1 // SPDX-License-Identifier: GPL-2.0-only
3 * Kernel-based Virtual Machine driver for Linux
5 * This module enables machines with Intel VT-x extensions to run virtual
6 * machines without emulation or binary translation.
8 * Copyright (C) 2006 Qumranet, Inc.
9 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
12 * Avi Kivity <avi@qumranet.com>
13 * Yaniv Kamay <yaniv@qumranet.com>
16 #include <kvm/iodev.h>
18 #include <linux/kvm_host.h>
19 #include <linux/kvm.h>
20 #include <linux/module.h>
21 #include <linux/errno.h>
22 #include <linux/percpu.h>
24 #include <linux/miscdevice.h>
25 #include <linux/vmalloc.h>
26 #include <linux/reboot.h>
27 #include <linux/debugfs.h>
28 #include <linux/highmem.h>
29 #include <linux/file.h>
30 #include <linux/syscore_ops.h>
31 #include <linux/cpu.h>
32 #include <linux/sched/signal.h>
33 #include <linux/sched/mm.h>
34 #include <linux/sched/stat.h>
35 #include <linux/cpumask.h>
36 #include <linux/smp.h>
37 #include <linux/anon_inodes.h>
38 #include <linux/profile.h>
39 #include <linux/kvm_para.h>
40 #include <linux/pagemap.h>
41 #include <linux/mman.h>
42 #include <linux/swap.h>
43 #include <linux/bitops.h>
44 #include <linux/spinlock.h>
45 #include <linux/compat.h>
46 #include <linux/srcu.h>
47 #include <linux/hugetlb.h>
48 #include <linux/slab.h>
49 #include <linux/sort.h>
50 #include <linux/bsearch.h>
52 #include <linux/lockdep.h>
53 #include <linux/kthread.h>
55 #include <asm/processor.h>
56 #include <asm/ioctl.h>
57 #include <linux/uaccess.h>
59 #include "coalesced_mmio.h"
63 #define CREATE_TRACE_POINTS
64 #include <trace/events/kvm.h>
66 /* Worst case buffer size needed for holding an integer. */
67 #define ITOA_MAX_LEN 12
69 MODULE_AUTHOR("Qumranet");
70 MODULE_LICENSE("GPL");
72 /* Architectures should define their poll value according to the halt latency */
73 unsigned int halt_poll_ns = KVM_HALT_POLL_NS_DEFAULT;
74 module_param(halt_poll_ns, uint, 0644);
75 EXPORT_SYMBOL_GPL(halt_poll_ns);
77 /* Default doubles per-vcpu halt_poll_ns. */
78 unsigned int halt_poll_ns_grow = 2;
79 module_param(halt_poll_ns_grow, uint, 0644);
80 EXPORT_SYMBOL_GPL(halt_poll_ns_grow);
82 /* The start value to grow halt_poll_ns from */
83 unsigned int halt_poll_ns_grow_start = 10000; /* 10us */
84 module_param(halt_poll_ns_grow_start, uint, 0644);
85 EXPORT_SYMBOL_GPL(halt_poll_ns_grow_start);
87 /* Default resets per-vcpu halt_poll_ns . */
88 unsigned int halt_poll_ns_shrink;
89 module_param(halt_poll_ns_shrink, uint, 0644);
90 EXPORT_SYMBOL_GPL(halt_poll_ns_shrink);
95 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
98 DEFINE_MUTEX(kvm_lock);
99 static DEFINE_RAW_SPINLOCK(kvm_count_lock);
102 static cpumask_var_t cpus_hardware_enabled;
103 static int kvm_usage_count;
104 static atomic_t hardware_enable_failed;
106 static struct kmem_cache *kvm_vcpu_cache;
108 static __read_mostly struct preempt_ops kvm_preempt_ops;
109 static DEFINE_PER_CPU(struct kvm_vcpu *, kvm_running_vcpu);
111 struct dentry *kvm_debugfs_dir;
112 EXPORT_SYMBOL_GPL(kvm_debugfs_dir);
114 static int kvm_debugfs_num_entries;
115 static const struct file_operations stat_fops_per_vm;
117 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
119 #ifdef CONFIG_KVM_COMPAT
120 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
122 #define KVM_COMPAT(c) .compat_ioctl = (c)
125 * For architectures that don't implement a compat infrastructure,
126 * adopt a double line of defense:
127 * - Prevent a compat task from opening /dev/kvm
128 * - If the open has been done by a 64bit task, and the KVM fd
129 * passed to a compat task, let the ioctls fail.
131 static long kvm_no_compat_ioctl(struct file *file, unsigned int ioctl,
132 unsigned long arg) { return -EINVAL; }
134 static int kvm_no_compat_open(struct inode *inode, struct file *file)
136 return is_compat_task() ? -ENODEV : 0;
138 #define KVM_COMPAT(c) .compat_ioctl = kvm_no_compat_ioctl, \
139 .open = kvm_no_compat_open
141 static int hardware_enable_all(void);
142 static void hardware_disable_all(void);
144 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
146 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot, gfn_t gfn);
148 __visible bool kvm_rebooting;
149 EXPORT_SYMBOL_GPL(kvm_rebooting);
151 #define KVM_EVENT_CREATE_VM 0
152 #define KVM_EVENT_DESTROY_VM 1
153 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm);
154 static unsigned long long kvm_createvm_count;
155 static unsigned long long kvm_active_vms;
157 __weak int kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
158 unsigned long start, unsigned long end, bool blockable)
163 bool kvm_is_zone_device_pfn(kvm_pfn_t pfn)
166 * The metadata used by is_zone_device_page() to determine whether or
167 * not a page is ZONE_DEVICE is guaranteed to be valid if and only if
168 * the device has been pinned, e.g. by get_user_pages(). WARN if the
169 * page_count() is zero to help detect bad usage of this helper.
171 if (!pfn_valid(pfn) || WARN_ON_ONCE(!page_count(pfn_to_page(pfn))))
174 return is_zone_device_page(pfn_to_page(pfn));
177 bool kvm_is_reserved_pfn(kvm_pfn_t pfn)
180 * ZONE_DEVICE pages currently set PG_reserved, but from a refcounting
181 * perspective they are "normal" pages, albeit with slightly different
185 return PageReserved(pfn_to_page(pfn)) &&
187 !kvm_is_zone_device_pfn(pfn);
192 bool kvm_is_transparent_hugepage(kvm_pfn_t pfn)
194 struct page *page = pfn_to_page(pfn);
196 if (!PageTransCompoundMap(page))
199 return is_transparent_hugepage(compound_head(page));
203 * Switches to specified vcpu, until a matching vcpu_put()
205 void vcpu_load(struct kvm_vcpu *vcpu)
209 __this_cpu_write(kvm_running_vcpu, vcpu);
210 preempt_notifier_register(&vcpu->preempt_notifier);
211 kvm_arch_vcpu_load(vcpu, cpu);
214 EXPORT_SYMBOL_GPL(vcpu_load);
216 void vcpu_put(struct kvm_vcpu *vcpu)
219 kvm_arch_vcpu_put(vcpu);
220 preempt_notifier_unregister(&vcpu->preempt_notifier);
221 __this_cpu_write(kvm_running_vcpu, NULL);
224 EXPORT_SYMBOL_GPL(vcpu_put);
226 /* TODO: merge with kvm_arch_vcpu_should_kick */
227 static bool kvm_request_needs_ipi(struct kvm_vcpu *vcpu, unsigned req)
229 int mode = kvm_vcpu_exiting_guest_mode(vcpu);
232 * We need to wait for the VCPU to reenable interrupts and get out of
233 * READING_SHADOW_PAGE_TABLES mode.
235 if (req & KVM_REQUEST_WAIT)
236 return mode != OUTSIDE_GUEST_MODE;
239 * Need to kick a running VCPU, but otherwise there is nothing to do.
241 return mode == IN_GUEST_MODE;
244 static void ack_flush(void *_completed)
248 static inline bool kvm_kick_many_cpus(const struct cpumask *cpus, bool wait)
251 cpus = cpu_online_mask;
253 if (cpumask_empty(cpus))
256 smp_call_function_many(cpus, ack_flush, NULL, wait);
260 bool kvm_make_vcpus_request_mask(struct kvm *kvm, unsigned int req,
261 struct kvm_vcpu *except,
262 unsigned long *vcpu_bitmap, cpumask_var_t tmp)
265 struct kvm_vcpu *vcpu;
270 kvm_for_each_vcpu(i, vcpu, kvm) {
271 if ((vcpu_bitmap && !test_bit(i, vcpu_bitmap)) ||
275 kvm_make_request(req, vcpu);
278 if (!(req & KVM_REQUEST_NO_WAKEUP) && kvm_vcpu_wake_up(vcpu))
281 if (tmp != NULL && cpu != -1 && cpu != me &&
282 kvm_request_needs_ipi(vcpu, req))
283 __cpumask_set_cpu(cpu, tmp);
286 called = kvm_kick_many_cpus(tmp, !!(req & KVM_REQUEST_WAIT));
292 bool kvm_make_all_cpus_request_except(struct kvm *kvm, unsigned int req,
293 struct kvm_vcpu *except)
298 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
300 called = kvm_make_vcpus_request_mask(kvm, req, except, NULL, cpus);
302 free_cpumask_var(cpus);
306 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
308 return kvm_make_all_cpus_request_except(kvm, req, NULL);
311 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
312 void kvm_flush_remote_tlbs(struct kvm *kvm)
315 * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in
316 * kvm_make_all_cpus_request.
318 long dirty_count = smp_load_acquire(&kvm->tlbs_dirty);
321 * We want to publish modifications to the page tables before reading
322 * mode. Pairs with a memory barrier in arch-specific code.
323 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
324 * and smp_mb in walk_shadow_page_lockless_begin/end.
325 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
327 * There is already an smp_mb__after_atomic() before
328 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
331 if (!kvm_arch_flush_remote_tlb(kvm)
332 || kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
333 ++kvm->stat.remote_tlb_flush;
334 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
336 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
339 void kvm_reload_remote_mmus(struct kvm *kvm)
341 kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
344 static void kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
346 mutex_init(&vcpu->mutex);
351 rcuwait_init(&vcpu->wait);
352 kvm_async_pf_vcpu_init(vcpu);
355 INIT_LIST_HEAD(&vcpu->blocked_vcpu_list);
357 kvm_vcpu_set_in_spin_loop(vcpu, false);
358 kvm_vcpu_set_dy_eligible(vcpu, false);
359 vcpu->preempted = false;
361 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
364 void kvm_vcpu_destroy(struct kvm_vcpu *vcpu)
366 kvm_arch_vcpu_destroy(vcpu);
369 * No need for rcu_read_lock as VCPU_RUN is the only place that changes
370 * the vcpu->pid pointer, and at destruction time all file descriptors
373 put_pid(rcu_dereference_protected(vcpu->pid, 1));
375 free_page((unsigned long)vcpu->run);
376 kmem_cache_free(kvm_vcpu_cache, vcpu);
378 EXPORT_SYMBOL_GPL(kvm_vcpu_destroy);
380 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
381 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
383 return container_of(mn, struct kvm, mmu_notifier);
386 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
387 struct mm_struct *mm,
388 unsigned long address,
391 struct kvm *kvm = mmu_notifier_to_kvm(mn);
394 idx = srcu_read_lock(&kvm->srcu);
395 spin_lock(&kvm->mmu_lock);
396 kvm->mmu_notifier_seq++;
398 if (kvm_set_spte_hva(kvm, address, pte))
399 kvm_flush_remote_tlbs(kvm);
401 spin_unlock(&kvm->mmu_lock);
402 srcu_read_unlock(&kvm->srcu, idx);
405 static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
406 const struct mmu_notifier_range *range)
408 struct kvm *kvm = mmu_notifier_to_kvm(mn);
409 int need_tlb_flush = 0, idx;
412 idx = srcu_read_lock(&kvm->srcu);
413 spin_lock(&kvm->mmu_lock);
415 * The count increase must become visible at unlock time as no
416 * spte can be established without taking the mmu_lock and
417 * count is also read inside the mmu_lock critical section.
419 kvm->mmu_notifier_count++;
420 need_tlb_flush = kvm_unmap_hva_range(kvm, range->start, range->end);
421 need_tlb_flush |= kvm->tlbs_dirty;
422 /* we've to flush the tlb before the pages can be freed */
424 kvm_flush_remote_tlbs(kvm);
426 spin_unlock(&kvm->mmu_lock);
428 ret = kvm_arch_mmu_notifier_invalidate_range(kvm, range->start,
430 mmu_notifier_range_blockable(range));
432 srcu_read_unlock(&kvm->srcu, idx);
437 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
438 const struct mmu_notifier_range *range)
440 struct kvm *kvm = mmu_notifier_to_kvm(mn);
442 spin_lock(&kvm->mmu_lock);
444 * This sequence increase will notify the kvm page fault that
445 * the page that is going to be mapped in the spte could have
448 kvm->mmu_notifier_seq++;
451 * The above sequence increase must be visible before the
452 * below count decrease, which is ensured by the smp_wmb above
453 * in conjunction with the smp_rmb in mmu_notifier_retry().
455 kvm->mmu_notifier_count--;
456 spin_unlock(&kvm->mmu_lock);
458 BUG_ON(kvm->mmu_notifier_count < 0);
461 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
462 struct mm_struct *mm,
466 struct kvm *kvm = mmu_notifier_to_kvm(mn);
469 idx = srcu_read_lock(&kvm->srcu);
470 spin_lock(&kvm->mmu_lock);
472 young = kvm_age_hva(kvm, start, end);
474 kvm_flush_remote_tlbs(kvm);
476 spin_unlock(&kvm->mmu_lock);
477 srcu_read_unlock(&kvm->srcu, idx);
482 static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
483 struct mm_struct *mm,
487 struct kvm *kvm = mmu_notifier_to_kvm(mn);
490 idx = srcu_read_lock(&kvm->srcu);
491 spin_lock(&kvm->mmu_lock);
493 * Even though we do not flush TLB, this will still adversely
494 * affect performance on pre-Haswell Intel EPT, where there is
495 * no EPT Access Bit to clear so that we have to tear down EPT
496 * tables instead. If we find this unacceptable, we can always
497 * add a parameter to kvm_age_hva so that it effectively doesn't
498 * do anything on clear_young.
