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 mark_page_dirty_in_slot(struct kvm_memory_slot *memslot, gfn_t gfn);
127 __visible bool kvm_rebooting;
128 EXPORT_SYMBOL_GPL(kvm_rebooting);
130 static bool largepages_enabled = true;
132 #define KVM_EVENT_CREATE_VM 0
133 #define KVM_EVENT_DESTROY_VM 1
134 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm);
135 static unsigned long long kvm_createvm_count;
136 static unsigned long long kvm_active_vms;
138 bool kvm_is_reserved_pfn(kvm_pfn_t pfn)
141 return PageReserved(pfn_to_page(pfn));
147 * Switches to specified vcpu, until a matching vcpu_put()
149 int vcpu_load(struct kvm_vcpu *vcpu)
153 if (mutex_lock_killable(&vcpu->mutex))
156 preempt_notifier_register(&vcpu->preempt_notifier);
157 kvm_arch_vcpu_load(vcpu, cpu);
161 EXPORT_SYMBOL_GPL(vcpu_load);
163 void vcpu_put(struct kvm_vcpu *vcpu)
166 kvm_arch_vcpu_put(vcpu);
167 preempt_notifier_unregister(&vcpu->preempt_notifier);
169 mutex_unlock(&vcpu->mutex);
171 EXPORT_SYMBOL_GPL(vcpu_put);
173 /* TODO: merge with kvm_arch_vcpu_should_kick */
174 static bool kvm_request_needs_ipi(struct kvm_vcpu *vcpu, unsigned req)
176 int mode = kvm_vcpu_exiting_guest_mode(vcpu);
179 * We need to wait for the VCPU to reenable interrupts and get out of
180 * READING_SHADOW_PAGE_TABLES mode.
182 if (req & KVM_REQUEST_WAIT)
183 return mode != OUTSIDE_GUEST_MODE;
186 * Need to kick a running VCPU, but otherwise there is nothing to do.
188 return mode == IN_GUEST_MODE;
191 static void ack_flush(void *_completed)
195 static inline bool kvm_kick_many_cpus(const struct cpumask *cpus, bool wait)
198 cpus = cpu_online_mask;
200 if (cpumask_empty(cpus))
203 smp_call_function_many(cpus, ack_flush, NULL, wait);
207 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
212 struct kvm_vcpu *vcpu;
214 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
217 kvm_for_each_vcpu(i, vcpu, kvm) {
218 kvm_make_request(req, vcpu);
221 if (!(req & KVM_REQUEST_NO_WAKEUP) && kvm_vcpu_wake_up(vcpu))
224 if (cpus != NULL && cpu != -1 && cpu != me &&
225 kvm_request_needs_ipi(vcpu, req))
226 __cpumask_set_cpu(cpu, cpus);
228 called = kvm_kick_many_cpus(cpus, !!(req & KVM_REQUEST_WAIT));
230 free_cpumask_var(cpus);
234 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
235 void kvm_flush_remote_tlbs(struct kvm *kvm)
238 * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in
239 * kvm_make_all_cpus_request.
241 long dirty_count = smp_load_acquire(&kvm->tlbs_dirty);
244 * We want to publish modifications to the page tables before reading
245 * mode. Pairs with a memory barrier in arch-specific code.
246 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
247 * and smp_mb in walk_shadow_page_lockless_begin/end.
248 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
250 * There is already an smp_mb__after_atomic() before
251 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
254 if (kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
255 ++kvm->stat.remote_tlb_flush;
256 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
258 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
261 void kvm_reload_remote_mmus(struct kvm *kvm)
263 kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
266 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
271 mutex_init(&vcpu->mutex);
276 init_swait_queue_head(&vcpu->wq);
277 kvm_async_pf_vcpu_init(vcpu);
280 INIT_LIST_HEAD(&vcpu->blocked_vcpu_list);
282 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
287 vcpu->run = page_address(page);
289 kvm_vcpu_set_in_spin_loop(vcpu, false);
290 kvm_vcpu_set_dy_eligible(vcpu, false);
291 vcpu->preempted = false;
293 r = kvm_arch_vcpu_init(vcpu);
299 free_page((unsigned long)vcpu->run);
303 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
305 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
308 * no need for rcu_read_lock as VCPU_RUN is the only place that
309 * will change the vcpu->pid pointer and on uninit all file
310 * descriptors are already gone.
312 put_pid(rcu_dereference_protected(vcpu->pid, 1));
313 kvm_arch_vcpu_uninit(vcpu);
314 free_page((unsigned long)vcpu->run);
316 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
318 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
319 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
321 return container_of(mn, struct kvm, mmu_notifier);
324 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
325 struct mm_struct *mm,
326 unsigned long address,
329 struct kvm *kvm = mmu_notifier_to_kvm(mn);
332 idx = srcu_read_lock(&kvm->srcu);
333 spin_lock(&kvm->mmu_lock);
334 kvm->mmu_notifier_seq++;
335 kvm_set_spte_hva(kvm, address, pte);
336 spin_unlock(&kvm->mmu_lock);
337 srcu_read_unlock(&kvm->srcu, idx);
340 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
341 struct mm_struct *mm,
345 struct kvm *kvm = mmu_notifier_to_kvm(mn);
346 int need_tlb_flush = 0, idx;
348 idx = srcu_read_lock(&kvm->srcu);
349 spin_lock(&kvm->mmu_lock);
351 * The count increase must become visible at unlock time as no
352 * spte can be established without taking the mmu_lock and
353 * count is also read inside the mmu_lock critical section.
355 kvm->mmu_notifier_count++;
356 need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
357 need_tlb_flush |= kvm->tlbs_dirty;
358 /* we've to flush the tlb before the pages can be freed */
360 kvm_flush_remote_tlbs(kvm);
362 spin_unlock(&kvm->mmu_lock);
363 srcu_read_unlock(&kvm->srcu, idx);
366 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
367 struct mm_struct *mm,
371 struct kvm *kvm = mmu_notifier_to_kvm(mn);
373 spin_lock(&kvm->mmu_lock);
375 * This sequence increase will notify the kvm page fault that
376 * the page that is going to be mapped in the spte could have
379 kvm->mmu_notifier_seq++;
382 * The above sequence increase must be visible before the
383 * below count decrease, which is ensured by the smp_wmb above
384 * in conjunction with the smp_rmb in mmu_notifier_retry().
386 kvm->mmu_notifier_count--;
387 spin_unlock(&kvm->mmu_lock);
389 BUG_ON(kvm->mmu_notifier_count < 0);
392 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
393 struct mm_struct *mm,
397 struct kvm *kvm = mmu_notifier_to_kvm(mn);
400 idx = srcu_read_lock(&kvm->srcu);
401 spin_lock(&kvm->mmu_lock);
403 young = kvm_age_hva(kvm, start, end);
405 kvm_flush_remote_tlbs(kvm);
407 spin_unlock(&kvm->mmu_lock);
408 srcu_read_unlock(&kvm->srcu, idx);
413 static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
414 struct mm_struct *mm,
418 struct kvm *kvm = mmu_notifier_to_kvm(mn);
421 idx = srcu_read_lock(&kvm->srcu);
422 spin_lock(&kvm->mmu_lock);
424 * Even though we do not flush TLB, this will still adversely
425 * affect performance on pre-Haswell Intel EPT, where there is
426 * no EPT Access Bit to clear so that we have to tear down EPT
427 * tables instead. If we find this unacceptable, we can always
428 * add a parameter to kvm_age_hva so that it effectively doesn't
429 * do anything on clear_young.
431 * Also note that currently we never issue secondary TLB flushes
432 * from clear_young, leaving this job up to the regular system
433 * cadence. If we find this inaccurate, we might come up with a
434 * more sophisticated heuristic later.
436 young = kvm_age_hva(kvm, start, end);
437 spin_unlock(&kvm->mmu_lock);
438 srcu_read_unlock(&kvm->srcu, idx);
443 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
444 struct mm_struct *mm,
445 unsigned long address)
447 struct kvm *kvm = mmu_notifier_to_kvm(mn);
450 idx = srcu_read_lock(&kvm->srcu);
451 spin_lock(&kvm->mmu_lock);
452 young = kvm_test_age_hva(kvm, address);
453 spin_unlock(&kvm->mmu_lock);
454 srcu_read_unlock(&kvm->srcu, idx);
459 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
460 struct mm_struct *mm)
462 struct kvm *kvm = mmu_notifier_to_kvm(mn);
465 idx = srcu_read_lock(&kvm->srcu);
466 kvm_arch_flush_shadow_all(kvm);
467 srcu_read_unlock(&kvm->srcu, idx);
470 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
471 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
472 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
473 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
474 .clear_young = kvm_mmu_notifier_clear_young,
475 .test_young = kvm_mmu_notifier_test_young,
476 .change_pte = kvm_mmu_notifier_change_pte,
477 .release = kvm_mmu_notifier_release,
480 static int kvm_init_mmu_notifier(struct kvm *kvm)
482 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
483 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
486 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
488 static int kvm_init_mmu_notifier(struct kvm *kvm)
493 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
495 static struct kvm_memslots *kvm_alloc_memslots(void)
498 struct kvm_memslots *slots;
500 slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
504 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
505 slots->id_to_index[i] = slots->memslots[i].id = i;
510 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
512 if (!memslot->dirty_bitmap)
515 kvfree(memslot->dirty_bitmap);
516 memslot->dirty_bitmap = NULL;
520 * Free any memory in @free but not in @dont.
