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.h>
36 #include <linux/cpumask.h>
37 #include <linux/smp.h>
38 #include <linux/anon_inodes.h>
39 #include <linux/profile.h>
40 #include <linux/kvm_para.h>
41 #include <linux/pagemap.h>
42 #include <linux/mman.h>
43 #include <linux/swap.h>
44 #include <linux/bitops.h>
45 #include <linux/spinlock.h>
46 #include <linux/compat.h>
47 #include <linux/srcu.h>
48 #include <linux/hugetlb.h>
49 #include <linux/slab.h>
50 #include <linux/sort.h>
51 #include <linux/bsearch.h>
53 #include <asm/processor.h>
55 #include <asm/ioctl.h>
56 #include <linux/uaccess.h>
57 #include <asm/pgtable.h>
59 #include "coalesced_mmio.h"
63 #define CREATE_TRACE_POINTS
64 #include <trace/events/kvm.h>
66 /* Worst case buffer size needed for holding an integer. */
67 #define ITOA_MAX_LEN 12
69 MODULE_AUTHOR("Qumranet");
70 MODULE_LICENSE("GPL");
72 /* Architectures should define their poll value according to the halt latency */
73 unsigned int halt_poll_ns = KVM_HALT_POLL_NS_DEFAULT;
74 module_param(halt_poll_ns, uint, S_IRUGO | S_IWUSR);
75 EXPORT_SYMBOL_GPL(halt_poll_ns);
77 /* Default doubles per-vcpu halt_poll_ns. */
78 unsigned int halt_poll_ns_grow = 2;
79 module_param(halt_poll_ns_grow, uint, S_IRUGO | S_IWUSR);
80 EXPORT_SYMBOL_GPL(halt_poll_ns_grow);
82 /* Default resets per-vcpu halt_poll_ns . */
83 unsigned int halt_poll_ns_shrink;
84 module_param(halt_poll_ns_shrink, uint, S_IRUGO | S_IWUSR);
85 EXPORT_SYMBOL_GPL(halt_poll_ns_shrink);
90 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
93 DEFINE_SPINLOCK(kvm_lock);
94 static DEFINE_RAW_SPINLOCK(kvm_count_lock);
97 static cpumask_var_t cpus_hardware_enabled;
98 static int kvm_usage_count;
99 static atomic_t hardware_enable_failed;
101 struct kmem_cache *kvm_vcpu_cache;
102 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
104 static __read_mostly struct preempt_ops kvm_preempt_ops;
106 struct dentry *kvm_debugfs_dir;
107 EXPORT_SYMBOL_GPL(kvm_debugfs_dir);
109 static int kvm_debugfs_num_entries;
110 static const struct file_operations *stat_fops_per_vm[];
112 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
114 #ifdef CONFIG_KVM_COMPAT
115 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
118 static int hardware_enable_all(void);
119 static void hardware_disable_all(void);
121 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
123 static void kvm_release_pfn_dirty(kvm_pfn_t pfn);
124 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot, gfn_t gfn);
126 __visible bool kvm_rebooting;
127 EXPORT_SYMBOL_GPL(kvm_rebooting);
129 static bool largepages_enabled = true;
131 bool kvm_is_reserved_pfn(kvm_pfn_t pfn)
134 return PageReserved(pfn_to_page(pfn));
140 * Switches to specified vcpu, until a matching vcpu_put()
142 int vcpu_load(struct kvm_vcpu *vcpu)
146 if (mutex_lock_killable(&vcpu->mutex))
149 preempt_notifier_register(&vcpu->preempt_notifier);
150 kvm_arch_vcpu_load(vcpu, cpu);
154 EXPORT_SYMBOL_GPL(vcpu_load);
156 void vcpu_put(struct kvm_vcpu *vcpu)
159 kvm_arch_vcpu_put(vcpu);
160 preempt_notifier_unregister(&vcpu->preempt_notifier);
162 mutex_unlock(&vcpu->mutex);
164 EXPORT_SYMBOL_GPL(vcpu_put);
166 static void ack_flush(void *_completed)
170 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
175 struct kvm_vcpu *vcpu;
177 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
180 kvm_for_each_vcpu(i, vcpu, kvm) {
181 kvm_make_request(req, vcpu);
184 /* Set ->requests bit before we read ->mode. */
185 smp_mb__after_atomic();
187 if (cpus != NULL && cpu != -1 && cpu != me &&
188 kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
189 cpumask_set_cpu(cpu, cpus);
191 if (unlikely(cpus == NULL))
192 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
193 else if (!cpumask_empty(cpus))
194 smp_call_function_many(cpus, ack_flush, NULL, 1);
198 free_cpumask_var(cpus);
202 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
203 void kvm_flush_remote_tlbs(struct kvm *kvm)
206 * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in
207 * kvm_make_all_cpus_request.
209 long dirty_count = smp_load_acquire(&kvm->tlbs_dirty);
212 * We want to publish modifications to the page tables before reading
213 * mode. Pairs with a memory barrier in arch-specific code.
214 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
215 * and smp_mb in walk_shadow_page_lockless_begin/end.
216 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
218 * There is already an smp_mb__after_atomic() before
219 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
222 if (kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
223 ++kvm->stat.remote_tlb_flush;
224 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
226 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
229 void kvm_reload_remote_mmus(struct kvm *kvm)
231 kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
234 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
239 mutex_init(&vcpu->mutex);
244 init_swait_queue_head(&vcpu->wq);
245 kvm_async_pf_vcpu_init(vcpu);
248 INIT_LIST_HEAD(&vcpu->blocked_vcpu_list);
250 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
255 vcpu->run = page_address(page);
257 kvm_vcpu_set_in_spin_loop(vcpu, false);
258 kvm_vcpu_set_dy_eligible(vcpu, false);
259 vcpu->preempted = false;
261 r = kvm_arch_vcpu_init(vcpu);
267 free_page((unsigned long)vcpu->run);
271 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
273 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
276 kvm_arch_vcpu_uninit(vcpu);
277 free_page((unsigned long)vcpu->run);
279 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
281 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
282 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
284 return container_of(mn, struct kvm, mmu_notifier);
287 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
288 struct mm_struct *mm,
289 unsigned long address)
291 struct kvm *kvm = mmu_notifier_to_kvm(mn);
292 int need_tlb_flush, idx;
295 * When ->invalidate_page runs, the linux pte has been zapped
296 * already but the page is still allocated until
297 * ->invalidate_page returns. So if we increase the sequence
298 * here the kvm page fault will notice if the spte can't be
299 * established because the page is going to be freed. If
300 * instead the kvm page fault establishes the spte before
301 * ->invalidate_page runs, kvm_unmap_hva will release it
304 * The sequence increase only need to be seen at spin_unlock
305 * time, and not at spin_lock time.
307 * Increasing the sequence after the spin_unlock would be
308 * unsafe because the kvm page fault could then establish the
309 * pte after kvm_unmap_hva returned, without noticing the page
310 * is going to be freed.
312 idx = srcu_read_lock(&kvm->srcu);
313 spin_lock(&kvm->mmu_lock);
315 kvm->mmu_notifier_seq++;
316 need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
317 /* we've to flush the tlb before the pages can be freed */
319 kvm_flush_remote_tlbs(kvm);
321 spin_unlock(&kvm->mmu_lock);
323 kvm_arch_mmu_notifier_invalidate_page(kvm, address);
325 srcu_read_unlock(&kvm->srcu, idx);
328 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
329 struct mm_struct *mm,
330 unsigned long address,
333 struct kvm *kvm = mmu_notifier_to_kvm(mn);
336 idx = srcu_read_lock(&kvm->srcu);
337 spin_lock(&kvm->mmu_lock);
338 kvm->mmu_notifier_seq++;
339 kvm_set_spte_hva(kvm, address, pte);
340 spin_unlock(&kvm->mmu_lock);
341 srcu_read_unlock(&kvm->srcu, idx);
344 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
345 struct mm_struct *mm,
349 struct kvm *kvm = mmu_notifier_to_kvm(mn);
350 int need_tlb_flush = 0, idx;
352 idx = srcu_read_lock(&kvm->srcu);
353 spin_lock(&kvm->mmu_lock);
355 * The count increase must become visible at unlock time as no
356 * spte can be established without taking the mmu_lock and
357 * count is also read inside the mmu_lock critical section.
359 kvm->mmu_notifier_count++;
360 need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
361 need_tlb_flush |= kvm->tlbs_dirty;
362 /* we've to flush the tlb before the pages can be freed */
364 kvm_flush_remote_tlbs(kvm);
366 spin_unlock(&kvm->mmu_lock);
367 srcu_read_unlock(&kvm->srcu, idx);
370 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
371 struct mm_struct *mm,
375 struct kvm *kvm = mmu_notifier_to_kvm(mn);
377 spin_lock(&kvm->mmu_lock);
379 * This sequence increase will notify the kvm page fault that
380 * the page that is going to be mapped in the spte could have
383 kvm->mmu_notifier_seq++;
386 * The above sequence increase must be visible before the
387 * below count decrease, which is ensured by the smp_wmb above
388 * in conjunction with the smp_rmb in mmu_notifier_retry().
390 kvm->mmu_notifier_count--;
391 spin_unlock(&kvm->mmu_lock);
393 BUG_ON(kvm->mmu_notifier_count < 0);
396 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
397 struct mm_struct *mm,
401 struct kvm *kvm = mmu_notifier_to_kvm(mn);
404 idx = srcu_read_lock(&kvm->srcu);
405 spin_lock(&kvm->mmu_lock);
407 young = kvm_age_hva(kvm, start, end);
409 kvm_flush_remote_tlbs(kvm);
411 spin_unlock(&kvm->mmu_lock);
412 srcu_read_unlock(&kvm->srcu, idx);
417 static int kvm_mmu_notifier_clear_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 * Even though we do not flush TLB, this will still adversely
429 * affect performance on pre-Haswell Intel EPT, where there is
430 * no EPT Access Bit to clear so that we have to tear down EPT
431 * tables instead. If we find this unacceptable, we can always
432 * add a parameter to kvm_age_hva so that it effectively doesn't
433 * do anything on clear_young.
435 * Also note that currently we never issue secondary TLB flushes
436 * from clear_young, leaving this job up to the regular system
437 * cadence. If we find this inaccurate, we might come up with a
438 * more sophisticated heuristic later.
