1 // SPDX-License-Identifier: GPL-2.0-only
3 * Kernel-based Virtual Machine driver for Linux
5 * This module enables machines with Intel VT-x extensions to run virtual
6 * machines without emulation or binary translation.
8 * Copyright (C) 2006 Qumranet, Inc.
9 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
12 * Avi Kivity <avi@qumranet.com>
13 * Yaniv Kamay <yaniv@qumranet.com>
16 #include <kvm/iodev.h>
18 #include <linux/kvm_host.h>
19 #include <linux/kvm.h>
20 #include <linux/module.h>
21 #include <linux/errno.h>
22 #include <linux/percpu.h>
24 #include <linux/miscdevice.h>
25 #include <linux/vmalloc.h>
26 #include <linux/reboot.h>
27 #include <linux/debugfs.h>
28 #include <linux/highmem.h>
29 #include <linux/file.h>
30 #include <linux/syscore_ops.h>
31 #include <linux/cpu.h>
32 #include <linux/sched/signal.h>
33 #include <linux/sched/mm.h>
34 #include <linux/sched/stat.h>
35 #include <linux/cpumask.h>
36 #include <linux/smp.h>
37 #include <linux/anon_inodes.h>
38 #include <linux/profile.h>
39 #include <linux/kvm_para.h>
40 #include <linux/pagemap.h>
41 #include <linux/mman.h>
42 #include <linux/swap.h>
43 #include <linux/bitops.h>
44 #include <linux/spinlock.h>
45 #include <linux/compat.h>
46 #include <linux/srcu.h>
47 #include <linux/hugetlb.h>
48 #include <linux/slab.h>
49 #include <linux/sort.h>
50 #include <linux/bsearch.h>
52 #include <linux/lockdep.h>
53 #include <linux/kthread.h>
55 #include <asm/processor.h>
56 #include <asm/ioctl.h>
57 #include <linux/uaccess.h>
59 #include "coalesced_mmio.h"
64 #define CREATE_TRACE_POINTS
65 #include <trace/events/kvm.h>
67 #include <linux/kvm_dirty_ring.h>
69 /* Worst case buffer size needed for holding an integer. */
70 #define ITOA_MAX_LEN 12
72 MODULE_AUTHOR("Qumranet");
73 MODULE_LICENSE("GPL");
75 /* Architectures should define their poll value according to the halt latency */
76 unsigned int halt_poll_ns = KVM_HALT_POLL_NS_DEFAULT;
77 module_param(halt_poll_ns, uint, 0644);
78 EXPORT_SYMBOL_GPL(halt_poll_ns);
80 /* Default doubles per-vcpu halt_poll_ns. */
81 unsigned int halt_poll_ns_grow = 2;
82 module_param(halt_poll_ns_grow, uint, 0644);
83 EXPORT_SYMBOL_GPL(halt_poll_ns_grow);
85 /* The start value to grow halt_poll_ns from */
86 unsigned int halt_poll_ns_grow_start = 10000; /* 10us */
87 module_param(halt_poll_ns_grow_start, uint, 0644);
88 EXPORT_SYMBOL_GPL(halt_poll_ns_grow_start);
90 /* Default resets per-vcpu halt_poll_ns . */
91 unsigned int halt_poll_ns_shrink;
92 module_param(halt_poll_ns_shrink, uint, 0644);
93 EXPORT_SYMBOL_GPL(halt_poll_ns_shrink);
98 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
101 DEFINE_MUTEX(kvm_lock);
102 static DEFINE_RAW_SPINLOCK(kvm_count_lock);
105 static cpumask_var_t cpus_hardware_enabled;
106 static int kvm_usage_count;
107 static atomic_t hardware_enable_failed;
109 static struct kmem_cache *kvm_vcpu_cache;
111 static __read_mostly struct preempt_ops kvm_preempt_ops;
112 static DEFINE_PER_CPU(struct kvm_vcpu *, kvm_running_vcpu);
114 struct dentry *kvm_debugfs_dir;
115 EXPORT_SYMBOL_GPL(kvm_debugfs_dir);
117 static int kvm_debugfs_num_entries;
118 static const struct file_operations stat_fops_per_vm;
120 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
122 #ifdef CONFIG_KVM_COMPAT
123 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
125 #define KVM_COMPAT(c) .compat_ioctl = (c)
128 * For architectures that don't implement a compat infrastructure,
129 * adopt a double line of defense:
130 * - Prevent a compat task from opening /dev/kvm
131 * - If the open has been done by a 64bit task, and the KVM fd
132 * passed to a compat task, let the ioctls fail.