522 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
523 struct kvm_memory_slot *dont)
525 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
526 kvm_destroy_dirty_bitmap(free);
528 kvm_arch_free_memslot(kvm, free, dont);
533 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
535 struct kvm_memory_slot *memslot;
540 kvm_for_each_memslot(memslot, slots)
541 kvm_free_memslot(kvm, memslot, NULL);
546 static void kvm_destroy_vm_debugfs(struct kvm *kvm)
550 if (!kvm->debugfs_dentry)
553 debugfs_remove_recursive(kvm->debugfs_dentry);
555 if (kvm->debugfs_stat_data) {
556 for (i = 0; i < kvm_debugfs_num_entries; i++)
557 kfree(kvm->debugfs_stat_data[i]);
558 kfree(kvm->debugfs_stat_data);
562 static int kvm_create_vm_debugfs(struct kvm *kvm, int fd)
564 char dir_name[ITOA_MAX_LEN * 2];
565 struct kvm_stat_data *stat_data;
566 struct kvm_stats_debugfs_item *p;
568 if (!debugfs_initialized())
571 snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd);
572 kvm->debugfs_dentry = debugfs_create_dir(dir_name,
574 if (!kvm->debugfs_dentry)
577 kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries,
578 sizeof(*kvm->debugfs_stat_data),
580 if (!kvm->debugfs_stat_data)
583 for (p = debugfs_entries; p->name; p++) {
584 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL);
588 stat_data->kvm = kvm;
589 stat_data->offset = p->offset;
590 kvm->debugfs_stat_data[p - debugfs_entries] = stat_data;
591 if (!debugfs_create_file(p->name, 0644,
594 stat_fops_per_vm[p->kind]))
600 static struct kvm *kvm_create_vm(unsigned long type)
603 struct kvm *kvm = kvm_arch_alloc_vm();
606 return ERR_PTR(-ENOMEM);
608 spin_lock_init(&kvm->mmu_lock);
610 kvm->mm = current->mm;
611 kvm_eventfd_init(kvm);
612 mutex_init(&kvm->lock);
613 mutex_init(&kvm->irq_lock);
614 mutex_init(&kvm->slots_lock);
615 refcount_set(&kvm->users_count, 1);
616 INIT_LIST_HEAD(&kvm->devices);
618 r = kvm_arch_init_vm(kvm, type);
620 goto out_err_no_disable;
622 r = hardware_enable_all();
624 goto out_err_no_disable;
626 #ifdef CONFIG_HAVE_KVM_IRQFD
627 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
630 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
633 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
634 struct kvm_memslots *slots = kvm_alloc_memslots();
636 goto out_err_no_srcu;
638 * Generations must be different for each address space.
639 * Init kvm generation close to the maximum to easily test the
640 * code of handling generation number wrap-around.
642 slots->generation = i * 2 - 150;
643 rcu_assign_pointer(kvm->memslots[i], slots);
646 if (init_srcu_struct(&kvm->srcu))
647 goto out_err_no_srcu;
648 if (init_srcu_struct(&kvm->irq_srcu))
649 goto out_err_no_irq_srcu;
650 for (i = 0; i < KVM_NR_BUSES; i++) {
651 rcu_assign_pointer(kvm->buses[i],
652 kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL));
657 r = kvm_init_mmu_notifier(kvm);
661 spin_lock(&kvm_lock);
662 list_add(&kvm->vm_list, &vm_list);
663 spin_unlock(&kvm_lock);
665 preempt_notifier_inc();
670 cleanup_srcu_struct(&kvm->irq_srcu);
672 cleanup_srcu_struct(&kvm->srcu);
674 hardware_disable_all();
676 refcount_set(&kvm->users_count, 0);
677 for (i = 0; i < KVM_NR_BUSES; i++)
678 kfree(kvm_get_bus(kvm, i));
679 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
680 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
681 kvm_arch_free_vm(kvm);
686 static void kvm_destroy_devices(struct kvm *kvm)
688 struct kvm_device *dev, *tmp;
691 * We do not need to take the kvm->lock here, because nobody else
692 * has a reference to the struct kvm at this point and therefore
693 * cannot access the devices list anyhow.
695 list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
696 list_del(&dev->vm_node);
697 dev->ops->destroy(dev);
701 static void kvm_destroy_vm(struct kvm *kvm)
704 struct mm_struct *mm = kvm->mm;
706 kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm);
707 kvm_destroy_vm_debugfs(kvm);
708 kvm_arch_sync_events(kvm);
709 spin_lock(&kvm_lock);
710 list_del(&kvm->vm_list);
711 spin_unlock(&kvm_lock);
712 kvm_free_irq_routing(kvm);
713 for (i = 0; i < KVM_NR_BUSES; i++) {
714 struct kvm_io_bus *bus = kvm_get_bus(kvm, i);
717 kvm_io_bus_destroy(bus);
718 kvm->buses[i] = NULL;
720 kvm_coalesced_mmio_free(kvm);
721 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
722 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
724 kvm_arch_flush_shadow_all(kvm);
726 kvm_arch_destroy_vm(kvm);
727 kvm_destroy_devices(kvm);
728 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
729 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
730 cleanup_srcu_struct(&kvm->irq_srcu);
731 cleanup_srcu_struct(&kvm->srcu);
732 kvm_arch_free_vm(kvm);
733 preempt_notifier_dec();
734 hardware_disable_all();
738 void kvm_get_kvm(struct kvm *kvm)
740 refcount_inc(&kvm->users_count);
742 EXPORT_SYMBOL_GPL(kvm_get_kvm);
744 void kvm_put_kvm(struct kvm *kvm)
746 if (refcount_dec_and_test(&kvm->users_count))
749 EXPORT_SYMBOL_GPL(kvm_put_kvm);
752 static int kvm_vm_release(struct inode *inode, struct file *filp)
754 struct kvm *kvm = filp->private_data;
756 kvm_irqfd_release(kvm);
763 * Allocation size is twice as large as the actual dirty bitmap size.
764 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
766 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
768 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
770 memslot->dirty_bitmap = kvzalloc(dirty_bytes, GFP_KERNEL);
771 if (!memslot->dirty_bitmap)
778 * Insert memslot and re-sort memslots based on their GFN,
779 * so binary search could be used to lookup GFN.
780 * Sorting algorithm takes advantage of having initially
781 * sorted array and known changed memslot position.
783 static void update_memslots(struct kvm_memslots *slots,
784 struct kvm_memory_slot *new)
787 int i = slots->id_to_index[id];
788 struct kvm_memory_slot *mslots = slots->memslots;
790 WARN_ON(mslots[i].id != id);
792 WARN_ON(!mslots[i].npages);
793 if (mslots[i].npages)
796 if (!mslots[i].npages)
800 while (i < KVM_MEM_SLOTS_NUM - 1 &&
801 new->base_gfn <= mslots[i + 1].base_gfn) {
802 if (!mslots[i + 1].npages)
804 mslots[i] = mslots[i + 1];
805 slots->id_to_index[mslots[i].id] = i;
810 * The ">=" is needed when creating a slot with base_gfn == 0,
811 * so that it moves before all those with base_gfn == npages == 0.
813 * On the other hand, if new->npages is zero, the above loop has
814 * already left i pointing to the beginning of the empty part of
815 * mslots, and the ">=" would move the hole backwards in this
816 * case---which is wrong. So skip the loop when deleting a slot.
820 new->base_gfn >= mslots[i - 1].base_gfn) {
821 mslots[i] = mslots[i - 1];
822 slots->id_to_index[mslots[i].id] = i;
826 WARN_ON_ONCE(i != slots->used_slots);
829 slots->id_to_index[mslots[i].id] = i;
832 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
834 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
836 #ifdef __KVM_HAVE_READONLY_MEM
837 valid_flags |= KVM_MEM_READONLY;
840 if (mem->flags & ~valid_flags)
846 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
847 int as_id, struct kvm_memslots *slots)
849 struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id);
852 * Set the low bit in the generation, which disables SPTE caching
853 * until the end of synchronize_srcu_expedited.
855 WARN_ON(old_memslots->generation & 1);
856 slots->generation = old_memslots->generation + 1;
858 rcu_assign_pointer(kvm->memslots[as_id], slots);
859 synchronize_srcu_expedited(&kvm->srcu);
862 * Increment the new memslot generation a second time. This prevents
863 * vm exits that race with memslot updates from caching a memslot
864 * generation that will (potentially) be valid forever.
866 * Generations must be unique even across address spaces. We do not need
867 * a global counter for that, instead the generation space is evenly split
868 * across address spaces. For example, with two address spaces, address
869 * space 0 will use generations 0, 4, 8, ... while * address space 1 will
870 * use generations 2, 6, 10, 14, ...
872 slots->generation += KVM_ADDRESS_SPACE_NUM * 2 - 1;
874 kvm_arch_memslots_updated(kvm, slots);
880 * Allocate some memory and give it an address in the guest physical address
883 * Discontiguous memory is allowed, mostly for framebuffers.
885 * Must be called holding kvm->slots_lock for write.
887 int __kvm_set_memory_region(struct kvm *kvm,
888 const struct kvm_userspace_memory_region *mem)
892 unsigned long npages;
893 struct kvm_memory_slot *slot;
894 struct kvm_memory_slot old, new;
895 struct kvm_memslots *slots = NULL, *old_memslots;
897 enum kvm_mr_change change;
899 r = check_memory_region_flags(mem);
904 as_id = mem->slot >> 16;
907 /* General sanity checks */
908 if (mem->memory_size & (PAGE_SIZE - 1))
910 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
912 /* We can read the guest memory with __xxx_user() later on. */
913 if ((id < KVM_USER_MEM_SLOTS) &&
914 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
915 !access_ok(VERIFY_WRITE,
916 (void __user *)(unsigned long)mem->userspace_addr,
919 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
921 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
924 slot = id_to_memslot(__kvm_memslots(kvm, as_id), id);
925 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
926 npages = mem->memory_size >> PAGE_SHIFT;
928 if (npages > KVM_MEM_MAX_NR_PAGES)
934 new.base_gfn = base_gfn;
936 new.flags = mem->flags;
940 change = KVM_MR_CREATE;
941 else { /* Modify an existing slot. */
942 if ((mem->userspace_addr != old.userspace_addr) ||
943 (npages != old.npages) ||
944 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
947 if (base_gfn != old.base_gfn)
948 change = KVM_MR_MOVE;
949 else if (new.flags != old.flags)
950 change = KVM_MR_FLAGS_ONLY;
951 else { /* Nothing to change. */
960 change = KVM_MR_DELETE;
965 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
966 /* Check for overlaps */
968 kvm_for_each_memslot(slot, __kvm_memslots(kvm, as_id)) {
969 if ((slot->id >= KVM_USER_MEM_SLOTS) ||
972 if (!((base_gfn + npages <= slot->base_gfn) ||
973 (base_gfn >= slot->base_gfn + slot->npages)))
978 /* Free page dirty bitmap if unneeded */
979 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
980 new.dirty_bitmap = NULL;
983 if (change == KVM_MR_CREATE) {
984 new.userspace_addr = mem->userspace_addr;
986 if (kvm_arch_create_memslot(kvm, &new, npages))
990 /* Allocate page dirty bitmap if needed */
991 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
992 if (kvm_create_dirty_bitmap(&new) < 0)
996 slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
999 memcpy(slots, __kvm_memslots(kvm, as_id), sizeof(struct kvm_memslots));
1001 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
1002 slot = id_to_memslot(slots, id);
1003 slot->flags |= KVM_MEMSLOT_INVALID;
1005 old_memslots = install_new_memslots(kvm, as_id, slots);
1007 /* From this point no new shadow pages pointing to a deleted,
1008 * or moved, memslot will be created.
