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)) {
978 if ((slot->id >= KVM_USER_MEM_SLOTS) ||
981 if (!((base_gfn + npages <= slot->base_gfn) ||
982 (base_gfn >= slot->base_gfn + slot->npages)))
987 /* Free page dirty bitmap if unneeded */
988 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
989 new.dirty_bitmap = NULL;
992 if (change == KVM_MR_CREATE) {
993 new.userspace_addr = mem->userspace_addr;
995 if (kvm_arch_create_memslot(kvm, &new, npages))
999 /* Allocate page dirty bitmap if needed */
1000 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
1001 if (kvm_create_dirty_bitmap(&new) < 0)
1005 slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
1008 memcpy(slots, __kvm_memslots(kvm, as_id), sizeof(struct kvm_memslots));
1010 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
1011 slot = id_to_memslot(slots, id);
1012 slot->flags |= KVM_MEMSLOT_INVALID;
1014 old_memslots = install_new_memslots(kvm, as_id, slots);
1016 /* From this point no new shadow pages pointing to a deleted,
1017 * or moved, memslot will be created.
1019 * validation of sp->gfn happens in:
1020 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1021 * - kvm_is_visible_gfn (mmu_check_roots)
1023 kvm_arch_flush_shadow_memslot(kvm, slot);
1026 * We can re-use the old_memslots from above, the only difference
1027 * from the currently installed memslots is the invalid flag. This
1028 * will get overwritten by update_memslots anyway.
1030 slots = old_memslots;
1033 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
1037 /* actual memory is freed via old in kvm_free_memslot below */
1038 if (change == KVM_MR_DELETE) {
1039 new.dirty_bitmap = NULL;
1040 memset(&new.arch, 0, sizeof(new.arch));
1043 update_memslots(slots, &new);
1044 old_memslots = install_new_memslots(kvm, as_id, slots);
1046 kvm_arch_commit_memory_region(kvm, mem, &old, &new, change);
1048 kvm_free_memslot(kvm, &old, &new);
1049 kvfree(old_memslots);
1055 kvm_free_memslot(kvm, &new, &old);
1059 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1061 int kvm_set_memory_region(struct kvm *kvm,
1062 const struct kvm_userspace_memory_region *mem)
1066 mutex_lock(&kvm->slots_lock);
1067 r = __kvm_set_memory_region(kvm, mem);
1068 mutex_unlock(&kvm->slots_lock);
1071 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1073 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1074 struct kvm_userspace_memory_region *mem)
1076 if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1079 return kvm_set_memory_region(kvm, mem);
1082 int kvm_get_dirty_log(struct kvm *kvm,
1083 struct kvm_dirty_log *log, int *is_dirty)
1085 struct kvm_memslots *slots;
1086 struct kvm_memory_slot *memslot;
1089 unsigned long any = 0;
1091 as_id = log->slot >> 16;
1092 id = (u16)log->slot;
1093 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1096 slots = __kvm_memslots(kvm, as_id);
1097 memslot = id_to_memslot(slots, id);
1098 if (!memslot->dirty_bitmap)
1101 n = kvm_dirty_bitmap_bytes(memslot);
1103 for (i = 0; !any && i < n/sizeof(long); ++i)
1104 any = memslot->dirty_bitmap[i];
1106 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1113 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1115 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1117 * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages
1118 * are dirty write protect them for next write.
1119 * @kvm: pointer to kvm instance
1120 * @log: slot id and address to which we copy the log
1121 * @is_dirty: flag set if any page is dirty
1123 * We need to keep it in mind that VCPU threads can write to the bitmap
1124 * concurrently. So, to avoid losing track of dirty pages we keep the
1127 * 1. Take a snapshot of the bit and clear it if needed.
1128 * 2. Write protect the corresponding page.
1129 * 3. Copy the snapshot to the userspace.
1130 * 4. Upon return caller flushes TLB's if needed.
1132 * Between 2 and 4, the guest may write to the page using the remaining TLB
1133 * entry. This is not a problem because the page is reported dirty using
1134 * the snapshot taken before and step 4 ensures that writes done after
1135 * exiting to userspace will be logged for the next call.
1138 int kvm_get_dirty_log_protect(struct kvm *kvm,
1139 struct kvm_dirty_log *log, bool *is_dirty)
1141 struct kvm_memslots *slots;
1142 struct kvm_memory_slot *memslot;
1145 unsigned long *dirty_bitmap;
1146 unsigned long *dirty_bitmap_buffer;
1148 as_id = log->slot >> 16;
1149 id = (u16)log->slot;
1150 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1153 slots = __kvm_memslots(kvm, as_id);
1154 memslot = id_to_memslot(slots, id);
1156 dirty_bitmap = memslot->dirty_bitmap;
1160 n = kvm_dirty_bitmap_bytes(memslot);
1162 dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long);
1163 memset(dirty_bitmap_buffer, 0, n);
1165 spin_lock(&kvm->mmu_lock);
1167 for (i = 0; i < n / sizeof(long); i++) {
1171 if (!dirty_bitmap[i])
1176 mask = xchg(&dirty_bitmap[i], 0);
1177 dirty_bitmap_buffer[i] = mask;
1180 offset = i * BITS_PER_LONG;
1181 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1186 spin_unlock(&kvm->mmu_lock);
1187 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1191 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect);
1194 bool kvm_largepages_enabled(void)
1196 return largepages_enabled;
1199 void kvm_disable_largepages(void)
1201 largepages_enabled = false;
1203 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1205 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1207 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1209 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1211 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
1213 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn);
1216 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1218 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1220 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1221 memslot->flags & KVM_MEMSLOT_INVALID)
1226 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1228 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1230 struct vm_area_struct *vma;
1231 unsigned long addr, size;
1235 addr = gfn_to_hva(kvm, gfn);
1236 if (kvm_is_error_hva(addr))
1239 down_read(¤t->mm->mmap_sem);
1240 vma = find_vma(current->mm, addr);
1244 size = vma_kernel_pagesize(vma);
1247 up_read(¤t->mm->mmap_sem);
1252 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1254 return slot->flags & KVM_MEM_READONLY;
1257 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1258 gfn_t *nr_pages, bool write)
1260 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1261 return KVM_HVA_ERR_BAD;
1263 if (memslot_is_readonly(slot) && write)
1264 return KVM_HVA_ERR_RO_BAD;
1267 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1269 return __gfn_to_hva_memslot(slot, gfn);
1272 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1275 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1278 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1281 return gfn_to_hva_many(slot, gfn, NULL);
1283 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1285 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1287 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1289 EXPORT_SYMBOL_GPL(gfn_to_hva);
1291 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
1293 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
1295 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
1298 * If writable is set to false, the hva returned by this function is only
1299 * allowed to be read.
