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 #define KVM_EVENT_CREATE_VM 0
134 #define KVM_EVENT_DESTROY_VM 1
135 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm);
136 static unsigned long long kvm_createvm_count;
137 static unsigned long long kvm_active_vms;
139 __weak void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
140 unsigned long start, unsigned long end)
144 bool kvm_is_reserved_pfn(kvm_pfn_t pfn)
147 return PageReserved(pfn_to_page(pfn));
153 * Switches to specified vcpu, until a matching vcpu_put()
155 int vcpu_load(struct kvm_vcpu *vcpu)
159 if (mutex_lock_killable(&vcpu->mutex))
162 preempt_notifier_register(&vcpu->preempt_notifier);
163 kvm_arch_vcpu_load(vcpu, cpu);
167 EXPORT_SYMBOL_GPL(vcpu_load);
169 void vcpu_put(struct kvm_vcpu *vcpu)
172 kvm_arch_vcpu_put(vcpu);
173 preempt_notifier_unregister(&vcpu->preempt_notifier);
175 mutex_unlock(&vcpu->mutex);
177 EXPORT_SYMBOL_GPL(vcpu_put);
179 /* TODO: merge with kvm_arch_vcpu_should_kick */
180 static bool kvm_request_needs_ipi(struct kvm_vcpu *vcpu, unsigned req)
182 int mode = kvm_vcpu_exiting_guest_mode(vcpu);
185 * We need to wait for the VCPU to reenable interrupts and get out of
186 * READING_SHADOW_PAGE_TABLES mode.
188 if (req & KVM_REQUEST_WAIT)
189 return mode != OUTSIDE_GUEST_MODE;
192 * Need to kick a running VCPU, but otherwise there is nothing to do.
194 return mode == IN_GUEST_MODE;
197 static void ack_flush(void *_completed)
201 static inline bool kvm_kick_many_cpus(const struct cpumask *cpus, bool wait)
204 cpus = cpu_online_mask;
206 if (cpumask_empty(cpus))
209 smp_call_function_many(cpus, ack_flush, NULL, wait);
213 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
218 struct kvm_vcpu *vcpu;
220 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
223 kvm_for_each_vcpu(i, vcpu, kvm) {
224 kvm_make_request(req, vcpu);
227 if (!(req & KVM_REQUEST_NO_WAKEUP) && kvm_vcpu_wake_up(vcpu))
230 if (cpus != NULL && cpu != -1 && cpu != me &&
231 kvm_request_needs_ipi(vcpu, req))
232 __cpumask_set_cpu(cpu, cpus);
234 called = kvm_kick_many_cpus(cpus, !!(req & KVM_REQUEST_WAIT));
236 free_cpumask_var(cpus);
240 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
241 void kvm_flush_remote_tlbs(struct kvm *kvm)
244 * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in
245 * kvm_make_all_cpus_request.
247 long dirty_count = smp_load_acquire(&kvm->tlbs_dirty);
250 * We want to publish modifications to the page tables before reading
251 * mode. Pairs with a memory barrier in arch-specific code.
252 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
253 * and smp_mb in walk_shadow_page_lockless_begin/end.
254 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
256 * There is already an smp_mb__after_atomic() before
257 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
260 if (kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
261 ++kvm->stat.remote_tlb_flush;
262 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
264 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
267 void kvm_reload_remote_mmus(struct kvm *kvm)
269 kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
272 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
277 mutex_init(&vcpu->mutex);
282 init_swait_queue_head(&vcpu->wq);
283 kvm_async_pf_vcpu_init(vcpu);
286 INIT_LIST_HEAD(&vcpu->blocked_vcpu_list);
288 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
293 vcpu->run = page_address(page);
295 kvm_vcpu_set_in_spin_loop(vcpu, false);
296 kvm_vcpu_set_dy_eligible(vcpu, false);
297 vcpu->preempted = false;
299 r = kvm_arch_vcpu_init(vcpu);
305 free_page((unsigned long)vcpu->run);
309 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
311 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
314 * no need for rcu_read_lock as VCPU_RUN is the only place that
315 * will change the vcpu->pid pointer and on uninit all file
316 * descriptors are already gone.
318 put_pid(rcu_dereference_protected(vcpu->pid, 1));
319 kvm_arch_vcpu_uninit(vcpu);
320 free_page((unsigned long)vcpu->run);
322 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
324 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
325 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
327 return container_of(mn, struct kvm, mmu_notifier);
330 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
331 struct mm_struct *mm,
332 unsigned long address,
335 struct kvm *kvm = mmu_notifier_to_kvm(mn);
338 idx = srcu_read_lock(&kvm->srcu);
339 spin_lock(&kvm->mmu_lock);
340 kvm->mmu_notifier_seq++;
341 kvm_set_spte_hva(kvm, address, pte);
342 spin_unlock(&kvm->mmu_lock);
343 srcu_read_unlock(&kvm->srcu, idx);
346 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
347 struct mm_struct *mm,
351 struct kvm *kvm = mmu_notifier_to_kvm(mn);
352 int need_tlb_flush = 0, idx;
354 idx = srcu_read_lock(&kvm->srcu);
355 spin_lock(&kvm->mmu_lock);
357 * The count increase must become visible at unlock time as no
358 * spte can be established without taking the mmu_lock and
359 * count is also read inside the mmu_lock critical section.
361 kvm->mmu_notifier_count++;
362 need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
363 need_tlb_flush |= kvm->tlbs_dirty;
364 /* we've to flush the tlb before the pages can be freed */
366 kvm_flush_remote_tlbs(kvm);
368 spin_unlock(&kvm->mmu_lock);
370 kvm_arch_mmu_notifier_invalidate_range(kvm, start, end);
372 srcu_read_unlock(&kvm->srcu, idx);
375 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
376 struct mm_struct *mm,
380 struct kvm *kvm = mmu_notifier_to_kvm(mn);
382 spin_lock(&kvm->mmu_lock);
384 * This sequence increase will notify the kvm page fault that
385 * the page that is going to be mapped in the spte could have
388 kvm->mmu_notifier_seq++;
391 * The above sequence increase must be visible before the
392 * below count decrease, which is ensured by the smp_wmb above
393 * in conjunction with the smp_rmb in mmu_notifier_retry().
395 kvm->mmu_notifier_count--;
396 spin_unlock(&kvm->mmu_lock);
398 BUG_ON(kvm->mmu_notifier_count < 0);
401 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
402 struct mm_struct *mm,
406 struct kvm *kvm = mmu_notifier_to_kvm(mn);
409 idx = srcu_read_lock(&kvm->srcu);
410 spin_lock(&kvm->mmu_lock);
412 young = kvm_age_hva(kvm, start, end);
414 kvm_flush_remote_tlbs(kvm);
416 spin_unlock(&kvm->mmu_lock);
417 srcu_read_unlock(&kvm->srcu, idx);
422 static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
423 struct mm_struct *mm,
427 struct kvm *kvm = mmu_notifier_to_kvm(mn);
430 idx = srcu_read_lock(&kvm->srcu);
431 spin_lock(&kvm->mmu_lock);
433 * Even though we do not flush TLB, this will still adversely
434 * affect performance on pre-Haswell Intel EPT, where there is
435 * no EPT Access Bit to clear so that we have to tear down EPT
436 * tables instead. If we find this unacceptable, we can always
437 * add a parameter to kvm_age_hva so that it effectively doesn't
438 * do anything on clear_young.
440 * Also note that currently we never issue secondary TLB flushes
441 * from clear_young, leaving this job up to the regular system
442 * cadence. If we find this inaccurate, we might come up with a
443 * more sophisticated heuristic later.
445 young = kvm_age_hva(kvm, start, end);
446 spin_unlock(&kvm->mmu_lock);
447 srcu_read_unlock(&kvm->srcu, idx);
452 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
453 struct mm_struct *mm,
454 unsigned long address)
456 struct kvm *kvm = mmu_notifier_to_kvm(mn);
459 idx = srcu_read_lock(&kvm->srcu);
460 spin_lock(&kvm->mmu_lock);
461 young = kvm_test_age_hva(kvm, address);
462 spin_unlock(&kvm->mmu_lock);
463 srcu_read_unlock(&kvm->srcu, idx);
468 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
469 struct mm_struct *mm)
471 struct kvm *kvm = mmu_notifier_to_kvm(mn);
474 idx = srcu_read_lock(&kvm->srcu);
475 kvm_arch_flush_shadow_all(kvm);
476 srcu_read_unlock(&kvm->srcu, idx);
479 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
480 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
481 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
482 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
483 .clear_young = kvm_mmu_notifier_clear_young,
484 .test_young = kvm_mmu_notifier_test_young,
485 .change_pte = kvm_mmu_notifier_change_pte,
486 .release = kvm_mmu_notifier_release,
489 static int kvm_init_mmu_notifier(struct kvm *kvm)
491 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
492 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
495 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
497 static int kvm_init_mmu_notifier(struct kvm *kvm)
502 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
504 static struct kvm_memslots *kvm_alloc_memslots(void)
507 struct kvm_memslots *slots;
509 slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
513 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
514 slots->id_to_index[i] = slots->memslots[i].id = i;
519 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
521 if (!memslot->dirty_bitmap)
524 kvfree(memslot->dirty_bitmap);
525 memslot->dirty_bitmap = NULL;
529 * Free any memory in @free but not in @dont.
