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++;
358 if (kvm_set_spte_hva(kvm, address, pte))
359 kvm_flush_remote_tlbs(kvm);
361 spin_unlock(&kvm->mmu_lock);
362 srcu_read_unlock(&kvm->srcu, idx);
365 static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
366 const struct mmu_notifier_range *range)
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, range->start, range->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, range->start,
389 range->end, range->blockable);
391 srcu_read_unlock(&kvm->srcu, idx);
396 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
397 const struct mmu_notifier_range *range)
399 struct kvm *kvm = mmu_notifier_to_kvm(mn);
401 spin_lock(&kvm->mmu_lock);
403 * This sequence increase will notify the kvm page fault that
404 * the page that is going to be mapped in the spte could have
407 kvm->mmu_notifier_seq++;
410 * The above sequence increase must be visible before the
411 * below count decrease, which is ensured by the smp_wmb above
412 * in conjunction with the smp_rmb in mmu_notifier_retry().
414 kvm->mmu_notifier_count--;
415 spin_unlock(&kvm->mmu_lock);
417 BUG_ON(kvm->mmu_notifier_count < 0);
420 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
421 struct mm_struct *mm,
425 struct kvm *kvm = mmu_notifier_to_kvm(mn);
428 idx = srcu_read_lock(&kvm->srcu);
429 spin_lock(&kvm->mmu_lock);
431 young = kvm_age_hva(kvm, start, end);
433 kvm_flush_remote_tlbs(kvm);
435 spin_unlock(&kvm->mmu_lock);
436 srcu_read_unlock(&kvm->srcu, idx);
441 static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
442 struct mm_struct *mm,
446 struct kvm *kvm = mmu_notifier_to_kvm(mn);
449 idx = srcu_read_lock(&kvm->srcu);
450 spin_lock(&kvm->mmu_lock);
452 * Even though we do not flush TLB, this will still adversely
453 * affect performance on pre-Haswell Intel EPT, where there is
454 * no EPT Access Bit to clear so that we have to tear down EPT
455 * tables instead. If we find this unacceptable, we can always
456 * add a parameter to kvm_age_hva so that it effectively doesn't
457 * do anything on clear_young.
459 * Also note that currently we never issue secondary TLB flushes
460 * from clear_young, leaving this job up to the regular system
461 * cadence. If we find this inaccurate, we might come up with a
462 * more sophisticated heuristic later.
464 young = kvm_age_hva(kvm, start, end);
465 spin_unlock(&kvm->mmu_lock);
466 srcu_read_unlock(&kvm->srcu, idx);
471 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
472 struct mm_struct *mm,
473 unsigned long address)
475 struct kvm *kvm = mmu_notifier_to_kvm(mn);
478 idx = srcu_read_lock(&kvm->srcu);
479 spin_lock(&kvm->mmu_lock);
480 young = kvm_test_age_hva(kvm, address);
481 spin_unlock(&kvm->mmu_lock);
482 srcu_read_unlock(&kvm->srcu, idx);
487 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
488 struct mm_struct *mm)
490 struct kvm *kvm = mmu_notifier_to_kvm(mn);
493 idx = srcu_read_lock(&kvm->srcu);
494 kvm_arch_flush_shadow_all(kvm);
495 srcu_read_unlock(&kvm->srcu, idx);
498 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
499 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
500 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
501 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
502 .clear_young = kvm_mmu_notifier_clear_young,
503 .test_young = kvm_mmu_notifier_test_young,
504 .change_pte = kvm_mmu_notifier_change_pte,
505 .release = kvm_mmu_notifier_release,
508 static int kvm_init_mmu_notifier(struct kvm *kvm)
510 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
511 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
514 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
516 static int kvm_init_mmu_notifier(struct kvm *kvm)
521 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
523 static struct kvm_memslots *kvm_alloc_memslots(void)
526 struct kvm_memslots *slots;
528 slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
532 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
533 slots->id_to_index[i] = slots->memslots[i].id = i;
538 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
540 if (!memslot->dirty_bitmap)
543 kvfree(memslot->dirty_bitmap);
544 memslot->dirty_bitmap = NULL;
548 * Free any memory in @free but not in @dont.
550 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
551 struct kvm_memory_slot *dont)
553 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
554 kvm_destroy_dirty_bitmap(free);
556 kvm_arch_free_memslot(kvm, free, dont);
561 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
563 struct kvm_memory_slot *memslot;
568 kvm_for_each_memslot(memslot, slots)
569 kvm_free_memslot(kvm, memslot, NULL);
574 static void kvm_destroy_vm_debugfs(struct kvm *kvm)
578 if (!kvm->debugfs_dentry)
581 debugfs_remove_recursive(kvm->debugfs_dentry);
583 if (kvm->debugfs_stat_data) {
584 for (i = 0; i < kvm_debugfs_num_entries; i++)
585 kfree(kvm->debugfs_stat_data[i]);
586 kfree(kvm->debugfs_stat_data);
590 static int kvm_create_vm_debugfs(struct kvm *kvm, int fd)
592 char dir_name[ITOA_MAX_LEN * 2];
593 struct kvm_stat_data *stat_data;
594 struct kvm_stats_debugfs_item *p;
596 if (!debugfs_initialized())
599 snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd);
600 kvm->debugfs_dentry = debugfs_create_dir(dir_name, kvm_debugfs_dir);
602 kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries,
603 sizeof(*kvm->debugfs_stat_data),
605 if (!kvm->debugfs_stat_data)
608 for (p = debugfs_entries; p->name; p++) {
609 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL);
613 stat_data->kvm = kvm;
614 stat_data->offset = p->offset;
615 kvm->debugfs_stat_data[p - debugfs_entries] = stat_data;
616 debugfs_create_file(p->name, 0644, kvm->debugfs_dentry,
617 stat_data, stat_fops_per_vm[p->kind]);
622 static struct kvm *kvm_create_vm(unsigned long type)
625 struct kvm *kvm = kvm_arch_alloc_vm();
628 return ERR_PTR(-ENOMEM);
630 spin_lock_init(&kvm->mmu_lock);
632 kvm->mm = current->mm;
633 kvm_eventfd_init(kvm);
634 mutex_init(&kvm->lock);
635 mutex_init(&kvm->irq_lock);
636 mutex_init(&kvm->slots_lock);
637 refcount_set(&kvm->users_count, 1);
638 INIT_LIST_HEAD(&kvm->devices);
640 r = kvm_arch_init_vm(kvm, type);
642 goto out_err_no_disable;
644 r = hardware_enable_all();
646 goto out_err_no_disable;
648 #ifdef CONFIG_HAVE_KVM_IRQFD
649 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
652 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
655 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
656 struct kvm_memslots *slots = kvm_alloc_memslots();
658 goto out_err_no_srcu;
660 * Generations must be different for each address space.
661 * Init kvm generation close to the maximum to easily test the
662 * code of handling generation number wrap-around.
664 slots->generation = i * 2 - 150;
665 rcu_assign_pointer(kvm->memslots[i], slots);
668 if (init_srcu_struct(&kvm->srcu))
669 goto out_err_no_srcu;
670 if (init_srcu_struct(&kvm->irq_srcu))
671 goto out_err_no_irq_srcu;
672 for (i = 0; i < KVM_NR_BUSES; i++) {
673 rcu_assign_pointer(kvm->buses[i],
674 kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL));
679 r = kvm_init_mmu_notifier(kvm);
683 spin_lock(&kvm_lock);
684 list_add(&kvm->vm_list, &vm_list);
685 spin_unlock(&kvm_lock);
687 preempt_notifier_inc();
692 cleanup_srcu_struct(&kvm->irq_srcu);
694 cleanup_srcu_struct(&kvm->srcu);
696 hardware_disable_all();
698 refcount_set(&kvm->users_count, 0);
699 for (i = 0; i < KVM_NR_BUSES; i++)
700 kfree(kvm_get_bus(kvm, i));
701 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
702 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
703 kvm_arch_free_vm(kvm);
708 static void kvm_destroy_devices(struct kvm *kvm)
710 struct kvm_device *dev, *tmp;
713 * We do not need to take the kvm->lock here, because nobody else
714 * has a reference to the struct kvm at this point and therefore
715 * cannot access the devices list anyhow.
717 list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
718 list_del(&dev->vm_node);
719 dev->ops->destroy(dev);
723 static void kvm_destroy_vm(struct kvm *kvm)
726 struct mm_struct *mm = kvm->mm;
728 kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm);
729 kvm_destroy_vm_debugfs(kvm);
730 kvm_arch_sync_events(kvm);
731 spin_lock(&kvm_lock);
732 list_del(&kvm->vm_list);
733 spin_unlock(&kvm_lock);
734 kvm_free_irq_routing(kvm);
735 for (i = 0; i < KVM_NR_BUSES; i++) {
736 struct kvm_io_bus *bus = kvm_get_bus(kvm, i);
739 kvm_io_bus_destroy(bus);
740 kvm->buses[i] = NULL;
742 kvm_coalesced_mmio_free(kvm);
743 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
744 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
746 kvm_arch_flush_shadow_all(kvm);
748 kvm_arch_destroy_vm(kvm);
749 kvm_destroy_devices(kvm);
750 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
751 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
752 cleanup_srcu_struct(&kvm->irq_srcu);
753 cleanup_srcu_struct(&kvm->srcu);
754 kvm_arch_free_vm(kvm);
755 preempt_notifier_dec();
756 hardware_disable_all();
760 void kvm_get_kvm(struct kvm *kvm)
762 refcount_inc(&kvm->users_count);
764 EXPORT_SYMBOL_GPL(kvm_get_kvm);
766 void kvm_put_kvm(struct kvm *kvm)
768 if (refcount_dec_and_test(&kvm->users_count))
771 EXPORT_SYMBOL_GPL(kvm_put_kvm);
774 static int kvm_vm_release(struct inode *inode, struct file *filp)
776 struct kvm *kvm = filp->private_data;
778 kvm_irqfd_release(kvm);
785 * Allocation size is twice as large as the actual dirty bitmap size.
