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>
54 #include <linux/kthread.h>
56 #include <asm/processor.h>
58 #include <asm/ioctl.h>
59 #include <linux/uaccess.h>
60 #include <asm/pgtable.h>
62 #include "coalesced_mmio.h"
66 #define CREATE_TRACE_POINTS
67 #include <trace/events/kvm.h>
69 /* Worst case buffer size needed for holding an integer. */
70 #define ITOA_MAX_LEN 12
72 MODULE_AUTHOR("Qumranet");
73 MODULE_LICENSE("GPL");
75 /* Architectures should define their poll value according to the halt latency */
76 unsigned int halt_poll_ns = KVM_HALT_POLL_NS_DEFAULT;
77 module_param(halt_poll_ns, uint, 0644);
78 EXPORT_SYMBOL_GPL(halt_poll_ns);
80 /* Default doubles per-vcpu halt_poll_ns. */
81 unsigned int halt_poll_ns_grow = 2;
82 module_param(halt_poll_ns_grow, uint, 0644);
83 EXPORT_SYMBOL_GPL(halt_poll_ns_grow);
85 /* Default resets per-vcpu halt_poll_ns . */
86 unsigned int halt_poll_ns_shrink;
87 module_param(halt_poll_ns_shrink, uint, 0644);
88 EXPORT_SYMBOL_GPL(halt_poll_ns_shrink);
93 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
96 DEFINE_MUTEX(kvm_lock);
97 static DEFINE_RAW_SPINLOCK(kvm_count_lock);
100 static cpumask_var_t cpus_hardware_enabled;
101 static int kvm_usage_count;
102 static atomic_t hardware_enable_failed;
104 struct kmem_cache *kvm_vcpu_cache;
105 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
107 static __read_mostly struct preempt_ops kvm_preempt_ops;
109 struct dentry *kvm_debugfs_dir;
110 EXPORT_SYMBOL_GPL(kvm_debugfs_dir);
112 static int kvm_debugfs_num_entries;
113 static const struct file_operations *stat_fops_per_vm[];
115 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
117 #ifdef CONFIG_KVM_COMPAT
118 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
121 static int hardware_enable_all(void);
122 static void hardware_disable_all(void);
124 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
126 static void kvm_release_pfn_dirty(kvm_pfn_t pfn);
127 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot, gfn_t gfn);
129 __visible bool kvm_rebooting;
130 EXPORT_SYMBOL_GPL(kvm_rebooting);
132 static bool largepages_enabled = true;
134 #define KVM_EVENT_CREATE_VM 0
135 #define KVM_EVENT_DESTROY_VM 1
136 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm);
137 static unsigned long long kvm_createvm_count;
138 static unsigned long long kvm_active_vms;
140 __weak void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
141 unsigned long start, unsigned long end)
145 bool kvm_is_zone_device_pfn(kvm_pfn_t pfn)
148 * The metadata used by is_zone_device_page() to determine whether or
149 * not a page is ZONE_DEVICE is guaranteed to be valid if and only if
150 * the device has been pinned, e.g. by get_user_pages(). WARN if the
151 * page_count() is zero to help detect bad usage of this helper.
153 if (!pfn_valid(pfn) || WARN_ON_ONCE(!page_count(pfn_to_page(pfn))))
156 return is_zone_device_page(pfn_to_page(pfn));
159 bool kvm_is_reserved_pfn(kvm_pfn_t pfn)
162 * ZONE_DEVICE pages currently set PG_reserved, but from a refcounting
163 * perspective they are "normal" pages, albeit with slightly different
167 return PageReserved(pfn_to_page(pfn)) &&
169 !kvm_is_zone_device_pfn(pfn);
175 * Switches to specified vcpu, until a matching vcpu_put()
177 int vcpu_load(struct kvm_vcpu *vcpu)
181 if (mutex_lock_killable(&vcpu->mutex))
184 preempt_notifier_register(&vcpu->preempt_notifier);
185 kvm_arch_vcpu_load(vcpu, cpu);
189 EXPORT_SYMBOL_GPL(vcpu_load);
191 void vcpu_put(struct kvm_vcpu *vcpu)
194 kvm_arch_vcpu_put(vcpu);
195 preempt_notifier_unregister(&vcpu->preempt_notifier);
197 mutex_unlock(&vcpu->mutex);
199 EXPORT_SYMBOL_GPL(vcpu_put);
201 /* TODO: merge with kvm_arch_vcpu_should_kick */
202 static bool kvm_request_needs_ipi(struct kvm_vcpu *vcpu, unsigned req)
204 int mode = kvm_vcpu_exiting_guest_mode(vcpu);
207 * We need to wait for the VCPU to reenable interrupts and get out of
208 * READING_SHADOW_PAGE_TABLES mode.
210 if (req & KVM_REQUEST_WAIT)
211 return mode != OUTSIDE_GUEST_MODE;
214 * Need to kick a running VCPU, but otherwise there is nothing to do.
216 return mode == IN_GUEST_MODE;
219 static void ack_flush(void *_completed)
223 static inline bool kvm_kick_many_cpus(const struct cpumask *cpus, bool wait)
226 cpus = cpu_online_mask;
228 if (cpumask_empty(cpus))
231 smp_call_function_many(cpus, ack_flush, NULL, wait);
235 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
240 struct kvm_vcpu *vcpu;
242 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
245 kvm_for_each_vcpu(i, vcpu, kvm) {
246 kvm_make_request(req, vcpu);
249 if (!(req & KVM_REQUEST_NO_WAKEUP) && kvm_vcpu_wake_up(vcpu))
252 if (cpus != NULL && cpu != -1 && cpu != me &&
253 kvm_request_needs_ipi(vcpu, req))
254 __cpumask_set_cpu(cpu, cpus);
256 called = kvm_kick_many_cpus(cpus, !!(req & KVM_REQUEST_WAIT));
258 free_cpumask_var(cpus);
262 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
263 void kvm_flush_remote_tlbs(struct kvm *kvm)
266 * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in
267 * kvm_make_all_cpus_request.
269 long dirty_count = smp_load_acquire(&kvm->tlbs_dirty);
272 * We want to publish modifications to the page tables before reading
273 * mode. Pairs with a memory barrier in arch-specific code.
274 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
275 * and smp_mb in walk_shadow_page_lockless_begin/end.
276 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
278 * There is already an smp_mb__after_atomic() before
279 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
282 if (kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
283 ++kvm->stat.remote_tlb_flush;
284 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
286 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
289 void kvm_reload_remote_mmus(struct kvm *kvm)
291 kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
294 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
299 mutex_init(&vcpu->mutex);
304 init_swait_queue_head(&vcpu->wq);
305 kvm_async_pf_vcpu_init(vcpu);
308 INIT_LIST_HEAD(&vcpu->blocked_vcpu_list);
310 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
315 vcpu->run = page_address(page);
317 kvm_vcpu_set_in_spin_loop(vcpu, false);
318 kvm_vcpu_set_dy_eligible(vcpu, false);
319 vcpu->preempted = false;
321 r = kvm_arch_vcpu_init(vcpu);
327 free_page((unsigned long)vcpu->run);
331 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
333 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
336 * no need for rcu_read_lock as VCPU_RUN is the only place that
337 * will change the vcpu->pid pointer and on uninit all file
338 * descriptors are already gone.
340 put_pid(rcu_dereference_protected(vcpu->pid, 1));
341 kvm_arch_vcpu_uninit(vcpu);
342 free_page((unsigned long)vcpu->run);
344 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
346 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
347 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
349 return container_of(mn, struct kvm, mmu_notifier);
352 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
353 struct mm_struct *mm,
354 unsigned long address,
357 struct kvm *kvm = mmu_notifier_to_kvm(mn);
360 idx = srcu_read_lock(&kvm->srcu);
361 spin_lock(&kvm->mmu_lock);
362 kvm->mmu_notifier_seq++;
363 kvm_set_spte_hva(kvm, address, pte);
364 spin_unlock(&kvm->mmu_lock);
365 srcu_read_unlock(&kvm->srcu, idx);
368 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
369 struct mm_struct *mm,
373 struct kvm *kvm = mmu_notifier_to_kvm(mn);
374 int need_tlb_flush = 0, idx;
376 idx = srcu_read_lock(&kvm->srcu);
377 spin_lock(&kvm->mmu_lock);
379 * The count increase must become visible at unlock time as no
380 * spte can be established without taking the mmu_lock and
381 * count is also read inside the mmu_lock critical section.
383 kvm->mmu_notifier_count++;
384 need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
385 /* we've to flush the tlb before the pages can be freed */
386 if (need_tlb_flush || kvm->tlbs_dirty)
387 kvm_flush_remote_tlbs(kvm);
389 spin_unlock(&kvm->mmu_lock);
391 kvm_arch_mmu_notifier_invalidate_range(kvm, start, end);
393 srcu_read_unlock(&kvm->srcu, idx);
396 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
397 struct mm_struct *mm,
401 struct kvm *kvm = mmu_notifier_to_kvm(mn);
403 spin_lock(&kvm->mmu_lock);
405 * This sequence increase will notify the kvm page fault that
406 * the page that is going to be mapped in the spte could have
409 kvm->mmu_notifier_seq++;
412 * The above sequence increase must be visible before the
413 * below count decrease, which is ensured by the smp_wmb above
414 * in conjunction with the smp_rmb in mmu_notifier_retry().
416 kvm->mmu_notifier_count--;
417 spin_unlock(&kvm->mmu_lock);
419 BUG_ON(kvm->mmu_notifier_count < 0);
422 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
423 struct mm_struct *mm,
427 struct kvm *kvm = mmu_notifier_to_kvm(mn);
430 idx = srcu_read_lock(&kvm->srcu);
431 spin_lock(&kvm->mmu_lock);
433 young = kvm_age_hva(kvm, start, end);
435 kvm_flush_remote_tlbs(kvm);
437 spin_unlock(&kvm->mmu_lock);
438 srcu_read_unlock(&kvm->srcu, idx);
443 static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
444 struct mm_struct *mm,
448 struct kvm *kvm = mmu_notifier_to_kvm(mn);
451 idx = srcu_read_lock(&kvm->srcu);
452 spin_lock(&kvm->mmu_lock);
454 * Even though we do not flush TLB, this will still adversely
455 * affect performance on pre-Haswell Intel EPT, where there is
456 * no EPT Access Bit to clear so that we have to tear down EPT
457 * tables instead. If we find this unacceptable, we can always
458 * add a parameter to kvm_age_hva so that it effectively doesn't
459 * do anything on clear_young.
461 * Also note that currently we never issue secondary TLB flushes
462 * from clear_young, leaving this job up to the regular system
463 * cadence. If we find this inaccurate, we might come up with a
464 * more sophisticated heuristic later.
