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 #define KVM_EVENT_CREATE_VM 0
153 #define KVM_EVENT_DESTROY_VM 1
154 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm);
155 static unsigned long long kvm_createvm_count;
156 static unsigned long long kvm_active_vms;
158 __weak int kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
159 unsigned long start, unsigned long end, bool blockable)
164 bool kvm_is_zone_device_pfn(kvm_pfn_t pfn)
167 * The metadata used by is_zone_device_page() to determine whether or
168 * not a page is ZONE_DEVICE is guaranteed to be valid if and only if
169 * the device has been pinned, e.g. by get_user_pages(). WARN if the
170 * page_count() is zero to help detect bad usage of this helper.
172 if (!pfn_valid(pfn) || WARN_ON_ONCE(!page_count(pfn_to_page(pfn))))
175 return is_zone_device_page(pfn_to_page(pfn));
178 bool kvm_is_reserved_pfn(kvm_pfn_t pfn)
181 * ZONE_DEVICE pages currently set PG_reserved, but from a refcounting
182 * perspective they are "normal" pages, albeit with slightly different
186 return PageReserved(pfn_to_page(pfn)) &&
188 !kvm_is_zone_device_pfn(pfn);
193 bool kvm_is_transparent_hugepage(kvm_pfn_t pfn)
195 struct page *page = pfn_to_page(pfn);
197 if (!PageTransCompoundMap(page))
200 return is_transparent_hugepage(compound_head(page));
204 * Switches to specified vcpu, until a matching vcpu_put()
206 void vcpu_load(struct kvm_vcpu *vcpu)
210 __this_cpu_write(kvm_running_vcpu, vcpu);
211 preempt_notifier_register(&vcpu->preempt_notifier);
212 kvm_arch_vcpu_load(vcpu, cpu);
215 EXPORT_SYMBOL_GPL(vcpu_load);
217 void vcpu_put(struct kvm_vcpu *vcpu)
220 kvm_arch_vcpu_put(vcpu);
221 preempt_notifier_unregister(&vcpu->preempt_notifier);
222 __this_cpu_write(kvm_running_vcpu, NULL);
225 EXPORT_SYMBOL_GPL(vcpu_put);
227 /* TODO: merge with kvm_arch_vcpu_should_kick */
228 static bool kvm_request_needs_ipi(struct kvm_vcpu *vcpu, unsigned req)
230 int mode = kvm_vcpu_exiting_guest_mode(vcpu);
233 * We need to wait for the VCPU to reenable interrupts and get out of
234 * READING_SHADOW_PAGE_TABLES mode.
236 if (req & KVM_REQUEST_WAIT)
237 return mode != OUTSIDE_GUEST_MODE;
240 * Need to kick a running VCPU, but otherwise there is nothing to do.
242 return mode == IN_GUEST_MODE;
245 static void ack_flush(void *_completed)
249 static inline bool kvm_kick_many_cpus(const struct cpumask *cpus, bool wait)
252 cpus = cpu_online_mask;
254 if (cpumask_empty(cpus))
257 smp_call_function_many(cpus, ack_flush, NULL, wait);
261 bool kvm_make_vcpus_request_mask(struct kvm *kvm, unsigned int req,
262 struct kvm_vcpu *except,
263 unsigned long *vcpu_bitmap, cpumask_var_t tmp)
266 struct kvm_vcpu *vcpu;
271 kvm_for_each_vcpu(i, vcpu, kvm) {
272 if ((vcpu_bitmap && !test_bit(i, vcpu_bitmap)) ||
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_except(struct kvm *kvm, unsigned int req,
294 struct kvm_vcpu *except)
299 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
301 called = kvm_make_vcpus_request_mask(kvm, req, except, NULL, cpus);
303 free_cpumask_var(cpus);
307 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
309 return kvm_make_all_cpus_request_except(kvm, req, NULL);
312 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
313 void kvm_flush_remote_tlbs(struct kvm *kvm)
316 * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in
317 * kvm_make_all_cpus_request.
319 long dirty_count = smp_load_acquire(&kvm->tlbs_dirty);
322 * We want to publish modifications to the page tables before reading
323 * mode. Pairs with a memory barrier in arch-specific code.
324 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
325 * and smp_mb in walk_shadow_page_lockless_begin/end.
326 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
328 * There is already an smp_mb__after_atomic() before
329 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
332 if (!kvm_arch_flush_remote_tlb(kvm)
333 || kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
334 ++kvm->stat.remote_tlb_flush;
335 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
337 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
340 void kvm_reload_remote_mmus(struct kvm *kvm)
342 kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
345 static void kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
347 mutex_init(&vcpu->mutex);
352 rcuwait_init(&vcpu->wait);
353 kvm_async_pf_vcpu_init(vcpu);
356 INIT_LIST_HEAD(&vcpu->blocked_vcpu_list);
358 kvm_vcpu_set_in_spin_loop(vcpu, false);
359 kvm_vcpu_set_dy_eligible(vcpu, false);
360 vcpu->preempted = false;
362 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
365 void kvm_vcpu_destroy(struct kvm_vcpu *vcpu)
367 kvm_arch_vcpu_destroy(vcpu);
370 * No need for rcu_read_lock as VCPU_RUN is the only place that changes
371 * the vcpu->pid pointer, and at destruction time all file descriptors
374 put_pid(rcu_dereference_protected(vcpu->pid, 1));
376 free_page((unsigned long)vcpu->run);
377 kmem_cache_free(kvm_vcpu_cache, vcpu);
379 EXPORT_SYMBOL_GPL(kvm_vcpu_destroy);
381 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
382 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
384 return container_of(mn, struct kvm, mmu_notifier);
387 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
388 struct mm_struct *mm,
389 unsigned long address,
392 struct kvm *kvm = mmu_notifier_to_kvm(mn);
395 idx = srcu_read_lock(&kvm->srcu);
396 spin_lock(&kvm->mmu_lock);
397 kvm->mmu_notifier_seq++;
399 if (kvm_set_spte_hva(kvm, address, pte))
400 kvm_flush_remote_tlbs(kvm);
402 spin_unlock(&kvm->mmu_lock);
403 srcu_read_unlock(&kvm->srcu, idx);
406 static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
407 const struct mmu_notifier_range *range)
409 struct kvm *kvm = mmu_notifier_to_kvm(mn);
410 int need_tlb_flush = 0, idx;
413 idx = srcu_read_lock(&kvm->srcu);
414 spin_lock(&kvm->mmu_lock);
416 * The count increase must become visible at unlock time as no
417 * spte can be established without taking the mmu_lock and
418 * count is also read inside the mmu_lock critical section.
420 kvm->mmu_notifier_count++;
421 need_tlb_flush = kvm_unmap_hva_range(kvm, range->start, range->end);
422 need_tlb_flush |= kvm->tlbs_dirty;
423 /* we've to flush the tlb before the pages can be freed */
425 kvm_flush_remote_tlbs(kvm);
427 spin_unlock(&kvm->mmu_lock);
429 ret = kvm_arch_mmu_notifier_invalidate_range(kvm, range->start,
431 mmu_notifier_range_blockable(range));
433 srcu_read_unlock(&kvm->srcu, idx);
438 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
439 const struct mmu_notifier_range *range)
441 struct kvm *kvm = mmu_notifier_to_kvm(mn);
443 spin_lock(&kvm->mmu_lock);
445 * This sequence increase will notify the kvm page fault that
446 * the page that is going to be mapped in the spte could have
449 kvm->mmu_notifier_seq++;
452 * The above sequence increase must be visible before the
453 * below count decrease, which is ensured by the smp_wmb above
454 * in conjunction with the smp_rmb in mmu_notifier_retry().
456 kvm->mmu_notifier_count--;
457 spin_unlock(&kvm->mmu_lock);
459 BUG_ON(kvm->mmu_notifier_count < 0);
462 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
463 struct mm_struct *mm,
467 struct kvm *kvm = mmu_notifier_to_kvm(mn);
470 idx = srcu_read_lock(&kvm->srcu);
471 spin_lock(&kvm->mmu_lock);
473 young = kvm_age_hva(kvm, start, end);
475 kvm_flush_remote_tlbs(kvm);
477 spin_unlock(&kvm->mmu_lock);
478 srcu_read_unlock(&kvm->srcu, idx);
483 static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
484 struct mm_struct *mm,
488 struct kvm *kvm = mmu_notifier_to_kvm(mn);
491 idx = srcu_read_lock(&kvm->srcu);
492 spin_lock(&kvm->mmu_lock);
494 * Even though we do not flush TLB, this will still adversely
495 * affect performance on pre-Haswell Intel EPT, where there is
496 * no EPT Access Bit to clear so that we have to tear down EPT
497 * tables instead. If we find this unacceptable, we can always
498 * add a parameter to kvm_age_hva so that it effectively doesn't
499 * do anything on clear_young.
501 * Also note that currently we never issue secondary TLB flushes
502 * from clear_young, leaving this job up to the regular system
503 * cadence. If we find this inaccurate, we might come up with a
504 * more sophisticated heuristic later.
506 young = kvm_age_hva(kvm, start, end);
507 spin_unlock(&kvm->mmu_lock);
508 srcu_read_unlock(&kvm->srcu, idx);
513 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
514 struct mm_struct *mm,
515 unsigned long address)
517 struct kvm *kvm = mmu_notifier_to_kvm(mn);
520 idx = srcu_read_lock(&kvm->srcu);
521 spin_lock(&kvm->mmu_lock);
522 young = kvm_test_age_hva(kvm, address);
523 spin_unlock(&kvm->mmu_lock);
524 srcu_read_unlock(&kvm->srcu, idx);
529 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
530 struct mm_struct *mm)
532 struct kvm *kvm = mmu_notifier_to_kvm(mn);
535 idx = srcu_read_lock(&kvm->srcu);
536 kvm_arch_flush_shadow_all(kvm);
537 srcu_read_unlock(&kvm->srcu, idx);
540 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
541 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
542 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
543 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
544 .clear_young = kvm_mmu_notifier_clear_young,
545 .test_young = kvm_mmu_notifier_test_young,
546 .change_pte = kvm_mmu_notifier_change_pte,
547 .release = kvm_mmu_notifier_release,
550 static int kvm_init_mmu_notifier(struct kvm *kvm)
552 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
553 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
556 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
558 static int kvm_init_mmu_notifier(struct kvm *kvm)
563 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
565 static struct kvm_memslots *kvm_alloc_memslots(void)
568 struct kvm_memslots *slots;
570 slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL_ACCOUNT);
574 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
575 slots->id_to_index[i] = -1;
580 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
582 if (!memslot->dirty_bitmap)
585 kvfree(memslot->dirty_bitmap);
586 memslot->dirty_bitmap = NULL;
589 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
591 kvm_destroy_dirty_bitmap(slot);
593 kvm_arch_free_memslot(kvm, slot);
599 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
601 struct kvm_memory_slot *memslot;
606 kvm_for_each_memslot(memslot, slots)
607 kvm_free_memslot(kvm, memslot);
612 static void kvm_destroy_vm_debugfs(struct kvm *kvm)
616 if (!kvm->debugfs_dentry)
619 debugfs_remove_recursive(kvm->debugfs_dentry);
621 if (kvm->debugfs_stat_data) {
622 for (i = 0; i < kvm_debugfs_num_entries; i++)
623 kfree(kvm->debugfs_stat_data[i]);
624 kfree(kvm->debugfs_stat_data);
628 static int kvm_create_vm_debugfs(struct kvm *kvm, int fd)
630 char dir_name[ITOA_MAX_LEN * 2];
631 struct kvm_stat_data *stat_data;
632 struct kvm_stats_debugfs_item *p;
634 if (!debugfs_initialized())
637 snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd);
638 kvm->debugfs_dentry = debugfs_create_dir(dir_name, kvm_debugfs_dir);
640 kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries,
641 sizeof(*kvm->debugfs_stat_data),
643 if (!kvm->debugfs_stat_data)
646 for (p = debugfs_entries; p->name; p++) {
647 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT);
651 stat_data->kvm = kvm;
652 stat_data->dbgfs_item = p;
653 kvm->debugfs_stat_data[p - debugfs_entries] = stat_data;
654 debugfs_create_file(p->name, KVM_DBGFS_GET_MODE(p),
655 kvm->debugfs_dentry, stat_data,
662 * Called after the VM is otherwise initialized, but just before adding it to
665 int __weak kvm_arch_post_init_vm(struct kvm *kvm)
671 * Called just after removing the VM from the vm_list, but before doing any
674 void __weak kvm_arch_pre_destroy_vm(struct kvm *kvm)
678 static struct kvm *kvm_create_vm(unsigned long type)
680 struct kvm *kvm = kvm_arch_alloc_vm();
685 return ERR_PTR(-ENOMEM);
687 spin_lock_init(&kvm->mmu_lock);
689 kvm->mm = current->mm;
690 kvm_eventfd_init(kvm);
691 mutex_init(&kvm->lock);
692 mutex_init(&kvm->irq_lock);
693 mutex_init(&kvm->slots_lock);
694 INIT_LIST_HEAD(&kvm->devices);
696 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
698 if (init_srcu_struct(&kvm->srcu))
699 goto out_err_no_srcu;
700 if (init_srcu_struct(&kvm->irq_srcu))
701 goto out_err_no_irq_srcu;
703 refcount_set(&kvm->users_count, 1);
704 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
705 struct kvm_memslots *slots = kvm_alloc_memslots();
708 goto out_err_no_arch_destroy_vm;
709 /* Generations must be different for each address space. */
710 slots->generation = i;
711 rcu_assign_pointer(kvm->memslots[i], slots);
714 for (i = 0; i < KVM_NR_BUSES; i++) {
715 rcu_assign_pointer(kvm->buses[i],
716 kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL_ACCOUNT));
718 goto out_err_no_arch_destroy_vm;
721 kvm->max_halt_poll_ns = halt_poll_ns;
723 r = kvm_arch_init_vm(kvm, type);
725 goto out_err_no_arch_destroy_vm;
727 r = hardware_enable_all();
729 goto out_err_no_disable;
731 #ifdef CONFIG_HAVE_KVM_IRQFD
732 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
735 r = kvm_init_mmu_notifier(kvm);
737 goto out_err_no_mmu_notifier;
739 r = kvm_arch_post_init_vm(kvm);
743 mutex_lock(&kvm_lock);
744 list_add(&kvm->vm_list, &vm_list);
745 mutex_unlock(&kvm_lock);
747 preempt_notifier_inc();
752 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
753 if (kvm->mmu_notifier.ops)
754 mmu_notifier_unregister(&kvm->mmu_notifier, current->mm);
756 out_err_no_mmu_notifier:
757 hardware_disable_all();
759 kvm_arch_destroy_vm(kvm);
760 out_err_no_arch_destroy_vm:
761 WARN_ON_ONCE(!refcount_dec_and_test(&kvm->users_count));
762 for (i = 0; i < KVM_NR_BUSES; i++)
763 kfree(kvm_get_bus(kvm, i));
764 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
765 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
766 cleanup_srcu_struct(&kvm->irq_srcu);
768 cleanup_srcu_struct(&kvm->srcu);
770 kvm_arch_free_vm(kvm);
775 static void kvm_destroy_devices(struct kvm *kvm)
777 struct kvm_device *dev, *tmp;
780 * We do not need to take the kvm->lock here, because nobody else
781 * has a reference to the struct kvm at this point and therefore
782 * cannot access the devices list anyhow.
