2 * Kernel-based Virtual Machine driver for Linux
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
7 * Copyright (C) 2006 Qumranet, Inc.
8 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
11 * Avi Kivity <avi@qumranet.com>
12 * Yaniv Kamay <yaniv@qumranet.com>
14 * This work is licensed under the terms of the GNU GPL, version 2. See
15 * the COPYING file in the top-level directory.
19 #include <kvm/iodev.h>
21 #include <linux/kvm_host.h>
22 #include <linux/kvm.h>
23 #include <linux/module.h>
24 #include <linux/errno.h>
25 #include <linux/percpu.h>
27 #include <linux/miscdevice.h>
28 #include <linux/vmalloc.h>
29 #include <linux/reboot.h>
30 #include <linux/debugfs.h>
31 #include <linux/highmem.h>
32 #include <linux/file.h>
33 #include <linux/syscore_ops.h>
34 #include <linux/cpu.h>
35 #include <linux/sched/signal.h>
36 #include <linux/sched/mm.h>
37 #include <linux/sched/stat.h>
38 #include <linux/cpumask.h>
39 #include <linux/smp.h>
40 #include <linux/anon_inodes.h>
41 #include <linux/profile.h>
42 #include <linux/kvm_para.h>
43 #include <linux/pagemap.h>
44 #include <linux/mman.h>
45 #include <linux/swap.h>
46 #include <linux/bitops.h>
47 #include <linux/spinlock.h>
48 #include <linux/compat.h>
49 #include <linux/srcu.h>
50 #include <linux/hugetlb.h>
51 #include <linux/slab.h>
52 #include <linux/sort.h>
53 #include <linux/bsearch.h>
55 #include <asm/processor.h>
57 #include <asm/ioctl.h>
58 #include <linux/uaccess.h>
59 #include <asm/pgtable.h>
61 #include "coalesced_mmio.h"
65 #define CREATE_TRACE_POINTS
66 #include <trace/events/kvm.h>
68 /* Worst case buffer size needed for holding an integer. */
69 #define ITOA_MAX_LEN 12
71 MODULE_AUTHOR("Qumranet");
72 MODULE_LICENSE("GPL");
74 /* Architectures should define their poll value according to the halt latency */
75 unsigned int halt_poll_ns = KVM_HALT_POLL_NS_DEFAULT;
76 module_param(halt_poll_ns, uint, 0644);
77 EXPORT_SYMBOL_GPL(halt_poll_ns);
79 /* Default doubles per-vcpu halt_poll_ns. */
80 unsigned int halt_poll_ns_grow = 2;
81 module_param(halt_poll_ns_grow, uint, 0644);
82 EXPORT_SYMBOL_GPL(halt_poll_ns_grow);
84 /* Default resets per-vcpu halt_poll_ns . */
85 unsigned int halt_poll_ns_shrink;
86 module_param(halt_poll_ns_shrink, uint, 0644);
87 EXPORT_SYMBOL_GPL(halt_poll_ns_shrink);
92 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
95 DEFINE_SPINLOCK(kvm_lock);
96 static DEFINE_RAW_SPINLOCK(kvm_count_lock);
99 static cpumask_var_t cpus_hardware_enabled;
100 static int kvm_usage_count;
101 static atomic_t hardware_enable_failed;
103 struct kmem_cache *kvm_vcpu_cache;
104 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
106 static __read_mostly struct preempt_ops kvm_preempt_ops;
108 struct dentry *kvm_debugfs_dir;
109 EXPORT_SYMBOL_GPL(kvm_debugfs_dir);
111 static int kvm_debugfs_num_entries;
112 static const struct file_operations *stat_fops_per_vm[];
114 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
116 #ifdef CONFIG_KVM_COMPAT
117 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
120 static int hardware_enable_all(void);
121 static void hardware_disable_all(void);
123 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
125 static void kvm_release_pfn_dirty(kvm_pfn_t pfn);
126 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot, gfn_t gfn);
128 __visible bool kvm_rebooting;
129 EXPORT_SYMBOL_GPL(kvm_rebooting);
131 static bool largepages_enabled = true;
133 bool kvm_is_reserved_pfn(kvm_pfn_t pfn)
136 return PageReserved(pfn_to_page(pfn));
142 * Switches to specified vcpu, until a matching vcpu_put()
144 int vcpu_load(struct kvm_vcpu *vcpu)
148 if (mutex_lock_killable(&vcpu->mutex))
151 preempt_notifier_register(&vcpu->preempt_notifier);
152 kvm_arch_vcpu_load(vcpu, cpu);
156 EXPORT_SYMBOL_GPL(vcpu_load);
158 void vcpu_put(struct kvm_vcpu *vcpu)
161 kvm_arch_vcpu_put(vcpu);
162 preempt_notifier_unregister(&vcpu->preempt_notifier);
164 mutex_unlock(&vcpu->mutex);
166 EXPORT_SYMBOL_GPL(vcpu_put);
168 /* TODO: merge with kvm_arch_vcpu_should_kick */
169 static bool kvm_request_needs_ipi(struct kvm_vcpu *vcpu, unsigned req)
171 int mode = kvm_vcpu_exiting_guest_mode(vcpu);
174 * We need to wait for the VCPU to reenable interrupts and get out of
175 * READING_SHADOW_PAGE_TABLES mode.
177 if (req & KVM_REQUEST_WAIT)
178 return mode != OUTSIDE_GUEST_MODE;
181 * Need to kick a running VCPU, but otherwise there is nothing to do.
183 return mode == IN_GUEST_MODE;
186 static void ack_flush(void *_completed)
190 static inline bool kvm_kick_many_cpus(const struct cpumask *cpus, bool wait)
193 cpus = cpu_online_mask;
195 if (cpumask_empty(cpus))
198 smp_call_function_many(cpus, ack_flush, NULL, wait);
202 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
207 struct kvm_vcpu *vcpu;
209 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
212 kvm_for_each_vcpu(i, vcpu, kvm) {
213 kvm_make_request(req, vcpu);
216 if (!(req & KVM_REQUEST_NO_WAKEUP) && kvm_vcpu_wake_up(vcpu))
219 if (cpus != NULL && cpu != -1 && cpu != me &&
220 kvm_request_needs_ipi(vcpu, req))
221 __cpumask_set_cpu(cpu, cpus);
223 called = kvm_kick_many_cpus(cpus, !!(req & KVM_REQUEST_WAIT));
225 free_cpumask_var(cpus);
229 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
230 void kvm_flush_remote_tlbs(struct kvm *kvm)
233 * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in
234 * kvm_make_all_cpus_request.
236 long dirty_count = smp_load_acquire(&kvm->tlbs_dirty);
239 * We want to publish modifications to the page tables before reading
240 * mode. Pairs with a memory barrier in arch-specific code.
241 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
242 * and smp_mb in walk_shadow_page_lockless_begin/end.
243 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
245 * There is already an smp_mb__after_atomic() before
246 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
249 if (kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
250 ++kvm->stat.remote_tlb_flush;
251 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
253 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
256 void kvm_reload_remote_mmus(struct kvm *kvm)
258 kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
261 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
266 mutex_init(&vcpu->mutex);
271 init_swait_queue_head(&vcpu->wq);
272 kvm_async_pf_vcpu_init(vcpu);
275 INIT_LIST_HEAD(&vcpu->blocked_vcpu_list);
277 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
282 vcpu->run = page_address(page);
284 kvm_vcpu_set_in_spin_loop(vcpu, false);
285 kvm_vcpu_set_dy_eligible(vcpu, false);
286 vcpu->preempted = false;
288 r = kvm_arch_vcpu_init(vcpu);
294 free_page((unsigned long)vcpu->run);
298 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
300 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
303 * no need for rcu_read_lock as VCPU_RUN is the only place that
304 * will change the vcpu->pid pointer and on uninit all file
305 * descriptors are already gone.
307 put_pid(rcu_dereference_protected(vcpu->pid, 1));
308 kvm_arch_vcpu_uninit(vcpu);
309 free_page((unsigned long)vcpu->run);
311 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
313 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
314 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
316 return container_of(mn, struct kvm, mmu_notifier);
319 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
320 struct mm_struct *mm,
321 unsigned long address)
323 struct kvm *kvm = mmu_notifier_to_kvm(mn);
324 int need_tlb_flush, idx;
327 * When ->invalidate_page runs, the linux pte has been zapped
328 * already but the page is still allocated until
329 * ->invalidate_page returns. So if we increase the sequence
330 * here the kvm page fault will notice if the spte can't be
331 * established because the page is going to be freed. If
332 * instead the kvm page fault establishes the spte before
333 * ->invalidate_page runs, kvm_unmap_hva will release it
336 * The sequence increase only need to be seen at spin_unlock
337 * time, and not at spin_lock time.
339 * Increasing the sequence after the spin_unlock would be
340 * unsafe because the kvm page fault could then establish the
341 * pte after kvm_unmap_hva returned, without noticing the page
342 * is going to be freed.
344 idx = srcu_read_lock(&kvm->srcu);
345 spin_lock(&kvm->mmu_lock);
347 kvm->mmu_notifier_seq++;
348 need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
349 /* we've to flush the tlb before the pages can be freed */
351 kvm_flush_remote_tlbs(kvm);
353 spin_unlock(&kvm->mmu_lock);
355 kvm_arch_mmu_notifier_invalidate_page(kvm, address);
357 srcu_read_unlock(&kvm->srcu, idx);
360 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
361 struct mm_struct *mm,
362 unsigned long address,
365 struct kvm *kvm = mmu_notifier_to_kvm(mn);
368 idx = srcu_read_lock(&kvm->srcu);
369 spin_lock(&kvm->mmu_lock);
370 kvm->mmu_notifier_seq++;
371 kvm_set_spte_hva(kvm, address, pte);
372 spin_unlock(&kvm->mmu_lock);
373 srcu_read_unlock(&kvm->srcu, idx);
376 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
377 struct mm_struct *mm,
381 struct kvm *kvm = mmu_notifier_to_kvm(mn);
382 int need_tlb_flush = 0, idx;
384 idx = srcu_read_lock(&kvm->srcu);
385 spin_lock(&kvm->mmu_lock);
387 * The count increase must become visible at unlock time as no
388 * spte can be established without taking the mmu_lock and
389 * count is also read inside the mmu_lock critical section.
