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"
64 #define CREATE_TRACE_POINTS
65 #include <trace/events/kvm.h>
67 #include <linux/kvm_dirty_ring.h>
69 /* Worst case buffer size needed for holding an integer. */
70 #define ITOA_MAX_LEN 12
72 MODULE_AUTHOR("Qumranet");
73 MODULE_LICENSE("GPL");
75 /* Architectures should define their poll value according to the halt latency */
76 unsigned int halt_poll_ns = KVM_HALT_POLL_NS_DEFAULT;
77 module_param(halt_poll_ns, uint, 0644);
78 EXPORT_SYMBOL_GPL(halt_poll_ns);
80 /* Default doubles per-vcpu halt_poll_ns. */
81 unsigned int halt_poll_ns_grow = 2;
82 module_param(halt_poll_ns_grow, uint, 0644);
83 EXPORT_SYMBOL_GPL(halt_poll_ns_grow);
85 /* The start value to grow halt_poll_ns from */
86 unsigned int halt_poll_ns_grow_start = 10000; /* 10us */
87 module_param(halt_poll_ns_grow_start, uint, 0644);
88 EXPORT_SYMBOL_GPL(halt_poll_ns_grow_start);
90 /* Default resets per-vcpu halt_poll_ns . */
91 unsigned int halt_poll_ns_shrink;
92 module_param(halt_poll_ns_shrink, uint, 0644);
93 EXPORT_SYMBOL_GPL(halt_poll_ns_shrink);
98 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
101 DEFINE_MUTEX(kvm_lock);
102 static DEFINE_RAW_SPINLOCK(kvm_count_lock);
105 static cpumask_var_t cpus_hardware_enabled;
106 static int kvm_usage_count;
107 static atomic_t hardware_enable_failed;
109 static struct kmem_cache *kvm_vcpu_cache;
111 static __read_mostly struct preempt_ops kvm_preempt_ops;
112 static DEFINE_PER_CPU(struct kvm_vcpu *, kvm_running_vcpu);
114 struct dentry *kvm_debugfs_dir;
115 EXPORT_SYMBOL_GPL(kvm_debugfs_dir);
117 static int kvm_debugfs_num_entries;
118 static const struct file_operations stat_fops_per_vm;
120 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
122 #ifdef CONFIG_KVM_COMPAT
123 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
125 #define KVM_COMPAT(c) .compat_ioctl = (c)
128 * For architectures that don't implement a compat infrastructure,
129 * adopt a double line of defense:
130 * - Prevent a compat task from opening /dev/kvm
131 * - If the open has been done by a 64bit task, and the KVM fd
132 * passed to a compat task, let the ioctls fail.
134 static long kvm_no_compat_ioctl(struct file *file, unsigned int ioctl,
135 unsigned long arg) { return -EINVAL; }
137 static int kvm_no_compat_open(struct inode *inode, struct file *file)
139 return is_compat_task() ? -ENODEV : 0;
141 #define KVM_COMPAT(c) .compat_ioctl = kvm_no_compat_ioctl, \
142 .open = kvm_no_compat_open
144 static int hardware_enable_all(void);
145 static void hardware_disable_all(void);
147 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
149 __visible bool kvm_rebooting;
150 EXPORT_SYMBOL_GPL(kvm_rebooting);
152 #define KVM_EVENT_CREATE_VM 0
153 #define KVM_EVENT_DESTROY_VM 1
154 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm);
155 static unsigned long long kvm_createvm_count;
156 static unsigned long long kvm_active_vms;
158 __weak void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
159 unsigned long start, unsigned long end)
163 bool kvm_is_zone_device_pfn(kvm_pfn_t pfn)
166 * The metadata used by is_zone_device_page() to determine whether or
167 * not a page is ZONE_DEVICE is guaranteed to be valid if and only if
168 * the device has been pinned, e.g. by get_user_pages(). WARN if the
169 * page_count() is zero to help detect bad usage of this helper.
171 if (!pfn_valid(pfn) || WARN_ON_ONCE(!page_count(pfn_to_page(pfn))))
174 return is_zone_device_page(pfn_to_page(pfn));
177 bool kvm_is_reserved_pfn(kvm_pfn_t pfn)
180 * ZONE_DEVICE pages currently set PG_reserved, but from a refcounting
181 * perspective they are "normal" pages, albeit with slightly different
185 return PageReserved(pfn_to_page(pfn)) &&
187 !kvm_is_zone_device_pfn(pfn);
192 bool kvm_is_transparent_hugepage(kvm_pfn_t pfn)
194 struct page *page = pfn_to_page(pfn);
196 if (!PageTransCompoundMap(page))
199 return is_transparent_hugepage(compound_head(page));
203 * Switches to specified vcpu, until a matching vcpu_put()
205 void vcpu_load(struct kvm_vcpu *vcpu)
209 __this_cpu_write(kvm_running_vcpu, vcpu);
210 preempt_notifier_register(&vcpu->preempt_notifier);
211 kvm_arch_vcpu_load(vcpu, cpu);
214 EXPORT_SYMBOL_GPL(vcpu_load);
216 void vcpu_put(struct kvm_vcpu *vcpu)
219 kvm_arch_vcpu_put(vcpu);
220 preempt_notifier_unregister(&vcpu->preempt_notifier);
221 __this_cpu_write(kvm_running_vcpu, NULL);
224 EXPORT_SYMBOL_GPL(vcpu_put);
226 /* TODO: merge with kvm_arch_vcpu_should_kick */
227 static bool kvm_request_needs_ipi(struct kvm_vcpu *vcpu, unsigned req)
229 int mode = kvm_vcpu_exiting_guest_mode(vcpu);
232 * We need to wait for the VCPU to reenable interrupts and get out of
233 * READING_SHADOW_PAGE_TABLES mode.
235 if (req & KVM_REQUEST_WAIT)
236 return mode != OUTSIDE_GUEST_MODE;
239 * Need to kick a running VCPU, but otherwise there is nothing to do.
241 return mode == IN_GUEST_MODE;
244 static void ack_flush(void *_completed)
248 static inline bool kvm_kick_many_cpus(const struct cpumask *cpus, bool wait)
251 cpus = cpu_online_mask;
253 if (cpumask_empty(cpus))
256 smp_call_function_many(cpus, ack_flush, NULL, wait);
260 bool kvm_make_vcpus_request_mask(struct kvm *kvm, unsigned int req,
261 struct kvm_vcpu *except,
262 unsigned long *vcpu_bitmap, cpumask_var_t tmp)
265 struct kvm_vcpu *vcpu;
270 kvm_for_each_vcpu(i, vcpu, kvm) {
271 if ((vcpu_bitmap && !test_bit(i, vcpu_bitmap)) ||
275 kvm_make_request(req, vcpu);
278 if (!(req & KVM_REQUEST_NO_WAKEUP) && kvm_vcpu_wake_up(vcpu))
281 if (tmp != NULL && cpu != -1 && cpu != me &&
282 kvm_request_needs_ipi(vcpu, req))
283 __cpumask_set_cpu(cpu, tmp);
286 called = kvm_kick_many_cpus(tmp, !!(req & KVM_REQUEST_WAIT));
292 bool kvm_make_all_cpus_request_except(struct kvm *kvm, unsigned int req,
293 struct kvm_vcpu *except)
298 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
300 called = kvm_make_vcpus_request_mask(kvm, req, except, NULL, cpus);
302 free_cpumask_var(cpus);
306 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
308 return kvm_make_all_cpus_request_except(kvm, req, NULL);
310 EXPORT_SYMBOL_GPL(kvm_make_all_cpus_request);
312 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
313 void kvm_flush_remote_tlbs(struct kvm *kvm)
316 * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in
317 * kvm_make_all_cpus_request.
319 long dirty_count = smp_load_acquire(&kvm->tlbs_dirty);
322 * We want to publish modifications to the page tables before reading
323 * mode. Pairs with a memory barrier in arch-specific code.
324 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
325 * and smp_mb in walk_shadow_page_lockless_begin/end.
326 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
328 * There is already an smp_mb__after_atomic() before
329 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
332 if (!kvm_arch_flush_remote_tlb(kvm)
333 || kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
334 ++kvm->stat.remote_tlb_flush;
335 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
337 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
340 void kvm_reload_remote_mmus(struct kvm *kvm)
342 kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
345 #ifdef KVM_ARCH_NR_OBJS_PER_MEMORY_CACHE
346 static inline void *mmu_memory_cache_alloc_obj(struct kvm_mmu_memory_cache *mc,
349 gfp_flags |= mc->gfp_zero;
352 return kmem_cache_alloc(mc->kmem_cache, gfp_flags);
354 return (void *)__get_free_page(gfp_flags);
357 int kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache *mc, int min)
361 if (mc->nobjs >= min)
363 while (mc->nobjs < ARRAY_SIZE(mc->objects)) {
364 obj = mmu_memory_cache_alloc_obj(mc, GFP_KERNEL_ACCOUNT);
366 return mc->nobjs >= min ? 0 : -ENOMEM;
367 mc->objects[mc->nobjs++] = obj;
372 int kvm_mmu_memory_cache_nr_free_objects(struct kvm_mmu_memory_cache *mc)
377 void kvm_mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc)
381 kmem_cache_free(mc->kmem_cache, mc->objects[--mc->nobjs]);
383 free_page((unsigned long)mc->objects[--mc->nobjs]);
387 void *kvm_mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc)
391 if (WARN_ON(!mc->nobjs))
392 p = mmu_memory_cache_alloc_obj(mc, GFP_ATOMIC | __GFP_ACCOUNT);
394 p = mc->objects[--mc->nobjs];
400 static void kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
402 mutex_init(&vcpu->mutex);
407 rcuwait_init(&vcpu->wait);
408 kvm_async_pf_vcpu_init(vcpu);
411 INIT_LIST_HEAD(&vcpu->blocked_vcpu_list);
413 kvm_vcpu_set_in_spin_loop(vcpu, false);
414 kvm_vcpu_set_dy_eligible(vcpu, false);
415 vcpu->preempted = false;
417 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
420 void kvm_vcpu_destroy(struct kvm_vcpu *vcpu)
422 kvm_dirty_ring_free(&vcpu->dirty_ring);
423 kvm_arch_vcpu_destroy(vcpu);
426 * No need for rcu_read_lock as VCPU_RUN is the only place that changes
427 * the vcpu->pid pointer, and at destruction time all file descriptors
430 put_pid(rcu_dereference_protected(vcpu->pid, 1));
432 free_page((unsigned long)vcpu->run);
433 kmem_cache_free(kvm_vcpu_cache, vcpu);
435 EXPORT_SYMBOL_GPL(kvm_vcpu_destroy);
437 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
438 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
440 return container_of(mn, struct kvm, mmu_notifier);
443 static void kvm_mmu_notifier_invalidate_range(struct mmu_notifier *mn,
444 struct mm_struct *mm,
445 unsigned long start, unsigned long end)
447 struct kvm *kvm = mmu_notifier_to_kvm(mn);
450 idx = srcu_read_lock(&kvm->srcu);
451 kvm_arch_mmu_notifier_invalidate_range(kvm, start, end);
452 srcu_read_unlock(&kvm->srcu, idx);
455 typedef bool (*hva_handler_t)(struct kvm *kvm, struct kvm_gfn_range *range);
457 typedef void (*on_lock_fn_t)(struct kvm *kvm, unsigned long start,
460 struct kvm_hva_range {
464 hva_handler_t handler;
465 on_lock_fn_t on_lock;
471 * Use a dedicated stub instead of NULL to indicate that there is no callback
472 * function/handler. The compiler technically can't guarantee that a real
473 * function will have a non-zero address, and so it will generate code to
474 * check for !NULL, whereas comparing against a stub will be elided at compile
475 * time (unless the compiler is getting long in the tooth, e.g. gcc 4.9).
477 static void kvm_null_fn(void)
481 #define IS_KVM_NULL_FN(fn) ((fn) == (void *)kvm_null_fn)
483 static __always_inline int __kvm_handle_hva_range(struct kvm *kvm,
484 const struct kvm_hva_range *range)
486 bool ret = false, locked = false;
487 struct kvm_gfn_range gfn_range;
488 struct kvm_memory_slot *slot;
489 struct kvm_memslots *slots;
492 /* A null handler is allowed if and only if on_lock() is provided. */
493 if (WARN_ON_ONCE(IS_KVM_NULL_FN(range->on_lock) &&
494 IS_KVM_NULL_FN(range->handler)))
497 idx = srcu_read_lock(&kvm->srcu);
499 /* The on_lock() path does not yet support lock elision. */
500 if (!IS_KVM_NULL_FN(range->on_lock)) {
504 range->on_lock(kvm, range->start, range->end);
506 if (IS_KVM_NULL_FN(range->handler))
510 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
511 slots = __kvm_memslots(kvm, i);
512 kvm_for_each_memslot(slot, slots) {
513 unsigned long hva_start, hva_end;
515 hva_start = max(range->start, slot->userspace_addr);
516 hva_end = min(range->end, slot->userspace_addr +
517 (slot->npages << PAGE_SHIFT));
518 if (hva_start >= hva_end)
522 * To optimize for the likely case where the address
523 * range is covered by zero or one memslots, don't
524 * bother making these conditional (to avoid writes on
525 * the second or later invocation of the handler).
527 gfn_range.pte = range->pte;
528 gfn_range.may_block = range->may_block;
531 * {gfn(page) | page intersects with [hva_start, hva_end)} =
532 * {gfn_start, gfn_start+1, ..., gfn_end-1}.
534 gfn_range.start = hva_to_gfn_memslot(hva_start, slot);
535 gfn_range.end = hva_to_gfn_memslot(hva_end + PAGE_SIZE - 1, slot);
536 gfn_range.slot = slot;
542 ret |= range->handler(kvm, &gfn_range);
546 if (range->flush_on_ret && (ret || kvm->tlbs_dirty))
547 kvm_flush_remote_tlbs(kvm);
553 srcu_read_unlock(&kvm->srcu, idx);
555 /* The notifiers are averse to booleans. :-( */
559 static __always_inline int kvm_handle_hva_range(struct mmu_notifier *mn,
563 hva_handler_t handler)
565 struct kvm *kvm = mmu_notifier_to_kvm(mn);
566 const struct kvm_hva_range range = {
571 .on_lock = (void *)kvm_null_fn,
572 .flush_on_ret = true,
576 return __kvm_handle_hva_range(kvm, &range);
579 static __always_inline int kvm_handle_hva_range_no_flush(struct mmu_notifier *mn,
582 hva_handler_t handler)
584 struct kvm *kvm = mmu_notifier_to_kvm(mn);
585 const struct kvm_hva_range range = {
590 .on_lock = (void *)kvm_null_fn,
591 .flush_on_ret = false,
595 return __kvm_handle_hva_range(kvm, &range);
597 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
598 struct mm_struct *mm,
599 unsigned long address,
602 struct kvm *kvm = mmu_notifier_to_kvm(mn);
604 trace_kvm_set_spte_hva(address);
607 * .change_pte() must be surrounded by .invalidate_range_{start,end}(),
608 * and so always runs with an elevated notifier count. This obviates
609 * the need to bump the sequence count.
611 WARN_ON_ONCE(!kvm->mmu_notifier_count);
613 kvm_handle_hva_range(mn, address, address + 1, pte, kvm_set_spte_gfn);
616 static void kvm_inc_notifier_count(struct kvm *kvm, unsigned long start,
620 * The count increase must become visible at unlock time as no
621 * spte can be established without taking the mmu_lock and
622 * count is also read inside the mmu_lock critical section.
