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>
58 #include <asm/pgtable.h>
60 #include "coalesced_mmio.h"
64 #define CREATE_TRACE_POINTS
65 #include <trace/events/kvm.h>
67 /* Worst case buffer size needed for holding an integer. */
68 #define ITOA_MAX_LEN 12
70 MODULE_AUTHOR("Qumranet");
71 MODULE_LICENSE("GPL");
73 /* Architectures should define their poll value according to the halt latency */
74 unsigned int halt_poll_ns = KVM_HALT_POLL_NS_DEFAULT;
75 module_param(halt_poll_ns, uint, 0644);
76 EXPORT_SYMBOL_GPL(halt_poll_ns);
78 /* Default doubles per-vcpu halt_poll_ns. */
79 unsigned int halt_poll_ns_grow = 2;
80 module_param(halt_poll_ns_grow, uint, 0644);
81 EXPORT_SYMBOL_GPL(halt_poll_ns_grow);
83 /* The start value to grow halt_poll_ns from */
84 unsigned int halt_poll_ns_grow_start = 10000; /* 10us */
85 module_param(halt_poll_ns_grow_start, uint, 0644);
86 EXPORT_SYMBOL_GPL(halt_poll_ns_grow_start);
88 /* Default resets per-vcpu halt_poll_ns . */
89 unsigned int halt_poll_ns_shrink;
90 module_param(halt_poll_ns_shrink, uint, 0644);
91 EXPORT_SYMBOL_GPL(halt_poll_ns_shrink);
96 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
99 DEFINE_MUTEX(kvm_lock);
100 static DEFINE_RAW_SPINLOCK(kvm_count_lock);
103 static cpumask_var_t cpus_hardware_enabled;
104 static int kvm_usage_count;
105 static atomic_t hardware_enable_failed;
107 static struct kmem_cache *kvm_vcpu_cache;
109 static __read_mostly struct preempt_ops kvm_preempt_ops;
110 static DEFINE_PER_CPU(struct kvm_vcpu *, kvm_running_vcpu);
112 struct dentry *kvm_debugfs_dir;
113 EXPORT_SYMBOL_GPL(kvm_debugfs_dir);
115 static int kvm_debugfs_num_entries;
116 static const struct file_operations stat_fops_per_vm;
118 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
120 #ifdef CONFIG_KVM_COMPAT
121 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
123 #define KVM_COMPAT(c) .compat_ioctl = (c)
126 * For architectures that don't implement a compat infrastructure,
127 * adopt a double line of defense:
128 * - Prevent a compat task from opening /dev/kvm
129 * - If the open has been done by a 64bit task, and the KVM fd
130 * passed to a compat task, let the ioctls fail.
132 static long kvm_no_compat_ioctl(struct file *file, unsigned int ioctl,
133 unsigned long arg) { return -EINVAL; }
135 static int kvm_no_compat_open(struct inode *inode, struct file *file)
137 return is_compat_task() ? -ENODEV : 0;
139 #define KVM_COMPAT(c) .compat_ioctl = kvm_no_compat_ioctl, \
140 .open = kvm_no_compat_open
142 static int hardware_enable_all(void);
143 static void hardware_disable_all(void);
145 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
147 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot, gfn_t gfn);
149 __visible bool kvm_rebooting;
150 EXPORT_SYMBOL_GPL(kvm_rebooting);
152 #define KVM_EVENT_CREATE_VM 0
153 #define KVM_EVENT_DESTROY_VM 1
154 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm);
155 static unsigned long long kvm_createvm_count;
156 static unsigned long long kvm_active_vms;
158 __weak int kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
159 unsigned long start, unsigned long end, bool blockable)
164 bool kvm_is_zone_device_pfn(kvm_pfn_t pfn)
167 * The metadata used by is_zone_device_page() to determine whether or
168 * not a page is ZONE_DEVICE is guaranteed to be valid if and only if
169 * the device has been pinned, e.g. by get_user_pages(). WARN if the
170 * page_count() is zero to help detect bad usage of this helper.
172 if (!pfn_valid(pfn) || WARN_ON_ONCE(!page_count(pfn_to_page(pfn))))
175 return is_zone_device_page(pfn_to_page(pfn));
178 bool kvm_is_reserved_pfn(kvm_pfn_t pfn)
181 * ZONE_DEVICE pages currently set PG_reserved, but from a refcounting
182 * perspective they are "normal" pages, albeit with slightly different
186 return PageReserved(pfn_to_page(pfn)) &&
188 !kvm_is_zone_device_pfn(pfn);
193 bool kvm_is_transparent_hugepage(kvm_pfn_t pfn)
195 struct page *page = pfn_to_page(pfn);
197 if (!PageTransCompoundMap(page))
200 return is_transparent_hugepage(compound_head(page));
204 * Switches to specified vcpu, until a matching vcpu_put()
206 void vcpu_load(struct kvm_vcpu *vcpu)
210 __this_cpu_write(kvm_running_vcpu, vcpu);
211 preempt_notifier_register(&vcpu->preempt_notifier);
212 kvm_arch_vcpu_load(vcpu, cpu);
215 EXPORT_SYMBOL_GPL(vcpu_load);
217 void vcpu_put(struct kvm_vcpu *vcpu)
220 kvm_arch_vcpu_put(vcpu);
221 preempt_notifier_unregister(&vcpu->preempt_notifier);
222 __this_cpu_write(kvm_running_vcpu, NULL);
225 EXPORT_SYMBOL_GPL(vcpu_put);
227 /* TODO: merge with kvm_arch_vcpu_should_kick */
228 static bool kvm_request_needs_ipi(struct kvm_vcpu *vcpu, unsigned req)
230 int mode = kvm_vcpu_exiting_guest_mode(vcpu);
233 * We need to wait for the VCPU to reenable interrupts and get out of
234 * READING_SHADOW_PAGE_TABLES mode.
236 if (req & KVM_REQUEST_WAIT)
237 return mode != OUTSIDE_GUEST_MODE;
240 * Need to kick a running VCPU, but otherwise there is nothing to do.
242 return mode == IN_GUEST_MODE;
245 static void ack_flush(void *_completed)
249 static inline bool kvm_kick_many_cpus(const struct cpumask *cpus, bool wait)
252 cpus = cpu_online_mask;
254 if (cpumask_empty(cpus))
257 smp_call_function_many(cpus, ack_flush, NULL, wait);
261 bool kvm_make_vcpus_request_mask(struct kvm *kvm, unsigned int req,
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))
274 kvm_make_request(req, vcpu);
277 if (!(req & KVM_REQUEST_NO_WAKEUP) && kvm_vcpu_wake_up(vcpu))
280 if (tmp != NULL && cpu != -1 && cpu != me &&
281 kvm_request_needs_ipi(vcpu, req))
282 __cpumask_set_cpu(cpu, tmp);
285 called = kvm_kick_many_cpus(tmp, !!(req & KVM_REQUEST_WAIT));
291 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
296 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
298 called = kvm_make_vcpus_request_mask(kvm, req, NULL, cpus);
300 free_cpumask_var(cpus);
304 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
305 void kvm_flush_remote_tlbs(struct kvm *kvm)
308 * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in
309 * kvm_make_all_cpus_request.
311 long dirty_count = smp_load_acquire(&kvm->tlbs_dirty);
314 * We want to publish modifications to the page tables before reading
315 * mode. Pairs with a memory barrier in arch-specific code.
316 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
317 * and smp_mb in walk_shadow_page_lockless_begin/end.
318 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
320 * There is already an smp_mb__after_atomic() before
321 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
324 if (!kvm_arch_flush_remote_tlb(kvm)
325 || kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
326 ++kvm->stat.remote_tlb_flush;
327 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
329 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
332 void kvm_reload_remote_mmus(struct kvm *kvm)
334 kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
337 static void kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
339 mutex_init(&vcpu->mutex);
344 init_swait_queue_head(&vcpu->wq);
345 kvm_async_pf_vcpu_init(vcpu);
348 INIT_LIST_HEAD(&vcpu->blocked_vcpu_list);
350 kvm_vcpu_set_in_spin_loop(vcpu, false);
351 kvm_vcpu_set_dy_eligible(vcpu, false);
352 vcpu->preempted = false;
354 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
357 void kvm_vcpu_destroy(struct kvm_vcpu *vcpu)
359 kvm_arch_vcpu_destroy(vcpu);
362 * No need for rcu_read_lock as VCPU_RUN is the only place that changes
363 * the vcpu->pid pointer, and at destruction time all file descriptors
366 put_pid(rcu_dereference_protected(vcpu->pid, 1));
368 free_page((unsigned long)vcpu->run);
369 kmem_cache_free(kvm_vcpu_cache, vcpu);
371 EXPORT_SYMBOL_GPL(kvm_vcpu_destroy);
373 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
374 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
376 return container_of(mn, struct kvm, mmu_notifier);
379 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
380 struct mm_struct *mm,
381 unsigned long address,
384 struct kvm *kvm = mmu_notifier_to_kvm(mn);
387 idx = srcu_read_lock(&kvm->srcu);
388 spin_lock(&kvm->mmu_lock);
389 kvm->mmu_notifier_seq++;
391 if (kvm_set_spte_hva(kvm, address, pte))
392 kvm_flush_remote_tlbs(kvm);
394 spin_unlock(&kvm->mmu_lock);
395 srcu_read_unlock(&kvm->srcu, idx);
398 static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
399 const struct mmu_notifier_range *range)
401 struct kvm *kvm = mmu_notifier_to_kvm(mn);
402 int need_tlb_flush = 0, idx;
405 idx = srcu_read_lock(&kvm->srcu);
406 spin_lock(&kvm->mmu_lock);
408 * The count increase must become visible at unlock time as no
409 * spte can be established without taking the mmu_lock and
410 * count is also read inside the mmu_lock critical section.
412 kvm->mmu_notifier_count++;
413 need_tlb_flush = kvm_unmap_hva_range(kvm, range->start, range->end);
414 need_tlb_flush |= kvm->tlbs_dirty;
415 /* we've to flush the tlb before the pages can be freed */
417 kvm_flush_remote_tlbs(kvm);
419 spin_unlock(&kvm->mmu_lock);
421 ret = kvm_arch_mmu_notifier_invalidate_range(kvm, range->start,
423 mmu_notifier_range_blockable(range));
425 srcu_read_unlock(&kvm->srcu, idx);
430 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
431 const struct mmu_notifier_range *range)
433 struct kvm *kvm = mmu_notifier_to_kvm(mn);
435 spin_lock(&kvm->mmu_lock);
437 * This sequence increase will notify the kvm page fault that
438 * the page that is going to be mapped in the spte could have
441 kvm->mmu_notifier_seq++;
444 * The above sequence increase must be visible before the
445 * below count decrease, which is ensured by the smp_wmb above
446 * in conjunction with the smp_rmb in mmu_notifier_retry().
448 kvm->mmu_notifier_count--;
449 spin_unlock(&kvm->mmu_lock);
451 BUG_ON(kvm->mmu_notifier_count < 0);
454 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
455 struct mm_struct *mm,
459 struct kvm *kvm = mmu_notifier_to_kvm(mn);
462 idx = srcu_read_lock(&kvm->srcu);
463 spin_lock(&kvm->mmu_lock);
465 young = kvm_age_hva(kvm, start, end);
467 kvm_flush_remote_tlbs(kvm);
469 spin_unlock(&kvm->mmu_lock);
470 srcu_read_unlock(&kvm->srcu, idx);
475 static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
476 struct mm_struct *mm,
480 struct kvm *kvm = mmu_notifier_to_kvm(mn);
483 idx = srcu_read_lock(&kvm->srcu);
484 spin_lock(&kvm->mmu_lock);
486 * Even though we do not flush TLB, this will still adversely
487 * affect performance on pre-Haswell Intel EPT, where there is
488 * no EPT Access Bit to clear so that we have to tear down EPT
489 * tables instead. If we find this unacceptable, we can always
490 * add a parameter to kvm_age_hva so that it effectively doesn't
491 * do anything on clear_young.
493 * Also note that currently we never issue secondary TLB flushes
494 * from clear_young, leaving this job up to the regular system
495 * cadence. If we find this inaccurate, we might come up with a
496 * more sophisticated heuristic later.