134 static long kvm_no_compat_ioctl(struct file *file, unsigned int ioctl,
135 unsigned long arg) { return -EINVAL; }
137 static int kvm_no_compat_open(struct inode *inode, struct file *file)
139 return is_compat_task() ? -ENODEV : 0;
141 #define KVM_COMPAT(c) .compat_ioctl = kvm_no_compat_ioctl, \
142 .open = kvm_no_compat_open
144 static int hardware_enable_all(void);
145 static void hardware_disable_all(void);
147 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
149 __visible bool kvm_rebooting;
150 EXPORT_SYMBOL_GPL(kvm_rebooting);
152 #define KVM_EVENT_CREATE_VM 0
153 #define KVM_EVENT_DESTROY_VM 1
154 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm);
155 static unsigned long long kvm_createvm_count;
156 static unsigned long long kvm_active_vms;
158 __weak void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
159 unsigned long start, unsigned long end)
163 bool kvm_is_zone_device_pfn(kvm_pfn_t pfn)
166 * The metadata used by is_zone_device_page() to determine whether or
167 * not a page is ZONE_DEVICE is guaranteed to be valid if and only if
168 * the device has been pinned, e.g. by get_user_pages(). WARN if the
169 * page_count() is zero to help detect bad usage of this helper.
171 if (!pfn_valid(pfn) || WARN_ON_ONCE(!page_count(pfn_to_page(pfn))))
174 return is_zone_device_page(pfn_to_page(pfn));
177 bool kvm_is_reserved_pfn(kvm_pfn_t pfn)
180 * ZONE_DEVICE pages currently set PG_reserved, but from a refcounting
181 * perspective they are "normal" pages, albeit with slightly different
185 return PageReserved(pfn_to_page(pfn)) &&
187 !kvm_is_zone_device_pfn(pfn);
192 bool kvm_is_transparent_hugepage(kvm_pfn_t pfn)
194 struct page *page = pfn_to_page(pfn);
196 if (!PageTransCompoundMap(page))
199 return is_transparent_hugepage(compound_head(page));
203 * Switches to specified vcpu, until a matching vcpu_put()
205 void vcpu_load(struct kvm_vcpu *vcpu)
209 __this_cpu_write(kvm_running_vcpu, vcpu);
210 preempt_notifier_register(&vcpu->preempt_notifier);
211 kvm_arch_vcpu_load(vcpu, cpu);
214 EXPORT_SYMBOL_GPL(vcpu_load);
216 void vcpu_put(struct kvm_vcpu *vcpu)
219 kvm_arch_vcpu_put(vcpu);
220 preempt_notifier_unregister(&vcpu->preempt_notifier);
221 __this_cpu_write(kvm_running_vcpu, NULL);
224 EXPORT_SYMBOL_GPL(vcpu_put);
226 /* TODO: merge with kvm_arch_vcpu_should_kick */
227 static bool kvm_request_needs_ipi(struct kvm_vcpu *vcpu, unsigned req)
229 int mode = kvm_vcpu_exiting_guest_mode(vcpu);
232 * We need to wait for the VCPU to reenable interrupts and get out of
233 * READING_SHADOW_PAGE_TABLES mode.
235 if (req & KVM_REQUEST_WAIT)
236 return mode != OUTSIDE_GUEST_MODE;
239 * Need to kick a running VCPU, but otherwise there is nothing to do.
241 return mode == IN_GUEST_MODE;
244 static void ack_flush(void *_completed)
248 static inline bool kvm_kick_many_cpus(const struct cpumask *cpus, bool wait)
251 cpus = cpu_online_mask;
253 if (cpumask_empty(cpus))
256 smp_call_function_many(cpus, ack_flush, NULL, wait);
260 bool kvm_make_vcpus_request_mask(struct kvm *kvm, unsigned int req,
261 struct kvm_vcpu *except,
262 unsigned long *vcpu_bitmap, cpumask_var_t tmp)
265 struct kvm_vcpu *vcpu;
270 kvm_for_each_vcpu(i, vcpu, kvm) {
271 if ((vcpu_bitmap && !test_bit(i, vcpu_bitmap)) ||
275 kvm_make_request(req, vcpu);
278 if (!(req & KVM_REQUEST_NO_WAKEUP) && kvm_vcpu_wake_up(vcpu))
281 if (tmp != NULL && cpu != -1 && cpu != me &&
282 kvm_request_needs_ipi(vcpu, req))
283 __cpumask_set_cpu(cpu, tmp);
286 called = kvm_kick_many_cpus(tmp, !!(req & KVM_REQUEST_WAIT));
292 bool kvm_make_all_cpus_request_except(struct kvm *kvm, unsigned int req,
293 struct kvm_vcpu *except)
298 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
300 called = kvm_make_vcpus_request_mask(kvm, req, except, NULL, cpus);
302 free_cpumask_var(cpus);
306 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
308 return kvm_make_all_cpus_request_except(kvm, req, NULL);
311 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
312 void kvm_flush_remote_tlbs(struct kvm *kvm)
315 * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in
316 * kvm_make_all_cpus_request.
