2 * Kernel-based Virtual Machine driver for Linux
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
7 * Copyright (C) 2006 Qumranet, Inc.
8 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
11 * Avi Kivity <avi@qumranet.com>
12 * Yaniv Kamay <yaniv@qumranet.com>
14 * This work is licensed under the terms of the GNU GPL, version 2. See
15 * the COPYING file in the top-level directory.
19 #include <kvm/iodev.h>
21 #include <linux/kvm_host.h>
22 #include <linux/kvm.h>
23 #include <linux/module.h>
24 #include <linux/errno.h>
25 #include <linux/percpu.h>
27 #include <linux/miscdevice.h>
28 #include <linux/vmalloc.h>
29 #include <linux/reboot.h>
30 #include <linux/debugfs.h>
31 #include <linux/highmem.h>
32 #include <linux/file.h>
33 #include <linux/syscore_ops.h>
34 #include <linux/cpu.h>
35 #include <linux/sched/signal.h>
36 #include <linux/sched/mm.h>
37 #include <linux/sched/stat.h>
38 #include <linux/cpumask.h>
39 #include <linux/smp.h>
40 #include <linux/anon_inodes.h>
41 #include <linux/profile.h>
42 #include <linux/kvm_para.h>
43 #include <linux/pagemap.h>
44 #include <linux/mman.h>
45 #include <linux/swap.h>
46 #include <linux/bitops.h>
47 #include <linux/spinlock.h>
48 #include <linux/compat.h>
49 #include <linux/srcu.h>
50 #include <linux/hugetlb.h>
51 #include <linux/slab.h>
52 #include <linux/sort.h>
53 #include <linux/bsearch.h>
55 #include <asm/processor.h>
57 #include <asm/ioctl.h>
58 #include <linux/uaccess.h>
59 #include <asm/pgtable.h>
61 #include "coalesced_mmio.h"
65 #define CREATE_TRACE_POINTS
66 #include <trace/events/kvm.h>
68 /* Worst case buffer size needed for holding an integer. */
69 #define ITOA_MAX_LEN 12
71 MODULE_AUTHOR("Qumranet");
72 MODULE_LICENSE("GPL");
74 /* Architectures should define their poll value according to the halt latency */
75 unsigned int halt_poll_ns = KVM_HALT_POLL_NS_DEFAULT;
76 module_param(halt_poll_ns, uint, 0644);
77 EXPORT_SYMBOL_GPL(halt_poll_ns);
79 /* Default doubles per-vcpu halt_poll_ns. */
80 unsigned int halt_poll_ns_grow = 2;
81 module_param(halt_poll_ns_grow, uint, 0644);
82 EXPORT_SYMBOL_GPL(halt_poll_ns_grow);
84 /* Default resets per-vcpu halt_poll_ns . */
85 unsigned int halt_poll_ns_shrink;
86 module_param(halt_poll_ns_shrink, uint, 0644);
87 EXPORT_SYMBOL_GPL(halt_poll_ns_shrink);
92 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
95 DEFINE_SPINLOCK(kvm_lock);
96 static DEFINE_RAW_SPINLOCK(kvm_count_lock);
99 static cpumask_var_t cpus_hardware_enabled;
100 static int kvm_usage_count;
101 static atomic_t hardware_enable_failed;
103 struct kmem_cache *kvm_vcpu_cache;
104 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
106 static __read_mostly struct preempt_ops kvm_preempt_ops;
108 struct dentry *kvm_debugfs_dir;
109 EXPORT_SYMBOL_GPL(kvm_debugfs_dir);
111 static int kvm_debugfs_num_entries;
112 static const struct file_operations *stat_fops_per_vm[];
114 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
116 #ifdef CONFIG_KVM_COMPAT
117 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
120 static int hardware_enable_all(void);
121 static void hardware_disable_all(void);
123 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
125 static void 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 #define KVM_EVENT_CREATE_VM 0
133 #define KVM_EVENT_DESTROY_VM 1
134 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm);
135 static unsigned long long kvm_createvm_count;
136 static unsigned long long kvm_active_vms;
138 __weak void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
139 unsigned long start, unsigned long end)
143 bool kvm_is_reserved_pfn(kvm_pfn_t pfn)
146 return PageReserved(pfn_to_page(pfn));
152 * Switches to specified vcpu, until a matching vcpu_put()
154 int vcpu_load(struct kvm_vcpu *vcpu)
158 if (mutex_lock_killable(&vcpu->mutex))
161 preempt_notifier_register(&vcpu->preempt_notifier);
162 kvm_arch_vcpu_load(vcpu, cpu);
166 EXPORT_SYMBOL_GPL(vcpu_load);
168 void vcpu_put(struct kvm_vcpu *vcpu)
171 kvm_arch_vcpu_put(vcpu);
172 preempt_notifier_unregister(&vcpu->preempt_notifier);
174 mutex_unlock(&vcpu->mutex);
176 EXPORT_SYMBOL_GPL(vcpu_put);
178 /* TODO: merge with kvm_arch_vcpu_should_kick */
179 static bool kvm_request_needs_ipi(struct kvm_vcpu *vcpu, unsigned req)
181 int mode = kvm_vcpu_exiting_guest_mode(vcpu);
184 * We need to wait for the VCPU to reenable interrupts and get out of
185 * READING_SHADOW_PAGE_TABLES mode.
187 if (req & KVM_REQUEST_WAIT)
188 return mode != OUTSIDE_GUEST_MODE;
191 * Need to kick a running VCPU, but otherwise there is nothing to do.
193 return mode == IN_GUEST_MODE;
196 static void ack_flush(void *_completed)
200 static inline bool kvm_kick_many_cpus(const struct cpumask *cpus, bool wait)
203 cpus = cpu_online_mask;
205 if (cpumask_empty(cpus))
208 smp_call_function_many(cpus, ack_flush, NULL, wait);
212 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
217 struct kvm_vcpu *vcpu;
219 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
222 kvm_for_each_vcpu(i, vcpu, kvm) {
223 kvm_make_request(req, vcpu);
226 if (!(req & KVM_REQUEST_NO_WAKEUP) && kvm_vcpu_wake_up(vcpu))
229 if (cpus != NULL && cpu != -1 && cpu != me &&
230 kvm_request_needs_ipi(vcpu, req))
231 __cpumask_set_cpu(cpu, cpus);
233 called = kvm_kick_many_cpus(cpus, !!(req & KVM_REQUEST_WAIT));
235 free_cpumask_var(cpus);
239 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
240 void kvm_flush_remote_tlbs(struct kvm *kvm)
243 * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in
244 * kvm_make_all_cpus_request.
246 long dirty_count = smp_load_acquire(&kvm->tlbs_dirty);
249 * We want to publish modifications to the page tables before reading
250 * mode. Pairs with a memory barrier in arch-specific code.
251 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
252 * and smp_mb in walk_shadow_page_lockless_begin/end.
253 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
255 * There is already an smp_mb__after_atomic() before
256 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
259 if (kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
260 ++kvm->stat.remote_tlb_flush;
261 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
263 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
266 void kvm_reload_remote_mmus(struct kvm *kvm)
268 kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
271 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
276 mutex_init(&vcpu->mutex);
281 init_swait_queue_head(&vcpu->wq);
282 kvm_async_pf_vcpu_init(vcpu);
285 INIT_LIST_HEAD(&vcpu->blocked_vcpu_list);
287 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
292 vcpu->run = page_address(page);
294 kvm_vcpu_set_in_spin_loop(vcpu, false);
295 kvm_vcpu_set_dy_eligible(vcpu, false);
296 vcpu->preempted = false;
298 r = kvm_arch_vcpu_init(vcpu);
304 free_page((unsigned long)vcpu->run);
308 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
310 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
313 * no need for rcu_read_lock as VCPU_RUN is the only place that
314 * will change the vcpu->pid pointer and on uninit all file
315 * descriptors are already gone.
317 put_pid(rcu_dereference_protected(vcpu->pid, 1));
318 kvm_arch_vcpu_uninit(vcpu);
319 free_page((unsigned long)vcpu->run);
321 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
323 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
324 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
326 return container_of(mn, struct kvm, mmu_notifier);
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);
369 kvm_arch_mmu_notifier_invalidate_range(kvm, start, end);
371 srcu_read_unlock(&kvm->srcu, idx);
374 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
375 struct mm_struct *mm,
379 struct kvm *kvm = mmu_notifier_to_kvm(mn);
381 spin_lock(&kvm->mmu_lock);
383 * This sequence increase will notify the kvm page fault that
384 * the page that is going to be mapped in the spte could have
387 kvm->mmu_notifier_seq++;
390 * The above sequence increase must be visible before the
391 * below count decrease, which is ensured by the smp_wmb above
392 * in conjunction with the smp_rmb in mmu_notifier_retry().
394 kvm->mmu_notifier_count--;
395 spin_unlock(&kvm->mmu_lock);
397 BUG_ON(kvm->mmu_notifier_count < 0);
400 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
401 struct mm_struct *mm,
405 struct kvm *kvm = mmu_notifier_to_kvm(mn);
408 idx = srcu_read_lock(&kvm->srcu);
409 spin_lock(&kvm->mmu_lock);
411 young = kvm_age_hva(kvm, start, end);
413 kvm_flush_remote_tlbs(kvm);
415 spin_unlock(&kvm->mmu_lock);
416 srcu_read_unlock(&kvm->srcu, idx);
421 static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
422 struct mm_struct *mm,
426 struct kvm *kvm = mmu_notifier_to_kvm(mn);
429 idx = srcu_read_lock(&kvm->srcu);
430 spin_lock(&kvm->mmu_lock);
432 * Even though we do not flush TLB, this will still adversely
433 * affect performance on pre-Haswell Intel EPT, where there is
434 * no EPT Access Bit to clear so that we have to tear down EPT
435 * tables instead. If we find this unacceptable, we can always
436 * add a parameter to kvm_age_hva so that it effectively doesn't
437 * do anything on clear_young.
439 * Also note that currently we never issue secondary TLB flushes
440 * from clear_young, leaving this job up to the regular system
441 * cadence. If we find this inaccurate, we might come up with a
442 * more sophisticated heuristic later.
