2 * Kernel-based Virtual Machine driver for Linux
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
7 * Copyright (C) 2006 Qumranet, Inc.
8 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
11 * Avi Kivity <avi@qumranet.com>
12 * Yaniv Kamay <yaniv@qumranet.com>
14 * This work is licensed under the terms of the GNU GPL, version 2. See
15 * the COPYING file in the top-level directory.
19 #include <kvm/iodev.h>
21 #include <linux/kvm_host.h>
22 #include <linux/kvm.h>
23 #include <linux/module.h>
24 #include <linux/errno.h>
25 #include <linux/percpu.h>
27 #include <linux/miscdevice.h>
28 #include <linux/vmalloc.h>
29 #include <linux/reboot.h>
30 #include <linux/debugfs.h>
31 #include <linux/highmem.h>
32 #include <linux/file.h>
33 #include <linux/syscore_ops.h>
34 #include <linux/cpu.h>
35 #include <linux/sched/signal.h>
36 #include <linux/sched/mm.h>
37 #include <linux/sched/stat.h>
38 #include <linux/cpumask.h>
39 #include <linux/smp.h>
40 #include <linux/anon_inodes.h>
41 #include <linux/profile.h>
42 #include <linux/kvm_para.h>
43 #include <linux/pagemap.h>
44 #include <linux/mman.h>
45 #include <linux/swap.h>
46 #include <linux/bitops.h>
47 #include <linux/spinlock.h>
48 #include <linux/compat.h>
49 #include <linux/srcu.h>
50 #include <linux/hugetlb.h>
51 #include <linux/slab.h>
52 #include <linux/sort.h>
53 #include <linux/bsearch.h>
55 #include <asm/processor.h>
57 #include <asm/ioctl.h>
58 #include <linux/uaccess.h>
59 #include <asm/pgtable.h>
61 #include "coalesced_mmio.h"
65 #define CREATE_TRACE_POINTS
66 #include <trace/events/kvm.h>
68 /* Worst case buffer size needed for holding an integer. */
69 #define ITOA_MAX_LEN 12
71 MODULE_AUTHOR("Qumranet");
72 MODULE_LICENSE("GPL");
74 /* Architectures should define their poll value according to the halt latency */
75 unsigned int halt_poll_ns = KVM_HALT_POLL_NS_DEFAULT;
76 module_param(halt_poll_ns, uint, 0644);
77 EXPORT_SYMBOL_GPL(halt_poll_ns);
79 /* Default doubles per-vcpu halt_poll_ns. */
80 unsigned int halt_poll_ns_grow = 2;
81 module_param(halt_poll_ns_grow, uint, 0644);
82 EXPORT_SYMBOL_GPL(halt_poll_ns_grow);
84 /* Default resets per-vcpu halt_poll_ns . */
85 unsigned int halt_poll_ns_shrink;
86 module_param(halt_poll_ns_shrink, uint, 0644);
87 EXPORT_SYMBOL_GPL(halt_poll_ns_shrink);
92 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
95 DEFINE_SPINLOCK(kvm_lock);
96 static DEFINE_RAW_SPINLOCK(kvm_count_lock);
99 static cpumask_var_t cpus_hardware_enabled;
100 static int kvm_usage_count;
101 static atomic_t hardware_enable_failed;
103 struct kmem_cache *kvm_vcpu_cache;
104 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
106 static __read_mostly struct preempt_ops kvm_preempt_ops;
108 struct dentry *kvm_debugfs_dir;
109 EXPORT_SYMBOL_GPL(kvm_debugfs_dir);
111 static int kvm_debugfs_num_entries;
112 static const struct file_operations *stat_fops_per_vm[];
114 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
116 #ifdef CONFIG_KVM_COMPAT
117 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
120 static int hardware_enable_all(void);
121 static void hardware_disable_all(void);
123 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
125 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot, gfn_t gfn);
127 __visible bool kvm_rebooting;
128 EXPORT_SYMBOL_GPL(kvm_rebooting);
130 static bool largepages_enabled = true;
132 #define KVM_EVENT_CREATE_VM 0
133 #define KVM_EVENT_DESTROY_VM 1
134 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm);
135 static unsigned long long kvm_createvm_count;
136 static unsigned long long kvm_active_vms;
138 __weak void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
139 unsigned long start, unsigned long end)
143 bool kvm_is_reserved_pfn(kvm_pfn_t pfn)
146 return PageReserved(pfn_to_page(pfn));
152 * Switches to specified vcpu, until a matching vcpu_put()
154 void vcpu_load(struct kvm_vcpu *vcpu)
157 preempt_notifier_register(&vcpu->preempt_notifier);
158 kvm_arch_vcpu_load(vcpu, cpu);
161 EXPORT_SYMBOL_GPL(vcpu_load);
163 void vcpu_put(struct kvm_vcpu *vcpu)
166 kvm_arch_vcpu_put(vcpu);
167 preempt_notifier_unregister(&vcpu->preempt_notifier);
170 EXPORT_SYMBOL_GPL(vcpu_put);
172 /* TODO: merge with kvm_arch_vcpu_should_kick */
173 static bool kvm_request_needs_ipi(struct kvm_vcpu *vcpu, unsigned req)
175 int mode = kvm_vcpu_exiting_guest_mode(vcpu);
178 * We need to wait for the VCPU to reenable interrupts and get out of
179 * READING_SHADOW_PAGE_TABLES mode.
181 if (req & KVM_REQUEST_WAIT)
182 return mode != OUTSIDE_GUEST_MODE;
185 * Need to kick a running VCPU, but otherwise there is nothing to do.
187 return mode == IN_GUEST_MODE;
190 static void ack_flush(void *_completed)
194 static inline bool kvm_kick_many_cpus(const struct cpumask *cpus, bool wait)
197 cpus = cpu_online_mask;
199 if (cpumask_empty(cpus))
202 smp_call_function_many(cpus, ack_flush, NULL, wait);
206 bool kvm_make_vcpus_request_mask(struct kvm *kvm, unsigned int req,
207 unsigned long *vcpu_bitmap, cpumask_var_t tmp)
210 struct kvm_vcpu *vcpu;
215 kvm_for_each_vcpu(i, vcpu, kvm) {
216 if (!test_bit(i, vcpu_bitmap))
219 kvm_make_request(req, vcpu);
222 if (!(req & KVM_REQUEST_NO_WAKEUP) && kvm_vcpu_wake_up(vcpu))
225 if (tmp != NULL && cpu != -1 && cpu != me &&
226 kvm_request_needs_ipi(vcpu, req))
227 __cpumask_set_cpu(cpu, tmp);
230 called = kvm_kick_many_cpus(tmp, !!(req & KVM_REQUEST_WAIT));
236 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
240 static unsigned long vcpu_bitmap[BITS_TO_LONGS(KVM_MAX_VCPUS)]
241 = {[0 ... BITS_TO_LONGS(KVM_MAX_VCPUS)-1] = ULONG_MAX};
243 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
245 called = kvm_make_vcpus_request_mask(kvm, req, vcpu_bitmap, cpus);
247 free_cpumask_var(cpus);
251 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
252 void kvm_flush_remote_tlbs(struct kvm *kvm)
255 * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in
256 * kvm_make_all_cpus_request.
258 long dirty_count = smp_load_acquire(&kvm->tlbs_dirty);
261 * We want to publish modifications to the page tables before reading
262 * mode. Pairs with a memory barrier in arch-specific code.
263 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
264 * and smp_mb in walk_shadow_page_lockless_begin/end.
265 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
267 * There is already an smp_mb__after_atomic() before
268 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
271 if (kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
272 ++kvm->stat.remote_tlb_flush;
273 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
275 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
278 void kvm_reload_remote_mmus(struct kvm *kvm)
280 kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
283 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
288 mutex_init(&vcpu->mutex);
293 init_swait_queue_head(&vcpu->wq);
294 kvm_async_pf_vcpu_init(vcpu);
297 INIT_LIST_HEAD(&vcpu->blocked_vcpu_list);
299 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
304 vcpu->run = page_address(page);
306 kvm_vcpu_set_in_spin_loop(vcpu, false);
307 kvm_vcpu_set_dy_eligible(vcpu, false);
308 vcpu->preempted = false;
310 r = kvm_arch_vcpu_init(vcpu);
316 free_page((unsigned long)vcpu->run);
320 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
322 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
325 * no need for rcu_read_lock as VCPU_RUN is the only place that
326 * will change the vcpu->pid pointer and on uninit all file
327 * descriptors are already gone.
329 put_pid(rcu_dereference_protected(vcpu->pid, 1));
330 kvm_arch_vcpu_uninit(vcpu);
331 free_page((unsigned long)vcpu->run);
333 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
335 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
336 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
338 return container_of(mn, struct kvm, mmu_notifier);
341 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
342 struct mm_struct *mm,
343 unsigned long address,
346 struct kvm *kvm = mmu_notifier_to_kvm(mn);
349 idx = srcu_read_lock(&kvm->srcu);
350 spin_lock(&kvm->mmu_lock);
351 kvm->mmu_notifier_seq++;
352 kvm_set_spte_hva(kvm, address, pte);
353 spin_unlock(&kvm->mmu_lock);
354 srcu_read_unlock(&kvm->srcu, idx);
357 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
358 struct mm_struct *mm,
362 struct kvm *kvm = mmu_notifier_to_kvm(mn);
363 int need_tlb_flush = 0, idx;
365 idx = srcu_read_lock(&kvm->srcu);
366 spin_lock(&kvm->mmu_lock);
368 * The count increase must become visible at unlock time as no
369 * spte can be established without taking the mmu_lock and
370 * count is also read inside the mmu_lock critical section.
372 kvm->mmu_notifier_count++;
373 need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
374 need_tlb_flush |= kvm->tlbs_dirty;
375 /* we've to flush the tlb before the pages can be freed */
377 kvm_flush_remote_tlbs(kvm);
379 spin_unlock(&kvm->mmu_lock);
381 kvm_arch_mmu_notifier_invalidate_range(kvm, start, end);
383 srcu_read_unlock(&kvm->srcu, idx);
386 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
387 struct mm_struct *mm,
391 struct kvm *kvm = mmu_notifier_to_kvm(mn);
393 spin_lock(&kvm->mmu_lock);
395 * This sequence increase will notify the kvm page fault that
396 * the page that is going to be mapped in the spte could have
399 kvm->mmu_notifier_seq++;
402 * The above sequence increase must be visible before the
403 * below count decrease, which is ensured by the smp_wmb above
404 * in conjunction with the smp_rmb in mmu_notifier_retry().
