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,
120 static int hardware_enable_all(void);
121 static void hardware_disable_all(void);
123 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
125 static void kvm_release_pfn_dirty(kvm_pfn_t pfn);
126 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot, gfn_t gfn);
128 __visible bool kvm_rebooting;
129 EXPORT_SYMBOL_GPL(kvm_rebooting);
131 static bool largepages_enabled = true;
133 bool kvm_is_reserved_pfn(kvm_pfn_t pfn)
136 return PageReserved(pfn_to_page(pfn));
142 * Switches to specified vcpu, until a matching vcpu_put()
144 int vcpu_load(struct kvm_vcpu *vcpu)
148 if (mutex_lock_killable(&vcpu->mutex))
151 preempt_notifier_register(&vcpu->preempt_notifier);
152 kvm_arch_vcpu_load(vcpu, cpu);
156 EXPORT_SYMBOL_GPL(vcpu_load);
158 void vcpu_put(struct kvm_vcpu *vcpu)
161 kvm_arch_vcpu_put(vcpu);
162 preempt_notifier_unregister(&vcpu->preempt_notifier);
164 mutex_unlock(&vcpu->mutex);
166 EXPORT_SYMBOL_GPL(vcpu_put);
168 /* TODO: merge with kvm_arch_vcpu_should_kick */
169 static bool kvm_request_needs_ipi(struct kvm_vcpu *vcpu, unsigned req)
171 int mode = kvm_vcpu_exiting_guest_mode(vcpu);
174 * We need to wait for the VCPU to reenable interrupts and get out of
175 * READING_SHADOW_PAGE_TABLES mode.
177 if (req & KVM_REQUEST_WAIT)
178 return mode != OUTSIDE_GUEST_MODE;
181 * Need to kick a running VCPU, but otherwise there is nothing to do.
183 return mode == IN_GUEST_MODE;
186 static void ack_flush(void *_completed)
190 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
195 bool wait = req & KVM_REQUEST_WAIT;
196 struct kvm_vcpu *vcpu;
198 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
201 kvm_for_each_vcpu(i, vcpu, kvm) {
202 kvm_make_request(req, vcpu);
205 if (!(req & KVM_REQUEST_NO_WAKEUP) && kvm_vcpu_wake_up(vcpu))
208 if (cpus != NULL && cpu != -1 && cpu != me &&
209 kvm_request_needs_ipi(vcpu, req))
210 cpumask_set_cpu(cpu, cpus);
212 if (unlikely(cpus == NULL))
213 smp_call_function_many(cpu_online_mask, ack_flush, NULL, wait);
214 else if (!cpumask_empty(cpus))
215 smp_call_function_many(cpus, ack_flush, NULL, wait);
219 free_cpumask_var(cpus);
223 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
224 void kvm_flush_remote_tlbs(struct kvm *kvm)
227 * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in
228 * kvm_make_all_cpus_request.
230 long dirty_count = smp_load_acquire(&kvm->tlbs_dirty);
233 * We want to publish modifications to the page tables before reading
234 * mode. Pairs with a memory barrier in arch-specific code.
235 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
236 * and smp_mb in walk_shadow_page_lockless_begin/end.
237 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
239 * There is already an smp_mb__after_atomic() before
240 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
243 if (kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
244 ++kvm->stat.remote_tlb_flush;
245 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
247 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
250 void kvm_reload_remote_mmus(struct kvm *kvm)
252 kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
255 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
260 mutex_init(&vcpu->mutex);
265 init_swait_queue_head(&vcpu->wq);
266 kvm_async_pf_vcpu_init(vcpu);
269 INIT_LIST_HEAD(&vcpu->blocked_vcpu_list);
271 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
276 vcpu->run = page_address(page);
278 kvm_vcpu_set_in_spin_loop(vcpu, false);
279 kvm_vcpu_set_dy_eligible(vcpu, false);
280 vcpu->preempted = false;
282 r = kvm_arch_vcpu_init(vcpu);
288 free_page((unsigned long)vcpu->run);
292 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
294 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
297 kvm_arch_vcpu_uninit(vcpu);
298 free_page((unsigned long)vcpu->run);
300 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
302 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
303 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
305 return container_of(mn, struct kvm, mmu_notifier);
308 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
309 struct mm_struct *mm,
310 unsigned long address)
312 struct kvm *kvm = mmu_notifier_to_kvm(mn);
313 int need_tlb_flush, idx;
316 * When ->invalidate_page runs, the linux pte has been zapped
317 * already but the page is still allocated until
318 * ->invalidate_page returns. So if we increase the sequence
319 * here the kvm page fault will notice if the spte can't be
320 * established because the page is going to be freed. If
321 * instead the kvm page fault establishes the spte before
322 * ->invalidate_page runs, kvm_unmap_hva will release it
325 * The sequence increase only need to be seen at spin_unlock
326 * time, and not at spin_lock time.
328 * Increasing the sequence after the spin_unlock would be
329 * unsafe because the kvm page fault could then establish the
330 * pte after kvm_unmap_hva returned, without noticing the page
331 * is going to be freed.
333 idx = srcu_read_lock(&kvm->srcu);
334 spin_lock(&kvm->mmu_lock);
336 kvm->mmu_notifier_seq++;
337 need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
338 /* we've to flush the tlb before the pages can be freed */
340 kvm_flush_remote_tlbs(kvm);
342 spin_unlock(&kvm->mmu_lock);
344 kvm_arch_mmu_notifier_invalidate_page(kvm, address);
346 srcu_read_unlock(&kvm->srcu, idx);
349 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
350 struct mm_struct *mm,
351 unsigned long address,
354 struct kvm *kvm = mmu_notifier_to_kvm(mn);
357 idx = srcu_read_lock(&kvm->srcu);
358 spin_lock(&kvm->mmu_lock);
359 kvm->mmu_notifier_seq++;
360 kvm_set_spte_hva(kvm, address, pte);
361 spin_unlock(&kvm->mmu_lock);
362 srcu_read_unlock(&kvm->srcu, idx);
365 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
366 struct mm_struct *mm,
370 struct kvm *kvm = mmu_notifier_to_kvm(mn);
371 int need_tlb_flush = 0, idx;
373 idx = srcu_read_lock(&kvm->srcu);
374 spin_lock(&kvm->mmu_lock);
376 * The count increase must become visible at unlock time as no
377 * spte can be established without taking the mmu_lock and
378 * count is also read inside the mmu_lock critical section.
380 kvm->mmu_notifier_count++;
381 need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
382 need_tlb_flush |= kvm->tlbs_dirty;
383 /* we've to flush the tlb before the pages can be freed */
385 kvm_flush_remote_tlbs(kvm);
387 spin_unlock(&kvm->mmu_lock);
388 srcu_read_unlock(&kvm->srcu, idx);
391 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
392 struct mm_struct *mm,
396 struct kvm *kvm = mmu_notifier_to_kvm(mn);
398 spin_lock(&kvm->mmu_lock);
400 * This sequence increase will notify the kvm page fault that
401 * the page that is going to be mapped in the spte could have
404 kvm->mmu_notifier_seq++;
407 * The above sequence increase must be visible before the
408 * below count decrease, which is ensured by the smp_wmb above
409 * in conjunction with the smp_rmb in mmu_notifier_retry().
411 kvm->mmu_notifier_count--;
412 spin_unlock(&kvm->mmu_lock);
414 BUG_ON(kvm->mmu_notifier_count < 0);
417 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
418 struct mm_struct *mm,
422 struct kvm *kvm = mmu_notifier_to_kvm(mn);
425 idx = srcu_read_lock(&kvm->srcu);
426 spin_lock(&kvm->mmu_lock);
428 young = kvm_age_hva(kvm, start, end);
430 kvm_flush_remote_tlbs(kvm);
432 spin_unlock(&kvm->mmu_lock);
433 srcu_read_unlock(&kvm->srcu, idx);
438 static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
439 struct mm_struct *mm,
443 struct kvm *kvm = mmu_notifier_to_kvm(mn);
446 idx = srcu_read_lock(&kvm->srcu);
447 spin_lock(&kvm->mmu_lock);
449 * Even though we do not flush TLB, this will still adversely
450 * affect performance on pre-Haswell Intel EPT, where there is
451 * no EPT Access Bit to clear so that we have to tear down EPT
452 * tables instead. If we find this unacceptable, we can always
453 * add a parameter to kvm_age_hva so that it effectively doesn't
454 * do anything on clear_young.
456 * Also note that currently we never issue secondary TLB flushes
457 * from clear_young, leaving this job up to the regular system
458 * cadence. If we find this inaccurate, we might come up with a
459 * more sophisticated heuristic later.
461 young = kvm_age_hva(kvm, start, end);
462 spin_unlock(&kvm->mmu_lock);
463 srcu_read_unlock(&kvm->srcu, idx);
468 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
469 struct mm_struct *mm,
470 unsigned long address)
472 struct kvm *kvm = mmu_notifier_to_kvm(mn);
475 idx = srcu_read_lock(&kvm->srcu);
476 spin_lock(&kvm->mmu_lock);
477 young = kvm_test_age_hva(kvm, address);
478 spin_unlock(&kvm->mmu_lock);
479 srcu_read_unlock(&kvm->srcu, idx);
484 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
485 struct mm_struct *mm)
487 struct kvm *kvm = mmu_notifier_to_kvm(mn);
490 idx = srcu_read_lock(&kvm->srcu);
491 kvm_arch_flush_shadow_all(kvm);
492 srcu_read_unlock(&kvm->srcu, idx);
495 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
496 .invalidate_page = kvm_mmu_notifier_invalidate_page,
497 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
498 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
499 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
500 .clear_young = kvm_mmu_notifier_clear_young,
501 .test_young = kvm_mmu_notifier_test_young,
502 .change_pte = kvm_mmu_notifier_change_pte,
503 .release = kvm_mmu_notifier_release,
506 static int kvm_init_mmu_notifier(struct kvm *kvm)
508 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
509 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
512 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
514 static int kvm_init_mmu_notifier(struct kvm *kvm)
519 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
521 static struct kvm_memslots *kvm_alloc_memslots(void)
524 struct kvm_memslots *slots;
526 slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
530 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
531 slots->id_to_index[i] = slots->memslots[i].id = i;
536 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
538 if (!memslot->dirty_bitmap)
541 kvfree(memslot->dirty_bitmap);
542 memslot->dirty_bitmap = NULL;
546 * Free any memory in @free but not in @dont.
