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.h>
36 #include <linux/cpumask.h>
37 #include <linux/smp.h>
38 #include <linux/anon_inodes.h>
39 #include <linux/profile.h>
40 #include <linux/kvm_para.h>
41 #include <linux/pagemap.h>
42 #include <linux/mman.h>
43 #include <linux/swap.h>
44 #include <linux/bitops.h>
45 #include <linux/spinlock.h>
46 #include <linux/compat.h>
47 #include <linux/srcu.h>
48 #include <linux/hugetlb.h>
49 #include <linux/slab.h>
50 #include <linux/sort.h>
51 #include <linux/bsearch.h>
53 #include <asm/processor.h>
55 #include <asm/ioctl.h>
56 #include <linux/uaccess.h>
57 #include <asm/pgtable.h>
59 #include "coalesced_mmio.h"
63 #define CREATE_TRACE_POINTS
64 #include <trace/events/kvm.h>
66 /* Worst case buffer size needed for holding an integer. */
67 #define ITOA_MAX_LEN 12
69 MODULE_AUTHOR("Qumranet");
70 MODULE_LICENSE("GPL");
72 /* Architectures should define their poll value according to the halt latency */
73 unsigned int halt_poll_ns = KVM_HALT_POLL_NS_DEFAULT;
74 module_param(halt_poll_ns, uint, S_IRUGO | S_IWUSR);
75 EXPORT_SYMBOL_GPL(halt_poll_ns);
77 /* Default doubles per-vcpu halt_poll_ns. */
78 unsigned int halt_poll_ns_grow = 2;
79 module_param(halt_poll_ns_grow, uint, S_IRUGO | S_IWUSR);
80 EXPORT_SYMBOL_GPL(halt_poll_ns_grow);
82 /* Default resets per-vcpu halt_poll_ns . */
83 unsigned int halt_poll_ns_shrink;
84 module_param(halt_poll_ns_shrink, uint, S_IRUGO | S_IWUSR);
85 EXPORT_SYMBOL_GPL(halt_poll_ns_shrink);
90 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
93 DEFINE_SPINLOCK(kvm_lock);
94 static DEFINE_RAW_SPINLOCK(kvm_count_lock);
97 static cpumask_var_t cpus_hardware_enabled;
98 static int kvm_usage_count;
99 static atomic_t hardware_enable_failed;
101 struct kmem_cache *kvm_vcpu_cache;
102 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
104 static __read_mostly struct preempt_ops kvm_preempt_ops;
106 struct dentry *kvm_debugfs_dir;
107 EXPORT_SYMBOL_GPL(kvm_debugfs_dir);
109 static int kvm_debugfs_num_entries;
110 static const struct file_operations *stat_fops_per_vm[];
112 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
114 #ifdef CONFIG_KVM_COMPAT
115 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
118 static int hardware_enable_all(void);
119 static void hardware_disable_all(void);
121 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
123 static void kvm_release_pfn_dirty(kvm_pfn_t pfn);
124 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot, gfn_t gfn);
126 __visible bool kvm_rebooting;
127 EXPORT_SYMBOL_GPL(kvm_rebooting);
129 static bool largepages_enabled = true;
131 bool kvm_is_reserved_pfn(kvm_pfn_t pfn)
134 return PageReserved(pfn_to_page(pfn));
140 * Switches to specified vcpu, until a matching vcpu_put()
142 int vcpu_load(struct kvm_vcpu *vcpu)
146 if (mutex_lock_killable(&vcpu->mutex))
149 preempt_notifier_register(&vcpu->preempt_notifier);
150 kvm_arch_vcpu_load(vcpu, cpu);
154 EXPORT_SYMBOL_GPL(vcpu_load);
156 void vcpu_put(struct kvm_vcpu *vcpu)
159 kvm_arch_vcpu_put(vcpu);
160 preempt_notifier_unregister(&vcpu->preempt_notifier);
162 mutex_unlock(&vcpu->mutex);
164 EXPORT_SYMBOL_GPL(vcpu_put);
166 static void ack_flush(void *_completed)
170 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
175 struct kvm_vcpu *vcpu;
177 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
180 kvm_for_each_vcpu(i, vcpu, kvm) {
181 kvm_make_request(req, vcpu);
184 /* Set ->requests bit before we read ->mode. */
185 smp_mb__after_atomic();
187 if (cpus != NULL && cpu != -1 && cpu != me &&
188 kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
189 cpumask_set_cpu(cpu, cpus);
191 if (unlikely(cpus == NULL))
192 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
193 else if (!cpumask_empty(cpus))
194 smp_call_function_many(cpus, ack_flush, NULL, 1);
198 free_cpumask_var(cpus);
202 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
203 void kvm_flush_remote_tlbs(struct kvm *kvm)
206 * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in
207 * kvm_make_all_cpus_request.
209 long dirty_count = smp_load_acquire(&kvm->tlbs_dirty);
212 * We want to publish modifications to the page tables before reading
213 * mode. Pairs with a memory barrier in arch-specific code.
214 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
215 * and smp_mb in walk_shadow_page_lockless_begin/end.
216 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
218 * There is already an smp_mb__after_atomic() before
219 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
222 if (kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
223 ++kvm->stat.remote_tlb_flush;
224 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
226 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
229 void kvm_reload_remote_mmus(struct kvm *kvm)
231 kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
234 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
239 mutex_init(&vcpu->mutex);
244 init_swait_queue_head(&vcpu->wq);
245 kvm_async_pf_vcpu_init(vcpu);
248 INIT_LIST_HEAD(&vcpu->blocked_vcpu_list);
250 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
255 vcpu->run = page_address(page);
257 kvm_vcpu_set_in_spin_loop(vcpu, false);
258 kvm_vcpu_set_dy_eligible(vcpu, false);
259 vcpu->preempted = false;
261 r = kvm_arch_vcpu_init(vcpu);
267 free_page((unsigned long)vcpu->run);
271 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
273 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
276 kvm_arch_vcpu_uninit(vcpu);
277 free_page((unsigned long)vcpu->run);
279 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
281 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
282 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
284 return container_of(mn, struct kvm, mmu_notifier);
287 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
288 struct mm_struct *mm,
289 unsigned long address)
291 struct kvm *kvm = mmu_notifier_to_kvm(mn);
292 int need_tlb_flush, idx;
295 * When ->invalidate_page runs, the linux pte has been zapped
296 * already but the page is still allocated until
297 * ->invalidate_page returns. So if we increase the sequence
298 * here the kvm page fault will notice if the spte can't be
299 * established because the page is going to be freed. If
300 * instead the kvm page fault establishes the spte before
301 * ->invalidate_page runs, kvm_unmap_hva will release it
304 * The sequence increase only need to be seen at spin_unlock
305 * time, and not at spin_lock time.
307 * Increasing the sequence after the spin_unlock would be
308 * unsafe because the kvm page fault could then establish the
309 * pte after kvm_unmap_hva returned, without noticing the page
310 * is going to be freed.
312 idx = srcu_read_lock(&kvm->srcu);
313 spin_lock(&kvm->mmu_lock);
315 kvm->mmu_notifier_seq++;
316 need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
317 /* we've to flush the tlb before the pages can be freed */
319 kvm_flush_remote_tlbs(kvm);
321 spin_unlock(&kvm->mmu_lock);
323 kvm_arch_mmu_notifier_invalidate_page(kvm, address);
325 srcu_read_unlock(&kvm->srcu, idx);
328 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
329 struct mm_struct *mm,
330 unsigned long address,
333 struct kvm *kvm = mmu_notifier_to_kvm(mn);
336 idx = srcu_read_lock(&kvm->srcu);
337 spin_lock(&kvm->mmu_lock);
338 kvm->mmu_notifier_seq++;
339 kvm_set_spte_hva(kvm, address, pte);
340 spin_unlock(&kvm->mmu_lock);
341 srcu_read_unlock(&kvm->srcu, idx);
344 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
345 struct mm_struct *mm,
349 struct kvm *kvm = mmu_notifier_to_kvm(mn);
350 int need_tlb_flush = 0, idx;
352 idx = srcu_read_lock(&kvm->srcu);
353 spin_lock(&kvm->mmu_lock);
355 * The count increase must become visible at unlock time as no
356 * spte can be established without taking the mmu_lock and
357 * count is also read inside the mmu_lock critical section.