318 long dirty_count = smp_load_acquire(&kvm->tlbs_dirty);
321 * We want to publish modifications to the page tables before reading
322 * mode. Pairs with a memory barrier in arch-specific code.
323 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
324 * and smp_mb in walk_shadow_page_lockless_begin/end.
325 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
327 * There is already an smp_mb__after_atomic() before
328 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
331 if (!kvm_arch_flush_remote_tlb(kvm)
332 || kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
333 ++kvm->stat.remote_tlb_flush;
334 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
336 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
339 void kvm_reload_remote_mmus(struct kvm *kvm)
341 kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
344 #ifdef KVM_ARCH_NR_OBJS_PER_MEMORY_CACHE
345 static inline void *mmu_memory_cache_alloc_obj(struct kvm_mmu_memory_cache *mc,
348 gfp_flags |= mc->gfp_zero;
351 return kmem_cache_alloc(mc->kmem_cache, gfp_flags);
353 return (void *)__get_free_page(gfp_flags);
356 int kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache *mc, int min)
360 if (mc->nobjs >= min)
362 while (mc->nobjs < ARRAY_SIZE(mc->objects)) {
363 obj = mmu_memory_cache_alloc_obj(mc, GFP_KERNEL_ACCOUNT);
365 return mc->nobjs >= min ? 0 : -ENOMEM;
366 mc->objects[mc->nobjs++] = obj;
371 int kvm_mmu_memory_cache_nr_free_objects(struct kvm_mmu_memory_cache *mc)
376 void kvm_mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc)
380 kmem_cache_free(mc->kmem_cache, mc->objects[--mc->nobjs]);
382 free_page((unsigned long)mc->objects[--mc->nobjs]);
386 void *kvm_mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc)
390 if (WARN_ON(!mc->nobjs))
391 p = mmu_memory_cache_alloc_obj(mc, GFP_ATOMIC | __GFP_ACCOUNT);
393 p = mc->objects[--mc->nobjs];
399 static void kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
401 mutex_init(&vcpu->mutex);
406 rcuwait_init(&vcpu->wait);
407 kvm_async_pf_vcpu_init(vcpu);
410 INIT_LIST_HEAD(&vcpu->blocked_vcpu_list);
412 kvm_vcpu_set_in_spin_loop(vcpu, false);
413 kvm_vcpu_set_dy_eligible(vcpu, false);
414 vcpu->preempted = false;
416 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
419 void kvm_vcpu_destroy(struct kvm_vcpu *vcpu)
421 kvm_dirty_ring_free(&vcpu->dirty_ring);
422 kvm_arch_vcpu_destroy(vcpu);
425 * No need for rcu_read_lock as VCPU_RUN is the only place that changes
426 * the vcpu->pid pointer, and at destruction time all file descriptors
429 put_pid(rcu_dereference_protected(vcpu->pid, 1));
431 free_page((unsigned long)vcpu->run);
432 kmem_cache_free(kvm_vcpu_cache, vcpu);
434 EXPORT_SYMBOL_GPL(kvm_vcpu_destroy);
436 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
437 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
439 return container_of(mn, struct kvm, mmu_notifier);
442 static void kvm_mmu_notifier_invalidate_range(struct mmu_notifier *mn,
443 struct mm_struct *mm,
444 unsigned long start, unsigned long end)
446 struct kvm *kvm = mmu_notifier_to_kvm(mn);
449 idx = srcu_read_lock(&kvm->srcu);
450 kvm_arch_mmu_notifier_invalidate_range(kvm, start, end);
451 srcu_read_unlock(&kvm->srcu, idx);
454 typedef bool (*hva_handler_t)(struct kvm *kvm, struct kvm_gfn_range *range);
456 typedef void (*on_lock_fn_t)(struct kvm *kvm, unsigned long start,
459 struct kvm_hva_range {
463 hva_handler_t handler;
464 on_lock_fn_t on_lock;
470 * Use a dedicated stub instead of NULL to indicate that there is no callback
471 * function/handler. The compiler technically can't guarantee that a real
472 * function will have a non-zero address, and so it will generate code to
473 * check for !NULL, whereas comparing against a stub will be elided at compile
474 * time (unless the compiler is getting long in the tooth, e.g. gcc 4.9).