786 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
788 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
790 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
792 memslot->dirty_bitmap = kvzalloc(dirty_bytes, GFP_KERNEL);
793 if (!memslot->dirty_bitmap)
800 * Insert memslot and re-sort memslots based on their GFN,
801 * so binary search could be used to lookup GFN.
802 * Sorting algorithm takes advantage of having initially
803 * sorted array and known changed memslot position.
805 static void update_memslots(struct kvm_memslots *slots,
806 struct kvm_memory_slot *new,
807 enum kvm_mr_change change)
810 int i = slots->id_to_index[id];
811 struct kvm_memory_slot *mslots = slots->memslots;
813 WARN_ON(mslots[i].id != id);
817 WARN_ON(mslots[i].npages || !new->npages);
821 WARN_ON(new->npages || !mslots[i].npages);
827 while (i < KVM_MEM_SLOTS_NUM - 1 &&
828 new->base_gfn <= mslots[i + 1].base_gfn) {
829 if (!mslots[i + 1].npages)
831 mslots[i] = mslots[i + 1];
832 slots->id_to_index[mslots[i].id] = i;
837 * The ">=" is needed when creating a slot with base_gfn == 0,
838 * so that it moves before all those with base_gfn == npages == 0.
840 * On the other hand, if new->npages is zero, the above loop has
841 * already left i pointing to the beginning of the empty part of
842 * mslots, and the ">=" would move the hole backwards in this
843 * case---which is wrong. So skip the loop when deleting a slot.
847 new->base_gfn >= mslots[i - 1].base_gfn) {
848 mslots[i] = mslots[i - 1];
849 slots->id_to_index[mslots[i].id] = i;
853 WARN_ON_ONCE(i != slots->used_slots);
856 slots->id_to_index[mslots[i].id] = i;
859 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
861 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
863 #ifdef __KVM_HAVE_READONLY_MEM
864 valid_flags |= KVM_MEM_READONLY;
867 if (mem->flags & ~valid_flags)
873 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
874 int as_id, struct kvm_memslots *slots)
876 struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id);
879 * Set the low bit in the generation, which disables SPTE caching
880 * until the end of synchronize_srcu_expedited.
882 WARN_ON(old_memslots->generation & 1);
883 slots->generation = old_memslots->generation + 1;
885 rcu_assign_pointer(kvm->memslots[as_id], slots);
886 synchronize_srcu_expedited(&kvm->srcu);
889 * Increment the new memslot generation a second time. This prevents
890 * vm exits that race with memslot updates from caching a memslot
891 * generation that will (potentially) be valid forever.
893 * Generations must be unique even across address spaces. We do not need
894 * a global counter for that, instead the generation space is evenly split
895 * across address spaces. For example, with two address spaces, address
896 * space 0 will use generations 0, 4, 8, ... while * address space 1 will
897 * use generations 2, 6, 10, 14, ...
899 slots->generation += KVM_ADDRESS_SPACE_NUM * 2 - 1;
901 kvm_arch_memslots_updated(kvm, slots);
907 * Allocate some memory and give it an address in the guest physical address
910 * Discontiguous memory is allowed, mostly for framebuffers.
912 * Must be called holding kvm->slots_lock for write.
914 int __kvm_set_memory_region(struct kvm *kvm,
915 const struct kvm_userspace_memory_region *mem)
919 unsigned long npages;
920 struct kvm_memory_slot *slot;
921 struct kvm_memory_slot old, new;
922 struct kvm_memslots *slots = NULL, *old_memslots;
924 enum kvm_mr_change change;
926 r = check_memory_region_flags(mem);
931 as_id = mem->slot >> 16;
934 /* General sanity checks */
935 if (mem->memory_size & (PAGE_SIZE - 1))
937 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
939 /* We can read the guest memory with __xxx_user() later on. */
940 if ((id < KVM_USER_MEM_SLOTS) &&
941 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
942 !access_ok((void __user *)(unsigned long)mem->userspace_addr,
945 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
947 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
950 slot = id_to_memslot(__kvm_memslots(kvm, as_id), id);
951 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
952 npages = mem->memory_size >> PAGE_SHIFT;
954 if (npages > KVM_MEM_MAX_NR_PAGES)
960 new.base_gfn = base_gfn;
962 new.flags = mem->flags;
966 change = KVM_MR_CREATE;
967 else { /* Modify an existing slot. */
968 if ((mem->userspace_addr != old.userspace_addr) ||
969 (npages != old.npages) ||
970 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
973 if (base_gfn != old.base_gfn)
974 change = KVM_MR_MOVE;
975 else if (new.flags != old.flags)
976 change = KVM_MR_FLAGS_ONLY;
977 else { /* Nothing to change. */
986 change = KVM_MR_DELETE;
991 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
992 /* Check for overlaps */
994 kvm_for_each_memslot(slot, __kvm_memslots(kvm, as_id)) {
997 if (!((base_gfn + npages <= slot->base_gfn) ||
998 (base_gfn >= slot->base_gfn + slot->npages)))
1003 /* Free page dirty bitmap if unneeded */
1004 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
1005 new.dirty_bitmap = NULL;
1008 if (change == KVM_MR_CREATE) {
1009 new.userspace_addr = mem->userspace_addr;
1011 if (kvm_arch_create_memslot(kvm, &new, npages))
1015 /* Allocate page dirty bitmap if needed */
1016 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
1017 if (kvm_create_dirty_bitmap(&new) < 0)
1021 slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
1024 memcpy(slots, __kvm_memslots(kvm, as_id), sizeof(struct kvm_memslots));
1026 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
1027 slot = id_to_memslot(slots, id);
1028 slot->flags |= KVM_MEMSLOT_INVALID;
1030 old_memslots = install_new_memslots(kvm, as_id, slots);
1032 /* From this point no new shadow pages pointing to a deleted,
1033 * or moved, memslot will be created.
1035 * validation of sp->gfn happens in:
1036 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1037 * - kvm_is_visible_gfn (mmu_check_roots)
1039 kvm_arch_flush_shadow_memslot(kvm, slot);
1042 * We can re-use the old_memslots from above, the only difference
1043 * from the currently installed memslots is the invalid flag. This
1044 * will get overwritten by update_memslots anyway.
1046 slots = old_memslots;
1049 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
1053 /* actual memory is freed via old in kvm_free_memslot below */
1054 if (change == KVM_MR_DELETE) {
1055 new.dirty_bitmap = NULL;
1056 memset(&new.arch, 0, sizeof(new.arch));
1059 update_memslots(slots, &new, change);
1060 old_memslots = install_new_memslots(kvm, as_id, slots);
1062 kvm_arch_commit_memory_region(kvm, mem, &old, &new, change);
1064 kvm_free_memslot(kvm, &old, &new);
1065 kvfree(old_memslots);
1071 kvm_free_memslot(kvm, &new, &old);
1075 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1077 int kvm_set_memory_region(struct kvm *kvm,
1078 const struct kvm_userspace_memory_region *mem)
1082 mutex_lock(&kvm->slots_lock);
1083 r = __kvm_set_memory_region(kvm, mem);
1084 mutex_unlock(&kvm->slots_lock);
1087 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1089 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1090 struct kvm_userspace_memory_region *mem)
1092 if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1095 return kvm_set_memory_region(kvm, mem);
1098 int kvm_get_dirty_log(struct kvm *kvm,
1099 struct kvm_dirty_log *log, int *is_dirty)
1101 struct kvm_memslots *slots;
1102 struct kvm_memory_slot *memslot;
1105 unsigned long any = 0;
1107 as_id = log->slot >> 16;
1108 id = (u16)log->slot;
1109 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1112 slots = __kvm_memslots(kvm, as_id);
1113 memslot = id_to_memslot(slots, id);
1114 if (!memslot->dirty_bitmap)
1117 n = kvm_dirty_bitmap_bytes(memslot);
1119 for (i = 0; !any && i < n/sizeof(long); ++i)
1120 any = memslot->dirty_bitmap[i];
1122 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1129 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1131 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1133 * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages
1134 * and reenable dirty page tracking for the corresponding pages.
1135 * @kvm: pointer to kvm instance
1136 * @log: slot id and address to which we copy the log
1137 * @is_dirty: flag set if any page is dirty
1139 * We need to keep it in mind that VCPU threads can write to the bitmap
1140 * concurrently. So, to avoid losing track of dirty pages we keep the
1143 * 1. Take a snapshot of the bit and clear it if needed.
1144 * 2. Write protect the corresponding page.
1145 * 3. Copy the snapshot to the userspace.
1146 * 4. Upon return caller flushes TLB's if needed.
1148 * Between 2 and 4, the guest may write to the page using the remaining TLB
1149 * entry. This is not a problem because the page is reported dirty using
1150 * the snapshot taken before and step 4 ensures that writes done after
1151 * exiting to userspace will be logged for the next call.
1154 int kvm_get_dirty_log_protect(struct kvm *kvm,
1155 struct kvm_dirty_log *log, bool *flush)
1157 struct kvm_memslots *slots;
1158 struct kvm_memory_slot *memslot;
1161 unsigned long *dirty_bitmap;
1162 unsigned long *dirty_bitmap_buffer;
1164 as_id = log->slot >> 16;
1165 id = (u16)log->slot;
1166 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1169 slots = __kvm_memslots(kvm, as_id);
1170 memslot = id_to_memslot(slots, id);
1172 dirty_bitmap = memslot->dirty_bitmap;
1176 n = kvm_dirty_bitmap_bytes(memslot);
1178 if (kvm->manual_dirty_log_protect) {
1180 * Unlike kvm_get_dirty_log, we always return false in *flush,
1181 * because no flush is needed until KVM_CLEAR_DIRTY_LOG. There
1182 * is some code duplication between this function and
1183 * kvm_get_dirty_log, but hopefully all architecture
1184 * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log
1185 * can be eliminated.