406 kvm->mmu_notifier_count--;
407 spin_unlock(&kvm->mmu_lock);
409 BUG_ON(kvm->mmu_notifier_count < 0);
412 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
413 struct mm_struct *mm,
417 struct kvm *kvm = mmu_notifier_to_kvm(mn);
420 idx = srcu_read_lock(&kvm->srcu);
421 spin_lock(&kvm->mmu_lock);
423 young = kvm_age_hva(kvm, start, end);
425 kvm_flush_remote_tlbs(kvm);
427 spin_unlock(&kvm->mmu_lock);
428 srcu_read_unlock(&kvm->srcu, idx);
433 static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
434 struct mm_struct *mm,
438 struct kvm *kvm = mmu_notifier_to_kvm(mn);
441 idx = srcu_read_lock(&kvm->srcu);
442 spin_lock(&kvm->mmu_lock);
444 * Even though we do not flush TLB, this will still adversely
445 * affect performance on pre-Haswell Intel EPT, where there is
446 * no EPT Access Bit to clear so that we have to tear down EPT
447 * tables instead. If we find this unacceptable, we can always
448 * add a parameter to kvm_age_hva so that it effectively doesn't
449 * do anything on clear_young.
451 * Also note that currently we never issue secondary TLB flushes
452 * from clear_young, leaving this job up to the regular system
453 * cadence. If we find this inaccurate, we might come up with a
454 * more sophisticated heuristic later.
456 young = kvm_age_hva(kvm, start, end);
457 spin_unlock(&kvm->mmu_lock);
458 srcu_read_unlock(&kvm->srcu, idx);
463 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
464 struct mm_struct *mm,
465 unsigned long address)
467 struct kvm *kvm = mmu_notifier_to_kvm(mn);
470 idx = srcu_read_lock(&kvm->srcu);
471 spin_lock(&kvm->mmu_lock);
472 young = kvm_test_age_hva(kvm, address);
473 spin_unlock(&kvm->mmu_lock);
474 srcu_read_unlock(&kvm->srcu, idx);
479 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
480 struct mm_struct *mm)
482 struct kvm *kvm = mmu_notifier_to_kvm(mn);
485 idx = srcu_read_lock(&kvm->srcu);
486 kvm_arch_flush_shadow_all(kvm);
487 srcu_read_unlock(&kvm->srcu, idx);
490 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
491 .flags = MMU_INVALIDATE_DOES_NOT_BLOCK,
492 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
493 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
494 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
495 .clear_young = kvm_mmu_notifier_clear_young,
496 .test_young = kvm_mmu_notifier_test_young,
497 .change_pte = kvm_mmu_notifier_change_pte,
498 .release = kvm_mmu_notifier_release,
501 static int kvm_init_mmu_notifier(struct kvm *kvm)
503 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
504 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
507 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
509 static int kvm_init_mmu_notifier(struct kvm *kvm)
514 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
516 static struct kvm_memslots *kvm_alloc_memslots(void)
519 struct kvm_memslots *slots;
521 slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
525 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
526 slots->id_to_index[i] = slots->memslots[i].id = i;
531 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
533 if (!memslot->dirty_bitmap)
536 kvfree(memslot->dirty_bitmap);
537 memslot->dirty_bitmap = NULL;
541 * Free any memory in @free but not in @dont.
543 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
544 struct kvm_memory_slot *dont)
546 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
547 kvm_destroy_dirty_bitmap(free);
549 kvm_arch_free_memslot(kvm, free, dont);
554 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
556 struct kvm_memory_slot *memslot;
561 kvm_for_each_memslot(memslot, slots)
562 kvm_free_memslot(kvm, memslot, NULL);
567 static void kvm_destroy_vm_debugfs(struct kvm *kvm)
571 if (!kvm->debugfs_dentry)
574 debugfs_remove_recursive(kvm->debugfs_dentry);
576 if (kvm->debugfs_stat_data) {
577 for (i = 0; i < kvm_debugfs_num_entries; i++)
578 kfree(kvm->debugfs_stat_data[i]);
579 kfree(kvm->debugfs_stat_data);
583 static int kvm_create_vm_debugfs(struct kvm *kvm, int fd)
585 char dir_name[ITOA_MAX_LEN * 2];
586 struct kvm_stat_data *stat_data;
587 struct kvm_stats_debugfs_item *p;
589 if (!debugfs_initialized())
592 snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd);
593 kvm->debugfs_dentry = debugfs_create_dir(dir_name, kvm_debugfs_dir);
595 kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries,
596 sizeof(*kvm->debugfs_stat_data),
598 if (!kvm->debugfs_stat_data)
601 for (p = debugfs_entries; p->name; p++) {
602 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL);
606 stat_data->kvm = kvm;
607 stat_data->offset = p->offset;
608 kvm->debugfs_stat_data[p - debugfs_entries] = stat_data;
609 debugfs_create_file(p->name, 0644, kvm->debugfs_dentry,
610 stat_data, stat_fops_per_vm[p->kind]);
615 static struct kvm *kvm_create_vm(unsigned long type)
618 struct kvm *kvm = kvm_arch_alloc_vm();
621 return ERR_PTR(-ENOMEM);
623 spin_lock_init(&kvm->mmu_lock);
625 kvm->mm = current->mm;
626 kvm_eventfd_init(kvm);
627 mutex_init(&kvm->lock);
628 mutex_init(&kvm->irq_lock);
629 mutex_init(&kvm->slots_lock);
630 refcount_set(&kvm->users_count, 1);
631 INIT_LIST_HEAD(&kvm->devices);
633 r = kvm_arch_init_vm(kvm, type);
635 goto out_err_no_disable;
637 r = hardware_enable_all();
639 goto out_err_no_disable;
641 #ifdef CONFIG_HAVE_KVM_IRQFD
642 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
645 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
648 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
649 struct kvm_memslots *slots = kvm_alloc_memslots();
651 goto out_err_no_srcu;
653 * Generations must be different for each address space.
654 * Init kvm generation close to the maximum to easily test the
655 * code of handling generation number wrap-around.
657 slots->generation = i * 2 - 150;
658 rcu_assign_pointer(kvm->memslots[i], slots);
661 if (init_srcu_struct(&kvm->srcu))
662 goto out_err_no_srcu;
663 if (init_srcu_struct(&kvm->irq_srcu))
664 goto out_err_no_irq_srcu;
665 for (i = 0; i < KVM_NR_BUSES; i++) {
666 rcu_assign_pointer(kvm->buses[i],
667 kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL));
672 r = kvm_init_mmu_notifier(kvm);
676 spin_lock(&kvm_lock);
677 list_add(&kvm->vm_list, &vm_list);
678 spin_unlock(&kvm_lock);
680 preempt_notifier_inc();
685 cleanup_srcu_struct(&kvm->irq_srcu);
687 cleanup_srcu_struct(&kvm->srcu);
689 hardware_disable_all();
691 refcount_set(&kvm->users_count, 0);
692 for (i = 0; i < KVM_NR_BUSES; i++)
693 kfree(kvm_get_bus(kvm, i));
694 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
695 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
696 kvm_arch_free_vm(kvm);
701 static void kvm_destroy_devices(struct kvm *kvm)
703 struct kvm_device *dev, *tmp;
706 * We do not need to take the kvm->lock here, because nobody else
707 * has a reference to the struct kvm at this point and therefore
708 * cannot access the devices list anyhow.
710 list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
711 list_del(&dev->vm_node);
712 dev->ops->destroy(dev);
716 static void kvm_destroy_vm(struct kvm *kvm)
719 struct mm_struct *mm = kvm->mm;
721 kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm);
722 kvm_destroy_vm_debugfs(kvm);
723 kvm_arch_sync_events(kvm);
724 spin_lock(&kvm_lock);
725 list_del(&kvm->vm_list);
726 spin_unlock(&kvm_lock);
727 kvm_free_irq_routing(kvm);
728 for (i = 0; i < KVM_NR_BUSES; i++) {
729 struct kvm_io_bus *bus = kvm_get_bus(kvm, i);
732 kvm_io_bus_destroy(bus);
733 kvm->buses[i] = NULL;
735 kvm_coalesced_mmio_free(kvm);
736 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
737 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
739 kvm_arch_flush_shadow_all(kvm);
741 kvm_arch_destroy_vm(kvm);
742 kvm_destroy_devices(kvm);
743 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
744 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
745 cleanup_srcu_struct(&kvm->irq_srcu);
746 cleanup_srcu_struct(&kvm->srcu);
747 kvm_arch_free_vm(kvm);
748 preempt_notifier_dec();
749 hardware_disable_all();
753 void kvm_get_kvm(struct kvm *kvm)
755 refcount_inc(&kvm->users_count);
757 EXPORT_SYMBOL_GPL(kvm_get_kvm);
759 void kvm_put_kvm(struct kvm *kvm)
761 if (refcount_dec_and_test(&kvm->users_count))
764 EXPORT_SYMBOL_GPL(kvm_put_kvm);
767 static int kvm_vm_release(struct inode *inode, struct file *filp)
769 struct kvm *kvm = filp->private_data;
771 kvm_irqfd_release(kvm);
778 * Allocation size is twice as large as the actual dirty bitmap size.
779 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
781 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
783 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
785 memslot->dirty_bitmap = kvzalloc(dirty_bytes, GFP_KERNEL);
786 if (!memslot->dirty_bitmap)
793 * Insert memslot and re-sort memslots based on their GFN,
794 * so binary search could be used to lookup GFN.
795 * Sorting algorithm takes advantage of having initially
796 * sorted array and known changed memslot position.
798 static void update_memslots(struct kvm_memslots *slots,
799 struct kvm_memory_slot *new)
802 int i = slots->id_to_index[id];
803 struct kvm_memory_slot *mslots = slots->memslots;
805 WARN_ON(mslots[i].id != id);
807 WARN_ON(!mslots[i].npages);
808 if (mslots[i].npages)
811 if (!mslots[i].npages)
815 while (i < KVM_MEM_SLOTS_NUM - 1 &&
816 new->base_gfn <= mslots[i + 1].base_gfn) {
817 if (!mslots[i + 1].npages)
819 mslots[i] = mslots[i + 1];
820 slots->id_to_index[mslots[i].id] = i;
825 * The ">=" is needed when creating a slot with base_gfn == 0,
826 * so that it moves before all those with base_gfn == npages == 0.
828 * On the other hand, if new->npages is zero, the above loop has
829 * already left i pointing to the beginning of the empty part of
830 * mslots, and the ">=" would move the hole backwards in this
831 * case---which is wrong. So skip the loop when deleting a slot.