391 kvm->mmu_notifier_count++;
392 need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
393 need_tlb_flush |= kvm->tlbs_dirty;
394 /* we've to flush the tlb before the pages can be freed */
396 kvm_flush_remote_tlbs(kvm);
398 spin_unlock(&kvm->mmu_lock);
399 srcu_read_unlock(&kvm->srcu, idx);
402 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
403 struct mm_struct *mm,
407 struct kvm *kvm = mmu_notifier_to_kvm(mn);
409 spin_lock(&kvm->mmu_lock);
411 * This sequence increase will notify the kvm page fault that
412 * the page that is going to be mapped in the spte could have
415 kvm->mmu_notifier_seq++;
418 * The above sequence increase must be visible before the
419 * below count decrease, which is ensured by the smp_wmb above
420 * in conjunction with the smp_rmb in mmu_notifier_retry().
422 kvm->mmu_notifier_count--;
423 spin_unlock(&kvm->mmu_lock);
425 BUG_ON(kvm->mmu_notifier_count < 0);
428 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
429 struct mm_struct *mm,
433 struct kvm *kvm = mmu_notifier_to_kvm(mn);
436 idx = srcu_read_lock(&kvm->srcu);
437 spin_lock(&kvm->mmu_lock);
439 young = kvm_age_hva(kvm, start, end);
441 kvm_flush_remote_tlbs(kvm);
443 spin_unlock(&kvm->mmu_lock);
444 srcu_read_unlock(&kvm->srcu, idx);
449 static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
450 struct mm_struct *mm,
454 struct kvm *kvm = mmu_notifier_to_kvm(mn);
457 idx = srcu_read_lock(&kvm->srcu);
458 spin_lock(&kvm->mmu_lock);
460 * Even though we do not flush TLB, this will still adversely
461 * affect performance on pre-Haswell Intel EPT, where there is
462 * no EPT Access Bit to clear so that we have to tear down EPT
463 * tables instead. If we find this unacceptable, we can always
464 * add a parameter to kvm_age_hva so that it effectively doesn't
465 * do anything on clear_young.
467 * Also note that currently we never issue secondary TLB flushes
468 * from clear_young, leaving this job up to the regular system
469 * cadence. If we find this inaccurate, we might come up with a
470 * more sophisticated heuristic later.
472 young = kvm_age_hva(kvm, start, end);
473 spin_unlock(&kvm->mmu_lock);
474 srcu_read_unlock(&kvm->srcu, idx);
479 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
480 struct mm_struct *mm,
481 unsigned long address)
483 struct kvm *kvm = mmu_notifier_to_kvm(mn);
486 idx = srcu_read_lock(&kvm->srcu);
487 spin_lock(&kvm->mmu_lock);
488 young = kvm_test_age_hva(kvm, address);
489 spin_unlock(&kvm->mmu_lock);
490 srcu_read_unlock(&kvm->srcu, idx);
495 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
496 struct mm_struct *mm)
498 struct kvm *kvm = mmu_notifier_to_kvm(mn);
501 idx = srcu_read_lock(&kvm->srcu);
502 kvm_arch_flush_shadow_all(kvm);
503 srcu_read_unlock(&kvm->srcu, idx);
506 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
507 .invalidate_page = kvm_mmu_notifier_invalidate_page,
508 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
509 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
510 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
511 .clear_young = kvm_mmu_notifier_clear_young,
512 .test_young = kvm_mmu_notifier_test_young,
513 .change_pte = kvm_mmu_notifier_change_pte,
514 .release = kvm_mmu_notifier_release,
517 static int kvm_init_mmu_notifier(struct kvm *kvm)
519 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
520 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
523 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
525 static int kvm_init_mmu_notifier(struct kvm *kvm)
530 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
532 static struct kvm_memslots *kvm_alloc_memslots(void)
535 struct kvm_memslots *slots;
537 slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
541 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
542 slots->id_to_index[i] = slots->memslots[i].id = i;
547 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
549 if (!memslot->dirty_bitmap)
552 kvfree(memslot->dirty_bitmap);
553 memslot->dirty_bitmap = NULL;
557 * Free any memory in @free but not in @dont.
559 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
560 struct kvm_memory_slot *dont)
562 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
563 kvm_destroy_dirty_bitmap(free);
565 kvm_arch_free_memslot(kvm, free, dont);
570 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
572 struct kvm_memory_slot *memslot;
577 kvm_for_each_memslot(memslot, slots)
578 kvm_free_memslot(kvm, memslot, NULL);
583 static void kvm_destroy_vm_debugfs(struct kvm *kvm)
587 if (!kvm->debugfs_dentry)
590 debugfs_remove_recursive(kvm->debugfs_dentry);
592 if (kvm->debugfs_stat_data) {
593 for (i = 0; i < kvm_debugfs_num_entries; i++)
594 kfree(kvm->debugfs_stat_data[i]);
595 kfree(kvm->debugfs_stat_data);
599 static int kvm_create_vm_debugfs(struct kvm *kvm, int fd)
601 char dir_name[ITOA_MAX_LEN * 2];
602 struct kvm_stat_data *stat_data;
603 struct kvm_stats_debugfs_item *p;
605 if (!debugfs_initialized())
608 snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd);
609 kvm->debugfs_dentry = debugfs_create_dir(dir_name,
611 if (!kvm->debugfs_dentry)
614 kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries,
615 sizeof(*kvm->debugfs_stat_data),
617 if (!kvm->debugfs_stat_data)
620 for (p = debugfs_entries; p->name; p++) {
621 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL);
625 stat_data->kvm = kvm;
626 stat_data->offset = p->offset;
627 kvm->debugfs_stat_data[p - debugfs_entries] = stat_data;
628 if (!debugfs_create_file(p->name, 0644,
631 stat_fops_per_vm[p->kind]))
637 static struct kvm *kvm_create_vm(unsigned long type)
640 struct kvm *kvm = kvm_arch_alloc_vm();
643 return ERR_PTR(-ENOMEM);
645 spin_lock_init(&kvm->mmu_lock);
647 kvm->mm = current->mm;
648 kvm_eventfd_init(kvm);
649 mutex_init(&kvm->lock);
650 mutex_init(&kvm->irq_lock);
651 mutex_init(&kvm->slots_lock);
652 refcount_set(&kvm->users_count, 1);
653 INIT_LIST_HEAD(&kvm->devices);
655 r = kvm_arch_init_vm(kvm, type);
657 goto out_err_no_disable;
659 r = hardware_enable_all();
661 goto out_err_no_disable;
663 #ifdef CONFIG_HAVE_KVM_IRQFD
664 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
667 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
670 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
671 struct kvm_memslots *slots = kvm_alloc_memslots();
673 goto out_err_no_srcu;
675 * Generations must be different for each address space.
676 * Init kvm generation close to the maximum to easily test the
677 * code of handling generation number wrap-around.
679 slots->generation = i * 2 - 150;
680 rcu_assign_pointer(kvm->memslots[i], slots);
683 if (init_srcu_struct(&kvm->srcu))
684 goto out_err_no_srcu;
685 if (init_srcu_struct(&kvm->irq_srcu))
686 goto out_err_no_irq_srcu;
687 for (i = 0; i < KVM_NR_BUSES; i++) {
688 rcu_assign_pointer(kvm->buses[i],
689 kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL));
694 r = kvm_init_mmu_notifier(kvm);
698 spin_lock(&kvm_lock);
699 list_add(&kvm->vm_list, &vm_list);
700 spin_unlock(&kvm_lock);
702 preempt_notifier_inc();
707 cleanup_srcu_struct(&kvm->irq_srcu);
709 cleanup_srcu_struct(&kvm->srcu);
711 hardware_disable_all();
713 for (i = 0; i < KVM_NR_BUSES; i++)
714 kfree(rcu_access_pointer(kvm->buses[i]));
715 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
716 kvm_free_memslots(kvm,
717 rcu_dereference_protected(kvm->memslots[i], 1));
718 kvm_arch_free_vm(kvm);
723 static void kvm_destroy_devices(struct kvm *kvm)
725 struct kvm_device *dev, *tmp;
728 * We do not need to take the kvm->lock here, because nobody else
729 * has a reference to the struct kvm at this point and therefore
730 * cannot access the devices list anyhow.
732 list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
733 list_del(&dev->vm_node);
734 dev->ops->destroy(dev);
738 static void kvm_destroy_vm(struct kvm *kvm)
741 struct mm_struct *mm = kvm->mm;
743 kvm_destroy_vm_debugfs(kvm);
744 kvm_arch_sync_events(kvm);
745 spin_lock(&kvm_lock);
746 list_del(&kvm->vm_list);
747 spin_unlock(&kvm_lock);
748 kvm_free_irq_routing(kvm);
749 for (i = 0; i < KVM_NR_BUSES; i++) {
750 struct kvm_io_bus *bus;
752 bus = rcu_dereference_protected(kvm->buses[i], 1);
754 kvm_io_bus_destroy(bus);
755 kvm->buses[i] = NULL;
757 kvm_coalesced_mmio_free(kvm);
758 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
759 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
761 kvm_arch_flush_shadow_all(kvm);
763 kvm_arch_destroy_vm(kvm);
764 kvm_destroy_devices(kvm);
765 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
766 kvm_free_memslots(kvm,
767 rcu_dereference_protected(kvm->memslots[i], 1));
768 cleanup_srcu_struct(&kvm->irq_srcu);
769 cleanup_srcu_struct(&kvm->srcu);
770 kvm_arch_free_vm(kvm);
771 preempt_notifier_dec();
772 hardware_disable_all();
776 void kvm_get_kvm(struct kvm *kvm)
778 refcount_inc(&kvm->users_count);
780 EXPORT_SYMBOL_GPL(kvm_get_kvm);
782 void kvm_put_kvm(struct kvm *kvm)
784 if (refcount_dec_and_test(&kvm->users_count))
787 EXPORT_SYMBOL_GPL(kvm_put_kvm);
790 static int kvm_vm_release(struct inode *inode, struct file *filp)
792 struct kvm *kvm = filp->private_data;
794 kvm_irqfd_release(kvm);
801 * Allocation size is twice as large as the actual dirty bitmap size.
802 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
804 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
806 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
808 memslot->dirty_bitmap = kvzalloc(dirty_bytes, GFP_KERNEL);
809 if (!memslot->dirty_bitmap)
816 * Insert memslot and re-sort memslots based on their GFN,
817 * so binary search could be used to lookup GFN.
818 * Sorting algorithm takes advantage of having initially
819 * sorted array and known changed memslot position.
