1 // SPDX-License-Identifier: GPL-2.0-only
3 * Kernel-based Virtual Machine driver for Linux
5 * This module enables machines with Intel VT-x extensions to run virtual
6 * machines without emulation or binary translation.
8 * Copyright (C) 2006 Qumranet, Inc.
9 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
12 * Avi Kivity <avi@qumranet.com>
13 * Yaniv Kamay <yaniv@qumranet.com>
16 #include <kvm/iodev.h>
18 #include <linux/kvm_host.h>
19 #include <linux/kvm.h>
20 #include <linux/module.h>
21 #include <linux/errno.h>
22 #include <linux/percpu.h>
24 #include <linux/miscdevice.h>
25 #include <linux/vmalloc.h>
26 #include <linux/reboot.h>
27 #include <linux/debugfs.h>
28 #include <linux/highmem.h>
29 #include <linux/file.h>
30 #include <linux/syscore_ops.h>
31 #include <linux/cpu.h>
32 #include <linux/sched/signal.h>
33 #include <linux/sched/mm.h>
34 #include <linux/sched/stat.h>
35 #include <linux/cpumask.h>
36 #include <linux/smp.h>
37 #include <linux/anon_inodes.h>
38 #include <linux/profile.h>
39 #include <linux/kvm_para.h>
40 #include <linux/pagemap.h>
41 #include <linux/mman.h>
42 #include <linux/swap.h>
43 #include <linux/bitops.h>
44 #include <linux/spinlock.h>
45 #include <linux/compat.h>
46 #include <linux/srcu.h>
47 #include <linux/hugetlb.h>
48 #include <linux/slab.h>
49 #include <linux/sort.h>
50 #include <linux/bsearch.h>
52 #include <linux/lockdep.h>
54 #include <asm/processor.h>
55 #include <asm/ioctl.h>
56 #include <linux/uaccess.h>
57 #include <asm/pgtable.h>
59 #include "coalesced_mmio.h"
63 #define CREATE_TRACE_POINTS
64 #include <trace/events/kvm.h>
66 /* Worst case buffer size needed for holding an integer. */
67 #define ITOA_MAX_LEN 12
69 MODULE_AUTHOR("Qumranet");
70 MODULE_LICENSE("GPL");
72 /* Architectures should define their poll value according to the halt latency */
73 unsigned int halt_poll_ns = KVM_HALT_POLL_NS_DEFAULT;
74 module_param(halt_poll_ns, uint, 0644);
75 EXPORT_SYMBOL_GPL(halt_poll_ns);
77 /* Default doubles per-vcpu halt_poll_ns. */
78 unsigned int halt_poll_ns_grow = 2;
79 module_param(halt_poll_ns_grow, uint, 0644);
80 EXPORT_SYMBOL_GPL(halt_poll_ns_grow);
82 /* The start value to grow halt_poll_ns from */
83 unsigned int halt_poll_ns_grow_start = 10000; /* 10us */
84 module_param(halt_poll_ns_grow_start, uint, 0644);
85 EXPORT_SYMBOL_GPL(halt_poll_ns_grow_start);
87 /* Default resets per-vcpu halt_poll_ns . */
88 unsigned int halt_poll_ns_shrink;
89 module_param(halt_poll_ns_shrink, uint, 0644);
90 EXPORT_SYMBOL_GPL(halt_poll_ns_shrink);
95 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
98 DEFINE_MUTEX(kvm_lock);
99 static DEFINE_RAW_SPINLOCK(kvm_count_lock);
102 static cpumask_var_t cpus_hardware_enabled;
103 static int kvm_usage_count;
104 static atomic_t hardware_enable_failed;
106 struct kmem_cache *kvm_vcpu_cache;
107 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
109 static __read_mostly struct preempt_ops kvm_preempt_ops;
111 struct dentry *kvm_debugfs_dir;
112 EXPORT_SYMBOL_GPL(kvm_debugfs_dir);
114 static int kvm_debugfs_num_entries;
115 static const struct file_operations *stat_fops_per_vm[];
117 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
119 #ifdef CONFIG_KVM_COMPAT
120 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
122 #define KVM_COMPAT(c) .compat_ioctl = (c)
124 static long kvm_no_compat_ioctl(struct file *file, unsigned int ioctl,
125 unsigned long arg) { return -EINVAL; }
126 #define KVM_COMPAT(c) .compat_ioctl = kvm_no_compat_ioctl
128 static int hardware_enable_all(void);
129 static void hardware_disable_all(void);
131 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
133 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot, gfn_t gfn);
135 __visible bool kvm_rebooting;
136 EXPORT_SYMBOL_GPL(kvm_rebooting);
138 static bool largepages_enabled = true;
140 #define KVM_EVENT_CREATE_VM 0
141 #define KVM_EVENT_DESTROY_VM 1
142 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm);
143 static unsigned long long kvm_createvm_count;
144 static unsigned long long kvm_active_vms;
146 __weak int kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
147 unsigned long start, unsigned long end, bool blockable)
152 bool kvm_is_reserved_pfn(kvm_pfn_t pfn)
155 return PageReserved(pfn_to_page(pfn));
161 * Switches to specified vcpu, until a matching vcpu_put()
163 void vcpu_load(struct kvm_vcpu *vcpu)
166 preempt_notifier_register(&vcpu->preempt_notifier);
167 kvm_arch_vcpu_load(vcpu, cpu);
170 EXPORT_SYMBOL_GPL(vcpu_load);
172 void vcpu_put(struct kvm_vcpu *vcpu)
175 kvm_arch_vcpu_put(vcpu);
176 preempt_notifier_unregister(&vcpu->preempt_notifier);
179 EXPORT_SYMBOL_GPL(vcpu_put);
181 /* TODO: merge with kvm_arch_vcpu_should_kick */
182 static bool kvm_request_needs_ipi(struct kvm_vcpu *vcpu, unsigned req)
184 int mode = kvm_vcpu_exiting_guest_mode(vcpu);
187 * We need to wait for the VCPU to reenable interrupts and get out of
188 * READING_SHADOW_PAGE_TABLES mode.
190 if (req & KVM_REQUEST_WAIT)
191 return mode != OUTSIDE_GUEST_MODE;
194 * Need to kick a running VCPU, but otherwise there is nothing to do.
196 return mode == IN_GUEST_MODE;
199 static void ack_flush(void *_completed)
203 static inline bool kvm_kick_many_cpus(const struct cpumask *cpus, bool wait)
206 cpus = cpu_online_mask;
208 if (cpumask_empty(cpus))
211 smp_call_function_many(cpus, ack_flush, NULL, wait);
215 bool kvm_make_vcpus_request_mask(struct kvm *kvm, unsigned int req,
216 unsigned long *vcpu_bitmap, cpumask_var_t tmp)
219 struct kvm_vcpu *vcpu;
224 kvm_for_each_vcpu(i, vcpu, kvm) {
225 if (vcpu_bitmap && !test_bit(i, vcpu_bitmap))
228 kvm_make_request(req, vcpu);
231 if (!(req & KVM_REQUEST_NO_WAKEUP) && kvm_vcpu_wake_up(vcpu))
234 if (tmp != NULL && cpu != -1 && cpu != me &&
235 kvm_request_needs_ipi(vcpu, req))
236 __cpumask_set_cpu(cpu, tmp);
239 called = kvm_kick_many_cpus(tmp, !!(req & KVM_REQUEST_WAIT));
245 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
250 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
252 called = kvm_make_vcpus_request_mask(kvm, req, NULL, cpus);
254 free_cpumask_var(cpus);
258 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
259 void kvm_flush_remote_tlbs(struct kvm *kvm)
262 * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in
263 * kvm_make_all_cpus_request.
265 long dirty_count = smp_load_acquire(&kvm->tlbs_dirty);
268 * We want to publish modifications to the page tables before reading
269 * mode. Pairs with a memory barrier in arch-specific code.
270 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
271 * and smp_mb in walk_shadow_page_lockless_begin/end.
272 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
274 * There is already an smp_mb__after_atomic() before
275 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
278 if (!kvm_arch_flush_remote_tlb(kvm)
279 || kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
280 ++kvm->stat.remote_tlb_flush;
281 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
283 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
286 void kvm_reload_remote_mmus(struct kvm *kvm)
288 kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
291 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
296 mutex_init(&vcpu->mutex);
301 init_swait_queue_head(&vcpu->wq);
302 kvm_async_pf_vcpu_init(vcpu);
305 INIT_LIST_HEAD(&vcpu->blocked_vcpu_list);
307 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
312 vcpu->run = page_address(page);
314 kvm_vcpu_set_in_spin_loop(vcpu, false);
315 kvm_vcpu_set_dy_eligible(vcpu, false);
316 vcpu->preempted = false;
319 r = kvm_arch_vcpu_init(vcpu);
325 free_page((unsigned long)vcpu->run);
329 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
331 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
334 * no need for rcu_read_lock as VCPU_RUN is the only place that
335 * will change the vcpu->pid pointer and on uninit all file
336 * descriptors are already gone.
338 put_pid(rcu_dereference_protected(vcpu->pid, 1));
339 kvm_arch_vcpu_uninit(vcpu);
340 free_page((unsigned long)vcpu->run);
342 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
344 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
345 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
347 return container_of(mn, struct kvm, mmu_notifier);
350 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
351 struct mm_struct *mm,
352 unsigned long address,
355 struct kvm *kvm = mmu_notifier_to_kvm(mn);
358 idx = srcu_read_lock(&kvm->srcu);
359 spin_lock(&kvm->mmu_lock);
360 kvm->mmu_notifier_seq++;
362 if (kvm_set_spte_hva(kvm, address, pte))
363 kvm_flush_remote_tlbs(kvm);
365 spin_unlock(&kvm->mmu_lock);
366 srcu_read_unlock(&kvm->srcu, idx);
369 static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
370 const struct mmu_notifier_range *range)
372 struct kvm *kvm = mmu_notifier_to_kvm(mn);
373 int need_tlb_flush = 0, idx;
376 idx = srcu_read_lock(&kvm->srcu);
377 spin_lock(&kvm->mmu_lock);
379 * The count increase must become visible at unlock time as no
380 * spte can be established without taking the mmu_lock and
381 * count is also read inside the mmu_lock critical section.
383 kvm->mmu_notifier_count++;
384 need_tlb_flush = kvm_unmap_hva_range(kvm, range->start, range->end);
385 need_tlb_flush |= kvm->tlbs_dirty;
386 /* we've to flush the tlb before the pages can be freed */
388 kvm_flush_remote_tlbs(kvm);
390 spin_unlock(&kvm->mmu_lock);
392 ret = kvm_arch_mmu_notifier_invalidate_range(kvm, range->start,
394 mmu_notifier_range_blockable(range));
396 srcu_read_unlock(&kvm->srcu, idx);
401 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
402 const struct mmu_notifier_range *range)
404 struct kvm *kvm = mmu_notifier_to_kvm(mn);
406 spin_lock(&kvm->mmu_lock);
408 * This sequence increase will notify the kvm page fault that
409 * the page that is going to be mapped in the spte could have
412 kvm->mmu_notifier_seq++;
415 * The above sequence increase must be visible before the
416 * below count decrease, which is ensured by the smp_wmb above
417 * in conjunction with the smp_rmb in mmu_notifier_retry().
