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;
313 cpus = this_cpu_cpumask_var_ptr(cpu_kick_mask);
316 kvm_for_each_vcpu(i, vcpu, kvm) {
319 kvm_make_vcpu_request(kvm, vcpu, req, cpus, me);
322 called = kvm_kick_many_cpus(cpus, !!(req & KVM_REQUEST_WAIT));
328 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
330 return kvm_make_all_cpus_request_except(kvm, req, NULL);
332 EXPORT_SYMBOL_GPL(kvm_make_all_cpus_request);
334 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
335 void kvm_flush_remote_tlbs(struct kvm *kvm)
337 ++kvm->stat.generic.remote_tlb_flush_requests;
340 * We want to publish modifications to the page tables before reading
341 * mode. Pairs with a memory barrier in arch-specific code.
342 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
343 * and smp_mb in walk_shadow_page_lockless_begin/end.
344 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
346 * There is already an smp_mb__after_atomic() before
347 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
350 if (!kvm_arch_flush_remote_tlb(kvm)
351 || kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
352 ++kvm->stat.generic.remote_tlb_flush;
354 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
357 void kvm_reload_remote_mmus(struct kvm *kvm)
359 kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
362 #ifdef KVM_ARCH_NR_OBJS_PER_MEMORY_CACHE
363 static inline void *mmu_memory_cache_alloc_obj(struct kvm_mmu_memory_cache *mc,
366 gfp_flags |= mc->gfp_zero;
369 return kmem_cache_alloc(mc->kmem_cache, gfp_flags);
371 return (void *)__get_free_page(gfp_flags);
374 int kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache *mc, int min)
378 if (mc->nobjs >= min)
380 while (mc->nobjs < ARRAY_SIZE(mc->objects)) {
381 obj = mmu_memory_cache_alloc_obj(mc, GFP_KERNEL_ACCOUNT);
383 return mc->nobjs >= min ? 0 : -ENOMEM;
384 mc->objects[mc->nobjs++] = obj;
389 int kvm_mmu_memory_cache_nr_free_objects(struct kvm_mmu_memory_cache *mc)
394 void kvm_mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc)
398 kmem_cache_free(mc->kmem_cache, mc->objects[--mc->nobjs]);
400 free_page((unsigned long)mc->objects[--mc->nobjs]);
404 void *kvm_mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc)
408 if (WARN_ON(!mc->nobjs))
409 p = mmu_memory_cache_alloc_obj(mc, GFP_ATOMIC | __GFP_ACCOUNT);
411 p = mc->objects[--mc->nobjs];
417 static void kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
419 mutex_init(&vcpu->mutex);
424 rcuwait_init(&vcpu->wait);
425 kvm_async_pf_vcpu_init(vcpu);
428 INIT_LIST_HEAD(&vcpu->blocked_vcpu_list);
430 kvm_vcpu_set_in_spin_loop(vcpu, false);
431 kvm_vcpu_set_dy_eligible(vcpu, false);
432 vcpu->preempted = false;
434 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
435 vcpu->last_used_slot = 0;
438 void kvm_vcpu_destroy(struct kvm_vcpu *vcpu)
440 kvm_dirty_ring_free(&vcpu->dirty_ring);
441 kvm_arch_vcpu_destroy(vcpu);
444 * No need for rcu_read_lock as VCPU_RUN is the only place that changes
445 * the vcpu->pid pointer, and at destruction time all file descriptors
448 put_pid(rcu_dereference_protected(vcpu->pid, 1));
450 free_page((unsigned long)vcpu->run);
451 kmem_cache_free(kvm_vcpu_cache, vcpu);
453 EXPORT_SYMBOL_GPL(kvm_vcpu_destroy);
455 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
456 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
458 return container_of(mn, struct kvm, mmu_notifier);
461 static void kvm_mmu_notifier_invalidate_range(struct mmu_notifier *mn,
462 struct mm_struct *mm,
463 unsigned long start, unsigned long end)
465 struct kvm *kvm = mmu_notifier_to_kvm(mn);
468 idx = srcu_read_lock(&kvm->srcu);
469 kvm_arch_mmu_notifier_invalidate_range(kvm, start, end);
470 srcu_read_unlock(&kvm->srcu, idx);
473 typedef bool (*hva_handler_t)(struct kvm *kvm, struct kvm_gfn_range *range);
475 typedef void (*on_lock_fn_t)(struct kvm *kvm, unsigned long start,
478 struct kvm_hva_range {
482 hva_handler_t handler;
483 on_lock_fn_t on_lock;
489 * Use a dedicated stub instead of NULL to indicate that there is no callback
490 * function/handler. The compiler technically can't guarantee that a real
491 * function will have a non-zero address, and so it will generate code to
492 * check for !NULL, whereas comparing against a stub will be elided at compile
493 * time (unless the compiler is getting long in the tooth, e.g. gcc 4.9).
495 static void kvm_null_fn(void)
499 #define IS_KVM_NULL_FN(fn) ((fn) == (void *)kvm_null_fn)
501 static __always_inline int __kvm_handle_hva_range(struct kvm *kvm,
502 const struct kvm_hva_range *range)
504 bool ret = false, locked = false;
505 struct kvm_gfn_range gfn_range;
506 struct kvm_memory_slot *slot;
507 struct kvm_memslots *slots;
510 /* A null handler is allowed if and only if on_lock() is provided. */
511 if (WARN_ON_ONCE(IS_KVM_NULL_FN(range->on_lock) &&
512 IS_KVM_NULL_FN(range->handler)))
515 idx = srcu_read_lock(&kvm->srcu);
517 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
518 slots = __kvm_memslots(kvm, i);
519 kvm_for_each_memslot(slot, slots) {
520 unsigned long hva_start, hva_end;
522 hva_start = max(range->start, slot->userspace_addr);
523 hva_end = min(range->end, slot->userspace_addr +
524 (slot->npages << PAGE_SHIFT));
525 if (hva_start >= hva_end)
529 * To optimize for the likely case where the address
530 * range is covered by zero or one memslots, don't
531 * bother making these conditional (to avoid writes on
532 * the second or later invocation of the handler).
534 gfn_range.pte = range->pte;
535 gfn_range.may_block = range->may_block;
538 * {gfn(page) | page intersects with [hva_start, hva_end)} =
539 * {gfn_start, gfn_start+1, ..., gfn_end-1}.
541 gfn_range.start = hva_to_gfn_memslot(hva_start, slot);
542 gfn_range.end = hva_to_gfn_memslot(hva_end + PAGE_SIZE - 1, slot);
543 gfn_range.slot = slot;
548 if (!IS_KVM_NULL_FN(range->on_lock))
549 range->on_lock(kvm, range->start, range->end);
550 if (IS_KVM_NULL_FN(range->handler))
553 ret |= range->handler(kvm, &gfn_range);
557 if (range->flush_on_ret && ret)
558 kvm_flush_remote_tlbs(kvm);
563 srcu_read_unlock(&kvm->srcu, idx);
565 /* The notifiers are averse to booleans. :-( */
569 static __always_inline int kvm_handle_hva_range(struct mmu_notifier *mn,
573 hva_handler_t handler)
575 struct kvm *kvm = mmu_notifier_to_kvm(mn);
576 const struct kvm_hva_range range = {
581 .on_lock = (void *)kvm_null_fn,
582 .flush_on_ret = true,
586 return __kvm_handle_hva_range(kvm, &range);
589 static __always_inline int kvm_handle_hva_range_no_flush(struct mmu_notifier *mn,
592 hva_handler_t handler)
594 struct kvm *kvm = mmu_notifier_to_kvm(mn);
595 const struct kvm_hva_range range = {
600 .on_lock = (void *)kvm_null_fn,
601 .flush_on_ret = false,
605 return __kvm_handle_hva_range(kvm, &range);
607 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
608 struct mm_struct *mm,
609 unsigned long address,
612 struct kvm *kvm = mmu_notifier_to_kvm(mn);
614 trace_kvm_set_spte_hva(address);
617 * .change_pte() must be surrounded by .invalidate_range_{start,end}().
618 * If mmu_notifier_count is zero, then no in-progress invalidations,
619 * including this one, found a relevant memslot at start(); rechecking
620 * memslots here is unnecessary. Note, a false positive (count elevated
621 * by a different invalidation) is sub-optimal but functionally ok.
623 WARN_ON_ONCE(!READ_ONCE(kvm->mn_active_invalidate_count));
624 if (!READ_ONCE(kvm->mmu_notifier_count))
627 kvm_handle_hva_range(mn, address, address + 1, pte, kvm_set_spte_gfn);
630 void kvm_inc_notifier_count(struct kvm *kvm, unsigned long start,
634 * The count increase must become visible at unlock time as no
635 * spte can be established without taking the mmu_lock and
636 * count is also read inside the mmu_lock critical section.
638 kvm->mmu_notifier_count++;
639 if (likely(kvm->mmu_notifier_count == 1)) {
640 kvm->mmu_notifier_range_start = start;
641 kvm->mmu_notifier_range_end = end;
644 * Fully tracking multiple concurrent ranges has dimishing
645 * returns. Keep things simple and just find the minimal range
646 * which includes the current and new ranges. As there won't be
647 * enough information to subtract a range after its invalidate
648 * completes, any ranges invalidated concurrently will
649 * accumulate and persist until all outstanding invalidates
652 kvm->mmu_notifier_range_start =
653 min(kvm->mmu_notifier_range_start, start);
654 kvm->mmu_notifier_range_end =
655 max(kvm->mmu_notifier_range_end, end);
659 static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
660 const struct mmu_notifier_range *range)
662 struct kvm *kvm = mmu_notifier_to_kvm(mn);
663 const struct kvm_hva_range hva_range = {
664 .start = range->start,
667 .handler = kvm_unmap_gfn_range,
668 .on_lock = kvm_inc_notifier_count,
669 .flush_on_ret = true,
670 .may_block = mmu_notifier_range_blockable(range),
673 trace_kvm_unmap_hva_range(range->start, range->end);
676 * Prevent memslot modification between range_start() and range_end()
677 * so that conditionally locking provides the same result in both
678 * functions. Without that guarantee, the mmu_notifier_count
679 * adjustments will be imbalanced.
681 * Pairs with the decrement in range_end().
683 spin_lock(&kvm->mn_invalidate_lock);
684 kvm->mn_active_invalidate_count++;
685 spin_unlock(&kvm->mn_invalidate_lock);
687 __kvm_handle_hva_range(kvm, &hva_range);
692 void kvm_dec_notifier_count(struct kvm *kvm, unsigned long start,
696 * This sequence increase will notify the kvm page fault that
697 * the page that is going to be mapped in the spte could have
700 kvm->mmu_notifier_seq++;
703 * The above sequence increase must be visible before the
704 * below count decrease, which is ensured by the smp_wmb above
705 * in conjunction with the smp_rmb in mmu_notifier_retry().
707 kvm->mmu_notifier_count--;
710 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
711 const struct mmu_notifier_range *range)
713 struct kvm *kvm = mmu_notifier_to_kvm(mn);
714 const struct kvm_hva_range hva_range = {
715 .start = range->start,
718 .handler = (void *)kvm_null_fn,
719 .on_lock = kvm_dec_notifier_count,
720 .flush_on_ret = false,
721 .may_block = mmu_notifier_range_blockable(range),
725 __kvm_handle_hva_range(kvm, &hva_range);
727 /* Pairs with the increment in range_start(). */
728 spin_lock(&kvm->mn_invalidate_lock);
729 wake = (--kvm->mn_active_invalidate_count == 0);
730 spin_unlock(&kvm->mn_invalidate_lock);
733 * There can only be one waiter, since the wait happens under
737 rcuwait_wake_up(&kvm->mn_memslots_update_rcuwait);
739 BUG_ON(kvm->mmu_notifier_count < 0);
742 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
743 struct mm_struct *mm,
747 trace_kvm_age_hva(start, end);
749 return kvm_handle_hva_range(mn, start, end, __pte(0), kvm_age_gfn);
752 static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
753 struct mm_struct *mm,
757 trace_kvm_age_hva(start, end);
760 * Even though we do not flush TLB, this will still adversely
761 * affect performance on pre-Haswell Intel EPT, where there is
762 * no EPT Access Bit to clear so that we have to tear down EPT
763 * tables instead. If we find this unacceptable, we can always
764 * add a parameter to kvm_age_hva so that it effectively doesn't
765 * do anything on clear_young.
767 * Also note that currently we never issue secondary TLB flushes
768 * from clear_young, leaving this job up to the regular system
769 * cadence. If we find this inaccurate, we might come up with a
770 * more sophisticated heuristic later.
772 return kvm_handle_hva_range_no_flush(mn, start, end, kvm_age_gfn);
775 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
776 struct mm_struct *mm,
777 unsigned long address)
779 trace_kvm_test_age_hva(address);
781 return kvm_handle_hva_range_no_flush(mn, address, address + 1,
785 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
786 struct mm_struct *mm)
788 struct kvm *kvm = mmu_notifier_to_kvm(mn);
791 idx = srcu_read_lock(&kvm->srcu);
792 kvm_arch_flush_shadow_all(kvm);
793 srcu_read_unlock(&kvm->srcu, idx);
796 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
797 .invalidate_range = kvm_mmu_notifier_invalidate_range,
798 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
799 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
800 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
801 .clear_young = kvm_mmu_notifier_clear_young,
802 .test_young = kvm_mmu_notifier_test_young,
803 .change_pte = kvm_mmu_notifier_change_pte,
804 .release = kvm_mmu_notifier_release,
807 static int kvm_init_mmu_notifier(struct kvm *kvm)
809 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
810 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
813 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
815 static int kvm_init_mmu_notifier(struct kvm *kvm)
820 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
822 #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER
823 static int kvm_pm_notifier_call(struct notifier_block *bl,
827 struct kvm *kvm = container_of(bl, struct kvm, pm_notifier);
829 return kvm_arch_pm_notifier(kvm, state);
832 static void kvm_init_pm_notifier(struct kvm *kvm)
834 kvm->pm_notifier.notifier_call = kvm_pm_notifier_call;
835 /* Suspend KVM before we suspend ftrace, RCU, etc. */
836 kvm->pm_notifier.priority = INT_MAX;
837 register_pm_notifier(&kvm->pm_notifier);
840 static void kvm_destroy_pm_notifier(struct kvm *kvm)
842 unregister_pm_notifier(&kvm->pm_notifier);
844 #else /* !CONFIG_HAVE_KVM_PM_NOTIFIER */
845 static void kvm_init_pm_notifier(struct kvm *kvm)
849 static void kvm_destroy_pm_notifier(struct kvm *kvm)
852 #endif /* CONFIG_HAVE_KVM_PM_NOTIFIER */
854 static struct kvm_memslots *kvm_alloc_memslots(void)
857 struct kvm_memslots *slots;
859 slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL_ACCOUNT);
863 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
864 slots->id_to_index[i] = -1;
869 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
871 if (!memslot->dirty_bitmap)
874 kvfree(memslot->dirty_bitmap);
875 memslot->dirty_bitmap = NULL;
878 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
880 kvm_destroy_dirty_bitmap(slot);
882 kvm_arch_free_memslot(kvm, slot);
888 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
890 struct kvm_memory_slot *memslot;
895 kvm_for_each_memslot(memslot, slots)
896 kvm_free_memslot(kvm, memslot);
901 static umode_t kvm_stats_debugfs_mode(const struct _kvm_stats_desc *pdesc)
903 switch (pdesc->desc.flags & KVM_STATS_TYPE_MASK) {
904 case KVM_STATS_TYPE_INSTANT:
906 case KVM_STATS_TYPE_CUMULATIVE:
907 case KVM_STATS_TYPE_PEAK:
914 static void kvm_destroy_vm_debugfs(struct kvm *kvm)
917 int kvm_debugfs_num_entries = kvm_vm_stats_header.num_desc +
918 kvm_vcpu_stats_header.num_desc;
920 if (!kvm->debugfs_dentry)
923 debugfs_remove_recursive(kvm->debugfs_dentry);
925 if (kvm->debugfs_stat_data) {
926 for (i = 0; i < kvm_debugfs_num_entries; i++)
927 kfree(kvm->debugfs_stat_data[i]);
928 kfree(kvm->debugfs_stat_data);
932 static int kvm_create_vm_debugfs(struct kvm *kvm, int fd)
934 static DEFINE_MUTEX(kvm_debugfs_lock);
936 char dir_name[ITOA_MAX_LEN * 2];
937 struct kvm_stat_data *stat_data;
938 const struct _kvm_stats_desc *pdesc;
940 int kvm_debugfs_num_entries = kvm_vm_stats_header.num_desc +
941 kvm_vcpu_stats_header.num_desc;
943 if (!debugfs_initialized())
946 snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd);
947 mutex_lock(&kvm_debugfs_lock);
948 dent = debugfs_lookup(dir_name, kvm_debugfs_dir);
950 pr_warn_ratelimited("KVM: debugfs: duplicate directory %s\n", dir_name);
952 mutex_unlock(&kvm_debugfs_lock);
955 dent = debugfs_create_dir(dir_name, kvm_debugfs_dir);
956 mutex_unlock(&kvm_debugfs_lock);
960 kvm->debugfs_dentry = dent;
961 kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries,
962 sizeof(*kvm->debugfs_stat_data),
964 if (!kvm->debugfs_stat_data)
967 for (i = 0; i < kvm_vm_stats_header.num_desc; ++i) {
968 pdesc = &kvm_vm_stats_desc[i];
969 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT);
973 stat_data->kvm = kvm;
974 stat_data->desc = pdesc;
975 stat_data->kind = KVM_STAT_VM;
976 kvm->debugfs_stat_data[i] = stat_data;
977 debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
978 kvm->debugfs_dentry, stat_data,
982 for (i = 0; i < kvm_vcpu_stats_header.num_desc; ++i) {
983 pdesc = &kvm_vcpu_stats_desc[i];
984 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT);
988 stat_data->kvm = kvm;
989 stat_data->desc = pdesc;
990 stat_data->kind = KVM_STAT_VCPU;
991 kvm->debugfs_stat_data[i + kvm_vm_stats_header.num_desc] = stat_data;
992 debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
993 kvm->debugfs_dentry, stat_data,
997 ret = kvm_arch_create_vm_debugfs(kvm);
999 kvm_destroy_vm_debugfs(kvm);
1007 * Called after the VM is otherwise initialized, but just before adding it to
1010 int __weak kvm_arch_post_init_vm(struct kvm *kvm)
1016 * Called just after removing the VM from the vm_list, but before doing any
1017 * other destruction.
