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, S_IRUGO | S_IWUSR);
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, S_IRUGO | S_IWUSR);
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, S_IRUGO | S_IWUSR);
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 static void ack_flush(void *_completed)
172 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
177 struct kvm_vcpu *vcpu;
179 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
182 kvm_for_each_vcpu(i, vcpu, kvm) {
183 kvm_make_request(req, vcpu);
186 /* Set ->requests bit before we read ->mode. */
187 smp_mb__after_atomic();
189 if (cpus != NULL && cpu != -1 && cpu != me &&
190 kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
191 cpumask_set_cpu(cpu, cpus);
193 if (unlikely(cpus == NULL))
194 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
195 else if (!cpumask_empty(cpus))
196 smp_call_function_many(cpus, ack_flush, NULL, 1);
200 free_cpumask_var(cpus);
204 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
205 void kvm_flush_remote_tlbs(struct kvm *kvm)
208 * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in
209 * kvm_make_all_cpus_request.
211 long dirty_count = smp_load_acquire(&kvm->tlbs_dirty);
214 * We want to publish modifications to the page tables before reading
215 * mode. Pairs with a memory barrier in arch-specific code.
216 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
217 * and smp_mb in walk_shadow_page_lockless_begin/end.
218 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
220 * There is already an smp_mb__after_atomic() before
221 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
224 if (kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
225 ++kvm->stat.remote_tlb_flush;
226 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
228 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
231 void kvm_reload_remote_mmus(struct kvm *kvm)
233 kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
236 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
241 mutex_init(&vcpu->mutex);
246 init_swait_queue_head(&vcpu->wq);
247 kvm_async_pf_vcpu_init(vcpu);
250 INIT_LIST_HEAD(&vcpu->blocked_vcpu_list);
252 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
257 vcpu->run = page_address(page);
259 kvm_vcpu_set_in_spin_loop(vcpu, false);
260 kvm_vcpu_set_dy_eligible(vcpu, false);
261 vcpu->preempted = false;
263 r = kvm_arch_vcpu_init(vcpu);
269 free_page((unsigned long)vcpu->run);
273 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
275 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
278 kvm_arch_vcpu_uninit(vcpu);
279 free_page((unsigned long)vcpu->run);
281 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
283 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
284 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
286 return container_of(mn, struct kvm, mmu_notifier);
289 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
290 struct mm_struct *mm,
291 unsigned long address)
293 struct kvm *kvm = mmu_notifier_to_kvm(mn);
294 int need_tlb_flush, idx;
297 * When ->invalidate_page runs, the linux pte has been zapped
298 * already but the page is still allocated until
299 * ->invalidate_page returns. So if we increase the sequence
300 * here the kvm page fault will notice if the spte can't be
301 * established because the page is going to be freed. If
302 * instead the kvm page fault establishes the spte before
303 * ->invalidate_page runs, kvm_unmap_hva will release it
306 * The sequence increase only need to be seen at spin_unlock
307 * time, and not at spin_lock time.
309 * Increasing the sequence after the spin_unlock would be
310 * unsafe because the kvm page fault could then establish the
311 * pte after kvm_unmap_hva returned, without noticing the page
312 * is going to be freed.
314 idx = srcu_read_lock(&kvm->srcu);
315 spin_lock(&kvm->mmu_lock);
317 kvm->mmu_notifier_seq++;
318 need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
319 /* we've to flush the tlb before the pages can be freed */
321 kvm_flush_remote_tlbs(kvm);
323 spin_unlock(&kvm->mmu_lock);
325 kvm_arch_mmu_notifier_invalidate_page(kvm, address);
327 srcu_read_unlock(&kvm->srcu, idx);
330 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
331 struct mm_struct *mm,
332 unsigned long address,
335 struct kvm *kvm = mmu_notifier_to_kvm(mn);
338 idx = srcu_read_lock(&kvm->srcu);
339 spin_lock(&kvm->mmu_lock);
340 kvm->mmu_notifier_seq++;
341 kvm_set_spte_hva(kvm, address, pte);
342 spin_unlock(&kvm->mmu_lock);
343 srcu_read_unlock(&kvm->srcu, idx);
346 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
347 struct mm_struct *mm,
351 struct kvm *kvm = mmu_notifier_to_kvm(mn);
352 int need_tlb_flush = 0, idx;
354 idx = srcu_read_lock(&kvm->srcu);
355 spin_lock(&kvm->mmu_lock);
357 * The count increase must become visible at unlock time as no
358 * spte can be established without taking the mmu_lock and
359 * count is also read inside the mmu_lock critical section.
361 kvm->mmu_notifier_count++;
362 need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
363 need_tlb_flush |= kvm->tlbs_dirty;
364 /* we've to flush the tlb before the pages can be freed */
366 kvm_flush_remote_tlbs(kvm);
368 spin_unlock(&kvm->mmu_lock);
369 srcu_read_unlock(&kvm->srcu, idx);
372 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
373 struct mm_struct *mm,
377 struct kvm *kvm = mmu_notifier_to_kvm(mn);
379 spin_lock(&kvm->mmu_lock);
381 * This sequence increase will notify the kvm page fault that
382 * the page that is going to be mapped in the spte could have
385 kvm->mmu_notifier_seq++;
388 * The above sequence increase must be visible before the
389 * below count decrease, which is ensured by the smp_wmb above
390 * in conjunction with the smp_rmb in mmu_notifier_retry().
392 kvm->mmu_notifier_count--;
393 spin_unlock(&kvm->mmu_lock);
395 BUG_ON(kvm->mmu_notifier_count < 0);
398 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
399 struct mm_struct *mm,
403 struct kvm *kvm = mmu_notifier_to_kvm(mn);
406 idx = srcu_read_lock(&kvm->srcu);
407 spin_lock(&kvm->mmu_lock);
409 young = kvm_age_hva(kvm, start, end);
411 kvm_flush_remote_tlbs(kvm);
413 spin_unlock(&kvm->mmu_lock);
414 srcu_read_unlock(&kvm->srcu, idx);
419 static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
420 struct mm_struct *mm,
424 struct kvm *kvm = mmu_notifier_to_kvm(mn);
427 idx = srcu_read_lock(&kvm->srcu);
428 spin_lock(&kvm->mmu_lock);
430 * Even though we do not flush TLB, this will still adversely
431 * affect performance on pre-Haswell Intel EPT, where there is
432 * no EPT Access Bit to clear so that we have to tear down EPT
433 * tables instead. If we find this unacceptable, we can always
434 * add a parameter to kvm_age_hva so that it effectively doesn't
435 * do anything on clear_young.
437 * Also note that currently we never issue secondary TLB flushes
438 * from clear_young, leaving this job up to the regular system
439 * cadence. If we find this inaccurate, we might come up with a
440 * more sophisticated heuristic later.
442 young = kvm_age_hva(kvm, start, end);
443 spin_unlock(&kvm->mmu_lock);
444 srcu_read_unlock(&kvm->srcu, idx);
449 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
450 struct mm_struct *mm,
451 unsigned long address)
453 struct kvm *kvm = mmu_notifier_to_kvm(mn);
456 idx = srcu_read_lock(&kvm->srcu);
457 spin_lock(&kvm->mmu_lock);
458 young = kvm_test_age_hva(kvm, address);
459 spin_unlock(&kvm->mmu_lock);
460 srcu_read_unlock(&kvm->srcu, idx);
465 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
466 struct mm_struct *mm)
468 struct kvm *kvm = mmu_notifier_to_kvm(mn);
471 idx = srcu_read_lock(&kvm->srcu);
472 kvm_arch_flush_shadow_all(kvm);
473 srcu_read_unlock(&kvm->srcu, idx);
476 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
477 .invalidate_page = kvm_mmu_notifier_invalidate_page,
478 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
479 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
480 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
481 .clear_young = kvm_mmu_notifier_clear_young,
482 .test_young = kvm_mmu_notifier_test_young,
483 .change_pte = kvm_mmu_notifier_change_pte,
484 .release = kvm_mmu_notifier_release,
487 static int kvm_init_mmu_notifier(struct kvm *kvm)
489 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
490 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
493 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
495 static int kvm_init_mmu_notifier(struct kvm *kvm)
500 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
502 static struct kvm_memslots *kvm_alloc_memslots(void)
505 struct kvm_memslots *slots;
507 slots = kvm_kvzalloc(sizeof(struct kvm_memslots));
511 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
512 slots->id_to_index[i] = slots->memslots[i].id = i;
517 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
519 if (!memslot->dirty_bitmap)
522 kvfree(memslot->dirty_bitmap);
523 memslot->dirty_bitmap = NULL;
527 * Free any memory in @free but not in @dont.
