1 // SPDX-License-Identifier: GPL-2.0-only
3 * Kernel-based Virtual Machine driver for Linux
5 * This module enables machines with Intel VT-x extensions to run virtual
6 * machines without emulation or binary translation.
8 * Copyright (C) 2006 Qumranet, Inc.
9 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
12 * Avi Kivity <avi@qumranet.com>
13 * Yaniv Kamay <yaniv@qumranet.com>
16 #include <kvm/iodev.h>
18 #include <linux/kvm_host.h>
19 #include <linux/kvm.h>
20 #include <linux/module.h>
21 #include <linux/errno.h>
22 #include <linux/percpu.h>
24 #include <linux/miscdevice.h>
25 #include <linux/vmalloc.h>
26 #include <linux/reboot.h>
27 #include <linux/debugfs.h>
28 #include <linux/highmem.h>
29 #include <linux/file.h>
30 #include <linux/syscore_ops.h>
31 #include <linux/cpu.h>
32 #include <linux/sched/signal.h>
33 #include <linux/sched/mm.h>
34 #include <linux/sched/stat.h>
35 #include <linux/cpumask.h>
36 #include <linux/smp.h>
37 #include <linux/anon_inodes.h>
38 #include <linux/profile.h>
39 #include <linux/kvm_para.h>
40 #include <linux/pagemap.h>
41 #include <linux/mman.h>
42 #include <linux/swap.h>
43 #include <linux/bitops.h>
44 #include <linux/spinlock.h>
45 #include <linux/compat.h>
46 #include <linux/srcu.h>
47 #include <linux/hugetlb.h>
48 #include <linux/slab.h>
49 #include <linux/sort.h>
50 #include <linux/bsearch.h>
52 #include <linux/lockdep.h>
53 #include <linux/kthread.h>
55 #include <asm/processor.h>
56 #include <asm/ioctl.h>
57 #include <linux/uaccess.h>
59 #include "coalesced_mmio.h"
63 #define CREATE_TRACE_POINTS
64 #include <trace/events/kvm.h>
66 /* Worst case buffer size needed for holding an integer. */
67 #define ITOA_MAX_LEN 12
69 MODULE_AUTHOR("Qumranet");
70 MODULE_LICENSE("GPL");
72 /* Architectures should define their poll value according to the halt latency */
73 unsigned int halt_poll_ns = KVM_HALT_POLL_NS_DEFAULT;
74 module_param(halt_poll_ns, uint, 0644);
75 EXPORT_SYMBOL_GPL(halt_poll_ns);
77 /* Default doubles per-vcpu halt_poll_ns. */
78 unsigned int halt_poll_ns_grow = 2;
79 module_param(halt_poll_ns_grow, uint, 0644);
80 EXPORT_SYMBOL_GPL(halt_poll_ns_grow);
82 /* The start value to grow halt_poll_ns from */
83 unsigned int halt_poll_ns_grow_start = 10000; /* 10us */
84 module_param(halt_poll_ns_grow_start, uint, 0644);
85 EXPORT_SYMBOL_GPL(halt_poll_ns_grow_start);
87 /* Default resets per-vcpu halt_poll_ns . */
88 unsigned int halt_poll_ns_shrink;
89 module_param(halt_poll_ns_shrink, uint, 0644);
90 EXPORT_SYMBOL_GPL(halt_poll_ns_shrink);
95 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
98 DEFINE_MUTEX(kvm_lock);
99 static DEFINE_RAW_SPINLOCK(kvm_count_lock);
102 static cpumask_var_t cpus_hardware_enabled;
103 static int kvm_usage_count;
104 static atomic_t hardware_enable_failed;
106 static struct kmem_cache *kvm_vcpu_cache;
108 static __read_mostly struct preempt_ops kvm_preempt_ops;
109 static DEFINE_PER_CPU(struct kvm_vcpu *, kvm_running_vcpu);
111 struct dentry *kvm_debugfs_dir;
112 EXPORT_SYMBOL_GPL(kvm_debugfs_dir);
114 static int kvm_debugfs_num_entries;
115 static const struct file_operations stat_fops_per_vm;
117 static struct file_operations kvm_chardev_ops;
119 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
121 #ifdef CONFIG_KVM_COMPAT
122 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
124 #define KVM_COMPAT(c) .compat_ioctl = (c)
127 * For architectures that don't implement a compat infrastructure,
128 * adopt a double line of defense:
129 * - Prevent a compat task from opening /dev/kvm
130 * - If the open has been done by a 64bit task, and the KVM fd
131 * passed to a compat task, let the ioctls fail.
133 static long kvm_no_compat_ioctl(struct file *file, unsigned int ioctl,
134 unsigned long arg) { return -EINVAL; }
136 static int kvm_no_compat_open(struct inode *inode, struct file *file)
138 return is_compat_task() ? -ENODEV : 0;
140 #define KVM_COMPAT(c) .compat_ioctl = kvm_no_compat_ioctl, \
141 .open = kvm_no_compat_open
143 static int hardware_enable_all(void);
144 static void hardware_disable_all(void);
146 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
148 __visible bool kvm_rebooting;
149 EXPORT_SYMBOL_GPL(kvm_rebooting);
151 #define KVM_EVENT_CREATE_VM 0
152 #define KVM_EVENT_DESTROY_VM 1
153 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm);
154 static unsigned long long kvm_createvm_count;
155 static unsigned long long kvm_active_vms;
157 static DEFINE_PER_CPU(cpumask_var_t, cpu_kick_mask);
159 __weak void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
160 unsigned long start, unsigned long end)
164 __weak void kvm_arch_guest_memory_reclaimed(struct kvm *kvm)
168 bool kvm_is_zone_device_pfn(kvm_pfn_t pfn)
171 * The metadata used by is_zone_device_page() to determine whether or
172 * not a page is ZONE_DEVICE is guaranteed to be valid if and only if
173 * the device has been pinned, e.g. by get_user_pages(). WARN if the
174 * page_count() is zero to help detect bad usage of this helper.
176 if (!pfn_valid(pfn) || WARN_ON_ONCE(!page_count(pfn_to_page(pfn))))
179 return is_zone_device_page(pfn_to_page(pfn));
182 bool kvm_is_reserved_pfn(kvm_pfn_t pfn)
185 * ZONE_DEVICE pages currently set PG_reserved, but from a refcounting
186 * perspective they are "normal" pages, albeit with slightly different
190 return PageReserved(pfn_to_page(pfn)) &&
192 !kvm_is_zone_device_pfn(pfn);
197 bool kvm_is_transparent_hugepage(kvm_pfn_t pfn)
199 struct page *page = pfn_to_page(pfn);
201 if (!PageTransCompoundMap(page))
204 return is_transparent_hugepage(compound_head(page));
208 * Switches to specified vcpu, until a matching vcpu_put()
210 void vcpu_load(struct kvm_vcpu *vcpu)
214 __this_cpu_write(kvm_running_vcpu, vcpu);
215 preempt_notifier_register(&vcpu->preempt_notifier);
216 kvm_arch_vcpu_load(vcpu, cpu);
219 EXPORT_SYMBOL_GPL(vcpu_load);
221 void vcpu_put(struct kvm_vcpu *vcpu)
224 kvm_arch_vcpu_put(vcpu);
225 preempt_notifier_unregister(&vcpu->preempt_notifier);
226 __this_cpu_write(kvm_running_vcpu, NULL);
229 EXPORT_SYMBOL_GPL(vcpu_put);
231 /* TODO: merge with kvm_arch_vcpu_should_kick */
232 static bool kvm_request_needs_ipi(struct kvm_vcpu *vcpu, unsigned req)
234 int mode = kvm_vcpu_exiting_guest_mode(vcpu);
237 * We need to wait for the VCPU to reenable interrupts and get out of
238 * READING_SHADOW_PAGE_TABLES mode.
240 if (req & KVM_REQUEST_WAIT)
241 return mode != OUTSIDE_GUEST_MODE;
244 * Need to kick a running VCPU, but otherwise there is nothing to do.
246 return mode == IN_GUEST_MODE;
249 static void ack_flush(void *_completed)
253 static inline bool kvm_kick_many_cpus(cpumask_var_t tmp, bool wait)
255 const struct cpumask *cpus;
257 if (likely(cpumask_available(tmp)))
260 cpus = cpu_online_mask;
262 if (cpumask_empty(cpus))
265 smp_call_function_many(cpus, ack_flush, NULL, wait);
269 static void kvm_make_vcpu_request(struct kvm *kvm, struct kvm_vcpu *vcpu,
270 unsigned int req, cpumask_var_t tmp,
275 kvm_make_request(req, vcpu);
277 if (!(req & KVM_REQUEST_NO_WAKEUP) && kvm_vcpu_wake_up(vcpu))
281 * tmp can be "unavailable" if cpumasks are allocated off stack as
282 * allocation of the mask is deliberately not fatal and is handled by
283 * falling back to kicking all online CPUs.
285 if (!cpumask_available(tmp))
289 * Note, the vCPU could get migrated to a different pCPU at any point
290 * after kvm_request_needs_ipi(), which could result in sending an IPI
291 * to the previous pCPU. But, that's OK because the purpose of the IPI
292 * is to ensure the vCPU returns to OUTSIDE_GUEST_MODE, which is
293 * satisfied if the vCPU migrates. Entering READING_SHADOW_PAGE_TABLES
294 * after this point is also OK, as the requirement is only that KVM wait
295 * for vCPUs that were reading SPTEs _before_ any changes were
296 * finalized. See kvm_vcpu_kick() for more details on handling requests.
298 if (kvm_request_needs_ipi(vcpu, req)) {
299 cpu = READ_ONCE(vcpu->cpu);
300 if (cpu != -1 && cpu != current_cpu)
301 __cpumask_set_cpu(cpu, tmp);
305 bool kvm_make_vcpus_request_mask(struct kvm *kvm, unsigned int req,
306 struct kvm_vcpu *except,
307 unsigned long *vcpu_bitmap, cpumask_var_t tmp)
309 struct kvm_vcpu *vcpu;
315 for_each_set_bit(i, vcpu_bitmap, KVM_MAX_VCPUS) {
316 vcpu = kvm_get_vcpu(kvm, i);
317 if (!vcpu || vcpu == except)
319 kvm_make_vcpu_request(kvm, vcpu, req, tmp, me);
322 called = kvm_kick_many_cpus(tmp, !!(req & KVM_REQUEST_WAIT));
328 bool kvm_make_all_cpus_request_except(struct kvm *kvm, unsigned int req,
329 struct kvm_vcpu *except)
331 struct kvm_vcpu *vcpu;
332 struct cpumask *cpus;
338 cpus = this_cpu_cpumask_var_ptr(cpu_kick_mask);
341 kvm_for_each_vcpu(i, vcpu, kvm) {
344 kvm_make_vcpu_request(kvm, vcpu, req, cpus, me);
347 called = kvm_kick_many_cpus(cpus, !!(req & KVM_REQUEST_WAIT));
353 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
355 return kvm_make_all_cpus_request_except(kvm, req, NULL);
358 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
359 void kvm_flush_remote_tlbs(struct kvm *kvm)
362 * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in
363 * kvm_make_all_cpus_request.
365 long dirty_count = smp_load_acquire(&kvm->tlbs_dirty);
368 * We want to publish modifications to the page tables before reading
369 * mode. Pairs with a memory barrier in arch-specific code.
370 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
371 * and smp_mb in walk_shadow_page_lockless_begin/end.
372 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
374 * There is already an smp_mb__after_atomic() before
375 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
378 if (!kvm_arch_flush_remote_tlb(kvm)
379 || kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
380 ++kvm->stat.remote_tlb_flush;
381 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
383 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
386 void kvm_reload_remote_mmus(struct kvm *kvm)
388 kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
391 static void kvm_flush_shadow_all(struct kvm *kvm)
393 kvm_arch_flush_shadow_all(kvm);
394 kvm_arch_guest_memory_reclaimed(kvm);
397 #ifdef KVM_ARCH_NR_OBJS_PER_MEMORY_CACHE
398 static inline void *mmu_memory_cache_alloc_obj(struct kvm_mmu_memory_cache *mc,
401 gfp_flags |= mc->gfp_zero;
404 return kmem_cache_alloc(mc->kmem_cache, gfp_flags);
406 return (void *)__get_free_page(gfp_flags);
409 int kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache *mc, int min)
413 if (mc->nobjs >= min)
415 while (mc->nobjs < ARRAY_SIZE(mc->objects)) {
416 obj = mmu_memory_cache_alloc_obj(mc, GFP_KERNEL_ACCOUNT);
418 return mc->nobjs >= min ? 0 : -ENOMEM;
419 mc->objects[mc->nobjs++] = obj;
424 int kvm_mmu_memory_cache_nr_free_objects(struct kvm_mmu_memory_cache *mc)
429 void kvm_mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc)
433 kmem_cache_free(mc->kmem_cache, mc->objects[--mc->nobjs]);
435 free_page((unsigned long)mc->objects[--mc->nobjs]);
439 void *kvm_mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc)
443 if (WARN_ON(!mc->nobjs))
444 p = mmu_memory_cache_alloc_obj(mc, GFP_ATOMIC | __GFP_ACCOUNT);
446 p = mc->objects[--mc->nobjs];
452 static void kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
454 mutex_init(&vcpu->mutex);
459 rcuwait_init(&vcpu->wait);
460 kvm_async_pf_vcpu_init(vcpu);
463 INIT_LIST_HEAD(&vcpu->blocked_vcpu_list);
465 kvm_vcpu_set_in_spin_loop(vcpu, false);
466 kvm_vcpu_set_dy_eligible(vcpu, false);
467 vcpu->preempted = false;
469 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
472 void kvm_vcpu_destroy(struct kvm_vcpu *vcpu)
474 kvm_arch_vcpu_destroy(vcpu);
477 * No need for rcu_read_lock as VCPU_RUN is the only place that changes
478 * the vcpu->pid pointer, and at destruction time all file descriptors
481 put_pid(rcu_dereference_protected(vcpu->pid, 1));
483 free_page((unsigned long)vcpu->run);
484 kmem_cache_free(kvm_vcpu_cache, vcpu);
486 EXPORT_SYMBOL_GPL(kvm_vcpu_destroy);
488 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
489 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
491 return container_of(mn, struct kvm, mmu_notifier);
494 static void kvm_mmu_notifier_invalidate_range(struct mmu_notifier *mn,
495 struct mm_struct *mm,
496 unsigned long start, unsigned long end)
498 struct kvm *kvm = mmu_notifier_to_kvm(mn);
501 idx = srcu_read_lock(&kvm->srcu);
502 kvm_arch_mmu_notifier_invalidate_range(kvm, start, end);
503 srcu_read_unlock(&kvm->srcu, idx);
506 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
507 struct mm_struct *mm,
508 unsigned long address,
511 struct kvm *kvm = mmu_notifier_to_kvm(mn);
514 idx = srcu_read_lock(&kvm->srcu);
515 spin_lock(&kvm->mmu_lock);
516 kvm->mmu_notifier_seq++;
518 if (kvm_set_spte_hva(kvm, address, pte))
519 kvm_flush_remote_tlbs(kvm);
521 spin_unlock(&kvm->mmu_lock);
522 srcu_read_unlock(&kvm->srcu, idx);
525 static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
526 const struct mmu_notifier_range *range)
528 struct kvm *kvm = mmu_notifier_to_kvm(mn);
529 int need_tlb_flush = 0, idx;
531 idx = srcu_read_lock(&kvm->srcu);
532 spin_lock(&kvm->mmu_lock);
534 * The count increase must become visible at unlock time as no
535 * spte can be established without taking the mmu_lock and
536 * count is also read inside the mmu_lock critical section.