784 list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
785 list_del(&dev->vm_node);
786 dev->ops->destroy(dev);
790 static void kvm_destroy_vm(struct kvm *kvm)
793 struct mm_struct *mm = kvm->mm;
795 kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm);
796 kvm_destroy_vm_debugfs(kvm);
797 kvm_arch_sync_events(kvm);
798 mutex_lock(&kvm_lock);
799 list_del(&kvm->vm_list);
800 mutex_unlock(&kvm_lock);
801 kvm_arch_pre_destroy_vm(kvm);
803 kvm_free_irq_routing(kvm);
804 for (i = 0; i < KVM_NR_BUSES; i++) {
805 struct kvm_io_bus *bus = kvm_get_bus(kvm, i);
808 kvm_io_bus_destroy(bus);
809 kvm->buses[i] = NULL;
811 kvm_coalesced_mmio_free(kvm);
812 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
813 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
815 kvm_arch_flush_shadow_all(kvm);
817 kvm_arch_destroy_vm(kvm);
818 kvm_destroy_devices(kvm);
819 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
820 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
821 cleanup_srcu_struct(&kvm->irq_srcu);
822 cleanup_srcu_struct(&kvm->srcu);
823 kvm_arch_free_vm(kvm);
824 preempt_notifier_dec();
825 hardware_disable_all();
829 void kvm_get_kvm(struct kvm *kvm)
831 refcount_inc(&kvm->users_count);
833 EXPORT_SYMBOL_GPL(kvm_get_kvm);
835 void kvm_put_kvm(struct kvm *kvm)
837 if (refcount_dec_and_test(&kvm->users_count))
840 EXPORT_SYMBOL_GPL(kvm_put_kvm);
843 * Used to put a reference that was taken on behalf of an object associated
844 * with a user-visible file descriptor, e.g. a vcpu or device, if installation
845 * of the new file descriptor fails and the reference cannot be transferred to
846 * its final owner. In such cases, the caller is still actively using @kvm and
847 * will fail miserably if the refcount unexpectedly hits zero.
849 void kvm_put_kvm_no_destroy(struct kvm *kvm)
851 WARN_ON(refcount_dec_and_test(&kvm->users_count));
853 EXPORT_SYMBOL_GPL(kvm_put_kvm_no_destroy);
855 static int kvm_vm_release(struct inode *inode, struct file *filp)
857 struct kvm *kvm = filp->private_data;
859 kvm_irqfd_release(kvm);
866 * Allocation size is twice as large as the actual dirty bitmap size.
867 * See kvm_vm_ioctl_get_dirty_log() why this is needed.
869 static int kvm_alloc_dirty_bitmap(struct kvm_memory_slot *memslot)
871 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
873 memslot->dirty_bitmap = kvzalloc(dirty_bytes, GFP_KERNEL_ACCOUNT);
874 if (!memslot->dirty_bitmap)
881 * Delete a memslot by decrementing the number of used slots and shifting all
882 * other entries in the array forward one spot.
884 static inline void kvm_memslot_delete(struct kvm_memslots *slots,
885 struct kvm_memory_slot *memslot)
887 struct kvm_memory_slot *mslots = slots->memslots;
890 if (WARN_ON(slots->id_to_index[memslot->id] == -1))
895 if (atomic_read(&slots->lru_slot) >= slots->used_slots)
896 atomic_set(&slots->lru_slot, 0);
898 for (i = slots->id_to_index[memslot->id]; i < slots->used_slots; i++) {
899 mslots[i] = mslots[i + 1];
900 slots->id_to_index[mslots[i].id] = i;
902 mslots[i] = *memslot;
903 slots->id_to_index[memslot->id] = -1;
907 * "Insert" a new memslot by incrementing the number of used slots. Returns
908 * the new slot's initial index into the memslots array.
910 static inline int kvm_memslot_insert_back(struct kvm_memslots *slots)
912 return slots->used_slots++;
916 * Move a changed memslot backwards in the array by shifting existing slots
917 * with a higher GFN toward the front of the array. Note, the changed memslot
918 * itself is not preserved in the array, i.e. not swapped at this time, only
919 * its new index into the array is tracked. Returns the changed memslot's
920 * current index into the memslots array.
922 static inline int kvm_memslot_move_backward(struct kvm_memslots *slots,
923 struct kvm_memory_slot *memslot)
925 struct kvm_memory_slot *mslots = slots->memslots;
928 if (WARN_ON_ONCE(slots->id_to_index[memslot->id] == -1) ||
929 WARN_ON_ONCE(!slots->used_slots))
933 * Move the target memslot backward in the array by shifting existing
934 * memslots with a higher GFN (than the target memslot) towards the
935 * front of the array.
937 for (i = slots->id_to_index[memslot->id]; i < slots->used_slots - 1; i++) {
938 if (memslot->base_gfn > mslots[i + 1].base_gfn)
941 WARN_ON_ONCE(memslot->base_gfn == mslots[i + 1].base_gfn);
943 /* Shift the next memslot forward one and update its index. */
944 mslots[i] = mslots[i + 1];
945 slots->id_to_index[mslots[i].id] = i;
951 * Move a changed memslot forwards in the array by shifting existing slots with
952 * a lower GFN toward the back of the array. Note, the changed memslot itself
953 * is not preserved in the array, i.e. not swapped at this time, only its new
954 * index into the array is tracked. Returns the changed memslot's final index
955 * into the memslots array.
957 static inline int kvm_memslot_move_forward(struct kvm_memslots *slots,
958 struct kvm_memory_slot *memslot,
961 struct kvm_memory_slot *mslots = slots->memslots;
964 for (i = start; i > 0; i--) {
965 if (memslot->base_gfn < mslots[i - 1].base_gfn)
968 WARN_ON_ONCE(memslot->base_gfn == mslots[i - 1].base_gfn);
970 /* Shift the next memslot back one and update its index. */
971 mslots[i] = mslots[i - 1];
972 slots->id_to_index[mslots[i].id] = i;
978 * Re-sort memslots based on their GFN to account for an added, deleted, or
979 * moved memslot. Sorting memslots by GFN allows using a binary search during
982 * IMPORTANT: Slots are sorted from highest GFN to lowest GFN! I.e. the entry
983 * at memslots[0] has the highest GFN.
985 * The sorting algorithm takes advantage of having initially sorted memslots
986 * and knowing the position of the changed memslot. Sorting is also optimized
987 * by not swapping the updated memslot and instead only shifting other memslots
988 * and tracking the new index for the update memslot. Only once its final
989 * index is known is the updated memslot copied into its position in the array.
991 * - When deleting a memslot, the deleted memslot simply needs to be moved to
992 * the end of the array.
994 * - When creating a memslot, the algorithm "inserts" the new memslot at the
995 * end of the array and then it forward to its correct location.
997 * - When moving a memslot, the algorithm first moves the updated memslot
998 * backward to handle the scenario where the memslot's GFN was changed to a
999 * lower value. update_memslots() then falls through and runs the same flow
1000 * as creating a memslot to move the memslot forward to handle the scenario
1001 * where its GFN was changed to a higher value.
1003 * Note, slots are sorted from highest->lowest instead of lowest->highest for
1004 * historical reasons. Originally, invalid memslots where denoted by having
1005 * GFN=0, thus sorting from highest->lowest naturally sorted invalid memslots
1006 * to the end of the array. The current algorithm uses dedicated logic to
1007 * delete a memslot and thus does not rely on invalid memslots having GFN=0.
1009 * The other historical motiviation for highest->lowest was to improve the
1010 * performance of memslot lookup. KVM originally used a linear search starting
1011 * at memslots[0]. On x86, the largest memslot usually has one of the highest,
1012 * if not *the* highest, GFN, as the bulk of the guest's RAM is located in a
1013 * single memslot above the 4gb boundary. As the largest memslot is also the
1014 * most likely to be referenced, sorting it to the front of the array was
1015 * advantageous. The current binary search starts from the middle of the array
1016 * and uses an LRU pointer to improve performance for all memslots and GFNs.
1018 static void update_memslots(struct kvm_memslots *slots,
1019 struct kvm_memory_slot *memslot,
1020 enum kvm_mr_change change)
1024 if (change == KVM_MR_DELETE) {
1025 kvm_memslot_delete(slots, memslot);
1027 if (change == KVM_MR_CREATE)
1028 i = kvm_memslot_insert_back(slots);
1030 i = kvm_memslot_move_backward(slots, memslot);
1031 i = kvm_memslot_move_forward(slots, memslot, i);
1034 * Copy the memslot to its new position in memslots and update
1035 * its index accordingly.
1037 slots->memslots[i] = *memslot;
1038 slots->id_to_index[memslot->id] = i;
1042 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
1044 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
1046 #ifdef __KVM_HAVE_READONLY_MEM
1047 valid_flags |= KVM_MEM_READONLY;
1050 if (mem->flags & ~valid_flags)
1056 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
1057 int as_id, struct kvm_memslots *slots)
1059 struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id);
1060 u64 gen = old_memslots->generation;
1062 WARN_ON(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS);
1063 slots->generation = gen | KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
1065 rcu_assign_pointer(kvm->memslots[as_id], slots);
1066 synchronize_srcu_expedited(&kvm->srcu);
1069 * Increment the new memslot generation a second time, dropping the
1070 * update in-progress flag and incrementing the generation based on
1071 * the number of address spaces. This provides a unique and easily
1072 * identifiable generation number while the memslots are in flux.
1074 gen = slots->generation & ~KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
1077 * Generations must be unique even across address spaces. We do not need
1078 * a global counter for that, instead the generation space is evenly split
1079 * across address spaces. For example, with two address spaces, address
1080 * space 0 will use generations 0, 2, 4, ... while address space 1 will
1081 * use generations 1, 3, 5, ...
1083 gen += KVM_ADDRESS_SPACE_NUM;
1085 kvm_arch_memslots_updated(kvm, gen);
1087 slots->generation = gen;
1089 return old_memslots;
1093 * Note, at a minimum, the current number of used slots must be allocated, even
1094 * when deleting a memslot, as we need a complete duplicate of the memslots for
1095 * use when invalidating a memslot prior to deleting/moving the memslot.
1097 static struct kvm_memslots *kvm_dup_memslots(struct kvm_memslots *old,
1098 enum kvm_mr_change change)
1100 struct kvm_memslots *slots;
1101 size_t old_size, new_size;
1103 old_size = sizeof(struct kvm_memslots) +
1104 (sizeof(struct kvm_memory_slot) * old->used_slots);
1106 if (change == KVM_MR_CREATE)
1107 new_size = old_size + sizeof(struct kvm_memory_slot);
1109 new_size = old_size;
1111 slots = kvzalloc(new_size, GFP_KERNEL_ACCOUNT);
1113 memcpy(slots, old, old_size);
1118 static int kvm_set_memslot(struct kvm *kvm,
1119 const struct kvm_userspace_memory_region *mem,
1120 struct kvm_memory_slot *old,
1121 struct kvm_memory_slot *new, int as_id,
1122 enum kvm_mr_change change)
1124 struct kvm_memory_slot *slot;
1125 struct kvm_memslots *slots;
1128 slots = kvm_dup_memslots(__kvm_memslots(kvm, as_id), change);
1132 if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) {
1134 * Note, the INVALID flag needs to be in the appropriate entry
1135 * in the freshly allocated memslots, not in @old or @new.