835 new->base_gfn >= mslots[i - 1].base_gfn) {
836 mslots[i] = mslots[i - 1];
837 slots->id_to_index[mslots[i].id] = i;
841 WARN_ON_ONCE(i != slots->used_slots);
844 slots->id_to_index[mslots[i].id] = i;
847 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
849 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
851 #ifdef __KVM_HAVE_READONLY_MEM
852 valid_flags |= KVM_MEM_READONLY;
855 if (mem->flags & ~valid_flags)
861 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
862 int as_id, struct kvm_memslots *slots)
864 struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id);
867 * Set the low bit in the generation, which disables SPTE caching
868 * until the end of synchronize_srcu_expedited.
870 WARN_ON(old_memslots->generation & 1);
871 slots->generation = old_memslots->generation + 1;
873 rcu_assign_pointer(kvm->memslots[as_id], slots);
874 synchronize_srcu_expedited(&kvm->srcu);
877 * Increment the new memslot generation a second time. This prevents
878 * vm exits that race with memslot updates from caching a memslot
879 * generation that will (potentially) be valid forever.
881 * Generations must be unique even across address spaces. We do not need
882 * a global counter for that, instead the generation space is evenly split
883 * across address spaces. For example, with two address spaces, address
884 * space 0 will use generations 0, 4, 8, ... while * address space 1 will
885 * use generations 2, 6, 10, 14, ...
887 slots->generation += KVM_ADDRESS_SPACE_NUM * 2 - 1;
889 kvm_arch_memslots_updated(kvm, slots);
895 * Allocate some memory and give it an address in the guest physical address
898 * Discontiguous memory is allowed, mostly for framebuffers.
900 * Must be called holding kvm->slots_lock for write.
902 int __kvm_set_memory_region(struct kvm *kvm,
903 const struct kvm_userspace_memory_region *mem)
907 unsigned long npages;
908 struct kvm_memory_slot *slot;
909 struct kvm_memory_slot old, new;
910 struct kvm_memslots *slots = NULL, *old_memslots;
912 enum kvm_mr_change change;
914 r = check_memory_region_flags(mem);
919 as_id = mem->slot >> 16;
922 /* General sanity checks */
923 if (mem->memory_size & (PAGE_SIZE - 1))
925 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
927 /* We can read the guest memory with __xxx_user() later on. */
928 if ((id < KVM_USER_MEM_SLOTS) &&
929 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
930 !access_ok(VERIFY_WRITE,
931 (void __user *)(unsigned long)mem->userspace_addr,
934 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
936 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
939 slot = id_to_memslot(__kvm_memslots(kvm, as_id), id);
940 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
941 npages = mem->memory_size >> PAGE_SHIFT;
943 if (npages > KVM_MEM_MAX_NR_PAGES)
949 new.base_gfn = base_gfn;
951 new.flags = mem->flags;
955 change = KVM_MR_CREATE;
956 else { /* Modify an existing slot. */
957 if ((mem->userspace_addr != old.userspace_addr) ||
958 (npages != old.npages) ||
959 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
962 if (base_gfn != old.base_gfn)
963 change = KVM_MR_MOVE;
964 else if (new.flags != old.flags)
965 change = KVM_MR_FLAGS_ONLY;
966 else { /* Nothing to change. */
975 change = KVM_MR_DELETE;
980 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
981 /* Check for overlaps */
983 kvm_for_each_memslot(slot, __kvm_memslots(kvm, as_id)) {
986 if (!((base_gfn + npages <= slot->base_gfn) ||
987 (base_gfn >= slot->base_gfn + slot->npages)))
992 /* Free page dirty bitmap if unneeded */
993 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
994 new.dirty_bitmap = NULL;
997 if (change == KVM_MR_CREATE) {
998 new.userspace_addr = mem->userspace_addr;
1000 if (kvm_arch_create_memslot(kvm, &new, npages))
1004 /* Allocate page dirty bitmap if needed */
1005 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
1006 if (kvm_create_dirty_bitmap(&new) < 0)
1010 slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
1013 memcpy(slots, __kvm_memslots(kvm, as_id), sizeof(struct kvm_memslots));
1015 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
1016 slot = id_to_memslot(slots, id);
1017 slot->flags |= KVM_MEMSLOT_INVALID;
1019 old_memslots = install_new_memslots(kvm, as_id, slots);
1021 /* From this point no new shadow pages pointing to a deleted,
1022 * or moved, memslot will be created.
1024 * validation of sp->gfn happens in:
1025 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1026 * - kvm_is_visible_gfn (mmu_check_roots)
1028 kvm_arch_flush_shadow_memslot(kvm, slot);
1031 * We can re-use the old_memslots from above, the only difference
1032 * from the currently installed memslots is the invalid flag. This
1033 * will get overwritten by update_memslots anyway.
1035 slots = old_memslots;
1038 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
1042 /* actual memory is freed via old in kvm_free_memslot below */
1043 if (change == KVM_MR_DELETE) {
1044 new.dirty_bitmap = NULL;
1045 memset(&new.arch, 0, sizeof(new.arch));
1048 update_memslots(slots, &new);
1049 old_memslots = install_new_memslots(kvm, as_id, slots);
1051 kvm_arch_commit_memory_region(kvm, mem, &old, &new, change);
1053 kvm_free_memslot(kvm, &old, &new);
1054 kvfree(old_memslots);
1060 kvm_free_memslot(kvm, &new, &old);
1064 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1066 int kvm_set_memory_region(struct kvm *kvm,
1067 const struct kvm_userspace_memory_region *mem)
1071 mutex_lock(&kvm->slots_lock);
1072 r = __kvm_set_memory_region(kvm, mem);
1073 mutex_unlock(&kvm->slots_lock);
1076 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1078 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1079 struct kvm_userspace_memory_region *mem)
1081 if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1084 return kvm_set_memory_region(kvm, mem);
1087 int kvm_get_dirty_log(struct kvm *kvm,
1088 struct kvm_dirty_log *log, int *is_dirty)
1090 struct kvm_memslots *slots;
1091 struct kvm_memory_slot *memslot;
1094 unsigned long any = 0;
1096 as_id = log->slot >> 16;
1097 id = (u16)log->slot;
1098 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1101 slots = __kvm_memslots(kvm, as_id);
1102 memslot = id_to_memslot(slots, id);
1103 if (!memslot->dirty_bitmap)
1106 n = kvm_dirty_bitmap_bytes(memslot);
1108 for (i = 0; !any && i < n/sizeof(long); ++i)
1109 any = memslot->dirty_bitmap[i];
1111 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1118 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1120 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1122 * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages
1123 * are dirty write protect them for next write.
1124 * @kvm: pointer to kvm instance
1125 * @log: slot id and address to which we copy the log
1126 * @is_dirty: flag set if any page is dirty
1128 * We need to keep it in mind that VCPU threads can write to the bitmap
1129 * concurrently. So, to avoid losing track of dirty pages we keep the
1132 * 1. Take a snapshot of the bit and clear it if needed.
1133 * 2. Write protect the corresponding page.
1134 * 3. Copy the snapshot to the userspace.
1135 * 4. Upon return caller flushes TLB's if needed.
1137 * Between 2 and 4, the guest may write to the page using the remaining TLB
1138 * entry. This is not a problem because the page is reported dirty using
1139 * the snapshot taken before and step 4 ensures that writes done after
1140 * exiting to userspace will be logged for the next call.
1143 int kvm_get_dirty_log_protect(struct kvm *kvm,
1144 struct kvm_dirty_log *log, bool *is_dirty)
1146 struct kvm_memslots *slots;
1147 struct kvm_memory_slot *memslot;
1150 unsigned long *dirty_bitmap;
1151 unsigned long *dirty_bitmap_buffer;
1153 as_id = log->slot >> 16;
1154 id = (u16)log->slot;
1155 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1158 slots = __kvm_memslots(kvm, as_id);
1159 memslot = id_to_memslot(slots, id);
1161 dirty_bitmap = memslot->dirty_bitmap;
1165 n = kvm_dirty_bitmap_bytes(memslot);
1167 dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long);
1168 memset(dirty_bitmap_buffer, 0, n);
1170 spin_lock(&kvm->mmu_lock);
1172 for (i = 0; i < n / sizeof(long); i++) {
1176 if (!dirty_bitmap[i])
1181 mask = xchg(&dirty_bitmap[i], 0);
1182 dirty_bitmap_buffer[i] = mask;
1185 offset = i * BITS_PER_LONG;
1186 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1191 spin_unlock(&kvm->mmu_lock);
1192 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1196 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect);
1199 bool kvm_largepages_enabled(void)
1201 return largepages_enabled;
1204 void kvm_disable_largepages(void)
1206 largepages_enabled = false;
1208 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1210 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1212 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1214 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1216 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
1218 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn);
1221 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1223 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1225 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1226 memslot->flags & KVM_MEMSLOT_INVALID)
1231 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1233 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1235 struct vm_area_struct *vma;
1236 unsigned long addr, size;
1240 addr = gfn_to_hva(kvm, gfn);
1241 if (kvm_is_error_hva(addr))
1244 down_read(¤t->mm->mmap_sem);
1245 vma = find_vma(current->mm, addr);
1249 size = vma_kernel_pagesize(vma);
1252 up_read(¤t->mm->mmap_sem);
1257 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1259 return slot->flags & KVM_MEM_READONLY;
1262 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1263 gfn_t *nr_pages, bool write)
1265 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1266 return KVM_HVA_ERR_BAD;
1268 if (memslot_is_readonly(slot) && write)
1269 return KVM_HVA_ERR_RO_BAD;
1272 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1274 return __gfn_to_hva_memslot(slot, gfn);
1277 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1280 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1283 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1286 return gfn_to_hva_many(slot, gfn, NULL);
1288 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1290 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1292 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1294 EXPORT_SYMBOL_GPL(gfn_to_hva);
1296 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
1298 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
1300 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
1303 * If writable is set to false, the hva returned by this function is only
1304 * allowed to be read.
1306 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1307 gfn_t gfn, bool *writable)
1309 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1311 if (!kvm_is_error_hva(hva) && writable)
1312 *writable = !memslot_is_readonly(slot);
1317 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1319 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1321 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1324 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
1326 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1328 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1331 static inline int check_user_page_hwpoison(unsigned long addr)
1333 int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
1335 rc = get_user_pages(addr, 1, flags, NULL, NULL);
1336 return rc == -EHWPOISON;
1340 * The atomic path to get the writable pfn which will be stored in @pfn,
1341 * true indicates success, otherwise false is returned.
1343 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1344 bool write_fault, bool *writable, kvm_pfn_t *pfn)
1346 struct page *page[1];
1349 if (!(async || atomic))
1353 * Fast pin a writable pfn only if it is a write fault request
1354 * or the caller allows to map a writable pfn for a read fault
1357 if (!(write_fault || writable))
1360 npages = __get_user_pages_fast(addr, 1, 1, page);
1362 *pfn = page_to_pfn(page[0]);
1373 * The slow path to get the pfn of the specified host virtual address,
1374 * 1 indicates success, -errno is returned if error is detected.
