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_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
277 ++kvm->stat.remote_tlb_flush;
278 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
280 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
283 void kvm_reload_remote_mmus(struct kvm *kvm)
285 kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
288 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
293 mutex_init(&vcpu->mutex);
298 init_swait_queue_head(&vcpu->wq);
299 kvm_async_pf_vcpu_init(vcpu);
302 INIT_LIST_HEAD(&vcpu->blocked_vcpu_list);
304 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
309 vcpu->run = page_address(page);
311 kvm_vcpu_set_in_spin_loop(vcpu, false);
312 kvm_vcpu_set_dy_eligible(vcpu, false);
313 vcpu->preempted = false;
315 r = kvm_arch_vcpu_init(vcpu);
321 free_page((unsigned long)vcpu->run);
325 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
327 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
330 * no need for rcu_read_lock as VCPU_RUN is the only place that
331 * will change the vcpu->pid pointer and on uninit all file
332 * descriptors are already gone.
334 put_pid(rcu_dereference_protected(vcpu->pid, 1));
335 kvm_arch_vcpu_uninit(vcpu);
336 free_page((unsigned long)vcpu->run);
338 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
340 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
341 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
343 return container_of(mn, struct kvm, mmu_notifier);
346 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
347 struct mm_struct *mm,
348 unsigned long address,
351 struct kvm *kvm = mmu_notifier_to_kvm(mn);
354 idx = srcu_read_lock(&kvm->srcu);
355 spin_lock(&kvm->mmu_lock);
356 kvm->mmu_notifier_seq++;
357 kvm_set_spte_hva(kvm, address, pte);
358 spin_unlock(&kvm->mmu_lock);
359 srcu_read_unlock(&kvm->srcu, idx);
362 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
363 struct mm_struct *mm,
367 struct kvm *kvm = mmu_notifier_to_kvm(mn);
368 int need_tlb_flush = 0, idx;
370 idx = srcu_read_lock(&kvm->srcu);
371 spin_lock(&kvm->mmu_lock);
373 * The count increase must become visible at unlock time as no
374 * spte can be established without taking the mmu_lock and
375 * count is also read inside the mmu_lock critical section.
377 kvm->mmu_notifier_count++;
378 need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
379 need_tlb_flush |= kvm->tlbs_dirty;
380 /* we've to flush the tlb before the pages can be freed */
382 kvm_flush_remote_tlbs(kvm);
384 spin_unlock(&kvm->mmu_lock);
386 kvm_arch_mmu_notifier_invalidate_range(kvm, start, end);
388 srcu_read_unlock(&kvm->srcu, idx);
391 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
392 struct mm_struct *mm,
396 struct kvm *kvm = mmu_notifier_to_kvm(mn);
398 spin_lock(&kvm->mmu_lock);
400 * This sequence increase will notify the kvm page fault that
401 * the page that is going to be mapped in the spte could have
404 kvm->mmu_notifier_seq++;
407 * The above sequence increase must be visible before the
408 * below count decrease, which is ensured by the smp_wmb above
409 * in conjunction with the smp_rmb in mmu_notifier_retry().
411 kvm->mmu_notifier_count--;
412 spin_unlock(&kvm->mmu_lock);
414 BUG_ON(kvm->mmu_notifier_count < 0);
417 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
418 struct mm_struct *mm,
422 struct kvm *kvm = mmu_notifier_to_kvm(mn);
425 idx = srcu_read_lock(&kvm->srcu);
426 spin_lock(&kvm->mmu_lock);
428 young = kvm_age_hva(kvm, start, end);
430 kvm_flush_remote_tlbs(kvm);
432 spin_unlock(&kvm->mmu_lock);
433 srcu_read_unlock(&kvm->srcu, idx);
438 static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
439 struct mm_struct *mm,
443 struct kvm *kvm = mmu_notifier_to_kvm(mn);
446 idx = srcu_read_lock(&kvm->srcu);
447 spin_lock(&kvm->mmu_lock);
449 * Even though we do not flush TLB, this will still adversely
450 * affect performance on pre-Haswell Intel EPT, where there is
451 * no EPT Access Bit to clear so that we have to tear down EPT
452 * tables instead. If we find this unacceptable, we can always
453 * add a parameter to kvm_age_hva so that it effectively doesn't
454 * do anything on clear_young.
456 * Also note that currently we never issue secondary TLB flushes
457 * from clear_young, leaving this job up to the regular system
458 * cadence. If we find this inaccurate, we might come up with a
459 * more sophisticated heuristic later.
461 young = kvm_age_hva(kvm, start, end);
462 spin_unlock(&kvm->mmu_lock);
463 srcu_read_unlock(&kvm->srcu, idx);
468 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
469 struct mm_struct *mm,
470 unsigned long address)
472 struct kvm *kvm = mmu_notifier_to_kvm(mn);
475 idx = srcu_read_lock(&kvm->srcu);
476 spin_lock(&kvm->mmu_lock);
477 young = kvm_test_age_hva(kvm, address);
478 spin_unlock(&kvm->mmu_lock);
479 srcu_read_unlock(&kvm->srcu, idx);
484 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
485 struct mm_struct *mm)
487 struct kvm *kvm = mmu_notifier_to_kvm(mn);
490 idx = srcu_read_lock(&kvm->srcu);
491 kvm_arch_flush_shadow_all(kvm);
492 srcu_read_unlock(&kvm->srcu, idx);
495 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
496 .flags = MMU_INVALIDATE_DOES_NOT_BLOCK,
497 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
498 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
499 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
500 .clear_young = kvm_mmu_notifier_clear_young,
501 .test_young = kvm_mmu_notifier_test_young,
502 .change_pte = kvm_mmu_notifier_change_pte,
503 .release = kvm_mmu_notifier_release,
506 static int kvm_init_mmu_notifier(struct kvm *kvm)
508 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
509 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
512 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
514 static int kvm_init_mmu_notifier(struct kvm *kvm)
519 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
521 static struct kvm_memslots *kvm_alloc_memslots(void)
524 struct kvm_memslots *slots;
526 slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
530 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
531 slots->id_to_index[i] = slots->memslots[i].id = i;
536 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
538 if (!memslot->dirty_bitmap)
541 kvfree(memslot->dirty_bitmap);
542 memslot->dirty_bitmap = NULL;
546 * Free any memory in @free but not in @dont.
548 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
549 struct kvm_memory_slot *dont)
551 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
552 kvm_destroy_dirty_bitmap(free);
554 kvm_arch_free_memslot(kvm, free, dont);
559 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
561 struct kvm_memory_slot *memslot;
566 kvm_for_each_memslot(memslot, slots)
567 kvm_free_memslot(kvm, memslot, NULL);
572 static void kvm_destroy_vm_debugfs(struct kvm *kvm)
576 if (!kvm->debugfs_dentry)
579 debugfs_remove_recursive(kvm->debugfs_dentry);
581 if (kvm->debugfs_stat_data) {
582 for (i = 0; i < kvm_debugfs_num_entries; i++)
583 kfree(kvm->debugfs_stat_data[i]);
584 kfree(kvm->debugfs_stat_data);
588 static int kvm_create_vm_debugfs(struct kvm *kvm, int fd)
590 char dir_name[ITOA_MAX_LEN * 2];
591 struct kvm_stat_data *stat_data;
592 struct kvm_stats_debugfs_item *p;
594 if (!debugfs_initialized())
597 snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd);
598 kvm->debugfs_dentry = debugfs_create_dir(dir_name, kvm_debugfs_dir);
600 kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries,
601 sizeof(*kvm->debugfs_stat_data),
603 if (!kvm->debugfs_stat_data)
606 for (p = debugfs_entries; p->name; p++) {
607 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL);
611 stat_data->kvm = kvm;
612 stat_data->offset = p->offset;
613 kvm->debugfs_stat_data[p - debugfs_entries] = stat_data;
614 debugfs_create_file(p->name, 0644, kvm->debugfs_dentry,
615 stat_data, stat_fops_per_vm[p->kind]);
620 static struct kvm *kvm_create_vm(unsigned long type)
623 struct kvm *kvm = kvm_arch_alloc_vm();
626 return ERR_PTR(-ENOMEM);
628 spin_lock_init(&kvm->mmu_lock);
630 kvm->mm = current->mm;
631 kvm_eventfd_init(kvm);
632 mutex_init(&kvm->lock);
633 mutex_init(&kvm->irq_lock);
634 mutex_init(&kvm->slots_lock);
635 refcount_set(&kvm->users_count, 1);
636 INIT_LIST_HEAD(&kvm->devices);
638 r = kvm_arch_init_vm(kvm, type);
640 goto out_err_no_disable;
642 r = hardware_enable_all();
644 goto out_err_no_disable;
646 #ifdef CONFIG_HAVE_KVM_IRQFD
647 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
650 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
653 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
654 struct kvm_memslots *slots = kvm_alloc_memslots();
656 goto out_err_no_srcu;
658 * Generations must be different for each address space.
659 * Init kvm generation close to the maximum to easily test the
660 * code of handling generation number wrap-around.
662 slots->generation = i * 2 - 150;
663 rcu_assign_pointer(kvm->memslots[i], slots);
666 if (init_srcu_struct(&kvm->srcu))
667 goto out_err_no_srcu;
668 if (init_srcu_struct(&kvm->irq_srcu))
669 goto out_err_no_irq_srcu;
670 for (i = 0; i < KVM_NR_BUSES; i++) {
671 rcu_assign_pointer(kvm->buses[i],
672 kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL));
677 r = kvm_init_mmu_notifier(kvm);
681 spin_lock(&kvm_lock);
682 list_add(&kvm->vm_list, &vm_list);
683 spin_unlock(&kvm_lock);
685 preempt_notifier_inc();
690 cleanup_srcu_struct(&kvm->irq_srcu);
692 cleanup_srcu_struct(&kvm->srcu);
694 hardware_disable_all();
696 refcount_set(&kvm->users_count, 0);
697 for (i = 0; i < KVM_NR_BUSES; i++)
698 kfree(kvm_get_bus(kvm, i));
699 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
700 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
701 kvm_arch_free_vm(kvm);
706 static void kvm_destroy_devices(struct kvm *kvm)
708 struct kvm_device *dev, *tmp;
711 * We do not need to take the kvm->lock here, because nobody else
712 * has a reference to the struct kvm at this point and therefore
713 * cannot access the devices list anyhow.
