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 = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
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);
692 static void kvm_destroy_devices(struct kvm *kvm)
694 struct kvm_device *dev, *tmp;
697 * We do not need to take the kvm->lock here, because nobody else
698 * has a reference to the struct kvm at this point and therefore
699 * cannot access the devices list anyhow.
701 list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
702 list_del(&dev->vm_node);
703 dev->ops->destroy(dev);
707 static void kvm_destroy_vm(struct kvm *kvm)
710 struct mm_struct *mm = kvm->mm;
712 kvm_destroy_vm_debugfs(kvm);
713 kvm_arch_sync_events(kvm);
714 spin_lock(&kvm_lock);
715 list_del(&kvm->vm_list);
716 spin_unlock(&kvm_lock);
717 kvm_free_irq_routing(kvm);
718 for (i = 0; i < KVM_NR_BUSES; i++) {
720 kvm_io_bus_destroy(kvm->buses[i]);
721 kvm->buses[i] = NULL;
723 kvm_coalesced_mmio_free(kvm);
724 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
725 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
727 kvm_arch_flush_shadow_all(kvm);
729 kvm_arch_destroy_vm(kvm);
730 kvm_destroy_devices(kvm);
731 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
732 kvm_free_memslots(kvm, kvm->memslots[i]);
733 cleanup_srcu_struct(&kvm->irq_srcu);
734 cleanup_srcu_struct(&kvm->srcu);
735 kvm_arch_free_vm(kvm);
736 preempt_notifier_dec();
737 hardware_disable_all();
741 void kvm_get_kvm(struct kvm *kvm)
743 refcount_inc(&kvm->users_count);
745 EXPORT_SYMBOL_GPL(kvm_get_kvm);
747 void kvm_put_kvm(struct kvm *kvm)
749 if (refcount_dec_and_test(&kvm->users_count))
752 EXPORT_SYMBOL_GPL(kvm_put_kvm);
755 static int kvm_vm_release(struct inode *inode, struct file *filp)
757 struct kvm *kvm = filp->private_data;
759 kvm_irqfd_release(kvm);
766 * Allocation size is twice as large as the actual dirty bitmap size.
767 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
769 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
771 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
773 memslot->dirty_bitmap = kvzalloc(dirty_bytes, GFP_KERNEL);
774 if (!memslot->dirty_bitmap)
781 * Insert memslot and re-sort memslots based on their GFN,
782 * so binary search could be used to lookup GFN.
783 * Sorting algorithm takes advantage of having initially
784 * sorted array and known changed memslot position.
786 static void update_memslots(struct kvm_memslots *slots,
787 struct kvm_memory_slot *new)
790 int i = slots->id_to_index[id];
791 struct kvm_memory_slot *mslots = slots->memslots;
793 WARN_ON(mslots[i].id != id);
795 WARN_ON(!mslots[i].npages);
796 if (mslots[i].npages)
799 if (!mslots[i].npages)
803 while (i < KVM_MEM_SLOTS_NUM - 1 &&
804 new->base_gfn <= mslots[i + 1].base_gfn) {
805 if (!mslots[i + 1].npages)
807 mslots[i] = mslots[i + 1];
808 slots->id_to_index[mslots[i].id] = i;
813 * The ">=" is needed when creating a slot with base_gfn == 0,
814 * so that it moves before all those with base_gfn == npages == 0.
816 * On the other hand, if new->npages is zero, the above loop has
817 * already left i pointing to the beginning of the empty part of
818 * mslots, and the ">=" would move the hole backwards in this
819 * case---which is wrong. So skip the loop when deleting a slot.
823 new->base_gfn >= mslots[i - 1].base_gfn) {
824 mslots[i] = mslots[i - 1];
825 slots->id_to_index[mslots[i].id] = i;
829 WARN_ON_ONCE(i != slots->used_slots);
832 slots->id_to_index[mslots[i].id] = i;
835 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
837 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
839 #ifdef __KVM_HAVE_READONLY_MEM
840 valid_flags |= KVM_MEM_READONLY;
843 if (mem->flags & ~valid_flags)
849 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
850 int as_id, struct kvm_memslots *slots)
852 struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id);
855 * Set the low bit in the generation, which disables SPTE caching
856 * until the end of synchronize_srcu_expedited.
858 WARN_ON(old_memslots->generation & 1);
859 slots->generation = old_memslots->generation + 1;
861 rcu_assign_pointer(kvm->memslots[as_id], slots);
862 synchronize_srcu_expedited(&kvm->srcu);
865 * Increment the new memslot generation a second time. This prevents
866 * vm exits that race with memslot updates from caching a memslot
867 * generation that will (potentially) be valid forever.
869 * Generations must be unique even across address spaces. We do not need
870 * a global counter for that, instead the generation space is evenly split
871 * across address spaces. For example, with two address spaces, address
872 * space 0 will use generations 0, 4, 8, ... while * address space 1 will
873 * use generations 2, 6, 10, 14, ...
875 slots->generation += KVM_ADDRESS_SPACE_NUM * 2 - 1;
877 kvm_arch_memslots_updated(kvm, slots);
883 * Allocate some memory and give it an address in the guest physical address
886 * Discontiguous memory is allowed, mostly for framebuffers.
888 * Must be called holding kvm->slots_lock for write.
890 int __kvm_set_memory_region(struct kvm *kvm,
891 const struct kvm_userspace_memory_region *mem)
895 unsigned long npages;
896 struct kvm_memory_slot *slot;
897 struct kvm_memory_slot old, new;
898 struct kvm_memslots *slots = NULL, *old_memslots;
900 enum kvm_mr_change change;
902 r = check_memory_region_flags(mem);
907 as_id = mem->slot >> 16;
910 /* General sanity checks */
911 if (mem->memory_size & (PAGE_SIZE - 1))
913 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
915 /* We can read the guest memory with __xxx_user() later on. */
916 if ((id < KVM_USER_MEM_SLOTS) &&
917 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
918 !access_ok(VERIFY_WRITE,
919 (void __user *)(unsigned long)mem->userspace_addr,
922 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
924 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
927 slot = id_to_memslot(__kvm_memslots(kvm, as_id), id);
928 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
929 npages = mem->memory_size >> PAGE_SHIFT;
931 if (npages > KVM_MEM_MAX_NR_PAGES)
937 new.base_gfn = base_gfn;
939 new.flags = mem->flags;
943 change = KVM_MR_CREATE;
944 else { /* Modify an existing slot. */
945 if ((mem->userspace_addr != old.userspace_addr) ||
946 (npages != old.npages) ||
947 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
950 if (base_gfn != old.base_gfn)
951 change = KVM_MR_MOVE;
952 else if (new.flags != old.flags)
953 change = KVM_MR_FLAGS_ONLY;
954 else { /* Nothing to change. */
963 change = KVM_MR_DELETE;
968 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
969 /* Check for overlaps */
971 kvm_for_each_memslot(slot, __kvm_memslots(kvm, as_id)) {
972 if ((slot->id >= KVM_USER_MEM_SLOTS) ||
975 if (!((base_gfn + npages <= slot->base_gfn) ||
976 (base_gfn >= slot->base_gfn + slot->npages)))
981 /* Free page dirty bitmap if unneeded */
982 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
983 new.dirty_bitmap = NULL;
986 if (change == KVM_MR_CREATE) {
987 new.userspace_addr = mem->userspace_addr;
989 if (kvm_arch_create_memslot(kvm, &new, npages))
993 /* Allocate page dirty bitmap if needed */
994 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
995 if (kvm_create_dirty_bitmap(&new) < 0)
999 slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
1002 memcpy(slots, __kvm_memslots(kvm, as_id), sizeof(struct kvm_memslots));
1004 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
1005 slot = id_to_memslot(slots, id);
1006 slot->flags |= KVM_MEMSLOT_INVALID;
1008 old_memslots = install_new_memslots(kvm, as_id, slots);
1010 /* slot was deleted or moved, clear iommu mapping */
1011 kvm_iommu_unmap_pages(kvm, &old);
1012 /* From this point no new shadow pages pointing to a deleted,
1013 * or moved, memslot will be created.
1015 * validation of sp->gfn happens in:
1016 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1017 * - kvm_is_visible_gfn (mmu_check_roots)
1019 kvm_arch_flush_shadow_memslot(kvm, slot);
1022 * We can re-use the old_memslots from above, the only difference
1023 * from the currently installed memslots is the invalid flag. This
1024 * will get overwritten by update_memslots anyway.
