1 // SPDX-License-Identifier: GPL-2.0-only
3 * Kernel-based Virtual Machine driver for Linux
5 * This module enables machines with Intel VT-x extensions to run virtual
6 * machines without emulation or binary translation.
8 * Copyright (C) 2006 Qumranet, Inc.
9 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
12 * Avi Kivity <avi@qumranet.com>
13 * Yaniv Kamay <yaniv@qumranet.com>
16 #include <kvm/iodev.h>
18 #include <linux/kvm_host.h>
19 #include <linux/kvm.h>
20 #include <linux/module.h>
21 #include <linux/errno.h>
22 #include <linux/percpu.h>
24 #include <linux/miscdevice.h>
25 #include <linux/vmalloc.h>
26 #include <linux/reboot.h>
27 #include <linux/debugfs.h>
28 #include <linux/highmem.h>
29 #include <linux/file.h>
30 #include <linux/syscore_ops.h>
31 #include <linux/cpu.h>
32 #include <linux/sched/signal.h>
33 #include <linux/sched/mm.h>
34 #include <linux/sched/stat.h>
35 #include <linux/cpumask.h>
36 #include <linux/smp.h>
37 #include <linux/anon_inodes.h>
38 #include <linux/profile.h>
39 #include <linux/kvm_para.h>
40 #include <linux/pagemap.h>
41 #include <linux/mman.h>
42 #include <linux/swap.h>
43 #include <linux/bitops.h>
44 #include <linux/spinlock.h>
45 #include <linux/compat.h>
46 #include <linux/srcu.h>
47 #include <linux/hugetlb.h>
48 #include <linux/slab.h>
49 #include <linux/sort.h>
50 #include <linux/bsearch.h>
52 #include <linux/lockdep.h>
53 #include <linux/kthread.h>
55 #include <asm/processor.h>
56 #include <asm/ioctl.h>
57 #include <linux/uaccess.h>
58 #include <asm/pgtable.h>
60 #include "coalesced_mmio.h"
64 #define CREATE_TRACE_POINTS
65 #include <trace/events/kvm.h>
67 /* Worst case buffer size needed for holding an integer. */
68 #define ITOA_MAX_LEN 12
70 MODULE_AUTHOR("Qumranet");
71 MODULE_LICENSE("GPL");
73 /* Architectures should define their poll value according to the halt latency */
74 unsigned int halt_poll_ns = KVM_HALT_POLL_NS_DEFAULT;
75 module_param(halt_poll_ns, uint, 0644);
76 EXPORT_SYMBOL_GPL(halt_poll_ns);
78 /* Default doubles per-vcpu halt_poll_ns. */
79 unsigned int halt_poll_ns_grow = 2;
80 module_param(halt_poll_ns_grow, uint, 0644);
81 EXPORT_SYMBOL_GPL(halt_poll_ns_grow);
83 /* The start value to grow halt_poll_ns from */
84 unsigned int halt_poll_ns_grow_start = 10000; /* 10us */
85 module_param(halt_poll_ns_grow_start, uint, 0644);
86 EXPORT_SYMBOL_GPL(halt_poll_ns_grow_start);
88 /* Default resets per-vcpu halt_poll_ns . */
89 unsigned int halt_poll_ns_shrink;
90 module_param(halt_poll_ns_shrink, uint, 0644);
91 EXPORT_SYMBOL_GPL(halt_poll_ns_shrink);
96 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
99 DEFINE_MUTEX(kvm_lock);
100 static DEFINE_RAW_SPINLOCK(kvm_count_lock);
103 static cpumask_var_t cpus_hardware_enabled;
104 static int kvm_usage_count;
105 static atomic_t hardware_enable_failed;
107 static struct kmem_cache *kvm_vcpu_cache;
109 static __read_mostly struct preempt_ops kvm_preempt_ops;
110 static DEFINE_PER_CPU(struct kvm_vcpu *, kvm_running_vcpu);
112 struct dentry *kvm_debugfs_dir;
113 EXPORT_SYMBOL_GPL(kvm_debugfs_dir);
115 static int kvm_debugfs_num_entries;
116 static const struct file_operations stat_fops_per_vm;
118 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
120 #ifdef CONFIG_KVM_COMPAT
121 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
123 #define KVM_COMPAT(c) .compat_ioctl = (c)
126 * For architectures that don't implement a compat infrastructure,
127 * adopt a double line of defense:
128 * - Prevent a compat task from opening /dev/kvm
129 * - If the open has been done by a 64bit task, and the KVM fd
130 * passed to a compat task, let the ioctls fail.
132 static long kvm_no_compat_ioctl(struct file *file, unsigned int ioctl,
133 unsigned long arg) { return -EINVAL; }
135 static int kvm_no_compat_open(struct inode *inode, struct file *file)
137 return is_compat_task() ? -ENODEV : 0;
139 #define KVM_COMPAT(c) .compat_ioctl = kvm_no_compat_ioctl, \
140 .open = kvm_no_compat_open
142 static int hardware_enable_all(void);
143 static void hardware_disable_all(void);
145 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
147 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot, gfn_t gfn);
149 __visible bool kvm_rebooting;
150 EXPORT_SYMBOL_GPL(kvm_rebooting);
152 static bool largepages_enabled = true;
154 #define KVM_EVENT_CREATE_VM 0
155 #define KVM_EVENT_DESTROY_VM 1
156 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm);
157 static unsigned long long kvm_createvm_count;
158 static unsigned long long kvm_active_vms;
160 __weak int kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
161 unsigned long start, unsigned long end, bool blockable)
166 bool kvm_is_zone_device_pfn(kvm_pfn_t pfn)
169 * The metadata used by is_zone_device_page() to determine whether or
170 * not a page is ZONE_DEVICE is guaranteed to be valid if and only if
171 * the device has been pinned, e.g. by get_user_pages(). WARN if the
172 * page_count() is zero to help detect bad usage of this helper.
174 if (!pfn_valid(pfn) || WARN_ON_ONCE(!page_count(pfn_to_page(pfn))))
177 return is_zone_device_page(pfn_to_page(pfn));
180 bool kvm_is_reserved_pfn(kvm_pfn_t pfn)
183 * ZONE_DEVICE pages currently set PG_reserved, but from a refcounting
184 * perspective they are "normal" pages, albeit with slightly different
188 return PageReserved(pfn_to_page(pfn)) &&
189 !kvm_is_zone_device_pfn(pfn);
195 * Switches to specified vcpu, until a matching vcpu_put()
197 void vcpu_load(struct kvm_vcpu *vcpu)
201 __this_cpu_write(kvm_running_vcpu, vcpu);
202 preempt_notifier_register(&vcpu->preempt_notifier);
203 kvm_arch_vcpu_load(vcpu, cpu);
206 EXPORT_SYMBOL_GPL(vcpu_load);
208 void vcpu_put(struct kvm_vcpu *vcpu)
211 kvm_arch_vcpu_put(vcpu);
212 preempt_notifier_unregister(&vcpu->preempt_notifier);
213 __this_cpu_write(kvm_running_vcpu, NULL);
216 EXPORT_SYMBOL_GPL(vcpu_put);
218 /* TODO: merge with kvm_arch_vcpu_should_kick */
219 static bool kvm_request_needs_ipi(struct kvm_vcpu *vcpu, unsigned req)
221 int mode = kvm_vcpu_exiting_guest_mode(vcpu);
224 * We need to wait for the VCPU to reenable interrupts and get out of
225 * READING_SHADOW_PAGE_TABLES mode.
227 if (req & KVM_REQUEST_WAIT)
228 return mode != OUTSIDE_GUEST_MODE;
231 * Need to kick a running VCPU, but otherwise there is nothing to do.
233 return mode == IN_GUEST_MODE;
236 static void ack_flush(void *_completed)
240 static inline bool kvm_kick_many_cpus(const struct cpumask *cpus, bool wait)
243 cpus = cpu_online_mask;
245 if (cpumask_empty(cpus))
248 smp_call_function_many(cpus, ack_flush, NULL, wait);
252 bool kvm_make_vcpus_request_mask(struct kvm *kvm, unsigned int req,
253 unsigned long *vcpu_bitmap, cpumask_var_t tmp)
256 struct kvm_vcpu *vcpu;
261 kvm_for_each_vcpu(i, vcpu, kvm) {
262 if (vcpu_bitmap && !test_bit(i, vcpu_bitmap))
265 kvm_make_request(req, vcpu);
268 if (!(req & KVM_REQUEST_NO_WAKEUP) && kvm_vcpu_wake_up(vcpu))
271 if (tmp != NULL && cpu != -1 && cpu != me &&
272 kvm_request_needs_ipi(vcpu, req))
273 __cpumask_set_cpu(cpu, tmp);
276 called = kvm_kick_many_cpus(tmp, !!(req & KVM_REQUEST_WAIT));
282 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
287 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
289 called = kvm_make_vcpus_request_mask(kvm, req, NULL, cpus);
291 free_cpumask_var(cpus);
295 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
296 void kvm_flush_remote_tlbs(struct kvm *kvm)
299 * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in
300 * kvm_make_all_cpus_request.
302 long dirty_count = smp_load_acquire(&kvm->tlbs_dirty);
305 * We want to publish modifications to the page tables before reading
306 * mode. Pairs with a memory barrier in arch-specific code.
307 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
308 * and smp_mb in walk_shadow_page_lockless_begin/end.
309 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
311 * There is already an smp_mb__after_atomic() before
312 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
315 if (!kvm_arch_flush_remote_tlb(kvm)
316 || kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
317 ++kvm->stat.remote_tlb_flush;
318 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
320 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
323 void kvm_reload_remote_mmus(struct kvm *kvm)
325 kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
328 static void kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
330 mutex_init(&vcpu->mutex);
335 init_swait_queue_head(&vcpu->wq);
336 kvm_async_pf_vcpu_init(vcpu);
339 INIT_LIST_HEAD(&vcpu->blocked_vcpu_list);
341 kvm_vcpu_set_in_spin_loop(vcpu, false);
342 kvm_vcpu_set_dy_eligible(vcpu, false);
343 vcpu->preempted = false;
345 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
348 void kvm_vcpu_destroy(struct kvm_vcpu *vcpu)
350 kvm_arch_vcpu_destroy(vcpu);
353 * No need for rcu_read_lock as VCPU_RUN is the only place that changes
354 * the vcpu->pid pointer, and at destruction time all file descriptors
357 put_pid(rcu_dereference_protected(vcpu->pid, 1));
359 free_page((unsigned long)vcpu->run);
360 kmem_cache_free(kvm_vcpu_cache, vcpu);
362 EXPORT_SYMBOL_GPL(kvm_vcpu_destroy);
364 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
365 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
367 return container_of(mn, struct kvm, mmu_notifier);
370 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
371 struct mm_struct *mm,
372 unsigned long address,
375 struct kvm *kvm = mmu_notifier_to_kvm(mn);
378 idx = srcu_read_lock(&kvm->srcu);
379 spin_lock(&kvm->mmu_lock);
380 kvm->mmu_notifier_seq++;
382 if (kvm_set_spte_hva(kvm, address, pte))
383 kvm_flush_remote_tlbs(kvm);
385 spin_unlock(&kvm->mmu_lock);
386 srcu_read_unlock(&kvm->srcu, idx);
389 static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
390 const struct mmu_notifier_range *range)
392 struct kvm *kvm = mmu_notifier_to_kvm(mn);
393 int need_tlb_flush = 0, idx;
396 idx = srcu_read_lock(&kvm->srcu);
397 spin_lock(&kvm->mmu_lock);
399 * The count increase must become visible at unlock time as no
400 * spte can be established without taking the mmu_lock and
401 * count is also read inside the mmu_lock critical section.
403 kvm->mmu_notifier_count++;
404 need_tlb_flush = kvm_unmap_hva_range(kvm, range->start, range->end);
405 need_tlb_flush |= kvm->tlbs_dirty;
406 /* we've to flush the tlb before the pages can be freed */
408 kvm_flush_remote_tlbs(kvm);
410 spin_unlock(&kvm->mmu_lock);
412 ret = kvm_arch_mmu_notifier_invalidate_range(kvm, range->start,
414 mmu_notifier_range_blockable(range));
416 srcu_read_unlock(&kvm->srcu, idx);
421 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
422 const struct mmu_notifier_range *range)
424 struct kvm *kvm = mmu_notifier_to_kvm(mn);
426 spin_lock(&kvm->mmu_lock);
428 * This sequence increase will notify the kvm page fault that
429 * the page that is going to be mapped in the spte could have
432 kvm->mmu_notifier_seq++;
435 * The above sequence increase must be visible before the
436 * below count decrease, which is ensured by the smp_wmb above
437 * in conjunction with the smp_rmb in mmu_notifier_retry().