444 young = kvm_age_hva(kvm, start, end);
445 spin_unlock(&kvm->mmu_lock);
446 srcu_read_unlock(&kvm->srcu, idx);
451 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
452 struct mm_struct *mm,
453 unsigned long address)
455 struct kvm *kvm = mmu_notifier_to_kvm(mn);
458 idx = srcu_read_lock(&kvm->srcu);
459 spin_lock(&kvm->mmu_lock);
460 young = kvm_test_age_hva(kvm, address);
461 spin_unlock(&kvm->mmu_lock);
462 srcu_read_unlock(&kvm->srcu, idx);
467 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
468 struct mm_struct *mm)
470 struct kvm *kvm = mmu_notifier_to_kvm(mn);
473 idx = srcu_read_lock(&kvm->srcu);
474 kvm_arch_flush_shadow_all(kvm);
475 srcu_read_unlock(&kvm->srcu, idx);
478 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
479 .flags = MMU_INVALIDATE_DOES_NOT_BLOCK,
480 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
481 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
482 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
483 .clear_young = kvm_mmu_notifier_clear_young,
484 .test_young = kvm_mmu_notifier_test_young,
485 .change_pte = kvm_mmu_notifier_change_pte,
486 .release = kvm_mmu_notifier_release,
489 static int kvm_init_mmu_notifier(struct kvm *kvm)
491 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
492 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
495 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
497 static int kvm_init_mmu_notifier(struct kvm *kvm)
502 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
504 static struct kvm_memslots *kvm_alloc_memslots(void)
507 struct kvm_memslots *slots;
509 slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
513 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
514 slots->id_to_index[i] = slots->memslots[i].id = i;
519 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
521 if (!memslot->dirty_bitmap)
524 kvfree(memslot->dirty_bitmap);
525 memslot->dirty_bitmap = NULL;
529 * Free any memory in @free but not in @dont.
531 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
532 struct kvm_memory_slot *dont)
534 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
535 kvm_destroy_dirty_bitmap(free);
537 kvm_arch_free_memslot(kvm, free, dont);
542 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
544 struct kvm_memory_slot *memslot;
549 kvm_for_each_memslot(memslot, slots)
550 kvm_free_memslot(kvm, memslot, NULL);
555 static void kvm_destroy_vm_debugfs(struct kvm *kvm)
559 if (!kvm->debugfs_dentry)
562 debugfs_remove_recursive(kvm->debugfs_dentry);
564 if (kvm->debugfs_stat_data) {
565 for (i = 0; i < kvm_debugfs_num_entries; i++)
566 kfree(kvm->debugfs_stat_data[i]);
567 kfree(kvm->debugfs_stat_data);
571 static int kvm_create_vm_debugfs(struct kvm *kvm, int fd)
573 char dir_name[ITOA_MAX_LEN * 2];
574 struct kvm_stat_data *stat_data;
575 struct kvm_stats_debugfs_item *p;
577 if (!debugfs_initialized())
580 snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd);
581 kvm->debugfs_dentry = debugfs_create_dir(dir_name,
583 if (!kvm->debugfs_dentry)
586 kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries,
587 sizeof(*kvm->debugfs_stat_data),
589 if (!kvm->debugfs_stat_data)
592 for (p = debugfs_entries; p->name; p++) {
593 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL);
597 stat_data->kvm = kvm;
598 stat_data->offset = p->offset;
599 kvm->debugfs_stat_data[p - debugfs_entries] = stat_data;
600 if (!debugfs_create_file(p->name, 0644,
603 stat_fops_per_vm[p->kind]))
609 static struct kvm *kvm_create_vm(unsigned long type)
612 struct kvm *kvm = kvm_arch_alloc_vm();
615 return ERR_PTR(-ENOMEM);
617 spin_lock_init(&kvm->mmu_lock);
619 kvm->mm = current->mm;
620 kvm_eventfd_init(kvm);
621 mutex_init(&kvm->lock);
622 mutex_init(&kvm->irq_lock);
623 mutex_init(&kvm->slots_lock);
624 refcount_set(&kvm->users_count, 1);
625 INIT_LIST_HEAD(&kvm->devices);
627 r = kvm_arch_init_vm(kvm, type);
629 goto out_err_no_disable;
631 r = hardware_enable_all();
633 goto out_err_no_disable;
635 #ifdef CONFIG_HAVE_KVM_IRQFD
636 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
639 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
642 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
643 struct kvm_memslots *slots = kvm_alloc_memslots();
645 goto out_err_no_srcu;
647 * Generations must be different for each address space.
648 * Init kvm generation close to the maximum to easily test the
649 * code of handling generation number wrap-around.
651 slots->generation = i * 2 - 150;
652 rcu_assign_pointer(kvm->memslots[i], slots);
655 if (init_srcu_struct(&kvm->srcu))
656 goto out_err_no_srcu;
657 if (init_srcu_struct(&kvm->irq_srcu))
658 goto out_err_no_irq_srcu;
659 for (i = 0; i < KVM_NR_BUSES; i++) {
660 rcu_assign_pointer(kvm->buses[i],
661 kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL));
666 r = kvm_init_mmu_notifier(kvm);
670 spin_lock(&kvm_lock);
671 list_add(&kvm->vm_list, &vm_list);
672 spin_unlock(&kvm_lock);
674 preempt_notifier_inc();
679 cleanup_srcu_struct(&kvm->irq_srcu);
681 cleanup_srcu_struct(&kvm->srcu);
683 hardware_disable_all();
685 refcount_set(&kvm->users_count, 0);
686 for (i = 0; i < KVM_NR_BUSES; i++)
687 kfree(kvm_get_bus(kvm, i));
688 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
689 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
690 kvm_arch_free_vm(kvm);
695 static void kvm_destroy_devices(struct kvm *kvm)
697 struct kvm_device *dev, *tmp;
700 * We do not need to take the kvm->lock here, because nobody else
701 * has a reference to the struct kvm at this point and therefore
702 * cannot access the devices list anyhow.
704 list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
705 list_del(&dev->vm_node);
706 dev->ops->destroy(dev);
710 static void kvm_destroy_vm(struct kvm *kvm)
713 struct mm_struct *mm = kvm->mm;
715 kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm);
716 kvm_destroy_vm_debugfs(kvm);
717 kvm_arch_sync_events(kvm);
718 spin_lock(&kvm_lock);
719 list_del(&kvm->vm_list);
720 spin_unlock(&kvm_lock);
721 kvm_free_irq_routing(kvm);
722 for (i = 0; i < KVM_NR_BUSES; i++) {
723 struct kvm_io_bus *bus = kvm_get_bus(kvm, i);
726 kvm_io_bus_destroy(bus);
727 kvm->buses[i] = NULL;
729 kvm_coalesced_mmio_free(kvm);
730 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
731 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
733 kvm_arch_flush_shadow_all(kvm);
735 kvm_arch_destroy_vm(kvm);
736 kvm_destroy_devices(kvm);
737 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
738 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
739 cleanup_srcu_struct(&kvm->irq_srcu);
740 cleanup_srcu_struct(&kvm->srcu);
741 kvm_arch_free_vm(kvm);
742 preempt_notifier_dec();
743 hardware_disable_all();
747 void kvm_get_kvm(struct kvm *kvm)
749 refcount_inc(&kvm->users_count);
751 EXPORT_SYMBOL_GPL(kvm_get_kvm);
753 void kvm_put_kvm(struct kvm *kvm)
755 if (refcount_dec_and_test(&kvm->users_count))
758 EXPORT_SYMBOL_GPL(kvm_put_kvm);
761 static int kvm_vm_release(struct inode *inode, struct file *filp)
763 struct kvm *kvm = filp->private_data;
765 kvm_irqfd_release(kvm);
772 * Allocation size is twice as large as the actual dirty bitmap size.
773 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
775 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
777 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
779 memslot->dirty_bitmap = kvzalloc(dirty_bytes, GFP_KERNEL);
780 if (!memslot->dirty_bitmap)
787 * Insert memslot and re-sort memslots based on their GFN,
788 * so binary search could be used to lookup GFN.
789 * Sorting algorithm takes advantage of having initially
790 * sorted array and known changed memslot position.
792 static void update_memslots(struct kvm_memslots *slots,
793 struct kvm_memory_slot *new)
796 int i = slots->id_to_index[id];
797 struct kvm_memory_slot *mslots = slots->memslots;
799 WARN_ON(mslots[i].id != id);
801 WARN_ON(!mslots[i].npages);
802 if (mslots[i].npages)
805 if (!mslots[i].npages)
809 while (i < KVM_MEM_SLOTS_NUM - 1 &&
810 new->base_gfn <= mslots[i + 1].base_gfn) {
811 if (!mslots[i + 1].npages)
813 mslots[i] = mslots[i + 1];
814 slots->id_to_index[mslots[i].id] = i;
819 * The ">=" is needed when creating a slot with base_gfn == 0,
820 * so that it moves before all those with base_gfn == npages == 0.
822 * On the other hand, if new->npages is zero, the above loop has
823 * already left i pointing to the beginning of the empty part of
824 * mslots, and the ">=" would move the hole backwards in this
825 * case---which is wrong. So skip the loop when deleting a slot.
829 new->base_gfn >= mslots[i - 1].base_gfn) {
830 mslots[i] = mslots[i - 1];
831 slots->id_to_index[mslots[i].id] = i;
835 WARN_ON_ONCE(i != slots->used_slots);
838 slots->id_to_index[mslots[i].id] = i;
841 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
843 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
845 #ifdef __KVM_HAVE_READONLY_MEM
846 valid_flags |= KVM_MEM_READONLY;
849 if (mem->flags & ~valid_flags)
855 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
856 int as_id, struct kvm_memslots *slots)
858 struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id);
861 * Set the low bit in the generation, which disables SPTE caching
862 * until the end of synchronize_srcu_expedited.
864 WARN_ON(old_memslots->generation & 1);
865 slots->generation = old_memslots->generation + 1;
867 rcu_assign_pointer(kvm->memslots[as_id], slots);
868 synchronize_srcu_expedited(&kvm->srcu);
871 * Increment the new memslot generation a second time. This prevents
872 * vm exits that race with memslot updates from caching a memslot
873 * generation that will (potentially) be valid forever.