531 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
532 struct kvm_memory_slot *dont)
534 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
535 kvm_destroy_dirty_bitmap(free);
537 kvm_arch_free_memslot(kvm, free, dont);
542 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
544 struct kvm_memory_slot *memslot;
549 kvm_for_each_memslot(memslot, slots)
550 kvm_free_memslot(kvm, memslot, NULL);
555 static void kvm_destroy_vm_debugfs(struct kvm *kvm)
559 if (!kvm->debugfs_dentry)
562 debugfs_remove_recursive(kvm->debugfs_dentry);
564 if (kvm->debugfs_stat_data) {
565 for (i = 0; i < kvm_debugfs_num_entries; i++)
566 kfree(kvm->debugfs_stat_data[i]);
567 kfree(kvm->debugfs_stat_data);
571 static int kvm_create_vm_debugfs(struct kvm *kvm, int fd)
573 char dir_name[ITOA_MAX_LEN * 2];
574 struct kvm_stat_data *stat_data;
575 struct kvm_stats_debugfs_item *p;
577 if (!debugfs_initialized())
580 snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd);
581 kvm->debugfs_dentry = debugfs_create_dir(dir_name,
583 if (!kvm->debugfs_dentry)
586 kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries,
587 sizeof(*kvm->debugfs_stat_data),
589 if (!kvm->debugfs_stat_data)
592 for (p = debugfs_entries; p->name; p++) {
593 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL);
597 stat_data->kvm = kvm;
598 stat_data->offset = p->offset;
599 kvm->debugfs_stat_data[p - debugfs_entries] = stat_data;
600 if (!debugfs_create_file(p->name, 0644,
603 stat_fops_per_vm[p->kind]))
609 static struct kvm *kvm_create_vm(unsigned long type)
612 struct kvm *kvm = kvm_arch_alloc_vm();
615 return ERR_PTR(-ENOMEM);
617 spin_lock_init(&kvm->mmu_lock);
619 kvm->mm = current->mm;
620 kvm_eventfd_init(kvm);
621 mutex_init(&kvm->lock);
622 mutex_init(&kvm->irq_lock);
623 mutex_init(&kvm->slots_lock);
624 refcount_set(&kvm->users_count, 1);
625 INIT_LIST_HEAD(&kvm->devices);
627 r = kvm_arch_init_vm(kvm, type);
629 goto out_err_no_disable;
631 r = hardware_enable_all();
633 goto out_err_no_disable;
635 #ifdef CONFIG_HAVE_KVM_IRQFD
636 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
639 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
642 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
643 struct kvm_memslots *slots = kvm_alloc_memslots();
645 goto out_err_no_srcu;
647 * Generations must be different for each address space.
648 * Init kvm generation close to the maximum to easily test the
649 * code of handling generation number wrap-around.
651 slots->generation = i * 2 - 150;
652 rcu_assign_pointer(kvm->memslots[i], slots);
655 if (init_srcu_struct(&kvm->srcu))
656 goto out_err_no_srcu;
657 if (init_srcu_struct(&kvm->irq_srcu))
658 goto out_err_no_irq_srcu;
659 for (i = 0; i < KVM_NR_BUSES; i++) {
660 rcu_assign_pointer(kvm->buses[i],
661 kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL));
666 r = kvm_init_mmu_notifier(kvm);
670 spin_lock(&kvm_lock);
671 list_add(&kvm->vm_list, &vm_list);
672 spin_unlock(&kvm_lock);
674 preempt_notifier_inc();
679 cleanup_srcu_struct(&kvm->irq_srcu);
681 cleanup_srcu_struct(&kvm->srcu);
683 hardware_disable_all();
685 refcount_set(&kvm->users_count, 0);
686 for (i = 0; i < KVM_NR_BUSES; i++)
687 kfree(kvm_get_bus(kvm, i));
688 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
689 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
690 kvm_arch_free_vm(kvm);
695 static void kvm_destroy_devices(struct kvm *kvm)
697 struct kvm_device *dev, *tmp;
700 * We do not need to take the kvm->lock here, because nobody else
701 * has a reference to the struct kvm at this point and therefore
702 * cannot access the devices list anyhow.
704 list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
705 list_del(&dev->vm_node);
706 dev->ops->destroy(dev);
710 static void kvm_destroy_vm(struct kvm *kvm)
713 struct mm_struct *mm = kvm->mm;
715 kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm);
716 kvm_destroy_vm_debugfs(kvm);
717 kvm_arch_sync_events(kvm);
718 spin_lock(&kvm_lock);
719 list_del(&kvm->vm_list);
720 spin_unlock(&kvm_lock);
721 kvm_free_irq_routing(kvm);
722 for (i = 0; i < KVM_NR_BUSES; i++) {
723 struct kvm_io_bus *bus = kvm_get_bus(kvm, i);
726 kvm_io_bus_destroy(bus);
727 kvm->buses[i] = NULL;
729 kvm_coalesced_mmio_free(kvm);
730 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
731 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
733 kvm_arch_flush_shadow_all(kvm);
735 kvm_arch_destroy_vm(kvm);
736 kvm_destroy_devices(kvm);
737 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
738 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
739 cleanup_srcu_struct(&kvm->irq_srcu);
740 cleanup_srcu_struct(&kvm->srcu);
741 kvm_arch_free_vm(kvm);
742 preempt_notifier_dec();
743 hardware_disable_all();
747 void kvm_get_kvm(struct kvm *kvm)
749 refcount_inc(&kvm->users_count);
751 EXPORT_SYMBOL_GPL(kvm_get_kvm);
753 void kvm_put_kvm(struct kvm *kvm)
755 if (refcount_dec_and_test(&kvm->users_count))
758 EXPORT_SYMBOL_GPL(kvm_put_kvm);
761 static int kvm_vm_release(struct inode *inode, struct file *filp)
763 struct kvm *kvm = filp->private_data;
765 kvm_irqfd_release(kvm);
772 * Allocation size is twice as large as the actual dirty bitmap size.
773 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
775 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
777 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
779 memslot->dirty_bitmap = kvzalloc(dirty_bytes, GFP_KERNEL);
780 if (!memslot->dirty_bitmap)
787 * Insert memslot and re-sort memslots based on their GFN,
788 * so binary search could be used to lookup GFN.
789 * Sorting algorithm takes advantage of having initially
790 * sorted array and known changed memslot position.
792 static void update_memslots(struct kvm_memslots *slots,
793 struct kvm_memory_slot *new)
796 int i = slots->id_to_index[id];
797 struct kvm_memory_slot *mslots = slots->memslots;
799 WARN_ON(mslots[i].id != id);
801 WARN_ON(!mslots[i].npages);
802 if (mslots[i].npages)
805 if (!mslots[i].npages)
809 while (i < KVM_MEM_SLOTS_NUM - 1 &&
810 new->base_gfn <= mslots[i + 1].base_gfn) {
811 if (!mslots[i + 1].npages)
813 mslots[i] = mslots[i + 1];
814 slots->id_to_index[mslots[i].id] = i;
819 * The ">=" is needed when creating a slot with base_gfn == 0,
820 * so that it moves before all those with base_gfn == npages == 0.
822 * On the other hand, if new->npages is zero, the above loop has
823 * already left i pointing to the beginning of the empty part of
824 * mslots, and the ">=" would move the hole backwards in this
825 * case---which is wrong. So skip the loop when deleting a slot.
829 new->base_gfn >= mslots[i - 1].base_gfn) {
830 mslots[i] = mslots[i - 1];
831 slots->id_to_index[mslots[i].id] = i;
835 WARN_ON_ONCE(i != slots->used_slots);
838 slots->id_to_index[mslots[i].id] = i;
841 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
843 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
845 #ifdef __KVM_HAVE_READONLY_MEM
846 valid_flags |= KVM_MEM_READONLY;
849 if (mem->flags & ~valid_flags)
855 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
856 int as_id, struct kvm_memslots *slots)
858 struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id);
861 * Set the low bit in the generation, which disables SPTE caching
862 * until the end of synchronize_srcu_expedited.
864 WARN_ON(old_memslots->generation & 1);
865 slots->generation = old_memslots->generation + 1;
867 rcu_assign_pointer(kvm->memslots[as_id], slots);
868 synchronize_srcu_expedited(&kvm->srcu);
871 * Increment the new memslot generation a second time. This prevents
872 * vm exits that race with memslot updates from caching a memslot
873 * generation that will (potentially) be valid forever.
875 * Generations must be unique even across address spaces. We do not need
876 * a global counter for that, instead the generation space is evenly split
877 * across address spaces. For example, with two address spaces, address
878 * space 0 will use generations 0, 4, 8, ... while * address space 1 will
879 * use generations 2, 6, 10, 14, ...
881 slots->generation += KVM_ADDRESS_SPACE_NUM * 2 - 1;
883 kvm_arch_memslots_updated(kvm, slots);
889 * Allocate some memory and give it an address in the guest physical address
892 * Discontiguous memory is allowed, mostly for framebuffers.
894 * Must be called holding kvm->slots_lock for write.
896 int __kvm_set_memory_region(struct kvm *kvm,
897 const struct kvm_userspace_memory_region *mem)
901 unsigned long npages;
902 struct kvm_memory_slot *slot;
903 struct kvm_memory_slot old, new;
904 struct kvm_memslots *slots = NULL, *old_memslots;
906 enum kvm_mr_change change;
908 r = check_memory_region_flags(mem);
913 as_id = mem->slot >> 16;
916 /* General sanity checks */
917 if (mem->memory_size & (PAGE_SIZE - 1))
919 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
921 /* We can read the guest memory with __xxx_user() later on. */
922 if ((id < KVM_USER_MEM_SLOTS) &&
923 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
924 !access_ok(VERIFY_WRITE,
925 (void __user *)(unsigned long)mem->userspace_addr,
928 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
930 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
933 slot = id_to_memslot(__kvm_memslots(kvm, as_id), id);
934 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
935 npages = mem->memory_size >> PAGE_SHIFT;
937 if (npages > KVM_MEM_MAX_NR_PAGES)
943 new.base_gfn = base_gfn;
945 new.flags = mem->flags;
949 change = KVM_MR_CREATE;
950 else { /* Modify an existing slot. */
951 if ((mem->userspace_addr != old.userspace_addr) ||
952 (npages != old.npages) ||
953 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
956 if (base_gfn != old.base_gfn)
957 change = KVM_MR_MOVE;
958 else if (new.flags != old.flags)
959 change = KVM_MR_FLAGS_ONLY;
960 else { /* Nothing to change. */
969 change = KVM_MR_DELETE;
974 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
975 /* Check for overlaps */
977 kvm_for_each_memslot(slot, __kvm_memslots(kvm, as_id)) {
980 if (!((base_gfn + npages <= slot->base_gfn) ||
981 (base_gfn >= slot->base_gfn + slot->npages)))
986 /* Free page dirty bitmap if unneeded */
987 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
988 new.dirty_bitmap = NULL;
991 if (change == KVM_MR_CREATE) {
992 new.userspace_addr = mem->userspace_addr;
994 if (kvm_arch_create_memslot(kvm, &new, npages))
998 /* Allocate page dirty bitmap if needed */
999 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
1000 if (kvm_create_dirty_bitmap(&new) < 0)
1004 slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
1007 memcpy(slots, __kvm_memslots(kvm, as_id), sizeof(struct kvm_memslots));
1009 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
1010 slot = id_to_memslot(slots, id);
1011 slot->flags |= KVM_MEMSLOT_INVALID;
1013 old_memslots = install_new_memslots(kvm, as_id, slots);
1015 /* From this point no new shadow pages pointing to a deleted,
1016 * or moved, memslot will be created.
