock(&kvm->srcu);
390 spin_lock(&kvm->mmu_lock);
391 kvm->mmu_notifier_seq++;
393 if (kvm_set_spte_hva(kvm, address, pte))
394 kvm_flush_remote_tlbs(kvm);
396 spin_unlock(&kvm->mmu_lock);
397 srcu_read_unlock(&kvm->srcu, idx);
400 static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
401 const struct mmu_notifier_range *range)
403 struct kvm *kvm = mmu_notifier_to_kvm(mn);
404 int need_tlb_flush = 0, idx;
407 idx = srcu_read_lock(&kvm->srcu);
408 spin_lock(&kvm->mmu_lock);
410 * The count increase must become visible at unlock time as no
411 * spte can be established without taking the mmu_lock and
412 * count is also read inside the mmu_lock critical section.
414 kvm->mmu_notifier_count++;
415 need_tlb_flush = kvm_unmap_hva_range(kvm, range->start, range->end);
416 need_tlb_flush |= kvm->tlbs_dirty;
417 /* we've to flush the tlb before the pages can be freed */
419 kvm_flush_remote_tlbs(kvm);
421 spin_unlock(&kvm->mmu_lock);
423 ret = kvm_arch_mmu_notifier_invalidate_range(kvm, range->start,
425 mmu_notifier_range_blockable(range));
427 srcu_read_unlock(&kvm->srcu, idx);
432 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
433 const struct mmu_notifier_range *range)
435 struct kvm *kvm = mmu_notifier_to_kvm(mn);
437 spin_lock(&kvm->mmu_lock);
439 * This sequence increase will notify the kvm page fault that
440 * the page that is going to be mapped in the spte could have
443 kvm->mmu_notifier_seq++;
446 * The above sequence increase must be visible before the
447 * below count decrease, which is ensured by the smp_wmb above
448 * in conjunction with the smp_rmb in mmu_notifier_retry().
450 kvm->mmu_notifier_count--;
451 spin_unlock(&kvm->mmu_lock);
453 BUG_ON(kvm->mmu_notifier_count < 0);
456 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
457 struct mm_struct *mm,
461 struct kvm *kvm = mmu_notifier_to_kvm(mn);
464 idx = srcu_read_lock(&kvm->srcu);
465 spin_lock(&kvm->mmu_lock);
467 young = kvm_age_hva(kvm, start, end);
469 kvm_flush_remote_tlbs(kvm);
471 spin_unlock(&kvm->mmu_lock);
472 srcu_read_unlock(&kvm->srcu, idx);
477 static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
478 struct mm_struct *mm,
482 struct kvm *kvm = mmu_notifier_to_kvm(mn);
485 idx = srcu_read_lock(&kvm->srcu);
486 spin_lock(&kvm->mmu_lock);
488 * Even though we do not flush TLB, this will still adversely
489 * affect performance on pre-Haswell Intel EPT, where there is
490 * no EPT Access Bit to clear so that we have to tear down EPT
491 * tables instead. If we find this unacceptable, we can always
492 * add a parameter to kvm_age_hva so that it effectively doesn't
493 * do anything on clear_young.
495 * Also note that currently we never issue secondary TLB flushes
496 * from clear_young, leaving this job up to the regular system
497 * cadence. If we find this inaccurate, we might come up with a
498 * more sophisticated heuristic later.
500 young = kvm_age_hva(kvm, start, end);
501 spin_unlock(&kvm->mmu_lock);
502 srcu_read_unlock(&kvm->srcu, idx);
507 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
508 struct mm_struct *mm,
509 unsigned long address)
511 struct kvm *kvm = mmu_notifier_to_kvm(mn);
514 idx = srcu_read_lock(&kvm->srcu);
515 spin_lock(&kvm->mmu_lock);
516 young = kvm_test_age_hva(kvm, address);
517 spin_unlock(&kvm->mmu_lock);
518 srcu_read_unlock(&kvm->srcu, idx);
523 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
524 struct mm_struct *mm)
526 struct kvm *kvm = mmu_notifier_to_kvm(mn);
529 idx = srcu_read_lock(&kvm->srcu);
530 kvm_arch_flush_shadow_all(kvm);
531 srcu_read_unlock(&kvm->srcu, idx);
534 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
535 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
536 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
537 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
538 .clear_young = kvm_mmu_notifier_clear_young,
539 .test_young = kvm_mmu_notifier_test_young,
540 .change_pte = kvm_mmu_notifier_change_pte,
541 .release = kvm_mmu_notifier_release,
544 static int kvm_init_mmu_notifier(struct kvm *kvm)
546 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
547 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
550 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
552 static int kvm_init_mmu_notifier(struct kvm *kvm)
557 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
559 static struct kvm_memslots *kvm_alloc_memslots(void)
562 struct kvm_memslots *slots;
564 slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL_ACCOUNT);
568 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
569 slots->id_to_index[i] = slots->memslots[i].id = i;
574 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
576 if (!memslot->dirty_bitmap)
579 kvfree(memslot->dirty_bitmap);
580 memslot->dirty_bitmap = NULL;
584 * Free any memory in @free but not in @dont.
586 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
587 struct kvm_memory_slot *dont)
589 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
590 kvm_destroy_dirty_bitmap(free);
592 kvm_arch_free_memslot(kvm, free, dont);
597 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
599 struct kvm_memory_slot *memslot;
604 kvm_for_each_memslot(memslot, slots)
605 kvm_free_memslot(kvm, memslot, NULL);
610 static void kvm_destroy_vm_debugfs(struct kvm *kvm)
614 if (!kvm->debugfs_dentry)
617 debugfs_remove_recursive(kvm->debugfs_dentry);
619 if (kvm->debugfs_stat_data) {
620 for (i = 0; i < kvm_debugfs_num_entries; i++)
621 kfree(kvm->debugfs_stat_data[i]);
622 kfree(kvm->debugfs_stat_data);
626 static int kvm_create_vm_debugfs(struct kvm *kvm, int fd)
628 char dir_name[ITOA_MAX_LEN * 2];
629 struct kvm_stat_data *stat_data;
630 struct kvm_stats_debugfs_item *p;
632 if (!debugfs_initialized())
635 snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd);
636 kvm->debugfs_dentry = debugfs_create_dir(dir_name, kvm_debugfs_dir);
638 kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries,
639 sizeof(*kvm->debugfs_stat_data),
641 if (!kvm->debugfs_stat_data)
644 for (p = debugfs_entries; p->name; p++) {
645 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT);
649 stat_data->kvm = kvm;
650 stat_data->dbgfs_item = p;
651 kvm->debugfs_stat_data[p - debugfs_entries] = stat_data;
652 debugfs_create_file(p->name, KVM_DBGFS_GET_MODE(p),
653 kvm->debugfs_dentry, stat_data,
660 * Called after the VM is otherwise initialized, but just before adding it to
663 int __weak kvm_arch_post_init_vm(struct kvm *kvm)
669 * Called just after removing the VM from the vm_list, but before doing any
672 void __weak kvm_arch_pre_destroy_vm(struct kvm *kvm)
676 static struct kvm *kvm_create_vm(unsigned long type)
678 struct kvm *kvm = kvm_arch_alloc_vm();
683 return ERR_PTR(-ENOMEM);
685 spin_lock_init(&kvm->mmu_lock);
687 kvm->mm = current->mm;
688 kvm_eventfd_init(kvm);
689 mutex_init(&kvm->lock);
690 mutex_init(&kvm->irq_lock);
691 mutex_init(&kvm->slots_lock);
692 INIT_LIST_HEAD(&kvm->devices);
694 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
696 if (init_srcu_struct(&kvm->srcu))
697 goto out_err_no_srcu;
698 if (init_srcu_struct(&kvm->irq_srcu))
699 goto out_err_no_irq_srcu;
701 refcount_set(&kvm->users_count, 1);
702 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
703 struct kvm_memslots *slots = kvm_alloc_memslots();
706 goto out_err_no_arch_destroy_vm;
707 /* Generations must be different for each address space. */
708 slots->generation = i;
709 rcu_assign_pointer(kvm->memslots[i], slots);
712 for (i = 0; i < KVM_NR_BUSES; i++) {
713 rcu_assign_pointer(kvm->buses[i],
714 kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL_ACCOUNT));
716 goto out_err_no_arch_destroy_vm;
719 r = kvm_arch_init_vm(kvm, type);
721 goto out_err_no_arch_destroy_vm;
723 r = hardware_enable_all();
725 goto out_err_no_disable;
727 #ifdef CONFIG_HAVE_KVM_IRQFD
728 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
731 r = kvm_init_mmu_notifier(kvm);
733 goto out_err_no_mmu_notifier;
735 r = kvm_arch_post_init_vm(kvm);
739 mutex_lock(&kvm_lock);
740 list_add(&kvm->vm_list, &vm_list);
741 mutex_unlock(&kvm_lock);
743 preempt_notifier_inc();
748 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
749 if (kvm->mmu_notifier.ops)
750 mmu_notifier_unregister(&kvm->mmu_notifier, current->mm);
752 out_err_no_mmu_notifier:
753 hardware_disable_all();
755 kvm_arch_destroy_vm(kvm);
756 out_err_no_arch_destroy_vm:
757 WARN_ON_ONCE(!refcount_dec_and_test(&kvm->users_count));
758 for (i = 0; i < KVM_NR_BUSES; i++)
759 kfree(kvm_get_bus(kvm, i));
760 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
761 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
762 cleanup_srcu_struct(&kvm->irq_srcu);
764 cleanup_srcu_struct(&kvm->srcu);
766 kvm_arch_free_vm(kvm);
771 static void kvm_destroy_devices(struct kvm *kvm)
773 struct kvm_device *dev, *tmp;
776 * We do not need to take the kvm->lock here, because nobody else
777 * has a reference to the struct kvm at this point and therefore
778 * cannot access the devices list anyhow.
780 list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
781 list_del(&dev->vm_node);
782 dev->ops->destroy(dev);
786 static void kvm_destroy_vm(struct kvm *kvm)
789 struct mm_struct *mm = kvm->mm;
791 kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm);
792 kvm_destroy_vm_debugfs(kvm);
793 kvm_arch_sync_events(kvm);
794 mutex_lock(&kvm_lock);
795 list_del(&kvm->vm_list);
796 mutex_unlock(&kvm_lock);
797 kvm_arch_pre_destroy_vm(kvm);
799 kvm_free_irq_routing(kvm);
800 for (i = 0; i < KVM_NR_BUSES; i++) {
801 struct kvm_io_bus *bus = kvm_get_bus(kvm, i);
804 kvm_io_bus_destroy(bus);
805 kvm->buses[i] = NULL;
807 kvm_coalesced_mmio_free(kvm);
808 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
809 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
811 kvm_arch_flush_shadow_all(kvm);
813 kvm_arch_destroy_vm(kvm);
814 kvm_destroy_devices(kvm);
815 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
816 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
817 cleanup_srcu_struct(&kvm->irq_srcu);
818 cleanup_srcu_struct(&kvm->srcu);
819 kvm_arch_free_vm(kvm);
820 preempt_notifier_dec();
821 hardware_disable_all();
825 void kvm_get_kvm(struct kvm *kvm)
827 refcount_inc(&kvm->users_count);
829 EXPORT_SYMBOL_GPL(kvm_get_kvm);
831 void kvm_put_kvm(struct kvm *kvm)
833 if (refcount_dec_and_test(&kvm->users_count))
836 EXPORT_SYMBOL_GPL(kvm_put_kvm);
839 * Used to put a reference that was taken on behalf of an object associated
840 * with a user-visible file descriptor, e.g. a vcpu or device, if installation
841 * of the new file descriptor fails and the reference cannot be transferred to
842 * its final owner. In such cases, the caller is still actively using @kvm and
843 * will fail miserably if the refcount unexpectedly hits zero.