1019 void __weak kvm_arch_pre_destroy_vm(struct kvm *kvm)
1024 * Called after per-vm debugfs created. When called kvm->debugfs_dentry should
1025 * be setup already, so we can create arch-specific debugfs entries under it.
1026 * Cleanup should be automatic done in kvm_destroy_vm_debugfs() recursively, so
1027 * a per-arch destroy interface is not needed.
1029 int __weak kvm_arch_create_vm_debugfs(struct kvm *kvm)
1034 static struct kvm *kvm_create_vm(unsigned long type)
1036 struct kvm *kvm = kvm_arch_alloc_vm();
1041 return ERR_PTR(-ENOMEM);
1043 KVM_MMU_LOCK_INIT(kvm);
1044 mmgrab(current->mm);
1045 kvm->mm = current->mm;
1046 kvm_eventfd_init(kvm);
1047 mutex_init(&kvm->lock);
1048 mutex_init(&kvm->irq_lock);
1049 mutex_init(&kvm->slots_lock);
1050 mutex_init(&kvm->slots_arch_lock);
1051 spin_lock_init(&kvm->mn_invalidate_lock);
1052 rcuwait_init(&kvm->mn_memslots_update_rcuwait);
1054 INIT_LIST_HEAD(&kvm->devices);
1056 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
1058 if (init_srcu_struct(&kvm->srcu))
1059 goto out_err_no_srcu;
1060 if (init_srcu_struct(&kvm->irq_srcu))
1061 goto out_err_no_irq_srcu;
1063 refcount_set(&kvm->users_count, 1);
1064 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
1065 struct kvm_memslots *slots = kvm_alloc_memslots();
1068 goto out_err_no_arch_destroy_vm;
1069 /* Generations must be different for each address space. */
1070 slots->generation = i;
1071 rcu_assign_pointer(kvm->memslots[i], slots);
1074 for (i = 0; i < KVM_NR_BUSES; i++) {
1075 rcu_assign_pointer(kvm->buses[i],
1076 kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL_ACCOUNT));
1078 goto out_err_no_arch_destroy_vm;
1081 kvm->max_halt_poll_ns = halt_poll_ns;
1083 r = kvm_arch_init_vm(kvm, type);
1085 goto out_err_no_arch_destroy_vm;
1087 r = hardware_enable_all();
1089 goto out_err_no_disable;
1091 #ifdef CONFIG_HAVE_KVM_IRQFD
1092 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
1095 r = kvm_init_mmu_notifier(kvm);
1097 goto out_err_no_mmu_notifier;
1099 r = kvm_arch_post_init_vm(kvm);
1103 mutex_lock(&kvm_lock);
1104 list_add(&kvm->vm_list, &vm_list);
1105 mutex_unlock(&kvm_lock);
1107 preempt_notifier_inc();
1108 kvm_init_pm_notifier(kvm);
1113 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1114 if (kvm->mmu_notifier.ops)
1115 mmu_notifier_unregister(&kvm->mmu_notifier, current->mm);
1117 out_err_no_mmu_notifier:
1118 hardware_disable_all();
1120 kvm_arch_destroy_vm(kvm);
1121 out_err_no_arch_destroy_vm:
1122 WARN_ON_ONCE(!refcount_dec_and_test(&kvm->users_count));
1123 for (i = 0; i < KVM_NR_BUSES; i++)
1124 kfree(kvm_get_bus(kvm, i));
1125 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
1126 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
1127 cleanup_srcu_struct(&kvm->irq_srcu);
1128 out_err_no_irq_srcu:
1129 cleanup_srcu_struct(&kvm->srcu);
1131 kvm_arch_free_vm(kvm);
1132 mmdrop(current->mm);
1136 static void kvm_destroy_devices(struct kvm *kvm)
1138 struct kvm_device *dev, *tmp;
1141 * We do not need to take the kvm->lock here, because nobody else
1142 * has a reference to the struct kvm at this point and therefore
1143 * cannot access the devices list anyhow.
1145 list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
1146 list_del(&dev->vm_node);
1147 dev->ops->destroy(dev);
1151 static void kvm_destroy_vm(struct kvm *kvm)
1154 struct mm_struct *mm = kvm->mm;
1156 kvm_destroy_pm_notifier(kvm);
1157 kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm);
1158 kvm_destroy_vm_debugfs(kvm);
1159 kvm_arch_sync_events(kvm);
1160 mutex_lock(&kvm_lock);
1161 list_del(&kvm->vm_list);
1162 mutex_unlock(&kvm_lock);
1163 kvm_arch_pre_destroy_vm(kvm);
1165 kvm_free_irq_routing(kvm);
1166 for (i = 0; i < KVM_NR_BUSES; i++) {
1167 struct kvm_io_bus *bus = kvm_get_bus(kvm, i);
1170 kvm_io_bus_destroy(bus);
1171 kvm->buses[i] = NULL;
1173 kvm_coalesced_mmio_free(kvm);
1174 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1175 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
1177 * At this point, pending calls to invalidate_range_start()
1178 * have completed but no more MMU notifiers will run, so
1179 * mn_active_invalidate_count may remain unbalanced.
1180 * No threads can be waiting in install_new_memslots as the
1181 * last reference on KVM has been dropped, but freeing
1182 * memslots would deadlock without this manual intervention.
1184 WARN_ON(rcuwait_active(&kvm->mn_memslots_update_rcuwait));
1185 kvm->mn_active_invalidate_count = 0;
1187 kvm_arch_flush_shadow_all(kvm);
1189 kvm_arch_destroy_vm(kvm);
1190 kvm_destroy_devices(kvm);
1191 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
1192 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
1193 cleanup_srcu_struct(&kvm->irq_srcu);
1194 cleanup_srcu_struct(&kvm->srcu);
1195 kvm_arch_free_vm(kvm);
1196 preempt_notifier_dec();
1197 hardware_disable_all();
1201 void kvm_get_kvm(struct kvm *kvm)
1203 refcount_inc(&kvm->users_count);
1205 EXPORT_SYMBOL_GPL(kvm_get_kvm);
1208 * Make sure the vm is not during destruction, which is a safe version of
1209 * kvm_get_kvm(). Return true if kvm referenced successfully, false otherwise.
1211 bool kvm_get_kvm_safe(struct kvm *kvm)
1213 return refcount_inc_not_zero(&kvm->users_count);
1215 EXPORT_SYMBOL_GPL(kvm_get_kvm_safe);
1217 void kvm_put_kvm(struct kvm *kvm)
1219 if (refcount_dec_and_test(&kvm->users_count))
1220 kvm_destroy_vm(kvm);
1222 EXPORT_SYMBOL_GPL(kvm_put_kvm);
1225 * Used to put a reference that was taken on behalf of an object associated
1226 * with a user-visible file descriptor, e.g. a vcpu or device, if installation
1227 * of the new file descriptor fails and the reference cannot be transferred to
1228 * its final owner. In such cases, the caller is still actively using @kvm and
1229 * will fail miserably if the refcount unexpectedly hits zero.
1231 void kvm_put_kvm_no_destroy(struct kvm *kvm)
1233 WARN_ON(refcount_dec_and_test(&kvm->users_count));
1235 EXPORT_SYMBOL_GPL(kvm_put_kvm_no_destroy);
1237 static int kvm_vm_release(struct inode *inode, struct file *filp)
1239 struct kvm *kvm = filp->private_data;
1241 kvm_irqfd_release(kvm);
1248 * Allocation size is twice as large as the actual dirty bitmap size.
1249 * See kvm_vm_ioctl_get_dirty_log() why this is needed.
1251 static int kvm_alloc_dirty_bitmap(struct kvm_memory_slot *memslot)
1253 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
1255 memslot->dirty_bitmap = kvzalloc(dirty_bytes, GFP_KERNEL_ACCOUNT);
1256 if (!memslot->dirty_bitmap)
1263 * Delete a memslot by decrementing the number of used slots and shifting all
1264 * other entries in the array forward one spot.
1266 static inline void kvm_memslot_delete(struct kvm_memslots *slots,
1267 struct kvm_memory_slot *memslot)
1269 struct kvm_memory_slot *mslots = slots->memslots;
1272 if (WARN_ON(slots->id_to_index[memslot->id] == -1))
1275 slots->used_slots--;
1277 if (atomic_read(&slots->last_used_slot) >= slots->used_slots)
1278 atomic_set(&slots->last_used_slot, 0);
1280 for (i = slots->id_to_index[memslot->id]; i < slots->used_slots; i++) {
1281 mslots[i] = mslots[i + 1];
1282 slots->id_to_index[mslots[i].id] = i;
1284 mslots[i] = *memslot;
1285 slots->id_to_index[memslot->id] = -1;
1289 * "Insert" a new memslot by incrementing the number of used slots. Returns
1290 * the new slot's initial index into the memslots array.
1292 static inline int kvm_memslot_insert_back(struct kvm_memslots *slots)
1294 return slots->used_slots++;
1298 * Move a changed memslot backwards in the array by shifting existing slots
1299 * with a higher GFN toward the front of the array. Note, the changed memslot
1300 * itself is not preserved in the array, i.e. not swapped at this time, only
1301 * its new index into the array is tracked. Returns the changed memslot's
1302 * current index into the memslots array.
1304 static inline int kvm_memslot_move_backward(struct kvm_memslots *slots,
1305 struct kvm_memory_slot *memslot)
1307 struct kvm_memory_slot *mslots = slots->memslots;
1310 if (WARN_ON_ONCE(slots->id_to_index[memslot->id] == -1) ||
1311 WARN_ON_ONCE(!slots->used_slots))
1315 * Move the target memslot backward in the array by shifting existing
1316 * memslots with a higher GFN (than the target memslot) towards the
1317 * front of the array.
1319 for (i = slots->id_to_index[memslot->id]; i < slots->used_slots - 1; i++) {
1320 if (memslot->base_gfn > mslots[i + 1].base_gfn)
1323 WARN_ON_ONCE(memslot->base_gfn == mslots[i + 1].base_gfn);
1325 /* Shift the next memslot forward one and update its index. */
1326 mslots[i] = mslots[i + 1];
1327 slots->id_to_index[mslots[i].id] = i;
1333 * Move a changed memslot forwards in the array by shifting existing slots with
1334 * a lower GFN toward the back of the array. Note, the changed memslot itself
1335 * is not preserved in the array, i.e. not swapped at this time, only its new
1336 * index into the array is tracked. Returns the changed memslot's final index
1337 * into the memslots array.
1339 static inline int kvm_memslot_move_forward(struct kvm_memslots *slots,
1340 struct kvm_memory_slot *memslot,
1343 struct kvm_memory_slot *mslots = slots->memslots;
1346 for (i = start; i > 0; i--) {
1347 if (memslot->base_gfn < mslots[i - 1].base_gfn)
1350 WARN_ON_ONCE(memslot->base_gfn == mslots[i - 1].base_gfn);
1352 /* Shift the next memslot back one and update its index. */
1353 mslots[i] = mslots[i - 1];
1354 slots->id_to_index[mslots[i].id] = i;
1360 * Re-sort memslots based on their GFN to account for an added, deleted, or
1361 * moved memslot. Sorting memslots by GFN allows using a binary search during
1364 * IMPORTANT: Slots are sorted from highest GFN to lowest GFN! I.e. the entry
1365 * at memslots[0] has the highest GFN.
1367 * The sorting algorithm takes advantage of having initially sorted memslots
1368 * and knowing the position of the changed memslot. Sorting is also optimized
1369 * by not swapping the updated memslot and instead only shifting other memslots
1370 * and tracking the new index for the update memslot. Only once its final
1371 * index is known is the updated memslot copied into its position in the array.
1373 * - When deleting a memslot, the deleted memslot simply needs to be moved to
1374 * the end of the array.
1376 * - When creating a memslot, the algorithm "inserts" the new memslot at the
1377 * end of the array and then it forward to its correct location.
1379 * - When moving a memslot, the algorithm first moves the updated memslot
1380 * backward to handle the scenario where the memslot's GFN was changed to a
1381 * lower value. update_memslots() then falls through and runs the same flow
1382 * as creating a memslot to move the memslot forward to handle the scenario
1383 * where its GFN was changed to a higher value.
1385 * Note, slots are sorted from highest->lowest instead of lowest->highest for
1386 * historical reasons. Originally, invalid memslots where denoted by having
1387 * GFN=0, thus sorting from highest->lowest naturally sorted invalid memslots
1388 * to the end of the array. The current algorithm uses dedicated logic to
1389 * delete a memslot and thus does not rely on invalid memslots having GFN=0.
1391 * The other historical motiviation for highest->lowest was to improve the
1392 * performance of memslot lookup. KVM originally used a linear search starting
1393 * at memslots[0]. On x86, the largest memslot usually has one of the highest,
1394 * if not *the* highest, GFN, as the bulk of the guest's RAM is located in a
1395 * single memslot above the 4gb boundary. As the largest memslot is also the
1396 * most likely to be referenced, sorting it to the front of the array was
1397 * advantageous. The current binary search starts from the middle of the array
1398 * and uses an LRU pointer to improve performance for all memslots and GFNs.
1400 static void update_memslots(struct kvm_memslots *slots,
1401 struct kvm_memory_slot *memslot,
1402 enum kvm_mr_change change)
1406 if (change == KVM_MR_DELETE) {
1407 kvm_memslot_delete(slots, memslot);
1409 if (change == KVM_MR_CREATE)
1410 i = kvm_memslot_insert_back(slots);
1412 i = kvm_memslot_move_backward(slots, memslot);
1413 i = kvm_memslot_move_forward(slots, memslot, i);
1416 * Copy the memslot to its new position in memslots and update
1417 * its index accordingly.