1018 * validation of sp->gfn happens in:
1019 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1020 * - kvm_is_visible_gfn (mmu_check_roots)
1022 kvm_arch_flush_shadow_memslot(kvm, slot);
1025 * We can re-use the old_memslots from above, the only difference
1026 * from the currently installed memslots is the invalid flag. This
1027 * will get overwritten by update_memslots anyway.
1029 slots = old_memslots;
1032 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
1036 /* actual memory is freed via old in kvm_free_memslot below */
1037 if (change == KVM_MR_DELETE) {
1038 new.dirty_bitmap = NULL;
1039 memset(&new.arch, 0, sizeof(new.arch));
1042 update_memslots(slots, &new);
1043 old_memslots = install_new_memslots(kvm, as_id, slots);
1045 kvm_arch_commit_memory_region(kvm, mem, &old, &new, change);
1047 kvm_free_memslot(kvm, &old, &new);
1048 kvfree(old_memslots);
1054 kvm_free_memslot(kvm, &new, &old);
1058 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1060 int kvm_set_memory_region(struct kvm *kvm,
1061 const struct kvm_userspace_memory_region *mem)
1065 mutex_lock(&kvm->slots_lock);
1066 r = __kvm_set_memory_region(kvm, mem);
1067 mutex_unlock(&kvm->slots_lock);
1070 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1072 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1073 struct kvm_userspace_memory_region *mem)
1075 if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1078 return kvm_set_memory_region(kvm, mem);
1081 int kvm_get_dirty_log(struct kvm *kvm,
1082 struct kvm_dirty_log *log, int *is_dirty)
1084 struct kvm_memslots *slots;
1085 struct kvm_memory_slot *memslot;
1088 unsigned long any = 0;
1090 as_id = log->slot >> 16;
1091 id = (u16)log->slot;
1092 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1095 slots = __kvm_memslots(kvm, as_id);
1096 memslot = id_to_memslot(slots, id);
1097 if (!memslot->dirty_bitmap)
1100 n = kvm_dirty_bitmap_bytes(memslot);
1102 for (i = 0; !any && i < n/sizeof(long); ++i)
1103 any = memslot->dirty_bitmap[i];
1105 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1112 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1114 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1116 * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages
1117 * are dirty write protect them for next write.
1118 * @kvm: pointer to kvm instance
1119 * @log: slot id and address to which we copy the log
1120 * @is_dirty: flag set if any page is dirty
1122 * We need to keep it in mind that VCPU threads can write to the bitmap
1123 * concurrently. So, to avoid losing track of dirty pages we keep the
1126 * 1. Take a snapshot of the bit and clear it if needed.
1127 * 2. Write protect the corresponding page.
1128 * 3. Copy the snapshot to the userspace.
1129 * 4. Upon return caller flushes TLB's if needed.
1131 * Between 2 and 4, the guest may write to the page using the remaining TLB
1132 * entry. This is not a problem because the page is reported dirty using
1133 * the snapshot taken before and step 4 ensures that writes done after
1134 * exiting to userspace will be logged for the next call.
1137 int kvm_get_dirty_log_protect(struct kvm *kvm,
1138 struct kvm_dirty_log *log, bool *is_dirty)
1140 struct kvm_memslots *slots;
1141 struct kvm_memory_slot *memslot;
1144 unsigned long *dirty_bitmap;
1145 unsigned long *dirty_bitmap_buffer;
1147 as_id = log->slot >> 16;
1148 id = (u16)log->slot;
1149 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1152 slots = __kvm_memslots(kvm, as_id);
1153 memslot = id_to_memslot(slots, id);
1155 dirty_bitmap = memslot->dirty_bitmap;
1159 n = kvm_dirty_bitmap_bytes(memslot);
1161 dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long);
1162 memset(dirty_bitmap_buffer, 0, n);
1164 spin_lock(&kvm->mmu_lock);
1166 for (i = 0; i < n / sizeof(long); i++) {
1170 if (!dirty_bitmap[i])
1175 mask = xchg(&dirty_bitmap[i], 0);
1176 dirty_bitmap_buffer[i] = mask;
1179 offset = i * BITS_PER_LONG;
1180 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1185 spin_unlock(&kvm->mmu_lock);
1186 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1190 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect);
1193 bool kvm_largepages_enabled(void)
1195 return largepages_enabled;
1198 void kvm_disable_largepages(void)
1200 largepages_enabled = false;
1202 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1204 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1206 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1208 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1210 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
1212 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn);
1215 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1217 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1219 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1220 memslot->flags & KVM_MEMSLOT_INVALID)
1225 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1227 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1229 struct vm_area_struct *vma;
1230 unsigned long addr, size;
1234 addr = gfn_to_hva(kvm, gfn);
1235 if (kvm_is_error_hva(addr))
1238 down_read(¤t->mm->mmap_sem);
1239 vma = find_vma(current->mm, addr);
1243 size = vma_kernel_pagesize(vma);
1246 up_read(¤t->mm->mmap_sem);
1251 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1253 return slot->flags & KVM_MEM_READONLY;
1256 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1257 gfn_t *nr_pages, bool write)
1259 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1260 return KVM_HVA_ERR_BAD;
1262 if (memslot_is_readonly(slot) && write)
1263 return KVM_HVA_ERR_RO_BAD;
1266 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1268 return __gfn_to_hva_memslot(slot, gfn);
1271 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1274 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1277 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1280 return gfn_to_hva_many(slot, gfn, NULL);
1282 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1284 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1286 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1288 EXPORT_SYMBOL_GPL(gfn_to_hva);
1290 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
1292 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
1294 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
1297 * If writable is set to false, the hva returned by this function is only
1298 * allowed to be read.
1300 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1301 gfn_t gfn, bool *writable)
1303 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1305 if (!kvm_is_error_hva(hva) && writable)
1306 *writable = !memslot_is_readonly(slot);
1311 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1313 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1315 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1318 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
1320 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1322 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1325 static int get_user_page_nowait(unsigned long start, int write,
1328 int flags = FOLL_NOWAIT | FOLL_HWPOISON;
1331 flags |= FOLL_WRITE;
1333 return get_user_pages(start, 1, flags, page, NULL);
1336 static inline int check_user_page_hwpoison(unsigned long addr)
1338 int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
1340 rc = get_user_pages(addr, 1, flags, NULL, NULL);
1341 return rc == -EHWPOISON;
1345 * The atomic path to get the writable pfn which will be stored in @pfn,
1346 * true indicates success, otherwise false is returned.
1348 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1349 bool write_fault, bool *writable, kvm_pfn_t *pfn)
1351 struct page *page[1];
1354 if (!(async || atomic))
1358 * Fast pin a writable pfn only if it is a write fault request
1359 * or the caller allows to map a writable pfn for a read fault
1362 if (!(write_fault || writable))
1365 npages = __get_user_pages_fast(addr, 1, 1, page);
1367 *pfn = page_to_pfn(page[0]);
1378 * The slow path to get the pfn of the specified host virtual address,
1379 * 1 indicates success, -errno is returned if error is detected.
1381 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1382 bool *writable, kvm_pfn_t *pfn)
1384 struct page *page[1];
1390 *writable = write_fault;
1393 down_read(¤t->mm->mmap_sem);
1394 npages = get_user_page_nowait(addr, write_fault, page);
1395 up_read(¤t->mm->mmap_sem);
1397 unsigned int flags = FOLL_HWPOISON;
1400 flags |= FOLL_WRITE;
1402 npages = get_user_pages_unlocked(addr, 1, page, flags);
1407 /* map read fault as writable if possible */
1408 if (unlikely(!write_fault) && writable) {
1409 struct page *wpage[1];
1411 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1420 *pfn = page_to_pfn(page[0]);
1424 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1426 if (unlikely(!(vma->vm_flags & VM_READ)))
1429 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1435 static int hva_to_pfn_remapped(struct vm_area_struct *vma,
1436 unsigned long addr, bool *async,
1437 bool write_fault, kvm_pfn_t *p_pfn)
1442 r = follow_pfn(vma, addr, &pfn);
1445 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
1446 * not call the fault handler, so do it here.
1448 bool unlocked = false;
1449 r = fixup_user_fault(current, current->mm, addr,
1450 (write_fault ? FAULT_FLAG_WRITE : 0),
1457 r = follow_pfn(vma, addr, &pfn);
1465 * Get a reference here because callers of *hva_to_pfn* and
1466 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
1467 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
1468 * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will
1469 * simply do nothing for reserved pfns.
1471 * Whoever called remap_pfn_range is also going to call e.g.
1472 * unmap_mapping_range before the underlying pages are freed,
1473 * causing a call to our MMU notifier.
1482 * Pin guest page in memory and return its pfn.
1483 * @addr: host virtual address which maps memory to the guest
1484 * @atomic: whether this function can sleep
1485 * @async: whether this function need to wait IO complete if the
1486 * host page is not in the memory
1487 * @write_fault: whether we should get a writable host page
1488 * @writable: whether it allows to map a writable host page for !@write_fault
1490 * The function will map a writable host page for these two cases:
1491 * 1): @write_fault = true
1492 * 2): @write_fault = false && @writable, @writable will tell the caller
1493 * whether the mapping is writable.