466 young = kvm_age_hva(kvm, start, end);
467 spin_unlock(&kvm->mmu_lock);
468 srcu_read_unlock(&kvm->srcu, idx);
473 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
474 struct mm_struct *mm,
475 unsigned long address)
477 struct kvm *kvm = mmu_notifier_to_kvm(mn);
480 idx = srcu_read_lock(&kvm->srcu);
481 spin_lock(&kvm->mmu_lock);
482 young = kvm_test_age_hva(kvm, address);
483 spin_unlock(&kvm->mmu_lock);
484 srcu_read_unlock(&kvm->srcu, idx);
489 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
490 struct mm_struct *mm)
492 struct kvm *kvm = mmu_notifier_to_kvm(mn);
495 idx = srcu_read_lock(&kvm->srcu);
496 kvm_arch_flush_shadow_all(kvm);
497 srcu_read_unlock(&kvm->srcu, idx);
500 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
501 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
502 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
503 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
504 .clear_young = kvm_mmu_notifier_clear_young,
505 .test_young = kvm_mmu_notifier_test_young,
506 .change_pte = kvm_mmu_notifier_change_pte,
507 .release = kvm_mmu_notifier_release,
510 static int kvm_init_mmu_notifier(struct kvm *kvm)
512 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
513 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
516 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
518 static int kvm_init_mmu_notifier(struct kvm *kvm)
523 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
525 static struct kvm_memslots *kvm_alloc_memslots(void)
528 struct kvm_memslots *slots;
530 slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
534 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
535 slots->id_to_index[i] = slots->memslots[i].id = i;
540 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
542 if (!memslot->dirty_bitmap)
545 kvfree(memslot->dirty_bitmap);
546 memslot->dirty_bitmap = NULL;
550 * Free any memory in @free but not in @dont.
552 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
553 struct kvm_memory_slot *dont)
555 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
556 kvm_destroy_dirty_bitmap(free);
558 kvm_arch_free_memslot(kvm, free, dont);
563 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
565 struct kvm_memory_slot *memslot;
570 kvm_for_each_memslot(memslot, slots)
571 kvm_free_memslot(kvm, memslot, NULL);
576 static void kvm_destroy_vm_debugfs(struct kvm *kvm)
580 if (!kvm->debugfs_dentry)
583 debugfs_remove_recursive(kvm->debugfs_dentry);
585 if (kvm->debugfs_stat_data) {
586 for (i = 0; i < kvm_debugfs_num_entries; i++)
587 kfree(kvm->debugfs_stat_data[i]);
588 kfree(kvm->debugfs_stat_data);
592 static int kvm_create_vm_debugfs(struct kvm *kvm, int fd)
594 char dir_name[ITOA_MAX_LEN * 2];
595 struct kvm_stat_data *stat_data;
596 struct kvm_stats_debugfs_item *p;
598 if (!debugfs_initialized())
601 snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd);
602 kvm->debugfs_dentry = debugfs_create_dir(dir_name,
604 if (!kvm->debugfs_dentry)
607 kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries,
608 sizeof(*kvm->debugfs_stat_data),
610 if (!kvm->debugfs_stat_data)
613 for (p = debugfs_entries; p->name; p++) {
614 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL);
618 stat_data->kvm = kvm;
619 stat_data->offset = p->offset;
620 stat_data->mode = p->mode ? p->mode : 0644;
621 kvm->debugfs_stat_data[p - debugfs_entries] = stat_data;
622 if (!debugfs_create_file(p->name, stat_data->mode,
625 stat_fops_per_vm[p->kind]))
632 * Called after the VM is otherwise initialized, but just before adding it to
635 int __weak kvm_arch_post_init_vm(struct kvm *kvm)
641 * Called just after removing the VM from the vm_list, but before doing any
644 void __weak kvm_arch_pre_destroy_vm(struct kvm *kvm)
648 static struct kvm *kvm_create_vm(unsigned long type)
651 struct kvm *kvm = kvm_arch_alloc_vm();
654 return ERR_PTR(-ENOMEM);
656 spin_lock_init(&kvm->mmu_lock);
658 kvm->mm = current->mm;
659 kvm_eventfd_init(kvm);
660 mutex_init(&kvm->lock);
661 mutex_init(&kvm->irq_lock);
662 mutex_init(&kvm->slots_lock);
663 refcount_set(&kvm->users_count, 1);
664 INIT_LIST_HEAD(&kvm->devices);
666 r = kvm_arch_init_vm(kvm, type);
668 goto out_err_no_disable;
670 r = hardware_enable_all();
672 goto out_err_no_disable;
674 #ifdef CONFIG_HAVE_KVM_IRQFD
675 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
678 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
681 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
682 struct kvm_memslots *slots = kvm_alloc_memslots();
684 goto out_err_no_srcu;
686 * Generations must be different for each address space.
687 * Init kvm generation close to the maximum to easily test the
688 * code of handling generation number wrap-around.
690 slots->generation = i * 2 - 150;
691 rcu_assign_pointer(kvm->memslots[i], slots);
694 if (init_srcu_struct(&kvm->srcu))
695 goto out_err_no_srcu;
696 if (init_srcu_struct(&kvm->irq_srcu))
697 goto out_err_no_irq_srcu;
698 for (i = 0; i < KVM_NR_BUSES; i++) {
699 rcu_assign_pointer(kvm->buses[i],
700 kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL));
702 goto out_err_no_mmu_notifier;
705 r = kvm_init_mmu_notifier(kvm);
707 goto out_err_no_mmu_notifier;
709 r = kvm_arch_post_init_vm(kvm);
713 mutex_lock(&kvm_lock);
714 list_add(&kvm->vm_list, &vm_list);
715 mutex_unlock(&kvm_lock);
717 preempt_notifier_inc();
722 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
723 if (kvm->mmu_notifier.ops)
724 mmu_notifier_unregister(&kvm->mmu_notifier, current->mm);
726 out_err_no_mmu_notifier:
727 cleanup_srcu_struct(&kvm->irq_srcu);
729 cleanup_srcu_struct(&kvm->srcu);
731 hardware_disable_all();
733 refcount_set(&kvm->users_count, 0);
734 for (i = 0; i < KVM_NR_BUSES; i++)
735 kfree(kvm_get_bus(kvm, i));
736 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
737 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
738 kvm_arch_free_vm(kvm);
743 static void kvm_destroy_devices(struct kvm *kvm)
745 struct kvm_device *dev, *tmp;
748 * We do not need to take the kvm->lock here, because nobody else
749 * has a reference to the struct kvm at this point and therefore
750 * cannot access the devices list anyhow.
752 list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
753 list_del(&dev->vm_node);
754 dev->ops->destroy(dev);
758 static void kvm_destroy_vm(struct kvm *kvm)
761 struct mm_struct *mm = kvm->mm;
763 kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm);
764 kvm_destroy_vm_debugfs(kvm);
765 kvm_arch_sync_events(kvm);
766 mutex_lock(&kvm_lock);
767 list_del(&kvm->vm_list);
768 mutex_unlock(&kvm_lock);
769 kvm_arch_pre_destroy_vm(kvm);
771 kvm_free_irq_routing(kvm);
772 for (i = 0; i < KVM_NR_BUSES; i++) {
773 struct kvm_io_bus *bus = kvm_get_bus(kvm, i);
776 kvm_io_bus_destroy(bus);
777 kvm->buses[i] = NULL;
779 kvm_coalesced_mmio_free(kvm);
780 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
781 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
783 kvm_arch_flush_shadow_all(kvm);
785 kvm_arch_destroy_vm(kvm);
786 kvm_destroy_devices(kvm);
787 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
788 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
789 cleanup_srcu_struct(&kvm->irq_srcu);
790 cleanup_srcu_struct(&kvm->srcu);
791 kvm_arch_free_vm(kvm);
792 preempt_notifier_dec();
793 hardware_disable_all();
797 void kvm_get_kvm(struct kvm *kvm)
799 refcount_inc(&kvm->users_count);
801 EXPORT_SYMBOL_GPL(kvm_get_kvm);
803 void kvm_put_kvm(struct kvm *kvm)
805 if (refcount_dec_and_test(&kvm->users_count))
808 EXPORT_SYMBOL_GPL(kvm_put_kvm);
811 static int kvm_vm_release(struct inode *inode, struct file *filp)
813 struct kvm *kvm = filp->private_data;
815 kvm_irqfd_release(kvm);
822 * Allocation size is twice as large as the actual dirty bitmap size.
823 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
825 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
827 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
829 memslot->dirty_bitmap = kvzalloc(dirty_bytes, GFP_KERNEL);
830 if (!memslot->dirty_bitmap)
837 * Insert memslot and re-sort memslots based on their GFN,
838 * so binary search could be used to lookup GFN.
839 * Sorting algorithm takes advantage of having initially
840 * sorted array and known changed memslot position.
842 static void update_memslots(struct kvm_memslots *slots,
843 struct kvm_memory_slot *new)
846 int i = slots->id_to_index[id];
847 struct kvm_memory_slot *mslots = slots->memslots;
849 WARN_ON(mslots[i].id != id);
851 WARN_ON(!mslots[i].npages);
852 if (mslots[i].npages)
855 if (!mslots[i].npages)
859 while (i < KVM_MEM_SLOTS_NUM - 1 &&
860 new->base_gfn <= mslots[i + 1].base_gfn) {
861 if (!mslots[i + 1].npages)
863 mslots[i] = mslots[i + 1];
864 slots->id_to_index[mslots[i].id] = i;
869 * The ">=" is needed when creating a slot with base_gfn == 0,
870 * so that it moves before all those with base_gfn == npages == 0.
872 * On the other hand, if new->npages is zero, the above loop has
873 * already left i pointing to the beginning of the empty part of
874 * mslots, and the ">=" would move the hole backwards in this
875 * case---which is wrong. So skip the loop when deleting a slot.
879 new->base_gfn >= mslots[i - 1].base_gfn) {
880 mslots[i] = mslots[i - 1];
881 slots->id_to_index[mslots[i].id] = i;
885 WARN_ON_ONCE(i != slots->used_slots);
888 slots->id_to_index[mslots[i].id] = i;
891 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
893 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
895 #ifdef __KVM_HAVE_READONLY_MEM
896 valid_flags |= KVM_MEM_READONLY;
899 if (mem->flags & ~valid_flags)
905 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
906 int as_id, struct kvm_memslots *slots)
908 struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id);
912 * Set the low bit in the generation, which disables SPTE caching
913 * until the end of synchronize_srcu_expedited.
915 WARN_ON(old_memslots->generation & 1);
916 slots->generation = old_memslots->generation + 1;
918 rcu_assign_pointer(kvm->memslots[as_id], slots);
919 synchronize_srcu_expedited(&kvm->srcu);
922 * Increment the new memslot generation a second time. This prevents
923 * vm exits that race with memslot updates from caching a memslot
924 * generation that will (potentially) be valid forever.
926 * Generations must be unique even across address spaces. We do not need
927 * a global counter for that, instead the generation space is evenly split
928 * across address spaces. For example, with two address spaces, address
929 * space 0 will use generations 0, 4, 8, ... while * address space 1 will
930 * use generations 2, 6, 10, 14, ...
932 gen = slots->generation + KVM_ADDRESS_SPACE_NUM * 2 - 1;
934 kvm_arch_memslots_updated(kvm, gen);
936 slots->generation = gen;
942 * Allocate some memory and give it an address in the guest physical address
945 * Discontiguous memory is allowed, mostly for framebuffers.
947 * Must be called holding kvm->slots_lock for write.