845 void kvm_put_kvm_no_destroy(struct kvm *kvm)
847 WARN_ON(refcount_dec_and_test(&kvm->users_count));
849 EXPORT_SYMBOL_GPL(kvm_put_kvm_no_destroy);
851 static int kvm_vm_release(struct inode *inode, struct file *filp)
853 struct kvm *kvm = filp->private_data;
855 kvm_irqfd_release(kvm);
862 * Allocation size is twice as large as the actual dirty bitmap size.
863 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
865 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
867 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
869 memslot->dirty_bitmap = kvzalloc(dirty_bytes, GFP_KERNEL_ACCOUNT);
870 if (!memslot->dirty_bitmap)
877 * Insert memslot and re-sort memslots based on their GFN,
878 * so binary search could be used to lookup GFN.
879 * Sorting algorithm takes advantage of having initially
880 * sorted array and known changed memslot position.
882 static void update_memslots(struct kvm_memslots *slots,
883 struct kvm_memory_slot *new,
884 enum kvm_mr_change change)
887 int i = slots->id_to_index[id];
888 struct kvm_memory_slot *mslots = slots->memslots;
890 WARN_ON(mslots[i].id != id);
894 WARN_ON(mslots[i].npages || !new->npages);
898 WARN_ON(new->npages || !mslots[i].npages);
904 while (i < KVM_MEM_SLOTS_NUM - 1 &&
905 new->base_gfn <= mslots[i + 1].base_gfn) {
906 if (!mslots[i + 1].npages)
908 mslots[i] = mslots[i + 1];
909 slots->id_to_index[mslots[i].id] = i;
914 * The ">=" is needed when creating a slot with base_gfn == 0,
915 * so that it moves before all those with base_gfn == npages == 0.
917 * On the other hand, if new->npages is zero, the above loop has
918 * already left i pointing to the beginning of the empty part of
919 * mslots, and the ">=" would move the hole backwards in this
920 * case---which is wrong. So skip the loop when deleting a slot.
924 new->base_gfn >= mslots[i - 1].base_gfn) {
925 mslots[i] = mslots[i - 1];
926 slots->id_to_index[mslots[i].id] = i;
930 WARN_ON_ONCE(i != slots->used_slots);
933 slots->id_to_index[mslots[i].id] = i;
936 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
938 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
940 #ifdef __KVM_HAVE_READONLY_MEM
941 valid_flags |= KVM_MEM_READONLY;
944 if (mem->flags & ~valid_flags)
950 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
951 int as_id, struct kvm_memslots *slots)
953 struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id);
954 u64 gen = old_memslots->generation;
956 WARN_ON(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS);
957 slots->generation = gen | KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
959 rcu_assign_pointer(kvm->memslots[as_id], slots);
960 synchronize_srcu_expedited(&kvm->srcu);
963 * Increment the new memslot generation a second time, dropping the
964 * update in-progress flag and incrementing the generation based on
965 * the number of address spaces. This provides a unique and easily
966 * identifiable generation number while the memslots are in flux.
968 gen = slots->generation & ~KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
971 * Generations must be unique even across address spaces. We do not need
972 * a global counter for that, instead the generation space is evenly split
973 * across address spaces. For example, with two address spaces, address
974 * space 0 will use generations 0, 2, 4, ... while address space 1 will
975 * use generations 1, 3, 5, ...
977 gen += KVM_ADDRESS_SPACE_NUM;
979 kvm_arch_memslots_updated(kvm, gen);
981 slots->generation = gen;
987 * Allocate some memory and give it an address in the guest physical address
990 * Discontiguous memory is allowed, mostly for framebuffers.
992 * Must be called holding kvm->slots_lock for write.
994 int __kvm_set_memory_region(struct kvm *kvm,
995 const struct kvm_userspace_memory_region *mem)
999 unsigned long npages;
1000 struct kvm_memory_slot *slot;
1001 struct kvm_memory_slot old, new;
1002 struct kvm_memslots *slots = NULL, *old_memslots;
1004 enum kvm_mr_change change;
1006 r = check_memory_region_flags(mem);
1011 as_id = mem->slot >> 16;
1012 id = (u16)mem->slot;
1014 /* General sanity checks */
1015 if (mem->memory_size & (PAGE_SIZE - 1))
1017 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
1019 /* We can read the guest memory with __xxx_user() later on. */
1020 if ((id < KVM_USER_MEM_SLOTS) &&
1021 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
1022 !access_ok((void __user *)(unsigned long)mem->userspace_addr,
1025 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
1027 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
1030 slot = id_to_memslot(__kvm_memslots(kvm, as_id), id);
1031 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
1032 npages = mem->memory_size >> PAGE_SHIFT;
1034 if (npages > KVM_MEM_MAX_NR_PAGES)
1040 new.base_gfn = base_gfn;
1041 new.npages = npages;
1042 new.flags = mem->flags;
1046 change = KVM_MR_CREATE;
1047 else { /* Modify an existing slot. */
1048 if ((mem->userspace_addr != old.userspace_addr) ||
1049 (npages != old.npages) ||
1050 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
1053 if (base_gfn != old.base_gfn)
1054 change = KVM_MR_MOVE;
1055 else if (new.flags != old.flags)
1056 change = KVM_MR_FLAGS_ONLY;
1057 else { /* Nothing to change. */
1066 change = KVM_MR_DELETE;
1071 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
1072 /* Check for overlaps */
1074 kvm_for_each_memslot(slot, __kvm_memslots(kvm, as_id)) {
1077 if (!((base_gfn + npages <= slot->base_gfn) ||
1078 (base_gfn >= slot->base_gfn + slot->npages)))
1083 /* Free page dirty bitmap if unneeded */
1084 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
1085 new.dirty_bitmap = NULL;
1088 if (change == KVM_MR_CREATE) {
1089 new.userspace_addr = mem->userspace_addr;
1091 if (kvm_arch_create_memslot(kvm, &new, npages))
1095 /* Allocate page dirty bitmap if needed */
1096 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
1097 if (kvm_create_dirty_bitmap(&new) < 0)
1101 slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL_ACCOUNT);
1104 memcpy(slots, __kvm_memslots(kvm, as_id), sizeof(struct kvm_memslots));
1106 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
1107 slot = id_to_memslot(slots, id);
1108 slot->flags |= KVM_MEMSLOT_INVALID;
1110 old_memslots = install_new_memslots(kvm, as_id, slots);
1112 /* From this point no new shadow pages pointing to a deleted,
1113 * or moved, memslot will be created.
1115 * validation of sp->gfn happens in:
1116 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1117 * - kvm_is_visible_gfn (mmu_check_root)
1119 kvm_arch_flush_shadow_memslot(kvm, slot);
1122 * We can re-use the old_memslots from above, the only difference
1123 * from the currently installed memslots is the invalid flag. This
1124 * will get overwritten by update_memslots anyway.
1126 slots = old_memslots;
1129 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
1133 /* actual memory is freed via old in kvm_free_memslot below */
1134 if (change == KVM_MR_DELETE) {
1135 new.dirty_bitmap = NULL;
1136 memset(&new.arch, 0, sizeof(new.arch));
1139 update_memslots(slots, &new, change);
1140 old_memslots = install_new_memslots(kvm, as_id, slots);
1142 kvm_arch_commit_memory_region(kvm, mem, &old, &new, change);
1144 kvm_free_memslot(kvm, &old, &new);
1145 kvfree(old_memslots);
1151 kvm_free_memslot(kvm, &new, &old);
1155 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1157 int kvm_set_memory_region(struct kvm *kvm,
1158 const struct kvm_userspace_memory_region *mem)
1162 mutex_lock(&kvm->slots_lock);
1163 r = __kvm_set_memory_region(kvm, mem);
1164 mutex_unlock(&kvm->slots_lock);
1167 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1169 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1170 struct kvm_userspace_memory_region *mem)
1172 if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1175 return kvm_set_memory_region(kvm, mem);
1178 int kvm_get_dirty_log(struct kvm *kvm,
1179 struct kvm_dirty_log *log, int *is_dirty)
1181 struct kvm_memslots *slots;
1182 struct kvm_memory_slot *memslot;
1185 unsigned long any = 0;
1187 as_id = log->slot >> 16;
1188 id = (u16)log->slot;
1189 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1192 slots = __kvm_memslots(kvm, as_id);
1193 memslot = id_to_memslot(slots, id);
1194 if (!memslot->dirty_bitmap)
1197 n = kvm_dirty_bitmap_bytes(memslot);
1199 for (i = 0; !any && i < n/sizeof(long); ++i)
1200 any = memslot->dirty_bitmap[i];
1202 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1209 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1211 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1213 * kvm_get_dirty_log_protect - get a snapshot of dirty pages
1214 * and reenable dirty page tracking for the corresponding pages.
1215 * @kvm: pointer to kvm instance
1216 * @log: slot id and address to which we copy the log
1217 * @flush: true if TLB flush is needed by caller
1219 * We need to keep it in mind that VCPU threads can write to the bitmap
1220 * concurrently. So, to avoid losing track of dirty pages we keep the
1223 * 1. Take a snapshot of the bit and clear it if needed.
1224 * 2. Write protect the corresponding page.
1225 * 3. Copy the snapshot to the userspace.
1226 * 4. Upon return caller flushes TLB's if needed.
1228 * Between 2 and 4, the guest may write to the page using the remaining TLB
1229 * entry. This is not a problem because the page is reported dirty using
1230 * the snapshot taken before and step 4 ensures that writes done after
1231 * exiting to userspace will be logged for the next call.
1234 int kvm_get_dirty_log_protect(struct kvm *kvm,
1235 struct kvm_dirty_log *log, bool *flush)
1237 struct kvm_memslots *slots;
1238 struct kvm_memory_slot *memslot;
1241 unsigned long *dirty_bitmap;
1242 unsigned long *dirty_bitmap_buffer;
1244 as_id = log->slot >> 16;
1245 id = (u16)log->slot;
1246 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1249 slots = __kvm_memslots(kvm, as_id);
1250 memslot = id_to_memslot(slots, id);
1252 dirty_bitmap = memslot->dirty_bitmap;
1256 n = kvm_dirty_bitmap_bytes(memslot);
1258 if (kvm->manual_dirty_log_protect) {
1260 * Unlike kvm_get_dirty_log, we always return false in *flush,
1261 * because no flush is needed until KVM_CLEAR_DIRTY_LOG. There
1262 * is some code duplication between this function and
1263 * kvm_get_dirty_log, but hopefully all architecture
1264 * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log
1265 * can be eliminated.
1267 dirty_bitmap_buffer = dirty_bitmap;
1269 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
1270 memset(dirty_bitmap_buffer, 0, n);
1272 spin_lock(&kvm->mmu_lock);
1273 for (i = 0; i < n / sizeof(long); i++) {
1277 if (!dirty_bitmap[i])
1281 mask = xchg(&dirty_bitmap[i], 0);
1282 dirty_bitmap_buffer[i] = mask;
1284 offset = i * BITS_PER_LONG;
1285 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1288 spin_unlock(&kvm->mmu_lock);
1291 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1295 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect);
1298 * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap
1299 * and reenable dirty page tracking for the corresponding pages.