538 kvm->mmu_notifier_count++;
539 need_tlb_flush = kvm_unmap_hva_range(kvm, range->start, range->end,
541 /* we've to flush the tlb before the pages can be freed */
542 if (need_tlb_flush || kvm->tlbs_dirty)
543 kvm_flush_remote_tlbs(kvm);
545 spin_unlock(&kvm->mmu_lock);
546 kvm_arch_guest_memory_reclaimed(kvm);
547 srcu_read_unlock(&kvm->srcu, idx);
552 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
553 const struct mmu_notifier_range *range)
555 struct kvm *kvm = mmu_notifier_to_kvm(mn);
557 spin_lock(&kvm->mmu_lock);
559 * This sequence increase will notify the kvm page fault that
560 * the page that is going to be mapped in the spte could have
563 kvm->mmu_notifier_seq++;
566 * The above sequence increase must be visible before the
567 * below count decrease, which is ensured by the smp_wmb above
568 * in conjunction with the smp_rmb in mmu_notifier_retry().
570 kvm->mmu_notifier_count--;
571 spin_unlock(&kvm->mmu_lock);
573 BUG_ON(kvm->mmu_notifier_count < 0);
576 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
577 struct mm_struct *mm,
581 struct kvm *kvm = mmu_notifier_to_kvm(mn);
584 idx = srcu_read_lock(&kvm->srcu);
585 spin_lock(&kvm->mmu_lock);
587 young = kvm_age_hva(kvm, start, end);
589 kvm_flush_remote_tlbs(kvm);
591 spin_unlock(&kvm->mmu_lock);
592 srcu_read_unlock(&kvm->srcu, idx);
597 static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
598 struct mm_struct *mm,
602 struct kvm *kvm = mmu_notifier_to_kvm(mn);
605 idx = srcu_read_lock(&kvm->srcu);
606 spin_lock(&kvm->mmu_lock);
608 * Even though we do not flush TLB, this will still adversely
609 * affect performance on pre-Haswell Intel EPT, where there is
610 * no EPT Access Bit to clear so that we have to tear down EPT
611 * tables instead. If we find this unacceptable, we can always
612 * add a parameter to kvm_age_hva so that it effectively doesn't
613 * do anything on clear_young.
615 * Also note that currently we never issue secondary TLB flushes
616 * from clear_young, leaving this job up to the regular system
617 * cadence. If we find this inaccurate, we might come up with a
618 * more sophisticated heuristic later.
620 young = kvm_age_hva(kvm, start, end);
621 spin_unlock(&kvm->mmu_lock);
622 srcu_read_unlock(&kvm->srcu, idx);
627 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
628 struct mm_struct *mm,
629 unsigned long address)
631 struct kvm *kvm = mmu_notifier_to_kvm(mn);
634 idx = srcu_read_lock(&kvm->srcu);
635 spin_lock(&kvm->mmu_lock);
636 young = kvm_test_age_hva(kvm, address);
637 spin_unlock(&kvm->mmu_lock);
638 srcu_read_unlock(&kvm->srcu, idx);
643 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
644 struct mm_struct *mm)
646 struct kvm *kvm = mmu_notifier_to_kvm(mn);
649 idx = srcu_read_lock(&kvm->srcu);
650 kvm_flush_shadow_all(kvm);
651 srcu_read_unlock(&kvm->srcu, idx);
654 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
655 .invalidate_range = kvm_mmu_notifier_invalidate_range,
656 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
657 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
658 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
659 .clear_young = kvm_mmu_notifier_clear_young,
660 .test_young = kvm_mmu_notifier_test_young,
661 .change_pte = kvm_mmu_notifier_change_pte,
662 .release = kvm_mmu_notifier_release,
665 static int kvm_init_mmu_notifier(struct kvm *kvm)
667 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
668 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
671 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
673 static int kvm_init_mmu_notifier(struct kvm *kvm)
678 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
680 static struct kvm_memslots *kvm_alloc_memslots(void)
683 struct kvm_memslots *slots;
685 slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL_ACCOUNT);
689 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
690 slots->id_to_index[i] = -1;
695 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
697 if (!memslot->dirty_bitmap)
700 kvfree(memslot->dirty_bitmap);
701 memslot->dirty_bitmap = NULL;
704 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
706 kvm_destroy_dirty_bitmap(slot);
708 kvm_arch_free_memslot(kvm, slot);
714 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
716 struct kvm_memory_slot *memslot;
721 kvm_for_each_memslot(memslot, slots)
722 kvm_free_memslot(kvm, memslot);
727 static void kvm_destroy_vm_debugfs(struct kvm *kvm)
731 if (!kvm->debugfs_dentry)
734 debugfs_remove_recursive(kvm->debugfs_dentry);
736 if (kvm->debugfs_stat_data) {
737 for (i = 0; i < kvm_debugfs_num_entries; i++)
738 kfree(kvm->debugfs_stat_data[i]);
739 kfree(kvm->debugfs_stat_data);
743 static int kvm_create_vm_debugfs(struct kvm *kvm, int fd)
745 static DEFINE_MUTEX(kvm_debugfs_lock);
747 char dir_name[ITOA_MAX_LEN * 2];
748 struct kvm_stat_data *stat_data;
749 struct kvm_stats_debugfs_item *p;
751 if (!debugfs_initialized())
754 snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd);
755 mutex_lock(&kvm_debugfs_lock);
756 dent = debugfs_lookup(dir_name, kvm_debugfs_dir);
758 pr_warn_ratelimited("KVM: debugfs: duplicate directory %s\n", dir_name);
760 mutex_unlock(&kvm_debugfs_lock);
763 dent = debugfs_create_dir(dir_name, kvm_debugfs_dir);
764 mutex_unlock(&kvm_debugfs_lock);
768 kvm->debugfs_dentry = dent;
769 kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries,
770 sizeof(*kvm->debugfs_stat_data),
772 if (!kvm->debugfs_stat_data)
775 for (p = debugfs_entries; p->name; p++) {
776 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT);
780 stat_data->kvm = kvm;
781 stat_data->dbgfs_item = p;
782 kvm->debugfs_stat_data[p - debugfs_entries] = stat_data;
783 debugfs_create_file(p->name, KVM_DBGFS_GET_MODE(p),
784 kvm->debugfs_dentry, stat_data,
791 * Called after the VM is otherwise initialized, but just before adding it to
794 int __weak kvm_arch_post_init_vm(struct kvm *kvm)
800 * Called just after removing the VM from the vm_list, but before doing any
803 void __weak kvm_arch_pre_destroy_vm(struct kvm *kvm)
807 static struct kvm *kvm_create_vm(unsigned long type)
809 struct kvm *kvm = kvm_arch_alloc_vm();
814 return ERR_PTR(-ENOMEM);
816 spin_lock_init(&kvm->mmu_lock);
818 kvm->mm = current->mm;
819 kvm_eventfd_init(kvm);
820 mutex_init(&kvm->lock);
821 mutex_init(&kvm->irq_lock);
822 mutex_init(&kvm->slots_lock);
823 INIT_LIST_HEAD(&kvm->devices);
825 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
827 if (init_srcu_struct(&kvm->srcu))
828 goto out_err_no_srcu;
829 if (init_srcu_struct(&kvm->irq_srcu))
830 goto out_err_no_irq_srcu;
832 refcount_set(&kvm->users_count, 1);
833 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
834 struct kvm_memslots *slots = kvm_alloc_memslots();
837 goto out_err_no_arch_destroy_vm;
838 /* Generations must be different for each address space. */
839 slots->generation = i;
840 rcu_assign_pointer(kvm->memslots[i], slots);
843 for (i = 0; i < KVM_NR_BUSES; i++) {
844 rcu_assign_pointer(kvm->buses[i],
845 kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL_ACCOUNT));
847 goto out_err_no_arch_destroy_vm;
850 kvm->max_halt_poll_ns = halt_poll_ns;
852 r = kvm_arch_init_vm(kvm, type);
854 goto out_err_no_arch_destroy_vm;
856 r = hardware_enable_all();
858 goto out_err_no_disable;
860 #ifdef CONFIG_HAVE_KVM_IRQFD
861 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
864 r = kvm_init_mmu_notifier(kvm);
866 goto out_err_no_mmu_notifier;
868 r = kvm_arch_post_init_vm(kvm);
872 mutex_lock(&kvm_lock);
873 list_add(&kvm->vm_list, &vm_list);
874 mutex_unlock(&kvm_lock);
876 preempt_notifier_inc();
879 * When the fd passed to this ioctl() is opened it pins the module,
880 * but try_module_get() also prevents getting a reference if the module
881 * is in MODULE_STATE_GOING (e.g. if someone ran "rmmod --wait").
883 if (!try_module_get(kvm_chardev_ops.owner)) {
891 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
892 if (kvm->mmu_notifier.ops)
893 mmu_notifier_unregister(&kvm->mmu_notifier, current->mm);
895 out_err_no_mmu_notifier:
896 hardware_disable_all();
898 kvm_arch_destroy_vm(kvm);
899 out_err_no_arch_destroy_vm:
900 WARN_ON_ONCE(!refcount_dec_and_test(&kvm->users_count));
901 for (i = 0; i < KVM_NR_BUSES; i++)
902 kfree(kvm_get_bus(kvm, i));
903 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
904 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
905 cleanup_srcu_struct(&kvm->irq_srcu);
907 cleanup_srcu_struct(&kvm->srcu);
909 kvm_arch_free_vm(kvm);
914 static void kvm_destroy_devices(struct kvm *kvm)
916 struct kvm_device *dev, *tmp;
919 * We do not need to take the kvm->lock here, because nobody else
920 * has a reference to the struct kvm at this point and therefore
921 * cannot access the devices list anyhow.
923 list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
924 list_del(&dev->vm_node);
925 dev->ops->destroy(dev);
929 static void kvm_destroy_vm(struct kvm *kvm)
932 struct mm_struct *mm = kvm->mm;
934 kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm);
935 kvm_destroy_vm_debugfs(kvm);
936 kvm_arch_sync_events(kvm);
937 mutex_lock(&kvm_lock);
938 list_del(&kvm->vm_list);
939 mutex_unlock(&kvm_lock);
940 kvm_arch_pre_destroy_vm(kvm);
942 kvm_free_irq_routing(kvm);
943 for (i = 0; i < KVM_NR_BUSES; i++) {
944 struct kvm_io_bus *bus = kvm_get_bus(kvm, i);
947 kvm_io_bus_destroy(bus);
948 kvm->buses[i] = NULL;
950 kvm_coalesced_mmio_free(kvm);
951 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
952 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
954 kvm_flush_shadow_all(kvm);
956 kvm_arch_destroy_vm(kvm);
957 kvm_destroy_devices(kvm);
958 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
959 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
960 cleanup_srcu_struct(&kvm->irq_srcu);
961 cleanup_srcu_struct(&kvm->srcu);
962 kvm_arch_free_vm(kvm);
963 preempt_notifier_dec();
964 hardware_disable_all();
966 module_put(kvm_chardev_ops.owner);
969 void kvm_get_kvm(struct kvm *kvm)
971 refcount_inc(&kvm->users_count);
973 EXPORT_SYMBOL_GPL(kvm_get_kvm);
975 void kvm_put_kvm(struct kvm *kvm)
977 if (refcount_dec_and_test(&kvm->users_count))
980 EXPORT_SYMBOL_GPL(kvm_put_kvm);
983 * Used to put a reference that was taken on behalf of an object associated
984 * with a user-visible file descriptor, e.g. a vcpu or device, if installation
985 * of the new file descriptor fails and the reference cannot be transferred to
986 * its final owner. In such cases, the caller is still actively using @kvm and
987 * will fail miserably if the refcount unexpectedly hits zero.
989 void kvm_put_kvm_no_destroy(struct kvm *kvm)
991 WARN_ON(refcount_dec_and_test(&kvm->users_count));
993 EXPORT_SYMBOL_GPL(kvm_put_kvm_no_destroy);
995 static int kvm_vm_release(struct inode *inode, struct file *filp)
997 struct kvm *kvm = filp->private_data;
999 kvm_irqfd_release(kvm);
1006 * Allocation size is twice as large as the actual dirty bitmap size.
1007 * See kvm_vm_ioctl_get_dirty_log() why this is needed.
1009 static int kvm_alloc_dirty_bitmap(struct kvm_memory_slot *memslot)
1011 unsigned long dirty_bytes = kvm_dirty_bitmap_bytes(memslot);
1013 memslot->dirty_bitmap = __vcalloc(2, dirty_bytes, GFP_KERNEL_ACCOUNT);
1014 if (!memslot->dirty_bitmap)
1021 * Delete a memslot by decrementing the number of used slots and shifting all
1022 * other entries in the array forward one spot.
1024 static inline void kvm_memslot_delete(struct kvm_memslots *slots,
1025 struct kvm_memory_slot *memslot)
1027 struct kvm_memory_slot *mslots = slots->memslots;
1030 if (WARN_ON(slots->id_to_index[memslot->id] == -1))
1033 slots->used_slots--;
1035 if (atomic_read(&slots->lru_slot) >= slots->used_slots)
1036 atomic_set(&slots->lru_slot, 0);
1038 for (i = slots->id_to_index[memslot->id]; i < slots->used_slots; i++) {
1039 mslots[i] = mslots[i + 1];
1040 slots->id_to_index[mslots[i].id] = i;
1042 mslots[i] = *memslot;
1043 slots->id_to_index[memslot->id] = -1;
1047 * "Insert" a new memslot by incrementing the number of used slots. Returns
1048 * the new slot's initial index into the memslots array.
1050 static inline int kvm_memslot_insert_back(struct kvm_memslots *slots)
1052 return slots->used_slots++;
1056 * Move a changed memslot backwards in the array by shifting existing slots
1057 * with a higher GFN toward the front of the array. Note, the changed memslot
1058 * itself is not preserved in the array, i.e. not swapped at this time, only
1059 * its new index into the array is tracked. Returns the changed memslot's
1060 * current index into the memslots array.
1062 static inline int kvm_memslot_move_backward(struct kvm_memslots *slots,
1063 struct kvm_memory_slot *memslot)
1065 struct kvm_memory_slot *mslots = slots->memslots;
1068 if (WARN_ON_ONCE(slots->id_to_index[memslot->id] == -1) ||
1069 WARN_ON_ONCE(!slots->used_slots))
1073 * Move the target memslot backward in the array by shifting existing
1074 * memslots with a higher GFN (than the target memslot) towards the
1075 * front of the array.
1077 for (i = slots->id_to_index[memslot->id]; i < slots->used_slots - 1; i++) {
1078 if (memslot->base_gfn > mslots[i + 1].base_gfn)
1081 WARN_ON_ONCE(memslot->base_gfn == mslots[i + 1].base_gfn);
1083 /* Shift the next memslot forward one and update its index. */
1084 mslots[i] = mslots[i + 1];
1085 slots->id_to_index[mslots[i].id] = i;
1091 * Move a changed memslot forwards in the array by shifting existing slots with
1092 * a lower GFN toward the back of the array. Note, the changed memslot itself
1093 * is not preserved in the array, i.e. not swapped at this time, only its new
1094 * index into the array is tracked. Returns the changed memslot's final index
1095 * into the memslots array.