1137 slot = id_to_memslot(slots, old->id);
1138 slot->flags |= KVM_MEMSLOT_INVALID;
1141 * We can re-use the old memslots, the only difference from the
1142 * newly installed memslots is the invalid flag, which will get
1143 * dropped by update_memslots anyway. We'll also revert to the
1144 * old memslots if preparing the new memory region fails.
1146 slots = install_new_memslots(kvm, as_id, slots);
1148 /* From this point no new shadow pages pointing to a deleted,
1149 * or moved, memslot will be created.
1151 * validation of sp->gfn happens in:
1152 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1153 * - kvm_is_visible_gfn (mmu_check_root)
1155 kvm_arch_flush_shadow_memslot(kvm, slot);
1158 r = kvm_arch_prepare_memory_region(kvm, new, mem, change);
1162 update_memslots(slots, new, change);
1163 slots = install_new_memslots(kvm, as_id, slots);
1165 kvm_arch_commit_memory_region(kvm, mem, old, new, change);
1171 if (change == KVM_MR_DELETE || change == KVM_MR_MOVE)
1172 slots = install_new_memslots(kvm, as_id, slots);
1177 static int kvm_delete_memslot(struct kvm *kvm,
1178 const struct kvm_userspace_memory_region *mem,
1179 struct kvm_memory_slot *old, int as_id)
1181 struct kvm_memory_slot new;
1187 memset(&new, 0, sizeof(new));
1190 r = kvm_set_memslot(kvm, mem, old, &new, as_id, KVM_MR_DELETE);
1194 kvm_free_memslot(kvm, old);
1199 * Allocate some memory and give it an address in the guest physical address
1202 * Discontiguous memory is allowed, mostly for framebuffers.
1204 * Must be called holding kvm->slots_lock for write.
1206 int __kvm_set_memory_region(struct kvm *kvm,
1207 const struct kvm_userspace_memory_region *mem)
1209 struct kvm_memory_slot old, new;
1210 struct kvm_memory_slot *tmp;
1211 enum kvm_mr_change change;
1215 r = check_memory_region_flags(mem);
1219 as_id = mem->slot >> 16;
1220 id = (u16)mem->slot;
1222 /* General sanity checks */
1223 if (mem->memory_size & (PAGE_SIZE - 1))
1225 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
1227 /* We can read the guest memory with __xxx_user() later on. */
1228 if ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
1229 !access_ok((void __user *)(unsigned long)mem->userspace_addr,
1232 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
1234 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
1238 * Make a full copy of the old memslot, the pointer will become stale
1239 * when the memslots are re-sorted by update_memslots(), and the old
1240 * memslot needs to be referenced after calling update_memslots(), e.g.
1241 * to free its resources and for arch specific behavior.
1243 tmp = id_to_memslot(__kvm_memslots(kvm, as_id), id);
1248 memset(&old, 0, sizeof(old));
1252 if (!mem->memory_size)
1253 return kvm_delete_memslot(kvm, mem, &old, as_id);
1256 new.base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
1257 new.npages = mem->memory_size >> PAGE_SHIFT;
1258 new.flags = mem->flags;
1259 new.userspace_addr = mem->userspace_addr;
1261 if (new.npages > KVM_MEM_MAX_NR_PAGES)
1265 change = KVM_MR_CREATE;
1266 new.dirty_bitmap = NULL;
1267 memset(&new.arch, 0, sizeof(new.arch));
1268 } else { /* Modify an existing slot. */
1269 if ((new.userspace_addr != old.userspace_addr) ||
1270 (new.npages != old.npages) ||
1271 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
1274 if (new.base_gfn != old.base_gfn)
1275 change = KVM_MR_MOVE;
1276 else if (new.flags != old.flags)
1277 change = KVM_MR_FLAGS_ONLY;
1278 else /* Nothing to change. */
1281 /* Copy dirty_bitmap and arch from the current memslot. */
1282 new.dirty_bitmap = old.dirty_bitmap;
1283 memcpy(&new.arch, &old.arch, sizeof(new.arch));
1286 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
1287 /* Check for overlaps */
1288 kvm_for_each_memslot(tmp, __kvm_memslots(kvm, as_id)) {
1291 if (!((new.base_gfn + new.npages <= tmp->base_gfn) ||
1292 (new.base_gfn >= tmp->base_gfn + tmp->npages)))
1297 /* Allocate/free page dirty bitmap as needed */
1298 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
1299 new.dirty_bitmap = NULL;
1300 else if (!new.dirty_bitmap) {
1301 r = kvm_alloc_dirty_bitmap(&new);
1305 if (kvm_dirty_log_manual_protect_and_init_set(kvm))
1306 bitmap_set(new.dirty_bitmap, 0, new.npages);
1309 r = kvm_set_memslot(kvm, mem, &old, &new, as_id, change);
1313 if (old.dirty_bitmap && !new.dirty_bitmap)
1314 kvm_destroy_dirty_bitmap(&old);
1318 if (new.dirty_bitmap && !old.dirty_bitmap)
1319 kvm_destroy_dirty_bitmap(&new);
1322 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1324 int kvm_set_memory_region(struct kvm *kvm,
1325 const struct kvm_userspace_memory_region *mem)
1329 mutex_lock(&kvm->slots_lock);
1330 r = __kvm_set_memory_region(kvm, mem);
1331 mutex_unlock(&kvm->slots_lock);
1334 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1336 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1337 struct kvm_userspace_memory_region *mem)
1339 if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1342 return kvm_set_memory_region(kvm, mem);
1345 #ifndef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1347 * kvm_get_dirty_log - get a snapshot of dirty pages
1348 * @kvm: pointer to kvm instance
1349 * @log: slot id and address to which we copy the log
1350 * @is_dirty: set to '1' if any dirty pages were found
1351 * @memslot: set to the associated memslot, always valid on success
1353 int kvm_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log,
1354 int *is_dirty, struct kvm_memory_slot **memslot)
1356 struct kvm_memslots *slots;
1359 unsigned long any = 0;
1364 as_id = log->slot >> 16;
1365 id = (u16)log->slot;
1366 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1369 slots = __kvm_memslots(kvm, as_id);
1370 *memslot = id_to_memslot(slots, id);
1371 if (!(*memslot) || !(*memslot)->dirty_bitmap)
1374 kvm_arch_sync_dirty_log(kvm, *memslot);
1376 n = kvm_dirty_bitmap_bytes(*memslot);
1378 for (i = 0; !any && i < n/sizeof(long); ++i)
1379 any = (*memslot)->dirty_bitmap[i];
1381 if (copy_to_user(log->dirty_bitmap, (*memslot)->dirty_bitmap, n))
1388 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1390 #else /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
1392 * kvm_get_dirty_log_protect - get a snapshot of dirty pages
1393 * and reenable dirty page tracking for the corresponding pages.
1394 * @kvm: pointer to kvm instance
1395 * @log: slot id and address to which we copy the log
1397 * We need to keep it in mind that VCPU threads can write to the bitmap
1398 * concurrently. So, to avoid losing track of dirty pages we keep the
1401 * 1. Take a snapshot of the bit and clear it if needed.
1402 * 2. Write protect the corresponding page.
1403 * 3. Copy the snapshot to the userspace.
1404 * 4. Upon return caller flushes TLB's if needed.
1406 * Between 2 and 4, the guest may write to the page using the remaining TLB
1407 * entry. This is not a problem because the page is reported dirty using
1408 * the snapshot taken before and step 4 ensures that writes done after
1409 * exiting to userspace will be logged for the next call.
1412 static int kvm_get_dirty_log_protect(struct kvm *kvm, struct kvm_dirty_log *log)
1414 struct kvm_memslots *slots;
1415 struct kvm_memory_slot *memslot;
1418 unsigned long *dirty_bitmap;
1419 unsigned long *dirty_bitmap_buffer;
1422 as_id = log->slot >> 16;
1423 id = (u16)log->slot;
1424 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1427 slots = __kvm_memslots(kvm, as_id);
1428 memslot = id_to_memslot(slots, id);
1429 if (!memslot || !memslot->dirty_bitmap)
1432 dirty_bitmap = memslot->dirty_bitmap;
1434 kvm_arch_sync_dirty_log(kvm, memslot);
1436 n = kvm_dirty_bitmap_bytes(memslot);
1438 if (kvm->manual_dirty_log_protect) {
1440 * Unlike kvm_get_dirty_log, we always return false in *flush,
1441 * because no flush is needed until KVM_CLEAR_DIRTY_LOG. There
1442 * is some code duplication between this function and
1443 * kvm_get_dirty_log, but hopefully all architecture
1444 * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log
1445 * can be eliminated.
1447 dirty_bitmap_buffer = dirty_bitmap;
1449 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
1450 memset(dirty_bitmap_buffer, 0, n);
1452 spin_lock(&kvm->mmu_lock);
1453 for (i = 0; i < n / sizeof(long); i++) {
1457 if (!dirty_bitmap[i])
1461 mask = xchg(&dirty_bitmap[i], 0);
1462 dirty_bitmap_buffer[i] = mask;
1464 offset = i * BITS_PER_LONG;
1465 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1468 spin_unlock(&kvm->mmu_lock);
1472 kvm_arch_flush_remote_tlbs_memslot(kvm, memslot);
1474 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1481 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
1482 * @kvm: kvm instance
1483 * @log: slot id and address to which we copy the log
1485 * Steps 1-4 below provide general overview of dirty page logging. See
1486 * kvm_get_dirty_log_protect() function description for additional details.
1488 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
1489 * always flush the TLB (step 4) even if previous step failed and the dirty
1490 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
1491 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
1492 * writes will be marked dirty for next log read.
1494 * 1. Take a snapshot of the bit and clear it if needed.
1495 * 2. Write protect the corresponding page.
1496 * 3. Copy the snapshot to the userspace.
1497 * 4. Flush TLB's if needed.
1499 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1500 struct kvm_dirty_log *log)
1504 mutex_lock(&kvm->slots_lock);
1506 r = kvm_get_dirty_log_protect(kvm, log);
1508 mutex_unlock(&kvm->slots_lock);
1513 * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap
1514 * and reenable dirty page tracking for the corresponding pages.
1515 * @kvm: pointer to kvm instance
1516 * @log: slot id and address from which to fetch the bitmap of dirty pages
1518 static int kvm_clear_dirty_log_protect(struct kvm *kvm,
1519 struct kvm_clear_dirty_log *log)
1521 struct kvm_memslots *slots;
1522 struct kvm_memory_slot *memslot;
1526 unsigned long *dirty_bitmap;
1527 unsigned long *dirty_bitmap_buffer;
1530 as_id = log->slot >> 16;
1531 id = (u16)log->slot;
1532 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1535 if (log->first_page & 63)
1538 slots = __kvm_memslots(kvm, as_id);
1539 memslot = id_to_memslot(slots, id);
1540 if (!memslot || !memslot->dirty_bitmap)
1543 dirty_bitmap = memslot->dirty_bitmap;
1545 n = ALIGN(log->num_pages, BITS_PER_LONG) / 8;
1547 if (log->first_page > memslot->npages ||
1548 log->num_pages > memslot->npages - log->first_page ||
1549 (log->num_pages < memslot->npages - log->first_page && (log->num_pages & 63)))
1552 kvm_arch_sync_dirty_log(kvm, memslot);
1555 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
1556 if (copy_from_user(dirty_bitmap_buffer, log->dirty_bitmap, n))
1559 spin_lock(&kvm->mmu_lock);
1560 for (offset = log->first_page, i = offset / BITS_PER_LONG,
1561 n = DIV_ROUND_UP(log->num_pages, BITS_PER_LONG); n--;
1562 i++, offset += BITS_PER_LONG) {
1563 unsigned long mask = *dirty_bitmap_buffer++;
1564 atomic_long_t *p = (atomic_long_t *) &dirty_bitmap[i];
1568 mask &= atomic_long_fetch_andnot(mask, p);
1571 * mask contains the bits that really have been cleared. This
1572 * never includes any bits beyond the length of the memslot (if
1573 * the length is not aligned to 64 pages), therefore it is not
1574 * a problem if userspace sets them in log->dirty_bitmap.
1578 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1582 spin_unlock(&kvm->mmu_lock);
1585 kvm_arch_flush_remote_tlbs_memslot(kvm, memslot);
1590 static int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm,
1591 struct kvm_clear_dirty_log *log)
1595 mutex_lock(&kvm->slots_lock);
1597 r = kvm_clear_dirty_log_protect(kvm, log);
1599 mutex_unlock(&kvm->slots_lock);
1602 #endif /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
1604 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1606 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1608 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1610 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
1612 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn);
1614 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_memslot);
1616 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1618 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1620 return kvm_is_visible_memslot(memslot);
1622 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1624 unsigned long kvm_host_page_size(struct kvm_vcpu *vcpu, gfn_t gfn)
1626 struct vm_area_struct *vma;
1627 unsigned long addr, size;
1631 addr = kvm_vcpu_gfn_to_hva_prot(vcpu, gfn, NULL);
1632 if (kvm_is_error_hva(addr))
1635 down_read(¤t->mm->mmap_sem);
1636 vma = find_vma(current->mm, addr);
1640 size = vma_kernel_pagesize(vma);
1643 up_read(¤t->mm->mmap_sem);
1648 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1650 return slot->flags & KVM_MEM_READONLY;
1653 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1654 gfn_t *nr_pages, bool write)
1656 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1657 return KVM_HVA_ERR_BAD;
1659 if (memslot_is_readonly(slot) && write)
1660 return KVM_HVA_ERR_RO_BAD;
1663 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1665 return __gfn_to_hva_memslot(slot, gfn);
1668 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1671 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1674 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1677 return gfn_to_hva_many(slot, gfn, NULL);
1679 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1681 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1683 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1685 EXPORT_SYMBOL_GPL(gfn_to_hva);
1687 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
1689 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
1691 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
1694 * Return the hva of a @gfn and the R/W attribute if possible.
