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 <asm/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 static unsigned int halt_poll_ns = KVM_HALT_POLL_NS_DEFAULT;
74 module_param(halt_poll_ns, uint, S_IRUGO | S_IWUSR);
76 /* Default doubles per-vcpu halt_poll_ns. */
77 static unsigned int halt_poll_ns_grow = 2;
78 module_param(halt_poll_ns_grow, uint, S_IRUGO | S_IWUSR);
80 /* Default resets per-vcpu halt_poll_ns . */
81 static unsigned int halt_poll_ns_shrink;
82 module_param(halt_poll_ns_shrink, uint, S_IRUGO | S_IWUSR);
87 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
90 DEFINE_SPINLOCK(kvm_lock);
91 static DEFINE_RAW_SPINLOCK(kvm_count_lock);
94 static cpumask_var_t cpus_hardware_enabled;
95 static int kvm_usage_count;
96 static atomic_t hardware_enable_failed;
98 struct kmem_cache *kvm_vcpu_cache;
99 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
101 static __read_mostly struct preempt_ops kvm_preempt_ops;
103 struct dentry *kvm_debugfs_dir;
104 EXPORT_SYMBOL_GPL(kvm_debugfs_dir);
106 static int kvm_debugfs_num_entries;
107 static const struct file_operations *stat_fops_per_vm[];
109 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
111 #ifdef CONFIG_KVM_COMPAT
112 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
115 static int hardware_enable_all(void);
116 static void hardware_disable_all(void);
118 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
120 static void kvm_release_pfn_dirty(kvm_pfn_t pfn);
121 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot, gfn_t gfn);
123 __visible bool kvm_rebooting;
124 EXPORT_SYMBOL_GPL(kvm_rebooting);
126 static bool largepages_enabled = true;
128 bool kvm_is_reserved_pfn(kvm_pfn_t pfn)
131 return PageReserved(pfn_to_page(pfn));
137 * Switches to specified vcpu, until a matching vcpu_put()
139 int vcpu_load(struct kvm_vcpu *vcpu)
143 if (mutex_lock_killable(&vcpu->mutex))
146 preempt_notifier_register(&vcpu->preempt_notifier);
147 kvm_arch_vcpu_load(vcpu, cpu);
151 EXPORT_SYMBOL_GPL(vcpu_load);
153 void vcpu_put(struct kvm_vcpu *vcpu)
156 kvm_arch_vcpu_put(vcpu);
157 preempt_notifier_unregister(&vcpu->preempt_notifier);
159 mutex_unlock(&vcpu->mutex);
161 EXPORT_SYMBOL_GPL(vcpu_put);
163 static void ack_flush(void *_completed)
167 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
172 struct kvm_vcpu *vcpu;
174 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
177 kvm_for_each_vcpu(i, vcpu, kvm) {
178 kvm_make_request(req, vcpu);
181 /* Set ->requests bit before we read ->mode. */
182 smp_mb__after_atomic();
184 if (cpus != NULL && cpu != -1 && cpu != me &&
185 kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
186 cpumask_set_cpu(cpu, cpus);
188 if (unlikely(cpus == NULL))
189 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
190 else if (!cpumask_empty(cpus))
191 smp_call_function_many(cpus, ack_flush, NULL, 1);
195 free_cpumask_var(cpus);
199 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
200 void kvm_flush_remote_tlbs(struct kvm *kvm)
203 * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in
204 * kvm_make_all_cpus_request.
206 long dirty_count = smp_load_acquire(&kvm->tlbs_dirty);
209 * We want to publish modifications to the page tables before reading
210 * mode. Pairs with a memory barrier in arch-specific code.
211 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
212 * and smp_mb in walk_shadow_page_lockless_begin/end.
213 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
215 * There is already an smp_mb__after_atomic() before
216 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
219 if (kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
220 ++kvm->stat.remote_tlb_flush;
221 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
223 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
226 void kvm_reload_remote_mmus(struct kvm *kvm)
228 kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
231 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
236 mutex_init(&vcpu->mutex);
241 init_swait_queue_head(&vcpu->wq);
242 kvm_async_pf_vcpu_init(vcpu);
245 INIT_LIST_HEAD(&vcpu->blocked_vcpu_list);
247 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
252 vcpu->run = page_address(page);
254 kvm_vcpu_set_in_spin_loop(vcpu, false);
255 kvm_vcpu_set_dy_eligible(vcpu, false);
256 vcpu->preempted = false;
258 r = kvm_arch_vcpu_init(vcpu);
264 free_page((unsigned long)vcpu->run);
268 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
270 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
273 kvm_arch_vcpu_uninit(vcpu);
274 free_page((unsigned long)vcpu->run);
276 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
278 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
279 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
281 return container_of(mn, struct kvm, mmu_notifier);
284 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
285 struct mm_struct *mm,
286 unsigned long address)
288 struct kvm *kvm = mmu_notifier_to_kvm(mn);
289 int need_tlb_flush, idx;
292 * When ->invalidate_page runs, the linux pte has been zapped
293 * already but the page is still allocated until
294 * ->invalidate_page returns. So if we increase the sequence
295 * here the kvm page fault will notice if the spte can't be
296 * established because the page is going to be freed. If
297 * instead the kvm page fault establishes the spte before
298 * ->invalidate_page runs, kvm_unmap_hva will release it
301 * The sequence increase only need to be seen at spin_unlock
302 * time, and not at spin_lock time.
304 * Increasing the sequence after the spin_unlock would be
305 * unsafe because the kvm page fault could then establish the
306 * pte after kvm_unmap_hva returned, without noticing the page
307 * is going to be freed.
309 idx = srcu_read_lock(&kvm->srcu);
310 spin_lock(&kvm->mmu_lock);
312 kvm->mmu_notifier_seq++;
313 need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
314 /* we've to flush the tlb before the pages can be freed */
316 kvm_flush_remote_tlbs(kvm);
318 spin_unlock(&kvm->mmu_lock);
320 kvm_arch_mmu_notifier_invalidate_page(kvm, address);
322 srcu_read_unlock(&kvm->srcu, idx);
325 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
326 struct mm_struct *mm,
327 unsigned long address,
330 struct kvm *kvm = mmu_notifier_to_kvm(mn);
333 idx = srcu_read_lock(&kvm->srcu);
334 spin_lock(&kvm->mmu_lock);
335 kvm->mmu_notifier_seq++;
336 kvm_set_spte_hva(kvm, address, pte);
337 spin_unlock(&kvm->mmu_lock);
338 srcu_read_unlock(&kvm->srcu, idx);
341 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
342 struct mm_struct *mm,
346 struct kvm *kvm = mmu_notifier_to_kvm(mn);
347 int need_tlb_flush = 0, idx;
349 idx = srcu_read_lock(&kvm->srcu);
350 spin_lock(&kvm->mmu_lock);
352 * The count increase must become visible at unlock time as no
353 * spte can be established without taking the mmu_lock and
354 * count is also read inside the mmu_lock critical section.
356 kvm->mmu_notifier_count++;
357 need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
358 need_tlb_flush |= kvm->tlbs_dirty;
359 /* we've to flush the tlb before the pages can be freed */
361 kvm_flush_remote_tlbs(kvm);
363 spin_unlock(&kvm->mmu_lock);
364 srcu_read_unlock(&kvm->srcu, idx);
367 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
368 struct mm_struct *mm,
372 struct kvm *kvm = mmu_notifier_to_kvm(mn);
374 spin_lock(&kvm->mmu_lock);
376 * This sequence increase will notify the kvm page fault that
377 * the page that is going to be mapped in the spte could have
380 kvm->mmu_notifier_seq++;
383 * The above sequence increase must be visible before the
384 * below count decrease, which is ensured by the smp_wmb above
385 * in conjunction with the smp_rmb in mmu_notifier_retry().
387 kvm->mmu_notifier_count--;
388 spin_unlock(&kvm->mmu_lock);
390 BUG_ON(kvm->mmu_notifier_count < 0);
393 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
394 struct mm_struct *mm,
398 struct kvm *kvm = mmu_notifier_to_kvm(mn);
401 idx = srcu_read_lock(&kvm->srcu);
402 spin_lock(&kvm->mmu_lock);
404 young = kvm_age_hva(kvm, start, end);
406 kvm_flush_remote_tlbs(kvm);
408 spin_unlock(&kvm->mmu_lock);
409 srcu_read_unlock(&kvm->srcu, idx);
414 static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
415 struct mm_struct *mm,
419 struct kvm *kvm = mmu_notifier_to_kvm(mn);
422 idx = srcu_read_lock(&kvm->srcu);
423 spin_lock(&kvm->mmu_lock);
425 * Even though we do not flush TLB, this will still adversely
426 * affect performance on pre-Haswell Intel EPT, where there is
427 * no EPT Access Bit to clear so that we have to tear down EPT
428 * tables instead. If we find this unacceptable, we can always
429 * add a parameter to kvm_age_hva so that it effectively doesn't
430 * do anything on clear_young.
432 * Also note that currently we never issue secondary TLB flushes
433 * from clear_young, leaving this job up to the regular system
434 * cadence. If we find this inaccurate, we might come up with a
435 * more sophisticated heuristic later.