949 int __kvm_set_memory_region(struct kvm *kvm,
950 const struct kvm_userspace_memory_region *mem)
954 unsigned long npages;
955 struct kvm_memory_slot *slot;
956 struct kvm_memory_slot old, new;
957 struct kvm_memslots *slots = NULL, *old_memslots;
959 enum kvm_mr_change change;
961 r = check_memory_region_flags(mem);
966 as_id = mem->slot >> 16;
969 /* General sanity checks */
970 if (mem->memory_size & (PAGE_SIZE - 1))
972 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
974 /* We can read the guest memory with __xxx_user() later on. */
975 if ((id < KVM_USER_MEM_SLOTS) &&
976 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
977 !access_ok(VERIFY_WRITE,
978 (void __user *)(unsigned long)mem->userspace_addr,
981 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
983 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
986 slot = id_to_memslot(__kvm_memslots(kvm, as_id), id);
987 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
988 npages = mem->memory_size >> PAGE_SHIFT;
990 if (npages > KVM_MEM_MAX_NR_PAGES)
996 new.base_gfn = base_gfn;
998 new.flags = mem->flags;
1002 change = KVM_MR_CREATE;
1003 else { /* Modify an existing slot. */
1004 if ((mem->userspace_addr != old.userspace_addr) ||
1005 (npages != old.npages) ||
1006 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
1009 if (base_gfn != old.base_gfn)
1010 change = KVM_MR_MOVE;
1011 else if (new.flags != old.flags)
1012 change = KVM_MR_FLAGS_ONLY;
1013 else { /* Nothing to change. */
1022 change = KVM_MR_DELETE;
1027 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
1028 /* Check for overlaps */
1030 kvm_for_each_memslot(slot, __kvm_memslots(kvm, as_id)) {
1033 if (!((base_gfn + npages <= slot->base_gfn) ||
1034 (base_gfn >= slot->base_gfn + slot->npages)))
1039 /* Free page dirty bitmap if unneeded */
1040 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
1041 new.dirty_bitmap = NULL;
1044 if (change == KVM_MR_CREATE) {
1045 new.userspace_addr = mem->userspace_addr;
1047 if (kvm_arch_create_memslot(kvm, &new, npages))
1051 /* Allocate page dirty bitmap if needed */
1052 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
1053 if (kvm_create_dirty_bitmap(&new) < 0)
1057 slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
1060 memcpy(slots, __kvm_memslots(kvm, as_id), sizeof(struct kvm_memslots));
1062 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
1063 slot = id_to_memslot(slots, id);
1064 slot->flags |= KVM_MEMSLOT_INVALID;
1066 old_memslots = install_new_memslots(kvm, as_id, slots);
1068 /* From this point no new shadow pages pointing to a deleted,
1069 * or moved, memslot will be created.
1071 * validation of sp->gfn happens in:
1072 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1073 * - kvm_is_visible_gfn (mmu_check_roots)
1075 kvm_arch_flush_shadow_memslot(kvm, slot);
1078 * We can re-use the old_memslots from above, the only difference
1079 * from the currently installed memslots is the invalid flag. This
1080 * will get overwritten by update_memslots anyway.
1082 slots = old_memslots;
1085 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
1089 /* actual memory is freed via old in kvm_free_memslot below */
1090 if (change == KVM_MR_DELETE) {
1091 new.dirty_bitmap = NULL;
1092 memset(&new.arch, 0, sizeof(new.arch));
1095 update_memslots(slots, &new);
1096 old_memslots = install_new_memslots(kvm, as_id, slots);
1098 kvm_arch_commit_memory_region(kvm, mem, &old, &new, change);
1100 kvm_free_memslot(kvm, &old, &new);
1101 kvfree(old_memslots);
1107 kvm_free_memslot(kvm, &new, &old);
1111 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1113 int kvm_set_memory_region(struct kvm *kvm,
1114 const struct kvm_userspace_memory_region *mem)
1118 mutex_lock(&kvm->slots_lock);
1119 r = __kvm_set_memory_region(kvm, mem);
1120 mutex_unlock(&kvm->slots_lock);
1123 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1125 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1126 struct kvm_userspace_memory_region *mem)
1128 if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1131 return kvm_set_memory_region(kvm, mem);
1134 int kvm_get_dirty_log(struct kvm *kvm,
1135 struct kvm_dirty_log *log, int *is_dirty)
1137 struct kvm_memslots *slots;
1138 struct kvm_memory_slot *memslot;
1141 unsigned long any = 0;
1143 as_id = log->slot >> 16;
1144 id = (u16)log->slot;
1145 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1148 slots = __kvm_memslots(kvm, as_id);
1149 memslot = id_to_memslot(slots, id);
1150 if (!memslot->dirty_bitmap)
1153 n = kvm_dirty_bitmap_bytes(memslot);
1155 for (i = 0; !any && i < n/sizeof(long); ++i)
1156 any = memslot->dirty_bitmap[i];
1158 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1165 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1167 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1169 * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages
1170 * are dirty write protect them for next write.
1171 * @kvm: pointer to kvm instance
1172 * @log: slot id and address to which we copy the log
1173 * @is_dirty: flag set if any page is dirty
1175 * We need to keep it in mind that VCPU threads can write to the bitmap
1176 * concurrently. So, to avoid losing track of dirty pages we keep the
1179 * 1. Take a snapshot of the bit and clear it if needed.
1180 * 2. Write protect the corresponding page.
1181 * 3. Copy the snapshot to the userspace.
1182 * 4. Upon return caller flushes TLB's if needed.
1184 * Between 2 and 4, the guest may write to the page using the remaining TLB
1185 * entry. This is not a problem because the page is reported dirty using
1186 * the snapshot taken before and step 4 ensures that writes done after
1187 * exiting to userspace will be logged for the next call.
1190 int kvm_get_dirty_log_protect(struct kvm *kvm,
1191 struct kvm_dirty_log *log, bool *is_dirty)
1193 struct kvm_memslots *slots;
1194 struct kvm_memory_slot *memslot;
1197 unsigned long *dirty_bitmap;
1198 unsigned long *dirty_bitmap_buffer;
1200 as_id = log->slot >> 16;
1201 id = (u16)log->slot;
1202 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1205 slots = __kvm_memslots(kvm, as_id);
1206 memslot = id_to_memslot(slots, id);
1208 dirty_bitmap = memslot->dirty_bitmap;
1212 n = kvm_dirty_bitmap_bytes(memslot);
1214 dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long);
1215 memset(dirty_bitmap_buffer, 0, n);
1217 spin_lock(&kvm->mmu_lock);
1219 for (i = 0; i < n / sizeof(long); i++) {
1223 if (!dirty_bitmap[i])
1228 mask = xchg(&dirty_bitmap[i], 0);
1229 dirty_bitmap_buffer[i] = mask;
1232 offset = i * BITS_PER_LONG;
1233 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1238 spin_unlock(&kvm->mmu_lock);
1239 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1243 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect);
1246 bool kvm_largepages_enabled(void)
1248 return largepages_enabled;
1251 void kvm_disable_largepages(void)
1253 largepages_enabled = false;
1255 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1257 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1259 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1261 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1263 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
1265 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn);
1268 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1270 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1272 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1273 memslot->flags & KVM_MEMSLOT_INVALID)
1278 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1280 unsigned long kvm_host_page_size(struct kvm_vcpu *vcpu, gfn_t gfn)
1282 struct vm_area_struct *vma;
1283 unsigned long addr, size;
1287 addr = kvm_vcpu_gfn_to_hva_prot(vcpu, gfn, NULL);
1288 if (kvm_is_error_hva(addr))
1291 down_read(¤t->mm->mmap_sem);
1292 vma = find_vma(current->mm, addr);
1296 size = vma_kernel_pagesize(vma);
1299 up_read(¤t->mm->mmap_sem);
1304 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1306 return slot->flags & KVM_MEM_READONLY;
1309 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1310 gfn_t *nr_pages, bool write)
1312 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1313 return KVM_HVA_ERR_BAD;
1315 if (memslot_is_readonly(slot) && write)
1316 return KVM_HVA_ERR_RO_BAD;
1319 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1321 return __gfn_to_hva_memslot(slot, gfn);
1324 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1327 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1330 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1333 return gfn_to_hva_many(slot, gfn, NULL);
1335 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1337 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1339 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1341 EXPORT_SYMBOL_GPL(gfn_to_hva);
1343 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
1345 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
1347 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
1350 * If writable is set to false, the hva returned by this function is only
1351 * allowed to be read.
1353 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1354 gfn_t gfn, bool *writable)
1356 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1358 if (!kvm_is_error_hva(hva) && writable)
1359 *writable = !memslot_is_readonly(slot);
1364 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1366 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1368 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1371 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
1373 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1375 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1378 static int get_user_page_nowait(unsigned long start, int write,
1381 int flags = FOLL_NOWAIT | FOLL_HWPOISON;
1384 flags |= FOLL_WRITE;
1386 return get_user_pages(start, 1, flags, page, NULL);
1389 static inline int check_user_page_hwpoison(unsigned long addr)
1391 int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
1393 rc = get_user_pages(addr, 1, flags, NULL, NULL);
1394 return rc == -EHWPOISON;
1398 * The atomic path to get the writable pfn which will be stored in @pfn,
1399 * true indicates success, otherwise false is returned.
1401 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1402 bool write_fault, bool *writable, kvm_pfn_t *pfn)
1404 struct page *page[1];
1407 if (!(async || atomic))
1411 * Fast pin a writable pfn only if it is a write fault request
1412 * or the caller allows to map a writable pfn for a read fault
1415 if (!(write_fault || writable))
1418 npages = __get_user_pages_fast(addr, 1, 1, page);
1420 *pfn = page_to_pfn(page[0]);
1431 * The slow path to get the pfn of the specified host virtual address,
1432 * 1 indicates success, -errno is returned if error is detected.
1434 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1435 bool *writable, kvm_pfn_t *pfn)
1437 struct page *page[1];
1443 *writable = write_fault;
1446 down_read(¤t->mm->mmap_sem);
1447 npages = get_user_page_nowait(addr, write_fault, page);
1448 up_read(¤t->mm->mmap_sem);
1450 unsigned int flags = FOLL_HWPOISON;
1453 flags |= FOLL_WRITE;
1455 npages = get_user_pages_unlocked(addr, 1, page, flags);
1460 /* map read fault as writable if possible */
1461 if (unlikely(!write_fault) && writable) {
1462 struct page *wpage[1];
1464 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1473 *pfn = page_to_pfn(page[0]);
1477 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1479 if (unlikely(!(vma->vm_flags & VM_READ)))
1482 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1488 static int kvm_try_get_pfn(kvm_pfn_t pfn)
1490 if (kvm_is_reserved_pfn(pfn))
1492 return get_page_unless_zero(pfn_to_page(pfn));
1495 static int hva_to_pfn_remapped(struct vm_area_struct *vma,
1496 unsigned long addr, bool *async,
1497 bool write_fault, bool *writable,
1505 r = follow_pte_pmd(vma->vm_mm, addr, NULL, NULL, &ptep, NULL, &ptl);
1508 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
1509 * not call the fault handler, so do it here.
