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 kvm->debugfs_stat_data[p - debugfs_entries] = stat_data;
621 debugfs_create_file(p->name, 0644, kvm->debugfs_dentry,
622 stat_data, stat_fops_per_vm[p->kind]);
627 static struct kvm *kvm_create_vm(unsigned long type)
630 struct kvm *kvm = kvm_arch_alloc_vm();
633 return ERR_PTR(-ENOMEM);
635 spin_lock_init(&kvm->mmu_lock);
637 kvm->mm = current->mm;
638 kvm_eventfd_init(kvm);
639 mutex_init(&kvm->lock);
640 mutex_init(&kvm->irq_lock);
641 mutex_init(&kvm->slots_lock);
642 refcount_set(&kvm->users_count, 1);
643 INIT_LIST_HEAD(&kvm->devices);
645 r = kvm_arch_init_vm(kvm, type);
647 goto out_err_no_disable;
649 r = hardware_enable_all();
651 goto out_err_no_disable;
653 #ifdef CONFIG_HAVE_KVM_IRQFD
654 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
657 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
660 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
661 struct kvm_memslots *slots = kvm_alloc_memslots();
663 goto out_err_no_srcu;
664 /* Generations must be different for each address space. */
665 slots->generation = i;
666 rcu_assign_pointer(kvm->memslots[i], slots);
669 if (init_srcu_struct(&kvm->srcu))
670 goto out_err_no_srcu;
671 if (init_srcu_struct(&kvm->irq_srcu))
672 goto out_err_no_irq_srcu;
673 for (i = 0; i < KVM_NR_BUSES; i++) {
674 rcu_assign_pointer(kvm->buses[i],
675 kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL_ACCOUNT));
680 r = kvm_init_mmu_notifier(kvm);
684 mutex_lock(&kvm_lock);
685 list_add(&kvm->vm_list, &vm_list);
686 mutex_unlock(&kvm_lock);
688 preempt_notifier_inc();
693 cleanup_srcu_struct(&kvm->irq_srcu);
695 cleanup_srcu_struct(&kvm->srcu);
697 hardware_disable_all();
699 refcount_set(&kvm->users_count, 0);
700 for (i = 0; i < KVM_NR_BUSES; i++)
701 kfree(kvm_get_bus(kvm, i));
702 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
703 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
704 kvm_arch_free_vm(kvm);
709 static void kvm_destroy_devices(struct kvm *kvm)
711 struct kvm_device *dev, *tmp;
714 * We do not need to take the kvm->lock here, because nobody else
715 * has a reference to the struct kvm at this point and therefore
716 * cannot access the devices list anyhow.
718 list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
719 list_del(&dev->vm_node);
720 dev->ops->destroy(dev);
724 static void kvm_destroy_vm(struct kvm *kvm)
727 struct mm_struct *mm = kvm->mm;
729 kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm);
730 kvm_destroy_vm_debugfs(kvm);
731 kvm_arch_sync_events(kvm);
732 mutex_lock(&kvm_lock);
733 list_del(&kvm->vm_list);
734 mutex_unlock(&kvm_lock);
735 kvm_free_irq_routing(kvm);
736 for (i = 0; i < KVM_NR_BUSES; i++) {
737 struct kvm_io_bus *bus = kvm_get_bus(kvm, i);
740 kvm_io_bus_destroy(bus);
741 kvm->buses[i] = NULL;
743 kvm_coalesced_mmio_free(kvm);
744 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
745 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
747 kvm_arch_flush_shadow_all(kvm);
749 kvm_arch_destroy_vm(kvm);
750 kvm_destroy_devices(kvm);
751 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
752 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
753 cleanup_srcu_struct(&kvm->irq_srcu);
754 cleanup_srcu_struct(&kvm->srcu);
755 kvm_arch_free_vm(kvm);
756 preempt_notifier_dec();
757 hardware_disable_all();
761 void kvm_get_kvm(struct kvm *kvm)
763 refcount_inc(&kvm->users_count);
765 EXPORT_SYMBOL_GPL(kvm_get_kvm);
767 void kvm_put_kvm(struct kvm *kvm)
769 if (refcount_dec_and_test(&kvm->users_count))
772 EXPORT_SYMBOL_GPL(kvm_put_kvm);
775 static int kvm_vm_release(struct inode *inode, struct file *filp)
777 struct kvm *kvm = filp->private_data;
779 kvm_irqfd_release(kvm);
786 * Allocation size is twice as large as the actual dirty bitmap size.
787 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
789 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
791 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
793 memslot->dirty_bitmap = kvzalloc(dirty_bytes, GFP_KERNEL_ACCOUNT);
794 if (!memslot->dirty_bitmap)
801 * Insert memslot and re-sort memslots based on their GFN,
802 * so binary search could be used to lookup GFN.
803 * Sorting algorithm takes advantage of having initially
804 * sorted array and known changed memslot position.
806 static void update_memslots(struct kvm_memslots *slots,
807 struct kvm_memory_slot *new,
808 enum kvm_mr_change change)
811 int i = slots->id_to_index[id];
812 struct kvm_memory_slot *mslots = slots->memslots;
814 WARN_ON(mslots[i].id != id);
818 WARN_ON(mslots[i].npages || !new->npages);
822 WARN_ON(new->npages || !mslots[i].npages);
828 while (i < KVM_MEM_SLOTS_NUM - 1 &&
829 new->base_gfn <= mslots[i + 1].base_gfn) {
830 if (!mslots[i + 1].npages)
832 mslots[i] = mslots[i + 1];
833 slots->id_to_index[mslots[i].id] = i;
838 * The ">=" is needed when creating a slot with base_gfn == 0,
839 * so that it moves before all those with base_gfn == npages == 0.
841 * On the other hand, if new->npages is zero, the above loop has
842 * already left i pointing to the beginning of the empty part of
843 * mslots, and the ">=" would move the hole backwards in this
844 * case---which is wrong. So skip the loop when deleting a slot.
848 new->base_gfn >= mslots[i - 1].base_gfn) {
849 mslots[i] = mslots[i - 1];
850 slots->id_to_index[mslots[i].id] = i;
854 WARN_ON_ONCE(i != slots->used_slots);
857 slots->id_to_index[mslots[i].id] = i;
860 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
862 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
864 #ifdef __KVM_HAVE_READONLY_MEM
865 valid_flags |= KVM_MEM_READONLY;
868 if (mem->flags & ~valid_flags)
874 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
875 int as_id, struct kvm_memslots *slots)
877 struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id);
878 u64 gen = old_memslots->generation;
880 WARN_ON(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS);
881 slots->generation = gen | KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
883 rcu_assign_pointer(kvm->memslots[as_id], slots);
884 synchronize_srcu_expedited(&kvm->srcu);
887 * Increment the new memslot generation a second time, dropping the
888 * update in-progress flag and incrementing then generation based on
889 * the number of address spaces. This provides a unique and easily
890 * identifiable generation number while the memslots are in flux.
892 gen = slots->generation & ~KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
895 * Generations must be unique even across address spaces. We do not need
896 * a global counter for that, instead the generation space is evenly split
897 * across address spaces. For example, with two address spaces, address
898 * space 0 will use generations 0, 2, 4, ... while address space 1 will
899 * use generations 1, 3, 5, ...
901 gen += KVM_ADDRESS_SPACE_NUM;
903 kvm_arch_memslots_updated(kvm, gen);
905 slots->generation = gen;
911 * Allocate some memory and give it an address in the guest physical address
914 * Discontiguous memory is allowed, mostly for framebuffers.
916 * Must be called holding kvm->slots_lock for write.
