1 // SPDX-License-Identifier: GPL-2.0-only
3 * Kernel-based Virtual Machine driver for Linux
5 * This module enables machines with Intel VT-x extensions to run virtual
6 * machines without emulation or binary translation.
8 * Copyright (C) 2006 Qumranet, Inc.
9 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
12 * Avi Kivity <avi@qumranet.com>
13 * Yaniv Kamay <yaniv@qumranet.com>
16 #include <kvm/iodev.h>
18 #include <linux/kvm_host.h>
19 #include <linux/kvm.h>
20 #include <linux/module.h>
21 #include <linux/errno.h>
22 #include <linux/percpu.h>
24 #include <linux/miscdevice.h>
25 #include <linux/vmalloc.h>
26 #include <linux/reboot.h>
27 #include <linux/debugfs.h>
28 #include <linux/highmem.h>
29 #include <linux/file.h>
30 #include <linux/syscore_ops.h>
31 #include <linux/cpu.h>
32 #include <linux/sched/signal.h>
33 #include <linux/sched/mm.h>
34 #include <linux/sched/stat.h>
35 #include <linux/cpumask.h>
36 #include <linux/smp.h>
37 #include <linux/anon_inodes.h>
38 #include <linux/profile.h>
39 #include <linux/kvm_para.h>
40 #include <linux/pagemap.h>
41 #include <linux/mman.h>
42 #include <linux/swap.h>
43 #include <linux/bitops.h>
44 #include <linux/spinlock.h>
45 #include <linux/compat.h>
46 #include <linux/srcu.h>
47 #include <linux/hugetlb.h>
48 #include <linux/slab.h>
49 #include <linux/sort.h>
50 #include <linux/bsearch.h>
52 #include <linux/lockdep.h>
53 #include <linux/kthread.h>
55 #include <asm/processor.h>
56 #include <asm/ioctl.h>
57 #include <linux/uaccess.h>
59 #include "coalesced_mmio.h"
63 #define CREATE_TRACE_POINTS
64 #include <trace/events/kvm.h>
66 /* Worst case buffer size needed for holding an integer. */
67 #define ITOA_MAX_LEN 12
69 MODULE_AUTHOR("Qumranet");
70 MODULE_LICENSE("GPL");
72 /* Architectures should define their poll value according to the halt latency */
73 unsigned int halt_poll_ns = KVM_HALT_POLL_NS_DEFAULT;
74 module_param(halt_poll_ns, uint, 0644);
75 EXPORT_SYMBOL_GPL(halt_poll_ns);
77 /* Default doubles per-vcpu halt_poll_ns. */
78 unsigned int halt_poll_ns_grow = 2;
79 module_param(halt_poll_ns_grow, uint, 0644);
80 EXPORT_SYMBOL_GPL(halt_poll_ns_grow);
82 /* The start value to grow halt_poll_ns from */
83 unsigned int halt_poll_ns_grow_start = 10000; /* 10us */
84 module_param(halt_poll_ns_grow_start, uint, 0644);
85 EXPORT_SYMBOL_GPL(halt_poll_ns_grow_start);
87 /* Default resets per-vcpu halt_poll_ns . */
88 unsigned int halt_poll_ns_shrink;
89 module_param(halt_poll_ns_shrink, uint, 0644);
90 EXPORT_SYMBOL_GPL(halt_poll_ns_shrink);
95 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
98 DEFINE_MUTEX(kvm_lock);
99 static DEFINE_RAW_SPINLOCK(kvm_count_lock);
102 static cpumask_var_t cpus_hardware_enabled;
103 static int kvm_usage_count;
104 static atomic_t hardware_enable_failed;
106 static struct kmem_cache *kvm_vcpu_cache;
108 static __read_mostly struct preempt_ops kvm_preempt_ops;
109 static DEFINE_PER_CPU(struct kvm_vcpu *, kvm_running_vcpu);
111 struct dentry *kvm_debugfs_dir;
112 EXPORT_SYMBOL_GPL(kvm_debugfs_dir);
114 static int kvm_debugfs_num_entries;
115 static const struct file_operations stat_fops_per_vm;
117 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
119 #ifdef CONFIG_KVM_COMPAT
120 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
122 #define KVM_COMPAT(c) .compat_ioctl = (c)
125 * For architectures that don't implement a compat infrastructure,
126 * adopt a double line of defense:
127 * - Prevent a compat task from opening /dev/kvm
128 * - If the open has been done by a 64bit task, and the KVM fd
129 * passed to a compat task, let the ioctls fail.
131 static long kvm_no_compat_ioctl(struct file *file, unsigned int ioctl,
132 unsigned long arg) { return -EINVAL; }
134 static int kvm_no_compat_open(struct inode *inode, struct file *file)
136 return is_compat_task() ? -ENODEV : 0;
138 #define KVM_COMPAT(c) .compat_ioctl = kvm_no_compat_ioctl, \
139 .open = kvm_no_compat_open
141 static int hardware_enable_all(void);
142 static void hardware_disable_all(void);
144 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
146 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot, gfn_t gfn);
148 __visible bool kvm_rebooting;
149 EXPORT_SYMBOL_GPL(kvm_rebooting);
151 #define KVM_EVENT_CREATE_VM 0
152 #define KVM_EVENT_DESTROY_VM 1
153 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm);
154 static unsigned long long kvm_createvm_count;
155 static unsigned long long kvm_active_vms;
157 __weak void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
158 unsigned long start, unsigned long end)
162 bool kvm_is_zone_device_pfn(kvm_pfn_t pfn)
165 * The metadata used by is_zone_device_page() to determine whether or
166 * not a page is ZONE_DEVICE is guaranteed to be valid if and only if
167 * the device has been pinned, e.g. by get_user_pages(). WARN if the
168 * page_count() is zero to help detect bad usage of this helper.
170 if (!pfn_valid(pfn) || WARN_ON_ONCE(!page_count(pfn_to_page(pfn))))
173 return is_zone_device_page(pfn_to_page(pfn));
176 bool kvm_is_reserved_pfn(kvm_pfn_t pfn)
179 * ZONE_DEVICE pages currently set PG_reserved, but from a refcounting
180 * perspective they are "normal" pages, albeit with slightly different
184 return PageReserved(pfn_to_page(pfn)) &&
186 !kvm_is_zone_device_pfn(pfn);
191 bool kvm_is_transparent_hugepage(kvm_pfn_t pfn)
193 struct page *page = pfn_to_page(pfn);
195 if (!PageTransCompoundMap(page))
198 return is_transparent_hugepage(compound_head(page));
202 * Switches to specified vcpu, until a matching vcpu_put()
204 void vcpu_load(struct kvm_vcpu *vcpu)
208 __this_cpu_write(kvm_running_vcpu, vcpu);
209 preempt_notifier_register(&vcpu->preempt_notifier);
210 kvm_arch_vcpu_load(vcpu, cpu);
213 EXPORT_SYMBOL_GPL(vcpu_load);
215 void vcpu_put(struct kvm_vcpu *vcpu)
218 kvm_arch_vcpu_put(vcpu);
219 preempt_notifier_unregister(&vcpu->preempt_notifier);
220 __this_cpu_write(kvm_running_vcpu, NULL);
223 EXPORT_SYMBOL_GPL(vcpu_put);
225 /* TODO: merge with kvm_arch_vcpu_should_kick */
226 static bool kvm_request_needs_ipi(struct kvm_vcpu *vcpu, unsigned req)
228 int mode = kvm_vcpu_exiting_guest_mode(vcpu);
231 * We need to wait for the VCPU to reenable interrupts and get out of
232 * READING_SHADOW_PAGE_TABLES mode.
234 if (req & KVM_REQUEST_WAIT)
235 return mode != OUTSIDE_GUEST_MODE;
238 * Need to kick a running VCPU, but otherwise there is nothing to do.
240 return mode == IN_GUEST_MODE;
243 static void ack_flush(void *_completed)
247 static inline bool kvm_kick_many_cpus(const struct cpumask *cpus, bool wait)
250 cpus = cpu_online_mask;
252 if (cpumask_empty(cpus))
255 smp_call_function_many(cpus, ack_flush, NULL, wait);
259 bool kvm_make_vcpus_request_mask(struct kvm *kvm, unsigned int req,
260 struct kvm_vcpu *except,
261 unsigned long *vcpu_bitmap, cpumask_var_t tmp)
264 struct kvm_vcpu *vcpu;
269 kvm_for_each_vcpu(i, vcpu, kvm) {
270 if ((vcpu_bitmap && !test_bit(i, vcpu_bitmap)) ||
274 kvm_make_request(req, vcpu);
277 if (!(req & KVM_REQUEST_NO_WAKEUP) && kvm_vcpu_wake_up(vcpu))
280 if (tmp != NULL && cpu != -1 && cpu != me &&
281 kvm_request_needs_ipi(vcpu, req))
282 __cpumask_set_cpu(cpu, tmp);
285 called = kvm_kick_many_cpus(tmp, !!(req & KVM_REQUEST_WAIT));
291 bool kvm_make_all_cpus_request_except(struct kvm *kvm, unsigned int req,
292 struct kvm_vcpu *except)
297 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
299 called = kvm_make_vcpus_request_mask(kvm, req, except, NULL, cpus);
301 free_cpumask_var(cpus);
305 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
307 return kvm_make_all_cpus_request_except(kvm, req, NULL);
310 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
311 void kvm_flush_remote_tlbs(struct kvm *kvm)
314 * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in
315 * kvm_make_all_cpus_request.
317 long dirty_count = smp_load_acquire(&kvm->tlbs_dirty);
320 * We want to publish modifications to the page tables before reading
321 * mode. Pairs with a memory barrier in arch-specific code.
322 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
323 * and smp_mb in walk_shadow_page_lockless_begin/end.
324 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
326 * There is already an smp_mb__after_atomic() before
327 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
330 if (!kvm_arch_flush_remote_tlb(kvm)
331 || kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
332 ++kvm->stat.remote_tlb_flush;
333 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
335 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
338 void kvm_reload_remote_mmus(struct kvm *kvm)
340 kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
343 #ifdef KVM_ARCH_NR_OBJS_PER_MEMORY_CACHE
344 static inline void *mmu_memory_cache_alloc_obj(struct kvm_mmu_memory_cache *mc,
347 gfp_flags |= mc->gfp_zero;
350 return kmem_cache_alloc(mc->kmem_cache, gfp_flags);
352 return (void *)__get_free_page(gfp_flags);
355 int kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache *mc, int min)
359 if (mc->nobjs >= min)
361 while (mc->nobjs < ARRAY_SIZE(mc->objects)) {
362 obj = mmu_memory_cache_alloc_obj(mc, GFP_KERNEL_ACCOUNT);
364 return mc->nobjs >= min ? 0 : -ENOMEM;
365 mc->objects[mc->nobjs++] = obj;
370 int kvm_mmu_memory_cache_nr_free_objects(struct kvm_mmu_memory_cache *mc)
375 void kvm_mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc)
379 kmem_cache_free(mc->kmem_cache, mc->objects[--mc->nobjs]);
381 free_page((unsigned long)mc->objects[--mc->nobjs]);
385 void *kvm_mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc)
389 if (WARN_ON(!mc->nobjs))
390 p = mmu_memory_cache_alloc_obj(mc, GFP_ATOMIC | __GFP_ACCOUNT);
392 p = mc->objects[--mc->nobjs];
398 static void kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
400 mutex_init(&vcpu->mutex);
405 rcuwait_init(&vcpu->wait);
406 kvm_async_pf_vcpu_init(vcpu);
409 INIT_LIST_HEAD(&vcpu->blocked_vcpu_list);
411 kvm_vcpu_set_in_spin_loop(vcpu, false);
412 kvm_vcpu_set_dy_eligible(vcpu, false);
413 vcpu->preempted = false;
415 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
418 void kvm_vcpu_destroy(struct kvm_vcpu *vcpu)
420 kvm_arch_vcpu_destroy(vcpu);
423 * No need for rcu_read_lock as VCPU_RUN is the only place that changes
424 * the vcpu->pid pointer, and at destruction time all file descriptors
427 put_pid(rcu_dereference_protected(vcpu->pid, 1));
429 free_page((unsigned long)vcpu->run);
430 kmem_cache_free(kvm_vcpu_cache, vcpu);
432 EXPORT_SYMBOL_GPL(kvm_vcpu_destroy);
434 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
435 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
437 return container_of(mn, struct kvm, mmu_notifier);
440 static void kvm_mmu_notifier_invalidate_range(struct mmu_notifier *mn,
441 struct mm_struct *mm,
442 unsigned long start, unsigned long end)
444 struct kvm *kvm = mmu_notifier_to_kvm(mn);
447 idx = srcu_read_lock(&kvm->srcu);
448 kvm_arch_mmu_notifier_invalidate_range(kvm, start, end);
449 srcu_read_unlock(&kvm->srcu, idx);
452 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
453 struct mm_struct *mm,
454 unsigned long address,
457 struct kvm *kvm = mmu_notifier_to_kvm(mn);
460 idx = srcu_read_lock(&kvm->srcu);
461 spin_lock(&kvm->mmu_lock);
462 kvm->mmu_notifier_seq++;
464 if (kvm_set_spte_hva(kvm, address, pte))
465 kvm_flush_remote_tlbs(kvm);
467 spin_unlock(&kvm->mmu_lock);
468 srcu_read_unlock(&kvm->srcu, idx);
471 static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
472 const struct mmu_notifier_range *range)
474 struct kvm *kvm = mmu_notifier_to_kvm(mn);
475 int need_tlb_flush = 0, idx;
477 idx = srcu_read_lock(&kvm->srcu);
478 spin_lock(&kvm->mmu_lock);
480 * The count increase must become visible at unlock time as no
481 * spte can be established without taking the mmu_lock and
482 * count is also read inside the mmu_lock critical section.
484 kvm->mmu_notifier_count++;
485 need_tlb_flush = kvm_unmap_hva_range(kvm, range->start, range->end,
487 need_tlb_flush |= kvm->tlbs_dirty;
488 /* we've to flush the tlb before the pages can be freed */
490 kvm_flush_remote_tlbs(kvm);
492 spin_unlock(&kvm->mmu_lock);
493 srcu_read_unlock(&kvm->srcu, idx);
498 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
499 const struct mmu_notifier_range *range)
501 struct kvm *kvm = mmu_notifier_to_kvm(mn);
503 spin_lock(&kvm->mmu_lock);
505 * This sequence increase will notify the kvm page fault that
506 * the page that is going to be mapped in the spte could have
509 kvm->mmu_notifier_seq++;
512 * The above sequence increase must be visible before the
513 * below count decrease, which is ensured by the smp_wmb above
514 * in conjunction with the smp_rmb in mmu_notifier_retry().