1026 slots = old_memslots;
1029 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
1033 /* actual memory is freed via old in kvm_free_memslot below */
1034 if (change == KVM_MR_DELETE) {
1035 new.dirty_bitmap = NULL;
1036 memset(&new.arch, 0, sizeof(new.arch));
1039 update_memslots(slots, &new);
1040 old_memslots = install_new_memslots(kvm, as_id, slots);
1042 kvm_arch_commit_memory_region(kvm, mem, &old, &new, change);
1044 kvm_free_memslot(kvm, &old, &new);
1045 kvfree(old_memslots);
1048 * IOMMU mapping: New slots need to be mapped. Old slots need to be
1049 * un-mapped and re-mapped if their base changes. Since base change
1050 * unmapping is handled above with slot deletion, mapping alone is
1051 * needed here. Anything else the iommu might care about for existing
1052 * slots (size changes, userspace addr changes and read-only flag
1053 * changes) is disallowed above, so any other attribute changes getting
1054 * here can be skipped.
1056 if (as_id == 0 && (change == KVM_MR_CREATE || change == KVM_MR_MOVE)) {
1057 r = kvm_iommu_map_pages(kvm, &new);
1066 kvm_free_memslot(kvm, &new, &old);
1070 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1072 int kvm_set_memory_region(struct kvm *kvm,
1073 const struct kvm_userspace_memory_region *mem)
1077 mutex_lock(&kvm->slots_lock);
1078 r = __kvm_set_memory_region(kvm, mem);
1079 mutex_unlock(&kvm->slots_lock);
1082 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1084 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1085 struct kvm_userspace_memory_region *mem)
1087 if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1090 return kvm_set_memory_region(kvm, mem);
1093 int kvm_get_dirty_log(struct kvm *kvm,
1094 struct kvm_dirty_log *log, int *is_dirty)
1096 struct kvm_memslots *slots;
1097 struct kvm_memory_slot *memslot;
1100 unsigned long any = 0;
1102 as_id = log->slot >> 16;
1103 id = (u16)log->slot;
1104 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1107 slots = __kvm_memslots(kvm, as_id);
1108 memslot = id_to_memslot(slots, id);
1109 if (!memslot->dirty_bitmap)
1112 n = kvm_dirty_bitmap_bytes(memslot);
1114 for (i = 0; !any && i < n/sizeof(long); ++i)
1115 any = memslot->dirty_bitmap[i];
1117 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1124 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1126 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1128 * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages
1129 * are dirty write protect them for next write.
1130 * @kvm: pointer to kvm instance
1131 * @log: slot id and address to which we copy the log
1132 * @is_dirty: flag set if any page is dirty
1134 * We need to keep it in mind that VCPU threads can write to the bitmap
1135 * concurrently. So, to avoid losing track of dirty pages we keep the
1138 * 1. Take a snapshot of the bit and clear it if needed.
1139 * 2. Write protect the corresponding page.
1140 * 3. Copy the snapshot to the userspace.
1141 * 4. Upon return caller flushes TLB's if needed.
1143 * Between 2 and 4, the guest may write to the page using the remaining TLB
1144 * entry. This is not a problem because the page is reported dirty using
1145 * the snapshot taken before and step 4 ensures that writes done after
1146 * exiting to userspace will be logged for the next call.
1149 int kvm_get_dirty_log_protect(struct kvm *kvm,
1150 struct kvm_dirty_log *log, bool *is_dirty)
1152 struct kvm_memslots *slots;
1153 struct kvm_memory_slot *memslot;
1156 unsigned long *dirty_bitmap;
1157 unsigned long *dirty_bitmap_buffer;
1159 as_id = log->slot >> 16;
1160 id = (u16)log->slot;
1161 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1164 slots = __kvm_memslots(kvm, as_id);
1165 memslot = id_to_memslot(slots, id);
1167 dirty_bitmap = memslot->dirty_bitmap;
1171 n = kvm_dirty_bitmap_bytes(memslot);
1173 dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long);
1174 memset(dirty_bitmap_buffer, 0, n);
1176 spin_lock(&kvm->mmu_lock);
1178 for (i = 0; i < n / sizeof(long); i++) {
1182 if (!dirty_bitmap[i])
1187 mask = xchg(&dirty_bitmap[i], 0);
1188 dirty_bitmap_buffer[i] = mask;
1191 offset = i * BITS_PER_LONG;
1192 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1197 spin_unlock(&kvm->mmu_lock);
1198 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1202 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect);
1205 bool kvm_largepages_enabled(void)
1207 return largepages_enabled;
1210 void kvm_disable_largepages(void)
1212 largepages_enabled = false;
1214 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1216 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1218 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1220 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1222 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
1224 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn);
1227 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1229 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1231 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1232 memslot->flags & KVM_MEMSLOT_INVALID)
1237 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1239 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1241 struct vm_area_struct *vma;
1242 unsigned long addr, size;
1246 addr = gfn_to_hva(kvm, gfn);
1247 if (kvm_is_error_hva(addr))
1250 down_read(¤t->mm->mmap_sem);
1251 vma = find_vma(current->mm, addr);
1255 size = vma_kernel_pagesize(vma);
1258 up_read(¤t->mm->mmap_sem);
1263 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1265 return slot->flags & KVM_MEM_READONLY;
1268 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1269 gfn_t *nr_pages, bool write)
1271 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1272 return KVM_HVA_ERR_BAD;
1274 if (memslot_is_readonly(slot) && write)
1275 return KVM_HVA_ERR_RO_BAD;
1278 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1280 return __gfn_to_hva_memslot(slot, gfn);
1283 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1286 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1289 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1292 return gfn_to_hva_many(slot, gfn, NULL);
1294 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1296 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1298 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1300 EXPORT_SYMBOL_GPL(gfn_to_hva);
1302 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
1304 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
1306 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
1309 * If writable is set to false, the hva returned by this function is only
1310 * allowed to be read.
1312 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1313 gfn_t gfn, bool *writable)
1315 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1317 if (!kvm_is_error_hva(hva) && writable)
1318 *writable = !memslot_is_readonly(slot);
1323 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1325 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1327 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1330 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
1332 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1334 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1337 static int get_user_page_nowait(unsigned long start, int write,
1340 int flags = FOLL_NOWAIT | FOLL_HWPOISON;
1343 flags |= FOLL_WRITE;
1345 return get_user_pages(start, 1, flags, page, NULL);
1348 static inline int check_user_page_hwpoison(unsigned long addr)
1350 int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
1352 rc = get_user_pages(addr, 1, flags, NULL, NULL);
1353 return rc == -EHWPOISON;
1357 * The atomic path to get the writable pfn which will be stored in @pfn,
1358 * true indicates success, otherwise false is returned.
1360 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1361 bool write_fault, bool *writable, kvm_pfn_t *pfn)
1363 struct page *page[1];
1366 if (!(async || atomic))
1370 * Fast pin a writable pfn only if it is a write fault request
1371 * or the caller allows to map a writable pfn for a read fault
1374 if (!(write_fault || writable))
1377 npages = __get_user_pages_fast(addr, 1, 1, page);
1379 *pfn = page_to_pfn(page[0]);
1390 * The slow path to get the pfn of the specified host virtual address,
1391 * 1 indicates success, -errno is returned if error is detected.
1393 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1394 bool *writable, kvm_pfn_t *pfn)
1396 struct page *page[1];
1402 *writable = write_fault;
1405 down_read(¤t->mm->mmap_sem);
1406 npages = get_user_page_nowait(addr, write_fault, page);
1407 up_read(¤t->mm->mmap_sem);
1409 unsigned int flags = FOLL_HWPOISON;
1412 flags |= FOLL_WRITE;
1414 npages = get_user_pages_unlocked(addr, 1, page, flags);
1419 /* map read fault as writable if possible */
1420 if (unlikely(!write_fault) && writable) {
1421 struct page *wpage[1];
1423 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1432 *pfn = page_to_pfn(page[0]);
1436 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1438 if (unlikely(!(vma->vm_flags & VM_READ)))
1441 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1447 static int hva_to_pfn_remapped(struct vm_area_struct *vma,
1448 unsigned long addr, bool *async,
1449 bool write_fault, kvm_pfn_t *p_pfn)
1454 r = follow_pfn(vma, addr, &pfn);
1457 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
1458 * not call the fault handler, so do it here.