1511 bool unlocked = false;
1512 r = fixup_user_fault(current, current->mm, addr,
1513 (write_fault ? FAULT_FLAG_WRITE : 0),
1520 r = follow_pte_pmd(vma->vm_mm, addr, NULL, NULL, &ptep, NULL, &ptl);
1525 if (write_fault && !pte_write(*ptep)) {
1526 pfn = KVM_PFN_ERR_RO_FAULT;
1531 *writable = pte_write(*ptep);
1532 pfn = pte_pfn(*ptep);
1535 * Get a reference here because callers of *hva_to_pfn* and
1536 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
1537 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
1538 * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will
1539 * simply do nothing for reserved pfns.
1541 * Whoever called remap_pfn_range is also going to call e.g.
1542 * unmap_mapping_range before the underlying pages are freed,
1543 * causing a call to our MMU notifier.
1545 * Certain IO or PFNMAP mappings can be backed with valid
1546 * struct pages, but be allocated without refcounting e.g.,
1547 * tail pages of non-compound higher order allocations, which
1548 * would then underflow the refcount when the caller does the
1549 * required put_page. Don't allow those pages here.
1551 if (!kvm_try_get_pfn(pfn))
1555 pte_unmap_unlock(ptep, ptl);
1562 * Pin guest page in memory and return its pfn.
1563 * @addr: host virtual address which maps memory to the guest
1564 * @atomic: whether this function can sleep
1565 * @async: whether this function need to wait IO complete if the
1566 * host page is not in the memory
1567 * @write_fault: whether we should get a writable host page
1568 * @writable: whether it allows to map a writable host page for !@write_fault
1570 * The function will map a writable host page for these two cases:
1571 * 1): @write_fault = true
1572 * 2): @write_fault = false && @writable, @writable will tell the caller
1573 * whether the mapping is writable.
1575 static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1576 bool write_fault, bool *writable)
1578 struct vm_area_struct *vma;
1582 /* we can do it either atomically or asynchronously, not both */
1583 BUG_ON(atomic && async);
1585 if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1589 return KVM_PFN_ERR_FAULT;
1591 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1595 down_read(¤t->mm->mmap_sem);
1596 if (npages == -EHWPOISON ||
1597 (!async && check_user_page_hwpoison(addr))) {
1598 pfn = KVM_PFN_ERR_HWPOISON;
1603 vma = find_vma_intersection(current->mm, addr, addr + 1);
1606 pfn = KVM_PFN_ERR_FAULT;
1607 else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
1608 r = hva_to_pfn_remapped(vma, addr, async, write_fault, writable, &pfn);
1612 pfn = KVM_PFN_ERR_FAULT;
1614 if (async && vma_is_valid(vma, write_fault))
1616 pfn = KVM_PFN_ERR_FAULT;
1619 up_read(¤t->mm->mmap_sem);
1623 kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn,
1624 bool atomic, bool *async, bool write_fault,
1627 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1629 if (addr == KVM_HVA_ERR_RO_BAD) {
1632 return KVM_PFN_ERR_RO_FAULT;
1635 if (kvm_is_error_hva(addr)) {
1638 return KVM_PFN_NOSLOT;
1641 /* Do not map writable pfn in the readonly memslot. */
1642 if (writable && memslot_is_readonly(slot)) {
1647 return hva_to_pfn(addr, atomic, async, write_fault,
1650 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
1652 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1655 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
1656 write_fault, writable);
1658 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1660 kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1662 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1664 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
1666 kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1668 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1670 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1672 kvm_pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1674 return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm, gfn), gfn);
1676 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1678 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
1680 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1682 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
1684 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1686 return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
1688 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1690 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
1692 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1694 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
1696 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1697 struct page **pages, int nr_pages)
1702 addr = gfn_to_hva_many(slot, gfn, &entry);
1703 if (kvm_is_error_hva(addr))
1706 if (entry < nr_pages)
1709 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1711 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1713 static struct page *kvm_pfn_to_page(kvm_pfn_t pfn)
1715 if (is_error_noslot_pfn(pfn))
1716 return KVM_ERR_PTR_BAD_PAGE;
1718 if (kvm_is_reserved_pfn(pfn)) {
1720 return KVM_ERR_PTR_BAD_PAGE;
1723 return pfn_to_page(pfn);
1726 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1730 pfn = gfn_to_pfn(kvm, gfn);
1732 return kvm_pfn_to_page(pfn);
1734 EXPORT_SYMBOL_GPL(gfn_to_page);
1736 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
1740 pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
1742 return kvm_pfn_to_page(pfn);
1744 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
1746 void kvm_release_page_clean(struct page *page)
1748 WARN_ON(is_error_page(page));
1750 kvm_release_pfn_clean(page_to_pfn(page));
1752 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1754 void kvm_release_pfn_clean(kvm_pfn_t pfn)
1756 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
1757 put_page(pfn_to_page(pfn));
1759 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1761 void kvm_release_page_dirty(struct page *page)
1763 WARN_ON(is_error_page(page));
1765 kvm_release_pfn_dirty(page_to_pfn(page));
1767 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1769 static void kvm_release_pfn_dirty(kvm_pfn_t pfn)
1771 kvm_set_pfn_dirty(pfn);
1772 kvm_release_pfn_clean(pfn);
1775 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
1777 if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn)) {
1778 struct page *page = pfn_to_page(pfn);
1780 if (!PageReserved(page))
1784 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1786 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
1788 if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn))
1789 mark_page_accessed(pfn_to_page(pfn));
1791 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1793 void kvm_get_pfn(kvm_pfn_t pfn)
1795 if (!kvm_is_reserved_pfn(pfn))
1796 get_page(pfn_to_page(pfn));
1798 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1800 static int next_segment(unsigned long len, int offset)
1802 if (len > PAGE_SIZE - offset)
1803 return PAGE_SIZE - offset;
1808 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
1809 void *data, int offset, int len)
1814 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1815 if (kvm_is_error_hva(addr))
1817 r = __copy_from_user(data, (void __user *)addr + offset, len);
1823 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1826 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1828 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1830 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1832 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
1833 int offset, int len)
1835 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1837 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1839 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
1841 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1843 gfn_t gfn = gpa >> PAGE_SHIFT;
1845 int offset = offset_in_page(gpa);
1848 while ((seg = next_segment(len, offset)) != 0) {
1849 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1859 EXPORT_SYMBOL_GPL(kvm_read_guest);
1861 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
1863 gfn_t gfn = gpa >> PAGE_SHIFT;
1865 int offset = offset_in_page(gpa);
1868 while ((seg = next_segment(len, offset)) != 0) {
1869 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
1879 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
1881 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1882 void *data, int offset, unsigned long len)
1887 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1888 if (kvm_is_error_hva(addr))
1890 pagefault_disable();
1891 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1898 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1901 gfn_t gfn = gpa >> PAGE_SHIFT;
1902 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1903 int offset = offset_in_page(gpa);
1905 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1907 EXPORT_SYMBOL_GPL(kvm_read_guest_atomic);
1909 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
1910 void *data, unsigned long len)
1912 gfn_t gfn = gpa >> PAGE_SHIFT;
1913 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1914 int offset = offset_in_page(gpa);
1916 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1918 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
1920 static int __kvm_write_guest_page(struct kvm_memory_slot *memslot, gfn_t gfn,
1921 const void *data, int offset, int len)
1926 addr = gfn_to_hva_memslot(memslot, gfn);
1927 if (kvm_is_error_hva(addr))
1929 r = __copy_to_user((void __user *)addr + offset, data, len);
1932 mark_page_dirty_in_slot(memslot, gfn);
1936 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
1937 const void *data, int offset, int len)
1939 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1941 return __kvm_write_guest_page(slot, gfn, data, offset, len);
1943 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1945 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
1946 const void *data, int offset, int len)
1948 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1950 return __kvm_write_guest_page(slot, gfn, data, offset, len);
1952 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
1954 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1957 gfn_t gfn = gpa >> PAGE_SHIFT;
1959 int offset = offset_in_page(gpa);
1962 while ((seg = next_segment(len, offset)) != 0) {
1963 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1973 EXPORT_SYMBOL_GPL(kvm_write_guest);
1975 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
1978 gfn_t gfn = gpa >> PAGE_SHIFT;
1980 int offset = offset_in_page(gpa);
1983 while ((seg = next_segment(len, offset)) != 0) {
1984 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
1994 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
1996 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
1997 struct gfn_to_hva_cache *ghc,
1998 gpa_t gpa, unsigned long len)
2000 int offset = offset_in_page(gpa);
2001 gfn_t start_gfn = gpa >> PAGE_SHIFT;
2002 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
2003 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
2004 gfn_t nr_pages_avail;
2007 ghc->generation = slots->generation;
2009 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
2010 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, NULL);
2011 if (!kvm_is_error_hva(ghc->hva) && nr_pages_needed <= 1) {
2015 * If the requested region crosses two memslots, we still
2016 * verify that the entire region is valid here.