1419 slots->memslots[i] = *memslot;
1420 slots->id_to_index[memslot->id] = i;
1424 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
1426 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
1428 #ifdef __KVM_HAVE_READONLY_MEM
1429 valid_flags |= KVM_MEM_READONLY;
1432 if (mem->flags & ~valid_flags)
1438 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
1439 int as_id, struct kvm_memslots *slots)
1441 struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id);
1442 u64 gen = old_memslots->generation;
1444 WARN_ON(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS);
1445 slots->generation = gen | KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
1448 * Do not store the new memslots while there are invalidations in
1449 * progress, otherwise the locking in invalidate_range_start and
1450 * invalidate_range_end will be unbalanced.
1452 spin_lock(&kvm->mn_invalidate_lock);
1453 prepare_to_rcuwait(&kvm->mn_memslots_update_rcuwait);
1454 while (kvm->mn_active_invalidate_count) {
1455 set_current_state(TASK_UNINTERRUPTIBLE);
1456 spin_unlock(&kvm->mn_invalidate_lock);
1458 spin_lock(&kvm->mn_invalidate_lock);
1460 finish_rcuwait(&kvm->mn_memslots_update_rcuwait);
1461 rcu_assign_pointer(kvm->memslots[as_id], slots);
1462 spin_unlock(&kvm->mn_invalidate_lock);
1465 * Acquired in kvm_set_memslot. Must be released before synchronize
1466 * SRCU below in order to avoid deadlock with another thread
1467 * acquiring the slots_arch_lock in an srcu critical section.
1469 mutex_unlock(&kvm->slots_arch_lock);
1471 synchronize_srcu_expedited(&kvm->srcu);
1474 * Increment the new memslot generation a second time, dropping the
1475 * update in-progress flag and incrementing the generation based on
1476 * the number of address spaces. This provides a unique and easily
1477 * identifiable generation number while the memslots are in flux.
1479 gen = slots->generation & ~KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
1482 * Generations must be unique even across address spaces. We do not need
1483 * a global counter for that, instead the generation space is evenly split
1484 * across address spaces. For example, with two address spaces, address
1485 * space 0 will use generations 0, 2, 4, ... while address space 1 will
1486 * use generations 1, 3, 5, ...
1488 gen += KVM_ADDRESS_SPACE_NUM;
1490 kvm_arch_memslots_updated(kvm, gen);
1492 slots->generation = gen;
1494 return old_memslots;
1497 static size_t kvm_memslots_size(int slots)
1499 return sizeof(struct kvm_memslots) +
1500 (sizeof(struct kvm_memory_slot) * slots);
1503 static void kvm_copy_memslots(struct kvm_memslots *to,
1504 struct kvm_memslots *from)
1506 memcpy(to, from, kvm_memslots_size(from->used_slots));
1510 * Note, at a minimum, the current number of used slots must be allocated, even
1511 * when deleting a memslot, as we need a complete duplicate of the memslots for
1512 * use when invalidating a memslot prior to deleting/moving the memslot.
1514 static struct kvm_memslots *kvm_dup_memslots(struct kvm_memslots *old,
1515 enum kvm_mr_change change)
1517 struct kvm_memslots *slots;
1520 if (change == KVM_MR_CREATE)
1521 new_size = kvm_memslots_size(old->used_slots + 1);
1523 new_size = kvm_memslots_size(old->used_slots);
1525 slots = kvzalloc(new_size, GFP_KERNEL_ACCOUNT);
1527 kvm_copy_memslots(slots, old);
1532 static int kvm_set_memslot(struct kvm *kvm,
1533 const struct kvm_userspace_memory_region *mem,
1534 struct kvm_memory_slot *old,
1535 struct kvm_memory_slot *new, int as_id,
1536 enum kvm_mr_change change)
1538 struct kvm_memory_slot *slot;
1539 struct kvm_memslots *slots;
1543 * Released in install_new_memslots.
1545 * Must be held from before the current memslots are copied until
1546 * after the new memslots are installed with rcu_assign_pointer,
1547 * then released before the synchronize srcu in install_new_memslots.
1549 * When modifying memslots outside of the slots_lock, must be held
1550 * before reading the pointer to the current memslots until after all
1551 * changes to those memslots are complete.
1553 * These rules ensure that installing new memslots does not lose
1554 * changes made to the previous memslots.
1556 mutex_lock(&kvm->slots_arch_lock);
1558 slots = kvm_dup_memslots(__kvm_memslots(kvm, as_id), change);
1560 mutex_unlock(&kvm->slots_arch_lock);
1564 if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) {
1566 * Note, the INVALID flag needs to be in the appropriate entry
1567 * in the freshly allocated memslots, not in @old or @new.
1569 slot = id_to_memslot(slots, old->id);
1570 slot->flags |= KVM_MEMSLOT_INVALID;
1573 * We can re-use the memory from the old memslots.
1574 * It will be overwritten with a copy of the new memslots
1575 * after reacquiring the slots_arch_lock below.
1577 slots = install_new_memslots(kvm, as_id, slots);
1579 /* From this point no new shadow pages pointing to a deleted,
1580 * or moved, memslot will be created.
1582 * validation of sp->gfn happens in:
1583 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1584 * - kvm_is_visible_gfn (mmu_check_root)
1586 kvm_arch_flush_shadow_memslot(kvm, slot);
1588 /* Released in install_new_memslots. */
1589 mutex_lock(&kvm->slots_arch_lock);
1592 * The arch-specific fields of the memslots could have changed
1593 * between releasing the slots_arch_lock in
1594 * install_new_memslots and here, so get a fresh copy of the
1597 kvm_copy_memslots(slots, __kvm_memslots(kvm, as_id));
1600 r = kvm_arch_prepare_memory_region(kvm, new, mem, change);
1604 update_memslots(slots, new, change);
1605 slots = install_new_memslots(kvm, as_id, slots);
1607 kvm_arch_commit_memory_region(kvm, mem, old, new, change);
1613 if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) {
1614 slot = id_to_memslot(slots, old->id);
1615 slot->flags &= ~KVM_MEMSLOT_INVALID;
1616 slots = install_new_memslots(kvm, as_id, slots);
1618 mutex_unlock(&kvm->slots_arch_lock);
1624 static int kvm_delete_memslot(struct kvm *kvm,
1625 const struct kvm_userspace_memory_region *mem,
1626 struct kvm_memory_slot *old, int as_id)
1628 struct kvm_memory_slot new;
1634 memset(&new, 0, sizeof(new));
1637 * This is only for debugging purpose; it should never be referenced
1638 * for a removed memslot.
1642 r = kvm_set_memslot(kvm, mem, old, &new, as_id, KVM_MR_DELETE);
1646 kvm_free_memslot(kvm, old);
1651 * Allocate some memory and give it an address in the guest physical address
1654 * Discontiguous memory is allowed, mostly for framebuffers.
1656 * Must be called holding kvm->slots_lock for write.
1658 int __kvm_set_memory_region(struct kvm *kvm,
1659 const struct kvm_userspace_memory_region *mem)
1661 struct kvm_memory_slot old, new;
1662 struct kvm_memory_slot *tmp;
1663 enum kvm_mr_change change;
1667 r = check_memory_region_flags(mem);
1671 as_id = mem->slot >> 16;
1672 id = (u16)mem->slot;
1674 /* General sanity checks */
1675 if (mem->memory_size & (PAGE_SIZE - 1))
1677 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
1679 /* We can read the guest memory with __xxx_user() later on. */
1680 if ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
1681 (mem->userspace_addr != untagged_addr(mem->userspace_addr)) ||
1682 !access_ok((void __user *)(unsigned long)mem->userspace_addr,
1685 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
1687 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
1691 * Make a full copy of the old memslot, the pointer will become stale
1692 * when the memslots are re-sorted by update_memslots(), and the old
1693 * memslot needs to be referenced after calling update_memslots(), e.g.
1694 * to free its resources and for arch specific behavior.
1696 tmp = id_to_memslot(__kvm_memslots(kvm, as_id), id);
1701 memset(&old, 0, sizeof(old));
1705 if (!mem->memory_size)
1706 return kvm_delete_memslot(kvm, mem, &old, as_id);
1710 new.base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
1711 new.npages = mem->memory_size >> PAGE_SHIFT;
1712 new.flags = mem->flags;
1713 new.userspace_addr = mem->userspace_addr;
1715 if (new.npages > KVM_MEM_MAX_NR_PAGES)
1719 change = KVM_MR_CREATE;
1720 new.dirty_bitmap = NULL;
1721 memset(&new.arch, 0, sizeof(new.arch));
1722 } else { /* Modify an existing slot. */
1723 if ((new.userspace_addr != old.userspace_addr) ||
1724 (new.npages != old.npages) ||
1725 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
1728 if (new.base_gfn != old.base_gfn)
1729 change = KVM_MR_MOVE;
1730 else if (new.flags != old.flags)
1731 change = KVM_MR_FLAGS_ONLY;
1732 else /* Nothing to change. */
1735 /* Copy dirty_bitmap and arch from the current memslot. */
1736 new.dirty_bitmap = old.dirty_bitmap;
1737 memcpy(&new.arch, &old.arch, sizeof(new.arch));
1740 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
1741 /* Check for overlaps */
1742 kvm_for_each_memslot(tmp, __kvm_memslots(kvm, as_id)) {
1745 if (!((new.base_gfn + new.npages <= tmp->base_gfn) ||
1746 (new.base_gfn >= tmp->base_gfn + tmp->npages)))
1751 /* Allocate/free page dirty bitmap as needed */
1752 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
1753 new.dirty_bitmap = NULL;
1754 else if (!new.dirty_bitmap && !kvm->dirty_ring_size) {
1755 r = kvm_alloc_dirty_bitmap(&new);
1759 if (kvm_dirty_log_manual_protect_and_init_set(kvm))
1760 bitmap_set(new.dirty_bitmap, 0, new.npages);
1763 r = kvm_set_memslot(kvm, mem, &old, &new, as_id, change);
1767 if (old.dirty_bitmap && !new.dirty_bitmap)
1768 kvm_destroy_dirty_bitmap(&old);
1772 if (new.dirty_bitmap && !old.dirty_bitmap)
1773 kvm_destroy_dirty_bitmap(&new);
1776 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1778 int kvm_set_memory_region(struct kvm *kvm,
1779 const struct kvm_userspace_memory_region *mem)
1783 mutex_lock(&kvm->slots_lock);
1784 r = __kvm_set_memory_region(kvm, mem);
1785 mutex_unlock(&kvm->slots_lock);
1788 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1790 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1791 struct kvm_userspace_memory_region *mem)
1793 if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1796 return kvm_set_memory_region(kvm, mem);
1799 #ifndef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1801 * kvm_get_dirty_log - get a snapshot of dirty pages
1802 * @kvm: pointer to kvm instance
1803 * @log: slot id and address to which we copy the log
1804 * @is_dirty: set to '1' if any dirty pages were found
1805 * @memslot: set to the associated memslot, always valid on success
1807 int kvm_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log,
1808 int *is_dirty, struct kvm_memory_slot **memslot)
1810 struct kvm_memslots *slots;
1813 unsigned long any = 0;
1815 /* Dirty ring tracking is exclusive to dirty log tracking */
1816 if (kvm->dirty_ring_size)
1822 as_id = log->slot >> 16;
1823 id = (u16)log->slot;
1824 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1827 slots = __kvm_memslots(kvm, as_id);
1828 *memslot = id_to_memslot(slots, id);
1829 if (!(*memslot) || !(*memslot)->dirty_bitmap)
1832 kvm_arch_sync_dirty_log(kvm, *memslot);
1834 n = kvm_dirty_bitmap_bytes(*memslot);
1836 for (i = 0; !any && i < n/sizeof(long); ++i)
1837 any = (*memslot)->dirty_bitmap[i];
1839 if (copy_to_user(log->dirty_bitmap, (*memslot)->dirty_bitmap, n))
1846 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1848 #else /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
1850 * kvm_get_dirty_log_protect - get a snapshot of dirty pages
1851 * and reenable dirty page tracking for the corresponding pages.
1852 * @kvm: pointer to kvm instance
1853 * @log: slot id and address to which we copy the log
1855 * We need to keep it in mind that VCPU threads can write to the bitmap
1856 * concurrently. So, to avoid losing track of dirty pages we keep the
1859 * 1. Take a snapshot of the bit and clear it if needed.
1860 * 2. Write protect the corresponding page.
1861 * 3. Copy the snapshot to the userspace.
1862 * 4. Upon return caller flushes TLB's if needed.
1864 * Between 2 and 4, the guest may write to the page using the remaining TLB
1865 * entry. This is not a problem because the page is reported dirty using
1866 * the snapshot taken before and step 4 ensures that writes done after
1867 * exiting to userspace will be logged for the next call.
1870 static int kvm_get_dirty_log_protect(struct kvm *kvm, struct kvm_dirty_log *log)
1872 struct kvm_memslots *slots;
1873 struct kvm_memory_slot *memslot;
1876 unsigned long *dirty_bitmap;
1877 unsigned long *dirty_bitmap_buffer;
1880 /* Dirty ring tracking is exclusive to dirty log tracking */
1881 if (kvm->dirty_ring_size)
1884 as_id = log->slot >> 16;
1885 id = (u16)log->slot;
1886 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1889 slots = __kvm_memslots(kvm, as_id);
1890 memslot = id_to_memslot(slots, id);
1891 if (!memslot || !memslot->dirty_bitmap)
1894 dirty_bitmap = memslot->dirty_bitmap;
1896 kvm_arch_sync_dirty_log(kvm, memslot);
1898 n = kvm_dirty_bitmap_bytes(memslot);
1900 if (kvm->manual_dirty_log_protect) {
1902 * Unlike kvm_get_dirty_log, we always return false in *flush,
1903 * because no flush is needed until KVM_CLEAR_DIRTY_LOG. There
1904 * is some code duplication between this function and
1905 * kvm_get_dirty_log, but hopefully all architecture
1906 * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log
1907 * can be eliminated.
1909 dirty_bitmap_buffer = dirty_bitmap;
1911 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
1912 memset(dirty_bitmap_buffer, 0, n);
1915 for (i = 0; i < n / sizeof(long); i++) {
1919 if (!dirty_bitmap[i])
1923 mask = xchg(&dirty_bitmap[i], 0);
1924 dirty_bitmap_buffer[i] = mask;
1926 offset = i * BITS_PER_LONG;
1927 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1930 KVM_MMU_UNLOCK(kvm);
1934 kvm_arch_flush_remote_tlbs_memslot(kvm, memslot);
1936 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1943 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
1944 * @kvm: kvm instance
1945 * @log: slot id and address to which we copy the log
1947 * Steps 1-4 below provide general overview of dirty page logging. See
1948 * kvm_get_dirty_log_protect() function description for additional details.
1950 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
1951 * always flush the TLB (step 4) even if previous step failed and the dirty
1952 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
1953 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
1954 * writes will be marked dirty for next log read.
1956 * 1. Take a snapshot of the bit and clear it if needed.
1957 * 2. Write protect the corresponding page.
1958 * 3. Copy the snapshot to the userspace.
1959 * 4. Flush TLB's if needed.
1961 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1962 struct kvm_dirty_log *log)
1966 mutex_lock(&kvm->slots_lock);
1968 r = kvm_get_dirty_log_protect(kvm, log);
1970 mutex_unlock(&kvm->slots_lock);
1975 * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap
1976 * and reenable dirty page tracking for the corresponding pages.