440 young = kvm_age_hva(kvm, start, end);
441 spin_unlock(&kvm->mmu_lock);
442 srcu_read_unlock(&kvm->srcu, idx);
447 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
448 struct mm_struct *mm,
449 unsigned long address)
451 struct kvm *kvm = mmu_notifier_to_kvm(mn);
454 idx = srcu_read_lock(&kvm->srcu);
455 spin_lock(&kvm->mmu_lock);
456 young = kvm_test_age_hva(kvm, address);
457 spin_unlock(&kvm->mmu_lock);
458 srcu_read_unlock(&kvm->srcu, idx);
463 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
464 struct mm_struct *mm)
466 struct kvm *kvm = mmu_notifier_to_kvm(mn);
469 idx = srcu_read_lock(&kvm->srcu);
470 kvm_arch_flush_shadow_all(kvm);
471 srcu_read_unlock(&kvm->srcu, idx);
474 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
475 .invalidate_page = kvm_mmu_notifier_invalidate_page,
476 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
477 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
478 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
479 .clear_young = kvm_mmu_notifier_clear_young,
480 .test_young = kvm_mmu_notifier_test_young,
481 .change_pte = kvm_mmu_notifier_change_pte,
482 .release = kvm_mmu_notifier_release,
485 static int kvm_init_mmu_notifier(struct kvm *kvm)
487 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
488 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
491 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
493 static int kvm_init_mmu_notifier(struct kvm *kvm)
498 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
500 static struct kvm_memslots *kvm_alloc_memslots(void)
503 struct kvm_memslots *slots;
505 slots = kvm_kvzalloc(sizeof(struct kvm_memslots));
509 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
510 slots->id_to_index[i] = slots->memslots[i].id = i;
515 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
517 if (!memslot->dirty_bitmap)
520 kvfree(memslot->dirty_bitmap);
521 memslot->dirty_bitmap = NULL;
525 * Free any memory in @free but not in @dont.
527 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
528 struct kvm_memory_slot *dont)
530 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
531 kvm_destroy_dirty_bitmap(free);
533 kvm_arch_free_memslot(kvm, free, dont);
538 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
540 struct kvm_memory_slot *memslot;
545 kvm_for_each_memslot(memslot, slots)
546 kvm_free_memslot(kvm, memslot, NULL);
551 static void kvm_destroy_vm_debugfs(struct kvm *kvm)
555 if (!kvm->debugfs_dentry)
558 debugfs_remove_recursive(kvm->debugfs_dentry);
560 if (kvm->debugfs_stat_data) {
561 for (i = 0; i < kvm_debugfs_num_entries; i++)
562 kfree(kvm->debugfs_stat_data[i]);
563 kfree(kvm->debugfs_stat_data);
567 static int kvm_create_vm_debugfs(struct kvm *kvm, int fd)
569 char dir_name[ITOA_MAX_LEN * 2];
570 struct kvm_stat_data *stat_data;
571 struct kvm_stats_debugfs_item *p;
573 if (!debugfs_initialized())
576 snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd);
577 kvm->debugfs_dentry = debugfs_create_dir(dir_name,
579 if (!kvm->debugfs_dentry)
582 kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries,
583 sizeof(*kvm->debugfs_stat_data),
585 if (!kvm->debugfs_stat_data)
588 for (p = debugfs_entries; p->name; p++) {
589 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL);
593 stat_data->kvm = kvm;
594 stat_data->offset = p->offset;
595 kvm->debugfs_stat_data[p - debugfs_entries] = stat_data;
596 if (!debugfs_create_file(p->name, 0644,
599 stat_fops_per_vm[p->kind]))
605 static struct kvm *kvm_create_vm(unsigned long type)
608 struct kvm *kvm = kvm_arch_alloc_vm();
611 return ERR_PTR(-ENOMEM);
613 spin_lock_init(&kvm->mmu_lock);
615 kvm->mm = current->mm;
616 kvm_eventfd_init(kvm);
617 mutex_init(&kvm->lock);
618 mutex_init(&kvm->irq_lock);
619 mutex_init(&kvm->slots_lock);
620 atomic_set(&kvm->users_count, 1);
621 INIT_LIST_HEAD(&kvm->devices);
623 r = kvm_arch_init_vm(kvm, type);
625 goto out_err_no_disable;
627 r = hardware_enable_all();
629 goto out_err_no_disable;
631 #ifdef CONFIG_HAVE_KVM_IRQFD
632 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
635 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
638 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
639 struct kvm_memslots *slots = kvm_alloc_memslots();
641 goto out_err_no_srcu;
643 * Generations must be different for each address space.
644 * Init kvm generation close to the maximum to easily test the
645 * code of handling generation number wrap-around.
647 slots->generation = i * 2 - 150;
648 rcu_assign_pointer(kvm->memslots[i], slots);
651 if (init_srcu_struct(&kvm->srcu))
652 goto out_err_no_srcu;
653 if (init_srcu_struct(&kvm->irq_srcu))
654 goto out_err_no_irq_srcu;
655 for (i = 0; i < KVM_NR_BUSES; i++) {
656 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
662 r = kvm_init_mmu_notifier(kvm);
666 spin_lock(&kvm_lock);
667 list_add(&kvm->vm_list, &vm_list);
668 spin_unlock(&kvm_lock);
670 preempt_notifier_inc();
675 cleanup_srcu_struct(&kvm->irq_srcu);
677 cleanup_srcu_struct(&kvm->srcu);
679 hardware_disable_all();
681 for (i = 0; i < KVM_NR_BUSES; i++)
682 kfree(kvm->buses[i]);
683 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
684 kvm_free_memslots(kvm, kvm->memslots[i]);
685 kvm_arch_free_vm(kvm);
691 * Avoid using vmalloc for a small buffer.
692 * Should not be used when the size is statically known.
694 void *kvm_kvzalloc(unsigned long size)
696 if (size > PAGE_SIZE)
697 return vzalloc(size);
699 return kzalloc(size, GFP_KERNEL);
702 static void kvm_destroy_devices(struct kvm *kvm)
704 struct kvm_device *dev, *tmp;
707 * We do not need to take the kvm->lock here, because nobody else
708 * has a reference to the struct kvm at this point and therefore
709 * cannot access the devices list anyhow.
711 list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
712 list_del(&dev->vm_node);
713 dev->ops->destroy(dev);
717 static void kvm_destroy_vm(struct kvm *kvm)
720 struct mm_struct *mm = kvm->mm;
722 kvm_destroy_vm_debugfs(kvm);
723 kvm_arch_sync_events(kvm);
724 spin_lock(&kvm_lock);
725 list_del(&kvm->vm_list);
726 spin_unlock(&kvm_lock);
727 kvm_free_irq_routing(kvm);
728 for (i = 0; i < KVM_NR_BUSES; i++)
729 kvm_io_bus_destroy(kvm->buses[i]);
730 kvm_coalesced_mmio_free(kvm);
731 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
732 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
734 kvm_arch_flush_shadow_all(kvm);
736 kvm_arch_destroy_vm(kvm);
737 kvm_destroy_devices(kvm);
738 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
739 kvm_free_memslots(kvm, kvm->memslots[i]);
740 cleanup_srcu_struct(&kvm->irq_srcu);
741 cleanup_srcu_struct(&kvm->srcu);
742 kvm_arch_free_vm(kvm);
743 preempt_notifier_dec();
744 hardware_disable_all();
748 void kvm_get_kvm(struct kvm *kvm)
750 atomic_inc(&kvm->users_count);
752 EXPORT_SYMBOL_GPL(kvm_get_kvm);
754 void kvm_put_kvm(struct kvm *kvm)
756 if (atomic_dec_and_test(&kvm->users_count))
759 EXPORT_SYMBOL_GPL(kvm_put_kvm);
762 static int kvm_vm_release(struct inode *inode, struct file *filp)
764 struct kvm *kvm = filp->private_data;
766 kvm_irqfd_release(kvm);
773 * Allocation size is twice as large as the actual dirty bitmap size.
774 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
776 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
778 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
780 memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
781 if (!memslot->dirty_bitmap)
788 * Insert memslot and re-sort memslots based on their GFN,
789 * so binary search could be used to lookup GFN.
790 * Sorting algorithm takes advantage of having initially
791 * sorted array and known changed memslot position.
793 static void update_memslots(struct kvm_memslots *slots,
794 struct kvm_memory_slot *new)
797 int i = slots->id_to_index[id];
798 struct kvm_memory_slot *mslots = slots->memslots;
800 WARN_ON(mslots[i].id != id);
802 WARN_ON(!mslots[i].npages);
803 if (mslots[i].npages)
806 if (!mslots[i].npages)
810 while (i < KVM_MEM_SLOTS_NUM - 1 &&
811 new->base_gfn <= mslots[i + 1].base_gfn) {
812 if (!mslots[i + 1].npages)
814 mslots[i] = mslots[i + 1];
815 slots->id_to_index[mslots[i].id] = i;
820 * The ">=" is needed when creating a slot with base_gfn == 0,
821 * so that it moves before all those with base_gfn == npages == 0.
823 * On the other hand, if new->npages is zero, the above loop has
824 * already left i pointing to the beginning of the empty part of
825 * mslots, and the ">=" would move the hole backwards in this
826 * case---which is wrong. So skip the loop when deleting a slot.
830 new->base_gfn >= mslots[i - 1].base_gfn) {
831 mslots[i] = mslots[i - 1];
832 slots->id_to_index[mslots[i].id] = i;
836 WARN_ON_ONCE(i != slots->used_slots);
839 slots->id_to_index[mslots[i].id] = i;
842 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
844 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
846 #ifdef __KVM_HAVE_READONLY_MEM
847 valid_flags |= KVM_MEM_READONLY;
850 if (mem->flags & ~valid_flags)
856 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
857 int as_id, struct kvm_memslots *slots)
859 struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id);
862 * Set the low bit in the generation, which disables SPTE caching
863 * until the end of synchronize_srcu_expedited.
865 WARN_ON(old_memslots->generation & 1);
866 slots->generation = old_memslots->generation + 1;
868 rcu_assign_pointer(kvm->memslots[as_id], slots);
869 synchronize_srcu_expedited(&kvm->srcu);
872 * Increment the new memslot generation a second time. This prevents
873 * vm exits that race with memslot updates from caching a memslot
874 * generation that will (potentially) be valid forever.