1495 static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1496 bool write_fault, bool *writable)
1498 struct vm_area_struct *vma;
1502 /* we can do it either atomically or asynchronously, not both */
1503 BUG_ON(atomic && async);
1505 if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1509 return KVM_PFN_ERR_FAULT;
1511 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1515 down_read(¤t->mm->mmap_sem);
1516 if (npages == -EHWPOISON ||
1517 (!async && check_user_page_hwpoison(addr))) {
1518 pfn = KVM_PFN_ERR_HWPOISON;
1523 vma = find_vma_intersection(current->mm, addr, addr + 1);
1526 pfn = KVM_PFN_ERR_FAULT;
1527 else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
1528 r = hva_to_pfn_remapped(vma, addr, async, write_fault, &pfn);
1532 pfn = KVM_PFN_ERR_FAULT;
1534 if (async && vma_is_valid(vma, write_fault))
1536 pfn = KVM_PFN_ERR_FAULT;
1539 up_read(¤t->mm->mmap_sem);
1543 kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn,
1544 bool atomic, bool *async, bool write_fault,
1547 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1549 if (addr == KVM_HVA_ERR_RO_BAD) {
1552 return KVM_PFN_ERR_RO_FAULT;
1555 if (kvm_is_error_hva(addr)) {
1558 return KVM_PFN_NOSLOT;
1561 /* Do not map writable pfn in the readonly memslot. */
1562 if (writable && memslot_is_readonly(slot)) {
1567 return hva_to_pfn(addr, atomic, async, write_fault,
1570 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
1572 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1575 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
1576 write_fault, writable);
1578 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1580 kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1582 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1584 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
1586 kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1588 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1590 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1592 kvm_pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1594 return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm, gfn), gfn);
1596 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1598 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
1600 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1602 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
1604 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1606 return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
1608 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1610 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
1612 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1614 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
1616 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1617 struct page **pages, int nr_pages)
1622 addr = gfn_to_hva_many(slot, gfn, &entry);
1623 if (kvm_is_error_hva(addr))
1626 if (entry < nr_pages)
1629 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1631 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1633 static struct page *kvm_pfn_to_page(kvm_pfn_t pfn)
1635 if (is_error_noslot_pfn(pfn))
1636 return KVM_ERR_PTR_BAD_PAGE;
1638 if (kvm_is_reserved_pfn(pfn)) {
1640 return KVM_ERR_PTR_BAD_PAGE;
1643 return pfn_to_page(pfn);
1646 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1650 pfn = gfn_to_pfn(kvm, gfn);
1652 return kvm_pfn_to_page(pfn);
1654 EXPORT_SYMBOL_GPL(gfn_to_page);
1656 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
1660 pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
1662 return kvm_pfn_to_page(pfn);
1664 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
1666 void kvm_release_page_clean(struct page *page)
1668 WARN_ON(is_error_page(page));
1670 kvm_release_pfn_clean(page_to_pfn(page));
1672 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1674 void kvm_release_pfn_clean(kvm_pfn_t pfn)
1676 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
1677 put_page(pfn_to_page(pfn));
1679 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1681 void kvm_release_page_dirty(struct page *page)
1683 WARN_ON(is_error_page(page));
1685 kvm_release_pfn_dirty(page_to_pfn(page));
1687 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1689 static void kvm_release_pfn_dirty(kvm_pfn_t pfn)
1691 kvm_set_pfn_dirty(pfn);
1692 kvm_release_pfn_clean(pfn);
1695 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
1697 if (!kvm_is_reserved_pfn(pfn)) {
1698 struct page *page = pfn_to_page(pfn);
1700 if (!PageReserved(page))
1704 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1706 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
1708 if (!kvm_is_reserved_pfn(pfn))
1709 mark_page_accessed(pfn_to_page(pfn));
1711 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1713 void kvm_get_pfn(kvm_pfn_t pfn)
1715 if (!kvm_is_reserved_pfn(pfn))
1716 get_page(pfn_to_page(pfn));
1718 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1720 static int next_segment(unsigned long len, int offset)
1722 if (len > PAGE_SIZE - offset)
1723 return PAGE_SIZE - offset;
1728 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
1729 void *data, int offset, int len)
1734 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1735 if (kvm_is_error_hva(addr))
1737 r = __copy_from_user(data, (void __user *)addr + offset, len);
1743 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1746 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1748 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1750 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1752 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
1753 int offset, int len)
1755 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1757 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1759 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
1761 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1763 gfn_t gfn = gpa >> PAGE_SHIFT;
1765 int offset = offset_in_page(gpa);
1768 while ((seg = next_segment(len, offset)) != 0) {
1769 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1779 EXPORT_SYMBOL_GPL(kvm_read_guest);
1781 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
1783 gfn_t gfn = gpa >> PAGE_SHIFT;
1785 int offset = offset_in_page(gpa);
1788 while ((seg = next_segment(len, offset)) != 0) {
1789 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
1799 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
1801 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1802 void *data, int offset, unsigned long len)
1807 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1808 if (kvm_is_error_hva(addr))
1810 pagefault_disable();
1811 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1818 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1821 gfn_t gfn = gpa >> PAGE_SHIFT;
1822 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1823 int offset = offset_in_page(gpa);
1825 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1827 EXPORT_SYMBOL_GPL(kvm_read_guest_atomic);
1829 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
1830 void *data, unsigned long len)
1832 gfn_t gfn = gpa >> PAGE_SHIFT;
1833 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1834 int offset = offset_in_page(gpa);
1836 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1838 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
1840 static int __kvm_write_guest_page(struct kvm_memory_slot *memslot, gfn_t gfn,
1841 const void *data, int offset, int len)
1846 addr = gfn_to_hva_memslot(memslot, gfn);
1847 if (kvm_is_error_hva(addr))
1849 r = __copy_to_user((void __user *)addr + offset, data, len);
1852 mark_page_dirty_in_slot(memslot, gfn);
1856 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
1857 const void *data, int offset, int len)
1859 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1861 return __kvm_write_guest_page(slot, gfn, data, offset, len);
1863 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1865 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
1866 const void *data, int offset, int len)
1868 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1870 return __kvm_write_guest_page(slot, gfn, data, offset, len);
1872 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
1874 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1877 gfn_t gfn = gpa >> PAGE_SHIFT;
1879 int offset = offset_in_page(gpa);
1882 while ((seg = next_segment(len, offset)) != 0) {
1883 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1893 EXPORT_SYMBOL_GPL(kvm_write_guest);
1895 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
1898 gfn_t gfn = gpa >> PAGE_SHIFT;
1900 int offset = offset_in_page(gpa);
1903 while ((seg = next_segment(len, offset)) != 0) {
1904 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
1914 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
1916 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
1917 struct gfn_to_hva_cache *ghc,
1918 gpa_t gpa, unsigned long len)
1920 int offset = offset_in_page(gpa);
1921 gfn_t start_gfn = gpa >> PAGE_SHIFT;
1922 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1923 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1924 gfn_t nr_pages_avail;
1927 ghc->generation = slots->generation;
1929 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
1930 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, NULL);
1931 if (!kvm_is_error_hva(ghc->hva) && nr_pages_needed <= 1) {
1935 * If the requested region crosses two memslots, we still
1936 * verify that the entire region is valid here.
1938 while (start_gfn <= end_gfn) {
1940 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
1941 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1943 if (kvm_is_error_hva(ghc->hva))
1945 start_gfn += nr_pages_avail;
1947 /* Use the slow path for cross page reads and writes. */
1948 ghc->memslot = NULL;
1953 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1954 gpa_t gpa, unsigned long len)
1956 struct kvm_memslots *slots = kvm_memslots(kvm);
1957 return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len);
1959 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1961 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1962 void *data, int offset, unsigned long len)
1964 struct kvm_memslots *slots = kvm_memslots(kvm);
1966 gpa_t gpa = ghc->gpa + offset;
1968 BUG_ON(len + offset > ghc->len);
1970 if (slots->generation != ghc->generation)
1971 __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len);
1973 if (unlikely(!ghc->memslot))
1974 return kvm_write_guest(kvm, gpa, data, len);
1976 if (kvm_is_error_hva(ghc->hva))
1979 r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
1982 mark_page_dirty_in_slot(ghc->memslot, gpa >> PAGE_SHIFT);
1986 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached);
1988 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1989 void *data, unsigned long len)
1991 return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len);
1993 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1995 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1996 void *data, unsigned long len)
1998 struct kvm_memslots *slots = kvm_memslots(kvm);
2001 BUG_ON(len > ghc->len);
2003 if (slots->generation != ghc->generation)
2004 __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len);
2006 if (unlikely(!ghc->memslot))
2007 return kvm_read_guest(kvm, ghc->gpa, data, len);
2009 if (kvm_is_error_hva(ghc->hva))
2012 r = __copy_from_user(data, (void __user *)ghc->hva, len);
2018 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
