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 init_swait_queue_head(&vcpu->wq);
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 ((id < KVM_USER_MEM_SLOTS) &&
1229 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
1230 !access_ok((void __user *)(unsigned long)mem->userspace_addr,
1233 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
1235 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
1239 * Make a full copy of the old memslot, the pointer will become stale
1240 * when the memslots are re-sorted by update_memslots(), and the old
1241 * memslot needs to be referenced after calling update_memslots(), e.g.
1242 * to free its resources and for arch specific behavior.
1244 tmp = id_to_memslot(__kvm_memslots(kvm, as_id), id);
1249 memset(&old, 0, sizeof(old));
1253 if (!mem->memory_size)
1254 return kvm_delete_memslot(kvm, mem, &old, as_id);
1257 new.base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
1258 new.npages = mem->memory_size >> PAGE_SHIFT;
1259 new.flags = mem->flags;
1260 new.userspace_addr = mem->userspace_addr;
1262 if (new.npages > KVM_MEM_MAX_NR_PAGES)
1266 change = KVM_MR_CREATE;
1267 new.dirty_bitmap = NULL;
1268 memset(&new.arch, 0, sizeof(new.arch));
1269 } else { /* Modify an existing slot. */
1270 if ((new.userspace_addr != old.userspace_addr) ||
1271 (new.npages != old.npages) ||
1272 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
1275 if (new.base_gfn != old.base_gfn)
1276 change = KVM_MR_MOVE;
1277 else if (new.flags != old.flags)
1278 change = KVM_MR_FLAGS_ONLY;
1279 else /* Nothing to change. */
1282 /* Copy dirty_bitmap and arch from the current memslot. */
1283 new.dirty_bitmap = old.dirty_bitmap;
1284 memcpy(&new.arch, &old.arch, sizeof(new.arch));
1287 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
1288 /* Check for overlaps */
1289 kvm_for_each_memslot(tmp, __kvm_memslots(kvm, as_id)) {
1292 if (!((new.base_gfn + new.npages <= tmp->base_gfn) ||
1293 (new.base_gfn >= tmp->base_gfn + tmp->npages)))
1298 /* Allocate/free page dirty bitmap as needed */
1299 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
1300 new.dirty_bitmap = NULL;
1301 else if (!new.dirty_bitmap) {
1302 r = kvm_alloc_dirty_bitmap(&new);
1306 if (kvm_dirty_log_manual_protect_and_init_set(kvm))
1307 bitmap_set(new.dirty_bitmap, 0, new.npages);
1310 r = kvm_set_memslot(kvm, mem, &old, &new, as_id, change);
1314 if (old.dirty_bitmap && !new.dirty_bitmap)
1315 kvm_destroy_dirty_bitmap(&old);
1319 if (new.dirty_bitmap && !old.dirty_bitmap)
1320 kvm_destroy_dirty_bitmap(&new);
1323 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1325 int kvm_set_memory_region(struct kvm *kvm,
1326 const struct kvm_userspace_memory_region *mem)
1330 mutex_lock(&kvm->slots_lock);
1331 r = __kvm_set_memory_region(kvm, mem);
1332 mutex_unlock(&kvm->slots_lock);
1335 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1337 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1338 struct kvm_userspace_memory_region *mem)
1340 if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1343 return kvm_set_memory_region(kvm, mem);
1346 #ifndef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1348 * kvm_get_dirty_log - get a snapshot of dirty pages
1349 * @kvm: pointer to kvm instance
1350 * @log: slot id and address to which we copy the log
1351 * @is_dirty: set to '1' if any dirty pages were found
1352 * @memslot: set to the associated memslot, always valid on success
1354 int kvm_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log,
1355 int *is_dirty, struct kvm_memory_slot **memslot)
1357 struct kvm_memslots *slots;
1360 unsigned long any = 0;
1365 as_id = log->slot >> 16;
1366 id = (u16)log->slot;
1367 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1370 slots = __kvm_memslots(kvm, as_id);
1371 *memslot = id_to_memslot(slots, id);
1372 if (!(*memslot) || !(*memslot)->dirty_bitmap)
1375 kvm_arch_sync_dirty_log(kvm, *memslot);
1377 n = kvm_dirty_bitmap_bytes(*memslot);
1379 for (i = 0; !any && i < n/sizeof(long); ++i)
1380 any = (*memslot)->dirty_bitmap[i];
1382 if (copy_to_user(log->dirty_bitmap, (*memslot)->dirty_bitmap, n))
1389 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1391 #else /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
1393 * kvm_get_dirty_log_protect - get a snapshot of dirty pages
1394 * and reenable dirty page tracking for the corresponding pages.
1395 * @kvm: pointer to kvm instance
1396 * @log: slot id and address to which we copy the log
1398 * We need to keep it in mind that VCPU threads can write to the bitmap
1399 * concurrently. So, to avoid losing track of dirty pages we keep the
1402 * 1. Take a snapshot of the bit and clear it if needed.
1403 * 2. Write protect the corresponding page.
1404 * 3. Copy the snapshot to the userspace.
1405 * 4. Upon return caller flushes TLB's if needed.
1407 * Between 2 and 4, the guest may write to the page using the remaining TLB
1408 * entry. This is not a problem because the page is reported dirty using
1409 * the snapshot taken before and step 4 ensures that writes done after
1410 * exiting to userspace will be logged for the next call.
1413 static int kvm_get_dirty_log_protect(struct kvm *kvm, struct kvm_dirty_log *log)
1415 struct kvm_memslots *slots;
1416 struct kvm_memory_slot *memslot;
1419 unsigned long *dirty_bitmap;
1420 unsigned long *dirty_bitmap_buffer;
1423 as_id = log->slot >> 16;
1424 id = (u16)log->slot;
1425 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1428 slots = __kvm_memslots(kvm, as_id);
1429 memslot = id_to_memslot(slots, id);
1430 if (!memslot || !memslot->dirty_bitmap)
1433 dirty_bitmap = memslot->dirty_bitmap;
1435 kvm_arch_sync_dirty_log(kvm, memslot);
1437 n = kvm_dirty_bitmap_bytes(memslot);
1439 if (kvm->manual_dirty_log_protect) {
1441 * Unlike kvm_get_dirty_log, we always return false in *flush,
1442 * because no flush is needed until KVM_CLEAR_DIRTY_LOG. There
1443 * is some code duplication between this function and
1444 * kvm_get_dirty_log, but hopefully all architecture
1445 * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log
1446 * can be eliminated.
1448 dirty_bitmap_buffer = dirty_bitmap;
1450 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
1451 memset(dirty_bitmap_buffer, 0, n);
1453 spin_lock(&kvm->mmu_lock);
1454 for (i = 0; i < n / sizeof(long); i++) {
1458 if (!dirty_bitmap[i])
1462 mask = xchg(&dirty_bitmap[i], 0);
1463 dirty_bitmap_buffer[i] = mask;
1465 offset = i * BITS_PER_LONG;
1466 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1469 spin_unlock(&kvm->mmu_lock);
1473 kvm_arch_flush_remote_tlbs_memslot(kvm, memslot);
1475 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1482 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
1483 * @kvm: kvm instance
1484 * @log: slot id and address to which we copy the log
1486 * Steps 1-4 below provide general overview of dirty page logging. See
1487 * kvm_get_dirty_log_protect() function description for additional details.
1489 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
1490 * always flush the TLB (step 4) even if previous step failed and the dirty
1491 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
1492 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
1493 * writes will be marked dirty for next log read.
1495 * 1. Take a snapshot of the bit and clear it if needed.
1496 * 2. Write protect the corresponding page.
1497 * 3. Copy the snapshot to the userspace.
1498 * 4. Flush TLB's if needed.
1500 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1501 struct kvm_dirty_log *log)
1505 mutex_lock(&kvm->slots_lock);
1507 r = kvm_get_dirty_log_protect(kvm, log);
1509 mutex_unlock(&kvm->slots_lock);
1514 * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap
1515 * and reenable dirty page tracking for the corresponding pages.
1516 * @kvm: pointer to kvm instance
1517 * @log: slot id and address from which to fetch the bitmap of dirty pages
1519 static int kvm_clear_dirty_log_protect(struct kvm *kvm,
1520 struct kvm_clear_dirty_log *log)
1522 struct kvm_memslots *slots;
1523 struct kvm_memory_slot *memslot;
1527 unsigned long *dirty_bitmap;
1528 unsigned long *dirty_bitmap_buffer;
1531 as_id = log->slot >> 16;
1532 id = (u16)log->slot;
1533 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1536 if (log->first_page & 63)
1539 slots = __kvm_memslots(kvm, as_id);
1540 memslot = id_to_memslot(slots, id);
1541 if (!memslot || !memslot->dirty_bitmap)
1544 dirty_bitmap = memslot->dirty_bitmap;
1546 n = ALIGN(log->num_pages, BITS_PER_LONG) / 8;
1548 if (log->first_page > memslot->npages ||
1549 log->num_pages > memslot->npages - log->first_page ||
1550 (log->num_pages < memslot->npages - log->first_page && (log->num_pages & 63)))
1553 kvm_arch_sync_dirty_log(kvm, memslot);
1556 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
1557 if (copy_from_user(dirty_bitmap_buffer, log->dirty_bitmap, n))
1560 spin_lock(&kvm->mmu_lock);
1561 for (offset = log->first_page, i = offset / BITS_PER_LONG,
1562 n = DIV_ROUND_UP(log->num_pages, BITS_PER_LONG); n--;
1563 i++, offset += BITS_PER_LONG) {
1564 unsigned long mask = *dirty_bitmap_buffer++;
1565 atomic_long_t *p = (atomic_long_t *) &dirty_bitmap[i];
1569 mask &= atomic_long_fetch_andnot(mask, p);
1572 * mask contains the bits that really have been cleared. This
1573 * never includes any bits beyond the length of the memslot (if
1574 * the length is not aligned to 64 pages), therefore it is not
1575 * a problem if userspace sets them in log->dirty_bitmap.