529 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
530 struct kvm_memory_slot *dont)
532 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
533 kvm_destroy_dirty_bitmap(free);
535 kvm_arch_free_memslot(kvm, free, dont);
540 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
542 struct kvm_memory_slot *memslot;
547 kvm_for_each_memslot(memslot, slots)
548 kvm_free_memslot(kvm, memslot, NULL);
553 static void kvm_destroy_vm_debugfs(struct kvm *kvm)
557 if (!kvm->debugfs_dentry)
560 debugfs_remove_recursive(kvm->debugfs_dentry);
562 if (kvm->debugfs_stat_data) {
563 for (i = 0; i < kvm_debugfs_num_entries; i++)
564 kfree(kvm->debugfs_stat_data[i]);
565 kfree(kvm->debugfs_stat_data);
569 static int kvm_create_vm_debugfs(struct kvm *kvm, int fd)
571 char dir_name[ITOA_MAX_LEN * 2];
572 struct kvm_stat_data *stat_data;
573 struct kvm_stats_debugfs_item *p;
575 if (!debugfs_initialized())
578 snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd);
579 kvm->debugfs_dentry = debugfs_create_dir(dir_name,
581 if (!kvm->debugfs_dentry)
584 kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries,
585 sizeof(*kvm->debugfs_stat_data),
587 if (!kvm->debugfs_stat_data)
590 for (p = debugfs_entries; p->name; p++) {
591 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL);
595 stat_data->kvm = kvm;
596 stat_data->offset = p->offset;
597 kvm->debugfs_stat_data[p - debugfs_entries] = stat_data;
598 if (!debugfs_create_file(p->name, 0644,
601 stat_fops_per_vm[p->kind]))
607 static struct kvm *kvm_create_vm(unsigned long type)
610 struct kvm *kvm = kvm_arch_alloc_vm();
613 return ERR_PTR(-ENOMEM);
615 spin_lock_init(&kvm->mmu_lock);
617 kvm->mm = current->mm;
618 kvm_eventfd_init(kvm);
619 mutex_init(&kvm->lock);
620 mutex_init(&kvm->irq_lock);
621 mutex_init(&kvm->slots_lock);
622 refcount_set(&kvm->users_count, 1);
623 INIT_LIST_HEAD(&kvm->devices);
625 r = kvm_arch_init_vm(kvm, type);
627 goto out_err_no_disable;
629 r = hardware_enable_all();
631 goto out_err_no_disable;
633 #ifdef CONFIG_HAVE_KVM_IRQFD
634 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
637 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
640 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
641 struct kvm_memslots *slots = kvm_alloc_memslots();
643 goto out_err_no_srcu;
645 * Generations must be different for each address space.
646 * Init kvm generation close to the maximum to easily test the
647 * code of handling generation number wrap-around.
649 slots->generation = i * 2 - 150;
650 rcu_assign_pointer(kvm->memslots[i], slots);
653 if (init_srcu_struct(&kvm->srcu))
654 goto out_err_no_srcu;
655 if (init_srcu_struct(&kvm->irq_srcu))
656 goto out_err_no_irq_srcu;
657 for (i = 0; i < KVM_NR_BUSES; i++) {
658 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
664 r = kvm_init_mmu_notifier(kvm);
668 spin_lock(&kvm_lock);
669 list_add(&kvm->vm_list, &vm_list);
670 spin_unlock(&kvm_lock);
672 preempt_notifier_inc();
677 cleanup_srcu_struct(&kvm->irq_srcu);
679 cleanup_srcu_struct(&kvm->srcu);
681 hardware_disable_all();
683 for (i = 0; i < KVM_NR_BUSES; i++)
684 kfree(kvm->buses[i]);
685 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
686 kvm_free_memslots(kvm, kvm->memslots[i]);
687 kvm_arch_free_vm(kvm);
693 * Avoid using vmalloc for a small buffer.
694 * Should not be used when the size is statically known.
696 void *kvm_kvzalloc(unsigned long size)
698 if (size > PAGE_SIZE)
699 return vzalloc(size);
701 return kzalloc(size, GFP_KERNEL);
704 static void kvm_destroy_devices(struct kvm *kvm)
706 struct kvm_device *dev, *tmp;
709 * We do not need to take the kvm->lock here, because nobody else
710 * has a reference to the struct kvm at this point and therefore
711 * cannot access the devices list anyhow.
713 list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
714 list_del(&dev->vm_node);
715 dev->ops->destroy(dev);
719 static void kvm_destroy_vm(struct kvm *kvm)
722 struct mm_struct *mm = kvm->mm;
724 kvm_destroy_vm_debugfs(kvm);
725 kvm_arch_sync_events(kvm);
726 spin_lock(&kvm_lock);
727 list_del(&kvm->vm_list);
728 spin_unlock(&kvm_lock);
729 kvm_free_irq_routing(kvm);
730 for (i = 0; i < KVM_NR_BUSES; i++) {
732 kvm_io_bus_destroy(kvm->buses[i]);
733 kvm->buses[i] = NULL;
735 kvm_coalesced_mmio_free(kvm);
736 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
737 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
739 kvm_arch_flush_shadow_all(kvm);
741 kvm_arch_destroy_vm(kvm);
742 kvm_destroy_devices(kvm);
743 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
744 kvm_free_memslots(kvm, kvm->memslots[i]);
745 cleanup_srcu_struct(&kvm->irq_srcu);
746 cleanup_srcu_struct(&kvm->srcu);
747 kvm_arch_free_vm(kvm);
748 preempt_notifier_dec();
749 hardware_disable_all();
753 void kvm_get_kvm(struct kvm *kvm)
755 refcount_inc(&kvm->users_count);
757 EXPORT_SYMBOL_GPL(kvm_get_kvm);
759 void kvm_put_kvm(struct kvm *kvm)
761 if (refcount_dec_and_test(&kvm->users_count))
764 EXPORT_SYMBOL_GPL(kvm_put_kvm);
767 static int kvm_vm_release(struct inode *inode, struct file *filp)
769 struct kvm *kvm = filp->private_data;
771 kvm_irqfd_release(kvm);
778 * Allocation size is twice as large as the actual dirty bitmap size.
779 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
781 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
783 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
785 memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
786 if (!memslot->dirty_bitmap)
793 * Insert memslot and re-sort memslots based on their GFN,
794 * so binary search could be used to lookup GFN.
795 * Sorting algorithm takes advantage of having initially
796 * sorted array and known changed memslot position.
798 static void update_memslots(struct kvm_memslots *slots,
799 struct kvm_memory_slot *new)
802 int i = slots->id_to_index[id];
803 struct kvm_memory_slot *mslots = slots->memslots;
805 WARN_ON(mslots[i].id != id);
807 WARN_ON(!mslots[i].npages);
808 if (mslots[i].npages)
811 if (!mslots[i].npages)
815 while (i < KVM_MEM_SLOTS_NUM - 1 &&
816 new->base_gfn <= mslots[i + 1].base_gfn) {
817 if (!mslots[i + 1].npages)
819 mslots[i] = mslots[i + 1];
820 slots->id_to_index[mslots[i].id] = i;
825 * The ">=" is needed when creating a slot with base_gfn == 0,
826 * so that it moves before all those with base_gfn == npages == 0.
828 * On the other hand, if new->npages is zero, the above loop has
829 * already left i pointing to the beginning of the empty part of
830 * mslots, and the ">=" would move the hole backwards in this
831 * case---which is wrong. So skip the loop when deleting a slot.
835 new->base_gfn >= mslots[i - 1].base_gfn) {
836 mslots[i] = mslots[i - 1];
837 slots->id_to_index[mslots[i].id] = i;
841 WARN_ON_ONCE(i != slots->used_slots);
844 slots->id_to_index[mslots[i].id] = i;
847 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
849 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
851 #ifdef __KVM_HAVE_READONLY_MEM
852 valid_flags |= KVM_MEM_READONLY;
855 if (mem->flags & ~valid_flags)
861 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
862 int as_id, struct kvm_memslots *slots)
864 struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id);
867 * Set the low bit in the generation, which disables SPTE caching
868 * until the end of synchronize_srcu_expedited.
870 WARN_ON(old_memslots->generation & 1);
871 slots->generation = old_memslots->generation + 1;
873 rcu_assign_pointer(kvm->memslots[as_id], slots);
874 synchronize_srcu_expedited(&kvm->srcu);
877 * Increment the new memslot generation a second time. This prevents
878 * vm exits that race with memslot updates from caching a memslot
879 * generation that will (potentially) be valid forever.
881 * Generations must be unique even across address spaces. We do not need
882 * a global counter for that, instead the generation space is evenly split
883 * across address spaces. For example, with two address spaces, address
884 * space 0 will use generations 0, 4, 8, ... while * address space 1 will
885 * use generations 2, 6, 10, 14, ...
887 slots->generation += KVM_ADDRESS_SPACE_NUM * 2 - 1;
889 kvm_arch_memslots_updated(kvm, slots);
895 * Allocate some memory and give it an address in the guest physical address
898 * Discontiguous memory is allowed, mostly for framebuffers.
900 * Must be called holding kvm->slots_lock for write.
902 int __kvm_set_memory_region(struct kvm *kvm,
903 const struct kvm_userspace_memory_region *mem)
907 unsigned long npages;
908 struct kvm_memory_slot *slot;
909 struct kvm_memory_slot old, new;
910 struct kvm_memslots *slots = NULL, *old_memslots;
912 enum kvm_mr_change change;
914 r = check_memory_region_flags(mem);
919 as_id = mem->slot >> 16;
922 /* General sanity checks */
923 if (mem->memory_size & (PAGE_SIZE - 1))
925 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
927 /* We can read the guest memory with __xxx_user() later on. */
928 if ((id < KVM_USER_MEM_SLOTS) &&
929 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
930 !access_ok(VERIFY_WRITE,
931 (void __user *)(unsigned long)mem->userspace_addr,
934 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
936 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
939 slot = id_to_memslot(__kvm_memslots(kvm, as_id), id);
940 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
941 npages = mem->memory_size >> PAGE_SHIFT;
943 if (npages > KVM_MEM_MAX_NR_PAGES)
949 new.base_gfn = base_gfn;
951 new.flags = mem->flags;
955 change = KVM_MR_CREATE;
956 else { /* Modify an existing slot. */
957 if ((mem->userspace_addr != old.userspace_addr) ||
958 (npages != old.npages) ||
959 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
962 if (base_gfn != old.base_gfn)
963 change = KVM_MR_MOVE;
964 else if (new.flags != old.flags)
965 change = KVM_MR_FLAGS_ONLY;
966 else { /* Nothing to change. */
975 change = KVM_MR_DELETE;
980 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
981 /* Check for overlaps */
983 kvm_for_each_memslot(slot, __kvm_memslots(kvm, as_id)) {
984 if ((slot->id >= KVM_USER_MEM_SLOTS) ||
987 if (!((base_gfn + npages <= slot->base_gfn) ||
988 (base_gfn >= slot->base_gfn + slot->npages)))
993 /* Free page dirty bitmap if unneeded */
994 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
995 new.dirty_bitmap = NULL;
998 if (change == KVM_MR_CREATE) {
999 new.userspace_addr = mem->userspace_addr;
1001 if (kvm_arch_create_memslot(kvm, &new, npages))
1005 /* Allocate page dirty bitmap if needed */
1006 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
1007 if (kvm_create_dirty_bitmap(&new) < 0)
1011 slots = kvm_kvzalloc(sizeof(struct kvm_memslots));
1014 memcpy(slots, __kvm_memslots(kvm, as_id), sizeof(struct kvm_memslots));
1016 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
1017 slot = id_to_memslot(slots, id);
1018 slot->flags |= KVM_MEMSLOT_INVALID;
1020 old_memslots = install_new_memslots(kvm, as_id, slots);
1022 /* slot was deleted or moved, clear iommu mapping */
1023 kvm_iommu_unmap_pages(kvm, &old);
1024 /* From this point no new shadow pages pointing to a deleted,
1025 * or moved, memslot will be created.