476 static void kvm_null_fn(void)
480 #define IS_KVM_NULL_FN(fn) ((fn) == (void *)kvm_null_fn)
482 static __always_inline int __kvm_handle_hva_range(struct kvm *kvm,
483 const struct kvm_hva_range *range)
485 bool ret = false, locked = false;
486 struct kvm_gfn_range gfn_range;
487 struct kvm_memory_slot *slot;
488 struct kvm_memslots *slots;
491 /* A null handler is allowed if and only if on_lock() is provided. */
492 if (WARN_ON_ONCE(IS_KVM_NULL_FN(range->on_lock) &&
493 IS_KVM_NULL_FN(range->handler)))
496 idx = srcu_read_lock(&kvm->srcu);
498 /* The on_lock() path does not yet support lock elision. */
499 if (!IS_KVM_NULL_FN(range->on_lock)) {
503 range->on_lock(kvm, range->start, range->end);
505 if (IS_KVM_NULL_FN(range->handler))
509 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
510 slots = __kvm_memslots(kvm, i);
511 kvm_for_each_memslot(slot, slots) {
512 unsigned long hva_start, hva_end;
514 hva_start = max(range->start, slot->userspace_addr);
515 hva_end = min(range->end, slot->userspace_addr +
516 (slot->npages << PAGE_SHIFT));
517 if (hva_start >= hva_end)
521 * To optimize for the likely case where the address
522 * range is covered by zero or one memslots, don't
523 * bother making these conditional (to avoid writes on
524 * the second or later invocation of the handler).
526 gfn_range.pte = range->pte;
527 gfn_range.may_block = range->may_block;
530 * {gfn(page) | page intersects with [hva_start, hva_end)} =
531 * {gfn_start, gfn_start+1, ..., gfn_end-1}.
533 gfn_range.start = hva_to_gfn_memslot(hva_start, slot);
534 gfn_range.end = hva_to_gfn_memslot(hva_end + PAGE_SIZE - 1, slot);
535 gfn_range.slot = slot;
541 ret |= range->handler(kvm, &gfn_range);
545 if (range->flush_on_ret && (ret || kvm->tlbs_dirty))
546 kvm_flush_remote_tlbs(kvm);
552 srcu_read_unlock(&kvm->srcu, idx);
554 /* The notifiers are averse to booleans. :-( */
558 static __always_inline int kvm_handle_hva_range(struct mmu_notifier *mn,
562 hva_handler_t handler)
564 struct kvm *kvm = mmu_notifier_to_kvm(mn);
565 const struct kvm_hva_range range = {
570 .on_lock = (void *)kvm_null_fn,
571 .flush_on_ret = true,
575 return __kvm_handle_hva_range(kvm, &range);
578 static __always_inline int kvm_handle_hva_range_no_flush(struct mmu_notifier *mn,
581 hva_handler_t handler)
583 struct kvm *kvm = mmu_notifier_to_kvm(mn);
584 const struct kvm_hva_range range = {
589 .on_lock = (void *)kvm_null_fn,
590 .flush_on_ret = false,
594 return __kvm_handle_hva_range(kvm, &range);
596 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
597 struct mm_struct *mm,
598 unsigned long address,
601 struct kvm *kvm = mmu_notifier_to_kvm(mn);
603 trace_kvm_set_spte_hva(address);
606 * .change_pte() must be surrounded by .invalidate_range_{start,end}(),
607 * and so always runs with an elevated notifier count. This obviates
608 * the need to bump the sequence count.
610 WARN_ON_ONCE(!kvm->mmu_notifier_count);
612 kvm_handle_hva_range(mn, address, address + 1, pte, kvm_set_spte_gfn);
615 static void kvm_inc_notifier_count(struct kvm *kvm, unsigned long start,
619 * The count increase must become visible at unlock time as no
620 * spte can be established without taking the mmu_lock and
621 * count is also read inside the mmu_lock critical section.