439 kvm->mmu_notifier_count--;
440 spin_unlock(&kvm->mmu_lock);
442 BUG_ON(kvm->mmu_notifier_count < 0);
445 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
446 struct mm_struct *mm,
450 struct kvm *kvm = mmu_notifier_to_kvm(mn);
453 idx = srcu_read_lock(&kvm->srcu);
454 spin_lock(&kvm->mmu_lock);
456 young = kvm_age_hva(kvm, start, end);
458 kvm_flush_remote_tlbs(kvm);
460 spin_unlock(&kvm->mmu_lock);
461 srcu_read_unlock(&kvm->srcu, idx);
466 static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
467 struct mm_struct *mm,
471 struct kvm *kvm = mmu_notifier_to_kvm(mn);
474 idx = srcu_read_lock(&kvm->srcu);
475 spin_lock(&kvm->mmu_lock);
477 * Even though we do not flush TLB, this will still adversely
478 * affect performance on pre-Haswell Intel EPT, where there is
479 * no EPT Access Bit to clear so that we have to tear down EPT
480 * tables instead. If we find this unacceptable, we can always
481 * add a parameter to kvm_age_hva so that it effectively doesn't
482 * do anything on clear_young.
484 * Also note that currently we never issue secondary TLB flushes
485 * from clear_young, leaving this job up to the regular system
486 * cadence. If we find this inaccurate, we might come up with a
487 * more sophisticated heuristic later.
489 young = kvm_age_hva(kvm, start, end);
490 spin_unlock(&kvm->mmu_lock);
491 srcu_read_unlock(&kvm->srcu, idx);
496 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
497 struct mm_struct *mm,
498 unsigned long address)
500 struct kvm *kvm = mmu_notifier_to_kvm(mn);
503 idx = srcu_read_lock(&kvm->srcu);
504 spin_lock(&kvm->mmu_lock);
505 young = kvm_test_age_hva(kvm, address);
506 spin_unlock(&kvm->mmu_lock);
507 srcu_read_unlock(&kvm->srcu, idx);
512 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
513 struct mm_struct *mm)
515 struct kvm *kvm = mmu_notifier_to_kvm(mn);
518 idx = srcu_read_lock(&kvm->srcu);
519 kvm_arch_flush_shadow_all(kvm);
520 srcu_read_unlock(&kvm->srcu, idx);
523 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
524 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
525 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
526 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
527 .clear_young = kvm_mmu_notifier_clear_young,
528 .test_young = kvm_mmu_notifier_test_young,
529 .change_pte = kvm_mmu_notifier_change_pte,
530 .release = kvm_mmu_notifier_release,
533 static int kvm_init_mmu_notifier(struct kvm *kvm)
535 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
536 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
539 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
541 static int kvm_init_mmu_notifier(struct kvm *kvm)
546 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
548 static struct kvm_memslots *kvm_alloc_memslots(void)
551 struct kvm_memslots *slots;
553 slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL_ACCOUNT);
557 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
558 slots->id_to_index[i] = slots->memslots[i].id = i;
563 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
565 if (!memslot->dirty_bitmap)
568 kvfree(memslot->dirty_bitmap);
569 memslot->dirty_bitmap = NULL;
573 * Free any memory in @free but not in @dont.
575 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
576 struct kvm_memory_slot *dont)
578 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
579 kvm_destroy_dirty_bitmap(free);
581 kvm_arch_free_memslot(kvm, free, dont);
586 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
588 struct kvm_memory_slot *memslot;
593 kvm_for_each_memslot(memslot, slots)
594 kvm_free_memslot(kvm, memslot, NULL);
599 static void kvm_destroy_vm_debugfs(struct kvm *kvm)
603 if (!kvm->debugfs_dentry)
606 debugfs_remove_recursive(kvm->debugfs_dentry);
608 if (kvm->debugfs_stat_data) {
609 for (i = 0; i < kvm_debugfs_num_entries; i++)
610 kfree(kvm->debugfs_stat_data[i]);
611 kfree(kvm->debugfs_stat_data);
615 static int kvm_create_vm_debugfs(struct kvm *kvm, int fd)
617 char dir_name[ITOA_MAX_LEN * 2];
618 struct kvm_stat_data *stat_data;
619 struct kvm_stats_debugfs_item *p;
621 if (!debugfs_initialized())
624 snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd);
625 kvm->debugfs_dentry = debugfs_create_dir(dir_name, kvm_debugfs_dir);
627 kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries,
628 sizeof(*kvm->debugfs_stat_data),
630 if (!kvm->debugfs_stat_data)
633 for (p = debugfs_entries; p->name; p++) {
634 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT);
638 stat_data->kvm = kvm;
639 stat_data->dbgfs_item = p;
640 kvm->debugfs_stat_data[p - debugfs_entries] = stat_data;
641 debugfs_create_file(p->name, KVM_DBGFS_GET_MODE(p),
642 kvm->debugfs_dentry, stat_data,
649 * Called after the VM is otherwise initialized, but just before adding it to
652 int __weak kvm_arch_post_init_vm(struct kvm *kvm)
658 * Called just after removing the VM from the vm_list, but before doing any
661 void __weak kvm_arch_pre_destroy_vm(struct kvm *kvm)
665 static struct kvm *kvm_create_vm(unsigned long type)
667 struct kvm *kvm = kvm_arch_alloc_vm();
672 return ERR_PTR(-ENOMEM);
674 spin_lock_init(&kvm->mmu_lock);
676 kvm->mm = current->mm;
677 kvm_eventfd_init(kvm);
678 mutex_init(&kvm->lock);
679 mutex_init(&kvm->irq_lock);
680 mutex_init(&kvm->slots_lock);
681 INIT_LIST_HEAD(&kvm->devices);
683 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
685 if (init_srcu_struct(&kvm->srcu))
686 goto out_err_no_srcu;
687 if (init_srcu_struct(&kvm->irq_srcu))
688 goto out_err_no_irq_srcu;
690 refcount_set(&kvm->users_count, 1);
691 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
692 struct kvm_memslots *slots = kvm_alloc_memslots();
695 goto out_err_no_arch_destroy_vm;
696 /* Generations must be different for each address space. */
697 slots->generation = i;
698 rcu_assign_pointer(kvm->memslots[i], slots);
701 for (i = 0; i < KVM_NR_BUSES; i++) {
702 rcu_assign_pointer(kvm->buses[i],
703 kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL_ACCOUNT));
705 goto out_err_no_arch_destroy_vm;
708 r = kvm_arch_init_vm(kvm, type);
710 goto out_err_no_arch_destroy_vm;
712 r = hardware_enable_all();
714 goto out_err_no_disable;
716 #ifdef CONFIG_HAVE_KVM_IRQFD
717 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
720 r = kvm_init_mmu_notifier(kvm);
722 goto out_err_no_mmu_notifier;
724 r = kvm_arch_post_init_vm(kvm);
728 mutex_lock(&kvm_lock);
729 list_add(&kvm->vm_list, &vm_list);
730 mutex_unlock(&kvm_lock);
732 preempt_notifier_inc();
737 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
738 if (kvm->mmu_notifier.ops)
739 mmu_notifier_unregister(&kvm->mmu_notifier, current->mm);
741 out_err_no_mmu_notifier:
742 hardware_disable_all();
744 kvm_arch_destroy_vm(kvm);
745 out_err_no_arch_destroy_vm:
746 WARN_ON_ONCE(!refcount_dec_and_test(&kvm->users_count));
747 for (i = 0; i < KVM_NR_BUSES; i++)
748 kfree(kvm_get_bus(kvm, i));
749 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
750 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
751 cleanup_srcu_struct(&kvm->irq_srcu);
753 cleanup_srcu_struct(&kvm->srcu);
755 kvm_arch_free_vm(kvm);
760 static void kvm_destroy_devices(struct kvm *kvm)
762 struct kvm_device *dev, *tmp;
765 * We do not need to take the kvm->lock here, because nobody else
766 * has a reference to the struct kvm at this point and therefore
767 * cannot access the devices list anyhow.
769 list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
770 list_del(&dev->vm_node);
771 dev->ops->destroy(dev);
775 static void kvm_destroy_vm(struct kvm *kvm)
778 struct mm_struct *mm = kvm->mm;
780 kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm);
781 kvm_destroy_vm_debugfs(kvm);
782 kvm_arch_sync_events(kvm);
783 mutex_lock(&kvm_lock);
784 list_del(&kvm->vm_list);
785 mutex_unlock(&kvm_lock);
786 kvm_arch_pre_destroy_vm(kvm);
788 kvm_free_irq_routing(kvm);
789 for (i = 0; i < KVM_NR_BUSES; i++) {
790 struct kvm_io_bus *bus = kvm_get_bus(kvm, i);
793 kvm_io_bus_destroy(bus);
794 kvm->buses[i] = NULL;
796 kvm_coalesced_mmio_free(kvm);
797 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
798 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
800 kvm_arch_flush_shadow_all(kvm);
802 kvm_arch_destroy_vm(kvm);
803 kvm_destroy_devices(kvm);
804 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
805 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
806 cleanup_srcu_struct(&kvm->irq_srcu);
807 cleanup_srcu_struct(&kvm->srcu);
808 kvm_arch_free_vm(kvm);
809 preempt_notifier_dec();
810 hardware_disable_all();
814 void kvm_get_kvm(struct kvm *kvm)
816 refcount_inc(&kvm->users_count);
818 EXPORT_SYMBOL_GPL(kvm_get_kvm);
820 void kvm_put_kvm(struct kvm *kvm)
822 if (refcount_dec_and_test(&kvm->users_count))
825 EXPORT_SYMBOL_GPL(kvm_put_kvm);
828 * Used to put a reference that was taken on behalf of an object associated
829 * with a user-visible file descriptor, e.g. a vcpu or device, if installation
830 * of the new file descriptor fails and the reference cannot be transferred to
831 * its final owner. In such cases, the caller is still actively using @kvm and
832 * will fail miserably if the refcount unexpectedly hits zero.
834 void kvm_put_kvm_no_destroy(struct kvm *kvm)
836 WARN_ON(refcount_dec_and_test(&kvm->users_count));
838 EXPORT_SYMBOL_GPL(kvm_put_kvm_no_destroy);
840 static int kvm_vm_release(struct inode *inode, struct file *filp)
842 struct kvm *kvm = filp->private_data;
844 kvm_irqfd_release(kvm);
851 * Allocation size is twice as large as the actual dirty bitmap size.
852 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
854 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
856 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
858 memslot->dirty_bitmap = kvzalloc(dirty_bytes, GFP_KERNEL_ACCOUNT);
859 if (!memslot->dirty_bitmap)
866 * Insert memslot and re-sort memslots based on their GFN,
867 * so binary search could be used to lookup GFN.
868 * Sorting algorithm takes advantage of having initially
869 * sorted array and known changed memslot position.
871 static void update_memslots(struct kvm_memslots *slots,
872 struct kvm_memory_slot *new,
873 enum kvm_mr_change change)
876 int i = slots->id_to_index[id];
877 struct kvm_memory_slot *mslots = slots->memslots;
879 WARN_ON(mslots[i].id != id);
883 WARN_ON(mslots[i].npages || !new->npages);
887 WARN_ON(new->npages || !mslots[i].npages);
893 while (i < KVM_MEM_SLOTS_NUM - 1 &&
894 new->base_gfn <= mslots[i + 1].base_gfn) {
895 if (!mslots[i + 1].npages)
897 mslots[i] = mslots[i + 1];
898 slots->id_to_index[mslots[i].id] = i;
903 * The ">=" is needed when creating a slot with base_gfn == 0,
904 * so that it moves before all those with base_gfn == npages == 0.
906 * On the other hand, if new->npages is zero, the above loop has
907 * already left i pointing to the beginning of the empty part of
908 * mslots, and the ">=" would move the hole backwards in this
909 * case---which is wrong. So skip the loop when deleting a slot.