1460 bool unlocked = false;
1461 r = fixup_user_fault(current, current->mm, addr,
1462 (write_fault ? FAULT_FLAG_WRITE : 0),
1469 r = follow_pfn(vma, addr, &pfn);
1477 * Get a reference here because callers of *hva_to_pfn* and
1478 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
1479 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
1480 * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will
1481 * simply do nothing for reserved pfns.
1483 * Whoever called remap_pfn_range is also going to call e.g.
1484 * unmap_mapping_range before the underlying pages are freed,
1485 * causing a call to our MMU notifier.
1494 * Pin guest page in memory and return its pfn.
1495 * @addr: host virtual address which maps memory to the guest
1496 * @atomic: whether this function can sleep
1497 * @async: whether this function need to wait IO complete if the
1498 * host page is not in the memory
1499 * @write_fault: whether we should get a writable host page
1500 * @writable: whether it allows to map a writable host page for !@write_fault
1502 * The function will map a writable host page for these two cases:
1503 * 1): @write_fault = true
1504 * 2): @write_fault = false && @writable, @writable will tell the caller
1505 * whether the mapping is writable.
1507 static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1508 bool write_fault, bool *writable)
1510 struct vm_area_struct *vma;
1514 /* we can do it either atomically or asynchronously, not both */
1515 BUG_ON(atomic && async);
1517 if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1521 return KVM_PFN_ERR_FAULT;
1523 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1527 down_read(¤t->mm->mmap_sem);
1528 if (npages == -EHWPOISON ||
1529 (!async && check_user_page_hwpoison(addr))) {
1530 pfn = KVM_PFN_ERR_HWPOISON;
1535 vma = find_vma_intersection(current->mm, addr, addr + 1);
1538 pfn = KVM_PFN_ERR_FAULT;
1539 else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
1540 r = hva_to_pfn_remapped(vma, addr, async, write_fault, &pfn);
1544 pfn = KVM_PFN_ERR_FAULT;
1546 if (async && vma_is_valid(vma, write_fault))
1548 pfn = KVM_PFN_ERR_FAULT;
1551 up_read(¤t->mm->mmap_sem);
1555 kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn,
1556 bool atomic, bool *async, bool write_fault,
1559 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1561 if (addr == KVM_HVA_ERR_RO_BAD) {
1564 return KVM_PFN_ERR_RO_FAULT;
1567 if (kvm_is_error_hva(addr)) {
1570 return KVM_PFN_NOSLOT;
1573 /* Do not map writable pfn in the readonly memslot. */
1574 if (writable && memslot_is_readonly(slot)) {
1579 return hva_to_pfn(addr, atomic, async, write_fault,
1582 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
1584 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1587 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
1588 write_fault, writable);
1590 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1592 kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1594 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1596 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
1598 kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1600 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1602 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1604 kvm_pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1606 return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm, gfn), gfn);
1608 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1610 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
1612 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1614 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
1616 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1618 return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
1620 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1622 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
1624 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1626 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
1628 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1629 struct page **pages, int nr_pages)
1634 addr = gfn_to_hva_many(slot, gfn, &entry);
1635 if (kvm_is_error_hva(addr))
1638 if (entry < nr_pages)
1641 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1643 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1645 static struct page *kvm_pfn_to_page(kvm_pfn_t pfn)
1647 if (is_error_noslot_pfn(pfn))
1648 return KVM_ERR_PTR_BAD_PAGE;
1650 if (kvm_is_reserved_pfn(pfn)) {
1652 return KVM_ERR_PTR_BAD_PAGE;
1655 return pfn_to_page(pfn);
1658 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1662 pfn = gfn_to_pfn(kvm, gfn);
1664 return kvm_pfn_to_page(pfn);
1666 EXPORT_SYMBOL_GPL(gfn_to_page);
1668 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
1672 pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
1674 return kvm_pfn_to_page(pfn);
1676 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
1678 void kvm_release_page_clean(struct page *page)
1680 WARN_ON(is_error_page(page));
1682 kvm_release_pfn_clean(page_to_pfn(page));
1684 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1686 void kvm_release_pfn_clean(kvm_pfn_t pfn)
1688 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
1689 put_page(pfn_to_page(pfn));
1691 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1693 void kvm_release_page_dirty(struct page *page)
1695 WARN_ON(is_error_page(page));
1697 kvm_release_pfn_dirty(page_to_pfn(page));
1699 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1701 static void kvm_release_pfn_dirty(kvm_pfn_t pfn)
1703 kvm_set_pfn_dirty(pfn);
1704 kvm_release_pfn_clean(pfn);
1707 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
1709 if (!kvm_is_reserved_pfn(pfn)) {
1710 struct page *page = pfn_to_page(pfn);
1712 if (!PageReserved(page))
1716 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1718 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
1720 if (!kvm_is_reserved_pfn(pfn))
1721 mark_page_accessed(pfn_to_page(pfn));
1723 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1725 void kvm_get_pfn(kvm_pfn_t pfn)
1727 if (!kvm_is_reserved_pfn(pfn))
1728 get_page(pfn_to_page(pfn));
1730 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1732 static int next_segment(unsigned long len, int offset)
1734 if (len > PAGE_SIZE - offset)
1735 return PAGE_SIZE - offset;
1740 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
1741 void *data, int offset, int len)
1746 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1747 if (kvm_is_error_hva(addr))
1749 r = __copy_from_user(data, (void __user *)addr + offset, len);
1755 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1758 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1760 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1762 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1764 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
1765 int offset, int len)
1767 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1769 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1771 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
1773 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1775 gfn_t gfn = gpa >> PAGE_SHIFT;
1777 int offset = offset_in_page(gpa);
1780 while ((seg = next_segment(len, offset)) != 0) {
1781 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1791 EXPORT_SYMBOL_GPL(kvm_read_guest);
1793 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
1795 gfn_t gfn = gpa >> PAGE_SHIFT;
1797 int offset = offset_in_page(gpa);
1800 while ((seg = next_segment(len, offset)) != 0) {
1801 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
1811 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
1813 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1814 void *data, int offset, unsigned long len)
1819 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1820 if (kvm_is_error_hva(addr))
1822 pagefault_disable();
1823 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1830 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1833 gfn_t gfn = gpa >> PAGE_SHIFT;
1834 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1835 int offset = offset_in_page(gpa);
1837 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1839 EXPORT_SYMBOL_GPL(kvm_read_guest_atomic);
1841 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
1842 void *data, unsigned long len)
1844 gfn_t gfn = gpa >> PAGE_SHIFT;
1845 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1846 int offset = offset_in_page(gpa);
1848 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1850 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
1852 static int __kvm_write_guest_page(struct kvm_memory_slot *memslot, gfn_t gfn,
1853 const void *data, int offset, int len)
1858 addr = gfn_to_hva_memslot(memslot, gfn);
1859 if (kvm_is_error_hva(addr))
1861 r = __copy_to_user((void __user *)addr + offset, data, len);
1864 mark_page_dirty_in_slot(memslot, gfn);
1868 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
1869 const void *data, int offset, int len)
1871 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1873 return __kvm_write_guest_page(slot, gfn, data, offset, len);
1875 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1877 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
1878 const void *data, int offset, int len)
1880 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1882 return __kvm_write_guest_page(slot, gfn, data, offset, len);
1884 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
1886 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1889 gfn_t gfn = gpa >> PAGE_SHIFT;
1891 int offset = offset_in_page(gpa);
1894 while ((seg = next_segment(len, offset)) != 0) {
1895 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1905 EXPORT_SYMBOL_GPL(kvm_write_guest);
1907 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
1910 gfn_t gfn = gpa >> PAGE_SHIFT;
1912 int offset = offset_in_page(gpa);
1915 while ((seg = next_segment(len, offset)) != 0) {
1916 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
1926 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
1928 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
1929 struct gfn_to_hva_cache *ghc,
1930 gpa_t gpa, unsigned long len)
1932 int offset = offset_in_page(gpa);
1933 gfn_t start_gfn = gpa >> PAGE_SHIFT;
1934 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1935 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1936 gfn_t nr_pages_avail;
1939 ghc->generation = slots->generation;
1941 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
1942 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, NULL);
1943 if (!kvm_is_error_hva(ghc->hva) && nr_pages_needed <= 1) {
1947 * If the requested region crosses two memslots, we still
1948 * verify that the entire region is valid here.