918 int __kvm_set_memory_region(struct kvm *kvm,
919 const struct kvm_userspace_memory_region *mem)
923 unsigned long npages;
924 struct kvm_memory_slot *slot;
925 struct kvm_memory_slot old, new;
926 struct kvm_memslots *slots = NULL, *old_memslots;
928 enum kvm_mr_change change;
930 r = check_memory_region_flags(mem);
935 as_id = mem->slot >> 16;
938 /* General sanity checks */
939 if (mem->memory_size & (PAGE_SIZE - 1))
941 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
943 /* We can read the guest memory with __xxx_user() later on. */
944 if ((id < KVM_USER_MEM_SLOTS) &&
945 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
946 !access_ok((void __user *)(unsigned long)mem->userspace_addr,
949 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
951 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
954 slot = id_to_memslot(__kvm_memslots(kvm, as_id), id);
955 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
956 npages = mem->memory_size >> PAGE_SHIFT;
958 if (npages > KVM_MEM_MAX_NR_PAGES)
964 new.base_gfn = base_gfn;
966 new.flags = mem->flags;
970 change = KVM_MR_CREATE;
971 else { /* Modify an existing slot. */
972 if ((mem->userspace_addr != old.userspace_addr) ||
973 (npages != old.npages) ||
974 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
977 if (base_gfn != old.base_gfn)
978 change = KVM_MR_MOVE;
979 else if (new.flags != old.flags)
980 change = KVM_MR_FLAGS_ONLY;
981 else { /* Nothing to change. */
990 change = KVM_MR_DELETE;
995 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
996 /* Check for overlaps */
998 kvm_for_each_memslot(slot, __kvm_memslots(kvm, as_id)) {
1001 if (!((base_gfn + npages <= slot->base_gfn) ||
1002 (base_gfn >= slot->base_gfn + slot->npages)))
1007 /* Free page dirty bitmap if unneeded */
1008 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
1009 new.dirty_bitmap = NULL;
1012 if (change == KVM_MR_CREATE) {
1013 new.userspace_addr = mem->userspace_addr;
1015 if (kvm_arch_create_memslot(kvm, &new, npages))
1019 /* Allocate page dirty bitmap if needed */
1020 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
1021 if (kvm_create_dirty_bitmap(&new) < 0)
1025 slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL_ACCOUNT);
1028 memcpy(slots, __kvm_memslots(kvm, as_id), sizeof(struct kvm_memslots));
1030 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
1031 slot = id_to_memslot(slots, id);
1032 slot->flags |= KVM_MEMSLOT_INVALID;
1034 old_memslots = install_new_memslots(kvm, as_id, slots);
1036 /* From this point no new shadow pages pointing to a deleted,
1037 * or moved, memslot will be created.
1039 * validation of sp->gfn happens in:
1040 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1041 * - kvm_is_visible_gfn (mmu_check_roots)
1043 kvm_arch_flush_shadow_memslot(kvm, slot);
1046 * We can re-use the old_memslots from above, the only difference
1047 * from the currently installed memslots is the invalid flag. This
1048 * will get overwritten by update_memslots anyway.
1050 slots = old_memslots;
1053 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
1057 /* actual memory is freed via old in kvm_free_memslot below */
1058 if (change == KVM_MR_DELETE) {
1059 new.dirty_bitmap = NULL;
1060 memset(&new.arch, 0, sizeof(new.arch));
1063 update_memslots(slots, &new, change);
1064 old_memslots = install_new_memslots(kvm, as_id, slots);
1066 kvm_arch_commit_memory_region(kvm, mem, &old, &new, change);
1068 kvm_free_memslot(kvm, &old, &new);
1069 kvfree(old_memslots);
1075 kvm_free_memslot(kvm, &new, &old);
1079 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1081 int kvm_set_memory_region(struct kvm *kvm,
1082 const struct kvm_userspace_memory_region *mem)
1086 mutex_lock(&kvm->slots_lock);
1087 r = __kvm_set_memory_region(kvm, mem);
1088 mutex_unlock(&kvm->slots_lock);
1091 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1093 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1094 struct kvm_userspace_memory_region *mem)
1096 if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1099 return kvm_set_memory_region(kvm, mem);
1102 int kvm_get_dirty_log(struct kvm *kvm,
1103 struct kvm_dirty_log *log, int *is_dirty)
1105 struct kvm_memslots *slots;
1106 struct kvm_memory_slot *memslot;
1109 unsigned long any = 0;
1111 as_id = log->slot >> 16;
1112 id = (u16)log->slot;
1113 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1116 slots = __kvm_memslots(kvm, as_id);
1117 memslot = id_to_memslot(slots, id);
1118 if (!memslot->dirty_bitmap)
1121 n = kvm_dirty_bitmap_bytes(memslot);
1123 for (i = 0; !any && i < n/sizeof(long); ++i)
1124 any = memslot->dirty_bitmap[i];
1126 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1133 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1135 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1137 * kvm_get_dirty_log_protect - get a snapshot of dirty pages
1138 * and reenable dirty page tracking for the corresponding pages.
1139 * @kvm: pointer to kvm instance
1140 * @log: slot id and address to which we copy the log
1141 * @flush: true if TLB flush is needed by caller
1143 * We need to keep it in mind that VCPU threads can write to the bitmap
1144 * concurrently. So, to avoid losing track of dirty pages we keep the
1147 * 1. Take a snapshot of the bit and clear it if needed.
1148 * 2. Write protect the corresponding page.
1149 * 3. Copy the snapshot to the userspace.
1150 * 4. Upon return caller flushes TLB's if needed.
1152 * Between 2 and 4, the guest may write to the page using the remaining TLB
1153 * entry. This is not a problem because the page is reported dirty using
1154 * the snapshot taken before and step 4 ensures that writes done after
1155 * exiting to userspace will be logged for the next call.
1158 int kvm_get_dirty_log_protect(struct kvm *kvm,
1159 struct kvm_dirty_log *log, bool *flush)
1161 struct kvm_memslots *slots;
1162 struct kvm_memory_slot *memslot;
1165 unsigned long *dirty_bitmap;
1166 unsigned long *dirty_bitmap_buffer;
1168 as_id = log->slot >> 16;
1169 id = (u16)log->slot;
1170 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1173 slots = __kvm_memslots(kvm, as_id);
1174 memslot = id_to_memslot(slots, id);
1176 dirty_bitmap = memslot->dirty_bitmap;
1180 n = kvm_dirty_bitmap_bytes(memslot);
1182 if (kvm->manual_dirty_log_protect) {
1184 * Unlike kvm_get_dirty_log, we always return false in *flush,
1185 * because no flush is needed until KVM_CLEAR_DIRTY_LOG. There
1186 * is some code duplication between this function and
1187 * kvm_get_dirty_log, but hopefully all architecture
1188 * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log
1189 * can be eliminated.
1191 dirty_bitmap_buffer = dirty_bitmap;
1193 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
1194 memset(dirty_bitmap_buffer, 0, n);
1196 spin_lock(&kvm->mmu_lock);
1197 for (i = 0; i < n / sizeof(long); i++) {
1201 if (!dirty_bitmap[i])
1205 mask = xchg(&dirty_bitmap[i], 0);
1206 dirty_bitmap_buffer[i] = mask;
1208 offset = i * BITS_PER_LONG;
1209 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1212 spin_unlock(&kvm->mmu_lock);
1215 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1219 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect);
1222 * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap
1223 * and reenable dirty page tracking for the corresponding pages.
1224 * @kvm: pointer to kvm instance
1225 * @log: slot id and address from which to fetch the bitmap of dirty pages
1226 * @flush: true if TLB flush is needed by caller
1228 int kvm_clear_dirty_log_protect(struct kvm *kvm,
1229 struct kvm_clear_dirty_log *log, bool *flush)
1231 struct kvm_memslots *slots;
1232 struct kvm_memory_slot *memslot;
1236 unsigned long *dirty_bitmap;
1237 unsigned long *dirty_bitmap_buffer;
1239 as_id = log->slot >> 16;
1240 id = (u16)log->slot;
1241 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1244 if (log->first_page & 63)
1247 slots = __kvm_memslots(kvm, as_id);
1248 memslot = id_to_memslot(slots, id);
1250 dirty_bitmap = memslot->dirty_bitmap;
1254 n = ALIGN(log->num_pages, BITS_PER_LONG) / 8;
1256 if (log->first_page > memslot->npages ||
1257 log->num_pages > memslot->npages - log->first_page ||
1258 (log->num_pages < memslot->npages - log->first_page && (log->num_pages & 63)))
1262 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
1263 if (copy_from_user(dirty_bitmap_buffer, log->dirty_bitmap, n))
1266 spin_lock(&kvm->mmu_lock);
1267 for (offset = log->first_page, i = offset / BITS_PER_LONG,
1268 n = DIV_ROUND_UP(log->num_pages, BITS_PER_LONG); n--;
1269 i++, offset += BITS_PER_LONG) {
1270 unsigned long mask = *dirty_bitmap_buffer++;
1271 atomic_long_t *p = (atomic_long_t *) &dirty_bitmap[i];
1275 mask &= atomic_long_fetch_andnot(mask, p);
1278 * mask contains the bits that really have been cleared. This
1279 * never includes any bits beyond the length of the memslot (if
1280 * the length is not aligned to 64 pages), therefore it is not
1281 * a problem if userspace sets them in log->dirty_bitmap.