516 kvm->mmu_notifier_count--;
517 spin_unlock(&kvm->mmu_lock);
519 BUG_ON(kvm->mmu_notifier_count < 0);
522 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
523 struct mm_struct *mm,
527 struct kvm *kvm = mmu_notifier_to_kvm(mn);
530 idx = srcu_read_lock(&kvm->srcu);
531 spin_lock(&kvm->mmu_lock);
533 young = kvm_age_hva(kvm, start, end);
535 kvm_flush_remote_tlbs(kvm);
537 spin_unlock(&kvm->mmu_lock);
538 srcu_read_unlock(&kvm->srcu, idx);
543 static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
544 struct mm_struct *mm,
548 struct kvm *kvm = mmu_notifier_to_kvm(mn);
551 idx = srcu_read_lock(&kvm->srcu);
552 spin_lock(&kvm->mmu_lock);
554 * Even though we do not flush TLB, this will still adversely
555 * affect performance on pre-Haswell Intel EPT, where there is
556 * no EPT Access Bit to clear so that we have to tear down EPT
557 * tables instead. If we find this unacceptable, we can always
558 * add a parameter to kvm_age_hva so that it effectively doesn't
559 * do anything on clear_young.
561 * Also note that currently we never issue secondary TLB flushes
562 * from clear_young, leaving this job up to the regular system
563 * cadence. If we find this inaccurate, we might come up with a
564 * more sophisticated heuristic later.
566 young = kvm_age_hva(kvm, start, end);
567 spin_unlock(&kvm->mmu_lock);
568 srcu_read_unlock(&kvm->srcu, idx);
573 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
574 struct mm_struct *mm,
575 unsigned long address)
577 struct kvm *kvm = mmu_notifier_to_kvm(mn);
580 idx = srcu_read_lock(&kvm->srcu);
581 spin_lock(&kvm->mmu_lock);
582 young = kvm_test_age_hva(kvm, address);
583 spin_unlock(&kvm->mmu_lock);
584 srcu_read_unlock(&kvm->srcu, idx);
589 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
590 struct mm_struct *mm)
592 struct kvm *kvm = mmu_notifier_to_kvm(mn);
595 idx = srcu_read_lock(&kvm->srcu);
596 kvm_arch_flush_shadow_all(kvm);
597 srcu_read_unlock(&kvm->srcu, idx);
600 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
601 .invalidate_range = kvm_mmu_notifier_invalidate_range,
602 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
603 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
604 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
605 .clear_young = kvm_mmu_notifier_clear_young,
606 .test_young = kvm_mmu_notifier_test_young,
607 .change_pte = kvm_mmu_notifier_change_pte,
608 .release = kvm_mmu_notifier_release,
611 static int kvm_init_mmu_notifier(struct kvm *kvm)
613 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
614 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
617 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
619 static int kvm_init_mmu_notifier(struct kvm *kvm)
624 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
626 static struct kvm_memslots *kvm_alloc_memslots(void)
629 struct kvm_memslots *slots;
631 slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL_ACCOUNT);
635 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
636 slots->id_to_index[i] = -1;
641 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
643 if (!memslot->dirty_bitmap)
646 kvfree(memslot->dirty_bitmap);
647 memslot->dirty_bitmap = NULL;
650 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
652 kvm_destroy_dirty_bitmap(slot);
654 kvm_arch_free_memslot(kvm, slot);
660 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
662 struct kvm_memory_slot *memslot;
667 kvm_for_each_memslot(memslot, slots)
668 kvm_free_memslot(kvm, memslot);
673 static void kvm_destroy_vm_debugfs(struct kvm *kvm)
677 if (!kvm->debugfs_dentry)
680 debugfs_remove_recursive(kvm->debugfs_dentry);
682 if (kvm->debugfs_stat_data) {
683 for (i = 0; i < kvm_debugfs_num_entries; i++)
684 kfree(kvm->debugfs_stat_data[i]);
685 kfree(kvm->debugfs_stat_data);
689 static int kvm_create_vm_debugfs(struct kvm *kvm, int fd)
691 char dir_name[ITOA_MAX_LEN * 2];
692 struct kvm_stat_data *stat_data;
693 struct kvm_stats_debugfs_item *p;
695 if (!debugfs_initialized())
698 snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd);
699 kvm->debugfs_dentry = debugfs_create_dir(dir_name, kvm_debugfs_dir);
701 kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries,
702 sizeof(*kvm->debugfs_stat_data),
704 if (!kvm->debugfs_stat_data)
707 for (p = debugfs_entries; p->name; p++) {
708 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT);
712 stat_data->kvm = kvm;
713 stat_data->dbgfs_item = p;
714 kvm->debugfs_stat_data[p - debugfs_entries] = stat_data;
715 debugfs_create_file(p->name, KVM_DBGFS_GET_MODE(p),
716 kvm->debugfs_dentry, stat_data,
723 * Called after the VM is otherwise initialized, but just before adding it to
726 int __weak kvm_arch_post_init_vm(struct kvm *kvm)
732 * Called just after removing the VM from the vm_list, but before doing any
735 void __weak kvm_arch_pre_destroy_vm(struct kvm *kvm)
739 static struct kvm *kvm_create_vm(unsigned long type)
741 struct kvm *kvm = kvm_arch_alloc_vm();
746 return ERR_PTR(-ENOMEM);
748 spin_lock_init(&kvm->mmu_lock);
750 kvm->mm = current->mm;
751 kvm_eventfd_init(kvm);
752 mutex_init(&kvm->lock);
753 mutex_init(&kvm->irq_lock);
754 mutex_init(&kvm->slots_lock);
755 INIT_LIST_HEAD(&kvm->devices);
757 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
759 if (init_srcu_struct(&kvm->srcu))
760 goto out_err_no_srcu;
761 if (init_srcu_struct(&kvm->irq_srcu))
762 goto out_err_no_irq_srcu;
764 refcount_set(&kvm->users_count, 1);
765 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
766 struct kvm_memslots *slots = kvm_alloc_memslots();
769 goto out_err_no_arch_destroy_vm;
770 /* Generations must be different for each address space. */
771 slots->generation = i;
772 rcu_assign_pointer(kvm->memslots[i], slots);
775 for (i = 0; i < KVM_NR_BUSES; i++) {
776 rcu_assign_pointer(kvm->buses[i],
777 kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL_ACCOUNT));
779 goto out_err_no_arch_destroy_vm;
782 kvm->max_halt_poll_ns = halt_poll_ns;
784 r = kvm_arch_init_vm(kvm, type);
786 goto out_err_no_arch_destroy_vm;
788 r = hardware_enable_all();
790 goto out_err_no_disable;
792 #ifdef CONFIG_HAVE_KVM_IRQFD
793 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
796 r = kvm_init_mmu_notifier(kvm);
798 goto out_err_no_mmu_notifier;
800 r = kvm_arch_post_init_vm(kvm);
804 mutex_lock(&kvm_lock);
805 list_add(&kvm->vm_list, &vm_list);
806 mutex_unlock(&kvm_lock);
808 preempt_notifier_inc();
813 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
814 if (kvm->mmu_notifier.ops)
815 mmu_notifier_unregister(&kvm->mmu_notifier, current->mm);
817 out_err_no_mmu_notifier:
818 hardware_disable_all();
820 kvm_arch_destroy_vm(kvm);
821 out_err_no_arch_destroy_vm:
822 WARN_ON_ONCE(!refcount_dec_and_test(&kvm->users_count));
823 for (i = 0; i < KVM_NR_BUSES; i++)
824 kfree(kvm_get_bus(kvm, i));
825 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
826 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
827 cleanup_srcu_struct(&kvm->irq_srcu);
829 cleanup_srcu_struct(&kvm->srcu);
831 kvm_arch_free_vm(kvm);
836 static void kvm_destroy_devices(struct kvm *kvm)
838 struct kvm_device *dev, *tmp;
841 * We do not need to take the kvm->lock here, because nobody else
842 * has a reference to the struct kvm at this point and therefore
843 * cannot access the devices list anyhow.
845 list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
846 list_del(&dev->vm_node);
847 dev->ops->destroy(dev);
851 static void kvm_destroy_vm(struct kvm *kvm)
854 struct mm_struct *mm = kvm->mm;
856 kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm);
857 kvm_destroy_vm_debugfs(kvm);
858 kvm_arch_sync_events(kvm);
859 mutex_lock(&kvm_lock);
860 list_del(&kvm->vm_list);
861 mutex_unlock(&kvm_lock);
862 kvm_arch_pre_destroy_vm(kvm);
864 kvm_free_irq_routing(kvm);
865 for (i = 0; i < KVM_NR_BUSES; i++) {
866 struct kvm_io_bus *bus = kvm_get_bus(kvm, i);
869 kvm_io_bus_destroy(bus);
870 kvm->buses[i] = NULL;
872 kvm_coalesced_mmio_free(kvm);
873 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
874 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
876 kvm_arch_flush_shadow_all(kvm);
878 kvm_arch_destroy_vm(kvm);
879 kvm_destroy_devices(kvm);
880 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
881 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
882 cleanup_srcu_struct(&kvm->irq_srcu);
883 cleanup_srcu_struct(&kvm->srcu);
884 kvm_arch_free_vm(kvm);
885 preempt_notifier_dec();
886 hardware_disable_all();
890 void kvm_get_kvm(struct kvm *kvm)
892 refcount_inc(&kvm->users_count);
894 EXPORT_SYMBOL_GPL(kvm_get_kvm);
896 void kvm_put_kvm(struct kvm *kvm)
898 if (refcount_dec_and_test(&kvm->users_count))
901 EXPORT_SYMBOL_GPL(kvm_put_kvm);
904 * Used to put a reference that was taken on behalf of an object associated
905 * with a user-visible file descriptor, e.g. a vcpu or device, if installation
906 * of the new file descriptor fails and the reference cannot be transferred to
907 * its final owner. In such cases, the caller is still actively using @kvm and
908 * will fail miserably if the refcount unexpectedly hits zero.
910 void kvm_put_kvm_no_destroy(struct kvm *kvm)
912 WARN_ON(refcount_dec_and_test(&kvm->users_count));
914 EXPORT_SYMBOL_GPL(kvm_put_kvm_no_destroy);
916 static int kvm_vm_release(struct inode *inode, struct file *filp)
918 struct kvm *kvm = filp->private_data;
920 kvm_irqfd_release(kvm);
927 * Allocation size is twice as large as the actual dirty bitmap size.
928 * See kvm_vm_ioctl_get_dirty_log() why this is needed.
930 static int kvm_alloc_dirty_bitmap(struct kvm_memory_slot *memslot)
932 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
934 memslot->dirty_bitmap = kvzalloc(dirty_bytes, GFP_KERNEL_ACCOUNT);
935 if (!memslot->dirty_bitmap)
942 * Delete a memslot by decrementing the number of used slots and shifting all
943 * other entries in the array forward one spot.
945 static inline void kvm_memslot_delete(struct kvm_memslots *slots,
946 struct kvm_memory_slot *memslot)
948 struct kvm_memory_slot *mslots = slots->memslots;
951 if (WARN_ON(slots->id_to_index[memslot->id] == -1))
956 if (atomic_read(&slots->lru_slot) >= slots->used_slots)
957 atomic_set(&slots->lru_slot, 0);
959 for (i = slots->id_to_index[memslot->id]; i < slots->used_slots; i++) {
960 mslots[i] = mslots[i + 1];
961 slots->id_to_index[mslots[i].id] = i;
963 mslots[i] = *memslot;
964 slots->id_to_index[memslot->id] = -1;
968 * "Insert" a new memslot by incrementing the number of used slots. Returns
969 * the new slot's initial index into the memslots array.
971 static inline int kvm_memslot_insert_back(struct kvm_memslots *slots)
973 return slots->used_slots++;
977 * Move a changed memslot backwards in the array by shifting existing slots
978 * with a higher GFN toward the front of the array. Note, the changed memslot
979 * itself is not preserved in the array, i.e. not swapped at this time, only
980 * its new index into the array is tracked. Returns the changed memslot's
981 * current index into the memslots array.
983 static inline int kvm_memslot_move_backward(struct kvm_memslots *slots,
984 struct kvm_memory_slot *memslot)
986 struct kvm_memory_slot *mslots = slots->memslots;
989 if (WARN_ON_ONCE(slots->id_to_index[memslot->id] == -1) ||
990 WARN_ON_ONCE(!slots->used_slots))
994 * Move the target memslot backward in the array by shifting existing
995 * memslots with a higher GFN (than the target memslot) towards the
996 * front of the array.
998 for (i = slots->id_to_index[memslot->id]; i < slots->used_slots - 1; i++) {
999 if (memslot->base_gfn > mslots[i + 1].base_gfn)
1002 WARN_ON_ONCE(memslot->base_gfn == mslots[i + 1].base_gfn);
1004 /* Shift the next memslot forward one and update its index. */
1005 mslots[i] = mslots[i + 1];
1006 slots->id_to_index[mslots[i].id] = i;
1012 * Move a changed memslot forwards in the array by shifting existing slots with
1013 * a lower GFN toward the back of the array. Note, the changed memslot itself
1014 * is not preserved in the array, i.e. not swapped at this time, only its new
1015 * index into the array is tracked. Returns the changed memslot's final index
1016 * into the memslots array.
1018 static inline int kvm_memslot_move_forward(struct kvm_memslots *slots,
1019 struct kvm_memory_slot *memslot,
1022 struct kvm_memory_slot *mslots = slots->memslots;
1025 for (i = start; i > 0; i--) {
1026 if (memslot->base_gfn < mslots[i - 1].base_gfn)
1029 WARN_ON_ONCE(memslot->base_gfn == mslots[i - 1].base_gfn);
1031 /* Shift the next memslot back one and update its index. */
1032 mslots[i] = mslots[i - 1];
1033 slots->id_to_index[mslots[i].id] = i;
1039 * Re-sort memslots based on their GFN to account for an added, deleted, or
1040 * moved memslot. Sorting memslots by GFN allows using a binary search during
1043 * IMPORTANT: Slots are sorted from highest GFN to lowest GFN! I.e. the entry
1044 * at memslots[0] has the highest GFN.