623 kvm->mmu_notifier_count++;
624 if (likely(kvm->mmu_notifier_count == 1)) {
625 kvm->mmu_notifier_range_start = start;
626 kvm->mmu_notifier_range_end = end;
629 * Fully tracking multiple concurrent ranges has dimishing
630 * returns. Keep things simple and just find the minimal range
631 * which includes the current and new ranges. As there won't be
632 * enough information to subtract a range after its invalidate
633 * completes, any ranges invalidated concurrently will
634 * accumulate and persist until all outstanding invalidates
637 kvm->mmu_notifier_range_start =
638 min(kvm->mmu_notifier_range_start, start);
639 kvm->mmu_notifier_range_end =
640 max(kvm->mmu_notifier_range_end, end);
644 static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
645 const struct mmu_notifier_range *range)
647 struct kvm *kvm = mmu_notifier_to_kvm(mn);
648 const struct kvm_hva_range hva_range = {
649 .start = range->start,
652 .handler = kvm_unmap_gfn_range,
653 .on_lock = kvm_inc_notifier_count,
654 .flush_on_ret = true,
655 .may_block = mmu_notifier_range_blockable(range),
658 trace_kvm_unmap_hva_range(range->start, range->end);
660 __kvm_handle_hva_range(kvm, &hva_range);
665 static void kvm_dec_notifier_count(struct kvm *kvm, unsigned long start,
669 * This sequence increase will notify the kvm page fault that
670 * the page that is going to be mapped in the spte could have
673 kvm->mmu_notifier_seq++;
676 * The above sequence increase must be visible before the
677 * below count decrease, which is ensured by the smp_wmb above
678 * in conjunction with the smp_rmb in mmu_notifier_retry().
680 kvm->mmu_notifier_count--;
683 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
684 const struct mmu_notifier_range *range)
686 struct kvm *kvm = mmu_notifier_to_kvm(mn);
687 const struct kvm_hva_range hva_range = {
688 .start = range->start,
691 .handler = (void *)kvm_null_fn,
692 .on_lock = kvm_dec_notifier_count,
693 .flush_on_ret = false,
694 .may_block = mmu_notifier_range_blockable(range),
697 __kvm_handle_hva_range(kvm, &hva_range);
699 BUG_ON(kvm->mmu_notifier_count < 0);
702 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
703 struct mm_struct *mm,
707 trace_kvm_age_hva(start, end);
709 return kvm_handle_hva_range(mn, start, end, __pte(0), kvm_age_gfn);
712 static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
713 struct mm_struct *mm,
717 trace_kvm_age_hva(start, end);
720 * Even though we do not flush TLB, this will still adversely
721 * affect performance on pre-Haswell Intel EPT, where there is
722 * no EPT Access Bit to clear so that we have to tear down EPT
723 * tables instead. If we find this unacceptable, we can always
724 * add a parameter to kvm_age_hva so that it effectively doesn't
725 * do anything on clear_young.
727 * Also note that currently we never issue secondary TLB flushes
728 * from clear_young, leaving this job up to the regular system
729 * cadence. If we find this inaccurate, we might come up with a
730 * more sophisticated heuristic later.