437 young = kvm_age_hva(kvm, start, end);
438 spin_unlock(&kvm->mmu_lock);
439 srcu_read_unlock(&kvm->srcu, idx);
444 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
445 struct mm_struct *mm,
446 unsigned long address)
448 struct kvm *kvm = mmu_notifier_to_kvm(mn);
451 idx = srcu_read_lock(&kvm->srcu);
452 spin_lock(&kvm->mmu_lock);
453 young = kvm_test_age_hva(kvm, address);
454 spin_unlock(&kvm->mmu_lock);
455 srcu_read_unlock(&kvm->srcu, idx);
460 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
461 struct mm_struct *mm)
463 struct kvm *kvm = mmu_notifier_to_kvm(mn);
466 idx = srcu_read_lock(&kvm->srcu);
467 kvm_arch_flush_shadow_all(kvm);
468 srcu_read_unlock(&kvm->srcu, idx);
471 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
472 .invalidate_page = kvm_mmu_notifier_invalidate_page,
473 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
474 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
475 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
476 .clear_young = kvm_mmu_notifier_clear_young,
477 .test_young = kvm_mmu_notifier_test_young,
478 .change_pte = kvm_mmu_notifier_change_pte,
479 .release = kvm_mmu_notifier_release,
482 static int kvm_init_mmu_notifier(struct kvm *kvm)
484 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
485 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
488 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
490 static int kvm_init_mmu_notifier(struct kvm *kvm)
495 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
497 static struct kvm_memslots *kvm_alloc_memslots(void)
500 struct kvm_memslots *slots;
502 slots = kvm_kvzalloc(sizeof(struct kvm_memslots));
507 * Init kvm generation close to the maximum to easily test the
508 * code of handling generation number wrap-around.
510 slots->generation = -150;
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, 0444,
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);
616 atomic_inc(¤t->mm->mm_count);
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 atomic_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 kvm->memslots[i] = kvm_alloc_memslots();
642 if (!kvm->memslots[i])
643 goto out_err_no_srcu;
646 if (init_srcu_struct(&kvm->srcu))
647 goto out_err_no_srcu;
648 if (init_srcu_struct(&kvm->irq_srcu))
649 goto out_err_no_irq_srcu;
650 for (i = 0; i < KVM_NR_BUSES; i++) {
651 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
657 r = kvm_init_mmu_notifier(kvm);
661 spin_lock(&kvm_lock);
662 list_add(&kvm->vm_list, &vm_list);
663 spin_unlock(&kvm_lock);
665 preempt_notifier_inc();
670 cleanup_srcu_struct(&kvm->irq_srcu);
672 cleanup_srcu_struct(&kvm->srcu);
674 hardware_disable_all();
676 for (i = 0; i < KVM_NR_BUSES; i++)
677 kfree(kvm->buses[i]);
678 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
679 kvm_free_memslots(kvm, kvm->memslots[i]);
680 kvm_arch_free_vm(kvm);
686 * Avoid using vmalloc for a small buffer.
687 * Should not be used when the size is statically known.
689 void *kvm_kvzalloc(unsigned long size)
691 if (size > PAGE_SIZE)
692 return vzalloc(size);
694 return kzalloc(size, GFP_KERNEL);
697 static void kvm_destroy_devices(struct kvm *kvm)
699 struct kvm_device *dev, *tmp;
702 * We do not need to take the kvm->lock here, because nobody else
703 * has a reference to the struct kvm at this point and therefore
704 * cannot access the devices list anyhow.
706 list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
707 list_del(&dev->vm_node);
708 dev->ops->destroy(dev);
712 static void kvm_destroy_vm(struct kvm *kvm)
715 struct mm_struct *mm = kvm->mm;
717 kvm_destroy_vm_debugfs(kvm);
718 kvm_arch_sync_events(kvm);
719 spin_lock(&kvm_lock);
720 list_del(&kvm->vm_list);
721 spin_unlock(&kvm_lock);
722 kvm_free_irq_routing(kvm);
723 for (i = 0; i < KVM_NR_BUSES; i++)
724 kvm_io_bus_destroy(kvm->buses[i]);
725 kvm_coalesced_mmio_free(kvm);
726 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
727 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
729 kvm_arch_flush_shadow_all(kvm);
731 kvm_arch_destroy_vm(kvm);
732 kvm_destroy_devices(kvm);
733 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
734 kvm_free_memslots(kvm, kvm->memslots[i]);
735 cleanup_srcu_struct(&kvm->irq_srcu);
736 cleanup_srcu_struct(&kvm->srcu);
737 kvm_arch_free_vm(kvm);
738 preempt_notifier_dec();
739 hardware_disable_all();
743 void kvm_get_kvm(struct kvm *kvm)
745 atomic_inc(&kvm->users_count);
747 EXPORT_SYMBOL_GPL(kvm_get_kvm);
749 void kvm_put_kvm(struct kvm *kvm)
751 if (atomic_dec_and_test(&kvm->users_count))
754 EXPORT_SYMBOL_GPL(kvm_put_kvm);
757 static int kvm_vm_release(struct inode *inode, struct file *filp)
759 struct kvm *kvm = filp->private_data;
761 kvm_irqfd_release(kvm);
768 * Allocation size is twice as large as the actual dirty bitmap size.
769 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
771 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
773 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
775 memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
776 if (!memslot->dirty_bitmap)
783 * Insert memslot and re-sort memslots based on their GFN,
784 * so binary search could be used to lookup GFN.
785 * Sorting algorithm takes advantage of having initially
786 * sorted array and known changed memslot position.
788 static void update_memslots(struct kvm_memslots *slots,
789 struct kvm_memory_slot *new)
792 int i = slots->id_to_index[id];
793 struct kvm_memory_slot *mslots = slots->memslots;
795 WARN_ON(mslots[i].id != id);
797 WARN_ON(!mslots[i].npages);
798 if (mslots[i].npages)
801 if (!mslots[i].npages)
805 while (i < KVM_MEM_SLOTS_NUM - 1 &&
806 new->base_gfn <= mslots[i + 1].base_gfn) {
807 if (!mslots[i + 1].npages)
809 mslots[i] = mslots[i + 1];
810 slots->id_to_index[mslots[i].id] = i;
815 * The ">=" is needed when creating a slot with base_gfn == 0,
816 * so that it moves before all those with base_gfn == npages == 0.
818 * On the other hand, if new->npages is zero, the above loop has
819 * already left i pointing to the beginning of the empty part of
820 * mslots, and the ">=" would move the hole backwards in this
821 * case---which is wrong. So skip the loop when deleting a slot.
825 new->base_gfn >= mslots[i - 1].base_gfn) {
826 mslots[i] = mslots[i - 1];
827 slots->id_to_index[mslots[i].id] = i;
831 WARN_ON_ONCE(i != slots->used_slots);
834 slots->id_to_index[mslots[i].id] = i;
837 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
839 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
841 #ifdef __KVM_HAVE_READONLY_MEM
842 valid_flags |= KVM_MEM_READONLY;
845 if (mem->flags & ~valid_flags)
851 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
852 int as_id, struct kvm_memslots *slots)
854 struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id);
857 * Set the low bit in the generation, which disables SPTE caching
858 * until the end of synchronize_srcu_expedited.
860 WARN_ON(old_memslots->generation & 1);
861 slots->generation = old_memslots->generation + 1;
863 rcu_assign_pointer(kvm->memslots[as_id], slots);
864 synchronize_srcu_expedited(&kvm->srcu);
867 * Increment the new memslot generation a second time. This prevents
868 * vm exits that race with memslot updates from caching a memslot
869 * generation that will (potentially) be valid forever.
873 kvm_arch_memslots_updated(kvm, slots);
879 * Allocate some memory and give it an address in the guest physical address
882 * Discontiguous memory is allowed, mostly for framebuffers.
884 * Must be called holding kvm->slots_lock for write.
886 int __kvm_set_memory_region(struct kvm *kvm,
887 const struct kvm_userspace_memory_region *mem)
891 unsigned long npages;
892 struct kvm_memory_slot *slot;
893 struct kvm_memory_slot old, new;
894 struct kvm_memslots *slots = NULL, *old_memslots;
896 enum kvm_mr_change change;
898 r = check_memory_region_flags(mem);
903 as_id = mem->slot >> 16;
906 /* General sanity checks */
907 if (mem->memory_size & (PAGE_SIZE - 1))
909 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
911 /* We can read the guest memory with __xxx_user() later on. */
912 if ((id < KVM_USER_MEM_SLOTS) &&
913 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
914 !access_ok(VERIFY_WRITE,
915 (void __user *)(unsigned long)mem->userspace_addr,
918 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
920 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
923 slot = id_to_memslot(__kvm_memslots(kvm, as_id), id);
924 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
925 npages = mem->memory_size >> PAGE_SHIFT;
927 if (npages > KVM_MEM_MAX_NR_PAGES)
933 new.base_gfn = base_gfn;
935 new.flags = mem->flags;
939 change = KVM_MR_CREATE;
940 else { /* Modify an existing slot. */
941 if ((mem->userspace_addr != old.userspace_addr) ||
942 (npages != old.npages) ||
943 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
946 if (base_gfn != old.base_gfn)
947 change = KVM_MR_MOVE;
948 else if (new.flags != old.flags)
949 change = KVM_MR_FLAGS_ONLY;
950 else { /* Nothing to change. */
959 change = KVM_MR_DELETE;
964 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
965 /* Check for overlaps */
967 kvm_for_each_memslot(slot, __kvm_memslots(kvm, as_id)) {
968 if ((slot->id >= KVM_USER_MEM_SLOTS) ||
971 if (!((base_gfn + npages <= slot->base_gfn) ||
972 (base_gfn >= slot->base_gfn + slot->npages)))
977 /* Free page dirty bitmap if unneeded */
978 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
979 new.dirty_bitmap = NULL;
982 if (change == KVM_MR_CREATE) {
983 new.userspace_addr = mem->userspace_addr;
985 if (kvm_arch_create_memslot(kvm, &new, npages))
989 /* Allocate page dirty bitmap if needed */
990 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
991 if (kvm_create_dirty_bitmap(&new) < 0)
995 slots = kvm_kvzalloc(sizeof(struct kvm_memslots));
998 memcpy(slots, __kvm_memslots(kvm, as_id), sizeof(struct kvm_memslots));
1000 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
1001 slot = id_to_memslot(slots, id);
1002 slot->flags |= KVM_MEMSLOT_INVALID;
1004 old_memslots = install_new_memslots(kvm, as_id, slots);
1006 /* slot was deleted or moved, clear iommu mapping */
1007 kvm_iommu_unmap_pages(kvm, &old);
1008 /* From this point no new shadow pages pointing to a deleted,
1009 * or moved, memslot will be created.
