2 * Kernel-based Virtual Machine driver for Linux
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
7 * Copyright (C) 2006 Qumranet, Inc.
8 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
11 * Avi Kivity <avi@qumranet.com>
12 * Yaniv Kamay <yaniv@qumranet.com>
14 * This work is licensed under the terms of the GNU GPL, version 2. See
15 * the COPYING file in the top-level directory.
19 #include <kvm/iodev.h>
21 #include <linux/kvm_host.h>
22 #include <linux/kvm.h>
23 #include <linux/module.h>
24 #include <linux/errno.h>
25 #include <linux/percpu.h>
27 #include <linux/miscdevice.h>
28 #include <linux/vmalloc.h>
29 #include <linux/reboot.h>
30 #include <linux/debugfs.h>
31 #include <linux/highmem.h>
32 #include <linux/file.h>
33 #include <linux/syscore_ops.h>
34 #include <linux/cpu.h>
35 #include <linux/sched/signal.h>
36 #include <linux/sched/mm.h>
37 #include <linux/sched/stat.h>
38 #include <linux/cpumask.h>
39 #include <linux/smp.h>
40 #include <linux/anon_inodes.h>
41 #include <linux/profile.h>
42 #include <linux/kvm_para.h>
43 #include <linux/pagemap.h>
44 #include <linux/mman.h>
45 #include <linux/swap.h>
46 #include <linux/bitops.h>
47 #include <linux/spinlock.h>
48 #include <linux/compat.h>
49 #include <linux/srcu.h>
50 #include <linux/hugetlb.h>
51 #include <linux/slab.h>
52 #include <linux/sort.h>
53 #include <linux/bsearch.h>
55 #include <linux/lockdep.h>
57 #include <asm/processor.h>
58 #include <asm/ioctl.h>
59 #include <linux/uaccess.h>
60 #include <asm/pgtable.h>
62 #include "coalesced_mmio.h"
66 #define CREATE_TRACE_POINTS
67 #include <trace/events/kvm.h>
69 /* Worst case buffer size needed for holding an integer. */
70 #define ITOA_MAX_LEN 12
72 MODULE_AUTHOR("Qumranet");
73 MODULE_LICENSE("GPL");
75 /* Architectures should define their poll value according to the halt latency */
76 unsigned int halt_poll_ns = KVM_HALT_POLL_NS_DEFAULT;
77 module_param(halt_poll_ns, uint, 0644);
78 EXPORT_SYMBOL_GPL(halt_poll_ns);
80 /* Default doubles per-vcpu halt_poll_ns. */
81 unsigned int halt_poll_ns_grow = 2;
82 module_param(halt_poll_ns_grow, uint, 0644);
83 EXPORT_SYMBOL_GPL(halt_poll_ns_grow);
85 /* The start value to grow halt_poll_ns from */
86 unsigned int halt_poll_ns_grow_start = 10000; /* 10us */
87 module_param(halt_poll_ns_grow_start, uint, 0644);
88 EXPORT_SYMBOL_GPL(halt_poll_ns_grow_start);
90 /* Default resets per-vcpu halt_poll_ns . */
91 unsigned int halt_poll_ns_shrink;
92 module_param(halt_poll_ns_shrink, uint, 0644);
93 EXPORT_SYMBOL_GPL(halt_poll_ns_shrink);
98 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
101 DEFINE_SPINLOCK(kvm_lock);
102 static DEFINE_RAW_SPINLOCK(kvm_count_lock);
105 static cpumask_var_t cpus_hardware_enabled;
106 static int kvm_usage_count;
107 static atomic_t hardware_enable_failed;
109 struct kmem_cache *kvm_vcpu_cache;
110 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
112 static __read_mostly struct preempt_ops kvm_preempt_ops;
114 struct dentry *kvm_debugfs_dir;
115 EXPORT_SYMBOL_GPL(kvm_debugfs_dir);
117 static int kvm_debugfs_num_entries;
118 static const struct file_operations *stat_fops_per_vm[];
120 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
122 #ifdef CONFIG_KVM_COMPAT
123 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
125 #define KVM_COMPAT(c) .compat_ioctl = (c)
127 static long kvm_no_compat_ioctl(struct file *file, unsigned int ioctl,
128 unsigned long arg) { return -EINVAL; }
129 #define KVM_COMPAT(c) .compat_ioctl = kvm_no_compat_ioctl
131 static int hardware_enable_all(void);
132 static void hardware_disable_all(void);
134 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
136 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot, gfn_t gfn);
138 __visible bool kvm_rebooting;
139 EXPORT_SYMBOL_GPL(kvm_rebooting);
141 static bool largepages_enabled = true;
143 #define KVM_EVENT_CREATE_VM 0
144 #define KVM_EVENT_DESTROY_VM 1
145 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm);
146 static unsigned long long kvm_createvm_count;
147 static unsigned long long kvm_active_vms;
149 __weak int kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
150 unsigned long start, unsigned long end, bool blockable)
155 bool kvm_is_reserved_pfn(kvm_pfn_t pfn)
158 return PageReserved(pfn_to_page(pfn));
164 * Switches to specified vcpu, until a matching vcpu_put()
166 void vcpu_load(struct kvm_vcpu *vcpu)
169 preempt_notifier_register(&vcpu->preempt_notifier);
170 kvm_arch_vcpu_load(vcpu, cpu);
173 EXPORT_SYMBOL_GPL(vcpu_load);
175 void vcpu_put(struct kvm_vcpu *vcpu)
178 kvm_arch_vcpu_put(vcpu);
179 preempt_notifier_unregister(&vcpu->preempt_notifier);
182 EXPORT_SYMBOL_GPL(vcpu_put);
184 /* TODO: merge with kvm_arch_vcpu_should_kick */
185 static bool kvm_request_needs_ipi(struct kvm_vcpu *vcpu, unsigned req)
187 int mode = kvm_vcpu_exiting_guest_mode(vcpu);
190 * We need to wait for the VCPU to reenable interrupts and get out of
191 * READING_SHADOW_PAGE_TABLES mode.
193 if (req & KVM_REQUEST_WAIT)
194 return mode != OUTSIDE_GUEST_MODE;
197 * Need to kick a running VCPU, but otherwise there is nothing to do.
199 return mode == IN_GUEST_MODE;
202 static void ack_flush(void *_completed)
206 static inline bool kvm_kick_many_cpus(const struct cpumask *cpus, bool wait)
209 cpus = cpu_online_mask;
211 if (cpumask_empty(cpus))
214 smp_call_function_many(cpus, ack_flush, NULL, wait);
218 bool kvm_make_vcpus_request_mask(struct kvm *kvm, unsigned int req,
219 unsigned long *vcpu_bitmap, cpumask_var_t tmp)
222 struct kvm_vcpu *vcpu;
227 kvm_for_each_vcpu(i, vcpu, kvm) {
228 if (vcpu_bitmap && !test_bit(i, vcpu_bitmap))
231 kvm_make_request(req, vcpu);
234 if (!(req & KVM_REQUEST_NO_WAKEUP) && kvm_vcpu_wake_up(vcpu))
237 if (tmp != NULL && cpu != -1 && cpu != me &&
238 kvm_request_needs_ipi(vcpu, req))
239 __cpumask_set_cpu(cpu, tmp);
242 called = kvm_kick_many_cpus(tmp, !!(req & KVM_REQUEST_WAIT));
248 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
253 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
255 called = kvm_make_vcpus_request_mask(kvm, req, NULL, cpus);
257 free_cpumask_var(cpus);
261 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
262 void kvm_flush_remote_tlbs(struct kvm *kvm)
265 * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in
266 * kvm_make_all_cpus_request.
268 long dirty_count = smp_load_acquire(&kvm->tlbs_dirty);
271 * We want to publish modifications to the page tables before reading
272 * mode. Pairs with a memory barrier in arch-specific code.
273 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
274 * and smp_mb in walk_shadow_page_lockless_begin/end.
275 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
277 * There is already an smp_mb__after_atomic() before
278 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
281 if (!kvm_arch_flush_remote_tlb(kvm)
282 || kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
283 ++kvm->stat.remote_tlb_flush;
284 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
286 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
289 void kvm_reload_remote_mmus(struct kvm *kvm)
291 kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
294 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
299 mutex_init(&vcpu->mutex);
304 init_swait_queue_head(&vcpu->wq);
305 kvm_async_pf_vcpu_init(vcpu);
308 INIT_LIST_HEAD(&vcpu->blocked_vcpu_list);
310 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
315 vcpu->run = page_address(page);
317 kvm_vcpu_set_in_spin_loop(vcpu, false);
318 kvm_vcpu_set_dy_eligible(vcpu, false);
319 vcpu->preempted = false;
321 r = kvm_arch_vcpu_init(vcpu);
327 free_page((unsigned long)vcpu->run);
331 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
333 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
336 * no need for rcu_read_lock as VCPU_RUN is the only place that
337 * will change the vcpu->pid pointer and on uninit all file
338 * descriptors are already gone.
340 put_pid(rcu_dereference_protected(vcpu->pid, 1));
341 kvm_arch_vcpu_uninit(vcpu);
342 free_page((unsigned long)vcpu->run);
344 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
346 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
347 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
349 return container_of(mn, struct kvm, mmu_notifier);
352 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
353 struct mm_struct *mm,
354 unsigned long address,
357 struct kvm *kvm = mmu_notifier_to_kvm(mn);
360 idx = srcu_read_lock(&kvm->srcu);
361 spin_lock(&kvm->mmu_lock);
362 kvm->mmu_notifier_seq++;
364 if (kvm_set_spte_hva(kvm, address, pte))
365 kvm_flush_remote_tlbs(kvm);
367 spin_unlock(&kvm->mmu_lock);
368 srcu_read_unlock(&kvm->srcu, idx);
371 static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
372 const struct mmu_notifier_range *range)
374 struct kvm *kvm = mmu_notifier_to_kvm(mn);
375 int need_tlb_flush = 0, idx;
378 idx = srcu_read_lock(&kvm->srcu);
379 spin_lock(&kvm->mmu_lock);
381 * The count increase must become visible at unlock time as no
382 * spte can be established without taking the mmu_lock and
383 * count is also read inside the mmu_lock critical section.
385 kvm->mmu_notifier_count++;
386 need_tlb_flush = kvm_unmap_hva_range(kvm, range->start, range->end);
387 need_tlb_flush |= kvm->tlbs_dirty;
388 /* we've to flush the tlb before the pages can be freed */
390 kvm_flush_remote_tlbs(kvm);
392 spin_unlock(&kvm->mmu_lock);
394 ret = kvm_arch_mmu_notifier_invalidate_range(kvm, range->start,
396 mmu_notifier_range_blockable(range));
398 srcu_read_unlock(&kvm->srcu, idx);
403 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
404 const struct mmu_notifier_range *range)
406 struct kvm *kvm = mmu_notifier_to_kvm(mn);
408 spin_lock(&kvm->mmu_lock);
410 * This sequence increase will notify the kvm page fault that
411 * the page that is going to be mapped in the spte could have
414 kvm->mmu_notifier_seq++;
417 * The above sequence increase must be visible before the
418 * below count decrease, which is ensured by the smp_wmb above
419 * in conjunction with the smp_rmb in mmu_notifier_retry().