1010 * validation of sp->gfn happens in:
1011 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1012 * - kvm_is_visible_gfn (mmu_check_roots)
1014 kvm_arch_flush_shadow_memslot(kvm, slot);
1017 * We can re-use the old_memslots from above, the only difference
1018 * from the currently installed memslots is the invalid flag. This
1019 * will get overwritten by update_memslots anyway.
1021 slots = old_memslots;
1024 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
1028 /* actual memory is freed via old in kvm_free_memslot below */
1029 if (change == KVM_MR_DELETE) {
1030 new.dirty_bitmap = NULL;
1031 memset(&new.arch, 0, sizeof(new.arch));
1034 update_memslots(slots, &new);
1035 old_memslots = install_new_memslots(kvm, as_id, slots);
1037 kvm_arch_commit_memory_region(kvm, mem, &old, &new, change);
1039 kvm_free_memslot(kvm, &old, &new);
1040 kvfree(old_memslots);
1046 kvm_free_memslot(kvm, &new, &old);
1050 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1052 int kvm_set_memory_region(struct kvm *kvm,
1053 const struct kvm_userspace_memory_region *mem)
1057 mutex_lock(&kvm->slots_lock);
1058 r = __kvm_set_memory_region(kvm, mem);
1059 mutex_unlock(&kvm->slots_lock);
1062 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1064 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1065 struct kvm_userspace_memory_region *mem)
1067 if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1070 return kvm_set_memory_region(kvm, mem);
1073 int kvm_get_dirty_log(struct kvm *kvm,
1074 struct kvm_dirty_log *log, int *is_dirty)
1076 struct kvm_memslots *slots;
1077 struct kvm_memory_slot *memslot;
1080 unsigned long any = 0;
1082 as_id = log->slot >> 16;
1083 id = (u16)log->slot;
1084 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1087 slots = __kvm_memslots(kvm, as_id);
1088 memslot = id_to_memslot(slots, id);
1089 if (!memslot->dirty_bitmap)
1092 n = kvm_dirty_bitmap_bytes(memslot);
1094 for (i = 0; !any && i < n/sizeof(long); ++i)
1095 any = memslot->dirty_bitmap[i];
1097 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1104 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1106 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1108 * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages
1109 * are dirty write protect them for next write.
1110 * @kvm: pointer to kvm instance
1111 * @log: slot id and address to which we copy the log
1112 * @is_dirty: flag set if any page is dirty
1114 * We need to keep it in mind that VCPU threads can write to the bitmap
1115 * concurrently. So, to avoid losing track of dirty pages we keep the
1118 * 1. Take a snapshot of the bit and clear it if needed.
1119 * 2. Write protect the corresponding page.
1120 * 3. Copy the snapshot to the userspace.
1121 * 4. Upon return caller flushes TLB's if needed.
1123 * Between 2 and 4, the guest may write to the page using the remaining TLB
1124 * entry. This is not a problem because the page is reported dirty using
1125 * the snapshot taken before and step 4 ensures that writes done after
1126 * exiting to userspace will be logged for the next call.
1129 int kvm_get_dirty_log_protect(struct kvm *kvm,
1130 struct kvm_dirty_log *log, bool *is_dirty)
1132 struct kvm_memslots *slots;
1133 struct kvm_memory_slot *memslot;
1136 unsigned long *dirty_bitmap;
1137 unsigned long *dirty_bitmap_buffer;
1139 as_id = log->slot >> 16;
1140 id = (u16)log->slot;
1141 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1144 slots = __kvm_memslots(kvm, as_id);
1145 memslot = id_to_memslot(slots, id);
1147 dirty_bitmap = memslot->dirty_bitmap;
1151 n = kvm_dirty_bitmap_bytes(memslot);
1153 dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long);
1154 memset(dirty_bitmap_buffer, 0, n);
1156 spin_lock(&kvm->mmu_lock);
1158 for (i = 0; i < n / sizeof(long); i++) {
1162 if (!dirty_bitmap[i])
1167 mask = xchg(&dirty_bitmap[i], 0);
1168 dirty_bitmap_buffer[i] = mask;
1171 offset = i * BITS_PER_LONG;
1172 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1177 spin_unlock(&kvm->mmu_lock);
1178 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1182 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect);
1185 bool kvm_largepages_enabled(void)
1187 return largepages_enabled;
1190 void kvm_disable_largepages(void)
1192 largepages_enabled = false;
1194 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1196 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1198 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1200 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1202 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
1204 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn);
1207 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1209 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1211 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1212 memslot->flags & KVM_MEMSLOT_INVALID)
1217 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1219 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1221 struct vm_area_struct *vma;
1222 unsigned long addr, size;
1226 addr = gfn_to_hva(kvm, gfn);
1227 if (kvm_is_error_hva(addr))
1230 down_read(¤t->mm->mmap_sem);
1231 vma = find_vma(current->mm, addr);
1235 size = vma_kernel_pagesize(vma);
1238 up_read(¤t->mm->mmap_sem);
1243 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1245 return slot->flags & KVM_MEM_READONLY;
1248 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1249 gfn_t *nr_pages, bool write)
1251 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1252 return KVM_HVA_ERR_BAD;
1254 if (memslot_is_readonly(slot) && write)
1255 return KVM_HVA_ERR_RO_BAD;
1258 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1260 return __gfn_to_hva_memslot(slot, gfn);
1263 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1266 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1269 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1272 return gfn_to_hva_many(slot, gfn, NULL);
1274 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1276 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1278 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1280 EXPORT_SYMBOL_GPL(gfn_to_hva);
1282 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
1284 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
1286 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
1289 * If writable is set to false, the hva returned by this function is only
1290 * allowed to be read.
1292 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1293 gfn_t gfn, bool *writable)
1295 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1297 if (!kvm_is_error_hva(hva) && writable)
1298 *writable = !memslot_is_readonly(slot);
1303 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1305 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1307 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1310 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
1312 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1314 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1317 static int get_user_page_nowait(unsigned long start, int write,
1320 int flags = FOLL_NOWAIT | FOLL_HWPOISON;
1323 flags |= FOLL_WRITE;
1325 return get_user_pages(start, 1, flags, page, NULL);
1328 static inline int check_user_page_hwpoison(unsigned long addr)
1330 int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
1332 rc = get_user_pages(addr, 1, flags, NULL, NULL);
1333 return rc == -EHWPOISON;
1337 * The atomic path to get the writable pfn which will be stored in @pfn,
1338 * true indicates success, otherwise false is returned.
1340 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1341 bool write_fault, bool *writable, kvm_pfn_t *pfn)
1343 struct page *page[1];
1346 if (!(async || atomic))
1350 * Fast pin a writable pfn only if it is a write fault request
1351 * or the caller allows to map a writable pfn for a read fault
1354 if (!(write_fault || writable))
1357 npages = __get_user_pages_fast(addr, 1, 1, page);
1359 *pfn = page_to_pfn(page[0]);
1370 * The slow path to get the pfn of the specified host virtual address,
1371 * 1 indicates success, -errno is returned if error is detected.
1373 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1374 bool *writable, kvm_pfn_t *pfn)
1376 struct page *page[1];
1382 *writable = write_fault;
1385 down_read(¤t->mm->mmap_sem);
1386 npages = get_user_page_nowait(addr, write_fault, page);
1387 up_read(¤t->mm->mmap_sem);
1389 unsigned int flags = FOLL_HWPOISON;
1392 flags |= FOLL_WRITE;
1394 npages = get_user_pages_unlocked(addr, 1, page, flags);
1399 /* map read fault as writable if possible */
1400 if (unlikely(!write_fault) && writable) {
1401 struct page *wpage[1];
1403 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1412 *pfn = page_to_pfn(page[0]);
1416 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1418 if (unlikely(!(vma->vm_flags & VM_READ)))
1421 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1427 static int hva_to_pfn_remapped(struct vm_area_struct *vma,
1428 unsigned long addr, bool *async,
1429 bool write_fault, kvm_pfn_t *p_pfn)
1434 r = follow_pfn(vma, addr, &pfn);
1437 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
1438 * not call the fault handler, so do it here.
1440 bool unlocked = false;
1441 r = fixup_user_fault(current, current->mm, addr,
1442 (write_fault ? FAULT_FLAG_WRITE : 0),
1449 r = follow_pfn(vma, addr, &pfn);
1457 * Get a reference here because callers of *hva_to_pfn* and
1458 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
1459 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
1460 * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will
1461 * simply do nothing for reserved pfns.
1463 * Whoever called remap_pfn_range is also going to call e.g.
1464 * unmap_mapping_range before the underlying pages are freed,
1465 * causing a call to our MMU notifier.
1474 * Pin guest page in memory and return its pfn.
1475 * @addr: host virtual address which maps memory to the guest
1476 * @atomic: whether this function can sleep
1477 * @async: whether this function need to wait IO complete if the
1478 * host page is not in the memory
1479 * @write_fault: whether we should get a writable host page
1480 * @writable: whether it allows to map a writable host page for !@write_fault
1482 * The function will map a writable host page for these two cases:
1483 * 1): @write_fault = true
1484 * 2): @write_fault = false && @writable, @writable will tell the caller
1485 * whether the mapping is writable.
