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/sched/mm.h>
37 #include <linux/cpumask.h>
38 #include <linux/smp.h>
39 #include <linux/anon_inodes.h>
40 #include <linux/profile.h>
41 #include <linux/kvm_para.h>
42 #include <linux/pagemap.h>
43 #include <linux/mman.h>
44 #include <linux/swap.h>
45 #include <linux/bitops.h>
46 #include <linux/spinlock.h>
47 #include <linux/compat.h>
48 #include <linux/srcu.h>
49 #include <linux/hugetlb.h>
50 #include <linux/slab.h>
51 #include <linux/sort.h>
52 #include <linux/bsearch.h>
54 #include <asm/processor.h>
56 #include <asm/ioctl.h>
57 #include <linux/uaccess.h>
58 #include <asm/pgtable.h>
60 #include "coalesced_mmio.h"
64 #define CREATE_TRACE_POINTS
65 #include <trace/events/kvm.h>
67 /* Worst case buffer size needed for holding an integer. */
68 #define ITOA_MAX_LEN 12
70 MODULE_AUTHOR("Qumranet");
71 MODULE_LICENSE("GPL");
73 /* Architectures should define their poll value according to the halt latency */
74 unsigned int halt_poll_ns = KVM_HALT_POLL_NS_DEFAULT;
75 module_param(halt_poll_ns, uint, S_IRUGO | S_IWUSR);
76 EXPORT_SYMBOL_GPL(halt_poll_ns);
78 /* Default doubles per-vcpu halt_poll_ns. */
79 unsigned int halt_poll_ns_grow = 2;
80 module_param(halt_poll_ns_grow, uint, S_IRUGO | S_IWUSR);
81 EXPORT_SYMBOL_GPL(halt_poll_ns_grow);
83 /* Default resets per-vcpu halt_poll_ns . */
84 unsigned int halt_poll_ns_shrink;
85 module_param(halt_poll_ns_shrink, uint, S_IRUGO | S_IWUSR);
86 EXPORT_SYMBOL_GPL(halt_poll_ns_shrink);
91 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
94 DEFINE_SPINLOCK(kvm_lock);
95 static DEFINE_RAW_SPINLOCK(kvm_count_lock);
98 static cpumask_var_t cpus_hardware_enabled;
99 static int kvm_usage_count;
100 static atomic_t hardware_enable_failed;
102 struct kmem_cache *kvm_vcpu_cache;
103 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
105 static __read_mostly struct preempt_ops kvm_preempt_ops;
107 struct dentry *kvm_debugfs_dir;
108 EXPORT_SYMBOL_GPL(kvm_debugfs_dir);
110 static int kvm_debugfs_num_entries;
111 static const struct file_operations *stat_fops_per_vm[];
113 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
115 #ifdef CONFIG_KVM_COMPAT
116 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
119 static int hardware_enable_all(void);
120 static void hardware_disable_all(void);
122 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
124 static void kvm_release_pfn_dirty(kvm_pfn_t pfn);
125 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot, gfn_t gfn);
127 __visible bool kvm_rebooting;
128 EXPORT_SYMBOL_GPL(kvm_rebooting);
130 static bool largepages_enabled = true;
132 bool kvm_is_reserved_pfn(kvm_pfn_t pfn)
135 return PageReserved(pfn_to_page(pfn));
141 * Switches to specified vcpu, until a matching vcpu_put()
143 int vcpu_load(struct kvm_vcpu *vcpu)
147 if (mutex_lock_killable(&vcpu->mutex))
150 preempt_notifier_register(&vcpu->preempt_notifier);
151 kvm_arch_vcpu_load(vcpu, cpu);
155 EXPORT_SYMBOL_GPL(vcpu_load);
157 void vcpu_put(struct kvm_vcpu *vcpu)
160 kvm_arch_vcpu_put(vcpu);
161 preempt_notifier_unregister(&vcpu->preempt_notifier);
163 mutex_unlock(&vcpu->mutex);
165 EXPORT_SYMBOL_GPL(vcpu_put);
167 static void ack_flush(void *_completed)
171 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
176 struct kvm_vcpu *vcpu;
178 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
181 kvm_for_each_vcpu(i, vcpu, kvm) {
182 kvm_make_request(req, vcpu);
185 /* Set ->requests bit before we read ->mode. */
186 smp_mb__after_atomic();
188 if (cpus != NULL && cpu != -1 && cpu != me &&
189 kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
190 cpumask_set_cpu(cpu, cpus);
192 if (unlikely(cpus == NULL))
193 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
194 else if (!cpumask_empty(cpus))
195 smp_call_function_many(cpus, ack_flush, NULL, 1);
199 free_cpumask_var(cpus);
203 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
204 void kvm_flush_remote_tlbs(struct kvm *kvm)
207 * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in
208 * kvm_make_all_cpus_request.
210 long dirty_count = smp_load_acquire(&kvm->tlbs_dirty);
213 * We want to publish modifications to the page tables before reading
214 * mode. Pairs with a memory barrier in arch-specific code.
215 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
216 * and smp_mb in walk_shadow_page_lockless_begin/end.
217 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
219 * There is already an smp_mb__after_atomic() before
220 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
223 if (kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
224 ++kvm->stat.remote_tlb_flush;
225 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
227 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
230 void kvm_reload_remote_mmus(struct kvm *kvm)
232 kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
235 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
240 mutex_init(&vcpu->mutex);
245 init_swait_queue_head(&vcpu->wq);
246 kvm_async_pf_vcpu_init(vcpu);
249 INIT_LIST_HEAD(&vcpu->blocked_vcpu_list);
251 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
256 vcpu->run = page_address(page);
258 kvm_vcpu_set_in_spin_loop(vcpu, false);
259 kvm_vcpu_set_dy_eligible(vcpu, false);
260 vcpu->preempted = false;
262 r = kvm_arch_vcpu_init(vcpu);
268 free_page((unsigned long)vcpu->run);
272 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
274 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
277 kvm_arch_vcpu_uninit(vcpu);
278 free_page((unsigned long)vcpu->run);
280 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
282 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
283 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
285 return container_of(mn, struct kvm, mmu_notifier);
288 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
289 struct mm_struct *mm,
290 unsigned long address)
292 struct kvm *kvm = mmu_notifier_to_kvm(mn);
293 int need_tlb_flush, idx;
296 * When ->invalidate_page runs, the linux pte has been zapped
297 * already but the page is still allocated until
298 * ->invalidate_page returns. So if we increase the sequence
299 * here the kvm page fault will notice if the spte can't be
300 * established because the page is going to be freed. If
301 * instead the kvm page fault establishes the spte before
302 * ->invalidate_page runs, kvm_unmap_hva will release it
305 * The sequence increase only need to be seen at spin_unlock
306 * time, and not at spin_lock time.
308 * Increasing the sequence after the spin_unlock would be
309 * unsafe because the kvm page fault could then establish the
310 * pte after kvm_unmap_hva returned, without noticing the page
311 * is going to be freed.
313 idx = srcu_read_lock(&kvm->srcu);
314 spin_lock(&kvm->mmu_lock);
316 kvm->mmu_notifier_seq++;
317 need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
318 /* we've to flush the tlb before the pages can be freed */
320 kvm_flush_remote_tlbs(kvm);
322 spin_unlock(&kvm->mmu_lock);
324 kvm_arch_mmu_notifier_invalidate_page(kvm, address);
326 srcu_read_unlock(&kvm->srcu, idx);
329 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
330 struct mm_struct *mm,
331 unsigned long address,
334 struct kvm *kvm = mmu_notifier_to_kvm(mn);
337 idx = srcu_read_lock(&kvm->srcu);
338 spin_lock(&kvm->mmu_lock);
339 kvm->mmu_notifier_seq++;
340 kvm_set_spte_hva(kvm, address, pte);
341 spin_unlock(&kvm->mmu_lock);
342 srcu_read_unlock(&kvm->srcu, idx);
345 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
346 struct mm_struct *mm,
350 struct kvm *kvm = mmu_notifier_to_kvm(mn);
351 int need_tlb_flush = 0, idx;
353 idx = srcu_read_lock(&kvm->srcu);
354 spin_lock(&kvm->mmu_lock);
356 * The count increase must become visible at unlock time as no
357 * spte can be established without taking the mmu_lock and
358 * count is also read inside the mmu_lock critical section.