1097 static inline int kvm_memslot_move_forward(struct kvm_memslots *slots,
1098 struct kvm_memory_slot *memslot,
1101 struct kvm_memory_slot *mslots = slots->memslots;
1104 for (i = start; i > 0; i--) {
1105 if (memslot->base_gfn < mslots[i - 1].base_gfn)
1108 WARN_ON_ONCE(memslot->base_gfn == mslots[i - 1].base_gfn);
1110 /* Shift the next memslot back one and update its index. */
1111 mslots[i] = mslots[i - 1];
1112 slots->id_to_index[mslots[i].id] = i;
1118 * Re-sort memslots based on their GFN to account for an added, deleted, or
1119 * moved memslot. Sorting memslots by GFN allows using a binary search during
1122 * IMPORTANT: Slots are sorted from highest GFN to lowest GFN! I.e. the entry
1123 * at memslots[0] has the highest GFN.
1125 * The sorting algorithm takes advantage of having initially sorted memslots
1126 * and knowing the position of the changed memslot. Sorting is also optimized
1127 * by not swapping the updated memslot and instead only shifting other memslots
1128 * and tracking the new index for the update memslot. Only once its final
1129 * index is known is the updated memslot copied into its position in the array.
1131 * - When deleting a memslot, the deleted memslot simply needs to be moved to
1132 * the end of the array.
1134 * - When creating a memslot, the algorithm "inserts" the new memslot at the
1135 * end of the array and then it forward to its correct location.
1137 * - When moving a memslot, the algorithm first moves the updated memslot
1138 * backward to handle the scenario where the memslot's GFN was changed to a
1139 * lower value. update_memslots() then falls through and runs the same flow
1140 * as creating a memslot to move the memslot forward to handle the scenario
1141 * where its GFN was changed to a higher value.
1143 * Note, slots are sorted from highest->lowest instead of lowest->highest for
1144 * historical reasons. Originally, invalid memslots where denoted by having
1145 * GFN=0, thus sorting from highest->lowest naturally sorted invalid memslots
1146 * to the end of the array. The current algorithm uses dedicated logic to
1147 * delete a memslot and thus does not rely on invalid memslots having GFN=0.
1149 * The other historical motiviation for highest->lowest was to improve the
1150 * performance of memslot lookup. KVM originally used a linear search starting
1151 * at memslots[0]. On x86, the largest memslot usually has one of the highest,
1152 * if not *the* highest, GFN, as the bulk of the guest's RAM is located in a
1153 * single memslot above the 4gb boundary. As the largest memslot is also the
1154 * most likely to be referenced, sorting it to the front of the array was
1155 * advantageous. The current binary search starts from the middle of the array
1156 * and uses an LRU pointer to improve performance for all memslots and GFNs.
1158 static void update_memslots(struct kvm_memslots *slots,
1159 struct kvm_memory_slot *memslot,
1160 enum kvm_mr_change change)
1164 if (change == KVM_MR_DELETE) {
1165 kvm_memslot_delete(slots, memslot);
1167 if (change == KVM_MR_CREATE)
1168 i = kvm_memslot_insert_back(slots);
1170 i = kvm_memslot_move_backward(slots, memslot);
1171 i = kvm_memslot_move_forward(slots, memslot, i);
1174 * Copy the memslot to its new position in memslots and update
1175 * its index accordingly.
1177 slots->memslots[i] = *memslot;
1178 slots->id_to_index[memslot->id] = i;
1182 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
1184 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
1186 #ifdef __KVM_HAVE_READONLY_MEM
1187 valid_flags |= KVM_MEM_READONLY;
1190 if (mem->flags & ~valid_flags)
1196 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
1197 int as_id, struct kvm_memslots *slots)
1199 struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id);
1200 u64 gen = old_memslots->generation;
1202 WARN_ON(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS);
1203 slots->generation = gen | KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
1205 rcu_assign_pointer(kvm->memslots[as_id], slots);
1206 synchronize_srcu_expedited(&kvm->srcu);
1209 * Increment the new memslot generation a second time, dropping the
1210 * update in-progress flag and incrementing the generation based on
1211 * the number of address spaces. This provides a unique and easily
1212 * identifiable generation number while the memslots are in flux.
1214 gen = slots->generation & ~KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
1217 * Generations must be unique even across address spaces. We do not need
1218 * a global counter for that, instead the generation space is evenly split
1219 * across address spaces. For example, with two address spaces, address
1220 * space 0 will use generations 0, 2, 4, ... while address space 1 will
1221 * use generations 1, 3, 5, ...
1223 gen += KVM_ADDRESS_SPACE_NUM;
1225 kvm_arch_memslots_updated(kvm, gen);
1227 slots->generation = gen;
1229 return old_memslots;
1233 * Note, at a minimum, the current number of used slots must be allocated, even
1234 * when deleting a memslot, as we need a complete duplicate of the memslots for
1235 * use when invalidating a memslot prior to deleting/moving the memslot.
1237 static struct kvm_memslots *kvm_dup_memslots(struct kvm_memslots *old,
1238 enum kvm_mr_change change)
1240 struct kvm_memslots *slots;
1241 size_t old_size, new_size;
1243 old_size = sizeof(struct kvm_memslots) +
1244 (sizeof(struct kvm_memory_slot) * old->used_slots);
1246 if (change == KVM_MR_CREATE)
1247 new_size = old_size + sizeof(struct kvm_memory_slot);
1249 new_size = old_size;
1251 slots = kvzalloc(new_size, GFP_KERNEL_ACCOUNT);
1253 memcpy(slots, old, old_size);
1258 static int kvm_set_memslot(struct kvm *kvm,
1259 const struct kvm_userspace_memory_region *mem,
1260 struct kvm_memory_slot *old,
1261 struct kvm_memory_slot *new, int as_id,
1262 enum kvm_mr_change change)
1264 struct kvm_memory_slot *slot;
1265 struct kvm_memslots *slots;
1268 slots = kvm_dup_memslots(__kvm_memslots(kvm, as_id), change);
1272 if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) {
1274 * Note, the INVALID flag needs to be in the appropriate entry
1275 * in the freshly allocated memslots, not in @old or @new.
1277 slot = id_to_memslot(slots, old->id);
1278 slot->flags |= KVM_MEMSLOT_INVALID;
1281 * We can re-use the old memslots, the only difference from the
1282 * newly installed memslots is the invalid flag, which will get
1283 * dropped by update_memslots anyway. We'll also revert to the
1284 * old memslots if preparing the new memory region fails.
1286 slots = install_new_memslots(kvm, as_id, slots);
1288 /* From this point no new shadow pages pointing to a deleted,
1289 * or moved, memslot will be created.
1291 * validation of sp->gfn happens in:
1292 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1293 * - kvm_is_visible_gfn (mmu_check_root)
1295 kvm_arch_flush_shadow_memslot(kvm, slot);
1296 kvm_arch_guest_memory_reclaimed(kvm);
1299 r = kvm_arch_prepare_memory_region(kvm, new, mem, change);
1303 update_memslots(slots, new, change);
1304 slots = install_new_memslots(kvm, as_id, slots);
1306 kvm_arch_commit_memory_region(kvm, mem, old, new, change);
1312 if (change == KVM_MR_DELETE || change == KVM_MR_MOVE)
1313 slots = install_new_memslots(kvm, as_id, slots);
1318 static int kvm_delete_memslot(struct kvm *kvm,
1319 const struct kvm_userspace_memory_region *mem,
1320 struct kvm_memory_slot *old, int as_id)
1322 struct kvm_memory_slot new;
1328 memset(&new, 0, sizeof(new));
1331 * This is only for debugging purpose; it should never be referenced
1332 * for a removed memslot.
1336 r = kvm_set_memslot(kvm, mem, old, &new, as_id, KVM_MR_DELETE);
1340 kvm_free_memslot(kvm, old);
1345 * Allocate some memory and give it an address in the guest physical address
1348 * Discontiguous memory is allowed, mostly for framebuffers.
1350 * Must be called holding kvm->slots_lock for write.
1352 int __kvm_set_memory_region(struct kvm *kvm,
1353 const struct kvm_userspace_memory_region *mem)
1355 struct kvm_memory_slot old, new;
1356 struct kvm_memory_slot *tmp;
1357 enum kvm_mr_change change;
1361 r = check_memory_region_flags(mem);
1365 as_id = mem->slot >> 16;
1366 id = (u16)mem->slot;
1368 /* General sanity checks */
1369 if ((mem->memory_size & (PAGE_SIZE - 1)) ||
1370 (mem->memory_size != (unsigned long)mem->memory_size))
1372 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
1374 /* We can read the guest memory with __xxx_user() later on. */
1375 if ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
1376 (mem->userspace_addr != untagged_addr(mem->userspace_addr)) ||
1377 !access_ok((void __user *)(unsigned long)mem->userspace_addr,
1380 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
1382 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
1386 * Make a full copy of the old memslot, the pointer will become stale
1387 * when the memslots are re-sorted by update_memslots(), and the old
1388 * memslot needs to be referenced after calling update_memslots(), e.g.
1389 * to free its resources and for arch specific behavior.
1391 tmp = id_to_memslot(__kvm_memslots(kvm, as_id), id);
1396 memset(&old, 0, sizeof(old));
1400 if (!mem->memory_size)
1401 return kvm_delete_memslot(kvm, mem, &old, as_id);
1405 new.base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
1406 new.npages = mem->memory_size >> PAGE_SHIFT;
1407 new.flags = mem->flags;
1408 new.userspace_addr = mem->userspace_addr;
1410 if (new.npages > KVM_MEM_MAX_NR_PAGES)
1414 change = KVM_MR_CREATE;
1415 new.dirty_bitmap = NULL;
1416 memset(&new.arch, 0, sizeof(new.arch));
1417 } else { /* Modify an existing slot. */
1418 if ((new.userspace_addr != old.userspace_addr) ||
1419 (new.npages != old.npages) ||
1420 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
1423 if (new.base_gfn != old.base_gfn)
1424 change = KVM_MR_MOVE;
1425 else if (new.flags != old.flags)
1426 change = KVM_MR_FLAGS_ONLY;
1427 else /* Nothing to change. */
1430 /* Copy dirty_bitmap and arch from the current memslot. */
1431 new.dirty_bitmap = old.dirty_bitmap;
1432 memcpy(&new.arch, &old.arch, sizeof(new.arch));
1435 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
1436 /* Check for overlaps */
1437 kvm_for_each_memslot(tmp, __kvm_memslots(kvm, as_id)) {
1440 if (!((new.base_gfn + new.npages <= tmp->base_gfn) ||
1441 (new.base_gfn >= tmp->base_gfn + tmp->npages)))
1446 /* Allocate/free page dirty bitmap as needed */
1447 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
1448 new.dirty_bitmap = NULL;
1449 else if (!new.dirty_bitmap) {
1450 r = kvm_alloc_dirty_bitmap(&new);
1454 if (kvm_dirty_log_manual_protect_and_init_set(kvm))
1455 bitmap_set(new.dirty_bitmap, 0, new.npages);
1458 r = kvm_set_memslot(kvm, mem, &old, &new, as_id, change);
1462 if (old.dirty_bitmap && !new.dirty_bitmap)
1463 kvm_destroy_dirty_bitmap(&old);
1467 if (new.dirty_bitmap && !old.dirty_bitmap)
1468 kvm_destroy_dirty_bitmap(&new);
1471 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1473 int kvm_set_memory_region(struct kvm *kvm,
1474 const struct kvm_userspace_memory_region *mem)
1478 mutex_lock(&kvm->slots_lock);
1479 r = __kvm_set_memory_region(kvm, mem);
1480 mutex_unlock(&kvm->slots_lock);
1483 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1485 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1486 struct kvm_userspace_memory_region *mem)
1488 if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1491 return kvm_set_memory_region(kvm, mem);
1494 #ifndef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1496 * kvm_get_dirty_log - get a snapshot of dirty pages
1497 * @kvm: pointer to kvm instance
1498 * @log: slot id and address to which we copy the log
1499 * @is_dirty: set to '1' if any dirty pages were found
1500 * @memslot: set to the associated memslot, always valid on success
1502 int kvm_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log,
1503 int *is_dirty, struct kvm_memory_slot **memslot)
1505 struct kvm_memslots *slots;
1508 unsigned long any = 0;
1513 as_id = log->slot >> 16;
1514 id = (u16)log->slot;
1515 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1518 slots = __kvm_memslots(kvm, as_id);
1519 *memslot = id_to_memslot(slots, id);
1520 if (!(*memslot) || !(*memslot)->dirty_bitmap)
1523 kvm_arch_sync_dirty_log(kvm, *memslot);
1525 n = kvm_dirty_bitmap_bytes(*memslot);
1527 for (i = 0; !any && i < n/sizeof(long); ++i)
1528 any = (*memslot)->dirty_bitmap[i];
1530 if (copy_to_user(log->dirty_bitmap, (*memslot)->dirty_bitmap, n))
1537 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1539 #else /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
1541 * kvm_get_dirty_log_protect - get a snapshot of dirty pages
1542 * and reenable dirty page tracking for the corresponding pages.
1543 * @kvm: pointer to kvm instance
1544 * @log: slot id and address to which we copy the log
1546 * We need to keep it in mind that VCPU threads can write to the bitmap
1547 * concurrently. So, to avoid losing track of dirty pages we keep the
1550 * 1. Take a snapshot of the bit and clear it if needed.
1551 * 2. Write protect the corresponding page.
1552 * 3. Copy the snapshot to the userspace.
1553 * 4. Upon return caller flushes TLB's if needed.
1555 * Between 2 and 4, the guest may write to the page using the remaining TLB
1556 * entry. This is not a problem because the page is reported dirty using
1557 * the snapshot taken before and step 4 ensures that writes done after
1558 * exiting to userspace will be logged for the next call.
1561 static int kvm_get_dirty_log_protect(struct kvm *kvm, struct kvm_dirty_log *log)
1563 struct kvm_memslots *slots;
1564 struct kvm_memory_slot *memslot;
1567 unsigned long *dirty_bitmap;
1568 unsigned long *dirty_bitmap_buffer;
1571 as_id = log->slot >> 16;
1572 id = (u16)log->slot;
1573 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1576 slots = __kvm_memslots(kvm, as_id);
1577 memslot = id_to_memslot(slots, id);
1578 if (!memslot || !memslot->dirty_bitmap)
1581 dirty_bitmap = memslot->dirty_bitmap;
1583 kvm_arch_sync_dirty_log(kvm, memslot);
1585 n = kvm_dirty_bitmap_bytes(memslot);
1587 if (kvm->manual_dirty_log_protect) {
1589 * Unlike kvm_get_dirty_log, we always return false in *flush,
1590 * because no flush is needed until KVM_CLEAR_DIRTY_LOG. There
1591 * is some code duplication between this function and
1592 * kvm_get_dirty_log, but hopefully all architecture
1593 * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log
1594 * can be eliminated.
