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 long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
122 #ifdef CONFIG_KVM_COMPAT
123 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
125 #define KVM_COMPAT(c) .compat_ioctl = (c)
128 * For architectures that don't implement a compat infrastructure,
129 * adopt a double line of defense:
130 * - Prevent a compat task from opening /dev/kvm
131 * - If the open has been done by a 64bit task, and the KVM fd
132 * passed to a compat task, let the ioctls fail.
134 static long kvm_no_compat_ioctl(struct file *file, unsigned int ioctl,
135 unsigned long arg) { return -EINVAL; }
137 static int kvm_no_compat_open(struct inode *inode, struct file *file)
139 return is_compat_task() ? -ENODEV : 0;
141 #define KVM_COMPAT(c) .compat_ioctl = kvm_no_compat_ioctl, \
142 .open = kvm_no_compat_open
144 static int hardware_enable_all(void);
145 static void hardware_disable_all(void);
147 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
149 __visible bool kvm_rebooting;
150 EXPORT_SYMBOL_GPL(kvm_rebooting);
152 #define KVM_EVENT_CREATE_VM 0
153 #define KVM_EVENT_DESTROY_VM 1
154 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm);
155 static unsigned long long kvm_createvm_count;
156 static unsigned long long kvm_active_vms;
158 static DEFINE_PER_CPU(cpumask_var_t, cpu_kick_mask);
160 __weak void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
161 unsigned long start, unsigned long end)
165 bool kvm_is_zone_device_pfn(kvm_pfn_t pfn)
168 * The metadata used by is_zone_device_page() to determine whether or
169 * not a page is ZONE_DEVICE is guaranteed to be valid if and only if
170 * the device has been pinned, e.g. by get_user_pages(). WARN if the
171 * page_count() is zero to help detect bad usage of this helper.
173 if (!pfn_valid(pfn) || WARN_ON_ONCE(!page_count(pfn_to_page(pfn))))
176 return is_zone_device_page(pfn_to_page(pfn));
179 bool kvm_is_reserved_pfn(kvm_pfn_t pfn)
182 * ZONE_DEVICE pages currently set PG_reserved, but from a refcounting
183 * perspective they are "normal" pages, albeit with slightly different
187 return PageReserved(pfn_to_page(pfn)) &&
189 !kvm_is_zone_device_pfn(pfn);
195 * Switches to specified vcpu, until a matching vcpu_put()
197 void vcpu_load(struct kvm_vcpu *vcpu)
201 __this_cpu_write(kvm_running_vcpu, vcpu);
202 preempt_notifier_register(&vcpu->preempt_notifier);
203 kvm_arch_vcpu_load(vcpu, cpu);
206 EXPORT_SYMBOL_GPL(vcpu_load);
208 void vcpu_put(struct kvm_vcpu *vcpu)
211 kvm_arch_vcpu_put(vcpu);
212 preempt_notifier_unregister(&vcpu->preempt_notifier);
213 __this_cpu_write(kvm_running_vcpu, NULL);
216 EXPORT_SYMBOL_GPL(vcpu_put);
218 /* TODO: merge with kvm_arch_vcpu_should_kick */
219 static bool kvm_request_needs_ipi(struct kvm_vcpu *vcpu, unsigned req)
221 int mode = kvm_vcpu_exiting_guest_mode(vcpu);
224 * We need to wait for the VCPU to reenable interrupts and get out of
225 * READING_SHADOW_PAGE_TABLES mode.
227 if (req & KVM_REQUEST_WAIT)
228 return mode != OUTSIDE_GUEST_MODE;
231 * Need to kick a running VCPU, but otherwise there is nothing to do.
233 return mode == IN_GUEST_MODE;
236 static void ack_flush(void *_completed)
240 static inline bool kvm_kick_many_cpus(struct cpumask *cpus, bool wait)
242 if (cpumask_empty(cpus))
245 smp_call_function_many(cpus, ack_flush, NULL, wait);
249 static void kvm_make_vcpu_request(struct kvm *kvm, struct kvm_vcpu *vcpu,
250 unsigned int req, struct cpumask *tmp,
255 kvm_make_request(req, vcpu);
257 if (!(req & KVM_REQUEST_NO_WAKEUP) && kvm_vcpu_wake_up(vcpu))
261 * Note, the vCPU could get migrated to a different pCPU at any point
262 * after kvm_request_needs_ipi(), which could result in sending an IPI
263 * to the previous pCPU. But, that's OK because the purpose of the IPI
264 * is to ensure the vCPU returns to OUTSIDE_GUEST_MODE, which is
265 * satisfied if the vCPU migrates. Entering READING_SHADOW_PAGE_TABLES
266 * after this point is also OK, as the requirement is only that KVM wait
267 * for vCPUs that were reading SPTEs _before_ any changes were
268 * finalized. See kvm_vcpu_kick() for more details on handling requests.
270 if (kvm_request_needs_ipi(vcpu, req)) {
271 cpu = READ_ONCE(vcpu->cpu);
272 if (cpu != -1 && cpu != current_cpu)
273 __cpumask_set_cpu(cpu, tmp);
277 bool kvm_make_vcpus_request_mask(struct kvm *kvm, unsigned int req,
278 unsigned long *vcpu_bitmap)
280 struct kvm_vcpu *vcpu;
281 struct cpumask *cpus;
287 cpus = this_cpu_cpumask_var_ptr(cpu_kick_mask);
290 for_each_set_bit(i, vcpu_bitmap, KVM_MAX_VCPUS) {
291 vcpu = kvm_get_vcpu(kvm, i);
294 kvm_make_vcpu_request(kvm, vcpu, req, cpus, me);
297 called = kvm_kick_many_cpus(cpus, !!(req & KVM_REQUEST_WAIT));
303 bool kvm_make_all_cpus_request_except(struct kvm *kvm, unsigned int req,
304 struct kvm_vcpu *except)
306 struct kvm_vcpu *vcpu;
307 struct cpumask *cpus;
314 cpus = this_cpu_cpumask_var_ptr(cpu_kick_mask);
317 kvm_for_each_vcpu(i, vcpu, kvm) {
320 kvm_make_vcpu_request(kvm, vcpu, req, cpus, me);
323 called = kvm_kick_many_cpus(cpus, !!(req & KVM_REQUEST_WAIT));
329 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
331 return kvm_make_all_cpus_request_except(kvm, req, NULL);
333 EXPORT_SYMBOL_GPL(kvm_make_all_cpus_request);
335 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
336 void kvm_flush_remote_tlbs(struct kvm *kvm)
338 ++kvm->stat.generic.remote_tlb_flush_requests;
341 * We want to publish modifications to the page tables before reading
342 * mode. Pairs with a memory barrier in arch-specific code.
343 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
344 * and smp_mb in walk_shadow_page_lockless_begin/end.
345 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
347 * There is already an smp_mb__after_atomic() before
348 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
351 if (!kvm_arch_flush_remote_tlb(kvm)
352 || kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
353 ++kvm->stat.generic.remote_tlb_flush;
355 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
358 void kvm_reload_remote_mmus(struct kvm *kvm)
360 kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
363 #ifdef KVM_ARCH_NR_OBJS_PER_MEMORY_CACHE
364 static inline void *mmu_memory_cache_alloc_obj(struct kvm_mmu_memory_cache *mc,
367 gfp_flags |= mc->gfp_zero;
370 return kmem_cache_alloc(mc->kmem_cache, gfp_flags);
372 return (void *)__get_free_page(gfp_flags);
375 int kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache *mc, int min)
379 if (mc->nobjs >= min)
381 while (mc->nobjs < ARRAY_SIZE(mc->objects)) {
382 obj = mmu_memory_cache_alloc_obj(mc, GFP_KERNEL_ACCOUNT);
384 return mc->nobjs >= min ? 0 : -ENOMEM;
385 mc->objects[mc->nobjs++] = obj;
390 int kvm_mmu_memory_cache_nr_free_objects(struct kvm_mmu_memory_cache *mc)
395 void kvm_mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc)
399 kmem_cache_free(mc->kmem_cache, mc->objects[--mc->nobjs]);
401 free_page((unsigned long)mc->objects[--mc->nobjs]);
405 void *kvm_mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc)
409 if (WARN_ON(!mc->nobjs))
410 p = mmu_memory_cache_alloc_obj(mc, GFP_ATOMIC | __GFP_ACCOUNT);
412 p = mc->objects[--mc->nobjs];
418 static void kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
420 mutex_init(&vcpu->mutex);
425 #ifndef __KVM_HAVE_ARCH_WQP
426 rcuwait_init(&vcpu->wait);
428 kvm_async_pf_vcpu_init(vcpu);
431 INIT_LIST_HEAD(&vcpu->blocked_vcpu_list);
433 kvm_vcpu_set_in_spin_loop(vcpu, false);
434 kvm_vcpu_set_dy_eligible(vcpu, false);
435 vcpu->preempted = false;
437 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
438 vcpu->last_used_slot = NULL;
441 static void kvm_vcpu_destroy(struct kvm_vcpu *vcpu)
443 kvm_dirty_ring_free(&vcpu->dirty_ring);
444 kvm_arch_vcpu_destroy(vcpu);
447 * No need for rcu_read_lock as VCPU_RUN is the only place that changes
448 * the vcpu->pid pointer, and at destruction time all file descriptors
451 put_pid(rcu_dereference_protected(vcpu->pid, 1));
453 free_page((unsigned long)vcpu->run);
454 kmem_cache_free(kvm_vcpu_cache, vcpu);
457 void kvm_destroy_vcpus(struct kvm *kvm)
460 struct kvm_vcpu *vcpu;
462 kvm_for_each_vcpu(i, vcpu, kvm) {
463 kvm_vcpu_destroy(vcpu);
464 xa_erase(&kvm->vcpu_array, i);
467 atomic_set(&kvm->online_vcpus, 0);
469 EXPORT_SYMBOL_GPL(kvm_destroy_vcpus);
471 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
472 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
474 return container_of(mn, struct kvm, mmu_notifier);
477 static void kvm_mmu_notifier_invalidate_range(struct mmu_notifier *mn,
478 struct mm_struct *mm,
479 unsigned long start, unsigned long end)
481 struct kvm *kvm = mmu_notifier_to_kvm(mn);
484 idx = srcu_read_lock(&kvm->srcu);
485 kvm_arch_mmu_notifier_invalidate_range(kvm, start, end);
486 srcu_read_unlock(&kvm->srcu, idx);
489 typedef bool (*hva_handler_t)(struct kvm *kvm, struct kvm_gfn_range *range);
491 typedef void (*on_lock_fn_t)(struct kvm *kvm, unsigned long start,
494 struct kvm_hva_range {
498 hva_handler_t handler;
499 on_lock_fn_t on_lock;
505 * Use a dedicated stub instead of NULL to indicate that there is no callback
506 * function/handler. The compiler technically can't guarantee that a real
507 * function will have a non-zero address, and so it will generate code to
508 * check for !NULL, whereas comparing against a stub will be elided at compile
509 * time (unless the compiler is getting long in the tooth, e.g. gcc 4.9).
511 static void kvm_null_fn(void)
515 #define IS_KVM_NULL_FN(fn) ((fn) == (void *)kvm_null_fn)
517 /* Iterate over each memslot intersecting [start, last] (inclusive) range */
518 #define kvm_for_each_memslot_in_hva_range(node, slots, start, last) \
519 for (node = interval_tree_iter_first(&slots->hva_tree, start, last); \
521 node = interval_tree_iter_next(node, start, last)) \
523 static __always_inline int __kvm_handle_hva_range(struct kvm *kvm,
524 const struct kvm_hva_range *range)
526 bool ret = false, locked = false;
527 struct kvm_gfn_range gfn_range;
528 struct kvm_memory_slot *slot;
529 struct kvm_memslots *slots;
532 if (WARN_ON_ONCE(range->end <= range->start))
535 /* A null handler is allowed if and only if on_lock() is provided. */
536 if (WARN_ON_ONCE(IS_KVM_NULL_FN(range->on_lock) &&
537 IS_KVM_NULL_FN(range->handler)))
540 idx = srcu_read_lock(&kvm->srcu);
542 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
543 struct interval_tree_node *node;
545 slots = __kvm_memslots(kvm, i);
546 kvm_for_each_memslot_in_hva_range(node, slots,
547 range->start, range->end - 1) {
548 unsigned long hva_start, hva_end;
550 slot = container_of(node, struct kvm_memory_slot, hva_node[slots->node_idx]);
551 hva_start = max(range->start, slot->userspace_addr);
552 hva_end = min(range->end, slot->userspace_addr +
553 (slot->npages << PAGE_SHIFT));
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);
590 srcu_read_unlock(&kvm->srcu, idx);
592 /* The notifiers are averse to booleans. :-( */
596 static __always_inline int kvm_handle_hva_range(struct mmu_notifier *mn,
600 hva_handler_t handler)
602 struct kvm *kvm = mmu_notifier_to_kvm(mn);
603 const struct kvm_hva_range range = {
608 .on_lock = (void *)kvm_null_fn,
609 .flush_on_ret = true,
613 return __kvm_handle_hva_range(kvm, &range);
616 static __always_inline int kvm_handle_hva_range_no_flush(struct mmu_notifier *mn,
619 hva_handler_t handler)
621 struct kvm *kvm = mmu_notifier_to_kvm(mn);
622 const struct kvm_hva_range range = {
627 .on_lock = (void *)kvm_null_fn,
628 .flush_on_ret = false,
632 return __kvm_handle_hva_range(kvm, &range);
634 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
635 struct mm_struct *mm,
636 unsigned long address,
639 struct kvm *kvm = mmu_notifier_to_kvm(mn);
641 trace_kvm_set_spte_hva(address);
644 * .change_pte() must be surrounded by .invalidate_range_{start,end}().
645 * If mmu_notifier_count is zero, then no in-progress invalidations,
646 * including this one, found a relevant memslot at start(); rechecking
647 * memslots here is unnecessary. Note, a false positive (count elevated
648 * by a different invalidation) is sub-optimal but functionally ok.
650 WARN_ON_ONCE(!READ_ONCE(kvm->mn_active_invalidate_count));
651 if (!READ_ONCE(kvm->mmu_notifier_count))
654 kvm_handle_hva_range(mn, address, address + 1, pte, kvm_set_spte_gfn);
657 void kvm_inc_notifier_count(struct kvm *kvm, unsigned long start,
661 * The count increase must become visible at unlock time as no
662 * spte can be established without taking the mmu_lock and
663 * count is also read inside the mmu_lock critical section.
665 kvm->mmu_notifier_count++;
666 if (likely(kvm->mmu_notifier_count == 1)) {
667 kvm->mmu_notifier_range_start = start;
668 kvm->mmu_notifier_range_end = end;
671 * Fully tracking multiple concurrent ranges has dimishing
672 * returns. Keep things simple and just find the minimal range
673 * which includes the current and new ranges. As there won't be
674 * enough information to subtract a range after its invalidate
675 * completes, any ranges invalidated concurrently will
676 * accumulate and persist until all outstanding invalidates
679 kvm->mmu_notifier_range_start =
680 min(kvm->mmu_notifier_range_start, start);
681 kvm->mmu_notifier_range_end =
682 max(kvm->mmu_notifier_range_end, end);
686 static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
687 const struct mmu_notifier_range *range)
689 struct kvm *kvm = mmu_notifier_to_kvm(mn);
690 const struct kvm_hva_range hva_range = {
691 .start = range->start,
694 .handler = kvm_unmap_gfn_range,
695 .on_lock = kvm_inc_notifier_count,
696 .flush_on_ret = true,
697 .may_block = mmu_notifier_range_blockable(range),
700 trace_kvm_unmap_hva_range(range->start, range->end);
703 * Prevent memslot modification between range_start() and range_end()
704 * so that conditionally locking provides the same result in both
705 * functions. Without that guarantee, the mmu_notifier_count
706 * adjustments will be imbalanced.
708 * Pairs with the decrement in range_end().
710 spin_lock(&kvm->mn_invalidate_lock);
711 kvm->mn_active_invalidate_count++;
712 spin_unlock(&kvm->mn_invalidate_lock);
714 gfn_to_pfn_cache_invalidate_start(kvm, range->start, range->end,
715 hva_range.may_block);
717 __kvm_handle_hva_range(kvm, &hva_range);
722 void kvm_dec_notifier_count(struct kvm *kvm, unsigned long start,
726 * This sequence increase will notify the kvm page fault that
727 * the page that is going to be mapped in the spte could have
730 kvm->mmu_notifier_seq++;
733 * The above sequence increase must be visible before the
734 * below count decrease, which is ensured by the smp_wmb above
735 * in conjunction with the smp_rmb in mmu_notifier_retry().
737 kvm->mmu_notifier_count--;
740 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
741 const struct mmu_notifier_range *range)
743 struct kvm *kvm = mmu_notifier_to_kvm(mn);
744 const struct kvm_hva_range hva_range = {
745 .start = range->start,
748 .handler = (void *)kvm_null_fn,
749 .on_lock = kvm_dec_notifier_count,
750 .flush_on_ret = false,
751 .may_block = mmu_notifier_range_blockable(range),
755 __kvm_handle_hva_range(kvm, &hva_range);
757 /* Pairs with the increment in range_start(). */
758 spin_lock(&kvm->mn_invalidate_lock);
759 wake = (--kvm->mn_active_invalidate_count == 0);
760 spin_unlock(&kvm->mn_invalidate_lock);
763 * There can only be one waiter, since the wait happens under
767 rcuwait_wake_up(&kvm->mn_memslots_update_rcuwait);
769 BUG_ON(kvm->mmu_notifier_count < 0);
772 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
773 struct mm_struct *mm,
777 trace_kvm_age_hva(start, end);
779 return kvm_handle_hva_range(mn, start, end, __pte(0), kvm_age_gfn);
782 static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
783 struct mm_struct *mm,
787 trace_kvm_age_hva(start, end);
790 * Even though we do not flush TLB, this will still adversely
791 * affect performance on pre-Haswell Intel EPT, where there is
792 * no EPT Access Bit to clear so that we have to tear down EPT
793 * tables instead. If we find this unacceptable, we can always
794 * add a parameter to kvm_age_hva so that it effectively doesn't
795 * do anything on clear_young.