500 * Also note that currently we never issue secondary TLB flushes
501 * from clear_young, leaving this job up to the regular system
502 * cadence. If we find this inaccurate, we might come up with a
503 * more sophisticated heuristic later.
505 young = kvm_age_hva(kvm, start, end);
506 spin_unlock(&kvm->mmu_lock);
507 srcu_read_unlock(&kvm->srcu, idx);
512 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
513 struct mm_struct *mm,
514 unsigned long address)
516 struct kvm *kvm = mmu_notifier_to_kvm(mn);
519 idx = srcu_read_lock(&kvm->srcu);
520 spin_lock(&kvm->mmu_lock);
521 young = kvm_test_age_hva(kvm, address);
522 spin_unlock(&kvm->mmu_lock);
523 srcu_read_unlock(&kvm->srcu, idx);
528 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
529 struct mm_struct *mm)
531 struct kvm *kvm = mmu_notifier_to_kvm(mn);
534 idx = srcu_read_lock(&kvm->srcu);
535 kvm_arch_flush_shadow_all(kvm);
536 srcu_read_unlock(&kvm->srcu, idx);
539 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
540 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
541 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
542 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
543 .clear_young = kvm_mmu_notifier_clear_young,
544 .test_young = kvm_mmu_notifier_test_young,
545 .change_pte = kvm_mmu_notifier_change_pte,
546 .release = kvm_mmu_notifier_release,
549 static int kvm_init_mmu_notifier(struct kvm *kvm)
551 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
552 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
555 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
557 static int kvm_init_mmu_notifier(struct kvm *kvm)
562 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
564 static struct kvm_memslots *kvm_alloc_memslots(void)
567 struct kvm_memslots *slots;
569 slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL_ACCOUNT);
573 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
574 slots->id_to_index[i] = -1;
579 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
581 if (!memslot->dirty_bitmap)
584 kvfree(memslot->dirty_bitmap);
585 memslot->dirty_bitmap = NULL;
588 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
590 kvm_destroy_dirty_bitmap(slot);
592 kvm_arch_free_memslot(kvm, slot);
598 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
600 struct kvm_memory_slot *memslot;
605 kvm_for_each_memslot(memslot, slots)
606 kvm_free_memslot(kvm, memslot);
611 static void kvm_destroy_vm_debugfs(struct kvm *kvm)
615 if (!kvm->debugfs_dentry)
618 debugfs_remove_recursive(kvm->debugfs_dentry);
620 if (kvm->debugfs_stat_data) {
621 for (i = 0; i < kvm_debugfs_num_entries; i++)
622 kfree(kvm->debugfs_stat_data[i]);
623 kfree(kvm->debugfs_stat_data);
627 static int kvm_create_vm_debugfs(struct kvm *kvm, int fd)
629 char dir_name[ITOA_MAX_LEN * 2];
630 struct kvm_stat_data *stat_data;
631 struct kvm_stats_debugfs_item *p;
633 if (!debugfs_initialized())
636 snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd);
637 kvm->debugfs_dentry = debugfs_create_dir(dir_name, kvm_debugfs_dir);
639 kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries,
640 sizeof(*kvm->debugfs_stat_data),
642 if (!kvm->debugfs_stat_data)
645 for (p = debugfs_entries; p->name; p++) {
646 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT);
650 stat_data->kvm = kvm;
651 stat_data->dbgfs_item = p;
652 kvm->debugfs_stat_data[p - debugfs_entries] = stat_data;
653 debugfs_create_file(p->name, KVM_DBGFS_GET_MODE(p),
654 kvm->debugfs_dentry, stat_data,
661 * Called after the VM is otherwise initialized, but just before adding it to
664 int __weak kvm_arch_post_init_vm(struct kvm *kvm)
670 * Called just after removing the VM from the vm_list, but before doing any
673 void __weak kvm_arch_pre_destroy_vm(struct kvm *kvm)
677 static struct kvm *kvm_create_vm(unsigned long type)
679 struct kvm *kvm = kvm_arch_alloc_vm();
684 return ERR_PTR(-ENOMEM);
686 spin_lock_init(&kvm->mmu_lock);
688 kvm->mm = current->mm;
689 kvm_eventfd_init(kvm);
690 mutex_init(&kvm->lock);
691 mutex_init(&kvm->irq_lock);
692 mutex_init(&kvm->slots_lock);
693 INIT_LIST_HEAD(&kvm->devices);
695 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
697 if (init_srcu_struct(&kvm->srcu))
698 goto out_err_no_srcu;
699 if (init_srcu_struct(&kvm->irq_srcu))
700 goto out_err_no_irq_srcu;
702 refcount_set(&kvm->users_count, 1);
703 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
704 struct kvm_memslots *slots = kvm_alloc_memslots();
707 goto out_err_no_arch_destroy_vm;
708 /* Generations must be different for each address space. */
709 slots->generation = i;
710 rcu_assign_pointer(kvm->memslots[i], slots);
713 for (i = 0; i < KVM_NR_BUSES; i++) {
714 rcu_assign_pointer(kvm->buses[i],
715 kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL_ACCOUNT));
717 goto out_err_no_arch_destroy_vm;
720 kvm->max_halt_poll_ns = halt_poll_ns;
722 r = kvm_arch_init_vm(kvm, type);
724 goto out_err_no_arch_destroy_vm;
726 r = hardware_enable_all();
728 goto out_err_no_disable;
730 #ifdef CONFIG_HAVE_KVM_IRQFD
731 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
734 r = kvm_init_mmu_notifier(kvm);
736 goto out_err_no_mmu_notifier;
738 r = kvm_arch_post_init_vm(kvm);
742 mutex_lock(&kvm_lock);
743 list_add(&kvm->vm_list, &vm_list);
744 mutex_unlock(&kvm_lock);
746 preempt_notifier_inc();
751 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
752 if (kvm->mmu_notifier.ops)
753 mmu_notifier_unregister(&kvm->mmu_notifier, current->mm);
755 out_err_no_mmu_notifier:
756 hardware_disable_all();
758 kvm_arch_destroy_vm(kvm);
759 out_err_no_arch_destroy_vm:
760 WARN_ON_ONCE(!refcount_dec_and_test(&kvm->users_count));
761 for (i = 0; i < KVM_NR_BUSES; i++)
762 kfree(kvm_get_bus(kvm, i));
763 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
764 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
765 cleanup_srcu_struct(&kvm->irq_srcu);
767 cleanup_srcu_struct(&kvm->srcu);
769 kvm_arch_free_vm(kvm);
774 static void kvm_destroy_devices(struct kvm *kvm)
776 struct kvm_device *dev, *tmp;
779 * We do not need to take the kvm->lock here, because nobody else
780 * has a reference to the struct kvm at this point and therefore
781 * cannot access the devices list anyhow.
783 list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
784 list_del(&dev->vm_node);
785 dev->ops->destroy(dev);
789 static void kvm_destroy_vm(struct kvm *kvm)
792 struct mm_struct *mm = kvm->mm;
794 kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm);
795 kvm_destroy_vm_debugfs(kvm);
796 kvm_arch_sync_events(kvm);
797 mutex_lock(&kvm_lock);
798 list_del(&kvm->vm_list);
799 mutex_unlock(&kvm_lock);
800 kvm_arch_pre_destroy_vm(kvm);
802 kvm_free_irq_routing(kvm);
803 for (i = 0; i < KVM_NR_BUSES; i++) {
804 struct kvm_io_bus *bus = kvm_get_bus(kvm, i);
807 kvm_io_bus_destroy(bus);
808 kvm->buses[i] = NULL;
810 kvm_coalesced_mmio_free(kvm);
811 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
812 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
814 kvm_arch_flush_shadow_all(kvm);
816 kvm_arch_destroy_vm(kvm);
817 kvm_destroy_devices(kvm);
818 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
819 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
820 cleanup_srcu_struct(&kvm->irq_srcu);
821 cleanup_srcu_struct(&kvm->srcu);
822 kvm_arch_free_vm(kvm);
823 preempt_notifier_dec();
824 hardware_disable_all();
828 void kvm_get_kvm(struct kvm *kvm)
830 refcount_inc(&kvm->users_count);
832 EXPORT_SYMBOL_GPL(kvm_get_kvm);
834 void kvm_put_kvm(struct kvm *kvm)
836 if (refcount_dec_and_test(&kvm->users_count))
839 EXPORT_SYMBOL_GPL(kvm_put_kvm);
842 * Used to put a reference that was taken on behalf of an object associated
843 * with a user-visible file descriptor, e.g. a vcpu or device, if installation
844 * of the new file descriptor fails and the reference cannot be transferred to
845 * its final owner. In such cases, the caller is still actively using @kvm and
846 * will fail miserably if the refcount unexpectedly hits zero.
848 void kvm_put_kvm_no_destroy(struct kvm *kvm)
850 WARN_ON(refcount_dec_and_test(&kvm->users_count));
852 EXPORT_SYMBOL_GPL(kvm_put_kvm_no_destroy);
854 static int kvm_vm_release(struct inode *inode, struct file *filp)
856 struct kvm *kvm = filp->private_data;
858 kvm_irqfd_release(kvm);
865 * Allocation size is twice as large as the actual dirty bitmap size.
866 * See kvm_vm_ioctl_get_dirty_log() why this is needed.
868 static int kvm_alloc_dirty_bitmap(struct kvm_memory_slot *memslot)
870 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
872 memslot->dirty_bitmap = kvzalloc(dirty_bytes, GFP_KERNEL_ACCOUNT);
873 if (!memslot->dirty_bitmap)
880 * Delete a memslot by decrementing the number of used slots and shifting all
881 * other entries in the array forward one spot.
883 static inline void kvm_memslot_delete(struct kvm_memslots *slots,
884 struct kvm_memory_slot *memslot)
886 struct kvm_memory_slot *mslots = slots->memslots;
889 if (WARN_ON(slots->id_to_index[memslot->id] == -1))
894 if (atomic_read(&slots->lru_slot) >= slots->used_slots)
895 atomic_set(&slots->lru_slot, 0);
897 for (i = slots->id_to_index[memslot->id]; i < slots->used_slots; i++) {
898 mslots[i] = mslots[i + 1];
899 slots->id_to_index[mslots[i].id] = i;
901 mslots[i] = *memslot;
902 slots->id_to_index[memslot->id] = -1;
906 * "Insert" a new memslot by incrementing the number of used slots. Returns
907 * the new slot's initial index into the memslots array.
909 static inline int kvm_memslot_insert_back(struct kvm_memslots *slots)
911 return slots->used_slots++;
915 * Move a changed memslot backwards in the array by shifting existing slots
916 * with a higher GFN toward the front of the array. Note, the changed memslot
917 * itself is not preserved in the array, i.e. not swapped at this time, only
918 * its new index into the array is tracked. Returns the changed memslot's
919 * current index into the memslots array.
921 static inline int kvm_memslot_move_backward(struct kvm_memslots *slots,
922 struct kvm_memory_slot *memslot)
924 struct kvm_memory_slot *mslots = slots->memslots;
927 if (WARN_ON_ONCE(slots->id_to_index[memslot->id] == -1) ||
928 WARN_ON_ONCE(!slots->used_slots))
932 * Move the target memslot backward in the array by shifting existing
933 * memslots with a higher GFN (than the target memslot) towards the
934 * front of the array.
936 for (i = slots->id_to_index[memslot->id]; i < slots->used_slots - 1; i++) {
937 if (memslot->base_gfn > mslots[i + 1].base_gfn)
940 WARN_ON_ONCE(memslot->base_gfn == mslots[i + 1].base_gfn);
942 /* Shift the next memslot forward one and update its index. */
943 mslots[i] = mslots[i + 1];
944 slots->id_to_index[mslots[i].id] = i;
950 * Move a changed memslot forwards in the array by shifting existing slots with
951 * a lower GFN toward the back of the array. Note, the changed memslot itself
952 * is not preserved in the array, i.e. not swapped at this time, only its new
953 * index into the array is tracked. Returns the changed memslot's final index
954 * into the memslots array.
956 static inline int kvm_memslot_move_forward(struct kvm_memslots *slots,
957 struct kvm_memory_slot *memslot,
960 struct kvm_memory_slot *mslots = slots->memslots;
963 for (i = start; i > 0; i--) {
964 if (memslot->base_gfn < mslots[i - 1].base_gfn)
967 WARN_ON_ONCE(memslot->base_gfn == mslots[i - 1].base_gfn);
969 /* Shift the next memslot back one and update its index. */
970 mslots[i] = mslots[i - 1];
971 slots->id_to_index[mslots[i].id] = i;
977 * Re-sort memslots based on their GFN to account for an added, deleted, or
978 * moved memslot. Sorting memslots by GFN allows using a binary search during
981 * IMPORTANT: Slots are sorted from highest GFN to lowest GFN! I.e. the entry
982 * at memslots[0] has the highest GFN.
984 * The sorting algorithm takes advantage of having initially sorted memslots
985 * and knowing the position of the changed memslot. Sorting is also optimized
986 * by not swapping the updated memslot and instead only shifting other memslots
987 * and tracking the new index for the update memslot. Only once its final
988 * index is known is the updated memslot copied into its position in the array.
990 * - When deleting a memslot, the deleted memslot simply needs to be moved to
991 * the end of the array.
993 * - When creating a memslot, the algorithm "inserts" the new memslot at the
994 * end of the array and then it forward to its correct location.
996 * - When moving a memslot, the algorithm first moves the updated memslot
997 * backward to handle the scenario where the memslot's GFN was changed to a
998 * lower value. update_memslots() then falls through and runs the same flow
999 * as creating a memslot to move the memslot forward to handle the scenario
1000 * where its GFN was changed to a higher value.
1002 * Note, slots are sorted from highest->lowest instead of lowest->highest for
1003 * historical reasons. Originally, invalid memslots where denoted by having
1004 * GFN=0, thus sorting from highest->lowest naturally sorted invalid memslots
1005 * to the end of the array. The current algorithm uses dedicated logic to
1006 * delete a memslot and thus does not rely on invalid memslots having GFN=0.
