2 * Kernel-based Virtual Machine driver for Linux
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
7 * Copyright (C) 2006 Qumranet, Inc.
8 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
11 * Avi Kivity <avi@qumranet.com>
12 * Yaniv Kamay <yaniv@qumranet.com>
14 * This work is licensed under the terms of the GNU GPL, version 2. See
15 * the COPYING file in the top-level directory.
19 #include <kvm/iodev.h>
21 #include <linux/kvm_host.h>
22 #include <linux/kvm.h>
23 #include <linux/module.h>
24 #include <linux/errno.h>
25 #include <linux/percpu.h>
27 #include <linux/miscdevice.h>
28 #include <linux/vmalloc.h>
29 #include <linux/reboot.h>
30 #include <linux/debugfs.h>
31 #include <linux/highmem.h>
32 #include <linux/file.h>
33 #include <linux/syscore_ops.h>
34 #include <linux/cpu.h>
35 #include <linux/sched/signal.h>
36 #include <linux/sched/mm.h>
37 #include <linux/sched/stat.h>
38 #include <linux/cpumask.h>
39 #include <linux/smp.h>
40 #include <linux/anon_inodes.h>
41 #include <linux/profile.h>
42 #include <linux/kvm_para.h>
43 #include <linux/pagemap.h>
44 #include <linux/mman.h>
45 #include <linux/swap.h>
46 #include <linux/bitops.h>
47 #include <linux/spinlock.h>
48 #include <linux/compat.h>
49 #include <linux/srcu.h>
50 #include <linux/hugetlb.h>
51 #include <linux/slab.h>
52 #include <linux/sort.h>
53 #include <linux/bsearch.h>
55 #include <asm/processor.h>
57 #include <asm/ioctl.h>
58 #include <linux/uaccess.h>
59 #include <asm/pgtable.h>
61 #include "coalesced_mmio.h"
65 #define CREATE_TRACE_POINTS
66 #include <trace/events/kvm.h>
68 /* Worst case buffer size needed for holding an integer. */
69 #define ITOA_MAX_LEN 12
71 MODULE_AUTHOR("Qumranet");
72 MODULE_LICENSE("GPL");
74 /* Architectures should define their poll value according to the halt latency */
75 unsigned int halt_poll_ns = KVM_HALT_POLL_NS_DEFAULT;
76 module_param(halt_poll_ns, uint, 0644);
77 EXPORT_SYMBOL_GPL(halt_poll_ns);
79 /* Default doubles per-vcpu halt_poll_ns. */
80 unsigned int halt_poll_ns_grow = 2;
81 module_param(halt_poll_ns_grow, uint, 0644);
82 EXPORT_SYMBOL_GPL(halt_poll_ns_grow);
84 /* The start value to grow halt_poll_ns from */
85 unsigned int halt_poll_ns_grow_start = 10000; /* 10us */
86 module_param(halt_poll_ns_grow_start, uint, 0644);
87 EXPORT_SYMBOL_GPL(halt_poll_ns_grow_start);
89 /* Default resets per-vcpu halt_poll_ns . */
90 unsigned int halt_poll_ns_shrink;
91 module_param(halt_poll_ns_shrink, uint, 0644);
92 EXPORT_SYMBOL_GPL(halt_poll_ns_shrink);
97 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
100 DEFINE_SPINLOCK(kvm_lock);
101 static DEFINE_RAW_SPINLOCK(kvm_count_lock);
104 static cpumask_var_t cpus_hardware_enabled;
105 static int kvm_usage_count;
106 static atomic_t hardware_enable_failed;
108 struct kmem_cache *kvm_vcpu_cache;
109 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
111 static __read_mostly struct preempt_ops kvm_preempt_ops;
113 struct dentry *kvm_debugfs_dir;
114 EXPORT_SYMBOL_GPL(kvm_debugfs_dir);
116 static int kvm_debugfs_num_entries;
117 static const struct file_operations *stat_fops_per_vm[];
119 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
121 #ifdef CONFIG_KVM_COMPAT
122 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
124 #define KVM_COMPAT(c) .compat_ioctl = (c)
126 static long kvm_no_compat_ioctl(struct file *file, unsigned int ioctl,
127 unsigned long arg) { return -EINVAL; }
128 #define KVM_COMPAT(c) .compat_ioctl = kvm_no_compat_ioctl
130 static int hardware_enable_all(void);
131 static void hardware_disable_all(void);
133 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
135 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot, gfn_t gfn);
137 __visible bool kvm_rebooting;
138 EXPORT_SYMBOL_GPL(kvm_rebooting);
140 static bool largepages_enabled = true;
142 #define KVM_EVENT_CREATE_VM 0
143 #define KVM_EVENT_DESTROY_VM 1
144 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm);
145 static unsigned long long kvm_createvm_count;
146 static unsigned long long kvm_active_vms;
148 __weak int kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
149 unsigned long start, unsigned long end, bool blockable)
154 bool kvm_is_reserved_pfn(kvm_pfn_t pfn)
157 return PageReserved(pfn_to_page(pfn));
163 * Switches to specified vcpu, until a matching vcpu_put()
165 void vcpu_load(struct kvm_vcpu *vcpu)
168 preempt_notifier_register(&vcpu->preempt_notifier);
169 kvm_arch_vcpu_load(vcpu, cpu);
172 EXPORT_SYMBOL_GPL(vcpu_load);
174 void vcpu_put(struct kvm_vcpu *vcpu)
177 kvm_arch_vcpu_put(vcpu);
178 preempt_notifier_unregister(&vcpu->preempt_notifier);
181 EXPORT_SYMBOL_GPL(vcpu_put);
183 /* TODO: merge with kvm_arch_vcpu_should_kick */
184 static bool kvm_request_needs_ipi(struct kvm_vcpu *vcpu, unsigned req)
186 int mode = kvm_vcpu_exiting_guest_mode(vcpu);
189 * We need to wait for the VCPU to reenable interrupts and get out of
190 * READING_SHADOW_PAGE_TABLES mode.
192 if (req & KVM_REQUEST_WAIT)
193 return mode != OUTSIDE_GUEST_MODE;
196 * Need to kick a running VCPU, but otherwise there is nothing to do.
198 return mode == IN_GUEST_MODE;
201 static void ack_flush(void *_completed)
205 static inline bool kvm_kick_many_cpus(const struct cpumask *cpus, bool wait)
208 cpus = cpu_online_mask;
210 if (cpumask_empty(cpus))
213 smp_call_function_many(cpus, ack_flush, NULL, wait);
217 bool kvm_make_vcpus_request_mask(struct kvm *kvm, unsigned int req,
218 unsigned long *vcpu_bitmap, cpumask_var_t tmp)
221 struct kvm_vcpu *vcpu;
226 kvm_for_each_vcpu(i, vcpu, kvm) {
227 if (vcpu_bitmap && !test_bit(i, vcpu_bitmap))
230 kvm_make_request(req, vcpu);
233 if (!(req & KVM_REQUEST_NO_WAKEUP) && kvm_vcpu_wake_up(vcpu))
236 if (tmp != NULL && cpu != -1 && cpu != me &&
237 kvm_request_needs_ipi(vcpu, req))
238 __cpumask_set_cpu(cpu, tmp);
241 called = kvm_kick_many_cpus(tmp, !!(req & KVM_REQUEST_WAIT));
247 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
252 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
254 called = kvm_make_vcpus_request_mask(kvm, req, NULL, cpus);
256 free_cpumask_var(cpus);
260 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
261 void kvm_flush_remote_tlbs(struct kvm *kvm)
264 * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in
265 * kvm_make_all_cpus_request.
267 long dirty_count = smp_load_acquire(&kvm->tlbs_dirty);
270 * We want to publish modifications to the page tables before reading
271 * mode. Pairs with a memory barrier in arch-specific code.
272 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
273 * and smp_mb in walk_shadow_page_lockless_begin/end.
274 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
276 * There is already an smp_mb__after_atomic() before
277 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
280 if (!kvm_arch_flush_remote_tlb(kvm)
281 || kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
282 ++kvm->stat.remote_tlb_flush;
283 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
285 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
288 void kvm_reload_remote_mmus(struct kvm *kvm)
290 kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
293 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
298 mutex_init(&vcpu->mutex);
303 init_swait_queue_head(&vcpu->wq);
304 kvm_async_pf_vcpu_init(vcpu);
307 INIT_LIST_HEAD(&vcpu->blocked_vcpu_list);
309 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
314 vcpu->run = page_address(page);
316 kvm_vcpu_set_in_spin_loop(vcpu, false);
317 kvm_vcpu_set_dy_eligible(vcpu, false);
318 vcpu->preempted = false;
320 r = kvm_arch_vcpu_init(vcpu);
326 free_page((unsigned long)vcpu->run);
330 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
332 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
335 * no need for rcu_read_lock as VCPU_RUN is the only place that
336 * will change the vcpu->pid pointer and on uninit all file
337 * descriptors are already gone.
339 put_pid(rcu_dereference_protected(vcpu->pid, 1));
340 kvm_arch_vcpu_uninit(vcpu);
341 free_page((unsigned long)vcpu->run);
343 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
345 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
346 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
348 return container_of(mn, struct kvm, mmu_notifier);
351 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
352 struct mm_struct *mm,
353 unsigned long address,
356 struct kvm *kvm = mmu_notifier_to_kvm(mn);
359 idx = srcu_read_lock(&kvm->srcu);
360 spin_lock(&kvm->mmu_lock);
361 kvm->mmu_notifier_seq++;
363 if (kvm_set_spte_hva(kvm, address, pte))
364 kvm_flush_remote_tlbs(kvm);
366 spin_unlock(&kvm->mmu_lock);
367 srcu_read_unlock(&kvm->srcu, idx);
370 static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
371 const struct mmu_notifier_range *range)
373 struct kvm *kvm = mmu_notifier_to_kvm(mn);
374 int need_tlb_flush = 0, idx;
377 idx = srcu_read_lock(&kvm->srcu);
378 spin_lock(&kvm->mmu_lock);
380 * The count increase must become visible at unlock time as no
381 * spte can be established without taking the mmu_lock and
382 * count is also read inside the mmu_lock critical section.
384 kvm->mmu_notifier_count++;
385 need_tlb_flush = kvm_unmap_hva_range(kvm, range->start, range->end);
386 need_tlb_flush |= kvm->tlbs_dirty;
387 /* we've to flush the tlb before the pages can be freed */
389 kvm_flush_remote_tlbs(kvm);
391 spin_unlock(&kvm->mmu_lock);
393 ret = kvm_arch_mmu_notifier_invalidate_range(kvm, range->start,
394 range->end, range->blockable);
396 srcu_read_unlock(&kvm->srcu, idx);
401 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
402 const struct mmu_notifier_range *range)
404 struct kvm *kvm = mmu_notifier_to_kvm(mn);
406 spin_lock(&kvm->mmu_lock);
408 * This sequence increase will notify the kvm page fault that
409 * the page that is going to be mapped in the spte could have
412 kvm->mmu_notifier_seq++;
415 * The above sequence increase must be visible before the
416 * below count decrease, which is ensured by the smp_wmb above
417 * in conjunction with the smp_rmb in mmu_notifier_retry().
