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)
811 int i = slots->id_to_index[id];
812 struct kvm_memory_slot *mslots = slots->memslots;
814 WARN_ON(mslots[i].id != id);
816 WARN_ON(!mslots[i].npages);
817 if (mslots[i].npages)
820 if (!mslots[i].npages)
824 while (i < KVM_MEM_SLOTS_NUM - 1 &&
825 new->base_gfn <= mslots[i + 1].base_gfn) {
826 if (!mslots[i + 1].npages)
828 mslots[i] = mslots[i + 1];
829 slots->id_to_index[mslots[i].id] = i;
834 * The ">=" is needed when creating a slot with base_gfn == 0,
835 * so that it moves before all those with base_gfn == npages == 0.
837 * On the other hand, if new->npages is zero, the above loop has
838 * already left i pointing to the beginning of the empty part of
839 * mslots, and the ">=" would move the hole backwards in this
840 * case---which is wrong. So skip the loop when deleting a slot.
844 new->base_gfn >= mslots[i - 1].base_gfn) {
845 mslots[i] = mslots[i - 1];
846 slots->id_to_index[mslots[i].id] = i;
850 WARN_ON_ONCE(i != slots->used_slots);
853 slots->id_to_index[mslots[i].id] = i;
856 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
858 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
860 #ifdef __KVM_HAVE_READONLY_MEM
861 valid_flags |= KVM_MEM_READONLY;
864 if (mem->flags & ~valid_flags)
870 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
871 int as_id, struct kvm_memslots *slots)
873 struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id);
876 * Set the low bit in the generation, which disables SPTE caching
877 * until the end of synchronize_srcu_expedited.
879 WARN_ON(old_memslots->generation & 1);
880 slots->generation = old_memslots->generation + 1;
882 rcu_assign_pointer(kvm->memslots[as_id], slots);
883 synchronize_srcu_expedited(&kvm->srcu);
886 * Increment the new memslot generation a second time. This prevents
887 * vm exits that race with memslot updates from caching a memslot
888 * generation that will (potentially) be valid forever.
890 * Generations must be unique even across address spaces. We do not need
891 * a global counter for that, instead the generation space is evenly split
892 * across address spaces. For example, with two address spaces, address
893 * space 0 will use generations 0, 4, 8, ... while * address space 1 will
894 * use generations 2, 6, 10, 14, ...
896 slots->generation += KVM_ADDRESS_SPACE_NUM * 2 - 1;
898 kvm_arch_memslots_updated(kvm, slots);
904 * Allocate some memory and give it an address in the guest physical address
907 * Discontiguous memory is allowed, mostly for framebuffers.
909 * Must be called holding kvm->slots_lock for write.
911 int __kvm_set_memory_region(struct kvm *kvm,
912 const struct kvm_userspace_memory_region *mem)
916 unsigned long npages;
917 struct kvm_memory_slot *slot;
918 struct kvm_memory_slot old, new;
919 struct kvm_memslots *slots = NULL, *old_memslots;
921 enum kvm_mr_change change;
923 r = check_memory_region_flags(mem);
928 as_id = mem->slot >> 16;
931 /* General sanity checks */
932 if (mem->memory_size & (PAGE_SIZE - 1))
934 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
936 /* We can read the guest memory with __xxx_user() later on. */
937 if ((id < KVM_USER_MEM_SLOTS) &&
938 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
939 !access_ok(VERIFY_WRITE,
940 (void __user *)(unsigned long)mem->userspace_addr,
943 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
945 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
948 slot = id_to_memslot(__kvm_memslots(kvm, as_id), id);
949 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
950 npages = mem->memory_size >> PAGE_SHIFT;
952 if (npages > KVM_MEM_MAX_NR_PAGES)
958 new.base_gfn = base_gfn;
960 new.flags = mem->flags;
964 change = KVM_MR_CREATE;
965 else { /* Modify an existing slot. */
966 if ((mem->userspace_addr != old.userspace_addr) ||
967 (npages != old.npages) ||
968 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
971 if (base_gfn != old.base_gfn)
972 change = KVM_MR_MOVE;
973 else if (new.flags != old.flags)
974 change = KVM_MR_FLAGS_ONLY;
975 else { /* Nothing to change. */
984 change = KVM_MR_DELETE;
989 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
990 /* Check for overlaps */
992 kvm_for_each_memslot(slot, __kvm_memslots(kvm, as_id)) {
995 if (!((base_gfn + npages <= slot->base_gfn) ||
996 (base_gfn >= slot->base_gfn + slot->npages)))
1001 /* Free page dirty bitmap if unneeded */
1002 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
1003 new.dirty_bitmap = NULL;
1006 if (change == KVM_MR_CREATE) {
1007 new.userspace_addr = mem->userspace_addr;
1009 if (kvm_arch_create_memslot(kvm, &new, npages))
1013 /* Allocate page dirty bitmap if needed */
1014 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
1015 if (kvm_create_dirty_bitmap(&new) < 0)
1019 slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
1022 memcpy(slots, __kvm_memslots(kvm, as_id), sizeof(struct kvm_memslots));
1024 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
1025 slot = id_to_memslot(slots, id);
1026 slot->flags |= KVM_MEMSLOT_INVALID;
1028 old_memslots = install_new_memslots(kvm, as_id, slots);
1030 /* From this point no new shadow pages pointing to a deleted,
1031 * or moved, memslot will be created.
1033 * validation of sp->gfn happens in:
1034 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1035 * - kvm_is_visible_gfn (mmu_check_roots)
1037 kvm_arch_flush_shadow_memslot(kvm, slot);
1040 * We can re-use the old_memslots from above, the only difference
1041 * from the currently installed memslots is the invalid flag. This
1042 * will get overwritten by update_memslots anyway.
1044 slots = old_memslots;
1047 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
1051 /* actual memory is freed via old in kvm_free_memslot below */
1052 if (change == KVM_MR_DELETE) {
1053 new.dirty_bitmap = NULL;
1054 memset(&new.arch, 0, sizeof(new.arch));
1057 update_memslots(slots, &new);
1058 old_memslots = install_new_memslots(kvm, as_id, slots);
1060 kvm_arch_commit_memory_region(kvm, mem, &old, &new, change);
1062 kvm_free_memslot(kvm, &old, &new);
1063 kvfree(old_memslots);
1069 kvm_free_memslot(kvm, &new, &old);
1073 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1075 int kvm_set_memory_region(struct kvm *kvm,
1076 const struct kvm_userspace_memory_region *mem)
1080 mutex_lock(&kvm->slots_lock);
1081 r = __kvm_set_memory_region(kvm, mem);
1082 mutex_unlock(&kvm->slots_lock);
1085 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1087 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1088 struct kvm_userspace_memory_region *mem)
1090 if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1093 return kvm_set_memory_region(kvm, mem);
1096 int kvm_get_dirty_log(struct kvm *kvm,
1097 struct kvm_dirty_log *log, int *is_dirty)
1099 struct kvm_memslots *slots;
1100 struct kvm_memory_slot *memslot;
1103 unsigned long any = 0;
1105 as_id = log->slot >> 16;
1106 id = (u16)log->slot;
1107 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1110 slots = __kvm_memslots(kvm, as_id);
1111 memslot = id_to_memslot(slots, id);
1112 if (!memslot->dirty_bitmap)
1115 n = kvm_dirty_bitmap_bytes(memslot);
1117 for (i = 0; !any && i < n/sizeof(long); ++i)
1118 any = memslot->dirty_bitmap[i];
1120 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1127 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1129 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1131 * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages
1132 * are dirty write protect them for next write.
1133 * @kvm: pointer to kvm instance
1134 * @log: slot id and address to which we copy the log
1135 * @is_dirty: flag set if any page is dirty
1137 * We need to keep it in mind that VCPU threads can write to the bitmap
1138 * concurrently. So, to avoid losing track of dirty pages we keep the
1141 * 1. Take a snapshot of the bit and clear it if needed.
1142 * 2. Write protect the corresponding page.
1143 * 3. Copy the snapshot to the userspace.
1144 * 4. Upon return caller flushes TLB's if needed.