1376 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1377 bool *writable, kvm_pfn_t *pfn)
1379 unsigned int flags = FOLL_HWPOISON;
1386 *writable = write_fault;
1389 flags |= FOLL_WRITE;
1391 flags |= FOLL_NOWAIT;
1393 npages = get_user_pages_unlocked(addr, 1, &page, flags);
1397 /* map read fault as writable if possible */
1398 if (unlikely(!write_fault) && writable) {
1401 if (__get_user_pages_fast(addr, 1, 1, &wpage) == 1) {
1407 *pfn = page_to_pfn(page);
1411 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1413 if (unlikely(!(vma->vm_flags & VM_READ)))
1416 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1422 static int hva_to_pfn_remapped(struct vm_area_struct *vma,
1423 unsigned long addr, bool *async,
1424 bool write_fault, bool *writable,
1430 r = follow_pfn(vma, addr, &pfn);
1433 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
1434 * not call the fault handler, so do it here.
1436 bool unlocked = false;
1437 r = fixup_user_fault(current, current->mm, addr,
1438 (write_fault ? FAULT_FLAG_WRITE : 0),
1445 r = follow_pfn(vma, addr, &pfn);
1455 * Get a reference here because callers of *hva_to_pfn* and
1456 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
1457 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
1458 * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will
1459 * simply do nothing for reserved pfns.
1461 * Whoever called remap_pfn_range is also going to call e.g.
1462 * unmap_mapping_range before the underlying pages are freed,
1463 * causing a call to our MMU notifier.
1472 * Pin guest page in memory and return its pfn.
1473 * @addr: host virtual address which maps memory to the guest
1474 * @atomic: whether this function can sleep
1475 * @async: whether this function need to wait IO complete if the
1476 * host page is not in the memory
1477 * @write_fault: whether we should get a writable host page
1478 * @writable: whether it allows to map a writable host page for !@write_fault
1480 * The function will map a writable host page for these two cases:
1481 * 1): @write_fault = true
1482 * 2): @write_fault = false && @writable, @writable will tell the caller
1483 * whether the mapping is writable.
1485 static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1486 bool write_fault, bool *writable)
1488 struct vm_area_struct *vma;
1492 /* we can do it either atomically or asynchronously, not both */
1493 BUG_ON(atomic && async);
1495 if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1499 return KVM_PFN_ERR_FAULT;
1501 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1505 down_read(¤t->mm->mmap_sem);
1506 if (npages == -EHWPOISON ||
1507 (!async && check_user_page_hwpoison(addr))) {
1508 pfn = KVM_PFN_ERR_HWPOISON;
1513 vma = find_vma_intersection(current->mm, addr, addr + 1);
1516 pfn = KVM_PFN_ERR_FAULT;
1517 else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
1518 r = hva_to_pfn_remapped(vma, addr, async, write_fault, writable, &pfn);
1522 pfn = KVM_PFN_ERR_FAULT;
1524 if (async && vma_is_valid(vma, write_fault))
1526 pfn = KVM_PFN_ERR_FAULT;
1529 up_read(¤t->mm->mmap_sem);
1533 kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn,
1534 bool atomic, bool *async, bool write_fault,
1537 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1539 if (addr == KVM_HVA_ERR_RO_BAD) {
1542 return KVM_PFN_ERR_RO_FAULT;
1545 if (kvm_is_error_hva(addr)) {
1548 return KVM_PFN_NOSLOT;
1551 /* Do not map writable pfn in the readonly memslot. */
1552 if (writable && memslot_is_readonly(slot)) {
1557 return hva_to_pfn(addr, atomic, async, write_fault,
1560 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
1562 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1565 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
1566 write_fault, writable);
1568 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1570 kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1572 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1574 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
1576 kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1578 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1580 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1582 kvm_pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1584 return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm, gfn), gfn);
1586 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1588 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
1590 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1592 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
1594 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1596 return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
1598 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1600 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
1602 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1604 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
1606 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1607 struct page **pages, int nr_pages)
1612 addr = gfn_to_hva_many(slot, gfn, &entry);
1613 if (kvm_is_error_hva(addr))
1616 if (entry < nr_pages)
1619 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1621 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1623 static struct page *kvm_pfn_to_page(kvm_pfn_t pfn)
1625 if (is_error_noslot_pfn(pfn))
1626 return KVM_ERR_PTR_BAD_PAGE;
1628 if (kvm_is_reserved_pfn(pfn)) {
1630 return KVM_ERR_PTR_BAD_PAGE;
1633 return pfn_to_page(pfn);
1636 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1640 pfn = gfn_to_pfn(kvm, gfn);
1642 return kvm_pfn_to_page(pfn);
1644 EXPORT_SYMBOL_GPL(gfn_to_page);
1646 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
1650 pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
1652 return kvm_pfn_to_page(pfn);
1654 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
1656 void kvm_release_page_clean(struct page *page)
1658 WARN_ON(is_error_page(page));
1660 kvm_release_pfn_clean(page_to_pfn(page));
1662 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1664 void kvm_release_pfn_clean(kvm_pfn_t pfn)
1666 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
1667 put_page(pfn_to_page(pfn));
1669 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1671 void kvm_release_page_dirty(struct page *page)
1673 WARN_ON(is_error_page(page));
1675 kvm_release_pfn_dirty(page_to_pfn(page));
1677 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1679 void kvm_release_pfn_dirty(kvm_pfn_t pfn)
1681 kvm_set_pfn_dirty(pfn);
1682 kvm_release_pfn_clean(pfn);
1684 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1686 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
1688 if (!kvm_is_reserved_pfn(pfn)) {
1689 struct page *page = pfn_to_page(pfn);
1691 if (!PageReserved(page))
1695 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1697 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
1699 if (!kvm_is_reserved_pfn(pfn))
1700 mark_page_accessed(pfn_to_page(pfn));
1702 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1704 void kvm_get_pfn(kvm_pfn_t pfn)
1706 if (!kvm_is_reserved_pfn(pfn))
1707 get_page(pfn_to_page(pfn));
1709 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1711 static int next_segment(unsigned long len, int offset)
1713 if (len > PAGE_SIZE - offset)
1714 return PAGE_SIZE - offset;
1719 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
1720 void *data, int offset, int len)
1725 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1726 if (kvm_is_error_hva(addr))
1728 r = __copy_from_user(data, (void __user *)addr + offset, len);
1734 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1737 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1739 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1741 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1743 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
1744 int offset, int len)
1746 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1748 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1750 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
1752 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1754 gfn_t gfn = gpa >> PAGE_SHIFT;
1756 int offset = offset_in_page(gpa);
1759 while ((seg = next_segment(len, offset)) != 0) {
1760 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1770 EXPORT_SYMBOL_GPL(kvm_read_guest);
1772 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
1774 gfn_t gfn = gpa >> PAGE_SHIFT;
1776 int offset = offset_in_page(gpa);
1779 while ((seg = next_segment(len, offset)) != 0) {
1780 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
1790 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
1792 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1793 void *data, int offset, unsigned long len)
1798 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1799 if (kvm_is_error_hva(addr))
1801 pagefault_disable();
1802 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1809 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1812 gfn_t gfn = gpa >> PAGE_SHIFT;
1813 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1814 int offset = offset_in_page(gpa);
1816 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1818 EXPORT_SYMBOL_GPL(kvm_read_guest_atomic);
1820 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
1821 void *data, unsigned long len)
1823 gfn_t gfn = gpa >> PAGE_SHIFT;
1824 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1825 int offset = offset_in_page(gpa);
1827 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1829 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
1831 static int __kvm_write_guest_page(struct kvm_memory_slot *memslot, gfn_t gfn,
1832 const void *data, int offset, int len)
1837 addr = gfn_to_hva_memslot(memslot, gfn);
1838 if (kvm_is_error_hva(addr))
1840 r = __copy_to_user((void __user *)addr + offset, data, len);
1843 mark_page_dirty_in_slot(memslot, gfn);
1847 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
1848 const void *data, int offset, int len)
1850 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1852 return __kvm_write_guest_page(slot, gfn, data, offset, len);
1854 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1856 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
1857 const void *data, int offset, int len)
1859 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1861 return __kvm_write_guest_page(slot, gfn, data, offset, len);
1863 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
1865 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1868 gfn_t gfn = gpa >> PAGE_SHIFT;
1870 int offset = offset_in_page(gpa);
1873 while ((seg = next_segment(len, offset)) != 0) {
1874 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1884 EXPORT_SYMBOL_GPL(kvm_write_guest);
1886 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
1889 gfn_t gfn = gpa >> PAGE_SHIFT;
1891 int offset = offset_in_page(gpa);
1894 while ((seg = next_segment(len, offset)) != 0) {
1895 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
1905 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
1907 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
1908 struct gfn_to_hva_cache *ghc,
1909 gpa_t gpa, unsigned long len)
1911 int offset = offset_in_page(gpa);
1912 gfn_t start_gfn = gpa >> PAGE_SHIFT;
1913 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1914 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1915 gfn_t nr_pages_avail;
1918 ghc->generation = slots->generation;
1920 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
1921 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, NULL);
1922 if (!kvm_is_error_hva(ghc->hva) && nr_pages_needed <= 1) {
1926 * If the requested region crosses two memslots, we still
1927 * verify that the entire region is valid here.
