2 * Kernel-based Virtual Machine driver for Linux
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
7 * Copyright (C) 2006 Qumranet, Inc.
8 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
11 * Avi Kivity <avi@qumranet.com>
12 * Yaniv Kamay <yaniv@qumranet.com>
14 * This work is licensed under the terms of the GNU GPL, version 2. See
15 * the COPYING file in the top-level directory.
19 #include <kvm/iodev.h>
21 #include <linux/kvm_host.h>
22 #include <linux/kvm.h>
23 #include <linux/module.h>
24 #include <linux/errno.h>
25 #include <linux/percpu.h>
27 #include <linux/miscdevice.h>
28 #include <linux/vmalloc.h>
29 #include <linux/reboot.h>
30 #include <linux/debugfs.h>
31 #include <linux/highmem.h>
32 #include <linux/file.h>
33 #include <linux/syscore_ops.h>
34 #include <linux/cpu.h>
35 #include <linux/sched/signal.h>
36 #include <linux/sched/mm.h>
37 #include <linux/sched/stat.h>
38 #include <linux/cpumask.h>
39 #include <linux/smp.h>
40 #include <linux/anon_inodes.h>
41 #include <linux/profile.h>
42 #include <linux/kvm_para.h>
43 #include <linux/pagemap.h>
44 #include <linux/mman.h>
45 #include <linux/swap.h>
46 #include <linux/bitops.h>
47 #include <linux/spinlock.h>
48 #include <linux/compat.h>
49 #include <linux/srcu.h>
50 #include <linux/hugetlb.h>
51 #include <linux/slab.h>
52 #include <linux/sort.h>
53 #include <linux/bsearch.h>
55 #include <asm/processor.h>
57 #include <asm/ioctl.h>
58 #include <linux/uaccess.h>
59 #include <asm/pgtable.h>
61 #include "coalesced_mmio.h"
65 #define CREATE_TRACE_POINTS
66 #include <trace/events/kvm.h>
68 /* Worst case buffer size needed for holding an integer. */
69 #define ITOA_MAX_LEN 12
71 MODULE_AUTHOR("Qumranet");
72 MODULE_LICENSE("GPL");
74 /* Architectures should define their poll value according to the halt latency */
75 unsigned int halt_poll_ns = KVM_HALT_POLL_NS_DEFAULT;
76 module_param(halt_poll_ns, uint, 0644);
77 EXPORT_SYMBOL_GPL(halt_poll_ns);
79 /* Default doubles per-vcpu halt_poll_ns. */
80 unsigned int halt_poll_ns_grow = 2;
81 module_param(halt_poll_ns_grow, uint, 0644);
82 EXPORT_SYMBOL_GPL(halt_poll_ns_grow);
84 /* Default resets per-vcpu halt_poll_ns . */
85 unsigned int halt_poll_ns_shrink;
86 module_param(halt_poll_ns_shrink, uint, 0644);
87 EXPORT_SYMBOL_GPL(halt_poll_ns_shrink);
92 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
95 DEFINE_SPINLOCK(kvm_lock);
96 static DEFINE_RAW_SPINLOCK(kvm_count_lock);
99 static cpumask_var_t cpus_hardware_enabled;
100 static int kvm_usage_count;
101 static atomic_t hardware_enable_failed;
103 struct kmem_cache *kvm_vcpu_cache;
104 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
106 static __read_mostly struct preempt_ops kvm_preempt_ops;
108 struct dentry *kvm_debugfs_dir;
109 EXPORT_SYMBOL_GPL(kvm_debugfs_dir);
111 static int kvm_debugfs_num_entries;
112 static const struct file_operations *stat_fops_per_vm[];
114 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
116 #ifdef CONFIG_KVM_COMPAT
117 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
119 #define KVM_COMPAT(c) .compat_ioctl = (c)
121 static long kvm_no_compat_ioctl(struct file *file, unsigned int ioctl,
122 unsigned long arg) { return -EINVAL; }
123 #define KVM_COMPAT(c) .compat_ioctl = kvm_no_compat_ioctl
125 static int hardware_enable_all(void);
126 static void hardware_disable_all(void);
128 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
130 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot, gfn_t gfn);
132 __visible bool kvm_rebooting;
133 EXPORT_SYMBOL_GPL(kvm_rebooting);
135 static bool largepages_enabled = true;
137 #define KVM_EVENT_CREATE_VM 0
138 #define KVM_EVENT_DESTROY_VM 1
139 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm);
140 static unsigned long long kvm_createvm_count;
141 static unsigned long long kvm_active_vms;
143 __weak int kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
144 unsigned long start, unsigned long end, bool blockable)
149 bool kvm_is_reserved_pfn(kvm_pfn_t pfn)
152 return PageReserved(pfn_to_page(pfn));
158 * Switches to specified vcpu, until a matching vcpu_put()
160 void vcpu_load(struct kvm_vcpu *vcpu)
163 preempt_notifier_register(&vcpu->preempt_notifier);
164 kvm_arch_vcpu_load(vcpu, cpu);
167 EXPORT_SYMBOL_GPL(vcpu_load);
169 void vcpu_put(struct kvm_vcpu *vcpu)
172 kvm_arch_vcpu_put(vcpu);
173 preempt_notifier_unregister(&vcpu->preempt_notifier);
176 EXPORT_SYMBOL_GPL(vcpu_put);
178 /* TODO: merge with kvm_arch_vcpu_should_kick */
179 static bool kvm_request_needs_ipi(struct kvm_vcpu *vcpu, unsigned req)
181 int mode = kvm_vcpu_exiting_guest_mode(vcpu);
184 * We need to wait for the VCPU to reenable interrupts and get out of
185 * READING_SHADOW_PAGE_TABLES mode.
187 if (req & KVM_REQUEST_WAIT)
188 return mode != OUTSIDE_GUEST_MODE;
191 * Need to kick a running VCPU, but otherwise there is nothing to do.
193 return mode == IN_GUEST_MODE;
196 static void ack_flush(void *_completed)
200 static inline bool kvm_kick_many_cpus(const struct cpumask *cpus, bool wait)
203 cpus = cpu_online_mask;
205 if (cpumask_empty(cpus))
208 smp_call_function_many(cpus, ack_flush, NULL, wait);
212 bool kvm_make_vcpus_request_mask(struct kvm *kvm, unsigned int req,
213 unsigned long *vcpu_bitmap, cpumask_var_t tmp)
216 struct kvm_vcpu *vcpu;
221 kvm_for_each_vcpu(i, vcpu, kvm) {
222 if (vcpu_bitmap && !test_bit(i, vcpu_bitmap))
225 kvm_make_request(req, vcpu);
228 if (!(req & KVM_REQUEST_NO_WAKEUP) && kvm_vcpu_wake_up(vcpu))
231 if (tmp != NULL && cpu != -1 && cpu != me &&
232 kvm_request_needs_ipi(vcpu, req))
233 __cpumask_set_cpu(cpu, tmp);
236 called = kvm_kick_many_cpus(tmp, !!(req & KVM_REQUEST_WAIT));
242 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
247 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
249 called = kvm_make_vcpus_request_mask(kvm, req, NULL, cpus);
251 free_cpumask_var(cpus);
255 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
256 void kvm_flush_remote_tlbs(struct kvm *kvm)
259 * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in
260 * kvm_make_all_cpus_request.
262 long dirty_count = smp_load_acquire(&kvm->tlbs_dirty);
265 * We want to publish modifications to the page tables before reading
266 * mode. Pairs with a memory barrier in arch-specific code.
267 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
268 * and smp_mb in walk_shadow_page_lockless_begin/end.
269 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
271 * There is already an smp_mb__after_atomic() before
272 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
275 if (!kvm_arch_flush_remote_tlb(kvm)
276 || kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
277 ++kvm->stat.remote_tlb_flush;
278 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
280 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
283 void kvm_reload_remote_mmus(struct kvm *kvm)
285 kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
288 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
293 mutex_init(&vcpu->mutex);
298 init_swait_queue_head(&vcpu->wq);
299 kvm_async_pf_vcpu_init(vcpu);
302 INIT_LIST_HEAD(&vcpu->blocked_vcpu_list);
304 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
309 vcpu->run = page_address(page);
311 kvm_vcpu_set_in_spin_loop(vcpu, false);
312 kvm_vcpu_set_dy_eligible(vcpu, false);
313 vcpu->preempted = false;
315 r = kvm_arch_vcpu_init(vcpu);
321 free_page((unsigned long)vcpu->run);
325 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
327 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
330 * no need for rcu_read_lock as VCPU_RUN is the only place that
331 * will change the vcpu->pid pointer and on uninit all file
332 * descriptors are already gone.
334 put_pid(rcu_dereference_protected(vcpu->pid, 1));
335 kvm_arch_vcpu_uninit(vcpu);
336 free_page((unsigned long)vcpu->run);
338 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
340 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
341 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
343 return container_of(mn, struct kvm, mmu_notifier);
346 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
347 struct mm_struct *mm,
348 unsigned long address,
351 struct kvm *kvm = mmu_notifier_to_kvm(mn);
354 idx = srcu_read_lock(&kvm->srcu);
355 spin_lock(&kvm->mmu_lock);
356 kvm->mmu_notifier_seq++;
357 kvm_set_spte_hva(kvm, address, pte);
358 spin_unlock(&kvm->mmu_lock);
359 srcu_read_unlock(&kvm->srcu, idx);
362 static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
363 struct mm_struct *mm,
368 struct kvm *kvm = mmu_notifier_to_kvm(mn);
369 int need_tlb_flush = 0, idx;
372 idx = srcu_read_lock(&kvm->srcu);
373 spin_lock(&kvm->mmu_lock);
375 * The count increase must become visible at unlock time as no
376 * spte can be established without taking the mmu_lock and
377 * count is also read inside the mmu_lock critical section.
379 kvm->mmu_notifier_count++;
380 need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
381 need_tlb_flush |= kvm->tlbs_dirty;
382 /* we've to flush the tlb before the pages can be freed */
384 kvm_flush_remote_tlbs(kvm);
386 spin_unlock(&kvm->mmu_lock);
388 ret = kvm_arch_mmu_notifier_invalidate_range(kvm, start, end, blockable);
390 srcu_read_unlock(&kvm->srcu, idx);
395 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
396 struct mm_struct *mm,
400 struct kvm *kvm = mmu_notifier_to_kvm(mn);
402 spin_lock(&kvm->mmu_lock);
404 * This sequence increase will notify the kvm page fault that
405 * the page that is going to be mapped in the spte could have
408 kvm->mmu_notifier_seq++;
411 * The above sequence increase must be visible before the
412 * below count decrease, which is ensured by the smp_wmb above
413 * in conjunction with the smp_rmb in mmu_notifier_retry().