421 kvm->mmu_notifier_count--;
422 spin_unlock(&kvm->mmu_lock);
424 BUG_ON(kvm->mmu_notifier_count < 0);
427 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
428 struct mm_struct *mm,
432 struct kvm *kvm = mmu_notifier_to_kvm(mn);
435 idx = srcu_read_lock(&kvm->srcu);
436 spin_lock(&kvm->mmu_lock);
438 young = kvm_age_hva(kvm, start, end);
440 kvm_flush_remote_tlbs(kvm);
442 spin_unlock(&kvm->mmu_lock);
443 srcu_read_unlock(&kvm->srcu, idx);
448 static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
449 struct mm_struct *mm,
453 struct kvm *kvm = mmu_notifier_to_kvm(mn);
456 idx = srcu_read_lock(&kvm->srcu);
457 spin_lock(&kvm->mmu_lock);
459 * Even though we do not flush TLB, this will still adversely
460 * affect performance on pre-Haswell Intel EPT, where there is
461 * no EPT Access Bit to clear so that we have to tear down EPT
462 * tables instead. If we find this unacceptable, we can always
463 * add a parameter to kvm_age_hva so that it effectively doesn't
464 * do anything on clear_young.
466 * Also note that currently we never issue secondary TLB flushes
467 * from clear_young, leaving this job up to the regular system
468 * cadence. If we find this inaccurate, we might come up with a
469 * more sophisticated heuristic later.
471 young = kvm_age_hva(kvm, start, end);
472 spin_unlock(&kvm->mmu_lock);
473 srcu_read_unlock(&kvm->srcu, idx);
478 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
479 struct mm_struct *mm,
480 unsigned long address)
482 struct kvm *kvm = mmu_notifier_to_kvm(mn);
485 idx = srcu_read_lock(&kvm->srcu);
486 spin_lock(&kvm->mmu_lock);
487 young = kvm_test_age_hva(kvm, address);
488 spin_unlock(&kvm->mmu_lock);
489 srcu_read_unlock(&kvm->srcu, idx);
494 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
495 struct mm_struct *mm)
497 struct kvm *kvm = mmu_notifier_to_kvm(mn);
500 idx = srcu_read_lock(&kvm->srcu);
501 kvm_arch_flush_shadow_all(kvm);
502 srcu_read_unlock(&kvm->srcu, idx);
505 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
506 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
507 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
508 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
509 .clear_young = kvm_mmu_notifier_clear_young,
510 .test_young = kvm_mmu_notifier_test_young,
511 .change_pte = kvm_mmu_notifier_change_pte,
512 .release = kvm_mmu_notifier_release,
515 static int kvm_init_mmu_notifier(struct kvm *kvm)
517 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
518 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
521 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
523 static int kvm_init_mmu_notifier(struct kvm *kvm)
528 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
530 static struct kvm_memslots *kvm_alloc_memslots(void)
533 struct kvm_memslots *slots;
535 slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL_ACCOUNT);
539 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
540 slots->id_to_index[i] = slots->memslots[i].id = i;
545 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
547 if (!memslot->dirty_bitmap)
550 kvfree(memslot->dirty_bitmap);
551 memslot->dirty_bitmap = NULL;
555 * Free any memory in @free but not in @dont.
557 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
558 struct kvm_memory_slot *dont)
560 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
561 kvm_destroy_dirty_bitmap(free);
563 kvm_arch_free_memslot(kvm, free, dont);
568 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
570 struct kvm_memory_slot *memslot;
575 kvm_for_each_memslot(memslot, slots)
576 kvm_free_memslot(kvm, memslot, NULL);
581 static void kvm_destroy_vm_debugfs(struct kvm *kvm)
585 if (!kvm->debugfs_dentry)
588 debugfs_remove_recursive(kvm->debugfs_dentry);
590 if (kvm->debugfs_stat_data) {
591 for (i = 0; i < kvm_debugfs_num_entries; i++)
592 kfree(kvm->debugfs_stat_data[i]);
593 kfree(kvm->debugfs_stat_data);
597 static int kvm_create_vm_debugfs(struct kvm *kvm, int fd)
599 char dir_name[ITOA_MAX_LEN * 2];
600 struct kvm_stat_data *stat_data;
601 struct kvm_stats_debugfs_item *p;
603 if (!debugfs_initialized())
606 snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd);
607 kvm->debugfs_dentry = debugfs_create_dir(dir_name, kvm_debugfs_dir);
609 kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries,
610 sizeof(*kvm->debugfs_stat_data),
612 if (!kvm->debugfs_stat_data)
615 for (p = debugfs_entries; p->name; p++) {
616 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT);
620 stat_data->kvm = kvm;
621 stat_data->offset = p->offset;
622 kvm->debugfs_stat_data[p - debugfs_entries] = stat_data;
623 debugfs_create_file(p->name, 0644, kvm->debugfs_dentry,
624 stat_data, stat_fops_per_vm[p->kind]);
629 static struct kvm *kvm_create_vm(unsigned long type)
632 struct kvm *kvm = kvm_arch_alloc_vm();
635 return ERR_PTR(-ENOMEM);
637 spin_lock_init(&kvm->mmu_lock);
639 kvm->mm = current->mm;
640 kvm_eventfd_init(kvm);
641 mutex_init(&kvm->lock);
642 mutex_init(&kvm->irq_lock);
643 mutex_init(&kvm->slots_lock);
644 refcount_set(&kvm->users_count, 1);
645 INIT_LIST_HEAD(&kvm->devices);
647 r = kvm_arch_init_vm(kvm, type);
649 goto out_err_no_disable;
651 r = hardware_enable_all();
653 goto out_err_no_disable;
655 #ifdef CONFIG_HAVE_KVM_IRQFD
656 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
659 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
662 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
663 struct kvm_memslots *slots = kvm_alloc_memslots();
665 goto out_err_no_srcu;
666 /* Generations must be different for each address space. */
667 slots->generation = i;
668 rcu_assign_pointer(kvm->memslots[i], slots);
671 if (init_srcu_struct(&kvm->srcu))
672 goto out_err_no_srcu;
673 if (init_srcu_struct(&kvm->irq_srcu))
674 goto out_err_no_irq_srcu;
675 for (i = 0; i < KVM_NR_BUSES; i++) {
676 rcu_assign_pointer(kvm->buses[i],
677 kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL_ACCOUNT));
682 r = kvm_init_mmu_notifier(kvm);
686 spin_lock(&kvm_lock);
687 list_add(&kvm->vm_list, &vm_list);
688 spin_unlock(&kvm_lock);
690 preempt_notifier_inc();
695 cleanup_srcu_struct(&kvm->irq_srcu);
697 cleanup_srcu_struct(&kvm->srcu);
699 hardware_disable_all();
701 refcount_set(&kvm->users_count, 0);
702 for (i = 0; i < KVM_NR_BUSES; i++)
703 kfree(kvm_get_bus(kvm, i));
704 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
705 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
706 kvm_arch_free_vm(kvm);
711 static void kvm_destroy_devices(struct kvm *kvm)
713 struct kvm_device *dev, *tmp;
716 * We do not need to take the kvm->lock here, because nobody else
717 * has a reference to the struct kvm at this point and therefore
718 * cannot access the devices list anyhow.
720 list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
721 list_del(&dev->vm_node);
722 dev->ops->destroy(dev);
726 static void kvm_destroy_vm(struct kvm *kvm)
729 struct mm_struct *mm = kvm->mm;
731 kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm);
732 kvm_destroy_vm_debugfs(kvm);
733 kvm_arch_sync_events(kvm);
734 spin_lock(&kvm_lock);
735 list_del(&kvm->vm_list);
736 spin_unlock(&kvm_lock);
737 kvm_free_irq_routing(kvm);
738 for (i = 0; i < KVM_NR_BUSES; i++) {
739 struct kvm_io_bus *bus = kvm_get_bus(kvm, i);
742 kvm_io_bus_destroy(bus);
743 kvm->buses[i] = NULL;
745 kvm_coalesced_mmio_free(kvm);
746 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
747 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
749 kvm_arch_flush_shadow_all(kvm);
751 kvm_arch_destroy_vm(kvm);
752 kvm_destroy_devices(kvm);
753 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
754 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
755 cleanup_srcu_struct(&kvm->irq_srcu);
756 cleanup_srcu_struct(&kvm->srcu);
757 kvm_arch_free_vm(kvm);
758 preempt_notifier_dec();
759 hardware_disable_all();
763 void kvm_get_kvm(struct kvm *kvm)
765 refcount_inc(&kvm->users_count);
767 EXPORT_SYMBOL_GPL(kvm_get_kvm);
769 void kvm_put_kvm(struct kvm *kvm)
771 if (refcount_dec_and_test(&kvm->users_count))
774 EXPORT_SYMBOL_GPL(kvm_put_kvm);
777 static int kvm_vm_release(struct inode *inode, struct file *filp)
779 struct kvm *kvm = filp->private_data;
781 kvm_irqfd_release(kvm);
788 * Allocation size is twice as large as the actual dirty bitmap size.
789 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
791 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
793 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
795 memslot->dirty_bitmap = kvzalloc(dirty_bytes, GFP_KERNEL_ACCOUNT);
796 if (!memslot->dirty_bitmap)
803 * Insert memslot and re-sort memslots based on their GFN,
804 * so binary search could be used to lookup GFN.
805 * Sorting algorithm takes advantage of having initially
806 * sorted array and known changed memslot position.
808 static void update_memslots(struct kvm_memslots *slots,
809 struct kvm_memory_slot *new,
810 enum kvm_mr_change change)
813 int i = slots->id_to_index[id];
814 struct kvm_memory_slot *mslots = slots->memslots;
816 WARN_ON(mslots[i].id != id);
820 WARN_ON(mslots[i].npages || !new->npages);
824 WARN_ON(new->npages || !mslots[i].npages);
830 while (i < KVM_MEM_SLOTS_NUM - 1 &&
831 new->base_gfn <= mslots[i + 1].base_gfn) {
832 if (!mslots[i + 1].npages)
834 mslots[i] = mslots[i + 1];
835 slots->id_to_index[mslots[i].id] = i;
840 * The ">=" is needed when creating a slot with base_gfn == 0,
841 * so that it moves before all those with base_gfn == npages == 0.
843 * On the other hand, if new->npages is zero, the above loop has
844 * already left i pointing to the beginning of the empty part of
845 * mslots, and the ">=" would move the hole backwards in this
846 * case---which is wrong. So skip the loop when deleting a slot.
850 new->base_gfn >= mslots[i - 1].base_gfn) {
851 mslots[i] = mslots[i - 1];
852 slots->id_to_index[mslots[i].id] = i;
856 WARN_ON_ONCE(i != slots->used_slots);
859 slots->id_to_index[mslots[i].id] = i;
862 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
864 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
866 #ifdef __KVM_HAVE_READONLY_MEM
867 valid_flags |= KVM_MEM_READONLY;
870 if (mem->flags & ~valid_flags)
876 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
877 int as_id, struct kvm_memslots *slots)
879 struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id);
880 u64 gen = old_memslots->generation;
882 WARN_ON(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS);
883 slots->generation = gen | KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
885 rcu_assign_pointer(kvm->memslots[as_id], slots);
886 synchronize_srcu_expedited(&kvm->srcu);
889 * Increment the new memslot generation a second time, dropping the
890 * update in-progress flag and incrementing then generation based on
891 * the number of address spaces. This provides a unique and easily
892 * identifiable generation number while the memslots are in flux.
894 gen = slots->generation & ~KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
897 * Generations must be unique even across address spaces. We do not need
898 * a global counter for that, instead the generation space is evenly split
899 * across address spaces. For example, with two address spaces, address
900 * space 0 will use generations 0, 2, 4, ... while address space 1 will
901 * use generations 1, 3, 5, ...
