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"
64 #define CREATE_TRACE_POINTS
65 #include <trace/events/kvm.h>
67 #include <linux/kvm_dirty_ring.h>
69 /* Worst case buffer size needed for holding an integer. */
70 #define ITOA_MAX_LEN 12
72 MODULE_AUTHOR("Qumranet");
73 MODULE_LICENSE("GPL");
75 /* Architectures should define their poll value according to the halt latency */
76 unsigned int halt_poll_ns = KVM_HALT_POLL_NS_DEFAULT;
77 module_param(halt_poll_ns, uint, 0644);
78 EXPORT_SYMBOL_GPL(halt_poll_ns);
80 /* Default doubles per-vcpu halt_poll_ns. */
81 unsigned int halt_poll_ns_grow = 2;
82 module_param(halt_poll_ns_grow, uint, 0644);
83 EXPORT_SYMBOL_GPL(halt_poll_ns_grow);
85 /* The start value to grow halt_poll_ns from */
86 unsigned int halt_poll_ns_grow_start = 10000; /* 10us */
87 module_param(halt_poll_ns_grow_start, uint, 0644);
88 EXPORT_SYMBOL_GPL(halt_poll_ns_grow_start);
90 /* Default resets per-vcpu halt_poll_ns . */
91 unsigned int halt_poll_ns_shrink;
92 module_param(halt_poll_ns_shrink, uint, 0644);
93 EXPORT_SYMBOL_GPL(halt_poll_ns_shrink);
98 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
101 DEFINE_MUTEX(kvm_lock);
102 static DEFINE_RAW_SPINLOCK(kvm_count_lock);
105 static cpumask_var_t cpus_hardware_enabled;
106 static int kvm_usage_count;
107 static atomic_t hardware_enable_failed;
109 static struct kmem_cache *kvm_vcpu_cache;
111 static __read_mostly struct preempt_ops kvm_preempt_ops;
112 static DEFINE_PER_CPU(struct kvm_vcpu *, kvm_running_vcpu);
114 struct dentry *kvm_debugfs_dir;
115 EXPORT_SYMBOL_GPL(kvm_debugfs_dir);
117 static int kvm_debugfs_num_entries;
118 static const struct file_operations stat_fops_per_vm;
120 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
122 #ifdef CONFIG_KVM_COMPAT
123 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
125 #define KVM_COMPAT(c) .compat_ioctl = (c)
128 * For architectures that don't implement a compat infrastructure,
129 * adopt a double line of defense:
130 * - Prevent a compat task from opening /dev/kvm
131 * - If the open has been done by a 64bit task, and the KVM fd
132 * passed to a compat task, let the ioctls fail.
134 static long kvm_no_compat_ioctl(struct file *file, unsigned int ioctl,
135 unsigned long arg) { return -EINVAL; }
137 static int kvm_no_compat_open(struct inode *inode, struct file *file)
139 return is_compat_task() ? -ENODEV : 0;
141 #define KVM_COMPAT(c) .compat_ioctl = kvm_no_compat_ioctl, \
142 .open = kvm_no_compat_open
144 static int hardware_enable_all(void);
145 static void hardware_disable_all(void);
147 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
149 __visible bool kvm_rebooting;
150 EXPORT_SYMBOL_GPL(kvm_rebooting);
152 #define KVM_EVENT_CREATE_VM 0
153 #define KVM_EVENT_DESTROY_VM 1
154 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm);
155 static unsigned long long kvm_createvm_count;
156 static unsigned long long kvm_active_vms;
158 __weak void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
159 unsigned long start, unsigned long end)
163 bool kvm_is_zone_device_pfn(kvm_pfn_t pfn)
166 * The metadata used by is_zone_device_page() to determine whether or
167 * not a page is ZONE_DEVICE is guaranteed to be valid if and only if
168 * the device has been pinned, e.g. by get_user_pages(). WARN if the
169 * page_count() is zero to help detect bad usage of this helper.
171 if (!pfn_valid(pfn) || WARN_ON_ONCE(!page_count(pfn_to_page(pfn))))
174 return is_zone_device_page(pfn_to_page(pfn));
177 bool kvm_is_reserved_pfn(kvm_pfn_t pfn)
180 * ZONE_DEVICE pages currently set PG_reserved, but from a refcounting
181 * perspective they are "normal" pages, albeit with slightly different
185 return PageReserved(pfn_to_page(pfn)) &&
187 !kvm_is_zone_device_pfn(pfn);
192 bool kvm_is_transparent_hugepage(kvm_pfn_t pfn)
194 struct page *page = pfn_to_page(pfn);
196 if (!PageTransCompoundMap(page))
199 return is_transparent_hugepage(compound_head(page));
203 * Switches to specified vcpu, until a matching vcpu_put()
205 void vcpu_load(struct kvm_vcpu *vcpu)
209 __this_cpu_write(kvm_running_vcpu, vcpu);
210 preempt_notifier_register(&vcpu->preempt_notifier);
211 kvm_arch_vcpu_load(vcpu, cpu);
214 EXPORT_SYMBOL_GPL(vcpu_load);
216 void vcpu_put(struct kvm_vcpu *vcpu)
219 kvm_arch_vcpu_put(vcpu);
220 preempt_notifier_unregister(&vcpu->preempt_notifier);
221 __this_cpu_write(kvm_running_vcpu, NULL);
224 EXPORT_SYMBOL_GPL(vcpu_put);
226 /* TODO: merge with kvm_arch_vcpu_should_kick */
227 static bool kvm_request_needs_ipi(struct kvm_vcpu *vcpu, unsigned req)
229 int mode = kvm_vcpu_exiting_guest_mode(vcpu);
232 * We need to wait for the VCPU to reenable interrupts and get out of
233 * READING_SHADOW_PAGE_TABLES mode.
235 if (req & KVM_REQUEST_WAIT)
236 return mode != OUTSIDE_GUEST_MODE;
239 * Need to kick a running VCPU, but otherwise there is nothing to do.
241 return mode == IN_GUEST_MODE;
244 static void ack_flush(void *_completed)
248 static inline bool kvm_kick_many_cpus(const struct cpumask *cpus, bool wait)
251 cpus = cpu_online_mask;
253 if (cpumask_empty(cpus))
256 smp_call_function_many(cpus, ack_flush, NULL, wait);
260 bool kvm_make_vcpus_request_mask(struct kvm *kvm, unsigned int req,
261 struct kvm_vcpu *except,
262 unsigned long *vcpu_bitmap, cpumask_var_t tmp)
265 struct kvm_vcpu *vcpu;
270 kvm_for_each_vcpu(i, vcpu, kvm) {
271 if ((vcpu_bitmap && !test_bit(i, vcpu_bitmap)) ||
275 kvm_make_request(req, vcpu);
278 if (!(req & KVM_REQUEST_NO_WAKEUP) && kvm_vcpu_wake_up(vcpu))
281 if (tmp != NULL && cpu != -1 && cpu != me &&
282 kvm_request_needs_ipi(vcpu, req))
283 __cpumask_set_cpu(cpu, tmp);
286 called = kvm_kick_many_cpus(tmp, !!(req & KVM_REQUEST_WAIT));
292 bool kvm_make_all_cpus_request_except(struct kvm *kvm, unsigned int req,
293 struct kvm_vcpu *except)
298 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
300 called = kvm_make_vcpus_request_mask(kvm, req, except, NULL, cpus);
302 free_cpumask_var(cpus);
306 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
308 return kvm_make_all_cpus_request_except(kvm, req, NULL);
311 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
312 void kvm_flush_remote_tlbs(struct kvm *kvm)
315 * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in
316 * kvm_make_all_cpus_request.
318 long dirty_count = smp_load_acquire(&kvm->tlbs_dirty);
321 * We want to publish modifications to the page tables before reading
322 * mode. Pairs with a memory barrier in arch-specific code.
323 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
324 * and smp_mb in walk_shadow_page_lockless_begin/end.
325 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
327 * There is already an smp_mb__after_atomic() before
328 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
331 if (!kvm_arch_flush_remote_tlb(kvm)
332 || kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
333 ++kvm->stat.remote_tlb_flush;
334 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
336 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
339 void kvm_reload_remote_mmus(struct kvm *kvm)
341 kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
344 #ifdef KVM_ARCH_NR_OBJS_PER_MEMORY_CACHE
345 static inline void *mmu_memory_cache_alloc_obj(struct kvm_mmu_memory_cache *mc,
348 gfp_flags |= mc->gfp_zero;
351 return kmem_cache_alloc(mc->kmem_cache, gfp_flags);
353 return (void *)__get_free_page(gfp_flags);
356 int kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache *mc, int min)
360 if (mc->nobjs >= min)
362 while (mc->nobjs < ARRAY_SIZE(mc->objects)) {
363 obj = mmu_memory_cache_alloc_obj(mc, GFP_KERNEL_ACCOUNT);
365 return mc->nobjs >= min ? 0 : -ENOMEM;
366 mc->objects[mc->nobjs++] = obj;
371 int kvm_mmu_memory_cache_nr_free_objects(struct kvm_mmu_memory_cache *mc)
376 void kvm_mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc)
380 kmem_cache_free(mc->kmem_cache, mc->objects[--mc->nobjs]);
382 free_page((unsigned long)mc->objects[--mc->nobjs]);
386 void *kvm_mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc)
390 if (WARN_ON(!mc->nobjs))
391 p = mmu_memory_cache_alloc_obj(mc, GFP_ATOMIC | __GFP_ACCOUNT);
393 p = mc->objects[--mc->nobjs];
399 static void kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
401 mutex_init(&vcpu->mutex);
406 rcuwait_init(&vcpu->wait);
407 kvm_async_pf_vcpu_init(vcpu);
410 INIT_LIST_HEAD(&vcpu->blocked_vcpu_list);
412 kvm_vcpu_set_in_spin_loop(vcpu, false);
413 kvm_vcpu_set_dy_eligible(vcpu, false);
414 vcpu->preempted = false;
416 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
419 void kvm_vcpu_destroy(struct kvm_vcpu *vcpu)
421 kvm_dirty_ring_free(&vcpu->dirty_ring);
422 kvm_arch_vcpu_destroy(vcpu);
425 * No need for rcu_read_lock as VCPU_RUN is the only place that changes
426 * the vcpu->pid pointer, and at destruction time all file descriptors
429 put_pid(rcu_dereference_protected(vcpu->pid, 1));
431 free_page((unsigned long)vcpu->run);
432 kmem_cache_free(kvm_vcpu_cache, vcpu);
434 EXPORT_SYMBOL_GPL(kvm_vcpu_destroy);
436 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
437 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
439 return container_of(mn, struct kvm, mmu_notifier);
442 static void kvm_mmu_notifier_invalidate_range(struct mmu_notifier *mn,
443 struct mm_struct *mm,
444 unsigned long start, unsigned long end)
446 struct kvm *kvm = mmu_notifier_to_kvm(mn);
449 idx = srcu_read_lock(&kvm->srcu);
450 kvm_arch_mmu_notifier_invalidate_range(kvm, start, end);
451 srcu_read_unlock(&kvm->srcu, idx);
454 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
455 struct mm_struct *mm,
456 unsigned long address,
459 struct kvm *kvm = mmu_notifier_to_kvm(mn);
462 idx = srcu_read_lock(&kvm->srcu);
466 kvm->mmu_notifier_seq++;
468 if (kvm_set_spte_hva(kvm, address, pte))
469 kvm_flush_remote_tlbs(kvm);
472 srcu_read_unlock(&kvm->srcu, idx);
475 static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
476 const struct mmu_notifier_range *range)
478 struct kvm *kvm = mmu_notifier_to_kvm(mn);
479 int need_tlb_flush = 0, idx;
481 idx = srcu_read_lock(&kvm->srcu);
484 * The count increase must become visible at unlock time as no
485 * spte can be established without taking the mmu_lock and
486 * count is also read inside the mmu_lock critical section.
488 kvm->mmu_notifier_count++;
489 if (likely(kvm->mmu_notifier_count == 1)) {
490 kvm->mmu_notifier_range_start = range->start;
491 kvm->mmu_notifier_range_end = range->end;
494 * Fully tracking multiple concurrent ranges has dimishing
495 * returns. Keep things simple and just find the minimal range
496 * which includes the current and new ranges. As there won't be
497 * enough information to subtract a range after its invalidate
498 * completes, any ranges invalidated concurrently will
499 * accumulate and persist until all outstanding invalidates
502 kvm->mmu_notifier_range_start =
503 min(kvm->mmu_notifier_range_start, range->start);
504 kvm->mmu_notifier_range_end =
505 max(kvm->mmu_notifier_range_end, range->end);
507 need_tlb_flush = kvm_unmap_hva_range(kvm, range->start, range->end,
509 /* we've to flush the tlb before the pages can be freed */
510 if (need_tlb_flush || kvm->tlbs_dirty)
511 kvm_flush_remote_tlbs(kvm);
514 srcu_read_unlock(&kvm->srcu, idx);
519 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
520 const struct mmu_notifier_range *range)
522 struct kvm *kvm = mmu_notifier_to_kvm(mn);
526 * This sequence increase will notify the kvm page fault that
527 * the page that is going to be mapped in the spte could have
530 kvm->mmu_notifier_seq++;
533 * The above sequence increase must be visible before the
534 * below count decrease, which is ensured by the smp_wmb above
535 * in conjunction with the smp_rmb in mmu_notifier_retry().
537 kvm->mmu_notifier_count--;
540 BUG_ON(kvm->mmu_notifier_count < 0);
543 static int kvm_mmu_notifier_clear_flush_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);
554 young = kvm_age_hva(kvm, start, end);
556 kvm_flush_remote_tlbs(kvm);
559 srcu_read_unlock(&kvm->srcu, idx);
564 static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
565 struct mm_struct *mm,
569 struct kvm *kvm = mmu_notifier_to_kvm(mn);
572 idx = srcu_read_lock(&kvm->srcu);
575 * Even though we do not flush TLB, this will still adversely
576 * affect performance on pre-Haswell Intel EPT, where there is
577 * no EPT Access Bit to clear so that we have to tear down EPT
578 * tables instead. If we find this unacceptable, we can always
579 * add a parameter to kvm_age_hva so that it effectively doesn't
580 * do anything on clear_young.
582 * Also note that currently we never issue secondary TLB flushes
583 * from clear_young, leaving this job up to the regular system
584 * cadence. If we find this inaccurate, we might come up with a
585 * more sophisticated heuristic later.