548 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
549 struct kvm_memory_slot *dont)
551 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
552 kvm_destroy_dirty_bitmap(free);
554 kvm_arch_free_memslot(kvm, free, dont);
559 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
561 struct kvm_memory_slot *memslot;
566 kvm_for_each_memslot(memslot, slots)
567 kvm_free_memslot(kvm, memslot, NULL);
572 static void kvm_destroy_vm_debugfs(struct kvm *kvm)
576 if (!kvm->debugfs_dentry)
579 debugfs_remove_recursive(kvm->debugfs_dentry);
581 if (kvm->debugfs_stat_data) {
582 for (i = 0; i < kvm_debugfs_num_entries; i++)
583 kfree(kvm->debugfs_stat_data[i]);
584 kfree(kvm->debugfs_stat_data);
588 static int kvm_create_vm_debugfs(struct kvm *kvm, int fd)
590 char dir_name[ITOA_MAX_LEN * 2];
591 struct kvm_stat_data *stat_data;
592 struct kvm_stats_debugfs_item *p;
594 if (!debugfs_initialized())
597 snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd);
598 kvm->debugfs_dentry = debugfs_create_dir(dir_name,
600 if (!kvm->debugfs_dentry)
603 kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries,
604 sizeof(*kvm->debugfs_stat_data),
606 if (!kvm->debugfs_stat_data)
609 for (p = debugfs_entries; p->name; p++) {
610 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL);
614 stat_data->kvm = kvm;
615 stat_data->offset = p->offset;
616 kvm->debugfs_stat_data[p - debugfs_entries] = stat_data;
617 if (!debugfs_create_file(p->name, 0644,
620 stat_fops_per_vm[p->kind]))
626 static struct kvm *kvm_create_vm(unsigned long type)
629 struct kvm *kvm = kvm_arch_alloc_vm();
632 return ERR_PTR(-ENOMEM);
634 spin_lock_init(&kvm->mmu_lock);
636 kvm->mm = current->mm;
637 kvm_eventfd_init(kvm);
638 mutex_init(&kvm->lock);
639 mutex_init(&kvm->irq_lock);
640 mutex_init(&kvm->slots_lock);
641 refcount_set(&kvm->users_count, 1);
642 INIT_LIST_HEAD(&kvm->devices);
644 r = kvm_arch_init_vm(kvm, type);
646 goto out_err_no_disable;
648 r = hardware_enable_all();
650 goto out_err_no_disable;
652 #ifdef CONFIG_HAVE_KVM_IRQFD
653 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
656 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
659 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
660 struct kvm_memslots *slots = kvm_alloc_memslots();
662 goto out_err_no_srcu;
664 * Generations must be different for each address space.
665 * Init kvm generation close to the maximum to easily test the
666 * code of handling generation number wrap-around.
668 slots->generation = i * 2 - 150;
669 rcu_assign_pointer(kvm->memslots[i], slots);
672 if (init_srcu_struct(&kvm->srcu))
673 goto out_err_no_srcu;
674 if (init_srcu_struct(&kvm->irq_srcu))
675 goto out_err_no_irq_srcu;
676 for (i = 0; i < KVM_NR_BUSES; i++) {
677 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
683 r = kvm_init_mmu_notifier(kvm);
687 spin_lock(&kvm_lock);
688 list_add(&kvm->vm_list, &vm_list);
689 spin_unlock(&kvm_lock);
691 preempt_notifier_inc();
696 cleanup_srcu_struct(&kvm->irq_srcu);
698 cleanup_srcu_struct(&kvm->srcu);
700 hardware_disable_all();
702 for (i = 0; i < KVM_NR_BUSES; i++)
703 kfree(kvm->buses[i]);
704 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
705 kvm_free_memslots(kvm, kvm->memslots[i]);
706 kvm_arch_free_vm(kvm);
711 static void kvm_destroy_devices(struct kvm *kvm)
713 struct kvm_device *dev, *tmp;
716 * We do not need to take the kvm->lock here, because nobody else
717 * has a reference to the struct kvm at this point and therefore
718 * cannot access the devices list anyhow.
720 list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
721 list_del(&dev->vm_node);
722 dev->ops->destroy(dev);
726 static void kvm_destroy_vm(struct kvm *kvm)
729 struct mm_struct *mm = kvm->mm;
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++) {
739 kvm_io_bus_destroy(kvm->buses[i]);
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[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)
809 int i = slots->id_to_index[id];
810 struct kvm_memory_slot *mslots = slots->memslots;
812 WARN_ON(mslots[i].id != id);
814 WARN_ON(!mslots[i].npages);
815 if (mslots[i].npages)
818 if (!mslots[i].npages)
822 while (i < KVM_MEM_SLOTS_NUM - 1 &&
823 new->base_gfn <= mslots[i + 1].base_gfn) {
824 if (!mslots[i + 1].npages)
826 mslots[i] = mslots[i + 1];
827 slots->id_to_index[mslots[i].id] = i;
832 * The ">=" is needed when creating a slot with base_gfn == 0,
833 * so that it moves before all those with base_gfn == npages == 0.
835 * On the other hand, if new->npages is zero, the above loop has
836 * already left i pointing to the beginning of the empty part of
837 * mslots, and the ">=" would move the hole backwards in this
838 * case---which is wrong. So skip the loop when deleting a slot.
842 new->base_gfn >= mslots[i - 1].base_gfn) {
843 mslots[i] = mslots[i - 1];
844 slots->id_to_index[mslots[i].id] = i;
848 WARN_ON_ONCE(i != slots->used_slots);
851 slots->id_to_index[mslots[i].id] = i;
854 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
856 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
858 #ifdef __KVM_HAVE_READONLY_MEM
859 valid_flags |= KVM_MEM_READONLY;
862 if (mem->flags & ~valid_flags)
868 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
869 int as_id, struct kvm_memslots *slots)
871 struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id);
874 * Set the low bit in the generation, which disables SPTE caching
875 * until the end of synchronize_srcu_expedited.
877 WARN_ON(old_memslots->generation & 1);
878 slots->generation = old_memslots->generation + 1;
880 rcu_assign_pointer(kvm->memslots[as_id], slots);
881 synchronize_srcu_expedited(&kvm->srcu);
884 * Increment the new memslot generation a second time. This prevents
885 * vm exits that race with memslot updates from caching a memslot
886 * generation that will (potentially) be valid forever.
888 * Generations must be unique even across address spaces. We do not need
889 * a global counter for that, instead the generation space is evenly split
890 * across address spaces. For example, with two address spaces, address
891 * space 0 will use generations 0, 4, 8, ... while * address space 1 will
892 * use generations 2, 6, 10, 14, ...
894 slots->generation += KVM_ADDRESS_SPACE_NUM * 2 - 1;
896 kvm_arch_memslots_updated(kvm, slots);
902 * Allocate some memory and give it an address in the guest physical address
905 * Discontiguous memory is allowed, mostly for framebuffers.
907 * Must be called holding kvm->slots_lock for write.
909 int __kvm_set_memory_region(struct kvm *kvm,
910 const struct kvm_userspace_memory_region *mem)
914 unsigned long npages;
915 struct kvm_memory_slot *slot;
916 struct kvm_memory_slot old, new;
917 struct kvm_memslots *slots = NULL, *old_memslots;
919 enum kvm_mr_change change;
921 r = check_memory_region_flags(mem);
926 as_id = mem->slot >> 16;
929 /* General sanity checks */
930 if (mem->memory_size & (PAGE_SIZE - 1))
932 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
934 /* We can read the guest memory with __xxx_user() later on. */
935 if ((id < KVM_USER_MEM_SLOTS) &&
936 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
937 !access_ok(VERIFY_WRITE,
938 (void __user *)(unsigned long)mem->userspace_addr,
941 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
943 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
946 slot = id_to_memslot(__kvm_memslots(kvm, as_id), id);
947 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
948 npages = mem->memory_size >> PAGE_SHIFT;
950 if (npages > KVM_MEM_MAX_NR_PAGES)
956 new.base_gfn = base_gfn;
958 new.flags = mem->flags;
962 change = KVM_MR_CREATE;
963 else { /* Modify an existing slot. */
964 if ((mem->userspace_addr != old.userspace_addr) ||
965 (npages != old.npages) ||
966 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
969 if (base_gfn != old.base_gfn)
970 change = KVM_MR_MOVE;
971 else if (new.flags != old.flags)
972 change = KVM_MR_FLAGS_ONLY;
973 else { /* Nothing to change. */
982 change = KVM_MR_DELETE;
987 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
988 /* Check for overlaps */
990 kvm_for_each_memslot(slot, __kvm_memslots(kvm, as_id)) {
991 if ((slot->id >= KVM_USER_MEM_SLOTS) ||
994 if (!((base_gfn + npages <= slot->base_gfn) ||
995 (base_gfn >= slot->base_gfn + slot->npages)))
1000 /* Free page dirty bitmap if unneeded */
1001 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
1002 new.dirty_bitmap = NULL;
1005 if (change == KVM_MR_CREATE) {
1006 new.userspace_addr = mem->userspace_addr;
1008 if (kvm_arch_create_memslot(kvm, &new, npages))
1012 /* Allocate page dirty bitmap if needed */
1013 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
1014 if (kvm_create_dirty_bitmap(&new) < 0)
1018 slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
1021 memcpy(slots, __kvm_memslots(kvm, as_id), sizeof(struct kvm_memslots));
1023 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
1024 slot = id_to_memslot(slots, id);
1025 slot->flags |= KVM_MEMSLOT_INVALID;
1027 old_memslots = install_new_memslots(kvm, as_id, slots);
1029 /* From this point no new shadow pages pointing to a deleted,
1030 * or moved, memslot will be created.