1596 dirty_bitmap_buffer = dirty_bitmap;
1598 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
1599 memset(dirty_bitmap_buffer, 0, n);
1601 spin_lock(&kvm->mmu_lock);
1602 for (i = 0; i < n / sizeof(long); i++) {
1606 if (!dirty_bitmap[i])
1610 mask = xchg(&dirty_bitmap[i], 0);
1611 dirty_bitmap_buffer[i] = mask;
1613 offset = i * BITS_PER_LONG;
1614 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1617 spin_unlock(&kvm->mmu_lock);
1621 kvm_arch_flush_remote_tlbs_memslot(kvm, memslot);
1623 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1630 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
1631 * @kvm: kvm instance
1632 * @log: slot id and address to which we copy the log
1634 * Steps 1-4 below provide general overview of dirty page logging. See
1635 * kvm_get_dirty_log_protect() function description for additional details.
1637 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
1638 * always flush the TLB (step 4) even if previous step failed and the dirty
1639 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
1640 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
1641 * writes will be marked dirty for next log read.
1643 * 1. Take a snapshot of the bit and clear it if needed.
1644 * 2. Write protect the corresponding page.
1645 * 3. Copy the snapshot to the userspace.
1646 * 4. Flush TLB's if needed.
1648 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1649 struct kvm_dirty_log *log)
1653 mutex_lock(&kvm->slots_lock);
1655 r = kvm_get_dirty_log_protect(kvm, log);
1657 mutex_unlock(&kvm->slots_lock);
1662 * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap
1663 * and reenable dirty page tracking for the corresponding pages.
1664 * @kvm: pointer to kvm instance
1665 * @log: slot id and address from which to fetch the bitmap of dirty pages
1667 static int kvm_clear_dirty_log_protect(struct kvm *kvm,
1668 struct kvm_clear_dirty_log *log)
1670 struct kvm_memslots *slots;
1671 struct kvm_memory_slot *memslot;
1675 unsigned long *dirty_bitmap;
1676 unsigned long *dirty_bitmap_buffer;
1679 as_id = log->slot >> 16;
1680 id = (u16)log->slot;
1681 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1684 if (log->first_page & 63)
1687 slots = __kvm_memslots(kvm, as_id);
1688 memslot = id_to_memslot(slots, id);
1689 if (!memslot || !memslot->dirty_bitmap)
1692 dirty_bitmap = memslot->dirty_bitmap;
1694 n = ALIGN(log->num_pages, BITS_PER_LONG) / 8;
1696 if (log->first_page > memslot->npages ||
1697 log->num_pages > memslot->npages - log->first_page ||
1698 (log->num_pages < memslot->npages - log->first_page && (log->num_pages & 63)))
1701 kvm_arch_sync_dirty_log(kvm, memslot);
1704 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
1705 if (copy_from_user(dirty_bitmap_buffer, log->dirty_bitmap, n))
1708 spin_lock(&kvm->mmu_lock);
1709 for (offset = log->first_page, i = offset / BITS_PER_LONG,
1710 n = DIV_ROUND_UP(log->num_pages, BITS_PER_LONG); n--;
1711 i++, offset += BITS_PER_LONG) {
1712 unsigned long mask = *dirty_bitmap_buffer++;
1713 atomic_long_t *p = (atomic_long_t *) &dirty_bitmap[i];
1717 mask &= atomic_long_fetch_andnot(mask, p);
1720 * mask contains the bits that really have been cleared. This
1721 * never includes any bits beyond the length of the memslot (if
1722 * the length is not aligned to 64 pages), therefore it is not
1723 * a problem if userspace sets them in log->dirty_bitmap.
1727 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1731 spin_unlock(&kvm->mmu_lock);
1734 kvm_arch_flush_remote_tlbs_memslot(kvm, memslot);
1739 static int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm,
1740 struct kvm_clear_dirty_log *log)
1744 mutex_lock(&kvm->slots_lock);
1746 r = kvm_clear_dirty_log_protect(kvm, log);
1748 mutex_unlock(&kvm->slots_lock);
1751 #endif /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
1753 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1755 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1757 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1759 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
1761 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn);
1764 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1766 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1768 return kvm_is_visible_memslot(memslot);
1770 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1772 bool kvm_vcpu_is_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
1774 struct kvm_memory_slot *memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1776 return kvm_is_visible_memslot(memslot);
1778 EXPORT_SYMBOL_GPL(kvm_vcpu_is_visible_gfn);
1780 unsigned long kvm_host_page_size(struct kvm_vcpu *vcpu, gfn_t gfn)
1782 struct vm_area_struct *vma;
1783 unsigned long addr, size;
1787 addr = kvm_vcpu_gfn_to_hva_prot(vcpu, gfn, NULL);
1788 if (kvm_is_error_hva(addr))
1791 mmap_read_lock(current->mm);
1792 vma = find_vma(current->mm, addr);
1796 size = vma_kernel_pagesize(vma);
1799 mmap_read_unlock(current->mm);
1804 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1806 return slot->flags & KVM_MEM_READONLY;
1809 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1810 gfn_t *nr_pages, bool write)
1812 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1813 return KVM_HVA_ERR_BAD;
1815 if (memslot_is_readonly(slot) && write)
1816 return KVM_HVA_ERR_RO_BAD;
1819 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1821 return __gfn_to_hva_memslot(slot, gfn);
1824 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1827 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1830 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1833 return gfn_to_hva_many(slot, gfn, NULL);
1835 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1837 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1839 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1841 EXPORT_SYMBOL_GPL(gfn_to_hva);
1843 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
1845 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
1847 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
1850 * Return the hva of a @gfn and the R/W attribute if possible.
1852 * @slot: the kvm_memory_slot which contains @gfn
1853 * @gfn: the gfn to be translated
1854 * @writable: used to return the read/write attribute of the @slot if the hva
1855 * is valid and @writable is not NULL
1857 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1858 gfn_t gfn, bool *writable)
1860 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1862 if (!kvm_is_error_hva(hva) && writable)
1863 *writable = !memslot_is_readonly(slot);
1868 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1870 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1872 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1875 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
1877 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1879 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1882 static inline int check_user_page_hwpoison(unsigned long addr)
1884 int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
1886 rc = get_user_pages(addr, 1, flags, NULL, NULL);
1887 return rc == -EHWPOISON;
1891 * The fast path to get the writable pfn which will be stored in @pfn,
1892 * true indicates success, otherwise false is returned. It's also the
1893 * only part that runs if we can in atomic context.
1895 static bool hva_to_pfn_fast(unsigned long addr, bool write_fault,
1896 bool *writable, kvm_pfn_t *pfn)
1898 struct page *page[1];
1901 * Fast pin a writable pfn only if it is a write fault request
1902 * or the caller allows to map a writable pfn for a read fault
1905 if (!(write_fault || writable))
1908 if (get_user_page_fast_only(addr, FOLL_WRITE, page)) {
1909 *pfn = page_to_pfn(page[0]);
1920 * The slow path to get the pfn of the specified host virtual address,
1921 * 1 indicates success, -errno is returned if error is detected.
1923 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1924 bool *writable, kvm_pfn_t *pfn)
1926 unsigned int flags = FOLL_HWPOISON;
1933 *writable = write_fault;
1936 flags |= FOLL_WRITE;
1938 flags |= FOLL_NOWAIT;
1940 npages = get_user_pages_unlocked(addr, 1, &page, flags);
1944 /* map read fault as writable if possible */
1945 if (unlikely(!write_fault) && writable) {
1948 if (get_user_page_fast_only(addr, FOLL_WRITE, &wpage)) {
1954 *pfn = page_to_pfn(page);
1958 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1960 if (unlikely(!(vma->vm_flags & VM_READ)))
1963 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1969 static int kvm_try_get_pfn(kvm_pfn_t pfn)
1971 if (kvm_is_reserved_pfn(pfn))
1973 return get_page_unless_zero(pfn_to_page(pfn));
1976 static int hva_to_pfn_remapped(struct vm_area_struct *vma,
1977 unsigned long addr, bool *async,
1978 bool write_fault, bool *writable,
1986 r = follow_pte(vma->vm_mm, addr, &ptep, &ptl);
1989 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
1990 * not call the fault handler, so do it here.
1992 bool unlocked = false;
1993 r = fixup_user_fault(current->mm, addr,
1994 (write_fault ? FAULT_FLAG_WRITE : 0),
2001 r = follow_pte(vma->vm_mm, addr, &ptep, &ptl);
2006 if (write_fault && !pte_write(*ptep)) {
2007 pfn = KVM_PFN_ERR_RO_FAULT;
2012 *writable = pte_write(*ptep);
2013 pfn = pte_pfn(*ptep);
2016 * Get a reference here because callers of *hva_to_pfn* and
2017 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
2018 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
2019 * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will
2020 * simply do nothing for reserved pfns.
2022 * Whoever called remap_pfn_range is also going to call e.g.
2023 * unmap_mapping_range before the underlying pages are freed,
2024 * causing a call to our MMU notifier.
2026 * Certain IO or PFNMAP mappings can be backed with valid
2027 * struct pages, but be allocated without refcounting e.g.,
2028 * tail pages of non-compound higher order allocations, which
2029 * would then underflow the refcount when the caller does the
2030 * required put_page. Don't allow those pages here.
2032 if (!kvm_try_get_pfn(pfn))
2036 pte_unmap_unlock(ptep, ptl);
2043 * Pin guest page in memory and return its pfn.
2044 * @addr: host virtual address which maps memory to the guest
2045 * @atomic: whether this function can sleep
2046 * @async: whether this function need to wait IO complete if the
2047 * host page is not in the memory
2048 * @write_fault: whether we should get a writable host page
2049 * @writable: whether it allows to map a writable host page for !@write_fault
2051 * The function will map a writable host page for these two cases:
2052 * 1): @write_fault = true
2053 * 2): @write_fault = false && @writable, @writable will tell the caller
2054 * whether the mapping is writable.
2056 static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
2057 bool write_fault, bool *writable)
2059 struct vm_area_struct *vma;
2063 /* we can do it either atomically or asynchronously, not both */
2064 BUG_ON(atomic && async);
2066 if (hva_to_pfn_fast(addr, write_fault, writable, &pfn))
2070 return KVM_PFN_ERR_FAULT;
2072 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
2076 mmap_read_lock(current->mm);
2077 if (npages == -EHWPOISON ||
2078 (!async && check_user_page_hwpoison(addr))) {
2079 pfn = KVM_PFN_ERR_HWPOISON;
2084 vma = find_vma_intersection(current->mm, addr, addr + 1);
2087 pfn = KVM_PFN_ERR_FAULT;
2088 else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
2089 r = hva_to_pfn_remapped(vma, addr, async, write_fault, writable, &pfn);
2093 pfn = KVM_PFN_ERR_FAULT;
2095 if (async && vma_is_valid(vma, write_fault))
2097 pfn = KVM_PFN_ERR_FAULT;
2100 mmap_read_unlock(current->mm);
2104 kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn,
2105 bool atomic, bool *async, bool write_fault,
2108 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
2110 if (addr == KVM_HVA_ERR_RO_BAD) {
2113 return KVM_PFN_ERR_RO_FAULT;
2116 if (kvm_is_error_hva(addr)) {
2119 return KVM_PFN_NOSLOT;
2122 /* Do not map writable pfn in the readonly memslot. */
2123 if (writable && memslot_is_readonly(slot)) {
2128 return hva_to_pfn(addr, atomic, async, write_fault,
2131 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
2133 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
2136 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
2137 write_fault, writable);
2139 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
2141 kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
2143 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
2145 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
2147 kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
2149 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
2151 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
2153 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
2155 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
2157 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
2159 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
2161 return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
2163 EXPORT_SYMBOL_GPL(gfn_to_pfn);
2165 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
2167 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
2169 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
2171 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
2172 struct page **pages, int nr_pages)
2177 addr = gfn_to_hva_many(slot, gfn, &entry);
2178 if (kvm_is_error_hva(addr))
2181 if (entry < nr_pages)
2184 return get_user_pages_fast_only(addr, nr_pages, FOLL_WRITE, pages);
2186 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
2188 static struct page *kvm_pfn_to_page(kvm_pfn_t pfn)
2190 if (is_error_noslot_pfn(pfn))
2191 return KVM_ERR_PTR_BAD_PAGE;
2193 if (kvm_is_reserved_pfn(pfn)) {
2195 return KVM_ERR_PTR_BAD_PAGE;
2198 return pfn_to_page(pfn);
2201 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
2205 pfn = gfn_to_pfn(kvm, gfn);
2207 return kvm_pfn_to_page(pfn);
2209 EXPORT_SYMBOL_GPL(gfn_to_page);
2211 void kvm_release_pfn(kvm_pfn_t pfn, bool dirty, struct gfn_to_pfn_cache *cache)
2217 cache->pfn = cache->gfn = 0;
2220 kvm_release_pfn_dirty(pfn);
2222 kvm_release_pfn_clean(pfn);
2225 static void kvm_cache_gfn_to_pfn(struct kvm_memory_slot *slot, gfn_t gfn,
2226 struct gfn_to_pfn_cache *cache, u64 gen)
2228 kvm_release_pfn(cache->pfn, cache->dirty, cache);
2230 cache->pfn = gfn_to_pfn_memslot(slot, gfn);
2232 cache->dirty = false;
2233 cache->generation = gen;
2236 static int __kvm_map_gfn(struct kvm_memslots *slots, gfn_t gfn,
2237 struct kvm_host_map *map,
2238 struct gfn_to_pfn_cache *cache,
2243 struct page *page = KVM_UNMAPPED_PAGE;
2244 struct kvm_memory_slot *slot = __gfn_to_memslot(slots, gfn);
2245 u64 gen = slots->generation;
2251 if (!cache->pfn || cache->gfn != gfn ||
2252 cache->generation != gen) {
2255 kvm_cache_gfn_to_pfn(slot, gfn, cache, gen);
2261 pfn = gfn_to_pfn_memslot(slot, gfn);
2263 if (is_error_noslot_pfn(pfn))
2266 if (pfn_valid(pfn)) {
2267 page = pfn_to_page(pfn);
2269 hva = kmap_atomic(page);
2272 #ifdef CONFIG_HAS_IOMEM
2273 } else if (!atomic) {
2274 hva = memremap(pfn_to_hpa(pfn), PAGE_SIZE, MEMREMAP_WB);
2291 int kvm_map_gfn(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map,
2292 struct gfn_to_pfn_cache *cache, bool atomic)
2294 return __kvm_map_gfn(kvm_memslots(vcpu->kvm), gfn, map,
2297 EXPORT_SYMBOL_GPL(kvm_map_gfn);
2299 int kvm_vcpu_map(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map)
2301 return __kvm_map_gfn(kvm_vcpu_memslots(vcpu), gfn, map,
2304 EXPORT_SYMBOL_GPL(kvm_vcpu_map);
2306 static void __kvm_unmap_gfn(struct kvm_memory_slot *memslot,
2307 struct kvm_host_map *map,
2308 struct gfn_to_pfn_cache *cache,
2309 bool dirty, bool atomic)
2317 if (map->page != KVM_UNMAPPED_PAGE) {
2319 kunmap_atomic(map->hva);
2323 #ifdef CONFIG_HAS_IOMEM
2327 WARN_ONCE(1, "Unexpected unmapping in atomic context");
2331 mark_page_dirty_in_slot(memslot, map->gfn);
2334 cache->dirty |= dirty;
2336 kvm_release_pfn(map->pfn, dirty, NULL);
2342 int kvm_unmap_gfn(struct kvm_vcpu *vcpu, struct kvm_host_map *map,
2343 struct gfn_to_pfn_cache *cache, bool dirty, bool atomic)
2345 __kvm_unmap_gfn(gfn_to_memslot(vcpu->kvm, map->gfn), map,
2346 cache, dirty, atomic);
2349 EXPORT_SYMBOL_GPL(kvm_unmap_gfn);
2351 void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map, bool dirty)
2353 __kvm_unmap_gfn(kvm_vcpu_gfn_to_memslot(vcpu, map->gfn), map, NULL,
2356 EXPORT_SYMBOL_GPL(kvm_vcpu_unmap);
2358 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
2362 pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
2364 return kvm_pfn_to_page(pfn);
2366 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
2368 void kvm_release_page_clean(struct page *page)
2370 WARN_ON(is_error_page(page));
2372 kvm_release_pfn_clean(page_to_pfn(page));
2374 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
2376 void kvm_release_pfn_clean(kvm_pfn_t pfn)
2378 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
2379 put_page(pfn_to_page(pfn));
2381 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
2383 void kvm_release_page_dirty(struct page *page)
2385 WARN_ON(is_error_page(page));
2387 kvm_release_pfn_dirty(page_to_pfn(page));
2389 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
2391 void kvm_release_pfn_dirty(kvm_pfn_t pfn)
2393 kvm_set_pfn_dirty(pfn);
2394 kvm_release_pfn_clean(pfn);
2396 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
2398 static bool kvm_is_ad_tracked_pfn(kvm_pfn_t pfn)
2400 if (!pfn_valid(pfn))
2404 * Per page-flags.h, pages tagged PG_reserved "should in general not be
2405 * touched (e.g. set dirty) except by its owner".
