2 * Kernel-based Virtual Machine driver for Linux
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
7 * Copyright (C) 2006 Qumranet, Inc.
8 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
11 * Avi Kivity <avi@qumranet.com>
12 * Yaniv Kamay <yaniv@qumranet.com>
14 * This work is licensed under the terms of the GNU GPL, version 2. See
15 * the COPYING file in the top-level directory.
19 #include <kvm/iodev.h>
21 #include <linux/kvm_host.h>
22 #include <linux/kvm.h>
23 #include <linux/module.h>
24 #include <linux/errno.h>
25 #include <linux/percpu.h>
27 #include <linux/miscdevice.h>
28 #include <linux/vmalloc.h>
29 #include <linux/reboot.h>
30 #include <linux/debugfs.h>
31 #include <linux/highmem.h>
32 #include <linux/file.h>
33 #include <linux/syscore_ops.h>
34 #include <linux/cpu.h>
35 #include <linux/sched/signal.h>
36 #include <linux/sched/mm.h>
37 #include <linux/sched/stat.h>
38 #include <linux/cpumask.h>
39 #include <linux/smp.h>
40 #include <linux/anon_inodes.h>
41 #include <linux/profile.h>
42 #include <linux/kvm_para.h>
43 #include <linux/pagemap.h>
44 #include <linux/mman.h>
45 #include <linux/swap.h>
46 #include <linux/bitops.h>
47 #include <linux/spinlock.h>
48 #include <linux/compat.h>
49 #include <linux/srcu.h>
50 #include <linux/hugetlb.h>
51 #include <linux/slab.h>
52 #include <linux/sort.h>
53 #include <linux/bsearch.h>
55 #include <asm/processor.h>
57 #include <asm/ioctl.h>
58 #include <linux/uaccess.h>
59 #include <asm/pgtable.h>
61 #include "coalesced_mmio.h"
65 #define CREATE_TRACE_POINTS
66 #include <trace/events/kvm.h>
68 /* Worst case buffer size needed for holding an integer. */
69 #define ITOA_MAX_LEN 12
71 MODULE_AUTHOR("Qumranet");
72 MODULE_LICENSE("GPL");
74 /* Architectures should define their poll value according to the halt latency */
75 unsigned int halt_poll_ns = KVM_HALT_POLL_NS_DEFAULT;
76 module_param(halt_poll_ns, uint, 0644);
77 EXPORT_SYMBOL_GPL(halt_poll_ns);
79 /* Default doubles per-vcpu halt_poll_ns. */
80 unsigned int halt_poll_ns_grow = 2;
81 module_param(halt_poll_ns_grow, uint, 0644);
82 EXPORT_SYMBOL_GPL(halt_poll_ns_grow);
84 /* Default resets per-vcpu halt_poll_ns . */
85 unsigned int halt_poll_ns_shrink;
86 module_param(halt_poll_ns_shrink, uint, 0644);
87 EXPORT_SYMBOL_GPL(halt_poll_ns_shrink);
92 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
95 DEFINE_SPINLOCK(kvm_lock);
96 static DEFINE_RAW_SPINLOCK(kvm_count_lock);
99 static cpumask_var_t cpus_hardware_enabled;
100 static int kvm_usage_count;
101 static atomic_t hardware_enable_failed;
103 struct kmem_cache *kvm_vcpu_cache;
104 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
106 static __read_mostly struct preempt_ops kvm_preempt_ops;
108 struct dentry *kvm_debugfs_dir;
109 EXPORT_SYMBOL_GPL(kvm_debugfs_dir);
111 static int kvm_debugfs_num_entries;
112 static const struct file_operations *stat_fops_per_vm[];
114 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
116 #ifdef CONFIG_KVM_COMPAT
117 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
119 #define KVM_COMPAT(c) .compat_ioctl = (c)
121 static long kvm_no_compat_ioctl(struct file *file, unsigned int ioctl,
122 unsigned long arg) { return -EINVAL; }
123 #define KVM_COMPAT(c) .compat_ioctl = kvm_no_compat_ioctl
125 static int hardware_enable_all(void);
126 static void hardware_disable_all(void);
128 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
130 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot, gfn_t gfn);
132 __visible bool kvm_rebooting;
133 EXPORT_SYMBOL_GPL(kvm_rebooting);
135 static bool largepages_enabled = true;
137 #define KVM_EVENT_CREATE_VM 0
138 #define KVM_EVENT_DESTROY_VM 1
139 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm);
140 static unsigned long long kvm_createvm_count;
141 static unsigned long long kvm_active_vms;
143 __weak void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
144 unsigned long start, unsigned long end)
148 bool kvm_is_reserved_pfn(kvm_pfn_t pfn)
151 return PageReserved(pfn_to_page(pfn));
157 * Switches to specified vcpu, until a matching vcpu_put()
159 void vcpu_load(struct kvm_vcpu *vcpu)
162 preempt_notifier_register(&vcpu->preempt_notifier);
163 kvm_arch_vcpu_load(vcpu, cpu);
166 EXPORT_SYMBOL_GPL(vcpu_load);
168 void vcpu_put(struct kvm_vcpu *vcpu)
171 kvm_arch_vcpu_put(vcpu);
172 preempt_notifier_unregister(&vcpu->preempt_notifier);
175 EXPORT_SYMBOL_GPL(vcpu_put);
177 /* TODO: merge with kvm_arch_vcpu_should_kick */
178 static bool kvm_request_needs_ipi(struct kvm_vcpu *vcpu, unsigned req)
180 int mode = kvm_vcpu_exiting_guest_mode(vcpu);
183 * We need to wait for the VCPU to reenable interrupts and get out of
184 * READING_SHADOW_PAGE_TABLES mode.
186 if (req & KVM_REQUEST_WAIT)
187 return mode != OUTSIDE_GUEST_MODE;
190 * Need to kick a running VCPU, but otherwise there is nothing to do.
192 return mode == IN_GUEST_MODE;
195 static void ack_flush(void *_completed)
199 static inline bool kvm_kick_many_cpus(const struct cpumask *cpus, bool wait)
202 cpus = cpu_online_mask;
204 if (cpumask_empty(cpus))
207 smp_call_function_many(cpus, ack_flush, NULL, wait);
211 bool kvm_make_vcpus_request_mask(struct kvm *kvm, unsigned int req,
212 unsigned long *vcpu_bitmap, cpumask_var_t tmp)
215 struct kvm_vcpu *vcpu;
220 kvm_for_each_vcpu(i, vcpu, kvm) {
221 if (!test_bit(i, vcpu_bitmap))
224 kvm_make_request(req, vcpu);
227 if (!(req & KVM_REQUEST_NO_WAKEUP) && kvm_vcpu_wake_up(vcpu))
230 if (tmp != NULL && cpu != -1 && cpu != me &&
231 kvm_request_needs_ipi(vcpu, req))
232 __cpumask_set_cpu(cpu, tmp);
235 called = kvm_kick_many_cpus(tmp, !!(req & KVM_REQUEST_WAIT));
241 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
245 static unsigned long vcpu_bitmap[BITS_TO_LONGS(KVM_MAX_VCPUS)]
246 = {[0 ... BITS_TO_LONGS(KVM_MAX_VCPUS)-1] = ULONG_MAX};
248 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
250 called = kvm_make_vcpus_request_mask(kvm, req, vcpu_bitmap, cpus);
252 free_cpumask_var(cpus);
256 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
257 void kvm_flush_remote_tlbs(struct kvm *kvm)
260 * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in
261 * kvm_make_all_cpus_request.
263 long dirty_count = smp_load_acquire(&kvm->tlbs_dirty);
266 * We want to publish modifications to the page tables before reading
267 * mode. Pairs with a memory barrier in arch-specific code.
268 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
269 * and smp_mb in walk_shadow_page_lockless_begin/end.
270 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
272 * There is already an smp_mb__after_atomic() before
273 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
276 if (!kvm_arch_flush_remote_tlb(kvm)
277 || kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
278 ++kvm->stat.remote_tlb_flush;
279 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
281 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
284 void kvm_reload_remote_mmus(struct kvm *kvm)
286 kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
289 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
294 mutex_init(&vcpu->mutex);
299 init_swait_queue_head(&vcpu->wq);
300 kvm_async_pf_vcpu_init(vcpu);
303 INIT_LIST_HEAD(&vcpu->blocked_vcpu_list);
305 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
310 vcpu->run = page_address(page);
312 kvm_vcpu_set_in_spin_loop(vcpu, false);
313 kvm_vcpu_set_dy_eligible(vcpu, false);
314 vcpu->preempted = false;
316 r = kvm_arch_vcpu_init(vcpu);
322 free_page((unsigned long)vcpu->run);
326 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
328 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
331 * no need for rcu_read_lock as VCPU_RUN is the only place that
332 * will change the vcpu->pid pointer and on uninit all file
333 * descriptors are already gone.
335 put_pid(rcu_dereference_protected(vcpu->pid, 1));
336 kvm_arch_vcpu_uninit(vcpu);
337 free_page((unsigned long)vcpu->run);
339 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
341 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
342 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
344 return container_of(mn, struct kvm, mmu_notifier);
347 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
348 struct mm_struct *mm,
349 unsigned long address,
352 struct kvm *kvm = mmu_notifier_to_kvm(mn);
355 idx = srcu_read_lock(&kvm->srcu);
356 spin_lock(&kvm->mmu_lock);
357 kvm->mmu_notifier_seq++;
358 kvm_set_spte_hva(kvm, address, pte);
359 spin_unlock(&kvm->mmu_lock);
360 srcu_read_unlock(&kvm->srcu, idx);
363 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
364 struct mm_struct *mm,
368 struct kvm *kvm = mmu_notifier_to_kvm(mn);
369 int need_tlb_flush = 0, idx;
371 idx = srcu_read_lock(&kvm->srcu);
372 spin_lock(&kvm->mmu_lock);
374 * The count increase must become visible at unlock time as no
375 * spte can be established without taking the mmu_lock and
376 * count is also read inside the mmu_lock critical section.
378 kvm->mmu_notifier_count++;
379 need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
380 need_tlb_flush |= kvm->tlbs_dirty;
381 /* we've to flush the tlb before the pages can be freed */
383 kvm_flush_remote_tlbs(kvm);
385 spin_unlock(&kvm->mmu_lock);
387 kvm_arch_mmu_notifier_invalidate_range(kvm, start, end);
389 srcu_read_unlock(&kvm->srcu, idx);
392 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
393 struct mm_struct *mm,
397 struct kvm *kvm = mmu_notifier_to_kvm(mn);
399 spin_lock(&kvm->mmu_lock);
401 * This sequence increase will notify the kvm page fault that
402 * the page that is going to be mapped in the spte could have
405 kvm->mmu_notifier_seq++;
408 * The above sequence increase must be visible before the
409 * below count decrease, which is ensured by the smp_wmb above
410 * in conjunction with the smp_rmb in mmu_notifier_retry().