1032 * validation of sp->gfn happens in:
1033 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1034 * - kvm_is_visible_gfn (mmu_check_roots)
1036 kvm_arch_flush_shadow_memslot(kvm, slot);
1039 * We can re-use the old_memslots from above, the only difference
1040 * from the currently installed memslots is the invalid flag. This
1041 * will get overwritten by update_memslots anyway.
1043 slots = old_memslots;
1046 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
1050 /* actual memory is freed via old in kvm_free_memslot below */
1051 if (change == KVM_MR_DELETE) {
1052 new.dirty_bitmap = NULL;
1053 memset(&new.arch, 0, sizeof(new.arch));
1056 update_memslots(slots, &new);
1057 old_memslots = install_new_memslots(kvm, as_id, slots);
1059 kvm_arch_commit_memory_region(kvm, mem, &old, &new, change);
1061 kvm_free_memslot(kvm, &old, &new);
1062 kvfree(old_memslots);
1068 kvm_free_memslot(kvm, &new, &old);
1072 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1074 int kvm_set_memory_region(struct kvm *kvm,
1075 const struct kvm_userspace_memory_region *mem)
1079 mutex_lock(&kvm->slots_lock);
1080 r = __kvm_set_memory_region(kvm, mem);
1081 mutex_unlock(&kvm->slots_lock);
1084 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1086 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1087 struct kvm_userspace_memory_region *mem)
1089 if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1092 return kvm_set_memory_region(kvm, mem);
1095 int kvm_get_dirty_log(struct kvm *kvm,
1096 struct kvm_dirty_log *log, int *is_dirty)
1098 struct kvm_memslots *slots;
1099 struct kvm_memory_slot *memslot;
1102 unsigned long any = 0;
1104 as_id = log->slot >> 16;
1105 id = (u16)log->slot;
1106 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1109 slots = __kvm_memslots(kvm, as_id);
1110 memslot = id_to_memslot(slots, id);
1111 if (!memslot->dirty_bitmap)
1114 n = kvm_dirty_bitmap_bytes(memslot);
1116 for (i = 0; !any && i < n/sizeof(long); ++i)
1117 any = memslot->dirty_bitmap[i];
1119 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1126 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1128 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1130 * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages
1131 * are dirty write protect them for next write.
1132 * @kvm: pointer to kvm instance
1133 * @log: slot id and address to which we copy the log
1134 * @is_dirty: flag set if any page is dirty
1136 * We need to keep it in mind that VCPU threads can write to the bitmap
1137 * concurrently. So, to avoid losing track of dirty pages we keep the
1140 * 1. Take a snapshot of the bit and clear it if needed.
1141 * 2. Write protect the corresponding page.
1142 * 3. Copy the snapshot to the userspace.
1143 * 4. Upon return caller flushes TLB's if needed.
1145 * Between 2 and 4, the guest may write to the page using the remaining TLB
1146 * entry. This is not a problem because the page is reported dirty using
1147 * the snapshot taken before and step 4 ensures that writes done after
1148 * exiting to userspace will be logged for the next call.
1151 int kvm_get_dirty_log_protect(struct kvm *kvm,
1152 struct kvm_dirty_log *log, bool *is_dirty)
1154 struct kvm_memslots *slots;
1155 struct kvm_memory_slot *memslot;
1158 unsigned long *dirty_bitmap;
1159 unsigned long *dirty_bitmap_buffer;
1161 as_id = log->slot >> 16;
1162 id = (u16)log->slot;
1163 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1166 slots = __kvm_memslots(kvm, as_id);
1167 memslot = id_to_memslot(slots, id);
1169 dirty_bitmap = memslot->dirty_bitmap;
1173 n = kvm_dirty_bitmap_bytes(memslot);
1175 dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long);
1176 memset(dirty_bitmap_buffer, 0, n);
1178 spin_lock(&kvm->mmu_lock);
1180 for (i = 0; i < n / sizeof(long); i++) {
1184 if (!dirty_bitmap[i])
1189 mask = xchg(&dirty_bitmap[i], 0);
1190 dirty_bitmap_buffer[i] = mask;
1193 offset = i * BITS_PER_LONG;
1194 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1199 spin_unlock(&kvm->mmu_lock);
1200 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1204 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect);
1207 bool kvm_largepages_enabled(void)
1209 return largepages_enabled;
1212 void kvm_disable_largepages(void)
1214 largepages_enabled = false;
1216 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1218 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1220 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1222 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1224 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
1226 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn);
1229 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1231 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1233 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1234 memslot->flags & KVM_MEMSLOT_INVALID)
1239 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1241 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1243 struct vm_area_struct *vma;
1244 unsigned long addr, size;
1248 addr = gfn_to_hva(kvm, gfn);
1249 if (kvm_is_error_hva(addr))
1252 down_read(¤t->mm->mmap_sem);
1253 vma = find_vma(current->mm, addr);
1257 size = vma_kernel_pagesize(vma);
1260 up_read(¤t->mm->mmap_sem);
1265 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1267 return slot->flags & KVM_MEM_READONLY;
1270 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1271 gfn_t *nr_pages, bool write)
1273 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1274 return KVM_HVA_ERR_BAD;
1276 if (memslot_is_readonly(slot) && write)
1277 return KVM_HVA_ERR_RO_BAD;
1280 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1282 return __gfn_to_hva_memslot(slot, gfn);
1285 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1288 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1291 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1294 return gfn_to_hva_many(slot, gfn, NULL);
1296 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1298 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1300 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1302 EXPORT_SYMBOL_GPL(gfn_to_hva);
1304 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
1306 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
1308 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
1311 * If writable is set to false, the hva returned by this function is only
1312 * allowed to be read.
1314 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1315 gfn_t gfn, bool *writable)
1317 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1319 if (!kvm_is_error_hva(hva) && writable)
1320 *writable = !memslot_is_readonly(slot);
1325 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1327 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1329 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1332 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
1334 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1336 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1339 static int get_user_page_nowait(unsigned long start, int write,
1342 int flags = FOLL_NOWAIT | FOLL_HWPOISON;
1345 flags |= FOLL_WRITE;
1347 return get_user_pages(start, 1, flags, page, NULL);
1350 static inline int check_user_page_hwpoison(unsigned long addr)
1352 int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
1354 rc = get_user_pages(addr, 1, flags, NULL, NULL);
1355 return rc == -EHWPOISON;
1359 * The atomic path to get the writable pfn which will be stored in @pfn,
1360 * true indicates success, otherwise false is returned.
1362 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1363 bool write_fault, bool *writable, kvm_pfn_t *pfn)
1365 struct page *page[1];
1368 if (!(async || atomic))
1372 * Fast pin a writable pfn only if it is a write fault request
1373 * or the caller allows to map a writable pfn for a read fault
1376 if (!(write_fault || writable))
1379 npages = __get_user_pages_fast(addr, 1, 1, page);
1381 *pfn = page_to_pfn(page[0]);
1392 * The slow path to get the pfn of the specified host virtual address,
1393 * 1 indicates success, -errno is returned if error is detected.
1395 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1396 bool *writable, kvm_pfn_t *pfn)
1398 struct page *page[1];
1404 *writable = write_fault;
1407 down_read(¤t->mm->mmap_sem);
1408 npages = get_user_page_nowait(addr, write_fault, page);
1409 up_read(¤t->mm->mmap_sem);
1411 unsigned int flags = FOLL_HWPOISON;
1414 flags |= FOLL_WRITE;
1416 npages = get_user_pages_unlocked(addr, 1, page, flags);
1421 /* map read fault as writable if possible */
1422 if (unlikely(!write_fault) && writable) {
1423 struct page *wpage[1];
1425 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1434 *pfn = page_to_pfn(page[0]);
1438 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1440 if (unlikely(!(vma->vm_flags & VM_READ)))
1443 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1449 static int hva_to_pfn_remapped(struct vm_area_struct *vma,
1450 unsigned long addr, bool *async,
1451 bool write_fault, kvm_pfn_t *p_pfn)
1456 r = follow_pfn(vma, addr, &pfn);
1459 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
1460 * not call the fault handler, so do it here.
1462 bool unlocked = false;
1463 r = fixup_user_fault(current, current->mm, addr,
1464 (write_fault ? FAULT_FLAG_WRITE : 0),
1471 r = follow_pfn(vma, addr, &pfn);
1479 * Get a reference here because callers of *hva_to_pfn* and
1480 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
1481 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
1482 * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will
1483 * simply do nothing for reserved pfns.
1485 * Whoever called remap_pfn_range is also going to call e.g.
1486 * unmap_mapping_range before the underlying pages are freed,
1487 * causing a call to our MMU notifier.
1496 * Pin guest page in memory and return its pfn.
1497 * @addr: host virtual address which maps memory to the guest
1498 * @atomic: whether this function can sleep
1499 * @async: whether this function need to wait IO complete if the
1500 * host page is not in the memory
1501 * @write_fault: whether we should get a writable host page
1502 * @writable: whether it allows to map a writable host page for !@write_fault
1504 * The function will map a writable host page for these two cases:
1505 * 1): @write_fault = true
1506 * 2): @write_fault = false && @writable, @writable will tell the caller
1507 * whether the mapping is writable.