419 kvm->mmu_notifier_count--;
420 spin_unlock(&kvm->mmu_lock);
422 BUG_ON(kvm->mmu_notifier_count < 0);
425 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
426 struct mm_struct *mm,
430 struct kvm *kvm = mmu_notifier_to_kvm(mn);
433 idx = srcu_read_lock(&kvm->srcu);
434 spin_lock(&kvm->mmu_lock);
436 young = kvm_age_hva(kvm, start, end);
438 kvm_flush_remote_tlbs(kvm);
440 spin_unlock(&kvm->mmu_lock);
441 srcu_read_unlock(&kvm->srcu, idx);
446 static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
447 struct mm_struct *mm,
451 struct kvm *kvm = mmu_notifier_to_kvm(mn);
454 idx = srcu_read_lock(&kvm->srcu);
455 spin_lock(&kvm->mmu_lock);
457 * Even though we do not flush TLB, this will still adversely
458 * affect performance on pre-Haswell Intel EPT, where there is
459 * no EPT Access Bit to clear so that we have to tear down EPT
460 * tables instead. If we find this unacceptable, we can always
461 * add a parameter to kvm_age_hva so that it effectively doesn't
462 * do anything on clear_young.
464 * Also note that currently we never issue secondary TLB flushes
465 * from clear_young, leaving this job up to the regular system
466 * cadence. If we find this inaccurate, we might come up with a
467 * more sophisticated heuristic later.
469 young = kvm_age_hva(kvm, start, end);
470 spin_unlock(&kvm->mmu_lock);
471 srcu_read_unlock(&kvm->srcu, idx);
476 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
477 struct mm_struct *mm,
478 unsigned long address)
480 struct kvm *kvm = mmu_notifier_to_kvm(mn);
483 idx = srcu_read_lock(&kvm->srcu);
484 spin_lock(&kvm->mmu_lock);
485 young = kvm_test_age_hva(kvm, address);
486 spin_unlock(&kvm->mmu_lock);
487 srcu_read_unlock(&kvm->srcu, idx);
492 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
493 struct mm_struct *mm)
495 struct kvm *kvm = mmu_notifier_to_kvm(mn);
498 idx = srcu_read_lock(&kvm->srcu);
499 kvm_arch_flush_shadow_all(kvm);
500 srcu_read_unlock(&kvm->srcu, idx);
503 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
504 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
505 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
506 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
507 .clear_young = kvm_mmu_notifier_clear_young,
508 .test_young = kvm_mmu_notifier_test_young,
509 .change_pte = kvm_mmu_notifier_change_pte,
510 .release = kvm_mmu_notifier_release,
513 static int kvm_init_mmu_notifier(struct kvm *kvm)
515 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
516 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
519 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
521 static int kvm_init_mmu_notifier(struct kvm *kvm)
526 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
528 static struct kvm_memslots *kvm_alloc_memslots(void)
531 struct kvm_memslots *slots;
533 slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL_ACCOUNT);
537 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
538 slots->id_to_index[i] = slots->memslots[i].id = i;
543 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
545 if (!memslot->dirty_bitmap)
548 kvfree(memslot->dirty_bitmap);
549 memslot->dirty_bitmap = NULL;
553 * Free any memory in @free but not in @dont.
555 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
556 struct kvm_memory_slot *dont)
558 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
559 kvm_destroy_dirty_bitmap(free);
561 kvm_arch_free_memslot(kvm, free, dont);
566 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
568 struct kvm_memory_slot *memslot;
573 kvm_for_each_memslot(memslot, slots)
574 kvm_free_memslot(kvm, memslot, NULL);
579 static void kvm_destroy_vm_debugfs(struct kvm *kvm)
583 if (!kvm->debugfs_dentry)
586 debugfs_remove_recursive(kvm->debugfs_dentry);
588 if (kvm->debugfs_stat_data) {
589 for (i = 0; i < kvm_debugfs_num_entries; i++)
590 kfree(kvm->debugfs_stat_data[i]);
591 kfree(kvm->debugfs_stat_data);
595 static int kvm_create_vm_debugfs(struct kvm *kvm, int fd)
597 char dir_name[ITOA_MAX_LEN * 2];
598 struct kvm_stat_data *stat_data;
599 struct kvm_stats_debugfs_item *p;
601 if (!debugfs_initialized())
604 snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd);
605 kvm->debugfs_dentry = debugfs_create_dir(dir_name, kvm_debugfs_dir);
607 kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries,
608 sizeof(*kvm->debugfs_stat_data),
610 if (!kvm->debugfs_stat_data)
613 for (p = debugfs_entries; p->name; p++) {
614 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT);
618 stat_data->kvm = kvm;
619 stat_data->offset = p->offset;
620 kvm->debugfs_stat_data[p - debugfs_entries] = stat_data;
621 debugfs_create_file(p->name, 0644, kvm->debugfs_dentry,
622 stat_data, stat_fops_per_vm[p->kind]);
627 static struct kvm *kvm_create_vm(unsigned long type)
630 struct kvm *kvm = kvm_arch_alloc_vm();
633 return ERR_PTR(-ENOMEM);
635 spin_lock_init(&kvm->mmu_lock);
637 kvm->mm = current->mm;
638 kvm_eventfd_init(kvm);
639 mutex_init(&kvm->lock);
640 mutex_init(&kvm->irq_lock);
641 mutex_init(&kvm->slots_lock);
642 refcount_set(&kvm->users_count, 1);
643 INIT_LIST_HEAD(&kvm->devices);
645 r = kvm_arch_init_vm(kvm, type);
647 goto out_err_no_disable;
649 r = hardware_enable_all();
651 goto out_err_no_disable;
653 #ifdef CONFIG_HAVE_KVM_IRQFD
654 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
657 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
660 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
661 struct kvm_memslots *slots = kvm_alloc_memslots();
663 goto out_err_no_srcu;
664 /* Generations must be different for each address space. */
665 slots->generation = i;
666 rcu_assign_pointer(kvm->memslots[i], slots);
669 if (init_srcu_struct(&kvm->srcu))
670 goto out_err_no_srcu;
671 if (init_srcu_struct(&kvm->irq_srcu))
672 goto out_err_no_irq_srcu;
673 for (i = 0; i < KVM_NR_BUSES; i++) {
674 rcu_assign_pointer(kvm->buses[i],
675 kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL_ACCOUNT));
680 r = kvm_init_mmu_notifier(kvm);
684 spin_lock(&kvm_lock);
685 list_add(&kvm->vm_list, &vm_list);
686 spin_unlock(&kvm_lock);
688 preempt_notifier_inc();
693 cleanup_srcu_struct(&kvm->irq_srcu);
695 cleanup_srcu_struct(&kvm->srcu);
697 hardware_disable_all();
699 refcount_set(&kvm->users_count, 0);
700 for (i = 0; i < KVM_NR_BUSES; i++)
701 kfree(kvm_get_bus(kvm, i));
702 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
703 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
704 kvm_arch_free_vm(kvm);
709 static void kvm_destroy_devices(struct kvm *kvm)
711 struct kvm_device *dev, *tmp;
714 * We do not need to take the kvm->lock here, because nobody else
715 * has a reference to the struct kvm at this point and therefore
716 * cannot access the devices list anyhow.
718 list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
719 list_del(&dev->vm_node);
720 dev->ops->destroy(dev);
724 static void kvm_destroy_vm(struct kvm *kvm)
727 struct mm_struct *mm = kvm->mm;
729 kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm);
730 kvm_destroy_vm_debugfs(kvm);
731 kvm_arch_sync_events(kvm);
732 spin_lock(&kvm_lock);
733 list_del(&kvm->vm_list);
734 spin_unlock(&kvm_lock);
735 kvm_free_irq_routing(kvm);
736 for (i = 0; i < KVM_NR_BUSES; i++) {
737 struct kvm_io_bus *bus = kvm_get_bus(kvm, i);
740 kvm_io_bus_destroy(bus);
741 kvm->buses[i] = NULL;
743 kvm_coalesced_mmio_free(kvm);
744 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
745 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
747 kvm_arch_flush_shadow_all(kvm);
749 kvm_arch_destroy_vm(kvm);
750 kvm_destroy_devices(kvm);
751 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
752 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
753 cleanup_srcu_struct(&kvm->irq_srcu);
754 cleanup_srcu_struct(&kvm->srcu);
755 kvm_arch_free_vm(kvm);
756 preempt_notifier_dec();
757 hardware_disable_all();
761 void kvm_get_kvm(struct kvm *kvm)
763 refcount_inc(&kvm->users_count);
765 EXPORT_SYMBOL_GPL(kvm_get_kvm);
767 void kvm_put_kvm(struct kvm *kvm)
769 if (refcount_dec_and_test(&kvm->users_count))
772 EXPORT_SYMBOL_GPL(kvm_put_kvm);
775 static int kvm_vm_release(struct inode *inode, struct file *filp)
777 struct kvm *kvm = filp->private_data;
779 kvm_irqfd_release(kvm);
786 * Allocation size is twice as large as the actual dirty bitmap size.
787 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
789 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
791 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
793 memslot->dirty_bitmap = kvzalloc(dirty_bytes, GFP_KERNEL_ACCOUNT);
794 if (!memslot->dirty_bitmap)
801 * Insert memslot and re-sort memslots based on their GFN,
802 * so binary search could be used to lookup GFN.
803 * Sorting algorithm takes advantage of having initially
804 * sorted array and known changed memslot position.
806 static void update_memslots(struct kvm_memslots *slots,
807 struct kvm_memory_slot *new,
808 enum kvm_mr_change change)
811 int i = slots->id_to_index[id];
812 struct kvm_memory_slot *mslots = slots->memslots;
814 WARN_ON(mslots[i].id != id);
818 WARN_ON(mslots[i].npages || !new->npages);
822 WARN_ON(new->npages || !mslots[i].npages);
828 while (i < KVM_MEM_SLOTS_NUM - 1 &&
829 new->base_gfn <= mslots[i + 1].base_gfn) {
830 if (!mslots[i + 1].npages)
832 mslots[i] = mslots[i + 1];
833 slots->id_to_index[mslots[i].id] = i;
838 * The ">=" is needed when creating a slot with base_gfn == 0,
839 * so that it moves before all those with base_gfn == npages == 0.
841 * On the other hand, if new->npages is zero, the above loop has
842 * already left i pointing to the beginning of the empty part of
843 * mslots, and the ">=" would move the hole backwards in this
844 * case---which is wrong. So skip the loop when deleting a slot.
848 new->base_gfn >= mslots[i - 1].base_gfn) {
849 mslots[i] = mslots[i - 1];
850 slots->id_to_index[mslots[i].id] = i;
854 WARN_ON_ONCE(i != slots->used_slots);
857 slots->id_to_index[mslots[i].id] = i;
860 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
862 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
864 #ifdef __KVM_HAVE_READONLY_MEM
865 valid_flags |= KVM_MEM_READONLY;
868 if (mem->flags & ~valid_flags)
874 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
875 int as_id, struct kvm_memslots *slots)
877 struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id);
878 u64 gen = old_memslots->generation;
880 WARN_ON(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS);
881 slots->generation = gen | KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
883 rcu_assign_pointer(kvm->memslots[as_id], slots);
884 synchronize_srcu_expedited(&kvm->srcu);
887 * Increment the new memslot generation a second time, dropping the
888 * update in-progress flag and incrementing then generation based on
889 * the number of address spaces. This provides a unique and easily
890 * identifiable generation number while the memslots are in flux.