875 * Generations must be unique even across address spaces. We do not need
876 * a global counter for that, instead the generation space is evenly split
877 * across address spaces. For example, with two address spaces, address
878 * space 0 will use generations 0, 4, 8, ... while * address space 1 will
879 * use generations 2, 6, 10, 14, ...
881 slots->generation += KVM_ADDRESS_SPACE_NUM * 2 - 1;
883 kvm_arch_memslots_updated(kvm, slots);
889 * Allocate some memory and give it an address in the guest physical address
892 * Discontiguous memory is allowed, mostly for framebuffers.
894 * Must be called holding kvm->slots_lock for write.
896 int __kvm_set_memory_region(struct kvm *kvm,
897 const struct kvm_userspace_memory_region *mem)
901 unsigned long npages;
902 struct kvm_memory_slot *slot;
903 struct kvm_memory_slot old, new;
904 struct kvm_memslots *slots = NULL, *old_memslots;
906 enum kvm_mr_change change;
908 r = check_memory_region_flags(mem);
913 as_id = mem->slot >> 16;
916 /* General sanity checks */
917 if (mem->memory_size & (PAGE_SIZE - 1))
919 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
921 /* We can read the guest memory with __xxx_user() later on. */
922 if ((id < KVM_USER_MEM_SLOTS) &&
923 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
924 !access_ok(VERIFY_WRITE,
925 (void __user *)(unsigned long)mem->userspace_addr,
928 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
930 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
933 slot = id_to_memslot(__kvm_memslots(kvm, as_id), id);
934 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
935 npages = mem->memory_size >> PAGE_SHIFT;
937 if (npages > KVM_MEM_MAX_NR_PAGES)
943 new.base_gfn = base_gfn;
945 new.flags = mem->flags;
949 change = KVM_MR_CREATE;
950 else { /* Modify an existing slot. */
951 if ((mem->userspace_addr != old.userspace_addr) ||
952 (npages != old.npages) ||
953 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
956 if (base_gfn != old.base_gfn)
957 change = KVM_MR_MOVE;
958 else if (new.flags != old.flags)
959 change = KVM_MR_FLAGS_ONLY;
960 else { /* Nothing to change. */
969 change = KVM_MR_DELETE;
974 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
975 /* Check for overlaps */
977 kvm_for_each_memslot(slot, __kvm_memslots(kvm, as_id)) {
978 if ((slot->id >= KVM_USER_MEM_SLOTS) ||
981 if (!((base_gfn + npages <= slot->base_gfn) ||
982 (base_gfn >= slot->base_gfn + slot->npages)))
987 /* Free page dirty bitmap if unneeded */
988 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
989 new.dirty_bitmap = NULL;
992 if (change == KVM_MR_CREATE) {
993 new.userspace_addr = mem->userspace_addr;
995 if (kvm_arch_create_memslot(kvm, &new, npages))
999 /* Allocate page dirty bitmap if needed */
1000 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
1001 if (kvm_create_dirty_bitmap(&new) < 0)
1005 slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
1008 memcpy(slots, __kvm_memslots(kvm, as_id), sizeof(struct kvm_memslots));
1010 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
1011 slot = id_to_memslot(slots, id);
1012 slot->flags |= KVM_MEMSLOT_INVALID;
1014 old_memslots = install_new_memslots(kvm, as_id, slots);
1016 /* From this point no new shadow pages pointing to a deleted,
1017 * or moved, memslot will be created.
1019 * validation of sp->gfn happens in:
1020 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1021 * - kvm_is_visible_gfn (mmu_check_roots)
1023 kvm_arch_flush_shadow_memslot(kvm, slot);
1026 * We can re-use the old_memslots from above, the only difference
1027 * from the currently installed memslots is the invalid flag. This
1028 * will get overwritten by update_memslots anyway.
1030 slots = old_memslots;
1033 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
1037 /* actual memory is freed via old in kvm_free_memslot below */
1038 if (change == KVM_MR_DELETE) {
1039 new.dirty_bitmap = NULL;
1040 memset(&new.arch, 0, sizeof(new.arch));
1043 update_memslots(slots, &new);
1044 old_memslots = install_new_memslots(kvm, as_id, slots);
1046 kvm_arch_commit_memory_region(kvm, mem, &old, &new, change);
1048 kvm_free_memslot(kvm, &old, &new);
1049 kvfree(old_memslots);
1055 kvm_free_memslot(kvm, &new, &old);
1059 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1061 int kvm_set_memory_region(struct kvm *kvm,
1062 const struct kvm_userspace_memory_region *mem)
1066 mutex_lock(&kvm->slots_lock);
1067 r = __kvm_set_memory_region(kvm, mem);
1068 mutex_unlock(&kvm->slots_lock);
1071 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1073 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1074 struct kvm_userspace_memory_region *mem)
1076 if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1079 return kvm_set_memory_region(kvm, mem);
1082 int kvm_get_dirty_log(struct kvm *kvm,
1083 struct kvm_dirty_log *log, int *is_dirty)
1085 struct kvm_memslots *slots;
1086 struct kvm_memory_slot *memslot;
1089 unsigned long any = 0;
1091 as_id = log->slot >> 16;
1092 id = (u16)log->slot;
1093 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1096 slots = __kvm_memslots(kvm, as_id);
1097 memslot = id_to_memslot(slots, id);
1098 if (!memslot->dirty_bitmap)
1101 n = kvm_dirty_bitmap_bytes(memslot);
1103 for (i = 0; !any && i < n/sizeof(long); ++i)
1104 any = memslot->dirty_bitmap[i];
1106 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1113 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1115 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1117 * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages
1118 * are dirty write protect them for next write.
1119 * @kvm: pointer to kvm instance
1120 * @log: slot id and address to which we copy the log
1121 * @is_dirty: flag set if any page is dirty
1123 * We need to keep it in mind that VCPU threads can write to the bitmap
1124 * concurrently. So, to avoid losing track of dirty pages we keep the
1127 * 1. Take a snapshot of the bit and clear it if needed.
1128 * 2. Write protect the corresponding page.
1129 * 3. Copy the snapshot to the userspace.
1130 * 4. Upon return caller flushes TLB's if needed.
1132 * Between 2 and 4, the guest may write to the page using the remaining TLB
1133 * entry. This is not a problem because the page is reported dirty using
1134 * the snapshot taken before and step 4 ensures that writes done after
1135 * exiting to userspace will be logged for the next call.
1138 int kvm_get_dirty_log_protect(struct kvm *kvm,
1139 struct kvm_dirty_log *log, bool *is_dirty)
1141 struct kvm_memslots *slots;
1142 struct kvm_memory_slot *memslot;
1145 unsigned long *dirty_bitmap;
1146 unsigned long *dirty_bitmap_buffer;
1148 as_id = log->slot >> 16;
1149 id = (u16)log->slot;
1150 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1153 slots = __kvm_memslots(kvm, as_id);
1154 memslot = id_to_memslot(slots, id);
1156 dirty_bitmap = memslot->dirty_bitmap;
1160 n = kvm_dirty_bitmap_bytes(memslot);
1162 dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long);
1163 memset(dirty_bitmap_buffer, 0, n);
1165 spin_lock(&kvm->mmu_lock);
1167 for (i = 0; i < n / sizeof(long); i++) {
1171 if (!dirty_bitmap[i])
1176 mask = xchg(&dirty_bitmap[i], 0);
1177 dirty_bitmap_buffer[i] = mask;
1180 offset = i * BITS_PER_LONG;
1181 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1186 spin_unlock(&kvm->mmu_lock);
1187 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1191 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect);
1194 bool kvm_largepages_enabled(void)
1196 return largepages_enabled;
1199 void kvm_disable_largepages(void)
1201 largepages_enabled = false;
1203 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1205 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1207 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1209 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1211 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
1213 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn);
1216 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1218 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1220 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1221 memslot->flags & KVM_MEMSLOT_INVALID)
1226 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1228 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1230 struct vm_area_struct *vma;
1231 unsigned long addr, size;
1235 addr = gfn_to_hva(kvm, gfn);
1236 if (kvm_is_error_hva(addr))
1239 down_read(¤t->mm->mmap_sem);
1240 vma = find_vma(current->mm, addr);
1244 size = vma_kernel_pagesize(vma);
1247 up_read(¤t->mm->mmap_sem);
1252 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1254 return slot->flags & KVM_MEM_READONLY;
1257 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1258 gfn_t *nr_pages, bool write)
1260 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1261 return KVM_HVA_ERR_BAD;
1263 if (memslot_is_readonly(slot) && write)
1264 return KVM_HVA_ERR_RO_BAD;
1267 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1269 return __gfn_to_hva_memslot(slot, gfn);
1272 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1275 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1278 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1281 return gfn_to_hva_many(slot, gfn, NULL);
1283 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1285 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1287 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1289 EXPORT_SYMBOL_GPL(gfn_to_hva);
1291 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
1293 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
1295 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
1298 * If writable is set to false, the hva returned by this function is only
1299 * allowed to be read.
1301 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1302 gfn_t gfn, bool *writable)
1304 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1306 if (!kvm_is_error_hva(hva) && writable)
1307 *writable = !memslot_is_readonly(slot);
1312 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1314 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1316 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1319 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
1321 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1323 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1326 static inline int check_user_page_hwpoison(unsigned long addr)
1328 int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
1330 rc = get_user_pages(addr, 1, flags, NULL, NULL);
1331 return rc == -EHWPOISON;
1335 * The atomic path to get the writable pfn which will be stored in @pfn,
1336 * true indicates success, otherwise false is returned.
1338 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1339 bool write_fault, bool *writable, kvm_pfn_t *pfn)
1341 struct page *page[1];
1344 if (!(async || atomic))
1348 * Fast pin a writable pfn only if it is a write fault request
1349 * or the caller allows to map a writable pfn for a read fault
1352 if (!(write_fault || writable))
1355 npages = __get_user_pages_fast(addr, 1, 1, page);
1357 *pfn = page_to_pfn(page[0]);
1368 * The slow path to get the pfn of the specified host virtual address,
1369 * 1 indicates success, -errno is returned if error is detected.
