1 // SPDX-License-Identifier: GPL-2.0-only
3 * Kernel-based Virtual Machine driver for Linux
5 * This module enables machines with Intel VT-x extensions to run virtual
6 * machines without emulation or binary translation.
8 * Copyright (C) 2006 Qumranet, Inc.
9 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
12 * Avi Kivity <avi@qumranet.com>
13 * Yaniv Kamay <yaniv@qumranet.com>
16 #include <kvm/iodev.h>
18 #include <linux/kvm_host.h>
19 #include <linux/kvm.h>
20 #include <linux/module.h>
21 #include <linux/errno.h>
22 #include <linux/percpu.h>
24 #include <linux/miscdevice.h>
25 #include <linux/vmalloc.h>
26 #include <linux/reboot.h>
27 #include <linux/debugfs.h>
28 #include <linux/highmem.h>
29 #include <linux/file.h>
30 #include <linux/syscore_ops.h>
31 #include <linux/cpu.h>
32 #include <linux/sched/signal.h>
33 #include <linux/sched/mm.h>
34 #include <linux/sched/stat.h>
35 #include <linux/cpumask.h>
36 #include <linux/smp.h>
37 #include <linux/anon_inodes.h>
38 #include <linux/profile.h>
39 #include <linux/kvm_para.h>
40 #include <linux/pagemap.h>
41 #include <linux/mman.h>
42 #include <linux/swap.h>
43 #include <linux/bitops.h>
44 #include <linux/spinlock.h>
45 #include <linux/compat.h>
46 #include <linux/srcu.h>
47 #include <linux/hugetlb.h>
48 #include <linux/slab.h>
49 #include <linux/sort.h>
50 #include <linux/bsearch.h>
52 #include <linux/lockdep.h>
53 #include <linux/kthread.h>
55 #include <asm/processor.h>
56 #include <asm/ioctl.h>
57 #include <linux/uaccess.h>
58 #include <asm/pgtable.h>
60 #include "coalesced_mmio.h"
64 #define CREATE_TRACE_POINTS
65 #include <trace/events/kvm.h>
67 /* Worst case buffer size needed for holding an integer. */
68 #define ITOA_MAX_LEN 12
70 MODULE_AUTHOR("Qumranet");
71 MODULE_LICENSE("GPL");
73 /* Architectures should define their poll value according to the halt latency */
74 unsigned int halt_poll_ns = KVM_HALT_POLL_NS_DEFAULT;
75 module_param(halt_poll_ns, uint, 0644);
76 EXPORT_SYMBOL_GPL(halt_poll_ns);
78 /* Default doubles per-vcpu halt_poll_ns. */
79 unsigned int halt_poll_ns_grow = 2;
80 module_param(halt_poll_ns_grow, uint, 0644);
81 EXPORT_SYMBOL_GPL(halt_poll_ns_grow);
83 /* The start value to grow halt_poll_ns from */
84 unsigned int halt_poll_ns_grow_start = 10000; /* 10us */
85 module_param(halt_poll_ns_grow_start, uint, 0644);
86 EXPORT_SYMBOL_GPL(halt_poll_ns_grow_start);
88 /* Default resets per-vcpu halt_poll_ns . */
89 unsigned int halt_poll_ns_shrink;
90 module_param(halt_poll_ns_shrink, uint, 0644);
91 EXPORT_SYMBOL_GPL(halt_poll_ns_shrink);
96 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
99 DEFINE_MUTEX(kvm_lock);
100 static DEFINE_RAW_SPINLOCK(kvm_count_lock);
103 static cpumask_var_t cpus_hardware_enabled;
104 static int kvm_usage_count;
105 static atomic_t hardware_enable_failed;
107 struct kmem_cache *kvm_vcpu_cache;
108 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
110 static __read_mostly struct preempt_ops kvm_preempt_ops;
112 struct dentry *kvm_debugfs_dir;
113 EXPORT_SYMBOL_GPL(kvm_debugfs_dir);
115 static int kvm_debugfs_num_entries;
116 static const struct file_operations *stat_fops_per_vm[];
118 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
120 #ifdef CONFIG_KVM_COMPAT
121 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
123 #define KVM_COMPAT(c) .compat_ioctl = (c)
125 static long kvm_no_compat_ioctl(struct file *file, unsigned int ioctl,
126 unsigned long arg) { return -EINVAL; }
127 #define KVM_COMPAT(c) .compat_ioctl = kvm_no_compat_ioctl
129 static int hardware_enable_all(void);
130 static void hardware_disable_all(void);
132 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
134 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot, gfn_t gfn);
136 __visible bool kvm_rebooting;
137 EXPORT_SYMBOL_GPL(kvm_rebooting);
139 static bool largepages_enabled = true;
141 #define KVM_EVENT_CREATE_VM 0
142 #define KVM_EVENT_DESTROY_VM 1
143 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm);
144 static unsigned long long kvm_createvm_count;
145 static unsigned long long kvm_active_vms;
147 __weak int kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
148 unsigned long start, unsigned long end, bool blockable)
153 bool kvm_is_reserved_pfn(kvm_pfn_t pfn)
156 return PageReserved(pfn_to_page(pfn));
162 * Switches to specified vcpu, until a matching vcpu_put()
164 void vcpu_load(struct kvm_vcpu *vcpu)
167 preempt_notifier_register(&vcpu->preempt_notifier);
168 kvm_arch_vcpu_load(vcpu, cpu);
171 EXPORT_SYMBOL_GPL(vcpu_load);
173 void vcpu_put(struct kvm_vcpu *vcpu)
176 kvm_arch_vcpu_put(vcpu);
177 preempt_notifier_unregister(&vcpu->preempt_notifier);
180 EXPORT_SYMBOL_GPL(vcpu_put);
182 /* TODO: merge with kvm_arch_vcpu_should_kick */
183 static bool kvm_request_needs_ipi(struct kvm_vcpu *vcpu, unsigned req)
185 int mode = kvm_vcpu_exiting_guest_mode(vcpu);
188 * We need to wait for the VCPU to reenable interrupts and get out of
189 * READING_SHADOW_PAGE_TABLES mode.
191 if (req & KVM_REQUEST_WAIT)
192 return mode != OUTSIDE_GUEST_MODE;
195 * Need to kick a running VCPU, but otherwise there is nothing to do.
197 return mode == IN_GUEST_MODE;
200 static void ack_flush(void *_completed)
204 static inline bool kvm_kick_many_cpus(const struct cpumask *cpus, bool wait)
207 cpus = cpu_online_mask;
209 if (cpumask_empty(cpus))
212 smp_call_function_many(cpus, ack_flush, NULL, wait);
216 bool kvm_make_vcpus_request_mask(struct kvm *kvm, unsigned int req,
217 unsigned long *vcpu_bitmap, cpumask_var_t tmp)
220 struct kvm_vcpu *vcpu;
225 kvm_for_each_vcpu(i, vcpu, kvm) {
226 if (vcpu_bitmap && !test_bit(i, vcpu_bitmap))
229 kvm_make_request(req, vcpu);
232 if (!(req & KVM_REQUEST_NO_WAKEUP) && kvm_vcpu_wake_up(vcpu))
235 if (tmp != NULL && cpu != -1 && cpu != me &&
236 kvm_request_needs_ipi(vcpu, req))
237 __cpumask_set_cpu(cpu, tmp);
240 called = kvm_kick_many_cpus(tmp, !!(req & KVM_REQUEST_WAIT));
246 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
251 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
253 called = kvm_make_vcpus_request_mask(kvm, req, NULL, cpus);
255 free_cpumask_var(cpus);
259 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
260 void kvm_flush_remote_tlbs(struct kvm *kvm)
263 * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in
264 * kvm_make_all_cpus_request.
266 long dirty_count = smp_load_acquire(&kvm->tlbs_dirty);
269 * We want to publish modifications to the page tables before reading
270 * mode. Pairs with a memory barrier in arch-specific code.
271 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
272 * and smp_mb in walk_shadow_page_lockless_begin/end.
273 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
275 * There is already an smp_mb__after_atomic() before
276 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
279 if (!kvm_arch_flush_remote_tlb(kvm)
280 || kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
281 ++kvm->stat.remote_tlb_flush;
282 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
284 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
287 void kvm_reload_remote_mmus(struct kvm *kvm)
289 kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
292 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
297 mutex_init(&vcpu->mutex);
302 init_swait_queue_head(&vcpu->wq);
303 kvm_async_pf_vcpu_init(vcpu);
306 INIT_LIST_HEAD(&vcpu->blocked_vcpu_list);
308 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
313 vcpu->run = page_address(page);
315 kvm_vcpu_set_in_spin_loop(vcpu, false);
316 kvm_vcpu_set_dy_eligible(vcpu, false);
317 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,
395 mmu_notifier_range_blockable(range));
397 srcu_read_unlock(&kvm->srcu, idx);
402 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
403 const struct mmu_notifier_range *range)
405 struct kvm *kvm = mmu_notifier_to_kvm(mn);
407 spin_lock(&kvm->mmu_lock);
409 * This sequence increase will notify the kvm page fault that
410 * the page that is going to be mapped in the spte could have
413 kvm->mmu_notifier_seq++;
416 * The above sequence increase must be visible before the
417 * below count decrease, which is ensured by the smp_wmb above
418 * in conjunction with the smp_rmb in mmu_notifier_retry().
420 kvm->mmu_notifier_count--;
421 spin_unlock(&kvm->mmu_lock);
423 BUG_ON(kvm->mmu_notifier_count < 0);
426 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
427 struct mm_struct *mm,
431 struct kvm *kvm = mmu_notifier_to_kvm(mn);
434 idx = srcu_read_lock(&kvm->srcu);
435 spin_lock(&kvm->mmu_lock);
437 young = kvm_age_hva(kvm, start, end);
439 kvm_flush_remote_tlbs(kvm);
441 spin_unlock(&kvm->mmu_lock);
442 srcu_read_unlock(&kvm->srcu, idx);
447 static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
448 struct mm_struct *mm,
452 struct kvm *kvm = mmu_notifier_to_kvm(mn);
455 idx = srcu_read_lock(&kvm->srcu);
456 spin_lock(&kvm->mmu_lock);
458 * Even though we do not flush TLB, this will still adversely
459 * affect performance on pre-Haswell Intel EPT, where there is
460 * no EPT Access Bit to clear so that we have to tear down EPT
461 * tables instead. If we find this unacceptable, we can always
462 * add a parameter to kvm_age_hva so that it effectively doesn't
463 * do anything on clear_young.
465 * Also note that currently we never issue secondary TLB flushes
466 * from clear_young, leaving this job up to the regular system
467 * cadence. If we find this inaccurate, we might come up with a
468 * more sophisticated heuristic later.
470 young = kvm_age_hva(kvm, start, end);
471 spin_unlock(&kvm->mmu_lock);
472 srcu_read_unlock(&kvm->srcu, idx);
477 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
478 struct mm_struct *mm,
479 unsigned long address)
481 struct kvm *kvm = mmu_notifier_to_kvm(mn);
484 idx = srcu_read_lock(&kvm->srcu);
485 spin_lock(&kvm->mmu_lock);
486 young = kvm_test_age_hva(kvm, address);
487 spin_unlock(&kvm->mmu_lock);
488 srcu_read_unlock(&kvm->srcu, idx);
493 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
494 struct mm_struct *mm)
496 struct kvm *kvm = mmu_notifier_to_kvm(mn);
499 idx = srcu_read_lock(&kvm->srcu);
500 kvm_arch_flush_shadow_all(kvm);
501 srcu_read_unlock(&kvm->srcu, idx);
504 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
505 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
506 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
507 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
508 .clear_young = kvm_mmu_notifier_clear_young,
509 .test_young = kvm_mmu_notifier_test_young,
510 .change_pte = kvm_mmu_notifier_change_pte,
511 .release = kvm_mmu_notifier_release,
514 static int kvm_init_mmu_notifier(struct kvm *kvm)
516 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
517 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
520 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
522 static int kvm_init_mmu_notifier(struct kvm *kvm)
527 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
529 static struct kvm_memslots *kvm_alloc_memslots(void)
532 struct kvm_memslots *slots;
534 slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL_ACCOUNT);
538 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
539 slots->id_to_index[i] = slots->memslots[i].id = i;
544 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
546 if (!memslot->dirty_bitmap)
549 kvfree(memslot->dirty_bitmap);
550 memslot->dirty_bitmap = NULL;
554 * Free any memory in @free but not in @dont.