498 young = kvm_age_hva(kvm, start, end);
499 spin_unlock(&kvm->mmu_lock);
500 srcu_read_unlock(&kvm->srcu, idx);
505 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
506 struct mm_struct *mm,
507 unsigned long address)
509 struct kvm *kvm = mmu_notifier_to_kvm(mn);
512 idx = srcu_read_lock(&kvm->srcu);
513 spin_lock(&kvm->mmu_lock);
514 young = kvm_test_age_hva(kvm, address);
515 spin_unlock(&kvm->mmu_lock);
516 srcu_read_unlock(&kvm->srcu, idx);
521 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
522 struct mm_struct *mm)
524 struct kvm *kvm = mmu_notifier_to_kvm(mn);
527 idx = srcu_read_lock(&kvm->srcu);
528 kvm_arch_flush_shadow_all(kvm);
529 srcu_read_unlock(&kvm->srcu, idx);
532 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
533 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
534 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
535 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
536 .clear_young = kvm_mmu_notifier_clear_young,
537 .test_young = kvm_mmu_notifier_test_young,
538 .change_pte = kvm_mmu_notifier_change_pte,
539 .release = kvm_mmu_notifier_release,
542 static int kvm_init_mmu_notifier(struct kvm *kvm)
544 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
545 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
548 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
550 static int kvm_init_mmu_notifier(struct kvm *kvm)
555 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
557 static struct kvm_memslots *kvm_alloc_memslots(void)
560 struct kvm_memslots *slots;
562 slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL_ACCOUNT);
566 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
567 slots->id_to_index[i] = -1;
572 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
574 if (!memslot->dirty_bitmap)
577 kvfree(memslot->dirty_bitmap);
578 memslot->dirty_bitmap = NULL;
581 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
583 kvm_destroy_dirty_bitmap(slot);
585 kvm_arch_free_memslot(kvm, slot);
591 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
593 struct kvm_memory_slot *memslot;
598 kvm_for_each_memslot(memslot, slots)
599 kvm_free_memslot(kvm, memslot);
604 static void kvm_destroy_vm_debugfs(struct kvm *kvm)
608 if (!kvm->debugfs_dentry)
611 debugfs_remove_recursive(kvm->debugfs_dentry);
613 if (kvm->debugfs_stat_data) {
614 for (i = 0; i < kvm_debugfs_num_entries; i++)
615 kfree(kvm->debugfs_stat_data[i]);
616 kfree(kvm->debugfs_stat_data);
620 static int kvm_create_vm_debugfs(struct kvm *kvm, int fd)
622 char dir_name[ITOA_MAX_LEN * 2];
623 struct kvm_stat_data *stat_data;
624 struct kvm_stats_debugfs_item *p;
626 if (!debugfs_initialized())
629 snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd);
630 kvm->debugfs_dentry = debugfs_create_dir(dir_name, kvm_debugfs_dir);
632 kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries,
633 sizeof(*kvm->debugfs_stat_data),
635 if (!kvm->debugfs_stat_data)
638 for (p = debugfs_entries; p->name; p++) {
639 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT);
643 stat_data->kvm = kvm;
644 stat_data->dbgfs_item = p;
645 kvm->debugfs_stat_data[p - debugfs_entries] = stat_data;
646 debugfs_create_file(p->name, KVM_DBGFS_GET_MODE(p),
647 kvm->debugfs_dentry, stat_data,
654 * Called after the VM is otherwise initialized, but just before adding it to
657 int __weak kvm_arch_post_init_vm(struct kvm *kvm)
663 * Called just after removing the VM from the vm_list, but before doing any
666 void __weak kvm_arch_pre_destroy_vm(struct kvm *kvm)
670 static struct kvm *kvm_create_vm(unsigned long type)
672 struct kvm *kvm = kvm_arch_alloc_vm();
677 return ERR_PTR(-ENOMEM);
679 spin_lock_init(&kvm->mmu_lock);
681 kvm->mm = current->mm;
682 kvm_eventfd_init(kvm);
683 mutex_init(&kvm->lock);
684 mutex_init(&kvm->irq_lock);
685 mutex_init(&kvm->slots_lock);
686 INIT_LIST_HEAD(&kvm->devices);
688 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
690 if (init_srcu_struct(&kvm->srcu))
691 goto out_err_no_srcu;
692 if (init_srcu_struct(&kvm->irq_srcu))
693 goto out_err_no_irq_srcu;
695 refcount_set(&kvm->users_count, 1);
696 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
697 struct kvm_memslots *slots = kvm_alloc_memslots();
700 goto out_err_no_arch_destroy_vm;
701 /* Generations must be different for each address space. */
702 slots->generation = i;
703 rcu_assign_pointer(kvm->memslots[i], slots);
706 for (i = 0; i < KVM_NR_BUSES; i++) {
707 rcu_assign_pointer(kvm->buses[i],
708 kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL_ACCOUNT));
710 goto out_err_no_arch_destroy_vm;
713 r = kvm_arch_init_vm(kvm, type);
715 goto out_err_no_arch_destroy_vm;
717 r = hardware_enable_all();
719 goto out_err_no_disable;
721 #ifdef CONFIG_HAVE_KVM_IRQFD
722 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
725 r = kvm_init_mmu_notifier(kvm);
727 goto out_err_no_mmu_notifier;
729 r = kvm_arch_post_init_vm(kvm);
733 mutex_lock(&kvm_lock);
734 list_add(&kvm->vm_list, &vm_list);
735 mutex_unlock(&kvm_lock);
737 preempt_notifier_inc();
742 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
743 if (kvm->mmu_notifier.ops)
744 mmu_notifier_unregister(&kvm->mmu_notifier, current->mm);
746 out_err_no_mmu_notifier:
747 hardware_disable_all();
749 kvm_arch_destroy_vm(kvm);
750 out_err_no_arch_destroy_vm:
751 WARN_ON_ONCE(!refcount_dec_and_test(&kvm->users_count));
752 for (i = 0; i < KVM_NR_BUSES; i++)
753 kfree(kvm_get_bus(kvm, i));
754 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
755 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
756 cleanup_srcu_struct(&kvm->irq_srcu);
758 cleanup_srcu_struct(&kvm->srcu);
760 kvm_arch_free_vm(kvm);
765 static void kvm_destroy_devices(struct kvm *kvm)
767 struct kvm_device *dev, *tmp;
770 * We do not need to take the kvm->lock here, because nobody else
771 * has a reference to the struct kvm at this point and therefore
772 * cannot access the devices list anyhow.
774 list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
775 list_del(&dev->vm_node);
776 dev->ops->destroy(dev);
780 static void kvm_destroy_vm(struct kvm *kvm)
783 struct mm_struct *mm = kvm->mm;
785 kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm);
786 kvm_destroy_vm_debugfs(kvm);
787 kvm_arch_sync_events(kvm);
788 mutex_lock(&kvm_lock);
789 list_del(&kvm->vm_list);
790 mutex_unlock(&kvm_lock);
791 kvm_arch_pre_destroy_vm(kvm);
793 kvm_free_irq_routing(kvm);
794 for (i = 0; i < KVM_NR_BUSES; i++) {
795 struct kvm_io_bus *bus = kvm_get_bus(kvm, i);
798 kvm_io_bus_destroy(bus);
799 kvm->buses[i] = NULL;
801 kvm_coalesced_mmio_free(kvm);
802 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
803 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
805 kvm_arch_flush_shadow_all(kvm);
807 kvm_arch_destroy_vm(kvm);
808 kvm_destroy_devices(kvm);
809 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
810 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
811 cleanup_srcu_struct(&kvm->irq_srcu);
812 cleanup_srcu_struct(&kvm->srcu);
813 kvm_arch_free_vm(kvm);
814 preempt_notifier_dec();
815 hardware_disable_all();
819 void kvm_get_kvm(struct kvm *kvm)
821 refcount_inc(&kvm->users_count);
823 EXPORT_SYMBOL_GPL(kvm_get_kvm);
825 void kvm_put_kvm(struct kvm *kvm)
827 if (refcount_dec_and_test(&kvm->users_count))
830 EXPORT_SYMBOL_GPL(kvm_put_kvm);
833 * Used to put a reference that was taken on behalf of an object associated
834 * with a user-visible file descriptor, e.g. a vcpu or device, if installation
835 * of the new file descriptor fails and the reference cannot be transferred to
836 * its final owner. In such cases, the caller is still actively using @kvm and
837 * will fail miserably if the refcount unexpectedly hits zero.
839 void kvm_put_kvm_no_destroy(struct kvm *kvm)
841 WARN_ON(refcount_dec_and_test(&kvm->users_count));
843 EXPORT_SYMBOL_GPL(kvm_put_kvm_no_destroy);
845 static int kvm_vm_release(struct inode *inode, struct file *filp)
847 struct kvm *kvm = filp->private_data;
849 kvm_irqfd_release(kvm);
856 * Allocation size is twice as large as the actual dirty bitmap size.
857 * See kvm_vm_ioctl_get_dirty_log() why this is needed.
859 static int kvm_alloc_dirty_bitmap(struct kvm_memory_slot *memslot)
861 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
863 memslot->dirty_bitmap = kvzalloc(dirty_bytes, GFP_KERNEL_ACCOUNT);
864 if (!memslot->dirty_bitmap)
871 * Delete a memslot by decrementing the number of used slots and shifting all
872 * other entries in the array forward one spot.
874 static inline void kvm_memslot_delete(struct kvm_memslots *slots,
875 struct kvm_memory_slot *memslot)
877 struct kvm_memory_slot *mslots = slots->memslots;
880 if (WARN_ON(slots->id_to_index[memslot->id] == -1))
885 if (atomic_read(&slots->lru_slot) >= slots->used_slots)
886 atomic_set(&slots->lru_slot, 0);
888 for (i = slots->id_to_index[memslot->id]; i < slots->used_slots; i++) {
889 mslots[i] = mslots[i + 1];
890 slots->id_to_index[mslots[i].id] = i;
892 mslots[i] = *memslot;
893 slots->id_to_index[memslot->id] = -1;
897 * "Insert" a new memslot by incrementing the number of used slots. Returns
898 * the new slot's initial index into the memslots array.
900 static inline int kvm_memslot_insert_back(struct kvm_memslots *slots)
902 return slots->used_slots++;
906 * Move a changed memslot backwards in the array by shifting existing slots
907 * with a higher GFN toward the front of the array. Note, the changed memslot
908 * itself is not preserved in the array, i.e. not swapped at this time, only
909 * its new index into the array is tracked. Returns the changed memslot's
910 * current index into the memslots array.
912 static inline int kvm_memslot_move_backward(struct kvm_memslots *slots,
913 struct kvm_memory_slot *memslot)
915 struct kvm_memory_slot *mslots = slots->memslots;
918 if (WARN_ON_ONCE(slots->id_to_index[memslot->id] == -1) ||
919 WARN_ON_ONCE(!slots->used_slots))
923 * Move the target memslot backward in the array by shifting existing
924 * memslots with a higher GFN (than the target memslot) towards the
925 * front of the array.
927 for (i = slots->id_to_index[memslot->id]; i < slots->used_slots - 1; i++) {
928 if (memslot->base_gfn > mslots[i + 1].base_gfn)
931 WARN_ON_ONCE(memslot->base_gfn == mslots[i + 1].base_gfn);
933 /* Shift the next memslot forward one and update its index. */
934 mslots[i] = mslots[i + 1];
935 slots->id_to_index[mslots[i].id] = i;
941 * Move a changed memslot forwards in the array by shifting existing slots with
942 * a lower GFN toward the back of the array. Note, the changed memslot itself
943 * is not preserved in the array, i.e. not swapped at this time, only its new
944 * index into the array is tracked. Returns the changed memslot's final index
945 * into the memslots array.
947 static inline int kvm_memslot_move_forward(struct kvm_memslots *slots,
948 struct kvm_memory_slot *memslot,
951 struct kvm_memory_slot *mslots = slots->memslots;
954 for (i = start; i > 0; i--) {
955 if (memslot->base_gfn < mslots[i - 1].base_gfn)
958 WARN_ON_ONCE(memslot->base_gfn == mslots[i - 1].base_gfn);
960 /* Shift the next memslot back one and update its index. */
961 mslots[i] = mslots[i - 1];
962 slots->id_to_index[mslots[i].id] = i;
968 * Re-sort memslots based on their GFN to account for an added, deleted, or
969 * moved memslot. Sorting memslots by GFN allows using a binary search during
972 * IMPORTANT: Slots are sorted from highest GFN to lowest GFN! I.e. the entry
973 * at memslots[0] has the highest GFN.
975 * The sorting algorithm takes advantage of having initially sorted memslots
976 * and knowing the position of the changed memslot. Sorting is also optimized
977 * by not swapping the updated memslot and instead only shifting other memslots
978 * and tracking the new index for the update memslot. Only once its final
979 * index is known is the updated memslot copied into its position in the array.
981 * - When deleting a memslot, the deleted memslot simply needs to be moved to
982 * the end of the array.
984 * - When creating a memslot, the algorithm "inserts" the new memslot at the
985 * end of the array and then it forward to its correct location.
987 * - When moving a memslot, the algorithm first moves the updated memslot
988 * backward to handle the scenario where the memslot's GFN was changed to a
989 * lower value. update_memslots() then falls through and runs the same flow
990 * as creating a memslot to move the memslot forward to handle the scenario
991 * where its GFN was changed to a higher value.
993 * Note, slots are sorted from highest->lowest instead of lowest->highest for
994 * historical reasons. Originally, invalid memslots where denoted by having
995 * GFN=0, thus sorting from highest->lowest naturally sorted invalid memslots
996 * to the end of the array. The current algorithm uses dedicated logic to
997 * delete a memslot and thus does not rely on invalid memslots having GFN=0.
999 * The other historical motiviation for highest->lowest was to improve the
1000 * performance of memslot lookup. KVM originally used a linear search starting
1001 * at memslots[0]. On x86, the largest memslot usually has one of the highest,
1002 * if not *the* highest, GFN, as the bulk of the guest's RAM is located in a
1003 * single memslot above the 4gb boundary. As the largest memslot is also the
1004 * most likely to be referenced, sorting it to the front of the array was
1005 * advantageous. The current binary search starts from the middle of the array
1006 * and uses an LRU pointer to improve performance for all memslots and GFNs.
1008 static void update_memslots(struct kvm_memslots *slots,
1009 struct kvm_memory_slot *memslot,
1010 enum kvm_mr_change change)
1014 if (change == KVM_MR_DELETE) {
1015 kvm_memslot_delete(slots, memslot);
1017 if (change == KVM_MR_CREATE)
1018 i = kvm_memslot_insert_back(slots);
1020 i = kvm_memslot_move_backward(slots, memslot);
1021 i = kvm_memslot_move_forward(slots, memslot, i);
1024 * Copy the memslot to its new position in memslots and update
1025 * its index accordingly.