1046 * The sorting algorithm takes advantage of having initially sorted memslots
1047 * and knowing the position of the changed memslot. Sorting is also optimized
1048 * by not swapping the updated memslot and instead only shifting other memslots
1049 * and tracking the new index for the update memslot. Only once its final
1050 * index is known is the updated memslot copied into its position in the array.
1052 * - When deleting a memslot, the deleted memslot simply needs to be moved to
1053 * the end of the array.
1055 * - When creating a memslot, the algorithm "inserts" the new memslot at the
1056 * end of the array and then it forward to its correct location.
1058 * - When moving a memslot, the algorithm first moves the updated memslot
1059 * backward to handle the scenario where the memslot's GFN was changed to a
1060 * lower value. update_memslots() then falls through and runs the same flow
1061 * as creating a memslot to move the memslot forward to handle the scenario
1062 * where its GFN was changed to a higher value.
1064 * Note, slots are sorted from highest->lowest instead of lowest->highest for
1065 * historical reasons. Originally, invalid memslots where denoted by having
1066 * GFN=0, thus sorting from highest->lowest naturally sorted invalid memslots
1067 * to the end of the array. The current algorithm uses dedicated logic to
1068 * delete a memslot and thus does not rely on invalid memslots having GFN=0.
1070 * The other historical motiviation for highest->lowest was to improve the
1071 * performance of memslot lookup. KVM originally used a linear search starting
1072 * at memslots[0]. On x86, the largest memslot usually has one of the highest,
1073 * if not *the* highest, GFN, as the bulk of the guest's RAM is located in a
1074 * single memslot above the 4gb boundary. As the largest memslot is also the
1075 * most likely to be referenced, sorting it to the front of the array was
1076 * advantageous. The current binary search starts from the middle of the array
1077 * and uses an LRU pointer to improve performance for all memslots and GFNs.
1079 static void update_memslots(struct kvm_memslots *slots,
1080 struct kvm_memory_slot *memslot,
1081 enum kvm_mr_change change)
1085 if (change == KVM_MR_DELETE) {
1086 kvm_memslot_delete(slots, memslot);
1088 if (change == KVM_MR_CREATE)
1089 i = kvm_memslot_insert_back(slots);
1091 i = kvm_memslot_move_backward(slots, memslot);
1092 i = kvm_memslot_move_forward(slots, memslot, i);
1095 * Copy the memslot to its new position in memslots and update
1096 * its index accordingly.
1098 slots->memslots[i] = *memslot;
1099 slots->id_to_index[memslot->id] = i;
1103 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
1105 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
1107 #ifdef __KVM_HAVE_READONLY_MEM
1108 valid_flags |= KVM_MEM_READONLY;
1111 if (mem->flags & ~valid_flags)
1117 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
1118 int as_id, struct kvm_memslots *slots)
1120 struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id);
1121 u64 gen = old_memslots->generation;
1123 WARN_ON(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS);
1124 slots->generation = gen | KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
1126 rcu_assign_pointer(kvm->memslots[as_id], slots);
1127 synchronize_srcu_expedited(&kvm->srcu);
1130 * Increment the new memslot generation a second time, dropping the
1131 * update in-progress flag and incrementing the generation based on
1132 * the number of address spaces. This provides a unique and easily
1133 * identifiable generation number while the memslots are in flux.
1135 gen = slots->generation & ~KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
1138 * Generations must be unique even across address spaces. We do not need
1139 * a global counter for that, instead the generation space is evenly split
1140 * across address spaces. For example, with two address spaces, address
1141 * space 0 will use generations 0, 2, 4, ... while address space 1 will
1142 * use generations 1, 3, 5, ...
1144 gen += KVM_ADDRESS_SPACE_NUM;
1146 kvm_arch_memslots_updated(kvm, gen);
1148 slots->generation = gen;
1150 return old_memslots;
1154 * Note, at a minimum, the current number of used slots must be allocated, even
1155 * when deleting a memslot, as we need a complete duplicate of the memslots for
1156 * use when invalidating a memslot prior to deleting/moving the memslot.
1158 static struct kvm_memslots *kvm_dup_memslots(struct kvm_memslots *old,
1159 enum kvm_mr_change change)
1161 struct kvm_memslots *slots;
1162 size_t old_size, new_size;
1164 old_size = sizeof(struct kvm_memslots) +
1165 (sizeof(struct kvm_memory_slot) * old->used_slots);
1167 if (change == KVM_MR_CREATE)
1168 new_size = old_size + sizeof(struct kvm_memory_slot);
1170 new_size = old_size;
1172 slots = kvzalloc(new_size, GFP_KERNEL_ACCOUNT);
1174 memcpy(slots, old, old_size);
1179 static int kvm_set_memslot(struct kvm *kvm,
1180 const struct kvm_userspace_memory_region *mem,
1181 struct kvm_memory_slot *old,
1182 struct kvm_memory_slot *new, int as_id,
1183 enum kvm_mr_change change)
1185 struct kvm_memory_slot *slot;
1186 struct kvm_memslots *slots;
1189 slots = kvm_dup_memslots(__kvm_memslots(kvm, as_id), change);
1193 if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) {
1195 * Note, the INVALID flag needs to be in the appropriate entry
1196 * in the freshly allocated memslots, not in @old or @new.
1198 slot = id_to_memslot(slots, old->id);
1199 slot->flags |= KVM_MEMSLOT_INVALID;
1202 * We can re-use the old memslots, the only difference from the
1203 * newly installed memslots is the invalid flag, which will get
1204 * dropped by update_memslots anyway. We'll also revert to the
1205 * old memslots if preparing the new memory region fails.
1207 slots = install_new_memslots(kvm, as_id, slots);
1209 /* From this point no new shadow pages pointing to a deleted,
1210 * or moved, memslot will be created.
1212 * validation of sp->gfn happens in:
1213 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1214 * - kvm_is_visible_gfn (mmu_check_root)
1216 kvm_arch_flush_shadow_memslot(kvm, slot);
1219 r = kvm_arch_prepare_memory_region(kvm, new, mem, change);
1223 update_memslots(slots, new, change);
1224 slots = install_new_memslots(kvm, as_id, slots);
1226 kvm_arch_commit_memory_region(kvm, mem, old, new, change);
1232 if (change == KVM_MR_DELETE || change == KVM_MR_MOVE)
1233 slots = install_new_memslots(kvm, as_id, slots);
1238 static int kvm_delete_memslot(struct kvm *kvm,
1239 const struct kvm_userspace_memory_region *mem,
1240 struct kvm_memory_slot *old, int as_id)
1242 struct kvm_memory_slot new;
1248 memset(&new, 0, sizeof(new));
1251 r = kvm_set_memslot(kvm, mem, old, &new, as_id, KVM_MR_DELETE);
1255 kvm_free_memslot(kvm, old);
1260 * Allocate some memory and give it an address in the guest physical address
1263 * Discontiguous memory is allowed, mostly for framebuffers.
1265 * Must be called holding kvm->slots_lock for write.
1267 int __kvm_set_memory_region(struct kvm *kvm,
1268 const struct kvm_userspace_memory_region *mem)
1270 struct kvm_memory_slot old, new;
1271 struct kvm_memory_slot *tmp;
1272 enum kvm_mr_change change;
1276 r = check_memory_region_flags(mem);
1280 as_id = mem->slot >> 16;
1281 id = (u16)mem->slot;
1283 /* General sanity checks */
1284 if (mem->memory_size & (PAGE_SIZE - 1))
1286 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
1288 /* We can read the guest memory with __xxx_user() later on. */
1289 if ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
1290 !access_ok((void __user *)(unsigned long)mem->userspace_addr,
1293 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
1295 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
1299 * Make a full copy of the old memslot, the pointer will become stale
1300 * when the memslots are re-sorted by update_memslots(), and the old
1301 * memslot needs to be referenced after calling update_memslots(), e.g.
1302 * to free its resources and for arch specific behavior.
1304 tmp = id_to_memslot(__kvm_memslots(kvm, as_id), id);
1309 memset(&old, 0, sizeof(old));
1313 if (!mem->memory_size)
1314 return kvm_delete_memslot(kvm, mem, &old, as_id);
1317 new.base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
1318 new.npages = mem->memory_size >> PAGE_SHIFT;
1319 new.flags = mem->flags;
1320 new.userspace_addr = mem->userspace_addr;
1322 if (new.npages > KVM_MEM_MAX_NR_PAGES)
1326 change = KVM_MR_CREATE;
1327 new.dirty_bitmap = NULL;
1328 memset(&new.arch, 0, sizeof(new.arch));
1329 } else { /* Modify an existing slot. */
1330 if ((new.userspace_addr != old.userspace_addr) ||
1331 (new.npages != old.npages) ||
1332 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
1335 if (new.base_gfn != old.base_gfn)
1336 change = KVM_MR_MOVE;
1337 else if (new.flags != old.flags)
1338 change = KVM_MR_FLAGS_ONLY;
1339 else /* Nothing to change. */
1342 /* Copy dirty_bitmap and arch from the current memslot. */
1343 new.dirty_bitmap = old.dirty_bitmap;
1344 memcpy(&new.arch, &old.arch, sizeof(new.arch));
1347 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
1348 /* Check for overlaps */
1349 kvm_for_each_memslot(tmp, __kvm_memslots(kvm, as_id)) {
1352 if (!((new.base_gfn + new.npages <= tmp->base_gfn) ||
1353 (new.base_gfn >= tmp->base_gfn + tmp->npages)))
1358 /* Allocate/free page dirty bitmap as needed */
1359 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
1360 new.dirty_bitmap = NULL;
1361 else if (!new.dirty_bitmap) {
1362 r = kvm_alloc_dirty_bitmap(&new);
1366 if (kvm_dirty_log_manual_protect_and_init_set(kvm))
1367 bitmap_set(new.dirty_bitmap, 0, new.npages);
1370 r = kvm_set_memslot(kvm, mem, &old, &new, as_id, change);
1374 if (old.dirty_bitmap && !new.dirty_bitmap)
1375 kvm_destroy_dirty_bitmap(&old);
1379 if (new.dirty_bitmap && !old.dirty_bitmap)
1380 kvm_destroy_dirty_bitmap(&new);
1383 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1385 int kvm_set_memory_region(struct kvm *kvm,
1386 const struct kvm_userspace_memory_region *mem)
1390 mutex_lock(&kvm->slots_lock);
1391 r = __kvm_set_memory_region(kvm, mem);
1392 mutex_unlock(&kvm->slots_lock);
1395 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1397 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1398 struct kvm_userspace_memory_region *mem)
1400 if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1403 return kvm_set_memory_region(kvm, mem);
1406 #ifndef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1408 * kvm_get_dirty_log - get a snapshot of dirty pages
1409 * @kvm: pointer to kvm instance
1410 * @log: slot id and address to which we copy the log
1411 * @is_dirty: set to '1' if any dirty pages were found
1412 * @memslot: set to the associated memslot, always valid on success
1414 int kvm_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log,
1415 int *is_dirty, struct kvm_memory_slot **memslot)
1417 struct kvm_memslots *slots;
1420 unsigned long any = 0;
1425 as_id = log->slot >> 16;
1426 id = (u16)log->slot;
1427 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1430 slots = __kvm_memslots(kvm, as_id);
1431 *memslot = id_to_memslot(slots, id);
1432 if (!(*memslot) || !(*memslot)->dirty_bitmap)
1435 kvm_arch_sync_dirty_log(kvm, *memslot);
1437 n = kvm_dirty_bitmap_bytes(*memslot);
1439 for (i = 0; !any && i < n/sizeof(long); ++i)
1440 any = (*memslot)->dirty_bitmap[i];
1442 if (copy_to_user(log->dirty_bitmap, (*memslot)->dirty_bitmap, n))
1449 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1451 #else /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
1453 * kvm_get_dirty_log_protect - get a snapshot of dirty pages
1454 * and reenable dirty page tracking for the corresponding pages.
1455 * @kvm: pointer to kvm instance
1456 * @log: slot id and address to which we copy the log
1458 * We need to keep it in mind that VCPU threads can write to the bitmap
1459 * concurrently. So, to avoid losing track of dirty pages we keep the
1462 * 1. Take a snapshot of the bit and clear it if needed.
1463 * 2. Write protect the corresponding page.
1464 * 3. Copy the snapshot to the userspace.
1465 * 4. Upon return caller flushes TLB's if needed.
1467 * Between 2 and 4, the guest may write to the page using the remaining TLB
1468 * entry. This is not a problem because the page is reported dirty using
1469 * the snapshot taken before and step 4 ensures that writes done after
1470 * exiting to userspace will be logged for the next call.
1473 static int kvm_get_dirty_log_protect(struct kvm *kvm, struct kvm_dirty_log *log)
1475 struct kvm_memslots *slots;
1476 struct kvm_memory_slot *memslot;
1479 unsigned long *dirty_bitmap;
1480 unsigned long *dirty_bitmap_buffer;
1483 as_id = log->slot >> 16;
1484 id = (u16)log->slot;
1485 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1488 slots = __kvm_memslots(kvm, as_id);
1489 memslot = id_to_memslot(slots, id);
1490 if (!memslot || !memslot->dirty_bitmap)
1493 dirty_bitmap = memslot->dirty_bitmap;
1495 kvm_arch_sync_dirty_log(kvm, memslot);
1497 n = kvm_dirty_bitmap_bytes(memslot);
1499 if (kvm->manual_dirty_log_protect) {
1501 * Unlike kvm_get_dirty_log, we always return false in *flush,
1502 * because no flush is needed until KVM_CLEAR_DIRTY_LOG. There
1503 * is some code duplication between this function and
1504 * kvm_get_dirty_log, but hopefully all architecture
1505 * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log
1506 * can be eliminated.
1508 dirty_bitmap_buffer = dirty_bitmap;
1510 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
1511 memset(dirty_bitmap_buffer, 0, n);
1513 spin_lock(&kvm->mmu_lock);
1514 for (i = 0; i < n / sizeof(long); i++) {
1518 if (!dirty_bitmap[i])
1522 mask = xchg(&dirty_bitmap[i], 0);
1523 dirty_bitmap_buffer[i] = mask;
1525 offset = i * BITS_PER_LONG;
1526 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1529 spin_unlock(&kvm->mmu_lock);
1533 kvm_arch_flush_remote_tlbs_memslot(kvm, memslot);
1535 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1542 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
1543 * @kvm: kvm instance
1544 * @log: slot id and address to which we copy the log
1546 * Steps 1-4 below provide general overview of dirty page logging. See
1547 * kvm_get_dirty_log_protect() function description for additional details.