412 kvm->mmu_notifier_count--;
413 spin_unlock(&kvm->mmu_lock);
415 BUG_ON(kvm->mmu_notifier_count < 0);
418 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
419 struct mm_struct *mm,
423 struct kvm *kvm = mmu_notifier_to_kvm(mn);
426 idx = srcu_read_lock(&kvm->srcu);
427 spin_lock(&kvm->mmu_lock);
429 young = kvm_age_hva(kvm, start, end);
431 kvm_flush_remote_tlbs(kvm);
433 spin_unlock(&kvm->mmu_lock);
434 srcu_read_unlock(&kvm->srcu, idx);
439 static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
440 struct mm_struct *mm,
444 struct kvm *kvm = mmu_notifier_to_kvm(mn);
447 idx = srcu_read_lock(&kvm->srcu);
448 spin_lock(&kvm->mmu_lock);
450 * Even though we do not flush TLB, this will still adversely
451 * affect performance on pre-Haswell Intel EPT, where there is
452 * no EPT Access Bit to clear so that we have to tear down EPT
453 * tables instead. If we find this unacceptable, we can always
454 * add a parameter to kvm_age_hva so that it effectively doesn't
455 * do anything on clear_young.
457 * Also note that currently we never issue secondary TLB flushes
458 * from clear_young, leaving this job up to the regular system
459 * cadence. If we find this inaccurate, we might come up with a
460 * more sophisticated heuristic later.
462 young = kvm_age_hva(kvm, start, end);
463 spin_unlock(&kvm->mmu_lock);
464 srcu_read_unlock(&kvm->srcu, idx);
469 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
470 struct mm_struct *mm,
471 unsigned long address)
473 struct kvm *kvm = mmu_notifier_to_kvm(mn);
476 idx = srcu_read_lock(&kvm->srcu);
477 spin_lock(&kvm->mmu_lock);
478 young = kvm_test_age_hva(kvm, address);
479 spin_unlock(&kvm->mmu_lock);
480 srcu_read_unlock(&kvm->srcu, idx);
485 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
486 struct mm_struct *mm)
488 struct kvm *kvm = mmu_notifier_to_kvm(mn);
491 idx = srcu_read_lock(&kvm->srcu);
492 kvm_arch_flush_shadow_all(kvm);
493 srcu_read_unlock(&kvm->srcu, idx);
496 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
497 .flags = MMU_INVALIDATE_DOES_NOT_BLOCK,
498 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
499 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
500 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
501 .clear_young = kvm_mmu_notifier_clear_young,
502 .test_young = kvm_mmu_notifier_test_young,
503 .change_pte = kvm_mmu_notifier_change_pte,
504 .release = kvm_mmu_notifier_release,
507 static int kvm_init_mmu_notifier(struct kvm *kvm)
509 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
510 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
513 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
515 static int kvm_init_mmu_notifier(struct kvm *kvm)
520 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
522 static struct kvm_memslots *kvm_alloc_memslots(void)
525 struct kvm_memslots *slots;
527 slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
531 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
532 slots->id_to_index[i] = slots->memslots[i].id = i;
537 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
539 if (!memslot->dirty_bitmap)
542 kvfree(memslot->dirty_bitmap);
543 memslot->dirty_bitmap = NULL;
547 * Free any memory in @free but not in @dont.
549 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
550 struct kvm_memory_slot *dont)
552 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
553 kvm_destroy_dirty_bitmap(free);
555 kvm_arch_free_memslot(kvm, free, dont);
560 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
562 struct kvm_memory_slot *memslot;
567 kvm_for_each_memslot(memslot, slots)
568 kvm_free_memslot(kvm, memslot, NULL);
573 static void kvm_destroy_vm_debugfs(struct kvm *kvm)
577 if (!kvm->debugfs_dentry)
580 debugfs_remove_recursive(kvm->debugfs_dentry);
582 if (kvm->debugfs_stat_data) {
583 for (i = 0; i < kvm_debugfs_num_entries; i++)
584 kfree(kvm->debugfs_stat_data[i]);
585 kfree(kvm->debugfs_stat_data);
589 static int kvm_create_vm_debugfs(struct kvm *kvm, int fd)
591 char dir_name[ITOA_MAX_LEN * 2];
592 struct kvm_stat_data *stat_data;
593 struct kvm_stats_debugfs_item *p;
595 if (!debugfs_initialized())
598 snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd);
599 kvm->debugfs_dentry = debugfs_create_dir(dir_name, kvm_debugfs_dir);
601 kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries,
602 sizeof(*kvm->debugfs_stat_data),
604 if (!kvm->debugfs_stat_data)
607 for (p = debugfs_entries; p->name; p++) {
608 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL);
612 stat_data->kvm = kvm;
613 stat_data->offset = p->offset;
614 kvm->debugfs_stat_data[p - debugfs_entries] = stat_data;
615 debugfs_create_file(p->name, 0644, kvm->debugfs_dentry,
616 stat_data, stat_fops_per_vm[p->kind]);
621 static struct kvm *kvm_create_vm(unsigned long type)
624 struct kvm *kvm = kvm_arch_alloc_vm();
627 return ERR_PTR(-ENOMEM);
629 spin_lock_init(&kvm->mmu_lock);
631 kvm->mm = current->mm;
632 kvm_eventfd_init(kvm);
633 mutex_init(&kvm->lock);
634 mutex_init(&kvm->irq_lock);
635 mutex_init(&kvm->slots_lock);
636 refcount_set(&kvm->users_count, 1);
637 INIT_LIST_HEAD(&kvm->devices);
639 r = kvm_arch_init_vm(kvm, type);
641 goto out_err_no_disable;
643 r = hardware_enable_all();
645 goto out_err_no_disable;
647 #ifdef CONFIG_HAVE_KVM_IRQFD
648 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
651 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
654 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
655 struct kvm_memslots *slots = kvm_alloc_memslots();
657 goto out_err_no_srcu;
659 * Generations must be different for each address space.
660 * Init kvm generation close to the maximum to easily test the
661 * code of handling generation number wrap-around.
663 slots->generation = i * 2 - 150;
664 rcu_assign_pointer(kvm->memslots[i], slots);
667 if (init_srcu_struct(&kvm->srcu))
668 goto out_err_no_srcu;
669 if (init_srcu_struct(&kvm->irq_srcu))
670 goto out_err_no_irq_srcu;
671 for (i = 0; i < KVM_NR_BUSES; i++) {
672 rcu_assign_pointer(kvm->buses[i],
673 kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL));
678 r = kvm_init_mmu_notifier(kvm);
682 spin_lock(&kvm_lock);
683 list_add(&kvm->vm_list, &vm_list);
684 spin_unlock(&kvm_lock);
686 preempt_notifier_inc();
691 cleanup_srcu_struct(&kvm->irq_srcu);
693 cleanup_srcu_struct(&kvm->srcu);
695 hardware_disable_all();
697 refcount_set(&kvm->users_count, 0);
698 for (i = 0; i < KVM_NR_BUSES; i++)
699 kfree(kvm_get_bus(kvm, i));
700 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
701 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
702 kvm_arch_free_vm(kvm);
707 static void kvm_destroy_devices(struct kvm *kvm)
709 struct kvm_device *dev, *tmp;
712 * We do not need to take the kvm->lock here, because nobody else
713 * has a reference to the struct kvm at this point and therefore
714 * cannot access the devices list anyhow.
716 list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
717 list_del(&dev->vm_node);
718 dev->ops->destroy(dev);
722 static void kvm_destroy_vm(struct kvm *kvm)
725 struct mm_struct *mm = kvm->mm;
727 kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm);
728 kvm_destroy_vm_debugfs(kvm);
729 kvm_arch_sync_events(kvm);
730 spin_lock(&kvm_lock);
731 list_del(&kvm->vm_list);
732 spin_unlock(&kvm_lock);
733 kvm_free_irq_routing(kvm);
734 for (i = 0; i < KVM_NR_BUSES; i++) {
735 struct kvm_io_bus *bus = kvm_get_bus(kvm, i);
738 kvm_io_bus_destroy(bus);
739 kvm->buses[i] = NULL;
741 kvm_coalesced_mmio_free(kvm);
742 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
743 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
745 kvm_arch_flush_shadow_all(kvm);
747 kvm_arch_destroy_vm(kvm);
748 kvm_destroy_devices(kvm);
749 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
750 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
751 cleanup_srcu_struct(&kvm->irq_srcu);
752 cleanup_srcu_struct(&kvm->srcu);
753 kvm_arch_free_vm(kvm);
754 preempt_notifier_dec();
755 hardware_disable_all();
759 void kvm_get_kvm(struct kvm *kvm)
761 refcount_inc(&kvm->users_count);
763 EXPORT_SYMBOL_GPL(kvm_get_kvm);
765 void kvm_put_kvm(struct kvm *kvm)
767 if (refcount_dec_and_test(&kvm->users_count))
770 EXPORT_SYMBOL_GPL(kvm_put_kvm);
773 static int kvm_vm_release(struct inode *inode, struct file *filp)
775 struct kvm *kvm = filp->private_data;
777 kvm_irqfd_release(kvm);
784 * Allocation size is twice as large as the actual dirty bitmap size.
785 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
787 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
789 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
791 memslot->dirty_bitmap = kvzalloc(dirty_bytes, GFP_KERNEL);
792 if (!memslot->dirty_bitmap)
799 * Insert memslot and re-sort memslots based on their GFN,
800 * so binary search could be used to lookup GFN.
801 * Sorting algorithm takes advantage of having initially
802 * sorted array and known changed memslot position.
804 static void update_memslots(struct kvm_memslots *slots,
805 struct kvm_memory_slot *new)
808 int i = slots->id_to_index[id];
809 struct kvm_memory_slot *mslots = slots->memslots;
811 WARN_ON(mslots[i].id != id);
813 WARN_ON(!mslots[i].npages);
814 if (mslots[i].npages)
817 if (!mslots[i].npages)
821 while (i < KVM_MEM_SLOTS_NUM - 1 &&
822 new->base_gfn <= mslots[i + 1].base_gfn) {
823 if (!mslots[i + 1].npages)
825 mslots[i] = mslots[i + 1];
826 slots->id_to_index[mslots[i].id] = i;
831 * The ">=" is needed when creating a slot with base_gfn == 0,
832 * so that it moves before all those with base_gfn == npages == 0.
834 * On the other hand, if new->npages is zero, the above loop has
835 * already left i pointing to the beginning of the empty part of
836 * mslots, and the ">=" would move the hole backwards in this
837 * case---which is wrong. So skip the loop when deleting a slot.