797 * Also note that currently we never issue secondary TLB flushes
798 * from clear_young, leaving this job up to the regular system
799 * cadence. If we find this inaccurate, we might come up with a
800 * more sophisticated heuristic later.
802 return kvm_handle_hva_range_no_flush(mn, start, end, kvm_age_gfn);
805 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
806 struct mm_struct *mm,
807 unsigned long address)
809 trace_kvm_test_age_hva(address);
811 return kvm_handle_hva_range_no_flush(mn, address, address + 1,
815 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
816 struct mm_struct *mm)
818 struct kvm *kvm = mmu_notifier_to_kvm(mn);
821 idx = srcu_read_lock(&kvm->srcu);
822 kvm_arch_flush_shadow_all(kvm);
823 srcu_read_unlock(&kvm->srcu, idx);
826 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
827 .invalidate_range = kvm_mmu_notifier_invalidate_range,
828 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
829 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
830 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
831 .clear_young = kvm_mmu_notifier_clear_young,
832 .test_young = kvm_mmu_notifier_test_young,
833 .change_pte = kvm_mmu_notifier_change_pte,
834 .release = kvm_mmu_notifier_release,
837 static int kvm_init_mmu_notifier(struct kvm *kvm)
839 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
840 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
843 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
845 static int kvm_init_mmu_notifier(struct kvm *kvm)
850 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
852 #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER
853 static int kvm_pm_notifier_call(struct notifier_block *bl,
857 struct kvm *kvm = container_of(bl, struct kvm, pm_notifier);
859 return kvm_arch_pm_notifier(kvm, state);
862 static void kvm_init_pm_notifier(struct kvm *kvm)
864 kvm->pm_notifier.notifier_call = kvm_pm_notifier_call;
865 /* Suspend KVM before we suspend ftrace, RCU, etc. */
866 kvm->pm_notifier.priority = INT_MAX;
867 register_pm_notifier(&kvm->pm_notifier);
870 static void kvm_destroy_pm_notifier(struct kvm *kvm)
872 unregister_pm_notifier(&kvm->pm_notifier);
874 #else /* !CONFIG_HAVE_KVM_PM_NOTIFIER */
875 static void kvm_init_pm_notifier(struct kvm *kvm)
879 static void kvm_destroy_pm_notifier(struct kvm *kvm)
882 #endif /* CONFIG_HAVE_KVM_PM_NOTIFIER */
884 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
886 if (!memslot->dirty_bitmap)
889 kvfree(memslot->dirty_bitmap);
890 memslot->dirty_bitmap = NULL;
893 /* This does not remove the slot from struct kvm_memslots data structures */
894 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
896 kvm_destroy_dirty_bitmap(slot);
898 kvm_arch_free_memslot(kvm, slot);
903 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
905 struct hlist_node *idnode;
906 struct kvm_memory_slot *memslot;
910 * The same memslot objects live in both active and inactive sets,
911 * arbitrarily free using index '1' so the second invocation of this
912 * function isn't operating over a structure with dangling pointers
913 * (even though this function isn't actually touching them).
915 if (!slots->node_idx)
918 hash_for_each_safe(slots->id_hash, bkt, idnode, memslot, id_node[1])
919 kvm_free_memslot(kvm, memslot);
922 static umode_t kvm_stats_debugfs_mode(const struct _kvm_stats_desc *pdesc)
924 switch (pdesc->desc.flags & KVM_STATS_TYPE_MASK) {
925 case KVM_STATS_TYPE_INSTANT:
927 case KVM_STATS_TYPE_CUMULATIVE:
928 case KVM_STATS_TYPE_PEAK:
935 static void kvm_destroy_vm_debugfs(struct kvm *kvm)
938 int kvm_debugfs_num_entries = kvm_vm_stats_header.num_desc +
939 kvm_vcpu_stats_header.num_desc;
941 if (!kvm->debugfs_dentry)
944 debugfs_remove_recursive(kvm->debugfs_dentry);
946 if (kvm->debugfs_stat_data) {
947 for (i = 0; i < kvm_debugfs_num_entries; i++)
948 kfree(kvm->debugfs_stat_data[i]);
949 kfree(kvm->debugfs_stat_data);
953 static int kvm_create_vm_debugfs(struct kvm *kvm, int fd)
955 static DEFINE_MUTEX(kvm_debugfs_lock);
957 char dir_name[ITOA_MAX_LEN * 2];
958 struct kvm_stat_data *stat_data;
959 const struct _kvm_stats_desc *pdesc;
961 int kvm_debugfs_num_entries = kvm_vm_stats_header.num_desc +
962 kvm_vcpu_stats_header.num_desc;
964 if (!debugfs_initialized())
967 snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd);
968 mutex_lock(&kvm_debugfs_lock);
969 dent = debugfs_lookup(dir_name, kvm_debugfs_dir);
971 pr_warn_ratelimited("KVM: debugfs: duplicate directory %s\n", dir_name);
973 mutex_unlock(&kvm_debugfs_lock);
976 dent = debugfs_create_dir(dir_name, kvm_debugfs_dir);
977 mutex_unlock(&kvm_debugfs_lock);
981 kvm->debugfs_dentry = dent;
982 kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries,
983 sizeof(*kvm->debugfs_stat_data),
985 if (!kvm->debugfs_stat_data)
988 for (i = 0; i < kvm_vm_stats_header.num_desc; ++i) {
989 pdesc = &kvm_vm_stats_desc[i];
990 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT);
994 stat_data->kvm = kvm;
995 stat_data->desc = pdesc;
996 stat_data->kind = KVM_STAT_VM;
997 kvm->debugfs_stat_data[i] = stat_data;
998 debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
999 kvm->debugfs_dentry, stat_data,
1003 for (i = 0; i < kvm_vcpu_stats_header.num_desc; ++i) {
1004 pdesc = &kvm_vcpu_stats_desc[i];
1005 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT);
1009 stat_data->kvm = kvm;
1010 stat_data->desc = pdesc;
1011 stat_data->kind = KVM_STAT_VCPU;
1012 kvm->debugfs_stat_data[i + kvm_vm_stats_header.num_desc] = stat_data;
1013 debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
1014 kvm->debugfs_dentry, stat_data,
1018 ret = kvm_arch_create_vm_debugfs(kvm);
1020 kvm_destroy_vm_debugfs(kvm);
1028 * Called after the VM is otherwise initialized, but just before adding it to
1031 int __weak kvm_arch_post_init_vm(struct kvm *kvm)
1037 * Called just after removing the VM from the vm_list, but before doing any
1038 * other destruction.
1040 void __weak kvm_arch_pre_destroy_vm(struct kvm *kvm)
1045 * Called after per-vm debugfs created. When called kvm->debugfs_dentry should
1046 * be setup already, so we can create arch-specific debugfs entries under it.
1047 * Cleanup should be automatic done in kvm_destroy_vm_debugfs() recursively, so
1048 * a per-arch destroy interface is not needed.
1050 int __weak kvm_arch_create_vm_debugfs(struct kvm *kvm)
1055 static struct kvm *kvm_create_vm(unsigned long type)
1057 struct kvm *kvm = kvm_arch_alloc_vm();
1058 struct kvm_memslots *slots;
1063 return ERR_PTR(-ENOMEM);
1065 KVM_MMU_LOCK_INIT(kvm);
1066 mmgrab(current->mm);
1067 kvm->mm = current->mm;
1068 kvm_eventfd_init(kvm);
1069 mutex_init(&kvm->lock);
1070 mutex_init(&kvm->irq_lock);
1071 mutex_init(&kvm->slots_lock);
1072 mutex_init(&kvm->slots_arch_lock);
1073 spin_lock_init(&kvm->mn_invalidate_lock);
1074 rcuwait_init(&kvm->mn_memslots_update_rcuwait);
1075 xa_init(&kvm->vcpu_array);
1077 INIT_LIST_HEAD(&kvm->gpc_list);
1078 spin_lock_init(&kvm->gpc_lock);
1080 INIT_LIST_HEAD(&kvm->devices);
1082 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
1084 if (init_srcu_struct(&kvm->srcu))
1085 goto out_err_no_srcu;
1086 if (init_srcu_struct(&kvm->irq_srcu))
1087 goto out_err_no_irq_srcu;
1089 refcount_set(&kvm->users_count, 1);
1090 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
1091 for (j = 0; j < 2; j++) {
1092 slots = &kvm->__memslots[i][j];
1094 atomic_long_set(&slots->last_used_slot, (unsigned long)NULL);
1095 slots->hva_tree = RB_ROOT_CACHED;
1096 slots->gfn_tree = RB_ROOT;
1097 hash_init(slots->id_hash);
1098 slots->node_idx = j;
1100 /* Generations must be different for each address space. */
1101 slots->generation = i;
1104 rcu_assign_pointer(kvm->memslots[i], &kvm->__memslots[i][0]);
1107 for (i = 0; i < KVM_NR_BUSES; i++) {
1108 rcu_assign_pointer(kvm->buses[i],
1109 kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL_ACCOUNT));
1111 goto out_err_no_arch_destroy_vm;
1114 kvm->max_halt_poll_ns = halt_poll_ns;
1116 r = kvm_arch_init_vm(kvm, type);
1118 goto out_err_no_arch_destroy_vm;
1120 r = hardware_enable_all();
1122 goto out_err_no_disable;
1124 #ifdef CONFIG_HAVE_KVM_IRQFD
1125 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
1128 r = kvm_init_mmu_notifier(kvm);
1130 goto out_err_no_mmu_notifier;
1132 r = kvm_arch_post_init_vm(kvm);
1136 mutex_lock(&kvm_lock);
1137 list_add(&kvm->vm_list, &vm_list);
1138 mutex_unlock(&kvm_lock);
1140 preempt_notifier_inc();
1141 kvm_init_pm_notifier(kvm);
1146 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1147 if (kvm->mmu_notifier.ops)
1148 mmu_notifier_unregister(&kvm->mmu_notifier, current->mm);
1150 out_err_no_mmu_notifier:
1151 hardware_disable_all();
1153 kvm_arch_destroy_vm(kvm);
1154 out_err_no_arch_destroy_vm:
1155 WARN_ON_ONCE(!refcount_dec_and_test(&kvm->users_count));
1156 for (i = 0; i < KVM_NR_BUSES; i++)
1157 kfree(kvm_get_bus(kvm, i));
1158 cleanup_srcu_struct(&kvm->irq_srcu);
1159 out_err_no_irq_srcu:
1160 cleanup_srcu_struct(&kvm->srcu);
1162 kvm_arch_free_vm(kvm);
1163 mmdrop(current->mm);
1167 static void kvm_destroy_devices(struct kvm *kvm)
1169 struct kvm_device *dev, *tmp;
1172 * We do not need to take the kvm->lock here, because nobody else
1173 * has a reference to the struct kvm at this point and therefore
1174 * cannot access the devices list anyhow.
1176 list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
1177 list_del(&dev->vm_node);
1178 dev->ops->destroy(dev);
1182 static void kvm_destroy_vm(struct kvm *kvm)
1185 struct mm_struct *mm = kvm->mm;
1187 kvm_destroy_pm_notifier(kvm);
1188 kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm);
1189 kvm_destroy_vm_debugfs(kvm);
1190 kvm_arch_sync_events(kvm);
1191 mutex_lock(&kvm_lock);
1192 list_del(&kvm->vm_list);
1193 mutex_unlock(&kvm_lock);
1194 kvm_arch_pre_destroy_vm(kvm);
1196 kvm_free_irq_routing(kvm);
1197 for (i = 0; i < KVM_NR_BUSES; i++) {
1198 struct kvm_io_bus *bus = kvm_get_bus(kvm, i);
1201 kvm_io_bus_destroy(bus);
1202 kvm->buses[i] = NULL;
1204 kvm_coalesced_mmio_free(kvm);
1205 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1206 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
1208 * At this point, pending calls to invalidate_range_start()
1209 * have completed but no more MMU notifiers will run, so
1210 * mn_active_invalidate_count may remain unbalanced.
1211 * No threads can be waiting in install_new_memslots as the
1212 * last reference on KVM has been dropped, but freeing
1213 * memslots would deadlock without this manual intervention.
1215 WARN_ON(rcuwait_active(&kvm->mn_memslots_update_rcuwait));
1216 kvm->mn_active_invalidate_count = 0;
1218 kvm_arch_flush_shadow_all(kvm);
1220 kvm_arch_destroy_vm(kvm);
1221 kvm_destroy_devices(kvm);
1222 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
1223 kvm_free_memslots(kvm, &kvm->__memslots[i][0]);
1224 kvm_free_memslots(kvm, &kvm->__memslots[i][1]);
1226 cleanup_srcu_struct(&kvm->irq_srcu);
1227 cleanup_srcu_struct(&kvm->srcu);
1228 kvm_arch_free_vm(kvm);
1229 preempt_notifier_dec();
1230 hardware_disable_all();
1234 void kvm_get_kvm(struct kvm *kvm)
1236 refcount_inc(&kvm->users_count);
1238 EXPORT_SYMBOL_GPL(kvm_get_kvm);
1241 * Make sure the vm is not during destruction, which is a safe version of
1242 * kvm_get_kvm(). Return true if kvm referenced successfully, false otherwise.
1244 bool kvm_get_kvm_safe(struct kvm *kvm)
1246 return refcount_inc_not_zero(&kvm->users_count);
1248 EXPORT_SYMBOL_GPL(kvm_get_kvm_safe);
1250 void kvm_put_kvm(struct kvm *kvm)
1252 if (refcount_dec_and_test(&kvm->users_count))
1253 kvm_destroy_vm(kvm);
1255 EXPORT_SYMBOL_GPL(kvm_put_kvm);
1258 * Used to put a reference that was taken on behalf of an object associated
1259 * with a user-visible file descriptor, e.g. a vcpu or device, if installation
1260 * of the new file descriptor fails and the reference cannot be transferred to
1261 * its final owner. In such cases, the caller is still actively using @kvm and
1262 * will fail miserably if the refcount unexpectedly hits zero.
1264 void kvm_put_kvm_no_destroy(struct kvm *kvm)
1266 WARN_ON(refcount_dec_and_test(&kvm->users_count));
1268 EXPORT_SYMBOL_GPL(kvm_put_kvm_no_destroy);
1270 static int kvm_vm_release(struct inode *inode, struct file *filp)
1272 struct kvm *kvm = filp->private_data;
1274 kvm_irqfd_release(kvm);
1281 * Allocation size is twice as large as the actual dirty bitmap size.
1282 * See kvm_vm_ioctl_get_dirty_log() why this is needed.
1284 static int kvm_alloc_dirty_bitmap(struct kvm_memory_slot *memslot)
1286 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
1288 memslot->dirty_bitmap = kvzalloc(dirty_bytes, GFP_KERNEL_ACCOUNT);
1289 if (!memslot->dirty_bitmap)
1295 static struct kvm_memslots *kvm_get_inactive_memslots(struct kvm *kvm, int as_id)
1297 struct kvm_memslots *active = __kvm_memslots(kvm, as_id);
1298 int node_idx_inactive = active->node_idx ^ 1;
1300 return &kvm->__memslots[as_id][node_idx_inactive];
1304 * Helper to get the address space ID when one of memslot pointers may be NULL.
1305 * This also serves as a sanity that at least one of the pointers is non-NULL,
1306 * and that their address space IDs don't diverge.
1308 static int kvm_memslots_get_as_id(struct kvm_memory_slot *a,
1309 struct kvm_memory_slot *b)
1311 if (WARN_ON_ONCE(!a && !b))
1319 WARN_ON_ONCE(a->as_id != b->as_id);
1323 static void kvm_insert_gfn_node(struct kvm_memslots *slots,
1324 struct kvm_memory_slot *slot)
1326 struct rb_root *gfn_tree = &slots->gfn_tree;
1327 struct rb_node **node, *parent;
1328 int idx = slots->node_idx;
1331 for (node = &gfn_tree->rb_node; *node; ) {
1332 struct kvm_memory_slot *tmp;
1334 tmp = container_of(*node, struct kvm_memory_slot, gfn_node[idx]);
1336 if (slot->base_gfn < tmp->base_gfn)
1337 node = &(*node)->rb_left;
1338 else if (slot->base_gfn > tmp->base_gfn)
1339 node = &(*node)->rb_right;
1344 rb_link_node(&slot->gfn_node[idx], parent, node);
1345 rb_insert_color(&slot->gfn_node[idx], gfn_tree);
1348 static void kvm_erase_gfn_node(struct kvm_memslots *slots,
1349 struct kvm_memory_slot *slot)
1351 rb_erase(&slot->gfn_node[slots->node_idx], &slots->gfn_tree);
1354 static void kvm_replace_gfn_node(struct kvm_memslots *slots,
1355 struct kvm_memory_slot *old,
1356 struct kvm_memory_slot *new)
1358 int idx = slots->node_idx;
1360 WARN_ON_ONCE(old->base_gfn != new->base_gfn);
1362 rb_replace_node(&old->gfn_node[idx], &new->gfn_node[idx],
1367 * Replace @old with @new in the inactive memslots.
1369 * With NULL @old this simply adds @new.
1370 * With NULL @new this simply removes @old.