903 gen += KVM_ADDRESS_SPACE_NUM;
905 kvm_arch_memslots_updated(kvm, gen);
907 slots->generation = gen;
913 * Allocate some memory and give it an address in the guest physical address
916 * Discontiguous memory is allowed, mostly for framebuffers.
918 * Must be called holding kvm->slots_lock for write.
920 int __kvm_set_memory_region(struct kvm *kvm,
921 const struct kvm_userspace_memory_region *mem)
925 unsigned long npages;
926 struct kvm_memory_slot *slot;
927 struct kvm_memory_slot old, new;
928 struct kvm_memslots *slots = NULL, *old_memslots;
930 enum kvm_mr_change change;
932 r = check_memory_region_flags(mem);
937 as_id = mem->slot >> 16;
940 /* General sanity checks */
941 if (mem->memory_size & (PAGE_SIZE - 1))
943 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
945 /* We can read the guest memory with __xxx_user() later on. */
946 if ((id < KVM_USER_MEM_SLOTS) &&
947 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
948 !access_ok((void __user *)(unsigned long)mem->userspace_addr,
951 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
953 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
956 slot = id_to_memslot(__kvm_memslots(kvm, as_id), id);
957 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
958 npages = mem->memory_size >> PAGE_SHIFT;
960 if (npages > KVM_MEM_MAX_NR_PAGES)
966 new.base_gfn = base_gfn;
968 new.flags = mem->flags;
972 change = KVM_MR_CREATE;
973 else { /* Modify an existing slot. */
974 if ((mem->userspace_addr != old.userspace_addr) ||
975 (npages != old.npages) ||
976 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
979 if (base_gfn != old.base_gfn)
980 change = KVM_MR_MOVE;
981 else if (new.flags != old.flags)
982 change = KVM_MR_FLAGS_ONLY;
983 else { /* Nothing to change. */
992 change = KVM_MR_DELETE;
997 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
998 /* Check for overlaps */
1000 kvm_for_each_memslot(slot, __kvm_memslots(kvm, as_id)) {
1003 if (!((base_gfn + npages <= slot->base_gfn) ||
1004 (base_gfn >= slot->base_gfn + slot->npages)))
1009 /* Free page dirty bitmap if unneeded */
1010 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
1011 new.dirty_bitmap = NULL;
1014 if (change == KVM_MR_CREATE) {
1015 new.userspace_addr = mem->userspace_addr;
1017 if (kvm_arch_create_memslot(kvm, &new, npages))
1021 /* Allocate page dirty bitmap if needed */
1022 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
1023 if (kvm_create_dirty_bitmap(&new) < 0)
1027 slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL_ACCOUNT);
1030 memcpy(slots, __kvm_memslots(kvm, as_id), sizeof(struct kvm_memslots));
1032 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
1033 slot = id_to_memslot(slots, id);
1034 slot->flags |= KVM_MEMSLOT_INVALID;
1036 old_memslots = install_new_memslots(kvm, as_id, slots);
1038 /* From this point no new shadow pages pointing to a deleted,
1039 * or moved, memslot will be created.
1041 * validation of sp->gfn happens in:
1042 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1043 * - kvm_is_visible_gfn (mmu_check_roots)
1045 kvm_arch_flush_shadow_memslot(kvm, slot);
1048 * We can re-use the old_memslots from above, the only difference
1049 * from the currently installed memslots is the invalid flag. This
1050 * will get overwritten by update_memslots anyway.
1052 slots = old_memslots;
1055 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
1059 /* actual memory is freed via old in kvm_free_memslot below */
1060 if (change == KVM_MR_DELETE) {
1061 new.dirty_bitmap = NULL;
1062 memset(&new.arch, 0, sizeof(new.arch));
1065 update_memslots(slots, &new, change);
1066 old_memslots = install_new_memslots(kvm, as_id, slots);
1068 kvm_arch_commit_memory_region(kvm, mem, &old, &new, change);
1070 kvm_free_memslot(kvm, &old, &new);
1071 kvfree(old_memslots);
1077 kvm_free_memslot(kvm, &new, &old);
1081 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1083 int kvm_set_memory_region(struct kvm *kvm,
1084 const struct kvm_userspace_memory_region *mem)
1088 mutex_lock(&kvm->slots_lock);
1089 r = __kvm_set_memory_region(kvm, mem);
1090 mutex_unlock(&kvm->slots_lock);
1093 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1095 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1096 struct kvm_userspace_memory_region *mem)
1098 if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1101 return kvm_set_memory_region(kvm, mem);
1104 int kvm_get_dirty_log(struct kvm *kvm,
1105 struct kvm_dirty_log *log, int *is_dirty)
1107 struct kvm_memslots *slots;
1108 struct kvm_memory_slot *memslot;
1111 unsigned long any = 0;
1113 as_id = log->slot >> 16;
1114 id = (u16)log->slot;
1115 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1118 slots = __kvm_memslots(kvm, as_id);
1119 memslot = id_to_memslot(slots, id);
1120 if (!memslot->dirty_bitmap)
1123 n = kvm_dirty_bitmap_bytes(memslot);
1125 for (i = 0; !any && i < n/sizeof(long); ++i)
1126 any = memslot->dirty_bitmap[i];
1128 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1135 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1137 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1139 * kvm_get_dirty_log_protect - get a snapshot of dirty pages
1140 * and reenable dirty page tracking for the corresponding pages.
1141 * @kvm: pointer to kvm instance
1142 * @log: slot id and address to which we copy the log
1143 * @flush: true if TLB flush is needed by caller
1145 * We need to keep it in mind that VCPU threads can write to the bitmap
1146 * concurrently. So, to avoid losing track of dirty pages we keep the
1149 * 1. Take a snapshot of the bit and clear it if needed.
1150 * 2. Write protect the corresponding page.
1151 * 3. Copy the snapshot to the userspace.
1152 * 4. Upon return caller flushes TLB's if needed.
1154 * Between 2 and 4, the guest may write to the page using the remaining TLB
1155 * entry. This is not a problem because the page is reported dirty using
1156 * the snapshot taken before and step 4 ensures that writes done after
1157 * exiting to userspace will be logged for the next call.
1160 int kvm_get_dirty_log_protect(struct kvm *kvm,
1161 struct kvm_dirty_log *log, bool *flush)
1163 struct kvm_memslots *slots;
1164 struct kvm_memory_slot *memslot;
1167 unsigned long *dirty_bitmap;
1168 unsigned long *dirty_bitmap_buffer;
1170 as_id = log->slot >> 16;
1171 id = (u16)log->slot;
1172 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1175 slots = __kvm_memslots(kvm, as_id);
1176 memslot = id_to_memslot(slots, id);
1178 dirty_bitmap = memslot->dirty_bitmap;
1182 n = kvm_dirty_bitmap_bytes(memslot);
1184 if (kvm->manual_dirty_log_protect) {
1186 * Unlike kvm_get_dirty_log, we always return false in *flush,
1187 * because no flush is needed until KVM_CLEAR_DIRTY_LOG. There
1188 * is some code duplication between this function and
1189 * kvm_get_dirty_log, but hopefully all architecture
1190 * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log
1191 * can be eliminated.
1193 dirty_bitmap_buffer = dirty_bitmap;
1195 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
1196 memset(dirty_bitmap_buffer, 0, n);
1198 spin_lock(&kvm->mmu_lock);
1199 for (i = 0; i < n / sizeof(long); i++) {
1203 if (!dirty_bitmap[i])
1207 mask = xchg(&dirty_bitmap[i], 0);
1208 dirty_bitmap_buffer[i] = mask;
1210 offset = i * BITS_PER_LONG;
1211 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1214 spin_unlock(&kvm->mmu_lock);
1217 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1221 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect);
1224 * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap
1225 * and reenable dirty page tracking for the corresponding pages.
1226 * @kvm: pointer to kvm instance
1227 * @log: slot id and address from which to fetch the bitmap of dirty pages
1228 * @flush: true if TLB flush is needed by caller
1230 int kvm_clear_dirty_log_protect(struct kvm *kvm,
1231 struct kvm_clear_dirty_log *log, bool *flush)
1233 struct kvm_memslots *slots;
1234 struct kvm_memory_slot *memslot;
1238 unsigned long *dirty_bitmap;
1239 unsigned long *dirty_bitmap_buffer;
1241 as_id = log->slot >> 16;
1242 id = (u16)log->slot;
1243 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1246 if (log->first_page & 63)
1249 slots = __kvm_memslots(kvm, as_id);
1250 memslot = id_to_memslot(slots, id);
1252 dirty_bitmap = memslot->dirty_bitmap;
1256 n = ALIGN(log->num_pages, BITS_PER_LONG) / 8;
1258 if (log->first_page > memslot->npages ||
1259 log->num_pages > memslot->npages - log->first_page ||
1260 (log->num_pages < memslot->npages - log->first_page && (log->num_pages & 63)))
1264 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
1265 if (copy_from_user(dirty_bitmap_buffer, log->dirty_bitmap, n))
1268 spin_lock(&kvm->mmu_lock);
1269 for (offset = log->first_page, i = offset / BITS_PER_LONG,
1270 n = DIV_ROUND_UP(log->num_pages, BITS_PER_LONG); n--;
1271 i++, offset += BITS_PER_LONG) {
1272 unsigned long mask = *dirty_bitmap_buffer++;
1273 atomic_long_t *p = (atomic_long_t *) &dirty_bitmap[i];
1277 mask &= atomic_long_fetch_andnot(mask, p);
1280 * mask contains the bits that really have been cleared. This
1281 * never includes any bits beyond the length of the memslot (if
1282 * the length is not aligned to 64 pages), therefore it is not
1283 * a problem if userspace sets them in log->dirty_bitmap.
1287 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1291 spin_unlock(&kvm->mmu_lock);
1295 EXPORT_SYMBOL_GPL(kvm_clear_dirty_log_protect);
1298 bool kvm_largepages_enabled(void)
1300 return largepages_enabled;
1303 void kvm_disable_largepages(void)
1305 largepages_enabled = false;
1307 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1309 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1311 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1313 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1315 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
1317 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn);
1320 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1322 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1324 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1325 memslot->flags & KVM_MEMSLOT_INVALID)
1330 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1332 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1334 struct vm_area_struct *vma;
1335 unsigned long addr, size;
1339 addr = gfn_to_hva(kvm, gfn);
1340 if (kvm_is_error_hva(addr))
1343 down_read(¤t->mm->mmap_sem);
1344 vma = find_vma(current->mm, addr);
1348 size = vma_kernel_pagesize(vma);
1351 up_read(¤t->mm->mmap_sem);
1356 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1358 return slot->flags & KVM_MEM_READONLY;
1361 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1362 gfn_t *nr_pages, bool write)
1364 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1365 return KVM_HVA_ERR_BAD;
1367 if (memslot_is_readonly(slot) && write)
1368 return KVM_HVA_ERR_RO_BAD;
1371 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1373 return __gfn_to_hva_memslot(slot, gfn);
1376 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1379 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1382 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1385 return gfn_to_hva_many(slot, gfn, NULL);
1387 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1389 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1391 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1393 EXPORT_SYMBOL_GPL(gfn_to_hva);
1395 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
1397 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
1399 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
1402 * Return the hva of a @gfn and the R/W attribute if possible.