1950 while (start_gfn <= end_gfn) {
1951 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
1952 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1954 if (kvm_is_error_hva(ghc->hva))
1956 start_gfn += nr_pages_avail;
1958 /* Use the slow path for cross page reads and writes. */
1959 ghc->memslot = NULL;
1964 int kvm_vcpu_gfn_to_hva_cache_init(struct kvm_vcpu *vcpu, struct gfn_to_hva_cache *ghc,
1965 gpa_t gpa, unsigned long len)
1967 struct kvm_memslots *slots = kvm_vcpu_memslots(vcpu);
1968 return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len);
1970 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva_cache_init);
1972 int kvm_vcpu_write_guest_offset_cached(struct kvm_vcpu *vcpu, struct gfn_to_hva_cache *ghc,
1973 void *data, int offset, unsigned long len)
1975 struct kvm_memslots *slots = kvm_vcpu_memslots(vcpu);
1977 gpa_t gpa = ghc->gpa + offset;
1979 BUG_ON(len + offset > ghc->len);
1981 if (slots->generation != ghc->generation)
1982 __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len);
1984 if (unlikely(!ghc->memslot))
1985 return kvm_vcpu_write_guest(vcpu, gpa, data, len);
1987 if (kvm_is_error_hva(ghc->hva))
1990 r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
1993 mark_page_dirty_in_slot(ghc->memslot, gpa >> PAGE_SHIFT);
1997 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_offset_cached);
1999 int kvm_vcpu_write_guest_cached(struct kvm_vcpu *vcpu, struct gfn_to_hva_cache *ghc,
2000 void *data, unsigned long len)
2002 return kvm_vcpu_write_guest_offset_cached(vcpu, ghc, data, 0, len);
2004 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_cached);
2006 int kvm_vcpu_read_guest_cached(struct kvm_vcpu *vcpu, struct gfn_to_hva_cache *ghc,
2007 void *data, unsigned long len)
2009 struct kvm_memslots *slots = kvm_vcpu_memslots(vcpu);
2012 BUG_ON(len > ghc->len);
2014 if (slots->generation != ghc->generation)
2015 __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len);
2017 if (unlikely(!ghc->memslot))
2018 return kvm_vcpu_read_guest(vcpu, ghc->gpa, data, len);
2020 if (kvm_is_error_hva(ghc->hva))
2023 r = __copy_from_user(data, (void __user *)ghc->hva, len);
2029 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_cached);
2031 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
2033 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
2035 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
2037 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
2039 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
2041 gfn_t gfn = gpa >> PAGE_SHIFT;
2043 int offset = offset_in_page(gpa);
2046 while ((seg = next_segment(len, offset)) != 0) {
2047 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
2056 EXPORT_SYMBOL_GPL(kvm_clear_guest);
2058 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot,
2061 if (memslot && memslot->dirty_bitmap) {
2062 unsigned long rel_gfn = gfn - memslot->base_gfn;
2064 set_bit_le(rel_gfn, memslot->dirty_bitmap);
2068 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
2070 struct kvm_memory_slot *memslot;
2072 memslot = gfn_to_memslot(kvm, gfn);
2073 mark_page_dirty_in_slot(memslot, gfn);
2075 EXPORT_SYMBOL_GPL(mark_page_dirty);
2077 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
2079 struct kvm_memory_slot *memslot;
2081 memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2082 mark_page_dirty_in_slot(memslot, gfn);
2084 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
2086 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
2088 unsigned int old, val, grow;
2090 old = val = vcpu->halt_poll_ns;
2091 grow = READ_ONCE(halt_poll_ns_grow);
2093 if (val == 0 && grow)
2098 if (val > halt_poll_ns)
2101 vcpu->halt_poll_ns = val;
2102 trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
2105 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
2107 unsigned int old, val, shrink;
2109 old = val = vcpu->halt_poll_ns;
2110 shrink = READ_ONCE(halt_poll_ns_shrink);
2116 vcpu->halt_poll_ns = val;
2117 trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
2120 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
2122 if (kvm_arch_vcpu_runnable(vcpu)) {
2123 kvm_make_request(KVM_REQ_UNHALT, vcpu);
2126 if (kvm_cpu_has_pending_timer(vcpu))
2128 if (signal_pending(current))
2135 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2137 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
2140 DECLARE_SWAITQUEUE(wait);
2141 bool waited = false;
2144 start = cur = ktime_get();
2145 if (vcpu->halt_poll_ns) {
2146 ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns);
2148 ++vcpu->stat.halt_attempted_poll;
2151 * This sets KVM_REQ_UNHALT if an interrupt
2154 if (kvm_vcpu_check_block(vcpu) < 0) {
2155 ++vcpu->stat.halt_successful_poll;
2156 if (!vcpu_valid_wakeup(vcpu))
2157 ++vcpu->stat.halt_poll_invalid;
2161 } while (single_task_running() && ktime_before(cur, stop));
2164 kvm_arch_vcpu_blocking(vcpu);
2167 prepare_to_swait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
2169 if (kvm_vcpu_check_block(vcpu) < 0)
2176 finish_swait(&vcpu->wq, &wait);
2179 kvm_arch_vcpu_unblocking(vcpu);
2181 block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
2183 if (!vcpu_valid_wakeup(vcpu))
2184 shrink_halt_poll_ns(vcpu);
2185 else if (halt_poll_ns) {
2186 if (block_ns <= vcpu->halt_poll_ns)
2188 /* we had a long block, shrink polling */
2189 else if (vcpu->halt_poll_ns && block_ns > halt_poll_ns)
2190 shrink_halt_poll_ns(vcpu);
2191 /* we had a short halt and our poll time is too small */
2192 else if (vcpu->halt_poll_ns < halt_poll_ns &&
2193 block_ns < halt_poll_ns)
2194 grow_halt_poll_ns(vcpu);
2196 vcpu->halt_poll_ns = 0;
2198 trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu));
2199 kvm_arch_vcpu_block_finish(vcpu);
2201 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
2204 void kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
2206 struct swait_queue_head *wqp;
2208 wqp = kvm_arch_vcpu_wq(vcpu);
2209 if (swait_active(wqp)) {
2211 ++vcpu->stat.halt_wakeup;
2215 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
2218 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2220 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
2223 int cpu = vcpu->cpu;
2225 kvm_vcpu_wake_up(vcpu);
2227 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
2228 if (kvm_arch_vcpu_should_kick(vcpu))
2229 smp_send_reschedule(cpu);
2232 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
2233 #endif /* !CONFIG_S390 */
2235 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
2238 struct task_struct *task = NULL;
2242 pid = rcu_dereference(target->pid);
2244 task = get_pid_task(pid, PIDTYPE_PID);
2248 ret = yield_to(task, 1);
2249 put_task_struct(task);
2253 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
2256 * Helper that checks whether a VCPU is eligible for directed yield.
2257 * Most eligible candidate to yield is decided by following heuristics:
2259 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2260 * (preempted lock holder), indicated by @in_spin_loop.
2261 * Set at the beiginning and cleared at the end of interception/PLE handler.
2263 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2264 * chance last time (mostly it has become eligible now since we have probably
2265 * yielded to lockholder in last iteration. This is done by toggling
2266 * @dy_eligible each time a VCPU checked for eligibility.)
2268 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2269 * to preempted lock-holder could result in wrong VCPU selection and CPU
2270 * burning. Giving priority for a potential lock-holder increases lock
2273 * Since algorithm is based on heuristics, accessing another VCPU data without
2274 * locking does not harm. It may result in trying to yield to same VCPU, fail
2275 * and continue with next VCPU and so on.
2277 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
2279 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2282 eligible = !vcpu->spin_loop.in_spin_loop ||
2283 vcpu->spin_loop.dy_eligible;
2285 if (vcpu->spin_loop.in_spin_loop)
2286 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
2294 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
2296 struct kvm *kvm = me->kvm;
2297 struct kvm_vcpu *vcpu;
2298 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
2304 kvm_vcpu_set_in_spin_loop(me, true);
2306 * We boost the priority of a VCPU that is runnable but not
2307 * currently running, because it got preempted by something
2308 * else and called schedule in __vcpu_run. Hopefully that
2309 * VCPU is holding the lock that we need and will release it.
2310 * We approximate round-robin by starting at the last boosted VCPU.