732 return kvm_handle_hva_range_no_flush(mn, start, end, kvm_age_gfn);
735 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
736 struct mm_struct *mm,
737 unsigned long address)
739 trace_kvm_test_age_hva(address);
741 return kvm_handle_hva_range_no_flush(mn, address, address + 1,
745 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
746 struct mm_struct *mm)
748 struct kvm *kvm = mmu_notifier_to_kvm(mn);
751 idx = srcu_read_lock(&kvm->srcu);
752 kvm_arch_flush_shadow_all(kvm);
753 srcu_read_unlock(&kvm->srcu, idx);
756 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
757 .invalidate_range = kvm_mmu_notifier_invalidate_range,
758 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
759 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
760 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
761 .clear_young = kvm_mmu_notifier_clear_young,
762 .test_young = kvm_mmu_notifier_test_young,
763 .change_pte = kvm_mmu_notifier_change_pte,
764 .release = kvm_mmu_notifier_release,
767 static int kvm_init_mmu_notifier(struct kvm *kvm)
769 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
770 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
773 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
775 static int kvm_init_mmu_notifier(struct kvm *kvm)
780 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
782 static struct kvm_memslots *kvm_alloc_memslots(void)
785 struct kvm_memslots *slots;
787 slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL_ACCOUNT);
791 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
792 slots->id_to_index[i] = -1;
797 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
799 if (!memslot->dirty_bitmap)
802 kvfree(memslot->dirty_bitmap);
803 memslot->dirty_bitmap = NULL;
806 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
808 kvm_destroy_dirty_bitmap(slot);
810 kvm_arch_free_memslot(kvm, slot);
816 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
818 struct kvm_memory_slot *memslot;
823 kvm_for_each_memslot(memslot, slots)
824 kvm_free_memslot(kvm, memslot);
829 static void kvm_destroy_vm_debugfs(struct kvm *kvm)
833 if (!kvm->debugfs_dentry)
836 debugfs_remove_recursive(kvm->debugfs_dentry);
838 if (kvm->debugfs_stat_data) {
839 for (i = 0; i < kvm_debugfs_num_entries; i++)
840 kfree(kvm->debugfs_stat_data[i]);
841 kfree(kvm->debugfs_stat_data);
845 static int kvm_create_vm_debugfs(struct kvm *kvm, int fd)
847 char dir_name[ITOA_MAX_LEN * 2];
848 struct kvm_stat_data *stat_data;
849 struct kvm_stats_debugfs_item *p;
851 if (!debugfs_initialized())
854 snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd);
855 kvm->debugfs_dentry = debugfs_create_dir(dir_name, kvm_debugfs_dir);
857 kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries,
858 sizeof(*kvm->debugfs_stat_data),
860 if (!kvm->debugfs_stat_data)
863 for (p = debugfs_entries; p->name; p++) {
864 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT);
868 stat_data->kvm = kvm;
869 stat_data->dbgfs_item = p;
870 kvm->debugfs_stat_data[p - debugfs_entries] = stat_data;
871 debugfs_create_file(p->name, KVM_DBGFS_GET_MODE(p),
872 kvm->debugfs_dentry, stat_data,
879 * Called after the VM is otherwise initialized, but just before adding it to
882 int __weak kvm_arch_post_init_vm(struct kvm *kvm)
888 * Called just after removing the VM from the vm_list, but before doing any
891 void __weak kvm_arch_pre_destroy_vm(struct kvm *kvm)
895 static struct kvm *kvm_create_vm(unsigned long type)
897 struct kvm *kvm = kvm_arch_alloc_vm();
902 return ERR_PTR(-ENOMEM);
904 KVM_MMU_LOCK_INIT(kvm);
906 kvm->mm = current->mm;
907 kvm_eventfd_init(kvm);
908 mutex_init(&kvm->lock);
909 mutex_init(&kvm->irq_lock);
910 mutex_init(&kvm->slots_lock);
911 INIT_LIST_HEAD(&kvm->devices);
913 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
915 if (init_srcu_struct(&kvm->srcu))
916 goto out_err_no_srcu;
917 if (init_srcu_struct(&kvm->irq_srcu))
918 goto out_err_no_irq_srcu;
920 refcount_set(&kvm->users_count, 1);
921 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
922 struct kvm_memslots *slots = kvm_alloc_memslots();
925 goto out_err_no_arch_destroy_vm;
926 /* Generations must be different for each address space. */
927 slots->generation = i;
928 rcu_assign_pointer(kvm->memslots[i], slots);
931 for (i = 0; i < KVM_NR_BUSES; i++) {
932 rcu_assign_pointer(kvm->buses[i],
933 kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL_ACCOUNT));
935 goto out_err_no_arch_destroy_vm;
938 kvm->max_halt_poll_ns = halt_poll_ns;
940 r = kvm_arch_init_vm(kvm, type);
942 goto out_err_no_arch_destroy_vm;
944 r = hardware_enable_all();
946 goto out_err_no_disable;
948 #ifdef CONFIG_HAVE_KVM_IRQFD
949 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
952 r = kvm_init_mmu_notifier(kvm);
954 goto out_err_no_mmu_notifier;
956 r = kvm_arch_post_init_vm(kvm);
960 mutex_lock(&kvm_lock);
961 list_add(&kvm->vm_list, &vm_list);
962 mutex_unlock(&kvm_lock);
964 preempt_notifier_inc();
969 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
970 if (kvm->mmu_notifier.ops)
971 mmu_notifier_unregister(&kvm->mmu_notifier, current->mm);
973 out_err_no_mmu_notifier:
974 hardware_disable_all();
976 kvm_arch_destroy_vm(kvm);
977 out_err_no_arch_destroy_vm:
978 WARN_ON_ONCE(!refcount_dec_and_test(&kvm->users_count));
979 for (i = 0; i < KVM_NR_BUSES; i++)
980 kfree(kvm_get_bus(kvm, i));
981 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
982 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
983 cleanup_srcu_struct(&kvm->irq_srcu);
985 cleanup_srcu_struct(&kvm->srcu);
987 kvm_arch_free_vm(kvm);
992 static void kvm_destroy_devices(struct kvm *kvm)
994 struct kvm_device *dev, *tmp;
997 * We do not need to take the kvm->lock here, because nobody else
998 * has a reference to the struct kvm at this point and therefore
999 * cannot access the devices list anyhow.