359 kvm->mmu_notifier_count++;
360 need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
361 need_tlb_flush |= kvm->tlbs_dirty;
362 /* we've to flush the tlb before the pages can be freed */
364 kvm_flush_remote_tlbs(kvm);
366 spin_unlock(&kvm->mmu_lock);
367 srcu_read_unlock(&kvm->srcu, idx);
370 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
371 struct mm_struct *mm,
375 struct kvm *kvm = mmu_notifier_to_kvm(mn);
377 spin_lock(&kvm->mmu_lock);
379 * This sequence increase will notify the kvm page fault that
380 * the page that is going to be mapped in the spte could have
383 kvm->mmu_notifier_seq++;
386 * The above sequence increase must be visible before the
387 * below count decrease, which is ensured by the smp_wmb above
388 * in conjunction with the smp_rmb in mmu_notifier_retry().
390 kvm->mmu_notifier_count--;
391 spin_unlock(&kvm->mmu_lock);
393 BUG_ON(kvm->mmu_notifier_count < 0);
396 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
397 struct mm_struct *mm,
401 struct kvm *kvm = mmu_notifier_to_kvm(mn);
404 idx = srcu_read_lock(&kvm->srcu);
405 spin_lock(&kvm->mmu_lock);
407 young = kvm_age_hva(kvm, start, end);
409 kvm_flush_remote_tlbs(kvm);
411 spin_unlock(&kvm->mmu_lock);
412 srcu_read_unlock(&kvm->srcu, idx);
417 static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
418 struct mm_struct *mm,
422 struct kvm *kvm = mmu_notifier_to_kvm(mn);
425 idx = srcu_read_lock(&kvm->srcu);
426 spin_lock(&kvm->mmu_lock);
428 * Even though we do not flush TLB, this will still adversely
429 * affect performance on pre-Haswell Intel EPT, where there is
430 * no EPT Access Bit to clear so that we have to tear down EPT
431 * tables instead. If we find this unacceptable, we can always
432 * add a parameter to kvm_age_hva so that it effectively doesn't
433 * do anything on clear_young.
435 * Also note that currently we never issue secondary TLB flushes
436 * from clear_young, leaving this job up to the regular system
437 * cadence. If we find this inaccurate, we might come up with a
438 * more sophisticated heuristic later.
440 young = kvm_age_hva(kvm, start, end);
441 spin_unlock(&kvm->mmu_lock);
442 srcu_read_unlock(&kvm->srcu, idx);
447 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
448 struct mm_struct *mm,
449 unsigned long address)
451 struct kvm *kvm = mmu_notifier_to_kvm(mn);
454 idx = srcu_read_lock(&kvm->srcu);
455 spin_lock(&kvm->mmu_lock);
456 young = kvm_test_age_hva(kvm, address);
457 spin_unlock(&kvm->mmu_lock);
458 srcu_read_unlock(&kvm->srcu, idx);
463 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
464 struct mm_struct *mm)
466 struct kvm *kvm = mmu_notifier_to_kvm(mn);
469 idx = srcu_read_lock(&kvm->srcu);
470 kvm_arch_flush_shadow_all(kvm);
471 srcu_read_unlock(&kvm->srcu, idx);
474 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
475 .invalidate_page = kvm_mmu_notifier_invalidate_page,
476 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
477 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
478 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
479 .clear_young = kvm_mmu_notifier_clear_young,
480 .test_young = kvm_mmu_notifier_test_young,
481 .change_pte = kvm_mmu_notifier_change_pte,
482 .release = kvm_mmu_notifier_release,
485 static int kvm_init_mmu_notifier(struct kvm *kvm)
487 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
488 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
491 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
493 static int kvm_init_mmu_notifier(struct kvm *kvm)
498 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
500 static struct kvm_memslots *kvm_alloc_memslots(void)
503 struct kvm_memslots *slots;
505 slots = kvm_kvzalloc(sizeof(struct kvm_memslots));
510 * Init kvm generation close to the maximum to easily test the
511 * code of handling generation number wrap-around.
513 slots->generation = -150;
514 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
515 slots->id_to_index[i] = slots->memslots[i].id = i;
520 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
522 if (!memslot->dirty_bitmap)
525 kvfree(memslot->dirty_bitmap);
526 memslot->dirty_bitmap = NULL;
530 * Free any memory in @free but not in @dont.
532 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
533 struct kvm_memory_slot *dont)
535 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
536 kvm_destroy_dirty_bitmap(free);
538 kvm_arch_free_memslot(kvm, free, dont);
543 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
545 struct kvm_memory_slot *memslot;
550 kvm_for_each_memslot(memslot, slots)
551 kvm_free_memslot(kvm, memslot, NULL);
556 static void kvm_destroy_vm_debugfs(struct kvm *kvm)
560 if (!kvm->debugfs_dentry)
563 debugfs_remove_recursive(kvm->debugfs_dentry);
565 if (kvm->debugfs_stat_data) {
566 for (i = 0; i < kvm_debugfs_num_entries; i++)
567 kfree(kvm->debugfs_stat_data[i]);
568 kfree(kvm->debugfs_stat_data);
572 static int kvm_create_vm_debugfs(struct kvm *kvm, int fd)
574 char dir_name[ITOA_MAX_LEN * 2];
575 struct kvm_stat_data *stat_data;
576 struct kvm_stats_debugfs_item *p;
578 if (!debugfs_initialized())
581 snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd);
582 kvm->debugfs_dentry = debugfs_create_dir(dir_name,
584 if (!kvm->debugfs_dentry)
587 kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries,
588 sizeof(*kvm->debugfs_stat_data),
590 if (!kvm->debugfs_stat_data)
593 for (p = debugfs_entries; p->name; p++) {
594 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL);
598 stat_data->kvm = kvm;
599 stat_data->offset = p->offset;
600 kvm->debugfs_stat_data[p - debugfs_entries] = stat_data;
601 if (!debugfs_create_file(p->name, 0644,
604 stat_fops_per_vm[p->kind]))
610 static struct kvm *kvm_create_vm(unsigned long type)
613 struct kvm *kvm = kvm_arch_alloc_vm();
616 return ERR_PTR(-ENOMEM);
618 spin_lock_init(&kvm->mmu_lock);
619 atomic_inc(¤t->mm->mm_count);
620 kvm->mm = current->mm;
621 kvm_eventfd_init(kvm);
622 mutex_init(&kvm->lock);
623 mutex_init(&kvm->irq_lock);
624 mutex_init(&kvm->slots_lock);
625 atomic_set(&kvm->users_count, 1);
626 INIT_LIST_HEAD(&kvm->devices);
628 r = kvm_arch_init_vm(kvm, type);
630 goto out_err_no_disable;
632 r = hardware_enable_all();
634 goto out_err_no_disable;
636 #ifdef CONFIG_HAVE_KVM_IRQFD
637 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
640 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
643 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
644 kvm->memslots[i] = kvm_alloc_memslots();
645 if (!kvm->memslots[i])
646 goto out_err_no_srcu;
649 if (init_srcu_struct(&kvm->srcu))
650 goto out_err_no_srcu;
651 if (init_srcu_struct(&kvm->irq_srcu))
652 goto out_err_no_irq_srcu;
653 for (i = 0; i < KVM_NR_BUSES; i++) {
654 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
660 r = kvm_init_mmu_notifier(kvm);
664 spin_lock(&kvm_lock);
665 list_add(&kvm->vm_list, &vm_list);
666 spin_unlock(&kvm_lock);
668 preempt_notifier_inc();
673 cleanup_srcu_struct(&kvm->irq_srcu);
675 cleanup_srcu_struct(&kvm->srcu);
677 hardware_disable_all();
679 for (i = 0; i < KVM_NR_BUSES; i++)
680 kfree(kvm->buses[i]);
681 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
682 kvm_free_memslots(kvm, kvm->memslots[i]);
683 kvm_arch_free_vm(kvm);
689 * Avoid using vmalloc for a small buffer.
690 * Should not be used when the size is statically known.
692 void *kvm_kvzalloc(unsigned long size)
694 if (size > PAGE_SIZE)
695 return vzalloc(size);
697 return kzalloc(size, GFP_KERNEL);
700 static void kvm_destroy_devices(struct kvm *kvm)
702 struct kvm_device *dev, *tmp;
705 * We do not need to take the kvm->lock here, because nobody else
706 * has a reference to the struct kvm at this point and therefore
707 * cannot access the devices list anyhow.