1487 static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1488 bool write_fault, bool *writable)
1490 struct vm_area_struct *vma;
1494 /* we can do it either atomically or asynchronously, not both */
1495 BUG_ON(atomic && async);
1497 if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1501 return KVM_PFN_ERR_FAULT;
1503 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1507 down_read(¤t->mm->mmap_sem);
1508 if (npages == -EHWPOISON ||
1509 (!async && check_user_page_hwpoison(addr))) {
1510 pfn = KVM_PFN_ERR_HWPOISON;
1515 vma = find_vma_intersection(current->mm, addr, addr + 1);
1518 pfn = KVM_PFN_ERR_FAULT;
1519 else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
1520 r = hva_to_pfn_remapped(vma, addr, async, write_fault, &pfn);
1524 pfn = KVM_PFN_ERR_FAULT;
1526 if (async && vma_is_valid(vma, write_fault))
1528 pfn = KVM_PFN_ERR_FAULT;
1531 up_read(¤t->mm->mmap_sem);
1535 kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn,
1536 bool atomic, bool *async, bool write_fault,
1539 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1541 if (addr == KVM_HVA_ERR_RO_BAD) {
1544 return KVM_PFN_ERR_RO_FAULT;
1547 if (kvm_is_error_hva(addr)) {
1550 return KVM_PFN_NOSLOT;
1553 /* Do not map writable pfn in the readonly memslot. */
1554 if (writable && memslot_is_readonly(slot)) {
1559 return hva_to_pfn(addr, atomic, async, write_fault,
1562 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
1564 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1567 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
1568 write_fault, writable);
1570 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1572 kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1574 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1576 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
1578 kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1580 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1582 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1584 kvm_pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1586 return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm, gfn), gfn);
1588 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1590 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
1592 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1594 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
1596 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1598 return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
1600 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1602 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
1604 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1606 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
1608 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1609 struct page **pages, int nr_pages)
1614 addr = gfn_to_hva_many(slot, gfn, &entry);
1615 if (kvm_is_error_hva(addr))
1618 if (entry < nr_pages)
1621 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1623 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1625 static struct page *kvm_pfn_to_page(kvm_pfn_t pfn)
1627 if (is_error_noslot_pfn(pfn))
1628 return KVM_ERR_PTR_BAD_PAGE;
1630 if (kvm_is_reserved_pfn(pfn)) {
1632 return KVM_ERR_PTR_BAD_PAGE;
1635 return pfn_to_page(pfn);
1638 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1642 pfn = gfn_to_pfn(kvm, gfn);
1644 return kvm_pfn_to_page(pfn);
1646 EXPORT_SYMBOL_GPL(gfn_to_page);
1648 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
1652 pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
1654 return kvm_pfn_to_page(pfn);
1656 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
1658 void kvm_release_page_clean(struct page *page)
1660 WARN_ON(is_error_page(page));
1662 kvm_release_pfn_clean(page_to_pfn(page));
1664 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1666 void kvm_release_pfn_clean(kvm_pfn_t pfn)
1668 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
1669 put_page(pfn_to_page(pfn));
1671 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1673 void kvm_release_page_dirty(struct page *page)
1675 WARN_ON(is_error_page(page));
1677 kvm_release_pfn_dirty(page_to_pfn(page));
1679 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1681 void kvm_release_pfn_dirty(kvm_pfn_t pfn)
1683 kvm_set_pfn_dirty(pfn);
1684 kvm_release_pfn_clean(pfn);
1686 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1688 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
1690 if (!kvm_is_reserved_pfn(pfn)) {
1691 struct page *page = pfn_to_page(pfn);
1693 if (!PageReserved(page))
1697 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1699 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
1701 if (!kvm_is_reserved_pfn(pfn))
1702 mark_page_accessed(pfn_to_page(pfn));
1704 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1706 void kvm_get_pfn(kvm_pfn_t pfn)
1708 if (!kvm_is_reserved_pfn(pfn))
1709 get_page(pfn_to_page(pfn));
1711 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1713 static int next_segment(unsigned long len, int offset)
1715 if (len > PAGE_SIZE - offset)
1716 return PAGE_SIZE - offset;
1721 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
1722 void *data, int offset, int len)
1727 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1728 if (kvm_is_error_hva(addr))
1730 r = __copy_from_user(data, (void __user *)addr + offset, len);
1736 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1739 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1741 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1743 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1745 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
1746 int offset, int len)
1748 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1750 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1752 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
1754 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1756 gfn_t gfn = gpa >> PAGE_SHIFT;
1758 int offset = offset_in_page(gpa);
1761 while ((seg = next_segment(len, offset)) != 0) {
1762 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1772 EXPORT_SYMBOL_GPL(kvm_read_guest);
1774 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
1776 gfn_t gfn = gpa >> PAGE_SHIFT;
1778 int offset = offset_in_page(gpa);
1781 while ((seg = next_segment(len, offset)) != 0) {
1782 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
1792 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
1794 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1795 void *data, int offset, unsigned long len)
1800 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1801 if (kvm_is_error_hva(addr))
1803 pagefault_disable();
1804 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1811 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1814 gfn_t gfn = gpa >> PAGE_SHIFT;
1815 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1816 int offset = offset_in_page(gpa);
1818 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1820 EXPORT_SYMBOL_GPL(kvm_read_guest_atomic);
1822 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
1823 void *data, unsigned long len)
1825 gfn_t gfn = gpa >> PAGE_SHIFT;
1826 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1827 int offset = offset_in_page(gpa);
1829 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1831 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
1833 static int __kvm_write_guest_page(struct kvm_memory_slot *memslot, gfn_t gfn,
1834 const void *data, int offset, int len)
1839 addr = gfn_to_hva_memslot(memslot, gfn);
1840 if (kvm_is_error_hva(addr))
1842 r = __copy_to_user((void __user *)addr + offset, data, len);
1845 mark_page_dirty_in_slot(memslot, gfn);
1849 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
1850 const void *data, int offset, int len)
1852 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1854 return __kvm_write_guest_page(slot, gfn, data, offset, len);
1856 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1858 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
1859 const void *data, int offset, int len)
1861 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1863 return __kvm_write_guest_page(slot, gfn, data, offset, len);
1865 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
1867 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1870 gfn_t gfn = gpa >> PAGE_SHIFT;
1872 int offset = offset_in_page(gpa);
1875 while ((seg = next_segment(len, offset)) != 0) {
1876 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1886 EXPORT_SYMBOL_GPL(kvm_write_guest);
1888 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
1891 gfn_t gfn = gpa >> PAGE_SHIFT;
1893 int offset = offset_in_page(gpa);
1896 while ((seg = next_segment(len, offset)) != 0) {
1897 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
1907 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
1909 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
1910 struct gfn_to_hva_cache *ghc,
1911 gpa_t gpa, unsigned long len)
1913 int offset = offset_in_page(gpa);
1914 gfn_t start_gfn = gpa >> PAGE_SHIFT;
1915 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1916 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1917 gfn_t nr_pages_avail;
1920 ghc->generation = slots->generation;
1922 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
1923 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, NULL);
1924 if (!kvm_is_error_hva(ghc->hva) && nr_pages_needed <= 1) {
1928 * If the requested region crosses two memslots, we still
1929 * verify that the entire region is valid here.