892 gen = slots->generation & ~KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
895 * Generations must be unique even across address spaces. We do not need
896 * a global counter for that, instead the generation space is evenly split
897 * across address spaces. For example, with two address spaces, address
898 * space 0 will use generations 0, 2, 4, ... while address space 1 will
899 * use generations 1, 3, 5, ...
901 gen += KVM_ADDRESS_SPACE_NUM;
903 kvm_arch_memslots_updated(kvm, gen);
905 slots->generation = gen;
911 * Allocate some memory and give it an address in the guest physical address
914 * Discontiguous memory is allowed, mostly for framebuffers.
916 * Must be called holding kvm->slots_lock for write.
918 int __kvm_set_memory_region(struct kvm *kvm,
919 const struct kvm_userspace_memory_region *mem)
923 unsigned long npages;
924 struct kvm_memory_slot *slot;
925 struct kvm_memory_slot old, new;
926 struct kvm_memslots *slots = NULL, *old_memslots;
928 enum kvm_mr_change change;
930 r = check_memory_region_flags(mem);
935 as_id = mem->slot >> 16;
938 /* General sanity checks */
939 if (mem->memory_size & (PAGE_SIZE - 1))
941 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
943 /* We can read the guest memory with __xxx_user() later on. */
944 if ((id < KVM_USER_MEM_SLOTS) &&
945 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
946 !access_ok((void __user *)(unsigned long)mem->userspace_addr,
949 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
951 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
954 slot = id_to_memslot(__kvm_memslots(kvm, as_id), id);
955 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
956 npages = mem->memory_size >> PAGE_SHIFT;
958 if (npages > KVM_MEM_MAX_NR_PAGES)
964 new.base_gfn = base_gfn;
966 new.flags = mem->flags;
970 change = KVM_MR_CREATE;
971 else { /* Modify an existing slot. */
972 if ((mem->userspace_addr != old.userspace_addr) ||
973 (npages != old.npages) ||
974 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
977 if (base_gfn != old.base_gfn)
978 change = KVM_MR_MOVE;
979 else if (new.flags != old.flags)
980 change = KVM_MR_FLAGS_ONLY;
981 else { /* Nothing to change. */
990 change = KVM_MR_DELETE;
995 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
996 /* Check for overlaps */
998 kvm_for_each_memslot(slot, __kvm_memslots(kvm, as_id)) {
1001 if (!((base_gfn + npages <= slot->base_gfn) ||
1002 (base_gfn >= slot->base_gfn + slot->npages)))
1007 /* Free page dirty bitmap if unneeded */
1008 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
1009 new.dirty_bitmap = NULL;
1012 if (change == KVM_MR_CREATE) {
1013 new.userspace_addr = mem->userspace_addr;
1015 if (kvm_arch_create_memslot(kvm, &new, npages))
1019 /* Allocate page dirty bitmap if needed */
1020 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
1021 if (kvm_create_dirty_bitmap(&new) < 0)
1025 slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL_ACCOUNT);
1028 memcpy(slots, __kvm_memslots(kvm, as_id), sizeof(struct kvm_memslots));
1030 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
1031 slot = id_to_memslot(slots, id);
1032 slot->flags |= KVM_MEMSLOT_INVALID;
1034 old_memslots = install_new_memslots(kvm, as_id, slots);
1036 /* From this point no new shadow pages pointing to a deleted,
1037 * or moved, memslot will be created.
1039 * validation of sp->gfn happens in:
1040 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1041 * - kvm_is_visible_gfn (mmu_check_roots)
1043 kvm_arch_flush_shadow_memslot(kvm, slot);
1046 * We can re-use the old_memslots from above, the only difference
1047 * from the currently installed memslots is the invalid flag. This
1048 * will get overwritten by update_memslots anyway.
1050 slots = old_memslots;
1053 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
1057 /* actual memory is freed via old in kvm_free_memslot below */
1058 if (change == KVM_MR_DELETE) {
1059 new.dirty_bitmap = NULL;
1060 memset(&new.arch, 0, sizeof(new.arch));
1063 update_memslots(slots, &new, change);
1064 old_memslots = install_new_memslots(kvm, as_id, slots);
1066 kvm_arch_commit_memory_region(kvm, mem, &old, &new, change);
1068 kvm_free_memslot(kvm, &old, &new);
1069 kvfree(old_memslots);
1075 kvm_free_memslot(kvm, &new, &old);
1079 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1081 int kvm_set_memory_region(struct kvm *kvm,
1082 const struct kvm_userspace_memory_region *mem)
1086 mutex_lock(&kvm->slots_lock);
1087 r = __kvm_set_memory_region(kvm, mem);
1088 mutex_unlock(&kvm->slots_lock);
1091 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1093 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1094 struct kvm_userspace_memory_region *mem)
1096 if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1099 return kvm_set_memory_region(kvm, mem);
1102 int kvm_get_dirty_log(struct kvm *kvm,
1103 struct kvm_dirty_log *log, int *is_dirty)
1105 struct kvm_memslots *slots;
1106 struct kvm_memory_slot *memslot;
1109 unsigned long any = 0;
1111 as_id = log->slot >> 16;
1112 id = (u16)log->slot;
1113 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1116 slots = __kvm_memslots(kvm, as_id);
1117 memslot = id_to_memslot(slots, id);
1118 if (!memslot->dirty_bitmap)
1121 n = kvm_dirty_bitmap_bytes(memslot);
1123 for (i = 0; !any && i < n/sizeof(long); ++i)
1124 any = memslot->dirty_bitmap[i];
1126 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1133 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1135 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1137 * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages
1138 * and reenable dirty page tracking for the corresponding pages.
1139 * @kvm: pointer to kvm instance
1140 * @log: slot id and address to which we copy the log
1141 * @is_dirty: flag set if any page is dirty
1143 * We need to keep it in mind that VCPU threads can write to the bitmap
1144 * concurrently. So, to avoid losing track of dirty pages we keep the
1147 * 1. Take a snapshot of the bit and clear it if needed.
1148 * 2. Write protect the corresponding page.
1149 * 3. Copy the snapshot to the userspace.
1150 * 4. Upon return caller flushes TLB's if needed.
1152 * Between 2 and 4, the guest may write to the page using the remaining TLB
1153 * entry. This is not a problem because the page is reported dirty using
1154 * the snapshot taken before and step 4 ensures that writes done after
1155 * exiting to userspace will be logged for the next call.
1158 int kvm_get_dirty_log_protect(struct kvm *kvm,
1159 struct kvm_dirty_log *log, bool *flush)
1161 struct kvm_memslots *slots;
1162 struct kvm_memory_slot *memslot;
1165 unsigned long *dirty_bitmap;
1166 unsigned long *dirty_bitmap_buffer;
1168 as_id = log->slot >> 16;
1169 id = (u16)log->slot;
1170 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1173 slots = __kvm_memslots(kvm, as_id);
1174 memslot = id_to_memslot(slots, id);
1176 dirty_bitmap = memslot->dirty_bitmap;
1180 n = kvm_dirty_bitmap_bytes(memslot);
1182 if (kvm->manual_dirty_log_protect) {
1184 * Unlike kvm_get_dirty_log, we always return false in *flush,
1185 * because no flush is needed until KVM_CLEAR_DIRTY_LOG. There
1186 * is some code duplication between this function and
1187 * kvm_get_dirty_log, but hopefully all architecture
1188 * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log
1189 * can be eliminated.
1191 dirty_bitmap_buffer = dirty_bitmap;
1193 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
1194 memset(dirty_bitmap_buffer, 0, n);
1196 spin_lock(&kvm->mmu_lock);
1197 for (i = 0; i < n / sizeof(long); i++) {
1201 if (!dirty_bitmap[i])
1205 mask = xchg(&dirty_bitmap[i], 0);
1206 dirty_bitmap_buffer[i] = mask;
1208 offset = i * BITS_PER_LONG;
1209 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1212 spin_unlock(&kvm->mmu_lock);
1215 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1219 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect);
1222 * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap
1223 * and reenable dirty page tracking for the corresponding pages.
1224 * @kvm: pointer to kvm instance
1225 * @log: slot id and address from which to fetch the bitmap of dirty pages
1227 int kvm_clear_dirty_log_protect(struct kvm *kvm,
1228 struct kvm_clear_dirty_log *log, bool *flush)
1230 struct kvm_memslots *slots;
1231 struct kvm_memory_slot *memslot;
1235 unsigned long *dirty_bitmap;
1236 unsigned long *dirty_bitmap_buffer;
1238 as_id = log->slot >> 16;
1239 id = (u16)log->slot;
1240 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1243 if ((log->first_page & 63) || (log->num_pages & 63))
1246 slots = __kvm_memslots(kvm, as_id);
1247 memslot = id_to_memslot(slots, id);
1249 dirty_bitmap = memslot->dirty_bitmap;
1253 n = kvm_dirty_bitmap_bytes(memslot);
1255 if (log->first_page > memslot->npages ||
1256 log->num_pages > memslot->npages - log->first_page)
1260 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
1261 if (copy_from_user(dirty_bitmap_buffer, log->dirty_bitmap, n))
1264 spin_lock(&kvm->mmu_lock);
1265 for (offset = log->first_page,
1266 i = offset / BITS_PER_LONG, n = log->num_pages / BITS_PER_LONG; n--;
1267 i++, offset += BITS_PER_LONG) {
1268 unsigned long mask = *dirty_bitmap_buffer++;
1269 atomic_long_t *p = (atomic_long_t *) &dirty_bitmap[i];
1273 mask &= atomic_long_fetch_andnot(mask, p);
1276 * mask contains the bits that really have been cleared. This
1277 * never includes any bits beyond the length of the memslot (if
1278 * the length is not aligned to 64 pages), therefore it is not
1279 * a problem if userspace sets them in log->dirty_bitmap.
1283 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1287 spin_unlock(&kvm->mmu_lock);
1291 EXPORT_SYMBOL_GPL(kvm_clear_dirty_log_protect);
1294 bool kvm_largepages_enabled(void)
1296 return largepages_enabled;
1299 void kvm_disable_largepages(void)
1301 largepages_enabled = false;
1303 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1305 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1307 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1309 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1311 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
1313 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn);
1316 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1318 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1320 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1321 memslot->flags & KVM_MEMSLOT_INVALID)
1326 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1328 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1330 struct vm_area_struct *vma;
1331 unsigned long addr, size;
1335 addr = gfn_to_hva(kvm, gfn);
1336 if (kvm_is_error_hva(addr))
1339 down_read(¤t->mm->mmap_sem);
1340 vma = find_vma(current->mm, addr);
1344 size = vma_kernel_pagesize(vma);
1347 up_read(¤t->mm->mmap_sem);
1352 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1354 return slot->flags & KVM_MEM_READONLY;
1357 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1358 gfn_t *nr_pages, bool write)
1360 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1361 return KVM_HVA_ERR_BAD;
1363 if (memslot_is_readonly(slot) && write)
1364 return KVM_HVA_ERR_RO_BAD;
1367 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1369 return __gfn_to_hva_memslot(slot, gfn);
1372 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1375 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1378 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1381 return gfn_to_hva_many(slot, gfn, NULL);
1383 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1385 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1387 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1389 EXPORT_SYMBOL_GPL(gfn_to_hva);
1391 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
1393 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
1395 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
1398 * Return the hva of a @gfn and the R/W attribute if possible.