419 kvm->mmu_notifier_count--;
420 spin_unlock(&kvm->mmu_lock);
422 BUG_ON(kvm->mmu_notifier_count < 0);
425 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
426 struct mm_struct *mm,
430 struct kvm *kvm = mmu_notifier_to_kvm(mn);
433 idx = srcu_read_lock(&kvm->srcu);
434 spin_lock(&kvm->mmu_lock);
436 young = kvm_age_hva(kvm, start, end);
438 kvm_flush_remote_tlbs(kvm);
440 spin_unlock(&kvm->mmu_lock);
441 srcu_read_unlock(&kvm->srcu, idx);
446 static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
447 struct mm_struct *mm,
451 struct kvm *kvm = mmu_notifier_to_kvm(mn);
454 idx = srcu_read_lock(&kvm->srcu);
455 spin_lock(&kvm->mmu_lock);
457 * Even though we do not flush TLB, this will still adversely
458 * affect performance on pre-Haswell Intel EPT, where there is
459 * no EPT Access Bit to clear so that we have to tear down EPT
460 * tables instead. If we find this unacceptable, we can always
461 * add a parameter to kvm_age_hva so that it effectively doesn't
462 * do anything on clear_young.
464 * Also note that currently we never issue secondary TLB flushes
465 * from clear_young, leaving this job up to the regular system
466 * cadence. If we find this inaccurate, we might come up with a
467 * more sophisticated heuristic later.
469 young = kvm_age_hva(kvm, start, end);
470 spin_unlock(&kvm->mmu_lock);
471 srcu_read_unlock(&kvm->srcu, idx);
476 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
477 struct mm_struct *mm,
478 unsigned long address)
480 struct kvm *kvm = mmu_notifier_to_kvm(mn);
483 idx = srcu_read_lock(&kvm->srcu);
484 spin_lock(&kvm->mmu_lock);
485 young = kvm_test_age_hva(kvm, address);
486 spin_unlock(&kvm->mmu_lock);
487 srcu_read_unlock(&kvm->srcu, idx);
492 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
493 struct mm_struct *mm)
495 struct kvm *kvm = mmu_notifier_to_kvm(mn);
498 idx = srcu_read_lock(&kvm->srcu);
499 kvm_arch_flush_shadow_all(kvm);
500 srcu_read_unlock(&kvm->srcu, idx);
503 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
504 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
505 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
506 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
507 .clear_young = kvm_mmu_notifier_clear_young,
508 .test_young = kvm_mmu_notifier_test_young,
509 .change_pte = kvm_mmu_notifier_change_pte,
510 .release = kvm_mmu_notifier_release,
513 static int kvm_init_mmu_notifier(struct kvm *kvm)
515 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
516 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
519 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
521 static int kvm_init_mmu_notifier(struct kvm *kvm)
526 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
528 static struct kvm_memslots *kvm_alloc_memslots(void)
531 struct kvm_memslots *slots;
533 slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL_ACCOUNT);
537 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
538 slots->id_to_index[i] = slots->memslots[i].id = i;
543 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
545 if (!memslot->dirty_bitmap)
548 kvfree(memslot->dirty_bitmap);
549 memslot->dirty_bitmap = NULL;
553 * Free any memory in @free but not in @dont.
555 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
556 struct kvm_memory_slot *dont)
558 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
559 kvm_destroy_dirty_bitmap(free);
561 kvm_arch_free_memslot(kvm, free, dont);
566 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
568 struct kvm_memory_slot *memslot;
573 kvm_for_each_memslot(memslot, slots)
574 kvm_free_memslot(kvm, memslot, NULL);
579 static void kvm_destroy_vm_debugfs(struct kvm *kvm)
583 if (!kvm->debugfs_dentry)
586 debugfs_remove_recursive(kvm->debugfs_dentry);
588 if (kvm->debugfs_stat_data) {
589 for (i = 0; i < kvm_debugfs_num_entries; i++)
590 kfree(kvm->debugfs_stat_data[i]);
591 kfree(kvm->debugfs_stat_data);
595 static int kvm_create_vm_debugfs(struct kvm *kvm, int fd)
597 char dir_name[ITOA_MAX_LEN * 2];
598 struct kvm_stat_data *stat_data;
599 struct kvm_stats_debugfs_item *p;
601 if (!debugfs_initialized())
604 snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd);
605 kvm->debugfs_dentry = debugfs_create_dir(dir_name, kvm_debugfs_dir);
607 kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries,
608 sizeof(*kvm->debugfs_stat_data),
610 if (!kvm->debugfs_stat_data)
613 for (p = debugfs_entries; p->name; p++) {
614 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT);
618 stat_data->kvm = kvm;
619 stat_data->offset = p->offset;
620 stat_data->mode = p->mode ? p->mode : 0644;
621 kvm->debugfs_stat_data[p - debugfs_entries] = stat_data;
622 debugfs_create_file(p->name, stat_data->mode, kvm->debugfs_dentry,
623 stat_data, stat_fops_per_vm[p->kind]);
628 static struct kvm *kvm_create_vm(unsigned long type)
631 struct kvm *kvm = kvm_arch_alloc_vm();
634 return ERR_PTR(-ENOMEM);
636 spin_lock_init(&kvm->mmu_lock);
638 kvm->mm = current->mm;
639 kvm_eventfd_init(kvm);
640 mutex_init(&kvm->lock);
641 mutex_init(&kvm->irq_lock);
642 mutex_init(&kvm->slots_lock);
643 refcount_set(&kvm->users_count, 1);
644 INIT_LIST_HEAD(&kvm->devices);
646 r = kvm_arch_init_vm(kvm, type);
648 goto out_err_no_disable;
650 r = hardware_enable_all();
652 goto out_err_no_disable;
654 #ifdef CONFIG_HAVE_KVM_IRQFD
655 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
658 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
661 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
662 struct kvm_memslots *slots = kvm_alloc_memslots();
664 goto out_err_no_srcu;
665 /* Generations must be different for each address space. */
666 slots->generation = i;
667 rcu_assign_pointer(kvm->memslots[i], slots);
670 if (init_srcu_struct(&kvm->srcu))
671 goto out_err_no_srcu;
672 if (init_srcu_struct(&kvm->irq_srcu))
673 goto out_err_no_irq_srcu;
674 for (i = 0; i < KVM_NR_BUSES; i++) {
675 rcu_assign_pointer(kvm->buses[i],
676 kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL_ACCOUNT));
681 r = kvm_init_mmu_notifier(kvm);
685 mutex_lock(&kvm_lock);
686 list_add(&kvm->vm_list, &vm_list);
687 mutex_unlock(&kvm_lock);
689 preempt_notifier_inc();
694 cleanup_srcu_struct(&kvm->irq_srcu);
696 cleanup_srcu_struct(&kvm->srcu);
698 hardware_disable_all();
700 refcount_set(&kvm->users_count, 0);
701 for (i = 0; i < KVM_NR_BUSES; i++)
702 kfree(kvm_get_bus(kvm, i));
703 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
704 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
705 kvm_arch_free_vm(kvm);
710 static void kvm_destroy_devices(struct kvm *kvm)
712 struct kvm_device *dev, *tmp;
715 * We do not need to take the kvm->lock here, because nobody else
716 * has a reference to the struct kvm at this point and therefore
717 * cannot access the devices list anyhow.
719 list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
720 list_del(&dev->vm_node);
721 dev->ops->destroy(dev);
725 static void kvm_destroy_vm(struct kvm *kvm)
728 struct mm_struct *mm = kvm->mm;
730 kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm);
731 kvm_destroy_vm_debugfs(kvm);
732 kvm_arch_sync_events(kvm);
733 mutex_lock(&kvm_lock);
734 list_del(&kvm->vm_list);
735 mutex_unlock(&kvm_lock);
736 kvm_free_irq_routing(kvm);
737 for (i = 0; i < KVM_NR_BUSES; i++) {
738 struct kvm_io_bus *bus = kvm_get_bus(kvm, i);
741 kvm_io_bus_destroy(bus);
742 kvm->buses[i] = NULL;
744 kvm_coalesced_mmio_free(kvm);
745 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
746 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
748 kvm_arch_flush_shadow_all(kvm);
750 kvm_arch_destroy_vm(kvm);
751 kvm_destroy_devices(kvm);
752 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
753 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
754 cleanup_srcu_struct(&kvm->irq_srcu);
755 cleanup_srcu_struct(&kvm->srcu);
756 kvm_arch_free_vm(kvm);
757 preempt_notifier_dec();
758 hardware_disable_all();
762 void kvm_get_kvm(struct kvm *kvm)
764 refcount_inc(&kvm->users_count);
766 EXPORT_SYMBOL_GPL(kvm_get_kvm);
768 void kvm_put_kvm(struct kvm *kvm)
770 if (refcount_dec_and_test(&kvm->users_count))
773 EXPORT_SYMBOL_GPL(kvm_put_kvm);
776 static int kvm_vm_release(struct inode *inode, struct file *filp)
778 struct kvm *kvm = filp->private_data;
780 kvm_irqfd_release(kvm);
787 * Allocation size is twice as large as the actual dirty bitmap size.
788 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
790 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
792 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
794 memslot->dirty_bitmap = kvzalloc(dirty_bytes, GFP_KERNEL_ACCOUNT);
795 if (!memslot->dirty_bitmap)
802 * Insert memslot and re-sort memslots based on their GFN,
803 * so binary search could be used to lookup GFN.
804 * Sorting algorithm takes advantage of having initially
805 * sorted array and known changed memslot position.
807 static void update_memslots(struct kvm_memslots *slots,
808 struct kvm_memory_slot *new,
809 enum kvm_mr_change change)
812 int i = slots->id_to_index[id];
813 struct kvm_memory_slot *mslots = slots->memslots;
815 WARN_ON(mslots[i].id != id);
819 WARN_ON(mslots[i].npages || !new->npages);
823 WARN_ON(new->npages || !mslots[i].npages);
829 while (i < KVM_MEM_SLOTS_NUM - 1 &&
830 new->base_gfn <= mslots[i + 1].base_gfn) {
831 if (!mslots[i + 1].npages)
833 mslots[i] = mslots[i + 1];
834 slots->id_to_index[mslots[i].id] = i;
839 * The ">=" is needed when creating a slot with base_gfn == 0,
840 * so that it moves before all those with base_gfn == npages == 0.
842 * On the other hand, if new->npages is zero, the above loop has
843 * already left i pointing to the beginning of the empty part of
844 * mslots, and the ">=" would move the hole backwards in this
845 * case---which is wrong. So skip the loop when deleting a slot.
849 new->base_gfn >= mslots[i - 1].base_gfn) {
850 mslots[i] = mslots[i - 1];
851 slots->id_to_index[mslots[i].id] = i;
855 WARN_ON_ONCE(i != slots->used_slots);
858 slots->id_to_index[mslots[i].id] = i;
861 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
863 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
865 #ifdef __KVM_HAVE_READONLY_MEM
866 valid_flags |= KVM_MEM_READONLY;
869 if (mem->flags & ~valid_flags)
875 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
876 int as_id, struct kvm_memslots *slots)
878 struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id);
879 u64 gen = old_memslots->generation;
881 WARN_ON(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS);
882 slots->generation = gen | KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
884 rcu_assign_pointer(kvm->memslots[as_id], slots);
885 synchronize_srcu_expedited(&kvm->srcu);
888 * Increment the new memslot generation a second time, dropping the
889 * update in-progress flag and incrementing then generation based on
890 * the number of address spaces. This provides a unique and easily
891 * identifiable generation number while the memslots are in flux.
893 gen = slots->generation & ~KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
896 * Generations must be unique even across address spaces. We do not need
897 * a global counter for that, instead the generation space is evenly split
898 * across address spaces. For example, with two address spaces, address
899 * space 0 will use generations 0, 2, 4, ... while address space 1 will
900 * use generations 1, 3, 5, ...