2407 return !PageReserved(pfn_to_page(pfn));
2410 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
2412 if (kvm_is_ad_tracked_pfn(pfn))
2413 SetPageDirty(pfn_to_page(pfn));
2415 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
2417 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
2419 if (kvm_is_ad_tracked_pfn(pfn))
2420 mark_page_accessed(pfn_to_page(pfn));
2422 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
2424 void kvm_get_pfn(kvm_pfn_t pfn)
2426 if (!kvm_is_reserved_pfn(pfn))
2427 get_page(pfn_to_page(pfn));
2429 EXPORT_SYMBOL_GPL(kvm_get_pfn);
2431 static int next_segment(unsigned long len, int offset)
2433 if (len > PAGE_SIZE - offset)
2434 return PAGE_SIZE - offset;
2439 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
2440 void *data, int offset, int len)
2445 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
2446 if (kvm_is_error_hva(addr))
2448 r = __copy_from_user(data, (void __user *)addr + offset, len);
2454 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
2457 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2459 return __kvm_read_guest_page(slot, gfn, data, offset, len);
2461 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
2463 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
2464 int offset, int len)
2466 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2468 return __kvm_read_guest_page(slot, gfn, data, offset, len);
2470 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
2472 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
2474 gfn_t gfn = gpa >> PAGE_SHIFT;
2476 int offset = offset_in_page(gpa);
2479 while ((seg = next_segment(len, offset)) != 0) {
2480 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
2490 EXPORT_SYMBOL_GPL(kvm_read_guest);
2492 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
2494 gfn_t gfn = gpa >> PAGE_SHIFT;
2496 int offset = offset_in_page(gpa);
2499 while ((seg = next_segment(len, offset)) != 0) {
2500 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
2510 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
2512 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
2513 void *data, int offset, unsigned long len)
2518 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
2519 if (kvm_is_error_hva(addr))
2521 pagefault_disable();
2522 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
2529 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
2530 void *data, unsigned long len)
2532 gfn_t gfn = gpa >> PAGE_SHIFT;
2533 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2534 int offset = offset_in_page(gpa);
2536 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
2538 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
2540 static int __kvm_write_guest_page(struct kvm_memory_slot *memslot, gfn_t gfn,
2541 const void *data, int offset, int len)
2546 addr = gfn_to_hva_memslot(memslot, gfn);
2547 if (kvm_is_error_hva(addr))
2549 r = __copy_to_user((void __user *)addr + offset, data, len);
2552 mark_page_dirty_in_slot(memslot, gfn);
2556 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
2557 const void *data, int offset, int len)
2559 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2561 return __kvm_write_guest_page(slot, gfn, data, offset, len);
2563 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
2565 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
2566 const void *data, int offset, int len)
2568 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2570 return __kvm_write_guest_page(slot, gfn, data, offset, len);
2572 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
2574 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
2577 gfn_t gfn = gpa >> PAGE_SHIFT;
2579 int offset = offset_in_page(gpa);
2582 while ((seg = next_segment(len, offset)) != 0) {
2583 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
2593 EXPORT_SYMBOL_GPL(kvm_write_guest);
2595 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
2598 gfn_t gfn = gpa >> PAGE_SHIFT;
2600 int offset = offset_in_page(gpa);
2603 while ((seg = next_segment(len, offset)) != 0) {
2604 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
2614 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
2616 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
2617 struct gfn_to_hva_cache *ghc,
2618 gpa_t gpa, unsigned long len)
2620 int offset = offset_in_page(gpa);
2621 gfn_t start_gfn = gpa >> PAGE_SHIFT;
2622 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
2623 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
2624 gfn_t nr_pages_avail;
2626 /* Update ghc->generation before performing any error checks. */
2627 ghc->generation = slots->generation;
2629 if (start_gfn > end_gfn) {
2630 ghc->hva = KVM_HVA_ERR_BAD;
2635 * If the requested region crosses two memslots, we still
2636 * verify that the entire region is valid here.
2638 for ( ; start_gfn <= end_gfn; start_gfn += nr_pages_avail) {
2639 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
2640 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
2642 if (kvm_is_error_hva(ghc->hva))
2646 /* Use the slow path for cross page reads and writes. */
2647 if (nr_pages_needed == 1)
2650 ghc->memslot = NULL;
2657 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2658 gpa_t gpa, unsigned long len)
2660 struct kvm_memslots *slots = kvm_memslots(kvm);
2661 return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len);
2663 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
2665 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2666 void *data, unsigned int offset,
2669 struct kvm_memslots *slots = kvm_memslots(kvm);
2671 gpa_t gpa = ghc->gpa + offset;
2673 if (WARN_ON_ONCE(len + offset > ghc->len))
2676 if (slots->generation != ghc->generation) {
2677 if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len))
2681 if (kvm_is_error_hva(ghc->hva))
2684 if (unlikely(!ghc->memslot))
2685 return kvm_write_guest(kvm, gpa, data, len);
2687 r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
2690 mark_page_dirty_in_slot(ghc->memslot, gpa >> PAGE_SHIFT);
2694 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached);
2696 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2697 void *data, unsigned long len)
2699 return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len);
2701 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
2703 int kvm_read_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2704 void *data, unsigned int offset,
2707 struct kvm_memslots *slots = kvm_memslots(kvm);
2709 gpa_t gpa = ghc->gpa + offset;
2711 if (WARN_ON_ONCE(len + offset > ghc->len))
2714 if (slots->generation != ghc->generation) {
2715 if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len))
2719 if (kvm_is_error_hva(ghc->hva))
2722 if (unlikely(!ghc->memslot))
2723 return kvm_read_guest(kvm, gpa, data, len);
2725 r = __copy_from_user(data, (void __user *)ghc->hva + offset, len);
2731 EXPORT_SYMBOL_GPL(kvm_read_guest_offset_cached);
2733 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2734 void *data, unsigned long len)
2736 return kvm_read_guest_offset_cached(kvm, ghc, data, 0, len);
2738 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
2740 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
2742 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
2744 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
2746 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
2748 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
2750 gfn_t gfn = gpa >> PAGE_SHIFT;
2752 int offset = offset_in_page(gpa);
2755 while ((seg = next_segment(len, offset)) != 0) {
2756 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
2765 EXPORT_SYMBOL_GPL(kvm_clear_guest);
2767 void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot, gfn_t gfn)
2769 if (memslot && memslot->dirty_bitmap) {
2770 unsigned long rel_gfn = gfn - memslot->base_gfn;
2772 set_bit_le(rel_gfn, memslot->dirty_bitmap);
2775 EXPORT_SYMBOL_GPL(mark_page_dirty_in_slot);
2777 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
2779 struct kvm_memory_slot *memslot;
2781 memslot = gfn_to_memslot(kvm, gfn);
2782 mark_page_dirty_in_slot(memslot, gfn);
2784 EXPORT_SYMBOL_GPL(mark_page_dirty);
2786 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
2788 struct kvm_memory_slot *memslot;
2790 memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2791 mark_page_dirty_in_slot(memslot, gfn);
2793 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
2795 void kvm_sigset_activate(struct kvm_vcpu *vcpu)
2797 if (!vcpu->sigset_active)
2801 * This does a lockless modification of ->real_blocked, which is fine
2802 * because, only current can change ->real_blocked and all readers of
2803 * ->real_blocked don't care as long ->real_blocked is always a subset
2806 sigprocmask(SIG_SETMASK, &vcpu->sigset, ¤t->real_blocked);
2809 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu)
2811 if (!vcpu->sigset_active)
2814 sigprocmask(SIG_SETMASK, ¤t->real_blocked, NULL);
2815 sigemptyset(¤t->real_blocked);
2818 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
2820 unsigned int old, val, grow, grow_start;
2822 old = val = vcpu->halt_poll_ns;
2823 grow_start = READ_ONCE(halt_poll_ns_grow_start);
2824 grow = READ_ONCE(halt_poll_ns_grow);
2829 if (val < grow_start)
2832 if (val > vcpu->kvm->max_halt_poll_ns)
2833 val = vcpu->kvm->max_halt_poll_ns;
2835 vcpu->halt_poll_ns = val;
2837 trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
2840 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
2842 unsigned int old, val, shrink, grow_start;
2844 old = val = vcpu->halt_poll_ns;
2845 shrink = READ_ONCE(halt_poll_ns_shrink);
2846 grow_start = READ_ONCE(halt_poll_ns_grow_start);
2852 if (val < grow_start)
2855 vcpu->halt_poll_ns = val;
2856 trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
2859 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
2862 int idx = srcu_read_lock(&vcpu->kvm->srcu);
2864 if (kvm_arch_vcpu_runnable(vcpu)) {
2865 kvm_make_request(KVM_REQ_UNHALT, vcpu);
2868 if (kvm_cpu_has_pending_timer(vcpu))
2870 if (signal_pending(current))
2875 srcu_read_unlock(&vcpu->kvm->srcu, idx);
2880 update_halt_poll_stats(struct kvm_vcpu *vcpu, u64 poll_ns, bool waited)
2883 vcpu->stat.halt_poll_fail_ns += poll_ns;
2885 vcpu->stat.halt_poll_success_ns += poll_ns;
2889 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2891 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
2893 ktime_t start, cur, poll_end;
2894 bool waited = false;
2897 kvm_arch_vcpu_blocking(vcpu);
2899 start = cur = poll_end = ktime_get();
2900 if (vcpu->halt_poll_ns && !kvm_arch_no_poll(vcpu)) {
2901 ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns);
2903 ++vcpu->stat.halt_attempted_poll;
2906 * This sets KVM_REQ_UNHALT if an interrupt
2909 if (kvm_vcpu_check_block(vcpu) < 0) {
2910 ++vcpu->stat.halt_successful_poll;
2911 if (!vcpu_valid_wakeup(vcpu))
2912 ++vcpu->stat.halt_poll_invalid;
2915 poll_end = cur = ktime_get();
2916 } while (single_task_running() && !need_resched() &&
2917 ktime_before(cur, stop));
2920 prepare_to_rcuwait(&vcpu->wait);
2922 set_current_state(TASK_INTERRUPTIBLE);
2924 if (kvm_vcpu_check_block(vcpu) < 0)
2930 finish_rcuwait(&vcpu->wait);
2933 kvm_arch_vcpu_unblocking(vcpu);
2934 block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
2936 update_halt_poll_stats(
2937 vcpu, ktime_to_ns(ktime_sub(poll_end, start)), waited);
2939 if (!kvm_arch_no_poll(vcpu)) {
2940 if (!vcpu_valid_wakeup(vcpu)) {
2941 shrink_halt_poll_ns(vcpu);
2942 } else if (vcpu->kvm->max_halt_poll_ns) {
2943 if (block_ns <= vcpu->halt_poll_ns)
2945 /* we had a long block, shrink polling */
2946 else if (vcpu->halt_poll_ns &&
2947 block_ns > vcpu->kvm->max_halt_poll_ns)
2948 shrink_halt_poll_ns(vcpu);
2949 /* we had a short halt and our poll time is too small */
2950 else if (vcpu->halt_poll_ns < vcpu->kvm->max_halt_poll_ns &&
2951 block_ns < vcpu->kvm->max_halt_poll_ns)
2952 grow_halt_poll_ns(vcpu);
2954 vcpu->halt_poll_ns = 0;
2958 trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu));
2959 kvm_arch_vcpu_block_finish(vcpu);
2961 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
2963 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
2965 struct rcuwait *waitp;
2967 waitp = kvm_arch_vcpu_get_wait(vcpu);
2968 if (rcuwait_wake_up(waitp)) {
2969 WRITE_ONCE(vcpu->ready, true);
2970 ++vcpu->stat.halt_wakeup;
2976 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
2980 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2982 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
2986 if (kvm_vcpu_wake_up(vcpu))
2990 * Note, the vCPU could get migrated to a different pCPU at any point
2991 * after kvm_arch_vcpu_should_kick(), which could result in sending an
2992 * IPI to the previous pCPU. But, that's ok because the purpose of the
2993 * IPI is to force the vCPU to leave IN_GUEST_MODE, and migrating the
2994 * vCPU also requires it to leave IN_GUEST_MODE.
2997 if (kvm_arch_vcpu_should_kick(vcpu)) {
2998 cpu = READ_ONCE(vcpu->cpu);
2999 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
3000 smp_send_reschedule(cpu);
3004 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
3005 #endif /* !CONFIG_S390 */
3007 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
3010 struct task_struct *task = NULL;
3014 pid = rcu_dereference(target->pid);
3016 task = get_pid_task(pid, PIDTYPE_PID);
3020 ret = yield_to(task, 1);
3021 put_task_struct(task);
3025 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
3028 * Helper that checks whether a VCPU is eligible for directed yield.
3029 * Most eligible candidate to yield is decided by following heuristics:
3031 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
3032 * (preempted lock holder), indicated by @in_spin_loop.
3033 * Set at the beginning and cleared at the end of interception/PLE handler.
3035 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
3036 * chance last time (mostly it has become eligible now since we have probably
3037 * yielded to lockholder in last iteration. This is done by toggling
3038 * @dy_eligible each time a VCPU checked for eligibility.)
3040 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
3041 * to preempted lock-holder could result in wrong VCPU selection and CPU
3042 * burning. Giving priority for a potential lock-holder increases lock
3045 * Since algorithm is based on heuristics, accessing another VCPU data without
3046 * locking does not harm. It may result in trying to yield to same VCPU, fail
3047 * and continue with next VCPU and so on.
