1 // SPDX-License-Identifier: GPL-2.0-only
3 * Kernel-based Virtual Machine driver for Linux
5 * This module enables machines with Intel VT-x extensions to run virtual
6 * machines without emulation or binary translation.
8 * Copyright (C) 2006 Qumranet, Inc.
9 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
12 * Avi Kivity <avi@qumranet.com>
13 * Yaniv Kamay <yaniv@qumranet.com>
16 #include <kvm/iodev.h>
18 #include <linux/kvm_host.h>
19 #include <linux/kvm.h>
20 #include <linux/module.h>
21 #include <linux/errno.h>
22 #include <linux/percpu.h>
24 #include <linux/miscdevice.h>
25 #include <linux/vmalloc.h>
26 #include <linux/reboot.h>
27 #include <linux/debugfs.h>
28 #include <linux/highmem.h>
29 #include <linux/file.h>
30 #include <linux/syscore_ops.h>
31 #include <linux/cpu.h>
32 #include <linux/sched/signal.h>
33 #include <linux/sched/mm.h>
34 #include <linux/sched/stat.h>
35 #include <linux/cpumask.h>
36 #include <linux/smp.h>
37 #include <linux/anon_inodes.h>
38 #include <linux/profile.h>
39 #include <linux/kvm_para.h>
40 #include <linux/pagemap.h>
41 #include <linux/mman.h>
42 #include <linux/swap.h>
43 #include <linux/bitops.h>
44 #include <linux/spinlock.h>
45 #include <linux/compat.h>
46 #include <linux/srcu.h>
47 #include <linux/hugetlb.h>
48 #include <linux/slab.h>
49 #include <linux/sort.h>
50 #include <linux/bsearch.h>
52 #include <linux/lockdep.h>
54 #include <asm/processor.h>
55 #include <asm/ioctl.h>
56 #include <linux/uaccess.h>
57 #include <asm/pgtable.h>
59 #include "coalesced_mmio.h"
63 #define CREATE_TRACE_POINTS
64 #include <trace/events/kvm.h>
66 /* Worst case buffer size needed for holding an integer. */
67 #define ITOA_MAX_LEN 12
69 MODULE_AUTHOR("Qumranet");
70 MODULE_LICENSE("GPL");
72 /* Architectures should define their poll value according to the halt latency */
73 unsigned int halt_poll_ns = KVM_HALT_POLL_NS_DEFAULT;
74 module_param(halt_poll_ns, uint, 0644);
75 EXPORT_SYMBOL_GPL(halt_poll_ns);
77 /* Default doubles per-vcpu halt_poll_ns. */
78 unsigned int halt_poll_ns_grow = 2;
79 module_param(halt_poll_ns_grow, uint, 0644);
80 EXPORT_SYMBOL_GPL(halt_poll_ns_grow);
82 /* The start value to grow halt_poll_ns from */
83 unsigned int halt_poll_ns_grow_start = 10000; /* 10us */
84 module_param(halt_poll_ns_grow_start, uint, 0644);
85 EXPORT_SYMBOL_GPL(halt_poll_ns_grow_start);
87 /* Default resets per-vcpu halt_poll_ns . */
88 unsigned int halt_poll_ns_shrink;
89 module_param(halt_poll_ns_shrink, uint, 0644);
90 EXPORT_SYMBOL_GPL(halt_poll_ns_shrink);
95 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
98 DEFINE_MUTEX(kvm_lock);
99 static DEFINE_RAW_SPINLOCK(kvm_count_lock);
102 static cpumask_var_t cpus_hardware_enabled;
103 static int kvm_usage_count;
104 static atomic_t hardware_enable_failed;
106 struct kmem_cache *kvm_vcpu_cache;
107 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
109 static __read_mostly struct preempt_ops kvm_preempt_ops;
111 struct dentry *kvm_debugfs_dir;
112 EXPORT_SYMBOL_GPL(kvm_debugfs_dir);
114 static int kvm_debugfs_num_entries;
115 static const struct file_operations *stat_fops_per_vm[];
117 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
119 #ifdef CONFIG_KVM_COMPAT
120 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
122 #define KVM_COMPAT(c) .compat_ioctl = (c)
124 static long kvm_no_compat_ioctl(struct file *file, unsigned int ioctl,
125 unsigned long arg) { return -EINVAL; }
126 #define KVM_COMPAT(c) .compat_ioctl = kvm_no_compat_ioctl
128 static int hardware_enable_all(void);
129 static void hardware_disable_all(void);
131 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
133 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot, gfn_t gfn);
135 __visible bool kvm_rebooting;
136 EXPORT_SYMBOL_GPL(kvm_rebooting);
138 static bool largepages_enabled = true;
140 #define KVM_EVENT_CREATE_VM 0
141 #define KVM_EVENT_DESTROY_VM 1
142 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm);
143 static unsigned long long kvm_createvm_count;
144 static unsigned long long kvm_active_vms;
146 __weak int kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
147 unsigned long start, unsigned long end, bool blockable)
152 bool kvm_is_reserved_pfn(kvm_pfn_t pfn)
155 return PageReserved(pfn_to_page(pfn));
161 * Switches to specified vcpu, until a matching vcpu_put()
163 void vcpu_load(struct kvm_vcpu *vcpu)
166 preempt_notifier_register(&vcpu->preempt_notifier);
167 kvm_arch_vcpu_load(vcpu, cpu);
170 EXPORT_SYMBOL_GPL(vcpu_load);
172 void vcpu_put(struct kvm_vcpu *vcpu)
175 kvm_arch_vcpu_put(vcpu);
176 preempt_notifier_unregister(&vcpu->preempt_notifier);
179 EXPORT_SYMBOL_GPL(vcpu_put);
181 /* TODO: merge with kvm_arch_vcpu_should_kick */
182 static bool kvm_request_needs_ipi(struct kvm_vcpu *vcpu, unsigned req)
184 int mode = kvm_vcpu_exiting_guest_mode(vcpu);
187 * We need to wait for the VCPU to reenable interrupts and get out of
188 * READING_SHADOW_PAGE_TABLES mode.
190 if (req & KVM_REQUEST_WAIT)
191 return mode != OUTSIDE_GUEST_MODE;
194 * Need to kick a running VCPU, but otherwise there is nothing to do.
196 return mode == IN_GUEST_MODE;
199 static void ack_flush(void *_completed)
203 static inline bool kvm_kick_many_cpus(const struct cpumask *cpus, bool wait)
206 cpus = cpu_online_mask;
208 if (cpumask_empty(cpus))
211 smp_call_function_many(cpus, ack_flush, NULL, wait);
215 bool kvm_make_vcpus_request_mask(struct kvm *kvm, unsigned int req,
216 unsigned long *vcpu_bitmap, cpumask_var_t tmp)
219 struct kvm_vcpu *vcpu;
224 kvm_for_each_vcpu(i, vcpu, kvm) {
225 if (vcpu_bitmap && !test_bit(i, vcpu_bitmap))
228 kvm_make_request(req, vcpu);
231 if (!(req & KVM_REQUEST_NO_WAKEUP) && kvm_vcpu_wake_up(vcpu))
234 if (tmp != NULL && cpu != -1 && cpu != me &&
235 kvm_request_needs_ipi(vcpu, req))
236 __cpumask_set_cpu(cpu, tmp);
239 called = kvm_kick_many_cpus(tmp, !!(req & KVM_REQUEST_WAIT));
245 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
250 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
252 called = kvm_make_vcpus_request_mask(kvm, req, NULL, cpus);
254 free_cpumask_var(cpus);
258 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
259 void kvm_flush_remote_tlbs(struct kvm *kvm)
262 * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in
263 * kvm_make_all_cpus_request.
265 long dirty_count = smp_load_acquire(&kvm->tlbs_dirty);
268 * We want to publish modifications to the page tables before reading
269 * mode. Pairs with a memory barrier in arch-specific code.
270 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
271 * and smp_mb in walk_shadow_page_lockless_begin/end.
272 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
274 * There is already an smp_mb__after_atomic() before
275 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
278 if (!kvm_arch_flush_remote_tlb(kvm)
279 || kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
280 ++kvm->stat.remote_tlb_flush;
281 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
283 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
286 void kvm_reload_remote_mmus(struct kvm *kvm)
288 kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
291 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
296 mutex_init(&vcpu->mutex);
301 init_swait_queue_head(&vcpu->wq);
302 kvm_async_pf_vcpu_init(vcpu);
305 INIT_LIST_HEAD(&vcpu->blocked_vcpu_list);
307 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
312 vcpu->run = page_address(page);
314 kvm_vcpu_set_in_spin_loop(vcpu, false);
315 kvm_vcpu_set_dy_eligible(vcpu, false);
316 vcpu->preempted = false;
319 r = kvm_arch_vcpu_init(vcpu);
325 free_page((unsigned long)vcpu->run);
329 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
331 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
334 * no need for rcu_read_lock as VCPU_RUN is the only place that
335 * will change the vcpu->pid pointer and on uninit all file
336 * descriptors are already gone.
338 put_pid(rcu_dereference_protected(vcpu->pid, 1));
339 kvm_arch_vcpu_uninit(vcpu);
340 free_page((unsigned long)vcpu->run);
342 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
344 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
345 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
347 return container_of(mn, struct kvm, mmu_notifier);
350 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
351 struct mm_struct *mm,
352 unsigned long address,
355 struct kvm *kvm = mmu_notifier_to_kvm(mn);
358 idx = srcu_read_lock(&kvm->srcu);
359 spin_lock(&kvm->mmu_lock);
360 kvm->mmu_notifier_seq++;
362 if (kvm_set_spte_hva(kvm, address, pte))
363 kvm_flush_remote_tlbs(kvm);
365 spin_unlock(&kvm->mmu_lock);
366 srcu_read_unlock(&kvm->srcu, idx);
369 static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
370 const struct mmu_notifier_range *range)
372 struct kvm *kvm = mmu_notifier_to_kvm(mn);
373 int need_tlb_flush = 0, idx;
376 idx = srcu_read_lock(&kvm->srcu);
377 spin_lock(&kvm->mmu_lock);
379 * The count increase must become visible at unlock time as no
380 * spte can be established without taking the mmu_lock and
381 * count is also read inside the mmu_lock critical section.
383 kvm->mmu_notifier_count++;
384 need_tlb_flush = kvm_unmap_hva_range(kvm, range->start, range->end);
385 need_tlb_flush |= kvm->tlbs_dirty;
386 /* we've to flush the tlb before the pages can be freed */
388 kvm_flush_remote_tlbs(kvm);
390 spin_unlock(&kvm->mmu_lock);
392 ret = kvm_arch_mmu_notifier_invalidate_range(kvm, range->start,
394 mmu_notifier_range_blockable(range));
396 srcu_read_unlock(&kvm->srcu, idx);
401 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
402 const struct mmu_notifier_range *range)
404 struct kvm *kvm = mmu_notifier_to_kvm(mn);
406 spin_lock(&kvm->mmu_lock);
408 * This sequence increase will notify the kvm page fault that
409 * the page that is going to be mapped in the spte could have
412 kvm->mmu_notifier_seq++;
415 * The above sequence increase must be visible before the
416 * below count decrease, which is ensured by the smp_wmb above
417 * in conjunction with the smp_rmb in mmu_notifier_retry().