1011 * validation of sp->gfn happens in:
1012 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1013 * - kvm_is_visible_gfn (mmu_check_roots)
1015 kvm_arch_flush_shadow_memslot(kvm, slot);
1018 * We can re-use the old_memslots from above, the only difference
1019 * from the currently installed memslots is the invalid flag. This
1020 * will get overwritten by update_memslots anyway.
1022 slots = old_memslots;
1025 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
1029 /* actual memory is freed via old in kvm_free_memslot below */
1030 if (change == KVM_MR_DELETE) {
1031 new.dirty_bitmap = NULL;
1032 memset(&new.arch, 0, sizeof(new.arch));
1035 update_memslots(slots, &new);
1036 old_memslots = install_new_memslots(kvm, as_id, slots);
1038 kvm_arch_commit_memory_region(kvm, mem, &old, &new, change);
1040 kvm_free_memslot(kvm, &old, &new);
1041 kvfree(old_memslots);
1044 * IOMMU mapping: New slots need to be mapped. Old slots need to be
1045 * un-mapped and re-mapped if their base changes. Since base change
1046 * unmapping is handled above with slot deletion, mapping alone is
1047 * needed here. Anything else the iommu might care about for existing
1048 * slots (size changes, userspace addr changes and read-only flag
1049 * changes) is disallowed above, so any other attribute changes getting
1050 * here can be skipped.
1052 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
1053 r = kvm_iommu_map_pages(kvm, &new);
1062 kvm_free_memslot(kvm, &new, &old);
1066 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1068 int kvm_set_memory_region(struct kvm *kvm,
1069 const struct kvm_userspace_memory_region *mem)
1073 mutex_lock(&kvm->slots_lock);
1074 r = __kvm_set_memory_region(kvm, mem);
1075 mutex_unlock(&kvm->slots_lock);
1078 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1080 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1081 struct kvm_userspace_memory_region *mem)
1083 if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1086 return kvm_set_memory_region(kvm, mem);
1089 int kvm_get_dirty_log(struct kvm *kvm,
1090 struct kvm_dirty_log *log, int *is_dirty)
1092 struct kvm_memslots *slots;
1093 struct kvm_memory_slot *memslot;
1094 int r, i, as_id, id;
1096 unsigned long any = 0;
1099 as_id = log->slot >> 16;
1100 id = (u16)log->slot;
1101 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1104 slots = __kvm_memslots(kvm, as_id);
1105 memslot = id_to_memslot(slots, id);
1107 if (!memslot->dirty_bitmap)
1110 n = kvm_dirty_bitmap_bytes(memslot);
1112 for (i = 0; !any && i < n/sizeof(long); ++i)
1113 any = memslot->dirty_bitmap[i];
1116 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1126 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1128 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1130 * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages
1131 * are dirty write protect them for next write.
1132 * @kvm: pointer to kvm instance
1133 * @log: slot id and address to which we copy the log
1134 * @is_dirty: flag set if any page is dirty
1136 * We need to keep it in mind that VCPU threads can write to the bitmap
1137 * concurrently. So, to avoid losing track of dirty pages we keep the
1140 * 1. Take a snapshot of the bit and clear it if needed.
1141 * 2. Write protect the corresponding page.
1142 * 3. Copy the snapshot to the userspace.
1143 * 4. Upon return caller flushes TLB's if needed.
1145 * Between 2 and 4, the guest may write to the page using the remaining TLB
1146 * entry. This is not a problem because the page is reported dirty using
1147 * the snapshot taken before and step 4 ensures that writes done after
1148 * exiting to userspace will be logged for the next call.
1151 int kvm_get_dirty_log_protect(struct kvm *kvm,
1152 struct kvm_dirty_log *log, bool *is_dirty)
1154 struct kvm_memslots *slots;
1155 struct kvm_memory_slot *memslot;
1156 int r, i, as_id, id;
1158 unsigned long *dirty_bitmap;
1159 unsigned long *dirty_bitmap_buffer;
1162 as_id = log->slot >> 16;
1163 id = (u16)log->slot;
1164 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1167 slots = __kvm_memslots(kvm, as_id);
1168 memslot = id_to_memslot(slots, id);
1170 dirty_bitmap = memslot->dirty_bitmap;
1175 n = kvm_dirty_bitmap_bytes(memslot);
1177 dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long);
1178 memset(dirty_bitmap_buffer, 0, n);
1180 spin_lock(&kvm->mmu_lock);
1182 for (i = 0; i < n / sizeof(long); i++) {
1186 if (!dirty_bitmap[i])
1191 mask = xchg(&dirty_bitmap[i], 0);
1192 dirty_bitmap_buffer[i] = mask;
1195 offset = i * BITS_PER_LONG;
1196 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1201 spin_unlock(&kvm->mmu_lock);
1204 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1211 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect);
1214 bool kvm_largepages_enabled(void)
1216 return largepages_enabled;
1219 void kvm_disable_largepages(void)
1221 largepages_enabled = false;
1223 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1225 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1227 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1229 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1231 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
1233 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn);
1236 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1238 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1240 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1241 memslot->flags & KVM_MEMSLOT_INVALID)
1246 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1248 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1250 struct vm_area_struct *vma;
1251 unsigned long addr, size;
1255 addr = gfn_to_hva(kvm, gfn);
1256 if (kvm_is_error_hva(addr))
1259 down_read(¤t->mm->mmap_sem);
1260 vma = find_vma(current->mm, addr);
1264 size = vma_kernel_pagesize(vma);
1267 up_read(¤t->mm->mmap_sem);
1272 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1274 return slot->flags & KVM_MEM_READONLY;
1277 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1278 gfn_t *nr_pages, bool write)
1280 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1281 return KVM_HVA_ERR_BAD;
1283 if (memslot_is_readonly(slot) && write)
1284 return KVM_HVA_ERR_RO_BAD;
1287 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1289 return __gfn_to_hva_memslot(slot, gfn);
1292 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1295 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1298 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1301 return gfn_to_hva_many(slot, gfn, NULL);
1303 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1305 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1307 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1309 EXPORT_SYMBOL_GPL(gfn_to_hva);
1311 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
1313 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
1315 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
1318 * If writable is set to false, the hva returned by this function is only
1319 * allowed to be read.
1321 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1322 gfn_t gfn, bool *writable)
1324 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1326 if (!kvm_is_error_hva(hva) && writable)
1327 *writable = !memslot_is_readonly(slot);
1332 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1334 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1336 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1339 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
1341 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1343 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1346 static int get_user_page_nowait(unsigned long start, int write,
1349 int flags = FOLL_NOWAIT | FOLL_HWPOISON;
1352 flags |= FOLL_WRITE;
1354 return get_user_pages(start, 1, flags, page, NULL);
1357 static inline int check_user_page_hwpoison(unsigned long addr)
1359 int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
1361 rc = get_user_pages(addr, 1, flags, NULL, NULL);
1362 return rc == -EHWPOISON;
1366 * The atomic path to get the writable pfn which will be stored in @pfn,
1367 * true indicates success, otherwise false is returned.
1369 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1370 bool write_fault, bool *writable, kvm_pfn_t *pfn)
1372 struct page *page[1];
1375 if (!(async || atomic))
1379 * Fast pin a writable pfn only if it is a write fault request
1380 * or the caller allows to map a writable pfn for a read fault
1383 if (!(write_fault || writable))
1386 npages = __get_user_pages_fast(addr, 1, 1, page);
1388 *pfn = page_to_pfn(page[0]);
1399 * The slow path to get the pfn of the specified host virtual address,
1400 * 1 indicates success, -errno is returned if error is detected.
1402 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1403 bool *writable, kvm_pfn_t *pfn)
1405 struct page *page[1];
1411 *writable = write_fault;
1414 down_read(¤t->mm->mmap_sem);
1415 npages = get_user_page_nowait(addr, write_fault, page);
1416 up_read(¤t->mm->mmap_sem);
1418 unsigned int flags = FOLL_TOUCH | FOLL_HWPOISON;
1421 flags |= FOLL_WRITE;
1423 npages = __get_user_pages_unlocked(current, current->mm, addr, 1,
1429 /* map read fault as writable if possible */
1430 if (unlikely(!write_fault) && writable) {
1431 struct page *wpage[1];
1433 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1442 *pfn = page_to_pfn(page[0]);
1446 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1448 if (unlikely(!(vma->vm_flags & VM_READ)))
1451 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1457 static int hva_to_pfn_remapped(struct vm_area_struct *vma,
1458 unsigned long addr, bool *async,
1459 bool write_fault, kvm_pfn_t *p_pfn)
1464 r = follow_pfn(vma, addr, &pfn);
1467 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
1468 * not call the fault handler, so do it here.