1977 * @kvm: pointer to kvm instance
1978 * @log: slot id and address from which to fetch the bitmap of dirty pages
1980 static int kvm_clear_dirty_log_protect(struct kvm *kvm,
1981 struct kvm_clear_dirty_log *log)
1983 struct kvm_memslots *slots;
1984 struct kvm_memory_slot *memslot;
1988 unsigned long *dirty_bitmap;
1989 unsigned long *dirty_bitmap_buffer;
1992 /* Dirty ring tracking is exclusive to dirty log tracking */
1993 if (kvm->dirty_ring_size)
1996 as_id = log->slot >> 16;
1997 id = (u16)log->slot;
1998 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
2001 if (log->first_page & 63)
2004 slots = __kvm_memslots(kvm, as_id);
2005 memslot = id_to_memslot(slots, id);
2006 if (!memslot || !memslot->dirty_bitmap)
2009 dirty_bitmap = memslot->dirty_bitmap;
2011 n = ALIGN(log->num_pages, BITS_PER_LONG) / 8;
2013 if (log->first_page > memslot->npages ||
2014 log->num_pages > memslot->npages - log->first_page ||
2015 (log->num_pages < memslot->npages - log->first_page && (log->num_pages & 63)))
2018 kvm_arch_sync_dirty_log(kvm, memslot);
2021 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
2022 if (copy_from_user(dirty_bitmap_buffer, log->dirty_bitmap, n))
2026 for (offset = log->first_page, i = offset / BITS_PER_LONG,
2027 n = DIV_ROUND_UP(log->num_pages, BITS_PER_LONG); n--;
2028 i++, offset += BITS_PER_LONG) {
2029 unsigned long mask = *dirty_bitmap_buffer++;
2030 atomic_long_t *p = (atomic_long_t *) &dirty_bitmap[i];
2034 mask &= atomic_long_fetch_andnot(mask, p);
2037 * mask contains the bits that really have been cleared. This
2038 * never includes any bits beyond the length of the memslot (if
2039 * the length is not aligned to 64 pages), therefore it is not
2040 * a problem if userspace sets them in log->dirty_bitmap.
2044 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
2048 KVM_MMU_UNLOCK(kvm);
2051 kvm_arch_flush_remote_tlbs_memslot(kvm, memslot);
2056 static int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm,
2057 struct kvm_clear_dirty_log *log)
2061 mutex_lock(&kvm->slots_lock);
2063 r = kvm_clear_dirty_log_protect(kvm, log);
2065 mutex_unlock(&kvm->slots_lock);
2068 #endif /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
2070 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
2072 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
2074 EXPORT_SYMBOL_GPL(gfn_to_memslot);
2076 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
2078 struct kvm_memslots *slots = kvm_vcpu_memslots(vcpu);
2079 struct kvm_memory_slot *slot;
2082 slot = try_get_memslot(slots, vcpu->last_used_slot, gfn);
2087 * Fall back to searching all memslots. We purposely use
2088 * search_memslots() instead of __gfn_to_memslot() to avoid
2089 * thrashing the VM-wide last_used_index in kvm_memslots.
2091 slot = search_memslots(slots, gfn, &slot_index);
2093 vcpu->last_used_slot = slot_index;
2099 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_memslot);
2101 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
2103 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
2105 return kvm_is_visible_memslot(memslot);
2107 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
2109 bool kvm_vcpu_is_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
2111 struct kvm_memory_slot *memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2113 return kvm_is_visible_memslot(memslot);
2115 EXPORT_SYMBOL_GPL(kvm_vcpu_is_visible_gfn);
2117 unsigned long kvm_host_page_size(struct kvm_vcpu *vcpu, gfn_t gfn)
2119 struct vm_area_struct *vma;
2120 unsigned long addr, size;
2124 addr = kvm_vcpu_gfn_to_hva_prot(vcpu, gfn, NULL);
2125 if (kvm_is_error_hva(addr))
2128 mmap_read_lock(current->mm);
2129 vma = find_vma(current->mm, addr);
2133 size = vma_kernel_pagesize(vma);
2136 mmap_read_unlock(current->mm);
2141 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
2143 return slot->flags & KVM_MEM_READONLY;
2146 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
2147 gfn_t *nr_pages, bool write)
2149 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
2150 return KVM_HVA_ERR_BAD;
2152 if (memslot_is_readonly(slot) && write)
2153 return KVM_HVA_ERR_RO_BAD;
2156 *nr_pages = slot->npages - (gfn - slot->base_gfn);
2158 return __gfn_to_hva_memslot(slot, gfn);
2161 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
2164 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
2167 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
2170 return gfn_to_hva_many(slot, gfn, NULL);
2172 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
2174 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
2176 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
2178 EXPORT_SYMBOL_GPL(gfn_to_hva);
2180 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
2182 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
2184 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
2187 * Return the hva of a @gfn and the R/W attribute if possible.
2189 * @slot: the kvm_memory_slot which contains @gfn
2190 * @gfn: the gfn to be translated
2191 * @writable: used to return the read/write attribute of the @slot if the hva
2192 * is valid and @writable is not NULL
2194 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
2195 gfn_t gfn, bool *writable)
2197 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
2199 if (!kvm_is_error_hva(hva) && writable)
2200 *writable = !memslot_is_readonly(slot);
2205 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
2207 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2209 return gfn_to_hva_memslot_prot(slot, gfn, writable);
2212 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
2214 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2216 return gfn_to_hva_memslot_prot(slot, gfn, writable);
2219 static inline int check_user_page_hwpoison(unsigned long addr)
2221 int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
2223 rc = get_user_pages(addr, 1, flags, NULL, NULL);
2224 return rc == -EHWPOISON;
2228 * The fast path to get the writable pfn which will be stored in @pfn,
2229 * true indicates success, otherwise false is returned. It's also the
2230 * only part that runs if we can in atomic context.
2232 static bool hva_to_pfn_fast(unsigned long addr, bool write_fault,
2233 bool *writable, kvm_pfn_t *pfn)
2235 struct page *page[1];
2238 * Fast pin a writable pfn only if it is a write fault request
2239 * or the caller allows to map a writable pfn for a read fault
2242 if (!(write_fault || writable))
2245 if (get_user_page_fast_only(addr, FOLL_WRITE, page)) {
2246 *pfn = page_to_pfn(page[0]);
2257 * The slow path to get the pfn of the specified host virtual address,
2258 * 1 indicates success, -errno is returned if error is detected.
2260 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
2261 bool *writable, kvm_pfn_t *pfn)
2263 unsigned int flags = FOLL_HWPOISON;
2270 *writable = write_fault;
2273 flags |= FOLL_WRITE;
2275 flags |= FOLL_NOWAIT;
2277 npages = get_user_pages_unlocked(addr, 1, &page, flags);
2281 /* map read fault as writable if possible */
2282 if (unlikely(!write_fault) && writable) {
2285 if (get_user_page_fast_only(addr, FOLL_WRITE, &wpage)) {
2291 *pfn = page_to_pfn(page);
2295 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
2297 if (unlikely(!(vma->vm_flags & VM_READ)))
2300 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
2306 static int kvm_try_get_pfn(kvm_pfn_t pfn)
2308 if (kvm_is_reserved_pfn(pfn))
2310 return get_page_unless_zero(pfn_to_page(pfn));
2313 static int hva_to_pfn_remapped(struct vm_area_struct *vma,
2314 unsigned long addr, bool *async,
2315 bool write_fault, bool *writable,
2323 r = follow_pte(vma->vm_mm, addr, &ptep, &ptl);
2326 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
2327 * not call the fault handler, so do it here.
2329 bool unlocked = false;
2330 r = fixup_user_fault(current->mm, addr,
2331 (write_fault ? FAULT_FLAG_WRITE : 0),
2338 r = follow_pte(vma->vm_mm, addr, &ptep, &ptl);
2343 if (write_fault && !pte_write(*ptep)) {
2344 pfn = KVM_PFN_ERR_RO_FAULT;
2349 *writable = pte_write(*ptep);
2350 pfn = pte_pfn(*ptep);
2353 * Get a reference here because callers of *hva_to_pfn* and
2354 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
2355 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
2356 * set, but the kvm_try_get_pfn/kvm_release_pfn_clean pair will
2357 * simply do nothing for reserved pfns.
2359 * Whoever called remap_pfn_range is also going to call e.g.
2360 * unmap_mapping_range before the underlying pages are freed,
2361 * causing a call to our MMU notifier.
2363 * Certain IO or PFNMAP mappings can be backed with valid
2364 * struct pages, but be allocated without refcounting e.g.,
2365 * tail pages of non-compound higher order allocations, which
2366 * would then underflow the refcount when the caller does the
2367 * required put_page. Don't allow those pages here.
2369 if (!kvm_try_get_pfn(pfn))
2373 pte_unmap_unlock(ptep, ptl);
2380 * Pin guest page in memory and return its pfn.
2381 * @addr: host virtual address which maps memory to the guest
2382 * @atomic: whether this function can sleep
2383 * @async: whether this function need to wait IO complete if the
2384 * host page is not in the memory
2385 * @write_fault: whether we should get a writable host page
2386 * @writable: whether it allows to map a writable host page for !@write_fault
2388 * The function will map a writable host page for these two cases:
2389 * 1): @write_fault = true
2390 * 2): @write_fault = false && @writable, @writable will tell the caller
2391 * whether the mapping is writable.
2393 static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
2394 bool write_fault, bool *writable)
2396 struct vm_area_struct *vma;
2400 /* we can do it either atomically or asynchronously, not both */
2401 BUG_ON(atomic && async);
2403 if (hva_to_pfn_fast(addr, write_fault, writable, &pfn))
2407 return KVM_PFN_ERR_FAULT;
2409 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
2413 mmap_read_lock(current->mm);
2414 if (npages == -EHWPOISON ||
2415 (!async && check_user_page_hwpoison(addr))) {
2416 pfn = KVM_PFN_ERR_HWPOISON;
2421 vma = vma_lookup(current->mm, addr);
2424 pfn = KVM_PFN_ERR_FAULT;
2425 else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
2426 r = hva_to_pfn_remapped(vma, addr, async, write_fault, writable, &pfn);
2430 pfn = KVM_PFN_ERR_FAULT;
2432 if (async && vma_is_valid(vma, write_fault))
2434 pfn = KVM_PFN_ERR_FAULT;
2437 mmap_read_unlock(current->mm);
2441 kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn,
2442 bool atomic, bool *async, bool write_fault,
2443 bool *writable, hva_t *hva)
2445 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
2450 if (addr == KVM_HVA_ERR_RO_BAD) {
2453 return KVM_PFN_ERR_RO_FAULT;
2456 if (kvm_is_error_hva(addr)) {
2459 return KVM_PFN_NOSLOT;
2462 /* Do not map writable pfn in the readonly memslot. */
2463 if (writable && memslot_is_readonly(slot)) {
2468 return hva_to_pfn(addr, atomic, async, write_fault,
2471 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
2473 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
2476 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
2477 write_fault, writable, NULL);
2479 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
2481 kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
2483 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL, NULL);
2485 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
2487 kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
2489 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL, NULL);
2491 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
2493 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
2495 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
2497 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
2499 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
2501 return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
2503 EXPORT_SYMBOL_GPL(gfn_to_pfn);
2505 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
2507 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
2509 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
2511 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
2512 struct page **pages, int nr_pages)
2517 addr = gfn_to_hva_many(slot, gfn, &entry);
2518 if (kvm_is_error_hva(addr))
2521 if (entry < nr_pages)
2524 return get_user_pages_fast_only(addr, nr_pages, FOLL_WRITE, pages);
2526 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
2528 static struct page *kvm_pfn_to_page(kvm_pfn_t pfn)
2530 if (is_error_noslot_pfn(pfn))
2531 return KVM_ERR_PTR_BAD_PAGE;
2533 if (kvm_is_reserved_pfn(pfn)) {
2535 return KVM_ERR_PTR_BAD_PAGE;
2538 return pfn_to_page(pfn);
2541 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
2545 pfn = gfn_to_pfn(kvm, gfn);
2547 return kvm_pfn_to_page(pfn);
2549 EXPORT_SYMBOL_GPL(gfn_to_page);
2551 void kvm_release_pfn(kvm_pfn_t pfn, bool dirty)
2557 kvm_release_pfn_dirty(pfn);
2559 kvm_release_pfn_clean(pfn);
2562 int kvm_vcpu_map(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map)
2566 struct page *page = KVM_UNMAPPED_PAGE;
2571 pfn = gfn_to_pfn(vcpu->kvm, gfn);
2572 if (is_error_noslot_pfn(pfn))
2575 if (pfn_valid(pfn)) {
2576 page = pfn_to_page(pfn);
2578 #ifdef CONFIG_HAS_IOMEM
2580 hva = memremap(pfn_to_hpa(pfn), PAGE_SIZE, MEMREMAP_WB);
2594 EXPORT_SYMBOL_GPL(kvm_vcpu_map);
2596 void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map, bool dirty)
2604 if (map->page != KVM_UNMAPPED_PAGE)
2606 #ifdef CONFIG_HAS_IOMEM
2612 kvm_vcpu_mark_page_dirty(vcpu, map->gfn);
2614 kvm_release_pfn(map->pfn, dirty);
2619 EXPORT_SYMBOL_GPL(kvm_vcpu_unmap);
2621 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
2625 pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
2627 return kvm_pfn_to_page(pfn);
2629 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
2631 void kvm_release_page_clean(struct page *page)
2633 WARN_ON(is_error_page(page));
2635 kvm_release_pfn_clean(page_to_pfn(page));
2637 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
2639 void kvm_release_pfn_clean(kvm_pfn_t pfn)
2641 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
2642 put_page(pfn_to_page(pfn));
2644 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
2646 void kvm_release_page_dirty(struct page *page)
2648 WARN_ON(is_error_page(page));
2650 kvm_release_pfn_dirty(page_to_pfn(page));
2652 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
2654 void kvm_release_pfn_dirty(kvm_pfn_t pfn)
2656 kvm_set_pfn_dirty(pfn);
2657 kvm_release_pfn_clean(pfn);
2659 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
2661 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
2663 if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn))
2664 SetPageDirty(pfn_to_page(pfn));
2666 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
2668 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
2670 if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn))
2671 mark_page_accessed(pfn_to_page(pfn));
2673 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
2675 static int next_segment(unsigned long len, int offset)
2677 if (len > PAGE_SIZE - offset)
2678 return PAGE_SIZE - offset;
2683 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
2684 void *data, int offset, int len)
2689 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
2690 if (kvm_is_error_hva(addr))
2692 r = __copy_from_user(data, (void __user *)addr + offset, len);
2698 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
2701 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2703 return __kvm_read_guest_page(slot, gfn, data, offset, len);
2705 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
2707 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
2708 int offset, int len)
2710 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2712 return __kvm_read_guest_page(slot, gfn, data, offset, len);
2714 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
2716 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
2718 gfn_t gfn = gpa >> PAGE_SHIFT;
2720 int offset = offset_in_page(gpa);
2723 while ((seg = next_segment(len, offset)) != 0) {
2724 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
2734 EXPORT_SYMBOL_GPL(kvm_read_guest);
2736 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
2738 gfn_t gfn = gpa >> PAGE_SHIFT;
2740 int offset = offset_in_page(gpa);
2743 while ((seg = next_segment(len, offset)) != 0) {
2744 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
2754 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
2756 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
2757 void *data, int offset, unsigned long len)
2762 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
2763 if (kvm_is_error_hva(addr))
2765 pagefault_disable();
2766 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
2773 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
2774 void *data, unsigned long len)
2776 gfn_t gfn = gpa >> PAGE_SHIFT;
2777 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2778 int offset = offset_in_page(gpa);
2780 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
2782 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
2784 static int __kvm_write_guest_page(struct kvm *kvm,
2785 struct kvm_memory_slot *memslot, gfn_t gfn,
2786 const void *data, int offset, int len)
2791 addr = gfn_to_hva_memslot(memslot, gfn);
2792 if (kvm_is_error_hva(addr))
2794 r = __copy_to_user((void __user *)addr + offset, data, len);
2797 mark_page_dirty_in_slot(kvm, memslot, gfn);
2801 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
2802 const void *data, int offset, int len)
2804 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2806 return __kvm_write_guest_page(kvm, slot, gfn, data, offset, len);
2808 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
2810 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
2811 const void *data, int offset, int len)
2813 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2815 return __kvm_write_guest_page(vcpu->kvm, slot, gfn, data, offset, len);
2817 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
2819 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
2822 gfn_t gfn = gpa >> PAGE_SHIFT;
2824 int offset = offset_in_page(gpa);
2827 while ((seg = next_segment(len, offset)) != 0) {
2828 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
2838 EXPORT_SYMBOL_GPL(kvm_write_guest);
2840 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
2843 gfn_t gfn = gpa >> PAGE_SHIFT;
2845 int offset = offset_in_page(gpa);
2848 while ((seg = next_segment(len, offset)) != 0) {
2849 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
2859 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
2861 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
2862 struct gfn_to_hva_cache *ghc,
2863 gpa_t gpa, unsigned long len)
2865 int offset = offset_in_page(gpa);
2866 gfn_t start_gfn = gpa >> PAGE_SHIFT;
2867 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
2868 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
2869 gfn_t nr_pages_avail;
2871 /* Update ghc->generation before performing any error checks. */
2872 ghc->generation = slots->generation;
2874 if (start_gfn > end_gfn) {
2875 ghc->hva = KVM_HVA_ERR_BAD;
2880 * If the requested region crosses two memslots, we still
2881 * verify that the entire region is valid here.
