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>
54 #include <linux/suspend.h>
56 #include <asm/processor.h>
57 #include <asm/ioctl.h>
58 #include <linux/uaccess.h>
60 #include "coalesced_mmio.h"
65 #define CREATE_TRACE_POINTS
66 #include <trace/events/kvm.h>
68 #include <linux/kvm_dirty_ring.h>
70 /* Worst case buffer size needed for holding an integer. */
71 #define ITOA_MAX_LEN 12
73 MODULE_AUTHOR("Qumranet");
74 MODULE_LICENSE("GPL");
76 /* Architectures should define their poll value according to the halt latency */
77 unsigned int halt_poll_ns = KVM_HALT_POLL_NS_DEFAULT;
78 module_param(halt_poll_ns, uint, 0644);
79 EXPORT_SYMBOL_GPL(halt_poll_ns);
81 /* Default doubles per-vcpu halt_poll_ns. */
82 unsigned int halt_poll_ns_grow = 2;
83 module_param(halt_poll_ns_grow, uint, 0644);
84 EXPORT_SYMBOL_GPL(halt_poll_ns_grow);
86 /* The start value to grow halt_poll_ns from */
87 unsigned int halt_poll_ns_grow_start = 10000; /* 10us */
88 module_param(halt_poll_ns_grow_start, uint, 0644);
89 EXPORT_SYMBOL_GPL(halt_poll_ns_grow_start);
91 /* Default resets per-vcpu halt_poll_ns . */
92 unsigned int halt_poll_ns_shrink;
93 module_param(halt_poll_ns_shrink, uint, 0644);
94 EXPORT_SYMBOL_GPL(halt_poll_ns_shrink);
99 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
102 DEFINE_MUTEX(kvm_lock);
103 static DEFINE_RAW_SPINLOCK(kvm_count_lock);
106 static cpumask_var_t cpus_hardware_enabled;
107 static int kvm_usage_count;
108 static atomic_t hardware_enable_failed;
110 static struct kmem_cache *kvm_vcpu_cache;
112 static __read_mostly struct preempt_ops kvm_preempt_ops;
113 static DEFINE_PER_CPU(struct kvm_vcpu *, kvm_running_vcpu);
115 struct dentry *kvm_debugfs_dir;
116 EXPORT_SYMBOL_GPL(kvm_debugfs_dir);
118 static const struct file_operations stat_fops_per_vm;
120 static struct file_operations kvm_chardev_ops;
122 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
124 #ifdef CONFIG_KVM_COMPAT
125 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
127 #define KVM_COMPAT(c) .compat_ioctl = (c)
130 * For architectures that don't implement a compat infrastructure,
131 * adopt a double line of defense:
132 * - Prevent a compat task from opening /dev/kvm
133 * - If the open has been done by a 64bit task, and the KVM fd
134 * passed to a compat task, let the ioctls fail.
136 static long kvm_no_compat_ioctl(struct file *file, unsigned int ioctl,
137 unsigned long arg) { return -EINVAL; }
139 static int kvm_no_compat_open(struct inode *inode, struct file *file)
141 return is_compat_task() ? -ENODEV : 0;
143 #define KVM_COMPAT(c) .compat_ioctl = kvm_no_compat_ioctl, \
144 .open = kvm_no_compat_open
146 static int hardware_enable_all(void);
147 static void hardware_disable_all(void);
149 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
151 __visible bool kvm_rebooting;
152 EXPORT_SYMBOL_GPL(kvm_rebooting);
154 #define KVM_EVENT_CREATE_VM 0
155 #define KVM_EVENT_DESTROY_VM 1
156 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm);
157 static unsigned long long kvm_createvm_count;
158 static unsigned long long kvm_active_vms;
160 static DEFINE_PER_CPU(cpumask_var_t, cpu_kick_mask);
162 __weak void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
163 unsigned long start, unsigned long end)
167 __weak void kvm_arch_guest_memory_reclaimed(struct kvm *kvm)
171 bool kvm_is_zone_device_pfn(kvm_pfn_t pfn)
174 * The metadata used by is_zone_device_page() to determine whether or
175 * not a page is ZONE_DEVICE is guaranteed to be valid if and only if
176 * the device has been pinned, e.g. by get_user_pages(). WARN if the
177 * page_count() is zero to help detect bad usage of this helper.
179 if (!pfn_valid(pfn) || WARN_ON_ONCE(!page_count(pfn_to_page(pfn))))
182 return is_zone_device_page(pfn_to_page(pfn));
185 bool kvm_is_reserved_pfn(kvm_pfn_t pfn)
188 * ZONE_DEVICE pages currently set PG_reserved, but from a refcounting
189 * perspective they are "normal" pages, albeit with slightly different
193 return PageReserved(pfn_to_page(pfn)) &&
195 !kvm_is_zone_device_pfn(pfn);
201 * Switches to specified vcpu, until a matching vcpu_put()
203 void vcpu_load(struct kvm_vcpu *vcpu)
207 __this_cpu_write(kvm_running_vcpu, vcpu);
208 preempt_notifier_register(&vcpu->preempt_notifier);
209 kvm_arch_vcpu_load(vcpu, cpu);
212 EXPORT_SYMBOL_GPL(vcpu_load);
214 void vcpu_put(struct kvm_vcpu *vcpu)
217 kvm_arch_vcpu_put(vcpu);
218 preempt_notifier_unregister(&vcpu->preempt_notifier);
219 __this_cpu_write(kvm_running_vcpu, NULL);
222 EXPORT_SYMBOL_GPL(vcpu_put);
224 /* TODO: merge with kvm_arch_vcpu_should_kick */
225 static bool kvm_request_needs_ipi(struct kvm_vcpu *vcpu, unsigned req)
227 int mode = kvm_vcpu_exiting_guest_mode(vcpu);
230 * We need to wait for the VCPU to reenable interrupts and get out of
231 * READING_SHADOW_PAGE_TABLES mode.
233 if (req & KVM_REQUEST_WAIT)
234 return mode != OUTSIDE_GUEST_MODE;
237 * Need to kick a running VCPU, but otherwise there is nothing to do.
239 return mode == IN_GUEST_MODE;
242 static void ack_flush(void *_completed)
246 static inline bool kvm_kick_many_cpus(cpumask_var_t tmp, bool wait)
248 const struct cpumask *cpus;
250 if (likely(cpumask_available(tmp)))
253 cpus = cpu_online_mask;
255 if (cpumask_empty(cpus))
258 smp_call_function_many(cpus, ack_flush, NULL, wait);
262 static void kvm_make_vcpu_request(struct kvm *kvm, struct kvm_vcpu *vcpu,
263 unsigned int req, cpumask_var_t tmp,
268 kvm_make_request(req, vcpu);
270 if (!(req & KVM_REQUEST_NO_WAKEUP) && kvm_vcpu_wake_up(vcpu))
274 * tmp can be "unavailable" if cpumasks are allocated off stack as
275 * allocation of the mask is deliberately not fatal and is handled by
276 * falling back to kicking all online CPUs.
278 if (!cpumask_available(tmp))
282 * Note, the vCPU could get migrated to a different pCPU at any point
283 * after kvm_request_needs_ipi(), which could result in sending an IPI
284 * to the previous pCPU. But, that's OK because the purpose of the IPI
285 * is to ensure the vCPU returns to OUTSIDE_GUEST_MODE, which is
286 * satisfied if the vCPU migrates. Entering READING_SHADOW_PAGE_TABLES
287 * after this point is also OK, as the requirement is only that KVM wait
288 * for vCPUs that were reading SPTEs _before_ any changes were
289 * finalized. See kvm_vcpu_kick() for more details on handling requests.
291 if (kvm_request_needs_ipi(vcpu, req)) {
292 cpu = READ_ONCE(vcpu->cpu);
293 if (cpu != -1 && cpu != current_cpu)
294 __cpumask_set_cpu(cpu, tmp);
298 bool kvm_make_vcpus_request_mask(struct kvm *kvm, unsigned int req,
299 struct kvm_vcpu *except,
300 unsigned long *vcpu_bitmap, cpumask_var_t tmp)
302 struct kvm_vcpu *vcpu;
308 for_each_set_bit(i, vcpu_bitmap, KVM_MAX_VCPUS) {
309 vcpu = kvm_get_vcpu(kvm, i);
310 if (!vcpu || vcpu == except)
312 kvm_make_vcpu_request(kvm, vcpu, req, tmp, me);
315 called = kvm_kick_many_cpus(tmp, !!(req & KVM_REQUEST_WAIT));
321 bool kvm_make_all_cpus_request_except(struct kvm *kvm, unsigned int req,
322 struct kvm_vcpu *except)
324 struct kvm_vcpu *vcpu;
325 struct cpumask *cpus;
331 cpus = this_cpu_cpumask_var_ptr(cpu_kick_mask);
334 kvm_for_each_vcpu(i, vcpu, kvm) {
337 kvm_make_vcpu_request(kvm, vcpu, req, cpus, me);
340 called = kvm_kick_many_cpus(cpus, !!(req & KVM_REQUEST_WAIT));
346 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
348 return kvm_make_all_cpus_request_except(kvm, req, NULL);
350 EXPORT_SYMBOL_GPL(kvm_make_all_cpus_request);
352 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
353 void kvm_flush_remote_tlbs(struct kvm *kvm)
355 ++kvm->stat.generic.remote_tlb_flush_requests;
358 * We want to publish modifications to the page tables before reading
359 * mode. Pairs with a memory barrier in arch-specific code.
360 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
361 * and smp_mb in walk_shadow_page_lockless_begin/end.
362 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
364 * There is already an smp_mb__after_atomic() before
365 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
368 if (!kvm_arch_flush_remote_tlb(kvm)
369 || kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
370 ++kvm->stat.generic.remote_tlb_flush;
372 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
375 void kvm_reload_remote_mmus(struct kvm *kvm)
377 kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
380 static void kvm_flush_shadow_all(struct kvm *kvm)
382 kvm_arch_flush_shadow_all(kvm);
383 kvm_arch_guest_memory_reclaimed(kvm);
386 #ifdef KVM_ARCH_NR_OBJS_PER_MEMORY_CACHE
387 static inline void *mmu_memory_cache_alloc_obj(struct kvm_mmu_memory_cache *mc,
390 gfp_flags |= mc->gfp_zero;
393 return kmem_cache_alloc(mc->kmem_cache, gfp_flags);
395 return (void *)__get_free_page(gfp_flags);
398 int kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache *mc, int min)
402 if (mc->nobjs >= min)
404 while (mc->nobjs < ARRAY_SIZE(mc->objects)) {
405 obj = mmu_memory_cache_alloc_obj(mc, GFP_KERNEL_ACCOUNT);
407 return mc->nobjs >= min ? 0 : -ENOMEM;
408 mc->objects[mc->nobjs++] = obj;
413 int kvm_mmu_memory_cache_nr_free_objects(struct kvm_mmu_memory_cache *mc)
418 void kvm_mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc)
422 kmem_cache_free(mc->kmem_cache, mc->objects[--mc->nobjs]);
424 free_page((unsigned long)mc->objects[--mc->nobjs]);
428 void *kvm_mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc)
432 if (WARN_ON(!mc->nobjs))
433 p = mmu_memory_cache_alloc_obj(mc, GFP_ATOMIC | __GFP_ACCOUNT);
435 p = mc->objects[--mc->nobjs];
441 static void kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
443 mutex_init(&vcpu->mutex);
448 rcuwait_init(&vcpu->wait);
449 kvm_async_pf_vcpu_init(vcpu);
452 INIT_LIST_HEAD(&vcpu->blocked_vcpu_list);
454 kvm_vcpu_set_in_spin_loop(vcpu, false);
455 kvm_vcpu_set_dy_eligible(vcpu, false);
456 vcpu->preempted = false;
458 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
459 vcpu->last_used_slot = 0;
462 void kvm_vcpu_destroy(struct kvm_vcpu *vcpu)
464 kvm_arch_vcpu_destroy(vcpu);
465 kvm_dirty_ring_free(&vcpu->dirty_ring);
468 * No need for rcu_read_lock as VCPU_RUN is the only place that changes
469 * the vcpu->pid pointer, and at destruction time all file descriptors
472 put_pid(rcu_dereference_protected(vcpu->pid, 1));
474 free_page((unsigned long)vcpu->run);
475 kmem_cache_free(kvm_vcpu_cache, vcpu);
477 EXPORT_SYMBOL_GPL(kvm_vcpu_destroy);
479 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
480 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
482 return container_of(mn, struct kvm, mmu_notifier);
485 static void kvm_mmu_notifier_invalidate_range(struct mmu_notifier *mn,
486 struct mm_struct *mm,
487 unsigned long start, unsigned long end)
489 struct kvm *kvm = mmu_notifier_to_kvm(mn);
492 idx = srcu_read_lock(&kvm->srcu);
493 kvm_arch_mmu_notifier_invalidate_range(kvm, start, end);
494 srcu_read_unlock(&kvm->srcu, idx);
497 typedef bool (*hva_handler_t)(struct kvm *kvm, struct kvm_gfn_range *range);
499 typedef void (*on_lock_fn_t)(struct kvm *kvm, unsigned long start,
502 typedef void (*on_unlock_fn_t)(struct kvm *kvm);
504 struct kvm_hva_range {
508 hva_handler_t handler;
509 on_lock_fn_t on_lock;
510 on_unlock_fn_t on_unlock;
516 * Use a dedicated stub instead of NULL to indicate that there is no callback
517 * function/handler. The compiler technically can't guarantee that a real
518 * function will have a non-zero address, and so it will generate code to
519 * check for !NULL, whereas comparing against a stub will be elided at compile
520 * time (unless the compiler is getting long in the tooth, e.g. gcc 4.9).
522 static void kvm_null_fn(void)
526 #define IS_KVM_NULL_FN(fn) ((fn) == (void *)kvm_null_fn)
528 static __always_inline int __kvm_handle_hva_range(struct kvm *kvm,
529 const struct kvm_hva_range *range)
531 bool ret = false, locked = false;
532 struct kvm_gfn_range gfn_range;
533 struct kvm_memory_slot *slot;
534 struct kvm_memslots *slots;
537 /* A null handler is allowed if and only if on_lock() is provided. */
538 if (WARN_ON_ONCE(IS_KVM_NULL_FN(range->on_lock) &&
539 IS_KVM_NULL_FN(range->handler)))
542 idx = srcu_read_lock(&kvm->srcu);
544 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
545 slots = __kvm_memslots(kvm, i);
546 kvm_for_each_memslot(slot, slots) {
547 unsigned long hva_start, hva_end;
549 hva_start = max(range->start, slot->userspace_addr);
550 hva_end = min(range->end, slot->userspace_addr +
551 (slot->npages << PAGE_SHIFT));
552 if (hva_start >= hva_end)
556 * To optimize for the likely case where the address
557 * range is covered by zero or one memslots, don't
558 * bother making these conditional (to avoid writes on
559 * the second or later invocation of the handler).
561 gfn_range.pte = range->pte;
562 gfn_range.may_block = range->may_block;
565 * {gfn(page) | page intersects with [hva_start, hva_end)} =
566 * {gfn_start, gfn_start+1, ..., gfn_end-1}.
568 gfn_range.start = hva_to_gfn_memslot(hva_start, slot);
569 gfn_range.end = hva_to_gfn_memslot(hva_end + PAGE_SIZE - 1, slot);
570 gfn_range.slot = slot;
575 if (!IS_KVM_NULL_FN(range->on_lock))
576 range->on_lock(kvm, range->start, range->end);
577 if (IS_KVM_NULL_FN(range->handler))
580 ret |= range->handler(kvm, &gfn_range);
584 if (range->flush_on_ret && ret)
585 kvm_flush_remote_tlbs(kvm);
589 if (!IS_KVM_NULL_FN(range->on_unlock))
590 range->on_unlock(kvm);
593 srcu_read_unlock(&kvm->srcu, idx);
595 /* The notifiers are averse to booleans. :-( */
599 static __always_inline int kvm_handle_hva_range(struct mmu_notifier *mn,
603 hva_handler_t handler)
605 struct kvm *kvm = mmu_notifier_to_kvm(mn);
606 const struct kvm_hva_range range = {
611 .on_lock = (void *)kvm_null_fn,
612 .on_unlock = (void *)kvm_null_fn,
613 .flush_on_ret = true,
617 return __kvm_handle_hva_range(kvm, &range);
620 static __always_inline int kvm_handle_hva_range_no_flush(struct mmu_notifier *mn,
623 hva_handler_t handler)
625 struct kvm *kvm = mmu_notifier_to_kvm(mn);
626 const struct kvm_hva_range range = {
631 .on_lock = (void *)kvm_null_fn,
632 .on_unlock = (void *)kvm_null_fn,
633 .flush_on_ret = false,
637 return __kvm_handle_hva_range(kvm, &range);
640 static bool kvm_change_spte_gfn(struct kvm *kvm, struct kvm_gfn_range *range)
643 * Skipping invalid memslots is correct if and only change_pte() is
644 * surrounded by invalidate_range_{start,end}(), which is currently
645 * guaranteed by the primary MMU. If that ever changes, KVM needs to
646 * unmap the memslot instead of skipping the memslot to ensure that KVM
647 * doesn't hold references to the old PFN.
649 WARN_ON_ONCE(!READ_ONCE(kvm->mn_active_invalidate_count));
651 if (range->slot->flags & KVM_MEMSLOT_INVALID)
654 return kvm_set_spte_gfn(kvm, range);
657 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
658 struct mm_struct *mm,
659 unsigned long address,
662 struct kvm *kvm = mmu_notifier_to_kvm(mn);
664 trace_kvm_set_spte_hva(address);
667 * .change_pte() must be surrounded by .invalidate_range_{start,end}().
668 * If mmu_notifier_count is zero, then no in-progress invalidations,
669 * including this one, found a relevant memslot at start(); rechecking
670 * memslots here is unnecessary. Note, a false positive (count elevated
671 * by a different invalidation) is sub-optimal but functionally ok.