419 kvm->mmu_notifier_count--;
420 spin_unlock(&kvm->mmu_lock);
422 BUG_ON(kvm->mmu_notifier_count < 0);
425 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
426 struct mm_struct *mm,
430 struct kvm *kvm = mmu_notifier_to_kvm(mn);
433 idx = srcu_read_lock(&kvm->srcu);
434 spin_lock(&kvm->mmu_lock);
436 young = kvm_age_hva(kvm, start, end);
438 kvm_flush_remote_tlbs(kvm);
440 spin_unlock(&kvm->mmu_lock);
441 srcu_read_unlock(&kvm->srcu, idx);
446 static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
447 struct mm_struct *mm,
451 struct kvm *kvm = mmu_notifier_to_kvm(mn);
454 idx = srcu_read_lock(&kvm->srcu);
455 spin_lock(&kvm->mmu_lock);
457 * Even though we do not flush TLB, this will still adversely
458 * affect performance on pre-Haswell Intel EPT, where there is
459 * no EPT Access Bit to clear so that we have to tear down EPT
460 * tables instead. If we find this unacceptable, we can always
461 * add a parameter to kvm_age_hva so that it effectively doesn't
462 * do anything on clear_young.
464 * Also note that currently we never issue secondary TLB flushes
465 * from clear_young, leaving this job up to the regular system
466 * cadence. If we find this inaccurate, we might come up with a
467 * more sophisticated heuristic later.
469 young = kvm_age_hva(kvm, start, end);
470 spin_unlock(&kvm->mmu_lock);
471 srcu_read_unlock(&kvm->srcu, idx);
476 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
477 struct mm_struct *mm,
478 unsigned long address)
480 struct kvm *kvm = mmu_notifier_to_kvm(mn);
483 idx = srcu_read_lock(&kvm->srcu);
484 spin_lock(&kvm->mmu_lock);
485 young = kvm_test_age_hva(kvm, address);
486 spin_unlock(&kvm->mmu_lock);
487 srcu_read_unlock(&kvm->srcu, idx);
492 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
493 struct mm_struct *mm)
495 struct kvm *kvm = mmu_notifier_to_kvm(mn);
498 idx = srcu_read_lock(&kvm->srcu);
499 kvm_arch_flush_shadow_all(kvm);
500 srcu_read_unlock(&kvm->srcu, idx);
503 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
504 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
505 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
506 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
507 .clear_young = kvm_mmu_notifier_clear_young,
508 .test_young = kvm_mmu_notifier_test_young,
509 .change_pte = kvm_mmu_notifier_change_pte,
510 .release = kvm_mmu_notifier_release,
513 static int kvm_init_mmu_notifier(struct kvm *kvm)
515 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
516 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
519 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
521 static int kvm_init_mmu_notifier(struct kvm *kvm)
526 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
528 static struct kvm_memslots *kvm_alloc_memslots(void)
531 struct kvm_memslots *slots;
533 slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL_ACCOUNT);
537 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
538 slots->id_to_index[i] = slots->memslots[i].id = i;
543 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
545 if (!memslot->dirty_bitmap)
548 kvfree(memslot->dirty_bitmap);
549 memslot->dirty_bitmap = NULL;
553 * Free any memory in @free but not in @dont.
555 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
556 struct kvm_memory_slot *dont)
558 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
559 kvm_destroy_dirty_bitmap(free);
561 kvm_arch_free_memslot(kvm, free, dont);
566 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
568 struct kvm_memory_slot *memslot;
573 kvm_for_each_memslot(memslot, slots)
574 kvm_free_memslot(kvm, memslot, NULL);
579 static void kvm_destroy_vm_debugfs(struct kvm *kvm)
583 if (!kvm->debugfs_dentry)
586 debugfs_remove_recursive(kvm->debugfs_dentry);
588 if (kvm->debugfs_stat_data) {
589 for (i = 0; i < kvm_debugfs_num_entries; i++)
590 kfree(kvm->debugfs_stat_data[i]);
591 kfree(kvm->debugfs_stat_data);
595 static int kvm_create_vm_debugfs(struct kvm *kvm, int fd)
597 char dir_name[ITOA_MAX_LEN * 2];
598 struct kvm_stat_data *stat_data;
599 struct kvm_stats_debugfs_item *p;
601 if (!debugfs_initialized())
604 snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd);
605 kvm->debugfs_dentry = debugfs_create_dir(dir_name, kvm_debugfs_dir);
607 kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries,
608 sizeof(*kvm->debugfs_stat_data),
610 if (!kvm->debugfs_stat_data)
613 for (p = debugfs_entries; p->name; p++) {
614 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT);
618 stat_data->kvm = kvm;
619 stat_data->offset = p->offset;
620 stat_data->mode = p->mode ? p->mode : 0644;
621 kvm->debugfs_stat_data[p - debugfs_entries] = stat_data;
622 debugfs_create_file(p->name, stat_data->mode, kvm->debugfs_dentry,
623 stat_data, stat_fops_per_vm[p->kind]);
628 static struct kvm *kvm_create_vm(unsigned long type)
630 struct kvm *kvm = kvm_arch_alloc_vm();
635 return ERR_PTR(-ENOMEM);
637 spin_lock_init(&kvm->mmu_lock);
639 kvm->mm = current->mm;
640 kvm_eventfd_init(kvm);
641 mutex_init(&kvm->lock);
642 mutex_init(&kvm->irq_lock);
643 mutex_init(&kvm->slots_lock);
644 INIT_LIST_HEAD(&kvm->devices);
646 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
648 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
649 struct kvm_memslots *slots = kvm_alloc_memslots();
652 goto out_err_no_arch_destroy_vm;
653 /* Generations must be different for each address space. */
654 slots->generation = i;
655 rcu_assign_pointer(kvm->memslots[i], slots);
658 for (i = 0; i < KVM_NR_BUSES; i++) {
659 rcu_assign_pointer(kvm->buses[i],
660 kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL_ACCOUNT));
662 goto out_err_no_arch_destroy_vm;
665 refcount_set(&kvm->users_count, 1);
666 r = kvm_arch_init_vm(kvm, type);
668 goto out_err_no_arch_destroy_vm;
670 r = hardware_enable_all();
672 goto out_err_no_disable;
674 #ifdef CONFIG_HAVE_KVM_IRQFD
675 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
678 if (init_srcu_struct(&kvm->srcu))
679 goto out_err_no_srcu;
680 if (init_srcu_struct(&kvm->irq_srcu))
681 goto out_err_no_irq_srcu;
683 r = kvm_init_mmu_notifier(kvm);
687 mutex_lock(&kvm_lock);
688 list_add(&kvm->vm_list, &vm_list);
689 mutex_unlock(&kvm_lock);
691 preempt_notifier_inc();
696 cleanup_srcu_struct(&kvm->irq_srcu);
698 cleanup_srcu_struct(&kvm->srcu);
700 hardware_disable_all();
702 kvm_arch_destroy_vm(kvm);
703 WARN_ON_ONCE(!refcount_dec_and_test(&kvm->users_count));
704 out_err_no_arch_destroy_vm:
705 for (i = 0; i < KVM_NR_BUSES; i++)
706 kfree(kvm_get_bus(kvm, i));
707 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
708 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
709 kvm_arch_free_vm(kvm);
714 static void kvm_destroy_devices(struct kvm *kvm)
716 struct kvm_device *dev, *tmp;
719 * We do not need to take the kvm->lock here, because nobody else
720 * has a reference to the struct kvm at this point and therefore
721 * cannot access the devices list anyhow.
723 list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
724 list_del(&dev->vm_node);
725 dev->ops->destroy(dev);
729 static void kvm_destroy_vm(struct kvm *kvm)
732 struct mm_struct *mm = kvm->mm;
734 kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm);
735 kvm_destroy_vm_debugfs(kvm);
736 kvm_arch_sync_events(kvm);
737 mutex_lock(&kvm_lock);
738 list_del(&kvm->vm_list);
739 mutex_unlock(&kvm_lock);
740 kvm_free_irq_routing(kvm);
741 for (i = 0; i < KVM_NR_BUSES; i++) {
742 struct kvm_io_bus *bus = kvm_get_bus(kvm, i);
745 kvm_io_bus_destroy(bus);
746 kvm->buses[i] = NULL;
748 kvm_coalesced_mmio_free(kvm);
749 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
750 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
752 kvm_arch_flush_shadow_all(kvm);
754 kvm_arch_destroy_vm(kvm);
755 kvm_destroy_devices(kvm);
756 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
757 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
758 cleanup_srcu_struct(&kvm->irq_srcu);
759 cleanup_srcu_struct(&kvm->srcu);
760 kvm_arch_free_vm(kvm);
761 preempt_notifier_dec();
762 hardware_disable_all();
766 void kvm_get_kvm(struct kvm *kvm)
768 refcount_inc(&kvm->users_count);
770 EXPORT_SYMBOL_GPL(kvm_get_kvm);
772 void kvm_put_kvm(struct kvm *kvm)
774 if (refcount_dec_and_test(&kvm->users_count))
777 EXPORT_SYMBOL_GPL(kvm_put_kvm);
780 static int kvm_vm_release(struct inode *inode, struct file *filp)
782 struct kvm *kvm = filp->private_data;
784 kvm_irqfd_release(kvm);
791 * Allocation size is twice as large as the actual dirty bitmap size.
792 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
794 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
796 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
798 memslot->dirty_bitmap = kvzalloc(dirty_bytes, GFP_KERNEL_ACCOUNT);
799 if (!memslot->dirty_bitmap)
806 * Insert memslot and re-sort memslots based on their GFN,
807 * so binary search could be used to lookup GFN.
808 * Sorting algorithm takes advantage of having initially
809 * sorted array and known changed memslot position.
811 static void update_memslots(struct kvm_memslots *slots,
812 struct kvm_memory_slot *new,
813 enum kvm_mr_change change)
816 int i = slots->id_to_index[id];
817 struct kvm_memory_slot *mslots = slots->memslots;
819 WARN_ON(mslots[i].id != id);
823 WARN_ON(mslots[i].npages || !new->npages);
827 WARN_ON(new->npages || !mslots[i].npages);
833 while (i < KVM_MEM_SLOTS_NUM - 1 &&
834 new->base_gfn <= mslots[i + 1].base_gfn) {
835 if (!mslots[i + 1].npages)
837 mslots[i] = mslots[i + 1];
838 slots->id_to_index[mslots[i].id] = i;
843 * The ">=" is needed when creating a slot with base_gfn == 0,
844 * so that it moves before all those with base_gfn == npages == 0.
846 * On the other hand, if new->npages is zero, the above loop has
847 * already left i pointing to the beginning of the empty part of
848 * mslots, and the ">=" would move the hole backwards in this
849 * case---which is wrong. So skip the loop when deleting a slot.
853 new->base_gfn >= mslots[i - 1].base_gfn) {
854 mslots[i] = mslots[i - 1];
855 slots->id_to_index[mslots[i].id] = i;
859 WARN_ON_ONCE(i != slots->used_slots);
862 slots->id_to_index[mslots[i].id] = i;
865 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
867 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
869 #ifdef __KVM_HAVE_READONLY_MEM
870 valid_flags |= KVM_MEM_READONLY;
873 if (mem->flags & ~valid_flags)
879 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
880 int as_id, struct kvm_memslots *slots)
882 struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id);
883 u64 gen = old_memslots->generation;
885 WARN_ON(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS);
886 slots->generation = gen | KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
888 rcu_assign_pointer(kvm->memslots[as_id], slots);
889 synchronize_srcu_expedited(&kvm->srcu);
892 * Increment the new memslot generation a second time, dropping the
893 * update in-progress flag and incrementing then generation based on
894 * the number of address spaces. This provides a unique and easily
895 * identifiable generation number while the memslots are in flux.
897 gen = slots->generation & ~KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
900 * Generations must be unique even across address spaces. We do not need
901 * a global counter for that, instead the generation space is evenly split
902 * across address spaces. For example, with two address spaces, address
903 * space 0 will use generations 0, 2, 4, ... while address space 1 will
904 * use generations 1, 3, 5, ...