1404 * @slot: the kvm_memory_slot which contains @gfn
1405 * @gfn: the gfn to be translated
1406 * @writable: used to return the read/write attribute of the @slot if the hva
1407 * is valid and @writable is not NULL
1409 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1410 gfn_t gfn, bool *writable)
1412 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1414 if (!kvm_is_error_hva(hva) && writable)
1415 *writable = !memslot_is_readonly(slot);
1420 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1422 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1424 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1427 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
1429 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1431 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1434 static inline int check_user_page_hwpoison(unsigned long addr)
1436 int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
1438 rc = get_user_pages(addr, 1, flags, NULL, NULL);
1439 return rc == -EHWPOISON;
1443 * The fast path to get the writable pfn which will be stored in @pfn,
1444 * true indicates success, otherwise false is returned. It's also the
1445 * only part that runs if we can are in atomic context.
1447 static bool hva_to_pfn_fast(unsigned long addr, bool write_fault,
1448 bool *writable, kvm_pfn_t *pfn)
1450 struct page *page[1];
1454 * Fast pin a writable pfn only if it is a write fault request
1455 * or the caller allows to map a writable pfn for a read fault
1458 if (!(write_fault || writable))
1461 npages = __get_user_pages_fast(addr, 1, 1, page);
1463 *pfn = page_to_pfn(page[0]);
1474 * The slow path to get the pfn of the specified host virtual address,
1475 * 1 indicates success, -errno is returned if error is detected.
1477 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1478 bool *writable, kvm_pfn_t *pfn)
1480 unsigned int flags = FOLL_HWPOISON;
1487 *writable = write_fault;
1490 flags |= FOLL_WRITE;
1492 flags |= FOLL_NOWAIT;
1494 npages = get_user_pages_unlocked(addr, 1, &page, flags);
1498 /* map read fault as writable if possible */
1499 if (unlikely(!write_fault) && writable) {
1502 if (__get_user_pages_fast(addr, 1, 1, &wpage) == 1) {
1508 *pfn = page_to_pfn(page);
1512 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1514 if (unlikely(!(vma->vm_flags & VM_READ)))
1517 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1523 static int hva_to_pfn_remapped(struct vm_area_struct *vma,
1524 unsigned long addr, bool *async,
1525 bool write_fault, bool *writable,
1531 r = follow_pfn(vma, addr, &pfn);
1534 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
1535 * not call the fault handler, so do it here.
1537 bool unlocked = false;
1538 r = fixup_user_fault(current, current->mm, addr,
1539 (write_fault ? FAULT_FLAG_WRITE : 0),
1546 r = follow_pfn(vma, addr, &pfn);
1556 * Get a reference here because callers of *hva_to_pfn* and
1557 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
1558 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
1559 * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will
1560 * simply do nothing for reserved pfns.
1562 * Whoever called remap_pfn_range is also going to call e.g.
1563 * unmap_mapping_range before the underlying pages are freed,
1564 * causing a call to our MMU notifier.
1573 * Pin guest page in memory and return its pfn.
1574 * @addr: host virtual address which maps memory to the guest
1575 * @atomic: whether this function can sleep
1576 * @async: whether this function need to wait IO complete if the
1577 * host page is not in the memory
1578 * @write_fault: whether we should get a writable host page
1579 * @writable: whether it allows to map a writable host page for !@write_fault
1581 * The function will map a writable host page for these two cases:
1582 * 1): @write_fault = true
1583 * 2): @write_fault = false && @writable, @writable will tell the caller
1584 * whether the mapping is writable.
1586 static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1587 bool write_fault, bool *writable)
1589 struct vm_area_struct *vma;
1593 /* we can do it either atomically or asynchronously, not both */
1594 BUG_ON(atomic && async);
1596 if (hva_to_pfn_fast(addr, write_fault, writable, &pfn))
1600 return KVM_PFN_ERR_FAULT;
1602 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1606 down_read(¤t->mm->mmap_sem);
1607 if (npages == -EHWPOISON ||
1608 (!async && check_user_page_hwpoison(addr))) {
1609 pfn = KVM_PFN_ERR_HWPOISON;
1614 vma = find_vma_intersection(current->mm, addr, addr + 1);
1617 pfn = KVM_PFN_ERR_FAULT;
1618 else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
1619 r = hva_to_pfn_remapped(vma, addr, async, write_fault, writable, &pfn);
1623 pfn = KVM_PFN_ERR_FAULT;
1625 if (async && vma_is_valid(vma, write_fault))
1627 pfn = KVM_PFN_ERR_FAULT;
1630 up_read(¤t->mm->mmap_sem);
1634 kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn,
1635 bool atomic, bool *async, bool write_fault,
1638 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1640 if (addr == KVM_HVA_ERR_RO_BAD) {
1643 return KVM_PFN_ERR_RO_FAULT;
1646 if (kvm_is_error_hva(addr)) {
1649 return KVM_PFN_NOSLOT;
1652 /* Do not map writable pfn in the readonly memslot. */
1653 if (writable && memslot_is_readonly(slot)) {
1658 return hva_to_pfn(addr, atomic, async, write_fault,
1661 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
1663 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1666 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
1667 write_fault, writable);
1669 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1671 kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1673 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1675 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
1677 kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1679 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1681 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1683 kvm_pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1685 return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm, gfn), gfn);
1687 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1689 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
1691 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1693 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
1695 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1697 return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
1699 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1701 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
1703 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1705 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
1707 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1708 struct page **pages, int nr_pages)
1713 addr = gfn_to_hva_many(slot, gfn, &entry);
1714 if (kvm_is_error_hva(addr))
1717 if (entry < nr_pages)
1720 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1722 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1724 static struct page *kvm_pfn_to_page(kvm_pfn_t pfn)
1726 if (is_error_noslot_pfn(pfn))
1727 return KVM_ERR_PTR_BAD_PAGE;
1729 if (kvm_is_reserved_pfn(pfn)) {
1731 return KVM_ERR_PTR_BAD_PAGE;
1734 return pfn_to_page(pfn);
1737 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1741 pfn = gfn_to_pfn(kvm, gfn);
1743 return kvm_pfn_to_page(pfn);
1745 EXPORT_SYMBOL_GPL(gfn_to_page);
1747 static int __kvm_map_gfn(struct kvm_memory_slot *slot, gfn_t gfn,
1748 struct kvm_host_map *map)
1752 struct page *page = KVM_UNMAPPED_PAGE;
1757 pfn = gfn_to_pfn_memslot(slot, gfn);
1758 if (is_error_noslot_pfn(pfn))
1761 if (pfn_valid(pfn)) {
1762 page = pfn_to_page(pfn);
1764 #ifdef CONFIG_HAS_IOMEM
1766 hva = memremap(pfn_to_hpa(pfn), PAGE_SIZE, MEMREMAP_WB);
1781 int kvm_vcpu_map(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map)
1783 return __kvm_map_gfn(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, map);
1785 EXPORT_SYMBOL_GPL(kvm_vcpu_map);
1787 void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map,
1798 #ifdef CONFIG_HAS_IOMEM
1804 kvm_vcpu_mark_page_dirty(vcpu, map->gfn);
1805 kvm_release_pfn_dirty(map->pfn);
1807 kvm_release_pfn_clean(map->pfn);
1813 EXPORT_SYMBOL_GPL(kvm_vcpu_unmap);
1815 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
1819 pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
1821 return kvm_pfn_to_page(pfn);
1823 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
1825 void kvm_release_page_clean(struct page *page)
1827 WARN_ON(is_error_page(page));
1829 kvm_release_pfn_clean(page_to_pfn(page));
1831 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1833 void kvm_release_pfn_clean(kvm_pfn_t pfn)
1835 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
1836 put_page(pfn_to_page(pfn));
1838 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1840 void kvm_release_page_dirty(struct page *page)
1842 WARN_ON(is_error_page(page));
1844 kvm_release_pfn_dirty(page_to_pfn(page));
1846 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1848 void kvm_release_pfn_dirty(kvm_pfn_t pfn)
1850 kvm_set_pfn_dirty(pfn);
1851 kvm_release_pfn_clean(pfn);
1853 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1855 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
1857 if (!kvm_is_reserved_pfn(pfn)) {
1858 struct page *page = pfn_to_page(pfn);
1860 if (!PageReserved(page))
1864 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1866 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
1868 if (!kvm_is_reserved_pfn(pfn))
1869 mark_page_accessed(pfn_to_page(pfn));
1871 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1873 void kvm_get_pfn(kvm_pfn_t pfn)
1875 if (!kvm_is_reserved_pfn(pfn))
1876 get_page(pfn_to_page(pfn));
1878 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1880 static int next_segment(unsigned long len, int offset)
1882 if (len > PAGE_SIZE - offset)
1883 return PAGE_SIZE - offset;
1888 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
1889 void *data, int offset, int len)
1894 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1895 if (kvm_is_error_hva(addr))
1897 r = __copy_from_user(data, (void __user *)addr + offset, len);
1903 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1906 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1908 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1910 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1912 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
1913 int offset, int len)
1915 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1917 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1919 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
1921 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1923 gfn_t gfn = gpa >> PAGE_SHIFT;
1925 int offset = offset_in_page(gpa);
1928 while ((seg = next_segment(len, offset)) != 0) {
1929 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1939 EXPORT_SYMBOL_GPL(kvm_read_guest);
1941 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
1943 gfn_t gfn = gpa >> PAGE_SHIFT;
1945 int offset = offset_in_page(gpa);
1948 while ((seg = next_segment(len, offset)) != 0) {
1949 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
1959 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
1961 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1962 void *data, int offset, unsigned long len)
1967 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1968 if (kvm_is_error_hva(addr))
1970 pagefault_disable();
1971 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1978 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1981 gfn_t gfn = gpa >> PAGE_SHIFT;
1982 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1983 int offset = offset_in_page(gpa);
1985 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1987 EXPORT_SYMBOL_GPL(kvm_read_guest_atomic);
1989 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
1990 void *data, unsigned long len)
1992 gfn_t gfn = gpa >> PAGE_SHIFT;
1993 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1994 int offset = offset_in_page(gpa);
1996 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1998 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
2000 static int __kvm_write_guest_page(struct kvm_memory_slot *memslot, gfn_t gfn,
2001 const void *data, int offset, int len)
2006 addr = gfn_to_hva_memslot(memslot, gfn);
2007 if (kvm_is_error_hva(addr))
2009 r = __copy_to_user((void __user *)addr + offset, data, len);
2012 mark_page_dirty_in_slot(memslot, gfn);
2016 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
2017 const void *data, int offset, int len)
2019 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2021 return __kvm_write_guest_page(slot, gfn, data, offset, len);
2023 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
2025 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
2026 const void *data, int offset, int len)
2028 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2030 return __kvm_write_guest_page(slot, gfn, data, offset, len);
2032 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
2034 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
2037 gfn_t gfn = gpa >> PAGE_SHIFT;
2039 int offset = offset_in_page(gpa);
2042 while ((seg = next_segment(len, offset)) != 0) {
2043 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
2053 EXPORT_SYMBOL_GPL(kvm_write_guest);
2055 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
2058 gfn_t gfn = gpa >> PAGE_SHIFT;
2060 int offset = offset_in_page(gpa);
2063 while ((seg = next_segment(len, offset)) != 0) {
2064 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
2074 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
2076 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
2077 struct gfn_to_hva_cache *ghc,
2078 gpa_t gpa, unsigned long len)
2080 int offset = offset_in_page(gpa);
2081 gfn_t start_gfn = gpa >> PAGE_SHIFT;
2082 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
2083 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
2084 gfn_t nr_pages_avail;
2085 int r = start_gfn <= end_gfn ? 0 : -EINVAL;
2088 ghc->generation = slots->generation;
2090 ghc->hva = KVM_HVA_ERR_BAD;
2093 * If the requested region crosses two memslots, we still
2094 * verify that the entire region is valid here.