876 * Generations must be unique even across address spaces. We do not need
877 * a global counter for that, instead the generation space is evenly split
878 * across address spaces. For example, with two address spaces, address
879 * space 0 will use generations 0, 4, 8, ... while * address space 1 will
880 * use generations 2, 6, 10, 14, ...
882 slots->generation += KVM_ADDRESS_SPACE_NUM * 2 - 1;
884 kvm_arch_memslots_updated(kvm, slots);
890 * Allocate some memory and give it an address in the guest physical address
893 * Discontiguous memory is allowed, mostly for framebuffers.
895 * Must be called holding kvm->slots_lock for write.
897 int __kvm_set_memory_region(struct kvm *kvm,
898 const struct kvm_userspace_memory_region *mem)
902 unsigned long npages;
903 struct kvm_memory_slot *slot;
904 struct kvm_memory_slot old, new;
905 struct kvm_memslots *slots = NULL, *old_memslots;
907 enum kvm_mr_change change;
909 r = check_memory_region_flags(mem);
914 as_id = mem->slot >> 16;
917 /* General sanity checks */
918 if (mem->memory_size & (PAGE_SIZE - 1))
920 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
922 /* We can read the guest memory with __xxx_user() later on. */
923 if ((id < KVM_USER_MEM_SLOTS) &&
924 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
925 !access_ok(VERIFY_WRITE,
926 (void __user *)(unsigned long)mem->userspace_addr,
929 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
931 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
934 slot = id_to_memslot(__kvm_memslots(kvm, as_id), id);
935 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
936 npages = mem->memory_size >> PAGE_SHIFT;
938 if (npages > KVM_MEM_MAX_NR_PAGES)
944 new.base_gfn = base_gfn;
946 new.flags = mem->flags;
950 change = KVM_MR_CREATE;
951 else { /* Modify an existing slot. */
952 if ((mem->userspace_addr != old.userspace_addr) ||
953 (npages != old.npages) ||
954 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
957 if (base_gfn != old.base_gfn)
958 change = KVM_MR_MOVE;
959 else if (new.flags != old.flags)
960 change = KVM_MR_FLAGS_ONLY;
961 else { /* Nothing to change. */
970 change = KVM_MR_DELETE;
975 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
976 /* Check for overlaps */
978 kvm_for_each_memslot(slot, __kvm_memslots(kvm, as_id)) {
979 if ((slot->id >= KVM_USER_MEM_SLOTS) ||
982 if (!((base_gfn + npages <= slot->base_gfn) ||
983 (base_gfn >= slot->base_gfn + slot->npages)))
988 /* Free page dirty bitmap if unneeded */
989 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
990 new.dirty_bitmap = NULL;
993 if (change == KVM_MR_CREATE) {
994 new.userspace_addr = mem->userspace_addr;
996 if (kvm_arch_create_memslot(kvm, &new, npages))
1000 /* Allocate page dirty bitmap if needed */
1001 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
1002 if (kvm_create_dirty_bitmap(&new) < 0)
1006 slots = kvm_kvzalloc(sizeof(struct kvm_memslots));
1009 memcpy(slots, __kvm_memslots(kvm, as_id), sizeof(struct kvm_memslots));
1011 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
1012 slot = id_to_memslot(slots, id);
1013 slot->flags |= KVM_MEMSLOT_INVALID;
1015 old_memslots = install_new_memslots(kvm, as_id, slots);
1017 /* slot was deleted or moved, clear iommu mapping */
1018 kvm_iommu_unmap_pages(kvm, &old);
1019 /* From this point no new shadow pages pointing to a deleted,
1020 * or moved, memslot will be created.
1022 * validation of sp->gfn happens in:
1023 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1024 * - kvm_is_visible_gfn (mmu_check_roots)
1026 kvm_arch_flush_shadow_memslot(kvm, slot);
1029 * We can re-use the old_memslots from above, the only difference
1030 * from the currently installed memslots is the invalid flag. This
1031 * will get overwritten by update_memslots anyway.
1033 slots = old_memslots;
1036 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
1040 /* actual memory is freed via old in kvm_free_memslot below */
1041 if (change == KVM_MR_DELETE) {
1042 new.dirty_bitmap = NULL;
1043 memset(&new.arch, 0, sizeof(new.arch));
1046 update_memslots(slots, &new);
1047 old_memslots = install_new_memslots(kvm, as_id, slots);
1049 kvm_arch_commit_memory_region(kvm, mem, &old, &new, change);
1051 kvm_free_memslot(kvm, &old, &new);
1052 kvfree(old_memslots);
1055 * IOMMU mapping: New slots need to be mapped. Old slots need to be
1056 * un-mapped and re-mapped if their base changes. Since base change
1057 * unmapping is handled above with slot deletion, mapping alone is
1058 * needed here. Anything else the iommu might care about for existing
1059 * slots (size changes, userspace addr changes and read-only flag
1060 * changes) is disallowed above, so any other attribute changes getting
1061 * here can be skipped.
1063 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
1064 r = kvm_iommu_map_pages(kvm, &new);
1073 kvm_free_memslot(kvm, &new, &old);
1077 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1079 int kvm_set_memory_region(struct kvm *kvm,
1080 const struct kvm_userspace_memory_region *mem)
1084 mutex_lock(&kvm->slots_lock);
1085 r = __kvm_set_memory_region(kvm, mem);
1086 mutex_unlock(&kvm->slots_lock);
1089 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1091 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1092 struct kvm_userspace_memory_region *mem)
1094 if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1097 return kvm_set_memory_region(kvm, mem);
1100 int kvm_get_dirty_log(struct kvm *kvm,
1101 struct kvm_dirty_log *log, int *is_dirty)
1103 struct kvm_memslots *slots;
1104 struct kvm_memory_slot *memslot;
1107 unsigned long any = 0;
1109 as_id = log->slot >> 16;
1110 id = (u16)log->slot;
1111 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1114 slots = __kvm_memslots(kvm, as_id);
1115 memslot = id_to_memslot(slots, id);
1116 if (!memslot->dirty_bitmap)
1119 n = kvm_dirty_bitmap_bytes(memslot);
1121 for (i = 0; !any && i < n/sizeof(long); ++i)
1122 any = memslot->dirty_bitmap[i];
1124 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1131 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1133 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1135 * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages
1136 * are dirty write protect them for next write.
1137 * @kvm: pointer to kvm instance
1138 * @log: slot id and address to which we copy the log
1139 * @is_dirty: flag set if any page is dirty
1141 * We need to keep it in mind that VCPU threads can write to the bitmap
1142 * concurrently. So, to avoid losing track of dirty pages we keep the
1145 * 1. Take a snapshot of the bit and clear it if needed.
1146 * 2. Write protect the corresponding page.
1147 * 3. Copy the snapshot to the userspace.
1148 * 4. Upon return caller flushes TLB's if needed.
1150 * Between 2 and 4, the guest may write to the page using the remaining TLB
1151 * entry. This is not a problem because the page is reported dirty using
1152 * the snapshot taken before and step 4 ensures that writes done after
1153 * exiting to userspace will be logged for the next call.
1156 int kvm_get_dirty_log_protect(struct kvm *kvm,
1157 struct kvm_dirty_log *log, bool *is_dirty)
1159 struct kvm_memslots *slots;
1160 struct kvm_memory_slot *memslot;
1163 unsigned long *dirty_bitmap;
1164 unsigned long *dirty_bitmap_buffer;
1166 as_id = log->slot >> 16;
1167 id = (u16)log->slot;
1168 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1171 slots = __kvm_memslots(kvm, as_id);
1172 memslot = id_to_memslot(slots, id);
1174 dirty_bitmap = memslot->dirty_bitmap;
1178 n = kvm_dirty_bitmap_bytes(memslot);
1180 dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long);
1181 memset(dirty_bitmap_buffer, 0, n);
1183 spin_lock(&kvm->mmu_lock);
1185 for (i = 0; i < n / sizeof(long); i++) {
1189 if (!dirty_bitmap[i])
1194 mask = xchg(&dirty_bitmap[i], 0);
1195 dirty_bitmap_buffer[i] = mask;
1198 offset = i * BITS_PER_LONG;
1199 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1204 spin_unlock(&kvm->mmu_lock);
1205 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1209 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect);
1212 bool kvm_largepages_enabled(void)
1214 return largepages_enabled;
1217 void kvm_disable_largepages(void)
1219 largepages_enabled = false;
1221 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1223 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1225 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1227 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1229 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
1231 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn);
1234 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1236 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1238 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1239 memslot->flags & KVM_MEMSLOT_INVALID)
1244 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1246 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1248 struct vm_area_struct *vma;
1249 unsigned long addr, size;
1253 addr = gfn_to_hva(kvm, gfn);
1254 if (kvm_is_error_hva(addr))
1257 down_read(¤t->mm->mmap_sem);
1258 vma = find_vma(current->mm, addr);
1262 size = vma_kernel_pagesize(vma);
1265 up_read(¤t->mm->mmap_sem);
1270 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1272 return slot->flags & KVM_MEM_READONLY;
1275 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1276 gfn_t *nr_pages, bool write)
1278 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1279 return KVM_HVA_ERR_BAD;
1281 if (memslot_is_readonly(slot) && write)
1282 return KVM_HVA_ERR_RO_BAD;
1285 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1287 return __gfn_to_hva_memslot(slot, gfn);
1290 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1293 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1296 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1299 return gfn_to_hva_many(slot, gfn, NULL);
1301 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1303 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1305 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1307 EXPORT_SYMBOL_GPL(gfn_to_hva);
1309 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
1311 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
1313 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
1316 * If writable is set to false, the hva returned by this function is only
1317 * allowed to be read.
1319 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1320 gfn_t gfn, bool *writable)
1322 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1324 if (!kvm_is_error_hva(hva) && writable)
1325 *writable = !memslot_is_readonly(slot);
1330 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1332 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1334 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1337 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
1339 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1341 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1344 static int get_user_page_nowait(unsigned long start, int write,
1347 int flags = FOLL_NOWAIT | FOLL_HWPOISON;
1350 flags |= FOLL_WRITE;
1352 return get_user_pages(start, 1, flags, page, NULL);
1355 static inline int check_user_page_hwpoison(unsigned long addr)
1357 int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
1359 rc = get_user_pages(addr, 1, flags, NULL, NULL);
1360 return rc == -EHWPOISON;
1364 * The atomic path to get the writable pfn which will be stored in @pfn,
1365 * true indicates success, otherwise false is returned.