2018 while (start_gfn <= end_gfn) {
2020 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
2021 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
2023 if (kvm_is_error_hva(ghc->hva))
2025 start_gfn += nr_pages_avail;
2027 /* Use the slow path for cross page reads and writes. */
2028 ghc->memslot = NULL;
2033 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2034 gpa_t gpa, unsigned long len)
2036 struct kvm_memslots *slots = kvm_memslots(kvm);
2037 return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len);
2039 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
2041 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2042 void *data, unsigned int offset,
2045 struct kvm_memslots *slots = kvm_memslots(kvm);
2047 gpa_t gpa = ghc->gpa + offset;
2049 BUG_ON(len + offset > ghc->len);
2051 if (slots->generation != ghc->generation)
2052 __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len);
2054 if (kvm_is_error_hva(ghc->hva))
2057 if (unlikely(!ghc->memslot))
2058 return kvm_write_guest(kvm, gpa, data, len);
2060 r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
2063 mark_page_dirty_in_slot(ghc->memslot, gpa >> PAGE_SHIFT);
2067 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached);
2069 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2070 void *data, unsigned long len)
2072 return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len);
2074 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
2076 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2077 void *data, unsigned long len)
2079 struct kvm_memslots *slots = kvm_memslots(kvm);
2082 BUG_ON(len > ghc->len);
2084 if (slots->generation != ghc->generation)
2085 __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len);
2087 if (kvm_is_error_hva(ghc->hva))
2090 if (unlikely(!ghc->memslot))
2091 return kvm_read_guest(kvm, ghc->gpa, data, len);
2093 r = __copy_from_user(data, (void __user *)ghc->hva, len);
2099 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
2101 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
2103 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
2105 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
2107 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
2109 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
2111 gfn_t gfn = gpa >> PAGE_SHIFT;
2113 int offset = offset_in_page(gpa);
2116 while ((seg = next_segment(len, offset)) != 0) {
2117 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
2126 EXPORT_SYMBOL_GPL(kvm_clear_guest);
2128 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot,
2131 if (memslot && memslot->dirty_bitmap) {
2132 unsigned long rel_gfn = gfn - memslot->base_gfn;
2134 set_bit_le(rel_gfn, memslot->dirty_bitmap);
2138 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
2140 struct kvm_memory_slot *memslot;
2142 memslot = gfn_to_memslot(kvm, gfn);
2143 mark_page_dirty_in_slot(memslot, gfn);
2145 EXPORT_SYMBOL_GPL(mark_page_dirty);
2147 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
2149 struct kvm_memory_slot *memslot;
2151 memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2152 mark_page_dirty_in_slot(memslot, gfn);
2154 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
2156 void kvm_sigset_activate(struct kvm_vcpu *vcpu)
2158 if (!vcpu->sigset_active)
2162 * This does a lockless modification of ->real_blocked, which is fine
2163 * because, only current can change ->real_blocked and all readers of
2164 * ->real_blocked don't care as long ->real_blocked is always a subset
2167 sigprocmask(SIG_SETMASK, &vcpu->sigset, ¤t->real_blocked);
2170 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu)
2172 if (!vcpu->sigset_active)
2175 sigprocmask(SIG_SETMASK, ¤t->real_blocked, NULL);
2176 sigemptyset(¤t->real_blocked);
2179 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
2181 unsigned int old, val, grow;
2183 old = val = vcpu->halt_poll_ns;
2184 grow = READ_ONCE(halt_poll_ns_grow);
2186 if (val == 0 && grow)
2191 if (val > halt_poll_ns)
2194 vcpu->halt_poll_ns = val;
2195 trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
2198 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
2200 unsigned int old, val, shrink;
2202 old = val = vcpu->halt_poll_ns;
2203 shrink = READ_ONCE(halt_poll_ns_shrink);
2209 vcpu->halt_poll_ns = val;
2210 trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
2213 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
2215 if (kvm_arch_vcpu_runnable(vcpu)) {
2216 kvm_make_request(KVM_REQ_UNHALT, vcpu);
2219 if (kvm_cpu_has_pending_timer(vcpu))
2221 if (signal_pending(current))
2228 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2230 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
2233 DECLARE_SWAITQUEUE(wait);
2234 bool waited = false;
2237 start = cur = ktime_get();
2238 if (vcpu->halt_poll_ns) {
2239 ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns);
2241 ++vcpu->stat.halt_attempted_poll;
2244 * This sets KVM_REQ_UNHALT if an interrupt
2247 if (kvm_vcpu_check_block(vcpu) < 0) {
2248 ++vcpu->stat.halt_successful_poll;
2249 if (!vcpu_valid_wakeup(vcpu))
2250 ++vcpu->stat.halt_poll_invalid;
2254 } while (single_task_running() && ktime_before(cur, stop));
2257 kvm_arch_vcpu_blocking(vcpu);
2260 prepare_to_swait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
2262 if (kvm_vcpu_check_block(vcpu) < 0)
2269 finish_swait(&vcpu->wq, &wait);
2272 kvm_arch_vcpu_unblocking(vcpu);
2274 block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
2276 if (!vcpu_valid_wakeup(vcpu))
2277 shrink_halt_poll_ns(vcpu);
2278 else if (halt_poll_ns) {
2279 if (block_ns <= vcpu->halt_poll_ns)
2281 /* we had a long block, shrink polling */
2282 else if (vcpu->halt_poll_ns && block_ns > halt_poll_ns)
2283 shrink_halt_poll_ns(vcpu);
2284 /* we had a short halt and our poll time is too small */
2285 else if (vcpu->halt_poll_ns < halt_poll_ns &&
2286 block_ns < halt_poll_ns)
2287 grow_halt_poll_ns(vcpu);
2289 vcpu->halt_poll_ns = 0;
2291 trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu));
2292 kvm_arch_vcpu_block_finish(vcpu);
2294 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
2296 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
2298 struct swait_queue_head *wqp;
2300 wqp = kvm_arch_vcpu_wq(vcpu);
2301 if (swq_has_sleeper(wqp)) {
2303 ++vcpu->stat.halt_wakeup;
2309 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
2313 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2315 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
2318 int cpu = vcpu->cpu;
2320 if (kvm_vcpu_wake_up(vcpu))
2324 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
2325 if (kvm_arch_vcpu_should_kick(vcpu))
2326 smp_send_reschedule(cpu);
2329 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
2330 #endif /* !CONFIG_S390 */
2332 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
2335 struct task_struct *task = NULL;
2339 pid = rcu_dereference(target->pid);
2341 task = get_pid_task(pid, PIDTYPE_PID);
2345 ret = yield_to(task, 1);
2346 put_task_struct(task);
2350 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
2353 * Helper that checks whether a VCPU is eligible for directed yield.
2354 * Most eligible candidate to yield is decided by following heuristics:
2356 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2357 * (preempted lock holder), indicated by @in_spin_loop.
2358 * Set at the beiginning and cleared at the end of interception/PLE handler.
2360 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2361 * chance last time (mostly it has become eligible now since we have probably
2362 * yielded to lockholder in last iteration. This is done by toggling
2363 * @dy_eligible each time a VCPU checked for eligibility.)
2365 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2366 * to preempted lock-holder could result in wrong VCPU selection and CPU
2367 * burning. Giving priority for a potential lock-holder increases lock
2370 * Since algorithm is based on heuristics, accessing another VCPU data without
2371 * locking does not harm. It may result in trying to yield to same VCPU, fail
2372 * and continue with next VCPU and so on.
2374 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
2376 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2379 eligible = !vcpu->spin_loop.in_spin_loop ||
2380 vcpu->spin_loop.dy_eligible;
2382 if (vcpu->spin_loop.in_spin_loop)
2383 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
2392 * Unlike kvm_arch_vcpu_runnable, this function is called outside
2393 * a vcpu_load/vcpu_put pair. However, for most architectures
2394 * kvm_arch_vcpu_runnable does not require vcpu_load.
2396 bool __weak kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
2398 return kvm_arch_vcpu_runnable(vcpu);
2401 static bool vcpu_dy_runnable(struct kvm_vcpu *vcpu)
2403 if (kvm_arch_dy_runnable(vcpu))
2406 #ifdef CONFIG_KVM_ASYNC_PF
2407 if (!list_empty_careful(&vcpu->async_pf.done))
2414 void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode)
2416 struct kvm *kvm = me->kvm;
2417 struct kvm_vcpu *vcpu;
2418 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
2424 kvm_vcpu_set_in_spin_loop(me, true);
2426 * We boost the priority of a VCPU that is runnable but not
2427 * currently running, because it got preempted by something
2428 * else and called schedule in __vcpu_run. Hopefully that
2429 * VCPU is holding the lock that we need and will release it.
2430 * We approximate round-robin by starting at the last boosted VCPU.
2432 for (pass = 0; pass < 2 && !yielded && try; pass++) {
2433 kvm_for_each_vcpu(i, vcpu, kvm) {
2434 if (!pass && i <= last_boosted_vcpu) {
2435 i = last_boosted_vcpu;
2437 } else if (pass && i > last_boosted_vcpu)
2439 if (!ACCESS_ONCE(vcpu->preempted))
2443 if (swait_active(&vcpu->wq) && !vcpu_dy_runnable(vcpu))
2445 if (yield_to_kernel_mode && !kvm_arch_vcpu_in_kernel(vcpu))
2447 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
2450 yielded = kvm_vcpu_yield_to(vcpu);
2452 kvm->last_boosted_vcpu = i;
2454 } else if (yielded < 0) {
2461 kvm_vcpu_set_in_spin_loop(me, false);
2463 /* Ensure vcpu is not eligible during next spinloop */
2464 kvm_vcpu_set_dy_eligible(me, false);
2466 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
2468 static int kvm_vcpu_fault(struct vm_fault *vmf)
2470 struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data;
2473 if (vmf->pgoff == 0)
2474 page = virt_to_page(vcpu->run);
2476 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
2477 page = virt_to_page(vcpu->arch.pio_data);
2479 #ifdef CONFIG_KVM_MMIO
2480 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
2481 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
2484 return kvm_arch_vcpu_fault(vcpu, vmf);
2490 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
2491 .fault = kvm_vcpu_fault,
2494 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2496 vma->vm_ops = &kvm_vcpu_vm_ops;
2500 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2502 struct kvm_vcpu *vcpu = filp->private_data;
2504 debugfs_remove_recursive(vcpu->debugfs_dentry);
2505 kvm_put_kvm(vcpu->kvm);
2509 static struct file_operations kvm_vcpu_fops = {
2510 .release = kvm_vcpu_release,
2511 .unlocked_ioctl = kvm_vcpu_ioctl,
2512 #ifdef CONFIG_KVM_COMPAT
2513 .compat_ioctl = kvm_vcpu_compat_ioctl,
2515 .mmap = kvm_vcpu_mmap,
2516 .llseek = noop_llseek,
2520 * Allocates an inode for the vcpu.
2522 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2524 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
2527 static int kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
2529 char dir_name[ITOA_MAX_LEN * 2];
2532 if (!kvm_arch_has_vcpu_debugfs())
2535 if (!debugfs_initialized())
2538 snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
2539 vcpu->debugfs_dentry = debugfs_create_dir(dir_name,
2540 vcpu->kvm->debugfs_dentry);
2541 if (!vcpu->debugfs_dentry)
2544 ret = kvm_arch_create_vcpu_debugfs(vcpu);
2546 debugfs_remove_recursive(vcpu->debugfs_dentry);
2554 * Creates some virtual cpus. Good luck creating more than one.
2556 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
2559 struct kvm_vcpu *vcpu;
2561 if (id >= KVM_MAX_VCPU_ID)
2564 mutex_lock(&kvm->lock);
2565 if (kvm->created_vcpus == KVM_MAX_VCPUS) {
2566 mutex_unlock(&kvm->lock);
2570 kvm->created_vcpus++;
2571 mutex_unlock(&kvm->lock);
2573 vcpu = kvm_arch_vcpu_create(kvm, id);
2576 goto vcpu_decrement;
2579 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2581 r = kvm_arch_vcpu_setup(vcpu);
2585 r = kvm_create_vcpu_debugfs(vcpu);
2589 mutex_lock(&kvm->lock);
2590 if (kvm_get_vcpu_by_id(kvm, id)) {
2592 goto unlock_vcpu_destroy;
2595 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
2597 /* Now it's all set up, let userspace reach it */
2599 r = create_vcpu_fd(vcpu);
2602 goto unlock_vcpu_destroy;
2605 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
2608 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
2609 * before kvm->online_vcpu's incremented value.
2612 atomic_inc(&kvm->online_vcpus);
2614 mutex_unlock(&kvm->lock);
2615 kvm_arch_vcpu_postcreate(vcpu);
2618 unlock_vcpu_destroy:
2619 mutex_unlock(&kvm->lock);
2620 debugfs_remove_recursive(vcpu->debugfs_dentry);
2622 kvm_arch_vcpu_destroy(vcpu);
2624 mutex_lock(&kvm->lock);
2625 kvm->created_vcpus--;
2626 mutex_unlock(&kvm->lock);
2630 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2633 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2634 vcpu->sigset_active = 1;
2635 vcpu->sigset = *sigset;
2637 vcpu->sigset_active = 0;
2641 static long kvm_vcpu_ioctl(struct file *filp,
2642 unsigned int ioctl, unsigned long arg)
2644 struct kvm_vcpu *vcpu = filp->private_data;
2645 void __user *argp = (void __user *)arg;
2647 struct kvm_fpu *fpu = NULL;
2648 struct kvm_sregs *kvm_sregs = NULL;
2650 if (vcpu->kvm->mm != current->mm)
2653 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2656 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
2658 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
2659 * so vcpu_load() would break it.