1549 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
1550 * always flush the TLB (step 4) even if previous step failed and the dirty
1551 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
1552 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
1553 * writes will be marked dirty for next log read.
1555 * 1. Take a snapshot of the bit and clear it if needed.
1556 * 2. Write protect the corresponding page.
1557 * 3. Copy the snapshot to the userspace.
1558 * 4. Flush TLB's if needed.
1560 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1561 struct kvm_dirty_log *log)
1565 mutex_lock(&kvm->slots_lock);
1567 r = kvm_get_dirty_log_protect(kvm, log);
1569 mutex_unlock(&kvm->slots_lock);
1574 * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap
1575 * and reenable dirty page tracking for the corresponding pages.
1576 * @kvm: pointer to kvm instance
1577 * @log: slot id and address from which to fetch the bitmap of dirty pages
1579 static int kvm_clear_dirty_log_protect(struct kvm *kvm,
1580 struct kvm_clear_dirty_log *log)
1582 struct kvm_memslots *slots;
1583 struct kvm_memory_slot *memslot;
1587 unsigned long *dirty_bitmap;
1588 unsigned long *dirty_bitmap_buffer;
1591 as_id = log->slot >> 16;
1592 id = (u16)log->slot;
1593 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1596 if (log->first_page & 63)
1599 slots = __kvm_memslots(kvm, as_id);
1600 memslot = id_to_memslot(slots, id);
1601 if (!memslot || !memslot->dirty_bitmap)
1604 dirty_bitmap = memslot->dirty_bitmap;
1606 n = ALIGN(log->num_pages, BITS_PER_LONG) / 8;
1608 if (log->first_page > memslot->npages ||
1609 log->num_pages > memslot->npages - log->first_page ||
1610 (log->num_pages < memslot->npages - log->first_page && (log->num_pages & 63)))
1613 kvm_arch_sync_dirty_log(kvm, memslot);
1616 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
1617 if (copy_from_user(dirty_bitmap_buffer, log->dirty_bitmap, n))
1620 spin_lock(&kvm->mmu_lock);
1621 for (offset = log->first_page, i = offset / BITS_PER_LONG,
1622 n = DIV_ROUND_UP(log->num_pages, BITS_PER_LONG); n--;
1623 i++, offset += BITS_PER_LONG) {
1624 unsigned long mask = *dirty_bitmap_buffer++;
1625 atomic_long_t *p = (atomic_long_t *) &dirty_bitmap[i];
1629 mask &= atomic_long_fetch_andnot(mask, p);
1632 * mask contains the bits that really have been cleared. This
1633 * never includes any bits beyond the length of the memslot (if
1634 * the length is not aligned to 64 pages), therefore it is not
1635 * a problem if userspace sets them in log->dirty_bitmap.
1639 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1643 spin_unlock(&kvm->mmu_lock);
1646 kvm_arch_flush_remote_tlbs_memslot(kvm, memslot);
1651 static int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm,
1652 struct kvm_clear_dirty_log *log)
1656 mutex_lock(&kvm->slots_lock);
1658 r = kvm_clear_dirty_log_protect(kvm, log);
1660 mutex_unlock(&kvm->slots_lock);
1663 #endif /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
1665 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1667 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1669 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1671 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
1673 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn);
1675 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_memslot);
1677 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1679 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1681 return kvm_is_visible_memslot(memslot);
1683 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1685 bool kvm_vcpu_is_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
1687 struct kvm_memory_slot *memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1689 return kvm_is_visible_memslot(memslot);
1691 EXPORT_SYMBOL_GPL(kvm_vcpu_is_visible_gfn);
1693 unsigned long kvm_host_page_size(struct kvm_vcpu *vcpu, gfn_t gfn)
1695 struct vm_area_struct *vma;
1696 unsigned long addr, size;
1700 addr = kvm_vcpu_gfn_to_hva_prot(vcpu, gfn, NULL);
1701 if (kvm_is_error_hva(addr))
1704 mmap_read_lock(current->mm);
1705 vma = find_vma(current->mm, addr);
1709 size = vma_kernel_pagesize(vma);
1712 mmap_read_unlock(current->mm);
1717 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1719 return slot->flags & KVM_MEM_READONLY;
1722 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1723 gfn_t *nr_pages, bool write)
1725 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1726 return KVM_HVA_ERR_BAD;
1728 if (memslot_is_readonly(slot) && write)
1729 return KVM_HVA_ERR_RO_BAD;
1732 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1734 return __gfn_to_hva_memslot(slot, gfn);
1737 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1740 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1743 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1746 return gfn_to_hva_many(slot, gfn, NULL);
1748 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1750 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1752 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1754 EXPORT_SYMBOL_GPL(gfn_to_hva);
1756 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
1758 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
1760 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
1763 * Return the hva of a @gfn and the R/W attribute if possible.
1765 * @slot: the kvm_memory_slot which contains @gfn
1766 * @gfn: the gfn to be translated
1767 * @writable: used to return the read/write attribute of the @slot if the hva
1768 * is valid and @writable is not NULL
1770 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1771 gfn_t gfn, bool *writable)
1773 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1775 if (!kvm_is_error_hva(hva) && writable)
1776 *writable = !memslot_is_readonly(slot);
1781 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1783 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1785 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1788 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
1790 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1792 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1795 static inline int check_user_page_hwpoison(unsigned long addr)
1797 int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
1799 rc = get_user_pages(addr, 1, flags, NULL, NULL);
1800 return rc == -EHWPOISON;
1804 * The fast path to get the writable pfn which will be stored in @pfn,
1805 * true indicates success, otherwise false is returned. It's also the
1806 * only part that runs if we can in atomic context.
1808 static bool hva_to_pfn_fast(unsigned long addr, bool write_fault,
1809 bool *writable, kvm_pfn_t *pfn)
1811 struct page *page[1];
1814 * Fast pin a writable pfn only if it is a write fault request
1815 * or the caller allows to map a writable pfn for a read fault
1818 if (!(write_fault || writable))
1821 if (get_user_page_fast_only(addr, FOLL_WRITE, page)) {
1822 *pfn = page_to_pfn(page[0]);
1833 * The slow path to get the pfn of the specified host virtual address,
1834 * 1 indicates success, -errno is returned if error is detected.
1836 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1837 bool *writable, kvm_pfn_t *pfn)
1839 unsigned int flags = FOLL_HWPOISON;
1846 *writable = write_fault;
1849 flags |= FOLL_WRITE;
1851 flags |= FOLL_NOWAIT;
1853 npages = get_user_pages_unlocked(addr, 1, &page, flags);
1857 /* map read fault as writable if possible */
1858 if (unlikely(!write_fault) && writable) {
1861 if (get_user_page_fast_only(addr, FOLL_WRITE, &wpage)) {
1867 *pfn = page_to_pfn(page);
1871 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1873 if (unlikely(!(vma->vm_flags & VM_READ)))
1876 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1882 static int hva_to_pfn_remapped(struct vm_area_struct *vma,
1883 unsigned long addr, bool *async,
1884 bool write_fault, bool *writable,
1890 r = follow_pfn(vma, addr, &pfn);
1893 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
1894 * not call the fault handler, so do it here.
1896 bool unlocked = false;
1897 r = fixup_user_fault(current->mm, addr,
1898 (write_fault ? FAULT_FLAG_WRITE : 0),
1905 r = follow_pfn(vma, addr, &pfn);
1915 * Get a reference here because callers of *hva_to_pfn* and
1916 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
1917 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
1918 * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will
1919 * simply do nothing for reserved pfns.
1921 * Whoever called remap_pfn_range is also going to call e.g.
1922 * unmap_mapping_range before the underlying pages are freed,
1923 * causing a call to our MMU notifier.
1932 * Pin guest page in memory and return its pfn.
1933 * @addr: host virtual address which maps memory to the guest
1934 * @atomic: whether this function can sleep
1935 * @async: whether this function need to wait IO complete if the
1936 * host page is not in the memory
1937 * @write_fault: whether we should get a writable host page
1938 * @writable: whether it allows to map a writable host page for !@write_fault
1940 * The function will map a writable host page for these two cases:
1941 * 1): @write_fault = true
1942 * 2): @write_fault = false && @writable, @writable will tell the caller
1943 * whether the mapping is writable.
1945 static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1946 bool write_fault, bool *writable)
1948 struct vm_area_struct *vma;
1952 /* we can do it either atomically or asynchronously, not both */
1953 BUG_ON(atomic && async);
1955 if (hva_to_pfn_fast(addr, write_fault, writable, &pfn))
1959 return KVM_PFN_ERR_FAULT;
1961 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1965 mmap_read_lock(current->mm);
1966 if (npages == -EHWPOISON ||
1967 (!async && check_user_page_hwpoison(addr))) {
1968 pfn = KVM_PFN_ERR_HWPOISON;
1973 vma = find_vma_intersection(current->mm, addr, addr + 1);
1976 pfn = KVM_PFN_ERR_FAULT;
1977 else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
1978 r = hva_to_pfn_remapped(vma, addr, async, write_fault, writable, &pfn);
1982 pfn = KVM_PFN_ERR_FAULT;
1984 if (async && vma_is_valid(vma, write_fault))
1986 pfn = KVM_PFN_ERR_FAULT;
1989 mmap_read_unlock(current->mm);
1993 kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn,
1994 bool atomic, bool *async, bool write_fault,
1997 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1999 if (addr == KVM_HVA_ERR_RO_BAD) {
2002 return KVM_PFN_ERR_RO_FAULT;
2005 if (kvm_is_error_hva(addr)) {
2008 return KVM_PFN_NOSLOT;
2011 /* Do not map writable pfn in the readonly memslot. */
2012 if (writable && memslot_is_readonly(slot)) {
2017 return hva_to_pfn(addr, atomic, async, write_fault,
2020 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
2022 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
2025 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
2026 write_fault, writable);
2028 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
2030 kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
2032 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
2034 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
2036 kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
2038 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
2040 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
2042 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
2044 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
2046 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
2048 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
2050 return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
2052 EXPORT_SYMBOL_GPL(gfn_to_pfn);
2054 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
2056 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
2058 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
2060 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
2061 struct page **pages, int nr_pages)
2066 addr = gfn_to_hva_many(slot, gfn, &entry);
2067 if (kvm_is_error_hva(addr))
2070 if (entry < nr_pages)
2073 return get_user_pages_fast_only(addr, nr_pages, FOLL_WRITE, pages);
2075 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
2077 static struct page *kvm_pfn_to_page(kvm_pfn_t pfn)
2079 if (is_error_noslot_pfn(pfn))
2080 return KVM_ERR_PTR_BAD_PAGE;
2082 if (kvm_is_reserved_pfn(pfn)) {
2084 return KVM_ERR_PTR_BAD_PAGE;
2087 return pfn_to_page(pfn);
2090 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
2094 pfn = gfn_to_pfn(kvm, gfn);
2096 return kvm_pfn_to_page(pfn);
2098 EXPORT_SYMBOL_GPL(gfn_to_page);
2100 void kvm_release_pfn(kvm_pfn_t pfn, bool dirty, struct gfn_to_pfn_cache *cache)
2106 cache->pfn = cache->gfn = 0;
2109 kvm_release_pfn_dirty(pfn);
2111 kvm_release_pfn_clean(pfn);
2114 static void kvm_cache_gfn_to_pfn(struct kvm_memory_slot *slot, gfn_t gfn,
2115 struct gfn_to_pfn_cache *cache, u64 gen)
2117 kvm_release_pfn(cache->pfn, cache->dirty, cache);
2119 cache->pfn = gfn_to_pfn_memslot(slot, gfn);
2121 cache->dirty = false;
2122 cache->generation = gen;
2125 static int __kvm_map_gfn(struct kvm_memslots *slots, gfn_t gfn,
2126 struct kvm_host_map *map,
2127 struct gfn_to_pfn_cache *cache,
2132 struct page *page = KVM_UNMAPPED_PAGE;
2133 struct kvm_memory_slot *slot = __gfn_to_memslot(slots, gfn);
2134 u64 gen = slots->generation;
2140 if (!cache->pfn || cache->gfn != gfn ||
2141 cache->generation != gen) {
2144 kvm_cache_gfn_to_pfn(slot, gfn, cache, gen);
2150 pfn = gfn_to_pfn_memslot(slot, gfn);
2152 if (is_error_noslot_pfn(pfn))
2155 if (pfn_valid(pfn)) {
2156 page = pfn_to_page(pfn);
2158 hva = kmap_atomic(page);
2161 #ifdef CONFIG_HAS_IOMEM
2162 } else if (!atomic) {
2163 hva = memremap(pfn_to_hpa(pfn), PAGE_SIZE, MEMREMAP_WB);
2180 int kvm_map_gfn(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map,
2181 struct gfn_to_pfn_cache *cache, bool atomic)
2183 return __kvm_map_gfn(kvm_memslots(vcpu->kvm), gfn, map,
2186 EXPORT_SYMBOL_GPL(kvm_map_gfn);
2188 int kvm_vcpu_map(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map)
2190 return __kvm_map_gfn(kvm_vcpu_memslots(vcpu), gfn, map,
2193 EXPORT_SYMBOL_GPL(kvm_vcpu_map);
2195 static void __kvm_unmap_gfn(struct kvm_memory_slot *memslot,
2196 struct kvm_host_map *map,
2197 struct gfn_to_pfn_cache *cache,
2198 bool dirty, bool atomic)
2206 if (map->page != KVM_UNMAPPED_PAGE) {
2208 kunmap_atomic(map->hva);
2212 #ifdef CONFIG_HAS_IOMEM
2216 WARN_ONCE(1, "Unexpected unmapping in atomic context");
2220 mark_page_dirty_in_slot(memslot, map->gfn);
2223 cache->dirty |= dirty;
2225 kvm_release_pfn(map->pfn, dirty, NULL);
2231 int kvm_unmap_gfn(struct kvm_vcpu *vcpu, struct kvm_host_map *map,
2232 struct gfn_to_pfn_cache *cache, bool dirty, bool atomic)
2234 __kvm_unmap_gfn(gfn_to_memslot(vcpu->kvm, map->gfn), map,
2235 cache, dirty, atomic);
2238 EXPORT_SYMBOL_GPL(kvm_unmap_gfn);
2240 void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map, bool dirty)
2242 __kvm_unmap_gfn(kvm_vcpu_gfn_to_memslot(vcpu, map->gfn), map, NULL,
2245 EXPORT_SYMBOL_GPL(kvm_vcpu_unmap);
2247 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
2251 pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
2253 return kvm_pfn_to_page(pfn);