1285 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1289 spin_unlock(&kvm->mmu_lock);
1293 EXPORT_SYMBOL_GPL(kvm_clear_dirty_log_protect);
1296 bool kvm_largepages_enabled(void)
1298 return largepages_enabled;
1301 void kvm_disable_largepages(void)
1303 largepages_enabled = false;
1305 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1307 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1309 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1311 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1313 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
1315 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn);
1318 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1320 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1322 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1323 memslot->flags & KVM_MEMSLOT_INVALID)
1328 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1330 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1332 struct vm_area_struct *vma;
1333 unsigned long addr, size;
1337 addr = gfn_to_hva(kvm, gfn);
1338 if (kvm_is_error_hva(addr))
1341 down_read(¤t->mm->mmap_sem);
1342 vma = find_vma(current->mm, addr);
1346 size = vma_kernel_pagesize(vma);
1349 up_read(¤t->mm->mmap_sem);
1354 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1356 return slot->flags & KVM_MEM_READONLY;
1359 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1360 gfn_t *nr_pages, bool write)
1362 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1363 return KVM_HVA_ERR_BAD;
1365 if (memslot_is_readonly(slot) && write)
1366 return KVM_HVA_ERR_RO_BAD;
1369 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1371 return __gfn_to_hva_memslot(slot, gfn);
1374 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1377 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1380 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1383 return gfn_to_hva_many(slot, gfn, NULL);
1385 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1387 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1389 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1391 EXPORT_SYMBOL_GPL(gfn_to_hva);
1393 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
1395 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
1397 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
1400 * Return the hva of a @gfn and the R/W attribute if possible.
1402 * @slot: the kvm_memory_slot which contains @gfn
1403 * @gfn: the gfn to be translated
1404 * @writable: used to return the read/write attribute of the @slot if the hva
1405 * is valid and @writable is not NULL
1407 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1408 gfn_t gfn, bool *writable)
1410 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1412 if (!kvm_is_error_hva(hva) && writable)
1413 *writable = !memslot_is_readonly(slot);
1418 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1420 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1422 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1425 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
1427 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1429 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1432 static inline int check_user_page_hwpoison(unsigned long addr)
1434 int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
1436 rc = get_user_pages(addr, 1, flags, NULL, NULL);
1437 return rc == -EHWPOISON;
1441 * The fast path to get the writable pfn which will be stored in @pfn,
1442 * true indicates success, otherwise false is returned. It's also the
1443 * only part that runs if we can are in atomic context.
1445 static bool hva_to_pfn_fast(unsigned long addr, bool write_fault,
1446 bool *writable, kvm_pfn_t *pfn)
1448 struct page *page[1];
1452 * Fast pin a writable pfn only if it is a write fault request
1453 * or the caller allows to map a writable pfn for a read fault
1456 if (!(write_fault || writable))
1459 npages = __get_user_pages_fast(addr, 1, 1, page);
1461 *pfn = page_to_pfn(page[0]);
1472 * The slow path to get the pfn of the specified host virtual address,
1473 * 1 indicates success, -errno is returned if error is detected.
1475 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1476 bool *writable, kvm_pfn_t *pfn)
1478 unsigned int flags = FOLL_HWPOISON;
1485 *writable = write_fault;
1488 flags |= FOLL_WRITE;
1490 flags |= FOLL_NOWAIT;
1492 npages = get_user_pages_unlocked(addr, 1, &page, flags);
1496 /* map read fault as writable if possible */
1497 if (unlikely(!write_fault) && writable) {
1500 if (__get_user_pages_fast(addr, 1, 1, &wpage) == 1) {
1506 *pfn = page_to_pfn(page);
1510 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1512 if (unlikely(!(vma->vm_flags & VM_READ)))
1515 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1521 static int hva_to_pfn_remapped(struct vm_area_struct *vma,
1522 unsigned long addr, bool *async,
1523 bool write_fault, bool *writable,
1529 r = follow_pfn(vma, addr, &pfn);
1532 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
1533 * not call the fault handler, so do it here.
1535 bool unlocked = false;
1536 r = fixup_user_fault(current, current->mm, addr,
1537 (write_fault ? FAULT_FLAG_WRITE : 0),
1544 r = follow_pfn(vma, addr, &pfn);
1554 * Get a reference here because callers of *hva_to_pfn* and
1555 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
1556 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
1557 * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will
1558 * simply do nothing for reserved pfns.
1560 * Whoever called remap_pfn_range is also going to call e.g.
1561 * unmap_mapping_range before the underlying pages are freed,
1562 * causing a call to our MMU notifier.
1571 * Pin guest page in memory and return its pfn.
1572 * @addr: host virtual address which maps memory to the guest
1573 * @atomic: whether this function can sleep
1574 * @async: whether this function need to wait IO complete if the
1575 * host page is not in the memory
1576 * @write_fault: whether we should get a writable host page
1577 * @writable: whether it allows to map a writable host page for !@write_fault
1579 * The function will map a writable host page for these two cases:
1580 * 1): @write_fault = true
1581 * 2): @write_fault = false && @writable, @writable will tell the caller
1582 * whether the mapping is writable.
1584 static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1585 bool write_fault, bool *writable)
1587 struct vm_area_struct *vma;
1591 /* we can do it either atomically or asynchronously, not both */
1592 BUG_ON(atomic && async);
1594 if (hva_to_pfn_fast(addr, write_fault, writable, &pfn))
1598 return KVM_PFN_ERR_FAULT;
1600 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1604 down_read(¤t->mm->mmap_sem);
1605 if (npages == -EHWPOISON ||
1606 (!async && check_user_page_hwpoison(addr))) {
1607 pfn = KVM_PFN_ERR_HWPOISON;
1612 vma = find_vma_intersection(current->mm, addr, addr + 1);
1615 pfn = KVM_PFN_ERR_FAULT;
1616 else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
1617 r = hva_to_pfn_remapped(vma, addr, async, write_fault, writable, &pfn);
1621 pfn = KVM_PFN_ERR_FAULT;
1623 if (async && vma_is_valid(vma, write_fault))
1625 pfn = KVM_PFN_ERR_FAULT;
1628 up_read(¤t->mm->mmap_sem);
1632 kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn,
1633 bool atomic, bool *async, bool write_fault,
1636 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1638 if (addr == KVM_HVA_ERR_RO_BAD) {
1641 return KVM_PFN_ERR_RO_FAULT;
1644 if (kvm_is_error_hva(addr)) {
1647 return KVM_PFN_NOSLOT;
1650 /* Do not map writable pfn in the readonly memslot. */
1651 if (writable && memslot_is_readonly(slot)) {
1656 return hva_to_pfn(addr, atomic, async, write_fault,
1659 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
1661 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1664 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
1665 write_fault, writable);
1667 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1669 kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1671 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1673 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
1675 kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1677 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1679 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1681 kvm_pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1683 return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm, gfn), gfn);
1685 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1687 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
1689 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1691 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
1693 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1695 return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
1697 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1699 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
1701 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1703 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
1705 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1706 struct page **pages, int nr_pages)
1711 addr = gfn_to_hva_many(slot, gfn, &entry);
1712 if (kvm_is_error_hva(addr))
1715 if (entry < nr_pages)
1718 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1720 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1722 static struct page *kvm_pfn_to_page(kvm_pfn_t pfn)
1724 if (is_error_noslot_pfn(pfn))
1725 return KVM_ERR_PTR_BAD_PAGE;
1727 if (kvm_is_reserved_pfn(pfn)) {