587 young = kvm_age_hva(kvm, start, end);
589 srcu_read_unlock(&kvm->srcu, idx);
594 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
595 struct mm_struct *mm,
596 unsigned long address)
598 struct kvm *kvm = mmu_notifier_to_kvm(mn);
601 idx = srcu_read_lock(&kvm->srcu);
603 young = kvm_test_age_hva(kvm, address);
605 srcu_read_unlock(&kvm->srcu, idx);
610 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
611 struct mm_struct *mm)
613 struct kvm *kvm = mmu_notifier_to_kvm(mn);
616 idx = srcu_read_lock(&kvm->srcu);
617 kvm_arch_flush_shadow_all(kvm);
618 srcu_read_unlock(&kvm->srcu, idx);
621 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
622 .invalidate_range = kvm_mmu_notifier_invalidate_range,
623 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
624 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
625 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
626 .clear_young = kvm_mmu_notifier_clear_young,
627 .test_young = kvm_mmu_notifier_test_young,
628 .change_pte = kvm_mmu_notifier_change_pte,
629 .release = kvm_mmu_notifier_release,
632 static int kvm_init_mmu_notifier(struct kvm *kvm)
634 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
635 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
638 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
640 static int kvm_init_mmu_notifier(struct kvm *kvm)
645 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
647 static struct kvm_memslots *kvm_alloc_memslots(void)
650 struct kvm_memslots *slots;
652 slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL_ACCOUNT);
656 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
657 slots->id_to_index[i] = -1;
662 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
664 if (!memslot->dirty_bitmap)
667 kvfree(memslot->dirty_bitmap);
668 memslot->dirty_bitmap = NULL;
671 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
673 kvm_destroy_dirty_bitmap(slot);
675 kvm_arch_free_memslot(kvm, slot);
681 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
683 struct kvm_memory_slot *memslot;
688 kvm_for_each_memslot(memslot, slots)
689 kvm_free_memslot(kvm, memslot);
694 static void kvm_destroy_vm_debugfs(struct kvm *kvm)
698 if (!kvm->debugfs_dentry)
701 debugfs_remove_recursive(kvm->debugfs_dentry);
703 if (kvm->debugfs_stat_data) {
704 for (i = 0; i < kvm_debugfs_num_entries; i++)
705 kfree(kvm->debugfs_stat_data[i]);
706 kfree(kvm->debugfs_stat_data);
710 static int kvm_create_vm_debugfs(struct kvm *kvm, int fd)
712 char dir_name[ITOA_MAX_LEN * 2];
713 struct kvm_stat_data *stat_data;
714 struct kvm_stats_debugfs_item *p;
716 if (!debugfs_initialized())
719 snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd);
720 kvm->debugfs_dentry = debugfs_create_dir(dir_name, kvm_debugfs_dir);
722 kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries,
723 sizeof(*kvm->debugfs_stat_data),
725 if (!kvm->debugfs_stat_data)
728 for (p = debugfs_entries; p->name; p++) {
729 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT);
733 stat_data->kvm = kvm;
734 stat_data->dbgfs_item = p;
735 kvm->debugfs_stat_data[p - debugfs_entries] = stat_data;
736 debugfs_create_file(p->name, KVM_DBGFS_GET_MODE(p),
737 kvm->debugfs_dentry, stat_data,
744 * Called after the VM is otherwise initialized, but just before adding it to
747 int __weak kvm_arch_post_init_vm(struct kvm *kvm)
753 * Called just after removing the VM from the vm_list, but before doing any
756 void __weak kvm_arch_pre_destroy_vm(struct kvm *kvm)
760 static struct kvm *kvm_create_vm(unsigned long type)
762 struct kvm *kvm = kvm_arch_alloc_vm();
767 return ERR_PTR(-ENOMEM);
769 KVM_MMU_LOCK_INIT(kvm);
771 kvm->mm = current->mm;
772 kvm_eventfd_init(kvm);
773 mutex_init(&kvm->lock);
774 mutex_init(&kvm->irq_lock);
775 mutex_init(&kvm->slots_lock);
776 INIT_LIST_HEAD(&kvm->devices);
778 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
780 if (init_srcu_struct(&kvm->srcu))
781 goto out_err_no_srcu;
782 if (init_srcu_struct(&kvm->irq_srcu))
783 goto out_err_no_irq_srcu;
785 refcount_set(&kvm->users_count, 1);
786 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
787 struct kvm_memslots *slots = kvm_alloc_memslots();
790 goto out_err_no_arch_destroy_vm;
791 /* Generations must be different for each address space. */
792 slots->generation = i;
793 rcu_assign_pointer(kvm->memslots[i], slots);
796 for (i = 0; i < KVM_NR_BUSES; i++) {
797 rcu_assign_pointer(kvm->buses[i],
798 kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL_ACCOUNT));
800 goto out_err_no_arch_destroy_vm;
803 kvm->max_halt_poll_ns = halt_poll_ns;
805 r = kvm_arch_init_vm(kvm, type);
807 goto out_err_no_arch_destroy_vm;
809 r = hardware_enable_all();
811 goto out_err_no_disable;
813 #ifdef CONFIG_HAVE_KVM_IRQFD
814 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
817 r = kvm_init_mmu_notifier(kvm);
819 goto out_err_no_mmu_notifier;
821 r = kvm_arch_post_init_vm(kvm);
825 mutex_lock(&kvm_lock);
826 list_add(&kvm->vm_list, &vm_list);
827 mutex_unlock(&kvm_lock);
829 preempt_notifier_inc();
834 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
835 if (kvm->mmu_notifier.ops)
836 mmu_notifier_unregister(&kvm->mmu_notifier, current->mm);
838 out_err_no_mmu_notifier:
839 hardware_disable_all();
841 kvm_arch_destroy_vm(kvm);
842 out_err_no_arch_destroy_vm:
843 WARN_ON_ONCE(!refcount_dec_and_test(&kvm->users_count));
844 for (i = 0; i < KVM_NR_BUSES; i++)
845 kfree(kvm_get_bus(kvm, i));
846 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
847 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
848 cleanup_srcu_struct(&kvm->irq_srcu);
850 cleanup_srcu_struct(&kvm->srcu);
852 kvm_arch_free_vm(kvm);
857 static void kvm_destroy_devices(struct kvm *kvm)
859 struct kvm_device *dev, *tmp;
862 * We do not need to take the kvm->lock here, because nobody else
863 * has a reference to the struct kvm at this point and therefore
864 * cannot access the devices list anyhow.
866 list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
867 list_del(&dev->vm_node);
868 dev->ops->destroy(dev);
872 static void kvm_destroy_vm(struct kvm *kvm)
875 struct mm_struct *mm = kvm->mm;
877 kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm);
878 kvm_destroy_vm_debugfs(kvm);
879 kvm_arch_sync_events(kvm);
880 mutex_lock(&kvm_lock);
881 list_del(&kvm->vm_list);
882 mutex_unlock(&kvm_lock);
883 kvm_arch_pre_destroy_vm(kvm);
885 kvm_free_irq_routing(kvm);
886 for (i = 0; i < KVM_NR_BUSES; i++) {
887 struct kvm_io_bus *bus = kvm_get_bus(kvm, i);
890 kvm_io_bus_destroy(bus);
891 kvm->buses[i] = NULL;
893 kvm_coalesced_mmio_free(kvm);
894 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
895 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
897 kvm_arch_flush_shadow_all(kvm);
899 kvm_arch_destroy_vm(kvm);
900 kvm_destroy_devices(kvm);
901 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
902 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
903 cleanup_srcu_struct(&kvm->irq_srcu);
904 cleanup_srcu_struct(&kvm->srcu);
905 kvm_arch_free_vm(kvm);
906 preempt_notifier_dec();
907 hardware_disable_all();
911 void kvm_get_kvm(struct kvm *kvm)
913 refcount_inc(&kvm->users_count);
915 EXPORT_SYMBOL_GPL(kvm_get_kvm);
917 void kvm_put_kvm(struct kvm *kvm)
919 if (refcount_dec_and_test(&kvm->users_count))
922 EXPORT_SYMBOL_GPL(kvm_put_kvm);
925 * Used to put a reference that was taken on behalf of an object associated
926 * with a user-visible file descriptor, e.g. a vcpu or device, if installation
927 * of the new file descriptor fails and the reference cannot be transferred to
928 * its final owner. In such cases, the caller is still actively using @kvm and
929 * will fail miserably if the refcount unexpectedly hits zero.
931 void kvm_put_kvm_no_destroy(struct kvm *kvm)
933 WARN_ON(refcount_dec_and_test(&kvm->users_count));
935 EXPORT_SYMBOL_GPL(kvm_put_kvm_no_destroy);
937 static int kvm_vm_release(struct inode *inode, struct file *filp)
939 struct kvm *kvm = filp->private_data;
941 kvm_irqfd_release(kvm);
948 * Allocation size is twice as large as the actual dirty bitmap size.
949 * See kvm_vm_ioctl_get_dirty_log() why this is needed.
951 static int kvm_alloc_dirty_bitmap(struct kvm_memory_slot *memslot)
953 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
955 memslot->dirty_bitmap = kvzalloc(dirty_bytes, GFP_KERNEL_ACCOUNT);
956 if (!memslot->dirty_bitmap)
963 * Delete a memslot by decrementing the number of used slots and shifting all
964 * other entries in the array forward one spot.
966 static inline void kvm_memslot_delete(struct kvm_memslots *slots,
967 struct kvm_memory_slot *memslot)
969 struct kvm_memory_slot *mslots = slots->memslots;
972 if (WARN_ON(slots->id_to_index[memslot->id] == -1))
977 if (atomic_read(&slots->lru_slot) >= slots->used_slots)
978 atomic_set(&slots->lru_slot, 0);
980 for (i = slots->id_to_index[memslot->id]; i < slots->used_slots; i++) {
981 mslots[i] = mslots[i + 1];
982 slots->id_to_index[mslots[i].id] = i;
984 mslots[i] = *memslot;
985 slots->id_to_index[memslot->id] = -1;
989 * "Insert" a new memslot by incrementing the number of used slots. Returns
990 * the new slot's initial index into the memslots array.
992 static inline int kvm_memslot_insert_back(struct kvm_memslots *slots)
994 return slots->used_slots++;
998 * Move a changed memslot backwards in the array by shifting existing slots
999 * with a higher GFN toward the front of the array. Note, the changed memslot
1000 * itself is not preserved in the array, i.e. not swapped at this time, only
1001 * its new index into the array is tracked. Returns the changed memslot's
1002 * current index into the memslots array.
1004 static inline int kvm_memslot_move_backward(struct kvm_memslots *slots,
1005 struct kvm_memory_slot *memslot)
1007 struct kvm_memory_slot *mslots = slots->memslots;
1010 if (WARN_ON_ONCE(slots->id_to_index[memslot->id] == -1) ||
1011 WARN_ON_ONCE(!slots->used_slots))
1015 * Move the target memslot backward in the array by shifting existing
1016 * memslots with a higher GFN (than the target memslot) towards the
1017 * front of the array.
1019 for (i = slots->id_to_index[memslot->id]; i < slots->used_slots - 1; i++) {
1020 if (memslot->base_gfn > mslots[i + 1].base_gfn)
1023 WARN_ON_ONCE(memslot->base_gfn == mslots[i + 1].base_gfn);
1025 /* Shift the next memslot forward one and update its index. */
1026 mslots[i] = mslots[i + 1];
1027 slots->id_to_index[mslots[i].id] = i;
1033 * Move a changed memslot forwards in the array by shifting existing slots with
1034 * a lower GFN toward the back of the array. Note, the changed memslot itself
1035 * is not preserved in the array, i.e. not swapped at this time, only its new
1036 * index into the array is tracked. Returns the changed memslot's final index
1037 * into the memslots array.
1039 static inline int kvm_memslot_move_forward(struct kvm_memslots *slots,
1040 struct kvm_memory_slot *memslot,
1043 struct kvm_memory_slot *mslots = slots->memslots;
1046 for (i = start; i > 0; i--) {
1047 if (memslot->base_gfn < mslots[i - 1].base_gfn)
1050 WARN_ON_ONCE(memslot->base_gfn == mslots[i - 1].base_gfn);
1052 /* Shift the next memslot back one and update its index. */
1053 mslots[i] = mslots[i - 1];
1054 slots->id_to_index[mslots[i].id] = i;
1060 * Re-sort memslots based on their GFN to account for an added, deleted, or
1061 * moved memslot. Sorting memslots by GFN allows using a binary search during
1064 * IMPORTANT: Slots are sorted from highest GFN to lowest GFN! I.e. the entry
1065 * at memslots[0] has the highest GFN.
1067 * The sorting algorithm takes advantage of having initially sorted memslots
1068 * and knowing the position of the changed memslot. Sorting is also optimized
1069 * by not swapping the updated memslot and instead only shifting other memslots
1070 * and tracking the new index for the update memslot. Only once its final
1071 * index is known is the updated memslot copied into its position in the array.
1073 * - When deleting a memslot, the deleted memslot simply needs to be moved to
1074 * the end of the array.
1076 * - When creating a memslot, the algorithm "inserts" the new memslot at the
1077 * end of the array and then it forward to its correct location.
1079 * - When moving a memslot, the algorithm first moves the updated memslot
1080 * backward to handle the scenario where the memslot's GFN was changed to a
1081 * lower value. update_memslots() then falls through and runs the same flow
1082 * as creating a memslot to move the memslot forward to handle the scenario
1083 * where its GFN was changed to a higher value.
1085 * Note, slots are sorted from highest->lowest instead of lowest->highest for
1086 * historical reasons. Originally, invalid memslots where denoted by having
1087 * GFN=0, thus sorting from highest->lowest naturally sorted invalid memslots
1088 * to the end of the array. The current algorithm uses dedicated logic to
1089 * delete a memslot and thus does not rely on invalid memslots having GFN=0.
1091 * The other historical motiviation for highest->lowest was to improve the
1092 * performance of memslot lookup. KVM originally used a linear search starting
1093 * at memslots[0]. On x86, the largest memslot usually has one of the highest,
1094 * if not *the* highest, GFN, as the bulk of the guest's RAM is located in a
1095 * single memslot above the 4gb boundary. As the largest memslot is also the
1096 * most likely to be referenced, sorting it to the front of the array was
1097 * advantageous. The current binary search starts from the middle of the array
1098 * and uses an LRU pointer to improve performance for all memslots and GFNs.
1100 static void update_memslots(struct kvm_memslots *slots,
1101 struct kvm_memory_slot *memslot,
1102 enum kvm_mr_change change)
1106 if (change == KVM_MR_DELETE) {
1107 kvm_memslot_delete(slots, memslot);
1109 if (change == KVM_MR_CREATE)
1110 i = kvm_memslot_insert_back(slots);
1112 i = kvm_memslot_move_backward(slots, memslot);
1113 i = kvm_memslot_move_forward(slots, memslot, i);
1116 * Copy the memslot to its new position in memslots and update
1117 * its index accordingly.