1367 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1368 bool write_fault, bool *writable, kvm_pfn_t *pfn)
1370 struct page *page[1];
1373 if (!(async || atomic))
1377 * Fast pin a writable pfn only if it is a write fault request
1378 * or the caller allows to map a writable pfn for a read fault
1381 if (!(write_fault || writable))
1384 npages = __get_user_pages_fast(addr, 1, 1, page);
1386 *pfn = page_to_pfn(page[0]);
1397 * The slow path to get the pfn of the specified host virtual address,
1398 * 1 indicates success, -errno is returned if error is detected.
1400 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1401 bool *writable, kvm_pfn_t *pfn)
1403 struct page *page[1];
1409 *writable = write_fault;
1412 down_read(¤t->mm->mmap_sem);
1413 npages = get_user_page_nowait(addr, write_fault, page);
1414 up_read(¤t->mm->mmap_sem);
1416 unsigned int flags = FOLL_HWPOISON;
1419 flags |= FOLL_WRITE;
1421 npages = get_user_pages_unlocked(addr, 1, page, flags);
1426 /* map read fault as writable if possible */
1427 if (unlikely(!write_fault) && writable) {
1428 struct page *wpage[1];
1430 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1439 *pfn = page_to_pfn(page[0]);
1443 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1445 if (unlikely(!(vma->vm_flags & VM_READ)))
1448 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1454 static int hva_to_pfn_remapped(struct vm_area_struct *vma,
1455 unsigned long addr, bool *async,
1456 bool write_fault, kvm_pfn_t *p_pfn)
1461 r = follow_pfn(vma, addr, &pfn);
1464 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
1465 * not call the fault handler, so do it here.
1467 bool unlocked = false;
1468 r = fixup_user_fault(current, current->mm, addr,
1469 (write_fault ? FAULT_FLAG_WRITE : 0),
1476 r = follow_pfn(vma, addr, &pfn);
1484 * Get a reference here because callers of *hva_to_pfn* and
1485 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
1486 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
1487 * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will
1488 * simply do nothing for reserved pfns.
1490 * Whoever called remap_pfn_range is also going to call e.g.
1491 * unmap_mapping_range before the underlying pages are freed,
1492 * causing a call to our MMU notifier.
1501 * Pin guest page in memory and return its pfn.
1502 * @addr: host virtual address which maps memory to the guest
1503 * @atomic: whether this function can sleep
1504 * @async: whether this function need to wait IO complete if the
1505 * host page is not in the memory
1506 * @write_fault: whether we should get a writable host page
1507 * @writable: whether it allows to map a writable host page for !@write_fault
1509 * The function will map a writable host page for these two cases:
1510 * 1): @write_fault = true
1511 * 2): @write_fault = false && @writable, @writable will tell the caller
1512 * whether the mapping is writable.
1514 static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1515 bool write_fault, bool *writable)
1517 struct vm_area_struct *vma;
1521 /* we can do it either atomically or asynchronously, not both */
1522 BUG_ON(atomic && async);
1524 if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1528 return KVM_PFN_ERR_FAULT;
1530 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1534 down_read(¤t->mm->mmap_sem);
1535 if (npages == -EHWPOISON ||
1536 (!async && check_user_page_hwpoison(addr))) {
1537 pfn = KVM_PFN_ERR_HWPOISON;
1542 vma = find_vma_intersection(current->mm, addr, addr + 1);
1545 pfn = KVM_PFN_ERR_FAULT;
1546 else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
1547 r = hva_to_pfn_remapped(vma, addr, async, write_fault, &pfn);
1551 pfn = KVM_PFN_ERR_FAULT;
1553 if (async && vma_is_valid(vma, write_fault))
1555 pfn = KVM_PFN_ERR_FAULT;
1558 up_read(¤t->mm->mmap_sem);
1562 kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn,
1563 bool atomic, bool *async, bool write_fault,
1566 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1568 if (addr == KVM_HVA_ERR_RO_BAD) {
1571 return KVM_PFN_ERR_RO_FAULT;
1574 if (kvm_is_error_hva(addr)) {
1577 return KVM_PFN_NOSLOT;
1580 /* Do not map writable pfn in the readonly memslot. */
1581 if (writable && memslot_is_readonly(slot)) {
1586 return hva_to_pfn(addr, atomic, async, write_fault,
1589 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
1591 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1594 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
1595 write_fault, writable);
1597 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1599 kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1601 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1603 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
1605 kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1607 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1609 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1611 kvm_pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1613 return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm, gfn), gfn);
1615 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1617 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
1619 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1621 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
1623 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1625 return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
1627 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1629 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
1631 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1633 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
1635 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1636 struct page **pages, int nr_pages)
1641 addr = gfn_to_hva_many(slot, gfn, &entry);
1642 if (kvm_is_error_hva(addr))
1645 if (entry < nr_pages)
1648 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1650 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1652 static struct page *kvm_pfn_to_page(kvm_pfn_t pfn)
1654 if (is_error_noslot_pfn(pfn))
1655 return KVM_ERR_PTR_BAD_PAGE;
1657 if (kvm_is_reserved_pfn(pfn)) {
1659 return KVM_ERR_PTR_BAD_PAGE;
1662 return pfn_to_page(pfn);
1665 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1669 pfn = gfn_to_pfn(kvm, gfn);
1671 return kvm_pfn_to_page(pfn);
1673 EXPORT_SYMBOL_GPL(gfn_to_page);
1675 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
1679 pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
1681 return kvm_pfn_to_page(pfn);
1683 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
1685 void kvm_release_page_clean(struct page *page)
1687 WARN_ON(is_error_page(page));
1689 kvm_release_pfn_clean(page_to_pfn(page));
1691 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1693 void kvm_release_pfn_clean(kvm_pfn_t pfn)
1695 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
1696 put_page(pfn_to_page(pfn));
1698 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1700 void kvm_release_page_dirty(struct page *page)
1702 WARN_ON(is_error_page(page));
1704 kvm_release_pfn_dirty(page_to_pfn(page));
1706 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1708 static void kvm_release_pfn_dirty(kvm_pfn_t pfn)
1710 kvm_set_pfn_dirty(pfn);
1711 kvm_release_pfn_clean(pfn);
1714 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
1716 if (!kvm_is_reserved_pfn(pfn)) {
1717 struct page *page = pfn_to_page(pfn);
1719 if (!PageReserved(page))
1723 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1725 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
1727 if (!kvm_is_reserved_pfn(pfn))
1728 mark_page_accessed(pfn_to_page(pfn));
1730 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1732 void kvm_get_pfn(kvm_pfn_t pfn)
1734 if (!kvm_is_reserved_pfn(pfn))
1735 get_page(pfn_to_page(pfn));
1737 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1739 static int next_segment(unsigned long len, int offset)
1741 if (len > PAGE_SIZE - offset)
1742 return PAGE_SIZE - offset;
1747 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
1748 void *data, int offset, int len)
1753 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1754 if (kvm_is_error_hva(addr))
1756 r = __copy_from_user(data, (void __user *)addr + offset, len);
1762 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1765 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1767 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1769 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1771 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
1772 int offset, int len)
1774 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1776 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1778 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
1780 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1782 gfn_t gfn = gpa >> PAGE_SHIFT;
1784 int offset = offset_in_page(gpa);
1787 while ((seg = next_segment(len, offset)) != 0) {
1788 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1798 EXPORT_SYMBOL_GPL(kvm_read_guest);
1800 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
1802 gfn_t gfn = gpa >> PAGE_SHIFT;
1804 int offset = offset_in_page(gpa);
1807 while ((seg = next_segment(len, offset)) != 0) {
1808 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
1818 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
1820 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1821 void *data, int offset, unsigned long len)
1826 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1827 if (kvm_is_error_hva(addr))
1829 pagefault_disable();
1830 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1837 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1840 gfn_t gfn = gpa >> PAGE_SHIFT;
1841 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1842 int offset = offset_in_page(gpa);
1844 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1846 EXPORT_SYMBOL_GPL(kvm_read_guest_atomic);
1848 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
1849 void *data, unsigned long len)
1851 gfn_t gfn = gpa >> PAGE_SHIFT;
1852 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1853 int offset = offset_in_page(gpa);
1855 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1857 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
1859 static int __kvm_write_guest_page(struct kvm_memory_slot *memslot, gfn_t gfn,
1860 const void *data, int offset, int len)
1865 addr = gfn_to_hva_memslot(memslot, gfn);
1866 if (kvm_is_error_hva(addr))
1868 r = __copy_to_user((void __user *)addr + offset, data, len);
1871 mark_page_dirty_in_slot(memslot, gfn);
1875 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
1876 const void *data, int offset, int len)
1878 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1880 return __kvm_write_guest_page(slot, gfn, data, offset, len);
1882 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1884 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
1885 const void *data, int offset, int len)
1887 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1889 return __kvm_write_guest_page(slot, gfn, data, offset, len);
1891 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
1893 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1896 gfn_t gfn = gpa >> PAGE_SHIFT;
1898 int offset = offset_in_page(gpa);
1901 while ((seg = next_segment(len, offset)) != 0) {
1902 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1912 EXPORT_SYMBOL_GPL(kvm_write_guest);
1914 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
1917 gfn_t gfn = gpa >> PAGE_SHIFT;
1919 int offset = offset_in_page(gpa);
1922 while ((seg = next_segment(len, offset)) != 0) {
1923 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
1933 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
1935 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
1936 struct gfn_to_hva_cache *ghc,
1937 gpa_t gpa, unsigned long len)
1939 int offset = offset_in_page(gpa);
1940 gfn_t start_gfn = gpa >> PAGE_SHIFT;
1941 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1942 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1943 gfn_t nr_pages_avail;
1946 ghc->generation = slots->generation;
1948 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
1949 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, NULL);
1950 if (!kvm_is_error_hva(ghc->hva) && nr_pages_needed <= 1) {
1954 * If the requested region crosses two memslots, we still
1955 * verify that the entire region is valid here.