1027 * validation of sp->gfn happens in:
1028 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1029 * - kvm_is_visible_gfn (mmu_check_roots)
1031 kvm_arch_flush_shadow_memslot(kvm, slot);
1034 * We can re-use the old_memslots from above, the only difference
1035 * from the currently installed memslots is the invalid flag. This
1036 * will get overwritten by update_memslots anyway.
1038 slots = old_memslots;
1041 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
1045 /* actual memory is freed via old in kvm_free_memslot below */
1046 if (change == KVM_MR_DELETE) {
1047 new.dirty_bitmap = NULL;
1048 memset(&new.arch, 0, sizeof(new.arch));
1051 update_memslots(slots, &new);
1052 old_memslots = install_new_memslots(kvm, as_id, slots);
1054 kvm_arch_commit_memory_region(kvm, mem, &old, &new, change);
1056 kvm_free_memslot(kvm, &old, &new);
1057 kvfree(old_memslots);
1060 * IOMMU mapping: New slots need to be mapped. Old slots need to be
1061 * un-mapped and re-mapped if their base changes. Since base change
1062 * unmapping is handled above with slot deletion, mapping alone is
1063 * needed here. Anything else the iommu might care about for existing
1064 * slots (size changes, userspace addr changes and read-only flag
1065 * changes) is disallowed above, so any other attribute changes getting
1066 * here can be skipped.
1068 if (as_id == 0 && (change == KVM_MR_CREATE || change == KVM_MR_MOVE)) {
1069 r = kvm_iommu_map_pages(kvm, &new);
1078 kvm_free_memslot(kvm, &new, &old);
1082 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1084 int kvm_set_memory_region(struct kvm *kvm,
1085 const struct kvm_userspace_memory_region *mem)
1089 mutex_lock(&kvm->slots_lock);
1090 r = __kvm_set_memory_region(kvm, mem);
1091 mutex_unlock(&kvm->slots_lock);
1094 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1096 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1097 struct kvm_userspace_memory_region *mem)
1099 if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1102 return kvm_set_memory_region(kvm, mem);
1105 int kvm_get_dirty_log(struct kvm *kvm,
1106 struct kvm_dirty_log *log, int *is_dirty)
1108 struct kvm_memslots *slots;
1109 struct kvm_memory_slot *memslot;
1112 unsigned long any = 0;
1114 as_id = log->slot >> 16;
1115 id = (u16)log->slot;
1116 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1119 slots = __kvm_memslots(kvm, as_id);
1120 memslot = id_to_memslot(slots, id);
1121 if (!memslot->dirty_bitmap)
1124 n = kvm_dirty_bitmap_bytes(memslot);
1126 for (i = 0; !any && i < n/sizeof(long); ++i)
1127 any = memslot->dirty_bitmap[i];
1129 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1136 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1138 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1140 * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages
1141 * are dirty write protect them for next write.
1142 * @kvm: pointer to kvm instance
1143 * @log: slot id and address to which we copy the log
1144 * @is_dirty: flag set if any page is dirty
1146 * We need to keep it in mind that VCPU threads can write to the bitmap
1147 * concurrently. So, to avoid losing track of dirty pages we keep the
1150 * 1. Take a snapshot of the bit and clear it if needed.
1151 * 2. Write protect the corresponding page.
1152 * 3. Copy the snapshot to the userspace.
1153 * 4. Upon return caller flushes TLB's if needed.
1155 * Between 2 and 4, the guest may write to the page using the remaining TLB
1156 * entry. This is not a problem because the page is reported dirty using
1157 * the snapshot taken before and step 4 ensures that writes done after
1158 * exiting to userspace will be logged for the next call.
1161 int kvm_get_dirty_log_protect(struct kvm *kvm,
1162 struct kvm_dirty_log *log, bool *is_dirty)
1164 struct kvm_memslots *slots;
1165 struct kvm_memory_slot *memslot;
1168 unsigned long *dirty_bitmap;
1169 unsigned long *dirty_bitmap_buffer;
1171 as_id = log->slot >> 16;
1172 id = (u16)log->slot;
1173 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1176 slots = __kvm_memslots(kvm, as_id);
1177 memslot = id_to_memslot(slots, id);
1179 dirty_bitmap = memslot->dirty_bitmap;
1183 n = kvm_dirty_bitmap_bytes(memslot);
1185 dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long);
1186 memset(dirty_bitmap_buffer, 0, n);
1188 spin_lock(&kvm->mmu_lock);
1190 for (i = 0; i < n / sizeof(long); i++) {
1194 if (!dirty_bitmap[i])
1199 mask = xchg(&dirty_bitmap[i], 0);
1200 dirty_bitmap_buffer[i] = mask;
1203 offset = i * BITS_PER_LONG;
1204 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1209 spin_unlock(&kvm->mmu_lock);
1210 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1214 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect);
1217 bool kvm_largepages_enabled(void)
1219 return largepages_enabled;
1222 void kvm_disable_largepages(void)
1224 largepages_enabled = false;
1226 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1228 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1230 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1232 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1234 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
1236 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn);
1239 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1241 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1243 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1244 memslot->flags & KVM_MEMSLOT_INVALID)
1249 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1251 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1253 struct vm_area_struct *vma;
1254 unsigned long addr, size;
1258 addr = gfn_to_hva(kvm, gfn);
1259 if (kvm_is_error_hva(addr))
1262 down_read(¤t->mm->mmap_sem);
1263 vma = find_vma(current->mm, addr);
1267 size = vma_kernel_pagesize(vma);
1270 up_read(¤t->mm->mmap_sem);
1275 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1277 return slot->flags & KVM_MEM_READONLY;
1280 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1281 gfn_t *nr_pages, bool write)
1283 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1284 return KVM_HVA_ERR_BAD;
1286 if (memslot_is_readonly(slot) && write)
1287 return KVM_HVA_ERR_RO_BAD;
1290 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1292 return __gfn_to_hva_memslot(slot, gfn);
1295 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1298 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1301 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1304 return gfn_to_hva_many(slot, gfn, NULL);
1306 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1308 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1310 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1312 EXPORT_SYMBOL_GPL(gfn_to_hva);
1314 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
1316 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
1318 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
1321 * If writable is set to false, the hva returned by this function is only
1322 * allowed to be read.
1324 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1325 gfn_t gfn, bool *writable)
1327 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1329 if (!kvm_is_error_hva(hva) && writable)
1330 *writable = !memslot_is_readonly(slot);
1335 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1337 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1339 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1342 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
1344 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1346 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1349 static int get_user_page_nowait(unsigned long start, int write,
1352 int flags = FOLL_NOWAIT | FOLL_HWPOISON;
1355 flags |= FOLL_WRITE;
1357 return get_user_pages(start, 1, flags, page, NULL);
1360 static inline int check_user_page_hwpoison(unsigned long addr)
1362 int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
1364 rc = get_user_pages(addr, 1, flags, NULL, NULL);
1365 return rc == -EHWPOISON;
1369 * The atomic path to get the writable pfn which will be stored in @pfn,
1370 * true indicates success, otherwise false is returned.
1372 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1373 bool write_fault, bool *writable, kvm_pfn_t *pfn)
1375 struct page *page[1];
1378 if (!(async || atomic))
1382 * Fast pin a writable pfn only if it is a write fault request
1383 * or the caller allows to map a writable pfn for a read fault
1386 if (!(write_fault || writable))
1389 npages = __get_user_pages_fast(addr, 1, 1, page);
1391 *pfn = page_to_pfn(page[0]);
1402 * The slow path to get the pfn of the specified host virtual address,
1403 * 1 indicates success, -errno is returned if error is detected.
1405 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1406 bool *writable, kvm_pfn_t *pfn)
1408 struct page *page[1];
1414 *writable = write_fault;
1417 down_read(¤t->mm->mmap_sem);
1418 npages = get_user_page_nowait(addr, write_fault, page);
1419 up_read(¤t->mm->mmap_sem);
1421 unsigned int flags = FOLL_HWPOISON;
1424 flags |= FOLL_WRITE;
1426 npages = get_user_pages_unlocked(addr, 1, page, flags);
1431 /* map read fault as writable if possible */
1432 if (unlikely(!write_fault) && writable) {
1433 struct page *wpage[1];
1435 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1444 *pfn = page_to_pfn(page[0]);
1448 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1450 if (unlikely(!(vma->vm_flags & VM_READ)))
1453 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1459 static int hva_to_pfn_remapped(struct vm_area_struct *vma,
1460 unsigned long addr, bool *async,
1461 bool write_fault, kvm_pfn_t *p_pfn)
1466 r = follow_pfn(vma, addr, &pfn);
1469 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
1470 * not call the fault handler, so do it here.
1472 bool unlocked = false;
1473 r = fixup_user_fault(current, current->mm, addr,
1474 (write_fault ? FAULT_FLAG_WRITE : 0),
1481 r = follow_pfn(vma, addr, &pfn);
1489 * Get a reference here because callers of *hva_to_pfn* and
1490 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
1491 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
1492 * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will
1493 * simply do nothing for reserved pfns.