841 new->base_gfn >= mslots[i - 1].base_gfn) {
842 mslots[i] = mslots[i - 1];
843 slots->id_to_index[mslots[i].id] = i;
847 WARN_ON_ONCE(i != slots->used_slots);
850 slots->id_to_index[mslots[i].id] = i;
853 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
855 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
857 #ifdef __KVM_HAVE_READONLY_MEM
858 valid_flags |= KVM_MEM_READONLY;
861 if (mem->flags & ~valid_flags)
867 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
868 int as_id, struct kvm_memslots *slots)
870 struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id);
873 * Set the low bit in the generation, which disables SPTE caching
874 * until the end of synchronize_srcu_expedited.
876 WARN_ON(old_memslots->generation & 1);
877 slots->generation = old_memslots->generation + 1;
879 rcu_assign_pointer(kvm->memslots[as_id], slots);
880 synchronize_srcu_expedited(&kvm->srcu);
883 * Increment the new memslot generation a second time. This prevents
884 * vm exits that race with memslot updates from caching a memslot
885 * generation that will (potentially) be valid forever.
887 * Generations must be unique even across address spaces. We do not need
888 * a global counter for that, instead the generation space is evenly split
889 * across address spaces. For example, with two address spaces, address
890 * space 0 will use generations 0, 4, 8, ... while * address space 1 will
891 * use generations 2, 6, 10, 14, ...
893 slots->generation += KVM_ADDRESS_SPACE_NUM * 2 - 1;
895 kvm_arch_memslots_updated(kvm, slots);
901 * Allocate some memory and give it an address in the guest physical address
904 * Discontiguous memory is allowed, mostly for framebuffers.
906 * Must be called holding kvm->slots_lock for write.
908 int __kvm_set_memory_region(struct kvm *kvm,
909 const struct kvm_userspace_memory_region *mem)
913 unsigned long npages;
914 struct kvm_memory_slot *slot;
915 struct kvm_memory_slot old, new;
916 struct kvm_memslots *slots = NULL, *old_memslots;
918 enum kvm_mr_change change;
920 r = check_memory_region_flags(mem);
925 as_id = mem->slot >> 16;
928 /* General sanity checks */
929 if (mem->memory_size & (PAGE_SIZE - 1))
931 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
933 /* We can read the guest memory with __xxx_user() later on. */
934 if ((id < KVM_USER_MEM_SLOTS) &&
935 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
936 !access_ok(VERIFY_WRITE,
937 (void __user *)(unsigned long)mem->userspace_addr,
940 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
942 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
945 slot = id_to_memslot(__kvm_memslots(kvm, as_id), id);
946 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
947 npages = mem->memory_size >> PAGE_SHIFT;
949 if (npages > KVM_MEM_MAX_NR_PAGES)
955 new.base_gfn = base_gfn;
957 new.flags = mem->flags;
961 change = KVM_MR_CREATE;
962 else { /* Modify an existing slot. */
963 if ((mem->userspace_addr != old.userspace_addr) ||
964 (npages != old.npages) ||
965 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
968 if (base_gfn != old.base_gfn)
969 change = KVM_MR_MOVE;
970 else if (new.flags != old.flags)
971 change = KVM_MR_FLAGS_ONLY;
972 else { /* Nothing to change. */
981 change = KVM_MR_DELETE;
986 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
987 /* Check for overlaps */
989 kvm_for_each_memslot(slot, __kvm_memslots(kvm, as_id)) {
992 if (!((base_gfn + npages <= slot->base_gfn) ||
993 (base_gfn >= slot->base_gfn + slot->npages)))
998 /* Free page dirty bitmap if unneeded */
999 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
1000 new.dirty_bitmap = NULL;
1003 if (change == KVM_MR_CREATE) {
1004 new.userspace_addr = mem->userspace_addr;
1006 if (kvm_arch_create_memslot(kvm, &new, npages))
1010 /* Allocate page dirty bitmap if needed */
1011 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
1012 if (kvm_create_dirty_bitmap(&new) < 0)
1016 slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
1019 memcpy(slots, __kvm_memslots(kvm, as_id), sizeof(struct kvm_memslots));
1021 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
1022 slot = id_to_memslot(slots, id);
1023 slot->flags |= KVM_MEMSLOT_INVALID;
1025 old_memslots = install_new_memslots(kvm, as_id, slots);
1027 /* From this point no new shadow pages pointing to a deleted,
1028 * or moved, memslot will be created.
1030 * validation of sp->gfn happens in:
1031 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1032 * - kvm_is_visible_gfn (mmu_check_roots)
1034 kvm_arch_flush_shadow_memslot(kvm, slot);
1037 * We can re-use the old_memslots from above, the only difference
1038 * from the currently installed memslots is the invalid flag. This
1039 * will get overwritten by update_memslots anyway.
1041 slots = old_memslots;
1044 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
1048 /* actual memory is freed via old in kvm_free_memslot below */
1049 if (change == KVM_MR_DELETE) {
1050 new.dirty_bitmap = NULL;
1051 memset(&new.arch, 0, sizeof(new.arch));
1054 update_memslots(slots, &new);
1055 old_memslots = install_new_memslots(kvm, as_id, slots);
1057 kvm_arch_commit_memory_region(kvm, mem, &old, &new, change);
1059 kvm_free_memslot(kvm, &old, &new);
1060 kvfree(old_memslots);
1066 kvm_free_memslot(kvm, &new, &old);
1070 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1072 int kvm_set_memory_region(struct kvm *kvm,
1073 const struct kvm_userspace_memory_region *mem)
1077 mutex_lock(&kvm->slots_lock);
1078 r = __kvm_set_memory_region(kvm, mem);
1079 mutex_unlock(&kvm->slots_lock);
1082 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1084 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1085 struct kvm_userspace_memory_region *mem)
1087 if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1090 return kvm_set_memory_region(kvm, mem);
1093 int kvm_get_dirty_log(struct kvm *kvm,
1094 struct kvm_dirty_log *log, int *is_dirty)
1096 struct kvm_memslots *slots;
1097 struct kvm_memory_slot *memslot;
1100 unsigned long any = 0;
1102 as_id = log->slot >> 16;
1103 id = (u16)log->slot;
1104 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1107 slots = __kvm_memslots(kvm, as_id);
1108 memslot = id_to_memslot(slots, id);
1109 if (!memslot->dirty_bitmap)
1112 n = kvm_dirty_bitmap_bytes(memslot);
1114 for (i = 0; !any && i < n/sizeof(long); ++i)
1115 any = memslot->dirty_bitmap[i];
1117 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1124 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1126 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1128 * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages
1129 * are dirty write protect them for next write.
1130 * @kvm: pointer to kvm instance
1131 * @log: slot id and address to which we copy the log
1132 * @is_dirty: flag set if any page is dirty
1134 * We need to keep it in mind that VCPU threads can write to the bitmap
1135 * concurrently. So, to avoid losing track of dirty pages we keep the
1138 * 1. Take a snapshot of the bit and clear it if needed.
1139 * 2. Write protect the corresponding page.
1140 * 3. Copy the snapshot to the userspace.
1141 * 4. Upon return caller flushes TLB's if needed.
1143 * Between 2 and 4, the guest may write to the page using the remaining TLB
1144 * entry. This is not a problem because the page is reported dirty using
1145 * the snapshot taken before and step 4 ensures that writes done after
1146 * exiting to userspace will be logged for the next call.
1149 int kvm_get_dirty_log_protect(struct kvm *kvm,
1150 struct kvm_dirty_log *log, bool *is_dirty)
1152 struct kvm_memslots *slots;
1153 struct kvm_memory_slot *memslot;
1156 unsigned long *dirty_bitmap;
1157 unsigned long *dirty_bitmap_buffer;
1159 as_id = log->slot >> 16;
1160 id = (u16)log->slot;
1161 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1164 slots = __kvm_memslots(kvm, as_id);
1165 memslot = id_to_memslot(slots, id);
1167 dirty_bitmap = memslot->dirty_bitmap;
1171 n = kvm_dirty_bitmap_bytes(memslot);
1173 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
1174 memset(dirty_bitmap_buffer, 0, n);
1176 spin_lock(&kvm->mmu_lock);
1178 for (i = 0; i < n / sizeof(long); i++) {
1182 if (!dirty_bitmap[i])
1187 mask = xchg(&dirty_bitmap[i], 0);
1188 dirty_bitmap_buffer[i] = mask;
1191 offset = i * BITS_PER_LONG;
1192 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1197 spin_unlock(&kvm->mmu_lock);
1198 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1202 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect);
1205 bool kvm_largepages_enabled(void)
1207 return largepages_enabled;
1210 void kvm_disable_largepages(void)
1212 largepages_enabled = false;
1214 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1216 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1218 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1220 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1222 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
1224 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn);
1227 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1229 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1231 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1232 memslot->flags & KVM_MEMSLOT_INVALID)
1237 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1239 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1241 struct vm_area_struct *vma;
1242 unsigned long addr, size;
1246 addr = gfn_to_hva(kvm, gfn);
1247 if (kvm_is_error_hva(addr))
1250 down_read(¤t->mm->mmap_sem);
1251 vma = find_vma(current->mm, addr);
1255 size = vma_kernel_pagesize(vma);
1258 up_read(¤t->mm->mmap_sem);
1263 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1265 return slot->flags & KVM_MEM_READONLY;
1268 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1269 gfn_t *nr_pages, bool write)
1271 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1272 return KVM_HVA_ERR_BAD;
1274 if (memslot_is_readonly(slot) && write)
1275 return KVM_HVA_ERR_RO_BAD;
1278 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1280 return __gfn_to_hva_memslot(slot, gfn);
1283 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1286 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1289 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1292 return gfn_to_hva_many(slot, gfn, NULL);
1294 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1296 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1298 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1300 EXPORT_SYMBOL_GPL(gfn_to_hva);
1302 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
1304 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
1306 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
1309 * If writable is set to false, the hva returned by this function is only
1310 * allowed to be read.
1312 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1313 gfn_t gfn, bool *writable)
1315 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1317 if (!kvm_is_error_hva(hva) && writable)
1318 *writable = !memslot_is_readonly(slot);
1323 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1325 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1327 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1330 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
1332 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1334 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1337 static inline int check_user_page_hwpoison(unsigned long addr)
1339 int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
1341 rc = get_user_pages(addr, 1, flags, NULL, NULL);
1342 return rc == -EHWPOISON;
1346 * The fast path to get the writable pfn which will be stored in @pfn,
1347 * true indicates success, otherwise false is returned. It's also the
1348 * only part that runs if we can are in atomic context.