1509 static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1510 bool write_fault, bool *writable)
1512 struct vm_area_struct *vma;
1516 /* we can do it either atomically or asynchronously, not both */
1517 BUG_ON(atomic && async);
1519 if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1523 return KVM_PFN_ERR_FAULT;
1525 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1529 down_read(¤t->mm->mmap_sem);
1530 if (npages == -EHWPOISON ||
1531 (!async && check_user_page_hwpoison(addr))) {
1532 pfn = KVM_PFN_ERR_HWPOISON;
1537 vma = find_vma_intersection(current->mm, addr, addr + 1);
1540 pfn = KVM_PFN_ERR_FAULT;
1541 else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
1542 r = hva_to_pfn_remapped(vma, addr, async, write_fault, &pfn);
1546 pfn = KVM_PFN_ERR_FAULT;
1548 if (async && vma_is_valid(vma, write_fault))
1550 pfn = KVM_PFN_ERR_FAULT;
1553 up_read(¤t->mm->mmap_sem);
1557 kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn,
1558 bool atomic, bool *async, bool write_fault,
1561 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1563 if (addr == KVM_HVA_ERR_RO_BAD) {
1566 return KVM_PFN_ERR_RO_FAULT;
1569 if (kvm_is_error_hva(addr)) {
1572 return KVM_PFN_NOSLOT;
1575 /* Do not map writable pfn in the readonly memslot. */
1576 if (writable && memslot_is_readonly(slot)) {
1581 return hva_to_pfn(addr, atomic, async, write_fault,
1584 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
1586 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1589 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
1590 write_fault, writable);
1592 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1594 kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1596 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1598 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
1600 kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1602 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1604 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1606 kvm_pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1608 return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm, gfn), gfn);
1610 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1612 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
1614 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1616 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
1618 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1620 return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
1622 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1624 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
1626 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1628 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
1630 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1631 struct page **pages, int nr_pages)
1636 addr = gfn_to_hva_many(slot, gfn, &entry);
1637 if (kvm_is_error_hva(addr))
1640 if (entry < nr_pages)
1643 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1645 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1647 static struct page *kvm_pfn_to_page(kvm_pfn_t pfn)
1649 if (is_error_noslot_pfn(pfn))
1650 return KVM_ERR_PTR_BAD_PAGE;
1652 if (kvm_is_reserved_pfn(pfn)) {
1654 return KVM_ERR_PTR_BAD_PAGE;
1657 return pfn_to_page(pfn);
1660 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1664 pfn = gfn_to_pfn(kvm, gfn);
1666 return kvm_pfn_to_page(pfn);
1668 EXPORT_SYMBOL_GPL(gfn_to_page);
1670 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
1674 pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
1676 return kvm_pfn_to_page(pfn);
1678 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
1680 void kvm_release_page_clean(struct page *page)
1682 WARN_ON(is_error_page(page));
1684 kvm_release_pfn_clean(page_to_pfn(page));
1686 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1688 void kvm_release_pfn_clean(kvm_pfn_t pfn)
1690 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
1691 put_page(pfn_to_page(pfn));
1693 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1695 void kvm_release_page_dirty(struct page *page)
1697 WARN_ON(is_error_page(page));
1699 kvm_release_pfn_dirty(page_to_pfn(page));
1701 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1703 static void kvm_release_pfn_dirty(kvm_pfn_t pfn)
1705 kvm_set_pfn_dirty(pfn);
1706 kvm_release_pfn_clean(pfn);
1709 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
1711 if (!kvm_is_reserved_pfn(pfn)) {
1712 struct page *page = pfn_to_page(pfn);
1714 if (!PageReserved(page))
1718 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1720 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
1722 if (!kvm_is_reserved_pfn(pfn))
1723 mark_page_accessed(pfn_to_page(pfn));
1725 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1727 void kvm_get_pfn(kvm_pfn_t pfn)
1729 if (!kvm_is_reserved_pfn(pfn))
1730 get_page(pfn_to_page(pfn));
1732 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1734 static int next_segment(unsigned long len, int offset)
1736 if (len > PAGE_SIZE - offset)
1737 return PAGE_SIZE - offset;
1742 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
1743 void *data, int offset, int len)
1748 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1749 if (kvm_is_error_hva(addr))
1751 r = __copy_from_user(data, (void __user *)addr + offset, len);
1757 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1760 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1762 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1764 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1766 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
1767 int offset, int len)
1769 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1771 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1773 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
1775 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1777 gfn_t gfn = gpa >> PAGE_SHIFT;
1779 int offset = offset_in_page(gpa);
1782 while ((seg = next_segment(len, offset)) != 0) {
1783 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1793 EXPORT_SYMBOL_GPL(kvm_read_guest);
1795 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
1797 gfn_t gfn = gpa >> PAGE_SHIFT;
1799 int offset = offset_in_page(gpa);
1802 while ((seg = next_segment(len, offset)) != 0) {
1803 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
1813 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
1815 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1816 void *data, int offset, unsigned long len)
1821 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1822 if (kvm_is_error_hva(addr))
1824 pagefault_disable();
1825 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1832 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1835 gfn_t gfn = gpa >> PAGE_SHIFT;
1836 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1837 int offset = offset_in_page(gpa);
1839 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1841 EXPORT_SYMBOL_GPL(kvm_read_guest_atomic);
1843 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
1844 void *data, unsigned long len)
1846 gfn_t gfn = gpa >> PAGE_SHIFT;
1847 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1848 int offset = offset_in_page(gpa);
1850 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1852 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
1854 static int __kvm_write_guest_page(struct kvm_memory_slot *memslot, gfn_t gfn,
1855 const void *data, int offset, int len)
1860 addr = gfn_to_hva_memslot(memslot, gfn);
1861 if (kvm_is_error_hva(addr))
1863 r = __copy_to_user((void __user *)addr + offset, data, len);
1866 mark_page_dirty_in_slot(memslot, gfn);
1870 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
1871 const void *data, int offset, int len)
1873 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1875 return __kvm_write_guest_page(slot, gfn, data, offset, len);
1877 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1879 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
1880 const void *data, int offset, int len)
1882 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1884 return __kvm_write_guest_page(slot, gfn, data, offset, len);
1886 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
1888 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1891 gfn_t gfn = gpa >> PAGE_SHIFT;
1893 int offset = offset_in_page(gpa);
1896 while ((seg = next_segment(len, offset)) != 0) {
1897 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1907 EXPORT_SYMBOL_GPL(kvm_write_guest);
1909 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
1912 gfn_t gfn = gpa >> PAGE_SHIFT;
1914 int offset = offset_in_page(gpa);
1917 while ((seg = next_segment(len, offset)) != 0) {
1918 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
1928 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
1930 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
1931 struct gfn_to_hva_cache *ghc,
1932 gpa_t gpa, unsigned long len)
1934 int offset = offset_in_page(gpa);
1935 gfn_t start_gfn = gpa >> PAGE_SHIFT;
1936 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1937 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1938 gfn_t nr_pages_avail;
1941 ghc->generation = slots->generation;
1943 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
1944 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, NULL);
1945 if (!kvm_is_error_hva(ghc->hva) && nr_pages_needed <= 1) {
1949 * If the requested region crosses two memslots, we still
1950 * verify that the entire region is valid here.