821 static void update_memslots(struct kvm_memslots *slots,
822 struct kvm_memory_slot *new)
825 int i = slots->id_to_index[id];
826 struct kvm_memory_slot *mslots = slots->memslots;
828 WARN_ON(mslots[i].id != id);
830 WARN_ON(!mslots[i].npages);
831 if (mslots[i].npages)
834 if (!mslots[i].npages)
838 while (i < KVM_MEM_SLOTS_NUM - 1 &&
839 new->base_gfn <= mslots[i + 1].base_gfn) {
840 if (!mslots[i + 1].npages)
842 mslots[i] = mslots[i + 1];
843 slots->id_to_index[mslots[i].id] = i;
848 * The ">=" is needed when creating a slot with base_gfn == 0,
849 * so that it moves before all those with base_gfn == npages == 0.
851 * On the other hand, if new->npages is zero, the above loop has
852 * already left i pointing to the beginning of the empty part of
853 * mslots, and the ">=" would move the hole backwards in this
854 * case---which is wrong. So skip the loop when deleting a slot.
858 new->base_gfn >= mslots[i - 1].base_gfn) {
859 mslots[i] = mslots[i - 1];
860 slots->id_to_index[mslots[i].id] = i;
864 WARN_ON_ONCE(i != slots->used_slots);
867 slots->id_to_index[mslots[i].id] = i;
870 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
872 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
874 #ifdef __KVM_HAVE_READONLY_MEM
875 valid_flags |= KVM_MEM_READONLY;
878 if (mem->flags & ~valid_flags)
884 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
885 int as_id, struct kvm_memslots *slots)
887 struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id);
890 * Set the low bit in the generation, which disables SPTE caching
891 * until the end of synchronize_srcu_expedited.
893 WARN_ON(old_memslots->generation & 1);
894 slots->generation = old_memslots->generation + 1;
896 rcu_assign_pointer(kvm->memslots[as_id], slots);
897 synchronize_srcu_expedited(&kvm->srcu);
900 * Increment the new memslot generation a second time. This prevents
901 * vm exits that race with memslot updates from caching a memslot
902 * generation that will (potentially) be valid forever.
904 * Generations must be unique even across address spaces. We do not need
905 * a global counter for that, instead the generation space is evenly split
906 * across address spaces. For example, with two address spaces, address
907 * space 0 will use generations 0, 4, 8, ... while * address space 1 will
908 * use generations 2, 6, 10, 14, ...
910 slots->generation += KVM_ADDRESS_SPACE_NUM * 2 - 1;
912 kvm_arch_memslots_updated(kvm, slots);
918 * Allocate some memory and give it an address in the guest physical address
921 * Discontiguous memory is allowed, mostly for framebuffers.
923 * Must be called holding kvm->slots_lock for write.
925 int __kvm_set_memory_region(struct kvm *kvm,
926 const struct kvm_userspace_memory_region *mem)
930 unsigned long npages;
931 struct kvm_memory_slot *slot;
932 struct kvm_memory_slot old, new;
933 struct kvm_memslots *slots = NULL, *old_memslots;
935 enum kvm_mr_change change;
937 r = check_memory_region_flags(mem);
942 as_id = mem->slot >> 16;
945 /* General sanity checks */
946 if (mem->memory_size & (PAGE_SIZE - 1))
948 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
950 /* We can read the guest memory with __xxx_user() later on. */
951 if ((id < KVM_USER_MEM_SLOTS) &&
952 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
953 !access_ok(VERIFY_WRITE,
954 (void __user *)(unsigned long)mem->userspace_addr,
957 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
959 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
962 slot = id_to_memslot(__kvm_memslots(kvm, as_id), id);
963 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
964 npages = mem->memory_size >> PAGE_SHIFT;
966 if (npages > KVM_MEM_MAX_NR_PAGES)
972 new.base_gfn = base_gfn;
974 new.flags = mem->flags;
978 change = KVM_MR_CREATE;
979 else { /* Modify an existing slot. */
980 if ((mem->userspace_addr != old.userspace_addr) ||
981 (npages != old.npages) ||
982 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
985 if (base_gfn != old.base_gfn)
986 change = KVM_MR_MOVE;
987 else if (new.flags != old.flags)
988 change = KVM_MR_FLAGS_ONLY;
989 else { /* Nothing to change. */
998 change = KVM_MR_DELETE;
1003 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
1004 /* Check for overlaps */
1006 kvm_for_each_memslot(slot, __kvm_memslots(kvm, as_id)) {
1007 if ((slot->id >= KVM_USER_MEM_SLOTS) ||
1010 if (!((base_gfn + npages <= slot->base_gfn) ||
1011 (base_gfn >= slot->base_gfn + slot->npages)))
1016 /* Free page dirty bitmap if unneeded */
1017 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
1018 new.dirty_bitmap = NULL;
1021 if (change == KVM_MR_CREATE) {
1022 new.userspace_addr = mem->userspace_addr;
1024 if (kvm_arch_create_memslot(kvm, &new, npages))
1028 /* Allocate page dirty bitmap if needed */
1029 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
1030 if (kvm_create_dirty_bitmap(&new) < 0)
1034 slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
1037 memcpy(slots, __kvm_memslots(kvm, as_id), sizeof(struct kvm_memslots));
1039 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
1040 slot = id_to_memslot(slots, id);
1041 slot->flags |= KVM_MEMSLOT_INVALID;
1043 old_memslots = install_new_memslots(kvm, as_id, slots);
1045 /* From this point no new shadow pages pointing to a deleted,
1046 * or moved, memslot will be created.
1048 * validation of sp->gfn happens in:
1049 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1050 * - kvm_is_visible_gfn (mmu_check_roots)
1052 kvm_arch_flush_shadow_memslot(kvm, slot);
1055 * We can re-use the old_memslots from above, the only difference
1056 * from the currently installed memslots is the invalid flag. This
1057 * will get overwritten by update_memslots anyway.
1059 slots = old_memslots;
1062 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
1066 /* actual memory is freed via old in kvm_free_memslot below */
1067 if (change == KVM_MR_DELETE) {
1068 new.dirty_bitmap = NULL;
1069 memset(&new.arch, 0, sizeof(new.arch));
1072 update_memslots(slots, &new);
1073 old_memslots = install_new_memslots(kvm, as_id, slots);
1075 kvm_arch_commit_memory_region(kvm, mem, &old, &new, change);
1077 kvm_free_memslot(kvm, &old, &new);
1078 kvfree(old_memslots);
1084 kvm_free_memslot(kvm, &new, &old);
1088 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1090 int kvm_set_memory_region(struct kvm *kvm,
1091 const struct kvm_userspace_memory_region *mem)
1095 mutex_lock(&kvm->slots_lock);
1096 r = __kvm_set_memory_region(kvm, mem);
1097 mutex_unlock(&kvm->slots_lock);
1100 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1102 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1103 struct kvm_userspace_memory_region *mem)
1105 if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1108 return kvm_set_memory_region(kvm, mem);
1111 int kvm_get_dirty_log(struct kvm *kvm,
1112 struct kvm_dirty_log *log, int *is_dirty)
1114 struct kvm_memslots *slots;
1115 struct kvm_memory_slot *memslot;
1118 unsigned long any = 0;
1120 as_id = log->slot >> 16;
1121 id = (u16)log->slot;
1122 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1125 slots = __kvm_memslots(kvm, as_id);
1126 memslot = id_to_memslot(slots, id);
1127 if (!memslot->dirty_bitmap)
1130 n = kvm_dirty_bitmap_bytes(memslot);
1132 for (i = 0; !any && i < n/sizeof(long); ++i)
1133 any = memslot->dirty_bitmap[i];
1135 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1142 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1144 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1146 * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages
1147 * are dirty write protect them for next write.
1148 * @kvm: pointer to kvm instance
1149 * @log: slot id and address to which we copy the log
1150 * @is_dirty: flag set if any page is dirty
1152 * We need to keep it in mind that VCPU threads can write to the bitmap
1153 * concurrently. So, to avoid losing track of dirty pages we keep the
1156 * 1. Take a snapshot of the bit and clear it if needed.
1157 * 2. Write protect the corresponding page.
1158 * 3. Copy the snapshot to the userspace.
1159 * 4. Upon return caller flushes TLB's if needed.
1161 * Between 2 and 4, the guest may write to the page using the remaining TLB
1162 * entry. This is not a problem because the page is reported dirty using
1163 * the snapshot taken before and step 4 ensures that writes done after
1164 * exiting to userspace will be logged for the next call.
1167 int kvm_get_dirty_log_protect(struct kvm *kvm,
1168 struct kvm_dirty_log *log, bool *is_dirty)
1170 struct kvm_memslots *slots;
1171 struct kvm_memory_slot *memslot;
1174 unsigned long *dirty_bitmap;
1175 unsigned long *dirty_bitmap_buffer;
1177 as_id = log->slot >> 16;
1178 id = (u16)log->slot;
1179 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1182 slots = __kvm_memslots(kvm, as_id);
1183 memslot = id_to_memslot(slots, id);
1185 dirty_bitmap = memslot->dirty_bitmap;
1189 n = kvm_dirty_bitmap_bytes(memslot);
1191 dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long);
1192 memset(dirty_bitmap_buffer, 0, n);
1194 spin_lock(&kvm->mmu_lock);
1196 for (i = 0; i < n / sizeof(long); i++) {
1200 if (!dirty_bitmap[i])
1205 mask = xchg(&dirty_bitmap[i], 0);
1206 dirty_bitmap_buffer[i] = mask;
1209 offset = i * BITS_PER_LONG;
1210 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1215 spin_unlock(&kvm->mmu_lock);
1216 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1220 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect);
1223 bool kvm_largepages_enabled(void)
1225 return largepages_enabled;
1228 void kvm_disable_largepages(void)
1230 largepages_enabled = false;
1232 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1234 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1236 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1238 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1240 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
1242 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn);
1245 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1247 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1249 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1250 memslot->flags & KVM_MEMSLOT_INVALID)
1255 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1257 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1259 struct vm_area_struct *vma;
1260 unsigned long addr, size;
1264 addr = gfn_to_hva(kvm, gfn);
1265 if (kvm_is_error_hva(addr))
1268 down_read(¤t->mm->mmap_sem);
1269 vma = find_vma(current->mm, addr);
1273 size = vma_kernel_pagesize(vma);
1276 up_read(¤t->mm->mmap_sem);
1281 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1283 return slot->flags & KVM_MEM_READONLY;
1286 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1287 gfn_t *nr_pages, bool write)
1289 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1290 return KVM_HVA_ERR_BAD;
1292 if (memslot_is_readonly(slot) && write)
1293 return KVM_HVA_ERR_RO_BAD;
1296 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1298 return __gfn_to_hva_memslot(slot, gfn);
1301 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1304 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1307 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1310 return gfn_to_hva_many(slot, gfn, NULL);
1312 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1314 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1316 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1318 EXPORT_SYMBOL_GPL(gfn_to_hva);
1320 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
1322 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
1324 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
1327 * If writable is set to false, the hva returned by this function is only
1328 * allowed to be read.