415 kvm->mmu_notifier_count--;
416 spin_unlock(&kvm->mmu_lock);
418 BUG_ON(kvm->mmu_notifier_count < 0);
421 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
422 struct mm_struct *mm,
426 struct kvm *kvm = mmu_notifier_to_kvm(mn);
429 idx = srcu_read_lock(&kvm->srcu);
430 spin_lock(&kvm->mmu_lock);
432 young = kvm_age_hva(kvm, start, end);
434 kvm_flush_remote_tlbs(kvm);
436 spin_unlock(&kvm->mmu_lock);
437 srcu_read_unlock(&kvm->srcu, idx);
442 static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
443 struct mm_struct *mm,
447 struct kvm *kvm = mmu_notifier_to_kvm(mn);
450 idx = srcu_read_lock(&kvm->srcu);
451 spin_lock(&kvm->mmu_lock);
453 * Even though we do not flush TLB, this will still adversely
454 * affect performance on pre-Haswell Intel EPT, where there is
455 * no EPT Access Bit to clear so that we have to tear down EPT
456 * tables instead. If we find this unacceptable, we can always
457 * add a parameter to kvm_age_hva so that it effectively doesn't
458 * do anything on clear_young.
460 * Also note that currently we never issue secondary TLB flushes
461 * from clear_young, leaving this job up to the regular system
462 * cadence. If we find this inaccurate, we might come up with a
463 * more sophisticated heuristic later.
465 young = kvm_age_hva(kvm, start, end);
466 spin_unlock(&kvm->mmu_lock);
467 srcu_read_unlock(&kvm->srcu, idx);
472 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
473 struct mm_struct *mm,
474 unsigned long address)
476 struct kvm *kvm = mmu_notifier_to_kvm(mn);
479 idx = srcu_read_lock(&kvm->srcu);
480 spin_lock(&kvm->mmu_lock);
481 young = kvm_test_age_hva(kvm, address);
482 spin_unlock(&kvm->mmu_lock);
483 srcu_read_unlock(&kvm->srcu, idx);
488 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
489 struct mm_struct *mm)
491 struct kvm *kvm = mmu_notifier_to_kvm(mn);
494 idx = srcu_read_lock(&kvm->srcu);
495 kvm_arch_flush_shadow_all(kvm);
496 srcu_read_unlock(&kvm->srcu, idx);
499 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
500 .flags = MMU_INVALIDATE_DOES_NOT_BLOCK,
501 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
502 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
503 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
504 .clear_young = kvm_mmu_notifier_clear_young,
505 .test_young = kvm_mmu_notifier_test_young,
506 .change_pte = kvm_mmu_notifier_change_pte,
507 .release = kvm_mmu_notifier_release,
510 static int kvm_init_mmu_notifier(struct kvm *kvm)
512 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
513 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
516 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
518 static int kvm_init_mmu_notifier(struct kvm *kvm)
523 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
525 static struct kvm_memslots *kvm_alloc_memslots(void)
528 struct kvm_memslots *slots;
530 slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
534 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
535 slots->id_to_index[i] = slots->memslots[i].id = i;
540 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
542 if (!memslot->dirty_bitmap)
545 kvfree(memslot->dirty_bitmap);
546 memslot->dirty_bitmap = NULL;
550 * Free any memory in @free but not in @dont.
552 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
553 struct kvm_memory_slot *dont)
555 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
556 kvm_destroy_dirty_bitmap(free);
558 kvm_arch_free_memslot(kvm, free, dont);
563 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
565 struct kvm_memory_slot *memslot;
570 kvm_for_each_memslot(memslot, slots)
571 kvm_free_memslot(kvm, memslot, NULL);
576 static void kvm_destroy_vm_debugfs(struct kvm *kvm)
580 if (!kvm->debugfs_dentry)
583 debugfs_remove_recursive(kvm->debugfs_dentry);
585 if (kvm->debugfs_stat_data) {
586 for (i = 0; i < kvm_debugfs_num_entries; i++)
587 kfree(kvm->debugfs_stat_data[i]);
588 kfree(kvm->debugfs_stat_data);
592 static int kvm_create_vm_debugfs(struct kvm *kvm, int fd)
594 char dir_name[ITOA_MAX_LEN * 2];
595 struct kvm_stat_data *stat_data;
596 struct kvm_stats_debugfs_item *p;
598 if (!debugfs_initialized())
601 snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd);
602 kvm->debugfs_dentry = debugfs_create_dir(dir_name, kvm_debugfs_dir);
604 kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries,
605 sizeof(*kvm->debugfs_stat_data),
607 if (!kvm->debugfs_stat_data)
610 for (p = debugfs_entries; p->name; p++) {
611 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL);
615 stat_data->kvm = kvm;
616 stat_data->offset = p->offset;
617 kvm->debugfs_stat_data[p - debugfs_entries] = stat_data;
618 debugfs_create_file(p->name, 0644, kvm->debugfs_dentry,
619 stat_data, stat_fops_per_vm[p->kind]);
624 static struct kvm *kvm_create_vm(unsigned long type)
627 struct kvm *kvm = kvm_arch_alloc_vm();
630 return ERR_PTR(-ENOMEM);
632 spin_lock_init(&kvm->mmu_lock);
634 kvm->mm = current->mm;
635 kvm_eventfd_init(kvm);
636 mutex_init(&kvm->lock);
637 mutex_init(&kvm->irq_lock);
638 mutex_init(&kvm->slots_lock);
639 refcount_set(&kvm->users_count, 1);
640 INIT_LIST_HEAD(&kvm->devices);
642 r = kvm_arch_init_vm(kvm, type);
644 goto out_err_no_disable;
646 r = hardware_enable_all();
648 goto out_err_no_disable;
650 #ifdef CONFIG_HAVE_KVM_IRQFD
651 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
654 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
657 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
658 struct kvm_memslots *slots = kvm_alloc_memslots();
660 goto out_err_no_srcu;
662 * Generations must be different for each address space.
663 * Init kvm generation close to the maximum to easily test the
664 * code of handling generation number wrap-around.
666 slots->generation = i * 2 - 150;
667 rcu_assign_pointer(kvm->memslots[i], slots);
670 if (init_srcu_struct(&kvm->srcu))
671 goto out_err_no_srcu;
672 if (init_srcu_struct(&kvm->irq_srcu))
673 goto out_err_no_irq_srcu;
674 for (i = 0; i < KVM_NR_BUSES; i++) {
675 rcu_assign_pointer(kvm->buses[i],
676 kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL));
681 r = kvm_init_mmu_notifier(kvm);
685 spin_lock(&kvm_lock);
686 list_add(&kvm->vm_list, &vm_list);
687 spin_unlock(&kvm_lock);
689 preempt_notifier_inc();
694 cleanup_srcu_struct(&kvm->irq_srcu);
696 cleanup_srcu_struct(&kvm->srcu);
698 hardware_disable_all();
700 refcount_set(&kvm->users_count, 0);
701 for (i = 0; i < KVM_NR_BUSES; i++)
702 kfree(kvm_get_bus(kvm, i));
703 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
704 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
705 kvm_arch_free_vm(kvm);
710 static void kvm_destroy_devices(struct kvm *kvm)
712 struct kvm_device *dev, *tmp;
715 * We do not need to take the kvm->lock here, because nobody else
716 * has a reference to the struct kvm at this point and therefore
717 * cannot access the devices list anyhow.
719 list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
720 list_del(&dev->vm_node);
721 dev->ops->destroy(dev);
725 static void kvm_destroy_vm(struct kvm *kvm)
728 struct mm_struct *mm = kvm->mm;
730 kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm);
731 kvm_destroy_vm_debugfs(kvm);
732 kvm_arch_sync_events(kvm);
733 spin_lock(&kvm_lock);
734 list_del(&kvm->vm_list);
735 spin_unlock(&kvm_lock);
736 kvm_free_irq_routing(kvm);
737 for (i = 0; i < KVM_NR_BUSES; i++) {
738 struct kvm_io_bus *bus = kvm_get_bus(kvm, i);
741 kvm_io_bus_destroy(bus);
742 kvm->buses[i] = NULL;
744 kvm_coalesced_mmio_free(kvm);
745 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
746 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
748 kvm_arch_flush_shadow_all(kvm);
750 kvm_arch_destroy_vm(kvm);
751 kvm_destroy_devices(kvm);
752 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
753 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
754 cleanup_srcu_struct(&kvm->irq_srcu);
755 cleanup_srcu_struct(&kvm->srcu);
756 kvm_arch_free_vm(kvm);
757 preempt_notifier_dec();
758 hardware_disable_all();
762 void kvm_get_kvm(struct kvm *kvm)
764 refcount_inc(&kvm->users_count);
766 EXPORT_SYMBOL_GPL(kvm_get_kvm);
768 void kvm_put_kvm(struct kvm *kvm)
770 if (refcount_dec_and_test(&kvm->users_count))
773 EXPORT_SYMBOL_GPL(kvm_put_kvm);
776 static int kvm_vm_release(struct inode *inode, struct file *filp)
778 struct kvm *kvm = filp->private_data;
780 kvm_irqfd_release(kvm);
787 * Allocation size is twice as large as the actual dirty bitmap size.
788 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
790 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
792 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
794 memslot->dirty_bitmap = kvzalloc(dirty_bytes, GFP_KERNEL);
795 if (!memslot->dirty_bitmap)
802 * Insert memslot and re-sort memslots based on their GFN,
803 * so binary search could be used to lookup GFN.
804 * Sorting algorithm takes advantage of having initially
805 * sorted array and known changed memslot position.
807 static void update_memslots(struct kvm_memslots *slots,
808 struct kvm_memory_slot *new,
809 enum kvm_mr_change change)
812 int i = slots->id_to_index[id];
813 struct kvm_memory_slot *mslots = slots->memslots;
815 WARN_ON(mslots[i].id != id);
819 WARN_ON(mslots[i].npages || !new->npages);
823 WARN_ON(new->npages || !mslots[i].npages);
829 while (i < KVM_MEM_SLOTS_NUM - 1 &&
830 new->base_gfn <= mslots[i + 1].base_gfn) {
831 if (!mslots[i + 1].npages)
833 mslots[i] = mslots[i + 1];
834 slots->id_to_index[mslots[i].id] = i;
839 * The ">=" is needed when creating a slot with base_gfn == 0,
840 * so that it moves before all those with base_gfn == npages == 0.
842 * On the other hand, if new->npages is zero, the above loop has
843 * already left i pointing to the beginning of the empty part of
844 * mslots, and the ">=" would move the hole backwards in this
845 * case---which is wrong. So skip the loop when deleting a slot.