1729 return KVM_ERR_PTR_BAD_PAGE;
1732 return pfn_to_page(pfn);
1735 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1739 pfn = gfn_to_pfn(kvm, gfn);
1741 return kvm_pfn_to_page(pfn);
1743 EXPORT_SYMBOL_GPL(gfn_to_page);
1745 static int __kvm_map_gfn(struct kvm_memory_slot *slot, gfn_t gfn,
1746 struct kvm_host_map *map)
1750 struct page *page = KVM_UNMAPPED_PAGE;
1755 pfn = gfn_to_pfn_memslot(slot, gfn);
1756 if (is_error_noslot_pfn(pfn))
1759 if (pfn_valid(pfn)) {
1760 page = pfn_to_page(pfn);
1762 #ifdef CONFIG_HAS_IOMEM
1764 hva = memremap(pfn_to_hpa(pfn), PAGE_SIZE, MEMREMAP_WB);
1779 int kvm_vcpu_map(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map)
1781 return __kvm_map_gfn(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, map);
1783 EXPORT_SYMBOL_GPL(kvm_vcpu_map);
1785 void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map,
1794 if (map->page != KVM_UNMAPPED_PAGE)
1796 #ifdef CONFIG_HAS_IOMEM
1802 kvm_vcpu_mark_page_dirty(vcpu, map->gfn);
1803 kvm_release_pfn_dirty(map->pfn);
1805 kvm_release_pfn_clean(map->pfn);
1811 EXPORT_SYMBOL_GPL(kvm_vcpu_unmap);
1813 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
1817 pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
1819 return kvm_pfn_to_page(pfn);
1821 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
1823 void kvm_release_page_clean(struct page *page)
1825 WARN_ON(is_error_page(page));
1827 kvm_release_pfn_clean(page_to_pfn(page));
1829 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1831 void kvm_release_pfn_clean(kvm_pfn_t pfn)
1833 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
1834 put_page(pfn_to_page(pfn));
1836 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1838 void kvm_release_page_dirty(struct page *page)
1840 WARN_ON(is_error_page(page));
1842 kvm_release_pfn_dirty(page_to_pfn(page));
1844 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1846 void kvm_release_pfn_dirty(kvm_pfn_t pfn)
1848 kvm_set_pfn_dirty(pfn);
1849 kvm_release_pfn_clean(pfn);
1851 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1853 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
1855 if (!kvm_is_reserved_pfn(pfn)) {
1856 struct page *page = pfn_to_page(pfn);
1858 if (!PageReserved(page))
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 start = cur = ktime_get();
2326 if (vcpu->halt_poll_ns && !kvm_arch_no_poll(vcpu)) {
2327 ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns);
2329 ++vcpu->stat.halt_attempted_poll;
2332 * This sets KVM_REQ_UNHALT if an interrupt
2335 if (kvm_vcpu_check_block(vcpu) < 0) {
2336 ++vcpu->stat.halt_successful_poll;
2337 if (!vcpu_valid_wakeup(vcpu))
2338 ++vcpu->stat.halt_poll_invalid;
2342 } while (single_task_running() && ktime_before(cur, stop));
2345 kvm_arch_vcpu_blocking(vcpu);
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);
2362 block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
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;
2379 trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu));
2380 kvm_arch_vcpu_block_finish(vcpu);
2382 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
2384 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
2386 struct swait_queue_head *wqp;
2388 wqp = kvm_arch_vcpu_wq(vcpu);
2389 if (swq_has_sleeper(wqp)) {
2391 WRITE_ONCE(vcpu->ready, true);
2392 ++vcpu->stat.halt_wakeup;
2398 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
2402 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2404 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
2407 int cpu = vcpu->cpu;
2409 if (kvm_vcpu_wake_up(vcpu))
2413 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
2414 if (kvm_arch_vcpu_should_kick(vcpu))
2415 smp_send_reschedule(cpu);
2418 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
2419 #endif /* !CONFIG_S390 */
2421 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
2424 struct task_struct *task = NULL;
2428 pid = rcu_dereference(target->pid);
2430 task = get_pid_task(pid, PIDTYPE_PID);
2434 ret = yield_to(task, 1);
2435 put_task_struct(task);
2439 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
2442 * Helper that checks whether a VCPU is eligible for directed yield.
2443 * Most eligible candidate to yield is decided by following heuristics:
2445 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2446 * (preempted lock holder), indicated by @in_spin_loop.
2447 * Set at the beiginning and cleared at the end of interception/PLE handler.
2449 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2450 * chance last time (mostly it has become eligible now since we have probably
2451 * yielded to lockholder in last iteration. This is done by toggling
2452 * @dy_eligible each time a VCPU checked for eligibility.)
2454 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2455 * to preempted lock-holder could result in wrong VCPU selection and CPU
2456 * burning. Giving priority for a potential lock-holder increases lock
2459 * Since algorithm is based on heuristics, accessing another VCPU data without
2460 * locking does not harm. It may result in trying to yield to same VCPU, fail
2461 * and continue with next VCPU and so on.
2463 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
2465 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2468 eligible = !vcpu->spin_loop.in_spin_loop ||
2469 vcpu->spin_loop.dy_eligible;
2471 if (vcpu->spin_loop.in_spin_loop)
2472 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
2480 void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode)
2482 struct kvm *kvm = me->kvm;
2483 struct kvm_vcpu *vcpu;
2484 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
2490 kvm_vcpu_set_in_spin_loop(me, true);
2492 * We boost the priority of a VCPU that is runnable but not
2493 * currently running, because it got preempted by something
2494 * else and called schedule in __vcpu_run. Hopefully that
2495 * VCPU is holding the lock that we need and will release it.
2496 * We approximate round-robin by starting at the last boosted VCPU.
2498 for (pass = 0; pass < 2 && !yielded && try; pass++) {
2499 kvm_for_each_vcpu(i, vcpu, kvm) {
2500 if (!pass && i <= last_boosted_vcpu) {
2501 i = last_boosted_vcpu;
2503 } else if (pass && i > last_boosted_vcpu)
2505 if (!READ_ONCE(vcpu->ready))
2509 if (swait_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
2511 if (yield_to_kernel_mode && !kvm_arch_vcpu_in_kernel(vcpu))
2513 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
2516 yielded = kvm_vcpu_yield_to(vcpu);
2518 kvm->last_boosted_vcpu = i;
2520 } else if (yielded < 0) {
2527 kvm_vcpu_set_in_spin_loop(me, false);
2529 /* Ensure vcpu is not eligible during next spinloop */
2530 kvm_vcpu_set_dy_eligible(me, false);
2532 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
2534 static vm_fault_t kvm_vcpu_fault(struct vm_fault *vmf)
2536 struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data;
2539 if (vmf->pgoff == 0)
2540 page = virt_to_page(vcpu->run);
2542 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
2543 page = virt_to_page(vcpu->arch.pio_data);
2545 #ifdef CONFIG_KVM_MMIO
2546 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
2547 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
2550 return kvm_arch_vcpu_fault(vcpu, vmf);
2556 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
2557 .fault = kvm_vcpu_fault,
2560 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2562 vma->vm_ops = &kvm_vcpu_vm_ops;
2566 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2568 struct kvm_vcpu *vcpu = filp->private_data;
2570 debugfs_remove_recursive(vcpu->debugfs_dentry);
2571 kvm_put_kvm(vcpu->kvm);
2575 static struct file_operations kvm_vcpu_fops = {
2576 .release = kvm_vcpu_release,
2577 .unlocked_ioctl = kvm_vcpu_ioctl,
2578 .mmap = kvm_vcpu_mmap,
2579 .llseek = noop_llseek,
2580 KVM_COMPAT(kvm_vcpu_compat_ioctl),
2584 * Allocates an inode for the vcpu.
2586 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2588 char name[8 + 1 + ITOA_MAX_LEN + 1];
2590 snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id);
2591 return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
2594 static int kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
2596 char dir_name[ITOA_MAX_LEN * 2];
2599 if (!kvm_arch_has_vcpu_debugfs())
2602 if (!debugfs_initialized())
2605 snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
2606 vcpu->debugfs_dentry = debugfs_create_dir(dir_name,
2607 vcpu->kvm->debugfs_dentry);
2608 if (!vcpu->debugfs_dentry)
2611 ret = kvm_arch_create_vcpu_debugfs(vcpu);
2613 debugfs_remove_recursive(vcpu->debugfs_dentry);
2621 * Creates some virtual cpus. Good luck creating more than one.
2623 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
2626 struct kvm_vcpu *vcpu;
2628 if (id >= KVM_MAX_VCPU_ID)
2631 mutex_lock(&kvm->lock);
2632 if (kvm->created_vcpus == KVM_MAX_VCPUS) {
2633 mutex_unlock(&kvm->lock);
2637 kvm->created_vcpus++;
2638 mutex_unlock(&kvm->lock);
2640 vcpu = kvm_arch_vcpu_create(kvm, id);
2643 goto vcpu_decrement;
2646 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2648 r = kvm_arch_vcpu_setup(vcpu);
2652 r = kvm_create_vcpu_debugfs(vcpu);
2656 mutex_lock(&kvm->lock);
2657 if (kvm_get_vcpu_by_id(kvm, id)) {
2659 goto unlock_vcpu_destroy;
2662 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
2664 /* Now it's all set up, let userspace reach it */
2666 r = create_vcpu_fd(vcpu);
2669 goto unlock_vcpu_destroy;
2672 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
2675 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
2676 * before kvm->online_vcpu's incremented value.