1495 * Whoever called remap_pfn_range is also going to call e.g.
1496 * unmap_mapping_range before the underlying pages are freed,
1497 * causing a call to our MMU notifier.
1506 * Pin guest page in memory and return its pfn.
1507 * @addr: host virtual address which maps memory to the guest
1508 * @atomic: whether this function can sleep
1509 * @async: whether this function need to wait IO complete if the
1510 * host page is not in the memory
1511 * @write_fault: whether we should get a writable host page
1512 * @writable: whether it allows to map a writable host page for !@write_fault
1514 * The function will map a writable host page for these two cases:
1515 * 1): @write_fault = true
1516 * 2): @write_fault = false && @writable, @writable will tell the caller
1517 * whether the mapping is writable.
1519 static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1520 bool write_fault, bool *writable)
1522 struct vm_area_struct *vma;
1526 /* we can do it either atomically or asynchronously, not both */
1527 BUG_ON(atomic && async);
1529 if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1533 return KVM_PFN_ERR_FAULT;
1535 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1539 down_read(¤t->mm->mmap_sem);
1540 if (npages == -EHWPOISON ||
1541 (!async && check_user_page_hwpoison(addr))) {
1542 pfn = KVM_PFN_ERR_HWPOISON;
1547 vma = find_vma_intersection(current->mm, addr, addr + 1);
1550 pfn = KVM_PFN_ERR_FAULT;
1551 else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
1552 r = hva_to_pfn_remapped(vma, addr, async, write_fault, &pfn);
1556 pfn = KVM_PFN_ERR_FAULT;
1558 if (async && vma_is_valid(vma, write_fault))
1560 pfn = KVM_PFN_ERR_FAULT;
1563 up_read(¤t->mm->mmap_sem);
1567 kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn,
1568 bool atomic, bool *async, bool write_fault,
1571 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1573 if (addr == KVM_HVA_ERR_RO_BAD) {
1576 return KVM_PFN_ERR_RO_FAULT;
1579 if (kvm_is_error_hva(addr)) {
1582 return KVM_PFN_NOSLOT;
1585 /* Do not map writable pfn in the readonly memslot. */
1586 if (writable && memslot_is_readonly(slot)) {
1591 return hva_to_pfn(addr, atomic, async, write_fault,
1594 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
1596 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1599 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
1600 write_fault, writable);
1602 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1604 kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1606 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1608 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
1610 kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1612 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1614 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1616 kvm_pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1618 return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm, gfn), gfn);
1620 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1622 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
1624 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1626 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
1628 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1630 return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
1632 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1634 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
1636 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1638 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
1640 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1641 struct page **pages, int nr_pages)
1646 addr = gfn_to_hva_many(slot, gfn, &entry);
1647 if (kvm_is_error_hva(addr))
1650 if (entry < nr_pages)
1653 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1655 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1657 static struct page *kvm_pfn_to_page(kvm_pfn_t pfn)
1659 if (is_error_noslot_pfn(pfn))
1660 return KVM_ERR_PTR_BAD_PAGE;
1662 if (kvm_is_reserved_pfn(pfn)) {
1664 return KVM_ERR_PTR_BAD_PAGE;
1667 return pfn_to_page(pfn);
1670 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1674 pfn = gfn_to_pfn(kvm, gfn);
1676 return kvm_pfn_to_page(pfn);
1678 EXPORT_SYMBOL_GPL(gfn_to_page);
1680 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
1684 pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
1686 return kvm_pfn_to_page(pfn);
1688 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
1690 void kvm_release_page_clean(struct page *page)
1692 WARN_ON(is_error_page(page));
1694 kvm_release_pfn_clean(page_to_pfn(page));
1696 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1698 void kvm_release_pfn_clean(kvm_pfn_t pfn)
1700 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
1701 put_page(pfn_to_page(pfn));
1703 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1705 void kvm_release_page_dirty(struct page *page)
1707 WARN_ON(is_error_page(page));
1709 kvm_release_pfn_dirty(page_to_pfn(page));
1711 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1713 static void kvm_release_pfn_dirty(kvm_pfn_t pfn)
1715 kvm_set_pfn_dirty(pfn);
1716 kvm_release_pfn_clean(pfn);
1719 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
1721 if (!kvm_is_reserved_pfn(pfn)) {
1722 struct page *page = pfn_to_page(pfn);
1724 if (!PageReserved(page))
1728 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1730 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
1732 if (!kvm_is_reserved_pfn(pfn))
1733 mark_page_accessed(pfn_to_page(pfn));
1735 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1737 void kvm_get_pfn(kvm_pfn_t pfn)
1739 if (!kvm_is_reserved_pfn(pfn))
1740 get_page(pfn_to_page(pfn));
1742 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1744 static int next_segment(unsigned long len, int offset)
1746 if (len > PAGE_SIZE - offset)
1747 return PAGE_SIZE - offset;
1752 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
1753 void *data, int offset, int len)
1758 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1759 if (kvm_is_error_hva(addr))
1761 r = __copy_from_user(data, (void __user *)addr + offset, len);
1767 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1770 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1772 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1774 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1776 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
1777 int offset, int len)
1779 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1781 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1783 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
1785 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1787 gfn_t gfn = gpa >> PAGE_SHIFT;
1789 int offset = offset_in_page(gpa);
1792 while ((seg = next_segment(len, offset)) != 0) {
1793 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1803 EXPORT_SYMBOL_GPL(kvm_read_guest);
1805 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
1807 gfn_t gfn = gpa >> PAGE_SHIFT;
1809 int offset = offset_in_page(gpa);
1812 while ((seg = next_segment(len, offset)) != 0) {
1813 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
1823 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
1825 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1826 void *data, int offset, unsigned long len)
1831 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1832 if (kvm_is_error_hva(addr))
1834 pagefault_disable();
1835 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1842 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1845 gfn_t gfn = gpa >> PAGE_SHIFT;
1846 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1847 int offset = offset_in_page(gpa);
1849 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1851 EXPORT_SYMBOL_GPL(kvm_read_guest_atomic);
1853 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
1854 void *data, unsigned long len)
1856 gfn_t gfn = gpa >> PAGE_SHIFT;
1857 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1858 int offset = offset_in_page(gpa);
1860 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1862 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
1864 static int __kvm_write_guest_page(struct kvm_memory_slot *memslot, gfn_t gfn,
1865 const void *data, int offset, int len)
1870 addr = gfn_to_hva_memslot(memslot, gfn);
1871 if (kvm_is_error_hva(addr))
1873 r = __copy_to_user((void __user *)addr + offset, data, len);
1876 mark_page_dirty_in_slot(memslot, gfn);
1880 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
1881 const void *data, int offset, int len)
1883 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1885 return __kvm_write_guest_page(slot, gfn, data, offset, len);
1887 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1889 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
1890 const void *data, int offset, int len)
1892 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1894 return __kvm_write_guest_page(slot, gfn, data, offset, len);
1896 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
1898 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1901 gfn_t gfn = gpa >> PAGE_SHIFT;
1903 int offset = offset_in_page(gpa);
1906 while ((seg = next_segment(len, offset)) != 0) {
1907 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1917 EXPORT_SYMBOL_GPL(kvm_write_guest);
1919 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
1922 gfn_t gfn = gpa >> PAGE_SHIFT;
1924 int offset = offset_in_page(gpa);
1927 while ((seg = next_segment(len, offset)) != 0) {
1928 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
1938 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
1940 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
1941 struct gfn_to_hva_cache *ghc,
1942 gpa_t gpa, unsigned long len)
1944 int offset = offset_in_page(gpa);
1945 gfn_t start_gfn = gpa >> PAGE_SHIFT;
1946 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1947 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1948 gfn_t nr_pages_avail;
1951 ghc->generation = slots->generation;
1953 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
1954 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, NULL);
1955 if (!kvm_is_error_hva(ghc->hva) && nr_pages_needed <= 1) {
1959 * If the requested region crosses two memslots, we still
1960 * verify that the entire region is valid here.