1350 static bool hva_to_pfn_fast(unsigned long addr, bool write_fault,
1351 bool *writable, kvm_pfn_t *pfn)
1353 struct page *page[1];
1357 * Fast pin a writable pfn only if it is a write fault request
1358 * or the caller allows to map a writable pfn for a read fault
1361 if (!(write_fault || writable))
1364 npages = __get_user_pages_fast(addr, 1, 1, page);
1366 *pfn = page_to_pfn(page[0]);
1377 * The slow path to get the pfn of the specified host virtual address,
1378 * 1 indicates success, -errno is returned if error is detected.
1380 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1381 bool *writable, kvm_pfn_t *pfn)
1383 unsigned int flags = FOLL_HWPOISON;
1390 *writable = write_fault;
1393 flags |= FOLL_WRITE;
1395 flags |= FOLL_NOWAIT;
1397 npages = get_user_pages_unlocked(addr, 1, &page, flags);
1401 /* map read fault as writable if possible */
1402 if (unlikely(!write_fault) && writable) {
1405 if (__get_user_pages_fast(addr, 1, 1, &wpage) == 1) {
1411 *pfn = page_to_pfn(page);
1415 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1417 if (unlikely(!(vma->vm_flags & VM_READ)))
1420 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1426 static int hva_to_pfn_remapped(struct vm_area_struct *vma,
1427 unsigned long addr, bool *async,
1428 bool write_fault, bool *writable,
1434 r = follow_pfn(vma, addr, &pfn);
1437 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
1438 * not call the fault handler, so do it here.
1440 bool unlocked = false;
1441 r = fixup_user_fault(current, current->mm, addr,
1442 (write_fault ? FAULT_FLAG_WRITE : 0),
1449 r = follow_pfn(vma, addr, &pfn);
1459 * Get a reference here because callers of *hva_to_pfn* and
1460 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
1461 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
1462 * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will
1463 * simply do nothing for reserved pfns.
1465 * Whoever called remap_pfn_range is also going to call e.g.
1466 * unmap_mapping_range before the underlying pages are freed,
1467 * causing a call to our MMU notifier.
1476 * Pin guest page in memory and return its pfn.
1477 * @addr: host virtual address which maps memory to the guest
1478 * @atomic: whether this function can sleep
1479 * @async: whether this function need to wait IO complete if the
1480 * host page is not in the memory
1481 * @write_fault: whether we should get a writable host page
1482 * @writable: whether it allows to map a writable host page for !@write_fault
1484 * The function will map a writable host page for these two cases:
1485 * 1): @write_fault = true
1486 * 2): @write_fault = false && @writable, @writable will tell the caller
1487 * whether the mapping is writable.
1489 static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1490 bool write_fault, bool *writable)
1492 struct vm_area_struct *vma;
1496 /* we can do it either atomically or asynchronously, not both */
1497 BUG_ON(atomic && async);
1499 if (hva_to_pfn_fast(addr, write_fault, writable, &pfn))
1503 return KVM_PFN_ERR_FAULT;
1505 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1509 down_read(¤t->mm->mmap_sem);
1510 if (npages == -EHWPOISON ||
1511 (!async && check_user_page_hwpoison(addr))) {
1512 pfn = KVM_PFN_ERR_HWPOISON;
1517 vma = find_vma_intersection(current->mm, addr, addr + 1);
1520 pfn = KVM_PFN_ERR_FAULT;
1521 else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
1522 r = hva_to_pfn_remapped(vma, addr, async, write_fault, writable, &pfn);
1526 pfn = KVM_PFN_ERR_FAULT;
1528 if (async && vma_is_valid(vma, write_fault))
1530 pfn = KVM_PFN_ERR_FAULT;
1533 up_read(¤t->mm->mmap_sem);
1537 kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn,
1538 bool atomic, bool *async, bool write_fault,
1541 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1543 if (addr == KVM_HVA_ERR_RO_BAD) {
1546 return KVM_PFN_ERR_RO_FAULT;
1549 if (kvm_is_error_hva(addr)) {
1552 return KVM_PFN_NOSLOT;
1555 /* Do not map writable pfn in the readonly memslot. */
1556 if (writable && memslot_is_readonly(slot)) {
1561 return hva_to_pfn(addr, atomic, async, write_fault,
1564 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
1566 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1569 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
1570 write_fault, writable);
1572 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1574 kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1576 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1578 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
1580 kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1582 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1584 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1586 kvm_pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1588 return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm, gfn), gfn);
1590 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1592 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
1594 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1596 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
1598 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1600 return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
1602 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1604 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
1606 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1608 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
1610 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1611 struct page **pages, int nr_pages)
1616 addr = gfn_to_hva_many(slot, gfn, &entry);
1617 if (kvm_is_error_hva(addr))
1620 if (entry < nr_pages)
1623 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1625 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1627 static struct page *kvm_pfn_to_page(kvm_pfn_t pfn)
1629 if (is_error_noslot_pfn(pfn))
1630 return KVM_ERR_PTR_BAD_PAGE;
1632 if (kvm_is_reserved_pfn(pfn)) {
1634 return KVM_ERR_PTR_BAD_PAGE;
1637 return pfn_to_page(pfn);
1640 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1644 pfn = gfn_to_pfn(kvm, gfn);
1646 return kvm_pfn_to_page(pfn);
1648 EXPORT_SYMBOL_GPL(gfn_to_page);
1650 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
1654 pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
1656 return kvm_pfn_to_page(pfn);
1658 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
1660 void kvm_release_page_clean(struct page *page)
1662 WARN_ON(is_error_page(page));
1664 kvm_release_pfn_clean(page_to_pfn(page));
1666 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1668 void kvm_release_pfn_clean(kvm_pfn_t pfn)
1670 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
1671 put_page(pfn_to_page(pfn));
1673 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1675 void kvm_release_page_dirty(struct page *page)
1677 WARN_ON(is_error_page(page));
1679 kvm_release_pfn_dirty(page_to_pfn(page));
1681 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1683 void kvm_release_pfn_dirty(kvm_pfn_t pfn)
1685 kvm_set_pfn_dirty(pfn);
1686 kvm_release_pfn_clean(pfn);
1688 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1690 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
1692 if (!kvm_is_reserved_pfn(pfn)) {
1693 struct page *page = pfn_to_page(pfn);
1695 if (!PageReserved(page))
1699 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1701 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
1703 if (!kvm_is_reserved_pfn(pfn))
1704 mark_page_accessed(pfn_to_page(pfn));
1706 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1708 void kvm_get_pfn(kvm_pfn_t pfn)
1710 if (!kvm_is_reserved_pfn(pfn))
1711 get_page(pfn_to_page(pfn));
1713 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1715 static int next_segment(unsigned long len, int offset)
1717 if (len > PAGE_SIZE - offset)
1718 return PAGE_SIZE - offset;
1723 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
1724 void *data, int offset, int len)
1729 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1730 if (kvm_is_error_hva(addr))
1732 r = __copy_from_user(data, (void __user *)addr + offset, len);
1738 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1741 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1743 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1745 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1747 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
1748 int offset, int len)
1750 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1752 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1754 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
1756 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1758 gfn_t gfn = gpa >> PAGE_SHIFT;
1760 int offset = offset_in_page(gpa);
1763 while ((seg = next_segment(len, offset)) != 0) {
1764 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1774 EXPORT_SYMBOL_GPL(kvm_read_guest);
1776 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
1778 gfn_t gfn = gpa >> PAGE_SHIFT;
1780 int offset = offset_in_page(gpa);
1783 while ((seg = next_segment(len, offset)) != 0) {
1784 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
1794 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
1796 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1797 void *data, int offset, unsigned long len)
1802 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1803 if (kvm_is_error_hva(addr))
1805 pagefault_disable();
1806 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1813 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1816 gfn_t gfn = gpa >> PAGE_SHIFT;
1817 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1818 int offset = offset_in_page(gpa);
1820 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1822 EXPORT_SYMBOL_GPL(kvm_read_guest_atomic);
1824 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
1825 void *data, unsigned long len)
1827 gfn_t gfn = gpa >> PAGE_SHIFT;
1828 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1829 int offset = offset_in_page(gpa);
1831 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1833 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
1835 static int __kvm_write_guest_page(struct kvm_memory_slot *memslot, gfn_t gfn,
1836 const void *data, int offset, int len)
1841 addr = gfn_to_hva_memslot(memslot, gfn);
1842 if (kvm_is_error_hva(addr))
1844 r = __copy_to_user((void __user *)addr + offset, data, len);
1847 mark_page_dirty_in_slot(memslot, gfn);
1851 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
1852 const void *data, int offset, int len)
1854 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1856 return __kvm_write_guest_page(slot, gfn, data, offset, len);
1858 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1860 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
1861 const void *data, int offset, int len)
1863 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1865 return __kvm_write_guest_page(slot, gfn, data, offset, len);
1867 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
1869 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1872 gfn_t gfn = gpa >> PAGE_SHIFT;
1874 int offset = offset_in_page(gpa);
1877 while ((seg = next_segment(len, offset)) != 0) {
1878 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1888 EXPORT_SYMBOL_GPL(kvm_write_guest);
1890 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
1893 gfn_t gfn = gpa >> PAGE_SHIFT;
1895 int offset = offset_in_page(gpa);
1898 while ((seg = next_segment(len, offset)) != 0) {
1899 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
1909 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
1911 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
1912 struct gfn_to_hva_cache *ghc,
1913 gpa_t gpa, unsigned long len)
1915 int offset = offset_in_page(gpa);
1916 gfn_t start_gfn = gpa >> PAGE_SHIFT;
1917 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1918 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1919 gfn_t nr_pages_avail;
1922 ghc->generation = slots->generation;
1924 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
1925 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, NULL);
1926 if (!kvm_is_error_hva(ghc->hva) && nr_pages_needed <= 1) {
1930 * If the requested region crosses two memslots, we still
1931 * verify that the entire region is valid here.