1400 * @slot: the kvm_memory_slot which contains @gfn
1401 * @gfn: the gfn to be translated
1402 * @writable: used to return the read/write attribute of the @slot if the hva
1403 * is valid and @writable is not NULL
1405 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1406 gfn_t gfn, bool *writable)
1408 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1410 if (!kvm_is_error_hva(hva) && writable)
1411 *writable = !memslot_is_readonly(slot);
1416 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1418 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1420 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1423 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
1425 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1427 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1430 static inline int check_user_page_hwpoison(unsigned long addr)
1432 int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
1434 rc = get_user_pages(addr, 1, flags, NULL, NULL);
1435 return rc == -EHWPOISON;
1439 * The fast path to get the writable pfn which will be stored in @pfn,
1440 * true indicates success, otherwise false is returned. It's also the
1441 * only part that runs if we can are in atomic context.
1443 static bool hva_to_pfn_fast(unsigned long addr, bool write_fault,
1444 bool *writable, kvm_pfn_t *pfn)
1446 struct page *page[1];
1450 * Fast pin a writable pfn only if it is a write fault request
1451 * or the caller allows to map a writable pfn for a read fault
1454 if (!(write_fault || writable))
1457 npages = __get_user_pages_fast(addr, 1, 1, page);
1459 *pfn = page_to_pfn(page[0]);
1470 * The slow path to get the pfn of the specified host virtual address,
1471 * 1 indicates success, -errno is returned if error is detected.
1473 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1474 bool *writable, kvm_pfn_t *pfn)
1476 unsigned int flags = FOLL_HWPOISON;
1483 *writable = write_fault;
1486 flags |= FOLL_WRITE;
1488 flags |= FOLL_NOWAIT;
1490 npages = get_user_pages_unlocked(addr, 1, &page, flags);
1494 /* map read fault as writable if possible */
1495 if (unlikely(!write_fault) && writable) {
1498 if (__get_user_pages_fast(addr, 1, 1, &wpage) == 1) {
1504 *pfn = page_to_pfn(page);
1508 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1510 if (unlikely(!(vma->vm_flags & VM_READ)))
1513 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1519 static int hva_to_pfn_remapped(struct vm_area_struct *vma,
1520 unsigned long addr, bool *async,
1521 bool write_fault, bool *writable,
1527 r = follow_pfn(vma, addr, &pfn);
1530 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
1531 * not call the fault handler, so do it here.
1533 bool unlocked = false;
1534 r = fixup_user_fault(current, current->mm, addr,
1535 (write_fault ? FAULT_FLAG_WRITE : 0),
1542 r = follow_pfn(vma, addr, &pfn);
1552 * Get a reference here because callers of *hva_to_pfn* and
1553 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
1554 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
1555 * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will
1556 * simply do nothing for reserved pfns.
1558 * Whoever called remap_pfn_range is also going to call e.g.
1559 * unmap_mapping_range before the underlying pages are freed,
1560 * causing a call to our MMU notifier.
1569 * Pin guest page in memory and return its pfn.
1570 * @addr: host virtual address which maps memory to the guest
1571 * @atomic: whether this function can sleep
1572 * @async: whether this function need to wait IO complete if the
1573 * host page is not in the memory
1574 * @write_fault: whether we should get a writable host page
1575 * @writable: whether it allows to map a writable host page for !@write_fault
1577 * The function will map a writable host page for these two cases:
1578 * 1): @write_fault = true
1579 * 2): @write_fault = false && @writable, @writable will tell the caller
1580 * whether the mapping is writable.
1582 static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1583 bool write_fault, bool *writable)
1585 struct vm_area_struct *vma;
1589 /* we can do it either atomically or asynchronously, not both */
1590 BUG_ON(atomic && async);
1592 if (hva_to_pfn_fast(addr, write_fault, writable, &pfn))
1596 return KVM_PFN_ERR_FAULT;
1598 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1602 down_read(¤t->mm->mmap_sem);
1603 if (npages == -EHWPOISON ||
1604 (!async && check_user_page_hwpoison(addr))) {
1605 pfn = KVM_PFN_ERR_HWPOISON;
1610 vma = find_vma_intersection(current->mm, addr, addr + 1);
1613 pfn = KVM_PFN_ERR_FAULT;
1614 else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
1615 r = hva_to_pfn_remapped(vma, addr, async, write_fault, writable, &pfn);
1619 pfn = KVM_PFN_ERR_FAULT;
1621 if (async && vma_is_valid(vma, write_fault))
1623 pfn = KVM_PFN_ERR_FAULT;
1626 up_read(¤t->mm->mmap_sem);
1630 kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn,
1631 bool atomic, bool *async, bool write_fault,
1634 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1636 if (addr == KVM_HVA_ERR_RO_BAD) {
1639 return KVM_PFN_ERR_RO_FAULT;
1642 if (kvm_is_error_hva(addr)) {
1645 return KVM_PFN_NOSLOT;
1648 /* Do not map writable pfn in the readonly memslot. */
1649 if (writable && memslot_is_readonly(slot)) {
1654 return hva_to_pfn(addr, atomic, async, write_fault,
1657 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
1659 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1662 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
1663 write_fault, writable);
1665 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1667 kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1669 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1671 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
1673 kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1675 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1677 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1679 kvm_pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1681 return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm, gfn), gfn);
1683 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1685 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
1687 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1689 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
1691 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1693 return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
1695 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1697 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
1699 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1701 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
1703 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1704 struct page **pages, int nr_pages)
1709 addr = gfn_to_hva_many(slot, gfn, &entry);
1710 if (kvm_is_error_hva(addr))
1713 if (entry < nr_pages)
1716 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1718 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1720 static struct page *kvm_pfn_to_page(kvm_pfn_t pfn)
1722 if (is_error_noslot_pfn(pfn))
1723 return KVM_ERR_PTR_BAD_PAGE;
1725 if (kvm_is_reserved_pfn(pfn)) {
1727 return KVM_ERR_PTR_BAD_PAGE;
1730 return pfn_to_page(pfn);
1733 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1737 pfn = gfn_to_pfn(kvm, gfn);
1739 return kvm_pfn_to_page(pfn);
1741 EXPORT_SYMBOL_GPL(gfn_to_page);
1743 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
1747 pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
1749 return kvm_pfn_to_page(pfn);
1751 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
1753 void kvm_release_page_clean(struct page *page)
1755 WARN_ON(is_error_page(page));
1757 kvm_release_pfn_clean(page_to_pfn(page));
1759 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1761 void kvm_release_pfn_clean(kvm_pfn_t pfn)
1763 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
1764 put_page(pfn_to_page(pfn));
1766 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1768 void kvm_release_page_dirty(struct page *page)
1770 WARN_ON(is_error_page(page));
1772 kvm_release_pfn_dirty(page_to_pfn(page));
1774 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1776 void kvm_release_pfn_dirty(kvm_pfn_t pfn)
1778 kvm_set_pfn_dirty(pfn);
1779 kvm_release_pfn_clean(pfn);
1781 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1783 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
1785 if (!kvm_is_reserved_pfn(pfn)) {
1786 struct page *page = pfn_to_page(pfn);
1788 if (!PageReserved(page))
1792 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1794 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
1796 if (!kvm_is_reserved_pfn(pfn))
1797 mark_page_accessed(pfn_to_page(pfn));
1799 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1801 void kvm_get_pfn(kvm_pfn_t pfn)
1803 if (!kvm_is_reserved_pfn(pfn))
1804 get_page(pfn_to_page(pfn));
1806 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1808 static int next_segment(unsigned long len, int offset)
1810 if (len > PAGE_SIZE - offset)
1811 return PAGE_SIZE - offset;
1816 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
1817 void *data, int offset, int len)
1822 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1823 if (kvm_is_error_hva(addr))
1825 r = __copy_from_user(data, (void __user *)addr + offset, len);
1831 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1834 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1836 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1838 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1840 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
1841 int offset, int len)
1843 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1845 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1847 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
1849 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1851 gfn_t gfn = gpa >> PAGE_SHIFT;
1853 int offset = offset_in_page(gpa);
1856 while ((seg = next_segment(len, offset)) != 0) {
1857 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1867 EXPORT_SYMBOL_GPL(kvm_read_guest);
1869 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
1871 gfn_t gfn = gpa >> PAGE_SHIFT;
1873 int offset = offset_in_page(gpa);
1876 while ((seg = next_segment(len, offset)) != 0) {
1877 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
1887 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
1889 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1890 void *data, int offset, unsigned long len)
1895 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1896 if (kvm_is_error_hva(addr))
1898 pagefault_disable();
1899 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1906 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1909 gfn_t gfn = gpa >> PAGE_SHIFT;
1910 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1911 int offset = offset_in_page(gpa);
1913 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1915 EXPORT_SYMBOL_GPL(kvm_read_guest_atomic);
1917 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
1918 void *data, unsigned long len)
1920 gfn_t gfn = gpa >> PAGE_SHIFT;
1921 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1922 int offset = offset_in_page(gpa);
1924 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1926 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
1928 static int __kvm_write_guest_page(struct kvm_memory_slot *memslot, gfn_t gfn,
1929 const void *data, int offset, int len)
1934 addr = gfn_to_hva_memslot(memslot, gfn);
1935 if (kvm_is_error_hva(addr))
1937 r = __copy_to_user((void __user *)addr + offset, data, len);
1940 mark_page_dirty_in_slot(memslot, gfn);
1944 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
1945 const void *data, int offset, int len)
1947 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1949 return __kvm_write_guest_page(slot, gfn, data, offset, len);
1951 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1953 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
1954 const void *data, int offset, int len)
1956 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1958 return __kvm_write_guest_page(slot, gfn, data, offset, len);
1960 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
1962 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1965 gfn_t gfn = gpa >> PAGE_SHIFT;
1967 int offset = offset_in_page(gpa);
1970 while ((seg = next_segment(len, offset)) != 0) {
1971 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1981 EXPORT_SYMBOL_GPL(kvm_write_guest);
1983 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
1986 gfn_t gfn = gpa >> PAGE_SHIFT;
1988 int offset = offset_in_page(gpa);
1991 while ((seg = next_segment(len, offset)) != 0) {
1992 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
2002 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
2004 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
2005 struct gfn_to_hva_cache *ghc,
2006 gpa_t gpa, unsigned long len)
2008 int offset = offset_in_page(gpa);
2009 gfn_t start_gfn = gpa >> PAGE_SHIFT;
2010 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
2011 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
2012 gfn_t nr_pages_avail;
2013 int r = start_gfn <= end_gfn ? 0 : -EINVAL;
2016 ghc->generation = slots->generation;
2018 ghc->hva = KVM_HVA_ERR_BAD;
2021 * If the requested region crosses two memslots, we still
2022 * verify that the entire region is valid here.