1187 dirty_bitmap_buffer = dirty_bitmap;
1189 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
1190 memset(dirty_bitmap_buffer, 0, n);
1192 spin_lock(&kvm->mmu_lock);
1193 for (i = 0; i < n / sizeof(long); i++) {
1197 if (!dirty_bitmap[i])
1201 mask = xchg(&dirty_bitmap[i], 0);
1202 dirty_bitmap_buffer[i] = mask;
1205 offset = i * BITS_PER_LONG;
1206 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1210 spin_unlock(&kvm->mmu_lock);
1213 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1217 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect);
1220 * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap
1221 * and reenable dirty page tracking for the corresponding pages.
1222 * @kvm: pointer to kvm instance
1223 * @log: slot id and address from which to fetch the bitmap of dirty pages
1225 int kvm_clear_dirty_log_protect(struct kvm *kvm,
1226 struct kvm_clear_dirty_log *log, bool *flush)
1228 struct kvm_memslots *slots;
1229 struct kvm_memory_slot *memslot;
1233 unsigned long *dirty_bitmap;
1234 unsigned long *dirty_bitmap_buffer;
1236 as_id = log->slot >> 16;
1237 id = (u16)log->slot;
1238 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1241 if ((log->first_page & 63) || (log->num_pages & 63))
1244 slots = __kvm_memslots(kvm, as_id);
1245 memslot = id_to_memslot(slots, id);
1247 dirty_bitmap = memslot->dirty_bitmap;
1251 n = kvm_dirty_bitmap_bytes(memslot);
1253 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
1254 if (copy_from_user(dirty_bitmap_buffer, log->dirty_bitmap, n))
1257 spin_lock(&kvm->mmu_lock);
1258 for (offset = log->first_page,
1259 i = offset / BITS_PER_LONG, n = log->num_pages / BITS_PER_LONG; n--;
1260 i++, offset += BITS_PER_LONG) {
1261 unsigned long mask = *dirty_bitmap_buffer++;
1262 atomic_long_t *p = (atomic_long_t *) &dirty_bitmap[i];
1266 mask &= atomic_long_fetch_andnot(mask, p);
1269 * mask contains the bits that really have been cleared. This
1270 * never includes any bits beyond the length of the memslot (if
1271 * the length is not aligned to 64 pages), therefore it is not
1272 * a problem if userspace sets them in log->dirty_bitmap.
1276 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1280 spin_unlock(&kvm->mmu_lock);
1284 EXPORT_SYMBOL_GPL(kvm_clear_dirty_log_protect);
1287 bool kvm_largepages_enabled(void)
1289 return largepages_enabled;
1292 void kvm_disable_largepages(void)
1294 largepages_enabled = false;
1296 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1298 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1300 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1302 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1304 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
1306 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn);
1309 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1311 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1313 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1314 memslot->flags & KVM_MEMSLOT_INVALID)
1319 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1321 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1323 struct vm_area_struct *vma;
1324 unsigned long addr, size;
1328 addr = gfn_to_hva(kvm, gfn);
1329 if (kvm_is_error_hva(addr))
1332 down_read(¤t->mm->mmap_sem);
1333 vma = find_vma(current->mm, addr);
1337 size = vma_kernel_pagesize(vma);
1340 up_read(¤t->mm->mmap_sem);
1345 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1347 return slot->flags & KVM_MEM_READONLY;
1350 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1351 gfn_t *nr_pages, bool write)
1353 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1354 return KVM_HVA_ERR_BAD;
1356 if (memslot_is_readonly(slot) && write)
1357 return KVM_HVA_ERR_RO_BAD;
1360 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1362 return __gfn_to_hva_memslot(slot, gfn);
1365 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1368 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1371 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1374 return gfn_to_hva_many(slot, gfn, NULL);
1376 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1378 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1380 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1382 EXPORT_SYMBOL_GPL(gfn_to_hva);
1384 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
1386 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
1388 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
1391 * Return the hva of a @gfn and the R/W attribute if possible.
1393 * @slot: the kvm_memory_slot which contains @gfn
1394 * @gfn: the gfn to be translated
1395 * @writable: used to return the read/write attribute of the @slot if the hva
1396 * is valid and @writable is not NULL
1398 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1399 gfn_t gfn, bool *writable)
1401 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1403 if (!kvm_is_error_hva(hva) && writable)
1404 *writable = !memslot_is_readonly(slot);
1409 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1411 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1413 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1416 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
1418 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1420 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1423 static inline int check_user_page_hwpoison(unsigned long addr)
1425 int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
1427 rc = get_user_pages(addr, 1, flags, NULL, NULL);
1428 return rc == -EHWPOISON;
1432 * The fast path to get the writable pfn which will be stored in @pfn,
1433 * true indicates success, otherwise false is returned. It's also the
1434 * only part that runs if we can are in atomic context.
1436 static bool hva_to_pfn_fast(unsigned long addr, bool write_fault,
1437 bool *writable, kvm_pfn_t *pfn)
1439 struct page *page[1];
1443 * Fast pin a writable pfn only if it is a write fault request
1444 * or the caller allows to map a writable pfn for a read fault
1447 if (!(write_fault || writable))
1450 npages = __get_user_pages_fast(addr, 1, 1, page);
1452 *pfn = page_to_pfn(page[0]);
1463 * The slow path to get the pfn of the specified host virtual address,
1464 * 1 indicates success, -errno is returned if error is detected.
1466 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1467 bool *writable, kvm_pfn_t *pfn)
1469 unsigned int flags = FOLL_HWPOISON;
1476 *writable = write_fault;
1479 flags |= FOLL_WRITE;
1481 flags |= FOLL_NOWAIT;
1483 npages = get_user_pages_unlocked(addr, 1, &page, flags);
1487 /* map read fault as writable if possible */
1488 if (unlikely(!write_fault) && writable) {
1491 if (__get_user_pages_fast(addr, 1, 1, &wpage) == 1) {
1497 *pfn = page_to_pfn(page);
1501 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1503 if (unlikely(!(vma->vm_flags & VM_READ)))
1506 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1512 static int hva_to_pfn_remapped(struct vm_area_struct *vma,
1513 unsigned long addr, bool *async,
1514 bool write_fault, bool *writable,
1520 r = follow_pfn(vma, addr, &pfn);
1523 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
1524 * not call the fault handler, so do it here.
1526 bool unlocked = false;
1527 r = fixup_user_fault(current, current->mm, addr,
1528 (write_fault ? FAULT_FLAG_WRITE : 0),
1535 r = follow_pfn(vma, addr, &pfn);
1545 * Get a reference here because callers of *hva_to_pfn* and
1546 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
1547 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
1548 * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will
1549 * simply do nothing for reserved pfns.
1551 * Whoever called remap_pfn_range is also going to call e.g.
1552 * unmap_mapping_range before the underlying pages are freed,
1553 * causing a call to our MMU notifier.
1562 * Pin guest page in memory and return its pfn.
1563 * @addr: host virtual address which maps memory to the guest
1564 * @atomic: whether this function can sleep
1565 * @async: whether this function need to wait IO complete if the
1566 * host page is not in the memory
1567 * @write_fault: whether we should get a writable host page
1568 * @writable: whether it allows to map a writable host page for !@write_fault
1570 * The function will map a writable host page for these two cases:
1571 * 1): @write_fault = true
1572 * 2): @write_fault = false && @writable, @writable will tell the caller
1573 * whether the mapping is writable.