1372 * If @new is non-NULL its hva_node[slots_idx] range has to be set
1375 static void kvm_replace_memslot(struct kvm *kvm,
1376 struct kvm_memory_slot *old,
1377 struct kvm_memory_slot *new)
1379 int as_id = kvm_memslots_get_as_id(old, new);
1380 struct kvm_memslots *slots = kvm_get_inactive_memslots(kvm, as_id);
1381 int idx = slots->node_idx;
1384 hash_del(&old->id_node[idx]);
1385 interval_tree_remove(&old->hva_node[idx], &slots->hva_tree);
1387 if ((long)old == atomic_long_read(&slots->last_used_slot))
1388 atomic_long_set(&slots->last_used_slot, (long)new);
1391 kvm_erase_gfn_node(slots, old);
1397 * Initialize @new's hva range. Do this even when replacing an @old
1398 * slot, kvm_copy_memslot() deliberately does not touch node data.
1400 new->hva_node[idx].start = new->userspace_addr;
1401 new->hva_node[idx].last = new->userspace_addr +
1402 (new->npages << PAGE_SHIFT) - 1;
1405 * (Re)Add the new memslot. There is no O(1) interval_tree_replace(),
1406 * hva_node needs to be swapped with remove+insert even though hva can't
1407 * change when replacing an existing slot.
1409 hash_add(slots->id_hash, &new->id_node[idx], new->id);
1410 interval_tree_insert(&new->hva_node[idx], &slots->hva_tree);
1413 * If the memslot gfn is unchanged, rb_replace_node() can be used to
1414 * switch the node in the gfn tree instead of removing the old and
1415 * inserting the new as two separate operations. Replacement is a
1416 * single O(1) operation versus two O(log(n)) operations for
1419 if (old && old->base_gfn == new->base_gfn) {
1420 kvm_replace_gfn_node(slots, old, new);
1423 kvm_erase_gfn_node(slots, old);
1424 kvm_insert_gfn_node(slots, new);
1428 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
1430 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
1432 #ifdef __KVM_HAVE_READONLY_MEM
1433 valid_flags |= KVM_MEM_READONLY;
1436 if (mem->flags & ~valid_flags)
1442 static void kvm_swap_active_memslots(struct kvm *kvm, int as_id)
1444 struct kvm_memslots *slots = kvm_get_inactive_memslots(kvm, as_id);
1446 /* Grab the generation from the activate memslots. */
1447 u64 gen = __kvm_memslots(kvm, as_id)->generation;
1449 WARN_ON(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS);
1450 slots->generation = gen | KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
1453 * Do not store the new memslots while there are invalidations in
1454 * progress, otherwise the locking in invalidate_range_start and
1455 * invalidate_range_end will be unbalanced.
1457 spin_lock(&kvm->mn_invalidate_lock);
1458 prepare_to_rcuwait(&kvm->mn_memslots_update_rcuwait);
1459 while (kvm->mn_active_invalidate_count) {
1460 set_current_state(TASK_UNINTERRUPTIBLE);
1461 spin_unlock(&kvm->mn_invalidate_lock);
1463 spin_lock(&kvm->mn_invalidate_lock);
1465 finish_rcuwait(&kvm->mn_memslots_update_rcuwait);
1466 rcu_assign_pointer(kvm->memslots[as_id], slots);
1467 spin_unlock(&kvm->mn_invalidate_lock);
1470 * Acquired in kvm_set_memslot. Must be released before synchronize
1471 * SRCU below in order to avoid deadlock with another thread
1472 * acquiring the slots_arch_lock in an srcu critical section.
1474 mutex_unlock(&kvm->slots_arch_lock);
1476 synchronize_srcu_expedited(&kvm->srcu);
1479 * Increment the new memslot generation a second time, dropping the
1480 * update in-progress flag and incrementing the generation based on
1481 * the number of address spaces. This provides a unique and easily
1482 * identifiable generation number while the memslots are in flux.
1484 gen = slots->generation & ~KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
1487 * Generations must be unique even across address spaces. We do not need
1488 * a global counter for that, instead the generation space is evenly split
1489 * across address spaces. For example, with two address spaces, address
1490 * space 0 will use generations 0, 2, 4, ... while address space 1 will
1491 * use generations 1, 3, 5, ...
1493 gen += KVM_ADDRESS_SPACE_NUM;
1495 kvm_arch_memslots_updated(kvm, gen);
1497 slots->generation = gen;
1500 static int kvm_prepare_memory_region(struct kvm *kvm,
1501 const struct kvm_memory_slot *old,
1502 struct kvm_memory_slot *new,
1503 enum kvm_mr_change change)
1508 * If dirty logging is disabled, nullify the bitmap; the old bitmap
1509 * will be freed on "commit". If logging is enabled in both old and
1510 * new, reuse the existing bitmap. If logging is enabled only in the
1511 * new and KVM isn't using a ring buffer, allocate and initialize a
1514 if (change != KVM_MR_DELETE) {
1515 if (!(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
1516 new->dirty_bitmap = NULL;
1517 else if (old && old->dirty_bitmap)
1518 new->dirty_bitmap = old->dirty_bitmap;
1519 else if (!kvm->dirty_ring_size) {
1520 r = kvm_alloc_dirty_bitmap(new);
1524 if (kvm_dirty_log_manual_protect_and_init_set(kvm))
1525 bitmap_set(new->dirty_bitmap, 0, new->npages);
1529 r = kvm_arch_prepare_memory_region(kvm, old, new, change);
1531 /* Free the bitmap on failure if it was allocated above. */
1532 if (r && new && new->dirty_bitmap && old && !old->dirty_bitmap)
1533 kvm_destroy_dirty_bitmap(new);
1538 static void kvm_commit_memory_region(struct kvm *kvm,
1539 struct kvm_memory_slot *old,
1540 const struct kvm_memory_slot *new,
1541 enum kvm_mr_change change)
1544 * Update the total number of memslot pages before calling the arch
1545 * hook so that architectures can consume the result directly.
1547 if (change == KVM_MR_DELETE)
1548 kvm->nr_memslot_pages -= old->npages;
1549 else if (change == KVM_MR_CREATE)
1550 kvm->nr_memslot_pages += new->npages;
1552 kvm_arch_commit_memory_region(kvm, old, new, change);
1556 /* Nothing more to do. */
1559 /* Free the old memslot and all its metadata. */
1560 kvm_free_memslot(kvm, old);
1563 case KVM_MR_FLAGS_ONLY:
1565 * Free the dirty bitmap as needed; the below check encompasses
1566 * both the flags and whether a ring buffer is being used)
1568 if (old->dirty_bitmap && !new->dirty_bitmap)
1569 kvm_destroy_dirty_bitmap(old);
1572 * The final quirk. Free the detached, old slot, but only its
1573 * memory, not any metadata. Metadata, including arch specific
1574 * data, may be reused by @new.
1584 * Activate @new, which must be installed in the inactive slots by the caller,
1585 * by swapping the active slots and then propagating @new to @old once @old is
1586 * unreachable and can be safely modified.
1588 * With NULL @old this simply adds @new to @active (while swapping the sets).
1589 * With NULL @new this simply removes @old from @active and frees it
1590 * (while also swapping the sets).
1592 static void kvm_activate_memslot(struct kvm *kvm,
1593 struct kvm_memory_slot *old,
1594 struct kvm_memory_slot *new)
1596 int as_id = kvm_memslots_get_as_id(old, new);
1598 kvm_swap_active_memslots(kvm, as_id);
1600 /* Propagate the new memslot to the now inactive memslots. */
1601 kvm_replace_memslot(kvm, old, new);
1604 static void kvm_copy_memslot(struct kvm_memory_slot *dest,
1605 const struct kvm_memory_slot *src)
1607 dest->base_gfn = src->base_gfn;
1608 dest->npages = src->npages;
1609 dest->dirty_bitmap = src->dirty_bitmap;
1610 dest->arch = src->arch;
1611 dest->userspace_addr = src->userspace_addr;
1612 dest->flags = src->flags;
1614 dest->as_id = src->as_id;
1617 static void kvm_invalidate_memslot(struct kvm *kvm,
1618 struct kvm_memory_slot *old,
1619 struct kvm_memory_slot *invalid_slot)
1622 * Mark the current slot INVALID. As with all memslot modifications,
1623 * this must be done on an unreachable slot to avoid modifying the
1624 * current slot in the active tree.
1626 kvm_copy_memslot(invalid_slot, old);
1627 invalid_slot->flags |= KVM_MEMSLOT_INVALID;
1628 kvm_replace_memslot(kvm, old, invalid_slot);
1631 * Activate the slot that is now marked INVALID, but don't propagate
1632 * the slot to the now inactive slots. The slot is either going to be
1633 * deleted or recreated as a new slot.
1635 kvm_swap_active_memslots(kvm, old->as_id);
1638 * From this point no new shadow pages pointing to a deleted, or moved,
1639 * memslot will be created. Validation of sp->gfn happens in:
1640 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1641 * - kvm_is_visible_gfn (mmu_check_root)
1643 kvm_arch_flush_shadow_memslot(kvm, old);
1645 /* Was released by kvm_swap_active_memslots, reacquire. */
1646 mutex_lock(&kvm->slots_arch_lock);
1649 * Copy the arch-specific field of the newly-installed slot back to the
1650 * old slot as the arch data could have changed between releasing
1651 * slots_arch_lock in install_new_memslots() and re-acquiring the lock
1652 * above. Writers are required to retrieve memslots *after* acquiring
1653 * slots_arch_lock, thus the active slot's data is guaranteed to be fresh.
1655 old->arch = invalid_slot->arch;
1658 static void kvm_create_memslot(struct kvm *kvm,
1659 struct kvm_memory_slot *new)
1661 /* Add the new memslot to the inactive set and activate. */
1662 kvm_replace_memslot(kvm, NULL, new);
1663 kvm_activate_memslot(kvm, NULL, new);
1666 static void kvm_delete_memslot(struct kvm *kvm,
1667 struct kvm_memory_slot *old,
1668 struct kvm_memory_slot *invalid_slot)
1671 * Remove the old memslot (in the inactive memslots) by passing NULL as
1672 * the "new" slot, and for the invalid version in the active slots.
1674 kvm_replace_memslot(kvm, old, NULL);
1675 kvm_activate_memslot(kvm, invalid_slot, NULL);
1678 static void kvm_move_memslot(struct kvm *kvm,
1679 struct kvm_memory_slot *old,
1680 struct kvm_memory_slot *new,
1681 struct kvm_memory_slot *invalid_slot)
1684 * Replace the old memslot in the inactive slots, and then swap slots
1685 * and replace the current INVALID with the new as well.
1687 kvm_replace_memslot(kvm, old, new);
1688 kvm_activate_memslot(kvm, invalid_slot, new);
1691 static void kvm_update_flags_memslot(struct kvm *kvm,
1692 struct kvm_memory_slot *old,
1693 struct kvm_memory_slot *new)
1696 * Similar to the MOVE case, but the slot doesn't need to be zapped as
1697 * an intermediate step. Instead, the old memslot is simply replaced
1698 * with a new, updated copy in both memslot sets.
1700 kvm_replace_memslot(kvm, old, new);
1701 kvm_activate_memslot(kvm, old, new);
1704 static int kvm_set_memslot(struct kvm *kvm,
1705 struct kvm_memory_slot *old,
1706 struct kvm_memory_slot *new,
1707 enum kvm_mr_change change)
1709 struct kvm_memory_slot *invalid_slot;
1713 * Released in kvm_swap_active_memslots.
1715 * Must be held from before the current memslots are copied until
1716 * after the new memslots are installed with rcu_assign_pointer,
1717 * then released before the synchronize srcu in kvm_swap_active_memslots.
1719 * When modifying memslots outside of the slots_lock, must be held
1720 * before reading the pointer to the current memslots until after all
1721 * changes to those memslots are complete.
1723 * These rules ensure that installing new memslots does not lose
1724 * changes made to the previous memslots.
1726 mutex_lock(&kvm->slots_arch_lock);
1729 * Invalidate the old slot if it's being deleted or moved. This is
1730 * done prior to actually deleting/moving the memslot to allow vCPUs to
1731 * continue running by ensuring there are no mappings or shadow pages
1732 * for the memslot when it is deleted/moved. Without pre-invalidation
1733 * (and without a lock), a window would exist between effecting the
1734 * delete/move and committing the changes in arch code where KVM or a
1735 * guest could access a non-existent memslot.
1737 * Modifications are done on a temporary, unreachable slot. The old
1738 * slot needs to be preserved in case a later step fails and the
1739 * invalidation needs to be reverted.
1741 if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) {
1742 invalid_slot = kzalloc(sizeof(*invalid_slot), GFP_KERNEL_ACCOUNT);
1743 if (!invalid_slot) {
1744 mutex_unlock(&kvm->slots_arch_lock);
1747 kvm_invalidate_memslot(kvm, old, invalid_slot);
1750 r = kvm_prepare_memory_region(kvm, old, new, change);
1753 * For DELETE/MOVE, revert the above INVALID change. No
1754 * modifications required since the original slot was preserved
1755 * in the inactive slots. Changing the active memslots also
1756 * release slots_arch_lock.
1758 if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) {
1759 kvm_activate_memslot(kvm, invalid_slot, old);
1760 kfree(invalid_slot);
1762 mutex_unlock(&kvm->slots_arch_lock);
1768 * For DELETE and MOVE, the working slot is now active as the INVALID
1769 * version of the old slot. MOVE is particularly special as it reuses
1770 * the old slot and returns a copy of the old slot (in working_slot).
1771 * For CREATE, there is no old slot. For DELETE and FLAGS_ONLY, the
1772 * old slot is detached but otherwise preserved.
1774 if (change == KVM_MR_CREATE)
1775 kvm_create_memslot(kvm, new);
1776 else if (change == KVM_MR_DELETE)
1777 kvm_delete_memslot(kvm, old, invalid_slot);
1778 else if (change == KVM_MR_MOVE)
1779 kvm_move_memslot(kvm, old, new, invalid_slot);
1780 else if (change == KVM_MR_FLAGS_ONLY)
1781 kvm_update_flags_memslot(kvm, old, new);
1785 /* Free the temporary INVALID slot used for DELETE and MOVE. */
1786 if (change == KVM_MR_DELETE || change == KVM_MR_MOVE)
1787 kfree(invalid_slot);
1790 * No need to refresh new->arch, changes after dropping slots_arch_lock
1791 * will directly hit the final, active memsot. Architectures are
1792 * responsible for knowing that new->arch may be stale.
1794 kvm_commit_memory_region(kvm, old, new, change);
1799 static bool kvm_check_memslot_overlap(struct kvm_memslots *slots, int id,
1800 gfn_t start, gfn_t end)
1802 struct kvm_memslot_iter iter;
1804 kvm_for_each_memslot_in_gfn_range(&iter, slots, start, end) {
1805 if (iter.slot->id != id)
1813 * Allocate some memory and give it an address in the guest physical address
1816 * Discontiguous memory is allowed, mostly for framebuffers.
1818 * Must be called holding kvm->slots_lock for write.
1820 int __kvm_set_memory_region(struct kvm *kvm,
1821 const struct kvm_userspace_memory_region *mem)
1823 struct kvm_memory_slot *old, *new;
1824 struct kvm_memslots *slots;
1825 enum kvm_mr_change change;
1826 unsigned long npages;
1831 r = check_memory_region_flags(mem);
1835 as_id = mem->slot >> 16;
1836 id = (u16)mem->slot;
1838 /* General sanity checks */
1839 if ((mem->memory_size & (PAGE_SIZE - 1)) ||
1840 (mem->memory_size != (unsigned long)mem->memory_size))
1842 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
1844 /* We can read the guest memory with __xxx_user() later on. */
1845 if ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
1846 (mem->userspace_addr != untagged_addr(mem->userspace_addr)) ||
1847 !access_ok((void __user *)(unsigned long)mem->userspace_addr,
1850 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
1852 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
1854 if ((mem->memory_size >> PAGE_SHIFT) > KVM_MEM_MAX_NR_PAGES)
1857 slots = __kvm_memslots(kvm, as_id);
1860 * Note, the old memslot (and the pointer itself!) may be invalidated
1861 * and/or destroyed by kvm_set_memslot().
1863 old = id_to_memslot(slots, id);
1865 if (!mem->memory_size) {
1866 if (!old || !old->npages)
1869 if (WARN_ON_ONCE(kvm->nr_memslot_pages < old->npages))
1872 return kvm_set_memslot(kvm, old, NULL, KVM_MR_DELETE);
1875 base_gfn = (mem->guest_phys_addr >> PAGE_SHIFT);
1876 npages = (mem->memory_size >> PAGE_SHIFT);
1878 if (!old || !old->npages) {
1879 change = KVM_MR_CREATE;
1882 * To simplify KVM internals, the total number of pages across
1883 * all memslots must fit in an unsigned long.