1146 * Between 2 and 4, the guest may write to the page using the remaining TLB
1147 * entry. This is not a problem because the page is reported dirty using
1148 * the snapshot taken before and step 4 ensures that writes done after
1149 * exiting to userspace will be logged for the next call.
1152 int kvm_get_dirty_log_protect(struct kvm *kvm,
1153 struct kvm_dirty_log *log, bool *is_dirty)
1155 struct kvm_memslots *slots;
1156 struct kvm_memory_slot *memslot;
1159 unsigned long *dirty_bitmap;
1160 unsigned long *dirty_bitmap_buffer;
1162 as_id = log->slot >> 16;
1163 id = (u16)log->slot;
1164 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1167 slots = __kvm_memslots(kvm, as_id);
1168 memslot = id_to_memslot(slots, id);
1170 dirty_bitmap = memslot->dirty_bitmap;
1174 n = kvm_dirty_bitmap_bytes(memslot);
1176 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
1177 memset(dirty_bitmap_buffer, 0, n);
1179 spin_lock(&kvm->mmu_lock);
1181 for (i = 0; i < n / sizeof(long); i++) {
1185 if (!dirty_bitmap[i])
1190 mask = xchg(&dirty_bitmap[i], 0);
1191 dirty_bitmap_buffer[i] = mask;
1194 offset = i * BITS_PER_LONG;
1195 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1200 spin_unlock(&kvm->mmu_lock);
1201 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1205 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect);
1208 bool kvm_largepages_enabled(void)
1210 return largepages_enabled;
1213 void kvm_disable_largepages(void)
1215 largepages_enabled = false;
1217 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1219 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1221 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1223 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1225 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
1227 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn);
1230 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1232 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1234 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1235 memslot->flags & KVM_MEMSLOT_INVALID)
1240 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1242 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1244 struct vm_area_struct *vma;
1245 unsigned long addr, size;
1249 addr = gfn_to_hva(kvm, gfn);
1250 if (kvm_is_error_hva(addr))
1253 down_read(¤t->mm->mmap_sem);
1254 vma = find_vma(current->mm, addr);
1258 size = vma_kernel_pagesize(vma);
1261 up_read(¤t->mm->mmap_sem);
1266 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1268 return slot->flags & KVM_MEM_READONLY;
1271 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1272 gfn_t *nr_pages, bool write)
1274 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1275 return KVM_HVA_ERR_BAD;
1277 if (memslot_is_readonly(slot) && write)
1278 return KVM_HVA_ERR_RO_BAD;
1281 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1283 return __gfn_to_hva_memslot(slot, gfn);
1286 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1289 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1292 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1295 return gfn_to_hva_many(slot, gfn, NULL);
1297 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1299 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1301 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1303 EXPORT_SYMBOL_GPL(gfn_to_hva);
1305 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
1307 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
1309 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
1312 * If writable is set to false, the hva returned by this function is only
1313 * allowed to be read.
1315 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1316 gfn_t gfn, bool *writable)
1318 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1320 if (!kvm_is_error_hva(hva) && writable)
1321 *writable = !memslot_is_readonly(slot);
1326 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1328 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1330 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1333 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
1335 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1337 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1340 static inline int check_user_page_hwpoison(unsigned long addr)
1342 int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
1344 rc = get_user_pages(addr, 1, flags, NULL, NULL);
1345 return rc == -EHWPOISON;
1349 * The fast path to get the writable pfn which will be stored in @pfn,
1350 * true indicates success, otherwise false is returned. It's also the
1351 * only part that runs if we can are in atomic context.
1353 static bool hva_to_pfn_fast(unsigned long addr, bool write_fault,
1354 bool *writable, kvm_pfn_t *pfn)
1356 struct page *page[1];
1360 * Fast pin a writable pfn only if it is a write fault request
1361 * or the caller allows to map a writable pfn for a read fault
1364 if (!(write_fault || writable))
1367 npages = __get_user_pages_fast(addr, 1, 1, page);
1369 *pfn = page_to_pfn(page[0]);
1380 * The slow path to get the pfn of the specified host virtual address,
1381 * 1 indicates success, -errno is returned if error is detected.
1383 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1384 bool *writable, kvm_pfn_t *pfn)
1386 unsigned int flags = FOLL_HWPOISON;
1393 *writable = write_fault;
1396 flags |= FOLL_WRITE;
1398 flags |= FOLL_NOWAIT;
1400 npages = get_user_pages_unlocked(addr, 1, &page, flags);
1404 /* map read fault as writable if possible */
1405 if (unlikely(!write_fault) && writable) {
1408 if (__get_user_pages_fast(addr, 1, 1, &wpage) == 1) {
1414 *pfn = page_to_pfn(page);
1418 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1420 if (unlikely(!(vma->vm_flags & VM_READ)))
1423 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1429 static int hva_to_pfn_remapped(struct vm_area_struct *vma,
1430 unsigned long addr, bool *async,
1431 bool write_fault, bool *writable,
1437 r = follow_pfn(vma, addr, &pfn);
1440 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
1441 * not call the fault handler, so do it here.
1443 bool unlocked = false;
1444 r = fixup_user_fault(current, current->mm, addr,
1445 (write_fault ? FAULT_FLAG_WRITE : 0),
1452 r = follow_pfn(vma, addr, &pfn);
1462 * Get a reference here because callers of *hva_to_pfn* and
1463 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
1464 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
1465 * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will
1466 * simply do nothing for reserved pfns.
1468 * Whoever called remap_pfn_range is also going to call e.g.
1469 * unmap_mapping_range before the underlying pages are freed,
1470 * causing a call to our MMU notifier.
1479 * Pin guest page in memory and return its pfn.
1480 * @addr: host virtual address which maps memory to the guest
1481 * @atomic: whether this function can sleep
1482 * @async: whether this function need to wait IO complete if the
1483 * host page is not in the memory
1484 * @write_fault: whether we should get a writable host page
1485 * @writable: whether it allows to map a writable host page for !@write_fault
1487 * The function will map a writable host page for these two cases:
1488 * 1): @write_fault = true
1489 * 2): @write_fault = false && @writable, @writable will tell the caller
1490 * whether the mapping is writable.