2679 atomic_inc(&kvm->online_vcpus);
2681 mutex_unlock(&kvm->lock);
2682 kvm_arch_vcpu_postcreate(vcpu);
2685 unlock_vcpu_destroy:
2686 mutex_unlock(&kvm->lock);
2687 debugfs_remove_recursive(vcpu->debugfs_dentry);
2689 kvm_arch_vcpu_destroy(vcpu);
2691 mutex_lock(&kvm->lock);
2692 kvm->created_vcpus--;
2693 mutex_unlock(&kvm->lock);
2697 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2700 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2701 vcpu->sigset_active = 1;
2702 vcpu->sigset = *sigset;
2704 vcpu->sigset_active = 0;
2708 static long kvm_vcpu_ioctl(struct file *filp,
2709 unsigned int ioctl, unsigned long arg)
2711 struct kvm_vcpu *vcpu = filp->private_data;
2712 void __user *argp = (void __user *)arg;
2714 struct kvm_fpu *fpu = NULL;
2715 struct kvm_sregs *kvm_sregs = NULL;
2717 if (vcpu->kvm->mm != current->mm)
2720 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2724 * Some architectures have vcpu ioctls that are asynchronous to vcpu
2725 * execution; mutex_lock() would break them.
2727 r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg);
2728 if (r != -ENOIOCTLCMD)
2731 if (mutex_lock_killable(&vcpu->mutex))
2739 oldpid = rcu_access_pointer(vcpu->pid);
2740 if (unlikely(oldpid != task_pid(current))) {
2741 /* The thread running this VCPU changed. */
2744 r = kvm_arch_vcpu_run_pid_change(vcpu);
2748 newpid = get_task_pid(current, PIDTYPE_PID);
2749 rcu_assign_pointer(vcpu->pid, newpid);
2754 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2755 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2758 case KVM_GET_REGS: {
2759 struct kvm_regs *kvm_regs;
2762 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL_ACCOUNT);
2765 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2769 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2776 case KVM_SET_REGS: {
2777 struct kvm_regs *kvm_regs;
2780 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2781 if (IS_ERR(kvm_regs)) {
2782 r = PTR_ERR(kvm_regs);
2785 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2789 case KVM_GET_SREGS: {
2790 kvm_sregs = kzalloc(sizeof(struct kvm_sregs),
2791 GFP_KERNEL_ACCOUNT);
2795 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2799 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2804 case KVM_SET_SREGS: {
2805 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2806 if (IS_ERR(kvm_sregs)) {
2807 r = PTR_ERR(kvm_sregs);
2811 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2814 case KVM_GET_MP_STATE: {
2815 struct kvm_mp_state mp_state;
2817 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2821 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
2826 case KVM_SET_MP_STATE: {
2827 struct kvm_mp_state mp_state;
2830 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
2832 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2835 case KVM_TRANSLATE: {
2836 struct kvm_translation tr;
2839 if (copy_from_user(&tr, argp, sizeof(tr)))
2841 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2845 if (copy_to_user(argp, &tr, sizeof(tr)))
2850 case KVM_SET_GUEST_DEBUG: {
2851 struct kvm_guest_debug dbg;
2854 if (copy_from_user(&dbg, argp, sizeof(dbg)))
2856 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2859 case KVM_SET_SIGNAL_MASK: {
2860 struct kvm_signal_mask __user *sigmask_arg = argp;
2861 struct kvm_signal_mask kvm_sigmask;
2862 sigset_t sigset, *p;
2867 if (copy_from_user(&kvm_sigmask, argp,
2868 sizeof(kvm_sigmask)))
2871 if (kvm_sigmask.len != sizeof(sigset))
2874 if (copy_from_user(&sigset, sigmask_arg->sigset,
2879 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2883 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL_ACCOUNT);
2887 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2891 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2897 fpu = memdup_user(argp, sizeof(*fpu));
2903 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2907 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2910 mutex_unlock(&vcpu->mutex);
2916 #ifdef CONFIG_KVM_COMPAT
2917 static long kvm_vcpu_compat_ioctl(struct file *filp,
2918 unsigned int ioctl, unsigned long arg)
2920 struct kvm_vcpu *vcpu = filp->private_data;
2921 void __user *argp = compat_ptr(arg);
2924 if (vcpu->kvm->mm != current->mm)
2928 case KVM_SET_SIGNAL_MASK: {
2929 struct kvm_signal_mask __user *sigmask_arg = argp;
2930 struct kvm_signal_mask kvm_sigmask;
2935 if (copy_from_user(&kvm_sigmask, argp,
2936 sizeof(kvm_sigmask)))
2939 if (kvm_sigmask.len != sizeof(compat_sigset_t))
2942 if (get_compat_sigset(&sigset, (void *)sigmask_arg->sigset))
2944 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2946 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2950 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2958 static int kvm_device_mmap(struct file *filp, struct vm_area_struct *vma)
2960 struct kvm_device *dev = filp->private_data;
2963 return dev->ops->mmap(dev, vma);
2968 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2969 int (*accessor)(struct kvm_device *dev,
2970 struct kvm_device_attr *attr),
2973 struct kvm_device_attr attr;
2978 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2981 return accessor(dev, &attr);
2984 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2987 struct kvm_device *dev = filp->private_data;
2989 if (dev->kvm->mm != current->mm)
2993 case KVM_SET_DEVICE_ATTR:
2994 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2995 case KVM_GET_DEVICE_ATTR:
2996 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2997 case KVM_HAS_DEVICE_ATTR:
2998 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
3000 if (dev->ops->ioctl)
3001 return dev->ops->ioctl(dev, ioctl, arg);
3007 static int kvm_device_release(struct inode *inode, struct file *filp)
3009 struct kvm_device *dev = filp->private_data;
3010 struct kvm *kvm = dev->kvm;
3012 if (dev->ops->release) {
3013 mutex_lock(&kvm->lock);
3014 list_del(&dev->vm_node);
3015 dev->ops->release(dev);
3016 mutex_unlock(&kvm->lock);
3023 static const struct file_operations kvm_device_fops = {
3024 .unlocked_ioctl = kvm_device_ioctl,
3025 .release = kvm_device_release,
3026 KVM_COMPAT(kvm_device_ioctl),
3027 .mmap = kvm_device_mmap,
3030 struct kvm_device *kvm_device_from_filp(struct file *filp)
3032 if (filp->f_op != &kvm_device_fops)
3035 return filp->private_data;
3038 static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
3039 #ifdef CONFIG_KVM_MPIC
3040 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
3041 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
3045 int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
3047 if (type >= ARRAY_SIZE(kvm_device_ops_table))
3050 if (kvm_device_ops_table[type] != NULL)
3053 kvm_device_ops_table[type] = ops;
3057 void kvm_unregister_device_ops(u32 type)
3059 if (kvm_device_ops_table[type] != NULL)
3060 kvm_device_ops_table[type] = NULL;
3063 static int kvm_ioctl_create_device(struct kvm *kvm,
3064 struct kvm_create_device *cd)
3066 struct kvm_device_ops *ops = NULL;
3067 struct kvm_device *dev;
3068 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
3072 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
3075 type = array_index_nospec(cd->type, ARRAY_SIZE(kvm_device_ops_table));
3076 ops = kvm_device_ops_table[type];
3083 dev = kzalloc(sizeof(*dev), GFP_KERNEL_ACCOUNT);
3090 mutex_lock(&kvm->lock);
3091 ret = ops->create(dev, type);
3093 mutex_unlock(&kvm->lock);
3097 list_add(&dev->vm_node, &kvm->devices);
3098 mutex_unlock(&kvm->lock);
3104 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
3107 mutex_lock(&kvm->lock);
3108 list_del(&dev->vm_node);
3109 mutex_unlock(&kvm->lock);
3118 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
3121 case KVM_CAP_USER_MEMORY:
3122 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
3123 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
3124 case KVM_CAP_INTERNAL_ERROR_DATA:
3125 #ifdef CONFIG_HAVE_KVM_MSI
3126 case KVM_CAP_SIGNAL_MSI:
3128 #ifdef CONFIG_HAVE_KVM_IRQFD
3130 case KVM_CAP_IRQFD_RESAMPLE:
3132 case KVM_CAP_IOEVENTFD_ANY_LENGTH:
3133 case KVM_CAP_CHECK_EXTENSION_VM:
3134 case KVM_CAP_ENABLE_CAP_VM:
3135 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3136 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2:
3139 #ifdef CONFIG_KVM_MMIO
3140 case KVM_CAP_COALESCED_MMIO:
3141 return KVM_COALESCED_MMIO_PAGE_OFFSET;
3142 case KVM_CAP_COALESCED_PIO:
3145 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3146 case KVM_CAP_IRQ_ROUTING:
3147 return KVM_MAX_IRQ_ROUTES;
3149 #if KVM_ADDRESS_SPACE_NUM > 1
3150 case KVM_CAP_MULTI_ADDRESS_SPACE:
3151 return KVM_ADDRESS_SPACE_NUM;
3153 case KVM_CAP_NR_MEMSLOTS:
3154 return KVM_USER_MEM_SLOTS;
3158 return kvm_vm_ioctl_check_extension(kvm, arg);
3161 int __attribute__((weak)) kvm_vm_ioctl_enable_cap(struct kvm *kvm,
3162 struct kvm_enable_cap *cap)
3167 static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm,
3168 struct kvm_enable_cap *cap)
3171 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3172 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2:
3173 if (cap->flags || (cap->args[0] & ~1))
3175 kvm->manual_dirty_log_protect = cap->args[0];
3179 return kvm_vm_ioctl_enable_cap(kvm, cap);
3183 static long kvm_vm_ioctl(struct file *filp,
3184 unsigned int ioctl, unsigned long arg)
3186 struct kvm *kvm = filp->private_data;
3187 void __user *argp = (void __user *)arg;
3190 if (kvm->mm != current->mm)
3193 case KVM_CREATE_VCPU:
3194 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
3196 case KVM_ENABLE_CAP: {
3197 struct kvm_enable_cap cap;
3200 if (copy_from_user(&cap, argp, sizeof(cap)))
3202 r = kvm_vm_ioctl_enable_cap_generic(kvm, &cap);
3205 case KVM_SET_USER_MEMORY_REGION: {
3206 struct kvm_userspace_memory_region kvm_userspace_mem;
3209 if (copy_from_user(&kvm_userspace_mem, argp,
3210 sizeof(kvm_userspace_mem)))
3213 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
3216 case KVM_GET_DIRTY_LOG: {
3217 struct kvm_dirty_log log;
3220 if (copy_from_user(&log, argp, sizeof(log)))
3222 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3225 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3226 case KVM_CLEAR_DIRTY_LOG: {
3227 struct kvm_clear_dirty_log log;
3230 if (copy_from_user(&log, argp, sizeof(log)))
3232 r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
3236 #ifdef CONFIG_KVM_MMIO
3237 case KVM_REGISTER_COALESCED_MMIO: {
3238 struct kvm_coalesced_mmio_zone zone;
3241 if (copy_from_user(&zone, argp, sizeof(zone)))
3243 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
3246 case KVM_UNREGISTER_COALESCED_MMIO: {
3247 struct kvm_coalesced_mmio_zone zone;
3250 if (copy_from_user(&zone, argp, sizeof(zone)))
3252 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
3257 struct kvm_irqfd data;
3260 if (copy_from_user(&data, argp, sizeof(data)))
3262 r = kvm_irqfd(kvm, &data);
3265 case KVM_IOEVENTFD: {
3266 struct kvm_ioeventfd data;
3269 if (copy_from_user(&data, argp, sizeof(data)))
3271 r = kvm_ioeventfd(kvm, &data);
3274 #ifdef CONFIG_HAVE_KVM_MSI
3275 case KVM_SIGNAL_MSI: {
3279 if (copy_from_user(&msi, argp, sizeof(msi)))
3281 r = kvm_send_userspace_msi(kvm, &msi);
3285 #ifdef __KVM_HAVE_IRQ_LINE
3286 case KVM_IRQ_LINE_STATUS:
3287 case KVM_IRQ_LINE: {
3288 struct kvm_irq_level irq_event;
3291 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
3294 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
3295 ioctl == KVM_IRQ_LINE_STATUS);
3300 if (ioctl == KVM_IRQ_LINE_STATUS) {
3301 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
3309 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3310 case KVM_SET_GSI_ROUTING: {
3311 struct kvm_irq_routing routing;
3312 struct kvm_irq_routing __user *urouting;
3313 struct kvm_irq_routing_entry *entries = NULL;
3316 if (copy_from_user(&routing, argp, sizeof(routing)))
3319 if (!kvm_arch_can_set_irq_routing(kvm))
3321 if (routing.nr > KVM_MAX_IRQ_ROUTES)
3327 entries = vmalloc(array_size(sizeof(*entries),
3333 if (copy_from_user(entries, urouting->entries,
3334 routing.nr * sizeof(*entries)))
3335 goto out_free_irq_routing;
3337 r = kvm_set_irq_routing(kvm, entries, routing.nr,
3339 out_free_irq_routing:
3343 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
3344 case KVM_CREATE_DEVICE: {
3345 struct kvm_create_device cd;
3348 if (copy_from_user(&cd, argp, sizeof(cd)))
3351 r = kvm_ioctl_create_device(kvm, &cd);
3356 if (copy_to_user(argp, &cd, sizeof(cd)))
3362 case KVM_CHECK_EXTENSION:
3363 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
3366 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
3372 #ifdef CONFIG_KVM_COMPAT
3373 struct compat_kvm_dirty_log {
3377 compat_uptr_t dirty_bitmap; /* one bit per page */
3382 static long kvm_vm_compat_ioctl(struct file *filp,
3383 unsigned int ioctl, unsigned long arg)
3385 struct kvm *kvm = filp->private_data;
3388 if (kvm->mm != current->mm)
3391 case KVM_GET_DIRTY_LOG: {
3392 struct compat_kvm_dirty_log compat_log;
3393 struct kvm_dirty_log log;
3395 if (copy_from_user(&compat_log, (void __user *)arg,
3396 sizeof(compat_log)))
3398 log.slot = compat_log.slot;
3399 log.padding1 = compat_log.padding1;
3400 log.padding2 = compat_log.padding2;
3401 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
3403 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3407 r = kvm_vm_ioctl(filp, ioctl, arg);
3413 static struct file_operations kvm_vm_fops = {
3414 .release = kvm_vm_release,
3415 .unlocked_ioctl = kvm_vm_ioctl,
3416 .llseek = noop_llseek,
3417 KVM_COMPAT(kvm_vm_compat_ioctl),
3420 static int kvm_dev_ioctl_create_vm(unsigned long type)
3426 kvm = kvm_create_vm(type);
3428 return PTR_ERR(kvm);
3429 #ifdef CONFIG_KVM_MMIO
3430 r = kvm_coalesced_mmio_init(kvm);
3434 r = get_unused_fd_flags(O_CLOEXEC);
3438 file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
3446 * Don't call kvm_put_kvm anymore at this point; file->f_op is
3447 * already set, with ->release() being kvm_vm_release(). In error
3448 * cases it will be called by the final fput(file) and will take
3449 * care of doing kvm_put_kvm(kvm).
3451 if (kvm_create_vm_debugfs(kvm, r) < 0) {
3456 kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm);
3458 fd_install(r, file);
3466 static long kvm_dev_ioctl(struct file *filp,
3467 unsigned int ioctl, unsigned long arg)
3472 case KVM_GET_API_VERSION:
3475 r = KVM_API_VERSION;
3478 r = kvm_dev_ioctl_create_vm(arg);
3480 case KVM_CHECK_EXTENSION:
3481 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
3483 case KVM_GET_VCPU_MMAP_SIZE:
3486 r = PAGE_SIZE; /* struct kvm_run */
3488 r += PAGE_SIZE; /* pio data page */
3490 #ifdef CONFIG_KVM_MMIO
3491 r += PAGE_SIZE; /* coalesced mmio ring page */
3494 case KVM_TRACE_ENABLE:
3495 case KVM_TRACE_PAUSE:
3496 case KVM_TRACE_DISABLE:
3500 return kvm_arch_dev_ioctl(filp, ioctl, arg);
3506 static struct file_operations kvm_chardev_ops = {
3507 .unlocked_ioctl = kvm_dev_ioctl,
3508 .llseek = noop_llseek,
3509 KVM_COMPAT(kvm_dev_ioctl),
3512 static struct miscdevice kvm_dev = {
3518 static void hardware_enable_nolock(void *junk)
3520 int cpu = raw_smp_processor_id();
3523 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
3526 cpumask_set_cpu(cpu, cpus_hardware_enabled);
3528 r = kvm_arch_hardware_enable();
3531 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3532 atomic_inc(&hardware_enable_failed);
3533 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
3537 static int kvm_starting_cpu(unsigned int cpu)
3539 raw_spin_lock(&kvm_count_lock);
3540 if (kvm_usage_count)
3541 hardware_enable_nolock(NULL);
3542 raw_spin_unlock(&kvm_count_lock);
3546 static void hardware_disable_nolock(void *junk)
3548 int cpu = raw_smp_processor_id();
3550 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
3552 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3553 kvm_arch_hardware_disable();
3556 static int kvm_dying_cpu(unsigned int cpu)
3558 raw_spin_lock(&kvm_count_lock);
3559 if (kvm_usage_count)
3560 hardware_disable_nolock(NULL);
3561 raw_spin_unlock(&kvm_count_lock);
3565 static void hardware_disable_all_nolock(void)
3567 BUG_ON(!kvm_usage_count);
3570 if (!kvm_usage_count)
3571 on_each_cpu(hardware_disable_nolock, NULL, 1);
3574 static void hardware_disable_all(void)
3576 raw_spin_lock(&kvm_count_lock);
3577 hardware_disable_all_nolock();
3578 raw_spin_unlock(&kvm_count_lock);
3581 static int hardware_enable_all(void)
3585 raw_spin_lock(&kvm_count_lock);
3588 if (kvm_usage_count == 1) {
3589 atomic_set(&hardware_enable_failed, 0);
3590 on_each_cpu(hardware_enable_nolock, NULL, 1);
3592 if (atomic_read(&hardware_enable_failed)) {
3593 hardware_disable_all_nolock();
3598 raw_spin_unlock(&kvm_count_lock);
3603 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3607 * Some (well, at least mine) BIOSes hang on reboot if
3610 * And Intel TXT required VMX off for all cpu when system shutdown.