624 kvm->mmu_notifier_count++;
625 if (likely(kvm->mmu_notifier_count == 1)) {
626 kvm->mmu_notifier_range_start = start;
627 kvm->mmu_notifier_range_end = end;
630 * Fully tracking multiple concurrent ranges has dimishing
631 * returns. Keep things simple and just find the minimal range
632 * which includes the current and new ranges. As there won't be
633 * enough information to subtract a range after its invalidate
634 * completes, any ranges invalidated concurrently will
635 * accumulate and persist until all outstanding invalidates
638 kvm->mmu_notifier_range_start =
639 min(kvm->mmu_notifier_range_start, start);
640 kvm->mmu_notifier_range_end =
641 max(kvm->mmu_notifier_range_end, end);
645 static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
646 const struct mmu_notifier_range *range)
648 struct kvm *kvm = mmu_notifier_to_kvm(mn);
649 const struct kvm_hva_range hva_range = {
650 .start = range->start,
653 .handler = kvm_unmap_gfn_range,
654 .on_lock = kvm_inc_notifier_count,
655 .flush_on_ret = true,
656 .may_block = mmu_notifier_range_blockable(range),
659 trace_kvm_unmap_hva_range(range->start, range->end);
661 __kvm_handle_hva_range(kvm, &hva_range);
666 static void kvm_dec_notifier_count(struct kvm *kvm, unsigned long start,
670 * This sequence increase will notify the kvm page fault that
671 * the page that is going to be mapped in the spte could have
674 kvm->mmu_notifier_seq++;
677 * The above sequence increase must be visible before the
678 * below count decrease, which is ensured by the smp_wmb above
679 * in conjunction with the smp_rmb in mmu_notifier_retry().
681 kvm->mmu_notifier_count--;
684 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
685 const struct mmu_notifier_range *range)
687 struct kvm *kvm = mmu_notifier_to_kvm(mn);
688 const struct kvm_hva_range hva_range = {
689 .start = range->start,
692 .handler = (void *)kvm_null_fn,
693 .on_lock = kvm_dec_notifier_count,
694 .flush_on_ret = false,
695 .may_block = mmu_notifier_range_blockable(range),
698 __kvm_handle_hva_range(kvm, &hva_range);
700 BUG_ON(kvm->mmu_notifier_count < 0);
703 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
704 struct mm_struct *mm,
708 trace_kvm_age_hva(start, end);
710 return kvm_handle_hva_range(mn, start, end, __pte(0), kvm_age_gfn);
713 static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
714 struct mm_struct *mm,
718 trace_kvm_age_hva(start, end);
721 * Even though we do not flush TLB, this will still adversely
722 * affect performance on pre-Haswell Intel EPT, where there is
723 * no EPT Access Bit to clear so that we have to tear down EPT
724 * tables instead. If we find this unacceptable, we can always
725 * add a parameter to kvm_age_hva so that it effectively doesn't
726 * do anything on clear_young.
728 * Also note that currently we never issue secondary TLB flushes
729 * from clear_young, leaving this job up to the regular system
730 * cadence. If we find this inaccurate, we might come up with a
731 * more sophisticated heuristic later.
733 return kvm_handle_hva_range_no_flush(mn, start, end, kvm_age_gfn);
736 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
737 struct mm_struct *mm,
738 unsigned long address)
740 trace_kvm_test_age_hva(address);
742 return kvm_handle_hva_range_no_flush(mn, address, address + 1,
746 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
747 struct mm_struct *mm)
749 struct kvm *kvm = mmu_notifier_to_kvm(mn);
752 idx = srcu_read_lock(&kvm->srcu);
753 kvm_arch_flush_shadow_all(kvm);
754 srcu_read_unlock(&kvm->srcu, idx);
757 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
758 .invalidate_range = kvm_mmu_notifier_invalidate_range,
759 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
760 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
761 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
762 .clear_young = kvm_mmu_notifier_clear_young,
763 .test_young = kvm_mmu_notifier_test_young,
764 .change_pte = kvm_mmu_notifier_change_pte,
765 .release = kvm_mmu_notifier_release,
768 static int kvm_init_mmu_notifier(struct kvm *kvm)
770 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
771 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
774 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
776 static int kvm_init_mmu_notifier(struct kvm *kvm)
781 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
783 static struct kvm_memslots *kvm_alloc_memslots(void)
786 struct kvm_memslots *slots;
788 slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL_ACCOUNT);
792 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
793 slots->id_to_index[i] = -1;
798 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
800 if (!memslot->dirty_bitmap)
803 kvfree(memslot->dirty_bitmap);
804 memslot->dirty_bitmap = NULL;
807 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
809 kvm_destroy_dirty_bitmap(slot);
811 kvm_arch_free_memslot(kvm, slot);
817 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
819 struct kvm_memory_slot *memslot;
824 kvm_for_each_memslot(memslot, slots)
825 kvm_free_memslot(kvm, memslot);
830 static void kvm_destroy_vm_debugfs(struct kvm *kvm)
834 if (!kvm->debugfs_dentry)
837 debugfs_remove_recursive(kvm->debugfs_dentry);
839 if (kvm->debugfs_stat_data) {
840 for (i = 0; i < kvm_debugfs_num_entries; i++)
841 kfree(kvm->debugfs_stat_data[i]);
842 kfree(kvm->debugfs_stat_data);
846 static int kvm_create_vm_debugfs(struct kvm *kvm, int fd)
848 char dir_name[ITOA_MAX_LEN * 2];
849 struct kvm_stat_data *stat_data;
850 struct kvm_stats_debugfs_item *p;
852 if (!debugfs_initialized())
855 snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd);
856 kvm->debugfs_dentry = debugfs_create_dir(dir_name, kvm_debugfs_dir);
858 kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries,
859 sizeof(*kvm->debugfs_stat_data),
861 if (!kvm->debugfs_stat_data)
864 for (p = debugfs_entries; p->name; p++) {
865 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT);
869 stat_data->kvm = kvm;
870 stat_data->dbgfs_item = p;
871 kvm->debugfs_stat_data[p - debugfs_entries] = stat_data;
872 debugfs_create_file(p->name, KVM_DBGFS_GET_MODE(p),
873 kvm->debugfs_dentry, stat_data,
880 * Called after the VM is otherwise initialized, but just before adding it to
883 int __weak kvm_arch_post_init_vm(struct kvm *kvm)
889 * Called just after removing the VM from the vm_list, but before doing any
892 void __weak kvm_arch_pre_destroy_vm(struct kvm *kvm)
896 static struct kvm *kvm_create_vm(unsigned long type)
898 struct kvm *kvm = kvm_arch_alloc_vm();
903 return ERR_PTR(-ENOMEM);
905 KVM_MMU_LOCK_INIT(kvm);
907 kvm->mm = current->mm;
908 kvm_eventfd_init(kvm);
909 mutex_init(&kvm->lock);
910 mutex_init(&kvm->irq_lock);
911 mutex_init(&kvm->slots_lock);
912 INIT_LIST_HEAD(&kvm->devices);
914 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
916 if (init_srcu_struct(&kvm->srcu))
917 goto out_err_no_srcu;
918 if (init_srcu_struct(&kvm->irq_srcu))
919 goto out_err_no_irq_srcu;
921 refcount_set(&kvm->users_count, 1);
922 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
923 struct kvm_memslots *slots = kvm_alloc_memslots();
926 goto out_err_no_arch_destroy_vm;
927 /* Generations must be different for each address space. */
928 slots->generation = i;
929 rcu_assign_pointer(kvm->memslots[i], slots);
932 for (i = 0; i < KVM_NR_BUSES; i++) {
933 rcu_assign_pointer(kvm->buses[i],
934 kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL_ACCOUNT));
936 goto out_err_no_arch_destroy_vm;
939 kvm->max_halt_poll_ns = halt_poll_ns;
941 r = kvm_arch_init_vm(kvm, type);
943 goto out_err_no_arch_destroy_vm;
945 r = hardware_enable_all();
947 goto out_err_no_disable;
949 #ifdef CONFIG_HAVE_KVM_IRQFD
950 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
953 r = kvm_init_mmu_notifier(kvm);
955 goto out_err_no_mmu_notifier;
957 r = kvm_arch_post_init_vm(kvm);
961 mutex_lock(&kvm_lock);
962 list_add(&kvm->vm_list, &vm_list);
963 mutex_unlock(&kvm_lock);
965 preempt_notifier_inc();
970 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
971 if (kvm->mmu_notifier.ops)
972 mmu_notifier_unregister(&kvm->mmu_notifier, current->mm);
974 out_err_no_mmu_notifier:
975 hardware_disable_all();
977 kvm_arch_destroy_vm(kvm);
978 out_err_no_arch_destroy_vm:
979 WARN_ON_ONCE(!refcount_dec_and_test(&kvm->users_count));
980 for (i = 0; i < KVM_NR_BUSES; i++)
981 kfree(kvm_get_bus(kvm, i));
982 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
983 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
984 cleanup_srcu_struct(&kvm->irq_srcu);
986 cleanup_srcu_struct(&kvm->srcu);
988 kvm_arch_free_vm(kvm);
993 static void kvm_destroy_devices(struct kvm *kvm)
995 struct kvm_device *dev, *tmp;
998 * We do not need to take the kvm->lock here, because nobody else
999 * has a reference to the struct kvm at this point and therefore
1000 * cannot access the devices list anyhow.
1002 list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
1003 list_del(&dev->vm_node);
1004 dev->ops->destroy(dev);
1008 static void kvm_destroy_vm(struct kvm *kvm)
1011 struct mm_struct *mm = kvm->mm;
1013 kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm);
1014 kvm_destroy_vm_debugfs(kvm);
1015 kvm_arch_sync_events(kvm);
1016 mutex_lock(&kvm_lock);
1017 list_del(&kvm->vm_list);
1018 mutex_unlock(&kvm_lock);
1019 kvm_arch_pre_destroy_vm(kvm);
1021 kvm_free_irq_routing(kvm);
1022 for (i = 0; i < KVM_NR_BUSES; i++) {
1023 struct kvm_io_bus *bus = kvm_get_bus(kvm, i);
1026 kvm_io_bus_destroy(bus);
1027 kvm->buses[i] = NULL;
1029 kvm_coalesced_mmio_free(kvm);
1030 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1031 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
1033 kvm_arch_flush_shadow_all(kvm);
1035 kvm_arch_destroy_vm(kvm);
1036 kvm_destroy_devices(kvm);
1037 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
1038 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
1039 cleanup_srcu_struct(&kvm->irq_srcu);
1040 cleanup_srcu_struct(&kvm->srcu);
1041 kvm_arch_free_vm(kvm);
1042 preempt_notifier_dec();
1043 hardware_disable_all();
1047 void kvm_get_kvm(struct kvm *kvm)
1049 refcount_inc(&kvm->users_count);
1051 EXPORT_SYMBOL_GPL(kvm_get_kvm);
1053 void kvm_put_kvm(struct kvm *kvm)
1055 if (refcount_dec_and_test(&kvm->users_count))
1056 kvm_destroy_vm(kvm);
1058 EXPORT_SYMBOL_GPL(kvm_put_kvm);
1061 * Used to put a reference that was taken on behalf of an object associated
1062 * with a user-visible file descriptor, e.g. a vcpu or device, if installation
1063 * of the new file descriptor fails and the reference cannot be transferred to
1064 * its final owner. In such cases, the caller is still actively using @kvm and
1065 * will fail miserably if the refcount unexpectedly hits zero.
1067 void kvm_put_kvm_no_destroy(struct kvm *kvm)
1069 WARN_ON(refcount_dec_and_test(&kvm->users_count));
1071 EXPORT_SYMBOL_GPL(kvm_put_kvm_no_destroy);
1073 static int kvm_vm_release(struct inode *inode, struct file *filp)
1075 struct kvm *kvm = filp->private_data;
1077 kvm_irqfd_release(kvm);
1084 * Allocation size is twice as large as the actual dirty bitmap size.
1085 * See kvm_vm_ioctl_get_dirty_log() why this is needed.
1087 static int kvm_alloc_dirty_bitmap(struct kvm_memory_slot *memslot)
1089 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
1091 memslot->dirty_bitmap = kvzalloc(dirty_bytes, GFP_KERNEL_ACCOUNT);
1092 if (!memslot->dirty_bitmap)
1099 * Delete a memslot by decrementing the number of used slots and shifting all
1100 * other entries in the array forward one spot.
1102 static inline void kvm_memslot_delete(struct kvm_memslots *slots,
1103 struct kvm_memory_slot *memslot)
1105 struct kvm_memory_slot *mslots = slots->memslots;
1108 if (WARN_ON(slots->id_to_index[memslot->id] == -1))
1111 slots->used_slots--;
1113 if (atomic_read(&slots->lru_slot) >= slots->used_slots)
1114 atomic_set(&slots->lru_slot, 0);
1116 for (i = slots->id_to_index[memslot->id]; i < slots->used_slots; i++) {
1117 mslots[i] = mslots[i + 1];
1118 slots->id_to_index[mslots[i].id] = i;
1120 mslots[i] = *memslot;
1121 slots->id_to_index[memslot->id] = -1;
1125 * "Insert" a new memslot by incrementing the number of used slots. Returns
1126 * the new slot's initial index into the memslots array.
1128 static inline int kvm_memslot_insert_back(struct kvm_memslots *slots)
1130 return slots->used_slots++;
1134 * Move a changed memslot backwards in the array by shifting existing slots
1135 * with a higher GFN toward the front of the array. Note, the changed memslot
1136 * itself is not preserved in the array, i.e. not swapped at this time, only
1137 * its new index into the array is tracked. Returns the changed memslot's
1138 * current index into the memslots array.
1140 static inline int kvm_memslot_move_backward(struct kvm_memslots *slots,
1141 struct kvm_memory_slot *memslot)
1143 struct kvm_memory_slot *mslots = slots->memslots;
1146 if (WARN_ON_ONCE(slots->id_to_index[memslot->id] == -1) ||
1147 WARN_ON_ONCE(!slots->used_slots))
1151 * Move the target memslot backward in the array by shifting existing
1152 * memslots with a higher GFN (than the target memslot) towards the
1153 * front of the array.
1155 for (i = slots->id_to_index[memslot->id]; i < slots->used_slots - 1; i++) {
1156 if (memslot->base_gfn > mslots[i + 1].base_gfn)
1159 WARN_ON_ONCE(memslot->base_gfn == mslots[i + 1].base_gfn);
1161 /* Shift the next memslot forward one and update its index. */
1162 mslots[i] = mslots[i + 1];
1163 slots->id_to_index[mslots[i].id] = i;
1169 * Move a changed memslot forwards in the array by shifting existing slots with
1170 * a lower GFN toward the back of the array. Note, the changed memslot itself
1171 * is not preserved in the array, i.e. not swapped at this time, only its new
1172 * index into the array is tracked. Returns the changed memslot's final index
1173 * into the memslots array.