1301 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1302 gfn_t gfn, bool *writable)
1304 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1306 if (!kvm_is_error_hva(hva) && writable)
1307 *writable = !memslot_is_readonly(slot);
1312 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1314 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1316 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1319 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
1321 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1323 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1326 static int get_user_page_nowait(unsigned long start, int write,
1329 int flags = FOLL_NOWAIT | FOLL_HWPOISON;
1332 flags |= FOLL_WRITE;
1334 return get_user_pages(start, 1, flags, page, NULL);
1337 static inline int check_user_page_hwpoison(unsigned long addr)
1339 int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
1341 rc = get_user_pages(addr, 1, flags, NULL, NULL);
1342 return rc == -EHWPOISON;
1346 * The atomic path to get the writable pfn which will be stored in @pfn,
1347 * true indicates success, otherwise false is returned.
1349 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1350 bool write_fault, bool *writable, kvm_pfn_t *pfn)
1352 struct page *page[1];
1355 if (!(async || atomic))
1359 * Fast pin a writable pfn only if it is a write fault request
1360 * or the caller allows to map a writable pfn for a read fault
1363 if (!(write_fault || writable))
1366 npages = __get_user_pages_fast(addr, 1, 1, page);
1368 *pfn = page_to_pfn(page[0]);
1379 * The slow path to get the pfn of the specified host virtual address,
1380 * 1 indicates success, -errno is returned if error is detected.
1382 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1383 bool *writable, kvm_pfn_t *pfn)
1385 struct page *page[1];
1391 *writable = write_fault;
1394 down_read(¤t->mm->mmap_sem);
1395 npages = get_user_page_nowait(addr, write_fault, page);
1396 up_read(¤t->mm->mmap_sem);
1398 unsigned int flags = FOLL_HWPOISON;
1401 flags |= FOLL_WRITE;
1403 npages = get_user_pages_unlocked(addr, 1, page, flags);
1408 /* map read fault as writable if possible */
1409 if (unlikely(!write_fault) && writable) {
1410 struct page *wpage[1];
1412 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1421 *pfn = page_to_pfn(page[0]);
1425 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1427 if (unlikely(!(vma->vm_flags & VM_READ)))
1430 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1436 static int hva_to_pfn_remapped(struct vm_area_struct *vma,
1437 unsigned long addr, bool *async,
1438 bool write_fault, kvm_pfn_t *p_pfn)
1443 r = follow_pfn(vma, addr, &pfn);
1446 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
1447 * not call the fault handler, so do it here.
1449 bool unlocked = false;
1450 r = fixup_user_fault(current, current->mm, addr,
1451 (write_fault ? FAULT_FLAG_WRITE : 0),
1458 r = follow_pfn(vma, addr, &pfn);
1466 * Get a reference here because callers of *hva_to_pfn* and
1467 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
1468 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
1469 * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will
1470 * simply do nothing for reserved pfns.
1472 * Whoever called remap_pfn_range is also going to call e.g.
1473 * unmap_mapping_range before the underlying pages are freed,
1474 * causing a call to our MMU notifier.
1483 * Pin guest page in memory and return its pfn.
1484 * @addr: host virtual address which maps memory to the guest
1485 * @atomic: whether this function can sleep
1486 * @async: whether this function need to wait IO complete if the
1487 * host page is not in the memory
1488 * @write_fault: whether we should get a writable host page
1489 * @writable: whether it allows to map a writable host page for !@write_fault
1491 * The function will map a writable host page for these two cases:
1492 * 1): @write_fault = true
1493 * 2): @write_fault = false && @writable, @writable will tell the caller
1494 * whether the mapping is writable.
1496 static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1497 bool write_fault, bool *writable)
1499 struct vm_area_struct *vma;
1503 /* we can do it either atomically or asynchronously, not both */
1504 BUG_ON(atomic && async);
1506 if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1510 return KVM_PFN_ERR_FAULT;
1512 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1516 down_read(¤t->mm->mmap_sem);
1517 if (npages == -EHWPOISON ||
1518 (!async && check_user_page_hwpoison(addr))) {
1519 pfn = KVM_PFN_ERR_HWPOISON;
1524 vma = find_vma_intersection(current->mm, addr, addr + 1);
1527 pfn = KVM_PFN_ERR_FAULT;
1528 else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
1529 r = hva_to_pfn_remapped(vma, addr, async, write_fault, &pfn);
1533 pfn = KVM_PFN_ERR_FAULT;
1535 if (async && vma_is_valid(vma, write_fault))
1537 pfn = KVM_PFN_ERR_FAULT;
1540 up_read(¤t->mm->mmap_sem);
1544 kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn,
1545 bool atomic, bool *async, bool write_fault,
1548 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1550 if (addr == KVM_HVA_ERR_RO_BAD) {
1553 return KVM_PFN_ERR_RO_FAULT;
1556 if (kvm_is_error_hva(addr)) {
1559 return KVM_PFN_NOSLOT;
1562 /* Do not map writable pfn in the readonly memslot. */
1563 if (writable && memslot_is_readonly(slot)) {
1568 return hva_to_pfn(addr, atomic, async, write_fault,
1571 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
1573 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1576 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
1577 write_fault, writable);
1579 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1581 kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1583 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1585 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
1587 kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1589 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1591 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1593 kvm_pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1595 return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm, gfn), gfn);
1597 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1599 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
1601 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1603 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
1605 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1607 return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
1609 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1611 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
1613 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1615 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
1617 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1618 struct page **pages, int nr_pages)
1623 addr = gfn_to_hva_many(slot, gfn, &entry);
1624 if (kvm_is_error_hva(addr))
1627 if (entry < nr_pages)
1630 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1632 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1634 static struct page *kvm_pfn_to_page(kvm_pfn_t pfn)
1636 if (is_error_noslot_pfn(pfn))
1637 return KVM_ERR_PTR_BAD_PAGE;
1639 if (kvm_is_reserved_pfn(pfn)) {
1641 return KVM_ERR_PTR_BAD_PAGE;
1644 return pfn_to_page(pfn);
1647 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1651 pfn = gfn_to_pfn(kvm, gfn);
1653 return kvm_pfn_to_page(pfn);
1655 EXPORT_SYMBOL_GPL(gfn_to_page);
1657 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
1661 pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
1663 return kvm_pfn_to_page(pfn);
1665 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
1667 void kvm_release_page_clean(struct page *page)
1669 WARN_ON(is_error_page(page));
1671 kvm_release_pfn_clean(page_to_pfn(page));
1673 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1675 void kvm_release_pfn_clean(kvm_pfn_t pfn)
1677 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
1678 put_page(pfn_to_page(pfn));
1680 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1682 void kvm_release_page_dirty(struct page *page)
1684 WARN_ON(is_error_page(page));