1579 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1583 spin_unlock(&kvm->mmu_lock);
1586 kvm_arch_flush_remote_tlbs_memslot(kvm, memslot);
1591 static int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm,
1592 struct kvm_clear_dirty_log *log)
1596 mutex_lock(&kvm->slots_lock);
1598 r = kvm_clear_dirty_log_protect(kvm, log);
1600 mutex_unlock(&kvm->slots_lock);
1603 #endif /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
1605 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1607 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1609 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1611 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
1613 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn);
1615 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_memslot);
1617 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1619 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1621 return kvm_is_visible_memslot(memslot);
1623 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1625 unsigned long kvm_host_page_size(struct kvm_vcpu *vcpu, gfn_t gfn)
1627 struct vm_area_struct *vma;
1628 unsigned long addr, size;
1632 addr = kvm_vcpu_gfn_to_hva_prot(vcpu, gfn, NULL);
1633 if (kvm_is_error_hva(addr))
1636 down_read(¤t->mm->mmap_sem);
1637 vma = find_vma(current->mm, addr);
1641 size = vma_kernel_pagesize(vma);
1644 up_read(¤t->mm->mmap_sem);
1649 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1651 return slot->flags & KVM_MEM_READONLY;
1654 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1655 gfn_t *nr_pages, bool write)
1657 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1658 return KVM_HVA_ERR_BAD;
1660 if (memslot_is_readonly(slot) && write)
1661 return KVM_HVA_ERR_RO_BAD;
1664 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1666 return __gfn_to_hva_memslot(slot, gfn);
1669 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1672 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1675 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1678 return gfn_to_hva_many(slot, gfn, NULL);
1680 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1682 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1684 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1686 EXPORT_SYMBOL_GPL(gfn_to_hva);
1688 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
1690 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
1692 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
1695 * Return the hva of a @gfn and the R/W attribute if possible.
1697 * @slot: the kvm_memory_slot which contains @gfn
1698 * @gfn: the gfn to be translated
1699 * @writable: used to return the read/write attribute of the @slot if the hva
1700 * is valid and @writable is not NULL
1702 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1703 gfn_t gfn, bool *writable)
1705 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1707 if (!kvm_is_error_hva(hva) && writable)
1708 *writable = !memslot_is_readonly(slot);
1713 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1715 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1717 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1720 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
1722 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1724 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1727 static inline int check_user_page_hwpoison(unsigned long addr)
1729 int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
1731 rc = get_user_pages(addr, 1, flags, NULL, NULL);
1732 return rc == -EHWPOISON;
1736 * The fast path to get the writable pfn which will be stored in @pfn,
1737 * true indicates success, otherwise false is returned. It's also the
1738 * only part that runs if we can in atomic context.
1740 static bool hva_to_pfn_fast(unsigned long addr, bool write_fault,
1741 bool *writable, kvm_pfn_t *pfn)
1743 struct page *page[1];
1747 * Fast pin a writable pfn only if it is a write fault request
1748 * or the caller allows to map a writable pfn for a read fault
1751 if (!(write_fault || writable))
1754 npages = __get_user_pages_fast(addr, 1, 1, page);
1756 *pfn = page_to_pfn(page[0]);
1767 * The slow path to get the pfn of the specified host virtual address,
1768 * 1 indicates success, -errno is returned if error is detected.
1770 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1771 bool *writable, kvm_pfn_t *pfn)
1773 unsigned int flags = FOLL_HWPOISON;
1780 *writable = write_fault;
1783 flags |= FOLL_WRITE;
1785 flags |= FOLL_NOWAIT;
1787 npages = get_user_pages_unlocked(addr, 1, &page, flags);
1791 /* map read fault as writable if possible */
1792 if (unlikely(!write_fault) && writable) {
1795 if (__get_user_pages_fast(addr, 1, 1, &wpage) == 1) {
1801 *pfn = page_to_pfn(page);
1805 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1807 if (unlikely(!(vma->vm_flags & VM_READ)))
1810 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1816 static int hva_to_pfn_remapped(struct vm_area_struct *vma,
1817 unsigned long addr, bool *async,
1818 bool write_fault, bool *writable,
1824 r = follow_pfn(vma, addr, &pfn);
1827 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
1828 * not call the fault handler, so do it here.
1830 bool unlocked = false;
1831 r = fixup_user_fault(current, current->mm, addr,
1832 (write_fault ? FAULT_FLAG_WRITE : 0),
1839 r = follow_pfn(vma, addr, &pfn);
1849 * Get a reference here because callers of *hva_to_pfn* and
1850 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
1851 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
1852 * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will
1853 * simply do nothing for reserved pfns.
1855 * Whoever called remap_pfn_range is also going to call e.g.
1856 * unmap_mapping_range before the underlying pages are freed,
1857 * causing a call to our MMU notifier.
1866 * Pin guest page in memory and return its pfn.
1867 * @addr: host virtual address which maps memory to the guest
1868 * @atomic: whether this function can sleep
1869 * @async: whether this function need to wait IO complete if the
1870 * host page is not in the memory
1871 * @write_fault: whether we should get a writable host page
1872 * @writable: whether it allows to map a writable host page for !@write_fault
1874 * The function will map a writable host page for these two cases:
1875 * 1): @write_fault = true
1876 * 2): @write_fault = false && @writable, @writable will tell the caller
1877 * whether the mapping is writable.
1879 static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1880 bool write_fault, bool *writable)
1882 struct vm_area_struct *vma;
1886 /* we can do it either atomically or asynchronously, not both */
1887 BUG_ON(atomic && async);
1889 if (hva_to_pfn_fast(addr, write_fault, writable, &pfn))
1893 return KVM_PFN_ERR_FAULT;
1895 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1899 down_read(¤t->mm->mmap_sem);
1900 if (npages == -EHWPOISON ||
1901 (!async && check_user_page_hwpoison(addr))) {
1902 pfn = KVM_PFN_ERR_HWPOISON;
1907 vma = find_vma_intersection(current->mm, addr, addr + 1);
1910 pfn = KVM_PFN_ERR_FAULT;
1911 else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
1912 r = hva_to_pfn_remapped(vma, addr, async, write_fault, writable, &pfn);
1916 pfn = KVM_PFN_ERR_FAULT;
1918 if (async && vma_is_valid(vma, write_fault))
1920 pfn = KVM_PFN_ERR_FAULT;
1923 up_read(¤t->mm->mmap_sem);
1927 kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn,
1928 bool atomic, bool *async, bool write_fault,
1931 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1933 if (addr == KVM_HVA_ERR_RO_BAD) {
1936 return KVM_PFN_ERR_RO_FAULT;
1939 if (kvm_is_error_hva(addr)) {
1942 return KVM_PFN_NOSLOT;
1945 /* Do not map writable pfn in the readonly memslot. */
1946 if (writable && memslot_is_readonly(slot)) {
1951 return hva_to_pfn(addr, atomic, async, write_fault,
1954 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
1956 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1959 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
1960 write_fault, writable);
1962 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1964 kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1966 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1968 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
1970 kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1972 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1974 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1976 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
1978 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1980 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
1982 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1984 return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
1986 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1988 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
1990 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1992 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
1994 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1995 struct page **pages, int nr_pages)
2000 addr = gfn_to_hva_many(slot, gfn, &entry);
2001 if (kvm_is_error_hva(addr))
2004 if (entry < nr_pages)
2007 return __get_user_pages_fast(addr, nr_pages, 1, pages);
2009 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
2011 static struct page *kvm_pfn_to_page(kvm_pfn_t pfn)
2013 if (is_error_noslot_pfn(pfn))
2014 return KVM_ERR_PTR_BAD_PAGE;
2016 if (kvm_is_reserved_pfn(pfn)) {
2018 return KVM_ERR_PTR_BAD_PAGE;
2021 return pfn_to_page(pfn);
2024 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
2028 pfn = gfn_to_pfn(kvm, gfn);
2030 return kvm_pfn_to_page(pfn);
2032 EXPORT_SYMBOL_GPL(gfn_to_page);
2034 void kvm_release_pfn(kvm_pfn_t pfn, bool dirty, struct gfn_to_pfn_cache *cache)
2040 cache->pfn = cache->gfn = 0;
2043 kvm_release_pfn_dirty(pfn);
2045 kvm_release_pfn_clean(pfn);
2048 static void kvm_cache_gfn_to_pfn(struct kvm_memory_slot *slot, gfn_t gfn,
2049 struct gfn_to_pfn_cache *cache, u64 gen)
2051 kvm_release_pfn(cache->pfn, cache->dirty, cache);
2053 cache->pfn = gfn_to_pfn_memslot(slot, gfn);
2055 cache->dirty = false;
2056 cache->generation = gen;
2059 static int __kvm_map_gfn(struct kvm_memslots *slots, gfn_t gfn,
2060 struct kvm_host_map *map,
2061 struct gfn_to_pfn_cache *cache,
2066 struct page *page = KVM_UNMAPPED_PAGE;
2067 struct kvm_memory_slot *slot = __gfn_to_memslot(slots, gfn);
2068 u64 gen = slots->generation;
2074 if (!cache->pfn || cache->gfn != gfn ||