1300 * @kvm: pointer to kvm instance
1301 * @log: slot id and address from which to fetch the bitmap of dirty pages
1302 * @flush: true if TLB flush is needed by caller
1304 int kvm_clear_dirty_log_protect(struct kvm *kvm,
1305 struct kvm_clear_dirty_log *log, bool *flush)
1307 struct kvm_memslots *slots;
1308 struct kvm_memory_slot *memslot;
1312 unsigned long *dirty_bitmap;
1313 unsigned long *dirty_bitmap_buffer;
1315 as_id = log->slot >> 16;
1316 id = (u16)log->slot;
1317 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1320 if (log->first_page & 63)
1323 slots = __kvm_memslots(kvm, as_id);
1324 memslot = id_to_memslot(slots, id);
1326 dirty_bitmap = memslot->dirty_bitmap;
1330 n = ALIGN(log->num_pages, BITS_PER_LONG) / 8;
1332 if (log->first_page > memslot->npages ||
1333 log->num_pages > memslot->npages - log->first_page ||
1334 (log->num_pages < memslot->npages - log->first_page && (log->num_pages & 63)))
1338 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
1339 if (copy_from_user(dirty_bitmap_buffer, log->dirty_bitmap, n))
1342 spin_lock(&kvm->mmu_lock);
1343 for (offset = log->first_page, i = offset / BITS_PER_LONG,
1344 n = DIV_ROUND_UP(log->num_pages, BITS_PER_LONG); n--;
1345 i++, offset += BITS_PER_LONG) {
1346 unsigned long mask = *dirty_bitmap_buffer++;
1347 atomic_long_t *p = (atomic_long_t *) &dirty_bitmap[i];
1351 mask &= atomic_long_fetch_andnot(mask, p);
1354 * mask contains the bits that really have been cleared. This
1355 * never includes any bits beyond the length of the memslot (if
1356 * the length is not aligned to 64 pages), therefore it is not
1357 * a problem if userspace sets them in log->dirty_bitmap.
1361 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1365 spin_unlock(&kvm->mmu_lock);
1369 EXPORT_SYMBOL_GPL(kvm_clear_dirty_log_protect);
1372 bool kvm_largepages_enabled(void)
1374 return largepages_enabled;
1377 void kvm_disable_largepages(void)
1379 largepages_enabled = false;
1381 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1383 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1385 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1387 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1389 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
1391 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn);
1394 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1396 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1398 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1399 memslot->flags & KVM_MEMSLOT_INVALID)
1404 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1406 unsigned long kvm_host_page_size(struct kvm_vcpu *vcpu, gfn_t gfn)
1408 struct vm_area_struct *vma;
1409 unsigned long addr, size;
1413 addr = kvm_vcpu_gfn_to_hva_prot(vcpu, gfn, NULL);
1414 if (kvm_is_error_hva(addr))
1417 down_read(¤t->mm->mmap_sem);
1418 vma = find_vma(current->mm, addr);
1422 size = vma_kernel_pagesize(vma);
1425 up_read(¤t->mm->mmap_sem);
1430 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1432 return slot->flags & KVM_MEM_READONLY;
1435 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1436 gfn_t *nr_pages, bool write)
1438 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1439 return KVM_HVA_ERR_BAD;
1441 if (memslot_is_readonly(slot) && write)
1442 return KVM_HVA_ERR_RO_BAD;
1445 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1447 return __gfn_to_hva_memslot(slot, gfn);
1450 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1453 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1456 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1459 return gfn_to_hva_many(slot, gfn, NULL);
1461 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1463 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1465 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1467 EXPORT_SYMBOL_GPL(gfn_to_hva);
1469 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
1471 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
1473 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
1476 * Return the hva of a @gfn and the R/W attribute if possible.
1478 * @slot: the kvm_memory_slot which contains @gfn
1479 * @gfn: the gfn to be translated
1480 * @writable: used to return the read/write attribute of the @slot if the hva
1481 * is valid and @writable is not NULL
1483 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1484 gfn_t gfn, bool *writable)
1486 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1488 if (!kvm_is_error_hva(hva) && writable)
1489 *writable = !memslot_is_readonly(slot);
1494 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1496 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1498 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1501 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
1503 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1505 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1508 static inline int check_user_page_hwpoison(unsigned long addr)
1510 int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
1512 rc = get_user_pages(addr, 1, flags, NULL, NULL);
1513 return rc == -EHWPOISON;
1517 * The fast path to get the writable pfn which will be stored in @pfn,
1518 * true indicates success, otherwise false is returned. It's also the
1519 * only part that runs if we can in atomic context.
1521 static bool hva_to_pfn_fast(unsigned long addr, bool write_fault,
1522 bool *writable, kvm_pfn_t *pfn)
1524 struct page *page[1];
1528 * Fast pin a writable pfn only if it is a write fault request
1529 * or the caller allows to map a writable pfn for a read fault
1532 if (!(write_fault || writable))
1535 npages = __get_user_pages_fast(addr, 1, 1, page);
1537 *pfn = page_to_pfn(page[0]);
1548 * The slow path to get the pfn of the specified host virtual address,
1549 * 1 indicates success, -errno is returned if error is detected.
1551 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1552 bool *writable, kvm_pfn_t *pfn)
1554 unsigned int flags = FOLL_HWPOISON;
1561 *writable = write_fault;
1564 flags |= FOLL_WRITE;
1566 flags |= FOLL_NOWAIT;
1568 npages = get_user_pages_unlocked(addr, 1, &page, flags);
1572 /* map read fault as writable if possible */
1573 if (unlikely(!write_fault) && writable) {
1576 if (__get_user_pages_fast(addr, 1, 1, &wpage) == 1) {
1582 *pfn = page_to_pfn(page);
1586 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1588 if (unlikely(!(vma->vm_flags & VM_READ)))
1591 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1597 static int hva_to_pfn_remapped(struct vm_area_struct *vma,
1598 unsigned long addr, bool *async,
1599 bool write_fault, bool *writable,
1605 r = follow_pfn(vma, addr, &pfn);
1608 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
1609 * not call the fault handler, so do it here.
1611 bool unlocked = false;
1612 r = fixup_user_fault(current, current->mm, addr,
1613 (write_fault ? FAULT_FLAG_WRITE : 0),
1620 r = follow_pfn(vma, addr, &pfn);
1630 * Get a reference here because callers of *hva_to_pfn* and
1631 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
1632 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
1633 * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will
1634 * simply do nothing for reserved pfns.
1636 * Whoever called remap_pfn_range is also going to call e.g.
1637 * unmap_mapping_range before the underlying pages are freed,
1638 * causing a call to our MMU notifier.
1647 * Pin guest page in memory and return its pfn.
1648 * @addr: host virtual address which maps memory to the guest
1649 * @atomic: whether this function can sleep
1650 * @async: whether this function need to wait IO complete if the
1651 * host page is not in the memory
1652 * @write_fault: whether we should get a writable host page
1653 * @writable: whether it allows to map a writable host page for !@write_fault
1655 * The function will map a writable host page for these two cases:
1656 * 1): @write_fault = true
1657 * 2): @write_fault = false && @writable, @writable will tell the caller
1658 * whether the mapping is writable.
1660 static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1661 bool write_fault, bool *writable)
1663 struct vm_area_struct *vma;
1667 /* we can do it either atomically or asynchronously, not both */
1668 BUG_ON(atomic && async);
1670 if (hva_to_pfn_fast(addr, write_fault, writable, &pfn))
1674 return KVM_PFN_ERR_FAULT;
1676 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1680 down_read(¤t->mm->mmap_sem);
1681 if (npages == -EHWPOISON ||
1682 (!async && check_user_page_hwpoison(addr))) {
1683 pfn = KVM_PFN_ERR_HWPOISON;
1688 vma = find_vma_intersection(current->mm, addr, addr + 1);
1691 pfn = KVM_PFN_ERR_FAULT;
1692 else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
1693 r = hva_to_pfn_remapped(vma, addr, async, write_fault, writable, &pfn);
1697 pfn = KVM_PFN_ERR_FAULT;
1699 if (async && vma_is_valid(vma, write_fault))
1701 pfn = KVM_PFN_ERR_FAULT;
1704 up_read(¤t->mm->mmap_sem);
1708 kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn,
1709 bool atomic, bool *async, bool write_fault,
1712 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1714 if (addr == KVM_HVA_ERR_RO_BAD) {
1717 return KVM_PFN_ERR_RO_FAULT;
1720 if (kvm_is_error_hva(addr)) {
1723 return KVM_PFN_NOSLOT;
1726 /* Do not map writable pfn in the readonly memslot. */
1727 if (writable && memslot_is_readonly(slot)) {
1732 return hva_to_pfn(addr, atomic, async, write_fault,
1735 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
1737 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1740 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
1741 write_fault, writable);
1743 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1745 kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1747 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1749 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
1751 kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1753 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1755 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1757 kvm_pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1759 return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm, gfn), gfn);
1761 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1763 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
1765 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1767 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
1769 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1771 return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
1773 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1775 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
1777 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1779 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
1781 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1782 struct page **pages, int nr_pages)
1787 addr = gfn_to_hva_many(slot, gfn, &entry);
1788 if (kvm_is_error_hva(addr))
1791 if (entry < nr_pages)
1794 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1796 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1798 static struct page *kvm_pfn_to_page(kvm_pfn_t pfn)
1800 if (is_error_noslot_pfn(pfn))
1801 return KVM_ERR_PTR_BAD_PAGE;
1803 if (kvm_is_reserved_pfn(pfn)) {
1805 return KVM_ERR_PTR_BAD_PAGE;
1808 return pfn_to_page(pfn);
1811 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1815 pfn = gfn_to_pfn(kvm, gfn);
1817 return kvm_pfn_to_page(pfn);
1819 EXPORT_SYMBOL_GPL(gfn_to_page);
1821 void kvm_release_pfn(kvm_pfn_t pfn, bool dirty, struct gfn_to_pfn_cache *cache)
1827 cache->pfn = cache->gfn = 0;
1830 kvm_release_pfn_dirty(pfn);
1832 kvm_release_pfn_clean(pfn);
1835 static void kvm_cache_gfn_to_pfn(struct kvm_memory_slot *slot, gfn_t gfn,
1836 struct gfn_to_pfn_cache *cache, u64 gen)
1838 kvm_release_pfn(cache->pfn, cache->dirty, cache);
1840 cache->pfn = gfn_to_pfn_memslot(slot, gfn);
1842 cache->dirty = false;
1843 cache->generation = gen;
1846 static int __kvm_map_gfn(struct kvm_memslots *slots, gfn_t gfn,
1847 struct kvm_host_map *map,
1848 struct gfn_to_pfn_cache *cache,
1853 struct page *page = KVM_UNMAPPED_PAGE;
1854 struct kvm_memory_slot *slot = __gfn_to_memslot(slots, gfn);
1855 u64 gen = slots->generation;
1861 if (!cache->pfn || cache->gfn != gfn ||
1862 cache->generation != gen) {
1865 kvm_cache_gfn_to_pfn(slot, gfn, cache, gen);
1871 pfn = gfn_to_pfn_memslot(slot, gfn);
1873 if (is_error_noslot_pfn(pfn))
1876 if (pfn_valid(pfn)) {
1877 page = pfn_to_page(pfn);
1879 hva = kmap_atomic(page);
1882 #ifdef CONFIG_HAS_IOMEM
1883 } else if (!atomic) {
1884 hva = memremap(pfn_to_hpa(pfn), PAGE_SIZE, MEMREMAP_WB);
1901 int kvm_map_gfn(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map,
1902 struct gfn_to_pfn_cache *cache, bool atomic)
1904 return __kvm_map_gfn(kvm_memslots(vcpu->kvm), gfn, map,
1907 EXPORT_SYMBOL_GPL(kvm_map_gfn);
1909 int kvm_vcpu_map(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map)
1911 return __kvm_map_gfn(kvm_vcpu_memslots(vcpu), gfn, map,
1914 EXPORT_SYMBOL_GPL(kvm_vcpu_map);
1916 static void __kvm_unmap_gfn(struct kvm_memory_slot *memslot,
1917 struct kvm_host_map *map,
1918 struct gfn_to_pfn_cache *cache,
1919 bool dirty, bool atomic)
1927 if (map->page != KVM_UNMAPPED_PAGE) {
1929 kunmap_atomic(map->hva);
1933 #ifdef CONFIG_HAS_IOMEM
1937 WARN_ONCE(1, "Unexpected unmapping in atomic context");
1941 mark_page_dirty_in_slot(memslot, map->gfn);
1944 cache->dirty |= dirty;
1946 kvm_release_pfn(map->pfn, dirty, NULL);
1952 int kvm_unmap_gfn(struct kvm_vcpu *vcpu, struct kvm_host_map *map,