360 kvm->mmu_notifier_count++;
361 need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
362 need_tlb_flush |= kvm->tlbs_dirty;
363 /* we've to flush the tlb before the pages can be freed */
365 kvm_flush_remote_tlbs(kvm);
367 spin_unlock(&kvm->mmu_lock);
368 srcu_read_unlock(&kvm->srcu, idx);
371 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
372 struct mm_struct *mm,
376 struct kvm *kvm = mmu_notifier_to_kvm(mn);
378 spin_lock(&kvm->mmu_lock);
380 * This sequence increase will notify the kvm page fault that
381 * the page that is going to be mapped in the spte could have
384 kvm->mmu_notifier_seq++;
387 * The above sequence increase must be visible before the
388 * below count decrease, which is ensured by the smp_wmb above
389 * in conjunction with the smp_rmb in mmu_notifier_retry().
391 kvm->mmu_notifier_count--;
392 spin_unlock(&kvm->mmu_lock);
394 BUG_ON(kvm->mmu_notifier_count < 0);
397 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
398 struct mm_struct *mm,
402 struct kvm *kvm = mmu_notifier_to_kvm(mn);
405 idx = srcu_read_lock(&kvm->srcu);
406 spin_lock(&kvm->mmu_lock);
408 young = kvm_age_hva(kvm, start, end);
410 kvm_flush_remote_tlbs(kvm);
412 spin_unlock(&kvm->mmu_lock);
413 srcu_read_unlock(&kvm->srcu, idx);
418 static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
419 struct mm_struct *mm,
423 struct kvm *kvm = mmu_notifier_to_kvm(mn);
426 idx = srcu_read_lock(&kvm->srcu);
427 spin_lock(&kvm->mmu_lock);
429 * Even though we do not flush TLB, this will still adversely
430 * affect performance on pre-Haswell Intel EPT, where there is
431 * no EPT Access Bit to clear so that we have to tear down EPT
432 * tables instead. If we find this unacceptable, we can always
433 * add a parameter to kvm_age_hva so that it effectively doesn't
434 * do anything on clear_young.
436 * Also note that currently we never issue secondary TLB flushes
437 * from clear_young, leaving this job up to the regular system
438 * cadence. If we find this inaccurate, we might come up with a
439 * more sophisticated heuristic later.
441 young = kvm_age_hva(kvm, start, end);
442 spin_unlock(&kvm->mmu_lock);
443 srcu_read_unlock(&kvm->srcu, idx);
448 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
449 struct mm_struct *mm,
450 unsigned long address)
452 struct kvm *kvm = mmu_notifier_to_kvm(mn);
455 idx = srcu_read_lock(&kvm->srcu);
456 spin_lock(&kvm->mmu_lock);
457 young = kvm_test_age_hva(kvm, address);
458 spin_unlock(&kvm->mmu_lock);
459 srcu_read_unlock(&kvm->srcu, idx);
464 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
465 struct mm_struct *mm)
467 struct kvm *kvm = mmu_notifier_to_kvm(mn);
470 idx = srcu_read_lock(&kvm->srcu);
471 kvm_arch_flush_shadow_all(kvm);
472 srcu_read_unlock(&kvm->srcu, idx);
475 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
476 .invalidate_page = kvm_mmu_notifier_invalidate_page,
477 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
478 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
479 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
480 .clear_young = kvm_mmu_notifier_clear_young,
481 .test_young = kvm_mmu_notifier_test_young,
482 .change_pte = kvm_mmu_notifier_change_pte,
483 .release = kvm_mmu_notifier_release,
486 static int kvm_init_mmu_notifier(struct kvm *kvm)
488 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
489 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
492 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
494 static int kvm_init_mmu_notifier(struct kvm *kvm)
499 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
501 static struct kvm_memslots *kvm_alloc_memslots(void)
504 struct kvm_memslots *slots;
506 slots = kvm_kvzalloc(sizeof(struct kvm_memslots));
510 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
511 slots->id_to_index[i] = slots->memslots[i].id = i;
516 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
518 if (!memslot->dirty_bitmap)
521 kvfree(memslot->dirty_bitmap);
522 memslot->dirty_bitmap = NULL;
526 * Free any memory in @free but not in @dont.
528 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
529 struct kvm_memory_slot *dont)
531 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
532 kvm_destroy_dirty_bitmap(free);
534 kvm_arch_free_memslot(kvm, free, dont);
539 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
541 struct kvm_memory_slot *memslot;
546 kvm_for_each_memslot(memslot, slots)
547 kvm_free_memslot(kvm, memslot, NULL);
552 static void kvm_destroy_vm_debugfs(struct kvm *kvm)
556 if (!kvm->debugfs_dentry)
559 debugfs_remove_recursive(kvm->debugfs_dentry);
561 if (kvm->debugfs_stat_data) {
562 for (i = 0; i < kvm_debugfs_num_entries; i++)
563 kfree(kvm->debugfs_stat_data[i]);
564 kfree(kvm->debugfs_stat_data);
568 static int kvm_create_vm_debugfs(struct kvm *kvm, int fd)
570 char dir_name[ITOA_MAX_LEN * 2];
571 struct kvm_stat_data *stat_data;
572 struct kvm_stats_debugfs_item *p;
574 if (!debugfs_initialized())
577 snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd);
578 kvm->debugfs_dentry = debugfs_create_dir(dir_name,
580 if (!kvm->debugfs_dentry)
583 kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries,
584 sizeof(*kvm->debugfs_stat_data),
586 if (!kvm->debugfs_stat_data)
589 for (p = debugfs_entries; p->name; p++) {
590 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL);
594 stat_data->kvm = kvm;
595 stat_data->offset = p->offset;
596 kvm->debugfs_stat_data[p - debugfs_entries] = stat_data;
597 if (!debugfs_create_file(p->name, 0644,
600 stat_fops_per_vm[p->kind]))
606 static struct kvm *kvm_create_vm(unsigned long type)
609 struct kvm *kvm = kvm_arch_alloc_vm();
612 return ERR_PTR(-ENOMEM);
614 spin_lock_init(&kvm->mmu_lock);
616 kvm->mm = current->mm;
617 kvm_eventfd_init(kvm);
618 mutex_init(&kvm->lock);
619 mutex_init(&kvm->irq_lock);
620 mutex_init(&kvm->slots_lock);
621 atomic_set(&kvm->users_count, 1);
622 INIT_LIST_HEAD(&kvm->devices);
624 r = kvm_arch_init_vm(kvm, type);
626 goto out_err_no_disable;
628 r = hardware_enable_all();
630 goto out_err_no_disable;
632 #ifdef CONFIG_HAVE_KVM_IRQFD
633 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
636 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
639 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
640 struct kvm_memslots *slots = kvm_alloc_memslots();
642 goto out_err_no_srcu;
644 * Generations must be different for each address space.
645 * Init kvm generation close to the maximum to easily test the
646 * code of handling generation number wrap-around.
648 slots->generation = i * 2 - 150;
649 rcu_assign_pointer(kvm->memslots[i], slots);
652 if (init_srcu_struct(&kvm->srcu))
653 goto out_err_no_srcu;
654 if (init_srcu_struct(&kvm->irq_srcu))
655 goto out_err_no_irq_srcu;
656 for (i = 0; i < KVM_NR_BUSES; i++) {
657 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
663 r = kvm_init_mmu_notifier(kvm);
667 spin_lock(&kvm_lock);
668 list_add(&kvm->vm_list, &vm_list);
669 spin_unlock(&kvm_lock);
671 preempt_notifier_inc();
676 cleanup_srcu_struct(&kvm->irq_srcu);
678 cleanup_srcu_struct(&kvm->srcu);
680 hardware_disable_all();
682 for (i = 0; i < KVM_NR_BUSES; i++)
683 kfree(kvm->buses[i]);
684 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
685 kvm_free_memslots(kvm, kvm->memslots[i]);
686 kvm_arch_free_vm(kvm);
692 * Avoid using vmalloc for a small buffer.
693 * Should not be used when the size is statically known.