3612 pr_info("kvm: exiting hardware virtualization\n");
3613 kvm_rebooting = true;
3614 on_each_cpu(hardware_disable_nolock, NULL, 1);
3618 static struct notifier_block kvm_reboot_notifier = {
3619 .notifier_call = kvm_reboot,
3623 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3627 for (i = 0; i < bus->dev_count; i++) {
3628 struct kvm_io_device *pos = bus->range[i].dev;
3630 kvm_iodevice_destructor(pos);
3635 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
3636 const struct kvm_io_range *r2)
3638 gpa_t addr1 = r1->addr;
3639 gpa_t addr2 = r2->addr;
3644 /* If r2->len == 0, match the exact address. If r2->len != 0,
3645 * accept any overlapping write. Any order is acceptable for
3646 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
3647 * we process all of them.
3660 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
3662 return kvm_io_bus_cmp(p1, p2);
3665 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
3666 gpa_t addr, int len)
3668 struct kvm_io_range *range, key;
3671 key = (struct kvm_io_range) {
3676 range = bsearch(&key, bus->range, bus->dev_count,
3677 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
3681 off = range - bus->range;
3683 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
3689 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3690 struct kvm_io_range *range, const void *val)
3694 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3698 while (idx < bus->dev_count &&
3699 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3700 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
3709 /* kvm_io_bus_write - called under kvm->slots_lock */
3710 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3711 int len, const void *val)
3713 struct kvm_io_bus *bus;
3714 struct kvm_io_range range;
3717 range = (struct kvm_io_range) {
3722 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3725 r = __kvm_io_bus_write(vcpu, bus, &range, val);
3726 return r < 0 ? r : 0;
3728 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
3730 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3731 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
3732 gpa_t addr, int len, const void *val, long cookie)
3734 struct kvm_io_bus *bus;
3735 struct kvm_io_range range;
3737 range = (struct kvm_io_range) {
3742 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3746 /* First try the device referenced by cookie. */
3747 if ((cookie >= 0) && (cookie < bus->dev_count) &&
3748 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
3749 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
3754 * cookie contained garbage; fall back to search and return the
3755 * correct cookie value.
3757 return __kvm_io_bus_write(vcpu, bus, &range, val);
3760 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3761 struct kvm_io_range *range, void *val)
3765 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3769 while (idx < bus->dev_count &&
3770 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3771 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
3780 /* kvm_io_bus_read - called under kvm->slots_lock */
3781 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3784 struct kvm_io_bus *bus;
3785 struct kvm_io_range range;
3788 range = (struct kvm_io_range) {
3793 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3796 r = __kvm_io_bus_read(vcpu, bus, &range, val);
3797 return r < 0 ? r : 0;
3800 /* Caller must hold slots_lock. */
3801 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3802 int len, struct kvm_io_device *dev)
3805 struct kvm_io_bus *new_bus, *bus;
3806 struct kvm_io_range range;
3808 bus = kvm_get_bus(kvm, bus_idx);
3812 /* exclude ioeventfd which is limited by maximum fd */
3813 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3816 new_bus = kmalloc(struct_size(bus, range, bus->dev_count + 1),
3817 GFP_KERNEL_ACCOUNT);
3821 range = (struct kvm_io_range) {
3827 for (i = 0; i < bus->dev_count; i++)
3828 if (kvm_io_bus_cmp(&bus->range[i], &range) > 0)
3831 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3832 new_bus->dev_count++;
3833 new_bus->range[i] = range;
3834 memcpy(new_bus->range + i + 1, bus->range + i,
3835 (bus->dev_count - i) * sizeof(struct kvm_io_range));
3836 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3837 synchronize_srcu_expedited(&kvm->srcu);
3843 /* Caller must hold slots_lock. */
3844 void kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3845 struct kvm_io_device *dev)
3848 struct kvm_io_bus *new_bus, *bus;
3850 bus = kvm_get_bus(kvm, bus_idx);
3854 for (i = 0; i < bus->dev_count; i++)
3855 if (bus->range[i].dev == dev) {
3859 if (i == bus->dev_count)
3862 new_bus = kmalloc(struct_size(bus, range, bus->dev_count - 1),
3863 GFP_KERNEL_ACCOUNT);
3865 pr_err("kvm: failed to shrink bus, removing it completely\n");
3869 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3870 new_bus->dev_count--;
3871 memcpy(new_bus->range + i, bus->range + i + 1,
3872 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3875 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3876 synchronize_srcu_expedited(&kvm->srcu);
3881 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3884 struct kvm_io_bus *bus;
3885 int dev_idx, srcu_idx;
3886 struct kvm_io_device *iodev = NULL;
3888 srcu_idx = srcu_read_lock(&kvm->srcu);
3890 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
3894 dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
3898 iodev = bus->range[dev_idx].dev;
3901 srcu_read_unlock(&kvm->srcu, srcu_idx);
3905 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
3907 static int kvm_debugfs_open(struct inode *inode, struct file *file,
3908 int (*get)(void *, u64 *), int (*set)(void *, u64),
3911 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
3914 /* The debugfs files are a reference to the kvm struct which
3915 * is still valid when kvm_destroy_vm is called.
3916 * To avoid the race between open and the removal of the debugfs
3917 * directory we test against the users count.