1933 while (start_gfn <= end_gfn) {
1935 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
1936 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1938 if (kvm_is_error_hva(ghc->hva))
1940 start_gfn += nr_pages_avail;
1942 /* Use the slow path for cross page reads and writes. */
1943 ghc->memslot = NULL;
1948 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1949 gpa_t gpa, unsigned long len)
1951 struct kvm_memslots *slots = kvm_memslots(kvm);
1952 return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len);
1954 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1956 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1957 void *data, int offset, unsigned long len)
1959 struct kvm_memslots *slots = kvm_memslots(kvm);
1961 gpa_t gpa = ghc->gpa + offset;
1963 BUG_ON(len + offset > ghc->len);
1965 if (slots->generation != ghc->generation)
1966 __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len);
1968 if (unlikely(!ghc->memslot))
1969 return kvm_write_guest(kvm, gpa, data, len);
1971 if (kvm_is_error_hva(ghc->hva))
1974 r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
1977 mark_page_dirty_in_slot(ghc->memslot, gpa >> PAGE_SHIFT);
1981 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached);
1983 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1984 void *data, unsigned long len)
1986 return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len);
1988 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1990 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1991 void *data, unsigned long len)
1993 struct kvm_memslots *slots = kvm_memslots(kvm);
1996 BUG_ON(len > ghc->len);
1998 if (slots->generation != ghc->generation)
1999 __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len);
2001 if (unlikely(!ghc->memslot))
2002 return kvm_read_guest(kvm, ghc->gpa, data, len);
2004 if (kvm_is_error_hva(ghc->hva))
2007 r = __copy_from_user(data, (void __user *)ghc->hva, len);
2013 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
2015 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
2017 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
2019 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
2021 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
2023 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
2025 gfn_t gfn = gpa >> PAGE_SHIFT;
2027 int offset = offset_in_page(gpa);
2030 while ((seg = next_segment(len, offset)) != 0) {
2031 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
2040 EXPORT_SYMBOL_GPL(kvm_clear_guest);
2042 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot,
2045 if (memslot && memslot->dirty_bitmap) {
2046 unsigned long rel_gfn = gfn - memslot->base_gfn;
2048 set_bit_le(rel_gfn, memslot->dirty_bitmap);
2052 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
2054 struct kvm_memory_slot *memslot;
2056 memslot = gfn_to_memslot(kvm, gfn);
2057 mark_page_dirty_in_slot(memslot, gfn);
2059 EXPORT_SYMBOL_GPL(mark_page_dirty);
2061 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
2063 struct kvm_memory_slot *memslot;
2065 memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2066 mark_page_dirty_in_slot(memslot, gfn);
2068 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
2070 void kvm_sigset_activate(struct kvm_vcpu *vcpu)
2072 if (!vcpu->sigset_active)
2076 * This does a lockless modification of ->real_blocked, which is fine
2077 * because, only current can change ->real_blocked and all readers of
2078 * ->real_blocked don't care as long ->real_blocked is always a subset
2081 sigprocmask(SIG_SETMASK, &vcpu->sigset, ¤t->real_blocked);
2084 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu)
2086 if (!vcpu->sigset_active)
2089 sigprocmask(SIG_SETMASK, ¤t->real_blocked, NULL);
2090 sigemptyset(¤t->real_blocked);
2093 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
2095 unsigned int old, val, grow;
2097 old = val = vcpu->halt_poll_ns;
2098 grow = READ_ONCE(halt_poll_ns_grow);
2100 if (val == 0 && grow)
2105 if (val > halt_poll_ns)
2108 vcpu->halt_poll_ns = val;
2109 trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
2112 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
2114 unsigned int old, val, shrink;
2116 old = val = vcpu->halt_poll_ns;
2117 shrink = READ_ONCE(halt_poll_ns_shrink);
2123 vcpu->halt_poll_ns = val;
2124 trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
2127 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
2130 int idx = srcu_read_lock(&vcpu->kvm->srcu);
2132 if (kvm_arch_vcpu_runnable(vcpu)) {
2133 kvm_make_request(KVM_REQ_UNHALT, vcpu);
2136 if (kvm_cpu_has_pending_timer(vcpu))
2138 if (signal_pending(current))
2143 srcu_read_unlock(&vcpu->kvm->srcu, idx);
2148 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2150 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
2153 DECLARE_SWAITQUEUE(wait);
2154 bool waited = false;
2157 start = cur = ktime_get();
2158 if (vcpu->halt_poll_ns) {
2159 ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns);
2161 ++vcpu->stat.halt_attempted_poll;
2164 * This sets KVM_REQ_UNHALT if an interrupt
2167 if (kvm_vcpu_check_block(vcpu) < 0) {
2168 ++vcpu->stat.halt_successful_poll;
2169 if (!vcpu_valid_wakeup(vcpu))
2170 ++vcpu->stat.halt_poll_invalid;
2174 } while (single_task_running() && ktime_before(cur, stop));
2177 kvm_arch_vcpu_blocking(vcpu);
2180 prepare_to_swait_exclusive(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
2182 if (kvm_vcpu_check_block(vcpu) < 0)
2189 finish_swait(&vcpu->wq, &wait);
2192 kvm_arch_vcpu_unblocking(vcpu);
2194 block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
2196 if (!vcpu_valid_wakeup(vcpu))
2197 shrink_halt_poll_ns(vcpu);
2198 else if (halt_poll_ns) {
2199 if (block_ns <= vcpu->halt_poll_ns)
2201 /* we had a long block, shrink polling */
2202 else if (vcpu->halt_poll_ns && block_ns > halt_poll_ns)
2203 shrink_halt_poll_ns(vcpu);
2204 /* we had a short halt and our poll time is too small */
2205 else if (vcpu->halt_poll_ns < halt_poll_ns &&
2206 block_ns < halt_poll_ns)
2207 grow_halt_poll_ns(vcpu);
2209 vcpu->halt_poll_ns = 0;
2211 trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu));
2212 kvm_arch_vcpu_block_finish(vcpu);
2214 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
2216 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
2218 struct swait_queue_head *wqp;
2220 wqp = kvm_arch_vcpu_wq(vcpu);
2221 if (swq_has_sleeper(wqp)) {
2223 ++vcpu->stat.halt_wakeup;
2229 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
2233 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2235 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
2238 int cpu = vcpu->cpu;
2240 if (kvm_vcpu_wake_up(vcpu))
2244 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
2245 if (kvm_arch_vcpu_should_kick(vcpu))
2246 smp_send_reschedule(cpu);
2249 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
2250 #endif /* !CONFIG_S390 */
2252 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
2255 struct task_struct *task = NULL;
2259 pid = rcu_dereference(target->pid);
2261 task = get_pid_task(pid, PIDTYPE_PID);
2265 ret = yield_to(task, 1);
2266 put_task_struct(task);
2270 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
2273 * Helper that checks whether a VCPU is eligible for directed yield.
2274 * Most eligible candidate to yield is decided by following heuristics:
2276 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2277 * (preempted lock holder), indicated by @in_spin_loop.
2278 * Set at the beiginning and cleared at the end of interception/PLE handler.
2280 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2281 * chance last time (mostly it has become eligible now since we have probably
2282 * yielded to lockholder in last iteration. This is done by toggling
2283 * @dy_eligible each time a VCPU checked for eligibility.)
2285 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2286 * to preempted lock-holder could result in wrong VCPU selection and CPU
2287 * burning. Giving priority for a potential lock-holder increases lock
2290 * Since algorithm is based on heuristics, accessing another VCPU data without
2291 * locking does not harm. It may result in trying to yield to same VCPU, fail
2292 * and continue with next VCPU and so on.
2294 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
2296 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2299 eligible = !vcpu->spin_loop.in_spin_loop ||
2300 vcpu->spin_loop.dy_eligible;
2302 if (vcpu->spin_loop.in_spin_loop)
2303 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
2311 void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode)
2313 struct kvm *kvm = me->kvm;
2314 struct kvm_vcpu *vcpu;
2315 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
2321 kvm_vcpu_set_in_spin_loop(me, true);
2323 * We boost the priority of a VCPU that is runnable but not
2324 * currently running, because it got preempted by something
2325 * else and called schedule in __vcpu_run. Hopefully that
2326 * VCPU is holding the lock that we need and will release it.
2327 * We approximate round-robin by starting at the last boosted VCPU.
2329 for (pass = 0; pass < 2 && !yielded && try; pass++) {
2330 kvm_for_each_vcpu(i, vcpu, kvm) {
2331 if (!pass && i <= last_boosted_vcpu) {
2332 i = last_boosted_vcpu;
2334 } else if (pass && i > last_boosted_vcpu)
2336 if (!READ_ONCE(vcpu->preempted))
2340 if (swait_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
2342 if (yield_to_kernel_mode && !kvm_arch_vcpu_in_kernel(vcpu))
2344 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
2347 yielded = kvm_vcpu_yield_to(vcpu);
2349 kvm->last_boosted_vcpu = i;
2351 } else if (yielded < 0) {
2358 kvm_vcpu_set_in_spin_loop(me, false);
2360 /* Ensure vcpu is not eligible during next spinloop */
2361 kvm_vcpu_set_dy_eligible(me, false);
2363 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
2365 static vm_fault_t kvm_vcpu_fault(struct vm_fault *vmf)
2367 struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data;
2370 if (vmf->pgoff == 0)
2371 page = virt_to_page(vcpu->run);
2373 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
2374 page = virt_to_page(vcpu->arch.pio_data);
2376 #ifdef CONFIG_KVM_MMIO
2377 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
2378 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
2381 return kvm_arch_vcpu_fault(vcpu, vmf);
2387 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
2388 .fault = kvm_vcpu_fault,
2391 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2393 vma->vm_ops = &kvm_vcpu_vm_ops;
2397 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2399 struct kvm_vcpu *vcpu = filp->private_data;
2401 debugfs_remove_recursive(vcpu->debugfs_dentry);
2402 kvm_put_kvm(vcpu->kvm);
2406 static struct file_operations kvm_vcpu_fops = {
2407 .release = kvm_vcpu_release,
2408 .unlocked_ioctl = kvm_vcpu_ioctl,
2409 .mmap = kvm_vcpu_mmap,
2410 .llseek = noop_llseek,
2411 KVM_COMPAT(kvm_vcpu_compat_ioctl),
2415 * Allocates an inode for the vcpu.
2417 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2419 char name[8 + 1 + ITOA_MAX_LEN + 1];
2421 snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id);
2422 return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
2425 static int kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
2427 char dir_name[ITOA_MAX_LEN * 2];
2430 if (!kvm_arch_has_vcpu_debugfs())
2433 if (!debugfs_initialized())
2436 snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
2437 vcpu->debugfs_dentry = debugfs_create_dir(dir_name,
2438 vcpu->kvm->debugfs_dentry);
2439 if (!vcpu->debugfs_dentry)
2442 ret = kvm_arch_create_vcpu_debugfs(vcpu);
2444 debugfs_remove_recursive(vcpu->debugfs_dentry);
2452 * Creates some virtual cpus. Good luck creating more than one.