913 new->base_gfn >= mslots[i - 1].base_gfn) {
914 mslots[i] = mslots[i - 1];
915 slots->id_to_index[mslots[i].id] = i;
919 WARN_ON_ONCE(i != slots->used_slots);
922 slots->id_to_index[mslots[i].id] = i;
925 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
927 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
929 #ifdef __KVM_HAVE_READONLY_MEM
930 valid_flags |= KVM_MEM_READONLY;
933 if (mem->flags & ~valid_flags)
939 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
940 int as_id, struct kvm_memslots *slots)
942 struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id);
943 u64 gen = old_memslots->generation;
945 WARN_ON(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS);
946 slots->generation = gen | KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
948 rcu_assign_pointer(kvm->memslots[as_id], slots);
949 synchronize_srcu_expedited(&kvm->srcu);
952 * Increment the new memslot generation a second time, dropping the
953 * update in-progress flag and incrementing the generation based on
954 * the number of address spaces. This provides a unique and easily
955 * identifiable generation number while the memslots are in flux.
957 gen = slots->generation & ~KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
960 * Generations must be unique even across address spaces. We do not need
961 * a global counter for that, instead the generation space is evenly split
962 * across address spaces. For example, with two address spaces, address
963 * space 0 will use generations 0, 2, 4, ... while address space 1 will
964 * use generations 1, 3, 5, ...
966 gen += KVM_ADDRESS_SPACE_NUM;
968 kvm_arch_memslots_updated(kvm, gen);
970 slots->generation = gen;
976 * Allocate some memory and give it an address in the guest physical address
979 * Discontiguous memory is allowed, mostly for framebuffers.
981 * Must be called holding kvm->slots_lock for write.
983 int __kvm_set_memory_region(struct kvm *kvm,
984 const struct kvm_userspace_memory_region *mem)
988 unsigned long npages;
989 struct kvm_memory_slot *slot;
990 struct kvm_memory_slot old, new;
991 struct kvm_memslots *slots = NULL, *old_memslots;
993 enum kvm_mr_change change;
995 r = check_memory_region_flags(mem);
1000 as_id = mem->slot >> 16;
1001 id = (u16)mem->slot;
1003 /* General sanity checks */
1004 if (mem->memory_size & (PAGE_SIZE - 1))
1006 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
1008 /* We can read the guest memory with __xxx_user() later on. */
1009 if ((id < KVM_USER_MEM_SLOTS) &&
1010 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
1011 !access_ok((void __user *)(unsigned long)mem->userspace_addr,
1014 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
1016 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
1019 slot = id_to_memslot(__kvm_memslots(kvm, as_id), id);
1020 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
1021 npages = mem->memory_size >> PAGE_SHIFT;
1023 if (npages > KVM_MEM_MAX_NR_PAGES)
1029 new.base_gfn = base_gfn;
1030 new.npages = npages;
1031 new.flags = mem->flags;
1035 change = KVM_MR_CREATE;
1036 else { /* Modify an existing slot. */
1037 if ((mem->userspace_addr != old.userspace_addr) ||
1038 (npages != old.npages) ||
1039 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
1042 if (base_gfn != old.base_gfn)
1043 change = KVM_MR_MOVE;
1044 else if (new.flags != old.flags)
1045 change = KVM_MR_FLAGS_ONLY;
1046 else { /* Nothing to change. */
1055 change = KVM_MR_DELETE;
1060 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
1061 /* Check for overlaps */
1063 kvm_for_each_memslot(slot, __kvm_memslots(kvm, as_id)) {
1066 if (!((base_gfn + npages <= slot->base_gfn) ||
1067 (base_gfn >= slot->base_gfn + slot->npages)))
1072 /* Free page dirty bitmap if unneeded */
1073 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
1074 new.dirty_bitmap = NULL;
1077 if (change == KVM_MR_CREATE) {
1078 new.userspace_addr = mem->userspace_addr;
1080 if (kvm_arch_create_memslot(kvm, &new, npages))
1084 /* Allocate page dirty bitmap if needed */
1085 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
1086 if (kvm_create_dirty_bitmap(&new) < 0)
1090 slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL_ACCOUNT);
1093 memcpy(slots, __kvm_memslots(kvm, as_id), sizeof(struct kvm_memslots));
1095 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
1096 slot = id_to_memslot(slots, id);
1097 slot->flags |= KVM_MEMSLOT_INVALID;
1099 old_memslots = install_new_memslots(kvm, as_id, slots);
1101 /* From this point no new shadow pages pointing to a deleted,
1102 * or moved, memslot will be created.
1104 * validation of sp->gfn happens in:
1105 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1106 * - kvm_is_visible_gfn (mmu_check_root)
1108 kvm_arch_flush_shadow_memslot(kvm, slot);
1111 * We can re-use the old_memslots from above, the only difference
1112 * from the currently installed memslots is the invalid flag. This
1113 * will get overwritten by update_memslots anyway.
1115 slots = old_memslots;
1118 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
1122 /* actual memory is freed via old in kvm_free_memslot below */
1123 if (change == KVM_MR_DELETE) {
1124 new.dirty_bitmap = NULL;
1125 memset(&new.arch, 0, sizeof(new.arch));
1128 update_memslots(slots, &new, change);
1129 old_memslots = install_new_memslots(kvm, as_id, slots);
1131 kvm_arch_commit_memory_region(kvm, mem, &old, &new, change);
1133 kvm_free_memslot(kvm, &old, &new);
1134 kvfree(old_memslots);
1140 kvm_free_memslot(kvm, &new, &old);
1144 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1146 int kvm_set_memory_region(struct kvm *kvm,
1147 const struct kvm_userspace_memory_region *mem)
1151 mutex_lock(&kvm->slots_lock);
1152 r = __kvm_set_memory_region(kvm, mem);
1153 mutex_unlock(&kvm->slots_lock);
1156 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1158 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1159 struct kvm_userspace_memory_region *mem)
1161 if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1164 return kvm_set_memory_region(kvm, mem);
1167 int kvm_get_dirty_log(struct kvm *kvm,
1168 struct kvm_dirty_log *log, int *is_dirty)
1170 struct kvm_memslots *slots;
1171 struct kvm_memory_slot *memslot;
1174 unsigned long any = 0;
1176 as_id = log->slot >> 16;
1177 id = (u16)log->slot;
1178 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1181 slots = __kvm_memslots(kvm, as_id);
1182 memslot = id_to_memslot(slots, id);
1183 if (!memslot->dirty_bitmap)
1186 n = kvm_dirty_bitmap_bytes(memslot);
1188 for (i = 0; !any && i < n/sizeof(long); ++i)
1189 any = memslot->dirty_bitmap[i];
1191 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1198 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1200 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1202 * kvm_get_dirty_log_protect - get a snapshot of dirty pages
1203 * and reenable dirty page tracking for the corresponding pages.
1204 * @kvm: pointer to kvm instance
1205 * @log: slot id and address to which we copy the log
1206 * @flush: true if TLB flush is needed by caller
1208 * We need to keep it in mind that VCPU threads can write to the bitmap
1209 * concurrently. So, to avoid losing track of dirty pages we keep the
1212 * 1. Take a snapshot of the bit and clear it if needed.
1213 * 2. Write protect the corresponding page.
1214 * 3. Copy the snapshot to the userspace.
1215 * 4. Upon return caller flushes TLB's if needed.
1217 * Between 2 and 4, the guest may write to the page using the remaining TLB
1218 * entry. This is not a problem because the page is reported dirty using
1219 * the snapshot taken before and step 4 ensures that writes done after
1220 * exiting to userspace will be logged for the next call.
1223 int kvm_get_dirty_log_protect(struct kvm *kvm,
1224 struct kvm_dirty_log *log, bool *flush)
1226 struct kvm_memslots *slots;
1227 struct kvm_memory_slot *memslot;
1230 unsigned long *dirty_bitmap;
1231 unsigned long *dirty_bitmap_buffer;
1233 as_id = log->slot >> 16;
1234 id = (u16)log->slot;
1235 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1238 slots = __kvm_memslots(kvm, as_id);
1239 memslot = id_to_memslot(slots, id);
1241 dirty_bitmap = memslot->dirty_bitmap;
1245 n = kvm_dirty_bitmap_bytes(memslot);
1247 if (kvm->manual_dirty_log_protect) {
1249 * Unlike kvm_get_dirty_log, we always return false in *flush,
1250 * because no flush is needed until KVM_CLEAR_DIRTY_LOG. There
1251 * is some code duplication between this function and
1252 * kvm_get_dirty_log, but hopefully all architecture
1253 * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log
1254 * can be eliminated.
1256 dirty_bitmap_buffer = dirty_bitmap;
1258 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
1259 memset(dirty_bitmap_buffer, 0, n);
1261 spin_lock(&kvm->mmu_lock);
1262 for (i = 0; i < n / sizeof(long); i++) {
1266 if (!dirty_bitmap[i])
1270 mask = xchg(&dirty_bitmap[i], 0);
1271 dirty_bitmap_buffer[i] = mask;
1273 offset = i * BITS_PER_LONG;
1274 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1277 spin_unlock(&kvm->mmu_lock);
1280 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1284 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect);
1287 * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap
1288 * and reenable dirty page tracking for the corresponding pages.
1289 * @kvm: pointer to kvm instance
1290 * @log: slot id and address from which to fetch the bitmap of dirty pages
1291 * @flush: true if TLB flush is needed by caller
1293 int kvm_clear_dirty_log_protect(struct kvm *kvm,
1294 struct kvm_clear_dirty_log *log, bool *flush)
1296 struct kvm_memslots *slots;
1297 struct kvm_memory_slot *memslot;
1301 unsigned long *dirty_bitmap;
1302 unsigned long *dirty_bitmap_buffer;
1304 as_id = log->slot >> 16;
1305 id = (u16)log->slot;
1306 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1309 if (log->first_page & 63)
1312 slots = __kvm_memslots(kvm, as_id);
1313 memslot = id_to_memslot(slots, id);
1315 dirty_bitmap = memslot->dirty_bitmap;
1319 n = ALIGN(log->num_pages, BITS_PER_LONG) / 8;
1321 if (log->first_page > memslot->npages ||
1322 log->num_pages > memslot->npages - log->first_page ||
1323 (log->num_pages < memslot->npages - log->first_page && (log->num_pages & 63)))
1327 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
1328 if (copy_from_user(dirty_bitmap_buffer, log->dirty_bitmap, n))
1331 spin_lock(&kvm->mmu_lock);
1332 for (offset = log->first_page, i = offset / BITS_PER_LONG,
1333 n = DIV_ROUND_UP(log->num_pages, BITS_PER_LONG); n--;
1334 i++, offset += BITS_PER_LONG) {
1335 unsigned long mask = *dirty_bitmap_buffer++;
1336 atomic_long_t *p = (atomic_long_t *) &dirty_bitmap[i];
1340 mask &= atomic_long_fetch_andnot(mask, p);
1343 * mask contains the bits that really have been cleared. This
1344 * never includes any bits beyond the length of the memslot (if
1345 * the length is not aligned to 64 pages), therefore it is not
1346 * a problem if userspace sets them in log->dirty_bitmap.
1350 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1354 spin_unlock(&kvm->mmu_lock);
1358 EXPORT_SYMBOL_GPL(kvm_clear_dirty_log_protect);
1361 bool kvm_largepages_enabled(void)
1363 return largepages_enabled;
1366 void kvm_disable_largepages(void)
1368 largepages_enabled = false;
1370 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1372 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1374 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1376 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1378 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
1380 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn);
1383 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1385 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1387 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1388 memslot->flags & KVM_MEMSLOT_INVALID)
1393 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1395 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1397 struct vm_area_struct *vma;
1398 unsigned long addr, size;
1402 addr = gfn_to_hva(kvm, gfn);
1403 if (kvm_is_error_hva(addr))
1406 down_read(¤t->mm->mmap_sem);
1407 vma = find_vma(current->mm, addr);
1411 size = vma_kernel_pagesize(vma);
1414 up_read(¤t->mm->mmap_sem);
1419 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1421 return slot->flags & KVM_MEM_READONLY;
1424 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1425 gfn_t *nr_pages, bool write)
1427 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1428 return KVM_HVA_ERR_BAD;
1430 if (memslot_is_readonly(slot) && write)
1431 return KVM_HVA_ERR_RO_BAD;
1434 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1436 return __gfn_to_hva_memslot(slot, gfn);
1439 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1442 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1445 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1448 return gfn_to_hva_many(slot, gfn, NULL);
1450 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1452 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1454 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1456 EXPORT_SYMBOL_GPL(gfn_to_hva);
1458 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
1460 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
1462 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
1465 * Return the hva of a @gfn and the R/W attribute if possible.