1492 static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1493 bool write_fault, bool *writable)
1495 struct vm_area_struct *vma;
1499 /* we can do it either atomically or asynchronously, not both */
1500 BUG_ON(atomic && async);
1502 if (hva_to_pfn_fast(addr, write_fault, writable, &pfn))
1506 return KVM_PFN_ERR_FAULT;
1508 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1512 down_read(¤t->mm->mmap_sem);
1513 if (npages == -EHWPOISON ||
1514 (!async && check_user_page_hwpoison(addr))) {
1515 pfn = KVM_PFN_ERR_HWPOISON;
1520 vma = find_vma_intersection(current->mm, addr, addr + 1);
1523 pfn = KVM_PFN_ERR_FAULT;
1524 else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
1525 r = hva_to_pfn_remapped(vma, addr, async, write_fault, writable, &pfn);
1529 pfn = KVM_PFN_ERR_FAULT;
1531 if (async && vma_is_valid(vma, write_fault))
1533 pfn = KVM_PFN_ERR_FAULT;
1536 up_read(¤t->mm->mmap_sem);
1540 kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn,
1541 bool atomic, bool *async, bool write_fault,
1544 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1546 if (addr == KVM_HVA_ERR_RO_BAD) {
1549 return KVM_PFN_ERR_RO_FAULT;
1552 if (kvm_is_error_hva(addr)) {
1555 return KVM_PFN_NOSLOT;
1558 /* Do not map writable pfn in the readonly memslot. */
1559 if (writable && memslot_is_readonly(slot)) {
1564 return hva_to_pfn(addr, atomic, async, write_fault,
1567 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
1569 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1572 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
1573 write_fault, writable);
1575 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1577 kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1579 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1581 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
1583 kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1585 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1587 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1589 kvm_pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1591 return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm, gfn), gfn);
1593 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1595 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
1597 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1599 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
1601 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1603 return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
1605 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1607 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
1609 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1611 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
1613 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1614 struct page **pages, int nr_pages)
1619 addr = gfn_to_hva_many(slot, gfn, &entry);
1620 if (kvm_is_error_hva(addr))
1623 if (entry < nr_pages)
1626 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1628 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1630 static struct page *kvm_pfn_to_page(kvm_pfn_t pfn)
1632 if (is_error_noslot_pfn(pfn))
1633 return KVM_ERR_PTR_BAD_PAGE;
1635 if (kvm_is_reserved_pfn(pfn)) {
1637 return KVM_ERR_PTR_BAD_PAGE;
1640 return pfn_to_page(pfn);
1643 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1647 pfn = gfn_to_pfn(kvm, gfn);
1649 return kvm_pfn_to_page(pfn);
1651 EXPORT_SYMBOL_GPL(gfn_to_page);
1653 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
1657 pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
1659 return kvm_pfn_to_page(pfn);
1661 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
1663 void kvm_release_page_clean(struct page *page)
1665 WARN_ON(is_error_page(page));
1667 kvm_release_pfn_clean(page_to_pfn(page));
1669 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1671 void kvm_release_pfn_clean(kvm_pfn_t pfn)
1673 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
1674 put_page(pfn_to_page(pfn));
1676 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1678 void kvm_release_page_dirty(struct page *page)
1680 WARN_ON(is_error_page(page));
1682 kvm_release_pfn_dirty(page_to_pfn(page));
1684 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1686 void kvm_release_pfn_dirty(kvm_pfn_t pfn)
1688 kvm_set_pfn_dirty(pfn);
1689 kvm_release_pfn_clean(pfn);
1691 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1693 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
1695 if (!kvm_is_reserved_pfn(pfn)) {
1696 struct page *page = pfn_to_page(pfn);
1698 if (!PageReserved(page))
1702 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1704 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
1706 if (!kvm_is_reserved_pfn(pfn))
1707 mark_page_accessed(pfn_to_page(pfn));
1709 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1711 void kvm_get_pfn(kvm_pfn_t pfn)
1713 if (!kvm_is_reserved_pfn(pfn))
1714 get_page(pfn_to_page(pfn));
1716 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1718 static int next_segment(unsigned long len, int offset)
1720 if (len > PAGE_SIZE - offset)
1721 return PAGE_SIZE - offset;
1726 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
1727 void *data, int offset, int len)
1732 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1733 if (kvm_is_error_hva(addr))
1735 r = __copy_from_user(data, (void __user *)addr + offset, len);
1741 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1744 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1746 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1748 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1750 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
1751 int offset, int len)
1753 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1755 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1757 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
1759 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1761 gfn_t gfn = gpa >> PAGE_SHIFT;
1763 int offset = offset_in_page(gpa);
1766 while ((seg = next_segment(len, offset)) != 0) {
1767 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1777 EXPORT_SYMBOL_GPL(kvm_read_guest);
1779 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
1781 gfn_t gfn = gpa >> PAGE_SHIFT;
1783 int offset = offset_in_page(gpa);
1786 while ((seg = next_segment(len, offset)) != 0) {
1787 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
1797 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
1799 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1800 void *data, int offset, unsigned long len)
1805 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1806 if (kvm_is_error_hva(addr))
1808 pagefault_disable();
1809 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1816 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1819 gfn_t gfn = gpa >> PAGE_SHIFT;
1820 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1821 int offset = offset_in_page(gpa);
1823 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1825 EXPORT_SYMBOL_GPL(kvm_read_guest_atomic);
1827 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
1828 void *data, unsigned long len)
1830 gfn_t gfn = gpa >> PAGE_SHIFT;
1831 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1832 int offset = offset_in_page(gpa);
1834 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1836 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
1838 static int __kvm_write_guest_page(struct kvm_memory_slot *memslot, gfn_t gfn,
1839 const void *data, int offset, int len)
1844 addr = gfn_to_hva_memslot(memslot, gfn);
1845 if (kvm_is_error_hva(addr))
1847 r = __copy_to_user((void __user *)addr + offset, data, len);
1850 mark_page_dirty_in_slot(memslot, gfn);
1854 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
1855 const void *data, int offset, int len)
1857 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1859 return __kvm_write_guest_page(slot, gfn, data, offset, len);
1861 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1863 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
1864 const void *data, int offset, int len)
1866 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1868 return __kvm_write_guest_page(slot, gfn, data, offset, len);
1870 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
1872 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1875 gfn_t gfn = gpa >> PAGE_SHIFT;
1877 int offset = offset_in_page(gpa);
1880 while ((seg = next_segment(len, offset)) != 0) {
1881 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1891 EXPORT_SYMBOL_GPL(kvm_write_guest);
1893 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
1896 gfn_t gfn = gpa >> PAGE_SHIFT;
1898 int offset = offset_in_page(gpa);
1901 while ((seg = next_segment(len, offset)) != 0) {
1902 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
1912 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
1914 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
1915 struct gfn_to_hva_cache *ghc,
1916 gpa_t gpa, unsigned long len)
1918 int offset = offset_in_page(gpa);
1919 gfn_t start_gfn = gpa >> PAGE_SHIFT;
1920 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1921 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1922 gfn_t nr_pages_avail;
1925 ghc->generation = slots->generation;
1927 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
1928 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, NULL);
1929 if (!kvm_is_error_hva(ghc->hva) && nr_pages_needed <= 1) {
1933 * If the requested region crosses two memslots, we still
1934 * verify that the entire region is valid here.
1936 while (start_gfn <= end_gfn) {
1938 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
1939 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1941 if (kvm_is_error_hva(ghc->hva))
1943 start_gfn += nr_pages_avail;
1945 /* Use the slow path for cross page reads and writes. */
1946 ghc->memslot = NULL;
1951 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1952 gpa_t gpa, unsigned long len)
1954 struct kvm_memslots *slots = kvm_memslots(kvm);
1955 return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len);
1957 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1959 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1960 void *data, int offset, unsigned long len)
1962 struct kvm_memslots *slots = kvm_memslots(kvm);
1964 gpa_t gpa = ghc->gpa + offset;
1966 BUG_ON(len + offset > ghc->len);
1968 if (slots->generation != ghc->generation)
1969 __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len);
1971 if (unlikely(!ghc->memslot))
1972 return kvm_write_guest(kvm, gpa, data, len);
1974 if (kvm_is_error_hva(ghc->hva))