1953 struct gfn_to_pfn_cache *cache, bool dirty, bool atomic)
1955 __kvm_unmap_gfn(gfn_to_memslot(vcpu->kvm, map->gfn), map,
1956 cache, dirty, atomic);
1959 EXPORT_SYMBOL_GPL(kvm_unmap_gfn);
1961 void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map, bool dirty)
1963 __kvm_unmap_gfn(kvm_vcpu_gfn_to_memslot(vcpu, map->gfn), map, NULL,
1966 EXPORT_SYMBOL_GPL(kvm_vcpu_unmap);
1968 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
1972 pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
1974 return kvm_pfn_to_page(pfn);
1976 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
1978 void kvm_release_page_clean(struct page *page)
1980 WARN_ON(is_error_page(page));
1982 kvm_release_pfn_clean(page_to_pfn(page));
1984 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1986 void kvm_release_pfn_clean(kvm_pfn_t pfn)
1988 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
1989 put_page(pfn_to_page(pfn));
1991 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1993 void kvm_release_page_dirty(struct page *page)
1995 WARN_ON(is_error_page(page));
1997 kvm_release_pfn_dirty(page_to_pfn(page));
1999 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
2001 void kvm_release_pfn_dirty(kvm_pfn_t pfn)
2003 kvm_set_pfn_dirty(pfn);
2004 kvm_release_pfn_clean(pfn);
2006 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
2008 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
2010 if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn))
2011 SetPageDirty(pfn_to_page(pfn));
2013 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
2015 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
2017 if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn))
2018 mark_page_accessed(pfn_to_page(pfn));
2020 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
2022 void kvm_get_pfn(kvm_pfn_t pfn)
2024 if (!kvm_is_reserved_pfn(pfn))
2025 get_page(pfn_to_page(pfn));
2027 EXPORT_SYMBOL_GPL(kvm_get_pfn);
2029 static int next_segment(unsigned long len, int offset)
2031 if (len > PAGE_SIZE - offset)
2032 return PAGE_SIZE - offset;
2037 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
2038 void *data, int offset, int len)
2043 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
2044 if (kvm_is_error_hva(addr))
2046 r = __copy_from_user(data, (void __user *)addr + offset, len);
2052 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
2055 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2057 return __kvm_read_guest_page(slot, gfn, data, offset, len);
2059 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
2061 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
2062 int offset, int len)
2064 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2066 return __kvm_read_guest_page(slot, gfn, data, offset, len);
2068 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
2070 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
2072 gfn_t gfn = gpa >> PAGE_SHIFT;
2074 int offset = offset_in_page(gpa);
2077 while ((seg = next_segment(len, offset)) != 0) {
2078 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
2088 EXPORT_SYMBOL_GPL(kvm_read_guest);
2090 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
2092 gfn_t gfn = gpa >> PAGE_SHIFT;
2094 int offset = offset_in_page(gpa);
2097 while ((seg = next_segment(len, offset)) != 0) {
2098 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
2108 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
2110 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
2111 void *data, int offset, unsigned long len)
2116 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
2117 if (kvm_is_error_hva(addr))
2119 pagefault_disable();
2120 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
2127 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
2128 void *data, unsigned long len)
2130 gfn_t gfn = gpa >> PAGE_SHIFT;
2131 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2132 int offset = offset_in_page(gpa);
2134 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
2136 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
2138 static int __kvm_write_guest_page(struct kvm_memory_slot *memslot, gfn_t gfn,
2139 const void *data, int offset, int len)
2144 addr = gfn_to_hva_memslot(memslot, gfn);
2145 if (kvm_is_error_hva(addr))
2147 r = __copy_to_user((void __user *)addr + offset, data, len);
2150 mark_page_dirty_in_slot(memslot, gfn);
2154 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
2155 const void *data, int offset, int len)
2157 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2159 return __kvm_write_guest_page(slot, gfn, data, offset, len);
2161 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
2163 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
2164 const void *data, int offset, int len)
2166 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2168 return __kvm_write_guest_page(slot, gfn, data, offset, len);
2170 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
2172 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
2175 gfn_t gfn = gpa >> PAGE_SHIFT;
2177 int offset = offset_in_page(gpa);
2180 while ((seg = next_segment(len, offset)) != 0) {
2181 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
2191 EXPORT_SYMBOL_GPL(kvm_write_guest);
2193 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
2196 gfn_t gfn = gpa >> PAGE_SHIFT;
2198 int offset = offset_in_page(gpa);
2201 while ((seg = next_segment(len, offset)) != 0) {
2202 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
2212 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
2214 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
2215 struct gfn_to_hva_cache *ghc,
2216 gpa_t gpa, unsigned long len)
2218 int offset = offset_in_page(gpa);
2219 gfn_t start_gfn = gpa >> PAGE_SHIFT;
2220 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
2221 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
2222 gfn_t nr_pages_avail;
2224 /* Update ghc->generation before performing any error checks. */
2225 ghc->generation = slots->generation;
2227 if (start_gfn > end_gfn) {
2228 ghc->hva = KVM_HVA_ERR_BAD;
2233 * If the requested region crosses two memslots, we still
2234 * verify that the entire region is valid here.
2236 for ( ; start_gfn <= end_gfn; start_gfn += nr_pages_avail) {
2237 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
2238 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
2240 if (kvm_is_error_hva(ghc->hva))
2244 /* Use the slow path for cross page reads and writes. */
2245 if (nr_pages_needed == 1)
2248 ghc->memslot = NULL;
2255 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2256 gpa_t gpa, unsigned long len)
2258 struct kvm_memslots *slots = kvm_memslots(kvm);
2259 return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len);
2261 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
2263 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2264 void *data, unsigned int offset,
2267 struct kvm_memslots *slots = kvm_memslots(kvm);
2269 gpa_t gpa = ghc->gpa + offset;
2271 BUG_ON(len + offset > ghc->len);
2273 if (slots->generation != ghc->generation) {
2274 if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len))
2278 if (kvm_is_error_hva(ghc->hva))
2281 if (unlikely(!ghc->memslot))
2282 return kvm_write_guest(kvm, gpa, data, len);
2284 r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
2287 mark_page_dirty_in_slot(ghc->memslot, gpa >> PAGE_SHIFT);
2291 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached);
2293 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2294 void *data, unsigned long len)
2296 return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len);
2298 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
2300 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2301 void *data, unsigned long len)
2303 struct kvm_memslots *slots = kvm_memslots(kvm);
2306 BUG_ON(len > ghc->len);
2308 if (slots->generation != ghc->generation) {
2309 if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len))
2313 if (kvm_is_error_hva(ghc->hva))
2316 if (unlikely(!ghc->memslot))
2317 return kvm_read_guest(kvm, ghc->gpa, data, len);
2319 r = __copy_from_user(data, (void __user *)ghc->hva, len);
2325 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
2327 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
2329 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
2331 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
2333 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
2335 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
2337 gfn_t gfn = gpa >> PAGE_SHIFT;
2339 int offset = offset_in_page(gpa);
2342 while ((seg = next_segment(len, offset)) != 0) {
2343 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
2352 EXPORT_SYMBOL_GPL(kvm_clear_guest);
2354 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot,
2357 if (memslot && memslot->dirty_bitmap) {
2358 unsigned long rel_gfn = gfn - memslot->base_gfn;
2360 set_bit_le(rel_gfn, memslot->dirty_bitmap);
2364 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
2366 struct kvm_memory_slot *memslot;
2368 memslot = gfn_to_memslot(kvm, gfn);
2369 mark_page_dirty_in_slot(memslot, gfn);
2371 EXPORT_SYMBOL_GPL(mark_page_dirty);
2373 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
2375 struct kvm_memory_slot *memslot;
2377 memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2378 mark_page_dirty_in_slot(memslot, gfn);
2380 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
2382 void kvm_sigset_activate(struct kvm_vcpu *vcpu)
2384 if (!vcpu->sigset_active)
2388 * This does a lockless modification of ->real_blocked, which is fine
2389 * because, only current can change ->real_blocked and all readers of
2390 * ->real_blocked don't care as long ->real_blocked is always a subset
2393 sigprocmask(SIG_SETMASK, &vcpu->sigset, ¤t->real_blocked);
2396 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu)
2398 if (!vcpu->sigset_active)
2401 sigprocmask(SIG_SETMASK, ¤t->real_blocked, NULL);
2402 sigemptyset(¤t->real_blocked);
2405 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
2407 unsigned int old, val, grow, grow_start;
2409 old = val = vcpu->halt_poll_ns;
2410 grow_start = READ_ONCE(halt_poll_ns_grow_start);
2411 grow = READ_ONCE(halt_poll_ns_grow);
2416 if (val < grow_start)
2419 if (val > halt_poll_ns)
2422 vcpu->halt_poll_ns = val;
2424 trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
2427 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
2429 unsigned int old, val, shrink;
2431 old = val = vcpu->halt_poll_ns;
2432 shrink = READ_ONCE(halt_poll_ns_shrink);
2438 vcpu->halt_poll_ns = val;
2439 trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
2442 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
2445 int idx = srcu_read_lock(&vcpu->kvm->srcu);
2447 if (kvm_arch_vcpu_runnable(vcpu)) {
2448 kvm_make_request(KVM_REQ_UNHALT, vcpu);
2451 if (kvm_cpu_has_pending_timer(vcpu))
2453 if (signal_pending(current))
2458 srcu_read_unlock(&vcpu->kvm->srcu, idx);
2463 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2465 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
2468 DECLARE_SWAITQUEUE(wait);
2469 bool waited = false;
2472 kvm_arch_vcpu_blocking(vcpu);
2474 start = cur = ktime_get();
2475 if (vcpu->halt_poll_ns && !kvm_arch_no_poll(vcpu)) {
2476 ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns);
2478 ++vcpu->stat.halt_attempted_poll;
2481 * This sets KVM_REQ_UNHALT if an interrupt
2484 if (kvm_vcpu_check_block(vcpu) < 0) {
2485 ++vcpu->stat.halt_successful_poll;
2486 if (!vcpu_valid_wakeup(vcpu))
2487 ++vcpu->stat.halt_poll_invalid;
2491 } while (single_task_running() && ktime_before(cur, stop));
2495 prepare_to_swait_exclusive(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
2497 if (kvm_vcpu_check_block(vcpu) < 0)
2504 finish_swait(&vcpu->wq, &wait);
2507 kvm_arch_vcpu_unblocking(vcpu);
2508 block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
2510 if (!kvm_arch_no_poll(vcpu)) {
2511 if (!vcpu_valid_wakeup(vcpu)) {
2512 shrink_halt_poll_ns(vcpu);
2513 } else if (halt_poll_ns) {
2514 if (block_ns <= vcpu->halt_poll_ns)
2516 /* we had a long block, shrink polling */
2517 else if (vcpu->halt_poll_ns && block_ns > halt_poll_ns)
2518 shrink_halt_poll_ns(vcpu);
2519 /* we had a short halt and our poll time is too small */
2520 else if (vcpu->halt_poll_ns < halt_poll_ns &&
2521 block_ns < halt_poll_ns)
2522 grow_halt_poll_ns(vcpu);
2524 vcpu->halt_poll_ns = 0;
2528 trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu));
2529 kvm_arch_vcpu_block_finish(vcpu);
2531 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
2533 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
2535 struct swait_queue_head *wqp;
2537 wqp = kvm_arch_vcpu_wq(vcpu);
2538 if (swq_has_sleeper(wqp)) {
2540 WRITE_ONCE(vcpu->ready, true);
2541 ++vcpu->stat.halt_wakeup;
2547 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
2551 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2553 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
2556 int cpu = vcpu->cpu;
2558 if (kvm_vcpu_wake_up(vcpu))
2562 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
2563 if (kvm_arch_vcpu_should_kick(vcpu))
2564 smp_send_reschedule(cpu);
2567 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
2568 #endif /* !CONFIG_S390 */
2570 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
2573 struct task_struct *task = NULL;
2577 pid = rcu_dereference(target->pid);
2579 task = get_pid_task(pid, PIDTYPE_PID);
2583 ret = yield_to(task, 1);
2584 put_task_struct(task);
2588 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
2591 * Helper that checks whether a VCPU is eligible for directed yield.