1467 * @slot: the kvm_memory_slot which contains @gfn
1468 * @gfn: the gfn to be translated
1469 * @writable: used to return the read/write attribute of the @slot if the hva
1470 * is valid and @writable is not NULL
1472 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1473 gfn_t gfn, bool *writable)
1475 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1477 if (!kvm_is_error_hva(hva) && writable)
1478 *writable = !memslot_is_readonly(slot);
1483 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1485 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1487 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1490 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
1492 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1494 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1497 static inline int check_user_page_hwpoison(unsigned long addr)
1499 int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
1501 rc = get_user_pages(addr, 1, flags, NULL, NULL);
1502 return rc == -EHWPOISON;
1506 * The fast path to get the writable pfn which will be stored in @pfn,
1507 * true indicates success, otherwise false is returned. It's also the
1508 * only part that runs if we can in atomic context.
1510 static bool hva_to_pfn_fast(unsigned long addr, bool write_fault,
1511 bool *writable, kvm_pfn_t *pfn)
1513 struct page *page[1];
1517 * Fast pin a writable pfn only if it is a write fault request
1518 * or the caller allows to map a writable pfn for a read fault
1521 if (!(write_fault || writable))
1524 npages = __get_user_pages_fast(addr, 1, 1, page);
1526 *pfn = page_to_pfn(page[0]);
1537 * The slow path to get the pfn of the specified host virtual address,
1538 * 1 indicates success, -errno is returned if error is detected.
1540 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1541 bool *writable, kvm_pfn_t *pfn)
1543 unsigned int flags = FOLL_HWPOISON;
1550 *writable = write_fault;
1553 flags |= FOLL_WRITE;
1555 flags |= FOLL_NOWAIT;
1557 npages = get_user_pages_unlocked(addr, 1, &page, flags);
1561 /* map read fault as writable if possible */
1562 if (unlikely(!write_fault) && writable) {
1565 if (__get_user_pages_fast(addr, 1, 1, &wpage) == 1) {
1571 *pfn = page_to_pfn(page);
1575 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1577 if (unlikely(!(vma->vm_flags & VM_READ)))
1580 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1586 static int hva_to_pfn_remapped(struct vm_area_struct *vma,
1587 unsigned long addr, bool *async,
1588 bool write_fault, bool *writable,
1594 r = follow_pfn(vma, addr, &pfn);
1597 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
1598 * not call the fault handler, so do it here.
1600 bool unlocked = false;
1601 r = fixup_user_fault(current, current->mm, addr,
1602 (write_fault ? FAULT_FLAG_WRITE : 0),
1609 r = follow_pfn(vma, addr, &pfn);
1619 * Get a reference here because callers of *hva_to_pfn* and
1620 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
1621 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
1622 * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will
1623 * simply do nothing for reserved pfns.
1625 * Whoever called remap_pfn_range is also going to call e.g.
1626 * unmap_mapping_range before the underlying pages are freed,
1627 * causing a call to our MMU notifier.
1636 * Pin guest page in memory and return its pfn.
1637 * @addr: host virtual address which maps memory to the guest
1638 * @atomic: whether this function can sleep
1639 * @async: whether this function need to wait IO complete if the
1640 * host page is not in the memory
1641 * @write_fault: whether we should get a writable host page
1642 * @writable: whether it allows to map a writable host page for !@write_fault
1644 * The function will map a writable host page for these two cases:
1645 * 1): @write_fault = true
1646 * 2): @write_fault = false && @writable, @writable will tell the caller
1647 * whether the mapping is writable.
1649 static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1650 bool write_fault, bool *writable)
1652 struct vm_area_struct *vma;
1656 /* we can do it either atomically or asynchronously, not both */
1657 BUG_ON(atomic && async);
1659 if (hva_to_pfn_fast(addr, write_fault, writable, &pfn))
1663 return KVM_PFN_ERR_FAULT;
1665 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1669 down_read(¤t->mm->mmap_sem);
1670 if (npages == -EHWPOISON ||
1671 (!async && check_user_page_hwpoison(addr))) {
1672 pfn = KVM_PFN_ERR_HWPOISON;
1677 vma = find_vma_intersection(current->mm, addr, addr + 1);
1680 pfn = KVM_PFN_ERR_FAULT;
1681 else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
1682 r = hva_to_pfn_remapped(vma, addr, async, write_fault, writable, &pfn);
1686 pfn = KVM_PFN_ERR_FAULT;
1688 if (async && vma_is_valid(vma, write_fault))
1690 pfn = KVM_PFN_ERR_FAULT;
1693 up_read(¤t->mm->mmap_sem);
1697 kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn,
1698 bool atomic, bool *async, bool write_fault,
1701 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1703 if (addr == KVM_HVA_ERR_RO_BAD) {
1706 return KVM_PFN_ERR_RO_FAULT;
1709 if (kvm_is_error_hva(addr)) {
1712 return KVM_PFN_NOSLOT;
1715 /* Do not map writable pfn in the readonly memslot. */
1716 if (writable && memslot_is_readonly(slot)) {
1721 return hva_to_pfn(addr, atomic, async, write_fault,
1724 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
1726 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1729 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
1730 write_fault, writable);
1732 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1734 kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1736 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1738 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
1740 kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1742 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1744 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1746 kvm_pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1748 return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm, gfn), gfn);
1750 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1752 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
1754 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1756 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
1758 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1760 return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
1762 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1764 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
1766 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1768 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
1770 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1771 struct page **pages, int nr_pages)
1776 addr = gfn_to_hva_many(slot, gfn, &entry);
1777 if (kvm_is_error_hva(addr))
1780 if (entry < nr_pages)
1783 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1785 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1787 static struct page *kvm_pfn_to_page(kvm_pfn_t pfn)
1789 if (is_error_noslot_pfn(pfn))
1790 return KVM_ERR_PTR_BAD_PAGE;
1792 if (kvm_is_reserved_pfn(pfn)) {
1794 return KVM_ERR_PTR_BAD_PAGE;
1797 return pfn_to_page(pfn);
1800 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1804 pfn = gfn_to_pfn(kvm, gfn);
1806 return kvm_pfn_to_page(pfn);
1808 EXPORT_SYMBOL_GPL(gfn_to_page);
1810 static int __kvm_map_gfn(struct kvm_memory_slot *slot, gfn_t gfn,
1811 struct kvm_host_map *map)
1815 struct page *page = KVM_UNMAPPED_PAGE;
1820 pfn = gfn_to_pfn_memslot(slot, gfn);
1821 if (is_error_noslot_pfn(pfn))
1824 if (pfn_valid(pfn)) {
1825 page = pfn_to_page(pfn);
1827 #ifdef CONFIG_HAS_IOMEM
1829 hva = memremap(pfn_to_hpa(pfn), PAGE_SIZE, MEMREMAP_WB);
1844 int kvm_vcpu_map(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map)
1846 return __kvm_map_gfn(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, map);
1848 EXPORT_SYMBOL_GPL(kvm_vcpu_map);
1850 void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map,
1859 if (map->page != KVM_UNMAPPED_PAGE)
1861 #ifdef CONFIG_HAS_IOMEM
1867 kvm_vcpu_mark_page_dirty(vcpu, map->gfn);
1868 kvm_release_pfn_dirty(map->pfn);
1870 kvm_release_pfn_clean(map->pfn);
1876 EXPORT_SYMBOL_GPL(kvm_vcpu_unmap);
1878 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
1882 pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
1884 return kvm_pfn_to_page(pfn);
1886 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
1888 void kvm_release_page_clean(struct page *page)
1890 WARN_ON(is_error_page(page));
1892 kvm_release_pfn_clean(page_to_pfn(page));
1894 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1896 void kvm_release_pfn_clean(kvm_pfn_t pfn)
1898 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
1899 put_page(pfn_to_page(pfn));
1901 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1903 void kvm_release_page_dirty(struct page *page)
1905 WARN_ON(is_error_page(page));
1907 kvm_release_pfn_dirty(page_to_pfn(page));
1909 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1911 void kvm_release_pfn_dirty(kvm_pfn_t pfn)
1913 kvm_set_pfn_dirty(pfn);
1914 kvm_release_pfn_clean(pfn);
1916 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1918 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
1920 if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn))
1921 SetPageDirty(pfn_to_page(pfn));
1923 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1925 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
1927 if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn))
1928 mark_page_accessed(pfn_to_page(pfn));
1930 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1932 void kvm_get_pfn(kvm_pfn_t pfn)
1934 if (!kvm_is_reserved_pfn(pfn))
1935 get_page(pfn_to_page(pfn));
1937 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1939 static int next_segment(unsigned long len, int offset)
1941 if (len > PAGE_SIZE - offset)
1942 return PAGE_SIZE - offset;
1947 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
1948 void *data, int offset, int len)
1953 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1954 if (kvm_is_error_hva(addr))
1956 r = __copy_from_user(data, (void __user *)addr + offset, len);
1962 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1965 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1967 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1969 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1971 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
1972 int offset, int len)
1974 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1976 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1978 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
1980 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1982 gfn_t gfn = gpa >> PAGE_SHIFT;
1984 int offset = offset_in_page(gpa);
1987 while ((seg = next_segment(len, offset)) != 0) {
1988 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1998 EXPORT_SYMBOL_GPL(kvm_read_guest);
2000 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
2002 gfn_t gfn = gpa >> PAGE_SHIFT;
2004 int offset = offset_in_page(gpa);
2007 while ((seg = next_segment(len, offset)) != 0) {
2008 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
2018 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
2020 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
2021 void *data, int offset, unsigned long len)
2026 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
2027 if (kvm_is_error_hva(addr))
2029 pagefault_disable();
2030 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
2037 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
2038 void *data, unsigned long len)
2040 gfn_t gfn = gpa >> PAGE_SHIFT;
2041 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2042 int offset = offset_in_page(gpa);
2044 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
2046 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
2048 static int __kvm_write_guest_page(struct kvm_memory_slot *memslot, gfn_t gfn,
2049 const void *data, int offset, int len)
2054 addr = gfn_to_hva_memslot(memslot, gfn);
2055 if (kvm_is_error_hva(addr))
2057 r = __copy_to_user((void __user *)addr + offset, data, len);
2060 mark_page_dirty_in_slot(memslot, gfn);
2064 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
2065 const void *data, int offset, int len)
2067 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2069 return __kvm_write_guest_page(slot, gfn, data, offset, len);
2071 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
2073 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
2074 const void *data, int offset, int len)
2076 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2078 return __kvm_write_guest_page(slot, gfn, data, offset, len);
2080 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
2082 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
2085 gfn_t gfn = gpa >> PAGE_SHIFT;
2087 int offset = offset_in_page(gpa);
2090 while ((seg = next_segment(len, offset)) != 0) {
2091 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
2101 EXPORT_SYMBOL_GPL(kvm_write_guest);
2103 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
2106 gfn_t gfn = gpa >> PAGE_SHIFT;
2108 int offset = offset_in_page(gpa);
2111 while ((seg = next_segment(len, offset)) != 0) {
2112 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
2122 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
2124 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
2125 struct gfn_to_hva_cache *ghc,
2126 gpa_t gpa, unsigned long len)
2128 int offset = offset_in_page(gpa);
2129 gfn_t start_gfn = gpa >> PAGE_SHIFT;
2130 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
2131 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
2132 gfn_t nr_pages_avail;
2133 int r = start_gfn <= end_gfn ? 0 : -EINVAL;
2136 ghc->generation = slots->generation;
2138 ghc->hva = KVM_HVA_ERR_BAD;
2141 * If the requested region crosses two memslots, we still
2142 * verify that the entire region is valid here.