709 list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
710 list_del(&dev->vm_node);
711 dev->ops->destroy(dev);
715 static void kvm_destroy_vm(struct kvm *kvm)
718 struct mm_struct *mm = kvm->mm;
720 kvm_destroy_vm_debugfs(kvm);
721 kvm_arch_sync_events(kvm);
722 spin_lock(&kvm_lock);
723 list_del(&kvm->vm_list);
724 spin_unlock(&kvm_lock);
725 kvm_free_irq_routing(kvm);
726 for (i = 0; i < KVM_NR_BUSES; i++)
727 kvm_io_bus_destroy(kvm->buses[i]);
728 kvm_coalesced_mmio_free(kvm);
729 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
730 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
732 kvm_arch_flush_shadow_all(kvm);
734 kvm_arch_destroy_vm(kvm);
735 kvm_destroy_devices(kvm);
736 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
737 kvm_free_memslots(kvm, kvm->memslots[i]);
738 cleanup_srcu_struct(&kvm->irq_srcu);
739 cleanup_srcu_struct(&kvm->srcu);
740 kvm_arch_free_vm(kvm);
741 preempt_notifier_dec();
742 hardware_disable_all();
746 void kvm_get_kvm(struct kvm *kvm)
748 atomic_inc(&kvm->users_count);
750 EXPORT_SYMBOL_GPL(kvm_get_kvm);
752 void kvm_put_kvm(struct kvm *kvm)
754 if (atomic_dec_and_test(&kvm->users_count))
757 EXPORT_SYMBOL_GPL(kvm_put_kvm);
760 static int kvm_vm_release(struct inode *inode, struct file *filp)
762 struct kvm *kvm = filp->private_data;
764 kvm_irqfd_release(kvm);
771 * Allocation size is twice as large as the actual dirty bitmap size.
772 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
774 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
776 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
778 memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
779 if (!memslot->dirty_bitmap)
786 * Insert memslot and re-sort memslots based on their GFN,
787 * so binary search could be used to lookup GFN.
788 * Sorting algorithm takes advantage of having initially
789 * sorted array and known changed memslot position.
791 static void update_memslots(struct kvm_memslots *slots,
792 struct kvm_memory_slot *new)
795 int i = slots->id_to_index[id];
796 struct kvm_memory_slot *mslots = slots->memslots;
798 WARN_ON(mslots[i].id != id);
800 WARN_ON(!mslots[i].npages);
801 if (mslots[i].npages)
804 if (!mslots[i].npages)
808 while (i < KVM_MEM_SLOTS_NUM - 1 &&
809 new->base_gfn <= mslots[i + 1].base_gfn) {
810 if (!mslots[i + 1].npages)
812 mslots[i] = mslots[i + 1];
813 slots->id_to_index[mslots[i].id] = i;
818 * The ">=" is needed when creating a slot with base_gfn == 0,
819 * so that it moves before all those with base_gfn == npages == 0.
821 * On the other hand, if new->npages is zero, the above loop has
822 * already left i pointing to the beginning of the empty part of
823 * mslots, and the ">=" would move the hole backwards in this
824 * case---which is wrong. So skip the loop when deleting a slot.
828 new->base_gfn >= mslots[i - 1].base_gfn) {
829 mslots[i] = mslots[i - 1];
830 slots->id_to_index[mslots[i].id] = i;
834 WARN_ON_ONCE(i != slots->used_slots);
837 slots->id_to_index[mslots[i].id] = i;
840 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
842 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
844 #ifdef __KVM_HAVE_READONLY_MEM
845 valid_flags |= KVM_MEM_READONLY;
848 if (mem->flags & ~valid_flags)
854 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
855 int as_id, struct kvm_memslots *slots)
857 struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id);
860 * Set the low bit in the generation, which disables SPTE caching
861 * until the end of synchronize_srcu_expedited.
863 WARN_ON(old_memslots->generation & 1);
864 slots->generation = old_memslots->generation + 1;
866 rcu_assign_pointer(kvm->memslots[as_id], slots);
867 synchronize_srcu_expedited(&kvm->srcu);
870 * Increment the new memslot generation a second time. This prevents
871 * vm exits that race with memslot updates from caching a memslot
872 * generation that will (potentially) be valid forever.
876 kvm_arch_memslots_updated(kvm, slots);
882 * Allocate some memory and give it an address in the guest physical address
885 * Discontiguous memory is allowed, mostly for framebuffers.
887 * Must be called holding kvm->slots_lock for write.
889 int __kvm_set_memory_region(struct kvm *kvm,
890 const struct kvm_userspace_memory_region *mem)
894 unsigned long npages;
895 struct kvm_memory_slot *slot;
896 struct kvm_memory_slot old, new;
897 struct kvm_memslots *slots = NULL, *old_memslots;
899 enum kvm_mr_change change;
901 r = check_memory_region_flags(mem);
906 as_id = mem->slot >> 16;
909 /* General sanity checks */
910 if (mem->memory_size & (PAGE_SIZE - 1))
912 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
914 /* We can read the guest memory with __xxx_user() later on. */
915 if ((id < KVM_USER_MEM_SLOTS) &&
916 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
917 !access_ok(VERIFY_WRITE,
918 (void __user *)(unsigned long)mem->userspace_addr,
921 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
923 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
926 slot = id_to_memslot(__kvm_memslots(kvm, as_id), id);
927 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
928 npages = mem->memory_size >> PAGE_SHIFT;
930 if (npages > KVM_MEM_MAX_NR_PAGES)
936 new.base_gfn = base_gfn;
938 new.flags = mem->flags;
942 change = KVM_MR_CREATE;
943 else { /* Modify an existing slot. */
944 if ((mem->userspace_addr != old.userspace_addr) ||
945 (npages != old.npages) ||
946 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
949 if (base_gfn != old.base_gfn)
950 change = KVM_MR_MOVE;
951 else if (new.flags != old.flags)
952 change = KVM_MR_FLAGS_ONLY;
953 else { /* Nothing to change. */
962 change = KVM_MR_DELETE;
967 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
968 /* Check for overlaps */
970 kvm_for_each_memslot(slot, __kvm_memslots(kvm, as_id)) {
971 if ((slot->id >= KVM_USER_MEM_SLOTS) ||
974 if (!((base_gfn + npages <= slot->base_gfn) ||
975 (base_gfn >= slot->base_gfn + slot->npages)))
980 /* Free page dirty bitmap if unneeded */
981 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
982 new.dirty_bitmap = NULL;
985 if (change == KVM_MR_CREATE) {
986 new.userspace_addr = mem->userspace_addr;
988 if (kvm_arch_create_memslot(kvm, &new, npages))
992 /* Allocate page dirty bitmap if needed */
993 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
994 if (kvm_create_dirty_bitmap(&new) < 0)
998 slots = kvm_kvzalloc(sizeof(struct kvm_memslots));
1001 memcpy(slots, __kvm_memslots(kvm, as_id), sizeof(struct kvm_memslots));
1003 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
1004 slot = id_to_memslot(slots, id);
1005 slot->flags |= KVM_MEMSLOT_INVALID;
1007 old_memslots = install_new_memslots(kvm, as_id, slots);
1009 /* slot was deleted or moved, clear iommu mapping */
1010 kvm_iommu_unmap_pages(kvm, &old);
1011 /* From this point no new shadow pages pointing to a deleted,
1012 * or moved, memslot will be created.
1014 * validation of sp->gfn happens in:
1015 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1016 * - kvm_is_visible_gfn (mmu_check_roots)
1018 kvm_arch_flush_shadow_memslot(kvm, slot);
1021 * We can re-use the old_memslots from above, the only difference
1022 * from the currently installed memslots is the invalid flag. This
1023 * will get overwritten by update_memslots anyway.
1025 slots = old_memslots;
1028 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
1032 /* actual memory is freed via old in kvm_free_memslot below */
1033 if (change == KVM_MR_DELETE) {
1034 new.dirty_bitmap = NULL;
1035 memset(&new.arch, 0, sizeof(new.arch));
1038 update_memslots(slots, &new);
1039 old_memslots = install_new_memslots(kvm, as_id, slots);
1041 kvm_arch_commit_memory_region(kvm, mem, &old, &new, change);
1043 kvm_free_memslot(kvm, &old, &new);
1044 kvfree(old_memslots);
1047 * IOMMU mapping: New slots need to be mapped. Old slots need to be
1048 * un-mapped and re-mapped if their base changes. Since base change
1049 * unmapping is handled above with slot deletion, mapping alone is
1050 * needed here. Anything else the iommu might care about for existing
1051 * slots (size changes, userspace addr changes and read-only flag
1052 * changes) is disallowed above, so any other attribute changes getting
1053 * here can be skipped.