2020 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
2022 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
2024 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
2026 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
2028 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
2030 gfn_t gfn = gpa >> PAGE_SHIFT;
2032 int offset = offset_in_page(gpa);
2035 while ((seg = next_segment(len, offset)) != 0) {
2036 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
2045 EXPORT_SYMBOL_GPL(kvm_clear_guest);
2047 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot,
2050 if (memslot && memslot->dirty_bitmap) {
2051 unsigned long rel_gfn = gfn - memslot->base_gfn;
2053 set_bit_le(rel_gfn, memslot->dirty_bitmap);
2057 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
2059 struct kvm_memory_slot *memslot;
2061 memslot = gfn_to_memslot(kvm, gfn);
2062 mark_page_dirty_in_slot(memslot, gfn);
2064 EXPORT_SYMBOL_GPL(mark_page_dirty);
2066 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
2068 struct kvm_memory_slot *memslot;
2070 memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2071 mark_page_dirty_in_slot(memslot, gfn);
2073 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
2075 void kvm_sigset_activate(struct kvm_vcpu *vcpu)
2077 if (!vcpu->sigset_active)
2081 * This does a lockless modification of ->real_blocked, which is fine
2082 * because, only current can change ->real_blocked and all readers of
2083 * ->real_blocked don't care as long ->real_blocked is always a subset
2086 sigprocmask(SIG_SETMASK, &vcpu->sigset, ¤t->real_blocked);
2089 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu)
2091 if (!vcpu->sigset_active)
2094 sigprocmask(SIG_SETMASK, ¤t->real_blocked, NULL);
2095 sigemptyset(¤t->real_blocked);
2098 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
2100 unsigned int old, val, grow;
2102 old = val = vcpu->halt_poll_ns;
2103 grow = READ_ONCE(halt_poll_ns_grow);
2105 if (val == 0 && grow)
2110 if (val > halt_poll_ns)
2113 vcpu->halt_poll_ns = val;
2114 trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
2117 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
2119 unsigned int old, val, shrink;
2121 old = val = vcpu->halt_poll_ns;
2122 shrink = READ_ONCE(halt_poll_ns_shrink);
2128 vcpu->halt_poll_ns = val;
2129 trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
2132 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
2134 if (kvm_arch_vcpu_runnable(vcpu)) {
2135 kvm_make_request(KVM_REQ_UNHALT, vcpu);
2138 if (kvm_cpu_has_pending_timer(vcpu))
2140 if (signal_pending(current))
2147 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2149 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
2152 DECLARE_SWAITQUEUE(wait);
2153 bool waited = false;
2156 start = cur = ktime_get();
2157 if (vcpu->halt_poll_ns) {
2158 ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns);
2160 ++vcpu->stat.halt_attempted_poll;
2163 * This sets KVM_REQ_UNHALT if an interrupt
2166 if (kvm_vcpu_check_block(vcpu) < 0) {
2167 ++vcpu->stat.halt_successful_poll;
2168 if (!vcpu_valid_wakeup(vcpu))
2169 ++vcpu->stat.halt_poll_invalid;
2173 } while (single_task_running() && ktime_before(cur, stop));
2176 kvm_arch_vcpu_blocking(vcpu);
2179 prepare_to_swait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
2181 if (kvm_vcpu_check_block(vcpu) < 0)
2188 finish_swait(&vcpu->wq, &wait);
2191 kvm_arch_vcpu_unblocking(vcpu);
2193 block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
2195 if (!vcpu_valid_wakeup(vcpu))
2196 shrink_halt_poll_ns(vcpu);
2197 else if (halt_poll_ns) {
2198 if (block_ns <= vcpu->halt_poll_ns)
2200 /* we had a long block, shrink polling */
2201 else if (vcpu->halt_poll_ns && block_ns > halt_poll_ns)
2202 shrink_halt_poll_ns(vcpu);
2203 /* we had a short halt and our poll time is too small */
2204 else if (vcpu->halt_poll_ns < halt_poll_ns &&
2205 block_ns < halt_poll_ns)
2206 grow_halt_poll_ns(vcpu);
2208 vcpu->halt_poll_ns = 0;
2210 trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu));
2211 kvm_arch_vcpu_block_finish(vcpu);
2213 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
2215 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
2217 struct swait_queue_head *wqp;
2219 wqp = kvm_arch_vcpu_wq(vcpu);
2220 if (swq_has_sleeper(wqp)) {
2222 ++vcpu->stat.halt_wakeup;
2228 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
2232 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2234 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
2237 int cpu = vcpu->cpu;
2239 if (kvm_vcpu_wake_up(vcpu))
2243 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
2244 if (kvm_arch_vcpu_should_kick(vcpu))
2245 smp_send_reschedule(cpu);
2248 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
2249 #endif /* !CONFIG_S390 */
2251 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
2254 struct task_struct *task = NULL;
2258 pid = rcu_dereference(target->pid);
2260 task = get_pid_task(pid, PIDTYPE_PID);
2264 ret = yield_to(task, 1);
2265 put_task_struct(task);
2269 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
2272 * Helper that checks whether a VCPU is eligible for directed yield.
2273 * Most eligible candidate to yield is decided by following heuristics:
2275 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2276 * (preempted lock holder), indicated by @in_spin_loop.
2277 * Set at the beiginning and cleared at the end of interception/PLE handler.
2279 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2280 * chance last time (mostly it has become eligible now since we have probably
2281 * yielded to lockholder in last iteration. This is done by toggling
2282 * @dy_eligible each time a VCPU checked for eligibility.)
2284 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2285 * to preempted lock-holder could result in wrong VCPU selection and CPU
2286 * burning. Giving priority for a potential lock-holder increases lock
2289 * Since algorithm is based on heuristics, accessing another VCPU data without
2290 * locking does not harm. It may result in trying to yield to same VCPU, fail
2291 * and continue with next VCPU and so on.
2293 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
2295 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2298 eligible = !vcpu->spin_loop.in_spin_loop ||
2299 vcpu->spin_loop.dy_eligible;
2301 if (vcpu->spin_loop.in_spin_loop)
2302 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
2310 void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode)
2312 struct kvm *kvm = me->kvm;
2313 struct kvm_vcpu *vcpu;
2314 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
2320 kvm_vcpu_set_in_spin_loop(me, true);
2322 * We boost the priority of a VCPU that is runnable but not
2323 * currently running, because it got preempted by something
2324 * else and called schedule in __vcpu_run. Hopefully that
2325 * VCPU is holding the lock that we need and will release it.
2326 * We approximate round-robin by starting at the last boosted VCPU.
2328 for (pass = 0; pass < 2 && !yielded && try; pass++) {
2329 kvm_for_each_vcpu(i, vcpu, kvm) {
2330 if (!pass && i <= last_boosted_vcpu) {
2331 i = last_boosted_vcpu;
2333 } else if (pass && i > last_boosted_vcpu)
2335 if (!ACCESS_ONCE(vcpu->preempted))
2339 if (swait_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
2341 if (yield_to_kernel_mode && !kvm_arch_vcpu_in_kernel(vcpu))
2343 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
2346 yielded = kvm_vcpu_yield_to(vcpu);
2348 kvm->last_boosted_vcpu = i;
2350 } else if (yielded < 0) {
2357 kvm_vcpu_set_in_spin_loop(me, false);
2359 /* Ensure vcpu is not eligible during next spinloop */
2360 kvm_vcpu_set_dy_eligible(me, false);
2362 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
2364 static int kvm_vcpu_fault(struct vm_fault *vmf)
2366 struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data;
2369 if (vmf->pgoff == 0)
2370 page = virt_to_page(vcpu->run);
2372 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
2373 page = virt_to_page(vcpu->arch.pio_data);
2375 #ifdef CONFIG_KVM_MMIO
2376 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
2377 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
2380 return kvm_arch_vcpu_fault(vcpu, vmf);
2386 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
2387 .fault = kvm_vcpu_fault,
2390 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2392 vma->vm_ops = &kvm_vcpu_vm_ops;
2396 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2398 struct kvm_vcpu *vcpu = filp->private_data;
2400 debugfs_remove_recursive(vcpu->debugfs_dentry);
2401 kvm_put_kvm(vcpu->kvm);
2405 static struct file_operations kvm_vcpu_fops = {
2406 .release = kvm_vcpu_release,
2407 .unlocked_ioctl = kvm_vcpu_ioctl,
2408 #ifdef CONFIG_KVM_COMPAT
2409 .compat_ioctl = kvm_vcpu_compat_ioctl,
2411 .mmap = kvm_vcpu_mmap,
2412 .llseek = noop_llseek,
2416 * Allocates an inode for the vcpu.
2418 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2420 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
2423 static int kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
2425 char dir_name[ITOA_MAX_LEN * 2];
2428 if (!kvm_arch_has_vcpu_debugfs())
2431 if (!debugfs_initialized())
2434 snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
2435 vcpu->debugfs_dentry = debugfs_create_dir(dir_name,
2436 vcpu->kvm->debugfs_dentry);
2437 if (!vcpu->debugfs_dentry)
2440 ret = kvm_arch_create_vcpu_debugfs(vcpu);
2442 debugfs_remove_recursive(vcpu->debugfs_dentry);
2450 * Creates some virtual cpus. Good luck creating more than one.
2452 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
2455 struct kvm_vcpu *vcpu;
2457 if (id >= KVM_MAX_VCPU_ID)
2460 mutex_lock(&kvm->lock);
2461 if (kvm->created_vcpus == KVM_MAX_VCPUS) {
2462 mutex_unlock(&kvm->lock);
2466 kvm->created_vcpus++;
2467 mutex_unlock(&kvm->lock);
2469 vcpu = kvm_arch_vcpu_create(kvm, id);
2472 goto vcpu_decrement;
2475 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2477 r = kvm_arch_vcpu_setup(vcpu);
2481 r = kvm_create_vcpu_debugfs(vcpu);
2485 mutex_lock(&kvm->lock);
2486 if (kvm_get_vcpu_by_id(kvm, id)) {
2488 goto unlock_vcpu_destroy;
2491 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
2493 /* Now it's all set up, let userspace reach it */
2495 r = create_vcpu_fd(vcpu);
2498 goto unlock_vcpu_destroy;
2501 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
2504 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
2505 * before kvm->online_vcpu's incremented value.
2508 atomic_inc(&kvm->online_vcpus);
2510 mutex_unlock(&kvm->lock);
2511 kvm_arch_vcpu_postcreate(vcpu);
2514 unlock_vcpu_destroy:
2515 mutex_unlock(&kvm->lock);
2516 debugfs_remove_recursive(vcpu->debugfs_dentry);
2518 kvm_arch_vcpu_destroy(vcpu);
2520 mutex_lock(&kvm->lock);
2521 kvm->created_vcpus--;
2522 mutex_unlock(&kvm->lock);
2526 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2529 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2530 vcpu->sigset_active = 1;
2531 vcpu->sigset = *sigset;
2533 vcpu->sigset_active = 0;
2537 static long kvm_vcpu_ioctl(struct file *filp,
2538 unsigned int ioctl, unsigned long arg)
2540 struct kvm_vcpu *vcpu = filp->private_data;
2541 void __user *argp = (void __user *)arg;
2543 struct kvm_fpu *fpu = NULL;
2544 struct kvm_sregs *kvm_sregs = NULL;
2546 if (vcpu->kvm->mm != current->mm)
2549 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2552 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
2554 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
2555 * so vcpu_load() would break it.