1931 while (start_gfn <= end_gfn) {
1933 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
1934 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1936 if (kvm_is_error_hva(ghc->hva))
1938 start_gfn += nr_pages_avail;
1940 /* Use the slow path for cross page reads and writes. */
1941 ghc->memslot = NULL;
1946 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1947 gpa_t gpa, unsigned long len)
1949 struct kvm_memslots *slots = kvm_memslots(kvm);
1950 return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len);
1952 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1954 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1955 void *data, int offset, unsigned long len)
1957 struct kvm_memslots *slots = kvm_memslots(kvm);
1959 gpa_t gpa = ghc->gpa + offset;
1961 BUG_ON(len + offset > ghc->len);
1963 if (slots->generation != ghc->generation)
1964 __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len);
1966 if (unlikely(!ghc->memslot))
1967 return kvm_write_guest(kvm, gpa, data, len);
1969 if (kvm_is_error_hva(ghc->hva))
1972 r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
1975 mark_page_dirty_in_slot(ghc->memslot, gpa >> PAGE_SHIFT);
1979 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached);
1981 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1982 void *data, unsigned long len)
1984 return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len);
1986 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1988 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1989 void *data, unsigned long len)
1991 struct kvm_memslots *slots = kvm_memslots(kvm);
1994 BUG_ON(len > ghc->len);
1996 if (slots->generation != ghc->generation)
1997 __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len);
1999 if (unlikely(!ghc->memslot))
2000 return kvm_read_guest(kvm, ghc->gpa, data, len);
2002 if (kvm_is_error_hva(ghc->hva))
2005 r = __copy_from_user(data, (void __user *)ghc->hva, len);
2011 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
2013 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
2015 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
2017 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
2019 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
2021 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
2023 gfn_t gfn = gpa >> PAGE_SHIFT;
2025 int offset = offset_in_page(gpa);
2028 while ((seg = next_segment(len, offset)) != 0) {
2029 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
2038 EXPORT_SYMBOL_GPL(kvm_clear_guest);
2040 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot,
2043 if (memslot && memslot->dirty_bitmap) {
2044 unsigned long rel_gfn = gfn - memslot->base_gfn;
2046 set_bit_le(rel_gfn, memslot->dirty_bitmap);
2050 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
2052 struct kvm_memory_slot *memslot;
2054 memslot = gfn_to_memslot(kvm, gfn);
2055 mark_page_dirty_in_slot(memslot, gfn);
2057 EXPORT_SYMBOL_GPL(mark_page_dirty);
2059 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
2061 struct kvm_memory_slot *memslot;
2063 memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2064 mark_page_dirty_in_slot(memslot, gfn);
2066 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
2068 void kvm_sigset_activate(struct kvm_vcpu *vcpu)
2070 if (!vcpu->sigset_active)
2074 * This does a lockless modification of ->real_blocked, which is fine
2075 * because, only current can change ->real_blocked and all readers of
2076 * ->real_blocked don't care as long ->real_blocked is always a subset
2079 sigprocmask(SIG_SETMASK, &vcpu->sigset, ¤t->real_blocked);
2082 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu)
2084 if (!vcpu->sigset_active)
2087 sigprocmask(SIG_SETMASK, ¤t->real_blocked, NULL);
2088 sigemptyset(¤t->real_blocked);
2091 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
2093 unsigned int old, val, grow;
2095 old = val = vcpu->halt_poll_ns;
2096 grow = READ_ONCE(halt_poll_ns_grow);
2098 if (val == 0 && grow)
2103 if (val > halt_poll_ns)
2106 vcpu->halt_poll_ns = val;
2107 trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
2110 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
2112 unsigned int old, val, shrink;
2114 old = val = vcpu->halt_poll_ns;
2115 shrink = READ_ONCE(halt_poll_ns_shrink);
2121 vcpu->halt_poll_ns = val;
2122 trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
2125 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
2127 if (kvm_arch_vcpu_runnable(vcpu)) {
2128 kvm_make_request(KVM_REQ_UNHALT, vcpu);
2131 if (kvm_cpu_has_pending_timer(vcpu))
2133 if (signal_pending(current))
2140 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2142 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
2145 DECLARE_SWAITQUEUE(wait);
2146 bool waited = false;
2149 start = cur = ktime_get();
2150 if (vcpu->halt_poll_ns) {
2151 ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns);
2153 ++vcpu->stat.halt_attempted_poll;
2156 * This sets KVM_REQ_UNHALT if an interrupt
2159 if (kvm_vcpu_check_block(vcpu) < 0) {
2160 ++vcpu->stat.halt_successful_poll;
2161 if (!vcpu_valid_wakeup(vcpu))
2162 ++vcpu->stat.halt_poll_invalid;
2166 } while (single_task_running() && ktime_before(cur, stop));
2169 kvm_arch_vcpu_blocking(vcpu);
2172 prepare_to_swait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
2174 if (kvm_vcpu_check_block(vcpu) < 0)
2181 finish_swait(&vcpu->wq, &wait);
2184 kvm_arch_vcpu_unblocking(vcpu);
2186 block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
2188 if (!vcpu_valid_wakeup(vcpu))
2189 shrink_halt_poll_ns(vcpu);
2190 else if (halt_poll_ns) {
2191 if (block_ns <= vcpu->halt_poll_ns)
2193 /* we had a long block, shrink polling */
2194 else if (vcpu->halt_poll_ns && block_ns > halt_poll_ns)
2195 shrink_halt_poll_ns(vcpu);
2196 /* we had a short halt and our poll time is too small */
2197 else if (vcpu->halt_poll_ns < halt_poll_ns &&
2198 block_ns < halt_poll_ns)
2199 grow_halt_poll_ns(vcpu);
2201 vcpu->halt_poll_ns = 0;
2203 trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu));
2204 kvm_arch_vcpu_block_finish(vcpu);
2206 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
2208 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
2210 struct swait_queue_head *wqp;
2212 wqp = kvm_arch_vcpu_wq(vcpu);
2213 if (swq_has_sleeper(wqp)) {
2215 ++vcpu->stat.halt_wakeup;
2221 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
2225 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2227 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
2230 int cpu = vcpu->cpu;
2232 if (kvm_vcpu_wake_up(vcpu))
2236 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
2237 if (kvm_arch_vcpu_should_kick(vcpu))
2238 smp_send_reschedule(cpu);
2241 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
2242 #endif /* !CONFIG_S390 */
2244 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
2247 struct task_struct *task = NULL;
2251 pid = rcu_dereference(target->pid);
2253 task = get_pid_task(pid, PIDTYPE_PID);
2257 ret = yield_to(task, 1);
2258 put_task_struct(task);
2262 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
2265 * Helper that checks whether a VCPU is eligible for directed yield.
2266 * Most eligible candidate to yield is decided by following heuristics:
2268 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2269 * (preempted lock holder), indicated by @in_spin_loop.
2270 * Set at the beiginning and cleared at the end of interception/PLE handler.
2272 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2273 * chance last time (mostly it has become eligible now since we have probably
2274 * yielded to lockholder in last iteration. This is done by toggling
2275 * @dy_eligible each time a VCPU checked for eligibility.)
2277 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2278 * to preempted lock-holder could result in wrong VCPU selection and CPU
2279 * burning. Giving priority for a potential lock-holder increases lock
2282 * Since algorithm is based on heuristics, accessing another VCPU data without
2283 * locking does not harm. It may result in trying to yield to same VCPU, fail
2284 * and continue with next VCPU and so on.
2286 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
2288 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2291 eligible = !vcpu->spin_loop.in_spin_loop ||
2292 vcpu->spin_loop.dy_eligible;
2294 if (vcpu->spin_loop.in_spin_loop)
2295 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
2303 void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode)
2305 struct kvm *kvm = me->kvm;
2306 struct kvm_vcpu *vcpu;
2307 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
2313 kvm_vcpu_set_in_spin_loop(me, true);
2315 * We boost the priority of a VCPU that is runnable but not
2316 * currently running, because it got preempted by something
2317 * else and called schedule in __vcpu_run. Hopefully that
2318 * VCPU is holding the lock that we need and will release it.
2319 * We approximate round-robin by starting at the last boosted VCPU.
2321 for (pass = 0; pass < 2 && !yielded && try; pass++) {
2322 kvm_for_each_vcpu(i, vcpu, kvm) {
2323 if (!pass && i <= last_boosted_vcpu) {
2324 i = last_boosted_vcpu;
2326 } else if (pass && i > last_boosted_vcpu)
2328 if (!READ_ONCE(vcpu->preempted))
2332 if (swait_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
2334 if (yield_to_kernel_mode && !kvm_arch_vcpu_in_kernel(vcpu))
2336 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
2339 yielded = kvm_vcpu_yield_to(vcpu);
2341 kvm->last_boosted_vcpu = i;
2343 } else if (yielded < 0) {
2350 kvm_vcpu_set_in_spin_loop(me, false);
2352 /* Ensure vcpu is not eligible during next spinloop */
2353 kvm_vcpu_set_dy_eligible(me, false);
2355 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
2357 static int kvm_vcpu_fault(struct vm_fault *vmf)
2359 struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data;
2362 if (vmf->pgoff == 0)
2363 page = virt_to_page(vcpu->run);
2365 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
2366 page = virt_to_page(vcpu->arch.pio_data);
2368 #ifdef CONFIG_KVM_MMIO
2369 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
2370 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
2373 return kvm_arch_vcpu_fault(vcpu, vmf);
2379 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
2380 .fault = kvm_vcpu_fault,
2383 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2385 vma->vm_ops = &kvm_vcpu_vm_ops;
2389 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2391 struct kvm_vcpu *vcpu = filp->private_data;
2393 debugfs_remove_recursive(vcpu->debugfs_dentry);
2394 kvm_put_kvm(vcpu->kvm);
2398 static struct file_operations kvm_vcpu_fops = {
2399 .release = kvm_vcpu_release,
2400 .unlocked_ioctl = kvm_vcpu_ioctl,
2401 #ifdef CONFIG_KVM_COMPAT
2402 .compat_ioctl = kvm_vcpu_compat_ioctl,
2404 .mmap = kvm_vcpu_mmap,
2405 .llseek = noop_llseek,
2409 * Allocates an inode for the vcpu.
2411 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2413 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
2416 static int kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
2418 char dir_name[ITOA_MAX_LEN * 2];
2421 if (!kvm_arch_has_vcpu_debugfs())
2424 if (!debugfs_initialized())
2427 snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
2428 vcpu->debugfs_dentry = debugfs_create_dir(dir_name,
2429 vcpu->kvm->debugfs_dentry);
2430 if (!vcpu->debugfs_dentry)
2433 ret = kvm_arch_create_vcpu_debugfs(vcpu);
2435 debugfs_remove_recursive(vcpu->debugfs_dentry);
2443 * Creates some virtual cpus. Good luck creating more than one.
2445 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
2448 struct kvm_vcpu *vcpu;
2450 if (id >= KVM_MAX_VCPU_ID)
2453 mutex_lock(&kvm->lock);
2454 if (kvm->created_vcpus == KVM_MAX_VCPUS) {
2455 mutex_unlock(&kvm->lock);
2459 kvm->created_vcpus++;
2460 mutex_unlock(&kvm->lock);
2462 vcpu = kvm_arch_vcpu_create(kvm, id);
2465 goto vcpu_decrement;
2468 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2470 r = kvm_arch_vcpu_setup(vcpu);
2474 r = kvm_create_vcpu_debugfs(vcpu);
2478 mutex_lock(&kvm->lock);
2479 if (kvm_get_vcpu_by_id(kvm, id)) {
2481 goto unlock_vcpu_destroy;
2484 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
2486 /* Now it's all set up, let userspace reach it */
2488 r = create_vcpu_fd(vcpu);
2491 goto unlock_vcpu_destroy;
2494 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
2497 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
2498 * before kvm->online_vcpu's incremented value.
2501 atomic_inc(&kvm->online_vcpus);
2503 mutex_unlock(&kvm->lock);
2504 kvm_arch_vcpu_postcreate(vcpu);
2507 unlock_vcpu_destroy:
2508 mutex_unlock(&kvm->lock);
2509 debugfs_remove_recursive(vcpu->debugfs_dentry);
2511 kvm_arch_vcpu_destroy(vcpu);
2513 mutex_lock(&kvm->lock);
2514 kvm->created_vcpus--;
2515 mutex_unlock(&kvm->lock);
2519 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2522 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2523 vcpu->sigset_active = 1;
2524 vcpu->sigset = *sigset;
2526 vcpu->sigset_active = 0;
2530 static long kvm_vcpu_ioctl(struct file *filp,
2531 unsigned int ioctl, unsigned long arg)
2533 struct kvm_vcpu *vcpu = filp->private_data;
2534 void __user *argp = (void __user *)arg;
2536 struct kvm_fpu *fpu = NULL;
2537 struct kvm_sregs *kvm_sregs = NULL;
2539 if (vcpu->kvm->mm != current->mm)
2542 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2545 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
2547 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
2548 * so vcpu_load() would break it.