2312 for (pass = 0; pass < 2 && !yielded && try; pass++) {
2313 kvm_for_each_vcpu(i, vcpu, kvm) {
2314 if (!pass && i <= last_boosted_vcpu) {
2315 i = last_boosted_vcpu;
2317 } else if (pass && i > last_boosted_vcpu)
2319 if (!ACCESS_ONCE(vcpu->preempted))
2323 if (swait_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
2325 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
2328 yielded = kvm_vcpu_yield_to(vcpu);
2330 kvm->last_boosted_vcpu = i;
2332 } else if (yielded < 0) {
2339 kvm_vcpu_set_in_spin_loop(me, false);
2341 /* Ensure vcpu is not eligible during next spinloop */
2342 kvm_vcpu_set_dy_eligible(me, false);
2344 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
2346 static int kvm_vcpu_fault(struct vm_fault *vmf)
2348 struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data;
2351 if (vmf->pgoff == 0)
2352 page = virt_to_page(vcpu->run);
2354 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
2355 page = virt_to_page(vcpu->arch.pio_data);
2357 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2358 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
2359 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
2362 return kvm_arch_vcpu_fault(vcpu, vmf);
2368 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
2369 .fault = kvm_vcpu_fault,
2372 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2374 vma->vm_ops = &kvm_vcpu_vm_ops;
2378 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2380 struct kvm_vcpu *vcpu = filp->private_data;
2382 debugfs_remove_recursive(vcpu->debugfs_dentry);
2383 kvm_put_kvm(vcpu->kvm);
2387 static struct file_operations kvm_vcpu_fops = {
2388 .release = kvm_vcpu_release,
2389 .unlocked_ioctl = kvm_vcpu_ioctl,
2390 #ifdef CONFIG_KVM_COMPAT
2391 .compat_ioctl = kvm_vcpu_compat_ioctl,
2393 .mmap = kvm_vcpu_mmap,
2394 .llseek = noop_llseek,
2398 * Allocates an inode for the vcpu.
2400 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2402 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
2405 static int kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
2407 char dir_name[ITOA_MAX_LEN * 2];
2410 if (!kvm_arch_has_vcpu_debugfs())
2413 if (!debugfs_initialized())
2416 snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
2417 vcpu->debugfs_dentry = debugfs_create_dir(dir_name,
2418 vcpu->kvm->debugfs_dentry);
2419 if (!vcpu->debugfs_dentry)
2422 ret = kvm_arch_create_vcpu_debugfs(vcpu);
2424 debugfs_remove_recursive(vcpu->debugfs_dentry);
2432 * Creates some virtual cpus. Good luck creating more than one.
2434 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
2437 struct kvm_vcpu *vcpu;
2439 if (id >= KVM_MAX_VCPU_ID)
2442 mutex_lock(&kvm->lock);
2443 if (kvm->created_vcpus == KVM_MAX_VCPUS) {
2444 mutex_unlock(&kvm->lock);
2448 kvm->created_vcpus++;
2449 mutex_unlock(&kvm->lock);
2451 vcpu = kvm_arch_vcpu_create(kvm, id);
2454 goto vcpu_decrement;
2457 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2459 r = kvm_arch_vcpu_setup(vcpu);
2463 r = kvm_create_vcpu_debugfs(vcpu);
2467 mutex_lock(&kvm->lock);
2468 if (kvm_get_vcpu_by_id(kvm, id)) {
2470 goto unlock_vcpu_destroy;
2473 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
2475 /* Now it's all set up, let userspace reach it */
2477 r = create_vcpu_fd(vcpu);
2480 goto unlock_vcpu_destroy;
2483 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
2486 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
2487 * before kvm->online_vcpu's incremented value.
2490 atomic_inc(&kvm->online_vcpus);
2492 mutex_unlock(&kvm->lock);
2493 kvm_arch_vcpu_postcreate(vcpu);
2496 unlock_vcpu_destroy:
2497 mutex_unlock(&kvm->lock);
2498 debugfs_remove_recursive(vcpu->debugfs_dentry);
2500 kvm_arch_vcpu_destroy(vcpu);
2502 mutex_lock(&kvm->lock);
2503 kvm->created_vcpus--;
2504 mutex_unlock(&kvm->lock);
2508 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2511 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2512 vcpu->sigset_active = 1;
2513 vcpu->sigset = *sigset;
2515 vcpu->sigset_active = 0;
2519 static long kvm_vcpu_ioctl(struct file *filp,
2520 unsigned int ioctl, unsigned long arg)
2522 struct kvm_vcpu *vcpu = filp->private_data;
2523 void __user *argp = (void __user *)arg;
2525 struct kvm_fpu *fpu = NULL;
2526 struct kvm_sregs *kvm_sregs = NULL;
2528 if (vcpu->kvm->mm != current->mm)
2531 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2534 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
2536 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
2537 * so vcpu_load() would break it.
2539 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_S390_IRQ || ioctl == KVM_INTERRUPT)
2540 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2544 r = vcpu_load(vcpu);
2552 if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
2553 /* The thread running this VCPU changed. */
2554 struct pid *oldpid = vcpu->pid;
2555 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
2557 rcu_assign_pointer(vcpu->pid, newpid);
2562 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2563 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2565 case KVM_GET_REGS: {
2566 struct kvm_regs *kvm_regs;
2569 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2572 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2576 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2583 case KVM_SET_REGS: {
2584 struct kvm_regs *kvm_regs;
2587 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2588 if (IS_ERR(kvm_regs)) {
2589 r = PTR_ERR(kvm_regs);
2592 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2596 case KVM_GET_SREGS: {
2597 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2601 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2605 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2610 case KVM_SET_SREGS: {
2611 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2612 if (IS_ERR(kvm_sregs)) {
2613 r = PTR_ERR(kvm_sregs);
2617 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2620 case KVM_GET_MP_STATE: {
2621 struct kvm_mp_state mp_state;
2623 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2627 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
2632 case KVM_SET_MP_STATE: {
2633 struct kvm_mp_state mp_state;
2636 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
2638 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2641 case KVM_TRANSLATE: {
2642 struct kvm_translation tr;
2645 if (copy_from_user(&tr, argp, sizeof(tr)))
2647 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2651 if (copy_to_user(argp, &tr, sizeof(tr)))
2656 case KVM_SET_GUEST_DEBUG: {
2657 struct kvm_guest_debug dbg;
2660 if (copy_from_user(&dbg, argp, sizeof(dbg)))
2662 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2665 case KVM_SET_SIGNAL_MASK: {
2666 struct kvm_signal_mask __user *sigmask_arg = argp;
2667 struct kvm_signal_mask kvm_sigmask;
2668 sigset_t sigset, *p;
2673 if (copy_from_user(&kvm_sigmask, argp,
2674 sizeof(kvm_sigmask)))
2677 if (kvm_sigmask.len != sizeof(sigset))
2680 if (copy_from_user(&sigset, sigmask_arg->sigset,
2685 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2689 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2693 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2697 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2703 fpu = memdup_user(argp, sizeof(*fpu));
2709 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2713 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2722 #ifdef CONFIG_KVM_COMPAT
2723 static long kvm_vcpu_compat_ioctl(struct file *filp,
2724 unsigned int ioctl, unsigned long arg)
2726 struct kvm_vcpu *vcpu = filp->private_data;
2727 void __user *argp = compat_ptr(arg);
2730 if (vcpu->kvm->mm != current->mm)
2734 case KVM_SET_SIGNAL_MASK: {
2735 struct kvm_signal_mask __user *sigmask_arg = argp;
2736 struct kvm_signal_mask kvm_sigmask;
2737 compat_sigset_t csigset;
2742 if (copy_from_user(&kvm_sigmask, argp,
2743 sizeof(kvm_sigmask)))
2746 if (kvm_sigmask.len != sizeof(csigset))
2749 if (copy_from_user(&csigset, sigmask_arg->sigset,
2752 sigset_from_compat(&sigset, &csigset);
2753 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2755 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2759 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2767 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2768 int (*accessor)(struct kvm_device *dev,