1470 bool unlocked = false;
1471 r = fixup_user_fault(current, current->mm, addr,
1472 (write_fault ? FAULT_FLAG_WRITE : 0),
1479 r = follow_pfn(vma, addr, &pfn);
1487 * Get a reference here because callers of *hva_to_pfn* and
1488 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
1489 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
1490 * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will
1491 * simply do nothing for reserved pfns.
1493 * Whoever called remap_pfn_range is also going to call e.g.
1494 * unmap_mapping_range before the underlying pages are freed,
1495 * causing a call to our MMU notifier.
1504 * Pin guest page in memory and return its pfn.
1505 * @addr: host virtual address which maps memory to the guest
1506 * @atomic: whether this function can sleep
1507 * @async: whether this function need to wait IO complete if the
1508 * host page is not in the memory
1509 * @write_fault: whether we should get a writable host page
1510 * @writable: whether it allows to map a writable host page for !@write_fault
1512 * The function will map a writable host page for these two cases:
1513 * 1): @write_fault = true
1514 * 2): @write_fault = false && @writable, @writable will tell the caller
1515 * whether the mapping is writable.
1517 static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1518 bool write_fault, bool *writable)
1520 struct vm_area_struct *vma;
1524 /* we can do it either atomically or asynchronously, not both */
1525 BUG_ON(atomic && async);
1527 if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1531 return KVM_PFN_ERR_FAULT;
1533 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1537 down_read(¤t->mm->mmap_sem);
1538 if (npages == -EHWPOISON ||
1539 (!async && check_user_page_hwpoison(addr))) {
1540 pfn = KVM_PFN_ERR_HWPOISON;
1545 vma = find_vma_intersection(current->mm, addr, addr + 1);
1548 pfn = KVM_PFN_ERR_FAULT;
1549 else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
1550 r = hva_to_pfn_remapped(vma, addr, async, write_fault, &pfn);
1554 pfn = KVM_PFN_ERR_FAULT;
1556 if (async && vma_is_valid(vma, write_fault))
1558 pfn = KVM_PFN_ERR_FAULT;
1561 up_read(¤t->mm->mmap_sem);
1565 kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn,
1566 bool atomic, bool *async, bool write_fault,
1569 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1571 if (addr == KVM_HVA_ERR_RO_BAD) {
1574 return KVM_PFN_ERR_RO_FAULT;
1577 if (kvm_is_error_hva(addr)) {
1580 return KVM_PFN_NOSLOT;
1583 /* Do not map writable pfn in the readonly memslot. */
1584 if (writable && memslot_is_readonly(slot)) {
1589 return hva_to_pfn(addr, atomic, async, write_fault,
1592 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
1594 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1597 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
1598 write_fault, writable);
1600 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1602 kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1604 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1606 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
1608 kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1610 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1612 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1614 kvm_pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1616 return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm, gfn), gfn);
1618 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1620 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
1622 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1624 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
1626 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1628 return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
1630 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1632 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
1634 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1636 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
1638 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1639 struct page **pages, int nr_pages)
1644 addr = gfn_to_hva_many(slot, gfn, &entry);
1645 if (kvm_is_error_hva(addr))
1648 if (entry < nr_pages)
1651 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1653 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1655 static struct page *kvm_pfn_to_page(kvm_pfn_t pfn)
1657 if (is_error_noslot_pfn(pfn))
1658 return KVM_ERR_PTR_BAD_PAGE;
1660 if (kvm_is_reserved_pfn(pfn)) {
1662 return KVM_ERR_PTR_BAD_PAGE;
1665 return pfn_to_page(pfn);
1668 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1672 pfn = gfn_to_pfn(kvm, gfn);
1674 return kvm_pfn_to_page(pfn);
1676 EXPORT_SYMBOL_GPL(gfn_to_page);
1678 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
1682 pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
1684 return kvm_pfn_to_page(pfn);
1686 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
1688 void kvm_release_page_clean(struct page *page)
1690 WARN_ON(is_error_page(page));
1692 kvm_release_pfn_clean(page_to_pfn(page));
1694 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1696 void kvm_release_pfn_clean(kvm_pfn_t pfn)
1698 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
1699 put_page(pfn_to_page(pfn));
1701 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1703 void kvm_release_page_dirty(struct page *page)
1705 WARN_ON(is_error_page(page));
1707 kvm_release_pfn_dirty(page_to_pfn(page));
1709 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1711 static void kvm_release_pfn_dirty(kvm_pfn_t pfn)
1713 kvm_set_pfn_dirty(pfn);
1714 kvm_release_pfn_clean(pfn);
1717 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
1719 if (!kvm_is_reserved_pfn(pfn)) {
1720 struct page *page = pfn_to_page(pfn);
1722 if (!PageReserved(page))
1726 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1728 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
1730 if (!kvm_is_reserved_pfn(pfn))
1731 mark_page_accessed(pfn_to_page(pfn));
1733 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1735 void kvm_get_pfn(kvm_pfn_t pfn)
1737 if (!kvm_is_reserved_pfn(pfn))
1738 get_page(pfn_to_page(pfn));
1740 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1742 static int next_segment(unsigned long len, int offset)
1744 if (len > PAGE_SIZE - offset)
1745 return PAGE_SIZE - offset;
1750 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
1751 void *data, int offset, int len)
1756 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1757 if (kvm_is_error_hva(addr))
1759 r = __copy_from_user(data, (void __user *)addr + offset, len);
1765 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1768 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1770 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1772 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1774 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
1775 int offset, int len)
1777 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1779 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1781 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
1783 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1785 gfn_t gfn = gpa >> PAGE_SHIFT;
1787 int offset = offset_in_page(gpa);
1790 while ((seg = next_segment(len, offset)) != 0) {
1791 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1801 EXPORT_SYMBOL_GPL(kvm_read_guest);
1803 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
1805 gfn_t gfn = gpa >> PAGE_SHIFT;
1807 int offset = offset_in_page(gpa);
1810 while ((seg = next_segment(len, offset)) != 0) {
1811 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
1821 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
1823 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1824 void *data, int offset, unsigned long len)
1829 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1830 if (kvm_is_error_hva(addr))
1832 pagefault_disable();
1833 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1840 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1843 gfn_t gfn = gpa >> PAGE_SHIFT;
1844 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1845 int offset = offset_in_page(gpa);
1847 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1849 EXPORT_SYMBOL_GPL(kvm_read_guest_atomic);
1851 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
1852 void *data, unsigned long len)
1854 gfn_t gfn = gpa >> PAGE_SHIFT;
1855 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1856 int offset = offset_in_page(gpa);
1858 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1860 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
1862 static int __kvm_write_guest_page(struct kvm_memory_slot *memslot, gfn_t gfn,
1863 const void *data, int offset, int len)
1868 addr = gfn_to_hva_memslot(memslot, gfn);
1869 if (kvm_is_error_hva(addr))
1871 r = __copy_to_user((void __user *)addr + offset, data, len);
1874 mark_page_dirty_in_slot(memslot, gfn);
1878 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
1879 const void *data, int offset, int len)
1881 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1883 return __kvm_write_guest_page(slot, gfn, data, offset, len);
1885 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1887 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
1888 const void *data, int offset, int len)
1890 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1892 return __kvm_write_guest_page(slot, gfn, data, offset, len);
1894 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
1896 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1899 gfn_t gfn = gpa >> PAGE_SHIFT;
1901 int offset = offset_in_page(gpa);
1904 while ((seg = next_segment(len, offset)) != 0) {
1905 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1915 EXPORT_SYMBOL_GPL(kvm_write_guest);
1917 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
1920 gfn_t gfn = gpa >> PAGE_SHIFT;
1922 int offset = offset_in_page(gpa);
1925 while ((seg = next_segment(len, offset)) != 0) {
1926 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
1936 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
1938 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1939 gpa_t gpa, unsigned long len)
1941 struct kvm_memslots *slots = kvm_memslots(kvm);
1942 int offset = offset_in_page(gpa);
1943 gfn_t start_gfn = gpa >> PAGE_SHIFT;
1944 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1945 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1946 gfn_t nr_pages_avail;
1949 ghc->generation = slots->generation;
1951 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1952 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, NULL);
1953 if (!kvm_is_error_hva(ghc->hva) && nr_pages_needed <= 1) {
1957 * If the requested region crosses two memslots, we still
1958 * verify that the entire region is valid here.