1175 static inline int kvm_memslot_move_forward(struct kvm_memslots *slots,
1176 struct kvm_memory_slot *memslot,
1179 struct kvm_memory_slot *mslots = slots->memslots;
1182 for (i = start; i > 0; i--) {
1183 if (memslot->base_gfn < mslots[i - 1].base_gfn)
1186 WARN_ON_ONCE(memslot->base_gfn == mslots[i - 1].base_gfn);
1188 /* Shift the next memslot back one and update its index. */
1189 mslots[i] = mslots[i - 1];
1190 slots->id_to_index[mslots[i].id] = i;
1196 * Re-sort memslots based on their GFN to account for an added, deleted, or
1197 * moved memslot. Sorting memslots by GFN allows using a binary search during
1200 * IMPORTANT: Slots are sorted from highest GFN to lowest GFN! I.e. the entry
1201 * at memslots[0] has the highest GFN.
1203 * The sorting algorithm takes advantage of having initially sorted memslots
1204 * and knowing the position of the changed memslot. Sorting is also optimized
1205 * by not swapping the updated memslot and instead only shifting other memslots
1206 * and tracking the new index for the update memslot. Only once its final
1207 * index is known is the updated memslot copied into its position in the array.
1209 * - When deleting a memslot, the deleted memslot simply needs to be moved to
1210 * the end of the array.
1212 * - When creating a memslot, the algorithm "inserts" the new memslot at the
1213 * end of the array and then it forward to its correct location.
1215 * - When moving a memslot, the algorithm first moves the updated memslot
1216 * backward to handle the scenario where the memslot's GFN was changed to a
1217 * lower value. update_memslots() then falls through and runs the same flow
1218 * as creating a memslot to move the memslot forward to handle the scenario
1219 * where its GFN was changed to a higher value.
1221 * Note, slots are sorted from highest->lowest instead of lowest->highest for
1222 * historical reasons. Originally, invalid memslots where denoted by having
1223 * GFN=0, thus sorting from highest->lowest naturally sorted invalid memslots
1224 * to the end of the array. The current algorithm uses dedicated logic to
1225 * delete a memslot and thus does not rely on invalid memslots having GFN=0.
1227 * The other historical motiviation for highest->lowest was to improve the
1228 * performance of memslot lookup. KVM originally used a linear search starting
1229 * at memslots[0]. On x86, the largest memslot usually has one of the highest,
1230 * if not *the* highest, GFN, as the bulk of the guest's RAM is located in a
1231 * single memslot above the 4gb boundary. As the largest memslot is also the
1232 * most likely to be referenced, sorting it to the front of the array was
1233 * advantageous. The current binary search starts from the middle of the array
1234 * and uses an LRU pointer to improve performance for all memslots and GFNs.
1236 static void update_memslots(struct kvm_memslots *slots,
1237 struct kvm_memory_slot *memslot,
1238 enum kvm_mr_change change)
1242 if (change == KVM_MR_DELETE) {
1243 kvm_memslot_delete(slots, memslot);
1245 if (change == KVM_MR_CREATE)
1246 i = kvm_memslot_insert_back(slots);
1248 i = kvm_memslot_move_backward(slots, memslot);
1249 i = kvm_memslot_move_forward(slots, memslot, i);
1252 * Copy the memslot to its new position in memslots and update
1253 * its index accordingly.
1255 slots->memslots[i] = *memslot;
1256 slots->id_to_index[memslot->id] = i;
1260 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
1262 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
1264 #ifdef __KVM_HAVE_READONLY_MEM
1265 valid_flags |= KVM_MEM_READONLY;
1268 if (mem->flags & ~valid_flags)
1274 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
1275 int as_id, struct kvm_memslots *slots)
1277 struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id);
1278 u64 gen = old_memslots->generation;
1280 WARN_ON(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS);
1281 slots->generation = gen | KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
1283 rcu_assign_pointer(kvm->memslots[as_id], slots);
1284 synchronize_srcu_expedited(&kvm->srcu);
1287 * Increment the new memslot generation a second time, dropping the
1288 * update in-progress flag and incrementing the generation based on
1289 * the number of address spaces. This provides a unique and easily
1290 * identifiable generation number while the memslots are in flux.
1292 gen = slots->generation & ~KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
1295 * Generations must be unique even across address spaces. We do not need
1296 * a global counter for that, instead the generation space is evenly split
1297 * across address spaces. For example, with two address spaces, address
1298 * space 0 will use generations 0, 2, 4, ... while address space 1 will
1299 * use generations 1, 3, 5, ...
1301 gen += KVM_ADDRESS_SPACE_NUM;
1303 kvm_arch_memslots_updated(kvm, gen);
1305 slots->generation = gen;
1307 return old_memslots;
1311 * Note, at a minimum, the current number of used slots must be allocated, even
1312 * when deleting a memslot, as we need a complete duplicate of the memslots for
1313 * use when invalidating a memslot prior to deleting/moving the memslot.
1315 static struct kvm_memslots *kvm_dup_memslots(struct kvm_memslots *old,
1316 enum kvm_mr_change change)
1318 struct kvm_memslots *slots;
1319 size_t old_size, new_size;
1321 old_size = sizeof(struct kvm_memslots) +
1322 (sizeof(struct kvm_memory_slot) * old->used_slots);
1324 if (change == KVM_MR_CREATE)
1325 new_size = old_size + sizeof(struct kvm_memory_slot);
1327 new_size = old_size;
1329 slots = kvzalloc(new_size, GFP_KERNEL_ACCOUNT);
1331 memcpy(slots, old, old_size);
1336 static int kvm_set_memslot(struct kvm *kvm,
1337 const struct kvm_userspace_memory_region *mem,
1338 struct kvm_memory_slot *old,
1339 struct kvm_memory_slot *new, int as_id,
1340 enum kvm_mr_change change)
1342 struct kvm_memory_slot *slot;
1343 struct kvm_memslots *slots;
1346 slots = kvm_dup_memslots(__kvm_memslots(kvm, as_id), change);
1350 if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) {
1352 * Note, the INVALID flag needs to be in the appropriate entry
1353 * in the freshly allocated memslots, not in @old or @new.
1355 slot = id_to_memslot(slots, old->id);
1356 slot->flags |= KVM_MEMSLOT_INVALID;
1359 * We can re-use the old memslots, the only difference from the
1360 * newly installed memslots is the invalid flag, which will get
1361 * dropped by update_memslots anyway. We'll also revert to the
1362 * old memslots if preparing the new memory region fails.
1364 slots = install_new_memslots(kvm, as_id, slots);
1366 /* From this point no new shadow pages pointing to a deleted,
1367 * or moved, memslot will be created.
1369 * validation of sp->gfn happens in:
1370 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1371 * - kvm_is_visible_gfn (mmu_check_root)
1373 kvm_arch_flush_shadow_memslot(kvm, slot);
1376 r = kvm_arch_prepare_memory_region(kvm, new, mem, change);
1380 update_memslots(slots, new, change);
1381 slots = install_new_memslots(kvm, as_id, slots);
1383 kvm_arch_commit_memory_region(kvm, mem, old, new, change);
1389 if (change == KVM_MR_DELETE || change == KVM_MR_MOVE)
1390 slots = install_new_memslots(kvm, as_id, slots);
1395 static int kvm_delete_memslot(struct kvm *kvm,
1396 const struct kvm_userspace_memory_region *mem,
1397 struct kvm_memory_slot *old, int as_id)
1399 struct kvm_memory_slot new;
1405 memset(&new, 0, sizeof(new));
1408 * This is only for debugging purpose; it should never be referenced
1409 * for a removed memslot.
1413 r = kvm_set_memslot(kvm, mem, old, &new, as_id, KVM_MR_DELETE);
1417 kvm_free_memslot(kvm, old);
1422 * Allocate some memory and give it an address in the guest physical address
1425 * Discontiguous memory is allowed, mostly for framebuffers.
1427 * Must be called holding kvm->slots_lock for write.
1429 int __kvm_set_memory_region(struct kvm *kvm,
1430 const struct kvm_userspace_memory_region *mem)
1432 struct kvm_memory_slot old, new;
1433 struct kvm_memory_slot *tmp;
1434 enum kvm_mr_change change;
1438 r = check_memory_region_flags(mem);
1442 as_id = mem->slot >> 16;
1443 id = (u16)mem->slot;
1445 /* General sanity checks */
1446 if (mem->memory_size & (PAGE_SIZE - 1))
1448 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
1450 /* We can read the guest memory with __xxx_user() later on. */
1451 if ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
1452 (mem->userspace_addr != untagged_addr(mem->userspace_addr)) ||
1453 !access_ok((void __user *)(unsigned long)mem->userspace_addr,
1456 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
1458 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
1462 * Make a full copy of the old memslot, the pointer will become stale
1463 * when the memslots are re-sorted by update_memslots(), and the old
1464 * memslot needs to be referenced after calling update_memslots(), e.g.
1465 * to free its resources and for arch specific behavior.
1467 tmp = id_to_memslot(__kvm_memslots(kvm, as_id), id);
1472 memset(&old, 0, sizeof(old));
1476 if (!mem->memory_size)
1477 return kvm_delete_memslot(kvm, mem, &old, as_id);
1481 new.base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
1482 new.npages = mem->memory_size >> PAGE_SHIFT;
1483 new.flags = mem->flags;
1484 new.userspace_addr = mem->userspace_addr;
1486 if (new.npages > KVM_MEM_MAX_NR_PAGES)
1490 change = KVM_MR_CREATE;
1491 new.dirty_bitmap = NULL;
1492 memset(&new.arch, 0, sizeof(new.arch));
1493 } else { /* Modify an existing slot. */
1494 if ((new.userspace_addr != old.userspace_addr) ||
1495 (new.npages != old.npages) ||
1496 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
1499 if (new.base_gfn != old.base_gfn)
1500 change = KVM_MR_MOVE;
1501 else if (new.flags != old.flags)
1502 change = KVM_MR_FLAGS_ONLY;
1503 else /* Nothing to change. */
1506 /* Copy dirty_bitmap and arch from the current memslot. */
1507 new.dirty_bitmap = old.dirty_bitmap;
1508 memcpy(&new.arch, &old.arch, sizeof(new.arch));
1511 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
1512 /* Check for overlaps */
1513 kvm_for_each_memslot(tmp, __kvm_memslots(kvm, as_id)) {
1516 if (!((new.base_gfn + new.npages <= tmp->base_gfn) ||
1517 (new.base_gfn >= tmp->base_gfn + tmp->npages)))
1522 /* Allocate/free page dirty bitmap as needed */
1523 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
1524 new.dirty_bitmap = NULL;
1525 else if (!new.dirty_bitmap && !kvm->dirty_ring_size) {
1526 r = kvm_alloc_dirty_bitmap(&new);
1530 if (kvm_dirty_log_manual_protect_and_init_set(kvm))
1531 bitmap_set(new.dirty_bitmap, 0, new.npages);
1534 r = kvm_set_memslot(kvm, mem, &old, &new, as_id, change);
1538 if (old.dirty_bitmap && !new.dirty_bitmap)
1539 kvm_destroy_dirty_bitmap(&old);
1543 if (new.dirty_bitmap && !old.dirty_bitmap)
1544 kvm_destroy_dirty_bitmap(&new);
1547 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1549 int kvm_set_memory_region(struct kvm *kvm,
1550 const struct kvm_userspace_memory_region *mem)
1554 mutex_lock(&kvm->slots_lock);
1555 r = __kvm_set_memory_region(kvm, mem);
1556 mutex_unlock(&kvm->slots_lock);
1559 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1561 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1562 struct kvm_userspace_memory_region *mem)
1564 if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1567 return kvm_set_memory_region(kvm, mem);
1570 #ifndef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1572 * kvm_get_dirty_log - get a snapshot of dirty pages
1573 * @kvm: pointer to kvm instance
1574 * @log: slot id and address to which we copy the log
1575 * @is_dirty: set to '1' if any dirty pages were found
1576 * @memslot: set to the associated memslot, always valid on success
1578 int kvm_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log,
1579 int *is_dirty, struct kvm_memory_slot **memslot)
1581 struct kvm_memslots *slots;
1584 unsigned long any = 0;
1586 /* Dirty ring tracking is exclusive to dirty log tracking */
1587 if (kvm->dirty_ring_size)
1593 as_id = log->slot >> 16;
1594 id = (u16)log->slot;
1595 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1598 slots = __kvm_memslots(kvm, as_id);
1599 *memslot = id_to_memslot(slots, id);
1600 if (!(*memslot) || !(*memslot)->dirty_bitmap)
1603 kvm_arch_sync_dirty_log(kvm, *memslot);
1605 n = kvm_dirty_bitmap_bytes(*memslot);
1607 for (i = 0; !any && i < n/sizeof(long); ++i)
1608 any = (*memslot)->dirty_bitmap[i];
1610 if (copy_to_user(log->dirty_bitmap, (*memslot)->dirty_bitmap, n))
1617 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1619 #else /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
1621 * kvm_get_dirty_log_protect - get a snapshot of dirty pages
1622 * and reenable dirty page tracking for the corresponding pages.
1623 * @kvm: pointer to kvm instance
1624 * @log: slot id and address to which we copy the log
1626 * We need to keep it in mind that VCPU threads can write to the bitmap
1627 * concurrently. So, to avoid losing track of dirty pages we keep the
1630 * 1. Take a snapshot of the bit and clear it if needed.
1631 * 2. Write protect the corresponding page.
1632 * 3. Copy the snapshot to the userspace.
1633 * 4. Upon return caller flushes TLB's if needed.
1635 * Between 2 and 4, the guest may write to the page using the remaining TLB
1636 * entry. This is not a problem because the page is reported dirty using
1637 * the snapshot taken before and step 4 ensures that writes done after
1638 * exiting to userspace will be logged for the next call.
1641 static int kvm_get_dirty_log_protect(struct kvm *kvm, struct kvm_dirty_log *log)
1643 struct kvm_memslots *slots;
1644 struct kvm_memory_slot *memslot;
1647 unsigned long *dirty_bitmap;
1648 unsigned long *dirty_bitmap_buffer;
1651 /* Dirty ring tracking is exclusive to dirty log tracking */
1652 if (kvm->dirty_ring_size)
1655 as_id = log->slot >> 16;
1656 id = (u16)log->slot;
1657 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1660 slots = __kvm_memslots(kvm, as_id);
1661 memslot = id_to_memslot(slots, id);
1662 if (!memslot || !memslot->dirty_bitmap)
1665 dirty_bitmap = memslot->dirty_bitmap;
1667 kvm_arch_sync_dirty_log(kvm, memslot);
1669 n = kvm_dirty_bitmap_bytes(memslot);
1671 if (kvm->manual_dirty_log_protect) {
1673 * Unlike kvm_get_dirty_log, we always return false in *flush,
1674 * because no flush is needed until KVM_CLEAR_DIRTY_LOG. There
1675 * is some code duplication between this function and
1676 * kvm_get_dirty_log, but hopefully all architecture
1677 * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log
1678 * can be eliminated.
1680 dirty_bitmap_buffer = dirty_bitmap;
1682 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
1683 memset(dirty_bitmap_buffer, 0, n);
1686 for (i = 0; i < n / sizeof(long); i++) {
1690 if (!dirty_bitmap[i])
1694 mask = xchg(&dirty_bitmap[i], 0);
1695 dirty_bitmap_buffer[i] = mask;
1697 offset = i * BITS_PER_LONG;
1698 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1701 KVM_MMU_UNLOCK(kvm);
1705 kvm_arch_flush_remote_tlbs_memslot(kvm, memslot);
1707 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1714 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
1715 * @kvm: kvm instance
1716 * @log: slot id and address to which we copy the log
1718 * Steps 1-4 below provide general overview of dirty page logging. See
1719 * kvm_get_dirty_log_protect() function description for additional details.