673 WARN_ON_ONCE(!READ_ONCE(kvm->mn_active_invalidate_count));
674 if (!READ_ONCE(kvm->mmu_notifier_count))
677 kvm_handle_hva_range(mn, address, address + 1, pte, kvm_change_spte_gfn);
680 void kvm_inc_notifier_count(struct kvm *kvm, unsigned long start,
684 * The count increase must become visible at unlock time as no
685 * spte can be established without taking the mmu_lock and
686 * count is also read inside the mmu_lock critical section.
688 kvm->mmu_notifier_count++;
689 if (likely(kvm->mmu_notifier_count == 1)) {
690 kvm->mmu_notifier_range_start = start;
691 kvm->mmu_notifier_range_end = end;
694 * Fully tracking multiple concurrent ranges has dimishing
695 * returns. Keep things simple and just find the minimal range
696 * which includes the current and new ranges. As there won't be
697 * enough information to subtract a range after its invalidate
698 * completes, any ranges invalidated concurrently will
699 * accumulate and persist until all outstanding invalidates
702 kvm->mmu_notifier_range_start =
703 min(kvm->mmu_notifier_range_start, start);
704 kvm->mmu_notifier_range_end =
705 max(kvm->mmu_notifier_range_end, end);
709 static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
710 const struct mmu_notifier_range *range)
712 struct kvm *kvm = mmu_notifier_to_kvm(mn);
713 const struct kvm_hva_range hva_range = {
714 .start = range->start,
717 .handler = kvm_unmap_gfn_range,
718 .on_lock = kvm_inc_notifier_count,
719 .on_unlock = kvm_arch_guest_memory_reclaimed,
720 .flush_on_ret = true,
721 .may_block = mmu_notifier_range_blockable(range),
724 trace_kvm_unmap_hva_range(range->start, range->end);
727 * Prevent memslot modification between range_start() and range_end()
728 * so that conditionally locking provides the same result in both
729 * functions. Without that guarantee, the mmu_notifier_count
730 * adjustments will be imbalanced.
732 * Pairs with the decrement in range_end().
734 spin_lock(&kvm->mn_invalidate_lock);
735 kvm->mn_active_invalidate_count++;
736 spin_unlock(&kvm->mn_invalidate_lock);
738 __kvm_handle_hva_range(kvm, &hva_range);
743 void kvm_dec_notifier_count(struct kvm *kvm, unsigned long start,
747 * This sequence increase will notify the kvm page fault that
748 * the page that is going to be mapped in the spte could have
751 kvm->mmu_notifier_seq++;
754 * The above sequence increase must be visible before the
755 * below count decrease, which is ensured by the smp_wmb above
756 * in conjunction with the smp_rmb in mmu_notifier_retry().
758 kvm->mmu_notifier_count--;
761 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
762 const struct mmu_notifier_range *range)
764 struct kvm *kvm = mmu_notifier_to_kvm(mn);
765 const struct kvm_hva_range hva_range = {
766 .start = range->start,
769 .handler = (void *)kvm_null_fn,
770 .on_lock = kvm_dec_notifier_count,
771 .on_unlock = (void *)kvm_null_fn,
772 .flush_on_ret = false,
773 .may_block = mmu_notifier_range_blockable(range),
777 __kvm_handle_hva_range(kvm, &hva_range);
779 /* Pairs with the increment in range_start(). */
780 spin_lock(&kvm->mn_invalidate_lock);
781 wake = (--kvm->mn_active_invalidate_count == 0);
782 spin_unlock(&kvm->mn_invalidate_lock);
785 * There can only be one waiter, since the wait happens under
789 rcuwait_wake_up(&kvm->mn_memslots_update_rcuwait);
791 BUG_ON(kvm->mmu_notifier_count < 0);
794 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
795 struct mm_struct *mm,
799 trace_kvm_age_hva(start, end);
801 return kvm_handle_hva_range(mn, start, end, __pte(0), kvm_age_gfn);
804 static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
805 struct mm_struct *mm,
809 trace_kvm_age_hva(start, end);
812 * Even though we do not flush TLB, this will still adversely
813 * affect performance on pre-Haswell Intel EPT, where there is
814 * no EPT Access Bit to clear so that we have to tear down EPT
815 * tables instead. If we find this unacceptable, we can always
816 * add a parameter to kvm_age_hva so that it effectively doesn't
817 * do anything on clear_young.
819 * Also note that currently we never issue secondary TLB flushes
820 * from clear_young, leaving this job up to the regular system
821 * cadence. If we find this inaccurate, we might come up with a
822 * more sophisticated heuristic later.
824 return kvm_handle_hva_range_no_flush(mn, start, end, kvm_age_gfn);
827 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
828 struct mm_struct *mm,
829 unsigned long address)
831 trace_kvm_test_age_hva(address);
833 return kvm_handle_hva_range_no_flush(mn, address, address + 1,
837 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
838 struct mm_struct *mm)
840 struct kvm *kvm = mmu_notifier_to_kvm(mn);
843 idx = srcu_read_lock(&kvm->srcu);
844 kvm_flush_shadow_all(kvm);
845 srcu_read_unlock(&kvm->srcu, idx);
848 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
849 .invalidate_range = kvm_mmu_notifier_invalidate_range,
850 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
851 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
852 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
853 .clear_young = kvm_mmu_notifier_clear_young,
854 .test_young = kvm_mmu_notifier_test_young,
855 .change_pte = kvm_mmu_notifier_change_pte,
856 .release = kvm_mmu_notifier_release,
859 static int kvm_init_mmu_notifier(struct kvm *kvm)
861 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
862 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
865 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
867 static int kvm_init_mmu_notifier(struct kvm *kvm)
872 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
874 #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER
875 static int kvm_pm_notifier_call(struct notifier_block *bl,
879 struct kvm *kvm = container_of(bl, struct kvm, pm_notifier);
881 return kvm_arch_pm_notifier(kvm, state);
884 static void kvm_init_pm_notifier(struct kvm *kvm)
886 kvm->pm_notifier.notifier_call = kvm_pm_notifier_call;
887 /* Suspend KVM before we suspend ftrace, RCU, etc. */
888 kvm->pm_notifier.priority = INT_MAX;
889 register_pm_notifier(&kvm->pm_notifier);
892 static void kvm_destroy_pm_notifier(struct kvm *kvm)
894 unregister_pm_notifier(&kvm->pm_notifier);
896 #else /* !CONFIG_HAVE_KVM_PM_NOTIFIER */
897 static void kvm_init_pm_notifier(struct kvm *kvm)
901 static void kvm_destroy_pm_notifier(struct kvm *kvm)
904 #endif /* CONFIG_HAVE_KVM_PM_NOTIFIER */
906 static struct kvm_memslots *kvm_alloc_memslots(void)
909 struct kvm_memslots *slots;
911 slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL_ACCOUNT);
915 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
916 slots->id_to_index[i] = -1;
921 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
923 if (!memslot->dirty_bitmap)
926 kvfree(memslot->dirty_bitmap);
927 memslot->dirty_bitmap = NULL;
930 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
932 kvm_destroy_dirty_bitmap(slot);
934 kvm_arch_free_memslot(kvm, slot);
940 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
942 struct kvm_memory_slot *memslot;
947 kvm_for_each_memslot(memslot, slots)
948 kvm_free_memslot(kvm, memslot);
953 static umode_t kvm_stats_debugfs_mode(const struct _kvm_stats_desc *pdesc)
955 switch (pdesc->desc.flags & KVM_STATS_TYPE_MASK) {
956 case KVM_STATS_TYPE_INSTANT:
958 case KVM_STATS_TYPE_CUMULATIVE:
959 case KVM_STATS_TYPE_PEAK:
966 static void kvm_destroy_vm_debugfs(struct kvm *kvm)
969 int kvm_debugfs_num_entries = kvm_vm_stats_header.num_desc +
970 kvm_vcpu_stats_header.num_desc;
972 if (IS_ERR(kvm->debugfs_dentry))
975 debugfs_remove_recursive(kvm->debugfs_dentry);
977 if (kvm->debugfs_stat_data) {
978 for (i = 0; i < kvm_debugfs_num_entries; i++)
979 kfree(kvm->debugfs_stat_data[i]);
980 kfree(kvm->debugfs_stat_data);
984 static int kvm_create_vm_debugfs(struct kvm *kvm, int fd)
986 static DEFINE_MUTEX(kvm_debugfs_lock);
988 char dir_name[ITOA_MAX_LEN * 2];
989 struct kvm_stat_data *stat_data;
990 const struct _kvm_stats_desc *pdesc;
992 int kvm_debugfs_num_entries = kvm_vm_stats_header.num_desc +
993 kvm_vcpu_stats_header.num_desc;
995 if (!debugfs_initialized())
998 snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd);
999 mutex_lock(&kvm_debugfs_lock);
1000 dent = debugfs_lookup(dir_name, kvm_debugfs_dir);
1002 pr_warn_ratelimited("KVM: debugfs: duplicate directory %s\n", dir_name);
1004 mutex_unlock(&kvm_debugfs_lock);
1007 dent = debugfs_create_dir(dir_name, kvm_debugfs_dir);
1008 mutex_unlock(&kvm_debugfs_lock);
1012 kvm->debugfs_dentry = dent;
1013 kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries,
1014 sizeof(*kvm->debugfs_stat_data),
1015 GFP_KERNEL_ACCOUNT);
1016 if (!kvm->debugfs_stat_data)
1019 for (i = 0; i < kvm_vm_stats_header.num_desc; ++i) {
1020 pdesc = &kvm_vm_stats_desc[i];
1021 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT);
1025 stat_data->kvm = kvm;
1026 stat_data->desc = pdesc;
1027 stat_data->kind = KVM_STAT_VM;
1028 kvm->debugfs_stat_data[i] = stat_data;
1029 debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
1030 kvm->debugfs_dentry, stat_data,
1034 for (i = 0; i < kvm_vcpu_stats_header.num_desc; ++i) {
1035 pdesc = &kvm_vcpu_stats_desc[i];
1036 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT);
1040 stat_data->kvm = kvm;
1041 stat_data->desc = pdesc;
1042 stat_data->kind = KVM_STAT_VCPU;
1043 kvm->debugfs_stat_data[i + kvm_vm_stats_header.num_desc] = stat_data;
1044 debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
1045 kvm->debugfs_dentry, stat_data,
1049 ret = kvm_arch_create_vm_debugfs(kvm);
1051 kvm_destroy_vm_debugfs(kvm);
1059 * Called after the VM is otherwise initialized, but just before adding it to
1062 int __weak kvm_arch_post_init_vm(struct kvm *kvm)
1068 * Called just after removing the VM from the vm_list, but before doing any
1069 * other destruction.
1071 void __weak kvm_arch_pre_destroy_vm(struct kvm *kvm)
1076 * Called after per-vm debugfs created. When called kvm->debugfs_dentry should
1077 * be setup already, so we can create arch-specific debugfs entries under it.
1078 * Cleanup should be automatic done in kvm_destroy_vm_debugfs() recursively, so
1079 * a per-arch destroy interface is not needed.
1081 int __weak kvm_arch_create_vm_debugfs(struct kvm *kvm)
1086 static struct kvm *kvm_create_vm(unsigned long type)
1088 struct kvm *kvm = kvm_arch_alloc_vm();
1093 return ERR_PTR(-ENOMEM);
1095 /* KVM is pinned via open("/dev/kvm"), the fd passed to this ioctl(). */
1096 __module_get(kvm_chardev_ops.owner);
1098 KVM_MMU_LOCK_INIT(kvm);
1099 mmgrab(current->mm);
1100 kvm->mm = current->mm;
1101 kvm_eventfd_init(kvm);
1102 mutex_init(&kvm->lock);
1103 mutex_init(&kvm->irq_lock);
1104 mutex_init(&kvm->slots_lock);
1105 mutex_init(&kvm->slots_arch_lock);
1106 spin_lock_init(&kvm->mn_invalidate_lock);
1107 rcuwait_init(&kvm->mn_memslots_update_rcuwait);
1109 INIT_LIST_HEAD(&kvm->devices);
1111 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
1114 * Force subsequent debugfs file creations to fail if the VM directory
1115 * is not created (by kvm_create_vm_debugfs()).
1117 kvm->debugfs_dentry = ERR_PTR(-ENOENT);
1119 if (init_srcu_struct(&kvm->srcu))
1120 goto out_err_no_srcu;
1121 if (init_srcu_struct(&kvm->irq_srcu))
1122 goto out_err_no_irq_srcu;
1124 refcount_set(&kvm->users_count, 1);
1125 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
1126 struct kvm_memslots *slots = kvm_alloc_memslots();
1129 goto out_err_no_arch_destroy_vm;
1130 /* Generations must be different for each address space. */
1131 slots->generation = i;
1132 rcu_assign_pointer(kvm->memslots[i], slots);
1135 for (i = 0; i < KVM_NR_BUSES; i++) {
1136 rcu_assign_pointer(kvm->buses[i],
1137 kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL_ACCOUNT));
1139 goto out_err_no_arch_destroy_vm;
1142 kvm->max_halt_poll_ns = halt_poll_ns;
1144 r = kvm_arch_init_vm(kvm, type);
1146 goto out_err_no_arch_destroy_vm;
1148 r = hardware_enable_all();
1150 goto out_err_no_disable;
1152 #ifdef CONFIG_HAVE_KVM_IRQFD
1153 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
1156 r = kvm_init_mmu_notifier(kvm);
1158 goto out_err_no_mmu_notifier;
1160 r = kvm_arch_post_init_vm(kvm);
1164 mutex_lock(&kvm_lock);
1165 list_add(&kvm->vm_list, &vm_list);
1166 mutex_unlock(&kvm_lock);
1168 preempt_notifier_inc();
1169 kvm_init_pm_notifier(kvm);
1174 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1175 if (kvm->mmu_notifier.ops)
1176 mmu_notifier_unregister(&kvm->mmu_notifier, current->mm);
1178 out_err_no_mmu_notifier:
1179 hardware_disable_all();
1181 kvm_arch_destroy_vm(kvm);
1182 out_err_no_arch_destroy_vm:
1183 WARN_ON_ONCE(!refcount_dec_and_test(&kvm->users_count));
1184 for (i = 0; i < KVM_NR_BUSES; i++)
1185 kfree(kvm_get_bus(kvm, i));
1186 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
1187 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
1188 cleanup_srcu_struct(&kvm->irq_srcu);
1189 out_err_no_irq_srcu:
1190 cleanup_srcu_struct(&kvm->srcu);
1192 kvm_arch_free_vm(kvm);
1193 mmdrop(current->mm);
1194 module_put(kvm_chardev_ops.owner);
1198 static void kvm_destroy_devices(struct kvm *kvm)
1200 struct kvm_device *dev, *tmp;
1203 * We do not need to take the kvm->lock here, because nobody else
1204 * has a reference to the struct kvm at this point and therefore
1205 * cannot access the devices list anyhow.
1207 list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
1208 list_del(&dev->vm_node);
1209 dev->ops->destroy(dev);
1213 static void kvm_destroy_vm(struct kvm *kvm)
1216 struct mm_struct *mm = kvm->mm;
1218 kvm_destroy_pm_notifier(kvm);
1219 kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm);
1220 kvm_destroy_vm_debugfs(kvm);
1221 kvm_arch_sync_events(kvm);
1222 mutex_lock(&kvm_lock);
1223 list_del(&kvm->vm_list);
1224 mutex_unlock(&kvm_lock);
1225 kvm_arch_pre_destroy_vm(kvm);
1227 kvm_free_irq_routing(kvm);
1228 for (i = 0; i < KVM_NR_BUSES; i++) {
1229 struct kvm_io_bus *bus = kvm_get_bus(kvm, i);
1232 kvm_io_bus_destroy(bus);
1233 kvm->buses[i] = NULL;
1235 kvm_coalesced_mmio_free(kvm);
1236 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1237 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
1239 * At this point, pending calls to invalidate_range_start()
1240 * have completed but no more MMU notifiers will run, so
1241 * mn_active_invalidate_count may remain unbalanced.
1242 * No threads can be waiting in install_new_memslots as the
1243 * last reference on KVM has been dropped, but freeing
1244 * memslots would deadlock without this manual intervention.
1246 WARN_ON(rcuwait_active(&kvm->mn_memslots_update_rcuwait));
1247 kvm->mn_active_invalidate_count = 0;
1249 kvm_flush_shadow_all(kvm);
1251 kvm_arch_destroy_vm(kvm);
1252 kvm_destroy_devices(kvm);
1253 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
1254 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
1255 cleanup_srcu_struct(&kvm->irq_srcu);
1256 cleanup_srcu_struct(&kvm->srcu);
1257 kvm_arch_free_vm(kvm);
1258 preempt_notifier_dec();
1259 hardware_disable_all();
1261 module_put(kvm_chardev_ops.owner);
1264 void kvm_get_kvm(struct kvm *kvm)
1266 refcount_inc(&kvm->users_count);
1268 EXPORT_SYMBOL_GPL(kvm_get_kvm);
1271 * Make sure the vm is not during destruction, which is a safe version of
1272 * kvm_get_kvm(). Return true if kvm referenced successfully, false otherwise.
1274 bool kvm_get_kvm_safe(struct kvm *kvm)
1276 return refcount_inc_not_zero(&kvm->users_count);
1278 EXPORT_SYMBOL_GPL(kvm_get_kvm_safe);
1280 void kvm_put_kvm(struct kvm *kvm)
1282 if (refcount_dec_and_test(&kvm->users_count))
1283 kvm_destroy_vm(kvm);
1285 EXPORT_SYMBOL_GPL(kvm_put_kvm);
1288 * Used to put a reference that was taken on behalf of an object associated
1289 * with a user-visible file descriptor, e.g. a vcpu or device, if installation
1290 * of the new file descriptor fails and the reference cannot be transferred to
1291 * its final owner. In such cases, the caller is still actively using @kvm and
1292 * will fail miserably if the refcount unexpectedly hits zero.