2592 * Most eligible candidate to yield is decided by following heuristics:
2594 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2595 * (preempted lock holder), indicated by @in_spin_loop.
2596 * Set at the beiginning and cleared at the end of interception/PLE handler.
2598 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2599 * chance last time (mostly it has become eligible now since we have probably
2600 * yielded to lockholder in last iteration. This is done by toggling
2601 * @dy_eligible each time a VCPU checked for eligibility.)
2603 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2604 * to preempted lock-holder could result in wrong VCPU selection and CPU
2605 * burning. Giving priority for a potential lock-holder increases lock
2608 * Since algorithm is based on heuristics, accessing another VCPU data without
2609 * locking does not harm. It may result in trying to yield to same VCPU, fail
2610 * and continue with next VCPU and so on.
2612 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
2614 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2617 eligible = !vcpu->spin_loop.in_spin_loop ||
2618 vcpu->spin_loop.dy_eligible;
2620 if (vcpu->spin_loop.in_spin_loop)
2621 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
2630 * Unlike kvm_arch_vcpu_runnable, this function is called outside
2631 * a vcpu_load/vcpu_put pair. However, for most architectures
2632 * kvm_arch_vcpu_runnable does not require vcpu_load.
2634 bool __weak kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
2636 return kvm_arch_vcpu_runnable(vcpu);
2639 static bool vcpu_dy_runnable(struct kvm_vcpu *vcpu)
2641 if (kvm_arch_dy_runnable(vcpu))
2644 #ifdef CONFIG_KVM_ASYNC_PF
2645 if (!list_empty_careful(&vcpu->async_pf.done))
2652 void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode)
2654 struct kvm *kvm = me->kvm;
2655 struct kvm_vcpu *vcpu;
2656 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
2662 kvm_vcpu_set_in_spin_loop(me, true);
2664 * We boost the priority of a VCPU that is runnable but not
2665 * currently running, because it got preempted by something
2666 * else and called schedule in __vcpu_run. Hopefully that
2667 * VCPU is holding the lock that we need and will release it.
2668 * We approximate round-robin by starting at the last boosted VCPU.
2670 for (pass = 0; pass < 2 && !yielded && try; pass++) {
2671 kvm_for_each_vcpu(i, vcpu, kvm) {
2672 if (!pass && i <= last_boosted_vcpu) {
2673 i = last_boosted_vcpu;
2675 } else if (pass && i > last_boosted_vcpu)
2677 if (!READ_ONCE(vcpu->ready))
2681 if (swait_active(&vcpu->wq) && !vcpu_dy_runnable(vcpu))
2683 if (READ_ONCE(vcpu->preempted) && yield_to_kernel_mode &&
2684 !kvm_arch_vcpu_in_kernel(vcpu))
2686 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
2689 yielded = kvm_vcpu_yield_to(vcpu);
2691 kvm->last_boosted_vcpu = i;
2693 } else if (yielded < 0) {
2700 kvm_vcpu_set_in_spin_loop(me, false);
2702 /* Ensure vcpu is not eligible during next spinloop */
2703 kvm_vcpu_set_dy_eligible(me, false);
2705 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
2707 static vm_fault_t kvm_vcpu_fault(struct vm_fault *vmf)
2709 struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data;
2712 if (vmf->pgoff == 0)
2713 page = virt_to_page(vcpu->run);
2715 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
2716 page = virt_to_page(vcpu->arch.pio_data);
2718 #ifdef CONFIG_KVM_MMIO
2719 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
2720 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
2723 return kvm_arch_vcpu_fault(vcpu, vmf);
2729 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
2730 .fault = kvm_vcpu_fault,
2733 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2735 vma->vm_ops = &kvm_vcpu_vm_ops;
2739 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2741 struct kvm_vcpu *vcpu = filp->private_data;
2743 debugfs_remove_recursive(vcpu->debugfs_dentry);
2744 kvm_put_kvm(vcpu->kvm);
2748 static struct file_operations kvm_vcpu_fops = {
2749 .release = kvm_vcpu_release,
2750 .unlocked_ioctl = kvm_vcpu_ioctl,
2751 .mmap = kvm_vcpu_mmap,
2752 .llseek = noop_llseek,
2753 KVM_COMPAT(kvm_vcpu_compat_ioctl),
2757 * Allocates an inode for the vcpu.
2759 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2761 char name[8 + 1 + ITOA_MAX_LEN + 1];
2763 snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id);
2764 return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
2767 static void kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
2769 #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS
2770 char dir_name[ITOA_MAX_LEN * 2];
2772 if (!debugfs_initialized())
2775 snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
2776 vcpu->debugfs_dentry = debugfs_create_dir(dir_name,
2777 vcpu->kvm->debugfs_dentry);
2779 kvm_arch_create_vcpu_debugfs(vcpu);
2784 * Creates some virtual cpus. Good luck creating more than one.
2786 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
2789 struct kvm_vcpu *vcpu;
2792 if (id >= KVM_MAX_VCPU_ID)
2795 mutex_lock(&kvm->lock);
2796 if (kvm->created_vcpus == KVM_MAX_VCPUS) {
2797 mutex_unlock(&kvm->lock);
2801 kvm->created_vcpus++;
2802 mutex_unlock(&kvm->lock);
2804 r = kvm_arch_vcpu_precreate(kvm, id);
2806 goto vcpu_decrement;
2808 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
2811 goto vcpu_decrement;
2814 BUILD_BUG_ON(sizeof(struct kvm_run) > PAGE_SIZE);
2815 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2820 vcpu->run = page_address(page);
2822 kvm_vcpu_init(vcpu, kvm, id);
2824 r = kvm_arch_vcpu_create(vcpu);
2826 goto vcpu_free_run_page;
2828 kvm_create_vcpu_debugfs(vcpu);
2830 mutex_lock(&kvm->lock);
2831 if (kvm_get_vcpu_by_id(kvm, id)) {
2833 goto unlock_vcpu_destroy;
2836 vcpu->vcpu_idx = atomic_read(&kvm->online_vcpus);
2837 BUG_ON(kvm->vcpus[vcpu->vcpu_idx]);
2839 /* Now it's all set up, let userspace reach it */
2841 r = create_vcpu_fd(vcpu);
2843 kvm_put_kvm_no_destroy(kvm);
2844 goto unlock_vcpu_destroy;
2847 kvm->vcpus[vcpu->vcpu_idx] = vcpu;
2850 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
2851 * before kvm->online_vcpu's incremented value.
2854 atomic_inc(&kvm->online_vcpus);
2856 mutex_unlock(&kvm->lock);
2857 kvm_arch_vcpu_postcreate(vcpu);
2860 unlock_vcpu_destroy:
2861 mutex_unlock(&kvm->lock);
2862 debugfs_remove_recursive(vcpu->debugfs_dentry);
2863 kvm_arch_vcpu_destroy(vcpu);
2865 free_page((unsigned long)vcpu->run);
2867 kmem_cache_free(kvm_vcpu_cache, vcpu);
2869 mutex_lock(&kvm->lock);
2870 kvm->created_vcpus--;
2871 mutex_unlock(&kvm->lock);
2875 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2878 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2879 vcpu->sigset_active = 1;
2880 vcpu->sigset = *sigset;
2882 vcpu->sigset_active = 0;
2886 static long kvm_vcpu_ioctl(struct file *filp,
2887 unsigned int ioctl, unsigned long arg)
2889 struct kvm_vcpu *vcpu = filp->private_data;
2890 void __user *argp = (void __user *)arg;
2892 struct kvm_fpu *fpu = NULL;
2893 struct kvm_sregs *kvm_sregs = NULL;
2895 if (vcpu->kvm->mm != current->mm)
2898 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2902 * Some architectures have vcpu ioctls that are asynchronous to vcpu
2903 * execution; mutex_lock() would break them.