2075 cache->generation != gen) {
2078 kvm_cache_gfn_to_pfn(slot, gfn, cache, gen);
2084 pfn = gfn_to_pfn_memslot(slot, gfn);
2086 if (is_error_noslot_pfn(pfn))
2089 if (pfn_valid(pfn)) {
2090 page = pfn_to_page(pfn);
2092 hva = kmap_atomic(page);
2095 #ifdef CONFIG_HAS_IOMEM
2096 } else if (!atomic) {
2097 hva = memremap(pfn_to_hpa(pfn), PAGE_SIZE, MEMREMAP_WB);
2114 int kvm_map_gfn(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map,
2115 struct gfn_to_pfn_cache *cache, bool atomic)
2117 return __kvm_map_gfn(kvm_memslots(vcpu->kvm), gfn, map,
2120 EXPORT_SYMBOL_GPL(kvm_map_gfn);
2122 int kvm_vcpu_map(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map)
2124 return __kvm_map_gfn(kvm_vcpu_memslots(vcpu), gfn, map,
2127 EXPORT_SYMBOL_GPL(kvm_vcpu_map);
2129 static void __kvm_unmap_gfn(struct kvm_memory_slot *memslot,
2130 struct kvm_host_map *map,
2131 struct gfn_to_pfn_cache *cache,
2132 bool dirty, bool atomic)
2140 if (map->page != KVM_UNMAPPED_PAGE) {
2142 kunmap_atomic(map->hva);
2146 #ifdef CONFIG_HAS_IOMEM
2150 WARN_ONCE(1, "Unexpected unmapping in atomic context");
2154 mark_page_dirty_in_slot(memslot, map->gfn);
2157 cache->dirty |= dirty;
2159 kvm_release_pfn(map->pfn, dirty, NULL);
2165 int kvm_unmap_gfn(struct kvm_vcpu *vcpu, struct kvm_host_map *map,
2166 struct gfn_to_pfn_cache *cache, bool dirty, bool atomic)
2168 __kvm_unmap_gfn(gfn_to_memslot(vcpu->kvm, map->gfn), map,
2169 cache, dirty, atomic);
2172 EXPORT_SYMBOL_GPL(kvm_unmap_gfn);
2174 void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map, bool dirty)
2176 __kvm_unmap_gfn(kvm_vcpu_gfn_to_memslot(vcpu, map->gfn), map, NULL,
2179 EXPORT_SYMBOL_GPL(kvm_vcpu_unmap);
2181 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
2185 pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
2187 return kvm_pfn_to_page(pfn);
2189 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
2191 void kvm_release_page_clean(struct page *page)
2193 WARN_ON(is_error_page(page));
2195 kvm_release_pfn_clean(page_to_pfn(page));
2197 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
2199 void kvm_release_pfn_clean(kvm_pfn_t pfn)
2201 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
2202 put_page(pfn_to_page(pfn));
2204 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
2206 void kvm_release_page_dirty(struct page *page)
2208 WARN_ON(is_error_page(page));
2210 kvm_release_pfn_dirty(page_to_pfn(page));
2212 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
2214 void kvm_release_pfn_dirty(kvm_pfn_t pfn)
2216 kvm_set_pfn_dirty(pfn);
2217 kvm_release_pfn_clean(pfn);
2219 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
2221 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
2223 if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn))
2224 SetPageDirty(pfn_to_page(pfn));
2226 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
2228 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
2230 if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn))
2231 mark_page_accessed(pfn_to_page(pfn));
2233 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
2235 void kvm_get_pfn(kvm_pfn_t pfn)
2237 if (!kvm_is_reserved_pfn(pfn))
2238 get_page(pfn_to_page(pfn));
2240 EXPORT_SYMBOL_GPL(kvm_get_pfn);
2242 static int next_segment(unsigned long len, int offset)
2244 if (len > PAGE_SIZE - offset)
2245 return PAGE_SIZE - offset;
2250 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
2251 void *data, int offset, int len)
2256 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
2257 if (kvm_is_error_hva(addr))
2259 r = __copy_from_user(data, (void __user *)addr + offset, len);
2265 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
2268 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2270 return __kvm_read_guest_page(slot, gfn, data, offset, len);
2272 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
2274 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
2275 int offset, int len)
2277 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2279 return __kvm_read_guest_page(slot, gfn, data, offset, len);
2281 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
2283 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
2285 gfn_t gfn = gpa >> PAGE_SHIFT;
2287 int offset = offset_in_page(gpa);
2290 while ((seg = next_segment(len, offset)) != 0) {
2291 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
2301 EXPORT_SYMBOL_GPL(kvm_read_guest);
2303 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
2305 gfn_t gfn = gpa >> PAGE_SHIFT;
2307 int offset = offset_in_page(gpa);
2310 while ((seg = next_segment(len, offset)) != 0) {
2311 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
2321 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
2323 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
2324 void *data, int offset, unsigned long len)
2329 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
2330 if (kvm_is_error_hva(addr))
2332 pagefault_disable();
2333 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
2340 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
2341 void *data, unsigned long len)
2343 gfn_t gfn = gpa >> PAGE_SHIFT;
2344 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2345 int offset = offset_in_page(gpa);
2347 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
2349 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
2351 static int __kvm_write_guest_page(struct kvm_memory_slot *memslot, gfn_t gfn,
2352 const void *data, int offset, int len)
2357 addr = gfn_to_hva_memslot(memslot, gfn);
2358 if (kvm_is_error_hva(addr))
2360 r = __copy_to_user((void __user *)addr + offset, data, len);
2363 mark_page_dirty_in_slot(memslot, gfn);
2367 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
2368 const void *data, int offset, int len)
2370 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2372 return __kvm_write_guest_page(slot, gfn, data, offset, len);
2374 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
2376 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
2377 const void *data, int offset, int len)
2379 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2381 return __kvm_write_guest_page(slot, gfn, data, offset, len);
2383 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
2385 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
2388 gfn_t gfn = gpa >> PAGE_SHIFT;
2390 int offset = offset_in_page(gpa);
2393 while ((seg = next_segment(len, offset)) != 0) {
2394 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
2404 EXPORT_SYMBOL_GPL(kvm_write_guest);
2406 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
2409 gfn_t gfn = gpa >> PAGE_SHIFT;
2411 int offset = offset_in_page(gpa);
2414 while ((seg = next_segment(len, offset)) != 0) {
2415 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
2425 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
2427 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
2428 struct gfn_to_hva_cache *ghc,
2429 gpa_t gpa, unsigned long len)
2431 int offset = offset_in_page(gpa);
2432 gfn_t start_gfn = gpa >> PAGE_SHIFT;
2433 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
2434 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
2435 gfn_t nr_pages_avail;
2437 /* Update ghc->generation before performing any error checks. */
2438 ghc->generation = slots->generation;
2440 if (start_gfn > end_gfn) {
2441 ghc->hva = KVM_HVA_ERR_BAD;
2446 * If the requested region crosses two memslots, we still
2447 * verify that the entire region is valid here.
2449 for ( ; start_gfn <= end_gfn; start_gfn += nr_pages_avail) {
2450 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
2451 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
2453 if (kvm_is_error_hva(ghc->hva))
2457 /* Use the slow path for cross page reads and writes. */
2458 if (nr_pages_needed == 1)
2461 ghc->memslot = NULL;
2468 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2469 gpa_t gpa, unsigned long len)
2471 struct kvm_memslots *slots = kvm_memslots(kvm);
2472 return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len);
2474 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
2476 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2477 void *data, unsigned int offset,
2480 struct kvm_memslots *slots = kvm_memslots(kvm);
2482 gpa_t gpa = ghc->gpa + offset;
2484 BUG_ON(len + offset > ghc->len);
2486 if (slots->generation != ghc->generation) {
2487 if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len))
2491 if (kvm_is_error_hva(ghc->hva))
2494 if (unlikely(!ghc->memslot))
2495 return kvm_write_guest(kvm, gpa, data, len);
2497 r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
2500 mark_page_dirty_in_slot(ghc->memslot, gpa >> PAGE_SHIFT);
2504 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached);
2506 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2507 void *data, unsigned long len)
2509 return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len);
2511 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
2513 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2514 void *data, unsigned long len)
2516 struct kvm_memslots *slots = kvm_memslots(kvm);
2519 BUG_ON(len > ghc->len);
2521 if (slots->generation != ghc->generation) {
2522 if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len))
2526 if (kvm_is_error_hva(ghc->hva))
2529 if (unlikely(!ghc->memslot))
2530 return kvm_read_guest(kvm, ghc->gpa, data, len);
2532 r = __copy_from_user(data, (void __user *)ghc->hva, len);
2538 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
2540 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
2542 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
2544 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
2546 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
2548 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
2550 gfn_t gfn = gpa >> PAGE_SHIFT;
2552 int offset = offset_in_page(gpa);
2555 while ((seg = next_segment(len, offset)) != 0) {
2556 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
2565 EXPORT_SYMBOL_GPL(kvm_clear_guest);
2567 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot,
2570 if (memslot && memslot->dirty_bitmap) {
2571 unsigned long rel_gfn = gfn - memslot->base_gfn;
2573 set_bit_le(rel_gfn, memslot->dirty_bitmap);
2577 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
2579 struct kvm_memory_slot *memslot;
2581 memslot = gfn_to_memslot(kvm, gfn);
2582 mark_page_dirty_in_slot(memslot, gfn);
2584 EXPORT_SYMBOL_GPL(mark_page_dirty);
2586 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
2588 struct kvm_memory_slot *memslot;