1696 * @slot: the kvm_memory_slot which contains @gfn
1697 * @gfn: the gfn to be translated
1698 * @writable: used to return the read/write attribute of the @slot if the hva
1699 * is valid and @writable is not NULL
1701 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1702 gfn_t gfn, bool *writable)
1704 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1706 if (!kvm_is_error_hva(hva) && writable)
1707 *writable = !memslot_is_readonly(slot);
1712 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1714 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1716 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1719 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
1721 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1723 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1726 static inline int check_user_page_hwpoison(unsigned long addr)
1728 int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
1730 rc = get_user_pages(addr, 1, flags, NULL, NULL);
1731 return rc == -EHWPOISON;
1735 * The fast path to get the writable pfn which will be stored in @pfn,
1736 * true indicates success, otherwise false is returned. It's also the
1737 * only part that runs if we can in atomic context.
1739 static bool hva_to_pfn_fast(unsigned long addr, bool write_fault,
1740 bool *writable, kvm_pfn_t *pfn)
1742 struct page *page[1];
1746 * Fast pin a writable pfn only if it is a write fault request
1747 * or the caller allows to map a writable pfn for a read fault
1750 if (!(write_fault || writable))
1753 npages = __get_user_pages_fast(addr, 1, 1, page);
1755 *pfn = page_to_pfn(page[0]);
1766 * The slow path to get the pfn of the specified host virtual address,
1767 * 1 indicates success, -errno is returned if error is detected.
1769 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1770 bool *writable, kvm_pfn_t *pfn)
1772 unsigned int flags = FOLL_HWPOISON;
1779 *writable = write_fault;
1782 flags |= FOLL_WRITE;
1784 flags |= FOLL_NOWAIT;
1786 npages = get_user_pages_unlocked(addr, 1, &page, flags);
1790 /* map read fault as writable if possible */
1791 if (unlikely(!write_fault) && writable) {
1794 if (__get_user_pages_fast(addr, 1, 1, &wpage) == 1) {
1800 *pfn = page_to_pfn(page);
1804 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1806 if (unlikely(!(vma->vm_flags & VM_READ)))
1809 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1815 static int hva_to_pfn_remapped(struct vm_area_struct *vma,
1816 unsigned long addr, bool *async,
1817 bool write_fault, bool *writable,
1823 r = follow_pfn(vma, addr, &pfn);
1826 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
1827 * not call the fault handler, so do it here.
1829 bool unlocked = false;
1830 r = fixup_user_fault(current, current->mm, addr,
1831 (write_fault ? FAULT_FLAG_WRITE : 0),
1838 r = follow_pfn(vma, addr, &pfn);
1848 * Get a reference here because callers of *hva_to_pfn* and
1849 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
1850 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
1851 * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will
1852 * simply do nothing for reserved pfns.
1854 * Whoever called remap_pfn_range is also going to call e.g.
1855 * unmap_mapping_range before the underlying pages are freed,
1856 * causing a call to our MMU notifier.
1865 * Pin guest page in memory and return its pfn.
1866 * @addr: host virtual address which maps memory to the guest
1867 * @atomic: whether this function can sleep
1868 * @async: whether this function need to wait IO complete if the
1869 * host page is not in the memory
1870 * @write_fault: whether we should get a writable host page
1871 * @writable: whether it allows to map a writable host page for !@write_fault
1873 * The function will map a writable host page for these two cases:
1874 * 1): @write_fault = true
1875 * 2): @write_fault = false && @writable, @writable will tell the caller
1876 * whether the mapping is writable.
1878 static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1879 bool write_fault, bool *writable)
1881 struct vm_area_struct *vma;
1885 /* we can do it either atomically or asynchronously, not both */
1886 BUG_ON(atomic && async);
1888 if (hva_to_pfn_fast(addr, write_fault, writable, &pfn))
1892 return KVM_PFN_ERR_FAULT;
1894 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1898 down_read(¤t->mm->mmap_sem);
1899 if (npages == -EHWPOISON ||
1900 (!async && check_user_page_hwpoison(addr))) {
1901 pfn = KVM_PFN_ERR_HWPOISON;
1906 vma = find_vma_intersection(current->mm, addr, addr + 1);
1909 pfn = KVM_PFN_ERR_FAULT;
1910 else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
1911 r = hva_to_pfn_remapped(vma, addr, async, write_fault, writable, &pfn);
1915 pfn = KVM_PFN_ERR_FAULT;
1917 if (async && vma_is_valid(vma, write_fault))
1919 pfn = KVM_PFN_ERR_FAULT;
1922 up_read(¤t->mm->mmap_sem);
1926 kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn,
1927 bool atomic, bool *async, bool write_fault,
1930 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1932 if (addr == KVM_HVA_ERR_RO_BAD) {
1935 return KVM_PFN_ERR_RO_FAULT;
1938 if (kvm_is_error_hva(addr)) {
1941 return KVM_PFN_NOSLOT;
1944 /* Do not map writable pfn in the readonly memslot. */
1945 if (writable && memslot_is_readonly(slot)) {
1950 return hva_to_pfn(addr, atomic, async, write_fault,
1953 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
1955 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1958 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
1959 write_fault, writable);
1961 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1963 kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1965 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1967 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
1969 kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1971 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1973 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1975 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
1977 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1979 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
1981 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1983 return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
1985 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1987 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
1989 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1991 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
1993 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1994 struct page **pages, int nr_pages)
1999 addr = gfn_to_hva_many(slot, gfn, &entry);
2000 if (kvm_is_error_hva(addr))
2003 if (entry < nr_pages)
2006 return __get_user_pages_fast(addr, nr_pages, 1, pages);
2008 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
2010 static struct page *kvm_pfn_to_page(kvm_pfn_t pfn)
2012 if (is_error_noslot_pfn(pfn))
2013 return KVM_ERR_PTR_BAD_PAGE;
2015 if (kvm_is_reserved_pfn(pfn)) {
2017 return KVM_ERR_PTR_BAD_PAGE;
2020 return pfn_to_page(pfn);
2023 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
2027 pfn = gfn_to_pfn(kvm, gfn);
2029 return kvm_pfn_to_page(pfn);
2031 EXPORT_SYMBOL_GPL(gfn_to_page);
2033 void kvm_release_pfn(kvm_pfn_t pfn, bool dirty, struct gfn_to_pfn_cache *cache)
2039 cache->pfn = cache->gfn = 0;
2042 kvm_release_pfn_dirty(pfn);
2044 kvm_release_pfn_clean(pfn);
2047 static void kvm_cache_gfn_to_pfn(struct kvm_memory_slot *slot, gfn_t gfn,
2048 struct gfn_to_pfn_cache *cache, u64 gen)
2050 kvm_release_pfn(cache->pfn, cache->dirty, cache);
2052 cache->pfn = gfn_to_pfn_memslot(slot, gfn);
2054 cache->dirty = false;
2055 cache->generation = gen;
2058 static int __kvm_map_gfn(struct kvm_memslots *slots, gfn_t gfn,
2059 struct kvm_host_map *map,
2060 struct gfn_to_pfn_cache *cache,
2065 struct page *page = KVM_UNMAPPED_PAGE;
2066 struct kvm_memory_slot *slot = __gfn_to_memslot(slots, gfn);
2067 u64 gen = slots->generation;
2073 if (!cache->pfn || cache->gfn != gfn ||
2074 cache->generation != gen) {
2077 kvm_cache_gfn_to_pfn(slot, gfn, cache, gen);
2083 pfn = gfn_to_pfn_memslot(slot, gfn);
2085 if (is_error_noslot_pfn(pfn))
2088 if (pfn_valid(pfn)) {
2089 page = pfn_to_page(pfn);
2091 hva = kmap_atomic(page);
2094 #ifdef CONFIG_HAS_IOMEM
2095 } else if (!atomic) {
2096 hva = memremap(pfn_to_hpa(pfn), PAGE_SIZE, MEMREMAP_WB);
2113 int kvm_map_gfn(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map,
2114 struct gfn_to_pfn_cache *cache, bool atomic)
2116 return __kvm_map_gfn(kvm_memslots(vcpu->kvm), gfn, map,
2119 EXPORT_SYMBOL_GPL(kvm_map_gfn);
2121 int kvm_vcpu_map(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map)
2123 return __kvm_map_gfn(kvm_vcpu_memslots(vcpu), gfn, map,
2126 EXPORT_SYMBOL_GPL(kvm_vcpu_map);
2128 static void __kvm_unmap_gfn(struct kvm_memory_slot *memslot,
2129 struct kvm_host_map *map,
2130 struct gfn_to_pfn_cache *cache,
2131 bool dirty, bool atomic)
2139 if (map->page != KVM_UNMAPPED_PAGE) {
2141 kunmap_atomic(map->hva);
2145 #ifdef CONFIG_HAS_IOMEM
2149 WARN_ONCE(1, "Unexpected unmapping in atomic context");
2153 mark_page_dirty_in_slot(memslot, map->gfn);
2156 cache->dirty |= dirty;
2158 kvm_release_pfn(map->pfn, dirty, NULL);
2164 int kvm_unmap_gfn(struct kvm_vcpu *vcpu, struct kvm_host_map *map,
2165 struct gfn_to_pfn_cache *cache, bool dirty, bool atomic)
2167 __kvm_unmap_gfn(gfn_to_memslot(vcpu->kvm, map->gfn), map,
2168 cache, dirty, atomic);
2171 EXPORT_SYMBOL_GPL(kvm_unmap_gfn);
2173 void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map, bool dirty)
2175 __kvm_unmap_gfn(kvm_vcpu_gfn_to_memslot(vcpu, map->gfn), map, NULL,
2178 EXPORT_SYMBOL_GPL(kvm_vcpu_unmap);
2180 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
2184 pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
2186 return kvm_pfn_to_page(pfn);
2188 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
2190 void kvm_release_page_clean(struct page *page)
2192 WARN_ON(is_error_page(page));
2194 kvm_release_pfn_clean(page_to_pfn(page));
2196 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
2198 void kvm_release_pfn_clean(kvm_pfn_t pfn)
2200 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
2201 put_page(pfn_to_page(pfn));
2203 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
2205 void kvm_release_page_dirty(struct page *page)
2207 WARN_ON(is_error_page(page));
2209 kvm_release_pfn_dirty(page_to_pfn(page));
2211 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
2213 void kvm_release_pfn_dirty(kvm_pfn_t pfn)
2215 kvm_set_pfn_dirty(pfn);
2216 kvm_release_pfn_clean(pfn);
2218 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
2220 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
2222 if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn))
2223 SetPageDirty(pfn_to_page(pfn));
2225 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
2227 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
2229 if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn))
2230 mark_page_accessed(pfn_to_page(pfn));
2232 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
2234 void kvm_get_pfn(kvm_pfn_t pfn)
2236 if (!kvm_is_reserved_pfn(pfn))
2237 get_page(pfn_to_page(pfn));
2239 EXPORT_SYMBOL_GPL(kvm_get_pfn);
2241 static int next_segment(unsigned long len, int offset)
2243 if (len > PAGE_SIZE - offset)
2244 return PAGE_SIZE - offset;
2249 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
2250 void *data, int offset, int len)
2255 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
2256 if (kvm_is_error_hva(addr))
2258 r = __copy_from_user(data, (void __user *)addr + offset, len);
2264 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
2267 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2269 return __kvm_read_guest_page(slot, gfn, data, offset, len);
2271 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
2273 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
2274 int offset, int len)
2276 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2278 return __kvm_read_guest_page(slot, gfn, data, offset, len);
2280 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
2282 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
2284 gfn_t gfn = gpa >> PAGE_SHIFT;
2286 int offset = offset_in_page(gpa);
2289 while ((seg = next_segment(len, offset)) != 0) {
2290 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
2300 EXPORT_SYMBOL_GPL(kvm_read_guest);
2302 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
2304 gfn_t gfn = gpa >> PAGE_SHIFT;
2306 int offset = offset_in_page(gpa);
2309 while ((seg = next_segment(len, offset)) != 0) {
2310 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
2320 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
2322 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
2323 void *data, int offset, unsigned long len)
2328 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
2329 if (kvm_is_error_hva(addr))
2331 pagefault_disable();
2332 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
2339 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
2340 void *data, unsigned long len)
2342 gfn_t gfn = gpa >> PAGE_SHIFT;
2343 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2344 int offset = offset_in_page(gpa);
2346 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
2348 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
2350 static int __kvm_write_guest_page(struct kvm_memory_slot *memslot, gfn_t gfn,
2351 const void *data, int offset, int len)
2356 addr = gfn_to_hva_memslot(memslot, gfn);
2357 if (kvm_is_error_hva(addr))
2359 r = __copy_to_user((void __user *)addr + offset, data, len);
2362 mark_page_dirty_in_slot(memslot, gfn);
2366 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
2367 const void *data, int offset, int len)
2369 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2371 return __kvm_write_guest_page(slot, gfn, data, offset, len);
2373 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
2375 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
2376 const void *data, int offset, int len)
2378 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2380 return __kvm_write_guest_page(slot, gfn, data, offset, len);
2382 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
2384 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
2387 gfn_t gfn = gpa >> PAGE_SHIFT;
2389 int offset = offset_in_page(gpa);
2392 while ((seg = next_segment(len, offset)) != 0) {
2393 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
2403 EXPORT_SYMBOL_GPL(kvm_write_guest);
2405 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
2408 gfn_t gfn = gpa >> PAGE_SHIFT;
2410 int offset = offset_in_page(gpa);
2413 while ((seg = next_segment(len, offset)) != 0) {
2414 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
2424 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
2426 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
2427 struct gfn_to_hva_cache *ghc,
2428 gpa_t gpa, unsigned long len)
2430 int offset = offset_in_page(gpa);
2431 gfn_t start_gfn = gpa >> PAGE_SHIFT;
2432 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
2433 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
2434 gfn_t nr_pages_avail;
2436 /* Update ghc->generation before performing any error checks. */
2437 ghc->generation = slots->generation;
2439 if (start_gfn > end_gfn) {
2440 ghc->hva = KVM_HVA_ERR_BAD;
2445 * If the requested region crosses two memslots, we still
2446 * verify that the entire region is valid here.