902 gen += KVM_ADDRESS_SPACE_NUM;
904 kvm_arch_memslots_updated(kvm, gen);
906 slots->generation = gen;
912 * Allocate some memory and give it an address in the guest physical address
915 * Discontiguous memory is allowed, mostly for framebuffers.
917 * Must be called holding kvm->slots_lock for write.
919 int __kvm_set_memory_region(struct kvm *kvm,
920 const struct kvm_userspace_memory_region *mem)
924 unsigned long npages;
925 struct kvm_memory_slot *slot;
926 struct kvm_memory_slot old, new;
927 struct kvm_memslots *slots = NULL, *old_memslots;
929 enum kvm_mr_change change;
931 r = check_memory_region_flags(mem);
936 as_id = mem->slot >> 16;
939 /* General sanity checks */
940 if (mem->memory_size & (PAGE_SIZE - 1))
942 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
944 /* We can read the guest memory with __xxx_user() later on. */
945 if ((id < KVM_USER_MEM_SLOTS) &&
946 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
947 !access_ok((void __user *)(unsigned long)mem->userspace_addr,
950 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
952 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
955 slot = id_to_memslot(__kvm_memslots(kvm, as_id), id);
956 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
957 npages = mem->memory_size >> PAGE_SHIFT;
959 if (npages > KVM_MEM_MAX_NR_PAGES)
965 new.base_gfn = base_gfn;
967 new.flags = mem->flags;
971 change = KVM_MR_CREATE;
972 else { /* Modify an existing slot. */
973 if ((mem->userspace_addr != old.userspace_addr) ||
974 (npages != old.npages) ||
975 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
978 if (base_gfn != old.base_gfn)
979 change = KVM_MR_MOVE;
980 else if (new.flags != old.flags)
981 change = KVM_MR_FLAGS_ONLY;
982 else { /* Nothing to change. */
991 change = KVM_MR_DELETE;
996 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
997 /* Check for overlaps */
999 kvm_for_each_memslot(slot, __kvm_memslots(kvm, as_id)) {
1002 if (!((base_gfn + npages <= slot->base_gfn) ||
1003 (base_gfn >= slot->base_gfn + slot->npages)))
1008 /* Free page dirty bitmap if unneeded */
1009 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
1010 new.dirty_bitmap = NULL;
1013 if (change == KVM_MR_CREATE) {
1014 new.userspace_addr = mem->userspace_addr;
1016 if (kvm_arch_create_memslot(kvm, &new, npages))
1020 /* Allocate page dirty bitmap if needed */
1021 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
1022 if (kvm_create_dirty_bitmap(&new) < 0)
1026 slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL_ACCOUNT);
1029 memcpy(slots, __kvm_memslots(kvm, as_id), sizeof(struct kvm_memslots));
1031 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
1032 slot = id_to_memslot(slots, id);
1033 slot->flags |= KVM_MEMSLOT_INVALID;
1035 old_memslots = install_new_memslots(kvm, as_id, slots);
1037 /* From this point no new shadow pages pointing to a deleted,
1038 * or moved, memslot will be created.
1040 * validation of sp->gfn happens in:
1041 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1042 * - kvm_is_visible_gfn (mmu_check_roots)
1044 kvm_arch_flush_shadow_memslot(kvm, slot);
1047 * We can re-use the old_memslots from above, the only difference
1048 * from the currently installed memslots is the invalid flag. This
1049 * will get overwritten by update_memslots anyway.
1051 slots = old_memslots;
1054 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
1058 /* actual memory is freed via old in kvm_free_memslot below */
1059 if (change == KVM_MR_DELETE) {
1060 new.dirty_bitmap = NULL;
1061 memset(&new.arch, 0, sizeof(new.arch));
1064 update_memslots(slots, &new, change);
1065 old_memslots = install_new_memslots(kvm, as_id, slots);
1067 kvm_arch_commit_memory_region(kvm, mem, &old, &new, change);
1069 kvm_free_memslot(kvm, &old, &new);
1070 kvfree(old_memslots);
1076 kvm_free_memslot(kvm, &new, &old);
1080 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1082 int kvm_set_memory_region(struct kvm *kvm,
1083 const struct kvm_userspace_memory_region *mem)
1087 mutex_lock(&kvm->slots_lock);
1088 r = __kvm_set_memory_region(kvm, mem);
1089 mutex_unlock(&kvm->slots_lock);
1092 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1094 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1095 struct kvm_userspace_memory_region *mem)
1097 if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1100 return kvm_set_memory_region(kvm, mem);
1103 int kvm_get_dirty_log(struct kvm *kvm,
1104 struct kvm_dirty_log *log, int *is_dirty)
1106 struct kvm_memslots *slots;
1107 struct kvm_memory_slot *memslot;
1110 unsigned long any = 0;
1112 as_id = log->slot >> 16;
1113 id = (u16)log->slot;
1114 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1117 slots = __kvm_memslots(kvm, as_id);
1118 memslot = id_to_memslot(slots, id);
1119 if (!memslot->dirty_bitmap)
1122 n = kvm_dirty_bitmap_bytes(memslot);
1124 for (i = 0; !any && i < n/sizeof(long); ++i)
1125 any = memslot->dirty_bitmap[i];
1127 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1134 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1136 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1138 * kvm_get_dirty_log_protect - get a snapshot of dirty pages
1139 * and reenable dirty page tracking for the corresponding pages.
1140 * @kvm: pointer to kvm instance
1141 * @log: slot id and address to which we copy the log
1142 * @flush: true if TLB flush is needed by caller
1144 * We need to keep it in mind that VCPU threads can write to the bitmap
1145 * concurrently. So, to avoid losing track of dirty pages we keep the
1148 * 1. Take a snapshot of the bit and clear it if needed.
1149 * 2. Write protect the corresponding page.
1150 * 3. Copy the snapshot to the userspace.
1151 * 4. Upon return caller flushes TLB's if needed.
1153 * Between 2 and 4, the guest may write to the page using the remaining TLB
1154 * entry. This is not a problem because the page is reported dirty using
1155 * the snapshot taken before and step 4 ensures that writes done after
1156 * exiting to userspace will be logged for the next call.
1159 int kvm_get_dirty_log_protect(struct kvm *kvm,
1160 struct kvm_dirty_log *log, bool *flush)
1162 struct kvm_memslots *slots;
1163 struct kvm_memory_slot *memslot;
1166 unsigned long *dirty_bitmap;
1167 unsigned long *dirty_bitmap_buffer;
1169 as_id = log->slot >> 16;
1170 id = (u16)log->slot;
1171 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1174 slots = __kvm_memslots(kvm, as_id);
1175 memslot = id_to_memslot(slots, id);
1177 dirty_bitmap = memslot->dirty_bitmap;
1181 n = kvm_dirty_bitmap_bytes(memslot);
1183 if (kvm->manual_dirty_log_protect) {
1185 * Unlike kvm_get_dirty_log, we always return false in *flush,
1186 * because no flush is needed until KVM_CLEAR_DIRTY_LOG. There
1187 * is some code duplication between this function and
1188 * kvm_get_dirty_log, but hopefully all architecture
1189 * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log
1190 * can be eliminated.
1192 dirty_bitmap_buffer = dirty_bitmap;
1194 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
1195 memset(dirty_bitmap_buffer, 0, n);
1197 spin_lock(&kvm->mmu_lock);
1198 for (i = 0; i < n / sizeof(long); i++) {
1202 if (!dirty_bitmap[i])
1206 mask = xchg(&dirty_bitmap[i], 0);
1207 dirty_bitmap_buffer[i] = mask;
1209 offset = i * BITS_PER_LONG;
1210 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1213 spin_unlock(&kvm->mmu_lock);
1216 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1220 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect);
1223 * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap
1224 * and reenable dirty page tracking for the corresponding pages.
1225 * @kvm: pointer to kvm instance
1226 * @log: slot id and address from which to fetch the bitmap of dirty pages
1227 * @flush: true if TLB flush is needed by caller
1229 int kvm_clear_dirty_log_protect(struct kvm *kvm,
1230 struct kvm_clear_dirty_log *log, bool *flush)
1232 struct kvm_memslots *slots;
1233 struct kvm_memory_slot *memslot;
1237 unsigned long *dirty_bitmap;
1238 unsigned long *dirty_bitmap_buffer;
1240 as_id = log->slot >> 16;
1241 id = (u16)log->slot;
1242 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1245 if (log->first_page & 63)
1248 slots = __kvm_memslots(kvm, as_id);
1249 memslot = id_to_memslot(slots, id);
1251 dirty_bitmap = memslot->dirty_bitmap;
1255 n = ALIGN(log->num_pages, BITS_PER_LONG) / 8;
1257 if (log->first_page > memslot->npages ||
1258 log->num_pages > memslot->npages - log->first_page ||
1259 (log->num_pages < memslot->npages - log->first_page && (log->num_pages & 63)))
1263 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
1264 if (copy_from_user(dirty_bitmap_buffer, log->dirty_bitmap, n))
1267 spin_lock(&kvm->mmu_lock);
1268 for (offset = log->first_page, i = offset / BITS_PER_LONG,
1269 n = DIV_ROUND_UP(log->num_pages, BITS_PER_LONG); n--;
1270 i++, offset += BITS_PER_LONG) {
1271 unsigned long mask = *dirty_bitmap_buffer++;
1272 atomic_long_t *p = (atomic_long_t *) &dirty_bitmap[i];
1276 mask &= atomic_long_fetch_andnot(mask, p);
1279 * mask contains the bits that really have been cleared. This
1280 * never includes any bits beyond the length of the memslot (if
1281 * the length is not aligned to 64 pages), therefore it is not
1282 * a problem if userspace sets them in log->dirty_bitmap.
1286 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1290 spin_unlock(&kvm->mmu_lock);
1294 EXPORT_SYMBOL_GPL(kvm_clear_dirty_log_protect);
1297 bool kvm_largepages_enabled(void)
1299 return largepages_enabled;
1302 void kvm_disable_largepages(void)
1304 largepages_enabled = false;
1306 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1308 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1310 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1312 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1314 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
1316 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn);
1319 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1321 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1323 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1324 memslot->flags & KVM_MEMSLOT_INVALID)
1329 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1331 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1333 struct vm_area_struct *vma;
1334 unsigned long addr, size;
1338 addr = gfn_to_hva(kvm, gfn);
1339 if (kvm_is_error_hva(addr))
1342 down_read(¤t->mm->mmap_sem);
1343 vma = find_vma(current->mm, addr);
1347 size = vma_kernel_pagesize(vma);
1350 up_read(¤t->mm->mmap_sem);
1355 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1357 return slot->flags & KVM_MEM_READONLY;
1360 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1361 gfn_t *nr_pages, bool write)
1363 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1364 return KVM_HVA_ERR_BAD;
1366 if (memslot_is_readonly(slot) && write)
1367 return KVM_HVA_ERR_RO_BAD;
1370 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1372 return __gfn_to_hva_memslot(slot, gfn);
1375 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1378 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1381 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1384 return gfn_to_hva_many(slot, gfn, NULL);
1386 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1388 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1390 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1392 EXPORT_SYMBOL_GPL(gfn_to_hva);
1394 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
1396 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
1398 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
1401 * Return the hva of a @gfn and the R/W attribute if possible.