556 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
557 struct kvm_memory_slot *dont)
559 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
560 kvm_destroy_dirty_bitmap(free);
562 kvm_arch_free_memslot(kvm, free, dont);
567 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
569 struct kvm_memory_slot *memslot;
574 kvm_for_each_memslot(memslot, slots)
575 kvm_free_memslot(kvm, memslot, NULL);
580 static void kvm_destroy_vm_debugfs(struct kvm *kvm)
584 if (!kvm->debugfs_dentry)
587 debugfs_remove_recursive(kvm->debugfs_dentry);
589 if (kvm->debugfs_stat_data) {
590 for (i = 0; i < kvm_debugfs_num_entries; i++)
591 kfree(kvm->debugfs_stat_data[i]);
592 kfree(kvm->debugfs_stat_data);
596 static int kvm_create_vm_debugfs(struct kvm *kvm, int fd)
598 char dir_name[ITOA_MAX_LEN * 2];
599 struct kvm_stat_data *stat_data;
600 struct kvm_stats_debugfs_item *p;
602 if (!debugfs_initialized())
605 snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd);
606 kvm->debugfs_dentry = debugfs_create_dir(dir_name, kvm_debugfs_dir);
608 kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries,
609 sizeof(*kvm->debugfs_stat_data),
611 if (!kvm->debugfs_stat_data)
614 for (p = debugfs_entries; p->name; p++) {
615 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT);
619 stat_data->kvm = kvm;
620 stat_data->offset = p->offset;
621 stat_data->mode = p->mode ? p->mode : 0644;
622 kvm->debugfs_stat_data[p - debugfs_entries] = stat_data;
623 debugfs_create_file(p->name, stat_data->mode, kvm->debugfs_dentry,
624 stat_data, stat_fops_per_vm[p->kind]);
630 * Called after the VM is otherwise initialized, but just before adding it to
633 int __weak kvm_arch_post_init_vm(struct kvm *kvm)
639 * Called just after removing the VM from the vm_list, but before doing any
642 void __weak kvm_arch_pre_destroy_vm(struct kvm *kvm)
646 static struct kvm *kvm_create_vm(unsigned long type)
648 struct kvm *kvm = kvm_arch_alloc_vm();
653 return ERR_PTR(-ENOMEM);
655 spin_lock_init(&kvm->mmu_lock);
657 kvm->mm = current->mm;
658 kvm_eventfd_init(kvm);
659 mutex_init(&kvm->lock);
660 mutex_init(&kvm->irq_lock);
661 mutex_init(&kvm->slots_lock);
662 INIT_LIST_HEAD(&kvm->devices);
664 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
666 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
667 struct kvm_memslots *slots = kvm_alloc_memslots();
670 goto out_err_no_arch_destroy_vm;
671 /* Generations must be different for each address space. */
672 slots->generation = i;
673 rcu_assign_pointer(kvm->memslots[i], slots);
676 for (i = 0; i < KVM_NR_BUSES; i++) {
677 rcu_assign_pointer(kvm->buses[i],
678 kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL_ACCOUNT));
680 goto out_err_no_arch_destroy_vm;
683 refcount_set(&kvm->users_count, 1);
684 r = kvm_arch_init_vm(kvm, type);
686 goto out_err_no_arch_destroy_vm;
688 r = hardware_enable_all();
690 goto out_err_no_disable;
692 #ifdef CONFIG_HAVE_KVM_IRQFD
693 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
696 if (init_srcu_struct(&kvm->srcu))
697 goto out_err_no_srcu;
698 if (init_srcu_struct(&kvm->irq_srcu))
699 goto out_err_no_irq_srcu;
701 r = kvm_init_mmu_notifier(kvm);
703 goto out_err_no_mmu_notifier;
705 r = kvm_arch_post_init_vm(kvm);
709 mutex_lock(&kvm_lock);
710 list_add(&kvm->vm_list, &vm_list);
711 mutex_unlock(&kvm_lock);
713 preempt_notifier_inc();
718 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
719 if (kvm->mmu_notifier.ops)
720 mmu_notifier_unregister(&kvm->mmu_notifier, current->mm);
722 out_err_no_mmu_notifier:
723 cleanup_srcu_struct(&kvm->irq_srcu);
725 cleanup_srcu_struct(&kvm->srcu);
727 hardware_disable_all();
729 kvm_arch_destroy_vm(kvm);
730 WARN_ON_ONCE(!refcount_dec_and_test(&kvm->users_count));
731 out_err_no_arch_destroy_vm:
732 for (i = 0; i < KVM_NR_BUSES; i++)
733 kfree(kvm_get_bus(kvm, i));
734 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
735 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
736 kvm_arch_free_vm(kvm);
741 static void kvm_destroy_devices(struct kvm *kvm)
743 struct kvm_device *dev, *tmp;
746 * We do not need to take the kvm->lock here, because nobody else
747 * has a reference to the struct kvm at this point and therefore
748 * cannot access the devices list anyhow.
750 list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
751 list_del(&dev->vm_node);
752 dev->ops->destroy(dev);
756 static void kvm_destroy_vm(struct kvm *kvm)
759 struct mm_struct *mm = kvm->mm;
761 kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm);
762 kvm_destroy_vm_debugfs(kvm);
763 kvm_arch_sync_events(kvm);
764 mutex_lock(&kvm_lock);
765 list_del(&kvm->vm_list);
766 mutex_unlock(&kvm_lock);
767 kvm_arch_pre_destroy_vm(kvm);
769 kvm_free_irq_routing(kvm);
770 for (i = 0; i < KVM_NR_BUSES; i++) {
771 struct kvm_io_bus *bus = kvm_get_bus(kvm, i);
774 kvm_io_bus_destroy(bus);
775 kvm->buses[i] = NULL;
777 kvm_coalesced_mmio_free(kvm);
778 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
779 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
781 kvm_arch_flush_shadow_all(kvm);
783 kvm_arch_destroy_vm(kvm);
784 kvm_destroy_devices(kvm);
785 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
786 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
787 cleanup_srcu_struct(&kvm->irq_srcu);
788 cleanup_srcu_struct(&kvm->srcu);
789 kvm_arch_free_vm(kvm);
790 preempt_notifier_dec();
791 hardware_disable_all();
795 void kvm_get_kvm(struct kvm *kvm)
797 refcount_inc(&kvm->users_count);
799 EXPORT_SYMBOL_GPL(kvm_get_kvm);
801 void kvm_put_kvm(struct kvm *kvm)
803 if (refcount_dec_and_test(&kvm->users_count))
806 EXPORT_SYMBOL_GPL(kvm_put_kvm);
809 static int kvm_vm_release(struct inode *inode, struct file *filp)
811 struct kvm *kvm = filp->private_data;
813 kvm_irqfd_release(kvm);
820 * Allocation size is twice as large as the actual dirty bitmap size.
821 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
823 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
825 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
827 memslot->dirty_bitmap = kvzalloc(dirty_bytes, GFP_KERNEL_ACCOUNT);
828 if (!memslot->dirty_bitmap)
835 * Insert memslot and re-sort memslots based on their GFN,
836 * so binary search could be used to lookup GFN.
837 * Sorting algorithm takes advantage of having initially
838 * sorted array and known changed memslot position.
840 static void update_memslots(struct kvm_memslots *slots,
841 struct kvm_memory_slot *new,
842 enum kvm_mr_change change)
845 int i = slots->id_to_index[id];
846 struct kvm_memory_slot *mslots = slots->memslots;
848 WARN_ON(mslots[i].id != id);
852 WARN_ON(mslots[i].npages || !new->npages);
856 WARN_ON(new->npages || !mslots[i].npages);
862 while (i < KVM_MEM_SLOTS_NUM - 1 &&
863 new->base_gfn <= mslots[i + 1].base_gfn) {
864 if (!mslots[i + 1].npages)
866 mslots[i] = mslots[i + 1];
867 slots->id_to_index[mslots[i].id] = i;
872 * The ">=" is needed when creating a slot with base_gfn == 0,
873 * so that it moves before all those with base_gfn == npages == 0.
875 * On the other hand, if new->npages is zero, the above loop has
876 * already left i pointing to the beginning of the empty part of
877 * mslots, and the ">=" would move the hole backwards in this
878 * case---which is wrong. So skip the loop when deleting a slot.
882 new->base_gfn >= mslots[i - 1].base_gfn) {
883 mslots[i] = mslots[i - 1];
884 slots->id_to_index[mslots[i].id] = i;
888 WARN_ON_ONCE(i != slots->used_slots);
891 slots->id_to_index[mslots[i].id] = i;
894 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
896 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
898 #ifdef __KVM_HAVE_READONLY_MEM
899 valid_flags |= KVM_MEM_READONLY;
902 if (mem->flags & ~valid_flags)
908 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
909 int as_id, struct kvm_memslots *slots)
911 struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id);
912 u64 gen = old_memslots->generation;
914 WARN_ON(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS);
915 slots->generation = gen | KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
917 rcu_assign_pointer(kvm->memslots[as_id], slots);
918 synchronize_srcu_expedited(&kvm->srcu);
921 * Increment the new memslot generation a second time, dropping the
922 * update in-progress flag and incrementing then generation based on
923 * the number of address spaces. This provides a unique and easily
924 * identifiable generation number while the memslots are in flux.
926 gen = slots->generation & ~KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
929 * Generations must be unique even across address spaces. We do not need
930 * a global counter for that, instead the generation space is evenly split
931 * across address spaces. For example, with two address spaces, address
932 * space 0 will use generations 0, 2, 4, ... while address space 1 will
933 * use generations 1, 3, 5, ...
935 gen += KVM_ADDRESS_SPACE_NUM;
937 kvm_arch_memslots_updated(kvm, gen);
939 slots->generation = gen;
945 * Allocate some memory and give it an address in the guest physical address
948 * Discontiguous memory is allowed, mostly for framebuffers.
950 * Must be called holding kvm->slots_lock for write.