1371 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1372 bool *writable, kvm_pfn_t *pfn)
1374 unsigned int flags = FOLL_HWPOISON;
1381 *writable = write_fault;
1384 flags |= FOLL_WRITE;
1386 flags |= FOLL_NOWAIT;
1388 npages = get_user_pages_unlocked(addr, 1, &page, flags);
1392 /* map read fault as writable if possible */
1393 if (unlikely(!write_fault) && writable) {
1396 if (__get_user_pages_fast(addr, 1, 1, &wpage) == 1) {
1402 *pfn = page_to_pfn(page);
1406 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1408 if (unlikely(!(vma->vm_flags & VM_READ)))
1411 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1417 static int hva_to_pfn_remapped(struct vm_area_struct *vma,
1418 unsigned long addr, bool *async,
1419 bool write_fault, kvm_pfn_t *p_pfn)
1424 r = follow_pfn(vma, addr, &pfn);
1427 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
1428 * not call the fault handler, so do it here.
1430 bool unlocked = false;
1431 r = fixup_user_fault(current, current->mm, addr,
1432 (write_fault ? FAULT_FLAG_WRITE : 0),
1439 r = follow_pfn(vma, addr, &pfn);
1447 * Get a reference here because callers of *hva_to_pfn* and
1448 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
1449 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
1450 * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will
1451 * simply do nothing for reserved pfns.
1453 * Whoever called remap_pfn_range is also going to call e.g.
1454 * unmap_mapping_range before the underlying pages are freed,
1455 * causing a call to our MMU notifier.
1464 * Pin guest page in memory and return its pfn.
1465 * @addr: host virtual address which maps memory to the guest
1466 * @atomic: whether this function can sleep
1467 * @async: whether this function need to wait IO complete if the
1468 * host page is not in the memory
1469 * @write_fault: whether we should get a writable host page
1470 * @writable: whether it allows to map a writable host page for !@write_fault
1472 * The function will map a writable host page for these two cases:
1473 * 1): @write_fault = true
1474 * 2): @write_fault = false && @writable, @writable will tell the caller
1475 * whether the mapping is writable.
1477 static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1478 bool write_fault, bool *writable)
1480 struct vm_area_struct *vma;
1484 /* we can do it either atomically or asynchronously, not both */
1485 BUG_ON(atomic && async);
1487 if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1491 return KVM_PFN_ERR_FAULT;
1493 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1497 down_read(¤t->mm->mmap_sem);
1498 if (npages == -EHWPOISON ||
1499 (!async && check_user_page_hwpoison(addr))) {
1500 pfn = KVM_PFN_ERR_HWPOISON;
1505 vma = find_vma_intersection(current->mm, addr, addr + 1);
1508 pfn = KVM_PFN_ERR_FAULT;
1509 else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
1510 r = hva_to_pfn_remapped(vma, addr, async, write_fault, &pfn);
1514 pfn = KVM_PFN_ERR_FAULT;
1516 if (async && vma_is_valid(vma, write_fault))
1518 pfn = KVM_PFN_ERR_FAULT;
1521 up_read(¤t->mm->mmap_sem);
1525 kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn,
1526 bool atomic, bool *async, bool write_fault,
1529 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1531 if (addr == KVM_HVA_ERR_RO_BAD) {
1534 return KVM_PFN_ERR_RO_FAULT;
1537 if (kvm_is_error_hva(addr)) {
1540 return KVM_PFN_NOSLOT;
1543 /* Do not map writable pfn in the readonly memslot. */
1544 if (writable && memslot_is_readonly(slot)) {
1549 return hva_to_pfn(addr, atomic, async, write_fault,
1552 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
1554 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1557 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
1558 write_fault, writable);
1560 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1562 kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1564 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1566 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
1568 kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1570 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1572 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1574 kvm_pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1576 return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm, gfn), gfn);
1578 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1580 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
1582 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1584 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
1586 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1588 return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
1590 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1592 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
1594 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1596 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
1598 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1599 struct page **pages, int nr_pages)
1604 addr = gfn_to_hva_many(slot, gfn, &entry);
1605 if (kvm_is_error_hva(addr))
1608 if (entry < nr_pages)
1611 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1613 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1615 static struct page *kvm_pfn_to_page(kvm_pfn_t pfn)
1617 if (is_error_noslot_pfn(pfn))
1618 return KVM_ERR_PTR_BAD_PAGE;
1620 if (kvm_is_reserved_pfn(pfn)) {
1622 return KVM_ERR_PTR_BAD_PAGE;
1625 return pfn_to_page(pfn);
1628 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1632 pfn = gfn_to_pfn(kvm, gfn);
1634 return kvm_pfn_to_page(pfn);
1636 EXPORT_SYMBOL_GPL(gfn_to_page);
1638 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
1642 pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
1644 return kvm_pfn_to_page(pfn);
1646 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
1648 void kvm_release_page_clean(struct page *page)
1650 WARN_ON(is_error_page(page));
1652 kvm_release_pfn_clean(page_to_pfn(page));
1654 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1656 void kvm_release_pfn_clean(kvm_pfn_t pfn)
1658 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
1659 put_page(pfn_to_page(pfn));
1661 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1663 void kvm_release_page_dirty(struct page *page)
1665 WARN_ON(is_error_page(page));
1667 kvm_release_pfn_dirty(page_to_pfn(page));
1669 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1671 void kvm_release_pfn_dirty(kvm_pfn_t pfn)
1673 kvm_set_pfn_dirty(pfn);
1674 kvm_release_pfn_clean(pfn);
1676 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1678 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
1680 if (!kvm_is_reserved_pfn(pfn)) {
1681 struct page *page = pfn_to_page(pfn);
1683 if (!PageReserved(page))
1687 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1689 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
1691 if (!kvm_is_reserved_pfn(pfn))
1692 mark_page_accessed(pfn_to_page(pfn));
1694 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1696 void kvm_get_pfn(kvm_pfn_t pfn)
1698 if (!kvm_is_reserved_pfn(pfn))
1699 get_page(pfn_to_page(pfn));
1701 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1703 static int next_segment(unsigned long len, int offset)
1705 if (len > PAGE_SIZE - offset)
1706 return PAGE_SIZE - offset;
1711 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
1712 void *data, int offset, int len)
1717 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1718 if (kvm_is_error_hva(addr))
1720 r = __copy_from_user(data, (void __user *)addr + offset, len);
1726 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1729 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1731 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1733 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1735 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
1736 int offset, int len)
1738 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1740 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1742 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
1744 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1746 gfn_t gfn = gpa >> PAGE_SHIFT;
1748 int offset = offset_in_page(gpa);
1751 while ((seg = next_segment(len, offset)) != 0) {
1752 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1762 EXPORT_SYMBOL_GPL(kvm_read_guest);
1764 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
1766 gfn_t gfn = gpa >> PAGE_SHIFT;
1768 int offset = offset_in_page(gpa);
1771 while ((seg = next_segment(len, offset)) != 0) {
1772 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
1782 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
1784 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1785 void *data, int offset, unsigned long len)
1790 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1791 if (kvm_is_error_hva(addr))
1793 pagefault_disable();
1794 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1801 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1804 gfn_t gfn = gpa >> PAGE_SHIFT;
1805 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1806 int offset = offset_in_page(gpa);
1808 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1810 EXPORT_SYMBOL_GPL(kvm_read_guest_atomic);
1812 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
1813 void *data, unsigned long len)
1815 gfn_t gfn = gpa >> PAGE_SHIFT;
1816 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1817 int offset = offset_in_page(gpa);
1819 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1821 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
1823 static int __kvm_write_guest_page(struct kvm_memory_slot *memslot, gfn_t gfn,
1824 const void *data, int offset, int len)
1829 addr = gfn_to_hva_memslot(memslot, gfn);
1830 if (kvm_is_error_hva(addr))
1832 r = __copy_to_user((void __user *)addr + offset, data, len);
1835 mark_page_dirty_in_slot(memslot, gfn);
1839 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
1840 const void *data, int offset, int len)
1842 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1844 return __kvm_write_guest_page(slot, gfn, data, offset, len);
1846 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1848 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
1849 const void *data, int offset, int len)
1851 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1853 return __kvm_write_guest_page(slot, gfn, data, offset, len);
1855 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
1857 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1860 gfn_t gfn = gpa >> PAGE_SHIFT;
1862 int offset = offset_in_page(gpa);
1865 while ((seg = next_segment(len, offset)) != 0) {
1866 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1876 EXPORT_SYMBOL_GPL(kvm_write_guest);
1878 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
1881 gfn_t gfn = gpa >> PAGE_SHIFT;
1883 int offset = offset_in_page(gpa);
1886 while ((seg = next_segment(len, offset)) != 0) {
1887 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
1897 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
1899 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
1900 struct gfn_to_hva_cache *ghc,
1901 gpa_t gpa, unsigned long len)
1903 int offset = offset_in_page(gpa);
1904 gfn_t start_gfn = gpa >> PAGE_SHIFT;
1905 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1906 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1907 gfn_t nr_pages_avail;
1910 ghc->generation = slots->generation;
1912 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
1913 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, NULL);
1914 if (!kvm_is_error_hva(ghc->hva) && nr_pages_needed <= 1) {
1918 * If the requested region crosses two memslots, we still
1919 * verify that the entire region is valid here.