1008 * The other historical motiviation for highest->lowest was to improve the
1009 * performance of memslot lookup. KVM originally used a linear search starting
1010 * at memslots[0]. On x86, the largest memslot usually has one of the highest,
1011 * if not *the* highest, GFN, as the bulk of the guest's RAM is located in a
1012 * single memslot above the 4gb boundary. As the largest memslot is also the
1013 * most likely to be referenced, sorting it to the front of the array was
1014 * advantageous. The current binary search starts from the middle of the array
1015 * and uses an LRU pointer to improve performance for all memslots and GFNs.
1017 static void update_memslots(struct kvm_memslots *slots,
1018 struct kvm_memory_slot *memslot,
1019 enum kvm_mr_change change)
1023 if (change == KVM_MR_DELETE) {
1024 kvm_memslot_delete(slots, memslot);
1026 if (change == KVM_MR_CREATE)
1027 i = kvm_memslot_insert_back(slots);
1029 i = kvm_memslot_move_backward(slots, memslot);
1030 i = kvm_memslot_move_forward(slots, memslot, i);
1033 * Copy the memslot to its new position in memslots and update
1034 * its index accordingly.
1036 slots->memslots[i] = *memslot;
1037 slots->id_to_index[memslot->id] = i;
1041 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
1043 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
1045 #ifdef __KVM_HAVE_READONLY_MEM
1046 valid_flags |= KVM_MEM_READONLY;
1049 if (mem->flags & ~valid_flags)
1055 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
1056 int as_id, struct kvm_memslots *slots)
1058 struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id);
1059 u64 gen = old_memslots->generation;
1061 WARN_ON(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS);
1062 slots->generation = gen | KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
1064 rcu_assign_pointer(kvm->memslots[as_id], slots);
1065 synchronize_srcu_expedited(&kvm->srcu);
1068 * Increment the new memslot generation a second time, dropping the
1069 * update in-progress flag and incrementing the generation based on
1070 * the number of address spaces. This provides a unique and easily
1071 * identifiable generation number while the memslots are in flux.
1073 gen = slots->generation & ~KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
1076 * Generations must be unique even across address spaces. We do not need
1077 * a global counter for that, instead the generation space is evenly split
1078 * across address spaces. For example, with two address spaces, address
1079 * space 0 will use generations 0, 2, 4, ... while address space 1 will
1080 * use generations 1, 3, 5, ...
1082 gen += KVM_ADDRESS_SPACE_NUM;
1084 kvm_arch_memslots_updated(kvm, gen);
1086 slots->generation = gen;
1088 return old_memslots;
1092 * Note, at a minimum, the current number of used slots must be allocated, even
1093 * when deleting a memslot, as we need a complete duplicate of the memslots for
1094 * use when invalidating a memslot prior to deleting/moving the memslot.
1096 static struct kvm_memslots *kvm_dup_memslots(struct kvm_memslots *old,
1097 enum kvm_mr_change change)
1099 struct kvm_memslots *slots;
1100 size_t old_size, new_size;
1102 old_size = sizeof(struct kvm_memslots) +
1103 (sizeof(struct kvm_memory_slot) * old->used_slots);
1105 if (change == KVM_MR_CREATE)
1106 new_size = old_size + sizeof(struct kvm_memory_slot);
1108 new_size = old_size;
1110 slots = kvzalloc(new_size, GFP_KERNEL_ACCOUNT);
1112 memcpy(slots, old, old_size);
1117 static int kvm_set_memslot(struct kvm *kvm,
1118 const struct kvm_userspace_memory_region *mem,
1119 struct kvm_memory_slot *old,
1120 struct kvm_memory_slot *new, int as_id,
1121 enum kvm_mr_change change)
1123 struct kvm_memory_slot *slot;
1124 struct kvm_memslots *slots;
1127 slots = kvm_dup_memslots(__kvm_memslots(kvm, as_id), change);
1131 if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) {
1133 * Note, the INVALID flag needs to be in the appropriate entry
1134 * in the freshly allocated memslots, not in @old or @new.
1136 slot = id_to_memslot(slots, old->id);
1137 slot->flags |= KVM_MEMSLOT_INVALID;
1140 * We can re-use the old memslots, the only difference from the
1141 * newly installed memslots is the invalid flag, which will get
1142 * dropped by update_memslots anyway. We'll also revert to the
1143 * old memslots if preparing the new memory region fails.
1145 slots = install_new_memslots(kvm, as_id, slots);
1147 /* From this point no new shadow pages pointing to a deleted,
1148 * or moved, memslot will be created.
1150 * validation of sp->gfn happens in:
1151 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1152 * - kvm_is_visible_gfn (mmu_check_root)
1154 kvm_arch_flush_shadow_memslot(kvm, slot);
1157 r = kvm_arch_prepare_memory_region(kvm, new, mem, change);
1161 update_memslots(slots, new, change);
1162 slots = install_new_memslots(kvm, as_id, slots);
1164 kvm_arch_commit_memory_region(kvm, mem, old, new, change);
1170 if (change == KVM_MR_DELETE || change == KVM_MR_MOVE)
1171 slots = install_new_memslots(kvm, as_id, slots);
1176 static int kvm_delete_memslot(struct kvm *kvm,
1177 const struct kvm_userspace_memory_region *mem,
1178 struct kvm_memory_slot *old, int as_id)
1180 struct kvm_memory_slot new;
1186 memset(&new, 0, sizeof(new));
1189 r = kvm_set_memslot(kvm, mem, old, &new, as_id, KVM_MR_DELETE);
1193 kvm_free_memslot(kvm, old);
1198 * Allocate some memory and give it an address in the guest physical address
1201 * Discontiguous memory is allowed, mostly for framebuffers.
1203 * Must be called holding kvm->slots_lock for write.
1205 int __kvm_set_memory_region(struct kvm *kvm,
1206 const struct kvm_userspace_memory_region *mem)
1208 struct kvm_memory_slot old, new;
1209 struct kvm_memory_slot *tmp;
1210 enum kvm_mr_change change;
1214 r = check_memory_region_flags(mem);
1218 as_id = mem->slot >> 16;
1219 id = (u16)mem->slot;
1221 /* General sanity checks */
1222 if (mem->memory_size & (PAGE_SIZE - 1))
1224 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
1226 /* We can read the guest memory with __xxx_user() later on. */
1227 if ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
1228 !access_ok((void __user *)(unsigned long)mem->userspace_addr,
1231 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
1233 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
1237 * Make a full copy of the old memslot, the pointer will become stale
1238 * when the memslots are re-sorted by update_memslots(), and the old
1239 * memslot needs to be referenced after calling update_memslots(), e.g.
1240 * to free its resources and for arch specific behavior.
1242 tmp = id_to_memslot(__kvm_memslots(kvm, as_id), id);
1247 memset(&old, 0, sizeof(old));
1251 if (!mem->memory_size)
1252 return kvm_delete_memslot(kvm, mem, &old, as_id);
1255 new.base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
1256 new.npages = mem->memory_size >> PAGE_SHIFT;
1257 new.flags = mem->flags;
1258 new.userspace_addr = mem->userspace_addr;
1260 if (new.npages > KVM_MEM_MAX_NR_PAGES)
1264 change = KVM_MR_CREATE;
1265 new.dirty_bitmap = NULL;
1266 memset(&new.arch, 0, sizeof(new.arch));
1267 } else { /* Modify an existing slot. */
1268 if ((new.userspace_addr != old.userspace_addr) ||
1269 (new.npages != old.npages) ||
1270 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
1273 if (new.base_gfn != old.base_gfn)
1274 change = KVM_MR_MOVE;
1275 else if (new.flags != old.flags)
1276 change = KVM_MR_FLAGS_ONLY;
1277 else /* Nothing to change. */
1280 /* Copy dirty_bitmap and arch from the current memslot. */
1281 new.dirty_bitmap = old.dirty_bitmap;
1282 memcpy(&new.arch, &old.arch, sizeof(new.arch));
1285 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
1286 /* Check for overlaps */
1287 kvm_for_each_memslot(tmp, __kvm_memslots(kvm, as_id)) {
1290 if (!((new.base_gfn + new.npages <= tmp->base_gfn) ||
1291 (new.base_gfn >= tmp->base_gfn + tmp->npages)))
1296 /* Allocate/free page dirty bitmap as needed */
1297 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
1298 new.dirty_bitmap = NULL;
1299 else if (!new.dirty_bitmap) {
1300 r = kvm_alloc_dirty_bitmap(&new);
1304 if (kvm_dirty_log_manual_protect_and_init_set(kvm))
1305 bitmap_set(new.dirty_bitmap, 0, new.npages);
1308 r = kvm_set_memslot(kvm, mem, &old, &new, as_id, change);
1312 if (old.dirty_bitmap && !new.dirty_bitmap)
1313 kvm_destroy_dirty_bitmap(&old);
1317 if (new.dirty_bitmap && !old.dirty_bitmap)
1318 kvm_destroy_dirty_bitmap(&new);
1321 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1323 int kvm_set_memory_region(struct kvm *kvm,
1324 const struct kvm_userspace_memory_region *mem)
1328 mutex_lock(&kvm->slots_lock);
1329 r = __kvm_set_memory_region(kvm, mem);
1330 mutex_unlock(&kvm->slots_lock);
1333 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1335 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1336 struct kvm_userspace_memory_region *mem)
1338 if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1341 return kvm_set_memory_region(kvm, mem);
1344 #ifndef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1346 * kvm_get_dirty_log - get a snapshot of dirty pages
1347 * @kvm: pointer to kvm instance
1348 * @log: slot id and address to which we copy the log
1349 * @is_dirty: set to '1' if any dirty pages were found
1350 * @memslot: set to the associated memslot, always valid on success
1352 int kvm_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log,
1353 int *is_dirty, struct kvm_memory_slot **memslot)
1355 struct kvm_memslots *slots;
1358 unsigned long any = 0;
1363 as_id = log->slot >> 16;
1364 id = (u16)log->slot;
1365 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1368 slots = __kvm_memslots(kvm, as_id);
1369 *memslot = id_to_memslot(slots, id);
1370 if (!(*memslot) || !(*memslot)->dirty_bitmap)
1373 kvm_arch_sync_dirty_log(kvm, *memslot);
1375 n = kvm_dirty_bitmap_bytes(*memslot);
1377 for (i = 0; !any && i < n/sizeof(long); ++i)
1378 any = (*memslot)->dirty_bitmap[i];
1380 if (copy_to_user(log->dirty_bitmap, (*memslot)->dirty_bitmap, n))
1387 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1389 #else /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
1391 * kvm_get_dirty_log_protect - get a snapshot of dirty pages
1392 * and reenable dirty page tracking for the corresponding pages.
1393 * @kvm: pointer to kvm instance
1394 * @log: slot id and address to which we copy the log
1396 * We need to keep it in mind that VCPU threads can write to the bitmap
1397 * concurrently. So, to avoid losing track of dirty pages we keep the
1400 * 1. Take a snapshot of the bit and clear it if needed.
1401 * 2. Write protect the corresponding page.
1402 * 3. Copy the snapshot to the userspace.
1403 * 4. Upon return caller flushes TLB's if needed.
1405 * Between 2 and 4, the guest may write to the page using the remaining TLB
1406 * entry. This is not a problem because the page is reported dirty using
1407 * the snapshot taken before and step 4 ensures that writes done after
1408 * exiting to userspace will be logged for the next call.
1411 static int kvm_get_dirty_log_protect(struct kvm *kvm, struct kvm_dirty_log *log)
1413 struct kvm_memslots *slots;
1414 struct kvm_memory_slot *memslot;
1417 unsigned long *dirty_bitmap;
1418 unsigned long *dirty_bitmap_buffer;
1421 as_id = log->slot >> 16;
1422 id = (u16)log->slot;
1423 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1426 slots = __kvm_memslots(kvm, as_id);
1427 memslot = id_to_memslot(slots, id);
1428 if (!memslot || !memslot->dirty_bitmap)
1431 dirty_bitmap = memslot->dirty_bitmap;
1433 kvm_arch_sync_dirty_log(kvm, memslot);
1435 n = kvm_dirty_bitmap_bytes(memslot);
1437 if (kvm->manual_dirty_log_protect) {
1439 * Unlike kvm_get_dirty_log, we always return false in *flush,
1440 * because no flush is needed until KVM_CLEAR_DIRTY_LOG. There
1441 * is some code duplication between this function and
1442 * kvm_get_dirty_log, but hopefully all architecture
1443 * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log
1444 * can be eliminated.
1446 dirty_bitmap_buffer = dirty_bitmap;
1448 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
1449 memset(dirty_bitmap_buffer, 0, n);
1451 spin_lock(&kvm->mmu_lock);
1452 for (i = 0; i < n / sizeof(long); i++) {
1456 if (!dirty_bitmap[i])
1460 mask = xchg(&dirty_bitmap[i], 0);
1461 dirty_bitmap_buffer[i] = mask;
1463 offset = i * BITS_PER_LONG;
1464 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1467 spin_unlock(&kvm->mmu_lock);
1471 kvm_arch_flush_remote_tlbs_memslot(kvm, memslot);
1473 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1480 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
1481 * @kvm: kvm instance
1482 * @log: slot id and address to which we copy the log
1484 * Steps 1-4 below provide general overview of dirty page logging. See
1485 * kvm_get_dirty_log_protect() function description for additional details.
1487 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
1488 * always flush the TLB (step 4) even if previous step failed and the dirty
1489 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
1490 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
1491 * writes will be marked dirty for next log read.
1493 * 1. Take a snapshot of the bit and clear it if needed.
1494 * 2. Write protect the corresponding page.
1495 * 3. Copy the snapshot to the userspace.
1496 * 4. Flush TLB's if needed.
1498 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1499 struct kvm_dirty_log *log)
1503 mutex_lock(&kvm->slots_lock);
1505 r = kvm_get_dirty_log_protect(kvm, log);
1507 mutex_unlock(&kvm->slots_lock);
1512 * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap
1513 * and reenable dirty page tracking for the corresponding pages.