695 void *kvm_kvzalloc(unsigned long size)
697 if (size > PAGE_SIZE)
698 return vzalloc(size);
700 return kzalloc(size, GFP_KERNEL);
703 static void kvm_destroy_devices(struct kvm *kvm)
705 struct kvm_device *dev, *tmp;
708 * We do not need to take the kvm->lock here, because nobody else
709 * has a reference to the struct kvm at this point and therefore
710 * cannot access the devices list anyhow.
712 list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
713 list_del(&dev->vm_node);
714 dev->ops->destroy(dev);
718 static void kvm_destroy_vm(struct kvm *kvm)
721 struct mm_struct *mm = kvm->mm;
723 kvm_destroy_vm_debugfs(kvm);
724 kvm_arch_sync_events(kvm);
725 spin_lock(&kvm_lock);
726 list_del(&kvm->vm_list);
727 spin_unlock(&kvm_lock);
728 kvm_free_irq_routing(kvm);
729 for (i = 0; i < KVM_NR_BUSES; i++)
730 kvm_io_bus_destroy(kvm->buses[i]);
731 kvm_coalesced_mmio_free(kvm);
732 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
733 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
735 kvm_arch_flush_shadow_all(kvm);
737 kvm_arch_destroy_vm(kvm);
738 kvm_destroy_devices(kvm);
739 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
740 kvm_free_memslots(kvm, kvm->memslots[i]);
741 cleanup_srcu_struct(&kvm->irq_srcu);
742 cleanup_srcu_struct(&kvm->srcu);
743 kvm_arch_free_vm(kvm);
744 preempt_notifier_dec();
745 hardware_disable_all();
749 void kvm_get_kvm(struct kvm *kvm)
751 atomic_inc(&kvm->users_count);
753 EXPORT_SYMBOL_GPL(kvm_get_kvm);
755 void kvm_put_kvm(struct kvm *kvm)
757 if (atomic_dec_and_test(&kvm->users_count))
760 EXPORT_SYMBOL_GPL(kvm_put_kvm);
763 static int kvm_vm_release(struct inode *inode, struct file *filp)
765 struct kvm *kvm = filp->private_data;
767 kvm_irqfd_release(kvm);
774 * Allocation size is twice as large as the actual dirty bitmap size.
775 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
777 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
779 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
781 memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
782 if (!memslot->dirty_bitmap)
789 * Insert memslot and re-sort memslots based on their GFN,
790 * so binary search could be used to lookup GFN.
791 * Sorting algorithm takes advantage of having initially
792 * sorted array and known changed memslot position.
794 static void update_memslots(struct kvm_memslots *slots,
795 struct kvm_memory_slot *new)
798 int i = slots->id_to_index[id];
799 struct kvm_memory_slot *mslots = slots->memslots;
801 WARN_ON(mslots[i].id != id);
803 WARN_ON(!mslots[i].npages);
804 if (mslots[i].npages)
807 if (!mslots[i].npages)
811 while (i < KVM_MEM_SLOTS_NUM - 1 &&
812 new->base_gfn <= mslots[i + 1].base_gfn) {
813 if (!mslots[i + 1].npages)
815 mslots[i] = mslots[i + 1];
816 slots->id_to_index[mslots[i].id] = i;
821 * The ">=" is needed when creating a slot with base_gfn == 0,
822 * so that it moves before all those with base_gfn == npages == 0.
824 * On the other hand, if new->npages is zero, the above loop has
825 * already left i pointing to the beginning of the empty part of
826 * mslots, and the ">=" would move the hole backwards in this
827 * case---which is wrong. So skip the loop when deleting a slot.
831 new->base_gfn >= mslots[i - 1].base_gfn) {
832 mslots[i] = mslots[i - 1];
833 slots->id_to_index[mslots[i].id] = i;
837 WARN_ON_ONCE(i != slots->used_slots);
840 slots->id_to_index[mslots[i].id] = i;
843 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
845 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
847 #ifdef __KVM_HAVE_READONLY_MEM
848 valid_flags |= KVM_MEM_READONLY;
851 if (mem->flags & ~valid_flags)
857 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
858 int as_id, struct kvm_memslots *slots)
860 struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id);
863 * Set the low bit in the generation, which disables SPTE caching
864 * until the end of synchronize_srcu_expedited.
866 WARN_ON(old_memslots->generation & 1);
867 slots->generation = old_memslots->generation + 1;
869 rcu_assign_pointer(kvm->memslots[as_id], slots);
870 synchronize_srcu_expedited(&kvm->srcu);
873 * Increment the new memslot generation a second time. This prevents
874 * vm exits that race with memslot updates from caching a memslot
875 * generation that will (potentially) be valid forever.
877 * Generations must be unique even across address spaces. We do not need
878 * a global counter for that, instead the generation space is evenly split
879 * across address spaces. For example, with two address spaces, address
880 * space 0 will use generations 0, 4, 8, ... while * address space 1 will
881 * use generations 2, 6, 10, 14, ...
883 slots->generation += KVM_ADDRESS_SPACE_NUM * 2 - 1;
885 kvm_arch_memslots_updated(kvm, slots);
891 * Allocate some memory and give it an address in the guest physical address
894 * Discontiguous memory is allowed, mostly for framebuffers.
896 * Must be called holding kvm->slots_lock for write.
898 int __kvm_set_memory_region(struct kvm *kvm,
899 const struct kvm_userspace_memory_region *mem)
903 unsigned long npages;
904 struct kvm_memory_slot *slot;
905 struct kvm_memory_slot old, new;
906 struct kvm_memslots *slots = NULL, *old_memslots;
908 enum kvm_mr_change change;
910 r = check_memory_region_flags(mem);
915 as_id = mem->slot >> 16;
918 /* General sanity checks */
919 if (mem->memory_size & (PAGE_SIZE - 1))
921 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
923 /* We can read the guest memory with __xxx_user() later on. */
924 if ((id < KVM_USER_MEM_SLOTS) &&
925 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
926 !access_ok(VERIFY_WRITE,
927 (void __user *)(unsigned long)mem->userspace_addr,
930 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
932 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
935 slot = id_to_memslot(__kvm_memslots(kvm, as_id), id);
936 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
937 npages = mem->memory_size >> PAGE_SHIFT;
939 if (npages > KVM_MEM_MAX_NR_PAGES)
945 new.base_gfn = base_gfn;
947 new.flags = mem->flags;
951 change = KVM_MR_CREATE;
952 else { /* Modify an existing slot. */
953 if ((mem->userspace_addr != old.userspace_addr) ||
954 (npages != old.npages) ||
955 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
958 if (base_gfn != old.base_gfn)
959 change = KVM_MR_MOVE;
960 else if (new.flags != old.flags)
961 change = KVM_MR_FLAGS_ONLY;
962 else { /* Nothing to change. */
971 change = KVM_MR_DELETE;
976 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
977 /* Check for overlaps */
979 kvm_for_each_memslot(slot, __kvm_memslots(kvm, as_id)) {
980 if ((slot->id >= KVM_USER_MEM_SLOTS) ||
983 if (!((base_gfn + npages <= slot->base_gfn) ||
984 (base_gfn >= slot->base_gfn + slot->npages)))
989 /* Free page dirty bitmap if unneeded */
990 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
991 new.dirty_bitmap = NULL;
994 if (change == KVM_MR_CREATE) {
995 new.userspace_addr = mem->userspace_addr;
997 if (kvm_arch_create_memslot(kvm, &new, npages))
1001 /* Allocate page dirty bitmap if needed */
1002 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
1003 if (kvm_create_dirty_bitmap(&new) < 0)
1007 slots = kvm_kvzalloc(sizeof(struct kvm_memslots));
1010 memcpy(slots, __kvm_memslots(kvm, as_id), sizeof(struct kvm_memslots));
1012 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
1013 slot = id_to_memslot(slots, id);
1014 slot->flags |= KVM_MEMSLOT_INVALID;
1016 old_memslots = install_new_memslots(kvm, as_id, slots);
1018 /* slot was deleted or moved, clear iommu mapping */
1019 kvm_iommu_unmap_pages(kvm, &old);
1020 /* From this point no new shadow pages pointing to a deleted,
1021 * or moved, memslot will be created.