2590 memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2591 mark_page_dirty_in_slot(memslot, gfn);
2593 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
2595 void kvm_sigset_activate(struct kvm_vcpu *vcpu)
2597 if (!vcpu->sigset_active)
2601 * This does a lockless modification of ->real_blocked, which is fine
2602 * because, only current can change ->real_blocked and all readers of
2603 * ->real_blocked don't care as long ->real_blocked is always a subset
2606 sigprocmask(SIG_SETMASK, &vcpu->sigset, ¤t->real_blocked);
2609 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu)
2611 if (!vcpu->sigset_active)
2614 sigprocmask(SIG_SETMASK, ¤t->real_blocked, NULL);
2615 sigemptyset(¤t->real_blocked);
2618 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
2620 unsigned int old, val, grow, grow_start;
2622 old = val = vcpu->halt_poll_ns;
2623 grow_start = READ_ONCE(halt_poll_ns_grow_start);
2624 grow = READ_ONCE(halt_poll_ns_grow);
2629 if (val < grow_start)
2632 if (val > halt_poll_ns)
2635 vcpu->halt_poll_ns = val;
2637 trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
2640 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
2642 unsigned int old, val, shrink;
2644 old = val = vcpu->halt_poll_ns;
2645 shrink = READ_ONCE(halt_poll_ns_shrink);
2651 vcpu->halt_poll_ns = val;
2652 trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
2655 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
2658 int idx = srcu_read_lock(&vcpu->kvm->srcu);
2660 if (kvm_arch_vcpu_runnable(vcpu)) {
2661 kvm_make_request(KVM_REQ_UNHALT, vcpu);
2664 if (kvm_cpu_has_pending_timer(vcpu))
2666 if (signal_pending(current))
2671 srcu_read_unlock(&vcpu->kvm->srcu, idx);
2676 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2678 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
2681 DECLARE_SWAITQUEUE(wait);
2682 bool waited = false;
2685 kvm_arch_vcpu_blocking(vcpu);
2687 start = cur = ktime_get();
2688 if (vcpu->halt_poll_ns && !kvm_arch_no_poll(vcpu)) {
2689 ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns);
2691 ++vcpu->stat.halt_attempted_poll;
2694 * This sets KVM_REQ_UNHALT if an interrupt
2697 if (kvm_vcpu_check_block(vcpu) < 0) {
2698 ++vcpu->stat.halt_successful_poll;
2699 if (!vcpu_valid_wakeup(vcpu))
2700 ++vcpu->stat.halt_poll_invalid;
2704 } while (single_task_running() && ktime_before(cur, stop));
2708 prepare_to_swait_exclusive(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
2710 if (kvm_vcpu_check_block(vcpu) < 0)
2717 finish_swait(&vcpu->wq, &wait);
2720 kvm_arch_vcpu_unblocking(vcpu);
2721 block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
2723 if (!kvm_arch_no_poll(vcpu)) {
2724 if (!vcpu_valid_wakeup(vcpu)) {
2725 shrink_halt_poll_ns(vcpu);
2726 } else if (vcpu->kvm->max_halt_poll_ns) {
2727 if (block_ns <= vcpu->halt_poll_ns)
2729 /* we had a long block, shrink polling */
2730 else if (vcpu->halt_poll_ns &&
2731 block_ns > vcpu->kvm->max_halt_poll_ns)
2732 shrink_halt_poll_ns(vcpu);
2733 /* we had a short halt and our poll time is too small */
2734 else if (vcpu->halt_poll_ns < vcpu->kvm->max_halt_poll_ns &&
2735 block_ns < vcpu->kvm->max_halt_poll_ns)
2736 grow_halt_poll_ns(vcpu);
2738 vcpu->halt_poll_ns = 0;
2742 trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu));
2743 kvm_arch_vcpu_block_finish(vcpu);
2745 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
2747 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
2749 struct swait_queue_head *wqp;
2751 wqp = kvm_arch_vcpu_wq(vcpu);
2752 if (swq_has_sleeper(wqp)) {
2754 WRITE_ONCE(vcpu->ready, true);
2755 ++vcpu->stat.halt_wakeup;
2761 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
2765 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2767 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
2770 int cpu = vcpu->cpu;
2772 if (kvm_vcpu_wake_up(vcpu))
2776 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
2777 if (kvm_arch_vcpu_should_kick(vcpu))
2778 smp_send_reschedule(cpu);
2781 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
2782 #endif /* !CONFIG_S390 */
2784 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
2787 struct task_struct *task = NULL;
2791 pid = rcu_dereference(target->pid);
2793 task = get_pid_task(pid, PIDTYPE_PID);
2797 ret = yield_to(task, 1);
2798 put_task_struct(task);
2802 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
2805 * Helper that checks whether a VCPU is eligible for directed yield.
2806 * Most eligible candidate to yield is decided by following heuristics:
2808 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2809 * (preempted lock holder), indicated by @in_spin_loop.
2810 * Set at the beiginning and cleared at the end of interception/PLE handler.
2812 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2813 * chance last time (mostly it has become eligible now since we have probably
2814 * yielded to lockholder in last iteration. This is done by toggling
2815 * @dy_eligible each time a VCPU checked for eligibility.)
2817 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2818 * to preempted lock-holder could result in wrong VCPU selection and CPU
2819 * burning. Giving priority for a potential lock-holder increases lock
2822 * Since algorithm is based on heuristics, accessing another VCPU data without
2823 * locking does not harm. It may result in trying to yield to same VCPU, fail
2824 * and continue with next VCPU and so on.
2826 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
2828 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2831 eligible = !vcpu->spin_loop.in_spin_loop ||
2832 vcpu->spin_loop.dy_eligible;
2834 if (vcpu->spin_loop.in_spin_loop)
2835 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
2844 * Unlike kvm_arch_vcpu_runnable, this function is called outside
2845 * a vcpu_load/vcpu_put pair. However, for most architectures
2846 * kvm_arch_vcpu_runnable does not require vcpu_load.
2848 bool __weak kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
2850 return kvm_arch_vcpu_runnable(vcpu);
2853 static bool vcpu_dy_runnable(struct kvm_vcpu *vcpu)
2855 if (kvm_arch_dy_runnable(vcpu))
2858 #ifdef CONFIG_KVM_ASYNC_PF
2859 if (!list_empty_careful(&vcpu->async_pf.done))
2866 void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode)
2868 struct kvm *kvm = me->kvm;
2869 struct kvm_vcpu *vcpu;
2870 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
2876 kvm_vcpu_set_in_spin_loop(me, true);
2878 * We boost the priority of a VCPU that is runnable but not
2879 * currently running, because it got preempted by something
2880 * else and called schedule in __vcpu_run. Hopefully that
2881 * VCPU is holding the lock that we need and will release it.
2882 * We approximate round-robin by starting at the last boosted VCPU.
2884 for (pass = 0; pass < 2 && !yielded && try; pass++) {
2885 kvm_for_each_vcpu(i, vcpu, kvm) {
2886 if (!pass && i <= last_boosted_vcpu) {
2887 i = last_boosted_vcpu;
2889 } else if (pass && i > last_boosted_vcpu)
2891 if (!READ_ONCE(vcpu->ready))
2895 if (swait_active(&vcpu->wq) && !vcpu_dy_runnable(vcpu))
2897 if (READ_ONCE(vcpu->preempted) && yield_to_kernel_mode &&
2898 !kvm_arch_vcpu_in_kernel(vcpu))
2900 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
2903 yielded = kvm_vcpu_yield_to(vcpu);
2905 kvm->last_boosted_vcpu = i;
2907 } else if (yielded < 0) {
2914 kvm_vcpu_set_in_spin_loop(me, false);
2916 /* Ensure vcpu is not eligible during next spinloop */
2917 kvm_vcpu_set_dy_eligible(me, false);
2919 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
2921 static vm_fault_t kvm_vcpu_fault(struct vm_fault *vmf)
2923 struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data;
2926 if (vmf->pgoff == 0)
2927 page = virt_to_page(vcpu->run);
2929 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
2930 page = virt_to_page(vcpu->arch.pio_data);
2932 #ifdef CONFIG_KVM_MMIO
2933 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
2934 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
2937 return kvm_arch_vcpu_fault(vcpu, vmf);
2943 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
2944 .fault = kvm_vcpu_fault,
2947 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2949 vma->vm_ops = &kvm_vcpu_vm_ops;
2953 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2955 struct kvm_vcpu *vcpu = filp->private_data;
2957 debugfs_remove_recursive(vcpu->debugfs_dentry);
2958 kvm_put_kvm(vcpu->kvm);
2962 static struct file_operations kvm_vcpu_fops = {
2963 .release = kvm_vcpu_release,
2964 .unlocked_ioctl = kvm_vcpu_ioctl,
2965 .mmap = kvm_vcpu_mmap,
2966 .llseek = noop_llseek,
2967 KVM_COMPAT(kvm_vcpu_compat_ioctl),
2971 * Allocates an inode for the vcpu.
2973 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2975 char name[8 + 1 + ITOA_MAX_LEN + 1];
2977 snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id);
2978 return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
2981 static void kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
2983 #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS
2984 char dir_name[ITOA_MAX_LEN * 2];
2986 if (!debugfs_initialized())
2989 snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
2990 vcpu->debugfs_dentry = debugfs_create_dir(dir_name,
2991 vcpu->kvm->debugfs_dentry);
2993 kvm_arch_create_vcpu_debugfs(vcpu);
2998 * Creates some virtual cpus. Good luck creating more than one.
3000 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
3003 struct kvm_vcpu *vcpu;
3006 if (id >= KVM_MAX_VCPU_ID)
3009 mutex_lock(&kvm->lock);
3010 if (kvm->created_vcpus == KVM_MAX_VCPUS) {
3011 mutex_unlock(&kvm->lock);
3015 kvm->created_vcpus++;
3016 mutex_unlock(&kvm->lock);
3018 r = kvm_arch_vcpu_precreate(kvm, id);
3020 goto vcpu_decrement;
3022 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
3025 goto vcpu_decrement;
3028 BUILD_BUG_ON(sizeof(struct kvm_run) > PAGE_SIZE);
3029 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
3034 vcpu->run = page_address(page);
3036 kvm_vcpu_init(vcpu, kvm, id);
3038 r = kvm_arch_vcpu_create(vcpu);
3040 goto vcpu_free_run_page;
3042 mutex_lock(&kvm->lock);
3043 if (kvm_get_vcpu_by_id(kvm, id)) {
3045 goto unlock_vcpu_destroy;
3048 vcpu->vcpu_idx = atomic_read(&kvm->online_vcpus);
3049 BUG_ON(kvm->vcpus[vcpu->vcpu_idx]);
3051 /* Now it's all set up, let userspace reach it */
3053 r = create_vcpu_fd(vcpu);
3055 kvm_put_kvm_no_destroy(kvm);
3056 goto unlock_vcpu_destroy;
3059 kvm->vcpus[vcpu->vcpu_idx] = vcpu;
3062 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
3063 * before kvm->online_vcpu's incremented value.