1055 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
1056 r = kvm_iommu_map_pages(kvm, &new);
1065 kvm_free_memslot(kvm, &new, &old);
1069 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1071 int kvm_set_memory_region(struct kvm *kvm,
1072 const struct kvm_userspace_memory_region *mem)
1076 mutex_lock(&kvm->slots_lock);
1077 r = __kvm_set_memory_region(kvm, mem);
1078 mutex_unlock(&kvm->slots_lock);
1081 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1083 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1084 struct kvm_userspace_memory_region *mem)
1086 if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1089 return kvm_set_memory_region(kvm, mem);
1092 int kvm_get_dirty_log(struct kvm *kvm,
1093 struct kvm_dirty_log *log, int *is_dirty)
1095 struct kvm_memslots *slots;
1096 struct kvm_memory_slot *memslot;
1097 int r, i, as_id, id;
1099 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);
1110 if (!memslot->dirty_bitmap)
1113 n = kvm_dirty_bitmap_bytes(memslot);
1115 for (i = 0; !any && i < n/sizeof(long); ++i)
1116 any = memslot->dirty_bitmap[i];
1119 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1129 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1131 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1133 * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages
1134 * are dirty write protect them for next write.
1135 * @kvm: pointer to kvm instance
1136 * @log: slot id and address to which we copy the log
1137 * @is_dirty: flag set if any page is dirty
1139 * We need to keep it in mind that VCPU threads can write to the bitmap
1140 * concurrently. So, to avoid losing track of dirty pages we keep the
1143 * 1. Take a snapshot of the bit and clear it if needed.
1144 * 2. Write protect the corresponding page.
1145 * 3. Copy the snapshot to the userspace.
1146 * 4. Upon return caller flushes TLB's if needed.
1148 * Between 2 and 4, the guest may write to the page using the remaining TLB
1149 * entry. This is not a problem because the page is reported dirty using
1150 * the snapshot taken before and step 4 ensures that writes done after
1151 * exiting to userspace will be logged for the next call.
1154 int kvm_get_dirty_log_protect(struct kvm *kvm,
1155 struct kvm_dirty_log *log, bool *is_dirty)
1157 struct kvm_memslots *slots;
1158 struct kvm_memory_slot *memslot;
1159 int r, i, as_id, id;
1161 unsigned long *dirty_bitmap;
1162 unsigned long *dirty_bitmap_buffer;
1165 as_id = log->slot >> 16;
1166 id = (u16)log->slot;
1167 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1170 slots = __kvm_memslots(kvm, as_id);
1171 memslot = id_to_memslot(slots, id);
1173 dirty_bitmap = memslot->dirty_bitmap;
1178 n = kvm_dirty_bitmap_bytes(memslot);
1180 dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long);
1181 memset(dirty_bitmap_buffer, 0, n);
1183 spin_lock(&kvm->mmu_lock);
1185 for (i = 0; i < n / sizeof(long); i++) {
1189 if (!dirty_bitmap[i])
1194 mask = xchg(&dirty_bitmap[i], 0);
1195 dirty_bitmap_buffer[i] = mask;
1198 offset = i * BITS_PER_LONG;
1199 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1204 spin_unlock(&kvm->mmu_lock);
1207 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 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1941 gpa_t gpa, unsigned long len)
1943 struct kvm_memslots *slots = kvm_memslots(kvm);
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(kvm, 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(kvm, 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;
1975 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1977 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1978 void *data, int offset, unsigned long len)
1980 struct kvm_memslots *slots = kvm_memslots(kvm);
1982 gpa_t gpa = ghc->gpa + offset;
1984 BUG_ON(len + offset > ghc->len);
1986 if (slots->generation != ghc->generation)
1987 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1989 if (unlikely(!ghc->memslot))
1990 return kvm_write_guest(kvm, gpa, data, len);
1992 if (kvm_is_error_hva(ghc->hva))
1995 r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
1998 mark_page_dirty_in_slot(ghc->memslot, gpa >> PAGE_SHIFT);
2002 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached);
2004 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2005 void *data, unsigned long len)
2007 return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len);
2009 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
2011 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2012 void *data, unsigned long len)
2014 struct kvm_memslots *slots = kvm_memslots(kvm);
2017 BUG_ON(len > ghc->len);
2019 if (slots->generation != ghc->generation)
2020 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
2022 if (unlikely(!ghc->memslot))
2023 return kvm_read_guest(kvm, ghc->gpa, data, len);
2025 if (kvm_is_error_hva(ghc->hva))
2028 r = __copy_from_user(data, (void __user *)ghc->hva, len);
2034 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
2036 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
2038 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
2040 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
2042 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
2044 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
2046 gfn_t gfn = gpa >> PAGE_SHIFT;
2048 int offset = offset_in_page(gpa);
2051 while ((seg = next_segment(len, offset)) != 0) {
2052 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
2061 EXPORT_SYMBOL_GPL(kvm_clear_guest);
2063 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot,
2066 if (memslot && memslot->dirty_bitmap) {
2067 unsigned long rel_gfn = gfn - memslot->base_gfn;
2069 set_bit_le(rel_gfn, memslot->dirty_bitmap);
2073 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
2075 struct kvm_memory_slot *memslot;
2077 memslot = gfn_to_memslot(kvm, gfn);
2078 mark_page_dirty_in_slot(memslot, gfn);
2080 EXPORT_SYMBOL_GPL(mark_page_dirty);
2082 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
2084 struct kvm_memory_slot *memslot;
2086 memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2087 mark_page_dirty_in_slot(memslot, gfn);
2089 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
2091 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
2093 unsigned int old, val, grow;
2095 old = val = vcpu->halt_poll_ns;
2096 grow = READ_ONCE(halt_poll_ns_grow);
2098 if (val == 0 && grow)
2103 if (val > halt_poll_ns)
2106 vcpu->halt_poll_ns = val;
2107 trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
2110 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
2112 unsigned int old, val, shrink;
2114 old = val = vcpu->halt_poll_ns;
2115 shrink = READ_ONCE(halt_poll_ns_shrink);
2121 vcpu->halt_poll_ns = val;
2122 trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
2125 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
2127 if (kvm_arch_vcpu_runnable(vcpu)) {
2128 kvm_make_request(KVM_REQ_UNHALT, vcpu);
2131 if (kvm_cpu_has_pending_timer(vcpu))
2133 if (signal_pending(current))
2140 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2142 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
2145 DECLARE_SWAITQUEUE(wait);
2146 bool waited = false;
2149 start = cur = ktime_get();
2150 if (vcpu->halt_poll_ns) {
2151 ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns);
2153 ++vcpu->stat.halt_attempted_poll;
2156 * This sets KVM_REQ_UNHALT if an interrupt
2159 if (kvm_vcpu_check_block(vcpu) < 0) {
2160 ++vcpu->stat.halt_successful_poll;
2161 if (!vcpu_valid_wakeup(vcpu))
2162 ++vcpu->stat.halt_poll_invalid;
2166 } while (single_task_running() && ktime_before(cur, stop));
2169 kvm_arch_vcpu_blocking(vcpu);
2172 prepare_to_swait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
2174 if (kvm_vcpu_check_block(vcpu) < 0)
2181 finish_swait(&vcpu->wq, &wait);
2184 kvm_arch_vcpu_unblocking(vcpu);
2186 block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
2188 if (!vcpu_valid_wakeup(vcpu))
2189 shrink_halt_poll_ns(vcpu);
2190 else if (halt_poll_ns) {
2191 if (block_ns <= vcpu->halt_poll_ns)
2193 /* we had a long block, shrink polling */
2194 else if (vcpu->halt_poll_ns && block_ns > halt_poll_ns)
2195 shrink_halt_poll_ns(vcpu);
2196 /* we had a short halt and our poll time is too small */
2197 else if (vcpu->halt_poll_ns < halt_poll_ns &&
2198 block_ns < halt_poll_ns)
2199 grow_halt_poll_ns(vcpu);
2201 vcpu->halt_poll_ns = 0;
2203 trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu));
2204 kvm_arch_vcpu_block_finish(vcpu);
2206 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
2209 void kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
2211 struct swait_queue_head *wqp;
2213 wqp = kvm_arch_vcpu_wq(vcpu);
2214 if (swait_active(wqp)) {
2216 ++vcpu->stat.halt_wakeup;
2220 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
2223 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2225 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
2228 int cpu = vcpu->cpu;
2230 kvm_vcpu_wake_up(vcpu);
2232 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
2233 if (kvm_arch_vcpu_should_kick(vcpu))
2234 smp_send_reschedule(cpu);
2237 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
2238 #endif /* !CONFIG_S390 */
2240 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
2243 struct task_struct *task = NULL;
2247 pid = rcu_dereference(target->pid);
2249 task = get_pid_task(pid, PIDTYPE_PID);
2253 ret = yield_to(task, 1);
2254 put_task_struct(task);
2258 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
2261 * Helper that checks whether a VCPU is eligible for directed yield.