1686 kvm_release_pfn_dirty(page_to_pfn(page));
1688 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1690 static void kvm_release_pfn_dirty(kvm_pfn_t pfn)
1692 kvm_set_pfn_dirty(pfn);
1693 kvm_release_pfn_clean(pfn);
1696 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
1698 if (!kvm_is_reserved_pfn(pfn)) {
1699 struct page *page = pfn_to_page(pfn);
1701 if (!PageReserved(page))
1705 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1707 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
1709 if (!kvm_is_reserved_pfn(pfn))
1710 mark_page_accessed(pfn_to_page(pfn));
1712 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1714 void kvm_get_pfn(kvm_pfn_t pfn)
1716 if (!kvm_is_reserved_pfn(pfn))
1717 get_page(pfn_to_page(pfn));
1719 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1721 static int next_segment(unsigned long len, int offset)
1723 if (len > PAGE_SIZE - offset)
1724 return PAGE_SIZE - offset;
1729 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
1730 void *data, int offset, int len)
1735 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1736 if (kvm_is_error_hva(addr))
1738 r = __copy_from_user(data, (void __user *)addr + offset, len);
1744 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1747 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1749 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1751 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1753 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
1754 int offset, int len)
1756 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1758 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1760 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
1762 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1764 gfn_t gfn = gpa >> PAGE_SHIFT;
1766 int offset = offset_in_page(gpa);
1769 while ((seg = next_segment(len, offset)) != 0) {
1770 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1780 EXPORT_SYMBOL_GPL(kvm_read_guest);
1782 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
1784 gfn_t gfn = gpa >> PAGE_SHIFT;
1786 int offset = offset_in_page(gpa);
1789 while ((seg = next_segment(len, offset)) != 0) {
1790 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
1800 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
1802 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1803 void *data, int offset, unsigned long len)
1808 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1809 if (kvm_is_error_hva(addr))
1811 pagefault_disable();
1812 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1819 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1822 gfn_t gfn = gpa >> PAGE_SHIFT;
1823 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1824 int offset = offset_in_page(gpa);
1826 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1828 EXPORT_SYMBOL_GPL(kvm_read_guest_atomic);
1830 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
1831 void *data, unsigned long len)
1833 gfn_t gfn = gpa >> PAGE_SHIFT;
1834 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1835 int offset = offset_in_page(gpa);
1837 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1839 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
1841 static int __kvm_write_guest_page(struct kvm_memory_slot *memslot, gfn_t gfn,
1842 const void *data, int offset, int len)
1847 addr = gfn_to_hva_memslot(memslot, gfn);
1848 if (kvm_is_error_hva(addr))
1850 r = __copy_to_user((void __user *)addr + offset, data, len);
1853 mark_page_dirty_in_slot(memslot, gfn);
1857 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
1858 const void *data, int offset, int len)
1860 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1862 return __kvm_write_guest_page(slot, gfn, data, offset, len);
1864 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1866 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
1867 const void *data, int offset, int len)
1869 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1871 return __kvm_write_guest_page(slot, gfn, data, offset, len);
1873 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
1875 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1878 gfn_t gfn = gpa >> PAGE_SHIFT;
1880 int offset = offset_in_page(gpa);
1883 while ((seg = next_segment(len, offset)) != 0) {
1884 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1894 EXPORT_SYMBOL_GPL(kvm_write_guest);
1896 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
1899 gfn_t gfn = gpa >> PAGE_SHIFT;
1901 int offset = offset_in_page(gpa);
1904 while ((seg = next_segment(len, offset)) != 0) {
1905 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
1915 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
1917 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
1918 struct gfn_to_hva_cache *ghc,
1919 gpa_t gpa, unsigned long len)
1921 int offset = offset_in_page(gpa);
1922 gfn_t start_gfn = gpa >> PAGE_SHIFT;
1923 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1924 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1925 gfn_t nr_pages_avail;
1928 ghc->generation = slots->generation;
1930 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
1931 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, NULL);
1932 if (!kvm_is_error_hva(ghc->hva) && nr_pages_needed <= 1) {
1936 * If the requested region crosses two memslots, we still
1937 * verify that the entire region is valid here.
1939 while (start_gfn <= end_gfn) {
1941 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
1942 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1944 if (kvm_is_error_hva(ghc->hva))
1946 start_gfn += nr_pages_avail;
1948 /* Use the slow path for cross page reads and writes. */
1949 ghc->memslot = NULL;
1954 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1955 gpa_t gpa, unsigned long len)
1957 struct kvm_memslots *slots = kvm_memslots(kvm);
1958 return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len);
1960 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1962 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1963 void *data, int offset, unsigned long len)
1965 struct kvm_memslots *slots = kvm_memslots(kvm);
1967 gpa_t gpa = ghc->gpa + offset;
1969 BUG_ON(len + offset > ghc->len);
1971 if (slots->generation != ghc->generation)
1972 __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len);
1974 if (unlikely(!ghc->memslot))
1975 return kvm_write_guest(kvm, gpa, data, len);
1977 if (kvm_is_error_hva(ghc->hva))
1980 r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
1983 mark_page_dirty_in_slot(ghc->memslot, gpa >> PAGE_SHIFT);
1987 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached);
1989 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1990 void *data, unsigned long len)
1992 return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len);
1994 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1996 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1997 void *data, unsigned long len)
1999 struct kvm_memslots *slots = kvm_memslots(kvm);
2002 BUG_ON(len > ghc->len);
2004 if (slots->generation != ghc->generation)
2005 __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len);
2007 if (unlikely(!ghc->memslot))
2008 return kvm_read_guest(kvm, ghc->gpa, data, len);
2010 if (kvm_is_error_hva(ghc->hva))
2013 r = __copy_from_user(data, (void __user *)ghc->hva, len);
2019 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
2021 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
2023 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
2025 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
2027 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
2029 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
2031 gfn_t gfn = gpa >> PAGE_SHIFT;
2033 int offset = offset_in_page(gpa);
2036 while ((seg = next_segment(len, offset)) != 0) {
2037 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
2046 EXPORT_SYMBOL_GPL(kvm_clear_guest);
2048 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot,
2051 if (memslot && memslot->dirty_bitmap) {
2052 unsigned long rel_gfn = gfn - memslot->base_gfn;
2054 set_bit_le(rel_gfn, memslot->dirty_bitmap);
2058 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