1575 static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1576 bool write_fault, bool *writable)
1578 struct vm_area_struct *vma;
1582 /* we can do it either atomically or asynchronously, not both */
1583 BUG_ON(atomic && async);
1585 if (hva_to_pfn_fast(addr, write_fault, writable, &pfn))
1589 return KVM_PFN_ERR_FAULT;
1591 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1595 down_read(¤t->mm->mmap_sem);
1596 if (npages == -EHWPOISON ||
1597 (!async && check_user_page_hwpoison(addr))) {
1598 pfn = KVM_PFN_ERR_HWPOISON;
1603 vma = find_vma_intersection(current->mm, addr, addr + 1);
1606 pfn = KVM_PFN_ERR_FAULT;
1607 else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
1608 r = hva_to_pfn_remapped(vma, addr, async, write_fault, writable, &pfn);
1612 pfn = KVM_PFN_ERR_FAULT;
1614 if (async && vma_is_valid(vma, write_fault))
1616 pfn = KVM_PFN_ERR_FAULT;
1619 up_read(¤t->mm->mmap_sem);
1623 kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn,
1624 bool atomic, bool *async, bool write_fault,
1627 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1629 if (addr == KVM_HVA_ERR_RO_BAD) {
1632 return KVM_PFN_ERR_RO_FAULT;
1635 if (kvm_is_error_hva(addr)) {
1638 return KVM_PFN_NOSLOT;
1641 /* Do not map writable pfn in the readonly memslot. */
1642 if (writable && memslot_is_readonly(slot)) {
1647 return hva_to_pfn(addr, atomic, async, write_fault,
1650 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
1652 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1655 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
1656 write_fault, writable);
1658 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1660 kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1662 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1664 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
1666 kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1668 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1670 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1672 kvm_pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1674 return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm, gfn), gfn);
1676 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1678 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
1680 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1682 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
1684 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1686 return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
1688 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1690 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
1692 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1694 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
1696 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1697 struct page **pages, int nr_pages)
1702 addr = gfn_to_hva_many(slot, gfn, &entry);
1703 if (kvm_is_error_hva(addr))
1706 if (entry < nr_pages)
1709 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1711 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1713 static struct page *kvm_pfn_to_page(kvm_pfn_t pfn)
1715 if (is_error_noslot_pfn(pfn))
1716 return KVM_ERR_PTR_BAD_PAGE;
1718 if (kvm_is_reserved_pfn(pfn)) {
1720 return KVM_ERR_PTR_BAD_PAGE;
1723 return pfn_to_page(pfn);
1726 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1730 pfn = gfn_to_pfn(kvm, gfn);
1732 return kvm_pfn_to_page(pfn);
1734 EXPORT_SYMBOL_GPL(gfn_to_page);
1736 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
1740 pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
1742 return kvm_pfn_to_page(pfn);
1744 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
1746 void kvm_release_page_clean(struct page *page)
1748 WARN_ON(is_error_page(page));
1750 kvm_release_pfn_clean(page_to_pfn(page));
1752 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1754 void kvm_release_pfn_clean(kvm_pfn_t pfn)
1756 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
1757 put_page(pfn_to_page(pfn));
1759 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1761 void kvm_release_page_dirty(struct page *page)
1763 WARN_ON(is_error_page(page));
1765 kvm_release_pfn_dirty(page_to_pfn(page));
1767 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1769 void kvm_release_pfn_dirty(kvm_pfn_t pfn)
1771 kvm_set_pfn_dirty(pfn);
1772 kvm_release_pfn_clean(pfn);
1774 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1776 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
1778 if (!kvm_is_reserved_pfn(pfn)) {
1779 struct page *page = pfn_to_page(pfn);
1781 if (!PageReserved(page))
1785 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1787 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
1789 if (!kvm_is_reserved_pfn(pfn))
1790 mark_page_accessed(pfn_to_page(pfn));
1792 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1794 void kvm_get_pfn(kvm_pfn_t pfn)
1796 if (!kvm_is_reserved_pfn(pfn))
1797 get_page(pfn_to_page(pfn));
1799 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1801 static int next_segment(unsigned long len, int offset)
1803 if (len > PAGE_SIZE - offset)
1804 return PAGE_SIZE - offset;
1809 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
1810 void *data, int offset, int len)
1815 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1816 if (kvm_is_error_hva(addr))
1818 r = __copy_from_user(data, (void __user *)addr + offset, len);
1824 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1827 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1829 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1831 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1833 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
1834 int offset, int len)
1836 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1838 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1840 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
1842 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1844 gfn_t gfn = gpa >> PAGE_SHIFT;
1846 int offset = offset_in_page(gpa);
1849 while ((seg = next_segment(len, offset)) != 0) {
1850 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1860 EXPORT_SYMBOL_GPL(kvm_read_guest);
1862 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
1864 gfn_t gfn = gpa >> PAGE_SHIFT;
1866 int offset = offset_in_page(gpa);
1869 while ((seg = next_segment(len, offset)) != 0) {
1870 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
1880 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
1882 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1883 void *data, int offset, unsigned long len)
1888 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1889 if (kvm_is_error_hva(addr))
1891 pagefault_disable();
1892 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1899 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1902 gfn_t gfn = gpa >> PAGE_SHIFT;
1903 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1904 int offset = offset_in_page(gpa);
1906 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1908 EXPORT_SYMBOL_GPL(kvm_read_guest_atomic);
1910 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
1911 void *data, unsigned long len)
1913 gfn_t gfn = gpa >> PAGE_SHIFT;
1914 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1915 int offset = offset_in_page(gpa);
1917 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1919 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
1921 static int __kvm_write_guest_page(struct kvm_memory_slot *memslot, gfn_t gfn,
1922 const void *data, int offset, int len)
1927 addr = gfn_to_hva_memslot(memslot, gfn);
1928 if (kvm_is_error_hva(addr))
1930 r = __copy_to_user((void __user *)addr + offset, data, len);
1933 mark_page_dirty_in_slot(memslot, gfn);
1937 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
1938 const void *data, int offset, int len)
1940 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1942 return __kvm_write_guest_page(slot, gfn, data, offset, len);
1944 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1946 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
1947 const void *data, int offset, int len)
1949 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1951 return __kvm_write_guest_page(slot, gfn, data, offset, len);
1953 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
1955 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1958 gfn_t gfn = gpa >> PAGE_SHIFT;
1960 int offset = offset_in_page(gpa);
1963 while ((seg = next_segment(len, offset)) != 0) {
1964 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1974 EXPORT_SYMBOL_GPL(kvm_write_guest);
1976 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
1979 gfn_t gfn = gpa >> PAGE_SHIFT;
1981 int offset = offset_in_page(gpa);
1984 while ((seg = next_segment(len, offset)) != 0) {
1985 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
1995 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
1997 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
1998 struct gfn_to_hva_cache *ghc,
1999 gpa_t gpa, unsigned long len)
2001 int offset = offset_in_page(gpa);
2002 gfn_t start_gfn = gpa >> PAGE_SHIFT;
2003 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
2004 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
2005 gfn_t nr_pages_avail;
2006 int r = start_gfn <= end_gfn ? 0 : -EINVAL;
2009 ghc->generation = slots->generation;
2011 ghc->hva = KVM_HVA_ERR_BAD;
2014 * If the requested region crosses two memslots, we still
2015 * verify that the entire region is valid here.
2017 while (!r && start_gfn <= end_gfn) {
2018 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
2019 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
2021 if (kvm_is_error_hva(ghc->hva))
2023 start_gfn += nr_pages_avail;
2026 /* Use the slow path for cross page reads and writes. */
2027 if (!r && nr_pages_needed == 1)
2030 ghc->memslot = NULL;
2035 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2036 gpa_t gpa, unsigned long len)
2038 struct kvm_memslots *slots = kvm_memslots(kvm);
2039 return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len);
2041 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
2043 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2044 void *data, unsigned int offset,
2047 struct kvm_memslots *slots = kvm_memslots(kvm);
2049 gpa_t gpa = ghc->gpa + offset;
2051 BUG_ON(len + offset > ghc->len);
2053 if (slots->generation != ghc->generation)
2054 __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len);
2056 if (unlikely(!ghc->memslot))
2057 return kvm_write_guest(kvm, gpa, data, len);
2059 if (kvm_is_error_hva(ghc->hva))
2062 r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
2065 mark_page_dirty_in_slot(ghc->memslot, gpa >> PAGE_SHIFT);
2069 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached);
2071 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2072 void *data, unsigned long len)
2074 return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len);
2076 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
2078 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2079 void *data, unsigned long len)
2081 struct kvm_memslots *slots = kvm_memslots(kvm);
2084 BUG_ON(len > ghc->len);
2086 if (slots->generation != ghc->generation)
2087 __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len);
2089 if (unlikely(!ghc->memslot))
2090 return kvm_read_guest(kvm, ghc->gpa, data, len);
2092 if (kvm_is_error_hva(ghc->hva))
2095 r = __copy_from_user(data, (void __user *)ghc->hva, len);
2101 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
2103 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
2105 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
2107 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
2109 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
2111 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
2113 gfn_t gfn = gpa >> PAGE_SHIFT;
2115 int offset = offset_in_page(gpa);
2118 while ((seg = next_segment(len, offset)) != 0) {
2119 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
2128 EXPORT_SYMBOL_GPL(kvm_clear_guest);
2130 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot,
2133 if (memslot && memslot->dirty_bitmap) {
2134 unsigned long rel_gfn = gfn - memslot->base_gfn;
2136 set_bit_le(rel_gfn, memslot->dirty_bitmap);
2140 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
2142 struct kvm_memory_slot *memslot;
2144 memslot = gfn_to_memslot(kvm, gfn);
2145 mark_page_dirty_in_slot(memslot, gfn);
2147 EXPORT_SYMBOL_GPL(mark_page_dirty);
2149 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
2151 struct kvm_memory_slot *memslot;
2153 memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2154 mark_page_dirty_in_slot(memslot, gfn);
2156 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
2158 void kvm_sigset_activate(struct kvm_vcpu *vcpu)
2160 if (!vcpu->sigset_active)
2164 * This does a lockless modification of ->real_blocked, which is fine
2165 * because, only current can change ->real_blocked and all readers of
2166 * ->real_blocked don't care as long ->real_blocked is always a subset
2169 sigprocmask(SIG_SETMASK, &vcpu->sigset, ¤t->real_blocked);
2172 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu)
2174 if (!vcpu->sigset_active)
2177 sigprocmask(SIG_SETMASK, ¤t->real_blocked, NULL);
2178 sigemptyset(¤t->real_blocked);
2181 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
2183 unsigned int old, val, grow;
2185 old = val = vcpu->halt_poll_ns;
2186 grow = READ_ONCE(halt_poll_ns_grow);
2188 if (val == 0 && grow)
2193 if (val > halt_poll_ns)
2196 vcpu->halt_poll_ns = val;
2197 trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
2200 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
2202 unsigned int old, val, shrink;
2204 old = val = vcpu->halt_poll_ns;
2205 shrink = READ_ONCE(halt_poll_ns_shrink);
2211 vcpu->halt_poll_ns = val;
2212 trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
2215 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
2218 int idx = srcu_read_lock(&vcpu->kvm->srcu);
2220 if (kvm_arch_vcpu_runnable(vcpu)) {
2221 kvm_make_request(KVM_REQ_UNHALT, vcpu);
2224 if (kvm_cpu_has_pending_timer(vcpu))
2226 if (signal_pending(current))
2231 srcu_read_unlock(&vcpu->kvm->srcu, idx);
2236 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2238 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
2241 DECLARE_SWAITQUEUE(wait);
2242 bool waited = false;
2245 start = cur = ktime_get();
2246 if (vcpu->halt_poll_ns) {
2247 ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns);
2249 ++vcpu->stat.halt_attempted_poll;
2252 * This sets KVM_REQ_UNHALT if an interrupt
2255 if (kvm_vcpu_check_block(vcpu) < 0) {
2256 ++vcpu->stat.halt_successful_poll;
2257 if (!vcpu_valid_wakeup(vcpu))
2258 ++vcpu->stat.halt_poll_invalid;
2262 } while (single_task_running() && ktime_before(cur, stop));
2265 kvm_arch_vcpu_blocking(vcpu);
2268 prepare_to_swait_exclusive(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
2270 if (kvm_vcpu_check_block(vcpu) < 0)
2277 finish_swait(&vcpu->wq, &wait);
2280 kvm_arch_vcpu_unblocking(vcpu);
2282 block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
2284 if (!vcpu_valid_wakeup(vcpu))
2285 shrink_halt_poll_ns(vcpu);
2286 else if (halt_poll_ns) {
2287 if (block_ns <= vcpu->halt_poll_ns)
2289 /* we had a long block, shrink polling */
2290 else if (vcpu->halt_poll_ns && block_ns > halt_poll_ns)
2291 shrink_halt_poll_ns(vcpu);
2292 /* we had a short halt and our poll time is too small */
2293 else if (vcpu->halt_poll_ns < halt_poll_ns &&
2294 block_ns < halt_poll_ns)
2295 grow_halt_poll_ns(vcpu);
2297 vcpu->halt_poll_ns = 0;
2299 trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu));
2300 kvm_arch_vcpu_block_finish(vcpu);
2302 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
2304 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
2306 struct swait_queue_head *wqp;
2308 wqp = kvm_arch_vcpu_wq(vcpu);
2309 if (swq_has_sleeper(wqp)) {
2311 ++vcpu->stat.halt_wakeup;
2317 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
2321 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2323 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
2326 int cpu = vcpu->cpu;
2328 if (kvm_vcpu_wake_up(vcpu))
2332 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
2333 if (kvm_arch_vcpu_should_kick(vcpu))
2334 smp_send_reschedule(cpu);
2337 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
2338 #endif /* !CONFIG_S390 */
2340 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
2343 struct task_struct *task = NULL;
2347 pid = rcu_dereference(target->pid);
2349 task = get_pid_task(pid, PIDTYPE_PID);
2353 ret = yield_to(task, 1);
2354 put_task_struct(task);
2358 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
2361 * Helper that checks whether a VCPU is eligible for directed yield.