1885 if ((kvm->nr_memslot_pages + npages) < kvm->nr_memslot_pages)
1887 } else { /* Modify an existing slot. */
1888 if ((mem->userspace_addr != old->userspace_addr) ||
1889 (npages != old->npages) ||
1890 ((mem->flags ^ old->flags) & KVM_MEM_READONLY))
1893 if (base_gfn != old->base_gfn)
1894 change = KVM_MR_MOVE;
1895 else if (mem->flags != old->flags)
1896 change = KVM_MR_FLAGS_ONLY;
1897 else /* Nothing to change. */
1901 if ((change == KVM_MR_CREATE || change == KVM_MR_MOVE) &&
1902 kvm_check_memslot_overlap(slots, id, base_gfn, base_gfn + npages))
1905 /* Allocate a slot that will persist in the memslot. */
1906 new = kzalloc(sizeof(*new), GFP_KERNEL_ACCOUNT);
1912 new->base_gfn = base_gfn;
1913 new->npages = npages;
1914 new->flags = mem->flags;
1915 new->userspace_addr = mem->userspace_addr;
1917 r = kvm_set_memslot(kvm, old, new, change);
1922 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1924 int kvm_set_memory_region(struct kvm *kvm,
1925 const struct kvm_userspace_memory_region *mem)
1929 mutex_lock(&kvm->slots_lock);
1930 r = __kvm_set_memory_region(kvm, mem);
1931 mutex_unlock(&kvm->slots_lock);
1934 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1936 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1937 struct kvm_userspace_memory_region *mem)
1939 if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1942 return kvm_set_memory_region(kvm, mem);
1945 #ifndef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1947 * kvm_get_dirty_log - get a snapshot of dirty pages
1948 * @kvm: pointer to kvm instance
1949 * @log: slot id and address to which we copy the log
1950 * @is_dirty: set to '1' if any dirty pages were found
1951 * @memslot: set to the associated memslot, always valid on success
1953 int kvm_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log,
1954 int *is_dirty, struct kvm_memory_slot **memslot)
1956 struct kvm_memslots *slots;
1959 unsigned long any = 0;
1961 /* Dirty ring tracking is exclusive to dirty log tracking */
1962 if (kvm->dirty_ring_size)
1968 as_id = log->slot >> 16;
1969 id = (u16)log->slot;
1970 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1973 slots = __kvm_memslots(kvm, as_id);
1974 *memslot = id_to_memslot(slots, id);
1975 if (!(*memslot) || !(*memslot)->dirty_bitmap)
1978 kvm_arch_sync_dirty_log(kvm, *memslot);
1980 n = kvm_dirty_bitmap_bytes(*memslot);
1982 for (i = 0; !any && i < n/sizeof(long); ++i)
1983 any = (*memslot)->dirty_bitmap[i];
1985 if (copy_to_user(log->dirty_bitmap, (*memslot)->dirty_bitmap, n))
1992 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1994 #else /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
1996 * kvm_get_dirty_log_protect - get a snapshot of dirty pages
1997 * and reenable dirty page tracking for the corresponding pages.
1998 * @kvm: pointer to kvm instance
1999 * @log: slot id and address to which we copy the log
2001 * We need to keep it in mind that VCPU threads can write to the bitmap
2002 * concurrently. So, to avoid losing track of dirty pages we keep the
2005 * 1. Take a snapshot of the bit and clear it if needed.
2006 * 2. Write protect the corresponding page.
2007 * 3. Copy the snapshot to the userspace.
2008 * 4. Upon return caller flushes TLB's if needed.
2010 * Between 2 and 4, the guest may write to the page using the remaining TLB
2011 * entry. This is not a problem because the page is reported dirty using
2012 * the snapshot taken before and step 4 ensures that writes done after
2013 * exiting to userspace will be logged for the next call.
2016 static int kvm_get_dirty_log_protect(struct kvm *kvm, struct kvm_dirty_log *log)
2018 struct kvm_memslots *slots;
2019 struct kvm_memory_slot *memslot;
2022 unsigned long *dirty_bitmap;
2023 unsigned long *dirty_bitmap_buffer;
2026 /* Dirty ring tracking is exclusive to dirty log tracking */
2027 if (kvm->dirty_ring_size)
2030 as_id = log->slot >> 16;
2031 id = (u16)log->slot;
2032 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
2035 slots = __kvm_memslots(kvm, as_id);
2036 memslot = id_to_memslot(slots, id);
2037 if (!memslot || !memslot->dirty_bitmap)
2040 dirty_bitmap = memslot->dirty_bitmap;
2042 kvm_arch_sync_dirty_log(kvm, memslot);
2044 n = kvm_dirty_bitmap_bytes(memslot);
2046 if (kvm->manual_dirty_log_protect) {
2048 * Unlike kvm_get_dirty_log, we always return false in *flush,
2049 * because no flush is needed until KVM_CLEAR_DIRTY_LOG. There
2050 * is some code duplication between this function and
2051 * kvm_get_dirty_log, but hopefully all architecture
2052 * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log
2053 * can be eliminated.
2055 dirty_bitmap_buffer = dirty_bitmap;
2057 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
2058 memset(dirty_bitmap_buffer, 0, n);
2061 for (i = 0; i < n / sizeof(long); i++) {
2065 if (!dirty_bitmap[i])
2069 mask = xchg(&dirty_bitmap[i], 0);
2070 dirty_bitmap_buffer[i] = mask;
2072 offset = i * BITS_PER_LONG;
2073 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
2076 KVM_MMU_UNLOCK(kvm);
2080 kvm_arch_flush_remote_tlbs_memslot(kvm, memslot);
2082 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
2089 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
2090 * @kvm: kvm instance
2091 * @log: slot id and address to which we copy the log
2093 * Steps 1-4 below provide general overview of dirty page logging. See
2094 * kvm_get_dirty_log_protect() function description for additional details.
2096 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
2097 * always flush the TLB (step 4) even if previous step failed and the dirty
2098 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
2099 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
2100 * writes will be marked dirty for next log read.
2102 * 1. Take a snapshot of the bit and clear it if needed.
2103 * 2. Write protect the corresponding page.
2104 * 3. Copy the snapshot to the userspace.
2105 * 4. Flush TLB's if needed.
2107 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
2108 struct kvm_dirty_log *log)
2112 mutex_lock(&kvm->slots_lock);
2114 r = kvm_get_dirty_log_protect(kvm, log);
2116 mutex_unlock(&kvm->slots_lock);
2121 * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap
2122 * and reenable dirty page tracking for the corresponding pages.
2123 * @kvm: pointer to kvm instance
2124 * @log: slot id and address from which to fetch the bitmap of dirty pages
2126 static int kvm_clear_dirty_log_protect(struct kvm *kvm,
2127 struct kvm_clear_dirty_log *log)
2129 struct kvm_memslots *slots;
2130 struct kvm_memory_slot *memslot;
2134 unsigned long *dirty_bitmap;
2135 unsigned long *dirty_bitmap_buffer;
2138 /* Dirty ring tracking is exclusive to dirty log tracking */
2139 if (kvm->dirty_ring_size)
2142 as_id = log->slot >> 16;
2143 id = (u16)log->slot;
2144 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
2147 if (log->first_page & 63)
2150 slots = __kvm_memslots(kvm, as_id);
2151 memslot = id_to_memslot(slots, id);
2152 if (!memslot || !memslot->dirty_bitmap)
2155 dirty_bitmap = memslot->dirty_bitmap;
2157 n = ALIGN(log->num_pages, BITS_PER_LONG) / 8;
2159 if (log->first_page > memslot->npages ||
2160 log->num_pages > memslot->npages - log->first_page ||
2161 (log->num_pages < memslot->npages - log->first_page && (log->num_pages & 63)))
2164 kvm_arch_sync_dirty_log(kvm, memslot);
2167 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
2168 if (copy_from_user(dirty_bitmap_buffer, log->dirty_bitmap, n))
2172 for (offset = log->first_page, i = offset / BITS_PER_LONG,
2173 n = DIV_ROUND_UP(log->num_pages, BITS_PER_LONG); n--;
2174 i++, offset += BITS_PER_LONG) {
2175 unsigned long mask = *dirty_bitmap_buffer++;
2176 atomic_long_t *p = (atomic_long_t *) &dirty_bitmap[i];
2180 mask &= atomic_long_fetch_andnot(mask, p);
2183 * mask contains the bits that really have been cleared. This
2184 * never includes any bits beyond the length of the memslot (if
2185 * the length is not aligned to 64 pages), therefore it is not
2186 * a problem if userspace sets them in log->dirty_bitmap.
2190 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
2194 KVM_MMU_UNLOCK(kvm);
2197 kvm_arch_flush_remote_tlbs_memslot(kvm, memslot);
2202 static int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm,
2203 struct kvm_clear_dirty_log *log)
2207 mutex_lock(&kvm->slots_lock);
2209 r = kvm_clear_dirty_log_protect(kvm, log);
2211 mutex_unlock(&kvm->slots_lock);
2214 #endif /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
2216 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
2218 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
2220 EXPORT_SYMBOL_GPL(gfn_to_memslot);
2222 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
2224 struct kvm_memslots *slots = kvm_vcpu_memslots(vcpu);
2225 u64 gen = slots->generation;
2226 struct kvm_memory_slot *slot;
2229 * This also protects against using a memslot from a different address space,
2230 * since different address spaces have different generation numbers.
2232 if (unlikely(gen != vcpu->last_used_slot_gen)) {
2233 vcpu->last_used_slot = NULL;
2234 vcpu->last_used_slot_gen = gen;
2237 slot = try_get_memslot(vcpu->last_used_slot, gfn);
2242 * Fall back to searching all memslots. We purposely use
2243 * search_memslots() instead of __gfn_to_memslot() to avoid
2244 * thrashing the VM-wide last_used_slot in kvm_memslots.
2246 slot = search_memslots(slots, gfn, false);
2248 vcpu->last_used_slot = slot;
2254 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_memslot);
2256 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
2258 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
2260 return kvm_is_visible_memslot(memslot);
2262 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
2264 bool kvm_vcpu_is_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
2266 struct kvm_memory_slot *memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2268 return kvm_is_visible_memslot(memslot);
2270 EXPORT_SYMBOL_GPL(kvm_vcpu_is_visible_gfn);
2272 unsigned long kvm_host_page_size(struct kvm_vcpu *vcpu, gfn_t gfn)
2274 struct vm_area_struct *vma;
2275 unsigned long addr, size;
2279 addr = kvm_vcpu_gfn_to_hva_prot(vcpu, gfn, NULL);
2280 if (kvm_is_error_hva(addr))
2283 mmap_read_lock(current->mm);
2284 vma = find_vma(current->mm, addr);
2288 size = vma_kernel_pagesize(vma);
2291 mmap_read_unlock(current->mm);
2296 static bool memslot_is_readonly(const struct kvm_memory_slot *slot)
2298 return slot->flags & KVM_MEM_READONLY;
2301 static unsigned long __gfn_to_hva_many(const struct kvm_memory_slot *slot, gfn_t gfn,
2302 gfn_t *nr_pages, bool write)
2304 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
2305 return KVM_HVA_ERR_BAD;
2307 if (memslot_is_readonly(slot) && write)
2308 return KVM_HVA_ERR_RO_BAD;
2311 *nr_pages = slot->npages - (gfn - slot->base_gfn);
2313 return __gfn_to_hva_memslot(slot, gfn);
2316 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
2319 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
2322 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
2325 return gfn_to_hva_many(slot, gfn, NULL);
2327 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
2329 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
2331 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
2333 EXPORT_SYMBOL_GPL(gfn_to_hva);
2335 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
2337 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
2339 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
2342 * Return the hva of a @gfn and the R/W attribute if possible.
2344 * @slot: the kvm_memory_slot which contains @gfn
2345 * @gfn: the gfn to be translated
2346 * @writable: used to return the read/write attribute of the @slot if the hva
2347 * is valid and @writable is not NULL
2349 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
2350 gfn_t gfn, bool *writable)
2352 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
2354 if (!kvm_is_error_hva(hva) && writable)
2355 *writable = !memslot_is_readonly(slot);
2360 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
2362 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2364 return gfn_to_hva_memslot_prot(slot, gfn, writable);
2367 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
2369 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2371 return gfn_to_hva_memslot_prot(slot, gfn, writable);
2374 static inline int check_user_page_hwpoison(unsigned long addr)
2376 int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
2378 rc = get_user_pages(addr, 1, flags, NULL, NULL);
2379 return rc == -EHWPOISON;
2383 * The fast path to get the writable pfn which will be stored in @pfn,
2384 * true indicates success, otherwise false is returned. It's also the
2385 * only part that runs if we can in atomic context.
2387 static bool hva_to_pfn_fast(unsigned long addr, bool write_fault,
2388 bool *writable, kvm_pfn_t *pfn)
2390 struct page *page[1];
2393 * Fast pin a writable pfn only if it is a write fault request
2394 * or the caller allows to map a writable pfn for a read fault
2397 if (!(write_fault || writable))
2400 if (get_user_page_fast_only(addr, FOLL_WRITE, page)) {
2401 *pfn = page_to_pfn(page[0]);
2412 * The slow path to get the pfn of the specified host virtual address,
2413 * 1 indicates success, -errno is returned if error is detected.
2415 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
2416 bool *writable, kvm_pfn_t *pfn)
2418 unsigned int flags = FOLL_HWPOISON;
2425 *writable = write_fault;
2428 flags |= FOLL_WRITE;
2430 flags |= FOLL_NOWAIT;
2432 npages = get_user_pages_unlocked(addr, 1, &page, flags);
2436 /* map read fault as writable if possible */
2437 if (unlikely(!write_fault) && writable) {
2440 if (get_user_page_fast_only(addr, FOLL_WRITE, &wpage)) {
2446 *pfn = page_to_pfn(page);
2450 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
2452 if (unlikely(!(vma->vm_flags & VM_READ)))
2455 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
2461 static int kvm_try_get_pfn(kvm_pfn_t pfn)
2463 if (kvm_is_reserved_pfn(pfn))
2465 return get_page_unless_zero(pfn_to_page(pfn));
2468 static int hva_to_pfn_remapped(struct vm_area_struct *vma,
2469 unsigned long addr, bool *async,
2470 bool write_fault, bool *writable,
2478 r = follow_pte(vma->vm_mm, addr, &ptep, &ptl);
2481 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
2482 * not call the fault handler, so do it here.
2484 bool unlocked = false;
2485 r = fixup_user_fault(current->mm, addr,
2486 (write_fault ? FAULT_FLAG_WRITE : 0),
2493 r = follow_pte(vma->vm_mm, addr, &ptep, &ptl);
2498 if (write_fault && !pte_write(*ptep)) {
2499 pfn = KVM_PFN_ERR_RO_FAULT;
2504 *writable = pte_write(*ptep);
2505 pfn = pte_pfn(*ptep);
2508 * Get a reference here because callers of *hva_to_pfn* and
2509 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
2510 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
2511 * set, but the kvm_try_get_pfn/kvm_release_pfn_clean pair will
2512 * simply do nothing for reserved pfns.
2514 * Whoever called remap_pfn_range is also going to call e.g.
2515 * unmap_mapping_range before the underlying pages are freed,
2516 * causing a call to our MMU notifier.
2518 * Certain IO or PFNMAP mappings can be backed with valid
2519 * struct pages, but be allocated without refcounting e.g.,
2520 * tail pages of non-compound higher order allocations, which
2521 * would then underflow the refcount when the caller does the
2522 * required put_page. Don't allow those pages here.
2524 if (!kvm_try_get_pfn(pfn))
2528 pte_unmap_unlock(ptep, ptl);
2535 * Pin guest page in memory and return its pfn.
2536 * @addr: host virtual address which maps memory to the guest
2537 * @atomic: whether this function can sleep
2538 * @async: whether this function need to wait IO complete if the
2539 * host page is not in the memory
2540 * @write_fault: whether we should get a writable host page
2541 * @writable: whether it allows to map a writable host page for !@write_fault
2543 * The function will map a writable host page for these two cases:
2544 * 1): @write_fault = true
2545 * 2): @write_fault = false && @writable, @writable will tell the caller
2546 * whether the mapping is writable.