1721 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
1722 * always flush the TLB (step 4) even if previous step failed and the dirty
1723 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
1724 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
1725 * writes will be marked dirty for next log read.
1727 * 1. Take a snapshot of the bit and clear it if needed.
1728 * 2. Write protect the corresponding page.
1729 * 3. Copy the snapshot to the userspace.
1730 * 4. Flush TLB's if needed.
1732 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1733 struct kvm_dirty_log *log)
1737 mutex_lock(&kvm->slots_lock);
1739 r = kvm_get_dirty_log_protect(kvm, log);
1741 mutex_unlock(&kvm->slots_lock);
1746 * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap
1747 * and reenable dirty page tracking for the corresponding pages.
1748 * @kvm: pointer to kvm instance
1749 * @log: slot id and address from which to fetch the bitmap of dirty pages
1751 static int kvm_clear_dirty_log_protect(struct kvm *kvm,
1752 struct kvm_clear_dirty_log *log)
1754 struct kvm_memslots *slots;
1755 struct kvm_memory_slot *memslot;
1759 unsigned long *dirty_bitmap;
1760 unsigned long *dirty_bitmap_buffer;
1763 /* Dirty ring tracking is exclusive to dirty log tracking */
1764 if (kvm->dirty_ring_size)
1767 as_id = log->slot >> 16;
1768 id = (u16)log->slot;
1769 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1772 if (log->first_page & 63)
1775 slots = __kvm_memslots(kvm, as_id);
1776 memslot = id_to_memslot(slots, id);
1777 if (!memslot || !memslot->dirty_bitmap)
1780 dirty_bitmap = memslot->dirty_bitmap;
1782 n = ALIGN(log->num_pages, BITS_PER_LONG) / 8;
1784 if (log->first_page > memslot->npages ||
1785 log->num_pages > memslot->npages - log->first_page ||
1786 (log->num_pages < memslot->npages - log->first_page && (log->num_pages & 63)))
1789 kvm_arch_sync_dirty_log(kvm, memslot);
1792 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
1793 if (copy_from_user(dirty_bitmap_buffer, log->dirty_bitmap, n))
1797 for (offset = log->first_page, i = offset / BITS_PER_LONG,
1798 n = DIV_ROUND_UP(log->num_pages, BITS_PER_LONG); n--;
1799 i++, offset += BITS_PER_LONG) {
1800 unsigned long mask = *dirty_bitmap_buffer++;
1801 atomic_long_t *p = (atomic_long_t *) &dirty_bitmap[i];
1805 mask &= atomic_long_fetch_andnot(mask, p);
1808 * mask contains the bits that really have been cleared. This
1809 * never includes any bits beyond the length of the memslot (if
1810 * the length is not aligned to 64 pages), therefore it is not
1811 * a problem if userspace sets them in log->dirty_bitmap.
1815 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1819 KVM_MMU_UNLOCK(kvm);
1822 kvm_arch_flush_remote_tlbs_memslot(kvm, memslot);
1827 static int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm,
1828 struct kvm_clear_dirty_log *log)
1832 mutex_lock(&kvm->slots_lock);
1834 r = kvm_clear_dirty_log_protect(kvm, log);
1836 mutex_unlock(&kvm->slots_lock);
1839 #endif /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
1841 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1843 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1845 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1847 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
1849 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn);
1851 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_memslot);
1853 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1855 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1857 return kvm_is_visible_memslot(memslot);
1859 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1861 bool kvm_vcpu_is_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
1863 struct kvm_memory_slot *memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1865 return kvm_is_visible_memslot(memslot);
1867 EXPORT_SYMBOL_GPL(kvm_vcpu_is_visible_gfn);
1869 unsigned long kvm_host_page_size(struct kvm_vcpu *vcpu, gfn_t gfn)
1871 struct vm_area_struct *vma;
1872 unsigned long addr, size;
1876 addr = kvm_vcpu_gfn_to_hva_prot(vcpu, gfn, NULL);
1877 if (kvm_is_error_hva(addr))
1880 mmap_read_lock(current->mm);
1881 vma = find_vma(current->mm, addr);
1885 size = vma_kernel_pagesize(vma);
1888 mmap_read_unlock(current->mm);
1893 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1895 return slot->flags & KVM_MEM_READONLY;
1898 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1899 gfn_t *nr_pages, bool write)
1901 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1902 return KVM_HVA_ERR_BAD;
1904 if (memslot_is_readonly(slot) && write)
1905 return KVM_HVA_ERR_RO_BAD;
1908 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1910 return __gfn_to_hva_memslot(slot, gfn);
1913 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1916 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1919 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1922 return gfn_to_hva_many(slot, gfn, NULL);
1924 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1926 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1928 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1930 EXPORT_SYMBOL_GPL(gfn_to_hva);
1932 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
1934 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
1936 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
1939 * Return the hva of a @gfn and the R/W attribute if possible.
1941 * @slot: the kvm_memory_slot which contains @gfn
1942 * @gfn: the gfn to be translated
1943 * @writable: used to return the read/write attribute of the @slot if the hva
1944 * is valid and @writable is not NULL
1946 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1947 gfn_t gfn, bool *writable)
1949 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1951 if (!kvm_is_error_hva(hva) && writable)
1952 *writable = !memslot_is_readonly(slot);
1957 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1959 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1961 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1964 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
1966 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1968 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1971 static inline int check_user_page_hwpoison(unsigned long addr)
1973 int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
1975 rc = get_user_pages(addr, 1, flags, NULL, NULL);
1976 return rc == -EHWPOISON;
1980 * The fast path to get the writable pfn which will be stored in @pfn,
1981 * true indicates success, otherwise false is returned. It's also the
1982 * only part that runs if we can in atomic context.
1984 static bool hva_to_pfn_fast(unsigned long addr, bool write_fault,
1985 bool *writable, kvm_pfn_t *pfn)
1987 struct page *page[1];
1990 * Fast pin a writable pfn only if it is a write fault request
1991 * or the caller allows to map a writable pfn for a read fault
1994 if (!(write_fault || writable))
1997 if (get_user_page_fast_only(addr, FOLL_WRITE, page)) {
1998 *pfn = page_to_pfn(page[0]);
2009 * The slow path to get the pfn of the specified host virtual address,
2010 * 1 indicates success, -errno is returned if error is detected.
2012 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
2013 bool *writable, kvm_pfn_t *pfn)
2015 unsigned int flags = FOLL_HWPOISON;
2022 *writable = write_fault;
2025 flags |= FOLL_WRITE;
2027 flags |= FOLL_NOWAIT;
2029 npages = get_user_pages_unlocked(addr, 1, &page, flags);
2033 /* map read fault as writable if possible */
2034 if (unlikely(!write_fault) && writable) {
2037 if (get_user_page_fast_only(addr, FOLL_WRITE, &wpage)) {
2043 *pfn = page_to_pfn(page);
2047 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
2049 if (unlikely(!(vma->vm_flags & VM_READ)))
2052 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
2058 static int kvm_try_get_pfn(kvm_pfn_t pfn)
2060 if (kvm_is_reserved_pfn(pfn))
2062 return get_page_unless_zero(pfn_to_page(pfn));
2065 static int hva_to_pfn_remapped(struct vm_area_struct *vma,
2066 unsigned long addr, bool *async,
2067 bool write_fault, bool *writable,
2075 r = follow_pte(vma->vm_mm, addr, &ptep, &ptl);
2078 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
2079 * not call the fault handler, so do it here.
2081 bool unlocked = false;
2082 r = fixup_user_fault(current->mm, addr,
2083 (write_fault ? FAULT_FLAG_WRITE : 0),
2090 r = follow_pte(vma->vm_mm, addr, &ptep, &ptl);
2095 if (write_fault && !pte_write(*ptep)) {
2096 pfn = KVM_PFN_ERR_RO_FAULT;
2101 *writable = pte_write(*ptep);
2102 pfn = pte_pfn(*ptep);
2105 * Get a reference here because callers of *hva_to_pfn* and
2106 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
2107 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
2108 * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will
2109 * simply do nothing for reserved pfns.
2111 * Whoever called remap_pfn_range is also going to call e.g.
2112 * unmap_mapping_range before the underlying pages are freed,
2113 * causing a call to our MMU notifier.
2115 * Certain IO or PFNMAP mappings can be backed with valid
2116 * struct pages, but be allocated without refcounting e.g.,
2117 * tail pages of non-compound higher order allocations, which
2118 * would then underflow the refcount when the caller does the
2119 * required put_page. Don't allow those pages here.
2121 if (!kvm_try_get_pfn(pfn))
2125 pte_unmap_unlock(ptep, ptl);
2132 * Pin guest page in memory and return its pfn.
2133 * @addr: host virtual address which maps memory to the guest
2134 * @atomic: whether this function can sleep
2135 * @async: whether this function need to wait IO complete if the
2136 * host page is not in the memory
2137 * @write_fault: whether we should get a writable host page
2138 * @writable: whether it allows to map a writable host page for !@write_fault
2140 * The function will map a writable host page for these two cases:
2141 * 1): @write_fault = true
2142 * 2): @write_fault = false && @writable, @writable will tell the caller
2143 * whether the mapping is writable.
2145 static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
2146 bool write_fault, bool *writable)
2148 struct vm_area_struct *vma;
2152 /* we can do it either atomically or asynchronously, not both */
2153 BUG_ON(atomic && async);
2155 if (hva_to_pfn_fast(addr, write_fault, writable, &pfn))
2159 return KVM_PFN_ERR_FAULT;
2161 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
2165 mmap_read_lock(current->mm);
2166 if (npages == -EHWPOISON ||
2167 (!async && check_user_page_hwpoison(addr))) {
2168 pfn = KVM_PFN_ERR_HWPOISON;
2173 vma = find_vma_intersection(current->mm, addr, addr + 1);
2176 pfn = KVM_PFN_ERR_FAULT;
2177 else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
2178 r = hva_to_pfn_remapped(vma, addr, async, write_fault, writable, &pfn);
2182 pfn = KVM_PFN_ERR_FAULT;
2184 if (async && vma_is_valid(vma, write_fault))
2186 pfn = KVM_PFN_ERR_FAULT;
2189 mmap_read_unlock(current->mm);
2193 kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn,
2194 bool atomic, bool *async, bool write_fault,
2195 bool *writable, hva_t *hva)
2197 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
2202 if (addr == KVM_HVA_ERR_RO_BAD) {
2205 return KVM_PFN_ERR_RO_FAULT;
2208 if (kvm_is_error_hva(addr)) {
2211 return KVM_PFN_NOSLOT;
2214 /* Do not map writable pfn in the readonly memslot. */
2215 if (writable && memslot_is_readonly(slot)) {
2220 return hva_to_pfn(addr, atomic, async, write_fault,
2223 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
2225 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
2228 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
2229 write_fault, writable, NULL);
2231 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
2233 kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
2235 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL, NULL);
2237 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
2239 kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
2241 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL, NULL);
2243 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
2245 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
2247 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
2249 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
2251 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
2253 return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
2255 EXPORT_SYMBOL_GPL(gfn_to_pfn);
2257 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
2259 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
2261 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
2263 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
2264 struct page **pages, int nr_pages)
2269 addr = gfn_to_hva_many(slot, gfn, &entry);
2270 if (kvm_is_error_hva(addr))
2273 if (entry < nr_pages)
2276 return get_user_pages_fast_only(addr, nr_pages, FOLL_WRITE, pages);
2278 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
2280 static struct page *kvm_pfn_to_page(kvm_pfn_t pfn)
2282 if (is_error_noslot_pfn(pfn))
2283 return KVM_ERR_PTR_BAD_PAGE;
2285 if (kvm_is_reserved_pfn(pfn)) {
2287 return KVM_ERR_PTR_BAD_PAGE;
2290 return pfn_to_page(pfn);
2293 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
2297 pfn = gfn_to_pfn(kvm, gfn);
2299 return kvm_pfn_to_page(pfn);
2301 EXPORT_SYMBOL_GPL(gfn_to_page);
2303 void kvm_release_pfn(kvm_pfn_t pfn, bool dirty, struct gfn_to_pfn_cache *cache)
2309 cache->pfn = cache->gfn = 0;
2312 kvm_release_pfn_dirty(pfn);
2314 kvm_release_pfn_clean(pfn);
2317 static void kvm_cache_gfn_to_pfn(struct kvm_memory_slot *slot, gfn_t gfn,
2318 struct gfn_to_pfn_cache *cache, u64 gen)
2320 kvm_release_pfn(cache->pfn, cache->dirty, cache);
2322 cache->pfn = gfn_to_pfn_memslot(slot, gfn);
2324 cache->dirty = false;
2325 cache->generation = gen;
2328 static int __kvm_map_gfn(struct kvm_memslots *slots, gfn_t gfn,
2329 struct kvm_host_map *map,
2330 struct gfn_to_pfn_cache *cache,
2335 struct page *page = KVM_UNMAPPED_PAGE;
2336 struct kvm_memory_slot *slot = __gfn_to_memslot(slots, gfn);
2337 u64 gen = slots->generation;
2343 if (!cache->pfn || cache->gfn != gfn ||
2344 cache->generation != gen) {
2347 kvm_cache_gfn_to_pfn(slot, gfn, cache, gen);
2353 pfn = gfn_to_pfn_memslot(slot, gfn);
2355 if (is_error_noslot_pfn(pfn))
2358 if (pfn_valid(pfn)) {
2359 page = pfn_to_page(pfn);
2361 hva = kmap_atomic(page);
2364 #ifdef CONFIG_HAS_IOMEM
2365 } else if (!atomic) {
2366 hva = memremap(pfn_to_hpa(pfn), PAGE_SIZE, MEMREMAP_WB);
2383 int kvm_map_gfn(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map,
2384 struct gfn_to_pfn_cache *cache, bool atomic)
2386 return __kvm_map_gfn(kvm_memslots(vcpu->kvm), gfn, map,
2389 EXPORT_SYMBOL_GPL(kvm_map_gfn);
2391 int kvm_vcpu_map(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map)
2393 return __kvm_map_gfn(kvm_vcpu_memslots(vcpu), gfn, map,
2396 EXPORT_SYMBOL_GPL(kvm_vcpu_map);
2398 static void __kvm_unmap_gfn(struct kvm *kvm,
2399 struct kvm_memory_slot *memslot,
2400 struct kvm_host_map *map,
2401 struct gfn_to_pfn_cache *cache,
2402 bool dirty, bool atomic)
2410 if (map->page != KVM_UNMAPPED_PAGE) {
2412 kunmap_atomic(map->hva);
2416 #ifdef CONFIG_HAS_IOMEM
2420 WARN_ONCE(1, "Unexpected unmapping in atomic context");
2424 mark_page_dirty_in_slot(kvm, memslot, map->gfn);
2427 cache->dirty |= dirty;
2429 kvm_release_pfn(map->pfn, dirty, NULL);
2435 int kvm_unmap_gfn(struct kvm_vcpu *vcpu, struct kvm_host_map *map,
2436 struct gfn_to_pfn_cache *cache, bool dirty, bool atomic)
2438 __kvm_unmap_gfn(vcpu->kvm, gfn_to_memslot(vcpu->kvm, map->gfn), map,
2439 cache, dirty, atomic);
2442 EXPORT_SYMBOL_GPL(kvm_unmap_gfn);
2444 void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map, bool dirty)
2446 __kvm_unmap_gfn(vcpu->kvm, kvm_vcpu_gfn_to_memslot(vcpu, map->gfn),
2447 map, NULL, dirty, false);
2449 EXPORT_SYMBOL_GPL(kvm_vcpu_unmap);
2451 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
2455 pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
2457 return kvm_pfn_to_page(pfn);
2459 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
2461 void kvm_release_page_clean(struct page *page)