2550 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_S390_IRQ || ioctl == KVM_INTERRUPT)
2551 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2555 r = vcpu_load(vcpu);
2564 oldpid = rcu_access_pointer(vcpu->pid);
2565 if (unlikely(oldpid != current->pids[PIDTYPE_PID].pid)) {
2566 /* The thread running this VCPU changed. */
2567 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
2569 rcu_assign_pointer(vcpu->pid, newpid);
2574 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2575 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2578 case KVM_GET_REGS: {
2579 struct kvm_regs *kvm_regs;
2582 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2585 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2589 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2596 case KVM_SET_REGS: {
2597 struct kvm_regs *kvm_regs;
2600 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2601 if (IS_ERR(kvm_regs)) {
2602 r = PTR_ERR(kvm_regs);
2605 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2609 case KVM_GET_SREGS: {
2610 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2614 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2618 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2623 case KVM_SET_SREGS: {
2624 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2625 if (IS_ERR(kvm_sregs)) {
2626 r = PTR_ERR(kvm_sregs);
2630 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2633 case KVM_GET_MP_STATE: {
2634 struct kvm_mp_state mp_state;
2636 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2640 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
2645 case KVM_SET_MP_STATE: {
2646 struct kvm_mp_state mp_state;
2649 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
2651 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2654 case KVM_TRANSLATE: {
2655 struct kvm_translation tr;
2658 if (copy_from_user(&tr, argp, sizeof(tr)))
2660 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2664 if (copy_to_user(argp, &tr, sizeof(tr)))
2669 case KVM_SET_GUEST_DEBUG: {
2670 struct kvm_guest_debug dbg;
2673 if (copy_from_user(&dbg, argp, sizeof(dbg)))
2675 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2678 case KVM_SET_SIGNAL_MASK: {
2679 struct kvm_signal_mask __user *sigmask_arg = argp;
2680 struct kvm_signal_mask kvm_sigmask;
2681 sigset_t sigset, *p;
2686 if (copy_from_user(&kvm_sigmask, argp,
2687 sizeof(kvm_sigmask)))
2690 if (kvm_sigmask.len != sizeof(sigset))
2693 if (copy_from_user(&sigset, sigmask_arg->sigset,
2698 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2702 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2706 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2710 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2716 fpu = memdup_user(argp, sizeof(*fpu));
2722 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2726 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2735 #ifdef CONFIG_KVM_COMPAT
2736 static long kvm_vcpu_compat_ioctl(struct file *filp,
2737 unsigned int ioctl, unsigned long arg)
2739 struct kvm_vcpu *vcpu = filp->private_data;
2740 void __user *argp = compat_ptr(arg);
2743 if (vcpu->kvm->mm != current->mm)
2747 case KVM_SET_SIGNAL_MASK: {
2748 struct kvm_signal_mask __user *sigmask_arg = argp;
2749 struct kvm_signal_mask kvm_sigmask;
2754 if (copy_from_user(&kvm_sigmask, argp,
2755 sizeof(kvm_sigmask)))
2758 if (kvm_sigmask.len != sizeof(compat_sigset_t))
2761 if (get_compat_sigset(&sigset, (void *)sigmask_arg->sigset))
2763 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2765 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2769 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2777 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2778 int (*accessor)(struct kvm_device *dev,
2779 struct kvm_device_attr *attr),
2782 struct kvm_device_attr attr;
2787 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2790 return accessor(dev, &attr);
2793 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2796 struct kvm_device *dev = filp->private_data;
2799 case KVM_SET_DEVICE_ATTR:
2800 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2801 case KVM_GET_DEVICE_ATTR:
2802 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2803 case KVM_HAS_DEVICE_ATTR:
2804 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2806 if (dev->ops->ioctl)
2807 return dev->ops->ioctl(dev, ioctl, arg);
2813 static int kvm_device_release(struct inode *inode, struct file *filp)
2815 struct kvm_device *dev = filp->private_data;
2816 struct kvm *kvm = dev->kvm;
2822 static const struct file_operations kvm_device_fops = {
2823 .unlocked_ioctl = kvm_device_ioctl,
2824 #ifdef CONFIG_KVM_COMPAT
2825 .compat_ioctl = kvm_device_ioctl,
2827 .release = kvm_device_release,
2830 struct kvm_device *kvm_device_from_filp(struct file *filp)
2832 if (filp->f_op != &kvm_device_fops)
2835 return filp->private_data;
2838 static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
2839 #ifdef CONFIG_KVM_MPIC
2840 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
2841 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
2845 int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
2847 if (type >= ARRAY_SIZE(kvm_device_ops_table))
2850 if (kvm_device_ops_table[type] != NULL)
2853 kvm_device_ops_table[type] = ops;
2857 void kvm_unregister_device_ops(u32 type)
2859 if (kvm_device_ops_table[type] != NULL)
2860 kvm_device_ops_table[type] = NULL;
2863 static int kvm_ioctl_create_device(struct kvm *kvm,
2864 struct kvm_create_device *cd)
2866 struct kvm_device_ops *ops = NULL;
2867 struct kvm_device *dev;
2868 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2871 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
2874 ops = kvm_device_ops_table[cd->type];
2881 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2888 mutex_lock(&kvm->lock);
2889 ret = ops->create(dev, cd->type);
2891 mutex_unlock(&kvm->lock);
2895 list_add(&dev->vm_node, &kvm->devices);
2896 mutex_unlock(&kvm->lock);
2901 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2903 mutex_lock(&kvm->lock);
2904 list_del(&dev->vm_node);
2905 mutex_unlock(&kvm->lock);
2915 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2918 case KVM_CAP_USER_MEMORY:
2919 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2920 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2921 case KVM_CAP_INTERNAL_ERROR_DATA:
2922 #ifdef CONFIG_HAVE_KVM_MSI
2923 case KVM_CAP_SIGNAL_MSI:
2925 #ifdef CONFIG_HAVE_KVM_IRQFD
2927 case KVM_CAP_IRQFD_RESAMPLE:
2929 case KVM_CAP_IOEVENTFD_ANY_LENGTH:
2930 case KVM_CAP_CHECK_EXTENSION_VM:
2932 #ifdef CONFIG_KVM_MMIO
2933 case KVM_CAP_COALESCED_MMIO:
2934 return KVM_COALESCED_MMIO_PAGE_OFFSET;
2936 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2937 case KVM_CAP_IRQ_ROUTING:
2938 return KVM_MAX_IRQ_ROUTES;
2940 #if KVM_ADDRESS_SPACE_NUM > 1
2941 case KVM_CAP_MULTI_ADDRESS_SPACE:
2942 return KVM_ADDRESS_SPACE_NUM;
2944 case KVM_CAP_MAX_VCPU_ID:
2945 return KVM_MAX_VCPU_ID;
2949 return kvm_vm_ioctl_check_extension(kvm, arg);
2952 static long kvm_vm_ioctl(struct file *filp,
2953 unsigned int ioctl, unsigned long arg)
2955 struct kvm *kvm = filp->private_data;
2956 void __user *argp = (void __user *)arg;
2959 if (kvm->mm != current->mm)
2962 case KVM_CREATE_VCPU:
2963 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2965 case KVM_SET_USER_MEMORY_REGION: {
2966 struct kvm_userspace_memory_region kvm_userspace_mem;
2969 if (copy_from_user(&kvm_userspace_mem, argp,
2970 sizeof(kvm_userspace_mem)))
2973 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2976 case KVM_GET_DIRTY_LOG: {
2977 struct kvm_dirty_log log;
2980 if (copy_from_user(&log, argp, sizeof(log)))
2982 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2985 #ifdef CONFIG_KVM_MMIO
2986 case KVM_REGISTER_COALESCED_MMIO: {
2987 struct kvm_coalesced_mmio_zone zone;
2990 if (copy_from_user(&zone, argp, sizeof(zone)))
2992 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2995 case KVM_UNREGISTER_COALESCED_MMIO: {
2996 struct kvm_coalesced_mmio_zone zone;
2999 if (copy_from_user(&zone, argp, sizeof(zone)))
3001 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
3006 struct kvm_irqfd data;
3009 if (copy_from_user(&data, argp, sizeof(data)))
3011 r = kvm_irqfd(kvm, &data);
3014 case KVM_IOEVENTFD: {
3015 struct kvm_ioeventfd data;
3018 if (copy_from_user(&data, argp, sizeof(data)))
3020 r = kvm_ioeventfd(kvm, &data);
3023 #ifdef CONFIG_HAVE_KVM_MSI
3024 case KVM_SIGNAL_MSI: {
3028 if (copy_from_user(&msi, argp, sizeof(msi)))
3030 r = kvm_send_userspace_msi(kvm, &msi);
3034 #ifdef __KVM_HAVE_IRQ_LINE
3035 case KVM_IRQ_LINE_STATUS:
3036 case KVM_IRQ_LINE: {
3037 struct kvm_irq_level irq_event;
3040 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
3043 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
3044 ioctl == KVM_IRQ_LINE_STATUS);
3049 if (ioctl == KVM_IRQ_LINE_STATUS) {
3050 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
3058 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3059 case KVM_SET_GSI_ROUTING: {
3060 struct kvm_irq_routing routing;
3061 struct kvm_irq_routing __user *urouting;
3062 struct kvm_irq_routing_entry *entries = NULL;
3065 if (copy_from_user(&routing, argp, sizeof(routing)))
3068 if (!kvm_arch_can_set_irq_routing(kvm))
3070 if (routing.nr > KVM_MAX_IRQ_ROUTES)
3076 entries = vmalloc(routing.nr * sizeof(*entries));
3081 if (copy_from_user(entries, urouting->entries,
3082 routing.nr * sizeof(*entries)))
3083 goto out_free_irq_routing;
3085 r = kvm_set_irq_routing(kvm, entries, routing.nr,
3087 out_free_irq_routing:
3091 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
3092 case KVM_CREATE_DEVICE: {
3093 struct kvm_create_device cd;
3096 if (copy_from_user(&cd, argp, sizeof(cd)))
3099 r = kvm_ioctl_create_device(kvm, &cd);
3104 if (copy_to_user(argp, &cd, sizeof(cd)))
3110 case KVM_CHECK_EXTENSION:
3111 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
3114 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
3120 #ifdef CONFIG_KVM_COMPAT
3121 struct compat_kvm_dirty_log {
3125 compat_uptr_t dirty_bitmap; /* one bit per page */
3130 static long kvm_vm_compat_ioctl(struct file *filp,
3131 unsigned int ioctl, unsigned long arg)
3133 struct kvm *kvm = filp->private_data;
3136 if (kvm->mm != current->mm)
3139 case KVM_GET_DIRTY_LOG: {
3140 struct compat_kvm_dirty_log compat_log;
3141 struct kvm_dirty_log log;
3143 if (copy_from_user(&compat_log, (void __user *)arg,
3144 sizeof(compat_log)))
3146 log.slot = compat_log.slot;
3147 log.padding1 = compat_log.padding1;
3148 log.padding2 = compat_log.padding2;
3149 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
3151 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3155 r = kvm_vm_ioctl(filp, ioctl, arg);
3161 static struct file_operations kvm_vm_fops = {
3162 .release = kvm_vm_release,
3163 .unlocked_ioctl = kvm_vm_ioctl,
3164 #ifdef CONFIG_KVM_COMPAT
3165 .compat_ioctl = kvm_vm_compat_ioctl,
3167 .llseek = noop_llseek,
3170 static int kvm_dev_ioctl_create_vm(unsigned long type)
3176 kvm = kvm_create_vm(type);
3178 return PTR_ERR(kvm);
3179 #ifdef CONFIG_KVM_MMIO
3180 r = kvm_coalesced_mmio_init(kvm);
3186 r = get_unused_fd_flags(O_CLOEXEC);
3191 file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
3195 return PTR_ERR(file);
3199 * Don't call kvm_put_kvm anymore at this point; file->f_op is
3200 * already set, with ->release() being kvm_vm_release(). In error
3201 * cases it will be called by the final fput(file) and will take
3202 * care of doing kvm_put_kvm(kvm).