2454 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
2457 struct kvm_vcpu *vcpu;
2459 if (id >= KVM_MAX_VCPU_ID)
2462 mutex_lock(&kvm->lock);
2463 if (kvm->created_vcpus == KVM_MAX_VCPUS) {
2464 mutex_unlock(&kvm->lock);
2468 kvm->created_vcpus++;
2469 mutex_unlock(&kvm->lock);
2471 vcpu = kvm_arch_vcpu_create(kvm, id);
2474 goto vcpu_decrement;
2477 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2479 r = kvm_arch_vcpu_setup(vcpu);
2483 r = kvm_create_vcpu_debugfs(vcpu);
2487 mutex_lock(&kvm->lock);
2488 if (kvm_get_vcpu_by_id(kvm, id)) {
2490 goto unlock_vcpu_destroy;
2493 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
2495 /* Now it's all set up, let userspace reach it */
2497 r = create_vcpu_fd(vcpu);
2500 goto unlock_vcpu_destroy;
2503 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
2506 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
2507 * before kvm->online_vcpu's incremented value.
2510 atomic_inc(&kvm->online_vcpus);
2512 mutex_unlock(&kvm->lock);
2513 kvm_arch_vcpu_postcreate(vcpu);
2516 unlock_vcpu_destroy:
2517 mutex_unlock(&kvm->lock);
2518 debugfs_remove_recursive(vcpu->debugfs_dentry);
2520 kvm_arch_vcpu_destroy(vcpu);
2522 mutex_lock(&kvm->lock);
2523 kvm->created_vcpus--;
2524 mutex_unlock(&kvm->lock);
2528 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2531 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2532 vcpu->sigset_active = 1;
2533 vcpu->sigset = *sigset;
2535 vcpu->sigset_active = 0;
2539 static long kvm_vcpu_ioctl(struct file *filp,
2540 unsigned int ioctl, unsigned long arg)
2542 struct kvm_vcpu *vcpu = filp->private_data;
2543 void __user *argp = (void __user *)arg;
2545 struct kvm_fpu *fpu = NULL;
2546 struct kvm_sregs *kvm_sregs = NULL;
2548 if (vcpu->kvm->mm != current->mm)
2551 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2555 * Some architectures have vcpu ioctls that are asynchronous to vcpu
2556 * execution; mutex_lock() would break them.
2558 r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg);
2559 if (r != -ENOIOCTLCMD)
2562 if (mutex_lock_killable(&vcpu->mutex))
2570 oldpid = rcu_access_pointer(vcpu->pid);
2571 if (unlikely(oldpid != task_pid(current))) {
2572 /* The thread running this VCPU changed. */
2575 r = kvm_arch_vcpu_run_pid_change(vcpu);
2579 newpid = get_task_pid(current, PIDTYPE_PID);
2580 rcu_assign_pointer(vcpu->pid, newpid);
2585 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2586 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2589 case KVM_GET_REGS: {
2590 struct kvm_regs *kvm_regs;
2593 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2596 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2600 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2607 case KVM_SET_REGS: {
2608 struct kvm_regs *kvm_regs;
2611 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2612 if (IS_ERR(kvm_regs)) {
2613 r = PTR_ERR(kvm_regs);
2616 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2620 case KVM_GET_SREGS: {
2621 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2625 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2629 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2634 case KVM_SET_SREGS: {
2635 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2636 if (IS_ERR(kvm_sregs)) {
2637 r = PTR_ERR(kvm_sregs);
2641 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2644 case KVM_GET_MP_STATE: {
2645 struct kvm_mp_state mp_state;
2647 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2651 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
2656 case KVM_SET_MP_STATE: {
2657 struct kvm_mp_state mp_state;
2660 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
2662 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2665 case KVM_TRANSLATE: {
2666 struct kvm_translation tr;
2669 if (copy_from_user(&tr, argp, sizeof(tr)))
2671 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2675 if (copy_to_user(argp, &tr, sizeof(tr)))
2680 case KVM_SET_GUEST_DEBUG: {
2681 struct kvm_guest_debug dbg;
2684 if (copy_from_user(&dbg, argp, sizeof(dbg)))
2686 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2689 case KVM_SET_SIGNAL_MASK: {
2690 struct kvm_signal_mask __user *sigmask_arg = argp;
2691 struct kvm_signal_mask kvm_sigmask;
2692 sigset_t sigset, *p;
2697 if (copy_from_user(&kvm_sigmask, argp,
2698 sizeof(kvm_sigmask)))
2701 if (kvm_sigmask.len != sizeof(sigset))
2704 if (copy_from_user(&sigset, sigmask_arg->sigset,
2709 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2713 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2717 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2721 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2727 fpu = memdup_user(argp, sizeof(*fpu));
2733 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2737 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2740 mutex_unlock(&vcpu->mutex);
2746 #ifdef CONFIG_KVM_COMPAT
2747 static long kvm_vcpu_compat_ioctl(struct file *filp,
2748 unsigned int ioctl, unsigned long arg)
2750 struct kvm_vcpu *vcpu = filp->private_data;
2751 void __user *argp = compat_ptr(arg);
2754 if (vcpu->kvm->mm != current->mm)
2758 case KVM_SET_SIGNAL_MASK: {
2759 struct kvm_signal_mask __user *sigmask_arg = argp;
2760 struct kvm_signal_mask kvm_sigmask;
2765 if (copy_from_user(&kvm_sigmask, argp,
2766 sizeof(kvm_sigmask)))
2769 if (kvm_sigmask.len != sizeof(compat_sigset_t))
2772 if (get_compat_sigset(&sigset, (void *)sigmask_arg->sigset))
2774 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2776 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2780 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2788 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2789 int (*accessor)(struct kvm_device *dev,
2790 struct kvm_device_attr *attr),
2793 struct kvm_device_attr attr;
2798 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2801 return accessor(dev, &attr);
2804 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2807 struct kvm_device *dev = filp->private_data;
2810 case KVM_SET_DEVICE_ATTR:
2811 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2812 case KVM_GET_DEVICE_ATTR:
2813 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2814 case KVM_HAS_DEVICE_ATTR:
2815 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2817 if (dev->ops->ioctl)
2818 return dev->ops->ioctl(dev, ioctl, arg);
2824 static int kvm_device_release(struct inode *inode, struct file *filp)
2826 struct kvm_device *dev = filp->private_data;
2827 struct kvm *kvm = dev->kvm;
2833 static const struct file_operations kvm_device_fops = {
2834 .unlocked_ioctl = kvm_device_ioctl,
2835 .release = kvm_device_release,
2836 KVM_COMPAT(kvm_device_ioctl),
2839 struct kvm_device *kvm_device_from_filp(struct file *filp)
2841 if (filp->f_op != &kvm_device_fops)
2844 return filp->private_data;
2847 static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
2848 #ifdef CONFIG_KVM_MPIC
2849 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
2850 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
2854 int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
2856 if (type >= ARRAY_SIZE(kvm_device_ops_table))
2859 if (kvm_device_ops_table[type] != NULL)
2862 kvm_device_ops_table[type] = ops;
2866 void kvm_unregister_device_ops(u32 type)
2868 if (kvm_device_ops_table[type] != NULL)
2869 kvm_device_ops_table[type] = NULL;
2872 static int kvm_ioctl_create_device(struct kvm *kvm,
2873 struct kvm_create_device *cd)
2875 struct kvm_device_ops *ops = NULL;
2876 struct kvm_device *dev;
2877 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2880 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
2883 ops = kvm_device_ops_table[cd->type];
2890 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2897 mutex_lock(&kvm->lock);
2898 ret = ops->create(dev, cd->type);
2900 mutex_unlock(&kvm->lock);
2904 list_add(&dev->vm_node, &kvm->devices);
2905 mutex_unlock(&kvm->lock);
2910 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2912 mutex_lock(&kvm->lock);
2913 list_del(&dev->vm_node);
2914 mutex_unlock(&kvm->lock);
2924 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2927 case KVM_CAP_USER_MEMORY:
2928 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2929 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2930 case KVM_CAP_INTERNAL_ERROR_DATA:
2931 #ifdef CONFIG_HAVE_KVM_MSI
2932 case KVM_CAP_SIGNAL_MSI:
2934 #ifdef CONFIG_HAVE_KVM_IRQFD
2936 case KVM_CAP_IRQFD_RESAMPLE:
2938 case KVM_CAP_IOEVENTFD_ANY_LENGTH:
2939 case KVM_CAP_CHECK_EXTENSION_VM:
2941 #ifdef CONFIG_KVM_MMIO
2942 case KVM_CAP_COALESCED_MMIO:
2943 return KVM_COALESCED_MMIO_PAGE_OFFSET;
2945 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2946 case KVM_CAP_IRQ_ROUTING:
2947 return KVM_MAX_IRQ_ROUTES;
2949 #if KVM_ADDRESS_SPACE_NUM > 1
2950 case KVM_CAP_MULTI_ADDRESS_SPACE:
2951 return KVM_ADDRESS_SPACE_NUM;
2953 case KVM_CAP_MAX_VCPU_ID:
2954 return KVM_MAX_VCPU_ID;
2958 return kvm_vm_ioctl_check_extension(kvm, arg);
2961 static long kvm_vm_ioctl(struct file *filp,
2962 unsigned int ioctl, unsigned long arg)
2964 struct kvm *kvm = filp->private_data;
2965 void __user *argp = (void __user *)arg;
2968 if (kvm->mm != current->mm)
2971 case KVM_CREATE_VCPU:
2972 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2974 case KVM_SET_USER_MEMORY_REGION: {
2975 struct kvm_userspace_memory_region kvm_userspace_mem;
2978 if (copy_from_user(&kvm_userspace_mem, argp,
2979 sizeof(kvm_userspace_mem)))
2982 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2985 case KVM_GET_DIRTY_LOG: {
2986 struct kvm_dirty_log log;
2989 if (copy_from_user(&log, argp, sizeof(log)))
2991 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2994 #ifdef CONFIG_KVM_MMIO
2995 case KVM_REGISTER_COALESCED_MMIO: {
2996 struct kvm_coalesced_mmio_zone zone;
2999 if (copy_from_user(&zone, argp, sizeof(zone)))
3001 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
3004 case KVM_UNREGISTER_COALESCED_MMIO: {
3005 struct kvm_coalesced_mmio_zone zone;
3008 if (copy_from_user(&zone, argp, sizeof(zone)))
3010 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
3015 struct kvm_irqfd data;
3018 if (copy_from_user(&data, argp, sizeof(data)))
3020 r = kvm_irqfd(kvm, &data);
3023 case KVM_IOEVENTFD: {
3024 struct kvm_ioeventfd data;
3027 if (copy_from_user(&data, argp, sizeof(data)))
3029 r = kvm_ioeventfd(kvm, &data);
3032 #ifdef CONFIG_HAVE_KVM_MSI
3033 case KVM_SIGNAL_MSI: {
3037 if (copy_from_user(&msi, argp, sizeof(msi)))
3039 r = kvm_send_userspace_msi(kvm, &msi);
3043 #ifdef __KVM_HAVE_IRQ_LINE
3044 case KVM_IRQ_LINE_STATUS:
3045 case KVM_IRQ_LINE: {
3046 struct kvm_irq_level irq_event;
3049 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
3052 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
3053 ioctl == KVM_IRQ_LINE_STATUS);
3058 if (ioctl == KVM_IRQ_LINE_STATUS) {
3059 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
3067 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3068 case KVM_SET_GSI_ROUTING: {
3069 struct kvm_irq_routing routing;
3070 struct kvm_irq_routing __user *urouting;
3071 struct kvm_irq_routing_entry *entries = NULL;
3074 if (copy_from_user(&routing, argp, sizeof(routing)))
3077 if (!kvm_arch_can_set_irq_routing(kvm))
3079 if (routing.nr > KVM_MAX_IRQ_ROUTES)
3085 entries = vmalloc(array_size(sizeof(*entries),
3091 if (copy_from_user(entries, urouting->entries,
3092 routing.nr * sizeof(*entries)))
3093 goto out_free_irq_routing;
3095 r = kvm_set_irq_routing(kvm, entries, routing.nr,
3097 out_free_irq_routing:
3101 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
3102 case KVM_CREATE_DEVICE: {
3103 struct kvm_create_device cd;
3106 if (copy_from_user(&cd, argp, sizeof(cd)))
3109 r = kvm_ioctl_create_device(kvm, &cd);
3114 if (copy_to_user(argp, &cd, sizeof(cd)))
3120 case KVM_CHECK_EXTENSION:
3121 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
3124 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
3130 #ifdef CONFIG_KVM_COMPAT
3131 struct compat_kvm_dirty_log {
3135 compat_uptr_t dirty_bitmap; /* one bit per page */
3140 static long kvm_vm_compat_ioctl(struct file *filp,
3141 unsigned int ioctl, unsigned long arg)
3143 struct kvm *kvm = filp->private_data;
3146 if (kvm->mm != current->mm)
3149 case KVM_GET_DIRTY_LOG: {
3150 struct compat_kvm_dirty_log compat_log;
3151 struct kvm_dirty_log log;
3153 if (copy_from_user(&compat_log, (void __user *)arg,
3154 sizeof(compat_log)))
3156 log.slot = compat_log.slot;
3157 log.padding1 = compat_log.padding1;
3158 log.padding2 = compat_log.padding2;
3159 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
3161 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3165 r = kvm_vm_ioctl(filp, ioctl, arg);
3171 static struct file_operations kvm_vm_fops = {
3172 .release = kvm_vm_release,
3173 .unlocked_ioctl = kvm_vm_ioctl,
3174 .llseek = noop_llseek,
3175 KVM_COMPAT(kvm_vm_compat_ioctl),
3178 static int kvm_dev_ioctl_create_vm(unsigned long type)
3184 kvm = kvm_create_vm(type);
3186 return PTR_ERR(kvm);
3187 #ifdef CONFIG_KVM_MMIO
3188 r = kvm_coalesced_mmio_init(kvm);
3192 r = get_unused_fd_flags(O_CLOEXEC);
3196 file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
3204 * Don't call kvm_put_kvm anymore at this point; file->f_op is
3205 * already set, with ->release() being kvm_vm_release(). In error
3206 * cases it will be called by the final fput(file) and will take
3207 * care of doing kvm_put_kvm(kvm).