849 new->base_gfn >= mslots[i - 1].base_gfn) {
850 mslots[i] = mslots[i - 1];
851 slots->id_to_index[mslots[i].id] = i;
855 WARN_ON_ONCE(i != slots->used_slots);
858 slots->id_to_index[mslots[i].id] = i;
861 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
863 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
865 #ifdef __KVM_HAVE_READONLY_MEM
866 valid_flags |= KVM_MEM_READONLY;
869 if (mem->flags & ~valid_flags)
875 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
876 int as_id, struct kvm_memslots *slots)
878 struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id);
881 * Set the low bit in the generation, which disables SPTE caching
882 * until the end of synchronize_srcu_expedited.
884 WARN_ON(old_memslots->generation & 1);
885 slots->generation = old_memslots->generation + 1;
887 rcu_assign_pointer(kvm->memslots[as_id], slots);
888 synchronize_srcu_expedited(&kvm->srcu);
891 * Increment the new memslot generation a second time. This prevents
892 * vm exits that race with memslot updates from caching a memslot
893 * generation that will (potentially) be valid forever.
895 * Generations must be unique even across address spaces. We do not need
896 * a global counter for that, instead the generation space is evenly split
897 * across address spaces. For example, with two address spaces, address
898 * space 0 will use generations 0, 4, 8, ... while * address space 1 will
899 * use generations 2, 6, 10, 14, ...
901 slots->generation += KVM_ADDRESS_SPACE_NUM * 2 - 1;
903 kvm_arch_memslots_updated(kvm, slots);
909 * Allocate some memory and give it an address in the guest physical address
912 * Discontiguous memory is allowed, mostly for framebuffers.
914 * Must be called holding kvm->slots_lock for write.
916 int __kvm_set_memory_region(struct kvm *kvm,
917 const struct kvm_userspace_memory_region *mem)
921 unsigned long npages;
922 struct kvm_memory_slot *slot;
923 struct kvm_memory_slot old, new;
924 struct kvm_memslots *slots = NULL, *old_memslots;
926 enum kvm_mr_change change;
928 r = check_memory_region_flags(mem);
933 as_id = mem->slot >> 16;
936 /* General sanity checks */
937 if (mem->memory_size & (PAGE_SIZE - 1))
939 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
941 /* We can read the guest memory with __xxx_user() later on. */
942 if ((id < KVM_USER_MEM_SLOTS) &&
943 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
944 !access_ok(VERIFY_WRITE,
945 (void __user *)(unsigned long)mem->userspace_addr,
948 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
950 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
953 slot = id_to_memslot(__kvm_memslots(kvm, as_id), id);
954 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
955 npages = mem->memory_size >> PAGE_SHIFT;
957 if (npages > KVM_MEM_MAX_NR_PAGES)
963 new.base_gfn = base_gfn;
965 new.flags = mem->flags;
969 change = KVM_MR_CREATE;
970 else { /* Modify an existing slot. */
971 if ((mem->userspace_addr != old.userspace_addr) ||
972 (npages != old.npages) ||
973 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
976 if (base_gfn != old.base_gfn)
977 change = KVM_MR_MOVE;
978 else if (new.flags != old.flags)
979 change = KVM_MR_FLAGS_ONLY;
980 else { /* Nothing to change. */
989 change = KVM_MR_DELETE;
994 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
995 /* Check for overlaps */
997 kvm_for_each_memslot(slot, __kvm_memslots(kvm, as_id)) {
1000 if (!((base_gfn + npages <= slot->base_gfn) ||
1001 (base_gfn >= slot->base_gfn + slot->npages)))
1006 /* Free page dirty bitmap if unneeded */
1007 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
1008 new.dirty_bitmap = NULL;
1011 if (change == KVM_MR_CREATE) {
1012 new.userspace_addr = mem->userspace_addr;
1014 if (kvm_arch_create_memslot(kvm, &new, npages))
1018 /* Allocate page dirty bitmap if needed */
1019 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
1020 if (kvm_create_dirty_bitmap(&new) < 0)
1024 slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
1027 memcpy(slots, __kvm_memslots(kvm, as_id), sizeof(struct kvm_memslots));
1029 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
1030 slot = id_to_memslot(slots, id);
1031 slot->flags |= KVM_MEMSLOT_INVALID;
1033 old_memslots = install_new_memslots(kvm, as_id, slots);
1035 /* From this point no new shadow pages pointing to a deleted,
1036 * or moved, memslot will be created.
1038 * validation of sp->gfn happens in:
1039 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1040 * - kvm_is_visible_gfn (mmu_check_roots)
1042 kvm_arch_flush_shadow_memslot(kvm, slot);
1045 * We can re-use the old_memslots from above, the only difference
1046 * from the currently installed memslots is the invalid flag. This
1047 * will get overwritten by update_memslots anyway.
1049 slots = old_memslots;
1052 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
1056 /* actual memory is freed via old in kvm_free_memslot below */
1057 if (change == KVM_MR_DELETE) {
1058 new.dirty_bitmap = NULL;
1059 memset(&new.arch, 0, sizeof(new.arch));
1062 update_memslots(slots, &new, change);
1063 old_memslots = install_new_memslots(kvm, as_id, slots);
1065 kvm_arch_commit_memory_region(kvm, mem, &old, &new, change);
1067 kvm_free_memslot(kvm, &old, &new);
1068 kvfree(old_memslots);
1074 kvm_free_memslot(kvm, &new, &old);
1078 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1080 int kvm_set_memory_region(struct kvm *kvm,
1081 const struct kvm_userspace_memory_region *mem)
1085 mutex_lock(&kvm->slots_lock);
1086 r = __kvm_set_memory_region(kvm, mem);
1087 mutex_unlock(&kvm->slots_lock);
1090 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1092 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1093 struct kvm_userspace_memory_region *mem)
1095 if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1098 return kvm_set_memory_region(kvm, mem);
1101 int kvm_get_dirty_log(struct kvm *kvm,
1102 struct kvm_dirty_log *log, int *is_dirty)
1104 struct kvm_memslots *slots;
1105 struct kvm_memory_slot *memslot;
1108 unsigned long any = 0;
1110 as_id = log->slot >> 16;
1111 id = (u16)log->slot;
1112 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1115 slots = __kvm_memslots(kvm, as_id);
1116 memslot = id_to_memslot(slots, id);
1117 if (!memslot->dirty_bitmap)
1120 n = kvm_dirty_bitmap_bytes(memslot);
1122 for (i = 0; !any && i < n/sizeof(long); ++i)
1123 any = memslot->dirty_bitmap[i];
1125 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1132 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1134 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1136 * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages
1137 * are dirty write protect them for next write.
1138 * @kvm: pointer to kvm instance
1139 * @log: slot id and address to which we copy the log
1140 * @is_dirty: flag set if any page is dirty
1142 * We need to keep it in mind that VCPU threads can write to the bitmap
1143 * concurrently. So, to avoid losing track of dirty pages we keep the
1146 * 1. Take a snapshot of the bit and clear it if needed.
1147 * 2. Write protect the corresponding page.
1148 * 3. Copy the snapshot to the userspace.
1149 * 4. Upon return caller flushes TLB's if needed.
1151 * Between 2 and 4, the guest may write to the page using the remaining TLB
1152 * entry. This is not a problem because the page is reported dirty using
1153 * the snapshot taken before and step 4 ensures that writes done after
1154 * exiting to userspace will be logged for the next call.
1157 int kvm_get_dirty_log_protect(struct kvm *kvm,
1158 struct kvm_dirty_log *log, bool *is_dirty)
1160 struct kvm_memslots *slots;
1161 struct kvm_memory_slot *memslot;
1164 unsigned long *dirty_bitmap;
1165 unsigned long *dirty_bitmap_buffer;
1167 as_id = log->slot >> 16;
1168 id = (u16)log->slot;
1169 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1172 slots = __kvm_memslots(kvm, as_id);
1173 memslot = id_to_memslot(slots, id);
1175 dirty_bitmap = memslot->dirty_bitmap;
1179 n = kvm_dirty_bitmap_bytes(memslot);
1181 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
1182 memset(dirty_bitmap_buffer, 0, n);
1184 spin_lock(&kvm->mmu_lock);
1186 for (i = 0; i < n / sizeof(long); i++) {
1190 if (!dirty_bitmap[i])
1195 mask = xchg(&dirty_bitmap[i], 0);
1196 dirty_bitmap_buffer[i] = mask;
1199 offset = i * BITS_PER_LONG;
1200 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1205 spin_unlock(&kvm->mmu_lock);
1206 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1210 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect);
1213 bool kvm_largepages_enabled(void)
1215 return largepages_enabled;
1218 void kvm_disable_largepages(void)
1220 largepages_enabled = false;
1222 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1224 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1226 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1228 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1230 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
1232 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn);
1235 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1237 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1239 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1240 memslot->flags & KVM_MEMSLOT_INVALID)
1245 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1247 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1249 struct vm_area_struct *vma;
1250 unsigned long addr, size;
1254 addr = gfn_to_hva(kvm, gfn);
1255 if (kvm_is_error_hva(addr))
1258 down_read(¤t->mm->mmap_sem);
1259 vma = find_vma(current->mm, addr);
1263 size = vma_kernel_pagesize(vma);
1266 up_read(¤t->mm->mmap_sem);
1271 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1273 return slot->flags & KVM_MEM_READONLY;
1276 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1277 gfn_t *nr_pages, bool write)
1279 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1280 return KVM_HVA_ERR_BAD;
1282 if (memslot_is_readonly(slot) && write)
1283 return KVM_HVA_ERR_RO_BAD;
1286 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1288 return __gfn_to_hva_memslot(slot, gfn);
1291 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1294 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1297 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1300 return gfn_to_hva_many(slot, gfn, NULL);
1302 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1304 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1306 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1308 EXPORT_SYMBOL_GPL(gfn_to_hva);
1310 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
1312 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
1314 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
1317 * Return the hva of a @gfn and the R/W attribute if possible.
1319 * @slot: the kvm_memory_slot which contains @gfn
1320 * @gfn: the gfn to be translated
1321 * @writable: used to return the read/write attribute of the @slot if the hva
1322 * is valid and @writable is not NULL
1324 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1325 gfn_t gfn, bool *writable)
1327 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1329 if (!kvm_is_error_hva(hva) && writable)
1330 *writable = !memslot_is_readonly(slot);
1335 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1337 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1339 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1342 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
1344 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1346 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1349 static inline int check_user_page_hwpoison(unsigned long addr)
1351 int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
1353 rc = get_user_pages(addr, 1, flags, NULL, NULL);
1354 return rc == -EHWPOISON;
1358 * The fast path to get the writable pfn which will be stored in @pfn,
1359 * true indicates success, otherwise false is returned. It's also the
1360 * only part that runs if we can are in atomic context.