1294 void kvm_put_kvm_no_destroy(struct kvm *kvm)
1296 WARN_ON(refcount_dec_and_test(&kvm->users_count));
1298 EXPORT_SYMBOL_GPL(kvm_put_kvm_no_destroy);
1300 static int kvm_vm_release(struct inode *inode, struct file *filp)
1302 struct kvm *kvm = filp->private_data;
1304 kvm_irqfd_release(kvm);
1311 * Allocation size is twice as large as the actual dirty bitmap size.
1312 * See kvm_vm_ioctl_get_dirty_log() why this is needed.
1314 static int kvm_alloc_dirty_bitmap(struct kvm_memory_slot *memslot)
1316 unsigned long dirty_bytes = kvm_dirty_bitmap_bytes(memslot);
1318 memslot->dirty_bitmap = __vcalloc(2, dirty_bytes, GFP_KERNEL_ACCOUNT);
1319 if (!memslot->dirty_bitmap)
1326 * Delete a memslot by decrementing the number of used slots and shifting all
1327 * other entries in the array forward one spot.
1329 static inline void kvm_memslot_delete(struct kvm_memslots *slots,
1330 struct kvm_memory_slot *memslot)
1332 struct kvm_memory_slot *mslots = slots->memslots;
1335 if (WARN_ON(slots->id_to_index[memslot->id] == -1))
1338 slots->used_slots--;
1340 if (atomic_read(&slots->last_used_slot) >= slots->used_slots)
1341 atomic_set(&slots->last_used_slot, 0);
1343 for (i = slots->id_to_index[memslot->id]; i < slots->used_slots; i++) {
1344 mslots[i] = mslots[i + 1];
1345 slots->id_to_index[mslots[i].id] = i;
1347 mslots[i] = *memslot;
1348 slots->id_to_index[memslot->id] = -1;
1352 * "Insert" a new memslot by incrementing the number of used slots. Returns
1353 * the new slot's initial index into the memslots array.
1355 static inline int kvm_memslot_insert_back(struct kvm_memslots *slots)
1357 return slots->used_slots++;
1361 * Move a changed memslot backwards in the array by shifting existing slots
1362 * with a higher GFN toward the front of the array. Note, the changed memslot
1363 * itself is not preserved in the array, i.e. not swapped at this time, only
1364 * its new index into the array is tracked. Returns the changed memslot's
1365 * current index into the memslots array.
1367 static inline int kvm_memslot_move_backward(struct kvm_memslots *slots,
1368 struct kvm_memory_slot *memslot)
1370 struct kvm_memory_slot *mslots = slots->memslots;
1373 if (WARN_ON_ONCE(slots->id_to_index[memslot->id] == -1) ||
1374 WARN_ON_ONCE(!slots->used_slots))
1378 * Move the target memslot backward in the array by shifting existing
1379 * memslots with a higher GFN (than the target memslot) towards the
1380 * front of the array.
1382 for (i = slots->id_to_index[memslot->id]; i < slots->used_slots - 1; i++) {
1383 if (memslot->base_gfn > mslots[i + 1].base_gfn)
1386 WARN_ON_ONCE(memslot->base_gfn == mslots[i + 1].base_gfn);
1388 /* Shift the next memslot forward one and update its index. */
1389 mslots[i] = mslots[i + 1];
1390 slots->id_to_index[mslots[i].id] = i;
1396 * Move a changed memslot forwards in the array by shifting existing slots with
1397 * a lower GFN toward the back of the array. Note, the changed memslot itself
1398 * is not preserved in the array, i.e. not swapped at this time, only its new
1399 * index into the array is tracked. Returns the changed memslot's final index
1400 * into the memslots array.
1402 static inline int kvm_memslot_move_forward(struct kvm_memslots *slots,
1403 struct kvm_memory_slot *memslot,
1406 struct kvm_memory_slot *mslots = slots->memslots;
1409 for (i = start; i > 0; i--) {
1410 if (memslot->base_gfn < mslots[i - 1].base_gfn)
1413 WARN_ON_ONCE(memslot->base_gfn == mslots[i - 1].base_gfn);
1415 /* Shift the next memslot back one and update its index. */
1416 mslots[i] = mslots[i - 1];
1417 slots->id_to_index[mslots[i].id] = i;
1423 * Re-sort memslots based on their GFN to account for an added, deleted, or
1424 * moved memslot. Sorting memslots by GFN allows using a binary search during
1427 * IMPORTANT: Slots are sorted from highest GFN to lowest GFN! I.e. the entry
1428 * at memslots[0] has the highest GFN.
1430 * The sorting algorithm takes advantage of having initially sorted memslots
1431 * and knowing the position of the changed memslot. Sorting is also optimized
1432 * by not swapping the updated memslot and instead only shifting other memslots
1433 * and tracking the new index for the update memslot. Only once its final
1434 * index is known is the updated memslot copied into its position in the array.
1436 * - When deleting a memslot, the deleted memslot simply needs to be moved to
1437 * the end of the array.
1439 * - When creating a memslot, the algorithm "inserts" the new memslot at the
1440 * end of the array and then it forward to its correct location.
1442 * - When moving a memslot, the algorithm first moves the updated memslot
1443 * backward to handle the scenario where the memslot's GFN was changed to a
1444 * lower value. update_memslots() then falls through and runs the same flow
1445 * as creating a memslot to move the memslot forward to handle the scenario
1446 * where its GFN was changed to a higher value.
1448 * Note, slots are sorted from highest->lowest instead of lowest->highest for
1449 * historical reasons. Originally, invalid memslots where denoted by having
1450 * GFN=0, thus sorting from highest->lowest naturally sorted invalid memslots
1451 * to the end of the array. The current algorithm uses dedicated logic to
1452 * delete a memslot and thus does not rely on invalid memslots having GFN=0.
1454 * The other historical motiviation for highest->lowest was to improve the
1455 * performance of memslot lookup. KVM originally used a linear search starting
1456 * at memslots[0]. On x86, the largest memslot usually has one of the highest,
1457 * if not *the* highest, GFN, as the bulk of the guest's RAM is located in a
1458 * single memslot above the 4gb boundary. As the largest memslot is also the
1459 * most likely to be referenced, sorting it to the front of the array was
1460 * advantageous. The current binary search starts from the middle of the array
1461 * and uses an LRU pointer to improve performance for all memslots and GFNs.
1463 static void update_memslots(struct kvm_memslots *slots,
1464 struct kvm_memory_slot *memslot,
1465 enum kvm_mr_change change)
1469 if (change == KVM_MR_DELETE) {
1470 kvm_memslot_delete(slots, memslot);
1472 if (change == KVM_MR_CREATE)
1473 i = kvm_memslot_insert_back(slots);
1475 i = kvm_memslot_move_backward(slots, memslot);
1476 i = kvm_memslot_move_forward(slots, memslot, i);
1479 * Copy the memslot to its new position in memslots and update
1480 * its index accordingly.
1482 slots->memslots[i] = *memslot;
1483 slots->id_to_index[memslot->id] = i;
1487 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
1489 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
1491 #ifdef __KVM_HAVE_READONLY_MEM
1492 valid_flags |= KVM_MEM_READONLY;
1495 if (mem->flags & ~valid_flags)
1501 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
1502 int as_id, struct kvm_memslots *slots)
1504 struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id);
1505 u64 gen = old_memslots->generation;
1507 WARN_ON(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS);
1508 slots->generation = gen | KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
1511 * Do not store the new memslots while there are invalidations in
1512 * progress, otherwise the locking in invalidate_range_start and
1513 * invalidate_range_end will be unbalanced.
1515 spin_lock(&kvm->mn_invalidate_lock);
1516 prepare_to_rcuwait(&kvm->mn_memslots_update_rcuwait);
1517 while (kvm->mn_active_invalidate_count) {
1518 set_current_state(TASK_UNINTERRUPTIBLE);
1519 spin_unlock(&kvm->mn_invalidate_lock);
1521 spin_lock(&kvm->mn_invalidate_lock);
1523 finish_rcuwait(&kvm->mn_memslots_update_rcuwait);
1524 rcu_assign_pointer(kvm->memslots[as_id], slots);
1525 spin_unlock(&kvm->mn_invalidate_lock);
1528 * Acquired in kvm_set_memslot. Must be released before synchronize
1529 * SRCU below in order to avoid deadlock with another thread
1530 * acquiring the slots_arch_lock in an srcu critical section.
1532 mutex_unlock(&kvm->slots_arch_lock);
1534 synchronize_srcu_expedited(&kvm->srcu);
1537 * Increment the new memslot generation a second time, dropping the
1538 * update in-progress flag and incrementing the generation based on
1539 * the number of address spaces. This provides a unique and easily
1540 * identifiable generation number while the memslots are in flux.
1542 gen = slots->generation & ~KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
1545 * Generations must be unique even across address spaces. We do not need
1546 * a global counter for that, instead the generation space is evenly split
1547 * across address spaces. For example, with two address spaces, address
1548 * space 0 will use generations 0, 2, 4, ... while address space 1 will
1549 * use generations 1, 3, 5, ...
1551 gen += KVM_ADDRESS_SPACE_NUM;
1553 kvm_arch_memslots_updated(kvm, gen);
1555 slots->generation = gen;
1557 return old_memslots;
1560 static size_t kvm_memslots_size(int slots)
1562 return sizeof(struct kvm_memslots) +
1563 (sizeof(struct kvm_memory_slot) * slots);
1566 static void kvm_copy_memslots(struct kvm_memslots *to,
1567 struct kvm_memslots *from)
1569 memcpy(to, from, kvm_memslots_size(from->used_slots));
1573 * Note, at a minimum, the current number of used slots must be allocated, even
1574 * when deleting a memslot, as we need a complete duplicate of the memslots for
1575 * use when invalidating a memslot prior to deleting/moving the memslot.
1577 static struct kvm_memslots *kvm_dup_memslots(struct kvm_memslots *old,
1578 enum kvm_mr_change change)
1580 struct kvm_memslots *slots;
1583 if (change == KVM_MR_CREATE)
1584 new_size = kvm_memslots_size(old->used_slots + 1);
1586 new_size = kvm_memslots_size(old->used_slots);
1588 slots = kvzalloc(new_size, GFP_KERNEL_ACCOUNT);
1590 kvm_copy_memslots(slots, old);
1595 static int kvm_set_memslot(struct kvm *kvm,
1596 const struct kvm_userspace_memory_region *mem,
1597 struct kvm_memory_slot *new, int as_id,
1598 enum kvm_mr_change change)
1600 struct kvm_memory_slot *slot, old;
1601 struct kvm_memslots *slots;
1605 * Released in install_new_memslots.
1607 * Must be held from before the current memslots are copied until
1608 * after the new memslots are installed with rcu_assign_pointer,
1609 * then released before the synchronize srcu in install_new_memslots.
1611 * When modifying memslots outside of the slots_lock, must be held
1612 * before reading the pointer to the current memslots until after all
1613 * changes to those memslots are complete.
1615 * These rules ensure that installing new memslots does not lose
1616 * changes made to the previous memslots.
1618 mutex_lock(&kvm->slots_arch_lock);
1620 slots = kvm_dup_memslots(__kvm_memslots(kvm, as_id), change);
1622 mutex_unlock(&kvm->slots_arch_lock);
1626 if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) {
1628 * Note, the INVALID flag needs to be in the appropriate entry
1629 * in the freshly allocated memslots, not in @old or @new.
1631 slot = id_to_memslot(slots, new->id);
1632 slot->flags |= KVM_MEMSLOT_INVALID;
1635 * We can re-use the memory from the old memslots.
1636 * It will be overwritten with a copy of the new memslots
1637 * after reacquiring the slots_arch_lock below.
1639 slots = install_new_memslots(kvm, as_id, slots);
1641 /* From this point no new shadow pages pointing to a deleted,
1642 * or moved, memslot will be created.
1644 * validation of sp->gfn happens in:
1645 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1646 * - kvm_is_visible_gfn (mmu_check_root)
1648 kvm_arch_flush_shadow_memslot(kvm, slot);
1649 kvm_arch_guest_memory_reclaimed(kvm);
1651 /* Released in install_new_memslots. */
1652 mutex_lock(&kvm->slots_arch_lock);
1655 * The arch-specific fields of the memslots could have changed
1656 * between releasing the slots_arch_lock in
1657 * install_new_memslots and here, so get a fresh copy of the
1660 kvm_copy_memslots(slots, __kvm_memslots(kvm, as_id));
1664 * Make a full copy of the old memslot, the pointer will become stale
1665 * when the memslots are re-sorted by update_memslots(), and the old
1666 * memslot needs to be referenced after calling update_memslots(), e.g.
1667 * to free its resources and for arch specific behavior. This needs to
1668 * happen *after* (re)acquiring slots_arch_lock.
1670 slot = id_to_memslot(slots, new->id);
1674 WARN_ON_ONCE(change != KVM_MR_CREATE);
1675 memset(&old, 0, sizeof(old));
1680 /* Copy the arch-specific data, again after (re)acquiring slots_arch_lock. */
1681 memcpy(&new->arch, &old.arch, sizeof(old.arch));
1683 r = kvm_arch_prepare_memory_region(kvm, new, mem, change);
1687 update_memslots(slots, new, change);
1688 slots = install_new_memslots(kvm, as_id, slots);
1690 kvm_arch_commit_memory_region(kvm, mem, &old, new, change);
1692 /* Free the old memslot's metadata. Note, this is the full copy!!! */
1693 if (change == KVM_MR_DELETE)
1694 kvm_free_memslot(kvm, &old);
1700 if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) {
1701 slot = id_to_memslot(slots, new->id);
1702 slot->flags &= ~KVM_MEMSLOT_INVALID;
1703 slots = install_new_memslots(kvm, as_id, slots);
1705 mutex_unlock(&kvm->slots_arch_lock);
1711 static int kvm_delete_memslot(struct kvm *kvm,
1712 const struct kvm_userspace_memory_region *mem,
1713 struct kvm_memory_slot *old, int as_id)
1715 struct kvm_memory_slot new;
1720 memset(&new, 0, sizeof(new));
1723 * This is only for debugging purpose; it should never be referenced
1724 * for a removed memslot.
1728 return kvm_set_memslot(kvm, mem, &new, as_id, KVM_MR_DELETE);
1732 * Allocate some memory and give it an address in the guest physical address
1735 * Discontiguous memory is allowed, mostly for framebuffers.
1737 * Must be called holding kvm->slots_lock for write.
1739 int __kvm_set_memory_region(struct kvm *kvm,
1740 const struct kvm_userspace_memory_region *mem)
1742 struct kvm_memory_slot old, new;
1743 struct kvm_memory_slot *tmp;
1744 enum kvm_mr_change change;
1748 r = check_memory_region_flags(mem);
1752 as_id = mem->slot >> 16;
1753 id = (u16)mem->slot;
1755 /* General sanity checks */
1756 if ((mem->memory_size & (PAGE_SIZE - 1)) ||
1757 (mem->memory_size != (unsigned long)mem->memory_size))
1759 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
1761 /* We can read the guest memory with __xxx_user() later on. */
1762 if ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
1763 (mem->userspace_addr != untagged_addr(mem->userspace_addr)) ||
1764 !access_ok((void __user *)(unsigned long)mem->userspace_addr,
1767 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
1769 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
1773 * Make a full copy of the old memslot, the pointer will become stale
1774 * when the memslots are re-sorted by update_memslots(), and the old
1775 * memslot needs to be referenced after calling update_memslots(), e.g.
1776 * to free its resources and for arch specific behavior.
1778 tmp = id_to_memslot(__kvm_memslots(kvm, as_id), id);
1783 memset(&old, 0, sizeof(old));
1787 if (!mem->memory_size)
1788 return kvm_delete_memslot(kvm, mem, &old, as_id);
1792 new.base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
1793 new.npages = mem->memory_size >> PAGE_SHIFT;
1794 new.flags = mem->flags;
1795 new.userspace_addr = mem->userspace_addr;
1797 if (new.npages > KVM_MEM_MAX_NR_PAGES)
1801 change = KVM_MR_CREATE;
1802 new.dirty_bitmap = NULL;
1803 } else { /* Modify an existing slot. */
1804 if ((new.userspace_addr != old.userspace_addr) ||
1805 (new.npages != old.npages) ||
1806 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
1809 if (new.base_gfn != old.base_gfn)
1810 change = KVM_MR_MOVE;
1811 else if (new.flags != old.flags)
1812 change = KVM_MR_FLAGS_ONLY;
1813 else /* Nothing to change. */
1816 /* Copy dirty_bitmap from the current memslot. */
1817 new.dirty_bitmap = old.dirty_bitmap;
1820 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
1821 /* Check for overlaps */
1822 kvm_for_each_memslot(tmp, __kvm_memslots(kvm, as_id)) {
1825 if (!((new.base_gfn + new.npages <= tmp->base_gfn) ||
1826 (new.base_gfn >= tmp->base_gfn + tmp->npages)))
1831 /* Allocate/free page dirty bitmap as needed */
1832 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
1833 new.dirty_bitmap = NULL;
1834 else if (!new.dirty_bitmap && !kvm->dirty_ring_size) {
1835 r = kvm_alloc_dirty_bitmap(&new);
1839 if (kvm_dirty_log_manual_protect_and_init_set(kvm))
1840 bitmap_set(new.dirty_bitmap, 0, new.npages);
1843 r = kvm_set_memslot(kvm, mem, &new, as_id, change);
1847 if (old.dirty_bitmap && !new.dirty_bitmap)
1848 kvm_destroy_dirty_bitmap(&old);
1852 if (new.dirty_bitmap && !old.dirty_bitmap)
1853 kvm_destroy_dirty_bitmap(&new);
1856 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1858 int kvm_set_memory_region(struct kvm *kvm,
1859 const struct kvm_userspace_memory_region *mem)
1863 mutex_lock(&kvm->slots_lock);
1864 r = __kvm_set_memory_region(kvm, mem);
1865 mutex_unlock(&kvm->slots_lock);
1868 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1870 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1871 struct kvm_userspace_memory_region *mem)
1873 if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1876 return kvm_set_memory_region(kvm, mem);
1879 #ifndef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1881 * kvm_get_dirty_log - get a snapshot of dirty pages
1882 * @kvm: pointer to kvm instance
1883 * @log: slot id and address to which we copy the log
1884 * @is_dirty: set to '1' if any dirty pages were found
1885 * @memslot: set to the associated memslot, always valid on success
1887 int kvm_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log,
1888 int *is_dirty, struct kvm_memory_slot **memslot)
1890 struct kvm_memslots *slots;
1893 unsigned long any = 0;
1895 /* Dirty ring tracking is exclusive to dirty log tracking */
1896 if (kvm->dirty_ring_size)
1902 as_id = log->slot >> 16;
1903 id = (u16)log->slot;
1904 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1907 slots = __kvm_memslots(kvm, as_id);
1908 *memslot = id_to_memslot(slots, id);
1909 if (!(*memslot) || !(*memslot)->dirty_bitmap)
1912 kvm_arch_sync_dirty_log(kvm, *memslot);
1914 n = kvm_dirty_bitmap_bytes(*memslot);
1916 for (i = 0; !any && i < n/sizeof(long); ++i)
1917 any = (*memslot)->dirty_bitmap[i];
1919 if (copy_to_user(log->dirty_bitmap, (*memslot)->dirty_bitmap, n))
1926 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1928 #else /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
1930 * kvm_get_dirty_log_protect - get a snapshot of dirty pages
1931 * and reenable dirty page tracking for the corresponding pages.