2769 struct kvm_device_attr *attr),
2772 struct kvm_device_attr attr;
2777 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2780 return accessor(dev, &attr);
2783 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2786 struct kvm_device *dev = filp->private_data;
2789 case KVM_SET_DEVICE_ATTR:
2790 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2791 case KVM_GET_DEVICE_ATTR:
2792 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2793 case KVM_HAS_DEVICE_ATTR:
2794 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2796 if (dev->ops->ioctl)
2797 return dev->ops->ioctl(dev, ioctl, arg);
2803 static int kvm_device_release(struct inode *inode, struct file *filp)
2805 struct kvm_device *dev = filp->private_data;
2806 struct kvm *kvm = dev->kvm;
2812 static const struct file_operations kvm_device_fops = {
2813 .unlocked_ioctl = kvm_device_ioctl,
2814 #ifdef CONFIG_KVM_COMPAT
2815 .compat_ioctl = kvm_device_ioctl,
2817 .release = kvm_device_release,
2820 struct kvm_device *kvm_device_from_filp(struct file *filp)
2822 if (filp->f_op != &kvm_device_fops)
2825 return filp->private_data;
2828 static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
2829 #ifdef CONFIG_KVM_MPIC
2830 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
2831 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
2834 #ifdef CONFIG_KVM_XICS
2835 [KVM_DEV_TYPE_XICS] = &kvm_xics_ops,
2839 int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
2841 if (type >= ARRAY_SIZE(kvm_device_ops_table))
2844 if (kvm_device_ops_table[type] != NULL)
2847 kvm_device_ops_table[type] = ops;
2851 void kvm_unregister_device_ops(u32 type)
2853 if (kvm_device_ops_table[type] != NULL)
2854 kvm_device_ops_table[type] = NULL;
2857 static int kvm_ioctl_create_device(struct kvm *kvm,
2858 struct kvm_create_device *cd)
2860 struct kvm_device_ops *ops = NULL;
2861 struct kvm_device *dev;
2862 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2865 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
2868 ops = kvm_device_ops_table[cd->type];
2875 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2882 mutex_lock(&kvm->lock);
2883 ret = ops->create(dev, cd->type);
2885 mutex_unlock(&kvm->lock);
2889 list_add(&dev->vm_node, &kvm->devices);
2890 mutex_unlock(&kvm->lock);
2895 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2897 mutex_lock(&kvm->lock);
2898 list_del(&dev->vm_node);
2899 mutex_unlock(&kvm->lock);
2909 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2912 case KVM_CAP_USER_MEMORY:
2913 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2914 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2915 case KVM_CAP_INTERNAL_ERROR_DATA:
2916 #ifdef CONFIG_HAVE_KVM_MSI
2917 case KVM_CAP_SIGNAL_MSI:
2919 #ifdef CONFIG_HAVE_KVM_IRQFD
2921 case KVM_CAP_IRQFD_RESAMPLE:
2923 case KVM_CAP_IOEVENTFD_ANY_LENGTH:
2924 case KVM_CAP_CHECK_EXTENSION_VM:
2926 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2927 case KVM_CAP_IRQ_ROUTING:
2928 return KVM_MAX_IRQ_ROUTES;
2930 #if KVM_ADDRESS_SPACE_NUM > 1
2931 case KVM_CAP_MULTI_ADDRESS_SPACE:
2932 return KVM_ADDRESS_SPACE_NUM;
2934 case KVM_CAP_MAX_VCPU_ID:
2935 return KVM_MAX_VCPU_ID;
2939 return kvm_vm_ioctl_check_extension(kvm, arg);
2942 static long kvm_vm_ioctl(struct file *filp,
2943 unsigned int ioctl, unsigned long arg)
2945 struct kvm *kvm = filp->private_data;
2946 void __user *argp = (void __user *)arg;
2949 if (kvm->mm != current->mm)
2952 case KVM_CREATE_VCPU:
2953 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2955 case KVM_SET_USER_MEMORY_REGION: {
2956 struct kvm_userspace_memory_region kvm_userspace_mem;
2959 if (copy_from_user(&kvm_userspace_mem, argp,
2960 sizeof(kvm_userspace_mem)))
2963 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2966 case KVM_GET_DIRTY_LOG: {
2967 struct kvm_dirty_log log;
2970 if (copy_from_user(&log, argp, sizeof(log)))
2972 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2975 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2976 case KVM_REGISTER_COALESCED_MMIO: {
2977 struct kvm_coalesced_mmio_zone zone;
2980 if (copy_from_user(&zone, argp, sizeof(zone)))
2982 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2985 case KVM_UNREGISTER_COALESCED_MMIO: {
2986 struct kvm_coalesced_mmio_zone zone;
2989 if (copy_from_user(&zone, argp, sizeof(zone)))
2991 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2996 struct kvm_irqfd data;
2999 if (copy_from_user(&data, argp, sizeof(data)))
3001 r = kvm_irqfd(kvm, &data);
3004 case KVM_IOEVENTFD: {
3005 struct kvm_ioeventfd data;
3008 if (copy_from_user(&data, argp, sizeof(data)))
3010 r = kvm_ioeventfd(kvm, &data);
3013 #ifdef CONFIG_HAVE_KVM_MSI
3014 case KVM_SIGNAL_MSI: {
3018 if (copy_from_user(&msi, argp, sizeof(msi)))
3020 r = kvm_send_userspace_msi(kvm, &msi);
3024 #ifdef __KVM_HAVE_IRQ_LINE
3025 case KVM_IRQ_LINE_STATUS:
3026 case KVM_IRQ_LINE: {
3027 struct kvm_irq_level irq_event;
3030 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
3033 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
3034 ioctl == KVM_IRQ_LINE_STATUS);
3039 if (ioctl == KVM_IRQ_LINE_STATUS) {
3040 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
3048 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3049 case KVM_SET_GSI_ROUTING: {
3050 struct kvm_irq_routing routing;
3051 struct kvm_irq_routing __user *urouting;
3052 struct kvm_irq_routing_entry *entries = NULL;
3055 if (copy_from_user(&routing, argp, sizeof(routing)))
3058 if (routing.nr > KVM_MAX_IRQ_ROUTES)
3064 entries = vmalloc(routing.nr * sizeof(*entries));
3069 if (copy_from_user(entries, urouting->entries,
3070 routing.nr * sizeof(*entries)))
3071 goto out_free_irq_routing;
3073 r = kvm_set_irq_routing(kvm, entries, routing.nr,
3075 out_free_irq_routing:
3079 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
3080 case KVM_CREATE_DEVICE: {
3081 struct kvm_create_device cd;
3084 if (copy_from_user(&cd, argp, sizeof(cd)))
3087 r = kvm_ioctl_create_device(kvm, &cd);
3092 if (copy_to_user(argp, &cd, sizeof(cd)))
3098 case KVM_CHECK_EXTENSION:
3099 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
3102 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
3108 #ifdef CONFIG_KVM_COMPAT
3109 struct compat_kvm_dirty_log {
3113 compat_uptr_t dirty_bitmap; /* one bit per page */
3118 static long kvm_vm_compat_ioctl(struct file *filp,
3119 unsigned int ioctl, unsigned long arg)
3121 struct kvm *kvm = filp->private_data;
3124 if (kvm->mm != current->mm)
3127 case KVM_GET_DIRTY_LOG: {
3128 struct compat_kvm_dirty_log compat_log;
3129 struct kvm_dirty_log log;
3131 if (copy_from_user(&compat_log, (void __user *)arg,
3132 sizeof(compat_log)))
3134 log.slot = compat_log.slot;
3135 log.padding1 = compat_log.padding1;
3136 log.padding2 = compat_log.padding2;
3137 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
3139 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3143 r = kvm_vm_ioctl(filp, ioctl, arg);
3149 static struct file_operations kvm_vm_fops = {
3150 .release = kvm_vm_release,
3151 .unlocked_ioctl = kvm_vm_ioctl,
3152 #ifdef CONFIG_KVM_COMPAT
3153 .compat_ioctl = kvm_vm_compat_ioctl,
3155 .llseek = noop_llseek,
3158 static int kvm_dev_ioctl_create_vm(unsigned long type)
3164 kvm = kvm_create_vm(type);
3166 return PTR_ERR(kvm);
3167 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
3168 r = kvm_coalesced_mmio_init(kvm);
3174 r = get_unused_fd_flags(O_CLOEXEC);
3179 file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
3183 return PTR_ERR(file);
3186 if (kvm_create_vm_debugfs(kvm, r) < 0) {
3192 fd_install(r, file);
3196 static long kvm_dev_ioctl(struct file *filp,
3197 unsigned int ioctl, unsigned long arg)
3202 case KVM_GET_API_VERSION:
3205 r = KVM_API_VERSION;
3208 r = kvm_dev_ioctl_create_vm(arg);
3210 case KVM_CHECK_EXTENSION:
3211 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
3213 case KVM_GET_VCPU_MMAP_SIZE:
3216 r = PAGE_SIZE; /* struct kvm_run */
3218 r += PAGE_SIZE; /* pio data page */
3220 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
3221 r += PAGE_SIZE; /* coalesced mmio ring page */