952 int __kvm_set_memory_region(struct kvm *kvm,
953 const struct kvm_userspace_memory_region *mem)
957 unsigned long npages;
958 struct kvm_memory_slot *slot;
959 struct kvm_memory_slot old, new;
960 struct kvm_memslots *slots = NULL, *old_memslots;
962 enum kvm_mr_change change;
964 r = check_memory_region_flags(mem);
969 as_id = mem->slot >> 16;
972 /* General sanity checks */
973 if (mem->memory_size & (PAGE_SIZE - 1))
975 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
977 /* We can read the guest memory with __xxx_user() later on. */
978 if ((id < KVM_USER_MEM_SLOTS) &&
979 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
980 !access_ok((void __user *)(unsigned long)mem->userspace_addr,
983 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
985 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
988 slot = id_to_memslot(__kvm_memslots(kvm, as_id), id);
989 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
990 npages = mem->memory_size >> PAGE_SHIFT;
992 if (npages > KVM_MEM_MAX_NR_PAGES)
998 new.base_gfn = base_gfn;
1000 new.flags = mem->flags;
1004 change = KVM_MR_CREATE;
1005 else { /* Modify an existing slot. */
1006 if ((mem->userspace_addr != old.userspace_addr) ||
1007 (npages != old.npages) ||
1008 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
1011 if (base_gfn != old.base_gfn)
1012 change = KVM_MR_MOVE;
1013 else if (new.flags != old.flags)
1014 change = KVM_MR_FLAGS_ONLY;
1015 else { /* Nothing to change. */
1024 change = KVM_MR_DELETE;
1029 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
1030 /* Check for overlaps */
1032 kvm_for_each_memslot(slot, __kvm_memslots(kvm, as_id)) {
1035 if (!((base_gfn + npages <= slot->base_gfn) ||
1036 (base_gfn >= slot->base_gfn + slot->npages)))
1041 /* Free page dirty bitmap if unneeded */
1042 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
1043 new.dirty_bitmap = NULL;
1046 if (change == KVM_MR_CREATE) {
1047 new.userspace_addr = mem->userspace_addr;
1049 if (kvm_arch_create_memslot(kvm, &new, npages))
1053 /* Allocate page dirty bitmap if needed */
1054 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
1055 if (kvm_create_dirty_bitmap(&new) < 0)
1059 slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL_ACCOUNT);
1062 memcpy(slots, __kvm_memslots(kvm, as_id), sizeof(struct kvm_memslots));
1064 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
1065 slot = id_to_memslot(slots, id);
1066 slot->flags |= KVM_MEMSLOT_INVALID;
1068 old_memslots = install_new_memslots(kvm, as_id, slots);
1070 /* From this point no new shadow pages pointing to a deleted,
1071 * or moved, memslot will be created.
1073 * validation of sp->gfn happens in:
1074 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1075 * - kvm_is_visible_gfn (mmu_check_roots)
1077 kvm_arch_flush_shadow_memslot(kvm, slot);
1080 * We can re-use the old_memslots from above, the only difference
1081 * from the currently installed memslots is the invalid flag. This
1082 * will get overwritten by update_memslots anyway.
1084 slots = old_memslots;
1087 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
1091 /* actual memory is freed via old in kvm_free_memslot below */
1092 if (change == KVM_MR_DELETE) {
1093 new.dirty_bitmap = NULL;
1094 memset(&new.arch, 0, sizeof(new.arch));
1097 update_memslots(slots, &new, change);
1098 old_memslots = install_new_memslots(kvm, as_id, slots);
1100 kvm_arch_commit_memory_region(kvm, mem, &old, &new, change);
1102 kvm_free_memslot(kvm, &old, &new);
1103 kvfree(old_memslots);
1109 kvm_free_memslot(kvm, &new, &old);
1113 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1115 int kvm_set_memory_region(struct kvm *kvm,
1116 const struct kvm_userspace_memory_region *mem)
1120 mutex_lock(&kvm->slots_lock);
1121 r = __kvm_set_memory_region(kvm, mem);
1122 mutex_unlock(&kvm->slots_lock);
1125 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1127 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1128 struct kvm_userspace_memory_region *mem)
1130 if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1133 return kvm_set_memory_region(kvm, mem);
1136 int kvm_get_dirty_log(struct kvm *kvm,
1137 struct kvm_dirty_log *log, int *is_dirty)
1139 struct kvm_memslots *slots;
1140 struct kvm_memory_slot *memslot;
1143 unsigned long any = 0;
1145 as_id = log->slot >> 16;
1146 id = (u16)log->slot;
1147 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1150 slots = __kvm_memslots(kvm, as_id);
1151 memslot = id_to_memslot(slots, id);
1152 if (!memslot->dirty_bitmap)
1155 n = kvm_dirty_bitmap_bytes(memslot);
1157 for (i = 0; !any && i < n/sizeof(long); ++i)
1158 any = memslot->dirty_bitmap[i];
1160 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1167 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1169 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1171 * kvm_get_dirty_log_protect - get a snapshot of dirty pages
1172 * and reenable dirty page tracking for the corresponding pages.
1173 * @kvm: pointer to kvm instance
1174 * @log: slot id and address to which we copy the log
1175 * @flush: true if TLB flush is needed by caller
1177 * We need to keep it in mind that VCPU threads can write to the bitmap
1178 * concurrently. So, to avoid losing track of dirty pages we keep the
1181 * 1. Take a snapshot of the bit and clear it if needed.
1182 * 2. Write protect the corresponding page.
1183 * 3. Copy the snapshot to the userspace.
1184 * 4. Upon return caller flushes TLB's if needed.
1186 * Between 2 and 4, the guest may write to the page using the remaining TLB
1187 * entry. This is not a problem because the page is reported dirty using
1188 * the snapshot taken before and step 4 ensures that writes done after
1189 * exiting to userspace will be logged for the next call.
1192 int kvm_get_dirty_log_protect(struct kvm *kvm,
1193 struct kvm_dirty_log *log, bool *flush)
1195 struct kvm_memslots *slots;
1196 struct kvm_memory_slot *memslot;
1199 unsigned long *dirty_bitmap;
1200 unsigned long *dirty_bitmap_buffer;
1202 as_id = log->slot >> 16;
1203 id = (u16)log->slot;
1204 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1207 slots = __kvm_memslots(kvm, as_id);
1208 memslot = id_to_memslot(slots, id);
1210 dirty_bitmap = memslot->dirty_bitmap;
1214 n = kvm_dirty_bitmap_bytes(memslot);
1216 if (kvm->manual_dirty_log_protect) {
1218 * Unlike kvm_get_dirty_log, we always return false in *flush,
1219 * because no flush is needed until KVM_CLEAR_DIRTY_LOG. There
1220 * is some code duplication between this function and
1221 * kvm_get_dirty_log, but hopefully all architecture
1222 * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log
1223 * can be eliminated.
1225 dirty_bitmap_buffer = dirty_bitmap;
1227 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
1228 memset(dirty_bitmap_buffer, 0, n);
1230 spin_lock(&kvm->mmu_lock);
1231 for (i = 0; i < n / sizeof(long); i++) {
1235 if (!dirty_bitmap[i])
1239 mask = xchg(&dirty_bitmap[i], 0);
1240 dirty_bitmap_buffer[i] = mask;
1242 offset = i * BITS_PER_LONG;
1243 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1246 spin_unlock(&kvm->mmu_lock);
1249 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1253 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect);
1256 * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap
1257 * and reenable dirty page tracking for the corresponding pages.
1258 * @kvm: pointer to kvm instance
1259 * @log: slot id and address from which to fetch the bitmap of dirty pages
1260 * @flush: true if TLB flush is needed by caller
1262 int kvm_clear_dirty_log_protect(struct kvm *kvm,
1263 struct kvm_clear_dirty_log *log, bool *flush)
1265 struct kvm_memslots *slots;
1266 struct kvm_memory_slot *memslot;
1270 unsigned long *dirty_bitmap;
1271 unsigned long *dirty_bitmap_buffer;
1273 as_id = log->slot >> 16;
1274 id = (u16)log->slot;
1275 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1278 if (log->first_page & 63)
1281 slots = __kvm_memslots(kvm, as_id);
1282 memslot = id_to_memslot(slots, id);
1284 dirty_bitmap = memslot->dirty_bitmap;
1288 n = ALIGN(log->num_pages, BITS_PER_LONG) / 8;
1290 if (log->first_page > memslot->npages ||
1291 log->num_pages > memslot->npages - log->first_page ||
1292 (log->num_pages < memslot->npages - log->first_page && (log->num_pages & 63)))
1296 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
1297 if (copy_from_user(dirty_bitmap_buffer, log->dirty_bitmap, n))
1300 spin_lock(&kvm->mmu_lock);
1301 for (offset = log->first_page, i = offset / BITS_PER_LONG,
1302 n = DIV_ROUND_UP(log->num_pages, BITS_PER_LONG); n--;
1303 i++, offset += BITS_PER_LONG) {
1304 unsigned long mask = *dirty_bitmap_buffer++;
1305 atomic_long_t *p = (atomic_long_t *) &dirty_bitmap[i];
1309 mask &= atomic_long_fetch_andnot(mask, p);
1312 * mask contains the bits that really have been cleared. This
1313 * never includes any bits beyond the length of the memslot (if
1314 * the length is not aligned to 64 pages), therefore it is not
1315 * a problem if userspace sets them in log->dirty_bitmap.
1319 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1323 spin_unlock(&kvm->mmu_lock);
1327 EXPORT_SYMBOL_GPL(kvm_clear_dirty_log_protect);
1330 bool kvm_largepages_enabled(void)
1332 return largepages_enabled;
1335 void kvm_disable_largepages(void)
1337 largepages_enabled = false;
1339 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1341 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1343 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1345 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1347 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
1349 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn);
1352 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1354 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1356 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1357 memslot->flags & KVM_MEMSLOT_INVALID)
1362 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1364 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1366 struct vm_area_struct *vma;
1367 unsigned long addr, size;
1371 addr = gfn_to_hva(kvm, gfn);
1372 if (kvm_is_error_hva(addr))
1375 down_read(¤t->mm->mmap_sem);
1376 vma = find_vma(current->mm, addr);
1380 size = vma_kernel_pagesize(vma);
1383 up_read(¤t->mm->mmap_sem);
1388 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1390 return slot->flags & KVM_MEM_READONLY;
1393 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1394 gfn_t *nr_pages, bool write)
1396 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1397 return KVM_HVA_ERR_BAD;
1399 if (memslot_is_readonly(slot) && write)
1400 return KVM_HVA_ERR_RO_BAD;
1403 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1405 return __gfn_to_hva_memslot(slot, gfn);
1408 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1411 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1414 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1417 return gfn_to_hva_many(slot, gfn, NULL);
1419 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1421 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1423 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1425 EXPORT_SYMBOL_GPL(gfn_to_hva);
1427 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
1429 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
1431 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
1434 * Return the hva of a @gfn and the R/W attribute if possible.
1436 * @slot: the kvm_memory_slot which contains @gfn
1437 * @gfn: the gfn to be translated
1438 * @writable: used to return the read/write attribute of the @slot if the hva
1439 * is valid and @writable is not NULL
1441 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1442 gfn_t gfn, bool *writable)
1444 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1446 if (!kvm_is_error_hva(hva) && writable)
1447 *writable = !memslot_is_readonly(slot);
1452 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1454 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1456 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1459 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
1461 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1463 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1466 static inline int check_user_page_hwpoison(unsigned long addr)
1468 int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
1470 rc = get_user_pages(addr, 1, flags, NULL, NULL);
1471 return rc == -EHWPOISON;
1475 * The fast path to get the writable pfn which will be stored in @pfn,
1476 * true indicates success, otherwise false is returned. It's also the
1477 * only part that runs if we can are in atomic context.
1479 static bool hva_to_pfn_fast(unsigned long addr, bool write_fault,
1480 bool *writable, kvm_pfn_t *pfn)
1482 struct page *page[1];
1486 * Fast pin a writable pfn only if it is a write fault request
1487 * or the caller allows to map a writable pfn for a read fault
1490 if (!(write_fault || writable))
1493 npages = __get_user_pages_fast(addr, 1, 1, page);
1495 *pfn = page_to_pfn(page[0]);
1506 * The slow path to get the pfn of the specified host virtual address,
1507 * 1 indicates success, -errno is returned if error is detected.
1509 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1510 bool *writable, kvm_pfn_t *pfn)
1512 unsigned int flags = FOLL_HWPOISON;
1519 *writable = write_fault;
1522 flags |= FOLL_WRITE;
1524 flags |= FOLL_NOWAIT;
1526 npages = get_user_pages_unlocked(addr, 1, &page, flags);
1530 /* map read fault as writable if possible */
1531 if (unlikely(!write_fault) && writable) {
1534 if (__get_user_pages_fast(addr, 1, 1, &wpage) == 1) {
1540 *pfn = page_to_pfn(page);
1544 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1546 if (unlikely(!(vma->vm_flags & VM_READ)))
1549 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1555 static int hva_to_pfn_remapped(struct vm_area_struct *vma,
1556 unsigned long addr, bool *async,
1557 bool write_fault, bool *writable,
1563 r = follow_pfn(vma, addr, &pfn);
1566 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
1567 * not call the fault handler, so do it here.