1330 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1331 gfn_t gfn, bool *writable)
1333 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1335 if (!kvm_is_error_hva(hva) && writable)
1336 *writable = !memslot_is_readonly(slot);
1341 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1343 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1345 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1348 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
1350 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1352 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1355 static int get_user_page_nowait(unsigned long start, int write,
1358 int flags = FOLL_NOWAIT | FOLL_HWPOISON;
1361 flags |= FOLL_WRITE;
1363 return get_user_pages(start, 1, flags, page, NULL);
1366 static inline int check_user_page_hwpoison(unsigned long addr)
1368 int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
1370 rc = get_user_pages(addr, 1, flags, NULL, NULL);
1371 return rc == -EHWPOISON;
1375 * The atomic path to get the writable pfn which will be stored in @pfn,
1376 * true indicates success, otherwise false is returned.
1378 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1379 bool write_fault, bool *writable, kvm_pfn_t *pfn)
1381 struct page *page[1];
1384 if (!(async || atomic))
1388 * Fast pin a writable pfn only if it is a write fault request
1389 * or the caller allows to map a writable pfn for a read fault
1392 if (!(write_fault || writable))
1395 npages = __get_user_pages_fast(addr, 1, 1, page);
1397 *pfn = page_to_pfn(page[0]);
1408 * The slow path to get the pfn of the specified host virtual address,
1409 * 1 indicates success, -errno is returned if error is detected.
1411 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1412 bool *writable, kvm_pfn_t *pfn)
1414 struct page *page[1];
1420 *writable = write_fault;
1423 down_read(¤t->mm->mmap_sem);
1424 npages = get_user_page_nowait(addr, write_fault, page);
1425 up_read(¤t->mm->mmap_sem);
1427 unsigned int flags = FOLL_HWPOISON;
1430 flags |= FOLL_WRITE;
1432 npages = get_user_pages_unlocked(addr, 1, page, flags);
1437 /* map read fault as writable if possible */
1438 if (unlikely(!write_fault) && writable) {
1439 struct page *wpage[1];
1441 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1450 *pfn = page_to_pfn(page[0]);
1454 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1456 if (unlikely(!(vma->vm_flags & VM_READ)))
1459 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1465 static int hva_to_pfn_remapped(struct vm_area_struct *vma,
1466 unsigned long addr, bool *async,
1467 bool write_fault, kvm_pfn_t *p_pfn)
1472 r = follow_pfn(vma, addr, &pfn);
1475 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
1476 * not call the fault handler, so do it here.
1478 bool unlocked = false;
1479 r = fixup_user_fault(current, current->mm, addr,
1480 (write_fault ? FAULT_FLAG_WRITE : 0),
1487 r = follow_pfn(vma, addr, &pfn);
1495 * Get a reference here because callers of *hva_to_pfn* and
1496 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
1497 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
1498 * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will
1499 * simply do nothing for reserved pfns.
1501 * Whoever called remap_pfn_range is also going to call e.g.
1502 * unmap_mapping_range before the underlying pages are freed,
1503 * causing a call to our MMU notifier.
1512 * Pin guest page in memory and return its pfn.
1513 * @addr: host virtual address which maps memory to the guest
1514 * @atomic: whether this function can sleep
1515 * @async: whether this function need to wait IO complete if the
1516 * host page is not in the memory
1517 * @write_fault: whether we should get a writable host page
1518 * @writable: whether it allows to map a writable host page for !@write_fault
1520 * The function will map a writable host page for these two cases:
1521 * 1): @write_fault = true
1522 * 2): @write_fault = false && @writable, @writable will tell the caller
1523 * whether the mapping is writable.
1525 static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1526 bool write_fault, bool *writable)
1528 struct vm_area_struct *vma;
1532 /* we can do it either atomically or asynchronously, not both */
1533 BUG_ON(atomic && async);
1535 if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1539 return KVM_PFN_ERR_FAULT;
1541 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1545 down_read(¤t->mm->mmap_sem);
1546 if (npages == -EHWPOISON ||
1547 (!async && check_user_page_hwpoison(addr))) {
1548 pfn = KVM_PFN_ERR_HWPOISON;
1553 vma = find_vma_intersection(current->mm, addr, addr + 1);
1556 pfn = KVM_PFN_ERR_FAULT;
1557 else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
1558 r = hva_to_pfn_remapped(vma, addr, async, write_fault, &pfn);
1562 pfn = KVM_PFN_ERR_FAULT;
1564 if (async && vma_is_valid(vma, write_fault))
1566 pfn = KVM_PFN_ERR_FAULT;
1569 up_read(¤t->mm->mmap_sem);
1573 kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn,
1574 bool atomic, bool *async, bool write_fault,
1577 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1579 if (addr == KVM_HVA_ERR_RO_BAD) {
1582 return KVM_PFN_ERR_RO_FAULT;
1585 if (kvm_is_error_hva(addr)) {
1588 return KVM_PFN_NOSLOT;
1591 /* Do not map writable pfn in the readonly memslot. */
1592 if (writable && memslot_is_readonly(slot)) {
1597 return hva_to_pfn(addr, atomic, async, write_fault,
1600 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
1602 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1605 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
1606 write_fault, writable);
1608 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1610 kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1612 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1614 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
1616 kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1618 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1620 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1622 kvm_pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1624 return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm, gfn), gfn);
1626 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1628 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
1630 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1632 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
1634 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1636 return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
1638 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1640 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
1642 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1644 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
1646 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1647 struct page **pages, int nr_pages)
1652 addr = gfn_to_hva_many(slot, gfn, &entry);
1653 if (kvm_is_error_hva(addr))
1656 if (entry < nr_pages)
1659 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1661 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1663 static struct page *kvm_pfn_to_page(kvm_pfn_t pfn)
1665 if (is_error_noslot_pfn(pfn))
1666 return KVM_ERR_PTR_BAD_PAGE;
1668 if (kvm_is_reserved_pfn(pfn)) {
1670 return KVM_ERR_PTR_BAD_PAGE;
1673 return pfn_to_page(pfn);
1676 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1680 pfn = gfn_to_pfn(kvm, gfn);
1682 return kvm_pfn_to_page(pfn);
1684 EXPORT_SYMBOL_GPL(gfn_to_page);
1686 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
1690 pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
1692 return kvm_pfn_to_page(pfn);
1694 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
1696 void kvm_release_page_clean(struct page *page)
1698 WARN_ON(is_error_page(page));
1700 kvm_release_pfn_clean(page_to_pfn(page));
1702 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1704 void kvm_release_pfn_clean(kvm_pfn_t pfn)
1706 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
1707 put_page(pfn_to_page(pfn));
1709 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1711 void kvm_release_page_dirty(struct page *page)
1713 WARN_ON(is_error_page(page));
1715 kvm_release_pfn_dirty(page_to_pfn(page));
1717 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1719 static void kvm_release_pfn_dirty(kvm_pfn_t pfn)
1721 kvm_set_pfn_dirty(pfn);
1722 kvm_release_pfn_clean(pfn);
1725 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
1727 if (!kvm_is_reserved_pfn(pfn)) {
1728 struct page *page = pfn_to_page(pfn);
1730 if (!PageReserved(page))
1734 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1736 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
1738 if (!kvm_is_reserved_pfn(pfn))
1739 mark_page_accessed(pfn_to_page(pfn));
1741 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1743 void kvm_get_pfn(kvm_pfn_t pfn)
1745 if (!kvm_is_reserved_pfn(pfn))
1746 get_page(pfn_to_page(pfn));
1748 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1750 static int next_segment(unsigned long len, int offset)
1752 if (len > PAGE_SIZE - offset)
1753 return PAGE_SIZE - offset;
1758 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
1759 void *data, int offset, int len)
1764 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1765 if (kvm_is_error_hva(addr))
1767 r = __copy_from_user(data, (void __user *)addr + offset, len);
1773 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1776 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1778 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1780 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1782 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
1783 int offset, int len)
1785 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1787 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1789 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
1791 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1793 gfn_t gfn = gpa >> PAGE_SHIFT;
1795 int offset = offset_in_page(gpa);
1798 while ((seg = next_segment(len, offset)) != 0) {
1799 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1809 EXPORT_SYMBOL_GPL(kvm_read_guest);
1811 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
1813 gfn_t gfn = gpa >> PAGE_SHIFT;
1815 int offset = offset_in_page(gpa);
1818 while ((seg = next_segment(len, offset)) != 0) {
1819 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
1829 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
1831 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1832 void *data, int offset, unsigned long len)
1837 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1838 if (kvm_is_error_hva(addr))
1840 pagefault_disable();
1841 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1848 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1851 gfn_t gfn = gpa >> PAGE_SHIFT;
1852 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1853 int offset = offset_in_page(gpa);
1855 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1857 EXPORT_SYMBOL_GPL(kvm_read_guest_atomic);
1859 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
1860 void *data, unsigned long len)
1862 gfn_t gfn = gpa >> PAGE_SHIFT;
1863 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1864 int offset = offset_in_page(gpa);
1866 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1868 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
1870 static int __kvm_write_guest_page(struct kvm_memory_slot *memslot, gfn_t gfn,
1871 const void *data, int offset, int len)
1876 addr = gfn_to_hva_memslot(memslot, gfn);
1877 if (kvm_is_error_hva(addr))
1879 r = __copy_to_user((void __user *)addr + offset, data, len);
1882 mark_page_dirty_in_slot(memslot, gfn);
1886 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
1887 const void *data, int offset, int len)
1889 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1891 return __kvm_write_guest_page(slot, gfn, data, offset, len);
1893 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1895 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
1896 const void *data, int offset, int len)
1898 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1900 return __kvm_write_guest_page(slot, gfn, data, offset, len);
1902 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
1904 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1907 gfn_t gfn = gpa >> PAGE_SHIFT;
1909 int offset = offset_in_page(gpa);
1912 while ((seg = next_segment(len, offset)) != 0) {
1913 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1923 EXPORT_SYMBOL_GPL(kvm_write_guest);
1925 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
1928 gfn_t gfn = gpa >> PAGE_SHIFT;
1930 int offset = offset_in_page(gpa);
1933 while ((seg = next_segment(len, offset)) != 0) {
1934 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
1944 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
1946 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
1947 struct gfn_to_hva_cache *ghc,
1948 gpa_t gpa, unsigned long len)
1950 int offset = offset_in_page(gpa);
1951 gfn_t start_gfn = gpa >> PAGE_SHIFT;
1952 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1953 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1954 gfn_t nr_pages_avail;
1957 ghc->generation = slots->generation;
1959 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
1960 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, NULL);
1961 if (!kvm_is_error_hva(ghc->hva) && nr_pages_needed <= 1) {
1965 * If the requested region crosses two memslots, we still
1966 * verify that the entire region is valid here.