715 list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
716 list_del(&dev->vm_node);
717 dev->ops->destroy(dev);
721 static void kvm_destroy_vm(struct kvm *kvm)
724 struct mm_struct *mm = kvm->mm;
726 kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm);
727 kvm_destroy_vm_debugfs(kvm);
728 kvm_arch_sync_events(kvm);
729 spin_lock(&kvm_lock);
730 list_del(&kvm->vm_list);
731 spin_unlock(&kvm_lock);
732 kvm_free_irq_routing(kvm);
733 for (i = 0; i < KVM_NR_BUSES; i++) {
734 struct kvm_io_bus *bus = kvm_get_bus(kvm, i);
737 kvm_io_bus_destroy(bus);
738 kvm->buses[i] = NULL;
740 kvm_coalesced_mmio_free(kvm);
741 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
742 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
744 kvm_arch_flush_shadow_all(kvm);
746 kvm_arch_destroy_vm(kvm);
747 kvm_destroy_devices(kvm);
748 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
749 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
750 cleanup_srcu_struct(&kvm->irq_srcu);
751 cleanup_srcu_struct(&kvm->srcu);
752 kvm_arch_free_vm(kvm);
753 preempt_notifier_dec();
754 hardware_disable_all();
758 void kvm_get_kvm(struct kvm *kvm)
760 refcount_inc(&kvm->users_count);
762 EXPORT_SYMBOL_GPL(kvm_get_kvm);
764 void kvm_put_kvm(struct kvm *kvm)
766 if (refcount_dec_and_test(&kvm->users_count))
769 EXPORT_SYMBOL_GPL(kvm_put_kvm);
772 static int kvm_vm_release(struct inode *inode, struct file *filp)
774 struct kvm *kvm = filp->private_data;
776 kvm_irqfd_release(kvm);
783 * Allocation size is twice as large as the actual dirty bitmap size.
784 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
786 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
788 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
790 memslot->dirty_bitmap = kvzalloc(dirty_bytes, GFP_KERNEL);
791 if (!memslot->dirty_bitmap)
798 * Insert memslot and re-sort memslots based on their GFN,
799 * so binary search could be used to lookup GFN.
800 * Sorting algorithm takes advantage of having initially
801 * sorted array and known changed memslot position.
803 static void update_memslots(struct kvm_memslots *slots,
804 struct kvm_memory_slot *new)
807 int i = slots->id_to_index[id];
808 struct kvm_memory_slot *mslots = slots->memslots;
810 WARN_ON(mslots[i].id != id);
812 WARN_ON(!mslots[i].npages);
813 if (mslots[i].npages)
816 if (!mslots[i].npages)
820 while (i < KVM_MEM_SLOTS_NUM - 1 &&
821 new->base_gfn <= mslots[i + 1].base_gfn) {
822 if (!mslots[i + 1].npages)
824 mslots[i] = mslots[i + 1];
825 slots->id_to_index[mslots[i].id] = i;
830 * The ">=" is needed when creating a slot with base_gfn == 0,
831 * so that it moves before all those with base_gfn == npages == 0.
833 * On the other hand, if new->npages is zero, the above loop has
834 * already left i pointing to the beginning of the empty part of
835 * mslots, and the ">=" would move the hole backwards in this
836 * case---which is wrong. So skip the loop when deleting a slot.
840 new->base_gfn >= mslots[i - 1].base_gfn) {
841 mslots[i] = mslots[i - 1];
842 slots->id_to_index[mslots[i].id] = i;
846 WARN_ON_ONCE(i != slots->used_slots);
849 slots->id_to_index[mslots[i].id] = i;
852 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
854 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
856 #ifdef __KVM_HAVE_READONLY_MEM
857 valid_flags |= KVM_MEM_READONLY;
860 if (mem->flags & ~valid_flags)
866 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
867 int as_id, struct kvm_memslots *slots)
869 struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id);
872 * Set the low bit in the generation, which disables SPTE caching
873 * until the end of synchronize_srcu_expedited.
875 WARN_ON(old_memslots->generation & 1);
876 slots->generation = old_memslots->generation + 1;
878 rcu_assign_pointer(kvm->memslots[as_id], slots);
879 synchronize_srcu_expedited(&kvm->srcu);
882 * Increment the new memslot generation a second time. This prevents
883 * vm exits that race with memslot updates from caching a memslot
884 * generation that will (potentially) be valid forever.
886 * Generations must be unique even across address spaces. We do not need
887 * a global counter for that, instead the generation space is evenly split
888 * across address spaces. For example, with two address spaces, address
889 * space 0 will use generations 0, 4, 8, ... while * address space 1 will
890 * use generations 2, 6, 10, 14, ...
892 slots->generation += KVM_ADDRESS_SPACE_NUM * 2 - 1;
894 kvm_arch_memslots_updated(kvm, slots);
900 * Allocate some memory and give it an address in the guest physical address
903 * Discontiguous memory is allowed, mostly for framebuffers.
905 * Must be called holding kvm->slots_lock for write.
907 int __kvm_set_memory_region(struct kvm *kvm,
908 const struct kvm_userspace_memory_region *mem)
912 unsigned long npages;
913 struct kvm_memory_slot *slot;
914 struct kvm_memory_slot old, new;
915 struct kvm_memslots *slots = NULL, *old_memslots;
917 enum kvm_mr_change change;
919 r = check_memory_region_flags(mem);
924 as_id = mem->slot >> 16;
927 /* General sanity checks */
928 if (mem->memory_size & (PAGE_SIZE - 1))
930 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
932 /* We can read the guest memory with __xxx_user() later on. */
933 if ((id < KVM_USER_MEM_SLOTS) &&
934 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
935 !access_ok(VERIFY_WRITE,
936 (void __user *)(unsigned long)mem->userspace_addr,
939 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
941 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
944 slot = id_to_memslot(__kvm_memslots(kvm, as_id), id);
945 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
946 npages = mem->memory_size >> PAGE_SHIFT;
948 if (npages > KVM_MEM_MAX_NR_PAGES)
954 new.base_gfn = base_gfn;
956 new.flags = mem->flags;
960 change = KVM_MR_CREATE;
961 else { /* Modify an existing slot. */
962 if ((mem->userspace_addr != old.userspace_addr) ||
963 (npages != old.npages) ||
964 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
967 if (base_gfn != old.base_gfn)
968 change = KVM_MR_MOVE;
969 else if (new.flags != old.flags)
970 change = KVM_MR_FLAGS_ONLY;
971 else { /* Nothing to change. */
980 change = KVM_MR_DELETE;
985 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
986 /* Check for overlaps */
988 kvm_for_each_memslot(slot, __kvm_memslots(kvm, as_id)) {
991 if (!((base_gfn + npages <= slot->base_gfn) ||
992 (base_gfn >= slot->base_gfn + slot->npages)))
997 /* Free page dirty bitmap if unneeded */
998 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
999 new.dirty_bitmap = NULL;
1002 if (change == KVM_MR_CREATE) {
1003 new.userspace_addr = mem->userspace_addr;
1005 if (kvm_arch_create_memslot(kvm, &new, npages))
1009 /* Allocate page dirty bitmap if needed */
1010 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
1011 if (kvm_create_dirty_bitmap(&new) < 0)
1015 slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
1018 memcpy(slots, __kvm_memslots(kvm, as_id), sizeof(struct kvm_memslots));
1020 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
1021 slot = id_to_memslot(slots, id);
1022 slot->flags |= KVM_MEMSLOT_INVALID;
1024 old_memslots = install_new_memslots(kvm, as_id, slots);
1026 /* From this point no new shadow pages pointing to a deleted,
1027 * or moved, memslot will be created.
1029 * validation of sp->gfn happens in:
1030 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1031 * - kvm_is_visible_gfn (mmu_check_roots)
1033 kvm_arch_flush_shadow_memslot(kvm, slot);
1036 * We can re-use the old_memslots from above, the only difference
1037 * from the currently installed memslots is the invalid flag. This
1038 * will get overwritten by update_memslots anyway.
1040 slots = old_memslots;
1043 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
1047 /* actual memory is freed via old in kvm_free_memslot below */
1048 if (change == KVM_MR_DELETE) {
1049 new.dirty_bitmap = NULL;
1050 memset(&new.arch, 0, sizeof(new.arch));
1053 update_memslots(slots, &new);
1054 old_memslots = install_new_memslots(kvm, as_id, slots);
1056 kvm_arch_commit_memory_region(kvm, mem, &old, &new, change);
1058 kvm_free_memslot(kvm, &old, &new);
1059 kvfree(old_memslots);
1065 kvm_free_memslot(kvm, &new, &old);
1069 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1071 int kvm_set_memory_region(struct kvm *kvm,
1072 const struct kvm_userspace_memory_region *mem)
1076 mutex_lock(&kvm->slots_lock);
1077 r = __kvm_set_memory_region(kvm, mem);
1078 mutex_unlock(&kvm->slots_lock);
1081 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1083 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1084 struct kvm_userspace_memory_region *mem)
1086 if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1089 return kvm_set_memory_region(kvm, mem);
1092 int kvm_get_dirty_log(struct kvm *kvm,
1093 struct kvm_dirty_log *log, int *is_dirty)
1095 struct kvm_memslots *slots;
1096 struct kvm_memory_slot *memslot;
1099 unsigned long any = 0;
1101 as_id = log->slot >> 16;
1102 id = (u16)log->slot;
1103 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1106 slots = __kvm_memslots(kvm, as_id);
1107 memslot = id_to_memslot(slots, id);
1108 if (!memslot->dirty_bitmap)
1111 n = kvm_dirty_bitmap_bytes(memslot);
1113 for (i = 0; !any && i < n/sizeof(long); ++i)
1114 any = memslot->dirty_bitmap[i];
1116 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1123 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1125 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1127 * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages
1128 * are dirty write protect them for next write.