2362 * Most eligible candidate to yield is decided by following heuristics:
2364 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2365 * (preempted lock holder), indicated by @in_spin_loop.
2366 * Set at the beiginning and cleared at the end of interception/PLE handler.
2368 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2369 * chance last time (mostly it has become eligible now since we have probably
2370 * yielded to lockholder in last iteration. This is done by toggling
2371 * @dy_eligible each time a VCPU checked for eligibility.)
2373 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2374 * to preempted lock-holder could result in wrong VCPU selection and CPU
2375 * burning. Giving priority for a potential lock-holder increases lock
2378 * Since algorithm is based on heuristics, accessing another VCPU data without
2379 * locking does not harm. It may result in trying to yield to same VCPU, fail
2380 * and continue with next VCPU and so on.
2382 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
2384 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2387 eligible = !vcpu->spin_loop.in_spin_loop ||
2388 vcpu->spin_loop.dy_eligible;
2390 if (vcpu->spin_loop.in_spin_loop)
2391 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
2399 void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode)
2401 struct kvm *kvm = me->kvm;
2402 struct kvm_vcpu *vcpu;
2403 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
2409 kvm_vcpu_set_in_spin_loop(me, true);
2411 * We boost the priority of a VCPU that is runnable but not
2412 * currently running, because it got preempted by something
2413 * else and called schedule in __vcpu_run. Hopefully that
2414 * VCPU is holding the lock that we need and will release it.
2415 * We approximate round-robin by starting at the last boosted VCPU.
2417 for (pass = 0; pass < 2 && !yielded && try; pass++) {
2418 kvm_for_each_vcpu(i, vcpu, kvm) {
2419 if (!pass && i <= last_boosted_vcpu) {
2420 i = last_boosted_vcpu;
2422 } else if (pass && i > last_boosted_vcpu)
2424 if (!READ_ONCE(vcpu->preempted))
2428 if (swait_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
2430 if (yield_to_kernel_mode && !kvm_arch_vcpu_in_kernel(vcpu))
2432 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
2435 yielded = kvm_vcpu_yield_to(vcpu);
2437 kvm->last_boosted_vcpu = i;
2439 } else if (yielded < 0) {
2446 kvm_vcpu_set_in_spin_loop(me, false);
2448 /* Ensure vcpu is not eligible during next spinloop */
2449 kvm_vcpu_set_dy_eligible(me, false);
2451 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
2453 static vm_fault_t kvm_vcpu_fault(struct vm_fault *vmf)
2455 struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data;
2458 if (vmf->pgoff == 0)
2459 page = virt_to_page(vcpu->run);
2461 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
2462 page = virt_to_page(vcpu->arch.pio_data);
2464 #ifdef CONFIG_KVM_MMIO
2465 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
2466 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
2469 return kvm_arch_vcpu_fault(vcpu, vmf);
2475 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
2476 .fault = kvm_vcpu_fault,
2479 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2481 vma->vm_ops = &kvm_vcpu_vm_ops;
2485 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2487 struct kvm_vcpu *vcpu = filp->private_data;
2489 debugfs_remove_recursive(vcpu->debugfs_dentry);
2490 kvm_put_kvm(vcpu->kvm);
2494 static struct file_operations kvm_vcpu_fops = {
2495 .release = kvm_vcpu_release,
2496 .unlocked_ioctl = kvm_vcpu_ioctl,
2497 .mmap = kvm_vcpu_mmap,
2498 .llseek = noop_llseek,
2499 KVM_COMPAT(kvm_vcpu_compat_ioctl),
2503 * Allocates an inode for the vcpu.
2505 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2507 char name[8 + 1 + ITOA_MAX_LEN + 1];
2509 snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id);
2510 return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
2513 static int kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
2515 char dir_name[ITOA_MAX_LEN * 2];
2518 if (!kvm_arch_has_vcpu_debugfs())
2521 if (!debugfs_initialized())
2524 snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
2525 vcpu->debugfs_dentry = debugfs_create_dir(dir_name,
2526 vcpu->kvm->debugfs_dentry);
2527 if (!vcpu->debugfs_dentry)
2530 ret = kvm_arch_create_vcpu_debugfs(vcpu);
2532 debugfs_remove_recursive(vcpu->debugfs_dentry);
2540 * Creates some virtual cpus. Good luck creating more than one.
2542 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
2545 struct kvm_vcpu *vcpu;
2547 if (id >= KVM_MAX_VCPU_ID)
2550 mutex_lock(&kvm->lock);
2551 if (kvm->created_vcpus == KVM_MAX_VCPUS) {
2552 mutex_unlock(&kvm->lock);
2556 kvm->created_vcpus++;
2557 mutex_unlock(&kvm->lock);
2559 vcpu = kvm_arch_vcpu_create(kvm, id);
2562 goto vcpu_decrement;
2565 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2567 r = kvm_arch_vcpu_setup(vcpu);
2571 r = kvm_create_vcpu_debugfs(vcpu);
2575 mutex_lock(&kvm->lock);
2576 if (kvm_get_vcpu_by_id(kvm, id)) {
2578 goto unlock_vcpu_destroy;
2581 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
2583 /* Now it's all set up, let userspace reach it */
2585 r = create_vcpu_fd(vcpu);
2588 goto unlock_vcpu_destroy;
2591 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
2594 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
2595 * before kvm->online_vcpu's incremented value.
2598 atomic_inc(&kvm->online_vcpus);
2600 mutex_unlock(&kvm->lock);
2601 kvm_arch_vcpu_postcreate(vcpu);
2604 unlock_vcpu_destroy:
2605 mutex_unlock(&kvm->lock);
2606 debugfs_remove_recursive(vcpu->debugfs_dentry);
2608 kvm_arch_vcpu_destroy(vcpu);
2610 mutex_lock(&kvm->lock);
2611 kvm->created_vcpus--;
2612 mutex_unlock(&kvm->lock);
2616 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2619 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2620 vcpu->sigset_active = 1;
2621 vcpu->sigset = *sigset;
2623 vcpu->sigset_active = 0;
2627 static long kvm_vcpu_ioctl(struct file *filp,
2628 unsigned int ioctl, unsigned long arg)
2630 struct kvm_vcpu *vcpu = filp->private_data;
2631 void __user *argp = (void __user *)arg;
2633 struct kvm_fpu *fpu = NULL;
2634 struct kvm_sregs *kvm_sregs = NULL;
2636 if (vcpu->kvm->mm != current->mm)
2639 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2643 * Some architectures have vcpu ioctls that are asynchronous to vcpu
2644 * execution; mutex_lock() would break them.