1027 slots->memslots[i] = *memslot;
1028 slots->id_to_index[memslot->id] = i;
1032 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
1034 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
1036 #ifdef __KVM_HAVE_READONLY_MEM
1037 valid_flags |= KVM_MEM_READONLY;
1040 if (mem->flags & ~valid_flags)
1046 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
1047 int as_id, struct kvm_memslots *slots)
1049 struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id);
1050 u64 gen = old_memslots->generation;
1052 WARN_ON(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS);
1053 slots->generation = gen | KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
1055 rcu_assign_pointer(kvm->memslots[as_id], slots);
1056 synchronize_srcu_expedited(&kvm->srcu);
1059 * Increment the new memslot generation a second time, dropping the
1060 * update in-progress flag and incrementing the generation based on
1061 * the number of address spaces. This provides a unique and easily
1062 * identifiable generation number while the memslots are in flux.
1064 gen = slots->generation & ~KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
1067 * Generations must be unique even across address spaces. We do not need
1068 * a global counter for that, instead the generation space is evenly split
1069 * across address spaces. For example, with two address spaces, address
1070 * space 0 will use generations 0, 2, 4, ... while address space 1 will
1071 * use generations 1, 3, 5, ...
1073 gen += KVM_ADDRESS_SPACE_NUM;
1075 kvm_arch_memslots_updated(kvm, gen);
1077 slots->generation = gen;
1079 return old_memslots;
1083 * Note, at a minimum, the current number of used slots must be allocated, even
1084 * when deleting a memslot, as we need a complete duplicate of the memslots for
1085 * use when invalidating a memslot prior to deleting/moving the memslot.
1087 static struct kvm_memslots *kvm_dup_memslots(struct kvm_memslots *old,
1088 enum kvm_mr_change change)
1090 struct kvm_memslots *slots;
1091 size_t old_size, new_size;
1093 old_size = sizeof(struct kvm_memslots) +
1094 (sizeof(struct kvm_memory_slot) * old->used_slots);
1096 if (change == KVM_MR_CREATE)
1097 new_size = old_size + sizeof(struct kvm_memory_slot);
1099 new_size = old_size;
1101 slots = kvzalloc(new_size, GFP_KERNEL_ACCOUNT);
1103 memcpy(slots, old, old_size);
1108 static int kvm_set_memslot(struct kvm *kvm,
1109 const struct kvm_userspace_memory_region *mem,
1110 struct kvm_memory_slot *old,
1111 struct kvm_memory_slot *new, int as_id,
1112 enum kvm_mr_change change)
1114 struct kvm_memory_slot *slot;
1115 struct kvm_memslots *slots;
1118 slots = kvm_dup_memslots(__kvm_memslots(kvm, as_id), change);
1122 if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) {
1124 * Note, the INVALID flag needs to be in the appropriate entry
1125 * in the freshly allocated memslots, not in @old or @new.
1127 slot = id_to_memslot(slots, old->id);
1128 slot->flags |= KVM_MEMSLOT_INVALID;
1131 * We can re-use the old memslots, the only difference from the
1132 * newly installed memslots is the invalid flag, which will get
1133 * dropped by update_memslots anyway. We'll also revert to the
1134 * old memslots if preparing the new memory region fails.
1136 slots = install_new_memslots(kvm, as_id, slots);
1138 /* From this point no new shadow pages pointing to a deleted,
1139 * or moved, memslot will be created.
1141 * validation of sp->gfn happens in:
1142 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1143 * - kvm_is_visible_gfn (mmu_check_root)
1145 kvm_arch_flush_shadow_memslot(kvm, slot);
1148 r = kvm_arch_prepare_memory_region(kvm, new, mem, change);
1152 update_memslots(slots, new, change);
1153 slots = install_new_memslots(kvm, as_id, slots);
1155 kvm_arch_commit_memory_region(kvm, mem, old, new, change);
1161 if (change == KVM_MR_DELETE || change == KVM_MR_MOVE)
1162 slots = install_new_memslots(kvm, as_id, slots);
1167 static int kvm_delete_memslot(struct kvm *kvm,
1168 const struct kvm_userspace_memory_region *mem,
1169 struct kvm_memory_slot *old, int as_id)
1171 struct kvm_memory_slot new;
1177 memset(&new, 0, sizeof(new));
1180 r = kvm_set_memslot(kvm, mem, old, &new, as_id, KVM_MR_DELETE);
1184 kvm_free_memslot(kvm, old);
1189 * Allocate some memory and give it an address in the guest physical address
1192 * Discontiguous memory is allowed, mostly for framebuffers.
1194 * Must be called holding kvm->slots_lock for write.
1196 int __kvm_set_memory_region(struct kvm *kvm,
1197 const struct kvm_userspace_memory_region *mem)
1199 struct kvm_memory_slot old, new;
1200 struct kvm_memory_slot *tmp;
1201 enum kvm_mr_change change;
1205 r = check_memory_region_flags(mem);
1209 as_id = mem->slot >> 16;
1210 id = (u16)mem->slot;
1212 /* General sanity checks */
1213 if (mem->memory_size & (PAGE_SIZE - 1))
1215 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
1217 /* We can read the guest memory with __xxx_user() later on. */
1218 if ((id < KVM_USER_MEM_SLOTS) &&
1219 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
1220 !access_ok((void __user *)(unsigned long)mem->userspace_addr,
1223 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
1225 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
1229 * Make a full copy of the old memslot, the pointer will become stale
1230 * when the memslots are re-sorted by update_memslots(), and the old
1231 * memslot needs to be referenced after calling update_memslots(), e.g.
1232 * to free its resources and for arch specific behavior.
1234 tmp = id_to_memslot(__kvm_memslots(kvm, as_id), id);
1239 memset(&old, 0, sizeof(old));
1243 if (!mem->memory_size)
1244 return kvm_delete_memslot(kvm, mem, &old, as_id);
1247 new.base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
1248 new.npages = mem->memory_size >> PAGE_SHIFT;
1249 new.flags = mem->flags;
1250 new.userspace_addr = mem->userspace_addr;
1252 if (new.npages > KVM_MEM_MAX_NR_PAGES)
1256 change = KVM_MR_CREATE;
1257 new.dirty_bitmap = NULL;
1258 memset(&new.arch, 0, sizeof(new.arch));
1259 } else { /* Modify an existing slot. */
1260 if ((new.userspace_addr != old.userspace_addr) ||
1261 (new.npages != old.npages) ||
1262 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
1265 if (new.base_gfn != old.base_gfn)
1266 change = KVM_MR_MOVE;
1267 else if (new.flags != old.flags)
1268 change = KVM_MR_FLAGS_ONLY;
1269 else /* Nothing to change. */
1272 /* Copy dirty_bitmap and arch from the current memslot. */
1273 new.dirty_bitmap = old.dirty_bitmap;
1274 memcpy(&new.arch, &old.arch, sizeof(new.arch));
1277 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
1278 /* Check for overlaps */
1279 kvm_for_each_memslot(tmp, __kvm_memslots(kvm, as_id)) {
1282 if (!((new.base_gfn + new.npages <= tmp->base_gfn) ||
1283 (new.base_gfn >= tmp->base_gfn + tmp->npages)))
1288 /* Allocate/free page dirty bitmap as needed */
1289 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
1290 new.dirty_bitmap = NULL;
1291 else if (!new.dirty_bitmap) {
1292 r = kvm_alloc_dirty_bitmap(&new);
1296 if (kvm_dirty_log_manual_protect_and_init_set(kvm))
1297 bitmap_set(new.dirty_bitmap, 0, new.npages);
1300 r = kvm_set_memslot(kvm, mem, &old, &new, as_id, change);
1304 if (old.dirty_bitmap && !new.dirty_bitmap)
1305 kvm_destroy_dirty_bitmap(&old);
1309 if (new.dirty_bitmap && !old.dirty_bitmap)
1310 kvm_destroy_dirty_bitmap(&new);
1313 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1315 int kvm_set_memory_region(struct kvm *kvm,
1316 const struct kvm_userspace_memory_region *mem)
1320 mutex_lock(&kvm->slots_lock);
1321 r = __kvm_set_memory_region(kvm, mem);
1322 mutex_unlock(&kvm->slots_lock);
1325 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1327 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1328 struct kvm_userspace_memory_region *mem)
1330 if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1333 return kvm_set_memory_region(kvm, mem);
1336 #ifndef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1338 * kvm_get_dirty_log - get a snapshot of dirty pages
1339 * @kvm: pointer to kvm instance
1340 * @log: slot id and address to which we copy the log
1341 * @is_dirty: set to '1' if any dirty pages were found
1342 * @memslot: set to the associated memslot, always valid on success
1344 int kvm_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log,
1345 int *is_dirty, struct kvm_memory_slot **memslot)
1347 struct kvm_memslots *slots;
1350 unsigned long any = 0;
1355 as_id = log->slot >> 16;
1356 id = (u16)log->slot;
1357 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1360 slots = __kvm_memslots(kvm, as_id);
1361 *memslot = id_to_memslot(slots, id);
1362 if (!(*memslot) || !(*memslot)->dirty_bitmap)
1365 kvm_arch_sync_dirty_log(kvm, *memslot);
1367 n = kvm_dirty_bitmap_bytes(*memslot);
1369 for (i = 0; !any && i < n/sizeof(long); ++i)
1370 any = (*memslot)->dirty_bitmap[i];
1372 if (copy_to_user(log->dirty_bitmap, (*memslot)->dirty_bitmap, n))
1379 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1381 #else /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
1383 * kvm_get_dirty_log_protect - get a snapshot of dirty pages
1384 * and reenable dirty page tracking for the corresponding pages.
1385 * @kvm: pointer to kvm instance
1386 * @log: slot id and address to which we copy the log
1388 * We need to keep it in mind that VCPU threads can write to the bitmap
1389 * concurrently. So, to avoid losing track of dirty pages we keep the
1392 * 1. Take a snapshot of the bit and clear it if needed.
1393 * 2. Write protect the corresponding page.
1394 * 3. Copy the snapshot to the userspace.
1395 * 4. Upon return caller flushes TLB's if needed.
1397 * Between 2 and 4, the guest may write to the page using the remaining TLB
1398 * entry. This is not a problem because the page is reported dirty using
1399 * the snapshot taken before and step 4 ensures that writes done after
1400 * exiting to userspace will be logged for the next call.
1403 static int kvm_get_dirty_log_protect(struct kvm *kvm, struct kvm_dirty_log *log)
1405 struct kvm_memslots *slots;
1406 struct kvm_memory_slot *memslot;
1409 unsigned long *dirty_bitmap;
1410 unsigned long *dirty_bitmap_buffer;
1413 as_id = log->slot >> 16;
1414 id = (u16)log->slot;
1415 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1418 slots = __kvm_memslots(kvm, as_id);
1419 memslot = id_to_memslot(slots, id);
1420 if (!memslot || !memslot->dirty_bitmap)
1423 dirty_bitmap = memslot->dirty_bitmap;
1425 kvm_arch_sync_dirty_log(kvm, memslot);
1427 n = kvm_dirty_bitmap_bytes(memslot);
1429 if (kvm->manual_dirty_log_protect) {
1431 * Unlike kvm_get_dirty_log, we always return false in *flush,
1432 * because no flush is needed until KVM_CLEAR_DIRTY_LOG. There
1433 * is some code duplication between this function and
1434 * kvm_get_dirty_log, but hopefully all architecture
1435 * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log
1436 * can be eliminated.
1438 dirty_bitmap_buffer = dirty_bitmap;
1440 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
1441 memset(dirty_bitmap_buffer, 0, n);
1443 spin_lock(&kvm->mmu_lock);
1444 for (i = 0; i < n / sizeof(long); i++) {
1448 if (!dirty_bitmap[i])
1452 mask = xchg(&dirty_bitmap[i], 0);
1453 dirty_bitmap_buffer[i] = mask;
1455 offset = i * BITS_PER_LONG;
1456 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1459 spin_unlock(&kvm->mmu_lock);
1463 kvm_arch_flush_remote_tlbs_memslot(kvm, memslot);
1465 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1472 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
1473 * @kvm: kvm instance
1474 * @log: slot id and address to which we copy the log
1476 * Steps 1-4 below provide general overview of dirty page logging. See
1477 * kvm_get_dirty_log_protect() function description for additional details.
1479 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
1480 * always flush the TLB (step 4) even if previous step failed and the dirty
1481 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
1482 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
1483 * writes will be marked dirty for next log read.
1485 * 1. Take a snapshot of the bit and clear it if needed.
1486 * 2. Write protect the corresponding page.
1487 * 3. Copy the snapshot to the userspace.
1488 * 4. Flush TLB's if needed.
1490 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1491 struct kvm_dirty_log *log)
1495 mutex_lock(&kvm->slots_lock);
1497 r = kvm_get_dirty_log_protect(kvm, log);
1499 mutex_unlock(&kvm->slots_lock);
1504 * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap
1505 * and reenable dirty page tracking for the corresponding pages.