906 gen += KVM_ADDRESS_SPACE_NUM;
908 kvm_arch_memslots_updated(kvm, gen);
910 slots->generation = gen;
916 * Allocate some memory and give it an address in the guest physical address
919 * Discontiguous memory is allowed, mostly for framebuffers.
921 * Must be called holding kvm->slots_lock for write.
923 int __kvm_set_memory_region(struct kvm *kvm,
924 const struct kvm_userspace_memory_region *mem)
928 unsigned long npages;
929 struct kvm_memory_slot *slot;
930 struct kvm_memory_slot old, new;
931 struct kvm_memslots *slots = NULL, *old_memslots;
933 enum kvm_mr_change change;
935 r = check_memory_region_flags(mem);
940 as_id = mem->slot >> 16;
943 /* General sanity checks */
944 if (mem->memory_size & (PAGE_SIZE - 1))
946 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
948 /* We can read the guest memory with __xxx_user() later on. */
949 if ((id < KVM_USER_MEM_SLOTS) &&
950 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
951 !access_ok((void __user *)(unsigned long)mem->userspace_addr,
954 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
956 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
959 slot = id_to_memslot(__kvm_memslots(kvm, as_id), id);
960 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
961 npages = mem->memory_size >> PAGE_SHIFT;
963 if (npages > KVM_MEM_MAX_NR_PAGES)
969 new.base_gfn = base_gfn;
971 new.flags = mem->flags;
975 change = KVM_MR_CREATE;
976 else { /* Modify an existing slot. */
977 if ((mem->userspace_addr != old.userspace_addr) ||
978 (npages != old.npages) ||
979 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
982 if (base_gfn != old.base_gfn)
983 change = KVM_MR_MOVE;
984 else if (new.flags != old.flags)
985 change = KVM_MR_FLAGS_ONLY;
986 else { /* Nothing to change. */
995 change = KVM_MR_DELETE;
1000 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
1001 /* Check for overlaps */
1003 kvm_for_each_memslot(slot, __kvm_memslots(kvm, as_id)) {
1006 if (!((base_gfn + npages <= slot->base_gfn) ||
1007 (base_gfn >= slot->base_gfn + slot->npages)))
1012 /* Free page dirty bitmap if unneeded */
1013 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
1014 new.dirty_bitmap = NULL;
1017 if (change == KVM_MR_CREATE) {
1018 new.userspace_addr = mem->userspace_addr;
1020 if (kvm_arch_create_memslot(kvm, &new, npages))
1024 /* Allocate page dirty bitmap if needed */
1025 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
1026 if (kvm_create_dirty_bitmap(&new) < 0)
1030 slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL_ACCOUNT);
1033 memcpy(slots, __kvm_memslots(kvm, as_id), sizeof(struct kvm_memslots));
1035 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
1036 slot = id_to_memslot(slots, id);
1037 slot->flags |= KVM_MEMSLOT_INVALID;
1039 old_memslots = install_new_memslots(kvm, as_id, slots);
1041 /* From this point no new shadow pages pointing to a deleted,
1042 * or moved, memslot will be created.
1044 * validation of sp->gfn happens in:
1045 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1046 * - kvm_is_visible_gfn (mmu_check_roots)
1048 kvm_arch_flush_shadow_memslot(kvm, slot);
1051 * We can re-use the old_memslots from above, the only difference
1052 * from the currently installed memslots is the invalid flag. This
1053 * will get overwritten by update_memslots anyway.
1055 slots = old_memslots;
1058 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
1062 /* actual memory is freed via old in kvm_free_memslot below */
1063 if (change == KVM_MR_DELETE) {
1064 new.dirty_bitmap = NULL;
1065 memset(&new.arch, 0, sizeof(new.arch));
1068 update_memslots(slots, &new, change);
1069 old_memslots = install_new_memslots(kvm, as_id, slots);
1071 kvm_arch_commit_memory_region(kvm, mem, &old, &new, change);
1073 kvm_free_memslot(kvm, &old, &new);
1074 kvfree(old_memslots);
1080 kvm_free_memslot(kvm, &new, &old);
1084 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1086 int kvm_set_memory_region(struct kvm *kvm,
1087 const struct kvm_userspace_memory_region *mem)
1091 mutex_lock(&kvm->slots_lock);
1092 r = __kvm_set_memory_region(kvm, mem);
1093 mutex_unlock(&kvm->slots_lock);
1096 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1098 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1099 struct kvm_userspace_memory_region *mem)
1101 if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1104 return kvm_set_memory_region(kvm, mem);
1107 int kvm_get_dirty_log(struct kvm *kvm,
1108 struct kvm_dirty_log *log, int *is_dirty)
1110 struct kvm_memslots *slots;
1111 struct kvm_memory_slot *memslot;
1114 unsigned long any = 0;
1116 as_id = log->slot >> 16;
1117 id = (u16)log->slot;
1118 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1121 slots = __kvm_memslots(kvm, as_id);
1122 memslot = id_to_memslot(slots, id);
1123 if (!memslot->dirty_bitmap)
1126 n = kvm_dirty_bitmap_bytes(memslot);
1128 for (i = 0; !any && i < n/sizeof(long); ++i)
1129 any = memslot->dirty_bitmap[i];
1131 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1138 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1140 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1142 * kvm_get_dirty_log_protect - get a snapshot of dirty pages
1143 * and reenable dirty page tracking for the corresponding pages.
1144 * @kvm: pointer to kvm instance
1145 * @log: slot id and address to which we copy the log
1146 * @flush: true if TLB flush is needed by caller
1148 * We need to keep it in mind that VCPU threads can write to the bitmap
1149 * concurrently. So, to avoid losing track of dirty pages we keep the
1152 * 1. Take a snapshot of the bit and clear it if needed.
1153 * 2. Write protect the corresponding page.
1154 * 3. Copy the snapshot to the userspace.
1155 * 4. Upon return caller flushes TLB's if needed.
1157 * Between 2 and 4, the guest may write to the page using the remaining TLB
1158 * entry. This is not a problem because the page is reported dirty using
1159 * the snapshot taken before and step 4 ensures that writes done after
1160 * exiting to userspace will be logged for the next call.
1163 int kvm_get_dirty_log_protect(struct kvm *kvm,
1164 struct kvm_dirty_log *log, bool *flush)
1166 struct kvm_memslots *slots;
1167 struct kvm_memory_slot *memslot;
1170 unsigned long *dirty_bitmap;
1171 unsigned long *dirty_bitmap_buffer;
1173 as_id = log->slot >> 16;
1174 id = (u16)log->slot;
1175 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1178 slots = __kvm_memslots(kvm, as_id);
1179 memslot = id_to_memslot(slots, id);
1181 dirty_bitmap = memslot->dirty_bitmap;
1185 n = kvm_dirty_bitmap_bytes(memslot);
1187 if (kvm->manual_dirty_log_protect) {
1189 * Unlike kvm_get_dirty_log, we always return false in *flush,
1190 * because no flush is needed until KVM_CLEAR_DIRTY_LOG. There
1191 * is some code duplication between this function and
1192 * kvm_get_dirty_log, but hopefully all architecture
1193 * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log
1194 * can be eliminated.
1196 dirty_bitmap_buffer = dirty_bitmap;
1198 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
1199 memset(dirty_bitmap_buffer, 0, n);
1201 spin_lock(&kvm->mmu_lock);
1202 for (i = 0; i < n / sizeof(long); i++) {
1206 if (!dirty_bitmap[i])
1210 mask = xchg(&dirty_bitmap[i], 0);
1211 dirty_bitmap_buffer[i] = mask;
1213 offset = i * BITS_PER_LONG;
1214 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1217 spin_unlock(&kvm->mmu_lock);
1220 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1224 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect);
1227 * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap
1228 * and reenable dirty page tracking for the corresponding pages.
1229 * @kvm: pointer to kvm instance
1230 * @log: slot id and address from which to fetch the bitmap of dirty pages
1231 * @flush: true if TLB flush is needed by caller
1233 int kvm_clear_dirty_log_protect(struct kvm *kvm,
1234 struct kvm_clear_dirty_log *log, bool *flush)
1236 struct kvm_memslots *slots;
1237 struct kvm_memory_slot *memslot;
1241 unsigned long *dirty_bitmap;
1242 unsigned long *dirty_bitmap_buffer;
1244 as_id = log->slot >> 16;
1245 id = (u16)log->slot;
1246 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1249 if (log->first_page & 63)
1252 slots = __kvm_memslots(kvm, as_id);
1253 memslot = id_to_memslot(slots, id);
1255 dirty_bitmap = memslot->dirty_bitmap;
1259 n = ALIGN(log->num_pages, BITS_PER_LONG) / 8;
1261 if (log->first_page > memslot->npages ||
1262 log->num_pages > memslot->npages - log->first_page ||
1263 (log->num_pages < memslot->npages - log->first_page && (log->num_pages & 63)))
1267 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
1268 if (copy_from_user(dirty_bitmap_buffer, log->dirty_bitmap, n))
1271 spin_lock(&kvm->mmu_lock);
1272 for (offset = log->first_page, i = offset / BITS_PER_LONG,
1273 n = DIV_ROUND_UP(log->num_pages, BITS_PER_LONG); n--;
1274 i++, offset += BITS_PER_LONG) {
1275 unsigned long mask = *dirty_bitmap_buffer++;
1276 atomic_long_t *p = (atomic_long_t *) &dirty_bitmap[i];
1280 mask &= atomic_long_fetch_andnot(mask, p);
1283 * mask contains the bits that really have been cleared. This
1284 * never includes any bits beyond the length of the memslot (if
1285 * the length is not aligned to 64 pages), therefore it is not
1286 * a problem if userspace sets them in log->dirty_bitmap.
1290 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1294 spin_unlock(&kvm->mmu_lock);
1298 EXPORT_SYMBOL_GPL(kvm_clear_dirty_log_protect);
1301 bool kvm_largepages_enabled(void)
1303 return largepages_enabled;
1306 void kvm_disable_largepages(void)
1308 largepages_enabled = false;
1310 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1312 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1314 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1316 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1318 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
1320 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn);
1323 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1325 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1327 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1328 memslot->flags & KVM_MEMSLOT_INVALID)
1333 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1335 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1337 struct vm_area_struct *vma;
1338 unsigned long addr, size;
1342 addr = gfn_to_hva(kvm, gfn);
1343 if (kvm_is_error_hva(addr))
1346 down_read(¤t->mm->mmap_sem);
1347 vma = find_vma(current->mm, addr);
1351 size = vma_kernel_pagesize(vma);
1354 up_read(¤t->mm->mmap_sem);
1359 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1361 return slot->flags & KVM_MEM_READONLY;
1364 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1365 gfn_t *nr_pages, bool write)
1367 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1368 return KVM_HVA_ERR_BAD;
1370 if (memslot_is_readonly(slot) && write)
1371 return KVM_HVA_ERR_RO_BAD;
1374 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1376 return __gfn_to_hva_memslot(slot, gfn);
1379 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1382 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1385 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1388 return gfn_to_hva_many(slot, gfn, NULL);
1390 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1392 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1394 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1396 EXPORT_SYMBOL_GPL(gfn_to_hva);
1398 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
1400 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
1402 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
1405 * Return the hva of a @gfn and the R/W attribute if possible.
1407 * @slot: the kvm_memory_slot which contains @gfn
1408 * @gfn: the gfn to be translated
1409 * @writable: used to return the read/write attribute of the @slot if the hva
1410 * is valid and @writable is not NULL
1412 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1413 gfn_t gfn, bool *writable)
1415 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1417 if (!kvm_is_error_hva(hva) && writable)
1418 *writable = !memslot_is_readonly(slot);
1423 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1425 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1427 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1430 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
1432 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1434 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1437 static inline int check_user_page_hwpoison(unsigned long addr)
1439 int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
1441 rc = get_user_pages(addr, 1, flags, NULL, NULL);
1442 return rc == -EHWPOISON;
1446 * The fast path to get the writable pfn which will be stored in @pfn,
1447 * true indicates success, otherwise false is returned. It's also the
1448 * only part that runs if we can are in atomic context.