1514 * @kvm: pointer to kvm instance
1515 * @log: slot id and address from which to fetch the bitmap of dirty pages
1517 static int kvm_clear_dirty_log_protect(struct kvm *kvm,
1518 struct kvm_clear_dirty_log *log)
1520 struct kvm_memslots *slots;
1521 struct kvm_memory_slot *memslot;
1525 unsigned long *dirty_bitmap;
1526 unsigned long *dirty_bitmap_buffer;
1529 as_id = log->slot >> 16;
1530 id = (u16)log->slot;
1531 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1534 if (log->first_page & 63)
1537 slots = __kvm_memslots(kvm, as_id);
1538 memslot = id_to_memslot(slots, id);
1539 if (!memslot || !memslot->dirty_bitmap)
1542 dirty_bitmap = memslot->dirty_bitmap;
1544 n = ALIGN(log->num_pages, BITS_PER_LONG) / 8;
1546 if (log->first_page > memslot->npages ||
1547 log->num_pages > memslot->npages - log->first_page ||
1548 (log->num_pages < memslot->npages - log->first_page && (log->num_pages & 63)))
1551 kvm_arch_sync_dirty_log(kvm, memslot);
1554 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
1555 if (copy_from_user(dirty_bitmap_buffer, log->dirty_bitmap, n))
1558 spin_lock(&kvm->mmu_lock);
1559 for (offset = log->first_page, i = offset / BITS_PER_LONG,
1560 n = DIV_ROUND_UP(log->num_pages, BITS_PER_LONG); n--;
1561 i++, offset += BITS_PER_LONG) {
1562 unsigned long mask = *dirty_bitmap_buffer++;
1563 atomic_long_t *p = (atomic_long_t *) &dirty_bitmap[i];
1567 mask &= atomic_long_fetch_andnot(mask, p);
1570 * mask contains the bits that really have been cleared. This
1571 * never includes any bits beyond the length of the memslot (if
1572 * the length is not aligned to 64 pages), therefore it is not
1573 * a problem if userspace sets them in log->dirty_bitmap.
1577 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1581 spin_unlock(&kvm->mmu_lock);
1584 kvm_arch_flush_remote_tlbs_memslot(kvm, memslot);
1589 static int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm,
1590 struct kvm_clear_dirty_log *log)
1594 mutex_lock(&kvm->slots_lock);
1596 r = kvm_clear_dirty_log_protect(kvm, log);
1598 mutex_unlock(&kvm->slots_lock);
1601 #endif /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
1603 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1605 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1607 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1609 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
1611 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn);
1613 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_memslot);
1615 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1617 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1619 return kvm_is_visible_memslot(memslot);
1621 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1623 unsigned long kvm_host_page_size(struct kvm_vcpu *vcpu, gfn_t gfn)
1625 struct vm_area_struct *vma;
1626 unsigned long addr, size;
1630 addr = kvm_vcpu_gfn_to_hva_prot(vcpu, gfn, NULL);
1631 if (kvm_is_error_hva(addr))
1634 mmap_read_lock(current->mm);
1635 vma = find_vma(current->mm, addr);
1639 size = vma_kernel_pagesize(vma);
1642 mmap_read_unlock(current->mm);
1647 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1649 return slot->flags & KVM_MEM_READONLY;
1652 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1653 gfn_t *nr_pages, bool write)
1655 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1656 return KVM_HVA_ERR_BAD;
1658 if (memslot_is_readonly(slot) && write)
1659 return KVM_HVA_ERR_RO_BAD;
1662 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1664 return __gfn_to_hva_memslot(slot, gfn);
1667 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1670 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1673 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1676 return gfn_to_hva_many(slot, gfn, NULL);
1678 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1680 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1682 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1684 EXPORT_SYMBOL_GPL(gfn_to_hva);
1686 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
1688 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
1690 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
1693 * Return the hva of a @gfn and the R/W attribute if possible.
1695 * @slot: the kvm_memory_slot which contains @gfn
1696 * @gfn: the gfn to be translated
1697 * @writable: used to return the read/write attribute of the @slot if the hva
1698 * is valid and @writable is not NULL
1700 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1701 gfn_t gfn, bool *writable)
1703 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1705 if (!kvm_is_error_hva(hva) && writable)
1706 *writable = !memslot_is_readonly(slot);
1711 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1713 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1715 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1718 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
1720 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1722 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1725 static inline int check_user_page_hwpoison(unsigned long addr)
1727 int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
1729 rc = get_user_pages(addr, 1, flags, NULL, NULL);
1730 return rc == -EHWPOISON;
1734 * The fast path to get the writable pfn which will be stored in @pfn,
1735 * true indicates success, otherwise false is returned. It's also the
1736 * only part that runs if we can in atomic context.
1738 static bool hva_to_pfn_fast(unsigned long addr, bool write_fault,
1739 bool *writable, kvm_pfn_t *pfn)
1741 struct page *page[1];
1744 * Fast pin a writable pfn only if it is a write fault request
1745 * or the caller allows to map a writable pfn for a read fault
1748 if (!(write_fault || writable))
1751 if (get_user_page_fast_only(addr, FOLL_WRITE, page)) {
1752 *pfn = page_to_pfn(page[0]);
1763 * The slow path to get the pfn of the specified host virtual address,
1764 * 1 indicates success, -errno is returned if error is detected.
1766 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1767 bool *writable, kvm_pfn_t *pfn)
1769 unsigned int flags = FOLL_HWPOISON;
1776 *writable = write_fault;
1779 flags |= FOLL_WRITE;
1781 flags |= FOLL_NOWAIT;
1783 npages = get_user_pages_unlocked(addr, 1, &page, flags);
1787 /* map read fault as writable if possible */
1788 if (unlikely(!write_fault) && writable) {
1791 if (get_user_page_fast_only(addr, FOLL_WRITE, &wpage)) {
1797 *pfn = page_to_pfn(page);
1801 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1803 if (unlikely(!(vma->vm_flags & VM_READ)))
1806 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1812 static int hva_to_pfn_remapped(struct vm_area_struct *vma,
1813 unsigned long addr, bool *async,
1814 bool write_fault, bool *writable,
1820 r = follow_pfn(vma, addr, &pfn);
1823 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
1824 * not call the fault handler, so do it here.
1826 bool unlocked = false;
1827 r = fixup_user_fault(current, current->mm, addr,
1828 (write_fault ? FAULT_FLAG_WRITE : 0),
1835 r = follow_pfn(vma, addr, &pfn);
1845 * Get a reference here because callers of *hva_to_pfn* and
1846 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
1847 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
1848 * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will
1849 * simply do nothing for reserved pfns.
1851 * Whoever called remap_pfn_range is also going to call e.g.
1852 * unmap_mapping_range before the underlying pages are freed,
1853 * causing a call to our MMU notifier.
1862 * Pin guest page in memory and return its pfn.
1863 * @addr: host virtual address which maps memory to the guest
1864 * @atomic: whether this function can sleep
1865 * @async: whether this function need to wait IO complete if the
1866 * host page is not in the memory
1867 * @write_fault: whether we should get a writable host page
1868 * @writable: whether it allows to map a writable host page for !@write_fault
1870 * The function will map a writable host page for these two cases:
1871 * 1): @write_fault = true
1872 * 2): @write_fault = false && @writable, @writable will tell the caller
1873 * whether the mapping is writable.
1875 static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1876 bool write_fault, bool *writable)
1878 struct vm_area_struct *vma;
1882 /* we can do it either atomically or asynchronously, not both */
1883 BUG_ON(atomic && async);
1885 if (hva_to_pfn_fast(addr, write_fault, writable, &pfn))
1889 return KVM_PFN_ERR_FAULT;
1891 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1895 mmap_read_lock(current->mm);
1896 if (npages == -EHWPOISON ||
1897 (!async && check_user_page_hwpoison(addr))) {
1898 pfn = KVM_PFN_ERR_HWPOISON;
1903 vma = find_vma_intersection(current->mm, addr, addr + 1);
1906 pfn = KVM_PFN_ERR_FAULT;
1907 else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
1908 r = hva_to_pfn_remapped(vma, addr, async, write_fault, writable, &pfn);
1912 pfn = KVM_PFN_ERR_FAULT;
1914 if (async && vma_is_valid(vma, write_fault))
1916 pfn = KVM_PFN_ERR_FAULT;
1919 mmap_read_unlock(current->mm);
1923 kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn,
1924 bool atomic, bool *async, bool write_fault,
1927 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1929 if (addr == KVM_HVA_ERR_RO_BAD) {
1932 return KVM_PFN_ERR_RO_FAULT;
1935 if (kvm_is_error_hva(addr)) {
1938 return KVM_PFN_NOSLOT;
1941 /* Do not map writable pfn in the readonly memslot. */
1942 if (writable && memslot_is_readonly(slot)) {
1947 return hva_to_pfn(addr, atomic, async, write_fault,
1950 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
1952 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1955 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
1956 write_fault, writable);
1958 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1960 kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1962 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1964 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
1966 kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1968 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1970 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1972 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
1974 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1976 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
1978 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1980 return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
1982 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1984 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
1986 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1988 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
1990 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1991 struct page **pages, int nr_pages)
1996 addr = gfn_to_hva_many(slot, gfn, &entry);
1997 if (kvm_is_error_hva(addr))
2000 if (entry < nr_pages)
2003 return get_user_pages_fast_only(addr, nr_pages, FOLL_WRITE, pages);
2005 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
2007 static struct page *kvm_pfn_to_page(kvm_pfn_t pfn)
2009 if (is_error_noslot_pfn(pfn))
2010 return KVM_ERR_PTR_BAD_PAGE;
2012 if (kvm_is_reserved_pfn(pfn)) {
2014 return KVM_ERR_PTR_BAD_PAGE;
2017 return pfn_to_page(pfn);
2020 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
2024 pfn = gfn_to_pfn(kvm, gfn);
2026 return kvm_pfn_to_page(pfn);
2028 EXPORT_SYMBOL_GPL(gfn_to_page);
2030 void kvm_release_pfn(kvm_pfn_t pfn, bool dirty, struct gfn_to_pfn_cache *cache)
2036 cache->pfn = cache->gfn = 0;
2039 kvm_release_pfn_dirty(pfn);
2041 kvm_release_pfn_clean(pfn);
2044 static void kvm_cache_gfn_to_pfn(struct kvm_memory_slot *slot, gfn_t gfn,
2045 struct gfn_to_pfn_cache *cache, u64 gen)
2047 kvm_release_pfn(cache->pfn, cache->dirty, cache);
2049 cache->pfn = gfn_to_pfn_memslot(slot, gfn);
2051 cache->dirty = false;
2052 cache->generation = gen;
2055 static int __kvm_map_gfn(struct kvm_memslots *slots, gfn_t gfn,
2056 struct kvm_host_map *map,
2057 struct gfn_to_pfn_cache *cache,
2062 struct page *page = KVM_UNMAPPED_PAGE;
2063 struct kvm_memory_slot *slot = __gfn_to_memslot(slots, gfn);
2064 u64 gen = slots->generation;
2070 if (!cache->pfn || cache->gfn != gfn ||
2071 cache->generation != gen) {
2074 kvm_cache_gfn_to_pfn(slot, gfn, cache, gen);
2080 pfn = gfn_to_pfn_memslot(slot, gfn);
2082 if (is_error_noslot_pfn(pfn))
2085 if (pfn_valid(pfn)) {
2086 page = pfn_to_page(pfn);
2088 hva = kmap_atomic(page);
2091 #ifdef CONFIG_HAS_IOMEM
2092 } else if (!atomic) {
2093 hva = memremap(pfn_to_hpa(pfn), PAGE_SIZE, MEMREMAP_WB);
2110 int kvm_map_gfn(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map,
2111 struct gfn_to_pfn_cache *cache, bool atomic)
2113 return __kvm_map_gfn(kvm_memslots(vcpu->kvm), gfn, map,
2116 EXPORT_SYMBOL_GPL(kvm_map_gfn);
2118 int kvm_vcpu_map(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map)
2120 return __kvm_map_gfn(kvm_vcpu_memslots(vcpu), gfn, map,
2123 EXPORT_SYMBOL_GPL(kvm_vcpu_map);
2125 static void __kvm_unmap_gfn(struct kvm_memory_slot *memslot,
2126 struct kvm_host_map *map,
2127 struct gfn_to_pfn_cache *cache,
2128 bool dirty, bool atomic)
2136 if (map->page != KVM_UNMAPPED_PAGE) {
2138 kunmap_atomic(map->hva);
2142 #ifdef CONFIG_HAS_IOMEM
2146 WARN_ONCE(1, "Unexpected unmapping in atomic context");
2150 mark_page_dirty_in_slot(memslot, map->gfn);
2153 cache->dirty |= dirty;
2155 kvm_release_pfn(map->pfn, dirty, NULL);
2161 int kvm_unmap_gfn(struct kvm_vcpu *vcpu, struct kvm_host_map *map,
2162 struct gfn_to_pfn_cache *cache, bool dirty, bool atomic)
2164 __kvm_unmap_gfn(gfn_to_memslot(vcpu->kvm, map->gfn), map,
2165 cache, dirty, atomic);
2168 EXPORT_SYMBOL_GPL(kvm_unmap_gfn);
2170 void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map, bool dirty)
2172 __kvm_unmap_gfn(kvm_vcpu_gfn_to_memslot(vcpu, map->gfn), map, NULL,
2175 EXPORT_SYMBOL_GPL(kvm_vcpu_unmap);
2177 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
2181 pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
2183 return kvm_pfn_to_page(pfn);
2185 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
2187 void kvm_release_page_clean(struct page *page)
2189 WARN_ON(is_error_page(page));
2191 kvm_release_pfn_clean(page_to_pfn(page));
2193 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
2195 void kvm_release_pfn_clean(kvm_pfn_t pfn)
2197 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
2198 put_page(pfn_to_page(pfn));
2200 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
2202 void kvm_release_page_dirty(struct page *page)
2204 WARN_ON(is_error_page(page));
2206 kvm_release_pfn_dirty(page_to_pfn(page));
2208 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
2210 void kvm_release_pfn_dirty(kvm_pfn_t pfn)
2212 kvm_set_pfn_dirty(pfn);
2213 kvm_release_pfn_clean(pfn);
2215 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
2217 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
2219 if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn))
2220 SetPageDirty(pfn_to_page(pfn));
2222 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
2224 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
2226 if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn))
2227 mark_page_accessed(pfn_to_page(pfn));
2229 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
2231 void kvm_get_pfn(kvm_pfn_t pfn)
2233 if (!kvm_is_reserved_pfn(pfn))
2234 get_page(pfn_to_page(pfn));
2236 EXPORT_SYMBOL_GPL(kvm_get_pfn);
2238 static int next_segment(unsigned long len, int offset)
2240 if (len > PAGE_SIZE - offset)
2241 return PAGE_SIZE - offset;
2246 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
2247 void *data, int offset, int len)
2252 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
2253 if (kvm_is_error_hva(addr))
2255 r = __copy_from_user(data, (void __user *)addr + offset, len);
2261 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
2264 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2266 return __kvm_read_guest_page(slot, gfn, data, offset, len);
2268 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
2270 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
2271 int offset, int len)
2273 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2275 return __kvm_read_guest_page(slot, gfn, data, offset, len);
2277 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
2279 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
2281 gfn_t gfn = gpa >> PAGE_SHIFT;
2283 int offset = offset_in_page(gpa);
2286 while ((seg = next_segment(len, offset)) != 0) {
2287 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
2297 EXPORT_SYMBOL_GPL(kvm_read_guest);
2299 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
2301 gfn_t gfn = gpa >> PAGE_SHIFT;
2303 int offset = offset_in_page(gpa);
2306 while ((seg = next_segment(len, offset)) != 0) {
2307 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
2317 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
2319 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
2320 void *data, int offset, unsigned long len)
2325 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
2326 if (kvm_is_error_hva(addr))
2328 pagefault_disable();
2329 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
2336 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
2337 void *data, unsigned long len)
2339 gfn_t gfn = gpa >> PAGE_SHIFT;
2340 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2341 int offset = offset_in_page(gpa);
2343 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
2345 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
2347 static int __kvm_write_guest_page(struct kvm_memory_slot *memslot, gfn_t gfn,
2348 const void *data, int offset, int len)
2353 addr = gfn_to_hva_memslot(memslot, gfn);
2354 if (kvm_is_error_hva(addr))
2356 r = __copy_to_user((void __user *)addr + offset, data, len);
2359 mark_page_dirty_in_slot(memslot, gfn);
2363 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
2364 const void *data, int offset, int len)
2366 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2368 return __kvm_write_guest_page(slot, gfn, data, offset, len);
2370 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
2372 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
2373 const void *data, int offset, int len)
2375 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2377 return __kvm_write_guest_page(slot, gfn, data, offset, len);
2379 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
2381 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
2384 gfn_t gfn = gpa >> PAGE_SHIFT;
2386 int offset = offset_in_page(gpa);
2389 while ((seg = next_segment(len, offset)) != 0) {
2390 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
2400 EXPORT_SYMBOL_GPL(kvm_write_guest);
2402 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
2405 gfn_t gfn = gpa >> PAGE_SHIFT;
2407 int offset = offset_in_page(gpa);
2410 while ((seg = next_segment(len, offset)) != 0) {
2411 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
2421 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
2423 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
2424 struct gfn_to_hva_cache *ghc,
2425 gpa_t gpa, unsigned long len)
2427 int offset = offset_in_page(gpa);
2428 gfn_t start_gfn = gpa >> PAGE_SHIFT;
2429 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
2430 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
2431 gfn_t nr_pages_avail;
2433 /* Update ghc->generation before performing any error checks. */
2434 ghc->generation = slots->generation;
2436 if (start_gfn > end_gfn) {
2437 ghc->hva = KVM_HVA_ERR_BAD;
2442 * If the requested region crosses two memslots, we still
2443 * verify that the entire region is valid here.