1362 static bool hva_to_pfn_fast(unsigned long addr, bool write_fault,
1363 bool *writable, kvm_pfn_t *pfn)
1365 struct page *page[1];
1369 * Fast pin a writable pfn only if it is a write fault request
1370 * or the caller allows to map a writable pfn for a read fault
1373 if (!(write_fault || writable))
1376 npages = __get_user_pages_fast(addr, 1, 1, page);
1378 *pfn = page_to_pfn(page[0]);
1389 * The slow path to get the pfn of the specified host virtual address,
1390 * 1 indicates success, -errno is returned if error is detected.
1392 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1393 bool *writable, kvm_pfn_t *pfn)
1395 unsigned int flags = FOLL_HWPOISON;
1402 *writable = write_fault;
1405 flags |= FOLL_WRITE;
1407 flags |= FOLL_NOWAIT;
1409 npages = get_user_pages_unlocked(addr, 1, &page, flags);
1413 /* map read fault as writable if possible */
1414 if (unlikely(!write_fault) && writable) {
1417 if (__get_user_pages_fast(addr, 1, 1, &wpage) == 1) {
1423 *pfn = page_to_pfn(page);
1427 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1429 if (unlikely(!(vma->vm_flags & VM_READ)))
1432 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1438 static int hva_to_pfn_remapped(struct vm_area_struct *vma,
1439 unsigned long addr, bool *async,
1440 bool write_fault, bool *writable,
1446 r = follow_pfn(vma, addr, &pfn);
1449 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
1450 * not call the fault handler, so do it here.
1452 bool unlocked = false;
1453 r = fixup_user_fault(current, current->mm, addr,
1454 (write_fault ? FAULT_FLAG_WRITE : 0),
1461 r = follow_pfn(vma, addr, &pfn);
1471 * Get a reference here because callers of *hva_to_pfn* and
1472 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
1473 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
1474 * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will
1475 * simply do nothing for reserved pfns.
1477 * Whoever called remap_pfn_range is also going to call e.g.
1478 * unmap_mapping_range before the underlying pages are freed,
1479 * causing a call to our MMU notifier.
1488 * Pin guest page in memory and return its pfn.
1489 * @addr: host virtual address which maps memory to the guest
1490 * @atomic: whether this function can sleep
1491 * @async: whether this function need to wait IO complete if the
1492 * host page is not in the memory
1493 * @write_fault: whether we should get a writable host page
1494 * @writable: whether it allows to map a writable host page for !@write_fault
1496 * The function will map a writable host page for these two cases:
1497 * 1): @write_fault = true
1498 * 2): @write_fault = false && @writable, @writable will tell the caller
1499 * whether the mapping is writable.
1501 static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1502 bool write_fault, bool *writable)
1504 struct vm_area_struct *vma;
1508 /* we can do it either atomically or asynchronously, not both */
1509 BUG_ON(atomic && async);
1511 if (hva_to_pfn_fast(addr, write_fault, writable, &pfn))
1515 return KVM_PFN_ERR_FAULT;
1517 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1521 down_read(¤t->mm->mmap_sem);
1522 if (npages == -EHWPOISON ||
1523 (!async && check_user_page_hwpoison(addr))) {
1524 pfn = KVM_PFN_ERR_HWPOISON;
1529 vma = find_vma_intersection(current->mm, addr, addr + 1);
1532 pfn = KVM_PFN_ERR_FAULT;
1533 else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
1534 r = hva_to_pfn_remapped(vma, addr, async, write_fault, writable, &pfn);
1538 pfn = KVM_PFN_ERR_FAULT;
1540 if (async && vma_is_valid(vma, write_fault))
1542 pfn = KVM_PFN_ERR_FAULT;
1545 up_read(¤t->mm->mmap_sem);
1549 kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn,
1550 bool atomic, bool *async, bool write_fault,
1553 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1555 if (addr == KVM_HVA_ERR_RO_BAD) {
1558 return KVM_PFN_ERR_RO_FAULT;
1561 if (kvm_is_error_hva(addr)) {
1564 return KVM_PFN_NOSLOT;
1567 /* Do not map writable pfn in the readonly memslot. */
1568 if (writable && memslot_is_readonly(slot)) {
1573 return hva_to_pfn(addr, atomic, async, write_fault,
1576 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
1578 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1581 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
1582 write_fault, writable);
1584 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1586 kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1588 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1590 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
1592 kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1594 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1596 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1598 kvm_pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1600 return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm, gfn), gfn);
1602 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1604 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
1606 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1608 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
1610 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1612 return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
1614 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1616 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
1618 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1620 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
1622 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1623 struct page **pages, int nr_pages)
1628 addr = gfn_to_hva_many(slot, gfn, &entry);
1629 if (kvm_is_error_hva(addr))
1632 if (entry < nr_pages)
1635 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1637 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1639 static struct page *kvm_pfn_to_page(kvm_pfn_t pfn)
1641 if (is_error_noslot_pfn(pfn))
1642 return KVM_ERR_PTR_BAD_PAGE;
1644 if (kvm_is_reserved_pfn(pfn)) {
1646 return KVM_ERR_PTR_BAD_PAGE;
1649 return pfn_to_page(pfn);
1652 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1656 pfn = gfn_to_pfn(kvm, gfn);
1658 return kvm_pfn_to_page(pfn);
1660 EXPORT_SYMBOL_GPL(gfn_to_page);
1662 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
1666 pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
1668 return kvm_pfn_to_page(pfn);
1670 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
1672 void kvm_release_page_clean(struct page *page)
1674 WARN_ON(is_error_page(page));
1676 kvm_release_pfn_clean(page_to_pfn(page));
1678 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1680 void kvm_release_pfn_clean(kvm_pfn_t pfn)
1682 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
1683 put_page(pfn_to_page(pfn));
1685 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1687 void kvm_release_page_dirty(struct page *page)
1689 WARN_ON(is_error_page(page));
1691 kvm_release_pfn_dirty(page_to_pfn(page));
1693 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1695 void kvm_release_pfn_dirty(kvm_pfn_t pfn)
1697 kvm_set_pfn_dirty(pfn);
1698 kvm_release_pfn_clean(pfn);
1700 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1702 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
1704 if (!kvm_is_reserved_pfn(pfn)) {
1705 struct page *page = pfn_to_page(pfn);
1707 if (!PageReserved(page))
1711 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1713 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
1715 if (!kvm_is_reserved_pfn(pfn))
1716 mark_page_accessed(pfn_to_page(pfn));
1718 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1720 void kvm_get_pfn(kvm_pfn_t pfn)
1722 if (!kvm_is_reserved_pfn(pfn))
1723 get_page(pfn_to_page(pfn));
1725 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1727 static int next_segment(unsigned long len, int offset)
1729 if (len > PAGE_SIZE - offset)
1730 return PAGE_SIZE - offset;
1735 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
1736 void *data, int offset, int len)
1741 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1742 if (kvm_is_error_hva(addr))
1744 r = __copy_from_user(data, (void __user *)addr + offset, len);
1750 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1753 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1755 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1757 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1759 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
1760 int offset, int len)
1762 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1764 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1766 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
1768 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1770 gfn_t gfn = gpa >> PAGE_SHIFT;
1772 int offset = offset_in_page(gpa);
1775 while ((seg = next_segment(len, offset)) != 0) {
1776 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1786 EXPORT_SYMBOL_GPL(kvm_read_guest);
1788 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
1790 gfn_t gfn = gpa >> PAGE_SHIFT;
1792 int offset = offset_in_page(gpa);
1795 while ((seg = next_segment(len, offset)) != 0) {
1796 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
1806 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
1808 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1809 void *data, int offset, unsigned long len)
1814 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1815 if (kvm_is_error_hva(addr))
1817 pagefault_disable();
1818 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1825 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1828 gfn_t gfn = gpa >> PAGE_SHIFT;
1829 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1830 int offset = offset_in_page(gpa);
1832 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1834 EXPORT_SYMBOL_GPL(kvm_read_guest_atomic);
1836 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
1837 void *data, unsigned long len)
1839 gfn_t gfn = gpa >> PAGE_SHIFT;
1840 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1841 int offset = offset_in_page(gpa);
1843 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1845 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
1847 static int __kvm_write_guest_page(struct kvm_memory_slot *memslot, gfn_t gfn,
1848 const void *data, int offset, int len)
1853 addr = gfn_to_hva_memslot(memslot, gfn);
1854 if (kvm_is_error_hva(addr))
1856 r = __copy_to_user((void __user *)addr + offset, data, len);
1859 mark_page_dirty_in_slot(memslot, gfn);
1863 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
1864 const void *data, int offset, int len)
1866 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1868 return __kvm_write_guest_page(slot, gfn, data, offset, len);
1870 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1872 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
1873 const void *data, int offset, int len)
1875 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1877 return __kvm_write_guest_page(slot, gfn, data, offset, len);
1879 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
1881 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1884 gfn_t gfn = gpa >> PAGE_SHIFT;
1886 int offset = offset_in_page(gpa);
1889 while ((seg = next_segment(len, offset)) != 0) {
1890 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1900 EXPORT_SYMBOL_GPL(kvm_write_guest);
1902 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
1905 gfn_t gfn = gpa >> PAGE_SHIFT;
1907 int offset = offset_in_page(gpa);
1910 while ((seg = next_segment(len, offset)) != 0) {
1911 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
1921 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
1923 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
1924 struct gfn_to_hva_cache *ghc,
1925 gpa_t gpa, unsigned long len)
1927 int offset = offset_in_page(gpa);
1928 gfn_t start_gfn = gpa >> PAGE_SHIFT;
1929 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1930 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1931 gfn_t nr_pages_avail;
1934 ghc->generation = slots->generation;
1936 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
1937 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, NULL);
1938 if (!kvm_is_error_hva(ghc->hva) && nr_pages_needed <= 1) {
1942 * If the requested region crosses two memslots, we still
1943 * verify that the entire region is valid here.