2024 while (!r && start_gfn <= end_gfn) {
2025 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
2026 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
2028 if (kvm_is_error_hva(ghc->hva))
2030 start_gfn += nr_pages_avail;
2033 /* Use the slow path for cross page reads and writes. */
2034 if (!r && nr_pages_needed == 1)
2037 ghc->memslot = NULL;
2042 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2043 gpa_t gpa, unsigned long len)
2045 struct kvm_memslots *slots = kvm_memslots(kvm);
2046 return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len);
2048 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
2050 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2051 void *data, unsigned int offset,
2054 struct kvm_memslots *slots = kvm_memslots(kvm);
2056 gpa_t gpa = ghc->gpa + offset;
2058 BUG_ON(len + offset > ghc->len);
2060 if (slots->generation != ghc->generation)
2061 __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len);
2063 if (unlikely(!ghc->memslot))
2064 return kvm_write_guest(kvm, gpa, data, len);
2066 if (kvm_is_error_hva(ghc->hva))
2069 r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
2072 mark_page_dirty_in_slot(ghc->memslot, gpa >> PAGE_SHIFT);
2076 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached);
2078 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2079 void *data, unsigned long len)
2081 return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len);
2083 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
2085 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2086 void *data, unsigned long len)
2088 struct kvm_memslots *slots = kvm_memslots(kvm);
2091 BUG_ON(len > ghc->len);
2093 if (slots->generation != ghc->generation)
2094 __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len);
2096 if (unlikely(!ghc->memslot))
2097 return kvm_read_guest(kvm, ghc->gpa, data, len);
2099 if (kvm_is_error_hva(ghc->hva))
2102 r = __copy_from_user(data, (void __user *)ghc->hva, len);
2108 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
2110 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
2112 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
2114 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
2116 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
2118 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
2120 gfn_t gfn = gpa >> PAGE_SHIFT;
2122 int offset = offset_in_page(gpa);
2125 while ((seg = next_segment(len, offset)) != 0) {
2126 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
2135 EXPORT_SYMBOL_GPL(kvm_clear_guest);
2137 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot,
2140 if (memslot && memslot->dirty_bitmap) {
2141 unsigned long rel_gfn = gfn - memslot->base_gfn;
2143 set_bit_le(rel_gfn, memslot->dirty_bitmap);
2147 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
2149 struct kvm_memory_slot *memslot;
2151 memslot = gfn_to_memslot(kvm, gfn);
2152 mark_page_dirty_in_slot(memslot, gfn);
2154 EXPORT_SYMBOL_GPL(mark_page_dirty);
2156 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
2158 struct kvm_memory_slot *memslot;
2160 memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2161 mark_page_dirty_in_slot(memslot, gfn);
2163 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
2165 void kvm_sigset_activate(struct kvm_vcpu *vcpu)
2167 if (!vcpu->sigset_active)
2171 * This does a lockless modification of ->real_blocked, which is fine
2172 * because, only current can change ->real_blocked and all readers of
2173 * ->real_blocked don't care as long ->real_blocked is always a subset
2176 sigprocmask(SIG_SETMASK, &vcpu->sigset, ¤t->real_blocked);
2179 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu)
2181 if (!vcpu->sigset_active)
2184 sigprocmask(SIG_SETMASK, ¤t->real_blocked, NULL);
2185 sigemptyset(¤t->real_blocked);
2188 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
2190 unsigned int old, val, grow, grow_start;
2192 old = val = vcpu->halt_poll_ns;
2193 grow_start = READ_ONCE(halt_poll_ns_grow_start);
2194 grow = READ_ONCE(halt_poll_ns_grow);
2199 if (val < grow_start)
2202 if (val > halt_poll_ns)
2205 vcpu->halt_poll_ns = val;
2207 trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
2210 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
2212 unsigned int old, val, shrink;
2214 old = val = vcpu->halt_poll_ns;
2215 shrink = READ_ONCE(halt_poll_ns_shrink);
2221 vcpu->halt_poll_ns = val;
2222 trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
2225 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
2228 int idx = srcu_read_lock(&vcpu->kvm->srcu);
2230 if (kvm_arch_vcpu_runnable(vcpu)) {
2231 kvm_make_request(KVM_REQ_UNHALT, vcpu);
2234 if (kvm_cpu_has_pending_timer(vcpu))
2236 if (signal_pending(current))
2241 srcu_read_unlock(&vcpu->kvm->srcu, idx);
2246 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2248 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
2251 DECLARE_SWAITQUEUE(wait);
2252 bool waited = false;
2255 start = cur = ktime_get();
2256 if (vcpu->halt_poll_ns) {
2257 ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns);
2259 ++vcpu->stat.halt_attempted_poll;
2262 * This sets KVM_REQ_UNHALT if an interrupt
2265 if (kvm_vcpu_check_block(vcpu) < 0) {
2266 ++vcpu->stat.halt_successful_poll;
2267 if (!vcpu_valid_wakeup(vcpu))
2268 ++vcpu->stat.halt_poll_invalid;
2272 } while (single_task_running() && ktime_before(cur, stop));
2275 kvm_arch_vcpu_blocking(vcpu);
2278 prepare_to_swait_exclusive(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
2280 if (kvm_vcpu_check_block(vcpu) < 0)
2287 finish_swait(&vcpu->wq, &wait);
2290 kvm_arch_vcpu_unblocking(vcpu);
2292 block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
2294 if (!vcpu_valid_wakeup(vcpu))
2295 shrink_halt_poll_ns(vcpu);
2296 else if (halt_poll_ns) {
2297 if (block_ns <= vcpu->halt_poll_ns)
2299 /* we had a long block, shrink polling */
2300 else if (vcpu->halt_poll_ns && block_ns > halt_poll_ns)
2301 shrink_halt_poll_ns(vcpu);
2302 /* we had a short halt and our poll time is too small */
2303 else if (vcpu->halt_poll_ns < halt_poll_ns &&
2304 block_ns < halt_poll_ns)
2305 grow_halt_poll_ns(vcpu);
2307 vcpu->halt_poll_ns = 0;
2309 trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu));
2310 kvm_arch_vcpu_block_finish(vcpu);
2312 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
2314 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
2316 struct swait_queue_head *wqp;
2318 wqp = kvm_arch_vcpu_wq(vcpu);
2319 if (swq_has_sleeper(wqp)) {
2321 ++vcpu->stat.halt_wakeup;
2327 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
2331 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2333 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
2336 int cpu = vcpu->cpu;
2338 if (kvm_vcpu_wake_up(vcpu))
2342 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
2343 if (kvm_arch_vcpu_should_kick(vcpu))
2344 smp_send_reschedule(cpu);
2347 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
2348 #endif /* !CONFIG_S390 */
2350 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
2353 struct task_struct *task = NULL;
2357 pid = rcu_dereference(target->pid);
2359 task = get_pid_task(pid, PIDTYPE_PID);
2363 ret = yield_to(task, 1);
2364 put_task_struct(task);
2368 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
2371 * Helper that checks whether a VCPU is eligible for directed yield.
2372 * Most eligible candidate to yield is decided by following heuristics:
2374 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2375 * (preempted lock holder), indicated by @in_spin_loop.
2376 * Set at the beiginning and cleared at the end of interception/PLE handler.
2378 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2379 * chance last time (mostly it has become eligible now since we have probably
2380 * yielded to lockholder in last iteration. This is done by toggling
2381 * @dy_eligible each time a VCPU checked for eligibility.)
2383 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2384 * to preempted lock-holder could result in wrong VCPU selection and CPU
2385 * burning. Giving priority for a potential lock-holder increases lock
2388 * Since algorithm is based on heuristics, accessing another VCPU data without
2389 * locking does not harm. It may result in trying to yield to same VCPU, fail
2390 * and continue with next VCPU and so on.
2392 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
2394 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2397 eligible = !vcpu->spin_loop.in_spin_loop ||
2398 vcpu->spin_loop.dy_eligible;
2400 if (vcpu->spin_loop.in_spin_loop)
2401 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
2409 void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode)
2411 struct kvm *kvm = me->kvm;
2412 struct kvm_vcpu *vcpu;
2413 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
2419 kvm_vcpu_set_in_spin_loop(me, true);
2421 * We boost the priority of a VCPU that is runnable but not
2422 * currently running, because it got preempted by something
2423 * else and called schedule in __vcpu_run. Hopefully that
2424 * VCPU is holding the lock that we need and will release it.
2425 * We approximate round-robin by starting at the last boosted VCPU.
2427 for (pass = 0; pass < 2 && !yielded && try; pass++) {
2428 kvm_for_each_vcpu(i, vcpu, kvm) {
2429 if (!pass && i <= last_boosted_vcpu) {
2430 i = last_boosted_vcpu;
2432 } else if (pass && i > last_boosted_vcpu)
2434 if (!READ_ONCE(vcpu->preempted))
2438 if (swait_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
2440 if (yield_to_kernel_mode && !kvm_arch_vcpu_in_kernel(vcpu))
2442 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
2445 yielded = kvm_vcpu_yield_to(vcpu);
2447 kvm->last_boosted_vcpu = i;
2449 } else if (yielded < 0) {
2456 kvm_vcpu_set_in_spin_loop(me, false);
2458 /* Ensure vcpu is not eligible during next spinloop */
2459 kvm_vcpu_set_dy_eligible(me, false);
2461 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
2463 static vm_fault_t kvm_vcpu_fault(struct vm_fault *vmf)
2465 struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data;
2468 if (vmf->pgoff == 0)
2469 page = virt_to_page(vcpu->run);
2471 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
2472 page = virt_to_page(vcpu->arch.pio_data);
2474 #ifdef CONFIG_KVM_MMIO
2475 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
2476 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
2479 return kvm_arch_vcpu_fault(vcpu, vmf);
2485 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
2486 .fault = kvm_vcpu_fault,
2489 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2491 vma->vm_ops = &kvm_vcpu_vm_ops;
2495 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2497 struct kvm_vcpu *vcpu = filp->private_data;
2499 debugfs_remove_recursive(vcpu->debugfs_dentry);
2500 kvm_put_kvm(vcpu->kvm);
2504 static struct file_operations kvm_vcpu_fops = {
2505 .release = kvm_vcpu_release,
2506 .unlocked_ioctl = kvm_vcpu_ioctl,
2507 .mmap = kvm_vcpu_mmap,
2508 .llseek = noop_llseek,
2509 KVM_COMPAT(kvm_vcpu_compat_ioctl),
2513 * Allocates an inode for the vcpu.
2515 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2517 char name[8 + 1 + ITOA_MAX_LEN + 1];
2519 snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id);
2520 return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
2523 static int kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
2525 char dir_name[ITOA_MAX_LEN * 2];
2528 if (!kvm_arch_has_vcpu_debugfs())
2531 if (!debugfs_initialized())
2534 snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
2535 vcpu->debugfs_dentry = debugfs_create_dir(dir_name,
2536 vcpu->kvm->debugfs_dentry);
2537 if (!vcpu->debugfs_dentry)
2540 ret = kvm_arch_create_vcpu_debugfs(vcpu);
2542 debugfs_remove_recursive(vcpu->debugfs_dentry);
2550 * Creates some virtual cpus. Good luck creating more than one.
2552 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
2555 struct kvm_vcpu *vcpu;
2557 if (id >= KVM_MAX_VCPU_ID)
2560 mutex_lock(&kvm->lock);
2561 if (kvm->created_vcpus == KVM_MAX_VCPUS) {
2562 mutex_unlock(&kvm->lock);
2566 kvm->created_vcpus++;
2567 mutex_unlock(&kvm->lock);
2569 vcpu = kvm_arch_vcpu_create(kvm, id);
2572 goto vcpu_decrement;
2575 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2577 r = kvm_arch_vcpu_setup(vcpu);
2581 r = kvm_create_vcpu_debugfs(vcpu);
2585 mutex_lock(&kvm->lock);
2586 if (kvm_get_vcpu_by_id(kvm, id)) {
2588 goto unlock_vcpu_destroy;
2591 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
2593 /* Now it's all set up, let userspace reach it */
2595 r = create_vcpu_fd(vcpu);
2598 goto unlock_vcpu_destroy;
2601 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
2604 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
2605 * before kvm->online_vcpu's incremented value.