1952 while (start_gfn <= end_gfn) {
1953 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
1954 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1956 if (kvm_is_error_hva(ghc->hva))
1958 start_gfn += nr_pages_avail;
1960 /* Use the slow path for cross page reads and writes. */
1961 ghc->memslot = NULL;
1966 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1967 gpa_t gpa, unsigned long len)
1969 struct kvm_memslots *slots = kvm_memslots(kvm);
1970 return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len);
1972 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1974 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1975 void *data, int offset, unsigned long len)
1977 struct kvm_memslots *slots = kvm_memslots(kvm);
1979 gpa_t gpa = ghc->gpa + offset;
1981 BUG_ON(len + offset > ghc->len);
1983 if (slots->generation != ghc->generation)
1984 __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len);
1986 if (unlikely(!ghc->memslot))
1987 return kvm_write_guest(kvm, gpa, data, len);
1989 if (kvm_is_error_hva(ghc->hva))
1992 r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
1995 mark_page_dirty_in_slot(ghc->memslot, gpa >> PAGE_SHIFT);
1999 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached);
2001 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2002 void *data, unsigned long len)
2004 return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len);
2006 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
2008 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2009 void *data, unsigned long len)
2011 struct kvm_memslots *slots = kvm_memslots(kvm);
2014 BUG_ON(len > ghc->len);
2016 if (slots->generation != ghc->generation)
2017 __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len);
2019 if (unlikely(!ghc->memslot))
2020 return kvm_read_guest(kvm, ghc->gpa, data, len);
2022 if (kvm_is_error_hva(ghc->hva))
2025 r = __copy_from_user(data, (void __user *)ghc->hva, len);
2031 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
2033 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
2035 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
2037 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
2039 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
2041 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
2043 gfn_t gfn = gpa >> PAGE_SHIFT;
2045 int offset = offset_in_page(gpa);
2048 while ((seg = next_segment(len, offset)) != 0) {
2049 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
2058 EXPORT_SYMBOL_GPL(kvm_clear_guest);
2060 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot,
2063 if (memslot && memslot->dirty_bitmap) {
2064 unsigned long rel_gfn = gfn - memslot->base_gfn;
2066 set_bit_le(rel_gfn, memslot->dirty_bitmap);
2070 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
2072 struct kvm_memory_slot *memslot;
2074 memslot = gfn_to_memslot(kvm, gfn);
2075 mark_page_dirty_in_slot(memslot, gfn);
2077 EXPORT_SYMBOL_GPL(mark_page_dirty);
2079 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
2081 struct kvm_memory_slot *memslot;
2083 memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2084 mark_page_dirty_in_slot(memslot, gfn);
2086 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
2088 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
2090 unsigned int old, val, grow;
2092 old = val = vcpu->halt_poll_ns;
2093 grow = READ_ONCE(halt_poll_ns_grow);
2095 if (val == 0 && grow)
2100 if (val > halt_poll_ns)
2103 vcpu->halt_poll_ns = val;
2104 trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
2107 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
2109 unsigned int old, val, shrink;
2111 old = val = vcpu->halt_poll_ns;
2112 shrink = READ_ONCE(halt_poll_ns_shrink);
2118 vcpu->halt_poll_ns = val;
2119 trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
2122 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
2124 if (kvm_arch_vcpu_runnable(vcpu)) {
2125 kvm_make_request(KVM_REQ_UNHALT, vcpu);
2128 if (kvm_cpu_has_pending_timer(vcpu))
2130 if (signal_pending(current))
2137 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2139 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
2142 DECLARE_SWAITQUEUE(wait);
2143 bool waited = false;
2146 start = cur = ktime_get();
2147 if (vcpu->halt_poll_ns) {
2148 ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns);
2150 ++vcpu->stat.halt_attempted_poll;
2153 * This sets KVM_REQ_UNHALT if an interrupt
2156 if (kvm_vcpu_check_block(vcpu) < 0) {
2157 ++vcpu->stat.halt_successful_poll;
2158 if (!vcpu_valid_wakeup(vcpu))
2159 ++vcpu->stat.halt_poll_invalid;
2163 } while (single_task_running() && ktime_before(cur, stop));
2166 kvm_arch_vcpu_blocking(vcpu);
2169 prepare_to_swait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
2171 if (kvm_vcpu_check_block(vcpu) < 0)
2178 finish_swait(&vcpu->wq, &wait);
2181 kvm_arch_vcpu_unblocking(vcpu);
2183 block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
2185 if (!vcpu_valid_wakeup(vcpu))
2186 shrink_halt_poll_ns(vcpu);
2187 else if (halt_poll_ns) {
2188 if (block_ns <= vcpu->halt_poll_ns)
2190 /* we had a long block, shrink polling */
2191 else if (vcpu->halt_poll_ns && block_ns > halt_poll_ns)
2192 shrink_halt_poll_ns(vcpu);
2193 /* we had a short halt and our poll time is too small */
2194 else if (vcpu->halt_poll_ns < halt_poll_ns &&
2195 block_ns < halt_poll_ns)
2196 grow_halt_poll_ns(vcpu);
2198 vcpu->halt_poll_ns = 0;
2200 trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu));
2201 kvm_arch_vcpu_block_finish(vcpu);
2203 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
2205 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
2207 struct swait_queue_head *wqp;
2209 wqp = kvm_arch_vcpu_wq(vcpu);
2210 if (swait_active(wqp)) {
2212 ++vcpu->stat.halt_wakeup;
2218 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
2222 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2224 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
2227 int cpu = vcpu->cpu;
2229 if (kvm_vcpu_wake_up(vcpu))
2233 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
2234 if (kvm_arch_vcpu_should_kick(vcpu))
2235 smp_send_reschedule(cpu);
2238 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
2239 #endif /* !CONFIG_S390 */
2241 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
2244 struct task_struct *task = NULL;
2248 pid = rcu_dereference(target->pid);
2250 task = get_pid_task(pid, PIDTYPE_PID);
2254 ret = yield_to(task, 1);
2255 put_task_struct(task);
2259 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
2262 * Helper that checks whether a VCPU is eligible for directed yield.
2263 * Most eligible candidate to yield is decided by following heuristics:
2265 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2266 * (preempted lock holder), indicated by @in_spin_loop.
2267 * Set at the beiginning and cleared at the end of interception/PLE handler.
2269 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2270 * chance last time (mostly it has become eligible now since we have probably
2271 * yielded to lockholder in last iteration. This is done by toggling
2272 * @dy_eligible each time a VCPU checked for eligibility.)
2274 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2275 * to preempted lock-holder could result in wrong VCPU selection and CPU
2276 * burning. Giving priority for a potential lock-holder increases lock
2279 * Since algorithm is based on heuristics, accessing another VCPU data without
2280 * locking does not harm. It may result in trying to yield to same VCPU, fail
2281 * and continue with next VCPU and so on.
2283 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
2285 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2288 eligible = !vcpu->spin_loop.in_spin_loop ||
2289 vcpu->spin_loop.dy_eligible;
2291 if (vcpu->spin_loop.in_spin_loop)
2292 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
2300 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
2302 struct kvm *kvm = me->kvm;
2303 struct kvm_vcpu *vcpu;
2304 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
2310 kvm_vcpu_set_in_spin_loop(me, true);
2312 * We boost the priority of a VCPU that is runnable but not
2313 * currently running, because it got preempted by something
2314 * else and called schedule in __vcpu_run. Hopefully that
2315 * VCPU is holding the lock that we need and will release it.
2316 * We approximate round-robin by starting at the last boosted VCPU.
2318 for (pass = 0; pass < 2 && !yielded && try; pass++) {
2319 kvm_for_each_vcpu(i, vcpu, kvm) {
2320 if (!pass && i <= last_boosted_vcpu) {
2321 i = last_boosted_vcpu;
2323 } else if (pass && i > last_boosted_vcpu)
2325 if (!ACCESS_ONCE(vcpu->preempted))
2329 if (swait_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
2331 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
2334 yielded = kvm_vcpu_yield_to(vcpu);
2336 kvm->last_boosted_vcpu = i;
2338 } else if (yielded < 0) {
2345 kvm_vcpu_set_in_spin_loop(me, false);
2347 /* Ensure vcpu is not eligible during next spinloop */
2348 kvm_vcpu_set_dy_eligible(me, false);
2350 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
2352 static int kvm_vcpu_fault(struct vm_fault *vmf)
2354 struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data;
2357 if (vmf->pgoff == 0)
2358 page = virt_to_page(vcpu->run);
2360 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
2361 page = virt_to_page(vcpu->arch.pio_data);
2363 #ifdef CONFIG_KVM_MMIO
2364 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
2365 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
2368 return kvm_arch_vcpu_fault(vcpu, vmf);
2374 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
2375 .fault = kvm_vcpu_fault,
2378 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2380 vma->vm_ops = &kvm_vcpu_vm_ops;
2384 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2386 struct kvm_vcpu *vcpu = filp->private_data;
2388 debugfs_remove_recursive(vcpu->debugfs_dentry);
2389 kvm_put_kvm(vcpu->kvm);
2393 static struct file_operations kvm_vcpu_fops = {
2394 .release = kvm_vcpu_release,
2395 .unlocked_ioctl = kvm_vcpu_ioctl,
2396 #ifdef CONFIG_KVM_COMPAT
2397 .compat_ioctl = kvm_vcpu_compat_ioctl,
2399 .mmap = kvm_vcpu_mmap,
2400 .llseek = noop_llseek,
2404 * Allocates an inode for the vcpu.
2406 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2408 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
2411 static int kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
2413 char dir_name[ITOA_MAX_LEN * 2];
2416 if (!kvm_arch_has_vcpu_debugfs())
2419 if (!debugfs_initialized())
2422 snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
2423 vcpu->debugfs_dentry = debugfs_create_dir(dir_name,
2424 vcpu->kvm->debugfs_dentry);
2425 if (!vcpu->debugfs_dentry)
2428 ret = kvm_arch_create_vcpu_debugfs(vcpu);
2430 debugfs_remove_recursive(vcpu->debugfs_dentry);
2438 * Creates some virtual cpus. Good luck creating more than one.
2440 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
2443 struct kvm_vcpu *vcpu;
2445 if (id >= KVM_MAX_VCPU_ID)
2448 mutex_lock(&kvm->lock);
2449 if (kvm->created_vcpus == KVM_MAX_VCPUS) {
2450 mutex_unlock(&kvm->lock);
2454 kvm->created_vcpus++;
2455 mutex_unlock(&kvm->lock);
2457 vcpu = kvm_arch_vcpu_create(kvm, id);
2460 goto vcpu_decrement;
2463 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2465 r = kvm_arch_vcpu_setup(vcpu);
2469 r = kvm_create_vcpu_debugfs(vcpu);
2473 mutex_lock(&kvm->lock);
2474 if (kvm_get_vcpu_by_id(kvm, id)) {
2476 goto unlock_vcpu_destroy;
2479 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
2481 /* Now it's all set up, let userspace reach it */
2483 r = create_vcpu_fd(vcpu);
2486 goto unlock_vcpu_destroy;
2489 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
2492 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
2493 * before kvm->online_vcpu's incremented value.
2496 atomic_inc(&kvm->online_vcpus);
2498 mutex_unlock(&kvm->lock);
2499 kvm_arch_vcpu_postcreate(vcpu);
2502 unlock_vcpu_destroy:
2503 mutex_unlock(&kvm->lock);
2504 debugfs_remove_recursive(vcpu->debugfs_dentry);
2506 kvm_arch_vcpu_destroy(vcpu);
2508 mutex_lock(&kvm->lock);
2509 kvm->created_vcpus--;
2510 mutex_unlock(&kvm->lock);
2514 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2517 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2518 vcpu->sigset_active = 1;
2519 vcpu->sigset = *sigset;
2521 vcpu->sigset_active = 0;
2525 static long kvm_vcpu_ioctl(struct file *filp,
2526 unsigned int ioctl, unsigned long arg)
2528 struct kvm_vcpu *vcpu = filp->private_data;
2529 void __user *argp = (void __user *)arg;
2531 struct kvm_fpu *fpu = NULL;
2532 struct kvm_sregs *kvm_sregs = NULL;
2534 if (vcpu->kvm->mm != current->mm)
2537 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2540 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
2542 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
2543 * so vcpu_load() would break it.