3204 if (kvm_create_vm_debugfs(kvm, r) < 0) {
3209 kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm);
3211 fd_install(r, file);
3215 static long kvm_dev_ioctl(struct file *filp,
3216 unsigned int ioctl, unsigned long arg)
3221 case KVM_GET_API_VERSION:
3224 r = KVM_API_VERSION;
3227 r = kvm_dev_ioctl_create_vm(arg);
3229 case KVM_CHECK_EXTENSION:
3230 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
3232 case KVM_GET_VCPU_MMAP_SIZE:
3235 r = PAGE_SIZE; /* struct kvm_run */
3237 r += PAGE_SIZE; /* pio data page */
3239 #ifdef CONFIG_KVM_MMIO
3240 r += PAGE_SIZE; /* coalesced mmio ring page */
3243 case KVM_TRACE_ENABLE:
3244 case KVM_TRACE_PAUSE:
3245 case KVM_TRACE_DISABLE:
3249 return kvm_arch_dev_ioctl(filp, ioctl, arg);
3255 static struct file_operations kvm_chardev_ops = {
3256 .unlocked_ioctl = kvm_dev_ioctl,
3257 .compat_ioctl = kvm_dev_ioctl,
3258 .llseek = noop_llseek,
3261 static struct miscdevice kvm_dev = {
3267 static void hardware_enable_nolock(void *junk)
3269 int cpu = raw_smp_processor_id();
3272 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
3275 cpumask_set_cpu(cpu, cpus_hardware_enabled);
3277 r = kvm_arch_hardware_enable();
3280 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3281 atomic_inc(&hardware_enable_failed);
3282 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
3286 static int kvm_starting_cpu(unsigned int cpu)
3288 raw_spin_lock(&kvm_count_lock);
3289 if (kvm_usage_count)
3290 hardware_enable_nolock(NULL);
3291 raw_spin_unlock(&kvm_count_lock);
3295 static void hardware_disable_nolock(void *junk)
3297 int cpu = raw_smp_processor_id();
3299 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
3301 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3302 kvm_arch_hardware_disable();
3305 static int kvm_dying_cpu(unsigned int cpu)
3307 raw_spin_lock(&kvm_count_lock);
3308 if (kvm_usage_count)
3309 hardware_disable_nolock(NULL);
3310 raw_spin_unlock(&kvm_count_lock);
3314 static void hardware_disable_all_nolock(void)
3316 BUG_ON(!kvm_usage_count);
3319 if (!kvm_usage_count)
3320 on_each_cpu(hardware_disable_nolock, NULL, 1);
3323 static void hardware_disable_all(void)
3325 raw_spin_lock(&kvm_count_lock);
3326 hardware_disable_all_nolock();
3327 raw_spin_unlock(&kvm_count_lock);
3330 static int hardware_enable_all(void)
3334 raw_spin_lock(&kvm_count_lock);
3337 if (kvm_usage_count == 1) {
3338 atomic_set(&hardware_enable_failed, 0);
3339 on_each_cpu(hardware_enable_nolock, NULL, 1);
3341 if (atomic_read(&hardware_enable_failed)) {
3342 hardware_disable_all_nolock();
3347 raw_spin_unlock(&kvm_count_lock);
3352 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3356 * Some (well, at least mine) BIOSes hang on reboot if
3359 * And Intel TXT required VMX off for all cpu when system shutdown.
3361 pr_info("kvm: exiting hardware virtualization\n");
3362 kvm_rebooting = true;
3363 on_each_cpu(hardware_disable_nolock, NULL, 1);
3367 static struct notifier_block kvm_reboot_notifier = {
3368 .notifier_call = kvm_reboot,
3372 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3376 for (i = 0; i < bus->dev_count; i++) {
3377 struct kvm_io_device *pos = bus->range[i].dev;
3379 kvm_iodevice_destructor(pos);
3384 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
3385 const struct kvm_io_range *r2)
3387 gpa_t addr1 = r1->addr;
3388 gpa_t addr2 = r2->addr;
3393 /* If r2->len == 0, match the exact address. If r2->len != 0,
3394 * accept any overlapping write. Any order is acceptable for
3395 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
3396 * we process all of them.
3409 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
3411 return kvm_io_bus_cmp(p1, p2);
3414 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
3415 gpa_t addr, int len)
3417 bus->range[bus->dev_count++] = (struct kvm_io_range) {
3423 sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
3424 kvm_io_bus_sort_cmp, NULL);
3429 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
3430 gpa_t addr, int len)
3432 struct kvm_io_range *range, key;
3435 key = (struct kvm_io_range) {
3440 range = bsearch(&key, bus->range, bus->dev_count,
3441 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
3445 off = range - bus->range;
3447 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
3453 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3454 struct kvm_io_range *range, const void *val)
3458 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3462 while (idx < bus->dev_count &&
3463 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3464 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
3473 /* kvm_io_bus_write - called under kvm->slots_lock */
3474 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3475 int len, const void *val)
3477 struct kvm_io_bus *bus;
3478 struct kvm_io_range range;
3481 range = (struct kvm_io_range) {
3486 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3489 r = __kvm_io_bus_write(vcpu, bus, &range, val);
3490 return r < 0 ? r : 0;
3493 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3494 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
3495 gpa_t addr, int len, const void *val, long cookie)
3497 struct kvm_io_bus *bus;
3498 struct kvm_io_range range;
3500 range = (struct kvm_io_range) {
3505 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3509 /* First try the device referenced by cookie. */
3510 if ((cookie >= 0) && (cookie < bus->dev_count) &&
3511 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
3512 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
3517 * cookie contained garbage; fall back to search and return the
3518 * correct cookie value.
3520 return __kvm_io_bus_write(vcpu, bus, &range, val);
3523 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3524 struct kvm_io_range *range, void *val)
3528 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3532 while (idx < bus->dev_count &&
3533 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3534 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
3542 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
3544 /* kvm_io_bus_read - called under kvm->slots_lock */
3545 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3548 struct kvm_io_bus *bus;
3549 struct kvm_io_range range;
3552 range = (struct kvm_io_range) {
3557 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3560 r = __kvm_io_bus_read(vcpu, bus, &range, val);
3561 return r < 0 ? r : 0;
3565 /* Caller must hold slots_lock. */
3566 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3567 int len, struct kvm_io_device *dev)
3569 struct kvm_io_bus *new_bus, *bus;
3571 bus = kvm_get_bus(kvm, bus_idx);
3575 /* exclude ioeventfd which is limited by maximum fd */
3576 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3579 new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count + 1) *
3580 sizeof(struct kvm_io_range)), GFP_KERNEL);
3583 memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
3584 sizeof(struct kvm_io_range)));
3585 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
3586 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3587 synchronize_srcu_expedited(&kvm->srcu);
3593 /* Caller must hold slots_lock. */
3594 void kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3595 struct kvm_io_device *dev)
3598 struct kvm_io_bus *new_bus, *bus;
3600 bus = kvm_get_bus(kvm, bus_idx);
3604 for (i = 0; i < bus->dev_count; i++)
3605 if (bus->range[i].dev == dev) {
3609 if (i == bus->dev_count)
3612 new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3613 sizeof(struct kvm_io_range)), GFP_KERNEL);
3615 pr_err("kvm: failed to shrink bus, removing it completely\n");
3619 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3620 new_bus->dev_count--;
3621 memcpy(new_bus->range + i, bus->range + i + 1,
3622 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3625 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3626 synchronize_srcu_expedited(&kvm->srcu);
3631 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3634 struct kvm_io_bus *bus;
3635 int dev_idx, srcu_idx;
3636 struct kvm_io_device *iodev = NULL;
3638 srcu_idx = srcu_read_lock(&kvm->srcu);
3640 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
3644 dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
3648 iodev = bus->range[dev_idx].dev;
3651 srcu_read_unlock(&kvm->srcu, srcu_idx);
3655 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
3657 static int kvm_debugfs_open(struct inode *inode, struct file *file,
3658 int (*get)(void *, u64 *), int (*set)(void *, u64),
3661 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
3664 /* The debugfs files are a reference to the kvm struct which
3665 * is still valid when kvm_destroy_vm is called.
3666 * To avoid the race between open and the removal of the debugfs
3667 * directory we test against the users count.