1450 static bool hva_to_pfn_fast(unsigned long addr, bool write_fault,
1451 bool *writable, kvm_pfn_t *pfn)
1453 struct page *page[1];
1457 * Fast pin a writable pfn only if it is a write fault request
1458 * or the caller allows to map a writable pfn for a read fault
1461 if (!(write_fault || writable))
1464 npages = __get_user_pages_fast(addr, 1, 1, page);
1466 *pfn = page_to_pfn(page[0]);
1477 * The slow path to get the pfn of the specified host virtual address,
1478 * 1 indicates success, -errno is returned if error is detected.
1480 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1481 bool *writable, kvm_pfn_t *pfn)
1483 unsigned int flags = FOLL_HWPOISON;
1490 *writable = write_fault;
1493 flags |= FOLL_WRITE;
1495 flags |= FOLL_NOWAIT;
1497 npages = get_user_pages_unlocked(addr, 1, &page, flags);
1501 /* map read fault as writable if possible */
1502 if (unlikely(!write_fault) && writable) {
1505 if (__get_user_pages_fast(addr, 1, 1, &wpage) == 1) {
1511 *pfn = page_to_pfn(page);
1515 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1517 if (unlikely(!(vma->vm_flags & VM_READ)))
1520 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1526 static int hva_to_pfn_remapped(struct vm_area_struct *vma,
1527 unsigned long addr, bool *async,
1528 bool write_fault, bool *writable,
1534 r = follow_pfn(vma, addr, &pfn);
1537 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
1538 * not call the fault handler, so do it here.
1540 bool unlocked = false;
1541 r = fixup_user_fault(current, current->mm, addr,
1542 (write_fault ? FAULT_FLAG_WRITE : 0),
1549 r = follow_pfn(vma, addr, &pfn);
1559 * Get a reference here because callers of *hva_to_pfn* and
1560 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
1561 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
1562 * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will
1563 * simply do nothing for reserved pfns.
1565 * Whoever called remap_pfn_range is also going to call e.g.
1566 * unmap_mapping_range before the underlying pages are freed,
1567 * causing a call to our MMU notifier.
1576 * Pin guest page in memory and return its pfn.
1577 * @addr: host virtual address which maps memory to the guest
1578 * @atomic: whether this function can sleep
1579 * @async: whether this function need to wait IO complete if the
1580 * host page is not in the memory
1581 * @write_fault: whether we should get a writable host page
1582 * @writable: whether it allows to map a writable host page for !@write_fault
1584 * The function will map a writable host page for these two cases:
1585 * 1): @write_fault = true
1586 * 2): @write_fault = false && @writable, @writable will tell the caller
1587 * whether the mapping is writable.
1589 static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1590 bool write_fault, bool *writable)
1592 struct vm_area_struct *vma;
1596 /* we can do it either atomically or asynchronously, not both */
1597 BUG_ON(atomic && async);
1599 if (hva_to_pfn_fast(addr, write_fault, writable, &pfn))
1603 return KVM_PFN_ERR_FAULT;
1605 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1609 down_read(¤t->mm->mmap_sem);
1610 if (npages == -EHWPOISON ||
1611 (!async && check_user_page_hwpoison(addr))) {
1612 pfn = KVM_PFN_ERR_HWPOISON;
1617 vma = find_vma_intersection(current->mm, addr, addr + 1);
1620 pfn = KVM_PFN_ERR_FAULT;
1621 else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
1622 r = hva_to_pfn_remapped(vma, addr, async, write_fault, writable, &pfn);
1626 pfn = KVM_PFN_ERR_FAULT;
1628 if (async && vma_is_valid(vma, write_fault))
1630 pfn = KVM_PFN_ERR_FAULT;
1633 up_read(¤t->mm->mmap_sem);
1637 kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn,
1638 bool atomic, bool *async, bool write_fault,
1641 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1643 if (addr == KVM_HVA_ERR_RO_BAD) {
1646 return KVM_PFN_ERR_RO_FAULT;
1649 if (kvm_is_error_hva(addr)) {
1652 return KVM_PFN_NOSLOT;
1655 /* Do not map writable pfn in the readonly memslot. */
1656 if (writable && memslot_is_readonly(slot)) {
1661 return hva_to_pfn(addr, atomic, async, write_fault,
1664 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
1666 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1669 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
1670 write_fault, writable);
1672 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1674 kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1676 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1678 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
1680 kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1682 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1684 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1686 kvm_pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1688 return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm, gfn), gfn);
1690 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1692 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
1694 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1696 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
1698 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1700 return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
1702 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1704 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
1706 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1708 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
1710 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1711 struct page **pages, int nr_pages)
1716 addr = gfn_to_hva_many(slot, gfn, &entry);
1717 if (kvm_is_error_hva(addr))
1720 if (entry < nr_pages)
1723 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1725 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1727 static struct page *kvm_pfn_to_page(kvm_pfn_t pfn)
1729 if (is_error_noslot_pfn(pfn))
1730 return KVM_ERR_PTR_BAD_PAGE;
1732 if (kvm_is_reserved_pfn(pfn)) {
1734 return KVM_ERR_PTR_BAD_PAGE;
1737 return pfn_to_page(pfn);
1740 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1744 pfn = gfn_to_pfn(kvm, gfn);
1746 return kvm_pfn_to_page(pfn);
1748 EXPORT_SYMBOL_GPL(gfn_to_page);
1750 static int __kvm_map_gfn(struct kvm_memory_slot *slot, gfn_t gfn,
1751 struct kvm_host_map *map)
1755 struct page *page = KVM_UNMAPPED_PAGE;
1760 pfn = gfn_to_pfn_memslot(slot, gfn);
1761 if (is_error_noslot_pfn(pfn))
1764 if (pfn_valid(pfn)) {
1765 page = pfn_to_page(pfn);
1767 #ifdef CONFIG_HAS_IOMEM
1769 hva = memremap(pfn_to_hpa(pfn), PAGE_SIZE, MEMREMAP_WB);
1784 int kvm_vcpu_map(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map)
1786 return __kvm_map_gfn(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, map);
1788 EXPORT_SYMBOL_GPL(kvm_vcpu_map);
1790 void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map,
1799 if (map->page != KVM_UNMAPPED_PAGE)
1801 #ifdef CONFIG_HAS_IOMEM
1807 kvm_vcpu_mark_page_dirty(vcpu, map->gfn);
1808 kvm_release_pfn_dirty(map->pfn);
1810 kvm_release_pfn_clean(map->pfn);
1816 EXPORT_SYMBOL_GPL(kvm_vcpu_unmap);
1818 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
1822 pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
1824 return kvm_pfn_to_page(pfn);
1826 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
1828 void kvm_release_page_clean(struct page *page)
1830 WARN_ON(is_error_page(page));
1832 kvm_release_pfn_clean(page_to_pfn(page));
1834 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1836 void kvm_release_pfn_clean(kvm_pfn_t pfn)
1838 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
1839 put_page(pfn_to_page(pfn));
1841 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1843 void kvm_release_page_dirty(struct page *page)
1845 WARN_ON(is_error_page(page));
1847 kvm_release_pfn_dirty(page_to_pfn(page));
1849 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1851 void kvm_release_pfn_dirty(kvm_pfn_t pfn)
1853 kvm_set_pfn_dirty(pfn);
1854 kvm_release_pfn_clean(pfn);
1856 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1858 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
1860 if (!kvm_is_reserved_pfn(pfn)) {
1861 struct page *page = pfn_to_page(pfn);
1866 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1868 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
1870 if (!kvm_is_reserved_pfn(pfn))
1871 mark_page_accessed(pfn_to_page(pfn));
1873 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1875 void kvm_get_pfn(kvm_pfn_t pfn)
1877 if (!kvm_is_reserved_pfn(pfn))
1878 get_page(pfn_to_page(pfn));
1880 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1882 static int next_segment(unsigned long len, int offset)
1884 if (len > PAGE_SIZE - offset)
1885 return PAGE_SIZE - offset;
1890 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
1891 void *data, int offset, int len)
1896 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1897 if (kvm_is_error_hva(addr))
1899 r = __copy_from_user(data, (void __user *)addr + offset, len);
1905 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1908 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1910 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1912 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1914 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
1915 int offset, int len)
1917 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1919 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1921 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
1923 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1925 gfn_t gfn = gpa >> PAGE_SHIFT;
1927 int offset = offset_in_page(gpa);
1930 while ((seg = next_segment(len, offset)) != 0) {
1931 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1941 EXPORT_SYMBOL_GPL(kvm_read_guest);
1943 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
1945 gfn_t gfn = gpa >> PAGE_SHIFT;
1947 int offset = offset_in_page(gpa);
1950 while ((seg = next_segment(len, offset)) != 0) {
1951 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
1961 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
1963 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1964 void *data, int offset, unsigned long len)
1969 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1970 if (kvm_is_error_hva(addr))
1972 pagefault_disable();
1973 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1980 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1983 gfn_t gfn = gpa >> PAGE_SHIFT;
1984 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1985 int offset = offset_in_page(gpa);
1987 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1989 EXPORT_SYMBOL_GPL(kvm_read_guest_atomic);
1991 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
1992 void *data, unsigned long len)
1994 gfn_t gfn = gpa >> PAGE_SHIFT;
1995 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1996 int offset = offset_in_page(gpa);
1998 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
2000 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
2002 static int __kvm_write_guest_page(struct kvm_memory_slot *memslot, gfn_t gfn,
2003 const void *data, int offset, int len)
2008 addr = gfn_to_hva_memslot(memslot, gfn);
2009 if (kvm_is_error_hva(addr))
2011 r = __copy_to_user((void __user *)addr + offset, data, len);
2014 mark_page_dirty_in_slot(memslot, gfn);
2018 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
2019 const void *data, int offset, int len)
2021 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2023 return __kvm_write_guest_page(slot, gfn, data, offset, len);
2025 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
2027 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
2028 const void *data, int offset, int len)
2030 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2032 return __kvm_write_guest_page(slot, gfn, data, offset, len);
2034 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
2036 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
2039 gfn_t gfn = gpa >> PAGE_SHIFT;
2041 int offset = offset_in_page(gpa);
2044 while ((seg = next_segment(len, offset)) != 0) {
2045 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
2055 EXPORT_SYMBOL_GPL(kvm_write_guest);
2057 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
2060 gfn_t gfn = gpa >> PAGE_SHIFT;
2062 int offset = offset_in_page(gpa);
2065 while ((seg = next_segment(len, offset)) != 0) {
2066 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
2076 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
2078 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
2079 struct gfn_to_hva_cache *ghc,
2080 gpa_t gpa, unsigned long len)
2082 int offset = offset_in_page(gpa);
2083 gfn_t start_gfn = gpa >> PAGE_SHIFT;
2084 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
2085 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
2086 gfn_t nr_pages_avail;
2087 int r = start_gfn <= end_gfn ? 0 : -EINVAL;
2090 ghc->generation = slots->generation;
2092 ghc->hva = KVM_HVA_ERR_BAD;
2095 * If the requested region crosses two memslots, we still
2096 * verify that the entire region is valid here.