3066 atomic_inc(&kvm->online_vcpus);
3068 mutex_unlock(&kvm->lock);
3069 kvm_arch_vcpu_postcreate(vcpu);
3070 kvm_create_vcpu_debugfs(vcpu);
3073 unlock_vcpu_destroy:
3074 mutex_unlock(&kvm->lock);
3075 kvm_arch_vcpu_destroy(vcpu);
3077 free_page((unsigned long)vcpu->run);
3079 kmem_cache_free(kvm_vcpu_cache, vcpu);
3081 mutex_lock(&kvm->lock);
3082 kvm->created_vcpus--;
3083 mutex_unlock(&kvm->lock);
3087 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
3090 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
3091 vcpu->sigset_active = 1;
3092 vcpu->sigset = *sigset;
3094 vcpu->sigset_active = 0;
3098 static long kvm_vcpu_ioctl(struct file *filp,
3099 unsigned int ioctl, unsigned long arg)
3101 struct kvm_vcpu *vcpu = filp->private_data;
3102 void __user *argp = (void __user *)arg;
3104 struct kvm_fpu *fpu = NULL;
3105 struct kvm_sregs *kvm_sregs = NULL;
3107 if (vcpu->kvm->mm != current->mm)
3110 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
3114 * Some architectures have vcpu ioctls that are asynchronous to vcpu
3115 * execution; mutex_lock() would break them.
3117 r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg);
3118 if (r != -ENOIOCTLCMD)
3121 if (mutex_lock_killable(&vcpu->mutex))
3129 oldpid = rcu_access_pointer(vcpu->pid);
3130 if (unlikely(oldpid != task_pid(current))) {
3131 /* The thread running this VCPU changed. */
3134 r = kvm_arch_vcpu_run_pid_change(vcpu);
3138 newpid = get_task_pid(current, PIDTYPE_PID);
3139 rcu_assign_pointer(vcpu->pid, newpid);
3144 r = kvm_arch_vcpu_ioctl_run(vcpu);
3145 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
3148 case KVM_GET_REGS: {
3149 struct kvm_regs *kvm_regs;
3152 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL_ACCOUNT);
3155 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
3159 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
3166 case KVM_SET_REGS: {
3167 struct kvm_regs *kvm_regs;
3169 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
3170 if (IS_ERR(kvm_regs)) {
3171 r = PTR_ERR(kvm_regs);
3174 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
3178 case KVM_GET_SREGS: {
3179 kvm_sregs = kzalloc(sizeof(struct kvm_sregs),
3180 GFP_KERNEL_ACCOUNT);
3184 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
3188 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
3193 case KVM_SET_SREGS: {
3194 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
3195 if (IS_ERR(kvm_sregs)) {
3196 r = PTR_ERR(kvm_sregs);
3200 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
3203 case KVM_GET_MP_STATE: {
3204 struct kvm_mp_state mp_state;
3206 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
3210 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
3215 case KVM_SET_MP_STATE: {
3216 struct kvm_mp_state mp_state;
3219 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
3221 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
3224 case KVM_TRANSLATE: {
3225 struct kvm_translation tr;
3228 if (copy_from_user(&tr, argp, sizeof(tr)))
3230 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
3234 if (copy_to_user(argp, &tr, sizeof(tr)))
3239 case KVM_SET_GUEST_DEBUG: {
3240 struct kvm_guest_debug dbg;
3243 if (copy_from_user(&dbg, argp, sizeof(dbg)))
3245 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
3248 case KVM_SET_SIGNAL_MASK: {
3249 struct kvm_signal_mask __user *sigmask_arg = argp;
3250 struct kvm_signal_mask kvm_sigmask;
3251 sigset_t sigset, *p;
3256 if (copy_from_user(&kvm_sigmask, argp,
3257 sizeof(kvm_sigmask)))
3260 if (kvm_sigmask.len != sizeof(sigset))
3263 if (copy_from_user(&sigset, sigmask_arg->sigset,
3268 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
3272 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL_ACCOUNT);
3276 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
3280 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
3286 fpu = memdup_user(argp, sizeof(*fpu));
3292 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
3296 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
3299 mutex_unlock(&vcpu->mutex);
3305 #ifdef CONFIG_KVM_COMPAT
3306 static long kvm_vcpu_compat_ioctl(struct file *filp,
3307 unsigned int ioctl, unsigned long arg)
3309 struct kvm_vcpu *vcpu = filp->private_data;
3310 void __user *argp = compat_ptr(arg);
3313 if (vcpu->kvm->mm != current->mm)
3317 case KVM_SET_SIGNAL_MASK: {
3318 struct kvm_signal_mask __user *sigmask_arg = argp;
3319 struct kvm_signal_mask kvm_sigmask;
3324 if (copy_from_user(&kvm_sigmask, argp,
3325 sizeof(kvm_sigmask)))
3328 if (kvm_sigmask.len != sizeof(compat_sigset_t))
3331 if (get_compat_sigset(&sigset, (void *)sigmask_arg->sigset))
3333 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
3335 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
3339 r = kvm_vcpu_ioctl(filp, ioctl, arg);
3347 static int kvm_device_mmap(struct file *filp, struct vm_area_struct *vma)
3349 struct kvm_device *dev = filp->private_data;
3352 return dev->ops->mmap(dev, vma);
3357 static int kvm_device_ioctl_attr(struct kvm_device *dev,
3358 int (*accessor)(struct kvm_device *dev,
3359 struct kvm_device_attr *attr),
3362 struct kvm_device_attr attr;
3367 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
3370 return accessor(dev, &attr);
3373 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
3376 struct kvm_device *dev = filp->private_data;
3378 if (dev->kvm->mm != current->mm)
3382 case KVM_SET_DEVICE_ATTR:
3383 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
3384 case KVM_GET_DEVICE_ATTR:
3385 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
3386 case KVM_HAS_DEVICE_ATTR:
3387 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
3389 if (dev->ops->ioctl)
3390 return dev->ops->ioctl(dev, ioctl, arg);
3396 static int kvm_device_release(struct inode *inode, struct file *filp)
3398 struct kvm_device *dev = filp->private_data;
3399 struct kvm *kvm = dev->kvm;
3401 if (dev->ops->release) {
3402 mutex_lock(&kvm->lock);
3403 list_del(&dev->vm_node);
3404 dev->ops->release(dev);
3405 mutex_unlock(&kvm->lock);
3412 static const struct file_operations kvm_device_fops = {
3413 .unlocked_ioctl = kvm_device_ioctl,
3414 .release = kvm_device_release,
3415 KVM_COMPAT(kvm_device_ioctl),
3416 .mmap = kvm_device_mmap,
3419 struct kvm_device *kvm_device_from_filp(struct file *filp)
3421 if (filp->f_op != &kvm_device_fops)
3424 return filp->private_data;
3427 static const struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
3428 #ifdef CONFIG_KVM_MPIC
3429 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
3430 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
3434 int kvm_register_device_ops(const struct kvm_device_ops *ops, u32 type)
3436 if (type >= ARRAY_SIZE(kvm_device_ops_table))
3439 if (kvm_device_ops_table[type] != NULL)
3442 kvm_device_ops_table[type] = ops;
3446 void kvm_unregister_device_ops(u32 type)
3448 if (kvm_device_ops_table[type] != NULL)
3449 kvm_device_ops_table[type] = NULL;
3452 static int kvm_ioctl_create_device(struct kvm *kvm,
3453 struct kvm_create_device *cd)
3455 const struct kvm_device_ops *ops = NULL;
3456 struct kvm_device *dev;
3457 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
3461 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
3464 type = array_index_nospec(cd->type, ARRAY_SIZE(kvm_device_ops_table));
3465 ops = kvm_device_ops_table[type];
3472 dev = kzalloc(sizeof(*dev), GFP_KERNEL_ACCOUNT);
3479 mutex_lock(&kvm->lock);
3480 ret = ops->create(dev, type);
3482 mutex_unlock(&kvm->lock);
3486 list_add(&dev->vm_node, &kvm->devices);
3487 mutex_unlock(&kvm->lock);
3493 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
3495 kvm_put_kvm_no_destroy(kvm);
3496 mutex_lock(&kvm->lock);
3497 list_del(&dev->vm_node);
3498 mutex_unlock(&kvm->lock);
3507 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
3510 case KVM_CAP_USER_MEMORY:
3511 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
3512 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
3513 case KVM_CAP_INTERNAL_ERROR_DATA:
3514 #ifdef CONFIG_HAVE_KVM_MSI
3515 case KVM_CAP_SIGNAL_MSI:
3517 #ifdef CONFIG_HAVE_KVM_IRQFD
3519 case KVM_CAP_IRQFD_RESAMPLE:
3521 case KVM_CAP_IOEVENTFD_ANY_LENGTH:
3522 case KVM_CAP_CHECK_EXTENSION_VM:
3523 case KVM_CAP_ENABLE_CAP_VM:
3524 case KVM_CAP_HALT_POLL:
3526 #ifdef CONFIG_KVM_MMIO
3527 case KVM_CAP_COALESCED_MMIO:
3528 return KVM_COALESCED_MMIO_PAGE_OFFSET;
3529 case KVM_CAP_COALESCED_PIO:
3532 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3533 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2:
3534 return KVM_DIRTY_LOG_MANUAL_CAPS;
3536 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3537 case KVM_CAP_IRQ_ROUTING:
3538 return KVM_MAX_IRQ_ROUTES;