1403 * @slot: the kvm_memory_slot which contains @gfn
1404 * @gfn: the gfn to be translated
1405 * @writable: used to return the read/write attribute of the @slot if the hva
1406 * is valid and @writable is not NULL
1408 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1409 gfn_t gfn, bool *writable)
1411 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1413 if (!kvm_is_error_hva(hva) && writable)
1414 *writable = !memslot_is_readonly(slot);
1419 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1421 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1423 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1426 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
1428 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1430 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1433 static inline int check_user_page_hwpoison(unsigned long addr)
1435 int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
1437 rc = get_user_pages(addr, 1, flags, NULL, NULL);
1438 return rc == -EHWPOISON;
1442 * The fast path to get the writable pfn which will be stored in @pfn,
1443 * true indicates success, otherwise false is returned. It's also the
1444 * only part that runs if we can are in atomic context.
1446 static bool hva_to_pfn_fast(unsigned long addr, bool write_fault,
1447 bool *writable, kvm_pfn_t *pfn)
1449 struct page *page[1];
1453 * Fast pin a writable pfn only if it is a write fault request
1454 * or the caller allows to map a writable pfn for a read fault
1457 if (!(write_fault || writable))
1460 npages = __get_user_pages_fast(addr, 1, 1, page);
1462 *pfn = page_to_pfn(page[0]);
1473 * The slow path to get the pfn of the specified host virtual address,
1474 * 1 indicates success, -errno is returned if error is detected.
1476 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1477 bool *writable, kvm_pfn_t *pfn)
1479 unsigned int flags = FOLL_HWPOISON;
1486 *writable = write_fault;
1489 flags |= FOLL_WRITE;
1491 flags |= FOLL_NOWAIT;
1493 npages = get_user_pages_unlocked(addr, 1, &page, flags);
1497 /* map read fault as writable if possible */
1498 if (unlikely(!write_fault) && writable) {
1501 if (__get_user_pages_fast(addr, 1, 1, &wpage) == 1) {
1507 *pfn = page_to_pfn(page);
1511 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1513 if (unlikely(!(vma->vm_flags & VM_READ)))
1516 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1522 static int hva_to_pfn_remapped(struct vm_area_struct *vma,
1523 unsigned long addr, bool *async,
1524 bool write_fault, bool *writable,
1530 r = follow_pfn(vma, addr, &pfn);
1533 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
1534 * not call the fault handler, so do it here.
1536 bool unlocked = false;
1537 r = fixup_user_fault(current, current->mm, addr,
1538 (write_fault ? FAULT_FLAG_WRITE : 0),
1545 r = follow_pfn(vma, addr, &pfn);
1555 * Get a reference here because callers of *hva_to_pfn* and
1556 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
1557 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
1558 * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will
1559 * simply do nothing for reserved pfns.
1561 * Whoever called remap_pfn_range is also going to call e.g.
1562 * unmap_mapping_range before the underlying pages are freed,
1563 * causing a call to our MMU notifier.
1572 * Pin guest page in memory and return its pfn.
1573 * @addr: host virtual address which maps memory to the guest
1574 * @atomic: whether this function can sleep
1575 * @async: whether this function need to wait IO complete if the
1576 * host page is not in the memory
1577 * @write_fault: whether we should get a writable host page
1578 * @writable: whether it allows to map a writable host page for !@write_fault
1580 * The function will map a writable host page for these two cases:
1581 * 1): @write_fault = true
1582 * 2): @write_fault = false && @writable, @writable will tell the caller
1583 * whether the mapping is writable.
1585 static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1586 bool write_fault, bool *writable)
1588 struct vm_area_struct *vma;
1592 /* we can do it either atomically or asynchronously, not both */
1593 BUG_ON(atomic && async);
1595 if (hva_to_pfn_fast(addr, write_fault, writable, &pfn))
1599 return KVM_PFN_ERR_FAULT;
1601 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1605 down_read(¤t->mm->mmap_sem);
1606 if (npages == -EHWPOISON ||
1607 (!async && check_user_page_hwpoison(addr))) {
1608 pfn = KVM_PFN_ERR_HWPOISON;
1613 vma = find_vma_intersection(current->mm, addr, addr + 1);
1616 pfn = KVM_PFN_ERR_FAULT;
1617 else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
1618 r = hva_to_pfn_remapped(vma, addr, async, write_fault, writable, &pfn);
1622 pfn = KVM_PFN_ERR_FAULT;
1624 if (async && vma_is_valid(vma, write_fault))
1626 pfn = KVM_PFN_ERR_FAULT;
1629 up_read(¤t->mm->mmap_sem);
1633 kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn,
1634 bool atomic, bool *async, bool write_fault,
1637 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1639 if (addr == KVM_HVA_ERR_RO_BAD) {
1642 return KVM_PFN_ERR_RO_FAULT;
1645 if (kvm_is_error_hva(addr)) {
1648 return KVM_PFN_NOSLOT;
1651 /* Do not map writable pfn in the readonly memslot. */
1652 if (writable && memslot_is_readonly(slot)) {
1657 return hva_to_pfn(addr, atomic, async, write_fault,
1660 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
1662 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1665 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
1666 write_fault, writable);
1668 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1670 kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1672 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1674 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
1676 kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1678 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1680 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1682 kvm_pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1684 return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm, gfn), gfn);
1686 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1688 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
1690 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1692 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
1694 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1696 return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
1698 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1700 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
1702 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1704 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
1706 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1707 struct page **pages, int nr_pages)
1712 addr = gfn_to_hva_many(slot, gfn, &entry);
1713 if (kvm_is_error_hva(addr))
1716 if (entry < nr_pages)
1719 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1721 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1723 static struct page *kvm_pfn_to_page(kvm_pfn_t pfn)
1725 if (is_error_noslot_pfn(pfn))
1726 return KVM_ERR_PTR_BAD_PAGE;
1728 if (kvm_is_reserved_pfn(pfn)) {
1730 return KVM_ERR_PTR_BAD_PAGE;
1733 return pfn_to_page(pfn);
1736 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1740 pfn = gfn_to_pfn(kvm, gfn);
1742 return kvm_pfn_to_page(pfn);
1744 EXPORT_SYMBOL_GPL(gfn_to_page);
1746 static int __kvm_map_gfn(struct kvm_memory_slot *slot, gfn_t gfn,
1747 struct kvm_host_map *map)
1751 struct page *page = KVM_UNMAPPED_PAGE;
1756 pfn = gfn_to_pfn_memslot(slot, gfn);
1757 if (is_error_noslot_pfn(pfn))
1760 if (pfn_valid(pfn)) {
1761 page = pfn_to_page(pfn);
1763 #ifdef CONFIG_HAS_IOMEM
1765 hva = memremap(pfn_to_hpa(pfn), PAGE_SIZE, MEMREMAP_WB);
1780 int kvm_vcpu_map(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map)
1782 return __kvm_map_gfn(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, map);
1784 EXPORT_SYMBOL_GPL(kvm_vcpu_map);
1786 void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map,
1795 if (map->page != KVM_UNMAPPED_PAGE)
1797 #ifdef CONFIG_HAS_IOMEM
1803 kvm_vcpu_mark_page_dirty(vcpu, map->gfn);
1804 kvm_release_pfn_dirty(map->pfn);
1806 kvm_release_pfn_clean(map->pfn);
1812 EXPORT_SYMBOL_GPL(kvm_vcpu_unmap);
1814 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
1818 pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
1820 return kvm_pfn_to_page(pfn);
1822 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
1824 void kvm_release_page_clean(struct page *page)
1826 WARN_ON(is_error_page(page));
1828 kvm_release_pfn_clean(page_to_pfn(page));
1830 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1832 void kvm_release_pfn_clean(kvm_pfn_t pfn)
1834 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
1835 put_page(pfn_to_page(pfn));
1837 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1839 void kvm_release_page_dirty(struct page *page)
1841 WARN_ON(is_error_page(page));
1843 kvm_release_pfn_dirty(page_to_pfn(page));
1845 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1847 void kvm_release_pfn_dirty(kvm_pfn_t pfn)
1849 kvm_set_pfn_dirty(pfn);
1850 kvm_release_pfn_clean(pfn);
1852 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1854 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
1856 if (!kvm_is_reserved_pfn(pfn)) {
1857 struct page *page = pfn_to_page(pfn);
1862 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1864 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
1866 if (!kvm_is_reserved_pfn(pfn))
1867 mark_page_accessed(pfn_to_page(pfn));
1869 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1871 void kvm_get_pfn(kvm_pfn_t pfn)
1873 if (!kvm_is_reserved_pfn(pfn))
1874 get_page(pfn_to_page(pfn));
1876 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1878 static int next_segment(unsigned long len, int offset)
1880 if (len > PAGE_SIZE - offset)
1881 return PAGE_SIZE - offset;
1886 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
1887 void *data, int offset, int len)
1892 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1893 if (kvm_is_error_hva(addr))
1895 r = __copy_from_user(data, (void __user *)addr + offset, len);
1901 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1904 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1906 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1908 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1910 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
1911 int offset, int len)
1913 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1915 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1917 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
1919 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1921 gfn_t gfn = gpa >> PAGE_SHIFT;
1923 int offset = offset_in_page(gpa);
1926 while ((seg = next_segment(len, offset)) != 0) {
1927 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1937 EXPORT_SYMBOL_GPL(kvm_read_guest);
1939 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
1941 gfn_t gfn = gpa >> PAGE_SHIFT;
1943 int offset = offset_in_page(gpa);
1946 while ((seg = next_segment(len, offset)) != 0) {
1947 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
1957 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
1959 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1960 void *data, int offset, unsigned long len)
1965 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1966 if (kvm_is_error_hva(addr))
1968 pagefault_disable();
1969 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1976 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1979 gfn_t gfn = gpa >> PAGE_SHIFT;
1980 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1981 int offset = offset_in_page(gpa);
1983 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1985 EXPORT_SYMBOL_GPL(kvm_read_guest_atomic);
1987 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
1988 void *data, unsigned long len)
1990 gfn_t gfn = gpa >> PAGE_SHIFT;
1991 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1992 int offset = offset_in_page(gpa);
1994 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1996 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
1998 static int __kvm_write_guest_page(struct kvm_memory_slot *memslot, gfn_t gfn,
1999 const void *data, int offset, int len)
2004 addr = gfn_to_hva_memslot(memslot, gfn);
2005 if (kvm_is_error_hva(addr))
2007 r = __copy_to_user((void __user *)addr + offset, data, len);
2010 mark_page_dirty_in_slot(memslot, gfn);
2014 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
2015 const void *data, int offset, int len)
2017 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2019 return __kvm_write_guest_page(slot, gfn, data, offset, len);
2021 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
2023 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
2024 const void *data, int offset, int len)
2026 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2028 return __kvm_write_guest_page(slot, gfn, data, offset, len);
2030 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
2032 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
2035 gfn_t gfn = gpa >> PAGE_SHIFT;
2037 int offset = offset_in_page(gpa);
2040 while ((seg = next_segment(len, offset)) != 0) {
2041 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
2051 EXPORT_SYMBOL_GPL(kvm_write_guest);
2053 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
2056 gfn_t gfn = gpa >> PAGE_SHIFT;
2058 int offset = offset_in_page(gpa);
2061 while ((seg = next_segment(len, offset)) != 0) {
2062 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
2072 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
2074 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
2075 struct gfn_to_hva_cache *ghc,
2076 gpa_t gpa, unsigned long len)
2078 int offset = offset_in_page(gpa);
2079 gfn_t start_gfn = gpa >> PAGE_SHIFT;
2080 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
2081 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
2082 gfn_t nr_pages_avail;
2083 int r = start_gfn <= end_gfn ? 0 : -EINVAL;
2086 ghc->generation = slots->generation;
2088 ghc->hva = KVM_HVA_ERR_BAD;
2091 * If the requested region crosses two memslots, we still
2092 * verify that the entire region is valid here.