2445 for ( ; start_gfn <= end_gfn; start_gfn += nr_pages_avail) {
2446 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
2447 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
2449 if (kvm_is_error_hva(ghc->hva))
2453 /* Use the slow path for cross page reads and writes. */
2454 if (nr_pages_needed == 1)
2457 ghc->memslot = NULL;
2464 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2465 gpa_t gpa, unsigned long len)
2467 struct kvm_memslots *slots = kvm_memslots(kvm);
2468 return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len);
2470 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
2472 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2473 void *data, unsigned int offset,
2476 struct kvm_memslots *slots = kvm_memslots(kvm);
2478 gpa_t gpa = ghc->gpa + offset;
2480 BUG_ON(len + offset > ghc->len);
2482 if (slots->generation != ghc->generation) {
2483 if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len))
2487 if (kvm_is_error_hva(ghc->hva))
2490 if (unlikely(!ghc->memslot))
2491 return kvm_write_guest(kvm, gpa, data, len);
2493 r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
2496 mark_page_dirty_in_slot(ghc->memslot, gpa >> PAGE_SHIFT);
2500 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached);
2502 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2503 void *data, unsigned long len)
2505 return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len);
2507 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
2509 int kvm_read_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2510 void *data, unsigned int offset,
2513 struct kvm_memslots *slots = kvm_memslots(kvm);
2515 gpa_t gpa = ghc->gpa + offset;
2517 BUG_ON(len + offset > ghc->len);
2519 if (slots->generation != ghc->generation) {
2520 if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len))
2524 if (kvm_is_error_hva(ghc->hva))
2527 if (unlikely(!ghc->memslot))
2528 return kvm_read_guest(kvm, gpa, data, len);
2530 r = __copy_from_user(data, (void __user *)ghc->hva + offset, len);
2536 EXPORT_SYMBOL_GPL(kvm_read_guest_offset_cached);
2538 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2539 void *data, unsigned long len)
2541 return kvm_read_guest_offset_cached(kvm, ghc, data, 0, len);
2543 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
2545 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
2547 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
2549 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
2551 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
2553 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
2555 gfn_t gfn = gpa >> PAGE_SHIFT;
2557 int offset = offset_in_page(gpa);
2560 while ((seg = next_segment(len, offset)) != 0) {
2561 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
2570 EXPORT_SYMBOL_GPL(kvm_clear_guest);
2572 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot,
2575 if (memslot && memslot->dirty_bitmap) {
2576 unsigned long rel_gfn = gfn - memslot->base_gfn;
2578 set_bit_le(rel_gfn, memslot->dirty_bitmap);
2582 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
2584 struct kvm_memory_slot *memslot;
2586 memslot = gfn_to_memslot(kvm, gfn);
2587 mark_page_dirty_in_slot(memslot, gfn);
2589 EXPORT_SYMBOL_GPL(mark_page_dirty);
2591 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
2593 struct kvm_memory_slot *memslot;
2595 memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2596 mark_page_dirty_in_slot(memslot, gfn);
2598 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
2600 void kvm_sigset_activate(struct kvm_vcpu *vcpu)
2602 if (!vcpu->sigset_active)
2606 * This does a lockless modification of ->real_blocked, which is fine
2607 * because, only current can change ->real_blocked and all readers of
2608 * ->real_blocked don't care as long ->real_blocked is always a subset
2611 sigprocmask(SIG_SETMASK, &vcpu->sigset, ¤t->real_blocked);
2614 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu)
2616 if (!vcpu->sigset_active)
2619 sigprocmask(SIG_SETMASK, ¤t->real_blocked, NULL);
2620 sigemptyset(¤t->real_blocked);
2623 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
2625 unsigned int old, val, grow, grow_start;
2627 old = val = vcpu->halt_poll_ns;
2628 grow_start = READ_ONCE(halt_poll_ns_grow_start);
2629 grow = READ_ONCE(halt_poll_ns_grow);
2634 if (val < grow_start)
2637 if (val > halt_poll_ns)
2640 vcpu->halt_poll_ns = val;
2642 trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
2645 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
2647 unsigned int old, val, shrink;
2649 old = val = vcpu->halt_poll_ns;
2650 shrink = READ_ONCE(halt_poll_ns_shrink);
2656 vcpu->halt_poll_ns = val;
2657 trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
2660 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
2663 int idx = srcu_read_lock(&vcpu->kvm->srcu);
2665 if (kvm_arch_vcpu_runnable(vcpu)) {
2666 kvm_make_request(KVM_REQ_UNHALT, vcpu);
2669 if (kvm_cpu_has_pending_timer(vcpu))
2671 if (signal_pending(current))
2676 srcu_read_unlock(&vcpu->kvm->srcu, idx);
2681 update_halt_poll_stats(struct kvm_vcpu *vcpu, u64 poll_ns, bool waited)
2684 vcpu->stat.halt_poll_fail_ns += poll_ns;
2686 vcpu->stat.halt_poll_success_ns += poll_ns;
2690 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2692 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
2694 ktime_t start, cur, poll_end;
2695 bool waited = false;
2698 kvm_arch_vcpu_blocking(vcpu);
2700 start = cur = poll_end = ktime_get();
2701 if (vcpu->halt_poll_ns && !kvm_arch_no_poll(vcpu)) {
2702 ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns);
2704 ++vcpu->stat.halt_attempted_poll;
2707 * This sets KVM_REQ_UNHALT if an interrupt
2710 if (kvm_vcpu_check_block(vcpu) < 0) {
2711 ++vcpu->stat.halt_successful_poll;
2712 if (!vcpu_valid_wakeup(vcpu))
2713 ++vcpu->stat.halt_poll_invalid;
2716 poll_end = cur = ktime_get();
2717 } while (single_task_running() && ktime_before(cur, stop));
2720 prepare_to_rcuwait(&vcpu->wait);
2722 set_current_state(TASK_INTERRUPTIBLE);
2724 if (kvm_vcpu_check_block(vcpu) < 0)
2730 finish_rcuwait(&vcpu->wait);
2733 kvm_arch_vcpu_unblocking(vcpu);
2734 block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
2736 update_halt_poll_stats(
2737 vcpu, ktime_to_ns(ktime_sub(poll_end, start)), waited);
2739 if (!kvm_arch_no_poll(vcpu)) {
2740 if (!vcpu_valid_wakeup(vcpu)) {
2741 shrink_halt_poll_ns(vcpu);
2742 } else if (vcpu->kvm->max_halt_poll_ns) {
2743 if (block_ns <= vcpu->halt_poll_ns)
2745 /* we had a long block, shrink polling */
2746 else if (vcpu->halt_poll_ns &&
2747 block_ns > vcpu->kvm->max_halt_poll_ns)
2748 shrink_halt_poll_ns(vcpu);
2749 /* we had a short halt and our poll time is too small */
2750 else if (vcpu->halt_poll_ns < vcpu->kvm->max_halt_poll_ns &&
2751 block_ns < vcpu->kvm->max_halt_poll_ns)
2752 grow_halt_poll_ns(vcpu);
2754 vcpu->halt_poll_ns = 0;
2758 trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu));
2759 kvm_arch_vcpu_block_finish(vcpu);
2761 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
2763 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
2765 struct rcuwait *waitp;
2767 waitp = kvm_arch_vcpu_get_wait(vcpu);
2768 if (rcuwait_wake_up(waitp)) {
2769 WRITE_ONCE(vcpu->ready, true);
2770 ++vcpu->stat.halt_wakeup;
2776 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
2780 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2782 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
2785 int cpu = vcpu->cpu;
2787 if (kvm_vcpu_wake_up(vcpu))
2791 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
2792 if (kvm_arch_vcpu_should_kick(vcpu))
2793 smp_send_reschedule(cpu);
2796 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
2797 #endif /* !CONFIG_S390 */
2799 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
2802 struct task_struct *task = NULL;
2806 pid = rcu_dereference(target->pid);
2808 task = get_pid_task(pid, PIDTYPE_PID);
2812 ret = yield_to(task, 1);
2813 put_task_struct(task);
2817 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
2820 * Helper that checks whether a VCPU is eligible for directed yield.
2821 * Most eligible candidate to yield is decided by following heuristics:
2823 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2824 * (preempted lock holder), indicated by @in_spin_loop.
2825 * Set at the beginning and cleared at the end of interception/PLE handler.
2827 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2828 * chance last time (mostly it has become eligible now since we have probably
2829 * yielded to lockholder in last iteration. This is done by toggling
2830 * @dy_eligible each time a VCPU checked for eligibility.)
2832 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2833 * to preempted lock-holder could result in wrong VCPU selection and CPU
2834 * burning. Giving priority for a potential lock-holder increases lock
2837 * Since algorithm is based on heuristics, accessing another VCPU data without
2838 * locking does not harm. It may result in trying to yield to same VCPU, fail
2839 * and continue with next VCPU and so on.
2841 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
2843 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2846 eligible = !vcpu->spin_loop.in_spin_loop ||
2847 vcpu->spin_loop.dy_eligible;
2849 if (vcpu->spin_loop.in_spin_loop)
2850 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
2859 * Unlike kvm_arch_vcpu_runnable, this function is called outside
2860 * a vcpu_load/vcpu_put pair. However, for most architectures
2861 * kvm_arch_vcpu_runnable does not require vcpu_load.
2863 bool __weak kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
2865 return kvm_arch_vcpu_runnable(vcpu);
2868 static bool vcpu_dy_runnable(struct kvm_vcpu *vcpu)
2870 if (kvm_arch_dy_runnable(vcpu))
2873 #ifdef CONFIG_KVM_ASYNC_PF
2874 if (!list_empty_careful(&vcpu->async_pf.done))
2881 void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode)
2883 struct kvm *kvm = me->kvm;
2884 struct kvm_vcpu *vcpu;
2885 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
2891 kvm_vcpu_set_in_spin_loop(me, true);
2893 * We boost the priority of a VCPU that is runnable but not
2894 * currently running, because it got preempted by something
2895 * else and called schedule in __vcpu_run. Hopefully that
2896 * VCPU is holding the lock that we need and will release it.
2897 * We approximate round-robin by starting at the last boosted VCPU.