1945 while (start_gfn <= end_gfn) {
1947 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
1948 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1950 if (kvm_is_error_hva(ghc->hva))
1952 start_gfn += nr_pages_avail;
1954 /* Use the slow path for cross page reads and writes. */
1955 ghc->memslot = NULL;
1960 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1961 gpa_t gpa, unsigned long len)
1963 struct kvm_memslots *slots = kvm_memslots(kvm);
1964 return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len);
1966 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1968 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1969 void *data, int offset, unsigned long len)
1971 struct kvm_memslots *slots = kvm_memslots(kvm);
1973 gpa_t gpa = ghc->gpa + offset;
1975 BUG_ON(len + offset > ghc->len);
1977 if (slots->generation != ghc->generation)
1978 __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len);
1980 if (unlikely(!ghc->memslot))
1981 return kvm_write_guest(kvm, gpa, data, len);
1983 if (kvm_is_error_hva(ghc->hva))
1986 r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
1989 mark_page_dirty_in_slot(ghc->memslot, gpa >> PAGE_SHIFT);
1993 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached);
1995 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1996 void *data, unsigned long len)
1998 return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len);
2000 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
2002 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2003 void *data, unsigned long len)
2005 struct kvm_memslots *slots = kvm_memslots(kvm);
2008 BUG_ON(len > ghc->len);
2010 if (slots->generation != ghc->generation)
2011 __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len);
2013 if (unlikely(!ghc->memslot))
2014 return kvm_read_guest(kvm, ghc->gpa, data, len);
2016 if (kvm_is_error_hva(ghc->hva))
2019 r = __copy_from_user(data, (void __user *)ghc->hva, len);
2025 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
2027 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
2029 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
2031 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
2033 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
2035 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
2037 gfn_t gfn = gpa >> PAGE_SHIFT;
2039 int offset = offset_in_page(gpa);
2042 while ((seg = next_segment(len, offset)) != 0) {
2043 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
2052 EXPORT_SYMBOL_GPL(kvm_clear_guest);
2054 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot,
2057 if (memslot && memslot->dirty_bitmap) {
2058 unsigned long rel_gfn = gfn - memslot->base_gfn;
2060 set_bit_le(rel_gfn, memslot->dirty_bitmap);
2064 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
2066 struct kvm_memory_slot *memslot;
2068 memslot = gfn_to_memslot(kvm, gfn);
2069 mark_page_dirty_in_slot(memslot, gfn);
2071 EXPORT_SYMBOL_GPL(mark_page_dirty);
2073 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
2075 struct kvm_memory_slot *memslot;
2077 memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2078 mark_page_dirty_in_slot(memslot, gfn);
2080 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
2082 void kvm_sigset_activate(struct kvm_vcpu *vcpu)
2084 if (!vcpu->sigset_active)
2088 * This does a lockless modification of ->real_blocked, which is fine
2089 * because, only current can change ->real_blocked and all readers of
2090 * ->real_blocked don't care as long ->real_blocked is always a subset
2093 sigprocmask(SIG_SETMASK, &vcpu->sigset, ¤t->real_blocked);
2096 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu)
2098 if (!vcpu->sigset_active)
2101 sigprocmask(SIG_SETMASK, ¤t->real_blocked, NULL);
2102 sigemptyset(¤t->real_blocked);
2105 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
2107 unsigned int old, val, grow;
2109 old = val = vcpu->halt_poll_ns;
2110 grow = READ_ONCE(halt_poll_ns_grow);
2112 if (val == 0 && grow)
2117 if (val > halt_poll_ns)
2120 vcpu->halt_poll_ns = val;
2121 trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
2124 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
2126 unsigned int old, val, shrink;
2128 old = val = vcpu->halt_poll_ns;
2129 shrink = READ_ONCE(halt_poll_ns_shrink);
2135 vcpu->halt_poll_ns = val;
2136 trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
2139 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
2142 int idx = srcu_read_lock(&vcpu->kvm->srcu);
2144 if (kvm_arch_vcpu_runnable(vcpu)) {
2145 kvm_make_request(KVM_REQ_UNHALT, vcpu);
2148 if (kvm_cpu_has_pending_timer(vcpu))
2150 if (signal_pending(current))
2155 srcu_read_unlock(&vcpu->kvm->srcu, idx);
2160 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2162 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
2165 DECLARE_SWAITQUEUE(wait);
2166 bool waited = false;
2169 start = cur = ktime_get();
2170 if (vcpu->halt_poll_ns) {
2171 ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns);
2173 ++vcpu->stat.halt_attempted_poll;
2176 * This sets KVM_REQ_UNHALT if an interrupt
2179 if (kvm_vcpu_check_block(vcpu) < 0) {
2180 ++vcpu->stat.halt_successful_poll;
2181 if (!vcpu_valid_wakeup(vcpu))
2182 ++vcpu->stat.halt_poll_invalid;
2186 } while (single_task_running() && ktime_before(cur, stop));
2189 kvm_arch_vcpu_blocking(vcpu);
2192 prepare_to_swait_exclusive(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
2194 if (kvm_vcpu_check_block(vcpu) < 0)
2201 finish_swait(&vcpu->wq, &wait);
2204 kvm_arch_vcpu_unblocking(vcpu);
2206 block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
2208 if (!vcpu_valid_wakeup(vcpu))
2209 shrink_halt_poll_ns(vcpu);
2210 else if (halt_poll_ns) {
2211 if (block_ns <= vcpu->halt_poll_ns)
2213 /* we had a long block, shrink polling */
2214 else if (vcpu->halt_poll_ns && block_ns > halt_poll_ns)
2215 shrink_halt_poll_ns(vcpu);
2216 /* we had a short halt and our poll time is too small */
2217 else if (vcpu->halt_poll_ns < halt_poll_ns &&
2218 block_ns < halt_poll_ns)
2219 grow_halt_poll_ns(vcpu);
2221 vcpu->halt_poll_ns = 0;
2223 trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu));
2224 kvm_arch_vcpu_block_finish(vcpu);
2226 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
2228 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
2230 struct swait_queue_head *wqp;
2232 wqp = kvm_arch_vcpu_wq(vcpu);
2233 if (swq_has_sleeper(wqp)) {
2235 ++vcpu->stat.halt_wakeup;
2241 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
2245 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2247 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
2250 int cpu = vcpu->cpu;
2252 if (kvm_vcpu_wake_up(vcpu))
2256 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
2257 if (kvm_arch_vcpu_should_kick(vcpu))
2258 smp_send_reschedule(cpu);
2261 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
2262 #endif /* !CONFIG_S390 */
2264 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
2267 struct task_struct *task = NULL;
2271 pid = rcu_dereference(target->pid);
2273 task = get_pid_task(pid, PIDTYPE_PID);
2277 ret = yield_to(task, 1);
2278 put_task_struct(task);
2282 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
2285 * Helper that checks whether a VCPU is eligible for directed yield.
2286 * Most eligible candidate to yield is decided by following heuristics:
2288 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2289 * (preempted lock holder), indicated by @in_spin_loop.
2290 * Set at the beiginning and cleared at the end of interception/PLE handler.
2292 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2293 * chance last time (mostly it has become eligible now since we have probably
2294 * yielded to lockholder in last iteration. This is done by toggling
2295 * @dy_eligible each time a VCPU checked for eligibility.)
2297 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2298 * to preempted lock-holder could result in wrong VCPU selection and CPU
2299 * burning. Giving priority for a potential lock-holder increases lock
2302 * Since algorithm is based on heuristics, accessing another VCPU data without
2303 * locking does not harm. It may result in trying to yield to same VCPU, fail
2304 * and continue with next VCPU and so on.
2306 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
2308 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2311 eligible = !vcpu->spin_loop.in_spin_loop ||
2312 vcpu->spin_loop.dy_eligible;
2314 if (vcpu->spin_loop.in_spin_loop)
2315 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
2323 void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode)
2325 struct kvm *kvm = me->kvm;
2326 struct kvm_vcpu *vcpu;
2327 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
2333 kvm_vcpu_set_in_spin_loop(me, true);
2335 * We boost the priority of a VCPU that is runnable but not
2336 * currently running, because it got preempted by something
2337 * else and called schedule in __vcpu_run. Hopefully that
2338 * VCPU is holding the lock that we need and will release it.
2339 * We approximate round-robin by starting at the last boosted VCPU.
2341 for (pass = 0; pass < 2 && !yielded && try; pass++) {
2342 kvm_for_each_vcpu(i, vcpu, kvm) {
2343 if (!pass && i <= last_boosted_vcpu) {
2344 i = last_boosted_vcpu;
2346 } else if (pass && i > last_boosted_vcpu)
2348 if (!READ_ONCE(vcpu->preempted))
2352 if (swait_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
2354 if (yield_to_kernel_mode && !kvm_arch_vcpu_in_kernel(vcpu))
2356 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
2359 yielded = kvm_vcpu_yield_to(vcpu);
2361 kvm->last_boosted_vcpu = i;
2363 } else if (yielded < 0) {
2370 kvm_vcpu_set_in_spin_loop(me, false);
2372 /* Ensure vcpu is not eligible during next spinloop */
2373 kvm_vcpu_set_dy_eligible(me, false);
2375 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
2377 static vm_fault_t kvm_vcpu_fault(struct vm_fault *vmf)
2379 struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data;
2382 if (vmf->pgoff == 0)
2383 page = virt_to_page(vcpu->run);
2385 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
2386 page = virt_to_page(vcpu->arch.pio_data);
2388 #ifdef CONFIG_KVM_MMIO
2389 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
2390 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
2393 return kvm_arch_vcpu_fault(vcpu, vmf);
2399 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
2400 .fault = kvm_vcpu_fault,
2403 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2405 vma->vm_ops = &kvm_vcpu_vm_ops;
2409 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2411 struct kvm_vcpu *vcpu = filp->private_data;
2413 debugfs_remove_recursive(vcpu->debugfs_dentry);
2414 kvm_put_kvm(vcpu->kvm);
2418 static struct file_operations kvm_vcpu_fops = {
2419 .release = kvm_vcpu_release,
2420 .unlocked_ioctl = kvm_vcpu_ioctl,
2421 .mmap = kvm_vcpu_mmap,
2422 .llseek = noop_llseek,
2423 KVM_COMPAT(kvm_vcpu_compat_ioctl),
2427 * Allocates an inode for the vcpu.
2429 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2431 char name[8 + 1 + ITOA_MAX_LEN + 1];
2433 snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id);
2434 return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
2437 static int kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
2439 char dir_name[ITOA_MAX_LEN * 2];
2442 if (!kvm_arch_has_vcpu_debugfs())
2445 if (!debugfs_initialized())
2448 snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
2449 vcpu->debugfs_dentry = debugfs_create_dir(dir_name,
2450 vcpu->kvm->debugfs_dentry);
2451 if (!vcpu->debugfs_dentry)
2454 ret = kvm_arch_create_vcpu_debugfs(vcpu);
2456 debugfs_remove_recursive(vcpu->debugfs_dentry);
2464 * Creates some virtual cpus. Good luck creating more than one.
2466 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
2469 struct kvm_vcpu *vcpu;
2471 if (id >= KVM_MAX_VCPU_ID)
2474 mutex_lock(&kvm->lock);
2475 if (kvm->created_vcpus == KVM_MAX_VCPUS) {
2476 mutex_unlock(&kvm->lock);
2480 kvm->created_vcpus++;
2481 mutex_unlock(&kvm->lock);
2483 vcpu = kvm_arch_vcpu_create(kvm, id);
2486 goto vcpu_decrement;
2489 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2491 r = kvm_arch_vcpu_setup(vcpu);
2495 r = kvm_create_vcpu_debugfs(vcpu);
2499 mutex_lock(&kvm->lock);
2500 if (kvm_get_vcpu_by_id(kvm, id)) {
2502 goto unlock_vcpu_destroy;
2505 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
2507 /* Now it's all set up, let userspace reach it */
2509 r = create_vcpu_fd(vcpu);
2512 goto unlock_vcpu_destroy;
2515 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
2518 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
2519 * before kvm->online_vcpu's incremented value.