2463 WARN_ON(is_error_page(page));
2465 kvm_release_pfn_clean(page_to_pfn(page));
2467 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
2469 void kvm_release_pfn_clean(kvm_pfn_t pfn)
2471 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
2472 put_page(pfn_to_page(pfn));
2474 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
2476 void kvm_release_page_dirty(struct page *page)
2478 WARN_ON(is_error_page(page));
2480 kvm_release_pfn_dirty(page_to_pfn(page));
2482 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
2484 void kvm_release_pfn_dirty(kvm_pfn_t pfn)
2486 kvm_set_pfn_dirty(pfn);
2487 kvm_release_pfn_clean(pfn);
2489 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
2491 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
2493 if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn))
2494 SetPageDirty(pfn_to_page(pfn));
2496 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
2498 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
2500 if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn))
2501 mark_page_accessed(pfn_to_page(pfn));
2503 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
2505 void kvm_get_pfn(kvm_pfn_t pfn)
2507 if (!kvm_is_reserved_pfn(pfn))
2508 get_page(pfn_to_page(pfn));
2510 EXPORT_SYMBOL_GPL(kvm_get_pfn);
2512 static int next_segment(unsigned long len, int offset)
2514 if (len > PAGE_SIZE - offset)
2515 return PAGE_SIZE - offset;
2520 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
2521 void *data, int offset, int len)
2526 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
2527 if (kvm_is_error_hva(addr))
2529 r = __copy_from_user(data, (void __user *)addr + offset, len);
2535 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
2538 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2540 return __kvm_read_guest_page(slot, gfn, data, offset, len);
2542 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
2544 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
2545 int offset, int len)
2547 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2549 return __kvm_read_guest_page(slot, gfn, data, offset, len);
2551 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
2553 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
2555 gfn_t gfn = gpa >> PAGE_SHIFT;
2557 int offset = offset_in_page(gpa);
2560 while ((seg = next_segment(len, offset)) != 0) {
2561 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
2571 EXPORT_SYMBOL_GPL(kvm_read_guest);
2573 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
2575 gfn_t gfn = gpa >> PAGE_SHIFT;
2577 int offset = offset_in_page(gpa);
2580 while ((seg = next_segment(len, offset)) != 0) {
2581 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
2591 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
2593 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
2594 void *data, int offset, unsigned long len)
2599 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
2600 if (kvm_is_error_hva(addr))
2602 pagefault_disable();
2603 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
2610 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
2611 void *data, unsigned long len)
2613 gfn_t gfn = gpa >> PAGE_SHIFT;
2614 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2615 int offset = offset_in_page(gpa);
2617 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
2619 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
2621 static int __kvm_write_guest_page(struct kvm *kvm,
2622 struct kvm_memory_slot *memslot, gfn_t gfn,
2623 const void *data, int offset, int len)
2628 addr = gfn_to_hva_memslot(memslot, gfn);
2629 if (kvm_is_error_hva(addr))
2631 r = __copy_to_user((void __user *)addr + offset, data, len);
2634 mark_page_dirty_in_slot(kvm, memslot, gfn);
2638 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
2639 const void *data, int offset, int len)
2641 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2643 return __kvm_write_guest_page(kvm, slot, gfn, data, offset, len);
2645 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
2647 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
2648 const void *data, int offset, int len)
2650 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2652 return __kvm_write_guest_page(vcpu->kvm, slot, gfn, data, offset, len);
2654 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
2656 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
2659 gfn_t gfn = gpa >> PAGE_SHIFT;
2661 int offset = offset_in_page(gpa);
2664 while ((seg = next_segment(len, offset)) != 0) {
2665 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
2675 EXPORT_SYMBOL_GPL(kvm_write_guest);
2677 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
2680 gfn_t gfn = gpa >> PAGE_SHIFT;
2682 int offset = offset_in_page(gpa);
2685 while ((seg = next_segment(len, offset)) != 0) {
2686 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
2696 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
2698 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
2699 struct gfn_to_hva_cache *ghc,
2700 gpa_t gpa, unsigned long len)
2702 int offset = offset_in_page(gpa);
2703 gfn_t start_gfn = gpa >> PAGE_SHIFT;
2704 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
2705 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
2706 gfn_t nr_pages_avail;
2708 /* Update ghc->generation before performing any error checks. */
2709 ghc->generation = slots->generation;
2711 if (start_gfn > end_gfn) {
2712 ghc->hva = KVM_HVA_ERR_BAD;
2717 * If the requested region crosses two memslots, we still
2718 * verify that the entire region is valid here.
2720 for ( ; start_gfn <= end_gfn; start_gfn += nr_pages_avail) {
2721 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
2722 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
2724 if (kvm_is_error_hva(ghc->hva))
2728 /* Use the slow path for cross page reads and writes. */
2729 if (nr_pages_needed == 1)
2732 ghc->memslot = NULL;
2739 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2740 gpa_t gpa, unsigned long len)
2742 struct kvm_memslots *slots = kvm_memslots(kvm);
2743 return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len);
2745 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
2747 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2748 void *data, unsigned int offset,
2751 struct kvm_memslots *slots = kvm_memslots(kvm);
2753 gpa_t gpa = ghc->gpa + offset;
2755 BUG_ON(len + offset > ghc->len);
2757 if (slots->generation != ghc->generation) {
2758 if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len))
2762 if (kvm_is_error_hva(ghc->hva))
2765 if (unlikely(!ghc->memslot))
2766 return kvm_write_guest(kvm, gpa, data, len);
2768 r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
2771 mark_page_dirty_in_slot(kvm, ghc->memslot, gpa >> PAGE_SHIFT);
2775 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached);
2777 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2778 void *data, unsigned long len)
2780 return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len);
2782 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
2784 int kvm_read_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2785 void *data, unsigned int offset,
2788 struct kvm_memslots *slots = kvm_memslots(kvm);
2790 gpa_t gpa = ghc->gpa + offset;
2792 BUG_ON(len + offset > ghc->len);
2794 if (slots->generation != ghc->generation) {
2795 if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len))
2799 if (kvm_is_error_hva(ghc->hva))
2802 if (unlikely(!ghc->memslot))
2803 return kvm_read_guest(kvm, gpa, data, len);
2805 r = __copy_from_user(data, (void __user *)ghc->hva + offset, len);
2811 EXPORT_SYMBOL_GPL(kvm_read_guest_offset_cached);
2813 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2814 void *data, unsigned long len)
2816 return kvm_read_guest_offset_cached(kvm, ghc, data, 0, len);
2818 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
2820 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
2822 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
2823 gfn_t gfn = gpa >> PAGE_SHIFT;
2825 int offset = offset_in_page(gpa);
2828 while ((seg = next_segment(len, offset)) != 0) {
2829 ret = kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
2838 EXPORT_SYMBOL_GPL(kvm_clear_guest);
2840 void mark_page_dirty_in_slot(struct kvm *kvm,
2841 struct kvm_memory_slot *memslot,
2844 if (memslot && kvm_slot_dirty_track_enabled(memslot)) {
2845 unsigned long rel_gfn = gfn - memslot->base_gfn;
2846 u32 slot = (memslot->as_id << 16) | memslot->id;
2848 if (kvm->dirty_ring_size)
2849 kvm_dirty_ring_push(kvm_dirty_ring_get(kvm),
2852 set_bit_le(rel_gfn, memslot->dirty_bitmap);
2855 EXPORT_SYMBOL_GPL(mark_page_dirty_in_slot);
2857 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
2859 struct kvm_memory_slot *memslot;
2861 memslot = gfn_to_memslot(kvm, gfn);
2862 mark_page_dirty_in_slot(kvm, memslot, gfn);
2864 EXPORT_SYMBOL_GPL(mark_page_dirty);
2866 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
2868 struct kvm_memory_slot *memslot;
2870 memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2871 mark_page_dirty_in_slot(vcpu->kvm, memslot, gfn);
2873 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
2875 void kvm_sigset_activate(struct kvm_vcpu *vcpu)
2877 if (!vcpu->sigset_active)
2881 * This does a lockless modification of ->real_blocked, which is fine
2882 * because, only current can change ->real_blocked and all readers of
2883 * ->real_blocked don't care as long ->real_blocked is always a subset
2886 sigprocmask(SIG_SETMASK, &vcpu->sigset, ¤t->real_blocked);
2889 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu)
2891 if (!vcpu->sigset_active)
2894 sigprocmask(SIG_SETMASK, ¤t->real_blocked, NULL);
2895 sigemptyset(¤t->real_blocked);
2898 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
2900 unsigned int old, val, grow, grow_start;
2902 old = val = vcpu->halt_poll_ns;
2903 grow_start = READ_ONCE(halt_poll_ns_grow_start);
2904 grow = READ_ONCE(halt_poll_ns_grow);
2909 if (val < grow_start)
2912 if (val > vcpu->kvm->max_halt_poll_ns)
2913 val = vcpu->kvm->max_halt_poll_ns;
2915 vcpu->halt_poll_ns = val;
2917 trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
2920 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
2922 unsigned int old, val, shrink;
2924 old = val = vcpu->halt_poll_ns;
2925 shrink = READ_ONCE(halt_poll_ns_shrink);
2931 vcpu->halt_poll_ns = val;
2932 trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
2935 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
2938 int idx = srcu_read_lock(&vcpu->kvm->srcu);
2940 if (kvm_arch_vcpu_runnable(vcpu)) {
2941 kvm_make_request(KVM_REQ_UNHALT, vcpu);
2944 if (kvm_cpu_has_pending_timer(vcpu))
2946 if (signal_pending(current))
2948 if (kvm_check_request(KVM_REQ_UNBLOCK, vcpu))
2953 srcu_read_unlock(&vcpu->kvm->srcu, idx);
2958 update_halt_poll_stats(struct kvm_vcpu *vcpu, u64 poll_ns, bool waited)
2961 vcpu->stat.halt_poll_fail_ns += poll_ns;
2963 vcpu->stat.halt_poll_success_ns += poll_ns;
2967 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2969 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
2971 ktime_t start, cur, poll_end;
2972 bool waited = false;
2975 kvm_arch_vcpu_blocking(vcpu);
2977 start = cur = poll_end = ktime_get();
2978 if (vcpu->halt_poll_ns && !kvm_arch_no_poll(vcpu)) {
2979 ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns);
2981 ++vcpu->stat.halt_attempted_poll;
2984 * This sets KVM_REQ_UNHALT if an interrupt
2987 if (kvm_vcpu_check_block(vcpu) < 0) {
2988 ++vcpu->stat.halt_successful_poll;
2989 if (!vcpu_valid_wakeup(vcpu))
2990 ++vcpu->stat.halt_poll_invalid;
2993 poll_end = cur = ktime_get();
2994 } while (kvm_vcpu_can_poll(cur, stop));
2997 prepare_to_rcuwait(&vcpu->wait);
2999 set_current_state(TASK_INTERRUPTIBLE);
3001 if (kvm_vcpu_check_block(vcpu) < 0)
3007 finish_rcuwait(&vcpu->wait);
3010 kvm_arch_vcpu_unblocking(vcpu);
3011 block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
3013 update_halt_poll_stats(
3014 vcpu, ktime_to_ns(ktime_sub(poll_end, start)), waited);
3016 if (!kvm_arch_no_poll(vcpu)) {
3017 if (!vcpu_valid_wakeup(vcpu)) {
3018 shrink_halt_poll_ns(vcpu);
3019 } else if (vcpu->kvm->max_halt_poll_ns) {
3020 if (block_ns <= vcpu->halt_poll_ns)
3022 /* we had a long block, shrink polling */
3023 else if (vcpu->halt_poll_ns &&
3024 block_ns > vcpu->kvm->max_halt_poll_ns)
3025 shrink_halt_poll_ns(vcpu);
3026 /* we had a short halt and our poll time is too small */
3027 else if (vcpu->halt_poll_ns < vcpu->kvm->max_halt_poll_ns &&
3028 block_ns < vcpu->kvm->max_halt_poll_ns)
3029 grow_halt_poll_ns(vcpu);
3031 vcpu->halt_poll_ns = 0;
3035 trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu));
3036 kvm_arch_vcpu_block_finish(vcpu);
3038 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
3040 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
3042 struct rcuwait *waitp;
3044 waitp = kvm_arch_vcpu_get_wait(vcpu);
3045 if (rcuwait_wake_up(waitp)) {
3046 WRITE_ONCE(vcpu->ready, true);
3047 ++vcpu->stat.halt_wakeup;
3053 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
3057 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
3059 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
3062 int cpu = vcpu->cpu;
3064 if (kvm_vcpu_wake_up(vcpu))
3068 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
3069 if (kvm_arch_vcpu_should_kick(vcpu))
3070 smp_send_reschedule(cpu);
3073 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
3074 #endif /* !CONFIG_S390 */
3076 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
3079 struct task_struct *task = NULL;
3083 pid = rcu_dereference(target->pid);
3085 task = get_pid_task(pid, PIDTYPE_PID);
3089 ret = yield_to(task, 1);
3090 put_task_struct(task);
3094 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
3097 * Helper that checks whether a VCPU is eligible for directed yield.
3098 * Most eligible candidate to yield is decided by following heuristics:
3100 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
3101 * (preempted lock holder), indicated by @in_spin_loop.
3102 * Set at the beginning and cleared at the end of interception/PLE handler.
3104 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
3105 * chance last time (mostly it has become eligible now since we have probably
3106 * yielded to lockholder in last iteration. This is done by toggling
3107 * @dy_eligible each time a VCPU checked for eligibility.)
3109 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
3110 * to preempted lock-holder could result in wrong VCPU selection and CPU
3111 * burning. Giving priority for a potential lock-holder increases lock
3114 * Since algorithm is based on heuristics, accessing another VCPU data without
3115 * locking does not harm. It may result in trying to yield to same VCPU, fail
3116 * and continue with next VCPU and so on.
3118 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
3120 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
3123 eligible = !vcpu->spin_loop.in_spin_loop ||
3124 vcpu->spin_loop.dy_eligible;
3126 if (vcpu->spin_loop.in_spin_loop)
3127 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
3136 * Unlike kvm_arch_vcpu_runnable, this function is called outside
3137 * a vcpu_load/vcpu_put pair. However, for most architectures
3138 * kvm_arch_vcpu_runnable does not require vcpu_load.
3140 bool __weak kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
3142 return kvm_arch_vcpu_runnable(vcpu);
3145 static bool vcpu_dy_runnable(struct kvm_vcpu *vcpu)
3147 if (kvm_arch_dy_runnable(vcpu))
3150 #ifdef CONFIG_KVM_ASYNC_PF
3151 if (!list_empty_careful(&vcpu->async_pf.done))
3158 bool __weak kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu)
3163 void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode)
3165 struct kvm *kvm = me->kvm;
3166 struct kvm_vcpu *vcpu;
3167 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
3173 kvm_vcpu_set_in_spin_loop(me, true);
3175 * We boost the priority of a VCPU that is runnable but not
3176 * currently running, because it got preempted by something
3177 * else and called schedule in __vcpu_run. Hopefully that
3178 * VCPU is holding the lock that we need and will release it.