1023 * validation of sp->gfn happens in:
1024 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1025 * - kvm_is_visible_gfn (mmu_check_roots)
1027 kvm_arch_flush_shadow_memslot(kvm, slot);
1030 * We can re-use the old_memslots from above, the only difference
1031 * from the currently installed memslots is the invalid flag. This
1032 * will get overwritten by update_memslots anyway.
1034 slots = old_memslots;
1037 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
1041 /* actual memory is freed via old in kvm_free_memslot below */
1042 if (change == KVM_MR_DELETE) {
1043 new.dirty_bitmap = NULL;
1044 memset(&new.arch, 0, sizeof(new.arch));
1047 update_memslots(slots, &new);
1048 old_memslots = install_new_memslots(kvm, as_id, slots);
1050 kvm_arch_commit_memory_region(kvm, mem, &old, &new, change);
1052 kvm_free_memslot(kvm, &old, &new);
1053 kvfree(old_memslots);
1056 * IOMMU mapping: New slots need to be mapped. Old slots need to be
1057 * un-mapped and re-mapped if their base changes. Since base change
1058 * unmapping is handled above with slot deletion, mapping alone is
1059 * needed here. Anything else the iommu might care about for existing
1060 * slots (size changes, userspace addr changes and read-only flag
1061 * changes) is disallowed above, so any other attribute changes getting
1062 * here can be skipped.
1064 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
1065 r = kvm_iommu_map_pages(kvm, &new);
1074 kvm_free_memslot(kvm, &new, &old);
1078 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1080 int kvm_set_memory_region(struct kvm *kvm,
1081 const struct kvm_userspace_memory_region *mem)
1085 mutex_lock(&kvm->slots_lock);
1086 r = __kvm_set_memory_region(kvm, mem);
1087 mutex_unlock(&kvm->slots_lock);
1090 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1092 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1093 struct kvm_userspace_memory_region *mem)
1095 if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1098 return kvm_set_memory_region(kvm, mem);
1101 int kvm_get_dirty_log(struct kvm *kvm,
1102 struct kvm_dirty_log *log, int *is_dirty)
1104 struct kvm_memslots *slots;
1105 struct kvm_memory_slot *memslot;
1108 unsigned long any = 0;
1110 as_id = log->slot >> 16;
1111 id = (u16)log->slot;
1112 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1115 slots = __kvm_memslots(kvm, as_id);
1116 memslot = id_to_memslot(slots, id);
1117 if (!memslot->dirty_bitmap)
1120 n = kvm_dirty_bitmap_bytes(memslot);
1122 for (i = 0; !any && i < n/sizeof(long); ++i)
1123 any = memslot->dirty_bitmap[i];
1125 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1132 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1134 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1136 * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages
1137 * are dirty write protect them for next write.
1138 * @kvm: pointer to kvm instance
1139 * @log: slot id and address to which we copy the log
1140 * @is_dirty: flag set if any page is dirty
1142 * We need to keep it in mind that VCPU threads can write to the bitmap
1143 * concurrently. So, to avoid losing track of dirty pages we keep the
1146 * 1. Take a snapshot of the bit and clear it if needed.
1147 * 2. Write protect the corresponding page.
1148 * 3. Copy the snapshot to the userspace.
1149 * 4. Upon return caller flushes TLB's if needed.
1151 * Between 2 and 4, the guest may write to the page using the remaining TLB
1152 * entry. This is not a problem because the page is reported dirty using
1153 * the snapshot taken before and step 4 ensures that writes done after
1154 * exiting to userspace will be logged for the next call.
1157 int kvm_get_dirty_log_protect(struct kvm *kvm,
1158 struct kvm_dirty_log *log, bool *is_dirty)
1160 struct kvm_memslots *slots;
1161 struct kvm_memory_slot *memslot;
1164 unsigned long *dirty_bitmap;
1165 unsigned long *dirty_bitmap_buffer;
1167 as_id = log->slot >> 16;
1168 id = (u16)log->slot;
1169 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1172 slots = __kvm_memslots(kvm, as_id);
1173 memslot = id_to_memslot(slots, id);
1175 dirty_bitmap = memslot->dirty_bitmap;
1179 n = kvm_dirty_bitmap_bytes(memslot);
1181 dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long);
1182 memset(dirty_bitmap_buffer, 0, n);
1184 spin_lock(&kvm->mmu_lock);
1186 for (i = 0; i < n / sizeof(long); i++) {
1190 if (!dirty_bitmap[i])
1195 mask = xchg(&dirty_bitmap[i], 0);
1196 dirty_bitmap_buffer[i] = mask;
1199 offset = i * BITS_PER_LONG;
1200 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1205 spin_unlock(&kvm->mmu_lock);
1206 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1210 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect);
1213 bool kvm_largepages_enabled(void)
1215 return largepages_enabled;
1218 void kvm_disable_largepages(void)
1220 largepages_enabled = false;
1222 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1224 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1226 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1228 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1230 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
1232 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn);
1235 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1237 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1239 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1240 memslot->flags & KVM_MEMSLOT_INVALID)
1245 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1247 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1249 struct vm_area_struct *vma;
1250 unsigned long addr, size;
1254 addr = gfn_to_hva(kvm, gfn);
1255 if (kvm_is_error_hva(addr))
1258 down_read(¤t->mm->mmap_sem);
1259 vma = find_vma(current->mm, addr);
1263 size = vma_kernel_pagesize(vma);
1266 up_read(¤t->mm->mmap_sem);
1271 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1273 return slot->flags & KVM_MEM_READONLY;
1276 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1277 gfn_t *nr_pages, bool write)
1279 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1280 return KVM_HVA_ERR_BAD;
1282 if (memslot_is_readonly(slot) && write)
1283 return KVM_HVA_ERR_RO_BAD;
1286 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1288 return __gfn_to_hva_memslot(slot, gfn);
1291 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1294 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1297 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1300 return gfn_to_hva_many(slot, gfn, NULL);
1302 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1304 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1306 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1308 EXPORT_SYMBOL_GPL(gfn_to_hva);
1310 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
1312 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
1314 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
1317 * If writable is set to false, the hva returned by this function is only
1318 * allowed to be read.
1320 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1321 gfn_t gfn, bool *writable)
1323 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1325 if (!kvm_is_error_hva(hva) && writable)
1326 *writable = !memslot_is_readonly(slot);
1331 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1333 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1335 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1338 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
1340 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1342 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1345 static int get_user_page_nowait(unsigned long start, int write,
1348 int flags = FOLL_NOWAIT | FOLL_HWPOISON;
1351 flags |= FOLL_WRITE;
1353 return get_user_pages(start, 1, flags, page, NULL);
1356 static inline int check_user_page_hwpoison(unsigned long addr)
1358 int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
1360 rc = get_user_pages(addr, 1, flags, NULL, NULL);
1361 return rc == -EHWPOISON;
1365 * The atomic path to get the writable pfn which will be stored in @pfn,
1366 * true indicates success, otherwise false is returned.
1368 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1369 bool write_fault, bool *writable, kvm_pfn_t *pfn)
1371 struct page *page[1];
1374 if (!(async || atomic))
1378 * Fast pin a writable pfn only if it is a write fault request
1379 * or the caller allows to map a writable pfn for a read fault
1382 if (!(write_fault || writable))
1385 npages = __get_user_pages_fast(addr, 1, 1, page);
1387 *pfn = page_to_pfn(page[0]);
1398 * The slow path to get the pfn of the specified host virtual address,
1399 * 1 indicates success, -errno is returned if error is detected.
1401 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1402 bool *writable, kvm_pfn_t *pfn)
1404 struct page *page[1];
1410 *writable = write_fault;
1413 down_read(¤t->mm->mmap_sem);
1414 npages = get_user_page_nowait(addr, write_fault, page);
1415 up_read(¤t->mm->mmap_sem);
1417 unsigned int flags = FOLL_HWPOISON;
1420 flags |= FOLL_WRITE;
1422 npages = get_user_pages_unlocked(addr, 1, page, flags);
1427 /* map read fault as writable if possible */
1428 if (unlikely(!write_fault) && writable) {
1429 struct page *wpage[1];
1431 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1440 *pfn = page_to_pfn(page[0]);
1444 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1446 if (unlikely(!(vma->vm_flags & VM_READ)))
1449 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1455 static int hva_to_pfn_remapped(struct vm_area_struct *vma,
1456 unsigned long addr, bool *async,
1457 bool write_fault, kvm_pfn_t *p_pfn)
1462 r = follow_pfn(vma, addr, &pfn);
1465 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
1466 * not call the fault handler, so do it here.