2545 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_S390_IRQ || ioctl == KVM_INTERRUPT)
2546 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2550 r = vcpu_load(vcpu);
2558 if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
2559 /* The thread running this VCPU changed. */
2560 struct pid *oldpid = vcpu->pid;
2561 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
2563 rcu_assign_pointer(vcpu->pid, newpid);
2568 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2569 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2571 case KVM_GET_REGS: {
2572 struct kvm_regs *kvm_regs;
2575 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2578 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2582 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2589 case KVM_SET_REGS: {
2590 struct kvm_regs *kvm_regs;
2593 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2594 if (IS_ERR(kvm_regs)) {
2595 r = PTR_ERR(kvm_regs);
2598 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2602 case KVM_GET_SREGS: {
2603 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2607 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2611 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2616 case KVM_SET_SREGS: {
2617 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2618 if (IS_ERR(kvm_sregs)) {
2619 r = PTR_ERR(kvm_sregs);
2623 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2626 case KVM_GET_MP_STATE: {
2627 struct kvm_mp_state mp_state;
2629 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2633 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
2638 case KVM_SET_MP_STATE: {
2639 struct kvm_mp_state mp_state;
2642 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
2644 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2647 case KVM_TRANSLATE: {
2648 struct kvm_translation tr;
2651 if (copy_from_user(&tr, argp, sizeof(tr)))
2653 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2657 if (copy_to_user(argp, &tr, sizeof(tr)))
2662 case KVM_SET_GUEST_DEBUG: {
2663 struct kvm_guest_debug dbg;
2666 if (copy_from_user(&dbg, argp, sizeof(dbg)))
2668 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2671 case KVM_SET_SIGNAL_MASK: {
2672 struct kvm_signal_mask __user *sigmask_arg = argp;
2673 struct kvm_signal_mask kvm_sigmask;
2674 sigset_t sigset, *p;
2679 if (copy_from_user(&kvm_sigmask, argp,
2680 sizeof(kvm_sigmask)))
2683 if (kvm_sigmask.len != sizeof(sigset))
2686 if (copy_from_user(&sigset, sigmask_arg->sigset,
2691 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2695 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2699 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2703 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2709 fpu = memdup_user(argp, sizeof(*fpu));
2715 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2719 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2728 #ifdef CONFIG_KVM_COMPAT
2729 static long kvm_vcpu_compat_ioctl(struct file *filp,
2730 unsigned int ioctl, unsigned long arg)
2732 struct kvm_vcpu *vcpu = filp->private_data;
2733 void __user *argp = compat_ptr(arg);
2736 if (vcpu->kvm->mm != current->mm)
2740 case KVM_SET_SIGNAL_MASK: {
2741 struct kvm_signal_mask __user *sigmask_arg = argp;
2742 struct kvm_signal_mask kvm_sigmask;
2743 compat_sigset_t csigset;
2748 if (copy_from_user(&kvm_sigmask, argp,
2749 sizeof(kvm_sigmask)))
2752 if (kvm_sigmask.len != sizeof(csigset))
2755 if (copy_from_user(&csigset, sigmask_arg->sigset,
2758 sigset_from_compat(&sigset, &csigset);
2759 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2761 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2765 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2773 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2774 int (*accessor)(struct kvm_device *dev,
2775 struct kvm_device_attr *attr),
2778 struct kvm_device_attr attr;
2783 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2786 return accessor(dev, &attr);
2789 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2792 struct kvm_device *dev = filp->private_data;
2795 case KVM_SET_DEVICE_ATTR:
2796 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2797 case KVM_GET_DEVICE_ATTR:
2798 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2799 case KVM_HAS_DEVICE_ATTR:
2800 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2802 if (dev->ops->ioctl)
2803 return dev->ops->ioctl(dev, ioctl, arg);
2809 static int kvm_device_release(struct inode *inode, struct file *filp)
2811 struct kvm_device *dev = filp->private_data;
2812 struct kvm *kvm = dev->kvm;
2818 static const struct file_operations kvm_device_fops = {
2819 .unlocked_ioctl = kvm_device_ioctl,
2820 #ifdef CONFIG_KVM_COMPAT
2821 .compat_ioctl = kvm_device_ioctl,
2823 .release = kvm_device_release,
2826 struct kvm_device *kvm_device_from_filp(struct file *filp)
2828 if (filp->f_op != &kvm_device_fops)
2831 return filp->private_data;
2834 static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
2835 #ifdef CONFIG_KVM_MPIC
2836 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
2837 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
2841 int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
2843 if (type >= ARRAY_SIZE(kvm_device_ops_table))
2846 if (kvm_device_ops_table[type] != NULL)
2849 kvm_device_ops_table[type] = ops;
2853 void kvm_unregister_device_ops(u32 type)
2855 if (kvm_device_ops_table[type] != NULL)
2856 kvm_device_ops_table[type] = NULL;
2859 static int kvm_ioctl_create_device(struct kvm *kvm,
2860 struct kvm_create_device *cd)
2862 struct kvm_device_ops *ops = NULL;
2863 struct kvm_device *dev;
2864 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2867 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
2870 ops = kvm_device_ops_table[cd->type];
2877 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2884 mutex_lock(&kvm->lock);
2885 ret = ops->create(dev, cd->type);
2887 mutex_unlock(&kvm->lock);
2891 list_add(&dev->vm_node, &kvm->devices);
2892 mutex_unlock(&kvm->lock);
2897 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2899 mutex_lock(&kvm->lock);
2900 list_del(&dev->vm_node);
2901 mutex_unlock(&kvm->lock);
2911 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2914 case KVM_CAP_USER_MEMORY:
2915 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2916 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2917 case KVM_CAP_INTERNAL_ERROR_DATA:
2918 #ifdef CONFIG_HAVE_KVM_MSI
2919 case KVM_CAP_SIGNAL_MSI:
2921 #ifdef CONFIG_HAVE_KVM_IRQFD
2923 case KVM_CAP_IRQFD_RESAMPLE:
2925 case KVM_CAP_IOEVENTFD_ANY_LENGTH:
2926 case KVM_CAP_CHECK_EXTENSION_VM:
2928 #ifdef CONFIG_KVM_MMIO
2929 case KVM_CAP_COALESCED_MMIO:
2930 return KVM_COALESCED_MMIO_PAGE_OFFSET;
2932 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2933 case KVM_CAP_IRQ_ROUTING:
2934 return KVM_MAX_IRQ_ROUTES;
2936 #if KVM_ADDRESS_SPACE_NUM > 1
2937 case KVM_CAP_MULTI_ADDRESS_SPACE:
2938 return KVM_ADDRESS_SPACE_NUM;
2940 case KVM_CAP_MAX_VCPU_ID:
2941 return KVM_MAX_VCPU_ID;
2945 return kvm_vm_ioctl_check_extension(kvm, arg);
2948 static long kvm_vm_ioctl(struct file *filp,
2949 unsigned int ioctl, unsigned long arg)
2951 struct kvm *kvm = filp->private_data;
2952 void __user *argp = (void __user *)arg;
2955 if (kvm->mm != current->mm)
2958 case KVM_CREATE_VCPU:
2959 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2961 case KVM_SET_USER_MEMORY_REGION: {
2962 struct kvm_userspace_memory_region kvm_userspace_mem;
2965 if (copy_from_user(&kvm_userspace_mem, argp,
2966 sizeof(kvm_userspace_mem)))
2969 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2972 case KVM_GET_DIRTY_LOG: {
2973 struct kvm_dirty_log log;
2976 if (copy_from_user(&log, argp, sizeof(log)))
2978 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2981 #ifdef CONFIG_KVM_MMIO
2982 case KVM_REGISTER_COALESCED_MMIO: {
2983 struct kvm_coalesced_mmio_zone zone;
2986 if (copy_from_user(&zone, argp, sizeof(zone)))
2988 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2991 case KVM_UNREGISTER_COALESCED_MMIO: {
2992 struct kvm_coalesced_mmio_zone zone;
2995 if (copy_from_user(&zone, argp, sizeof(zone)))
2997 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
3002 struct kvm_irqfd data;
3005 if (copy_from_user(&data, argp, sizeof(data)))
3007 r = kvm_irqfd(kvm, &data);
3010 case KVM_IOEVENTFD: {
3011 struct kvm_ioeventfd data;
3014 if (copy_from_user(&data, argp, sizeof(data)))
3016 r = kvm_ioeventfd(kvm, &data);
3019 #ifdef CONFIG_HAVE_KVM_MSI
3020 case KVM_SIGNAL_MSI: {
3024 if (copy_from_user(&msi, argp, sizeof(msi)))
3026 r = kvm_send_userspace_msi(kvm, &msi);
3030 #ifdef __KVM_HAVE_IRQ_LINE
3031 case KVM_IRQ_LINE_STATUS:
3032 case KVM_IRQ_LINE: {
3033 struct kvm_irq_level irq_event;
3036 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
3039 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
3040 ioctl == KVM_IRQ_LINE_STATUS);
3045 if (ioctl == KVM_IRQ_LINE_STATUS) {
3046 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
3054 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3055 case KVM_SET_GSI_ROUTING: {
3056 struct kvm_irq_routing routing;
3057 struct kvm_irq_routing __user *urouting;
3058 struct kvm_irq_routing_entry *entries = NULL;
3061 if (copy_from_user(&routing, argp, sizeof(routing)))
3064 if (!kvm_arch_can_set_irq_routing(kvm))
3066 if (routing.nr > KVM_MAX_IRQ_ROUTES)
3072 entries = vmalloc(routing.nr * sizeof(*entries));
3077 if (copy_from_user(entries, urouting->entries,
3078 routing.nr * sizeof(*entries)))
3079 goto out_free_irq_routing;
3081 r = kvm_set_irq_routing(kvm, entries, routing.nr,
3083 out_free_irq_routing:
3087 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
3088 case KVM_CREATE_DEVICE: {
3089 struct kvm_create_device cd;
3092 if (copy_from_user(&cd, argp, sizeof(cd)))
3095 r = kvm_ioctl_create_device(kvm, &cd);
3100 if (copy_to_user(argp, &cd, sizeof(cd)))
3106 case KVM_CHECK_EXTENSION:
3107 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
3110 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
3116 #ifdef CONFIG_KVM_COMPAT
3117 struct compat_kvm_dirty_log {
3121 compat_uptr_t dirty_bitmap; /* one bit per page */
3126 static long kvm_vm_compat_ioctl(struct file *filp,
3127 unsigned int ioctl, unsigned long arg)
3129 struct kvm *kvm = filp->private_data;
3132 if (kvm->mm != current->mm)
3135 case KVM_GET_DIRTY_LOG: {
3136 struct compat_kvm_dirty_log compat_log;
3137 struct kvm_dirty_log log;
3139 if (copy_from_user(&compat_log, (void __user *)arg,
3140 sizeof(compat_log)))
3142 log.slot = compat_log.slot;
3143 log.padding1 = compat_log.padding1;
3144 log.padding2 = compat_log.padding2;
3145 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
3147 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3151 r = kvm_vm_ioctl(filp, ioctl, arg);
3157 static struct file_operations kvm_vm_fops = {
3158 .release = kvm_vm_release,
3159 .unlocked_ioctl = kvm_vm_ioctl,
3160 #ifdef CONFIG_KVM_COMPAT
3161 .compat_ioctl = kvm_vm_compat_ioctl,
3163 .llseek = noop_llseek,
3166 static int kvm_dev_ioctl_create_vm(unsigned long type)
3172 kvm = kvm_create_vm(type);
3174 return PTR_ERR(kvm);
3175 #ifdef CONFIG_KVM_MMIO
3176 r = kvm_coalesced_mmio_init(kvm);
3182 r = get_unused_fd_flags(O_CLOEXEC);
3187 file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
3191 return PTR_ERR(file);
3195 * Don't call kvm_put_kvm anymore at this point; file->f_op is
3196 * already set, with ->release() being kvm_vm_release(). In error
3197 * cases it will be called by the final fput(file) and will take
3198 * care of doing kvm_put_kvm(kvm).