2899 for (pass = 0; pass < 2 && !yielded && try; pass++) {
2900 kvm_for_each_vcpu(i, vcpu, kvm) {
2901 if (!pass && i <= last_boosted_vcpu) {
2902 i = last_boosted_vcpu;
2904 } else if (pass && i > last_boosted_vcpu)
2906 if (!READ_ONCE(vcpu->ready))
2910 if (rcuwait_active(&vcpu->wait) &&
2911 !vcpu_dy_runnable(vcpu))
2913 if (READ_ONCE(vcpu->preempted) && yield_to_kernel_mode &&
2914 !kvm_arch_vcpu_in_kernel(vcpu))
2916 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
2919 yielded = kvm_vcpu_yield_to(vcpu);
2921 kvm->last_boosted_vcpu = i;
2923 } else if (yielded < 0) {
2930 kvm_vcpu_set_in_spin_loop(me, false);
2932 /* Ensure vcpu is not eligible during next spinloop */
2933 kvm_vcpu_set_dy_eligible(me, false);
2935 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
2937 static vm_fault_t kvm_vcpu_fault(struct vm_fault *vmf)
2939 struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data;
2942 if (vmf->pgoff == 0)
2943 page = virt_to_page(vcpu->run);
2945 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
2946 page = virt_to_page(vcpu->arch.pio_data);
2948 #ifdef CONFIG_KVM_MMIO
2949 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
2950 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
2953 return kvm_arch_vcpu_fault(vcpu, vmf);
2959 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
2960 .fault = kvm_vcpu_fault,
2963 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2965 vma->vm_ops = &kvm_vcpu_vm_ops;
2969 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2971 struct kvm_vcpu *vcpu = filp->private_data;
2973 debugfs_remove_recursive(vcpu->debugfs_dentry);
2974 kvm_put_kvm(vcpu->kvm);
2978 static struct file_operations kvm_vcpu_fops = {
2979 .release = kvm_vcpu_release,
2980 .unlocked_ioctl = kvm_vcpu_ioctl,
2981 .mmap = kvm_vcpu_mmap,
2982 .llseek = noop_llseek,
2983 KVM_COMPAT(kvm_vcpu_compat_ioctl),
2987 * Allocates an inode for the vcpu.
2989 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2991 char name[8 + 1 + ITOA_MAX_LEN + 1];
2993 snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id);
2994 return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
2997 static void kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
2999 #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS
3000 char dir_name[ITOA_MAX_LEN * 2];
3002 if (!debugfs_initialized())
3005 snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
3006 vcpu->debugfs_dentry = debugfs_create_dir(dir_name,
3007 vcpu->kvm->debugfs_dentry);
3009 kvm_arch_create_vcpu_debugfs(vcpu);
3014 * Creates some virtual cpus. Good luck creating more than one.
3016 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
3019 struct kvm_vcpu *vcpu;
3022 if (id >= KVM_MAX_VCPU_ID)
3025 mutex_lock(&kvm->lock);
3026 if (kvm->created_vcpus == KVM_MAX_VCPUS) {
3027 mutex_unlock(&kvm->lock);
3031 kvm->created_vcpus++;
3032 mutex_unlock(&kvm->lock);
3034 r = kvm_arch_vcpu_precreate(kvm, id);
3036 goto vcpu_decrement;
3038 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
3041 goto vcpu_decrement;
3044 BUILD_BUG_ON(sizeof(struct kvm_run) > PAGE_SIZE);
3045 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
3050 vcpu->run = page_address(page);
3052 kvm_vcpu_init(vcpu, kvm, id);
3054 r = kvm_arch_vcpu_create(vcpu);
3056 goto vcpu_free_run_page;
3058 mutex_lock(&kvm->lock);
3059 if (kvm_get_vcpu_by_id(kvm, id)) {
3061 goto unlock_vcpu_destroy;
3064 vcpu->vcpu_idx = atomic_read(&kvm->online_vcpus);
3065 BUG_ON(kvm->vcpus[vcpu->vcpu_idx]);
3067 /* Now it's all set up, let userspace reach it */
3069 r = create_vcpu_fd(vcpu);
3071 kvm_put_kvm_no_destroy(kvm);
3072 goto unlock_vcpu_destroy;
3075 kvm->vcpus[vcpu->vcpu_idx] = vcpu;
3078 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
3079 * before kvm->online_vcpu's incremented value.
3082 atomic_inc(&kvm->online_vcpus);
3084 mutex_unlock(&kvm->lock);
3085 kvm_arch_vcpu_postcreate(vcpu);
3086 kvm_create_vcpu_debugfs(vcpu);
3089 unlock_vcpu_destroy:
3090 mutex_unlock(&kvm->lock);
3091 kvm_arch_vcpu_destroy(vcpu);
3093 free_page((unsigned long)vcpu->run);
3095 kmem_cache_free(kvm_vcpu_cache, vcpu);
3097 mutex_lock(&kvm->lock);
3098 kvm->created_vcpus--;
3099 mutex_unlock(&kvm->lock);
3103 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
3106 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
3107 vcpu->sigset_active = 1;
3108 vcpu->sigset = *sigset;
3110 vcpu->sigset_active = 0;
3114 static long kvm_vcpu_ioctl(struct file *filp,
3115 unsigned int ioctl, unsigned long arg)
3117 struct kvm_vcpu *vcpu = filp->private_data;
3118 void __user *argp = (void __user *)arg;
3120 struct kvm_fpu *fpu = NULL;
3121 struct kvm_sregs *kvm_sregs = NULL;
3123 if (vcpu->kvm->mm != current->mm)
3126 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
3130 * Some architectures have vcpu ioctls that are asynchronous to vcpu
3131 * execution; mutex_lock() would break them.
3133 r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg);
3134 if (r != -ENOIOCTLCMD)
3137 if (mutex_lock_killable(&vcpu->mutex))
3145 oldpid = rcu_access_pointer(vcpu->pid);
3146 if (unlikely(oldpid != task_pid(current))) {
3147 /* The thread running this VCPU changed. */
3150 r = kvm_arch_vcpu_run_pid_change(vcpu);
3154 newpid = get_task_pid(current, PIDTYPE_PID);
3155 rcu_assign_pointer(vcpu->pid, newpid);
3160 r = kvm_arch_vcpu_ioctl_run(vcpu);
3161 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
3164 case KVM_GET_REGS: {
3165 struct kvm_regs *kvm_regs;
3168 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL_ACCOUNT);
3171 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
3175 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
3182 case KVM_SET_REGS: {
3183 struct kvm_regs *kvm_regs;
3185 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
3186 if (IS_ERR(kvm_regs)) {
3187 r = PTR_ERR(kvm_regs);
3190 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
3194 case KVM_GET_SREGS: {
3195 kvm_sregs = kzalloc(sizeof(struct kvm_sregs),
3196 GFP_KERNEL_ACCOUNT);
3200 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
3204 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
3209 case KVM_SET_SREGS: {
3210 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
3211 if (IS_ERR(kvm_sregs)) {
3212 r = PTR_ERR(kvm_sregs);
3216 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
3219 case KVM_GET_MP_STATE: {
3220 struct kvm_mp_state mp_state;
3222 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
3226 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
3231 case KVM_SET_MP_STATE: {
3232 struct kvm_mp_state mp_state;
3235 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
3237 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
3240 case KVM_TRANSLATE: {
3241 struct kvm_translation tr;
3244 if (copy_from_user(&tr, argp, sizeof(tr)))
3246 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
3250 if (copy_to_user(argp, &tr, sizeof(tr)))
3255 case KVM_SET_GUEST_DEBUG: {
3256 struct kvm_guest_debug dbg;
3259 if (copy_from_user(&dbg, argp, sizeof(dbg)))
3261 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
3264 case KVM_SET_SIGNAL_MASK: {
3265 struct kvm_signal_mask __user *sigmask_arg = argp;
3266 struct kvm_signal_mask kvm_sigmask;
3267 sigset_t sigset, *p;
3272 if (copy_from_user(&kvm_sigmask, argp,
3273 sizeof(kvm_sigmask)))
3276 if (kvm_sigmask.len != sizeof(sigset))
3279 if (copy_from_user(&sigset, sigmask_arg->sigset,
3284 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
3288 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL_ACCOUNT);
3292 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
3296 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
3302 fpu = memdup_user(argp, sizeof(*fpu));
3308 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
3312 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
3315 mutex_unlock(&vcpu->mutex);
3321 #ifdef CONFIG_KVM_COMPAT
3322 static long kvm_vcpu_compat_ioctl(struct file *filp,
3323 unsigned int ioctl, unsigned long arg)
3325 struct kvm_vcpu *vcpu = filp->private_data;
3326 void __user *argp = compat_ptr(arg);
3329 if (vcpu->kvm->mm != current->mm)
3333 case KVM_SET_SIGNAL_MASK: {
3334 struct kvm_signal_mask __user *sigmask_arg = argp;
3335 struct kvm_signal_mask kvm_sigmask;
3340 if (copy_from_user(&kvm_sigmask, argp,
3341 sizeof(kvm_sigmask)))
3344 if (kvm_sigmask.len != sizeof(compat_sigset_t))
3347 if (get_compat_sigset(&sigset, (void *)sigmask_arg->sigset))
3349 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
3351 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
3355 r = kvm_vcpu_ioctl(filp, ioctl, arg);
3363 static int kvm_device_mmap(struct file *filp, struct vm_area_struct *vma)
3365 struct kvm_device *dev = filp->private_data;
3368 return dev->ops->mmap(dev, vma);
3373 static int kvm_device_ioctl_attr(struct kvm_device *dev,
3374 int (*accessor)(struct kvm_device *dev,
3375 struct kvm_device_attr *attr),
3378 struct kvm_device_attr attr;
3383 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
3386 return accessor(dev, &attr);
3389 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
3392 struct kvm_device *dev = filp->private_data;
3394 if (dev->kvm->mm != current->mm)
3398 case KVM_SET_DEVICE_ATTR:
3399 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
3400 case KVM_GET_DEVICE_ATTR:
3401 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
3402 case KVM_HAS_DEVICE_ATTR:
3403 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
3405 if (dev->ops->ioctl)
3406 return dev->ops->ioctl(dev, ioctl, arg);
3412 static int kvm_device_release(struct inode *inode, struct file *filp)
3414 struct kvm_device *dev = filp->private_data;
3415 struct kvm *kvm = dev->kvm;
3417 if (dev->ops->release) {
3418 mutex_lock(&kvm->lock);
3419 list_del(&dev->vm_node);
3420 dev->ops->release(dev);
3421 mutex_unlock(&kvm->lock);
3428 static const struct file_operations kvm_device_fops = {
3429 .unlocked_ioctl = kvm_device_ioctl,
3430 .release = kvm_device_release,
3431 KVM_COMPAT(kvm_device_ioctl),
3432 .mmap = kvm_device_mmap,
3435 struct kvm_device *kvm_device_from_filp(struct file *filp)
3437 if (filp->f_op != &kvm_device_fops)
3440 return filp->private_data;
3443 static const struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
3444 #ifdef CONFIG_KVM_MPIC
3445 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
3446 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
3450 int kvm_register_device_ops(const struct kvm_device_ops *ops, u32 type)
3452 if (type >= ARRAY_SIZE(kvm_device_ops_table))
3455 if (kvm_device_ops_table[type] != NULL)
3458 kvm_device_ops_table[type] = ops;
3462 void kvm_unregister_device_ops(u32 type)
3464 if (kvm_device_ops_table[type] != NULL)
3465 kvm_device_ops_table[type] = NULL;
3468 static int kvm_ioctl_create_device(struct kvm *kvm,
3469 struct kvm_create_device *cd)
3471 const struct kvm_device_ops *ops = NULL;
3472 struct kvm_device *dev;
3473 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
3477 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
3480 type = array_index_nospec(cd->type, ARRAY_SIZE(kvm_device_ops_table));
3481 ops = kvm_device_ops_table[type];
3488 dev = kzalloc(sizeof(*dev), GFP_KERNEL_ACCOUNT);
3495 mutex_lock(&kvm->lock);
3496 ret = ops->create(dev, type);
3498 mutex_unlock(&kvm->lock);
3502 list_add(&dev->vm_node, &kvm->devices);
3503 mutex_unlock(&kvm->lock);
3509 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
3511 kvm_put_kvm_no_destroy(kvm);
3512 mutex_lock(&kvm->lock);
3513 list_del(&dev->vm_node);
3514 mutex_unlock(&kvm->lock);
3523 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
3526 case KVM_CAP_USER_MEMORY:
3527 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
3528 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
3529 case KVM_CAP_INTERNAL_ERROR_DATA:
3530 #ifdef CONFIG_HAVE_KVM_MSI
3531 case KVM_CAP_SIGNAL_MSI:
3533 #ifdef CONFIG_HAVE_KVM_IRQFD
3535 case KVM_CAP_IRQFD_RESAMPLE:
3537 case KVM_CAP_IOEVENTFD_ANY_LENGTH:
3538 case KVM_CAP_CHECK_EXTENSION_VM:
3539 case KVM_CAP_ENABLE_CAP_VM:
3540 case KVM_CAP_HALT_POLL:
3542 #ifdef CONFIG_KVM_MMIO
3543 case KVM_CAP_COALESCED_MMIO:
3544 return KVM_COALESCED_MMIO_PAGE_OFFSET;
3545 case KVM_CAP_COALESCED_PIO:
3548 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3549 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2:
3550 return KVM_DIRTY_LOG_MANUAL_CAPS;
3552 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3553 case KVM_CAP_IRQ_ROUTING:
3554 return KVM_MAX_IRQ_ROUTES;
3556 #if KVM_ADDRESS_SPACE_NUM > 1
3557 case KVM_CAP_MULTI_ADDRESS_SPACE:
3558 return KVM_ADDRESS_SPACE_NUM;
3560 case KVM_CAP_NR_MEMSLOTS:
3561 return KVM_USER_MEM_SLOTS;
3565 return kvm_vm_ioctl_check_extension(kvm, arg);
3568 int __attribute__((weak)) kvm_vm_ioctl_enable_cap(struct kvm *kvm,
3569 struct kvm_enable_cap *cap)
3574 static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm,
3575 struct kvm_enable_cap *cap)
3578 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3579 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2: {
3580 u64 allowed_options = KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE;
3582 if (cap->args[0] & KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE)
3583 allowed_options = KVM_DIRTY_LOG_MANUAL_CAPS;
3585 if (cap->flags || (cap->args[0] & ~allowed_options))
3587 kvm->manual_dirty_log_protect = cap->args[0];
3591 case KVM_CAP_HALT_POLL: {
3592 if (cap->flags || cap->args[0] != (unsigned int)cap->args[0])
3595 kvm->max_halt_poll_ns = cap->args[0];
3599 return kvm_vm_ioctl_enable_cap(kvm, cap);