1960 while (start_gfn <= end_gfn) {
1961 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1962 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1964 if (kvm_is_error_hva(ghc->hva))
1966 start_gfn += nr_pages_avail;
1968 /* Use the slow path for cross page reads and writes. */
1969 ghc->memslot = NULL;
1973 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1975 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1976 void *data, int offset, unsigned long len)
1978 struct kvm_memslots *slots = kvm_memslots(kvm);
1980 gpa_t gpa = ghc->gpa + offset;
1982 BUG_ON(len + offset > ghc->len);
1984 if (slots->generation != ghc->generation)
1985 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1987 if (unlikely(!ghc->memslot))
1988 return kvm_write_guest(kvm, gpa, data, len);
1990 if (kvm_is_error_hva(ghc->hva))
1993 r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
1996 mark_page_dirty_in_slot(ghc->memslot, gpa >> PAGE_SHIFT);
2000 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached);
2002 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2003 void *data, unsigned long len)
2005 return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len);
2007 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
2009 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2010 void *data, unsigned long len)
2012 struct kvm_memslots *slots = kvm_memslots(kvm);
2015 BUG_ON(len > ghc->len);
2017 if (slots->generation != ghc->generation)
2018 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
2020 if (unlikely(!ghc->memslot))
2021 return kvm_read_guest(kvm, ghc->gpa, data, len);
2023 if (kvm_is_error_hva(ghc->hva))
2026 r = __copy_from_user(data, (void __user *)ghc->hva, len);
2032 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
2034 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
2036 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
2038 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
2040 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
2042 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
2044 gfn_t gfn = gpa >> PAGE_SHIFT;
2046 int offset = offset_in_page(gpa);
2049 while ((seg = next_segment(len, offset)) != 0) {
2050 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
2059 EXPORT_SYMBOL_GPL(kvm_clear_guest);
2061 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot,
2064 if (memslot && memslot->dirty_bitmap) {
2065 unsigned long rel_gfn = gfn - memslot->base_gfn;
2067 set_bit_le(rel_gfn, memslot->dirty_bitmap);
2071 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
2073 struct kvm_memory_slot *memslot;
2075 memslot = gfn_to_memslot(kvm, gfn);
2076 mark_page_dirty_in_slot(memslot, gfn);
2078 EXPORT_SYMBOL_GPL(mark_page_dirty);
2080 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
2082 struct kvm_memory_slot *memslot;
2084 memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2085 mark_page_dirty_in_slot(memslot, gfn);
2087 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
2089 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
2091 unsigned int old, val, grow;
2093 old = val = vcpu->halt_poll_ns;
2094 grow = READ_ONCE(halt_poll_ns_grow);
2096 if (val == 0 && grow)
2101 if (val > halt_poll_ns)
2104 vcpu->halt_poll_ns = val;
2105 trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
2108 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
2110 unsigned int old, val, shrink;
2112 old = val = vcpu->halt_poll_ns;
2113 shrink = READ_ONCE(halt_poll_ns_shrink);
2119 vcpu->halt_poll_ns = val;
2120 trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
2123 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
2125 if (kvm_arch_vcpu_runnable(vcpu)) {
2126 kvm_make_request(KVM_REQ_UNHALT, vcpu);
2129 if (kvm_cpu_has_pending_timer(vcpu))
2131 if (signal_pending(current))
2138 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2140 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
2143 DECLARE_SWAITQUEUE(wait);
2144 bool waited = false;
2147 start = cur = ktime_get();
2148 if (vcpu->halt_poll_ns) {
2149 ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns);
2151 ++vcpu->stat.halt_attempted_poll;
2154 * This sets KVM_REQ_UNHALT if an interrupt
2157 if (kvm_vcpu_check_block(vcpu) < 0) {
2158 ++vcpu->stat.halt_successful_poll;
2159 if (!vcpu_valid_wakeup(vcpu))
2160 ++vcpu->stat.halt_poll_invalid;
2164 } while (single_task_running() && ktime_before(cur, stop));
2167 kvm_arch_vcpu_blocking(vcpu);
2170 prepare_to_swait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
2172 if (kvm_vcpu_check_block(vcpu) < 0)
2179 finish_swait(&vcpu->wq, &wait);
2182 kvm_arch_vcpu_unblocking(vcpu);
2184 block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
2186 if (!vcpu_valid_wakeup(vcpu))
2187 shrink_halt_poll_ns(vcpu);
2188 else if (halt_poll_ns) {
2189 if (block_ns <= vcpu->halt_poll_ns)
2191 /* we had a long block, shrink polling */
2192 else if (vcpu->halt_poll_ns && block_ns > halt_poll_ns)
2193 shrink_halt_poll_ns(vcpu);
2194 /* we had a short halt and our poll time is too small */
2195 else if (vcpu->halt_poll_ns < halt_poll_ns &&
2196 block_ns < halt_poll_ns)
2197 grow_halt_poll_ns(vcpu);
2199 vcpu->halt_poll_ns = 0;
2201 trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu));
2202 kvm_arch_vcpu_block_finish(vcpu);
2204 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
2207 void kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
2209 struct swait_queue_head *wqp;
2211 wqp = kvm_arch_vcpu_wq(vcpu);
2212 if (swait_active(wqp)) {
2214 ++vcpu->stat.halt_wakeup;
2218 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
2221 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2223 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
2226 int cpu = vcpu->cpu;
2228 kvm_vcpu_wake_up(vcpu);
2230 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
2231 if (kvm_arch_vcpu_should_kick(vcpu))
2232 smp_send_reschedule(cpu);
2235 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
2236 #endif /* !CONFIG_S390 */
2238 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
2241 struct task_struct *task = NULL;
2245 pid = rcu_dereference(target->pid);
2247 task = get_pid_task(pid, PIDTYPE_PID);
2251 ret = yield_to(task, 1);
2252 put_task_struct(task);
2256 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
2259 * Helper that checks whether a VCPU is eligible for directed yield.
2260 * Most eligible candidate to yield is decided by following heuristics:
2262 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2263 * (preempted lock holder), indicated by @in_spin_loop.
2264 * Set at the beiginning and cleared at the end of interception/PLE handler.
2266 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2267 * chance last time (mostly it has become eligible now since we have probably
2268 * yielded to lockholder in last iteration. This is done by toggling
2269 * @dy_eligible each time a VCPU checked for eligibility.)
2271 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2272 * to preempted lock-holder could result in wrong VCPU selection and CPU
2273 * burning. Giving priority for a potential lock-holder increases lock
2276 * Since algorithm is based on heuristics, accessing another VCPU data without
2277 * locking does not harm. It may result in trying to yield to same VCPU, fail
2278 * and continue with next VCPU and so on.
2280 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
2282 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2285 eligible = !vcpu->spin_loop.in_spin_loop ||
2286 vcpu->spin_loop.dy_eligible;
2288 if (vcpu->spin_loop.in_spin_loop)
2289 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
2297 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
2299 struct kvm *kvm = me->kvm;
2300 struct kvm_vcpu *vcpu;
2301 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
2307 kvm_vcpu_set_in_spin_loop(me, true);
2309 * We boost the priority of a VCPU that is runnable but not
2310 * currently running, because it got preempted by something
2311 * else and called schedule in __vcpu_run. Hopefully that
2312 * VCPU is holding the lock that we need and will release it.
2313 * We approximate round-robin by starting at the last boosted VCPU.
2315 for (pass = 0; pass < 2 && !yielded && try; pass++) {
2316 kvm_for_each_vcpu(i, vcpu, kvm) {
2317 if (!pass && i <= last_boosted_vcpu) {
2318 i = last_boosted_vcpu;
2320 } else if (pass && i > last_boosted_vcpu)
2322 if (!ACCESS_ONCE(vcpu->preempted))
2326 if (swait_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
2328 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
2331 yielded = kvm_vcpu_yield_to(vcpu);
2333 kvm->last_boosted_vcpu = i;
2335 } else if (yielded < 0) {
2342 kvm_vcpu_set_in_spin_loop(me, false);
2344 /* Ensure vcpu is not eligible during next spinloop */
2345 kvm_vcpu_set_dy_eligible(me, false);
2347 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
2349 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2351 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
2354 if (vmf->pgoff == 0)
2355 page = virt_to_page(vcpu->run);
2357 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
2358 page = virt_to_page(vcpu->arch.pio_data);
2360 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2361 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
2362 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
2365 return kvm_arch_vcpu_fault(vcpu, vmf);
2371 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
2372 .fault = kvm_vcpu_fault,
2375 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2377 vma->vm_ops = &kvm_vcpu_vm_ops;
2381 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2383 struct kvm_vcpu *vcpu = filp->private_data;
2385 debugfs_remove_recursive(vcpu->debugfs_dentry);
2386 kvm_put_kvm(vcpu->kvm);
2390 static struct file_operations kvm_vcpu_fops = {
2391 .release = kvm_vcpu_release,
2392 .unlocked_ioctl = kvm_vcpu_ioctl,
2393 #ifdef CONFIG_KVM_COMPAT
2394 .compat_ioctl = kvm_vcpu_compat_ioctl,
2396 .mmap = kvm_vcpu_mmap,
2397 .llseek = noop_llseek,
2401 * Allocates an inode for the vcpu.
2403 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2405 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
2408 static int kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
2410 char dir_name[ITOA_MAX_LEN * 2];
2413 if (!kvm_arch_has_vcpu_debugfs())
2416 if (!debugfs_initialized())
2419 snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
2420 vcpu->debugfs_dentry = debugfs_create_dir(dir_name,
2421 vcpu->kvm->debugfs_dentry);
2422 if (!vcpu->debugfs_dentry)
2425 ret = kvm_arch_create_vcpu_debugfs(vcpu);
2427 debugfs_remove_recursive(vcpu->debugfs_dentry);
2435 * Creates some virtual cpus. Good luck creating more than one.