1929 while (start_gfn <= end_gfn) {
1931 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
1932 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1934 if (kvm_is_error_hva(ghc->hva))
1936 start_gfn += nr_pages_avail;
1938 /* Use the slow path for cross page reads and writes. */
1939 ghc->memslot = NULL;
1944 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1945 gpa_t gpa, unsigned long len)
1947 struct kvm_memslots *slots = kvm_memslots(kvm);
1948 return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len);
1950 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1952 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1953 void *data, int offset, unsigned long len)
1955 struct kvm_memslots *slots = kvm_memslots(kvm);
1957 gpa_t gpa = ghc->gpa + offset;
1959 BUG_ON(len + offset > ghc->len);
1961 if (slots->generation != ghc->generation)
1962 __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len);
1964 if (unlikely(!ghc->memslot))
1965 return kvm_write_guest(kvm, gpa, data, len);
1967 if (kvm_is_error_hva(ghc->hva))
1970 r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
1973 mark_page_dirty_in_slot(ghc->memslot, gpa >> PAGE_SHIFT);
1977 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached);
1979 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1980 void *data, unsigned long len)
1982 return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len);
1984 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1986 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1987 void *data, unsigned long len)
1989 struct kvm_memslots *slots = kvm_memslots(kvm);
1992 BUG_ON(len > ghc->len);
1994 if (slots->generation != ghc->generation)
1995 __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len);
1997 if (unlikely(!ghc->memslot))
1998 return kvm_read_guest(kvm, ghc->gpa, data, len);
2000 if (kvm_is_error_hva(ghc->hva))
2003 r = __copy_from_user(data, (void __user *)ghc->hva, len);
2009 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
2011 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
2013 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
2015 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
2017 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
2019 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
2021 gfn_t gfn = gpa >> PAGE_SHIFT;
2023 int offset = offset_in_page(gpa);
2026 while ((seg = next_segment(len, offset)) != 0) {
2027 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
2036 EXPORT_SYMBOL_GPL(kvm_clear_guest);
2038 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot,
2041 if (memslot && memslot->dirty_bitmap) {
2042 unsigned long rel_gfn = gfn - memslot->base_gfn;
2044 set_bit_le(rel_gfn, memslot->dirty_bitmap);
2048 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
2050 struct kvm_memory_slot *memslot;
2052 memslot = gfn_to_memslot(kvm, gfn);
2053 mark_page_dirty_in_slot(memslot, gfn);
2055 EXPORT_SYMBOL_GPL(mark_page_dirty);
2057 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
2059 struct kvm_memory_slot *memslot;
2061 memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2062 mark_page_dirty_in_slot(memslot, gfn);
2064 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
2066 void kvm_sigset_activate(struct kvm_vcpu *vcpu)
2068 if (!vcpu->sigset_active)
2072 * This does a lockless modification of ->real_blocked, which is fine
2073 * because, only current can change ->real_blocked and all readers of
2074 * ->real_blocked don't care as long ->real_blocked is always a subset
2077 sigprocmask(SIG_SETMASK, &vcpu->sigset, ¤t->real_blocked);
2080 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu)
2082 if (!vcpu->sigset_active)
2085 sigprocmask(SIG_SETMASK, ¤t->real_blocked, NULL);
2086 sigemptyset(¤t->real_blocked);
2089 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
2091 unsigned int old, val, grow;
2093 old = val = vcpu->halt_poll_ns;
2094 grow = READ_ONCE(halt_poll_ns_grow);
2096 if (val == 0 && grow)
2101 if (val > halt_poll_ns)
2104 vcpu->halt_poll_ns = val;
2105 trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
2108 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
2110 unsigned int old, val, shrink;
2112 old = val = vcpu->halt_poll_ns;
2113 shrink = READ_ONCE(halt_poll_ns_shrink);
2119 vcpu->halt_poll_ns = val;
2120 trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
2123 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
2125 if (kvm_arch_vcpu_runnable(vcpu)) {
2126 kvm_make_request(KVM_REQ_UNHALT, vcpu);
2129 if (kvm_cpu_has_pending_timer(vcpu))
2131 if (signal_pending(current))
2138 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2140 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
2143 DECLARE_SWAITQUEUE(wait);
2144 bool waited = false;
2147 start = cur = ktime_get();
2148 if (vcpu->halt_poll_ns) {
2149 ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns);
2151 ++vcpu->stat.halt_attempted_poll;
2154 * This sets KVM_REQ_UNHALT if an interrupt
2157 if (kvm_vcpu_check_block(vcpu) < 0) {
2158 ++vcpu->stat.halt_successful_poll;
2159 if (!vcpu_valid_wakeup(vcpu))
2160 ++vcpu->stat.halt_poll_invalid;
2164 } while (single_task_running() && ktime_before(cur, stop));
2167 kvm_arch_vcpu_blocking(vcpu);
2170 prepare_to_swait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
2172 if (kvm_vcpu_check_block(vcpu) < 0)
2179 finish_swait(&vcpu->wq, &wait);
2182 kvm_arch_vcpu_unblocking(vcpu);
2184 block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
2186 if (!vcpu_valid_wakeup(vcpu))
2187 shrink_halt_poll_ns(vcpu);
2188 else if (halt_poll_ns) {
2189 if (block_ns <= vcpu->halt_poll_ns)
2191 /* we had a long block, shrink polling */
2192 else if (vcpu->halt_poll_ns && block_ns > halt_poll_ns)
2193 shrink_halt_poll_ns(vcpu);
2194 /* we had a short halt and our poll time is too small */
2195 else if (vcpu->halt_poll_ns < halt_poll_ns &&
2196 block_ns < halt_poll_ns)
2197 grow_halt_poll_ns(vcpu);
2199 vcpu->halt_poll_ns = 0;
2201 trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu));
2202 kvm_arch_vcpu_block_finish(vcpu);
2204 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
2206 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
2208 struct swait_queue_head *wqp;
2210 wqp = kvm_arch_vcpu_wq(vcpu);
2211 if (swq_has_sleeper(wqp)) {
2213 ++vcpu->stat.halt_wakeup;
2219 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
2223 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2225 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
2228 int cpu = vcpu->cpu;
2230 if (kvm_vcpu_wake_up(vcpu))
2234 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
2235 if (kvm_arch_vcpu_should_kick(vcpu))
2236 smp_send_reschedule(cpu);
2239 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
2240 #endif /* !CONFIG_S390 */
2242 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
2245 struct task_struct *task = NULL;
2249 pid = rcu_dereference(target->pid);
2251 task = get_pid_task(pid, PIDTYPE_PID);
2255 ret = yield_to(task, 1);
2256 put_task_struct(task);
2260 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
2263 * Helper that checks whether a VCPU is eligible for directed yield.
2264 * Most eligible candidate to yield is decided by following heuristics:
2266 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2267 * (preempted lock holder), indicated by @in_spin_loop.
2268 * Set at the beiginning and cleared at the end of interception/PLE handler.
2270 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2271 * chance last time (mostly it has become eligible now since we have probably
2272 * yielded to lockholder in last iteration. This is done by toggling
2273 * @dy_eligible each time a VCPU checked for eligibility.)
2275 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2276 * to preempted lock-holder could result in wrong VCPU selection and CPU
2277 * burning. Giving priority for a potential lock-holder increases lock
2280 * Since algorithm is based on heuristics, accessing another VCPU data without
2281 * locking does not harm. It may result in trying to yield to same VCPU, fail
2282 * and continue with next VCPU and so on.
2284 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
2286 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2289 eligible = !vcpu->spin_loop.in_spin_loop ||
2290 vcpu->spin_loop.dy_eligible;
2292 if (vcpu->spin_loop.in_spin_loop)
2293 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
2301 void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode)
2303 struct kvm *kvm = me->kvm;
2304 struct kvm_vcpu *vcpu;
2305 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
2311 kvm_vcpu_set_in_spin_loop(me, true);
2313 * We boost the priority of a VCPU that is runnable but not
2314 * currently running, because it got preempted by something
2315 * else and called schedule in __vcpu_run. Hopefully that
2316 * VCPU is holding the lock that we need and will release it.
2317 * We approximate round-robin by starting at the last boosted VCPU.
2319 for (pass = 0; pass < 2 && !yielded && try; pass++) {
2320 kvm_for_each_vcpu(i, vcpu, kvm) {
2321 if (!pass && i <= last_boosted_vcpu) {
2322 i = last_boosted_vcpu;
2324 } else if (pass && i > last_boosted_vcpu)
2326 if (!READ_ONCE(vcpu->preempted))
2330 if (swait_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
2332 if (yield_to_kernel_mode && !kvm_arch_vcpu_in_kernel(vcpu))
2334 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
2337 yielded = kvm_vcpu_yield_to(vcpu);
2339 kvm->last_boosted_vcpu = i;
2341 } else if (yielded < 0) {
2348 kvm_vcpu_set_in_spin_loop(me, false);
2350 /* Ensure vcpu is not eligible during next spinloop */
2351 kvm_vcpu_set_dy_eligible(me, false);
2353 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
2355 static vm_fault_t kvm_vcpu_fault(struct vm_fault *vmf)
2357 struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data;
2360 if (vmf->pgoff == 0)
2361 page = virt_to_page(vcpu->run);
2363 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
2364 page = virt_to_page(vcpu->arch.pio_data);
2366 #ifdef CONFIG_KVM_MMIO
2367 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
2368 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
2371 return kvm_arch_vcpu_fault(vcpu, vmf);
2377 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
2378 .fault = kvm_vcpu_fault,
2381 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2383 vma->vm_ops = &kvm_vcpu_vm_ops;
2387 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2389 struct kvm_vcpu *vcpu = filp->private_data;
2391 debugfs_remove_recursive(vcpu->debugfs_dentry);
2392 kvm_put_kvm(vcpu->kvm);
2396 static struct file_operations kvm_vcpu_fops = {
2397 .release = kvm_vcpu_release,
2398 .unlocked_ioctl = kvm_vcpu_ioctl,
2399 #ifdef CONFIG_KVM_COMPAT
2400 .compat_ioctl = kvm_vcpu_compat_ioctl,
2402 .mmap = kvm_vcpu_mmap,
2403 .llseek = noop_llseek,
2407 * Allocates an inode for the vcpu.
2409 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2411 char name[8 + 1 + ITOA_MAX_LEN + 1];
2413 snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id);
2414 return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
2417 static int kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
2419 char dir_name[ITOA_MAX_LEN * 2];
2422 if (!kvm_arch_has_vcpu_debugfs())
2425 if (!debugfs_initialized())
2428 snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
2429 vcpu->debugfs_dentry = debugfs_create_dir(dir_name,
2430 vcpu->kvm->debugfs_dentry);
2431 if (!vcpu->debugfs_dentry)
2434 ret = kvm_arch_create_vcpu_debugfs(vcpu);
2436 debugfs_remove_recursive(vcpu->debugfs_dentry);
2444 * Creates some virtual cpus. Good luck creating more than one.
2446 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
2449 struct kvm_vcpu *vcpu;
2451 if (id >= KVM_MAX_VCPU_ID)
2454 mutex_lock(&kvm->lock);
2455 if (kvm->created_vcpus == KVM_MAX_VCPUS) {
2456 mutex_unlock(&kvm->lock);
2460 kvm->created_vcpus++;
2461 mutex_unlock(&kvm->lock);
2463 vcpu = kvm_arch_vcpu_create(kvm, id);
2466 goto vcpu_decrement;
2469 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2471 r = kvm_arch_vcpu_setup(vcpu);
2475 r = kvm_create_vcpu_debugfs(vcpu);
2479 mutex_lock(&kvm->lock);
2480 if (kvm_get_vcpu_by_id(kvm, id)) {
2482 goto unlock_vcpu_destroy;
2485 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
2487 /* Now it's all set up, let userspace reach it */
2489 r = create_vcpu_fd(vcpu);
2492 goto unlock_vcpu_destroy;
2495 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
2498 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
2499 * before kvm->online_vcpu's incremented value.