3603 static long kvm_vm_ioctl(struct file *filp,
3604 unsigned int ioctl, unsigned long arg)
3606 struct kvm *kvm = filp->private_data;
3607 void __user *argp = (void __user *)arg;
3610 if (kvm->mm != current->mm)
3613 case KVM_CREATE_VCPU:
3614 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
3616 case KVM_ENABLE_CAP: {
3617 struct kvm_enable_cap cap;
3620 if (copy_from_user(&cap, argp, sizeof(cap)))
3622 r = kvm_vm_ioctl_enable_cap_generic(kvm, &cap);
3625 case KVM_SET_USER_MEMORY_REGION: {
3626 struct kvm_userspace_memory_region kvm_userspace_mem;
3629 if (copy_from_user(&kvm_userspace_mem, argp,
3630 sizeof(kvm_userspace_mem)))
3633 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
3636 case KVM_GET_DIRTY_LOG: {
3637 struct kvm_dirty_log log;
3640 if (copy_from_user(&log, argp, sizeof(log)))
3642 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3645 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3646 case KVM_CLEAR_DIRTY_LOG: {
3647 struct kvm_clear_dirty_log log;
3650 if (copy_from_user(&log, argp, sizeof(log)))
3652 r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
3656 #ifdef CONFIG_KVM_MMIO
3657 case KVM_REGISTER_COALESCED_MMIO: {
3658 struct kvm_coalesced_mmio_zone zone;
3661 if (copy_from_user(&zone, argp, sizeof(zone)))
3663 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
3666 case KVM_UNREGISTER_COALESCED_MMIO: {
3667 struct kvm_coalesced_mmio_zone zone;
3670 if (copy_from_user(&zone, argp, sizeof(zone)))
3672 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
3677 struct kvm_irqfd data;
3680 if (copy_from_user(&data, argp, sizeof(data)))
3682 r = kvm_irqfd(kvm, &data);
3685 case KVM_IOEVENTFD: {
3686 struct kvm_ioeventfd data;
3689 if (copy_from_user(&data, argp, sizeof(data)))
3691 r = kvm_ioeventfd(kvm, &data);
3694 #ifdef CONFIG_HAVE_KVM_MSI
3695 case KVM_SIGNAL_MSI: {
3699 if (copy_from_user(&msi, argp, sizeof(msi)))
3701 r = kvm_send_userspace_msi(kvm, &msi);
3705 #ifdef __KVM_HAVE_IRQ_LINE
3706 case KVM_IRQ_LINE_STATUS:
3707 case KVM_IRQ_LINE: {
3708 struct kvm_irq_level irq_event;
3711 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
3714 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
3715 ioctl == KVM_IRQ_LINE_STATUS);
3720 if (ioctl == KVM_IRQ_LINE_STATUS) {
3721 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
3729 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3730 case KVM_SET_GSI_ROUTING: {
3731 struct kvm_irq_routing routing;
3732 struct kvm_irq_routing __user *urouting;
3733 struct kvm_irq_routing_entry *entries = NULL;
3736 if (copy_from_user(&routing, argp, sizeof(routing)))
3739 if (!kvm_arch_can_set_irq_routing(kvm))
3741 if (routing.nr > KVM_MAX_IRQ_ROUTES)
3747 entries = vmalloc(array_size(sizeof(*entries),
3753 if (copy_from_user(entries, urouting->entries,
3754 routing.nr * sizeof(*entries)))
3755 goto out_free_irq_routing;
3757 r = kvm_set_irq_routing(kvm, entries, routing.nr,
3759 out_free_irq_routing:
3763 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
3764 case KVM_CREATE_DEVICE: {
3765 struct kvm_create_device cd;
3768 if (copy_from_user(&cd, argp, sizeof(cd)))
3771 r = kvm_ioctl_create_device(kvm, &cd);
3776 if (copy_to_user(argp, &cd, sizeof(cd)))
3782 case KVM_CHECK_EXTENSION:
3783 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
3786 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
3792 #ifdef CONFIG_KVM_COMPAT
3793 struct compat_kvm_dirty_log {
3797 compat_uptr_t dirty_bitmap; /* one bit per page */
3802 static long kvm_vm_compat_ioctl(struct file *filp,
3803 unsigned int ioctl, unsigned long arg)
3805 struct kvm *kvm = filp->private_data;
3808 if (kvm->mm != current->mm)
3811 case KVM_GET_DIRTY_LOG: {
3812 struct compat_kvm_dirty_log compat_log;
3813 struct kvm_dirty_log log;
3815 if (copy_from_user(&compat_log, (void __user *)arg,
3816 sizeof(compat_log)))
3818 log.slot = compat_log.slot;
3819 log.padding1 = compat_log.padding1;
3820 log.padding2 = compat_log.padding2;
3821 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
3823 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3827 r = kvm_vm_ioctl(filp, ioctl, arg);
3833 static struct file_operations kvm_vm_fops = {
3834 .release = kvm_vm_release,
3835 .unlocked_ioctl = kvm_vm_ioctl,
3836 .llseek = noop_llseek,
3837 KVM_COMPAT(kvm_vm_compat_ioctl),
3840 static int kvm_dev_ioctl_create_vm(unsigned long type)
3846 kvm = kvm_create_vm(type);
3848 return PTR_ERR(kvm);
3849 #ifdef CONFIG_KVM_MMIO
3850 r = kvm_coalesced_mmio_init(kvm);
3854 r = get_unused_fd_flags(O_CLOEXEC);
3858 file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
3866 * Don't call kvm_put_kvm anymore at this point; file->f_op is
3867 * already set, with ->release() being kvm_vm_release(). In error
3868 * cases it will be called by the final fput(file) and will take
3869 * care of doing kvm_put_kvm(kvm).
3871 if (kvm_create_vm_debugfs(kvm, r) < 0) {
3876 kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm);
3878 fd_install(r, file);
3886 static long kvm_dev_ioctl(struct file *filp,
3887 unsigned int ioctl, unsigned long arg)
3892 case KVM_GET_API_VERSION:
3895 r = KVM_API_VERSION;
3898 r = kvm_dev_ioctl_create_vm(arg);
3900 case KVM_CHECK_EXTENSION:
3901 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
3903 case KVM_GET_VCPU_MMAP_SIZE:
3906 r = PAGE_SIZE; /* struct kvm_run */
3908 r += PAGE_SIZE; /* pio data page */
3910 #ifdef CONFIG_KVM_MMIO
3911 r += PAGE_SIZE; /* coalesced mmio ring page */
3914 case KVM_TRACE_ENABLE:
3915 case KVM_TRACE_PAUSE:
3916 case KVM_TRACE_DISABLE:
3920 return kvm_arch_dev_ioctl(filp, ioctl, arg);
3926 static struct file_operations kvm_chardev_ops = {
3927 .unlocked_ioctl = kvm_dev_ioctl,
3928 .llseek = noop_llseek,
3929 KVM_COMPAT(kvm_dev_ioctl),
3932 static struct miscdevice kvm_dev = {
3938 static void hardware_enable_nolock(void *junk)
3940 int cpu = raw_smp_processor_id();
3943 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
3946 cpumask_set_cpu(cpu, cpus_hardware_enabled);
3948 r = kvm_arch_hardware_enable();
3951 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3952 atomic_inc(&hardware_enable_failed);
3953 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
3957 static int kvm_starting_cpu(unsigned int cpu)
3959 raw_spin_lock(&kvm_count_lock);
3960 if (kvm_usage_count)
3961 hardware_enable_nolock(NULL);
3962 raw_spin_unlock(&kvm_count_lock);
3966 static void hardware_disable_nolock(void *junk)
3968 int cpu = raw_smp_processor_id();
3970 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
3972 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3973 kvm_arch_hardware_disable();
3976 static int kvm_dying_cpu(unsigned int cpu)
3978 raw_spin_lock(&kvm_count_lock);
3979 if (kvm_usage_count)
3980 hardware_disable_nolock(NULL);
3981 raw_spin_unlock(&kvm_count_lock);
3985 static void hardware_disable_all_nolock(void)
3987 BUG_ON(!kvm_usage_count);
3990 if (!kvm_usage_count)
3991 on_each_cpu(hardware_disable_nolock, NULL, 1);
3994 static void hardware_disable_all(void)
3996 raw_spin_lock(&kvm_count_lock);
3997 hardware_disable_all_nolock();
3998 raw_spin_unlock(&kvm_count_lock);
4001 static int hardware_enable_all(void)
4005 raw_spin_lock(&kvm_count_lock);
4008 if (kvm_usage_count == 1) {
4009 atomic_set(&hardware_enable_failed, 0);
4010 on_each_cpu(hardware_enable_nolock, NULL, 1);
4012 if (atomic_read(&hardware_enable_failed)) {
4013 hardware_disable_all_nolock();
4018 raw_spin_unlock(&kvm_count_lock);
4023 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
4027 * Some (well, at least mine) BIOSes hang on reboot if
4030 * And Intel TXT required VMX off for all cpu when system shutdown.
4032 pr_info("kvm: exiting hardware virtualization\n");
4033 kvm_rebooting = true;
4034 on_each_cpu(hardware_disable_nolock, NULL, 1);
4038 static struct notifier_block kvm_reboot_notifier = {
4039 .notifier_call = kvm_reboot,
4043 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
4047 for (i = 0; i < bus->dev_count; i++) {
4048 struct kvm_io_device *pos = bus->range[i].dev;
4050 kvm_iodevice_destructor(pos);
4055 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
4056 const struct kvm_io_range *r2)
4058 gpa_t addr1 = r1->addr;
4059 gpa_t addr2 = r2->addr;
4064 /* If r2->len == 0, match the exact address. If r2->len != 0,
4065 * accept any overlapping write. Any order is acceptable for
4066 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
4067 * we process all of them.
4080 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
4082 return kvm_io_bus_cmp(p1, p2);
4085 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
4086 gpa_t addr, int len)
4088 struct kvm_io_range *range, key;
4091 key = (struct kvm_io_range) {
4096 range = bsearch(&key, bus->range, bus->dev_count,
4097 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
4101 off = range - bus->range;
4103 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
4109 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
4110 struct kvm_io_range *range, const void *val)
4114 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
4118 while (idx < bus->dev_count &&
4119 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
4120 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
4129 /* kvm_io_bus_write - called under kvm->slots_lock */
4130 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
4131 int len, const void *val)
4133 struct kvm_io_bus *bus;
4134 struct kvm_io_range range;
4137 range = (struct kvm_io_range) {
4142 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
4145 r = __kvm_io_bus_write(vcpu, bus, &range, val);
4146 return r < 0 ? r : 0;
4148 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
4150 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
4151 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
4152 gpa_t addr, int len, const void *val, long cookie)
4154 struct kvm_io_bus *bus;
4155 struct kvm_io_range range;
4157 range = (struct kvm_io_range) {
4162 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
4166 /* First try the device referenced by cookie. */
4167 if ((cookie >= 0) && (cookie < bus->dev_count) &&
4168 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
4169 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
4174 * cookie contained garbage; fall back to search and return the
4175 * correct cookie value.
4177 return __kvm_io_bus_write(vcpu, bus, &range, val);
4180 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
4181 struct kvm_io_range *range, void *val)
4185 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
4189 while (idx < bus->dev_count &&
4190 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
4191 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
4200 /* kvm_io_bus_read - called under kvm->slots_lock */
4201 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
4204 struct kvm_io_bus *bus;
4205 struct kvm_io_range range;
4208 range = (struct kvm_io_range) {
4213 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
4216 r = __kvm_io_bus_read(vcpu, bus, &range, val);
4217 return r < 0 ? r : 0;
4220 /* Caller must hold slots_lock. */
4221 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
4222 int len, struct kvm_io_device *dev)
4225 struct kvm_io_bus *new_bus, *bus;
4226 struct kvm_io_range range;
4228 bus = kvm_get_bus(kvm, bus_idx);
4232 /* exclude ioeventfd which is limited by maximum fd */
4233 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
4236 new_bus = kmalloc(struct_size(bus, range, bus->dev_count + 1),
4237 GFP_KERNEL_ACCOUNT);
4241 range = (struct kvm_io_range) {
4247 for (i = 0; i < bus->dev_count; i++)
4248 if (kvm_io_bus_cmp(&bus->range[i], &range) > 0)
4251 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
4252 new_bus->dev_count++;
4253 new_bus->range[i] = range;
4254 memcpy(new_bus->range + i + 1, bus->range + i,
4255 (bus->dev_count - i) * sizeof(struct kvm_io_range));
4256 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
4257 synchronize_srcu_expedited(&kvm->srcu);
4263 /* Caller must hold slots_lock. */
4264 void kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
4265 struct kvm_io_device *dev)
4268 struct kvm_io_bus *new_bus, *bus;
4270 bus = kvm_get_bus(kvm, bus_idx);
4274 for (i = 0; i < bus->dev_count; i++)
4275 if (bus->range[i].dev == dev) {
4279 if (i == bus->dev_count)
4282 new_bus = kmalloc(struct_size(bus, range, bus->dev_count - 1),
4283 GFP_KERNEL_ACCOUNT);
4285 pr_err("kvm: failed to shrink bus, removing it completely\n");
4289 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
4290 new_bus->dev_count--;
4291 memcpy(new_bus->range + i, bus->range + i + 1,
4292 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
4295 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
4296 synchronize_srcu_expedited(&kvm->srcu);
4301 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
4304 struct kvm_io_bus *bus;
4305 int dev_idx, srcu_idx;
4306 struct kvm_io_device *iodev = NULL;
4308 srcu_idx = srcu_read_lock(&kvm->srcu);
4310 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
4314 dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
4318 iodev = bus->range[dev_idx].dev;
4321 srcu_read_unlock(&kvm->srcu, srcu_idx);
4325 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
4327 static int kvm_debugfs_open(struct inode *inode, struct file *file,
4328 int (*get)(void *, u64 *), int (*set)(void *, u64),
4331 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
4334 /* The debugfs files are a reference to the kvm struct which
4335 * is still valid when kvm_destroy_vm is called.