2608 atomic_inc(&kvm->online_vcpus);
2610 mutex_unlock(&kvm->lock);
2611 kvm_arch_vcpu_postcreate(vcpu);
2614 unlock_vcpu_destroy:
2615 mutex_unlock(&kvm->lock);
2616 debugfs_remove_recursive(vcpu->debugfs_dentry);
2618 kvm_arch_vcpu_destroy(vcpu);
2620 mutex_lock(&kvm->lock);
2621 kvm->created_vcpus--;
2622 mutex_unlock(&kvm->lock);
2626 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2629 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2630 vcpu->sigset_active = 1;
2631 vcpu->sigset = *sigset;
2633 vcpu->sigset_active = 0;
2637 static long kvm_vcpu_ioctl(struct file *filp,
2638 unsigned int ioctl, unsigned long arg)
2640 struct kvm_vcpu *vcpu = filp->private_data;
2641 void __user *argp = (void __user *)arg;
2643 struct kvm_fpu *fpu = NULL;
2644 struct kvm_sregs *kvm_sregs = NULL;
2646 if (vcpu->kvm->mm != current->mm)
2649 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2653 * Some architectures have vcpu ioctls that are asynchronous to vcpu
2654 * execution; mutex_lock() would break them.
2656 r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg);
2657 if (r != -ENOIOCTLCMD)
2660 if (mutex_lock_killable(&vcpu->mutex))
2668 oldpid = rcu_access_pointer(vcpu->pid);
2669 if (unlikely(oldpid != task_pid(current))) {
2670 /* The thread running this VCPU changed. */
2673 r = kvm_arch_vcpu_run_pid_change(vcpu);
2677 newpid = get_task_pid(current, PIDTYPE_PID);
2678 rcu_assign_pointer(vcpu->pid, newpid);
2683 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2684 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2687 case KVM_GET_REGS: {
2688 struct kvm_regs *kvm_regs;
2691 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL_ACCOUNT);
2694 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2698 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2705 case KVM_SET_REGS: {
2706 struct kvm_regs *kvm_regs;
2709 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2710 if (IS_ERR(kvm_regs)) {
2711 r = PTR_ERR(kvm_regs);
2714 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2718 case KVM_GET_SREGS: {
2719 kvm_sregs = kzalloc(sizeof(struct kvm_sregs),
2720 GFP_KERNEL_ACCOUNT);
2724 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2728 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2733 case KVM_SET_SREGS: {
2734 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2735 if (IS_ERR(kvm_sregs)) {
2736 r = PTR_ERR(kvm_sregs);
2740 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2743 case KVM_GET_MP_STATE: {
2744 struct kvm_mp_state mp_state;
2746 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2750 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
2755 case KVM_SET_MP_STATE: {
2756 struct kvm_mp_state mp_state;
2759 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
2761 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2764 case KVM_TRANSLATE: {
2765 struct kvm_translation tr;
2768 if (copy_from_user(&tr, argp, sizeof(tr)))
2770 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2774 if (copy_to_user(argp, &tr, sizeof(tr)))
2779 case KVM_SET_GUEST_DEBUG: {
2780 struct kvm_guest_debug dbg;
2783 if (copy_from_user(&dbg, argp, sizeof(dbg)))
2785 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2788 case KVM_SET_SIGNAL_MASK: {
2789 struct kvm_signal_mask __user *sigmask_arg = argp;
2790 struct kvm_signal_mask kvm_sigmask;
2791 sigset_t sigset, *p;
2796 if (copy_from_user(&kvm_sigmask, argp,
2797 sizeof(kvm_sigmask)))
2800 if (kvm_sigmask.len != sizeof(sigset))
2803 if (copy_from_user(&sigset, sigmask_arg->sigset,
2808 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2812 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL_ACCOUNT);
2816 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2820 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2826 fpu = memdup_user(argp, sizeof(*fpu));
2832 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2836 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2839 mutex_unlock(&vcpu->mutex);
2845 #ifdef CONFIG_KVM_COMPAT
2846 static long kvm_vcpu_compat_ioctl(struct file *filp,
2847 unsigned int ioctl, unsigned long arg)
2849 struct kvm_vcpu *vcpu = filp->private_data;
2850 void __user *argp = compat_ptr(arg);
2853 if (vcpu->kvm->mm != current->mm)
2857 case KVM_SET_SIGNAL_MASK: {
2858 struct kvm_signal_mask __user *sigmask_arg = argp;
2859 struct kvm_signal_mask kvm_sigmask;
2864 if (copy_from_user(&kvm_sigmask, argp,
2865 sizeof(kvm_sigmask)))
2868 if (kvm_sigmask.len != sizeof(compat_sigset_t))
2871 if (get_compat_sigset(&sigset, (void *)sigmask_arg->sigset))
2873 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2875 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2879 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2887 static int kvm_device_mmap(struct file *filp, struct vm_area_struct *vma)
2889 struct kvm_device *dev = filp->private_data;
2892 return dev->ops->mmap(dev, vma);
2897 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2898 int (*accessor)(struct kvm_device *dev,
2899 struct kvm_device_attr *attr),
2902 struct kvm_device_attr attr;
2907 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2910 return accessor(dev, &attr);
2913 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2916 struct kvm_device *dev = filp->private_data;
2918 if (dev->kvm->mm != current->mm)
2922 case KVM_SET_DEVICE_ATTR:
2923 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2924 case KVM_GET_DEVICE_ATTR:
2925 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2926 case KVM_HAS_DEVICE_ATTR:
2927 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2929 if (dev->ops->ioctl)
2930 return dev->ops->ioctl(dev, ioctl, arg);
2936 static int kvm_device_release(struct inode *inode, struct file *filp)
2938 struct kvm_device *dev = filp->private_data;
2939 struct kvm *kvm = dev->kvm;
2941 if (dev->ops->release) {
2942 mutex_lock(&kvm->lock);
2943 list_del(&dev->vm_node);
2944 dev->ops->release(dev);
2945 mutex_unlock(&kvm->lock);
2952 static const struct file_operations kvm_device_fops = {
2953 .unlocked_ioctl = kvm_device_ioctl,
2954 .release = kvm_device_release,
2955 KVM_COMPAT(kvm_device_ioctl),
2956 .mmap = kvm_device_mmap,
2959 struct kvm_device *kvm_device_from_filp(struct file *filp)
2961 if (filp->f_op != &kvm_device_fops)
2964 return filp->private_data;
2967 static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
2968 #ifdef CONFIG_KVM_MPIC
2969 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
2970 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
2974 int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
2976 if (type >= ARRAY_SIZE(kvm_device_ops_table))
2979 if (kvm_device_ops_table[type] != NULL)
2982 kvm_device_ops_table[type] = ops;
2986 void kvm_unregister_device_ops(u32 type)
2988 if (kvm_device_ops_table[type] != NULL)
2989 kvm_device_ops_table[type] = NULL;
2992 static int kvm_ioctl_create_device(struct kvm *kvm,
2993 struct kvm_create_device *cd)
2995 struct kvm_device_ops *ops = NULL;
2996 struct kvm_device *dev;
2997 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
3000 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
3003 ops = kvm_device_ops_table[cd->type];
3010 dev = kzalloc(sizeof(*dev), GFP_KERNEL_ACCOUNT);
3017 mutex_lock(&kvm->lock);
3018 ret = ops->create(dev, cd->type);
3020 mutex_unlock(&kvm->lock);
3024 list_add(&dev->vm_node, &kvm->devices);
3025 mutex_unlock(&kvm->lock);
3031 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
3034 mutex_lock(&kvm->lock);
3035 list_del(&dev->vm_node);
3036 mutex_unlock(&kvm->lock);
3045 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
3048 case KVM_CAP_USER_MEMORY:
3049 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
3050 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
3051 case KVM_CAP_INTERNAL_ERROR_DATA:
3052 #ifdef CONFIG_HAVE_KVM_MSI
3053 case KVM_CAP_SIGNAL_MSI:
3055 #ifdef CONFIG_HAVE_KVM_IRQFD
3057 case KVM_CAP_IRQFD_RESAMPLE:
3059 case KVM_CAP_IOEVENTFD_ANY_LENGTH:
3060 case KVM_CAP_CHECK_EXTENSION_VM:
3061 case KVM_CAP_ENABLE_CAP_VM:
3062 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3063 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT:
3066 #ifdef CONFIG_KVM_MMIO
3067 case KVM_CAP_COALESCED_MMIO:
3068 return KVM_COALESCED_MMIO_PAGE_OFFSET;
3069 case KVM_CAP_COALESCED_PIO:
3072 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3073 case KVM_CAP_IRQ_ROUTING:
3074 return KVM_MAX_IRQ_ROUTES;
3076 #if KVM_ADDRESS_SPACE_NUM > 1
3077 case KVM_CAP_MULTI_ADDRESS_SPACE:
3078 return KVM_ADDRESS_SPACE_NUM;
3080 case KVM_CAP_MAX_VCPU_ID:
3081 return KVM_MAX_VCPU_ID;
3085 return kvm_vm_ioctl_check_extension(kvm, arg);
3088 int __attribute__((weak)) kvm_vm_ioctl_enable_cap(struct kvm *kvm,
3089 struct kvm_enable_cap *cap)
3094 static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm,
3095 struct kvm_enable_cap *cap)
3098 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3099 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT:
3100 if (cap->flags || (cap->args[0] & ~1))
3102 kvm->manual_dirty_log_protect = cap->args[0];
3106 return kvm_vm_ioctl_enable_cap(kvm, cap);
3110 static long kvm_vm_ioctl(struct file *filp,
3111 unsigned int ioctl, unsigned long arg)
3113 struct kvm *kvm = filp->private_data;
3114 void __user *argp = (void __user *)arg;
3117 if (kvm->mm != current->mm)
3120 case KVM_CREATE_VCPU:
3121 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
3123 case KVM_ENABLE_CAP: {
3124 struct kvm_enable_cap cap;
3127 if (copy_from_user(&cap, argp, sizeof(cap)))
3129 r = kvm_vm_ioctl_enable_cap_generic(kvm, &cap);
3132 case KVM_SET_USER_MEMORY_REGION: {
3133 struct kvm_userspace_memory_region kvm_userspace_mem;
3136 if (copy_from_user(&kvm_userspace_mem, argp,
3137 sizeof(kvm_userspace_mem)))
3140 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
3143 case KVM_GET_DIRTY_LOG: {
3144 struct kvm_dirty_log log;
3147 if (copy_from_user(&log, argp, sizeof(log)))
3149 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3152 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3153 case KVM_CLEAR_DIRTY_LOG: {
3154 struct kvm_clear_dirty_log log;
3157 if (copy_from_user(&log, argp, sizeof(log)))
3159 r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
3163 #ifdef CONFIG_KVM_MMIO
3164 case KVM_REGISTER_COALESCED_MMIO: {
3165 struct kvm_coalesced_mmio_zone zone;
3168 if (copy_from_user(&zone, argp, sizeof(zone)))
3170 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
3173 case KVM_UNREGISTER_COALESCED_MMIO: {
3174 struct kvm_coalesced_mmio_zone zone;
3177 if (copy_from_user(&zone, argp, sizeof(zone)))
3179 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
3184 struct kvm_irqfd data;
3187 if (copy_from_user(&data, argp, sizeof(data)))
3189 r = kvm_irqfd(kvm, &data);
3192 case KVM_IOEVENTFD: {
3193 struct kvm_ioeventfd data;
3196 if (copy_from_user(&data, argp, sizeof(data)))
3198 r = kvm_ioeventfd(kvm, &data);
3201 #ifdef CONFIG_HAVE_KVM_MSI
3202 case KVM_SIGNAL_MSI: {
3206 if (copy_from_user(&msi, argp, sizeof(msi)))
3208 r = kvm_send_userspace_msi(kvm, &msi);
3212 #ifdef __KVM_HAVE_IRQ_LINE
3213 case KVM_IRQ_LINE_STATUS:
3214 case KVM_IRQ_LINE: {
3215 struct kvm_irq_level irq_event;
3218 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
3221 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