2262 * Most eligible candidate to yield is decided by following heuristics:
2264 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2265 * (preempted lock holder), indicated by @in_spin_loop.
2266 * Set at the beiginning and cleared at the end of interception/PLE handler.
2268 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2269 * chance last time (mostly it has become eligible now since we have probably
2270 * yielded to lockholder in last iteration. This is done by toggling
2271 * @dy_eligible each time a VCPU checked for eligibility.)
2273 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2274 * to preempted lock-holder could result in wrong VCPU selection and CPU
2275 * burning. Giving priority for a potential lock-holder increases lock
2278 * Since algorithm is based on heuristics, accessing another VCPU data without
2279 * locking does not harm. It may result in trying to yield to same VCPU, fail
2280 * and continue with next VCPU and so on.
2282 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
2284 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2287 eligible = !vcpu->spin_loop.in_spin_loop ||
2288 vcpu->spin_loop.dy_eligible;
2290 if (vcpu->spin_loop.in_spin_loop)
2291 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
2299 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
2301 struct kvm *kvm = me->kvm;
2302 struct kvm_vcpu *vcpu;
2303 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
2309 kvm_vcpu_set_in_spin_loop(me, true);
2311 * We boost the priority of a VCPU that is runnable but not
2312 * currently running, because it got preempted by something
2313 * else and called schedule in __vcpu_run. Hopefully that
2314 * VCPU is holding the lock that we need and will release it.
2315 * We approximate round-robin by starting at the last boosted VCPU.
2317 for (pass = 0; pass < 2 && !yielded && try; pass++) {
2318 kvm_for_each_vcpu(i, vcpu, kvm) {
2319 if (!pass && i <= last_boosted_vcpu) {
2320 i = last_boosted_vcpu;
2322 } else if (pass && i > last_boosted_vcpu)
2324 if (!ACCESS_ONCE(vcpu->preempted))
2328 if (swait_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
2330 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
2333 yielded = kvm_vcpu_yield_to(vcpu);
2335 kvm->last_boosted_vcpu = i;
2337 } else if (yielded < 0) {
2344 kvm_vcpu_set_in_spin_loop(me, false);
2346 /* Ensure vcpu is not eligible during next spinloop */
2347 kvm_vcpu_set_dy_eligible(me, false);
2349 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
2351 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2353 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
2356 if (vmf->pgoff == 0)
2357 page = virt_to_page(vcpu->run);
2359 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
2360 page = virt_to_page(vcpu->arch.pio_data);
2362 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2363 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
2364 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
2367 return kvm_arch_vcpu_fault(vcpu, vmf);
2373 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
2374 .fault = kvm_vcpu_fault,
2377 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2379 vma->vm_ops = &kvm_vcpu_vm_ops;
2383 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2385 struct kvm_vcpu *vcpu = filp->private_data;
2387 debugfs_remove_recursive(vcpu->debugfs_dentry);
2388 kvm_put_kvm(vcpu->kvm);
2392 static struct file_operations kvm_vcpu_fops = {
2393 .release = kvm_vcpu_release,
2394 .unlocked_ioctl = kvm_vcpu_ioctl,
2395 #ifdef CONFIG_KVM_COMPAT
2396 .compat_ioctl = kvm_vcpu_compat_ioctl,
2398 .mmap = kvm_vcpu_mmap,
2399 .llseek = noop_llseek,
2403 * Allocates an inode for the vcpu.
2405 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2407 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
2410 static int kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
2412 char dir_name[ITOA_MAX_LEN * 2];
2415 if (!kvm_arch_has_vcpu_debugfs())
2418 if (!debugfs_initialized())
2421 snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
2422 vcpu->debugfs_dentry = debugfs_create_dir(dir_name,
2423 vcpu->kvm->debugfs_dentry);
2424 if (!vcpu->debugfs_dentry)
2427 ret = kvm_arch_create_vcpu_debugfs(vcpu);
2429 debugfs_remove_recursive(vcpu->debugfs_dentry);
2437 * Creates some virtual cpus. Good luck creating more than one.
2439 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
2442 struct kvm_vcpu *vcpu;
2444 if (id >= KVM_MAX_VCPU_ID)
2447 mutex_lock(&kvm->lock);
2448 if (kvm->created_vcpus == KVM_MAX_VCPUS) {
2449 mutex_unlock(&kvm->lock);
2453 kvm->created_vcpus++;
2454 mutex_unlock(&kvm->lock);
2456 vcpu = kvm_arch_vcpu_create(kvm, id);
2459 goto vcpu_decrement;
2462 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2464 r = kvm_arch_vcpu_setup(vcpu);
2468 r = kvm_create_vcpu_debugfs(vcpu);
2472 mutex_lock(&kvm->lock);
2473 if (kvm_get_vcpu_by_id(kvm, id)) {
2475 goto unlock_vcpu_destroy;
2478 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
2480 /* Now it's all set up, let userspace reach it */
2482 r = create_vcpu_fd(vcpu);
2485 goto unlock_vcpu_destroy;
2488 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
2491 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
2492 * before kvm->online_vcpu's incremented value.
2495 atomic_inc(&kvm->online_vcpus);
2497 mutex_unlock(&kvm->lock);
2498 kvm_arch_vcpu_postcreate(vcpu);
2501 unlock_vcpu_destroy:
2502 mutex_unlock(&kvm->lock);
2503 debugfs_remove_recursive(vcpu->debugfs_dentry);
2505 kvm_arch_vcpu_destroy(vcpu);
2507 mutex_lock(&kvm->lock);
2508 kvm->created_vcpus--;
2509 mutex_unlock(&kvm->lock);
2513 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2516 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2517 vcpu->sigset_active = 1;
2518 vcpu->sigset = *sigset;
2520 vcpu->sigset_active = 0;
2524 static long kvm_vcpu_ioctl(struct file *filp,
2525 unsigned int ioctl, unsigned long arg)
2527 struct kvm_vcpu *vcpu = filp->private_data;
2528 void __user *argp = (void __user *)arg;
2530 struct kvm_fpu *fpu = NULL;
2531 struct kvm_sregs *kvm_sregs = NULL;
2533 if (vcpu->kvm->mm != current->mm)
2536 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2539 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
2541 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
2542 * so vcpu_load() would break it.