2437 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
2440 struct kvm_vcpu *vcpu;
2442 if (id >= KVM_MAX_VCPU_ID)
2445 mutex_lock(&kvm->lock);
2446 if (kvm->created_vcpus == KVM_MAX_VCPUS) {
2447 mutex_unlock(&kvm->lock);
2451 kvm->created_vcpus++;
2452 mutex_unlock(&kvm->lock);
2454 vcpu = kvm_arch_vcpu_create(kvm, id);
2457 goto vcpu_decrement;
2460 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2462 r = kvm_arch_vcpu_setup(vcpu);
2466 r = kvm_create_vcpu_debugfs(vcpu);
2470 mutex_lock(&kvm->lock);
2471 if (kvm_get_vcpu_by_id(kvm, id)) {
2473 goto unlock_vcpu_destroy;
2476 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
2478 /* Now it's all set up, let userspace reach it */
2480 r = create_vcpu_fd(vcpu);
2483 goto unlock_vcpu_destroy;
2486 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
2489 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
2490 * before kvm->online_vcpu's incremented value.
2493 atomic_inc(&kvm->online_vcpus);
2495 mutex_unlock(&kvm->lock);
2496 kvm_arch_vcpu_postcreate(vcpu);
2499 unlock_vcpu_destroy:
2500 mutex_unlock(&kvm->lock);
2501 debugfs_remove_recursive(vcpu->debugfs_dentry);
2503 kvm_arch_vcpu_destroy(vcpu);
2505 mutex_lock(&kvm->lock);
2506 kvm->created_vcpus--;
2507 mutex_unlock(&kvm->lock);
2511 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2514 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2515 vcpu->sigset_active = 1;
2516 vcpu->sigset = *sigset;
2518 vcpu->sigset_active = 0;
2522 static long kvm_vcpu_ioctl(struct file *filp,
2523 unsigned int ioctl, unsigned long arg)
2525 struct kvm_vcpu *vcpu = filp->private_data;
2526 void __user *argp = (void __user *)arg;
2528 struct kvm_fpu *fpu = NULL;
2529 struct kvm_sregs *kvm_sregs = NULL;
2531 if (vcpu->kvm->mm != current->mm)
2534 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2537 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
2539 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
2540 * so vcpu_load() would break it.
2542 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_S390_IRQ || ioctl == KVM_INTERRUPT)
2543 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2547 r = vcpu_load(vcpu);
2555 if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
2556 /* The thread running this VCPU changed. */
2557 struct pid *oldpid = vcpu->pid;
2558 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
2560 rcu_assign_pointer(vcpu->pid, newpid);
2565 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2566 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2568 case KVM_GET_REGS: {
2569 struct kvm_regs *kvm_regs;
2572 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2575 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2579 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2586 case KVM_SET_REGS: {
2587 struct kvm_regs *kvm_regs;
2590 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2591 if (IS_ERR(kvm_regs)) {
2592 r = PTR_ERR(kvm_regs);
2595 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2599 case KVM_GET_SREGS: {
2600 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2604 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2608 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2613 case KVM_SET_SREGS: {
2614 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2615 if (IS_ERR(kvm_sregs)) {
2616 r = PTR_ERR(kvm_sregs);
2620 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2623 case KVM_GET_MP_STATE: {
2624 struct kvm_mp_state mp_state;
2626 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2630 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
2635 case KVM_SET_MP_STATE: {
2636 struct kvm_mp_state mp_state;
2639 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
2641 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2644 case KVM_TRANSLATE: {
2645 struct kvm_translation tr;
2648 if (copy_from_user(&tr, argp, sizeof(tr)))
2650 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2654 if (copy_to_user(argp, &tr, sizeof(tr)))
2659 case KVM_SET_GUEST_DEBUG: {
2660 struct kvm_guest_debug dbg;
2663 if (copy_from_user(&dbg, argp, sizeof(dbg)))
2665 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2668 case KVM_SET_SIGNAL_MASK: {
2669 struct kvm_signal_mask __user *sigmask_arg = argp;
2670 struct kvm_signal_mask kvm_sigmask;
2671 sigset_t sigset, *p;
2676 if (copy_from_user(&kvm_sigmask, argp,
2677 sizeof(kvm_sigmask)))
2680 if (kvm_sigmask.len != sizeof(sigset))
2683 if (copy_from_user(&sigset, sigmask_arg->sigset,
2688 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2692 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2696 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2700 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2706 fpu = memdup_user(argp, sizeof(*fpu));
2712 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2716 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2725 #ifdef CONFIG_KVM_COMPAT
2726 static long kvm_vcpu_compat_ioctl(struct file *filp,
2727 unsigned int ioctl, unsigned long arg)
2729 struct kvm_vcpu *vcpu = filp->private_data;
2730 void __user *argp = compat_ptr(arg);
2733 if (vcpu->kvm->mm != current->mm)
2737 case KVM_SET_SIGNAL_MASK: {
2738 struct kvm_signal_mask __user *sigmask_arg = argp;
2739 struct kvm_signal_mask kvm_sigmask;
2740 compat_sigset_t csigset;
2745 if (copy_from_user(&kvm_sigmask, argp,
2746 sizeof(kvm_sigmask)))
2749 if (kvm_sigmask.len != sizeof(csigset))
2752 if (copy_from_user(&csigset, sigmask_arg->sigset,
2755 sigset_from_compat(&sigset, &csigset);
2756 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2758 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2762 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2770 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2771 int (*accessor)(struct kvm_device *dev,
2772 struct kvm_device_attr *attr),
2775 struct kvm_device_attr attr;
2780 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2783 return accessor(dev, &attr);
2786 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2789 struct kvm_device *dev = filp->private_data;
2792 case KVM_SET_DEVICE_ATTR:
2793 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2794 case KVM_GET_DEVICE_ATTR:
2795 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2796 case KVM_HAS_DEVICE_ATTR:
2797 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2799 if (dev->ops->ioctl)
2800 return dev->ops->ioctl(dev, ioctl, arg);
2806 static int kvm_device_release(struct inode *inode, struct file *filp)
2808 struct kvm_device *dev = filp->private_data;
2809 struct kvm *kvm = dev->kvm;
2815 static const struct file_operations kvm_device_fops = {
2816 .unlocked_ioctl = kvm_device_ioctl,
2817 #ifdef CONFIG_KVM_COMPAT
2818 .compat_ioctl = kvm_device_ioctl,
2820 .release = kvm_device_release,
2823 struct kvm_device *kvm_device_from_filp(struct file *filp)
2825 if (filp->f_op != &kvm_device_fops)
2828 return filp->private_data;
2831 static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
2832 #ifdef CONFIG_KVM_MPIC
2833 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
2834 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
2837 #ifdef CONFIG_KVM_XICS
2838 [KVM_DEV_TYPE_XICS] = &kvm_xics_ops,
2842 int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
2844 if (type >= ARRAY_SIZE(kvm_device_ops_table))
2847 if (kvm_device_ops_table[type] != NULL)
2850 kvm_device_ops_table[type] = ops;
2854 void kvm_unregister_device_ops(u32 type)
2856 if (kvm_device_ops_table[type] != NULL)
2857 kvm_device_ops_table[type] = NULL;
2860 static int kvm_ioctl_create_device(struct kvm *kvm,
2861 struct kvm_create_device *cd)
2863 struct kvm_device_ops *ops = NULL;
2864 struct kvm_device *dev;
2865 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2868 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
2871 ops = kvm_device_ops_table[cd->type];
2878 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2885 mutex_lock(&kvm->lock);
2886 ret = ops->create(dev, cd->type);
2888 mutex_unlock(&kvm->lock);
2892 list_add(&dev->vm_node, &kvm->devices);
2893 mutex_unlock(&kvm->lock);
2898 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2900 mutex_lock(&kvm->lock);
2901 list_del(&dev->vm_node);
2902 mutex_unlock(&kvm->lock);
2912 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2915 case KVM_CAP_USER_MEMORY:
2916 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2917 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2918 case KVM_CAP_INTERNAL_ERROR_DATA:
2919 #ifdef CONFIG_HAVE_KVM_MSI
2920 case KVM_CAP_SIGNAL_MSI:
2922 #ifdef CONFIG_HAVE_KVM_IRQFD
2924 case KVM_CAP_IRQFD_RESAMPLE:
2926 case KVM_CAP_IOEVENTFD_ANY_LENGTH:
2927 case KVM_CAP_CHECK_EXTENSION_VM:
2929 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2930 case KVM_CAP_IRQ_ROUTING:
2931 return KVM_MAX_IRQ_ROUTES;
2933 #if KVM_ADDRESS_SPACE_NUM > 1
2934 case KVM_CAP_MULTI_ADDRESS_SPACE:
2935 return KVM_ADDRESS_SPACE_NUM;
2937 case KVM_CAP_MAX_VCPU_ID:
2938 return KVM_MAX_VCPU_ID;
2942 return kvm_vm_ioctl_check_extension(kvm, arg);
2945 static long kvm_vm_ioctl(struct file *filp,
2946 unsigned int ioctl, unsigned long arg)
2948 struct kvm *kvm = filp->private_data;
2949 void __user *argp = (void __user *)arg;
2952 if (kvm->mm != current->mm)
2955 case KVM_CREATE_VCPU:
2956 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2958 case KVM_SET_USER_MEMORY_REGION: {
2959 struct kvm_userspace_memory_region kvm_userspace_mem;
2962 if (copy_from_user(&kvm_userspace_mem, argp,