2548 kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
2549 bool write_fault, bool *writable)
2551 struct vm_area_struct *vma;
2555 /* we can do it either atomically or asynchronously, not both */
2556 BUG_ON(atomic && async);
2558 if (hva_to_pfn_fast(addr, write_fault, writable, &pfn))
2562 return KVM_PFN_ERR_FAULT;
2564 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
2568 mmap_read_lock(current->mm);
2569 if (npages == -EHWPOISON ||
2570 (!async && check_user_page_hwpoison(addr))) {
2571 pfn = KVM_PFN_ERR_HWPOISON;
2576 vma = vma_lookup(current->mm, addr);
2579 pfn = KVM_PFN_ERR_FAULT;
2580 else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
2581 r = hva_to_pfn_remapped(vma, addr, async, write_fault, writable, &pfn);
2585 pfn = KVM_PFN_ERR_FAULT;
2587 if (async && vma_is_valid(vma, write_fault))
2589 pfn = KVM_PFN_ERR_FAULT;
2592 mmap_read_unlock(current->mm);
2596 kvm_pfn_t __gfn_to_pfn_memslot(const struct kvm_memory_slot *slot, gfn_t gfn,
2597 bool atomic, bool *async, bool write_fault,
2598 bool *writable, hva_t *hva)
2600 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
2605 if (addr == KVM_HVA_ERR_RO_BAD) {
2608 return KVM_PFN_ERR_RO_FAULT;
2611 if (kvm_is_error_hva(addr)) {
2614 return KVM_PFN_NOSLOT;
2617 /* Do not map writable pfn in the readonly memslot. */
2618 if (writable && memslot_is_readonly(slot)) {
2623 return hva_to_pfn(addr, atomic, async, write_fault,
2626 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
2628 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
2631 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
2632 write_fault, writable, NULL);
2634 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
2636 kvm_pfn_t gfn_to_pfn_memslot(const struct kvm_memory_slot *slot, gfn_t gfn)
2638 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL, NULL);
2640 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
2642 kvm_pfn_t gfn_to_pfn_memslot_atomic(const struct kvm_memory_slot *slot, gfn_t gfn)
2644 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL, NULL);
2646 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
2648 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
2650 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
2652 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
2654 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
2656 return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
2658 EXPORT_SYMBOL_GPL(gfn_to_pfn);
2660 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
2662 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
2664 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
2666 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
2667 struct page **pages, int nr_pages)
2672 addr = gfn_to_hva_many(slot, gfn, &entry);
2673 if (kvm_is_error_hva(addr))
2676 if (entry < nr_pages)
2679 return get_user_pages_fast_only(addr, nr_pages, FOLL_WRITE, pages);
2681 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
2683 static struct page *kvm_pfn_to_page(kvm_pfn_t pfn)
2685 if (is_error_noslot_pfn(pfn))
2686 return KVM_ERR_PTR_BAD_PAGE;
2688 if (kvm_is_reserved_pfn(pfn)) {
2690 return KVM_ERR_PTR_BAD_PAGE;
2693 return pfn_to_page(pfn);
2696 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
2700 pfn = gfn_to_pfn(kvm, gfn);
2702 return kvm_pfn_to_page(pfn);
2704 EXPORT_SYMBOL_GPL(gfn_to_page);
2706 void kvm_release_pfn(kvm_pfn_t pfn, bool dirty)
2712 kvm_release_pfn_dirty(pfn);
2714 kvm_release_pfn_clean(pfn);
2717 int kvm_vcpu_map(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map)
2721 struct page *page = KVM_UNMAPPED_PAGE;
2726 pfn = gfn_to_pfn(vcpu->kvm, gfn);
2727 if (is_error_noslot_pfn(pfn))
2730 if (pfn_valid(pfn)) {
2731 page = pfn_to_page(pfn);
2733 #ifdef CONFIG_HAS_IOMEM
2735 hva = memremap(pfn_to_hpa(pfn), PAGE_SIZE, MEMREMAP_WB);
2749 EXPORT_SYMBOL_GPL(kvm_vcpu_map);
2751 void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map, bool dirty)
2759 if (map->page != KVM_UNMAPPED_PAGE)
2761 #ifdef CONFIG_HAS_IOMEM
2767 kvm_vcpu_mark_page_dirty(vcpu, map->gfn);
2769 kvm_release_pfn(map->pfn, dirty);
2774 EXPORT_SYMBOL_GPL(kvm_vcpu_unmap);
2776 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
2780 pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
2782 return kvm_pfn_to_page(pfn);
2784 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
2786 void kvm_release_page_clean(struct page *page)
2788 WARN_ON(is_error_page(page));
2790 kvm_release_pfn_clean(page_to_pfn(page));
2792 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
2794 void kvm_release_pfn_clean(kvm_pfn_t pfn)
2796 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
2797 put_page(pfn_to_page(pfn));
2799 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
2801 void kvm_release_page_dirty(struct page *page)
2803 WARN_ON(is_error_page(page));
2805 kvm_release_pfn_dirty(page_to_pfn(page));
2807 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
2809 void kvm_release_pfn_dirty(kvm_pfn_t pfn)
2811 kvm_set_pfn_dirty(pfn);
2812 kvm_release_pfn_clean(pfn);
2814 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
2816 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
2818 if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn))
2819 SetPageDirty(pfn_to_page(pfn));
2821 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
2823 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
2825 if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn))
2826 mark_page_accessed(pfn_to_page(pfn));
2828 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
2830 static int next_segment(unsigned long len, int offset)
2832 if (len > PAGE_SIZE - offset)
2833 return PAGE_SIZE - offset;
2838 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
2839 void *data, int offset, int len)
2844 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
2845 if (kvm_is_error_hva(addr))
2847 r = __copy_from_user(data, (void __user *)addr + offset, len);
2853 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
2856 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2858 return __kvm_read_guest_page(slot, gfn, data, offset, len);
2860 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
2862 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
2863 int offset, int len)
2865 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2867 return __kvm_read_guest_page(slot, gfn, data, offset, len);
2869 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
2871 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
2873 gfn_t gfn = gpa >> PAGE_SHIFT;
2875 int offset = offset_in_page(gpa);
2878 while ((seg = next_segment(len, offset)) != 0) {
2879 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
2889 EXPORT_SYMBOL_GPL(kvm_read_guest);
2891 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
2893 gfn_t gfn = gpa >> PAGE_SHIFT;
2895 int offset = offset_in_page(gpa);
2898 while ((seg = next_segment(len, offset)) != 0) {
2899 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
2909 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
2911 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
2912 void *data, int offset, unsigned long len)
2917 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
2918 if (kvm_is_error_hva(addr))
2920 pagefault_disable();
2921 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
2928 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
2929 void *data, unsigned long len)
2931 gfn_t gfn = gpa >> PAGE_SHIFT;
2932 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2933 int offset = offset_in_page(gpa);
2935 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
2937 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
2939 static int __kvm_write_guest_page(struct kvm *kvm,
2940 struct kvm_memory_slot *memslot, gfn_t gfn,
2941 const void *data, int offset, int len)
2946 addr = gfn_to_hva_memslot(memslot, gfn);
2947 if (kvm_is_error_hva(addr))
2949 r = __copy_to_user((void __user *)addr + offset, data, len);
2952 mark_page_dirty_in_slot(kvm, memslot, gfn);
2956 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
2957 const void *data, int offset, int len)
2959 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2961 return __kvm_write_guest_page(kvm, slot, gfn, data, offset, len);
2963 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
2965 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
2966 const void *data, int offset, int len)
2968 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2970 return __kvm_write_guest_page(vcpu->kvm, slot, gfn, data, offset, len);
2972 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
2974 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
2977 gfn_t gfn = gpa >> PAGE_SHIFT;
2979 int offset = offset_in_page(gpa);
2982 while ((seg = next_segment(len, offset)) != 0) {
2983 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
2993 EXPORT_SYMBOL_GPL(kvm_write_guest);
2995 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
2998 gfn_t gfn = gpa >> PAGE_SHIFT;
3000 int offset = offset_in_page(gpa);
3003 while ((seg = next_segment(len, offset)) != 0) {
3004 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
3014 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
3016 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
3017 struct gfn_to_hva_cache *ghc,
3018 gpa_t gpa, unsigned long len)
3020 int offset = offset_in_page(gpa);
3021 gfn_t start_gfn = gpa >> PAGE_SHIFT;
3022 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
3023 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
3024 gfn_t nr_pages_avail;
3026 /* Update ghc->generation before performing any error checks. */
3027 ghc->generation = slots->generation;
3029 if (start_gfn > end_gfn) {
3030 ghc->hva = KVM_HVA_ERR_BAD;
3035 * If the requested region crosses two memslots, we still
3036 * verify that the entire region is valid here.
3038 for ( ; start_gfn <= end_gfn; start_gfn += nr_pages_avail) {
3039 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
3040 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
3042 if (kvm_is_error_hva(ghc->hva))
3046 /* Use the slow path for cross page reads and writes. */
3047 if (nr_pages_needed == 1)
3050 ghc->memslot = NULL;
3057 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3058 gpa_t gpa, unsigned long len)
3060 struct kvm_memslots *slots = kvm_memslots(kvm);
3061 return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len);
3063 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
3065 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3066 void *data, unsigned int offset,
3069 struct kvm_memslots *slots = kvm_memslots(kvm);
3071 gpa_t gpa = ghc->gpa + offset;
3073 if (WARN_ON_ONCE(len + offset > ghc->len))
3076 if (slots->generation != ghc->generation) {
3077 if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len))
3081 if (kvm_is_error_hva(ghc->hva))
3084 if (unlikely(!ghc->memslot))
3085 return kvm_write_guest(kvm, gpa, data, len);
3087 r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
3090 mark_page_dirty_in_slot(kvm, ghc->memslot, gpa >> PAGE_SHIFT);
3094 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached);
3096 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3097 void *data, unsigned long len)
3099 return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len);
3101 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
3103 int kvm_read_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3104 void *data, unsigned int offset,
3107 struct kvm_memslots *slots = kvm_memslots(kvm);
3109 gpa_t gpa = ghc->gpa + offset;
3111 if (WARN_ON_ONCE(len + offset > ghc->len))
3114 if (slots->generation != ghc->generation) {
3115 if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len))
3119 if (kvm_is_error_hva(ghc->hva))
3122 if (unlikely(!ghc->memslot))
3123 return kvm_read_guest(kvm, gpa, data, len);
3125 r = __copy_from_user(data, (void __user *)ghc->hva + offset, len);
3131 EXPORT_SYMBOL_GPL(kvm_read_guest_offset_cached);
3133 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3134 void *data, unsigned long len)
3136 return kvm_read_guest_offset_cached(kvm, ghc, data, 0, len);
3138 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
3140 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
3142 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
3143 gfn_t gfn = gpa >> PAGE_SHIFT;
3145 int offset = offset_in_page(gpa);
3148 while ((seg = next_segment(len, offset)) != 0) {
3149 ret = kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
3158 EXPORT_SYMBOL_GPL(kvm_clear_guest);
3160 void mark_page_dirty_in_slot(struct kvm *kvm,
3161 const struct kvm_memory_slot *memslot,
3164 struct kvm_vcpu *vcpu = kvm_get_running_vcpu();
3166 if (WARN_ON_ONCE(!vcpu) || WARN_ON_ONCE(vcpu->kvm != kvm))
3169 if (memslot && kvm_slot_dirty_track_enabled(memslot)) {
3170 unsigned long rel_gfn = gfn - memslot->base_gfn;
3171 u32 slot = (memslot->as_id << 16) | memslot->id;
3173 if (kvm->dirty_ring_size)
3174 kvm_dirty_ring_push(&vcpu->dirty_ring,
3177 set_bit_le(rel_gfn, memslot->dirty_bitmap);
3180 EXPORT_SYMBOL_GPL(mark_page_dirty_in_slot);
3182 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
3184 struct kvm_memory_slot *memslot;
3186 memslot = gfn_to_memslot(kvm, gfn);
3187 mark_page_dirty_in_slot(kvm, memslot, gfn);
3189 EXPORT_SYMBOL_GPL(mark_page_dirty);
3191 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
3193 struct kvm_memory_slot *memslot;
3195 memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
3196 mark_page_dirty_in_slot(vcpu->kvm, memslot, gfn);
3198 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
3200 void kvm_sigset_activate(struct kvm_vcpu *vcpu)
3202 if (!vcpu->sigset_active)
3206 * This does a lockless modification of ->real_blocked, which is fine
3207 * because, only current can change ->real_blocked and all readers of
3208 * ->real_blocked don't care as long ->real_blocked is always a subset
3211 sigprocmask(SIG_SETMASK, &vcpu->sigset, ¤t->real_blocked);
3214 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu)
3216 if (!vcpu->sigset_active)
3219 sigprocmask(SIG_SETMASK, ¤t->real_blocked, NULL);
3220 sigemptyset(¤t->real_blocked);
3223 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
3225 unsigned int old, val, grow, grow_start;
3227 old = val = vcpu->halt_poll_ns;
3228 grow_start = READ_ONCE(halt_poll_ns_grow_start);
3229 grow = READ_ONCE(halt_poll_ns_grow);
3234 if (val < grow_start)
3237 if (val > vcpu->kvm->max_halt_poll_ns)
3238 val = vcpu->kvm->max_halt_poll_ns;
3240 vcpu->halt_poll_ns = val;
3242 trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
3245 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
3247 unsigned int old, val, shrink, grow_start;
3249 old = val = vcpu->halt_poll_ns;
3250 shrink = READ_ONCE(halt_poll_ns_shrink);
3251 grow_start = READ_ONCE(halt_poll_ns_grow_start);
3257 if (val < grow_start)
3260 vcpu->halt_poll_ns = val;
3261 trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
3264 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
3267 int idx = srcu_read_lock(&vcpu->kvm->srcu);
3269 if (kvm_arch_vcpu_runnable(vcpu)) {
3270 kvm_make_request(KVM_REQ_UNHALT, vcpu);
3273 if (kvm_cpu_has_pending_timer(vcpu))
3275 if (signal_pending(current))
3277 if (kvm_check_request(KVM_REQ_UNBLOCK, vcpu))
3282 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3287 * Block the vCPU until the vCPU is runnable, an event arrives, or a signal is
3288 * pending. This is mostly used when halting a vCPU, but may also be used
3289 * directly for other vCPU non-runnable states, e.g. x86's Wait-For-SIPI.
3291 bool kvm_vcpu_block(struct kvm_vcpu *vcpu)
3293 struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu);
3294 bool waited = false;
3296 vcpu->stat.generic.blocking = 1;
3298 kvm_arch_vcpu_blocking(vcpu);
3300 prepare_to_rcuwait(wait);
3302 set_current_state(TASK_INTERRUPTIBLE);
3304 if (kvm_vcpu_check_block(vcpu) < 0)
3310 finish_rcuwait(wait);
3312 kvm_arch_vcpu_unblocking(vcpu);
3314 vcpu->stat.generic.blocking = 0;
3319 static inline void update_halt_poll_stats(struct kvm_vcpu *vcpu, ktime_t start,
3320 ktime_t end, bool success)
3322 struct kvm_vcpu_stat_generic *stats = &vcpu->stat.generic;
3323 u64 poll_ns = ktime_to_ns(ktime_sub(end, start));
3325 ++vcpu->stat.generic.halt_attempted_poll;
3328 ++vcpu->stat.generic.halt_successful_poll;
3330 if (!vcpu_valid_wakeup(vcpu))
3331 ++vcpu->stat.generic.halt_poll_invalid;
3333 stats->halt_poll_success_ns += poll_ns;
3334 KVM_STATS_LOG_HIST_UPDATE(stats->halt_poll_success_hist, poll_ns);
3336 stats->halt_poll_fail_ns += poll_ns;
3337 KVM_STATS_LOG_HIST_UPDATE(stats->halt_poll_fail_hist, poll_ns);
3342 * Emulate a vCPU halt condition, e.g. HLT on x86, WFI on arm, etc... If halt
3343 * polling is enabled, busy wait for a short time before blocking to avoid the
3344 * expensive block+unblock sequence if a wake event arrives soon after the vCPU
3347 void kvm_vcpu_halt(struct kvm_vcpu *vcpu)
3349 bool halt_poll_allowed = !kvm_arch_no_poll(vcpu);
3350 bool do_halt_poll = halt_poll_allowed && vcpu->halt_poll_ns;
3351 ktime_t start, cur, poll_end;
3352 bool waited = false;
3355 start = cur = poll_end = ktime_get();
3357 ktime_t stop = ktime_add_ns(start, vcpu->halt_poll_ns);
3361 * This sets KVM_REQ_UNHALT if an interrupt
3364 if (kvm_vcpu_check_block(vcpu) < 0)
3367 poll_end = cur = ktime_get();
3368 } while (kvm_vcpu_can_poll(cur, stop));
3371 waited = kvm_vcpu_block(vcpu);
3375 vcpu->stat.generic.halt_wait_ns +=
3376 ktime_to_ns(cur) - ktime_to_ns(poll_end);
3377 KVM_STATS_LOG_HIST_UPDATE(vcpu->stat.generic.halt_wait_hist,
3378 ktime_to_ns(cur) - ktime_to_ns(poll_end));
3381 /* The total time the vCPU was "halted", including polling time. */
3382 halt_ns = ktime_to_ns(cur) - ktime_to_ns(start);
3385 * Note, halt-polling is considered successful so long as the vCPU was
3386 * never actually scheduled out, i.e. even if the wake event arrived
3387 * after of the halt-polling loop itself, but before the full wait.
3390 update_halt_poll_stats(vcpu, start, poll_end, !waited);
3392 if (halt_poll_allowed) {
3393 if (!vcpu_valid_wakeup(vcpu)) {
3394 shrink_halt_poll_ns(vcpu);
3395 } else if (vcpu->kvm->max_halt_poll_ns) {
3396 if (halt_ns <= vcpu->halt_poll_ns)
3398 /* we had a long block, shrink polling */
3399 else if (vcpu->halt_poll_ns &&
3400 halt_ns > vcpu->kvm->max_halt_poll_ns)
3401 shrink_halt_poll_ns(vcpu);
3402 /* we had a short halt and our poll time is too small */
3403 else if (vcpu->halt_poll_ns < vcpu->kvm->max_halt_poll_ns &&
3404 halt_ns < vcpu->kvm->max_halt_poll_ns)
3405 grow_halt_poll_ns(vcpu);
3407 vcpu->halt_poll_ns = 0;
3411 trace_kvm_vcpu_wakeup(halt_ns, waited, vcpu_valid_wakeup(vcpu));
3413 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
3415 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
3417 if (__kvm_vcpu_wake_up(vcpu)) {
3418 WRITE_ONCE(vcpu->ready, true);
3419 ++vcpu->stat.generic.halt_wakeup;
3425 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
3429 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
3431 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
3435 if (kvm_vcpu_wake_up(vcpu))
3440 * The only state change done outside the vcpu mutex is IN_GUEST_MODE
3441 * to EXITING_GUEST_MODE. Therefore the moderately expensive "should
3442 * kick" check does not need atomic operations if kvm_vcpu_kick is used
3443 * within the vCPU thread itself.
3445 if (vcpu == __this_cpu_read(kvm_running_vcpu)) {
3446 if (vcpu->mode == IN_GUEST_MODE)
3447 WRITE_ONCE(vcpu->mode, EXITING_GUEST_MODE);
3452 * Note, the vCPU could get migrated to a different pCPU at any point
3453 * after kvm_arch_vcpu_should_kick(), which could result in sending an
3454 * IPI to the previous pCPU. But, that's ok because the purpose of the
3455 * IPI is to force the vCPU to leave IN_GUEST_MODE, and migrating the
3456 * vCPU also requires it to leave IN_GUEST_MODE.
3458 if (kvm_arch_vcpu_should_kick(vcpu)) {
3459 cpu = READ_ONCE(vcpu->cpu);
3460 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
3461 smp_send_reschedule(cpu);
3466 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
3467 #endif /* !CONFIG_S390 */
3469 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
3472 struct task_struct *task = NULL;
3476 pid = rcu_dereference(target->pid);
3478 task = get_pid_task(pid, PIDTYPE_PID);
3482 ret = yield_to(task, 1);
3483 put_task_struct(task);
3487 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
3490 * Helper that checks whether a VCPU is eligible for directed yield.
3491 * Most eligible candidate to yield is decided by following heuristics:
3493 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
3494 * (preempted lock holder), indicated by @in_spin_loop.
3495 * Set at the beginning and cleared at the end of interception/PLE handler.
3497 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
3498 * chance last time (mostly it has become eligible now since we have probably
3499 * yielded to lockholder in last iteration. This is done by toggling
3500 * @dy_eligible each time a VCPU checked for eligibility.)
3502 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
3503 * to preempted lock-holder could result in wrong VCPU selection and CPU
3504 * burning. Giving priority for a potential lock-holder increases lock
3507 * Since algorithm is based on heuristics, accessing another VCPU data without
3508 * locking does not harm. It may result in trying to yield to same VCPU, fail
3509 * and continue with next VCPU and so on.