3209 if (kvm_create_vm_debugfs(kvm, r) < 0) {
3214 kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm);
3216 fd_install(r, file);
3224 static long kvm_dev_ioctl(struct file *filp,
3225 unsigned int ioctl, unsigned long arg)
3230 case KVM_GET_API_VERSION:
3233 r = KVM_API_VERSION;
3236 r = kvm_dev_ioctl_create_vm(arg);
3238 case KVM_CHECK_EXTENSION:
3239 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
3241 case KVM_GET_VCPU_MMAP_SIZE:
3244 r = PAGE_SIZE; /* struct kvm_run */
3246 r += PAGE_SIZE; /* pio data page */
3248 #ifdef CONFIG_KVM_MMIO
3249 r += PAGE_SIZE; /* coalesced mmio ring page */
3252 case KVM_TRACE_ENABLE:
3253 case KVM_TRACE_PAUSE:
3254 case KVM_TRACE_DISABLE:
3258 return kvm_arch_dev_ioctl(filp, ioctl, arg);
3264 static struct file_operations kvm_chardev_ops = {
3265 .unlocked_ioctl = kvm_dev_ioctl,
3266 .llseek = noop_llseek,
3267 KVM_COMPAT(kvm_dev_ioctl),
3270 static struct miscdevice kvm_dev = {
3276 static void hardware_enable_nolock(void *junk)
3278 int cpu = raw_smp_processor_id();
3281 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
3284 cpumask_set_cpu(cpu, cpus_hardware_enabled);
3286 r = kvm_arch_hardware_enable();
3289 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3290 atomic_inc(&hardware_enable_failed);
3291 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
3295 static int kvm_starting_cpu(unsigned int cpu)
3297 raw_spin_lock(&kvm_count_lock);
3298 if (kvm_usage_count)
3299 hardware_enable_nolock(NULL);
3300 raw_spin_unlock(&kvm_count_lock);
3304 static void hardware_disable_nolock(void *junk)
3306 int cpu = raw_smp_processor_id();
3308 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
3310 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3311 kvm_arch_hardware_disable();
3314 static int kvm_dying_cpu(unsigned int cpu)
3316 raw_spin_lock(&kvm_count_lock);
3317 if (kvm_usage_count)
3318 hardware_disable_nolock(NULL);
3319 raw_spin_unlock(&kvm_count_lock);
3323 static void hardware_disable_all_nolock(void)
3325 BUG_ON(!kvm_usage_count);
3328 if (!kvm_usage_count)
3329 on_each_cpu(hardware_disable_nolock, NULL, 1);
3332 static void hardware_disable_all(void)
3334 raw_spin_lock(&kvm_count_lock);
3335 hardware_disable_all_nolock();
3336 raw_spin_unlock(&kvm_count_lock);
3339 static int hardware_enable_all(void)
3343 raw_spin_lock(&kvm_count_lock);
3346 if (kvm_usage_count == 1) {
3347 atomic_set(&hardware_enable_failed, 0);
3348 on_each_cpu(hardware_enable_nolock, NULL, 1);
3350 if (atomic_read(&hardware_enable_failed)) {
3351 hardware_disable_all_nolock();
3356 raw_spin_unlock(&kvm_count_lock);
3361 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3365 * Some (well, at least mine) BIOSes hang on reboot if
3368 * And Intel TXT required VMX off for all cpu when system shutdown.
3370 pr_info("kvm: exiting hardware virtualization\n");
3371 kvm_rebooting = true;
3372 on_each_cpu(hardware_disable_nolock, NULL, 1);
3376 static struct notifier_block kvm_reboot_notifier = {
3377 .notifier_call = kvm_reboot,
3381 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3385 for (i = 0; i < bus->dev_count; i++) {
3386 struct kvm_io_device *pos = bus->range[i].dev;
3388 kvm_iodevice_destructor(pos);
3393 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
3394 const struct kvm_io_range *r2)
3396 gpa_t addr1 = r1->addr;
3397 gpa_t addr2 = r2->addr;
3402 /* If r2->len == 0, match the exact address. If r2->len != 0,
3403 * accept any overlapping write. Any order is acceptable for
3404 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
3405 * we process all of them.
3418 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
3420 return kvm_io_bus_cmp(p1, p2);
3423 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
3424 gpa_t addr, int len)
3426 struct kvm_io_range *range, key;
3429 key = (struct kvm_io_range) {
3434 range = bsearch(&key, bus->range, bus->dev_count,
3435 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
3439 off = range - bus->range;
3441 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
3447 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3448 struct kvm_io_range *range, const void *val)
3452 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3456 while (idx < bus->dev_count &&
3457 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3458 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
3467 /* kvm_io_bus_write - called under kvm->slots_lock */
3468 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3469 int len, const void *val)
3471 struct kvm_io_bus *bus;
3472 struct kvm_io_range range;
3475 range = (struct kvm_io_range) {
3480 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3483 r = __kvm_io_bus_write(vcpu, bus, &range, val);
3484 return r < 0 ? r : 0;
3487 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3488 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
3489 gpa_t addr, int len, const void *val, long cookie)
3491 struct kvm_io_bus *bus;
3492 struct kvm_io_range range;
3494 range = (struct kvm_io_range) {
3499 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3503 /* First try the device referenced by cookie. */
3504 if ((cookie >= 0) && (cookie < bus->dev_count) &&
3505 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
3506 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
3511 * cookie contained garbage; fall back to search and return the
3512 * correct cookie value.
3514 return __kvm_io_bus_write(vcpu, bus, &range, val);
3517 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3518 struct kvm_io_range *range, void *val)
3522 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3526 while (idx < bus->dev_count &&
3527 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3528 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
3536 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
3538 /* kvm_io_bus_read - called under kvm->slots_lock */
3539 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3542 struct kvm_io_bus *bus;
3543 struct kvm_io_range range;
3546 range = (struct kvm_io_range) {
3551 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3554 r = __kvm_io_bus_read(vcpu, bus, &range, val);
3555 return r < 0 ? r : 0;
3559 /* Caller must hold slots_lock. */
3560 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3561 int len, struct kvm_io_device *dev)
3564 struct kvm_io_bus *new_bus, *bus;
3565 struct kvm_io_range range;
3567 bus = kvm_get_bus(kvm, bus_idx);
3571 /* exclude ioeventfd which is limited by maximum fd */
3572 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3575 new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count + 1) *
3576 sizeof(struct kvm_io_range)), GFP_KERNEL);
3580 range = (struct kvm_io_range) {
3586 for (i = 0; i < bus->dev_count; i++)
3587 if (kvm_io_bus_cmp(&bus->range[i], &range) > 0)
3590 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3591 new_bus->dev_count++;
3592 new_bus->range[i] = range;
3593 memcpy(new_bus->range + i + 1, bus->range + i,
3594 (bus->dev_count - i) * sizeof(struct kvm_io_range));
3595 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3596 synchronize_srcu_expedited(&kvm->srcu);
3602 /* Caller must hold slots_lock. */
3603 void kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3604 struct kvm_io_device *dev)
3607 struct kvm_io_bus *new_bus, *bus;
3609 bus = kvm_get_bus(kvm, bus_idx);
3613 for (i = 0; i < bus->dev_count; i++)
3614 if (bus->range[i].dev == dev) {
3618 if (i == bus->dev_count)
3621 new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3622 sizeof(struct kvm_io_range)), GFP_KERNEL);
3624 pr_err("kvm: failed to shrink bus, removing it completely\n");
3628 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3629 new_bus->dev_count--;
3630 memcpy(new_bus->range + i, bus->range + i + 1,
3631 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3634 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3635 synchronize_srcu_expedited(&kvm->srcu);
3640 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3643 struct kvm_io_bus *bus;
3644 int dev_idx, srcu_idx;
3645 struct kvm_io_device *iodev = NULL;
3647 srcu_idx = srcu_read_lock(&kvm->srcu);
3649 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
3653 dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
3657 iodev = bus->range[dev_idx].dev;
3660 srcu_read_unlock(&kvm->srcu, srcu_idx);
3664 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
3666 static int kvm_debugfs_open(struct inode *inode, struct file *file,
3667 int (*get)(void *, u64 *), int (*set)(void *, u64),
3670 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
3673 /* The debugfs files are a reference to the kvm struct which
3674 * is still valid when kvm_destroy_vm is called.