2502 atomic_inc(&kvm->online_vcpus);
2504 mutex_unlock(&kvm->lock);
2505 kvm_arch_vcpu_postcreate(vcpu);
2508 unlock_vcpu_destroy:
2509 mutex_unlock(&kvm->lock);
2510 debugfs_remove_recursive(vcpu->debugfs_dentry);
2512 kvm_arch_vcpu_destroy(vcpu);
2514 mutex_lock(&kvm->lock);
2515 kvm->created_vcpus--;
2516 mutex_unlock(&kvm->lock);
2520 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2523 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2524 vcpu->sigset_active = 1;
2525 vcpu->sigset = *sigset;
2527 vcpu->sigset_active = 0;
2531 static long kvm_vcpu_ioctl(struct file *filp,
2532 unsigned int ioctl, unsigned long arg)
2534 struct kvm_vcpu *vcpu = filp->private_data;
2535 void __user *argp = (void __user *)arg;
2537 struct kvm_fpu *fpu = NULL;
2538 struct kvm_sregs *kvm_sregs = NULL;
2540 if (vcpu->kvm->mm != current->mm)
2543 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2547 * Some architectures have vcpu ioctls that are asynchronous to vcpu
2548 * execution; mutex_lock() would break them.
2550 r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg);
2551 if (r != -ENOIOCTLCMD)
2554 if (mutex_lock_killable(&vcpu->mutex))
2562 oldpid = rcu_access_pointer(vcpu->pid);
2563 if (unlikely(oldpid != current->pids[PIDTYPE_PID].pid)) {
2564 /* The thread running this VCPU changed. */
2567 r = kvm_arch_vcpu_run_pid_change(vcpu);
2571 newpid = get_task_pid(current, PIDTYPE_PID);
2572 rcu_assign_pointer(vcpu->pid, newpid);
2577 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2578 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2581 case KVM_GET_REGS: {
2582 struct kvm_regs *kvm_regs;
2585 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2588 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2592 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2599 case KVM_SET_REGS: {
2600 struct kvm_regs *kvm_regs;
2603 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2604 if (IS_ERR(kvm_regs)) {
2605 r = PTR_ERR(kvm_regs);
2608 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2612 case KVM_GET_SREGS: {
2613 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2617 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2621 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2626 case KVM_SET_SREGS: {
2627 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2628 if (IS_ERR(kvm_sregs)) {
2629 r = PTR_ERR(kvm_sregs);
2633 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2636 case KVM_GET_MP_STATE: {
2637 struct kvm_mp_state mp_state;
2639 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2643 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
2648 case KVM_SET_MP_STATE: {
2649 struct kvm_mp_state mp_state;
2652 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
2654 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2657 case KVM_TRANSLATE: {
2658 struct kvm_translation tr;
2661 if (copy_from_user(&tr, argp, sizeof(tr)))
2663 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2667 if (copy_to_user(argp, &tr, sizeof(tr)))
2672 case KVM_SET_GUEST_DEBUG: {
2673 struct kvm_guest_debug dbg;
2676 if (copy_from_user(&dbg, argp, sizeof(dbg)))
2678 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2681 case KVM_SET_SIGNAL_MASK: {
2682 struct kvm_signal_mask __user *sigmask_arg = argp;
2683 struct kvm_signal_mask kvm_sigmask;
2684 sigset_t sigset, *p;
2689 if (copy_from_user(&kvm_sigmask, argp,
2690 sizeof(kvm_sigmask)))
2693 if (kvm_sigmask.len != sizeof(sigset))
2696 if (copy_from_user(&sigset, sigmask_arg->sigset,
2701 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2705 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2709 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2713 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2719 fpu = memdup_user(argp, sizeof(*fpu));
2725 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2729 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2732 mutex_unlock(&vcpu->mutex);
2738 #ifdef CONFIG_KVM_COMPAT
2739 static long kvm_vcpu_compat_ioctl(struct file *filp,
2740 unsigned int ioctl, unsigned long arg)
2742 struct kvm_vcpu *vcpu = filp->private_data;
2743 void __user *argp = compat_ptr(arg);
2746 if (vcpu->kvm->mm != current->mm)
2750 case KVM_SET_SIGNAL_MASK: {
2751 struct kvm_signal_mask __user *sigmask_arg = argp;
2752 struct kvm_signal_mask kvm_sigmask;
2757 if (copy_from_user(&kvm_sigmask, argp,
2758 sizeof(kvm_sigmask)))
2761 if (kvm_sigmask.len != sizeof(compat_sigset_t))
2764 if (get_compat_sigset(&sigset, (void *)sigmask_arg->sigset))
2766 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2768 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2772 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2780 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2781 int (*accessor)(struct kvm_device *dev,
2782 struct kvm_device_attr *attr),
2785 struct kvm_device_attr attr;
2790 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2793 return accessor(dev, &attr);
2796 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2799 struct kvm_device *dev = filp->private_data;
2802 case KVM_SET_DEVICE_ATTR:
2803 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2804 case KVM_GET_DEVICE_ATTR:
2805 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2806 case KVM_HAS_DEVICE_ATTR:
2807 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2809 if (dev->ops->ioctl)
2810 return dev->ops->ioctl(dev, ioctl, arg);
2816 static int kvm_device_release(struct inode *inode, struct file *filp)
2818 struct kvm_device *dev = filp->private_data;
2819 struct kvm *kvm = dev->kvm;
2825 static const struct file_operations kvm_device_fops = {
2826 .unlocked_ioctl = kvm_device_ioctl,
2827 #ifdef CONFIG_KVM_COMPAT
2828 .compat_ioctl = kvm_device_ioctl,
2830 .release = kvm_device_release,
2833 struct kvm_device *kvm_device_from_filp(struct file *filp)
2835 if (filp->f_op != &kvm_device_fops)
2838 return filp->private_data;
2841 static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
2842 #ifdef CONFIG_KVM_MPIC
2843 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
2844 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
2848 int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
2850 if (type >= ARRAY_SIZE(kvm_device_ops_table))
2853 if (kvm_device_ops_table[type] != NULL)
2856 kvm_device_ops_table[type] = ops;
2860 void kvm_unregister_device_ops(u32 type)
2862 if (kvm_device_ops_table[type] != NULL)
2863 kvm_device_ops_table[type] = NULL;
2866 static int kvm_ioctl_create_device(struct kvm *kvm,
2867 struct kvm_create_device *cd)
2869 struct kvm_device_ops *ops = NULL;
2870 struct kvm_device *dev;
2871 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2874 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
2877 ops = kvm_device_ops_table[cd->type];
2884 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2891 mutex_lock(&kvm->lock);
2892 ret = ops->create(dev, cd->type);
2894 mutex_unlock(&kvm->lock);
2898 list_add(&dev->vm_node, &kvm->devices);
2899 mutex_unlock(&kvm->lock);
2904 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2906 mutex_lock(&kvm->lock);
2907 list_del(&dev->vm_node);
2908 mutex_unlock(&kvm->lock);
2918 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2921 case KVM_CAP_USER_MEMORY:
2922 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2923 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2924 case KVM_CAP_INTERNAL_ERROR_DATA:
2925 #ifdef CONFIG_HAVE_KVM_MSI
2926 case KVM_CAP_SIGNAL_MSI:
2928 #ifdef CONFIG_HAVE_KVM_IRQFD
2930 case KVM_CAP_IRQFD_RESAMPLE:
2932 case KVM_CAP_IOEVENTFD_ANY_LENGTH:
2933 case KVM_CAP_CHECK_EXTENSION_VM:
2935 #ifdef CONFIG_KVM_MMIO
2936 case KVM_CAP_COALESCED_MMIO:
2937 return KVM_COALESCED_MMIO_PAGE_OFFSET;
2939 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2940 case KVM_CAP_IRQ_ROUTING:
2941 return KVM_MAX_IRQ_ROUTES;
2943 #if KVM_ADDRESS_SPACE_NUM > 1
2944 case KVM_CAP_MULTI_ADDRESS_SPACE:
2945 return KVM_ADDRESS_SPACE_NUM;
2947 case KVM_CAP_MAX_VCPU_ID:
2948 return KVM_MAX_VCPU_ID;
2952 return kvm_vm_ioctl_check_extension(kvm, arg);
2955 static long kvm_vm_ioctl(struct file *filp,
2956 unsigned int ioctl, unsigned long arg)
2958 struct kvm *kvm = filp->private_data;
2959 void __user *argp = (void __user *)arg;
2962 if (kvm->mm != current->mm)
2965 case KVM_CREATE_VCPU:
2966 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2968 case KVM_SET_USER_MEMORY_REGION: {
2969 struct kvm_userspace_memory_region kvm_userspace_mem;
2972 if (copy_from_user(&kvm_userspace_mem, argp,
2973 sizeof(kvm_userspace_mem)))
2976 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2979 case KVM_GET_DIRTY_LOG: {
2980 struct kvm_dirty_log log;
2983 if (copy_from_user(&log, argp, sizeof(log)))
2985 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2988 #ifdef CONFIG_KVM_MMIO
2989 case KVM_REGISTER_COALESCED_MMIO: {
2990 struct kvm_coalesced_mmio_zone zone;
2993 if (copy_from_user(&zone, argp, sizeof(zone)))
2995 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2998 case KVM_UNREGISTER_COALESCED_MMIO: {
2999 struct kvm_coalesced_mmio_zone zone;
3002 if (copy_from_user(&zone, argp, sizeof(zone)))
3004 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
3009 struct kvm_irqfd data;
3012 if (copy_from_user(&data, argp, sizeof(data)))
3014 r = kvm_irqfd(kvm, &data);
3017 case KVM_IOEVENTFD: {
3018 struct kvm_ioeventfd data;
3021 if (copy_from_user(&data, argp, sizeof(data)))
3023 r = kvm_ioeventfd(kvm, &data);
3026 #ifdef CONFIG_HAVE_KVM_MSI
3027 case KVM_SIGNAL_MSI: {
3031 if (copy_from_user(&msi, argp, sizeof(msi)))
3033 r = kvm_send_userspace_msi(kvm, &msi);
3037 #ifdef __KVM_HAVE_IRQ_LINE
3038 case KVM_IRQ_LINE_STATUS:
3039 case KVM_IRQ_LINE: {
3040 struct kvm_irq_level irq_event;
3043 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
3046 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
3047 ioctl == KVM_IRQ_LINE_STATUS);
3052 if (ioctl == KVM_IRQ_LINE_STATUS) {
3053 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
3061 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3062 case KVM_SET_GSI_ROUTING: {
3063 struct kvm_irq_routing routing;
3064 struct kvm_irq_routing __user *urouting;
3065 struct kvm_irq_routing_entry *entries = NULL;
3068 if (copy_from_user(&routing, argp, sizeof(routing)))
3071 if (!kvm_arch_can_set_irq_routing(kvm))
3073 if (routing.nr > KVM_MAX_IRQ_ROUTES)
3079 entries = vmalloc(array_size(sizeof(*entries),
3085 if (copy_from_user(entries, urouting->entries,
3086 routing.nr * sizeof(*entries)))
3087 goto out_free_irq_routing;
3089 r = kvm_set_irq_routing(kvm, entries, routing.nr,
3091 out_free_irq_routing:
3095 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
3096 case KVM_CREATE_DEVICE: {
3097 struct kvm_create_device cd;
3100 if (copy_from_user(&cd, argp, sizeof(cd)))
3103 r = kvm_ioctl_create_device(kvm, &cd);
3108 if (copy_to_user(argp, &cd, sizeof(cd)))
3114 case KVM_CHECK_EXTENSION:
3115 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
3118 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
3124 #ifdef CONFIG_KVM_COMPAT
3125 struct compat_kvm_dirty_log {
3129 compat_uptr_t dirty_bitmap; /* one bit per page */
3134 static long kvm_vm_compat_ioctl(struct file *filp,
3135 unsigned int ioctl, unsigned long arg)
3137 struct kvm *kvm = filp->private_data;
3140 if (kvm->mm != current->mm)
3143 case KVM_GET_DIRTY_LOG: {
3144 struct compat_kvm_dirty_log compat_log;
3145 struct kvm_dirty_log log;
3147 if (copy_from_user(&compat_log, (void __user *)arg,
3148 sizeof(compat_log)))
3150 log.slot = compat_log.slot;
3151 log.padding1 = compat_log.padding1;
3152 log.padding2 = compat_log.padding2;
3153 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
3155 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3159 r = kvm_vm_ioctl(filp, ioctl, arg);
3165 static struct file_operations kvm_vm_fops = {
3166 .release = kvm_vm_release,
3167 .unlocked_ioctl = kvm_vm_ioctl,
3168 #ifdef CONFIG_KVM_COMPAT
3169 .compat_ioctl = kvm_vm_compat_ioctl,
3171 .llseek = noop_llseek,
3174 static int kvm_dev_ioctl_create_vm(unsigned long type)
3180 kvm = kvm_create_vm(type);
3182 return PTR_ERR(kvm);
3183 #ifdef CONFIG_KVM_MMIO
3184 r = kvm_coalesced_mmio_init(kvm);
3188 r = get_unused_fd_flags(O_CLOEXEC);
3192 file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
3200 * Don't call kvm_put_kvm anymore at this point; file->f_op is
3201 * already set, with ->release() being kvm_vm_release(). In error
3202 * cases it will be called by the final fput(file) and will take
3203 * care of doing kvm_put_kvm(kvm).