1957 while (start_gfn <= end_gfn) {
1958 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
1959 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1961 if (kvm_is_error_hva(ghc->hva))
1963 start_gfn += nr_pages_avail;
1965 /* Use the slow path for cross page reads and writes. */
1966 ghc->memslot = NULL;
1971 int kvm_vcpu_gfn_to_hva_cache_init(struct kvm_vcpu *vcpu, struct gfn_to_hva_cache *ghc,
1972 gpa_t gpa, unsigned long len)
1974 struct kvm_memslots *slots = kvm_vcpu_memslots(vcpu);
1975 return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len);
1977 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva_cache_init);
1979 int kvm_vcpu_write_guest_offset_cached(struct kvm_vcpu *vcpu, struct gfn_to_hva_cache *ghc,
1980 void *data, int offset, unsigned long len)
1982 struct kvm_memslots *slots = kvm_vcpu_memslots(vcpu);
1984 gpa_t gpa = ghc->gpa + offset;
1986 BUG_ON(len + offset > ghc->len);
1988 if (slots->generation != ghc->generation)
1989 __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len);
1991 if (unlikely(!ghc->memslot))
1992 return kvm_vcpu_write_guest(vcpu, gpa, data, len);
1994 if (kvm_is_error_hva(ghc->hva))
1997 r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
2000 mark_page_dirty_in_slot(ghc->memslot, gpa >> PAGE_SHIFT);
2004 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_offset_cached);
2006 int kvm_vcpu_write_guest_cached(struct kvm_vcpu *vcpu, struct gfn_to_hva_cache *ghc,
2007 void *data, unsigned long len)
2009 return kvm_vcpu_write_guest_offset_cached(vcpu, ghc, data, 0, len);
2011 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_cached);
2013 int kvm_vcpu_read_guest_cached(struct kvm_vcpu *vcpu, struct gfn_to_hva_cache *ghc,
2014 void *data, unsigned long len)
2016 struct kvm_memslots *slots = kvm_vcpu_memslots(vcpu);
2019 BUG_ON(len > ghc->len);
2021 if (slots->generation != ghc->generation)
2022 __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len);
2024 if (unlikely(!ghc->memslot))
2025 return kvm_vcpu_read_guest(vcpu, ghc->gpa, data, len);
2027 if (kvm_is_error_hva(ghc->hva))
2030 r = __copy_from_user(data, (void __user *)ghc->hva, len);
2036 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_cached);
2038 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
2040 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
2042 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
2044 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
2046 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
2048 gfn_t gfn = gpa >> PAGE_SHIFT;
2050 int offset = offset_in_page(gpa);
2053 while ((seg = next_segment(len, offset)) != 0) {
2054 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
2063 EXPORT_SYMBOL_GPL(kvm_clear_guest);
2065 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot,
2068 if (memslot && memslot->dirty_bitmap) {
2069 unsigned long rel_gfn = gfn - memslot->base_gfn;
2071 set_bit_le(rel_gfn, memslot->dirty_bitmap);
2075 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
2077 struct kvm_memory_slot *memslot;
2079 memslot = gfn_to_memslot(kvm, gfn);
2080 mark_page_dirty_in_slot(memslot, gfn);
2082 EXPORT_SYMBOL_GPL(mark_page_dirty);
2084 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
2086 struct kvm_memory_slot *memslot;
2088 memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2089 mark_page_dirty_in_slot(memslot, gfn);
2091 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
2093 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
2095 unsigned int old, val, grow;
2097 old = val = vcpu->halt_poll_ns;
2098 grow = READ_ONCE(halt_poll_ns_grow);
2100 if (val == 0 && grow)
2105 if (val > halt_poll_ns)
2108 vcpu->halt_poll_ns = val;
2109 trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
2112 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
2114 unsigned int old, val, shrink;
2116 old = val = vcpu->halt_poll_ns;
2117 shrink = READ_ONCE(halt_poll_ns_shrink);
2123 vcpu->halt_poll_ns = val;
2124 trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
2127 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
2129 if (kvm_arch_vcpu_runnable(vcpu)) {
2130 kvm_make_request(KVM_REQ_UNHALT, vcpu);
2133 if (kvm_cpu_has_pending_timer(vcpu))
2135 if (signal_pending(current))
2142 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2144 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
2147 DECLARE_SWAITQUEUE(wait);
2148 bool waited = false;
2151 start = cur = ktime_get();
2152 if (vcpu->halt_poll_ns) {
2153 ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns);
2155 ++vcpu->stat.halt_attempted_poll;
2158 * This sets KVM_REQ_UNHALT if an interrupt
2161 if (kvm_vcpu_check_block(vcpu) < 0) {
2162 ++vcpu->stat.halt_successful_poll;
2163 if (!vcpu_valid_wakeup(vcpu))
2164 ++vcpu->stat.halt_poll_invalid;
2168 } while (single_task_running() && ktime_before(cur, stop));
2171 kvm_arch_vcpu_blocking(vcpu);
2174 prepare_to_swait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
2176 if (kvm_vcpu_check_block(vcpu) < 0)
2183 finish_swait(&vcpu->wq, &wait);
2186 kvm_arch_vcpu_unblocking(vcpu);
2188 block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
2190 if (!vcpu_valid_wakeup(vcpu))
2191 shrink_halt_poll_ns(vcpu);
2192 else if (halt_poll_ns) {
2193 if (block_ns <= vcpu->halt_poll_ns)
2195 /* we had a long block, shrink polling */
2196 else if (vcpu->halt_poll_ns && block_ns > halt_poll_ns)
2197 shrink_halt_poll_ns(vcpu);
2198 /* we had a short halt and our poll time is too small */
2199 else if (vcpu->halt_poll_ns < halt_poll_ns &&
2200 block_ns < halt_poll_ns)
2201 grow_halt_poll_ns(vcpu);
2203 vcpu->halt_poll_ns = 0;
2205 trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu));
2206 kvm_arch_vcpu_block_finish(vcpu);
2208 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
2211 void kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
2213 struct swait_queue_head *wqp;
2215 wqp = kvm_arch_vcpu_wq(vcpu);
2216 if (swait_active(wqp)) {
2218 ++vcpu->stat.halt_wakeup;
2222 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
2225 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2227 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
2230 int cpu = vcpu->cpu;
2232 kvm_vcpu_wake_up(vcpu);
2234 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
2235 if (kvm_arch_vcpu_should_kick(vcpu))
2236 smp_send_reschedule(cpu);
2239 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
2240 #endif /* !CONFIG_S390 */
2242 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
2245 struct task_struct *task = NULL;
2249 pid = rcu_dereference(target->pid);
2251 task = get_pid_task(pid, PIDTYPE_PID);
2255 ret = yield_to(task, 1);
2256 put_task_struct(task);
2260 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
2263 * Helper that checks whether a VCPU is eligible for directed yield.
2264 * Most eligible candidate to yield is decided by following heuristics:
2266 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2267 * (preempted lock holder), indicated by @in_spin_loop.
2268 * Set at the beiginning and cleared at the end of interception/PLE handler.
2270 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2271 * chance last time (mostly it has become eligible now since we have probably
2272 * yielded to lockholder in last iteration. This is done by toggling
2273 * @dy_eligible each time a VCPU checked for eligibility.)
2275 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2276 * to preempted lock-holder could result in wrong VCPU selection and CPU
2277 * burning. Giving priority for a potential lock-holder increases lock
2280 * Since algorithm is based on heuristics, accessing another VCPU data without
2281 * locking does not harm. It may result in trying to yield to same VCPU, fail
2282 * and continue with next VCPU and so on.
2284 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
2286 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2289 eligible = !vcpu->spin_loop.in_spin_loop ||
2290 vcpu->spin_loop.dy_eligible;
2292 if (vcpu->spin_loop.in_spin_loop)
2293 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
2301 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
2303 struct kvm *kvm = me->kvm;
2304 struct kvm_vcpu *vcpu;
2305 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
2311 kvm_vcpu_set_in_spin_loop(me, true);
2313 * We boost the priority of a VCPU that is runnable but not
2314 * currently running, because it got preempted by something
2315 * else and called schedule in __vcpu_run. Hopefully that
2316 * VCPU is holding the lock that we need and will release it.
2317 * We approximate round-robin by starting at the last boosted VCPU.
2319 for (pass = 0; pass < 2 && !yielded && try; pass++) {
2320 kvm_for_each_vcpu(i, vcpu, kvm) {
2321 if (!pass && i <= last_boosted_vcpu) {
2322 i = last_boosted_vcpu;
2324 } else if (pass && i > last_boosted_vcpu)
2326 if (!ACCESS_ONCE(vcpu->preempted))
2330 if (swait_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
2332 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
2335 yielded = kvm_vcpu_yield_to(vcpu);
2337 kvm->last_boosted_vcpu = i;
2339 } else if (yielded < 0) {
2346 kvm_vcpu_set_in_spin_loop(me, false);
2348 /* Ensure vcpu is not eligible during next spinloop */
2349 kvm_vcpu_set_dy_eligible(me, false);
2351 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
2353 static int kvm_vcpu_fault(struct vm_fault *vmf)
2355 struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data;
2358 if (vmf->pgoff == 0)
2359 page = virt_to_page(vcpu->run);
2361 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
2362 page = virt_to_page(vcpu->arch.pio_data);
2364 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2365 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
2366 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
2369 return kvm_arch_vcpu_fault(vcpu, vmf);
2375 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
2376 .fault = kvm_vcpu_fault,
2379 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2381 vma->vm_ops = &kvm_vcpu_vm_ops;
2385 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2387 struct kvm_vcpu *vcpu = filp->private_data;
2389 debugfs_remove_recursive(vcpu->debugfs_dentry);
2390 kvm_put_kvm(vcpu->kvm);
2394 static struct file_operations kvm_vcpu_fops = {
2395 .release = kvm_vcpu_release,
2396 .unlocked_ioctl = kvm_vcpu_ioctl,
2397 #ifdef CONFIG_KVM_COMPAT
2398 .compat_ioctl = kvm_vcpu_compat_ioctl,
2400 .mmap = kvm_vcpu_mmap,
2401 .llseek = noop_llseek,
2405 * Allocates an inode for the vcpu.