2060 struct kvm_memory_slot *memslot;
2062 memslot = gfn_to_memslot(kvm, gfn);
2063 mark_page_dirty_in_slot(memslot, gfn);
2065 EXPORT_SYMBOL_GPL(mark_page_dirty);
2067 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
2069 struct kvm_memory_slot *memslot;
2071 memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2072 mark_page_dirty_in_slot(memslot, gfn);
2074 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
2076 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
2078 unsigned int old, val, grow;
2080 old = val = vcpu->halt_poll_ns;
2081 grow = READ_ONCE(halt_poll_ns_grow);
2083 if (val == 0 && grow)
2088 if (val > halt_poll_ns)
2091 vcpu->halt_poll_ns = val;
2092 trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
2095 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
2097 unsigned int old, val, shrink;
2099 old = val = vcpu->halt_poll_ns;
2100 shrink = READ_ONCE(halt_poll_ns_shrink);
2106 vcpu->halt_poll_ns = val;
2107 trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
2110 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
2112 if (kvm_arch_vcpu_runnable(vcpu)) {
2113 kvm_make_request(KVM_REQ_UNHALT, vcpu);
2116 if (kvm_cpu_has_pending_timer(vcpu))
2118 if (signal_pending(current))
2125 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2127 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
2130 DECLARE_SWAITQUEUE(wait);
2131 bool waited = false;
2134 start = cur = ktime_get();
2135 if (vcpu->halt_poll_ns) {
2136 ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns);
2138 ++vcpu->stat.halt_attempted_poll;
2141 * This sets KVM_REQ_UNHALT if an interrupt
2144 if (kvm_vcpu_check_block(vcpu) < 0) {
2145 ++vcpu->stat.halt_successful_poll;
2146 if (!vcpu_valid_wakeup(vcpu))
2147 ++vcpu->stat.halt_poll_invalid;
2151 } while (single_task_running() && ktime_before(cur, stop));
2154 kvm_arch_vcpu_blocking(vcpu);
2157 prepare_to_swait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
2159 if (kvm_vcpu_check_block(vcpu) < 0)
2166 finish_swait(&vcpu->wq, &wait);
2169 kvm_arch_vcpu_unblocking(vcpu);
2171 block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
2173 if (!vcpu_valid_wakeup(vcpu))
2174 shrink_halt_poll_ns(vcpu);
2175 else if (halt_poll_ns) {
2176 if (block_ns <= vcpu->halt_poll_ns)
2178 /* we had a long block, shrink polling */
2179 else if (vcpu->halt_poll_ns && block_ns > halt_poll_ns)
2180 shrink_halt_poll_ns(vcpu);
2181 /* we had a short halt and our poll time is too small */
2182 else if (vcpu->halt_poll_ns < halt_poll_ns &&
2183 block_ns < halt_poll_ns)
2184 grow_halt_poll_ns(vcpu);
2186 vcpu->halt_poll_ns = 0;
2188 trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu));
2189 kvm_arch_vcpu_block_finish(vcpu);
2191 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
2193 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
2195 struct swait_queue_head *wqp;
2197 wqp = kvm_arch_vcpu_wq(vcpu);
2198 if (swq_has_sleeper(wqp)) {
2200 ++vcpu->stat.halt_wakeup;
2206 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
2210 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2212 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
2215 int cpu = vcpu->cpu;
2217 if (kvm_vcpu_wake_up(vcpu))
2221 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
2222 if (kvm_arch_vcpu_should_kick(vcpu))
2223 smp_send_reschedule(cpu);
2226 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
2227 #endif /* !CONFIG_S390 */
2229 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
2232 struct task_struct *task = NULL;
2236 pid = rcu_dereference(target->pid);
2238 task = get_pid_task(pid, PIDTYPE_PID);
2242 ret = yield_to(task, 1);
2243 put_task_struct(task);
2247 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
2250 * Helper that checks whether a VCPU is eligible for directed yield.
2251 * Most eligible candidate to yield is decided by following heuristics:
2253 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2254 * (preempted lock holder), indicated by @in_spin_loop.
2255 * Set at the beiginning and cleared at the end of interception/PLE handler.
2257 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2258 * chance last time (mostly it has become eligible now since we have probably
2259 * yielded to lockholder in last iteration. This is done by toggling
2260 * @dy_eligible each time a VCPU checked for eligibility.)
2262 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2263 * to preempted lock-holder could result in wrong VCPU selection and CPU
2264 * burning. Giving priority for a potential lock-holder increases lock
2267 * Since algorithm is based on heuristics, accessing another VCPU data without
2268 * locking does not harm. It may result in trying to yield to same VCPU, fail
2269 * and continue with next VCPU and so on.
2271 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
2273 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2276 eligible = !vcpu->spin_loop.in_spin_loop ||
2277 vcpu->spin_loop.dy_eligible;
2279 if (vcpu->spin_loop.in_spin_loop)
2280 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
2288 void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode)
2290 struct kvm *kvm = me->kvm;
2291 struct kvm_vcpu *vcpu;
2292 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
2298 kvm_vcpu_set_in_spin_loop(me, true);
2300 * We boost the priority of a VCPU that is runnable but not
2301 * currently running, because it got preempted by something
2302 * else and called schedule in __vcpu_run. Hopefully that
2303 * VCPU is holding the lock that we need and will release it.
2304 * We approximate round-robin by starting at the last boosted VCPU.
2306 for (pass = 0; pass < 2 && !yielded && try; pass++) {
2307 kvm_for_each_vcpu(i, vcpu, kvm) {
2308 if (!pass && i <= last_boosted_vcpu) {
2309 i = last_boosted_vcpu;
2311 } else if (pass && i > last_boosted_vcpu)
2313 if (!ACCESS_ONCE(vcpu->preempted))
2317 if (swait_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
2319 if (yield_to_kernel_mode && !kvm_arch_vcpu_in_kernel(vcpu))
2321 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
2324 yielded = kvm_vcpu_yield_to(vcpu);
2326 kvm->last_boosted_vcpu = i;
2328 } else if (yielded < 0) {
2335 kvm_vcpu_set_in_spin_loop(me, false);
2337 /* Ensure vcpu is not eligible during next spinloop */
2338 kvm_vcpu_set_dy_eligible(me, false);
2340 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
2342 static int kvm_vcpu_fault(struct vm_fault *vmf)
2344 struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data;
2347 if (vmf->pgoff == 0)
2348 page = virt_to_page(vcpu->run);
2350 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
2351 page = virt_to_page(vcpu->arch.pio_data);
2353 #ifdef CONFIG_KVM_MMIO
2354 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
2355 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
2358 return kvm_arch_vcpu_fault(vcpu, vmf);
2364 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
2365 .fault = kvm_vcpu_fault,
2368 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2370 vma->vm_ops = &kvm_vcpu_vm_ops;
2374 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2376 struct kvm_vcpu *vcpu = filp->private_data;
2378 debugfs_remove_recursive(vcpu->debugfs_dentry);
2379 kvm_put_kvm(vcpu->kvm);
2383 static struct file_operations kvm_vcpu_fops = {
2384 .release = kvm_vcpu_release,
2385 .unlocked_ioctl = kvm_vcpu_ioctl,
2386 #ifdef CONFIG_KVM_COMPAT
2387 .compat_ioctl = kvm_vcpu_compat_ioctl,
2389 .mmap = kvm_vcpu_mmap,
2390 .llseek = noop_llseek,
2394 * Allocates an inode for the vcpu.
2396 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2398 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
2401 static int kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
2403 char dir_name[ITOA_MAX_LEN * 2];
2406 if (!kvm_arch_has_vcpu_debugfs())
2409 if (!debugfs_initialized())
2412 snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
2413 vcpu->debugfs_dentry = debugfs_create_dir(dir_name,
2414 vcpu->kvm->debugfs_dentry);
2415 if (!vcpu->debugfs_dentry)
2418 ret = kvm_arch_create_vcpu_debugfs(vcpu);
2420 debugfs_remove_recursive(vcpu->debugfs_dentry);
2428 * Creates some virtual cpus. Good luck creating more than one.