1129 * @kvm: pointer to kvm instance
1130 * @log: slot id and address to which we copy the log
1131 * @is_dirty: flag set if any page is dirty
1133 * We need to keep it in mind that VCPU threads can write to the bitmap
1134 * concurrently. So, to avoid losing track of dirty pages we keep the
1137 * 1. Take a snapshot of the bit and clear it if needed.
1138 * 2. Write protect the corresponding page.
1139 * 3. Copy the snapshot to the userspace.
1140 * 4. Upon return caller flushes TLB's if needed.
1142 * Between 2 and 4, the guest may write to the page using the remaining TLB
1143 * entry. This is not a problem because the page is reported dirty using
1144 * the snapshot taken before and step 4 ensures that writes done after
1145 * exiting to userspace will be logged for the next call.
1148 int kvm_get_dirty_log_protect(struct kvm *kvm,
1149 struct kvm_dirty_log *log, bool *is_dirty)
1151 struct kvm_memslots *slots;
1152 struct kvm_memory_slot *memslot;
1155 unsigned long *dirty_bitmap;
1156 unsigned long *dirty_bitmap_buffer;
1158 as_id = log->slot >> 16;
1159 id = (u16)log->slot;
1160 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1163 slots = __kvm_memslots(kvm, as_id);
1164 memslot = id_to_memslot(slots, id);
1166 dirty_bitmap = memslot->dirty_bitmap;
1170 n = kvm_dirty_bitmap_bytes(memslot);
1172 dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long);
1173 memset(dirty_bitmap_buffer, 0, n);
1175 spin_lock(&kvm->mmu_lock);
1177 for (i = 0; i < n / sizeof(long); i++) {
1181 if (!dirty_bitmap[i])
1186 mask = xchg(&dirty_bitmap[i], 0);
1187 dirty_bitmap_buffer[i] = mask;
1190 offset = i * BITS_PER_LONG;
1191 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1196 spin_unlock(&kvm->mmu_lock);
1197 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1201 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect);
1204 bool kvm_largepages_enabled(void)
1206 return largepages_enabled;
1209 void kvm_disable_largepages(void)
1211 largepages_enabled = false;
1213 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1215 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1217 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1219 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1221 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
1223 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn);
1226 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1228 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1230 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1231 memslot->flags & KVM_MEMSLOT_INVALID)
1236 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1238 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1240 struct vm_area_struct *vma;
1241 unsigned long addr, size;
1245 addr = gfn_to_hva(kvm, gfn);
1246 if (kvm_is_error_hva(addr))
1249 down_read(¤t->mm->mmap_sem);
1250 vma = find_vma(current->mm, addr);
1254 size = vma_kernel_pagesize(vma);
1257 up_read(¤t->mm->mmap_sem);
1262 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1264 return slot->flags & KVM_MEM_READONLY;
1267 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1268 gfn_t *nr_pages, bool write)
1270 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1271 return KVM_HVA_ERR_BAD;
1273 if (memslot_is_readonly(slot) && write)
1274 return KVM_HVA_ERR_RO_BAD;
1277 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1279 return __gfn_to_hva_memslot(slot, gfn);
1282 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1285 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1288 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1291 return gfn_to_hva_many(slot, gfn, NULL);
1293 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1295 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1297 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1299 EXPORT_SYMBOL_GPL(gfn_to_hva);
1301 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
1303 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
1305 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
1308 * If writable is set to false, the hva returned by this function is only
1309 * allowed to be read.
1311 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1312 gfn_t gfn, bool *writable)
1314 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1316 if (!kvm_is_error_hva(hva) && writable)
1317 *writable = !memslot_is_readonly(slot);
1322 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1324 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1326 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1329 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
1331 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1333 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1336 static inline int check_user_page_hwpoison(unsigned long addr)
1338 int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
1340 rc = get_user_pages(addr, 1, flags, NULL, NULL);
1341 return rc == -EHWPOISON;
1345 * The atomic path to get the writable pfn which will be stored in @pfn,
1346 * true indicates success, otherwise false is returned.
1348 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1349 bool write_fault, bool *writable, kvm_pfn_t *pfn)
1351 struct page *page[1];
1354 if (!(async || atomic))
1358 * Fast pin a writable pfn only if it is a write fault request
1359 * or the caller allows to map a writable pfn for a read fault
1362 if (!(write_fault || writable))
1365 npages = __get_user_pages_fast(addr, 1, 1, page);
1367 *pfn = page_to_pfn(page[0]);
1378 * The slow path to get the pfn of the specified host virtual address,
1379 * 1 indicates success, -errno is returned if error is detected.
1381 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1382 bool *writable, kvm_pfn_t *pfn)
1384 unsigned int flags = FOLL_HWPOISON;
1391 *writable = write_fault;
1394 flags |= FOLL_WRITE;
1396 flags |= FOLL_NOWAIT;
1398 npages = get_user_pages_unlocked(addr, 1, &page, flags);
1402 /* map read fault as writable if possible */
1403 if (unlikely(!write_fault) && writable) {
1406 if (__get_user_pages_fast(addr, 1, 1, &wpage) == 1) {
1412 *pfn = page_to_pfn(page);
1416 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1418 if (unlikely(!(vma->vm_flags & VM_READ)))
1421 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1427 static int hva_to_pfn_remapped(struct vm_area_struct *vma,
1428 unsigned long addr, bool *async,
1429 bool write_fault, bool *writable,
1435 r = follow_pfn(vma, addr, &pfn);
1438 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
1439 * not call the fault handler, so do it here.
1441 bool unlocked = false;
1442 r = fixup_user_fault(current, current->mm, addr,
1443 (write_fault ? FAULT_FLAG_WRITE : 0),
1450 r = follow_pfn(vma, addr, &pfn);
1460 * Get a reference here because callers of *hva_to_pfn* and
1461 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
1462 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
1463 * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will
1464 * simply do nothing for reserved pfns.
1466 * Whoever called remap_pfn_range is also going to call e.g.
1467 * unmap_mapping_range before the underlying pages are freed,
1468 * causing a call to our MMU notifier.
1477 * Pin guest page in memory and return its pfn.
1478 * @addr: host virtual address which maps memory to the guest
1479 * @atomic: whether this function can sleep
1480 * @async: whether this function need to wait IO complete if the
1481 * host page is not in the memory
1482 * @write_fault: whether we should get a writable host page
1483 * @writable: whether it allows to map a writable host page for !@write_fault
1485 * The function will map a writable host page for these two cases:
1486 * 1): @write_fault = true
1487 * 2): @write_fault = false && @writable, @writable will tell the caller
1488 * whether the mapping is writable.