2557 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_S390_IRQ || ioctl == KVM_INTERRUPT)
2558 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2562 r = vcpu_load(vcpu);
2571 oldpid = rcu_access_pointer(vcpu->pid);
2572 if (unlikely(oldpid != current->pids[PIDTYPE_PID].pid)) {
2573 /* The thread running this VCPU changed. */
2574 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
2576 rcu_assign_pointer(vcpu->pid, newpid);
2581 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2582 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2585 case KVM_GET_REGS: {
2586 struct kvm_regs *kvm_regs;
2589 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2592 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2596 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2603 case KVM_SET_REGS: {
2604 struct kvm_regs *kvm_regs;
2607 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2608 if (IS_ERR(kvm_regs)) {
2609 r = PTR_ERR(kvm_regs);
2612 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2616 case KVM_GET_SREGS: {
2617 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2621 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2625 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2630 case KVM_SET_SREGS: {
2631 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2632 if (IS_ERR(kvm_sregs)) {
2633 r = PTR_ERR(kvm_sregs);
2637 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2640 case KVM_GET_MP_STATE: {
2641 struct kvm_mp_state mp_state;
2643 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2647 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
2652 case KVM_SET_MP_STATE: {
2653 struct kvm_mp_state mp_state;
2656 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
2658 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2661 case KVM_TRANSLATE: {
2662 struct kvm_translation tr;
2665 if (copy_from_user(&tr, argp, sizeof(tr)))
2667 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2671 if (copy_to_user(argp, &tr, sizeof(tr)))
2676 case KVM_SET_GUEST_DEBUG: {
2677 struct kvm_guest_debug dbg;
2680 if (copy_from_user(&dbg, argp, sizeof(dbg)))
2682 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2685 case KVM_SET_SIGNAL_MASK: {
2686 struct kvm_signal_mask __user *sigmask_arg = argp;
2687 struct kvm_signal_mask kvm_sigmask;
2688 sigset_t sigset, *p;
2693 if (copy_from_user(&kvm_sigmask, argp,
2694 sizeof(kvm_sigmask)))
2697 if (kvm_sigmask.len != sizeof(sigset))
2700 if (copy_from_user(&sigset, sigmask_arg->sigset,
2705 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2709 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2713 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2717 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2723 fpu = memdup_user(argp, sizeof(*fpu));
2729 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2733 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2742 #ifdef CONFIG_KVM_COMPAT
2743 static long kvm_vcpu_compat_ioctl(struct file *filp,
2744 unsigned int ioctl, unsigned long arg)
2746 struct kvm_vcpu *vcpu = filp->private_data;
2747 void __user *argp = compat_ptr(arg);
2750 if (vcpu->kvm->mm != current->mm)
2754 case KVM_SET_SIGNAL_MASK: {
2755 struct kvm_signal_mask __user *sigmask_arg = argp;
2756 struct kvm_signal_mask kvm_sigmask;
2757 compat_sigset_t csigset;
2762 if (copy_from_user(&kvm_sigmask, argp,
2763 sizeof(kvm_sigmask)))
2766 if (kvm_sigmask.len != sizeof(csigset))
2769 if (copy_from_user(&csigset, sigmask_arg->sigset,
2772 sigset_from_compat(&sigset, &csigset);
2773 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2775 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2779 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2787 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2788 int (*accessor)(struct kvm_device *dev,
2789 struct kvm_device_attr *attr),
2792 struct kvm_device_attr attr;
2797 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2800 return accessor(dev, &attr);
2803 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2806 struct kvm_device *dev = filp->private_data;
2809 case KVM_SET_DEVICE_ATTR:
2810 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2811 case KVM_GET_DEVICE_ATTR:
2812 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2813 case KVM_HAS_DEVICE_ATTR:
2814 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2816 if (dev->ops->ioctl)
2817 return dev->ops->ioctl(dev, ioctl, arg);
2823 static int kvm_device_release(struct inode *inode, struct file *filp)
2825 struct kvm_device *dev = filp->private_data;
2826 struct kvm *kvm = dev->kvm;
2832 static const struct file_operations kvm_device_fops = {
2833 .unlocked_ioctl = kvm_device_ioctl,
2834 #ifdef CONFIG_KVM_COMPAT
2835 .compat_ioctl = kvm_device_ioctl,
2837 .release = kvm_device_release,
2840 struct kvm_device *kvm_device_from_filp(struct file *filp)
2842 if (filp->f_op != &kvm_device_fops)
2845 return filp->private_data;
2848 static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
2849 #ifdef CONFIG_KVM_MPIC
2850 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
2851 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
2855 int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
2857 if (type >= ARRAY_SIZE(kvm_device_ops_table))
2860 if (kvm_device_ops_table[type] != NULL)
2863 kvm_device_ops_table[type] = ops;
2867 void kvm_unregister_device_ops(u32 type)
2869 if (kvm_device_ops_table[type] != NULL)
2870 kvm_device_ops_table[type] = NULL;
2873 static int kvm_ioctl_create_device(struct kvm *kvm,
2874 struct kvm_create_device *cd)
2876 struct kvm_device_ops *ops = NULL;
2877 struct kvm_device *dev;
2878 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2881 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
2884 ops = kvm_device_ops_table[cd->type];
2891 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2898 mutex_lock(&kvm->lock);
2899 ret = ops->create(dev, cd->type);
2901 mutex_unlock(&kvm->lock);
2905 list_add(&dev->vm_node, &kvm->devices);
2906 mutex_unlock(&kvm->lock);
2911 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2913 mutex_lock(&kvm->lock);
2914 list_del(&dev->vm_node);
2915 mutex_unlock(&kvm->lock);
2925 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2928 case KVM_CAP_USER_MEMORY:
2929 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2930 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2931 case KVM_CAP_INTERNAL_ERROR_DATA:
2932 #ifdef CONFIG_HAVE_KVM_MSI
2933 case KVM_CAP_SIGNAL_MSI:
2935 #ifdef CONFIG_HAVE_KVM_IRQFD
2937 case KVM_CAP_IRQFD_RESAMPLE:
2939 case KVM_CAP_IOEVENTFD_ANY_LENGTH:
2940 case KVM_CAP_CHECK_EXTENSION_VM:
2942 #ifdef CONFIG_KVM_MMIO
2943 case KVM_CAP_COALESCED_MMIO:
2944 return KVM_COALESCED_MMIO_PAGE_OFFSET;
2946 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2947 case KVM_CAP_IRQ_ROUTING:
2948 return KVM_MAX_IRQ_ROUTES;
2950 #if KVM_ADDRESS_SPACE_NUM > 1
2951 case KVM_CAP_MULTI_ADDRESS_SPACE:
2952 return KVM_ADDRESS_SPACE_NUM;
2954 case KVM_CAP_MAX_VCPU_ID:
2955 return KVM_MAX_VCPU_ID;
2959 return kvm_vm_ioctl_check_extension(kvm, arg);
2962 static long kvm_vm_ioctl(struct file *filp,
2963 unsigned int ioctl, unsigned long arg)
2965 struct kvm *kvm = filp->private_data;
2966 void __user *argp = (void __user *)arg;
2969 if (kvm->mm != current->mm)
2972 case KVM_CREATE_VCPU:
2973 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2975 case KVM_SET_USER_MEMORY_REGION: {
2976 struct kvm_userspace_memory_region kvm_userspace_mem;
2979 if (copy_from_user(&kvm_userspace_mem, argp,
2980 sizeof(kvm_userspace_mem)))
2983 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2986 case KVM_GET_DIRTY_LOG: {
2987 struct kvm_dirty_log log;
2990 if (copy_from_user(&log, argp, sizeof(log)))
2992 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2995 #ifdef CONFIG_KVM_MMIO
2996 case KVM_REGISTER_COALESCED_MMIO: {
2997 struct kvm_coalesced_mmio_zone zone;
3000 if (copy_from_user(&zone, argp, sizeof(zone)))
3002 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
3005 case KVM_UNREGISTER_COALESCED_MMIO: {
3006 struct kvm_coalesced_mmio_zone zone;
3009 if (copy_from_user(&zone, argp, sizeof(zone)))
3011 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
3016 struct kvm_irqfd data;
3019 if (copy_from_user(&data, argp, sizeof(data)))
3021 r = kvm_irqfd(kvm, &data);
3024 case KVM_IOEVENTFD: {
3025 struct kvm_ioeventfd data;
3028 if (copy_from_user(&data, argp, sizeof(data)))
3030 r = kvm_ioeventfd(kvm, &data);
3033 #ifdef CONFIG_HAVE_KVM_MSI
3034 case KVM_SIGNAL_MSI: {
3038 if (copy_from_user(&msi, argp, sizeof(msi)))
3040 r = kvm_send_userspace_msi(kvm, &msi);
3044 #ifdef __KVM_HAVE_IRQ_LINE
3045 case KVM_IRQ_LINE_STATUS:
3046 case KVM_IRQ_LINE: {
3047 struct kvm_irq_level irq_event;
3050 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
3053 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
3054 ioctl == KVM_IRQ_LINE_STATUS);
3059 if (ioctl == KVM_IRQ_LINE_STATUS) {
3060 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
3068 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3069 case KVM_SET_GSI_ROUTING: {
3070 struct kvm_irq_routing routing;
3071 struct kvm_irq_routing __user *urouting;
3072 struct kvm_irq_routing_entry *entries = NULL;
3075 if (copy_from_user(&routing, argp, sizeof(routing)))
3078 if (!kvm_arch_can_set_irq_routing(kvm))
3080 if (routing.nr > KVM_MAX_IRQ_ROUTES)
3086 entries = vmalloc(routing.nr * sizeof(*entries));
3091 if (copy_from_user(entries, urouting->entries,
3092 routing.nr * sizeof(*entries)))
3093 goto out_free_irq_routing;
3095 r = kvm_set_irq_routing(kvm, entries, routing.nr,
3097 out_free_irq_routing:
3101 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
3102 case KVM_CREATE_DEVICE: {
3103 struct kvm_create_device cd;
3106 if (copy_from_user(&cd, argp, sizeof(cd)))
3109 r = kvm_ioctl_create_device(kvm, &cd);
3114 if (copy_to_user(argp, &cd, sizeof(cd)))
3120 case KVM_CHECK_EXTENSION:
3121 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
3124 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
3130 #ifdef CONFIG_KVM_COMPAT
3131 struct compat_kvm_dirty_log {
3135 compat_uptr_t dirty_bitmap; /* one bit per page */
3140 static long kvm_vm_compat_ioctl(struct file *filp,
3141 unsigned int ioctl, unsigned long arg)
3143 struct kvm *kvm = filp->private_data;
3146 if (kvm->mm != current->mm)
3149 case KVM_GET_DIRTY_LOG: {
3150 struct compat_kvm_dirty_log compat_log;
3151 struct kvm_dirty_log log;
3153 if (copy_from_user(&compat_log, (void __user *)arg,
3154 sizeof(compat_log)))
3156 log.slot = compat_log.slot;
3157 log.padding1 = compat_log.padding1;
3158 log.padding2 = compat_log.padding2;
3159 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
3161 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3165 r = kvm_vm_ioctl(filp, ioctl, arg);
3171 static struct file_operations kvm_vm_fops = {
3172 .release = kvm_vm_release,
3173 .unlocked_ioctl = kvm_vm_ioctl,
3174 #ifdef CONFIG_KVM_COMPAT
3175 .compat_ioctl = kvm_vm_compat_ioctl,
3177 .llseek = noop_llseek,
3180 static int kvm_dev_ioctl_create_vm(unsigned long type)
3186 kvm = kvm_create_vm(type);
3188 return PTR_ERR(kvm);
3189 #ifdef CONFIG_KVM_MMIO
3190 r = kvm_coalesced_mmio_init(kvm);
3196 r = get_unused_fd_flags(O_CLOEXEC);
3201 file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
3205 return PTR_ERR(file);
3209 * Don't call kvm_put_kvm anymore at this point; file->f_op is
3210 * already set, with ->release() being kvm_vm_release(). In error
3211 * cases it will be called by the final fput(file) and will take
3212 * care of doing kvm_put_kvm(kvm).