1001 list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
1002 list_del(&dev->vm_node);
1003 dev->ops->destroy(dev);
1007 static void kvm_destroy_vm(struct kvm *kvm)
1010 struct mm_struct *mm = kvm->mm;
1012 kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm);
1013 kvm_destroy_vm_debugfs(kvm);
1014 kvm_arch_sync_events(kvm);
1015 mutex_lock(&kvm_lock);
1016 list_del(&kvm->vm_list);
1017 mutex_unlock(&kvm_lock);
1018 kvm_arch_pre_destroy_vm(kvm);
1020 kvm_free_irq_routing(kvm);
1021 for (i = 0; i < KVM_NR_BUSES; i++) {
1022 struct kvm_io_bus *bus = kvm_get_bus(kvm, i);
1025 kvm_io_bus_destroy(bus);
1026 kvm->buses[i] = NULL;
1028 kvm_coalesced_mmio_free(kvm);
1029 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1030 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
1032 kvm_arch_flush_shadow_all(kvm);
1034 kvm_arch_destroy_vm(kvm);
1035 kvm_destroy_devices(kvm);
1036 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
1037 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
1038 cleanup_srcu_struct(&kvm->irq_srcu);
1039 cleanup_srcu_struct(&kvm->srcu);
1040 kvm_arch_free_vm(kvm);
1041 preempt_notifier_dec();
1042 hardware_disable_all();
1046 void kvm_get_kvm(struct kvm *kvm)
1048 refcount_inc(&kvm->users_count);
1050 EXPORT_SYMBOL_GPL(kvm_get_kvm);
1052 void kvm_put_kvm(struct kvm *kvm)
1054 if (refcount_dec_and_test(&kvm->users_count))
1055 kvm_destroy_vm(kvm);
1057 EXPORT_SYMBOL_GPL(kvm_put_kvm);
1060 * Used to put a reference that was taken on behalf of an object associated
1061 * with a user-visible file descriptor, e.g. a vcpu or device, if installation
1062 * of the new file descriptor fails and the reference cannot be transferred to
1063 * its final owner. In such cases, the caller is still actively using @kvm and
1064 * will fail miserably if the refcount unexpectedly hits zero.
1066 void kvm_put_kvm_no_destroy(struct kvm *kvm)
1068 WARN_ON(refcount_dec_and_test(&kvm->users_count));
1070 EXPORT_SYMBOL_GPL(kvm_put_kvm_no_destroy);
1072 static int kvm_vm_release(struct inode *inode, struct file *filp)
1074 struct kvm *kvm = filp->private_data;
1076 kvm_irqfd_release(kvm);
1083 * Allocation size is twice as large as the actual dirty bitmap size.
1084 * See kvm_vm_ioctl_get_dirty_log() why this is needed.
1086 static int kvm_alloc_dirty_bitmap(struct kvm_memory_slot *memslot)
1088 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
1090 memslot->dirty_bitmap = kvzalloc(dirty_bytes, GFP_KERNEL_ACCOUNT);
1091 if (!memslot->dirty_bitmap)
1098 * Delete a memslot by decrementing the number of used slots and shifting all
1099 * other entries in the array forward one spot.
1101 static inline void kvm_memslot_delete(struct kvm_memslots *slots,
1102 struct kvm_memory_slot *memslot)
1104 struct kvm_memory_slot *mslots = slots->memslots;
1107 if (WARN_ON(slots->id_to_index[memslot->id] == -1))
1110 slots->used_slots--;
1112 if (atomic_read(&slots->lru_slot) >= slots->used_slots)
1113 atomic_set(&slots->lru_slot, 0);
1115 for (i = slots->id_to_index[memslot->id]; i < slots->used_slots; i++) {
1116 mslots[i] = mslots[i + 1];
1117 slots->id_to_index[mslots[i].id] = i;
1119 mslots[i] = *memslot;
1120 slots->id_to_index[memslot->id] = -1;
1124 * "Insert" a new memslot by incrementing the number of used slots. Returns
1125 * the new slot's initial index into the memslots array.