2407 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2409 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
2412 static int kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
2414 char dir_name[ITOA_MAX_LEN * 2];
2417 if (!kvm_arch_has_vcpu_debugfs())
2420 if (!debugfs_initialized())
2423 snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
2424 vcpu->debugfs_dentry = debugfs_create_dir(dir_name,
2425 vcpu->kvm->debugfs_dentry);
2426 if (!vcpu->debugfs_dentry)
2429 ret = kvm_arch_create_vcpu_debugfs(vcpu);
2431 debugfs_remove_recursive(vcpu->debugfs_dentry);
2439 * Creates some virtual cpus. Good luck creating more than one.
2441 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
2444 struct kvm_vcpu *vcpu;
2446 if (id >= KVM_MAX_VCPU_ID)
2449 mutex_lock(&kvm->lock);
2450 if (kvm->created_vcpus == KVM_MAX_VCPUS) {
2451 mutex_unlock(&kvm->lock);
2455 kvm->created_vcpus++;
2456 mutex_unlock(&kvm->lock);
2458 vcpu = kvm_arch_vcpu_create(kvm, id);
2461 goto vcpu_decrement;
2464 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2466 r = kvm_arch_vcpu_setup(vcpu);
2470 r = kvm_create_vcpu_debugfs(vcpu);
2474 mutex_lock(&kvm->lock);
2475 if (kvm_get_vcpu_by_id(kvm, id)) {
2477 goto unlock_vcpu_destroy;
2480 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
2482 /* Now it's all set up, let userspace reach it */
2484 r = create_vcpu_fd(vcpu);
2487 goto unlock_vcpu_destroy;
2490 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
2493 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
2494 * before kvm->online_vcpu's incremented value.
2497 atomic_inc(&kvm->online_vcpus);
2499 mutex_unlock(&kvm->lock);
2500 kvm_arch_vcpu_postcreate(vcpu);
2503 unlock_vcpu_destroy:
2504 mutex_unlock(&kvm->lock);
2505 debugfs_remove_recursive(vcpu->debugfs_dentry);
2507 kvm_arch_vcpu_destroy(vcpu);
2509 mutex_lock(&kvm->lock);
2510 kvm->created_vcpus--;
2511 mutex_unlock(&kvm->lock);
2515 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2518 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2519 vcpu->sigset_active = 1;
2520 vcpu->sigset = *sigset;
2522 vcpu->sigset_active = 0;
2526 static long kvm_vcpu_ioctl(struct file *filp,
2527 unsigned int ioctl, unsigned long arg)
2529 struct kvm_vcpu *vcpu = filp->private_data;
2530 void __user *argp = (void __user *)arg;
2532 struct kvm_fpu *fpu = NULL;
2533 struct kvm_sregs *kvm_sregs = NULL;
2535 if (vcpu->kvm->mm != current->mm)
2538 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2541 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
2543 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
2544 * so vcpu_load() would break it.
2546 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_S390_IRQ || ioctl == KVM_INTERRUPT)
2547 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2551 r = vcpu_load(vcpu);
2559 if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
2560 /* The thread running this VCPU changed. */
2561 struct pid *oldpid = vcpu->pid;
2562 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
2564 rcu_assign_pointer(vcpu->pid, newpid);
2569 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2570 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2572 case KVM_GET_REGS: {
2573 struct kvm_regs *kvm_regs;
2576 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2579 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2583 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2590 case KVM_SET_REGS: {
2591 struct kvm_regs *kvm_regs;
2594 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2595 if (IS_ERR(kvm_regs)) {
2596 r = PTR_ERR(kvm_regs);
2599 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2603 case KVM_GET_SREGS: {
2604 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2608 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2612 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2617 case KVM_SET_SREGS: {
2618 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2619 if (IS_ERR(kvm_sregs)) {
2620 r = PTR_ERR(kvm_sregs);
2624 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2627 case KVM_GET_MP_STATE: {
2628 struct kvm_mp_state mp_state;
2630 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2634 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
2639 case KVM_SET_MP_STATE: {
2640 struct kvm_mp_state mp_state;
2643 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
2645 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2648 case KVM_TRANSLATE: {
2649 struct kvm_translation tr;
2652 if (copy_from_user(&tr, argp, sizeof(tr)))
2654 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2658 if (copy_to_user(argp, &tr, sizeof(tr)))
2663 case KVM_SET_GUEST_DEBUG: {
2664 struct kvm_guest_debug dbg;
2667 if (copy_from_user(&dbg, argp, sizeof(dbg)))
2669 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2672 case KVM_SET_SIGNAL_MASK: {
2673 struct kvm_signal_mask __user *sigmask_arg = argp;
2674 struct kvm_signal_mask kvm_sigmask;
2675 sigset_t sigset, *p;
2680 if (copy_from_user(&kvm_sigmask, argp,
2681 sizeof(kvm_sigmask)))
2684 if (kvm_sigmask.len != sizeof(sigset))
2687 if (copy_from_user(&sigset, sigmask_arg->sigset,
2692 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2696 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2700 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2704 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2710 fpu = memdup_user(argp, sizeof(*fpu));
2716 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2720 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2729 #ifdef CONFIG_KVM_COMPAT
2730 static long kvm_vcpu_compat_ioctl(struct file *filp,
2731 unsigned int ioctl, unsigned long arg)
2733 struct kvm_vcpu *vcpu = filp->private_data;
2734 void __user *argp = compat_ptr(arg);
2737 if (vcpu->kvm->mm != current->mm)
2741 case KVM_SET_SIGNAL_MASK: {
2742 struct kvm_signal_mask __user *sigmask_arg = argp;
2743 struct kvm_signal_mask kvm_sigmask;
2744 compat_sigset_t csigset;
2749 if (copy_from_user(&kvm_sigmask, argp,
2750 sizeof(kvm_sigmask)))
2753 if (kvm_sigmask.len != sizeof(csigset))
2756 if (copy_from_user(&csigset, sigmask_arg->sigset,
2759 sigset_from_compat(&sigset, &csigset);
2760 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2762 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2766 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2774 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2775 int (*accessor)(struct kvm_device *dev,
2776 struct kvm_device_attr *attr),
2779 struct kvm_device_attr attr;
2784 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2787 return accessor(dev, &attr);
2790 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2793 struct kvm_device *dev = filp->private_data;
2796 case KVM_SET_DEVICE_ATTR:
2797 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2798 case KVM_GET_DEVICE_ATTR:
2799 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2800 case KVM_HAS_DEVICE_ATTR:
2801 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2803 if (dev->ops->ioctl)
2804 return dev->ops->ioctl(dev, ioctl, arg);
2810 static int kvm_device_release(struct inode *inode, struct file *filp)
2812 struct kvm_device *dev = filp->private_data;
2813 struct kvm *kvm = dev->kvm;
2819 static const struct file_operations kvm_device_fops = {
2820 .unlocked_ioctl = kvm_device_ioctl,
2821 #ifdef CONFIG_KVM_COMPAT
2822 .compat_ioctl = kvm_device_ioctl,
2824 .release = kvm_device_release,
2827 struct kvm_device *kvm_device_from_filp(struct file *filp)
2829 if (filp->f_op != &kvm_device_fops)
2832 return filp->private_data;
2835 static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
2836 #ifdef CONFIG_KVM_MPIC
2837 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
2838 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
2841 #ifdef CONFIG_KVM_XICS
2842 [KVM_DEV_TYPE_XICS] = &kvm_xics_ops,
2846 int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
2848 if (type >= ARRAY_SIZE(kvm_device_ops_table))
2851 if (kvm_device_ops_table[type] != NULL)
2854 kvm_device_ops_table[type] = ops;
2858 void kvm_unregister_device_ops(u32 type)
2860 if (kvm_device_ops_table[type] != NULL)
2861 kvm_device_ops_table[type] = NULL;
2864 static int kvm_ioctl_create_device(struct kvm *kvm,
2865 struct kvm_create_device *cd)
2867 struct kvm_device_ops *ops = NULL;
2868 struct kvm_device *dev;
2869 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2872 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
2875 ops = kvm_device_ops_table[cd->type];
2882 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2889 mutex_lock(&kvm->lock);
2890 ret = ops->create(dev, cd->type);
2892 mutex_unlock(&kvm->lock);
2896 list_add(&dev->vm_node, &kvm->devices);
2897 mutex_unlock(&kvm->lock);
2902 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2904 mutex_lock(&kvm->lock);
2905 list_del(&dev->vm_node);
2906 mutex_unlock(&kvm->lock);
2916 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2919 case KVM_CAP_USER_MEMORY:
2920 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2921 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2922 case KVM_CAP_INTERNAL_ERROR_DATA:
2923 #ifdef CONFIG_HAVE_KVM_MSI
2924 case KVM_CAP_SIGNAL_MSI:
2926 #ifdef CONFIG_HAVE_KVM_IRQFD
2928 case KVM_CAP_IRQFD_RESAMPLE:
2930 case KVM_CAP_IOEVENTFD_ANY_LENGTH:
2931 case KVM_CAP_CHECK_EXTENSION_VM:
2933 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2934 case KVM_CAP_IRQ_ROUTING:
2935 return KVM_MAX_IRQ_ROUTES;
2937 #if KVM_ADDRESS_SPACE_NUM > 1
2938 case KVM_CAP_MULTI_ADDRESS_SPACE:
2939 return KVM_ADDRESS_SPACE_NUM;
2941 case KVM_CAP_MAX_VCPU_ID:
2942 return KVM_MAX_VCPU_ID;
2946 return kvm_vm_ioctl_check_extension(kvm, arg);
2949 static long kvm_vm_ioctl(struct file *filp,
2950 unsigned int ioctl, unsigned long arg)