1119 slots->memslots[i] = *memslot;
1120 slots->id_to_index[memslot->id] = i;
1124 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
1126 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
1128 #ifdef __KVM_HAVE_READONLY_MEM
1129 valid_flags |= KVM_MEM_READONLY;
1132 if (mem->flags & ~valid_flags)
1138 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
1139 int as_id, struct kvm_memslots *slots)
1141 struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id);
1142 u64 gen = old_memslots->generation;
1144 WARN_ON(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS);
1145 slots->generation = gen | KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
1147 rcu_assign_pointer(kvm->memslots[as_id], slots);
1148 synchronize_srcu_expedited(&kvm->srcu);
1151 * Increment the new memslot generation a second time, dropping the
1152 * update in-progress flag and incrementing the generation based on
1153 * the number of address spaces. This provides a unique and easily
1154 * identifiable generation number while the memslots are in flux.
1156 gen = slots->generation & ~KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
1159 * Generations must be unique even across address spaces. We do not need
1160 * a global counter for that, instead the generation space is evenly split
1161 * across address spaces. For example, with two address spaces, address
1162 * space 0 will use generations 0, 2, 4, ... while address space 1 will
1163 * use generations 1, 3, 5, ...
1165 gen += KVM_ADDRESS_SPACE_NUM;
1167 kvm_arch_memslots_updated(kvm, gen);
1169 slots->generation = gen;
1171 return old_memslots;
1175 * Note, at a minimum, the current number of used slots must be allocated, even
1176 * when deleting a memslot, as we need a complete duplicate of the memslots for
1177 * use when invalidating a memslot prior to deleting/moving the memslot.
1179 static struct kvm_memslots *kvm_dup_memslots(struct kvm_memslots *old,
1180 enum kvm_mr_change change)
1182 struct kvm_memslots *slots;
1183 size_t old_size, new_size;
1185 old_size = sizeof(struct kvm_memslots) +
1186 (sizeof(struct kvm_memory_slot) * old->used_slots);
1188 if (change == KVM_MR_CREATE)
1189 new_size = old_size + sizeof(struct kvm_memory_slot);
1191 new_size = old_size;
1193 slots = kvzalloc(new_size, GFP_KERNEL_ACCOUNT);
1195 memcpy(slots, old, old_size);
1200 static int kvm_set_memslot(struct kvm *kvm,
1201 const struct kvm_userspace_memory_region *mem,
1202 struct kvm_memory_slot *old,
1203 struct kvm_memory_slot *new, int as_id,
1204 enum kvm_mr_change change)
1206 struct kvm_memory_slot *slot;
1207 struct kvm_memslots *slots;
1210 slots = kvm_dup_memslots(__kvm_memslots(kvm, as_id), change);
1214 if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) {
1216 * Note, the INVALID flag needs to be in the appropriate entry
1217 * in the freshly allocated memslots, not in @old or @new.
1219 slot = id_to_memslot(slots, old->id);
1220 slot->flags |= KVM_MEMSLOT_INVALID;
1223 * We can re-use the old memslots, the only difference from the
1224 * newly installed memslots is the invalid flag, which will get
1225 * dropped by update_memslots anyway. We'll also revert to the
1226 * old memslots if preparing the new memory region fails.
1228 slots = install_new_memslots(kvm, as_id, slots);
1230 /* From this point no new shadow pages pointing to a deleted,
1231 * or moved, memslot will be created.
1233 * validation of sp->gfn happens in:
1234 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1235 * - kvm_is_visible_gfn (mmu_check_root)
1237 kvm_arch_flush_shadow_memslot(kvm, slot);
1240 r = kvm_arch_prepare_memory_region(kvm, new, mem, change);
1244 update_memslots(slots, new, change);
1245 slots = install_new_memslots(kvm, as_id, slots);
1247 kvm_arch_commit_memory_region(kvm, mem, old, new, change);
1253 if (change == KVM_MR_DELETE || change == KVM_MR_MOVE)
1254 slots = install_new_memslots(kvm, as_id, slots);
1259 static int kvm_delete_memslot(struct kvm *kvm,
1260 const struct kvm_userspace_memory_region *mem,
1261 struct kvm_memory_slot *old, int as_id)
1263 struct kvm_memory_slot new;
1269 memset(&new, 0, sizeof(new));
1272 * This is only for debugging purpose; it should never be referenced
1273 * for a removed memslot.
1277 r = kvm_set_memslot(kvm, mem, old, &new, as_id, KVM_MR_DELETE);
1281 kvm_free_memslot(kvm, old);
1286 * Allocate some memory and give it an address in the guest physical address
1289 * Discontiguous memory is allowed, mostly for framebuffers.
1291 * Must be called holding kvm->slots_lock for write.
1293 int __kvm_set_memory_region(struct kvm *kvm,
1294 const struct kvm_userspace_memory_region *mem)
1296 struct kvm_memory_slot old, new;
1297 struct kvm_memory_slot *tmp;
1298 enum kvm_mr_change change;
1302 r = check_memory_region_flags(mem);
1306 as_id = mem->slot >> 16;
1307 id = (u16)mem->slot;
1309 /* General sanity checks */
1310 if (mem->memory_size & (PAGE_SIZE - 1))
1312 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
1314 /* We can read the guest memory with __xxx_user() later on. */
1315 if ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
1316 (mem->userspace_addr != untagged_addr(mem->userspace_addr)) ||
1317 !access_ok((void __user *)(unsigned long)mem->userspace_addr,
1320 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
1322 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
1326 * Make a full copy of the old memslot, the pointer will become stale
1327 * when the memslots are re-sorted by update_memslots(), and the old
1328 * memslot needs to be referenced after calling update_memslots(), e.g.
1329 * to free its resources and for arch specific behavior.
1331 tmp = id_to_memslot(__kvm_memslots(kvm, as_id), id);
1336 memset(&old, 0, sizeof(old));
1340 if (!mem->memory_size)
1341 return kvm_delete_memslot(kvm, mem, &old, as_id);
1345 new.base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
1346 new.npages = mem->memory_size >> PAGE_SHIFT;
1347 new.flags = mem->flags;
1348 new.userspace_addr = mem->userspace_addr;
1350 if (new.npages > KVM_MEM_MAX_NR_PAGES)
1354 change = KVM_MR_CREATE;
1355 new.dirty_bitmap = NULL;
1356 memset(&new.arch, 0, sizeof(new.arch));
1357 } else { /* Modify an existing slot. */
1358 if ((new.userspace_addr != old.userspace_addr) ||
1359 (new.npages != old.npages) ||
1360 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
1363 if (new.base_gfn != old.base_gfn)
1364 change = KVM_MR_MOVE;
1365 else if (new.flags != old.flags)
1366 change = KVM_MR_FLAGS_ONLY;
1367 else /* Nothing to change. */
1370 /* Copy dirty_bitmap and arch from the current memslot. */
1371 new.dirty_bitmap = old.dirty_bitmap;
1372 memcpy(&new.arch, &old.arch, sizeof(new.arch));
1375 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
1376 /* Check for overlaps */
1377 kvm_for_each_memslot(tmp, __kvm_memslots(kvm, as_id)) {
1380 if (!((new.base_gfn + new.npages <= tmp->base_gfn) ||
1381 (new.base_gfn >= tmp->base_gfn + tmp->npages)))
1386 /* Allocate/free page dirty bitmap as needed */
1387 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
1388 new.dirty_bitmap = NULL;
1389 else if (!new.dirty_bitmap && !kvm->dirty_ring_size) {
1390 r = kvm_alloc_dirty_bitmap(&new);
1394 if (kvm_dirty_log_manual_protect_and_init_set(kvm))
1395 bitmap_set(new.dirty_bitmap, 0, new.npages);
1398 r = kvm_set_memslot(kvm, mem, &old, &new, as_id, change);
1402 if (old.dirty_bitmap && !new.dirty_bitmap)
1403 kvm_destroy_dirty_bitmap(&old);
1407 if (new.dirty_bitmap && !old.dirty_bitmap)
1408 kvm_destroy_dirty_bitmap(&new);
1411 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1413 int kvm_set_memory_region(struct kvm *kvm,
1414 const struct kvm_userspace_memory_region *mem)
1418 mutex_lock(&kvm->slots_lock);
1419 r = __kvm_set_memory_region(kvm, mem);
1420 mutex_unlock(&kvm->slots_lock);
1423 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1425 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1426 struct kvm_userspace_memory_region *mem)
1428 if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1431 return kvm_set_memory_region(kvm, mem);
1434 #ifndef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1436 * kvm_get_dirty_log - get a snapshot of dirty pages
1437 * @kvm: pointer to kvm instance
1438 * @log: slot id and address to which we copy the log
1439 * @is_dirty: set to '1' if any dirty pages were found
1440 * @memslot: set to the associated memslot, always valid on success
1442 int kvm_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log,
1443 int *is_dirty, struct kvm_memory_slot **memslot)
1445 struct kvm_memslots *slots;
1448 unsigned long any = 0;
1450 /* Dirty ring tracking is exclusive to dirty log tracking */
1451 if (kvm->dirty_ring_size)
1457 as_id = log->slot >> 16;
1458 id = (u16)log->slot;
1459 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1462 slots = __kvm_memslots(kvm, as_id);
1463 *memslot = id_to_memslot(slots, id);
1464 if (!(*memslot) || !(*memslot)->dirty_bitmap)
1467 kvm_arch_sync_dirty_log(kvm, *memslot);
1469 n = kvm_dirty_bitmap_bytes(*memslot);
1471 for (i = 0; !any && i < n/sizeof(long); ++i)
1472 any = (*memslot)->dirty_bitmap[i];
1474 if (copy_to_user(log->dirty_bitmap, (*memslot)->dirty_bitmap, n))
1481 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1483 #else /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
1485 * kvm_get_dirty_log_protect - get a snapshot of dirty pages
1486 * and reenable dirty page tracking for the corresponding pages.
1487 * @kvm: pointer to kvm instance
1488 * @log: slot id and address to which we copy the log
1490 * We need to keep it in mind that VCPU threads can write to the bitmap
1491 * concurrently. So, to avoid losing track of dirty pages we keep the
1494 * 1. Take a snapshot of the bit and clear it if needed.
1495 * 2. Write protect the corresponding page.
1496 * 3. Copy the snapshot to the userspace.
1497 * 4. Upon return caller flushes TLB's if needed.
1499 * Between 2 and 4, the guest may write to the page using the remaining TLB
1500 * entry. This is not a problem because the page is reported dirty using
1501 * the snapshot taken before and step 4 ensures that writes done after
1502 * exiting to userspace will be logged for the next call.
1505 static int kvm_get_dirty_log_protect(struct kvm *kvm, struct kvm_dirty_log *log)
1507 struct kvm_memslots *slots;
1508 struct kvm_memory_slot *memslot;
1511 unsigned long *dirty_bitmap;
1512 unsigned long *dirty_bitmap_buffer;
1515 /* Dirty ring tracking is exclusive to dirty log tracking */
1516 if (kvm->dirty_ring_size)
1519 as_id = log->slot >> 16;
1520 id = (u16)log->slot;
1521 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1524 slots = __kvm_memslots(kvm, as_id);
1525 memslot = id_to_memslot(slots, id);
1526 if (!memslot || !memslot->dirty_bitmap)
1529 dirty_bitmap = memslot->dirty_bitmap;
1531 kvm_arch_sync_dirty_log(kvm, memslot);
1533 n = kvm_dirty_bitmap_bytes(memslot);
1535 if (kvm->manual_dirty_log_protect) {
1537 * Unlike kvm_get_dirty_log, we always return false in *flush,
1538 * because no flush is needed until KVM_CLEAR_DIRTY_LOG. There
1539 * is some code duplication between this function and
1540 * kvm_get_dirty_log, but hopefully all architecture
1541 * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log
1542 * can be eliminated.
1544 dirty_bitmap_buffer = dirty_bitmap;
1546 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
1547 memset(dirty_bitmap_buffer, 0, n);
1550 for (i = 0; i < n / sizeof(long); i++) {
1554 if (!dirty_bitmap[i])
1558 mask = xchg(&dirty_bitmap[i], 0);
1559 dirty_bitmap_buffer[i] = mask;
1561 offset = i * BITS_PER_LONG;
1562 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1565 KVM_MMU_UNLOCK(kvm);
1569 kvm_arch_flush_remote_tlbs_memslot(kvm, memslot);
1571 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1578 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
1579 * @kvm: kvm instance
1580 * @log: slot id and address to which we copy the log
1582 * Steps 1-4 below provide general overview of dirty page logging. See
1583 * kvm_get_dirty_log_protect() function description for additional details.
1585 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
1586 * always flush the TLB (step 4) even if previous step failed and the dirty
1587 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
1588 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
1589 * writes will be marked dirty for next log read.
1591 * 1. Take a snapshot of the bit and clear it if needed.
1592 * 2. Write protect the corresponding page.
1593 * 3. Copy the snapshot to the userspace.
1594 * 4. Flush TLB's if needed.
1596 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1597 struct kvm_dirty_log *log)
1601 mutex_lock(&kvm->slots_lock);
1603 r = kvm_get_dirty_log_protect(kvm, log);
1605 mutex_unlock(&kvm->slots_lock);
1610 * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap
1611 * and reenable dirty page tracking for the corresponding pages.
1612 * @kvm: pointer to kvm instance
1613 * @log: slot id and address from which to fetch the bitmap of dirty pages
1615 static int kvm_clear_dirty_log_protect(struct kvm *kvm,
1616 struct kvm_clear_dirty_log *log)
1618 struct kvm_memslots *slots;
1619 struct kvm_memory_slot *memslot;
1623 unsigned long *dirty_bitmap;
1624 unsigned long *dirty_bitmap_buffer;
1627 /* Dirty ring tracking is exclusive to dirty log tracking */
1628 if (kvm->dirty_ring_size)
1631 as_id = log->slot >> 16;
1632 id = (u16)log->slot;
1633 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1636 if (log->first_page & 63)
1639 slots = __kvm_memslots(kvm, as_id);
1640 memslot = id_to_memslot(slots, id);
1641 if (!memslot || !memslot->dirty_bitmap)
1644 dirty_bitmap = memslot->dirty_bitmap;
1646 n = ALIGN(log->num_pages, BITS_PER_LONG) / 8;
1648 if (log->first_page > memslot->npages ||
1649 log->num_pages > memslot->npages - log->first_page ||
1650 (log->num_pages < memslot->npages - log->first_page && (log->num_pages & 63)))
1653 kvm_arch_sync_dirty_log(kvm, memslot);
1656 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
1657 if (copy_from_user(dirty_bitmap_buffer, log->dirty_bitmap, n))
1661 for (offset = log->first_page, i = offset / BITS_PER_LONG,
1662 n = DIV_ROUND_UP(log->num_pages, BITS_PER_LONG); n--;
1663 i++, offset += BITS_PER_LONG) {
1664 unsigned long mask = *dirty_bitmap_buffer++;
1665 atomic_long_t *p = (atomic_long_t *) &dirty_bitmap[i];
1669 mask &= atomic_long_fetch_andnot(mask, p);
1672 * mask contains the bits that really have been cleared. This
1673 * never includes any bits beyond the length of the memslot (if
1674 * the length is not aligned to 64 pages), therefore it is not
1675 * a problem if userspace sets them in log->dirty_bitmap.