1921 while (start_gfn <= end_gfn) {
1923 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
1924 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1926 if (kvm_is_error_hva(ghc->hva))
1928 start_gfn += nr_pages_avail;
1930 /* Use the slow path for cross page reads and writes. */
1931 ghc->memslot = NULL;
1936 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1937 gpa_t gpa, unsigned long len)
1939 struct kvm_memslots *slots = kvm_memslots(kvm);
1940 return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len);
1942 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1944 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1945 void *data, int offset, unsigned long len)
1947 struct kvm_memslots *slots = kvm_memslots(kvm);
1949 gpa_t gpa = ghc->gpa + offset;
1951 BUG_ON(len + offset > ghc->len);
1953 if (slots->generation != ghc->generation)
1954 __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len);
1956 if (unlikely(!ghc->memslot))
1957 return kvm_write_guest(kvm, gpa, data, len);
1959 if (kvm_is_error_hva(ghc->hva))
1962 r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
1965 mark_page_dirty_in_slot(ghc->memslot, gpa >> PAGE_SHIFT);
1969 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached);
1971 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1972 void *data, unsigned long len)
1974 return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len);
1976 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1978 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1979 void *data, unsigned long len)
1981 struct kvm_memslots *slots = kvm_memslots(kvm);
1984 BUG_ON(len > ghc->len);
1986 if (slots->generation != ghc->generation)
1987 __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len);
1989 if (unlikely(!ghc->memslot))
1990 return kvm_read_guest(kvm, ghc->gpa, data, len);
1992 if (kvm_is_error_hva(ghc->hva))
1995 r = __copy_from_user(data, (void __user *)ghc->hva, len);
2001 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
2003 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
2005 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
2007 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
2009 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
2011 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
2013 gfn_t gfn = gpa >> PAGE_SHIFT;
2015 int offset = offset_in_page(gpa);
2018 while ((seg = next_segment(len, offset)) != 0) {
2019 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
2028 EXPORT_SYMBOL_GPL(kvm_clear_guest);
2030 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot,
2033 if (memslot && memslot->dirty_bitmap) {
2034 unsigned long rel_gfn = gfn - memslot->base_gfn;
2036 set_bit_le(rel_gfn, memslot->dirty_bitmap);
2040 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
2042 struct kvm_memory_slot *memslot;
2044 memslot = gfn_to_memslot(kvm, gfn);
2045 mark_page_dirty_in_slot(memslot, gfn);
2047 EXPORT_SYMBOL_GPL(mark_page_dirty);
2049 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
2051 struct kvm_memory_slot *memslot;
2053 memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2054 mark_page_dirty_in_slot(memslot, gfn);
2056 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
2058 void kvm_sigset_activate(struct kvm_vcpu *vcpu)
2060 if (!vcpu->sigset_active)
2064 * This does a lockless modification of ->real_blocked, which is fine
2065 * because, only current can change ->real_blocked and all readers of
2066 * ->real_blocked don't care as long ->real_blocked is always a subset
2069 sigprocmask(SIG_SETMASK, &vcpu->sigset, ¤t->real_blocked);
2072 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu)
2074 if (!vcpu->sigset_active)
2077 sigprocmask(SIG_SETMASK, ¤t->real_blocked, NULL);
2078 sigemptyset(¤t->real_blocked);
2081 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
2083 unsigned int old, val, grow;
2085 old = val = vcpu->halt_poll_ns;
2086 grow = READ_ONCE(halt_poll_ns_grow);
2088 if (val == 0 && grow)
2093 if (val > halt_poll_ns)
2096 vcpu->halt_poll_ns = val;
2097 trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
2100 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
2102 unsigned int old, val, shrink;
2104 old = val = vcpu->halt_poll_ns;
2105 shrink = READ_ONCE(halt_poll_ns_shrink);
2111 vcpu->halt_poll_ns = val;
2112 trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
2115 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
2117 if (kvm_arch_vcpu_runnable(vcpu)) {
2118 kvm_make_request(KVM_REQ_UNHALT, vcpu);
2121 if (kvm_cpu_has_pending_timer(vcpu))
2123 if (signal_pending(current))
2130 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2132 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
2135 DECLARE_SWAITQUEUE(wait);
2136 bool waited = false;
2139 start = cur = ktime_get();
2140 if (vcpu->halt_poll_ns) {
2141 ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns);
2143 ++vcpu->stat.halt_attempted_poll;
2146 * This sets KVM_REQ_UNHALT if an interrupt
2149 if (kvm_vcpu_check_block(vcpu) < 0) {
2150 ++vcpu->stat.halt_successful_poll;
2151 if (!vcpu_valid_wakeup(vcpu))
2152 ++vcpu->stat.halt_poll_invalid;
2156 } while (single_task_running() && ktime_before(cur, stop));
2159 kvm_arch_vcpu_blocking(vcpu);
2162 prepare_to_swait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
2164 if (kvm_vcpu_check_block(vcpu) < 0)
2171 finish_swait(&vcpu->wq, &wait);
2174 kvm_arch_vcpu_unblocking(vcpu);
2176 block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
2178 if (!vcpu_valid_wakeup(vcpu))
2179 shrink_halt_poll_ns(vcpu);
2180 else if (halt_poll_ns) {
2181 if (block_ns <= vcpu->halt_poll_ns)
2183 /* we had a long block, shrink polling */
2184 else if (vcpu->halt_poll_ns && block_ns > halt_poll_ns)
2185 shrink_halt_poll_ns(vcpu);
2186 /* we had a short halt and our poll time is too small */
2187 else if (vcpu->halt_poll_ns < halt_poll_ns &&
2188 block_ns < halt_poll_ns)
2189 grow_halt_poll_ns(vcpu);
2191 vcpu->halt_poll_ns = 0;
2193 trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu));
2194 kvm_arch_vcpu_block_finish(vcpu);
2196 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
2198 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
2200 struct swait_queue_head *wqp;
2202 wqp = kvm_arch_vcpu_wq(vcpu);
2203 if (swq_has_sleeper(wqp)) {
2205 ++vcpu->stat.halt_wakeup;
2211 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
2215 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2217 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
2220 int cpu = vcpu->cpu;
2222 if (kvm_vcpu_wake_up(vcpu))
2226 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
2227 if (kvm_arch_vcpu_should_kick(vcpu))
2228 smp_send_reschedule(cpu);
2231 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
2232 #endif /* !CONFIG_S390 */
2234 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
2237 struct task_struct *task = NULL;
2241 pid = rcu_dereference(target->pid);
2243 task = get_pid_task(pid, PIDTYPE_PID);
2247 ret = yield_to(task, 1);
2248 put_task_struct(task);
2252 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
2255 * Helper that checks whether a VCPU is eligible for directed yield.
2256 * Most eligible candidate to yield is decided by following heuristics:
2258 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2259 * (preempted lock holder), indicated by @in_spin_loop.
2260 * Set at the beiginning and cleared at the end of interception/PLE handler.
2262 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2263 * chance last time (mostly it has become eligible now since we have probably
2264 * yielded to lockholder in last iteration. This is done by toggling
2265 * @dy_eligible each time a VCPU checked for eligibility.)
2267 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2268 * to preempted lock-holder could result in wrong VCPU selection and CPU
2269 * burning. Giving priority for a potential lock-holder increases lock
2272 * Since algorithm is based on heuristics, accessing another VCPU data without
2273 * locking does not harm. It may result in trying to yield to same VCPU, fail
2274 * and continue with next VCPU and so on.
2276 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
2278 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2281 eligible = !vcpu->spin_loop.in_spin_loop ||
2282 vcpu->spin_loop.dy_eligible;
2284 if (vcpu->spin_loop.in_spin_loop)
2285 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
2293 void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode)
2295 struct kvm *kvm = me->kvm;
2296 struct kvm_vcpu *vcpu;
2297 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
2303 kvm_vcpu_set_in_spin_loop(me, true);
2305 * We boost the priority of a VCPU that is runnable but not
2306 * currently running, because it got preempted by something
2307 * else and called schedule in __vcpu_run. Hopefully that
2308 * VCPU is holding the lock that we need and will release it.
2309 * We approximate round-robin by starting at the last boosted VCPU.
2311 for (pass = 0; pass < 2 && !yielded && try; pass++) {
2312 kvm_for_each_vcpu(i, vcpu, kvm) {
2313 if (!pass && i <= last_boosted_vcpu) {
2314 i = last_boosted_vcpu;
2316 } else if (pass && i > last_boosted_vcpu)
2318 if (!READ_ONCE(vcpu->preempted))
2322 if (swait_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
2324 if (yield_to_kernel_mode && !kvm_arch_vcpu_in_kernel(vcpu))
2326 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
2329 yielded = kvm_vcpu_yield_to(vcpu);
2331 kvm->last_boosted_vcpu = i;
2333 } else if (yielded < 0) {
2340 kvm_vcpu_set_in_spin_loop(me, false);
2342 /* Ensure vcpu is not eligible during next spinloop */
2343 kvm_vcpu_set_dy_eligible(me, false);
2345 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
2347 static int kvm_vcpu_fault(struct vm_fault *vmf)
2349 struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data;
2352 if (vmf->pgoff == 0)
2353 page = virt_to_page(vcpu->run);
2355 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
2356 page = virt_to_page(vcpu->arch.pio_data);
2358 #ifdef CONFIG_KVM_MMIO
2359 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
2360 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
2363 return kvm_arch_vcpu_fault(vcpu, vmf);
2369 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
2370 .fault = kvm_vcpu_fault,
2373 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2375 vma->vm_ops = &kvm_vcpu_vm_ops;
2379 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2381 struct kvm_vcpu *vcpu = filp->private_data;
2383 debugfs_remove_recursive(vcpu->debugfs_dentry);
2384 kvm_put_kvm(vcpu->kvm);
2388 static struct file_operations kvm_vcpu_fops = {
2389 .release = kvm_vcpu_release,
2390 .unlocked_ioctl = kvm_vcpu_ioctl,
2391 #ifdef CONFIG_KVM_COMPAT
2392 .compat_ioctl = kvm_vcpu_compat_ioctl,
2394 .mmap = kvm_vcpu_mmap,
2395 .llseek = noop_llseek,
2399 * Allocates an inode for the vcpu.
2401 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2403 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
2406 static int kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
2408 char dir_name[ITOA_MAX_LEN * 2];
2411 if (!kvm_arch_has_vcpu_debugfs())
2414 if (!debugfs_initialized())
2417 snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
2418 vcpu->debugfs_dentry = debugfs_create_dir(dir_name,
2419 vcpu->kvm->debugfs_dentry);
2420 if (!vcpu->debugfs_dentry)
2423 ret = kvm_arch_create_vcpu_debugfs(vcpu);
2425 debugfs_remove_recursive(vcpu->debugfs_dentry);
2433 * Creates some virtual cpus. Good luck creating more than one.