2448 for ( ; start_gfn <= end_gfn; start_gfn += nr_pages_avail) {
2449 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
2450 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
2452 if (kvm_is_error_hva(ghc->hva))
2456 /* Use the slow path for cross page reads and writes. */
2457 if (nr_pages_needed == 1)
2460 ghc->memslot = NULL;
2467 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2468 gpa_t gpa, unsigned long len)
2470 struct kvm_memslots *slots = kvm_memslots(kvm);
2471 return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len);
2473 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
2475 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2476 void *data, unsigned int offset,
2479 struct kvm_memslots *slots = kvm_memslots(kvm);
2481 gpa_t gpa = ghc->gpa + offset;
2483 BUG_ON(len + offset > ghc->len);
2485 if (slots->generation != ghc->generation) {
2486 if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len))
2490 if (kvm_is_error_hva(ghc->hva))
2493 if (unlikely(!ghc->memslot))
2494 return kvm_write_guest(kvm, gpa, data, len);
2496 r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
2499 mark_page_dirty_in_slot(ghc->memslot, gpa >> PAGE_SHIFT);
2503 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached);
2505 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2506 void *data, unsigned long len)
2508 return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len);
2510 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
2512 int kvm_read_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2513 void *data, unsigned int offset,
2516 struct kvm_memslots *slots = kvm_memslots(kvm);
2518 gpa_t gpa = ghc->gpa + offset;
2520 BUG_ON(len + offset > ghc->len);
2522 if (slots->generation != ghc->generation) {
2523 if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len))
2527 if (kvm_is_error_hva(ghc->hva))
2530 if (unlikely(!ghc->memslot))
2531 return kvm_read_guest(kvm, gpa, data, len);
2533 r = __copy_from_user(data, (void __user *)ghc->hva + offset, len);
2539 EXPORT_SYMBOL_GPL(kvm_read_guest_offset_cached);
2541 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2542 void *data, unsigned long len)
2544 return kvm_read_guest_offset_cached(kvm, ghc, data, 0, len);
2546 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
2548 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
2550 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
2552 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
2554 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
2556 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
2558 gfn_t gfn = gpa >> PAGE_SHIFT;
2560 int offset = offset_in_page(gpa);
2563 while ((seg = next_segment(len, offset)) != 0) {
2564 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
2573 EXPORT_SYMBOL_GPL(kvm_clear_guest);
2575 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot,
2578 if (memslot && memslot->dirty_bitmap) {
2579 unsigned long rel_gfn = gfn - memslot->base_gfn;
2581 set_bit_le(rel_gfn, memslot->dirty_bitmap);
2585 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
2587 struct kvm_memory_slot *memslot;
2589 memslot = gfn_to_memslot(kvm, gfn);
2590 mark_page_dirty_in_slot(memslot, gfn);
2592 EXPORT_SYMBOL_GPL(mark_page_dirty);
2594 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
2596 struct kvm_memory_slot *memslot;
2598 memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2599 mark_page_dirty_in_slot(memslot, gfn);
2601 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
2603 void kvm_sigset_activate(struct kvm_vcpu *vcpu)
2605 if (!vcpu->sigset_active)
2609 * This does a lockless modification of ->real_blocked, which is fine
2610 * because, only current can change ->real_blocked and all readers of
2611 * ->real_blocked don't care as long ->real_blocked is always a subset
2614 sigprocmask(SIG_SETMASK, &vcpu->sigset, ¤t->real_blocked);
2617 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu)
2619 if (!vcpu->sigset_active)
2622 sigprocmask(SIG_SETMASK, ¤t->real_blocked, NULL);
2623 sigemptyset(¤t->real_blocked);
2626 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
2628 unsigned int old, val, grow, grow_start;
2630 old = val = vcpu->halt_poll_ns;
2631 grow_start = READ_ONCE(halt_poll_ns_grow_start);
2632 grow = READ_ONCE(halt_poll_ns_grow);
2637 if (val < grow_start)
2640 if (val > halt_poll_ns)
2643 vcpu->halt_poll_ns = val;
2645 trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
2648 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
2650 unsigned int old, val, shrink;
2652 old = val = vcpu->halt_poll_ns;
2653 shrink = READ_ONCE(halt_poll_ns_shrink);
2659 vcpu->halt_poll_ns = val;
2660 trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
2663 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
2666 int idx = srcu_read_lock(&vcpu->kvm->srcu);
2668 if (kvm_arch_vcpu_runnable(vcpu)) {
2669 kvm_make_request(KVM_REQ_UNHALT, vcpu);
2672 if (kvm_cpu_has_pending_timer(vcpu))
2674 if (signal_pending(current))
2679 srcu_read_unlock(&vcpu->kvm->srcu, idx);
2684 update_halt_poll_stats(struct kvm_vcpu *vcpu, u64 poll_ns, bool waited)
2687 vcpu->stat.halt_poll_fail_ns += poll_ns;
2689 vcpu->stat.halt_poll_success_ns += poll_ns;
2693 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2695 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
2697 ktime_t start, cur, poll_end;
2698 bool waited = false;
2701 kvm_arch_vcpu_blocking(vcpu);
2703 start = cur = poll_end = ktime_get();
2704 if (vcpu->halt_poll_ns && !kvm_arch_no_poll(vcpu)) {
2705 ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns);
2707 ++vcpu->stat.halt_attempted_poll;
2710 * This sets KVM_REQ_UNHALT if an interrupt
2713 if (kvm_vcpu_check_block(vcpu) < 0) {
2714 ++vcpu->stat.halt_successful_poll;
2715 if (!vcpu_valid_wakeup(vcpu))
2716 ++vcpu->stat.halt_poll_invalid;
2719 poll_end = cur = ktime_get();
2720 } while (single_task_running() && ktime_before(cur, stop));
2723 prepare_to_rcuwait(&vcpu->wait);
2725 set_current_state(TASK_INTERRUPTIBLE);
2727 if (kvm_vcpu_check_block(vcpu) < 0)
2733 finish_rcuwait(&vcpu->wait);
2736 kvm_arch_vcpu_unblocking(vcpu);
2737 block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
2739 update_halt_poll_stats(
2740 vcpu, ktime_to_ns(ktime_sub(poll_end, start)), waited);
2742 if (!kvm_arch_no_poll(vcpu)) {
2743 if (!vcpu_valid_wakeup(vcpu)) {
2744 shrink_halt_poll_ns(vcpu);
2745 } else if (vcpu->kvm->max_halt_poll_ns) {
2746 if (block_ns <= vcpu->halt_poll_ns)
2748 /* we had a long block, shrink polling */
2749 else if (vcpu->halt_poll_ns &&
2750 block_ns > vcpu->kvm->max_halt_poll_ns)
2751 shrink_halt_poll_ns(vcpu);
2752 /* we had a short halt and our poll time is too small */
2753 else if (vcpu->halt_poll_ns < vcpu->kvm->max_halt_poll_ns &&
2754 block_ns < vcpu->kvm->max_halt_poll_ns)
2755 grow_halt_poll_ns(vcpu);
2757 vcpu->halt_poll_ns = 0;
2761 trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu));
2762 kvm_arch_vcpu_block_finish(vcpu);
2764 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
2766 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
2768 struct rcuwait *waitp;
2770 waitp = kvm_arch_vcpu_get_wait(vcpu);
2771 if (rcuwait_wake_up(waitp)) {
2772 WRITE_ONCE(vcpu->ready, true);
2773 ++vcpu->stat.halt_wakeup;
2779 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
2783 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2785 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
2788 int cpu = vcpu->cpu;
2790 if (kvm_vcpu_wake_up(vcpu))
2794 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
2795 if (kvm_arch_vcpu_should_kick(vcpu))
2796 smp_send_reschedule(cpu);
2799 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
2800 #endif /* !CONFIG_S390 */
2802 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
2805 struct task_struct *task = NULL;
2809 pid = rcu_dereference(target->pid);
2811 task = get_pid_task(pid, PIDTYPE_PID);
2815 ret = yield_to(task, 1);
2816 put_task_struct(task);
2820 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
2823 * Helper that checks whether a VCPU is eligible for directed yield.
2824 * Most eligible candidate to yield is decided by following heuristics:
2826 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2827 * (preempted lock holder), indicated by @in_spin_loop.
2828 * Set at the beginning and cleared at the end of interception/PLE handler.
2830 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2831 * chance last time (mostly it has become eligible now since we have probably
2832 * yielded to lockholder in last iteration. This is done by toggling
2833 * @dy_eligible each time a VCPU checked for eligibility.)
2835 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2836 * to preempted lock-holder could result in wrong VCPU selection and CPU
2837 * burning. Giving priority for a potential lock-holder increases lock
2840 * Since algorithm is based on heuristics, accessing another VCPU data without
2841 * locking does not harm. It may result in trying to yield to same VCPU, fail
2842 * and continue with next VCPU and so on.
2844 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
2846 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2849 eligible = !vcpu->spin_loop.in_spin_loop ||
2850 vcpu->spin_loop.dy_eligible;
2852 if (vcpu->spin_loop.in_spin_loop)
2853 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
2862 * Unlike kvm_arch_vcpu_runnable, this function is called outside
2863 * a vcpu_load/vcpu_put pair. However, for most architectures
2864 * kvm_arch_vcpu_runnable does not require vcpu_load.
2866 bool __weak kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
2868 return kvm_arch_vcpu_runnable(vcpu);
2871 static bool vcpu_dy_runnable(struct kvm_vcpu *vcpu)
2873 if (kvm_arch_dy_runnable(vcpu))
2876 #ifdef CONFIG_KVM_ASYNC_PF
2877 if (!list_empty_careful(&vcpu->async_pf.done))
2884 void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode)
2886 struct kvm *kvm = me->kvm;
2887 struct kvm_vcpu *vcpu;
2888 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
2894 kvm_vcpu_set_in_spin_loop(me, true);
2896 * We boost the priority of a VCPU that is runnable but not
2897 * currently running, because it got preempted by something
2898 * else and called schedule in __vcpu_run. Hopefully that
2899 * VCPU is holding the lock that we need and will release it.
2900 * We approximate round-robin by starting at the last boosted VCPU.
2902 for (pass = 0; pass < 2 && !yielded && try; pass++) {
2903 kvm_for_each_vcpu(i, vcpu, kvm) {
2904 if (!pass && i <= last_boosted_vcpu) {
2905 i = last_boosted_vcpu;
2907 } else if (pass && i > last_boosted_vcpu)
2909 if (!READ_ONCE(vcpu->ready))
2913 if (rcuwait_active(&vcpu->wait) &&
2914 !vcpu_dy_runnable(vcpu))
2916 if (READ_ONCE(vcpu->preempted) && yield_to_kernel_mode &&
2917 !kvm_arch_vcpu_in_kernel(vcpu))
2919 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
2922 yielded = kvm_vcpu_yield_to(vcpu);
2924 kvm->last_boosted_vcpu = i;
2926 } else if (yielded < 0) {
2933 kvm_vcpu_set_in_spin_loop(me, false);
2935 /* Ensure vcpu is not eligible during next spinloop */
2936 kvm_vcpu_set_dy_eligible(me, false);
2938 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
2940 static vm_fault_t kvm_vcpu_fault(struct vm_fault *vmf)
2942 struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data;
2945 if (vmf->pgoff == 0)
2946 page = virt_to_page(vcpu->run);
2948 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
2949 page = virt_to_page(vcpu->arch.pio_data);
2951 #ifdef CONFIG_KVM_MMIO
2952 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
2953 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
2956 return kvm_arch_vcpu_fault(vcpu, vmf);
2962 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
2963 .fault = kvm_vcpu_fault,
2966 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2968 vma->vm_ops = &kvm_vcpu_vm_ops;
2972 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2974 struct kvm_vcpu *vcpu = filp->private_data;
2976 debugfs_remove_recursive(vcpu->debugfs_dentry);
2977 kvm_put_kvm(vcpu->kvm);
2981 static struct file_operations kvm_vcpu_fops = {
2982 .release = kvm_vcpu_release,
2983 .unlocked_ioctl = kvm_vcpu_ioctl,
2984 .mmap = kvm_vcpu_mmap,
2985 .llseek = noop_llseek,
2986 KVM_COMPAT(kvm_vcpu_compat_ioctl),
2990 * Allocates an inode for the vcpu.