1490 static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1491 bool write_fault, bool *writable)
1493 struct vm_area_struct *vma;
1497 /* we can do it either atomically or asynchronously, not both */
1498 BUG_ON(atomic && async);
1500 if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1504 return KVM_PFN_ERR_FAULT;
1506 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1510 down_read(¤t->mm->mmap_sem);
1511 if (npages == -EHWPOISON ||
1512 (!async && check_user_page_hwpoison(addr))) {
1513 pfn = KVM_PFN_ERR_HWPOISON;
1518 vma = find_vma_intersection(current->mm, addr, addr + 1);
1521 pfn = KVM_PFN_ERR_FAULT;
1522 else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
1523 r = hva_to_pfn_remapped(vma, addr, async, write_fault, writable, &pfn);
1527 pfn = KVM_PFN_ERR_FAULT;
1529 if (async && vma_is_valid(vma, write_fault))
1531 pfn = KVM_PFN_ERR_FAULT;
1534 up_read(¤t->mm->mmap_sem);
1538 kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn,
1539 bool atomic, bool *async, bool write_fault,
1542 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1544 if (addr == KVM_HVA_ERR_RO_BAD) {
1547 return KVM_PFN_ERR_RO_FAULT;
1550 if (kvm_is_error_hva(addr)) {
1553 return KVM_PFN_NOSLOT;
1556 /* Do not map writable pfn in the readonly memslot. */
1557 if (writable && memslot_is_readonly(slot)) {
1562 return hva_to_pfn(addr, atomic, async, write_fault,
1565 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
1567 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1570 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
1571 write_fault, writable);
1573 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1575 kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1577 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1579 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
1581 kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1583 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1585 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1587 kvm_pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1589 return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm, gfn), gfn);
1591 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1593 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
1595 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1597 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
1599 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1601 return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
1603 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1605 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
1607 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1609 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
1611 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1612 struct page **pages, int nr_pages)
1617 addr = gfn_to_hva_many(slot, gfn, &entry);
1618 if (kvm_is_error_hva(addr))
1621 if (entry < nr_pages)
1624 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1626 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1628 static struct page *kvm_pfn_to_page(kvm_pfn_t pfn)
1630 if (is_error_noslot_pfn(pfn))
1631 return KVM_ERR_PTR_BAD_PAGE;
1633 if (kvm_is_reserved_pfn(pfn)) {
1635 return KVM_ERR_PTR_BAD_PAGE;
1638 return pfn_to_page(pfn);
1641 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1645 pfn = gfn_to_pfn(kvm, gfn);
1647 return kvm_pfn_to_page(pfn);
1649 EXPORT_SYMBOL_GPL(gfn_to_page);
1651 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
1655 pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
1657 return kvm_pfn_to_page(pfn);
1659 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
1661 void kvm_release_page_clean(struct page *page)
1663 WARN_ON(is_error_page(page));
1665 kvm_release_pfn_clean(page_to_pfn(page));
1667 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1669 void kvm_release_pfn_clean(kvm_pfn_t pfn)
1671 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
1672 put_page(pfn_to_page(pfn));
1674 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1676 void kvm_release_page_dirty(struct page *page)
1678 WARN_ON(is_error_page(page));
1680 kvm_release_pfn_dirty(page_to_pfn(page));
1682 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1684 void kvm_release_pfn_dirty(kvm_pfn_t pfn)
1686 kvm_set_pfn_dirty(pfn);
1687 kvm_release_pfn_clean(pfn);
1689 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1691 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
1693 if (!kvm_is_reserved_pfn(pfn)) {
1694 struct page *page = pfn_to_page(pfn);
1696 if (!PageReserved(page))
1700 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1702 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
1704 if (!kvm_is_reserved_pfn(pfn))
1705 mark_page_accessed(pfn_to_page(pfn));
1707 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1709 void kvm_get_pfn(kvm_pfn_t pfn)
1711 if (!kvm_is_reserved_pfn(pfn))
1712 get_page(pfn_to_page(pfn));
1714 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1716 static int next_segment(unsigned long len, int offset)
1718 if (len > PAGE_SIZE - offset)
1719 return PAGE_SIZE - offset;
1724 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
1725 void *data, int offset, int len)
1730 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1731 if (kvm_is_error_hva(addr))
1733 r = __copy_from_user(data, (void __user *)addr + offset, len);
1739 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1742 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1744 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1746 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1748 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
1749 int offset, int len)
1751 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1753 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1755 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
1757 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1759 gfn_t gfn = gpa >> PAGE_SHIFT;
1761 int offset = offset_in_page(gpa);
1764 while ((seg = next_segment(len, offset)) != 0) {
1765 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1775 EXPORT_SYMBOL_GPL(kvm_read_guest);
1777 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
1779 gfn_t gfn = gpa >> PAGE_SHIFT;
1781 int offset = offset_in_page(gpa);
1784 while ((seg = next_segment(len, offset)) != 0) {
1785 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
1795 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
1797 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1798 void *data, int offset, unsigned long len)
1803 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1804 if (kvm_is_error_hva(addr))
1806 pagefault_disable();
1807 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1814 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1817 gfn_t gfn = gpa >> PAGE_SHIFT;
1818 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1819 int offset = offset_in_page(gpa);
1821 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1823 EXPORT_SYMBOL_GPL(kvm_read_guest_atomic);
1825 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
1826 void *data, unsigned long len)
1828 gfn_t gfn = gpa >> PAGE_SHIFT;
1829 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1830 int offset = offset_in_page(gpa);
1832 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1834 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
1836 static int __kvm_write_guest_page(struct kvm_memory_slot *memslot, gfn_t gfn,
1837 const void *data, int offset, int len)
1842 addr = gfn_to_hva_memslot(memslot, gfn);
1843 if (kvm_is_error_hva(addr))
1845 r = __copy_to_user((void __user *)addr + offset, data, len);
1848 mark_page_dirty_in_slot(memslot, gfn);
1852 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
1853 const void *data, int offset, int len)
1855 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1857 return __kvm_write_guest_page(slot, gfn, data, offset, len);
1859 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1861 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
1862 const void *data, int offset, int len)
1864 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1866 return __kvm_write_guest_page(slot, gfn, data, offset, len);
1868 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
1870 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1873 gfn_t gfn = gpa >> PAGE_SHIFT;
1875 int offset = offset_in_page(gpa);
1878 while ((seg = next_segment(len, offset)) != 0) {
1879 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1889 EXPORT_SYMBOL_GPL(kvm_write_guest);
1891 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
1894 gfn_t gfn = gpa >> PAGE_SHIFT;
1896 int offset = offset_in_page(gpa);
1899 while ((seg = next_segment(len, offset)) != 0) {
1900 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
1910 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
1912 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
1913 struct gfn_to_hva_cache *ghc,
1914 gpa_t gpa, unsigned long len)
1916 int offset = offset_in_page(gpa);
1917 gfn_t start_gfn = gpa >> PAGE_SHIFT;
1918 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1919 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1920 gfn_t nr_pages_avail;
1923 ghc->generation = slots->generation;
1925 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
1926 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, NULL);
1927 if (!kvm_is_error_hva(ghc->hva) && nr_pages_needed <= 1) {
1931 * If the requested region crosses two memslots, we still
1932 * verify that the entire region is valid here.
1934 while (start_gfn <= end_gfn) {
1936 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
1937 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1939 if (kvm_is_error_hva(ghc->hva))
1941 start_gfn += nr_pages_avail;
1943 /* Use the slow path for cross page reads and writes. */
1944 ghc->memslot = NULL;
1949 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1950 gpa_t gpa, unsigned long len)
1952 struct kvm_memslots *slots = kvm_memslots(kvm);
1953 return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len);
1955 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1957 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1958 void *data, int offset, unsigned long len)
1960 struct kvm_memslots *slots = kvm_memslots(kvm);
1962 gpa_t gpa = ghc->gpa + offset;
1964 BUG_ON(len + offset > ghc->len);
1966 if (slots->generation != ghc->generation)
1967 __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len);
1969 if (unlikely(!ghc->memslot))
1970 return kvm_write_guest(kvm, gpa, data, len);
1972 if (kvm_is_error_hva(ghc->hva))
1975 r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
1978 mark_page_dirty_in_slot(ghc->memslot, gpa >> PAGE_SHIFT);
1982 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached);
1984 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1985 void *data, unsigned long len)
1987 return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len);
1989 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1991 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1992 void *data, unsigned long len)
1994 struct kvm_memslots *slots = kvm_memslots(kvm);
1997 BUG_ON(len > ghc->len);
1999 if (slots->generation != ghc->generation)
2000 __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len);
2002 if (unlikely(!ghc->memslot))
2003 return kvm_read_guest(kvm, ghc->gpa, data, len);
2005 if (kvm_is_error_hva(ghc->hva))
2008 r = __copy_from_user(data, (void __user *)ghc->hva, len);
2014 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
2016 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
2018 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
2020 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
2022 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
2024 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
2026 gfn_t gfn = gpa >> PAGE_SHIFT;
2028 int offset = offset_in_page(gpa);
2031 while ((seg = next_segment(len, offset)) != 0) {
2032 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
2041 EXPORT_SYMBOL_GPL(kvm_clear_guest);
2043 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot,
2046 if (memslot && memslot->dirty_bitmap) {
2047 unsigned long rel_gfn = gfn - memslot->base_gfn;
2049 set_bit_le(rel_gfn, memslot->dirty_bitmap);
2053 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
2055 struct kvm_memory_slot *memslot;
2057 memslot = gfn_to_memslot(kvm, gfn);
2058 mark_page_dirty_in_slot(memslot, gfn);
2060 EXPORT_SYMBOL_GPL(mark_page_dirty);
2062 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
2064 struct kvm_memory_slot *memslot;
2066 memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2067 mark_page_dirty_in_slot(memslot, gfn);
2069 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
2071 void kvm_sigset_activate(struct kvm_vcpu *vcpu)
2073 if (!vcpu->sigset_active)
2077 * This does a lockless modification of ->real_blocked, which is fine
2078 * because, only current can change ->real_blocked and all readers of
2079 * ->real_blocked don't care as long ->real_blocked is always a subset
2082 sigprocmask(SIG_SETMASK, &vcpu->sigset, ¤t->real_blocked);
2085 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu)
2087 if (!vcpu->sigset_active)
2090 sigprocmask(SIG_SETMASK, ¤t->real_blocked, NULL);
2091 sigemptyset(¤t->real_blocked);
2094 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
2096 unsigned int old, val, grow;
2098 old = val = vcpu->halt_poll_ns;
2099 grow = READ_ONCE(halt_poll_ns_grow);
2101 if (val == 0 && grow)
2106 if (val > halt_poll_ns)
2109 vcpu->halt_poll_ns = val;
2110 trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
2113 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
2115 unsigned int old, val, shrink;
2117 old = val = vcpu->halt_poll_ns;
2118 shrink = READ_ONCE(halt_poll_ns_shrink);
2124 vcpu->halt_poll_ns = val;
2125 trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
2128 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
2130 if (kvm_arch_vcpu_runnable(vcpu)) {
2131 kvm_make_request(KVM_REQ_UNHALT, vcpu);
2134 if (kvm_cpu_has_pending_timer(vcpu))
2136 if (signal_pending(current))
2143 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2145 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
2148 DECLARE_SWAITQUEUE(wait);
2149 bool waited = false;
2152 start = cur = ktime_get();
2153 if (vcpu->halt_poll_ns) {
2154 ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns);
2156 ++vcpu->stat.halt_attempted_poll;
2159 * This sets KVM_REQ_UNHALT if an interrupt
2162 if (kvm_vcpu_check_block(vcpu) < 0) {
2163 ++vcpu->stat.halt_successful_poll;
2164 if (!vcpu_valid_wakeup(vcpu))
2165 ++vcpu->stat.halt_poll_invalid;
2169 } while (single_task_running() && ktime_before(cur, stop));
2172 kvm_arch_vcpu_blocking(vcpu);
2175 prepare_to_swait_exclusive(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
2177 if (kvm_vcpu_check_block(vcpu) < 0)
2184 finish_swait(&vcpu->wq, &wait);
2187 kvm_arch_vcpu_unblocking(vcpu);
2189 block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
2191 if (!vcpu_valid_wakeup(vcpu))
2192 shrink_halt_poll_ns(vcpu);
2193 else if (halt_poll_ns) {
2194 if (block_ns <= vcpu->halt_poll_ns)
2196 /* we had a long block, shrink polling */
2197 else if (vcpu->halt_poll_ns && block_ns > halt_poll_ns)
2198 shrink_halt_poll_ns(vcpu);
2199 /* we had a short halt and our poll time is too small */
2200 else if (vcpu->halt_poll_ns < halt_poll_ns &&
2201 block_ns < halt_poll_ns)
2202 grow_halt_poll_ns(vcpu);
2204 vcpu->halt_poll_ns = 0;
2206 trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu));
2207 kvm_arch_vcpu_block_finish(vcpu);
2209 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
2211 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
2213 struct swait_queue_head *wqp;
2215 wqp = kvm_arch_vcpu_wq(vcpu);
2216 if (swq_has_sleeper(wqp)) {
2218 ++vcpu->stat.halt_wakeup;
2224 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
2228 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2230 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
2233 int cpu = vcpu->cpu;
2235 if (kvm_vcpu_wake_up(vcpu))
2239 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
2240 if (kvm_arch_vcpu_should_kick(vcpu))
2241 smp_send_reschedule(cpu);
2244 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
2245 #endif /* !CONFIG_S390 */
2247 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
2250 struct task_struct *task = NULL;
2254 pid = rcu_dereference(target->pid);
2256 task = get_pid_task(pid, PIDTYPE_PID);
2260 ret = yield_to(task, 1);
2261 put_task_struct(task);
2265 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
2268 * Helper that checks whether a VCPU is eligible for directed yield.