2144 while (!r && start_gfn <= end_gfn) {
2145 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
2146 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
2148 if (kvm_is_error_hva(ghc->hva))
2150 start_gfn += nr_pages_avail;
2153 /* Use the slow path for cross page reads and writes. */
2154 if (!r && nr_pages_needed == 1)
2157 ghc->memslot = NULL;
2162 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2163 gpa_t gpa, unsigned long len)
2165 struct kvm_memslots *slots = kvm_memslots(kvm);
2166 return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len);
2168 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
2170 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2171 void *data, unsigned int offset,
2174 struct kvm_memslots *slots = kvm_memslots(kvm);
2176 gpa_t gpa = ghc->gpa + offset;
2178 BUG_ON(len + offset > ghc->len);
2180 if (slots->generation != ghc->generation)
2181 __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len);
2183 if (unlikely(!ghc->memslot))
2184 return kvm_write_guest(kvm, gpa, data, len);
2186 if (kvm_is_error_hva(ghc->hva))
2189 r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
2192 mark_page_dirty_in_slot(ghc->memslot, gpa >> PAGE_SHIFT);
2196 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached);
2198 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2199 void *data, unsigned long len)
2201 return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len);
2203 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
2205 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2206 void *data, unsigned long len)
2208 struct kvm_memslots *slots = kvm_memslots(kvm);
2211 BUG_ON(len > ghc->len);
2213 if (slots->generation != ghc->generation)
2214 __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len);
2216 if (unlikely(!ghc->memslot))
2217 return kvm_read_guest(kvm, ghc->gpa, data, len);
2219 if (kvm_is_error_hva(ghc->hva))
2222 r = __copy_from_user(data, (void __user *)ghc->hva, len);
2228 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
2230 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
2232 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
2234 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
2236 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
2238 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
2240 gfn_t gfn = gpa >> PAGE_SHIFT;
2242 int offset = offset_in_page(gpa);
2245 while ((seg = next_segment(len, offset)) != 0) {
2246 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
2255 EXPORT_SYMBOL_GPL(kvm_clear_guest);
2257 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot,
2260 if (memslot && memslot->dirty_bitmap) {
2261 unsigned long rel_gfn = gfn - memslot->base_gfn;
2263 set_bit_le(rel_gfn, memslot->dirty_bitmap);
2267 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
2269 struct kvm_memory_slot *memslot;
2271 memslot = gfn_to_memslot(kvm, gfn);
2272 mark_page_dirty_in_slot(memslot, gfn);
2274 EXPORT_SYMBOL_GPL(mark_page_dirty);
2276 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
2278 struct kvm_memory_slot *memslot;
2280 memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2281 mark_page_dirty_in_slot(memslot, gfn);
2283 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
2285 void kvm_sigset_activate(struct kvm_vcpu *vcpu)
2287 if (!vcpu->sigset_active)
2291 * This does a lockless modification of ->real_blocked, which is fine
2292 * because, only current can change ->real_blocked and all readers of
2293 * ->real_blocked don't care as long ->real_blocked is always a subset
2296 sigprocmask(SIG_SETMASK, &vcpu->sigset, ¤t->real_blocked);
2299 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu)
2301 if (!vcpu->sigset_active)
2304 sigprocmask(SIG_SETMASK, ¤t->real_blocked, NULL);
2305 sigemptyset(¤t->real_blocked);
2308 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
2310 unsigned int old, val, grow, grow_start;
2312 old = val = vcpu->halt_poll_ns;
2313 grow_start = READ_ONCE(halt_poll_ns_grow_start);
2314 grow = READ_ONCE(halt_poll_ns_grow);
2319 if (val < grow_start)
2322 if (val > halt_poll_ns)
2325 vcpu->halt_poll_ns = val;
2327 trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
2330 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
2332 unsigned int old, val, shrink;
2334 old = val = vcpu->halt_poll_ns;
2335 shrink = READ_ONCE(halt_poll_ns_shrink);
2341 vcpu->halt_poll_ns = val;
2342 trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
2345 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
2348 int idx = srcu_read_lock(&vcpu->kvm->srcu);
2350 if (kvm_arch_vcpu_runnable(vcpu)) {
2351 kvm_make_request(KVM_REQ_UNHALT, vcpu);
2354 if (kvm_cpu_has_pending_timer(vcpu))
2356 if (signal_pending(current))
2361 srcu_read_unlock(&vcpu->kvm->srcu, idx);
2366 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2368 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
2371 DECLARE_SWAITQUEUE(wait);
2372 bool waited = false;
2375 kvm_arch_vcpu_blocking(vcpu);
2377 start = cur = ktime_get();
2378 if (vcpu->halt_poll_ns && !kvm_arch_no_poll(vcpu)) {
2379 ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns);
2381 ++vcpu->stat.halt_attempted_poll;
2384 * This sets KVM_REQ_UNHALT if an interrupt
2387 if (kvm_vcpu_check_block(vcpu) < 0) {
2388 ++vcpu->stat.halt_successful_poll;
2389 if (!vcpu_valid_wakeup(vcpu))
2390 ++vcpu->stat.halt_poll_invalid;
2394 } while (single_task_running() && ktime_before(cur, stop));
2398 prepare_to_swait_exclusive(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
2400 if (kvm_vcpu_check_block(vcpu) < 0)
2407 finish_swait(&vcpu->wq, &wait);
2410 kvm_arch_vcpu_unblocking(vcpu);
2411 block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
2413 if (!kvm_arch_no_poll(vcpu)) {
2414 if (!vcpu_valid_wakeup(vcpu)) {
2415 shrink_halt_poll_ns(vcpu);
2416 } else if (halt_poll_ns) {
2417 if (block_ns <= vcpu->halt_poll_ns)
2419 /* we had a long block, shrink polling */
2420 else if (vcpu->halt_poll_ns && block_ns > halt_poll_ns)
2421 shrink_halt_poll_ns(vcpu);
2422 /* we had a short halt and our poll time is too small */
2423 else if (vcpu->halt_poll_ns < halt_poll_ns &&
2424 block_ns < halt_poll_ns)
2425 grow_halt_poll_ns(vcpu);
2427 vcpu->halt_poll_ns = 0;
2431 trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu));
2432 kvm_arch_vcpu_block_finish(vcpu);
2434 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
2436 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
2438 struct swait_queue_head *wqp;
2440 wqp = kvm_arch_vcpu_wq(vcpu);
2441 if (swq_has_sleeper(wqp)) {
2443 WRITE_ONCE(vcpu->ready, true);
2444 ++vcpu->stat.halt_wakeup;
2450 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
2454 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2456 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
2459 int cpu = vcpu->cpu;
2461 if (kvm_vcpu_wake_up(vcpu))
2465 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
2466 if (kvm_arch_vcpu_should_kick(vcpu))
2467 smp_send_reschedule(cpu);
2470 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
2471 #endif /* !CONFIG_S390 */
2473 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
2476 struct task_struct *task = NULL;
2480 pid = rcu_dereference(target->pid);
2482 task = get_pid_task(pid, PIDTYPE_PID);
2486 ret = yield_to(task, 1);
2487 put_task_struct(task);
2491 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
2494 * Helper that checks whether a VCPU is eligible for directed yield.
2495 * Most eligible candidate to yield is decided by following heuristics:
2497 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2498 * (preempted lock holder), indicated by @in_spin_loop.
2499 * Set at the beiginning and cleared at the end of interception/PLE handler.
2501 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2502 * chance last time (mostly it has become eligible now since we have probably
2503 * yielded to lockholder in last iteration. This is done by toggling
2504 * @dy_eligible each time a VCPU checked for eligibility.)
2506 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2507 * to preempted lock-holder could result in wrong VCPU selection and CPU
2508 * burning. Giving priority for a potential lock-holder increases lock
2511 * Since algorithm is based on heuristics, accessing another VCPU data without
2512 * locking does not harm. It may result in trying to yield to same VCPU, fail
2513 * and continue with next VCPU and so on.
2515 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
2517 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2520 eligible = !vcpu->spin_loop.in_spin_loop ||
2521 vcpu->spin_loop.dy_eligible;
2523 if (vcpu->spin_loop.in_spin_loop)
2524 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
2533 * Unlike kvm_arch_vcpu_runnable, this function is called outside
2534 * a vcpu_load/vcpu_put pair. However, for most architectures
2535 * kvm_arch_vcpu_runnable does not require vcpu_load.
2537 bool __weak kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
2539 return kvm_arch_vcpu_runnable(vcpu);
2542 static bool vcpu_dy_runnable(struct kvm_vcpu *vcpu)
2544 if (kvm_arch_dy_runnable(vcpu))
2547 #ifdef CONFIG_KVM_ASYNC_PF
2548 if (!list_empty_careful(&vcpu->async_pf.done))
2555 void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode)
2557 struct kvm *kvm = me->kvm;
2558 struct kvm_vcpu *vcpu;
2559 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
2565 kvm_vcpu_set_in_spin_loop(me, true);
2567 * We boost the priority of a VCPU that is runnable but not
2568 * currently running, because it got preempted by something
2569 * else and called schedule in __vcpu_run. Hopefully that
2570 * VCPU is holding the lock that we need and will release it.
2571 * We approximate round-robin by starting at the last boosted VCPU.
2573 for (pass = 0; pass < 2 && !yielded && try; pass++) {
2574 kvm_for_each_vcpu(i, vcpu, kvm) {
2575 if (!pass && i <= last_boosted_vcpu) {
2576 i = last_boosted_vcpu;
2578 } else if (pass && i > last_boosted_vcpu)
2580 if (!READ_ONCE(vcpu->ready))
2584 if (swait_active(&vcpu->wq) && !vcpu_dy_runnable(vcpu))
2586 if (READ_ONCE(vcpu->preempted) && yield_to_kernel_mode &&
2587 !kvm_arch_vcpu_in_kernel(vcpu))
2589 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
2592 yielded = kvm_vcpu_yield_to(vcpu);
2594 kvm->last_boosted_vcpu = i;
2596 } else if (yielded < 0) {
2603 kvm_vcpu_set_in_spin_loop(me, false);
2605 /* Ensure vcpu is not eligible during next spinloop */
2606 kvm_vcpu_set_dy_eligible(me, false);
2608 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
2610 static vm_fault_t kvm_vcpu_fault(struct vm_fault *vmf)
2612 struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data;
2615 if (vmf->pgoff == 0)
2616 page = virt_to_page(vcpu->run);
2618 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
2619 page = virt_to_page(vcpu->arch.pio_data);
2621 #ifdef CONFIG_KVM_MMIO
2622 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
2623 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
2626 return kvm_arch_vcpu_fault(vcpu, vmf);
2632 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
2633 .fault = kvm_vcpu_fault,
2636 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2638 vma->vm_ops = &kvm_vcpu_vm_ops;
2642 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2644 struct kvm_vcpu *vcpu = filp->private_data;
2646 debugfs_remove_recursive(vcpu->debugfs_dentry);
2647 kvm_put_kvm(vcpu->kvm);
2651 static struct file_operations kvm_vcpu_fops = {
2652 .release = kvm_vcpu_release,
2653 .unlocked_ioctl = kvm_vcpu_ioctl,
2654 .mmap = kvm_vcpu_mmap,
2655 .llseek = noop_llseek,
2656 KVM_COMPAT(kvm_vcpu_compat_ioctl),
2660 * Allocates an inode for the vcpu.
2662 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2664 char name[8 + 1 + ITOA_MAX_LEN + 1];
2666 snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id);
2667 return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
2670 static void kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
2672 #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS
2673 char dir_name[ITOA_MAX_LEN * 2];
2675 if (!debugfs_initialized())
2678 snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
2679 vcpu->debugfs_dentry = debugfs_create_dir(dir_name,
2680 vcpu->kvm->debugfs_dentry);
2682 kvm_arch_create_vcpu_debugfs(vcpu);
2687 * Creates some virtual cpus. Good luck creating more than one.
2689 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
2692 struct kvm_vcpu *vcpu;
2695 if (id >= KVM_MAX_VCPU_ID)
2698 mutex_lock(&kvm->lock);
2699 if (kvm->created_vcpus == KVM_MAX_VCPUS) {
2700 mutex_unlock(&kvm->lock);
2704 kvm->created_vcpus++;
2705 mutex_unlock(&kvm->lock);
2707 r = kvm_arch_vcpu_precreate(kvm, id);
2709 goto vcpu_decrement;
2711 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
2714 goto vcpu_decrement;
2717 BUILD_BUG_ON(sizeof(struct kvm_run) > PAGE_SIZE);
2718 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2723 vcpu->run = page_address(page);
2725 kvm_vcpu_init(vcpu, kvm, id);
2727 r = kvm_arch_vcpu_create(vcpu);
2729 goto vcpu_free_run_page;
2731 kvm_create_vcpu_debugfs(vcpu);
2733 mutex_lock(&kvm->lock);
2734 if (kvm_get_vcpu_by_id(kvm, id)) {
2736 goto unlock_vcpu_destroy;
2739 vcpu->vcpu_idx = atomic_read(&kvm->online_vcpus);
2740 BUG_ON(kvm->vcpus[vcpu->vcpu_idx]);
2742 /* Now it's all set up, let userspace reach it */
2744 r = create_vcpu_fd(vcpu);
2746 kvm_put_kvm_no_destroy(kvm);
2747 goto unlock_vcpu_destroy;
2750 kvm->vcpus[vcpu->vcpu_idx] = vcpu;
2753 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
2754 * before kvm->online_vcpu's incremented value.