2096 while (!r && start_gfn <= end_gfn) {
2097 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
2098 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
2100 if (kvm_is_error_hva(ghc->hva))
2102 start_gfn += nr_pages_avail;
2105 /* Use the slow path for cross page reads and writes. */
2106 if (!r && nr_pages_needed == 1)
2109 ghc->memslot = NULL;
2114 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2115 gpa_t gpa, unsigned long len)
2117 struct kvm_memslots *slots = kvm_memslots(kvm);
2118 return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len);
2120 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
2122 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2123 void *data, unsigned int offset,
2126 struct kvm_memslots *slots = kvm_memslots(kvm);
2128 gpa_t gpa = ghc->gpa + offset;
2130 BUG_ON(len + offset > ghc->len);
2132 if (slots->generation != ghc->generation)
2133 __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len);
2135 if (unlikely(!ghc->memslot))
2136 return kvm_write_guest(kvm, gpa, data, len);
2138 if (kvm_is_error_hva(ghc->hva))
2141 r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
2144 mark_page_dirty_in_slot(ghc->memslot, gpa >> PAGE_SHIFT);
2148 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached);
2150 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2151 void *data, unsigned long len)
2153 return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len);
2155 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
2157 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2158 void *data, unsigned long len)
2160 struct kvm_memslots *slots = kvm_memslots(kvm);
2163 BUG_ON(len > ghc->len);
2165 if (slots->generation != ghc->generation)
2166 __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len);
2168 if (unlikely(!ghc->memslot))
2169 return kvm_read_guest(kvm, ghc->gpa, data, len);
2171 if (kvm_is_error_hva(ghc->hva))
2174 r = __copy_from_user(data, (void __user *)ghc->hva, len);
2180 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
2182 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
2184 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
2186 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
2188 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
2190 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
2192 gfn_t gfn = gpa >> PAGE_SHIFT;
2194 int offset = offset_in_page(gpa);
2197 while ((seg = next_segment(len, offset)) != 0) {
2198 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
2207 EXPORT_SYMBOL_GPL(kvm_clear_guest);
2209 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot,
2212 if (memslot && memslot->dirty_bitmap) {
2213 unsigned long rel_gfn = gfn - memslot->base_gfn;
2215 set_bit_le(rel_gfn, memslot->dirty_bitmap);
2219 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
2221 struct kvm_memory_slot *memslot;
2223 memslot = gfn_to_memslot(kvm, gfn);
2224 mark_page_dirty_in_slot(memslot, gfn);
2226 EXPORT_SYMBOL_GPL(mark_page_dirty);
2228 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
2230 struct kvm_memory_slot *memslot;
2232 memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2233 mark_page_dirty_in_slot(memslot, gfn);
2235 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
2237 void kvm_sigset_activate(struct kvm_vcpu *vcpu)
2239 if (!vcpu->sigset_active)
2243 * This does a lockless modification of ->real_blocked, which is fine
2244 * because, only current can change ->real_blocked and all readers of
2245 * ->real_blocked don't care as long ->real_blocked is always a subset
2248 sigprocmask(SIG_SETMASK, &vcpu->sigset, ¤t->real_blocked);
2251 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu)
2253 if (!vcpu->sigset_active)
2256 sigprocmask(SIG_SETMASK, ¤t->real_blocked, NULL);
2257 sigemptyset(¤t->real_blocked);
2260 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
2262 unsigned int old, val, grow, grow_start;
2264 old = val = vcpu->halt_poll_ns;
2265 grow_start = READ_ONCE(halt_poll_ns_grow_start);
2266 grow = READ_ONCE(halt_poll_ns_grow);
2271 if (val < grow_start)
2274 if (val > halt_poll_ns)
2277 vcpu->halt_poll_ns = val;
2279 trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
2282 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
2284 unsigned int old, val, shrink;
2286 old = val = vcpu->halt_poll_ns;
2287 shrink = READ_ONCE(halt_poll_ns_shrink);
2293 vcpu->halt_poll_ns = val;
2294 trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
2297 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
2300 int idx = srcu_read_lock(&vcpu->kvm->srcu);
2302 if (kvm_arch_vcpu_runnable(vcpu)) {
2303 kvm_make_request(KVM_REQ_UNHALT, vcpu);
2306 if (kvm_cpu_has_pending_timer(vcpu))
2308 if (signal_pending(current))
2313 srcu_read_unlock(&vcpu->kvm->srcu, idx);
2318 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2320 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
2323 DECLARE_SWAITQUEUE(wait);
2324 bool waited = false;
2327 start = cur = ktime_get();
2328 if (vcpu->halt_poll_ns && !kvm_arch_no_poll(vcpu)) {
2329 ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns);
2331 ++vcpu->stat.halt_attempted_poll;
2334 * This sets KVM_REQ_UNHALT if an interrupt
2337 if (kvm_vcpu_check_block(vcpu) < 0) {
2338 ++vcpu->stat.halt_successful_poll;
2339 if (!vcpu_valid_wakeup(vcpu))
2340 ++vcpu->stat.halt_poll_invalid;
2344 } while (single_task_running() && ktime_before(cur, stop));
2347 kvm_arch_vcpu_blocking(vcpu);
2350 prepare_to_swait_exclusive(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
2352 if (kvm_vcpu_check_block(vcpu) < 0)
2359 finish_swait(&vcpu->wq, &wait);
2362 kvm_arch_vcpu_unblocking(vcpu);
2364 block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
2366 if (!vcpu_valid_wakeup(vcpu))
2367 shrink_halt_poll_ns(vcpu);
2368 else if (halt_poll_ns) {
2369 if (block_ns <= vcpu->halt_poll_ns)
2371 /* we had a long block, shrink polling */
2372 else if (vcpu->halt_poll_ns && block_ns > halt_poll_ns)
2373 shrink_halt_poll_ns(vcpu);
2374 /* we had a short halt and our poll time is too small */
2375 else if (vcpu->halt_poll_ns < halt_poll_ns &&
2376 block_ns < halt_poll_ns)
2377 grow_halt_poll_ns(vcpu);
2379 vcpu->halt_poll_ns = 0;
2381 trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu));
2382 kvm_arch_vcpu_block_finish(vcpu);
2384 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
2386 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
2388 struct swait_queue_head *wqp;
2390 wqp = kvm_arch_vcpu_wq(vcpu);
2391 if (swq_has_sleeper(wqp)) {
2393 ++vcpu->stat.halt_wakeup;
2399 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
2403 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2405 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
2408 int cpu = vcpu->cpu;
2410 if (kvm_vcpu_wake_up(vcpu))
2414 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
2415 if (kvm_arch_vcpu_should_kick(vcpu))
2416 smp_send_reschedule(cpu);
2419 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
2420 #endif /* !CONFIG_S390 */
2422 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
2425 struct task_struct *task = NULL;
2429 pid = rcu_dereference(target->pid);
2431 task = get_pid_task(pid, PIDTYPE_PID);
2435 ret = yield_to(task, 1);
2436 put_task_struct(task);
2440 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
2443 * Helper that checks whether a VCPU is eligible for directed yield.
2444 * Most eligible candidate to yield is decided by following heuristics:
2446 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2447 * (preempted lock holder), indicated by @in_spin_loop.
2448 * Set at the beiginning and cleared at the end of interception/PLE handler.
2450 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2451 * chance last time (mostly it has become eligible now since we have probably
2452 * yielded to lockholder in last iteration. This is done by toggling
2453 * @dy_eligible each time a VCPU checked for eligibility.)
2455 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2456 * to preempted lock-holder could result in wrong VCPU selection and CPU
2457 * burning. Giving priority for a potential lock-holder increases lock
2460 * Since algorithm is based on heuristics, accessing another VCPU data without
2461 * locking does not harm. It may result in trying to yield to same VCPU, fail
2462 * and continue with next VCPU and so on.
2464 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
2466 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2469 eligible = !vcpu->spin_loop.in_spin_loop ||
2470 vcpu->spin_loop.dy_eligible;
2472 if (vcpu->spin_loop.in_spin_loop)
2473 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
2481 void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode)
2483 struct kvm *kvm = me->kvm;
2484 struct kvm_vcpu *vcpu;
2485 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
2491 kvm_vcpu_set_in_spin_loop(me, true);
2493 * We boost the priority of a VCPU that is runnable but not
2494 * currently running, because it got preempted by something
2495 * else and called schedule in __vcpu_run. Hopefully that
2496 * VCPU is holding the lock that we need and will release it.
2497 * We approximate round-robin by starting at the last boosted VCPU.
2499 for (pass = 0; pass < 2 && !yielded && try; pass++) {
2500 kvm_for_each_vcpu(i, vcpu, kvm) {
2501 if (!pass && i <= last_boosted_vcpu) {
2502 i = last_boosted_vcpu;
2504 } else if (pass && i > last_boosted_vcpu)
2506 if (!READ_ONCE(vcpu->preempted))
2510 if (swait_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
2512 if (yield_to_kernel_mode && !kvm_arch_vcpu_in_kernel(vcpu))
2514 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
2517 yielded = kvm_vcpu_yield_to(vcpu);
2519 kvm->last_boosted_vcpu = i;
2521 } else if (yielded < 0) {
2528 kvm_vcpu_set_in_spin_loop(me, false);
2530 /* Ensure vcpu is not eligible during next spinloop */
2531 kvm_vcpu_set_dy_eligible(me, false);
2533 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
2535 static vm_fault_t kvm_vcpu_fault(struct vm_fault *vmf)
2537 struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data;
2540 if (vmf->pgoff == 0)
2541 page = virt_to_page(vcpu->run);
2543 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
2544 page = virt_to_page(vcpu->arch.pio_data);
2546 #ifdef CONFIG_KVM_MMIO
2547 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
2548 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
2551 return kvm_arch_vcpu_fault(vcpu, vmf);
2557 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
2558 .fault = kvm_vcpu_fault,
2561 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2563 vma->vm_ops = &kvm_vcpu_vm_ops;
2567 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2569 struct kvm_vcpu *vcpu = filp->private_data;
2571 debugfs_remove_recursive(vcpu->debugfs_dentry);
2572 kvm_put_kvm(vcpu->kvm);
2576 static struct file_operations kvm_vcpu_fops = {
2577 .release = kvm_vcpu_release,
2578 .unlocked_ioctl = kvm_vcpu_ioctl,
2579 .mmap = kvm_vcpu_mmap,
2580 .llseek = noop_llseek,
2581 KVM_COMPAT(kvm_vcpu_compat_ioctl),
2585 * Allocates an inode for the vcpu.
2587 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2589 char name[8 + 1 + ITOA_MAX_LEN + 1];
2591 snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id);
2592 return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
2595 static int kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
2597 char dir_name[ITOA_MAX_LEN * 2];
2600 if (!kvm_arch_has_vcpu_debugfs())
2603 if (!debugfs_initialized())
2606 snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
2607 vcpu->debugfs_dentry = debugfs_create_dir(dir_name,
2608 vcpu->kvm->debugfs_dentry);
2609 if (!vcpu->debugfs_dentry)
2612 ret = kvm_arch_create_vcpu_debugfs(vcpu);
2614 debugfs_remove_recursive(vcpu->debugfs_dentry);
2622 * Creates some virtual cpus. Good luck creating more than one.