2905 r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg);
2906 if (r != -ENOIOCTLCMD)
2909 if (mutex_lock_killable(&vcpu->mutex))
2917 oldpid = rcu_access_pointer(vcpu->pid);
2918 if (unlikely(oldpid != task_pid(current))) {
2919 /* The thread running this VCPU changed. */
2922 r = kvm_arch_vcpu_run_pid_change(vcpu);
2926 newpid = get_task_pid(current, PIDTYPE_PID);
2927 rcu_assign_pointer(vcpu->pid, newpid);
2932 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2933 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2936 case KVM_GET_REGS: {
2937 struct kvm_regs *kvm_regs;
2940 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL_ACCOUNT);
2943 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2947 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2954 case KVM_SET_REGS: {
2955 struct kvm_regs *kvm_regs;
2958 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2959 if (IS_ERR(kvm_regs)) {
2960 r = PTR_ERR(kvm_regs);
2963 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2967 case KVM_GET_SREGS: {
2968 kvm_sregs = kzalloc(sizeof(struct kvm_sregs),
2969 GFP_KERNEL_ACCOUNT);
2973 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2977 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2982 case KVM_SET_SREGS: {
2983 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2984 if (IS_ERR(kvm_sregs)) {
2985 r = PTR_ERR(kvm_sregs);
2989 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2992 case KVM_GET_MP_STATE: {
2993 struct kvm_mp_state mp_state;
2995 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2999 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
3004 case KVM_SET_MP_STATE: {
3005 struct kvm_mp_state mp_state;
3008 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
3010 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
3013 case KVM_TRANSLATE: {
3014 struct kvm_translation tr;
3017 if (copy_from_user(&tr, argp, sizeof(tr)))
3019 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
3023 if (copy_to_user(argp, &tr, sizeof(tr)))
3028 case KVM_SET_GUEST_DEBUG: {
3029 struct kvm_guest_debug dbg;
3032 if (copy_from_user(&dbg, argp, sizeof(dbg)))
3034 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
3037 case KVM_SET_SIGNAL_MASK: {
3038 struct kvm_signal_mask __user *sigmask_arg = argp;
3039 struct kvm_signal_mask kvm_sigmask;
3040 sigset_t sigset, *p;
3045 if (copy_from_user(&kvm_sigmask, argp,
3046 sizeof(kvm_sigmask)))
3049 if (kvm_sigmask.len != sizeof(sigset))
3052 if (copy_from_user(&sigset, sigmask_arg->sigset,
3057 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
3061 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL_ACCOUNT);
3065 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
3069 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
3075 fpu = memdup_user(argp, sizeof(*fpu));
3081 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
3085 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
3088 mutex_unlock(&vcpu->mutex);
3094 #ifdef CONFIG_KVM_COMPAT
3095 static long kvm_vcpu_compat_ioctl(struct file *filp,
3096 unsigned int ioctl, unsigned long arg)
3098 struct kvm_vcpu *vcpu = filp->private_data;
3099 void __user *argp = compat_ptr(arg);
3102 if (vcpu->kvm->mm != current->mm)
3106 case KVM_SET_SIGNAL_MASK: {
3107 struct kvm_signal_mask __user *sigmask_arg = argp;
3108 struct kvm_signal_mask kvm_sigmask;
3113 if (copy_from_user(&kvm_sigmask, argp,
3114 sizeof(kvm_sigmask)))
3117 if (kvm_sigmask.len != sizeof(compat_sigset_t))
3120 if (get_compat_sigset(&sigset, (void *)sigmask_arg->sigset))
3122 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
3124 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
3128 r = kvm_vcpu_ioctl(filp, ioctl, arg);
3136 static int kvm_device_mmap(struct file *filp, struct vm_area_struct *vma)
3138 struct kvm_device *dev = filp->private_data;
3141 return dev->ops->mmap(dev, vma);
3146 static int kvm_device_ioctl_attr(struct kvm_device *dev,
3147 int (*accessor)(struct kvm_device *dev,
3148 struct kvm_device_attr *attr),
3151 struct kvm_device_attr attr;
3156 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
3159 return accessor(dev, &attr);
3162 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
3165 struct kvm_device *dev = filp->private_data;
3167 if (dev->kvm->mm != current->mm)
3171 case KVM_SET_DEVICE_ATTR:
3172 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
3173 case KVM_GET_DEVICE_ATTR:
3174 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
3175 case KVM_HAS_DEVICE_ATTR:
3176 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
3178 if (dev->ops->ioctl)
3179 return dev->ops->ioctl(dev, ioctl, arg);
3185 static int kvm_device_release(struct inode *inode, struct file *filp)
3187 struct kvm_device *dev = filp->private_data;
3188 struct kvm *kvm = dev->kvm;
3190 if (dev->ops->release) {
3191 mutex_lock(&kvm->lock);
3192 list_del(&dev->vm_node);
3193 dev->ops->release(dev);
3194 mutex_unlock(&kvm->lock);
3201 static const struct file_operations kvm_device_fops = {
3202 .unlocked_ioctl = kvm_device_ioctl,
3203 .release = kvm_device_release,
3204 KVM_COMPAT(kvm_device_ioctl),
3205 .mmap = kvm_device_mmap,
3208 struct kvm_device *kvm_device_from_filp(struct file *filp)
3210 if (filp->f_op != &kvm_device_fops)
3213 return filp->private_data;
3216 static const struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
3217 #ifdef CONFIG_KVM_MPIC
3218 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
3219 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
3223 int kvm_register_device_ops(const struct kvm_device_ops *ops, u32 type)
3225 if (type >= ARRAY_SIZE(kvm_device_ops_table))
3228 if (kvm_device_ops_table[type] != NULL)
3231 kvm_device_ops_table[type] = ops;
3235 void kvm_unregister_device_ops(u32 type)
3237 if (kvm_device_ops_table[type] != NULL)
3238 kvm_device_ops_table[type] = NULL;
3241 static int kvm_ioctl_create_device(struct kvm *kvm,
3242 struct kvm_create_device *cd)
3244 const struct kvm_device_ops *ops = NULL;
3245 struct kvm_device *dev;
3246 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
3250 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
3253 type = array_index_nospec(cd->type, ARRAY_SIZE(kvm_device_ops_table));
3254 ops = kvm_device_ops_table[type];
3261 dev = kzalloc(sizeof(*dev), GFP_KERNEL_ACCOUNT);
3268 mutex_lock(&kvm->lock);
3269 ret = ops->create(dev, type);
3271 mutex_unlock(&kvm->lock);
3275 list_add(&dev->vm_node, &kvm->devices);
3276 mutex_unlock(&kvm->lock);
3282 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
3284 kvm_put_kvm_no_destroy(kvm);
3285 mutex_lock(&kvm->lock);
3286 list_del(&dev->vm_node);
3287 mutex_unlock(&kvm->lock);
3296 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
3299 case KVM_CAP_USER_MEMORY:
3300 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
3301 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
3302 case KVM_CAP_INTERNAL_ERROR_DATA:
3303 #ifdef CONFIG_HAVE_KVM_MSI
3304 case KVM_CAP_SIGNAL_MSI:
3306 #ifdef CONFIG_HAVE_KVM_IRQFD
3308 case KVM_CAP_IRQFD_RESAMPLE:
3310 case KVM_CAP_IOEVENTFD_ANY_LENGTH:
3311 case KVM_CAP_CHECK_EXTENSION_VM:
3312 case KVM_CAP_ENABLE_CAP_VM:
3313 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3314 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2:
3317 #ifdef CONFIG_KVM_MMIO
3318 case KVM_CAP_COALESCED_MMIO:
3319 return KVM_COALESCED_MMIO_PAGE_OFFSET;
3320 case KVM_CAP_COALESCED_PIO:
3323 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3324 case KVM_CAP_IRQ_ROUTING:
3325 return KVM_MAX_IRQ_ROUTES;
3327 #if KVM_ADDRESS_SPACE_NUM > 1
3328 case KVM_CAP_MULTI_ADDRESS_SPACE:
3329 return KVM_ADDRESS_SPACE_NUM;
3331 case KVM_CAP_NR_MEMSLOTS:
3332 return KVM_USER_MEM_SLOTS;
3336 return kvm_vm_ioctl_check_extension(kvm, arg);
3339 int __attribute__((weak)) kvm_vm_ioctl_enable_cap(struct kvm *kvm,
3340 struct kvm_enable_cap *cap)
3345 static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm,
3346 struct kvm_enable_cap *cap)
3349 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3350 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2:
3351 if (cap->flags || (cap->args[0] & ~1))
3353 kvm->manual_dirty_log_protect = cap->args[0];
3357 return kvm_vm_ioctl_enable_cap(kvm, cap);
3361 static long kvm_vm_ioctl(struct file *filp,
3362 unsigned int ioctl, unsigned long arg)
3364 struct kvm *kvm = filp->private_data;
3365 void __user *argp = (void __user *)arg;
3368 if (kvm->mm != current->mm)
3371 case KVM_CREATE_VCPU:
3372 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
3374 case KVM_ENABLE_CAP: {
3375 struct kvm_enable_cap cap;
3378 if (copy_from_user(&cap, argp, sizeof(cap)))
3380 r = kvm_vm_ioctl_enable_cap_generic(kvm, &cap);
3383 case KVM_SET_USER_MEMORY_REGION: {
3384 struct kvm_userspace_memory_region kvm_userspace_mem;
3387 if (copy_from_user(&kvm_userspace_mem, argp,
3388 sizeof(kvm_userspace_mem)))
3391 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
3394 case KVM_GET_DIRTY_LOG: {
3395 struct kvm_dirty_log log;
3398 if (copy_from_user(&log, argp, sizeof(log)))
3400 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3403 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3404 case KVM_CLEAR_DIRTY_LOG: {
3405 struct kvm_clear_dirty_log log;
3408 if (copy_from_user(&log, argp, sizeof(log)))
3410 r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
3414 #ifdef CONFIG_KVM_MMIO
3415 case KVM_REGISTER_COALESCED_MMIO: {
3416 struct kvm_coalesced_mmio_zone zone;
3419 if (copy_from_user(&zone, argp, sizeof(zone)))
3421 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
3424 case KVM_UNREGISTER_COALESCED_MMIO: {
3425 struct kvm_coalesced_mmio_zone zone;
3428 if (copy_from_user(&zone, argp, sizeof(zone)))
3430 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
3435 struct kvm_irqfd data;
3438 if (copy_from_user(&data, argp, sizeof(data)))
3440 r = kvm_irqfd(kvm, &data);
3443 case KVM_IOEVENTFD: {
3444 struct kvm_ioeventfd data;
3447 if (copy_from_user(&data, argp, sizeof(data)))
3449 r = kvm_ioeventfd(kvm, &data);
3452 #ifdef CONFIG_HAVE_KVM_MSI
3453 case KVM_SIGNAL_MSI: {
3457 if (copy_from_user(&msi, argp, sizeof(msi)))
3459 r = kvm_send_userspace_msi(kvm, &msi);
3463 #ifdef __KVM_HAVE_IRQ_LINE
3464 case KVM_IRQ_LINE_STATUS:
3465 case KVM_IRQ_LINE: {
3466 struct kvm_irq_level irq_event;
3469 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
3472 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
3473 ioctl == KVM_IRQ_LINE_STATUS);
3478 if (ioctl == KVM_IRQ_LINE_STATUS) {
3479 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
3487 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3488 case KVM_SET_GSI_ROUTING: {
3489 struct kvm_irq_routing routing;
3490 struct kvm_irq_routing __user *urouting;
3491 struct kvm_irq_routing_entry *entries = NULL;
3494 if (copy_from_user(&routing, argp, sizeof(routing)))
3497 if (!kvm_arch_can_set_irq_routing(kvm))
3499 if (routing.nr > KVM_MAX_IRQ_ROUTES)
3505 entries = vmalloc(array_size(sizeof(*entries),
3511 if (copy_from_user(entries, urouting->entries,
3512 routing.nr * sizeof(*entries)))
3513 goto out_free_irq_routing;
3515 r = kvm_set_irq_routing(kvm, entries, routing.nr,
3517 out_free_irq_routing:
3521 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
3522 case KVM_CREATE_DEVICE: {
3523 struct kvm_create_device cd;
3526 if (copy_from_user(&cd, argp, sizeof(cd)))
3529 r = kvm_ioctl_create_device(kvm, &cd);
3534 if (copy_to_user(argp, &cd, sizeof(cd)))
3540 case KVM_CHECK_EXTENSION:
3541 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
3544 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
3550 #ifdef CONFIG_KVM_COMPAT
3551 struct compat_kvm_dirty_log {
3555 compat_uptr_t dirty_bitmap; /* one bit per page */
3560 static long kvm_vm_compat_ioctl(struct file *filp,
3561 unsigned int ioctl, unsigned long arg)
3563 struct kvm *kvm = filp->private_data;
3566 if (kvm->mm != current->mm)
3569 case KVM_GET_DIRTY_LOG: {
3570 struct compat_kvm_dirty_log compat_log;
3571 struct kvm_dirty_log log;
3573 if (copy_from_user(&compat_log, (void __user *)arg,
3574 sizeof(compat_log)))
3576 log.slot = compat_log.slot;
3577 log.padding1 = compat_log.padding1;
3578 log.padding2 = compat_log.padding2;
3579 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
3581 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3585 r = kvm_vm_ioctl(filp, ioctl, arg);
3591 static struct file_operations kvm_vm_fops = {
3592 .release = kvm_vm_release,
3593 .unlocked_ioctl = kvm_vm_ioctl,
3594 .llseek = noop_llseek,
3595 KVM_COMPAT(kvm_vm_compat_ioctl),
3598 static int kvm_dev_ioctl_create_vm(unsigned long type)
3604 kvm = kvm_create_vm(type);
3606 return PTR_ERR(kvm);
3607 #ifdef CONFIG_KVM_MMIO
3608 r = kvm_coalesced_mmio_init(kvm);
3612 r = get_unused_fd_flags(O_CLOEXEC);
3616 file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
3624 * Don't call kvm_put_kvm anymore at this point; file->f_op is
3625 * already set, with ->release() being kvm_vm_release(). In error
3626 * cases it will be called by the final fput(file) and will take
3627 * care of doing kvm_put_kvm(kvm).