1932 * @kvm: pointer to kvm instance
1933 * @log: slot id and address to which we copy the log
1935 * We need to keep it in mind that VCPU threads can write to the bitmap
1936 * concurrently. So, to avoid losing track of dirty pages we keep the
1939 * 1. Take a snapshot of the bit and clear it if needed.
1940 * 2. Write protect the corresponding page.
1941 * 3. Copy the snapshot to the userspace.
1942 * 4. Upon return caller flushes TLB's if needed.
1944 * Between 2 and 4, the guest may write to the page using the remaining TLB
1945 * entry. This is not a problem because the page is reported dirty using
1946 * the snapshot taken before and step 4 ensures that writes done after
1947 * exiting to userspace will be logged for the next call.
1950 static int kvm_get_dirty_log_protect(struct kvm *kvm, struct kvm_dirty_log *log)
1952 struct kvm_memslots *slots;
1953 struct kvm_memory_slot *memslot;
1956 unsigned long *dirty_bitmap;
1957 unsigned long *dirty_bitmap_buffer;
1960 /* Dirty ring tracking is exclusive to dirty log tracking */
1961 if (kvm->dirty_ring_size)
1964 as_id = log->slot >> 16;
1965 id = (u16)log->slot;
1966 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1969 slots = __kvm_memslots(kvm, as_id);
1970 memslot = id_to_memslot(slots, id);
1971 if (!memslot || !memslot->dirty_bitmap)
1974 dirty_bitmap = memslot->dirty_bitmap;
1976 kvm_arch_sync_dirty_log(kvm, memslot);
1978 n = kvm_dirty_bitmap_bytes(memslot);
1980 if (kvm->manual_dirty_log_protect) {
1982 * Unlike kvm_get_dirty_log, we always return false in *flush,
1983 * because no flush is needed until KVM_CLEAR_DIRTY_LOG. There
1984 * is some code duplication between this function and
1985 * kvm_get_dirty_log, but hopefully all architecture
1986 * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log
1987 * can be eliminated.
1989 dirty_bitmap_buffer = dirty_bitmap;
1991 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
1992 memset(dirty_bitmap_buffer, 0, n);
1995 for (i = 0; i < n / sizeof(long); i++) {
1999 if (!dirty_bitmap[i])
2003 mask = xchg(&dirty_bitmap[i], 0);
2004 dirty_bitmap_buffer[i] = mask;
2006 offset = i * BITS_PER_LONG;
2007 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
2010 KVM_MMU_UNLOCK(kvm);
2014 kvm_arch_flush_remote_tlbs_memslot(kvm, memslot);
2016 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
2023 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
2024 * @kvm: kvm instance
2025 * @log: slot id and address to which we copy the log
2027 * Steps 1-4 below provide general overview of dirty page logging. See
2028 * kvm_get_dirty_log_protect() function description for additional details.
2030 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
2031 * always flush the TLB (step 4) even if previous step failed and the dirty
2032 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
2033 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
2034 * writes will be marked dirty for next log read.
2036 * 1. Take a snapshot of the bit and clear it if needed.
2037 * 2. Write protect the corresponding page.
2038 * 3. Copy the snapshot to the userspace.
2039 * 4. Flush TLB's if needed.
2041 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
2042 struct kvm_dirty_log *log)
2046 mutex_lock(&kvm->slots_lock);
2048 r = kvm_get_dirty_log_protect(kvm, log);
2050 mutex_unlock(&kvm->slots_lock);
2055 * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap
2056 * and reenable dirty page tracking for the corresponding pages.
2057 * @kvm: pointer to kvm instance
2058 * @log: slot id and address from which to fetch the bitmap of dirty pages
2060 static int kvm_clear_dirty_log_protect(struct kvm *kvm,
2061 struct kvm_clear_dirty_log *log)
2063 struct kvm_memslots *slots;
2064 struct kvm_memory_slot *memslot;
2068 unsigned long *dirty_bitmap;
2069 unsigned long *dirty_bitmap_buffer;
2072 /* Dirty ring tracking is exclusive to dirty log tracking */
2073 if (kvm->dirty_ring_size)
2076 as_id = log->slot >> 16;
2077 id = (u16)log->slot;
2078 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
2081 if (log->first_page & 63)
2084 slots = __kvm_memslots(kvm, as_id);
2085 memslot = id_to_memslot(slots, id);
2086 if (!memslot || !memslot->dirty_bitmap)
2089 dirty_bitmap = memslot->dirty_bitmap;
2091 n = ALIGN(log->num_pages, BITS_PER_LONG) / 8;
2093 if (log->first_page > memslot->npages ||
2094 log->num_pages > memslot->npages - log->first_page ||
2095 (log->num_pages < memslot->npages - log->first_page && (log->num_pages & 63)))
2098 kvm_arch_sync_dirty_log(kvm, memslot);
2101 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
2102 if (copy_from_user(dirty_bitmap_buffer, log->dirty_bitmap, n))
2106 for (offset = log->first_page, i = offset / BITS_PER_LONG,
2107 n = DIV_ROUND_UP(log->num_pages, BITS_PER_LONG); n--;
2108 i++, offset += BITS_PER_LONG) {
2109 unsigned long mask = *dirty_bitmap_buffer++;
2110 atomic_long_t *p = (atomic_long_t *) &dirty_bitmap[i];
2114 mask &= atomic_long_fetch_andnot(mask, p);
2117 * mask contains the bits that really have been cleared. This
2118 * never includes any bits beyond the length of the memslot (if
2119 * the length is not aligned to 64 pages), therefore it is not
2120 * a problem if userspace sets them in log->dirty_bitmap.
2124 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
2128 KVM_MMU_UNLOCK(kvm);
2131 kvm_arch_flush_remote_tlbs_memslot(kvm, memslot);
2136 static int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm,
2137 struct kvm_clear_dirty_log *log)
2141 mutex_lock(&kvm->slots_lock);
2143 r = kvm_clear_dirty_log_protect(kvm, log);
2145 mutex_unlock(&kvm->slots_lock);
2148 #endif /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
2150 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
2152 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
2154 EXPORT_SYMBOL_GPL(gfn_to_memslot);
2156 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
2158 struct kvm_memslots *slots = kvm_vcpu_memslots(vcpu);
2159 struct kvm_memory_slot *slot;
2162 slot = try_get_memslot(slots, vcpu->last_used_slot, gfn);
2167 * Fall back to searching all memslots. We purposely use
2168 * search_memslots() instead of __gfn_to_memslot() to avoid
2169 * thrashing the VM-wide last_used_index in kvm_memslots.
2171 slot = search_memslots(slots, gfn, &slot_index);
2173 vcpu->last_used_slot = slot_index;
2180 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
2182 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
2184 return kvm_is_visible_memslot(memslot);
2186 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
2188 bool kvm_vcpu_is_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
2190 struct kvm_memory_slot *memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2192 return kvm_is_visible_memslot(memslot);
2194 EXPORT_SYMBOL_GPL(kvm_vcpu_is_visible_gfn);
2196 unsigned long kvm_host_page_size(struct kvm_vcpu *vcpu, gfn_t gfn)
2198 struct vm_area_struct *vma;
2199 unsigned long addr, size;
2203 addr = kvm_vcpu_gfn_to_hva_prot(vcpu, gfn, NULL);
2204 if (kvm_is_error_hva(addr))
2207 mmap_read_lock(current->mm);
2208 vma = find_vma(current->mm, addr);
2212 size = vma_kernel_pagesize(vma);
2215 mmap_read_unlock(current->mm);
2220 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
2222 return slot->flags & KVM_MEM_READONLY;
2225 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
2226 gfn_t *nr_pages, bool write)
2228 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
2229 return KVM_HVA_ERR_BAD;
2231 if (memslot_is_readonly(slot) && write)
2232 return KVM_HVA_ERR_RO_BAD;
2235 *nr_pages = slot->npages - (gfn - slot->base_gfn);
2237 return __gfn_to_hva_memslot(slot, gfn);
2240 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
2243 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
2246 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
2249 return gfn_to_hva_many(slot, gfn, NULL);
2251 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
2253 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
2255 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
2257 EXPORT_SYMBOL_GPL(gfn_to_hva);
2259 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
2261 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
2263 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
2266 * Return the hva of a @gfn and the R/W attribute if possible.
2268 * @slot: the kvm_memory_slot which contains @gfn
2269 * @gfn: the gfn to be translated
2270 * @writable: used to return the read/write attribute of the @slot if the hva
2271 * is valid and @writable is not NULL
2273 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
2274 gfn_t gfn, bool *writable)
2276 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
2278 if (!kvm_is_error_hva(hva) && writable)
2279 *writable = !memslot_is_readonly(slot);
2284 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
2286 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2288 return gfn_to_hva_memslot_prot(slot, gfn, writable);
2291 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
2293 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2295 return gfn_to_hva_memslot_prot(slot, gfn, writable);
2298 static inline int check_user_page_hwpoison(unsigned long addr)
2300 int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
2302 rc = get_user_pages(addr, 1, flags, NULL, NULL);
2303 return rc == -EHWPOISON;
2307 * The fast path to get the writable pfn which will be stored in @pfn,
2308 * true indicates success, otherwise false is returned. It's also the
2309 * only part that runs if we can in atomic context.
2311 static bool hva_to_pfn_fast(unsigned long addr, bool write_fault,
2312 bool *writable, kvm_pfn_t *pfn)
2314 struct page *page[1];
2317 * Fast pin a writable pfn only if it is a write fault request
2318 * or the caller allows to map a writable pfn for a read fault
2321 if (!(write_fault || writable))
2324 if (get_user_page_fast_only(addr, FOLL_WRITE, page)) {
2325 *pfn = page_to_pfn(page[0]);
2336 * The slow path to get the pfn of the specified host virtual address,
2337 * 1 indicates success, -errno is returned if error is detected.
2339 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
2340 bool *writable, kvm_pfn_t *pfn)
2342 unsigned int flags = FOLL_HWPOISON;
2349 *writable = write_fault;
2352 flags |= FOLL_WRITE;
2354 flags |= FOLL_NOWAIT;
2356 npages = get_user_pages_unlocked(addr, 1, &page, flags);
2360 /* map read fault as writable if possible */
2361 if (unlikely(!write_fault) && writable) {
2364 if (get_user_page_fast_only(addr, FOLL_WRITE, &wpage)) {
2370 *pfn = page_to_pfn(page);
2374 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
2376 if (unlikely(!(vma->vm_flags & VM_READ)))
2379 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
2385 static int kvm_try_get_pfn(kvm_pfn_t pfn)
2387 if (kvm_is_reserved_pfn(pfn))
2389 return get_page_unless_zero(pfn_to_page(pfn));
2392 static int hva_to_pfn_remapped(struct vm_area_struct *vma,
2393 unsigned long addr, bool *async,
2394 bool write_fault, bool *writable,
2402 r = follow_pte(vma->vm_mm, addr, &ptep, &ptl);
2405 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
2406 * not call the fault handler, so do it here.
2408 bool unlocked = false;
2409 r = fixup_user_fault(current->mm, addr,
2410 (write_fault ? FAULT_FLAG_WRITE : 0),
2417 r = follow_pte(vma->vm_mm, addr, &ptep, &ptl);
2422 if (write_fault && !pte_write(*ptep)) {
2423 pfn = KVM_PFN_ERR_RO_FAULT;
2428 *writable = pte_write(*ptep);
2429 pfn = pte_pfn(*ptep);
2432 * Get a reference here because callers of *hva_to_pfn* and
2433 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
2434 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
2435 * set, but the kvm_try_get_pfn/kvm_release_pfn_clean pair will
2436 * simply do nothing for reserved pfns.
2438 * Whoever called remap_pfn_range is also going to call e.g.
2439 * unmap_mapping_range before the underlying pages are freed,
2440 * causing a call to our MMU notifier.
2442 * Certain IO or PFNMAP mappings can be backed with valid
2443 * struct pages, but be allocated without refcounting e.g.,
2444 * tail pages of non-compound higher order allocations, which
2445 * would then underflow the refcount when the caller does the
2446 * required put_page. Don't allow those pages here.
2448 if (!kvm_try_get_pfn(pfn))
2452 pte_unmap_unlock(ptep, ptl);
2459 * Pin guest page in memory and return its pfn.
2460 * @addr: host virtual address which maps memory to the guest
2461 * @atomic: whether this function can sleep
2462 * @async: whether this function need to wait IO complete if the
2463 * host page is not in the memory
2464 * @write_fault: whether we should get a writable host page
2465 * @writable: whether it allows to map a writable host page for !@write_fault
2467 * The function will map a writable host page for these two cases:
2468 * 1): @write_fault = true
2469 * 2): @write_fault = false && @writable, @writable will tell the caller
2470 * whether the mapping is writable.