1468 bool unlocked = false;
1469 r = fixup_user_fault(current, current->mm, addr,
1470 (write_fault ? FAULT_FLAG_WRITE : 0),
1477 r = follow_pfn(vma, addr, &pfn);
1485 * Get a reference here because callers of *hva_to_pfn* and
1486 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
1487 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
1488 * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will
1489 * simply do nothing for reserved pfns.
1491 * Whoever called remap_pfn_range is also going to call e.g.
1492 * unmap_mapping_range before the underlying pages are freed,
1493 * causing a call to our MMU notifier.
1502 * Pin guest page in memory and return its pfn.
1503 * @addr: host virtual address which maps memory to the guest
1504 * @atomic: whether this function can sleep
1505 * @async: whether this function need to wait IO complete if the
1506 * host page is not in the memory
1507 * @write_fault: whether we should get a writable host page
1508 * @writable: whether it allows to map a writable host page for !@write_fault
1510 * The function will map a writable host page for these two cases:
1511 * 1): @write_fault = true
1512 * 2): @write_fault = false && @writable, @writable will tell the caller
1513 * whether the mapping is writable.
1515 static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1516 bool write_fault, bool *writable)
1518 struct vm_area_struct *vma;
1522 /* we can do it either atomically or asynchronously, not both */
1523 BUG_ON(atomic && async);
1525 if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1529 return KVM_PFN_ERR_FAULT;
1531 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1535 down_read(¤t->mm->mmap_sem);
1536 if (npages == -EHWPOISON ||
1537 (!async && check_user_page_hwpoison(addr))) {
1538 pfn = KVM_PFN_ERR_HWPOISON;
1543 vma = find_vma_intersection(current->mm, addr, addr + 1);
1546 pfn = KVM_PFN_ERR_FAULT;
1547 else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
1548 r = hva_to_pfn_remapped(vma, addr, async, write_fault, &pfn);
1552 pfn = KVM_PFN_ERR_FAULT;
1554 if (async && vma_is_valid(vma, write_fault))
1556 pfn = KVM_PFN_ERR_FAULT;
1559 up_read(¤t->mm->mmap_sem);
1563 kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn,
1564 bool atomic, bool *async, bool write_fault,
1567 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1569 if (addr == KVM_HVA_ERR_RO_BAD) {
1572 return KVM_PFN_ERR_RO_FAULT;
1575 if (kvm_is_error_hva(addr)) {
1578 return KVM_PFN_NOSLOT;
1581 /* Do not map writable pfn in the readonly memslot. */
1582 if (writable && memslot_is_readonly(slot)) {
1587 return hva_to_pfn(addr, atomic, async, write_fault,
1590 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
1592 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1595 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
1596 write_fault, writable);
1598 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1600 kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1602 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1604 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
1606 kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1608 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1610 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1612 kvm_pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1614 return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm, gfn), gfn);
1616 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1618 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
1620 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1622 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
1624 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1626 return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
1628 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1630 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
1632 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1634 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
1636 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1637 struct page **pages, int nr_pages)
1642 addr = gfn_to_hva_many(slot, gfn, &entry);
1643 if (kvm_is_error_hva(addr))
1646 if (entry < nr_pages)
1649 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1651 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1653 static struct page *kvm_pfn_to_page(kvm_pfn_t pfn)
1655 if (is_error_noslot_pfn(pfn))
1656 return KVM_ERR_PTR_BAD_PAGE;
1658 if (kvm_is_reserved_pfn(pfn)) {
1660 return KVM_ERR_PTR_BAD_PAGE;
1663 return pfn_to_page(pfn);
1666 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1670 pfn = gfn_to_pfn(kvm, gfn);
1672 return kvm_pfn_to_page(pfn);
1674 EXPORT_SYMBOL_GPL(gfn_to_page);
1676 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
1680 pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
1682 return kvm_pfn_to_page(pfn);
1684 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
1686 void kvm_release_page_clean(struct page *page)
1688 WARN_ON(is_error_page(page));
1690 kvm_release_pfn_clean(page_to_pfn(page));
1692 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1694 void kvm_release_pfn_clean(kvm_pfn_t pfn)
1696 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
1697 put_page(pfn_to_page(pfn));
1699 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1701 void kvm_release_page_dirty(struct page *page)
1703 WARN_ON(is_error_page(page));
1705 kvm_release_pfn_dirty(page_to_pfn(page));
1707 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1709 static void kvm_release_pfn_dirty(kvm_pfn_t pfn)
1711 kvm_set_pfn_dirty(pfn);
1712 kvm_release_pfn_clean(pfn);
1715 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
1717 if (!kvm_is_reserved_pfn(pfn)) {
1718 struct page *page = pfn_to_page(pfn);
1720 if (!PageReserved(page))
1724 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1726 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
1728 if (!kvm_is_reserved_pfn(pfn))
1729 mark_page_accessed(pfn_to_page(pfn));
1731 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1733 void kvm_get_pfn(kvm_pfn_t pfn)
1735 if (!kvm_is_reserved_pfn(pfn))
1736 get_page(pfn_to_page(pfn));
1738 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1740 static int next_segment(unsigned long len, int offset)
1742 if (len > PAGE_SIZE - offset)
1743 return PAGE_SIZE - offset;
1748 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
1749 void *data, int offset, int len)
1754 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1755 if (kvm_is_error_hva(addr))
1757 r = __copy_from_user(data, (void __user *)addr + offset, len);
1763 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1766 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1768 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1770 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1772 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
1773 int offset, int len)
1775 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1777 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1779 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
1781 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1783 gfn_t gfn = gpa >> PAGE_SHIFT;
1785 int offset = offset_in_page(gpa);
1788 while ((seg = next_segment(len, offset)) != 0) {
1789 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1799 EXPORT_SYMBOL_GPL(kvm_read_guest);
1801 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
1803 gfn_t gfn = gpa >> PAGE_SHIFT;
1805 int offset = offset_in_page(gpa);
1808 while ((seg = next_segment(len, offset)) != 0) {
1809 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
1819 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
1821 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1822 void *data, int offset, unsigned long len)
1827 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1828 if (kvm_is_error_hva(addr))
1830 pagefault_disable();
1831 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1838 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1841 gfn_t gfn = gpa >> PAGE_SHIFT;
1842 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1843 int offset = offset_in_page(gpa);
1845 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1847 EXPORT_SYMBOL_GPL(kvm_read_guest_atomic);
1849 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
1850 void *data, unsigned long len)
1852 gfn_t gfn = gpa >> PAGE_SHIFT;
1853 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1854 int offset = offset_in_page(gpa);
1856 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1858 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
1860 static int __kvm_write_guest_page(struct kvm_memory_slot *memslot, gfn_t gfn,
1861 const void *data, int offset, int len)
1866 addr = gfn_to_hva_memslot(memslot, gfn);
1867 if (kvm_is_error_hva(addr))
1869 r = __copy_to_user((void __user *)addr + offset, data, len);
1872 mark_page_dirty_in_slot(memslot, gfn);
1876 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
1877 const void *data, int offset, int len)
1879 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1881 return __kvm_write_guest_page(slot, gfn, data, offset, len);
1883 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1885 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
1886 const void *data, int offset, int len)
1888 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1890 return __kvm_write_guest_page(slot, gfn, data, offset, len);
1892 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
1894 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1897 gfn_t gfn = gpa >> PAGE_SHIFT;
1899 int offset = offset_in_page(gpa);
1902 while ((seg = next_segment(len, offset)) != 0) {
1903 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1913 EXPORT_SYMBOL_GPL(kvm_write_guest);
1915 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
1918 gfn_t gfn = gpa >> PAGE_SHIFT;
1920 int offset = offset_in_page(gpa);
1923 while ((seg = next_segment(len, offset)) != 0) {
1924 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
1934 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
1936 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
1937 struct gfn_to_hva_cache *ghc,
1938 gpa_t gpa, unsigned long len)
1940 int offset = offset_in_page(gpa);
1941 gfn_t start_gfn = gpa >> PAGE_SHIFT;
1942 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1943 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1944 gfn_t nr_pages_avail;
1947 ghc->generation = slots->generation;
1949 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
1950 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, NULL);
1951 if (!kvm_is_error_hva(ghc->hva) && nr_pages_needed <= 1) {
1955 * If the requested region crosses two memslots, we still
1956 * verify that the entire region is valid here.