2094 while (!r && start_gfn <= end_gfn) {
2095 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
2096 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
2098 if (kvm_is_error_hva(ghc->hva))
2100 start_gfn += nr_pages_avail;
2103 /* Use the slow path for cross page reads and writes. */
2104 if (!r && nr_pages_needed == 1)
2107 ghc->memslot = NULL;
2112 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2113 gpa_t gpa, unsigned long len)
2115 struct kvm_memslots *slots = kvm_memslots(kvm);
2116 return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len);
2118 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
2120 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2121 void *data, unsigned int offset,
2124 struct kvm_memslots *slots = kvm_memslots(kvm);
2126 gpa_t gpa = ghc->gpa + offset;
2128 BUG_ON(len + offset > ghc->len);
2130 if (slots->generation != ghc->generation)
2131 __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len);
2133 if (unlikely(!ghc->memslot))
2134 return kvm_write_guest(kvm, gpa, data, len);
2136 if (kvm_is_error_hva(ghc->hva))
2139 r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
2142 mark_page_dirty_in_slot(ghc->memslot, gpa >> PAGE_SHIFT);
2146 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached);
2148 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2149 void *data, unsigned long len)
2151 return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len);
2153 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
2155 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2156 void *data, unsigned long len)
2158 struct kvm_memslots *slots = kvm_memslots(kvm);
2161 BUG_ON(len > ghc->len);
2163 if (slots->generation != ghc->generation)
2164 __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len);
2166 if (unlikely(!ghc->memslot))
2167 return kvm_read_guest(kvm, ghc->gpa, data, len);
2169 if (kvm_is_error_hva(ghc->hva))
2172 r = __copy_from_user(data, (void __user *)ghc->hva, len);
2178 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
2180 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
2182 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
2184 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
2186 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
2188 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
2190 gfn_t gfn = gpa >> PAGE_SHIFT;
2192 int offset = offset_in_page(gpa);
2195 while ((seg = next_segment(len, offset)) != 0) {
2196 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
2205 EXPORT_SYMBOL_GPL(kvm_clear_guest);
2207 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot,
2210 if (memslot && memslot->dirty_bitmap) {
2211 unsigned long rel_gfn = gfn - memslot->base_gfn;
2213 set_bit_le(rel_gfn, memslot->dirty_bitmap);
2217 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
2219 struct kvm_memory_slot *memslot;
2221 memslot = gfn_to_memslot(kvm, gfn);
2222 mark_page_dirty_in_slot(memslot, gfn);
2224 EXPORT_SYMBOL_GPL(mark_page_dirty);
2226 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
2228 struct kvm_memory_slot *memslot;
2230 memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2231 mark_page_dirty_in_slot(memslot, gfn);
2233 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
2235 void kvm_sigset_activate(struct kvm_vcpu *vcpu)
2237 if (!vcpu->sigset_active)
2241 * This does a lockless modification of ->real_blocked, which is fine
2242 * because, only current can change ->real_blocked and all readers of
2243 * ->real_blocked don't care as long ->real_blocked is always a subset
2246 sigprocmask(SIG_SETMASK, &vcpu->sigset, ¤t->real_blocked);
2249 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu)
2251 if (!vcpu->sigset_active)
2254 sigprocmask(SIG_SETMASK, ¤t->real_blocked, NULL);
2255 sigemptyset(¤t->real_blocked);
2258 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
2260 unsigned int old, val, grow, grow_start;
2262 old = val = vcpu->halt_poll_ns;
2263 grow_start = READ_ONCE(halt_poll_ns_grow_start);
2264 grow = READ_ONCE(halt_poll_ns_grow);
2269 if (val < grow_start)
2272 if (val > halt_poll_ns)
2275 vcpu->halt_poll_ns = val;
2277 trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
2280 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
2282 unsigned int old, val, shrink;
2284 old = val = vcpu->halt_poll_ns;
2285 shrink = READ_ONCE(halt_poll_ns_shrink);
2291 vcpu->halt_poll_ns = val;
2292 trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
2295 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
2298 int idx = srcu_read_lock(&vcpu->kvm->srcu);
2300 if (kvm_arch_vcpu_runnable(vcpu)) {
2301 kvm_make_request(KVM_REQ_UNHALT, vcpu);
2304 if (kvm_cpu_has_pending_timer(vcpu))
2306 if (signal_pending(current))
2311 srcu_read_unlock(&vcpu->kvm->srcu, idx);
2316 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2318 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
2321 DECLARE_SWAITQUEUE(wait);
2322 bool waited = false;
2325 kvm_arch_vcpu_blocking(vcpu);
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));
2348 prepare_to_swait_exclusive(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
2350 if (kvm_vcpu_check_block(vcpu) < 0)
2357 finish_swait(&vcpu->wq, &wait);
2360 kvm_arch_vcpu_unblocking(vcpu);
2361 block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
2363 if (!kvm_arch_no_poll(vcpu)) {
2364 if (!vcpu_valid_wakeup(vcpu)) {
2365 shrink_halt_poll_ns(vcpu);
2366 } else if (halt_poll_ns) {
2367 if (block_ns <= vcpu->halt_poll_ns)
2369 /* we had a long block, shrink polling */
2370 else if (vcpu->halt_poll_ns && block_ns > halt_poll_ns)
2371 shrink_halt_poll_ns(vcpu);
2372 /* we had a short halt and our poll time is too small */
2373 else if (vcpu->halt_poll_ns < halt_poll_ns &&
2374 block_ns < halt_poll_ns)
2375 grow_halt_poll_ns(vcpu);
2377 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 WRITE_ONCE(vcpu->ready, true);
2394 ++vcpu->stat.halt_wakeup;
2400 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
2404 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2406 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
2409 int cpu = vcpu->cpu;
2411 if (kvm_vcpu_wake_up(vcpu))
2415 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
2416 if (kvm_arch_vcpu_should_kick(vcpu))
2417 smp_send_reschedule(cpu);
2420 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
2421 #endif /* !CONFIG_S390 */
2423 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
2426 struct task_struct *task = NULL;
2430 pid = rcu_dereference(target->pid);
2432 task = get_pid_task(pid, PIDTYPE_PID);
2436 ret = yield_to(task, 1);
2437 put_task_struct(task);
2441 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
2444 * Helper that checks whether a VCPU is eligible for directed yield.
2445 * Most eligible candidate to yield is decided by following heuristics:
2447 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2448 * (preempted lock holder), indicated by @in_spin_loop.
2449 * Set at the beiginning and cleared at the end of interception/PLE handler.
2451 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2452 * chance last time (mostly it has become eligible now since we have probably
2453 * yielded to lockholder in last iteration. This is done by toggling
2454 * @dy_eligible each time a VCPU checked for eligibility.)
2456 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2457 * to preempted lock-holder could result in wrong VCPU selection and CPU
2458 * burning. Giving priority for a potential lock-holder increases lock
2461 * Since algorithm is based on heuristics, accessing another VCPU data without
2462 * locking does not harm. It may result in trying to yield to same VCPU, fail
2463 * and continue with next VCPU and so on.
2465 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
2467 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2470 eligible = !vcpu->spin_loop.in_spin_loop ||
2471 vcpu->spin_loop.dy_eligible;
2473 if (vcpu->spin_loop.in_spin_loop)
2474 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
2483 * Unlike kvm_arch_vcpu_runnable, this function is called outside
2484 * a vcpu_load/vcpu_put pair. However, for most architectures
2485 * kvm_arch_vcpu_runnable does not require vcpu_load.
2487 bool __weak kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
2489 return kvm_arch_vcpu_runnable(vcpu);
2492 static bool vcpu_dy_runnable(struct kvm_vcpu *vcpu)
2494 if (kvm_arch_dy_runnable(vcpu))
2497 #ifdef CONFIG_KVM_ASYNC_PF
2498 if (!list_empty_careful(&vcpu->async_pf.done))
2505 void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode)
2507 struct kvm *kvm = me->kvm;
2508 struct kvm_vcpu *vcpu;
2509 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
2515 kvm_vcpu_set_in_spin_loop(me, true);
2517 * We boost the priority of a VCPU that is runnable but not
2518 * currently running, because it got preempted by something
2519 * else and called schedule in __vcpu_run. Hopefully that
2520 * VCPU is holding the lock that we need and will release it.
2521 * We approximate round-robin by starting at the last boosted VCPU.
2523 for (pass = 0; pass < 2 && !yielded && try; pass++) {
2524 kvm_for_each_vcpu(i, vcpu, kvm) {
2525 if (!pass && i <= last_boosted_vcpu) {
2526 i = last_boosted_vcpu;
2528 } else if (pass && i > last_boosted_vcpu)
2530 if (!READ_ONCE(vcpu->ready))
2534 if (swait_active(&vcpu->wq) && !vcpu_dy_runnable(vcpu))
2536 if (READ_ONCE(vcpu->preempted) && yield_to_kernel_mode &&
2537 !kvm_arch_vcpu_in_kernel(vcpu))
2539 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
2542 yielded = kvm_vcpu_yield_to(vcpu);
2544 kvm->last_boosted_vcpu = i;
2546 } else if (yielded < 0) {
2553 kvm_vcpu_set_in_spin_loop(me, false);
2555 /* Ensure vcpu is not eligible during next spinloop */
2556 kvm_vcpu_set_dy_eligible(me, false);
2558 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
2560 static vm_fault_t kvm_vcpu_fault(struct vm_fault *vmf)
2562 struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data;
2565 if (vmf->pgoff == 0)
2566 page = virt_to_page(vcpu->run);
2568 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
2569 page = virt_to_page(vcpu->arch.pio_data);
2571 #ifdef CONFIG_KVM_MMIO
2572 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
2573 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
2576 return kvm_arch_vcpu_fault(vcpu, vmf);
2582 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
2583 .fault = kvm_vcpu_fault,
2586 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2588 vma->vm_ops = &kvm_vcpu_vm_ops;
2592 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2594 struct kvm_vcpu *vcpu = filp->private_data;
2596 debugfs_remove_recursive(vcpu->debugfs_dentry);
2597 kvm_put_kvm(vcpu->kvm);
2601 static struct file_operations kvm_vcpu_fops = {
2602 .release = kvm_vcpu_release,
2603 .unlocked_ioctl = kvm_vcpu_ioctl,
2604 .mmap = kvm_vcpu_mmap,
2605 .llseek = noop_llseek,
2606 KVM_COMPAT(kvm_vcpu_compat_ioctl),
2610 * Allocates an inode for the vcpu.
2612 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2614 char name[8 + 1 + ITOA_MAX_LEN + 1];
2616 snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id);
2617 return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
2620 static void kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
2622 #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS
2623 char dir_name[ITOA_MAX_LEN * 2];
2625 if (!debugfs_initialized())
2628 snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
2629 vcpu->debugfs_dentry = debugfs_create_dir(dir_name,
2630 vcpu->kvm->debugfs_dentry);
2632 kvm_arch_create_vcpu_debugfs(vcpu);
2637 * Creates some virtual cpus. Good luck creating more than one.
2639 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
2642 struct kvm_vcpu *vcpu;
2644 if (id >= KVM_MAX_VCPU_ID)
2647 mutex_lock(&kvm->lock);
2648 if (kvm->created_vcpus == KVM_MAX_VCPUS) {
2649 mutex_unlock(&kvm->lock);
2653 kvm->created_vcpus++;
2654 mutex_unlock(&kvm->lock);
2656 vcpu = kvm_arch_vcpu_create(kvm, id);
2659 goto vcpu_decrement;
2662 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2664 r = kvm_arch_vcpu_setup(vcpu);
2668 kvm_create_vcpu_debugfs(vcpu);
2670 mutex_lock(&kvm->lock);
2671 if (kvm_get_vcpu_by_id(kvm, id)) {
2673 goto unlock_vcpu_destroy;
2676 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
2678 /* Now it's all set up, let userspace reach it */
2680 r = create_vcpu_fd(vcpu);
2683 goto unlock_vcpu_destroy;
2686 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
2689 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
2690 * before kvm->online_vcpu's incremented value.