1679 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1683 KVM_MMU_UNLOCK(kvm);
1686 kvm_arch_flush_remote_tlbs_memslot(kvm, memslot);
1691 static int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm,
1692 struct kvm_clear_dirty_log *log)
1696 mutex_lock(&kvm->slots_lock);
1698 r = kvm_clear_dirty_log_protect(kvm, log);
1700 mutex_unlock(&kvm->slots_lock);
1703 #endif /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
1705 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1707 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1709 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1711 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
1713 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn);
1715 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_memslot);
1717 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1719 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1721 return kvm_is_visible_memslot(memslot);
1723 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1725 bool kvm_vcpu_is_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
1727 struct kvm_memory_slot *memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1729 return kvm_is_visible_memslot(memslot);
1731 EXPORT_SYMBOL_GPL(kvm_vcpu_is_visible_gfn);
1733 unsigned long kvm_host_page_size(struct kvm_vcpu *vcpu, gfn_t gfn)
1735 struct vm_area_struct *vma;
1736 unsigned long addr, size;
1740 addr = kvm_vcpu_gfn_to_hva_prot(vcpu, gfn, NULL);
1741 if (kvm_is_error_hva(addr))
1744 mmap_read_lock(current->mm);
1745 vma = find_vma(current->mm, addr);
1749 size = vma_kernel_pagesize(vma);
1752 mmap_read_unlock(current->mm);
1757 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1759 return slot->flags & KVM_MEM_READONLY;
1762 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1763 gfn_t *nr_pages, bool write)
1765 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1766 return KVM_HVA_ERR_BAD;
1768 if (memslot_is_readonly(slot) && write)
1769 return KVM_HVA_ERR_RO_BAD;
1772 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1774 return __gfn_to_hva_memslot(slot, gfn);
1777 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1780 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1783 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1786 return gfn_to_hva_many(slot, gfn, NULL);
1788 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1790 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1792 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1794 EXPORT_SYMBOL_GPL(gfn_to_hva);
1796 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
1798 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
1800 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
1803 * Return the hva of a @gfn and the R/W attribute if possible.
1805 * @slot: the kvm_memory_slot which contains @gfn
1806 * @gfn: the gfn to be translated
1807 * @writable: used to return the read/write attribute of the @slot if the hva
1808 * is valid and @writable is not NULL
1810 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1811 gfn_t gfn, bool *writable)
1813 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1815 if (!kvm_is_error_hva(hva) && writable)
1816 *writable = !memslot_is_readonly(slot);
1821 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1823 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1825 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1828 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
1830 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1832 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1835 static inline int check_user_page_hwpoison(unsigned long addr)
1837 int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
1839 rc = get_user_pages(addr, 1, flags, NULL, NULL);
1840 return rc == -EHWPOISON;
1844 * The fast path to get the writable pfn which will be stored in @pfn,
1845 * true indicates success, otherwise false is returned. It's also the
1846 * only part that runs if we can in atomic context.
1848 static bool hva_to_pfn_fast(unsigned long addr, bool write_fault,
1849 bool *writable, kvm_pfn_t *pfn)
1851 struct page *page[1];
1854 * Fast pin a writable pfn only if it is a write fault request
1855 * or the caller allows to map a writable pfn for a read fault
1858 if (!(write_fault || writable))
1861 if (get_user_page_fast_only(addr, FOLL_WRITE, page)) {
1862 *pfn = page_to_pfn(page[0]);
1873 * The slow path to get the pfn of the specified host virtual address,
1874 * 1 indicates success, -errno is returned if error is detected.
1876 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1877 bool *writable, kvm_pfn_t *pfn)
1879 unsigned int flags = FOLL_HWPOISON;
1886 *writable = write_fault;
1889 flags |= FOLL_WRITE;
1891 flags |= FOLL_NOWAIT;
1893 npages = get_user_pages_unlocked(addr, 1, &page, flags);
1897 /* map read fault as writable if possible */
1898 if (unlikely(!write_fault) && writable) {
1901 if (get_user_page_fast_only(addr, FOLL_WRITE, &wpage)) {
1907 *pfn = page_to_pfn(page);
1911 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1913 if (unlikely(!(vma->vm_flags & VM_READ)))
1916 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1922 static int hva_to_pfn_remapped(struct vm_area_struct *vma,
1923 unsigned long addr, bool *async,
1924 bool write_fault, bool *writable,
1932 r = follow_pte(vma->vm_mm, addr, &ptep, &ptl);
1935 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
1936 * not call the fault handler, so do it here.
1938 bool unlocked = false;
1939 r = fixup_user_fault(current->mm, addr,
1940 (write_fault ? FAULT_FLAG_WRITE : 0),
1947 r = follow_pte(vma->vm_mm, addr, &ptep, &ptl);
1952 if (write_fault && !pte_write(*ptep)) {
1953 pfn = KVM_PFN_ERR_RO_FAULT;
1958 *writable = pte_write(*ptep);
1959 pfn = pte_pfn(*ptep);
1962 * Get a reference here because callers of *hva_to_pfn* and
1963 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
1964 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
1965 * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will
1966 * simply do nothing for reserved pfns.
1968 * Whoever called remap_pfn_range is also going to call e.g.
1969 * unmap_mapping_range before the underlying pages are freed,
1970 * causing a call to our MMU notifier.
1975 pte_unmap_unlock(ptep, ptl);
1981 * Pin guest page in memory and return its pfn.
1982 * @addr: host virtual address which maps memory to the guest
1983 * @atomic: whether this function can sleep
1984 * @async: whether this function need to wait IO complete if the
1985 * host page is not in the memory
1986 * @write_fault: whether we should get a writable host page
1987 * @writable: whether it allows to map a writable host page for !@write_fault
1989 * The function will map a writable host page for these two cases:
1990 * 1): @write_fault = true
1991 * 2): @write_fault = false && @writable, @writable will tell the caller
1992 * whether the mapping is writable.
1994 static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1995 bool write_fault, bool *writable)
1997 struct vm_area_struct *vma;
2001 /* we can do it either atomically or asynchronously, not both */
2002 BUG_ON(atomic && async);
2004 if (hva_to_pfn_fast(addr, write_fault, writable, &pfn))
2008 return KVM_PFN_ERR_FAULT;
2010 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
2014 mmap_read_lock(current->mm);
2015 if (npages == -EHWPOISON ||
2016 (!async && check_user_page_hwpoison(addr))) {
2017 pfn = KVM_PFN_ERR_HWPOISON;
2022 vma = find_vma_intersection(current->mm, addr, addr + 1);
2025 pfn = KVM_PFN_ERR_FAULT;
2026 else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
2027 r = hva_to_pfn_remapped(vma, addr, async, write_fault, writable, &pfn);
2031 pfn = KVM_PFN_ERR_FAULT;
2033 if (async && vma_is_valid(vma, write_fault))
2035 pfn = KVM_PFN_ERR_FAULT;
2038 mmap_read_unlock(current->mm);
2042 kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn,
2043 bool atomic, bool *async, bool write_fault,
2044 bool *writable, hva_t *hva)
2046 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
2051 if (addr == KVM_HVA_ERR_RO_BAD) {
2054 return KVM_PFN_ERR_RO_FAULT;
2057 if (kvm_is_error_hva(addr)) {
2060 return KVM_PFN_NOSLOT;
2063 /* Do not map writable pfn in the readonly memslot. */
2064 if (writable && memslot_is_readonly(slot)) {
2069 return hva_to_pfn(addr, atomic, async, write_fault,
2072 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
2074 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
2077 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
2078 write_fault, writable, NULL);
2080 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
2082 kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
2084 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL, NULL);
2086 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
2088 kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
2090 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL, NULL);
2092 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
2094 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
2096 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
2098 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
2100 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
2102 return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
2104 EXPORT_SYMBOL_GPL(gfn_to_pfn);
2106 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
2108 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
2110 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
2112 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
2113 struct page **pages, int nr_pages)
2118 addr = gfn_to_hva_many(slot, gfn, &entry);
2119 if (kvm_is_error_hva(addr))
2122 if (entry < nr_pages)
2125 return get_user_pages_fast_only(addr, nr_pages, FOLL_WRITE, pages);
2127 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
2129 static struct page *kvm_pfn_to_page(kvm_pfn_t pfn)
2131 if (is_error_noslot_pfn(pfn))
2132 return KVM_ERR_PTR_BAD_PAGE;
2134 if (kvm_is_reserved_pfn(pfn)) {
2136 return KVM_ERR_PTR_BAD_PAGE;
2139 return pfn_to_page(pfn);
2142 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
2146 pfn = gfn_to_pfn(kvm, gfn);
2148 return kvm_pfn_to_page(pfn);
2150 EXPORT_SYMBOL_GPL(gfn_to_page);
2152 void kvm_release_pfn(kvm_pfn_t pfn, bool dirty, struct gfn_to_pfn_cache *cache)
2158 cache->pfn = cache->gfn = 0;
2161 kvm_release_pfn_dirty(pfn);
2163 kvm_release_pfn_clean(pfn);
2166 static void kvm_cache_gfn_to_pfn(struct kvm_memory_slot *slot, gfn_t gfn,
2167 struct gfn_to_pfn_cache *cache, u64 gen)
2169 kvm_release_pfn(cache->pfn, cache->dirty, cache);
2171 cache->pfn = gfn_to_pfn_memslot(slot, gfn);
2173 cache->dirty = false;
2174 cache->generation = gen;
2177 static int __kvm_map_gfn(struct kvm_memslots *slots, gfn_t gfn,
2178 struct kvm_host_map *map,
2179 struct gfn_to_pfn_cache *cache,
2184 struct page *page = KVM_UNMAPPED_PAGE;
2185 struct kvm_memory_slot *slot = __gfn_to_memslot(slots, gfn);
2186 u64 gen = slots->generation;
2192 if (!cache->pfn || cache->gfn != gfn ||
2193 cache->generation != gen) {
2196 kvm_cache_gfn_to_pfn(slot, gfn, cache, gen);
2202 pfn = gfn_to_pfn_memslot(slot, gfn);
2204 if (is_error_noslot_pfn(pfn))
2207 if (pfn_valid(pfn)) {
2208 page = pfn_to_page(pfn);
2210 hva = kmap_atomic(page);
2213 #ifdef CONFIG_HAS_IOMEM
2214 } else if (!atomic) {
2215 hva = memremap(pfn_to_hpa(pfn), PAGE_SIZE, MEMREMAP_WB);
2232 int kvm_map_gfn(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map,
2233 struct gfn_to_pfn_cache *cache, bool atomic)
2235 return __kvm_map_gfn(kvm_memslots(vcpu->kvm), gfn, map,
2238 EXPORT_SYMBOL_GPL(kvm_map_gfn);
2240 int kvm_vcpu_map(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map)
2242 return __kvm_map_gfn(kvm_vcpu_memslots(vcpu), gfn, map,
2245 EXPORT_SYMBOL_GPL(kvm_vcpu_map);
2247 static void __kvm_unmap_gfn(struct kvm *kvm,
2248 struct kvm_memory_slot *memslot,
2249 struct kvm_host_map *map,
2250 struct gfn_to_pfn_cache *cache,
2251 bool dirty, bool atomic)
2259 if (map->page != KVM_UNMAPPED_PAGE) {
2261 kunmap_atomic(map->hva);
2265 #ifdef CONFIG_HAS_IOMEM
2269 WARN_ONCE(1, "Unexpected unmapping in atomic context");
2273 mark_page_dirty_in_slot(kvm, memslot, map->gfn);
2276 cache->dirty |= dirty;
2278 kvm_release_pfn(map->pfn, dirty, NULL);
2284 int kvm_unmap_gfn(struct kvm_vcpu *vcpu, struct kvm_host_map *map,
2285 struct gfn_to_pfn_cache *cache, bool dirty, bool atomic)
2287 __kvm_unmap_gfn(vcpu->kvm, gfn_to_memslot(vcpu->kvm, map->gfn), map,
2288 cache, dirty, atomic);
2291 EXPORT_SYMBOL_GPL(kvm_unmap_gfn);
2293 void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map, bool dirty)
2295 __kvm_unmap_gfn(vcpu->kvm, kvm_vcpu_gfn_to_memslot(vcpu, map->gfn),
2296 map, NULL, dirty, false);
2298 EXPORT_SYMBOL_GPL(kvm_vcpu_unmap);
2300 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
2304 pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
2306 return kvm_pfn_to_page(pfn);
2308 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
2310 void kvm_release_page_clean(struct page *page)
2312 WARN_ON(is_error_page(page));
2314 kvm_release_pfn_clean(page_to_pfn(page));
2316 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
2318 void kvm_release_pfn_clean(kvm_pfn_t pfn)
2320 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
2321 put_page(pfn_to_page(pfn));
2323 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
2325 void kvm_release_page_dirty(struct page *page)
2327 WARN_ON(is_error_page(page));
2329 kvm_release_pfn_dirty(page_to_pfn(page));
2331 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
2333 void kvm_release_pfn_dirty(kvm_pfn_t pfn)
2335 kvm_set_pfn_dirty(pfn);
2336 kvm_release_pfn_clean(pfn);
2338 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
2340 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
2342 if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn))
2343 SetPageDirty(pfn_to_page(pfn));
2345 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
2347 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
2349 if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn))
2350 mark_page_accessed(pfn_to_page(pfn));
2352 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
2354 void kvm_get_pfn(kvm_pfn_t pfn)
2356 if (!kvm_is_reserved_pfn(pfn))
2357 get_page(pfn_to_page(pfn));
2359 EXPORT_SYMBOL_GPL(kvm_get_pfn);
2361 static int next_segment(unsigned long len, int offset)
2363 if (len > PAGE_SIZE - offset)
2364 return PAGE_SIZE - offset;
2369 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
2370 void *data, int offset, int len)
2375 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
2376 if (kvm_is_error_hva(addr))
2378 r = __copy_from_user(data, (void __user *)addr + offset, len);
2384 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
2387 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2389 return __kvm_read_guest_page(slot, gfn, data, offset, len);
2391 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
2393 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
2394 int offset, int len)
2396 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2398 return __kvm_read_guest_page(slot, gfn, data, offset, len);
2400 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
2402 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
2404 gfn_t gfn = gpa >> PAGE_SHIFT;
2406 int offset = offset_in_page(gpa);
2409 while ((seg = next_segment(len, offset)) != 0) {
2410 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
2420 EXPORT_SYMBOL_GPL(kvm_read_guest);
2422 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
2424 gfn_t gfn = gpa >> PAGE_SHIFT;
2426 int offset = offset_in_page(gpa);
2429 while ((seg = next_segment(len, offset)) != 0) {
2430 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
2440 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
2442 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
2443 void *data, int offset, unsigned long len)
2448 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
2449 if (kvm_is_error_hva(addr))
2451 pagefault_disable();
2452 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
2459 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
2460 void *data, unsigned long len)
2462 gfn_t gfn = gpa >> PAGE_SHIFT;
2463 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2464 int offset = offset_in_page(gpa);
2466 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
2468 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
2470 static int __kvm_write_guest_page(struct kvm *kvm,
2471 struct kvm_memory_slot *memslot, gfn_t gfn,
2472 const void *data, int offset, int len)
2477 addr = gfn_to_hva_memslot(memslot, gfn);
2478 if (kvm_is_error_hva(addr))
2480 r = __copy_to_user((void __user *)addr + offset, data, len);
2483 mark_page_dirty_in_slot(kvm, memslot, gfn);
2487 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
2488 const void *data, int offset, int len)
2490 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2492 return __kvm_write_guest_page(kvm, slot, gfn, data, offset, len);
2494 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
2496 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
2497 const void *data, int offset, int len)
2499 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2501 return __kvm_write_guest_page(vcpu->kvm, slot, gfn, data, offset, len);
2503 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
2505 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
2508 gfn_t gfn = gpa >> PAGE_SHIFT;
2510 int offset = offset_in_page(gpa);
2513 while ((seg = next_segment(len, offset)) != 0) {
2514 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
2524 EXPORT_SYMBOL_GPL(kvm_write_guest);
2526 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
2529 gfn_t gfn = gpa >> PAGE_SHIFT;
2531 int offset = offset_in_page(gpa);
2534 while ((seg = next_segment(len, offset)) != 0) {
2535 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
2545 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
2547 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
2548 struct gfn_to_hva_cache *ghc,
2549 gpa_t gpa, unsigned long len)
2551 int offset = offset_in_page(gpa);
2552 gfn_t start_gfn = gpa >> PAGE_SHIFT;
2553 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
2554 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
2555 gfn_t nr_pages_avail;
2557 /* Update ghc->generation before performing any error checks. */
2558 ghc->generation = slots->generation;
2560 if (start_gfn > end_gfn) {
2561 ghc->hva = KVM_HVA_ERR_BAD;
2566 * If the requested region crosses two memslots, we still
2567 * verify that the entire region is valid here.