2883 for ( ; start_gfn <= end_gfn; start_gfn += nr_pages_avail) {
2884 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
2885 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
2887 if (kvm_is_error_hva(ghc->hva))
2891 /* Use the slow path for cross page reads and writes. */
2892 if (nr_pages_needed == 1)
2895 ghc->memslot = NULL;
2902 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2903 gpa_t gpa, unsigned long len)
2905 struct kvm_memslots *slots = kvm_memslots(kvm);
2906 return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len);
2908 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
2910 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2911 void *data, unsigned int offset,
2914 struct kvm_memslots *slots = kvm_memslots(kvm);
2916 gpa_t gpa = ghc->gpa + offset;
2918 BUG_ON(len + offset > ghc->len);
2920 if (slots->generation != ghc->generation) {
2921 if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len))
2925 if (kvm_is_error_hva(ghc->hva))
2928 if (unlikely(!ghc->memslot))
2929 return kvm_write_guest(kvm, gpa, data, len);
2931 r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
2934 mark_page_dirty_in_slot(kvm, ghc->memslot, gpa >> PAGE_SHIFT);
2938 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached);
2940 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2941 void *data, unsigned long len)
2943 return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len);
2945 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
2947 int kvm_read_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2948 void *data, unsigned int offset,
2951 struct kvm_memslots *slots = kvm_memslots(kvm);
2953 gpa_t gpa = ghc->gpa + offset;
2955 BUG_ON(len + offset > ghc->len);
2957 if (slots->generation != ghc->generation) {
2958 if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len))
2962 if (kvm_is_error_hva(ghc->hva))
2965 if (unlikely(!ghc->memslot))
2966 return kvm_read_guest(kvm, gpa, data, len);
2968 r = __copy_from_user(data, (void __user *)ghc->hva + offset, len);
2974 EXPORT_SYMBOL_GPL(kvm_read_guest_offset_cached);
2976 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2977 void *data, unsigned long len)
2979 return kvm_read_guest_offset_cached(kvm, ghc, data, 0, len);
2981 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
2983 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
2985 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
2986 gfn_t gfn = gpa >> PAGE_SHIFT;
2988 int offset = offset_in_page(gpa);
2991 while ((seg = next_segment(len, offset)) != 0) {
2992 ret = kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
3001 EXPORT_SYMBOL_GPL(kvm_clear_guest);
3003 void mark_page_dirty_in_slot(struct kvm *kvm,
3004 struct kvm_memory_slot *memslot,
3007 if (memslot && kvm_slot_dirty_track_enabled(memslot)) {
3008 unsigned long rel_gfn = gfn - memslot->base_gfn;
3009 u32 slot = (memslot->as_id << 16) | memslot->id;
3011 if (kvm->dirty_ring_size)
3012 kvm_dirty_ring_push(kvm_dirty_ring_get(kvm),
3015 set_bit_le(rel_gfn, memslot->dirty_bitmap);
3018 EXPORT_SYMBOL_GPL(mark_page_dirty_in_slot);
3020 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
3022 struct kvm_memory_slot *memslot;
3024 memslot = gfn_to_memslot(kvm, gfn);
3025 mark_page_dirty_in_slot(kvm, memslot, gfn);
3027 EXPORT_SYMBOL_GPL(mark_page_dirty);
3029 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
3031 struct kvm_memory_slot *memslot;
3033 memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
3034 mark_page_dirty_in_slot(vcpu->kvm, memslot, gfn);
3036 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
3038 void kvm_sigset_activate(struct kvm_vcpu *vcpu)
3040 if (!vcpu->sigset_active)
3044 * This does a lockless modification of ->real_blocked, which is fine
3045 * because, only current can change ->real_blocked and all readers of
3046 * ->real_blocked don't care as long ->real_blocked is always a subset
3049 sigprocmask(SIG_SETMASK, &vcpu->sigset, ¤t->real_blocked);
3052 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu)
3054 if (!vcpu->sigset_active)
3057 sigprocmask(SIG_SETMASK, ¤t->real_blocked, NULL);
3058 sigemptyset(¤t->real_blocked);
3061 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
3063 unsigned int old, val, grow, grow_start;
3065 old = val = vcpu->halt_poll_ns;
3066 grow_start = READ_ONCE(halt_poll_ns_grow_start);
3067 grow = READ_ONCE(halt_poll_ns_grow);
3072 if (val < grow_start)
3075 if (val > vcpu->kvm->max_halt_poll_ns)
3076 val = vcpu->kvm->max_halt_poll_ns;
3078 vcpu->halt_poll_ns = val;
3080 trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
3083 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
3085 unsigned int old, val, shrink, grow_start;
3087 old = val = vcpu->halt_poll_ns;
3088 shrink = READ_ONCE(halt_poll_ns_shrink);
3089 grow_start = READ_ONCE(halt_poll_ns_grow_start);
3095 if (val < grow_start)
3098 vcpu->halt_poll_ns = val;
3099 trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
3102 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
3105 int idx = srcu_read_lock(&vcpu->kvm->srcu);
3107 if (kvm_arch_vcpu_runnable(vcpu)) {
3108 kvm_make_request(KVM_REQ_UNHALT, vcpu);
3111 if (kvm_cpu_has_pending_timer(vcpu))
3113 if (signal_pending(current))
3115 if (kvm_check_request(KVM_REQ_UNBLOCK, vcpu))
3120 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3125 update_halt_poll_stats(struct kvm_vcpu *vcpu, u64 poll_ns, bool waited)
3128 vcpu->stat.generic.halt_poll_fail_ns += poll_ns;
3130 vcpu->stat.generic.halt_poll_success_ns += poll_ns;
3134 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
3136 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
3138 ktime_t start, cur, poll_end;
3139 bool waited = false;
3142 kvm_arch_vcpu_blocking(vcpu);
3144 start = cur = poll_end = ktime_get();
3145 if (vcpu->halt_poll_ns && !kvm_arch_no_poll(vcpu)) {
3146 ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns);
3148 ++vcpu->stat.generic.halt_attempted_poll;
3151 * This sets KVM_REQ_UNHALT if an interrupt
3154 if (kvm_vcpu_check_block(vcpu) < 0) {
3155 ++vcpu->stat.generic.halt_successful_poll;
3156 if (!vcpu_valid_wakeup(vcpu))
3157 ++vcpu->stat.generic.halt_poll_invalid;
3159 KVM_STATS_LOG_HIST_UPDATE(
3160 vcpu->stat.generic.halt_poll_success_hist,
3161 ktime_to_ns(ktime_get()) -
3162 ktime_to_ns(start));
3166 poll_end = cur = ktime_get();
3167 } while (kvm_vcpu_can_poll(cur, stop));
3169 KVM_STATS_LOG_HIST_UPDATE(
3170 vcpu->stat.generic.halt_poll_fail_hist,
3171 ktime_to_ns(ktime_get()) - ktime_to_ns(start));
3175 prepare_to_rcuwait(&vcpu->wait);
3177 set_current_state(TASK_INTERRUPTIBLE);
3179 if (kvm_vcpu_check_block(vcpu) < 0)
3185 finish_rcuwait(&vcpu->wait);
3188 vcpu->stat.generic.halt_wait_ns +=
3189 ktime_to_ns(cur) - ktime_to_ns(poll_end);
3190 KVM_STATS_LOG_HIST_UPDATE(vcpu->stat.generic.halt_wait_hist,
3191 ktime_to_ns(cur) - ktime_to_ns(poll_end));
3194 kvm_arch_vcpu_unblocking(vcpu);
3195 block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
3197 update_halt_poll_stats(
3198 vcpu, ktime_to_ns(ktime_sub(poll_end, start)), waited);
3200 if (!kvm_arch_no_poll(vcpu)) {
3201 if (!vcpu_valid_wakeup(vcpu)) {
3202 shrink_halt_poll_ns(vcpu);
3203 } else if (vcpu->kvm->max_halt_poll_ns) {
3204 if (block_ns <= vcpu->halt_poll_ns)
3206 /* we had a long block, shrink polling */
3207 else if (vcpu->halt_poll_ns &&
3208 block_ns > vcpu->kvm->max_halt_poll_ns)
3209 shrink_halt_poll_ns(vcpu);
3210 /* we had a short halt and our poll time is too small */
3211 else if (vcpu->halt_poll_ns < vcpu->kvm->max_halt_poll_ns &&
3212 block_ns < vcpu->kvm->max_halt_poll_ns)
3213 grow_halt_poll_ns(vcpu);
3215 vcpu->halt_poll_ns = 0;
3219 trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu));
3220 kvm_arch_vcpu_block_finish(vcpu);
3222 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
3224 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
3226 struct rcuwait *waitp;
3228 waitp = kvm_arch_vcpu_get_wait(vcpu);
3229 if (rcuwait_wake_up(waitp)) {
3230 WRITE_ONCE(vcpu->ready, true);
3231 ++vcpu->stat.generic.halt_wakeup;
3237 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
3241 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
3243 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
3247 if (kvm_vcpu_wake_up(vcpu))
3251 * Note, the vCPU could get migrated to a different pCPU at any point
3252 * after kvm_arch_vcpu_should_kick(), which could result in sending an
3253 * IPI to the previous pCPU. But, that's ok because the purpose of the
3254 * IPI is to force the vCPU to leave IN_GUEST_MODE, and migrating the
3255 * vCPU also requires it to leave IN_GUEST_MODE.
3258 if (kvm_arch_vcpu_should_kick(vcpu)) {
3259 cpu = READ_ONCE(vcpu->cpu);
3260 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
3261 smp_send_reschedule(cpu);
3265 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
3266 #endif /* !CONFIG_S390 */
3268 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
3271 struct task_struct *task = NULL;
3275 pid = rcu_dereference(target->pid);
3277 task = get_pid_task(pid, PIDTYPE_PID);
3281 ret = yield_to(task, 1);
3282 put_task_struct(task);
3286 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
3289 * Helper that checks whether a VCPU is eligible for directed yield.
3290 * Most eligible candidate to yield is decided by following heuristics:
3292 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
3293 * (preempted lock holder), indicated by @in_spin_loop.
3294 * Set at the beginning and cleared at the end of interception/PLE handler.
3296 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
3297 * chance last time (mostly it has become eligible now since we have probably
3298 * yielded to lockholder in last iteration. This is done by toggling
3299 * @dy_eligible each time a VCPU checked for eligibility.)
3301 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
3302 * to preempted lock-holder could result in wrong VCPU selection and CPU
3303 * burning. Giving priority for a potential lock-holder increases lock
3306 * Since algorithm is based on heuristics, accessing another VCPU data without
3307 * locking does not harm. It may result in trying to yield to same VCPU, fail
3308 * and continue with next VCPU and so on.
3310 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
3312 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
3315 eligible = !vcpu->spin_loop.in_spin_loop ||
3316 vcpu->spin_loop.dy_eligible;
3318 if (vcpu->spin_loop.in_spin_loop)
3319 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
3328 * Unlike kvm_arch_vcpu_runnable, this function is called outside
3329 * a vcpu_load/vcpu_put pair. However, for most architectures
3330 * kvm_arch_vcpu_runnable does not require vcpu_load.
3332 bool __weak kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
3334 return kvm_arch_vcpu_runnable(vcpu);
3337 static bool vcpu_dy_runnable(struct kvm_vcpu *vcpu)
3339 if (kvm_arch_dy_runnable(vcpu))
3342 #ifdef CONFIG_KVM_ASYNC_PF
3343 if (!list_empty_careful(&vcpu->async_pf.done))
3350 bool __weak kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu)
3355 void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode)
3357 struct kvm *kvm = me->kvm;
3358 struct kvm_vcpu *vcpu;
3359 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
3365 kvm_vcpu_set_in_spin_loop(me, true);
3367 * We boost the priority of a VCPU that is runnable but not
3368 * currently running, because it got preempted by something
3369 * else and called schedule in __vcpu_run. Hopefully that
3370 * VCPU is holding the lock that we need and will release it.
3371 * We approximate round-robin by starting at the last boosted VCPU.
3373 for (pass = 0; pass < 2 && !yielded && try; pass++) {
3374 kvm_for_each_vcpu(i, vcpu, kvm) {
3375 if (!pass && i <= last_boosted_vcpu) {
3376 i = last_boosted_vcpu;
3378 } else if (pass && i > last_boosted_vcpu)
3380 if (!READ_ONCE(vcpu->ready))
3384 if (rcuwait_active(&vcpu->wait) &&
3385 !vcpu_dy_runnable(vcpu))
3387 if (READ_ONCE(vcpu->preempted) && yield_to_kernel_mode &&
3388 !kvm_arch_dy_has_pending_interrupt(vcpu) &&
3389 !kvm_arch_vcpu_in_kernel(vcpu))
3391 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
3394 yielded = kvm_vcpu_yield_to(vcpu);
3396 kvm->last_boosted_vcpu = i;
3398 } else if (yielded < 0) {
3405 kvm_vcpu_set_in_spin_loop(me, false);
3407 /* Ensure vcpu is not eligible during next spinloop */
3408 kvm_vcpu_set_dy_eligible(me, false);
3410 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
3412 static bool kvm_page_in_dirty_ring(struct kvm *kvm, unsigned long pgoff)
3414 #if KVM_DIRTY_LOG_PAGE_OFFSET > 0
3415 return (pgoff >= KVM_DIRTY_LOG_PAGE_OFFSET) &&
3416 (pgoff < KVM_DIRTY_LOG_PAGE_OFFSET +
3417 kvm->dirty_ring_size / PAGE_SIZE);
3423 static vm_fault_t kvm_vcpu_fault(struct vm_fault *vmf)
3425 struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data;
3428 if (vmf->pgoff == 0)
3429 page = virt_to_page(vcpu->run);
3431 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
3432 page = virt_to_page(vcpu->arch.pio_data);
3434 #ifdef CONFIG_KVM_MMIO
3435 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
3436 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
3438 else if (kvm_page_in_dirty_ring(vcpu->kvm, vmf->pgoff))
3439 page = kvm_dirty_ring_get_page(
3441 vmf->pgoff - KVM_DIRTY_LOG_PAGE_OFFSET);
3443 return kvm_arch_vcpu_fault(vcpu, vmf);
3449 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
3450 .fault = kvm_vcpu_fault,
3453 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
3455 struct kvm_vcpu *vcpu = file->private_data;
3456 unsigned long pages = vma_pages(vma);
3458 if ((kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff) ||
3459 kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff + pages - 1)) &&
3460 ((vma->vm_flags & VM_EXEC) || !(vma->vm_flags & VM_SHARED)))
3463 vma->vm_ops = &kvm_vcpu_vm_ops;
3467 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
3469 struct kvm_vcpu *vcpu = filp->private_data;
3471 kvm_put_kvm(vcpu->kvm);
3475 static struct file_operations kvm_vcpu_fops = {
3476 .release = kvm_vcpu_release,
3477 .unlocked_ioctl = kvm_vcpu_ioctl,
3478 .mmap = kvm_vcpu_mmap,
3479 .llseek = noop_llseek,
3480 KVM_COMPAT(kvm_vcpu_compat_ioctl),
3484 * Allocates an inode for the vcpu.