3214 if (kvm_create_vm_debugfs(kvm, r) < 0) {
3219 kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm);
3221 fd_install(r, file);
3225 static long kvm_dev_ioctl(struct file *filp,
3226 unsigned int ioctl, unsigned long arg)
3231 case KVM_GET_API_VERSION:
3234 r = KVM_API_VERSION;
3237 r = kvm_dev_ioctl_create_vm(arg);
3239 case KVM_CHECK_EXTENSION:
3240 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
3242 case KVM_GET_VCPU_MMAP_SIZE:
3245 r = PAGE_SIZE; /* struct kvm_run */
3247 r += PAGE_SIZE; /* pio data page */
3249 #ifdef CONFIG_KVM_MMIO
3250 r += PAGE_SIZE; /* coalesced mmio ring page */
3253 case KVM_TRACE_ENABLE:
3254 case KVM_TRACE_PAUSE:
3255 case KVM_TRACE_DISABLE:
3259 return kvm_arch_dev_ioctl(filp, ioctl, arg);
3265 static struct file_operations kvm_chardev_ops = {
3266 .unlocked_ioctl = kvm_dev_ioctl,
3267 .compat_ioctl = kvm_dev_ioctl,
3268 .llseek = noop_llseek,
3271 static struct miscdevice kvm_dev = {
3277 static void hardware_enable_nolock(void *junk)
3279 int cpu = raw_smp_processor_id();
3282 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
3285 cpumask_set_cpu(cpu, cpus_hardware_enabled);
3287 r = kvm_arch_hardware_enable();
3290 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3291 atomic_inc(&hardware_enable_failed);
3292 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
3296 static int kvm_starting_cpu(unsigned int cpu)
3298 raw_spin_lock(&kvm_count_lock);
3299 if (kvm_usage_count)
3300 hardware_enable_nolock(NULL);
3301 raw_spin_unlock(&kvm_count_lock);
3305 static void hardware_disable_nolock(void *junk)
3307 int cpu = raw_smp_processor_id();
3309 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
3311 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3312 kvm_arch_hardware_disable();
3315 static int kvm_dying_cpu(unsigned int cpu)
3317 raw_spin_lock(&kvm_count_lock);
3318 if (kvm_usage_count)
3319 hardware_disable_nolock(NULL);
3320 raw_spin_unlock(&kvm_count_lock);
3324 static void hardware_disable_all_nolock(void)
3326 BUG_ON(!kvm_usage_count);
3329 if (!kvm_usage_count)
3330 on_each_cpu(hardware_disable_nolock, NULL, 1);
3333 static void hardware_disable_all(void)
3335 raw_spin_lock(&kvm_count_lock);
3336 hardware_disable_all_nolock();
3337 raw_spin_unlock(&kvm_count_lock);
3340 static int hardware_enable_all(void)
3344 raw_spin_lock(&kvm_count_lock);
3347 if (kvm_usage_count == 1) {
3348 atomic_set(&hardware_enable_failed, 0);
3349 on_each_cpu(hardware_enable_nolock, NULL, 1);
3351 if (atomic_read(&hardware_enable_failed)) {
3352 hardware_disable_all_nolock();
3357 raw_spin_unlock(&kvm_count_lock);
3362 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3366 * Some (well, at least mine) BIOSes hang on reboot if
3369 * And Intel TXT required VMX off for all cpu when system shutdown.
3371 pr_info("kvm: exiting hardware virtualization\n");
3372 kvm_rebooting = true;
3373 on_each_cpu(hardware_disable_nolock, NULL, 1);
3377 static struct notifier_block kvm_reboot_notifier = {
3378 .notifier_call = kvm_reboot,
3382 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3386 for (i = 0; i < bus->dev_count; i++) {
3387 struct kvm_io_device *pos = bus->range[i].dev;
3389 kvm_iodevice_destructor(pos);
3394 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
3395 const struct kvm_io_range *r2)
3397 gpa_t addr1 = r1->addr;
3398 gpa_t addr2 = r2->addr;
3403 /* If r2->len == 0, match the exact address. If r2->len != 0,
3404 * accept any overlapping write. Any order is acceptable for
3405 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
3406 * we process all of them.
3419 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
3421 return kvm_io_bus_cmp(p1, p2);
3424 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
3425 gpa_t addr, int len)
3427 bus->range[bus->dev_count++] = (struct kvm_io_range) {
3433 sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
3434 kvm_io_bus_sort_cmp, NULL);
3439 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
3440 gpa_t addr, int len)
3442 struct kvm_io_range *range, key;
3445 key = (struct kvm_io_range) {
3450 range = bsearch(&key, bus->range, bus->dev_count,
3451 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
3455 off = range - bus->range;
3457 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
3463 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3464 struct kvm_io_range *range, const void *val)
3468 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3472 while (idx < bus->dev_count &&
3473 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3474 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
3483 /* kvm_io_bus_write - called under kvm->slots_lock */
3484 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3485 int len, const void *val)
3487 struct kvm_io_bus *bus;
3488 struct kvm_io_range range;
3491 range = (struct kvm_io_range) {
3496 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3499 r = __kvm_io_bus_write(vcpu, bus, &range, val);
3500 return r < 0 ? r : 0;
3503 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3504 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
3505 gpa_t addr, int len, const void *val, long cookie)
3507 struct kvm_io_bus *bus;
3508 struct kvm_io_range range;
3510 range = (struct kvm_io_range) {
3515 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3519 /* First try the device referenced by cookie. */
3520 if ((cookie >= 0) && (cookie < bus->dev_count) &&
3521 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
3522 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
3527 * cookie contained garbage; fall back to search and return the
3528 * correct cookie value.
3530 return __kvm_io_bus_write(vcpu, bus, &range, val);
3533 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3534 struct kvm_io_range *range, void *val)
3538 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3542 while (idx < bus->dev_count &&
3543 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3544 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
3552 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
3554 /* kvm_io_bus_read - called under kvm->slots_lock */
3555 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3558 struct kvm_io_bus *bus;
3559 struct kvm_io_range range;
3562 range = (struct kvm_io_range) {
3567 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3570 r = __kvm_io_bus_read(vcpu, bus, &range, val);
3571 return r < 0 ? r : 0;
3575 /* Caller must hold slots_lock. */
3576 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3577 int len, struct kvm_io_device *dev)
3579 struct kvm_io_bus *new_bus, *bus;
3581 bus = kvm_get_bus(kvm, bus_idx);
3585 /* exclude ioeventfd which is limited by maximum fd */
3586 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3589 new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count + 1) *
3590 sizeof(struct kvm_io_range)), GFP_KERNEL);
3593 memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
3594 sizeof(struct kvm_io_range)));
3595 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
3596 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3597 synchronize_srcu_expedited(&kvm->srcu);
3603 /* Caller must hold slots_lock. */
3604 void kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3605 struct kvm_io_device *dev)
3608 struct kvm_io_bus *new_bus, *bus;
3610 bus = kvm_get_bus(kvm, bus_idx);
3614 for (i = 0; i < bus->dev_count; i++)
3615 if (bus->range[i].dev == dev) {
3619 if (i == bus->dev_count)
3622 new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3623 sizeof(struct kvm_io_range)), GFP_KERNEL);
3625 pr_err("kvm: failed to shrink bus, removing it completely\n");
3629 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3630 new_bus->dev_count--;
3631 memcpy(new_bus->range + i, bus->range + i + 1,
3632 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3635 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3636 synchronize_srcu_expedited(&kvm->srcu);
3641 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3644 struct kvm_io_bus *bus;
3645 int dev_idx, srcu_idx;
3646 struct kvm_io_device *iodev = NULL;
3648 srcu_idx = srcu_read_lock(&kvm->srcu);
3650 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
3654 dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
3658 iodev = bus->range[dev_idx].dev;
3661 srcu_read_unlock(&kvm->srcu, srcu_idx);
3665 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
3667 static int kvm_debugfs_open(struct inode *inode, struct file *file,
3668 int (*get)(void *, u64 *), int (*set)(void *, u64),
3671 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
3674 /* The debugfs files are a reference to the kvm struct which
3675 * is still valid when kvm_destroy_vm is called.
3676 * To avoid the race between open and the removal of the debugfs
3677 * directory we test against the users count.