2255 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
2257 void kvm_release_page_clean(struct page *page)
2259 WARN_ON(is_error_page(page));
2261 kvm_release_pfn_clean(page_to_pfn(page));
2263 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
2265 void kvm_release_pfn_clean(kvm_pfn_t pfn)
2267 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
2268 put_page(pfn_to_page(pfn));
2270 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
2272 void kvm_release_page_dirty(struct page *page)
2274 WARN_ON(is_error_page(page));
2276 kvm_release_pfn_dirty(page_to_pfn(page));
2278 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
2280 void kvm_release_pfn_dirty(kvm_pfn_t pfn)
2282 kvm_set_pfn_dirty(pfn);
2283 kvm_release_pfn_clean(pfn);
2285 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
2287 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
2289 if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn))
2290 SetPageDirty(pfn_to_page(pfn));
2292 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
2294 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
2296 if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn))
2297 mark_page_accessed(pfn_to_page(pfn));
2299 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
2301 void kvm_get_pfn(kvm_pfn_t pfn)
2303 if (!kvm_is_reserved_pfn(pfn))
2304 get_page(pfn_to_page(pfn));
2306 EXPORT_SYMBOL_GPL(kvm_get_pfn);
2308 static int next_segment(unsigned long len, int offset)
2310 if (len > PAGE_SIZE - offset)
2311 return PAGE_SIZE - offset;
2316 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
2317 void *data, int offset, int len)
2322 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
2323 if (kvm_is_error_hva(addr))
2325 r = __copy_from_user(data, (void __user *)addr + offset, len);
2331 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
2334 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2336 return __kvm_read_guest_page(slot, gfn, data, offset, len);
2338 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
2340 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
2341 int offset, int len)
2343 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2345 return __kvm_read_guest_page(slot, gfn, data, offset, len);
2347 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
2349 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
2351 gfn_t gfn = gpa >> PAGE_SHIFT;
2353 int offset = offset_in_page(gpa);
2356 while ((seg = next_segment(len, offset)) != 0) {
2357 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
2367 EXPORT_SYMBOL_GPL(kvm_read_guest);
2369 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
2371 gfn_t gfn = gpa >> PAGE_SHIFT;
2373 int offset = offset_in_page(gpa);
2376 while ((seg = next_segment(len, offset)) != 0) {
2377 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
2387 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
2389 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
2390 void *data, int offset, unsigned long len)
2395 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
2396 if (kvm_is_error_hva(addr))
2398 pagefault_disable();
2399 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
2406 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
2407 void *data, unsigned long len)
2409 gfn_t gfn = gpa >> PAGE_SHIFT;
2410 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2411 int offset = offset_in_page(gpa);
2413 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
2415 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
2417 static int __kvm_write_guest_page(struct kvm_memory_slot *memslot, gfn_t gfn,
2418 const void *data, int offset, int len)
2423 addr = gfn_to_hva_memslot(memslot, gfn);
2424 if (kvm_is_error_hva(addr))
2426 r = __copy_to_user((void __user *)addr + offset, data, len);
2429 mark_page_dirty_in_slot(memslot, gfn);
2433 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
2434 const void *data, int offset, int len)
2436 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2438 return __kvm_write_guest_page(slot, gfn, data, offset, len);
2440 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
2442 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
2443 const void *data, int offset, int len)
2445 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2447 return __kvm_write_guest_page(slot, gfn, data, offset, len);
2449 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
2451 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
2454 gfn_t gfn = gpa >> PAGE_SHIFT;
2456 int offset = offset_in_page(gpa);
2459 while ((seg = next_segment(len, offset)) != 0) {
2460 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
2470 EXPORT_SYMBOL_GPL(kvm_write_guest);
2472 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
2475 gfn_t gfn = gpa >> PAGE_SHIFT;
2477 int offset = offset_in_page(gpa);
2480 while ((seg = next_segment(len, offset)) != 0) {
2481 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
2491 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
2493 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
2494 struct gfn_to_hva_cache *ghc,
2495 gpa_t gpa, unsigned long len)
2497 int offset = offset_in_page(gpa);
2498 gfn_t start_gfn = gpa >> PAGE_SHIFT;
2499 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
2500 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
2501 gfn_t nr_pages_avail;
2503 /* Update ghc->generation before performing any error checks. */
2504 ghc->generation = slots->generation;
2506 if (start_gfn > end_gfn) {
2507 ghc->hva = KVM_HVA_ERR_BAD;
2512 * If the requested region crosses two memslots, we still
2513 * verify that the entire region is valid here.
2515 for ( ; start_gfn <= end_gfn; start_gfn += nr_pages_avail) {
2516 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
2517 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
2519 if (kvm_is_error_hva(ghc->hva))
2523 /* Use the slow path for cross page reads and writes. */
2524 if (nr_pages_needed == 1)
2527 ghc->memslot = NULL;
2534 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2535 gpa_t gpa, unsigned long len)
2537 struct kvm_memslots *slots = kvm_memslots(kvm);
2538 return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len);
2540 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
2542 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2543 void *data, unsigned int offset,
2546 struct kvm_memslots *slots = kvm_memslots(kvm);
2548 gpa_t gpa = ghc->gpa + offset;
2550 BUG_ON(len + offset > ghc->len);
2552 if (slots->generation != ghc->generation) {
2553 if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len))
2557 if (kvm_is_error_hva(ghc->hva))
2560 if (unlikely(!ghc->memslot))
2561 return kvm_write_guest(kvm, gpa, data, len);
2563 r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
2566 mark_page_dirty_in_slot(ghc->memslot, gpa >> PAGE_SHIFT);
2570 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached);
2572 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2573 void *data, unsigned long len)
2575 return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len);
2577 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
2579 int kvm_read_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2580 void *data, unsigned int offset,
2583 struct kvm_memslots *slots = kvm_memslots(kvm);
2585 gpa_t gpa = ghc->gpa + offset;
2587 BUG_ON(len + offset > ghc->len);
2589 if (slots->generation != ghc->generation) {
2590 if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len))
2594 if (kvm_is_error_hva(ghc->hva))
2597 if (unlikely(!ghc->memslot))
2598 return kvm_read_guest(kvm, gpa, data, len);
2600 r = __copy_from_user(data, (void __user *)ghc->hva + offset, len);
2606 EXPORT_SYMBOL_GPL(kvm_read_guest_offset_cached);
2608 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2609 void *data, unsigned long len)
2611 return kvm_read_guest_offset_cached(kvm, ghc, data, 0, len);
2613 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
2615 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
2617 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
2619 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
2621 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
2623 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
2625 gfn_t gfn = gpa >> PAGE_SHIFT;
2627 int offset = offset_in_page(gpa);
2630 while ((seg = next_segment(len, offset)) != 0) {
2631 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
2640 EXPORT_SYMBOL_GPL(kvm_clear_guest);
2642 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot,
2645 if (memslot && memslot->dirty_bitmap) {
2646 unsigned long rel_gfn = gfn - memslot->base_gfn;
2648 set_bit_le(rel_gfn, memslot->dirty_bitmap);
2652 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
2654 struct kvm_memory_slot *memslot;
2656 memslot = gfn_to_memslot(kvm, gfn);
2657 mark_page_dirty_in_slot(memslot, gfn);
2659 EXPORT_SYMBOL_GPL(mark_page_dirty);
2661 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
2663 struct kvm_memory_slot *memslot;
2665 memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2666 mark_page_dirty_in_slot(memslot, gfn);
2668 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
2670 void kvm_sigset_activate(struct kvm_vcpu *vcpu)
2672 if (!vcpu->sigset_active)
2676 * This does a lockless modification of ->real_blocked, which is fine
2677 * because, only current can change ->real_blocked and all readers of
2678 * ->real_blocked don't care as long ->real_blocked is always a subset
2681 sigprocmask(SIG_SETMASK, &vcpu->sigset, ¤t->real_blocked);
2684 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu)
2686 if (!vcpu->sigset_active)
2689 sigprocmask(SIG_SETMASK, ¤t->real_blocked, NULL);
2690 sigemptyset(¤t->real_blocked);
2693 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
2695 unsigned int old, val, grow, grow_start;
2697 old = val = vcpu->halt_poll_ns;
2698 grow_start = READ_ONCE(halt_poll_ns_grow_start);
2699 grow = READ_ONCE(halt_poll_ns_grow);
2704 if (val < grow_start)
2707 if (val > halt_poll_ns)
2710 vcpu->halt_poll_ns = val;
2712 trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
2715 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
2717 unsigned int old, val, shrink;
2719 old = val = vcpu->halt_poll_ns;
2720 shrink = READ_ONCE(halt_poll_ns_shrink);
2726 vcpu->halt_poll_ns = val;
2727 trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
2730 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
2733 int idx = srcu_read_lock(&vcpu->kvm->srcu);
2735 if (kvm_arch_vcpu_runnable(vcpu)) {
2736 kvm_make_request(KVM_REQ_UNHALT, vcpu);
2739 if (kvm_cpu_has_pending_timer(vcpu))
2741 if (signal_pending(current))
2746 srcu_read_unlock(&vcpu->kvm->srcu, idx);
2751 update_halt_poll_stats(struct kvm_vcpu *vcpu, u64 poll_ns, bool waited)
2754 vcpu->stat.halt_poll_fail_ns += poll_ns;
2756 vcpu->stat.halt_poll_success_ns += poll_ns;
2760 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2762 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
2764 ktime_t start, cur, poll_end;
2765 bool waited = false;
2768 kvm_arch_vcpu_blocking(vcpu);
2770 start = cur = poll_end = ktime_get();
2771 if (vcpu->halt_poll_ns && !kvm_arch_no_poll(vcpu)) {
2772 ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns);
2774 ++vcpu->stat.halt_attempted_poll;
2777 * This sets KVM_REQ_UNHALT if an interrupt
2780 if (kvm_vcpu_check_block(vcpu) < 0) {
2781 ++vcpu->stat.halt_successful_poll;
2782 if (!vcpu_valid_wakeup(vcpu))
2783 ++vcpu->stat.halt_poll_invalid;
2786 poll_end = cur = ktime_get();
2787 } while (single_task_running() && ktime_before(cur, stop));
2790 prepare_to_rcuwait(&vcpu->wait);
2792 set_current_state(TASK_INTERRUPTIBLE);
2794 if (kvm_vcpu_check_block(vcpu) < 0)
2800 finish_rcuwait(&vcpu->wait);
2803 kvm_arch_vcpu_unblocking(vcpu);
2804 block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
2806 update_halt_poll_stats(
2807 vcpu, ktime_to_ns(ktime_sub(poll_end, start)), waited);
2809 if (!kvm_arch_no_poll(vcpu)) {
2810 if (!vcpu_valid_wakeup(vcpu)) {
2811 shrink_halt_poll_ns(vcpu);
2812 } else if (vcpu->kvm->max_halt_poll_ns) {
2813 if (block_ns <= vcpu->halt_poll_ns)
2815 /* we had a long block, shrink polling */
2816 else if (vcpu->halt_poll_ns &&
2817 block_ns > vcpu->kvm->max_halt_poll_ns)
2818 shrink_halt_poll_ns(vcpu);
2819 /* we had a short halt and our poll time is too small */
2820 else if (vcpu->halt_poll_ns < vcpu->kvm->max_halt_poll_ns &&
2821 block_ns < vcpu->kvm->max_halt_poll_ns)
2822 grow_halt_poll_ns(vcpu);
2824 vcpu->halt_poll_ns = 0;
2828 trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu));
2829 kvm_arch_vcpu_block_finish(vcpu);
2831 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
2833 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
2835 struct rcuwait *waitp;
2837 waitp = kvm_arch_vcpu_get_wait(vcpu);
2838 if (rcuwait_wake_up(waitp)) {
2839 WRITE_ONCE(vcpu->ready, true);
2840 ++vcpu->stat.halt_wakeup;
2846 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
2850 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2852 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
2855 int cpu = vcpu->cpu;
2857 if (kvm_vcpu_wake_up(vcpu))
2861 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
2862 if (kvm_arch_vcpu_should_kick(vcpu))
2863 smp_send_reschedule(cpu);
2866 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
2867 #endif /* !CONFIG_S390 */
2869 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
2872 struct task_struct *task = NULL;
2876 pid = rcu_dereference(target->pid);
2878 task = get_pid_task(pid, PIDTYPE_PID);
2882 ret = yield_to(task, 1);
2883 put_task_struct(task);
2887 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
2890 * Helper that checks whether a VCPU is eligible for directed yield.
2891 * Most eligible candidate to yield is decided by following heuristics:
2893 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2894 * (preempted lock holder), indicated by @in_spin_loop.
2895 * Set at the beginning and cleared at the end of interception/PLE handler.
2897 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2898 * chance last time (mostly it has become eligible now since we have probably
2899 * yielded to lockholder in last iteration. This is done by toggling
2900 * @dy_eligible each time a VCPU checked for eligibility.)
2902 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2903 * to preempted lock-holder could result in wrong VCPU selection and CPU
2904 * burning. Giving priority for a potential lock-holder increases lock
2907 * Since algorithm is based on heuristics, accessing another VCPU data without
2908 * locking does not harm. It may result in trying to yield to same VCPU, fail
2909 * and continue with next VCPU and so on.
2911 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
2913 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2916 eligible = !vcpu->spin_loop.in_spin_loop ||
2917 vcpu->spin_loop.dy_eligible;
2919 if (vcpu->spin_loop.in_spin_loop)
2920 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
2929 * Unlike kvm_arch_vcpu_runnable, this function is called outside
2930 * a vcpu_load/vcpu_put pair. However, for most architectures
2931 * kvm_arch_vcpu_runnable does not require vcpu_load.