2661 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_S390_IRQ || ioctl == KVM_INTERRUPT)
2662 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2666 r = vcpu_load(vcpu);
2675 oldpid = rcu_access_pointer(vcpu->pid);
2676 if (unlikely(oldpid != current->pids[PIDTYPE_PID].pid)) {
2677 /* The thread running this VCPU changed. */
2678 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
2680 rcu_assign_pointer(vcpu->pid, newpid);
2685 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2686 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2689 case KVM_GET_REGS: {
2690 struct kvm_regs *kvm_regs;
2693 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2696 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2700 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2707 case KVM_SET_REGS: {
2708 struct kvm_regs *kvm_regs;
2711 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2712 if (IS_ERR(kvm_regs)) {
2713 r = PTR_ERR(kvm_regs);
2716 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2720 case KVM_GET_SREGS: {
2721 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2725 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2729 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2734 case KVM_SET_SREGS: {
2735 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2736 if (IS_ERR(kvm_sregs)) {
2737 r = PTR_ERR(kvm_sregs);
2741 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2744 case KVM_GET_MP_STATE: {
2745 struct kvm_mp_state mp_state;
2747 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2751 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
2756 case KVM_SET_MP_STATE: {
2757 struct kvm_mp_state mp_state;
2760 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
2762 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2765 case KVM_TRANSLATE: {
2766 struct kvm_translation tr;
2769 if (copy_from_user(&tr, argp, sizeof(tr)))
2771 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2775 if (copy_to_user(argp, &tr, sizeof(tr)))
2780 case KVM_SET_GUEST_DEBUG: {
2781 struct kvm_guest_debug dbg;
2784 if (copy_from_user(&dbg, argp, sizeof(dbg)))
2786 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2789 case KVM_SET_SIGNAL_MASK: {
2790 struct kvm_signal_mask __user *sigmask_arg = argp;
2791 struct kvm_signal_mask kvm_sigmask;
2792 sigset_t sigset, *p;
2797 if (copy_from_user(&kvm_sigmask, argp,
2798 sizeof(kvm_sigmask)))
2801 if (kvm_sigmask.len != sizeof(sigset))
2804 if (copy_from_user(&sigset, sigmask_arg->sigset,
2809 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2813 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2817 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2821 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2827 fpu = memdup_user(argp, sizeof(*fpu));
2833 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2837 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2846 #ifdef CONFIG_KVM_COMPAT
2847 static long kvm_vcpu_compat_ioctl(struct file *filp,
2848 unsigned int ioctl, unsigned long arg)
2850 struct kvm_vcpu *vcpu = filp->private_data;
2851 void __user *argp = compat_ptr(arg);
2854 if (vcpu->kvm->mm != current->mm)
2858 case KVM_SET_SIGNAL_MASK: {
2859 struct kvm_signal_mask __user *sigmask_arg = argp;
2860 struct kvm_signal_mask kvm_sigmask;
2861 compat_sigset_t csigset;
2866 if (copy_from_user(&kvm_sigmask, argp,
2867 sizeof(kvm_sigmask)))
2870 if (kvm_sigmask.len != sizeof(csigset))
2873 if (copy_from_user(&csigset, sigmask_arg->sigset,
2876 sigset_from_compat(&sigset, &csigset);
2877 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2879 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2883 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2891 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2892 int (*accessor)(struct kvm_device *dev,
2893 struct kvm_device_attr *attr),
2896 struct kvm_device_attr attr;
2901 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2904 return accessor(dev, &attr);
2907 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2910 struct kvm_device *dev = filp->private_data;
2912 if (dev->kvm->mm != current->mm)
2916 case KVM_SET_DEVICE_ATTR:
2917 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2918 case KVM_GET_DEVICE_ATTR:
2919 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2920 case KVM_HAS_DEVICE_ATTR:
2921 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2923 if (dev->ops->ioctl)
2924 return dev->ops->ioctl(dev, ioctl, arg);
2930 static int kvm_device_release(struct inode *inode, struct file *filp)
2932 struct kvm_device *dev = filp->private_data;
2933 struct kvm *kvm = dev->kvm;
2939 static const struct file_operations kvm_device_fops = {
2940 .unlocked_ioctl = kvm_device_ioctl,
2941 #ifdef CONFIG_KVM_COMPAT
2942 .compat_ioctl = kvm_device_ioctl,
2944 .release = kvm_device_release,
2947 struct kvm_device *kvm_device_from_filp(struct file *filp)
2949 if (filp->f_op != &kvm_device_fops)
2952 return filp->private_data;
2955 static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
2956 #ifdef CONFIG_KVM_MPIC
2957 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
2958 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
2962 int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
2964 if (type >= ARRAY_SIZE(kvm_device_ops_table))
2967 if (kvm_device_ops_table[type] != NULL)
2970 kvm_device_ops_table[type] = ops;
2974 void kvm_unregister_device_ops(u32 type)
2976 if (kvm_device_ops_table[type] != NULL)
2977 kvm_device_ops_table[type] = NULL;
2980 static int kvm_ioctl_create_device(struct kvm *kvm,
2981 struct kvm_create_device *cd)
2983 struct kvm_device_ops *ops = NULL;
2984 struct kvm_device *dev;
2985 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2989 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
2992 type = array_index_nospec(cd->type, ARRAY_SIZE(kvm_device_ops_table));
2993 ops = kvm_device_ops_table[type];
3000 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
3007 mutex_lock(&kvm->lock);
3008 ret = ops->create(dev, type);
3010 mutex_unlock(&kvm->lock);
3014 list_add(&dev->vm_node, &kvm->devices);
3015 mutex_unlock(&kvm->lock);
3021 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
3024 mutex_lock(&kvm->lock);
3025 list_del(&dev->vm_node);
3026 mutex_unlock(&kvm->lock);
3035 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
3038 case KVM_CAP_USER_MEMORY:
3039 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
3040 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
3041 case KVM_CAP_INTERNAL_ERROR_DATA:
3042 #ifdef CONFIG_HAVE_KVM_MSI
3043 case KVM_CAP_SIGNAL_MSI:
3045 #ifdef CONFIG_HAVE_KVM_IRQFD
3047 case KVM_CAP_IRQFD_RESAMPLE:
3049 case KVM_CAP_IOEVENTFD_ANY_LENGTH:
3050 case KVM_CAP_CHECK_EXTENSION_VM:
3052 #ifdef CONFIG_KVM_MMIO
3053 case KVM_CAP_COALESCED_MMIO:
3054 return KVM_COALESCED_MMIO_PAGE_OFFSET;
3056 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3057 case KVM_CAP_IRQ_ROUTING:
3058 return KVM_MAX_IRQ_ROUTES;
3060 #if KVM_ADDRESS_SPACE_NUM > 1
3061 case KVM_CAP_MULTI_ADDRESS_SPACE:
3062 return KVM_ADDRESS_SPACE_NUM;
3067 return kvm_vm_ioctl_check_extension(kvm, arg);
3070 static long kvm_vm_ioctl(struct file *filp,
3071 unsigned int ioctl, unsigned long arg)
3073 struct kvm *kvm = filp->private_data;
3074 void __user *argp = (void __user *)arg;
3077 if (kvm->mm != current->mm)
3080 case KVM_CREATE_VCPU:
3081 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
3083 case KVM_SET_USER_MEMORY_REGION: {
3084 struct kvm_userspace_memory_region kvm_userspace_mem;
3087 if (copy_from_user(&kvm_userspace_mem, argp,
3088 sizeof(kvm_userspace_mem)))
3091 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
3094 case KVM_GET_DIRTY_LOG: {
3095 struct kvm_dirty_log log;
3098 if (copy_from_user(&log, argp, sizeof(log)))
3100 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3103 #ifdef CONFIG_KVM_MMIO
3104 case KVM_REGISTER_COALESCED_MMIO: {
3105 struct kvm_coalesced_mmio_zone zone;
3108 if (copy_from_user(&zone, argp, sizeof(zone)))
3110 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
3113 case KVM_UNREGISTER_COALESCED_MMIO: {
3114 struct kvm_coalesced_mmio_zone zone;
3117 if (copy_from_user(&zone, argp, sizeof(zone)))
3119 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
3124 struct kvm_irqfd data;
3127 if (copy_from_user(&data, argp, sizeof(data)))
3129 r = kvm_irqfd(kvm, &data);
3132 case KVM_IOEVENTFD: {
3133 struct kvm_ioeventfd data;
3136 if (copy_from_user(&data, argp, sizeof(data)))
3138 r = kvm_ioeventfd(kvm, &data);
3141 #ifdef CONFIG_HAVE_KVM_MSI
3142 case KVM_SIGNAL_MSI: {
3146 if (copy_from_user(&msi, argp, sizeof(msi)))
3148 r = kvm_send_userspace_msi(kvm, &msi);
3152 #ifdef __KVM_HAVE_IRQ_LINE
3153 case KVM_IRQ_LINE_STATUS:
3154 case KVM_IRQ_LINE: {
3155 struct kvm_irq_level irq_event;
3158 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
3161 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
3162 ioctl == KVM_IRQ_LINE_STATUS);
3167 if (ioctl == KVM_IRQ_LINE_STATUS) {
3168 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
3176 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3177 case KVM_SET_GSI_ROUTING: {
3178 struct kvm_irq_routing routing;
3179 struct kvm_irq_routing __user *urouting;
3180 struct kvm_irq_routing_entry *entries = NULL;
3183 if (copy_from_user(&routing, argp, sizeof(routing)))
3186 if (!kvm_arch_can_set_irq_routing(kvm))
3188 if (routing.nr > KVM_MAX_IRQ_ROUTES)
3194 entries = vmalloc(routing.nr * sizeof(*entries));
3199 if (copy_from_user(entries, urouting->entries,
3200 routing.nr * sizeof(*entries)))
3201 goto out_free_irq_routing;
3203 r = kvm_set_irq_routing(kvm, entries, routing.nr,
3205 out_free_irq_routing:
3209 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
3210 case KVM_CREATE_DEVICE: {
3211 struct kvm_create_device cd;
3214 if (copy_from_user(&cd, argp, sizeof(cd)))
3217 r = kvm_ioctl_create_device(kvm, &cd);
3222 if (copy_to_user(argp, &cd, sizeof(cd)))
3228 case KVM_CHECK_EXTENSION:
3229 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
3232 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
3238 #ifdef CONFIG_KVM_COMPAT
3239 struct compat_kvm_dirty_log {
3243 compat_uptr_t dirty_bitmap; /* one bit per page */
3248 static long kvm_vm_compat_ioctl(struct file *filp,
3249 unsigned int ioctl, unsigned long arg)
3251 struct kvm *kvm = filp->private_data;
3254 if (kvm->mm != current->mm)
3257 case KVM_GET_DIRTY_LOG: {
3258 struct compat_kvm_dirty_log compat_log;
3259 struct kvm_dirty_log log;
3261 if (copy_from_user(&compat_log, (void __user *)arg,
3262 sizeof(compat_log)))
3264 log.slot = compat_log.slot;
3265 log.padding1 = compat_log.padding1;
3266 log.padding2 = compat_log.padding2;
3267 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
3269 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3273 r = kvm_vm_ioctl(filp, ioctl, arg);
3279 static struct file_operations kvm_vm_fops = {
3280 .release = kvm_vm_release,
3281 .unlocked_ioctl = kvm_vm_ioctl,
3282 #ifdef CONFIG_KVM_COMPAT
3283 .compat_ioctl = kvm_vm_compat_ioctl,
3285 .llseek = noop_llseek,
3288 static int kvm_dev_ioctl_create_vm(unsigned long type)
3294 kvm = kvm_create_vm(type);
3296 return PTR_ERR(kvm);
3297 #ifdef CONFIG_KVM_MMIO
3298 r = kvm_coalesced_mmio_init(kvm);
3304 r = get_unused_fd_flags(O_CLOEXEC);
3309 file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
3313 return PTR_ERR(file);
3317 * Don't call kvm_put_kvm anymore at this point; file->f_op is
3318 * already set, with ->release() being kvm_vm_release(). In error
3319 * cases it will be called by the final fput(file) and will take
3320 * care of doing kvm_put_kvm(kvm).