1506 * @kvm: pointer to kvm instance
1507 * @log: slot id and address from which to fetch the bitmap of dirty pages
1509 static int kvm_clear_dirty_log_protect(struct kvm *kvm,
1510 struct kvm_clear_dirty_log *log)
1512 struct kvm_memslots *slots;
1513 struct kvm_memory_slot *memslot;
1517 unsigned long *dirty_bitmap;
1518 unsigned long *dirty_bitmap_buffer;
1521 as_id = log->slot >> 16;
1522 id = (u16)log->slot;
1523 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1526 if (log->first_page & 63)
1529 slots = __kvm_memslots(kvm, as_id);
1530 memslot = id_to_memslot(slots, id);
1531 if (!memslot || !memslot->dirty_bitmap)
1534 dirty_bitmap = memslot->dirty_bitmap;
1536 n = ALIGN(log->num_pages, BITS_PER_LONG) / 8;
1538 if (log->first_page > memslot->npages ||
1539 log->num_pages > memslot->npages - log->first_page ||
1540 (log->num_pages < memslot->npages - log->first_page && (log->num_pages & 63)))
1543 kvm_arch_sync_dirty_log(kvm, memslot);
1546 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
1547 if (copy_from_user(dirty_bitmap_buffer, log->dirty_bitmap, n))
1550 spin_lock(&kvm->mmu_lock);
1551 for (offset = log->first_page, i = offset / BITS_PER_LONG,
1552 n = DIV_ROUND_UP(log->num_pages, BITS_PER_LONG); n--;
1553 i++, offset += BITS_PER_LONG) {
1554 unsigned long mask = *dirty_bitmap_buffer++;
1555 atomic_long_t *p = (atomic_long_t *) &dirty_bitmap[i];
1559 mask &= atomic_long_fetch_andnot(mask, p);
1562 * mask contains the bits that really have been cleared. This
1563 * never includes any bits beyond the length of the memslot (if
1564 * the length is not aligned to 64 pages), therefore it is not
1565 * a problem if userspace sets them in log->dirty_bitmap.
1569 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1573 spin_unlock(&kvm->mmu_lock);
1576 kvm_arch_flush_remote_tlbs_memslot(kvm, memslot);
1581 static int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm,
1582 struct kvm_clear_dirty_log *log)
1586 mutex_lock(&kvm->slots_lock);
1588 r = kvm_clear_dirty_log_protect(kvm, log);
1590 mutex_unlock(&kvm->slots_lock);
1593 #endif /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
1595 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1597 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1599 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1601 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
1603 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn);
1606 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1608 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1610 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1611 memslot->flags & KVM_MEMSLOT_INVALID)
1616 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1618 unsigned long kvm_host_page_size(struct kvm_vcpu *vcpu, gfn_t gfn)
1620 struct vm_area_struct *vma;
1621 unsigned long addr, size;
1625 addr = kvm_vcpu_gfn_to_hva_prot(vcpu, gfn, NULL);
1626 if (kvm_is_error_hva(addr))
1629 down_read(¤t->mm->mmap_sem);
1630 vma = find_vma(current->mm, addr);
1634 size = vma_kernel_pagesize(vma);
1637 up_read(¤t->mm->mmap_sem);
1642 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1644 return slot->flags & KVM_MEM_READONLY;
1647 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1648 gfn_t *nr_pages, bool write)
1650 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1651 return KVM_HVA_ERR_BAD;
1653 if (memslot_is_readonly(slot) && write)
1654 return KVM_HVA_ERR_RO_BAD;
1657 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1659 return __gfn_to_hva_memslot(slot, gfn);
1662 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1665 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1668 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1671 return gfn_to_hva_many(slot, gfn, NULL);
1673 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1675 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1677 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1679 EXPORT_SYMBOL_GPL(gfn_to_hva);
1681 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
1683 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
1685 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
1688 * Return the hva of a @gfn and the R/W attribute if possible.
1690 * @slot: the kvm_memory_slot which contains @gfn
1691 * @gfn: the gfn to be translated
1692 * @writable: used to return the read/write attribute of the @slot if the hva
1693 * is valid and @writable is not NULL
1695 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1696 gfn_t gfn, bool *writable)
1698 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1700 if (!kvm_is_error_hva(hva) && writable)
1701 *writable = !memslot_is_readonly(slot);
1706 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1708 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1710 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1713 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
1715 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1717 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1720 static inline int check_user_page_hwpoison(unsigned long addr)
1722 int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
1724 rc = get_user_pages(addr, 1, flags, NULL, NULL);
1725 return rc == -EHWPOISON;
1729 * The fast path to get the writable pfn which will be stored in @pfn,
1730 * true indicates success, otherwise false is returned. It's also the
1731 * only part that runs if we can in atomic context.
1733 static bool hva_to_pfn_fast(unsigned long addr, bool write_fault,
1734 bool *writable, kvm_pfn_t *pfn)
1736 struct page *page[1];
1740 * Fast pin a writable pfn only if it is a write fault request
1741 * or the caller allows to map a writable pfn for a read fault
1744 if (!(write_fault || writable))
1747 npages = __get_user_pages_fast(addr, 1, 1, page);
1749 *pfn = page_to_pfn(page[0]);
1760 * The slow path to get the pfn of the specified host virtual address,
1761 * 1 indicates success, -errno is returned if error is detected.
1763 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1764 bool *writable, kvm_pfn_t *pfn)
1766 unsigned int flags = FOLL_HWPOISON;
1773 *writable = write_fault;
1776 flags |= FOLL_WRITE;
1778 flags |= FOLL_NOWAIT;
1780 npages = get_user_pages_unlocked(addr, 1, &page, flags);
1784 /* map read fault as writable if possible */
1785 if (unlikely(!write_fault) && writable) {
1788 if (__get_user_pages_fast(addr, 1, 1, &wpage) == 1) {
1794 *pfn = page_to_pfn(page);
1798 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1800 if (unlikely(!(vma->vm_flags & VM_READ)))
1803 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1809 static int hva_to_pfn_remapped(struct vm_area_struct *vma,
1810 unsigned long addr, bool *async,
1811 bool write_fault, bool *writable,
1817 r = follow_pfn(vma, addr, &pfn);
1820 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
1821 * not call the fault handler, so do it here.
1823 bool unlocked = false;
1824 r = fixup_user_fault(current, current->mm, addr,
1825 (write_fault ? FAULT_FLAG_WRITE : 0),
1832 r = follow_pfn(vma, addr, &pfn);
1842 * Get a reference here because callers of *hva_to_pfn* and
1843 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
1844 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
1845 * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will
1846 * simply do nothing for reserved pfns.
1848 * Whoever called remap_pfn_range is also going to call e.g.
1849 * unmap_mapping_range before the underlying pages are freed,
1850 * causing a call to our MMU notifier.
1859 * Pin guest page in memory and return its pfn.
1860 * @addr: host virtual address which maps memory to the guest
1861 * @atomic: whether this function can sleep
1862 * @async: whether this function need to wait IO complete if the
1863 * host page is not in the memory
1864 * @write_fault: whether we should get a writable host page
1865 * @writable: whether it allows to map a writable host page for !@write_fault
1867 * The function will map a writable host page for these two cases:
1868 * 1): @write_fault = true
1869 * 2): @write_fault = false && @writable, @writable will tell the caller
1870 * whether the mapping is writable.
1872 static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1873 bool write_fault, bool *writable)
1875 struct vm_area_struct *vma;
1879 /* we can do it either atomically or asynchronously, not both */
1880 BUG_ON(atomic && async);
1882 if (hva_to_pfn_fast(addr, write_fault, writable, &pfn))
1886 return KVM_PFN_ERR_FAULT;
1888 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1892 down_read(¤t->mm->mmap_sem);
1893 if (npages == -EHWPOISON ||
1894 (!async && check_user_page_hwpoison(addr))) {
1895 pfn = KVM_PFN_ERR_HWPOISON;
1900 vma = find_vma_intersection(current->mm, addr, addr + 1);
1903 pfn = KVM_PFN_ERR_FAULT;
1904 else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
1905 r = hva_to_pfn_remapped(vma, addr, async, write_fault, writable, &pfn);
1909 pfn = KVM_PFN_ERR_FAULT;
1911 if (async && vma_is_valid(vma, write_fault))
1913 pfn = KVM_PFN_ERR_FAULT;
1916 up_read(¤t->mm->mmap_sem);
1920 kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn,
1921 bool atomic, bool *async, bool write_fault,
1924 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1926 if (addr == KVM_HVA_ERR_RO_BAD) {
1929 return KVM_PFN_ERR_RO_FAULT;
1932 if (kvm_is_error_hva(addr)) {
1935 return KVM_PFN_NOSLOT;
1938 /* Do not map writable pfn in the readonly memslot. */
1939 if (writable && memslot_is_readonly(slot)) {
1944 return hva_to_pfn(addr, atomic, async, write_fault,
1947 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
1949 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1952 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
1953 write_fault, writable);
1955 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1957 kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1959 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1961 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
1963 kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1965 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1967 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1969 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
1971 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1973 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
1975 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1977 return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
1979 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1981 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
1983 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1985 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
1987 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1988 struct page **pages, int nr_pages)
1993 addr = gfn_to_hva_many(slot, gfn, &entry);
1994 if (kvm_is_error_hva(addr))
1997 if (entry < nr_pages)
2000 return __get_user_pages_fast(addr, nr_pages, 1, pages);
2002 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
2004 static struct page *kvm_pfn_to_page(kvm_pfn_t pfn)
2006 if (is_error_noslot_pfn(pfn))
2007 return KVM_ERR_PTR_BAD_PAGE;
2009 if (kvm_is_reserved_pfn(pfn)) {
2011 return KVM_ERR_PTR_BAD_PAGE;
2014 return pfn_to_page(pfn);
2017 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
2021 pfn = gfn_to_pfn(kvm, gfn);
2023 return kvm_pfn_to_page(pfn);
2025 EXPORT_SYMBOL_GPL(gfn_to_page);
2027 void kvm_release_pfn(kvm_pfn_t pfn, bool dirty, struct gfn_to_pfn_cache *cache)
2033 cache->pfn = cache->gfn = 0;
2036 kvm_release_pfn_dirty(pfn);
2038 kvm_release_pfn_clean(pfn);
2041 static void kvm_cache_gfn_to_pfn(struct kvm_memory_slot *slot, gfn_t gfn,
2042 struct gfn_to_pfn_cache *cache, u64 gen)
2044 kvm_release_pfn(cache->pfn, cache->dirty, cache);
2046 cache->pfn = gfn_to_pfn_memslot(slot, gfn);
2048 cache->dirty = false;
2049 cache->generation = gen;
2052 static int __kvm_map_gfn(struct kvm_memslots *slots, gfn_t gfn,
2053 struct kvm_host_map *map,
2054 struct gfn_to_pfn_cache *cache,
2059 struct page *page = KVM_UNMAPPED_PAGE;
2060 struct kvm_memory_slot *slot = __gfn_to_memslot(slots, gfn);
2061 u64 gen = slots->generation;
2067 if (!cache->pfn || cache->gfn != gfn ||
2068 cache->generation != gen) {
2071 kvm_cache_gfn_to_pfn(slot, gfn, cache, gen);
2077 pfn = gfn_to_pfn_memslot(slot, gfn);
2079 if (is_error_noslot_pfn(pfn))
2082 if (pfn_valid(pfn)) {
2083 page = pfn_to_page(pfn);
2085 hva = kmap_atomic(page);
2088 #ifdef CONFIG_HAS_IOMEM
2089 } else if (!atomic) {
2090 hva = memremap(pfn_to_hpa(pfn), PAGE_SIZE, MEMREMAP_WB);
2107 int kvm_map_gfn(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map,
2108 struct gfn_to_pfn_cache *cache, bool atomic)
2110 return __kvm_map_gfn(kvm_memslots(vcpu->kvm), gfn, map,
2113 EXPORT_SYMBOL_GPL(kvm_map_gfn);
2115 int kvm_vcpu_map(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map)
2117 return __kvm_map_gfn(kvm_vcpu_memslots(vcpu), gfn, map,
2120 EXPORT_SYMBOL_GPL(kvm_vcpu_map);
2122 static void __kvm_unmap_gfn(struct kvm_memory_slot *memslot,
2123 struct kvm_host_map *map,
2124 struct gfn_to_pfn_cache *cache,
2125 bool dirty, bool atomic)
2133 if (map->page != KVM_UNMAPPED_PAGE) {
2135 kunmap_atomic(map->hva);
2139 #ifdef CONFIG_HAS_IOMEM
2143 WARN_ONCE(1, "Unexpected unmapping in atomic context");
2147 mark_page_dirty_in_slot(memslot, map->gfn);
2150 cache->dirty |= dirty;
2152 kvm_release_pfn(map->pfn, dirty, NULL);
2158 int kvm_unmap_gfn(struct kvm_vcpu *vcpu, struct kvm_host_map *map,
2159 struct gfn_to_pfn_cache *cache, bool dirty, bool atomic)
2161 __kvm_unmap_gfn(gfn_to_memslot(vcpu->kvm, map->gfn), map,
2162 cache, dirty, atomic);
2165 EXPORT_SYMBOL_GPL(kvm_unmap_gfn);
2167 void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map, bool dirty)
2169 __kvm_unmap_gfn(kvm_vcpu_gfn_to_memslot(vcpu, map->gfn), map, NULL,
2172 EXPORT_SYMBOL_GPL(kvm_vcpu_unmap);
2174 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
2178 pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
2180 return kvm_pfn_to_page(pfn);
2182 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
2184 void kvm_release_page_clean(struct page *page)
2186 WARN_ON(is_error_page(page));
2188 kvm_release_pfn_clean(page_to_pfn(page));
2190 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
2192 void kvm_release_pfn_clean(kvm_pfn_t pfn)
2194 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
2195 put_page(pfn_to_page(pfn));
2197 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
2199 void kvm_release_page_dirty(struct page *page)
2201 WARN_ON(is_error_page(page));
2203 kvm_release_pfn_dirty(page_to_pfn(page));
2205 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
2207 void kvm_release_pfn_dirty(kvm_pfn_t pfn)
2209 kvm_set_pfn_dirty(pfn);
2210 kvm_release_pfn_clean(pfn);
2212 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
2214 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
2216 if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn))
2217 SetPageDirty(pfn_to_page(pfn));
2219 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
2221 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
2223 if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn))
2224 mark_page_accessed(pfn_to_page(pfn));
2226 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
2228 void kvm_get_pfn(kvm_pfn_t pfn)
2230 if (!kvm_is_reserved_pfn(pfn))
2231 get_page(pfn_to_page(pfn));
2233 EXPORT_SYMBOL_GPL(kvm_get_pfn);
2235 static int next_segment(unsigned long len, int offset)
2237 if (len > PAGE_SIZE - offset)
2238 return PAGE_SIZE - offset;
2243 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
2244 void *data, int offset, int len)
2249 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
2250 if (kvm_is_error_hva(addr))
2252 r = __copy_from_user(data, (void __user *)addr + offset, len);
2258 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
2261 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2263 return __kvm_read_guest_page(slot, gfn, data, offset, len);
2265 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
2267 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
2268 int offset, int len)
2270 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2272 return __kvm_read_guest_page(slot, gfn, data, offset, len);
2274 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
2276 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
2278 gfn_t gfn = gpa >> PAGE_SHIFT;
2280 int offset = offset_in_page(gpa);
2283 while ((seg = next_segment(len, offset)) != 0) {
2284 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
2294 EXPORT_SYMBOL_GPL(kvm_read_guest);
2296 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
2298 gfn_t gfn = gpa >> PAGE_SHIFT;
2300 int offset = offset_in_page(gpa);
2303 while ((seg = next_segment(len, offset)) != 0) {
2304 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
2314 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
2316 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
2317 void *data, int offset, unsigned long len)
2322 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
2323 if (kvm_is_error_hva(addr))
2325 pagefault_disable();
2326 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
2333 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
2334 void *data, unsigned long len)
2336 gfn_t gfn = gpa >> PAGE_SHIFT;
2337 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2338 int offset = offset_in_page(gpa);
2340 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
2342 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
2344 static int __kvm_write_guest_page(struct kvm_memory_slot *memslot, gfn_t gfn,
2345 const void *data, int offset, int len)
2350 addr = gfn_to_hva_memslot(memslot, gfn);
2351 if (kvm_is_error_hva(addr))
2353 r = __copy_to_user((void __user *)addr + offset, data, len);
2356 mark_page_dirty_in_slot(memslot, gfn);
2360 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
2361 const void *data, int offset, int len)
2363 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2365 return __kvm_write_guest_page(slot, gfn, data, offset, len);
2367 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
2369 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
2370 const void *data, int offset, int len)
2372 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2374 return __kvm_write_guest_page(slot, gfn, data, offset, len);
2376 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
2378 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
2381 gfn_t gfn = gpa >> PAGE_SHIFT;
2383 int offset = offset_in_page(gpa);
2386 while ((seg = next_segment(len, offset)) != 0) {
2387 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
2397 EXPORT_SYMBOL_GPL(kvm_write_guest);
2399 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
2402 gfn_t gfn = gpa >> PAGE_SHIFT;
2404 int offset = offset_in_page(gpa);
2407 while ((seg = next_segment(len, offset)) != 0) {
2408 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
2418 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
2420 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
2421 struct gfn_to_hva_cache *ghc,
2422 gpa_t gpa, unsigned long len)
2424 int offset = offset_in_page(gpa);
2425 gfn_t start_gfn = gpa >> PAGE_SHIFT;
2426 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
2427 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
2428 gfn_t nr_pages_avail;
2430 /* Update ghc->generation before performing any error checks. */
2431 ghc->generation = slots->generation;
2433 if (start_gfn > end_gfn) {
2434 ghc->hva = KVM_HVA_ERR_BAD;
2439 * If the requested region crosses two memslots, we still
2440 * verify that the entire region is valid here.