2472 static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
2473 bool write_fault, bool *writable)
2475 struct vm_area_struct *vma;
2479 /* we can do it either atomically or asynchronously, not both */
2480 BUG_ON(atomic && async);
2482 if (hva_to_pfn_fast(addr, write_fault, writable, &pfn))
2486 return KVM_PFN_ERR_FAULT;
2488 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
2492 mmap_read_lock(current->mm);
2493 if (npages == -EHWPOISON ||
2494 (!async && check_user_page_hwpoison(addr))) {
2495 pfn = KVM_PFN_ERR_HWPOISON;
2500 vma = vma_lookup(current->mm, addr);
2503 pfn = KVM_PFN_ERR_FAULT;
2504 else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
2505 r = hva_to_pfn_remapped(vma, addr, async, write_fault, writable, &pfn);
2509 pfn = KVM_PFN_ERR_FAULT;
2511 if (async && vma_is_valid(vma, write_fault))
2513 pfn = KVM_PFN_ERR_FAULT;
2516 mmap_read_unlock(current->mm);
2520 kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn,
2521 bool atomic, bool *async, bool write_fault,
2522 bool *writable, hva_t *hva)
2524 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
2529 if (addr == KVM_HVA_ERR_RO_BAD) {
2532 return KVM_PFN_ERR_RO_FAULT;
2535 if (kvm_is_error_hva(addr)) {
2538 return KVM_PFN_NOSLOT;
2541 /* Do not map writable pfn in the readonly memslot. */
2542 if (writable && memslot_is_readonly(slot)) {
2547 return hva_to_pfn(addr, atomic, async, write_fault,
2550 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
2552 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
2555 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
2556 write_fault, writable, NULL);
2558 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
2560 kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
2562 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL, NULL);
2564 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
2566 kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
2568 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL, NULL);
2570 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
2572 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
2574 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
2576 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
2578 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
2580 return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
2582 EXPORT_SYMBOL_GPL(gfn_to_pfn);
2584 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
2586 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
2588 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
2590 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
2591 struct page **pages, int nr_pages)
2596 addr = gfn_to_hva_many(slot, gfn, &entry);
2597 if (kvm_is_error_hva(addr))
2600 if (entry < nr_pages)
2603 return get_user_pages_fast_only(addr, nr_pages, FOLL_WRITE, pages);
2605 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
2607 static struct page *kvm_pfn_to_page(kvm_pfn_t pfn)
2609 if (is_error_noslot_pfn(pfn))
2610 return KVM_ERR_PTR_BAD_PAGE;
2612 if (kvm_is_reserved_pfn(pfn)) {
2614 return KVM_ERR_PTR_BAD_PAGE;
2617 return pfn_to_page(pfn);
2620 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
2624 pfn = gfn_to_pfn(kvm, gfn);
2626 return kvm_pfn_to_page(pfn);
2628 EXPORT_SYMBOL_GPL(gfn_to_page);
2630 void kvm_release_pfn(kvm_pfn_t pfn, bool dirty, struct gfn_to_pfn_cache *cache)
2636 cache->pfn = cache->gfn = 0;
2639 kvm_release_pfn_dirty(pfn);
2641 kvm_release_pfn_clean(pfn);
2644 static void kvm_cache_gfn_to_pfn(struct kvm_memory_slot *slot, gfn_t gfn,
2645 struct gfn_to_pfn_cache *cache, u64 gen)
2647 kvm_release_pfn(cache->pfn, cache->dirty, cache);
2649 cache->pfn = gfn_to_pfn_memslot(slot, gfn);
2651 cache->dirty = false;
2652 cache->generation = gen;
2655 static int __kvm_map_gfn(struct kvm_memslots *slots, gfn_t gfn,
2656 struct kvm_host_map *map,
2657 struct gfn_to_pfn_cache *cache,
2662 struct page *page = KVM_UNMAPPED_PAGE;
2663 struct kvm_memory_slot *slot = __gfn_to_memslot(slots, gfn);
2664 u64 gen = slots->generation;
2670 if (!cache->pfn || cache->gfn != gfn ||
2671 cache->generation != gen) {
2674 kvm_cache_gfn_to_pfn(slot, gfn, cache, gen);
2680 pfn = gfn_to_pfn_memslot(slot, gfn);
2682 if (is_error_noslot_pfn(pfn))
2685 if (pfn_valid(pfn)) {
2686 page = pfn_to_page(pfn);
2688 hva = kmap_atomic(page);
2691 #ifdef CONFIG_HAS_IOMEM
2692 } else if (!atomic) {
2693 hva = memremap(pfn_to_hpa(pfn), PAGE_SIZE, MEMREMAP_WB);
2710 int kvm_map_gfn(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map,
2711 struct gfn_to_pfn_cache *cache, bool atomic)
2713 return __kvm_map_gfn(kvm_memslots(vcpu->kvm), gfn, map,
2716 EXPORT_SYMBOL_GPL(kvm_map_gfn);
2718 int kvm_vcpu_map(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map)
2720 return __kvm_map_gfn(kvm_vcpu_memslots(vcpu), gfn, map,
2723 EXPORT_SYMBOL_GPL(kvm_vcpu_map);
2725 static void __kvm_unmap_gfn(struct kvm *kvm,
2726 struct kvm_memory_slot *memslot,
2727 struct kvm_host_map *map,
2728 struct gfn_to_pfn_cache *cache,
2729 bool dirty, bool atomic)
2737 if (map->page != KVM_UNMAPPED_PAGE) {
2739 kunmap_atomic(map->hva);
2743 #ifdef CONFIG_HAS_IOMEM
2747 WARN_ONCE(1, "Unexpected unmapping in atomic context");
2751 mark_page_dirty_in_slot(kvm, memslot, map->gfn);
2754 cache->dirty |= dirty;
2756 kvm_release_pfn(map->pfn, dirty, NULL);
2762 int kvm_unmap_gfn(struct kvm_vcpu *vcpu, struct kvm_host_map *map,
2763 struct gfn_to_pfn_cache *cache, bool dirty, bool atomic)
2765 __kvm_unmap_gfn(vcpu->kvm, gfn_to_memslot(vcpu->kvm, map->gfn), map,
2766 cache, dirty, atomic);
2769 EXPORT_SYMBOL_GPL(kvm_unmap_gfn);
2771 void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map, bool dirty)
2773 __kvm_unmap_gfn(vcpu->kvm, kvm_vcpu_gfn_to_memslot(vcpu, map->gfn),
2774 map, NULL, dirty, false);
2776 EXPORT_SYMBOL_GPL(kvm_vcpu_unmap);
2778 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
2782 pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
2784 return kvm_pfn_to_page(pfn);
2786 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
2788 void kvm_release_page_clean(struct page *page)
2790 WARN_ON(is_error_page(page));
2792 kvm_release_pfn_clean(page_to_pfn(page));
2794 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
2796 void kvm_release_pfn_clean(kvm_pfn_t pfn)
2798 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
2799 put_page(pfn_to_page(pfn));
2801 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
2803 void kvm_release_page_dirty(struct page *page)
2805 WARN_ON(is_error_page(page));
2807 kvm_release_pfn_dirty(page_to_pfn(page));
2809 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
2811 void kvm_release_pfn_dirty(kvm_pfn_t pfn)
2813 kvm_set_pfn_dirty(pfn);
2814 kvm_release_pfn_clean(pfn);
2816 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
2818 static bool kvm_is_ad_tracked_pfn(kvm_pfn_t pfn)
2820 if (!pfn_valid(pfn))
2824 * Per page-flags.h, pages tagged PG_reserved "should in general not be
2825 * touched (e.g. set dirty) except by its owner".
2827 return !PageReserved(pfn_to_page(pfn));
2830 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
2832 if (kvm_is_ad_tracked_pfn(pfn))
2833 SetPageDirty(pfn_to_page(pfn));
2835 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
2837 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
2839 if (kvm_is_ad_tracked_pfn(pfn))
2840 mark_page_accessed(pfn_to_page(pfn));
2842 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
2844 static int next_segment(unsigned long len, int offset)
2846 if (len > PAGE_SIZE - offset)
2847 return PAGE_SIZE - offset;
2852 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
2853 void *data, int offset, int len)
2858 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
2859 if (kvm_is_error_hva(addr))
2861 r = __copy_from_user(data, (void __user *)addr + offset, len);
2867 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
2870 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2872 return __kvm_read_guest_page(slot, gfn, data, offset, len);
2874 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
2876 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
2877 int offset, int len)
2879 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2881 return __kvm_read_guest_page(slot, gfn, data, offset, len);
2883 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
2885 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
2887 gfn_t gfn = gpa >> PAGE_SHIFT;
2889 int offset = offset_in_page(gpa);
2892 while ((seg = next_segment(len, offset)) != 0) {
2893 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
2903 EXPORT_SYMBOL_GPL(kvm_read_guest);
2905 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
2907 gfn_t gfn = gpa >> PAGE_SHIFT;
2909 int offset = offset_in_page(gpa);
2912 while ((seg = next_segment(len, offset)) != 0) {
2913 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
2923 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
2925 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
2926 void *data, int offset, unsigned long len)
2931 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
2932 if (kvm_is_error_hva(addr))
2934 pagefault_disable();
2935 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
2942 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
2943 void *data, unsigned long len)
2945 gfn_t gfn = gpa >> PAGE_SHIFT;
2946 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2947 int offset = offset_in_page(gpa);
2949 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
2951 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
2953 static int __kvm_write_guest_page(struct kvm *kvm,
2954 struct kvm_memory_slot *memslot, gfn_t gfn,
2955 const void *data, int offset, int len)
2960 addr = gfn_to_hva_memslot(memslot, gfn);
2961 if (kvm_is_error_hva(addr))
2963 r = __copy_to_user((void __user *)addr + offset, data, len);
2966 mark_page_dirty_in_slot(kvm, memslot, gfn);
2970 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
2971 const void *data, int offset, int len)
2973 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2975 return __kvm_write_guest_page(kvm, slot, gfn, data, offset, len);
2977 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
2979 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
2980 const void *data, int offset, int len)
2982 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2984 return __kvm_write_guest_page(vcpu->kvm, slot, gfn, data, offset, len);
2986 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
2988 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
2991 gfn_t gfn = gpa >> PAGE_SHIFT;
2993 int offset = offset_in_page(gpa);
2996 while ((seg = next_segment(len, offset)) != 0) {
2997 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
3007 EXPORT_SYMBOL_GPL(kvm_write_guest);
3009 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
3012 gfn_t gfn = gpa >> PAGE_SHIFT;
3014 int offset = offset_in_page(gpa);
3017 while ((seg = next_segment(len, offset)) != 0) {
3018 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
3028 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
3030 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
3031 struct gfn_to_hva_cache *ghc,
3032 gpa_t gpa, unsigned long len)
3034 int offset = offset_in_page(gpa);
3035 gfn_t start_gfn = gpa >> PAGE_SHIFT;
3036 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
3037 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
3038 gfn_t nr_pages_avail;
3040 /* Update ghc->generation before performing any error checks. */
3041 ghc->generation = slots->generation;
3043 if (start_gfn > end_gfn) {
3044 ghc->hva = KVM_HVA_ERR_BAD;
3049 * If the requested region crosses two memslots, we still
3050 * verify that the entire region is valid here.
3052 for ( ; start_gfn <= end_gfn; start_gfn += nr_pages_avail) {
3053 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
3054 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
3056 if (kvm_is_error_hva(ghc->hva))
3060 /* Use the slow path for cross page reads and writes. */
3061 if (nr_pages_needed == 1)
3064 ghc->memslot = NULL;
3071 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3072 gpa_t gpa, unsigned long len)
3074 struct kvm_memslots *slots = kvm_memslots(kvm);
3075 return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len);
3077 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
3079 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3080 void *data, unsigned int offset,
3083 struct kvm_memslots *slots = kvm_memslots(kvm);
3085 gpa_t gpa = ghc->gpa + offset;
3087 if (WARN_ON_ONCE(len + offset > ghc->len))
3090 if (slots->generation != ghc->generation) {
3091 if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len))
3095 if (kvm_is_error_hva(ghc->hva))
3098 if (unlikely(!ghc->memslot))
3099 return kvm_write_guest(kvm, gpa, data, len);
3101 r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
3104 mark_page_dirty_in_slot(kvm, ghc->memslot, gpa >> PAGE_SHIFT);
3108 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached);
3110 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3111 void *data, unsigned long len)
3113 return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len);
3115 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
3117 int kvm_read_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3118 void *data, unsigned int offset,
3121 struct kvm_memslots *slots = kvm_memslots(kvm);
3123 gpa_t gpa = ghc->gpa + offset;
3125 if (WARN_ON_ONCE(len + offset > ghc->len))
3128 if (slots->generation != ghc->generation) {
3129 if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len))
3133 if (kvm_is_error_hva(ghc->hva))
3136 if (unlikely(!ghc->memslot))
3137 return kvm_read_guest(kvm, gpa, data, len);
3139 r = __copy_from_user(data, (void __user *)ghc->hva + offset, len);
3145 EXPORT_SYMBOL_GPL(kvm_read_guest_offset_cached);
3147 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3148 void *data, unsigned long len)
3150 return kvm_read_guest_offset_cached(kvm, ghc, data, 0, len);
3152 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
3154 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
3156 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
3157 gfn_t gfn = gpa >> PAGE_SHIFT;
3159 int offset = offset_in_page(gpa);
3162 while ((seg = next_segment(len, offset)) != 0) {
3163 ret = kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
3172 EXPORT_SYMBOL_GPL(kvm_clear_guest);
3174 void mark_page_dirty_in_slot(struct kvm *kvm,
3175 struct kvm_memory_slot *memslot,
3178 if (memslot && kvm_slot_dirty_track_enabled(memslot)) {
3179 unsigned long rel_gfn = gfn - memslot->base_gfn;
3180 u32 slot = (memslot->as_id << 16) | memslot->id;
3182 if (kvm->dirty_ring_size)
3183 kvm_dirty_ring_push(kvm_dirty_ring_get(kvm),
3186 set_bit_le(rel_gfn, memslot->dirty_bitmap);
3189 EXPORT_SYMBOL_GPL(mark_page_dirty_in_slot);
3191 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
3193 struct kvm_memory_slot *memslot;
3195 memslot = gfn_to_memslot(kvm, gfn);
3196 mark_page_dirty_in_slot(kvm, memslot, gfn);
3198 EXPORT_SYMBOL_GPL(mark_page_dirty);
3200 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
3202 struct kvm_memory_slot *memslot;
3204 memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
3205 mark_page_dirty_in_slot(vcpu->kvm, memslot, gfn);
3207 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
3209 void kvm_sigset_activate(struct kvm_vcpu *vcpu)
3211 if (!vcpu->sigset_active)
3215 * This does a lockless modification of ->real_blocked, which is fine
3216 * because, only current can change ->real_blocked and all readers of
3217 * ->real_blocked don't care as long ->real_blocked is always a subset
3220 sigprocmask(SIG_SETMASK, &vcpu->sigset, ¤t->real_blocked);
3223 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu)
3225 if (!vcpu->sigset_active)
3228 sigprocmask(SIG_SETMASK, ¤t->real_blocked, NULL);
3229 sigemptyset(¤t->real_blocked);
3232 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
3234 unsigned int old, val, grow, grow_start;
3236 old = val = vcpu->halt_poll_ns;
3237 grow_start = READ_ONCE(halt_poll_ns_grow_start);
3238 grow = READ_ONCE(halt_poll_ns_grow);
3243 if (val < grow_start)
3246 if (val > vcpu->kvm->max_halt_poll_ns)
3247 val = vcpu->kvm->max_halt_poll_ns;
3249 vcpu->halt_poll_ns = val;
3251 trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
3254 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
3256 unsigned int old, val, shrink, grow_start;
3258 old = val = vcpu->halt_poll_ns;
3259 shrink = READ_ONCE(halt_poll_ns_shrink);
3260 grow_start = READ_ONCE(halt_poll_ns_grow_start);
3266 if (val < grow_start)
3269 vcpu->halt_poll_ns = val;
3270 trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
3273 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
3276 int idx = srcu_read_lock(&vcpu->kvm->srcu);
3278 if (kvm_arch_vcpu_runnable(vcpu)) {
3279 kvm_make_request(KVM_REQ_UNHALT, vcpu);
3282 if (kvm_cpu_has_pending_timer(vcpu))
3284 if (signal_pending(current))
3286 if (kvm_check_request(KVM_REQ_UNBLOCK, vcpu))
3291 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3296 update_halt_poll_stats(struct kvm_vcpu *vcpu, u64 poll_ns, bool waited)
3299 vcpu->stat.generic.halt_poll_fail_ns += poll_ns;
3301 vcpu->stat.generic.halt_poll_success_ns += poll_ns;
3305 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
3307 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
3309 bool halt_poll_allowed = !kvm_arch_no_poll(vcpu);
3310 ktime_t start, cur, poll_end;
3311 bool waited = false;
3314 kvm_arch_vcpu_blocking(vcpu);
3316 start = cur = poll_end = ktime_get();
3317 if (vcpu->halt_poll_ns && halt_poll_allowed) {
3318 ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns);
3320 ++vcpu->stat.generic.halt_attempted_poll;
3323 * This sets KVM_REQ_UNHALT if an interrupt
3326 if (kvm_vcpu_check_block(vcpu) < 0) {
3327 ++vcpu->stat.generic.halt_successful_poll;
3328 if (!vcpu_valid_wakeup(vcpu))
3329 ++vcpu->stat.generic.halt_poll_invalid;
3331 KVM_STATS_LOG_HIST_UPDATE(
3332 vcpu->stat.generic.halt_poll_success_hist,
3333 ktime_to_ns(ktime_get()) -
3334 ktime_to_ns(start));
3338 poll_end = cur = ktime_get();
3339 } while (kvm_vcpu_can_poll(cur, stop));
3341 KVM_STATS_LOG_HIST_UPDATE(
3342 vcpu->stat.generic.halt_poll_fail_hist,
3343 ktime_to_ns(ktime_get()) - ktime_to_ns(start));
3347 prepare_to_rcuwait(&vcpu->wait);
3349 set_current_state(TASK_INTERRUPTIBLE);
3351 if (kvm_vcpu_check_block(vcpu) < 0)
3357 finish_rcuwait(&vcpu->wait);
3360 vcpu->stat.generic.halt_wait_ns +=
3361 ktime_to_ns(cur) - ktime_to_ns(poll_end);
3362 KVM_STATS_LOG_HIST_UPDATE(vcpu->stat.generic.halt_wait_hist,
3363 ktime_to_ns(cur) - ktime_to_ns(poll_end));
3366 kvm_arch_vcpu_unblocking(vcpu);
3367 block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
3369 update_halt_poll_stats(
3370 vcpu, ktime_to_ns(ktime_sub(poll_end, start)), waited);
3372 if (halt_poll_allowed) {
3373 if (!vcpu_valid_wakeup(vcpu)) {
3374 shrink_halt_poll_ns(vcpu);
3375 } else if (vcpu->kvm->max_halt_poll_ns) {
3376 if (block_ns <= vcpu->halt_poll_ns)
3378 /* we had a long block, shrink polling */
3379 else if (vcpu->halt_poll_ns &&
3380 block_ns > vcpu->kvm->max_halt_poll_ns)
3381 shrink_halt_poll_ns(vcpu);
3382 /* we had a short halt and our poll time is too small */
3383 else if (vcpu->halt_poll_ns < vcpu->kvm->max_halt_poll_ns &&
3384 block_ns < vcpu->kvm->max_halt_poll_ns)
3385 grow_halt_poll_ns(vcpu);
3387 vcpu->halt_poll_ns = 0;
3391 trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu));
3392 kvm_arch_vcpu_block_finish(vcpu);
3394 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
3396 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
3398 struct rcuwait *waitp;
3400 waitp = kvm_arch_vcpu_get_wait(vcpu);
3401 if (rcuwait_wake_up(waitp)) {
3402 WRITE_ONCE(vcpu->ready, true);
3403 ++vcpu->stat.generic.halt_wakeup;
3409 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
3413 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
3415 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
3419 if (kvm_vcpu_wake_up(vcpu))
3423 * Note, the vCPU could get migrated to a different pCPU at any point
3424 * after kvm_arch_vcpu_should_kick(), which could result in sending an
3425 * IPI to the previous pCPU. But, that's ok because the purpose of the
3426 * IPI is to force the vCPU to leave IN_GUEST_MODE, and migrating the
3427 * vCPU also requires it to leave IN_GUEST_MODE.
3430 if (kvm_arch_vcpu_should_kick(vcpu)) {
3431 cpu = READ_ONCE(vcpu->cpu);
3432 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
3433 smp_send_reschedule(cpu);
3437 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
3438 #endif /* !CONFIG_S390 */
3440 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
3443 struct task_struct *task = NULL;
3447 pid = rcu_dereference(target->pid);
3449 task = get_pid_task(pid, PIDTYPE_PID);
3453 ret = yield_to(task, 1);
3454 put_task_struct(task);
3458 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
3461 * Helper that checks whether a VCPU is eligible for directed yield.
3462 * Most eligible candidate to yield is decided by following heuristics:
3464 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
3465 * (preempted lock holder), indicated by @in_spin_loop.