3679 if (!refcount_inc_not_zero(&stat_data->kvm->users_count))
3682 if (simple_attr_open(inode, file, get, set, fmt)) {
3683 kvm_put_kvm(stat_data->kvm);
3690 static int kvm_debugfs_release(struct inode *inode, struct file *file)
3692 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
3695 simple_attr_release(inode, file);
3696 kvm_put_kvm(stat_data->kvm);
3701 static int vm_stat_get_per_vm(void *data, u64 *val)
3703 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3705 *val = *(ulong *)((void *)stat_data->kvm + stat_data->offset);
3710 static int vm_stat_clear_per_vm(void *data, u64 val)
3712 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3717 *(ulong *)((void *)stat_data->kvm + stat_data->offset) = 0;
3722 static int vm_stat_get_per_vm_open(struct inode *inode, struct file *file)
3724 __simple_attr_check_format("%llu\n", 0ull);
3725 return kvm_debugfs_open(inode, file, vm_stat_get_per_vm,
3726 vm_stat_clear_per_vm, "%llu\n");
3729 static const struct file_operations vm_stat_get_per_vm_fops = {
3730 .owner = THIS_MODULE,
3731 .open = vm_stat_get_per_vm_open,
3732 .release = kvm_debugfs_release,
3733 .read = simple_attr_read,
3734 .write = simple_attr_write,
3735 .llseek = no_llseek,
3738 static int vcpu_stat_get_per_vm(void *data, u64 *val)
3741 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3742 struct kvm_vcpu *vcpu;
3746 kvm_for_each_vcpu(i, vcpu, stat_data->kvm)
3747 *val += *(u64 *)((void *)vcpu + stat_data->offset);
3752 static int vcpu_stat_clear_per_vm(void *data, u64 val)
3755 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3756 struct kvm_vcpu *vcpu;
3761 kvm_for_each_vcpu(i, vcpu, stat_data->kvm)
3762 *(u64 *)((void *)vcpu + stat_data->offset) = 0;
3767 static int vcpu_stat_get_per_vm_open(struct inode *inode, struct file *file)
3769 __simple_attr_check_format("%llu\n", 0ull);
3770 return kvm_debugfs_open(inode, file, vcpu_stat_get_per_vm,
3771 vcpu_stat_clear_per_vm, "%llu\n");
3774 static const struct file_operations vcpu_stat_get_per_vm_fops = {
3775 .owner = THIS_MODULE,
3776 .open = vcpu_stat_get_per_vm_open,
3777 .release = kvm_debugfs_release,
3778 .read = simple_attr_read,
3779 .write = simple_attr_write,
3780 .llseek = no_llseek,
3783 static const struct file_operations *stat_fops_per_vm[] = {
3784 [KVM_STAT_VCPU] = &vcpu_stat_get_per_vm_fops,
3785 [KVM_STAT_VM] = &vm_stat_get_per_vm_fops,
3788 static int vm_stat_get(void *_offset, u64 *val)
3790 unsigned offset = (long)_offset;
3792 struct kvm_stat_data stat_tmp = {.offset = offset};
3796 spin_lock(&kvm_lock);
3797 list_for_each_entry(kvm, &vm_list, vm_list) {
3799 vm_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
3802 spin_unlock(&kvm_lock);
3806 static int vm_stat_clear(void *_offset, u64 val)
3808 unsigned offset = (long)_offset;
3810 struct kvm_stat_data stat_tmp = {.offset = offset};
3815 spin_lock(&kvm_lock);
3816 list_for_each_entry(kvm, &vm_list, vm_list) {
3818 vm_stat_clear_per_vm((void *)&stat_tmp, 0);
3820 spin_unlock(&kvm_lock);
3825 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n");
3827 static int vcpu_stat_get(void *_offset, u64 *val)
3829 unsigned offset = (long)_offset;
3831 struct kvm_stat_data stat_tmp = {.offset = offset};
3835 spin_lock(&kvm_lock);
3836 list_for_each_entry(kvm, &vm_list, vm_list) {
3838 vcpu_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
3841 spin_unlock(&kvm_lock);
3845 static int vcpu_stat_clear(void *_offset, u64 val)
3847 unsigned offset = (long)_offset;
3849 struct kvm_stat_data stat_tmp = {.offset = offset};
3854 spin_lock(&kvm_lock);
3855 list_for_each_entry(kvm, &vm_list, vm_list) {
3857 vcpu_stat_clear_per_vm((void *)&stat_tmp, 0);
3859 spin_unlock(&kvm_lock);
3864 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear,
3867 static const struct file_operations *stat_fops[] = {
3868 [KVM_STAT_VCPU] = &vcpu_stat_fops,
3869 [KVM_STAT_VM] = &vm_stat_fops,
3872 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm)
3874 struct kobj_uevent_env *env;
3875 unsigned long long created, active;
3877 if (!kvm_dev.this_device || !kvm)
3880 spin_lock(&kvm_lock);
3881 if (type == KVM_EVENT_CREATE_VM) {
3882 kvm_createvm_count++;
3884 } else if (type == KVM_EVENT_DESTROY_VM) {
3887 created = kvm_createvm_count;
3888 active = kvm_active_vms;
3889 spin_unlock(&kvm_lock);
3891 env = kzalloc(sizeof(*env), GFP_KERNEL);
3895 add_uevent_var(env, "CREATED=%llu", created);
3896 add_uevent_var(env, "COUNT=%llu", active);
3898 if (type == KVM_EVENT_CREATE_VM) {
3899 add_uevent_var(env, "EVENT=create");
3900 kvm->userspace_pid = task_pid_nr(current);
3901 } else if (type == KVM_EVENT_DESTROY_VM) {
3902 add_uevent_var(env, "EVENT=destroy");
3904 add_uevent_var(env, "PID=%d", kvm->userspace_pid);
3906 if (kvm->debugfs_dentry) {
3907 char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL);
3910 tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX);
3912 add_uevent_var(env, "STATS_PATH=%s", tmp);
3916 /* no need for checks, since we are adding at most only 5 keys */
3917 env->envp[env->envp_idx++] = NULL;
3918 kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp);
3922 static int kvm_init_debug(void)
3925 struct kvm_stats_debugfs_item *p;
3927 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3928 if (kvm_debugfs_dir == NULL)
3931 kvm_debugfs_num_entries = 0;
3932 for (p = debugfs_entries; p->name; ++p, kvm_debugfs_num_entries++) {
3933 if (!debugfs_create_file(p->name, 0644, kvm_debugfs_dir,
3934 (void *)(long)p->offset,
3935 stat_fops[p->kind]))
3942 debugfs_remove_recursive(kvm_debugfs_dir);
3947 static int kvm_suspend(void)
3949 if (kvm_usage_count)
3950 hardware_disable_nolock(NULL);
3954 static void kvm_resume(void)
3956 if (kvm_usage_count) {
3957 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3958 hardware_enable_nolock(NULL);
3962 static struct syscore_ops kvm_syscore_ops = {
3963 .suspend = kvm_suspend,
3964 .resume = kvm_resume,
3968 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3970 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3973 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3975 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3977 if (vcpu->preempted)
3978 vcpu->preempted = false;
3980 kvm_arch_sched_in(vcpu, cpu);
3982 kvm_arch_vcpu_load(vcpu, cpu);
3985 static void kvm_sched_out(struct preempt_notifier *pn,
3986 struct task_struct *next)
3988 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3990 if (current->state == TASK_RUNNING)
3991 vcpu->preempted = true;
3992 kvm_arch_vcpu_put(vcpu);
3995 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3996 struct module *module)
4001 r = kvm_arch_init(opaque);
4006 * kvm_arch_init makes sure there's at most one caller
4007 * for architectures that support multiple implementations,
4008 * like intel and amd on x86.
4009 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
4010 * conflicts in case kvm is already setup for another implementation.
4012 r = kvm_irqfd_init();
4016 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
4021 r = kvm_arch_hardware_setup();
4025 for_each_online_cpu(cpu) {
4026 smp_call_function_single(cpu,
4027 kvm_arch_check_processor_compat,
4033 r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting",
4034 kvm_starting_cpu, kvm_dying_cpu);
4037 register_reboot_notifier(&kvm_reboot_notifier);
4039 /* A kmem cache lets us meet the alignment requirements of fx_save. */
4041 vcpu_align = __alignof__(struct kvm_vcpu);
4042 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
4043 SLAB_ACCOUNT, NULL);
4044 if (!kvm_vcpu_cache) {
4049 r = kvm_async_pf_init();
4053 kvm_chardev_ops.owner = module;
4054 kvm_vm_fops.owner = module;
4055 kvm_vcpu_fops.owner = module;
4057 r = misc_register(&kvm_dev);
4059 pr_err("kvm: misc device register failed\n");
4063 register_syscore_ops(&kvm_syscore_ops);
4065 kvm_preempt_ops.sched_in = kvm_sched_in;
4066 kvm_preempt_ops.sched_out = kvm_sched_out;
4068 r = kvm_init_debug();
4070 pr_err("kvm: create debugfs files failed\n");
4074 r = kvm_vfio_ops_init();
4080 unregister_syscore_ops(&kvm_syscore_ops);
4081 misc_deregister(&kvm_dev);
4083 kvm_async_pf_deinit();
4085 kmem_cache_destroy(kvm_vcpu_cache);
4087 unregister_reboot_notifier(&kvm_reboot_notifier);
4088 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4091 kvm_arch_hardware_unsetup();
4093 free_cpumask_var(cpus_hardware_enabled);
4101 EXPORT_SYMBOL_GPL(kvm_init);
4105 debugfs_remove_recursive(kvm_debugfs_dir);
4106 misc_deregister(&kvm_dev);
4107 kmem_cache_destroy(kvm_vcpu_cache);
4108 kvm_async_pf_deinit();
4109 unregister_syscore_ops(&kvm_syscore_ops);
4110 unregister_reboot_notifier(&kvm_reboot_notifier);
4111 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4112 on_each_cpu(hardware_disable_nolock, NULL, 1);
4113 kvm_arch_hardware_unsetup();
4116 free_cpumask_var(cpus_hardware_enabled);
4117 kvm_vfio_ops_exit();
4119 EXPORT_SYMBOL_GPL(kvm_exit);