3675 * To avoid the race between open and the removal of the debugfs
3676 * directory we test against the users count.
3678 if (!refcount_inc_not_zero(&stat_data->kvm->users_count))
3681 if (simple_attr_open(inode, file, get, set, fmt)) {
3682 kvm_put_kvm(stat_data->kvm);
3689 static int kvm_debugfs_release(struct inode *inode, struct file *file)
3691 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
3694 simple_attr_release(inode, file);
3695 kvm_put_kvm(stat_data->kvm);
3700 static int vm_stat_get_per_vm(void *data, u64 *val)
3702 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3704 *val = *(ulong *)((void *)stat_data->kvm + stat_data->offset);
3709 static int vm_stat_clear_per_vm(void *data, u64 val)
3711 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3716 *(ulong *)((void *)stat_data->kvm + stat_data->offset) = 0;
3721 static int vm_stat_get_per_vm_open(struct inode *inode, struct file *file)
3723 __simple_attr_check_format("%llu\n", 0ull);
3724 return kvm_debugfs_open(inode, file, vm_stat_get_per_vm,
3725 vm_stat_clear_per_vm, "%llu\n");
3728 static const struct file_operations vm_stat_get_per_vm_fops = {
3729 .owner = THIS_MODULE,
3730 .open = vm_stat_get_per_vm_open,
3731 .release = kvm_debugfs_release,
3732 .read = simple_attr_read,
3733 .write = simple_attr_write,
3734 .llseek = no_llseek,
3737 static int vcpu_stat_get_per_vm(void *data, u64 *val)
3740 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3741 struct kvm_vcpu *vcpu;
3745 kvm_for_each_vcpu(i, vcpu, stat_data->kvm)
3746 *val += *(u64 *)((void *)vcpu + stat_data->offset);
3751 static int vcpu_stat_clear_per_vm(void *data, u64 val)
3754 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3755 struct kvm_vcpu *vcpu;
3760 kvm_for_each_vcpu(i, vcpu, stat_data->kvm)
3761 *(u64 *)((void *)vcpu + stat_data->offset) = 0;
3766 static int vcpu_stat_get_per_vm_open(struct inode *inode, struct file *file)
3768 __simple_attr_check_format("%llu\n", 0ull);
3769 return kvm_debugfs_open(inode, file, vcpu_stat_get_per_vm,
3770 vcpu_stat_clear_per_vm, "%llu\n");
3773 static const struct file_operations vcpu_stat_get_per_vm_fops = {
3774 .owner = THIS_MODULE,
3775 .open = vcpu_stat_get_per_vm_open,
3776 .release = kvm_debugfs_release,
3777 .read = simple_attr_read,
3778 .write = simple_attr_write,
3779 .llseek = no_llseek,
3782 static const struct file_operations *stat_fops_per_vm[] = {
3783 [KVM_STAT_VCPU] = &vcpu_stat_get_per_vm_fops,
3784 [KVM_STAT_VM] = &vm_stat_get_per_vm_fops,
3787 static int vm_stat_get(void *_offset, u64 *val)
3789 unsigned offset = (long)_offset;
3791 struct kvm_stat_data stat_tmp = {.offset = offset};
3795 spin_lock(&kvm_lock);
3796 list_for_each_entry(kvm, &vm_list, vm_list) {
3798 vm_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
3801 spin_unlock(&kvm_lock);
3805 static int vm_stat_clear(void *_offset, u64 val)
3807 unsigned offset = (long)_offset;
3809 struct kvm_stat_data stat_tmp = {.offset = offset};
3814 spin_lock(&kvm_lock);
3815 list_for_each_entry(kvm, &vm_list, vm_list) {
3817 vm_stat_clear_per_vm((void *)&stat_tmp, 0);
3819 spin_unlock(&kvm_lock);
3824 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n");
3826 static int vcpu_stat_get(void *_offset, u64 *val)
3828 unsigned offset = (long)_offset;
3830 struct kvm_stat_data stat_tmp = {.offset = offset};
3834 spin_lock(&kvm_lock);
3835 list_for_each_entry(kvm, &vm_list, vm_list) {
3837 vcpu_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
3840 spin_unlock(&kvm_lock);
3844 static int vcpu_stat_clear(void *_offset, u64 val)
3846 unsigned offset = (long)_offset;
3848 struct kvm_stat_data stat_tmp = {.offset = offset};
3853 spin_lock(&kvm_lock);
3854 list_for_each_entry(kvm, &vm_list, vm_list) {
3856 vcpu_stat_clear_per_vm((void *)&stat_tmp, 0);
3858 spin_unlock(&kvm_lock);
3863 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear,
3866 static const struct file_operations *stat_fops[] = {
3867 [KVM_STAT_VCPU] = &vcpu_stat_fops,
3868 [KVM_STAT_VM] = &vm_stat_fops,
3871 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm)
3873 struct kobj_uevent_env *env;
3874 unsigned long long created, active;
3876 if (!kvm_dev.this_device || !kvm)
3879 spin_lock(&kvm_lock);
3880 if (type == KVM_EVENT_CREATE_VM) {
3881 kvm_createvm_count++;
3883 } else if (type == KVM_EVENT_DESTROY_VM) {
3886 created = kvm_createvm_count;
3887 active = kvm_active_vms;
3888 spin_unlock(&kvm_lock);
3890 env = kzalloc(sizeof(*env), GFP_KERNEL);
3894 add_uevent_var(env, "CREATED=%llu", created);
3895 add_uevent_var(env, "COUNT=%llu", active);
3897 if (type == KVM_EVENT_CREATE_VM) {
3898 add_uevent_var(env, "EVENT=create");
3899 kvm->userspace_pid = task_pid_nr(current);
3900 } else if (type == KVM_EVENT_DESTROY_VM) {
3901 add_uevent_var(env, "EVENT=destroy");
3903 add_uevent_var(env, "PID=%d", kvm->userspace_pid);
3905 if (kvm->debugfs_dentry) {
3906 char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL);
3909 tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX);
3911 add_uevent_var(env, "STATS_PATH=%s", tmp);
3915 /* no need for checks, since we are adding at most only 5 keys */
3916 env->envp[env->envp_idx++] = NULL;
3917 kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp);
3921 static void kvm_init_debug(void)
3923 struct kvm_stats_debugfs_item *p;
3925 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3927 kvm_debugfs_num_entries = 0;
3928 for (p = debugfs_entries; p->name; ++p, kvm_debugfs_num_entries++) {
3929 debugfs_create_file(p->name, 0644, kvm_debugfs_dir,
3930 (void *)(long)p->offset,
3931 stat_fops[p->kind]);
3935 static int kvm_suspend(void)
3937 if (kvm_usage_count)
3938 hardware_disable_nolock(NULL);
3942 static void kvm_resume(void)
3944 if (kvm_usage_count) {
3945 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3946 hardware_enable_nolock(NULL);
3950 static struct syscore_ops kvm_syscore_ops = {
3951 .suspend = kvm_suspend,
3952 .resume = kvm_resume,
3956 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3958 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3961 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3963 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3965 if (vcpu->preempted)
3966 vcpu->preempted = false;
3968 kvm_arch_sched_in(vcpu, cpu);
3970 kvm_arch_vcpu_load(vcpu, cpu);
3973 static void kvm_sched_out(struct preempt_notifier *pn,
3974 struct task_struct *next)
3976 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3978 if (current->state == TASK_RUNNING)
3979 vcpu->preempted = true;
3980 kvm_arch_vcpu_put(vcpu);
3983 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3984 struct module *module)
3989 r = kvm_arch_init(opaque);
3994 * kvm_arch_init makes sure there's at most one caller
3995 * for architectures that support multiple implementations,
3996 * like intel and amd on x86.
3997 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3998 * conflicts in case kvm is already setup for another implementation.
4000 r = kvm_irqfd_init();
4004 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
4009 r = kvm_arch_hardware_setup();
4013 for_each_online_cpu(cpu) {
4014 smp_call_function_single(cpu,
4015 kvm_arch_check_processor_compat,
4021 r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting",
4022 kvm_starting_cpu, kvm_dying_cpu);
4025 register_reboot_notifier(&kvm_reboot_notifier);
4027 /* A kmem cache lets us meet the alignment requirements of fx_save. */
4029 vcpu_align = __alignof__(struct kvm_vcpu);
4031 kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align,
4033 offsetof(struct kvm_vcpu, arch),
4034 sizeof_field(struct kvm_vcpu, arch),
4036 if (!kvm_vcpu_cache) {
4041 r = kvm_async_pf_init();
4045 kvm_chardev_ops.owner = module;
4046 kvm_vm_fops.owner = module;
4047 kvm_vcpu_fops.owner = module;
4049 r = misc_register(&kvm_dev);
4051 pr_err("kvm: misc device register failed\n");
4055 register_syscore_ops(&kvm_syscore_ops);
4057 kvm_preempt_ops.sched_in = kvm_sched_in;
4058 kvm_preempt_ops.sched_out = kvm_sched_out;
4062 r = kvm_vfio_ops_init();
4068 kvm_async_pf_deinit();
4070 kmem_cache_destroy(kvm_vcpu_cache);
4072 unregister_reboot_notifier(&kvm_reboot_notifier);
4073 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4076 kvm_arch_hardware_unsetup();
4078 free_cpumask_var(cpus_hardware_enabled);
4086 EXPORT_SYMBOL_GPL(kvm_init);
4090 debugfs_remove_recursive(kvm_debugfs_dir);
4091 misc_deregister(&kvm_dev);
4092 kmem_cache_destroy(kvm_vcpu_cache);
4093 kvm_async_pf_deinit();
4094 unregister_syscore_ops(&kvm_syscore_ops);
4095 unregister_reboot_notifier(&kvm_reboot_notifier);
4096 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4097 on_each_cpu(hardware_disable_nolock, NULL, 1);
4098 kvm_arch_hardware_unsetup();
4101 free_cpumask_var(cpus_hardware_enabled);
4102 kvm_vfio_ops_exit();
4104 EXPORT_SYMBOL_GPL(kvm_exit);