2098 while (!r && start_gfn <= end_gfn) {
2099 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
2100 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
2102 if (kvm_is_error_hva(ghc->hva))
2104 start_gfn += nr_pages_avail;
2107 /* Use the slow path for cross page reads and writes. */
2108 if (!r && nr_pages_needed == 1)
2111 ghc->memslot = NULL;
2116 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2117 gpa_t gpa, unsigned long len)
2119 struct kvm_memslots *slots = kvm_memslots(kvm);
2120 return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len);
2122 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
2124 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2125 void *data, unsigned int offset,
2128 struct kvm_memslots *slots = kvm_memslots(kvm);
2130 gpa_t gpa = ghc->gpa + offset;
2132 BUG_ON(len + offset > ghc->len);
2134 if (slots->generation != ghc->generation)
2135 __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len);
2137 if (unlikely(!ghc->memslot))
2138 return kvm_write_guest(kvm, gpa, data, len);
2140 if (kvm_is_error_hva(ghc->hva))
2143 r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
2146 mark_page_dirty_in_slot(ghc->memslot, gpa >> PAGE_SHIFT);
2150 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached);
2152 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2153 void *data, unsigned long len)
2155 return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len);
2157 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
2159 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2160 void *data, unsigned long len)
2162 struct kvm_memslots *slots = kvm_memslots(kvm);
2165 BUG_ON(len > ghc->len);
2167 if (slots->generation != ghc->generation)
2168 __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len);
2170 if (unlikely(!ghc->memslot))
2171 return kvm_read_guest(kvm, ghc->gpa, data, len);
2173 if (kvm_is_error_hva(ghc->hva))
2176 r = __copy_from_user(data, (void __user *)ghc->hva, len);
2182 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
2184 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
2186 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
2188 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
2190 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
2192 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
2194 gfn_t gfn = gpa >> PAGE_SHIFT;
2196 int offset = offset_in_page(gpa);
2199 while ((seg = next_segment(len, offset)) != 0) {
2200 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
2209 EXPORT_SYMBOL_GPL(kvm_clear_guest);
2211 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot,
2214 if (memslot && memslot->dirty_bitmap) {
2215 unsigned long rel_gfn = gfn - memslot->base_gfn;
2217 set_bit_le(rel_gfn, memslot->dirty_bitmap);
2221 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
2223 struct kvm_memory_slot *memslot;
2225 memslot = gfn_to_memslot(kvm, gfn);
2226 mark_page_dirty_in_slot(memslot, gfn);
2228 EXPORT_SYMBOL_GPL(mark_page_dirty);
2230 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
2232 struct kvm_memory_slot *memslot;
2234 memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2235 mark_page_dirty_in_slot(memslot, gfn);
2237 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
2239 void kvm_sigset_activate(struct kvm_vcpu *vcpu)
2241 if (!vcpu->sigset_active)
2245 * This does a lockless modification of ->real_blocked, which is fine
2246 * because, only current can change ->real_blocked and all readers of
2247 * ->real_blocked don't care as long ->real_blocked is always a subset
2250 sigprocmask(SIG_SETMASK, &vcpu->sigset, ¤t->real_blocked);
2253 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu)
2255 if (!vcpu->sigset_active)
2258 sigprocmask(SIG_SETMASK, ¤t->real_blocked, NULL);
2259 sigemptyset(¤t->real_blocked);
2262 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
2264 unsigned int old, val, grow, grow_start;
2266 old = val = vcpu->halt_poll_ns;
2267 grow_start = READ_ONCE(halt_poll_ns_grow_start);
2268 grow = READ_ONCE(halt_poll_ns_grow);
2273 if (val < grow_start)
2276 if (val > halt_poll_ns)
2279 vcpu->halt_poll_ns = val;
2281 trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
2284 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
2286 unsigned int old, val, shrink;
2288 old = val = vcpu->halt_poll_ns;
2289 shrink = READ_ONCE(halt_poll_ns_shrink);
2295 vcpu->halt_poll_ns = val;
2296 trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
2299 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
2302 int idx = srcu_read_lock(&vcpu->kvm->srcu);
2304 if (kvm_arch_vcpu_runnable(vcpu)) {
2305 kvm_make_request(KVM_REQ_UNHALT, vcpu);
2308 if (kvm_cpu_has_pending_timer(vcpu))
2310 if (signal_pending(current))
2315 srcu_read_unlock(&vcpu->kvm->srcu, idx);
2320 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2322 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
2325 DECLARE_SWAITQUEUE(wait);
2326 bool waited = false;
2329 kvm_arch_vcpu_blocking(vcpu);
2331 start = cur = ktime_get();
2332 if (vcpu->halt_poll_ns && !kvm_arch_no_poll(vcpu)) {
2333 ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns);
2335 ++vcpu->stat.halt_attempted_poll;
2338 * This sets KVM_REQ_UNHALT if an interrupt
2341 if (kvm_vcpu_check_block(vcpu) < 0) {
2342 ++vcpu->stat.halt_successful_poll;
2343 if (!vcpu_valid_wakeup(vcpu))
2344 ++vcpu->stat.halt_poll_invalid;
2348 } while (single_task_running() && ktime_before(cur, stop));
2352 prepare_to_swait_exclusive(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
2354 if (kvm_vcpu_check_block(vcpu) < 0)
2361 finish_swait(&vcpu->wq, &wait);
2364 kvm_arch_vcpu_unblocking(vcpu);
2365 block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
2367 if (!kvm_arch_no_poll(vcpu)) {
2368 if (!vcpu_valid_wakeup(vcpu)) {
2369 shrink_halt_poll_ns(vcpu);
2370 } else if (halt_poll_ns) {
2371 if (block_ns <= vcpu->halt_poll_ns)
2373 /* we had a long block, shrink polling */
2374 else if (vcpu->halt_poll_ns && block_ns > halt_poll_ns)
2375 shrink_halt_poll_ns(vcpu);
2376 /* we had a short halt and our poll time is too small */
2377 else if (vcpu->halt_poll_ns < halt_poll_ns &&
2378 block_ns < halt_poll_ns)
2379 grow_halt_poll_ns(vcpu);
2381 vcpu->halt_poll_ns = 0;
2385 trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu));
2386 kvm_arch_vcpu_block_finish(vcpu);
2388 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
2390 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
2392 struct swait_queue_head *wqp;
2394 wqp = kvm_arch_vcpu_wq(vcpu);
2395 if (swq_has_sleeper(wqp)) {
2397 WRITE_ONCE(vcpu->ready, true);
2398 ++vcpu->stat.halt_wakeup;
2404 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
2408 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2410 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
2413 int cpu = vcpu->cpu;
2415 if (kvm_vcpu_wake_up(vcpu))
2419 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
2420 if (kvm_arch_vcpu_should_kick(vcpu))
2421 smp_send_reschedule(cpu);
2424 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
2425 #endif /* !CONFIG_S390 */
2427 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
2430 struct task_struct *task = NULL;
2434 pid = rcu_dereference(target->pid);
2436 task = get_pid_task(pid, PIDTYPE_PID);
2440 ret = yield_to(task, 1);
2441 put_task_struct(task);
2445 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
2448 * Helper that checks whether a VCPU is eligible for directed yield.
2449 * Most eligible candidate to yield is decided by following heuristics:
2451 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2452 * (preempted lock holder), indicated by @in_spin_loop.
2453 * Set at the beiginning and cleared at the end of interception/PLE handler.
2455 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2456 * chance last time (mostly it has become eligible now since we have probably
2457 * yielded to lockholder in last iteration. This is done by toggling
2458 * @dy_eligible each time a VCPU checked for eligibility.)
2460 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2461 * to preempted lock-holder could result in wrong VCPU selection and CPU
2462 * burning. Giving priority for a potential lock-holder increases lock
2465 * Since algorithm is based on heuristics, accessing another VCPU data without
2466 * locking does not harm. It may result in trying to yield to same VCPU, fail
2467 * and continue with next VCPU and so on.
2469 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
2471 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2474 eligible = !vcpu->spin_loop.in_spin_loop ||
2475 vcpu->spin_loop.dy_eligible;
2477 if (vcpu->spin_loop.in_spin_loop)
2478 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
2487 * Unlike kvm_arch_vcpu_runnable, this function is called outside
2488 * a vcpu_load/vcpu_put pair. However, for most architectures
2489 * kvm_arch_vcpu_runnable does not require vcpu_load.
2491 bool __weak kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
2493 return kvm_arch_vcpu_runnable(vcpu);
2496 static bool vcpu_dy_runnable(struct kvm_vcpu *vcpu)
2498 if (kvm_arch_dy_runnable(vcpu))
2501 #ifdef CONFIG_KVM_ASYNC_PF
2502 if (!list_empty_careful(&vcpu->async_pf.done))
2509 void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode)
2511 struct kvm *kvm = me->kvm;
2512 struct kvm_vcpu *vcpu;
2513 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
2519 kvm_vcpu_set_in_spin_loop(me, true);
2521 * We boost the priority of a VCPU that is runnable but not
2522 * currently running, because it got preempted by something
2523 * else and called schedule in __vcpu_run. Hopefully that
2524 * VCPU is holding the lock that we need and will release it.
2525 * We approximate round-robin by starting at the last boosted VCPU.
2527 for (pass = 0; pass < 2 && !yielded && try; pass++) {
2528 kvm_for_each_vcpu(i, vcpu, kvm) {
2529 if (!pass && i <= last_boosted_vcpu) {
2530 i = last_boosted_vcpu;
2532 } else if (pass && i > last_boosted_vcpu)
2534 if (!READ_ONCE(vcpu->ready))
2538 if (swait_active(&vcpu->wq) && !vcpu_dy_runnable(vcpu))
2540 if (READ_ONCE(vcpu->preempted) && yield_to_kernel_mode &&
2541 !kvm_arch_vcpu_in_kernel(vcpu))
2543 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
2546 yielded = kvm_vcpu_yield_to(vcpu);
2548 kvm->last_boosted_vcpu = i;
2550 } else if (yielded < 0) {
2557 kvm_vcpu_set_in_spin_loop(me, false);
2559 /* Ensure vcpu is not eligible during next spinloop */
2560 kvm_vcpu_set_dy_eligible(me, false);
2562 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
2564 static vm_fault_t kvm_vcpu_fault(struct vm_fault *vmf)
2566 struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data;
2569 if (vmf->pgoff == 0)
2570 page = virt_to_page(vcpu->run);
2572 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
2573 page = virt_to_page(vcpu->arch.pio_data);
2575 #ifdef CONFIG_KVM_MMIO
2576 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
2577 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
2580 return kvm_arch_vcpu_fault(vcpu, vmf);
2586 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
2587 .fault = kvm_vcpu_fault,
2590 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2592 vma->vm_ops = &kvm_vcpu_vm_ops;
2596 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2598 struct kvm_vcpu *vcpu = filp->private_data;
2600 debugfs_remove_recursive(vcpu->debugfs_dentry);
2601 kvm_put_kvm(vcpu->kvm);
2605 static struct file_operations kvm_vcpu_fops = {
2606 .release = kvm_vcpu_release,
2607 .unlocked_ioctl = kvm_vcpu_ioctl,
2608 .mmap = kvm_vcpu_mmap,
2609 .llseek = noop_llseek,
2610 KVM_COMPAT(kvm_vcpu_compat_ioctl),
2614 * Allocates an inode for the vcpu.
2616 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2618 char name[8 + 1 + ITOA_MAX_LEN + 1];
2620 snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id);
2621 return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
2624 static void kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
2626 #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS
2627 char dir_name[ITOA_MAX_LEN * 2];
2629 if (!debugfs_initialized())
2632 snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
2633 vcpu->debugfs_dentry = debugfs_create_dir(dir_name,
2634 vcpu->kvm->debugfs_dentry);
2636 kvm_arch_create_vcpu_debugfs(vcpu);
2641 * Creates some virtual cpus. Good luck creating more than one.