3322 if (kvm_create_vm_debugfs(kvm, r) < 0) {
3327 kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm);
3329 fd_install(r, file);
3333 static long kvm_dev_ioctl(struct file *filp,
3334 unsigned int ioctl, unsigned long arg)
3339 case KVM_GET_API_VERSION:
3342 r = KVM_API_VERSION;
3345 r = kvm_dev_ioctl_create_vm(arg);
3347 case KVM_CHECK_EXTENSION:
3348 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
3350 case KVM_GET_VCPU_MMAP_SIZE:
3353 r = PAGE_SIZE; /* struct kvm_run */
3355 r += PAGE_SIZE; /* pio data page */
3357 #ifdef CONFIG_KVM_MMIO
3358 r += PAGE_SIZE; /* coalesced mmio ring page */
3361 case KVM_TRACE_ENABLE:
3362 case KVM_TRACE_PAUSE:
3363 case KVM_TRACE_DISABLE:
3367 return kvm_arch_dev_ioctl(filp, ioctl, arg);
3373 static struct file_operations kvm_chardev_ops = {
3374 .unlocked_ioctl = kvm_dev_ioctl,
3375 .compat_ioctl = kvm_dev_ioctl,
3376 .llseek = noop_llseek,
3379 static struct miscdevice kvm_dev = {
3385 static void hardware_enable_nolock(void *junk)
3387 int cpu = raw_smp_processor_id();
3390 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
3393 cpumask_set_cpu(cpu, cpus_hardware_enabled);
3395 r = kvm_arch_hardware_enable();
3398 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3399 atomic_inc(&hardware_enable_failed);
3400 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
3404 static int kvm_starting_cpu(unsigned int cpu)
3406 raw_spin_lock(&kvm_count_lock);
3407 if (kvm_usage_count)
3408 hardware_enable_nolock(NULL);
3409 raw_spin_unlock(&kvm_count_lock);
3413 static void hardware_disable_nolock(void *junk)
3415 int cpu = raw_smp_processor_id();
3417 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
3419 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3420 kvm_arch_hardware_disable();
3423 static int kvm_dying_cpu(unsigned int cpu)
3425 raw_spin_lock(&kvm_count_lock);
3426 if (kvm_usage_count)
3427 hardware_disable_nolock(NULL);
3428 raw_spin_unlock(&kvm_count_lock);
3432 static void hardware_disable_all_nolock(void)
3434 BUG_ON(!kvm_usage_count);
3437 if (!kvm_usage_count)
3438 on_each_cpu(hardware_disable_nolock, NULL, 1);
3441 static void hardware_disable_all(void)
3443 raw_spin_lock(&kvm_count_lock);
3444 hardware_disable_all_nolock();
3445 raw_spin_unlock(&kvm_count_lock);
3448 static int hardware_enable_all(void)
3452 raw_spin_lock(&kvm_count_lock);
3455 if (kvm_usage_count == 1) {
3456 atomic_set(&hardware_enable_failed, 0);
3457 on_each_cpu(hardware_enable_nolock, NULL, 1);
3459 if (atomic_read(&hardware_enable_failed)) {
3460 hardware_disable_all_nolock();
3465 raw_spin_unlock(&kvm_count_lock);
3470 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3474 * Some (well, at least mine) BIOSes hang on reboot if
3477 * And Intel TXT required VMX off for all cpu when system shutdown.
3479 pr_info("kvm: exiting hardware virtualization\n");
3480 kvm_rebooting = true;
3481 on_each_cpu(hardware_disable_nolock, NULL, 1);
3485 static struct notifier_block kvm_reboot_notifier = {
3486 .notifier_call = kvm_reboot,
3490 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3494 for (i = 0; i < bus->dev_count; i++) {
3495 struct kvm_io_device *pos = bus->range[i].dev;
3497 kvm_iodevice_destructor(pos);
3502 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
3503 const struct kvm_io_range *r2)
3505 gpa_t addr1 = r1->addr;
3506 gpa_t addr2 = r2->addr;
3511 /* If r2->len == 0, match the exact address. If r2->len != 0,
3512 * accept any overlapping write. Any order is acceptable for
3513 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
3514 * we process all of them.
3527 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
3529 return kvm_io_bus_cmp(p1, p2);
3532 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
3533 gpa_t addr, int len)
3535 bus->range[bus->dev_count++] = (struct kvm_io_range) {
3541 sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
3542 kvm_io_bus_sort_cmp, NULL);
3547 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
3548 gpa_t addr, int len)
3550 struct kvm_io_range *range, key;
3553 key = (struct kvm_io_range) {
3558 range = bsearch(&key, bus->range, bus->dev_count,
3559 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
3563 off = range - bus->range;
3565 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
3571 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3572 struct kvm_io_range *range, const void *val)
3576 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3580 while (idx < bus->dev_count &&
3581 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3582 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
3591 /* kvm_io_bus_write - called under kvm->slots_lock */
3592 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3593 int len, const void *val)
3595 struct kvm_io_bus *bus;
3596 struct kvm_io_range range;
3599 range = (struct kvm_io_range) {
3604 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3607 r = __kvm_io_bus_write(vcpu, bus, &range, val);
3608 return r < 0 ? r : 0;
3611 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3612 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
3613 gpa_t addr, int len, const void *val, long cookie)
3615 struct kvm_io_bus *bus;
3616 struct kvm_io_range range;
3618 range = (struct kvm_io_range) {
3623 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3627 /* First try the device referenced by cookie. */
3628 if ((cookie >= 0) && (cookie < bus->dev_count) &&
3629 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
3630 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
3635 * cookie contained garbage; fall back to search and return the
3636 * correct cookie value.
3638 return __kvm_io_bus_write(vcpu, bus, &range, val);
3641 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3642 struct kvm_io_range *range, void *val)
3646 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3650 while (idx < bus->dev_count &&
3651 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3652 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
3660 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
3662 /* kvm_io_bus_read - called under kvm->slots_lock */
3663 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3666 struct kvm_io_bus *bus;
3667 struct kvm_io_range range;
3670 range = (struct kvm_io_range) {
3675 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3678 r = __kvm_io_bus_read(vcpu, bus, &range, val);
3679 return r < 0 ? r : 0;
3683 /* Caller must hold slots_lock. */
3684 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3685 int len, struct kvm_io_device *dev)
3687 struct kvm_io_bus *new_bus, *bus;
3689 bus = kvm_get_bus(kvm, bus_idx);
3693 /* exclude ioeventfd which is limited by maximum fd */
3694 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3697 new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count + 1) *
3698 sizeof(struct kvm_io_range)), GFP_KERNEL);
3701 memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
3702 sizeof(struct kvm_io_range)));
3703 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
3704 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3705 synchronize_srcu_expedited(&kvm->srcu);
3711 /* Caller must hold slots_lock. */
3712 void kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3713 struct kvm_io_device *dev)
3716 struct kvm_io_bus *new_bus, *bus;
3718 bus = kvm_get_bus(kvm, bus_idx);
3722 for (i = 0; i < bus->dev_count; i++)
3723 if (bus->range[i].dev == dev) {
3727 if (i == bus->dev_count)
3730 new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3731 sizeof(struct kvm_io_range)), GFP_KERNEL);
3733 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3734 new_bus->dev_count--;
3735 memcpy(new_bus->range + i, bus->range + i + 1,
3736 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3738 pr_err("kvm: failed to shrink bus, removing it completely\n");
3739 for (j = 0; j < bus->dev_count; j++) {
3742 kvm_iodevice_destructor(bus->range[j].dev);
3746 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3747 synchronize_srcu_expedited(&kvm->srcu);
3752 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3755 struct kvm_io_bus *bus;
3756 int dev_idx, srcu_idx;
3757 struct kvm_io_device *iodev = NULL;
3759 srcu_idx = srcu_read_lock(&kvm->srcu);
3761 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
3765 dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
3769 iodev = bus->range[dev_idx].dev;
3772 srcu_read_unlock(&kvm->srcu, srcu_idx);
3776 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
3778 static int kvm_debugfs_open(struct inode *inode, struct file *file,
3779 int (*get)(void *, u64 *), int (*set)(void *, u64),
3782 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
3785 /* The debugfs files are a reference to the kvm struct which
3786 * is still valid when kvm_destroy_vm is called.
3787 * To avoid the race between open and the removal of the debugfs
3788 * directory we test against the users count.