3224 case KVM_TRACE_ENABLE:
3225 case KVM_TRACE_PAUSE:
3226 case KVM_TRACE_DISABLE:
3230 return kvm_arch_dev_ioctl(filp, ioctl, arg);
3236 static struct file_operations kvm_chardev_ops = {
3237 .unlocked_ioctl = kvm_dev_ioctl,
3238 .compat_ioctl = kvm_dev_ioctl,
3239 .llseek = noop_llseek,
3242 static struct miscdevice kvm_dev = {
3248 static void hardware_enable_nolock(void *junk)
3250 int cpu = raw_smp_processor_id();
3253 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
3256 cpumask_set_cpu(cpu, cpus_hardware_enabled);
3258 r = kvm_arch_hardware_enable();
3261 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3262 atomic_inc(&hardware_enable_failed);
3263 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
3267 static int kvm_starting_cpu(unsigned int cpu)
3269 raw_spin_lock(&kvm_count_lock);
3270 if (kvm_usage_count)
3271 hardware_enable_nolock(NULL);
3272 raw_spin_unlock(&kvm_count_lock);
3276 static void hardware_disable_nolock(void *junk)
3278 int cpu = raw_smp_processor_id();
3280 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
3282 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3283 kvm_arch_hardware_disable();
3286 static int kvm_dying_cpu(unsigned int cpu)
3288 raw_spin_lock(&kvm_count_lock);
3289 if (kvm_usage_count)
3290 hardware_disable_nolock(NULL);
3291 raw_spin_unlock(&kvm_count_lock);
3295 static void hardware_disable_all_nolock(void)
3297 BUG_ON(!kvm_usage_count);
3300 if (!kvm_usage_count)
3301 on_each_cpu(hardware_disable_nolock, NULL, 1);
3304 static void hardware_disable_all(void)
3306 raw_spin_lock(&kvm_count_lock);
3307 hardware_disable_all_nolock();
3308 raw_spin_unlock(&kvm_count_lock);
3311 static int hardware_enable_all(void)
3315 raw_spin_lock(&kvm_count_lock);
3318 if (kvm_usage_count == 1) {
3319 atomic_set(&hardware_enable_failed, 0);
3320 on_each_cpu(hardware_enable_nolock, NULL, 1);
3322 if (atomic_read(&hardware_enable_failed)) {
3323 hardware_disable_all_nolock();
3328 raw_spin_unlock(&kvm_count_lock);
3333 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3337 * Some (well, at least mine) BIOSes hang on reboot if
3340 * And Intel TXT required VMX off for all cpu when system shutdown.
3342 pr_info("kvm: exiting hardware virtualization\n");
3343 kvm_rebooting = true;
3344 on_each_cpu(hardware_disable_nolock, NULL, 1);
3348 static struct notifier_block kvm_reboot_notifier = {
3349 .notifier_call = kvm_reboot,
3353 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3357 for (i = 0; i < bus->dev_count; i++) {
3358 struct kvm_io_device *pos = bus->range[i].dev;
3360 kvm_iodevice_destructor(pos);
3365 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
3366 const struct kvm_io_range *r2)
3368 gpa_t addr1 = r1->addr;
3369 gpa_t addr2 = r2->addr;
3374 /* If r2->len == 0, match the exact address. If r2->len != 0,
3375 * accept any overlapping write. Any order is acceptable for
3376 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
3377 * we process all of them.
3390 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
3392 return kvm_io_bus_cmp(p1, p2);
3395 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
3396 gpa_t addr, int len)
3398 bus->range[bus->dev_count++] = (struct kvm_io_range) {
3404 sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
3405 kvm_io_bus_sort_cmp, NULL);
3410 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
3411 gpa_t addr, int len)
3413 struct kvm_io_range *range, key;
3416 key = (struct kvm_io_range) {
3421 range = bsearch(&key, bus->range, bus->dev_count,
3422 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
3426 off = range - bus->range;
3428 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
3434 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3435 struct kvm_io_range *range, const void *val)
3439 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3443 while (idx < bus->dev_count &&
3444 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3445 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
3454 /* kvm_io_bus_write - called under kvm->slots_lock */
3455 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3456 int len, const void *val)
3458 struct kvm_io_bus *bus;
3459 struct kvm_io_range range;
3462 range = (struct kvm_io_range) {
3467 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3470 r = __kvm_io_bus_write(vcpu, bus, &range, val);
3471 return r < 0 ? r : 0;
3474 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3475 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
3476 gpa_t addr, int len, const void *val, long cookie)
3478 struct kvm_io_bus *bus;
3479 struct kvm_io_range range;
3481 range = (struct kvm_io_range) {
3486 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3490 /* First try the device referenced by cookie. */
3491 if ((cookie >= 0) && (cookie < bus->dev_count) &&
3492 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
3493 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
3498 * cookie contained garbage; fall back to search and return the
3499 * correct cookie value.
3501 return __kvm_io_bus_write(vcpu, bus, &range, val);
3504 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3505 struct kvm_io_range *range, void *val)
3509 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3513 while (idx < bus->dev_count &&
3514 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3515 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
3523 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
3525 /* kvm_io_bus_read - called under kvm->slots_lock */
3526 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3529 struct kvm_io_bus *bus;
3530 struct kvm_io_range range;
3533 range = (struct kvm_io_range) {
3538 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3541 r = __kvm_io_bus_read(vcpu, bus, &range, val);
3542 return r < 0 ? r : 0;
3546 /* Caller must hold slots_lock. */
3547 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3548 int len, struct kvm_io_device *dev)
3550 struct kvm_io_bus *new_bus, *bus;
3552 bus = kvm->buses[bus_idx];
3556 /* exclude ioeventfd which is limited by maximum fd */
3557 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3560 new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count + 1) *
3561 sizeof(struct kvm_io_range)), GFP_KERNEL);
3564 memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
3565 sizeof(struct kvm_io_range)));
3566 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
3567 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3568 synchronize_srcu_expedited(&kvm->srcu);
3574 /* Caller must hold slots_lock. */
3575 void kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3576 struct kvm_io_device *dev)
3579 struct kvm_io_bus *new_bus, *bus;
3581 bus = kvm->buses[bus_idx];
3585 for (i = 0; i < bus->dev_count; i++)
3586 if (bus->range[i].dev == dev) {
3590 if (i == bus->dev_count)
3593 new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3594 sizeof(struct kvm_io_range)), GFP_KERNEL);
3596 pr_err("kvm: failed to shrink bus, removing it completely\n");
3600 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3601 new_bus->dev_count--;
3602 memcpy(new_bus->range + i, bus->range + i + 1,
3603 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3606 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3607 synchronize_srcu_expedited(&kvm->srcu);
3612 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3615 struct kvm_io_bus *bus;
3616 int dev_idx, srcu_idx;
3617 struct kvm_io_device *iodev = NULL;
3619 srcu_idx = srcu_read_lock(&kvm->srcu);
3621 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
3625 dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
3629 iodev = bus->range[dev_idx].dev;
3632 srcu_read_unlock(&kvm->srcu, srcu_idx);
3636 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
3638 static int kvm_debugfs_open(struct inode *inode, struct file *file,
3639 int (*get)(void *, u64 *), int (*set)(void *, u64),
3642 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
3645 /* The debugfs files are a reference to the kvm struct which
3646 * is still valid when kvm_destroy_vm is called.
3647 * To avoid the race between open and the removal of the debugfs
3648 * directory we test against the users count.