3200 if (kvm_create_vm_debugfs(kvm, r) < 0) {
3206 fd_install(r, file);
3210 static long kvm_dev_ioctl(struct file *filp,
3211 unsigned int ioctl, unsigned long arg)
3216 case KVM_GET_API_VERSION:
3219 r = KVM_API_VERSION;
3222 r = kvm_dev_ioctl_create_vm(arg);
3224 case KVM_CHECK_EXTENSION:
3225 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
3227 case KVM_GET_VCPU_MMAP_SIZE:
3230 r = PAGE_SIZE; /* struct kvm_run */
3232 r += PAGE_SIZE; /* pio data page */
3234 #ifdef CONFIG_KVM_MMIO
3235 r += PAGE_SIZE; /* coalesced mmio ring page */
3238 case KVM_TRACE_ENABLE:
3239 case KVM_TRACE_PAUSE:
3240 case KVM_TRACE_DISABLE:
3244 return kvm_arch_dev_ioctl(filp, ioctl, arg);
3250 static struct file_operations kvm_chardev_ops = {
3251 .unlocked_ioctl = kvm_dev_ioctl,
3252 .compat_ioctl = kvm_dev_ioctl,
3253 .llseek = noop_llseek,
3256 static struct miscdevice kvm_dev = {
3262 static void hardware_enable_nolock(void *junk)
3264 int cpu = raw_smp_processor_id();
3267 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
3270 cpumask_set_cpu(cpu, cpus_hardware_enabled);
3272 r = kvm_arch_hardware_enable();
3275 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3276 atomic_inc(&hardware_enable_failed);
3277 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
3281 static int kvm_starting_cpu(unsigned int cpu)
3283 raw_spin_lock(&kvm_count_lock);
3284 if (kvm_usage_count)
3285 hardware_enable_nolock(NULL);
3286 raw_spin_unlock(&kvm_count_lock);
3290 static void hardware_disable_nolock(void *junk)
3292 int cpu = raw_smp_processor_id();
3294 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
3296 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3297 kvm_arch_hardware_disable();
3300 static int kvm_dying_cpu(unsigned int cpu)
3302 raw_spin_lock(&kvm_count_lock);
3303 if (kvm_usage_count)
3304 hardware_disable_nolock(NULL);
3305 raw_spin_unlock(&kvm_count_lock);
3309 static void hardware_disable_all_nolock(void)
3311 BUG_ON(!kvm_usage_count);
3314 if (!kvm_usage_count)
3315 on_each_cpu(hardware_disable_nolock, NULL, 1);
3318 static void hardware_disable_all(void)
3320 raw_spin_lock(&kvm_count_lock);
3321 hardware_disable_all_nolock();
3322 raw_spin_unlock(&kvm_count_lock);
3325 static int hardware_enable_all(void)
3329 raw_spin_lock(&kvm_count_lock);
3332 if (kvm_usage_count == 1) {
3333 atomic_set(&hardware_enable_failed, 0);
3334 on_each_cpu(hardware_enable_nolock, NULL, 1);
3336 if (atomic_read(&hardware_enable_failed)) {
3337 hardware_disable_all_nolock();
3342 raw_spin_unlock(&kvm_count_lock);
3347 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3351 * Some (well, at least mine) BIOSes hang on reboot if
3354 * And Intel TXT required VMX off for all cpu when system shutdown.
3356 pr_info("kvm: exiting hardware virtualization\n");
3357 kvm_rebooting = true;
3358 on_each_cpu(hardware_disable_nolock, NULL, 1);
3362 static struct notifier_block kvm_reboot_notifier = {
3363 .notifier_call = kvm_reboot,
3367 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3371 for (i = 0; i < bus->dev_count; i++) {
3372 struct kvm_io_device *pos = bus->range[i].dev;
3374 kvm_iodevice_destructor(pos);
3379 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
3380 const struct kvm_io_range *r2)
3382 gpa_t addr1 = r1->addr;
3383 gpa_t addr2 = r2->addr;
3388 /* If r2->len == 0, match the exact address. If r2->len != 0,
3389 * accept any overlapping write. Any order is acceptable for
3390 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
3391 * we process all of them.
3404 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
3406 return kvm_io_bus_cmp(p1, p2);
3409 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
3410 gpa_t addr, int len)
3412 bus->range[bus->dev_count++] = (struct kvm_io_range) {
3418 sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
3419 kvm_io_bus_sort_cmp, NULL);
3424 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
3425 gpa_t addr, int len)
3427 struct kvm_io_range *range, key;
3430 key = (struct kvm_io_range) {
3435 range = bsearch(&key, bus->range, bus->dev_count,
3436 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
3440 off = range - bus->range;
3442 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
3448 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3449 struct kvm_io_range *range, const void *val)
3453 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3457 while (idx < bus->dev_count &&
3458 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3459 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
3468 /* kvm_io_bus_write - called under kvm->slots_lock */
3469 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3470 int len, const void *val)
3472 struct kvm_io_bus *bus;
3473 struct kvm_io_range range;
3476 range = (struct kvm_io_range) {
3481 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3484 r = __kvm_io_bus_write(vcpu, bus, &range, val);
3485 return r < 0 ? r : 0;
3488 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3489 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
3490 gpa_t addr, int len, const void *val, long cookie)
3492 struct kvm_io_bus *bus;
3493 struct kvm_io_range range;
3495 range = (struct kvm_io_range) {
3500 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3504 /* First try the device referenced by cookie. */
3505 if ((cookie >= 0) && (cookie < bus->dev_count) &&
3506 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
3507 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
3512 * cookie contained garbage; fall back to search and return the
3513 * correct cookie value.
3515 return __kvm_io_bus_write(vcpu, bus, &range, val);
3518 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3519 struct kvm_io_range *range, void *val)
3523 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3527 while (idx < bus->dev_count &&
3528 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3529 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
3537 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
3539 /* kvm_io_bus_read - called under kvm->slots_lock */
3540 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3543 struct kvm_io_bus *bus;
3544 struct kvm_io_range range;
3547 range = (struct kvm_io_range) {
3552 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3555 r = __kvm_io_bus_read(vcpu, bus, &range, val);
3556 return r < 0 ? r : 0;
3560 /* Caller must hold slots_lock. */
3561 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3562 int len, struct kvm_io_device *dev)
3564 struct kvm_io_bus *new_bus, *bus;
3566 bus = kvm->buses[bus_idx];
3570 /* exclude ioeventfd which is limited by maximum fd */
3571 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3574 new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count + 1) *
3575 sizeof(struct kvm_io_range)), GFP_KERNEL);
3578 memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
3579 sizeof(struct kvm_io_range)));
3580 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
3581 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3582 synchronize_srcu_expedited(&kvm->srcu);
3588 /* Caller must hold slots_lock. */
3589 void kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3590 struct kvm_io_device *dev)
3593 struct kvm_io_bus *new_bus, *bus;
3595 bus = kvm->buses[bus_idx];
3599 for (i = 0; i < bus->dev_count; i++)
3600 if (bus->range[i].dev == dev) {
3604 if (i == bus->dev_count)
3607 new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3608 sizeof(struct kvm_io_range)), GFP_KERNEL);
3610 pr_err("kvm: failed to shrink bus, removing it completely\n");
3614 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3615 new_bus->dev_count--;
3616 memcpy(new_bus->range + i, bus->range + i + 1,
3617 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3620 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3621 synchronize_srcu_expedited(&kvm->srcu);
3626 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3629 struct kvm_io_bus *bus;
3630 int dev_idx, srcu_idx;
3631 struct kvm_io_device *iodev = NULL;
3633 srcu_idx = srcu_read_lock(&kvm->srcu);
3635 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
3639 dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
3643 iodev = bus->range[dev_idx].dev;
3646 srcu_read_unlock(&kvm->srcu, srcu_idx);
3650 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
3652 static int kvm_debugfs_open(struct inode *inode, struct file *file,
3653 int (*get)(void *, u64 *), int (*set)(void *, u64),
3656 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
3659 /* The debugfs files are a reference to the kvm struct which
3660 * is still valid when kvm_destroy_vm is called.