3179 * We approximate round-robin by starting at the last boosted VCPU.
3181 for (pass = 0; pass < 2 && !yielded && try; pass++) {
3182 kvm_for_each_vcpu(i, vcpu, kvm) {
3183 if (!pass && i <= last_boosted_vcpu) {
3184 i = last_boosted_vcpu;
3186 } else if (pass && i > last_boosted_vcpu)
3188 if (!READ_ONCE(vcpu->ready))
3192 if (rcuwait_active(&vcpu->wait) &&
3193 !vcpu_dy_runnable(vcpu))
3195 if (READ_ONCE(vcpu->preempted) && yield_to_kernel_mode &&
3196 !kvm_arch_dy_has_pending_interrupt(vcpu) &&
3197 !kvm_arch_vcpu_in_kernel(vcpu))
3199 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
3202 yielded = kvm_vcpu_yield_to(vcpu);
3204 kvm->last_boosted_vcpu = i;
3206 } else if (yielded < 0) {
3213 kvm_vcpu_set_in_spin_loop(me, false);
3215 /* Ensure vcpu is not eligible during next spinloop */
3216 kvm_vcpu_set_dy_eligible(me, false);
3218 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
3220 static bool kvm_page_in_dirty_ring(struct kvm *kvm, unsigned long pgoff)
3222 #if KVM_DIRTY_LOG_PAGE_OFFSET > 0
3223 return (pgoff >= KVM_DIRTY_LOG_PAGE_OFFSET) &&
3224 (pgoff < KVM_DIRTY_LOG_PAGE_OFFSET +
3225 kvm->dirty_ring_size / PAGE_SIZE);
3231 static vm_fault_t kvm_vcpu_fault(struct vm_fault *vmf)
3233 struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data;
3236 if (vmf->pgoff == 0)
3237 page = virt_to_page(vcpu->run);
3239 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
3240 page = virt_to_page(vcpu->arch.pio_data);
3242 #ifdef CONFIG_KVM_MMIO
3243 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
3244 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
3246 else if (kvm_page_in_dirty_ring(vcpu->kvm, vmf->pgoff))
3247 page = kvm_dirty_ring_get_page(
3249 vmf->pgoff - KVM_DIRTY_LOG_PAGE_OFFSET);
3251 return kvm_arch_vcpu_fault(vcpu, vmf);
3257 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
3258 .fault = kvm_vcpu_fault,
3261 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
3263 struct kvm_vcpu *vcpu = file->private_data;
3264 unsigned long pages = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
3266 if ((kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff) ||
3267 kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff + pages - 1)) &&
3268 ((vma->vm_flags & VM_EXEC) || !(vma->vm_flags & VM_SHARED)))
3271 vma->vm_ops = &kvm_vcpu_vm_ops;
3275 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
3277 struct kvm_vcpu *vcpu = filp->private_data;
3279 kvm_put_kvm(vcpu->kvm);
3283 static struct file_operations kvm_vcpu_fops = {
3284 .release = kvm_vcpu_release,
3285 .unlocked_ioctl = kvm_vcpu_ioctl,
3286 .mmap = kvm_vcpu_mmap,
3287 .llseek = noop_llseek,
3288 KVM_COMPAT(kvm_vcpu_compat_ioctl),
3292 * Allocates an inode for the vcpu.
3294 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
3296 char name[8 + 1 + ITOA_MAX_LEN + 1];
3298 snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id);
3299 return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
3302 static void kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
3304 #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS
3305 struct dentry *debugfs_dentry;
3306 char dir_name[ITOA_MAX_LEN * 2];
3308 if (!debugfs_initialized())
3311 snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
3312 debugfs_dentry = debugfs_create_dir(dir_name,
3313 vcpu->kvm->debugfs_dentry);
3315 kvm_arch_create_vcpu_debugfs(vcpu, debugfs_dentry);
3320 * Creates some virtual cpus. Good luck creating more than one.
3322 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
3325 struct kvm_vcpu *vcpu;
3328 if (id >= KVM_MAX_VCPU_ID)
3331 mutex_lock(&kvm->lock);
3332 if (kvm->created_vcpus == KVM_MAX_VCPUS) {
3333 mutex_unlock(&kvm->lock);
3337 kvm->created_vcpus++;
3338 mutex_unlock(&kvm->lock);
3340 r = kvm_arch_vcpu_precreate(kvm, id);
3342 goto vcpu_decrement;
3344 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL_ACCOUNT);
3347 goto vcpu_decrement;
3350 BUILD_BUG_ON(sizeof(struct kvm_run) > PAGE_SIZE);
3351 page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
3356 vcpu->run = page_address(page);
3358 kvm_vcpu_init(vcpu, kvm, id);
3360 r = kvm_arch_vcpu_create(vcpu);
3362 goto vcpu_free_run_page;
3364 if (kvm->dirty_ring_size) {
3365 r = kvm_dirty_ring_alloc(&vcpu->dirty_ring,
3366 id, kvm->dirty_ring_size);
3368 goto arch_vcpu_destroy;
3371 mutex_lock(&kvm->lock);
3372 if (kvm_get_vcpu_by_id(kvm, id)) {
3374 goto unlock_vcpu_destroy;
3377 vcpu->vcpu_idx = atomic_read(&kvm->online_vcpus);
3378 BUG_ON(kvm->vcpus[vcpu->vcpu_idx]);
3380 /* Now it's all set up, let userspace reach it */
3382 r = create_vcpu_fd(vcpu);
3384 kvm_put_kvm_no_destroy(kvm);
3385 goto unlock_vcpu_destroy;
3388 kvm->vcpus[vcpu->vcpu_idx] = vcpu;
3391 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
3392 * before kvm->online_vcpu's incremented value.
3395 atomic_inc(&kvm->online_vcpus);
3397 mutex_unlock(&kvm->lock);
3398 kvm_arch_vcpu_postcreate(vcpu);
3399 kvm_create_vcpu_debugfs(vcpu);
3402 unlock_vcpu_destroy:
3403 mutex_unlock(&kvm->lock);
3404 kvm_dirty_ring_free(&vcpu->dirty_ring);
3406 kvm_arch_vcpu_destroy(vcpu);
3408 free_page((unsigned long)vcpu->run);
3410 kmem_cache_free(kvm_vcpu_cache, vcpu);
3412 mutex_lock(&kvm->lock);
3413 kvm->created_vcpus--;
3414 mutex_unlock(&kvm->lock);
3418 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
3421 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
3422 vcpu->sigset_active = 1;
3423 vcpu->sigset = *sigset;
3425 vcpu->sigset_active = 0;
3429 static long kvm_vcpu_ioctl(struct file *filp,
3430 unsigned int ioctl, unsigned long arg)
3432 struct kvm_vcpu *vcpu = filp->private_data;
3433 void __user *argp = (void __user *)arg;
3435 struct kvm_fpu *fpu = NULL;
3436 struct kvm_sregs *kvm_sregs = NULL;
3438 if (vcpu->kvm->mm != current->mm)
3441 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
3445 * Some architectures have vcpu ioctls that are asynchronous to vcpu
3446 * execution; mutex_lock() would break them.
3448 r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg);
3449 if (r != -ENOIOCTLCMD)
3452 if (mutex_lock_killable(&vcpu->mutex))
3460 oldpid = rcu_access_pointer(vcpu->pid);
3461 if (unlikely(oldpid != task_pid(current))) {
3462 /* The thread running this VCPU changed. */
3465 r = kvm_arch_vcpu_run_pid_change(vcpu);
3469 newpid = get_task_pid(current, PIDTYPE_PID);
3470 rcu_assign_pointer(vcpu->pid, newpid);
3475 r = kvm_arch_vcpu_ioctl_run(vcpu);
3476 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
3479 case KVM_GET_REGS: {
3480 struct kvm_regs *kvm_regs;
3483 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL_ACCOUNT);
3486 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
3490 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
3497 case KVM_SET_REGS: {
3498 struct kvm_regs *kvm_regs;
3500 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
3501 if (IS_ERR(kvm_regs)) {
3502 r = PTR_ERR(kvm_regs);
3505 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
3509 case KVM_GET_SREGS: {
3510 kvm_sregs = kzalloc(sizeof(struct kvm_sregs),
3511 GFP_KERNEL_ACCOUNT);
3515 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
3519 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
3524 case KVM_SET_SREGS: {
3525 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
3526 if (IS_ERR(kvm_sregs)) {
3527 r = PTR_ERR(kvm_sregs);
3531 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
3534 case KVM_GET_MP_STATE: {
3535 struct kvm_mp_state mp_state;
3537 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
3541 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
3546 case KVM_SET_MP_STATE: {
3547 struct kvm_mp_state mp_state;
3550 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
3552 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
3555 case KVM_TRANSLATE: {
3556 struct kvm_translation tr;
3559 if (copy_from_user(&tr, argp, sizeof(tr)))
3561 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
3565 if (copy_to_user(argp, &tr, sizeof(tr)))
3570 case KVM_SET_GUEST_DEBUG: {
3571 struct kvm_guest_debug dbg;
3574 if (copy_from_user(&dbg, argp, sizeof(dbg)))
3576 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
3579 case KVM_SET_SIGNAL_MASK: {
3580 struct kvm_signal_mask __user *sigmask_arg = argp;
3581 struct kvm_signal_mask kvm_sigmask;
3582 sigset_t sigset, *p;
3587 if (copy_from_user(&kvm_sigmask, argp,
3588 sizeof(kvm_sigmask)))
3591 if (kvm_sigmask.len != sizeof(sigset))
3594 if (copy_from_user(&sigset, sigmask_arg->sigset,
3599 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
3603 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL_ACCOUNT);
3607 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
3611 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
3617 fpu = memdup_user(argp, sizeof(*fpu));
3623 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
3627 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
3630 mutex_unlock(&vcpu->mutex);
3636 #ifdef CONFIG_KVM_COMPAT
3637 static long kvm_vcpu_compat_ioctl(struct file *filp,
3638 unsigned int ioctl, unsigned long arg)
3640 struct kvm_vcpu *vcpu = filp->private_data;
3641 void __user *argp = compat_ptr(arg);
3644 if (vcpu->kvm->mm != current->mm)
3648 case KVM_SET_SIGNAL_MASK: {
3649 struct kvm_signal_mask __user *sigmask_arg = argp;
3650 struct kvm_signal_mask kvm_sigmask;
3655 if (copy_from_user(&kvm_sigmask, argp,
3656 sizeof(kvm_sigmask)))
3659 if (kvm_sigmask.len != sizeof(compat_sigset_t))
3662 if (get_compat_sigset(&sigset,
3663 (compat_sigset_t __user *)sigmask_arg->sigset))
3665 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
3667 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
3671 r = kvm_vcpu_ioctl(filp, ioctl, arg);
3679 static int kvm_device_mmap(struct file *filp, struct vm_area_struct *vma)
3681 struct kvm_device *dev = filp->private_data;
3684 return dev->ops->mmap(dev, vma);
3689 static int kvm_device_ioctl_attr(struct kvm_device *dev,
3690 int (*accessor)(struct kvm_device *dev,
3691 struct kvm_device_attr *attr),
3694 struct kvm_device_attr attr;
3699 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
3702 return accessor(dev, &attr);
3705 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
3708 struct kvm_device *dev = filp->private_data;
3710 if (dev->kvm->mm != current->mm)
3714 case KVM_SET_DEVICE_ATTR:
3715 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
3716 case KVM_GET_DEVICE_ATTR:
3717 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
3718 case KVM_HAS_DEVICE_ATTR:
3719 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
3721 if (dev->ops->ioctl)
3722 return dev->ops->ioctl(dev, ioctl, arg);
3728 static int kvm_device_release(struct inode *inode, struct file *filp)
3730 struct kvm_device *dev = filp->private_data;
3731 struct kvm *kvm = dev->kvm;
3733 if (dev->ops->release) {
3734 mutex_lock(&kvm->lock);
3735 list_del(&dev->vm_node);
3736 dev->ops->release(dev);
3737 mutex_unlock(&kvm->lock);
3744 static const struct file_operations kvm_device_fops = {
3745 .unlocked_ioctl = kvm_device_ioctl,
3746 .release = kvm_device_release,
3747 KVM_COMPAT(kvm_device_ioctl),
3748 .mmap = kvm_device_mmap,
3751 struct kvm_device *kvm_device_from_filp(struct file *filp)
3753 if (filp->f_op != &kvm_device_fops)
3756 return filp->private_data;
3759 static const struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
3760 #ifdef CONFIG_KVM_MPIC
3761 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
3762 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
3766 int kvm_register_device_ops(const struct kvm_device_ops *ops, u32 type)
3768 if (type >= ARRAY_SIZE(kvm_device_ops_table))
3771 if (kvm_device_ops_table[type] != NULL)
3774 kvm_device_ops_table[type] = ops;
3778 void kvm_unregister_device_ops(u32 type)
3780 if (kvm_device_ops_table[type] != NULL)
3781 kvm_device_ops_table[type] = NULL;
3784 static int kvm_ioctl_create_device(struct kvm *kvm,
3785 struct kvm_create_device *cd)
3787 const struct kvm_device_ops *ops = NULL;
3788 struct kvm_device *dev;
3789 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
3793 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
3796 type = array_index_nospec(cd->type, ARRAY_SIZE(kvm_device_ops_table));
3797 ops = kvm_device_ops_table[type];
3804 dev = kzalloc(sizeof(*dev), GFP_KERNEL_ACCOUNT);
3811 mutex_lock(&kvm->lock);
3812 ret = ops->create(dev, type);
3814 mutex_unlock(&kvm->lock);
3818 list_add(&dev->vm_node, &kvm->devices);
3819 mutex_unlock(&kvm->lock);
3825 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
3827 kvm_put_kvm_no_destroy(kvm);
3828 mutex_lock(&kvm->lock);
3829 list_del(&dev->vm_node);
3830 mutex_unlock(&kvm->lock);
3839 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
3842 case KVM_CAP_USER_MEMORY:
3843 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
3844 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
3845 case KVM_CAP_INTERNAL_ERROR_DATA:
3846 #ifdef CONFIG_HAVE_KVM_MSI
3847 case KVM_CAP_SIGNAL_MSI:
3849 #ifdef CONFIG_HAVE_KVM_IRQFD
3851 case KVM_CAP_IRQFD_RESAMPLE:
3853 case KVM_CAP_IOEVENTFD_ANY_LENGTH:
3854 case KVM_CAP_CHECK_EXTENSION_VM:
3855 case KVM_CAP_ENABLE_CAP_VM:
3856 case KVM_CAP_HALT_POLL:
3858 #ifdef CONFIG_KVM_MMIO
3859 case KVM_CAP_COALESCED_MMIO:
3860 return KVM_COALESCED_MMIO_PAGE_OFFSET;
3861 case KVM_CAP_COALESCED_PIO:
3864 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3865 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2:
3866 return KVM_DIRTY_LOG_MANUAL_CAPS;
3868 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3869 case KVM_CAP_IRQ_ROUTING:
3870 return KVM_MAX_IRQ_ROUTES;
3872 #if KVM_ADDRESS_SPACE_NUM > 1
3873 case KVM_CAP_MULTI_ADDRESS_SPACE:
3874 return KVM_ADDRESS_SPACE_NUM;
3876 case KVM_CAP_NR_MEMSLOTS:
3877 return KVM_USER_MEM_SLOTS;
3878 case KVM_CAP_DIRTY_LOG_RING:
3879 #if KVM_DIRTY_LOG_PAGE_OFFSET > 0
3880 return KVM_DIRTY_RING_MAX_ENTRIES * sizeof(struct kvm_dirty_gfn);
3887 return kvm_vm_ioctl_check_extension(kvm, arg);
3890 static int kvm_vm_ioctl_enable_dirty_log_ring(struct kvm *kvm, u32 size)
3894 if (!KVM_DIRTY_LOG_PAGE_OFFSET)
3897 /* the size should be power of 2 */
3898 if (!size || (size & (size - 1)))
3901 /* Should be bigger to keep the reserved entries, or a page */
3902 if (size < kvm_dirty_ring_get_rsvd_entries() *
3903 sizeof(struct kvm_dirty_gfn) || size < PAGE_SIZE)
3906 if (size > KVM_DIRTY_RING_MAX_ENTRIES *
3907 sizeof(struct kvm_dirty_gfn))
3910 /* We only allow it to set once */
3911 if (kvm->dirty_ring_size)
3914 mutex_lock(&kvm->lock);
3916 if (kvm->created_vcpus) {
3917 /* We don't allow to change this value after vcpu created */
3920 kvm->dirty_ring_size = size;
3924 mutex_unlock(&kvm->lock);
3928 static int kvm_vm_ioctl_reset_dirty_pages(struct kvm *kvm)
3931 struct kvm_vcpu *vcpu;
3934 if (!kvm->dirty_ring_size)
3937 mutex_lock(&kvm->slots_lock);
3939 kvm_for_each_vcpu(i, vcpu, kvm)
3940 cleared += kvm_dirty_ring_reset(vcpu->kvm, &vcpu->dirty_ring);
3942 mutex_unlock(&kvm->slots_lock);
3945 kvm_flush_remote_tlbs(kvm);
3950 int __attribute__((weak)) kvm_vm_ioctl_enable_cap(struct kvm *kvm,
3951 struct kvm_enable_cap *cap)
3956 static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm,
3957 struct kvm_enable_cap *cap)
3960 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3961 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2: {
3962 u64 allowed_options = KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE;
3964 if (cap->args[0] & KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE)
3965 allowed_options = KVM_DIRTY_LOG_MANUAL_CAPS;
3967 if (cap->flags || (cap->args[0] & ~allowed_options))
3969 kvm->manual_dirty_log_protect = cap->args[0];
3973 case KVM_CAP_HALT_POLL: {
3974 if (cap->flags || cap->args[0] != (unsigned int)cap->args[0])
3977 kvm->max_halt_poll_ns = cap->args[0];
3980 case KVM_CAP_DIRTY_LOG_RING:
3981 return kvm_vm_ioctl_enable_dirty_log_ring(kvm, cap->args[0]);
3983 return kvm_vm_ioctl_enable_cap(kvm, cap);
3987 static long kvm_vm_ioctl(struct file *filp,
3988 unsigned int ioctl, unsigned long arg)
3990 struct kvm *kvm = filp->private_data;
3991 void __user *argp = (void __user *)arg;
3994 if (kvm->mm != current->mm)
3997 case KVM_CREATE_VCPU:
3998 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
4000 case KVM_ENABLE_CAP: {
4001 struct kvm_enable_cap cap;
4004 if (copy_from_user(&cap, argp, sizeof(cap)))
4006 r = kvm_vm_ioctl_enable_cap_generic(kvm, &cap);
4009 case KVM_SET_USER_MEMORY_REGION: {
4010 struct kvm_userspace_memory_region kvm_userspace_mem;
4013 if (copy_from_user(&kvm_userspace_mem, argp,
4014 sizeof(kvm_userspace_mem)))
4017 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
4020 case KVM_GET_DIRTY_LOG: {
4021 struct kvm_dirty_log log;
4024 if (copy_from_user(&log, argp, sizeof(log)))
4026 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
4029 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4030 case KVM_CLEAR_DIRTY_LOG: {
4031 struct kvm_clear_dirty_log log;
4034 if (copy_from_user(&log, argp, sizeof(log)))
4036 r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
4040 #ifdef CONFIG_KVM_MMIO
4041 case KVM_REGISTER_COALESCED_MMIO: {
4042 struct kvm_coalesced_mmio_zone zone;
4045 if (copy_from_user(&zone, argp, sizeof(zone)))
4047 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
4050 case KVM_UNREGISTER_COALESCED_MMIO: {
4051 struct kvm_coalesced_mmio_zone zone;
4054 if (copy_from_user(&zone, argp, sizeof(zone)))
4056 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
4061 struct kvm_irqfd data;
4064 if (copy_from_user(&data, argp, sizeof(data)))
4066 r = kvm_irqfd(kvm, &data);
4069 case KVM_IOEVENTFD: {
4070 struct kvm_ioeventfd data;
4073 if (copy_from_user(&data, argp, sizeof(data)))
4075 r = kvm_ioeventfd(kvm, &data);
4078 #ifdef CONFIG_HAVE_KVM_MSI
4079 case KVM_SIGNAL_MSI: {
4083 if (copy_from_user(&msi, argp, sizeof(msi)))
4085 r = kvm_send_userspace_msi(kvm, &msi);
4089 #ifdef __KVM_HAVE_IRQ_LINE
4090 case KVM_IRQ_LINE_STATUS:
4091 case KVM_IRQ_LINE: {
4092 struct kvm_irq_level irq_event;
4095 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
4098 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
4099 ioctl == KVM_IRQ_LINE_STATUS);
4104 if (ioctl == KVM_IRQ_LINE_STATUS) {
4105 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
4113 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
4114 case KVM_SET_GSI_ROUTING: {
4115 struct kvm_irq_routing routing;
4116 struct kvm_irq_routing __user *urouting;
4117 struct kvm_irq_routing_entry *entries = NULL;
4120 if (copy_from_user(&routing, argp, sizeof(routing)))
4123 if (!kvm_arch_can_set_irq_routing(kvm))
4125 if (routing.nr > KVM_MAX_IRQ_ROUTES)
4131 entries = vmemdup_user(urouting->entries,
4132 array_size(sizeof(*entries),
4134 if (IS_ERR(entries)) {
4135 r = PTR_ERR(entries);
4139 r = kvm_set_irq_routing(kvm, entries, routing.nr,
4144 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
4145 case KVM_CREATE_DEVICE: {
4146 struct kvm_create_device cd;
4149 if (copy_from_user(&cd, argp, sizeof(cd)))
4152 r = kvm_ioctl_create_device(kvm, &cd);
4157 if (copy_to_user(argp, &cd, sizeof(cd)))
4163 case KVM_CHECK_EXTENSION:
4164 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
4166 case KVM_RESET_DIRTY_RINGS:
4167 r = kvm_vm_ioctl_reset_dirty_pages(kvm);
4170 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
4176 #ifdef CONFIG_KVM_COMPAT
4177 struct compat_kvm_dirty_log {
4181 compat_uptr_t dirty_bitmap; /* one bit per page */
4186 static long kvm_vm_compat_ioctl(struct file *filp,
4187 unsigned int ioctl, unsigned long arg)
4189 struct kvm *kvm = filp->private_data;
4192 if (kvm->mm != current->mm)
4195 case KVM_GET_DIRTY_LOG: {
4196 struct compat_kvm_dirty_log compat_log;
4197 struct kvm_dirty_log log;
4199 if (copy_from_user(&compat_log, (void __user *)arg,
4200 sizeof(compat_log)))
4202 log.slot = compat_log.slot;
4203 log.padding1 = compat_log.padding1;
4204 log.padding2 = compat_log.padding2;
4205 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
4207 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
4211 r = kvm_vm_ioctl(filp, ioctl, arg);
4217 static struct file_operations kvm_vm_fops = {
4218 .release = kvm_vm_release,
4219 .unlocked_ioctl = kvm_vm_ioctl,
4220 .llseek = noop_llseek,
4221 KVM_COMPAT(kvm_vm_compat_ioctl),
4224 bool file_is_kvm(struct file *file)
4226 return file && file->f_op == &kvm_vm_fops;
4228 EXPORT_SYMBOL_GPL(file_is_kvm);
4230 static int kvm_dev_ioctl_create_vm(unsigned long type)
4236 kvm = kvm_create_vm(type);
4238 return PTR_ERR(kvm);
4239 #ifdef CONFIG_KVM_MMIO
4240 r = kvm_coalesced_mmio_init(kvm);
4244 r = get_unused_fd_flags(O_CLOEXEC);
4248 file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
4256 * Don't call kvm_put_kvm anymore at this point; file->f_op is
4257 * already set, with ->release() being kvm_vm_release(). In error
4258 * cases it will be called by the final fput(file) and will take
4259 * care of doing kvm_put_kvm(kvm).
4261 if (kvm_create_vm_debugfs(kvm, r) < 0) {
4266 kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm);
4268 fd_install(r, file);
4276 static long kvm_dev_ioctl(struct file *filp,
4277 unsigned int ioctl, unsigned long arg)
4282 case KVM_GET_API_VERSION:
4285 r = KVM_API_VERSION;
4288 r = kvm_dev_ioctl_create_vm(arg);
4290 case KVM_CHECK_EXTENSION:
4291 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
4293 case KVM_GET_VCPU_MMAP_SIZE:
4296 r = PAGE_SIZE; /* struct kvm_run */
4298 r += PAGE_SIZE; /* pio data page */
4300 #ifdef CONFIG_KVM_MMIO
4301 r += PAGE_SIZE; /* coalesced mmio ring page */
4304 case KVM_TRACE_ENABLE:
4305 case KVM_TRACE_PAUSE:
4306 case KVM_TRACE_DISABLE:
4310 return kvm_arch_dev_ioctl(filp, ioctl, arg);
4316 static struct file_operations kvm_chardev_ops = {
4317 .unlocked_ioctl = kvm_dev_ioctl,
4318 .llseek = noop_llseek,
4319 KVM_COMPAT(kvm_dev_ioctl),
4322 static struct miscdevice kvm_dev = {
4328 static void hardware_enable_nolock(void *junk)
4330 int cpu = raw_smp_processor_id();
4333 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
4336 cpumask_set_cpu(cpu, cpus_hardware_enabled);
4338 r = kvm_arch_hardware_enable();
4341 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
4342 atomic_inc(&hardware_enable_failed);
4343 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
4347 static int kvm_starting_cpu(unsigned int cpu)
4349 raw_spin_lock(&kvm_count_lock);
4350 if (kvm_usage_count)
4351 hardware_enable_nolock(NULL);
4352 raw_spin_unlock(&kvm_count_lock);
4356 static void hardware_disable_nolock(void *junk)
4358 int cpu = raw_smp_processor_id();
4360 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
4362 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
4363 kvm_arch_hardware_disable();
4366 static int kvm_dying_cpu(unsigned int cpu)
4368 raw_spin_lock(&kvm_count_lock);
4369 if (kvm_usage_count)
4370 hardware_disable_nolock(NULL);
4371 raw_spin_unlock(&kvm_count_lock);
4375 static void hardware_disable_all_nolock(void)
4377 BUG_ON(!kvm_usage_count);
4380 if (!kvm_usage_count)
4381 on_each_cpu(hardware_disable_nolock, NULL, 1);
4384 static void hardware_disable_all(void)
4386 raw_spin_lock(&kvm_count_lock);
4387 hardware_disable_all_nolock();
4388 raw_spin_unlock(&kvm_count_lock);
4391 static int hardware_enable_all(void)
4395 raw_spin_lock(&kvm_count_lock);
4398 if (kvm_usage_count == 1) {
4399 atomic_set(&hardware_enable_failed, 0);
4400 on_each_cpu(hardware_enable_nolock, NULL, 1);
4402 if (atomic_read(&hardware_enable_failed)) {
4403 hardware_disable_all_nolock();
4408 raw_spin_unlock(&kvm_count_lock);
4413 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
4417 * Some (well, at least mine) BIOSes hang on reboot if
4420 * And Intel TXT required VMX off for all cpu when system shutdown.
4422 pr_info("kvm: exiting hardware virtualization\n");
4423 kvm_rebooting = true;
4424 on_each_cpu(hardware_disable_nolock, NULL, 1);
4428 static struct notifier_block kvm_reboot_notifier = {
4429 .notifier_call = kvm_reboot,
4433 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
4437 for (i = 0; i < bus->dev_count; i++) {
4438 struct kvm_io_device *pos = bus->range[i].dev;
4440 kvm_iodevice_destructor(pos);
4445 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
4446 const struct kvm_io_range *r2)
4448 gpa_t addr1 = r1->addr;
4449 gpa_t addr2 = r2->addr;
4454 /* If r2->len == 0, match the exact address. If r2->len != 0,
4455 * accept any overlapping write. Any order is acceptable for
4456 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
4457 * we process all of them.
4470 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
4472 return kvm_io_bus_cmp(p1, p2);
4475 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
4476 gpa_t addr, int len)
4478 struct kvm_io_range *range, key;
4481 key = (struct kvm_io_range) {
4486 range = bsearch(&key, bus->range, bus->dev_count,
4487 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
4491 off = range - bus->range;
4493 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
4499 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
4500 struct kvm_io_range *range, const void *val)
4504 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
4508 while (idx < bus->dev_count &&
4509 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
4510 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
4519 /* kvm_io_bus_write - called under kvm->slots_lock */
4520 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
4521 int len, const void *val)
4523 struct kvm_io_bus *bus;
4524 struct kvm_io_range range;
4527 range = (struct kvm_io_range) {
4532 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
4535 r = __kvm_io_bus_write(vcpu, bus, &range, val);
4536 return r < 0 ? r : 0;
4538 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
4540 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
4541 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
4542 gpa_t addr, int len, const void *val, long cookie)
4544 struct kvm_io_bus *bus;
4545 struct kvm_io_range range;
4547 range = (struct kvm_io_range) {
4552 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
4556 /* First try the device referenced by cookie. */
4557 if ((cookie >= 0) && (cookie < bus->dev_count) &&
4558 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
4559 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
4564 * cookie contained garbage; fall back to search and return the
4565 * correct cookie value.