3669 if (!refcount_inc_not_zero(&stat_data->kvm->users_count))
3672 if (simple_attr_open(inode, file, get, set, fmt)) {
3673 kvm_put_kvm(stat_data->kvm);
3680 static int kvm_debugfs_release(struct inode *inode, struct file *file)
3682 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
3685 simple_attr_release(inode, file);
3686 kvm_put_kvm(stat_data->kvm);
3691 static int vm_stat_get_per_vm(void *data, u64 *val)
3693 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3695 *val = *(ulong *)((void *)stat_data->kvm + stat_data->offset);
3700 static int vm_stat_clear_per_vm(void *data, u64 val)
3702 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3707 *(ulong *)((void *)stat_data->kvm + stat_data->offset) = 0;
3712 static int vm_stat_get_per_vm_open(struct inode *inode, struct file *file)
3714 __simple_attr_check_format("%llu\n", 0ull);
3715 return kvm_debugfs_open(inode, file, vm_stat_get_per_vm,
3716 vm_stat_clear_per_vm, "%llu\n");
3719 static const struct file_operations vm_stat_get_per_vm_fops = {
3720 .owner = THIS_MODULE,
3721 .open = vm_stat_get_per_vm_open,
3722 .release = kvm_debugfs_release,
3723 .read = simple_attr_read,
3724 .write = simple_attr_write,
3725 .llseek = no_llseek,
3728 static int vcpu_stat_get_per_vm(void *data, u64 *val)
3731 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3732 struct kvm_vcpu *vcpu;
3736 kvm_for_each_vcpu(i, vcpu, stat_data->kvm)
3737 *val += *(u64 *)((void *)vcpu + stat_data->offset);
3742 static int vcpu_stat_clear_per_vm(void *data, u64 val)
3745 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3746 struct kvm_vcpu *vcpu;
3751 kvm_for_each_vcpu(i, vcpu, stat_data->kvm)
3752 *(u64 *)((void *)vcpu + stat_data->offset) = 0;
3757 static int vcpu_stat_get_per_vm_open(struct inode *inode, struct file *file)
3759 __simple_attr_check_format("%llu\n", 0ull);
3760 return kvm_debugfs_open(inode, file, vcpu_stat_get_per_vm,
3761 vcpu_stat_clear_per_vm, "%llu\n");
3764 static const struct file_operations vcpu_stat_get_per_vm_fops = {
3765 .owner = THIS_MODULE,
3766 .open = vcpu_stat_get_per_vm_open,
3767 .release = kvm_debugfs_release,
3768 .read = simple_attr_read,
3769 .write = simple_attr_write,
3770 .llseek = no_llseek,
3773 static const struct file_operations *stat_fops_per_vm[] = {
3774 [KVM_STAT_VCPU] = &vcpu_stat_get_per_vm_fops,
3775 [KVM_STAT_VM] = &vm_stat_get_per_vm_fops,
3778 static int vm_stat_get(void *_offset, u64 *val)
3780 unsigned offset = (long)_offset;
3782 struct kvm_stat_data stat_tmp = {.offset = offset};
3786 spin_lock(&kvm_lock);
3787 list_for_each_entry(kvm, &vm_list, vm_list) {
3789 vm_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
3792 spin_unlock(&kvm_lock);
3796 static int vm_stat_clear(void *_offset, u64 val)
3798 unsigned offset = (long)_offset;
3800 struct kvm_stat_data stat_tmp = {.offset = offset};
3805 spin_lock(&kvm_lock);
3806 list_for_each_entry(kvm, &vm_list, vm_list) {
3808 vm_stat_clear_per_vm((void *)&stat_tmp, 0);
3810 spin_unlock(&kvm_lock);
3815 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n");
3817 static int vcpu_stat_get(void *_offset, u64 *val)
3819 unsigned offset = (long)_offset;
3821 struct kvm_stat_data stat_tmp = {.offset = offset};
3825 spin_lock(&kvm_lock);
3826 list_for_each_entry(kvm, &vm_list, vm_list) {
3828 vcpu_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
3831 spin_unlock(&kvm_lock);
3835 static int vcpu_stat_clear(void *_offset, u64 val)
3837 unsigned offset = (long)_offset;
3839 struct kvm_stat_data stat_tmp = {.offset = offset};
3844 spin_lock(&kvm_lock);
3845 list_for_each_entry(kvm, &vm_list, vm_list) {
3847 vcpu_stat_clear_per_vm((void *)&stat_tmp, 0);
3849 spin_unlock(&kvm_lock);
3854 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear,
3857 static const struct file_operations *stat_fops[] = {
3858 [KVM_STAT_VCPU] = &vcpu_stat_fops,
3859 [KVM_STAT_VM] = &vm_stat_fops,
3862 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm)
3864 struct kobj_uevent_env *env;
3865 unsigned long long created, active;
3867 if (!kvm_dev.this_device || !kvm)
3870 spin_lock(&kvm_lock);
3871 if (type == KVM_EVENT_CREATE_VM) {
3872 kvm_createvm_count++;
3874 } else if (type == KVM_EVENT_DESTROY_VM) {
3877 created = kvm_createvm_count;
3878 active = kvm_active_vms;
3879 spin_unlock(&kvm_lock);
3881 env = kzalloc(sizeof(*env), GFP_KERNEL);
3885 add_uevent_var(env, "CREATED=%llu", created);
3886 add_uevent_var(env, "COUNT=%llu", active);
3888 if (type == KVM_EVENT_CREATE_VM) {
3889 add_uevent_var(env, "EVENT=create");
3890 kvm->userspace_pid = task_pid_nr(current);
3891 } else if (type == KVM_EVENT_DESTROY_VM) {
3892 add_uevent_var(env, "EVENT=destroy");
3894 add_uevent_var(env, "PID=%d", kvm->userspace_pid);
3896 if (kvm->debugfs_dentry) {
3897 char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL);
3900 tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX);
3902 add_uevent_var(env, "STATS_PATH=%s", tmp);
3906 /* no need for checks, since we are adding at most only 5 keys */
3907 env->envp[env->envp_idx++] = NULL;
3908 kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp);
3912 static int kvm_init_debug(void)
3915 struct kvm_stats_debugfs_item *p;
3917 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3918 if (kvm_debugfs_dir == NULL)
3921 kvm_debugfs_num_entries = 0;
3922 for (p = debugfs_entries; p->name; ++p, kvm_debugfs_num_entries++) {
3923 if (!debugfs_create_file(p->name, 0644, kvm_debugfs_dir,
3924 (void *)(long)p->offset,
3925 stat_fops[p->kind]))
3932 debugfs_remove_recursive(kvm_debugfs_dir);
3937 static int kvm_suspend(void)
3939 if (kvm_usage_count)
3940 hardware_disable_nolock(NULL);
3944 static void kvm_resume(void)
3946 if (kvm_usage_count) {
3947 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3948 hardware_enable_nolock(NULL);
3952 static struct syscore_ops kvm_syscore_ops = {
3953 .suspend = kvm_suspend,
3954 .resume = kvm_resume,
3958 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3960 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3963 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3965 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3967 if (vcpu->preempted)
3968 vcpu->preempted = false;
3970 kvm_arch_sched_in(vcpu, cpu);
3972 kvm_arch_vcpu_load(vcpu, cpu);
3975 static void kvm_sched_out(struct preempt_notifier *pn,
3976 struct task_struct *next)
3978 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3980 if (current->state == TASK_RUNNING)
3981 vcpu->preempted = true;
3982 kvm_arch_vcpu_put(vcpu);
3985 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3986 struct module *module)
3991 r = kvm_arch_init(opaque);
3996 * kvm_arch_init makes sure there's at most one caller
3997 * for architectures that support multiple implementations,
3998 * like intel and amd on x86.
3999 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
4000 * conflicts in case kvm is already setup for another implementation.
4002 r = kvm_irqfd_init();
4006 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
4011 r = kvm_arch_hardware_setup();
4015 for_each_online_cpu(cpu) {
4016 smp_call_function_single(cpu,
4017 kvm_arch_check_processor_compat,
4023 r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting",
4024 kvm_starting_cpu, kvm_dying_cpu);
4027 register_reboot_notifier(&kvm_reboot_notifier);
4029 /* A kmem cache lets us meet the alignment requirements of fx_save. */
4031 vcpu_align = __alignof__(struct kvm_vcpu);
4032 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
4033 SLAB_ACCOUNT, NULL);
4034 if (!kvm_vcpu_cache) {
4039 r = kvm_async_pf_init();
4043 kvm_chardev_ops.owner = module;
4044 kvm_vm_fops.owner = module;
4045 kvm_vcpu_fops.owner = module;
4047 r = misc_register(&kvm_dev);
4049 pr_err("kvm: misc device register failed\n");
4053 register_syscore_ops(&kvm_syscore_ops);
4055 kvm_preempt_ops.sched_in = kvm_sched_in;
4056 kvm_preempt_ops.sched_out = kvm_sched_out;
4058 r = kvm_init_debug();
4060 pr_err("kvm: create debugfs files failed\n");
4064 r = kvm_vfio_ops_init();
4070 unregister_syscore_ops(&kvm_syscore_ops);
4071 misc_deregister(&kvm_dev);
4073 kvm_async_pf_deinit();
4075 kmem_cache_destroy(kvm_vcpu_cache);
4077 unregister_reboot_notifier(&kvm_reboot_notifier);
4078 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4081 kvm_arch_hardware_unsetup();
4083 free_cpumask_var(cpus_hardware_enabled);
4091 EXPORT_SYMBOL_GPL(kvm_init);
4095 debugfs_remove_recursive(kvm_debugfs_dir);
4096 misc_deregister(&kvm_dev);
4097 kmem_cache_destroy(kvm_vcpu_cache);
4098 kvm_async_pf_deinit();
4099 unregister_syscore_ops(&kvm_syscore_ops);
4100 unregister_reboot_notifier(&kvm_reboot_notifier);
4101 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4102 on_each_cpu(hardware_disable_nolock, NULL, 1);
4103 kvm_arch_hardware_unsetup();
4106 free_cpumask_var(cpus_hardware_enabled);
4107 kvm_vfio_ops_exit();
4109 EXPORT_SYMBOL_GPL(kvm_exit);