2757 atomic_inc(&kvm->online_vcpus);
2759 mutex_unlock(&kvm->lock);
2760 kvm_arch_vcpu_postcreate(vcpu);
2763 unlock_vcpu_destroy:
2764 mutex_unlock(&kvm->lock);
2765 debugfs_remove_recursive(vcpu->debugfs_dentry);
2766 kvm_arch_vcpu_destroy(vcpu);
2768 free_page((unsigned long)vcpu->run);
2770 kmem_cache_free(kvm_vcpu_cache, vcpu);
2772 mutex_lock(&kvm->lock);
2773 kvm->created_vcpus--;
2774 mutex_unlock(&kvm->lock);
2778 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2781 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2782 vcpu->sigset_active = 1;
2783 vcpu->sigset = *sigset;
2785 vcpu->sigset_active = 0;
2789 static long kvm_vcpu_ioctl(struct file *filp,
2790 unsigned int ioctl, unsigned long arg)
2792 struct kvm_vcpu *vcpu = filp->private_data;
2793 void __user *argp = (void __user *)arg;
2795 struct kvm_fpu *fpu = NULL;
2796 struct kvm_sregs *kvm_sregs = NULL;
2798 if (vcpu->kvm->mm != current->mm)
2801 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2805 * Some architectures have vcpu ioctls that are asynchronous to vcpu
2806 * execution; mutex_lock() would break them.
2808 r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg);
2809 if (r != -ENOIOCTLCMD)
2812 if (mutex_lock_killable(&vcpu->mutex))
2820 oldpid = rcu_access_pointer(vcpu->pid);
2821 if (unlikely(oldpid != task_pid(current))) {
2822 /* The thread running this VCPU changed. */
2825 r = kvm_arch_vcpu_run_pid_change(vcpu);
2829 newpid = get_task_pid(current, PIDTYPE_PID);
2830 rcu_assign_pointer(vcpu->pid, newpid);
2835 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2836 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2839 case KVM_GET_REGS: {
2840 struct kvm_regs *kvm_regs;
2843 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL_ACCOUNT);
2846 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2850 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2857 case KVM_SET_REGS: {
2858 struct kvm_regs *kvm_regs;
2861 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2862 if (IS_ERR(kvm_regs)) {
2863 r = PTR_ERR(kvm_regs);
2866 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2870 case KVM_GET_SREGS: {
2871 kvm_sregs = kzalloc(sizeof(struct kvm_sregs),
2872 GFP_KERNEL_ACCOUNT);
2876 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2880 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2885 case KVM_SET_SREGS: {
2886 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2887 if (IS_ERR(kvm_sregs)) {
2888 r = PTR_ERR(kvm_sregs);
2892 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2895 case KVM_GET_MP_STATE: {
2896 struct kvm_mp_state mp_state;
2898 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2902 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
2907 case KVM_SET_MP_STATE: {
2908 struct kvm_mp_state mp_state;
2911 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
2913 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2916 case KVM_TRANSLATE: {
2917 struct kvm_translation tr;
2920 if (copy_from_user(&tr, argp, sizeof(tr)))
2922 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2926 if (copy_to_user(argp, &tr, sizeof(tr)))
2931 case KVM_SET_GUEST_DEBUG: {
2932 struct kvm_guest_debug dbg;
2935 if (copy_from_user(&dbg, argp, sizeof(dbg)))
2937 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2940 case KVM_SET_SIGNAL_MASK: {
2941 struct kvm_signal_mask __user *sigmask_arg = argp;
2942 struct kvm_signal_mask kvm_sigmask;
2943 sigset_t sigset, *p;
2948 if (copy_from_user(&kvm_sigmask, argp,
2949 sizeof(kvm_sigmask)))
2952 if (kvm_sigmask.len != sizeof(sigset))
2955 if (copy_from_user(&sigset, sigmask_arg->sigset,
2960 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2964 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL_ACCOUNT);
2968 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2972 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2978 fpu = memdup_user(argp, sizeof(*fpu));
2984 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2988 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2991 mutex_unlock(&vcpu->mutex);
2997 #ifdef CONFIG_KVM_COMPAT
2998 static long kvm_vcpu_compat_ioctl(struct file *filp,
2999 unsigned int ioctl, unsigned long arg)
3001 struct kvm_vcpu *vcpu = filp->private_data;
3002 void __user *argp = compat_ptr(arg);
3005 if (vcpu->kvm->mm != current->mm)
3009 case KVM_SET_SIGNAL_MASK: {
3010 struct kvm_signal_mask __user *sigmask_arg = argp;
3011 struct kvm_signal_mask kvm_sigmask;
3016 if (copy_from_user(&kvm_sigmask, argp,
3017 sizeof(kvm_sigmask)))
3020 if (kvm_sigmask.len != sizeof(compat_sigset_t))
3023 if (get_compat_sigset(&sigset, (void *)sigmask_arg->sigset))
3025 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
3027 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
3031 r = kvm_vcpu_ioctl(filp, ioctl, arg);
3039 static int kvm_device_mmap(struct file *filp, struct vm_area_struct *vma)
3041 struct kvm_device *dev = filp->private_data;
3044 return dev->ops->mmap(dev, vma);
3049 static int kvm_device_ioctl_attr(struct kvm_device *dev,
3050 int (*accessor)(struct kvm_device *dev,
3051 struct kvm_device_attr *attr),
3054 struct kvm_device_attr attr;
3059 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
3062 return accessor(dev, &attr);
3065 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
3068 struct kvm_device *dev = filp->private_data;
3070 if (dev->kvm->mm != current->mm)
3074 case KVM_SET_DEVICE_ATTR:
3075 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
3076 case KVM_GET_DEVICE_ATTR:
3077 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
3078 case KVM_HAS_DEVICE_ATTR:
3079 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
3081 if (dev->ops->ioctl)
3082 return dev->ops->ioctl(dev, ioctl, arg);
3088 static int kvm_device_release(struct inode *inode, struct file *filp)
3090 struct kvm_device *dev = filp->private_data;
3091 struct kvm *kvm = dev->kvm;
3093 if (dev->ops->release) {
3094 mutex_lock(&kvm->lock);
3095 list_del(&dev->vm_node);
3096 dev->ops->release(dev);
3097 mutex_unlock(&kvm->lock);
3104 static const struct file_operations kvm_device_fops = {
3105 .unlocked_ioctl = kvm_device_ioctl,
3106 .release = kvm_device_release,
3107 KVM_COMPAT(kvm_device_ioctl),
3108 .mmap = kvm_device_mmap,
3111 struct kvm_device *kvm_device_from_filp(struct file *filp)
3113 if (filp->f_op != &kvm_device_fops)
3116 return filp->private_data;
3119 static const struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
3120 #ifdef CONFIG_KVM_MPIC
3121 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
3122 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
3126 int kvm_register_device_ops(const struct kvm_device_ops *ops, u32 type)
3128 if (type >= ARRAY_SIZE(kvm_device_ops_table))
3131 if (kvm_device_ops_table[type] != NULL)
3134 kvm_device_ops_table[type] = ops;
3138 void kvm_unregister_device_ops(u32 type)
3140 if (kvm_device_ops_table[type] != NULL)
3141 kvm_device_ops_table[type] = NULL;
3144 static int kvm_ioctl_create_device(struct kvm *kvm,
3145 struct kvm_create_device *cd)
3147 const struct kvm_device_ops *ops = NULL;
3148 struct kvm_device *dev;
3149 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
3153 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
3156 type = array_index_nospec(cd->type, ARRAY_SIZE(kvm_device_ops_table));
3157 ops = kvm_device_ops_table[type];
3164 dev = kzalloc(sizeof(*dev), GFP_KERNEL_ACCOUNT);
3171 mutex_lock(&kvm->lock);
3172 ret = ops->create(dev, type);
3174 mutex_unlock(&kvm->lock);
3178 list_add(&dev->vm_node, &kvm->devices);
3179 mutex_unlock(&kvm->lock);
3185 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
3187 kvm_put_kvm_no_destroy(kvm);
3188 mutex_lock(&kvm->lock);
3189 list_del(&dev->vm_node);
3190 mutex_unlock(&kvm->lock);
3199 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
3202 case KVM_CAP_USER_MEMORY:
3203 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
3204 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
3205 case KVM_CAP_INTERNAL_ERROR_DATA:
3206 #ifdef CONFIG_HAVE_KVM_MSI
3207 case KVM_CAP_SIGNAL_MSI:
3209 #ifdef CONFIG_HAVE_KVM_IRQFD
3211 case KVM_CAP_IRQFD_RESAMPLE:
3213 case KVM_CAP_IOEVENTFD_ANY_LENGTH:
3214 case KVM_CAP_CHECK_EXTENSION_VM:
3215 case KVM_CAP_ENABLE_CAP_VM:
3216 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3217 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2:
3220 #ifdef CONFIG_KVM_MMIO
3221 case KVM_CAP_COALESCED_MMIO:
3222 return KVM_COALESCED_MMIO_PAGE_OFFSET;
3223 case KVM_CAP_COALESCED_PIO:
3226 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3227 case KVM_CAP_IRQ_ROUTING:
3228 return KVM_MAX_IRQ_ROUTES;
3230 #if KVM_ADDRESS_SPACE_NUM > 1
3231 case KVM_CAP_MULTI_ADDRESS_SPACE:
3232 return KVM_ADDRESS_SPACE_NUM;
3234 case KVM_CAP_NR_MEMSLOTS:
3235 return KVM_USER_MEM_SLOTS;
3239 return kvm_vm_ioctl_check_extension(kvm, arg);
3242 int __attribute__((weak)) kvm_vm_ioctl_enable_cap(struct kvm *kvm,
3243 struct kvm_enable_cap *cap)
3248 static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm,
3249 struct kvm_enable_cap *cap)
3252 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3253 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2:
3254 if (cap->flags || (cap->args[0] & ~1))
3256 kvm->manual_dirty_log_protect = cap->args[0];
3260 return kvm_vm_ioctl_enable_cap(kvm, cap);
3264 static long kvm_vm_ioctl(struct file *filp,
3265 unsigned int ioctl, unsigned long arg)
3267 struct kvm *kvm = filp->private_data;
3268 void __user *argp = (void __user *)arg;
3271 if (kvm->mm != current->mm)
3274 case KVM_CREATE_VCPU:
3275 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
3277 case KVM_ENABLE_CAP: {
3278 struct kvm_enable_cap cap;
3281 if (copy_from_user(&cap, argp, sizeof(cap)))
3283 r = kvm_vm_ioctl_enable_cap_generic(kvm, &cap);
3286 case KVM_SET_USER_MEMORY_REGION: {
3287 struct kvm_userspace_memory_region kvm_userspace_mem;
3290 if (copy_from_user(&kvm_userspace_mem, argp,
3291 sizeof(kvm_userspace_mem)))
3294 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
3297 case KVM_GET_DIRTY_LOG: {
3298 struct kvm_dirty_log log;
3301 if (copy_from_user(&log, argp, sizeof(log)))
3303 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3306 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3307 case KVM_CLEAR_DIRTY_LOG: {
3308 struct kvm_clear_dirty_log log;
3311 if (copy_from_user(&log, argp, sizeof(log)))
3313 r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
3317 #ifdef CONFIG_KVM_MMIO
3318 case KVM_REGISTER_COALESCED_MMIO: {
3319 struct kvm_coalesced_mmio_zone zone;
3322 if (copy_from_user(&zone, argp, sizeof(zone)))
3324 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
3327 case KVM_UNREGISTER_COALESCED_MMIO: {
3328 struct kvm_coalesced_mmio_zone zone;
3331 if (copy_from_user(&zone, argp, sizeof(zone)))
3333 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
3338 struct kvm_irqfd data;
3341 if (copy_from_user(&data, argp, sizeof(data)))
3343 r = kvm_irqfd(kvm, &data);
3346 case KVM_IOEVENTFD: {
3347 struct kvm_ioeventfd data;
3350 if (copy_from_user(&data, argp, sizeof(data)))
3352 r = kvm_ioeventfd(kvm, &data);
3355 #ifdef CONFIG_HAVE_KVM_MSI
3356 case KVM_SIGNAL_MSI: {
3360 if (copy_from_user(&msi, argp, sizeof(msi)))
3362 r = kvm_send_userspace_msi(kvm, &msi);
3366 #ifdef __KVM_HAVE_IRQ_LINE
3367 case KVM_IRQ_LINE_STATUS:
3368 case KVM_IRQ_LINE: {
3369 struct kvm_irq_level irq_event;
3372 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
3375 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
3376 ioctl == KVM_IRQ_LINE_STATUS);
3381 if (ioctl == KVM_IRQ_LINE_STATUS) {
3382 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
3390 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3391 case KVM_SET_GSI_ROUTING: {
3392 struct kvm_irq_routing routing;
3393 struct kvm_irq_routing __user *urouting;
3394 struct kvm_irq_routing_entry *entries = NULL;
3397 if (copy_from_user(&routing, argp, sizeof(routing)))
3400 if (!kvm_arch_can_set_irq_routing(kvm))
3402 if (routing.nr > KVM_MAX_IRQ_ROUTES)
3408 entries = vmalloc(array_size(sizeof(*entries),
3414 if (copy_from_user(entries, urouting->entries,
3415 routing.nr * sizeof(*entries)))
3416 goto out_free_irq_routing;
3418 r = kvm_set_irq_routing(kvm, entries, routing.nr,
3420 out_free_irq_routing:
3424 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
3425 case KVM_CREATE_DEVICE: {
3426 struct kvm_create_device cd;
3429 if (copy_from_user(&cd, argp, sizeof(cd)))
3432 r = kvm_ioctl_create_device(kvm, &cd);
3437 if (copy_to_user(argp, &cd, sizeof(cd)))
3443 case KVM_CHECK_EXTENSION:
3444 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
3447 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
3453 #ifdef CONFIG_KVM_COMPAT
3454 struct compat_kvm_dirty_log {
3458 compat_uptr_t dirty_bitmap; /* one bit per page */
3463 static long kvm_vm_compat_ioctl(struct file *filp,
3464 unsigned int ioctl, unsigned long arg)
3466 struct kvm *kvm = filp->private_data;
3469 if (kvm->mm != current->mm)
3472 case KVM_GET_DIRTY_LOG: {
3473 struct compat_kvm_dirty_log compat_log;
3474 struct kvm_dirty_log log;
3476 if (copy_from_user(&compat_log, (void __user *)arg,
3477 sizeof(compat_log)))
3479 log.slot = compat_log.slot;
3480 log.padding1 = compat_log.padding1;
3481 log.padding2 = compat_log.padding2;
3482 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
3484 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3488 r = kvm_vm_ioctl(filp, ioctl, arg);
3494 static struct file_operations kvm_vm_fops = {
3495 .release = kvm_vm_release,
3496 .unlocked_ioctl = kvm_vm_ioctl,
3497 .llseek = noop_llseek,
3498 KVM_COMPAT(kvm_vm_compat_ioctl),
3501 static int kvm_dev_ioctl_create_vm(unsigned long type)
3507 kvm = kvm_create_vm(type);
3509 return PTR_ERR(kvm);
3510 #ifdef CONFIG_KVM_MMIO
3511 r = kvm_coalesced_mmio_init(kvm);
3515 r = get_unused_fd_flags(O_CLOEXEC);
3519 file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
3527 * Don't call kvm_put_kvm anymore at this point; file->f_op is
3528 * already set, with ->release() being kvm_vm_release(). In error
3529 * cases it will be called by the final fput(file) and will take
3530 * care of doing kvm_put_kvm(kvm).