4567 return __kvm_io_bus_write(vcpu, bus, &range, val);
4570 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
4571 struct kvm_io_range *range, void *val)
4575 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
4579 while (idx < bus->dev_count &&
4580 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
4581 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
4590 /* kvm_io_bus_read - called under kvm->slots_lock */
4591 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
4594 struct kvm_io_bus *bus;
4595 struct kvm_io_range range;
4598 range = (struct kvm_io_range) {
4603 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
4606 r = __kvm_io_bus_read(vcpu, bus, &range, val);
4607 return r < 0 ? r : 0;
4610 /* Caller must hold slots_lock. */
4611 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
4612 int len, struct kvm_io_device *dev)
4615 struct kvm_io_bus *new_bus, *bus;
4616 struct kvm_io_range range;
4618 bus = kvm_get_bus(kvm, bus_idx);
4622 /* exclude ioeventfd which is limited by maximum fd */
4623 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
4626 new_bus = kmalloc(struct_size(bus, range, bus->dev_count + 1),
4627 GFP_KERNEL_ACCOUNT);
4631 range = (struct kvm_io_range) {
4637 for (i = 0; i < bus->dev_count; i++)
4638 if (kvm_io_bus_cmp(&bus->range[i], &range) > 0)
4641 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
4642 new_bus->dev_count++;
4643 new_bus->range[i] = range;
4644 memcpy(new_bus->range + i + 1, bus->range + i,
4645 (bus->dev_count - i) * sizeof(struct kvm_io_range));
4646 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
4647 synchronize_srcu_expedited(&kvm->srcu);
4653 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
4654 struct kvm_io_device *dev)
4657 struct kvm_io_bus *new_bus, *bus;
4659 lockdep_assert_held(&kvm->slots_lock);
4661 bus = kvm_get_bus(kvm, bus_idx);
4665 for (i = 0; i < bus->dev_count; i++) {
4666 if (bus->range[i].dev == dev) {
4671 if (i == bus->dev_count)
4674 new_bus = kmalloc(struct_size(bus, range, bus->dev_count - 1),
4675 GFP_KERNEL_ACCOUNT);
4677 memcpy(new_bus, bus, struct_size(bus, range, i));
4678 new_bus->dev_count--;
4679 memcpy(new_bus->range + i, bus->range + i + 1,
4680 flex_array_size(new_bus, range, new_bus->dev_count - i));
4683 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
4684 synchronize_srcu_expedited(&kvm->srcu);
4686 /* Destroy the old bus _after_ installing the (null) bus. */
4688 pr_err("kvm: failed to shrink bus, removing it completely\n");
4689 for (j = 0; j < bus->dev_count; j++) {
4692 kvm_iodevice_destructor(bus->range[j].dev);
4697 return new_bus ? 0 : -ENOMEM;
4700 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
4703 struct kvm_io_bus *bus;
4704 int dev_idx, srcu_idx;
4705 struct kvm_io_device *iodev = NULL;
4707 srcu_idx = srcu_read_lock(&kvm->srcu);
4709 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
4713 dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
4717 iodev = bus->range[dev_idx].dev;
4720 srcu_read_unlock(&kvm->srcu, srcu_idx);
4724 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
4726 static int kvm_debugfs_open(struct inode *inode, struct file *file,
4727 int (*get)(void *, u64 *), int (*set)(void *, u64),
4730 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
4733 /* The debugfs files are a reference to the kvm struct which
4734 * is still valid when kvm_destroy_vm is called.
4735 * To avoid the race between open and the removal of the debugfs
4736 * directory we test against the users count.
4738 if (!refcount_inc_not_zero(&stat_data->kvm->users_count))
4741 if (simple_attr_open(inode, file, get,
4742 KVM_DBGFS_GET_MODE(stat_data->dbgfs_item) & 0222
4745 kvm_put_kvm(stat_data->kvm);
4752 static int kvm_debugfs_release(struct inode *inode, struct file *file)
4754 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
4757 simple_attr_release(inode, file);
4758 kvm_put_kvm(stat_data->kvm);
4763 static int kvm_get_stat_per_vm(struct kvm *kvm, size_t offset, u64 *val)
4765 *val = *(ulong *)((void *)kvm + offset);
4770 static int kvm_clear_stat_per_vm(struct kvm *kvm, size_t offset)
4772 *(ulong *)((void *)kvm + offset) = 0;
4777 static int kvm_get_stat_per_vcpu(struct kvm *kvm, size_t offset, u64 *val)
4780 struct kvm_vcpu *vcpu;
4784 kvm_for_each_vcpu(i, vcpu, kvm)
4785 *val += *(u64 *)((void *)vcpu + offset);
4790 static int kvm_clear_stat_per_vcpu(struct kvm *kvm, size_t offset)
4793 struct kvm_vcpu *vcpu;
4795 kvm_for_each_vcpu(i, vcpu, kvm)
4796 *(u64 *)((void *)vcpu + offset) = 0;
4801 static int kvm_stat_data_get(void *data, u64 *val)
4804 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
4806 switch (stat_data->dbgfs_item->kind) {
4808 r = kvm_get_stat_per_vm(stat_data->kvm,
4809 stat_data->dbgfs_item->offset, val);
4812 r = kvm_get_stat_per_vcpu(stat_data->kvm,
4813 stat_data->dbgfs_item->offset, val);
4820 static int kvm_stat_data_clear(void *data, u64 val)
4823 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
4828 switch (stat_data->dbgfs_item->kind) {
4830 r = kvm_clear_stat_per_vm(stat_data->kvm,
4831 stat_data->dbgfs_item->offset);
4834 r = kvm_clear_stat_per_vcpu(stat_data->kvm,
4835 stat_data->dbgfs_item->offset);
4842 static int kvm_stat_data_open(struct inode *inode, struct file *file)
4844 __simple_attr_check_format("%llu\n", 0ull);
4845 return kvm_debugfs_open(inode, file, kvm_stat_data_get,
4846 kvm_stat_data_clear, "%llu\n");
4849 static const struct file_operations stat_fops_per_vm = {
4850 .owner = THIS_MODULE,
4851 .open = kvm_stat_data_open,
4852 .release = kvm_debugfs_release,
4853 .read = simple_attr_read,
4854 .write = simple_attr_write,
4855 .llseek = no_llseek,
4858 static int vm_stat_get(void *_offset, u64 *val)
4860 unsigned offset = (long)_offset;
4865 mutex_lock(&kvm_lock);
4866 list_for_each_entry(kvm, &vm_list, vm_list) {
4867 kvm_get_stat_per_vm(kvm, offset, &tmp_val);
4870 mutex_unlock(&kvm_lock);
4874 static int vm_stat_clear(void *_offset, u64 val)
4876 unsigned offset = (long)_offset;
4882 mutex_lock(&kvm_lock);
4883 list_for_each_entry(kvm, &vm_list, vm_list) {
4884 kvm_clear_stat_per_vm(kvm, offset);
4886 mutex_unlock(&kvm_lock);
4891 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n");
4893 static int vcpu_stat_get(void *_offset, u64 *val)
4895 unsigned offset = (long)_offset;
4900 mutex_lock(&kvm_lock);
4901 list_for_each_entry(kvm, &vm_list, vm_list) {
4902 kvm_get_stat_per_vcpu(kvm, offset, &tmp_val);
4905 mutex_unlock(&kvm_lock);
4909 static int vcpu_stat_clear(void *_offset, u64 val)
4911 unsigned offset = (long)_offset;
4917 mutex_lock(&kvm_lock);
4918 list_for_each_entry(kvm, &vm_list, vm_list) {
4919 kvm_clear_stat_per_vcpu(kvm, offset);
4921 mutex_unlock(&kvm_lock);
4926 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear,
4929 static const struct file_operations *stat_fops[] = {
4930 [KVM_STAT_VCPU] = &vcpu_stat_fops,
4931 [KVM_STAT_VM] = &vm_stat_fops,
4934 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm)
4936 struct kobj_uevent_env *env;
4937 unsigned long long created, active;
4939 if (!kvm_dev.this_device || !kvm)
4942 mutex_lock(&kvm_lock);
4943 if (type == KVM_EVENT_CREATE_VM) {
4944 kvm_createvm_count++;
4946 } else if (type == KVM_EVENT_DESTROY_VM) {
4949 created = kvm_createvm_count;
4950 active = kvm_active_vms;
4951 mutex_unlock(&kvm_lock);
4953 env = kzalloc(sizeof(*env), GFP_KERNEL_ACCOUNT);
4957 add_uevent_var(env, "CREATED=%llu", created);
4958 add_uevent_var(env, "COUNT=%llu", active);
4960 if (type == KVM_EVENT_CREATE_VM) {
4961 add_uevent_var(env, "EVENT=create");
4962 kvm->userspace_pid = task_pid_nr(current);
4963 } else if (type == KVM_EVENT_DESTROY_VM) {
4964 add_uevent_var(env, "EVENT=destroy");
4966 add_uevent_var(env, "PID=%d", kvm->userspace_pid);
4968 if (!IS_ERR_OR_NULL(kvm->debugfs_dentry)) {
4969 char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL_ACCOUNT);
4972 tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX);
4974 add_uevent_var(env, "STATS_PATH=%s", tmp);
4978 /* no need for checks, since we are adding at most only 5 keys */
4979 env->envp[env->envp_idx++] = NULL;
4980 kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp);
4984 static void kvm_init_debug(void)
4986 struct kvm_stats_debugfs_item *p;
4988 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
4990 kvm_debugfs_num_entries = 0;
4991 for (p = debugfs_entries; p->name; ++p, kvm_debugfs_num_entries++) {
4992 debugfs_create_file(p->name, KVM_DBGFS_GET_MODE(p),
4993 kvm_debugfs_dir, (void *)(long)p->offset,
4994 stat_fops[p->kind]);
4998 static int kvm_suspend(void)
5000 if (kvm_usage_count)
5001 hardware_disable_nolock(NULL);
5005 static void kvm_resume(void)
5007 if (kvm_usage_count) {
5008 #ifdef CONFIG_LOCKDEP
5009 WARN_ON(lockdep_is_held(&kvm_count_lock));
5011 hardware_enable_nolock(NULL);
5015 static struct syscore_ops kvm_syscore_ops = {
5016 .suspend = kvm_suspend,
5017 .resume = kvm_resume,
5021 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
5023 return container_of(pn, struct kvm_vcpu, preempt_notifier);
5026 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
5028 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
5030 WRITE_ONCE(vcpu->preempted, false);
5031 WRITE_ONCE(vcpu->ready, false);
5033 __this_cpu_write(kvm_running_vcpu, vcpu);
5034 kvm_arch_sched_in(vcpu, cpu);
5035 kvm_arch_vcpu_load(vcpu, cpu);
5038 static void kvm_sched_out(struct preempt_notifier *pn,
5039 struct task_struct *next)
5041 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
5043 if (current->on_rq) {
5044 WRITE_ONCE(vcpu->preempted, true);
5045 WRITE_ONCE(vcpu->ready, true);
5047 kvm_arch_vcpu_put(vcpu);
5048 __this_cpu_write(kvm_running_vcpu, NULL);
5052 * kvm_get_running_vcpu - get the vcpu running on the current CPU.
5054 * We can disable preemption locally around accessing the per-CPU variable,
5055 * and use the resolved vcpu pointer after enabling preemption again,
5056 * because even if the current thread is migrated to another CPU, reading
5057 * the per-CPU value later will give us the same value as we update the
5058 * per-CPU variable in the preempt notifier handlers.
5060 struct kvm_vcpu *kvm_get_running_vcpu(void)
5062 struct kvm_vcpu *vcpu;
5065 vcpu = __this_cpu_read(kvm_running_vcpu);
5070 EXPORT_SYMBOL_GPL(kvm_get_running_vcpu);
5073 * kvm_get_running_vcpus - get the per-CPU array of currently running vcpus.
5075 struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void)
5077 return &kvm_running_vcpu;
5080 struct kvm_cpu_compat_check {
5085 static void check_processor_compat(void *data)
5087 struct kvm_cpu_compat_check *c = data;
5089 *c->ret = kvm_arch_check_processor_compat(c->opaque);
5092 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
5093 struct module *module)
5095 struct kvm_cpu_compat_check c;
5099 r = kvm_arch_init(opaque);
5104 * kvm_arch_init makes sure there's at most one caller
5105 * for architectures that support multiple implementations,
5106 * like intel and amd on x86.
5107 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
5108 * conflicts in case kvm is already setup for another implementation.
5110 r = kvm_irqfd_init();
5114 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
5119 r = kvm_arch_hardware_setup(opaque);
5125 for_each_online_cpu(cpu) {
5126 smp_call_function_single(cpu, check_processor_compat, &c, 1);
5131 r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting",
5132 kvm_starting_cpu, kvm_dying_cpu);
5135 register_reboot_notifier(&kvm_reboot_notifier);
5137 /* A kmem cache lets us meet the alignment requirements of fx_save. */
5139 vcpu_align = __alignof__(struct kvm_vcpu);
5141 kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align,
5143 offsetof(struct kvm_vcpu, arch),
5144 sizeof_field(struct kvm_vcpu, arch),
5146 if (!kvm_vcpu_cache) {
5151 r = kvm_async_pf_init();
5155 kvm_chardev_ops.owner = module;
5156 kvm_vm_fops.owner = module;
5157 kvm_vcpu_fops.owner = module;
5159 r = misc_register(&kvm_dev);
5161 pr_err("kvm: misc device register failed\n");
5165 register_syscore_ops(&kvm_syscore_ops);
5167 kvm_preempt_ops.sched_in = kvm_sched_in;
5168 kvm_preempt_ops.sched_out = kvm_sched_out;
5172 r = kvm_vfio_ops_init();
5178 kvm_async_pf_deinit();
5180 kmem_cache_destroy(kvm_vcpu_cache);
5182 unregister_reboot_notifier(&kvm_reboot_notifier);
5183 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
5185 kvm_arch_hardware_unsetup();
5187 free_cpumask_var(cpus_hardware_enabled);
5195 EXPORT_SYMBOL_GPL(kvm_init);
5199 debugfs_remove_recursive(kvm_debugfs_dir);
5200 misc_deregister(&kvm_dev);
5201 kmem_cache_destroy(kvm_vcpu_cache);
5202 kvm_async_pf_deinit();
5203 unregister_syscore_ops(&kvm_syscore_ops);
5204 unregister_reboot_notifier(&kvm_reboot_notifier);
5205 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
5206 on_each_cpu(hardware_disable_nolock, NULL, 1);
5207 kvm_arch_hardware_unsetup();
5210 free_cpumask_var(cpus_hardware_enabled);
5211 kvm_vfio_ops_exit();
5213 EXPORT_SYMBOL_GPL(kvm_exit);
5215 struct kvm_vm_worker_thread_context {
5217 struct task_struct *parent;
5218 struct completion init_done;
5219 kvm_vm_thread_fn_t thread_fn;
5224 static int kvm_vm_worker_thread(void *context)
5227 * The init_context is allocated on the stack of the parent thread, so
5228 * we have to locally copy anything that is needed beyond initialization
5230 struct kvm_vm_worker_thread_context *init_context = context;
5231 struct kvm *kvm = init_context->kvm;
5232 kvm_vm_thread_fn_t thread_fn = init_context->thread_fn;
5233 uintptr_t data = init_context->data;
5236 err = kthread_park(current);
5237 /* kthread_park(current) is never supposed to return an error */
5242 err = cgroup_attach_task_all(init_context->parent, current);
5244 kvm_err("%s: cgroup_attach_task_all failed with err %d\n",
5249 set_user_nice(current, task_nice(init_context->parent));
5252 init_context->err = err;
5253 complete(&init_context->init_done);
5254 init_context = NULL;
5259 /* Wait to be woken up by the spawner before proceeding. */
5262 if (!kthread_should_stop())
5263 err = thread_fn(kvm, data);
5268 int kvm_vm_create_worker_thread(struct kvm *kvm, kvm_vm_thread_fn_t thread_fn,
5269 uintptr_t data, const char *name,
5270 struct task_struct **thread_ptr)
5272 struct kvm_vm_worker_thread_context init_context = {};
5273 struct task_struct *thread;
5276 init_context.kvm = kvm;
5277 init_context.parent = current;
5278 init_context.thread_fn = thread_fn;
5279 init_context.data = data;
5280 init_completion(&init_context.init_done);
5282 thread = kthread_run(kvm_vm_worker_thread, &init_context,
5283 "%s-%d", name, task_pid_nr(current));
5285 return PTR_ERR(thread);
5287 /* kthread_run is never supposed to return NULL */
5288 WARN_ON(thread == NULL);
5290 wait_for_completion(&init_context.init_done);
5292 if (!init_context.err)
5293 *thread_ptr = thread;
5295 return init_context.err;