2952 struct kvm *kvm = filp->private_data;
2953 void __user *argp = (void __user *)arg;
2956 if (kvm->mm != current->mm)
2959 case KVM_CREATE_VCPU:
2960 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2962 case KVM_SET_USER_MEMORY_REGION: {
2963 struct kvm_userspace_memory_region kvm_userspace_mem;
2966 if (copy_from_user(&kvm_userspace_mem, argp,
2967 sizeof(kvm_userspace_mem)))
2970 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2973 case KVM_GET_DIRTY_LOG: {
2974 struct kvm_dirty_log log;
2977 if (copy_from_user(&log, argp, sizeof(log)))
2979 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2982 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2983 case KVM_REGISTER_COALESCED_MMIO: {
2984 struct kvm_coalesced_mmio_zone zone;
2987 if (copy_from_user(&zone, argp, sizeof(zone)))
2989 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2992 case KVM_UNREGISTER_COALESCED_MMIO: {
2993 struct kvm_coalesced_mmio_zone zone;
2996 if (copy_from_user(&zone, argp, sizeof(zone)))
2998 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
3003 struct kvm_irqfd data;
3006 if (copy_from_user(&data, argp, sizeof(data)))
3008 r = kvm_irqfd(kvm, &data);
3011 case KVM_IOEVENTFD: {
3012 struct kvm_ioeventfd data;
3015 if (copy_from_user(&data, argp, sizeof(data)))
3017 r = kvm_ioeventfd(kvm, &data);
3020 #ifdef CONFIG_HAVE_KVM_MSI
3021 case KVM_SIGNAL_MSI: {
3025 if (copy_from_user(&msi, argp, sizeof(msi)))
3027 r = kvm_send_userspace_msi(kvm, &msi);
3031 #ifdef __KVM_HAVE_IRQ_LINE
3032 case KVM_IRQ_LINE_STATUS:
3033 case KVM_IRQ_LINE: {
3034 struct kvm_irq_level irq_event;
3037 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
3040 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
3041 ioctl == KVM_IRQ_LINE_STATUS);
3046 if (ioctl == KVM_IRQ_LINE_STATUS) {
3047 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
3055 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3056 case KVM_SET_GSI_ROUTING: {
3057 struct kvm_irq_routing routing;
3058 struct kvm_irq_routing __user *urouting;
3059 struct kvm_irq_routing_entry *entries = NULL;
3062 if (copy_from_user(&routing, argp, sizeof(routing)))
3065 if (routing.nr > KVM_MAX_IRQ_ROUTES)
3071 entries = vmalloc(routing.nr * sizeof(*entries));
3076 if (copy_from_user(entries, urouting->entries,
3077 routing.nr * sizeof(*entries)))
3078 goto out_free_irq_routing;
3080 r = kvm_set_irq_routing(kvm, entries, routing.nr,
3082 out_free_irq_routing:
3086 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
3087 case KVM_CREATE_DEVICE: {
3088 struct kvm_create_device cd;
3091 if (copy_from_user(&cd, argp, sizeof(cd)))
3094 r = kvm_ioctl_create_device(kvm, &cd);
3099 if (copy_to_user(argp, &cd, sizeof(cd)))
3105 case KVM_CHECK_EXTENSION:
3106 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
3109 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
3115 #ifdef CONFIG_KVM_COMPAT
3116 struct compat_kvm_dirty_log {
3120 compat_uptr_t dirty_bitmap; /* one bit per page */
3125 static long kvm_vm_compat_ioctl(struct file *filp,
3126 unsigned int ioctl, unsigned long arg)
3128 struct kvm *kvm = filp->private_data;
3131 if (kvm->mm != current->mm)
3134 case KVM_GET_DIRTY_LOG: {
3135 struct compat_kvm_dirty_log compat_log;
3136 struct kvm_dirty_log log;
3138 if (copy_from_user(&compat_log, (void __user *)arg,
3139 sizeof(compat_log)))
3141 log.slot = compat_log.slot;
3142 log.padding1 = compat_log.padding1;
3143 log.padding2 = compat_log.padding2;
3144 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
3146 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3150 r = kvm_vm_ioctl(filp, ioctl, arg);
3156 static struct file_operations kvm_vm_fops = {
3157 .release = kvm_vm_release,
3158 .unlocked_ioctl = kvm_vm_ioctl,
3159 #ifdef CONFIG_KVM_COMPAT
3160 .compat_ioctl = kvm_vm_compat_ioctl,
3162 .llseek = noop_llseek,
3165 static int kvm_dev_ioctl_create_vm(unsigned long type)
3171 kvm = kvm_create_vm(type);
3173 return PTR_ERR(kvm);
3174 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
3175 r = kvm_coalesced_mmio_init(kvm);
3181 r = get_unused_fd_flags(O_CLOEXEC);
3186 file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
3190 return PTR_ERR(file);
3193 if (kvm_create_vm_debugfs(kvm, r) < 0) {
3199 fd_install(r, file);
3203 static long kvm_dev_ioctl(struct file *filp,
3204 unsigned int ioctl, unsigned long arg)
3209 case KVM_GET_API_VERSION:
3212 r = KVM_API_VERSION;
3215 r = kvm_dev_ioctl_create_vm(arg);
3217 case KVM_CHECK_EXTENSION:
3218 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
3220 case KVM_GET_VCPU_MMAP_SIZE:
3223 r = PAGE_SIZE; /* struct kvm_run */
3225 r += PAGE_SIZE; /* pio data page */
3227 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
3228 r += PAGE_SIZE; /* coalesced mmio ring page */
3231 case KVM_TRACE_ENABLE:
3232 case KVM_TRACE_PAUSE:
3233 case KVM_TRACE_DISABLE:
3237 return kvm_arch_dev_ioctl(filp, ioctl, arg);
3243 static struct file_operations kvm_chardev_ops = {
3244 .unlocked_ioctl = kvm_dev_ioctl,
3245 .compat_ioctl = kvm_dev_ioctl,
3246 .llseek = noop_llseek,
3249 static struct miscdevice kvm_dev = {
3255 static void hardware_enable_nolock(void *junk)
3257 int cpu = raw_smp_processor_id();
3260 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
3263 cpumask_set_cpu(cpu, cpus_hardware_enabled);
3265 r = kvm_arch_hardware_enable();
3268 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3269 atomic_inc(&hardware_enable_failed);
3270 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
3274 static int kvm_starting_cpu(unsigned int cpu)
3276 raw_spin_lock(&kvm_count_lock);
3277 if (kvm_usage_count)
3278 hardware_enable_nolock(NULL);
3279 raw_spin_unlock(&kvm_count_lock);
3283 static void hardware_disable_nolock(void *junk)
3285 int cpu = raw_smp_processor_id();
3287 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
3289 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3290 kvm_arch_hardware_disable();
3293 static int kvm_dying_cpu(unsigned int cpu)
3295 raw_spin_lock(&kvm_count_lock);
3296 if (kvm_usage_count)
3297 hardware_disable_nolock(NULL);
3298 raw_spin_unlock(&kvm_count_lock);
3302 static void hardware_disable_all_nolock(void)
3304 BUG_ON(!kvm_usage_count);
3307 if (!kvm_usage_count)
3308 on_each_cpu(hardware_disable_nolock, NULL, 1);
3311 static void hardware_disable_all(void)
3313 raw_spin_lock(&kvm_count_lock);
3314 hardware_disable_all_nolock();
3315 raw_spin_unlock(&kvm_count_lock);
3318 static int hardware_enable_all(void)
3322 raw_spin_lock(&kvm_count_lock);
3325 if (kvm_usage_count == 1) {
3326 atomic_set(&hardware_enable_failed, 0);
3327 on_each_cpu(hardware_enable_nolock, NULL, 1);
3329 if (atomic_read(&hardware_enable_failed)) {
3330 hardware_disable_all_nolock();
3335 raw_spin_unlock(&kvm_count_lock);
3340 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3344 * Some (well, at least mine) BIOSes hang on reboot if
3347 * And Intel TXT required VMX off for all cpu when system shutdown.
3349 pr_info("kvm: exiting hardware virtualization\n");
3350 kvm_rebooting = true;
3351 on_each_cpu(hardware_disable_nolock, NULL, 1);
3355 static struct notifier_block kvm_reboot_notifier = {
3356 .notifier_call = kvm_reboot,
3360 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3364 for (i = 0; i < bus->dev_count; i++) {
3365 struct kvm_io_device *pos = bus->range[i].dev;
3367 kvm_iodevice_destructor(pos);
3372 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
3373 const struct kvm_io_range *r2)
3375 gpa_t addr1 = r1->addr;
3376 gpa_t addr2 = r2->addr;
3381 /* If r2->len == 0, match the exact address. If r2->len != 0,
3382 * accept any overlapping write. Any order is acceptable for
3383 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
3384 * we process all of them.
3397 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
3399 return kvm_io_bus_cmp(p1, p2);
3402 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
3403 gpa_t addr, int len)
3405 bus->range[bus->dev_count++] = (struct kvm_io_range) {
3411 sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
3412 kvm_io_bus_sort_cmp, NULL);
3417 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
3418 gpa_t addr, int len)
3420 struct kvm_io_range *range, key;
3423 key = (struct kvm_io_range) {
3428 range = bsearch(&key, bus->range, bus->dev_count,
3429 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
3433 off = range - bus->range;
3435 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
3441 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3442 struct kvm_io_range *range, const void *val)
3446 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3450 while (idx < bus->dev_count &&
3451 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3452 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
3461 /* kvm_io_bus_write - called under kvm->slots_lock */
3462 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3463 int len, const void *val)
3465 struct kvm_io_bus *bus;
3466 struct kvm_io_range range;
3469 range = (struct kvm_io_range) {
3474 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3475 r = __kvm_io_bus_write(vcpu, bus, &range, val);
3476 return r < 0 ? r : 0;
3479 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3480 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
3481 gpa_t addr, int len, const void *val, long cookie)
3483 struct kvm_io_bus *bus;
3484 struct kvm_io_range range;
3486 range = (struct kvm_io_range) {
3491 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3493 /* First try the device referenced by cookie. */
3494 if ((cookie >= 0) && (cookie < bus->dev_count) &&
3495 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
3496 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
3501 * cookie contained garbage; fall back to search and return the
3502 * correct cookie value.