2442 for ( ; start_gfn <= end_gfn; start_gfn += nr_pages_avail) {
2443 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
2444 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
2446 if (kvm_is_error_hva(ghc->hva))
2450 /* Use the slow path for cross page reads and writes. */
2451 if (nr_pages_needed == 1)
2454 ghc->memslot = NULL;
2461 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2462 gpa_t gpa, unsigned long len)
2464 struct kvm_memslots *slots = kvm_memslots(kvm);
2465 return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len);
2467 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
2469 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2470 void *data, unsigned int offset,
2473 struct kvm_memslots *slots = kvm_memslots(kvm);
2475 gpa_t gpa = ghc->gpa + offset;
2477 BUG_ON(len + offset > ghc->len);
2479 if (slots->generation != ghc->generation) {
2480 if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len))
2484 if (kvm_is_error_hva(ghc->hva))
2487 if (unlikely(!ghc->memslot))
2488 return kvm_write_guest(kvm, gpa, data, len);
2490 r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
2493 mark_page_dirty_in_slot(ghc->memslot, gpa >> PAGE_SHIFT);
2497 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached);
2499 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2500 void *data, unsigned long len)
2502 return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len);
2504 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
2506 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2507 void *data, unsigned long len)
2509 struct kvm_memslots *slots = kvm_memslots(kvm);
2512 BUG_ON(len > ghc->len);
2514 if (slots->generation != ghc->generation) {
2515 if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len))
2519 if (kvm_is_error_hva(ghc->hva))
2522 if (unlikely(!ghc->memslot))
2523 return kvm_read_guest(kvm, ghc->gpa, data, len);
2525 r = __copy_from_user(data, (void __user *)ghc->hva, len);
2531 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
2533 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
2535 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
2537 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
2539 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
2541 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
2543 gfn_t gfn = gpa >> PAGE_SHIFT;
2545 int offset = offset_in_page(gpa);
2548 while ((seg = next_segment(len, offset)) != 0) {
2549 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
2558 EXPORT_SYMBOL_GPL(kvm_clear_guest);
2560 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot,
2563 if (memslot && memslot->dirty_bitmap) {
2564 unsigned long rel_gfn = gfn - memslot->base_gfn;
2566 set_bit_le(rel_gfn, memslot->dirty_bitmap);
2570 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
2572 struct kvm_memory_slot *memslot;
2574 memslot = gfn_to_memslot(kvm, gfn);
2575 mark_page_dirty_in_slot(memslot, gfn);
2577 EXPORT_SYMBOL_GPL(mark_page_dirty);
2579 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
2581 struct kvm_memory_slot *memslot;
2583 memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2584 mark_page_dirty_in_slot(memslot, gfn);
2586 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
2588 void kvm_sigset_activate(struct kvm_vcpu *vcpu)
2590 if (!vcpu->sigset_active)
2594 * This does a lockless modification of ->real_blocked, which is fine
2595 * because, only current can change ->real_blocked and all readers of
2596 * ->real_blocked don't care as long ->real_blocked is always a subset
2599 sigprocmask(SIG_SETMASK, &vcpu->sigset, ¤t->real_blocked);
2602 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu)
2604 if (!vcpu->sigset_active)
2607 sigprocmask(SIG_SETMASK, ¤t->real_blocked, NULL);
2608 sigemptyset(¤t->real_blocked);
2611 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
2613 unsigned int old, val, grow, grow_start;
2615 old = val = vcpu->halt_poll_ns;
2616 grow_start = READ_ONCE(halt_poll_ns_grow_start);
2617 grow = READ_ONCE(halt_poll_ns_grow);
2622 if (val < grow_start)
2625 if (val > halt_poll_ns)
2628 vcpu->halt_poll_ns = val;
2630 trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
2633 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
2635 unsigned int old, val, shrink;
2637 old = val = vcpu->halt_poll_ns;
2638 shrink = READ_ONCE(halt_poll_ns_shrink);
2644 vcpu->halt_poll_ns = val;
2645 trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
2648 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
2651 int idx = srcu_read_lock(&vcpu->kvm->srcu);
2653 if (kvm_arch_vcpu_runnable(vcpu)) {
2654 kvm_make_request(KVM_REQ_UNHALT, vcpu);
2657 if (kvm_cpu_has_pending_timer(vcpu))
2659 if (signal_pending(current))
2664 srcu_read_unlock(&vcpu->kvm->srcu, idx);
2669 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2671 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
2674 DECLARE_SWAITQUEUE(wait);
2675 bool waited = false;
2678 kvm_arch_vcpu_blocking(vcpu);
2680 start = cur = ktime_get();
2681 if (vcpu->halt_poll_ns && !kvm_arch_no_poll(vcpu)) {
2682 ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns);
2684 ++vcpu->stat.halt_attempted_poll;
2687 * This sets KVM_REQ_UNHALT if an interrupt
2690 if (kvm_vcpu_check_block(vcpu) < 0) {
2691 ++vcpu->stat.halt_successful_poll;
2692 if (!vcpu_valid_wakeup(vcpu))
2693 ++vcpu->stat.halt_poll_invalid;
2697 } while (single_task_running() && ktime_before(cur, stop));
2701 prepare_to_swait_exclusive(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
2703 if (kvm_vcpu_check_block(vcpu) < 0)
2710 finish_swait(&vcpu->wq, &wait);
2713 kvm_arch_vcpu_unblocking(vcpu);
2714 block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
2716 if (!kvm_arch_no_poll(vcpu)) {
2717 if (!vcpu_valid_wakeup(vcpu)) {
2718 shrink_halt_poll_ns(vcpu);
2719 } else if (halt_poll_ns) {
2720 if (block_ns <= vcpu->halt_poll_ns)
2722 /* we had a long block, shrink polling */
2723 else if (vcpu->halt_poll_ns && block_ns > halt_poll_ns)
2724 shrink_halt_poll_ns(vcpu);
2725 /* we had a short halt and our poll time is too small */
2726 else if (vcpu->halt_poll_ns < halt_poll_ns &&
2727 block_ns < halt_poll_ns)
2728 grow_halt_poll_ns(vcpu);
2730 vcpu->halt_poll_ns = 0;
2734 trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu));
2735 kvm_arch_vcpu_block_finish(vcpu);
2737 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
2739 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
2741 struct swait_queue_head *wqp;
2743 wqp = kvm_arch_vcpu_wq(vcpu);
2744 if (swq_has_sleeper(wqp)) {
2746 WRITE_ONCE(vcpu->ready, true);
2747 ++vcpu->stat.halt_wakeup;
2753 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
2757 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2759 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
2762 int cpu = vcpu->cpu;
2764 if (kvm_vcpu_wake_up(vcpu))
2768 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
2769 if (kvm_arch_vcpu_should_kick(vcpu))
2770 smp_send_reschedule(cpu);
2773 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
2774 #endif /* !CONFIG_S390 */
2776 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
2779 struct task_struct *task = NULL;
2783 pid = rcu_dereference(target->pid);
2785 task = get_pid_task(pid, PIDTYPE_PID);
2789 ret = yield_to(task, 1);
2790 put_task_struct(task);
2794 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
2797 * Helper that checks whether a VCPU is eligible for directed yield.
2798 * Most eligible candidate to yield is decided by following heuristics:
2800 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2801 * (preempted lock holder), indicated by @in_spin_loop.
2802 * Set at the beiginning and cleared at the end of interception/PLE handler.
2804 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2805 * chance last time (mostly it has become eligible now since we have probably
2806 * yielded to lockholder in last iteration. This is done by toggling
2807 * @dy_eligible each time a VCPU checked for eligibility.)
2809 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2810 * to preempted lock-holder could result in wrong VCPU selection and CPU
2811 * burning. Giving priority for a potential lock-holder increases lock
2814 * Since algorithm is based on heuristics, accessing another VCPU data without
2815 * locking does not harm. It may result in trying to yield to same VCPU, fail
2816 * and continue with next VCPU and so on.
2818 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
2820 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2823 eligible = !vcpu->spin_loop.in_spin_loop ||
2824 vcpu->spin_loop.dy_eligible;
2826 if (vcpu->spin_loop.in_spin_loop)
2827 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
2836 * Unlike kvm_arch_vcpu_runnable, this function is called outside
2837 * a vcpu_load/vcpu_put pair. However, for most architectures
2838 * kvm_arch_vcpu_runnable does not require vcpu_load.
2840 bool __weak kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
2842 return kvm_arch_vcpu_runnable(vcpu);
2845 static bool vcpu_dy_runnable(struct kvm_vcpu *vcpu)
2847 if (kvm_arch_dy_runnable(vcpu))
2850 #ifdef CONFIG_KVM_ASYNC_PF
2851 if (!list_empty_careful(&vcpu->async_pf.done))
2858 void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode)
2860 struct kvm *kvm = me->kvm;
2861 struct kvm_vcpu *vcpu;
2862 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
2868 kvm_vcpu_set_in_spin_loop(me, true);
2870 * We boost the priority of a VCPU that is runnable but not
2871 * currently running, because it got preempted by something
2872 * else and called schedule in __vcpu_run. Hopefully that
2873 * VCPU is holding the lock that we need and will release it.
2874 * We approximate round-robin by starting at the last boosted VCPU.