3661 * To avoid the race between open and the removal of the debugfs
3662 * directory we test against the users count.
3664 if (!refcount_inc_not_zero(&stat_data->kvm->users_count))
3667 if (simple_attr_open(inode, file, get, set, fmt)) {
3668 kvm_put_kvm(stat_data->kvm);
3675 static int kvm_debugfs_release(struct inode *inode, struct file *file)
3677 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
3680 simple_attr_release(inode, file);
3681 kvm_put_kvm(stat_data->kvm);
3686 static int vm_stat_get_per_vm(void *data, u64 *val)
3688 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3690 *val = *(ulong *)((void *)stat_data->kvm + stat_data->offset);
3695 static int vm_stat_clear_per_vm(void *data, u64 val)
3697 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3702 *(ulong *)((void *)stat_data->kvm + stat_data->offset) = 0;
3707 static int vm_stat_get_per_vm_open(struct inode *inode, struct file *file)
3709 __simple_attr_check_format("%llu\n", 0ull);
3710 return kvm_debugfs_open(inode, file, vm_stat_get_per_vm,
3711 vm_stat_clear_per_vm, "%llu\n");
3714 static const struct file_operations vm_stat_get_per_vm_fops = {
3715 .owner = THIS_MODULE,
3716 .open = vm_stat_get_per_vm_open,
3717 .release = kvm_debugfs_release,
3718 .read = simple_attr_read,
3719 .write = simple_attr_write,
3720 .llseek = no_llseek,
3723 static int vcpu_stat_get_per_vm(void *data, u64 *val)
3726 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3727 struct kvm_vcpu *vcpu;
3731 kvm_for_each_vcpu(i, vcpu, stat_data->kvm)
3732 *val += *(u64 *)((void *)vcpu + stat_data->offset);
3737 static int vcpu_stat_clear_per_vm(void *data, u64 val)
3740 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3741 struct kvm_vcpu *vcpu;
3746 kvm_for_each_vcpu(i, vcpu, stat_data->kvm)
3747 *(u64 *)((void *)vcpu + stat_data->offset) = 0;
3752 static int vcpu_stat_get_per_vm_open(struct inode *inode, struct file *file)
3754 __simple_attr_check_format("%llu\n", 0ull);
3755 return kvm_debugfs_open(inode, file, vcpu_stat_get_per_vm,
3756 vcpu_stat_clear_per_vm, "%llu\n");
3759 static const struct file_operations vcpu_stat_get_per_vm_fops = {
3760 .owner = THIS_MODULE,
3761 .open = vcpu_stat_get_per_vm_open,
3762 .release = kvm_debugfs_release,
3763 .read = simple_attr_read,
3764 .write = simple_attr_write,
3765 .llseek = no_llseek,
3768 static const struct file_operations *stat_fops_per_vm[] = {
3769 [KVM_STAT_VCPU] = &vcpu_stat_get_per_vm_fops,
3770 [KVM_STAT_VM] = &vm_stat_get_per_vm_fops,
3773 static int vm_stat_get(void *_offset, u64 *val)
3775 unsigned offset = (long)_offset;
3777 struct kvm_stat_data stat_tmp = {.offset = offset};
3781 spin_lock(&kvm_lock);
3782 list_for_each_entry(kvm, &vm_list, vm_list) {
3784 vm_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
3787 spin_unlock(&kvm_lock);
3791 static int vm_stat_clear(void *_offset, u64 val)
3793 unsigned offset = (long)_offset;
3795 struct kvm_stat_data stat_tmp = {.offset = offset};
3800 spin_lock(&kvm_lock);
3801 list_for_each_entry(kvm, &vm_list, vm_list) {
3803 vm_stat_clear_per_vm((void *)&stat_tmp, 0);
3805 spin_unlock(&kvm_lock);
3810 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n");
3812 static int vcpu_stat_get(void *_offset, u64 *val)
3814 unsigned offset = (long)_offset;
3816 struct kvm_stat_data stat_tmp = {.offset = offset};
3820 spin_lock(&kvm_lock);
3821 list_for_each_entry(kvm, &vm_list, vm_list) {
3823 vcpu_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
3826 spin_unlock(&kvm_lock);
3830 static int vcpu_stat_clear(void *_offset, u64 val)
3832 unsigned offset = (long)_offset;
3834 struct kvm_stat_data stat_tmp = {.offset = offset};
3839 spin_lock(&kvm_lock);
3840 list_for_each_entry(kvm, &vm_list, vm_list) {
3842 vcpu_stat_clear_per_vm((void *)&stat_tmp, 0);
3844 spin_unlock(&kvm_lock);
3849 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear,
3852 static const struct file_operations *stat_fops[] = {
3853 [KVM_STAT_VCPU] = &vcpu_stat_fops,
3854 [KVM_STAT_VM] = &vm_stat_fops,
3857 static int kvm_init_debug(void)
3860 struct kvm_stats_debugfs_item *p;
3862 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3863 if (kvm_debugfs_dir == NULL)
3866 kvm_debugfs_num_entries = 0;
3867 for (p = debugfs_entries; p->name; ++p, kvm_debugfs_num_entries++) {
3868 if (!debugfs_create_file(p->name, 0644, kvm_debugfs_dir,
3869 (void *)(long)p->offset,
3870 stat_fops[p->kind]))
3877 debugfs_remove_recursive(kvm_debugfs_dir);
3882 static int kvm_suspend(void)
3884 if (kvm_usage_count)
3885 hardware_disable_nolock(NULL);
3889 static void kvm_resume(void)
3891 if (kvm_usage_count) {
3892 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3893 hardware_enable_nolock(NULL);
3897 static struct syscore_ops kvm_syscore_ops = {
3898 .suspend = kvm_suspend,
3899 .resume = kvm_resume,
3903 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3905 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3908 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3910 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3912 if (vcpu->preempted)
3913 vcpu->preempted = false;
3915 kvm_arch_sched_in(vcpu, cpu);
3917 kvm_arch_vcpu_load(vcpu, cpu);
3920 static void kvm_sched_out(struct preempt_notifier *pn,
3921 struct task_struct *next)
3923 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3925 if (current->state == TASK_RUNNING)
3926 vcpu->preempted = true;
3927 kvm_arch_vcpu_put(vcpu);
3930 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3931 struct module *module)
3936 r = kvm_arch_init(opaque);
3941 * kvm_arch_init makes sure there's at most one caller
3942 * for architectures that support multiple implementations,
3943 * like intel and amd on x86.
3944 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3945 * conflicts in case kvm is already setup for another implementation.
3947 r = kvm_irqfd_init();
3951 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
3956 r = kvm_arch_hardware_setup();
3960 for_each_online_cpu(cpu) {
3961 smp_call_function_single(cpu,
3962 kvm_arch_check_processor_compat,
3968 r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting",
3969 kvm_starting_cpu, kvm_dying_cpu);
3972 register_reboot_notifier(&kvm_reboot_notifier);
3974 /* A kmem cache lets us meet the alignment requirements of fx_save. */
3976 vcpu_align = __alignof__(struct kvm_vcpu);
3977 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
3979 if (!kvm_vcpu_cache) {
3984 r = kvm_async_pf_init();
3988 kvm_chardev_ops.owner = module;
3989 kvm_vm_fops.owner = module;
3990 kvm_vcpu_fops.owner = module;
3992 r = misc_register(&kvm_dev);
3994 pr_err("kvm: misc device register failed\n");
3998 register_syscore_ops(&kvm_syscore_ops);
4000 kvm_preempt_ops.sched_in = kvm_sched_in;
4001 kvm_preempt_ops.sched_out = kvm_sched_out;
4003 r = kvm_init_debug();
4005 pr_err("kvm: create debugfs files failed\n");
4009 r = kvm_vfio_ops_init();
4015 unregister_syscore_ops(&kvm_syscore_ops);
4016 misc_deregister(&kvm_dev);
4018 kvm_async_pf_deinit();
4020 kmem_cache_destroy(kvm_vcpu_cache);
4022 unregister_reboot_notifier(&kvm_reboot_notifier);
4023 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4026 kvm_arch_hardware_unsetup();
4028 free_cpumask_var(cpus_hardware_enabled);
4036 EXPORT_SYMBOL_GPL(kvm_init);
4040 debugfs_remove_recursive(kvm_debugfs_dir);
4041 misc_deregister(&kvm_dev);
4042 kmem_cache_destroy(kvm_vcpu_cache);
4043 kvm_async_pf_deinit();
4044 unregister_syscore_ops(&kvm_syscore_ops);
4045 unregister_reboot_notifier(&kvm_reboot_notifier);
4046 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4047 on_each_cpu(hardware_disable_nolock, NULL, 1);
4048 kvm_arch_hardware_unsetup();
4051 free_cpumask_var(cpus_hardware_enabled);
4052 kvm_vfio_ops_exit();
4054 EXPORT_SYMBOL_GPL(kvm_exit);