2430 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
2433 struct kvm_vcpu *vcpu;
2435 if (id >= KVM_MAX_VCPU_ID)
2438 mutex_lock(&kvm->lock);
2439 if (kvm->created_vcpus == KVM_MAX_VCPUS) {
2440 mutex_unlock(&kvm->lock);
2444 kvm->created_vcpus++;
2445 mutex_unlock(&kvm->lock);
2447 vcpu = kvm_arch_vcpu_create(kvm, id);
2450 goto vcpu_decrement;
2453 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2455 r = kvm_arch_vcpu_setup(vcpu);
2459 r = kvm_create_vcpu_debugfs(vcpu);
2463 mutex_lock(&kvm->lock);
2464 if (kvm_get_vcpu_by_id(kvm, id)) {
2466 goto unlock_vcpu_destroy;
2469 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
2471 /* Now it's all set up, let userspace reach it */
2473 r = create_vcpu_fd(vcpu);
2476 goto unlock_vcpu_destroy;
2479 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
2482 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
2483 * before kvm->online_vcpu's incremented value.
2486 atomic_inc(&kvm->online_vcpus);
2488 mutex_unlock(&kvm->lock);
2489 kvm_arch_vcpu_postcreate(vcpu);
2492 unlock_vcpu_destroy:
2493 mutex_unlock(&kvm->lock);
2494 debugfs_remove_recursive(vcpu->debugfs_dentry);
2496 kvm_arch_vcpu_destroy(vcpu);
2498 mutex_lock(&kvm->lock);
2499 kvm->created_vcpus--;
2500 mutex_unlock(&kvm->lock);
2504 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2507 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2508 vcpu->sigset_active = 1;
2509 vcpu->sigset = *sigset;
2511 vcpu->sigset_active = 0;
2515 static long kvm_vcpu_ioctl(struct file *filp,
2516 unsigned int ioctl, unsigned long arg)
2518 struct kvm_vcpu *vcpu = filp->private_data;
2519 void __user *argp = (void __user *)arg;
2521 struct kvm_fpu *fpu = NULL;
2522 struct kvm_sregs *kvm_sregs = NULL;
2524 if (vcpu->kvm->mm != current->mm)
2527 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2530 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
2532 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
2533 * so vcpu_load() would break it.
2535 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_S390_IRQ || ioctl == KVM_INTERRUPT)
2536 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2540 r = vcpu_load(vcpu);
2549 oldpid = rcu_access_pointer(vcpu->pid);
2550 if (unlikely(oldpid != current->pids[PIDTYPE_PID].pid)) {
2551 /* The thread running this VCPU changed. */
2552 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
2554 rcu_assign_pointer(vcpu->pid, newpid);
2559 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2560 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2563 case KVM_GET_REGS: {
2564 struct kvm_regs *kvm_regs;
2567 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2570 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2574 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2581 case KVM_SET_REGS: {
2582 struct kvm_regs *kvm_regs;
2585 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2586 if (IS_ERR(kvm_regs)) {
2587 r = PTR_ERR(kvm_regs);
2590 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2594 case KVM_GET_SREGS: {
2595 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2599 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2603 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2608 case KVM_SET_SREGS: {
2609 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2610 if (IS_ERR(kvm_sregs)) {
2611 r = PTR_ERR(kvm_sregs);
2615 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2618 case KVM_GET_MP_STATE: {
2619 struct kvm_mp_state mp_state;
2621 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2625 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
2630 case KVM_SET_MP_STATE: {
2631 struct kvm_mp_state mp_state;
2634 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
2636 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2639 case KVM_TRANSLATE: {
2640 struct kvm_translation tr;
2643 if (copy_from_user(&tr, argp, sizeof(tr)))
2645 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2649 if (copy_to_user(argp, &tr, sizeof(tr)))
2654 case KVM_SET_GUEST_DEBUG: {
2655 struct kvm_guest_debug dbg;
2658 if (copy_from_user(&dbg, argp, sizeof(dbg)))
2660 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2663 case KVM_SET_SIGNAL_MASK: {
2664 struct kvm_signal_mask __user *sigmask_arg = argp;
2665 struct kvm_signal_mask kvm_sigmask;
2666 sigset_t sigset, *p;
2671 if (copy_from_user(&kvm_sigmask, argp,
2672 sizeof(kvm_sigmask)))
2675 if (kvm_sigmask.len != sizeof(sigset))
2678 if (copy_from_user(&sigset, sigmask_arg->sigset,
2683 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2687 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2691 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2695 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2701 fpu = memdup_user(argp, sizeof(*fpu));
2707 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2711 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2720 #ifdef CONFIG_KVM_COMPAT
2721 static long kvm_vcpu_compat_ioctl(struct file *filp,
2722 unsigned int ioctl, unsigned long arg)
2724 struct kvm_vcpu *vcpu = filp->private_data;
2725 void __user *argp = compat_ptr(arg);
2728 if (vcpu->kvm->mm != current->mm)
2732 case KVM_SET_SIGNAL_MASK: {
2733 struct kvm_signal_mask __user *sigmask_arg = argp;
2734 struct kvm_signal_mask kvm_sigmask;
2735 compat_sigset_t csigset;
2740 if (copy_from_user(&kvm_sigmask, argp,
2741 sizeof(kvm_sigmask)))
2744 if (kvm_sigmask.len != sizeof(csigset))
2747 if (copy_from_user(&csigset, sigmask_arg->sigset,
2750 sigset_from_compat(&sigset, &csigset);
2751 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2753 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2757 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2765 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2766 int (*accessor)(struct kvm_device *dev,
2767 struct kvm_device_attr *attr),
2770 struct kvm_device_attr attr;
2775 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2778 return accessor(dev, &attr);
2781 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2784 struct kvm_device *dev = filp->private_data;
2787 case KVM_SET_DEVICE_ATTR:
2788 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2789 case KVM_GET_DEVICE_ATTR:
2790 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2791 case KVM_HAS_DEVICE_ATTR:
2792 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2794 if (dev->ops->ioctl)
2795 return dev->ops->ioctl(dev, ioctl, arg);
2801 static int kvm_device_release(struct inode *inode, struct file *filp)
2803 struct kvm_device *dev = filp->private_data;
2804 struct kvm *kvm = dev->kvm;
2810 static const struct file_operations kvm_device_fops = {
2811 .unlocked_ioctl = kvm_device_ioctl,
2812 #ifdef CONFIG_KVM_COMPAT
2813 .compat_ioctl = kvm_device_ioctl,
2815 .release = kvm_device_release,