2876 for (pass = 0; pass < 2 && !yielded && try; pass++) {
2877 kvm_for_each_vcpu(i, vcpu, kvm) {
2878 if (!pass && i <= last_boosted_vcpu) {
2879 i = last_boosted_vcpu;
2881 } else if (pass && i > last_boosted_vcpu)
2883 if (!READ_ONCE(vcpu->ready))
2887 if (swait_active(&vcpu->wq) && !vcpu_dy_runnable(vcpu))
2889 if (READ_ONCE(vcpu->preempted) && yield_to_kernel_mode &&
2890 !kvm_arch_vcpu_in_kernel(vcpu))
2892 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
2895 yielded = kvm_vcpu_yield_to(vcpu);
2897 kvm->last_boosted_vcpu = i;
2899 } else if (yielded < 0) {
2906 kvm_vcpu_set_in_spin_loop(me, false);
2908 /* Ensure vcpu is not eligible during next spinloop */
2909 kvm_vcpu_set_dy_eligible(me, false);
2911 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
2913 static vm_fault_t kvm_vcpu_fault(struct vm_fault *vmf)
2915 struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data;
2918 if (vmf->pgoff == 0)
2919 page = virt_to_page(vcpu->run);
2921 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
2922 page = virt_to_page(vcpu->arch.pio_data);
2924 #ifdef CONFIG_KVM_MMIO
2925 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
2926 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
2929 return kvm_arch_vcpu_fault(vcpu, vmf);
2935 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
2936 .fault = kvm_vcpu_fault,
2939 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2941 vma->vm_ops = &kvm_vcpu_vm_ops;
2945 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2947 struct kvm_vcpu *vcpu = filp->private_data;
2949 debugfs_remove_recursive(vcpu->debugfs_dentry);
2950 kvm_put_kvm(vcpu->kvm);
2954 static struct file_operations kvm_vcpu_fops = {
2955 .release = kvm_vcpu_release,
2956 .unlocked_ioctl = kvm_vcpu_ioctl,
2957 .mmap = kvm_vcpu_mmap,
2958 .llseek = noop_llseek,
2959 KVM_COMPAT(kvm_vcpu_compat_ioctl),
2963 * Allocates an inode for the vcpu.
2965 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2967 char name[8 + 1 + ITOA_MAX_LEN + 1];
2969 snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id);
2970 return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
2973 static void kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
2975 #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS
2976 char dir_name[ITOA_MAX_LEN * 2];
2978 if (!debugfs_initialized())
2981 snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
2982 vcpu->debugfs_dentry = debugfs_create_dir(dir_name,
2983 vcpu->kvm->debugfs_dentry);
2985 kvm_arch_create_vcpu_debugfs(vcpu);
2990 * Creates some virtual cpus. Good luck creating more than one.
2992 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
2995 struct kvm_vcpu *vcpu;
2998 if (id >= KVM_MAX_VCPU_ID)
3001 mutex_lock(&kvm->lock);
3002 if (kvm->created_vcpus == KVM_MAX_VCPUS) {
3003 mutex_unlock(&kvm->lock);
3007 kvm->created_vcpus++;
3008 mutex_unlock(&kvm->lock);
3010 r = kvm_arch_vcpu_precreate(kvm, id);
3012 goto vcpu_decrement;
3014 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
3017 goto vcpu_decrement;
3020 BUILD_BUG_ON(sizeof(struct kvm_run) > PAGE_SIZE);
3021 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
3026 vcpu->run = page_address(page);
3028 kvm_vcpu_init(vcpu, kvm, id);
3030 r = kvm_arch_vcpu_create(vcpu);
3032 goto vcpu_free_run_page;
3034 kvm_create_vcpu_debugfs(vcpu);
3036 mutex_lock(&kvm->lock);
3037 if (kvm_get_vcpu_by_id(kvm, id)) {
3039 goto unlock_vcpu_destroy;
3042 vcpu->vcpu_idx = atomic_read(&kvm->online_vcpus);
3043 BUG_ON(kvm->vcpus[vcpu->vcpu_idx]);
3045 /* Now it's all set up, let userspace reach it */
3047 r = create_vcpu_fd(vcpu);
3049 kvm_put_kvm_no_destroy(kvm);
3050 goto unlock_vcpu_destroy;
3053 kvm->vcpus[vcpu->vcpu_idx] = vcpu;
3056 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
3057 * before kvm->online_vcpu's incremented value.
3060 atomic_inc(&kvm->online_vcpus);
3062 mutex_unlock(&kvm->lock);
3063 kvm_arch_vcpu_postcreate(vcpu);
3066 unlock_vcpu_destroy:
3067 mutex_unlock(&kvm->lock);
3068 debugfs_remove_recursive(vcpu->debugfs_dentry);
3069 kvm_arch_vcpu_destroy(vcpu);
3071 free_page((unsigned long)vcpu->run);
3073 kmem_cache_free(kvm_vcpu_cache, vcpu);
3075 mutex_lock(&kvm->lock);
3076 kvm->created_vcpus--;
3077 mutex_unlock(&kvm->lock);
3081 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
3084 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
3085 vcpu->sigset_active = 1;
3086 vcpu->sigset = *sigset;
3088 vcpu->sigset_active = 0;
3092 static long kvm_vcpu_ioctl(struct file *filp,
3093 unsigned int ioctl, unsigned long arg)
3095 struct kvm_vcpu *vcpu = filp->private_data;
3096 void __user *argp = (void __user *)arg;
3098 struct kvm_fpu *fpu = NULL;
3099 struct kvm_sregs *kvm_sregs = NULL;
3101 if (vcpu->kvm->mm != current->mm)
3104 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
3108 * Some architectures have vcpu ioctls that are asynchronous to vcpu
3109 * execution; mutex_lock() would break them.
3111 r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg);
3112 if (r != -ENOIOCTLCMD)
3115 if (mutex_lock_killable(&vcpu->mutex))
3123 oldpid = rcu_access_pointer(vcpu->pid);
3124 if (unlikely(oldpid != task_pid(current))) {
3125 /* The thread running this VCPU changed. */
3128 r = kvm_arch_vcpu_run_pid_change(vcpu);
3132 newpid = get_task_pid(current, PIDTYPE_PID);
3133 rcu_assign_pointer(vcpu->pid, newpid);
3138 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
3139 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
3142 case KVM_GET_REGS: {
3143 struct kvm_regs *kvm_regs;
3146 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL_ACCOUNT);
3149 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
3153 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
3160 case KVM_SET_REGS: {
3161 struct kvm_regs *kvm_regs;
3164 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
3165 if (IS_ERR(kvm_regs)) {
3166 r = PTR_ERR(kvm_regs);
3169 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
3173 case KVM_GET_SREGS: {
3174 kvm_sregs = kzalloc(sizeof(struct kvm_sregs),
3175 GFP_KERNEL_ACCOUNT);
3179 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
3183 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
3188 case KVM_SET_SREGS: {
3189 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
3190 if (IS_ERR(kvm_sregs)) {
3191 r = PTR_ERR(kvm_sregs);
3195 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
3198 case KVM_GET_MP_STATE: {
3199 struct kvm_mp_state mp_state;
3201 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
3205 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
3210 case KVM_SET_MP_STATE: {
3211 struct kvm_mp_state mp_state;
3214 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
3216 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
3219 case KVM_TRANSLATE: {
3220 struct kvm_translation tr;
3223 if (copy_from_user(&tr, argp, sizeof(tr)))
3225 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
3229 if (copy_to_user(argp, &tr, sizeof(tr)))
3234 case KVM_SET_GUEST_DEBUG: {
3235 struct kvm_guest_debug dbg;
3238 if (copy_from_user(&dbg, argp, sizeof(dbg)))
3240 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
3243 case KVM_SET_SIGNAL_MASK: {
3244 struct kvm_signal_mask __user *sigmask_arg = argp;
3245 struct kvm_signal_mask kvm_sigmask;
3246 sigset_t sigset, *p;
3251 if (copy_from_user(&kvm_sigmask, argp,
3252 sizeof(kvm_sigmask)))
3255 if (kvm_sigmask.len != sizeof(sigset))
3258 if (copy_from_user(&sigset, sigmask_arg->sigset,
3263 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
3267 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL_ACCOUNT);
3271 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
3275 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
3281 fpu = memdup_user(argp, sizeof(*fpu));
3287 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
3291 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
3294 mutex_unlock(&vcpu->mutex);
3300 #ifdef CONFIG_KVM_COMPAT
3301 static long kvm_vcpu_compat_ioctl(struct file *filp,
3302 unsigned int ioctl, unsigned long arg)
3304 struct kvm_vcpu *vcpu = filp->private_data;
3305 void __user *argp = compat_ptr(arg);
3308 if (vcpu->kvm->mm != current->mm)
3312 case KVM_SET_SIGNAL_MASK: {
3313 struct kvm_signal_mask __user *sigmask_arg = argp;
3314 struct kvm_signal_mask kvm_sigmask;
3319 if (copy_from_user(&kvm_sigmask, argp,
3320 sizeof(kvm_sigmask)))
3323 if (kvm_sigmask.len != sizeof(compat_sigset_t))
3326 if (get_compat_sigset(&sigset, (void *)sigmask_arg->sigset))
3328 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
3330 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
3334 r = kvm_vcpu_ioctl(filp, ioctl, arg);
3342 static int kvm_device_mmap(struct file *filp, struct vm_area_struct *vma)
3344 struct kvm_device *dev = filp->private_data;
3347 return dev->ops->mmap(dev, vma);
3352 static int kvm_device_ioctl_attr(struct kvm_device *dev,
3353 int (*accessor)(struct kvm_device *dev,
3354 struct kvm_device_attr *attr),
3357 struct kvm_device_attr attr;
3362 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
3365 return accessor(dev, &attr);
3368 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
3371 struct kvm_device *dev = filp->private_data;
3373 if (dev->kvm->mm != current->mm)
3377 case KVM_SET_DEVICE_ATTR:
3378 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
3379 case KVM_GET_DEVICE_ATTR:
3380 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
3381 case KVM_HAS_DEVICE_ATTR:
3382 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
3384 if (dev->ops->ioctl)
3385 return dev->ops->ioctl(dev, ioctl, arg);
3391 static int kvm_device_release(struct inode *inode, struct file *filp)
3393 struct kvm_device *dev = filp->private_data;
3394 struct kvm *kvm = dev->kvm;
3396 if (dev->ops->release) {
3397 mutex_lock(&kvm->lock);
3398 list_del(&dev->vm_node);
3399 dev->ops->release(dev);
3400 mutex_unlock(&kvm->lock);
3407 static const struct file_operations kvm_device_fops = {
3408 .unlocked_ioctl = kvm_device_ioctl,
3409 .release = kvm_device_release,
3410 KVM_COMPAT(kvm_device_ioctl),
3411 .mmap = kvm_device_mmap,
3414 struct kvm_device *kvm_device_from_filp(struct file *filp)
3416 if (filp->f_op != &kvm_device_fops)
3419 return filp->private_data;
3422 static const struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
3423 #ifdef CONFIG_KVM_MPIC
3424 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
3425 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
3429 int kvm_register_device_ops(const struct kvm_device_ops *ops, u32 type)
3431 if (type >= ARRAY_SIZE(kvm_device_ops_table))
3434 if (kvm_device_ops_table[type] != NULL)
3437 kvm_device_ops_table[type] = ops;
3441 void kvm_unregister_device_ops(u32 type)
3443 if (kvm_device_ops_table[type] != NULL)
3444 kvm_device_ops_table[type] = NULL;
3447 static int kvm_ioctl_create_device(struct kvm *kvm,
3448 struct kvm_create_device *cd)
3450 const struct kvm_device_ops *ops = NULL;
3451 struct kvm_device *dev;
3452 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
3456 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
3459 type = array_index_nospec(cd->type, ARRAY_SIZE(kvm_device_ops_table));
3460 ops = kvm_device_ops_table[type];
3467 dev = kzalloc(sizeof(*dev), GFP_KERNEL_ACCOUNT);
3474 mutex_lock(&kvm->lock);
3475 ret = ops->create(dev, type);
3477 mutex_unlock(&kvm->lock);
3481 list_add(&dev->vm_node, &kvm->devices);
3482 mutex_unlock(&kvm->lock);
3488 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
3490 kvm_put_kvm_no_destroy(kvm);
3491 mutex_lock(&kvm->lock);
3492 list_del(&dev->vm_node);
3493 mutex_unlock(&kvm->lock);
3502 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
3505 case KVM_CAP_USER_MEMORY:
3506 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
3507 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
3508 case KVM_CAP_INTERNAL_ERROR_DATA:
3509 #ifdef CONFIG_HAVE_KVM_MSI
3510 case KVM_CAP_SIGNAL_MSI:
3512 #ifdef CONFIG_HAVE_KVM_IRQFD
3514 case KVM_CAP_IRQFD_RESAMPLE:
3516 case KVM_CAP_IOEVENTFD_ANY_LENGTH:
3517 case KVM_CAP_CHECK_EXTENSION_VM:
3518 case KVM_CAP_ENABLE_CAP_VM:
3520 #ifdef CONFIG_KVM_MMIO
3521 case KVM_CAP_COALESCED_MMIO:
3522 return KVM_COALESCED_MMIO_PAGE_OFFSET;
3523 case KVM_CAP_COALESCED_PIO:
3526 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3527 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2:
3528 return KVM_DIRTY_LOG_MANUAL_CAPS;
3530 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3531 case KVM_CAP_IRQ_ROUTING:
3532 return KVM_MAX_IRQ_ROUTES;
3534 #if KVM_ADDRESS_SPACE_NUM > 1
3535 case KVM_CAP_MULTI_ADDRESS_SPACE:
3536 return KVM_ADDRESS_SPACE_NUM;
3538 case KVM_CAP_NR_MEMSLOTS:
3539 return KVM_USER_MEM_SLOTS;
3543 return kvm_vm_ioctl_check_extension(kvm, arg);
3546 int __attribute__((weak)) kvm_vm_ioctl_enable_cap(struct kvm *kvm,
3547 struct kvm_enable_cap *cap)
3552 static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm,
3553 struct kvm_enable_cap *cap)
3556 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3557 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2: {
3558 u64 allowed_options = KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE;
3560 if (cap->args[0] & KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE)
3561 allowed_options = KVM_DIRTY_LOG_MANUAL_CAPS;
3563 if (cap->flags || (cap->args[0] & ~allowed_options))
3565 kvm->manual_dirty_log_protect = cap->args[0];
3570 return kvm_vm_ioctl_enable_cap(kvm, cap);
3574 static long kvm_vm_ioctl(struct file *filp,