2544 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_S390_IRQ || ioctl == KVM_INTERRUPT)
2545 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2549 r = vcpu_load(vcpu);
2557 if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
2558 /* The thread running this VCPU changed. */
2559 struct pid *oldpid = vcpu->pid;
2560 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
2562 rcu_assign_pointer(vcpu->pid, newpid);
2567 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2568 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2570 case KVM_GET_REGS: {
2571 struct kvm_regs *kvm_regs;
2574 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2577 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2581 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2588 case KVM_SET_REGS: {
2589 struct kvm_regs *kvm_regs;
2592 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2593 if (IS_ERR(kvm_regs)) {
2594 r = PTR_ERR(kvm_regs);
2597 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2601 case KVM_GET_SREGS: {
2602 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2606 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2610 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2615 case KVM_SET_SREGS: {
2616 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2617 if (IS_ERR(kvm_sregs)) {
2618 r = PTR_ERR(kvm_sregs);
2622 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2625 case KVM_GET_MP_STATE: {
2626 struct kvm_mp_state mp_state;
2628 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2632 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
2637 case KVM_SET_MP_STATE: {
2638 struct kvm_mp_state mp_state;
2641 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
2643 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2646 case KVM_TRANSLATE: {
2647 struct kvm_translation tr;
2650 if (copy_from_user(&tr, argp, sizeof(tr)))
2652 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2656 if (copy_to_user(argp, &tr, sizeof(tr)))
2661 case KVM_SET_GUEST_DEBUG: {
2662 struct kvm_guest_debug dbg;
2665 if (copy_from_user(&dbg, argp, sizeof(dbg)))
2667 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2670 case KVM_SET_SIGNAL_MASK: {
2671 struct kvm_signal_mask __user *sigmask_arg = argp;
2672 struct kvm_signal_mask kvm_sigmask;
2673 sigset_t sigset, *p;
2678 if (copy_from_user(&kvm_sigmask, argp,
2679 sizeof(kvm_sigmask)))
2682 if (kvm_sigmask.len != sizeof(sigset))
2685 if (copy_from_user(&sigset, sigmask_arg->sigset,
2690 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2694 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2698 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2702 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2708 fpu = memdup_user(argp, sizeof(*fpu));
2714 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2718 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2727 #ifdef CONFIG_KVM_COMPAT
2728 static long kvm_vcpu_compat_ioctl(struct file *filp,
2729 unsigned int ioctl, unsigned long arg)
2731 struct kvm_vcpu *vcpu = filp->private_data;
2732 void __user *argp = compat_ptr(arg);
2735 if (vcpu->kvm->mm != current->mm)
2739 case KVM_SET_SIGNAL_MASK: {
2740 struct kvm_signal_mask __user *sigmask_arg = argp;
2741 struct kvm_signal_mask kvm_sigmask;
2742 compat_sigset_t csigset;
2747 if (copy_from_user(&kvm_sigmask, argp,
2748 sizeof(kvm_sigmask)))
2751 if (kvm_sigmask.len != sizeof(csigset))
2754 if (copy_from_user(&csigset, sigmask_arg->sigset,
2757 sigset_from_compat(&sigset, &csigset);
2758 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2760 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2764 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2772 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2773 int (*accessor)(struct kvm_device *dev,
2774 struct kvm_device_attr *attr),
2777 struct kvm_device_attr attr;
2782 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2785 return accessor(dev, &attr);
2788 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2791 struct kvm_device *dev = filp->private_data;
2794 case KVM_SET_DEVICE_ATTR:
2795 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2796 case KVM_GET_DEVICE_ATTR:
2797 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2798 case KVM_HAS_DEVICE_ATTR:
2799 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2801 if (dev->ops->ioctl)
2802 return dev->ops->ioctl(dev, ioctl, arg);
2808 static int kvm_device_release(struct inode *inode, struct file *filp)
2810 struct kvm_device *dev = filp->private_data;
2811 struct kvm *kvm = dev->kvm;
2817 static const struct file_operations kvm_device_fops = {
2818 .unlocked_ioctl = kvm_device_ioctl,
2819 #ifdef CONFIG_KVM_COMPAT
2820 .compat_ioctl = kvm_device_ioctl,
2822 .release = kvm_device_release,
2825 struct kvm_device *kvm_device_from_filp(struct file *filp)
2827 if (filp->f_op != &kvm_device_fops)
2830 return filp->private_data;
2833 static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
2834 #ifdef CONFIG_KVM_MPIC
2835 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
2836 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
2839 #ifdef CONFIG_KVM_XICS
2840 [KVM_DEV_TYPE_XICS] = &kvm_xics_ops,
2844 int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
2846 if (type >= ARRAY_SIZE(kvm_device_ops_table))
2849 if (kvm_device_ops_table[type] != NULL)
2852 kvm_device_ops_table[type] = ops;
2856 void kvm_unregister_device_ops(u32 type)
2858 if (kvm_device_ops_table[type] != NULL)
2859 kvm_device_ops_table[type] = NULL;
2862 static int kvm_ioctl_create_device(struct kvm *kvm,
2863 struct kvm_create_device *cd)
2865 struct kvm_device_ops *ops = NULL;
2866 struct kvm_device *dev;
2867 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2870 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
2873 ops = kvm_device_ops_table[cd->type];
2880 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2887 mutex_lock(&kvm->lock);
2888 ret = ops->create(dev, cd->type);
2890 mutex_unlock(&kvm->lock);
2894 list_add(&dev->vm_node, &kvm->devices);
2895 mutex_unlock(&kvm->lock);
2900 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2902 mutex_lock(&kvm->lock);
2903 list_del(&dev->vm_node);
2904 mutex_unlock(&kvm->lock);
2914 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2917 case KVM_CAP_USER_MEMORY:
2918 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2919 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2920 case KVM_CAP_INTERNAL_ERROR_DATA:
2921 #ifdef CONFIG_HAVE_KVM_MSI
2922 case KVM_CAP_SIGNAL_MSI:
2924 #ifdef CONFIG_HAVE_KVM_IRQFD
2926 case KVM_CAP_IRQFD_RESAMPLE:
2928 case KVM_CAP_IOEVENTFD_ANY_LENGTH:
2929 case KVM_CAP_CHECK_EXTENSION_VM:
2931 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2932 case KVM_CAP_IRQ_ROUTING:
2933 return KVM_MAX_IRQ_ROUTES;
2935 #if KVM_ADDRESS_SPACE_NUM > 1
2936 case KVM_CAP_MULTI_ADDRESS_SPACE:
2937 return KVM_ADDRESS_SPACE_NUM;
2939 case KVM_CAP_MAX_VCPU_ID:
2940 return KVM_MAX_VCPU_ID;
2944 return kvm_vm_ioctl_check_extension(kvm, arg);
2947 static long kvm_vm_ioctl(struct file *filp,
2948 unsigned int ioctl, unsigned long arg)
2950 struct kvm *kvm = filp->private_data;
2951 void __user *argp = (void __user *)arg;
2954 if (kvm->mm != current->mm)
2957 case KVM_CREATE_VCPU:
2958 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2960 case KVM_SET_USER_MEMORY_REGION: {
2961 struct kvm_userspace_memory_region kvm_userspace_mem;
2964 if (copy_from_user(&kvm_userspace_mem, argp,
2965 sizeof(kvm_userspace_mem)))
2968 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2971 case KVM_GET_DIRTY_LOG: {
2972 struct kvm_dirty_log log;
2975 if (copy_from_user(&log, argp, sizeof(log)))
2977 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2980 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2981 case KVM_REGISTER_COALESCED_MMIO: {
2982 struct kvm_coalesced_mmio_zone zone;
2985 if (copy_from_user(&zone, argp, sizeof(zone)))
2987 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2990 case KVM_UNREGISTER_COALESCED_MMIO: {
2991 struct kvm_coalesced_mmio_zone zone;
2994 if (copy_from_user(&zone, argp, sizeof(zone)))
2996 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
3001 struct kvm_irqfd data;
3004 if (copy_from_user(&data, argp, sizeof(data)))
3006 r = kvm_irqfd(kvm, &data);
3009 case KVM_IOEVENTFD: {
3010 struct kvm_ioeventfd data;
3013 if (copy_from_user(&data, argp, sizeof(data)))
3015 r = kvm_ioeventfd(kvm, &data);
3018 #ifdef CONFIG_HAVE_KVM_MSI
3019 case KVM_SIGNAL_MSI: {
3023 if (copy_from_user(&msi, argp, sizeof(msi)))
3025 r = kvm_send_userspace_msi(kvm, &msi);
3029 #ifdef __KVM_HAVE_IRQ_LINE
3030 case KVM_IRQ_LINE_STATUS:
3031 case KVM_IRQ_LINE: {
3032 struct kvm_irq_level irq_event;
3035 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
3038 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
3039 ioctl == KVM_IRQ_LINE_STATUS);
3044 if (ioctl == KVM_IRQ_LINE_STATUS) {
3045 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
3053 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3054 case KVM_SET_GSI_ROUTING: {
3055 struct kvm_irq_routing routing;
3056 struct kvm_irq_routing __user *urouting;
3057 struct kvm_irq_routing_entry *entries = NULL;
3060 if (copy_from_user(&routing, argp, sizeof(routing)))
3063 if (routing.nr > KVM_MAX_IRQ_ROUTES)
3069 entries = vmalloc(routing.nr * sizeof(*entries));
3074 if (copy_from_user(entries, urouting->entries,
3075 routing.nr * sizeof(*entries)))
3076 goto out_free_irq_routing;
3078 r = kvm_set_irq_routing(kvm, entries, routing.nr,
3080 out_free_irq_routing:
3084 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
3085 case KVM_CREATE_DEVICE: {
3086 struct kvm_create_device cd;
3089 if (copy_from_user(&cd, argp, sizeof(cd)))
3092 r = kvm_ioctl_create_device(kvm, &cd);
3097 if (copy_to_user(argp, &cd, sizeof(cd)))
3103 case KVM_CHECK_EXTENSION:
3104 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
3107 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
3113 #ifdef CONFIG_KVM_COMPAT