2643 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
2646 struct kvm_vcpu *vcpu;
2648 if (id >= KVM_MAX_VCPU_ID)
2651 mutex_lock(&kvm->lock);
2652 if (kvm->created_vcpus == KVM_MAX_VCPUS) {
2653 mutex_unlock(&kvm->lock);
2657 kvm->created_vcpus++;
2658 mutex_unlock(&kvm->lock);
2660 vcpu = kvm_arch_vcpu_create(kvm, id);
2663 goto vcpu_decrement;
2666 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2668 r = kvm_arch_vcpu_setup(vcpu);
2672 kvm_create_vcpu_debugfs(vcpu);
2674 mutex_lock(&kvm->lock);
2675 if (kvm_get_vcpu_by_id(kvm, id)) {
2677 goto unlock_vcpu_destroy;
2680 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
2682 /* Now it's all set up, let userspace reach it */
2684 r = create_vcpu_fd(vcpu);
2687 goto unlock_vcpu_destroy;
2690 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
2693 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
2694 * before kvm->online_vcpu's incremented value.
2697 atomic_inc(&kvm->online_vcpus);
2699 mutex_unlock(&kvm->lock);
2700 kvm_arch_vcpu_postcreate(vcpu);
2703 unlock_vcpu_destroy:
2704 mutex_unlock(&kvm->lock);
2705 debugfs_remove_recursive(vcpu->debugfs_dentry);
2707 kvm_arch_vcpu_destroy(vcpu);
2709 mutex_lock(&kvm->lock);
2710 kvm->created_vcpus--;
2711 mutex_unlock(&kvm->lock);
2715 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2718 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2719 vcpu->sigset_active = 1;
2720 vcpu->sigset = *sigset;
2722 vcpu->sigset_active = 0;
2726 static long kvm_vcpu_ioctl(struct file *filp,
2727 unsigned int ioctl, unsigned long arg)
2729 struct kvm_vcpu *vcpu = filp->private_data;
2730 void __user *argp = (void __user *)arg;
2732 struct kvm_fpu *fpu = NULL;
2733 struct kvm_sregs *kvm_sregs = NULL;
2735 if (vcpu->kvm->mm != current->mm)
2738 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2742 * Some architectures have vcpu ioctls that are asynchronous to vcpu
2743 * execution; mutex_lock() would break them.
2745 r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg);
2746 if (r != -ENOIOCTLCMD)
2749 if (mutex_lock_killable(&vcpu->mutex))
2757 oldpid = rcu_access_pointer(vcpu->pid);
2758 if (unlikely(oldpid != task_pid(current))) {
2759 /* The thread running this VCPU changed. */
2762 r = kvm_arch_vcpu_run_pid_change(vcpu);
2766 newpid = get_task_pid(current, PIDTYPE_PID);
2767 rcu_assign_pointer(vcpu->pid, newpid);
2772 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2773 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2776 case KVM_GET_REGS: {
2777 struct kvm_regs *kvm_regs;
2780 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL_ACCOUNT);
2783 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2787 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2794 case KVM_SET_REGS: {
2795 struct kvm_regs *kvm_regs;
2798 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2799 if (IS_ERR(kvm_regs)) {
2800 r = PTR_ERR(kvm_regs);
2803 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2807 case KVM_GET_SREGS: {
2808 kvm_sregs = kzalloc(sizeof(struct kvm_sregs),
2809 GFP_KERNEL_ACCOUNT);
2813 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2817 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2822 case KVM_SET_SREGS: {
2823 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2824 if (IS_ERR(kvm_sregs)) {
2825 r = PTR_ERR(kvm_sregs);
2829 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2832 case KVM_GET_MP_STATE: {
2833 struct kvm_mp_state mp_state;
2835 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2839 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
2844 case KVM_SET_MP_STATE: {
2845 struct kvm_mp_state mp_state;
2848 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
2850 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2853 case KVM_TRANSLATE: {
2854 struct kvm_translation tr;
2857 if (copy_from_user(&tr, argp, sizeof(tr)))
2859 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2863 if (copy_to_user(argp, &tr, sizeof(tr)))
2868 case KVM_SET_GUEST_DEBUG: {
2869 struct kvm_guest_debug dbg;
2872 if (copy_from_user(&dbg, argp, sizeof(dbg)))
2874 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2877 case KVM_SET_SIGNAL_MASK: {
2878 struct kvm_signal_mask __user *sigmask_arg = argp;
2879 struct kvm_signal_mask kvm_sigmask;
2880 sigset_t sigset, *p;
2885 if (copy_from_user(&kvm_sigmask, argp,
2886 sizeof(kvm_sigmask)))
2889 if (kvm_sigmask.len != sizeof(sigset))
2892 if (copy_from_user(&sigset, sigmask_arg->sigset,
2897 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2901 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL_ACCOUNT);
2905 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2909 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2915 fpu = memdup_user(argp, sizeof(*fpu));
2921 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2925 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2928 mutex_unlock(&vcpu->mutex);
2934 #ifdef CONFIG_KVM_COMPAT
2935 static long kvm_vcpu_compat_ioctl(struct file *filp,
2936 unsigned int ioctl, unsigned long arg)
2938 struct kvm_vcpu *vcpu = filp->private_data;
2939 void __user *argp = compat_ptr(arg);
2942 if (vcpu->kvm->mm != current->mm)
2946 case KVM_SET_SIGNAL_MASK: {
2947 struct kvm_signal_mask __user *sigmask_arg = argp;
2948 struct kvm_signal_mask kvm_sigmask;
2953 if (copy_from_user(&kvm_sigmask, argp,
2954 sizeof(kvm_sigmask)))
2957 if (kvm_sigmask.len != sizeof(compat_sigset_t))
2960 if (get_compat_sigset(&sigset, (void *)sigmask_arg->sigset))
2962 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2964 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2968 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2976 static int kvm_device_mmap(struct file *filp, struct vm_area_struct *vma)
2978 struct kvm_device *dev = filp->private_data;
2981 return dev->ops->mmap(dev, vma);
2986 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2987 int (*accessor)(struct kvm_device *dev,
2988 struct kvm_device_attr *attr),
2991 struct kvm_device_attr attr;
2996 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2999 return accessor(dev, &attr);
3002 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
3005 struct kvm_device *dev = filp->private_data;
3007 if (dev->kvm->mm != current->mm)
3011 case KVM_SET_DEVICE_ATTR:
3012 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
3013 case KVM_GET_DEVICE_ATTR:
3014 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
3015 case KVM_HAS_DEVICE_ATTR:
3016 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
3018 if (dev->ops->ioctl)
3019 return dev->ops->ioctl(dev, ioctl, arg);
3025 static int kvm_device_release(struct inode *inode, struct file *filp)
3027 struct kvm_device *dev = filp->private_data;
3028 struct kvm *kvm = dev->kvm;
3030 if (dev->ops->release) {
3031 mutex_lock(&kvm->lock);
3032 list_del(&dev->vm_node);
3033 dev->ops->release(dev);
3034 mutex_unlock(&kvm->lock);
3041 static const struct file_operations kvm_device_fops = {
3042 .unlocked_ioctl = kvm_device_ioctl,
3043 .release = kvm_device_release,
3044 KVM_COMPAT(kvm_device_ioctl),
3045 .mmap = kvm_device_mmap,
3048 struct kvm_device *kvm_device_from_filp(struct file *filp)
3050 if (filp->f_op != &kvm_device_fops)
3053 return filp->private_data;
3056 static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
3057 #ifdef CONFIG_KVM_MPIC
3058 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
3059 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
3063 int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
3065 if (type >= ARRAY_SIZE(kvm_device_ops_table))
3068 if (kvm_device_ops_table[type] != NULL)
3071 kvm_device_ops_table[type] = ops;
3075 void kvm_unregister_device_ops(u32 type)
3077 if (kvm_device_ops_table[type] != NULL)
3078 kvm_device_ops_table[type] = NULL;
3081 static int kvm_ioctl_create_device(struct kvm *kvm,
3082 struct kvm_create_device *cd)
3084 struct kvm_device_ops *ops = NULL;
3085 struct kvm_device *dev;
3086 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
3090 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
3093 type = array_index_nospec(cd->type, ARRAY_SIZE(kvm_device_ops_table));
3094 ops = kvm_device_ops_table[type];
3101 dev = kzalloc(sizeof(*dev), GFP_KERNEL_ACCOUNT);
3108 mutex_lock(&kvm->lock);
3109 ret = ops->create(dev, type);
3111 mutex_unlock(&kvm->lock);
3115 list_add(&dev->vm_node, &kvm->devices);
3116 mutex_unlock(&kvm->lock);
3122 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
3125 mutex_lock(&kvm->lock);
3126 list_del(&dev->vm_node);
3127 mutex_unlock(&kvm->lock);
3136 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
3139 case KVM_CAP_USER_MEMORY:
3140 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
3141 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
3142 case KVM_CAP_INTERNAL_ERROR_DATA:
3143 #ifdef CONFIG_HAVE_KVM_MSI
3144 case KVM_CAP_SIGNAL_MSI:
3146 #ifdef CONFIG_HAVE_KVM_IRQFD
3148 case KVM_CAP_IRQFD_RESAMPLE:
3150 case KVM_CAP_IOEVENTFD_ANY_LENGTH:
3151 case KVM_CAP_CHECK_EXTENSION_VM:
3152 case KVM_CAP_ENABLE_CAP_VM:
3153 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3154 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2:
3157 #ifdef CONFIG_KVM_MMIO
3158 case KVM_CAP_COALESCED_MMIO:
3159 return KVM_COALESCED_MMIO_PAGE_OFFSET;
3160 case KVM_CAP_COALESCED_PIO:
3163 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3164 case KVM_CAP_IRQ_ROUTING:
3165 return KVM_MAX_IRQ_ROUTES;
3167 #if KVM_ADDRESS_SPACE_NUM > 1
3168 case KVM_CAP_MULTI_ADDRESS_SPACE:
3169 return KVM_ADDRESS_SPACE_NUM;
3171 case KVM_CAP_NR_MEMSLOTS:
3172 return KVM_USER_MEM_SLOTS;
3176 return kvm_vm_ioctl_check_extension(kvm, arg);
3179 int __attribute__((weak)) kvm_vm_ioctl_enable_cap(struct kvm *kvm,
3180 struct kvm_enable_cap *cap)
3185 static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm,
3186 struct kvm_enable_cap *cap)
3189 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3190 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2:
3191 if (cap->flags || (cap->args[0] & ~1))
3193 kvm->manual_dirty_log_protect = cap->args[0];
3197 return kvm_vm_ioctl_enable_cap(kvm, cap);
3201 static long kvm_vm_ioctl(struct file *filp,
3202 unsigned int ioctl, unsigned long arg)
3204 struct kvm *kvm = filp->private_data;
3205 void __user *argp = (void __user *)arg;
3208 if (kvm->mm != current->mm)
3211 case KVM_CREATE_VCPU:
3212 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
3214 case KVM_ENABLE_CAP: {
3215 struct kvm_enable_cap cap;
3218 if (copy_from_user(&cap, argp, sizeof(cap)))
3220 r = kvm_vm_ioctl_enable_cap_generic(kvm, &cap);
3223 case KVM_SET_USER_MEMORY_REGION: {
3224 struct kvm_userspace_memory_region kvm_userspace_mem;
3227 if (copy_from_user(&kvm_userspace_mem, argp,
3228 sizeof(kvm_userspace_mem)))
3231 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
3234 case KVM_GET_DIRTY_LOG: {
3235 struct kvm_dirty_log log;
3238 if (copy_from_user(&log, argp, sizeof(log)))
3240 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3243 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3244 case KVM_CLEAR_DIRTY_LOG: {
3245 struct kvm_clear_dirty_log log;
3248 if (copy_from_user(&log, argp, sizeof(log)))
3250 r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
3254 #ifdef CONFIG_KVM_MMIO
3255 case KVM_REGISTER_COALESCED_MMIO: {
3256 struct kvm_coalesced_mmio_zone zone;
3259 if (copy_from_user(&zone, argp, sizeof(zone)))
3261 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
3264 case KVM_UNREGISTER_COALESCED_MMIO: {
3265 struct kvm_coalesced_mmio_zone zone;
3268 if (copy_from_user(&zone, argp, sizeof(zone)))
3270 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
3275 struct kvm_irqfd data;
3278 if (copy_from_user(&data, argp, sizeof(data)))
3280 r = kvm_irqfd(kvm, &data);
3283 case KVM_IOEVENTFD: {
3284 struct kvm_ioeventfd data;