2992 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2994 char name[8 + 1 + ITOA_MAX_LEN + 1];
2996 snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id);
2997 return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
3000 static void kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
3002 #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS
3003 char dir_name[ITOA_MAX_LEN * 2];
3005 if (!debugfs_initialized())
3008 snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
3009 vcpu->debugfs_dentry = debugfs_create_dir(dir_name,
3010 vcpu->kvm->debugfs_dentry);
3012 kvm_arch_create_vcpu_debugfs(vcpu);
3017 * Creates some virtual cpus. Good luck creating more than one.
3019 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
3022 struct kvm_vcpu *vcpu;
3025 if (id >= KVM_MAX_VCPU_ID)
3028 mutex_lock(&kvm->lock);
3029 if (kvm->created_vcpus == KVM_MAX_VCPUS) {
3030 mutex_unlock(&kvm->lock);
3034 kvm->created_vcpus++;
3035 mutex_unlock(&kvm->lock);
3037 r = kvm_arch_vcpu_precreate(kvm, id);
3039 goto vcpu_decrement;
3041 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
3044 goto vcpu_decrement;
3047 BUILD_BUG_ON(sizeof(struct kvm_run) > PAGE_SIZE);
3048 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
3053 vcpu->run = page_address(page);
3055 kvm_vcpu_init(vcpu, kvm, id);
3057 r = kvm_arch_vcpu_create(vcpu);
3059 goto vcpu_free_run_page;
3061 mutex_lock(&kvm->lock);
3062 if (kvm_get_vcpu_by_id(kvm, id)) {
3064 goto unlock_vcpu_destroy;
3067 vcpu->vcpu_idx = atomic_read(&kvm->online_vcpus);
3068 BUG_ON(kvm->vcpus[vcpu->vcpu_idx]);
3070 /* Now it's all set up, let userspace reach it */
3072 r = create_vcpu_fd(vcpu);
3074 kvm_put_kvm_no_destroy(kvm);
3075 goto unlock_vcpu_destroy;
3078 kvm->vcpus[vcpu->vcpu_idx] = vcpu;
3081 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
3082 * before kvm->online_vcpu's incremented value.
3085 atomic_inc(&kvm->online_vcpus);
3087 mutex_unlock(&kvm->lock);
3088 kvm_arch_vcpu_postcreate(vcpu);
3089 kvm_create_vcpu_debugfs(vcpu);
3092 unlock_vcpu_destroy:
3093 mutex_unlock(&kvm->lock);
3094 kvm_arch_vcpu_destroy(vcpu);
3096 free_page((unsigned long)vcpu->run);
3098 kmem_cache_free(kvm_vcpu_cache, vcpu);
3100 mutex_lock(&kvm->lock);
3101 kvm->created_vcpus--;
3102 mutex_unlock(&kvm->lock);
3106 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
3109 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
3110 vcpu->sigset_active = 1;
3111 vcpu->sigset = *sigset;
3113 vcpu->sigset_active = 0;
3117 static long kvm_vcpu_ioctl(struct file *filp,
3118 unsigned int ioctl, unsigned long arg)
3120 struct kvm_vcpu *vcpu = filp->private_data;
3121 void __user *argp = (void __user *)arg;
3123 struct kvm_fpu *fpu = NULL;
3124 struct kvm_sregs *kvm_sregs = NULL;
3126 if (vcpu->kvm->mm != current->mm)
3129 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
3133 * Some architectures have vcpu ioctls that are asynchronous to vcpu
3134 * execution; mutex_lock() would break them.
3136 r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg);
3137 if (r != -ENOIOCTLCMD)
3140 if (mutex_lock_killable(&vcpu->mutex))
3148 oldpid = rcu_access_pointer(vcpu->pid);
3149 if (unlikely(oldpid != task_pid(current))) {
3150 /* The thread running this VCPU changed. */
3153 r = kvm_arch_vcpu_run_pid_change(vcpu);
3157 newpid = get_task_pid(current, PIDTYPE_PID);
3158 rcu_assign_pointer(vcpu->pid, newpid);
3163 r = kvm_arch_vcpu_ioctl_run(vcpu);
3164 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
3167 case KVM_GET_REGS: {
3168 struct kvm_regs *kvm_regs;
3171 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL_ACCOUNT);
3174 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
3178 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
3185 case KVM_SET_REGS: {
3186 struct kvm_regs *kvm_regs;
3188 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
3189 if (IS_ERR(kvm_regs)) {
3190 r = PTR_ERR(kvm_regs);
3193 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
3197 case KVM_GET_SREGS: {
3198 kvm_sregs = kzalloc(sizeof(struct kvm_sregs),
3199 GFP_KERNEL_ACCOUNT);
3203 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
3207 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
3212 case KVM_SET_SREGS: {
3213 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
3214 if (IS_ERR(kvm_sregs)) {
3215 r = PTR_ERR(kvm_sregs);
3219 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
3222 case KVM_GET_MP_STATE: {
3223 struct kvm_mp_state mp_state;
3225 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
3229 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
3234 case KVM_SET_MP_STATE: {
3235 struct kvm_mp_state mp_state;
3238 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
3240 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
3243 case KVM_TRANSLATE: {
3244 struct kvm_translation tr;
3247 if (copy_from_user(&tr, argp, sizeof(tr)))
3249 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
3253 if (copy_to_user(argp, &tr, sizeof(tr)))
3258 case KVM_SET_GUEST_DEBUG: {
3259 struct kvm_guest_debug dbg;
3262 if (copy_from_user(&dbg, argp, sizeof(dbg)))
3264 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
3267 case KVM_SET_SIGNAL_MASK: {
3268 struct kvm_signal_mask __user *sigmask_arg = argp;
3269 struct kvm_signal_mask kvm_sigmask;
3270 sigset_t sigset, *p;
3275 if (copy_from_user(&kvm_sigmask, argp,
3276 sizeof(kvm_sigmask)))
3279 if (kvm_sigmask.len != sizeof(sigset))
3282 if (copy_from_user(&sigset, sigmask_arg->sigset,
3287 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
3291 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL_ACCOUNT);
3295 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
3299 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
3305 fpu = memdup_user(argp, sizeof(*fpu));
3311 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
3315 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
3318 mutex_unlock(&vcpu->mutex);
3324 #ifdef CONFIG_KVM_COMPAT
3325 static long kvm_vcpu_compat_ioctl(struct file *filp,
3326 unsigned int ioctl, unsigned long arg)
3328 struct kvm_vcpu *vcpu = filp->private_data;
3329 void __user *argp = compat_ptr(arg);
3332 if (vcpu->kvm->mm != current->mm)
3336 case KVM_SET_SIGNAL_MASK: {
3337 struct kvm_signal_mask __user *sigmask_arg = argp;
3338 struct kvm_signal_mask kvm_sigmask;
3343 if (copy_from_user(&kvm_sigmask, argp,
3344 sizeof(kvm_sigmask)))
3347 if (kvm_sigmask.len != sizeof(compat_sigset_t))
3350 if (get_compat_sigset(&sigset, (void *)sigmask_arg->sigset))
3352 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
3354 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
3358 r = kvm_vcpu_ioctl(filp, ioctl, arg);
3366 static int kvm_device_mmap(struct file *filp, struct vm_area_struct *vma)
3368 struct kvm_device *dev = filp->private_data;
3371 return dev->ops->mmap(dev, vma);
3376 static int kvm_device_ioctl_attr(struct kvm_device *dev,
3377 int (*accessor)(struct kvm_device *dev,
3378 struct kvm_device_attr *attr),
3381 struct kvm_device_attr attr;
3386 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
3389 return accessor(dev, &attr);
3392 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
3395 struct kvm_device *dev = filp->private_data;
3397 if (dev->kvm->mm != current->mm)
3401 case KVM_SET_DEVICE_ATTR:
3402 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
3403 case KVM_GET_DEVICE_ATTR:
3404 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
3405 case KVM_HAS_DEVICE_ATTR:
3406 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
3408 if (dev->ops->ioctl)
3409 return dev->ops->ioctl(dev, ioctl, arg);
3415 static int kvm_device_release(struct inode *inode, struct file *filp)
3417 struct kvm_device *dev = filp->private_data;
3418 struct kvm *kvm = dev->kvm;
3420 if (dev->ops->release) {
3421 mutex_lock(&kvm->lock);
3422 list_del(&dev->vm_node);
3423 dev->ops->release(dev);
3424 mutex_unlock(&kvm->lock);
3431 static const struct file_operations kvm_device_fops = {
3432 .unlocked_ioctl = kvm_device_ioctl,
3433 .release = kvm_device_release,
3434 KVM_COMPAT(kvm_device_ioctl),
3435 .mmap = kvm_device_mmap,
3438 struct kvm_device *kvm_device_from_filp(struct file *filp)
3440 if (filp->f_op != &kvm_device_fops)
3443 return filp->private_data;
3446 static const struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
3447 #ifdef CONFIG_KVM_MPIC
3448 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
3449 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
3453 int kvm_register_device_ops(const struct kvm_device_ops *ops, u32 type)
3455 if (type >= ARRAY_SIZE(kvm_device_ops_table))
3458 if (kvm_device_ops_table[type] != NULL)
3461 kvm_device_ops_table[type] = ops;
3465 void kvm_unregister_device_ops(u32 type)
3467 if (kvm_device_ops_table[type] != NULL)
3468 kvm_device_ops_table[type] = NULL;
3471 static int kvm_ioctl_create_device(struct kvm *kvm,
3472 struct kvm_create_device *cd)
3474 const struct kvm_device_ops *ops = NULL;
3475 struct kvm_device *dev;
3476 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
3480 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
3483 type = array_index_nospec(cd->type, ARRAY_SIZE(kvm_device_ops_table));
3484 ops = kvm_device_ops_table[type];
3491 dev = kzalloc(sizeof(*dev), GFP_KERNEL_ACCOUNT);
3498 mutex_lock(&kvm->lock);
3499 ret = ops->create(dev, type);
3501 mutex_unlock(&kvm->lock);
3505 list_add(&dev->vm_node, &kvm->devices);
3506 mutex_unlock(&kvm->lock);
3512 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
3514 kvm_put_kvm_no_destroy(kvm);
3515 mutex_lock(&kvm->lock);
3516 list_del(&dev->vm_node);
3517 mutex_unlock(&kvm->lock);
3526 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
3529 case KVM_CAP_USER_MEMORY:
3530 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
3531 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
3532 case KVM_CAP_INTERNAL_ERROR_DATA:
3533 #ifdef CONFIG_HAVE_KVM_MSI
3534 case KVM_CAP_SIGNAL_MSI:
3536 #ifdef CONFIG_HAVE_KVM_IRQFD
3538 case KVM_CAP_IRQFD_RESAMPLE:
3540 case KVM_CAP_IOEVENTFD_ANY_LENGTH:
3541 case KVM_CAP_CHECK_EXTENSION_VM:
3542 case KVM_CAP_ENABLE_CAP_VM:
3543 case KVM_CAP_HALT_POLL:
3545 #ifdef CONFIG_KVM_MMIO
3546 case KVM_CAP_COALESCED_MMIO:
3547 return KVM_COALESCED_MMIO_PAGE_OFFSET;
3548 case KVM_CAP_COALESCED_PIO:
3551 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3552 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2:
3553 return KVM_DIRTY_LOG_MANUAL_CAPS;
3555 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3556 case KVM_CAP_IRQ_ROUTING:
3557 return KVM_MAX_IRQ_ROUTES;
3559 #if KVM_ADDRESS_SPACE_NUM > 1
3560 case KVM_CAP_MULTI_ADDRESS_SPACE:
3561 return KVM_ADDRESS_SPACE_NUM;
3563 case KVM_CAP_NR_MEMSLOTS:
3564 return KVM_USER_MEM_SLOTS;
3568 return kvm_vm_ioctl_check_extension(kvm, arg);
3571 int __attribute__((weak)) kvm_vm_ioctl_enable_cap(struct kvm *kvm,
3572 struct kvm_enable_cap *cap)
3577 static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm,
3578 struct kvm_enable_cap *cap)
3581 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3582 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2: {
3583 u64 allowed_options = KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE;
3585 if (cap->args[0] & KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE)
3586 allowed_options = KVM_DIRTY_LOG_MANUAL_CAPS;
3588 if (cap->flags || (cap->args[0] & ~allowed_options))
3590 kvm->manual_dirty_log_protect = cap->args[0];
3594 case KVM_CAP_HALT_POLL: {
3595 if (cap->flags || cap->args[0] != (unsigned int)cap->args[0])
3598 kvm->max_halt_poll_ns = cap->args[0];
3602 return kvm_vm_ioctl_enable_cap(kvm, cap);
3606 static long kvm_vm_ioctl(struct file *filp,
3607 unsigned int ioctl, unsigned long arg)
3609 struct kvm *kvm = filp->private_data;
3610 void __user *argp = (void __user *)arg;
3613 if (kvm->mm != current->mm)
3616 case KVM_CREATE_VCPU:
3617 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
3619 case KVM_ENABLE_CAP: {
3620 struct kvm_enable_cap cap;
3623 if (copy_from_user(&cap, argp, sizeof(cap)))
3625 r = kvm_vm_ioctl_enable_cap_generic(kvm, &cap);
3628 case KVM_SET_USER_MEMORY_REGION: {
3629 struct kvm_userspace_memory_region kvm_userspace_mem;
3632 if (copy_from_user(&kvm_userspace_mem, argp,
3633 sizeof(kvm_userspace_mem)))
3636 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
3639 case KVM_GET_DIRTY_LOG: {
3640 struct kvm_dirty_log log;
3643 if (copy_from_user(&log, argp, sizeof(log)))
3645 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3648 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3649 case KVM_CLEAR_DIRTY_LOG: {