2818 struct kvm_device *kvm_device_from_filp(struct file *filp)
2820 if (filp->f_op != &kvm_device_fops)
2823 return filp->private_data;
2826 static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
2827 #ifdef CONFIG_KVM_MPIC
2828 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
2829 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
2833 int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
2835 if (type >= ARRAY_SIZE(kvm_device_ops_table))
2838 if (kvm_device_ops_table[type] != NULL)
2841 kvm_device_ops_table[type] = ops;
2845 void kvm_unregister_device_ops(u32 type)
2847 if (kvm_device_ops_table[type] != NULL)
2848 kvm_device_ops_table[type] = NULL;
2851 static int kvm_ioctl_create_device(struct kvm *kvm,
2852 struct kvm_create_device *cd)
2854 struct kvm_device_ops *ops = NULL;
2855 struct kvm_device *dev;
2856 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2859 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
2862 ops = kvm_device_ops_table[cd->type];
2869 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2876 mutex_lock(&kvm->lock);
2877 ret = ops->create(dev, cd->type);
2879 mutex_unlock(&kvm->lock);
2883 list_add(&dev->vm_node, &kvm->devices);
2884 mutex_unlock(&kvm->lock);
2889 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2891 mutex_lock(&kvm->lock);
2892 list_del(&dev->vm_node);
2893 mutex_unlock(&kvm->lock);
2903 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2906 case KVM_CAP_USER_MEMORY:
2907 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2908 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2909 case KVM_CAP_INTERNAL_ERROR_DATA:
2910 #ifdef CONFIG_HAVE_KVM_MSI
2911 case KVM_CAP_SIGNAL_MSI:
2913 #ifdef CONFIG_HAVE_KVM_IRQFD
2915 case KVM_CAP_IRQFD_RESAMPLE:
2917 case KVM_CAP_IOEVENTFD_ANY_LENGTH:
2918 case KVM_CAP_CHECK_EXTENSION_VM:
2920 #ifdef CONFIG_KVM_MMIO
2921 case KVM_CAP_COALESCED_MMIO:
2922 return KVM_COALESCED_MMIO_PAGE_OFFSET;
2924 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2925 case KVM_CAP_IRQ_ROUTING:
2926 return KVM_MAX_IRQ_ROUTES;
2928 #if KVM_ADDRESS_SPACE_NUM > 1
2929 case KVM_CAP_MULTI_ADDRESS_SPACE:
2930 return KVM_ADDRESS_SPACE_NUM;
2932 case KVM_CAP_MAX_VCPU_ID:
2933 return KVM_MAX_VCPU_ID;
2937 return kvm_vm_ioctl_check_extension(kvm, arg);
2940 static long kvm_vm_ioctl(struct file *filp,
2941 unsigned int ioctl, unsigned long arg)
2943 struct kvm *kvm = filp->private_data;
2944 void __user *argp = (void __user *)arg;
2947 if (kvm->mm != current->mm)
2950 case KVM_CREATE_VCPU:
2951 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2953 case KVM_SET_USER_MEMORY_REGION: {
2954 struct kvm_userspace_memory_region kvm_userspace_mem;
2957 if (copy_from_user(&kvm_userspace_mem, argp,
2958 sizeof(kvm_userspace_mem)))
2961 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2964 case KVM_GET_DIRTY_LOG: {
2965 struct kvm_dirty_log log;
2968 if (copy_from_user(&log, argp, sizeof(log)))
2970 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2973 #ifdef CONFIG_KVM_MMIO
2974 case KVM_REGISTER_COALESCED_MMIO: {
2975 struct kvm_coalesced_mmio_zone zone;
2978 if (copy_from_user(&zone, argp, sizeof(zone)))
2980 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2983 case KVM_UNREGISTER_COALESCED_MMIO: {
2984 struct kvm_coalesced_mmio_zone zone;
2987 if (copy_from_user(&zone, argp, sizeof(zone)))
2989 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2994 struct kvm_irqfd data;
2997 if (copy_from_user(&data, argp, sizeof(data)))
2999 r = kvm_irqfd(kvm, &data);
3002 case KVM_IOEVENTFD: {
3003 struct kvm_ioeventfd data;
3006 if (copy_from_user(&data, argp, sizeof(data)))
3008 r = kvm_ioeventfd(kvm, &data);
3011 #ifdef CONFIG_HAVE_KVM_MSI
3012 case KVM_SIGNAL_MSI: {
3016 if (copy_from_user(&msi, argp, sizeof(msi)))
3018 r = kvm_send_userspace_msi(kvm, &msi);
3022 #ifdef __KVM_HAVE_IRQ_LINE
3023 case KVM_IRQ_LINE_STATUS:
3024 case KVM_IRQ_LINE: {
3025 struct kvm_irq_level irq_event;
3028 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
3031 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
3032 ioctl == KVM_IRQ_LINE_STATUS);
3037 if (ioctl == KVM_IRQ_LINE_STATUS) {
3038 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
3046 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3047 case KVM_SET_GSI_ROUTING: {
3048 struct kvm_irq_routing routing;
3049 struct kvm_irq_routing __user *urouting;
3050 struct kvm_irq_routing_entry *entries = NULL;
3053 if (copy_from_user(&routing, argp, sizeof(routing)))
3056 if (!kvm_arch_can_set_irq_routing(kvm))
3058 if (routing.nr > KVM_MAX_IRQ_ROUTES)
3064 entries = vmalloc(routing.nr * sizeof(*entries));
3069 if (copy_from_user(entries, urouting->entries,
3070 routing.nr * sizeof(*entries)))
3071 goto out_free_irq_routing;
3073 r = kvm_set_irq_routing(kvm, entries, routing.nr,
3075 out_free_irq_routing:
3079 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
3080 case KVM_CREATE_DEVICE: {
3081 struct kvm_create_device cd;
3084 if (copy_from_user(&cd, argp, sizeof(cd)))
3087 r = kvm_ioctl_create_device(kvm, &cd);
3092 if (copy_to_user(argp, &cd, sizeof(cd)))
3098 case KVM_CHECK_EXTENSION:
3099 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
3102 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
3108 #ifdef CONFIG_KVM_COMPAT
3109 struct compat_kvm_dirty_log {
3113 compat_uptr_t dirty_bitmap; /* one bit per page */
3118 static long kvm_vm_compat_ioctl(struct file *filp,
3119 unsigned int ioctl, unsigned long arg)
3121 struct kvm *kvm = filp->private_data;
3124 if (kvm->mm != current->mm)
3127 case KVM_GET_DIRTY_LOG: {
3128 struct compat_kvm_dirty_log compat_log;
3129 struct kvm_dirty_log log;
3131 if (copy_from_user(&compat_log, (void __user *)arg,
3132 sizeof(compat_log)))
3134 log.slot = compat_log.slot;
3135 log.padding1 = compat_log.padding1;
3136 log.padding2 = compat_log.padding2;
3137 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
3139 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3143 r = kvm_vm_ioctl(filp, ioctl, arg);
3149 static struct file_operations kvm_vm_fops = {
3150 .release = kvm_vm_release,
3151 .unlocked_ioctl = kvm_vm_ioctl,
3152 #ifdef CONFIG_KVM_COMPAT
3153 .compat_ioctl = kvm_vm_compat_ioctl,
3155 .llseek = noop_llseek,
3158 static int kvm_dev_ioctl_create_vm(unsigned long type)
3164 kvm = kvm_create_vm(type);
3166 return PTR_ERR(kvm);
3167 #ifdef CONFIG_KVM_MMIO
3168 r = kvm_coalesced_mmio_init(kvm);
3174 r = get_unused_fd_flags(O_CLOEXEC);
3179 file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
3183 return PTR_ERR(file);
3187 * Don't call kvm_put_kvm anymore at this point; file->f_op is
3188 * already set, with ->release() being kvm_vm_release(). In error
3189 * cases it will be called by the final fput(file) and will take
3190 * care of doing kvm_put_kvm(kvm).