3049 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
3051 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
3054 eligible = !vcpu->spin_loop.in_spin_loop ||
3055 vcpu->spin_loop.dy_eligible;
3057 if (vcpu->spin_loop.in_spin_loop)
3058 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
3067 * Unlike kvm_arch_vcpu_runnable, this function is called outside
3068 * a vcpu_load/vcpu_put pair. However, for most architectures
3069 * kvm_arch_vcpu_runnable does not require vcpu_load.
3071 bool __weak kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
3073 return kvm_arch_vcpu_runnable(vcpu);
3076 static bool vcpu_dy_runnable(struct kvm_vcpu *vcpu)
3078 if (kvm_arch_dy_runnable(vcpu))
3081 #ifdef CONFIG_KVM_ASYNC_PF
3082 if (!list_empty_careful(&vcpu->async_pf.done))
3089 void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode)
3091 struct kvm *kvm = me->kvm;
3092 struct kvm_vcpu *vcpu;
3093 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
3099 kvm_vcpu_set_in_spin_loop(me, true);
3101 * We boost the priority of a VCPU that is runnable but not
3102 * currently running, because it got preempted by something
3103 * else and called schedule in __vcpu_run. Hopefully that
3104 * VCPU is holding the lock that we need and will release it.
3105 * We approximate round-robin by starting at the last boosted VCPU.
3107 for (pass = 0; pass < 2 && !yielded && try; pass++) {
3108 kvm_for_each_vcpu(i, vcpu, kvm) {
3109 if (!pass && i <= last_boosted_vcpu) {
3110 i = last_boosted_vcpu;
3112 } else if (pass && i > last_boosted_vcpu)
3114 if (!READ_ONCE(vcpu->ready))
3118 if (rcuwait_active(&vcpu->wait) &&
3119 !vcpu_dy_runnable(vcpu))
3121 if (READ_ONCE(vcpu->preempted) && yield_to_kernel_mode &&
3122 !kvm_arch_vcpu_in_kernel(vcpu))
3124 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
3127 yielded = kvm_vcpu_yield_to(vcpu);
3129 kvm->last_boosted_vcpu = i;
3131 } else if (yielded < 0) {
3138 kvm_vcpu_set_in_spin_loop(me, false);
3140 /* Ensure vcpu is not eligible during next spinloop */
3141 kvm_vcpu_set_dy_eligible(me, false);
3143 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
3145 static vm_fault_t kvm_vcpu_fault(struct vm_fault *vmf)
3147 struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data;
3150 if (vmf->pgoff == 0)
3151 page = virt_to_page(vcpu->run);
3153 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
3154 page = virt_to_page(vcpu->arch.pio_data);
3156 #ifdef CONFIG_KVM_MMIO
3157 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
3158 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
3161 return kvm_arch_vcpu_fault(vcpu, vmf);
3167 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
3168 .fault = kvm_vcpu_fault,
3171 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
3173 vma->vm_ops = &kvm_vcpu_vm_ops;
3177 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
3179 struct kvm_vcpu *vcpu = filp->private_data;
3181 kvm_put_kvm(vcpu->kvm);
3185 static struct file_operations kvm_vcpu_fops = {
3186 .release = kvm_vcpu_release,
3187 .unlocked_ioctl = kvm_vcpu_ioctl,
3188 .mmap = kvm_vcpu_mmap,
3189 .llseek = noop_llseek,
3190 KVM_COMPAT(kvm_vcpu_compat_ioctl),
3194 * Allocates an inode for the vcpu.
3196 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
3198 char name[8 + 1 + ITOA_MAX_LEN + 1];
3200 snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id);
3201 return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
3204 static void kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
3206 #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS
3207 struct dentry *debugfs_dentry;
3208 char dir_name[ITOA_MAX_LEN * 2];
3210 if (!debugfs_initialized())
3213 snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
3214 debugfs_dentry = debugfs_create_dir(dir_name,
3215 vcpu->kvm->debugfs_dentry);
3217 kvm_arch_create_vcpu_debugfs(vcpu, debugfs_dentry);
3222 * Creates some virtual cpus. Good luck creating more than one.
3224 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
3227 struct kvm_vcpu *vcpu;
3230 if (id >= KVM_MAX_VCPU_ID)
3233 mutex_lock(&kvm->lock);
3234 if (kvm->created_vcpus == KVM_MAX_VCPUS) {
3235 mutex_unlock(&kvm->lock);
3239 kvm->created_vcpus++;
3240 mutex_unlock(&kvm->lock);
3242 r = kvm_arch_vcpu_precreate(kvm, id);
3244 goto vcpu_decrement;
3246 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
3249 goto vcpu_decrement;
3252 BUILD_BUG_ON(sizeof(struct kvm_run) > PAGE_SIZE);
3253 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
3258 vcpu->run = page_address(page);
3260 kvm_vcpu_init(vcpu, kvm, id);
3262 r = kvm_arch_vcpu_create(vcpu);
3264 goto vcpu_free_run_page;
3266 mutex_lock(&kvm->lock);
3267 if (kvm_get_vcpu_by_id(kvm, id)) {
3269 goto unlock_vcpu_destroy;
3272 vcpu->vcpu_idx = atomic_read(&kvm->online_vcpus);
3273 BUG_ON(kvm->vcpus[vcpu->vcpu_idx]);
3275 /* Now it's all set up, let userspace reach it */
3277 r = create_vcpu_fd(vcpu);
3279 kvm_put_kvm_no_destroy(kvm);
3280 goto unlock_vcpu_destroy;
3283 kvm->vcpus[vcpu->vcpu_idx] = vcpu;
3286 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
3287 * before kvm->online_vcpu's incremented value.
3290 atomic_inc(&kvm->online_vcpus);
3292 mutex_unlock(&kvm->lock);
3293 kvm_arch_vcpu_postcreate(vcpu);
3294 kvm_create_vcpu_debugfs(vcpu);
3297 unlock_vcpu_destroy:
3298 mutex_unlock(&kvm->lock);
3299 kvm_arch_vcpu_destroy(vcpu);
3301 free_page((unsigned long)vcpu->run);
3303 kmem_cache_free(kvm_vcpu_cache, vcpu);
3305 mutex_lock(&kvm->lock);
3306 kvm->created_vcpus--;
3307 mutex_unlock(&kvm->lock);
3311 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
3314 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
3315 vcpu->sigset_active = 1;
3316 vcpu->sigset = *sigset;
3318 vcpu->sigset_active = 0;
3322 static long kvm_vcpu_ioctl(struct file *filp,
3323 unsigned int ioctl, unsigned long arg)
3325 struct kvm_vcpu *vcpu = filp->private_data;
3326 void __user *argp = (void __user *)arg;
3328 struct kvm_fpu *fpu = NULL;
3329 struct kvm_sregs *kvm_sregs = NULL;
3331 if (vcpu->kvm->mm != current->mm || vcpu->kvm->vm_bugged)
3334 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
3338 * Some architectures have vcpu ioctls that are asynchronous to vcpu
3339 * execution; mutex_lock() would break them.
3341 r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg);
3342 if (r != -ENOIOCTLCMD)
3345 if (mutex_lock_killable(&vcpu->mutex))
3353 oldpid = rcu_access_pointer(vcpu->pid);
3354 if (unlikely(oldpid != task_pid(current))) {
3355 /* The thread running this VCPU changed. */
3358 r = kvm_arch_vcpu_run_pid_change(vcpu);
3362 newpid = get_task_pid(current, PIDTYPE_PID);
3363 rcu_assign_pointer(vcpu->pid, newpid);
3368 r = kvm_arch_vcpu_ioctl_run(vcpu);
3369 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
3372 case KVM_GET_REGS: {
3373 struct kvm_regs *kvm_regs;
3376 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL_ACCOUNT);
3379 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
3383 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
3390 case KVM_SET_REGS: {
3391 struct kvm_regs *kvm_regs;
3393 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
3394 if (IS_ERR(kvm_regs)) {
3395 r = PTR_ERR(kvm_regs);
3398 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
3402 case KVM_GET_SREGS: {
3403 kvm_sregs = kzalloc(sizeof(struct kvm_sregs),
3404 GFP_KERNEL_ACCOUNT);
3408 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
3412 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
3417 case KVM_SET_SREGS: {
3418 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
3419 if (IS_ERR(kvm_sregs)) {
3420 r = PTR_ERR(kvm_sregs);
3424 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
3427 case KVM_GET_MP_STATE: {
3428 struct kvm_mp_state mp_state;
3430 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
3434 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
3439 case KVM_SET_MP_STATE: {
3440 struct kvm_mp_state mp_state;
3443 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
3445 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
3448 case KVM_TRANSLATE: {
3449 struct kvm_translation tr;
3452 if (copy_from_user(&tr, argp, sizeof(tr)))
3454 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
3458 if (copy_to_user(argp, &tr, sizeof(tr)))
3463 case KVM_SET_GUEST_DEBUG: {
3464 struct kvm_guest_debug dbg;
3467 if (copy_from_user(&dbg, argp, sizeof(dbg)))
3469 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
3472 case KVM_SET_SIGNAL_MASK: {
3473 struct kvm_signal_mask __user *sigmask_arg = argp;
3474 struct kvm_signal_mask kvm_sigmask;
3475 sigset_t sigset, *p;
3480 if (copy_from_user(&kvm_sigmask, argp,
3481 sizeof(kvm_sigmask)))
3484 if (kvm_sigmask.len != sizeof(sigset))
3487 if (copy_from_user(&sigset, sigmask_arg->sigset,
3492 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
3496 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL_ACCOUNT);
3500 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
3504 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
3510 fpu = memdup_user(argp, sizeof(*fpu));
3516 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
3520 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
3523 mutex_unlock(&vcpu->mutex);
3529 #ifdef CONFIG_KVM_COMPAT
3530 static long kvm_vcpu_compat_ioctl(struct file *filp,
3531 unsigned int ioctl, unsigned long arg)
3533 struct kvm_vcpu *vcpu = filp->private_data;
3534 void __user *argp = compat_ptr(arg);
3537 if (vcpu->kvm->mm != current->mm || vcpu->kvm->vm_bugged)
3541 case KVM_SET_SIGNAL_MASK: {
3542 struct kvm_signal_mask __user *sigmask_arg = argp;
3543 struct kvm_signal_mask kvm_sigmask;
3548 if (copy_from_user(&kvm_sigmask, argp,
3549 sizeof(kvm_sigmask)))
3552 if (kvm_sigmask.len != sizeof(compat_sigset_t))
3555 if (get_compat_sigset(&sigset,
3556 (compat_sigset_t __user *)sigmask_arg->sigset))
3558 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
3560 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
3564 r = kvm_vcpu_ioctl(filp, ioctl, arg);
3572 static int kvm_device_mmap(struct file *filp, struct vm_area_struct *vma)
3574 struct kvm_device *dev = filp->private_data;
3577 return dev->ops->mmap(dev, vma);
3582 static int kvm_device_ioctl_attr(struct kvm_device *dev,
3583 int (*accessor)(struct kvm_device *dev,
3584 struct kvm_device_attr *attr),
3587 struct kvm_device_attr attr;
3592 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
3595 return accessor(dev, &attr);
3598 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
3601 struct kvm_device *dev = filp->private_data;
3603 if (dev->kvm->mm != current->mm || dev->kvm->vm_bugged)
3607 case KVM_SET_DEVICE_ATTR:
3608 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
3609 case KVM_GET_DEVICE_ATTR:
3610 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
3611 case KVM_HAS_DEVICE_ATTR:
3612 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
3614 if (dev->ops->ioctl)
3615 return dev->ops->ioctl(dev, ioctl, arg);
3621 static int kvm_device_release(struct inode *inode, struct file *filp)
3623 struct kvm_device *dev = filp->private_data;
3624 struct kvm *kvm = dev->kvm;
3626 if (dev->ops->release) {
3627 mutex_lock(&kvm->lock);
3628 list_del(&dev->vm_node);
3629 dev->ops->release(dev);
3630 mutex_unlock(&kvm->lock);
3637 static const struct file_operations kvm_device_fops = {
3638 .unlocked_ioctl = kvm_device_ioctl,
3639 .release = kvm_device_release,
3640 KVM_COMPAT(kvm_device_ioctl),
3641 .mmap = kvm_device_mmap,
3644 struct kvm_device *kvm_device_from_filp(struct file *filp)
3646 if (filp->f_op != &kvm_device_fops)
3649 return filp->private_data;
3652 static const struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
3653 #ifdef CONFIG_KVM_MPIC
3654 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
3655 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
3659 int kvm_register_device_ops(const struct kvm_device_ops *ops, u32 type)
3661 if (type >= ARRAY_SIZE(kvm_device_ops_table))
3664 if (kvm_device_ops_table[type] != NULL)
3667 kvm_device_ops_table[type] = ops;
3671 void kvm_unregister_device_ops(u32 type)
3673 if (kvm_device_ops_table[type] != NULL)
3674 kvm_device_ops_table[type] = NULL;
3677 static int kvm_ioctl_create_device(struct kvm *kvm,
3678 struct kvm_create_device *cd)
3680 const struct kvm_device_ops *ops = NULL;
3681 struct kvm_device *dev;
3682 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
3686 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
3689 type = array_index_nospec(cd->type, ARRAY_SIZE(kvm_device_ops_table));
3690 ops = kvm_device_ops_table[type];
3697 dev = kzalloc(sizeof(*dev), GFP_KERNEL_ACCOUNT);
3704 mutex_lock(&kvm->lock);
3705 ret = ops->create(dev, type);
3707 mutex_unlock(&kvm->lock);
3711 list_add(&dev->vm_node, &kvm->devices);
3712 mutex_unlock(&kvm->lock);
3718 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
3720 kvm_put_kvm_no_destroy(kvm);
3721 mutex_lock(&kvm->lock);
3722 list_del(&dev->vm_node);
3725 mutex_unlock(&kvm->lock);
3735 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
3738 case KVM_CAP_USER_MEMORY:
3739 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
3740 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
3741 case KVM_CAP_INTERNAL_ERROR_DATA:
3742 #ifdef CONFIG_HAVE_KVM_MSI
3743 case KVM_CAP_SIGNAL_MSI:
3745 #ifdef CONFIG_HAVE_KVM_IRQFD
3747 case KVM_CAP_IRQFD_RESAMPLE:
3749 case KVM_CAP_IOEVENTFD_ANY_LENGTH:
3750 case KVM_CAP_CHECK_EXTENSION_VM:
3751 case KVM_CAP_ENABLE_CAP_VM:
3752 case KVM_CAP_HALT_POLL:
3754 #ifdef CONFIG_KVM_MMIO
3755 case KVM_CAP_COALESCED_MMIO:
3756 return KVM_COALESCED_MMIO_PAGE_OFFSET;
3757 case KVM_CAP_COALESCED_PIO:
3760 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3761 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2:
3762 return KVM_DIRTY_LOG_MANUAL_CAPS;
3764 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3765 case KVM_CAP_IRQ_ROUTING:
3766 return KVM_MAX_IRQ_ROUTES;
3768 #if KVM_ADDRESS_SPACE_NUM > 1
3769 case KVM_CAP_MULTI_ADDRESS_SPACE:
3770 return KVM_ADDRESS_SPACE_NUM;
3772 case KVM_CAP_NR_MEMSLOTS:
3773 return KVM_USER_MEM_SLOTS;
3777 return kvm_vm_ioctl_check_extension(kvm, arg);
3780 int __attribute__((weak)) kvm_vm_ioctl_enable_cap(struct kvm *kvm,
3781 struct kvm_enable_cap *cap)
3786 static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm,
3787 struct kvm_enable_cap *cap)
3790 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3791 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2: {
3792 u64 allowed_options = KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE;
3794 if (cap->args[0] & KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE)
3795 allowed_options = KVM_DIRTY_LOG_MANUAL_CAPS;
3797 if (cap->flags || (cap->args[0] & ~allowed_options))