2522 atomic_inc(&kvm->online_vcpus);
2524 mutex_unlock(&kvm->lock);
2525 kvm_arch_vcpu_postcreate(vcpu);
2528 unlock_vcpu_destroy:
2529 mutex_unlock(&kvm->lock);
2530 debugfs_remove_recursive(vcpu->debugfs_dentry);
2532 kvm_arch_vcpu_destroy(vcpu);
2534 mutex_lock(&kvm->lock);
2535 kvm->created_vcpus--;
2536 mutex_unlock(&kvm->lock);
2540 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2543 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2544 vcpu->sigset_active = 1;
2545 vcpu->sigset = *sigset;
2547 vcpu->sigset_active = 0;
2551 static long kvm_vcpu_ioctl(struct file *filp,
2552 unsigned int ioctl, unsigned long arg)
2554 struct kvm_vcpu *vcpu = filp->private_data;
2555 void __user *argp = (void __user *)arg;
2557 struct kvm_fpu *fpu = NULL;
2558 struct kvm_sregs *kvm_sregs = NULL;
2560 if (vcpu->kvm->mm != current->mm)
2563 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2567 * Some architectures have vcpu ioctls that are asynchronous to vcpu
2568 * execution; mutex_lock() would break them.
2570 r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg);
2571 if (r != -ENOIOCTLCMD)
2574 if (mutex_lock_killable(&vcpu->mutex))
2582 oldpid = rcu_access_pointer(vcpu->pid);
2583 if (unlikely(oldpid != task_pid(current))) {
2584 /* The thread running this VCPU changed. */
2587 r = kvm_arch_vcpu_run_pid_change(vcpu);
2591 newpid = get_task_pid(current, PIDTYPE_PID);
2592 rcu_assign_pointer(vcpu->pid, newpid);
2597 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2598 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2601 case KVM_GET_REGS: {
2602 struct kvm_regs *kvm_regs;
2605 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2608 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2612 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2619 case KVM_SET_REGS: {
2620 struct kvm_regs *kvm_regs;
2623 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2624 if (IS_ERR(kvm_regs)) {
2625 r = PTR_ERR(kvm_regs);
2628 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2632 case KVM_GET_SREGS: {
2633 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2637 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2641 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2646 case KVM_SET_SREGS: {
2647 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2648 if (IS_ERR(kvm_sregs)) {
2649 r = PTR_ERR(kvm_sregs);
2653 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2656 case KVM_GET_MP_STATE: {
2657 struct kvm_mp_state mp_state;
2659 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2663 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
2668 case KVM_SET_MP_STATE: {
2669 struct kvm_mp_state mp_state;
2672 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
2674 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2677 case KVM_TRANSLATE: {
2678 struct kvm_translation tr;
2681 if (copy_from_user(&tr, argp, sizeof(tr)))
2683 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2687 if (copy_to_user(argp, &tr, sizeof(tr)))
2692 case KVM_SET_GUEST_DEBUG: {
2693 struct kvm_guest_debug dbg;
2696 if (copy_from_user(&dbg, argp, sizeof(dbg)))
2698 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2701 case KVM_SET_SIGNAL_MASK: {
2702 struct kvm_signal_mask __user *sigmask_arg = argp;
2703 struct kvm_signal_mask kvm_sigmask;
2704 sigset_t sigset, *p;
2709 if (copy_from_user(&kvm_sigmask, argp,
2710 sizeof(kvm_sigmask)))
2713 if (kvm_sigmask.len != sizeof(sigset))
2716 if (copy_from_user(&sigset, sigmask_arg->sigset,
2721 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2725 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2729 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2733 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2739 fpu = memdup_user(argp, sizeof(*fpu));
2745 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2749 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2752 mutex_unlock(&vcpu->mutex);
2758 #ifdef CONFIG_KVM_COMPAT
2759 static long kvm_vcpu_compat_ioctl(struct file *filp,
2760 unsigned int ioctl, unsigned long arg)
2762 struct kvm_vcpu *vcpu = filp->private_data;
2763 void __user *argp = compat_ptr(arg);
2766 if (vcpu->kvm->mm != current->mm)
2770 case KVM_SET_SIGNAL_MASK: {
2771 struct kvm_signal_mask __user *sigmask_arg = argp;
2772 struct kvm_signal_mask kvm_sigmask;
2777 if (copy_from_user(&kvm_sigmask, argp,
2778 sizeof(kvm_sigmask)))
2781 if (kvm_sigmask.len != sizeof(compat_sigset_t))
2784 if (get_compat_sigset(&sigset, (void *)sigmask_arg->sigset))
2786 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2788 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2792 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2800 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2801 int (*accessor)(struct kvm_device *dev,
2802 struct kvm_device_attr *attr),
2805 struct kvm_device_attr attr;
2810 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2813 return accessor(dev, &attr);
2816 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2819 struct kvm_device *dev = filp->private_data;
2822 case KVM_SET_DEVICE_ATTR:
2823 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2824 case KVM_GET_DEVICE_ATTR:
2825 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2826 case KVM_HAS_DEVICE_ATTR:
2827 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2829 if (dev->ops->ioctl)
2830 return dev->ops->ioctl(dev, ioctl, arg);
2836 static int kvm_device_release(struct inode *inode, struct file *filp)
2838 struct kvm_device *dev = filp->private_data;
2839 struct kvm *kvm = dev->kvm;
2845 static const struct file_operations kvm_device_fops = {
2846 .unlocked_ioctl = kvm_device_ioctl,
2847 .release = kvm_device_release,
2848 KVM_COMPAT(kvm_device_ioctl),
2851 struct kvm_device *kvm_device_from_filp(struct file *filp)
2853 if (filp->f_op != &kvm_device_fops)
2856 return filp->private_data;
2859 static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
2860 #ifdef CONFIG_KVM_MPIC
2861 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
2862 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
2866 int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
2868 if (type >= ARRAY_SIZE(kvm_device_ops_table))
2871 if (kvm_device_ops_table[type] != NULL)
2874 kvm_device_ops_table[type] = ops;
2878 void kvm_unregister_device_ops(u32 type)
2880 if (kvm_device_ops_table[type] != NULL)
2881 kvm_device_ops_table[type] = NULL;
2884 static int kvm_ioctl_create_device(struct kvm *kvm,
2885 struct kvm_create_device *cd)
2887 struct kvm_device_ops *ops = NULL;
2888 struct kvm_device *dev;
2889 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2892 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
2895 ops = kvm_device_ops_table[cd->type];
2902 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2909 mutex_lock(&kvm->lock);
2910 ret = ops->create(dev, cd->type);
2912 mutex_unlock(&kvm->lock);
2916 list_add(&dev->vm_node, &kvm->devices);
2917 mutex_unlock(&kvm->lock);
2922 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2924 mutex_lock(&kvm->lock);
2925 list_del(&dev->vm_node);
2926 mutex_unlock(&kvm->lock);
2936 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2939 case KVM_CAP_USER_MEMORY:
2940 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2941 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2942 case KVM_CAP_INTERNAL_ERROR_DATA:
2943 #ifdef CONFIG_HAVE_KVM_MSI
2944 case KVM_CAP_SIGNAL_MSI:
2946 #ifdef CONFIG_HAVE_KVM_IRQFD
2948 case KVM_CAP_IRQFD_RESAMPLE:
2950 case KVM_CAP_IOEVENTFD_ANY_LENGTH:
2951 case KVM_CAP_CHECK_EXTENSION_VM:
2953 #ifdef CONFIG_KVM_MMIO
2954 case KVM_CAP_COALESCED_MMIO:
2955 return KVM_COALESCED_MMIO_PAGE_OFFSET;
2956 case KVM_CAP_COALESCED_PIO:
2959 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2960 case KVM_CAP_IRQ_ROUTING:
2961 return KVM_MAX_IRQ_ROUTES;
2963 #if KVM_ADDRESS_SPACE_NUM > 1
2964 case KVM_CAP_MULTI_ADDRESS_SPACE:
2965 return KVM_ADDRESS_SPACE_NUM;
2967 case KVM_CAP_MAX_VCPU_ID:
2968 return KVM_MAX_VCPU_ID;
2972 return kvm_vm_ioctl_check_extension(kvm, arg);
2975 static long kvm_vm_ioctl(struct file *filp,
2976 unsigned int ioctl, unsigned long arg)
2978 struct kvm *kvm = filp->private_data;
2979 void __user *argp = (void __user *)arg;
2982 if (kvm->mm != current->mm)
2985 case KVM_CREATE_VCPU:
2986 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2988 case KVM_SET_USER_MEMORY_REGION: {
2989 struct kvm_userspace_memory_region kvm_userspace_mem;
2992 if (copy_from_user(&kvm_userspace_mem, argp,
2993 sizeof(kvm_userspace_mem)))
2996 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2999 case KVM_GET_DIRTY_LOG: {
3000 struct kvm_dirty_log log;
3003 if (copy_from_user(&log, argp, sizeof(log)))
3005 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3008 #ifdef CONFIG_KVM_MMIO
3009 case KVM_REGISTER_COALESCED_MMIO: {
3010 struct kvm_coalesced_mmio_zone zone;
3013 if (copy_from_user(&zone, argp, sizeof(zone)))
3015 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
3018 case KVM_UNREGISTER_COALESCED_MMIO: {
3019 struct kvm_coalesced_mmio_zone zone;
3022 if (copy_from_user(&zone, argp, sizeof(zone)))
3024 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
3029 struct kvm_irqfd data;
3032 if (copy_from_user(&data, argp, sizeof(data)))
3034 r = kvm_irqfd(kvm, &data);
3037 case KVM_IOEVENTFD: {
3038 struct kvm_ioeventfd data;
3041 if (copy_from_user(&data, argp, sizeof(data)))
3043 r = kvm_ioeventfd(kvm, &data);
3046 #ifdef CONFIG_HAVE_KVM_MSI
3047 case KVM_SIGNAL_MSI: {
3051 if (copy_from_user(&msi, argp, sizeof(msi)))
3053 r = kvm_send_userspace_msi(kvm, &msi);
3057 #ifdef __KVM_HAVE_IRQ_LINE
3058 case KVM_IRQ_LINE_STATUS:
3059 case KVM_IRQ_LINE: {
3060 struct kvm_irq_level irq_event;
3063 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
3066 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
3067 ioctl == KVM_IRQ_LINE_STATUS);
3072 if (ioctl == KVM_IRQ_LINE_STATUS) {
3073 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
3081 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3082 case KVM_SET_GSI_ROUTING: {
3083 struct kvm_irq_routing routing;
3084 struct kvm_irq_routing __user *urouting;
3085 struct kvm_irq_routing_entry *entries = NULL;
3088 if (copy_from_user(&routing, argp, sizeof(routing)))
3091 if (!kvm_arch_can_set_irq_routing(kvm))
3093 if (routing.nr > KVM_MAX_IRQ_ROUTES)
3099 entries = vmalloc(array_size(sizeof(*entries),
3105 if (copy_from_user(entries, urouting->entries,
3106 routing.nr * sizeof(*entries)))
3107 goto out_free_irq_routing;
3109 r = kvm_set_irq_routing(kvm, entries, routing.nr,
3111 out_free_irq_routing:
3115 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
3116 case KVM_CREATE_DEVICE: {
3117 struct kvm_create_device cd;
3120 if (copy_from_user(&cd, argp, sizeof(cd)))
3123 r = kvm_ioctl_create_device(kvm, &cd);
3128 if (copy_to_user(argp, &cd, sizeof(cd)))
3134 case KVM_CHECK_EXTENSION:
3135 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
3138 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
3144 #ifdef CONFIG_KVM_COMPAT
3145 struct compat_kvm_dirty_log {
3149 compat_uptr_t dirty_bitmap; /* one bit per page */
3154 static long kvm_vm_compat_ioctl(struct file *filp,
3155 unsigned int ioctl, unsigned long arg)
3157 struct kvm *kvm = filp->private_data;
3160 if (kvm->mm != current->mm)
3163 case KVM_GET_DIRTY_LOG: {
3164 struct compat_kvm_dirty_log compat_log;
3165 struct kvm_dirty_log log;
3167 if (copy_from_user(&compat_log, (void __user *)arg,
3168 sizeof(compat_log)))
3170 log.slot = compat_log.slot;
3171 log.padding1 = compat_log.padding1;
3172 log.padding2 = compat_log.padding2;
3173 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
3175 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3179 r = kvm_vm_ioctl(filp, ioctl, arg);
3185 static struct file_operations kvm_vm_fops = {
3186 .release = kvm_vm_release,
3187 .unlocked_ioctl = kvm_vm_ioctl,
3188 .llseek = noop_llseek,
3189 KVM_COMPAT(kvm_vm_compat_ioctl),
3192 static int kvm_dev_ioctl_create_vm(unsigned long type)
3198 kvm = kvm_create_vm(type);
3200 return PTR_ERR(kvm);
3201 #ifdef CONFIG_KVM_MMIO
3202 r = kvm_coalesced_mmio_init(kvm);
3206 r = get_unused_fd_flags(O_CLOEXEC);
3210 file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
3218 * Don't call kvm_put_kvm anymore at this point; file->f_op is
3219 * already set, with ->release() being kvm_vm_release(). In error
3220 * cases it will be called by the final fput(file) and will take
3221 * care of doing kvm_put_kvm(kvm).