3790 if (!refcount_inc_not_zero(&stat_data->kvm->users_count))
3793 if (simple_attr_open(inode, file, get,
3794 stat_data->mode & S_IWUGO ? set : NULL,
3796 kvm_put_kvm(stat_data->kvm);
3803 static int kvm_debugfs_release(struct inode *inode, struct file *file)
3805 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
3808 simple_attr_release(inode, file);
3809 kvm_put_kvm(stat_data->kvm);
3814 static int vm_stat_get_per_vm(void *data, u64 *val)
3816 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3818 *val = *(ulong *)((void *)stat_data->kvm + stat_data->offset);
3823 static int vm_stat_clear_per_vm(void *data, u64 val)
3825 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3830 *(ulong *)((void *)stat_data->kvm + stat_data->offset) = 0;
3835 static int vm_stat_get_per_vm_open(struct inode *inode, struct file *file)
3837 __simple_attr_check_format("%llu\n", 0ull);
3838 return kvm_debugfs_open(inode, file, vm_stat_get_per_vm,
3839 vm_stat_clear_per_vm, "%llu\n");
3842 static const struct file_operations vm_stat_get_per_vm_fops = {
3843 .owner = THIS_MODULE,
3844 .open = vm_stat_get_per_vm_open,
3845 .release = kvm_debugfs_release,
3846 .read = simple_attr_read,
3847 .write = simple_attr_write,
3848 .llseek = no_llseek,
3851 static int vcpu_stat_get_per_vm(void *data, u64 *val)
3854 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3855 struct kvm_vcpu *vcpu;
3859 kvm_for_each_vcpu(i, vcpu, stat_data->kvm)
3860 *val += *(u64 *)((void *)vcpu + stat_data->offset);
3865 static int vcpu_stat_clear_per_vm(void *data, u64 val)
3868 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3869 struct kvm_vcpu *vcpu;
3874 kvm_for_each_vcpu(i, vcpu, stat_data->kvm)
3875 *(u64 *)((void *)vcpu + stat_data->offset) = 0;
3880 static int vcpu_stat_get_per_vm_open(struct inode *inode, struct file *file)
3882 __simple_attr_check_format("%llu\n", 0ull);
3883 return kvm_debugfs_open(inode, file, vcpu_stat_get_per_vm,
3884 vcpu_stat_clear_per_vm, "%llu\n");
3887 static const struct file_operations vcpu_stat_get_per_vm_fops = {
3888 .owner = THIS_MODULE,
3889 .open = vcpu_stat_get_per_vm_open,
3890 .release = kvm_debugfs_release,
3891 .read = simple_attr_read,
3892 .write = simple_attr_write,
3893 .llseek = no_llseek,
3896 static const struct file_operations *stat_fops_per_vm[] = {
3897 [KVM_STAT_VCPU] = &vcpu_stat_get_per_vm_fops,
3898 [KVM_STAT_VM] = &vm_stat_get_per_vm_fops,
3901 static int vm_stat_get(void *_offset, u64 *val)
3903 unsigned offset = (long)_offset;
3905 struct kvm_stat_data stat_tmp = {.offset = offset};
3909 mutex_lock(&kvm_lock);
3910 list_for_each_entry(kvm, &vm_list, vm_list) {
3912 vm_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
3915 mutex_unlock(&kvm_lock);
3919 static int vm_stat_clear(void *_offset, u64 val)
3921 unsigned offset = (long)_offset;
3923 struct kvm_stat_data stat_tmp = {.offset = offset};
3928 mutex_lock(&kvm_lock);
3929 list_for_each_entry(kvm, &vm_list, vm_list) {
3931 vm_stat_clear_per_vm((void *)&stat_tmp, 0);
3933 mutex_unlock(&kvm_lock);
3938 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n");
3940 static int vcpu_stat_get(void *_offset, u64 *val)
3942 unsigned offset = (long)_offset;
3944 struct kvm_stat_data stat_tmp = {.offset = offset};
3948 mutex_lock(&kvm_lock);
3949 list_for_each_entry(kvm, &vm_list, vm_list) {
3951 vcpu_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
3954 mutex_unlock(&kvm_lock);
3958 static int vcpu_stat_clear(void *_offset, u64 val)
3960 unsigned offset = (long)_offset;
3962 struct kvm_stat_data stat_tmp = {.offset = offset};
3967 mutex_lock(&kvm_lock);
3968 list_for_each_entry(kvm, &vm_list, vm_list) {
3970 vcpu_stat_clear_per_vm((void *)&stat_tmp, 0);
3972 mutex_unlock(&kvm_lock);
3977 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear,
3980 static const struct file_operations *stat_fops[] = {
3981 [KVM_STAT_VCPU] = &vcpu_stat_fops,
3982 [KVM_STAT_VM] = &vm_stat_fops,
3985 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm)
3987 struct kobj_uevent_env *env;
3988 unsigned long long created, active;
3990 if (!kvm_dev.this_device || !kvm)
3993 mutex_lock(&kvm_lock);
3994 if (type == KVM_EVENT_CREATE_VM) {
3995 kvm_createvm_count++;
3997 } else if (type == KVM_EVENT_DESTROY_VM) {
4000 created = kvm_createvm_count;
4001 active = kvm_active_vms;
4002 mutex_unlock(&kvm_lock);
4004 env = kzalloc(sizeof(*env), GFP_KERNEL);
4008 add_uevent_var(env, "CREATED=%llu", created);
4009 add_uevent_var(env, "COUNT=%llu", active);
4011 if (type == KVM_EVENT_CREATE_VM) {
4012 add_uevent_var(env, "EVENT=create");
4013 kvm->userspace_pid = task_pid_nr(current);
4014 } else if (type == KVM_EVENT_DESTROY_VM) {
4015 add_uevent_var(env, "EVENT=destroy");
4017 add_uevent_var(env, "PID=%d", kvm->userspace_pid);
4019 if (!IS_ERR_OR_NULL(kvm->debugfs_dentry)) {
4020 char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL);
4023 tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX);
4025 add_uevent_var(env, "STATS_PATH=%s", tmp);
4029 /* no need for checks, since we are adding at most only 5 keys */
4030 env->envp[env->envp_idx++] = NULL;
4031 kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp);
4035 static int kvm_init_debug(void)
4038 struct kvm_stats_debugfs_item *p;
4040 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
4041 if (kvm_debugfs_dir == NULL)
4044 kvm_debugfs_num_entries = 0;
4045 for (p = debugfs_entries; p->name; ++p, kvm_debugfs_num_entries++) {
4046 int mode = p->mode ? p->mode : 0644;
4047 if (!debugfs_create_file(p->name, mode, kvm_debugfs_dir,
4048 (void *)(long)p->offset,
4049 stat_fops[p->kind]))
4056 debugfs_remove_recursive(kvm_debugfs_dir);
4061 static int kvm_suspend(void)
4063 if (kvm_usage_count)
4064 hardware_disable_nolock(NULL);
4068 static void kvm_resume(void)
4070 if (kvm_usage_count) {
4071 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
4072 hardware_enable_nolock(NULL);
4076 static struct syscore_ops kvm_syscore_ops = {
4077 .suspend = kvm_suspend,
4078 .resume = kvm_resume,
4082 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
4084 return container_of(pn, struct kvm_vcpu, preempt_notifier);
4087 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
4089 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
4091 if (vcpu->preempted)
4092 vcpu->preempted = false;
4094 kvm_arch_sched_in(vcpu, cpu);
4096 kvm_arch_vcpu_load(vcpu, cpu);
4099 static void kvm_sched_out(struct preempt_notifier *pn,
4100 struct task_struct *next)
4102 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
4104 if (current->state == TASK_RUNNING)
4105 vcpu->preempted = true;
4106 kvm_arch_vcpu_put(vcpu);
4109 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
4110 struct module *module)
4115 r = kvm_arch_init(opaque);
4120 * kvm_arch_init makes sure there's at most one caller
4121 * for architectures that support multiple implementations,
4122 * like intel and amd on x86.
4123 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
4124 * conflicts in case kvm is already setup for another implementation.
4126 r = kvm_irqfd_init();
4130 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
4135 r = kvm_arch_hardware_setup();
4139 for_each_online_cpu(cpu) {
4140 smp_call_function_single(cpu,
4141 kvm_arch_check_processor_compat,
4147 r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting",
4148 kvm_starting_cpu, kvm_dying_cpu);
4151 register_reboot_notifier(&kvm_reboot_notifier);
4153 /* A kmem cache lets us meet the alignment requirements of fx_save. */
4155 vcpu_align = __alignof__(struct kvm_vcpu);
4156 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
4157 SLAB_ACCOUNT, NULL);
4158 if (!kvm_vcpu_cache) {
4163 r = kvm_async_pf_init();
4167 kvm_chardev_ops.owner = module;
4168 kvm_vm_fops.owner = module;
4169 kvm_vcpu_fops.owner = module;
4171 r = misc_register(&kvm_dev);
4173 pr_err("kvm: misc device register failed\n");
4177 register_syscore_ops(&kvm_syscore_ops);
4179 kvm_preempt_ops.sched_in = kvm_sched_in;
4180 kvm_preempt_ops.sched_out = kvm_sched_out;
4182 r = kvm_init_debug();
4184 pr_err("kvm: create debugfs files failed\n");
4188 r = kvm_vfio_ops_init();
4194 unregister_syscore_ops(&kvm_syscore_ops);
4195 misc_deregister(&kvm_dev);
4197 kvm_async_pf_deinit();
4199 kmem_cache_destroy(kvm_vcpu_cache);
4201 unregister_reboot_notifier(&kvm_reboot_notifier);
4202 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4205 kvm_arch_hardware_unsetup();
4207 free_cpumask_var(cpus_hardware_enabled);
4215 EXPORT_SYMBOL_GPL(kvm_init);
4219 debugfs_remove_recursive(kvm_debugfs_dir);
4220 misc_deregister(&kvm_dev);
4221 kmem_cache_destroy(kvm_vcpu_cache);
4222 kvm_async_pf_deinit();
4223 unregister_syscore_ops(&kvm_syscore_ops);
4224 unregister_reboot_notifier(&kvm_reboot_notifier);
4225 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4226 on_each_cpu(hardware_disable_nolock, NULL, 1);
4227 kvm_arch_hardware_unsetup();
4230 free_cpumask_var(cpus_hardware_enabled);
4231 kvm_vfio_ops_exit();
4233 EXPORT_SYMBOL_GPL(kvm_exit);
4235 struct kvm_vm_worker_thread_context {
4237 struct task_struct *parent;
4238 struct completion init_done;
4239 kvm_vm_thread_fn_t thread_fn;
4244 static int kvm_vm_worker_thread(void *context)
4247 * The init_context is allocated on the stack of the parent thread, so
4248 * we have to locally copy anything that is needed beyond initialization
4250 struct kvm_vm_worker_thread_context *init_context = context;
4251 struct kvm *kvm = init_context->kvm;
4252 kvm_vm_thread_fn_t thread_fn = init_context->thread_fn;
4253 uintptr_t data = init_context->data;
4256 err = kthread_park(current);
4257 /* kthread_park(current) is never supposed to return an error */
4262 err = cgroup_attach_task_all(init_context->parent, current);
4264 kvm_err("%s: cgroup_attach_task_all failed with err %d\n",
4269 set_user_nice(current, task_nice(init_context->parent));
4272 init_context->err = err;
4273 complete(&init_context->init_done);
4274 init_context = NULL;
4279 /* Wait to be woken up by the spawner before proceeding. */
4282 if (!kthread_should_stop())
4283 err = thread_fn(kvm, data);
4288 int kvm_vm_create_worker_thread(struct kvm *kvm, kvm_vm_thread_fn_t thread_fn,
4289 uintptr_t data, const char *name,
4290 struct task_struct **thread_ptr)
4292 struct kvm_vm_worker_thread_context init_context = {};
4293 struct task_struct *thread;
4296 init_context.kvm = kvm;
4297 init_context.parent = current;
4298 init_context.thread_fn = thread_fn;
4299 init_context.data = data;
4300 init_completion(&init_context.init_done);
4302 thread = kthread_run(kvm_vm_worker_thread, &init_context,
4303 "%s-%d", name, task_pid_nr(current));
4305 return PTR_ERR(thread);
4307 /* kthread_run is never supposed to return NULL */
4308 WARN_ON(thread == NULL);
4310 wait_for_completion(&init_context.init_done);
4312 if (!init_context.err)
4313 *thread_ptr = thread;
4315 return init_context.err;