1569 bool unlocked = false;
1570 r = fixup_user_fault(current, current->mm, addr,
1571 (write_fault ? FAULT_FLAG_WRITE : 0),
1578 r = follow_pfn(vma, addr, &pfn);
1588 * Get a reference here because callers of *hva_to_pfn* and
1589 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
1590 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
1591 * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will
1592 * simply do nothing for reserved pfns.
1594 * Whoever called remap_pfn_range is also going to call e.g.
1595 * unmap_mapping_range before the underlying pages are freed,
1596 * causing a call to our MMU notifier.
1605 * Pin guest page in memory and return its pfn.
1606 * @addr: host virtual address which maps memory to the guest
1607 * @atomic: whether this function can sleep
1608 * @async: whether this function need to wait IO complete if the
1609 * host page is not in the memory
1610 * @write_fault: whether we should get a writable host page
1611 * @writable: whether it allows to map a writable host page for !@write_fault
1613 * The function will map a writable host page for these two cases:
1614 * 1): @write_fault = true
1615 * 2): @write_fault = false && @writable, @writable will tell the caller
1616 * whether the mapping is writable.
1618 static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1619 bool write_fault, bool *writable)
1621 struct vm_area_struct *vma;
1625 /* we can do it either atomically or asynchronously, not both */
1626 BUG_ON(atomic && async);
1628 if (hva_to_pfn_fast(addr, write_fault, writable, &pfn))
1632 return KVM_PFN_ERR_FAULT;
1634 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1638 down_read(¤t->mm->mmap_sem);
1639 if (npages == -EHWPOISON ||
1640 (!async && check_user_page_hwpoison(addr))) {
1641 pfn = KVM_PFN_ERR_HWPOISON;
1646 vma = find_vma_intersection(current->mm, addr, addr + 1);
1649 pfn = KVM_PFN_ERR_FAULT;
1650 else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
1651 r = hva_to_pfn_remapped(vma, addr, async, write_fault, writable, &pfn);
1655 pfn = KVM_PFN_ERR_FAULT;
1657 if (async && vma_is_valid(vma, write_fault))
1659 pfn = KVM_PFN_ERR_FAULT;
1662 up_read(¤t->mm->mmap_sem);
1666 kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn,
1667 bool atomic, bool *async, bool write_fault,
1670 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1672 if (addr == KVM_HVA_ERR_RO_BAD) {
1675 return KVM_PFN_ERR_RO_FAULT;
1678 if (kvm_is_error_hva(addr)) {
1681 return KVM_PFN_NOSLOT;
1684 /* Do not map writable pfn in the readonly memslot. */
1685 if (writable && memslot_is_readonly(slot)) {
1690 return hva_to_pfn(addr, atomic, async, write_fault,
1693 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
1695 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1698 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
1699 write_fault, writable);
1701 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1703 kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1705 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1707 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
1709 kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1711 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1713 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1715 kvm_pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1717 return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm, gfn), gfn);
1719 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1721 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
1723 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1725 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
1727 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1729 return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
1731 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1733 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
1735 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1737 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
1739 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1740 struct page **pages, int nr_pages)
1745 addr = gfn_to_hva_many(slot, gfn, &entry);
1746 if (kvm_is_error_hva(addr))
1749 if (entry < nr_pages)
1752 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1754 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1756 static struct page *kvm_pfn_to_page(kvm_pfn_t pfn)
1758 if (is_error_noslot_pfn(pfn))
1759 return KVM_ERR_PTR_BAD_PAGE;
1761 if (kvm_is_reserved_pfn(pfn)) {
1763 return KVM_ERR_PTR_BAD_PAGE;
1766 return pfn_to_page(pfn);
1769 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1773 pfn = gfn_to_pfn(kvm, gfn);
1775 return kvm_pfn_to_page(pfn);
1777 EXPORT_SYMBOL_GPL(gfn_to_page);
1779 static int __kvm_map_gfn(struct kvm_memory_slot *slot, gfn_t gfn,
1780 struct kvm_host_map *map)
1784 struct page *page = KVM_UNMAPPED_PAGE;
1789 pfn = gfn_to_pfn_memslot(slot, gfn);
1790 if (is_error_noslot_pfn(pfn))
1793 if (pfn_valid(pfn)) {
1794 page = pfn_to_page(pfn);
1796 #ifdef CONFIG_HAS_IOMEM
1798 hva = memremap(pfn_to_hpa(pfn), PAGE_SIZE, MEMREMAP_WB);
1813 int kvm_vcpu_map(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map)
1815 return __kvm_map_gfn(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, map);
1817 EXPORT_SYMBOL_GPL(kvm_vcpu_map);
1819 void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map,
1828 if (map->page != KVM_UNMAPPED_PAGE)
1830 #ifdef CONFIG_HAS_IOMEM
1836 kvm_vcpu_mark_page_dirty(vcpu, map->gfn);
1837 kvm_release_pfn_dirty(map->pfn);
1839 kvm_release_pfn_clean(map->pfn);
1845 EXPORT_SYMBOL_GPL(kvm_vcpu_unmap);
1847 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
1851 pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
1853 return kvm_pfn_to_page(pfn);
1855 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
1857 void kvm_release_page_clean(struct page *page)
1859 WARN_ON(is_error_page(page));
1861 kvm_release_pfn_clean(page_to_pfn(page));
1863 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1865 void kvm_release_pfn_clean(kvm_pfn_t pfn)
1867 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
1868 put_page(pfn_to_page(pfn));
1870 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1872 void kvm_release_page_dirty(struct page *page)
1874 WARN_ON(is_error_page(page));
1876 kvm_release_pfn_dirty(page_to_pfn(page));
1878 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1880 void kvm_release_pfn_dirty(kvm_pfn_t pfn)
1882 kvm_set_pfn_dirty(pfn);
1883 kvm_release_pfn_clean(pfn);
1885 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1887 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
1889 if (!kvm_is_reserved_pfn(pfn)) {
1890 struct page *page = pfn_to_page(pfn);
1895 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1897 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
1899 if (!kvm_is_reserved_pfn(pfn))
1900 mark_page_accessed(pfn_to_page(pfn));
1902 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1904 void kvm_get_pfn(kvm_pfn_t pfn)
1906 if (!kvm_is_reserved_pfn(pfn))
1907 get_page(pfn_to_page(pfn));
1909 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1911 static int next_segment(unsigned long len, int offset)
1913 if (len > PAGE_SIZE - offset)
1914 return PAGE_SIZE - offset;
1919 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
1920 void *data, int offset, int len)
1925 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1926 if (kvm_is_error_hva(addr))
1928 r = __copy_from_user(data, (void __user *)addr + offset, len);
1934 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1937 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1939 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1941 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1943 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
1944 int offset, int len)
1946 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1948 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1950 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
1952 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1954 gfn_t gfn = gpa >> PAGE_SHIFT;
1956 int offset = offset_in_page(gpa);
1959 while ((seg = next_segment(len, offset)) != 0) {
1960 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1970 EXPORT_SYMBOL_GPL(kvm_read_guest);
1972 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
1974 gfn_t gfn = gpa >> PAGE_SHIFT;
1976 int offset = offset_in_page(gpa);
1979 while ((seg = next_segment(len, offset)) != 0) {
1980 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
1990 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
1992 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1993 void *data, int offset, unsigned long len)
1998 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1999 if (kvm_is_error_hva(addr))
2001 pagefault_disable();
2002 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
2009 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
2012 gfn_t gfn = gpa >> PAGE_SHIFT;
2013 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2014 int offset = offset_in_page(gpa);
2016 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
2018 EXPORT_SYMBOL_GPL(kvm_read_guest_atomic);
2020 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
2021 void *data, unsigned long len)
2023 gfn_t gfn = gpa >> PAGE_SHIFT;
2024 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2025 int offset = offset_in_page(gpa);
2027 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
2029 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
2031 static int __kvm_write_guest_page(struct kvm_memory_slot *memslot, gfn_t gfn,
2032 const void *data, int offset, int len)
2037 addr = gfn_to_hva_memslot(memslot, gfn);
2038 if (kvm_is_error_hva(addr))
2040 r = __copy_to_user((void __user *)addr + offset, data, len);
2043 mark_page_dirty_in_slot(memslot, gfn);
2047 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
2048 const void *data, int offset, int len)
2050 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2052 return __kvm_write_guest_page(slot, gfn, data, offset, len);
2054 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
2056 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
2057 const void *data, int offset, int len)
2059 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2061 return __kvm_write_guest_page(slot, gfn, data, offset, len);
2063 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
2065 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
2068 gfn_t gfn = gpa >> PAGE_SHIFT;
2070 int offset = offset_in_page(gpa);
2073 while ((seg = next_segment(len, offset)) != 0) {
2074 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
2084 EXPORT_SYMBOL_GPL(kvm_write_guest);
2086 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
2089 gfn_t gfn = gpa >> PAGE_SHIFT;
2091 int offset = offset_in_page(gpa);
2094 while ((seg = next_segment(len, offset)) != 0) {
2095 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
2105 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
2107 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
2108 struct gfn_to_hva_cache *ghc,
2109 gpa_t gpa, unsigned long len)
2111 int offset = offset_in_page(gpa);
2112 gfn_t start_gfn = gpa >> PAGE_SHIFT;
2113 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
2114 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
2115 gfn_t nr_pages_avail;
2116 int r = start_gfn <= end_gfn ? 0 : -EINVAL;
2119 ghc->generation = slots->generation;
2121 ghc->hva = KVM_HVA_ERR_BAD;
2124 * If the requested region crosses two memslots, we still
2125 * verify that the entire region is valid here.