3511 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
3513 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
3516 eligible = !vcpu->spin_loop.in_spin_loop ||
3517 vcpu->spin_loop.dy_eligible;
3519 if (vcpu->spin_loop.in_spin_loop)
3520 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
3529 * Unlike kvm_arch_vcpu_runnable, this function is called outside
3530 * a vcpu_load/vcpu_put pair. However, for most architectures
3531 * kvm_arch_vcpu_runnable does not require vcpu_load.
3533 bool __weak kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
3535 return kvm_arch_vcpu_runnable(vcpu);
3538 static bool vcpu_dy_runnable(struct kvm_vcpu *vcpu)
3540 if (kvm_arch_dy_runnable(vcpu))
3543 #ifdef CONFIG_KVM_ASYNC_PF
3544 if (!list_empty_careful(&vcpu->async_pf.done))
3551 bool __weak kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu)
3556 void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode)
3558 struct kvm *kvm = me->kvm;
3559 struct kvm_vcpu *vcpu;
3560 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
3566 kvm_vcpu_set_in_spin_loop(me, true);
3568 * We boost the priority of a VCPU that is runnable but not
3569 * currently running, because it got preempted by something
3570 * else and called schedule in __vcpu_run. Hopefully that
3571 * VCPU is holding the lock that we need and will release it.
3572 * We approximate round-robin by starting at the last boosted VCPU.
3574 for (pass = 0; pass < 2 && !yielded && try; pass++) {
3575 kvm_for_each_vcpu(i, vcpu, kvm) {
3576 if (!pass && i <= last_boosted_vcpu) {
3577 i = last_boosted_vcpu;
3579 } else if (pass && i > last_boosted_vcpu)
3581 if (!READ_ONCE(vcpu->ready))
3585 if (kvm_vcpu_is_blocking(vcpu) && !vcpu_dy_runnable(vcpu))
3587 if (READ_ONCE(vcpu->preempted) && yield_to_kernel_mode &&
3588 !kvm_arch_dy_has_pending_interrupt(vcpu) &&
3589 !kvm_arch_vcpu_in_kernel(vcpu))
3591 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
3594 yielded = kvm_vcpu_yield_to(vcpu);
3596 kvm->last_boosted_vcpu = i;
3598 } else if (yielded < 0) {
3605 kvm_vcpu_set_in_spin_loop(me, false);
3607 /* Ensure vcpu is not eligible during next spinloop */
3608 kvm_vcpu_set_dy_eligible(me, false);
3610 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
3612 static bool kvm_page_in_dirty_ring(struct kvm *kvm, unsigned long pgoff)
3614 #ifdef CONFIG_HAVE_KVM_DIRTY_RING
3615 return (pgoff >= KVM_DIRTY_LOG_PAGE_OFFSET) &&
3616 (pgoff < KVM_DIRTY_LOG_PAGE_OFFSET +
3617 kvm->dirty_ring_size / PAGE_SIZE);
3623 static vm_fault_t kvm_vcpu_fault(struct vm_fault *vmf)
3625 struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data;
3628 if (vmf->pgoff == 0)
3629 page = virt_to_page(vcpu->run);
3631 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
3632 page = virt_to_page(vcpu->arch.pio_data);
3634 #ifdef CONFIG_KVM_MMIO
3635 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
3636 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
3638 else if (kvm_page_in_dirty_ring(vcpu->kvm, vmf->pgoff))
3639 page = kvm_dirty_ring_get_page(
3641 vmf->pgoff - KVM_DIRTY_LOG_PAGE_OFFSET);
3643 return kvm_arch_vcpu_fault(vcpu, vmf);
3649 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
3650 .fault = kvm_vcpu_fault,
3653 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
3655 struct kvm_vcpu *vcpu = file->private_data;
3656 unsigned long pages = vma_pages(vma);
3658 if ((kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff) ||
3659 kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff + pages - 1)) &&
3660 ((vma->vm_flags & VM_EXEC) || !(vma->vm_flags & VM_SHARED)))
3663 vma->vm_ops = &kvm_vcpu_vm_ops;
3667 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
3669 struct kvm_vcpu *vcpu = filp->private_data;
3671 kvm_put_kvm(vcpu->kvm);
3675 static struct file_operations kvm_vcpu_fops = {
3676 .release = kvm_vcpu_release,
3677 .unlocked_ioctl = kvm_vcpu_ioctl,
3678 .mmap = kvm_vcpu_mmap,
3679 .llseek = noop_llseek,
3680 KVM_COMPAT(kvm_vcpu_compat_ioctl),
3684 * Allocates an inode for the vcpu.
3686 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
3688 char name[8 + 1 + ITOA_MAX_LEN + 1];
3690 snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id);
3691 return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
3694 static void kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
3696 #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS
3697 struct dentry *debugfs_dentry;
3698 char dir_name[ITOA_MAX_LEN * 2];
3700 if (!debugfs_initialized())
3703 snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
3704 debugfs_dentry = debugfs_create_dir(dir_name,
3705 vcpu->kvm->debugfs_dentry);
3707 kvm_arch_create_vcpu_debugfs(vcpu, debugfs_dentry);
3712 * Creates some virtual cpus. Good luck creating more than one.
3714 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
3717 struct kvm_vcpu *vcpu;
3720 if (id >= KVM_MAX_VCPU_IDS)
3723 mutex_lock(&kvm->lock);
3724 if (kvm->created_vcpus == KVM_MAX_VCPUS) {
3725 mutex_unlock(&kvm->lock);
3729 kvm->created_vcpus++;
3730 mutex_unlock(&kvm->lock);
3732 r = kvm_arch_vcpu_precreate(kvm, id);
3734 goto vcpu_decrement;
3736 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL_ACCOUNT);
3739 goto vcpu_decrement;
3742 BUILD_BUG_ON(sizeof(struct kvm_run) > PAGE_SIZE);
3743 page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
3748 vcpu->run = page_address(page);
3750 kvm_vcpu_init(vcpu, kvm, id);
3752 r = kvm_arch_vcpu_create(vcpu);
3754 goto vcpu_free_run_page;
3756 if (kvm->dirty_ring_size) {
3757 r = kvm_dirty_ring_alloc(&vcpu->dirty_ring,
3758 id, kvm->dirty_ring_size);
3760 goto arch_vcpu_destroy;
3763 mutex_lock(&kvm->lock);
3764 if (kvm_get_vcpu_by_id(kvm, id)) {
3766 goto unlock_vcpu_destroy;
3769 vcpu->vcpu_idx = atomic_read(&kvm->online_vcpus);
3770 r = xa_insert(&kvm->vcpu_array, vcpu->vcpu_idx, vcpu, GFP_KERNEL_ACCOUNT);
3771 BUG_ON(r == -EBUSY);
3773 goto unlock_vcpu_destroy;
3775 /* Fill the stats id string for the vcpu */
3776 snprintf(vcpu->stats_id, sizeof(vcpu->stats_id), "kvm-%d/vcpu-%d",
3777 task_pid_nr(current), id);
3779 /* Now it's all set up, let userspace reach it */
3781 r = create_vcpu_fd(vcpu);
3783 xa_erase(&kvm->vcpu_array, vcpu->vcpu_idx);
3784 kvm_put_kvm_no_destroy(kvm);
3785 goto unlock_vcpu_destroy;
3789 * Pairs with smp_rmb() in kvm_get_vcpu. Store the vcpu
3790 * pointer before kvm->online_vcpu's incremented value.
3793 atomic_inc(&kvm->online_vcpus);
3795 mutex_unlock(&kvm->lock);
3796 kvm_arch_vcpu_postcreate(vcpu);
3797 kvm_create_vcpu_debugfs(vcpu);
3800 unlock_vcpu_destroy:
3801 mutex_unlock(&kvm->lock);
3802 kvm_dirty_ring_free(&vcpu->dirty_ring);
3804 kvm_arch_vcpu_destroy(vcpu);
3806 free_page((unsigned long)vcpu->run);
3808 kmem_cache_free(kvm_vcpu_cache, vcpu);
3810 mutex_lock(&kvm->lock);
3811 kvm->created_vcpus--;
3812 mutex_unlock(&kvm->lock);
3816 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
3819 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
3820 vcpu->sigset_active = 1;
3821 vcpu->sigset = *sigset;
3823 vcpu->sigset_active = 0;
3827 static ssize_t kvm_vcpu_stats_read(struct file *file, char __user *user_buffer,
3828 size_t size, loff_t *offset)
3830 struct kvm_vcpu *vcpu = file->private_data;
3832 return kvm_stats_read(vcpu->stats_id, &kvm_vcpu_stats_header,
3833 &kvm_vcpu_stats_desc[0], &vcpu->stat,
3834 sizeof(vcpu->stat), user_buffer, size, offset);
3837 static const struct file_operations kvm_vcpu_stats_fops = {
3838 .read = kvm_vcpu_stats_read,
3839 .llseek = noop_llseek,
3842 static int kvm_vcpu_ioctl_get_stats_fd(struct kvm_vcpu *vcpu)
3846 char name[15 + ITOA_MAX_LEN + 1];
3848 snprintf(name, sizeof(name), "kvm-vcpu-stats:%d", vcpu->vcpu_id);
3850 fd = get_unused_fd_flags(O_CLOEXEC);
3854 file = anon_inode_getfile(name, &kvm_vcpu_stats_fops, vcpu, O_RDONLY);
3857 return PTR_ERR(file);
3859 file->f_mode |= FMODE_PREAD;
3860 fd_install(fd, file);
3865 static long kvm_vcpu_ioctl(struct file *filp,
3866 unsigned int ioctl, unsigned long arg)
3868 struct kvm_vcpu *vcpu = filp->private_data;
3869 void __user *argp = (void __user *)arg;
3871 struct kvm_fpu *fpu = NULL;
3872 struct kvm_sregs *kvm_sregs = NULL;
3874 if (vcpu->kvm->mm != current->mm || vcpu->kvm->vm_dead)
3877 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
3881 * Some architectures have vcpu ioctls that are asynchronous to vcpu
3882 * execution; mutex_lock() would break them.
3884 r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg);
3885 if (r != -ENOIOCTLCMD)
3888 if (mutex_lock_killable(&vcpu->mutex))
3896 oldpid = rcu_access_pointer(vcpu->pid);
3897 if (unlikely(oldpid != task_pid(current))) {
3898 /* The thread running this VCPU changed. */
3901 r = kvm_arch_vcpu_run_pid_change(vcpu);
3905 newpid = get_task_pid(current, PIDTYPE_PID);
3906 rcu_assign_pointer(vcpu->pid, newpid);
3911 r = kvm_arch_vcpu_ioctl_run(vcpu);
3912 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
3915 case KVM_GET_REGS: {
3916 struct kvm_regs *kvm_regs;
3919 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL_ACCOUNT);
3922 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
3926 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
3933 case KVM_SET_REGS: {
3934 struct kvm_regs *kvm_regs;
3936 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
3937 if (IS_ERR(kvm_regs)) {
3938 r = PTR_ERR(kvm_regs);
3941 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
3945 case KVM_GET_SREGS: {
3946 kvm_sregs = kzalloc(sizeof(struct kvm_sregs),
3947 GFP_KERNEL_ACCOUNT);
3951 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
3955 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
3960 case KVM_SET_SREGS: {
3961 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
3962 if (IS_ERR(kvm_sregs)) {
3963 r = PTR_ERR(kvm_sregs);
3967 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
3970 case KVM_GET_MP_STATE: {
3971 struct kvm_mp_state mp_state;
3973 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
3977 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
3982 case KVM_SET_MP_STATE: {
3983 struct kvm_mp_state mp_state;
3986 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
3988 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
3991 case KVM_TRANSLATE: {
3992 struct kvm_translation tr;
3995 if (copy_from_user(&tr, argp, sizeof(tr)))
3997 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
4001 if (copy_to_user(argp, &tr, sizeof(tr)))
4006 case KVM_SET_GUEST_DEBUG: {
4007 struct kvm_guest_debug dbg;
4010 if (copy_from_user(&dbg, argp, sizeof(dbg)))
4012 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
4015 case KVM_SET_SIGNAL_MASK: {
4016 struct kvm_signal_mask __user *sigmask_arg = argp;
4017 struct kvm_signal_mask kvm_sigmask;
4018 sigset_t sigset, *p;
4023 if (copy_from_user(&kvm_sigmask, argp,
4024 sizeof(kvm_sigmask)))
4027 if (kvm_sigmask.len != sizeof(sigset))
4030 if (copy_from_user(&sigset, sigmask_arg->sigset,
4035 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
4039 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL_ACCOUNT);
4043 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
4047 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
4053 fpu = memdup_user(argp, sizeof(*fpu));
4059 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
4062 case KVM_GET_STATS_FD: {
4063 r = kvm_vcpu_ioctl_get_stats_fd(vcpu);
4067 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
4070 mutex_unlock(&vcpu->mutex);
4076 #ifdef CONFIG_KVM_COMPAT
4077 static long kvm_vcpu_compat_ioctl(struct file *filp,
4078 unsigned int ioctl, unsigned long arg)
4080 struct kvm_vcpu *vcpu = filp->private_data;
4081 void __user *argp = compat_ptr(arg);
4084 if (vcpu->kvm->mm != current->mm || vcpu->kvm->vm_dead)
4088 case KVM_SET_SIGNAL_MASK: {
4089 struct kvm_signal_mask __user *sigmask_arg = argp;
4090 struct kvm_signal_mask kvm_sigmask;
4095 if (copy_from_user(&kvm_sigmask, argp,
4096 sizeof(kvm_sigmask)))
4099 if (kvm_sigmask.len != sizeof(compat_sigset_t))
4102 if (get_compat_sigset(&sigset,
4103 (compat_sigset_t __user *)sigmask_arg->sigset))
4105 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
4107 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
4111 r = kvm_vcpu_ioctl(filp, ioctl, arg);
4119 static int kvm_device_mmap(struct file *filp, struct vm_area_struct *vma)
4121 struct kvm_device *dev = filp->private_data;
4124 return dev->ops->mmap(dev, vma);
4129 static int kvm_device_ioctl_attr(struct kvm_device *dev,
4130 int (*accessor)(struct kvm_device *dev,
4131 struct kvm_device_attr *attr),
4134 struct kvm_device_attr attr;
4139 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
4142 return accessor(dev, &attr);
4145 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
4148 struct kvm_device *dev = filp->private_data;
4150 if (dev->kvm->mm != current->mm || dev->kvm->vm_dead)
4154 case KVM_SET_DEVICE_ATTR:
4155 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
4156 case KVM_GET_DEVICE_ATTR:
4157 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
4158 case KVM_HAS_DEVICE_ATTR:
4159 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
4161 if (dev->ops->ioctl)
4162 return dev->ops->ioctl(dev, ioctl, arg);
4168 static int kvm_device_release(struct inode *inode, struct file *filp)
4170 struct kvm_device *dev = filp->private_data;
4171 struct kvm *kvm = dev->kvm;
4173 if (dev->ops->release) {
4174 mutex_lock(&kvm->lock);
4175 list_del(&dev->vm_node);
4176 dev->ops->release(dev);
4177 mutex_unlock(&kvm->lock);
4184 static const struct file_operations kvm_device_fops = {
4185 .unlocked_ioctl = kvm_device_ioctl,
4186 .release = kvm_device_release,
4187 KVM_COMPAT(kvm_device_ioctl),
4188 .mmap = kvm_device_mmap,
4191 struct kvm_device *kvm_device_from_filp(struct file *filp)
4193 if (filp->f_op != &kvm_device_fops)
4196 return filp->private_data;
4199 static const struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
4200 #ifdef CONFIG_KVM_MPIC
4201 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
4202 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
4206 int kvm_register_device_ops(const struct kvm_device_ops *ops, u32 type)
4208 if (type >= ARRAY_SIZE(kvm_device_ops_table))
4211 if (kvm_device_ops_table[type] != NULL)
4214 kvm_device_ops_table[type] = ops;
4218 void kvm_unregister_device_ops(u32 type)
4220 if (kvm_device_ops_table[type] != NULL)
4221 kvm_device_ops_table[type] = NULL;
4224 static int kvm_ioctl_create_device(struct kvm *kvm,
4225 struct kvm_create_device *cd)
4227 const struct kvm_device_ops *ops = NULL;
4228 struct kvm_device *dev;
4229 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
4233 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
4236 type = array_index_nospec(cd->type, ARRAY_SIZE(kvm_device_ops_table));
4237 ops = kvm_device_ops_table[type];
4244 dev = kzalloc(sizeof(*dev), GFP_KERNEL_ACCOUNT);
4251 mutex_lock(&kvm->lock);
4252 ret = ops->create(dev, type);
4254 mutex_unlock(&kvm->lock);
4258 list_add(&dev->vm_node, &kvm->devices);
4259 mutex_unlock(&kvm->lock);
4265 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
4267 kvm_put_kvm_no_destroy(kvm);
4268 mutex_lock(&kvm->lock);
4269 list_del(&dev->vm_node);
4270 mutex_unlock(&kvm->lock);
4279 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
4282 case KVM_CAP_USER_MEMORY:
4283 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
4284 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
4285 case KVM_CAP_INTERNAL_ERROR_DATA:
4286 #ifdef CONFIG_HAVE_KVM_MSI
4287 case KVM_CAP_SIGNAL_MSI:
4289 #ifdef CONFIG_HAVE_KVM_IRQFD
4291 case KVM_CAP_IRQFD_RESAMPLE:
4293 case KVM_CAP_IOEVENTFD_ANY_LENGTH:
4294 case KVM_CAP_CHECK_EXTENSION_VM:
4295 case KVM_CAP_ENABLE_CAP_VM:
4296 case KVM_CAP_HALT_POLL:
4298 #ifdef CONFIG_KVM_MMIO
4299 case KVM_CAP_COALESCED_MMIO:
4300 return KVM_COALESCED_MMIO_PAGE_OFFSET;
4301 case KVM_CAP_COALESCED_PIO:
4304 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4305 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2:
4306 return KVM_DIRTY_LOG_MANUAL_CAPS;
4308 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
4309 case KVM_CAP_IRQ_ROUTING:
4310 return KVM_MAX_IRQ_ROUTES;
4312 #if KVM_ADDRESS_SPACE_NUM > 1
4313 case KVM_CAP_MULTI_ADDRESS_SPACE:
4314 return KVM_ADDRESS_SPACE_NUM;
4316 case KVM_CAP_NR_MEMSLOTS:
4317 return KVM_USER_MEM_SLOTS;
4318 case KVM_CAP_DIRTY_LOG_RING:
4319 #ifdef CONFIG_HAVE_KVM_DIRTY_RING
4320 return KVM_DIRTY_RING_MAX_ENTRIES * sizeof(struct kvm_dirty_gfn);
4324 case KVM_CAP_BINARY_STATS_FD:
4329 return kvm_vm_ioctl_check_extension(kvm, arg);
4332 static int kvm_vm_ioctl_enable_dirty_log_ring(struct kvm *kvm, u32 size)
4336 if (!KVM_DIRTY_LOG_PAGE_OFFSET)
4339 /* the size should be power of 2 */
4340 if (!size || (size & (size - 1)))
4343 /* Should be bigger to keep the reserved entries, or a page */
4344 if (size < kvm_dirty_ring_get_rsvd_entries() *
4345 sizeof(struct kvm_dirty_gfn) || size < PAGE_SIZE)
4348 if (size > KVM_DIRTY_RING_MAX_ENTRIES *
4349 sizeof(struct kvm_dirty_gfn))
4352 /* We only allow it to set once */
4353 if (kvm->dirty_ring_size)
4356 mutex_lock(&kvm->lock);
4358 if (kvm->created_vcpus) {
4359 /* We don't allow to change this value after vcpu created */
4362 kvm->dirty_ring_size = size;
4366 mutex_unlock(&kvm->lock);
4370 static int kvm_vm_ioctl_reset_dirty_pages(struct kvm *kvm)
4373 struct kvm_vcpu *vcpu;
4376 if (!kvm->dirty_ring_size)
4379 mutex_lock(&kvm->slots_lock);
4381 kvm_for_each_vcpu(i, vcpu, kvm)
4382 cleared += kvm_dirty_ring_reset(vcpu->kvm, &vcpu->dirty_ring);
4384 mutex_unlock(&kvm->slots_lock);
4387 kvm_flush_remote_tlbs(kvm);
4392 int __attribute__((weak)) kvm_vm_ioctl_enable_cap(struct kvm *kvm,
4393 struct kvm_enable_cap *cap)
4398 static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm,
4399 struct kvm_enable_cap *cap)
4402 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4403 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2: {
4404 u64 allowed_options = KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE;
4406 if (cap->args[0] & KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE)
4407 allowed_options = KVM_DIRTY_LOG_MANUAL_CAPS;
4409 if (cap->flags || (cap->args[0] & ~allowed_options))
4411 kvm->manual_dirty_log_protect = cap->args[0];
4415 case KVM_CAP_HALT_POLL: {
4416 if (cap->flags || cap->args[0] != (unsigned int)cap->args[0])
4419 kvm->max_halt_poll_ns = cap->args[0];
4422 case KVM_CAP_DIRTY_LOG_RING:
4423 return kvm_vm_ioctl_enable_dirty_log_ring(kvm, cap->args[0]);
4425 return kvm_vm_ioctl_enable_cap(kvm, cap);
4429 static ssize_t kvm_vm_stats_read(struct file *file, char __user *user_buffer,
4430 size_t size, loff_t *offset)
4432 struct kvm *kvm = file->private_data;
4434 return kvm_stats_read(kvm->stats_id, &kvm_vm_stats_header,
4435 &kvm_vm_stats_desc[0], &kvm->stat,
4436 sizeof(kvm->stat), user_buffer, size, offset);
4439 static const struct file_operations kvm_vm_stats_fops = {
4440 .read = kvm_vm_stats_read,
4441 .llseek = noop_llseek,
4444 static int kvm_vm_ioctl_get_stats_fd(struct kvm *kvm)
4449 fd = get_unused_fd_flags(O_CLOEXEC);
4453 file = anon_inode_getfile("kvm-vm-stats",
4454 &kvm_vm_stats_fops, kvm, O_RDONLY);
4457 return PTR_ERR(file);
4459 file->f_mode |= FMODE_PREAD;
4460 fd_install(fd, file);
4465 static long kvm_vm_ioctl(struct file *filp,
4466 unsigned int ioctl, unsigned long arg)
4468 struct kvm *kvm = filp->private_data;
4469 void __user *argp = (void __user *)arg;
4472 if (kvm->mm != current->mm || kvm->vm_dead)
4475 case KVM_CREATE_VCPU:
4476 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
4478 case KVM_ENABLE_CAP: {
4479 struct kvm_enable_cap cap;
4482 if (copy_from_user(&cap, argp, sizeof(cap)))
4484 r = kvm_vm_ioctl_enable_cap_generic(kvm, &cap);
4487 case KVM_SET_USER_MEMORY_REGION: {
4488 struct kvm_userspace_memory_region kvm_userspace_mem;
4491 if (copy_from_user(&kvm_userspace_mem, argp,
4492 sizeof(kvm_userspace_mem)))
4495 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
4498 case KVM_GET_DIRTY_LOG: {
4499 struct kvm_dirty_log log;
4502 if (copy_from_user(&log, argp, sizeof(log)))
4504 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
4507 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4508 case KVM_CLEAR_DIRTY_LOG: {
4509 struct kvm_clear_dirty_log log;
4512 if (copy_from_user(&log, argp, sizeof(log)))
4514 r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
4518 #ifdef CONFIG_KVM_MMIO
4519 case KVM_REGISTER_COALESCED_MMIO: {
4520 struct kvm_coalesced_mmio_zone zone;
4523 if (copy_from_user(&zone, argp, sizeof(zone)))
4525 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
4528 case KVM_UNREGISTER_COALESCED_MMIO: {
4529 struct kvm_coalesced_mmio_zone zone;
4532 if (copy_from_user(&zone, argp, sizeof(zone)))
4534 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
4539 struct kvm_irqfd data;
4542 if (copy_from_user(&data, argp, sizeof(data)))
4544 r = kvm_irqfd(kvm, &data);
4547 case KVM_IOEVENTFD: {
4548 struct kvm_ioeventfd data;
4551 if (copy_from_user(&data, argp, sizeof(data)))
4553 r = kvm_ioeventfd(kvm, &data);
4556 #ifdef CONFIG_HAVE_KVM_MSI
4557 case KVM_SIGNAL_MSI: {
4561 if (copy_from_user(&msi, argp, sizeof(msi)))
4563 r = kvm_send_userspace_msi(kvm, &msi);
4567 #ifdef __KVM_HAVE_IRQ_LINE
4568 case KVM_IRQ_LINE_STATUS:
4569 case KVM_IRQ_LINE: {
4570 struct kvm_irq_level irq_event;
4573 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
4576 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
4577 ioctl == KVM_IRQ_LINE_STATUS);
4582 if (ioctl == KVM_IRQ_LINE_STATUS) {
4583 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
4591 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
4592 case KVM_SET_GSI_ROUTING: {
4593 struct kvm_irq_routing routing;
4594 struct kvm_irq_routing __user *urouting;
4595 struct kvm_irq_routing_entry *entries = NULL;
4598 if (copy_from_user(&routing, argp, sizeof(routing)))
4601 if (!kvm_arch_can_set_irq_routing(kvm))
4603 if (routing.nr > KVM_MAX_IRQ_ROUTES)
4609 entries = vmemdup_user(urouting->entries,
4610 array_size(sizeof(*entries),
4612 if (IS_ERR(entries)) {
4613 r = PTR_ERR(entries);
4617 r = kvm_set_irq_routing(kvm, entries, routing.nr,
4622 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
4623 case KVM_CREATE_DEVICE: {
4624 struct kvm_create_device cd;
4627 if (copy_from_user(&cd, argp, sizeof(cd)))
4630 r = kvm_ioctl_create_device(kvm, &cd);
4635 if (copy_to_user(argp, &cd, sizeof(cd)))
4641 case KVM_CHECK_EXTENSION:
4642 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
4644 case KVM_RESET_DIRTY_RINGS:
4645 r = kvm_vm_ioctl_reset_dirty_pages(kvm);
4647 case KVM_GET_STATS_FD:
4648 r = kvm_vm_ioctl_get_stats_fd(kvm);
4651 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
4657 #ifdef CONFIG_KVM_COMPAT
4658 struct compat_kvm_dirty_log {
4662 compat_uptr_t dirty_bitmap; /* one bit per page */
4667 struct compat_kvm_clear_dirty_log {
4672 compat_uptr_t dirty_bitmap; /* one bit per page */
4677 static long kvm_vm_compat_ioctl(struct file *filp,
4678 unsigned int ioctl, unsigned long arg)
4680 struct kvm *kvm = filp->private_data;
4683 if (kvm->mm != current->mm || kvm->vm_dead)
4686 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4687 case KVM_CLEAR_DIRTY_LOG: {
4688 struct compat_kvm_clear_dirty_log compat_log;
4689 struct kvm_clear_dirty_log log;
4691 if (copy_from_user(&compat_log, (void __user *)arg,
4692 sizeof(compat_log)))
4694 log.slot = compat_log.slot;
4695 log.num_pages = compat_log.num_pages;
4696 log.first_page = compat_log.first_page;
4697 log.padding2 = compat_log.padding2;
4698 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
4700 r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
4704 case KVM_GET_DIRTY_LOG: {
4705 struct compat_kvm_dirty_log compat_log;
4706 struct kvm_dirty_log log;
4708 if (copy_from_user(&compat_log, (void __user *)arg,
4709 sizeof(compat_log)))
4711 log.slot = compat_log.slot;
4712 log.padding1 = compat_log.padding1;
4713 log.padding2 = compat_log.padding2;
4714 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
4716 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
4720 r = kvm_vm_ioctl(filp, ioctl, arg);
4726 static struct file_operations kvm_vm_fops = {
4727 .release = kvm_vm_release,
4728 .unlocked_ioctl = kvm_vm_ioctl,
4729 .llseek = noop_llseek,
4730 KVM_COMPAT(kvm_vm_compat_ioctl),
4733 bool file_is_kvm(struct file *file)
4735 return file && file->f_op == &kvm_vm_fops;
4737 EXPORT_SYMBOL_GPL(file_is_kvm);
4739 static int kvm_dev_ioctl_create_vm(unsigned long type)
4745 kvm = kvm_create_vm(type);
4747 return PTR_ERR(kvm);
4748 #ifdef CONFIG_KVM_MMIO
4749 r = kvm_coalesced_mmio_init(kvm);
4753 r = get_unused_fd_flags(O_CLOEXEC);
4757 snprintf(kvm->stats_id, sizeof(kvm->stats_id),
4758 "kvm-%d", task_pid_nr(current));
4760 file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
4768 * Don't call kvm_put_kvm anymore at this point; file->f_op is
4769 * already set, with ->release() being kvm_vm_release(). In error
4770 * cases it will be called by the final fput(file) and will take
4771 * care of doing kvm_put_kvm(kvm).
4773 if (kvm_create_vm_debugfs(kvm, r) < 0) {
4778 kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm);
4780 fd_install(r, file);
4788 static long kvm_dev_ioctl(struct file *filp,
4789 unsigned int ioctl, unsigned long arg)
4794 case KVM_GET_API_VERSION:
4797 r = KVM_API_VERSION;
4800 r = kvm_dev_ioctl_create_vm(arg);
4802 case KVM_CHECK_EXTENSION:
4803 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
4805 case KVM_GET_VCPU_MMAP_SIZE:
4808 r = PAGE_SIZE; /* struct kvm_run */
4810 r += PAGE_SIZE; /* pio data page */
4812 #ifdef CONFIG_KVM_MMIO
4813 r += PAGE_SIZE; /* coalesced mmio ring page */
4816 case KVM_TRACE_ENABLE:
4817 case KVM_TRACE_PAUSE:
4818 case KVM_TRACE_DISABLE:
4822 return kvm_arch_dev_ioctl(filp, ioctl, arg);
4828 static struct file_operations kvm_chardev_ops = {
4829 .unlocked_ioctl = kvm_dev_ioctl,
4830 .llseek = noop_llseek,
4831 KVM_COMPAT(kvm_dev_ioctl),
4834 static struct miscdevice kvm_dev = {
4840 static void hardware_enable_nolock(void *junk)
4842 int cpu = raw_smp_processor_id();
4845 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
4848 cpumask_set_cpu(cpu, cpus_hardware_enabled);
4850 r = kvm_arch_hardware_enable();
4853 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
4854 atomic_inc(&hardware_enable_failed);
4855 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
4859 static int kvm_starting_cpu(unsigned int cpu)
4861 raw_spin_lock(&kvm_count_lock);
4862 if (kvm_usage_count)
4863 hardware_enable_nolock(NULL);
4864 raw_spin_unlock(&kvm_count_lock);
4868 static void hardware_disable_nolock(void *junk)
4870 int cpu = raw_smp_processor_id();
4872 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
4874 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
4875 kvm_arch_hardware_disable();
4878 static int kvm_dying_cpu(unsigned int cpu)
4880 raw_spin_lock(&kvm_count_lock);
4881 if (kvm_usage_count)
4882 hardware_disable_nolock(NULL);
4883 raw_spin_unlock(&kvm_count_lock);
4887 static void hardware_disable_all_nolock(void)
4889 BUG_ON(!kvm_usage_count);
4892 if (!kvm_usage_count)
4893 on_each_cpu(hardware_disable_nolock, NULL, 1);
4896 static void hardware_disable_all(void)
4898 raw_spin_lock(&kvm_count_lock);
4899 hardware_disable_all_nolock();
4900 raw_spin_unlock(&kvm_count_lock);
4903 static int hardware_enable_all(void)
4907 raw_spin_lock(&kvm_count_lock);
4910 if (kvm_usage_count == 1) {
4911 atomic_set(&hardware_enable_failed, 0);
4912 on_each_cpu(hardware_enable_nolock, NULL, 1);
4914 if (atomic_read(&hardware_enable_failed)) {
4915 hardware_disable_all_nolock();
4920 raw_spin_unlock(&kvm_count_lock);
4925 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
4929 * Some (well, at least mine) BIOSes hang on reboot if
4932 * And Intel TXT required VMX off for all cpu when system shutdown.
4934 pr_info("kvm: exiting hardware virtualization\n");
4935 kvm_rebooting = true;
4936 on_each_cpu(hardware_disable_nolock, NULL, 1);
4940 static struct notifier_block kvm_reboot_notifier = {
4941 .notifier_call = kvm_reboot,
4945 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
4949 for (i = 0; i < bus->dev_count; i++) {
4950 struct kvm_io_device *pos = bus->range[i].dev;
4952 kvm_iodevice_destructor(pos);
4957 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
4958 const struct kvm_io_range *r2)
4960 gpa_t addr1 = r1->addr;
4961 gpa_t addr2 = r2->addr;
4966 /* If r2->len == 0, match the exact address. If r2->len != 0,
4967 * accept any overlapping write. Any order is acceptable for
4968 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
4969 * we process all of them.
4982 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
4984 return kvm_io_bus_cmp(p1, p2);
4987 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
4988 gpa_t addr, int len)
4990 struct kvm_io_range *range, key;
4993 key = (struct kvm_io_range) {
4998 range = bsearch(&key, bus->range, bus->dev_count,
4999 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
5003 off = range - bus->range;
5005 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
5011 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
5012 struct kvm_io_range *range, const void *val)
5016 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
5020 while (idx < bus->dev_count &&
5021 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
5022 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
5031 /* kvm_io_bus_write - called under kvm->slots_lock */
5032 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
5033 int len, const void *val)
5035 struct kvm_io_bus *bus;
5036 struct kvm_io_range range;
5039 range = (struct kvm_io_range) {
5044 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
5047 r = __kvm_io_bus_write(vcpu, bus, &range, val);
5048 return r < 0 ? r : 0;
5050 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
5052 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
5053 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
5054 gpa_t addr, int len, const void *val, long cookie)
5056 struct kvm_io_bus *bus;
5057 struct kvm_io_range range;
5059 range = (struct kvm_io_range) {
5064 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
5068 /* First try the device referenced by cookie. */
5069 if ((cookie >= 0) && (cookie < bus->dev_count) &&
5070 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
5071 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
5076 * cookie contained garbage; fall back to search and return the
5077 * correct cookie value.