3486 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
3488 char name[8 + 1 + ITOA_MAX_LEN + 1];
3490 snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id);
3491 return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
3494 static void kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
3496 #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS
3497 struct dentry *debugfs_dentry;
3498 char dir_name[ITOA_MAX_LEN * 2];
3500 if (!debugfs_initialized())
3503 snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
3504 debugfs_dentry = debugfs_create_dir(dir_name,
3505 vcpu->kvm->debugfs_dentry);
3507 kvm_arch_create_vcpu_debugfs(vcpu, debugfs_dentry);
3512 * Creates some virtual cpus. Good luck creating more than one.
3514 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
3517 struct kvm_vcpu *vcpu;
3520 if (id >= KVM_MAX_VCPU_IDS)
3523 mutex_lock(&kvm->lock);
3524 if (kvm->created_vcpus == KVM_MAX_VCPUS) {
3525 mutex_unlock(&kvm->lock);
3529 kvm->created_vcpus++;
3530 mutex_unlock(&kvm->lock);
3532 r = kvm_arch_vcpu_precreate(kvm, id);
3534 goto vcpu_decrement;
3536 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL_ACCOUNT);
3539 goto vcpu_decrement;
3542 BUILD_BUG_ON(sizeof(struct kvm_run) > PAGE_SIZE);
3543 page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
3548 vcpu->run = page_address(page);
3550 kvm_vcpu_init(vcpu, kvm, id);
3552 r = kvm_arch_vcpu_create(vcpu);
3554 goto vcpu_free_run_page;
3556 if (kvm->dirty_ring_size) {
3557 r = kvm_dirty_ring_alloc(&vcpu->dirty_ring,
3558 id, kvm->dirty_ring_size);
3560 goto arch_vcpu_destroy;
3563 mutex_lock(&kvm->lock);
3564 if (kvm_get_vcpu_by_id(kvm, id)) {
3566 goto unlock_vcpu_destroy;
3569 vcpu->vcpu_idx = atomic_read(&kvm->online_vcpus);
3570 BUG_ON(kvm->vcpus[vcpu->vcpu_idx]);
3572 /* Fill the stats id string for the vcpu */
3573 snprintf(vcpu->stats_id, sizeof(vcpu->stats_id), "kvm-%d/vcpu-%d",
3574 task_pid_nr(current), id);
3576 /* Now it's all set up, let userspace reach it */
3578 r = create_vcpu_fd(vcpu);
3580 kvm_put_kvm_no_destroy(kvm);
3581 goto unlock_vcpu_destroy;
3584 kvm->vcpus[vcpu->vcpu_idx] = vcpu;
3587 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
3588 * before kvm->online_vcpu's incremented value.
3591 atomic_inc(&kvm->online_vcpus);
3593 mutex_unlock(&kvm->lock);
3594 kvm_arch_vcpu_postcreate(vcpu);
3595 kvm_create_vcpu_debugfs(vcpu);
3598 unlock_vcpu_destroy:
3599 mutex_unlock(&kvm->lock);
3600 kvm_dirty_ring_free(&vcpu->dirty_ring);
3602 kvm_arch_vcpu_destroy(vcpu);
3604 free_page((unsigned long)vcpu->run);
3606 kmem_cache_free(kvm_vcpu_cache, vcpu);
3608 mutex_lock(&kvm->lock);
3609 kvm->created_vcpus--;
3610 mutex_unlock(&kvm->lock);
3614 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
3617 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
3618 vcpu->sigset_active = 1;
3619 vcpu->sigset = *sigset;
3621 vcpu->sigset_active = 0;
3625 static ssize_t kvm_vcpu_stats_read(struct file *file, char __user *user_buffer,
3626 size_t size, loff_t *offset)
3628 struct kvm_vcpu *vcpu = file->private_data;
3630 return kvm_stats_read(vcpu->stats_id, &kvm_vcpu_stats_header,
3631 &kvm_vcpu_stats_desc[0], &vcpu->stat,
3632 sizeof(vcpu->stat), user_buffer, size, offset);
3635 static const struct file_operations kvm_vcpu_stats_fops = {
3636 .read = kvm_vcpu_stats_read,
3637 .llseek = noop_llseek,
3640 static int kvm_vcpu_ioctl_get_stats_fd(struct kvm_vcpu *vcpu)
3644 char name[15 + ITOA_MAX_LEN + 1];
3646 snprintf(name, sizeof(name), "kvm-vcpu-stats:%d", vcpu->vcpu_id);
3648 fd = get_unused_fd_flags(O_CLOEXEC);
3652 file = anon_inode_getfile(name, &kvm_vcpu_stats_fops, vcpu, O_RDONLY);
3655 return PTR_ERR(file);
3657 file->f_mode |= FMODE_PREAD;
3658 fd_install(fd, file);
3663 static long kvm_vcpu_ioctl(struct file *filp,
3664 unsigned int ioctl, unsigned long arg)
3666 struct kvm_vcpu *vcpu = filp->private_data;
3667 void __user *argp = (void __user *)arg;
3669 struct kvm_fpu *fpu = NULL;
3670 struct kvm_sregs *kvm_sregs = NULL;
3672 if (vcpu->kvm->mm != current->mm || vcpu->kvm->vm_dead)
3675 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
3679 * Some architectures have vcpu ioctls that are asynchronous to vcpu
3680 * execution; mutex_lock() would break them.
3682 r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg);
3683 if (r != -ENOIOCTLCMD)
3686 if (mutex_lock_killable(&vcpu->mutex))
3694 oldpid = rcu_access_pointer(vcpu->pid);
3695 if (unlikely(oldpid != task_pid(current))) {
3696 /* The thread running this VCPU changed. */
3699 r = kvm_arch_vcpu_run_pid_change(vcpu);
3703 newpid = get_task_pid(current, PIDTYPE_PID);
3704 rcu_assign_pointer(vcpu->pid, newpid);
3709 r = kvm_arch_vcpu_ioctl_run(vcpu);
3710 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
3713 case KVM_GET_REGS: {
3714 struct kvm_regs *kvm_regs;
3717 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL_ACCOUNT);
3720 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
3724 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
3731 case KVM_SET_REGS: {
3732 struct kvm_regs *kvm_regs;
3734 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
3735 if (IS_ERR(kvm_regs)) {
3736 r = PTR_ERR(kvm_regs);
3739 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
3743 case KVM_GET_SREGS: {
3744 kvm_sregs = kzalloc(sizeof(struct kvm_sregs),
3745 GFP_KERNEL_ACCOUNT);
3749 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
3753 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
3758 case KVM_SET_SREGS: {
3759 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
3760 if (IS_ERR(kvm_sregs)) {
3761 r = PTR_ERR(kvm_sregs);
3765 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
3768 case KVM_GET_MP_STATE: {
3769 struct kvm_mp_state mp_state;
3771 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
3775 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
3780 case KVM_SET_MP_STATE: {
3781 struct kvm_mp_state mp_state;
3784 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
3786 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
3789 case KVM_TRANSLATE: {
3790 struct kvm_translation tr;
3793 if (copy_from_user(&tr, argp, sizeof(tr)))
3795 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
3799 if (copy_to_user(argp, &tr, sizeof(tr)))
3804 case KVM_SET_GUEST_DEBUG: {
3805 struct kvm_guest_debug dbg;
3808 if (copy_from_user(&dbg, argp, sizeof(dbg)))
3810 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
3813 case KVM_SET_SIGNAL_MASK: {
3814 struct kvm_signal_mask __user *sigmask_arg = argp;
3815 struct kvm_signal_mask kvm_sigmask;
3816 sigset_t sigset, *p;
3821 if (copy_from_user(&kvm_sigmask, argp,
3822 sizeof(kvm_sigmask)))
3825 if (kvm_sigmask.len != sizeof(sigset))
3828 if (copy_from_user(&sigset, sigmask_arg->sigset,
3833 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
3837 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL_ACCOUNT);
3841 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
3845 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
3851 fpu = memdup_user(argp, sizeof(*fpu));
3857 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
3860 case KVM_GET_STATS_FD: {
3861 r = kvm_vcpu_ioctl_get_stats_fd(vcpu);
3865 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
3868 mutex_unlock(&vcpu->mutex);
3874 #ifdef CONFIG_KVM_COMPAT
3875 static long kvm_vcpu_compat_ioctl(struct file *filp,
3876 unsigned int ioctl, unsigned long arg)
3878 struct kvm_vcpu *vcpu = filp->private_data;
3879 void __user *argp = compat_ptr(arg);
3882 if (vcpu->kvm->mm != current->mm || vcpu->kvm->vm_dead)
3886 case KVM_SET_SIGNAL_MASK: {
3887 struct kvm_signal_mask __user *sigmask_arg = argp;
3888 struct kvm_signal_mask kvm_sigmask;
3893 if (copy_from_user(&kvm_sigmask, argp,
3894 sizeof(kvm_sigmask)))
3897 if (kvm_sigmask.len != sizeof(compat_sigset_t))
3900 if (get_compat_sigset(&sigset,
3901 (compat_sigset_t __user *)sigmask_arg->sigset))
3903 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
3905 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
3909 r = kvm_vcpu_ioctl(filp, ioctl, arg);
3917 static int kvm_device_mmap(struct file *filp, struct vm_area_struct *vma)
3919 struct kvm_device *dev = filp->private_data;
3922 return dev->ops->mmap(dev, vma);
3927 static int kvm_device_ioctl_attr(struct kvm_device *dev,
3928 int (*accessor)(struct kvm_device *dev,
3929 struct kvm_device_attr *attr),
3932 struct kvm_device_attr attr;
3937 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
3940 return accessor(dev, &attr);
3943 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
3946 struct kvm_device *dev = filp->private_data;
3948 if (dev->kvm->mm != current->mm || dev->kvm->vm_dead)
3952 case KVM_SET_DEVICE_ATTR:
3953 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
3954 case KVM_GET_DEVICE_ATTR:
3955 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
3956 case KVM_HAS_DEVICE_ATTR:
3957 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
3959 if (dev->ops->ioctl)
3960 return dev->ops->ioctl(dev, ioctl, arg);
3966 static int kvm_device_release(struct inode *inode, struct file *filp)
3968 struct kvm_device *dev = filp->private_data;
3969 struct kvm *kvm = dev->kvm;
3971 if (dev->ops->release) {
3972 mutex_lock(&kvm->lock);
3973 list_del(&dev->vm_node);
3974 dev->ops->release(dev);
3975 mutex_unlock(&kvm->lock);
3982 static const struct file_operations kvm_device_fops = {
3983 .unlocked_ioctl = kvm_device_ioctl,
3984 .release = kvm_device_release,
3985 KVM_COMPAT(kvm_device_ioctl),
3986 .mmap = kvm_device_mmap,
3989 struct kvm_device *kvm_device_from_filp(struct file *filp)
3991 if (filp->f_op != &kvm_device_fops)
3994 return filp->private_data;
3997 static const struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
3998 #ifdef CONFIG_KVM_MPIC
3999 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
4000 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
4004 int kvm_register_device_ops(const struct kvm_device_ops *ops, u32 type)
4006 if (type >= ARRAY_SIZE(kvm_device_ops_table))
4009 if (kvm_device_ops_table[type] != NULL)
4012 kvm_device_ops_table[type] = ops;
4016 void kvm_unregister_device_ops(u32 type)
4018 if (kvm_device_ops_table[type] != NULL)
4019 kvm_device_ops_table[type] = NULL;
4022 static int kvm_ioctl_create_device(struct kvm *kvm,
4023 struct kvm_create_device *cd)
4025 const struct kvm_device_ops *ops = NULL;
4026 struct kvm_device *dev;
4027 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
4031 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
4034 type = array_index_nospec(cd->type, ARRAY_SIZE(kvm_device_ops_table));
4035 ops = kvm_device_ops_table[type];
4042 dev = kzalloc(sizeof(*dev), GFP_KERNEL_ACCOUNT);
4049 mutex_lock(&kvm->lock);
4050 ret = ops->create(dev, type);
4052 mutex_unlock(&kvm->lock);
4056 list_add(&dev->vm_node, &kvm->devices);
4057 mutex_unlock(&kvm->lock);
4063 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
4065 kvm_put_kvm_no_destroy(kvm);
4066 mutex_lock(&kvm->lock);
4067 list_del(&dev->vm_node);
4068 mutex_unlock(&kvm->lock);
4077 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
4080 case KVM_CAP_USER_MEMORY:
4081 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
4082 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
4083 case KVM_CAP_INTERNAL_ERROR_DATA:
4084 #ifdef CONFIG_HAVE_KVM_MSI
4085 case KVM_CAP_SIGNAL_MSI:
4087 #ifdef CONFIG_HAVE_KVM_IRQFD
4089 case KVM_CAP_IRQFD_RESAMPLE:
4091 case KVM_CAP_IOEVENTFD_ANY_LENGTH:
4092 case KVM_CAP_CHECK_EXTENSION_VM:
4093 case KVM_CAP_ENABLE_CAP_VM:
4094 case KVM_CAP_HALT_POLL:
4096 #ifdef CONFIG_KVM_MMIO
4097 case KVM_CAP_COALESCED_MMIO:
4098 return KVM_COALESCED_MMIO_PAGE_OFFSET;
4099 case KVM_CAP_COALESCED_PIO:
4102 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4103 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2:
4104 return KVM_DIRTY_LOG_MANUAL_CAPS;
4106 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
4107 case KVM_CAP_IRQ_ROUTING:
4108 return KVM_MAX_IRQ_ROUTES;
4110 #if KVM_ADDRESS_SPACE_NUM > 1
4111 case KVM_CAP_MULTI_ADDRESS_SPACE:
4112 return KVM_ADDRESS_SPACE_NUM;
4114 case KVM_CAP_NR_MEMSLOTS:
4115 return KVM_USER_MEM_SLOTS;
4116 case KVM_CAP_DIRTY_LOG_RING:
4117 #if KVM_DIRTY_LOG_PAGE_OFFSET > 0
4118 return KVM_DIRTY_RING_MAX_ENTRIES * sizeof(struct kvm_dirty_gfn);
4122 case KVM_CAP_BINARY_STATS_FD:
4127 return kvm_vm_ioctl_check_extension(kvm, arg);
4130 static int kvm_vm_ioctl_enable_dirty_log_ring(struct kvm *kvm, u32 size)
4134 if (!KVM_DIRTY_LOG_PAGE_OFFSET)
4137 /* the size should be power of 2 */
4138 if (!size || (size & (size - 1)))
4141 /* Should be bigger to keep the reserved entries, or a page */
4142 if (size < kvm_dirty_ring_get_rsvd_entries() *
4143 sizeof(struct kvm_dirty_gfn) || size < PAGE_SIZE)
4146 if (size > KVM_DIRTY_RING_MAX_ENTRIES *
4147 sizeof(struct kvm_dirty_gfn))
4150 /* We only allow it to set once */
4151 if (kvm->dirty_ring_size)
4154 mutex_lock(&kvm->lock);
4156 if (kvm->created_vcpus) {
4157 /* We don't allow to change this value after vcpu created */
4160 kvm->dirty_ring_size = size;
4164 mutex_unlock(&kvm->lock);
4168 static int kvm_vm_ioctl_reset_dirty_pages(struct kvm *kvm)
4171 struct kvm_vcpu *vcpu;