3919 if (!refcount_inc_not_zero(&stat_data->kvm->users_count))
3922 if (simple_attr_open(inode, file, get, set, fmt)) {
3923 kvm_put_kvm(stat_data->kvm);
3930 static int kvm_debugfs_release(struct inode *inode, struct file *file)
3932 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
3935 simple_attr_release(inode, file);
3936 kvm_put_kvm(stat_data->kvm);
3941 static int vm_stat_get_per_vm(void *data, u64 *val)
3943 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3945 *val = *(ulong *)((void *)stat_data->kvm + stat_data->offset);
3950 static int vm_stat_clear_per_vm(void *data, u64 val)
3952 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3957 *(ulong *)((void *)stat_data->kvm + stat_data->offset) = 0;
3962 static int vm_stat_get_per_vm_open(struct inode *inode, struct file *file)
3964 __simple_attr_check_format("%llu\n", 0ull);
3965 return kvm_debugfs_open(inode, file, vm_stat_get_per_vm,
3966 vm_stat_clear_per_vm, "%llu\n");
3969 static const struct file_operations vm_stat_get_per_vm_fops = {
3970 .owner = THIS_MODULE,
3971 .open = vm_stat_get_per_vm_open,
3972 .release = kvm_debugfs_release,
3973 .read = simple_attr_read,
3974 .write = simple_attr_write,
3975 .llseek = no_llseek,
3978 static int vcpu_stat_get_per_vm(void *data, u64 *val)
3981 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3982 struct kvm_vcpu *vcpu;
3986 kvm_for_each_vcpu(i, vcpu, stat_data->kvm)
3987 *val += *(u64 *)((void *)vcpu + stat_data->offset);
3992 static int vcpu_stat_clear_per_vm(void *data, u64 val)
3995 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3996 struct kvm_vcpu *vcpu;
4001 kvm_for_each_vcpu(i, vcpu, stat_data->kvm)
4002 *(u64 *)((void *)vcpu + stat_data->offset) = 0;
4007 static int vcpu_stat_get_per_vm_open(struct inode *inode, struct file *file)
4009 __simple_attr_check_format("%llu\n", 0ull);
4010 return kvm_debugfs_open(inode, file, vcpu_stat_get_per_vm,
4011 vcpu_stat_clear_per_vm, "%llu\n");
4014 static const struct file_operations vcpu_stat_get_per_vm_fops = {
4015 .owner = THIS_MODULE,
4016 .open = vcpu_stat_get_per_vm_open,
4017 .release = kvm_debugfs_release,
4018 .read = simple_attr_read,
4019 .write = simple_attr_write,
4020 .llseek = no_llseek,
4023 static const struct file_operations *stat_fops_per_vm[] = {
4024 [KVM_STAT_VCPU] = &vcpu_stat_get_per_vm_fops,
4025 [KVM_STAT_VM] = &vm_stat_get_per_vm_fops,
4028 static int vm_stat_get(void *_offset, u64 *val)
4030 unsigned offset = (long)_offset;
4032 struct kvm_stat_data stat_tmp = {.offset = offset};
4036 mutex_lock(&kvm_lock);
4037 list_for_each_entry(kvm, &vm_list, vm_list) {
4039 vm_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
4042 mutex_unlock(&kvm_lock);
4046 static int vm_stat_clear(void *_offset, u64 val)
4048 unsigned offset = (long)_offset;
4050 struct kvm_stat_data stat_tmp = {.offset = offset};
4055 mutex_lock(&kvm_lock);
4056 list_for_each_entry(kvm, &vm_list, vm_list) {
4058 vm_stat_clear_per_vm((void *)&stat_tmp, 0);
4060 mutex_unlock(&kvm_lock);
4065 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n");
4067 static int vcpu_stat_get(void *_offset, u64 *val)
4069 unsigned offset = (long)_offset;
4071 struct kvm_stat_data stat_tmp = {.offset = offset};
4075 mutex_lock(&kvm_lock);
4076 list_for_each_entry(kvm, &vm_list, vm_list) {
4078 vcpu_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
4081 mutex_unlock(&kvm_lock);
4085 static int vcpu_stat_clear(void *_offset, u64 val)
4087 unsigned offset = (long)_offset;
4089 struct kvm_stat_data stat_tmp = {.offset = offset};
4094 mutex_lock(&kvm_lock);
4095 list_for_each_entry(kvm, &vm_list, vm_list) {
4097 vcpu_stat_clear_per_vm((void *)&stat_tmp, 0);
4099 mutex_unlock(&kvm_lock);
4104 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear,
4107 static const struct file_operations *stat_fops[] = {
4108 [KVM_STAT_VCPU] = &vcpu_stat_fops,
4109 [KVM_STAT_VM] = &vm_stat_fops,
4112 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm)
4114 struct kobj_uevent_env *env;
4115 unsigned long long created, active;
4117 if (!kvm_dev.this_device || !kvm)
4120 mutex_lock(&kvm_lock);
4121 if (type == KVM_EVENT_CREATE_VM) {
4122 kvm_createvm_count++;
4124 } else if (type == KVM_EVENT_DESTROY_VM) {
4127 created = kvm_createvm_count;
4128 active = kvm_active_vms;
4129 mutex_unlock(&kvm_lock);
4131 env = kzalloc(sizeof(*env), GFP_KERNEL_ACCOUNT);
4135 add_uevent_var(env, "CREATED=%llu", created);
4136 add_uevent_var(env, "COUNT=%llu", active);
4138 if (type == KVM_EVENT_CREATE_VM) {
4139 add_uevent_var(env, "EVENT=create");
4140 kvm->userspace_pid = task_pid_nr(current);
4141 } else if (type == KVM_EVENT_DESTROY_VM) {
4142 add_uevent_var(env, "EVENT=destroy");
4144 add_uevent_var(env, "PID=%d", kvm->userspace_pid);
4146 if (!IS_ERR_OR_NULL(kvm->debugfs_dentry)) {
4147 char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL_ACCOUNT);
4150 tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX);
4152 add_uevent_var(env, "STATS_PATH=%s", tmp);
4156 /* no need for checks, since we are adding at most only 5 keys */
4157 env->envp[env->envp_idx++] = NULL;
4158 kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp);
4162 static void kvm_init_debug(void)
4164 struct kvm_stats_debugfs_item *p;
4166 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
4168 kvm_debugfs_num_entries = 0;
4169 for (p = debugfs_entries; p->name; ++p, kvm_debugfs_num_entries++) {
4170 debugfs_create_file(p->name, 0644, kvm_debugfs_dir,
4171 (void *)(long)p->offset,
4172 stat_fops[p->kind]);
4176 static int kvm_suspend(void)
4178 if (kvm_usage_count)
4179 hardware_disable_nolock(NULL);
4183 static void kvm_resume(void)
4185 if (kvm_usage_count) {
4186 #ifdef CONFIG_LOCKDEP
4187 WARN_ON(lockdep_is_held(&kvm_count_lock));
4189 hardware_enable_nolock(NULL);
4193 static struct syscore_ops kvm_syscore_ops = {
4194 .suspend = kvm_suspend,
4195 .resume = kvm_resume,
4199 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
4201 return container_of(pn, struct kvm_vcpu, preempt_notifier);
4204 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
4206 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
4208 vcpu->preempted = false;
4209 WRITE_ONCE(vcpu->ready, false);
4211 kvm_arch_sched_in(vcpu, cpu);
4213 kvm_arch_vcpu_load(vcpu, cpu);
4216 static void kvm_sched_out(struct preempt_notifier *pn,
4217 struct task_struct *next)
4219 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
4221 if (current->state == TASK_RUNNING) {
4222 vcpu->preempted = true;
4223 WRITE_ONCE(vcpu->ready, true);
4225 kvm_arch_vcpu_put(vcpu);
4228 static void check_processor_compat(void *rtn)
4230 *(int *)rtn = kvm_arch_check_processor_compat();
4233 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
4234 struct module *module)
4239 r = kvm_arch_init(opaque);
4244 * kvm_arch_init makes sure there's at most one caller
4245 * for architectures that support multiple implementations,
4246 * like intel and amd on x86.
4247 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
4248 * conflicts in case kvm is already setup for another implementation.
4250 r = kvm_irqfd_init();
4254 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
4259 r = kvm_arch_hardware_setup();
4263 for_each_online_cpu(cpu) {
4264 smp_call_function_single(cpu, check_processor_compat, &r, 1);
4269 r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting",
4270 kvm_starting_cpu, kvm_dying_cpu);
4273 register_reboot_notifier(&kvm_reboot_notifier);
4275 /* A kmem cache lets us meet the alignment requirements of fx_save. */
4277 vcpu_align = __alignof__(struct kvm_vcpu);
4279 kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align,
4281 offsetof(struct kvm_vcpu, arch),
4282 sizeof_field(struct kvm_vcpu, arch),
4284 if (!kvm_vcpu_cache) {
4289 r = kvm_async_pf_init();
4293 kvm_chardev_ops.owner = module;
4294 kvm_vm_fops.owner = module;
4295 kvm_vcpu_fops.owner = module;
4297 r = misc_register(&kvm_dev);
4299 pr_err("kvm: misc device register failed\n");
4303 register_syscore_ops(&kvm_syscore_ops);
4305 kvm_preempt_ops.sched_in = kvm_sched_in;
4306 kvm_preempt_ops.sched_out = kvm_sched_out;
4310 r = kvm_vfio_ops_init();
4316 kvm_async_pf_deinit();
4318 kmem_cache_destroy(kvm_vcpu_cache);
4320 unregister_reboot_notifier(&kvm_reboot_notifier);
4321 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4324 kvm_arch_hardware_unsetup();
4326 free_cpumask_var(cpus_hardware_enabled);
4334 EXPORT_SYMBOL_GPL(kvm_init);
4338 debugfs_remove_recursive(kvm_debugfs_dir);
4339 misc_deregister(&kvm_dev);
4340 kmem_cache_destroy(kvm_vcpu_cache);
4341 kvm_async_pf_deinit();
4342 unregister_syscore_ops(&kvm_syscore_ops);
4343 unregister_reboot_notifier(&kvm_reboot_notifier);
4344 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4345 on_each_cpu(hardware_disable_nolock, NULL, 1);
4346 kvm_arch_hardware_unsetup();
4349 free_cpumask_var(cpus_hardware_enabled);
4350 kvm_vfio_ops_exit();
4352 EXPORT_SYMBOL_GPL(kvm_exit);