2624 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
2627 struct kvm_vcpu *vcpu;
2629 if (id >= KVM_MAX_VCPU_ID)
2632 mutex_lock(&kvm->lock);
2633 if (kvm->created_vcpus == KVM_MAX_VCPUS) {
2634 mutex_unlock(&kvm->lock);
2638 kvm->created_vcpus++;
2639 mutex_unlock(&kvm->lock);
2641 vcpu = kvm_arch_vcpu_create(kvm, id);
2644 goto vcpu_decrement;
2647 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2649 r = kvm_arch_vcpu_setup(vcpu);
2653 r = kvm_create_vcpu_debugfs(vcpu);
2657 mutex_lock(&kvm->lock);
2658 if (kvm_get_vcpu_by_id(kvm, id)) {
2660 goto unlock_vcpu_destroy;
2663 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
2665 /* Now it's all set up, let userspace reach it */
2667 r = create_vcpu_fd(vcpu);
2670 goto unlock_vcpu_destroy;
2673 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
2676 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
2677 * before kvm->online_vcpu's incremented value.
2680 atomic_inc(&kvm->online_vcpus);
2682 mutex_unlock(&kvm->lock);
2683 kvm_arch_vcpu_postcreate(vcpu);
2686 unlock_vcpu_destroy:
2687 mutex_unlock(&kvm->lock);
2688 debugfs_remove_recursive(vcpu->debugfs_dentry);
2690 kvm_arch_vcpu_destroy(vcpu);
2692 mutex_lock(&kvm->lock);
2693 kvm->created_vcpus--;
2694 mutex_unlock(&kvm->lock);
2698 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2701 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2702 vcpu->sigset_active = 1;
2703 vcpu->sigset = *sigset;
2705 vcpu->sigset_active = 0;
2709 static long kvm_vcpu_ioctl(struct file *filp,
2710 unsigned int ioctl, unsigned long arg)
2712 struct kvm_vcpu *vcpu = filp->private_data;
2713 void __user *argp = (void __user *)arg;
2715 struct kvm_fpu *fpu = NULL;
2716 struct kvm_sregs *kvm_sregs = NULL;
2718 if (vcpu->kvm->mm != current->mm)
2721 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2725 * Some architectures have vcpu ioctls that are asynchronous to vcpu
2726 * execution; mutex_lock() would break them.
2728 r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg);
2729 if (r != -ENOIOCTLCMD)
2732 if (mutex_lock_killable(&vcpu->mutex))
2740 oldpid = rcu_access_pointer(vcpu->pid);
2741 if (unlikely(oldpid != task_pid(current))) {
2742 /* The thread running this VCPU changed. */
2745 r = kvm_arch_vcpu_run_pid_change(vcpu);
2749 newpid = get_task_pid(current, PIDTYPE_PID);
2750 rcu_assign_pointer(vcpu->pid, newpid);
2755 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2756 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2759 case KVM_GET_REGS: {
2760 struct kvm_regs *kvm_regs;
2763 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL_ACCOUNT);
2766 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2770 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2777 case KVM_SET_REGS: {
2778 struct kvm_regs *kvm_regs;
2781 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2782 if (IS_ERR(kvm_regs)) {
2783 r = PTR_ERR(kvm_regs);
2786 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2790 case KVM_GET_SREGS: {
2791 kvm_sregs = kzalloc(sizeof(struct kvm_sregs),
2792 GFP_KERNEL_ACCOUNT);
2796 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2800 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2805 case KVM_SET_SREGS: {
2806 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2807 if (IS_ERR(kvm_sregs)) {
2808 r = PTR_ERR(kvm_sregs);
2812 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2815 case KVM_GET_MP_STATE: {
2816 struct kvm_mp_state mp_state;
2818 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2822 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
2827 case KVM_SET_MP_STATE: {
2828 struct kvm_mp_state mp_state;
2831 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
2833 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2836 case KVM_TRANSLATE: {
2837 struct kvm_translation tr;
2840 if (copy_from_user(&tr, argp, sizeof(tr)))
2842 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2846 if (copy_to_user(argp, &tr, sizeof(tr)))
2851 case KVM_SET_GUEST_DEBUG: {
2852 struct kvm_guest_debug dbg;
2855 if (copy_from_user(&dbg, argp, sizeof(dbg)))
2857 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2860 case KVM_SET_SIGNAL_MASK: {
2861 struct kvm_signal_mask __user *sigmask_arg = argp;
2862 struct kvm_signal_mask kvm_sigmask;
2863 sigset_t sigset, *p;
2868 if (copy_from_user(&kvm_sigmask, argp,
2869 sizeof(kvm_sigmask)))
2872 if (kvm_sigmask.len != sizeof(sigset))
2875 if (copy_from_user(&sigset, sigmask_arg->sigset,
2880 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2884 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL_ACCOUNT);
2888 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2892 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2898 fpu = memdup_user(argp, sizeof(*fpu));
2904 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2908 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2911 mutex_unlock(&vcpu->mutex);
2917 #ifdef CONFIG_KVM_COMPAT
2918 static long kvm_vcpu_compat_ioctl(struct file *filp,
2919 unsigned int ioctl, unsigned long arg)
2921 struct kvm_vcpu *vcpu = filp->private_data;
2922 void __user *argp = compat_ptr(arg);
2925 if (vcpu->kvm->mm != current->mm)
2929 case KVM_SET_SIGNAL_MASK: {
2930 struct kvm_signal_mask __user *sigmask_arg = argp;
2931 struct kvm_signal_mask kvm_sigmask;
2936 if (copy_from_user(&kvm_sigmask, argp,
2937 sizeof(kvm_sigmask)))
2940 if (kvm_sigmask.len != sizeof(compat_sigset_t))
2943 if (get_compat_sigset(&sigset, (void *)sigmask_arg->sigset))
2945 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2947 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2951 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2959 static int kvm_device_mmap(struct file *filp, struct vm_area_struct *vma)
2961 struct kvm_device *dev = filp->private_data;
2964 return dev->ops->mmap(dev, vma);
2969 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2970 int (*accessor)(struct kvm_device *dev,
2971 struct kvm_device_attr *attr),
2974 struct kvm_device_attr attr;
2979 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2982 return accessor(dev, &attr);
2985 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2988 struct kvm_device *dev = filp->private_data;
2990 if (dev->kvm->mm != current->mm)
2994 case KVM_SET_DEVICE_ATTR:
2995 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2996 case KVM_GET_DEVICE_ATTR:
2997 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2998 case KVM_HAS_DEVICE_ATTR:
2999 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
3001 if (dev->ops->ioctl)
3002 return dev->ops->ioctl(dev, ioctl, arg);
3008 static int kvm_device_release(struct inode *inode, struct file *filp)
3010 struct kvm_device *dev = filp->private_data;
3011 struct kvm *kvm = dev->kvm;
3013 if (dev->ops->release) {
3014 mutex_lock(&kvm->lock);
3015 list_del(&dev->vm_node);
3016 dev->ops->release(dev);
3017 mutex_unlock(&kvm->lock);
3024 static const struct file_operations kvm_device_fops = {
3025 .unlocked_ioctl = kvm_device_ioctl,
3026 .release = kvm_device_release,
3027 KVM_COMPAT(kvm_device_ioctl),
3028 .mmap = kvm_device_mmap,
3031 struct kvm_device *kvm_device_from_filp(struct file *filp)
3033 if (filp->f_op != &kvm_device_fops)
3036 return filp->private_data;
3039 static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
3040 #ifdef CONFIG_KVM_MPIC
3041 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
3042 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
3046 int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
3048 if (type >= ARRAY_SIZE(kvm_device_ops_table))
3051 if (kvm_device_ops_table[type] != NULL)
3054 kvm_device_ops_table[type] = ops;
3058 void kvm_unregister_device_ops(u32 type)
3060 if (kvm_device_ops_table[type] != NULL)
3061 kvm_device_ops_table[type] = NULL;
3064 static int kvm_ioctl_create_device(struct kvm *kvm,
3065 struct kvm_create_device *cd)
3067 struct kvm_device_ops *ops = NULL;
3068 struct kvm_device *dev;
3069 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
3073 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
3076 type = array_index_nospec(cd->type, ARRAY_SIZE(kvm_device_ops_table));
3077 ops = kvm_device_ops_table[type];
3084 dev = kzalloc(sizeof(*dev), GFP_KERNEL_ACCOUNT);
3091 mutex_lock(&kvm->lock);
3092 ret = ops->create(dev, type);
3094 mutex_unlock(&kvm->lock);
3098 list_add(&dev->vm_node, &kvm->devices);
3099 mutex_unlock(&kvm->lock);
3105 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
3108 mutex_lock(&kvm->lock);
3109 list_del(&dev->vm_node);
3110 mutex_unlock(&kvm->lock);
3119 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
3122 case KVM_CAP_USER_MEMORY:
3123 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
3124 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
3125 case KVM_CAP_INTERNAL_ERROR_DATA:
3126 #ifdef CONFIG_HAVE_KVM_MSI
3127 case KVM_CAP_SIGNAL_MSI:
3129 #ifdef CONFIG_HAVE_KVM_IRQFD
3131 case KVM_CAP_IRQFD_RESAMPLE:
3133 case KVM_CAP_IOEVENTFD_ANY_LENGTH:
3134 case KVM_CAP_CHECK_EXTENSION_VM:
3135 case KVM_CAP_ENABLE_CAP_VM:
3136 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3137 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2:
3140 #ifdef CONFIG_KVM_MMIO
3141 case KVM_CAP_COALESCED_MMIO:
3142 return KVM_COALESCED_MMIO_PAGE_OFFSET;
3143 case KVM_CAP_COALESCED_PIO:
3146 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3147 case KVM_CAP_IRQ_ROUTING:
3148 return KVM_MAX_IRQ_ROUTES;
3150 #if KVM_ADDRESS_SPACE_NUM > 1
3151 case KVM_CAP_MULTI_ADDRESS_SPACE:
3152 return KVM_ADDRESS_SPACE_NUM;
3154 case KVM_CAP_NR_MEMSLOTS:
3155 return KVM_USER_MEM_SLOTS;
3159 return kvm_vm_ioctl_check_extension(kvm, arg);
3162 int __attribute__((weak)) kvm_vm_ioctl_enable_cap(struct kvm *kvm,
3163 struct kvm_enable_cap *cap)
3168 static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm,
3169 struct kvm_enable_cap *cap)
3172 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3173 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2:
3174 if (cap->flags || (cap->args[0] & ~1))
3176 kvm->manual_dirty_log_protect = cap->args[0];
3180 return kvm_vm_ioctl_enable_cap(kvm, cap);
3184 static long kvm_vm_ioctl(struct file *filp,
3185 unsigned int ioctl, unsigned long arg)
3187 struct kvm *kvm = filp->private_data;
3188 void __user *argp = (void __user *)arg;
3191 if (kvm->mm != current->mm)
3194 case KVM_CREATE_VCPU:
3195 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
3197 case KVM_ENABLE_CAP: {
3198 struct kvm_enable_cap cap;
3201 if (copy_from_user(&cap, argp, sizeof(cap)))
3203 r = kvm_vm_ioctl_enable_cap_generic(kvm, &cap);
3206 case KVM_SET_USER_MEMORY_REGION: {
3207 struct kvm_userspace_memory_region kvm_userspace_mem;
3210 if (copy_from_user(&kvm_userspace_mem, argp,
3211 sizeof(kvm_userspace_mem)))
3214 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
3217 case KVM_GET_DIRTY_LOG: {
3218 struct kvm_dirty_log log;
3221 if (copy_from_user(&log, argp, sizeof(log)))
3223 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3226 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3227 case KVM_CLEAR_DIRTY_LOG: {
3228 struct kvm_clear_dirty_log log;
3231 if (copy_from_user(&log, argp, sizeof(log)))
3233 r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
3237 #ifdef CONFIG_KVM_MMIO
3238 case KVM_REGISTER_COALESCED_MMIO: {
3239 struct kvm_coalesced_mmio_zone zone;
3242 if (copy_from_user(&zone, argp, sizeof(zone)))
3244 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
3247 case KVM_UNREGISTER_COALESCED_MMIO: {
3248 struct kvm_coalesced_mmio_zone zone;
3251 if (copy_from_user(&zone, argp, sizeof(zone)))
3253 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
3258 struct kvm_irqfd data;
3261 if (copy_from_user(&data, argp, sizeof(data)))
3263 r = kvm_irqfd(kvm, &data);
3266 case KVM_IOEVENTFD: {
3267 struct kvm_ioeventfd data;