3629 if (kvm_create_vm_debugfs(kvm, r) < 0) {
3634 kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm);
3636 fd_install(r, file);
3644 static long kvm_dev_ioctl(struct file *filp,
3645 unsigned int ioctl, unsigned long arg)
3650 case KVM_GET_API_VERSION:
3653 r = KVM_API_VERSION;
3656 r = kvm_dev_ioctl_create_vm(arg);
3658 case KVM_CHECK_EXTENSION:
3659 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
3661 case KVM_GET_VCPU_MMAP_SIZE:
3664 r = PAGE_SIZE; /* struct kvm_run */
3666 r += PAGE_SIZE; /* pio data page */
3668 #ifdef CONFIG_KVM_MMIO
3669 r += PAGE_SIZE; /* coalesced mmio ring page */
3672 case KVM_TRACE_ENABLE:
3673 case KVM_TRACE_PAUSE:
3674 case KVM_TRACE_DISABLE:
3678 return kvm_arch_dev_ioctl(filp, ioctl, arg);
3684 static struct file_operations kvm_chardev_ops = {
3685 .unlocked_ioctl = kvm_dev_ioctl,
3686 .llseek = noop_llseek,
3687 KVM_COMPAT(kvm_dev_ioctl),
3690 static struct miscdevice kvm_dev = {
3696 static void hardware_enable_nolock(void *junk)
3698 int cpu = raw_smp_processor_id();
3701 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
3704 cpumask_set_cpu(cpu, cpus_hardware_enabled);
3706 r = kvm_arch_hardware_enable();
3709 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3710 atomic_inc(&hardware_enable_failed);
3711 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
3715 static int kvm_starting_cpu(unsigned int cpu)
3717 raw_spin_lock(&kvm_count_lock);
3718 if (kvm_usage_count)
3719 hardware_enable_nolock(NULL);
3720 raw_spin_unlock(&kvm_count_lock);
3724 static void hardware_disable_nolock(void *junk)
3726 int cpu = raw_smp_processor_id();
3728 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
3730 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3731 kvm_arch_hardware_disable();
3734 static int kvm_dying_cpu(unsigned int cpu)
3736 raw_spin_lock(&kvm_count_lock);
3737 if (kvm_usage_count)
3738 hardware_disable_nolock(NULL);
3739 raw_spin_unlock(&kvm_count_lock);
3743 static void hardware_disable_all_nolock(void)
3745 BUG_ON(!kvm_usage_count);
3748 if (!kvm_usage_count)
3749 on_each_cpu(hardware_disable_nolock, NULL, 1);
3752 static void hardware_disable_all(void)
3754 raw_spin_lock(&kvm_count_lock);
3755 hardware_disable_all_nolock();
3756 raw_spin_unlock(&kvm_count_lock);
3759 static int hardware_enable_all(void)
3763 raw_spin_lock(&kvm_count_lock);
3766 if (kvm_usage_count == 1) {
3767 atomic_set(&hardware_enable_failed, 0);
3768 on_each_cpu(hardware_enable_nolock, NULL, 1);
3770 if (atomic_read(&hardware_enable_failed)) {
3771 hardware_disable_all_nolock();
3776 raw_spin_unlock(&kvm_count_lock);
3781 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3785 * Some (well, at least mine) BIOSes hang on reboot if
3788 * And Intel TXT required VMX off for all cpu when system shutdown.
3790 pr_info("kvm: exiting hardware virtualization\n");
3791 kvm_rebooting = true;
3792 on_each_cpu(hardware_disable_nolock, NULL, 1);
3796 static struct notifier_block kvm_reboot_notifier = {
3797 .notifier_call = kvm_reboot,
3801 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3805 for (i = 0; i < bus->dev_count; i++) {
3806 struct kvm_io_device *pos = bus->range[i].dev;
3808 kvm_iodevice_destructor(pos);
3813 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
3814 const struct kvm_io_range *r2)
3816 gpa_t addr1 = r1->addr;
3817 gpa_t addr2 = r2->addr;
3822 /* If r2->len == 0, match the exact address. If r2->len != 0,
3823 * accept any overlapping write. Any order is acceptable for
3824 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
3825 * we process all of them.
3838 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
3840 return kvm_io_bus_cmp(p1, p2);
3843 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
3844 gpa_t addr, int len)
3846 struct kvm_io_range *range, key;
3849 key = (struct kvm_io_range) {
3854 range = bsearch(&key, bus->range, bus->dev_count,
3855 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
3859 off = range - bus->range;
3861 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
3867 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3868 struct kvm_io_range *range, const void *val)
3872 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3876 while (idx < bus->dev_count &&
3877 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3878 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
3887 /* kvm_io_bus_write - called under kvm->slots_lock */
3888 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3889 int len, const void *val)
3891 struct kvm_io_bus *bus;
3892 struct kvm_io_range range;
3895 range = (struct kvm_io_range) {
3900 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3903 r = __kvm_io_bus_write(vcpu, bus, &range, val);
3904 return r < 0 ? r : 0;
3906 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
3908 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3909 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
3910 gpa_t addr, int len, const void *val, long cookie)
3912 struct kvm_io_bus *bus;
3913 struct kvm_io_range range;
3915 range = (struct kvm_io_range) {
3920 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3924 /* First try the device referenced by cookie. */
3925 if ((cookie >= 0) && (cookie < bus->dev_count) &&
3926 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
3927 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
3932 * cookie contained garbage; fall back to search and return the
3933 * correct cookie value.
3935 return __kvm_io_bus_write(vcpu, bus, &range, val);
3938 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3939 struct kvm_io_range *range, void *val)
3943 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3947 while (idx < bus->dev_count &&
3948 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3949 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
3958 /* kvm_io_bus_read - called under kvm->slots_lock */
3959 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3962 struct kvm_io_bus *bus;
3963 struct kvm_io_range range;
3966 range = (struct kvm_io_range) {
3971 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3974 r = __kvm_io_bus_read(vcpu, bus, &range, val);
3975 return r < 0 ? r : 0;
3978 /* Caller must hold slots_lock. */
3979 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3980 int len, struct kvm_io_device *dev)
3983 struct kvm_io_bus *new_bus, *bus;
3984 struct kvm_io_range range;
3986 bus = kvm_get_bus(kvm, bus_idx);
3990 /* exclude ioeventfd which is limited by maximum fd */
3991 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3994 new_bus = kmalloc(struct_size(bus, range, bus->dev_count + 1),
3995 GFP_KERNEL_ACCOUNT);
3999 range = (struct kvm_io_range) {
4005 for (i = 0; i < bus->dev_count; i++)
4006 if (kvm_io_bus_cmp(&bus->range[i], &range) > 0)
4009 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
4010 new_bus->dev_count++;
4011 new_bus->range[i] = range;
4012 memcpy(new_bus->range + i + 1, bus->range + i,
4013 (bus->dev_count - i) * sizeof(struct kvm_io_range));
4014 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
4015 synchronize_srcu_expedited(&kvm->srcu);
4021 /* Caller must hold slots_lock. */
4022 void kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
4023 struct kvm_io_device *dev)
4026 struct kvm_io_bus *new_bus, *bus;
4028 bus = kvm_get_bus(kvm, bus_idx);
4032 for (i = 0; i < bus->dev_count; i++)
4033 if (bus->range[i].dev == dev) {
4037 if (i == bus->dev_count)
4040 new_bus = kmalloc(struct_size(bus, range, bus->dev_count - 1),
4041 GFP_KERNEL_ACCOUNT);
4043 pr_err("kvm: failed to shrink bus, removing it completely\n");
4047 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
4048 new_bus->dev_count--;
4049 memcpy(new_bus->range + i, bus->range + i + 1,
4050 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
4053 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
4054 synchronize_srcu_expedited(&kvm->srcu);
4059 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
4062 struct kvm_io_bus *bus;
4063 int dev_idx, srcu_idx;
4064 struct kvm_io_device *iodev = NULL;
4066 srcu_idx = srcu_read_lock(&kvm->srcu);
4068 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
4072 dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
4076 iodev = bus->range[dev_idx].dev;
4079 srcu_read_unlock(&kvm->srcu, srcu_idx);
4083 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
4085 static int kvm_debugfs_open(struct inode *inode, struct file *file,
4086 int (*get)(void *, u64 *), int (*set)(void *, u64),
4089 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
4092 /* The debugfs files are a reference to the kvm struct which
4093 * is still valid when kvm_destroy_vm is called.
4094 * To avoid the race between open and the removal of the debugfs
4095 * directory we test against the users count.