4336 * To avoid the race between open and the removal of the debugfs
4337 * directory we test against the users count.
4339 if (!refcount_inc_not_zero(&stat_data->kvm->users_count))
4342 if (simple_attr_open(inode, file, get,
4343 KVM_DBGFS_GET_MODE(stat_data->dbgfs_item) & 0222
4346 kvm_put_kvm(stat_data->kvm);
4353 static int kvm_debugfs_release(struct inode *inode, struct file *file)
4355 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
4358 simple_attr_release(inode, file);
4359 kvm_put_kvm(stat_data->kvm);
4364 static int kvm_get_stat_per_vm(struct kvm *kvm, size_t offset, u64 *val)
4366 *val = *(ulong *)((void *)kvm + offset);
4371 static int kvm_clear_stat_per_vm(struct kvm *kvm, size_t offset)
4373 *(ulong *)((void *)kvm + offset) = 0;
4378 static int kvm_get_stat_per_vcpu(struct kvm *kvm, size_t offset, u64 *val)
4381 struct kvm_vcpu *vcpu;
4385 kvm_for_each_vcpu(i, vcpu, kvm)
4386 *val += *(u64 *)((void *)vcpu + offset);
4391 static int kvm_clear_stat_per_vcpu(struct kvm *kvm, size_t offset)
4394 struct kvm_vcpu *vcpu;
4396 kvm_for_each_vcpu(i, vcpu, kvm)
4397 *(u64 *)((void *)vcpu + offset) = 0;
4402 static int kvm_stat_data_get(void *data, u64 *val)
4405 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
4407 switch (stat_data->dbgfs_item->kind) {
4409 r = kvm_get_stat_per_vm(stat_data->kvm,
4410 stat_data->dbgfs_item->offset, val);
4413 r = kvm_get_stat_per_vcpu(stat_data->kvm,
4414 stat_data->dbgfs_item->offset, val);
4421 static int kvm_stat_data_clear(void *data, u64 val)
4424 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
4429 switch (stat_data->dbgfs_item->kind) {
4431 r = kvm_clear_stat_per_vm(stat_data->kvm,
4432 stat_data->dbgfs_item->offset);
4435 r = kvm_clear_stat_per_vcpu(stat_data->kvm,
4436 stat_data->dbgfs_item->offset);
4443 static int kvm_stat_data_open(struct inode *inode, struct file *file)
4445 __simple_attr_check_format("%llu\n", 0ull);
4446 return kvm_debugfs_open(inode, file, kvm_stat_data_get,
4447 kvm_stat_data_clear, "%llu\n");
4450 static const struct file_operations stat_fops_per_vm = {
4451 .owner = THIS_MODULE,
4452 .open = kvm_stat_data_open,
4453 .release = kvm_debugfs_release,
4454 .read = simple_attr_read,
4455 .write = simple_attr_write,
4456 .llseek = no_llseek,
4459 static int vm_stat_get(void *_offset, u64 *val)
4461 unsigned offset = (long)_offset;
4466 mutex_lock(&kvm_lock);
4467 list_for_each_entry(kvm, &vm_list, vm_list) {
4468 kvm_get_stat_per_vm(kvm, offset, &tmp_val);
4471 mutex_unlock(&kvm_lock);
4475 static int vm_stat_clear(void *_offset, u64 val)
4477 unsigned offset = (long)_offset;
4483 mutex_lock(&kvm_lock);
4484 list_for_each_entry(kvm, &vm_list, vm_list) {
4485 kvm_clear_stat_per_vm(kvm, offset);
4487 mutex_unlock(&kvm_lock);
4492 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n");
4494 static int vcpu_stat_get(void *_offset, u64 *val)
4496 unsigned offset = (long)_offset;
4501 mutex_lock(&kvm_lock);
4502 list_for_each_entry(kvm, &vm_list, vm_list) {
4503 kvm_get_stat_per_vcpu(kvm, offset, &tmp_val);
4506 mutex_unlock(&kvm_lock);
4510 static int vcpu_stat_clear(void *_offset, u64 val)
4512 unsigned offset = (long)_offset;
4518 mutex_lock(&kvm_lock);
4519 list_for_each_entry(kvm, &vm_list, vm_list) {
4520 kvm_clear_stat_per_vcpu(kvm, offset);
4522 mutex_unlock(&kvm_lock);
4527 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear,
4530 static const struct file_operations *stat_fops[] = {
4531 [KVM_STAT_VCPU] = &vcpu_stat_fops,
4532 [KVM_STAT_VM] = &vm_stat_fops,
4535 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm)
4537 struct kobj_uevent_env *env;
4538 unsigned long long created, active;
4540 if (!kvm_dev.this_device || !kvm)
4543 mutex_lock(&kvm_lock);
4544 if (type == KVM_EVENT_CREATE_VM) {
4545 kvm_createvm_count++;
4547 } else if (type == KVM_EVENT_DESTROY_VM) {
4550 created = kvm_createvm_count;
4551 active = kvm_active_vms;
4552 mutex_unlock(&kvm_lock);
4554 env = kzalloc(sizeof(*env), GFP_KERNEL_ACCOUNT);
4558 add_uevent_var(env, "CREATED=%llu", created);
4559 add_uevent_var(env, "COUNT=%llu", active);
4561 if (type == KVM_EVENT_CREATE_VM) {
4562 add_uevent_var(env, "EVENT=create");
4563 kvm->userspace_pid = task_pid_nr(current);
4564 } else if (type == KVM_EVENT_DESTROY_VM) {
4565 add_uevent_var(env, "EVENT=destroy");
4567 add_uevent_var(env, "PID=%d", kvm->userspace_pid);
4569 if (!IS_ERR_OR_NULL(kvm->debugfs_dentry)) {
4570 char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL_ACCOUNT);
4573 tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX);
4575 add_uevent_var(env, "STATS_PATH=%s", tmp);
4579 /* no need for checks, since we are adding at most only 5 keys */
4580 env->envp[env->envp_idx++] = NULL;
4581 kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp);
4585 static void kvm_init_debug(void)
4587 struct kvm_stats_debugfs_item *p;
4589 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
4591 kvm_debugfs_num_entries = 0;
4592 for (p = debugfs_entries; p->name; ++p, kvm_debugfs_num_entries++) {
4593 debugfs_create_file(p->name, KVM_DBGFS_GET_MODE(p),
4594 kvm_debugfs_dir, (void *)(long)p->offset,
4595 stat_fops[p->kind]);
4599 static int kvm_suspend(void)
4601 if (kvm_usage_count)
4602 hardware_disable_nolock(NULL);
4606 static void kvm_resume(void)
4608 if (kvm_usage_count) {
4609 #ifdef CONFIG_LOCKDEP
4610 WARN_ON(lockdep_is_held(&kvm_count_lock));
4612 hardware_enable_nolock(NULL);
4616 static struct syscore_ops kvm_syscore_ops = {
4617 .suspend = kvm_suspend,
4618 .resume = kvm_resume,
4622 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
4624 return container_of(pn, struct kvm_vcpu, preempt_notifier);
4627 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
4629 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
4631 WRITE_ONCE(vcpu->preempted, false);
4632 WRITE_ONCE(vcpu->ready, false);
4634 __this_cpu_write(kvm_running_vcpu, vcpu);
4635 kvm_arch_sched_in(vcpu, cpu);
4636 kvm_arch_vcpu_load(vcpu, cpu);
4639 static void kvm_sched_out(struct preempt_notifier *pn,
4640 struct task_struct *next)
4642 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
4644 if (current->state == TASK_RUNNING) {
4645 WRITE_ONCE(vcpu->preempted, true);
4646 WRITE_ONCE(vcpu->ready, true);
4648 kvm_arch_vcpu_put(vcpu);
4649 __this_cpu_write(kvm_running_vcpu, NULL);
4653 * kvm_get_running_vcpu - get the vcpu running on the current CPU.
4655 * We can disable preemption locally around accessing the per-CPU variable,
4656 * and use the resolved vcpu pointer after enabling preemption again,
4657 * because even if the current thread is migrated to another CPU, reading
4658 * the per-CPU value later will give us the same value as we update the
4659 * per-CPU variable in the preempt notifier handlers.
4661 struct kvm_vcpu *kvm_get_running_vcpu(void)
4663 struct kvm_vcpu *vcpu;
4666 vcpu = __this_cpu_read(kvm_running_vcpu);
4671 EXPORT_SYMBOL_GPL(kvm_get_running_vcpu);
4674 * kvm_get_running_vcpus - get the per-CPU array of currently running vcpus.
4676 struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void)
4678 return &kvm_running_vcpu;
4681 struct kvm_cpu_compat_check {
4686 static void check_processor_compat(void *data)
4688 struct kvm_cpu_compat_check *c = data;
4690 *c->ret = kvm_arch_check_processor_compat(c->opaque);
4693 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
4694 struct module *module)
4696 struct kvm_cpu_compat_check c;
4700 r = kvm_arch_init(opaque);
4705 * kvm_arch_init makes sure there's at most one caller
4706 * for architectures that support multiple implementations,
4707 * like intel and amd on x86.
4708 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
4709 * conflicts in case kvm is already setup for another implementation.
4711 r = kvm_irqfd_init();
4715 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
4720 r = kvm_arch_hardware_setup(opaque);
4726 for_each_online_cpu(cpu) {
4727 smp_call_function_single(cpu, check_processor_compat, &c, 1);
4732 r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting",
4733 kvm_starting_cpu, kvm_dying_cpu);
4736 register_reboot_notifier(&kvm_reboot_notifier);
4738 /* A kmem cache lets us meet the alignment requirements of fx_save. */
4740 vcpu_align = __alignof__(struct kvm_vcpu);
4742 kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align,
4744 offsetof(struct kvm_vcpu, arch),
4745 sizeof_field(struct kvm_vcpu, arch),
4747 if (!kvm_vcpu_cache) {
4752 r = kvm_async_pf_init();
4756 kvm_chardev_ops.owner = module;
4757 kvm_vm_fops.owner = module;
4758 kvm_vcpu_fops.owner = module;
4760 r = misc_register(&kvm_dev);
4762 pr_err("kvm: misc device register failed\n");
4766 register_syscore_ops(&kvm_syscore_ops);
4768 kvm_preempt_ops.sched_in = kvm_sched_in;
4769 kvm_preempt_ops.sched_out = kvm_sched_out;
4773 r = kvm_vfio_ops_init();
4779 kvm_async_pf_deinit();
4781 kmem_cache_destroy(kvm_vcpu_cache);
4783 unregister_reboot_notifier(&kvm_reboot_notifier);
4784 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4786 kvm_arch_hardware_unsetup();
4788 free_cpumask_var(cpus_hardware_enabled);
4796 EXPORT_SYMBOL_GPL(kvm_init);
4800 debugfs_remove_recursive(kvm_debugfs_dir);
4801 misc_deregister(&kvm_dev);
4802 kmem_cache_destroy(kvm_vcpu_cache);
4803 kvm_async_pf_deinit();
4804 unregister_syscore_ops(&kvm_syscore_ops);
4805 unregister_reboot_notifier(&kvm_reboot_notifier);
4806 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4807 on_each_cpu(hardware_disable_nolock, NULL, 1);
4808 kvm_arch_hardware_unsetup();
4811 free_cpumask_var(cpus_hardware_enabled);
4812 kvm_vfio_ops_exit();
4814 EXPORT_SYMBOL_GPL(kvm_exit);
4816 struct kvm_vm_worker_thread_context {
4818 struct task_struct *parent;
4819 struct completion init_done;
4820 kvm_vm_thread_fn_t thread_fn;
4825 static int kvm_vm_worker_thread(void *context)
4828 * The init_context is allocated on the stack of the parent thread, so
4829 * we have to locally copy anything that is needed beyond initialization
4831 struct kvm_vm_worker_thread_context *init_context = context;
4832 struct kvm *kvm = init_context->kvm;
4833 kvm_vm_thread_fn_t thread_fn = init_context->thread_fn;
4834 uintptr_t data = init_context->data;
4837 err = kthread_park(current);
4838 /* kthread_park(current) is never supposed to return an error */
4843 err = cgroup_attach_task_all(init_context->parent, current);
4845 kvm_err("%s: cgroup_attach_task_all failed with err %d\n",
4850 set_user_nice(current, task_nice(init_context->parent));
4853 init_context->err = err;
4854 complete(&init_context->init_done);
4855 init_context = NULL;
4860 /* Wait to be woken up by the spawner before proceeding. */
4863 if (!kthread_should_stop())
4864 err = thread_fn(kvm, data);
4869 int kvm_vm_create_worker_thread(struct kvm *kvm, kvm_vm_thread_fn_t thread_fn,
4870 uintptr_t data, const char *name,
4871 struct task_struct **thread_ptr)
4873 struct kvm_vm_worker_thread_context init_context = {};
4874 struct task_struct *thread;
4877 init_context.kvm = kvm;
4878 init_context.parent = current;
4879 init_context.thread_fn = thread_fn;
4880 init_context.data = data;
4881 init_completion(&init_context.init_done);
4883 thread = kthread_run(kvm_vm_worker_thread, &init_context,
4884 "%s-%d", name, task_pid_nr(current));
4886 return PTR_ERR(thread);
4888 /* kthread_run is never supposed to return NULL */
4889 WARN_ON(thread == NULL);
4891 wait_for_completion(&init_context.init_done);
4893 if (!init_context.err)
4894 *thread_ptr = thread;
4896 return init_context.err;