5079 return __kvm_io_bus_write(vcpu, bus, &range, val);
5082 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
5083 struct kvm_io_range *range, void *val)
5087 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
5091 while (idx < bus->dev_count &&
5092 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
5093 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
5102 /* kvm_io_bus_read - called under kvm->slots_lock */
5103 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
5106 struct kvm_io_bus *bus;
5107 struct kvm_io_range range;
5110 range = (struct kvm_io_range) {
5115 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
5118 r = __kvm_io_bus_read(vcpu, bus, &range, val);
5119 return r < 0 ? r : 0;
5122 /* Caller must hold slots_lock. */
5123 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
5124 int len, struct kvm_io_device *dev)
5127 struct kvm_io_bus *new_bus, *bus;
5128 struct kvm_io_range range;
5130 bus = kvm_get_bus(kvm, bus_idx);
5134 /* exclude ioeventfd which is limited by maximum fd */
5135 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
5138 new_bus = kmalloc(struct_size(bus, range, bus->dev_count + 1),
5139 GFP_KERNEL_ACCOUNT);
5143 range = (struct kvm_io_range) {
5149 for (i = 0; i < bus->dev_count; i++)
5150 if (kvm_io_bus_cmp(&bus->range[i], &range) > 0)
5153 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
5154 new_bus->dev_count++;
5155 new_bus->range[i] = range;
5156 memcpy(new_bus->range + i + 1, bus->range + i,
5157 (bus->dev_count - i) * sizeof(struct kvm_io_range));
5158 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
5159 synchronize_srcu_expedited(&kvm->srcu);
5165 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
5166 struct kvm_io_device *dev)
5169 struct kvm_io_bus *new_bus, *bus;
5171 lockdep_assert_held(&kvm->slots_lock);
5173 bus = kvm_get_bus(kvm, bus_idx);
5177 for (i = 0; i < bus->dev_count; i++) {
5178 if (bus->range[i].dev == dev) {
5183 if (i == bus->dev_count)
5186 new_bus = kmalloc(struct_size(bus, range, bus->dev_count - 1),
5187 GFP_KERNEL_ACCOUNT);
5189 memcpy(new_bus, bus, struct_size(bus, range, i));
5190 new_bus->dev_count--;
5191 memcpy(new_bus->range + i, bus->range + i + 1,
5192 flex_array_size(new_bus, range, new_bus->dev_count - i));
5195 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
5196 synchronize_srcu_expedited(&kvm->srcu);
5198 /* Destroy the old bus _after_ installing the (null) bus. */
5200 pr_err("kvm: failed to shrink bus, removing it completely\n");
5201 for (j = 0; j < bus->dev_count; j++) {
5204 kvm_iodevice_destructor(bus->range[j].dev);
5209 return new_bus ? 0 : -ENOMEM;
5212 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
5215 struct kvm_io_bus *bus;
5216 int dev_idx, srcu_idx;
5217 struct kvm_io_device *iodev = NULL;
5219 srcu_idx = srcu_read_lock(&kvm->srcu);
5221 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
5225 dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
5229 iodev = bus->range[dev_idx].dev;
5232 srcu_read_unlock(&kvm->srcu, srcu_idx);
5236 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
5238 static int kvm_debugfs_open(struct inode *inode, struct file *file,
5239 int (*get)(void *, u64 *), int (*set)(void *, u64),
5242 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
5246 * The debugfs files are a reference to the kvm struct which
5247 * is still valid when kvm_destroy_vm is called. kvm_get_kvm_safe
5248 * avoids the race between open and the removal of the debugfs directory.
5250 if (!kvm_get_kvm_safe(stat_data->kvm))
5253 if (simple_attr_open(inode, file, get,
5254 kvm_stats_debugfs_mode(stat_data->desc) & 0222
5257 kvm_put_kvm(stat_data->kvm);
5264 static int kvm_debugfs_release(struct inode *inode, struct file *file)
5266 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
5269 simple_attr_release(inode, file);
5270 kvm_put_kvm(stat_data->kvm);
5275 static int kvm_get_stat_per_vm(struct kvm *kvm, size_t offset, u64 *val)
5277 *val = *(u64 *)((void *)(&kvm->stat) + offset);
5282 static int kvm_clear_stat_per_vm(struct kvm *kvm, size_t offset)
5284 *(u64 *)((void *)(&kvm->stat) + offset) = 0;
5289 static int kvm_get_stat_per_vcpu(struct kvm *kvm, size_t offset, u64 *val)
5292 struct kvm_vcpu *vcpu;
5296 kvm_for_each_vcpu(i, vcpu, kvm)
5297 *val += *(u64 *)((void *)(&vcpu->stat) + offset);
5302 static int kvm_clear_stat_per_vcpu(struct kvm *kvm, size_t offset)
5305 struct kvm_vcpu *vcpu;
5307 kvm_for_each_vcpu(i, vcpu, kvm)
5308 *(u64 *)((void *)(&vcpu->stat) + offset) = 0;
5313 static int kvm_stat_data_get(void *data, u64 *val)
5316 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
5318 switch (stat_data->kind) {
5320 r = kvm_get_stat_per_vm(stat_data->kvm,
5321 stat_data->desc->desc.offset, val);
5324 r = kvm_get_stat_per_vcpu(stat_data->kvm,
5325 stat_data->desc->desc.offset, val);
5332 static int kvm_stat_data_clear(void *data, u64 val)
5335 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
5340 switch (stat_data->kind) {
5342 r = kvm_clear_stat_per_vm(stat_data->kvm,
5343 stat_data->desc->desc.offset);
5346 r = kvm_clear_stat_per_vcpu(stat_data->kvm,
5347 stat_data->desc->desc.offset);
5354 static int kvm_stat_data_open(struct inode *inode, struct file *file)
5356 __simple_attr_check_format("%llu\n", 0ull);
5357 return kvm_debugfs_open(inode, file, kvm_stat_data_get,
5358 kvm_stat_data_clear, "%llu\n");
5361 static const struct file_operations stat_fops_per_vm = {
5362 .owner = THIS_MODULE,
5363 .open = kvm_stat_data_open,
5364 .release = kvm_debugfs_release,
5365 .read = simple_attr_read,
5366 .write = simple_attr_write,
5367 .llseek = no_llseek,
5370 static int vm_stat_get(void *_offset, u64 *val)
5372 unsigned offset = (long)_offset;
5377 mutex_lock(&kvm_lock);
5378 list_for_each_entry(kvm, &vm_list, vm_list) {
5379 kvm_get_stat_per_vm(kvm, offset, &tmp_val);
5382 mutex_unlock(&kvm_lock);
5386 static int vm_stat_clear(void *_offset, u64 val)
5388 unsigned offset = (long)_offset;
5394 mutex_lock(&kvm_lock);
5395 list_for_each_entry(kvm, &vm_list, vm_list) {
5396 kvm_clear_stat_per_vm(kvm, offset);
5398 mutex_unlock(&kvm_lock);
5403 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n");
5404 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_readonly_fops, vm_stat_get, NULL, "%llu\n");
5406 static int vcpu_stat_get(void *_offset, u64 *val)
5408 unsigned offset = (long)_offset;
5413 mutex_lock(&kvm_lock);
5414 list_for_each_entry(kvm, &vm_list, vm_list) {
5415 kvm_get_stat_per_vcpu(kvm, offset, &tmp_val);
5418 mutex_unlock(&kvm_lock);
5422 static int vcpu_stat_clear(void *_offset, u64 val)
5424 unsigned offset = (long)_offset;
5430 mutex_lock(&kvm_lock);
5431 list_for_each_entry(kvm, &vm_list, vm_list) {
5432 kvm_clear_stat_per_vcpu(kvm, offset);
5434 mutex_unlock(&kvm_lock);
5439 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear,
5441 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_readonly_fops, vcpu_stat_get, NULL, "%llu\n");
5443 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm)
5445 struct kobj_uevent_env *env;
5446 unsigned long long created, active;
5448 if (!kvm_dev.this_device || !kvm)
5451 mutex_lock(&kvm_lock);
5452 if (type == KVM_EVENT_CREATE_VM) {
5453 kvm_createvm_count++;
5455 } else if (type == KVM_EVENT_DESTROY_VM) {
5458 created = kvm_createvm_count;
5459 active = kvm_active_vms;
5460 mutex_unlock(&kvm_lock);
5462 env = kzalloc(sizeof(*env), GFP_KERNEL_ACCOUNT);
5466 add_uevent_var(env, "CREATED=%llu", created);
5467 add_uevent_var(env, "COUNT=%llu", active);
5469 if (type == KVM_EVENT_CREATE_VM) {
5470 add_uevent_var(env, "EVENT=create");
5471 kvm->userspace_pid = task_pid_nr(current);
5472 } else if (type == KVM_EVENT_DESTROY_VM) {
5473 add_uevent_var(env, "EVENT=destroy");
5475 add_uevent_var(env, "PID=%d", kvm->userspace_pid);
5477 if (kvm->debugfs_dentry) {
5478 char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL_ACCOUNT);
5481 tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX);
5483 add_uevent_var(env, "STATS_PATH=%s", tmp);
5487 /* no need for checks, since we are adding at most only 5 keys */
5488 env->envp[env->envp_idx++] = NULL;
5489 kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp);
5493 static void kvm_init_debug(void)
5495 const struct file_operations *fops;
5496 const struct _kvm_stats_desc *pdesc;
5499 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
5501 for (i = 0; i < kvm_vm_stats_header.num_desc; ++i) {
5502 pdesc = &kvm_vm_stats_desc[i];
5503 if (kvm_stats_debugfs_mode(pdesc) & 0222)
5504 fops = &vm_stat_fops;
5506 fops = &vm_stat_readonly_fops;
5507 debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
5509 (void *)(long)pdesc->desc.offset, fops);
5512 for (i = 0; i < kvm_vcpu_stats_header.num_desc; ++i) {
5513 pdesc = &kvm_vcpu_stats_desc[i];
5514 if (kvm_stats_debugfs_mode(pdesc) & 0222)
5515 fops = &vcpu_stat_fops;
5517 fops = &vcpu_stat_readonly_fops;
5518 debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
5520 (void *)(long)pdesc->desc.offset, fops);
5524 static int kvm_suspend(void)
5526 if (kvm_usage_count)
5527 hardware_disable_nolock(NULL);
5531 static void kvm_resume(void)
5533 if (kvm_usage_count) {
5534 #ifdef CONFIG_LOCKDEP
5535 WARN_ON(lockdep_is_held(&kvm_count_lock));
5537 hardware_enable_nolock(NULL);
5541 static struct syscore_ops kvm_syscore_ops = {
5542 .suspend = kvm_suspend,
5543 .resume = kvm_resume,
5547 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
5549 return container_of(pn, struct kvm_vcpu, preempt_notifier);
5552 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
5554 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
5556 WRITE_ONCE(vcpu->preempted, false);
5557 WRITE_ONCE(vcpu->ready, false);
5559 __this_cpu_write(kvm_running_vcpu, vcpu);
5560 kvm_arch_sched_in(vcpu, cpu);
5561 kvm_arch_vcpu_load(vcpu, cpu);
5564 static void kvm_sched_out(struct preempt_notifier *pn,
5565 struct task_struct *next)
5567 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
5569 if (current->on_rq) {
5570 WRITE_ONCE(vcpu->preempted, true);
5571 WRITE_ONCE(vcpu->ready, true);
5573 kvm_arch_vcpu_put(vcpu);
5574 __this_cpu_write(kvm_running_vcpu, NULL);
5578 * kvm_get_running_vcpu - get the vcpu running on the current CPU.
5580 * We can disable preemption locally around accessing the per-CPU variable,
5581 * and use the resolved vcpu pointer after enabling preemption again,
5582 * because even if the current thread is migrated to another CPU, reading
5583 * the per-CPU value later will give us the same value as we update the
5584 * per-CPU variable in the preempt notifier handlers.
5586 struct kvm_vcpu *kvm_get_running_vcpu(void)
5588 struct kvm_vcpu *vcpu;
5591 vcpu = __this_cpu_read(kvm_running_vcpu);
5596 EXPORT_SYMBOL_GPL(kvm_get_running_vcpu);
5599 * kvm_get_running_vcpus - get the per-CPU array of currently running vcpus.
5601 struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void)
5603 return &kvm_running_vcpu;
5606 #ifdef CONFIG_GUEST_PERF_EVENTS
5607 static unsigned int kvm_guest_state(void)
5609 struct kvm_vcpu *vcpu = kvm_get_running_vcpu();
5612 if (!kvm_arch_pmi_in_guest(vcpu))
5615 state = PERF_GUEST_ACTIVE;
5616 if (!kvm_arch_vcpu_in_kernel(vcpu))
5617 state |= PERF_GUEST_USER;
5622 static unsigned long kvm_guest_get_ip(void)
5624 struct kvm_vcpu *vcpu = kvm_get_running_vcpu();
5626 /* Retrieving the IP must be guarded by a call to kvm_guest_state(). */
5627 if (WARN_ON_ONCE(!kvm_arch_pmi_in_guest(vcpu)))
5630 return kvm_arch_vcpu_get_ip(vcpu);
5633 static struct perf_guest_info_callbacks kvm_guest_cbs = {
5634 .state = kvm_guest_state,
5635 .get_ip = kvm_guest_get_ip,
5636 .handle_intel_pt_intr = NULL,
5639 void kvm_register_perf_callbacks(unsigned int (*pt_intr_handler)(void))
5641 kvm_guest_cbs.handle_intel_pt_intr = pt_intr_handler;
5642 perf_register_guest_info_callbacks(&kvm_guest_cbs);
5644 void kvm_unregister_perf_callbacks(void)
5646 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
5650 struct kvm_cpu_compat_check {
5655 static void check_processor_compat(void *data)
5657 struct kvm_cpu_compat_check *c = data;
5659 *c->ret = kvm_arch_check_processor_compat(c->opaque);
5662 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
5663 struct module *module)
5665 struct kvm_cpu_compat_check c;
5669 r = kvm_arch_init(opaque);
5674 * kvm_arch_init makes sure there's at most one caller
5675 * for architectures that support multiple implementations,
5676 * like intel and amd on x86.
5677 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
5678 * conflicts in case kvm is already setup for another implementation.
5680 r = kvm_irqfd_init();
5684 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
5689 r = kvm_arch_hardware_setup(opaque);
5695 for_each_online_cpu(cpu) {
5696 smp_call_function_single(cpu, check_processor_compat, &c, 1);
5701 r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting",
5702 kvm_starting_cpu, kvm_dying_cpu);
5705 register_reboot_notifier(&kvm_reboot_notifier);
5707 /* A kmem cache lets us meet the alignment requirements of fx_save. */
5709 vcpu_align = __alignof__(struct kvm_vcpu);
5711 kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align,
5713 offsetof(struct kvm_vcpu, arch),
5714 offsetofend(struct kvm_vcpu, stats_id)
5715 - offsetof(struct kvm_vcpu, arch),
5717 if (!kvm_vcpu_cache) {
5722 for_each_possible_cpu(cpu) {
5723 if (!alloc_cpumask_var_node(&per_cpu(cpu_kick_mask, cpu),
5724 GFP_KERNEL, cpu_to_node(cpu))) {
5730 r = kvm_async_pf_init();
5734 kvm_chardev_ops.owner = module;
5735 kvm_vm_fops.owner = module;
5736 kvm_vcpu_fops.owner = module;
5738 r = misc_register(&kvm_dev);
5740 pr_err("kvm: misc device register failed\n");
5744 register_syscore_ops(&kvm_syscore_ops);
5746 kvm_preempt_ops.sched_in = kvm_sched_in;
5747 kvm_preempt_ops.sched_out = kvm_sched_out;
5751 r = kvm_vfio_ops_init();
5757 kvm_async_pf_deinit();
5759 for_each_possible_cpu(cpu)
5760 free_cpumask_var(per_cpu(cpu_kick_mask, cpu));
5762 kmem_cache_destroy(kvm_vcpu_cache);
5764 unregister_reboot_notifier(&kvm_reboot_notifier);
5765 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
5767 kvm_arch_hardware_unsetup();
5769 free_cpumask_var(cpus_hardware_enabled);
5777 EXPORT_SYMBOL_GPL(kvm_init);
5783 debugfs_remove_recursive(kvm_debugfs_dir);
5784 misc_deregister(&kvm_dev);
5785 for_each_possible_cpu(cpu)
5786 free_cpumask_var(per_cpu(cpu_kick_mask, cpu));
5787 kmem_cache_destroy(kvm_vcpu_cache);
5788 kvm_async_pf_deinit();
5789 unregister_syscore_ops(&kvm_syscore_ops);
5790 unregister_reboot_notifier(&kvm_reboot_notifier);
5791 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
5792 on_each_cpu(hardware_disable_nolock, NULL, 1);
5793 kvm_arch_hardware_unsetup();
5796 free_cpumask_var(cpus_hardware_enabled);
5797 kvm_vfio_ops_exit();
5799 EXPORT_SYMBOL_GPL(kvm_exit);
5801 struct kvm_vm_worker_thread_context {
5803 struct task_struct *parent;
5804 struct completion init_done;
5805 kvm_vm_thread_fn_t thread_fn;
5810 static int kvm_vm_worker_thread(void *context)
5813 * The init_context is allocated on the stack of the parent thread, so
5814 * we have to locally copy anything that is needed beyond initialization
5816 struct kvm_vm_worker_thread_context *init_context = context;
5817 struct kvm *kvm = init_context->kvm;
5818 kvm_vm_thread_fn_t thread_fn = init_context->thread_fn;
5819 uintptr_t data = init_context->data;
5822 err = kthread_park(current);
5823 /* kthread_park(current) is never supposed to return an error */
5828 err = cgroup_attach_task_all(init_context->parent, current);
5830 kvm_err("%s: cgroup_attach_task_all failed with err %d\n",
5835 set_user_nice(current, task_nice(init_context->parent));
5838 init_context->err = err;
5839 complete(&init_context->init_done);
5840 init_context = NULL;
5845 /* Wait to be woken up by the spawner before proceeding. */
5848 if (!kthread_should_stop())
5849 err = thread_fn(kvm, data);
5854 int kvm_vm_create_worker_thread(struct kvm *kvm, kvm_vm_thread_fn_t thread_fn,
5855 uintptr_t data, const char *name,
5856 struct task_struct **thread_ptr)
5858 struct kvm_vm_worker_thread_context init_context = {};
5859 struct task_struct *thread;
5862 init_context.kvm = kvm;
5863 init_context.parent = current;
5864 init_context.thread_fn = thread_fn;
5865 init_context.data = data;
5866 init_completion(&init_context.init_done);
5868 thread = kthread_run(kvm_vm_worker_thread, &init_context,
5869 "%s-%d", name, task_pid_nr(current));
5871 return PTR_ERR(thread);
5873 /* kthread_run is never supposed to return NULL */
5874 WARN_ON(thread == NULL);
5876 wait_for_completion(&init_context.init_done);
5878 if (!init_context.err)
5879 *thread_ptr = thread;
5881 return init_context.err;