3532 if (kvm_create_vm_debugfs(kvm, r) < 0) {
3537 kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm);
3539 fd_install(r, file);
3547 static long kvm_dev_ioctl(struct file *filp,
3548 unsigned int ioctl, unsigned long arg)
3553 case KVM_GET_API_VERSION:
3556 r = KVM_API_VERSION;
3559 r = kvm_dev_ioctl_create_vm(arg);
3561 case KVM_CHECK_EXTENSION:
3562 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
3564 case KVM_GET_VCPU_MMAP_SIZE:
3567 r = PAGE_SIZE; /* struct kvm_run */
3569 r += PAGE_SIZE; /* pio data page */
3571 #ifdef CONFIG_KVM_MMIO
3572 r += PAGE_SIZE; /* coalesced mmio ring page */
3575 case KVM_TRACE_ENABLE:
3576 case KVM_TRACE_PAUSE:
3577 case KVM_TRACE_DISABLE:
3581 return kvm_arch_dev_ioctl(filp, ioctl, arg);
3587 static struct file_operations kvm_chardev_ops = {
3588 .unlocked_ioctl = kvm_dev_ioctl,
3589 .llseek = noop_llseek,
3590 KVM_COMPAT(kvm_dev_ioctl),
3593 static struct miscdevice kvm_dev = {
3599 static void hardware_enable_nolock(void *junk)
3601 int cpu = raw_smp_processor_id();
3604 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
3607 cpumask_set_cpu(cpu, cpus_hardware_enabled);
3609 r = kvm_arch_hardware_enable();
3612 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3613 atomic_inc(&hardware_enable_failed);
3614 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
3618 static int kvm_starting_cpu(unsigned int cpu)
3620 raw_spin_lock(&kvm_count_lock);
3621 if (kvm_usage_count)
3622 hardware_enable_nolock(NULL);
3623 raw_spin_unlock(&kvm_count_lock);
3627 static void hardware_disable_nolock(void *junk)
3629 int cpu = raw_smp_processor_id();
3631 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
3633 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3634 kvm_arch_hardware_disable();
3637 static int kvm_dying_cpu(unsigned int cpu)
3639 raw_spin_lock(&kvm_count_lock);
3640 if (kvm_usage_count)
3641 hardware_disable_nolock(NULL);
3642 raw_spin_unlock(&kvm_count_lock);
3646 static void hardware_disable_all_nolock(void)
3648 BUG_ON(!kvm_usage_count);
3651 if (!kvm_usage_count)
3652 on_each_cpu(hardware_disable_nolock, NULL, 1);
3655 static void hardware_disable_all(void)
3657 raw_spin_lock(&kvm_count_lock);
3658 hardware_disable_all_nolock();
3659 raw_spin_unlock(&kvm_count_lock);
3662 static int hardware_enable_all(void)
3666 raw_spin_lock(&kvm_count_lock);
3669 if (kvm_usage_count == 1) {
3670 atomic_set(&hardware_enable_failed, 0);
3671 on_each_cpu(hardware_enable_nolock, NULL, 1);
3673 if (atomic_read(&hardware_enable_failed)) {
3674 hardware_disable_all_nolock();
3679 raw_spin_unlock(&kvm_count_lock);
3684 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3688 * Some (well, at least mine) BIOSes hang on reboot if
3691 * And Intel TXT required VMX off for all cpu when system shutdown.
3693 pr_info("kvm: exiting hardware virtualization\n");
3694 kvm_rebooting = true;
3695 on_each_cpu(hardware_disable_nolock, NULL, 1);
3699 static struct notifier_block kvm_reboot_notifier = {
3700 .notifier_call = kvm_reboot,
3704 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3708 for (i = 0; i < bus->dev_count; i++) {
3709 struct kvm_io_device *pos = bus->range[i].dev;
3711 kvm_iodevice_destructor(pos);
3716 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
3717 const struct kvm_io_range *r2)
3719 gpa_t addr1 = r1->addr;
3720 gpa_t addr2 = r2->addr;
3725 /* If r2->len == 0, match the exact address. If r2->len != 0,
3726 * accept any overlapping write. Any order is acceptable for
3727 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
3728 * we process all of them.
3741 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
3743 return kvm_io_bus_cmp(p1, p2);
3746 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
3747 gpa_t addr, int len)
3749 struct kvm_io_range *range, key;
3752 key = (struct kvm_io_range) {
3757 range = bsearch(&key, bus->range, bus->dev_count,
3758 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
3762 off = range - bus->range;
3764 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
3770 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3771 struct kvm_io_range *range, const void *val)
3775 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3779 while (idx < bus->dev_count &&
3780 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3781 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
3790 /* kvm_io_bus_write - called under kvm->slots_lock */
3791 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3792 int len, const void *val)
3794 struct kvm_io_bus *bus;
3795 struct kvm_io_range range;
3798 range = (struct kvm_io_range) {
3803 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3806 r = __kvm_io_bus_write(vcpu, bus, &range, val);
3807 return r < 0 ? r : 0;
3809 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
3811 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3812 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
3813 gpa_t addr, int len, const void *val, long cookie)
3815 struct kvm_io_bus *bus;
3816 struct kvm_io_range range;
3818 range = (struct kvm_io_range) {
3823 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3827 /* First try the device referenced by cookie. */
3828 if ((cookie >= 0) && (cookie < bus->dev_count) &&
3829 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
3830 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
3835 * cookie contained garbage; fall back to search and return the
3836 * correct cookie value.
3838 return __kvm_io_bus_write(vcpu, bus, &range, val);
3841 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3842 struct kvm_io_range *range, void *val)
3846 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3850 while (idx < bus->dev_count &&
3851 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3852 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
3861 /* kvm_io_bus_read - called under kvm->slots_lock */
3862 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3865 struct kvm_io_bus *bus;
3866 struct kvm_io_range range;
3869 range = (struct kvm_io_range) {
3874 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3877 r = __kvm_io_bus_read(vcpu, bus, &range, val);
3878 return r < 0 ? r : 0;
3881 /* Caller must hold slots_lock. */
3882 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3883 int len, struct kvm_io_device *dev)
3886 struct kvm_io_bus *new_bus, *bus;
3887 struct kvm_io_range range;
3889 bus = kvm_get_bus(kvm, bus_idx);
3893 /* exclude ioeventfd which is limited by maximum fd */
3894 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3897 new_bus = kmalloc(struct_size(bus, range, bus->dev_count + 1),
3898 GFP_KERNEL_ACCOUNT);
3902 range = (struct kvm_io_range) {
3908 for (i = 0; i < bus->dev_count; i++)
3909 if (kvm_io_bus_cmp(&bus->range[i], &range) > 0)
3912 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3913 new_bus->dev_count++;
3914 new_bus->range[i] = range;
3915 memcpy(new_bus->range + i + 1, bus->range + i,
3916 (bus->dev_count - i) * sizeof(struct kvm_io_range));
3917 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3918 synchronize_srcu_expedited(&kvm->srcu);
3924 /* Caller must hold slots_lock. */
3925 void kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3926 struct kvm_io_device *dev)
3929 struct kvm_io_bus *new_bus, *bus;
3931 bus = kvm_get_bus(kvm, bus_idx);
3935 for (i = 0; i < bus->dev_count; i++)
3936 if (bus->range[i].dev == dev) {
3940 if (i == bus->dev_count)
3943 new_bus = kmalloc(struct_size(bus, range, bus->dev_count - 1),
3944 GFP_KERNEL_ACCOUNT);
3946 pr_err("kvm: failed to shrink bus, removing it completely\n");
3950 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3951 new_bus->dev_count--;
3952 memcpy(new_bus->range + i, bus->range + i + 1,
3953 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3956 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3957 synchronize_srcu_expedited(&kvm->srcu);
3962 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3965 struct kvm_io_bus *bus;
3966 int dev_idx, srcu_idx;
3967 struct kvm_io_device *iodev = NULL;
3969 srcu_idx = srcu_read_lock(&kvm->srcu);
3971 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
3975 dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
3979 iodev = bus->range[dev_idx].dev;
3982 srcu_read_unlock(&kvm->srcu, srcu_idx);
3986 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
3988 static int kvm_debugfs_open(struct inode *inode, struct file *file,
3989 int (*get)(void *, u64 *), int (*set)(void *, u64),
3992 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
3995 /* The debugfs files are a reference to the kvm struct which
3996 * is still valid when kvm_destroy_vm is called.
3997 * To avoid the race between open and the removal of the debugfs
3998 * directory we test against the users count.