3205 if (kvm_create_vm_debugfs(kvm, r) < 0) {
3210 kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm);
3212 fd_install(r, file);
3220 static long kvm_dev_ioctl(struct file *filp,
3221 unsigned int ioctl, unsigned long arg)
3226 case KVM_GET_API_VERSION:
3229 r = KVM_API_VERSION;
3232 r = kvm_dev_ioctl_create_vm(arg);
3234 case KVM_CHECK_EXTENSION:
3235 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
3237 case KVM_GET_VCPU_MMAP_SIZE:
3240 r = PAGE_SIZE; /* struct kvm_run */
3242 r += PAGE_SIZE; /* pio data page */
3244 #ifdef CONFIG_KVM_MMIO
3245 r += PAGE_SIZE; /* coalesced mmio ring page */
3248 case KVM_TRACE_ENABLE:
3249 case KVM_TRACE_PAUSE:
3250 case KVM_TRACE_DISABLE:
3254 return kvm_arch_dev_ioctl(filp, ioctl, arg);
3260 static struct file_operations kvm_chardev_ops = {
3261 .unlocked_ioctl = kvm_dev_ioctl,
3262 .compat_ioctl = kvm_dev_ioctl,
3263 .llseek = noop_llseek,
3266 static struct miscdevice kvm_dev = {
3272 static void hardware_enable_nolock(void *junk)
3274 int cpu = raw_smp_processor_id();
3277 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
3280 cpumask_set_cpu(cpu, cpus_hardware_enabled);
3282 r = kvm_arch_hardware_enable();
3285 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3286 atomic_inc(&hardware_enable_failed);
3287 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
3291 static int kvm_starting_cpu(unsigned int cpu)
3293 raw_spin_lock(&kvm_count_lock);
3294 if (kvm_usage_count)
3295 hardware_enable_nolock(NULL);
3296 raw_spin_unlock(&kvm_count_lock);
3300 static void hardware_disable_nolock(void *junk)
3302 int cpu = raw_smp_processor_id();
3304 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
3306 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3307 kvm_arch_hardware_disable();
3310 static int kvm_dying_cpu(unsigned int cpu)
3312 raw_spin_lock(&kvm_count_lock);
3313 if (kvm_usage_count)
3314 hardware_disable_nolock(NULL);
3315 raw_spin_unlock(&kvm_count_lock);
3319 static void hardware_disable_all_nolock(void)
3321 BUG_ON(!kvm_usage_count);
3324 if (!kvm_usage_count)
3325 on_each_cpu(hardware_disable_nolock, NULL, 1);
3328 static void hardware_disable_all(void)
3330 raw_spin_lock(&kvm_count_lock);
3331 hardware_disable_all_nolock();
3332 raw_spin_unlock(&kvm_count_lock);
3335 static int hardware_enable_all(void)
3339 raw_spin_lock(&kvm_count_lock);
3342 if (kvm_usage_count == 1) {
3343 atomic_set(&hardware_enable_failed, 0);
3344 on_each_cpu(hardware_enable_nolock, NULL, 1);
3346 if (atomic_read(&hardware_enable_failed)) {
3347 hardware_disable_all_nolock();
3352 raw_spin_unlock(&kvm_count_lock);
3357 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3361 * Some (well, at least mine) BIOSes hang on reboot if
3364 * And Intel TXT required VMX off for all cpu when system shutdown.
3366 pr_info("kvm: exiting hardware virtualization\n");
3367 kvm_rebooting = true;
3368 on_each_cpu(hardware_disable_nolock, NULL, 1);
3372 static struct notifier_block kvm_reboot_notifier = {
3373 .notifier_call = kvm_reboot,
3377 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3381 for (i = 0; i < bus->dev_count; i++) {
3382 struct kvm_io_device *pos = bus->range[i].dev;
3384 kvm_iodevice_destructor(pos);
3389 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
3390 const struct kvm_io_range *r2)
3392 gpa_t addr1 = r1->addr;
3393 gpa_t addr2 = r2->addr;
3398 /* If r2->len == 0, match the exact address. If r2->len != 0,
3399 * accept any overlapping write. Any order is acceptable for
3400 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
3401 * we process all of them.
3414 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
3416 return kvm_io_bus_cmp(p1, p2);
3419 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
3420 gpa_t addr, int len)
3422 struct kvm_io_range *range, key;
3425 key = (struct kvm_io_range) {
3430 range = bsearch(&key, bus->range, bus->dev_count,
3431 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
3435 off = range - bus->range;
3437 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
3443 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3444 struct kvm_io_range *range, const void *val)
3448 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3452 while (idx < bus->dev_count &&
3453 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3454 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
3463 /* kvm_io_bus_write - called under kvm->slots_lock */
3464 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3465 int len, const void *val)
3467 struct kvm_io_bus *bus;
3468 struct kvm_io_range range;
3471 range = (struct kvm_io_range) {
3476 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3479 r = __kvm_io_bus_write(vcpu, bus, &range, val);
3480 return r < 0 ? r : 0;
3483 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3484 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
3485 gpa_t addr, int len, const void *val, long cookie)
3487 struct kvm_io_bus *bus;
3488 struct kvm_io_range range;
3490 range = (struct kvm_io_range) {
3495 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3499 /* First try the device referenced by cookie. */
3500 if ((cookie >= 0) && (cookie < bus->dev_count) &&
3501 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
3502 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
3507 * cookie contained garbage; fall back to search and return the
3508 * correct cookie value.
3510 return __kvm_io_bus_write(vcpu, bus, &range, val);
3513 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3514 struct kvm_io_range *range, void *val)
3518 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3522 while (idx < bus->dev_count &&
3523 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3524 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
3532 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
3534 /* kvm_io_bus_read - called under kvm->slots_lock */
3535 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3538 struct kvm_io_bus *bus;
3539 struct kvm_io_range range;
3542 range = (struct kvm_io_range) {
3547 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3550 r = __kvm_io_bus_read(vcpu, bus, &range, val);
3551 return r < 0 ? r : 0;
3555 /* Caller must hold slots_lock. */
3556 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3557 int len, struct kvm_io_device *dev)
3560 struct kvm_io_bus *new_bus, *bus;
3561 struct kvm_io_range range;
3563 bus = kvm_get_bus(kvm, bus_idx);
3567 /* exclude ioeventfd which is limited by maximum fd */
3568 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3571 new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count + 1) *
3572 sizeof(struct kvm_io_range)), GFP_KERNEL);
3576 range = (struct kvm_io_range) {
3582 for (i = 0; i < bus->dev_count; i++)
3583 if (kvm_io_bus_cmp(&bus->range[i], &range) > 0)
3586 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3587 new_bus->dev_count++;
3588 new_bus->range[i] = range;
3589 memcpy(new_bus->range + i + 1, bus->range + i,
3590 (bus->dev_count - i) * sizeof(struct kvm_io_range));
3591 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3592 synchronize_srcu_expedited(&kvm->srcu);
3598 /* Caller must hold slots_lock. */
3599 void kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3600 struct kvm_io_device *dev)
3603 struct kvm_io_bus *new_bus, *bus;
3605 bus = kvm_get_bus(kvm, bus_idx);
3609 for (i = 0; i < bus->dev_count; i++)
3610 if (bus->range[i].dev == dev) {
3614 if (i == bus->dev_count)
3617 new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3618 sizeof(struct kvm_io_range)), GFP_KERNEL);
3620 pr_err("kvm: failed to shrink bus, removing it completely\n");
3624 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3625 new_bus->dev_count--;
3626 memcpy(new_bus->range + i, bus->range + i + 1,
3627 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3630 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3631 synchronize_srcu_expedited(&kvm->srcu);
3636 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3639 struct kvm_io_bus *bus;
3640 int dev_idx, srcu_idx;
3641 struct kvm_io_device *iodev = NULL;
3643 srcu_idx = srcu_read_lock(&kvm->srcu);
3645 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
3649 dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
3653 iodev = bus->range[dev_idx].dev;
3656 srcu_read_unlock(&kvm->srcu, srcu_idx);
3660 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
3662 static int kvm_debugfs_open(struct inode *inode, struct file *file,
3663 int (*get)(void *, u64 *), int (*set)(void *, u64),
3666 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
3669 /* The debugfs files are a reference to the kvm struct which
3670 * is still valid when kvm_destroy_vm is called.