3223 if (kvm_create_vm_debugfs(kvm, r) < 0) {
3228 kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm);
3230 fd_install(r, file);
3238 static long kvm_dev_ioctl(struct file *filp,
3239 unsigned int ioctl, unsigned long arg)
3244 case KVM_GET_API_VERSION:
3247 r = KVM_API_VERSION;
3250 r = kvm_dev_ioctl_create_vm(arg);
3252 case KVM_CHECK_EXTENSION:
3253 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
3255 case KVM_GET_VCPU_MMAP_SIZE:
3258 r = PAGE_SIZE; /* struct kvm_run */
3260 r += PAGE_SIZE; /* pio data page */
3262 #ifdef CONFIG_KVM_MMIO
3263 r += PAGE_SIZE; /* coalesced mmio ring page */
3266 case KVM_TRACE_ENABLE:
3267 case KVM_TRACE_PAUSE:
3268 case KVM_TRACE_DISABLE:
3272 return kvm_arch_dev_ioctl(filp, ioctl, arg);
3278 static struct file_operations kvm_chardev_ops = {
3279 .unlocked_ioctl = kvm_dev_ioctl,
3280 .llseek = noop_llseek,
3281 KVM_COMPAT(kvm_dev_ioctl),
3284 static struct miscdevice kvm_dev = {
3290 static void hardware_enable_nolock(void *junk)
3292 int cpu = raw_smp_processor_id();
3295 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
3298 cpumask_set_cpu(cpu, cpus_hardware_enabled);
3300 r = kvm_arch_hardware_enable();
3303 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3304 atomic_inc(&hardware_enable_failed);
3305 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
3309 static int kvm_starting_cpu(unsigned int cpu)
3311 raw_spin_lock(&kvm_count_lock);
3312 if (kvm_usage_count)
3313 hardware_enable_nolock(NULL);
3314 raw_spin_unlock(&kvm_count_lock);
3318 static void hardware_disable_nolock(void *junk)
3320 int cpu = raw_smp_processor_id();
3322 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
3324 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3325 kvm_arch_hardware_disable();
3328 static int kvm_dying_cpu(unsigned int cpu)
3330 raw_spin_lock(&kvm_count_lock);
3331 if (kvm_usage_count)
3332 hardware_disable_nolock(NULL);
3333 raw_spin_unlock(&kvm_count_lock);
3337 static void hardware_disable_all_nolock(void)
3339 BUG_ON(!kvm_usage_count);
3342 if (!kvm_usage_count)
3343 on_each_cpu(hardware_disable_nolock, NULL, 1);
3346 static void hardware_disable_all(void)
3348 raw_spin_lock(&kvm_count_lock);
3349 hardware_disable_all_nolock();
3350 raw_spin_unlock(&kvm_count_lock);
3353 static int hardware_enable_all(void)
3357 raw_spin_lock(&kvm_count_lock);
3360 if (kvm_usage_count == 1) {
3361 atomic_set(&hardware_enable_failed, 0);
3362 on_each_cpu(hardware_enable_nolock, NULL, 1);
3364 if (atomic_read(&hardware_enable_failed)) {
3365 hardware_disable_all_nolock();
3370 raw_spin_unlock(&kvm_count_lock);
3375 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3379 * Some (well, at least mine) BIOSes hang on reboot if
3382 * And Intel TXT required VMX off for all cpu when system shutdown.
3384 pr_info("kvm: exiting hardware virtualization\n");
3385 kvm_rebooting = true;
3386 on_each_cpu(hardware_disable_nolock, NULL, 1);
3390 static struct notifier_block kvm_reboot_notifier = {
3391 .notifier_call = kvm_reboot,
3395 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3399 for (i = 0; i < bus->dev_count; i++) {
3400 struct kvm_io_device *pos = bus->range[i].dev;
3402 kvm_iodevice_destructor(pos);
3407 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
3408 const struct kvm_io_range *r2)
3410 gpa_t addr1 = r1->addr;
3411 gpa_t addr2 = r2->addr;
3416 /* If r2->len == 0, match the exact address. If r2->len != 0,
3417 * accept any overlapping write. Any order is acceptable for
3418 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
3419 * we process all of them.
3432 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
3434 return kvm_io_bus_cmp(p1, p2);
3437 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
3438 gpa_t addr, int len)
3440 struct kvm_io_range *range, key;
3443 key = (struct kvm_io_range) {
3448 range = bsearch(&key, bus->range, bus->dev_count,
3449 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
3453 off = range - bus->range;
3455 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
3461 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3462 struct kvm_io_range *range, const void *val)
3466 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3470 while (idx < bus->dev_count &&
3471 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3472 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
3481 /* kvm_io_bus_write - called under kvm->slots_lock */
3482 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3483 int len, const void *val)
3485 struct kvm_io_bus *bus;
3486 struct kvm_io_range range;
3489 range = (struct kvm_io_range) {
3494 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3497 r = __kvm_io_bus_write(vcpu, bus, &range, val);
3498 return r < 0 ? r : 0;
3501 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3502 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
3503 gpa_t addr, int len, const void *val, long cookie)
3505 struct kvm_io_bus *bus;
3506 struct kvm_io_range range;
3508 range = (struct kvm_io_range) {
3513 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3517 /* First try the device referenced by cookie. */
3518 if ((cookie >= 0) && (cookie < bus->dev_count) &&
3519 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
3520 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
3525 * cookie contained garbage; fall back to search and return the
3526 * correct cookie value.
3528 return __kvm_io_bus_write(vcpu, bus, &range, val);
3531 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3532 struct kvm_io_range *range, void *val)
3536 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3540 while (idx < bus->dev_count &&
3541 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3542 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
3550 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
3552 /* kvm_io_bus_read - called under kvm->slots_lock */
3553 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3556 struct kvm_io_bus *bus;
3557 struct kvm_io_range range;
3560 range = (struct kvm_io_range) {
3565 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3568 r = __kvm_io_bus_read(vcpu, bus, &range, val);
3569 return r < 0 ? r : 0;
3573 /* Caller must hold slots_lock. */
3574 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3575 int len, struct kvm_io_device *dev)
3578 struct kvm_io_bus *new_bus, *bus;
3579 struct kvm_io_range range;
3581 bus = kvm_get_bus(kvm, bus_idx);
3585 /* exclude ioeventfd which is limited by maximum fd */
3586 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3589 new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count + 1) *
3590 sizeof(struct kvm_io_range)), GFP_KERNEL);
3594 range = (struct kvm_io_range) {
3600 for (i = 0; i < bus->dev_count; i++)
3601 if (kvm_io_bus_cmp(&bus->range[i], &range) > 0)
3604 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3605 new_bus->dev_count++;
3606 new_bus->range[i] = range;
3607 memcpy(new_bus->range + i + 1, bus->range + i,
3608 (bus->dev_count - i) * sizeof(struct kvm_io_range));
3609 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3610 synchronize_srcu_expedited(&kvm->srcu);
3616 /* Caller must hold slots_lock. */
3617 void kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3618 struct kvm_io_device *dev)
3621 struct kvm_io_bus *new_bus, *bus;
3623 bus = kvm_get_bus(kvm, bus_idx);
3627 for (i = 0; i < bus->dev_count; i++)
3628 if (bus->range[i].dev == dev) {
3632 if (i == bus->dev_count)
3635 new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3636 sizeof(struct kvm_io_range)), GFP_KERNEL);
3638 pr_err("kvm: failed to shrink bus, removing it completely\n");
3642 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3643 new_bus->dev_count--;
3644 memcpy(new_bus->range + i, bus->range + i + 1,
3645 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3648 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3649 synchronize_srcu_expedited(&kvm->srcu);
3654 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3657 struct kvm_io_bus *bus;
3658 int dev_idx, srcu_idx;
3659 struct kvm_io_device *iodev = NULL;
3661 srcu_idx = srcu_read_lock(&kvm->srcu);
3663 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
3667 dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
3671 iodev = bus->range[dev_idx].dev;
3674 srcu_read_unlock(&kvm->srcu, srcu_idx);
3678 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
3680 static int kvm_debugfs_open(struct inode *inode, struct file *file,
3681 int (*get)(void *, u64 *), int (*set)(void *, u64),
3684 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
3687 /* The debugfs files are a reference to the kvm struct which
3688 * is still valid when kvm_destroy_vm is called.