2269 * Most eligible candidate to yield is decided by following heuristics:
2271 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2272 * (preempted lock holder), indicated by @in_spin_loop.
2273 * Set at the beiginning and cleared at the end of interception/PLE handler.
2275 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2276 * chance last time (mostly it has become eligible now since we have probably
2277 * yielded to lockholder in last iteration. This is done by toggling
2278 * @dy_eligible each time a VCPU checked for eligibility.)
2280 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2281 * to preempted lock-holder could result in wrong VCPU selection and CPU
2282 * burning. Giving priority for a potential lock-holder increases lock
2285 * Since algorithm is based on heuristics, accessing another VCPU data without
2286 * locking does not harm. It may result in trying to yield to same VCPU, fail
2287 * and continue with next VCPU and so on.
2289 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
2291 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2294 eligible = !vcpu->spin_loop.in_spin_loop ||
2295 vcpu->spin_loop.dy_eligible;
2297 if (vcpu->spin_loop.in_spin_loop)
2298 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
2306 void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode)
2308 struct kvm *kvm = me->kvm;
2309 struct kvm_vcpu *vcpu;
2310 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
2316 kvm_vcpu_set_in_spin_loop(me, true);
2318 * We boost the priority of a VCPU that is runnable but not
2319 * currently running, because it got preempted by something
2320 * else and called schedule in __vcpu_run. Hopefully that
2321 * VCPU is holding the lock that we need and will release it.
2322 * We approximate round-robin by starting at the last boosted VCPU.
2324 for (pass = 0; pass < 2 && !yielded && try; pass++) {
2325 kvm_for_each_vcpu(i, vcpu, kvm) {
2326 if (!pass && i <= last_boosted_vcpu) {
2327 i = last_boosted_vcpu;
2329 } else if (pass && i > last_boosted_vcpu)
2331 if (!READ_ONCE(vcpu->preempted))
2335 if (swait_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
2337 if (yield_to_kernel_mode && !kvm_arch_vcpu_in_kernel(vcpu))
2339 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
2342 yielded = kvm_vcpu_yield_to(vcpu);
2344 kvm->last_boosted_vcpu = i;
2346 } else if (yielded < 0) {
2353 kvm_vcpu_set_in_spin_loop(me, false);
2355 /* Ensure vcpu is not eligible during next spinloop */
2356 kvm_vcpu_set_dy_eligible(me, false);
2358 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
2360 static vm_fault_t kvm_vcpu_fault(struct vm_fault *vmf)
2362 struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data;
2365 if (vmf->pgoff == 0)
2366 page = virt_to_page(vcpu->run);
2368 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
2369 page = virt_to_page(vcpu->arch.pio_data);
2371 #ifdef CONFIG_KVM_MMIO
2372 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
2373 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
2376 return kvm_arch_vcpu_fault(vcpu, vmf);
2382 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
2383 .fault = kvm_vcpu_fault,
2386 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2388 vma->vm_ops = &kvm_vcpu_vm_ops;
2392 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2394 struct kvm_vcpu *vcpu = filp->private_data;
2396 debugfs_remove_recursive(vcpu->debugfs_dentry);
2397 kvm_put_kvm(vcpu->kvm);
2401 static struct file_operations kvm_vcpu_fops = {
2402 .release = kvm_vcpu_release,
2403 .unlocked_ioctl = kvm_vcpu_ioctl,
2404 .mmap = kvm_vcpu_mmap,
2405 .llseek = noop_llseek,
2406 KVM_COMPAT(kvm_vcpu_compat_ioctl),
2410 * Allocates an inode for the vcpu.
2412 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2414 char name[8 + 1 + ITOA_MAX_LEN + 1];
2416 snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id);
2417 return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
2420 static int kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
2422 char dir_name[ITOA_MAX_LEN * 2];
2425 if (!kvm_arch_has_vcpu_debugfs())
2428 if (!debugfs_initialized())
2431 snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
2432 vcpu->debugfs_dentry = debugfs_create_dir(dir_name,
2433 vcpu->kvm->debugfs_dentry);
2434 if (!vcpu->debugfs_dentry)
2437 ret = kvm_arch_create_vcpu_debugfs(vcpu);
2439 debugfs_remove_recursive(vcpu->debugfs_dentry);
2447 * Creates some virtual cpus. Good luck creating more than one.
2449 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
2452 struct kvm_vcpu *vcpu;
2454 if (id >= KVM_MAX_VCPU_ID)
2457 mutex_lock(&kvm->lock);
2458 if (kvm->created_vcpus == KVM_MAX_VCPUS) {
2459 mutex_unlock(&kvm->lock);
2463 kvm->created_vcpus++;
2464 mutex_unlock(&kvm->lock);
2466 vcpu = kvm_arch_vcpu_create(kvm, id);
2469 goto vcpu_decrement;
2472 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2474 r = kvm_arch_vcpu_setup(vcpu);
2478 r = kvm_create_vcpu_debugfs(vcpu);
2482 mutex_lock(&kvm->lock);
2483 if (kvm_get_vcpu_by_id(kvm, id)) {
2485 goto unlock_vcpu_destroy;
2488 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
2490 /* Now it's all set up, let userspace reach it */
2492 r = create_vcpu_fd(vcpu);
2495 goto unlock_vcpu_destroy;
2498 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
2501 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
2502 * before kvm->online_vcpu's incremented value.
2505 atomic_inc(&kvm->online_vcpus);
2507 mutex_unlock(&kvm->lock);
2508 kvm_arch_vcpu_postcreate(vcpu);
2511 unlock_vcpu_destroy:
2512 mutex_unlock(&kvm->lock);
2513 debugfs_remove_recursive(vcpu->debugfs_dentry);
2515 kvm_arch_vcpu_destroy(vcpu);
2517 mutex_lock(&kvm->lock);
2518 kvm->created_vcpus--;
2519 mutex_unlock(&kvm->lock);
2523 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2526 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2527 vcpu->sigset_active = 1;
2528 vcpu->sigset = *sigset;
2530 vcpu->sigset_active = 0;
2534 static long kvm_vcpu_ioctl(struct file *filp,
2535 unsigned int ioctl, unsigned long arg)
2537 struct kvm_vcpu *vcpu = filp->private_data;
2538 void __user *argp = (void __user *)arg;
2540 struct kvm_fpu *fpu = NULL;
2541 struct kvm_sregs *kvm_sregs = NULL;
2543 if (vcpu->kvm->mm != current->mm)
2546 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2550 * Some architectures have vcpu ioctls that are asynchronous to vcpu
2551 * execution; mutex_lock() would break them.