1962 while (start_gfn <= end_gfn) {
1963 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
1964 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1966 if (kvm_is_error_hva(ghc->hva))
1968 start_gfn += nr_pages_avail;
1970 /* Use the slow path for cross page reads and writes. */
1971 ghc->memslot = NULL;
1976 int kvm_vcpu_gfn_to_hva_cache_init(struct kvm_vcpu *vcpu, struct gfn_to_hva_cache *ghc,
1977 gpa_t gpa, unsigned long len)
1979 struct kvm_memslots *slots = kvm_vcpu_memslots(vcpu);
1980 return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len);
1982 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva_cache_init);
1984 int kvm_vcpu_write_guest_offset_cached(struct kvm_vcpu *vcpu, struct gfn_to_hva_cache *ghc,
1985 void *data, int offset, unsigned long len)
1987 struct kvm_memslots *slots = kvm_vcpu_memslots(vcpu);
1989 gpa_t gpa = ghc->gpa + offset;
1991 BUG_ON(len + offset > ghc->len);
1993 if (slots->generation != ghc->generation)
1994 __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len);
1996 if (unlikely(!ghc->memslot))
1997 return kvm_vcpu_write_guest(vcpu, gpa, data, len);
1999 if (kvm_is_error_hva(ghc->hva))
2002 r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
2005 mark_page_dirty_in_slot(ghc->memslot, gpa >> PAGE_SHIFT);
2009 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_offset_cached);
2011 int kvm_vcpu_write_guest_cached(struct kvm_vcpu *vcpu, struct gfn_to_hva_cache *ghc,
2012 void *data, unsigned long len)
2014 return kvm_vcpu_write_guest_offset_cached(vcpu, ghc, data, 0, len);
2016 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_cached);
2018 int kvm_vcpu_read_guest_cached(struct kvm_vcpu *vcpu, struct gfn_to_hva_cache *ghc,
2019 void *data, unsigned long len)
2021 struct kvm_memslots *slots = kvm_vcpu_memslots(vcpu);
2024 BUG_ON(len > ghc->len);
2026 if (slots->generation != ghc->generation)
2027 __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len);
2029 if (unlikely(!ghc->memslot))
2030 return kvm_vcpu_read_guest(vcpu, ghc->gpa, data, len);
2032 if (kvm_is_error_hva(ghc->hva))
2035 r = __copy_from_user(data, (void __user *)ghc->hva, len);
2041 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_cached);
2043 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
2045 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
2047 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
2049 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
2051 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
2053 gfn_t gfn = gpa >> PAGE_SHIFT;
2055 int offset = offset_in_page(gpa);
2058 while ((seg = next_segment(len, offset)) != 0) {
2059 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
2068 EXPORT_SYMBOL_GPL(kvm_clear_guest);
2070 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot,
2073 if (memslot && memslot->dirty_bitmap) {
2074 unsigned long rel_gfn = gfn - memslot->base_gfn;
2076 set_bit_le(rel_gfn, memslot->dirty_bitmap);
2080 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
2082 struct kvm_memory_slot *memslot;
2084 memslot = gfn_to_memslot(kvm, gfn);
2085 mark_page_dirty_in_slot(memslot, gfn);
2087 EXPORT_SYMBOL_GPL(mark_page_dirty);
2089 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
2091 struct kvm_memory_slot *memslot;
2093 memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2094 mark_page_dirty_in_slot(memslot, gfn);
2096 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
2098 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
2100 unsigned int old, val, grow;
2102 old = val = vcpu->halt_poll_ns;
2103 grow = READ_ONCE(halt_poll_ns_grow);
2105 if (val == 0 && grow)
2110 if (val > halt_poll_ns)
2113 vcpu->halt_poll_ns = val;
2114 trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
2117 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
2119 unsigned int old, val, shrink;
2121 old = val = vcpu->halt_poll_ns;
2122 shrink = READ_ONCE(halt_poll_ns_shrink);
2128 vcpu->halt_poll_ns = val;
2129 trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
2132 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
2134 if (kvm_arch_vcpu_runnable(vcpu)) {
2135 kvm_make_request(KVM_REQ_UNHALT, vcpu);
2138 if (kvm_cpu_has_pending_timer(vcpu))
2140 if (signal_pending(current))
2147 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2149 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
2152 DECLARE_SWAITQUEUE(wait);
2153 bool waited = false;
2156 start = cur = ktime_get();
2157 if (vcpu->halt_poll_ns) {
2158 ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns);
2160 ++vcpu->stat.halt_attempted_poll;
2163 * This sets KVM_REQ_UNHALT if an interrupt
2166 if (kvm_vcpu_check_block(vcpu) < 0) {
2167 ++vcpu->stat.halt_successful_poll;
2168 if (!vcpu_valid_wakeup(vcpu))
2169 ++vcpu->stat.halt_poll_invalid;
2173 } while (single_task_running() && ktime_before(cur, stop));
2176 kvm_arch_vcpu_blocking(vcpu);
2179 prepare_to_swait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
2181 if (kvm_vcpu_check_block(vcpu) < 0)
2188 finish_swait(&vcpu->wq, &wait);
2191 kvm_arch_vcpu_unblocking(vcpu);
2193 block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
2195 if (!vcpu_valid_wakeup(vcpu))
2196 shrink_halt_poll_ns(vcpu);
2197 else if (halt_poll_ns) {
2198 if (block_ns <= vcpu->halt_poll_ns)
2200 /* we had a long block, shrink polling */
2201 else if (vcpu->halt_poll_ns && block_ns > halt_poll_ns)
2202 shrink_halt_poll_ns(vcpu);
2203 /* we had a short halt and our poll time is too small */
2204 else if (vcpu->halt_poll_ns < halt_poll_ns &&
2205 block_ns < halt_poll_ns)
2206 grow_halt_poll_ns(vcpu);
2208 vcpu->halt_poll_ns = 0;
2210 trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu));
2211 kvm_arch_vcpu_block_finish(vcpu);
2213 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
2216 void kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
2218 struct swait_queue_head *wqp;
2220 wqp = kvm_arch_vcpu_wq(vcpu);
2221 if (swait_active(wqp)) {
2223 ++vcpu->stat.halt_wakeup;
2227 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
2230 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2232 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
2235 int cpu = vcpu->cpu;
2237 kvm_vcpu_wake_up(vcpu);
2239 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
2240 if (kvm_arch_vcpu_should_kick(vcpu))
2241 smp_send_reschedule(cpu);
2244 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
2245 #endif /* !CONFIG_S390 */
2247 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
2250 struct task_struct *task = NULL;
2254 pid = rcu_dereference(target->pid);
2256 task = get_pid_task(pid, PIDTYPE_PID);
2260 ret = yield_to(task, 1);
2261 put_task_struct(task);
2265 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
2268 * Helper that checks whether a VCPU is eligible for directed yield.
2269 * Most eligible candidate to yield is decided by following heuristics:
2271 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2272 * (preempted lock holder), indicated by @in_spin_loop.
2273 * Set at the beiginning and cleared at the end of interception/PLE handler.
2275 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2276 * chance last time (mostly it has become eligible now since we have probably
2277 * yielded to lockholder in last iteration. This is done by toggling
2278 * @dy_eligible each time a VCPU checked for eligibility.)
2280 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2281 * to preempted lock-holder could result in wrong VCPU selection and CPU
2282 * burning. Giving priority for a potential lock-holder increases lock
2285 * Since algorithm is based on heuristics, accessing another VCPU data without
2286 * locking does not harm. It may result in trying to yield to same VCPU, fail
2287 * and continue with next VCPU and so on.
2289 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
2291 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2294 eligible = !vcpu->spin_loop.in_spin_loop ||
2295 vcpu->spin_loop.dy_eligible;
2297 if (vcpu->spin_loop.in_spin_loop)
2298 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
2306 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
2308 struct kvm *kvm = me->kvm;
2309 struct kvm_vcpu *vcpu;
2310 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
2316 kvm_vcpu_set_in_spin_loop(me, true);
2318 * We boost the priority of a VCPU that is runnable but not
2319 * currently running, because it got preempted by something
2320 * else and called schedule in __vcpu_run. Hopefully that
2321 * VCPU is holding the lock that we need and will release it.
2322 * We approximate round-robin by starting at the last boosted VCPU.
2324 for (pass = 0; pass < 2 && !yielded && try; pass++) {
2325 kvm_for_each_vcpu(i, vcpu, kvm) {
2326 if (!pass && i <= last_boosted_vcpu) {
2327 i = last_boosted_vcpu;
2329 } else if (pass && i > last_boosted_vcpu)
2331 if (!ACCESS_ONCE(vcpu->preempted))
2335 if (swait_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
2337 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
2340 yielded = kvm_vcpu_yield_to(vcpu);
2342 kvm->last_boosted_vcpu = i;
2344 } else if (yielded < 0) {
2351 kvm_vcpu_set_in_spin_loop(me, false);
2353 /* Ensure vcpu is not eligible during next spinloop */
2354 kvm_vcpu_set_dy_eligible(me, false);
2356 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
2358 static int kvm_vcpu_fault(struct vm_fault *vmf)
2360 struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data;
2363 if (vmf->pgoff == 0)
2364 page = virt_to_page(vcpu->run);
2366 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
2367 page = virt_to_page(vcpu->arch.pio_data);
2369 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2370 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
2371 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
2374 return kvm_arch_vcpu_fault(vcpu, vmf);
2380 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
2381 .fault = kvm_vcpu_fault,
2384 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2386 vma->vm_ops = &kvm_vcpu_vm_ops;
2390 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2392 struct kvm_vcpu *vcpu = filp->private_data;
2394 debugfs_remove_recursive(vcpu->debugfs_dentry);
2395 kvm_put_kvm(vcpu->kvm);
2399 static struct file_operations kvm_vcpu_fops = {
2400 .release = kvm_vcpu_release,
2401 .unlocked_ioctl = kvm_vcpu_ioctl,
2402 #ifdef CONFIG_KVM_COMPAT
2403 .compat_ioctl = kvm_vcpu_compat_ioctl,
2405 .mmap = kvm_vcpu_mmap,
2406 .llseek = noop_llseek,
2410 * Allocates an inode for the vcpu.
2412 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2414 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
2417 static int kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
2419 char dir_name[ITOA_MAX_LEN * 2];
2422 if (!kvm_arch_has_vcpu_debugfs())
2425 if (!debugfs_initialized())
2428 snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
2429 vcpu->debugfs_dentry = debugfs_create_dir(dir_name,
2430 vcpu->kvm->debugfs_dentry);
2431 if (!vcpu->debugfs_dentry)
2434 ret = kvm_arch_create_vcpu_debugfs(vcpu);
2436 debugfs_remove_recursive(vcpu->debugfs_dentry);
2444 * Creates some virtual cpus. Good luck creating more than one.
2446 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
2449 struct kvm_vcpu *vcpu;
2451 if (id >= KVM_MAX_VCPU_ID)
2454 mutex_lock(&kvm->lock);
2455 if (kvm->created_vcpus == KVM_MAX_VCPUS) {
2456 mutex_unlock(&kvm->lock);
2460 kvm->created_vcpus++;
2461 mutex_unlock(&kvm->lock);
2463 vcpu = kvm_arch_vcpu_create(kvm, id);
2466 goto vcpu_decrement;
2469 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2471 r = kvm_arch_vcpu_setup(vcpu);
2475 r = kvm_create_vcpu_debugfs(vcpu);
2479 mutex_lock(&kvm->lock);
2480 if (kvm_get_vcpu_by_id(kvm, id)) {
2482 goto unlock_vcpu_destroy;
2485 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
2487 /* Now it's all set up, let userspace reach it */
2489 r = create_vcpu_fd(vcpu);
2492 goto unlock_vcpu_destroy;
2495 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
2498 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
2499 * before kvm->online_vcpu's incremented value.