3466 * Set at the beginning and cleared at the end of interception/PLE handler.
3468 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
3469 * chance last time (mostly it has become eligible now since we have probably
3470 * yielded to lockholder in last iteration. This is done by toggling
3471 * @dy_eligible each time a VCPU checked for eligibility.)
3473 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
3474 * to preempted lock-holder could result in wrong VCPU selection and CPU
3475 * burning. Giving priority for a potential lock-holder increases lock
3478 * Since algorithm is based on heuristics, accessing another VCPU data without
3479 * locking does not harm. It may result in trying to yield to same VCPU, fail
3480 * and continue with next VCPU and so on.
3482 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
3484 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
3487 eligible = !vcpu->spin_loop.in_spin_loop ||
3488 vcpu->spin_loop.dy_eligible;
3490 if (vcpu->spin_loop.in_spin_loop)
3491 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
3500 * Unlike kvm_arch_vcpu_runnable, this function is called outside
3501 * a vcpu_load/vcpu_put pair. However, for most architectures
3502 * kvm_arch_vcpu_runnable does not require vcpu_load.
3504 bool __weak kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
3506 return kvm_arch_vcpu_runnable(vcpu);
3509 static bool vcpu_dy_runnable(struct kvm_vcpu *vcpu)
3511 if (kvm_arch_dy_runnable(vcpu))
3514 #ifdef CONFIG_KVM_ASYNC_PF
3515 if (!list_empty_careful(&vcpu->async_pf.done))
3522 bool __weak kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu)
3527 void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode)
3529 struct kvm *kvm = me->kvm;
3530 struct kvm_vcpu *vcpu;
3531 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
3537 kvm_vcpu_set_in_spin_loop(me, true);
3539 * We boost the priority of a VCPU that is runnable but not
3540 * currently running, because it got preempted by something
3541 * else and called schedule in __vcpu_run. Hopefully that
3542 * VCPU is holding the lock that we need and will release it.
3543 * We approximate round-robin by starting at the last boosted VCPU.
3545 for (pass = 0; pass < 2 && !yielded && try; pass++) {
3546 kvm_for_each_vcpu(i, vcpu, kvm) {
3547 if (!pass && i <= last_boosted_vcpu) {
3548 i = last_boosted_vcpu;
3550 } else if (pass && i > last_boosted_vcpu)
3552 if (!READ_ONCE(vcpu->ready))
3556 if (rcuwait_active(&vcpu->wait) &&
3557 !vcpu_dy_runnable(vcpu))
3559 if (READ_ONCE(vcpu->preempted) && yield_to_kernel_mode &&
3560 !kvm_arch_dy_has_pending_interrupt(vcpu) &&
3561 !kvm_arch_vcpu_in_kernel(vcpu))
3563 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
3566 yielded = kvm_vcpu_yield_to(vcpu);
3568 kvm->last_boosted_vcpu = i;
3570 } else if (yielded < 0) {
3577 kvm_vcpu_set_in_spin_loop(me, false);
3579 /* Ensure vcpu is not eligible during next spinloop */
3580 kvm_vcpu_set_dy_eligible(me, false);
3582 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
3584 static bool kvm_page_in_dirty_ring(struct kvm *kvm, unsigned long pgoff)
3586 #if KVM_DIRTY_LOG_PAGE_OFFSET > 0
3587 return (pgoff >= KVM_DIRTY_LOG_PAGE_OFFSET) &&
3588 (pgoff < KVM_DIRTY_LOG_PAGE_OFFSET +
3589 kvm->dirty_ring_size / PAGE_SIZE);
3595 static vm_fault_t kvm_vcpu_fault(struct vm_fault *vmf)
3597 struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data;
3600 if (vmf->pgoff == 0)
3601 page = virt_to_page(vcpu->run);
3603 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
3604 page = virt_to_page(vcpu->arch.pio_data);
3606 #ifdef CONFIG_KVM_MMIO
3607 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
3608 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
3610 else if (kvm_page_in_dirty_ring(vcpu->kvm, vmf->pgoff))
3611 page = kvm_dirty_ring_get_page(
3613 vmf->pgoff - KVM_DIRTY_LOG_PAGE_OFFSET);
3615 return kvm_arch_vcpu_fault(vcpu, vmf);
3621 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
3622 .fault = kvm_vcpu_fault,
3625 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
3627 struct kvm_vcpu *vcpu = file->private_data;
3628 unsigned long pages = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
3630 if ((kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff) ||
3631 kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff + pages - 1)) &&
3632 ((vma->vm_flags & VM_EXEC) || !(vma->vm_flags & VM_SHARED)))
3635 vma->vm_ops = &kvm_vcpu_vm_ops;
3639 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
3641 struct kvm_vcpu *vcpu = filp->private_data;
3643 kvm_put_kvm(vcpu->kvm);
3647 static struct file_operations kvm_vcpu_fops = {
3648 .release = kvm_vcpu_release,
3649 .unlocked_ioctl = kvm_vcpu_ioctl,
3650 .mmap = kvm_vcpu_mmap,
3651 .llseek = noop_llseek,
3652 KVM_COMPAT(kvm_vcpu_compat_ioctl),
3656 * Allocates an inode for the vcpu.
3658 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
3660 char name[8 + 1 + ITOA_MAX_LEN + 1];
3662 snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id);
3663 return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
3666 static void kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
3668 #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS
3669 struct dentry *debugfs_dentry;
3670 char dir_name[ITOA_MAX_LEN * 2];
3672 if (!debugfs_initialized())
3675 snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
3676 debugfs_dentry = debugfs_create_dir(dir_name,
3677 vcpu->kvm->debugfs_dentry);
3679 kvm_arch_create_vcpu_debugfs(vcpu, debugfs_dentry);
3684 * Creates some virtual cpus. Good luck creating more than one.
3686 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
3689 struct kvm_vcpu *vcpu;
3692 if (id >= KVM_MAX_VCPU_ID)
3695 mutex_lock(&kvm->lock);
3696 if (kvm->created_vcpus == KVM_MAX_VCPUS) {
3697 mutex_unlock(&kvm->lock);
3701 kvm->created_vcpus++;
3702 mutex_unlock(&kvm->lock);
3704 r = kvm_arch_vcpu_precreate(kvm, id);
3706 goto vcpu_decrement;
3708 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL_ACCOUNT);
3711 goto vcpu_decrement;
3714 BUILD_BUG_ON(sizeof(struct kvm_run) > PAGE_SIZE);
3715 page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
3720 vcpu->run = page_address(page);
3722 kvm_vcpu_init(vcpu, kvm, id);
3724 r = kvm_arch_vcpu_create(vcpu);
3726 goto vcpu_free_run_page;
3728 if (kvm->dirty_ring_size) {
3729 r = kvm_dirty_ring_alloc(&vcpu->dirty_ring,
3730 id, kvm->dirty_ring_size);
3732 goto arch_vcpu_destroy;
3735 mutex_lock(&kvm->lock);
3736 if (kvm_get_vcpu_by_id(kvm, id)) {
3738 goto unlock_vcpu_destroy;
3741 vcpu->vcpu_idx = atomic_read(&kvm->online_vcpus);
3742 BUG_ON(kvm->vcpus[vcpu->vcpu_idx]);
3744 /* Fill the stats id string for the vcpu */
3745 snprintf(vcpu->stats_id, sizeof(vcpu->stats_id), "kvm-%d/vcpu-%d",
3746 task_pid_nr(current), id);
3748 /* Now it's all set up, let userspace reach it */
3750 r = create_vcpu_fd(vcpu);
3752 kvm_put_kvm_no_destroy(kvm);
3753 goto unlock_vcpu_destroy;
3756 kvm->vcpus[vcpu->vcpu_idx] = vcpu;
3759 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
3760 * before kvm->online_vcpu's incremented value.
3763 atomic_inc(&kvm->online_vcpus);
3765 mutex_unlock(&kvm->lock);
3766 kvm_arch_vcpu_postcreate(vcpu);
3767 kvm_create_vcpu_debugfs(vcpu);
3770 unlock_vcpu_destroy:
3771 mutex_unlock(&kvm->lock);
3772 kvm_dirty_ring_free(&vcpu->dirty_ring);
3774 kvm_arch_vcpu_destroy(vcpu);
3776 free_page((unsigned long)vcpu->run);
3778 kmem_cache_free(kvm_vcpu_cache, vcpu);
3780 mutex_lock(&kvm->lock);
3781 kvm->created_vcpus--;
3782 mutex_unlock(&kvm->lock);
3786 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
3789 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
3790 vcpu->sigset_active = 1;
3791 vcpu->sigset = *sigset;
3793 vcpu->sigset_active = 0;
3797 static ssize_t kvm_vcpu_stats_read(struct file *file, char __user *user_buffer,
3798 size_t size, loff_t *offset)
3800 struct kvm_vcpu *vcpu = file->private_data;
3802 return kvm_stats_read(vcpu->stats_id, &kvm_vcpu_stats_header,
3803 &kvm_vcpu_stats_desc[0], &vcpu->stat,
3804 sizeof(vcpu->stat), user_buffer, size, offset);
3807 static int kvm_vcpu_stats_release(struct inode *inode, struct file *file)
3809 struct kvm_vcpu *vcpu = file->private_data;
3811 kvm_put_kvm(vcpu->kvm);
3815 static const struct file_operations kvm_vcpu_stats_fops = {
3816 .read = kvm_vcpu_stats_read,
3817 .release = kvm_vcpu_stats_release,
3818 .llseek = noop_llseek,
3821 static int kvm_vcpu_ioctl_get_stats_fd(struct kvm_vcpu *vcpu)
3825 char name[15 + ITOA_MAX_LEN + 1];
3827 snprintf(name, sizeof(name), "kvm-vcpu-stats:%d", vcpu->vcpu_id);
3829 fd = get_unused_fd_flags(O_CLOEXEC);
3833 file = anon_inode_getfile(name, &kvm_vcpu_stats_fops, vcpu, O_RDONLY);
3836 return PTR_ERR(file);
3839 kvm_get_kvm(vcpu->kvm);
3841 file->f_mode |= FMODE_PREAD;
3842 fd_install(fd, file);
3847 static long kvm_vcpu_ioctl(struct file *filp,
3848 unsigned int ioctl, unsigned long arg)
3850 struct kvm_vcpu *vcpu = filp->private_data;
3851 void __user *argp = (void __user *)arg;
3853 struct kvm_fpu *fpu = NULL;
3854 struct kvm_sregs *kvm_sregs = NULL;
3856 if (vcpu->kvm->mm != current->mm || vcpu->kvm->vm_bugged)
3859 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
3863 * Some architectures have vcpu ioctls that are asynchronous to vcpu
3864 * execution; mutex_lock() would break them.
3866 r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg);
3867 if (r != -ENOIOCTLCMD)
3870 if (mutex_lock_killable(&vcpu->mutex))
3878 oldpid = rcu_access_pointer(vcpu->pid);
3879 if (unlikely(oldpid != task_pid(current))) {
3880 /* The thread running this VCPU changed. */
3883 r = kvm_arch_vcpu_run_pid_change(vcpu);
3887 newpid = get_task_pid(current, PIDTYPE_PID);
3888 rcu_assign_pointer(vcpu->pid, newpid);
3893 r = kvm_arch_vcpu_ioctl_run(vcpu);
3894 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
3897 case KVM_GET_REGS: {
3898 struct kvm_regs *kvm_regs;
3901 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL_ACCOUNT);
3904 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
3908 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
3915 case KVM_SET_REGS: {
3916 struct kvm_regs *kvm_regs;
3918 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
3919 if (IS_ERR(kvm_regs)) {
3920 r = PTR_ERR(kvm_regs);
3923 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
3927 case KVM_GET_SREGS: {
3928 kvm_sregs = kzalloc(sizeof(struct kvm_sregs),
3929 GFP_KERNEL_ACCOUNT);
3933 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
3937 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
3942 case KVM_SET_SREGS: {
3943 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
3944 if (IS_ERR(kvm_sregs)) {
3945 r = PTR_ERR(kvm_sregs);
3949 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
3952 case KVM_GET_MP_STATE: {
3953 struct kvm_mp_state mp_state;
3955 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
3959 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
3964 case KVM_SET_MP_STATE: {
3965 struct kvm_mp_state mp_state;
3968 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
3970 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
3973 case KVM_TRANSLATE: {
3974 struct kvm_translation tr;
3977 if (copy_from_user(&tr, argp, sizeof(tr)))
3979 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
3983 if (copy_to_user(argp, &tr, sizeof(tr)))
3988 case KVM_SET_GUEST_DEBUG: {
3989 struct kvm_guest_debug dbg;
3992 if (copy_from_user(&dbg, argp, sizeof(dbg)))
3994 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
3997 case KVM_SET_SIGNAL_MASK: {
3998 struct kvm_signal_mask __user *sigmask_arg = argp;
3999 struct kvm_signal_mask kvm_sigmask;
4000 sigset_t sigset, *p;
4005 if (copy_from_user(&kvm_sigmask, argp,
4006 sizeof(kvm_sigmask)))
4009 if (kvm_sigmask.len != sizeof(sigset))
4012 if (copy_from_user(&sigset, sigmask_arg->sigset,
4017 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
4021 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL_ACCOUNT);
4025 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
4029 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
4035 fpu = memdup_user(argp, sizeof(*fpu));
4041 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
4044 case KVM_GET_STATS_FD: {
4045 r = kvm_vcpu_ioctl_get_stats_fd(vcpu);
4049 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
4052 mutex_unlock(&vcpu->mutex);
4058 #ifdef CONFIG_KVM_COMPAT
4059 static long kvm_vcpu_compat_ioctl(struct file *filp,
4060 unsigned int ioctl, unsigned long arg)
4062 struct kvm_vcpu *vcpu = filp->private_data;
4063 void __user *argp = compat_ptr(arg);
4066 if (vcpu->kvm->mm != current->mm || vcpu->kvm->vm_bugged)
4070 case KVM_SET_SIGNAL_MASK: {
4071 struct kvm_signal_mask __user *sigmask_arg = argp;
4072 struct kvm_signal_mask kvm_sigmask;
4077 if (copy_from_user(&kvm_sigmask, argp,
4078 sizeof(kvm_sigmask)))
4081 if (kvm_sigmask.len != sizeof(compat_sigset_t))
4084 if (get_compat_sigset(&sigset,
4085 (compat_sigset_t __user *)sigmask_arg->sigset))
4087 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
4089 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
4093 r = kvm_vcpu_ioctl(filp, ioctl, arg);
4101 static int kvm_device_mmap(struct file *filp, struct vm_area_struct *vma)
4103 struct kvm_device *dev = filp->private_data;
4106 return dev->ops->mmap(dev, vma);
4111 static int kvm_device_ioctl_attr(struct kvm_device *dev,
4112 int (*accessor)(struct kvm_device *dev,
4113 struct kvm_device_attr *attr),
4116 struct kvm_device_attr attr;
4121 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
4124 return accessor(dev, &attr);
4127 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
4130 struct kvm_device *dev = filp->private_data;
4132 if (dev->kvm->mm != current->mm || dev->kvm->vm_bugged)
4136 case KVM_SET_DEVICE_ATTR:
4137 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
4138 case KVM_GET_DEVICE_ATTR:
4139 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
4140 case KVM_HAS_DEVICE_ATTR:
4141 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
4143 if (dev->ops->ioctl)
4144 return dev->ops->ioctl(dev, ioctl, arg);
4150 static int kvm_device_release(struct inode *inode, struct file *filp)
4152 struct kvm_device *dev = filp->private_data;
4153 struct kvm *kvm = dev->kvm;
4155 if (dev->ops->release) {
4156 mutex_lock(&kvm->lock);
4157 list_del(&dev->vm_node);
4158 dev->ops->release(dev);
4159 mutex_unlock(&kvm->lock);
4166 static const struct file_operations kvm_device_fops = {
4167 .unlocked_ioctl = kvm_device_ioctl,
4168 .release = kvm_device_release,
4169 KVM_COMPAT(kvm_device_ioctl),
4170 .mmap = kvm_device_mmap,
4173 struct kvm_device *kvm_device_from_filp(struct file *filp)
4175 if (filp->f_op != &kvm_device_fops)
4178 return filp->private_data;
4181 static const struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
4182 #ifdef CONFIG_KVM_MPIC
4183 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
4184 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
4188 int kvm_register_device_ops(const struct kvm_device_ops *ops, u32 type)
4190 if (type >= ARRAY_SIZE(kvm_device_ops_table))
4193 if (kvm_device_ops_table[type] != NULL)
4196 kvm_device_ops_table[type] = ops;
4200 void kvm_unregister_device_ops(u32 type)
4202 if (kvm_device_ops_table[type] != NULL)
4203 kvm_device_ops_table[type] = NULL;
4206 static int kvm_ioctl_create_device(struct kvm *kvm,
4207 struct kvm_create_device *cd)
4209 const struct kvm_device_ops *ops = NULL;
4210 struct kvm_device *dev;
4211 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
4215 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
4218 type = array_index_nospec(cd->type, ARRAY_SIZE(kvm_device_ops_table));
4219 ops = kvm_device_ops_table[type];
4226 dev = kzalloc(sizeof(*dev), GFP_KERNEL_ACCOUNT);
4233 mutex_lock(&kvm->lock);
4234 ret = ops->create(dev, type);
4236 mutex_unlock(&kvm->lock);