3270 if (copy_from_user(&data, argp, sizeof(data)))
3272 r = kvm_ioeventfd(kvm, &data);
3275 #ifdef CONFIG_HAVE_KVM_MSI
3276 case KVM_SIGNAL_MSI: {
3280 if (copy_from_user(&msi, argp, sizeof(msi)))
3282 r = kvm_send_userspace_msi(kvm, &msi);
3286 #ifdef __KVM_HAVE_IRQ_LINE
3287 case KVM_IRQ_LINE_STATUS:
3288 case KVM_IRQ_LINE: {
3289 struct kvm_irq_level irq_event;
3292 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
3295 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
3296 ioctl == KVM_IRQ_LINE_STATUS);
3301 if (ioctl == KVM_IRQ_LINE_STATUS) {
3302 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
3310 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3311 case KVM_SET_GSI_ROUTING: {
3312 struct kvm_irq_routing routing;
3313 struct kvm_irq_routing __user *urouting;
3314 struct kvm_irq_routing_entry *entries = NULL;
3317 if (copy_from_user(&routing, argp, sizeof(routing)))
3320 if (!kvm_arch_can_set_irq_routing(kvm))
3322 if (routing.nr > KVM_MAX_IRQ_ROUTES)
3328 entries = vmalloc(array_size(sizeof(*entries),
3334 if (copy_from_user(entries, urouting->entries,
3335 routing.nr * sizeof(*entries)))
3336 goto out_free_irq_routing;
3338 r = kvm_set_irq_routing(kvm, entries, routing.nr,
3340 out_free_irq_routing:
3344 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
3345 case KVM_CREATE_DEVICE: {
3346 struct kvm_create_device cd;
3349 if (copy_from_user(&cd, argp, sizeof(cd)))
3352 r = kvm_ioctl_create_device(kvm, &cd);
3357 if (copy_to_user(argp, &cd, sizeof(cd)))
3363 case KVM_CHECK_EXTENSION:
3364 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
3367 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
3373 #ifdef CONFIG_KVM_COMPAT
3374 struct compat_kvm_dirty_log {
3378 compat_uptr_t dirty_bitmap; /* one bit per page */
3383 static long kvm_vm_compat_ioctl(struct file *filp,
3384 unsigned int ioctl, unsigned long arg)
3386 struct kvm *kvm = filp->private_data;
3389 if (kvm->mm != current->mm)
3392 case KVM_GET_DIRTY_LOG: {
3393 struct compat_kvm_dirty_log compat_log;
3394 struct kvm_dirty_log log;
3396 if (copy_from_user(&compat_log, (void __user *)arg,
3397 sizeof(compat_log)))
3399 log.slot = compat_log.slot;
3400 log.padding1 = compat_log.padding1;
3401 log.padding2 = compat_log.padding2;
3402 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
3404 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3408 r = kvm_vm_ioctl(filp, ioctl, arg);
3414 static struct file_operations kvm_vm_fops = {
3415 .release = kvm_vm_release,
3416 .unlocked_ioctl = kvm_vm_ioctl,
3417 .llseek = noop_llseek,
3418 KVM_COMPAT(kvm_vm_compat_ioctl),
3421 static int kvm_dev_ioctl_create_vm(unsigned long type)
3427 kvm = kvm_create_vm(type);
3429 return PTR_ERR(kvm);
3430 #ifdef CONFIG_KVM_MMIO
3431 r = kvm_coalesced_mmio_init(kvm);
3435 r = get_unused_fd_flags(O_CLOEXEC);
3439 file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
3447 * Don't call kvm_put_kvm anymore at this point; file->f_op is
3448 * already set, with ->release() being kvm_vm_release(). In error
3449 * cases it will be called by the final fput(file) and will take
3450 * care of doing kvm_put_kvm(kvm).
3452 if (kvm_create_vm_debugfs(kvm, r) < 0) {
3457 kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm);
3459 fd_install(r, file);
3467 static long kvm_dev_ioctl(struct file *filp,
3468 unsigned int ioctl, unsigned long arg)
3473 case KVM_GET_API_VERSION:
3476 r = KVM_API_VERSION;
3479 r = kvm_dev_ioctl_create_vm(arg);
3481 case KVM_CHECK_EXTENSION:
3482 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
3484 case KVM_GET_VCPU_MMAP_SIZE:
3487 r = PAGE_SIZE; /* struct kvm_run */
3489 r += PAGE_SIZE; /* pio data page */
3491 #ifdef CONFIG_KVM_MMIO
3492 r += PAGE_SIZE; /* coalesced mmio ring page */
3495 case KVM_TRACE_ENABLE:
3496 case KVM_TRACE_PAUSE:
3497 case KVM_TRACE_DISABLE:
3501 return kvm_arch_dev_ioctl(filp, ioctl, arg);
3507 static struct file_operations kvm_chardev_ops = {
3508 .unlocked_ioctl = kvm_dev_ioctl,
3509 .llseek = noop_llseek,
3510 KVM_COMPAT(kvm_dev_ioctl),
3513 static struct miscdevice kvm_dev = {
3519 static void hardware_enable_nolock(void *junk)
3521 int cpu = raw_smp_processor_id();
3524 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
3527 cpumask_set_cpu(cpu, cpus_hardware_enabled);
3529 r = kvm_arch_hardware_enable();
3532 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3533 atomic_inc(&hardware_enable_failed);
3534 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
3538 static int kvm_starting_cpu(unsigned int cpu)
3540 raw_spin_lock(&kvm_count_lock);
3541 if (kvm_usage_count)
3542 hardware_enable_nolock(NULL);
3543 raw_spin_unlock(&kvm_count_lock);
3547 static void hardware_disable_nolock(void *junk)
3549 int cpu = raw_smp_processor_id();
3551 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
3553 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3554 kvm_arch_hardware_disable();
3557 static int kvm_dying_cpu(unsigned int cpu)
3559 raw_spin_lock(&kvm_count_lock);
3560 if (kvm_usage_count)
3561 hardware_disable_nolock(NULL);
3562 raw_spin_unlock(&kvm_count_lock);
3566 static void hardware_disable_all_nolock(void)
3568 BUG_ON(!kvm_usage_count);
3571 if (!kvm_usage_count)
3572 on_each_cpu(hardware_disable_nolock, NULL, 1);
3575 static void hardware_disable_all(void)
3577 raw_spin_lock(&kvm_count_lock);
3578 hardware_disable_all_nolock();
3579 raw_spin_unlock(&kvm_count_lock);
3582 static int hardware_enable_all(void)
3586 raw_spin_lock(&kvm_count_lock);
3589 if (kvm_usage_count == 1) {
3590 atomic_set(&hardware_enable_failed, 0);
3591 on_each_cpu(hardware_enable_nolock, NULL, 1);
3593 if (atomic_read(&hardware_enable_failed)) {
3594 hardware_disable_all_nolock();
3599 raw_spin_unlock(&kvm_count_lock);
3604 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3608 * Some (well, at least mine) BIOSes hang on reboot if
3611 * And Intel TXT required VMX off for all cpu when system shutdown.
3613 pr_info("kvm: exiting hardware virtualization\n");
3614 kvm_rebooting = true;
3615 on_each_cpu(hardware_disable_nolock, NULL, 1);
3619 static struct notifier_block kvm_reboot_notifier = {
3620 .notifier_call = kvm_reboot,
3624 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3628 for (i = 0; i < bus->dev_count; i++) {
3629 struct kvm_io_device *pos = bus->range[i].dev;
3631 kvm_iodevice_destructor(pos);
3636 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
3637 const struct kvm_io_range *r2)
3639 gpa_t addr1 = r1->addr;
3640 gpa_t addr2 = r2->addr;
3645 /* If r2->len == 0, match the exact address. If r2->len != 0,
3646 * accept any overlapping write. Any order is acceptable for
3647 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
3648 * we process all of them.
3661 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
3663 return kvm_io_bus_cmp(p1, p2);
3666 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
3667 gpa_t addr, int len)
3669 struct kvm_io_range *range, key;
3672 key = (struct kvm_io_range) {
3677 range = bsearch(&key, bus->range, bus->dev_count,
3678 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
3682 off = range - bus->range;
3684 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
3690 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3691 struct kvm_io_range *range, const void *val)
3695 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3699 while (idx < bus->dev_count &&
3700 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3701 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
3710 /* kvm_io_bus_write - called under kvm->slots_lock */
3711 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3712 int len, const void *val)
3714 struct kvm_io_bus *bus;
3715 struct kvm_io_range range;
3718 range = (struct kvm_io_range) {
3723 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3726 r = __kvm_io_bus_write(vcpu, bus, &range, val);
3727 return r < 0 ? r : 0;
3729 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
3731 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3732 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
3733 gpa_t addr, int len, const void *val, long cookie)
3735 struct kvm_io_bus *bus;
3736 struct kvm_io_range range;
3738 range = (struct kvm_io_range) {
3743 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3747 /* First try the device referenced by cookie. */
3748 if ((cookie >= 0) && (cookie < bus->dev_count) &&
3749 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
3750 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
3755 * cookie contained garbage; fall back to search and return the
3756 * correct cookie value.
3758 return __kvm_io_bus_write(vcpu, bus, &range, val);
3761 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3762 struct kvm_io_range *range, void *val)
3766 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3770 while (idx < bus->dev_count &&
3771 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3772 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
3781 /* kvm_io_bus_read - called under kvm->slots_lock */
3782 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3785 struct kvm_io_bus *bus;
3786 struct kvm_io_range range;
3789 range = (struct kvm_io_range) {
3794 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3797 r = __kvm_io_bus_read(vcpu, bus, &range, val);
3798 return r < 0 ? r : 0;
3801 /* Caller must hold slots_lock. */
3802 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3803 int len, struct kvm_io_device *dev)
3806 struct kvm_io_bus *new_bus, *bus;
3807 struct kvm_io_range range;
3809 bus = kvm_get_bus(kvm, bus_idx);
3813 /* exclude ioeventfd which is limited by maximum fd */
3814 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3817 new_bus = kmalloc(struct_size(bus, range, bus->dev_count + 1),
3818 GFP_KERNEL_ACCOUNT);
3822 range = (struct kvm_io_range) {
3828 for (i = 0; i < bus->dev_count; i++)
3829 if (kvm_io_bus_cmp(&bus->range[i], &range) > 0)
3832 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3833 new_bus->dev_count++;
3834 new_bus->range[i] = range;
3835 memcpy(new_bus->range + i + 1, bus->range + i,
3836 (bus->dev_count - i) * sizeof(struct kvm_io_range));
3837 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3838 synchronize_srcu_expedited(&kvm->srcu);
3844 /* Caller must hold slots_lock. */
3845 void kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3846 struct kvm_io_device *dev)
3849 struct kvm_io_bus *new_bus, *bus;
3851 bus = kvm_get_bus(kvm, bus_idx);
3855 for (i = 0; i < bus->dev_count; i++)
3856 if (bus->range[i].dev == dev) {
3860 if (i == bus->dev_count)
3863 new_bus = kmalloc(struct_size(bus, range, bus->dev_count - 1),
3864 GFP_KERNEL_ACCOUNT);
3866 pr_err("kvm: failed to shrink bus, removing it completely\n");
3870 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3871 new_bus->dev_count--;
3872 memcpy(new_bus->range + i, bus->range + i + 1,
3873 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3876 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3877 synchronize_srcu_expedited(&kvm->srcu);
3882 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3885 struct kvm_io_bus *bus;
3886 int dev_idx, srcu_idx;
3887 struct kvm_io_device *iodev = NULL;
3889 srcu_idx = srcu_read_lock(&kvm->srcu);
3891 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
3895 dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
3899 iodev = bus->range[dev_idx].dev;
3902 srcu_read_unlock(&kvm->srcu, srcu_idx);
3906 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
3908 static int kvm_debugfs_open(struct inode *inode, struct file *file,
3909 int (*get)(void *, u64 *), int (*set)(void *, u64),
3912 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
3915 /* The debugfs files are a reference to the kvm struct which
3916 * is still valid when kvm_destroy_vm is called.