3799 kvm->manual_dirty_log_protect = cap->args[0];
3803 case KVM_CAP_HALT_POLL: {
3804 if (cap->flags || cap->args[0] != (unsigned int)cap->args[0])
3807 kvm->max_halt_poll_ns = cap->args[0];
3811 return kvm_vm_ioctl_enable_cap(kvm, cap);
3815 static long kvm_vm_ioctl(struct file *filp,
3816 unsigned int ioctl, unsigned long arg)
3818 struct kvm *kvm = filp->private_data;
3819 void __user *argp = (void __user *)arg;
3822 if (kvm->mm != current->mm || kvm->vm_bugged)
3825 case KVM_CREATE_VCPU:
3826 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
3828 case KVM_ENABLE_CAP: {
3829 struct kvm_enable_cap cap;
3832 if (copy_from_user(&cap, argp, sizeof(cap)))
3834 r = kvm_vm_ioctl_enable_cap_generic(kvm, &cap);
3837 case KVM_SET_USER_MEMORY_REGION: {
3838 struct kvm_userspace_memory_region kvm_userspace_mem;
3841 if (copy_from_user(&kvm_userspace_mem, argp,
3842 sizeof(kvm_userspace_mem)))
3845 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
3848 case KVM_GET_DIRTY_LOG: {
3849 struct kvm_dirty_log log;
3852 if (copy_from_user(&log, argp, sizeof(log)))
3854 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3857 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3858 case KVM_CLEAR_DIRTY_LOG: {
3859 struct kvm_clear_dirty_log log;
3862 if (copy_from_user(&log, argp, sizeof(log)))
3864 r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
3868 #ifdef CONFIG_KVM_MMIO
3869 case KVM_REGISTER_COALESCED_MMIO: {
3870 struct kvm_coalesced_mmio_zone zone;
3873 if (copy_from_user(&zone, argp, sizeof(zone)))
3875 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
3878 case KVM_UNREGISTER_COALESCED_MMIO: {
3879 struct kvm_coalesced_mmio_zone zone;
3882 if (copy_from_user(&zone, argp, sizeof(zone)))
3884 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
3889 struct kvm_irqfd data;
3892 if (copy_from_user(&data, argp, sizeof(data)))
3894 r = kvm_irqfd(kvm, &data);
3897 case KVM_IOEVENTFD: {
3898 struct kvm_ioeventfd data;
3901 if (copy_from_user(&data, argp, sizeof(data)))
3903 r = kvm_ioeventfd(kvm, &data);
3906 #ifdef CONFIG_HAVE_KVM_MSI
3907 case KVM_SIGNAL_MSI: {
3911 if (copy_from_user(&msi, argp, sizeof(msi)))
3913 r = kvm_send_userspace_msi(kvm, &msi);
3917 #ifdef __KVM_HAVE_IRQ_LINE
3918 case KVM_IRQ_LINE_STATUS:
3919 case KVM_IRQ_LINE: {
3920 struct kvm_irq_level irq_event;
3923 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
3926 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
3927 ioctl == KVM_IRQ_LINE_STATUS);
3932 if (ioctl == KVM_IRQ_LINE_STATUS) {
3933 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
3941 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3942 case KVM_SET_GSI_ROUTING: {
3943 struct kvm_irq_routing routing;
3944 struct kvm_irq_routing __user *urouting;
3945 struct kvm_irq_routing_entry *entries = NULL;
3948 if (copy_from_user(&routing, argp, sizeof(routing)))
3951 if (!kvm_arch_can_set_irq_routing(kvm))
3953 if (routing.nr > KVM_MAX_IRQ_ROUTES)
3959 entries = vmemdup_user(urouting->entries,
3960 array_size(sizeof(*entries),
3962 if (IS_ERR(entries)) {
3963 r = PTR_ERR(entries);
3967 r = kvm_set_irq_routing(kvm, entries, routing.nr,
3972 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
3973 case KVM_CREATE_DEVICE: {
3974 struct kvm_create_device cd;
3977 if (copy_from_user(&cd, argp, sizeof(cd)))
3980 r = kvm_ioctl_create_device(kvm, &cd);
3985 if (copy_to_user(argp, &cd, sizeof(cd)))
3991 case KVM_CHECK_EXTENSION:
3992 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
3995 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
4001 #ifdef CONFIG_KVM_COMPAT
4002 struct compat_kvm_dirty_log {
4006 compat_uptr_t dirty_bitmap; /* one bit per page */
4011 struct compat_kvm_clear_dirty_log {
4016 compat_uptr_t dirty_bitmap; /* one bit per page */
4021 long __weak kvm_arch_vm_compat_ioctl(struct file *filp, unsigned int ioctl,
4027 static long kvm_vm_compat_ioctl(struct file *filp,
4028 unsigned int ioctl, unsigned long arg)
4030 struct kvm *kvm = filp->private_data;
4033 if (kvm->mm != current->mm || kvm->vm_bugged)
4036 r = kvm_arch_vm_compat_ioctl(filp, ioctl, arg);
4041 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4042 case KVM_CLEAR_DIRTY_LOG: {
4043 struct compat_kvm_clear_dirty_log compat_log;
4044 struct kvm_clear_dirty_log log;
4046 if (copy_from_user(&compat_log, (void __user *)arg,
4047 sizeof(compat_log)))
4049 log.slot = compat_log.slot;
4050 log.num_pages = compat_log.num_pages;
4051 log.first_page = compat_log.first_page;
4052 log.padding2 = compat_log.padding2;
4053 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
4055 r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
4059 case KVM_GET_DIRTY_LOG: {
4060 struct compat_kvm_dirty_log compat_log;
4061 struct kvm_dirty_log log;
4063 if (copy_from_user(&compat_log, (void __user *)arg,
4064 sizeof(compat_log)))
4066 log.slot = compat_log.slot;
4067 log.padding1 = compat_log.padding1;
4068 log.padding2 = compat_log.padding2;
4069 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
4071 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
4075 r = kvm_vm_ioctl(filp, ioctl, arg);
4081 static struct file_operations kvm_vm_fops = {
4082 .release = kvm_vm_release,
4083 .unlocked_ioctl = kvm_vm_ioctl,
4084 .llseek = noop_llseek,
4085 KVM_COMPAT(kvm_vm_compat_ioctl),
4088 static int kvm_dev_ioctl_create_vm(unsigned long type)
4094 kvm = kvm_create_vm(type);
4096 return PTR_ERR(kvm);
4097 #ifdef CONFIG_KVM_MMIO
4098 r = kvm_coalesced_mmio_init(kvm);
4102 r = get_unused_fd_flags(O_CLOEXEC);
4106 file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
4114 * Don't call kvm_put_kvm anymore at this point; file->f_op is
4115 * already set, with ->release() being kvm_vm_release(). In error
4116 * cases it will be called by the final fput(file) and will take
4117 * care of doing kvm_put_kvm(kvm).
4119 if (kvm_create_vm_debugfs(kvm, r) < 0) {
4124 kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm);
4126 fd_install(r, file);
4134 static long kvm_dev_ioctl(struct file *filp,
4135 unsigned int ioctl, unsigned long arg)
4140 case KVM_GET_API_VERSION:
4143 r = KVM_API_VERSION;
4146 r = kvm_dev_ioctl_create_vm(arg);
4148 case KVM_CHECK_EXTENSION:
4149 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
4151 case KVM_GET_VCPU_MMAP_SIZE:
4154 r = PAGE_SIZE; /* struct kvm_run */
4156 r += PAGE_SIZE; /* pio data page */
4158 #ifdef CONFIG_KVM_MMIO
4159 r += PAGE_SIZE; /* coalesced mmio ring page */
4162 case KVM_TRACE_ENABLE:
4163 case KVM_TRACE_PAUSE:
4164 case KVM_TRACE_DISABLE:
4168 return kvm_arch_dev_ioctl(filp, ioctl, arg);
4174 static struct file_operations kvm_chardev_ops = {
4175 .unlocked_ioctl = kvm_dev_ioctl,
4176 .llseek = noop_llseek,
4177 KVM_COMPAT(kvm_dev_ioctl),
4180 static struct miscdevice kvm_dev = {
4186 static void hardware_enable_nolock(void *junk)
4188 int cpu = raw_smp_processor_id();
4191 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
4194 cpumask_set_cpu(cpu, cpus_hardware_enabled);
4196 r = kvm_arch_hardware_enable();
4199 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
4200 atomic_inc(&hardware_enable_failed);
4201 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
4205 static int kvm_starting_cpu(unsigned int cpu)
4207 raw_spin_lock(&kvm_count_lock);
4208 if (kvm_usage_count)
4209 hardware_enable_nolock(NULL);
4210 raw_spin_unlock(&kvm_count_lock);
4214 static void hardware_disable_nolock(void *junk)
4216 int cpu = raw_smp_processor_id();
4218 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
4220 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
4221 kvm_arch_hardware_disable();
4224 static int kvm_dying_cpu(unsigned int cpu)
4226 raw_spin_lock(&kvm_count_lock);
4227 if (kvm_usage_count)
4228 hardware_disable_nolock(NULL);
4229 raw_spin_unlock(&kvm_count_lock);
4233 static void hardware_disable_all_nolock(void)
4235 BUG_ON(!kvm_usage_count);
4238 if (!kvm_usage_count)
4239 on_each_cpu(hardware_disable_nolock, NULL, 1);
4242 static void hardware_disable_all(void)
4244 raw_spin_lock(&kvm_count_lock);
4245 hardware_disable_all_nolock();
4246 raw_spin_unlock(&kvm_count_lock);
4249 static int hardware_enable_all(void)
4253 raw_spin_lock(&kvm_count_lock);
4256 if (kvm_usage_count == 1) {
4257 atomic_set(&hardware_enable_failed, 0);
4258 on_each_cpu(hardware_enable_nolock, NULL, 1);
4260 if (atomic_read(&hardware_enable_failed)) {
4261 hardware_disable_all_nolock();
4266 raw_spin_unlock(&kvm_count_lock);
4271 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
4275 * Some (well, at least mine) BIOSes hang on reboot if
4278 * And Intel TXT required VMX off for all cpu when system shutdown.
4280 pr_info("kvm: exiting hardware virtualization\n");
4281 kvm_rebooting = true;
4282 on_each_cpu(hardware_disable_nolock, NULL, 1);
4286 static struct notifier_block kvm_reboot_notifier = {
4287 .notifier_call = kvm_reboot,
4291 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
4295 for (i = 0; i < bus->dev_count; i++) {
4296 struct kvm_io_device *pos = bus->range[i].dev;
4298 kvm_iodevice_destructor(pos);
4303 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
4304 const struct kvm_io_range *r2)
4306 gpa_t addr1 = r1->addr;
4307 gpa_t addr2 = r2->addr;
4312 /* If r2->len == 0, match the exact address. If r2->len != 0,
4313 * accept any overlapping write. Any order is acceptable for
4314 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
4315 * we process all of them.
4328 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
4330 return kvm_io_bus_cmp(p1, p2);
4333 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
4334 gpa_t addr, int len)
4336 struct kvm_io_range *range, key;
4339 key = (struct kvm_io_range) {
4344 range = bsearch(&key, bus->range, bus->dev_count,
4345 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
4349 off = range - bus->range;
4351 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
4357 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
4358 struct kvm_io_range *range, const void *val)
4362 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
4366 while (idx < bus->dev_count &&
4367 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
4368 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
4377 /* kvm_io_bus_write - called under kvm->slots_lock */
4378 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
4379 int len, const void *val)
4381 struct kvm_io_bus *bus;
4382 struct kvm_io_range range;
4385 range = (struct kvm_io_range) {
4390 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
4393 r = __kvm_io_bus_write(vcpu, bus, &range, val);
4394 return r < 0 ? r : 0;
4396 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
4398 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
4399 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
4400 gpa_t addr, int len, const void *val, long cookie)
4402 struct kvm_io_bus *bus;
4403 struct kvm_io_range range;
4405 range = (struct kvm_io_range) {
4410 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
4414 /* First try the device referenced by cookie. */
4415 if ((cookie >= 0) && (cookie < bus->dev_count) &&
4416 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
4417 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
4422 * cookie contained garbage; fall back to search and return the
4423 * correct cookie value.
4425 return __kvm_io_bus_write(vcpu, bus, &range, val);
4428 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
4429 struct kvm_io_range *range, void *val)
4433 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
4437 while (idx < bus->dev_count &&
4438 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
4439 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
4448 /* kvm_io_bus_read - called under kvm->slots_lock */
4449 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
4452 struct kvm_io_bus *bus;
4453 struct kvm_io_range range;
4456 range = (struct kvm_io_range) {
4461 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
4464 r = __kvm_io_bus_read(vcpu, bus, &range, val);
4465 return r < 0 ? r : 0;
4468 /* Caller must hold slots_lock. */
4469 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
4470 int len, struct kvm_io_device *dev)
4473 struct kvm_io_bus *new_bus, *bus;
4474 struct kvm_io_range range;
4476 bus = kvm_get_bus(kvm, bus_idx);
4480 /* exclude ioeventfd which is limited by maximum fd */
4481 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
4484 new_bus = kmalloc(struct_size(bus, range, bus->dev_count + 1),
4485 GFP_KERNEL_ACCOUNT);
4489 range = (struct kvm_io_range) {
4495 for (i = 0; i < bus->dev_count; i++)
4496 if (kvm_io_bus_cmp(&bus->range[i], &range) > 0)
4499 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
4500 new_bus->dev_count++;
4501 new_bus->range[i] = range;
4502 memcpy(new_bus->range + i + 1, bus->range + i,
4503 (bus->dev_count - i) * sizeof(struct kvm_io_range));
4504 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
4505 synchronize_srcu_expedited(&kvm->srcu);
4511 /* Caller must hold slots_lock. */
4512 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
4513 struct kvm_io_device *dev)
4516 struct kvm_io_bus *new_bus, *bus;
4518 bus = kvm_get_bus(kvm, bus_idx);
4522 for (i = 0; i < bus->dev_count; i++)
4523 if (bus->range[i].dev == dev) {
4527 if (i == bus->dev_count)
4530 new_bus = kmalloc(struct_size(bus, range, bus->dev_count - 1),
4531 GFP_KERNEL_ACCOUNT);
4533 memcpy(new_bus, bus, struct_size(bus, range, i));
4534 new_bus->dev_count--;
4535 memcpy(new_bus->range + i, bus->range + i + 1,
4536 flex_array_size(new_bus, range, new_bus->dev_count - i));
4539 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
4540 synchronize_srcu_expedited(&kvm->srcu);
4542 /* Destroy the old bus _after_ installing the (null) bus. */
4544 pr_err("kvm: failed to shrink bus, removing it completely\n");
4545 for (j = 0; j < bus->dev_count; j++) {
4548 kvm_iodevice_destructor(bus->range[j].dev);
4553 return new_bus ? 0 : -ENOMEM;
4556 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
4559 struct kvm_io_bus *bus;
4560 int dev_idx, srcu_idx;
4561 struct kvm_io_device *iodev = NULL;
4563 srcu_idx = srcu_read_lock(&kvm->srcu);
4565 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
4569 dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
4573 iodev = bus->range[dev_idx].dev;
4576 srcu_read_unlock(&kvm->srcu, srcu_idx);
4580 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
4582 static int kvm_debugfs_open(struct inode *inode, struct file *file,
4583 int (*get)(void *, u64 *), int (*set)(void *, u64),
4586 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
4589 /* The debugfs files are a reference to the kvm struct which
4590 * is still valid when kvm_destroy_vm is called.