3192 if (kvm_create_vm_debugfs(kvm, r) < 0) {
3197 kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm);
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 CONFIG_KVM_MMIO
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);
3477 r = __kvm_io_bus_write(vcpu, bus, &range, val);
3478 return r < 0 ? r : 0;
3481 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3482 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
3483 gpa_t addr, int len, const void *val, long cookie)
3485 struct kvm_io_bus *bus;
3486 struct kvm_io_range range;
3488 range = (struct kvm_io_range) {
3493 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3497 /* First try the device referenced by cookie. */
3498 if ((cookie >= 0) && (cookie < bus->dev_count) &&
3499 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
3500 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
3505 * cookie contained garbage; fall back to search and return the
3506 * correct cookie value.
3508 return __kvm_io_bus_write(vcpu, bus, &range, val);
3511 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3512 struct kvm_io_range *range, void *val)
3516 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3520 while (idx < bus->dev_count &&
3521 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3522 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
3530 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
3532 /* kvm_io_bus_read - called under kvm->slots_lock */
3533 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3536 struct kvm_io_bus *bus;
3537 struct kvm_io_range range;
3540 range = (struct kvm_io_range) {
3545 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3548 r = __kvm_io_bus_read(vcpu, bus, &range, val);
3549 return r < 0 ? r : 0;
3553 /* Caller must hold slots_lock. */
3554 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3555 int len, struct kvm_io_device *dev)
3557 struct kvm_io_bus *new_bus, *bus;
3559 bus = kvm_get_bus(kvm, bus_idx);
3563 /* exclude ioeventfd which is limited by maximum fd */
3564 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3567 new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count + 1) *
3568 sizeof(struct kvm_io_range)), GFP_KERNEL);
3571 memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
3572 sizeof(struct kvm_io_range)));
3573 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
3574 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3575 synchronize_srcu_expedited(&kvm->srcu);
3581 /* Caller must hold slots_lock. */
3582 void kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3583 struct kvm_io_device *dev)
3586 struct kvm_io_bus *new_bus, *bus;
3588 bus = kvm_get_bus(kvm, bus_idx);
3592 for (i = 0; i < bus->dev_count; i++)
3593 if (bus->range[i].dev == dev) {
3597 if (i == bus->dev_count)
3600 new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3601 sizeof(struct kvm_io_range)), GFP_KERNEL);
3603 pr_err("kvm: failed to shrink bus, removing it completely\n");
3607 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3608 new_bus->dev_count--;
3609 memcpy(new_bus->range + i, bus->range + i + 1,
3610 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3613 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3614 synchronize_srcu_expedited(&kvm->srcu);
3619 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3622 struct kvm_io_bus *bus;
3623 int dev_idx, srcu_idx;
3624 struct kvm_io_device *iodev = NULL;
3626 srcu_idx = srcu_read_lock(&kvm->srcu);
3628 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
3632 dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
3636 iodev = bus->range[dev_idx].dev;
3639 srcu_read_unlock(&kvm->srcu, srcu_idx);
3643 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
3645 static int kvm_debugfs_open(struct inode *inode, struct file *file,
3646 int (*get)(void *, u64 *), int (*set)(void *, u64),
3649 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
3652 /* The debugfs files are a reference to the kvm struct which
3653 * is still valid when kvm_destroy_vm is called.
3654 * To avoid the race between open and the removal of the debugfs
3655 * directory we test against the users count.
3657 if (!refcount_inc_not_zero(&stat_data->kvm->users_count))
3660 if (simple_attr_open(inode, file, get, set, fmt)) {
3661 kvm_put_kvm(stat_data->kvm);
3668 static int kvm_debugfs_release(struct inode *inode, struct file *file)
3670 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
3673 simple_attr_release(inode, file);
3674 kvm_put_kvm(stat_data->kvm);
3679 static int vm_stat_get_per_vm(void *data, u64 *val)
3681 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3683 *val = *(ulong *)((void *)stat_data->kvm + stat_data->offset);
3688 static int vm_stat_clear_per_vm(void *data, u64 val)
3690 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3695 *(ulong *)((void *)stat_data->kvm + stat_data->offset) = 0;
3700 static int vm_stat_get_per_vm_open(struct inode *inode, struct file *file)
3702 __simple_attr_check_format("%llu\n", 0ull);
3703 return kvm_debugfs_open(inode, file, vm_stat_get_per_vm,
3704 vm_stat_clear_per_vm, "%llu\n");
3707 static const struct file_operations vm_stat_get_per_vm_fops = {
3708 .owner = THIS_MODULE,
3709 .open = vm_stat_get_per_vm_open,
3710 .release = kvm_debugfs_release,
3711 .read = simple_attr_read,
3712 .write = simple_attr_write,
3713 .llseek = no_llseek,
3716 static int vcpu_stat_get_per_vm(void *data, u64 *val)
3719 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3720 struct kvm_vcpu *vcpu;
3724 kvm_for_each_vcpu(i, vcpu, stat_data->kvm)
3725 *val += *(u64 *)((void *)vcpu + stat_data->offset);
3730 static int vcpu_stat_clear_per_vm(void *data, u64 val)
3733 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3734 struct kvm_vcpu *vcpu;
3739 kvm_for_each_vcpu(i, vcpu, stat_data->kvm)
3740 *(u64 *)((void *)vcpu + stat_data->offset) = 0;
3745 static int vcpu_stat_get_per_vm_open(struct inode *inode, struct file *file)
3747 __simple_attr_check_format("%llu\n", 0ull);
3748 return kvm_debugfs_open(inode, file, vcpu_stat_get_per_vm,
3749 vcpu_stat_clear_per_vm, "%llu\n");
3752 static const struct file_operations vcpu_stat_get_per_vm_fops = {
3753 .owner = THIS_MODULE,
3754 .open = vcpu_stat_get_per_vm_open,
3755 .release = kvm_debugfs_release,
3756 .read = simple_attr_read,
3757 .write = simple_attr_write,
3758 .llseek = no_llseek,
3761 static const struct file_operations *stat_fops_per_vm[] = {
3762 [KVM_STAT_VCPU] = &vcpu_stat_get_per_vm_fops,
3763 [KVM_STAT_VM] = &vm_stat_get_per_vm_fops,