2502 atomic_inc(&kvm->online_vcpus);
2504 mutex_unlock(&kvm->lock);
2505 kvm_arch_vcpu_postcreate(vcpu);
2508 unlock_vcpu_destroy:
2509 mutex_unlock(&kvm->lock);
2510 debugfs_remove_recursive(vcpu->debugfs_dentry);
2512 kvm_arch_vcpu_destroy(vcpu);
2514 mutex_lock(&kvm->lock);
2515 kvm->created_vcpus--;
2516 mutex_unlock(&kvm->lock);
2520 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2523 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2524 vcpu->sigset_active = 1;
2525 vcpu->sigset = *sigset;
2527 vcpu->sigset_active = 0;
2531 static long kvm_vcpu_ioctl(struct file *filp,
2532 unsigned int ioctl, unsigned long arg)
2534 struct kvm_vcpu *vcpu = filp->private_data;
2535 void __user *argp = (void __user *)arg;
2537 struct kvm_fpu *fpu = NULL;
2538 struct kvm_sregs *kvm_sregs = NULL;
2540 if (vcpu->kvm->mm != current->mm)
2543 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2546 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
2548 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
2549 * so vcpu_load() would break it.
2551 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_S390_IRQ || ioctl == KVM_INTERRUPT)
2552 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2556 r = vcpu_load(vcpu);
2564 if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
2565 /* The thread running this VCPU changed. */
2566 struct pid *oldpid = vcpu->pid;
2567 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
2569 rcu_assign_pointer(vcpu->pid, newpid);
2574 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2575 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2577 case KVM_GET_REGS: {
2578 struct kvm_regs *kvm_regs;
2581 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2584 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2588 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2595 case KVM_SET_REGS: {
2596 struct kvm_regs *kvm_regs;
2599 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2600 if (IS_ERR(kvm_regs)) {
2601 r = PTR_ERR(kvm_regs);
2604 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2608 case KVM_GET_SREGS: {
2609 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2613 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2617 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2622 case KVM_SET_SREGS: {
2623 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2624 if (IS_ERR(kvm_sregs)) {
2625 r = PTR_ERR(kvm_sregs);
2629 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2632 case KVM_GET_MP_STATE: {
2633 struct kvm_mp_state mp_state;
2635 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2639 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
2644 case KVM_SET_MP_STATE: {
2645 struct kvm_mp_state mp_state;
2648 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
2650 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2653 case KVM_TRANSLATE: {
2654 struct kvm_translation tr;
2657 if (copy_from_user(&tr, argp, sizeof(tr)))
2659 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2663 if (copy_to_user(argp, &tr, sizeof(tr)))
2668 case KVM_SET_GUEST_DEBUG: {
2669 struct kvm_guest_debug dbg;
2672 if (copy_from_user(&dbg, argp, sizeof(dbg)))
2674 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2677 case KVM_SET_SIGNAL_MASK: {
2678 struct kvm_signal_mask __user *sigmask_arg = argp;
2679 struct kvm_signal_mask kvm_sigmask;
2680 sigset_t sigset, *p;
2685 if (copy_from_user(&kvm_sigmask, argp,
2686 sizeof(kvm_sigmask)))
2689 if (kvm_sigmask.len != sizeof(sigset))
2692 if (copy_from_user(&sigset, sigmask_arg->sigset,
2697 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2701 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2705 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2709 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2715 fpu = memdup_user(argp, sizeof(*fpu));
2721 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2725 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2734 #ifdef CONFIG_KVM_COMPAT
2735 static long kvm_vcpu_compat_ioctl(struct file *filp,
2736 unsigned int ioctl, unsigned long arg)
2738 struct kvm_vcpu *vcpu = filp->private_data;
2739 void __user *argp = compat_ptr(arg);
2742 if (vcpu->kvm->mm != current->mm)
2746 case KVM_SET_SIGNAL_MASK: {
2747 struct kvm_signal_mask __user *sigmask_arg = argp;
2748 struct kvm_signal_mask kvm_sigmask;
2749 compat_sigset_t csigset;
2754 if (copy_from_user(&kvm_sigmask, argp,
2755 sizeof(kvm_sigmask)))
2758 if (kvm_sigmask.len != sizeof(csigset))
2761 if (copy_from_user(&csigset, sigmask_arg->sigset,
2764 sigset_from_compat(&sigset, &csigset);
2765 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2767 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2771 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2779 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2780 int (*accessor)(struct kvm_device *dev,
2781 struct kvm_device_attr *attr),
2784 struct kvm_device_attr attr;
2789 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2792 return accessor(dev, &attr);
2795 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2798 struct kvm_device *dev = filp->private_data;
2801 case KVM_SET_DEVICE_ATTR:
2802 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2803 case KVM_GET_DEVICE_ATTR:
2804 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2805 case KVM_HAS_DEVICE_ATTR:
2806 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2808 if (dev->ops->ioctl)
2809 return dev->ops->ioctl(dev, ioctl, arg);
2815 static int kvm_device_release(struct inode *inode, struct file *filp)
2817 struct kvm_device *dev = filp->private_data;
2818 struct kvm *kvm = dev->kvm;
2824 static const struct file_operations kvm_device_fops = {
2825 .unlocked_ioctl = kvm_device_ioctl,
2826 #ifdef CONFIG_KVM_COMPAT
2827 .compat_ioctl = kvm_device_ioctl,
2829 .release = kvm_device_release,
2832 struct kvm_device *kvm_device_from_filp(struct file *filp)
2834 if (filp->f_op != &kvm_device_fops)
2837 return filp->private_data;
2840 static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
2841 #ifdef CONFIG_KVM_MPIC
2842 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
2843 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
2846 #ifdef CONFIG_KVM_XICS
2847 [KVM_DEV_TYPE_XICS] = &kvm_xics_ops,
2851 int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
2853 if (type >= ARRAY_SIZE(kvm_device_ops_table))
2856 if (kvm_device_ops_table[type] != NULL)
2859 kvm_device_ops_table[type] = ops;
2863 void kvm_unregister_device_ops(u32 type)
2865 if (kvm_device_ops_table[type] != NULL)
2866 kvm_device_ops_table[type] = NULL;
2869 static int kvm_ioctl_create_device(struct kvm *kvm,
2870 struct kvm_create_device *cd)
2872 struct kvm_device_ops *ops = NULL;
2873 struct kvm_device *dev;
2874 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2877 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
2880 ops = kvm_device_ops_table[cd->type];
2887 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2894 mutex_lock(&kvm->lock);
2895 ret = ops->create(dev, cd->type);
2897 mutex_unlock(&kvm->lock);
2901 list_add(&dev->vm_node, &kvm->devices);
2902 mutex_unlock(&kvm->lock);
2907 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2909 mutex_lock(&kvm->lock);
2910 list_del(&dev->vm_node);
2911 mutex_unlock(&kvm->lock);
2921 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2924 case KVM_CAP_USER_MEMORY:
2925 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2926 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2927 case KVM_CAP_INTERNAL_ERROR_DATA:
2928 #ifdef CONFIG_HAVE_KVM_MSI
2929 case KVM_CAP_SIGNAL_MSI:
2931 #ifdef CONFIG_HAVE_KVM_IRQFD
2933 case KVM_CAP_IRQFD_RESAMPLE:
2935 case KVM_CAP_IOEVENTFD_ANY_LENGTH:
2936 case KVM_CAP_CHECK_EXTENSION_VM:
2938 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2939 case KVM_CAP_IRQ_ROUTING:
2940 return KVM_MAX_IRQ_ROUTES;
2942 #if KVM_ADDRESS_SPACE_NUM > 1
2943 case KVM_CAP_MULTI_ADDRESS_SPACE:
2944 return KVM_ADDRESS_SPACE_NUM;
2946 case KVM_CAP_MAX_VCPU_ID:
2947 return KVM_MAX_VCPU_ID;
2951 return kvm_vm_ioctl_check_extension(kvm, arg);
2954 static long kvm_vm_ioctl(struct file *filp,
2955 unsigned int ioctl, unsigned long arg)
2957 struct kvm *kvm = filp->private_data;
2958 void __user *argp = (void __user *)arg;
2961 if (kvm->mm != current->mm)
2964 case KVM_CREATE_VCPU:
2965 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2967 case KVM_SET_USER_MEMORY_REGION: {
2968 struct kvm_userspace_memory_region kvm_userspace_mem;
2971 if (copy_from_user(&kvm_userspace_mem, argp,
2972 sizeof(kvm_userspace_mem)))
2975 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2978 case KVM_GET_DIRTY_LOG: {
2979 struct kvm_dirty_log log;
2982 if (copy_from_user(&log, argp, sizeof(log)))
2984 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2987 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2988 case KVM_REGISTER_COALESCED_MMIO: {
2989 struct kvm_coalesced_mmio_zone zone;
2992 if (copy_from_user(&zone, argp, sizeof(zone)))
2994 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2997 case KVM_UNREGISTER_COALESCED_MMIO: {
2998 struct kvm_coalesced_mmio_zone zone;
3001 if (copy_from_user(&zone, argp, sizeof(zone)))
3003 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
3008 struct kvm_irqfd data;
3011 if (copy_from_user(&data, argp, sizeof(data)))
3013 r = kvm_irqfd(kvm, &data);
3016 case KVM_IOEVENTFD: {
3017 struct kvm_ioeventfd data;
3020 if (copy_from_user(&data, argp, sizeof(data)))
3022 r = kvm_ioeventfd(kvm, &data);
3025 #ifdef CONFIG_HAVE_KVM_MSI
3026 case KVM_SIGNAL_MSI: {
3030 if (copy_from_user(&msi, argp, sizeof(msi)))
3032 r = kvm_send_userspace_msi(kvm, &msi);
3036 #ifdef __KVM_HAVE_IRQ_LINE
3037 case KVM_IRQ_LINE_STATUS:
3038 case KVM_IRQ_LINE: {
3039 struct kvm_irq_level irq_event;
3042 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
3045 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
3046 ioctl == KVM_IRQ_LINE_STATUS);
3051 if (ioctl == KVM_IRQ_LINE_STATUS) {
3052 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
3060 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3061 case KVM_SET_GSI_ROUTING: {
3062 struct kvm_irq_routing routing;
3063 struct kvm_irq_routing __user *urouting;
3064 struct kvm_irq_routing_entry *entries = NULL;
3067 if (copy_from_user(&routing, argp, sizeof(routing)))
3070 if (routing.nr > KVM_MAX_IRQ_ROUTES)
3076 entries = vmalloc(routing.nr * sizeof(*entries));
3081 if (copy_from_user(entries, urouting->entries,
3082 routing.nr * sizeof(*entries)))