3650 struct kvm_clear_dirty_log log;
3653 if (copy_from_user(&log, argp, sizeof(log)))
3655 r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
3659 #ifdef CONFIG_KVM_MMIO
3660 case KVM_REGISTER_COALESCED_MMIO: {
3661 struct kvm_coalesced_mmio_zone zone;
3664 if (copy_from_user(&zone, argp, sizeof(zone)))
3666 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
3669 case KVM_UNREGISTER_COALESCED_MMIO: {
3670 struct kvm_coalesced_mmio_zone zone;
3673 if (copy_from_user(&zone, argp, sizeof(zone)))
3675 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
3680 struct kvm_irqfd data;
3683 if (copy_from_user(&data, argp, sizeof(data)))
3685 r = kvm_irqfd(kvm, &data);
3688 case KVM_IOEVENTFD: {
3689 struct kvm_ioeventfd data;
3692 if (copy_from_user(&data, argp, sizeof(data)))
3694 r = kvm_ioeventfd(kvm, &data);
3697 #ifdef CONFIG_HAVE_KVM_MSI
3698 case KVM_SIGNAL_MSI: {
3702 if (copy_from_user(&msi, argp, sizeof(msi)))
3704 r = kvm_send_userspace_msi(kvm, &msi);
3708 #ifdef __KVM_HAVE_IRQ_LINE
3709 case KVM_IRQ_LINE_STATUS:
3710 case KVM_IRQ_LINE: {
3711 struct kvm_irq_level irq_event;
3714 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
3717 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
3718 ioctl == KVM_IRQ_LINE_STATUS);
3723 if (ioctl == KVM_IRQ_LINE_STATUS) {
3724 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
3732 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3733 case KVM_SET_GSI_ROUTING: {
3734 struct kvm_irq_routing routing;
3735 struct kvm_irq_routing __user *urouting;
3736 struct kvm_irq_routing_entry *entries = NULL;
3739 if (copy_from_user(&routing, argp, sizeof(routing)))
3742 if (!kvm_arch_can_set_irq_routing(kvm))
3744 if (routing.nr > KVM_MAX_IRQ_ROUTES)
3750 entries = vmalloc(array_size(sizeof(*entries),
3756 if (copy_from_user(entries, urouting->entries,
3757 routing.nr * sizeof(*entries)))
3758 goto out_free_irq_routing;
3760 r = kvm_set_irq_routing(kvm, entries, routing.nr,
3762 out_free_irq_routing:
3766 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
3767 case KVM_CREATE_DEVICE: {
3768 struct kvm_create_device cd;
3771 if (copy_from_user(&cd, argp, sizeof(cd)))
3774 r = kvm_ioctl_create_device(kvm, &cd);
3779 if (copy_to_user(argp, &cd, sizeof(cd)))
3785 case KVM_CHECK_EXTENSION:
3786 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
3789 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
3795 #ifdef CONFIG_KVM_COMPAT
3796 struct compat_kvm_dirty_log {
3800 compat_uptr_t dirty_bitmap; /* one bit per page */
3805 static long kvm_vm_compat_ioctl(struct file *filp,
3806 unsigned int ioctl, unsigned long arg)
3808 struct kvm *kvm = filp->private_data;
3811 if (kvm->mm != current->mm)
3814 case KVM_GET_DIRTY_LOG: {
3815 struct compat_kvm_dirty_log compat_log;
3816 struct kvm_dirty_log log;
3818 if (copy_from_user(&compat_log, (void __user *)arg,
3819 sizeof(compat_log)))
3821 log.slot = compat_log.slot;
3822 log.padding1 = compat_log.padding1;
3823 log.padding2 = compat_log.padding2;
3824 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
3826 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3830 r = kvm_vm_ioctl(filp, ioctl, arg);
3836 static struct file_operations kvm_vm_fops = {
3837 .release = kvm_vm_release,
3838 .unlocked_ioctl = kvm_vm_ioctl,
3839 .llseek = noop_llseek,
3840 KVM_COMPAT(kvm_vm_compat_ioctl),
3843 static int kvm_dev_ioctl_create_vm(unsigned long type)
3849 kvm = kvm_create_vm(type);
3851 return PTR_ERR(kvm);
3852 #ifdef CONFIG_KVM_MMIO
3853 r = kvm_coalesced_mmio_init(kvm);
3857 r = get_unused_fd_flags(O_CLOEXEC);
3861 file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
3869 * Don't call kvm_put_kvm anymore at this point; file->f_op is
3870 * already set, with ->release() being kvm_vm_release(). In error
3871 * cases it will be called by the final fput(file) and will take
3872 * care of doing kvm_put_kvm(kvm).
3874 if (kvm_create_vm_debugfs(kvm, r) < 0) {
3879 kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm);
3881 fd_install(r, file);
3889 static long kvm_dev_ioctl(struct file *filp,
3890 unsigned int ioctl, unsigned long arg)
3895 case KVM_GET_API_VERSION:
3898 r = KVM_API_VERSION;
3901 r = kvm_dev_ioctl_create_vm(arg);
3903 case KVM_CHECK_EXTENSION:
3904 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
3906 case KVM_GET_VCPU_MMAP_SIZE:
3909 r = PAGE_SIZE; /* struct kvm_run */
3911 r += PAGE_SIZE; /* pio data page */
3913 #ifdef CONFIG_KVM_MMIO
3914 r += PAGE_SIZE; /* coalesced mmio ring page */
3917 case KVM_TRACE_ENABLE:
3918 case KVM_TRACE_PAUSE:
3919 case KVM_TRACE_DISABLE:
3923 return kvm_arch_dev_ioctl(filp, ioctl, arg);
3929 static struct file_operations kvm_chardev_ops = {
3930 .unlocked_ioctl = kvm_dev_ioctl,
3931 .llseek = noop_llseek,
3932 KVM_COMPAT(kvm_dev_ioctl),
3935 static struct miscdevice kvm_dev = {
3941 static void hardware_enable_nolock(void *junk)
3943 int cpu = raw_smp_processor_id();
3946 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
3949 cpumask_set_cpu(cpu, cpus_hardware_enabled);
3951 r = kvm_arch_hardware_enable();
3954 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3955 atomic_inc(&hardware_enable_failed);
3956 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
3960 static int kvm_starting_cpu(unsigned int cpu)
3962 raw_spin_lock(&kvm_count_lock);
3963 if (kvm_usage_count)
3964 hardware_enable_nolock(NULL);
3965 raw_spin_unlock(&kvm_count_lock);
3969 static void hardware_disable_nolock(void *junk)
3971 int cpu = raw_smp_processor_id();
3973 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
3975 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3976 kvm_arch_hardware_disable();
3979 static int kvm_dying_cpu(unsigned int cpu)
3981 raw_spin_lock(&kvm_count_lock);
3982 if (kvm_usage_count)
3983 hardware_disable_nolock(NULL);
3984 raw_spin_unlock(&kvm_count_lock);
3988 static void hardware_disable_all_nolock(void)
3990 BUG_ON(!kvm_usage_count);
3993 if (!kvm_usage_count)
3994 on_each_cpu(hardware_disable_nolock, NULL, 1);
3997 static void hardware_disable_all(void)
3999 raw_spin_lock(&kvm_count_lock);
4000 hardware_disable_all_nolock();
4001 raw_spin_unlock(&kvm_count_lock);
4004 static int hardware_enable_all(void)
4008 raw_spin_lock(&kvm_count_lock);
4011 if (kvm_usage_count == 1) {
4012 atomic_set(&hardware_enable_failed, 0);
4013 on_each_cpu(hardware_enable_nolock, NULL, 1);
4015 if (atomic_read(&hardware_enable_failed)) {
4016 hardware_disable_all_nolock();
4021 raw_spin_unlock(&kvm_count_lock);
4026 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
4030 * Some (well, at least mine) BIOSes hang on reboot if
4033 * And Intel TXT required VMX off for all cpu when system shutdown.
4035 pr_info("kvm: exiting hardware virtualization\n");
4036 kvm_rebooting = true;
4037 on_each_cpu(hardware_disable_nolock, NULL, 1);
4041 static struct notifier_block kvm_reboot_notifier = {
4042 .notifier_call = kvm_reboot,
4046 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
4050 for (i = 0; i < bus->dev_count; i++) {
4051 struct kvm_io_device *pos = bus->range[i].dev;
4053 kvm_iodevice_destructor(pos);
4058 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
4059 const struct kvm_io_range *r2)
4061 gpa_t addr1 = r1->addr;
4062 gpa_t addr2 = r2->addr;
4067 /* If r2->len == 0, match the exact address. If r2->len != 0,
4068 * accept any overlapping write. Any order is acceptable for
4069 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
4070 * we process all of them.
4083 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
4085 return kvm_io_bus_cmp(p1, p2);
4088 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
4089 gpa_t addr, int len)
4091 struct kvm_io_range *range, key;
4094 key = (struct kvm_io_range) {
4099 range = bsearch(&key, bus->range, bus->dev_count,
4100 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
4104 off = range - bus->range;
4106 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
4112 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
4113 struct kvm_io_range *range, const void *val)
4117 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
4121 while (idx < bus->dev_count &&
4122 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
4123 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
4132 /* kvm_io_bus_write - called under kvm->slots_lock */
4133 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
4134 int len, const void *val)
4136 struct kvm_io_bus *bus;
4137 struct kvm_io_range range;
4140 range = (struct kvm_io_range) {
4145 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
4148 r = __kvm_io_bus_write(vcpu, bus, &range, val);
4149 return r < 0 ? r : 0;
4151 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
4153 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
4154 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
4155 gpa_t addr, int len, const void *val, long cookie)
4157 struct kvm_io_bus *bus;
4158 struct kvm_io_range range;
4160 range = (struct kvm_io_range) {
4165 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
4169 /* First try the device referenced by cookie. */
4170 if ((cookie >= 0) && (cookie < bus->dev_count) &&
4171 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
4172 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
4177 * cookie contained garbage; fall back to search and return the
4178 * correct cookie value.
4180 return __kvm_io_bus_write(vcpu, bus, &range, val);
4183 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
4184 struct kvm_io_range *range, void *val)
4188 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
4192 while (idx < bus->dev_count &&
4193 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
4194 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
4203 /* kvm_io_bus_read - called under kvm->slots_lock */
4204 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
4207 struct kvm_io_bus *bus;
4208 struct kvm_io_range range;
4211 range = (struct kvm_io_range) {
4216 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
4219 r = __kvm_io_bus_read(vcpu, bus, &range, val);
4220 return r < 0 ? r : 0;
4223 /* Caller must hold slots_lock. */
4224 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
4225 int len, struct kvm_io_device *dev)
4228 struct kvm_io_bus *new_bus, *bus;
4229 struct kvm_io_range range;
4231 bus = kvm_get_bus(kvm, bus_idx);
4235 /* exclude ioeventfd which is limited by maximum fd */
4236 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
4239 new_bus = kmalloc(struct_size(bus, range, bus->dev_count + 1),
4240 GFP_KERNEL_ACCOUNT);
4244 range = (struct kvm_io_range) {
4250 for (i = 0; i < bus->dev_count; i++)
4251 if (kvm_io_bus_cmp(&bus->range[i], &range) > 0)
4254 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
4255 new_bus->dev_count++;
4256 new_bus->range[i] = range;
4257 memcpy(new_bus->range + i + 1, bus->range + i,
4258 (bus->dev_count - i) * sizeof(struct kvm_io_range));
4259 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
4260 synchronize_srcu_expedited(&kvm->srcu);
4266 /* Caller must hold slots_lock. */
4267 void kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
4268 struct kvm_io_device *dev)
4271 struct kvm_io_bus *new_bus, *bus;
4273 bus = kvm_get_bus(kvm, bus_idx);
4277 for (i = 0; i < bus->dev_count; i++)
4278 if (bus->range[i].dev == dev) {
4282 if (i == bus->dev_count)
4285 new_bus = kmalloc(struct_size(bus, range, bus->dev_count - 1),
4286 GFP_KERNEL_ACCOUNT);
4288 pr_err("kvm: failed to shrink bus, removing it completely\n");
4292 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
4293 new_bus->dev_count--;
4294 memcpy(new_bus->range + i, bus->range + i + 1,
4295 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
4298 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
4299 synchronize_srcu_expedited(&kvm->srcu);
4304 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
4307 struct kvm_io_bus *bus;
4308 int dev_idx, srcu_idx;
4309 struct kvm_io_device *iodev = NULL;
4311 srcu_idx = srcu_read_lock(&kvm->srcu);
4313 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
4317 dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
4321 iodev = bus->range[dev_idx].dev;
4324 srcu_read_unlock(&kvm->srcu, srcu_idx);
4328 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
4330 static int kvm_debugfs_open(struct inode *inode, struct file *file,
4331 int (*get)(void *, u64 *), int (*set)(void *, u64),
4334 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
4337 /* The debugfs files are a reference to the kvm struct which
4338 * is still valid when kvm_destroy_vm is called.