2933 bool __weak kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
2935 return kvm_arch_vcpu_runnable(vcpu);
2938 static bool vcpu_dy_runnable(struct kvm_vcpu *vcpu)
2940 if (kvm_arch_dy_runnable(vcpu))
2943 #ifdef CONFIG_KVM_ASYNC_PF
2944 if (!list_empty_careful(&vcpu->async_pf.done))
2951 void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode)
2953 struct kvm *kvm = me->kvm;
2954 struct kvm_vcpu *vcpu;
2955 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
2961 kvm_vcpu_set_in_spin_loop(me, true);
2963 * We boost the priority of a VCPU that is runnable but not
2964 * currently running, because it got preempted by something
2965 * else and called schedule in __vcpu_run. Hopefully that
2966 * VCPU is holding the lock that we need and will release it.
2967 * We approximate round-robin by starting at the last boosted VCPU.
2969 for (pass = 0; pass < 2 && !yielded && try; pass++) {
2970 kvm_for_each_vcpu(i, vcpu, kvm) {
2971 if (!pass && i <= last_boosted_vcpu) {
2972 i = last_boosted_vcpu;
2974 } else if (pass && i > last_boosted_vcpu)
2976 if (!READ_ONCE(vcpu->ready))
2980 if (rcuwait_active(&vcpu->wait) &&
2981 !vcpu_dy_runnable(vcpu))
2983 if (READ_ONCE(vcpu->preempted) && yield_to_kernel_mode &&
2984 !kvm_arch_vcpu_in_kernel(vcpu))
2986 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
2989 yielded = kvm_vcpu_yield_to(vcpu);
2991 kvm->last_boosted_vcpu = i;
2993 } else if (yielded < 0) {
3000 kvm_vcpu_set_in_spin_loop(me, false);
3002 /* Ensure vcpu is not eligible during next spinloop */
3003 kvm_vcpu_set_dy_eligible(me, false);
3005 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
3007 static vm_fault_t kvm_vcpu_fault(struct vm_fault *vmf)
3009 struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data;
3012 if (vmf->pgoff == 0)
3013 page = virt_to_page(vcpu->run);
3015 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
3016 page = virt_to_page(vcpu->arch.pio_data);
3018 #ifdef CONFIG_KVM_MMIO
3019 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
3020 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
3023 return kvm_arch_vcpu_fault(vcpu, vmf);
3029 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
3030 .fault = kvm_vcpu_fault,
3033 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
3035 vma->vm_ops = &kvm_vcpu_vm_ops;
3039 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
3041 struct kvm_vcpu *vcpu = filp->private_data;
3043 kvm_put_kvm(vcpu->kvm);
3047 static struct file_operations kvm_vcpu_fops = {
3048 .release = kvm_vcpu_release,
3049 .unlocked_ioctl = kvm_vcpu_ioctl,
3050 .mmap = kvm_vcpu_mmap,
3051 .llseek = noop_llseek,
3052 KVM_COMPAT(kvm_vcpu_compat_ioctl),
3056 * Allocates an inode for the vcpu.
3058 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
3060 char name[8 + 1 + ITOA_MAX_LEN + 1];
3062 snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id);
3063 return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
3066 static void kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
3068 #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS
3069 struct dentry *debugfs_dentry;
3070 char dir_name[ITOA_MAX_LEN * 2];
3072 if (!debugfs_initialized())
3075 snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
3076 debugfs_dentry = debugfs_create_dir(dir_name,
3077 vcpu->kvm->debugfs_dentry);
3079 kvm_arch_create_vcpu_debugfs(vcpu, debugfs_dentry);
3084 * Creates some virtual cpus. Good luck creating more than one.
3086 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
3089 struct kvm_vcpu *vcpu;
3092 if (id >= KVM_MAX_VCPU_ID)
3095 mutex_lock(&kvm->lock);
3096 if (kvm->created_vcpus == KVM_MAX_VCPUS) {
3097 mutex_unlock(&kvm->lock);
3101 kvm->created_vcpus++;
3102 mutex_unlock(&kvm->lock);
3104 r = kvm_arch_vcpu_precreate(kvm, id);
3106 goto vcpu_decrement;
3108 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
3111 goto vcpu_decrement;
3114 BUILD_BUG_ON(sizeof(struct kvm_run) > PAGE_SIZE);
3115 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
3120 vcpu->run = page_address(page);
3122 kvm_vcpu_init(vcpu, kvm, id);
3124 r = kvm_arch_vcpu_create(vcpu);
3126 goto vcpu_free_run_page;
3128 mutex_lock(&kvm->lock);
3129 if (kvm_get_vcpu_by_id(kvm, id)) {
3131 goto unlock_vcpu_destroy;
3134 vcpu->vcpu_idx = atomic_read(&kvm->online_vcpus);
3135 BUG_ON(kvm->vcpus[vcpu->vcpu_idx]);
3137 /* Now it's all set up, let userspace reach it */
3139 r = create_vcpu_fd(vcpu);
3141 kvm_put_kvm_no_destroy(kvm);
3142 goto unlock_vcpu_destroy;
3145 kvm->vcpus[vcpu->vcpu_idx] = vcpu;
3148 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
3149 * before kvm->online_vcpu's incremented value.
3152 atomic_inc(&kvm->online_vcpus);
3154 mutex_unlock(&kvm->lock);
3155 kvm_arch_vcpu_postcreate(vcpu);
3156 kvm_create_vcpu_debugfs(vcpu);
3159 unlock_vcpu_destroy:
3160 mutex_unlock(&kvm->lock);
3161 kvm_arch_vcpu_destroy(vcpu);
3163 free_page((unsigned long)vcpu->run);
3165 kmem_cache_free(kvm_vcpu_cache, vcpu);
3167 mutex_lock(&kvm->lock);
3168 kvm->created_vcpus--;
3169 mutex_unlock(&kvm->lock);
3173 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
3176 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
3177 vcpu->sigset_active = 1;
3178 vcpu->sigset = *sigset;
3180 vcpu->sigset_active = 0;
3184 static long kvm_vcpu_ioctl(struct file *filp,
3185 unsigned int ioctl, unsigned long arg)
3187 struct kvm_vcpu *vcpu = filp->private_data;
3188 void __user *argp = (void __user *)arg;
3190 struct kvm_fpu *fpu = NULL;
3191 struct kvm_sregs *kvm_sregs = NULL;
3193 if (vcpu->kvm->mm != current->mm)
3196 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
3200 * Some architectures have vcpu ioctls that are asynchronous to vcpu
3201 * execution; mutex_lock() would break them.
3203 r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg);
3204 if (r != -ENOIOCTLCMD)
3207 if (mutex_lock_killable(&vcpu->mutex))
3215 oldpid = rcu_access_pointer(vcpu->pid);
3216 if (unlikely(oldpid != task_pid(current))) {
3217 /* The thread running this VCPU changed. */
3220 r = kvm_arch_vcpu_run_pid_change(vcpu);
3224 newpid = get_task_pid(current, PIDTYPE_PID);
3225 rcu_assign_pointer(vcpu->pid, newpid);
3230 r = kvm_arch_vcpu_ioctl_run(vcpu);
3231 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
3234 case KVM_GET_REGS: {
3235 struct kvm_regs *kvm_regs;
3238 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL_ACCOUNT);
3241 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
3245 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
3252 case KVM_SET_REGS: {
3253 struct kvm_regs *kvm_regs;
3255 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
3256 if (IS_ERR(kvm_regs)) {
3257 r = PTR_ERR(kvm_regs);
3260 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
3264 case KVM_GET_SREGS: {
3265 kvm_sregs = kzalloc(sizeof(struct kvm_sregs),
3266 GFP_KERNEL_ACCOUNT);
3270 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
3274 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
3279 case KVM_SET_SREGS: {
3280 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
3281 if (IS_ERR(kvm_sregs)) {
3282 r = PTR_ERR(kvm_sregs);
3286 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
3289 case KVM_GET_MP_STATE: {
3290 struct kvm_mp_state mp_state;
3292 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
3296 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
3301 case KVM_SET_MP_STATE: {
3302 struct kvm_mp_state mp_state;
3305 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
3307 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
3310 case KVM_TRANSLATE: {
3311 struct kvm_translation tr;
3314 if (copy_from_user(&tr, argp, sizeof(tr)))
3316 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
3320 if (copy_to_user(argp, &tr, sizeof(tr)))
3325 case KVM_SET_GUEST_DEBUG: {
3326 struct kvm_guest_debug dbg;
3329 if (copy_from_user(&dbg, argp, sizeof(dbg)))
3331 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
3334 case KVM_SET_SIGNAL_MASK: {
3335 struct kvm_signal_mask __user *sigmask_arg = argp;
3336 struct kvm_signal_mask kvm_sigmask;
3337 sigset_t sigset, *p;
3342 if (copy_from_user(&kvm_sigmask, argp,
3343 sizeof(kvm_sigmask)))
3346 if (kvm_sigmask.len != sizeof(sigset))
3349 if (copy_from_user(&sigset, sigmask_arg->sigset,
3354 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
3358 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL_ACCOUNT);
3362 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
3366 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
3372 fpu = memdup_user(argp, sizeof(*fpu));
3378 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
3382 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
3385 mutex_unlock(&vcpu->mutex);
3391 #ifdef CONFIG_KVM_COMPAT
3392 static long kvm_vcpu_compat_ioctl(struct file *filp,
3393 unsigned int ioctl, unsigned long arg)
3395 struct kvm_vcpu *vcpu = filp->private_data;
3396 void __user *argp = compat_ptr(arg);
3399 if (vcpu->kvm->mm != current->mm)
3403 case KVM_SET_SIGNAL_MASK: {
3404 struct kvm_signal_mask __user *sigmask_arg = argp;
3405 struct kvm_signal_mask kvm_sigmask;
3410 if (copy_from_user(&kvm_sigmask, argp,
3411 sizeof(kvm_sigmask)))
3414 if (kvm_sigmask.len != sizeof(compat_sigset_t))
3417 if (get_compat_sigset(&sigset,
3418 (compat_sigset_t __user *)sigmask_arg->sigset))
3420 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
3422 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
3426 r = kvm_vcpu_ioctl(filp, ioctl, arg);
3434 static int kvm_device_mmap(struct file *filp, struct vm_area_struct *vma)
3436 struct kvm_device *dev = filp->private_data;
3439 return dev->ops->mmap(dev, vma);
3444 static int kvm_device_ioctl_attr(struct kvm_device *dev,
3445 int (*accessor)(struct kvm_device *dev,
3446 struct kvm_device_attr *attr),
3449 struct kvm_device_attr attr;
3454 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
3457 return accessor(dev, &attr);
3460 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
3463 struct kvm_device *dev = filp->private_data;
3465 if (dev->kvm->mm != current->mm)
3469 case KVM_SET_DEVICE_ATTR:
3470 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
3471 case KVM_GET_DEVICE_ATTR:
3472 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
3473 case KVM_HAS_DEVICE_ATTR:
3474 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
3476 if (dev->ops->ioctl)
3477 return dev->ops->ioctl(dev, ioctl, arg);
3483 static int kvm_device_release(struct inode *inode, struct file *filp)
3485 struct kvm_device *dev = filp->private_data;
3486 struct kvm *kvm = dev->kvm;
3488 if (dev->ops->release) {
3489 mutex_lock(&kvm->lock);
3490 list_del(&dev->vm_node);
3491 dev->ops->release(dev);
3492 mutex_unlock(&kvm->lock);
3499 static const struct file_operations kvm_device_fops = {
3500 .unlocked_ioctl = kvm_device_ioctl,
3501 .release = kvm_device_release,
3502 KVM_COMPAT(kvm_device_ioctl),
3503 .mmap = kvm_device_mmap,
3506 struct kvm_device *kvm_device_from_filp(struct file *filp)
3508 if (filp->f_op != &kvm_device_fops)
3511 return filp->private_data;
3514 static const struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
3515 #ifdef CONFIG_KVM_MPIC
3516 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
3517 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
3521 int kvm_register_device_ops(const struct kvm_device_ops *ops, u32 type)
3523 if (type >= ARRAY_SIZE(kvm_device_ops_table))
3526 if (kvm_device_ops_table[type] != NULL)
3529 kvm_device_ops_table[type] = ops;
3533 void kvm_unregister_device_ops(u32 type)
3535 if (kvm_device_ops_table[type] != NULL)
3536 kvm_device_ops_table[type] = NULL;
3539 static int kvm_ioctl_create_device(struct kvm *kvm,
3540 struct kvm_create_device *cd)
3542 const struct kvm_device_ops *ops = NULL;
3543 struct kvm_device *dev;
3544 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
3548 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
3551 type = array_index_nospec(cd->type, ARRAY_SIZE(kvm_device_ops_table));
3552 ops = kvm_device_ops_table[type];
3559 dev = kzalloc(sizeof(*dev), GFP_KERNEL_ACCOUNT);
3566 mutex_lock(&kvm->lock);
3567 ret = ops->create(dev, type);
3569 mutex_unlock(&kvm->lock);
3573 list_add(&dev->vm_node, &kvm->devices);
3574 mutex_unlock(&kvm->lock);
3580 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
3582 kvm_put_kvm_no_destroy(kvm);
3583 mutex_lock(&kvm->lock);
3584 list_del(&dev->vm_node);
3585 mutex_unlock(&kvm->lock);
3594 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
3597 case KVM_CAP_USER_MEMORY:
3598 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
3599 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
3600 case KVM_CAP_INTERNAL_ERROR_DATA:
3601 #ifdef CONFIG_HAVE_KVM_MSI
3602 case KVM_CAP_SIGNAL_MSI:
3604 #ifdef CONFIG_HAVE_KVM_IRQFD
3606 case KVM_CAP_IRQFD_RESAMPLE:
3608 case KVM_CAP_IOEVENTFD_ANY_LENGTH:
3609 case KVM_CAP_CHECK_EXTENSION_VM:
3610 case KVM_CAP_ENABLE_CAP_VM:
3611 case KVM_CAP_HALT_POLL:
3613 #ifdef CONFIG_KVM_MMIO
3614 case KVM_CAP_COALESCED_MMIO:
3615 return KVM_COALESCED_MMIO_PAGE_OFFSET;
3616 case KVM_CAP_COALESCED_PIO:
3619 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3620 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2:
3621 return KVM_DIRTY_LOG_MANUAL_CAPS;
3623 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3624 case KVM_CAP_IRQ_ROUTING:
3625 return KVM_MAX_IRQ_ROUTES;
3627 #if KVM_ADDRESS_SPACE_NUM > 1
3628 case KVM_CAP_MULTI_ADDRESS_SPACE:
3629 return KVM_ADDRESS_SPACE_NUM;
3631 case KVM_CAP_NR_MEMSLOTS:
3632 return KVM_USER_MEM_SLOTS;
3636 return kvm_vm_ioctl_check_extension(kvm, arg);
3639 int __attribute__((weak)) kvm_vm_ioctl_enable_cap(struct kvm *kvm,
3640 struct kvm_enable_cap *cap)
3645 static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm,
3646 struct kvm_enable_cap *cap)
3649 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3650 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2: {
3651 u64 allowed_options = KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE;