2646 r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg);
2647 if (r != -ENOIOCTLCMD)
2650 if (mutex_lock_killable(&vcpu->mutex))
2658 oldpid = rcu_access_pointer(vcpu->pid);
2659 if (unlikely(oldpid != task_pid(current))) {
2660 /* The thread running this VCPU changed. */
2663 r = kvm_arch_vcpu_run_pid_change(vcpu);
2667 newpid = get_task_pid(current, PIDTYPE_PID);
2668 rcu_assign_pointer(vcpu->pid, newpid);
2673 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2674 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2677 case KVM_GET_REGS: {
2678 struct kvm_regs *kvm_regs;
2681 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2684 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2688 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2695 case KVM_SET_REGS: {
2696 struct kvm_regs *kvm_regs;
2699 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2700 if (IS_ERR(kvm_regs)) {
2701 r = PTR_ERR(kvm_regs);
2704 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2708 case KVM_GET_SREGS: {
2709 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2713 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2717 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2722 case KVM_SET_SREGS: {
2723 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2724 if (IS_ERR(kvm_sregs)) {
2725 r = PTR_ERR(kvm_sregs);
2729 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2732 case KVM_GET_MP_STATE: {
2733 struct kvm_mp_state mp_state;
2735 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2739 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
2744 case KVM_SET_MP_STATE: {
2745 struct kvm_mp_state mp_state;
2748 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
2750 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2753 case KVM_TRANSLATE: {
2754 struct kvm_translation tr;
2757 if (copy_from_user(&tr, argp, sizeof(tr)))
2759 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2763 if (copy_to_user(argp, &tr, sizeof(tr)))
2768 case KVM_SET_GUEST_DEBUG: {
2769 struct kvm_guest_debug dbg;
2772 if (copy_from_user(&dbg, argp, sizeof(dbg)))
2774 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2777 case KVM_SET_SIGNAL_MASK: {
2778 struct kvm_signal_mask __user *sigmask_arg = argp;
2779 struct kvm_signal_mask kvm_sigmask;
2780 sigset_t sigset, *p;
2785 if (copy_from_user(&kvm_sigmask, argp,
2786 sizeof(kvm_sigmask)))
2789 if (kvm_sigmask.len != sizeof(sigset))
2792 if (copy_from_user(&sigset, sigmask_arg->sigset,
2797 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2801 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2805 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2809 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2815 fpu = memdup_user(argp, sizeof(*fpu));
2821 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2825 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2828 mutex_unlock(&vcpu->mutex);
2834 #ifdef CONFIG_KVM_COMPAT
2835 static long kvm_vcpu_compat_ioctl(struct file *filp,
2836 unsigned int ioctl, unsigned long arg)
2838 struct kvm_vcpu *vcpu = filp->private_data;
2839 void __user *argp = compat_ptr(arg);
2842 if (vcpu->kvm->mm != current->mm)
2846 case KVM_SET_SIGNAL_MASK: {
2847 struct kvm_signal_mask __user *sigmask_arg = argp;
2848 struct kvm_signal_mask kvm_sigmask;
2853 if (copy_from_user(&kvm_sigmask, argp,
2854 sizeof(kvm_sigmask)))
2857 if (kvm_sigmask.len != sizeof(compat_sigset_t))
2860 if (get_compat_sigset(&sigset, (void *)sigmask_arg->sigset))
2862 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2864 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2868 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2876 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2877 int (*accessor)(struct kvm_device *dev,
2878 struct kvm_device_attr *attr),
2881 struct kvm_device_attr attr;
2886 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2889 return accessor(dev, &attr);
2892 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2895 struct kvm_device *dev = filp->private_data;
2898 case KVM_SET_DEVICE_ATTR:
2899 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2900 case KVM_GET_DEVICE_ATTR:
2901 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2902 case KVM_HAS_DEVICE_ATTR:
2903 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2905 if (dev->ops->ioctl)
2906 return dev->ops->ioctl(dev, ioctl, arg);
2912 static int kvm_device_release(struct inode *inode, struct file *filp)
2914 struct kvm_device *dev = filp->private_data;
2915 struct kvm *kvm = dev->kvm;
2921 static const struct file_operations kvm_device_fops = {
2922 .unlocked_ioctl = kvm_device_ioctl,
2923 .release = kvm_device_release,
2924 KVM_COMPAT(kvm_device_ioctl),
2927 struct kvm_device *kvm_device_from_filp(struct file *filp)
2929 if (filp->f_op != &kvm_device_fops)
2932 return filp->private_data;
2935 static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
2936 #ifdef CONFIG_KVM_MPIC
2937 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
2938 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
2942 int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
2944 if (type >= ARRAY_SIZE(kvm_device_ops_table))
2947 if (kvm_device_ops_table[type] != NULL)
2950 kvm_device_ops_table[type] = ops;
2954 void kvm_unregister_device_ops(u32 type)
2956 if (kvm_device_ops_table[type] != NULL)
2957 kvm_device_ops_table[type] = NULL;
2960 static int kvm_ioctl_create_device(struct kvm *kvm,
2961 struct kvm_create_device *cd)
2963 struct kvm_device_ops *ops = NULL;
2964 struct kvm_device *dev;
2965 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2968 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
2971 ops = kvm_device_ops_table[cd->type];
2978 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2985 mutex_lock(&kvm->lock);
2986 ret = ops->create(dev, cd->type);
2988 mutex_unlock(&kvm->lock);
2992 list_add(&dev->vm_node, &kvm->devices);
2993 mutex_unlock(&kvm->lock);
2998 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
3000 mutex_lock(&kvm->lock);
3001 list_del(&dev->vm_node);
3002 mutex_unlock(&kvm->lock);
3012 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
3015 case KVM_CAP_USER_MEMORY:
3016 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
3017 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
3018 case KVM_CAP_INTERNAL_ERROR_DATA:
3019 #ifdef CONFIG_HAVE_KVM_MSI
3020 case KVM_CAP_SIGNAL_MSI:
3022 #ifdef CONFIG_HAVE_KVM_IRQFD
3024 case KVM_CAP_IRQFD_RESAMPLE:
3026 case KVM_CAP_IOEVENTFD_ANY_LENGTH:
3027 case KVM_CAP_CHECK_EXTENSION_VM:
3028 case KVM_CAP_ENABLE_CAP_VM:
3029 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3030 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT:
3033 #ifdef CONFIG_KVM_MMIO
3034 case KVM_CAP_COALESCED_MMIO:
3035 return KVM_COALESCED_MMIO_PAGE_OFFSET;
3036 case KVM_CAP_COALESCED_PIO:
3039 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3040 case KVM_CAP_IRQ_ROUTING:
3041 return KVM_MAX_IRQ_ROUTES;
3043 #if KVM_ADDRESS_SPACE_NUM > 1
3044 case KVM_CAP_MULTI_ADDRESS_SPACE:
3045 return KVM_ADDRESS_SPACE_NUM;
3047 case KVM_CAP_MAX_VCPU_ID:
3048 return KVM_MAX_VCPU_ID;
3052 return kvm_vm_ioctl_check_extension(kvm, arg);
3055 int __attribute__((weak)) kvm_vm_ioctl_enable_cap(struct kvm *kvm,
3056 struct kvm_enable_cap *cap)
3061 static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm,
3062 struct kvm_enable_cap *cap)
3065 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3066 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT:
3067 if (cap->flags || (cap->args[0] & ~1))
3069 kvm->manual_dirty_log_protect = cap->args[0];
3073 return kvm_vm_ioctl_enable_cap(kvm, cap);
3077 static long kvm_vm_ioctl(struct file *filp,
3078 unsigned int ioctl, unsigned long arg)
3080 struct kvm *kvm = filp->private_data;
3081 void __user *argp = (void __user *)arg;
3084 if (kvm->mm != current->mm)
3087 case KVM_CREATE_VCPU:
3088 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
3090 case KVM_ENABLE_CAP: {
3091 struct kvm_enable_cap cap;
3094 if (copy_from_user(&cap, argp, sizeof(cap)))
3096 r = kvm_vm_ioctl_enable_cap_generic(kvm, &cap);
3099 case KVM_SET_USER_MEMORY_REGION: {
3100 struct kvm_userspace_memory_region kvm_userspace_mem;
3103 if (copy_from_user(&kvm_userspace_mem, argp,
3104 sizeof(kvm_userspace_mem)))
3107 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
3110 case KVM_GET_DIRTY_LOG: {
3111 struct kvm_dirty_log log;
3114 if (copy_from_user(&log, argp, sizeof(log)))
3116 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3119 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3120 case KVM_CLEAR_DIRTY_LOG: {
3121 struct kvm_clear_dirty_log log;
3124 if (copy_from_user(&log, argp, sizeof(log)))
3126 r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
3130 #ifdef CONFIG_KVM_MMIO
3131 case KVM_REGISTER_COALESCED_MMIO: {
3132 struct kvm_coalesced_mmio_zone zone;
3135 if (copy_from_user(&zone, argp, sizeof(zone)))
3137 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
3140 case KVM_UNREGISTER_COALESCED_MMIO: {
3141 struct kvm_coalesced_mmio_zone zone;
3144 if (copy_from_user(&zone, argp, sizeof(zone)))
3146 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
3151 struct kvm_irqfd data;
3154 if (copy_from_user(&data, argp, sizeof(data)))
3156 r = kvm_irqfd(kvm, &data);
3159 case KVM_IOEVENTFD: {
3160 struct kvm_ioeventfd data;
3163 if (copy_from_user(&data, argp, sizeof(data)))
3165 r = kvm_ioeventfd(kvm, &data);
3168 #ifdef CONFIG_HAVE_KVM_MSI
3169 case KVM_SIGNAL_MSI: {
3173 if (copy_from_user(&msi, argp, sizeof(msi)))
3175 r = kvm_send_userspace_msi(kvm, &msi);
3179 #ifdef __KVM_HAVE_IRQ_LINE
3180 case KVM_IRQ_LINE_STATUS:
3181 case KVM_IRQ_LINE: {
3182 struct kvm_irq_level irq_event;
3185 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
3188 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
3189 ioctl == KVM_IRQ_LINE_STATUS);
3194 if (ioctl == KVM_IRQ_LINE_STATUS) {
3195 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
3203 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3204 case KVM_SET_GSI_ROUTING: {
3205 struct kvm_irq_routing routing;
3206 struct kvm_irq_routing __user *urouting;
3207 struct kvm_irq_routing_entry *entries = NULL;
3210 if (copy_from_user(&routing, argp, sizeof(routing)))
3213 if (!kvm_arch_can_set_irq_routing(kvm))
3215 if (routing.nr > KVM_MAX_IRQ_ROUTES)
3221 entries = vmalloc(array_size(sizeof(*entries),
3227 if (copy_from_user(entries, urouting->entries,
3228 routing.nr * sizeof(*entries)))
3229 goto out_free_irq_routing;
3231 r = kvm_set_irq_routing(kvm, entries, routing.nr,
3233 out_free_irq_routing:
3237 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
3238 case KVM_CREATE_DEVICE: {
3239 struct kvm_create_device cd;
3242 if (copy_from_user(&cd, argp, sizeof(cd)))
3245 r = kvm_ioctl_create_device(kvm, &cd);
3250 if (copy_to_user(argp, &cd, sizeof(cd)))
3256 case KVM_CHECK_EXTENSION:
3257 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
3260 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
3266 #ifdef CONFIG_KVM_COMPAT
3267 struct compat_kvm_dirty_log {
3271 compat_uptr_t dirty_bitmap; /* one bit per page */
3276 static long kvm_vm_compat_ioctl(struct file *filp,
3277 unsigned int ioctl, unsigned long arg)
3279 struct kvm *kvm = filp->private_data;
3282 if (kvm->mm != current->mm)
3285 case KVM_GET_DIRTY_LOG: {
3286 struct compat_kvm_dirty_log compat_log;
3287 struct kvm_dirty_log log;
3289 if (copy_from_user(&compat_log, (void __user *)arg,
3290 sizeof(compat_log)))
3292 log.slot = compat_log.slot;
3293 log.padding1 = compat_log.padding1;
3294 log.padding2 = compat_log.padding2;
3295 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
3297 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3301 r = kvm_vm_ioctl(filp, ioctl, arg);
3307 static struct file_operations kvm_vm_fops = {
3308 .release = kvm_vm_release,
3309 .unlocked_ioctl = kvm_vm_ioctl,
3310 .llseek = noop_llseek,
3311 KVM_COMPAT(kvm_vm_compat_ioctl),
3314 static int kvm_dev_ioctl_create_vm(unsigned long type)
3320 kvm = kvm_create_vm(type);
3322 return PTR_ERR(kvm);
3323 #ifdef CONFIG_KVM_MMIO
3324 r = kvm_coalesced_mmio_init(kvm);
3328 r = get_unused_fd_flags(O_CLOEXEC);
3332 file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
3340 * Don't call kvm_put_kvm anymore at this point; file->f_op is
3341 * already set, with ->release() being kvm_vm_release(). In error
3342 * cases it will be called by the final fput(file) and will take
3343 * care of doing kvm_put_kvm(kvm).