2693 atomic_inc(&kvm->online_vcpus);
2695 mutex_unlock(&kvm->lock);
2696 kvm_arch_vcpu_postcreate(vcpu);
2699 unlock_vcpu_destroy:
2700 mutex_unlock(&kvm->lock);
2701 debugfs_remove_recursive(vcpu->debugfs_dentry);
2703 kvm_arch_vcpu_destroy(vcpu);
2705 mutex_lock(&kvm->lock);
2706 kvm->created_vcpus--;
2707 mutex_unlock(&kvm->lock);
2711 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2714 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2715 vcpu->sigset_active = 1;
2716 vcpu->sigset = *sigset;
2718 vcpu->sigset_active = 0;
2722 static long kvm_vcpu_ioctl(struct file *filp,
2723 unsigned int ioctl, unsigned long arg)
2725 struct kvm_vcpu *vcpu = filp->private_data;
2726 void __user *argp = (void __user *)arg;
2728 struct kvm_fpu *fpu = NULL;
2729 struct kvm_sregs *kvm_sregs = NULL;
2731 if (vcpu->kvm->mm != current->mm)
2734 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2738 * Some architectures have vcpu ioctls that are asynchronous to vcpu
2739 * execution; mutex_lock() would break them.
2741 r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg);
2742 if (r != -ENOIOCTLCMD)
2745 if (mutex_lock_killable(&vcpu->mutex))
2753 oldpid = rcu_access_pointer(vcpu->pid);
2754 if (unlikely(oldpid != task_pid(current))) {
2755 /* The thread running this VCPU changed. */
2758 r = kvm_arch_vcpu_run_pid_change(vcpu);
2762 newpid = get_task_pid(current, PIDTYPE_PID);
2763 rcu_assign_pointer(vcpu->pid, newpid);
2768 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2769 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2772 case KVM_GET_REGS: {
2773 struct kvm_regs *kvm_regs;
2776 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL_ACCOUNT);
2779 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2783 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2790 case KVM_SET_REGS: {
2791 struct kvm_regs *kvm_regs;
2794 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2795 if (IS_ERR(kvm_regs)) {
2796 r = PTR_ERR(kvm_regs);
2799 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2803 case KVM_GET_SREGS: {
2804 kvm_sregs = kzalloc(sizeof(struct kvm_sregs),
2805 GFP_KERNEL_ACCOUNT);
2809 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2813 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2818 case KVM_SET_SREGS: {
2819 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2820 if (IS_ERR(kvm_sregs)) {
2821 r = PTR_ERR(kvm_sregs);
2825 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2828 case KVM_GET_MP_STATE: {
2829 struct kvm_mp_state mp_state;
2831 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2835 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
2840 case KVM_SET_MP_STATE: {
2841 struct kvm_mp_state mp_state;
2844 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
2846 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2849 case KVM_TRANSLATE: {
2850 struct kvm_translation tr;
2853 if (copy_from_user(&tr, argp, sizeof(tr)))
2855 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2859 if (copy_to_user(argp, &tr, sizeof(tr)))
2864 case KVM_SET_GUEST_DEBUG: {
2865 struct kvm_guest_debug dbg;
2868 if (copy_from_user(&dbg, argp, sizeof(dbg)))
2870 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2873 case KVM_SET_SIGNAL_MASK: {
2874 struct kvm_signal_mask __user *sigmask_arg = argp;
2875 struct kvm_signal_mask kvm_sigmask;
2876 sigset_t sigset, *p;
2881 if (copy_from_user(&kvm_sigmask, argp,
2882 sizeof(kvm_sigmask)))
2885 if (kvm_sigmask.len != sizeof(sigset))
2888 if (copy_from_user(&sigset, sigmask_arg->sigset,
2893 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2897 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL_ACCOUNT);
2901 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2905 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2911 fpu = memdup_user(argp, sizeof(*fpu));
2917 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2921 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2924 mutex_unlock(&vcpu->mutex);
2930 #ifdef CONFIG_KVM_COMPAT
2931 static long kvm_vcpu_compat_ioctl(struct file *filp,
2932 unsigned int ioctl, unsigned long arg)
2934 struct kvm_vcpu *vcpu = filp->private_data;
2935 void __user *argp = compat_ptr(arg);
2938 if (vcpu->kvm->mm != current->mm)
2942 case KVM_SET_SIGNAL_MASK: {
2943 struct kvm_signal_mask __user *sigmask_arg = argp;
2944 struct kvm_signal_mask kvm_sigmask;
2949 if (copy_from_user(&kvm_sigmask, argp,
2950 sizeof(kvm_sigmask)))
2953 if (kvm_sigmask.len != sizeof(compat_sigset_t))
2956 if (get_compat_sigset(&sigset, (void *)sigmask_arg->sigset))
2958 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2960 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2964 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2972 static int kvm_device_mmap(struct file *filp, struct vm_area_struct *vma)
2974 struct kvm_device *dev = filp->private_data;
2977 return dev->ops->mmap(dev, vma);
2982 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2983 int (*accessor)(struct kvm_device *dev,
2984 struct kvm_device_attr *attr),
2987 struct kvm_device_attr attr;
2992 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2995 return accessor(dev, &attr);
2998 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
3001 struct kvm_device *dev = filp->private_data;
3003 if (dev->kvm->mm != current->mm)
3007 case KVM_SET_DEVICE_ATTR:
3008 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
3009 case KVM_GET_DEVICE_ATTR:
3010 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
3011 case KVM_HAS_DEVICE_ATTR:
3012 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
3014 if (dev->ops->ioctl)
3015 return dev->ops->ioctl(dev, ioctl, arg);
3021 static int kvm_device_release(struct inode *inode, struct file *filp)
3023 struct kvm_device *dev = filp->private_data;
3024 struct kvm *kvm = dev->kvm;
3026 if (dev->ops->release) {
3027 mutex_lock(&kvm->lock);
3028 list_del(&dev->vm_node);
3029 dev->ops->release(dev);
3030 mutex_unlock(&kvm->lock);
3037 static const struct file_operations kvm_device_fops = {
3038 .unlocked_ioctl = kvm_device_ioctl,
3039 .release = kvm_device_release,
3040 KVM_COMPAT(kvm_device_ioctl),
3041 .mmap = kvm_device_mmap,
3044 struct kvm_device *kvm_device_from_filp(struct file *filp)
3046 if (filp->f_op != &kvm_device_fops)
3049 return filp->private_data;
3052 static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
3053 #ifdef CONFIG_KVM_MPIC
3054 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
3055 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
3059 int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
3061 if (type >= ARRAY_SIZE(kvm_device_ops_table))
3064 if (kvm_device_ops_table[type] != NULL)
3067 kvm_device_ops_table[type] = ops;
3071 void kvm_unregister_device_ops(u32 type)
3073 if (kvm_device_ops_table[type] != NULL)
3074 kvm_device_ops_table[type] = NULL;
3077 static int kvm_ioctl_create_device(struct kvm *kvm,
3078 struct kvm_create_device *cd)
3080 struct kvm_device_ops *ops = NULL;
3081 struct kvm_device *dev;
3082 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
3086 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
3089 type = array_index_nospec(cd->type, ARRAY_SIZE(kvm_device_ops_table));
3090 ops = kvm_device_ops_table[type];
3097 dev = kzalloc(sizeof(*dev), GFP_KERNEL_ACCOUNT);
3104 mutex_lock(&kvm->lock);
3105 ret = ops->create(dev, type);
3107 mutex_unlock(&kvm->lock);
3111 list_add(&dev->vm_node, &kvm->devices);
3112 mutex_unlock(&kvm->lock);
3118 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
3121 mutex_lock(&kvm->lock);
3122 list_del(&dev->vm_node);
3123 mutex_unlock(&kvm->lock);
3132 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
3135 case KVM_CAP_USER_MEMORY:
3136 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
3137 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
3138 case KVM_CAP_INTERNAL_ERROR_DATA:
3139 #ifdef CONFIG_HAVE_KVM_MSI
3140 case KVM_CAP_SIGNAL_MSI:
3142 #ifdef CONFIG_HAVE_KVM_IRQFD
3144 case KVM_CAP_IRQFD_RESAMPLE:
3146 case KVM_CAP_IOEVENTFD_ANY_LENGTH:
3147 case KVM_CAP_CHECK_EXTENSION_VM:
3148 case KVM_CAP_ENABLE_CAP_VM:
3149 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3150 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2:
3153 #ifdef CONFIG_KVM_MMIO
3154 case KVM_CAP_COALESCED_MMIO:
3155 return KVM_COALESCED_MMIO_PAGE_OFFSET;
3156 case KVM_CAP_COALESCED_PIO:
3159 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3160 case KVM_CAP_IRQ_ROUTING:
3161 return KVM_MAX_IRQ_ROUTES;
3163 #if KVM_ADDRESS_SPACE_NUM > 1
3164 case KVM_CAP_MULTI_ADDRESS_SPACE:
3165 return KVM_ADDRESS_SPACE_NUM;
3167 case KVM_CAP_NR_MEMSLOTS:
3168 return KVM_USER_MEM_SLOTS;
3172 return kvm_vm_ioctl_check_extension(kvm, arg);
3175 int __attribute__((weak)) kvm_vm_ioctl_enable_cap(struct kvm *kvm,
3176 struct kvm_enable_cap *cap)
3181 static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm,
3182 struct kvm_enable_cap *cap)
3185 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3186 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2:
3187 if (cap->flags || (cap->args[0] & ~1))
3189 kvm->manual_dirty_log_protect = cap->args[0];
3193 return kvm_vm_ioctl_enable_cap(kvm, cap);
3197 static long kvm_vm_ioctl(struct file *filp,
3198 unsigned int ioctl, unsigned long arg)
3200 struct kvm *kvm = filp->private_data;
3201 void __user *argp = (void __user *)arg;
3204 if (kvm->mm != current->mm)
3207 case KVM_CREATE_VCPU:
3208 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
3210 case KVM_ENABLE_CAP: {
3211 struct kvm_enable_cap cap;
3214 if (copy_from_user(&cap, argp, sizeof(cap)))
3216 r = kvm_vm_ioctl_enable_cap_generic(kvm, &cap);
3219 case KVM_SET_USER_MEMORY_REGION: {
3220 struct kvm_userspace_memory_region kvm_userspace_mem;
3223 if (copy_from_user(&kvm_userspace_mem, argp,
3224 sizeof(kvm_userspace_mem)))
3227 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
3230 case KVM_GET_DIRTY_LOG: {
3231 struct kvm_dirty_log log;
3234 if (copy_from_user(&log, argp, sizeof(log)))
3236 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3239 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3240 case KVM_CLEAR_DIRTY_LOG: {
3241 struct kvm_clear_dirty_log log;
3244 if (copy_from_user(&log, argp, sizeof(log)))
3246 r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
3250 #ifdef CONFIG_KVM_MMIO
3251 case KVM_REGISTER_COALESCED_MMIO: {
3252 struct kvm_coalesced_mmio_zone zone;
3255 if (copy_from_user(&zone, argp, sizeof(zone)))
3257 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
3260 case KVM_UNREGISTER_COALESCED_MMIO: {
3261 struct kvm_coalesced_mmio_zone zone;
3264 if (copy_from_user(&zone, argp, sizeof(zone)))
3266 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
3271 struct kvm_irqfd data;
3274 if (copy_from_user(&data, argp, sizeof(data)))
3276 r = kvm_irqfd(kvm, &data);
3279 case KVM_IOEVENTFD: {
3280 struct kvm_ioeventfd data;
3283 if (copy_from_user(&data, argp, sizeof(data)))
3285 r = kvm_ioeventfd(kvm, &data);
3288 #ifdef CONFIG_HAVE_KVM_MSI
3289 case KVM_SIGNAL_MSI: {
3293 if (copy_from_user(&msi, argp, sizeof(msi)))
3295 r = kvm_send_userspace_msi(kvm, &msi);
3299 #ifdef __KVM_HAVE_IRQ_LINE
3300 case KVM_IRQ_LINE_STATUS:
3301 case KVM_IRQ_LINE: {
3302 struct kvm_irq_level irq_event;
3305 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
3308 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