3114 struct compat_kvm_dirty_log {
3118 compat_uptr_t dirty_bitmap; /* one bit per page */
3123 static long kvm_vm_compat_ioctl(struct file *filp,
3124 unsigned int ioctl, unsigned long arg)
3126 struct kvm *kvm = filp->private_data;
3129 if (kvm->mm != current->mm)
3132 case KVM_GET_DIRTY_LOG: {
3133 struct compat_kvm_dirty_log compat_log;
3134 struct kvm_dirty_log log;
3137 if (copy_from_user(&compat_log, (void __user *)arg,
3138 sizeof(compat_log)))
3140 log.slot = compat_log.slot;
3141 log.padding1 = compat_log.padding1;
3142 log.padding2 = compat_log.padding2;
3143 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
3145 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3149 r = kvm_vm_ioctl(filp, ioctl, arg);
3157 static struct file_operations kvm_vm_fops = {
3158 .release = kvm_vm_release,
3159 .unlocked_ioctl = kvm_vm_ioctl,
3160 #ifdef CONFIG_KVM_COMPAT
3161 .compat_ioctl = kvm_vm_compat_ioctl,
3163 .llseek = noop_llseek,
3166 static int kvm_dev_ioctl_create_vm(unsigned long type)
3172 kvm = kvm_create_vm(type);
3174 return PTR_ERR(kvm);
3175 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
3176 r = kvm_coalesced_mmio_init(kvm);
3182 r = get_unused_fd_flags(O_CLOEXEC);
3187 file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
3191 return PTR_ERR(file);
3194 if (kvm_create_vm_debugfs(kvm, r) < 0) {
3200 fd_install(r, file);
3204 static long kvm_dev_ioctl(struct file *filp,
3205 unsigned int ioctl, unsigned long arg)
3210 case KVM_GET_API_VERSION:
3213 r = KVM_API_VERSION;
3216 r = kvm_dev_ioctl_create_vm(arg);
3218 case KVM_CHECK_EXTENSION:
3219 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
3221 case KVM_GET_VCPU_MMAP_SIZE:
3224 r = PAGE_SIZE; /* struct kvm_run */
3226 r += PAGE_SIZE; /* pio data page */
3228 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
3229 r += PAGE_SIZE; /* coalesced mmio ring page */
3232 case KVM_TRACE_ENABLE:
3233 case KVM_TRACE_PAUSE:
3234 case KVM_TRACE_DISABLE:
3238 return kvm_arch_dev_ioctl(filp, ioctl, arg);
3244 static struct file_operations kvm_chardev_ops = {
3245 .unlocked_ioctl = kvm_dev_ioctl,
3246 .compat_ioctl = kvm_dev_ioctl,
3247 .llseek = noop_llseek,
3250 static struct miscdevice kvm_dev = {
3256 static void hardware_enable_nolock(void *junk)
3258 int cpu = raw_smp_processor_id();
3261 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
3264 cpumask_set_cpu(cpu, cpus_hardware_enabled);
3266 r = kvm_arch_hardware_enable();
3269 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3270 atomic_inc(&hardware_enable_failed);
3271 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
3275 static int kvm_starting_cpu(unsigned int cpu)
3277 raw_spin_lock(&kvm_count_lock);
3278 if (kvm_usage_count)
3279 hardware_enable_nolock(NULL);
3280 raw_spin_unlock(&kvm_count_lock);
3284 static void hardware_disable_nolock(void *junk)
3286 int cpu = raw_smp_processor_id();
3288 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
3290 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3291 kvm_arch_hardware_disable();
3294 static int kvm_dying_cpu(unsigned int cpu)
3296 raw_spin_lock(&kvm_count_lock);
3297 if (kvm_usage_count)
3298 hardware_disable_nolock(NULL);
3299 raw_spin_unlock(&kvm_count_lock);
3303 static void hardware_disable_all_nolock(void)
3305 BUG_ON(!kvm_usage_count);
3308 if (!kvm_usage_count)
3309 on_each_cpu(hardware_disable_nolock, NULL, 1);
3312 static void hardware_disable_all(void)
3314 raw_spin_lock(&kvm_count_lock);
3315 hardware_disable_all_nolock();
3316 raw_spin_unlock(&kvm_count_lock);
3319 static int hardware_enable_all(void)
3323 raw_spin_lock(&kvm_count_lock);
3326 if (kvm_usage_count == 1) {
3327 atomic_set(&hardware_enable_failed, 0);
3328 on_each_cpu(hardware_enable_nolock, NULL, 1);
3330 if (atomic_read(&hardware_enable_failed)) {
3331 hardware_disable_all_nolock();
3336 raw_spin_unlock(&kvm_count_lock);
3341 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3345 * Some (well, at least mine) BIOSes hang on reboot if
3348 * And Intel TXT required VMX off for all cpu when system shutdown.
3350 pr_info("kvm: exiting hardware virtualization\n");
3351 kvm_rebooting = true;
3352 on_each_cpu(hardware_disable_nolock, NULL, 1);
3356 static struct notifier_block kvm_reboot_notifier = {
3357 .notifier_call = kvm_reboot,
3361 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3365 for (i = 0; i < bus->dev_count; i++) {
3366 struct kvm_io_device *pos = bus->range[i].dev;
3368 kvm_iodevice_destructor(pos);
3373 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
3374 const struct kvm_io_range *r2)
3376 gpa_t addr1 = r1->addr;
3377 gpa_t addr2 = r2->addr;
3382 /* If r2->len == 0, match the exact address. If r2->len != 0,
3383 * accept any overlapping write. Any order is acceptable for
3384 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
3385 * we process all of them.
3398 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
3400 return kvm_io_bus_cmp(p1, p2);
3403 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
3404 gpa_t addr, int len)
3406 bus->range[bus->dev_count++] = (struct kvm_io_range) {
3412 sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
3413 kvm_io_bus_sort_cmp, NULL);
3418 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
3419 gpa_t addr, int len)
3421 struct kvm_io_range *range, key;
3424 key = (struct kvm_io_range) {
3429 range = bsearch(&key, bus->range, bus->dev_count,
3430 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
3434 off = range - bus->range;
3436 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
3442 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3443 struct kvm_io_range *range, const void *val)
3447 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3451 while (idx < bus->dev_count &&
3452 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3453 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
3462 /* kvm_io_bus_write - called under kvm->slots_lock */
3463 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3464 int len, const void *val)
3466 struct kvm_io_bus *bus;
3467 struct kvm_io_range range;
3470 range = (struct kvm_io_range) {
3475 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3476 r = __kvm_io_bus_write(vcpu, bus, &range, val);
3477 return r < 0 ? r : 0;
3480 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3481 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
3482 gpa_t addr, int len, const void *val, long cookie)
3484 struct kvm_io_bus *bus;
3485 struct kvm_io_range range;
3487 range = (struct kvm_io_range) {
3492 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3494 /* First try the device referenced by cookie. */
3495 if ((cookie >= 0) && (cookie < bus->dev_count) &&
3496 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
3497 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
3502 * cookie contained garbage; fall back to search and return the
3503 * correct cookie value.
3505 return __kvm_io_bus_write(vcpu, bus, &range, val);
3508 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3509 struct kvm_io_range *range, void *val)
3513 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3517 while (idx < bus->dev_count &&
3518 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3519 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
3527 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
3529 /* kvm_io_bus_read - called under kvm->slots_lock */
3530 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3533 struct kvm_io_bus *bus;
3534 struct kvm_io_range range;
3537 range = (struct kvm_io_range) {
3542 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3543 r = __kvm_io_bus_read(vcpu, bus, &range, val);
3544 return r < 0 ? r : 0;
3548 /* Caller must hold slots_lock. */
3549 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3550 int len, struct kvm_io_device *dev)
3552 struct kvm_io_bus *new_bus, *bus;
3554 bus = kvm->buses[bus_idx];
3555 /* exclude ioeventfd which is limited by maximum fd */
3556 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3559 new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count + 1) *
3560 sizeof(struct kvm_io_range)), GFP_KERNEL);
3563 memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
3564 sizeof(struct kvm_io_range)));
3565 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
3566 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3567 synchronize_srcu_expedited(&kvm->srcu);
3573 /* Caller must hold slots_lock. */
3574 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3575 struct kvm_io_device *dev)
3578 struct kvm_io_bus *new_bus, *bus;
3580 bus = kvm->buses[bus_idx];
3582 for (i = 0; i < bus->dev_count; i++)
3583 if (bus->range[i].dev == dev) {
3591 new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3592 sizeof(struct kvm_io_range)), GFP_KERNEL);
3596 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3597 new_bus->dev_count--;
3598 memcpy(new_bus->range + i, bus->range + i + 1,
3599 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3601 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3602 synchronize_srcu_expedited(&kvm->srcu);
3607 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3610 struct kvm_io_bus *bus;
3611 int dev_idx, srcu_idx;
3612 struct kvm_io_device *iodev = NULL;
3614 srcu_idx = srcu_read_lock(&kvm->srcu);
3616 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
3618 dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
3622 iodev = bus->range[dev_idx].dev;
3625 srcu_read_unlock(&kvm->srcu, srcu_idx);
3629 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
3631 static int kvm_debugfs_open(struct inode *inode, struct file *file,
3632 int (*get)(void *, u64 *), int (*set)(void *, u64),
3635 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
3638 /* The debugfs files are a reference to the kvm struct which
3639 * is still valid when kvm_destroy_vm is called.
3640 * To avoid the race between open and the removal of the debugfs
3641 * directory we test against the users count.