3345 if (kvm_create_vm_debugfs(kvm, r) < 0) {
3350 kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm);
3352 fd_install(r, file);
3360 static long kvm_dev_ioctl(struct file *filp,
3361 unsigned int ioctl, unsigned long arg)
3366 case KVM_GET_API_VERSION:
3369 r = KVM_API_VERSION;
3372 r = kvm_dev_ioctl_create_vm(arg);
3374 case KVM_CHECK_EXTENSION:
3375 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
3377 case KVM_GET_VCPU_MMAP_SIZE:
3380 r = PAGE_SIZE; /* struct kvm_run */
3382 r += PAGE_SIZE; /* pio data page */
3384 #ifdef CONFIG_KVM_MMIO
3385 r += PAGE_SIZE; /* coalesced mmio ring page */
3388 case KVM_TRACE_ENABLE:
3389 case KVM_TRACE_PAUSE:
3390 case KVM_TRACE_DISABLE:
3394 return kvm_arch_dev_ioctl(filp, ioctl, arg);
3400 static struct file_operations kvm_chardev_ops = {
3401 .unlocked_ioctl = kvm_dev_ioctl,
3402 .llseek = noop_llseek,
3403 KVM_COMPAT(kvm_dev_ioctl),
3406 static struct miscdevice kvm_dev = {
3412 static void hardware_enable_nolock(void *junk)
3414 int cpu = raw_smp_processor_id();
3417 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
3420 cpumask_set_cpu(cpu, cpus_hardware_enabled);
3422 r = kvm_arch_hardware_enable();
3425 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3426 atomic_inc(&hardware_enable_failed);
3427 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
3431 static int kvm_starting_cpu(unsigned int cpu)
3433 raw_spin_lock(&kvm_count_lock);
3434 if (kvm_usage_count)
3435 hardware_enable_nolock(NULL);
3436 raw_spin_unlock(&kvm_count_lock);
3440 static void hardware_disable_nolock(void *junk)
3442 int cpu = raw_smp_processor_id();
3444 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
3446 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3447 kvm_arch_hardware_disable();
3450 static int kvm_dying_cpu(unsigned int cpu)
3452 raw_spin_lock(&kvm_count_lock);
3453 if (kvm_usage_count)
3454 hardware_disable_nolock(NULL);
3455 raw_spin_unlock(&kvm_count_lock);
3459 static void hardware_disable_all_nolock(void)
3461 BUG_ON(!kvm_usage_count);
3464 if (!kvm_usage_count)
3465 on_each_cpu(hardware_disable_nolock, NULL, 1);
3468 static void hardware_disable_all(void)
3470 raw_spin_lock(&kvm_count_lock);
3471 hardware_disable_all_nolock();
3472 raw_spin_unlock(&kvm_count_lock);
3475 static int hardware_enable_all(void)
3479 raw_spin_lock(&kvm_count_lock);
3482 if (kvm_usage_count == 1) {
3483 atomic_set(&hardware_enable_failed, 0);
3484 on_each_cpu(hardware_enable_nolock, NULL, 1);
3486 if (atomic_read(&hardware_enable_failed)) {
3487 hardware_disable_all_nolock();
3492 raw_spin_unlock(&kvm_count_lock);
3497 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3501 * Some (well, at least mine) BIOSes hang on reboot if
3504 * And Intel TXT required VMX off for all cpu when system shutdown.
3506 pr_info("kvm: exiting hardware virtualization\n");
3507 kvm_rebooting = true;
3508 on_each_cpu(hardware_disable_nolock, NULL, 1);
3512 static struct notifier_block kvm_reboot_notifier = {
3513 .notifier_call = kvm_reboot,
3517 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3521 for (i = 0; i < bus->dev_count; i++) {
3522 struct kvm_io_device *pos = bus->range[i].dev;
3524 kvm_iodevice_destructor(pos);
3529 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
3530 const struct kvm_io_range *r2)
3532 gpa_t addr1 = r1->addr;
3533 gpa_t addr2 = r2->addr;
3538 /* If r2->len == 0, match the exact address. If r2->len != 0,
3539 * accept any overlapping write. Any order is acceptable for
3540 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
3541 * we process all of them.
3554 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
3556 return kvm_io_bus_cmp(p1, p2);
3559 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
3560 gpa_t addr, int len)
3562 struct kvm_io_range *range, key;
3565 key = (struct kvm_io_range) {
3570 range = bsearch(&key, bus->range, bus->dev_count,
3571 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
3575 off = range - bus->range;
3577 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
3583 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3584 struct kvm_io_range *range, const void *val)
3588 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3592 while (idx < bus->dev_count &&
3593 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3594 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
3603 /* kvm_io_bus_write - called under kvm->slots_lock */
3604 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3605 int len, const void *val)
3607 struct kvm_io_bus *bus;
3608 struct kvm_io_range range;
3611 range = (struct kvm_io_range) {
3616 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3619 r = __kvm_io_bus_write(vcpu, bus, &range, val);
3620 return r < 0 ? r : 0;
3623 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3624 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
3625 gpa_t addr, int len, const void *val, long cookie)
3627 struct kvm_io_bus *bus;
3628 struct kvm_io_range range;
3630 range = (struct kvm_io_range) {
3635 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3639 /* First try the device referenced by cookie. */
3640 if ((cookie >= 0) && (cookie < bus->dev_count) &&
3641 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
3642 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
3647 * cookie contained garbage; fall back to search and return the
3648 * correct cookie value.
3650 return __kvm_io_bus_write(vcpu, bus, &range, val);
3653 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3654 struct kvm_io_range *range, void *val)
3658 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3662 while (idx < bus->dev_count &&
3663 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3664 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
3672 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
3674 /* kvm_io_bus_read - called under kvm->slots_lock */
3675 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3678 struct kvm_io_bus *bus;
3679 struct kvm_io_range range;
3682 range = (struct kvm_io_range) {
3687 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3690 r = __kvm_io_bus_read(vcpu, bus, &range, val);
3691 return r < 0 ? r : 0;
3695 /* Caller must hold slots_lock. */
3696 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3697 int len, struct kvm_io_device *dev)
3700 struct kvm_io_bus *new_bus, *bus;
3701 struct kvm_io_range range;
3703 bus = kvm_get_bus(kvm, bus_idx);
3707 /* exclude ioeventfd which is limited by maximum fd */
3708 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3711 new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count + 1) *
3712 sizeof(struct kvm_io_range)), GFP_KERNEL);
3716 range = (struct kvm_io_range) {
3722 for (i = 0; i < bus->dev_count; i++)
3723 if (kvm_io_bus_cmp(&bus->range[i], &range) > 0)
3726 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3727 new_bus->dev_count++;
3728 new_bus->range[i] = range;
3729 memcpy(new_bus->range + i + 1, bus->range + i,
3730 (bus->dev_count - i) * sizeof(struct kvm_io_range));
3731 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3732 synchronize_srcu_expedited(&kvm->srcu);
3738 /* Caller must hold slots_lock. */
3739 void kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3740 struct kvm_io_device *dev)
3743 struct kvm_io_bus *new_bus, *bus;
3745 bus = kvm_get_bus(kvm, bus_idx);
3749 for (i = 0; i < bus->dev_count; i++)
3750 if (bus->range[i].dev == dev) {
3754 if (i == bus->dev_count)
3757 new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3758 sizeof(struct kvm_io_range)), GFP_KERNEL);
3760 pr_err("kvm: failed to shrink bus, removing it completely\n");
3764 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3765 new_bus->dev_count--;
3766 memcpy(new_bus->range + i, bus->range + i + 1,
3767 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3770 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3771 synchronize_srcu_expedited(&kvm->srcu);
3776 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3779 struct kvm_io_bus *bus;
3780 int dev_idx, srcu_idx;
3781 struct kvm_io_device *iodev = NULL;
3783 srcu_idx = srcu_read_lock(&kvm->srcu);
3785 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
3789 dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
3793 iodev = bus->range[dev_idx].dev;
3796 srcu_read_unlock(&kvm->srcu, srcu_idx);
3800 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
3802 static int kvm_debugfs_open(struct inode *inode, struct file *file,
3803 int (*get)(void *, u64 *), int (*set)(void *, u64),
3806 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
3809 /* The debugfs files are a reference to the kvm struct which
3810 * is still valid when kvm_destroy_vm is called.
3811 * To avoid the race between open and the removal of the debugfs
3812 * directory we test against the users count.