3309 ioctl == KVM_IRQ_LINE_STATUS);
3314 if (ioctl == KVM_IRQ_LINE_STATUS) {
3315 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
3323 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3324 case KVM_SET_GSI_ROUTING: {
3325 struct kvm_irq_routing routing;
3326 struct kvm_irq_routing __user *urouting;
3327 struct kvm_irq_routing_entry *entries = NULL;
3330 if (copy_from_user(&routing, argp, sizeof(routing)))
3333 if (!kvm_arch_can_set_irq_routing(kvm))
3335 if (routing.nr > KVM_MAX_IRQ_ROUTES)
3341 entries = vmalloc(array_size(sizeof(*entries),
3347 if (copy_from_user(entries, urouting->entries,
3348 routing.nr * sizeof(*entries)))
3349 goto out_free_irq_routing;
3351 r = kvm_set_irq_routing(kvm, entries, routing.nr,
3353 out_free_irq_routing:
3357 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
3358 case KVM_CREATE_DEVICE: {
3359 struct kvm_create_device cd;
3362 if (copy_from_user(&cd, argp, sizeof(cd)))
3365 r = kvm_ioctl_create_device(kvm, &cd);
3370 if (copy_to_user(argp, &cd, sizeof(cd)))
3376 case KVM_CHECK_EXTENSION:
3377 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
3380 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
3386 #ifdef CONFIG_KVM_COMPAT
3387 struct compat_kvm_dirty_log {
3391 compat_uptr_t dirty_bitmap; /* one bit per page */
3396 static long kvm_vm_compat_ioctl(struct file *filp,
3397 unsigned int ioctl, unsigned long arg)
3399 struct kvm *kvm = filp->private_data;
3402 if (kvm->mm != current->mm)
3405 case KVM_GET_DIRTY_LOG: {
3406 struct compat_kvm_dirty_log compat_log;
3407 struct kvm_dirty_log log;
3409 if (copy_from_user(&compat_log, (void __user *)arg,
3410 sizeof(compat_log)))
3412 log.slot = compat_log.slot;
3413 log.padding1 = compat_log.padding1;
3414 log.padding2 = compat_log.padding2;
3415 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
3417 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3421 r = kvm_vm_ioctl(filp, ioctl, arg);
3427 static struct file_operations kvm_vm_fops = {
3428 .release = kvm_vm_release,
3429 .unlocked_ioctl = kvm_vm_ioctl,
3430 .llseek = noop_llseek,
3431 KVM_COMPAT(kvm_vm_compat_ioctl),
3434 static int kvm_dev_ioctl_create_vm(unsigned long type)
3440 kvm = kvm_create_vm(type);
3442 return PTR_ERR(kvm);
3443 #ifdef CONFIG_KVM_MMIO
3444 r = kvm_coalesced_mmio_init(kvm);
3448 r = get_unused_fd_flags(O_CLOEXEC);
3452 file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
3460 * Don't call kvm_put_kvm anymore at this point; file->f_op is
3461 * already set, with ->release() being kvm_vm_release(). In error
3462 * cases it will be called by the final fput(file) and will take
3463 * care of doing kvm_put_kvm(kvm).
3465 if (kvm_create_vm_debugfs(kvm, r) < 0) {
3470 kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm);
3472 fd_install(r, file);
3480 static long kvm_dev_ioctl(struct file *filp,
3481 unsigned int ioctl, unsigned long arg)
3486 case KVM_GET_API_VERSION:
3489 r = KVM_API_VERSION;
3492 r = kvm_dev_ioctl_create_vm(arg);
3494 case KVM_CHECK_EXTENSION:
3495 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
3497 case KVM_GET_VCPU_MMAP_SIZE:
3500 r = PAGE_SIZE; /* struct kvm_run */
3502 r += PAGE_SIZE; /* pio data page */
3504 #ifdef CONFIG_KVM_MMIO
3505 r += PAGE_SIZE; /* coalesced mmio ring page */
3508 case KVM_TRACE_ENABLE:
3509 case KVM_TRACE_PAUSE:
3510 case KVM_TRACE_DISABLE:
3514 return kvm_arch_dev_ioctl(filp, ioctl, arg);
3520 static struct file_operations kvm_chardev_ops = {
3521 .unlocked_ioctl = kvm_dev_ioctl,
3522 .llseek = noop_llseek,
3523 KVM_COMPAT(kvm_dev_ioctl),
3526 static struct miscdevice kvm_dev = {
3532 static void hardware_enable_nolock(void *junk)
3534 int cpu = raw_smp_processor_id();
3537 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
3540 cpumask_set_cpu(cpu, cpus_hardware_enabled);
3542 r = kvm_arch_hardware_enable();
3545 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3546 atomic_inc(&hardware_enable_failed);
3547 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
3551 static int kvm_starting_cpu(unsigned int cpu)
3553 raw_spin_lock(&kvm_count_lock);
3554 if (kvm_usage_count)
3555 hardware_enable_nolock(NULL);
3556 raw_spin_unlock(&kvm_count_lock);
3560 static void hardware_disable_nolock(void *junk)
3562 int cpu = raw_smp_processor_id();
3564 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
3566 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3567 kvm_arch_hardware_disable();
3570 static int kvm_dying_cpu(unsigned int cpu)
3572 raw_spin_lock(&kvm_count_lock);
3573 if (kvm_usage_count)
3574 hardware_disable_nolock(NULL);
3575 raw_spin_unlock(&kvm_count_lock);
3579 static void hardware_disable_all_nolock(void)
3581 BUG_ON(!kvm_usage_count);
3584 if (!kvm_usage_count)
3585 on_each_cpu(hardware_disable_nolock, NULL, 1);
3588 static void hardware_disable_all(void)
3590 raw_spin_lock(&kvm_count_lock);
3591 hardware_disable_all_nolock();
3592 raw_spin_unlock(&kvm_count_lock);
3595 static int hardware_enable_all(void)
3599 raw_spin_lock(&kvm_count_lock);
3602 if (kvm_usage_count == 1) {
3603 atomic_set(&hardware_enable_failed, 0);
3604 on_each_cpu(hardware_enable_nolock, NULL, 1);
3606 if (atomic_read(&hardware_enable_failed)) {
3607 hardware_disable_all_nolock();
3612 raw_spin_unlock(&kvm_count_lock);
3617 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3621 * Some (well, at least mine) BIOSes hang on reboot if
3624 * And Intel TXT required VMX off for all cpu when system shutdown.
3626 pr_info("kvm: exiting hardware virtualization\n");
3627 kvm_rebooting = true;
3628 on_each_cpu(hardware_disable_nolock, NULL, 1);
3632 static struct notifier_block kvm_reboot_notifier = {
3633 .notifier_call = kvm_reboot,
3637 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3641 for (i = 0; i < bus->dev_count; i++) {
3642 struct kvm_io_device *pos = bus->range[i].dev;
3644 kvm_iodevice_destructor(pos);
3649 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
3650 const struct kvm_io_range *r2)
3652 gpa_t addr1 = r1->addr;
3653 gpa_t addr2 = r2->addr;
3658 /* If r2->len == 0, match the exact address. If r2->len != 0,
3659 * accept any overlapping write. Any order is acceptable for
3660 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
3661 * we process all of them.
3674 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
3676 return kvm_io_bus_cmp(p1, p2);
3679 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
3680 gpa_t addr, int len)
3682 struct kvm_io_range *range, key;
3685 key = (struct kvm_io_range) {
3690 range = bsearch(&key, bus->range, bus->dev_count,
3691 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
3695 off = range - bus->range;
3697 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
3703 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3704 struct kvm_io_range *range, const void *val)
3708 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3712 while (idx < bus->dev_count &&
3713 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3714 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
3723 /* kvm_io_bus_write - called under kvm->slots_lock */
3724 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3725 int len, const void *val)
3727 struct kvm_io_bus *bus;
3728 struct kvm_io_range range;
3731 range = (struct kvm_io_range) {
3736 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3739 r = __kvm_io_bus_write(vcpu, bus, &range, val);
3740 return r < 0 ? r : 0;
3742 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
3744 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3745 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
3746 gpa_t addr, int len, const void *val, long cookie)
3748 struct kvm_io_bus *bus;
3749 struct kvm_io_range range;
3751 range = (struct kvm_io_range) {
3756 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3760 /* First try the device referenced by cookie. */
3761 if ((cookie >= 0) && (cookie < bus->dev_count) &&
3762 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
3763 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
3768 * cookie contained garbage; fall back to search and return the
3769 * correct cookie value.
3771 return __kvm_io_bus_write(vcpu, bus, &range, val);
3774 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3775 struct kvm_io_range *range, void *val)
3779 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3783 while (idx < bus->dev_count &&
3784 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3785 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
3794 /* kvm_io_bus_read - called under kvm->slots_lock */
3795 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3798 struct kvm_io_bus *bus;
3799 struct kvm_io_range range;
3802 range = (struct kvm_io_range) {
3807 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3810 r = __kvm_io_bus_read(vcpu, bus, &range, val);
3811 return r < 0 ? r : 0;
3814 /* Caller must hold slots_lock. */
3815 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3816 int len, struct kvm_io_device *dev)
3819 struct kvm_io_bus *new_bus, *bus;
3820 struct kvm_io_range range;
3822 bus = kvm_get_bus(kvm, bus_idx);
3826 /* exclude ioeventfd which is limited by maximum fd */
3827 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3830 new_bus = kmalloc(struct_size(bus, range, bus->dev_count + 1),
3831 GFP_KERNEL_ACCOUNT);
3835 range = (struct kvm_io_range) {
3841 for (i = 0; i < bus->dev_count; i++)
3842 if (kvm_io_bus_cmp(&bus->range[i], &range) > 0)
3845 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3846 new_bus->dev_count++;
3847 new_bus->range[i] = range;
3848 memcpy(new_bus->range + i + 1, bus->range + i,
3849 (bus->dev_count - i) * sizeof(struct kvm_io_range));
3850 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3851 synchronize_srcu_expedited(&kvm->srcu);
3857 /* Caller must hold slots_lock. */
3858 void kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3859 struct kvm_io_device *dev)
3862 struct kvm_io_bus *new_bus, *bus;
3864 bus = kvm_get_bus(kvm, bus_idx);
3868 for (i = 0; i < bus->dev_count; i++)
3869 if (bus->range[i].dev == dev) {
3873 if (i == bus->dev_count)
3876 new_bus = kmalloc(struct_size(bus, range, bus->dev_count - 1),
3877 GFP_KERNEL_ACCOUNT);
3879 pr_err("kvm: failed to shrink bus, removing it completely\n");
3883 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3884 new_bus->dev_count--;
3885 memcpy(new_bus->range + i, bus->range + i + 1,
3886 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3889 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3890 synchronize_srcu_expedited(&kvm->srcu);
3895 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3898 struct kvm_io_bus *bus;
3899 int dev_idx, srcu_idx;
3900 struct kvm_io_device *iodev = NULL;
3902 srcu_idx = srcu_read_lock(&kvm->srcu);
3904 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
3908 dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
3912 iodev = bus->range[dev_idx].dev;
3915 srcu_read_unlock(&kvm->srcu, srcu_idx);
3919 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
3921 static int kvm_debugfs_open(struct inode *inode, struct file *file,
3922 int (*get)(void *, u64 *), int (*set)(void *, u64),
3925 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
3928 /* The debugfs files are a reference to the kvm struct which
3929 * is still valid when kvm_destroy_vm is called.