3671 * To avoid the race between open and the removal of the debugfs
3672 * directory we test against the users count.
3674 if (!refcount_inc_not_zero(&stat_data->kvm->users_count))
3677 if (simple_attr_open(inode, file, get, set, fmt)) {
3678 kvm_put_kvm(stat_data->kvm);
3685 static int kvm_debugfs_release(struct inode *inode, struct file *file)
3687 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
3690 simple_attr_release(inode, file);
3691 kvm_put_kvm(stat_data->kvm);
3696 static int vm_stat_get_per_vm(void *data, u64 *val)
3698 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3700 *val = *(ulong *)((void *)stat_data->kvm + stat_data->offset);
3705 static int vm_stat_clear_per_vm(void *data, u64 val)
3707 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3712 *(ulong *)((void *)stat_data->kvm + stat_data->offset) = 0;
3717 static int vm_stat_get_per_vm_open(struct inode *inode, struct file *file)
3719 __simple_attr_check_format("%llu\n", 0ull);
3720 return kvm_debugfs_open(inode, file, vm_stat_get_per_vm,
3721 vm_stat_clear_per_vm, "%llu\n");
3724 static const struct file_operations vm_stat_get_per_vm_fops = {
3725 .owner = THIS_MODULE,
3726 .open = vm_stat_get_per_vm_open,
3727 .release = kvm_debugfs_release,
3728 .read = simple_attr_read,
3729 .write = simple_attr_write,
3730 .llseek = no_llseek,
3733 static int vcpu_stat_get_per_vm(void *data, u64 *val)
3736 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3737 struct kvm_vcpu *vcpu;
3741 kvm_for_each_vcpu(i, vcpu, stat_data->kvm)
3742 *val += *(u64 *)((void *)vcpu + stat_data->offset);
3747 static int vcpu_stat_clear_per_vm(void *data, u64 val)
3750 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3751 struct kvm_vcpu *vcpu;
3756 kvm_for_each_vcpu(i, vcpu, stat_data->kvm)
3757 *(u64 *)((void *)vcpu + stat_data->offset) = 0;
3762 static int vcpu_stat_get_per_vm_open(struct inode *inode, struct file *file)
3764 __simple_attr_check_format("%llu\n", 0ull);
3765 return kvm_debugfs_open(inode, file, vcpu_stat_get_per_vm,
3766 vcpu_stat_clear_per_vm, "%llu\n");
3769 static const struct file_operations vcpu_stat_get_per_vm_fops = {
3770 .owner = THIS_MODULE,
3771 .open = vcpu_stat_get_per_vm_open,
3772 .release = kvm_debugfs_release,
3773 .read = simple_attr_read,
3774 .write = simple_attr_write,
3775 .llseek = no_llseek,
3778 static const struct file_operations *stat_fops_per_vm[] = {
3779 [KVM_STAT_VCPU] = &vcpu_stat_get_per_vm_fops,
3780 [KVM_STAT_VM] = &vm_stat_get_per_vm_fops,
3783 static int vm_stat_get(void *_offset, u64 *val)
3785 unsigned offset = (long)_offset;
3787 struct kvm_stat_data stat_tmp = {.offset = offset};
3791 spin_lock(&kvm_lock);
3792 list_for_each_entry(kvm, &vm_list, vm_list) {
3794 vm_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
3797 spin_unlock(&kvm_lock);
3801 static int vm_stat_clear(void *_offset, u64 val)
3803 unsigned offset = (long)_offset;
3805 struct kvm_stat_data stat_tmp = {.offset = offset};
3810 spin_lock(&kvm_lock);
3811 list_for_each_entry(kvm, &vm_list, vm_list) {
3813 vm_stat_clear_per_vm((void *)&stat_tmp, 0);
3815 spin_unlock(&kvm_lock);
3820 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n");
3822 static int vcpu_stat_get(void *_offset, u64 *val)
3824 unsigned offset = (long)_offset;
3826 struct kvm_stat_data stat_tmp = {.offset = offset};
3830 spin_lock(&kvm_lock);
3831 list_for_each_entry(kvm, &vm_list, vm_list) {
3833 vcpu_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
3836 spin_unlock(&kvm_lock);
3840 static int vcpu_stat_clear(void *_offset, u64 val)
3842 unsigned offset = (long)_offset;
3844 struct kvm_stat_data stat_tmp = {.offset = offset};
3849 spin_lock(&kvm_lock);
3850 list_for_each_entry(kvm, &vm_list, vm_list) {
3852 vcpu_stat_clear_per_vm((void *)&stat_tmp, 0);
3854 spin_unlock(&kvm_lock);
3859 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear,
3862 static const struct file_operations *stat_fops[] = {
3863 [KVM_STAT_VCPU] = &vcpu_stat_fops,
3864 [KVM_STAT_VM] = &vm_stat_fops,
3867 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm)
3869 struct kobj_uevent_env *env;
3870 unsigned long long created, active;
3872 if (!kvm_dev.this_device || !kvm)
3875 spin_lock(&kvm_lock);
3876 if (type == KVM_EVENT_CREATE_VM) {
3877 kvm_createvm_count++;
3879 } else if (type == KVM_EVENT_DESTROY_VM) {
3882 created = kvm_createvm_count;
3883 active = kvm_active_vms;
3884 spin_unlock(&kvm_lock);
3886 env = kzalloc(sizeof(*env), GFP_KERNEL);
3890 add_uevent_var(env, "CREATED=%llu", created);
3891 add_uevent_var(env, "COUNT=%llu", active);
3893 if (type == KVM_EVENT_CREATE_VM) {
3894 add_uevent_var(env, "EVENT=create");
3895 kvm->userspace_pid = task_pid_nr(current);
3896 } else if (type == KVM_EVENT_DESTROY_VM) {
3897 add_uevent_var(env, "EVENT=destroy");
3899 add_uevent_var(env, "PID=%d", kvm->userspace_pid);
3901 if (kvm->debugfs_dentry) {
3902 char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL);
3905 tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX);
3907 add_uevent_var(env, "STATS_PATH=%s", tmp);
3911 /* no need for checks, since we are adding at most only 5 keys */
3912 env->envp[env->envp_idx++] = NULL;
3913 kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp);
3917 static void kvm_init_debug(void)
3919 struct kvm_stats_debugfs_item *p;
3921 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3923 kvm_debugfs_num_entries = 0;
3924 for (p = debugfs_entries; p->name; ++p, kvm_debugfs_num_entries++) {
3925 debugfs_create_file(p->name, 0644, kvm_debugfs_dir,
3926 (void *)(long)p->offset,
3927 stat_fops[p->kind]);
3931 static int kvm_suspend(void)
3933 if (kvm_usage_count)
3934 hardware_disable_nolock(NULL);
3938 static void kvm_resume(void)
3940 if (kvm_usage_count) {
3941 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3942 hardware_enable_nolock(NULL);
3946 static struct syscore_ops kvm_syscore_ops = {
3947 .suspend = kvm_suspend,
3948 .resume = kvm_resume,
3952 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3954 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3957 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3959 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3961 if (vcpu->preempted)
3962 vcpu->preempted = false;
3964 kvm_arch_sched_in(vcpu, cpu);
3966 kvm_arch_vcpu_load(vcpu, cpu);
3969 static void kvm_sched_out(struct preempt_notifier *pn,
3970 struct task_struct *next)
3972 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3974 if (current->state == TASK_RUNNING)
3975 vcpu->preempted = true;
3976 kvm_arch_vcpu_put(vcpu);
3979 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3980 struct module *module)
3985 r = kvm_arch_init(opaque);
3990 * kvm_arch_init makes sure there's at most one caller
3991 * for architectures that support multiple implementations,
3992 * like intel and amd on x86.
3993 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3994 * conflicts in case kvm is already setup for another implementation.
3996 r = kvm_irqfd_init();
4000 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
4005 r = kvm_arch_hardware_setup();
4009 for_each_online_cpu(cpu) {
4010 smp_call_function_single(cpu,
4011 kvm_arch_check_processor_compat,
4017 r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting",
4018 kvm_starting_cpu, kvm_dying_cpu);
4021 register_reboot_notifier(&kvm_reboot_notifier);
4023 /* A kmem cache lets us meet the alignment requirements of fx_save. */
4025 vcpu_align = __alignof__(struct kvm_vcpu);
4027 kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align,
4029 offsetof(struct kvm_vcpu, arch),
4030 sizeof_field(struct kvm_vcpu, arch),
4032 if (!kvm_vcpu_cache) {
4037 r = kvm_async_pf_init();
4041 kvm_chardev_ops.owner = module;
4042 kvm_vm_fops.owner = module;
4043 kvm_vcpu_fops.owner = module;
4045 r = misc_register(&kvm_dev);
4047 pr_err("kvm: misc device register failed\n");
4051 register_syscore_ops(&kvm_syscore_ops);
4053 kvm_preempt_ops.sched_in = kvm_sched_in;
4054 kvm_preempt_ops.sched_out = kvm_sched_out;
4058 r = kvm_vfio_ops_init();
4064 kvm_async_pf_deinit();
4066 kmem_cache_destroy(kvm_vcpu_cache);
4068 unregister_reboot_notifier(&kvm_reboot_notifier);
4069 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4072 kvm_arch_hardware_unsetup();
4074 free_cpumask_var(cpus_hardware_enabled);
4082 EXPORT_SYMBOL_GPL(kvm_init);
4086 debugfs_remove_recursive(kvm_debugfs_dir);
4087 misc_deregister(&kvm_dev);
4088 kmem_cache_destroy(kvm_vcpu_cache);
4089 kvm_async_pf_deinit();
4090 unregister_syscore_ops(&kvm_syscore_ops);
4091 unregister_reboot_notifier(&kvm_reboot_notifier);
4092 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4093 on_each_cpu(hardware_disable_nolock, NULL, 1);
4094 kvm_arch_hardware_unsetup();
4097 free_cpumask_var(cpus_hardware_enabled);
4098 kvm_vfio_ops_exit();
4100 EXPORT_SYMBOL_GPL(kvm_exit);