1968 while (start_gfn <= end_gfn) {
1969 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
1970 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1972 if (kvm_is_error_hva(ghc->hva))
1974 start_gfn += nr_pages_avail;
1976 /* Use the slow path for cross page reads and writes. */
1977 ghc->memslot = NULL;
1982 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1983 gpa_t gpa, unsigned long len)
1985 struct kvm_memslots *slots = kvm_memslots(kvm);
1986 return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len);
1988 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1990 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1991 void *data, int offset, unsigned long len)
1993 struct kvm_memslots *slots = kvm_memslots(kvm);
1995 gpa_t gpa = ghc->gpa + offset;
1997 BUG_ON(len + offset > ghc->len);
1999 if (slots->generation != ghc->generation)
2000 __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len);
2002 if (unlikely(!ghc->memslot))
2003 return kvm_write_guest(kvm, gpa, data, len);
2005 if (kvm_is_error_hva(ghc->hva))
2008 r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
2011 mark_page_dirty_in_slot(ghc->memslot, gpa >> PAGE_SHIFT);
2015 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached);
2017 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2018 void *data, unsigned long len)
2020 return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len);
2022 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
2024 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2025 void *data, unsigned long len)
2027 struct kvm_memslots *slots = kvm_memslots(kvm);
2030 BUG_ON(len > ghc->len);
2032 if (slots->generation != ghc->generation)
2033 __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len);
2035 if (unlikely(!ghc->memslot))
2036 return kvm_read_guest(kvm, ghc->gpa, data, len);
2038 if (kvm_is_error_hva(ghc->hva))
2041 r = __copy_from_user(data, (void __user *)ghc->hva, len);
2047 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
2049 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
2051 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
2053 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
2055 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
2057 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
2059 gfn_t gfn = gpa >> PAGE_SHIFT;
2061 int offset = offset_in_page(gpa);
2064 while ((seg = next_segment(len, offset)) != 0) {
2065 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
2074 EXPORT_SYMBOL_GPL(kvm_clear_guest);
2076 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot,
2079 if (memslot && memslot->dirty_bitmap) {
2080 unsigned long rel_gfn = gfn - memslot->base_gfn;
2082 set_bit_le(rel_gfn, memslot->dirty_bitmap);
2086 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
2088 struct kvm_memory_slot *memslot;
2090 memslot = gfn_to_memslot(kvm, gfn);
2091 mark_page_dirty_in_slot(memslot, gfn);
2093 EXPORT_SYMBOL_GPL(mark_page_dirty);
2095 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
2097 struct kvm_memory_slot *memslot;
2099 memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2100 mark_page_dirty_in_slot(memslot, gfn);
2102 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
2104 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
2106 unsigned int old, val, grow;
2108 old = val = vcpu->halt_poll_ns;
2109 grow = READ_ONCE(halt_poll_ns_grow);
2111 if (val == 0 && grow)
2116 if (val > halt_poll_ns)
2119 vcpu->halt_poll_ns = val;
2120 trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
2123 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
2125 unsigned int old, val, shrink;
2127 old = val = vcpu->halt_poll_ns;
2128 shrink = READ_ONCE(halt_poll_ns_shrink);
2134 vcpu->halt_poll_ns = val;
2135 trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
2138 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
2140 if (kvm_arch_vcpu_runnable(vcpu)) {
2141 kvm_make_request(KVM_REQ_UNHALT, vcpu);
2144 if (kvm_cpu_has_pending_timer(vcpu))
2146 if (signal_pending(current))
2153 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2155 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
2158 DECLARE_SWAITQUEUE(wait);
2159 bool waited = false;
2162 start = cur = ktime_get();
2163 if (vcpu->halt_poll_ns) {
2164 ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns);
2166 ++vcpu->stat.halt_attempted_poll;
2169 * This sets KVM_REQ_UNHALT if an interrupt
2172 if (kvm_vcpu_check_block(vcpu) < 0) {
2173 ++vcpu->stat.halt_successful_poll;
2174 if (!vcpu_valid_wakeup(vcpu))
2175 ++vcpu->stat.halt_poll_invalid;
2179 } while (single_task_running() && ktime_before(cur, stop));
2182 kvm_arch_vcpu_blocking(vcpu);
2185 prepare_to_swait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
2187 if (kvm_vcpu_check_block(vcpu) < 0)
2194 finish_swait(&vcpu->wq, &wait);
2197 kvm_arch_vcpu_unblocking(vcpu);
2199 block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
2201 if (!vcpu_valid_wakeup(vcpu))
2202 shrink_halt_poll_ns(vcpu);
2203 else if (halt_poll_ns) {
2204 if (block_ns <= vcpu->halt_poll_ns)
2206 /* we had a long block, shrink polling */
2207 else if (vcpu->halt_poll_ns && block_ns > halt_poll_ns)
2208 shrink_halt_poll_ns(vcpu);
2209 /* we had a short halt and our poll time is too small */
2210 else if (vcpu->halt_poll_ns < halt_poll_ns &&
2211 block_ns < halt_poll_ns)
2212 grow_halt_poll_ns(vcpu);
2214 vcpu->halt_poll_ns = 0;
2216 trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu));
2217 kvm_arch_vcpu_block_finish(vcpu);
2219 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
2221 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
2223 struct swait_queue_head *wqp;
2225 wqp = kvm_arch_vcpu_wq(vcpu);
2226 if (swait_active(wqp)) {
2228 ++vcpu->stat.halt_wakeup;
2234 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
2238 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2240 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
2243 int cpu = vcpu->cpu;
2245 if (kvm_vcpu_wake_up(vcpu))
2249 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
2250 if (kvm_arch_vcpu_should_kick(vcpu))
2251 smp_send_reschedule(cpu);
2254 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
2255 #endif /* !CONFIG_S390 */
2257 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
2260 struct task_struct *task = NULL;
2264 pid = rcu_dereference(target->pid);
2266 task = get_pid_task(pid, PIDTYPE_PID);
2270 ret = yield_to(task, 1);
2271 put_task_struct(task);
2275 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
2278 * Helper that checks whether a VCPU is eligible for directed yield.
2279 * Most eligible candidate to yield is decided by following heuristics:
2281 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2282 * (preempted lock holder), indicated by @in_spin_loop.
2283 * Set at the beiginning and cleared at the end of interception/PLE handler.
2285 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2286 * chance last time (mostly it has become eligible now since we have probably
2287 * yielded to lockholder in last iteration. This is done by toggling
2288 * @dy_eligible each time a VCPU checked for eligibility.)
2290 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2291 * to preempted lock-holder could result in wrong VCPU selection and CPU
2292 * burning. Giving priority for a potential lock-holder increases lock
2295 * Since algorithm is based on heuristics, accessing another VCPU data without
2296 * locking does not harm. It may result in trying to yield to same VCPU, fail
2297 * and continue with next VCPU and so on.
2299 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
2301 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2304 eligible = !vcpu->spin_loop.in_spin_loop ||
2305 vcpu->spin_loop.dy_eligible;
2307 if (vcpu->spin_loop.in_spin_loop)
2308 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
2316 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
2318 struct kvm *kvm = me->kvm;
2319 struct kvm_vcpu *vcpu;
2320 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
2326 kvm_vcpu_set_in_spin_loop(me, true);
2328 * We boost the priority of a VCPU that is runnable but not
2329 * currently running, because it got preempted by something
2330 * else and called schedule in __vcpu_run. Hopefully that
2331 * VCPU is holding the lock that we need and will release it.
2332 * We approximate round-robin by starting at the last boosted VCPU.
2334 for (pass = 0; pass < 2 && !yielded && try; pass++) {
2335 kvm_for_each_vcpu(i, vcpu, kvm) {
2336 if (!pass && i <= last_boosted_vcpu) {
2337 i = last_boosted_vcpu;
2339 } else if (pass && i > last_boosted_vcpu)
2341 if (!ACCESS_ONCE(vcpu->preempted))
2345 if (swait_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
2347 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
2350 yielded = kvm_vcpu_yield_to(vcpu);
2352 kvm->last_boosted_vcpu = i;
2354 } else if (yielded < 0) {
2361 kvm_vcpu_set_in_spin_loop(me, false);
2363 /* Ensure vcpu is not eligible during next spinloop */
2364 kvm_vcpu_set_dy_eligible(me, false);
2366 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
2368 static int kvm_vcpu_fault(struct vm_fault *vmf)
2370 struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data;
2373 if (vmf->pgoff == 0)
2374 page = virt_to_page(vcpu->run);
2376 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
2377 page = virt_to_page(vcpu->arch.pio_data);
2379 #ifdef CONFIG_KVM_MMIO
2380 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
2381 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
2384 return kvm_arch_vcpu_fault(vcpu, vmf);
2390 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
2391 .fault = kvm_vcpu_fault,
2394 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2396 vma->vm_ops = &kvm_vcpu_vm_ops;
2400 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2402 struct kvm_vcpu *vcpu = filp->private_data;
2404 debugfs_remove_recursive(vcpu->debugfs_dentry);
2405 kvm_put_kvm(vcpu->kvm);
2409 static struct file_operations kvm_vcpu_fops = {
2410 .release = kvm_vcpu_release,
2411 .unlocked_ioctl = kvm_vcpu_ioctl,
2412 #ifdef CONFIG_KVM_COMPAT
2413 .compat_ioctl = kvm_vcpu_compat_ioctl,
2415 .mmap = kvm_vcpu_mmap,
2416 .llseek = noop_llseek,
2420 * Allocates an inode for the vcpu.
2422 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2424 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
2427 static int kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
2429 char dir_name[ITOA_MAX_LEN * 2];
2432 if (!kvm_arch_has_vcpu_debugfs())
2435 if (!debugfs_initialized())
2438 snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
2439 vcpu->debugfs_dentry = debugfs_create_dir(dir_name,
2440 vcpu->kvm->debugfs_dentry);
2441 if (!vcpu->debugfs_dentry)
2444 ret = kvm_arch_create_vcpu_debugfs(vcpu);
2446 debugfs_remove_recursive(vcpu->debugfs_dentry);
2454 * Creates some virtual cpus. Good luck creating more than one.