1977 r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
1980 mark_page_dirty_in_slot(ghc->memslot, gpa >> PAGE_SHIFT);
1984 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached);
1986 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1987 void *data, unsigned long len)
1989 return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len);
1991 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1993 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1994 void *data, unsigned long len)
1996 struct kvm_memslots *slots = kvm_memslots(kvm);
1999 BUG_ON(len > ghc->len);
2001 if (slots->generation != ghc->generation)
2002 __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len);
2004 if (unlikely(!ghc->memslot))
2005 return kvm_read_guest(kvm, ghc->gpa, data, len);
2007 if (kvm_is_error_hva(ghc->hva))
2010 r = __copy_from_user(data, (void __user *)ghc->hva, len);
2016 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
2018 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
2020 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
2022 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
2024 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
2026 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
2028 gfn_t gfn = gpa >> PAGE_SHIFT;
2030 int offset = offset_in_page(gpa);
2033 while ((seg = next_segment(len, offset)) != 0) {
2034 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
2043 EXPORT_SYMBOL_GPL(kvm_clear_guest);
2045 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot,
2048 if (memslot && memslot->dirty_bitmap) {
2049 unsigned long rel_gfn = gfn - memslot->base_gfn;
2051 set_bit_le(rel_gfn, memslot->dirty_bitmap);
2055 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
2057 struct kvm_memory_slot *memslot;
2059 memslot = gfn_to_memslot(kvm, gfn);
2060 mark_page_dirty_in_slot(memslot, gfn);
2062 EXPORT_SYMBOL_GPL(mark_page_dirty);
2064 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
2066 struct kvm_memory_slot *memslot;
2068 memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2069 mark_page_dirty_in_slot(memslot, gfn);
2071 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
2073 void kvm_sigset_activate(struct kvm_vcpu *vcpu)
2075 if (!vcpu->sigset_active)
2079 * This does a lockless modification of ->real_blocked, which is fine
2080 * because, only current can change ->real_blocked and all readers of
2081 * ->real_blocked don't care as long ->real_blocked is always a subset
2084 sigprocmask(SIG_SETMASK, &vcpu->sigset, ¤t->real_blocked);
2087 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu)
2089 if (!vcpu->sigset_active)
2092 sigprocmask(SIG_SETMASK, ¤t->real_blocked, NULL);
2093 sigemptyset(¤t->real_blocked);
2096 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
2098 unsigned int old, val, grow;
2100 old = val = vcpu->halt_poll_ns;
2101 grow = READ_ONCE(halt_poll_ns_grow);
2103 if (val == 0 && grow)
2108 if (val > halt_poll_ns)
2111 vcpu->halt_poll_ns = val;
2112 trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
2115 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
2117 unsigned int old, val, shrink;
2119 old = val = vcpu->halt_poll_ns;
2120 shrink = READ_ONCE(halt_poll_ns_shrink);
2126 vcpu->halt_poll_ns = val;
2127 trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
2130 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
2133 int idx = srcu_read_lock(&vcpu->kvm->srcu);
2135 if (kvm_arch_vcpu_runnable(vcpu)) {
2136 kvm_make_request(KVM_REQ_UNHALT, vcpu);
2139 if (kvm_cpu_has_pending_timer(vcpu))
2141 if (signal_pending(current))
2146 srcu_read_unlock(&vcpu->kvm->srcu, idx);
2151 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2153 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
2156 DECLARE_SWAITQUEUE(wait);
2157 bool waited = false;
2160 start = cur = ktime_get();
2161 if (vcpu->halt_poll_ns) {
2162 ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns);
2164 ++vcpu->stat.halt_attempted_poll;
2167 * This sets KVM_REQ_UNHALT if an interrupt
2170 if (kvm_vcpu_check_block(vcpu) < 0) {
2171 ++vcpu->stat.halt_successful_poll;
2172 if (!vcpu_valid_wakeup(vcpu))
2173 ++vcpu->stat.halt_poll_invalid;
2177 } while (single_task_running() && ktime_before(cur, stop));
2180 kvm_arch_vcpu_blocking(vcpu);
2183 prepare_to_swait_exclusive(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
2185 if (kvm_vcpu_check_block(vcpu) < 0)
2192 finish_swait(&vcpu->wq, &wait);
2195 kvm_arch_vcpu_unblocking(vcpu);
2197 block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
2199 if (!vcpu_valid_wakeup(vcpu))
2200 shrink_halt_poll_ns(vcpu);
2201 else if (halt_poll_ns) {
2202 if (block_ns <= vcpu->halt_poll_ns)
2204 /* we had a long block, shrink polling */
2205 else if (vcpu->halt_poll_ns && block_ns > halt_poll_ns)
2206 shrink_halt_poll_ns(vcpu);
2207 /* we had a short halt and our poll time is too small */
2208 else if (vcpu->halt_poll_ns < halt_poll_ns &&
2209 block_ns < halt_poll_ns)
2210 grow_halt_poll_ns(vcpu);
2212 vcpu->halt_poll_ns = 0;
2214 trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu));
2215 kvm_arch_vcpu_block_finish(vcpu);
2217 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
2219 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
2221 struct swait_queue_head *wqp;
2223 wqp = kvm_arch_vcpu_wq(vcpu);
2224 if (swq_has_sleeper(wqp)) {
2226 ++vcpu->stat.halt_wakeup;
2232 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
2236 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2238 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
2241 int cpu = vcpu->cpu;
2243 if (kvm_vcpu_wake_up(vcpu))
2247 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
2248 if (kvm_arch_vcpu_should_kick(vcpu))
2249 smp_send_reschedule(cpu);
2252 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
2253 #endif /* !CONFIG_S390 */
2255 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
2258 struct task_struct *task = NULL;
2262 pid = rcu_dereference(target->pid);
2264 task = get_pid_task(pid, PIDTYPE_PID);
2268 ret = yield_to(task, 1);
2269 put_task_struct(task);
2273 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
2276 * Helper that checks whether a VCPU is eligible for directed yield.
2277 * Most eligible candidate to yield is decided by following heuristics:
2279 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2280 * (preempted lock holder), indicated by @in_spin_loop.
2281 * Set at the beiginning and cleared at the end of interception/PLE handler.
2283 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2284 * chance last time (mostly it has become eligible now since we have probably
2285 * yielded to lockholder in last iteration. This is done by toggling
2286 * @dy_eligible each time a VCPU checked for eligibility.)
2288 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2289 * to preempted lock-holder could result in wrong VCPU selection and CPU
2290 * burning. Giving priority for a potential lock-holder increases lock
2293 * Since algorithm is based on heuristics, accessing another VCPU data without
2294 * locking does not harm. It may result in trying to yield to same VCPU, fail
2295 * and continue with next VCPU and so on.
2297 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
2299 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2302 eligible = !vcpu->spin_loop.in_spin_loop ||
2303 vcpu->spin_loop.dy_eligible;
2305 if (vcpu->spin_loop.in_spin_loop)
2306 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
2314 void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode)
2316 struct kvm *kvm = me->kvm;
2317 struct kvm_vcpu *vcpu;
2318 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
2324 kvm_vcpu_set_in_spin_loop(me, true);
2326 * We boost the priority of a VCPU that is runnable but not
2327 * currently running, because it got preempted by something
2328 * else and called schedule in __vcpu_run. Hopefully that
2329 * VCPU is holding the lock that we need and will release it.
2330 * We approximate round-robin by starting at the last boosted VCPU.
2332 for (pass = 0; pass < 2 && !yielded && try; pass++) {
2333 kvm_for_each_vcpu(i, vcpu, kvm) {
2334 if (!pass && i <= last_boosted_vcpu) {
2335 i = last_boosted_vcpu;
2337 } else if (pass && i > last_boosted_vcpu)
2339 if (!READ_ONCE(vcpu->preempted))
2343 if (swait_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
2345 if (yield_to_kernel_mode && !kvm_arch_vcpu_in_kernel(vcpu))
2347 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
2350 yielded = kvm_vcpu_yield_to(vcpu);
2352 kvm->last_boosted_vcpu = i;
2354 } else if (yielded < 0) {
2361 kvm_vcpu_set_in_spin_loop(me, false);
2363 /* Ensure vcpu is not eligible during next spinloop */
2364 kvm_vcpu_set_dy_eligible(me, false);
2366 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
2368 static vm_fault_t kvm_vcpu_fault(struct vm_fault *vmf)
2370 struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data;
2373 if (vmf->pgoff == 0)
2374 page = virt_to_page(vcpu->run);
2376 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
2377 page = virt_to_page(vcpu->arch.pio_data);
2379 #ifdef CONFIG_KVM_MMIO
2380 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
2381 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
2384 return kvm_arch_vcpu_fault(vcpu, vmf);
2390 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
2391 .fault = kvm_vcpu_fault,
2394 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2396 vma->vm_ops = &kvm_vcpu_vm_ops;
2400 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2402 struct kvm_vcpu *vcpu = filp->private_data;
2404 debugfs_remove_recursive(vcpu->debugfs_dentry);
2405 kvm_put_kvm(vcpu->kvm);
2409 static struct file_operations kvm_vcpu_fops = {
2410 .release = kvm_vcpu_release,
2411 .unlocked_ioctl = kvm_vcpu_ioctl,
2412 .mmap = kvm_vcpu_mmap,
2413 .llseek = noop_llseek,
2414 KVM_COMPAT(kvm_vcpu_compat_ioctl),
2418 * Allocates an inode for the vcpu.
2420 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2422 char name[8 + 1 + ITOA_MAX_LEN + 1];
2424 snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id);
2425 return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
2428 static int kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
2430 char dir_name[ITOA_MAX_LEN * 2];
2433 if (!kvm_arch_has_vcpu_debugfs())
2436 if (!debugfs_initialized())
2439 snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
2440 vcpu->debugfs_dentry = debugfs_create_dir(dir_name,
2441 vcpu->kvm->debugfs_dentry);
2442 if (!vcpu->debugfs_dentry)
2445 ret = kvm_arch_create_vcpu_debugfs(vcpu);
2447 debugfs_remove_recursive(vcpu->debugfs_dentry);
2455 * Creates some virtual cpus. Good luck creating more than one.