3653 if (cap->args[0] & KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE)
3654 allowed_options = KVM_DIRTY_LOG_MANUAL_CAPS;
3656 if (cap->flags || (cap->args[0] & ~allowed_options))
3658 kvm->manual_dirty_log_protect = cap->args[0];
3662 case KVM_CAP_HALT_POLL: {
3663 if (cap->flags || cap->args[0] != (unsigned int)cap->args[0])
3666 kvm->max_halt_poll_ns = cap->args[0];
3670 return kvm_vm_ioctl_enable_cap(kvm, cap);
3674 static long kvm_vm_ioctl(struct file *filp,
3675 unsigned int ioctl, unsigned long arg)
3677 struct kvm *kvm = filp->private_data;
3678 void __user *argp = (void __user *)arg;
3681 if (kvm->mm != current->mm)
3684 case KVM_CREATE_VCPU:
3685 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
3687 case KVM_ENABLE_CAP: {
3688 struct kvm_enable_cap cap;
3691 if (copy_from_user(&cap, argp, sizeof(cap)))
3693 r = kvm_vm_ioctl_enable_cap_generic(kvm, &cap);
3696 case KVM_SET_USER_MEMORY_REGION: {
3697 struct kvm_userspace_memory_region kvm_userspace_mem;
3700 if (copy_from_user(&kvm_userspace_mem, argp,
3701 sizeof(kvm_userspace_mem)))
3704 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
3707 case KVM_GET_DIRTY_LOG: {
3708 struct kvm_dirty_log log;
3711 if (copy_from_user(&log, argp, sizeof(log)))
3713 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3716 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3717 case KVM_CLEAR_DIRTY_LOG: {
3718 struct kvm_clear_dirty_log log;
3721 if (copy_from_user(&log, argp, sizeof(log)))
3723 r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
3727 #ifdef CONFIG_KVM_MMIO
3728 case KVM_REGISTER_COALESCED_MMIO: {
3729 struct kvm_coalesced_mmio_zone zone;
3732 if (copy_from_user(&zone, argp, sizeof(zone)))
3734 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
3737 case KVM_UNREGISTER_COALESCED_MMIO: {
3738 struct kvm_coalesced_mmio_zone zone;
3741 if (copy_from_user(&zone, argp, sizeof(zone)))
3743 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
3748 struct kvm_irqfd data;
3751 if (copy_from_user(&data, argp, sizeof(data)))
3753 r = kvm_irqfd(kvm, &data);
3756 case KVM_IOEVENTFD: {
3757 struct kvm_ioeventfd data;
3760 if (copy_from_user(&data, argp, sizeof(data)))
3762 r = kvm_ioeventfd(kvm, &data);
3765 #ifdef CONFIG_HAVE_KVM_MSI
3766 case KVM_SIGNAL_MSI: {
3770 if (copy_from_user(&msi, argp, sizeof(msi)))
3772 r = kvm_send_userspace_msi(kvm, &msi);
3776 #ifdef __KVM_HAVE_IRQ_LINE
3777 case KVM_IRQ_LINE_STATUS:
3778 case KVM_IRQ_LINE: {
3779 struct kvm_irq_level irq_event;
3782 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
3785 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
3786 ioctl == KVM_IRQ_LINE_STATUS);
3791 if (ioctl == KVM_IRQ_LINE_STATUS) {
3792 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
3800 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3801 case KVM_SET_GSI_ROUTING: {
3802 struct kvm_irq_routing routing;
3803 struct kvm_irq_routing __user *urouting;
3804 struct kvm_irq_routing_entry *entries = NULL;
3807 if (copy_from_user(&routing, argp, sizeof(routing)))
3810 if (!kvm_arch_can_set_irq_routing(kvm))
3812 if (routing.nr > KVM_MAX_IRQ_ROUTES)
3818 entries = vmemdup_user(urouting->entries,
3819 array_size(sizeof(*entries),
3821 if (IS_ERR(entries)) {
3822 r = PTR_ERR(entries);
3826 r = kvm_set_irq_routing(kvm, entries, routing.nr,
3831 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
3832 case KVM_CREATE_DEVICE: {
3833 struct kvm_create_device cd;
3836 if (copy_from_user(&cd, argp, sizeof(cd)))
3839 r = kvm_ioctl_create_device(kvm, &cd);
3844 if (copy_to_user(argp, &cd, sizeof(cd)))
3850 case KVM_CHECK_EXTENSION:
3851 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
3854 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
3860 #ifdef CONFIG_KVM_COMPAT
3861 struct compat_kvm_dirty_log {
3865 compat_uptr_t dirty_bitmap; /* one bit per page */
3870 static long kvm_vm_compat_ioctl(struct file *filp,
3871 unsigned int ioctl, unsigned long arg)
3873 struct kvm *kvm = filp->private_data;
3876 if (kvm->mm != current->mm)
3879 case KVM_GET_DIRTY_LOG: {
3880 struct compat_kvm_dirty_log compat_log;
3881 struct kvm_dirty_log log;
3883 if (copy_from_user(&compat_log, (void __user *)arg,
3884 sizeof(compat_log)))
3886 log.slot = compat_log.slot;
3887 log.padding1 = compat_log.padding1;
3888 log.padding2 = compat_log.padding2;
3889 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
3891 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3895 r = kvm_vm_ioctl(filp, ioctl, arg);
3901 static struct file_operations kvm_vm_fops = {
3902 .release = kvm_vm_release,
3903 .unlocked_ioctl = kvm_vm_ioctl,
3904 .llseek = noop_llseek,
3905 KVM_COMPAT(kvm_vm_compat_ioctl),
3908 static int kvm_dev_ioctl_create_vm(unsigned long type)
3914 kvm = kvm_create_vm(type);
3916 return PTR_ERR(kvm);
3917 #ifdef CONFIG_KVM_MMIO
3918 r = kvm_coalesced_mmio_init(kvm);
3922 r = get_unused_fd_flags(O_CLOEXEC);
3926 file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
3934 * Don't call kvm_put_kvm anymore at this point; file->f_op is
3935 * already set, with ->release() being kvm_vm_release(). In error
3936 * cases it will be called by the final fput(file) and will take
3937 * care of doing kvm_put_kvm(kvm).
3939 if (kvm_create_vm_debugfs(kvm, r) < 0) {
3944 kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm);
3946 fd_install(r, file);
3954 static long kvm_dev_ioctl(struct file *filp,
3955 unsigned int ioctl, unsigned long arg)
3960 case KVM_GET_API_VERSION:
3963 r = KVM_API_VERSION;
3966 r = kvm_dev_ioctl_create_vm(arg);
3968 case KVM_CHECK_EXTENSION:
3969 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
3971 case KVM_GET_VCPU_MMAP_SIZE:
3974 r = PAGE_SIZE; /* struct kvm_run */
3976 r += PAGE_SIZE; /* pio data page */
3978 #ifdef CONFIG_KVM_MMIO
3979 r += PAGE_SIZE; /* coalesced mmio ring page */
3982 case KVM_TRACE_ENABLE:
3983 case KVM_TRACE_PAUSE:
3984 case KVM_TRACE_DISABLE:
3988 return kvm_arch_dev_ioctl(filp, ioctl, arg);
3994 static struct file_operations kvm_chardev_ops = {
3995 .unlocked_ioctl = kvm_dev_ioctl,
3996 .llseek = noop_llseek,
3997 KVM_COMPAT(kvm_dev_ioctl),
4000 static struct miscdevice kvm_dev = {
4006 static void hardware_enable_nolock(void *junk)
4008 int cpu = raw_smp_processor_id();
4011 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
4014 cpumask_set_cpu(cpu, cpus_hardware_enabled);
4016 r = kvm_arch_hardware_enable();
4019 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
4020 atomic_inc(&hardware_enable_failed);
4021 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
4025 static int kvm_starting_cpu(unsigned int cpu)
4027 raw_spin_lock(&kvm_count_lock);
4028 if (kvm_usage_count)
4029 hardware_enable_nolock(NULL);
4030 raw_spin_unlock(&kvm_count_lock);
4034 static void hardware_disable_nolock(void *junk)
4036 int cpu = raw_smp_processor_id();
4038 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
4040 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
4041 kvm_arch_hardware_disable();
4044 static int kvm_dying_cpu(unsigned int cpu)
4046 raw_spin_lock(&kvm_count_lock);
4047 if (kvm_usage_count)
4048 hardware_disable_nolock(NULL);
4049 raw_spin_unlock(&kvm_count_lock);
4053 static void hardware_disable_all_nolock(void)
4055 BUG_ON(!kvm_usage_count);
4058 if (!kvm_usage_count)
4059 on_each_cpu(hardware_disable_nolock, NULL, 1);
4062 static void hardware_disable_all(void)
4064 raw_spin_lock(&kvm_count_lock);
4065 hardware_disable_all_nolock();
4066 raw_spin_unlock(&kvm_count_lock);
4069 static int hardware_enable_all(void)
4073 raw_spin_lock(&kvm_count_lock);
4076 if (kvm_usage_count == 1) {
4077 atomic_set(&hardware_enable_failed, 0);
4078 on_each_cpu(hardware_enable_nolock, NULL, 1);
4080 if (atomic_read(&hardware_enable_failed)) {
4081 hardware_disable_all_nolock();
4086 raw_spin_unlock(&kvm_count_lock);
4091 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
4095 * Some (well, at least mine) BIOSes hang on reboot if
4098 * And Intel TXT required VMX off for all cpu when system shutdown.
4100 pr_info("kvm: exiting hardware virtualization\n");
4101 kvm_rebooting = true;
4102 on_each_cpu(hardware_disable_nolock, NULL, 1);
4106 static struct notifier_block kvm_reboot_notifier = {
4107 .notifier_call = kvm_reboot,
4111 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
4115 for (i = 0; i < bus->dev_count; i++) {
4116 struct kvm_io_device *pos = bus->range[i].dev;
4118 kvm_iodevice_destructor(pos);
4123 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
4124 const struct kvm_io_range *r2)
4126 gpa_t addr1 = r1->addr;
4127 gpa_t addr2 = r2->addr;
4132 /* If r2->len == 0, match the exact address. If r2->len != 0,
4133 * accept any overlapping write. Any order is acceptable for
4134 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
4135 * we process all of them.
4148 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
4150 return kvm_io_bus_cmp(p1, p2);
4153 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
4154 gpa_t addr, int len)
4156 struct kvm_io_range *range, key;
4159 key = (struct kvm_io_range) {
4164 range = bsearch(&key, bus->range, bus->dev_count,
4165 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
4169 off = range - bus->range;
4171 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
4177 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
4178 struct kvm_io_range *range, const void *val)
4182 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
4186 while (idx < bus->dev_count &&
4187 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
4188 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
4197 /* kvm_io_bus_write - called under kvm->slots_lock */
4198 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
4199 int len, const void *val)
4201 struct kvm_io_bus *bus;
4202 struct kvm_io_range range;
4205 range = (struct kvm_io_range) {
4210 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
4213 r = __kvm_io_bus_write(vcpu, bus, &range, val);
4214 return r < 0 ? r : 0;
4216 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
4218 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
4219 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
4220 gpa_t addr, int len, const void *val, long cookie)
4222 struct kvm_io_bus *bus;
4223 struct kvm_io_range range;
4225 range = (struct kvm_io_range) {
4230 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
4234 /* First try the device referenced by cookie. */
4235 if ((cookie >= 0) && (cookie < bus->dev_count) &&
4236 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
4237 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
4242 * cookie contained garbage; fall back to search and return the
4243 * correct cookie value.
4245 return __kvm_io_bus_write(vcpu, bus, &range, val);
4248 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
4249 struct kvm_io_range *range, void *val)
4253 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
4257 while (idx < bus->dev_count &&
4258 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
4259 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
4268 /* kvm_io_bus_read - called under kvm->slots_lock */
4269 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
4272 struct kvm_io_bus *bus;
4273 struct kvm_io_range range;
4276 range = (struct kvm_io_range) {
4281 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
4284 r = __kvm_io_bus_read(vcpu, bus, &range, val);
4285 return r < 0 ? r : 0;
4288 /* Caller must hold slots_lock. */
4289 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
4290 int len, struct kvm_io_device *dev)
4293 struct kvm_io_bus *new_bus, *bus;
4294 struct kvm_io_range range;
4296 bus = kvm_get_bus(kvm, bus_idx);
4300 /* exclude ioeventfd which is limited by maximum fd */
4301 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
4304 new_bus = kmalloc(struct_size(bus, range, bus->dev_count + 1),
4305 GFP_KERNEL_ACCOUNT);
4309 range = (struct kvm_io_range) {
4315 for (i = 0; i < bus->dev_count; i++)
4316 if (kvm_io_bus_cmp(&bus->range[i], &range) > 0)
4319 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
4320 new_bus->dev_count++;
4321 new_bus->range[i] = range;
4322 memcpy(new_bus->range + i + 1, bus->range + i,
4323 (bus->dev_count - i) * sizeof(struct kvm_io_range));
4324 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
4325 synchronize_srcu_expedited(&kvm->srcu);
4331 /* Caller must hold slots_lock. */
4332 void kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
4333 struct kvm_io_device *dev)
4336 struct kvm_io_bus *new_bus, *bus;
4338 bus = kvm_get_bus(kvm, bus_idx);
4342 for (i = 0; i < bus->dev_count; i++)
4343 if (bus->range[i].dev == dev) {
4347 if (i == bus->dev_count)
4350 new_bus = kmalloc(struct_size(bus, range, bus->dev_count - 1),
4351 GFP_KERNEL_ACCOUNT);
4353 pr_err("kvm: failed to shrink bus, removing it completely\n");
4357 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
4358 new_bus->dev_count--;
4359 memcpy(new_bus->range + i, bus->range + i + 1,
4360 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
4363 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
4364 synchronize_srcu_expedited(&kvm->srcu);
4369 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
4372 struct kvm_io_bus *bus;
4373 int dev_idx, srcu_idx;
4374 struct kvm_io_device *iodev = NULL;
4376 srcu_idx = srcu_read_lock(&kvm->srcu);
4378 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
4382 dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
4386 iodev = bus->range[dev_idx].dev;
4389 srcu_read_unlock(&kvm->srcu, srcu_idx);
4393 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
4395 static int kvm_debugfs_open(struct inode *inode, struct file *file,
4396 int (*get)(void *, u64 *), int (*set)(void *, u64),
4399 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
4402 /* The debugfs files are a reference to the kvm struct which
4403 * is still valid when kvm_destroy_vm is called.