2456 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
2459 struct kvm_vcpu *vcpu;
2461 if (id >= KVM_MAX_VCPU_ID)
2464 mutex_lock(&kvm->lock);
2465 if (kvm->created_vcpus == KVM_MAX_VCPUS) {
2466 mutex_unlock(&kvm->lock);
2470 kvm->created_vcpus++;
2471 mutex_unlock(&kvm->lock);
2473 vcpu = kvm_arch_vcpu_create(kvm, id);
2476 goto vcpu_decrement;
2479 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2481 r = kvm_arch_vcpu_setup(vcpu);
2485 r = kvm_create_vcpu_debugfs(vcpu);
2489 mutex_lock(&kvm->lock);
2490 if (kvm_get_vcpu_by_id(kvm, id)) {
2492 goto unlock_vcpu_destroy;
2495 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
2497 /* Now it's all set up, let userspace reach it */
2499 r = create_vcpu_fd(vcpu);
2502 goto unlock_vcpu_destroy;
2505 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
2508 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
2509 * before kvm->online_vcpu's incremented value.
2512 atomic_inc(&kvm->online_vcpus);
2514 mutex_unlock(&kvm->lock);
2515 kvm_arch_vcpu_postcreate(vcpu);
2518 unlock_vcpu_destroy:
2519 mutex_unlock(&kvm->lock);
2520 debugfs_remove_recursive(vcpu->debugfs_dentry);
2522 kvm_arch_vcpu_destroy(vcpu);
2524 mutex_lock(&kvm->lock);
2525 kvm->created_vcpus--;
2526 mutex_unlock(&kvm->lock);
2530 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2533 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2534 vcpu->sigset_active = 1;
2535 vcpu->sigset = *sigset;
2537 vcpu->sigset_active = 0;
2541 static long kvm_vcpu_ioctl(struct file *filp,
2542 unsigned int ioctl, unsigned long arg)
2544 struct kvm_vcpu *vcpu = filp->private_data;
2545 void __user *argp = (void __user *)arg;
2547 struct kvm_fpu *fpu = NULL;
2548 struct kvm_sregs *kvm_sregs = NULL;
2550 if (vcpu->kvm->mm != current->mm)
2553 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2556 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
2558 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
2559 * so vcpu_load() would break it.
2561 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_S390_IRQ || ioctl == KVM_INTERRUPT)
2562 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2566 r = vcpu_load(vcpu);
2575 oldpid = rcu_access_pointer(vcpu->pid);
2576 if (unlikely(oldpid != current->pids[PIDTYPE_PID].pid)) {
2577 /* The thread running this VCPU changed. */
2578 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
2580 rcu_assign_pointer(vcpu->pid, newpid);
2585 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2586 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2589 case KVM_GET_REGS: {
2590 struct kvm_regs *kvm_regs;
2593 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2596 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2600 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2607 case KVM_SET_REGS: {
2608 struct kvm_regs *kvm_regs;
2611 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2612 if (IS_ERR(kvm_regs)) {
2613 r = PTR_ERR(kvm_regs);
2616 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2620 case KVM_GET_SREGS: {
2621 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2625 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2629 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2634 case KVM_SET_SREGS: {
2635 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2636 if (IS_ERR(kvm_sregs)) {
2637 r = PTR_ERR(kvm_sregs);
2641 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2644 case KVM_GET_MP_STATE: {
2645 struct kvm_mp_state mp_state;
2647 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2651 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
2656 case KVM_SET_MP_STATE: {
2657 struct kvm_mp_state mp_state;
2660 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
2662 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2665 case KVM_TRANSLATE: {
2666 struct kvm_translation tr;
2669 if (copy_from_user(&tr, argp, sizeof(tr)))
2671 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2675 if (copy_to_user(argp, &tr, sizeof(tr)))
2680 case KVM_SET_GUEST_DEBUG: {
2681 struct kvm_guest_debug dbg;
2684 if (copy_from_user(&dbg, argp, sizeof(dbg)))
2686 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2689 case KVM_SET_SIGNAL_MASK: {
2690 struct kvm_signal_mask __user *sigmask_arg = argp;
2691 struct kvm_signal_mask kvm_sigmask;
2692 sigset_t sigset, *p;
2697 if (copy_from_user(&kvm_sigmask, argp,
2698 sizeof(kvm_sigmask)))
2701 if (kvm_sigmask.len != sizeof(sigset))
2704 if (copy_from_user(&sigset, sigmask_arg->sigset,
2709 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2713 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2717 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2721 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2727 fpu = memdup_user(argp, sizeof(*fpu));
2733 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2737 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2746 #ifdef CONFIG_KVM_COMPAT
2747 static long kvm_vcpu_compat_ioctl(struct file *filp,
2748 unsigned int ioctl, unsigned long arg)
2750 struct kvm_vcpu *vcpu = filp->private_data;
2751 void __user *argp = compat_ptr(arg);
2754 if (vcpu->kvm->mm != current->mm)
2758 case KVM_SET_SIGNAL_MASK: {
2759 struct kvm_signal_mask __user *sigmask_arg = argp;
2760 struct kvm_signal_mask kvm_sigmask;
2761 compat_sigset_t csigset;
2766 if (copy_from_user(&kvm_sigmask, argp,
2767 sizeof(kvm_sigmask)))
2770 if (kvm_sigmask.len != sizeof(csigset))
2773 if (copy_from_user(&csigset, sigmask_arg->sigset,
2776 sigset_from_compat(&sigset, &csigset);
2777 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2779 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2783 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2791 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2792 int (*accessor)(struct kvm_device *dev,
2793 struct kvm_device_attr *attr),
2796 struct kvm_device_attr attr;
2801 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2804 return accessor(dev, &attr);
2807 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2810 struct kvm_device *dev = filp->private_data;
2813 case KVM_SET_DEVICE_ATTR:
2814 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2815 case KVM_GET_DEVICE_ATTR:
2816 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2817 case KVM_HAS_DEVICE_ATTR:
2818 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2820 if (dev->ops->ioctl)
2821 return dev->ops->ioctl(dev, ioctl, arg);
2827 static int kvm_device_release(struct inode *inode, struct file *filp)
2829 struct kvm_device *dev = filp->private_data;
2830 struct kvm *kvm = dev->kvm;
2836 static const struct file_operations kvm_device_fops = {
2837 .unlocked_ioctl = kvm_device_ioctl,
2838 #ifdef CONFIG_KVM_COMPAT
2839 .compat_ioctl = kvm_device_ioctl,
2841 .release = kvm_device_release,
2844 struct kvm_device *kvm_device_from_filp(struct file *filp)
2846 if (filp->f_op != &kvm_device_fops)
2849 return filp->private_data;
2852 static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
2853 #ifdef CONFIG_KVM_MPIC
2854 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
2855 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
2859 int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
2861 if (type >= ARRAY_SIZE(kvm_device_ops_table))
2864 if (kvm_device_ops_table[type] != NULL)
2867 kvm_device_ops_table[type] = ops;
2871 void kvm_unregister_device_ops(u32 type)
2873 if (kvm_device_ops_table[type] != NULL)
2874 kvm_device_ops_table[type] = NULL;
2877 static int kvm_ioctl_create_device(struct kvm *kvm,
2878 struct kvm_create_device *cd)
2880 struct kvm_device_ops *ops = NULL;
2881 struct kvm_device *dev;
2882 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2885 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
2888 ops = kvm_device_ops_table[cd->type];
2895 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2902 mutex_lock(&kvm->lock);
2903 ret = ops->create(dev, cd->type);
2905 mutex_unlock(&kvm->lock);
2909 list_add(&dev->vm_node, &kvm->devices);
2910 mutex_unlock(&kvm->lock);
2915 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2917 mutex_lock(&kvm->lock);
2918 list_del(&dev->vm_node);
2919 mutex_unlock(&kvm->lock);
2929 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2932 case KVM_CAP_USER_MEMORY:
2933 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2934 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2935 case KVM_CAP_INTERNAL_ERROR_DATA:
2936 #ifdef CONFIG_HAVE_KVM_MSI
2937 case KVM_CAP_SIGNAL_MSI:
2939 #ifdef CONFIG_HAVE_KVM_IRQFD
2941 case KVM_CAP_IRQFD_RESAMPLE:
2943 case KVM_CAP_IOEVENTFD_ANY_LENGTH:
2944 case KVM_CAP_CHECK_EXTENSION_VM:
2946 #ifdef CONFIG_KVM_MMIO
2947 case KVM_CAP_COALESCED_MMIO:
2948 return KVM_COALESCED_MMIO_PAGE_OFFSET;
2950 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2951 case KVM_CAP_IRQ_ROUTING:
2952 return KVM_MAX_IRQ_ROUTES;
2954 #if KVM_ADDRESS_SPACE_NUM > 1
2955 case KVM_CAP_MULTI_ADDRESS_SPACE:
2956 return KVM_ADDRESS_SPACE_NUM;
2958 case KVM_CAP_MAX_VCPU_ID:
2959 return KVM_MAX_VCPU_ID;
2963 return kvm_vm_ioctl_check_extension(kvm, arg);
2966 static long kvm_vm_ioctl(struct file *filp,
2967 unsigned int ioctl, unsigned long arg)
2969 struct kvm *kvm = filp->private_data;
2970 void __user *argp = (void __user *)arg;
2973 if (kvm->mm != current->mm)
2976 case KVM_CREATE_VCPU:
2977 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2979 case KVM_SET_USER_MEMORY_REGION: {
2980 struct kvm_userspace_memory_region kvm_userspace_mem;
2983 if (copy_from_user(&kvm_userspace_mem, argp,
2984 sizeof(kvm_userspace_mem)))