4174 if (!kvm->dirty_ring_size)
4177 mutex_lock(&kvm->slots_lock);
4179 kvm_for_each_vcpu(i, vcpu, kvm)
4180 cleared += kvm_dirty_ring_reset(vcpu->kvm, &vcpu->dirty_ring);
4182 mutex_unlock(&kvm->slots_lock);
4185 kvm_flush_remote_tlbs(kvm);
4190 int __attribute__((weak)) kvm_vm_ioctl_enable_cap(struct kvm *kvm,
4191 struct kvm_enable_cap *cap)
4196 static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm,
4197 struct kvm_enable_cap *cap)
4200 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4201 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2: {
4202 u64 allowed_options = KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE;
4204 if (cap->args[0] & KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE)
4205 allowed_options = KVM_DIRTY_LOG_MANUAL_CAPS;
4207 if (cap->flags || (cap->args[0] & ~allowed_options))
4209 kvm->manual_dirty_log_protect = cap->args[0];
4213 case KVM_CAP_HALT_POLL: {
4214 if (cap->flags || cap->args[0] != (unsigned int)cap->args[0])
4217 kvm->max_halt_poll_ns = cap->args[0];
4220 case KVM_CAP_DIRTY_LOG_RING:
4221 return kvm_vm_ioctl_enable_dirty_log_ring(kvm, cap->args[0]);
4223 return kvm_vm_ioctl_enable_cap(kvm, cap);
4227 static ssize_t kvm_vm_stats_read(struct file *file, char __user *user_buffer,
4228 size_t size, loff_t *offset)
4230 struct kvm *kvm = file->private_data;
4232 return kvm_stats_read(kvm->stats_id, &kvm_vm_stats_header,
4233 &kvm_vm_stats_desc[0], &kvm->stat,
4234 sizeof(kvm->stat), user_buffer, size, offset);
4237 static const struct file_operations kvm_vm_stats_fops = {
4238 .read = kvm_vm_stats_read,
4239 .llseek = noop_llseek,
4242 static int kvm_vm_ioctl_get_stats_fd(struct kvm *kvm)
4247 fd = get_unused_fd_flags(O_CLOEXEC);
4251 file = anon_inode_getfile("kvm-vm-stats",
4252 &kvm_vm_stats_fops, kvm, O_RDONLY);
4255 return PTR_ERR(file);
4257 file->f_mode |= FMODE_PREAD;
4258 fd_install(fd, file);
4263 static long kvm_vm_ioctl(struct file *filp,
4264 unsigned int ioctl, unsigned long arg)
4266 struct kvm *kvm = filp->private_data;
4267 void __user *argp = (void __user *)arg;
4270 if (kvm->mm != current->mm || kvm->vm_dead)
4273 case KVM_CREATE_VCPU:
4274 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
4276 case KVM_ENABLE_CAP: {
4277 struct kvm_enable_cap cap;
4280 if (copy_from_user(&cap, argp, sizeof(cap)))
4282 r = kvm_vm_ioctl_enable_cap_generic(kvm, &cap);
4285 case KVM_SET_USER_MEMORY_REGION: {
4286 struct kvm_userspace_memory_region kvm_userspace_mem;
4289 if (copy_from_user(&kvm_userspace_mem, argp,
4290 sizeof(kvm_userspace_mem)))
4293 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
4296 case KVM_GET_DIRTY_LOG: {
4297 struct kvm_dirty_log log;
4300 if (copy_from_user(&log, argp, sizeof(log)))
4302 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
4305 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4306 case KVM_CLEAR_DIRTY_LOG: {
4307 struct kvm_clear_dirty_log log;
4310 if (copy_from_user(&log, argp, sizeof(log)))
4312 r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
4316 #ifdef CONFIG_KVM_MMIO
4317 case KVM_REGISTER_COALESCED_MMIO: {
4318 struct kvm_coalesced_mmio_zone zone;
4321 if (copy_from_user(&zone, argp, sizeof(zone)))
4323 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
4326 case KVM_UNREGISTER_COALESCED_MMIO: {
4327 struct kvm_coalesced_mmio_zone zone;
4330 if (copy_from_user(&zone, argp, sizeof(zone)))
4332 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
4337 struct kvm_irqfd data;
4340 if (copy_from_user(&data, argp, sizeof(data)))
4342 r = kvm_irqfd(kvm, &data);
4345 case KVM_IOEVENTFD: {
4346 struct kvm_ioeventfd data;
4349 if (copy_from_user(&data, argp, sizeof(data)))
4351 r = kvm_ioeventfd(kvm, &data);
4354 #ifdef CONFIG_HAVE_KVM_MSI
4355 case KVM_SIGNAL_MSI: {
4359 if (copy_from_user(&msi, argp, sizeof(msi)))
4361 r = kvm_send_userspace_msi(kvm, &msi);
4365 #ifdef __KVM_HAVE_IRQ_LINE
4366 case KVM_IRQ_LINE_STATUS:
4367 case KVM_IRQ_LINE: {
4368 struct kvm_irq_level irq_event;
4371 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
4374 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
4375 ioctl == KVM_IRQ_LINE_STATUS);
4380 if (ioctl == KVM_IRQ_LINE_STATUS) {
4381 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
4389 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
4390 case KVM_SET_GSI_ROUTING: {
4391 struct kvm_irq_routing routing;
4392 struct kvm_irq_routing __user *urouting;
4393 struct kvm_irq_routing_entry *entries = NULL;
4396 if (copy_from_user(&routing, argp, sizeof(routing)))
4399 if (!kvm_arch_can_set_irq_routing(kvm))
4401 if (routing.nr > KVM_MAX_IRQ_ROUTES)
4407 entries = vmemdup_user(urouting->entries,
4408 array_size(sizeof(*entries),
4410 if (IS_ERR(entries)) {
4411 r = PTR_ERR(entries);
4415 r = kvm_set_irq_routing(kvm, entries, routing.nr,
4420 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
4421 case KVM_CREATE_DEVICE: {
4422 struct kvm_create_device cd;
4425 if (copy_from_user(&cd, argp, sizeof(cd)))
4428 r = kvm_ioctl_create_device(kvm, &cd);
4433 if (copy_to_user(argp, &cd, sizeof(cd)))
4439 case KVM_CHECK_EXTENSION:
4440 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
4442 case KVM_RESET_DIRTY_RINGS:
4443 r = kvm_vm_ioctl_reset_dirty_pages(kvm);
4445 case KVM_GET_STATS_FD:
4446 r = kvm_vm_ioctl_get_stats_fd(kvm);
4449 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
4455 #ifdef CONFIG_KVM_COMPAT
4456 struct compat_kvm_dirty_log {
4460 compat_uptr_t dirty_bitmap; /* one bit per page */
4465 struct compat_kvm_clear_dirty_log {
4470 compat_uptr_t dirty_bitmap; /* one bit per page */
4475 static long kvm_vm_compat_ioctl(struct file *filp,
4476 unsigned int ioctl, unsigned long arg)
4478 struct kvm *kvm = filp->private_data;
4481 if (kvm->mm != current->mm || kvm->vm_dead)
4484 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4485 case KVM_CLEAR_DIRTY_LOG: {
4486 struct compat_kvm_clear_dirty_log compat_log;
4487 struct kvm_clear_dirty_log log;
4489 if (copy_from_user(&compat_log, (void __user *)arg,
4490 sizeof(compat_log)))
4492 log.slot = compat_log.slot;
4493 log.num_pages = compat_log.num_pages;
4494 log.first_page = compat_log.first_page;
4495 log.padding2 = compat_log.padding2;
4496 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
4498 r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
4502 case KVM_GET_DIRTY_LOG: {
4503 struct compat_kvm_dirty_log compat_log;
4504 struct kvm_dirty_log log;
4506 if (copy_from_user(&compat_log, (void __user *)arg,
4507 sizeof(compat_log)))
4509 log.slot = compat_log.slot;
4510 log.padding1 = compat_log.padding1;
4511 log.padding2 = compat_log.padding2;
4512 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
4514 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
4518 r = kvm_vm_ioctl(filp, ioctl, arg);
4524 static struct file_operations kvm_vm_fops = {
4525 .release = kvm_vm_release,
4526 .unlocked_ioctl = kvm_vm_ioctl,
4527 .llseek = noop_llseek,
4528 KVM_COMPAT(kvm_vm_compat_ioctl),
4531 bool file_is_kvm(struct file *file)
4533 return file && file->f_op == &kvm_vm_fops;
4535 EXPORT_SYMBOL_GPL(file_is_kvm);
4537 static int kvm_dev_ioctl_create_vm(unsigned long type)
4543 kvm = kvm_create_vm(type);
4545 return PTR_ERR(kvm);
4546 #ifdef CONFIG_KVM_MMIO
4547 r = kvm_coalesced_mmio_init(kvm);
4551 r = get_unused_fd_flags(O_CLOEXEC);
4555 snprintf(kvm->stats_id, sizeof(kvm->stats_id),
4556 "kvm-%d", task_pid_nr(current));
4558 file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
4566 * Don't call kvm_put_kvm anymore at this point; file->f_op is
4567 * already set, with ->release() being kvm_vm_release(). In error
4568 * cases it will be called by the final fput(file) and will take
4569 * care of doing kvm_put_kvm(kvm).
4571 if (kvm_create_vm_debugfs(kvm, r) < 0) {
4576 kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm);
4578 fd_install(r, file);
4586 static long kvm_dev_ioctl(struct file *filp,
4587 unsigned int ioctl, unsigned long arg)
4592 case KVM_GET_API_VERSION:
4595 r = KVM_API_VERSION;
4598 r = kvm_dev_ioctl_create_vm(arg);
4600 case KVM_CHECK_EXTENSION:
4601 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
4603 case KVM_GET_VCPU_MMAP_SIZE:
4606 r = PAGE_SIZE; /* struct kvm_run */
4608 r += PAGE_SIZE; /* pio data page */
4610 #ifdef CONFIG_KVM_MMIO
4611 r += PAGE_SIZE; /* coalesced mmio ring page */
4614 case KVM_TRACE_ENABLE:
4615 case KVM_TRACE_PAUSE:
4616 case KVM_TRACE_DISABLE:
4620 return kvm_arch_dev_ioctl(filp, ioctl, arg);
4626 static struct file_operations kvm_chardev_ops = {
4627 .unlocked_ioctl = kvm_dev_ioctl,
4628 .llseek = noop_llseek,
4629 KVM_COMPAT(kvm_dev_ioctl),
4632 static struct miscdevice kvm_dev = {
4638 static void hardware_enable_nolock(void *junk)
4640 int cpu = raw_smp_processor_id();
4643 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
4646 cpumask_set_cpu(cpu, cpus_hardware_enabled);
4648 r = kvm_arch_hardware_enable();
4651 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
4652 atomic_inc(&hardware_enable_failed);
4653 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
4657 static int kvm_starting_cpu(unsigned int cpu)
4659 raw_spin_lock(&kvm_count_lock);
4660 if (kvm_usage_count)
4661 hardware_enable_nolock(NULL);
4662 raw_spin_unlock(&kvm_count_lock);
4666 static void hardware_disable_nolock(void *junk)
4668 int cpu = raw_smp_processor_id();
4670 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
4672 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
4673 kvm_arch_hardware_disable();
4676 static int kvm_dying_cpu(unsigned int cpu)
4678 raw_spin_lock(&kvm_count_lock);
4679 if (kvm_usage_count)
4680 hardware_disable_nolock(NULL);
4681 raw_spin_unlock(&kvm_count_lock);
4685 static void hardware_disable_all_nolock(void)
4687 BUG_ON(!kvm_usage_count);
4690 if (!kvm_usage_count)
4691 on_each_cpu(hardware_disable_nolock, NULL, 1);
4694 static void hardware_disable_all(void)
4696 raw_spin_lock(&kvm_count_lock);
4697 hardware_disable_all_nolock();
4698 raw_spin_unlock(&kvm_count_lock);
4701 static int hardware_enable_all(void)
4705 raw_spin_lock(&kvm_count_lock);
4708 if (kvm_usage_count == 1) {
4709 atomic_set(&hardware_enable_failed, 0);
4710 on_each_cpu(hardware_enable_nolock, NULL, 1);
4712 if (atomic_read(&hardware_enable_failed)) {
4713 hardware_disable_all_nolock();
4718 raw_spin_unlock(&kvm_count_lock);
4723 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
4727 * Some (well, at least mine) BIOSes hang on reboot if
4730 * And Intel TXT required VMX off for all cpu when system shutdown.
4732 pr_info("kvm: exiting hardware virtualization\n");
4733 kvm_rebooting = true;
4734 on_each_cpu(hardware_disable_nolock, NULL, 1);
4738 static struct notifier_block kvm_reboot_notifier = {
4739 .notifier_call = kvm_reboot,
4743 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
4747 for (i = 0; i < bus->dev_count; i++) {
4748 struct kvm_io_device *pos = bus->range[i].dev;
4750 kvm_iodevice_destructor(pos);
4755 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
4756 const struct kvm_io_range *r2)
4758 gpa_t addr1 = r1->addr;
4759 gpa_t addr2 = r2->addr;
4764 /* If r2->len == 0, match the exact address. If r2->len != 0,
4765 * accept any overlapping write. Any order is acceptable for
4766 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
4767 * we process all of them.
4780 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
4782 return kvm_io_bus_cmp(p1, p2);
4785 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
4786 gpa_t addr, int len)
4788 struct kvm_io_range *range, key;
4791 key = (struct kvm_io_range) {
4796 range = bsearch(&key, bus->range, bus->dev_count,
4797 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
4801 off = range - bus->range;
4803 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
4809 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
4810 struct kvm_io_range *range, const void *val)
4814 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
4818 while (idx < bus->dev_count &&
4819 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
4820 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
4829 /* kvm_io_bus_write - called under kvm->slots_lock */
4830 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
4831 int len, const void *val)
4833 struct kvm_io_bus *bus;
4834 struct kvm_io_range range;
4837 range = (struct kvm_io_range) {
4842 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
4845 r = __kvm_io_bus_write(vcpu, bus, &range, val);
4846 return r < 0 ? r : 0;
4848 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
4850 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
4851 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
4852 gpa_t addr, int len, const void *val, long cookie)
4854 struct kvm_io_bus *bus;
4855 struct kvm_io_range range;
4857 range = (struct kvm_io_range) {
4862 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
4866 /* First try the device referenced by cookie. */
4867 if ((cookie >= 0) && (cookie < bus->dev_count) &&
4868 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
4869 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
4874 * cookie contained garbage; fall back to search and return the
4875 * correct cookie value.