3650 if (!refcount_inc_not_zero(&stat_data->kvm->users_count))
3653 if (simple_attr_open(inode, file, get, set, fmt)) {
3654 kvm_put_kvm(stat_data->kvm);
3661 static int kvm_debugfs_release(struct inode *inode, struct file *file)
3663 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
3666 simple_attr_release(inode, file);
3667 kvm_put_kvm(stat_data->kvm);
3672 static int vm_stat_get_per_vm(void *data, u64 *val)
3674 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3676 *val = *(ulong *)((void *)stat_data->kvm + stat_data->offset);
3681 static int vm_stat_clear_per_vm(void *data, u64 val)
3683 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3688 *(ulong *)((void *)stat_data->kvm + stat_data->offset) = 0;
3693 static int vm_stat_get_per_vm_open(struct inode *inode, struct file *file)
3695 __simple_attr_check_format("%llu\n", 0ull);
3696 return kvm_debugfs_open(inode, file, vm_stat_get_per_vm,
3697 vm_stat_clear_per_vm, "%llu\n");
3700 static const struct file_operations vm_stat_get_per_vm_fops = {
3701 .owner = THIS_MODULE,
3702 .open = vm_stat_get_per_vm_open,
3703 .release = kvm_debugfs_release,
3704 .read = simple_attr_read,
3705 .write = simple_attr_write,
3706 .llseek = generic_file_llseek,
3709 static int vcpu_stat_get_per_vm(void *data, u64 *val)
3712 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3713 struct kvm_vcpu *vcpu;
3717 kvm_for_each_vcpu(i, vcpu, stat_data->kvm)
3718 *val += *(u64 *)((void *)vcpu + stat_data->offset);
3723 static int vcpu_stat_clear_per_vm(void *data, u64 val)
3726 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3727 struct kvm_vcpu *vcpu;
3732 kvm_for_each_vcpu(i, vcpu, stat_data->kvm)
3733 *(u64 *)((void *)vcpu + stat_data->offset) = 0;
3738 static int vcpu_stat_get_per_vm_open(struct inode *inode, struct file *file)
3740 __simple_attr_check_format("%llu\n", 0ull);
3741 return kvm_debugfs_open(inode, file, vcpu_stat_get_per_vm,
3742 vcpu_stat_clear_per_vm, "%llu\n");
3745 static const struct file_operations vcpu_stat_get_per_vm_fops = {
3746 .owner = THIS_MODULE,
3747 .open = vcpu_stat_get_per_vm_open,
3748 .release = kvm_debugfs_release,
3749 .read = simple_attr_read,
3750 .write = simple_attr_write,
3751 .llseek = generic_file_llseek,
3754 static const struct file_operations *stat_fops_per_vm[] = {
3755 [KVM_STAT_VCPU] = &vcpu_stat_get_per_vm_fops,
3756 [KVM_STAT_VM] = &vm_stat_get_per_vm_fops,
3759 static int vm_stat_get(void *_offset, u64 *val)
3761 unsigned offset = (long)_offset;
3763 struct kvm_stat_data stat_tmp = {.offset = offset};
3767 spin_lock(&kvm_lock);
3768 list_for_each_entry(kvm, &vm_list, vm_list) {
3770 vm_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
3773 spin_unlock(&kvm_lock);
3777 static int vm_stat_clear(void *_offset, u64 val)
3779 unsigned offset = (long)_offset;
3781 struct kvm_stat_data stat_tmp = {.offset = offset};
3786 spin_lock(&kvm_lock);
3787 list_for_each_entry(kvm, &vm_list, vm_list) {
3789 vm_stat_clear_per_vm((void *)&stat_tmp, 0);
3791 spin_unlock(&kvm_lock);
3796 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n");
3798 static int vcpu_stat_get(void *_offset, u64 *val)
3800 unsigned offset = (long)_offset;
3802 struct kvm_stat_data stat_tmp = {.offset = offset};
3806 spin_lock(&kvm_lock);
3807 list_for_each_entry(kvm, &vm_list, vm_list) {
3809 vcpu_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
3812 spin_unlock(&kvm_lock);
3816 static int vcpu_stat_clear(void *_offset, u64 val)
3818 unsigned offset = (long)_offset;
3820 struct kvm_stat_data stat_tmp = {.offset = offset};
3825 spin_lock(&kvm_lock);
3826 list_for_each_entry(kvm, &vm_list, vm_list) {
3828 vcpu_stat_clear_per_vm((void *)&stat_tmp, 0);
3830 spin_unlock(&kvm_lock);
3835 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear,
3838 static const struct file_operations *stat_fops[] = {
3839 [KVM_STAT_VCPU] = &vcpu_stat_fops,
3840 [KVM_STAT_VM] = &vm_stat_fops,
3843 static int kvm_init_debug(void)
3846 struct kvm_stats_debugfs_item *p;
3848 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3849 if (kvm_debugfs_dir == NULL)
3852 kvm_debugfs_num_entries = 0;
3853 for (p = debugfs_entries; p->name; ++p, kvm_debugfs_num_entries++) {
3854 if (!debugfs_create_file(p->name, 0644, kvm_debugfs_dir,
3855 (void *)(long)p->offset,
3856 stat_fops[p->kind]))
3863 debugfs_remove_recursive(kvm_debugfs_dir);
3868 static int kvm_suspend(void)
3870 if (kvm_usage_count)
3871 hardware_disable_nolock(NULL);
3875 static void kvm_resume(void)
3877 if (kvm_usage_count) {
3878 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3879 hardware_enable_nolock(NULL);
3883 static struct syscore_ops kvm_syscore_ops = {
3884 .suspend = kvm_suspend,
3885 .resume = kvm_resume,
3889 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3891 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3894 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3896 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3898 if (vcpu->preempted)
3899 vcpu->preempted = false;
3901 kvm_arch_sched_in(vcpu, cpu);
3903 kvm_arch_vcpu_load(vcpu, cpu);
3906 static void kvm_sched_out(struct preempt_notifier *pn,
3907 struct task_struct *next)
3909 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3911 if (current->state == TASK_RUNNING)
3912 vcpu->preempted = true;
3913 kvm_arch_vcpu_put(vcpu);
3916 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3917 struct module *module)
3922 r = kvm_arch_init(opaque);
3927 * kvm_arch_init makes sure there's at most one caller
3928 * for architectures that support multiple implementations,
3929 * like intel and amd on x86.
3930 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3931 * conflicts in case kvm is already setup for another implementation.
3933 r = kvm_irqfd_init();
3937 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
3942 r = kvm_arch_hardware_setup();
3946 for_each_online_cpu(cpu) {
3947 smp_call_function_single(cpu,
3948 kvm_arch_check_processor_compat,
3954 r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting",
3955 kvm_starting_cpu, kvm_dying_cpu);
3958 register_reboot_notifier(&kvm_reboot_notifier);
3960 /* A kmem cache lets us meet the alignment requirements of fx_save. */
3962 vcpu_align = __alignof__(struct kvm_vcpu);
3963 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
3965 if (!kvm_vcpu_cache) {
3970 r = kvm_async_pf_init();
3974 kvm_chardev_ops.owner = module;
3975 kvm_vm_fops.owner = module;
3976 kvm_vcpu_fops.owner = module;
3978 r = misc_register(&kvm_dev);
3980 pr_err("kvm: misc device register failed\n");
3984 register_syscore_ops(&kvm_syscore_ops);
3986 kvm_preempt_ops.sched_in = kvm_sched_in;
3987 kvm_preempt_ops.sched_out = kvm_sched_out;
3989 r = kvm_init_debug();
3991 pr_err("kvm: create debugfs files failed\n");
3995 r = kvm_vfio_ops_init();
4001 unregister_syscore_ops(&kvm_syscore_ops);
4002 misc_deregister(&kvm_dev);
4004 kvm_async_pf_deinit();
4006 kmem_cache_destroy(kvm_vcpu_cache);
4008 unregister_reboot_notifier(&kvm_reboot_notifier);
4009 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4012 kvm_arch_hardware_unsetup();
4014 free_cpumask_var(cpus_hardware_enabled);
4022 EXPORT_SYMBOL_GPL(kvm_init);
4026 debugfs_remove_recursive(kvm_debugfs_dir);
4027 misc_deregister(&kvm_dev);
4028 kmem_cache_destroy(kvm_vcpu_cache);
4029 kvm_async_pf_deinit();
4030 unregister_syscore_ops(&kvm_syscore_ops);
4031 unregister_reboot_notifier(&kvm_reboot_notifier);
4032 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4033 on_each_cpu(hardware_disable_nolock, NULL, 1);
4034 kvm_arch_hardware_unsetup();
4037 free_cpumask_var(cpus_hardware_enabled);
4038 kvm_vfio_ops_exit();
4040 EXPORT_SYMBOL_GPL(kvm_exit);