2457 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
2460 struct kvm_vcpu *vcpu;
2462 if (id >= KVM_MAX_VCPU_ID)
2465 mutex_lock(&kvm->lock);
2466 if (kvm->created_vcpus == KVM_MAX_VCPUS) {
2467 mutex_unlock(&kvm->lock);
2471 kvm->created_vcpus++;
2472 mutex_unlock(&kvm->lock);
2474 vcpu = kvm_arch_vcpu_create(kvm, id);
2477 goto vcpu_decrement;
2480 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2482 r = kvm_arch_vcpu_setup(vcpu);
2486 r = kvm_create_vcpu_debugfs(vcpu);
2490 mutex_lock(&kvm->lock);
2491 if (kvm_get_vcpu_by_id(kvm, id)) {
2493 goto unlock_vcpu_destroy;
2496 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
2498 /* Now it's all set up, let userspace reach it */
2500 r = create_vcpu_fd(vcpu);
2503 goto unlock_vcpu_destroy;
2506 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
2509 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
2510 * before kvm->online_vcpu's incremented value.
2513 atomic_inc(&kvm->online_vcpus);
2515 mutex_unlock(&kvm->lock);
2516 kvm_arch_vcpu_postcreate(vcpu);
2519 unlock_vcpu_destroy:
2520 mutex_unlock(&kvm->lock);
2521 debugfs_remove_recursive(vcpu->debugfs_dentry);
2523 kvm_arch_vcpu_destroy(vcpu);
2525 mutex_lock(&kvm->lock);
2526 kvm->created_vcpus--;
2527 mutex_unlock(&kvm->lock);
2531 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2534 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2535 vcpu->sigset_active = 1;
2536 vcpu->sigset = *sigset;
2538 vcpu->sigset_active = 0;
2542 static long kvm_vcpu_ioctl(struct file *filp,
2543 unsigned int ioctl, unsigned long arg)
2545 struct kvm_vcpu *vcpu = filp->private_data;
2546 void __user *argp = (void __user *)arg;
2548 struct kvm_fpu *fpu = NULL;
2549 struct kvm_sregs *kvm_sregs = NULL;
2551 if (vcpu->kvm->mm != current->mm)
2554 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2558 * Some architectures have vcpu ioctls that are asynchronous to vcpu
2559 * execution; mutex_lock() would break them.
2561 r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg);
2562 if (r != -ENOIOCTLCMD)
2565 if (mutex_lock_killable(&vcpu->mutex))
2573 oldpid = rcu_access_pointer(vcpu->pid);
2574 if (unlikely(oldpid != task_pid(current))) {
2575 /* The thread running this VCPU changed. */
2578 r = kvm_arch_vcpu_run_pid_change(vcpu);
2582 newpid = get_task_pid(current, PIDTYPE_PID);
2583 rcu_assign_pointer(vcpu->pid, newpid);
2588 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2589 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2592 case KVM_GET_REGS: {
2593 struct kvm_regs *kvm_regs;
2596 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2599 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2603 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2610 case KVM_SET_REGS: {
2611 struct kvm_regs *kvm_regs;
2614 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2615 if (IS_ERR(kvm_regs)) {
2616 r = PTR_ERR(kvm_regs);
2619 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2623 case KVM_GET_SREGS: {
2624 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2628 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2632 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2637 case KVM_SET_SREGS: {
2638 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2639 if (IS_ERR(kvm_sregs)) {
2640 r = PTR_ERR(kvm_sregs);
2644 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2647 case KVM_GET_MP_STATE: {
2648 struct kvm_mp_state mp_state;
2650 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2654 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
2659 case KVM_SET_MP_STATE: {
2660 struct kvm_mp_state mp_state;
2663 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
2665 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2668 case KVM_TRANSLATE: {
2669 struct kvm_translation tr;
2672 if (copy_from_user(&tr, argp, sizeof(tr)))
2674 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2678 if (copy_to_user(argp, &tr, sizeof(tr)))
2683 case KVM_SET_GUEST_DEBUG: {
2684 struct kvm_guest_debug dbg;
2687 if (copy_from_user(&dbg, argp, sizeof(dbg)))
2689 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2692 case KVM_SET_SIGNAL_MASK: {
2693 struct kvm_signal_mask __user *sigmask_arg = argp;
2694 struct kvm_signal_mask kvm_sigmask;
2695 sigset_t sigset, *p;
2700 if (copy_from_user(&kvm_sigmask, argp,
2701 sizeof(kvm_sigmask)))
2704 if (kvm_sigmask.len != sizeof(sigset))
2707 if (copy_from_user(&sigset, sigmask_arg->sigset,
2712 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2716 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2720 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2724 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2730 fpu = memdup_user(argp, sizeof(*fpu));
2736 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2740 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2743 mutex_unlock(&vcpu->mutex);
2749 #ifdef CONFIG_KVM_COMPAT
2750 static long kvm_vcpu_compat_ioctl(struct file *filp,
2751 unsigned int ioctl, unsigned long arg)
2753 struct kvm_vcpu *vcpu = filp->private_data;
2754 void __user *argp = compat_ptr(arg);
2757 if (vcpu->kvm->mm != current->mm)
2761 case KVM_SET_SIGNAL_MASK: {
2762 struct kvm_signal_mask __user *sigmask_arg = argp;
2763 struct kvm_signal_mask kvm_sigmask;
2768 if (copy_from_user(&kvm_sigmask, argp,
2769 sizeof(kvm_sigmask)))
2772 if (kvm_sigmask.len != sizeof(compat_sigset_t))
2775 if (get_compat_sigset(&sigset, (void *)sigmask_arg->sigset))
2777 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2779 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2783 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2791 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2792 int (*accessor)(struct kvm_device *dev,
2793 struct kvm_device_attr *attr),
2796 struct kvm_device_attr attr;
2801 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2804 return accessor(dev, &attr);
2807 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2810 struct kvm_device *dev = filp->private_data;
2813 case KVM_SET_DEVICE_ATTR:
2814 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2815 case KVM_GET_DEVICE_ATTR:
2816 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2817 case KVM_HAS_DEVICE_ATTR:
2818 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2820 if (dev->ops->ioctl)
2821 return dev->ops->ioctl(dev, ioctl, arg);
2827 static int kvm_device_release(struct inode *inode, struct file *filp)
2829 struct kvm_device *dev = filp->private_data;
2830 struct kvm *kvm = dev->kvm;
2836 static const struct file_operations kvm_device_fops = {
2837 .unlocked_ioctl = kvm_device_ioctl,
2838 .release = kvm_device_release,
2839 KVM_COMPAT(kvm_device_ioctl),
2842 struct kvm_device *kvm_device_from_filp(struct file *filp)
2844 if (filp->f_op != &kvm_device_fops)
2847 return filp->private_data;
2850 static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
2851 #ifdef CONFIG_KVM_MPIC
2852 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
2853 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
2857 int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
2859 if (type >= ARRAY_SIZE(kvm_device_ops_table))
2862 if (kvm_device_ops_table[type] != NULL)
2865 kvm_device_ops_table[type] = ops;
2869 void kvm_unregister_device_ops(u32 type)
2871 if (kvm_device_ops_table[type] != NULL)
2872 kvm_device_ops_table[type] = NULL;
2875 static int kvm_ioctl_create_device(struct kvm *kvm,
2876 struct kvm_create_device *cd)
2878 struct kvm_device_ops *ops = NULL;
2879 struct kvm_device *dev;
2880 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2883 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
2886 ops = kvm_device_ops_table[cd->type];
2893 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2900 mutex_lock(&kvm->lock);
2901 ret = ops->create(dev, cd->type);
2903 mutex_unlock(&kvm->lock);
2907 list_add(&dev->vm_node, &kvm->devices);
2908 mutex_unlock(&kvm->lock);
2913 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2915 mutex_lock(&kvm->lock);
2916 list_del(&dev->vm_node);
2917 mutex_unlock(&kvm->lock);
2927 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2930 case KVM_CAP_USER_MEMORY:
2931 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2932 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2933 case KVM_CAP_INTERNAL_ERROR_DATA:
2934 #ifdef CONFIG_HAVE_KVM_MSI
2935 case KVM_CAP_SIGNAL_MSI:
2937 #ifdef CONFIG_HAVE_KVM_IRQFD
2939 case KVM_CAP_IRQFD_RESAMPLE:
2941 case KVM_CAP_IOEVENTFD_ANY_LENGTH:
2942 case KVM_CAP_CHECK_EXTENSION_VM:
2944 #ifdef CONFIG_KVM_MMIO
2945 case KVM_CAP_COALESCED_MMIO:
2946 return KVM_COALESCED_MMIO_PAGE_OFFSET;
2948 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2949 case KVM_CAP_IRQ_ROUTING:
2950 return KVM_MAX_IRQ_ROUTES;
2952 #if KVM_ADDRESS_SPACE_NUM > 1
2953 case KVM_CAP_MULTI_ADDRESS_SPACE:
2954 return KVM_ADDRESS_SPACE_NUM;
2956 case KVM_CAP_MAX_VCPU_ID:
2957 return KVM_MAX_VCPU_ID;
2961 return kvm_vm_ioctl_check_extension(kvm, arg);
2964 static long kvm_vm_ioctl(struct file *filp,
2965 unsigned int ioctl, unsigned long arg)
2967 struct kvm *kvm = filp->private_data;
2968 void __user *argp = (void __user *)arg;
2971 if (kvm->mm != current->mm)
2974 case KVM_CREATE_VCPU:
2975 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2977 case KVM_SET_USER_MEMORY_REGION: {
2978 struct kvm_userspace_memory_region kvm_userspace_mem;
2981 if (copy_from_user(&kvm_userspace_mem, argp,
2982 sizeof(kvm_userspace_mem)))
2985 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2988 case KVM_GET_DIRTY_LOG: {
2989 struct kvm_dirty_log log;
2992 if (copy_from_user(&log, argp, sizeof(log)))
2994 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2997 #ifdef CONFIG_KVM_MMIO
2998 case KVM_REGISTER_COALESCED_MMIO: {
2999 struct kvm_coalesced_mmio_zone zone;
3002 if (copy_from_user(&zone, argp, sizeof(zone)))
3004 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
3007 case KVM_UNREGISTER_COALESCED_MMIO: {
3008 struct kvm_coalesced_mmio_zone zone;
3011 if (copy_from_user(&zone, argp, sizeof(zone)))
3013 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
3018 struct kvm_irqfd data;
3021 if (copy_from_user(&data, argp, sizeof(data)))
3023 r = kvm_irqfd(kvm, &data);
3026 case KVM_IOEVENTFD: {
3027 struct kvm_ioeventfd data;
3030 if (copy_from_user(&data, argp, sizeof(data)))
3032 r = kvm_ioeventfd(kvm, &data);
3035 #ifdef CONFIG_HAVE_KVM_MSI
3036 case KVM_SIGNAL_MSI: {
3040 if (copy_from_user(&msi, argp, sizeof(msi)))
3042 r = kvm_send_userspace_msi(kvm, &msi);
3046 #ifdef __KVM_HAVE_IRQ_LINE
3047 case KVM_IRQ_LINE_STATUS:
3048 case KVM_IRQ_LINE: {
3049 struct kvm_irq_level irq_event;
3052 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
3055 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
3056 ioctl == KVM_IRQ_LINE_STATUS);
3061 if (ioctl == KVM_IRQ_LINE_STATUS) {
3062 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
3070 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3071 case KVM_SET_GSI_ROUTING: {
3072 struct kvm_irq_routing routing;
3073 struct kvm_irq_routing __user *urouting;
3074 struct kvm_irq_routing_entry *entries = NULL;
3077 if (copy_from_user(&routing, argp, sizeof(routing)))
3080 if (!kvm_arch_can_set_irq_routing(kvm))
3082 if (routing.nr > KVM_MAX_IRQ_ROUTES)
3088 entries = vmalloc(array_size(sizeof(*entries),
3094 if (copy_from_user(entries, urouting->entries,
3095 routing.nr * sizeof(*entries)))
3096 goto out_free_irq_routing;
3098 r = kvm_set_irq_routing(kvm, entries, routing.nr,
3100 out_free_irq_routing:
3104 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
3105 case KVM_CREATE_DEVICE: {
3106 struct kvm_create_device cd;
3109 if (copy_from_user(&cd, argp, sizeof(cd)))
3112 r = kvm_ioctl_create_device(kvm, &cd);
3117 if (copy_to_user(argp, &cd, sizeof(cd)))
3123 case KVM_CHECK_EXTENSION:
3124 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
3127 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
3133 #ifdef CONFIG_KVM_COMPAT
3134 struct compat_kvm_dirty_log {
3138 compat_uptr_t dirty_bitmap; /* one bit per page */
3143 static long kvm_vm_compat_ioctl(struct file *filp,
3144 unsigned int ioctl, unsigned long arg)
3146 struct kvm *kvm = filp->private_data;
3149 if (kvm->mm != current->mm)
3152 case KVM_GET_DIRTY_LOG: {
3153 struct compat_kvm_dirty_log compat_log;
3154 struct kvm_dirty_log log;
3156 if (copy_from_user(&compat_log, (void __user *)arg,
3157 sizeof(compat_log)))
3159 log.slot = compat_log.slot;
3160 log.padding1 = compat_log.padding1;
3161 log.padding2 = compat_log.padding2;
3162 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
3164 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3168 r = kvm_vm_ioctl(filp, ioctl, arg);
3174 static struct file_operations kvm_vm_fops = {
3175 .release = kvm_vm_release,
3176 .unlocked_ioctl = kvm_vm_ioctl,
3177 .llseek = noop_llseek,
3178 KVM_COMPAT(kvm_vm_compat_ioctl),
3181 static int kvm_dev_ioctl_create_vm(unsigned long type)
3187 kvm = kvm_create_vm(type);
3189 return PTR_ERR(kvm);
3190 #ifdef CONFIG_KVM_MMIO
3191 r = kvm_coalesced_mmio_init(kvm);
3195 r = get_unused_fd_flags(O_CLOEXEC);
3199 file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
3207 * Don't call kvm_put_kvm anymore at this point; file->f_op is
3208 * already set, with ->release() being kvm_vm_release(). In error
3209 * cases it will be called by the final fput(file) and will take
3210 * care of doing kvm_put_kvm(kvm).
3212 if (kvm_create_vm_debugfs(kvm, r) < 0) {
3217 kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm);
3219 fd_install(r, file);
3227 static long kvm_dev_ioctl(struct file *filp,
3228 unsigned int ioctl, unsigned long arg)
3233 case KVM_GET_API_VERSION:
3236 r = KVM_API_VERSION;
3239 r = kvm_dev_ioctl_create_vm(arg);
3241 case KVM_CHECK_EXTENSION:
3242 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
3244 case KVM_GET_VCPU_MMAP_SIZE:
3247 r = PAGE_SIZE; /* struct kvm_run */
3249 r += PAGE_SIZE; /* pio data page */
3251 #ifdef CONFIG_KVM_MMIO
3252 r += PAGE_SIZE; /* coalesced mmio ring page */
3255 case KVM_TRACE_ENABLE:
3256 case KVM_TRACE_PAUSE:
3257 case KVM_TRACE_DISABLE:
3261 return kvm_arch_dev_ioctl(filp, ioctl, arg);
3267 static struct file_operations kvm_chardev_ops = {
3268 .unlocked_ioctl = kvm_dev_ioctl,
3269 .llseek = noop_llseek,
3270 KVM_COMPAT(kvm_dev_ioctl),
3273 static struct miscdevice kvm_dev = {
3279 static void hardware_enable_nolock(void *junk)
3281 int cpu = raw_smp_processor_id();
3284 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
3287 cpumask_set_cpu(cpu, cpus_hardware_enabled);
3289 r = kvm_arch_hardware_enable();
3292 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3293 atomic_inc(&hardware_enable_failed);
3294 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
3298 static int kvm_starting_cpu(unsigned int cpu)
3300 raw_spin_lock(&kvm_count_lock);
3301 if (kvm_usage_count)
3302 hardware_enable_nolock(NULL);
3303 raw_spin_unlock(&kvm_count_lock);
3307 static void hardware_disable_nolock(void *junk)
3309 int cpu = raw_smp_processor_id();
3311 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
3313 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3314 kvm_arch_hardware_disable();
3317 static int kvm_dying_cpu(unsigned int cpu)
3319 raw_spin_lock(&kvm_count_lock);
3320 if (kvm_usage_count)
3321 hardware_disable_nolock(NULL);
3322 raw_spin_unlock(&kvm_count_lock);
3326 static void hardware_disable_all_nolock(void)
3328 BUG_ON(!kvm_usage_count);
3331 if (!kvm_usage_count)
3332 on_each_cpu(hardware_disable_nolock, NULL, 1);
3335 static void hardware_disable_all(void)
3337 raw_spin_lock(&kvm_count_lock);
3338 hardware_disable_all_nolock();
3339 raw_spin_unlock(&kvm_count_lock);
3342 static int hardware_enable_all(void)
3346 raw_spin_lock(&kvm_count_lock);
3349 if (kvm_usage_count == 1) {
3350 atomic_set(&hardware_enable_failed, 0);
3351 on_each_cpu(hardware_enable_nolock, NULL, 1);
3353 if (atomic_read(&hardware_enable_failed)) {
3354 hardware_disable_all_nolock();
3359 raw_spin_unlock(&kvm_count_lock);
3364 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3368 * Some (well, at least mine) BIOSes hang on reboot if
3371 * And Intel TXT required VMX off for all cpu when system shutdown.