2963 sizeof(kvm_userspace_mem)))
2966 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2969 case KVM_GET_DIRTY_LOG: {
2970 struct kvm_dirty_log log;
2973 if (copy_from_user(&log, argp, sizeof(log)))
2975 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2978 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2979 case KVM_REGISTER_COALESCED_MMIO: {
2980 struct kvm_coalesced_mmio_zone zone;
2983 if (copy_from_user(&zone, argp, sizeof(zone)))
2985 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2988 case KVM_UNREGISTER_COALESCED_MMIO: {
2989 struct kvm_coalesced_mmio_zone zone;
2992 if (copy_from_user(&zone, argp, sizeof(zone)))
2994 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2999 struct kvm_irqfd data;
3002 if (copy_from_user(&data, argp, sizeof(data)))
3004 r = kvm_irqfd(kvm, &data);
3007 case KVM_IOEVENTFD: {
3008 struct kvm_ioeventfd data;
3011 if (copy_from_user(&data, argp, sizeof(data)))
3013 r = kvm_ioeventfd(kvm, &data);
3016 #ifdef CONFIG_HAVE_KVM_MSI
3017 case KVM_SIGNAL_MSI: {
3021 if (copy_from_user(&msi, argp, sizeof(msi)))
3023 r = kvm_send_userspace_msi(kvm, &msi);
3027 #ifdef __KVM_HAVE_IRQ_LINE
3028 case KVM_IRQ_LINE_STATUS:
3029 case KVM_IRQ_LINE: {
3030 struct kvm_irq_level irq_event;
3033 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
3036 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
3037 ioctl == KVM_IRQ_LINE_STATUS);
3042 if (ioctl == KVM_IRQ_LINE_STATUS) {
3043 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
3051 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3052 case KVM_SET_GSI_ROUTING: {
3053 struct kvm_irq_routing routing;
3054 struct kvm_irq_routing __user *urouting;
3055 struct kvm_irq_routing_entry *entries = NULL;
3058 if (copy_from_user(&routing, argp, sizeof(routing)))
3061 if (routing.nr > KVM_MAX_IRQ_ROUTES)
3067 entries = vmalloc(routing.nr * sizeof(*entries));
3072 if (copy_from_user(entries, urouting->entries,
3073 routing.nr * sizeof(*entries)))
3074 goto out_free_irq_routing;
3076 r = kvm_set_irq_routing(kvm, entries, routing.nr,
3078 out_free_irq_routing:
3082 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
3083 case KVM_CREATE_DEVICE: {
3084 struct kvm_create_device cd;
3087 if (copy_from_user(&cd, argp, sizeof(cd)))
3090 r = kvm_ioctl_create_device(kvm, &cd);
3095 if (copy_to_user(argp, &cd, sizeof(cd)))
3101 case KVM_CHECK_EXTENSION:
3102 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
3105 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
3111 #ifdef CONFIG_KVM_COMPAT
3112 struct compat_kvm_dirty_log {
3116 compat_uptr_t dirty_bitmap; /* one bit per page */
3121 static long kvm_vm_compat_ioctl(struct file *filp,
3122 unsigned int ioctl, unsigned long arg)
3124 struct kvm *kvm = filp->private_data;
3127 if (kvm->mm != current->mm)
3130 case KVM_GET_DIRTY_LOG: {
3131 struct compat_kvm_dirty_log compat_log;
3132 struct kvm_dirty_log log;
3135 if (copy_from_user(&compat_log, (void __user *)arg,
3136 sizeof(compat_log)))
3138 log.slot = compat_log.slot;
3139 log.padding1 = compat_log.padding1;
3140 log.padding2 = compat_log.padding2;
3141 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
3143 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3147 r = kvm_vm_ioctl(filp, ioctl, arg);
3155 static struct file_operations kvm_vm_fops = {
3156 .release = kvm_vm_release,
3157 .unlocked_ioctl = kvm_vm_ioctl,
3158 #ifdef CONFIG_KVM_COMPAT
3159 .compat_ioctl = kvm_vm_compat_ioctl,
3161 .llseek = noop_llseek,
3164 static int kvm_dev_ioctl_create_vm(unsigned long type)
3170 kvm = kvm_create_vm(type);
3172 return PTR_ERR(kvm);
3173 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
3174 r = kvm_coalesced_mmio_init(kvm);
3180 r = get_unused_fd_flags(O_CLOEXEC);
3185 file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
3189 return PTR_ERR(file);
3192 if (kvm_create_vm_debugfs(kvm, r) < 0) {
3198 fd_install(r, file);
3202 static long kvm_dev_ioctl(struct file *filp,
3203 unsigned int ioctl, unsigned long arg)
3208 case KVM_GET_API_VERSION:
3211 r = KVM_API_VERSION;
3214 r = kvm_dev_ioctl_create_vm(arg);
3216 case KVM_CHECK_EXTENSION:
3217 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
3219 case KVM_GET_VCPU_MMAP_SIZE:
3222 r = PAGE_SIZE; /* struct kvm_run */
3224 r += PAGE_SIZE; /* pio data page */
3226 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
3227 r += PAGE_SIZE; /* coalesced mmio ring page */
3230 case KVM_TRACE_ENABLE:
3231 case KVM_TRACE_PAUSE:
3232 case KVM_TRACE_DISABLE:
3236 return kvm_arch_dev_ioctl(filp, ioctl, arg);
3242 static struct file_operations kvm_chardev_ops = {
3243 .unlocked_ioctl = kvm_dev_ioctl,
3244 .compat_ioctl = kvm_dev_ioctl,
3245 .llseek = noop_llseek,
3248 static struct miscdevice kvm_dev = {
3254 static void hardware_enable_nolock(void *junk)
3256 int cpu = raw_smp_processor_id();
3259 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
3262 cpumask_set_cpu(cpu, cpus_hardware_enabled);
3264 r = kvm_arch_hardware_enable();
3267 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3268 atomic_inc(&hardware_enable_failed);
3269 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
3273 static int kvm_starting_cpu(unsigned int cpu)
3275 raw_spin_lock(&kvm_count_lock);
3276 if (kvm_usage_count)
3277 hardware_enable_nolock(NULL);
3278 raw_spin_unlock(&kvm_count_lock);
3282 static void hardware_disable_nolock(void *junk)
3284 int cpu = raw_smp_processor_id();
3286 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
3288 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3289 kvm_arch_hardware_disable();
3292 static int kvm_dying_cpu(unsigned int cpu)
3294 raw_spin_lock(&kvm_count_lock);
3295 if (kvm_usage_count)
3296 hardware_disable_nolock(NULL);
3297 raw_spin_unlock(&kvm_count_lock);
3301 static void hardware_disable_all_nolock(void)
3303 BUG_ON(!kvm_usage_count);
3306 if (!kvm_usage_count)
3307 on_each_cpu(hardware_disable_nolock, NULL, 1);
3310 static void hardware_disable_all(void)
3312 raw_spin_lock(&kvm_count_lock);
3313 hardware_disable_all_nolock();
3314 raw_spin_unlock(&kvm_count_lock);
3317 static int hardware_enable_all(void)
3321 raw_spin_lock(&kvm_count_lock);
3324 if (kvm_usage_count == 1) {
3325 atomic_set(&hardware_enable_failed, 0);
3326 on_each_cpu(hardware_enable_nolock, NULL, 1);
3328 if (atomic_read(&hardware_enable_failed)) {
3329 hardware_disable_all_nolock();
3334 raw_spin_unlock(&kvm_count_lock);
3339 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3343 * Some (well, at least mine) BIOSes hang on reboot if
3346 * And Intel TXT required VMX off for all cpu when system shutdown.
3348 pr_info("kvm: exiting hardware virtualization\n");
3349 kvm_rebooting = true;
3350 on_each_cpu(hardware_disable_nolock, NULL, 1);
3354 static struct notifier_block kvm_reboot_notifier = {
3355 .notifier_call = kvm_reboot,
3359 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3363 for (i = 0; i < bus->dev_count; i++) {
3364 struct kvm_io_device *pos = bus->range[i].dev;
3366 kvm_iodevice_destructor(pos);
3371 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
3372 const struct kvm_io_range *r2)
3374 gpa_t addr1 = r1->addr;
3375 gpa_t addr2 = r2->addr;
3380 /* If r2->len == 0, match the exact address. If r2->len != 0,
3381 * accept any overlapping write. Any order is acceptable for
3382 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
3383 * we process all of them.
3396 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
3398 return kvm_io_bus_cmp(p1, p2);
3401 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
3402 gpa_t addr, int len)
3404 bus->range[bus->dev_count++] = (struct kvm_io_range) {
3410 sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
3411 kvm_io_bus_sort_cmp, NULL);
3416 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
3417 gpa_t addr, int len)
3419 struct kvm_io_range *range, key;
3422 key = (struct kvm_io_range) {
3427 range = bsearch(&key, bus->range, bus->dev_count,
3428 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
3432 off = range - bus->range;
3434 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
3440 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3441 struct kvm_io_range *range, const void *val)
3445 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3449 while (idx < bus->dev_count &&
3450 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3451 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
3460 /* kvm_io_bus_write - called under kvm->slots_lock */
3461 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3462 int len, const void *val)
3464 struct kvm_io_bus *bus;
3465 struct kvm_io_range range;
3468 range = (struct kvm_io_range) {
3473 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3474 r = __kvm_io_bus_write(vcpu, bus, &range, val);
3475 return r < 0 ? r : 0;
3478 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3479 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
3480 gpa_t addr, int len, const void *val, long cookie)
3482 struct kvm_io_bus *bus;
3483 struct kvm_io_range range;
3485 range = (struct kvm_io_range) {
3490 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3492 /* First try the device referenced by cookie. */
3493 if ((cookie >= 0) && (cookie < bus->dev_count) &&
3494 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
3495 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
3500 * cookie contained garbage; fall back to search and return the
3501 * correct cookie value.