2569 for ( ; start_gfn <= end_gfn; start_gfn += nr_pages_avail) {
2570 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
2571 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
2573 if (kvm_is_error_hva(ghc->hva))
2577 /* Use the slow path for cross page reads and writes. */
2578 if (nr_pages_needed == 1)
2581 ghc->memslot = NULL;
2588 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2589 gpa_t gpa, unsigned long len)
2591 struct kvm_memslots *slots = kvm_memslots(kvm);
2592 return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len);
2594 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
2596 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2597 void *data, unsigned int offset,
2600 struct kvm_memslots *slots = kvm_memslots(kvm);
2602 gpa_t gpa = ghc->gpa + offset;
2604 BUG_ON(len + offset > ghc->len);
2606 if (slots->generation != ghc->generation) {
2607 if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len))
2611 if (kvm_is_error_hva(ghc->hva))
2614 if (unlikely(!ghc->memslot))
2615 return kvm_write_guest(kvm, gpa, data, len);
2617 r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
2620 mark_page_dirty_in_slot(kvm, ghc->memslot, gpa >> PAGE_SHIFT);
2624 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached);
2626 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2627 void *data, unsigned long len)
2629 return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len);
2631 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
2633 int kvm_read_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2634 void *data, unsigned int offset,
2637 struct kvm_memslots *slots = kvm_memslots(kvm);
2639 gpa_t gpa = ghc->gpa + offset;
2641 BUG_ON(len + offset > ghc->len);
2643 if (slots->generation != ghc->generation) {
2644 if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len))
2648 if (kvm_is_error_hva(ghc->hva))
2651 if (unlikely(!ghc->memslot))
2652 return kvm_read_guest(kvm, gpa, data, len);
2654 r = __copy_from_user(data, (void __user *)ghc->hva + offset, len);
2660 EXPORT_SYMBOL_GPL(kvm_read_guest_offset_cached);
2662 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2663 void *data, unsigned long len)
2665 return kvm_read_guest_offset_cached(kvm, ghc, data, 0, len);
2667 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
2669 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
2671 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
2672 gfn_t gfn = gpa >> PAGE_SHIFT;
2674 int offset = offset_in_page(gpa);
2677 while ((seg = next_segment(len, offset)) != 0) {
2678 ret = kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
2687 EXPORT_SYMBOL_GPL(kvm_clear_guest);
2689 void mark_page_dirty_in_slot(struct kvm *kvm,
2690 struct kvm_memory_slot *memslot,
2693 if (memslot && kvm_slot_dirty_track_enabled(memslot)) {
2694 unsigned long rel_gfn = gfn - memslot->base_gfn;
2695 u32 slot = (memslot->as_id << 16) | memslot->id;
2697 if (kvm->dirty_ring_size)
2698 kvm_dirty_ring_push(kvm_dirty_ring_get(kvm),
2701 set_bit_le(rel_gfn, memslot->dirty_bitmap);
2704 EXPORT_SYMBOL_GPL(mark_page_dirty_in_slot);
2706 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
2708 struct kvm_memory_slot *memslot;
2710 memslot = gfn_to_memslot(kvm, gfn);
2711 mark_page_dirty_in_slot(kvm, memslot, gfn);
2713 EXPORT_SYMBOL_GPL(mark_page_dirty);
2715 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
2717 struct kvm_memory_slot *memslot;
2719 memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2720 mark_page_dirty_in_slot(vcpu->kvm, memslot, gfn);
2722 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
2724 void kvm_sigset_activate(struct kvm_vcpu *vcpu)
2726 if (!vcpu->sigset_active)
2730 * This does a lockless modification of ->real_blocked, which is fine
2731 * because, only current can change ->real_blocked and all readers of
2732 * ->real_blocked don't care as long ->real_blocked is always a subset
2735 sigprocmask(SIG_SETMASK, &vcpu->sigset, ¤t->real_blocked);
2738 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu)
2740 if (!vcpu->sigset_active)
2743 sigprocmask(SIG_SETMASK, ¤t->real_blocked, NULL);
2744 sigemptyset(¤t->real_blocked);
2747 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
2749 unsigned int old, val, grow, grow_start;
2751 old = val = vcpu->halt_poll_ns;
2752 grow_start = READ_ONCE(halt_poll_ns_grow_start);
2753 grow = READ_ONCE(halt_poll_ns_grow);
2758 if (val < grow_start)
2761 if (val > halt_poll_ns)
2764 vcpu->halt_poll_ns = val;
2766 trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
2769 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
2771 unsigned int old, val, shrink;
2773 old = val = vcpu->halt_poll_ns;
2774 shrink = READ_ONCE(halt_poll_ns_shrink);
2780 vcpu->halt_poll_ns = val;
2781 trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
2784 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
2787 int idx = srcu_read_lock(&vcpu->kvm->srcu);
2789 if (kvm_arch_vcpu_runnable(vcpu)) {
2790 kvm_make_request(KVM_REQ_UNHALT, vcpu);
2793 if (kvm_cpu_has_pending_timer(vcpu))
2795 if (signal_pending(current))
2800 srcu_read_unlock(&vcpu->kvm->srcu, idx);
2805 update_halt_poll_stats(struct kvm_vcpu *vcpu, u64 poll_ns, bool waited)
2808 vcpu->stat.halt_poll_fail_ns += poll_ns;
2810 vcpu->stat.halt_poll_success_ns += poll_ns;
2814 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2816 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
2818 ktime_t start, cur, poll_end;
2819 bool waited = false;
2822 kvm_arch_vcpu_blocking(vcpu);
2824 start = cur = poll_end = ktime_get();
2825 if (vcpu->halt_poll_ns && !kvm_arch_no_poll(vcpu)) {
2826 ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns);
2828 ++vcpu->stat.halt_attempted_poll;
2831 * This sets KVM_REQ_UNHALT if an interrupt
2834 if (kvm_vcpu_check_block(vcpu) < 0) {
2835 ++vcpu->stat.halt_successful_poll;
2836 if (!vcpu_valid_wakeup(vcpu))
2837 ++vcpu->stat.halt_poll_invalid;
2840 poll_end = cur = ktime_get();
2841 } while (single_task_running() && ktime_before(cur, stop));
2844 prepare_to_rcuwait(&vcpu->wait);
2846 set_current_state(TASK_INTERRUPTIBLE);
2848 if (kvm_vcpu_check_block(vcpu) < 0)
2854 finish_rcuwait(&vcpu->wait);
2857 kvm_arch_vcpu_unblocking(vcpu);
2858 block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
2860 update_halt_poll_stats(
2861 vcpu, ktime_to_ns(ktime_sub(poll_end, start)), waited);
2863 if (!kvm_arch_no_poll(vcpu)) {
2864 if (!vcpu_valid_wakeup(vcpu)) {
2865 shrink_halt_poll_ns(vcpu);
2866 } else if (vcpu->kvm->max_halt_poll_ns) {
2867 if (block_ns <= vcpu->halt_poll_ns)
2869 /* we had a long block, shrink polling */
2870 else if (vcpu->halt_poll_ns &&
2871 block_ns > vcpu->kvm->max_halt_poll_ns)
2872 shrink_halt_poll_ns(vcpu);
2873 /* we had a short halt and our poll time is too small */
2874 else if (vcpu->halt_poll_ns < vcpu->kvm->max_halt_poll_ns &&
2875 block_ns < vcpu->kvm->max_halt_poll_ns)
2876 grow_halt_poll_ns(vcpu);
2878 vcpu->halt_poll_ns = 0;
2882 trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu));
2883 kvm_arch_vcpu_block_finish(vcpu);
2885 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
2887 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
2889 struct rcuwait *waitp;
2891 waitp = kvm_arch_vcpu_get_wait(vcpu);
2892 if (rcuwait_wake_up(waitp)) {
2893 WRITE_ONCE(vcpu->ready, true);
2894 ++vcpu->stat.halt_wakeup;
2900 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
2904 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2906 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
2909 int cpu = vcpu->cpu;
2911 if (kvm_vcpu_wake_up(vcpu))
2915 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
2916 if (kvm_arch_vcpu_should_kick(vcpu))
2917 smp_send_reschedule(cpu);
2920 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
2921 #endif /* !CONFIG_S390 */
2923 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
2926 struct task_struct *task = NULL;
2930 pid = rcu_dereference(target->pid);
2932 task = get_pid_task(pid, PIDTYPE_PID);
2936 ret = yield_to(task, 1);
2937 put_task_struct(task);
2941 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
2944 * Helper that checks whether a VCPU is eligible for directed yield.
2945 * Most eligible candidate to yield is decided by following heuristics:
2947 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2948 * (preempted lock holder), indicated by @in_spin_loop.
2949 * Set at the beginning and cleared at the end of interception/PLE handler.
2951 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2952 * chance last time (mostly it has become eligible now since we have probably
2953 * yielded to lockholder in last iteration. This is done by toggling
2954 * @dy_eligible each time a VCPU checked for eligibility.)
2956 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2957 * to preempted lock-holder could result in wrong VCPU selection and CPU
2958 * burning. Giving priority for a potential lock-holder increases lock
2961 * Since algorithm is based on heuristics, accessing another VCPU data without
2962 * locking does not harm. It may result in trying to yield to same VCPU, fail
2963 * and continue with next VCPU and so on.
2965 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
2967 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2970 eligible = !vcpu->spin_loop.in_spin_loop ||
2971 vcpu->spin_loop.dy_eligible;
2973 if (vcpu->spin_loop.in_spin_loop)
2974 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
2983 * Unlike kvm_arch_vcpu_runnable, this function is called outside
2984 * a vcpu_load/vcpu_put pair. However, for most architectures
2985 * kvm_arch_vcpu_runnable does not require vcpu_load.
2987 bool __weak kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
2989 return kvm_arch_vcpu_runnable(vcpu);
2992 static bool vcpu_dy_runnable(struct kvm_vcpu *vcpu)
2994 if (kvm_arch_dy_runnable(vcpu))
2997 #ifdef CONFIG_KVM_ASYNC_PF
2998 if (!list_empty_careful(&vcpu->async_pf.done))
3005 void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode)
3007 struct kvm *kvm = me->kvm;
3008 struct kvm_vcpu *vcpu;
3009 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
3015 kvm_vcpu_set_in_spin_loop(me, true);
3017 * We boost the priority of a VCPU that is runnable but not
3018 * currently running, because it got preempted by something
3019 * else and called schedule in __vcpu_run. Hopefully that
3020 * VCPU is holding the lock that we need and will release it.
3021 * We approximate round-robin by starting at the last boosted VCPU.
3023 for (pass = 0; pass < 2 && !yielded && try; pass++) {
3024 kvm_for_each_vcpu(i, vcpu, kvm) {
3025 if (!pass && i <= last_boosted_vcpu) {
3026 i = last_boosted_vcpu;
3028 } else if (pass && i > last_boosted_vcpu)
3030 if (!READ_ONCE(vcpu->ready))
3034 if (rcuwait_active(&vcpu->wait) &&
3035 !vcpu_dy_runnable(vcpu))
3037 if (READ_ONCE(vcpu->preempted) && yield_to_kernel_mode &&
3038 !kvm_arch_vcpu_in_kernel(vcpu))
3040 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
3043 yielded = kvm_vcpu_yield_to(vcpu);
3045 kvm->last_boosted_vcpu = i;
3047 } else if (yielded < 0) {
3054 kvm_vcpu_set_in_spin_loop(me, false);
3056 /* Ensure vcpu is not eligible during next spinloop */
3057 kvm_vcpu_set_dy_eligible(me, false);
3059 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
3061 static bool kvm_page_in_dirty_ring(struct kvm *kvm, unsigned long pgoff)
3063 #if KVM_DIRTY_LOG_PAGE_OFFSET > 0
3064 return (pgoff >= KVM_DIRTY_LOG_PAGE_OFFSET) &&
3065 (pgoff < KVM_DIRTY_LOG_PAGE_OFFSET +
3066 kvm->dirty_ring_size / PAGE_SIZE);
3072 static vm_fault_t kvm_vcpu_fault(struct vm_fault *vmf)
3074 struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data;
3077 if (vmf->pgoff == 0)
3078 page = virt_to_page(vcpu->run);
3080 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
3081 page = virt_to_page(vcpu->arch.pio_data);
3083 #ifdef CONFIG_KVM_MMIO
3084 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
3085 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
3087 else if (kvm_page_in_dirty_ring(vcpu->kvm, vmf->pgoff))
3088 page = kvm_dirty_ring_get_page(
3090 vmf->pgoff - KVM_DIRTY_LOG_PAGE_OFFSET);
3092 return kvm_arch_vcpu_fault(vcpu, vmf);
3098 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
3099 .fault = kvm_vcpu_fault,
3102 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
3104 struct kvm_vcpu *vcpu = file->private_data;
3105 unsigned long pages = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
3107 if ((kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff) ||
3108 kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff + pages - 1)) &&
3109 ((vma->vm_flags & VM_EXEC) || !(vma->vm_flags & VM_SHARED)))
3112 vma->vm_ops = &kvm_vcpu_vm_ops;
3116 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
3118 struct kvm_vcpu *vcpu = filp->private_data;
3120 kvm_put_kvm(vcpu->kvm);
3124 static struct file_operations kvm_vcpu_fops = {
3125 .release = kvm_vcpu_release,
3126 .unlocked_ioctl = kvm_vcpu_ioctl,
3127 .mmap = kvm_vcpu_mmap,
3128 .llseek = noop_llseek,
3129 KVM_COMPAT(kvm_vcpu_compat_ioctl),
3133 * Allocates an inode for the vcpu.