3930 * To avoid the race between open and the removal of the debugfs
3931 * directory we test against the users count.
3933 if (!refcount_inc_not_zero(&stat_data->kvm->users_count))
3936 if (simple_attr_open(inode, file, get,
3937 stat_data->mode & S_IWUGO ? set : NULL,
3939 kvm_put_kvm(stat_data->kvm);
3946 static int kvm_debugfs_release(struct inode *inode, struct file *file)
3948 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
3951 simple_attr_release(inode, file);
3952 kvm_put_kvm(stat_data->kvm);
3957 static int vm_stat_get_per_vm(void *data, u64 *val)
3959 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3961 *val = *(ulong *)((void *)stat_data->kvm + stat_data->offset);
3966 static int vm_stat_clear_per_vm(void *data, u64 val)
3968 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3973 *(ulong *)((void *)stat_data->kvm + stat_data->offset) = 0;
3978 static int vm_stat_get_per_vm_open(struct inode *inode, struct file *file)
3980 __simple_attr_check_format("%llu\n", 0ull);
3981 return kvm_debugfs_open(inode, file, vm_stat_get_per_vm,
3982 vm_stat_clear_per_vm, "%llu\n");
3985 static const struct file_operations vm_stat_get_per_vm_fops = {
3986 .owner = THIS_MODULE,
3987 .open = vm_stat_get_per_vm_open,
3988 .release = kvm_debugfs_release,
3989 .read = simple_attr_read,
3990 .write = simple_attr_write,
3991 .llseek = no_llseek,
3994 static int vcpu_stat_get_per_vm(void *data, u64 *val)
3997 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3998 struct kvm_vcpu *vcpu;
4002 kvm_for_each_vcpu(i, vcpu, stat_data->kvm)
4003 *val += *(u64 *)((void *)vcpu + stat_data->offset);
4008 static int vcpu_stat_clear_per_vm(void *data, u64 val)
4011 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
4012 struct kvm_vcpu *vcpu;
4017 kvm_for_each_vcpu(i, vcpu, stat_data->kvm)
4018 *(u64 *)((void *)vcpu + stat_data->offset) = 0;
4023 static int vcpu_stat_get_per_vm_open(struct inode *inode, struct file *file)
4025 __simple_attr_check_format("%llu\n", 0ull);
4026 return kvm_debugfs_open(inode, file, vcpu_stat_get_per_vm,
4027 vcpu_stat_clear_per_vm, "%llu\n");
4030 static const struct file_operations vcpu_stat_get_per_vm_fops = {
4031 .owner = THIS_MODULE,
4032 .open = vcpu_stat_get_per_vm_open,
4033 .release = kvm_debugfs_release,
4034 .read = simple_attr_read,
4035 .write = simple_attr_write,
4036 .llseek = no_llseek,
4039 static const struct file_operations *stat_fops_per_vm[] = {
4040 [KVM_STAT_VCPU] = &vcpu_stat_get_per_vm_fops,
4041 [KVM_STAT_VM] = &vm_stat_get_per_vm_fops,
4044 static int vm_stat_get(void *_offset, u64 *val)
4046 unsigned offset = (long)_offset;
4048 struct kvm_stat_data stat_tmp = {.offset = offset};
4052 mutex_lock(&kvm_lock);
4053 list_for_each_entry(kvm, &vm_list, vm_list) {
4055 vm_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
4058 mutex_unlock(&kvm_lock);
4062 static int vm_stat_clear(void *_offset, u64 val)
4064 unsigned offset = (long)_offset;
4066 struct kvm_stat_data stat_tmp = {.offset = offset};
4071 mutex_lock(&kvm_lock);
4072 list_for_each_entry(kvm, &vm_list, vm_list) {
4074 vm_stat_clear_per_vm((void *)&stat_tmp, 0);
4076 mutex_unlock(&kvm_lock);
4081 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n");
4083 static int vcpu_stat_get(void *_offset, u64 *val)
4085 unsigned offset = (long)_offset;
4087 struct kvm_stat_data stat_tmp = {.offset = offset};
4091 mutex_lock(&kvm_lock);
4092 list_for_each_entry(kvm, &vm_list, vm_list) {
4094 vcpu_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
4097 mutex_unlock(&kvm_lock);
4101 static int vcpu_stat_clear(void *_offset, u64 val)
4103 unsigned offset = (long)_offset;
4105 struct kvm_stat_data stat_tmp = {.offset = offset};
4110 mutex_lock(&kvm_lock);
4111 list_for_each_entry(kvm, &vm_list, vm_list) {
4113 vcpu_stat_clear_per_vm((void *)&stat_tmp, 0);
4115 mutex_unlock(&kvm_lock);
4120 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear,
4123 static const struct file_operations *stat_fops[] = {
4124 [KVM_STAT_VCPU] = &vcpu_stat_fops,
4125 [KVM_STAT_VM] = &vm_stat_fops,
4128 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm)
4130 struct kobj_uevent_env *env;
4131 unsigned long long created, active;
4133 if (!kvm_dev.this_device || !kvm)
4136 mutex_lock(&kvm_lock);
4137 if (type == KVM_EVENT_CREATE_VM) {
4138 kvm_createvm_count++;
4140 } else if (type == KVM_EVENT_DESTROY_VM) {
4143 created = kvm_createvm_count;
4144 active = kvm_active_vms;
4145 mutex_unlock(&kvm_lock);
4147 env = kzalloc(sizeof(*env), GFP_KERNEL_ACCOUNT);
4151 add_uevent_var(env, "CREATED=%llu", created);
4152 add_uevent_var(env, "COUNT=%llu", active);
4154 if (type == KVM_EVENT_CREATE_VM) {
4155 add_uevent_var(env, "EVENT=create");
4156 kvm->userspace_pid = task_pid_nr(current);
4157 } else if (type == KVM_EVENT_DESTROY_VM) {
4158 add_uevent_var(env, "EVENT=destroy");
4160 add_uevent_var(env, "PID=%d", kvm->userspace_pid);
4162 if (!IS_ERR_OR_NULL(kvm->debugfs_dentry)) {
4163 char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL_ACCOUNT);
4166 tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX);
4168 add_uevent_var(env, "STATS_PATH=%s", tmp);
4172 /* no need for checks, since we are adding at most only 5 keys */
4173 env->envp[env->envp_idx++] = NULL;
4174 kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp);
4178 static void kvm_init_debug(void)
4180 struct kvm_stats_debugfs_item *p;
4182 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
4184 kvm_debugfs_num_entries = 0;
4185 for (p = debugfs_entries; p->name; ++p, kvm_debugfs_num_entries++) {
4186 int mode = p->mode ? p->mode : 0644;
4187 debugfs_create_file(p->name, mode, kvm_debugfs_dir,
4188 (void *)(long)p->offset,
4189 stat_fops[p->kind]);
4193 static int kvm_suspend(void)
4195 if (kvm_usage_count)
4196 hardware_disable_nolock(NULL);
4200 static void kvm_resume(void)
4202 if (kvm_usage_count) {
4203 #ifdef CONFIG_LOCKDEP
4204 WARN_ON(lockdep_is_held(&kvm_count_lock));
4206 hardware_enable_nolock(NULL);
4210 static struct syscore_ops kvm_syscore_ops = {
4211 .suspend = kvm_suspend,
4212 .resume = kvm_resume,
4216 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
4218 return container_of(pn, struct kvm_vcpu, preempt_notifier);
4221 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
4223 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
4225 WRITE_ONCE(vcpu->preempted, false);
4226 WRITE_ONCE(vcpu->ready, false);
4228 kvm_arch_sched_in(vcpu, cpu);
4230 kvm_arch_vcpu_load(vcpu, cpu);
4233 static void kvm_sched_out(struct preempt_notifier *pn,
4234 struct task_struct *next)
4236 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
4238 if (current->state == TASK_RUNNING) {
4239 WRITE_ONCE(vcpu->preempted, true);
4240 WRITE_ONCE(vcpu->ready, true);
4242 kvm_arch_vcpu_put(vcpu);
4245 static void check_processor_compat(void *rtn)
4247 *(int *)rtn = kvm_arch_check_processor_compat();
4250 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
4251 struct module *module)
4256 r = kvm_arch_init(opaque);
4261 * kvm_arch_init makes sure there's at most one caller
4262 * for architectures that support multiple implementations,
4263 * like intel and amd on x86.
4264 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
4265 * conflicts in case kvm is already setup for another implementation.
4267 r = kvm_irqfd_init();
4271 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
4276 r = kvm_arch_hardware_setup();
4280 for_each_online_cpu(cpu) {
4281 smp_call_function_single(cpu, check_processor_compat, &r, 1);
4286 r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting",
4287 kvm_starting_cpu, kvm_dying_cpu);
4290 register_reboot_notifier(&kvm_reboot_notifier);
4292 /* A kmem cache lets us meet the alignment requirements of fx_save. */
4294 vcpu_align = __alignof__(struct kvm_vcpu);
4296 kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align,
4298 offsetof(struct kvm_vcpu, arch),
4299 sizeof_field(struct kvm_vcpu, arch),
4301 if (!kvm_vcpu_cache) {
4306 r = kvm_async_pf_init();
4310 kvm_chardev_ops.owner = module;
4311 kvm_vm_fops.owner = module;
4312 kvm_vcpu_fops.owner = module;
4314 r = misc_register(&kvm_dev);
4316 pr_err("kvm: misc device register failed\n");
4320 register_syscore_ops(&kvm_syscore_ops);
4322 kvm_preempt_ops.sched_in = kvm_sched_in;
4323 kvm_preempt_ops.sched_out = kvm_sched_out;
4327 r = kvm_vfio_ops_init();
4333 kvm_async_pf_deinit();
4335 kmem_cache_destroy(kvm_vcpu_cache);
4337 unregister_reboot_notifier(&kvm_reboot_notifier);
4338 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4341 kvm_arch_hardware_unsetup();
4343 free_cpumask_var(cpus_hardware_enabled);
4351 EXPORT_SYMBOL_GPL(kvm_init);
4355 debugfs_remove_recursive(kvm_debugfs_dir);
4356 misc_deregister(&kvm_dev);
4357 kmem_cache_destroy(kvm_vcpu_cache);
4358 kvm_async_pf_deinit();
4359 unregister_syscore_ops(&kvm_syscore_ops);
4360 unregister_reboot_notifier(&kvm_reboot_notifier);
4361 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4362 on_each_cpu(hardware_disable_nolock, NULL, 1);
4363 kvm_arch_hardware_unsetup();
4366 free_cpumask_var(cpus_hardware_enabled);
4367 kvm_vfio_ops_exit();
4369 EXPORT_SYMBOL_GPL(kvm_exit);