2435 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
2438 struct kvm_vcpu *vcpu;
2440 if (id >= KVM_MAX_VCPU_ID)
2443 mutex_lock(&kvm->lock);
2444 if (kvm->created_vcpus == KVM_MAX_VCPUS) {
2445 mutex_unlock(&kvm->lock);
2449 kvm->created_vcpus++;
2450 mutex_unlock(&kvm->lock);
2452 vcpu = kvm_arch_vcpu_create(kvm, id);
2455 goto vcpu_decrement;
2458 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2460 r = kvm_arch_vcpu_setup(vcpu);
2464 r = kvm_create_vcpu_debugfs(vcpu);
2468 mutex_lock(&kvm->lock);
2469 if (kvm_get_vcpu_by_id(kvm, id)) {
2471 goto unlock_vcpu_destroy;
2474 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
2476 /* Now it's all set up, let userspace reach it */
2478 r = create_vcpu_fd(vcpu);
2481 goto unlock_vcpu_destroy;
2484 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
2487 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
2488 * before kvm->online_vcpu's incremented value.
2491 atomic_inc(&kvm->online_vcpus);
2493 mutex_unlock(&kvm->lock);
2494 kvm_arch_vcpu_postcreate(vcpu);
2497 unlock_vcpu_destroy:
2498 mutex_unlock(&kvm->lock);
2499 debugfs_remove_recursive(vcpu->debugfs_dentry);
2501 kvm_arch_vcpu_destroy(vcpu);
2503 mutex_lock(&kvm->lock);
2504 kvm->created_vcpus--;
2505 mutex_unlock(&kvm->lock);
2509 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2512 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2513 vcpu->sigset_active = 1;
2514 vcpu->sigset = *sigset;
2516 vcpu->sigset_active = 0;
2520 static long kvm_vcpu_ioctl(struct file *filp,
2521 unsigned int ioctl, unsigned long arg)
2523 struct kvm_vcpu *vcpu = filp->private_data;
2524 void __user *argp = (void __user *)arg;
2526 struct kvm_fpu *fpu = NULL;
2527 struct kvm_sregs *kvm_sregs = NULL;
2529 if (vcpu->kvm->mm != current->mm)
2532 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2535 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
2537 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
2538 * so vcpu_load() would break it.
2540 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_S390_IRQ || ioctl == KVM_INTERRUPT)
2541 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2545 r = vcpu_load(vcpu);
2554 oldpid = rcu_access_pointer(vcpu->pid);
2555 if (unlikely(oldpid != current->pids[PIDTYPE_PID].pid)) {
2556 /* The thread running this VCPU changed. */
2557 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
2559 rcu_assign_pointer(vcpu->pid, newpid);
2564 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2565 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;
2744 if (copy_from_user(&kvm_sigmask, argp,
2745 sizeof(kvm_sigmask)))
2748 if (kvm_sigmask.len != sizeof(compat_sigset_t))
2751 if (get_compat_sigset(&sigset, (void *)sigmask_arg->sigset))
2753 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2755 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2759 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2767 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2768 int (*accessor)(struct kvm_device *dev,
2769 struct kvm_device_attr *attr),
2772 struct kvm_device_attr attr;
2777 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2780 return accessor(dev, &attr);
2783 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2786 struct kvm_device *dev = filp->private_data;
2789 case KVM_SET_DEVICE_ATTR:
2790 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2791 case KVM_GET_DEVICE_ATTR:
2792 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2793 case KVM_HAS_DEVICE_ATTR:
2794 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2796 if (dev->ops->ioctl)
2797 return dev->ops->ioctl(dev, ioctl, arg);
2803 static int kvm_device_release(struct inode *inode, struct file *filp)
2805 struct kvm_device *dev = filp->private_data;
2806 struct kvm *kvm = dev->kvm;
2812 static const struct file_operations kvm_device_fops = {
2813 .unlocked_ioctl = kvm_device_ioctl,
2814 #ifdef CONFIG_KVM_COMPAT
2815 .compat_ioctl = kvm_device_ioctl,
2817 .release = kvm_device_release,
2820 struct kvm_device *kvm_device_from_filp(struct file *filp)
2822 if (filp->f_op != &kvm_device_fops)
2825 return filp->private_data;
2828 static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
2829 #ifdef CONFIG_KVM_MPIC
2830 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
2831 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
2835 int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
2837 if (type >= ARRAY_SIZE(kvm_device_ops_table))
2840 if (kvm_device_ops_table[type] != NULL)
2843 kvm_device_ops_table[type] = ops;
2847 void kvm_unregister_device_ops(u32 type)
2849 if (kvm_device_ops_table[type] != NULL)
2850 kvm_device_ops_table[type] = NULL;
2853 static int kvm_ioctl_create_device(struct kvm *kvm,
2854 struct kvm_create_device *cd)
2856 struct kvm_device_ops *ops = NULL;
2857 struct kvm_device *dev;
2858 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2861 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
2864 ops = kvm_device_ops_table[cd->type];
2871 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2878 mutex_lock(&kvm->lock);
2879 ret = ops->create(dev, cd->type);
2881 mutex_unlock(&kvm->lock);
2885 list_add(&dev->vm_node, &kvm->devices);
2886 mutex_unlock(&kvm->lock);
2891 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2893 mutex_lock(&kvm->lock);
2894 list_del(&dev->vm_node);
2895 mutex_unlock(&kvm->lock);
2905 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2908 case KVM_CAP_USER_MEMORY:
2909 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2910 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2911 case KVM_CAP_INTERNAL_ERROR_DATA:
2912 #ifdef CONFIG_HAVE_KVM_MSI
2913 case KVM_CAP_SIGNAL_MSI:
2915 #ifdef CONFIG_HAVE_KVM_IRQFD
2917 case KVM_CAP_IRQFD_RESAMPLE:
2919 case KVM_CAP_IOEVENTFD_ANY_LENGTH:
2920 case KVM_CAP_CHECK_EXTENSION_VM:
2922 #ifdef CONFIG_KVM_MMIO
2923 case KVM_CAP_COALESCED_MMIO:
2924 return KVM_COALESCED_MMIO_PAGE_OFFSET;
2926 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2927 case KVM_CAP_IRQ_ROUTING:
2928 return KVM_MAX_IRQ_ROUTES;
2930 #if KVM_ADDRESS_SPACE_NUM > 1
2931 case KVM_CAP_MULTI_ADDRESS_SPACE:
2932 return KVM_ADDRESS_SPACE_NUM;
2934 case KVM_CAP_MAX_VCPU_ID:
2935 return KVM_MAX_VCPU_ID;
2939 return kvm_vm_ioctl_check_extension(kvm, arg);
2942 static long kvm_vm_ioctl(struct file *filp,
2943 unsigned int ioctl, unsigned long arg)
2945 struct kvm *kvm = filp->private_data;
2946 void __user *argp = (void __user *)arg;
2949 if (kvm->mm != current->mm)
2952 case KVM_CREATE_VCPU:
2953 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2955 case KVM_SET_USER_MEMORY_REGION: {
2956 struct kvm_userspace_memory_region kvm_userspace_mem;
2959 if (copy_from_user(&kvm_userspace_mem, argp,
2960 sizeof(kvm_userspace_mem)))
2963 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2966 case KVM_GET_DIRTY_LOG: {
2967 struct kvm_dirty_log log;
2970 if (copy_from_user(&log, argp, sizeof(log)))
2972 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2975 #ifdef CONFIG_KVM_MMIO
2976 case KVM_REGISTER_COALESCED_MMIO: {
2977 struct kvm_coalesced_mmio_zone zone;
2980 if (copy_from_user(&zone, argp, sizeof(zone)))
2982 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2985 case KVM_UNREGISTER_COALESCED_MMIO: {
2986 struct kvm_coalesced_mmio_zone zone;
2989 if (copy_from_user(&zone, argp, sizeof(zone)))
2991 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2996 struct kvm_irqfd data;
2999 if (copy_from_user(&data, argp, sizeof(data)))
3001 r = kvm_irqfd(kvm, &data);
3004 case KVM_IOEVENTFD: {
3005 struct kvm_ioeventfd data;
3008 if (copy_from_user(&data, argp, sizeof(data)))
3010 r = kvm_ioeventfd(kvm, &data);
3013 #ifdef CONFIG_HAVE_KVM_MSI
3014 case KVM_SIGNAL_MSI: {
3018 if (copy_from_user(&msi, argp, sizeof(msi)))
3020 r = kvm_send_userspace_msi(kvm, &msi);
3024 #ifdef __KVM_HAVE_IRQ_LINE
3025 case KVM_IRQ_LINE_STATUS:
3026 case KVM_IRQ_LINE: {
3027 struct kvm_irq_level irq_event;
3030 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
3033 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
3034 ioctl == KVM_IRQ_LINE_STATUS);
3039 if (ioctl == KVM_IRQ_LINE_STATUS) {
3040 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
3048 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3049 case KVM_SET_GSI_ROUTING: {
3050 struct kvm_irq_routing routing;
3051 struct kvm_irq_routing __user *urouting;
3052 struct kvm_irq_routing_entry *entries = NULL;
3055 if (copy_from_user(&routing, argp, sizeof(routing)))
3058 if (!kvm_arch_can_set_irq_routing(kvm))
3060 if (routing.nr > KVM_MAX_IRQ_ROUTES)
3066 entries = vmalloc(routing.nr * sizeof(*entries));
3071 if (copy_from_user(entries, urouting->entries,
3072 routing.nr * sizeof(*entries)))
3073 goto out_free_irq_routing;
3075 r = kvm_set_irq_routing(kvm, entries, routing.nr,
3077 out_free_irq_routing:
3081 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
3082 case KVM_CREATE_DEVICE: {
3083 struct kvm_create_device cd;
3086 if (copy_from_user(&cd, argp, sizeof(cd)))
3089 r = kvm_ioctl_create_device(kvm, &cd);
3094 if (copy_to_user(argp, &cd, sizeof(cd)))
3100 case KVM_CHECK_EXTENSION:
3101 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
3104 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
3110 #ifdef CONFIG_KVM_COMPAT
3111 struct compat_kvm_dirty_log {
3115 compat_uptr_t dirty_bitmap; /* one bit per page */
3120 static long kvm_vm_compat_ioctl(struct file *filp,
3121 unsigned int ioctl, unsigned long arg)
3123 struct kvm *kvm = filp->private_data;
3126 if (kvm->mm != current->mm)
3129 case KVM_GET_DIRTY_LOG: {
3130 struct compat_kvm_dirty_log compat_log;
3131 struct kvm_dirty_log log;
3133 if (copy_from_user(&compat_log, (void __user *)arg,
3134 sizeof(compat_log)))
3136 log.slot = compat_log.slot;
3137 log.padding1 = compat_log.padding1;
3138 log.padding2 = compat_log.padding2;
3139 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
3141 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3145 r = kvm_vm_ioctl(filp, ioctl, arg);
3151 static struct file_operations kvm_vm_fops = {
3152 .release = kvm_vm_release,
3153 .unlocked_ioctl = kvm_vm_ioctl,
3154 #ifdef CONFIG_KVM_COMPAT
3155 .compat_ioctl = kvm_vm_compat_ioctl,
3157 .llseek = noop_llseek,
3160 static int kvm_dev_ioctl_create_vm(unsigned long type)
3166 kvm = kvm_create_vm(type);
3168 return PTR_ERR(kvm);
3169 #ifdef CONFIG_KVM_MMIO
3170 r = kvm_coalesced_mmio_init(kvm);
3176 r = get_unused_fd_flags(O_CLOEXEC);
3181 file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
3185 return PTR_ERR(file);
3189 * Don't call kvm_put_kvm anymore at this point; file->f_op is
3190 * already set, with ->release() being kvm_vm_release(). In error
3191 * cases it will be called by the final fput(file) and will take
3192 * care of doing kvm_put_kvm(kvm).