2127 while (!r && start_gfn <= end_gfn) {
2128 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
2129 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
2131 if (kvm_is_error_hva(ghc->hva))
2133 start_gfn += nr_pages_avail;
2136 /* Use the slow path for cross page reads and writes. */
2137 if (!r && nr_pages_needed == 1)
2140 ghc->memslot = NULL;
2145 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2146 gpa_t gpa, unsigned long len)
2148 struct kvm_memslots *slots = kvm_memslots(kvm);
2149 return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len);
2151 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
2153 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2154 void *data, unsigned int offset,
2157 struct kvm_memslots *slots = kvm_memslots(kvm);
2159 gpa_t gpa = ghc->gpa + offset;
2161 BUG_ON(len + offset > ghc->len);
2163 if (slots->generation != ghc->generation)
2164 __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len);
2166 if (unlikely(!ghc->memslot))
2167 return kvm_write_guest(kvm, gpa, data, len);
2169 if (kvm_is_error_hva(ghc->hva))
2172 r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
2175 mark_page_dirty_in_slot(ghc->memslot, gpa >> PAGE_SHIFT);
2179 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached);
2181 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2182 void *data, unsigned long len)
2184 return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len);
2186 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
2188 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2189 void *data, unsigned long len)
2191 struct kvm_memslots *slots = kvm_memslots(kvm);
2194 BUG_ON(len > ghc->len);
2196 if (slots->generation != ghc->generation)
2197 __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len);
2199 if (unlikely(!ghc->memslot))
2200 return kvm_read_guest(kvm, ghc->gpa, data, len);
2202 if (kvm_is_error_hva(ghc->hva))
2205 r = __copy_from_user(data, (void __user *)ghc->hva, len);
2211 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
2213 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
2215 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
2217 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
2219 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
2221 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
2223 gfn_t gfn = gpa >> PAGE_SHIFT;
2225 int offset = offset_in_page(gpa);
2228 while ((seg = next_segment(len, offset)) != 0) {
2229 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
2238 EXPORT_SYMBOL_GPL(kvm_clear_guest);
2240 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot,
2243 if (memslot && memslot->dirty_bitmap) {
2244 unsigned long rel_gfn = gfn - memslot->base_gfn;
2246 set_bit_le(rel_gfn, memslot->dirty_bitmap);
2250 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
2252 struct kvm_memory_slot *memslot;
2254 memslot = gfn_to_memslot(kvm, gfn);
2255 mark_page_dirty_in_slot(memslot, gfn);
2257 EXPORT_SYMBOL_GPL(mark_page_dirty);
2259 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
2261 struct kvm_memory_slot *memslot;
2263 memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2264 mark_page_dirty_in_slot(memslot, gfn);
2266 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
2268 void kvm_sigset_activate(struct kvm_vcpu *vcpu)
2270 if (!vcpu->sigset_active)
2274 * This does a lockless modification of ->real_blocked, which is fine
2275 * because, only current can change ->real_blocked and all readers of
2276 * ->real_blocked don't care as long ->real_blocked is always a subset
2279 sigprocmask(SIG_SETMASK, &vcpu->sigset, ¤t->real_blocked);
2282 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu)
2284 if (!vcpu->sigset_active)
2287 sigprocmask(SIG_SETMASK, ¤t->real_blocked, NULL);
2288 sigemptyset(¤t->real_blocked);
2291 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
2293 unsigned int old, val, grow, grow_start;
2295 old = val = vcpu->halt_poll_ns;
2296 grow_start = READ_ONCE(halt_poll_ns_grow_start);
2297 grow = READ_ONCE(halt_poll_ns_grow);
2302 if (val < grow_start)
2305 if (val > halt_poll_ns)
2308 vcpu->halt_poll_ns = val;
2310 trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
2313 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
2315 unsigned int old, val, shrink;
2317 old = val = vcpu->halt_poll_ns;
2318 shrink = READ_ONCE(halt_poll_ns_shrink);
2324 vcpu->halt_poll_ns = val;
2325 trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
2328 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
2331 int idx = srcu_read_lock(&vcpu->kvm->srcu);
2333 if (kvm_arch_vcpu_runnable(vcpu)) {
2334 kvm_make_request(KVM_REQ_UNHALT, vcpu);
2337 if (kvm_cpu_has_pending_timer(vcpu))
2339 if (signal_pending(current))
2344 srcu_read_unlock(&vcpu->kvm->srcu, idx);
2349 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2351 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
2354 DECLARE_SWAITQUEUE(wait);
2355 bool waited = false;
2358 kvm_arch_vcpu_blocking(vcpu);
2360 start = cur = ktime_get();
2361 if (vcpu->halt_poll_ns && !kvm_arch_no_poll(vcpu)) {
2362 ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns);
2364 ++vcpu->stat.halt_attempted_poll;
2367 * This sets KVM_REQ_UNHALT if an interrupt
2370 if (kvm_vcpu_check_block(vcpu) < 0) {
2371 ++vcpu->stat.halt_successful_poll;
2372 if (!vcpu_valid_wakeup(vcpu))
2373 ++vcpu->stat.halt_poll_invalid;
2377 } while (single_task_running() && ktime_before(cur, stop));
2381 prepare_to_swait_exclusive(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
2383 if (kvm_vcpu_check_block(vcpu) < 0)
2390 finish_swait(&vcpu->wq, &wait);
2393 kvm_arch_vcpu_unblocking(vcpu);
2394 block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
2396 if (!kvm_arch_no_poll(vcpu)) {
2397 if (!vcpu_valid_wakeup(vcpu)) {
2398 shrink_halt_poll_ns(vcpu);
2399 } else if (halt_poll_ns) {
2400 if (block_ns <= vcpu->halt_poll_ns)
2402 /* we had a long block, shrink polling */
2403 else if (vcpu->halt_poll_ns && block_ns > halt_poll_ns)
2404 shrink_halt_poll_ns(vcpu);
2405 /* we had a short halt and our poll time is too small */
2406 else if (vcpu->halt_poll_ns < halt_poll_ns &&
2407 block_ns < halt_poll_ns)
2408 grow_halt_poll_ns(vcpu);
2410 vcpu->halt_poll_ns = 0;
2414 trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu));
2415 kvm_arch_vcpu_block_finish(vcpu);
2417 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
2419 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
2421 struct swait_queue_head *wqp;
2423 wqp = kvm_arch_vcpu_wq(vcpu);
2424 if (swq_has_sleeper(wqp)) {
2426 WRITE_ONCE(vcpu->ready, true);
2427 ++vcpu->stat.halt_wakeup;
2433 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
2437 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2439 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
2442 int cpu = vcpu->cpu;
2444 if (kvm_vcpu_wake_up(vcpu))
2448 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
2449 if (kvm_arch_vcpu_should_kick(vcpu))
2450 smp_send_reschedule(cpu);
2453 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
2454 #endif /* !CONFIG_S390 */
2456 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
2459 struct task_struct *task = NULL;
2463 pid = rcu_dereference(target->pid);
2465 task = get_pid_task(pid, PIDTYPE_PID);
2469 ret = yield_to(task, 1);
2470 put_task_struct(task);
2474 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
2477 * Helper that checks whether a VCPU is eligible for directed yield.
2478 * Most eligible candidate to yield is decided by following heuristics:
2480 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2481 * (preempted lock holder), indicated by @in_spin_loop.
2482 * Set at the beiginning and cleared at the end of interception/PLE handler.
2484 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2485 * chance last time (mostly it has become eligible now since we have probably
2486 * yielded to lockholder in last iteration. This is done by toggling
2487 * @dy_eligible each time a VCPU checked for eligibility.)
2489 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2490 * to preempted lock-holder could result in wrong VCPU selection and CPU
2491 * burning. Giving priority for a potential lock-holder increases lock
2494 * Since algorithm is based on heuristics, accessing another VCPU data without
2495 * locking does not harm. It may result in trying to yield to same VCPU, fail
2496 * and continue with next VCPU and so on.
2498 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
2500 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2503 eligible = !vcpu->spin_loop.in_spin_loop ||
2504 vcpu->spin_loop.dy_eligible;
2506 if (vcpu->spin_loop.in_spin_loop)
2507 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
2516 * Unlike kvm_arch_vcpu_runnable, this function is called outside
2517 * a vcpu_load/vcpu_put pair. However, for most architectures
2518 * kvm_arch_vcpu_runnable does not require vcpu_load.
2520 bool __weak kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
2522 return kvm_arch_vcpu_runnable(vcpu);
2525 static bool vcpu_dy_runnable(struct kvm_vcpu *vcpu)
2527 if (kvm_arch_dy_runnable(vcpu))
2530 #ifdef CONFIG_KVM_ASYNC_PF
2531 if (!list_empty_careful(&vcpu->async_pf.done))
2538 void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode)
2540 struct kvm *kvm = me->kvm;
2541 struct kvm_vcpu *vcpu;
2542 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
2548 kvm_vcpu_set_in_spin_loop(me, true);
2550 * We boost the priority of a VCPU that is runnable but not
2551 * currently running, because it got preempted by something
2552 * else and called schedule in __vcpu_run. Hopefully that
2553 * VCPU is holding the lock that we need and will release it.
2554 * We approximate round-robin by starting at the last boosted VCPU.
2556 for (pass = 0; pass < 2 && !yielded && try; pass++) {
2557 kvm_for_each_vcpu(i, vcpu, kvm) {
2558 if (!pass && i <= last_boosted_vcpu) {
2559 i = last_boosted_vcpu;
2561 } else if (pass && i > last_boosted_vcpu)
2563 if (!READ_ONCE(vcpu->ready))
2567 if (swait_active(&vcpu->wq) && !vcpu_dy_runnable(vcpu))
2569 if (READ_ONCE(vcpu->preempted) && yield_to_kernel_mode &&
2570 !kvm_arch_vcpu_in_kernel(vcpu))
2572 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
2575 yielded = kvm_vcpu_yield_to(vcpu);
2577 kvm->last_boosted_vcpu = i;
2579 } else if (yielded < 0) {
2586 kvm_vcpu_set_in_spin_loop(me, false);
2588 /* Ensure vcpu is not eligible during next spinloop */
2589 kvm_vcpu_set_dy_eligible(me, false);
2591 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
2593 static vm_fault_t kvm_vcpu_fault(struct vm_fault *vmf)
2595 struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data;
2598 if (vmf->pgoff == 0)
2599 page = virt_to_page(vcpu->run);
2601 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
2602 page = virt_to_page(vcpu->arch.pio_data);
2604 #ifdef CONFIG_KVM_MMIO
2605 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
2606 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
2609 return kvm_arch_vcpu_fault(vcpu, vmf);
2615 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
2616 .fault = kvm_vcpu_fault,
2619 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2621 vma->vm_ops = &kvm_vcpu_vm_ops;
2625 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2627 struct kvm_vcpu *vcpu = filp->private_data;
2629 debugfs_remove_recursive(vcpu->debugfs_dentry);
2630 kvm_put_kvm(vcpu->kvm);
2634 static struct file_operations kvm_vcpu_fops = {
2635 .release = kvm_vcpu_release,
2636 .unlocked_ioctl = kvm_vcpu_ioctl,
2637 .mmap = kvm_vcpu_mmap,
2638 .llseek = noop_llseek,
2639 KVM_COMPAT(kvm_vcpu_compat_ioctl),
2643 * Allocates an inode for the vcpu.
2645 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2647 char name[8 + 1 + ITOA_MAX_LEN + 1];
2649 snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id);
2650 return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
2653 static void kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
2655 #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS
2656 char dir_name[ITOA_MAX_LEN * 2];
2658 if (!debugfs_initialized())
2661 snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
2662 vcpu->debugfs_dentry = debugfs_create_dir(dir_name,
2663 vcpu->kvm->debugfs_dentry);
2665 kvm_arch_create_vcpu_debugfs(vcpu);
2670 * Creates some virtual cpus. Good luck creating more than one.
2672 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
2675 struct kvm_vcpu *vcpu;
2677 if (id >= KVM_MAX_VCPU_ID)
2680 mutex_lock(&kvm->lock);
2681 if (kvm->created_vcpus == KVM_MAX_VCPUS) {
2682 mutex_unlock(&kvm->lock);
2686 kvm->created_vcpus++;
2687 mutex_unlock(&kvm->lock);
2689 vcpu = kvm_arch_vcpu_create(kvm, id);
2692 goto vcpu_decrement;
2695 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2697 r = kvm_arch_vcpu_setup(vcpu);
2701 kvm_create_vcpu_debugfs(vcpu);
2703 mutex_lock(&kvm->lock);
2704 if (kvm_get_vcpu_by_id(kvm, id)) {
2706 goto unlock_vcpu_destroy;
2709 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
2711 /* Now it's all set up, let userspace reach it */
2713 r = create_vcpu_fd(vcpu);
2716 goto unlock_vcpu_destroy;
2719 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
2722 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
2723 * before kvm->online_vcpu's incremented value.
2726 atomic_inc(&kvm->online_vcpus);
2728 mutex_unlock(&kvm->lock);
2729 kvm_arch_vcpu_postcreate(vcpu);
2732 unlock_vcpu_destroy:
2733 mutex_unlock(&kvm->lock);
2734 debugfs_remove_recursive(vcpu->debugfs_dentry);
2736 kvm_arch_vcpu_destroy(vcpu);
2738 mutex_lock(&kvm->lock);
2739 kvm->created_vcpus--;
2740 mutex_unlock(&kvm->lock);
2744 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2747 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2748 vcpu->sigset_active = 1;
2749 vcpu->sigset = *sigset;
2751 vcpu->sigset_active = 0;
2755 static long kvm_vcpu_ioctl(struct file *filp,
2756 unsigned int ioctl, unsigned long arg)
2758 struct kvm_vcpu *vcpu = filp->private_data;
2759 void __user *argp = (void __user *)arg;
2761 struct kvm_fpu *fpu = NULL;
2762 struct kvm_sregs *kvm_sregs = NULL;
2764 if (vcpu->kvm->mm != current->mm)
2767 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2771 * Some architectures have vcpu ioctls that are asynchronous to vcpu
2772 * execution; mutex_lock() would break them.