3504 return __kvm_io_bus_write(vcpu, bus, &range, val);
3507 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3508 struct kvm_io_range *range, void *val)
3512 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3516 while (idx < bus->dev_count &&
3517 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3518 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
3526 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
3528 /* kvm_io_bus_read - called under kvm->slots_lock */
3529 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3532 struct kvm_io_bus *bus;
3533 struct kvm_io_range range;
3536 range = (struct kvm_io_range) {
3541 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3542 r = __kvm_io_bus_read(vcpu, bus, &range, val);
3543 return r < 0 ? r : 0;
3547 /* Caller must hold slots_lock. */
3548 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3549 int len, struct kvm_io_device *dev)
3551 struct kvm_io_bus *new_bus, *bus;
3553 bus = kvm->buses[bus_idx];
3554 /* exclude ioeventfd which is limited by maximum fd */
3555 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3558 new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count + 1) *
3559 sizeof(struct kvm_io_range)), GFP_KERNEL);
3562 memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
3563 sizeof(struct kvm_io_range)));
3564 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
3565 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3566 synchronize_srcu_expedited(&kvm->srcu);
3572 /* Caller must hold slots_lock. */
3573 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3574 struct kvm_io_device *dev)
3577 struct kvm_io_bus *new_bus, *bus;
3579 bus = kvm->buses[bus_idx];
3581 for (i = 0; i < bus->dev_count; i++)
3582 if (bus->range[i].dev == dev) {
3590 new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3591 sizeof(struct kvm_io_range)), GFP_KERNEL);
3595 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3596 new_bus->dev_count--;
3597 memcpy(new_bus->range + i, bus->range + i + 1,
3598 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3600 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3601 synchronize_srcu_expedited(&kvm->srcu);
3606 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3609 struct kvm_io_bus *bus;
3610 int dev_idx, srcu_idx;
3611 struct kvm_io_device *iodev = NULL;
3613 srcu_idx = srcu_read_lock(&kvm->srcu);
3615 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
3617 dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
3621 iodev = bus->range[dev_idx].dev;
3624 srcu_read_unlock(&kvm->srcu, srcu_idx);
3628 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
3630 static int kvm_debugfs_open(struct inode *inode, struct file *file,
3631 int (*get)(void *, u64 *), int (*set)(void *, u64),
3634 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
3637 /* The debugfs files are a reference to the kvm struct which
3638 * is still valid when kvm_destroy_vm is called.
3639 * To avoid the race between open and the removal of the debugfs
3640 * directory we test against the users count.
3642 if (!atomic_add_unless(&stat_data->kvm->users_count, 1, 0))
3645 if (simple_attr_open(inode, file, get, set, fmt)) {
3646 kvm_put_kvm(stat_data->kvm);
3653 static int kvm_debugfs_release(struct inode *inode, struct file *file)
3655 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
3658 simple_attr_release(inode, file);
3659 kvm_put_kvm(stat_data->kvm);
3664 static int vm_stat_get_per_vm(void *data, u64 *val)
3666 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3668 *val = *(ulong *)((void *)stat_data->kvm + stat_data->offset);
3673 static int vm_stat_clear_per_vm(void *data, u64 val)
3675 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3680 *(ulong *)((void *)stat_data->kvm + stat_data->offset) = 0;
3685 static int vm_stat_get_per_vm_open(struct inode *inode, struct file *file)
3687 __simple_attr_check_format("%llu\n", 0ull);
3688 return kvm_debugfs_open(inode, file, vm_stat_get_per_vm,
3689 vm_stat_clear_per_vm, "%llu\n");
3692 static const struct file_operations vm_stat_get_per_vm_fops = {
3693 .owner = THIS_MODULE,
3694 .open = vm_stat_get_per_vm_open,
3695 .release = kvm_debugfs_release,
3696 .read = simple_attr_read,
3697 .write = simple_attr_write,
3698 .llseek = generic_file_llseek,
3701 static int vcpu_stat_get_per_vm(void *data, u64 *val)
3704 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3705 struct kvm_vcpu *vcpu;
3709 kvm_for_each_vcpu(i, vcpu, stat_data->kvm)
3710 *val += *(u64 *)((void *)vcpu + stat_data->offset);
3715 static int vcpu_stat_clear_per_vm(void *data, u64 val)
3718 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3719 struct kvm_vcpu *vcpu;
3724 kvm_for_each_vcpu(i, vcpu, stat_data->kvm)
3725 *(u64 *)((void *)vcpu + stat_data->offset) = 0;
3730 static int vcpu_stat_get_per_vm_open(struct inode *inode, struct file *file)
3732 __simple_attr_check_format("%llu\n", 0ull);
3733 return kvm_debugfs_open(inode, file, vcpu_stat_get_per_vm,
3734 vcpu_stat_clear_per_vm, "%llu\n");
3737 static const struct file_operations vcpu_stat_get_per_vm_fops = {
3738 .owner = THIS_MODULE,
3739 .open = vcpu_stat_get_per_vm_open,
3740 .release = kvm_debugfs_release,
3741 .read = simple_attr_read,
3742 .write = simple_attr_write,
3743 .llseek = generic_file_llseek,
3746 static const struct file_operations *stat_fops_per_vm[] = {
3747 [KVM_STAT_VCPU] = &vcpu_stat_get_per_vm_fops,
3748 [KVM_STAT_VM] = &vm_stat_get_per_vm_fops,
3751 static int vm_stat_get(void *_offset, u64 *val)
3753 unsigned offset = (long)_offset;
3755 struct kvm_stat_data stat_tmp = {.offset = offset};
3759 spin_lock(&kvm_lock);
3760 list_for_each_entry(kvm, &vm_list, vm_list) {
3762 vm_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
3765 spin_unlock(&kvm_lock);
3769 static int vm_stat_clear(void *_offset, u64 val)
3771 unsigned offset = (long)_offset;
3773 struct kvm_stat_data stat_tmp = {.offset = offset};
3778 spin_lock(&kvm_lock);
3779 list_for_each_entry(kvm, &vm_list, vm_list) {
3781 vm_stat_clear_per_vm((void *)&stat_tmp, 0);
3783 spin_unlock(&kvm_lock);
3788 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n");
3790 static int vcpu_stat_get(void *_offset, u64 *val)
3792 unsigned offset = (long)_offset;
3794 struct kvm_stat_data stat_tmp = {.offset = offset};
3798 spin_lock(&kvm_lock);
3799 list_for_each_entry(kvm, &vm_list, vm_list) {
3801 vcpu_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
3804 spin_unlock(&kvm_lock);
3808 static int vcpu_stat_clear(void *_offset, u64 val)
3810 unsigned offset = (long)_offset;
3812 struct kvm_stat_data stat_tmp = {.offset = offset};
3817 spin_lock(&kvm_lock);
3818 list_for_each_entry(kvm, &vm_list, vm_list) {
3820 vcpu_stat_clear_per_vm((void *)&stat_tmp, 0);
3822 spin_unlock(&kvm_lock);
3827 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear,
3830 static const struct file_operations *stat_fops[] = {
3831 [KVM_STAT_VCPU] = &vcpu_stat_fops,
3832 [KVM_STAT_VM] = &vm_stat_fops,
3835 static int kvm_init_debug(void)
3838 struct kvm_stats_debugfs_item *p;
3840 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3841 if (kvm_debugfs_dir == NULL)
3844 kvm_debugfs_num_entries = 0;
3845 for (p = debugfs_entries; p->name; ++p, kvm_debugfs_num_entries++) {
3846 if (!debugfs_create_file(p->name, 0644, kvm_debugfs_dir,
3847 (void *)(long)p->offset,
3848 stat_fops[p->kind]))
3855 debugfs_remove_recursive(kvm_debugfs_dir);
3860 static int kvm_suspend(void)
3862 if (kvm_usage_count)
3863 hardware_disable_nolock(NULL);
3867 static void kvm_resume(void)
3869 if (kvm_usage_count) {
3870 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3871 hardware_enable_nolock(NULL);
3875 static struct syscore_ops kvm_syscore_ops = {
3876 .suspend = kvm_suspend,
3877 .resume = kvm_resume,
3881 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3883 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3886 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3888 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3890 if (vcpu->preempted)
3891 vcpu->preempted = false;
3893 kvm_arch_sched_in(vcpu, cpu);
3895 kvm_arch_vcpu_load(vcpu, cpu);
3898 static void kvm_sched_out(struct preempt_notifier *pn,
3899 struct task_struct *next)
3901 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3903 if (current->state == TASK_RUNNING)
3904 vcpu->preempted = true;
3905 kvm_arch_vcpu_put(vcpu);
3908 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3909 struct module *module)
3914 r = kvm_arch_init(opaque);
3919 * kvm_arch_init makes sure there's at most one caller
3920 * for architectures that support multiple implementations,
3921 * like intel and amd on x86.
3922 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3923 * conflicts in case kvm is already setup for another implementation.
3925 r = kvm_irqfd_init();
3929 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
3934 r = kvm_arch_hardware_setup();
3938 for_each_online_cpu(cpu) {
3939 smp_call_function_single(cpu,
3940 kvm_arch_check_processor_compat,
3946 r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting",
3947 kvm_starting_cpu, kvm_dying_cpu);
3950 register_reboot_notifier(&kvm_reboot_notifier);
3952 /* A kmem cache lets us meet the alignment requirements of fx_save. */
3954 vcpu_align = __alignof__(struct kvm_vcpu);
3955 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
3957 if (!kvm_vcpu_cache) {
3962 r = kvm_async_pf_init();
3966 kvm_chardev_ops.owner = module;
3967 kvm_vm_fops.owner = module;
3968 kvm_vcpu_fops.owner = module;
3970 r = misc_register(&kvm_dev);
3972 pr_err("kvm: misc device register failed\n");
3976 register_syscore_ops(&kvm_syscore_ops);
3978 kvm_preempt_ops.sched_in = kvm_sched_in;
3979 kvm_preempt_ops.sched_out = kvm_sched_out;
3981 r = kvm_init_debug();
3983 pr_err("kvm: create debugfs files failed\n");
3987 r = kvm_vfio_ops_init();
3993 unregister_syscore_ops(&kvm_syscore_ops);
3994 misc_deregister(&kvm_dev);
3996 kvm_async_pf_deinit();
3998 kmem_cache_destroy(kvm_vcpu_cache);
4000 unregister_reboot_notifier(&kvm_reboot_notifier);
4001 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4004 kvm_arch_hardware_unsetup();
4006 free_cpumask_var(cpus_hardware_enabled);
4014 EXPORT_SYMBOL_GPL(kvm_init);
4018 debugfs_remove_recursive(kvm_debugfs_dir);
4019 misc_deregister(&kvm_dev);
4020 kmem_cache_destroy(kvm_vcpu_cache);
4021 kvm_async_pf_deinit();
4022 unregister_syscore_ops(&kvm_syscore_ops);
4023 unregister_reboot_notifier(&kvm_reboot_notifier);
4024 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4025 on_each_cpu(hardware_disable_nolock, NULL, 1);
4026 kvm_arch_hardware_unsetup();
4029 free_cpumask_var(cpus_hardware_enabled);
4030 kvm_vfio_ops_exit();
4032 EXPORT_SYMBOL_GPL(kvm_exit);