4877 return __kvm_io_bus_write(vcpu, bus, &range, val);
4880 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
4881 struct kvm_io_range *range, void *val)
4885 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
4889 while (idx < bus->dev_count &&
4890 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
4891 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
4900 /* kvm_io_bus_read - called under kvm->slots_lock */
4901 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
4904 struct kvm_io_bus *bus;
4905 struct kvm_io_range range;
4908 range = (struct kvm_io_range) {
4913 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
4916 r = __kvm_io_bus_read(vcpu, bus, &range, val);
4917 return r < 0 ? r : 0;
4920 /* Caller must hold slots_lock. */
4921 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
4922 int len, struct kvm_io_device *dev)
4925 struct kvm_io_bus *new_bus, *bus;
4926 struct kvm_io_range range;
4928 bus = kvm_get_bus(kvm, bus_idx);
4932 /* exclude ioeventfd which is limited by maximum fd */
4933 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
4936 new_bus = kmalloc(struct_size(bus, range, bus->dev_count + 1),
4937 GFP_KERNEL_ACCOUNT);
4941 range = (struct kvm_io_range) {
4947 for (i = 0; i < bus->dev_count; i++)
4948 if (kvm_io_bus_cmp(&bus->range[i], &range) > 0)
4951 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
4952 new_bus->dev_count++;
4953 new_bus->range[i] = range;
4954 memcpy(new_bus->range + i + 1, bus->range + i,
4955 (bus->dev_count - i) * sizeof(struct kvm_io_range));
4956 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
4957 synchronize_srcu_expedited(&kvm->srcu);
4963 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
4964 struct kvm_io_device *dev)
4967 struct kvm_io_bus *new_bus, *bus;
4969 lockdep_assert_held(&kvm->slots_lock);
4971 bus = kvm_get_bus(kvm, bus_idx);
4975 for (i = 0; i < bus->dev_count; i++) {
4976 if (bus->range[i].dev == dev) {
4981 if (i == bus->dev_count)
4984 new_bus = kmalloc(struct_size(bus, range, bus->dev_count - 1),
4985 GFP_KERNEL_ACCOUNT);
4987 memcpy(new_bus, bus, struct_size(bus, range, i));
4988 new_bus->dev_count--;
4989 memcpy(new_bus->range + i, bus->range + i + 1,
4990 flex_array_size(new_bus, range, new_bus->dev_count - i));
4993 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
4994 synchronize_srcu_expedited(&kvm->srcu);
4996 /* Destroy the old bus _after_ installing the (null) bus. */
4998 pr_err("kvm: failed to shrink bus, removing it completely\n");
4999 for (j = 0; j < bus->dev_count; j++) {
5002 kvm_iodevice_destructor(bus->range[j].dev);
5007 return new_bus ? 0 : -ENOMEM;
5010 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
5013 struct kvm_io_bus *bus;
5014 int dev_idx, srcu_idx;
5015 struct kvm_io_device *iodev = NULL;
5017 srcu_idx = srcu_read_lock(&kvm->srcu);
5019 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
5023 dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
5027 iodev = bus->range[dev_idx].dev;
5030 srcu_read_unlock(&kvm->srcu, srcu_idx);
5034 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
5036 static int kvm_debugfs_open(struct inode *inode, struct file *file,
5037 int (*get)(void *, u64 *), int (*set)(void *, u64),
5040 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
5044 * The debugfs files are a reference to the kvm struct which
5045 * is still valid when kvm_destroy_vm is called. kvm_get_kvm_safe
5046 * avoids the race between open and the removal of the debugfs directory.
5048 if (!kvm_get_kvm_safe(stat_data->kvm))
5051 if (simple_attr_open(inode, file, get,
5052 kvm_stats_debugfs_mode(stat_data->desc) & 0222
5055 kvm_put_kvm(stat_data->kvm);
5062 static int kvm_debugfs_release(struct inode *inode, struct file *file)
5064 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
5067 simple_attr_release(inode, file);
5068 kvm_put_kvm(stat_data->kvm);
5073 static int kvm_get_stat_per_vm(struct kvm *kvm, size_t offset, u64 *val)
5075 *val = *(u64 *)((void *)(&kvm->stat) + offset);
5080 static int kvm_clear_stat_per_vm(struct kvm *kvm, size_t offset)
5082 *(u64 *)((void *)(&kvm->stat) + offset) = 0;
5087 static int kvm_get_stat_per_vcpu(struct kvm *kvm, size_t offset, u64 *val)
5090 struct kvm_vcpu *vcpu;
5094 kvm_for_each_vcpu(i, vcpu, kvm)
5095 *val += *(u64 *)((void *)(&vcpu->stat) + offset);
5100 static int kvm_clear_stat_per_vcpu(struct kvm *kvm, size_t offset)
5103 struct kvm_vcpu *vcpu;
5105 kvm_for_each_vcpu(i, vcpu, kvm)
5106 *(u64 *)((void *)(&vcpu->stat) + offset) = 0;
5111 static int kvm_stat_data_get(void *data, u64 *val)
5114 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
5116 switch (stat_data->kind) {
5118 r = kvm_get_stat_per_vm(stat_data->kvm,
5119 stat_data->desc->desc.offset, val);
5122 r = kvm_get_stat_per_vcpu(stat_data->kvm,
5123 stat_data->desc->desc.offset, val);
5130 static int kvm_stat_data_clear(void *data, u64 val)
5133 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
5138 switch (stat_data->kind) {
5140 r = kvm_clear_stat_per_vm(stat_data->kvm,
5141 stat_data->desc->desc.offset);
5144 r = kvm_clear_stat_per_vcpu(stat_data->kvm,
5145 stat_data->desc->desc.offset);
5152 static int kvm_stat_data_open(struct inode *inode, struct file *file)
5154 __simple_attr_check_format("%llu\n", 0ull);
5155 return kvm_debugfs_open(inode, file, kvm_stat_data_get,
5156 kvm_stat_data_clear, "%llu\n");
5159 static const struct file_operations stat_fops_per_vm = {
5160 .owner = THIS_MODULE,
5161 .open = kvm_stat_data_open,
5162 .release = kvm_debugfs_release,
5163 .read = simple_attr_read,
5164 .write = simple_attr_write,
5165 .llseek = no_llseek,
5168 static int vm_stat_get(void *_offset, u64 *val)
5170 unsigned offset = (long)_offset;
5175 mutex_lock(&kvm_lock);
5176 list_for_each_entry(kvm, &vm_list, vm_list) {
5177 kvm_get_stat_per_vm(kvm, offset, &tmp_val);
5180 mutex_unlock(&kvm_lock);
5184 static int vm_stat_clear(void *_offset, u64 val)
5186 unsigned offset = (long)_offset;
5192 mutex_lock(&kvm_lock);
5193 list_for_each_entry(kvm, &vm_list, vm_list) {
5194 kvm_clear_stat_per_vm(kvm, offset);
5196 mutex_unlock(&kvm_lock);
5201 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n");
5202 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_readonly_fops, vm_stat_get, NULL, "%llu\n");
5204 static int vcpu_stat_get(void *_offset, u64 *val)
5206 unsigned offset = (long)_offset;
5211 mutex_lock(&kvm_lock);
5212 list_for_each_entry(kvm, &vm_list, vm_list) {
5213 kvm_get_stat_per_vcpu(kvm, offset, &tmp_val);
5216 mutex_unlock(&kvm_lock);
5220 static int vcpu_stat_clear(void *_offset, u64 val)
5222 unsigned offset = (long)_offset;
5228 mutex_lock(&kvm_lock);
5229 list_for_each_entry(kvm, &vm_list, vm_list) {
5230 kvm_clear_stat_per_vcpu(kvm, offset);
5232 mutex_unlock(&kvm_lock);
5237 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear,
5239 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_readonly_fops, vcpu_stat_get, NULL, "%llu\n");
5241 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm)
5243 struct kobj_uevent_env *env;
5244 unsigned long long created, active;
5246 if (!kvm_dev.this_device || !kvm)
5249 mutex_lock(&kvm_lock);
5250 if (type == KVM_EVENT_CREATE_VM) {
5251 kvm_createvm_count++;
5253 } else if (type == KVM_EVENT_DESTROY_VM) {
5256 created = kvm_createvm_count;
5257 active = kvm_active_vms;
5258 mutex_unlock(&kvm_lock);
5260 env = kzalloc(sizeof(*env), GFP_KERNEL_ACCOUNT);
5264 add_uevent_var(env, "CREATED=%llu", created);
5265 add_uevent_var(env, "COUNT=%llu", active);
5267 if (type == KVM_EVENT_CREATE_VM) {
5268 add_uevent_var(env, "EVENT=create");
5269 kvm->userspace_pid = task_pid_nr(current);
5270 } else if (type == KVM_EVENT_DESTROY_VM) {
5271 add_uevent_var(env, "EVENT=destroy");
5273 add_uevent_var(env, "PID=%d", kvm->userspace_pid);
5275 if (kvm->debugfs_dentry) {
5276 char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL_ACCOUNT);
5279 tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX);
5281 add_uevent_var(env, "STATS_PATH=%s", tmp);
5285 /* no need for checks, since we are adding at most only 5 keys */
5286 env->envp[env->envp_idx++] = NULL;
5287 kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp);
5291 static void kvm_init_debug(void)
5293 const struct file_operations *fops;
5294 const struct _kvm_stats_desc *pdesc;
5297 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
5299 for (i = 0; i < kvm_vm_stats_header.num_desc; ++i) {
5300 pdesc = &kvm_vm_stats_desc[i];
5301 if (kvm_stats_debugfs_mode(pdesc) & 0222)
5302 fops = &vm_stat_fops;
5304 fops = &vm_stat_readonly_fops;
5305 debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
5307 (void *)(long)pdesc->desc.offset, fops);
5310 for (i = 0; i < kvm_vcpu_stats_header.num_desc; ++i) {
5311 pdesc = &kvm_vcpu_stats_desc[i];
5312 if (kvm_stats_debugfs_mode(pdesc) & 0222)
5313 fops = &vcpu_stat_fops;
5315 fops = &vcpu_stat_readonly_fops;
5316 debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
5318 (void *)(long)pdesc->desc.offset, fops);
5322 static int kvm_suspend(void)
5324 if (kvm_usage_count)
5325 hardware_disable_nolock(NULL);
5329 static void kvm_resume(void)
5331 if (kvm_usage_count) {
5332 #ifdef CONFIG_LOCKDEP
5333 WARN_ON(lockdep_is_held(&kvm_count_lock));
5335 hardware_enable_nolock(NULL);
5339 static struct syscore_ops kvm_syscore_ops = {
5340 .suspend = kvm_suspend,
5341 .resume = kvm_resume,
5345 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
5347 return container_of(pn, struct kvm_vcpu, preempt_notifier);
5350 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
5352 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
5354 WRITE_ONCE(vcpu->preempted, false);
5355 WRITE_ONCE(vcpu->ready, false);
5357 __this_cpu_write(kvm_running_vcpu, vcpu);
5358 kvm_arch_sched_in(vcpu, cpu);
5359 kvm_arch_vcpu_load(vcpu, cpu);
5362 static void kvm_sched_out(struct preempt_notifier *pn,
5363 struct task_struct *next)
5365 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
5367 if (current->on_rq) {
5368 WRITE_ONCE(vcpu->preempted, true);
5369 WRITE_ONCE(vcpu->ready, true);
5371 kvm_arch_vcpu_put(vcpu);
5372 __this_cpu_write(kvm_running_vcpu, NULL);
5376 * kvm_get_running_vcpu - get the vcpu running on the current CPU.
5378 * We can disable preemption locally around accessing the per-CPU variable,
5379 * and use the resolved vcpu pointer after enabling preemption again,
5380 * because even if the current thread is migrated to another CPU, reading
5381 * the per-CPU value later will give us the same value as we update the
5382 * per-CPU variable in the preempt notifier handlers.
5384 struct kvm_vcpu *kvm_get_running_vcpu(void)
5386 struct kvm_vcpu *vcpu;
5389 vcpu = __this_cpu_read(kvm_running_vcpu);
5394 EXPORT_SYMBOL_GPL(kvm_get_running_vcpu);
5397 * kvm_get_running_vcpus - get the per-CPU array of currently running vcpus.
5399 struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void)
5401 return &kvm_running_vcpu;
5404 struct kvm_cpu_compat_check {
5409 static void check_processor_compat(void *data)
5411 struct kvm_cpu_compat_check *c = data;
5413 *c->ret = kvm_arch_check_processor_compat(c->opaque);
5416 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
5417 struct module *module)
5419 struct kvm_cpu_compat_check c;
5423 r = kvm_arch_init(opaque);
5428 * kvm_arch_init makes sure there's at most one caller
5429 * for architectures that support multiple implementations,
5430 * like intel and amd on x86.
5431 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
5432 * conflicts in case kvm is already setup for another implementation.
5434 r = kvm_irqfd_init();
5438 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
5443 r = kvm_arch_hardware_setup(opaque);
5449 for_each_online_cpu(cpu) {
5450 smp_call_function_single(cpu, check_processor_compat, &c, 1);
5455 r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting",
5456 kvm_starting_cpu, kvm_dying_cpu);
5459 register_reboot_notifier(&kvm_reboot_notifier);
5461 /* A kmem cache lets us meet the alignment requirements of fx_save. */
5463 vcpu_align = __alignof__(struct kvm_vcpu);
5465 kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align,
5467 offsetof(struct kvm_vcpu, arch),
5468 offsetofend(struct kvm_vcpu, stats_id)
5469 - offsetof(struct kvm_vcpu, arch),
5471 if (!kvm_vcpu_cache) {
5476 for_each_possible_cpu(cpu) {
5477 if (!alloc_cpumask_var_node(&per_cpu(cpu_kick_mask, cpu),
5478 GFP_KERNEL, cpu_to_node(cpu))) {
5484 r = kvm_async_pf_init();
5488 kvm_chardev_ops.owner = module;
5489 kvm_vm_fops.owner = module;
5490 kvm_vcpu_fops.owner = module;
5492 r = misc_register(&kvm_dev);
5494 pr_err("kvm: misc device register failed\n");
5498 register_syscore_ops(&kvm_syscore_ops);
5500 kvm_preempt_ops.sched_in = kvm_sched_in;
5501 kvm_preempt_ops.sched_out = kvm_sched_out;
5505 r = kvm_vfio_ops_init();
5511 kvm_async_pf_deinit();
5513 for_each_possible_cpu(cpu)
5514 free_cpumask_var(per_cpu(cpu_kick_mask, cpu));
5516 kmem_cache_destroy(kvm_vcpu_cache);
5518 unregister_reboot_notifier(&kvm_reboot_notifier);
5519 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
5521 kvm_arch_hardware_unsetup();
5523 free_cpumask_var(cpus_hardware_enabled);
5531 EXPORT_SYMBOL_GPL(kvm_init);
5537 debugfs_remove_recursive(kvm_debugfs_dir);
5538 misc_deregister(&kvm_dev);
5539 for_each_possible_cpu(cpu)
5540 free_cpumask_var(per_cpu(cpu_kick_mask, cpu));
5541 kmem_cache_destroy(kvm_vcpu_cache);
5542 kvm_async_pf_deinit();
5543 unregister_syscore_ops(&kvm_syscore_ops);
5544 unregister_reboot_notifier(&kvm_reboot_notifier);
5545 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
5546 on_each_cpu(hardware_disable_nolock, NULL, 1);
5547 kvm_arch_hardware_unsetup();
5550 free_cpumask_var(cpus_hardware_enabled);
5551 kvm_vfio_ops_exit();
5553 EXPORT_SYMBOL_GPL(kvm_exit);
5555 struct kvm_vm_worker_thread_context {
5557 struct task_struct *parent;
5558 struct completion init_done;
5559 kvm_vm_thread_fn_t thread_fn;
5564 static int kvm_vm_worker_thread(void *context)
5567 * The init_context is allocated on the stack of the parent thread, so
5568 * we have to locally copy anything that is needed beyond initialization
5570 struct kvm_vm_worker_thread_context *init_context = context;
5571 struct kvm *kvm = init_context->kvm;
5572 kvm_vm_thread_fn_t thread_fn = init_context->thread_fn;
5573 uintptr_t data = init_context->data;
5576 err = kthread_park(current);
5577 /* kthread_park(current) is never supposed to return an error */
5582 err = cgroup_attach_task_all(init_context->parent, current);
5584 kvm_err("%s: cgroup_attach_task_all failed with err %d\n",
5589 set_user_nice(current, task_nice(init_context->parent));
5592 init_context->err = err;
5593 complete(&init_context->init_done);
5594 init_context = NULL;
5599 /* Wait to be woken up by the spawner before proceeding. */
5602 if (!kthread_should_stop())
5603 err = thread_fn(kvm, data);
5608 int kvm_vm_create_worker_thread(struct kvm *kvm, kvm_vm_thread_fn_t thread_fn,
5609 uintptr_t data, const char *name,
5610 struct task_struct **thread_ptr)
5612 struct kvm_vm_worker_thread_context init_context = {};
5613 struct task_struct *thread;
5616 init_context.kvm = kvm;
5617 init_context.parent = current;
5618 init_context.thread_fn = thread_fn;
5619 init_context.data = data;
5620 init_completion(&init_context.init_done);
5622 thread = kthread_run(kvm_vm_worker_thread, &init_context,
5623 "%s-%d", name, task_pid_nr(current));
5625 return PTR_ERR(thread);
5627 /* kthread_run is never supposed to return NULL */
5628 WARN_ON(thread == NULL);
5630 wait_for_completion(&init_context.init_done);
5632 if (!init_context.err)
5633 *thread_ptr = thread;
5635 return init_context.err;