2987 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2990 case KVM_GET_DIRTY_LOG: {
2991 struct kvm_dirty_log log;
2994 if (copy_from_user(&log, argp, sizeof(log)))
2996 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2999 #ifdef CONFIG_KVM_MMIO
3000 case KVM_REGISTER_COALESCED_MMIO: {
3001 struct kvm_coalesced_mmio_zone zone;
3004 if (copy_from_user(&zone, argp, sizeof(zone)))
3006 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
3009 case KVM_UNREGISTER_COALESCED_MMIO: {
3010 struct kvm_coalesced_mmio_zone zone;
3013 if (copy_from_user(&zone, argp, sizeof(zone)))
3015 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
3020 struct kvm_irqfd data;
3023 if (copy_from_user(&data, argp, sizeof(data)))
3025 r = kvm_irqfd(kvm, &data);
3028 case KVM_IOEVENTFD: {
3029 struct kvm_ioeventfd data;
3032 if (copy_from_user(&data, argp, sizeof(data)))
3034 r = kvm_ioeventfd(kvm, &data);
3037 #ifdef CONFIG_HAVE_KVM_MSI
3038 case KVM_SIGNAL_MSI: {
3042 if (copy_from_user(&msi, argp, sizeof(msi)))
3044 r = kvm_send_userspace_msi(kvm, &msi);
3048 #ifdef __KVM_HAVE_IRQ_LINE
3049 case KVM_IRQ_LINE_STATUS:
3050 case KVM_IRQ_LINE: {
3051 struct kvm_irq_level irq_event;
3054 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
3057 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
3058 ioctl == KVM_IRQ_LINE_STATUS);
3063 if (ioctl == KVM_IRQ_LINE_STATUS) {
3064 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
3072 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3073 case KVM_SET_GSI_ROUTING: {
3074 struct kvm_irq_routing routing;
3075 struct kvm_irq_routing __user *urouting;
3076 struct kvm_irq_routing_entry *entries = NULL;
3079 if (copy_from_user(&routing, argp, sizeof(routing)))
3082 if (!kvm_arch_can_set_irq_routing(kvm))
3084 if (routing.nr > KVM_MAX_IRQ_ROUTES)
3090 entries = vmalloc(routing.nr * sizeof(*entries));
3095 if (copy_from_user(entries, urouting->entries,
3096 routing.nr * sizeof(*entries)))
3097 goto out_free_irq_routing;
3099 r = kvm_set_irq_routing(kvm, entries, routing.nr,
3101 out_free_irq_routing:
3105 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
3106 case KVM_CREATE_DEVICE: {
3107 struct kvm_create_device cd;
3110 if (copy_from_user(&cd, argp, sizeof(cd)))
3113 r = kvm_ioctl_create_device(kvm, &cd);
3118 if (copy_to_user(argp, &cd, sizeof(cd)))
3124 case KVM_CHECK_EXTENSION:
3125 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
3128 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
3134 #ifdef CONFIG_KVM_COMPAT
3135 struct compat_kvm_dirty_log {
3139 compat_uptr_t dirty_bitmap; /* one bit per page */
3144 static long kvm_vm_compat_ioctl(struct file *filp,
3145 unsigned int ioctl, unsigned long arg)
3147 struct kvm *kvm = filp->private_data;
3150 if (kvm->mm != current->mm)
3153 case KVM_GET_DIRTY_LOG: {
3154 struct compat_kvm_dirty_log compat_log;
3155 struct kvm_dirty_log log;
3157 if (copy_from_user(&compat_log, (void __user *)arg,
3158 sizeof(compat_log)))
3160 log.slot = compat_log.slot;
3161 log.padding1 = compat_log.padding1;
3162 log.padding2 = compat_log.padding2;
3163 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
3165 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3169 r = kvm_vm_ioctl(filp, ioctl, arg);
3175 static struct file_operations kvm_vm_fops = {
3176 .release = kvm_vm_release,
3177 .unlocked_ioctl = kvm_vm_ioctl,
3178 #ifdef CONFIG_KVM_COMPAT
3179 .compat_ioctl = kvm_vm_compat_ioctl,
3181 .llseek = noop_llseek,
3184 static int kvm_dev_ioctl_create_vm(unsigned long type)
3190 kvm = kvm_create_vm(type);
3192 return PTR_ERR(kvm);
3193 #ifdef CONFIG_KVM_MMIO
3194 r = kvm_coalesced_mmio_init(kvm);
3200 r = get_unused_fd_flags(O_CLOEXEC);
3205 file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
3209 return PTR_ERR(file);
3213 * Don't call kvm_put_kvm anymore at this point; file->f_op is
3214 * already set, with ->release() being kvm_vm_release(). In error
3215 * cases it will be called by the final fput(file) and will take
3216 * care of doing kvm_put_kvm(kvm).
3218 if (kvm_create_vm_debugfs(kvm, r) < 0) {
3224 fd_install(r, file);
3228 static long kvm_dev_ioctl(struct file *filp,
3229 unsigned int ioctl, unsigned long arg)
3234 case KVM_GET_API_VERSION:
3237 r = KVM_API_VERSION;
3240 r = kvm_dev_ioctl_create_vm(arg);
3242 case KVM_CHECK_EXTENSION:
3243 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
3245 case KVM_GET_VCPU_MMAP_SIZE:
3248 r = PAGE_SIZE; /* struct kvm_run */
3250 r += PAGE_SIZE; /* pio data page */
3252 #ifdef CONFIG_KVM_MMIO
3253 r += PAGE_SIZE; /* coalesced mmio ring page */
3256 case KVM_TRACE_ENABLE:
3257 case KVM_TRACE_PAUSE:
3258 case KVM_TRACE_DISABLE:
3262 return kvm_arch_dev_ioctl(filp, ioctl, arg);
3268 static struct file_operations kvm_chardev_ops = {
3269 .unlocked_ioctl = kvm_dev_ioctl,
3270 .compat_ioctl = kvm_dev_ioctl,
3271 .llseek = noop_llseek,
3274 static struct miscdevice kvm_dev = {
3280 static void hardware_enable_nolock(void *junk)
3282 int cpu = raw_smp_processor_id();
3285 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
3288 cpumask_set_cpu(cpu, cpus_hardware_enabled);
3290 r = kvm_arch_hardware_enable();
3293 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3294 atomic_inc(&hardware_enable_failed);
3295 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
3299 static int kvm_starting_cpu(unsigned int cpu)
3301 raw_spin_lock(&kvm_count_lock);
3302 if (kvm_usage_count)
3303 hardware_enable_nolock(NULL);
3304 raw_spin_unlock(&kvm_count_lock);
3308 static void hardware_disable_nolock(void *junk)
3310 int cpu = raw_smp_processor_id();
3312 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
3314 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3315 kvm_arch_hardware_disable();
3318 static int kvm_dying_cpu(unsigned int cpu)
3320 raw_spin_lock(&kvm_count_lock);
3321 if (kvm_usage_count)
3322 hardware_disable_nolock(NULL);
3323 raw_spin_unlock(&kvm_count_lock);
3327 static void hardware_disable_all_nolock(void)
3329 BUG_ON(!kvm_usage_count);
3332 if (!kvm_usage_count)
3333 on_each_cpu(hardware_disable_nolock, NULL, 1);
3336 static void hardware_disable_all(void)
3338 raw_spin_lock(&kvm_count_lock);
3339 hardware_disable_all_nolock();
3340 raw_spin_unlock(&kvm_count_lock);
3343 static int hardware_enable_all(void)
3347 raw_spin_lock(&kvm_count_lock);
3350 if (kvm_usage_count == 1) {
3351 atomic_set(&hardware_enable_failed, 0);
3352 on_each_cpu(hardware_enable_nolock, NULL, 1);
3354 if (atomic_read(&hardware_enable_failed)) {
3355 hardware_disable_all_nolock();
3360 raw_spin_unlock(&kvm_count_lock);
3365 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3369 * Some (well, at least mine) BIOSes hang on reboot if
3372 * And Intel TXT required VMX off for all cpu when system shutdown.
3374 pr_info("kvm: exiting hardware virtualization\n");
3375 kvm_rebooting = true;
3376 on_each_cpu(hardware_disable_nolock, NULL, 1);
3380 static struct notifier_block kvm_reboot_notifier = {
3381 .notifier_call = kvm_reboot,
3385 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3389 for (i = 0; i < bus->dev_count; i++) {
3390 struct kvm_io_device *pos = bus->range[i].dev;
3392 kvm_iodevice_destructor(pos);
3397 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
3398 const struct kvm_io_range *r2)
3400 gpa_t addr1 = r1->addr;
3401 gpa_t addr2 = r2->addr;
3406 /* If r2->len == 0, match the exact address. If r2->len != 0,
3407 * accept any overlapping write. Any order is acceptable for
3408 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
3409 * we process all of them.
3422 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
3424 return kvm_io_bus_cmp(p1, p2);
3427 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
3428 gpa_t addr, int len)
3430 bus->range[bus->dev_count++] = (struct kvm_io_range) {
3436 sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
3437 kvm_io_bus_sort_cmp, NULL);
3442 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
3443 gpa_t addr, int len)
3445 struct kvm_io_range *range, key;
3448 key = (struct kvm_io_range) {
3453 range = bsearch(&key, bus->range, bus->dev_count,
3454 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
3458 off = range - bus->range;
3460 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
3466 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3467 struct kvm_io_range *range, const void *val)
3471 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3475 while (idx < bus->dev_count &&
3476 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3477 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
3486 /* kvm_io_bus_write - called under kvm->slots_lock */
3487 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3488 int len, const void *val)
3490 struct kvm_io_bus *bus;
3491 struct kvm_io_range range;
3494 range = (struct kvm_io_range) {
3499 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3502 r = __kvm_io_bus_write(vcpu, bus, &range, val);
3503 return r < 0 ? r : 0;
3506 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3507 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
3508 gpa_t addr, int len, const void *val, long cookie)
3510 struct kvm_io_bus *bus;
3511 struct kvm_io_range range;
3513 range = (struct kvm_io_range) {
3518 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3522 /* First try the device referenced by cookie. */
3523 if ((cookie >= 0) && (cookie < bus->dev_count) &&
3524 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
3525 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
3530 * cookie contained garbage; fall back to search and return the
3531 * correct cookie value.