3373 pr_info("kvm: exiting hardware virtualization\n");
3374 kvm_rebooting = true;
3375 on_each_cpu(hardware_disable_nolock, NULL, 1);
3379 static struct notifier_block kvm_reboot_notifier = {
3380 .notifier_call = kvm_reboot,
3384 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3388 for (i = 0; i < bus->dev_count; i++) {
3389 struct kvm_io_device *pos = bus->range[i].dev;
3391 kvm_iodevice_destructor(pos);
3396 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
3397 const struct kvm_io_range *r2)
3399 gpa_t addr1 = r1->addr;
3400 gpa_t addr2 = r2->addr;
3405 /* If r2->len == 0, match the exact address. If r2->len != 0,
3406 * accept any overlapping write. Any order is acceptable for
3407 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
3408 * we process all of them.
3421 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
3423 return kvm_io_bus_cmp(p1, p2);
3426 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
3427 gpa_t addr, int len)
3429 struct kvm_io_range *range, key;
3432 key = (struct kvm_io_range) {
3437 range = bsearch(&key, bus->range, bus->dev_count,
3438 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
3442 off = range - bus->range;
3444 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
3450 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3451 struct kvm_io_range *range, const void *val)
3455 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3459 while (idx < bus->dev_count &&
3460 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3461 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
3470 /* kvm_io_bus_write - called under kvm->slots_lock */
3471 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3472 int len, const void *val)
3474 struct kvm_io_bus *bus;
3475 struct kvm_io_range range;
3478 range = (struct kvm_io_range) {
3483 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3486 r = __kvm_io_bus_write(vcpu, bus, &range, val);
3487 return r < 0 ? r : 0;
3490 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3491 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
3492 gpa_t addr, int len, const void *val, long cookie)
3494 struct kvm_io_bus *bus;
3495 struct kvm_io_range range;
3497 range = (struct kvm_io_range) {
3502 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3506 /* First try the device referenced by cookie. */
3507 if ((cookie >= 0) && (cookie < bus->dev_count) &&
3508 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
3509 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
3514 * cookie contained garbage; fall back to search and return the
3515 * correct cookie value.
3517 return __kvm_io_bus_write(vcpu, bus, &range, val);
3520 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3521 struct kvm_io_range *range, void *val)
3525 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3529 while (idx < bus->dev_count &&
3530 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3531 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
3539 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
3541 /* kvm_io_bus_read - called under kvm->slots_lock */
3542 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3545 struct kvm_io_bus *bus;
3546 struct kvm_io_range range;
3549 range = (struct kvm_io_range) {
3554 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3557 r = __kvm_io_bus_read(vcpu, bus, &range, val);
3558 return r < 0 ? r : 0;
3562 /* Caller must hold slots_lock. */
3563 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3564 int len, struct kvm_io_device *dev)
3567 struct kvm_io_bus *new_bus, *bus;
3568 struct kvm_io_range range;
3570 bus = kvm_get_bus(kvm, bus_idx);
3574 /* exclude ioeventfd which is limited by maximum fd */
3575 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3578 new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count + 1) *
3579 sizeof(struct kvm_io_range)), GFP_KERNEL);
3583 range = (struct kvm_io_range) {
3589 for (i = 0; i < bus->dev_count; i++)
3590 if (kvm_io_bus_cmp(&bus->range[i], &range) > 0)
3593 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3594 new_bus->dev_count++;
3595 new_bus->range[i] = range;
3596 memcpy(new_bus->range + i + 1, bus->range + i,
3597 (bus->dev_count - i) * sizeof(struct kvm_io_range));
3598 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3599 synchronize_srcu_expedited(&kvm->srcu);
3605 /* Caller must hold slots_lock. */
3606 void kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3607 struct kvm_io_device *dev)
3610 struct kvm_io_bus *new_bus, *bus;
3612 bus = kvm_get_bus(kvm, bus_idx);
3616 for (i = 0; i < bus->dev_count; i++)
3617 if (bus->range[i].dev == dev) {
3621 if (i == bus->dev_count)
3624 new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3625 sizeof(struct kvm_io_range)), GFP_KERNEL);
3627 pr_err("kvm: failed to shrink bus, removing it completely\n");
3631 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3632 new_bus->dev_count--;
3633 memcpy(new_bus->range + i, bus->range + i + 1,
3634 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3637 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3638 synchronize_srcu_expedited(&kvm->srcu);
3643 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3646 struct kvm_io_bus *bus;
3647 int dev_idx, srcu_idx;
3648 struct kvm_io_device *iodev = NULL;
3650 srcu_idx = srcu_read_lock(&kvm->srcu);
3652 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
3656 dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
3660 iodev = bus->range[dev_idx].dev;
3663 srcu_read_unlock(&kvm->srcu, srcu_idx);
3667 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
3669 static int kvm_debugfs_open(struct inode *inode, struct file *file,
3670 int (*get)(void *, u64 *), int (*set)(void *, u64),
3673 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
3676 /* The debugfs files are a reference to the kvm struct which
3677 * is still valid when kvm_destroy_vm is called.
3678 * To avoid the race between open and the removal of the debugfs
3679 * directory we test against the users count.