3575 unsigned int ioctl, unsigned long arg)
3577 struct kvm *kvm = filp->private_data;
3578 void __user *argp = (void __user *)arg;
3581 if (kvm->mm != current->mm)
3584 case KVM_CREATE_VCPU:
3585 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
3587 case KVM_ENABLE_CAP: {
3588 struct kvm_enable_cap cap;
3591 if (copy_from_user(&cap, argp, sizeof(cap)))
3593 r = kvm_vm_ioctl_enable_cap_generic(kvm, &cap);
3596 case KVM_SET_USER_MEMORY_REGION: {
3597 struct kvm_userspace_memory_region kvm_userspace_mem;
3600 if (copy_from_user(&kvm_userspace_mem, argp,
3601 sizeof(kvm_userspace_mem)))
3604 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
3607 case KVM_GET_DIRTY_LOG: {
3608 struct kvm_dirty_log log;
3611 if (copy_from_user(&log, argp, sizeof(log)))
3613 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3616 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3617 case KVM_CLEAR_DIRTY_LOG: {
3618 struct kvm_clear_dirty_log log;
3621 if (copy_from_user(&log, argp, sizeof(log)))
3623 r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
3627 #ifdef CONFIG_KVM_MMIO
3628 case KVM_REGISTER_COALESCED_MMIO: {
3629 struct kvm_coalesced_mmio_zone zone;
3632 if (copy_from_user(&zone, argp, sizeof(zone)))
3634 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
3637 case KVM_UNREGISTER_COALESCED_MMIO: {
3638 struct kvm_coalesced_mmio_zone zone;
3641 if (copy_from_user(&zone, argp, sizeof(zone)))
3643 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
3648 struct kvm_irqfd data;
3651 if (copy_from_user(&data, argp, sizeof(data)))
3653 r = kvm_irqfd(kvm, &data);
3656 case KVM_IOEVENTFD: {
3657 struct kvm_ioeventfd data;
3660 if (copy_from_user(&data, argp, sizeof(data)))
3662 r = kvm_ioeventfd(kvm, &data);
3665 #ifdef CONFIG_HAVE_KVM_MSI
3666 case KVM_SIGNAL_MSI: {
3670 if (copy_from_user(&msi, argp, sizeof(msi)))
3672 r = kvm_send_userspace_msi(kvm, &msi);
3676 #ifdef __KVM_HAVE_IRQ_LINE
3677 case KVM_IRQ_LINE_STATUS:
3678 case KVM_IRQ_LINE: {
3679 struct kvm_irq_level irq_event;
3682 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
3685 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
3686 ioctl == KVM_IRQ_LINE_STATUS);
3691 if (ioctl == KVM_IRQ_LINE_STATUS) {
3692 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
3700 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3701 case KVM_SET_GSI_ROUTING: {
3702 struct kvm_irq_routing routing;
3703 struct kvm_irq_routing __user *urouting;
3704 struct kvm_irq_routing_entry *entries = NULL;
3707 if (copy_from_user(&routing, argp, sizeof(routing)))
3710 if (!kvm_arch_can_set_irq_routing(kvm))
3712 if (routing.nr > KVM_MAX_IRQ_ROUTES)
3718 entries = vmalloc(array_size(sizeof(*entries),
3724 if (copy_from_user(entries, urouting->entries,
3725 routing.nr * sizeof(*entries)))
3726 goto out_free_irq_routing;
3728 r = kvm_set_irq_routing(kvm, entries, routing.nr,
3730 out_free_irq_routing:
3734 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
3735 case KVM_CREATE_DEVICE: {
3736 struct kvm_create_device cd;
3739 if (copy_from_user(&cd, argp, sizeof(cd)))
3742 r = kvm_ioctl_create_device(kvm, &cd);
3747 if (copy_to_user(argp, &cd, sizeof(cd)))
3753 case KVM_CHECK_EXTENSION:
3754 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
3757 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
3763 #ifdef CONFIG_KVM_COMPAT
3764 struct compat_kvm_dirty_log {
3768 compat_uptr_t dirty_bitmap; /* one bit per page */
3773 static long kvm_vm_compat_ioctl(struct file *filp,
3774 unsigned int ioctl, unsigned long arg)
3776 struct kvm *kvm = filp->private_data;
3779 if (kvm->mm != current->mm)
3782 case KVM_GET_DIRTY_LOG: {
3783 struct compat_kvm_dirty_log compat_log;
3784 struct kvm_dirty_log log;
3786 if (copy_from_user(&compat_log, (void __user *)arg,
3787 sizeof(compat_log)))
3789 log.slot = compat_log.slot;
3790 log.padding1 = compat_log.padding1;
3791 log.padding2 = compat_log.padding2;
3792 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
3794 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3798 r = kvm_vm_ioctl(filp, ioctl, arg);
3804 static struct file_operations kvm_vm_fops = {
3805 .release = kvm_vm_release,
3806 .unlocked_ioctl = kvm_vm_ioctl,
3807 .llseek = noop_llseek,
3808 KVM_COMPAT(kvm_vm_compat_ioctl),
3811 static int kvm_dev_ioctl_create_vm(unsigned long type)
3817 kvm = kvm_create_vm(type);
3819 return PTR_ERR(kvm);
3820 #ifdef CONFIG_KVM_MMIO
3821 r = kvm_coalesced_mmio_init(kvm);
3825 r = get_unused_fd_flags(O_CLOEXEC);
3829 file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
3837 * Don't call kvm_put_kvm anymore at this point; file->f_op is
3838 * already set, with ->release() being kvm_vm_release(). In error
3839 * cases it will be called by the final fput(file) and will take
3840 * care of doing kvm_put_kvm(kvm).
3842 if (kvm_create_vm_debugfs(kvm, r) < 0) {
3847 kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm);
3849 fd_install(r, file);
3857 static long kvm_dev_ioctl(struct file *filp,
3858 unsigned int ioctl, unsigned long arg)
3863 case KVM_GET_API_VERSION:
3866 r = KVM_API_VERSION;
3869 r = kvm_dev_ioctl_create_vm(arg);
3871 case KVM_CHECK_EXTENSION:
3872 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
3874 case KVM_GET_VCPU_MMAP_SIZE:
3877 r = PAGE_SIZE; /* struct kvm_run */
3879 r += PAGE_SIZE; /* pio data page */
3881 #ifdef CONFIG_KVM_MMIO
3882 r += PAGE_SIZE; /* coalesced mmio ring page */
3885 case KVM_TRACE_ENABLE:
3886 case KVM_TRACE_PAUSE:
3887 case KVM_TRACE_DISABLE:
3891 return kvm_arch_dev_ioctl(filp, ioctl, arg);
3897 static struct file_operations kvm_chardev_ops = {
3898 .unlocked_ioctl = kvm_dev_ioctl,
3899 .llseek = noop_llseek,
3900 KVM_COMPAT(kvm_dev_ioctl),
3903 static struct miscdevice kvm_dev = {
3909 static void hardware_enable_nolock(void *junk)
3911 int cpu = raw_smp_processor_id();
3914 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
3917 cpumask_set_cpu(cpu, cpus_hardware_enabled);
3919 r = kvm_arch_hardware_enable();
3922 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3923 atomic_inc(&hardware_enable_failed);
3924 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
3928 static int kvm_starting_cpu(unsigned int cpu)
3930 raw_spin_lock(&kvm_count_lock);
3931 if (kvm_usage_count)
3932 hardware_enable_nolock(NULL);
3933 raw_spin_unlock(&kvm_count_lock);
3937 static void hardware_disable_nolock(void *junk)
3939 int cpu = raw_smp_processor_id();
3941 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
3943 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3944 kvm_arch_hardware_disable();
3947 static int kvm_dying_cpu(unsigned int cpu)
3949 raw_spin_lock(&kvm_count_lock);
3950 if (kvm_usage_count)
3951 hardware_disable_nolock(NULL);
3952 raw_spin_unlock(&kvm_count_lock);
3956 static void hardware_disable_all_nolock(void)
3958 BUG_ON(!kvm_usage_count);
3961 if (!kvm_usage_count)
3962 on_each_cpu(hardware_disable_nolock, NULL, 1);
3965 static void hardware_disable_all(void)
3967 raw_spin_lock(&kvm_count_lock);
3968 hardware_disable_all_nolock();
3969 raw_spin_unlock(&kvm_count_lock);
3972 static int hardware_enable_all(void)
3976 raw_spin_lock(&kvm_count_lock);
3979 if (kvm_usage_count == 1) {
3980 atomic_set(&hardware_enable_failed, 0);
3981 on_each_cpu(hardware_enable_nolock, NULL, 1);
3983 if (atomic_read(&hardware_enable_failed)) {
3984 hardware_disable_all_nolock();
3989 raw_spin_unlock(&kvm_count_lock);
3994 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3998 * Some (well, at least mine) BIOSes hang on reboot if
4001 * And Intel TXT required VMX off for all cpu when system shutdown.
4003 pr_info("kvm: exiting hardware virtualization\n");
4004 kvm_rebooting = true;
4005 on_each_cpu(hardware_disable_nolock, NULL, 1);
4009 static struct notifier_block kvm_reboot_notifier = {
4010 .notifier_call = kvm_reboot,
4014 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
4018 for (i = 0; i < bus->dev_count; i++) {
4019 struct kvm_io_device *pos = bus->range[i].dev;
4021 kvm_iodevice_destructor(pos);
4026 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
4027 const struct kvm_io_range *r2)
4029 gpa_t addr1 = r1->addr;
4030 gpa_t addr2 = r2->addr;
4035 /* If r2->len == 0, match the exact address. If r2->len != 0,
4036 * accept any overlapping write. Any order is acceptable for
4037 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
4038 * we process all of them.
4051 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
4053 return kvm_io_bus_cmp(p1, p2);
4056 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
4057 gpa_t addr, int len)
4059 struct kvm_io_range *range, key;
4062 key = (struct kvm_io_range) {
4067 range = bsearch(&key, bus->range, bus->dev_count,
4068 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
4072 off = range - bus->range;
4074 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
4080 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
4081 struct kvm_io_range *range, const void *val)
4085 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
4089 while (idx < bus->dev_count &&
4090 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
4091 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
4100 /* kvm_io_bus_write - called under kvm->slots_lock */
4101 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
4102 int len, const void *val)
4104 struct kvm_io_bus *bus;
4105 struct kvm_io_range range;
4108 range = (struct kvm_io_range) {
4113 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
4116 r = __kvm_io_bus_write(vcpu, bus, &range, val);
4117 return r < 0 ? r : 0;
4119 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
4121 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
4122 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
4123 gpa_t addr, int len, const void *val, long cookie)
4125 struct kvm_io_bus *bus;
4126 struct kvm_io_range range;
4128 range = (struct kvm_io_range) {
4133 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
4137 /* First try the device referenced by cookie. */
4138 if ((cookie >= 0) && (cookie < bus->dev_count) &&
4139 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
4140 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
4145 * cookie contained garbage; fall back to search and return the
4146 * correct cookie value.
4148 return __kvm_io_bus_write(vcpu, bus, &range, val);
4151 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
4152 struct kvm_io_range *range, void *val)
4156 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
4160 while (idx < bus->dev_count &&
4161 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
4162 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
4171 /* kvm_io_bus_read - called under kvm->slots_lock */
4172 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
4175 struct kvm_io_bus *bus;
4176 struct kvm_io_range range;
4179 range = (struct kvm_io_range) {
4184 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
4187 r = __kvm_io_bus_read(vcpu, bus, &range, val);
4188 return r < 0 ? r : 0;
4191 /* Caller must hold slots_lock. */
4192 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
4193 int len, struct kvm_io_device *dev)
4196 struct kvm_io_bus *new_bus, *bus;
4197 struct kvm_io_range range;
4199 bus = kvm_get_bus(kvm, bus_idx);
4203 /* exclude ioeventfd which is limited by maximum fd */
4204 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
4207 new_bus = kmalloc(struct_size(bus, range, bus->dev_count + 1),
4208 GFP_KERNEL_ACCOUNT);
4212 range = (struct kvm_io_range) {
4218 for (i = 0; i < bus->dev_count; i++)
4219 if (kvm_io_bus_cmp(&bus->range[i], &range) > 0)
4222 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
4223 new_bus->dev_count++;
4224 new_bus->range[i] = range;
4225 memcpy(new_bus->range + i + 1, bus->range + i,
4226 (bus->dev_count - i) * sizeof(struct kvm_io_range));
4227 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
4228 synchronize_srcu_expedited(&kvm->srcu);
4234 /* Caller must hold slots_lock. */
4235 void kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
4236 struct kvm_io_device *dev)
4239 struct kvm_io_bus *new_bus, *bus;
4241 bus = kvm_get_bus(kvm, bus_idx);
4245 for (i = 0; i < bus->dev_count; i++)
4246 if (bus->range[i].dev == dev) {
4250 if (i == bus->dev_count)
4253 new_bus = kmalloc(struct_size(bus, range, bus->dev_count - 1),
4254 GFP_KERNEL_ACCOUNT);
4256 pr_err("kvm: failed to shrink bus, removing it completely\n");
4260 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
4261 new_bus->dev_count--;
4262 memcpy(new_bus->range + i, bus->range + i + 1,
4263 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
4266 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
4267 synchronize_srcu_expedited(&kvm->srcu);
4272 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
4275 struct kvm_io_bus *bus;
4276 int dev_idx, srcu_idx;
4277 struct kvm_io_device *iodev = NULL;
4279 srcu_idx = srcu_read_lock(&kvm->srcu);
4281 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
4285 dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
4289 iodev = bus->range[dev_idx].dev;
4292 srcu_read_unlock(&kvm->srcu, srcu_idx);
4296 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
4298 static int kvm_debugfs_open(struct inode *inode, struct file *file,
4299 int (*get)(void *, u64 *), int (*set)(void *, u64),
4302 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
4305 /* The debugfs files are a reference to the kvm struct which
4306 * is still valid when kvm_destroy_vm is called.