4097 if (!refcount_inc_not_zero(&stat_data->kvm->users_count))
4100 if (simple_attr_open(inode, file, get,
4101 KVM_DBGFS_GET_MODE(stat_data->dbgfs_item) & 0222
4104 kvm_put_kvm(stat_data->kvm);
4111 static int kvm_debugfs_release(struct inode *inode, struct file *file)
4113 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
4116 simple_attr_release(inode, file);
4117 kvm_put_kvm(stat_data->kvm);
4122 static int kvm_get_stat_per_vm(struct kvm *kvm, size_t offset, u64 *val)
4124 *val = *(ulong *)((void *)kvm + offset);
4129 static int kvm_clear_stat_per_vm(struct kvm *kvm, size_t offset)
4131 *(ulong *)((void *)kvm + offset) = 0;
4136 static int kvm_get_stat_per_vcpu(struct kvm *kvm, size_t offset, u64 *val)
4139 struct kvm_vcpu *vcpu;
4143 kvm_for_each_vcpu(i, vcpu, kvm)
4144 *val += *(u64 *)((void *)vcpu + offset);
4149 static int kvm_clear_stat_per_vcpu(struct kvm *kvm, size_t offset)
4152 struct kvm_vcpu *vcpu;
4154 kvm_for_each_vcpu(i, vcpu, kvm)
4155 *(u64 *)((void *)vcpu + offset) = 0;
4160 static int kvm_stat_data_get(void *data, u64 *val)
4163 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
4165 switch (stat_data->dbgfs_item->kind) {
4167 r = kvm_get_stat_per_vm(stat_data->kvm,
4168 stat_data->dbgfs_item->offset, val);
4171 r = kvm_get_stat_per_vcpu(stat_data->kvm,
4172 stat_data->dbgfs_item->offset, val);
4179 static int kvm_stat_data_clear(void *data, u64 val)
4182 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
4187 switch (stat_data->dbgfs_item->kind) {
4189 r = kvm_clear_stat_per_vm(stat_data->kvm,
4190 stat_data->dbgfs_item->offset);
4193 r = kvm_clear_stat_per_vcpu(stat_data->kvm,
4194 stat_data->dbgfs_item->offset);
4201 static int kvm_stat_data_open(struct inode *inode, struct file *file)
4203 __simple_attr_check_format("%llu\n", 0ull);
4204 return kvm_debugfs_open(inode, file, kvm_stat_data_get,
4205 kvm_stat_data_clear, "%llu\n");
4208 static const struct file_operations stat_fops_per_vm = {
4209 .owner = THIS_MODULE,
4210 .open = kvm_stat_data_open,
4211 .release = kvm_debugfs_release,
4212 .read = simple_attr_read,
4213 .write = simple_attr_write,
4214 .llseek = no_llseek,
4217 static int vm_stat_get(void *_offset, u64 *val)
4219 unsigned offset = (long)_offset;
4224 mutex_lock(&kvm_lock);
4225 list_for_each_entry(kvm, &vm_list, vm_list) {
4226 kvm_get_stat_per_vm(kvm, offset, &tmp_val);
4229 mutex_unlock(&kvm_lock);
4233 static int vm_stat_clear(void *_offset, u64 val)
4235 unsigned offset = (long)_offset;
4241 mutex_lock(&kvm_lock);
4242 list_for_each_entry(kvm, &vm_list, vm_list) {
4243 kvm_clear_stat_per_vm(kvm, offset);
4245 mutex_unlock(&kvm_lock);
4250 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n");
4252 static int vcpu_stat_get(void *_offset, u64 *val)
4254 unsigned offset = (long)_offset;
4259 mutex_lock(&kvm_lock);
4260 list_for_each_entry(kvm, &vm_list, vm_list) {
4261 kvm_get_stat_per_vcpu(kvm, offset, &tmp_val);
4264 mutex_unlock(&kvm_lock);
4268 static int vcpu_stat_clear(void *_offset, u64 val)
4270 unsigned offset = (long)_offset;
4276 mutex_lock(&kvm_lock);
4277 list_for_each_entry(kvm, &vm_list, vm_list) {
4278 kvm_clear_stat_per_vcpu(kvm, offset);
4280 mutex_unlock(&kvm_lock);
4285 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear,
4288 static const struct file_operations *stat_fops[] = {
4289 [KVM_STAT_VCPU] = &vcpu_stat_fops,
4290 [KVM_STAT_VM] = &vm_stat_fops,
4293 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm)
4295 struct kobj_uevent_env *env;
4296 unsigned long long created, active;
4298 if (!kvm_dev.this_device || !kvm)
4301 mutex_lock(&kvm_lock);
4302 if (type == KVM_EVENT_CREATE_VM) {
4303 kvm_createvm_count++;
4305 } else if (type == KVM_EVENT_DESTROY_VM) {
4308 created = kvm_createvm_count;
4309 active = kvm_active_vms;
4310 mutex_unlock(&kvm_lock);
4312 env = kzalloc(sizeof(*env), GFP_KERNEL_ACCOUNT);
4316 add_uevent_var(env, "CREATED=%llu", created);
4317 add_uevent_var(env, "COUNT=%llu", active);
4319 if (type == KVM_EVENT_CREATE_VM) {
4320 add_uevent_var(env, "EVENT=create");
4321 kvm->userspace_pid = task_pid_nr(current);
4322 } else if (type == KVM_EVENT_DESTROY_VM) {
4323 add_uevent_var(env, "EVENT=destroy");
4325 add_uevent_var(env, "PID=%d", kvm->userspace_pid);
4327 if (!IS_ERR_OR_NULL(kvm->debugfs_dentry)) {
4328 char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL_ACCOUNT);
4331 tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX);
4333 add_uevent_var(env, "STATS_PATH=%s", tmp);
4337 /* no need for checks, since we are adding at most only 5 keys */
4338 env->envp[env->envp_idx++] = NULL;
4339 kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp);
4343 static void kvm_init_debug(void)
4345 struct kvm_stats_debugfs_item *p;
4347 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
4349 kvm_debugfs_num_entries = 0;
4350 for (p = debugfs_entries; p->name; ++p, kvm_debugfs_num_entries++) {
4351 debugfs_create_file(p->name, KVM_DBGFS_GET_MODE(p),
4352 kvm_debugfs_dir, (void *)(long)p->offset,
4353 stat_fops[p->kind]);
4357 static int kvm_suspend(void)
4359 if (kvm_usage_count)
4360 hardware_disable_nolock(NULL);
4364 static void kvm_resume(void)
4366 if (kvm_usage_count) {
4367 #ifdef CONFIG_LOCKDEP
4368 WARN_ON(lockdep_is_held(&kvm_count_lock));
4370 hardware_enable_nolock(NULL);
4374 static struct syscore_ops kvm_syscore_ops = {
4375 .suspend = kvm_suspend,
4376 .resume = kvm_resume,
4380 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
4382 return container_of(pn, struct kvm_vcpu, preempt_notifier);
4385 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
4387 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
4389 WRITE_ONCE(vcpu->preempted, false);
4390 WRITE_ONCE(vcpu->ready, false);
4392 __this_cpu_write(kvm_running_vcpu, vcpu);
4393 kvm_arch_sched_in(vcpu, cpu);
4394 kvm_arch_vcpu_load(vcpu, cpu);
4397 static void kvm_sched_out(struct preempt_notifier *pn,
4398 struct task_struct *next)
4400 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
4402 if (current->state == TASK_RUNNING) {
4403 WRITE_ONCE(vcpu->preempted, true);
4404 WRITE_ONCE(vcpu->ready, true);
4406 kvm_arch_vcpu_put(vcpu);
4407 __this_cpu_write(kvm_running_vcpu, NULL);
4411 * kvm_get_running_vcpu - get the vcpu running on the current CPU.
4413 * We can disable preemption locally around accessing the per-CPU variable,
4414 * and use the resolved vcpu pointer after enabling preemption again,
4415 * because even if the current thread is migrated to another CPU, reading
4416 * the per-CPU value later will give us the same value as we update the
4417 * per-CPU variable in the preempt notifier handlers.
4419 struct kvm_vcpu *kvm_get_running_vcpu(void)
4421 struct kvm_vcpu *vcpu;
4424 vcpu = __this_cpu_read(kvm_running_vcpu);
4431 * kvm_get_running_vcpus - get the per-CPU array of currently running vcpus.
4433 struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void)
4435 return &kvm_running_vcpu;
4438 static void check_processor_compat(void *rtn)
4440 *(int *)rtn = kvm_arch_check_processor_compat();
4443 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
4444 struct module *module)
4449 r = kvm_arch_init(opaque);
4454 * kvm_arch_init makes sure there's at most one caller
4455 * for architectures that support multiple implementations,
4456 * like intel and amd on x86.
4457 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
4458 * conflicts in case kvm is already setup for another implementation.
4460 r = kvm_irqfd_init();
4464 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
4469 r = kvm_arch_hardware_setup();
4473 for_each_online_cpu(cpu) {
4474 smp_call_function_single(cpu, check_processor_compat, &r, 1);
4479 r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting",
4480 kvm_starting_cpu, kvm_dying_cpu);
4483 register_reboot_notifier(&kvm_reboot_notifier);
4485 /* A kmem cache lets us meet the alignment requirements of fx_save. */
4487 vcpu_align = __alignof__(struct kvm_vcpu);
4489 kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align,
4491 offsetof(struct kvm_vcpu, arch),
4492 sizeof_field(struct kvm_vcpu, arch),
4494 if (!kvm_vcpu_cache) {
4499 r = kvm_async_pf_init();
4503 kvm_chardev_ops.owner = module;
4504 kvm_vm_fops.owner = module;
4505 kvm_vcpu_fops.owner = module;
4507 r = misc_register(&kvm_dev);
4509 pr_err("kvm: misc device register failed\n");
4513 register_syscore_ops(&kvm_syscore_ops);
4515 kvm_preempt_ops.sched_in = kvm_sched_in;
4516 kvm_preempt_ops.sched_out = kvm_sched_out;
4520 r = kvm_vfio_ops_init();
4526 kvm_async_pf_deinit();
4528 kmem_cache_destroy(kvm_vcpu_cache);
4530 unregister_reboot_notifier(&kvm_reboot_notifier);
4531 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4533 kvm_arch_hardware_unsetup();
4535 free_cpumask_var(cpus_hardware_enabled);
4543 EXPORT_SYMBOL_GPL(kvm_init);
4547 debugfs_remove_recursive(kvm_debugfs_dir);
4548 misc_deregister(&kvm_dev);
4549 kmem_cache_destroy(kvm_vcpu_cache);
4550 kvm_async_pf_deinit();
4551 unregister_syscore_ops(&kvm_syscore_ops);
4552 unregister_reboot_notifier(&kvm_reboot_notifier);
4553 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4554 on_each_cpu(hardware_disable_nolock, NULL, 1);
4555 kvm_arch_hardware_unsetup();
4558 free_cpumask_var(cpus_hardware_enabled);
4559 kvm_vfio_ops_exit();
4561 EXPORT_SYMBOL_GPL(kvm_exit);
4563 struct kvm_vm_worker_thread_context {
4565 struct task_struct *parent;
4566 struct completion init_done;
4567 kvm_vm_thread_fn_t thread_fn;
4572 static int kvm_vm_worker_thread(void *context)
4575 * The init_context is allocated on the stack of the parent thread, so
4576 * we have to locally copy anything that is needed beyond initialization
4578 struct kvm_vm_worker_thread_context *init_context = context;
4579 struct kvm *kvm = init_context->kvm;
4580 kvm_vm_thread_fn_t thread_fn = init_context->thread_fn;
4581 uintptr_t data = init_context->data;
4584 err = kthread_park(current);
4585 /* kthread_park(current) is never supposed to return an error */
4590 err = cgroup_attach_task_all(init_context->parent, current);
4592 kvm_err("%s: cgroup_attach_task_all failed with err %d\n",
4597 set_user_nice(current, task_nice(init_context->parent));
4600 init_context->err = err;
4601 complete(&init_context->init_done);
4602 init_context = NULL;
4607 /* Wait to be woken up by the spawner before proceeding. */
4610 if (!kthread_should_stop())
4611 err = thread_fn(kvm, data);
4616 int kvm_vm_create_worker_thread(struct kvm *kvm, kvm_vm_thread_fn_t thread_fn,
4617 uintptr_t data, const char *name,
4618 struct task_struct **thread_ptr)
4620 struct kvm_vm_worker_thread_context init_context = {};
4621 struct task_struct *thread;
4624 init_context.kvm = kvm;
4625 init_context.parent = current;
4626 init_context.thread_fn = thread_fn;
4627 init_context.data = data;
4628 init_completion(&init_context.init_done);
4630 thread = kthread_run(kvm_vm_worker_thread, &init_context,
4631 "%s-%d", name, task_pid_nr(current));
4633 return PTR_ERR(thread);
4635 /* kthread_run is never supposed to return NULL */
4636 WARN_ON(thread == NULL);
4638 wait_for_completion(&init_context.init_done);
4640 if (!init_context.err)
4641 *thread_ptr = thread;
4643 return init_context.err;