3681 if (!refcount_inc_not_zero(&stat_data->kvm->users_count))
3684 if (simple_attr_open(inode, file, get, set, fmt)) {
3685 kvm_put_kvm(stat_data->kvm);
3692 static int kvm_debugfs_release(struct inode *inode, struct file *file)
3694 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
3697 simple_attr_release(inode, file);
3698 kvm_put_kvm(stat_data->kvm);
3703 static int vm_stat_get_per_vm(void *data, u64 *val)
3705 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3707 *val = *(ulong *)((void *)stat_data->kvm + stat_data->offset);
3712 static int vm_stat_clear_per_vm(void *data, u64 val)
3714 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3719 *(ulong *)((void *)stat_data->kvm + stat_data->offset) = 0;
3724 static int vm_stat_get_per_vm_open(struct inode *inode, struct file *file)
3726 __simple_attr_check_format("%llu\n", 0ull);
3727 return kvm_debugfs_open(inode, file, vm_stat_get_per_vm,
3728 vm_stat_clear_per_vm, "%llu\n");
3731 static const struct file_operations vm_stat_get_per_vm_fops = {
3732 .owner = THIS_MODULE,
3733 .open = vm_stat_get_per_vm_open,
3734 .release = kvm_debugfs_release,
3735 .read = simple_attr_read,
3736 .write = simple_attr_write,
3737 .llseek = no_llseek,
3740 static int vcpu_stat_get_per_vm(void *data, u64 *val)
3743 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3744 struct kvm_vcpu *vcpu;
3748 kvm_for_each_vcpu(i, vcpu, stat_data->kvm)
3749 *val += *(u64 *)((void *)vcpu + stat_data->offset);
3754 static int vcpu_stat_clear_per_vm(void *data, u64 val)
3757 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3758 struct kvm_vcpu *vcpu;
3763 kvm_for_each_vcpu(i, vcpu, stat_data->kvm)
3764 *(u64 *)((void *)vcpu + stat_data->offset) = 0;
3769 static int vcpu_stat_get_per_vm_open(struct inode *inode, struct file *file)
3771 __simple_attr_check_format("%llu\n", 0ull);
3772 return kvm_debugfs_open(inode, file, vcpu_stat_get_per_vm,
3773 vcpu_stat_clear_per_vm, "%llu\n");
3776 static const struct file_operations vcpu_stat_get_per_vm_fops = {
3777 .owner = THIS_MODULE,
3778 .open = vcpu_stat_get_per_vm_open,
3779 .release = kvm_debugfs_release,
3780 .read = simple_attr_read,
3781 .write = simple_attr_write,
3782 .llseek = no_llseek,
3785 static const struct file_operations *stat_fops_per_vm[] = {
3786 [KVM_STAT_VCPU] = &vcpu_stat_get_per_vm_fops,
3787 [KVM_STAT_VM] = &vm_stat_get_per_vm_fops,
3790 static int vm_stat_get(void *_offset, u64 *val)
3792 unsigned offset = (long)_offset;
3794 struct kvm_stat_data stat_tmp = {.offset = offset};
3798 spin_lock(&kvm_lock);
3799 list_for_each_entry(kvm, &vm_list, vm_list) {
3801 vm_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
3804 spin_unlock(&kvm_lock);
3808 static int vm_stat_clear(void *_offset, u64 val)
3810 unsigned offset = (long)_offset;
3812 struct kvm_stat_data stat_tmp = {.offset = offset};
3817 spin_lock(&kvm_lock);
3818 list_for_each_entry(kvm, &vm_list, vm_list) {
3820 vm_stat_clear_per_vm((void *)&stat_tmp, 0);
3822 spin_unlock(&kvm_lock);
3827 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n");
3829 static int vcpu_stat_get(void *_offset, u64 *val)
3831 unsigned offset = (long)_offset;
3833 struct kvm_stat_data stat_tmp = {.offset = offset};
3837 spin_lock(&kvm_lock);
3838 list_for_each_entry(kvm, &vm_list, vm_list) {
3840 vcpu_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
3843 spin_unlock(&kvm_lock);
3847 static int vcpu_stat_clear(void *_offset, u64 val)
3849 unsigned offset = (long)_offset;
3851 struct kvm_stat_data stat_tmp = {.offset = offset};
3856 spin_lock(&kvm_lock);
3857 list_for_each_entry(kvm, &vm_list, vm_list) {
3859 vcpu_stat_clear_per_vm((void *)&stat_tmp, 0);
3861 spin_unlock(&kvm_lock);
3866 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear,
3869 static const struct file_operations *stat_fops[] = {
3870 [KVM_STAT_VCPU] = &vcpu_stat_fops,
3871 [KVM_STAT_VM] = &vm_stat_fops,
3874 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm)
3876 struct kobj_uevent_env *env;
3877 unsigned long long created, active;
3879 if (!kvm_dev.this_device || !kvm)
3882 spin_lock(&kvm_lock);
3883 if (type == KVM_EVENT_CREATE_VM) {
3884 kvm_createvm_count++;
3886 } else if (type == KVM_EVENT_DESTROY_VM) {
3889 created = kvm_createvm_count;
3890 active = kvm_active_vms;
3891 spin_unlock(&kvm_lock);
3893 env = kzalloc(sizeof(*env), GFP_KERNEL);
3897 add_uevent_var(env, "CREATED=%llu", created);
3898 add_uevent_var(env, "COUNT=%llu", active);
3900 if (type == KVM_EVENT_CREATE_VM) {
3901 add_uevent_var(env, "EVENT=create");
3902 kvm->userspace_pid = task_pid_nr(current);
3903 } else if (type == KVM_EVENT_DESTROY_VM) {
3904 add_uevent_var(env, "EVENT=destroy");
3906 add_uevent_var(env, "PID=%d", kvm->userspace_pid);
3908 if (kvm->debugfs_dentry) {
3909 char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL);
3912 tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX);
3914 add_uevent_var(env, "STATS_PATH=%s", tmp);
3918 /* no need for checks, since we are adding at most only 5 keys */
3919 env->envp[env->envp_idx++] = NULL;
3920 kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp);
3924 static void kvm_init_debug(void)
3926 struct kvm_stats_debugfs_item *p;
3928 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3930 kvm_debugfs_num_entries = 0;
3931 for (p = debugfs_entries; p->name; ++p, kvm_debugfs_num_entries++) {
3932 debugfs_create_file(p->name, 0644, kvm_debugfs_dir,
3933 (void *)(long)p->offset,
3934 stat_fops[p->kind]);
3938 static int kvm_suspend(void)
3940 if (kvm_usage_count)
3941 hardware_disable_nolock(NULL);
3945 static void kvm_resume(void)
3947 if (kvm_usage_count) {
3948 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3949 hardware_enable_nolock(NULL);
3953 static struct syscore_ops kvm_syscore_ops = {
3954 .suspend = kvm_suspend,
3955 .resume = kvm_resume,
3959 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3961 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3964 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3966 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3968 if (vcpu->preempted)
3969 vcpu->preempted = false;
3971 kvm_arch_sched_in(vcpu, cpu);
3973 kvm_arch_vcpu_load(vcpu, cpu);
3976 static void kvm_sched_out(struct preempt_notifier *pn,
3977 struct task_struct *next)
3979 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3981 if (current->state == TASK_RUNNING)
3982 vcpu->preempted = true;
3983 kvm_arch_vcpu_put(vcpu);
3986 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3987 struct module *module)
3992 r = kvm_arch_init(opaque);
3997 * kvm_arch_init makes sure there's at most one caller
3998 * for architectures that support multiple implementations,
3999 * like intel and amd on x86.
4000 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
4001 * conflicts in case kvm is already setup for another implementation.
4003 r = kvm_irqfd_init();
4007 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
4012 r = kvm_arch_hardware_setup();
4016 for_each_online_cpu(cpu) {
4017 smp_call_function_single(cpu,
4018 kvm_arch_check_processor_compat,
4024 r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting",
4025 kvm_starting_cpu, kvm_dying_cpu);
4028 register_reboot_notifier(&kvm_reboot_notifier);
4030 /* A kmem cache lets us meet the alignment requirements of fx_save. */
4032 vcpu_align = __alignof__(struct kvm_vcpu);
4034 kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align,
4036 offsetof(struct kvm_vcpu, arch),
4037 sizeof_field(struct kvm_vcpu, arch),
4039 if (!kvm_vcpu_cache) {
4044 r = kvm_async_pf_init();
4048 kvm_chardev_ops.owner = module;
4049 kvm_vm_fops.owner = module;
4050 kvm_vcpu_fops.owner = module;
4052 r = misc_register(&kvm_dev);
4054 pr_err("kvm: misc device register failed\n");
4058 register_syscore_ops(&kvm_syscore_ops);
4060 kvm_preempt_ops.sched_in = kvm_sched_in;
4061 kvm_preempt_ops.sched_out = kvm_sched_out;
4065 r = kvm_vfio_ops_init();
4071 kvm_async_pf_deinit();
4073 kmem_cache_destroy(kvm_vcpu_cache);
4075 unregister_reboot_notifier(&kvm_reboot_notifier);
4076 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4079 kvm_arch_hardware_unsetup();
4081 free_cpumask_var(cpus_hardware_enabled);
4089 EXPORT_SYMBOL_GPL(kvm_init);
4093 debugfs_remove_recursive(kvm_debugfs_dir);
4094 misc_deregister(&kvm_dev);
4095 kmem_cache_destroy(kvm_vcpu_cache);
4096 kvm_async_pf_deinit();
4097 unregister_syscore_ops(&kvm_syscore_ops);
4098 unregister_reboot_notifier(&kvm_reboot_notifier);
4099 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4100 on_each_cpu(hardware_disable_nolock, NULL, 1);
4101 kvm_arch_hardware_unsetup();
4104 free_cpumask_var(cpus_hardware_enabled);
4105 kvm_vfio_ops_exit();
4107 EXPORT_SYMBOL_GPL(kvm_exit);