3766 static int vm_stat_get(void *_offset, u64 *val)
3768 unsigned offset = (long)_offset;
3770 struct kvm_stat_data stat_tmp = {.offset = offset};
3774 spin_lock(&kvm_lock);
3775 list_for_each_entry(kvm, &vm_list, vm_list) {
3777 vm_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
3780 spin_unlock(&kvm_lock);
3784 static int vm_stat_clear(void *_offset, u64 val)
3786 unsigned offset = (long)_offset;
3788 struct kvm_stat_data stat_tmp = {.offset = offset};
3793 spin_lock(&kvm_lock);
3794 list_for_each_entry(kvm, &vm_list, vm_list) {
3796 vm_stat_clear_per_vm((void *)&stat_tmp, 0);
3798 spin_unlock(&kvm_lock);
3803 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n");
3805 static int vcpu_stat_get(void *_offset, u64 *val)
3807 unsigned offset = (long)_offset;
3809 struct kvm_stat_data stat_tmp = {.offset = offset};
3813 spin_lock(&kvm_lock);
3814 list_for_each_entry(kvm, &vm_list, vm_list) {
3816 vcpu_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
3819 spin_unlock(&kvm_lock);
3823 static int vcpu_stat_clear(void *_offset, u64 val)
3825 unsigned offset = (long)_offset;
3827 struct kvm_stat_data stat_tmp = {.offset = offset};
3832 spin_lock(&kvm_lock);
3833 list_for_each_entry(kvm, &vm_list, vm_list) {
3835 vcpu_stat_clear_per_vm((void *)&stat_tmp, 0);
3837 spin_unlock(&kvm_lock);
3842 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear,
3845 static const struct file_operations *stat_fops[] = {
3846 [KVM_STAT_VCPU] = &vcpu_stat_fops,
3847 [KVM_STAT_VM] = &vm_stat_fops,
3850 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm)
3852 struct kobj_uevent_env *env;
3853 unsigned long long created, active;
3855 if (!kvm_dev.this_device || !kvm)
3858 spin_lock(&kvm_lock);
3859 if (type == KVM_EVENT_CREATE_VM) {
3860 kvm_createvm_count++;
3862 } else if (type == KVM_EVENT_DESTROY_VM) {
3865 created = kvm_createvm_count;
3866 active = kvm_active_vms;
3867 spin_unlock(&kvm_lock);
3869 env = kzalloc(sizeof(*env), GFP_KERNEL);
3873 add_uevent_var(env, "CREATED=%llu", created);
3874 add_uevent_var(env, "COUNT=%llu", active);
3876 if (type == KVM_EVENT_CREATE_VM) {
3877 add_uevent_var(env, "EVENT=create");
3878 kvm->userspace_pid = task_pid_nr(current);
3879 } else if (type == KVM_EVENT_DESTROY_VM) {
3880 add_uevent_var(env, "EVENT=destroy");
3882 add_uevent_var(env, "PID=%d", kvm->userspace_pid);
3884 if (kvm->debugfs_dentry) {
3885 char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL);
3888 tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX);
3890 add_uevent_var(env, "STATS_PATH=%s", tmp);
3894 /* no need for checks, since we are adding at most only 5 keys */
3895 env->envp[env->envp_idx++] = NULL;
3896 kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp);
3900 static int kvm_init_debug(void)
3903 struct kvm_stats_debugfs_item *p;
3905 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3906 if (kvm_debugfs_dir == NULL)
3909 kvm_debugfs_num_entries = 0;
3910 for (p = debugfs_entries; p->name; ++p, kvm_debugfs_num_entries++) {
3911 if (!debugfs_create_file(p->name, 0644, kvm_debugfs_dir,
3912 (void *)(long)p->offset,
3913 stat_fops[p->kind]))
3920 debugfs_remove_recursive(kvm_debugfs_dir);
3925 static int kvm_suspend(void)
3927 if (kvm_usage_count)
3928 hardware_disable_nolock(NULL);
3932 static void kvm_resume(void)
3934 if (kvm_usage_count) {
3935 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3936 hardware_enable_nolock(NULL);
3940 static struct syscore_ops kvm_syscore_ops = {
3941 .suspend = kvm_suspend,
3942 .resume = kvm_resume,
3946 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3948 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3951 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3953 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3955 if (vcpu->preempted)
3956 vcpu->preempted = false;
3958 kvm_arch_sched_in(vcpu, cpu);
3960 kvm_arch_vcpu_load(vcpu, cpu);
3963 static void kvm_sched_out(struct preempt_notifier *pn,
3964 struct task_struct *next)
3966 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3968 if (current->state == TASK_RUNNING)
3969 vcpu->preempted = true;
3970 kvm_arch_vcpu_put(vcpu);
3973 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3974 struct module *module)
3979 r = kvm_arch_init(opaque);
3984 * kvm_arch_init makes sure there's at most one caller
3985 * for architectures that support multiple implementations,
3986 * like intel and amd on x86.
3987 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3988 * conflicts in case kvm is already setup for another implementation.
3990 r = kvm_irqfd_init();
3994 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
3999 r = kvm_arch_hardware_setup();
4003 for_each_online_cpu(cpu) {
4004 smp_call_function_single(cpu,
4005 kvm_arch_check_processor_compat,
4011 r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting",
4012 kvm_starting_cpu, kvm_dying_cpu);
4015 register_reboot_notifier(&kvm_reboot_notifier);
4017 /* A kmem cache lets us meet the alignment requirements of fx_save. */
4019 vcpu_align = __alignof__(struct kvm_vcpu);
4020 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
4021 SLAB_ACCOUNT, NULL);
4022 if (!kvm_vcpu_cache) {
4027 r = kvm_async_pf_init();
4031 kvm_chardev_ops.owner = module;
4032 kvm_vm_fops.owner = module;
4033 kvm_vcpu_fops.owner = module;
4035 r = misc_register(&kvm_dev);
4037 pr_err("kvm: misc device register failed\n");
4041 register_syscore_ops(&kvm_syscore_ops);
4043 kvm_preempt_ops.sched_in = kvm_sched_in;
4044 kvm_preempt_ops.sched_out = kvm_sched_out;
4046 r = kvm_init_debug();
4048 pr_err("kvm: create debugfs files failed\n");
4052 r = kvm_vfio_ops_init();
4058 unregister_syscore_ops(&kvm_syscore_ops);
4059 misc_deregister(&kvm_dev);
4061 kvm_async_pf_deinit();
4063 kmem_cache_destroy(kvm_vcpu_cache);
4065 unregister_reboot_notifier(&kvm_reboot_notifier);
4066 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4069 kvm_arch_hardware_unsetup();
4071 free_cpumask_var(cpus_hardware_enabled);
4079 EXPORT_SYMBOL_GPL(kvm_init);
4083 debugfs_remove_recursive(kvm_debugfs_dir);
4084 misc_deregister(&kvm_dev);
4085 kmem_cache_destroy(kvm_vcpu_cache);
4086 kvm_async_pf_deinit();
4087 unregister_syscore_ops(&kvm_syscore_ops);
4088 unregister_reboot_notifier(&kvm_reboot_notifier);
4089 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4090 on_each_cpu(hardware_disable_nolock, NULL, 1);
4091 kvm_arch_hardware_unsetup();
4094 free_cpumask_var(cpus_hardware_enabled);
4095 kvm_vfio_ops_exit();
4097 EXPORT_SYMBOL_GPL(kvm_exit);