4240 list_add(&dev->vm_node, &kvm->devices);
4241 mutex_unlock(&kvm->lock);
4247 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
4249 kvm_put_kvm_no_destroy(kvm);
4250 mutex_lock(&kvm->lock);
4251 list_del(&dev->vm_node);
4254 mutex_unlock(&kvm->lock);
4264 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
4267 case KVM_CAP_USER_MEMORY:
4268 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
4269 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
4270 case KVM_CAP_INTERNAL_ERROR_DATA:
4271 #ifdef CONFIG_HAVE_KVM_MSI
4272 case KVM_CAP_SIGNAL_MSI:
4274 #ifdef CONFIG_HAVE_KVM_IRQFD
4276 case KVM_CAP_IRQFD_RESAMPLE:
4278 case KVM_CAP_IOEVENTFD_ANY_LENGTH:
4279 case KVM_CAP_CHECK_EXTENSION_VM:
4280 case KVM_CAP_ENABLE_CAP_VM:
4281 case KVM_CAP_HALT_POLL:
4283 #ifdef CONFIG_KVM_MMIO
4284 case KVM_CAP_COALESCED_MMIO:
4285 return KVM_COALESCED_MMIO_PAGE_OFFSET;
4286 case KVM_CAP_COALESCED_PIO:
4289 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4290 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2:
4291 return KVM_DIRTY_LOG_MANUAL_CAPS;
4293 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
4294 case KVM_CAP_IRQ_ROUTING:
4295 return KVM_MAX_IRQ_ROUTES;
4297 #if KVM_ADDRESS_SPACE_NUM > 1
4298 case KVM_CAP_MULTI_ADDRESS_SPACE:
4299 return KVM_ADDRESS_SPACE_NUM;
4301 case KVM_CAP_NR_MEMSLOTS:
4302 return KVM_USER_MEM_SLOTS;
4303 case KVM_CAP_DIRTY_LOG_RING:
4304 #if KVM_DIRTY_LOG_PAGE_OFFSET > 0
4305 return KVM_DIRTY_RING_MAX_ENTRIES * sizeof(struct kvm_dirty_gfn);
4309 case KVM_CAP_BINARY_STATS_FD:
4314 return kvm_vm_ioctl_check_extension(kvm, arg);
4317 static int kvm_vm_ioctl_enable_dirty_log_ring(struct kvm *kvm, u32 size)
4321 if (!KVM_DIRTY_LOG_PAGE_OFFSET)
4324 /* the size should be power of 2 */
4325 if (!size || (size & (size - 1)))
4328 /* Should be bigger to keep the reserved entries, or a page */
4329 if (size < kvm_dirty_ring_get_rsvd_entries() *
4330 sizeof(struct kvm_dirty_gfn) || size < PAGE_SIZE)
4333 if (size > KVM_DIRTY_RING_MAX_ENTRIES *
4334 sizeof(struct kvm_dirty_gfn))
4337 /* We only allow it to set once */
4338 if (kvm->dirty_ring_size)
4341 mutex_lock(&kvm->lock);
4343 if (kvm->created_vcpus) {
4344 /* We don't allow to change this value after vcpu created */
4347 kvm->dirty_ring_size = size;
4351 mutex_unlock(&kvm->lock);
4355 static int kvm_vm_ioctl_reset_dirty_pages(struct kvm *kvm)
4358 struct kvm_vcpu *vcpu;
4361 if (!kvm->dirty_ring_size)
4364 mutex_lock(&kvm->slots_lock);
4366 kvm_for_each_vcpu(i, vcpu, kvm)
4367 cleared += kvm_dirty_ring_reset(vcpu->kvm, &vcpu->dirty_ring);
4369 mutex_unlock(&kvm->slots_lock);
4372 kvm_flush_remote_tlbs(kvm);
4377 int __attribute__((weak)) kvm_vm_ioctl_enable_cap(struct kvm *kvm,
4378 struct kvm_enable_cap *cap)
4383 static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm,
4384 struct kvm_enable_cap *cap)
4387 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4388 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2: {
4389 u64 allowed_options = KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE;
4391 if (cap->args[0] & KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE)
4392 allowed_options = KVM_DIRTY_LOG_MANUAL_CAPS;
4394 if (cap->flags || (cap->args[0] & ~allowed_options))
4396 kvm->manual_dirty_log_protect = cap->args[0];
4400 case KVM_CAP_HALT_POLL: {
4401 if (cap->flags || cap->args[0] != (unsigned int)cap->args[0])
4404 kvm->max_halt_poll_ns = cap->args[0];
4407 case KVM_CAP_DIRTY_LOG_RING:
4408 return kvm_vm_ioctl_enable_dirty_log_ring(kvm, cap->args[0]);
4410 return kvm_vm_ioctl_enable_cap(kvm, cap);
4414 static ssize_t kvm_vm_stats_read(struct file *file, char __user *user_buffer,
4415 size_t size, loff_t *offset)
4417 struct kvm *kvm = file->private_data;
4419 return kvm_stats_read(kvm->stats_id, &kvm_vm_stats_header,
4420 &kvm_vm_stats_desc[0], &kvm->stat,
4421 sizeof(kvm->stat), user_buffer, size, offset);
4424 static int kvm_vm_stats_release(struct inode *inode, struct file *file)
4426 struct kvm *kvm = file->private_data;
4432 static const struct file_operations kvm_vm_stats_fops = {
4433 .read = kvm_vm_stats_read,
4434 .release = kvm_vm_stats_release,
4435 .llseek = noop_llseek,
4438 static int kvm_vm_ioctl_get_stats_fd(struct kvm *kvm)
4443 fd = get_unused_fd_flags(O_CLOEXEC);
4447 file = anon_inode_getfile("kvm-vm-stats",
4448 &kvm_vm_stats_fops, kvm, O_RDONLY);
4451 return PTR_ERR(file);
4456 file->f_mode |= FMODE_PREAD;
4457 fd_install(fd, file);
4462 static long kvm_vm_ioctl(struct file *filp,
4463 unsigned int ioctl, unsigned long arg)
4465 struct kvm *kvm = filp->private_data;
4466 void __user *argp = (void __user *)arg;
4469 if (kvm->mm != current->mm || kvm->vm_bugged)
4472 case KVM_CREATE_VCPU:
4473 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
4475 case KVM_ENABLE_CAP: {
4476 struct kvm_enable_cap cap;
4479 if (copy_from_user(&cap, argp, sizeof(cap)))
4481 r = kvm_vm_ioctl_enable_cap_generic(kvm, &cap);
4484 case KVM_SET_USER_MEMORY_REGION: {
4485 struct kvm_userspace_memory_region kvm_userspace_mem;
4488 if (copy_from_user(&kvm_userspace_mem, argp,
4489 sizeof(kvm_userspace_mem)))
4492 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
4495 case KVM_GET_DIRTY_LOG: {
4496 struct kvm_dirty_log log;
4499 if (copy_from_user(&log, argp, sizeof(log)))
4501 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
4504 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4505 case KVM_CLEAR_DIRTY_LOG: {
4506 struct kvm_clear_dirty_log log;
4509 if (copy_from_user(&log, argp, sizeof(log)))
4511 r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
4515 #ifdef CONFIG_KVM_MMIO
4516 case KVM_REGISTER_COALESCED_MMIO: {
4517 struct kvm_coalesced_mmio_zone zone;
4520 if (copy_from_user(&zone, argp, sizeof(zone)))
4522 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
4525 case KVM_UNREGISTER_COALESCED_MMIO: {
4526 struct kvm_coalesced_mmio_zone zone;
4529 if (copy_from_user(&zone, argp, sizeof(zone)))
4531 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
4536 struct kvm_irqfd data;
4539 if (copy_from_user(&data, argp, sizeof(data)))
4541 r = kvm_irqfd(kvm, &data);
4544 case KVM_IOEVENTFD: {
4545 struct kvm_ioeventfd data;
4548 if (copy_from_user(&data, argp, sizeof(data)))
4550 r = kvm_ioeventfd(kvm, &data);
4553 #ifdef CONFIG_HAVE_KVM_MSI
4554 case KVM_SIGNAL_MSI: {
4558 if (copy_from_user(&msi, argp, sizeof(msi)))
4560 r = kvm_send_userspace_msi(kvm, &msi);
4564 #ifdef __KVM_HAVE_IRQ_LINE
4565 case KVM_IRQ_LINE_STATUS:
4566 case KVM_IRQ_LINE: {
4567 struct kvm_irq_level irq_event;
4570 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
4573 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
4574 ioctl == KVM_IRQ_LINE_STATUS);
4579 if (ioctl == KVM_IRQ_LINE_STATUS) {
4580 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
4588 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
4589 case KVM_SET_GSI_ROUTING: {
4590 struct kvm_irq_routing routing;
4591 struct kvm_irq_routing __user *urouting;
4592 struct kvm_irq_routing_entry *entries = NULL;
4595 if (copy_from_user(&routing, argp, sizeof(routing)))
4598 if (!kvm_arch_can_set_irq_routing(kvm))
4600 if (routing.nr > KVM_MAX_IRQ_ROUTES)
4606 entries = vmemdup_user(urouting->entries,
4607 array_size(sizeof(*entries),
4609 if (IS_ERR(entries)) {
4610 r = PTR_ERR(entries);
4614 r = kvm_set_irq_routing(kvm, entries, routing.nr,
4619 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
4620 case KVM_CREATE_DEVICE: {
4621 struct kvm_create_device cd;
4624 if (copy_from_user(&cd, argp, sizeof(cd)))
4627 r = kvm_ioctl_create_device(kvm, &cd);
4632 if (copy_to_user(argp, &cd, sizeof(cd)))
4638 case KVM_CHECK_EXTENSION:
4639 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
4641 case KVM_RESET_DIRTY_RINGS:
4642 r = kvm_vm_ioctl_reset_dirty_pages(kvm);
4644 case KVM_GET_STATS_FD:
4645 r = kvm_vm_ioctl_get_stats_fd(kvm);
4648 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
4654 #ifdef CONFIG_KVM_COMPAT
4655 struct compat_kvm_dirty_log {
4659 compat_uptr_t dirty_bitmap; /* one bit per page */
4664 struct compat_kvm_clear_dirty_log {
4669 compat_uptr_t dirty_bitmap; /* one bit per page */
4674 long __weak kvm_arch_vm_compat_ioctl(struct file *filp, unsigned int ioctl,
4680 static long kvm_vm_compat_ioctl(struct file *filp,
4681 unsigned int ioctl, unsigned long arg)
4683 struct kvm *kvm = filp->private_data;
4686 if (kvm->mm != current->mm || kvm->vm_bugged)
4689 r = kvm_arch_vm_compat_ioctl(filp, ioctl, arg);
4694 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4695 case KVM_CLEAR_DIRTY_LOG: {
4696 struct compat_kvm_clear_dirty_log compat_log;
4697 struct kvm_clear_dirty_log log;
4699 if (copy_from_user(&compat_log, (void __user *)arg,
4700 sizeof(compat_log)))
4702 log.slot = compat_log.slot;
4703 log.num_pages = compat_log.num_pages;
4704 log.first_page = compat_log.first_page;
4705 log.padding2 = compat_log.padding2;
4706 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
4708 r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
4712 case KVM_GET_DIRTY_LOG: {
4713 struct compat_kvm_dirty_log compat_log;
4714 struct kvm_dirty_log log;
4716 if (copy_from_user(&compat_log, (void __user *)arg,
4717 sizeof(compat_log)))
4719 log.slot = compat_log.slot;
4720 log.padding1 = compat_log.padding1;
4721 log.padding2 = compat_log.padding2;
4722 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
4724 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
4728 r = kvm_vm_ioctl(filp, ioctl, arg);
4734 static struct file_operations kvm_vm_fops = {
4735 .release = kvm_vm_release,
4736 .unlocked_ioctl = kvm_vm_ioctl,
4737 .llseek = noop_llseek,
4738 KVM_COMPAT(kvm_vm_compat_ioctl),
4741 bool file_is_kvm(struct file *file)
4743 return file && file->f_op == &kvm_vm_fops;
4745 EXPORT_SYMBOL_GPL(file_is_kvm);
4747 static int kvm_dev_ioctl_create_vm(unsigned long type)
4753 kvm = kvm_create_vm(type);
4755 return PTR_ERR(kvm);
4756 #ifdef CONFIG_KVM_MMIO
4757 r = kvm_coalesced_mmio_init(kvm);
4761 r = get_unused_fd_flags(O_CLOEXEC);
4765 snprintf(kvm->stats_id, sizeof(kvm->stats_id),
4766 "kvm-%d", task_pid_nr(current));
4768 file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
4776 * Don't call kvm_put_kvm anymore at this point; file->f_op is
4777 * already set, with ->release() being kvm_vm_release(). In error
4778 * cases it will be called by the final fput(file) and will take
4779 * care of doing kvm_put_kvm(kvm).
4781 if (kvm_create_vm_debugfs(kvm, r) < 0) {
4786 kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm);
4788 fd_install(r, file);
4796 static long kvm_dev_ioctl(struct file *filp,
4797 unsigned int ioctl, unsigned long arg)
4802 case KVM_GET_API_VERSION:
4805 r = KVM_API_VERSION;
4808 r = kvm_dev_ioctl_create_vm(arg);
4810 case KVM_CHECK_EXTENSION:
4811 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
4813 case KVM_GET_VCPU_MMAP_SIZE:
4816 r = PAGE_SIZE; /* struct kvm_run */
4818 r += PAGE_SIZE; /* pio data page */
4820 #ifdef CONFIG_KVM_MMIO
4821 r += PAGE_SIZE; /* coalesced mmio ring page */
4824 case KVM_TRACE_ENABLE:
4825 case KVM_TRACE_PAUSE:
4826 case KVM_TRACE_DISABLE:
4830 return kvm_arch_dev_ioctl(filp, ioctl, arg);
4836 static struct file_operations kvm_chardev_ops = {
4837 .unlocked_ioctl = kvm_dev_ioctl,
4838 .llseek = noop_llseek,
4839 KVM_COMPAT(kvm_dev_ioctl),
4842 static struct miscdevice kvm_dev = {
4848 static void hardware_enable_nolock(void *junk)
4850 int cpu = raw_smp_processor_id();
4853 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
4856 cpumask_set_cpu(cpu, cpus_hardware_enabled);
4858 r = kvm_arch_hardware_enable();
4861 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
4862 atomic_inc(&hardware_enable_failed);
4863 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
4867 static int kvm_starting_cpu(unsigned int cpu)
4869 raw_spin_lock(&kvm_count_lock);
4870 if (kvm_usage_count)
4871 hardware_enable_nolock(NULL);
4872 raw_spin_unlock(&kvm_count_lock);
4876 static void hardware_disable_nolock(void *junk)
4878 int cpu = raw_smp_processor_id();
4880 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
4882 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
4883 kvm_arch_hardware_disable();
4886 static int kvm_dying_cpu(unsigned int cpu)
4888 raw_spin_lock(&kvm_count_lock);
4889 if (kvm_usage_count)
4890 hardware_disable_nolock(NULL);
4891 raw_spin_unlock(&kvm_count_lock);
4895 static void hardware_disable_all_nolock(void)
4897 BUG_ON(!kvm_usage_count);
4900 if (!kvm_usage_count)
4901 on_each_cpu(hardware_disable_nolock, NULL, 1);
4904 static void hardware_disable_all(void)
4906 raw_spin_lock(&kvm_count_lock);
4907 hardware_disable_all_nolock();
4908 raw_spin_unlock(&kvm_count_lock);
4911 static int hardware_enable_all(void)
4915 raw_spin_lock(&kvm_count_lock);
4918 if (kvm_usage_count == 1) {
4919 atomic_set(&hardware_enable_failed, 0);
4920 on_each_cpu(hardware_enable_nolock, NULL, 1);
4922 if (atomic_read(&hardware_enable_failed)) {
4923 hardware_disable_all_nolock();
4928 raw_spin_unlock(&kvm_count_lock);
4933 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
4937 * Some (well, at least mine) BIOSes hang on reboot if
4940 * And Intel TXT required VMX off for all cpu when system shutdown.
4942 pr_info("kvm: exiting hardware virtualization\n");
4943 kvm_rebooting = true;
4944 on_each_cpu(hardware_disable_nolock, NULL, 1);
4948 static struct notifier_block kvm_reboot_notifier = {
4949 .notifier_call = kvm_reboot,
4953 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
4957 for (i = 0; i < bus->dev_count; i++) {
4958 struct kvm_io_device *pos = bus->range[i].dev;
4960 kvm_iodevice_destructor(pos);
4965 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
4966 const struct kvm_io_range *r2)
4968 gpa_t addr1 = r1->addr;
4969 gpa_t addr2 = r2->addr;
4974 /* If r2->len == 0, match the exact address. If r2->len != 0,
4975 * accept any overlapping write. Any order is acceptable for
4976 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
4977 * we process all of them.
4990 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
4992 return kvm_io_bus_cmp(p1, p2);
4995 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
4996 gpa_t addr, int len)
4998 struct kvm_io_range *range, key;
5001 key = (struct kvm_io_range) {
5006 range = bsearch(&key, bus->range, bus->dev_count,
5007 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
5011 off = range - bus->range;
5013 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
5019 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
5020 struct kvm_io_range *range, const void *val)
5024 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
5028 while (idx < bus->dev_count &&
5029 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
5030 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
5039 /* kvm_io_bus_write - called under kvm->slots_lock */
5040 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
5041 int len, const void *val)
5043 struct kvm_io_bus *bus;
5044 struct kvm_io_range range;
5047 range = (struct kvm_io_range) {
5052 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
5055 r = __kvm_io_bus_write(vcpu, bus, &range, val);
5056 return r < 0 ? r : 0;
5058 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
5060 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
5061 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
5062 gpa_t addr, int len, const void *val, long cookie)
5064 struct kvm_io_bus *bus;
5065 struct kvm_io_range range;
5067 range = (struct kvm_io_range) {
5072 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
5076 /* First try the device referenced by cookie. */
5077 if ((cookie >= 0) && (cookie < bus->dev_count) &&
5078 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
5079 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
5084 * cookie contained garbage; fall back to search and return the
5085 * correct cookie value.