2774 r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg);
2775 if (r != -ENOIOCTLCMD)
2778 if (mutex_lock_killable(&vcpu->mutex))
2786 oldpid = rcu_access_pointer(vcpu->pid);
2787 if (unlikely(oldpid != task_pid(current))) {
2788 /* The thread running this VCPU changed. */
2791 r = kvm_arch_vcpu_run_pid_change(vcpu);
2795 newpid = get_task_pid(current, PIDTYPE_PID);
2796 rcu_assign_pointer(vcpu->pid, newpid);
2801 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2802 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2805 case KVM_GET_REGS: {
2806 struct kvm_regs *kvm_regs;
2809 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL_ACCOUNT);
2812 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2816 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2823 case KVM_SET_REGS: {
2824 struct kvm_regs *kvm_regs;
2827 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2828 if (IS_ERR(kvm_regs)) {
2829 r = PTR_ERR(kvm_regs);
2832 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2836 case KVM_GET_SREGS: {
2837 kvm_sregs = kzalloc(sizeof(struct kvm_sregs),
2838 GFP_KERNEL_ACCOUNT);
2842 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2846 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2851 case KVM_SET_SREGS: {
2852 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2853 if (IS_ERR(kvm_sregs)) {
2854 r = PTR_ERR(kvm_sregs);
2858 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2861 case KVM_GET_MP_STATE: {
2862 struct kvm_mp_state mp_state;
2864 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2868 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
2873 case KVM_SET_MP_STATE: {
2874 struct kvm_mp_state mp_state;
2877 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
2879 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2882 case KVM_TRANSLATE: {
2883 struct kvm_translation tr;
2886 if (copy_from_user(&tr, argp, sizeof(tr)))
2888 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2892 if (copy_to_user(argp, &tr, sizeof(tr)))
2897 case KVM_SET_GUEST_DEBUG: {
2898 struct kvm_guest_debug dbg;
2901 if (copy_from_user(&dbg, argp, sizeof(dbg)))
2903 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2906 case KVM_SET_SIGNAL_MASK: {
2907 struct kvm_signal_mask __user *sigmask_arg = argp;
2908 struct kvm_signal_mask kvm_sigmask;
2909 sigset_t sigset, *p;
2914 if (copy_from_user(&kvm_sigmask, argp,
2915 sizeof(kvm_sigmask)))
2918 if (kvm_sigmask.len != sizeof(sigset))
2921 if (copy_from_user(&sigset, sigmask_arg->sigset,
2926 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2930 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL_ACCOUNT);
2934 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2938 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2944 fpu = memdup_user(argp, sizeof(*fpu));
2950 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2954 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2957 mutex_unlock(&vcpu->mutex);
2963 #ifdef CONFIG_KVM_COMPAT
2964 static long kvm_vcpu_compat_ioctl(struct file *filp,
2965 unsigned int ioctl, unsigned long arg)
2967 struct kvm_vcpu *vcpu = filp->private_data;
2968 void __user *argp = compat_ptr(arg);
2971 if (vcpu->kvm->mm != current->mm)
2975 case KVM_SET_SIGNAL_MASK: {
2976 struct kvm_signal_mask __user *sigmask_arg = argp;
2977 struct kvm_signal_mask kvm_sigmask;
2982 if (copy_from_user(&kvm_sigmask, argp,
2983 sizeof(kvm_sigmask)))
2986 if (kvm_sigmask.len != sizeof(compat_sigset_t))
2989 if (get_compat_sigset(&sigset, (void *)sigmask_arg->sigset))
2991 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2993 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2997 r = kvm_vcpu_ioctl(filp, ioctl, arg);
3005 static int kvm_device_mmap(struct file *filp, struct vm_area_struct *vma)
3007 struct kvm_device *dev = filp->private_data;
3010 return dev->ops->mmap(dev, vma);
3015 static int kvm_device_ioctl_attr(struct kvm_device *dev,
3016 int (*accessor)(struct kvm_device *dev,
3017 struct kvm_device_attr *attr),
3020 struct kvm_device_attr attr;
3025 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
3028 return accessor(dev, &attr);
3031 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
3034 struct kvm_device *dev = filp->private_data;
3036 if (dev->kvm->mm != current->mm)
3040 case KVM_SET_DEVICE_ATTR:
3041 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
3042 case KVM_GET_DEVICE_ATTR:
3043 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
3044 case KVM_HAS_DEVICE_ATTR:
3045 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
3047 if (dev->ops->ioctl)
3048 return dev->ops->ioctl(dev, ioctl, arg);
3054 static int kvm_device_release(struct inode *inode, struct file *filp)
3056 struct kvm_device *dev = filp->private_data;
3057 struct kvm *kvm = dev->kvm;
3059 if (dev->ops->release) {
3060 mutex_lock(&kvm->lock);
3061 list_del(&dev->vm_node);
3062 dev->ops->release(dev);
3063 mutex_unlock(&kvm->lock);
3070 static const struct file_operations kvm_device_fops = {
3071 .unlocked_ioctl = kvm_device_ioctl,
3072 .release = kvm_device_release,
3073 KVM_COMPAT(kvm_device_ioctl),
3074 .mmap = kvm_device_mmap,
3077 struct kvm_device *kvm_device_from_filp(struct file *filp)
3079 if (filp->f_op != &kvm_device_fops)
3082 return filp->private_data;
3085 static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
3086 #ifdef CONFIG_KVM_MPIC
3087 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
3088 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
3092 int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
3094 if (type >= ARRAY_SIZE(kvm_device_ops_table))
3097 if (kvm_device_ops_table[type] != NULL)
3100 kvm_device_ops_table[type] = ops;
3104 void kvm_unregister_device_ops(u32 type)
3106 if (kvm_device_ops_table[type] != NULL)
3107 kvm_device_ops_table[type] = NULL;
3110 static int kvm_ioctl_create_device(struct kvm *kvm,
3111 struct kvm_create_device *cd)
3113 struct kvm_device_ops *ops = NULL;
3114 struct kvm_device *dev;
3115 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
3119 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
3122 type = array_index_nospec(cd->type, ARRAY_SIZE(kvm_device_ops_table));
3123 ops = kvm_device_ops_table[type];
3130 dev = kzalloc(sizeof(*dev), GFP_KERNEL_ACCOUNT);
3137 mutex_lock(&kvm->lock);
3138 ret = ops->create(dev, type);
3140 mutex_unlock(&kvm->lock);
3144 list_add(&dev->vm_node, &kvm->devices);
3145 mutex_unlock(&kvm->lock);
3151 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
3154 mutex_lock(&kvm->lock);
3155 list_del(&dev->vm_node);
3156 mutex_unlock(&kvm->lock);
3165 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
3168 case KVM_CAP_USER_MEMORY:
3169 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
3170 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
3171 case KVM_CAP_INTERNAL_ERROR_DATA:
3172 #ifdef CONFIG_HAVE_KVM_MSI
3173 case KVM_CAP_SIGNAL_MSI:
3175 #ifdef CONFIG_HAVE_KVM_IRQFD
3177 case KVM_CAP_IRQFD_RESAMPLE:
3179 case KVM_CAP_IOEVENTFD_ANY_LENGTH:
3180 case KVM_CAP_CHECK_EXTENSION_VM:
3181 case KVM_CAP_ENABLE_CAP_VM:
3182 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3183 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2:
3186 #ifdef CONFIG_KVM_MMIO
3187 case KVM_CAP_COALESCED_MMIO:
3188 return KVM_COALESCED_MMIO_PAGE_OFFSET;
3189 case KVM_CAP_COALESCED_PIO:
3192 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3193 case KVM_CAP_IRQ_ROUTING:
3194 return KVM_MAX_IRQ_ROUTES;
3196 #if KVM_ADDRESS_SPACE_NUM > 1
3197 case KVM_CAP_MULTI_ADDRESS_SPACE:
3198 return KVM_ADDRESS_SPACE_NUM;
3200 case KVM_CAP_NR_MEMSLOTS:
3201 return KVM_USER_MEM_SLOTS;
3205 return kvm_vm_ioctl_check_extension(kvm, arg);
3208 int __attribute__((weak)) kvm_vm_ioctl_enable_cap(struct kvm *kvm,
3209 struct kvm_enable_cap *cap)
3214 static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm,
3215 struct kvm_enable_cap *cap)
3218 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3219 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2:
3220 if (cap->flags || (cap->args[0] & ~1))
3222 kvm->manual_dirty_log_protect = cap->args[0];
3226 return kvm_vm_ioctl_enable_cap(kvm, cap);
3230 static long kvm_vm_ioctl(struct file *filp,
3231 unsigned int ioctl, unsigned long arg)
3233 struct kvm *kvm = filp->private_data;
3234 void __user *argp = (void __user *)arg;
3237 if (kvm->mm != current->mm)
3240 case KVM_CREATE_VCPU:
3241 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
3243 case KVM_ENABLE_CAP: {
3244 struct kvm_enable_cap cap;
3247 if (copy_from_user(&cap, argp, sizeof(cap)))
3249 r = kvm_vm_ioctl_enable_cap_generic(kvm, &cap);
3252 case KVM_SET_USER_MEMORY_REGION: {
3253 struct kvm_userspace_memory_region kvm_userspace_mem;
3256 if (copy_from_user(&kvm_userspace_mem, argp,
3257 sizeof(kvm_userspace_mem)))
3260 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
3263 case KVM_GET_DIRTY_LOG: {
3264 struct kvm_dirty_log log;
3267 if (copy_from_user(&log, argp, sizeof(log)))
3269 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3272 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3273 case KVM_CLEAR_DIRTY_LOG: {
3274 struct kvm_clear_dirty_log log;
3277 if (copy_from_user(&log, argp, sizeof(log)))
3279 r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
3283 #ifdef CONFIG_KVM_MMIO
3284 case KVM_REGISTER_COALESCED_MMIO: {
3285 struct kvm_coalesced_mmio_zone zone;
3288 if (copy_from_user(&zone, argp, sizeof(zone)))
3290 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
3293 case KVM_UNREGISTER_COALESCED_MMIO: {
3294 struct kvm_coalesced_mmio_zone zone;
3297 if (copy_from_user(&zone, argp, sizeof(zone)))
3299 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
3304 struct kvm_irqfd data;
3307 if (copy_from_user(&data, argp, sizeof(data)))
3309 r = kvm_irqfd(kvm, &data);
3312 case KVM_IOEVENTFD: {
3313 struct kvm_ioeventfd data;
3316 if (copy_from_user(&data, argp, sizeof(data)))
3318 r = kvm_ioeventfd(kvm, &data);
3321 #ifdef CONFIG_HAVE_KVM_MSI
3322 case KVM_SIGNAL_MSI: {
3326 if (copy_from_user(&msi, argp, sizeof(msi)))
3328 r = kvm_send_userspace_msi(kvm, &msi);
3332 #ifdef __KVM_HAVE_IRQ_LINE
3333 case KVM_IRQ_LINE_STATUS:
3334 case KVM_IRQ_LINE: {
3335 struct kvm_irq_level irq_event;
3338 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
3341 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
3342 ioctl == KVM_IRQ_LINE_STATUS);
3347 if (ioctl == KVM_IRQ_LINE_STATUS) {
3348 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
3356 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3357 case KVM_SET_GSI_ROUTING: {
3358 struct kvm_irq_routing routing;
3359 struct kvm_irq_routing __user *urouting;
3360 struct kvm_irq_routing_entry *entries = NULL;
3363 if (copy_from_user(&routing, argp, sizeof(routing)))
3366 if (!kvm_arch_can_set_irq_routing(kvm))
3368 if (routing.nr > KVM_MAX_IRQ_ROUTES)
3374 entries = vmalloc(array_size(sizeof(*entries),
3380 if (copy_from_user(entries, urouting->entries,
3381 routing.nr * sizeof(*entries)))
3382 goto out_free_irq_routing;
3384 r = kvm_set_irq_routing(kvm, entries, routing.nr,
3386 out_free_irq_routing:
3390 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
3391 case KVM_CREATE_DEVICE: {
3392 struct kvm_create_device cd;
3395 if (copy_from_user(&cd, argp, sizeof(cd)))
3398 r = kvm_ioctl_create_device(kvm, &cd);
3403 if (copy_to_user(argp, &cd, sizeof(cd)))
3409 case KVM_CHECK_EXTENSION:
3410 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
3413 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
3419 #ifdef CONFIG_KVM_COMPAT
3420 struct compat_kvm_dirty_log {
3424 compat_uptr_t dirty_bitmap; /* one bit per page */
3429 static long kvm_vm_compat_ioctl(struct file *filp,
3430 unsigned int ioctl, unsigned long arg)
3432 struct kvm *kvm = filp->private_data;
3435 if (kvm->mm != current->mm)
3438 case KVM_GET_DIRTY_LOG: {
3439 struct compat_kvm_dirty_log compat_log;
3440 struct kvm_dirty_log log;
3442 if (copy_from_user(&compat_log, (void __user *)arg,
3443 sizeof(compat_log)))
3445 log.slot = compat_log.slot;
3446 log.padding1 = compat_log.padding1;
3447 log.padding2 = compat_log.padding2;
3448 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
3450 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3454 r = kvm_vm_ioctl(filp, ioctl, arg);
3460 static struct file_operations kvm_vm_fops = {
3461 .release = kvm_vm_release,
3462 .unlocked_ioctl = kvm_vm_ioctl,
3463 .llseek = noop_llseek,
3464 KVM_COMPAT(kvm_vm_compat_ioctl),
3467 static int kvm_dev_ioctl_create_vm(unsigned long type)
3473 kvm = kvm_create_vm(type);
3475 return PTR_ERR(kvm);
3476 #ifdef CONFIG_KVM_MMIO
3477 r = kvm_coalesced_mmio_init(kvm);
3481 r = get_unused_fd_flags(O_CLOEXEC);
3485 file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
3493 * Don't call kvm_put_kvm anymore at this point; file->f_op is
3494 * already set, with ->release() being kvm_vm_release(). In error
3495 * cases it will be called by the final fput(file) and will take
3496 * care of doing kvm_put_kvm(kvm).