4591 * To avoid the race between open and the removal of the debugfs
4592 * directory we test against the users count.
4594 if (!refcount_inc_not_zero(&stat_data->kvm->users_count))
4597 if (simple_attr_open(inode, file, get,
4598 KVM_DBGFS_GET_MODE(stat_data->dbgfs_item) & 0222
4601 kvm_put_kvm(stat_data->kvm);
4608 static int kvm_debugfs_release(struct inode *inode, struct file *file)
4610 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
4613 simple_attr_release(inode, file);
4614 kvm_put_kvm(stat_data->kvm);
4619 static int kvm_get_stat_per_vm(struct kvm *kvm, size_t offset, u64 *val)
4621 *val = *(ulong *)((void *)kvm + offset);
4626 static int kvm_clear_stat_per_vm(struct kvm *kvm, size_t offset)
4628 *(ulong *)((void *)kvm + offset) = 0;
4633 static int kvm_get_stat_per_vcpu(struct kvm *kvm, size_t offset, u64 *val)
4636 struct kvm_vcpu *vcpu;
4640 kvm_for_each_vcpu(i, vcpu, kvm)
4641 *val += *(u64 *)((void *)vcpu + offset);
4646 static int kvm_clear_stat_per_vcpu(struct kvm *kvm, size_t offset)
4649 struct kvm_vcpu *vcpu;
4651 kvm_for_each_vcpu(i, vcpu, kvm)
4652 *(u64 *)((void *)vcpu + offset) = 0;
4657 static int kvm_stat_data_get(void *data, u64 *val)
4660 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
4662 switch (stat_data->dbgfs_item->kind) {
4664 r = kvm_get_stat_per_vm(stat_data->kvm,
4665 stat_data->dbgfs_item->offset, val);
4668 r = kvm_get_stat_per_vcpu(stat_data->kvm,
4669 stat_data->dbgfs_item->offset, val);
4676 static int kvm_stat_data_clear(void *data, u64 val)
4679 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
4684 switch (stat_data->dbgfs_item->kind) {
4686 r = kvm_clear_stat_per_vm(stat_data->kvm,
4687 stat_data->dbgfs_item->offset);
4690 r = kvm_clear_stat_per_vcpu(stat_data->kvm,
4691 stat_data->dbgfs_item->offset);
4698 static int kvm_stat_data_open(struct inode *inode, struct file *file)
4700 __simple_attr_check_format("%llu\n", 0ull);
4701 return kvm_debugfs_open(inode, file, kvm_stat_data_get,
4702 kvm_stat_data_clear, "%llu\n");
4705 static const struct file_operations stat_fops_per_vm = {
4706 .owner = THIS_MODULE,
4707 .open = kvm_stat_data_open,
4708 .release = kvm_debugfs_release,
4709 .read = simple_attr_read,
4710 .write = simple_attr_write,
4711 .llseek = no_llseek,
4714 static int vm_stat_get(void *_offset, u64 *val)
4716 unsigned offset = (long)_offset;
4721 mutex_lock(&kvm_lock);
4722 list_for_each_entry(kvm, &vm_list, vm_list) {
4723 kvm_get_stat_per_vm(kvm, offset, &tmp_val);
4726 mutex_unlock(&kvm_lock);
4730 static int vm_stat_clear(void *_offset, u64 val)
4732 unsigned offset = (long)_offset;
4738 mutex_lock(&kvm_lock);
4739 list_for_each_entry(kvm, &vm_list, vm_list) {
4740 kvm_clear_stat_per_vm(kvm, offset);
4742 mutex_unlock(&kvm_lock);
4747 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n");
4749 static int vcpu_stat_get(void *_offset, u64 *val)
4751 unsigned offset = (long)_offset;
4756 mutex_lock(&kvm_lock);
4757 list_for_each_entry(kvm, &vm_list, vm_list) {
4758 kvm_get_stat_per_vcpu(kvm, offset, &tmp_val);
4761 mutex_unlock(&kvm_lock);
4765 static int vcpu_stat_clear(void *_offset, u64 val)
4767 unsigned offset = (long)_offset;
4773 mutex_lock(&kvm_lock);
4774 list_for_each_entry(kvm, &vm_list, vm_list) {
4775 kvm_clear_stat_per_vcpu(kvm, offset);
4777 mutex_unlock(&kvm_lock);
4782 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear,
4785 static const struct file_operations *stat_fops[] = {
4786 [KVM_STAT_VCPU] = &vcpu_stat_fops,
4787 [KVM_STAT_VM] = &vm_stat_fops,
4790 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm)
4792 struct kobj_uevent_env *env;
4793 unsigned long long created, active;
4795 if (!kvm_dev.this_device || !kvm)
4798 mutex_lock(&kvm_lock);
4799 if (type == KVM_EVENT_CREATE_VM) {
4800 kvm_createvm_count++;
4802 } else if (type == KVM_EVENT_DESTROY_VM) {
4805 created = kvm_createvm_count;
4806 active = kvm_active_vms;
4807 mutex_unlock(&kvm_lock);
4809 env = kzalloc(sizeof(*env), GFP_KERNEL_ACCOUNT);
4813 add_uevent_var(env, "CREATED=%llu", created);
4814 add_uevent_var(env, "COUNT=%llu", active);
4816 if (type == KVM_EVENT_CREATE_VM) {
4817 add_uevent_var(env, "EVENT=create");
4818 kvm->userspace_pid = task_pid_nr(current);
4819 } else if (type == KVM_EVENT_DESTROY_VM) {
4820 add_uevent_var(env, "EVENT=destroy");
4822 add_uevent_var(env, "PID=%d", kvm->userspace_pid);
4824 if (kvm->debugfs_dentry) {
4825 char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL_ACCOUNT);
4828 tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX);
4830 add_uevent_var(env, "STATS_PATH=%s", tmp);
4834 /* no need for checks, since we are adding at most only 5 keys */
4835 env->envp[env->envp_idx++] = NULL;
4836 kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp);
4840 static void kvm_init_debug(void)
4842 struct kvm_stats_debugfs_item *p;
4844 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
4846 kvm_debugfs_num_entries = 0;
4847 for (p = debugfs_entries; p->name; ++p, kvm_debugfs_num_entries++) {
4848 debugfs_create_file(p->name, KVM_DBGFS_GET_MODE(p),
4849 kvm_debugfs_dir, (void *)(long)p->offset,
4850 stat_fops[p->kind]);
4854 static int kvm_suspend(void)
4856 if (kvm_usage_count)
4857 hardware_disable_nolock(NULL);
4861 static void kvm_resume(void)
4863 if (kvm_usage_count) {
4864 #ifdef CONFIG_LOCKDEP
4865 WARN_ON(lockdep_is_held(&kvm_count_lock));
4867 hardware_enable_nolock(NULL);
4871 static struct syscore_ops kvm_syscore_ops = {
4872 .suspend = kvm_suspend,
4873 .resume = kvm_resume,
4877 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
4879 return container_of(pn, struct kvm_vcpu, preempt_notifier);
4882 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
4884 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
4886 WRITE_ONCE(vcpu->preempted, false);
4887 WRITE_ONCE(vcpu->ready, false);
4889 __this_cpu_write(kvm_running_vcpu, vcpu);
4890 kvm_arch_sched_in(vcpu, cpu);
4891 kvm_arch_vcpu_load(vcpu, cpu);
4894 static void kvm_sched_out(struct preempt_notifier *pn,
4895 struct task_struct *next)
4897 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
4899 if (current->state == TASK_RUNNING) {
4900 WRITE_ONCE(vcpu->preempted, true);
4901 WRITE_ONCE(vcpu->ready, true);
4903 kvm_arch_vcpu_put(vcpu);
4904 __this_cpu_write(kvm_running_vcpu, NULL);
4908 * kvm_get_running_vcpu - get the vcpu running on the current CPU.
4910 * We can disable preemption locally around accessing the per-CPU variable,
4911 * and use the resolved vcpu pointer after enabling preemption again,
4912 * because even if the current thread is migrated to another CPU, reading
4913 * the per-CPU value later will give us the same value as we update the
4914 * per-CPU variable in the preempt notifier handlers.
4916 struct kvm_vcpu *kvm_get_running_vcpu(void)
4918 struct kvm_vcpu *vcpu;
4921 vcpu = __this_cpu_read(kvm_running_vcpu);
4926 EXPORT_SYMBOL_GPL(kvm_get_running_vcpu);
4929 * kvm_get_running_vcpus - get the per-CPU array of currently running vcpus.
4931 struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void)
4933 return &kvm_running_vcpu;
4936 struct kvm_cpu_compat_check {
4941 static void check_processor_compat(void *data)
4943 struct kvm_cpu_compat_check *c = data;
4945 *c->ret = kvm_arch_check_processor_compat(c->opaque);
4948 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
4949 struct module *module)
4951 struct kvm_cpu_compat_check c;
4955 r = kvm_arch_init(opaque);
4960 * kvm_arch_init makes sure there's at most one caller
4961 * for architectures that support multiple implementations,
4962 * like intel and amd on x86.
4963 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
4964 * conflicts in case kvm is already setup for another implementation.
4966 r = kvm_irqfd_init();
4970 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
4975 r = kvm_arch_hardware_setup(opaque);
4981 for_each_online_cpu(cpu) {
4982 smp_call_function_single(cpu, check_processor_compat, &c, 1);
4987 r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting",
4988 kvm_starting_cpu, kvm_dying_cpu);
4991 register_reboot_notifier(&kvm_reboot_notifier);
4993 /* A kmem cache lets us meet the alignment requirements of fx_save. */
4995 vcpu_align = __alignof__(struct kvm_vcpu);
4997 kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align,
4999 offsetof(struct kvm_vcpu, arch),
5000 sizeof_field(struct kvm_vcpu, arch),
5002 if (!kvm_vcpu_cache) {
5007 for_each_possible_cpu(cpu) {
5008 if (!alloc_cpumask_var_node(&per_cpu(cpu_kick_mask, cpu),
5009 GFP_KERNEL, cpu_to_node(cpu))) {
5015 r = kvm_async_pf_init();
5019 kvm_chardev_ops.owner = module;
5020 kvm_vm_fops.owner = module;
5021 kvm_vcpu_fops.owner = module;
5023 register_syscore_ops(&kvm_syscore_ops);
5025 kvm_preempt_ops.sched_in = kvm_sched_in;
5026 kvm_preempt_ops.sched_out = kvm_sched_out;
5030 r = kvm_vfio_ops_init();
5031 if (WARN_ON_ONCE(r))
5035 * Registration _must_ be the very last thing done, as this exposes
5036 * /dev/kvm to userspace, i.e. all infrastructure must be setup!
5038 r = misc_register(&kvm_dev);
5040 pr_err("kvm: misc device register failed\n");
5047 kvm_vfio_ops_exit();
5049 kvm_async_pf_deinit();
5051 for_each_possible_cpu(cpu)
5052 free_cpumask_var(per_cpu(cpu_kick_mask, cpu));
5053 kmem_cache_destroy(kvm_vcpu_cache);
5055 unregister_reboot_notifier(&kvm_reboot_notifier);
5056 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
5058 kvm_arch_hardware_unsetup();
5060 free_cpumask_var(cpus_hardware_enabled);
5068 EXPORT_SYMBOL_GPL(kvm_init);
5075 * Note, unregistering /dev/kvm doesn't strictly need to come first,
5076 * fops_get(), a.k.a. try_module_get(), prevents acquiring references
5077 * to KVM while the module is being stopped.
5079 misc_deregister(&kvm_dev);
5081 debugfs_remove_recursive(kvm_debugfs_dir);
5082 for_each_possible_cpu(cpu)
5083 free_cpumask_var(per_cpu(cpu_kick_mask, cpu));
5084 kmem_cache_destroy(kvm_vcpu_cache);
5085 kvm_async_pf_deinit();
5086 unregister_syscore_ops(&kvm_syscore_ops);
5087 unregister_reboot_notifier(&kvm_reboot_notifier);
5088 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
5089 on_each_cpu(hardware_disable_nolock, NULL, 1);
5090 kvm_arch_hardware_unsetup();
5093 free_cpumask_var(cpus_hardware_enabled);
5094 kvm_vfio_ops_exit();
5096 EXPORT_SYMBOL_GPL(kvm_exit);
5098 struct kvm_vm_worker_thread_context {
5100 struct task_struct *parent;
5101 struct completion init_done;
5102 kvm_vm_thread_fn_t thread_fn;
5107 static int kvm_vm_worker_thread(void *context)
5110 * The init_context is allocated on the stack of the parent thread, so
5111 * we have to locally copy anything that is needed beyond initialization
5113 struct kvm_vm_worker_thread_context *init_context = context;
5114 struct kvm *kvm = init_context->kvm;
5115 kvm_vm_thread_fn_t thread_fn = init_context->thread_fn;
5116 uintptr_t data = init_context->data;
5119 err = kthread_park(current);
5120 /* kthread_park(current) is never supposed to return an error */
5125 err = cgroup_attach_task_all(init_context->parent, current);
5127 kvm_err("%s: cgroup_attach_task_all failed with err %d\n",
5132 set_user_nice(current, task_nice(init_context->parent));
5135 init_context->err = err;
5136 complete(&init_context->init_done);
5137 init_context = NULL;
5142 /* Wait to be woken up by the spawner before proceeding. */
5145 if (!kthread_should_stop())
5146 err = thread_fn(kvm, data);
5151 int kvm_vm_create_worker_thread(struct kvm *kvm, kvm_vm_thread_fn_t thread_fn,
5152 uintptr_t data, const char *name,
5153 struct task_struct **thread_ptr)
5155 struct kvm_vm_worker_thread_context init_context = {};
5156 struct task_struct *thread;
5159 init_context.kvm = kvm;
5160 init_context.parent = current;
5161 init_context.thread_fn = thread_fn;
5162 init_context.data = data;
5163 init_completion(&init_context.init_done);
5165 thread = kthread_run(kvm_vm_worker_thread, &init_context,
5166 "%s-%d", name, task_pid_nr(current));
5168 return PTR_ERR(thread);
5170 /* kthread_run is never supposed to return NULL */
5171 WARN_ON(thread == NULL);
5173 wait_for_completion(&init_context.init_done);
5175 if (!init_context.err)
5176 *thread_ptr = thread;
5178 return init_context.err;