3503 return __kvm_io_bus_write(vcpu, bus, &range, val);
3506 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3507 struct kvm_io_range *range, void *val)
3511 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3515 while (idx < bus->dev_count &&
3516 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3517 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
3525 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
3527 /* kvm_io_bus_read - called under kvm->slots_lock */
3528 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3531 struct kvm_io_bus *bus;
3532 struct kvm_io_range range;
3535 range = (struct kvm_io_range) {
3540 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];
3553 /* exclude ioeventfd which is limited by maximum fd */
3554 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3557 new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count + 1) *
3558 sizeof(struct kvm_io_range)), GFP_KERNEL);
3561 memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
3562 sizeof(struct kvm_io_range)));
3563 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
3564 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3565 synchronize_srcu_expedited(&kvm->srcu);
3571 /* Caller must hold slots_lock. */
3572 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3573 struct kvm_io_device *dev)
3576 struct kvm_io_bus *new_bus, *bus;
3578 bus = kvm->buses[bus_idx];
3580 for (i = 0; i < bus->dev_count; i++)
3581 if (bus->range[i].dev == dev) {
3589 new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3590 sizeof(struct kvm_io_range)), GFP_KERNEL);
3594 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3595 new_bus->dev_count--;
3596 memcpy(new_bus->range + i, bus->range + i + 1,
3597 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3599 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3600 synchronize_srcu_expedited(&kvm->srcu);
3605 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3608 struct kvm_io_bus *bus;
3609 int dev_idx, srcu_idx;
3610 struct kvm_io_device *iodev = NULL;
3612 srcu_idx = srcu_read_lock(&kvm->srcu);
3614 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
3616 dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
3620 iodev = bus->range[dev_idx].dev;
3623 srcu_read_unlock(&kvm->srcu, srcu_idx);
3627 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
3629 static int kvm_debugfs_open(struct inode *inode, struct file *file,
3630 int (*get)(void *, u64 *), int (*set)(void *, u64),
3633 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
3636 /* The debugfs files are a reference to the kvm struct which
3637 * is still valid when kvm_destroy_vm is called.
3638 * To avoid the race between open and the removal of the debugfs
3639 * directory we test against the users count.
3641 if (!atomic_add_unless(&stat_data->kvm->users_count, 1, 0))
3644 if (simple_attr_open(inode, file, get, set, fmt)) {
3645 kvm_put_kvm(stat_data->kvm);
3652 static int kvm_debugfs_release(struct inode *inode, struct file *file)
3654 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
3657 simple_attr_release(inode, file);
3658 kvm_put_kvm(stat_data->kvm);
3663 static int vm_stat_get_per_vm(void *data, u64 *val)
3665 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3667 *val = *(ulong *)((void *)stat_data->kvm + stat_data->offset);
3672 static int vm_stat_get_per_vm_open(struct inode *inode, struct file *file)
3674 __simple_attr_check_format("%llu\n", 0ull);
3675 return kvm_debugfs_open(inode, file, vm_stat_get_per_vm,
3679 static const struct file_operations vm_stat_get_per_vm_fops = {
3680 .owner = THIS_MODULE,
3681 .open = vm_stat_get_per_vm_open,
3682 .release = kvm_debugfs_release,
3683 .read = simple_attr_read,
3684 .write = simple_attr_write,
3685 .llseek = generic_file_llseek,
3688 static int vcpu_stat_get_per_vm(void *data, u64 *val)
3691 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3692 struct kvm_vcpu *vcpu;
3696 kvm_for_each_vcpu(i, vcpu, stat_data->kvm)
3697 *val += *(u64 *)((void *)vcpu + stat_data->offset);
3702 static int vcpu_stat_get_per_vm_open(struct inode *inode, struct file *file)
3704 __simple_attr_check_format("%llu\n", 0ull);
3705 return kvm_debugfs_open(inode, file, vcpu_stat_get_per_vm,
3709 static const struct file_operations vcpu_stat_get_per_vm_fops = {
3710 .owner = THIS_MODULE,
3711 .open = vcpu_stat_get_per_vm_open,
3712 .release = kvm_debugfs_release,
3713 .read = simple_attr_read,
3714 .write = simple_attr_write,
3715 .llseek = generic_file_llseek,
3718 static const struct file_operations *stat_fops_per_vm[] = {
3719 [KVM_STAT_VCPU] = &vcpu_stat_get_per_vm_fops,
3720 [KVM_STAT_VM] = &vm_stat_get_per_vm_fops,
3723 static int vm_stat_get(void *_offset, u64 *val)
3725 unsigned offset = (long)_offset;
3727 struct kvm_stat_data stat_tmp = {.offset = offset};
3731 spin_lock(&kvm_lock);
3732 list_for_each_entry(kvm, &vm_list, vm_list) {
3734 vm_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
3737 spin_unlock(&kvm_lock);
3741 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
3743 static int vcpu_stat_get(void *_offset, u64 *val)
3745 unsigned offset = (long)_offset;
3747 struct kvm_stat_data stat_tmp = {.offset = offset};
3751 spin_lock(&kvm_lock);
3752 list_for_each_entry(kvm, &vm_list, vm_list) {
3754 vcpu_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
3757 spin_unlock(&kvm_lock);
3761 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
3763 static const struct file_operations *stat_fops[] = {
3764 [KVM_STAT_VCPU] = &vcpu_stat_fops,
3765 [KVM_STAT_VM] = &vm_stat_fops,
3768 static int kvm_init_debug(void)
3771 struct kvm_stats_debugfs_item *p;
3773 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3774 if (kvm_debugfs_dir == NULL)
3777 kvm_debugfs_num_entries = 0;
3778 for (p = debugfs_entries; p->name; ++p, kvm_debugfs_num_entries++) {
3779 if (!debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
3780 (void *)(long)p->offset,
3781 stat_fops[p->kind]))
3788 debugfs_remove_recursive(kvm_debugfs_dir);
3793 static int kvm_suspend(void)
3795 if (kvm_usage_count)
3796 hardware_disable_nolock(NULL);
3800 static void kvm_resume(void)
3802 if (kvm_usage_count) {
3803 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3804 hardware_enable_nolock(NULL);
3808 static struct syscore_ops kvm_syscore_ops = {
3809 .suspend = kvm_suspend,
3810 .resume = kvm_resume,
3814 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3816 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3819 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3821 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3823 if (vcpu->preempted)
3824 vcpu->preempted = false;
3826 kvm_arch_sched_in(vcpu, cpu);
3828 kvm_arch_vcpu_load(vcpu, cpu);
3831 static void kvm_sched_out(struct preempt_notifier *pn,
3832 struct task_struct *next)
3834 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3836 if (current->state == TASK_RUNNING)
3837 vcpu->preempted = true;
3838 kvm_arch_vcpu_put(vcpu);
3841 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3842 struct module *module)
3847 r = kvm_arch_init(opaque);
3852 * kvm_arch_init makes sure there's at most one caller
3853 * for architectures that support multiple implementations,
3854 * like intel and amd on x86.
3855 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3856 * conflicts in case kvm is already setup for another implementation.
3858 r = kvm_irqfd_init();
3862 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
3867 r = kvm_arch_hardware_setup();
3871 for_each_online_cpu(cpu) {
3872 smp_call_function_single(cpu,
3873 kvm_arch_check_processor_compat,
3879 r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "AP_KVM_STARTING",
3880 kvm_starting_cpu, kvm_dying_cpu);
3883 register_reboot_notifier(&kvm_reboot_notifier);
3885 /* A kmem cache lets us meet the alignment requirements of fx_save. */
3887 vcpu_align = __alignof__(struct kvm_vcpu);
3888 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
3890 if (!kvm_vcpu_cache) {
3895 r = kvm_async_pf_init();
3899 kvm_chardev_ops.owner = module;
3900 kvm_vm_fops.owner = module;
3901 kvm_vcpu_fops.owner = module;
3903 r = misc_register(&kvm_dev);
3905 pr_err("kvm: misc device register failed\n");
3909 register_syscore_ops(&kvm_syscore_ops);
3911 kvm_preempt_ops.sched_in = kvm_sched_in;
3912 kvm_preempt_ops.sched_out = kvm_sched_out;
3914 r = kvm_init_debug();
3916 pr_err("kvm: create debugfs files failed\n");
3920 r = kvm_vfio_ops_init();
3926 unregister_syscore_ops(&kvm_syscore_ops);
3927 misc_deregister(&kvm_dev);
3929 kvm_async_pf_deinit();
3931 kmem_cache_destroy(kvm_vcpu_cache);
3933 unregister_reboot_notifier(&kvm_reboot_notifier);
3934 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
3937 kvm_arch_hardware_unsetup();
3939 free_cpumask_var(cpus_hardware_enabled);
3947 EXPORT_SYMBOL_GPL(kvm_init);
3951 debugfs_remove_recursive(kvm_debugfs_dir);
3952 misc_deregister(&kvm_dev);
3953 kmem_cache_destroy(kvm_vcpu_cache);
3954 kvm_async_pf_deinit();
3955 unregister_syscore_ops(&kvm_syscore_ops);
3956 unregister_reboot_notifier(&kvm_reboot_notifier);
3957 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
3958 on_each_cpu(hardware_disable_nolock, NULL, 1);
3959 kvm_arch_hardware_unsetup();
3962 free_cpumask_var(cpus_hardware_enabled);
3963 kvm_vfio_ops_exit();
3965 EXPORT_SYMBOL_GPL(kvm_exit);