3194 if (kvm_create_vm_debugfs(kvm, r) < 0) {
3199 kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm);
3201 fd_install(r, file);
3205 static long kvm_dev_ioctl(struct file *filp,
3206 unsigned int ioctl, unsigned long arg)
3211 case KVM_GET_API_VERSION:
3214 r = KVM_API_VERSION;
3217 r = kvm_dev_ioctl_create_vm(arg);
3219 case KVM_CHECK_EXTENSION:
3220 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
3222 case KVM_GET_VCPU_MMAP_SIZE:
3225 r = PAGE_SIZE; /* struct kvm_run */
3227 r += PAGE_SIZE; /* pio data page */
3229 #ifdef CONFIG_KVM_MMIO
3230 r += PAGE_SIZE; /* coalesced mmio ring page */
3233 case KVM_TRACE_ENABLE:
3234 case KVM_TRACE_PAUSE:
3235 case KVM_TRACE_DISABLE:
3239 return kvm_arch_dev_ioctl(filp, ioctl, arg);
3245 static struct file_operations kvm_chardev_ops = {
3246 .unlocked_ioctl = kvm_dev_ioctl,
3247 .compat_ioctl = kvm_dev_ioctl,
3248 .llseek = noop_llseek,
3251 static struct miscdevice kvm_dev = {
3257 static void hardware_enable_nolock(void *junk)
3259 int cpu = raw_smp_processor_id();
3262 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
3265 cpumask_set_cpu(cpu, cpus_hardware_enabled);
3267 r = kvm_arch_hardware_enable();
3270 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3271 atomic_inc(&hardware_enable_failed);
3272 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
3276 static int kvm_starting_cpu(unsigned int cpu)
3278 raw_spin_lock(&kvm_count_lock);
3279 if (kvm_usage_count)
3280 hardware_enable_nolock(NULL);
3281 raw_spin_unlock(&kvm_count_lock);
3285 static void hardware_disable_nolock(void *junk)
3287 int cpu = raw_smp_processor_id();
3289 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
3291 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3292 kvm_arch_hardware_disable();
3295 static int kvm_dying_cpu(unsigned int cpu)
3297 raw_spin_lock(&kvm_count_lock);
3298 if (kvm_usage_count)
3299 hardware_disable_nolock(NULL);
3300 raw_spin_unlock(&kvm_count_lock);
3304 static void hardware_disable_all_nolock(void)
3306 BUG_ON(!kvm_usage_count);
3309 if (!kvm_usage_count)
3310 on_each_cpu(hardware_disable_nolock, NULL, 1);
3313 static void hardware_disable_all(void)
3315 raw_spin_lock(&kvm_count_lock);
3316 hardware_disable_all_nolock();
3317 raw_spin_unlock(&kvm_count_lock);
3320 static int hardware_enable_all(void)
3324 raw_spin_lock(&kvm_count_lock);
3327 if (kvm_usage_count == 1) {
3328 atomic_set(&hardware_enable_failed, 0);
3329 on_each_cpu(hardware_enable_nolock, NULL, 1);
3331 if (atomic_read(&hardware_enable_failed)) {
3332 hardware_disable_all_nolock();
3337 raw_spin_unlock(&kvm_count_lock);
3342 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3346 * Some (well, at least mine) BIOSes hang on reboot if
3349 * And Intel TXT required VMX off for all cpu when system shutdown.
3351 pr_info("kvm: exiting hardware virtualization\n");
3352 kvm_rebooting = true;
3353 on_each_cpu(hardware_disable_nolock, NULL, 1);
3357 static struct notifier_block kvm_reboot_notifier = {
3358 .notifier_call = kvm_reboot,
3362 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3366 for (i = 0; i < bus->dev_count; i++) {
3367 struct kvm_io_device *pos = bus->range[i].dev;
3369 kvm_iodevice_destructor(pos);
3374 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
3375 const struct kvm_io_range *r2)
3377 gpa_t addr1 = r1->addr;
3378 gpa_t addr2 = r2->addr;
3383 /* If r2->len == 0, match the exact address. If r2->len != 0,
3384 * accept any overlapping write. Any order is acceptable for
3385 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
3386 * we process all of them.
3399 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
3401 return kvm_io_bus_cmp(p1, p2);
3404 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
3405 gpa_t addr, int len)
3407 bus->range[bus->dev_count++] = (struct kvm_io_range) {
3413 sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
3414 kvm_io_bus_sort_cmp, NULL);
3419 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
3420 gpa_t addr, int len)
3422 struct kvm_io_range *range, key;
3425 key = (struct kvm_io_range) {
3430 range = bsearch(&key, bus->range, bus->dev_count,
3431 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
3435 off = range - bus->range;
3437 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
3443 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3444 struct kvm_io_range *range, const void *val)
3448 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3452 while (idx < bus->dev_count &&
3453 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3454 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
3463 /* kvm_io_bus_write - called under kvm->slots_lock */
3464 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3465 int len, const void *val)
3467 struct kvm_io_bus *bus;
3468 struct kvm_io_range range;
3471 range = (struct kvm_io_range) {
3476 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3479 r = __kvm_io_bus_write(vcpu, bus, &range, val);
3480 return r < 0 ? r : 0;
3483 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3484 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
3485 gpa_t addr, int len, const void *val, long cookie)
3487 struct kvm_io_bus *bus;
3488 struct kvm_io_range range;
3490 range = (struct kvm_io_range) {
3495 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3499 /* First try the device referenced by cookie. */
3500 if ((cookie >= 0) && (cookie < bus->dev_count) &&
3501 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
3502 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
3507 * cookie contained garbage; fall back to search and return the
3508 * correct cookie value.
3510 return __kvm_io_bus_write(vcpu, bus, &range, val);
3513 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3514 struct kvm_io_range *range, void *val)
3518 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3522 while (idx < bus->dev_count &&
3523 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3524 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
3532 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
3534 /* kvm_io_bus_read - called under kvm->slots_lock */
3535 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3538 struct kvm_io_bus *bus;
3539 struct kvm_io_range range;
3542 range = (struct kvm_io_range) {
3547 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3550 r = __kvm_io_bus_read(vcpu, bus, &range, val);
3551 return r < 0 ? r : 0;
3555 /* Caller must hold slots_lock. */
3556 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3557 int len, struct kvm_io_device *dev)
3559 struct kvm_io_bus *new_bus, *bus;
3561 bus = kvm_get_bus(kvm, bus_idx);
3565 /* exclude ioeventfd which is limited by maximum fd */
3566 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3569 new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count + 1) *
3570 sizeof(struct kvm_io_range)), GFP_KERNEL);
3573 memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
3574 sizeof(struct kvm_io_range)));
3575 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
3576 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3577 synchronize_srcu_expedited(&kvm->srcu);
3583 /* Caller must hold slots_lock. */
3584 void kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3585 struct kvm_io_device *dev)
3588 struct kvm_io_bus *new_bus, *bus;
3590 bus = kvm_get_bus(kvm, bus_idx);
3594 for (i = 0; i < bus->dev_count; i++)
3595 if (bus->range[i].dev == dev) {
3599 if (i == bus->dev_count)
3602 new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3603 sizeof(struct kvm_io_range)), GFP_KERNEL);
3605 pr_err("kvm: failed to shrink bus, removing it completely\n");
3609 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3610 new_bus->dev_count--;
3611 memcpy(new_bus->range + i, bus->range + i + 1,
3612 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3615 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3616 synchronize_srcu_expedited(&kvm->srcu);
3621 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3624 struct kvm_io_bus *bus;
3625 int dev_idx, srcu_idx;
3626 struct kvm_io_device *iodev = NULL;
3628 srcu_idx = srcu_read_lock(&kvm->srcu);
3630 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
3634 dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
3638 iodev = bus->range[dev_idx].dev;
3641 srcu_read_unlock(&kvm->srcu, srcu_idx);
3645 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
3647 static int kvm_debugfs_open(struct inode *inode, struct file *file,
3648 int (*get)(void *, u64 *), int (*set)(void *, u64),
3651 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
3654 /* The debugfs files are a reference to the kvm struct which
3655 * is still valid when kvm_destroy_vm is called.
3656 * To avoid the race between open and the removal of the debugfs
3657 * directory we test against the users count.