3498 if (kvm_create_vm_debugfs(kvm, r) < 0) {
3503 kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm);
3505 fd_install(r, file);
3513 static long kvm_dev_ioctl(struct file *filp,
3514 unsigned int ioctl, unsigned long arg)
3519 case KVM_GET_API_VERSION:
3522 r = KVM_API_VERSION;
3525 r = kvm_dev_ioctl_create_vm(arg);
3527 case KVM_CHECK_EXTENSION:
3528 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
3530 case KVM_GET_VCPU_MMAP_SIZE:
3533 r = PAGE_SIZE; /* struct kvm_run */
3535 r += PAGE_SIZE; /* pio data page */
3537 #ifdef CONFIG_KVM_MMIO
3538 r += PAGE_SIZE; /* coalesced mmio ring page */
3541 case KVM_TRACE_ENABLE:
3542 case KVM_TRACE_PAUSE:
3543 case KVM_TRACE_DISABLE:
3547 return kvm_arch_dev_ioctl(filp, ioctl, arg);
3553 static struct file_operations kvm_chardev_ops = {
3554 .unlocked_ioctl = kvm_dev_ioctl,
3555 .llseek = noop_llseek,
3556 KVM_COMPAT(kvm_dev_ioctl),
3559 static struct miscdevice kvm_dev = {
3565 static void hardware_enable_nolock(void *junk)
3567 int cpu = raw_smp_processor_id();
3570 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
3573 cpumask_set_cpu(cpu, cpus_hardware_enabled);
3575 r = kvm_arch_hardware_enable();
3578 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3579 atomic_inc(&hardware_enable_failed);
3580 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
3584 static int kvm_starting_cpu(unsigned int cpu)
3586 raw_spin_lock(&kvm_count_lock);
3587 if (kvm_usage_count)
3588 hardware_enable_nolock(NULL);
3589 raw_spin_unlock(&kvm_count_lock);
3593 static void hardware_disable_nolock(void *junk)
3595 int cpu = raw_smp_processor_id();
3597 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
3599 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3600 kvm_arch_hardware_disable();
3603 static int kvm_dying_cpu(unsigned int cpu)
3605 raw_spin_lock(&kvm_count_lock);
3606 if (kvm_usage_count)
3607 hardware_disable_nolock(NULL);
3608 raw_spin_unlock(&kvm_count_lock);
3612 static void hardware_disable_all_nolock(void)
3614 BUG_ON(!kvm_usage_count);
3617 if (!kvm_usage_count)
3618 on_each_cpu(hardware_disable_nolock, NULL, 1);
3621 static void hardware_disable_all(void)
3623 raw_spin_lock(&kvm_count_lock);
3624 hardware_disable_all_nolock();
3625 raw_spin_unlock(&kvm_count_lock);
3628 static int hardware_enable_all(void)
3632 raw_spin_lock(&kvm_count_lock);
3635 if (kvm_usage_count == 1) {
3636 atomic_set(&hardware_enable_failed, 0);
3637 on_each_cpu(hardware_enable_nolock, NULL, 1);
3639 if (atomic_read(&hardware_enable_failed)) {
3640 hardware_disable_all_nolock();
3645 raw_spin_unlock(&kvm_count_lock);
3650 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3654 * Some (well, at least mine) BIOSes hang on reboot if
3657 * And Intel TXT required VMX off for all cpu when system shutdown.
3659 pr_info("kvm: exiting hardware virtualization\n");
3660 kvm_rebooting = true;
3661 on_each_cpu(hardware_disable_nolock, NULL, 1);
3665 static struct notifier_block kvm_reboot_notifier = {
3666 .notifier_call = kvm_reboot,
3670 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3674 for (i = 0; i < bus->dev_count; i++) {
3675 struct kvm_io_device *pos = bus->range[i].dev;
3677 kvm_iodevice_destructor(pos);
3682 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
3683 const struct kvm_io_range *r2)
3685 gpa_t addr1 = r1->addr;
3686 gpa_t addr2 = r2->addr;
3691 /* If r2->len == 0, match the exact address. If r2->len != 0,
3692 * accept any overlapping write. Any order is acceptable for
3693 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
3694 * we process all of them.
3707 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
3709 return kvm_io_bus_cmp(p1, p2);
3712 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
3713 gpa_t addr, int len)
3715 struct kvm_io_range *range, key;
3718 key = (struct kvm_io_range) {
3723 range = bsearch(&key, bus->range, bus->dev_count,
3724 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
3728 off = range - bus->range;
3730 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
3736 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3737 struct kvm_io_range *range, const void *val)
3741 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3745 while (idx < bus->dev_count &&
3746 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3747 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
3756 /* kvm_io_bus_write - called under kvm->slots_lock */
3757 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3758 int len, const void *val)
3760 struct kvm_io_bus *bus;
3761 struct kvm_io_range range;
3764 range = (struct kvm_io_range) {
3769 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3772 r = __kvm_io_bus_write(vcpu, bus, &range, val);
3773 return r < 0 ? r : 0;
3775 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
3777 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3778 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
3779 gpa_t addr, int len, const void *val, long cookie)
3781 struct kvm_io_bus *bus;
3782 struct kvm_io_range range;
3784 range = (struct kvm_io_range) {
3789 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3793 /* First try the device referenced by cookie. */
3794 if ((cookie >= 0) && (cookie < bus->dev_count) &&
3795 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
3796 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
3801 * cookie contained garbage; fall back to search and return the
3802 * correct cookie value.
3804 return __kvm_io_bus_write(vcpu, bus, &range, val);
3807 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3808 struct kvm_io_range *range, void *val)
3812 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3816 while (idx < bus->dev_count &&
3817 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3818 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
3827 /* kvm_io_bus_read - called under kvm->slots_lock */
3828 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3831 struct kvm_io_bus *bus;
3832 struct kvm_io_range range;
3835 range = (struct kvm_io_range) {
3840 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3843 r = __kvm_io_bus_read(vcpu, bus, &range, val);
3844 return r < 0 ? r : 0;
3847 /* Caller must hold slots_lock. */
3848 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3849 int len, struct kvm_io_device *dev)
3852 struct kvm_io_bus *new_bus, *bus;
3853 struct kvm_io_range range;
3855 bus = kvm_get_bus(kvm, bus_idx);
3859 /* exclude ioeventfd which is limited by maximum fd */
3860 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3863 new_bus = kmalloc(struct_size(bus, range, bus->dev_count + 1),
3864 GFP_KERNEL_ACCOUNT);
3868 range = (struct kvm_io_range) {
3874 for (i = 0; i < bus->dev_count; i++)
3875 if (kvm_io_bus_cmp(&bus->range[i], &range) > 0)
3878 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3879 new_bus->dev_count++;
3880 new_bus->range[i] = range;
3881 memcpy(new_bus->range + i + 1, bus->range + i,
3882 (bus->dev_count - i) * sizeof(struct kvm_io_range));
3883 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3884 synchronize_srcu_expedited(&kvm->srcu);
3890 /* Caller must hold slots_lock. */
3891 void kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3892 struct kvm_io_device *dev)
3895 struct kvm_io_bus *new_bus, *bus;
3897 bus = kvm_get_bus(kvm, bus_idx);
3901 for (i = 0; i < bus->dev_count; i++)
3902 if (bus->range[i].dev == dev) {
3906 if (i == bus->dev_count)
3909 new_bus = kmalloc(struct_size(bus, range, bus->dev_count - 1),
3910 GFP_KERNEL_ACCOUNT);
3912 pr_err("kvm: failed to shrink bus, removing it completely\n");
3916 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3917 new_bus->dev_count--;
3918 memcpy(new_bus->range + i, bus->range + i + 1,
3919 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3922 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3923 synchronize_srcu_expedited(&kvm->srcu);
3928 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3931 struct kvm_io_bus *bus;
3932 int dev_idx, srcu_idx;
3933 struct kvm_io_device *iodev = NULL;
3935 srcu_idx = srcu_read_lock(&kvm->srcu);
3937 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
3941 dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
3945 iodev = bus->range[dev_idx].dev;
3948 srcu_read_unlock(&kvm->srcu, srcu_idx);
3952 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
3954 static int kvm_debugfs_open(struct inode *inode, struct file *file,
3955 int (*get)(void *, u64 *), int (*set)(void *, u64),
3958 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
3961 /* The debugfs files are a reference to the kvm struct which
3962 * is still valid when kvm_destroy_vm is called.
3963 * To avoid the race between open and the removal of the debugfs
3964 * directory we test against the users count.