2553 r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg);
2554 if (r != -ENOIOCTLCMD)
2557 if (mutex_lock_killable(&vcpu->mutex))
2565 oldpid = rcu_access_pointer(vcpu->pid);
2566 if (unlikely(oldpid != current->pids[PIDTYPE_PID].pid)) {
2567 /* The thread running this VCPU changed. */
2570 r = kvm_arch_vcpu_run_pid_change(vcpu);
2574 newpid = get_task_pid(current, PIDTYPE_PID);
2575 rcu_assign_pointer(vcpu->pid, newpid);
2580 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2581 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2584 case KVM_GET_REGS: {
2585 struct kvm_regs *kvm_regs;
2588 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2591 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2595 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2602 case KVM_SET_REGS: {
2603 struct kvm_regs *kvm_regs;
2606 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2607 if (IS_ERR(kvm_regs)) {
2608 r = PTR_ERR(kvm_regs);
2611 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2615 case KVM_GET_SREGS: {
2616 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2620 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2624 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2629 case KVM_SET_SREGS: {
2630 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2631 if (IS_ERR(kvm_sregs)) {
2632 r = PTR_ERR(kvm_sregs);
2636 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2639 case KVM_GET_MP_STATE: {
2640 struct kvm_mp_state mp_state;
2642 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2646 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
2651 case KVM_SET_MP_STATE: {
2652 struct kvm_mp_state mp_state;
2655 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
2657 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2660 case KVM_TRANSLATE: {
2661 struct kvm_translation tr;
2664 if (copy_from_user(&tr, argp, sizeof(tr)))
2666 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2670 if (copy_to_user(argp, &tr, sizeof(tr)))
2675 case KVM_SET_GUEST_DEBUG: {
2676 struct kvm_guest_debug dbg;
2679 if (copy_from_user(&dbg, argp, sizeof(dbg)))
2681 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2684 case KVM_SET_SIGNAL_MASK: {
2685 struct kvm_signal_mask __user *sigmask_arg = argp;
2686 struct kvm_signal_mask kvm_sigmask;
2687 sigset_t sigset, *p;
2692 if (copy_from_user(&kvm_sigmask, argp,
2693 sizeof(kvm_sigmask)))
2696 if (kvm_sigmask.len != sizeof(sigset))
2699 if (copy_from_user(&sigset, sigmask_arg->sigset,
2704 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2708 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2712 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2716 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2722 fpu = memdup_user(argp, sizeof(*fpu));
2728 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2732 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2735 mutex_unlock(&vcpu->mutex);
2741 #ifdef CONFIG_KVM_COMPAT
2742 static long kvm_vcpu_compat_ioctl(struct file *filp,
2743 unsigned int ioctl, unsigned long arg)
2745 struct kvm_vcpu *vcpu = filp->private_data;
2746 void __user *argp = compat_ptr(arg);
2749 if (vcpu->kvm->mm != current->mm)
2753 case KVM_SET_SIGNAL_MASK: {
2754 struct kvm_signal_mask __user *sigmask_arg = argp;
2755 struct kvm_signal_mask kvm_sigmask;
2760 if (copy_from_user(&kvm_sigmask, argp,
2761 sizeof(kvm_sigmask)))
2764 if (kvm_sigmask.len != sizeof(compat_sigset_t))
2767 if (get_compat_sigset(&sigset, (void *)sigmask_arg->sigset))
2769 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2771 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2775 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2783 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2784 int (*accessor)(struct kvm_device *dev,
2785 struct kvm_device_attr *attr),
2788 struct kvm_device_attr attr;
2793 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2796 return accessor(dev, &attr);
2799 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2802 struct kvm_device *dev = filp->private_data;
2805 case KVM_SET_DEVICE_ATTR:
2806 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2807 case KVM_GET_DEVICE_ATTR:
2808 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2809 case KVM_HAS_DEVICE_ATTR:
2810 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2812 if (dev->ops->ioctl)
2813 return dev->ops->ioctl(dev, ioctl, arg);
2819 static int kvm_device_release(struct inode *inode, struct file *filp)
2821 struct kvm_device *dev = filp->private_data;
2822 struct kvm *kvm = dev->kvm;
2828 static const struct file_operations kvm_device_fops = {
2829 .unlocked_ioctl = kvm_device_ioctl,
2830 .release = kvm_device_release,
2831 KVM_COMPAT(kvm_device_ioctl),
2834 struct kvm_device *kvm_device_from_filp(struct file *filp)
2836 if (filp->f_op != &kvm_device_fops)
2839 return filp->private_data;
2842 static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
2843 #ifdef CONFIG_KVM_MPIC
2844 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
2845 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
2849 int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
2851 if (type >= ARRAY_SIZE(kvm_device_ops_table))
2854 if (kvm_device_ops_table[type] != NULL)
2857 kvm_device_ops_table[type] = ops;
2861 void kvm_unregister_device_ops(u32 type)
2863 if (kvm_device_ops_table[type] != NULL)
2864 kvm_device_ops_table[type] = NULL;
2867 static int kvm_ioctl_create_device(struct kvm *kvm,
2868 struct kvm_create_device *cd)
2870 struct kvm_device_ops *ops = NULL;
2871 struct kvm_device *dev;
2872 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2875 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
2878 ops = kvm_device_ops_table[cd->type];
2885 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2892 mutex_lock(&kvm->lock);
2893 ret = ops->create(dev, cd->type);
2895 mutex_unlock(&kvm->lock);
2899 list_add(&dev->vm_node, &kvm->devices);
2900 mutex_unlock(&kvm->lock);
2905 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2907 mutex_lock(&kvm->lock);
2908 list_del(&dev->vm_node);
2909 mutex_unlock(&kvm->lock);
2919 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2922 case KVM_CAP_USER_MEMORY:
2923 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2924 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2925 case KVM_CAP_INTERNAL_ERROR_DATA:
2926 #ifdef CONFIG_HAVE_KVM_MSI
2927 case KVM_CAP_SIGNAL_MSI:
2929 #ifdef CONFIG_HAVE_KVM_IRQFD
2931 case KVM_CAP_IRQFD_RESAMPLE:
2933 case KVM_CAP_IOEVENTFD_ANY_LENGTH:
2934 case KVM_CAP_CHECK_EXTENSION_VM:
2936 #ifdef CONFIG_KVM_MMIO
2937 case KVM_CAP_COALESCED_MMIO:
2938 return KVM_COALESCED_MMIO_PAGE_OFFSET;
2940 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2941 case KVM_CAP_IRQ_ROUTING:
2942 return KVM_MAX_IRQ_ROUTES;
2944 #if KVM_ADDRESS_SPACE_NUM > 1
2945 case KVM_CAP_MULTI_ADDRESS_SPACE:
2946 return KVM_ADDRESS_SPACE_NUM;
2948 case KVM_CAP_MAX_VCPU_ID:
2949 return KVM_MAX_VCPU_ID;
2953 return kvm_vm_ioctl_check_extension(kvm, arg);
2956 static long kvm_vm_ioctl(struct file *filp,
2957 unsigned int ioctl, unsigned long arg)
2959 struct kvm *kvm = filp->private_data;
2960 void __user *argp = (void __user *)arg;
2963 if (kvm->mm != current->mm)
2966 case KVM_CREATE_VCPU:
2967 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2969 case KVM_SET_USER_MEMORY_REGION: {
2970 struct kvm_userspace_memory_region kvm_userspace_mem;
2973 if (copy_from_user(&kvm_userspace_mem, argp,
2974 sizeof(kvm_userspace_mem)))
2977 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2980 case KVM_GET_DIRTY_LOG: {
2981 struct kvm_dirty_log log;
2984 if (copy_from_user(&log, argp, sizeof(log)))
2986 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2989 #ifdef CONFIG_KVM_MMIO
2990 case KVM_REGISTER_COALESCED_MMIO: {
2991 struct kvm_coalesced_mmio_zone zone;
2994 if (copy_from_user(&zone, argp, sizeof(zone)))
2996 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2999 case KVM_UNREGISTER_COALESCED_MMIO: {
3000 struct kvm_coalesced_mmio_zone zone;
3003 if (copy_from_user(&zone, argp, sizeof(zone)))
3005 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
3010 struct kvm_irqfd data;
3013 if (copy_from_user(&data, argp, sizeof(data)))
3015 r = kvm_irqfd(kvm, &data);
3018 case KVM_IOEVENTFD: {
3019 struct kvm_ioeventfd data;
3022 if (copy_from_user(&data, argp, sizeof(data)))
3024 r = kvm_ioeventfd(kvm, &data);
3027 #ifdef CONFIG_HAVE_KVM_MSI
3028 case KVM_SIGNAL_MSI: {
3032 if (copy_from_user(&msi, argp, sizeof(msi)))
3034 r = kvm_send_userspace_msi(kvm, &msi);
3038 #ifdef __KVM_HAVE_IRQ_LINE
3039 case KVM_IRQ_LINE_STATUS:
3040 case KVM_IRQ_LINE: {
3041 struct kvm_irq_level irq_event;
3044 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
3047 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
3048 ioctl == KVM_IRQ_LINE_STATUS);
3053 if (ioctl == KVM_IRQ_LINE_STATUS) {
3054 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
3062 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3063 case KVM_SET_GSI_ROUTING: {
3064 struct kvm_irq_routing routing;
3065 struct kvm_irq_routing __user *urouting;
3066 struct kvm_irq_routing_entry *entries = NULL;
3069 if (copy_from_user(&routing, argp, sizeof(routing)))
3072 if (!kvm_arch_can_set_irq_routing(kvm))
3074 if (routing.nr > KVM_MAX_IRQ_ROUTES)
3080 entries = vmalloc(array_size(sizeof(*entries),
3086 if (copy_from_user(entries, urouting->entries,
3087 routing.nr * sizeof(*entries)))
3088 goto out_free_irq_routing;
3090 r = kvm_set_irq_routing(kvm, entries, routing.nr,
3092 out_free_irq_routing:
3096 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
3097 case KVM_CREATE_DEVICE: {
3098 struct kvm_create_device cd;
3101 if (copy_from_user(&cd, argp, sizeof(cd)))
3104 r = kvm_ioctl_create_device(kvm, &cd);
3109 if (copy_to_user(argp, &cd, sizeof(cd)))
3115 case KVM_CHECK_EXTENSION:
3116 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
3119 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
3125 #ifdef CONFIG_KVM_COMPAT
3126 struct compat_kvm_dirty_log {
3130 compat_uptr_t dirty_bitmap; /* one bit per page */
3135 static long kvm_vm_compat_ioctl(struct file *filp,
3136 unsigned int ioctl, unsigned long arg)
3138 struct kvm *kvm = filp->private_data;
3141 if (kvm->mm != current->mm)
3144 case KVM_GET_DIRTY_LOG: {
3145 struct compat_kvm_dirty_log compat_log;
3146 struct kvm_dirty_log log;
3148 if (copy_from_user(&compat_log, (void __user *)arg,
3149 sizeof(compat_log)))
3151 log.slot = compat_log.slot;
3152 log.padding1 = compat_log.padding1;
3153 log.padding2 = compat_log.padding2;
3154 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
3156 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3160 r = kvm_vm_ioctl(filp, ioctl, arg);
3166 static struct file_operations kvm_vm_fops = {
3167 .release = kvm_vm_release,
3168 .unlocked_ioctl = kvm_vm_ioctl,
3169 .llseek = noop_llseek,
3170 KVM_COMPAT(kvm_vm_compat_ioctl),
3173 static int kvm_dev_ioctl_create_vm(unsigned long type)
3179 kvm = kvm_create_vm(type);
3181 return PTR_ERR(kvm);
3182 #ifdef CONFIG_KVM_MMIO
3183 r = kvm_coalesced_mmio_init(kvm);
3187 r = get_unused_fd_flags(O_CLOEXEC);
3191 file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
3199 * Don't call kvm_put_kvm anymore at this point; file->f_op is
3200 * already set, with ->release() being kvm_vm_release(). In error
3201 * cases it will be called by the final fput(file) and will take
3202 * care of doing kvm_put_kvm(kvm).