3643 if (!atomic_add_unless(&stat_data->kvm->users_count, 1, 0))
3646 if (simple_attr_open(inode, file, get, set, fmt)) {
3647 kvm_put_kvm(stat_data->kvm);
3654 static int kvm_debugfs_release(struct inode *inode, struct file *file)
3656 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
3659 simple_attr_release(inode, file);
3660 kvm_put_kvm(stat_data->kvm);
3665 static int vm_stat_get_per_vm(void *data, u64 *val)
3667 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3669 *val = *(ulong *)((void *)stat_data->kvm + stat_data->offset);
3674 static int vm_stat_clear_per_vm(void *data, u64 val)
3676 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3681 *(ulong *)((void *)stat_data->kvm + stat_data->offset) = 0;
3686 static int vm_stat_get_per_vm_open(struct inode *inode, struct file *file)
3688 __simple_attr_check_format("%llu\n", 0ull);
3689 return kvm_debugfs_open(inode, file, vm_stat_get_per_vm,
3690 vm_stat_clear_per_vm, "%llu\n");
3693 static const struct file_operations vm_stat_get_per_vm_fops = {
3694 .owner = THIS_MODULE,
3695 .open = vm_stat_get_per_vm_open,
3696 .release = kvm_debugfs_release,
3697 .read = simple_attr_read,
3698 .write = simple_attr_write,
3699 .llseek = generic_file_llseek,
3702 static int vcpu_stat_get_per_vm(void *data, u64 *val)
3705 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3706 struct kvm_vcpu *vcpu;
3710 kvm_for_each_vcpu(i, vcpu, stat_data->kvm)
3711 *val += *(u64 *)((void *)vcpu + stat_data->offset);
3716 static int vcpu_stat_clear_per_vm(void *data, u64 val)
3719 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3720 struct kvm_vcpu *vcpu;
3725 kvm_for_each_vcpu(i, vcpu, stat_data->kvm)
3726 *(u64 *)((void *)vcpu + stat_data->offset) = 0;
3731 static int vcpu_stat_get_per_vm_open(struct inode *inode, struct file *file)
3733 __simple_attr_check_format("%llu\n", 0ull);
3734 return kvm_debugfs_open(inode, file, vcpu_stat_get_per_vm,
3735 vcpu_stat_clear_per_vm, "%llu\n");
3738 static const struct file_operations vcpu_stat_get_per_vm_fops = {
3739 .owner = THIS_MODULE,
3740 .open = vcpu_stat_get_per_vm_open,
3741 .release = kvm_debugfs_release,
3742 .read = simple_attr_read,
3743 .write = simple_attr_write,
3744 .llseek = generic_file_llseek,
3747 static const struct file_operations *stat_fops_per_vm[] = {
3748 [KVM_STAT_VCPU] = &vcpu_stat_get_per_vm_fops,
3749 [KVM_STAT_VM] = &vm_stat_get_per_vm_fops,
3752 static int vm_stat_get(void *_offset, u64 *val)
3754 unsigned offset = (long)_offset;
3756 struct kvm_stat_data stat_tmp = {.offset = offset};
3760 spin_lock(&kvm_lock);
3761 list_for_each_entry(kvm, &vm_list, vm_list) {
3763 vm_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
3766 spin_unlock(&kvm_lock);
3770 static int vm_stat_clear(void *_offset, u64 val)
3772 unsigned offset = (long)_offset;
3774 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_clear_per_vm((void *)&stat_tmp, 0);
3784 spin_unlock(&kvm_lock);
3789 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n");
3791 static int vcpu_stat_get(void *_offset, u64 *val)
3793 unsigned offset = (long)_offset;
3795 struct kvm_stat_data stat_tmp = {.offset = offset};
3799 spin_lock(&kvm_lock);
3800 list_for_each_entry(kvm, &vm_list, vm_list) {
3802 vcpu_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
3805 spin_unlock(&kvm_lock);
3809 static int vcpu_stat_clear(void *_offset, u64 val)
3811 unsigned offset = (long)_offset;
3813 struct kvm_stat_data stat_tmp = {.offset = offset};
3818 spin_lock(&kvm_lock);
3819 list_for_each_entry(kvm, &vm_list, vm_list) {
3821 vcpu_stat_clear_per_vm((void *)&stat_tmp, 0);
3823 spin_unlock(&kvm_lock);
3828 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear,
3831 static const struct file_operations *stat_fops[] = {
3832 [KVM_STAT_VCPU] = &vcpu_stat_fops,
3833 [KVM_STAT_VM] = &vm_stat_fops,
3836 static int kvm_init_debug(void)
3839 struct kvm_stats_debugfs_item *p;
3841 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3842 if (kvm_debugfs_dir == NULL)
3845 kvm_debugfs_num_entries = 0;
3846 for (p = debugfs_entries; p->name; ++p, kvm_debugfs_num_entries++) {
3847 if (!debugfs_create_file(p->name, 0644, kvm_debugfs_dir,
3848 (void *)(long)p->offset,
3849 stat_fops[p->kind]))
3856 debugfs_remove_recursive(kvm_debugfs_dir);
3861 static int kvm_suspend(void)
3863 if (kvm_usage_count)
3864 hardware_disable_nolock(NULL);
3868 static void kvm_resume(void)
3870 if (kvm_usage_count) {
3871 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3872 hardware_enable_nolock(NULL);
3876 static struct syscore_ops kvm_syscore_ops = {
3877 .suspend = kvm_suspend,
3878 .resume = kvm_resume,
3882 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3884 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3887 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3889 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3891 if (vcpu->preempted)
3892 vcpu->preempted = false;
3894 kvm_arch_sched_in(vcpu, cpu);
3896 kvm_arch_vcpu_load(vcpu, cpu);
3899 static void kvm_sched_out(struct preempt_notifier *pn,
3900 struct task_struct *next)
3902 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3904 if (current->state == TASK_RUNNING)
3905 vcpu->preempted = true;
3906 kvm_arch_vcpu_put(vcpu);
3909 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3910 struct module *module)
3915 r = kvm_arch_init(opaque);
3920 * kvm_arch_init makes sure there's at most one caller
3921 * for architectures that support multiple implementations,
3922 * like intel and amd on x86.
3923 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3924 * conflicts in case kvm is already setup for another implementation.
3926 r = kvm_irqfd_init();
3930 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
3935 r = kvm_arch_hardware_setup();
3939 for_each_online_cpu(cpu) {
3940 smp_call_function_single(cpu,
3941 kvm_arch_check_processor_compat,
3947 r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "AP_KVM_STARTING",
3948 kvm_starting_cpu, kvm_dying_cpu);
3951 register_reboot_notifier(&kvm_reboot_notifier);
3953 /* A kmem cache lets us meet the alignment requirements of fx_save. */
3955 vcpu_align = __alignof__(struct kvm_vcpu);
3956 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
3958 if (!kvm_vcpu_cache) {
3963 r = kvm_async_pf_init();
3967 kvm_chardev_ops.owner = module;
3968 kvm_vm_fops.owner = module;
3969 kvm_vcpu_fops.owner = module;
3971 r = misc_register(&kvm_dev);
3973 pr_err("kvm: misc device register failed\n");
3977 register_syscore_ops(&kvm_syscore_ops);
3979 kvm_preempt_ops.sched_in = kvm_sched_in;
3980 kvm_preempt_ops.sched_out = kvm_sched_out;
3982 r = kvm_init_debug();
3984 pr_err("kvm: create debugfs files failed\n");
3988 r = kvm_vfio_ops_init();
3994 unregister_syscore_ops(&kvm_syscore_ops);
3995 misc_deregister(&kvm_dev);
3997 kvm_async_pf_deinit();
3999 kmem_cache_destroy(kvm_vcpu_cache);
4001 unregister_reboot_notifier(&kvm_reboot_notifier);
4002 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4005 kvm_arch_hardware_unsetup();
4007 free_cpumask_var(cpus_hardware_enabled);
4015 EXPORT_SYMBOL_GPL(kvm_init);
4019 debugfs_remove_recursive(kvm_debugfs_dir);
4020 misc_deregister(&kvm_dev);
4021 kmem_cache_destroy(kvm_vcpu_cache);
4022 kvm_async_pf_deinit();
4023 unregister_syscore_ops(&kvm_syscore_ops);
4024 unregister_reboot_notifier(&kvm_reboot_notifier);
4025 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4026 on_each_cpu(hardware_disable_nolock, NULL, 1);
4027 kvm_arch_hardware_unsetup();
4030 free_cpumask_var(cpus_hardware_enabled);
4031 kvm_vfio_ops_exit();
4033 EXPORT_SYMBOL_GPL(kvm_exit);