3659 if (!refcount_inc_not_zero(&stat_data->kvm->users_count))
3662 if (simple_attr_open(inode, file, get, set, fmt)) {
3663 kvm_put_kvm(stat_data->kvm);
3670 static int kvm_debugfs_release(struct inode *inode, struct file *file)
3672 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
3675 simple_attr_release(inode, file);
3676 kvm_put_kvm(stat_data->kvm);
3681 static int vm_stat_get_per_vm(void *data, u64 *val)
3683 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3685 *val = *(ulong *)((void *)stat_data->kvm + stat_data->offset);
3690 static int vm_stat_clear_per_vm(void *data, u64 val)
3692 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3697 *(ulong *)((void *)stat_data->kvm + stat_data->offset) = 0;
3702 static int vm_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, vm_stat_get_per_vm,
3706 vm_stat_clear_per_vm, "%llu\n");
3709 static const struct file_operations vm_stat_get_per_vm_fops = {
3710 .owner = THIS_MODULE,
3711 .open = vm_stat_get_per_vm_open,
3712 .release = kvm_debugfs_release,
3713 .read = simple_attr_read,
3714 .write = simple_attr_write,
3715 .llseek = no_llseek,
3718 static int vcpu_stat_get_per_vm(void *data, u64 *val)
3721 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3722 struct kvm_vcpu *vcpu;
3726 kvm_for_each_vcpu(i, vcpu, stat_data->kvm)
3727 *val += *(u64 *)((void *)vcpu + stat_data->offset);
3732 static int vcpu_stat_clear_per_vm(void *data, u64 val)
3735 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3736 struct kvm_vcpu *vcpu;
3741 kvm_for_each_vcpu(i, vcpu, stat_data->kvm)
3742 *(u64 *)((void *)vcpu + stat_data->offset) = 0;
3747 static int vcpu_stat_get_per_vm_open(struct inode *inode, struct file *file)
3749 __simple_attr_check_format("%llu\n", 0ull);
3750 return kvm_debugfs_open(inode, file, vcpu_stat_get_per_vm,
3751 vcpu_stat_clear_per_vm, "%llu\n");
3754 static const struct file_operations vcpu_stat_get_per_vm_fops = {
3755 .owner = THIS_MODULE,
3756 .open = vcpu_stat_get_per_vm_open,
3757 .release = kvm_debugfs_release,
3758 .read = simple_attr_read,
3759 .write = simple_attr_write,
3760 .llseek = no_llseek,
3763 static const struct file_operations *stat_fops_per_vm[] = {
3764 [KVM_STAT_VCPU] = &vcpu_stat_get_per_vm_fops,
3765 [KVM_STAT_VM] = &vm_stat_get_per_vm_fops,
3768 static int vm_stat_get(void *_offset, u64 *val)
3770 unsigned offset = (long)_offset;
3772 struct kvm_stat_data stat_tmp = {.offset = offset};
3776 spin_lock(&kvm_lock);
3777 list_for_each_entry(kvm, &vm_list, vm_list) {
3779 vm_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
3782 spin_unlock(&kvm_lock);
3786 static int vm_stat_clear(void *_offset, u64 val)
3788 unsigned offset = (long)_offset;
3790 struct kvm_stat_data stat_tmp = {.offset = offset};
3795 spin_lock(&kvm_lock);
3796 list_for_each_entry(kvm, &vm_list, vm_list) {
3798 vm_stat_clear_per_vm((void *)&stat_tmp, 0);
3800 spin_unlock(&kvm_lock);
3805 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n");
3807 static int vcpu_stat_get(void *_offset, u64 *val)
3809 unsigned offset = (long)_offset;
3811 struct kvm_stat_data stat_tmp = {.offset = offset};
3815 spin_lock(&kvm_lock);
3816 list_for_each_entry(kvm, &vm_list, vm_list) {
3818 vcpu_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
3821 spin_unlock(&kvm_lock);
3825 static int vcpu_stat_clear(void *_offset, u64 val)
3827 unsigned offset = (long)_offset;
3829 struct kvm_stat_data stat_tmp = {.offset = offset};
3834 spin_lock(&kvm_lock);
3835 list_for_each_entry(kvm, &vm_list, vm_list) {
3837 vcpu_stat_clear_per_vm((void *)&stat_tmp, 0);
3839 spin_unlock(&kvm_lock);
3844 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear,
3847 static const struct file_operations *stat_fops[] = {
3848 [KVM_STAT_VCPU] = &vcpu_stat_fops,
3849 [KVM_STAT_VM] = &vm_stat_fops,
3852 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm)
3854 struct kobj_uevent_env *env;
3855 unsigned long long created, active;
3857 if (!kvm_dev.this_device || !kvm)
3860 spin_lock(&kvm_lock);
3861 if (type == KVM_EVENT_CREATE_VM) {
3862 kvm_createvm_count++;
3864 } else if (type == KVM_EVENT_DESTROY_VM) {
3867 created = kvm_createvm_count;
3868 active = kvm_active_vms;
3869 spin_unlock(&kvm_lock);
3871 env = kzalloc(sizeof(*env), GFP_KERNEL);
3875 add_uevent_var(env, "CREATED=%llu", created);
3876 add_uevent_var(env, "COUNT=%llu", active);
3878 if (type == KVM_EVENT_CREATE_VM) {
3879 add_uevent_var(env, "EVENT=create");
3880 kvm->userspace_pid = task_pid_nr(current);
3881 } else if (type == KVM_EVENT_DESTROY_VM) {
3882 add_uevent_var(env, "EVENT=destroy");
3884 add_uevent_var(env, "PID=%d", kvm->userspace_pid);
3886 if (kvm->debugfs_dentry) {
3887 char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL);
3890 tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX);
3892 add_uevent_var(env, "STATS_PATH=%s", tmp);
3896 /* no need for checks, since we are adding at most only 5 keys */
3897 env->envp[env->envp_idx++] = NULL;
3898 kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp);
3902 static int kvm_init_debug(void)
3905 struct kvm_stats_debugfs_item *p;
3907 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3908 if (kvm_debugfs_dir == NULL)
3911 kvm_debugfs_num_entries = 0;
3912 for (p = debugfs_entries; p->name; ++p, kvm_debugfs_num_entries++) {
3913 if (!debugfs_create_file(p->name, 0644, kvm_debugfs_dir,
3914 (void *)(long)p->offset,
3915 stat_fops[p->kind]))
3922 debugfs_remove_recursive(kvm_debugfs_dir);
3927 static int kvm_suspend(void)
3929 if (kvm_usage_count)
3930 hardware_disable_nolock(NULL);
3934 static void kvm_resume(void)
3936 if (kvm_usage_count) {
3937 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3938 hardware_enable_nolock(NULL);
3942 static struct syscore_ops kvm_syscore_ops = {
3943 .suspend = kvm_suspend,
3944 .resume = kvm_resume,
3948 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3950 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3953 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3955 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3957 if (vcpu->preempted)
3958 vcpu->preempted = false;
3960 kvm_arch_sched_in(vcpu, cpu);
3962 kvm_arch_vcpu_load(vcpu, cpu);
3965 static void kvm_sched_out(struct preempt_notifier *pn,
3966 struct task_struct *next)
3968 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3970 if (current->state == TASK_RUNNING)
3971 vcpu->preempted = true;
3972 kvm_arch_vcpu_put(vcpu);
3975 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3976 struct module *module)
3981 r = kvm_arch_init(opaque);
3986 * kvm_arch_init makes sure there's at most one caller
3987 * for architectures that support multiple implementations,
3988 * like intel and amd on x86.
3989 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3990 * conflicts in case kvm is already setup for another implementation.
3992 r = kvm_irqfd_init();
3996 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
4001 r = kvm_arch_hardware_setup();
4005 for_each_online_cpu(cpu) {
4006 smp_call_function_single(cpu,
4007 kvm_arch_check_processor_compat,
4013 r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting",
4014 kvm_starting_cpu, kvm_dying_cpu);
4017 register_reboot_notifier(&kvm_reboot_notifier);
4019 /* A kmem cache lets us meet the alignment requirements of fx_save. */
4021 vcpu_align = __alignof__(struct kvm_vcpu);
4023 kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align,
4025 offsetof(struct kvm_vcpu, arch),
4026 sizeof_field(struct kvm_vcpu, arch),
4028 if (!kvm_vcpu_cache) {
4033 r = kvm_async_pf_init();
4037 kvm_chardev_ops.owner = module;
4038 kvm_vm_fops.owner = module;
4039 kvm_vcpu_fops.owner = module;
4041 r = misc_register(&kvm_dev);
4043 pr_err("kvm: misc device register failed\n");
4047 register_syscore_ops(&kvm_syscore_ops);
4049 kvm_preempt_ops.sched_in = kvm_sched_in;
4050 kvm_preempt_ops.sched_out = kvm_sched_out;
4052 r = kvm_init_debug();
4054 pr_err("kvm: create debugfs files failed\n");
4058 r = kvm_vfio_ops_init();
4064 unregister_syscore_ops(&kvm_syscore_ops);
4065 misc_deregister(&kvm_dev);
4067 kvm_async_pf_deinit();
4069 kmem_cache_destroy(kvm_vcpu_cache);
4071 unregister_reboot_notifier(&kvm_reboot_notifier);
4072 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4075 kvm_arch_hardware_unsetup();
4077 free_cpumask_var(cpus_hardware_enabled);
4085 EXPORT_SYMBOL_GPL(kvm_init);
4089 debugfs_remove_recursive(kvm_debugfs_dir);
4090 misc_deregister(&kvm_dev);
4091 kmem_cache_destroy(kvm_vcpu_cache);
4092 kvm_async_pf_deinit();
4093 unregister_syscore_ops(&kvm_syscore_ops);
4094 unregister_reboot_notifier(&kvm_reboot_notifier);
4095 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4096 on_each_cpu(hardware_disable_nolock, NULL, 1);
4097 kvm_arch_hardware_unsetup();
4100 free_cpumask_var(cpus_hardware_enabled);
4101 kvm_vfio_ops_exit();
4103 EXPORT_SYMBOL_GPL(kvm_exit);