3814 if (!refcount_inc_not_zero(&stat_data->kvm->users_count))
3817 if (simple_attr_open(inode, file, get, set, fmt)) {
3818 kvm_put_kvm(stat_data->kvm);
3825 static int kvm_debugfs_release(struct inode *inode, struct file *file)
3827 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
3830 simple_attr_release(inode, file);
3831 kvm_put_kvm(stat_data->kvm);
3836 static int vm_stat_get_per_vm(void *data, u64 *val)
3838 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3840 *val = *(ulong *)((void *)stat_data->kvm + stat_data->offset);
3845 static int vm_stat_clear_per_vm(void *data, u64 val)
3847 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3852 *(ulong *)((void *)stat_data->kvm + stat_data->offset) = 0;
3857 static int vm_stat_get_per_vm_open(struct inode *inode, struct file *file)
3859 __simple_attr_check_format("%llu\n", 0ull);
3860 return kvm_debugfs_open(inode, file, vm_stat_get_per_vm,
3861 vm_stat_clear_per_vm, "%llu\n");
3864 static const struct file_operations vm_stat_get_per_vm_fops = {
3865 .owner = THIS_MODULE,
3866 .open = vm_stat_get_per_vm_open,
3867 .release = kvm_debugfs_release,
3868 .read = simple_attr_read,
3869 .write = simple_attr_write,
3870 .llseek = no_llseek,
3873 static int vcpu_stat_get_per_vm(void *data, u64 *val)
3876 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3877 struct kvm_vcpu *vcpu;
3881 kvm_for_each_vcpu(i, vcpu, stat_data->kvm)
3882 *val += *(u64 *)((void *)vcpu + stat_data->offset);
3887 static int vcpu_stat_clear_per_vm(void *data, u64 val)
3890 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3891 struct kvm_vcpu *vcpu;
3896 kvm_for_each_vcpu(i, vcpu, stat_data->kvm)
3897 *(u64 *)((void *)vcpu + stat_data->offset) = 0;
3902 static int vcpu_stat_get_per_vm_open(struct inode *inode, struct file *file)
3904 __simple_attr_check_format("%llu\n", 0ull);
3905 return kvm_debugfs_open(inode, file, vcpu_stat_get_per_vm,
3906 vcpu_stat_clear_per_vm, "%llu\n");
3909 static const struct file_operations vcpu_stat_get_per_vm_fops = {
3910 .owner = THIS_MODULE,
3911 .open = vcpu_stat_get_per_vm_open,
3912 .release = kvm_debugfs_release,
3913 .read = simple_attr_read,
3914 .write = simple_attr_write,
3915 .llseek = no_llseek,
3918 static const struct file_operations *stat_fops_per_vm[] = {
3919 [KVM_STAT_VCPU] = &vcpu_stat_get_per_vm_fops,
3920 [KVM_STAT_VM] = &vm_stat_get_per_vm_fops,
3923 static int vm_stat_get(void *_offset, u64 *val)
3925 unsigned offset = (long)_offset;
3927 struct kvm_stat_data stat_tmp = {.offset = offset};
3931 spin_lock(&kvm_lock);
3932 list_for_each_entry(kvm, &vm_list, vm_list) {
3934 vm_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
3937 spin_unlock(&kvm_lock);
3941 static int vm_stat_clear(void *_offset, u64 val)
3943 unsigned offset = (long)_offset;
3945 struct kvm_stat_data stat_tmp = {.offset = offset};
3950 spin_lock(&kvm_lock);
3951 list_for_each_entry(kvm, &vm_list, vm_list) {
3953 vm_stat_clear_per_vm((void *)&stat_tmp, 0);
3955 spin_unlock(&kvm_lock);
3960 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n");
3962 static int vcpu_stat_get(void *_offset, u64 *val)
3964 unsigned offset = (long)_offset;
3966 struct kvm_stat_data stat_tmp = {.offset = offset};
3970 spin_lock(&kvm_lock);
3971 list_for_each_entry(kvm, &vm_list, vm_list) {
3973 vcpu_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
3976 spin_unlock(&kvm_lock);
3980 static int vcpu_stat_clear(void *_offset, u64 val)
3982 unsigned offset = (long)_offset;
3984 struct kvm_stat_data stat_tmp = {.offset = offset};
3989 spin_lock(&kvm_lock);
3990 list_for_each_entry(kvm, &vm_list, vm_list) {
3992 vcpu_stat_clear_per_vm((void *)&stat_tmp, 0);
3994 spin_unlock(&kvm_lock);
3999 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear,
4002 static const struct file_operations *stat_fops[] = {
4003 [KVM_STAT_VCPU] = &vcpu_stat_fops,
4004 [KVM_STAT_VM] = &vm_stat_fops,
4007 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm)
4009 struct kobj_uevent_env *env;
4010 unsigned long long created, active;
4012 if (!kvm_dev.this_device || !kvm)
4015 spin_lock(&kvm_lock);
4016 if (type == KVM_EVENT_CREATE_VM) {
4017 kvm_createvm_count++;
4019 } else if (type == KVM_EVENT_DESTROY_VM) {
4022 created = kvm_createvm_count;
4023 active = kvm_active_vms;
4024 spin_unlock(&kvm_lock);
4026 env = kzalloc(sizeof(*env), GFP_KERNEL);
4030 add_uevent_var(env, "CREATED=%llu", created);
4031 add_uevent_var(env, "COUNT=%llu", active);
4033 if (type == KVM_EVENT_CREATE_VM) {
4034 add_uevent_var(env, "EVENT=create");
4035 kvm->userspace_pid = task_pid_nr(current);
4036 } else if (type == KVM_EVENT_DESTROY_VM) {
4037 add_uevent_var(env, "EVENT=destroy");
4039 add_uevent_var(env, "PID=%d", kvm->userspace_pid);
4041 if (kvm->debugfs_dentry) {
4042 char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL);
4045 tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX);
4047 add_uevent_var(env, "STATS_PATH=%s", tmp);
4051 /* no need for checks, since we are adding at most only 5 keys */
4052 env->envp[env->envp_idx++] = NULL;
4053 kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp);
4057 static void kvm_init_debug(void)
4059 struct kvm_stats_debugfs_item *p;
4061 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
4063 kvm_debugfs_num_entries = 0;
4064 for (p = debugfs_entries; p->name; ++p, kvm_debugfs_num_entries++) {
4065 debugfs_create_file(p->name, 0644, kvm_debugfs_dir,
4066 (void *)(long)p->offset,
4067 stat_fops[p->kind]);
4071 static int kvm_suspend(void)
4073 if (kvm_usage_count)
4074 hardware_disable_nolock(NULL);
4078 static void kvm_resume(void)
4080 if (kvm_usage_count) {
4081 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
4082 hardware_enable_nolock(NULL);
4086 static struct syscore_ops kvm_syscore_ops = {
4087 .suspend = kvm_suspend,
4088 .resume = kvm_resume,
4092 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
4094 return container_of(pn, struct kvm_vcpu, preempt_notifier);
4097 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
4099 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
4101 if (vcpu->preempted)
4102 vcpu->preempted = false;
4104 kvm_arch_sched_in(vcpu, cpu);
4106 kvm_arch_vcpu_load(vcpu, cpu);
4109 static void kvm_sched_out(struct preempt_notifier *pn,
4110 struct task_struct *next)
4112 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
4114 if (current->state == TASK_RUNNING)
4115 vcpu->preempted = true;
4116 kvm_arch_vcpu_put(vcpu);
4119 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
4120 struct module *module)
4125 r = kvm_arch_init(opaque);
4130 * kvm_arch_init makes sure there's at most one caller
4131 * for architectures that support multiple implementations,
4132 * like intel and amd on x86.
4133 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
4134 * conflicts in case kvm is already setup for another implementation.
4136 r = kvm_irqfd_init();
4140 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
4145 r = kvm_arch_hardware_setup();
4149 for_each_online_cpu(cpu) {
4150 smp_call_function_single(cpu,
4151 kvm_arch_check_processor_compat,
4157 r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting",
4158 kvm_starting_cpu, kvm_dying_cpu);
4161 register_reboot_notifier(&kvm_reboot_notifier);
4163 /* A kmem cache lets us meet the alignment requirements of fx_save. */
4165 vcpu_align = __alignof__(struct kvm_vcpu);
4167 kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align,
4169 offsetof(struct kvm_vcpu, arch),
4170 sizeof_field(struct kvm_vcpu, arch),
4172 if (!kvm_vcpu_cache) {
4177 r = kvm_async_pf_init();
4181 kvm_chardev_ops.owner = module;
4182 kvm_vm_fops.owner = module;
4183 kvm_vcpu_fops.owner = module;
4185 r = misc_register(&kvm_dev);
4187 pr_err("kvm: misc device register failed\n");
4191 register_syscore_ops(&kvm_syscore_ops);
4193 kvm_preempt_ops.sched_in = kvm_sched_in;
4194 kvm_preempt_ops.sched_out = kvm_sched_out;
4198 r = kvm_vfio_ops_init();
4204 kvm_async_pf_deinit();
4206 kmem_cache_destroy(kvm_vcpu_cache);
4208 unregister_reboot_notifier(&kvm_reboot_notifier);
4209 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4212 kvm_arch_hardware_unsetup();
4214 free_cpumask_var(cpus_hardware_enabled);
4222 EXPORT_SYMBOL_GPL(kvm_init);
4226 debugfs_remove_recursive(kvm_debugfs_dir);
4227 misc_deregister(&kvm_dev);
4228 kmem_cache_destroy(kvm_vcpu_cache);
4229 kvm_async_pf_deinit();
4230 unregister_syscore_ops(&kvm_syscore_ops);
4231 unregister_reboot_notifier(&kvm_reboot_notifier);
4232 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4233 on_each_cpu(hardware_disable_nolock, NULL, 1);
4234 kvm_arch_hardware_unsetup();
4237 free_cpumask_var(cpus_hardware_enabled);
4238 kvm_vfio_ops_exit();
4240 EXPORT_SYMBOL_GPL(kvm_exit);