4307 * To avoid the race between open and the removal of the debugfs
4308 * directory we test against the users count.
4310 if (!refcount_inc_not_zero(&stat_data->kvm->users_count))
4313 if (simple_attr_open(inode, file, get,
4314 KVM_DBGFS_GET_MODE(stat_data->dbgfs_item) & 0222
4317 kvm_put_kvm(stat_data->kvm);
4324 static int kvm_debugfs_release(struct inode *inode, struct file *file)
4326 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
4329 simple_attr_release(inode, file);
4330 kvm_put_kvm(stat_data->kvm);
4335 static int kvm_get_stat_per_vm(struct kvm *kvm, size_t offset, u64 *val)
4337 *val = *(ulong *)((void *)kvm + offset);
4342 static int kvm_clear_stat_per_vm(struct kvm *kvm, size_t offset)
4344 *(ulong *)((void *)kvm + offset) = 0;
4349 static int kvm_get_stat_per_vcpu(struct kvm *kvm, size_t offset, u64 *val)
4352 struct kvm_vcpu *vcpu;
4356 kvm_for_each_vcpu(i, vcpu, kvm)
4357 *val += *(u64 *)((void *)vcpu + offset);
4362 static int kvm_clear_stat_per_vcpu(struct kvm *kvm, size_t offset)
4365 struct kvm_vcpu *vcpu;
4367 kvm_for_each_vcpu(i, vcpu, kvm)
4368 *(u64 *)((void *)vcpu + offset) = 0;
4373 static int kvm_stat_data_get(void *data, u64 *val)
4376 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
4378 switch (stat_data->dbgfs_item->kind) {
4380 r = kvm_get_stat_per_vm(stat_data->kvm,
4381 stat_data->dbgfs_item->offset, val);
4384 r = kvm_get_stat_per_vcpu(stat_data->kvm,
4385 stat_data->dbgfs_item->offset, val);
4392 static int kvm_stat_data_clear(void *data, u64 val)
4395 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
4400 switch (stat_data->dbgfs_item->kind) {
4402 r = kvm_clear_stat_per_vm(stat_data->kvm,
4403 stat_data->dbgfs_item->offset);
4406 r = kvm_clear_stat_per_vcpu(stat_data->kvm,
4407 stat_data->dbgfs_item->offset);
4414 static int kvm_stat_data_open(struct inode *inode, struct file *file)
4416 __simple_attr_check_format("%llu\n", 0ull);
4417 return kvm_debugfs_open(inode, file, kvm_stat_data_get,
4418 kvm_stat_data_clear, "%llu\n");
4421 static const struct file_operations stat_fops_per_vm = {
4422 .owner = THIS_MODULE,
4423 .open = kvm_stat_data_open,
4424 .release = kvm_debugfs_release,
4425 .read = simple_attr_read,
4426 .write = simple_attr_write,
4427 .llseek = no_llseek,
4430 static int vm_stat_get(void *_offset, u64 *val)
4432 unsigned offset = (long)_offset;
4437 mutex_lock(&kvm_lock);
4438 list_for_each_entry(kvm, &vm_list, vm_list) {
4439 kvm_get_stat_per_vm(kvm, offset, &tmp_val);
4442 mutex_unlock(&kvm_lock);
4446 static int vm_stat_clear(void *_offset, u64 val)
4448 unsigned offset = (long)_offset;
4454 mutex_lock(&kvm_lock);
4455 list_for_each_entry(kvm, &vm_list, vm_list) {
4456 kvm_clear_stat_per_vm(kvm, offset);
4458 mutex_unlock(&kvm_lock);
4463 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n");
4465 static int vcpu_stat_get(void *_offset, u64 *val)
4467 unsigned offset = (long)_offset;
4472 mutex_lock(&kvm_lock);
4473 list_for_each_entry(kvm, &vm_list, vm_list) {
4474 kvm_get_stat_per_vcpu(kvm, offset, &tmp_val);
4477 mutex_unlock(&kvm_lock);
4481 static int vcpu_stat_clear(void *_offset, u64 val)
4483 unsigned offset = (long)_offset;
4489 mutex_lock(&kvm_lock);
4490 list_for_each_entry(kvm, &vm_list, vm_list) {
4491 kvm_clear_stat_per_vcpu(kvm, offset);
4493 mutex_unlock(&kvm_lock);
4498 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear,
4501 static const struct file_operations *stat_fops[] = {
4502 [KVM_STAT_VCPU] = &vcpu_stat_fops,
4503 [KVM_STAT_VM] = &vm_stat_fops,
4506 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm)
4508 struct kobj_uevent_env *env;
4509 unsigned long long created, active;
4511 if (!kvm_dev.this_device || !kvm)
4514 mutex_lock(&kvm_lock);
4515 if (type == KVM_EVENT_CREATE_VM) {
4516 kvm_createvm_count++;
4518 } else if (type == KVM_EVENT_DESTROY_VM) {
4521 created = kvm_createvm_count;
4522 active = kvm_active_vms;
4523 mutex_unlock(&kvm_lock);
4525 env = kzalloc(sizeof(*env), GFP_KERNEL_ACCOUNT);
4529 add_uevent_var(env, "CREATED=%llu", created);
4530 add_uevent_var(env, "COUNT=%llu", active);
4532 if (type == KVM_EVENT_CREATE_VM) {
4533 add_uevent_var(env, "EVENT=create");
4534 kvm->userspace_pid = task_pid_nr(current);
4535 } else if (type == KVM_EVENT_DESTROY_VM) {
4536 add_uevent_var(env, "EVENT=destroy");
4538 add_uevent_var(env, "PID=%d", kvm->userspace_pid);
4540 if (!IS_ERR_OR_NULL(kvm->debugfs_dentry)) {
4541 char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL_ACCOUNT);
4544 tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX);
4546 add_uevent_var(env, "STATS_PATH=%s", tmp);
4550 /* no need for checks, since we are adding at most only 5 keys */
4551 env->envp[env->envp_idx++] = NULL;
4552 kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp);
4556 static void kvm_init_debug(void)
4558 struct kvm_stats_debugfs_item *p;
4560 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
4562 kvm_debugfs_num_entries = 0;
4563 for (p = debugfs_entries; p->name; ++p, kvm_debugfs_num_entries++) {
4564 debugfs_create_file(p->name, KVM_DBGFS_GET_MODE(p),
4565 kvm_debugfs_dir, (void *)(long)p->offset,
4566 stat_fops[p->kind]);
4570 static int kvm_suspend(void)
4572 if (kvm_usage_count)
4573 hardware_disable_nolock(NULL);
4577 static void kvm_resume(void)
4579 if (kvm_usage_count) {
4580 #ifdef CONFIG_LOCKDEP
4581 WARN_ON(lockdep_is_held(&kvm_count_lock));
4583 hardware_enable_nolock(NULL);
4587 static struct syscore_ops kvm_syscore_ops = {
4588 .suspend = kvm_suspend,
4589 .resume = kvm_resume,
4593 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
4595 return container_of(pn, struct kvm_vcpu, preempt_notifier);
4598 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
4600 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
4602 WRITE_ONCE(vcpu->preempted, false);
4603 WRITE_ONCE(vcpu->ready, false);
4605 __this_cpu_write(kvm_running_vcpu, vcpu);
4606 kvm_arch_sched_in(vcpu, cpu);
4607 kvm_arch_vcpu_load(vcpu, cpu);
4610 static void kvm_sched_out(struct preempt_notifier *pn,
4611 struct task_struct *next)
4613 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
4615 if (current->state == TASK_RUNNING) {
4616 WRITE_ONCE(vcpu->preempted, true);
4617 WRITE_ONCE(vcpu->ready, true);
4619 kvm_arch_vcpu_put(vcpu);
4620 __this_cpu_write(kvm_running_vcpu, NULL);
4624 * kvm_get_running_vcpu - get the vcpu running on the current CPU.
4626 * We can disable preemption locally around accessing the per-CPU variable,
4627 * and use the resolved vcpu pointer after enabling preemption again,
4628 * because even if the current thread is migrated to another CPU, reading
4629 * the per-CPU value later will give us the same value as we update the
4630 * per-CPU variable in the preempt notifier handlers.
4632 struct kvm_vcpu *kvm_get_running_vcpu(void)
4634 struct kvm_vcpu *vcpu;
4637 vcpu = __this_cpu_read(kvm_running_vcpu);
4644 * kvm_get_running_vcpus - get the per-CPU array of currently running vcpus.
4646 struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void)
4648 return &kvm_running_vcpu;
4651 struct kvm_cpu_compat_check {
4656 static void check_processor_compat(void *data)
4658 struct kvm_cpu_compat_check *c = data;
4660 *c->ret = kvm_arch_check_processor_compat(c->opaque);
4663 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
4664 struct module *module)
4666 struct kvm_cpu_compat_check c;
4670 r = kvm_arch_init(opaque);
4675 * kvm_arch_init makes sure there's at most one caller
4676 * for architectures that support multiple implementations,
4677 * like intel and amd on x86.
4678 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
4679 * conflicts in case kvm is already setup for another implementation.
4681 r = kvm_irqfd_init();
4685 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
4690 r = kvm_arch_hardware_setup(opaque);
4696 for_each_online_cpu(cpu) {
4697 smp_call_function_single(cpu, check_processor_compat, &c, 1);
4702 r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting",
4703 kvm_starting_cpu, kvm_dying_cpu);
4706 register_reboot_notifier(&kvm_reboot_notifier);
4708 /* A kmem cache lets us meet the alignment requirements of fx_save. */
4710 vcpu_align = __alignof__(struct kvm_vcpu);
4712 kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align,
4714 offsetof(struct kvm_vcpu, arch),
4715 sizeof_field(struct kvm_vcpu, arch),
4717 if (!kvm_vcpu_cache) {
4722 r = kvm_async_pf_init();
4726 kvm_chardev_ops.owner = module;
4727 kvm_vm_fops.owner = module;
4728 kvm_vcpu_fops.owner = module;
4730 r = misc_register(&kvm_dev);
4732 pr_err("kvm: misc device register failed\n");
4736 register_syscore_ops(&kvm_syscore_ops);
4738 kvm_preempt_ops.sched_in = kvm_sched_in;
4739 kvm_preempt_ops.sched_out = kvm_sched_out;
4743 r = kvm_vfio_ops_init();
4749 kvm_async_pf_deinit();
4751 kmem_cache_destroy(kvm_vcpu_cache);
4753 unregister_reboot_notifier(&kvm_reboot_notifier);
4754 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4756 kvm_arch_hardware_unsetup();
4758 free_cpumask_var(cpus_hardware_enabled);
4766 EXPORT_SYMBOL_GPL(kvm_init);
4770 debugfs_remove_recursive(kvm_debugfs_dir);
4771 misc_deregister(&kvm_dev);
4772 kmem_cache_destroy(kvm_vcpu_cache);
4773 kvm_async_pf_deinit();
4774 unregister_syscore_ops(&kvm_syscore_ops);
4775 unregister_reboot_notifier(&kvm_reboot_notifier);
4776 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4777 on_each_cpu(hardware_disable_nolock, NULL, 1);
4778 kvm_arch_hardware_unsetup();
4781 free_cpumask_var(cpus_hardware_enabled);
4782 kvm_vfio_ops_exit();
4784 EXPORT_SYMBOL_GPL(kvm_exit);
4786 struct kvm_vm_worker_thread_context {
4788 struct task_struct *parent;
4789 struct completion init_done;
4790 kvm_vm_thread_fn_t thread_fn;
4795 static int kvm_vm_worker_thread(void *context)
4798 * The init_context is allocated on the stack of the parent thread, so
4799 * we have to locally copy anything that is needed beyond initialization
4801 struct kvm_vm_worker_thread_context *init_context = context;
4802 struct kvm *kvm = init_context->kvm;
4803 kvm_vm_thread_fn_t thread_fn = init_context->thread_fn;
4804 uintptr_t data = init_context->data;
4807 err = kthread_park(current);
4808 /* kthread_park(current) is never supposed to return an error */
4813 err = cgroup_attach_task_all(init_context->parent, current);
4815 kvm_err("%s: cgroup_attach_task_all failed with err %d\n",
4820 set_user_nice(current, task_nice(init_context->parent));
4823 init_context->err = err;
4824 complete(&init_context->init_done);
4825 init_context = NULL;
4830 /* Wait to be woken up by the spawner before proceeding. */
4833 if (!kthread_should_stop())
4834 err = thread_fn(kvm, data);
4839 int kvm_vm_create_worker_thread(struct kvm *kvm, kvm_vm_thread_fn_t thread_fn,
4840 uintptr_t data, const char *name,
4841 struct task_struct **thread_ptr)
4843 struct kvm_vm_worker_thread_context init_context = {};
4844 struct task_struct *thread;
4847 init_context.kvm = kvm;
4848 init_context.parent = current;
4849 init_context.thread_fn = thread_fn;
4850 init_context.data = data;
4851 init_completion(&init_context.init_done);
4853 thread = kthread_run(kvm_vm_worker_thread, &init_context,
4854 "%s-%d", name, task_pid_nr(current));
4856 return PTR_ERR(thread);
4858 /* kthread_run is never supposed to return NULL */
4859 WARN_ON(thread == NULL);
4861 wait_for_completion(&init_context.init_done);
4863 if (!init_context.err)
4864 *thread_ptr = thread;
4866 return init_context.err;