4339 * To avoid the race between open and the removal of the debugfs
4340 * directory we test against the users count.
4342 if (!refcount_inc_not_zero(&stat_data->kvm->users_count))
4345 if (simple_attr_open(inode, file, get,
4346 KVM_DBGFS_GET_MODE(stat_data->dbgfs_item) & 0222
4349 kvm_put_kvm(stat_data->kvm);
4356 static int kvm_debugfs_release(struct inode *inode, struct file *file)
4358 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
4361 simple_attr_release(inode, file);
4362 kvm_put_kvm(stat_data->kvm);
4367 static int kvm_get_stat_per_vm(struct kvm *kvm, size_t offset, u64 *val)
4369 *val = *(ulong *)((void *)kvm + offset);
4374 static int kvm_clear_stat_per_vm(struct kvm *kvm, size_t offset)
4376 *(ulong *)((void *)kvm + offset) = 0;
4381 static int kvm_get_stat_per_vcpu(struct kvm *kvm, size_t offset, u64 *val)
4384 struct kvm_vcpu *vcpu;
4388 kvm_for_each_vcpu(i, vcpu, kvm)
4389 *val += *(u64 *)((void *)vcpu + offset);
4394 static int kvm_clear_stat_per_vcpu(struct kvm *kvm, size_t offset)
4397 struct kvm_vcpu *vcpu;
4399 kvm_for_each_vcpu(i, vcpu, kvm)
4400 *(u64 *)((void *)vcpu + offset) = 0;
4405 static int kvm_stat_data_get(void *data, u64 *val)
4408 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
4410 switch (stat_data->dbgfs_item->kind) {
4412 r = kvm_get_stat_per_vm(stat_data->kvm,
4413 stat_data->dbgfs_item->offset, val);
4416 r = kvm_get_stat_per_vcpu(stat_data->kvm,
4417 stat_data->dbgfs_item->offset, val);
4424 static int kvm_stat_data_clear(void *data, u64 val)
4427 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
4432 switch (stat_data->dbgfs_item->kind) {
4434 r = kvm_clear_stat_per_vm(stat_data->kvm,
4435 stat_data->dbgfs_item->offset);
4438 r = kvm_clear_stat_per_vcpu(stat_data->kvm,
4439 stat_data->dbgfs_item->offset);
4446 static int kvm_stat_data_open(struct inode *inode, struct file *file)
4448 __simple_attr_check_format("%llu\n", 0ull);
4449 return kvm_debugfs_open(inode, file, kvm_stat_data_get,
4450 kvm_stat_data_clear, "%llu\n");
4453 static const struct file_operations stat_fops_per_vm = {
4454 .owner = THIS_MODULE,
4455 .open = kvm_stat_data_open,
4456 .release = kvm_debugfs_release,
4457 .read = simple_attr_read,
4458 .write = simple_attr_write,
4459 .llseek = no_llseek,
4462 static int vm_stat_get(void *_offset, u64 *val)
4464 unsigned offset = (long)_offset;
4469 mutex_lock(&kvm_lock);
4470 list_for_each_entry(kvm, &vm_list, vm_list) {
4471 kvm_get_stat_per_vm(kvm, offset, &tmp_val);
4474 mutex_unlock(&kvm_lock);
4478 static int vm_stat_clear(void *_offset, u64 val)
4480 unsigned offset = (long)_offset;
4486 mutex_lock(&kvm_lock);
4487 list_for_each_entry(kvm, &vm_list, vm_list) {
4488 kvm_clear_stat_per_vm(kvm, offset);
4490 mutex_unlock(&kvm_lock);
4495 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n");
4497 static int vcpu_stat_get(void *_offset, u64 *val)
4499 unsigned offset = (long)_offset;
4504 mutex_lock(&kvm_lock);
4505 list_for_each_entry(kvm, &vm_list, vm_list) {
4506 kvm_get_stat_per_vcpu(kvm, offset, &tmp_val);
4509 mutex_unlock(&kvm_lock);
4513 static int vcpu_stat_clear(void *_offset, u64 val)
4515 unsigned offset = (long)_offset;
4521 mutex_lock(&kvm_lock);
4522 list_for_each_entry(kvm, &vm_list, vm_list) {
4523 kvm_clear_stat_per_vcpu(kvm, offset);
4525 mutex_unlock(&kvm_lock);
4530 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear,
4533 static const struct file_operations *stat_fops[] = {
4534 [KVM_STAT_VCPU] = &vcpu_stat_fops,
4535 [KVM_STAT_VM] = &vm_stat_fops,
4538 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm)
4540 struct kobj_uevent_env *env;
4541 unsigned long long created, active;
4543 if (!kvm_dev.this_device || !kvm)
4546 mutex_lock(&kvm_lock);
4547 if (type == KVM_EVENT_CREATE_VM) {
4548 kvm_createvm_count++;
4550 } else if (type == KVM_EVENT_DESTROY_VM) {
4553 created = kvm_createvm_count;
4554 active = kvm_active_vms;
4555 mutex_unlock(&kvm_lock);
4557 env = kzalloc(sizeof(*env), GFP_KERNEL_ACCOUNT);
4561 add_uevent_var(env, "CREATED=%llu", created);
4562 add_uevent_var(env, "COUNT=%llu", active);
4564 if (type == KVM_EVENT_CREATE_VM) {
4565 add_uevent_var(env, "EVENT=create");
4566 kvm->userspace_pid = task_pid_nr(current);
4567 } else if (type == KVM_EVENT_DESTROY_VM) {
4568 add_uevent_var(env, "EVENT=destroy");
4570 add_uevent_var(env, "PID=%d", kvm->userspace_pid);
4572 if (!IS_ERR_OR_NULL(kvm->debugfs_dentry)) {
4573 char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL_ACCOUNT);
4576 tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX);
4578 add_uevent_var(env, "STATS_PATH=%s", tmp);
4582 /* no need for checks, since we are adding at most only 5 keys */
4583 env->envp[env->envp_idx++] = NULL;
4584 kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp);
4588 static void kvm_init_debug(void)
4590 struct kvm_stats_debugfs_item *p;
4592 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
4594 kvm_debugfs_num_entries = 0;
4595 for (p = debugfs_entries; p->name; ++p, kvm_debugfs_num_entries++) {
4596 debugfs_create_file(p->name, KVM_DBGFS_GET_MODE(p),
4597 kvm_debugfs_dir, (void *)(long)p->offset,
4598 stat_fops[p->kind]);
4602 static int kvm_suspend(void)
4604 if (kvm_usage_count)
4605 hardware_disable_nolock(NULL);
4609 static void kvm_resume(void)
4611 if (kvm_usage_count) {
4612 #ifdef CONFIG_LOCKDEP
4613 WARN_ON(lockdep_is_held(&kvm_count_lock));
4615 hardware_enable_nolock(NULL);
4619 static struct syscore_ops kvm_syscore_ops = {
4620 .suspend = kvm_suspend,
4621 .resume = kvm_resume,
4625 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
4627 return container_of(pn, struct kvm_vcpu, preempt_notifier);
4630 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
4632 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
4634 WRITE_ONCE(vcpu->preempted, false);
4635 WRITE_ONCE(vcpu->ready, false);
4637 __this_cpu_write(kvm_running_vcpu, vcpu);
4638 kvm_arch_sched_in(vcpu, cpu);
4639 kvm_arch_vcpu_load(vcpu, cpu);
4642 static void kvm_sched_out(struct preempt_notifier *pn,
4643 struct task_struct *next)
4645 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
4647 if (current->state == TASK_RUNNING) {
4648 WRITE_ONCE(vcpu->preempted, true);
4649 WRITE_ONCE(vcpu->ready, true);
4651 kvm_arch_vcpu_put(vcpu);
4652 __this_cpu_write(kvm_running_vcpu, NULL);
4656 * kvm_get_running_vcpu - get the vcpu running on the current CPU.
4658 * We can disable preemption locally around accessing the per-CPU variable,
4659 * and use the resolved vcpu pointer after enabling preemption again,
4660 * because even if the current thread is migrated to another CPU, reading
4661 * the per-CPU value later will give us the same value as we update the
4662 * per-CPU variable in the preempt notifier handlers.
4664 struct kvm_vcpu *kvm_get_running_vcpu(void)
4666 struct kvm_vcpu *vcpu;
4669 vcpu = __this_cpu_read(kvm_running_vcpu);
4674 EXPORT_SYMBOL_GPL(kvm_get_running_vcpu);
4677 * kvm_get_running_vcpus - get the per-CPU array of currently running vcpus.
4679 struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void)
4681 return &kvm_running_vcpu;
4684 struct kvm_cpu_compat_check {
4689 static void check_processor_compat(void *data)
4691 struct kvm_cpu_compat_check *c = data;
4693 *c->ret = kvm_arch_check_processor_compat(c->opaque);
4696 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
4697 struct module *module)
4699 struct kvm_cpu_compat_check c;
4703 r = kvm_arch_init(opaque);
4708 * kvm_arch_init makes sure there's at most one caller
4709 * for architectures that support multiple implementations,
4710 * like intel and amd on x86.
4711 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
4712 * conflicts in case kvm is already setup for another implementation.
4714 r = kvm_irqfd_init();
4718 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
4723 r = kvm_arch_hardware_setup(opaque);
4729 for_each_online_cpu(cpu) {
4730 smp_call_function_single(cpu, check_processor_compat, &c, 1);
4735 r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting",
4736 kvm_starting_cpu, kvm_dying_cpu);
4739 register_reboot_notifier(&kvm_reboot_notifier);
4741 /* A kmem cache lets us meet the alignment requirements of fx_save. */
4743 vcpu_align = __alignof__(struct kvm_vcpu);
4745 kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align,
4747 offsetof(struct kvm_vcpu, arch),
4748 sizeof_field(struct kvm_vcpu, arch),
4750 if (!kvm_vcpu_cache) {
4755 r = kvm_async_pf_init();
4759 kvm_chardev_ops.owner = module;
4760 kvm_vm_fops.owner = module;
4761 kvm_vcpu_fops.owner = module;
4763 r = misc_register(&kvm_dev);
4765 pr_err("kvm: misc device register failed\n");
4769 register_syscore_ops(&kvm_syscore_ops);
4771 kvm_preempt_ops.sched_in = kvm_sched_in;
4772 kvm_preempt_ops.sched_out = kvm_sched_out;
4776 r = kvm_vfio_ops_init();
4782 kvm_async_pf_deinit();
4784 kmem_cache_destroy(kvm_vcpu_cache);
4786 unregister_reboot_notifier(&kvm_reboot_notifier);
4787 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4789 kvm_arch_hardware_unsetup();
4791 free_cpumask_var(cpus_hardware_enabled);
4799 EXPORT_SYMBOL_GPL(kvm_init);
4803 debugfs_remove_recursive(kvm_debugfs_dir);
4804 misc_deregister(&kvm_dev);
4805 kmem_cache_destroy(kvm_vcpu_cache);
4806 kvm_async_pf_deinit();
4807 unregister_syscore_ops(&kvm_syscore_ops);
4808 unregister_reboot_notifier(&kvm_reboot_notifier);
4809 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4810 on_each_cpu(hardware_disable_nolock, NULL, 1);
4811 kvm_arch_hardware_unsetup();
4814 free_cpumask_var(cpus_hardware_enabled);
4815 kvm_vfio_ops_exit();
4817 EXPORT_SYMBOL_GPL(kvm_exit);
4819 struct kvm_vm_worker_thread_context {
4821 struct task_struct *parent;
4822 struct completion init_done;
4823 kvm_vm_thread_fn_t thread_fn;
4828 static int kvm_vm_worker_thread(void *context)
4831 * The init_context is allocated on the stack of the parent thread, so
4832 * we have to locally copy anything that is needed beyond initialization
4834 struct kvm_vm_worker_thread_context *init_context = context;
4835 struct kvm *kvm = init_context->kvm;
4836 kvm_vm_thread_fn_t thread_fn = init_context->thread_fn;
4837 uintptr_t data = init_context->data;
4840 err = kthread_park(current);
4841 /* kthread_park(current) is never supposed to return an error */
4846 err = cgroup_attach_task_all(init_context->parent, current);
4848 kvm_err("%s: cgroup_attach_task_all failed with err %d\n",
4853 set_user_nice(current, task_nice(init_context->parent));
4856 init_context->err = err;
4857 complete(&init_context->init_done);
4858 init_context = NULL;
4863 /* Wait to be woken up by the spawner before proceeding. */
4866 if (!kthread_should_stop())
4867 err = thread_fn(kvm, data);
4872 int kvm_vm_create_worker_thread(struct kvm *kvm, kvm_vm_thread_fn_t thread_fn,
4873 uintptr_t data, const char *name,
4874 struct task_struct **thread_ptr)
4876 struct kvm_vm_worker_thread_context init_context = {};
4877 struct task_struct *thread;
4880 init_context.kvm = kvm;
4881 init_context.parent = current;
4882 init_context.thread_fn = thread_fn;
4883 init_context.data = data;
4884 init_completion(&init_context.init_done);
4886 thread = kthread_run(kvm_vm_worker_thread, &init_context,
4887 "%s-%d", name, task_pid_nr(current));
4889 return PTR_ERR(thread);
4891 /* kthread_run is never supposed to return NULL */
4892 WARN_ON(thread == NULL);
4894 wait_for_completion(&init_context.init_done);
4896 if (!init_context.err)
4897 *thread_ptr = thread;
4899 return init_context.err;