3135 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
3137 char name[8 + 1 + ITOA_MAX_LEN + 1];
3139 snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id);
3140 return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
3143 static void kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
3145 #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS
3146 struct dentry *debugfs_dentry;
3147 char dir_name[ITOA_MAX_LEN * 2];
3149 if (!debugfs_initialized())
3152 snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
3153 debugfs_dentry = debugfs_create_dir(dir_name,
3154 vcpu->kvm->debugfs_dentry);
3156 kvm_arch_create_vcpu_debugfs(vcpu, debugfs_dentry);
3161 * Creates some virtual cpus. Good luck creating more than one.
3163 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
3166 struct kvm_vcpu *vcpu;
3169 if (id >= KVM_MAX_VCPU_ID)
3172 mutex_lock(&kvm->lock);
3173 if (kvm->created_vcpus == KVM_MAX_VCPUS) {
3174 mutex_unlock(&kvm->lock);
3178 kvm->created_vcpus++;
3179 mutex_unlock(&kvm->lock);
3181 r = kvm_arch_vcpu_precreate(kvm, id);
3183 goto vcpu_decrement;
3185 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
3188 goto vcpu_decrement;
3191 BUILD_BUG_ON(sizeof(struct kvm_run) > PAGE_SIZE);
3192 page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
3197 vcpu->run = page_address(page);
3199 kvm_vcpu_init(vcpu, kvm, id);
3201 r = kvm_arch_vcpu_create(vcpu);
3203 goto vcpu_free_run_page;
3205 if (kvm->dirty_ring_size) {
3206 r = kvm_dirty_ring_alloc(&vcpu->dirty_ring,
3207 id, kvm->dirty_ring_size);
3209 goto arch_vcpu_destroy;
3212 mutex_lock(&kvm->lock);
3213 if (kvm_get_vcpu_by_id(kvm, id)) {
3215 goto unlock_vcpu_destroy;
3218 vcpu->vcpu_idx = atomic_read(&kvm->online_vcpus);
3219 BUG_ON(kvm->vcpus[vcpu->vcpu_idx]);
3221 /* Now it's all set up, let userspace reach it */
3223 r = create_vcpu_fd(vcpu);
3225 kvm_put_kvm_no_destroy(kvm);
3226 goto unlock_vcpu_destroy;
3229 kvm->vcpus[vcpu->vcpu_idx] = vcpu;
3232 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
3233 * before kvm->online_vcpu's incremented value.
3236 atomic_inc(&kvm->online_vcpus);
3238 mutex_unlock(&kvm->lock);
3239 kvm_arch_vcpu_postcreate(vcpu);
3240 kvm_create_vcpu_debugfs(vcpu);
3243 unlock_vcpu_destroy:
3244 mutex_unlock(&kvm->lock);
3245 kvm_dirty_ring_free(&vcpu->dirty_ring);
3247 kvm_arch_vcpu_destroy(vcpu);
3249 free_page((unsigned long)vcpu->run);
3251 kmem_cache_free(kvm_vcpu_cache, vcpu);
3253 mutex_lock(&kvm->lock);
3254 kvm->created_vcpus--;
3255 mutex_unlock(&kvm->lock);
3259 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
3262 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
3263 vcpu->sigset_active = 1;
3264 vcpu->sigset = *sigset;
3266 vcpu->sigset_active = 0;
3270 static long kvm_vcpu_ioctl(struct file *filp,
3271 unsigned int ioctl, unsigned long arg)
3273 struct kvm_vcpu *vcpu = filp->private_data;
3274 void __user *argp = (void __user *)arg;
3276 struct kvm_fpu *fpu = NULL;
3277 struct kvm_sregs *kvm_sregs = NULL;
3279 if (vcpu->kvm->mm != current->mm)
3282 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
3286 * Some architectures have vcpu ioctls that are asynchronous to vcpu
3287 * execution; mutex_lock() would break them.
3289 r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg);
3290 if (r != -ENOIOCTLCMD)
3293 if (mutex_lock_killable(&vcpu->mutex))
3301 oldpid = rcu_access_pointer(vcpu->pid);
3302 if (unlikely(oldpid != task_pid(current))) {
3303 /* The thread running this VCPU changed. */
3306 r = kvm_arch_vcpu_run_pid_change(vcpu);
3310 newpid = get_task_pid(current, PIDTYPE_PID);
3311 rcu_assign_pointer(vcpu->pid, newpid);
3316 r = kvm_arch_vcpu_ioctl_run(vcpu);
3317 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
3320 case KVM_GET_REGS: {
3321 struct kvm_regs *kvm_regs;
3324 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL_ACCOUNT);
3327 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
3331 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
3338 case KVM_SET_REGS: {
3339 struct kvm_regs *kvm_regs;
3341 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
3342 if (IS_ERR(kvm_regs)) {
3343 r = PTR_ERR(kvm_regs);
3346 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
3350 case KVM_GET_SREGS: {
3351 kvm_sregs = kzalloc(sizeof(struct kvm_sregs),
3352 GFP_KERNEL_ACCOUNT);
3356 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
3360 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
3365 case KVM_SET_SREGS: {
3366 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
3367 if (IS_ERR(kvm_sregs)) {
3368 r = PTR_ERR(kvm_sregs);
3372 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
3375 case KVM_GET_MP_STATE: {
3376 struct kvm_mp_state mp_state;
3378 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
3382 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
3387 case KVM_SET_MP_STATE: {
3388 struct kvm_mp_state mp_state;
3391 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
3393 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
3396 case KVM_TRANSLATE: {
3397 struct kvm_translation tr;
3400 if (copy_from_user(&tr, argp, sizeof(tr)))
3402 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
3406 if (copy_to_user(argp, &tr, sizeof(tr)))
3411 case KVM_SET_GUEST_DEBUG: {
3412 struct kvm_guest_debug dbg;
3415 if (copy_from_user(&dbg, argp, sizeof(dbg)))
3417 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
3420 case KVM_SET_SIGNAL_MASK: {
3421 struct kvm_signal_mask __user *sigmask_arg = argp;
3422 struct kvm_signal_mask kvm_sigmask;
3423 sigset_t sigset, *p;
3428 if (copy_from_user(&kvm_sigmask, argp,
3429 sizeof(kvm_sigmask)))
3432 if (kvm_sigmask.len != sizeof(sigset))
3435 if (copy_from_user(&sigset, sigmask_arg->sigset,
3440 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
3444 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL_ACCOUNT);
3448 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
3452 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
3458 fpu = memdup_user(argp, sizeof(*fpu));
3464 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
3468 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
3471 mutex_unlock(&vcpu->mutex);
3477 #ifdef CONFIG_KVM_COMPAT
3478 static long kvm_vcpu_compat_ioctl(struct file *filp,
3479 unsigned int ioctl, unsigned long arg)
3481 struct kvm_vcpu *vcpu = filp->private_data;
3482 void __user *argp = compat_ptr(arg);
3485 if (vcpu->kvm->mm != current->mm)
3489 case KVM_SET_SIGNAL_MASK: {
3490 struct kvm_signal_mask __user *sigmask_arg = argp;
3491 struct kvm_signal_mask kvm_sigmask;
3496 if (copy_from_user(&kvm_sigmask, argp,
3497 sizeof(kvm_sigmask)))
3500 if (kvm_sigmask.len != sizeof(compat_sigset_t))
3503 if (get_compat_sigset(&sigset,
3504 (compat_sigset_t __user *)sigmask_arg->sigset))
3506 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
3508 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
3512 r = kvm_vcpu_ioctl(filp, ioctl, arg);
3520 static int kvm_device_mmap(struct file *filp, struct vm_area_struct *vma)
3522 struct kvm_device *dev = filp->private_data;
3525 return dev->ops->mmap(dev, vma);
3530 static int kvm_device_ioctl_attr(struct kvm_device *dev,
3531 int (*accessor)(struct kvm_device *dev,
3532 struct kvm_device_attr *attr),
3535 struct kvm_device_attr attr;
3540 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
3543 return accessor(dev, &attr);
3546 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
3549 struct kvm_device *dev = filp->private_data;
3551 if (dev->kvm->mm != current->mm)
3555 case KVM_SET_DEVICE_ATTR:
3556 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
3557 case KVM_GET_DEVICE_ATTR:
3558 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
3559 case KVM_HAS_DEVICE_ATTR:
3560 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
3562 if (dev->ops->ioctl)
3563 return dev->ops->ioctl(dev, ioctl, arg);
3569 static int kvm_device_release(struct inode *inode, struct file *filp)
3571 struct kvm_device *dev = filp->private_data;
3572 struct kvm *kvm = dev->kvm;
3574 if (dev->ops->release) {
3575 mutex_lock(&kvm->lock);
3576 list_del(&dev->vm_node);
3577 dev->ops->release(dev);
3578 mutex_unlock(&kvm->lock);
3585 static const struct file_operations kvm_device_fops = {
3586 .unlocked_ioctl = kvm_device_ioctl,
3587 .release = kvm_device_release,
3588 KVM_COMPAT(kvm_device_ioctl),
3589 .mmap = kvm_device_mmap,
3592 struct kvm_device *kvm_device_from_filp(struct file *filp)
3594 if (filp->f_op != &kvm_device_fops)
3597 return filp->private_data;
3600 static const struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
3601 #ifdef CONFIG_KVM_MPIC
3602 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
3603 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
3607 int kvm_register_device_ops(const struct kvm_device_ops *ops, u32 type)
3609 if (type >= ARRAY_SIZE(kvm_device_ops_table))
3612 if (kvm_device_ops_table[type] != NULL)
3615 kvm_device_ops_table[type] = ops;
3619 void kvm_unregister_device_ops(u32 type)
3621 if (kvm_device_ops_table[type] != NULL)
3622 kvm_device_ops_table[type] = NULL;
3625 static int kvm_ioctl_create_device(struct kvm *kvm,
3626 struct kvm_create_device *cd)
3628 const struct kvm_device_ops *ops = NULL;
3629 struct kvm_device *dev;
3630 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
3634 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
3637 type = array_index_nospec(cd->type, ARRAY_SIZE(kvm_device_ops_table));
3638 ops = kvm_device_ops_table[type];
3645 dev = kzalloc(sizeof(*dev), GFP_KERNEL_ACCOUNT);
3652 mutex_lock(&kvm->lock);
3653 ret = ops->create(dev, type);
3655 mutex_unlock(&kvm->lock);
3659 list_add(&dev->vm_node, &kvm->devices);
3660 mutex_unlock(&kvm->lock);
3666 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
3668 kvm_put_kvm_no_destroy(kvm);
3669 mutex_lock(&kvm->lock);
3670 list_del(&dev->vm_node);
3671 mutex_unlock(&kvm->lock);
3680 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
3683 case KVM_CAP_USER_MEMORY:
3684 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
3685 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
3686 case KVM_CAP_INTERNAL_ERROR_DATA:
3687 #ifdef CONFIG_HAVE_KVM_MSI
3688 case KVM_CAP_SIGNAL_MSI:
3690 #ifdef CONFIG_HAVE_KVM_IRQFD
3692 case KVM_CAP_IRQFD_RESAMPLE:
3694 case KVM_CAP_IOEVENTFD_ANY_LENGTH:
3695 case KVM_CAP_CHECK_EXTENSION_VM:
3696 case KVM_CAP_ENABLE_CAP_VM:
3697 case KVM_CAP_HALT_POLL:
3699 #ifdef CONFIG_KVM_MMIO
3700 case KVM_CAP_COALESCED_MMIO:
3701 return KVM_COALESCED_MMIO_PAGE_OFFSET;
3702 case KVM_CAP_COALESCED_PIO:
3705 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3706 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2:
3707 return KVM_DIRTY_LOG_MANUAL_CAPS;
3709 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3710 case KVM_CAP_IRQ_ROUTING:
3711 return KVM_MAX_IRQ_ROUTES;
3713 #if KVM_ADDRESS_SPACE_NUM > 1
3714 case KVM_CAP_MULTI_ADDRESS_SPACE:
3715 return KVM_ADDRESS_SPACE_NUM;
3717 case KVM_CAP_NR_MEMSLOTS:
3718 return KVM_USER_MEM_SLOTS;
3719 case KVM_CAP_DIRTY_LOG_RING:
3720 #if KVM_DIRTY_LOG_PAGE_OFFSET > 0
3721 return KVM_DIRTY_RING_MAX_ENTRIES * sizeof(struct kvm_dirty_gfn);
3728 return kvm_vm_ioctl_check_extension(kvm, arg);
3731 static int kvm_vm_ioctl_enable_dirty_log_ring(struct kvm *kvm, u32 size)
3735 if (!KVM_DIRTY_LOG_PAGE_OFFSET)
3738 /* the size should be power of 2 */
3739 if (!size || (size & (size - 1)))
3742 /* Should be bigger to keep the reserved entries, or a page */
3743 if (size < kvm_dirty_ring_get_rsvd_entries() *
3744 sizeof(struct kvm_dirty_gfn) || size < PAGE_SIZE)
3747 if (size > KVM_DIRTY_RING_MAX_ENTRIES *
3748 sizeof(struct kvm_dirty_gfn))
3751 /* We only allow it to set once */
3752 if (kvm->dirty_ring_size)
3755 mutex_lock(&kvm->lock);
3757 if (kvm->created_vcpus) {
3758 /* We don't allow to change this value after vcpu created */
3761 kvm->dirty_ring_size = size;
3765 mutex_unlock(&kvm->lock);
3769 static int kvm_vm_ioctl_reset_dirty_pages(struct kvm *kvm)
3772 struct kvm_vcpu *vcpu;
3775 if (!kvm->dirty_ring_size)
3778 mutex_lock(&kvm->slots_lock);
3780 kvm_for_each_vcpu(i, vcpu, kvm)
3781 cleared += kvm_dirty_ring_reset(vcpu->kvm, &vcpu->dirty_ring);
3783 mutex_unlock(&kvm->slots_lock);
3786 kvm_flush_remote_tlbs(kvm);
3791 int __attribute__((weak)) kvm_vm_ioctl_enable_cap(struct kvm *kvm,
3792 struct kvm_enable_cap *cap)
3797 static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm,
3798 struct kvm_enable_cap *cap)
3801 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3802 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2: {
3803 u64 allowed_options = KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE;
3805 if (cap->args[0] & KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE)
3806 allowed_options = KVM_DIRTY_LOG_MANUAL_CAPS;
3808 if (cap->flags || (cap->args[0] & ~allowed_options))
3810 kvm->manual_dirty_log_protect = cap->args[0];
3814 case KVM_CAP_HALT_POLL: {
3815 if (cap->flags || cap->args[0] != (unsigned int)cap->args[0])
3818 kvm->max_halt_poll_ns = cap->args[0];
3821 case KVM_CAP_DIRTY_LOG_RING:
3822 return kvm_vm_ioctl_enable_dirty_log_ring(kvm, cap->args[0]);
3824 return kvm_vm_ioctl_enable_cap(kvm, cap);
3828 static long kvm_vm_ioctl(struct file *filp,
3829 unsigned int ioctl, unsigned long arg)
3831 struct kvm *kvm = filp->private_data;
3832 void __user *argp = (void __user *)arg;
3835 if (kvm->mm != current->mm)
3838 case KVM_CREATE_VCPU:
3839 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
3841 case KVM_ENABLE_CAP: {
3842 struct kvm_enable_cap cap;
3845 if (copy_from_user(&cap, argp, sizeof(cap)))
3847 r = kvm_vm_ioctl_enable_cap_generic(kvm, &cap);
3850 case KVM_SET_USER_MEMORY_REGION: {
3851 struct kvm_userspace_memory_region kvm_userspace_mem;
3854 if (copy_from_user(&kvm_userspace_mem, argp,
3855 sizeof(kvm_userspace_mem)))
3858 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
3861 case KVM_GET_DIRTY_LOG: {
3862 struct kvm_dirty_log log;
3865 if (copy_from_user(&log, argp, sizeof(log)))
3867 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3870 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3871 case KVM_CLEAR_DIRTY_LOG: {
3872 struct kvm_clear_dirty_log log;
3875 if (copy_from_user(&log, argp, sizeof(log)))
3877 r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
3881 #ifdef CONFIG_KVM_MMIO
3882 case KVM_REGISTER_COALESCED_MMIO: {
3883 struct kvm_coalesced_mmio_zone zone;
3886 if (copy_from_user(&zone, argp, sizeof(zone)))
3888 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
3891 case KVM_UNREGISTER_COALESCED_MMIO: {
3892 struct kvm_coalesced_mmio_zone zone;
3895 if (copy_from_user(&zone, argp, sizeof(zone)))
3897 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
3902 struct kvm_irqfd data;
3905 if (copy_from_user(&data, argp, sizeof(data)))
3907 r = kvm_irqfd(kvm, &data);
3910 case KVM_IOEVENTFD: {
3911 struct kvm_ioeventfd data;
3914 if (copy_from_user(&data, argp, sizeof(data)))
3916 r = kvm_ioeventfd(kvm, &data);
3919 #ifdef CONFIG_HAVE_KVM_MSI
3920 case KVM_SIGNAL_MSI: {
3924 if (copy_from_user(&msi, argp, sizeof(msi)))
3926 r = kvm_send_userspace_msi(kvm, &msi);
3930 #ifdef __KVM_HAVE_IRQ_LINE
3931 case KVM_IRQ_LINE_STATUS:
3932 case KVM_IRQ_LINE: {
3933 struct kvm_irq_level irq_event;
3936 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
3939 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
3940 ioctl == KVM_IRQ_LINE_STATUS);
3945 if (ioctl == KVM_IRQ_LINE_STATUS) {
3946 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
3954 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3955 case KVM_SET_GSI_ROUTING: {
3956 struct kvm_irq_routing routing;
3957 struct kvm_irq_routing __user *urouting;
3958 struct kvm_irq_routing_entry *entries = NULL;
3961 if (copy_from_user(&routing, argp, sizeof(routing)))
3964 if (!kvm_arch_can_set_irq_routing(kvm))
3966 if (routing.nr > KVM_MAX_IRQ_ROUTES)
3972 entries = vmemdup_user(urouting->entries,
3973 array_size(sizeof(*entries),
3975 if (IS_ERR(entries)) {
3976 r = PTR_ERR(entries);
3980 r = kvm_set_irq_routing(kvm, entries, routing.nr,
3985 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
3986 case KVM_CREATE_DEVICE: {
3987 struct kvm_create_device cd;
3990 if (copy_from_user(&cd, argp, sizeof(cd)))
3993 r = kvm_ioctl_create_device(kvm, &cd);
3998 if (copy_to_user(argp, &cd, sizeof(cd)))
4004 case KVM_CHECK_EXTENSION:
4005 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
4007 case KVM_RESET_DIRTY_RINGS:
4008 r = kvm_vm_ioctl_reset_dirty_pages(kvm);
4011 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
4017 #ifdef CONFIG_KVM_COMPAT
4018 struct compat_kvm_dirty_log {
4022 compat_uptr_t dirty_bitmap; /* one bit per page */
4027 static long kvm_vm_compat_ioctl(struct file *filp,
4028 unsigned int ioctl, unsigned long arg)
4030 struct kvm *kvm = filp->private_data;
4033 if (kvm->mm != current->mm)
4036 case KVM_GET_DIRTY_LOG: {
4037 struct compat_kvm_dirty_log compat_log;
4038 struct kvm_dirty_log log;
4040 if (copy_from_user(&compat_log, (void __user *)arg,
4041 sizeof(compat_log)))
4043 log.slot = compat_log.slot;
4044 log.padding1 = compat_log.padding1;
4045 log.padding2 = compat_log.padding2;
4046 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
4048 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
4052 r = kvm_vm_ioctl(filp, ioctl, arg);
4058 static struct file_operations kvm_vm_fops = {
4059 .release = kvm_vm_release,
4060 .unlocked_ioctl = kvm_vm_ioctl,
4061 .llseek = noop_llseek,
4062 KVM_COMPAT(kvm_vm_compat_ioctl),
4065 static int kvm_dev_ioctl_create_vm(unsigned long type)
4071 kvm = kvm_create_vm(type);
4073 return PTR_ERR(kvm);
4074 #ifdef CONFIG_KVM_MMIO
4075 r = kvm_coalesced_mmio_init(kvm);
4079 r = get_unused_fd_flags(O_CLOEXEC);
4083 file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
4091 * Don't call kvm_put_kvm anymore at this point; file->f_op is
4092 * already set, with ->release() being kvm_vm_release(). In error
4093 * cases it will be called by the final fput(file) and will take
4094 * care of doing kvm_put_kvm(kvm).
4096 if (kvm_create_vm_debugfs(kvm, r) < 0) {
4101 kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm);
4103 fd_install(r, file);
4111 static long kvm_dev_ioctl(struct file *filp,
4112 unsigned int ioctl, unsigned long arg)
4117 case KVM_GET_API_VERSION:
4120 r = KVM_API_VERSION;
4123 r = kvm_dev_ioctl_create_vm(arg);
4125 case KVM_CHECK_EXTENSION:
4126 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
4128 case KVM_GET_VCPU_MMAP_SIZE:
4131 r = PAGE_SIZE; /* struct kvm_run */
4133 r += PAGE_SIZE; /* pio data page */
4135 #ifdef CONFIG_KVM_MMIO
4136 r += PAGE_SIZE; /* coalesced mmio ring page */
4139 case KVM_TRACE_ENABLE:
4140 case KVM_TRACE_PAUSE:
4141 case KVM_TRACE_DISABLE:
4145 return kvm_arch_dev_ioctl(filp, ioctl, arg);
4151 static struct file_operations kvm_chardev_ops = {
4152 .unlocked_ioctl = kvm_dev_ioctl,
4153 .llseek = noop_llseek,
4154 KVM_COMPAT(kvm_dev_ioctl),
4157 static struct miscdevice kvm_dev = {
4163 static void hardware_enable_nolock(void *junk)
4165 int cpu = raw_smp_processor_id();
4168 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
4171 cpumask_set_cpu(cpu, cpus_hardware_enabled);
4173 r = kvm_arch_hardware_enable();
4176 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
4177 atomic_inc(&hardware_enable_failed);
4178 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
4182 static int kvm_starting_cpu(unsigned int cpu)
4184 raw_spin_lock(&kvm_count_lock);
4185 if (kvm_usage_count)
4186 hardware_enable_nolock(NULL);
4187 raw_spin_unlock(&kvm_count_lock);
4191 static void hardware_disable_nolock(void *junk)
4193 int cpu = raw_smp_processor_id();
4195 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
4197 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
4198 kvm_arch_hardware_disable();
4201 static int kvm_dying_cpu(unsigned int cpu)
4203 raw_spin_lock(&kvm_count_lock);
4204 if (kvm_usage_count)
4205 hardware_disable_nolock(NULL);
4206 raw_spin_unlock(&kvm_count_lock);
4210 static void hardware_disable_all_nolock(void)
4212 BUG_ON(!kvm_usage_count);
4215 if (!kvm_usage_count)
4216 on_each_cpu(hardware_disable_nolock, NULL, 1);
4219 static void hardware_disable_all(void)
4221 raw_spin_lock(&kvm_count_lock);
4222 hardware_disable_all_nolock();
4223 raw_spin_unlock(&kvm_count_lock);
4226 static int hardware_enable_all(void)
4230 raw_spin_lock(&kvm_count_lock);
4233 if (kvm_usage_count == 1) {
4234 atomic_set(&hardware_enable_failed, 0);
4235 on_each_cpu(hardware_enable_nolock, NULL, 1);
4237 if (atomic_read(&hardware_enable_failed)) {
4238 hardware_disable_all_nolock();
4243 raw_spin_unlock(&kvm_count_lock);
4248 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
4252 * Some (well, at least mine) BIOSes hang on reboot if
4255 * And Intel TXT required VMX off for all cpu when system shutdown.
4257 pr_info("kvm: exiting hardware virtualization\n");
4258 kvm_rebooting = true;
4259 on_each_cpu(hardware_disable_nolock, NULL, 1);
4263 static struct notifier_block kvm_reboot_notifier = {
4264 .notifier_call = kvm_reboot,
4268 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
4272 for (i = 0; i < bus->dev_count; i++) {
4273 struct kvm_io_device *pos = bus->range[i].dev;
4275 kvm_iodevice_destructor(pos);
4280 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
4281 const struct kvm_io_range *r2)
4283 gpa_t addr1 = r1->addr;
4284 gpa_t addr2 = r2->addr;
4289 /* If r2->len == 0, match the exact address. If r2->len != 0,
4290 * accept any overlapping write. Any order is acceptable for
4291 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
4292 * we process all of them.
4305 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
4307 return kvm_io_bus_cmp(p1, p2);
4310 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
4311 gpa_t addr, int len)
4313 struct kvm_io_range *range, key;
4316 key = (struct kvm_io_range) {
4321 range = bsearch(&key, bus->range, bus->dev_count,
4322 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
4326 off = range - bus->range;
4328 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
4334 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
4335 struct kvm_io_range *range, const void *val)
4339 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
4343 while (idx < bus->dev_count &&
4344 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
4345 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
4354 /* kvm_io_bus_write - called under kvm->slots_lock */
4355 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
4356 int len, const void *val)
4358 struct kvm_io_bus *bus;
4359 struct kvm_io_range range;
4362 range = (struct kvm_io_range) {
4367 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
4370 r = __kvm_io_bus_write(vcpu, bus, &range, val);
4371 return r < 0 ? r : 0;
4373 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
4375 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
4376 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
4377 gpa_t addr, int len, const void *val, long cookie)
4379 struct kvm_io_bus *bus;
4380 struct kvm_io_range range;
4382 range = (struct kvm_io_range) {
4387 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
4391 /* First try the device referenced by cookie. */
4392 if ((cookie >= 0) && (cookie < bus->dev_count) &&
4393 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
4394 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
4399 * cookie contained garbage; fall back to search and return the
4400 * correct cookie value.
4402 return __kvm_io_bus_write(vcpu, bus, &range, val);
4405 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
4406 struct kvm_io_range *range, void *val)
4410 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
4414 while (idx < bus->dev_count &&
4415 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
4416 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
4425 /* kvm_io_bus_read - called under kvm->slots_lock */
4426 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
4429 struct kvm_io_bus *bus;
4430 struct kvm_io_range range;
4433 range = (struct kvm_io_range) {
4438 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
4441 r = __kvm_io_bus_read(vcpu, bus, &range, val);
4442 return r < 0 ? r : 0;
4445 /* Caller must hold slots_lock. */
4446 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
4447 int len, struct kvm_io_device *dev)
4450 struct kvm_io_bus *new_bus, *bus;
4451 struct kvm_io_range range;
4453 bus = kvm_get_bus(kvm, bus_idx);
4457 /* exclude ioeventfd which is limited by maximum fd */
4458 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
4461 new_bus = kmalloc(struct_size(bus, range, bus->dev_count + 1),
4462 GFP_KERNEL_ACCOUNT);
4466 range = (struct kvm_io_range) {
4472 for (i = 0; i < bus->dev_count; i++)
4473 if (kvm_io_bus_cmp(&bus->range[i], &range) > 0)
4476 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
4477 new_bus->dev_count++;
4478 new_bus->range[i] = range;
4479 memcpy(new_bus->range + i + 1, bus->range + i,
4480 (bus->dev_count - i) * sizeof(struct kvm_io_range));
4481 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
4482 synchronize_srcu_expedited(&kvm->srcu);
4488 /* Caller must hold slots_lock. */
4489 void kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
4490 struct kvm_io_device *dev)
4493 struct kvm_io_bus *new_bus, *bus;
4495 bus = kvm_get_bus(kvm, bus_idx);
4499 for (i = 0; i < bus->dev_count; i++)
4500 if (bus->range[i].dev == dev) {
4504 if (i == bus->dev_count)
4507 new_bus = kmalloc(struct_size(bus, range, bus->dev_count - 1),
4508 GFP_KERNEL_ACCOUNT);
4510 memcpy(new_bus, bus, struct_size(bus, range, i));
4511 new_bus->dev_count--;
4512 memcpy(new_bus->range + i, bus->range + i + 1,
4513 flex_array_size(new_bus, range, new_bus->dev_count - i));
4515 pr_err("kvm: failed to shrink bus, removing it completely\n");
4516 for (j = 0; j < bus->dev_count; j++) {
4519 kvm_iodevice_destructor(bus->range[j].dev);
4523 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
4524 synchronize_srcu_expedited(&kvm->srcu);
4529 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
4532 struct kvm_io_bus *bus;
4533 int dev_idx, srcu_idx;
4534 struct kvm_io_device *iodev = NULL;
4536 srcu_idx = srcu_read_lock(&kvm->srcu);
4538 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
4542 dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
4546 iodev = bus->range[dev_idx].dev;
4549 srcu_read_unlock(&kvm->srcu, srcu_idx);
4553 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
4555 static int kvm_debugfs_open(struct inode *inode, struct file *file,
4556 int (*get)(void *, u64 *), int (*set)(void *, u64),
4559 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
4562 /* The debugfs files are a reference to the kvm struct which
4563 * is still valid when kvm_destroy_vm is called.