3540 #if KVM_ADDRESS_SPACE_NUM > 1
3541 case KVM_CAP_MULTI_ADDRESS_SPACE:
3542 return KVM_ADDRESS_SPACE_NUM;
3544 case KVM_CAP_NR_MEMSLOTS:
3545 return KVM_USER_MEM_SLOTS;
3549 return kvm_vm_ioctl_check_extension(kvm, arg);
3552 int __attribute__((weak)) kvm_vm_ioctl_enable_cap(struct kvm *kvm,
3553 struct kvm_enable_cap *cap)
3558 static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm,
3559 struct kvm_enable_cap *cap)
3562 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3563 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2: {
3564 u64 allowed_options = KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE;
3566 if (cap->args[0] & KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE)
3567 allowed_options = KVM_DIRTY_LOG_MANUAL_CAPS;
3569 if (cap->flags || (cap->args[0] & ~allowed_options))
3571 kvm->manual_dirty_log_protect = cap->args[0];
3575 case KVM_CAP_HALT_POLL: {
3576 if (cap->flags || cap->args[0] != (unsigned int)cap->args[0])
3579 kvm->max_halt_poll_ns = cap->args[0];
3583 return kvm_vm_ioctl_enable_cap(kvm, cap);
3587 static long kvm_vm_ioctl(struct file *filp,
3588 unsigned int ioctl, unsigned long arg)
3590 struct kvm *kvm = filp->private_data;
3591 void __user *argp = (void __user *)arg;
3594 if (kvm->mm != current->mm)
3597 case KVM_CREATE_VCPU:
3598 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
3600 case KVM_ENABLE_CAP: {
3601 struct kvm_enable_cap cap;
3604 if (copy_from_user(&cap, argp, sizeof(cap)))
3606 r = kvm_vm_ioctl_enable_cap_generic(kvm, &cap);
3609 case KVM_SET_USER_MEMORY_REGION: {
3610 struct kvm_userspace_memory_region kvm_userspace_mem;
3613 if (copy_from_user(&kvm_userspace_mem, argp,
3614 sizeof(kvm_userspace_mem)))
3617 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
3620 case KVM_GET_DIRTY_LOG: {
3621 struct kvm_dirty_log log;
3624 if (copy_from_user(&log, argp, sizeof(log)))
3626 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3629 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3630 case KVM_CLEAR_DIRTY_LOG: {
3631 struct kvm_clear_dirty_log log;
3634 if (copy_from_user(&log, argp, sizeof(log)))
3636 r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
3640 #ifdef CONFIG_KVM_MMIO
3641 case KVM_REGISTER_COALESCED_MMIO: {
3642 struct kvm_coalesced_mmio_zone zone;
3645 if (copy_from_user(&zone, argp, sizeof(zone)))
3647 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
3650 case KVM_UNREGISTER_COALESCED_MMIO: {
3651 struct kvm_coalesced_mmio_zone zone;
3654 if (copy_from_user(&zone, argp, sizeof(zone)))
3656 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
3661 struct kvm_irqfd data;
3664 if (copy_from_user(&data, argp, sizeof(data)))
3666 r = kvm_irqfd(kvm, &data);
3669 case KVM_IOEVENTFD: {
3670 struct kvm_ioeventfd data;
3673 if (copy_from_user(&data, argp, sizeof(data)))
3675 r = kvm_ioeventfd(kvm, &data);
3678 #ifdef CONFIG_HAVE_KVM_MSI
3679 case KVM_SIGNAL_MSI: {
3683 if (copy_from_user(&msi, argp, sizeof(msi)))
3685 r = kvm_send_userspace_msi(kvm, &msi);
3689 #ifdef __KVM_HAVE_IRQ_LINE
3690 case KVM_IRQ_LINE_STATUS:
3691 case KVM_IRQ_LINE: {
3692 struct kvm_irq_level irq_event;
3695 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
3698 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
3699 ioctl == KVM_IRQ_LINE_STATUS);
3704 if (ioctl == KVM_IRQ_LINE_STATUS) {
3705 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
3713 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3714 case KVM_SET_GSI_ROUTING: {
3715 struct kvm_irq_routing routing;
3716 struct kvm_irq_routing __user *urouting;
3717 struct kvm_irq_routing_entry *entries = NULL;
3720 if (copy_from_user(&routing, argp, sizeof(routing)))
3723 if (!kvm_arch_can_set_irq_routing(kvm))
3725 if (routing.nr > KVM_MAX_IRQ_ROUTES)
3731 entries = vmalloc(array_size(sizeof(*entries),
3737 if (copy_from_user(entries, urouting->entries,
3738 routing.nr * sizeof(*entries)))
3739 goto out_free_irq_routing;
3741 r = kvm_set_irq_routing(kvm, entries, routing.nr,
3743 out_free_irq_routing:
3747 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
3748 case KVM_CREATE_DEVICE: {
3749 struct kvm_create_device cd;
3752 if (copy_from_user(&cd, argp, sizeof(cd)))
3755 r = kvm_ioctl_create_device(kvm, &cd);
3760 if (copy_to_user(argp, &cd, sizeof(cd)))
3766 case KVM_CHECK_EXTENSION:
3767 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
3770 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
3776 #ifdef CONFIG_KVM_COMPAT
3777 struct compat_kvm_dirty_log {
3781 compat_uptr_t dirty_bitmap; /* one bit per page */
3786 static long kvm_vm_compat_ioctl(struct file *filp,
3787 unsigned int ioctl, unsigned long arg)
3789 struct kvm *kvm = filp->private_data;
3792 if (kvm->mm != current->mm)
3795 case KVM_GET_DIRTY_LOG: {
3796 struct compat_kvm_dirty_log compat_log;
3797 struct kvm_dirty_log log;
3799 if (copy_from_user(&compat_log, (void __user *)arg,
3800 sizeof(compat_log)))
3802 log.slot = compat_log.slot;
3803 log.padding1 = compat_log.padding1;
3804 log.padding2 = compat_log.padding2;
3805 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
3807 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3811 r = kvm_vm_ioctl(filp, ioctl, arg);
3817 static struct file_operations kvm_vm_fops = {
3818 .release = kvm_vm_release,
3819 .unlocked_ioctl = kvm_vm_ioctl,
3820 .llseek = noop_llseek,
3821 KVM_COMPAT(kvm_vm_compat_ioctl),
3824 static int kvm_dev_ioctl_create_vm(unsigned long type)
3830 kvm = kvm_create_vm(type);
3832 return PTR_ERR(kvm);
3833 #ifdef CONFIG_KVM_MMIO
3834 r = kvm_coalesced_mmio_init(kvm);
3838 r = get_unused_fd_flags(O_CLOEXEC);
3842 file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
3850 * Don't call kvm_put_kvm anymore at this point; file->f_op is
3851 * already set, with ->release() being kvm_vm_release(). In error
3852 * cases it will be called by the final fput(file) and will take
3853 * care of doing kvm_put_kvm(kvm).
3855 if (kvm_create_vm_debugfs(kvm, r) < 0) {
3860 kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm);
3862 fd_install(r, file);
3870 static long kvm_dev_ioctl(struct file *filp,
3871 unsigned int ioctl, unsigned long arg)
3876 case KVM_GET_API_VERSION:
3879 r = KVM_API_VERSION;
3882 r = kvm_dev_ioctl_create_vm(arg);
3884 case KVM_CHECK_EXTENSION:
3885 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
3887 case KVM_GET_VCPU_MMAP_SIZE:
3890 r = PAGE_SIZE; /* struct kvm_run */
3892 r += PAGE_SIZE; /* pio data page */
3894 #ifdef CONFIG_KVM_MMIO
3895 r += PAGE_SIZE; /* coalesced mmio ring page */
3898 case KVM_TRACE_ENABLE:
3899 case KVM_TRACE_PAUSE:
3900 case KVM_TRACE_DISABLE:
3904 return kvm_arch_dev_ioctl(filp, ioctl, arg);
3910 static struct file_operations kvm_chardev_ops = {
3911 .unlocked_ioctl = kvm_dev_ioctl,
3912 .llseek = noop_llseek,
3913 KVM_COMPAT(kvm_dev_ioctl),
3916 static struct miscdevice kvm_dev = {
3922 static void hardware_enable_nolock(void *junk)
3924 int cpu = raw_smp_processor_id();
3927 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
3930 cpumask_set_cpu(cpu, cpus_hardware_enabled);
3932 r = kvm_arch_hardware_enable();
3935 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3936 atomic_inc(&hardware_enable_failed);
3937 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
3941 static int kvm_starting_cpu(unsigned int cpu)
3943 raw_spin_lock(&kvm_count_lock);
3944 if (kvm_usage_count)
3945 hardware_enable_nolock(NULL);
3946 raw_spin_unlock(&kvm_count_lock);
3950 static void hardware_disable_nolock(void *junk)
3952 int cpu = raw_smp_processor_id();
3954 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
3956 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3957 kvm_arch_hardware_disable();
3960 static int kvm_dying_cpu(unsigned int cpu)
3962 raw_spin_lock(&kvm_count_lock);
3963 if (kvm_usage_count)
3964 hardware_disable_nolock(NULL);
3965 raw_spin_unlock(&kvm_count_lock);
3969 static void hardware_disable_all_nolock(void)
3971 BUG_ON(!kvm_usage_count);
3974 if (!kvm_usage_count)
3975 on_each_cpu(hardware_disable_nolock, NULL, 1);
3978 static void hardware_disable_all(void)
3980 raw_spin_lock(&kvm_count_lock);
3981 hardware_disable_all_nolock();
3982 raw_spin_unlock(&kvm_count_lock);
3985 static int hardware_enable_all(void)
3989 raw_spin_lock(&kvm_count_lock);
3992 if (kvm_usage_count == 1) {
3993 atomic_set(&hardware_enable_failed, 0);
3994 on_each_cpu(hardware_enable_nolock, NULL, 1);
3996 if (atomic_read(&hardware_enable_failed)) {
3997 hardware_disable_all_nolock();
4002 raw_spin_unlock(&kvm_count_lock);
4007 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
4011 * Some (well, at least mine) BIOSes hang on reboot if
4014 * And Intel TXT required VMX off for all cpu when system shutdown.