1958 while (start_gfn <= end_gfn) {
1959 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
1960 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1962 if (kvm_is_error_hva(ghc->hva))
1964 start_gfn += nr_pages_avail;
1966 /* Use the slow path for cross page reads and writes. */
1967 ghc->memslot = NULL;
1972 int kvm_vcpu_gfn_to_hva_cache_init(struct kvm_vcpu *vcpu, struct gfn_to_hva_cache *ghc,
1973 gpa_t gpa, unsigned long len)
1975 struct kvm_memslots *slots = kvm_vcpu_memslots(vcpu);
1976 return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len);
1978 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva_cache_init);
1980 int kvm_vcpu_write_guest_offset_cached(struct kvm_vcpu *vcpu, struct gfn_to_hva_cache *ghc,
1981 void *data, int offset, unsigned long len)
1983 struct kvm_memslots *slots = kvm_vcpu_memslots(vcpu);
1985 gpa_t gpa = ghc->gpa + offset;
1987 BUG_ON(len + offset > ghc->len);
1989 if (slots->generation != ghc->generation)
1990 __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len);
1992 if (unlikely(!ghc->memslot))
1993 return kvm_vcpu_write_guest(vcpu, gpa, data, len);
1995 if (kvm_is_error_hva(ghc->hva))
1998 r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
2001 mark_page_dirty_in_slot(ghc->memslot, gpa >> PAGE_SHIFT);
2005 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_offset_cached);
2007 int kvm_vcpu_write_guest_cached(struct kvm_vcpu *vcpu, struct gfn_to_hva_cache *ghc,
2008 void *data, unsigned long len)
2010 return kvm_vcpu_write_guest_offset_cached(vcpu, ghc, data, 0, len);
2012 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_cached);
2014 int kvm_vcpu_read_guest_cached(struct kvm_vcpu *vcpu, struct gfn_to_hva_cache *ghc,
2015 void *data, unsigned long len)
2017 struct kvm_memslots *slots = kvm_vcpu_memslots(vcpu);
2020 BUG_ON(len > ghc->len);
2022 if (slots->generation != ghc->generation)
2023 __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len);
2025 if (unlikely(!ghc->memslot))
2026 return kvm_vcpu_read_guest(vcpu, ghc->gpa, data, len);
2028 if (kvm_is_error_hva(ghc->hva))
2031 r = __copy_from_user(data, (void __user *)ghc->hva, len);
2037 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_cached);
2039 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
2041 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
2043 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
2045 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
2047 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
2049 gfn_t gfn = gpa >> PAGE_SHIFT;
2051 int offset = offset_in_page(gpa);
2054 while ((seg = next_segment(len, offset)) != 0) {
2055 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
2064 EXPORT_SYMBOL_GPL(kvm_clear_guest);
2066 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot,
2069 if (memslot && memslot->dirty_bitmap) {
2070 unsigned long rel_gfn = gfn - memslot->base_gfn;
2072 set_bit_le(rel_gfn, memslot->dirty_bitmap);
2076 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
2078 struct kvm_memory_slot *memslot;
2080 memslot = gfn_to_memslot(kvm, gfn);
2081 mark_page_dirty_in_slot(memslot, gfn);
2083 EXPORT_SYMBOL_GPL(mark_page_dirty);
2085 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
2087 struct kvm_memory_slot *memslot;
2089 memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2090 mark_page_dirty_in_slot(memslot, gfn);
2092 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
2094 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
2096 unsigned int old, val, grow;
2098 old = val = vcpu->halt_poll_ns;
2099 grow = READ_ONCE(halt_poll_ns_grow);
2101 if (val == 0 && grow)
2106 if (val > halt_poll_ns)
2109 vcpu->halt_poll_ns = val;
2110 trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
2113 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
2115 unsigned int old, val, shrink;
2117 old = val = vcpu->halt_poll_ns;
2118 shrink = READ_ONCE(halt_poll_ns_shrink);
2124 vcpu->halt_poll_ns = val;
2125 trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
2128 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
2130 if (kvm_arch_vcpu_runnable(vcpu)) {
2131 kvm_make_request(KVM_REQ_UNHALT, vcpu);
2134 if (kvm_cpu_has_pending_timer(vcpu))
2136 if (signal_pending(current))
2143 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2145 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
2148 DECLARE_SWAITQUEUE(wait);
2149 bool waited = false;
2152 start = cur = ktime_get();
2153 if (vcpu->halt_poll_ns) {
2154 ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns);
2156 ++vcpu->stat.halt_attempted_poll;
2159 * This sets KVM_REQ_UNHALT if an interrupt
2162 if (kvm_vcpu_check_block(vcpu) < 0) {
2163 ++vcpu->stat.halt_successful_poll;
2164 if (!vcpu_valid_wakeup(vcpu))
2165 ++vcpu->stat.halt_poll_invalid;
2169 } while (single_task_running() && ktime_before(cur, stop));
2172 kvm_arch_vcpu_blocking(vcpu);
2175 prepare_to_swait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
2177 if (kvm_vcpu_check_block(vcpu) < 0)
2184 finish_swait(&vcpu->wq, &wait);
2187 kvm_arch_vcpu_unblocking(vcpu);
2189 block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
2191 if (!vcpu_valid_wakeup(vcpu))
2192 shrink_halt_poll_ns(vcpu);
2193 else if (halt_poll_ns) {
2194 if (block_ns <= vcpu->halt_poll_ns)
2196 /* we had a long block, shrink polling */
2197 else if (vcpu->halt_poll_ns && block_ns > halt_poll_ns)
2198 shrink_halt_poll_ns(vcpu);
2199 /* we had a short halt and our poll time is too small */
2200 else if (vcpu->halt_poll_ns < halt_poll_ns &&
2201 block_ns < halt_poll_ns)
2202 grow_halt_poll_ns(vcpu);
2204 vcpu->halt_poll_ns = 0;
2206 trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu));
2207 kvm_arch_vcpu_block_finish(vcpu);
2209 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
2212 void kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
2214 struct swait_queue_head *wqp;
2216 wqp = kvm_arch_vcpu_wq(vcpu);
2217 if (swait_active(wqp)) {
2219 ++vcpu->stat.halt_wakeup;
2223 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
2226 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2228 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
2231 int cpu = vcpu->cpu;
2233 kvm_vcpu_wake_up(vcpu);
2235 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
2236 if (kvm_arch_vcpu_should_kick(vcpu))
2237 smp_send_reschedule(cpu);
2240 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
2241 #endif /* !CONFIG_S390 */
2243 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
2246 struct task_struct *task = NULL;
2250 pid = rcu_dereference(target->pid);
2252 task = get_pid_task(pid, PIDTYPE_PID);
2256 ret = yield_to(task, 1);
2257 put_task_struct(task);
2261 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
2264 * Helper that checks whether a VCPU is eligible for directed yield.
2265 * Most eligible candidate to yield is decided by following heuristics:
2267 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2268 * (preempted lock holder), indicated by @in_spin_loop.
2269 * Set at the beiginning and cleared at the end of interception/PLE handler.
2271 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2272 * chance last time (mostly it has become eligible now since we have probably
2273 * yielded to lockholder in last iteration. This is done by toggling
2274 * @dy_eligible each time a VCPU checked for eligibility.)
2276 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2277 * to preempted lock-holder could result in wrong VCPU selection and CPU
2278 * burning. Giving priority for a potential lock-holder increases lock
2281 * Since algorithm is based on heuristics, accessing another VCPU data without
2282 * locking does not harm. It may result in trying to yield to same VCPU, fail
2283 * and continue with next VCPU and so on.