3083 goto out_free_irq_routing;
3085 r = kvm_set_irq_routing(kvm, entries, routing.nr,
3087 out_free_irq_routing:
3091 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
3092 case KVM_CREATE_DEVICE: {
3093 struct kvm_create_device cd;
3096 if (copy_from_user(&cd, argp, sizeof(cd)))
3099 r = kvm_ioctl_create_device(kvm, &cd);
3104 if (copy_to_user(argp, &cd, sizeof(cd)))
3110 case KVM_CHECK_EXTENSION:
3111 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
3114 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
3120 #ifdef CONFIG_KVM_COMPAT
3121 struct compat_kvm_dirty_log {
3125 compat_uptr_t dirty_bitmap; /* one bit per page */
3130 static long kvm_vm_compat_ioctl(struct file *filp,
3131 unsigned int ioctl, unsigned long arg)
3133 struct kvm *kvm = filp->private_data;
3136 if (kvm->mm != current->mm)
3139 case KVM_GET_DIRTY_LOG: {
3140 struct compat_kvm_dirty_log compat_log;
3141 struct kvm_dirty_log log;
3143 if (copy_from_user(&compat_log, (void __user *)arg,
3144 sizeof(compat_log)))
3146 log.slot = compat_log.slot;
3147 log.padding1 = compat_log.padding1;
3148 log.padding2 = compat_log.padding2;
3149 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
3151 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3155 r = kvm_vm_ioctl(filp, ioctl, arg);
3161 static struct file_operations kvm_vm_fops = {
3162 .release = kvm_vm_release,
3163 .unlocked_ioctl = kvm_vm_ioctl,
3164 #ifdef CONFIG_KVM_COMPAT
3165 .compat_ioctl = kvm_vm_compat_ioctl,
3167 .llseek = noop_llseek,
3170 static int kvm_dev_ioctl_create_vm(unsigned long type)
3176 kvm = kvm_create_vm(type);
3178 return PTR_ERR(kvm);
3179 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
3180 r = kvm_coalesced_mmio_init(kvm);
3186 r = get_unused_fd_flags(O_CLOEXEC);
3191 file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
3195 return PTR_ERR(file);
3198 if (kvm_create_vm_debugfs(kvm, r) < 0) {
3204 fd_install(r, file);
3208 static long kvm_dev_ioctl(struct file *filp,
3209 unsigned int ioctl, unsigned long arg)
3214 case KVM_GET_API_VERSION:
3217 r = KVM_API_VERSION;
3220 r = kvm_dev_ioctl_create_vm(arg);
3222 case KVM_CHECK_EXTENSION:
3223 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
3225 case KVM_GET_VCPU_MMAP_SIZE:
3228 r = PAGE_SIZE; /* struct kvm_run */
3230 r += PAGE_SIZE; /* pio data page */
3232 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
3233 r += PAGE_SIZE; /* coalesced mmio ring page */
3236 case KVM_TRACE_ENABLE:
3237 case KVM_TRACE_PAUSE:
3238 case KVM_TRACE_DISABLE:
3242 return kvm_arch_dev_ioctl(filp, ioctl, arg);
3248 static struct file_operations kvm_chardev_ops = {
3249 .unlocked_ioctl = kvm_dev_ioctl,
3250 .compat_ioctl = kvm_dev_ioctl,
3251 .llseek = noop_llseek,
3254 static struct miscdevice kvm_dev = {
3260 static void hardware_enable_nolock(void *junk)
3262 int cpu = raw_smp_processor_id();
3265 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
3268 cpumask_set_cpu(cpu, cpus_hardware_enabled);
3270 r = kvm_arch_hardware_enable();
3273 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3274 atomic_inc(&hardware_enable_failed);
3275 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
3279 static int kvm_starting_cpu(unsigned int cpu)
3281 raw_spin_lock(&kvm_count_lock);
3282 if (kvm_usage_count)
3283 hardware_enable_nolock(NULL);
3284 raw_spin_unlock(&kvm_count_lock);
3288 static void hardware_disable_nolock(void *junk)
3290 int cpu = raw_smp_processor_id();
3292 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
3294 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3295 kvm_arch_hardware_disable();
3298 static int kvm_dying_cpu(unsigned int cpu)
3300 raw_spin_lock(&kvm_count_lock);
3301 if (kvm_usage_count)
3302 hardware_disable_nolock(NULL);
3303 raw_spin_unlock(&kvm_count_lock);
3307 static void hardware_disable_all_nolock(void)
3309 BUG_ON(!kvm_usage_count);
3312 if (!kvm_usage_count)
3313 on_each_cpu(hardware_disable_nolock, NULL, 1);
3316 static void hardware_disable_all(void)
3318 raw_spin_lock(&kvm_count_lock);
3319 hardware_disable_all_nolock();
3320 raw_spin_unlock(&kvm_count_lock);
3323 static int hardware_enable_all(void)
3327 raw_spin_lock(&kvm_count_lock);
3330 if (kvm_usage_count == 1) {
3331 atomic_set(&hardware_enable_failed, 0);
3332 on_each_cpu(hardware_enable_nolock, NULL, 1);
3334 if (atomic_read(&hardware_enable_failed)) {
3335 hardware_disable_all_nolock();
3340 raw_spin_unlock(&kvm_count_lock);
3345 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3349 * Some (well, at least mine) BIOSes hang on reboot if
3352 * And Intel TXT required VMX off for all cpu when system shutdown.
3354 pr_info("kvm: exiting hardware virtualization\n");
3355 kvm_rebooting = true;
3356 on_each_cpu(hardware_disable_nolock, NULL, 1);
3360 static struct notifier_block kvm_reboot_notifier = {
3361 .notifier_call = kvm_reboot,
3365 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3369 for (i = 0; i < bus->dev_count; i++) {
3370 struct kvm_io_device *pos = bus->range[i].dev;
3372 kvm_iodevice_destructor(pos);
3377 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
3378 const struct kvm_io_range *r2)
3380 gpa_t addr1 = r1->addr;
3381 gpa_t addr2 = r2->addr;
3386 /* If r2->len == 0, match the exact address. If r2->len != 0,
3387 * accept any overlapping write. Any order is acceptable for
3388 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
3389 * we process all of them.
3402 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
3404 return kvm_io_bus_cmp(p1, p2);
3407 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
3408 gpa_t addr, int len)
3410 bus->range[bus->dev_count++] = (struct kvm_io_range) {
3416 sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
3417 kvm_io_bus_sort_cmp, NULL);
3422 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
3423 gpa_t addr, int len)
3425 struct kvm_io_range *range, key;
3428 key = (struct kvm_io_range) {
3433 range = bsearch(&key, bus->range, bus->dev_count,
3434 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
3438 off = range - bus->range;
3440 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
3446 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3447 struct kvm_io_range *range, const void *val)
3451 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3455 while (idx < bus->dev_count &&
3456 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3457 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
3466 /* kvm_io_bus_write - called under kvm->slots_lock */
3467 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3468 int len, const void *val)
3470 struct kvm_io_bus *bus;
3471 struct kvm_io_range range;
3474 range = (struct kvm_io_range) {
3479 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3482 r = __kvm_io_bus_write(vcpu, bus, &range, val);
3483 return r < 0 ? r : 0;
3486 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3487 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
3488 gpa_t addr, int len, const void *val, long cookie)
3490 struct kvm_io_bus *bus;
3491 struct kvm_io_range range;
3493 range = (struct kvm_io_range) {
3498 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3502 /* First try the device referenced by cookie. */
3503 if ((cookie >= 0) && (cookie < bus->dev_count) &&
3504 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
3505 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
3510 * cookie contained garbage; fall back to search and return the
3511 * correct cookie value.
3513 return __kvm_io_bus_write(vcpu, bus, &range, val);
3516 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3517 struct kvm_io_range *range, void *val)
3521 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3525 while (idx < bus->dev_count &&
3526 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3527 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
3535 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
3537 /* kvm_io_bus_read - called under kvm->slots_lock */
3538 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3541 struct kvm_io_bus *bus;
3542 struct kvm_io_range range;
3545 range = (struct kvm_io_range) {
3550 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3553 r = __kvm_io_bus_read(vcpu, bus, &range, val);
3554 return r < 0 ? r : 0;
3558 /* Caller must hold slots_lock. */
3559 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3560 int len, struct kvm_io_device *dev)
3562 struct kvm_io_bus *new_bus, *bus;
3564 bus = kvm->buses[bus_idx];
3568 /* exclude ioeventfd which is limited by maximum fd */
3569 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3572 new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count + 1) *
3573 sizeof(struct kvm_io_range)), GFP_KERNEL);
3576 memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
3577 sizeof(struct kvm_io_range)));
3578 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
3579 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3580 synchronize_srcu_expedited(&kvm->srcu);
3586 /* Caller must hold slots_lock. */
3587 void kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3588 struct kvm_io_device *dev)
3591 struct kvm_io_bus *new_bus, *bus;
3593 bus = kvm->buses[bus_idx];
3597 for (i = 0; i < bus->dev_count; i++)
3598 if (bus->range[i].dev == dev) {
3602 if (i == bus->dev_count)
3605 new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3606 sizeof(struct kvm_io_range)), GFP_KERNEL);
3608 pr_err("kvm: failed to shrink bus, removing it completely\n");
3612 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3613 new_bus->dev_count--;
3614 memcpy(new_bus->range + i, bus->range + i + 1,
3615 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3618 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3619 synchronize_srcu_expedited(&kvm->srcu);
3624 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3627 struct kvm_io_bus *bus;
3628 int dev_idx, srcu_idx;
3629 struct kvm_io_device *iodev = NULL;
3631 srcu_idx = srcu_read_lock(&kvm->srcu);
3633 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
3637 dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
3641 iodev = bus->range[dev_idx].dev;
3644 srcu_read_unlock(&kvm->srcu, srcu_idx);
3648 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
3650 static int kvm_debugfs_open(struct inode *inode, struct file *file,
3651 int (*get)(void *, u64 *), int (*set)(void *, u64),
3654 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
3657 /* The debugfs files are a reference to the kvm struct which
3658 * is still valid when kvm_destroy_vm is called.