4016 pr_info("kvm: exiting hardware virtualization\n");
4017 kvm_rebooting = true;
4018 on_each_cpu(hardware_disable_nolock, NULL, 1);
4022 static struct notifier_block kvm_reboot_notifier = {
4023 .notifier_call = kvm_reboot,
4027 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
4031 for (i = 0; i < bus->dev_count; i++) {
4032 struct kvm_io_device *pos = bus->range[i].dev;
4034 kvm_iodevice_destructor(pos);
4039 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
4040 const struct kvm_io_range *r2)
4042 gpa_t addr1 = r1->addr;
4043 gpa_t addr2 = r2->addr;
4048 /* If r2->len == 0, match the exact address. If r2->len != 0,
4049 * accept any overlapping write. Any order is acceptable for
4050 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
4051 * we process all of them.
4064 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
4066 return kvm_io_bus_cmp(p1, p2);
4069 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
4070 gpa_t addr, int len)
4072 struct kvm_io_range *range, key;
4075 key = (struct kvm_io_range) {
4080 range = bsearch(&key, bus->range, bus->dev_count,
4081 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
4085 off = range - bus->range;
4087 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
4093 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
4094 struct kvm_io_range *range, const void *val)
4098 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
4102 while (idx < bus->dev_count &&
4103 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
4104 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
4113 /* kvm_io_bus_write - called under kvm->slots_lock */
4114 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
4115 int len, const void *val)
4117 struct kvm_io_bus *bus;
4118 struct kvm_io_range range;
4121 range = (struct kvm_io_range) {
4126 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
4129 r = __kvm_io_bus_write(vcpu, bus, &range, val);
4130 return r < 0 ? r : 0;
4132 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
4134 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
4135 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
4136 gpa_t addr, int len, const void *val, long cookie)
4138 struct kvm_io_bus *bus;
4139 struct kvm_io_range range;
4141 range = (struct kvm_io_range) {
4146 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
4150 /* First try the device referenced by cookie. */
4151 if ((cookie >= 0) && (cookie < bus->dev_count) &&
4152 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
4153 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
4158 * cookie contained garbage; fall back to search and return the
4159 * correct cookie value.
4161 return __kvm_io_bus_write(vcpu, bus, &range, val);
4164 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
4165 struct kvm_io_range *range, void *val)
4169 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
4173 while (idx < bus->dev_count &&
4174 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
4175 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
4184 /* kvm_io_bus_read - called under kvm->slots_lock */
4185 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
4188 struct kvm_io_bus *bus;
4189 struct kvm_io_range range;
4192 range = (struct kvm_io_range) {
4197 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
4200 r = __kvm_io_bus_read(vcpu, bus, &range, val);
4201 return r < 0 ? r : 0;
4204 /* Caller must hold slots_lock. */
4205 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
4206 int len, struct kvm_io_device *dev)
4209 struct kvm_io_bus *new_bus, *bus;
4210 struct kvm_io_range range;
4212 bus = kvm_get_bus(kvm, bus_idx);
4216 /* exclude ioeventfd which is limited by maximum fd */
4217 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
4220 new_bus = kmalloc(struct_size(bus, range, bus->dev_count + 1),
4221 GFP_KERNEL_ACCOUNT);
4225 range = (struct kvm_io_range) {
4231 for (i = 0; i < bus->dev_count; i++)
4232 if (kvm_io_bus_cmp(&bus->range[i], &range) > 0)
4235 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
4236 new_bus->dev_count++;
4237 new_bus->range[i] = range;
4238 memcpy(new_bus->range + i + 1, bus->range + i,
4239 (bus->dev_count - i) * sizeof(struct kvm_io_range));
4240 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
4241 synchronize_srcu_expedited(&kvm->srcu);
4247 /* Caller must hold slots_lock. */
4248 void kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
4249 struct kvm_io_device *dev)
4252 struct kvm_io_bus *new_bus, *bus;
4254 bus = kvm_get_bus(kvm, bus_idx);
4258 for (i = 0; i < bus->dev_count; i++)
4259 if (bus->range[i].dev == dev) {
4263 if (i == bus->dev_count)
4266 new_bus = kmalloc(struct_size(bus, range, bus->dev_count - 1),
4267 GFP_KERNEL_ACCOUNT);
4269 pr_err("kvm: failed to shrink bus, removing it completely\n");
4273 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
4274 new_bus->dev_count--;
4275 memcpy(new_bus->range + i, bus->range + i + 1,
4276 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
4279 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
4280 synchronize_srcu_expedited(&kvm->srcu);
4285 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
4288 struct kvm_io_bus *bus;
4289 int dev_idx, srcu_idx;
4290 struct kvm_io_device *iodev = NULL;
4292 srcu_idx = srcu_read_lock(&kvm->srcu);
4294 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
4298 dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
4302 iodev = bus->range[dev_idx].dev;
4305 srcu_read_unlock(&kvm->srcu, srcu_idx);
4309 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
4311 static int kvm_debugfs_open(struct inode *inode, struct file *file,
4312 int (*get)(void *, u64 *), int (*set)(void *, u64),
4315 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
4318 /* The debugfs files are a reference to the kvm struct which
4319 * is still valid when kvm_destroy_vm is called.
4320 * To avoid the race between open and the removal of the debugfs
4321 * directory we test against the users count.
4323 if (!refcount_inc_not_zero(&stat_data->kvm->users_count))
4326 if (simple_attr_open(inode, file, get,
4327 KVM_DBGFS_GET_MODE(stat_data->dbgfs_item) & 0222
4330 kvm_put_kvm(stat_data->kvm);
4337 static int kvm_debugfs_release(struct inode *inode, struct file *file)
4339 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
4342 simple_attr_release(inode, file);
4343 kvm_put_kvm(stat_data->kvm);
4348 static int kvm_get_stat_per_vm(struct kvm *kvm, size_t offset, u64 *val)
4350 *val = *(ulong *)((void *)kvm + offset);
4355 static int kvm_clear_stat_per_vm(struct kvm *kvm, size_t offset)
4357 *(ulong *)((void *)kvm + offset) = 0;
4362 static int kvm_get_stat_per_vcpu(struct kvm *kvm, size_t offset, u64 *val)
4365 struct kvm_vcpu *vcpu;
4369 kvm_for_each_vcpu(i, vcpu, kvm)
4370 *val += *(u64 *)((void *)vcpu + offset);
4375 static int kvm_clear_stat_per_vcpu(struct kvm *kvm, size_t offset)
4378 struct kvm_vcpu *vcpu;
4380 kvm_for_each_vcpu(i, vcpu, kvm)
4381 *(u64 *)((void *)vcpu + offset) = 0;
4386 static int kvm_stat_data_get(void *data, u64 *val)
4389 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
4391 switch (stat_data->dbgfs_item->kind) {
4393 r = kvm_get_stat_per_vm(stat_data->kvm,
4394 stat_data->dbgfs_item->offset, val);
4397 r = kvm_get_stat_per_vcpu(stat_data->kvm,
4398 stat_data->dbgfs_item->offset, val);
4405 static int kvm_stat_data_clear(void *data, u64 val)
4408 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
4413 switch (stat_data->dbgfs_item->kind) {
4415 r = kvm_clear_stat_per_vm(stat_data->kvm,
4416 stat_data->dbgfs_item->offset);
4419 r = kvm_clear_stat_per_vcpu(stat_data->kvm,
4420 stat_data->dbgfs_item->offset);
4427 static int kvm_stat_data_open(struct inode *inode, struct file *file)
4429 __simple_attr_check_format("%llu\n", 0ull);
4430 return kvm_debugfs_open(inode, file, kvm_stat_data_get,
4431 kvm_stat_data_clear, "%llu\n");
4434 static const struct file_operations stat_fops_per_vm = {
4435 .owner = THIS_MODULE,
4436 .open = kvm_stat_data_open,
4437 .release = kvm_debugfs_release,
4438 .read = simple_attr_read,
4439 .write = simple_attr_write,
4440 .llseek = no_llseek,
4443 static int vm_stat_get(void *_offset, u64 *val)
4445 unsigned offset = (long)_offset;
4450 mutex_lock(&kvm_lock);
4451 list_for_each_entry(kvm, &vm_list, vm_list) {
4452 kvm_get_stat_per_vm(kvm, offset, &tmp_val);
4455 mutex_unlock(&kvm_lock);
4459 static int vm_stat_clear(void *_offset, u64 val)
4461 unsigned offset = (long)_offset;
4467 mutex_lock(&kvm_lock);
4468 list_for_each_entry(kvm, &vm_list, vm_list) {
4469 kvm_clear_stat_per_vm(kvm, offset);
4471 mutex_unlock(&kvm_lock);