3917 * To avoid the race between open and the removal of the debugfs
3918 * directory we test against the users count.
3920 if (!refcount_inc_not_zero(&stat_data->kvm->users_count))
3923 if (simple_attr_open(inode, file, get, set, fmt)) {
3924 kvm_put_kvm(stat_data->kvm);
3931 static int kvm_debugfs_release(struct inode *inode, struct file *file)
3933 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
3936 simple_attr_release(inode, file);
3937 kvm_put_kvm(stat_data->kvm);
3942 static int vm_stat_get_per_vm(void *data, u64 *val)
3944 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3946 *val = *(ulong *)((void *)stat_data->kvm + stat_data->offset);
3951 static int vm_stat_clear_per_vm(void *data, u64 val)
3953 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3958 *(ulong *)((void *)stat_data->kvm + stat_data->offset) = 0;
3963 static int vm_stat_get_per_vm_open(struct inode *inode, struct file *file)
3965 __simple_attr_check_format("%llu\n", 0ull);
3966 return kvm_debugfs_open(inode, file, vm_stat_get_per_vm,
3967 vm_stat_clear_per_vm, "%llu\n");
3970 static const struct file_operations vm_stat_get_per_vm_fops = {
3971 .owner = THIS_MODULE,
3972 .open = vm_stat_get_per_vm_open,
3973 .release = kvm_debugfs_release,
3974 .read = simple_attr_read,
3975 .write = simple_attr_write,
3976 .llseek = no_llseek,
3979 static int vcpu_stat_get_per_vm(void *data, u64 *val)
3982 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3983 struct kvm_vcpu *vcpu;
3987 kvm_for_each_vcpu(i, vcpu, stat_data->kvm)
3988 *val += *(u64 *)((void *)vcpu + stat_data->offset);
3993 static int vcpu_stat_clear_per_vm(void *data, u64 val)
3996 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3997 struct kvm_vcpu *vcpu;
4002 kvm_for_each_vcpu(i, vcpu, stat_data->kvm)
4003 *(u64 *)((void *)vcpu + stat_data->offset) = 0;
4008 static int vcpu_stat_get_per_vm_open(struct inode *inode, struct file *file)
4010 __simple_attr_check_format("%llu\n", 0ull);
4011 return kvm_debugfs_open(inode, file, vcpu_stat_get_per_vm,
4012 vcpu_stat_clear_per_vm, "%llu\n");
4015 static const struct file_operations vcpu_stat_get_per_vm_fops = {
4016 .owner = THIS_MODULE,
4017 .open = vcpu_stat_get_per_vm_open,
4018 .release = kvm_debugfs_release,
4019 .read = simple_attr_read,
4020 .write = simple_attr_write,
4021 .llseek = no_llseek,
4024 static const struct file_operations *stat_fops_per_vm[] = {
4025 [KVM_STAT_VCPU] = &vcpu_stat_get_per_vm_fops,
4026 [KVM_STAT_VM] = &vm_stat_get_per_vm_fops,
4029 static int vm_stat_get(void *_offset, u64 *val)
4031 unsigned offset = (long)_offset;
4033 struct kvm_stat_data stat_tmp = {.offset = offset};
4037 spin_lock(&kvm_lock);
4038 list_for_each_entry(kvm, &vm_list, vm_list) {
4040 vm_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
4043 spin_unlock(&kvm_lock);
4047 static int vm_stat_clear(void *_offset, u64 val)
4049 unsigned offset = (long)_offset;
4051 struct kvm_stat_data stat_tmp = {.offset = offset};
4056 spin_lock(&kvm_lock);
4057 list_for_each_entry(kvm, &vm_list, vm_list) {
4059 vm_stat_clear_per_vm((void *)&stat_tmp, 0);
4061 spin_unlock(&kvm_lock);
4066 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n");
4068 static int vcpu_stat_get(void *_offset, u64 *val)
4070 unsigned offset = (long)_offset;
4072 struct kvm_stat_data stat_tmp = {.offset = offset};
4076 spin_lock(&kvm_lock);
4077 list_for_each_entry(kvm, &vm_list, vm_list) {
4079 vcpu_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
4082 spin_unlock(&kvm_lock);
4086 static int vcpu_stat_clear(void *_offset, u64 val)
4088 unsigned offset = (long)_offset;
4090 struct kvm_stat_data stat_tmp = {.offset = offset};
4095 spin_lock(&kvm_lock);
4096 list_for_each_entry(kvm, &vm_list, vm_list) {
4098 vcpu_stat_clear_per_vm((void *)&stat_tmp, 0);
4100 spin_unlock(&kvm_lock);
4105 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear,
4108 static const struct file_operations *stat_fops[] = {
4109 [KVM_STAT_VCPU] = &vcpu_stat_fops,
4110 [KVM_STAT_VM] = &vm_stat_fops,
4113 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm)
4115 struct kobj_uevent_env *env;
4116 unsigned long long created, active;
4118 if (!kvm_dev.this_device || !kvm)
4121 spin_lock(&kvm_lock);
4122 if (type == KVM_EVENT_CREATE_VM) {
4123 kvm_createvm_count++;
4125 } else if (type == KVM_EVENT_DESTROY_VM) {
4128 created = kvm_createvm_count;
4129 active = kvm_active_vms;
4130 spin_unlock(&kvm_lock);
4132 env = kzalloc(sizeof(*env), GFP_KERNEL_ACCOUNT);
4136 add_uevent_var(env, "CREATED=%llu", created);
4137 add_uevent_var(env, "COUNT=%llu", active);
4139 if (type == KVM_EVENT_CREATE_VM) {
4140 add_uevent_var(env, "EVENT=create");
4141 kvm->userspace_pid = task_pid_nr(current);
4142 } else if (type == KVM_EVENT_DESTROY_VM) {
4143 add_uevent_var(env, "EVENT=destroy");
4145 add_uevent_var(env, "PID=%d", kvm->userspace_pid);
4147 if (!IS_ERR_OR_NULL(kvm->debugfs_dentry)) {
4148 char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL_ACCOUNT);
4151 tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX);
4153 add_uevent_var(env, "STATS_PATH=%s", tmp);
4157 /* no need for checks, since we are adding at most only 5 keys */
4158 env->envp[env->envp_idx++] = NULL;
4159 kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp);
4163 static void kvm_init_debug(void)
4165 struct kvm_stats_debugfs_item *p;
4167 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
4169 kvm_debugfs_num_entries = 0;
4170 for (p = debugfs_entries; p->name; ++p, kvm_debugfs_num_entries++) {
4171 debugfs_create_file(p->name, 0644, kvm_debugfs_dir,
4172 (void *)(long)p->offset,
4173 stat_fops[p->kind]);
4177 static int kvm_suspend(void)
4179 if (kvm_usage_count)
4180 hardware_disable_nolock(NULL);
4184 static void kvm_resume(void)
4186 if (kvm_usage_count) {
4187 #ifdef CONFIG_LOCKDEP
4188 WARN_ON(lockdep_is_held(&kvm_count_lock));
4190 hardware_enable_nolock(NULL);
4194 static struct syscore_ops kvm_syscore_ops = {
4195 .suspend = kvm_suspend,
4196 .resume = kvm_resume,
4200 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
4202 return container_of(pn, struct kvm_vcpu, preempt_notifier);
4205 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
4207 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
4209 if (vcpu->preempted)
4210 vcpu->preempted = false;
4212 kvm_arch_sched_in(vcpu, cpu);
4214 kvm_arch_vcpu_load(vcpu, cpu);
4217 static void kvm_sched_out(struct preempt_notifier *pn,
4218 struct task_struct *next)
4220 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
4222 if (current->state == TASK_RUNNING)
4223 vcpu->preempted = true;
4224 kvm_arch_vcpu_put(vcpu);
4227 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
4228 struct module *module)
4233 r = kvm_arch_init(opaque);
4238 * kvm_arch_init makes sure there's at most one caller
4239 * for architectures that support multiple implementations,
4240 * like intel and amd on x86.
4241 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
4242 * conflicts in case kvm is already setup for another implementation.
4244 r = kvm_irqfd_init();
4248 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
4253 r = kvm_arch_hardware_setup();
4257 for_each_online_cpu(cpu) {
4258 smp_call_function_single(cpu,
4259 kvm_arch_check_processor_compat,
4265 r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting",
4266 kvm_starting_cpu, kvm_dying_cpu);
4269 register_reboot_notifier(&kvm_reboot_notifier);
4271 /* A kmem cache lets us meet the alignment requirements of fx_save. */
4273 vcpu_align = __alignof__(struct kvm_vcpu);
4275 kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align,
4277 offsetof(struct kvm_vcpu, arch),
4278 sizeof_field(struct kvm_vcpu, arch),
4280 if (!kvm_vcpu_cache) {
4285 r = kvm_async_pf_init();
4289 kvm_chardev_ops.owner = module;
4290 kvm_vm_fops.owner = module;
4291 kvm_vcpu_fops.owner = module;
4293 r = misc_register(&kvm_dev);
4295 pr_err("kvm: misc device register failed\n");
4299 register_syscore_ops(&kvm_syscore_ops);
4301 kvm_preempt_ops.sched_in = kvm_sched_in;
4302 kvm_preempt_ops.sched_out = kvm_sched_out;
4306 r = kvm_vfio_ops_init();
4312 kvm_async_pf_deinit();
4314 kmem_cache_destroy(kvm_vcpu_cache);
4316 unregister_reboot_notifier(&kvm_reboot_notifier);
4317 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4320 kvm_arch_hardware_unsetup();
4322 free_cpumask_var(cpus_hardware_enabled);
4330 EXPORT_SYMBOL_GPL(kvm_init);
4334 debugfs_remove_recursive(kvm_debugfs_dir);
4335 misc_deregister(&kvm_dev);
4336 kmem_cache_destroy(kvm_vcpu_cache);
4337 kvm_async_pf_deinit();
4338 unregister_syscore_ops(&kvm_syscore_ops);
4339 unregister_reboot_notifier(&kvm_reboot_notifier);
4340 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4341 on_each_cpu(hardware_disable_nolock, NULL, 1);
4342 kvm_arch_hardware_unsetup();
4345 free_cpumask_var(cpus_hardware_enabled);
4346 kvm_vfio_ops_exit();
4348 EXPORT_SYMBOL_GPL(kvm_exit);