2285 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
2287 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2290 eligible = !vcpu->spin_loop.in_spin_loop ||
2291 vcpu->spin_loop.dy_eligible;
2293 if (vcpu->spin_loop.in_spin_loop)
2294 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
2302 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
2304 struct kvm *kvm = me->kvm;
2305 struct kvm_vcpu *vcpu;
2306 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
2312 kvm_vcpu_set_in_spin_loop(me, true);
2314 * We boost the priority of a VCPU that is runnable but not
2315 * currently running, because it got preempted by something
2316 * else and called schedule in __vcpu_run. Hopefully that
2317 * VCPU is holding the lock that we need and will release it.
2318 * We approximate round-robin by starting at the last boosted VCPU.
2320 for (pass = 0; pass < 2 && !yielded && try; pass++) {
2321 kvm_for_each_vcpu(i, vcpu, kvm) {
2322 if (!pass && i <= last_boosted_vcpu) {
2323 i = last_boosted_vcpu;
2325 } else if (pass && i > last_boosted_vcpu)
2327 if (!ACCESS_ONCE(vcpu->preempted))
2331 if (swait_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
2333 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
2336 yielded = kvm_vcpu_yield_to(vcpu);
2338 kvm->last_boosted_vcpu = i;
2340 } else if (yielded < 0) {
2347 kvm_vcpu_set_in_spin_loop(me, false);
2349 /* Ensure vcpu is not eligible during next spinloop */
2350 kvm_vcpu_set_dy_eligible(me, false);
2352 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
2354 static int kvm_vcpu_fault(struct vm_fault *vmf)
2356 struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data;
2359 if (vmf->pgoff == 0)
2360 page = virt_to_page(vcpu->run);
2362 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
2363 page = virt_to_page(vcpu->arch.pio_data);
2365 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2366 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
2367 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
2370 return kvm_arch_vcpu_fault(vcpu, vmf);
2376 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
2377 .fault = kvm_vcpu_fault,
2380 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2382 vma->vm_ops = &kvm_vcpu_vm_ops;
2386 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2388 struct kvm_vcpu *vcpu = filp->private_data;
2390 debugfs_remove_recursive(vcpu->debugfs_dentry);
2391 kvm_put_kvm(vcpu->kvm);
2395 static struct file_operations kvm_vcpu_fops = {
2396 .release = kvm_vcpu_release,
2397 .unlocked_ioctl = kvm_vcpu_ioctl,
2398 #ifdef CONFIG_KVM_COMPAT
2399 .compat_ioctl = kvm_vcpu_compat_ioctl,
2401 .mmap = kvm_vcpu_mmap,
2402 .llseek = noop_llseek,
2406 * Allocates an inode for the vcpu.
2408 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2410 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
2413 static int kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
2415 char dir_name[ITOA_MAX_LEN * 2];
2418 if (!kvm_arch_has_vcpu_debugfs())
2421 if (!debugfs_initialized())
2424 snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
2425 vcpu->debugfs_dentry = debugfs_create_dir(dir_name,
2426 vcpu->kvm->debugfs_dentry);
2427 if (!vcpu->debugfs_dentry)
2430 ret = kvm_arch_create_vcpu_debugfs(vcpu);
2432 debugfs_remove_recursive(vcpu->debugfs_dentry);
2440 * Creates some virtual cpus. Good luck creating more than one.
2442 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
2445 struct kvm_vcpu *vcpu;
2447 if (id >= KVM_MAX_VCPU_ID)
2450 mutex_lock(&kvm->lock);
2451 if (kvm->created_vcpus == KVM_MAX_VCPUS) {
2452 mutex_unlock(&kvm->lock);
2456 kvm->created_vcpus++;
2457 mutex_unlock(&kvm->lock);
2459 vcpu = kvm_arch_vcpu_create(kvm, id);
2462 goto vcpu_decrement;
2465 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2467 r = kvm_arch_vcpu_setup(vcpu);
2471 r = kvm_create_vcpu_debugfs(vcpu);
2475 mutex_lock(&kvm->lock);
2476 if (kvm_get_vcpu_by_id(kvm, id)) {
2478 goto unlock_vcpu_destroy;
2481 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
2483 /* Now it's all set up, let userspace reach it */
2485 r = create_vcpu_fd(vcpu);
2488 goto unlock_vcpu_destroy;
2491 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
2494 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
2495 * before kvm->online_vcpu's incremented value.
2498 atomic_inc(&kvm->online_vcpus);
2500 mutex_unlock(&kvm->lock);
2501 kvm_arch_vcpu_postcreate(vcpu);
2504 unlock_vcpu_destroy:
2505 mutex_unlock(&kvm->lock);
2506 debugfs_remove_recursive(vcpu->debugfs_dentry);
2508 kvm_arch_vcpu_destroy(vcpu);
2510 mutex_lock(&kvm->lock);
2511 kvm->created_vcpus--;
2512 mutex_unlock(&kvm->lock);
2516 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2519 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2520 vcpu->sigset_active = 1;
2521 vcpu->sigset = *sigset;
2523 vcpu->sigset_active = 0;
2527 static long kvm_vcpu_ioctl(struct file *filp,
2528 unsigned int ioctl, unsigned long arg)
2530 struct kvm_vcpu *vcpu = filp->private_data;
2531 void __user *argp = (void __user *)arg;
2533 struct kvm_fpu *fpu = NULL;
2534 struct kvm_sregs *kvm_sregs = NULL;
2536 if (vcpu->kvm->mm != current->mm)
2539 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2542 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
2544 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
2545 * so vcpu_load() would break it.
2547 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_S390_IRQ || ioctl == KVM_INTERRUPT)
2548 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2552 r = vcpu_load(vcpu);
2560 if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
2561 /* The thread running this VCPU changed. */
2562 struct pid *oldpid = vcpu->pid;
2563 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
2565 rcu_assign_pointer(vcpu->pid, newpid);
2570 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2571 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2573 case KVM_GET_REGS: {
2574 struct kvm_regs *kvm_regs;
2577 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2580 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2584 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2591 case KVM_SET_REGS: {
2592 struct kvm_regs *kvm_regs;
2595 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2596 if (IS_ERR(kvm_regs)) {
2597 r = PTR_ERR(kvm_regs);
2600 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2604 case KVM_GET_SREGS: {
2605 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2609 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2613 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2618 case KVM_SET_SREGS: {
2619 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2620 if (IS_ERR(kvm_sregs)) {
2621 r = PTR_ERR(kvm_sregs);
2625 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2628 case KVM_GET_MP_STATE: {
2629 struct kvm_mp_state mp_state;
2631 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2635 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
2640 case KVM_SET_MP_STATE: {
2641 struct kvm_mp_state mp_state;
2644 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
2646 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2649 case KVM_TRANSLATE: {
2650 struct kvm_translation tr;
2653 if (copy_from_user(&tr, argp, sizeof(tr)))
2655 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2659 if (copy_to_user(argp, &tr, sizeof(tr)))
2664 case KVM_SET_GUEST_DEBUG: {
2665 struct kvm_guest_debug dbg;
2668 if (copy_from_user(&dbg, argp, sizeof(dbg)))
2670 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2673 case KVM_SET_SIGNAL_MASK: {
2674 struct kvm_signal_mask __user *sigmask_arg = argp;
2675 struct kvm_signal_mask kvm_sigmask;
2676 sigset_t sigset, *p;
2681 if (copy_from_user(&kvm_sigmask, argp,
2682 sizeof(kvm_sigmask)))
2685 if (kvm_sigmask.len != sizeof(sigset))
2688 if (copy_from_user(&sigset, sigmask_arg->sigset,
2693 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2697 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2701 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2705 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2711 fpu = memdup_user(argp, sizeof(*fpu));