4476 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n");
4478 static int vcpu_stat_get(void *_offset, u64 *val)
4480 unsigned offset = (long)_offset;
4485 mutex_lock(&kvm_lock);
4486 list_for_each_entry(kvm, &vm_list, vm_list) {
4487 kvm_get_stat_per_vcpu(kvm, offset, &tmp_val);
4490 mutex_unlock(&kvm_lock);
4494 static int vcpu_stat_clear(void *_offset, u64 val)
4496 unsigned offset = (long)_offset;
4502 mutex_lock(&kvm_lock);
4503 list_for_each_entry(kvm, &vm_list, vm_list) {
4504 kvm_clear_stat_per_vcpu(kvm, offset);
4506 mutex_unlock(&kvm_lock);
4511 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear,
4514 static const struct file_operations *stat_fops[] = {
4515 [KVM_STAT_VCPU] = &vcpu_stat_fops,
4516 [KVM_STAT_VM] = &vm_stat_fops,
4519 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm)
4521 struct kobj_uevent_env *env;
4522 unsigned long long created, active;
4524 if (!kvm_dev.this_device || !kvm)
4527 mutex_lock(&kvm_lock);
4528 if (type == KVM_EVENT_CREATE_VM) {
4529 kvm_createvm_count++;
4531 } else if (type == KVM_EVENT_DESTROY_VM) {
4534 created = kvm_createvm_count;
4535 active = kvm_active_vms;
4536 mutex_unlock(&kvm_lock);
4538 env = kzalloc(sizeof(*env), GFP_KERNEL_ACCOUNT);
4542 add_uevent_var(env, "CREATED=%llu", created);
4543 add_uevent_var(env, "COUNT=%llu", active);
4545 if (type == KVM_EVENT_CREATE_VM) {
4546 add_uevent_var(env, "EVENT=create");
4547 kvm->userspace_pid = task_pid_nr(current);
4548 } else if (type == KVM_EVENT_DESTROY_VM) {
4549 add_uevent_var(env, "EVENT=destroy");
4551 add_uevent_var(env, "PID=%d", kvm->userspace_pid);
4553 if (!IS_ERR_OR_NULL(kvm->debugfs_dentry)) {
4554 char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL_ACCOUNT);
4557 tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX);
4559 add_uevent_var(env, "STATS_PATH=%s", tmp);
4563 /* no need for checks, since we are adding at most only 5 keys */
4564 env->envp[env->envp_idx++] = NULL;
4565 kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp);
4569 static void kvm_init_debug(void)
4571 struct kvm_stats_debugfs_item *p;
4573 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
4575 kvm_debugfs_num_entries = 0;
4576 for (p = debugfs_entries; p->name; ++p, kvm_debugfs_num_entries++) {
4577 debugfs_create_file(p->name, KVM_DBGFS_GET_MODE(p),
4578 kvm_debugfs_dir, (void *)(long)p->offset,
4579 stat_fops[p->kind]);
4583 static int kvm_suspend(void)
4585 if (kvm_usage_count)
4586 hardware_disable_nolock(NULL);
4590 static void kvm_resume(void)
4592 if (kvm_usage_count) {
4593 #ifdef CONFIG_LOCKDEP
4594 WARN_ON(lockdep_is_held(&kvm_count_lock));
4596 hardware_enable_nolock(NULL);
4600 static struct syscore_ops kvm_syscore_ops = {
4601 .suspend = kvm_suspend,
4602 .resume = kvm_resume,
4606 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
4608 return container_of(pn, struct kvm_vcpu, preempt_notifier);
4611 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
4613 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
4615 WRITE_ONCE(vcpu->preempted, false);
4616 WRITE_ONCE(vcpu->ready, false);
4618 __this_cpu_write(kvm_running_vcpu, vcpu);
4619 kvm_arch_sched_in(vcpu, cpu);
4620 kvm_arch_vcpu_load(vcpu, cpu);
4623 static void kvm_sched_out(struct preempt_notifier *pn,
4624 struct task_struct *next)
4626 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
4628 if (current->state == TASK_RUNNING) {
4629 WRITE_ONCE(vcpu->preempted, true);
4630 WRITE_ONCE(vcpu->ready, true);
4632 kvm_arch_vcpu_put(vcpu);
4633 __this_cpu_write(kvm_running_vcpu, NULL);
4637 * kvm_get_running_vcpu - get the vcpu running on the current CPU.
4639 * We can disable preemption locally around accessing the per-CPU variable,
4640 * and use the resolved vcpu pointer after enabling preemption again,
4641 * because even if the current thread is migrated to another CPU, reading
4642 * the per-CPU value later will give us the same value as we update the
4643 * per-CPU variable in the preempt notifier handlers.
4645 struct kvm_vcpu *kvm_get_running_vcpu(void)
4647 struct kvm_vcpu *vcpu;
4650 vcpu = __this_cpu_read(kvm_running_vcpu);
4657 * kvm_get_running_vcpus - get the per-CPU array of currently running vcpus.
4659 struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void)
4661 return &kvm_running_vcpu;
4664 struct kvm_cpu_compat_check {
4669 static void check_processor_compat(void *data)
4671 struct kvm_cpu_compat_check *c = data;
4673 *c->ret = kvm_arch_check_processor_compat(c->opaque);
4676 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
4677 struct module *module)
4679 struct kvm_cpu_compat_check c;
4683 r = kvm_arch_init(opaque);
4688 * kvm_arch_init makes sure there's at most one caller
4689 * for architectures that support multiple implementations,
4690 * like intel and amd on x86.
4691 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
4692 * conflicts in case kvm is already setup for another implementation.
4694 r = kvm_irqfd_init();
4698 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
4703 r = kvm_arch_hardware_setup(opaque);
4709 for_each_online_cpu(cpu) {
4710 smp_call_function_single(cpu, check_processor_compat, &c, 1);
4715 r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting",
4716 kvm_starting_cpu, kvm_dying_cpu);
4719 register_reboot_notifier(&kvm_reboot_notifier);
4721 /* A kmem cache lets us meet the alignment requirements of fx_save. */
4723 vcpu_align = __alignof__(struct kvm_vcpu);
4725 kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align,
4727 offsetof(struct kvm_vcpu, arch),
4728 sizeof_field(struct kvm_vcpu, arch),
4730 if (!kvm_vcpu_cache) {
4735 r = kvm_async_pf_init();
4739 kvm_chardev_ops.owner = module;
4740 kvm_vm_fops.owner = module;
4741 kvm_vcpu_fops.owner = module;
4743 r = misc_register(&kvm_dev);
4745 pr_err("kvm: misc device register failed\n");
4749 register_syscore_ops(&kvm_syscore_ops);
4751 kvm_preempt_ops.sched_in = kvm_sched_in;
4752 kvm_preempt_ops.sched_out = kvm_sched_out;
4756 r = kvm_vfio_ops_init();
4762 kvm_async_pf_deinit();
4764 kmem_cache_destroy(kvm_vcpu_cache);
4766 unregister_reboot_notifier(&kvm_reboot_notifier);
4767 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4769 kvm_arch_hardware_unsetup();
4771 free_cpumask_var(cpus_hardware_enabled);
4779 EXPORT_SYMBOL_GPL(kvm_init);
4783 debugfs_remove_recursive(kvm_debugfs_dir);
4784 misc_deregister(&kvm_dev);
4785 kmem_cache_destroy(kvm_vcpu_cache);
4786 kvm_async_pf_deinit();
4787 unregister_syscore_ops(&kvm_syscore_ops);
4788 unregister_reboot_notifier(&kvm_reboot_notifier);
4789 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4790 on_each_cpu(hardware_disable_nolock, NULL, 1);
4791 kvm_arch_hardware_unsetup();
4794 free_cpumask_var(cpus_hardware_enabled);
4795 kvm_vfio_ops_exit();
4797 EXPORT_SYMBOL_GPL(kvm_exit);
4799 struct kvm_vm_worker_thread_context {
4801 struct task_struct *parent;
4802 struct completion init_done;
4803 kvm_vm_thread_fn_t thread_fn;
4808 static int kvm_vm_worker_thread(void *context)
4811 * The init_context is allocated on the stack of the parent thread, so
4812 * we have to locally copy anything that is needed beyond initialization
4814 struct kvm_vm_worker_thread_context *init_context = context;
4815 struct kvm *kvm = init_context->kvm;
4816 kvm_vm_thread_fn_t thread_fn = init_context->thread_fn;
4817 uintptr_t data = init_context->data;
4820 err = kthread_park(current);
4821 /* kthread_park(current) is never supposed to return an error */
4826 err = cgroup_attach_task_all(init_context->parent, current);
4828 kvm_err("%s: cgroup_attach_task_all failed with err %d\n",
4833 set_user_nice(current, task_nice(init_context->parent));
4836 init_context->err = err;
4837 complete(&init_context->init_done);
4838 init_context = NULL;
4843 /* Wait to be woken up by the spawner before proceeding. */
4846 if (!kthread_should_stop())
4847 err = thread_fn(kvm, data);
4852 int kvm_vm_create_worker_thread(struct kvm *kvm, kvm_vm_thread_fn_t thread_fn,
4853 uintptr_t data, const char *name,
4854 struct task_struct **thread_ptr)
4856 struct kvm_vm_worker_thread_context init_context = {};
4857 struct task_struct *thread;
4860 init_context.kvm = kvm;
4861 init_context.parent = current;
4862 init_context.thread_fn = thread_fn;
4863 init_context.data = data;
4864 init_completion(&init_context.init_done);
4866 thread = kthread_run(kvm_vm_worker_thread, &init_context,
4867 "%s-%d", name, task_pid_nr(current));
4869 return PTR_ERR(thread);
4871 /* kthread_run is never supposed to return NULL */
4872 WARN_ON(thread == NULL);
4874 wait_for_completion(&init_context.init_done);
4876 if (!init_context.err)
4877 *thread_ptr = thread;
4879 return init_context.err;