5087 return __kvm_io_bus_write(vcpu, bus, &range, val);
5090 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
5091 struct kvm_io_range *range, void *val)
5095 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
5099 while (idx < bus->dev_count &&
5100 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
5101 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
5110 /* kvm_io_bus_read - called under kvm->slots_lock */
5111 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
5114 struct kvm_io_bus *bus;
5115 struct kvm_io_range range;
5118 range = (struct kvm_io_range) {
5123 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
5126 r = __kvm_io_bus_read(vcpu, bus, &range, val);
5127 return r < 0 ? r : 0;
5130 /* Caller must hold slots_lock. */
5131 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
5132 int len, struct kvm_io_device *dev)
5135 struct kvm_io_bus *new_bus, *bus;
5136 struct kvm_io_range range;
5138 bus = kvm_get_bus(kvm, bus_idx);
5142 /* exclude ioeventfd which is limited by maximum fd */
5143 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
5146 new_bus = kmalloc(struct_size(bus, range, bus->dev_count + 1),
5147 GFP_KERNEL_ACCOUNT);
5151 range = (struct kvm_io_range) {
5157 for (i = 0; i < bus->dev_count; i++)
5158 if (kvm_io_bus_cmp(&bus->range[i], &range) > 0)
5161 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
5162 new_bus->dev_count++;
5163 new_bus->range[i] = range;
5164 memcpy(new_bus->range + i + 1, bus->range + i,
5165 (bus->dev_count - i) * sizeof(struct kvm_io_range));
5166 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
5167 synchronize_srcu_expedited(&kvm->srcu);
5173 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
5174 struct kvm_io_device *dev)
5177 struct kvm_io_bus *new_bus, *bus;
5179 lockdep_assert_held(&kvm->slots_lock);
5181 bus = kvm_get_bus(kvm, bus_idx);
5185 for (i = 0; i < bus->dev_count; i++) {
5186 if (bus->range[i].dev == dev) {
5191 if (i == bus->dev_count)
5194 new_bus = kmalloc(struct_size(bus, range, bus->dev_count - 1),
5195 GFP_KERNEL_ACCOUNT);
5197 memcpy(new_bus, bus, struct_size(bus, range, i));
5198 new_bus->dev_count--;
5199 memcpy(new_bus->range + i, bus->range + i + 1,
5200 flex_array_size(new_bus, range, new_bus->dev_count - i));
5203 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
5204 synchronize_srcu_expedited(&kvm->srcu);
5206 /* Destroy the old bus _after_ installing the (null) bus. */
5208 pr_err("kvm: failed to shrink bus, removing it completely\n");
5209 for (j = 0; j < bus->dev_count; j++) {
5212 kvm_iodevice_destructor(bus->range[j].dev);
5217 return new_bus ? 0 : -ENOMEM;
5220 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
5223 struct kvm_io_bus *bus;
5224 int dev_idx, srcu_idx;
5225 struct kvm_io_device *iodev = NULL;
5227 srcu_idx = srcu_read_lock(&kvm->srcu);
5229 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
5233 dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
5237 iodev = bus->range[dev_idx].dev;
5240 srcu_read_unlock(&kvm->srcu, srcu_idx);
5244 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
5246 static int kvm_debugfs_open(struct inode *inode, struct file *file,
5247 int (*get)(void *, u64 *), int (*set)(void *, u64),
5250 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
5254 * The debugfs files are a reference to the kvm struct which
5255 * is still valid when kvm_destroy_vm is called. kvm_get_kvm_safe
5256 * avoids the race between open and the removal of the debugfs directory.
5258 if (!kvm_get_kvm_safe(stat_data->kvm))
5261 if (simple_attr_open(inode, file, get,
5262 kvm_stats_debugfs_mode(stat_data->desc) & 0222
5265 kvm_put_kvm(stat_data->kvm);
5272 static int kvm_debugfs_release(struct inode *inode, struct file *file)
5274 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
5277 simple_attr_release(inode, file);
5278 kvm_put_kvm(stat_data->kvm);
5283 static int kvm_get_stat_per_vm(struct kvm *kvm, size_t offset, u64 *val)
5285 *val = *(u64 *)((void *)(&kvm->stat) + offset);
5290 static int kvm_clear_stat_per_vm(struct kvm *kvm, size_t offset)
5292 *(u64 *)((void *)(&kvm->stat) + offset) = 0;
5297 static int kvm_get_stat_per_vcpu(struct kvm *kvm, size_t offset, u64 *val)
5300 struct kvm_vcpu *vcpu;
5304 kvm_for_each_vcpu(i, vcpu, kvm)
5305 *val += *(u64 *)((void *)(&vcpu->stat) + offset);
5310 static int kvm_clear_stat_per_vcpu(struct kvm *kvm, size_t offset)
5313 struct kvm_vcpu *vcpu;
5315 kvm_for_each_vcpu(i, vcpu, kvm)
5316 *(u64 *)((void *)(&vcpu->stat) + offset) = 0;
5321 static int kvm_stat_data_get(void *data, u64 *val)
5324 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
5326 switch (stat_data->kind) {
5328 r = kvm_get_stat_per_vm(stat_data->kvm,
5329 stat_data->desc->desc.offset, val);
5332 r = kvm_get_stat_per_vcpu(stat_data->kvm,
5333 stat_data->desc->desc.offset, val);
5340 static int kvm_stat_data_clear(void *data, u64 val)
5343 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
5348 switch (stat_data->kind) {
5350 r = kvm_clear_stat_per_vm(stat_data->kvm,
5351 stat_data->desc->desc.offset);
5354 r = kvm_clear_stat_per_vcpu(stat_data->kvm,
5355 stat_data->desc->desc.offset);
5362 static int kvm_stat_data_open(struct inode *inode, struct file *file)
5364 __simple_attr_check_format("%llu\n", 0ull);
5365 return kvm_debugfs_open(inode, file, kvm_stat_data_get,
5366 kvm_stat_data_clear, "%llu\n");
5369 static const struct file_operations stat_fops_per_vm = {
5370 .owner = THIS_MODULE,
5371 .open = kvm_stat_data_open,
5372 .release = kvm_debugfs_release,
5373 .read = simple_attr_read,
5374 .write = simple_attr_write,
5375 .llseek = no_llseek,
5378 static int vm_stat_get(void *_offset, u64 *val)
5380 unsigned offset = (long)_offset;
5385 mutex_lock(&kvm_lock);
5386 list_for_each_entry(kvm, &vm_list, vm_list) {
5387 kvm_get_stat_per_vm(kvm, offset, &tmp_val);
5390 mutex_unlock(&kvm_lock);
5394 static int vm_stat_clear(void *_offset, u64 val)
5396 unsigned offset = (long)_offset;
5402 mutex_lock(&kvm_lock);
5403 list_for_each_entry(kvm, &vm_list, vm_list) {
5404 kvm_clear_stat_per_vm(kvm, offset);
5406 mutex_unlock(&kvm_lock);
5411 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n");
5412 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_readonly_fops, vm_stat_get, NULL, "%llu\n");
5414 static int vcpu_stat_get(void *_offset, u64 *val)
5416 unsigned offset = (long)_offset;
5421 mutex_lock(&kvm_lock);
5422 list_for_each_entry(kvm, &vm_list, vm_list) {
5423 kvm_get_stat_per_vcpu(kvm, offset, &tmp_val);
5426 mutex_unlock(&kvm_lock);
5430 static int vcpu_stat_clear(void *_offset, u64 val)
5432 unsigned offset = (long)_offset;
5438 mutex_lock(&kvm_lock);
5439 list_for_each_entry(kvm, &vm_list, vm_list) {
5440 kvm_clear_stat_per_vcpu(kvm, offset);
5442 mutex_unlock(&kvm_lock);
5447 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear,
5449 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_readonly_fops, vcpu_stat_get, NULL, "%llu\n");
5451 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm)
5453 struct kobj_uevent_env *env;
5454 unsigned long long created, active;
5456 if (!kvm_dev.this_device || !kvm)
5459 mutex_lock(&kvm_lock);
5460 if (type == KVM_EVENT_CREATE_VM) {
5461 kvm_createvm_count++;
5463 } else if (type == KVM_EVENT_DESTROY_VM) {
5466 created = kvm_createvm_count;
5467 active = kvm_active_vms;
5468 mutex_unlock(&kvm_lock);
5470 env = kzalloc(sizeof(*env), GFP_KERNEL_ACCOUNT);
5474 add_uevent_var(env, "CREATED=%llu", created);
5475 add_uevent_var(env, "COUNT=%llu", active);
5477 if (type == KVM_EVENT_CREATE_VM) {
5478 add_uevent_var(env, "EVENT=create");
5479 kvm->userspace_pid = task_pid_nr(current);
5480 } else if (type == KVM_EVENT_DESTROY_VM) {
5481 add_uevent_var(env, "EVENT=destroy");
5483 add_uevent_var(env, "PID=%d", kvm->userspace_pid);
5485 if (!IS_ERR(kvm->debugfs_dentry)) {
5486 char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL_ACCOUNT);
5489 tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX);
5491 add_uevent_var(env, "STATS_PATH=%s", tmp);
5495 /* no need for checks, since we are adding at most only 5 keys */
5496 env->envp[env->envp_idx++] = NULL;
5497 kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp);
5501 static void kvm_init_debug(void)
5503 const struct file_operations *fops;
5504 const struct _kvm_stats_desc *pdesc;
5507 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
5509 for (i = 0; i < kvm_vm_stats_header.num_desc; ++i) {
5510 pdesc = &kvm_vm_stats_desc[i];
5511 if (kvm_stats_debugfs_mode(pdesc) & 0222)
5512 fops = &vm_stat_fops;
5514 fops = &vm_stat_readonly_fops;
5515 debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
5517 (void *)(long)pdesc->desc.offset, fops);
5520 for (i = 0; i < kvm_vcpu_stats_header.num_desc; ++i) {
5521 pdesc = &kvm_vcpu_stats_desc[i];
5522 if (kvm_stats_debugfs_mode(pdesc) & 0222)
5523 fops = &vcpu_stat_fops;
5525 fops = &vcpu_stat_readonly_fops;
5526 debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
5528 (void *)(long)pdesc->desc.offset, fops);
5532 static int kvm_suspend(void)
5534 if (kvm_usage_count)
5535 hardware_disable_nolock(NULL);
5539 static void kvm_resume(void)
5541 if (kvm_usage_count) {
5542 lockdep_assert_not_held(&kvm_count_lock);
5543 hardware_enable_nolock(NULL);
5547 static struct syscore_ops kvm_syscore_ops = {
5548 .suspend = kvm_suspend,
5549 .resume = kvm_resume,
5553 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
5555 return container_of(pn, struct kvm_vcpu, preempt_notifier);
5558 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
5560 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
5562 WRITE_ONCE(vcpu->preempted, false);
5563 WRITE_ONCE(vcpu->ready, false);
5565 __this_cpu_write(kvm_running_vcpu, vcpu);
5566 kvm_arch_sched_in(vcpu, cpu);
5567 kvm_arch_vcpu_load(vcpu, cpu);
5570 static void kvm_sched_out(struct preempt_notifier *pn,
5571 struct task_struct *next)
5573 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
5575 if (current->on_rq) {
5576 WRITE_ONCE(vcpu->preempted, true);
5577 WRITE_ONCE(vcpu->ready, true);
5579 kvm_arch_vcpu_put(vcpu);
5580 __this_cpu_write(kvm_running_vcpu, NULL);
5584 * kvm_get_running_vcpu - get the vcpu running on the current CPU.
5586 * We can disable preemption locally around accessing the per-CPU variable,
5587 * and use the resolved vcpu pointer after enabling preemption again,
5588 * because even if the current thread is migrated to another CPU, reading
5589 * the per-CPU value later will give us the same value as we update the
5590 * per-CPU variable in the preempt notifier handlers.
5592 struct kvm_vcpu *kvm_get_running_vcpu(void)
5594 struct kvm_vcpu *vcpu;
5597 vcpu = __this_cpu_read(kvm_running_vcpu);
5602 EXPORT_SYMBOL_GPL(kvm_get_running_vcpu);
5605 * kvm_get_running_vcpus - get the per-CPU array of currently running vcpus.
5607 struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void)
5609 return &kvm_running_vcpu;
5612 struct kvm_cpu_compat_check {
5617 static void check_processor_compat(void *data)
5619 struct kvm_cpu_compat_check *c = data;
5621 *c->ret = kvm_arch_check_processor_compat(c->opaque);
5624 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
5625 struct module *module)
5627 struct kvm_cpu_compat_check c;
5631 r = kvm_arch_init(opaque);
5636 * kvm_arch_init makes sure there's at most one caller
5637 * for architectures that support multiple implementations,
5638 * like intel and amd on x86.
5639 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
5640 * conflicts in case kvm is already setup for another implementation.
5642 r = kvm_irqfd_init();
5646 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
5651 r = kvm_arch_hardware_setup(opaque);
5657 for_each_online_cpu(cpu) {
5658 smp_call_function_single(cpu, check_processor_compat, &c, 1);
5663 r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting",
5664 kvm_starting_cpu, kvm_dying_cpu);
5667 register_reboot_notifier(&kvm_reboot_notifier);
5669 /* A kmem cache lets us meet the alignment requirements of fx_save. */
5671 vcpu_align = __alignof__(struct kvm_vcpu);
5673 kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align,
5675 offsetof(struct kvm_vcpu, arch),
5676 offsetofend(struct kvm_vcpu, stats_id)
5677 - offsetof(struct kvm_vcpu, arch),
5679 if (!kvm_vcpu_cache) {
5684 for_each_possible_cpu(cpu) {
5685 if (!alloc_cpumask_var_node(&per_cpu(cpu_kick_mask, cpu),
5686 GFP_KERNEL, cpu_to_node(cpu))) {
5692 r = kvm_async_pf_init();
5696 kvm_chardev_ops.owner = module;
5697 kvm_vm_fops.owner = module;
5698 kvm_vcpu_fops.owner = module;
5700 register_syscore_ops(&kvm_syscore_ops);
5702 kvm_preempt_ops.sched_in = kvm_sched_in;
5703 kvm_preempt_ops.sched_out = kvm_sched_out;
5707 r = kvm_vfio_ops_init();
5708 if (WARN_ON_ONCE(r))
5712 * Registration _must_ be the very last thing done, as this exposes
5713 * /dev/kvm to userspace, i.e. all infrastructure must be setup!
5715 r = misc_register(&kvm_dev);
5717 pr_err("kvm: misc device register failed\n");
5724 kvm_vfio_ops_exit();
5726 kvm_async_pf_deinit();
5728 for_each_possible_cpu(cpu)
5729 free_cpumask_var(per_cpu(cpu_kick_mask, cpu));
5730 kmem_cache_destroy(kvm_vcpu_cache);
5732 unregister_reboot_notifier(&kvm_reboot_notifier);
5733 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
5735 kvm_arch_hardware_unsetup();
5737 free_cpumask_var(cpus_hardware_enabled);
5745 EXPORT_SYMBOL_GPL(kvm_init);
5752 * Note, unregistering /dev/kvm doesn't strictly need to come first,
5753 * fops_get(), a.k.a. try_module_get(), prevents acquiring references
5754 * to KVM while the module is being stopped.
5756 misc_deregister(&kvm_dev);
5758 debugfs_remove_recursive(kvm_debugfs_dir);
5759 for_each_possible_cpu(cpu)
5760 free_cpumask_var(per_cpu(cpu_kick_mask, cpu));
5761 kmem_cache_destroy(kvm_vcpu_cache);
5762 kvm_async_pf_deinit();
5763 unregister_syscore_ops(&kvm_syscore_ops);
5764 unregister_reboot_notifier(&kvm_reboot_notifier);
5765 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
5766 on_each_cpu(hardware_disable_nolock, NULL, 1);
5767 kvm_arch_hardware_unsetup();
5770 free_cpumask_var(cpus_hardware_enabled);
5771 kvm_vfio_ops_exit();
5773 EXPORT_SYMBOL_GPL(kvm_exit);
5775 struct kvm_vm_worker_thread_context {
5777 struct task_struct *parent;
5778 struct completion init_done;
5779 kvm_vm_thread_fn_t thread_fn;
5784 static int kvm_vm_worker_thread(void *context)
5787 * The init_context is allocated on the stack of the parent thread, so
5788 * we have to locally copy anything that is needed beyond initialization
5790 struct kvm_vm_worker_thread_context *init_context = context;
5791 struct kvm *kvm = init_context->kvm;
5792 kvm_vm_thread_fn_t thread_fn = init_context->thread_fn;
5793 uintptr_t data = init_context->data;
5796 err = kthread_park(current);
5797 /* kthread_park(current) is never supposed to return an error */
5802 err = cgroup_attach_task_all(init_context->parent, current);
5804 kvm_err("%s: cgroup_attach_task_all failed with err %d\n",
5809 set_user_nice(current, task_nice(init_context->parent));
5812 init_context->err = err;
5813 complete(&init_context->init_done);
5814 init_context = NULL;
5819 /* Wait to be woken up by the spawner before proceeding. */
5822 if (!kthread_should_stop())
5823 err = thread_fn(kvm, data);
5828 int kvm_vm_create_worker_thread(struct kvm *kvm, kvm_vm_thread_fn_t thread_fn,
5829 uintptr_t data, const char *name,
5830 struct task_struct **thread_ptr)
5832 struct kvm_vm_worker_thread_context init_context = {};
5833 struct task_struct *thread;
5836 init_context.kvm = kvm;
5837 init_context.parent = current;
5838 init_context.thread_fn = thread_fn;
5839 init_context.data = data;
5840 init_completion(&init_context.init_done);
5842 thread = kthread_run(kvm_vm_worker_thread, &init_context,
5843 "%s-%d", name, task_pid_nr(current));
5845 return PTR_ERR(thread);
5847 /* kthread_run is never supposed to return NULL */
5848 WARN_ON(thread == NULL);
5850 wait_for_completion(&init_context.init_done);
5852 if (!init_context.err)
5853 *thread_ptr = thread;
5855 return init_context.err;