3659 * To avoid the race between open and the removal of the debugfs
3660 * directory we test against the users count.
3662 if (!refcount_inc_not_zero(&stat_data->kvm->users_count))
3665 if (simple_attr_open(inode, file, get, set, fmt)) {
3666 kvm_put_kvm(stat_data->kvm);
3673 static int kvm_debugfs_release(struct inode *inode, struct file *file)
3675 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
3678 simple_attr_release(inode, file);
3679 kvm_put_kvm(stat_data->kvm);
3684 static int vm_stat_get_per_vm(void *data, u64 *val)
3686 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3688 *val = *(ulong *)((void *)stat_data->kvm + stat_data->offset);
3693 static int vm_stat_clear_per_vm(void *data, u64 val)
3695 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3700 *(ulong *)((void *)stat_data->kvm + stat_data->offset) = 0;
3705 static int vm_stat_get_per_vm_open(struct inode *inode, struct file *file)
3707 __simple_attr_check_format("%llu\n", 0ull);
3708 return kvm_debugfs_open(inode, file, vm_stat_get_per_vm,
3709 vm_stat_clear_per_vm, "%llu\n");
3712 static const struct file_operations vm_stat_get_per_vm_fops = {
3713 .owner = THIS_MODULE,
3714 .open = vm_stat_get_per_vm_open,
3715 .release = kvm_debugfs_release,
3716 .read = simple_attr_read,
3717 .write = simple_attr_write,
3718 .llseek = generic_file_llseek,
3721 static int vcpu_stat_get_per_vm(void *data, u64 *val)
3724 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3725 struct kvm_vcpu *vcpu;
3729 kvm_for_each_vcpu(i, vcpu, stat_data->kvm)
3730 *val += *(u64 *)((void *)vcpu + stat_data->offset);
3735 static int vcpu_stat_clear_per_vm(void *data, u64 val)
3738 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3739 struct kvm_vcpu *vcpu;
3744 kvm_for_each_vcpu(i, vcpu, stat_data->kvm)
3745 *(u64 *)((void *)vcpu + stat_data->offset) = 0;
3750 static int vcpu_stat_get_per_vm_open(struct inode *inode, struct file *file)
3752 __simple_attr_check_format("%llu\n", 0ull);
3753 return kvm_debugfs_open(inode, file, vcpu_stat_get_per_vm,
3754 vcpu_stat_clear_per_vm, "%llu\n");
3757 static const struct file_operations vcpu_stat_get_per_vm_fops = {
3758 .owner = THIS_MODULE,
3759 .open = vcpu_stat_get_per_vm_open,
3760 .release = kvm_debugfs_release,
3761 .read = simple_attr_read,
3762 .write = simple_attr_write,
3763 .llseek = generic_file_llseek,
3766 static const struct file_operations *stat_fops_per_vm[] = {
3767 [KVM_STAT_VCPU] = &vcpu_stat_get_per_vm_fops,
3768 [KVM_STAT_VM] = &vm_stat_get_per_vm_fops,
3771 static int vm_stat_get(void *_offset, u64 *val)
3773 unsigned offset = (long)_offset;
3775 struct kvm_stat_data stat_tmp = {.offset = offset};
3779 spin_lock(&kvm_lock);
3780 list_for_each_entry(kvm, &vm_list, vm_list) {
3782 vm_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
3785 spin_unlock(&kvm_lock);
3789 static int vm_stat_clear(void *_offset, u64 val)
3791 unsigned offset = (long)_offset;
3793 struct kvm_stat_data stat_tmp = {.offset = offset};
3798 spin_lock(&kvm_lock);
3799 list_for_each_entry(kvm, &vm_list, vm_list) {
3801 vm_stat_clear_per_vm((void *)&stat_tmp, 0);
3803 spin_unlock(&kvm_lock);
3808 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n");
3810 static int vcpu_stat_get(void *_offset, u64 *val)
3812 unsigned offset = (long)_offset;
3814 struct kvm_stat_data stat_tmp = {.offset = offset};
3818 spin_lock(&kvm_lock);
3819 list_for_each_entry(kvm, &vm_list, vm_list) {
3821 vcpu_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
3824 spin_unlock(&kvm_lock);
3828 static int vcpu_stat_clear(void *_offset, u64 val)
3830 unsigned offset = (long)_offset;
3832 struct kvm_stat_data stat_tmp = {.offset = offset};
3837 spin_lock(&kvm_lock);
3838 list_for_each_entry(kvm, &vm_list, vm_list) {
3840 vcpu_stat_clear_per_vm((void *)&stat_tmp, 0);
3842 spin_unlock(&kvm_lock);
3847 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear,
3850 static const struct file_operations *stat_fops[] = {
3851 [KVM_STAT_VCPU] = &vcpu_stat_fops,
3852 [KVM_STAT_VM] = &vm_stat_fops,
3855 static int kvm_init_debug(void)
3858 struct kvm_stats_debugfs_item *p;
3860 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3861 if (kvm_debugfs_dir == NULL)
3864 kvm_debugfs_num_entries = 0;
3865 for (p = debugfs_entries; p->name; ++p, kvm_debugfs_num_entries++) {
3866 if (!debugfs_create_file(p->name, 0644, kvm_debugfs_dir,
3867 (void *)(long)p->offset,
3868 stat_fops[p->kind]))
3875 debugfs_remove_recursive(kvm_debugfs_dir);
3880 static int kvm_suspend(void)
3882 if (kvm_usage_count)
3883 hardware_disable_nolock(NULL);
3887 static void kvm_resume(void)
3889 if (kvm_usage_count) {
3890 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3891 hardware_enable_nolock(NULL);
3895 static struct syscore_ops kvm_syscore_ops = {
3896 .suspend = kvm_suspend,
3897 .resume = kvm_resume,
3901 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3903 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3906 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3908 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3910 if (vcpu->preempted)
3911 vcpu->preempted = false;
3913 kvm_arch_sched_in(vcpu, cpu);
3915 kvm_arch_vcpu_load(vcpu, cpu);
3918 static void kvm_sched_out(struct preempt_notifier *pn,
3919 struct task_struct *next)
3921 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3923 if (current->state == TASK_RUNNING)
3924 vcpu->preempted = true;
3925 kvm_arch_vcpu_put(vcpu);
3928 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3929 struct module *module)
3934 r = kvm_arch_init(opaque);
3939 * kvm_arch_init makes sure there's at most one caller
3940 * for architectures that support multiple implementations,
3941 * like intel and amd on x86.
3942 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3943 * conflicts in case kvm is already setup for another implementation.
3945 r = kvm_irqfd_init();
3949 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
3954 r = kvm_arch_hardware_setup();
3958 for_each_online_cpu(cpu) {
3959 smp_call_function_single(cpu,
3960 kvm_arch_check_processor_compat,
3966 r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting",
3967 kvm_starting_cpu, kvm_dying_cpu);
3970 register_reboot_notifier(&kvm_reboot_notifier);
3972 /* A kmem cache lets us meet the alignment requirements of fx_save. */
3974 vcpu_align = __alignof__(struct kvm_vcpu);
3975 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
3977 if (!kvm_vcpu_cache) {
3982 r = kvm_async_pf_init();
3986 kvm_chardev_ops.owner = module;
3987 kvm_vm_fops.owner = module;
3988 kvm_vcpu_fops.owner = module;
3990 r = misc_register(&kvm_dev);
3992 pr_err("kvm: misc device register failed\n");
3996 register_syscore_ops(&kvm_syscore_ops);
3998 kvm_preempt_ops.sched_in = kvm_sched_in;
3999 kvm_preempt_ops.sched_out = kvm_sched_out;
4001 r = kvm_init_debug();
4003 pr_err("kvm: create debugfs files failed\n");
4007 r = kvm_vfio_ops_init();
4013 unregister_syscore_ops(&kvm_syscore_ops);
4014 misc_deregister(&kvm_dev);
4016 kvm_async_pf_deinit();
4018 kmem_cache_destroy(kvm_vcpu_cache);
4020 unregister_reboot_notifier(&kvm_reboot_notifier);
4021 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4024 kvm_arch_hardware_unsetup();
4026 free_cpumask_var(cpus_hardware_enabled);
4034 EXPORT_SYMBOL_GPL(kvm_init);
4038 debugfs_remove_recursive(kvm_debugfs_dir);
4039 misc_deregister(&kvm_dev);
4040 kmem_cache_destroy(kvm_vcpu_cache);
4041 kvm_async_pf_deinit();
4042 unregister_syscore_ops(&kvm_syscore_ops);
4043 unregister_reboot_notifier(&kvm_reboot_notifier);
4044 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4045 on_each_cpu(hardware_disable_nolock, NULL, 1);
4046 kvm_arch_hardware_unsetup();
4049 free_cpumask_var(cpus_hardware_enabled);
4050 kvm_vfio_ops_exit();
4052 EXPORT_SYMBOL_GPL(kvm_exit);