3966 if (!refcount_inc_not_zero(&stat_data->kvm->users_count))
3969 if (simple_attr_open(inode, file, get,
3970 stat_data->mode & S_IWUGO ? set : NULL,
3972 kvm_put_kvm(stat_data->kvm);
3979 static int kvm_debugfs_release(struct inode *inode, struct file *file)
3981 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
3984 simple_attr_release(inode, file);
3985 kvm_put_kvm(stat_data->kvm);
3990 static int vm_stat_get_per_vm(void *data, u64 *val)
3992 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3994 *val = *(ulong *)((void *)stat_data->kvm + stat_data->offset);
3999 static int vm_stat_clear_per_vm(void *data, u64 val)
4001 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
4006 *(ulong *)((void *)stat_data->kvm + stat_data->offset) = 0;
4011 static int vm_stat_get_per_vm_open(struct inode *inode, struct file *file)
4013 __simple_attr_check_format("%llu\n", 0ull);
4014 return kvm_debugfs_open(inode, file, vm_stat_get_per_vm,
4015 vm_stat_clear_per_vm, "%llu\n");
4018 static const struct file_operations vm_stat_get_per_vm_fops = {
4019 .owner = THIS_MODULE,
4020 .open = vm_stat_get_per_vm_open,
4021 .release = kvm_debugfs_release,
4022 .read = simple_attr_read,
4023 .write = simple_attr_write,
4024 .llseek = no_llseek,
4027 static int vcpu_stat_get_per_vm(void *data, u64 *val)
4030 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
4031 struct kvm_vcpu *vcpu;
4035 kvm_for_each_vcpu(i, vcpu, stat_data->kvm)
4036 *val += *(u64 *)((void *)vcpu + stat_data->offset);
4041 static int vcpu_stat_clear_per_vm(void *data, u64 val)
4044 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
4045 struct kvm_vcpu *vcpu;
4050 kvm_for_each_vcpu(i, vcpu, stat_data->kvm)
4051 *(u64 *)((void *)vcpu + stat_data->offset) = 0;
4056 static int vcpu_stat_get_per_vm_open(struct inode *inode, struct file *file)
4058 __simple_attr_check_format("%llu\n", 0ull);
4059 return kvm_debugfs_open(inode, file, vcpu_stat_get_per_vm,
4060 vcpu_stat_clear_per_vm, "%llu\n");
4063 static const struct file_operations vcpu_stat_get_per_vm_fops = {
4064 .owner = THIS_MODULE,
4065 .open = vcpu_stat_get_per_vm_open,
4066 .release = kvm_debugfs_release,
4067 .read = simple_attr_read,
4068 .write = simple_attr_write,
4069 .llseek = no_llseek,
4072 static const struct file_operations *stat_fops_per_vm[] = {
4073 [KVM_STAT_VCPU] = &vcpu_stat_get_per_vm_fops,
4074 [KVM_STAT_VM] = &vm_stat_get_per_vm_fops,
4077 static int vm_stat_get(void *_offset, u64 *val)
4079 unsigned offset = (long)_offset;
4081 struct kvm_stat_data stat_tmp = {.offset = offset};
4085 mutex_lock(&kvm_lock);
4086 list_for_each_entry(kvm, &vm_list, vm_list) {
4088 vm_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
4091 mutex_unlock(&kvm_lock);
4095 static int vm_stat_clear(void *_offset, u64 val)
4097 unsigned offset = (long)_offset;
4099 struct kvm_stat_data stat_tmp = {.offset = offset};
4104 mutex_lock(&kvm_lock);
4105 list_for_each_entry(kvm, &vm_list, vm_list) {
4107 vm_stat_clear_per_vm((void *)&stat_tmp, 0);
4109 mutex_unlock(&kvm_lock);
4114 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n");
4116 static int vcpu_stat_get(void *_offset, u64 *val)
4118 unsigned offset = (long)_offset;
4120 struct kvm_stat_data stat_tmp = {.offset = offset};
4124 mutex_lock(&kvm_lock);
4125 list_for_each_entry(kvm, &vm_list, vm_list) {
4127 vcpu_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
4130 mutex_unlock(&kvm_lock);
4134 static int vcpu_stat_clear(void *_offset, u64 val)
4136 unsigned offset = (long)_offset;
4138 struct kvm_stat_data stat_tmp = {.offset = offset};
4143 mutex_lock(&kvm_lock);
4144 list_for_each_entry(kvm, &vm_list, vm_list) {
4146 vcpu_stat_clear_per_vm((void *)&stat_tmp, 0);
4148 mutex_unlock(&kvm_lock);
4153 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear,
4156 static const struct file_operations *stat_fops[] = {
4157 [KVM_STAT_VCPU] = &vcpu_stat_fops,
4158 [KVM_STAT_VM] = &vm_stat_fops,
4161 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm)
4163 struct kobj_uevent_env *env;
4164 unsigned long long created, active;
4166 if (!kvm_dev.this_device || !kvm)
4169 mutex_lock(&kvm_lock);
4170 if (type == KVM_EVENT_CREATE_VM) {
4171 kvm_createvm_count++;
4173 } else if (type == KVM_EVENT_DESTROY_VM) {
4176 created = kvm_createvm_count;
4177 active = kvm_active_vms;
4178 mutex_unlock(&kvm_lock);
4180 env = kzalloc(sizeof(*env), GFP_KERNEL_ACCOUNT);
4184 add_uevent_var(env, "CREATED=%llu", created);
4185 add_uevent_var(env, "COUNT=%llu", active);
4187 if (type == KVM_EVENT_CREATE_VM) {
4188 add_uevent_var(env, "EVENT=create");
4189 kvm->userspace_pid = task_pid_nr(current);
4190 } else if (type == KVM_EVENT_DESTROY_VM) {
4191 add_uevent_var(env, "EVENT=destroy");
4193 add_uevent_var(env, "PID=%d", kvm->userspace_pid);
4195 if (!IS_ERR_OR_NULL(kvm->debugfs_dentry)) {
4196 char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL_ACCOUNT);
4199 tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX);
4201 add_uevent_var(env, "STATS_PATH=%s", tmp);
4205 /* no need for checks, since we are adding at most only 5 keys */
4206 env->envp[env->envp_idx++] = NULL;
4207 kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp);
4211 static void kvm_init_debug(void)
4213 struct kvm_stats_debugfs_item *p;
4215 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
4217 kvm_debugfs_num_entries = 0;
4218 for (p = debugfs_entries; p->name; ++p, kvm_debugfs_num_entries++) {
4219 int mode = p->mode ? p->mode : 0644;
4220 debugfs_create_file(p->name, mode, kvm_debugfs_dir,
4221 (void *)(long)p->offset,
4222 stat_fops[p->kind]);
4226 static int kvm_suspend(void)
4228 if (kvm_usage_count)
4229 hardware_disable_nolock(NULL);
4233 static void kvm_resume(void)
4235 if (kvm_usage_count) {
4236 #ifdef CONFIG_LOCKDEP
4237 WARN_ON(lockdep_is_held(&kvm_count_lock));
4239 hardware_enable_nolock(NULL);
4243 static struct syscore_ops kvm_syscore_ops = {
4244 .suspend = kvm_suspend,
4245 .resume = kvm_resume,
4249 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
4251 return container_of(pn, struct kvm_vcpu, preempt_notifier);
4254 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
4256 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
4258 WRITE_ONCE(vcpu->preempted, false);
4259 WRITE_ONCE(vcpu->ready, false);
4261 kvm_arch_sched_in(vcpu, cpu);
4263 kvm_arch_vcpu_load(vcpu, cpu);
4266 static void kvm_sched_out(struct preempt_notifier *pn,
4267 struct task_struct *next)
4269 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
4271 if (current->state == TASK_RUNNING) {
4272 WRITE_ONCE(vcpu->preempted, true);
4273 WRITE_ONCE(vcpu->ready, true);
4275 kvm_arch_vcpu_put(vcpu);
4278 static void check_processor_compat(void *rtn)
4280 *(int *)rtn = kvm_arch_check_processor_compat();
4283 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
4284 struct module *module)
4289 r = kvm_arch_init(opaque);
4294 * kvm_arch_init makes sure there's at most one caller
4295 * for architectures that support multiple implementations,
4296 * like intel and amd on x86.
4297 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
4298 * conflicts in case kvm is already setup for another implementation.
4300 r = kvm_irqfd_init();
4304 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
4309 r = kvm_arch_hardware_setup();
4313 for_each_online_cpu(cpu) {
4314 smp_call_function_single(cpu, check_processor_compat, &r, 1);
4319 r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting",
4320 kvm_starting_cpu, kvm_dying_cpu);
4323 register_reboot_notifier(&kvm_reboot_notifier);
4325 /* A kmem cache lets us meet the alignment requirements of fx_save. */
4327 vcpu_align = __alignof__(struct kvm_vcpu);
4329 kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align,
4331 offsetof(struct kvm_vcpu, arch),
4332 sizeof_field(struct kvm_vcpu, arch),
4334 if (!kvm_vcpu_cache) {
4339 r = kvm_async_pf_init();
4343 kvm_chardev_ops.owner = module;
4344 kvm_vm_fops.owner = module;
4345 kvm_vcpu_fops.owner = module;
4347 r = misc_register(&kvm_dev);
4349 pr_err("kvm: misc device register failed\n");
4353 register_syscore_ops(&kvm_syscore_ops);
4355 kvm_preempt_ops.sched_in = kvm_sched_in;
4356 kvm_preempt_ops.sched_out = kvm_sched_out;
4360 r = kvm_vfio_ops_init();
4366 kvm_async_pf_deinit();
4368 kmem_cache_destroy(kvm_vcpu_cache);
4370 unregister_reboot_notifier(&kvm_reboot_notifier);
4371 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4374 kvm_arch_hardware_unsetup();
4376 free_cpumask_var(cpus_hardware_enabled);
4384 EXPORT_SYMBOL_GPL(kvm_init);
4388 debugfs_remove_recursive(kvm_debugfs_dir);
4389 misc_deregister(&kvm_dev);
4390 kmem_cache_destroy(kvm_vcpu_cache);
4391 kvm_async_pf_deinit();
4392 unregister_syscore_ops(&kvm_syscore_ops);
4393 unregister_reboot_notifier(&kvm_reboot_notifier);
4394 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4395 on_each_cpu(hardware_disable_nolock, NULL, 1);
4396 kvm_arch_hardware_unsetup();
4399 free_cpumask_var(cpus_hardware_enabled);
4400 kvm_vfio_ops_exit();
4402 EXPORT_SYMBOL_GPL(kvm_exit);
4404 struct kvm_vm_worker_thread_context {
4406 struct task_struct *parent;
4407 struct completion init_done;
4408 kvm_vm_thread_fn_t thread_fn;
4413 static int kvm_vm_worker_thread(void *context)
4416 * The init_context is allocated on the stack of the parent thread, so
4417 * we have to locally copy anything that is needed beyond initialization
4419 struct kvm_vm_worker_thread_context *init_context = context;
4420 struct kvm *kvm = init_context->kvm;
4421 kvm_vm_thread_fn_t thread_fn = init_context->thread_fn;
4422 uintptr_t data = init_context->data;
4425 err = kthread_park(current);
4426 /* kthread_park(current) is never supposed to return an error */
4431 err = cgroup_attach_task_all(init_context->parent, current);
4433 kvm_err("%s: cgroup_attach_task_all failed with err %d\n",
4438 set_user_nice(current, task_nice(init_context->parent));
4441 init_context->err = err;
4442 complete(&init_context->init_done);
4443 init_context = NULL;
4448 /* Wait to be woken up by the spawner before proceeding. */
4451 if (!kthread_should_stop())
4452 err = thread_fn(kvm, data);
4457 int kvm_vm_create_worker_thread(struct kvm *kvm, kvm_vm_thread_fn_t thread_fn,
4458 uintptr_t data, const char *name,
4459 struct task_struct **thread_ptr)
4461 struct kvm_vm_worker_thread_context init_context = {};
4462 struct task_struct *thread;
4465 init_context.kvm = kvm;
4466 init_context.parent = current;
4467 init_context.thread_fn = thread_fn;
4468 init_context.data = data;
4469 init_completion(&init_context.init_done);
4471 thread = kthread_run(kvm_vm_worker_thread, &init_context,
4472 "%s-%d", name, task_pid_nr(current));
4474 return PTR_ERR(thread);
4476 /* kthread_run is never supposed to return NULL */
4477 WARN_ON(thread == NULL);
4479 wait_for_completion(&init_context.init_done);
4481 if (!init_context.err)
4482 *thread_ptr = thread;
4484 return init_context.err;