3533 return __kvm_io_bus_write(vcpu, bus, &range, val);
3536 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3537 struct kvm_io_range *range, void *val)
3541 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3545 while (idx < bus->dev_count &&
3546 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3547 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
3555 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
3557 /* kvm_io_bus_read - called under kvm->slots_lock */
3558 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3561 struct kvm_io_bus *bus;
3562 struct kvm_io_range range;
3565 range = (struct kvm_io_range) {
3570 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3573 r = __kvm_io_bus_read(vcpu, bus, &range, val);
3574 return r < 0 ? r : 0;
3578 /* Caller must hold slots_lock. */
3579 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3580 int len, struct kvm_io_device *dev)
3582 struct kvm_io_bus *new_bus, *bus;
3584 bus = kvm_get_bus(kvm, bus_idx);
3588 /* exclude ioeventfd which is limited by maximum fd */
3589 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3592 new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count + 1) *
3593 sizeof(struct kvm_io_range)), GFP_KERNEL);
3596 memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
3597 sizeof(struct kvm_io_range)));
3598 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
3599 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3600 synchronize_srcu_expedited(&kvm->srcu);
3606 /* Caller must hold slots_lock. */
3607 void kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3608 struct kvm_io_device *dev)
3611 struct kvm_io_bus *new_bus, *bus;
3613 bus = kvm_get_bus(kvm, bus_idx);
3617 for (i = 0; i < bus->dev_count; i++)
3618 if (bus->range[i].dev == dev) {
3622 if (i == bus->dev_count)
3625 new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3626 sizeof(struct kvm_io_range)), GFP_KERNEL);
3628 pr_err("kvm: failed to shrink bus, removing it completely\n");
3632 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3633 new_bus->dev_count--;
3634 memcpy(new_bus->range + i, bus->range + i + 1,
3635 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3638 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3639 synchronize_srcu_expedited(&kvm->srcu);
3644 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3647 struct kvm_io_bus *bus;
3648 int dev_idx, srcu_idx;
3649 struct kvm_io_device *iodev = NULL;
3651 srcu_idx = srcu_read_lock(&kvm->srcu);
3653 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
3657 dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
3661 iodev = bus->range[dev_idx].dev;
3664 srcu_read_unlock(&kvm->srcu, srcu_idx);
3668 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
3670 static int kvm_debugfs_open(struct inode *inode, struct file *file,
3671 int (*get)(void *, u64 *), int (*set)(void *, u64),
3674 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
3677 /* The debugfs files are a reference to the kvm struct which
3678 * is still valid when kvm_destroy_vm is called.
3679 * To avoid the race between open and the removal of the debugfs
3680 * directory we test against the users count.
3682 if (!refcount_inc_not_zero(&stat_data->kvm->users_count))
3685 if (simple_attr_open(inode, file, get, set, fmt)) {
3686 kvm_put_kvm(stat_data->kvm);
3693 static int kvm_debugfs_release(struct inode *inode, struct file *file)
3695 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
3698 simple_attr_release(inode, file);
3699 kvm_put_kvm(stat_data->kvm);
3704 static int vm_stat_get_per_vm(void *data, u64 *val)
3706 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3708 *val = *(ulong *)((void *)stat_data->kvm + stat_data->offset);
3713 static int vm_stat_clear_per_vm(void *data, u64 val)
3715 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3720 *(ulong *)((void *)stat_data->kvm + stat_data->offset) = 0;
3725 static int vm_stat_get_per_vm_open(struct inode *inode, struct file *file)
3727 __simple_attr_check_format("%llu\n", 0ull);
3728 return kvm_debugfs_open(inode, file, vm_stat_get_per_vm,
3729 vm_stat_clear_per_vm, "%llu\n");
3732 static const struct file_operations vm_stat_get_per_vm_fops = {
3733 .owner = THIS_MODULE,
3734 .open = vm_stat_get_per_vm_open,
3735 .release = kvm_debugfs_release,
3736 .read = simple_attr_read,
3737 .write = simple_attr_write,
3738 .llseek = no_llseek,
3741 static int vcpu_stat_get_per_vm(void *data, u64 *val)
3744 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3745 struct kvm_vcpu *vcpu;
3749 kvm_for_each_vcpu(i, vcpu, stat_data->kvm)
3750 *val += *(u64 *)((void *)vcpu + stat_data->offset);
3755 static int vcpu_stat_clear_per_vm(void *data, u64 val)
3758 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3759 struct kvm_vcpu *vcpu;
3764 kvm_for_each_vcpu(i, vcpu, stat_data->kvm)
3765 *(u64 *)((void *)vcpu + stat_data->offset) = 0;
3770 static int vcpu_stat_get_per_vm_open(struct inode *inode, struct file *file)
3772 __simple_attr_check_format("%llu\n", 0ull);
3773 return kvm_debugfs_open(inode, file, vcpu_stat_get_per_vm,
3774 vcpu_stat_clear_per_vm, "%llu\n");
3777 static const struct file_operations vcpu_stat_get_per_vm_fops = {
3778 .owner = THIS_MODULE,
3779 .open = vcpu_stat_get_per_vm_open,
3780 .release = kvm_debugfs_release,
3781 .read = simple_attr_read,
3782 .write = simple_attr_write,
3783 .llseek = no_llseek,
3786 static const struct file_operations *stat_fops_per_vm[] = {
3787 [KVM_STAT_VCPU] = &vcpu_stat_get_per_vm_fops,
3788 [KVM_STAT_VM] = &vm_stat_get_per_vm_fops,
3791 static int vm_stat_get(void *_offset, u64 *val)
3793 unsigned offset = (long)_offset;
3795 struct kvm_stat_data stat_tmp = {.offset = offset};
3799 spin_lock(&kvm_lock);
3800 list_for_each_entry(kvm, &vm_list, vm_list) {
3802 vm_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
3805 spin_unlock(&kvm_lock);
3809 static int vm_stat_clear(void *_offset, u64 val)
3811 unsigned offset = (long)_offset;
3813 struct kvm_stat_data stat_tmp = {.offset = offset};
3818 spin_lock(&kvm_lock);
3819 list_for_each_entry(kvm, &vm_list, vm_list) {
3821 vm_stat_clear_per_vm((void *)&stat_tmp, 0);
3823 spin_unlock(&kvm_lock);
3828 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n");
3830 static int vcpu_stat_get(void *_offset, u64 *val)
3832 unsigned offset = (long)_offset;
3834 struct kvm_stat_data stat_tmp = {.offset = offset};
3838 spin_lock(&kvm_lock);
3839 list_for_each_entry(kvm, &vm_list, vm_list) {
3841 vcpu_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
3844 spin_unlock(&kvm_lock);
3848 static int vcpu_stat_clear(void *_offset, u64 val)
3850 unsigned offset = (long)_offset;
3852 struct kvm_stat_data stat_tmp = {.offset = offset};
3857 spin_lock(&kvm_lock);
3858 list_for_each_entry(kvm, &vm_list, vm_list) {
3860 vcpu_stat_clear_per_vm((void *)&stat_tmp, 0);
3862 spin_unlock(&kvm_lock);
3867 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear,
3870 static const struct file_operations *stat_fops[] = {
3871 [KVM_STAT_VCPU] = &vcpu_stat_fops,
3872 [KVM_STAT_VM] = &vm_stat_fops,
3875 static int kvm_init_debug(void)
3878 struct kvm_stats_debugfs_item *p;
3880 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3881 if (kvm_debugfs_dir == NULL)
3884 kvm_debugfs_num_entries = 0;
3885 for (p = debugfs_entries; p->name; ++p, kvm_debugfs_num_entries++) {
3886 if (!debugfs_create_file(p->name, 0644, kvm_debugfs_dir,
3887 (void *)(long)p->offset,
3888 stat_fops[p->kind]))
3895 debugfs_remove_recursive(kvm_debugfs_dir);
3900 static int kvm_suspend(void)
3902 if (kvm_usage_count)
3903 hardware_disable_nolock(NULL);
3907 static void kvm_resume(void)
3909 if (kvm_usage_count) {
3910 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3911 hardware_enable_nolock(NULL);
3915 static struct syscore_ops kvm_syscore_ops = {
3916 .suspend = kvm_suspend,
3917 .resume = kvm_resume,
3921 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3923 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3926 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3928 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3930 if (vcpu->preempted)
3931 vcpu->preempted = false;
3933 kvm_arch_sched_in(vcpu, cpu);
3935 kvm_arch_vcpu_load(vcpu, cpu);
3938 static void kvm_sched_out(struct preempt_notifier *pn,
3939 struct task_struct *next)
3941 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3943 if (current->state == TASK_RUNNING)
3944 vcpu->preempted = true;
3945 kvm_arch_vcpu_put(vcpu);
3948 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3949 struct module *module)
3954 r = kvm_arch_init(opaque);
3959 * kvm_arch_init makes sure there's at most one caller
3960 * for architectures that support multiple implementations,
3961 * like intel and amd on x86.
3962 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3963 * conflicts in case kvm is already setup for another implementation.
3965 r = kvm_irqfd_init();
3969 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
3974 r = kvm_arch_hardware_setup();
3978 for_each_online_cpu(cpu) {
3979 smp_call_function_single(cpu,
3980 kvm_arch_check_processor_compat,
3986 r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting",
3987 kvm_starting_cpu, kvm_dying_cpu);
3990 register_reboot_notifier(&kvm_reboot_notifier);
3992 /* A kmem cache lets us meet the alignment requirements of fx_save. */
3994 vcpu_align = __alignof__(struct kvm_vcpu);
3995 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
3997 if (!kvm_vcpu_cache) {
4002 r = kvm_async_pf_init();
4006 kvm_chardev_ops.owner = module;
4007 kvm_vm_fops.owner = module;
4008 kvm_vcpu_fops.owner = module;
4010 r = misc_register(&kvm_dev);
4012 pr_err("kvm: misc device register failed\n");
4016 register_syscore_ops(&kvm_syscore_ops);
4018 kvm_preempt_ops.sched_in = kvm_sched_in;
4019 kvm_preempt_ops.sched_out = kvm_sched_out;
4021 r = kvm_init_debug();
4023 pr_err("kvm: create debugfs files failed\n");
4027 r = kvm_vfio_ops_init();
4033 unregister_syscore_ops(&kvm_syscore_ops);
4034 misc_deregister(&kvm_dev);
4036 kvm_async_pf_deinit();
4038 kmem_cache_destroy(kvm_vcpu_cache);
4040 unregister_reboot_notifier(&kvm_reboot_notifier);
4041 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4044 kvm_arch_hardware_unsetup();
4046 free_cpumask_var(cpus_hardware_enabled);
4054 EXPORT_SYMBOL_GPL(kvm_init);
4058 debugfs_remove_recursive(kvm_debugfs_dir);
4059 misc_deregister(&kvm_dev);
4060 kmem_cache_destroy(kvm_vcpu_cache);
4061 kvm_async_pf_deinit();
4062 unregister_syscore_ops(&kvm_syscore_ops);
4063 unregister_reboot_notifier(&kvm_reboot_notifier);
4064 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4065 on_each_cpu(hardware_disable_nolock, NULL, 1);
4066 kvm_arch_hardware_unsetup();
4069 free_cpumask_var(cpus_hardware_enabled);
4070 kvm_vfio_ops_exit();
4072 EXPORT_SYMBOL_GPL(kvm_exit);