3204 if (kvm_create_vm_debugfs(kvm, r) < 0) {
3209 kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm);
3211 fd_install(r, file);
3219 static long kvm_dev_ioctl(struct file *filp,
3220 unsigned int ioctl, unsigned long arg)
3225 case KVM_GET_API_VERSION:
3228 r = KVM_API_VERSION;
3231 r = kvm_dev_ioctl_create_vm(arg);
3233 case KVM_CHECK_EXTENSION:
3234 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
3236 case KVM_GET_VCPU_MMAP_SIZE:
3239 r = PAGE_SIZE; /* struct kvm_run */
3241 r += PAGE_SIZE; /* pio data page */
3243 #ifdef CONFIG_KVM_MMIO
3244 r += PAGE_SIZE; /* coalesced mmio ring page */
3247 case KVM_TRACE_ENABLE:
3248 case KVM_TRACE_PAUSE:
3249 case KVM_TRACE_DISABLE:
3253 return kvm_arch_dev_ioctl(filp, ioctl, arg);
3259 static struct file_operations kvm_chardev_ops = {
3260 .unlocked_ioctl = kvm_dev_ioctl,
3261 .llseek = noop_llseek,
3262 KVM_COMPAT(kvm_dev_ioctl),
3265 static struct miscdevice kvm_dev = {
3271 static void hardware_enable_nolock(void *junk)
3273 int cpu = raw_smp_processor_id();
3276 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
3279 cpumask_set_cpu(cpu, cpus_hardware_enabled);
3281 r = kvm_arch_hardware_enable();
3284 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3285 atomic_inc(&hardware_enable_failed);
3286 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
3290 static int kvm_starting_cpu(unsigned int cpu)
3292 raw_spin_lock(&kvm_count_lock);
3293 if (kvm_usage_count)
3294 hardware_enable_nolock(NULL);
3295 raw_spin_unlock(&kvm_count_lock);
3299 static void hardware_disable_nolock(void *junk)
3301 int cpu = raw_smp_processor_id();
3303 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
3305 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3306 kvm_arch_hardware_disable();
3309 static int kvm_dying_cpu(unsigned int cpu)
3311 raw_spin_lock(&kvm_count_lock);
3312 if (kvm_usage_count)
3313 hardware_disable_nolock(NULL);
3314 raw_spin_unlock(&kvm_count_lock);
3318 static void hardware_disable_all_nolock(void)
3320 BUG_ON(!kvm_usage_count);
3323 if (!kvm_usage_count)
3324 on_each_cpu(hardware_disable_nolock, NULL, 1);
3327 static void hardware_disable_all(void)
3329 raw_spin_lock(&kvm_count_lock);
3330 hardware_disable_all_nolock();
3331 raw_spin_unlock(&kvm_count_lock);
3334 static int hardware_enable_all(void)
3338 raw_spin_lock(&kvm_count_lock);
3341 if (kvm_usage_count == 1) {
3342 atomic_set(&hardware_enable_failed, 0);
3343 on_each_cpu(hardware_enable_nolock, NULL, 1);
3345 if (atomic_read(&hardware_enable_failed)) {
3346 hardware_disable_all_nolock();
3351 raw_spin_unlock(&kvm_count_lock);
3356 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3360 * Some (well, at least mine) BIOSes hang on reboot if
3363 * And Intel TXT required VMX off for all cpu when system shutdown.
3365 pr_info("kvm: exiting hardware virtualization\n");
3366 kvm_rebooting = true;
3367 on_each_cpu(hardware_disable_nolock, NULL, 1);
3371 static struct notifier_block kvm_reboot_notifier = {
3372 .notifier_call = kvm_reboot,
3376 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3380 for (i = 0; i < bus->dev_count; i++) {
3381 struct kvm_io_device *pos = bus->range[i].dev;
3383 kvm_iodevice_destructor(pos);
3388 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
3389 const struct kvm_io_range *r2)
3391 gpa_t addr1 = r1->addr;
3392 gpa_t addr2 = r2->addr;
3397 /* If r2->len == 0, match the exact address. If r2->len != 0,
3398 * accept any overlapping write. Any order is acceptable for
3399 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
3400 * we process all of them.
3413 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
3415 return kvm_io_bus_cmp(p1, p2);
3418 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
3419 gpa_t addr, int len)
3421 struct kvm_io_range *range, key;
3424 key = (struct kvm_io_range) {
3429 range = bsearch(&key, bus->range, bus->dev_count,
3430 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
3434 off = range - bus->range;
3436 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
3442 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3443 struct kvm_io_range *range, const void *val)
3447 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3451 while (idx < bus->dev_count &&
3452 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3453 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
3462 /* kvm_io_bus_write - called under kvm->slots_lock */
3463 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3464 int len, const void *val)
3466 struct kvm_io_bus *bus;
3467 struct kvm_io_range range;
3470 range = (struct kvm_io_range) {
3475 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3478 r = __kvm_io_bus_write(vcpu, bus, &range, val);
3479 return r < 0 ? r : 0;
3482 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3483 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
3484 gpa_t addr, int len, const void *val, long cookie)
3486 struct kvm_io_bus *bus;
3487 struct kvm_io_range range;
3489 range = (struct kvm_io_range) {
3494 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3498 /* First try the device referenced by cookie. */
3499 if ((cookie >= 0) && (cookie < bus->dev_count) &&
3500 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
3501 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
3506 * cookie contained garbage; fall back to search and return the
3507 * correct cookie value.
3509 return __kvm_io_bus_write(vcpu, bus, &range, val);
3512 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3513 struct kvm_io_range *range, void *val)
3517 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3521 while (idx < bus->dev_count &&
3522 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3523 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
3531 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
3533 /* kvm_io_bus_read - called under kvm->slots_lock */
3534 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3537 struct kvm_io_bus *bus;
3538 struct kvm_io_range range;
3541 range = (struct kvm_io_range) {
3546 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3549 r = __kvm_io_bus_read(vcpu, bus, &range, val);
3550 return r < 0 ? r : 0;
3554 /* Caller must hold slots_lock. */
3555 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3556 int len, struct kvm_io_device *dev)
3559 struct kvm_io_bus *new_bus, *bus;
3560 struct kvm_io_range range;
3562 bus = kvm_get_bus(kvm, bus_idx);
3566 /* exclude ioeventfd which is limited by maximum fd */
3567 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3570 new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count + 1) *
3571 sizeof(struct kvm_io_range)), GFP_KERNEL);
3575 range = (struct kvm_io_range) {
3581 for (i = 0; i < bus->dev_count; i++)
3582 if (kvm_io_bus_cmp(&bus->range[i], &range) > 0)
3585 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3586 new_bus->dev_count++;
3587 new_bus->range[i] = range;
3588 memcpy(new_bus->range + i + 1, bus->range + i,
3589 (bus->dev_count - i) * sizeof(struct kvm_io_range));
3590 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3591 synchronize_srcu_expedited(&kvm->srcu);
3597 /* Caller must hold slots_lock. */
3598 void kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3599 struct kvm_io_device *dev)
3602 struct kvm_io_bus *new_bus, *bus;
3604 bus = kvm_get_bus(kvm, bus_idx);
3608 for (i = 0; i < bus->dev_count; i++)
3609 if (bus->range[i].dev == dev) {
3613 if (i == bus->dev_count)
3616 new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3617 sizeof(struct kvm_io_range)), GFP_KERNEL);
3619 pr_err("kvm: failed to shrink bus, removing it completely\n");
3623 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3624 new_bus->dev_count--;
3625 memcpy(new_bus->range + i, bus->range + i + 1,
3626 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3629 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3630 synchronize_srcu_expedited(&kvm->srcu);
3635 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3638 struct kvm_io_bus *bus;
3639 int dev_idx, srcu_idx;
3640 struct kvm_io_device *iodev = NULL;
3642 srcu_idx = srcu_read_lock(&kvm->srcu);
3644 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
3648 dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
3652 iodev = bus->range[dev_idx].dev;
3655 srcu_read_unlock(&kvm->srcu, srcu_idx);
3659 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
3661 static int kvm_debugfs_open(struct inode *inode, struct file *file,
3662 int (*get)(void *, u64 *), int (*set)(void *, u64),
3665 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
3668 /* The debugfs files are a reference to the kvm struct which
3669 * is still valid when kvm_destroy_vm is called.
3670 * To avoid the race between open and the removal of the debugfs
3671 * directory we test against the users count.