3222 ioctl == KVM_IRQ_LINE_STATUS);
3227 if (ioctl == KVM_IRQ_LINE_STATUS) {
3228 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
3236 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3237 case KVM_SET_GSI_ROUTING: {
3238 struct kvm_irq_routing routing;
3239 struct kvm_irq_routing __user *urouting;
3240 struct kvm_irq_routing_entry *entries = NULL;
3243 if (copy_from_user(&routing, argp, sizeof(routing)))
3246 if (!kvm_arch_can_set_irq_routing(kvm))
3248 if (routing.nr > KVM_MAX_IRQ_ROUTES)
3254 entries = vmalloc(array_size(sizeof(*entries),
3260 if (copy_from_user(entries, urouting->entries,
3261 routing.nr * sizeof(*entries)))
3262 goto out_free_irq_routing;
3264 r = kvm_set_irq_routing(kvm, entries, routing.nr,
3266 out_free_irq_routing:
3270 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
3271 case KVM_CREATE_DEVICE: {
3272 struct kvm_create_device cd;
3275 if (copy_from_user(&cd, argp, sizeof(cd)))
3278 r = kvm_ioctl_create_device(kvm, &cd);
3283 if (copy_to_user(argp, &cd, sizeof(cd)))
3289 case KVM_CHECK_EXTENSION:
3290 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
3293 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
3299 #ifdef CONFIG_KVM_COMPAT
3300 struct compat_kvm_dirty_log {
3304 compat_uptr_t dirty_bitmap; /* one bit per page */
3309 static long kvm_vm_compat_ioctl(struct file *filp,
3310 unsigned int ioctl, unsigned long arg)
3312 struct kvm *kvm = filp->private_data;
3315 if (kvm->mm != current->mm)
3318 case KVM_GET_DIRTY_LOG: {
3319 struct compat_kvm_dirty_log compat_log;
3320 struct kvm_dirty_log log;
3322 if (copy_from_user(&compat_log, (void __user *)arg,
3323 sizeof(compat_log)))
3325 log.slot = compat_log.slot;
3326 log.padding1 = compat_log.padding1;
3327 log.padding2 = compat_log.padding2;
3328 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
3330 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3334 r = kvm_vm_ioctl(filp, ioctl, arg);
3340 static struct file_operations kvm_vm_fops = {
3341 .release = kvm_vm_release,
3342 .unlocked_ioctl = kvm_vm_ioctl,
3343 .llseek = noop_llseek,
3344 KVM_COMPAT(kvm_vm_compat_ioctl),
3347 static int kvm_dev_ioctl_create_vm(unsigned long type)
3353 kvm = kvm_create_vm(type);
3355 return PTR_ERR(kvm);
3356 #ifdef CONFIG_KVM_MMIO
3357 r = kvm_coalesced_mmio_init(kvm);
3361 r = get_unused_fd_flags(O_CLOEXEC);
3365 file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
3373 * Don't call kvm_put_kvm anymore at this point; file->f_op is
3374 * already set, with ->release() being kvm_vm_release(). In error
3375 * cases it will be called by the final fput(file) and will take
3376 * care of doing kvm_put_kvm(kvm).
3378 if (kvm_create_vm_debugfs(kvm, r) < 0) {
3383 kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm);
3385 fd_install(r, file);
3393 static long kvm_dev_ioctl(struct file *filp,
3394 unsigned int ioctl, unsigned long arg)
3399 case KVM_GET_API_VERSION:
3402 r = KVM_API_VERSION;
3405 r = kvm_dev_ioctl_create_vm(arg);
3407 case KVM_CHECK_EXTENSION:
3408 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
3410 case KVM_GET_VCPU_MMAP_SIZE:
3413 r = PAGE_SIZE; /* struct kvm_run */
3415 r += PAGE_SIZE; /* pio data page */
3417 #ifdef CONFIG_KVM_MMIO
3418 r += PAGE_SIZE; /* coalesced mmio ring page */
3421 case KVM_TRACE_ENABLE:
3422 case KVM_TRACE_PAUSE:
3423 case KVM_TRACE_DISABLE:
3427 return kvm_arch_dev_ioctl(filp, ioctl, arg);
3433 static struct file_operations kvm_chardev_ops = {
3434 .unlocked_ioctl = kvm_dev_ioctl,
3435 .llseek = noop_llseek,
3436 KVM_COMPAT(kvm_dev_ioctl),
3439 static struct miscdevice kvm_dev = {
3445 static void hardware_enable_nolock(void *junk)
3447 int cpu = raw_smp_processor_id();
3450 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
3453 cpumask_set_cpu(cpu, cpus_hardware_enabled);
3455 r = kvm_arch_hardware_enable();
3458 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3459 atomic_inc(&hardware_enable_failed);
3460 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
3464 static int kvm_starting_cpu(unsigned int cpu)
3466 raw_spin_lock(&kvm_count_lock);
3467 if (kvm_usage_count)
3468 hardware_enable_nolock(NULL);
3469 raw_spin_unlock(&kvm_count_lock);
3473 static void hardware_disable_nolock(void *junk)
3475 int cpu = raw_smp_processor_id();
3477 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
3479 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3480 kvm_arch_hardware_disable();
3483 static int kvm_dying_cpu(unsigned int cpu)
3485 raw_spin_lock(&kvm_count_lock);
3486 if (kvm_usage_count)
3487 hardware_disable_nolock(NULL);
3488 raw_spin_unlock(&kvm_count_lock);
3492 static void hardware_disable_all_nolock(void)
3494 BUG_ON(!kvm_usage_count);
3497 if (!kvm_usage_count)
3498 on_each_cpu(hardware_disable_nolock, NULL, 1);
3501 static void hardware_disable_all(void)
3503 raw_spin_lock(&kvm_count_lock);
3504 hardware_disable_all_nolock();
3505 raw_spin_unlock(&kvm_count_lock);
3508 static int hardware_enable_all(void)
3512 raw_spin_lock(&kvm_count_lock);
3515 if (kvm_usage_count == 1) {
3516 atomic_set(&hardware_enable_failed, 0);
3517 on_each_cpu(hardware_enable_nolock, NULL, 1);
3519 if (atomic_read(&hardware_enable_failed)) {
3520 hardware_disable_all_nolock();
3525 raw_spin_unlock(&kvm_count_lock);
3530 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3534 * Some (well, at least mine) BIOSes hang on reboot if
3537 * And Intel TXT required VMX off for all cpu when system shutdown.
3539 pr_info("kvm: exiting hardware virtualization\n");
3540 kvm_rebooting = true;
3541 on_each_cpu(hardware_disable_nolock, NULL, 1);
3545 static struct notifier_block kvm_reboot_notifier = {
3546 .notifier_call = kvm_reboot,
3550 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3554 for (i = 0; i < bus->dev_count; i++) {
3555 struct kvm_io_device *pos = bus->range[i].dev;
3557 kvm_iodevice_destructor(pos);
3562 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
3563 const struct kvm_io_range *r2)
3565 gpa_t addr1 = r1->addr;
3566 gpa_t addr2 = r2->addr;
3571 /* If r2->len == 0, match the exact address. If r2->len != 0,
3572 * accept any overlapping write. Any order is acceptable for
3573 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
3574 * we process all of them.
3587 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
3589 return kvm_io_bus_cmp(p1, p2);
3592 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
3593 gpa_t addr, int len)
3595 struct kvm_io_range *range, key;
3598 key = (struct kvm_io_range) {
3603 range = bsearch(&key, bus->range, bus->dev_count,
3604 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
3608 off = range - bus->range;
3610 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
3616 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3617 struct kvm_io_range *range, const void *val)
3621 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3625 while (idx < bus->dev_count &&
3626 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3627 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
3636 /* kvm_io_bus_write - called under kvm->slots_lock */
3637 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3638 int len, const void *val)
3640 struct kvm_io_bus *bus;
3641 struct kvm_io_range range;
3644 range = (struct kvm_io_range) {
3649 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3652 r = __kvm_io_bus_write(vcpu, bus, &range, val);
3653 return r < 0 ? r : 0;
3655 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
3657 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3658 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
3659 gpa_t addr, int len, const void *val, long cookie)
3661 struct kvm_io_bus *bus;
3662 struct kvm_io_range range;
3664 range = (struct kvm_io_range) {
3669 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3673 /* First try the device referenced by cookie. */
3674 if ((cookie >= 0) && (cookie < bus->dev_count) &&
3675 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
3676 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
3681 * cookie contained garbage; fall back to search and return the
3682 * correct cookie value.
3684 return __kvm_io_bus_write(vcpu, bus, &range, val);
3687 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3688 struct kvm_io_range *range, void *val)
3692 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3696 while (idx < bus->dev_count &&
3697 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3698 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
3707 /* kvm_io_bus_read - called under kvm->slots_lock */
3708 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3711 struct kvm_io_bus *bus;
3712 struct kvm_io_range range;
3715 range = (struct kvm_io_range) {
3720 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3723 r = __kvm_io_bus_read(vcpu, bus, &range, val);
3724 return r < 0 ? r : 0;
3727 /* Caller must hold slots_lock. */
3728 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3729 int len, struct kvm_io_device *dev)
3732 struct kvm_io_bus *new_bus, *bus;
3733 struct kvm_io_range range;
3735 bus = kvm_get_bus(kvm, bus_idx);
3739 /* exclude ioeventfd which is limited by maximum fd */
3740 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3743 new_bus = kmalloc(struct_size(bus, range, bus->dev_count + 1),
3744 GFP_KERNEL_ACCOUNT);
3748 range = (struct kvm_io_range) {
3754 for (i = 0; i < bus->dev_count; i++)
3755 if (kvm_io_bus_cmp(&bus->range[i], &range) > 0)
3758 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3759 new_bus->dev_count++;
3760 new_bus->range[i] = range;
3761 memcpy(new_bus->range + i + 1, bus->range + i,
3762 (bus->dev_count - i) * sizeof(struct kvm_io_range));
3763 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3764 synchronize_srcu_expedited(&kvm->srcu);
3770 /* Caller must hold slots_lock. */
3771 void kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3772 struct kvm_io_device *dev)
3775 struct kvm_io_bus *new_bus, *bus;
3777 bus = kvm_get_bus(kvm, bus_idx);
3781 for (i = 0; i < bus->dev_count; i++)
3782 if (bus->range[i].dev == dev) {
3786 if (i == bus->dev_count)
3789 new_bus = kmalloc(struct_size(bus, range, bus->dev_count - 1),
3790 GFP_KERNEL_ACCOUNT);
3792 pr_err("kvm: failed to shrink bus, removing it completely\n");
3796 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3797 new_bus->dev_count--;
3798 memcpy(new_bus->range + i, bus->range + i + 1,
3799 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3802 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3803 synchronize_srcu_expedited(&kvm->srcu);
3808 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3811 struct kvm_io_bus *bus;
3812 int dev_idx, srcu_idx;
3813 struct kvm_io_device *iodev = NULL;
3815 srcu_idx = srcu_read_lock(&kvm->srcu);
3817 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
3821 dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
3825 iodev = bus->range[dev_idx].dev;
3828 srcu_read_unlock(&kvm->srcu, srcu_idx);
3832 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
3834 static int kvm_debugfs_open(struct inode *inode, struct file *file,
3835 int (*get)(void *, u64 *), int (*set)(void *, u64),
3838 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
3841 /* The debugfs files are a reference to the kvm struct which
3842 * is still valid when kvm_destroy_vm is called.