4564 * To avoid the race between open and the removal of the debugfs
4565 * directory we test against the users count.
4567 if (!refcount_inc_not_zero(&stat_data->kvm->users_count))
4570 if (simple_attr_open(inode, file, get,
4571 KVM_DBGFS_GET_MODE(stat_data->dbgfs_item) & 0222
4574 kvm_put_kvm(stat_data->kvm);
4581 static int kvm_debugfs_release(struct inode *inode, struct file *file)
4583 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
4586 simple_attr_release(inode, file);
4587 kvm_put_kvm(stat_data->kvm);
4592 static int kvm_get_stat_per_vm(struct kvm *kvm, size_t offset, u64 *val)
4594 *val = *(ulong *)((void *)kvm + offset);
4599 static int kvm_clear_stat_per_vm(struct kvm *kvm, size_t offset)
4601 *(ulong *)((void *)kvm + offset) = 0;
4606 static int kvm_get_stat_per_vcpu(struct kvm *kvm, size_t offset, u64 *val)
4609 struct kvm_vcpu *vcpu;
4613 kvm_for_each_vcpu(i, vcpu, kvm)
4614 *val += *(u64 *)((void *)vcpu + offset);
4619 static int kvm_clear_stat_per_vcpu(struct kvm *kvm, size_t offset)
4622 struct kvm_vcpu *vcpu;
4624 kvm_for_each_vcpu(i, vcpu, kvm)
4625 *(u64 *)((void *)vcpu + offset) = 0;
4630 static int kvm_stat_data_get(void *data, u64 *val)
4633 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
4635 switch (stat_data->dbgfs_item->kind) {
4637 r = kvm_get_stat_per_vm(stat_data->kvm,
4638 stat_data->dbgfs_item->offset, val);
4641 r = kvm_get_stat_per_vcpu(stat_data->kvm,
4642 stat_data->dbgfs_item->offset, val);
4649 static int kvm_stat_data_clear(void *data, u64 val)
4652 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
4657 switch (stat_data->dbgfs_item->kind) {
4659 r = kvm_clear_stat_per_vm(stat_data->kvm,
4660 stat_data->dbgfs_item->offset);
4663 r = kvm_clear_stat_per_vcpu(stat_data->kvm,
4664 stat_data->dbgfs_item->offset);
4671 static int kvm_stat_data_open(struct inode *inode, struct file *file)
4673 __simple_attr_check_format("%llu\n", 0ull);
4674 return kvm_debugfs_open(inode, file, kvm_stat_data_get,
4675 kvm_stat_data_clear, "%llu\n");
4678 static const struct file_operations stat_fops_per_vm = {
4679 .owner = THIS_MODULE,
4680 .open = kvm_stat_data_open,
4681 .release = kvm_debugfs_release,
4682 .read = simple_attr_read,
4683 .write = simple_attr_write,
4684 .llseek = no_llseek,
4687 static int vm_stat_get(void *_offset, u64 *val)
4689 unsigned offset = (long)_offset;
4694 mutex_lock(&kvm_lock);
4695 list_for_each_entry(kvm, &vm_list, vm_list) {
4696 kvm_get_stat_per_vm(kvm, offset, &tmp_val);
4699 mutex_unlock(&kvm_lock);
4703 static int vm_stat_clear(void *_offset, u64 val)
4705 unsigned offset = (long)_offset;
4711 mutex_lock(&kvm_lock);
4712 list_for_each_entry(kvm, &vm_list, vm_list) {
4713 kvm_clear_stat_per_vm(kvm, offset);
4715 mutex_unlock(&kvm_lock);
4720 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n");
4722 static int vcpu_stat_get(void *_offset, u64 *val)
4724 unsigned offset = (long)_offset;
4729 mutex_lock(&kvm_lock);
4730 list_for_each_entry(kvm, &vm_list, vm_list) {
4731 kvm_get_stat_per_vcpu(kvm, offset, &tmp_val);
4734 mutex_unlock(&kvm_lock);
4738 static int vcpu_stat_clear(void *_offset, u64 val)
4740 unsigned offset = (long)_offset;
4746 mutex_lock(&kvm_lock);
4747 list_for_each_entry(kvm, &vm_list, vm_list) {
4748 kvm_clear_stat_per_vcpu(kvm, offset);
4750 mutex_unlock(&kvm_lock);
4755 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear,
4758 static const struct file_operations *stat_fops[] = {
4759 [KVM_STAT_VCPU] = &vcpu_stat_fops,
4760 [KVM_STAT_VM] = &vm_stat_fops,
4763 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm)
4765 struct kobj_uevent_env *env;
4766 unsigned long long created, active;
4768 if (!kvm_dev.this_device || !kvm)
4771 mutex_lock(&kvm_lock);
4772 if (type == KVM_EVENT_CREATE_VM) {
4773 kvm_createvm_count++;
4775 } else if (type == KVM_EVENT_DESTROY_VM) {
4778 created = kvm_createvm_count;
4779 active = kvm_active_vms;
4780 mutex_unlock(&kvm_lock);
4782 env = kzalloc(sizeof(*env), GFP_KERNEL_ACCOUNT);
4786 add_uevent_var(env, "CREATED=%llu", created);
4787 add_uevent_var(env, "COUNT=%llu", active);
4789 if (type == KVM_EVENT_CREATE_VM) {
4790 add_uevent_var(env, "EVENT=create");
4791 kvm->userspace_pid = task_pid_nr(current);
4792 } else if (type == KVM_EVENT_DESTROY_VM) {
4793 add_uevent_var(env, "EVENT=destroy");
4795 add_uevent_var(env, "PID=%d", kvm->userspace_pid);
4797 if (!IS_ERR_OR_NULL(kvm->debugfs_dentry)) {
4798 char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL_ACCOUNT);
4801 tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX);
4803 add_uevent_var(env, "STATS_PATH=%s", tmp);
4807 /* no need for checks, since we are adding at most only 5 keys */
4808 env->envp[env->envp_idx++] = NULL;
4809 kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp);
4813 static void kvm_init_debug(void)
4815 struct kvm_stats_debugfs_item *p;
4817 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
4819 kvm_debugfs_num_entries = 0;
4820 for (p = debugfs_entries; p->name; ++p, kvm_debugfs_num_entries++) {
4821 debugfs_create_file(p->name, KVM_DBGFS_GET_MODE(p),
4822 kvm_debugfs_dir, (void *)(long)p->offset,
4823 stat_fops[p->kind]);
4827 static int kvm_suspend(void)
4829 if (kvm_usage_count)
4830 hardware_disable_nolock(NULL);
4834 static void kvm_resume(void)
4836 if (kvm_usage_count) {
4837 #ifdef CONFIG_LOCKDEP
4838 WARN_ON(lockdep_is_held(&kvm_count_lock));
4840 hardware_enable_nolock(NULL);
4844 static struct syscore_ops kvm_syscore_ops = {
4845 .suspend = kvm_suspend,
4846 .resume = kvm_resume,
4850 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
4852 return container_of(pn, struct kvm_vcpu, preempt_notifier);
4855 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
4857 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
4859 WRITE_ONCE(vcpu->preempted, false);
4860 WRITE_ONCE(vcpu->ready, false);
4862 __this_cpu_write(kvm_running_vcpu, vcpu);
4863 kvm_arch_sched_in(vcpu, cpu);
4864 kvm_arch_vcpu_load(vcpu, cpu);
4867 static void kvm_sched_out(struct preempt_notifier *pn,
4868 struct task_struct *next)
4870 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
4872 if (current->state == TASK_RUNNING) {
4873 WRITE_ONCE(vcpu->preempted, true);
4874 WRITE_ONCE(vcpu->ready, true);
4876 kvm_arch_vcpu_put(vcpu);
4877 __this_cpu_write(kvm_running_vcpu, NULL);
4881 * kvm_get_running_vcpu - get the vcpu running on the current CPU.
4883 * We can disable preemption locally around accessing the per-CPU variable,
4884 * and use the resolved vcpu pointer after enabling preemption again,
4885 * because even if the current thread is migrated to another CPU, reading
4886 * the per-CPU value later will give us the same value as we update the
4887 * per-CPU variable in the preempt notifier handlers.
4889 struct kvm_vcpu *kvm_get_running_vcpu(void)
4891 struct kvm_vcpu *vcpu;
4894 vcpu = __this_cpu_read(kvm_running_vcpu);
4899 EXPORT_SYMBOL_GPL(kvm_get_running_vcpu);
4902 * kvm_get_running_vcpus - get the per-CPU array of currently running vcpus.
4904 struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void)
4906 return &kvm_running_vcpu;
4909 struct kvm_cpu_compat_check {
4914 static void check_processor_compat(void *data)
4916 struct kvm_cpu_compat_check *c = data;
4918 *c->ret = kvm_arch_check_processor_compat(c->opaque);
4921 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
4922 struct module *module)
4924 struct kvm_cpu_compat_check c;
4928 r = kvm_arch_init(opaque);
4933 * kvm_arch_init makes sure there's at most one caller
4934 * for architectures that support multiple implementations,
4935 * like intel and amd on x86.
4936 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
4937 * conflicts in case kvm is already setup for another implementation.
4939 r = kvm_irqfd_init();
4943 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
4948 r = kvm_arch_hardware_setup(opaque);
4954 for_each_online_cpu(cpu) {
4955 smp_call_function_single(cpu, check_processor_compat, &c, 1);
4960 r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting",
4961 kvm_starting_cpu, kvm_dying_cpu);
4964 register_reboot_notifier(&kvm_reboot_notifier);
4966 /* A kmem cache lets us meet the alignment requirements of fx_save. */
4968 vcpu_align = __alignof__(struct kvm_vcpu);
4970 kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align,
4972 offsetof(struct kvm_vcpu, arch),
4973 sizeof_field(struct kvm_vcpu, arch),
4975 if (!kvm_vcpu_cache) {
4980 r = kvm_async_pf_init();
4984 kvm_chardev_ops.owner = module;
4985 kvm_vm_fops.owner = module;
4986 kvm_vcpu_fops.owner = module;
4988 r = misc_register(&kvm_dev);
4990 pr_err("kvm: misc device register failed\n");
4994 register_syscore_ops(&kvm_syscore_ops);
4996 kvm_preempt_ops.sched_in = kvm_sched_in;
4997 kvm_preempt_ops.sched_out = kvm_sched_out;
5001 r = kvm_vfio_ops_init();
5007 kvm_async_pf_deinit();
5009 kmem_cache_destroy(kvm_vcpu_cache);
5011 unregister_reboot_notifier(&kvm_reboot_notifier);
5012 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
5014 kvm_arch_hardware_unsetup();
5016 free_cpumask_var(cpus_hardware_enabled);
5024 EXPORT_SYMBOL_GPL(kvm_init);
5028 debugfs_remove_recursive(kvm_debugfs_dir);
5029 misc_deregister(&kvm_dev);
5030 kmem_cache_destroy(kvm_vcpu_cache);
5031 kvm_async_pf_deinit();
5032 unregister_syscore_ops(&kvm_syscore_ops);
5033 unregister_reboot_notifier(&kvm_reboot_notifier);
5034 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
5035 on_each_cpu(hardware_disable_nolock, NULL, 1);
5036 kvm_arch_hardware_unsetup();
5039 free_cpumask_var(cpus_hardware_enabled);
5040 kvm_vfio_ops_exit();
5042 EXPORT_SYMBOL_GPL(kvm_exit);
5044 struct kvm_vm_worker_thread_context {
5046 struct task_struct *parent;
5047 struct completion init_done;
5048 kvm_vm_thread_fn_t thread_fn;
5053 static int kvm_vm_worker_thread(void *context)
5056 * The init_context is allocated on the stack of the parent thread, so
5057 * we have to locally copy anything that is needed beyond initialization
5059 struct kvm_vm_worker_thread_context *init_context = context;
5060 struct kvm *kvm = init_context->kvm;
5061 kvm_vm_thread_fn_t thread_fn = init_context->thread_fn;
5062 uintptr_t data = init_context->data;
5065 err = kthread_park(current);
5066 /* kthread_park(current) is never supposed to return an error */
5071 err = cgroup_attach_task_all(init_context->parent, current);
5073 kvm_err("%s: cgroup_attach_task_all failed with err %d\n",
5078 set_user_nice(current, task_nice(init_context->parent));
5081 init_context->err = err;
5082 complete(&init_context->init_done);
5083 init_context = NULL;
5088 /* Wait to be woken up by the spawner before proceeding. */
5091 if (!kthread_should_stop())
5092 err = thread_fn(kvm, data);
5097 int kvm_vm_create_worker_thread(struct kvm *kvm, kvm_vm_thread_fn_t thread_fn,
5098 uintptr_t data, const char *name,
5099 struct task_struct **thread_ptr)
5101 struct kvm_vm_worker_thread_context init_context = {};
5102 struct task_struct *thread;
5105 init_context.kvm = kvm;
5106 init_context.parent = current;
5107 init_context.thread_fn = thread_fn;
5108 init_context.data = data;
5109 init_completion(&init_context.init_done);
5111 thread = kthread_run(kvm_vm_worker_thread, &init_context,
5112 "%s-%d", name, task_pid_nr(current));
5114 return PTR_ERR(thread);
5116 /* kthread_run is never supposed to return NULL */
5117 WARN_ON(thread == NULL);
5119 wait_for_completion(&init_context.init_done);
5121 if (!init_context.err)
5122 *thread_ptr = thread;
5124 return init_context.err;