4404 * To avoid the race between open and the removal of the debugfs
4405 * directory we test against the users count.
4407 if (!refcount_inc_not_zero(&stat_data->kvm->users_count))
4410 if (simple_attr_open(inode, file, get,
4411 KVM_DBGFS_GET_MODE(stat_data->dbgfs_item) & 0222
4414 kvm_put_kvm(stat_data->kvm);
4421 static int kvm_debugfs_release(struct inode *inode, struct file *file)
4423 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
4426 simple_attr_release(inode, file);
4427 kvm_put_kvm(stat_data->kvm);
4432 static int kvm_get_stat_per_vm(struct kvm *kvm, size_t offset, u64 *val)
4434 *val = *(ulong *)((void *)kvm + offset);
4439 static int kvm_clear_stat_per_vm(struct kvm *kvm, size_t offset)
4441 *(ulong *)((void *)kvm + offset) = 0;
4446 static int kvm_get_stat_per_vcpu(struct kvm *kvm, size_t offset, u64 *val)
4449 struct kvm_vcpu *vcpu;
4453 kvm_for_each_vcpu(i, vcpu, kvm)
4454 *val += *(u64 *)((void *)vcpu + offset);
4459 static int kvm_clear_stat_per_vcpu(struct kvm *kvm, size_t offset)
4462 struct kvm_vcpu *vcpu;
4464 kvm_for_each_vcpu(i, vcpu, kvm)
4465 *(u64 *)((void *)vcpu + offset) = 0;
4470 static int kvm_stat_data_get(void *data, u64 *val)
4473 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
4475 switch (stat_data->dbgfs_item->kind) {
4477 r = kvm_get_stat_per_vm(stat_data->kvm,
4478 stat_data->dbgfs_item->offset, val);
4481 r = kvm_get_stat_per_vcpu(stat_data->kvm,
4482 stat_data->dbgfs_item->offset, val);
4489 static int kvm_stat_data_clear(void *data, u64 val)
4492 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
4497 switch (stat_data->dbgfs_item->kind) {
4499 r = kvm_clear_stat_per_vm(stat_data->kvm,
4500 stat_data->dbgfs_item->offset);
4503 r = kvm_clear_stat_per_vcpu(stat_data->kvm,
4504 stat_data->dbgfs_item->offset);
4511 static int kvm_stat_data_open(struct inode *inode, struct file *file)
4513 __simple_attr_check_format("%llu\n", 0ull);
4514 return kvm_debugfs_open(inode, file, kvm_stat_data_get,
4515 kvm_stat_data_clear, "%llu\n");
4518 static const struct file_operations stat_fops_per_vm = {
4519 .owner = THIS_MODULE,
4520 .open = kvm_stat_data_open,
4521 .release = kvm_debugfs_release,
4522 .read = simple_attr_read,
4523 .write = simple_attr_write,
4524 .llseek = no_llseek,
4527 static int vm_stat_get(void *_offset, u64 *val)
4529 unsigned offset = (long)_offset;
4534 mutex_lock(&kvm_lock);
4535 list_for_each_entry(kvm, &vm_list, vm_list) {
4536 kvm_get_stat_per_vm(kvm, offset, &tmp_val);
4539 mutex_unlock(&kvm_lock);
4543 static int vm_stat_clear(void *_offset, u64 val)
4545 unsigned offset = (long)_offset;
4551 mutex_lock(&kvm_lock);
4552 list_for_each_entry(kvm, &vm_list, vm_list) {
4553 kvm_clear_stat_per_vm(kvm, offset);
4555 mutex_unlock(&kvm_lock);
4560 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n");
4562 static int vcpu_stat_get(void *_offset, u64 *val)
4564 unsigned offset = (long)_offset;
4569 mutex_lock(&kvm_lock);
4570 list_for_each_entry(kvm, &vm_list, vm_list) {
4571 kvm_get_stat_per_vcpu(kvm, offset, &tmp_val);
4574 mutex_unlock(&kvm_lock);
4578 static int vcpu_stat_clear(void *_offset, u64 val)
4580 unsigned offset = (long)_offset;
4586 mutex_lock(&kvm_lock);
4587 list_for_each_entry(kvm, &vm_list, vm_list) {
4588 kvm_clear_stat_per_vcpu(kvm, offset);
4590 mutex_unlock(&kvm_lock);
4595 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear,
4598 static const struct file_operations *stat_fops[] = {
4599 [KVM_STAT_VCPU] = &vcpu_stat_fops,
4600 [KVM_STAT_VM] = &vm_stat_fops,
4603 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm)
4605 struct kobj_uevent_env *env;
4606 unsigned long long created, active;
4608 if (!kvm_dev.this_device || !kvm)
4611 mutex_lock(&kvm_lock);
4612 if (type == KVM_EVENT_CREATE_VM) {
4613 kvm_createvm_count++;
4615 } else if (type == KVM_EVENT_DESTROY_VM) {
4618 created = kvm_createvm_count;
4619 active = kvm_active_vms;
4620 mutex_unlock(&kvm_lock);
4622 env = kzalloc(sizeof(*env), GFP_KERNEL_ACCOUNT);
4626 add_uevent_var(env, "CREATED=%llu", created);
4627 add_uevent_var(env, "COUNT=%llu", active);
4629 if (type == KVM_EVENT_CREATE_VM) {
4630 add_uevent_var(env, "EVENT=create");
4631 kvm->userspace_pid = task_pid_nr(current);
4632 } else if (type == KVM_EVENT_DESTROY_VM) {
4633 add_uevent_var(env, "EVENT=destroy");
4635 add_uevent_var(env, "PID=%d", kvm->userspace_pid);
4637 if (!IS_ERR_OR_NULL(kvm->debugfs_dentry)) {
4638 char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL_ACCOUNT);
4641 tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX);
4643 add_uevent_var(env, "STATS_PATH=%s", tmp);
4647 /* no need for checks, since we are adding at most only 5 keys */
4648 env->envp[env->envp_idx++] = NULL;
4649 kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp);
4653 static void kvm_init_debug(void)
4655 struct kvm_stats_debugfs_item *p;
4657 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
4659 kvm_debugfs_num_entries = 0;
4660 for (p = debugfs_entries; p->name; ++p, kvm_debugfs_num_entries++) {
4661 debugfs_create_file(p->name, KVM_DBGFS_GET_MODE(p),
4662 kvm_debugfs_dir, (void *)(long)p->offset,
4663 stat_fops[p->kind]);
4667 static int kvm_suspend(void)
4669 if (kvm_usage_count)
4670 hardware_disable_nolock(NULL);
4674 static void kvm_resume(void)
4676 if (kvm_usage_count) {
4677 #ifdef CONFIG_LOCKDEP
4678 WARN_ON(lockdep_is_held(&kvm_count_lock));
4680 hardware_enable_nolock(NULL);
4684 static struct syscore_ops kvm_syscore_ops = {
4685 .suspend = kvm_suspend,
4686 .resume = kvm_resume,
4690 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
4692 return container_of(pn, struct kvm_vcpu, preempt_notifier);
4695 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
4697 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
4699 WRITE_ONCE(vcpu->preempted, false);
4700 WRITE_ONCE(vcpu->ready, false);
4702 __this_cpu_write(kvm_running_vcpu, vcpu);
4703 kvm_arch_sched_in(vcpu, cpu);
4704 kvm_arch_vcpu_load(vcpu, cpu);
4707 static void kvm_sched_out(struct preempt_notifier *pn,
4708 struct task_struct *next)
4710 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
4712 if (current->state == TASK_RUNNING) {
4713 WRITE_ONCE(vcpu->preempted, true);
4714 WRITE_ONCE(vcpu->ready, true);
4716 kvm_arch_vcpu_put(vcpu);
4717 __this_cpu_write(kvm_running_vcpu, NULL);
4721 * kvm_get_running_vcpu - get the vcpu running on the current CPU.
4723 * We can disable preemption locally around accessing the per-CPU variable,
4724 * and use the resolved vcpu pointer after enabling preemption again,
4725 * because even if the current thread is migrated to another CPU, reading
4726 * the per-CPU value later will give us the same value as we update the
4727 * per-CPU variable in the preempt notifier handlers.
4729 struct kvm_vcpu *kvm_get_running_vcpu(void)
4731 struct kvm_vcpu *vcpu;
4734 vcpu = __this_cpu_read(kvm_running_vcpu);
4739 EXPORT_SYMBOL_GPL(kvm_get_running_vcpu);
4742 * kvm_get_running_vcpus - get the per-CPU array of currently running vcpus.
4744 struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void)
4746 return &kvm_running_vcpu;
4749 struct kvm_cpu_compat_check {
4754 static void check_processor_compat(void *data)
4756 struct kvm_cpu_compat_check *c = data;
4758 *c->ret = kvm_arch_check_processor_compat(c->opaque);
4761 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
4762 struct module *module)
4764 struct kvm_cpu_compat_check c;
4768 r = kvm_arch_init(opaque);
4773 * kvm_arch_init makes sure there's at most one caller
4774 * for architectures that support multiple implementations,
4775 * like intel and amd on x86.
4776 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
4777 * conflicts in case kvm is already setup for another implementation.
4779 r = kvm_irqfd_init();
4783 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
4788 r = kvm_arch_hardware_setup(opaque);
4794 for_each_online_cpu(cpu) {
4795 smp_call_function_single(cpu, check_processor_compat, &c, 1);
4800 r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting",
4801 kvm_starting_cpu, kvm_dying_cpu);
4804 register_reboot_notifier(&kvm_reboot_notifier);
4806 /* A kmem cache lets us meet the alignment requirements of fx_save. */
4808 vcpu_align = __alignof__(struct kvm_vcpu);
4810 kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align,
4812 offsetof(struct kvm_vcpu, arch),
4813 sizeof_field(struct kvm_vcpu, arch),
4815 if (!kvm_vcpu_cache) {
4820 r = kvm_async_pf_init();
4824 kvm_chardev_ops.owner = module;
4825 kvm_vm_fops.owner = module;
4826 kvm_vcpu_fops.owner = module;
4828 r = misc_register(&kvm_dev);
4830 pr_err("kvm: misc device register failed\n");
4834 register_syscore_ops(&kvm_syscore_ops);
4836 kvm_preempt_ops.sched_in = kvm_sched_in;
4837 kvm_preempt_ops.sched_out = kvm_sched_out;
4841 r = kvm_vfio_ops_init();
4847 kvm_async_pf_deinit();
4849 kmem_cache_destroy(kvm_vcpu_cache);
4851 unregister_reboot_notifier(&kvm_reboot_notifier);
4852 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4854 kvm_arch_hardware_unsetup();
4856 free_cpumask_var(cpus_hardware_enabled);
4864 EXPORT_SYMBOL_GPL(kvm_init);
4868 debugfs_remove_recursive(kvm_debugfs_dir);
4869 misc_deregister(&kvm_dev);
4870 kmem_cache_destroy(kvm_vcpu_cache);
4871 kvm_async_pf_deinit();
4872 unregister_syscore_ops(&kvm_syscore_ops);
4873 unregister_reboot_notifier(&kvm_reboot_notifier);
4874 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4875 on_each_cpu(hardware_disable_nolock, NULL, 1);
4876 kvm_arch_hardware_unsetup();
4879 free_cpumask_var(cpus_hardware_enabled);
4880 kvm_vfio_ops_exit();
4882 EXPORT_SYMBOL_GPL(kvm_exit);
4884 struct kvm_vm_worker_thread_context {
4886 struct task_struct *parent;
4887 struct completion init_done;
4888 kvm_vm_thread_fn_t thread_fn;
4893 static int kvm_vm_worker_thread(void *context)
4896 * The init_context is allocated on the stack of the parent thread, so
4897 * we have to locally copy anything that is needed beyond initialization
4899 struct kvm_vm_worker_thread_context *init_context = context;
4900 struct kvm *kvm = init_context->kvm;
4901 kvm_vm_thread_fn_t thread_fn = init_context->thread_fn;
4902 uintptr_t data = init_context->data;
4905 err = kthread_park(current);
4906 /* kthread_park(current) is never supposed to return an error */
4911 err = cgroup_attach_task_all(init_context->parent, current);
4913 kvm_err("%s: cgroup_attach_task_all failed with err %d\n",
4918 set_user_nice(current, task_nice(init_context->parent));
4921 init_context->err = err;
4922 complete(&init_context->init_done);
4923 init_context = NULL;
4928 /* Wait to be woken up by the spawner before proceeding. */
4931 if (!kthread_should_stop())
4932 err = thread_fn(kvm, data);
4937 int kvm_vm_create_worker_thread(struct kvm *kvm, kvm_vm_thread_fn_t thread_fn,
4938 uintptr_t data, const char *name,
4939 struct task_struct **thread_ptr)
4941 struct kvm_vm_worker_thread_context init_context = {};
4942 struct task_struct *thread;
4945 init_context.kvm = kvm;
4946 init_context.parent = current;
4947 init_context.thread_fn = thread_fn;
4948 init_context.data = data;
4949 init_completion(&init_context.init_done);
4951 thread = kthread_run(kvm_vm_worker_thread, &init_context,
4952 "%s-%d", name, task_pid_nr(current));
4954 return PTR_ERR(thread);
4956 /* kthread_run is never supposed to return NULL */
4957 WARN_ON(thread == NULL);
4959 wait_for_completion(&init_context.init_done);
4961 if (!init_context.err)
4962 *thread_ptr = thread;
4964 return init_context.err;