4000 if (!refcount_inc_not_zero(&stat_data->kvm->users_count))
4003 if (simple_attr_open(inode, file, get,
4004 KVM_DBGFS_GET_MODE(stat_data->dbgfs_item) & 0222
4007 kvm_put_kvm(stat_data->kvm);
4014 static int kvm_debugfs_release(struct inode *inode, struct file *file)
4016 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
4019 simple_attr_release(inode, file);
4020 kvm_put_kvm(stat_data->kvm);
4025 static int kvm_get_stat_per_vm(struct kvm *kvm, size_t offset, u64 *val)
4027 *val = *(ulong *)((void *)kvm + offset);
4032 static int kvm_clear_stat_per_vm(struct kvm *kvm, size_t offset)
4034 *(ulong *)((void *)kvm + offset) = 0;
4039 static int kvm_get_stat_per_vcpu(struct kvm *kvm, size_t offset, u64 *val)
4042 struct kvm_vcpu *vcpu;
4046 kvm_for_each_vcpu(i, vcpu, kvm)
4047 *val += *(u64 *)((void *)vcpu + offset);
4052 static int kvm_clear_stat_per_vcpu(struct kvm *kvm, size_t offset)
4055 struct kvm_vcpu *vcpu;
4057 kvm_for_each_vcpu(i, vcpu, kvm)
4058 *(u64 *)((void *)vcpu + offset) = 0;
4063 static int kvm_stat_data_get(void *data, u64 *val)
4066 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
4068 switch (stat_data->dbgfs_item->kind) {
4070 r = kvm_get_stat_per_vm(stat_data->kvm,
4071 stat_data->dbgfs_item->offset, val);
4074 r = kvm_get_stat_per_vcpu(stat_data->kvm,
4075 stat_data->dbgfs_item->offset, val);
4082 static int kvm_stat_data_clear(void *data, u64 val)
4085 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
4090 switch (stat_data->dbgfs_item->kind) {
4092 r = kvm_clear_stat_per_vm(stat_data->kvm,
4093 stat_data->dbgfs_item->offset);
4096 r = kvm_clear_stat_per_vcpu(stat_data->kvm,
4097 stat_data->dbgfs_item->offset);
4104 static int kvm_stat_data_open(struct inode *inode, struct file *file)
4106 __simple_attr_check_format("%llu\n", 0ull);
4107 return kvm_debugfs_open(inode, file, kvm_stat_data_get,
4108 kvm_stat_data_clear, "%llu\n");
4111 static const struct file_operations stat_fops_per_vm = {
4112 .owner = THIS_MODULE,
4113 .open = kvm_stat_data_open,
4114 .release = kvm_debugfs_release,
4115 .read = simple_attr_read,
4116 .write = simple_attr_write,
4117 .llseek = no_llseek,
4120 static int vm_stat_get(void *_offset, u64 *val)
4122 unsigned offset = (long)_offset;
4127 mutex_lock(&kvm_lock);
4128 list_for_each_entry(kvm, &vm_list, vm_list) {
4129 kvm_get_stat_per_vm(kvm, offset, &tmp_val);
4132 mutex_unlock(&kvm_lock);
4136 static int vm_stat_clear(void *_offset, u64 val)
4138 unsigned offset = (long)_offset;
4144 mutex_lock(&kvm_lock);
4145 list_for_each_entry(kvm, &vm_list, vm_list) {
4146 kvm_clear_stat_per_vm(kvm, offset);
4148 mutex_unlock(&kvm_lock);
4153 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n");
4155 static int vcpu_stat_get(void *_offset, u64 *val)
4157 unsigned offset = (long)_offset;
4162 mutex_lock(&kvm_lock);
4163 list_for_each_entry(kvm, &vm_list, vm_list) {
4164 kvm_get_stat_per_vcpu(kvm, offset, &tmp_val);
4167 mutex_unlock(&kvm_lock);
4171 static int vcpu_stat_clear(void *_offset, u64 val)
4173 unsigned offset = (long)_offset;
4179 mutex_lock(&kvm_lock);
4180 list_for_each_entry(kvm, &vm_list, vm_list) {
4181 kvm_clear_stat_per_vcpu(kvm, offset);
4183 mutex_unlock(&kvm_lock);
4188 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear,
4191 static const struct file_operations *stat_fops[] = {
4192 [KVM_STAT_VCPU] = &vcpu_stat_fops,
4193 [KVM_STAT_VM] = &vm_stat_fops,
4196 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm)
4198 struct kobj_uevent_env *env;
4199 unsigned long long created, active;
4201 if (!kvm_dev.this_device || !kvm)
4204 mutex_lock(&kvm_lock);
4205 if (type == KVM_EVENT_CREATE_VM) {
4206 kvm_createvm_count++;
4208 } else if (type == KVM_EVENT_DESTROY_VM) {
4211 created = kvm_createvm_count;
4212 active = kvm_active_vms;
4213 mutex_unlock(&kvm_lock);
4215 env = kzalloc(sizeof(*env), GFP_KERNEL_ACCOUNT);
4219 add_uevent_var(env, "CREATED=%llu", created);
4220 add_uevent_var(env, "COUNT=%llu", active);
4222 if (type == KVM_EVENT_CREATE_VM) {
4223 add_uevent_var(env, "EVENT=create");
4224 kvm->userspace_pid = task_pid_nr(current);
4225 } else if (type == KVM_EVENT_DESTROY_VM) {
4226 add_uevent_var(env, "EVENT=destroy");
4228 add_uevent_var(env, "PID=%d", kvm->userspace_pid);
4230 if (!IS_ERR_OR_NULL(kvm->debugfs_dentry)) {
4231 char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL_ACCOUNT);
4234 tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX);
4236 add_uevent_var(env, "STATS_PATH=%s", tmp);
4240 /* no need for checks, since we are adding at most only 5 keys */
4241 env->envp[env->envp_idx++] = NULL;
4242 kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp);
4246 static void kvm_init_debug(void)
4248 struct kvm_stats_debugfs_item *p;
4250 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
4252 kvm_debugfs_num_entries = 0;
4253 for (p = debugfs_entries; p->name; ++p, kvm_debugfs_num_entries++) {
4254 debugfs_create_file(p->name, KVM_DBGFS_GET_MODE(p),
4255 kvm_debugfs_dir, (void *)(long)p->offset,
4256 stat_fops[p->kind]);
4260 static int kvm_suspend(void)
4262 if (kvm_usage_count)
4263 hardware_disable_nolock(NULL);
4267 static void kvm_resume(void)
4269 if (kvm_usage_count) {
4270 #ifdef CONFIG_LOCKDEP
4271 WARN_ON(lockdep_is_held(&kvm_count_lock));
4273 hardware_enable_nolock(NULL);
4277 static struct syscore_ops kvm_syscore_ops = {
4278 .suspend = kvm_suspend,
4279 .resume = kvm_resume,
4283 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
4285 return container_of(pn, struct kvm_vcpu, preempt_notifier);
4288 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
4290 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
4292 WRITE_ONCE(vcpu->preempted, false);
4293 WRITE_ONCE(vcpu->ready, false);
4295 __this_cpu_write(kvm_running_vcpu, vcpu);
4296 kvm_arch_sched_in(vcpu, cpu);
4297 kvm_arch_vcpu_load(vcpu, cpu);
4300 static void kvm_sched_out(struct preempt_notifier *pn,
4301 struct task_struct *next)
4303 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
4305 if (current->state == TASK_RUNNING) {
4306 WRITE_ONCE(vcpu->preempted, true);
4307 WRITE_ONCE(vcpu->ready, true);
4309 kvm_arch_vcpu_put(vcpu);
4310 __this_cpu_write(kvm_running_vcpu, NULL);
4314 * kvm_get_running_vcpu - get the vcpu running on the current CPU.
4315 * Thanks to preempt notifiers, this can also be called from
4316 * preemptible context.
4318 struct kvm_vcpu *kvm_get_running_vcpu(void)
4320 return __this_cpu_read(kvm_running_vcpu);
4324 * kvm_get_running_vcpus - get the per-CPU array of currently running vcpus.
4326 struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void)
4328 return &kvm_running_vcpu;
4331 static void check_processor_compat(void *rtn)
4333 *(int *)rtn = kvm_arch_check_processor_compat();
4336 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
4337 struct module *module)
4342 r = kvm_arch_init(opaque);
4347 * kvm_arch_init makes sure there's at most one caller
4348 * for architectures that support multiple implementations,
4349 * like intel and amd on x86.
4350 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
4351 * conflicts in case kvm is already setup for another implementation.
4353 r = kvm_irqfd_init();
4357 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
4362 r = kvm_arch_hardware_setup();
4366 for_each_online_cpu(cpu) {
4367 smp_call_function_single(cpu, check_processor_compat, &r, 1);
4372 r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting",
4373 kvm_starting_cpu, kvm_dying_cpu);
4376 register_reboot_notifier(&kvm_reboot_notifier);
4378 /* A kmem cache lets us meet the alignment requirements of fx_save. */
4380 vcpu_align = __alignof__(struct kvm_vcpu);
4382 kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align,
4384 offsetof(struct kvm_vcpu, arch),
4385 sizeof_field(struct kvm_vcpu, arch),
4387 if (!kvm_vcpu_cache) {
4392 r = kvm_async_pf_init();
4396 kvm_chardev_ops.owner = module;
4397 kvm_vm_fops.owner = module;
4398 kvm_vcpu_fops.owner = module;
4400 r = misc_register(&kvm_dev);
4402 pr_err("kvm: misc device register failed\n");
4406 register_syscore_ops(&kvm_syscore_ops);
4408 kvm_preempt_ops.sched_in = kvm_sched_in;
4409 kvm_preempt_ops.sched_out = kvm_sched_out;
4413 r = kvm_vfio_ops_init();
4419 kvm_async_pf_deinit();
4421 kmem_cache_destroy(kvm_vcpu_cache);
4423 unregister_reboot_notifier(&kvm_reboot_notifier);
4424 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4426 kvm_arch_hardware_unsetup();
4428 free_cpumask_var(cpus_hardware_enabled);
4436 EXPORT_SYMBOL_GPL(kvm_init);
4440 debugfs_remove_recursive(kvm_debugfs_dir);
4441 misc_deregister(&kvm_dev);
4442 kmem_cache_destroy(kvm_vcpu_cache);
4443 kvm_async_pf_deinit();
4444 unregister_syscore_ops(&kvm_syscore_ops);
4445 unregister_reboot_notifier(&kvm_reboot_notifier);
4446 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4447 on_each_cpu(hardware_disable_nolock, NULL, 1);
4448 kvm_arch_hardware_unsetup();
4451 free_cpumask_var(cpus_hardware_enabled);
4452 kvm_vfio_ops_exit();
4454 EXPORT_SYMBOL_GPL(kvm_exit);
4456 struct kvm_vm_worker_thread_context {
4458 struct task_struct *parent;
4459 struct completion init_done;
4460 kvm_vm_thread_fn_t thread_fn;
4465 static int kvm_vm_worker_thread(void *context)
4468 * The init_context is allocated on the stack of the parent thread, so
4469 * we have to locally copy anything that is needed beyond initialization
4471 struct kvm_vm_worker_thread_context *init_context = context;
4472 struct kvm *kvm = init_context->kvm;
4473 kvm_vm_thread_fn_t thread_fn = init_context->thread_fn;
4474 uintptr_t data = init_context->data;
4477 err = kthread_park(current);
4478 /* kthread_park(current) is never supposed to return an error */
4483 err = cgroup_attach_task_all(init_context->parent, current);
4485 kvm_err("%s: cgroup_attach_task_all failed with err %d\n",
4490 set_user_nice(current, task_nice(init_context->parent));
4493 init_context->err = err;
4494 complete(&init_context->init_done);
4495 init_context = NULL;
4500 /* Wait to be woken up by the spawner before proceeding. */
4503 if (!kthread_should_stop())
4504 err = thread_fn(kvm, data);
4509 int kvm_vm_create_worker_thread(struct kvm *kvm, kvm_vm_thread_fn_t thread_fn,
4510 uintptr_t data, const char *name,
4511 struct task_struct **thread_ptr)
4513 struct kvm_vm_worker_thread_context init_context = {};
4514 struct task_struct *thread;
4517 init_context.kvm = kvm;
4518 init_context.parent = current;
4519 init_context.thread_fn = thread_fn;
4520 init_context.data = data;
4521 init_completion(&init_context.init_done);
4523 thread = kthread_run(kvm_vm_worker_thread, &init_context,
4524 "%s-%d", name, task_pid_nr(current));
4526 return PTR_ERR(thread);
4528 /* kthread_run is never supposed to return NULL */
4529 WARN_ON(thread == NULL);
4531 wait_for_completion(&init_context.init_done);
4533 if (!init_context.err)
4534 *thread_ptr = thread;
4536 return init_context.err;