3843 * To avoid the race between open and the removal of the debugfs
3844 * directory we test against the users count.
3846 if (!refcount_inc_not_zero(&stat_data->kvm->users_count))
3849 if (simple_attr_open(inode, file, get, set, fmt)) {
3850 kvm_put_kvm(stat_data->kvm);
3857 static int kvm_debugfs_release(struct inode *inode, struct file *file)
3859 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
3862 simple_attr_release(inode, file);
3863 kvm_put_kvm(stat_data->kvm);
3868 static int vm_stat_get_per_vm(void *data, u64 *val)
3870 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3872 *val = *(ulong *)((void *)stat_data->kvm + stat_data->offset);
3877 static int vm_stat_clear_per_vm(void *data, u64 val)
3879 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3884 *(ulong *)((void *)stat_data->kvm + stat_data->offset) = 0;
3889 static int vm_stat_get_per_vm_open(struct inode *inode, struct file *file)
3891 __simple_attr_check_format("%llu\n", 0ull);
3892 return kvm_debugfs_open(inode, file, vm_stat_get_per_vm,
3893 vm_stat_clear_per_vm, "%llu\n");
3896 static const struct file_operations vm_stat_get_per_vm_fops = {
3897 .owner = THIS_MODULE,
3898 .open = vm_stat_get_per_vm_open,
3899 .release = kvm_debugfs_release,
3900 .read = simple_attr_read,
3901 .write = simple_attr_write,
3902 .llseek = no_llseek,
3905 static int vcpu_stat_get_per_vm(void *data, u64 *val)
3908 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3909 struct kvm_vcpu *vcpu;
3913 kvm_for_each_vcpu(i, vcpu, stat_data->kvm)
3914 *val += *(u64 *)((void *)vcpu + stat_data->offset);
3919 static int vcpu_stat_clear_per_vm(void *data, u64 val)
3922 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3923 struct kvm_vcpu *vcpu;
3928 kvm_for_each_vcpu(i, vcpu, stat_data->kvm)
3929 *(u64 *)((void *)vcpu + stat_data->offset) = 0;
3934 static int vcpu_stat_get_per_vm_open(struct inode *inode, struct file *file)
3936 __simple_attr_check_format("%llu\n", 0ull);
3937 return kvm_debugfs_open(inode, file, vcpu_stat_get_per_vm,
3938 vcpu_stat_clear_per_vm, "%llu\n");
3941 static const struct file_operations vcpu_stat_get_per_vm_fops = {
3942 .owner = THIS_MODULE,
3943 .open = vcpu_stat_get_per_vm_open,
3944 .release = kvm_debugfs_release,
3945 .read = simple_attr_read,
3946 .write = simple_attr_write,
3947 .llseek = no_llseek,
3950 static const struct file_operations *stat_fops_per_vm[] = {
3951 [KVM_STAT_VCPU] = &vcpu_stat_get_per_vm_fops,
3952 [KVM_STAT_VM] = &vm_stat_get_per_vm_fops,
3955 static int vm_stat_get(void *_offset, u64 *val)
3957 unsigned offset = (long)_offset;
3959 struct kvm_stat_data stat_tmp = {.offset = offset};
3963 spin_lock(&kvm_lock);
3964 list_for_each_entry(kvm, &vm_list, vm_list) {
3966 vm_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
3969 spin_unlock(&kvm_lock);
3973 static int vm_stat_clear(void *_offset, u64 val)
3975 unsigned offset = (long)_offset;
3977 struct kvm_stat_data stat_tmp = {.offset = offset};
3982 spin_lock(&kvm_lock);
3983 list_for_each_entry(kvm, &vm_list, vm_list) {
3985 vm_stat_clear_per_vm((void *)&stat_tmp, 0);
3987 spin_unlock(&kvm_lock);
3992 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n");
3994 static int vcpu_stat_get(void *_offset, u64 *val)
3996 unsigned offset = (long)_offset;
3998 struct kvm_stat_data stat_tmp = {.offset = offset};
4002 spin_lock(&kvm_lock);
4003 list_for_each_entry(kvm, &vm_list, vm_list) {
4005 vcpu_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
4008 spin_unlock(&kvm_lock);
4012 static int vcpu_stat_clear(void *_offset, u64 val)
4014 unsigned offset = (long)_offset;
4016 struct kvm_stat_data stat_tmp = {.offset = offset};
4021 spin_lock(&kvm_lock);
4022 list_for_each_entry(kvm, &vm_list, vm_list) {
4024 vcpu_stat_clear_per_vm((void *)&stat_tmp, 0);
4026 spin_unlock(&kvm_lock);
4031 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear,
4034 static const struct file_operations *stat_fops[] = {
4035 [KVM_STAT_VCPU] = &vcpu_stat_fops,
4036 [KVM_STAT_VM] = &vm_stat_fops,
4039 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm)
4041 struct kobj_uevent_env *env;
4042 unsigned long long created, active;
4044 if (!kvm_dev.this_device || !kvm)
4047 spin_lock(&kvm_lock);
4048 if (type == KVM_EVENT_CREATE_VM) {
4049 kvm_createvm_count++;
4051 } else if (type == KVM_EVENT_DESTROY_VM) {
4054 created = kvm_createvm_count;
4055 active = kvm_active_vms;
4056 spin_unlock(&kvm_lock);
4058 env = kzalloc(sizeof(*env), GFP_KERNEL_ACCOUNT);
4062 add_uevent_var(env, "CREATED=%llu", created);
4063 add_uevent_var(env, "COUNT=%llu", active);
4065 if (type == KVM_EVENT_CREATE_VM) {
4066 add_uevent_var(env, "EVENT=create");
4067 kvm->userspace_pid = task_pid_nr(current);
4068 } else if (type == KVM_EVENT_DESTROY_VM) {
4069 add_uevent_var(env, "EVENT=destroy");
4071 add_uevent_var(env, "PID=%d", kvm->userspace_pid);
4073 if (!IS_ERR_OR_NULL(kvm->debugfs_dentry)) {
4074 char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL_ACCOUNT);
4077 tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX);
4079 add_uevent_var(env, "STATS_PATH=%s", tmp);
4083 /* no need for checks, since we are adding at most only 5 keys */
4084 env->envp[env->envp_idx++] = NULL;
4085 kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp);
4089 static void kvm_init_debug(void)
4091 struct kvm_stats_debugfs_item *p;
4093 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
4095 kvm_debugfs_num_entries = 0;
4096 for (p = debugfs_entries; p->name; ++p, kvm_debugfs_num_entries++) {
4097 debugfs_create_file(p->name, 0644, kvm_debugfs_dir,
4098 (void *)(long)p->offset,
4099 stat_fops[p->kind]);
4103 static int kvm_suspend(void)
4105 if (kvm_usage_count)
4106 hardware_disable_nolock(NULL);
4110 static void kvm_resume(void)
4112 if (kvm_usage_count) {
4113 lockdep_assert_held(&kvm_count_lock);
4114 hardware_enable_nolock(NULL);
4118 static struct syscore_ops kvm_syscore_ops = {
4119 .suspend = kvm_suspend,
4120 .resume = kvm_resume,
4124 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
4126 return container_of(pn, struct kvm_vcpu, preempt_notifier);
4129 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
4131 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
4133 if (vcpu->preempted)
4134 vcpu->preempted = false;
4136 kvm_arch_sched_in(vcpu, cpu);
4138 kvm_arch_vcpu_load(vcpu, cpu);
4141 static void kvm_sched_out(struct preempt_notifier *pn,
4142 struct task_struct *next)
4144 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
4146 if (current->state == TASK_RUNNING)
4147 vcpu->preempted = true;
4148 kvm_arch_vcpu_put(vcpu);
4151 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
4152 struct module *module)
4157 r = kvm_arch_init(opaque);
4162 * kvm_arch_init makes sure there's at most one caller
4163 * for architectures that support multiple implementations,
4164 * like intel and amd on x86.
4165 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
4166 * conflicts in case kvm is already setup for another implementation.
4168 r = kvm_irqfd_init();
4172 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
4177 r = kvm_arch_hardware_setup();
4181 for_each_online_cpu(cpu) {
4182 smp_call_function_single(cpu,
4183 kvm_arch_check_processor_compat,
4189 r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting",
4190 kvm_starting_cpu, kvm_dying_cpu);
4193 register_reboot_notifier(&kvm_reboot_notifier);
4195 /* A kmem cache lets us meet the alignment requirements of fx_save. */
4197 vcpu_align = __alignof__(struct kvm_vcpu);
4199 kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align,
4201 offsetof(struct kvm_vcpu, arch),
4202 sizeof_field(struct kvm_vcpu, arch),
4204 if (!kvm_vcpu_cache) {
4209 r = kvm_async_pf_init();
4213 kvm_chardev_ops.owner = module;
4214 kvm_vm_fops.owner = module;
4215 kvm_vcpu_fops.owner = module;
4217 r = misc_register(&kvm_dev);
4219 pr_err("kvm: misc device register failed\n");
4223 register_syscore_ops(&kvm_syscore_ops);
4225 kvm_preempt_ops.sched_in = kvm_sched_in;
4226 kvm_preempt_ops.sched_out = kvm_sched_out;
4230 r = kvm_vfio_ops_init();
4236 kvm_async_pf_deinit();
4238 kmem_cache_destroy(kvm_vcpu_cache);
4240 unregister_reboot_notifier(&kvm_reboot_notifier);
4241 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4244 kvm_arch_hardware_unsetup();
4246 free_cpumask_var(cpus_hardware_enabled);
4254 EXPORT_SYMBOL_GPL(kvm_init);
4258 debugfs_remove_recursive(kvm_debugfs_dir);
4259 misc_deregister(&kvm_dev);
4260 kmem_cache_destroy(kvm_vcpu_cache);
4261 kvm_async_pf_deinit();
4262 unregister_syscore_ops(&kvm_syscore_ops);
4263 unregister_reboot_notifier(&kvm_reboot_notifier);
4264 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4265 on_each_cpu(hardware_disable_nolock, NULL, 1);
4266 kvm_arch_hardware_unsetup();
4269 free_cpumask_var(cpus_hardware_enabled);
4270 kvm_vfio_ops_exit();
4272 EXPORT_SYMBOL_GPL(kvm_exit);