2717 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2721 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2730 #ifdef CONFIG_KVM_COMPAT
2731 static long kvm_vcpu_compat_ioctl(struct file *filp,
2732 unsigned int ioctl, unsigned long arg)
2734 struct kvm_vcpu *vcpu = filp->private_data;
2735 void __user *argp = compat_ptr(arg);
2738 if (vcpu->kvm->mm != current->mm)
2742 case KVM_SET_SIGNAL_MASK: {
2743 struct kvm_signal_mask __user *sigmask_arg = argp;
2744 struct kvm_signal_mask kvm_sigmask;
2745 compat_sigset_t csigset;
2750 if (copy_from_user(&kvm_sigmask, argp,
2751 sizeof(kvm_sigmask)))
2754 if (kvm_sigmask.len != sizeof(csigset))
2757 if (copy_from_user(&csigset, sigmask_arg->sigset,
2760 sigset_from_compat(&sigset, &csigset);
2761 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2763 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2767 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2775 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2776 int (*accessor)(struct kvm_device *dev,
2777 struct kvm_device_attr *attr),
2780 struct kvm_device_attr attr;
2785 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2788 return accessor(dev, &attr);
2791 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2794 struct kvm_device *dev = filp->private_data;
2797 case KVM_SET_DEVICE_ATTR:
2798 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2799 case KVM_GET_DEVICE_ATTR:
2800 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2801 case KVM_HAS_DEVICE_ATTR:
2802 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2804 if (dev->ops->ioctl)
2805 return dev->ops->ioctl(dev, ioctl, arg);
2811 static int kvm_device_release(struct inode *inode, struct file *filp)
2813 struct kvm_device *dev = filp->private_data;
2814 struct kvm *kvm = dev->kvm;
2820 static const struct file_operations kvm_device_fops = {
2821 .unlocked_ioctl = kvm_device_ioctl,
2822 #ifdef CONFIG_KVM_COMPAT
2823 .compat_ioctl = kvm_device_ioctl,
2825 .release = kvm_device_release,
2828 struct kvm_device *kvm_device_from_filp(struct file *filp)
2830 if (filp->f_op != &kvm_device_fops)
2833 return filp->private_data;
2836 static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
2837 #ifdef CONFIG_KVM_MPIC
2838 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
2839 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
2842 #ifdef CONFIG_KVM_XICS
2843 [KVM_DEV_TYPE_XICS] = &kvm_xics_ops,
2847 int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
2849 if (type >= ARRAY_SIZE(kvm_device_ops_table))
2852 if (kvm_device_ops_table[type] != NULL)
2855 kvm_device_ops_table[type] = ops;
2859 void kvm_unregister_device_ops(u32 type)
2861 if (kvm_device_ops_table[type] != NULL)
2862 kvm_device_ops_table[type] = NULL;
2865 static int kvm_ioctl_create_device(struct kvm *kvm,
2866 struct kvm_create_device *cd)
2868 struct kvm_device_ops *ops = NULL;
2869 struct kvm_device *dev;
2870 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2873 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
2876 ops = kvm_device_ops_table[cd->type];
2883 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2890 mutex_lock(&kvm->lock);
2891 ret = ops->create(dev, cd->type);
2893 mutex_unlock(&kvm->lock);
2897 list_add(&dev->vm_node, &kvm->devices);
2898 mutex_unlock(&kvm->lock);
2903 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2905 mutex_lock(&kvm->lock);
2906 list_del(&dev->vm_node);
2907 mutex_unlock(&kvm->lock);
2917 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2920 case KVM_CAP_USER_MEMORY:
2921 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2922 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2923 case KVM_CAP_INTERNAL_ERROR_DATA:
2924 #ifdef CONFIG_HAVE_KVM_MSI
2925 case KVM_CAP_SIGNAL_MSI:
2927 #ifdef CONFIG_HAVE_KVM_IRQFD
2929 case KVM_CAP_IRQFD_RESAMPLE:
2931 case KVM_CAP_IOEVENTFD_ANY_LENGTH:
2932 case KVM_CAP_CHECK_EXTENSION_VM:
2934 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2935 case KVM_CAP_IRQ_ROUTING:
2936 return KVM_MAX_IRQ_ROUTES;
2938 #if KVM_ADDRESS_SPACE_NUM > 1
2939 case KVM_CAP_MULTI_ADDRESS_SPACE:
2940 return KVM_ADDRESS_SPACE_NUM;
2942 case KVM_CAP_MAX_VCPU_ID:
2943 return KVM_MAX_VCPU_ID;
2947 return kvm_vm_ioctl_check_extension(kvm, arg);
2950 static long kvm_vm_ioctl(struct file *filp,
2951 unsigned int ioctl, unsigned long arg)
2953 struct kvm *kvm = filp->private_data;
2954 void __user *argp = (void __user *)arg;
2957 if (kvm->mm != current->mm)
2960 case KVM_CREATE_VCPU:
2961 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2963 case KVM_SET_USER_MEMORY_REGION: {
2964 struct kvm_userspace_memory_region kvm_userspace_mem;
2967 if (copy_from_user(&kvm_userspace_mem, argp,
2968 sizeof(kvm_userspace_mem)))
2971 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2974 case KVM_GET_DIRTY_LOG: {
2975 struct kvm_dirty_log log;
2978 if (copy_from_user(&log, argp, sizeof(log)))
2980 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2983 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2984 case KVM_REGISTER_COALESCED_MMIO: {
2985 struct kvm_coalesced_mmio_zone zone;
2988 if (copy_from_user(&zone, argp, sizeof(zone)))
2990 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2993 case KVM_UNREGISTER_COALESCED_MMIO: {
2994 struct kvm_coalesced_mmio_zone zone;
2997 if (copy_from_user(&zone, argp, sizeof(zone)))
2999 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
3004 struct kvm_irqfd data;
3007 if (copy_from_user(&data, argp, sizeof(data)))
3009 r = kvm_irqfd(kvm, &data);
3012 case KVM_IOEVENTFD: {
3013 struct kvm_ioeventfd data;
3016 if (copy_from_user(&data, argp, sizeof(data)))
3018 r = kvm_ioeventfd(kvm, &data);
3021 #ifdef CONFIG_HAVE_KVM_MSI
3022 case KVM_SIGNAL_MSI: {
3026 if (copy_from_user(&msi, argp, sizeof(msi)))
3028 r = kvm_send_userspace_msi(kvm, &msi);
3032 #ifdef __KVM_HAVE_IRQ_LINE
3033 case KVM_IRQ_LINE_STATUS:
3034 case KVM_IRQ_LINE: {
3035 struct kvm_irq_level irq_event;
3038 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
3041 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
3042 ioctl == KVM_IRQ_LINE_STATUS);
3047 if (ioctl == KVM_IRQ_LINE_STATUS) {
3048 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
3056 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3057 case KVM_SET_GSI_ROUTING: {
3058 struct kvm_irq_routing routing;
3059 struct kvm_irq_routing __user *urouting;
3060 struct kvm_irq_routing_entry *entries = NULL;
3063 if (copy_from_user(&routing, argp, sizeof(routing)))
3066 if (routing.nr > KVM_MAX_IRQ_ROUTES)
3072 entries = vmalloc(routing.nr * sizeof(*entries));
3077 if (copy_from_user(entries, urouting->entries,
3078 routing.nr * sizeof(*entries)))
3079 goto out_free_irq_routing;
3081 r = kvm_set_irq_routing(kvm, entries, routing.nr,
3083 out_free_irq_routing:
3087 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
3088 case KVM_CREATE_DEVICE: {
3089 struct kvm_create_device cd;
3092 if (copy_from_user(&cd, argp, sizeof(cd)))
3095 r = kvm_ioctl_create_device(kvm, &cd);
3100 if (copy_to_user(argp, &cd, sizeof(cd)))
3106 case KVM_CHECK_EXTENSION:
3107 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
3110 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
3116 #ifdef CONFIG_KVM_COMPAT
3117 struct compat_kvm_dirty_log {
3121 compat_uptr_t dirty_bitmap; /* one bit per page */
3126 static long kvm_vm_compat_ioctl(struct file *filp,
3127 unsigned int ioctl, unsigned long arg)
3129 struct kvm *kvm = filp->private_data;
3132 if (kvm->mm != current->mm)
3135 case KVM_GET_DIRTY_LOG: {
3136 struct compat_kvm_dirty_log compat_log;
3137 struct kvm_dirty_log log;
3139 if (copy_from_user(&compat_log, (void __user *)arg,
3140 sizeof(compat_log)))
3142 log.slot = compat_log.slot;
3143 log.padding1 = compat_log.padding1;
3144 log.padding2 = compat_log.padding2;
3145 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
3147 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3151 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, "kvm/cpu: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);