3287 if (copy_from_user(&data, argp, sizeof(data)))
3289 r = kvm_ioeventfd(kvm, &data);
3292 #ifdef CONFIG_HAVE_KVM_MSI
3293 case KVM_SIGNAL_MSI: {
3297 if (copy_from_user(&msi, argp, sizeof(msi)))
3299 r = kvm_send_userspace_msi(kvm, &msi);
3303 #ifdef __KVM_HAVE_IRQ_LINE
3304 case KVM_IRQ_LINE_STATUS:
3305 case KVM_IRQ_LINE: {
3306 struct kvm_irq_level irq_event;
3309 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
3312 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
3313 ioctl == KVM_IRQ_LINE_STATUS);
3318 if (ioctl == KVM_IRQ_LINE_STATUS) {
3319 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
3327 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3328 case KVM_SET_GSI_ROUTING: {
3329 struct kvm_irq_routing routing;
3330 struct kvm_irq_routing __user *urouting;
3331 struct kvm_irq_routing_entry *entries = NULL;
3334 if (copy_from_user(&routing, argp, sizeof(routing)))
3337 if (!kvm_arch_can_set_irq_routing(kvm))
3339 if (routing.nr > KVM_MAX_IRQ_ROUTES)
3345 entries = vmalloc(array_size(sizeof(*entries),
3351 if (copy_from_user(entries, urouting->entries,
3352 routing.nr * sizeof(*entries)))
3353 goto out_free_irq_routing;
3355 r = kvm_set_irq_routing(kvm, entries, routing.nr,
3357 out_free_irq_routing:
3361 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
3362 case KVM_CREATE_DEVICE: {
3363 struct kvm_create_device cd;
3366 if (copy_from_user(&cd, argp, sizeof(cd)))
3369 r = kvm_ioctl_create_device(kvm, &cd);
3374 if (copy_to_user(argp, &cd, sizeof(cd)))
3380 case KVM_CHECK_EXTENSION:
3381 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
3384 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
3390 #ifdef CONFIG_KVM_COMPAT
3391 struct compat_kvm_dirty_log {
3395 compat_uptr_t dirty_bitmap; /* one bit per page */
3400 static long kvm_vm_compat_ioctl(struct file *filp,
3401 unsigned int ioctl, unsigned long arg)
3403 struct kvm *kvm = filp->private_data;
3406 if (kvm->mm != current->mm)
3409 case KVM_GET_DIRTY_LOG: {
3410 struct compat_kvm_dirty_log compat_log;
3411 struct kvm_dirty_log log;
3413 if (copy_from_user(&compat_log, (void __user *)arg,
3414 sizeof(compat_log)))
3416 log.slot = compat_log.slot;
3417 log.padding1 = compat_log.padding1;
3418 log.padding2 = compat_log.padding2;
3419 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
3421 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3425 r = kvm_vm_ioctl(filp, ioctl, arg);
3431 static struct file_operations kvm_vm_fops = {
3432 .release = kvm_vm_release,
3433 .unlocked_ioctl = kvm_vm_ioctl,
3434 .llseek = noop_llseek,
3435 KVM_COMPAT(kvm_vm_compat_ioctl),
3438 static int kvm_dev_ioctl_create_vm(unsigned long type)
3444 kvm = kvm_create_vm(type);
3446 return PTR_ERR(kvm);
3447 #ifdef CONFIG_KVM_MMIO
3448 r = kvm_coalesced_mmio_init(kvm);
3452 r = get_unused_fd_flags(O_CLOEXEC);
3456 file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
3464 * Don't call kvm_put_kvm anymore at this point; file->f_op is
3465 * already set, with ->release() being kvm_vm_release(). In error
3466 * cases it will be called by the final fput(file) and will take
3467 * care of doing kvm_put_kvm(kvm).
3469 if (kvm_create_vm_debugfs(kvm, r) < 0) {
3474 kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm);
3476 fd_install(r, file);
3484 static long kvm_dev_ioctl(struct file *filp,
3485 unsigned int ioctl, unsigned long arg)
3490 case KVM_GET_API_VERSION:
3493 r = KVM_API_VERSION;
3496 r = kvm_dev_ioctl_create_vm(arg);
3498 case KVM_CHECK_EXTENSION:
3499 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
3501 case KVM_GET_VCPU_MMAP_SIZE:
3504 r = PAGE_SIZE; /* struct kvm_run */
3506 r += PAGE_SIZE; /* pio data page */
3508 #ifdef CONFIG_KVM_MMIO
3509 r += PAGE_SIZE; /* coalesced mmio ring page */
3512 case KVM_TRACE_ENABLE:
3513 case KVM_TRACE_PAUSE:
3514 case KVM_TRACE_DISABLE:
3518 return kvm_arch_dev_ioctl(filp, ioctl, arg);
3524 static struct file_operations kvm_chardev_ops = {
3525 .unlocked_ioctl = kvm_dev_ioctl,
3526 .llseek = noop_llseek,
3527 KVM_COMPAT(kvm_dev_ioctl),
3530 static struct miscdevice kvm_dev = {
3536 static void hardware_enable_nolock(void *junk)
3538 int cpu = raw_smp_processor_id();
3541 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
3544 cpumask_set_cpu(cpu, cpus_hardware_enabled);
3546 r = kvm_arch_hardware_enable();
3549 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3550 atomic_inc(&hardware_enable_failed);
3551 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
3555 static int kvm_starting_cpu(unsigned int cpu)
3557 raw_spin_lock(&kvm_count_lock);
3558 if (kvm_usage_count)
3559 hardware_enable_nolock(NULL);
3560 raw_spin_unlock(&kvm_count_lock);
3564 static void hardware_disable_nolock(void *junk)
3566 int cpu = raw_smp_processor_id();
3568 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
3570 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3571 kvm_arch_hardware_disable();
3574 static int kvm_dying_cpu(unsigned int cpu)
3576 raw_spin_lock(&kvm_count_lock);
3577 if (kvm_usage_count)
3578 hardware_disable_nolock(NULL);
3579 raw_spin_unlock(&kvm_count_lock);
3583 static void hardware_disable_all_nolock(void)
3585 BUG_ON(!kvm_usage_count);
3588 if (!kvm_usage_count)
3589 on_each_cpu(hardware_disable_nolock, NULL, 1);
3592 static void hardware_disable_all(void)
3594 raw_spin_lock(&kvm_count_lock);
3595 hardware_disable_all_nolock();
3596 raw_spin_unlock(&kvm_count_lock);
3599 static int hardware_enable_all(void)
3603 raw_spin_lock(&kvm_count_lock);
3606 if (kvm_usage_count == 1) {
3607 atomic_set(&hardware_enable_failed, 0);
3608 on_each_cpu(hardware_enable_nolock, NULL, 1);
3610 if (atomic_read(&hardware_enable_failed)) {
3611 hardware_disable_all_nolock();
3616 raw_spin_unlock(&kvm_count_lock);
3621 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3625 * Some (well, at least mine) BIOSes hang on reboot if
3628 * And Intel TXT required VMX off for all cpu when system shutdown.
3630 pr_info("kvm: exiting hardware virtualization\n");
3631 kvm_rebooting = true;
3632 on_each_cpu(hardware_disable_nolock, NULL, 1);
3636 static struct notifier_block kvm_reboot_notifier = {
3637 .notifier_call = kvm_reboot,
3641 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3645 for (i = 0; i < bus->dev_count; i++) {
3646 struct kvm_io_device *pos = bus->range[i].dev;
3648 kvm_iodevice_destructor(pos);
3653 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
3654 const struct kvm_io_range *r2)
3656 gpa_t addr1 = r1->addr;
3657 gpa_t addr2 = r2->addr;
3662 /* If r2->len == 0, match the exact address. If r2->len != 0,
3663 * accept any overlapping write. Any order is acceptable for
3664 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
3665 * we process all of them.
3678 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
3680 return kvm_io_bus_cmp(p1, p2);
3683 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
3684 gpa_t addr, int len)
3686 struct kvm_io_range *range, key;
3689 key = (struct kvm_io_range) {
3694 range = bsearch(&key, bus->range, bus->dev_count,
3695 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
3699 off = range - bus->range;
3701 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
3707 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3708 struct kvm_io_range *range, const void *val)
3712 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3716 while (idx < bus->dev_count &&
3717 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3718 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
3727 /* kvm_io_bus_write - called under kvm->slots_lock */
3728 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3729 int len, const void *val)
3731 struct kvm_io_bus *bus;
3732 struct kvm_io_range range;
3735 range = (struct kvm_io_range) {
3740 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3743 r = __kvm_io_bus_write(vcpu, bus, &range, val);
3744 return r < 0 ? r : 0;
3746 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
3748 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3749 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
3750 gpa_t addr, int len, const void *val, long cookie)
3752 struct kvm_io_bus *bus;
3753 struct kvm_io_range range;
3755 range = (struct kvm_io_range) {
3760 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3764 /* First try the device referenced by cookie. */
3765 if ((cookie >= 0) && (cookie < bus->dev_count) &&
3766 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
3767 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
3772 * cookie contained garbage; fall back to search and return the
3773 * correct cookie value.
3775 return __kvm_io_bus_write(vcpu, bus, &range, val);
3778 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3779 struct kvm_io_range *range, void *val)
3783 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3787 while (idx < bus->dev_count &&
3788 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3789 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
3798 /* kvm_io_bus_read - called under kvm->slots_lock */
3799 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3802 struct kvm_io_bus *bus;
3803 struct kvm_io_range range;
3806 range = (struct kvm_io_range) {
3811 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3814 r = __kvm_io_bus_read(vcpu, bus, &range, val);
3815 return r < 0 ? r : 0;
3818 /* Caller must hold slots_lock. */
3819 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3820 int len, struct kvm_io_device *dev)
3823 struct kvm_io_bus *new_bus, *bus;
3824 struct kvm_io_range range;
3826 bus = kvm_get_bus(kvm, bus_idx);
3830 /* exclude ioeventfd which is limited by maximum fd */
3831 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3834 new_bus = kmalloc(struct_size(bus, range, bus->dev_count + 1),
3835 GFP_KERNEL_ACCOUNT);
3839 range = (struct kvm_io_range) {
3845 for (i = 0; i < bus->dev_count; i++)
3846 if (kvm_io_bus_cmp(&bus->range[i], &range) > 0)
3849 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3850 new_bus->dev_count++;
3851 new_bus->range[i] = range;
3852 memcpy(new_bus->range + i + 1, bus->range + i,
3853 (bus->dev_count - i) * sizeof(struct kvm_io_range));
3854 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3855 synchronize_srcu_expedited(&kvm->srcu);
3861 /* Caller must hold slots_lock. */
3862 void kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3863 struct kvm_io_device *dev)
3866 struct kvm_io_bus *new_bus, *bus;
3868 bus = kvm_get_bus(kvm, bus_idx);
3872 for (i = 0; i < bus->dev_count; i++)
3873 if (bus->range[i].dev == dev) {
3877 if (i == bus->dev_count)
3880 new_bus = kmalloc(struct_size(bus, range, bus->dev_count - 1),
3881 GFP_KERNEL_ACCOUNT);
3883 pr_err("kvm: failed to shrink bus, removing it completely\n");
3887 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3888 new_bus->dev_count--;
3889 memcpy(new_bus->range + i, bus->range + i + 1,
3890 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3893 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3894 synchronize_srcu_expedited(&kvm->srcu);
3899 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3902 struct kvm_io_bus *bus;
3903 int dev_idx, srcu_idx;
3904 struct kvm_io_device *iodev = NULL;
3906 srcu_idx = srcu_read_lock(&kvm->srcu);
3908 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
3912 dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
3916 iodev = bus->range[dev_idx].dev;
3919 srcu_read_unlock(&kvm->srcu, srcu_idx);
3923 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
3925 static int kvm_debugfs_open(struct inode *inode, struct file *file,
3926 int (*get)(void *, u64 *), int (*set)(void *, u64),
3929 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
3932 /* The debugfs files are a reference to the kvm struct which
3933 * is still valid when kvm_destroy_vm is called.