3689 * To avoid the race between open and the removal of the debugfs
3690 * directory we test against the users count.
3692 if (!refcount_inc_not_zero(&stat_data->kvm->users_count))
3695 if (simple_attr_open(inode, file, get, set, fmt)) {
3696 kvm_put_kvm(stat_data->kvm);
3703 static int kvm_debugfs_release(struct inode *inode, struct file *file)
3705 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
3708 simple_attr_release(inode, file);
3709 kvm_put_kvm(stat_data->kvm);
3714 static int vm_stat_get_per_vm(void *data, u64 *val)
3716 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3718 *val = *(ulong *)((void *)stat_data->kvm + stat_data->offset);
3723 static int vm_stat_clear_per_vm(void *data, u64 val)
3725 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3730 *(ulong *)((void *)stat_data->kvm + stat_data->offset) = 0;
3735 static int vm_stat_get_per_vm_open(struct inode *inode, struct file *file)
3737 __simple_attr_check_format("%llu\n", 0ull);
3738 return kvm_debugfs_open(inode, file, vm_stat_get_per_vm,
3739 vm_stat_clear_per_vm, "%llu\n");
3742 static const struct file_operations vm_stat_get_per_vm_fops = {
3743 .owner = THIS_MODULE,
3744 .open = vm_stat_get_per_vm_open,
3745 .release = kvm_debugfs_release,
3746 .read = simple_attr_read,
3747 .write = simple_attr_write,
3748 .llseek = no_llseek,
3751 static int vcpu_stat_get_per_vm(void *data, u64 *val)
3754 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3755 struct kvm_vcpu *vcpu;
3759 kvm_for_each_vcpu(i, vcpu, stat_data->kvm)
3760 *val += *(u64 *)((void *)vcpu + stat_data->offset);
3765 static int vcpu_stat_clear_per_vm(void *data, u64 val)
3768 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3769 struct kvm_vcpu *vcpu;
3774 kvm_for_each_vcpu(i, vcpu, stat_data->kvm)
3775 *(u64 *)((void *)vcpu + stat_data->offset) = 0;
3780 static int vcpu_stat_get_per_vm_open(struct inode *inode, struct file *file)
3782 __simple_attr_check_format("%llu\n", 0ull);
3783 return kvm_debugfs_open(inode, file, vcpu_stat_get_per_vm,
3784 vcpu_stat_clear_per_vm, "%llu\n");
3787 static const struct file_operations vcpu_stat_get_per_vm_fops = {
3788 .owner = THIS_MODULE,
3789 .open = vcpu_stat_get_per_vm_open,
3790 .release = kvm_debugfs_release,
3791 .read = simple_attr_read,
3792 .write = simple_attr_write,
3793 .llseek = no_llseek,
3796 static const struct file_operations *stat_fops_per_vm[] = {
3797 [KVM_STAT_VCPU] = &vcpu_stat_get_per_vm_fops,
3798 [KVM_STAT_VM] = &vm_stat_get_per_vm_fops,
3801 static int vm_stat_get(void *_offset, u64 *val)
3803 unsigned offset = (long)_offset;
3805 struct kvm_stat_data stat_tmp = {.offset = offset};
3809 spin_lock(&kvm_lock);
3810 list_for_each_entry(kvm, &vm_list, vm_list) {
3812 vm_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
3815 spin_unlock(&kvm_lock);
3819 static int vm_stat_clear(void *_offset, u64 val)
3821 unsigned offset = (long)_offset;
3823 struct kvm_stat_data stat_tmp = {.offset = offset};
3828 spin_lock(&kvm_lock);
3829 list_for_each_entry(kvm, &vm_list, vm_list) {
3831 vm_stat_clear_per_vm((void *)&stat_tmp, 0);
3833 spin_unlock(&kvm_lock);
3838 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n");
3840 static int vcpu_stat_get(void *_offset, u64 *val)
3842 unsigned offset = (long)_offset;
3844 struct kvm_stat_data stat_tmp = {.offset = offset};
3848 spin_lock(&kvm_lock);
3849 list_for_each_entry(kvm, &vm_list, vm_list) {
3851 vcpu_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
3854 spin_unlock(&kvm_lock);
3858 static int vcpu_stat_clear(void *_offset, u64 val)
3860 unsigned offset = (long)_offset;
3862 struct kvm_stat_data stat_tmp = {.offset = offset};
3867 spin_lock(&kvm_lock);
3868 list_for_each_entry(kvm, &vm_list, vm_list) {
3870 vcpu_stat_clear_per_vm((void *)&stat_tmp, 0);
3872 spin_unlock(&kvm_lock);
3877 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear,
3880 static const struct file_operations *stat_fops[] = {
3881 [KVM_STAT_VCPU] = &vcpu_stat_fops,
3882 [KVM_STAT_VM] = &vm_stat_fops,
3885 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm)
3887 struct kobj_uevent_env *env;
3888 unsigned long long created, active;
3890 if (!kvm_dev.this_device || !kvm)
3893 spin_lock(&kvm_lock);
3894 if (type == KVM_EVENT_CREATE_VM) {
3895 kvm_createvm_count++;
3897 } else if (type == KVM_EVENT_DESTROY_VM) {
3900 created = kvm_createvm_count;
3901 active = kvm_active_vms;
3902 spin_unlock(&kvm_lock);
3904 env = kzalloc(sizeof(*env), GFP_KERNEL);
3908 add_uevent_var(env, "CREATED=%llu", created);
3909 add_uevent_var(env, "COUNT=%llu", active);
3911 if (type == KVM_EVENT_CREATE_VM) {
3912 add_uevent_var(env, "EVENT=create");
3913 kvm->userspace_pid = task_pid_nr(current);
3914 } else if (type == KVM_EVENT_DESTROY_VM) {
3915 add_uevent_var(env, "EVENT=destroy");
3917 add_uevent_var(env, "PID=%d", kvm->userspace_pid);
3919 if (kvm->debugfs_dentry) {
3920 char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL);
3923 tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX);
3925 add_uevent_var(env, "STATS_PATH=%s", tmp);
3929 /* no need for checks, since we are adding at most only 5 keys */
3930 env->envp[env->envp_idx++] = NULL;
3931 kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp);
3935 static void kvm_init_debug(void)
3937 struct kvm_stats_debugfs_item *p;
3939 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3941 kvm_debugfs_num_entries = 0;
3942 for (p = debugfs_entries; p->name; ++p, kvm_debugfs_num_entries++) {
3943 debugfs_create_file(p->name, 0644, kvm_debugfs_dir,
3944 (void *)(long)p->offset,
3945 stat_fops[p->kind]);
3949 static int kvm_suspend(void)
3951 if (kvm_usage_count)
3952 hardware_disable_nolock(NULL);
3956 static void kvm_resume(void)
3958 if (kvm_usage_count) {
3959 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3960 hardware_enable_nolock(NULL);
3964 static struct syscore_ops kvm_syscore_ops = {
3965 .suspend = kvm_suspend,
3966 .resume = kvm_resume,
3970 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3972 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3975 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3977 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3979 if (vcpu->preempted)
3980 vcpu->preempted = false;
3982 kvm_arch_sched_in(vcpu, cpu);
3984 kvm_arch_vcpu_load(vcpu, cpu);
3987 static void kvm_sched_out(struct preempt_notifier *pn,
3988 struct task_struct *next)
3990 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3992 if (current->state == TASK_RUNNING)
3993 vcpu->preempted = true;
3994 kvm_arch_vcpu_put(vcpu);
3997 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3998 struct module *module)
4003 r = kvm_arch_init(opaque);
4008 * kvm_arch_init makes sure there's at most one caller
4009 * for architectures that support multiple implementations,
4010 * like intel and amd on x86.
4011 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
4012 * conflicts in case kvm is already setup for another implementation.
4014 r = kvm_irqfd_init();
4018 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
4023 r = kvm_arch_hardware_setup();
4027 for_each_online_cpu(cpu) {
4028 smp_call_function_single(cpu,
4029 kvm_arch_check_processor_compat,
4035 r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting",
4036 kvm_starting_cpu, kvm_dying_cpu);
4039 register_reboot_notifier(&kvm_reboot_notifier);
4041 /* A kmem cache lets us meet the alignment requirements of fx_save. */
4043 vcpu_align = __alignof__(struct kvm_vcpu);
4045 kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align,
4047 offsetof(struct kvm_vcpu, arch),
4048 sizeof_field(struct kvm_vcpu, arch),
4050 if (!kvm_vcpu_cache) {
4055 r = kvm_async_pf_init();
4059 kvm_chardev_ops.owner = module;
4060 kvm_vm_fops.owner = module;
4061 kvm_vcpu_fops.owner = module;
4063 r = misc_register(&kvm_dev);
4065 pr_err("kvm: misc device register failed\n");
4069 register_syscore_ops(&kvm_syscore_ops);
4071 kvm_preempt_ops.sched_in = kvm_sched_in;
4072 kvm_preempt_ops.sched_out = kvm_sched_out;
4076 r = kvm_vfio_ops_init();
4082 kvm_async_pf_deinit();
4084 kmem_cache_destroy(kvm_vcpu_cache);
4086 unregister_reboot_notifier(&kvm_reboot_notifier);
4087 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4090 kvm_arch_hardware_unsetup();
4092 free_cpumask_var(cpus_hardware_enabled);
4100 EXPORT_SYMBOL_GPL(kvm_init);
4104 debugfs_remove_recursive(kvm_debugfs_dir);
4105 misc_deregister(&kvm_dev);
4106 kmem_cache_destroy(kvm_vcpu_cache);
4107 kvm_async_pf_deinit();
4108 unregister_syscore_ops(&kvm_syscore_ops);
4109 unregister_reboot_notifier(&kvm_reboot_notifier);
4110 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4111 on_each_cpu(hardware_disable_nolock, NULL, 1);
4112 kvm_arch_hardware_unsetup();
4115 free_cpumask_var(cpus_hardware_enabled);
4116 kvm_vfio_ops_exit();
4118 EXPORT_SYMBOL_GPL(kvm_exit);