3673 if (!refcount_inc_not_zero(&stat_data->kvm->users_count))
3676 if (simple_attr_open(inode, file, get, set, fmt)) {
3677 kvm_put_kvm(stat_data->kvm);
3684 static int kvm_debugfs_release(struct inode *inode, struct file *file)
3686 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
3689 simple_attr_release(inode, file);
3690 kvm_put_kvm(stat_data->kvm);
3695 static int vm_stat_get_per_vm(void *data, u64 *val)
3697 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3699 *val = *(ulong *)((void *)stat_data->kvm + stat_data->offset);
3704 static int vm_stat_clear_per_vm(void *data, u64 val)
3706 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3711 *(ulong *)((void *)stat_data->kvm + stat_data->offset) = 0;
3716 static int vm_stat_get_per_vm_open(struct inode *inode, struct file *file)
3718 __simple_attr_check_format("%llu\n", 0ull);
3719 return kvm_debugfs_open(inode, file, vm_stat_get_per_vm,
3720 vm_stat_clear_per_vm, "%llu\n");
3723 static const struct file_operations vm_stat_get_per_vm_fops = {
3724 .owner = THIS_MODULE,
3725 .open = vm_stat_get_per_vm_open,
3726 .release = kvm_debugfs_release,
3727 .read = simple_attr_read,
3728 .write = simple_attr_write,
3729 .llseek = no_llseek,
3732 static int vcpu_stat_get_per_vm(void *data, u64 *val)
3735 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3736 struct kvm_vcpu *vcpu;
3740 kvm_for_each_vcpu(i, vcpu, stat_data->kvm)
3741 *val += *(u64 *)((void *)vcpu + stat_data->offset);
3746 static int vcpu_stat_clear_per_vm(void *data, u64 val)
3749 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3750 struct kvm_vcpu *vcpu;
3755 kvm_for_each_vcpu(i, vcpu, stat_data->kvm)
3756 *(u64 *)((void *)vcpu + stat_data->offset) = 0;
3761 static int vcpu_stat_get_per_vm_open(struct inode *inode, struct file *file)
3763 __simple_attr_check_format("%llu\n", 0ull);
3764 return kvm_debugfs_open(inode, file, vcpu_stat_get_per_vm,
3765 vcpu_stat_clear_per_vm, "%llu\n");
3768 static const struct file_operations vcpu_stat_get_per_vm_fops = {
3769 .owner = THIS_MODULE,
3770 .open = vcpu_stat_get_per_vm_open,
3771 .release = kvm_debugfs_release,
3772 .read = simple_attr_read,
3773 .write = simple_attr_write,
3774 .llseek = no_llseek,
3777 static const struct file_operations *stat_fops_per_vm[] = {
3778 [KVM_STAT_VCPU] = &vcpu_stat_get_per_vm_fops,
3779 [KVM_STAT_VM] = &vm_stat_get_per_vm_fops,
3782 static int vm_stat_get(void *_offset, u64 *val)
3784 unsigned offset = (long)_offset;
3786 struct kvm_stat_data stat_tmp = {.offset = offset};
3790 spin_lock(&kvm_lock);
3791 list_for_each_entry(kvm, &vm_list, vm_list) {
3793 vm_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
3796 spin_unlock(&kvm_lock);
3800 static int vm_stat_clear(void *_offset, u64 val)
3802 unsigned offset = (long)_offset;
3804 struct kvm_stat_data stat_tmp = {.offset = offset};
3809 spin_lock(&kvm_lock);
3810 list_for_each_entry(kvm, &vm_list, vm_list) {
3812 vm_stat_clear_per_vm((void *)&stat_tmp, 0);
3814 spin_unlock(&kvm_lock);
3819 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n");
3821 static int vcpu_stat_get(void *_offset, u64 *val)
3823 unsigned offset = (long)_offset;
3825 struct kvm_stat_data stat_tmp = {.offset = offset};
3829 spin_lock(&kvm_lock);
3830 list_for_each_entry(kvm, &vm_list, vm_list) {
3832 vcpu_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
3835 spin_unlock(&kvm_lock);
3839 static int vcpu_stat_clear(void *_offset, u64 val)
3841 unsigned offset = (long)_offset;
3843 struct kvm_stat_data stat_tmp = {.offset = offset};
3848 spin_lock(&kvm_lock);
3849 list_for_each_entry(kvm, &vm_list, vm_list) {
3851 vcpu_stat_clear_per_vm((void *)&stat_tmp, 0);
3853 spin_unlock(&kvm_lock);
3858 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear,
3861 static const struct file_operations *stat_fops[] = {
3862 [KVM_STAT_VCPU] = &vcpu_stat_fops,
3863 [KVM_STAT_VM] = &vm_stat_fops,
3866 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm)
3868 struct kobj_uevent_env *env;
3869 unsigned long long created, active;
3871 if (!kvm_dev.this_device || !kvm)
3874 spin_lock(&kvm_lock);
3875 if (type == KVM_EVENT_CREATE_VM) {
3876 kvm_createvm_count++;
3878 } else if (type == KVM_EVENT_DESTROY_VM) {
3881 created = kvm_createvm_count;
3882 active = kvm_active_vms;
3883 spin_unlock(&kvm_lock);
3885 env = kzalloc(sizeof(*env), GFP_KERNEL);
3889 add_uevent_var(env, "CREATED=%llu", created);
3890 add_uevent_var(env, "COUNT=%llu", active);
3892 if (type == KVM_EVENT_CREATE_VM) {
3893 add_uevent_var(env, "EVENT=create");
3894 kvm->userspace_pid = task_pid_nr(current);
3895 } else if (type == KVM_EVENT_DESTROY_VM) {
3896 add_uevent_var(env, "EVENT=destroy");
3898 add_uevent_var(env, "PID=%d", kvm->userspace_pid);
3900 if (kvm->debugfs_dentry) {
3901 char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL);
3904 tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX);
3906 add_uevent_var(env, "STATS_PATH=%s", tmp);
3910 /* no need for checks, since we are adding at most only 5 keys */
3911 env->envp[env->envp_idx++] = NULL;
3912 kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp);
3916 static void kvm_init_debug(void)
3918 struct kvm_stats_debugfs_item *p;
3920 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3922 kvm_debugfs_num_entries = 0;
3923 for (p = debugfs_entries; p->name; ++p, kvm_debugfs_num_entries++) {
3924 debugfs_create_file(p->name, 0644, kvm_debugfs_dir,
3925 (void *)(long)p->offset,
3926 stat_fops[p->kind]);
3930 static int kvm_suspend(void)
3932 if (kvm_usage_count)
3933 hardware_disable_nolock(NULL);
3937 static void kvm_resume(void)
3939 if (kvm_usage_count) {
3940 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3941 hardware_enable_nolock(NULL);
3945 static struct syscore_ops kvm_syscore_ops = {
3946 .suspend = kvm_suspend,
3947 .resume = kvm_resume,
3951 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3953 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3956 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3958 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3960 if (vcpu->preempted)
3961 vcpu->preempted = false;
3963 kvm_arch_sched_in(vcpu, cpu);
3965 kvm_arch_vcpu_load(vcpu, cpu);
3968 static void kvm_sched_out(struct preempt_notifier *pn,
3969 struct task_struct *next)
3971 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3973 if (current->state == TASK_RUNNING)
3974 vcpu->preempted = true;
3975 kvm_arch_vcpu_put(vcpu);
3978 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3979 struct module *module)
3984 r = kvm_arch_init(opaque);
3989 * kvm_arch_init makes sure there's at most one caller
3990 * for architectures that support multiple implementations,
3991 * like intel and amd on x86.
3992 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3993 * conflicts in case kvm is already setup for another implementation.
3995 r = kvm_irqfd_init();
3999 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
4004 r = kvm_arch_hardware_setup();
4008 for_each_online_cpu(cpu) {
4009 smp_call_function_single(cpu,
4010 kvm_arch_check_processor_compat,
4016 r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting",
4017 kvm_starting_cpu, kvm_dying_cpu);
4020 register_reboot_notifier(&kvm_reboot_notifier);
4022 /* A kmem cache lets us meet the alignment requirements of fx_save. */
4024 vcpu_align = __alignof__(struct kvm_vcpu);
4026 kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align,
4028 offsetof(struct kvm_vcpu, arch),
4029 sizeof_field(struct kvm_vcpu, arch),
4031 if (!kvm_vcpu_cache) {
4036 r = kvm_async_pf_init();
4040 kvm_chardev_ops.owner = module;
4041 kvm_vm_fops.owner = module;
4042 kvm_vcpu_fops.owner = module;
4044 r = misc_register(&kvm_dev);
4046 pr_err("kvm: misc device register failed\n");
4050 register_syscore_ops(&kvm_syscore_ops);
4052 kvm_preempt_ops.sched_in = kvm_sched_in;
4053 kvm_preempt_ops.sched_out = kvm_sched_out;
4057 r = kvm_vfio_ops_init();
4063 kvm_async_pf_deinit();
4065 kmem_cache_destroy(kvm_vcpu_cache);
4067 unregister_reboot_notifier(&kvm_reboot_notifier);
4068 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4071 kvm_arch_hardware_unsetup();
4073 free_cpumask_var(cpus_hardware_enabled);
4081 EXPORT_SYMBOL_GPL(kvm_init);
4085 debugfs_remove_recursive(kvm_debugfs_dir);
4086 misc_deregister(&kvm_dev);
4087 kmem_cache_destroy(kvm_vcpu_cache);
4088 kvm_async_pf_deinit();
4089 unregister_syscore_ops(&kvm_syscore_ops);
4090 unregister_reboot_notifier(&kvm_reboot_notifier);
4091 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4092 on_each_cpu(hardware_disable_nolock, NULL, 1);
4093 kvm_arch_hardware_unsetup();
4096 free_cpumask_var(cpus_hardware_enabled);
4097 kvm_vfio_ops_exit();
4099 EXPORT_SYMBOL_GPL(kvm_exit);