3934 * To avoid the race between open and the removal of the debugfs
3935 * directory we test against the users count.
3937 if (!refcount_inc_not_zero(&stat_data->kvm->users_count))
3940 if (simple_attr_open(inode, file, get,
3941 stat_data->mode & S_IWUGO ? set : NULL,
3943 kvm_put_kvm(stat_data->kvm);
3950 static int kvm_debugfs_release(struct inode *inode, struct file *file)
3952 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
3955 simple_attr_release(inode, file);
3956 kvm_put_kvm(stat_data->kvm);
3961 static int vm_stat_get_per_vm(void *data, u64 *val)
3963 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3965 *val = *(ulong *)((void *)stat_data->kvm + stat_data->offset);
3970 static int vm_stat_clear_per_vm(void *data, u64 val)
3972 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3977 *(ulong *)((void *)stat_data->kvm + stat_data->offset) = 0;
3982 static int vm_stat_get_per_vm_open(struct inode *inode, struct file *file)
3984 __simple_attr_check_format("%llu\n", 0ull);
3985 return kvm_debugfs_open(inode, file, vm_stat_get_per_vm,
3986 vm_stat_clear_per_vm, "%llu\n");
3989 static const struct file_operations vm_stat_get_per_vm_fops = {
3990 .owner = THIS_MODULE,
3991 .open = vm_stat_get_per_vm_open,
3992 .release = kvm_debugfs_release,
3993 .read = simple_attr_read,
3994 .write = simple_attr_write,
3995 .llseek = no_llseek,
3998 static int vcpu_stat_get_per_vm(void *data, u64 *val)
4001 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
4002 struct kvm_vcpu *vcpu;
4006 kvm_for_each_vcpu(i, vcpu, stat_data->kvm)
4007 *val += *(u64 *)((void *)vcpu + stat_data->offset);
4012 static int vcpu_stat_clear_per_vm(void *data, u64 val)
4015 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
4016 struct kvm_vcpu *vcpu;
4021 kvm_for_each_vcpu(i, vcpu, stat_data->kvm)
4022 *(u64 *)((void *)vcpu + stat_data->offset) = 0;
4027 static int vcpu_stat_get_per_vm_open(struct inode *inode, struct file *file)
4029 __simple_attr_check_format("%llu\n", 0ull);
4030 return kvm_debugfs_open(inode, file, vcpu_stat_get_per_vm,
4031 vcpu_stat_clear_per_vm, "%llu\n");
4034 static const struct file_operations vcpu_stat_get_per_vm_fops = {
4035 .owner = THIS_MODULE,
4036 .open = vcpu_stat_get_per_vm_open,
4037 .release = kvm_debugfs_release,
4038 .read = simple_attr_read,
4039 .write = simple_attr_write,
4040 .llseek = no_llseek,
4043 static const struct file_operations *stat_fops_per_vm[] = {
4044 [KVM_STAT_VCPU] = &vcpu_stat_get_per_vm_fops,
4045 [KVM_STAT_VM] = &vm_stat_get_per_vm_fops,
4048 static int vm_stat_get(void *_offset, u64 *val)
4050 unsigned offset = (long)_offset;
4052 struct kvm_stat_data stat_tmp = {.offset = offset};
4056 mutex_lock(&kvm_lock);
4057 list_for_each_entry(kvm, &vm_list, vm_list) {
4059 vm_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
4062 mutex_unlock(&kvm_lock);
4066 static int vm_stat_clear(void *_offset, u64 val)
4068 unsigned offset = (long)_offset;
4070 struct kvm_stat_data stat_tmp = {.offset = offset};
4075 mutex_lock(&kvm_lock);
4076 list_for_each_entry(kvm, &vm_list, vm_list) {
4078 vm_stat_clear_per_vm((void *)&stat_tmp, 0);
4080 mutex_unlock(&kvm_lock);
4085 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n");
4087 static int vcpu_stat_get(void *_offset, u64 *val)
4089 unsigned offset = (long)_offset;
4091 struct kvm_stat_data stat_tmp = {.offset = offset};
4095 mutex_lock(&kvm_lock);
4096 list_for_each_entry(kvm, &vm_list, vm_list) {
4098 vcpu_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
4101 mutex_unlock(&kvm_lock);
4105 static int vcpu_stat_clear(void *_offset, u64 val)
4107 unsigned offset = (long)_offset;
4109 struct kvm_stat_data stat_tmp = {.offset = offset};
4114 mutex_lock(&kvm_lock);
4115 list_for_each_entry(kvm, &vm_list, vm_list) {
4117 vcpu_stat_clear_per_vm((void *)&stat_tmp, 0);
4119 mutex_unlock(&kvm_lock);
4124 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear,
4127 static const struct file_operations *stat_fops[] = {
4128 [KVM_STAT_VCPU] = &vcpu_stat_fops,
4129 [KVM_STAT_VM] = &vm_stat_fops,
4132 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm)
4134 struct kobj_uevent_env *env;
4135 unsigned long long created, active;
4137 if (!kvm_dev.this_device || !kvm)
4140 mutex_lock(&kvm_lock);
4141 if (type == KVM_EVENT_CREATE_VM) {
4142 kvm_createvm_count++;
4144 } else if (type == KVM_EVENT_DESTROY_VM) {
4147 created = kvm_createvm_count;
4148 active = kvm_active_vms;
4149 mutex_unlock(&kvm_lock);
4151 env = kzalloc(sizeof(*env), GFP_KERNEL_ACCOUNT);
4155 add_uevent_var(env, "CREATED=%llu", created);
4156 add_uevent_var(env, "COUNT=%llu", active);
4158 if (type == KVM_EVENT_CREATE_VM) {
4159 add_uevent_var(env, "EVENT=create");
4160 kvm->userspace_pid = task_pid_nr(current);
4161 } else if (type == KVM_EVENT_DESTROY_VM) {
4162 add_uevent_var(env, "EVENT=destroy");
4164 add_uevent_var(env, "PID=%d", kvm->userspace_pid);
4166 if (!IS_ERR_OR_NULL(kvm->debugfs_dentry)) {
4167 char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL_ACCOUNT);
4170 tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX);
4172 add_uevent_var(env, "STATS_PATH=%s", tmp);
4176 /* no need for checks, since we are adding at most only 5 keys */
4177 env->envp[env->envp_idx++] = NULL;
4178 kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp);
4182 static void kvm_init_debug(void)
4184 struct kvm_stats_debugfs_item *p;
4186 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
4188 kvm_debugfs_num_entries = 0;
4189 for (p = debugfs_entries; p->name; ++p, kvm_debugfs_num_entries++) {
4190 int mode = p->mode ? p->mode : 0644;
4191 debugfs_create_file(p->name, mode, kvm_debugfs_dir,
4192 (void *)(long)p->offset,
4193 stat_fops[p->kind]);
4197 static int kvm_suspend(void)
4199 if (kvm_usage_count)
4200 hardware_disable_nolock(NULL);
4204 static void kvm_resume(void)
4206 if (kvm_usage_count) {
4207 #ifdef CONFIG_LOCKDEP
4208 WARN_ON(lockdep_is_held(&kvm_count_lock));
4210 hardware_enable_nolock(NULL);
4214 static struct syscore_ops kvm_syscore_ops = {
4215 .suspend = kvm_suspend,
4216 .resume = kvm_resume,
4220 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
4222 return container_of(pn, struct kvm_vcpu, preempt_notifier);
4225 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
4227 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
4229 WRITE_ONCE(vcpu->preempted, false);
4230 WRITE_ONCE(vcpu->ready, false);
4232 kvm_arch_sched_in(vcpu, cpu);
4234 kvm_arch_vcpu_load(vcpu, cpu);
4237 static void kvm_sched_out(struct preempt_notifier *pn,
4238 struct task_struct *next)
4240 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
4242 if (current->state == TASK_RUNNING) {
4243 WRITE_ONCE(vcpu->preempted, true);
4244 WRITE_ONCE(vcpu->ready, true);
4246 kvm_arch_vcpu_put(vcpu);
4249 static void check_processor_compat(void *rtn)
4251 *(int *)rtn = kvm_arch_check_processor_compat();
4254 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
4255 struct module *module)
4260 r = kvm_arch_init(opaque);
4265 * kvm_arch_init makes sure there's at most one caller
4266 * for architectures that support multiple implementations,
4267 * like intel and amd on x86.
4268 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
4269 * conflicts in case kvm is already setup for another implementation.
4271 r = kvm_irqfd_init();
4275 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
4280 r = kvm_arch_hardware_setup();
4284 for_each_online_cpu(cpu) {
4285 smp_call_function_single(cpu, check_processor_compat, &r, 1);
4290 r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting",
4291 kvm_starting_cpu, kvm_dying_cpu);
4294 register_reboot_notifier(&kvm_reboot_notifier);
4296 /* A kmem cache lets us meet the alignment requirements of fx_save. */
4298 vcpu_align = __alignof__(struct kvm_vcpu);
4300 kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align,
4302 offsetof(struct kvm_vcpu, arch),
4303 sizeof_field(struct kvm_vcpu, arch),
4305 if (!kvm_vcpu_cache) {
4310 r = kvm_async_pf_init();
4314 kvm_chardev_ops.owner = module;
4315 kvm_vm_fops.owner = module;
4316 kvm_vcpu_fops.owner = module;
4318 r = misc_register(&kvm_dev);
4320 pr_err("kvm: misc device register failed\n");
4324 register_syscore_ops(&kvm_syscore_ops);
4326 kvm_preempt_ops.sched_in = kvm_sched_in;
4327 kvm_preempt_ops.sched_out = kvm_sched_out;
4331 r = kvm_vfio_ops_init();
4337 kvm_async_pf_deinit();
4339 kmem_cache_destroy(kvm_vcpu_cache);
4341 unregister_reboot_notifier(&kvm_reboot_notifier);
4342 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4345 kvm_arch_hardware_unsetup();
4347 free_cpumask_var(cpus_hardware_enabled);
4355 EXPORT_SYMBOL_GPL(kvm_init);
4359 debugfs_remove_recursive(kvm_debugfs_dir);
4360 misc_deregister(&kvm_dev);
4361 kmem_cache_destroy(kvm_vcpu_cache);
4362 kvm_async_pf_deinit();
4363 unregister_syscore_ops(&kvm_syscore_ops);
4364 unregister_reboot_notifier(&kvm_reboot_notifier);
4365 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4366 on_each_cpu(hardware_disable_nolock, NULL, 1);
4367 kvm_arch_hardware_unsetup();
4370 free_cpumask_var(cpus_hardware_enabled);
4371 kvm_vfio_ops_exit();
4373 EXPORT_SYMBOL_GPL(kvm_exit);