1127 static inline int kvm_memslot_insert_back(struct kvm_memslots *slots)
1129 return slots->used_slots++;
1133 * Move a changed memslot backwards in the array by shifting existing slots
1134 * with a higher GFN toward the front of the array. Note, the changed memslot
1135 * itself is not preserved in the array, i.e. not swapped at this time, only
1136 * its new index into the array is tracked. Returns the changed memslot's
1137 * current index into the memslots array.
1139 static inline int kvm_memslot_move_backward(struct kvm_memslots *slots,
1140 struct kvm_memory_slot *memslot)
1142 struct kvm_memory_slot *mslots = slots->memslots;
1145 if (WARN_ON_ONCE(slots->id_to_index[memslot->id] == -1) ||
1146 WARN_ON_ONCE(!slots->used_slots))
1150 * Move the target memslot backward in the array by shifting existing
1151 * memslots with a higher GFN (than the target memslot) towards the
1152 * front of the array.
1154 for (i = slots->id_to_index[memslot->id]; i < slots->used_slots - 1; i++) {
1155 if (memslot->base_gfn > mslots[i + 1].base_gfn)
1158 WARN_ON_ONCE(memslot->base_gfn == mslots[i + 1].base_gfn);
1160 /* Shift the next memslot forward one and update its index. */
1161 mslots[i] = mslots[i + 1];
1162 slots->id_to_index[mslots[i].id] = i;
1168 * Move a changed memslot forwards in the array by shifting existing slots with
1169 * a lower GFN toward the back of the array. Note, the changed memslot itself
1170 * is not preserved in the array, i.e. not swapped at this time, only its new
1171 * index into the array is tracked. Returns the changed memslot's final index
1172 * into the memslots array.
1174 static inline int kvm_memslot_move_forward(struct kvm_memslots *slots,
1175 struct kvm_memory_slot *memslot,
1178 struct kvm_memory_slot *mslots = slots->memslots;
1181 for (i = start; i > 0; i--) {
1182 if (memslot->base_gfn < mslots[i - 1].base_gfn)
1185 WARN_ON_ONCE(memslot->base_gfn == mslots[i - 1].base_gfn);
1187 /* Shift the next memslot back one and update its index. */
1188 mslots[i] = mslots[i - 1];
1189 slots->id_to_index[mslots[i].id] = i;
1195 * Re-sort memslots based on their GFN to account for an added, deleted, or
1196 * moved memslot. Sorting memslots by GFN allows using a binary search during
1199 * IMPORTANT: Slots are sorted from highest GFN to lowest GFN! I.e. the entry
1200 * at memslots[0] has the highest GFN.
1202 * The sorting algorithm takes advantage of having initially sorted memslots
1203 * and knowing the position of the changed memslot. Sorting is also optimized
1204 * by not swapping the updated memslot and instead only shifting other memslots
1205 * and tracking the new index for the update memslot. Only once its final
1206 * index is known is the updated memslot copied into its position in the array.
1208 * - When deleting a memslot, the deleted memslot simply needs to be moved to
1209 * the end of the array.
1211 * - When creating a memslot, the algorithm "inserts" the new memslot at the
1212 * end of the array and then it forward to its correct location.
1214 * - When moving a memslot, the algorithm first moves the updated memslot
1215 * backward to handle the scenario where the memslot's GFN was changed to a
1216 * lower value. update_memslots() then falls through and runs the same flow
1217 * as creating a memslot to move the memslot forward to handle the scenario
1218 * where its GFN was changed to a higher value.
1220 * Note, slots are sorted from highest->lowest instead of lowest->highest for
1221 * historical reasons. Originally, invalid memslots where denoted by having
1222 * GFN=0, thus sorting from highest->lowest naturally sorted invalid memslots
1223 * to the end of the array. The current algorithm uses dedicated logic to
1224 * delete a memslot and thus does not rely on invalid memslots having GFN=0.
1226 * The other historical motiviation for highest->lowest was to improve the
1227 * performance of memslot lookup. KVM originally used a linear search starting
1228 * at memslots[0]. On x86, the largest memslot usually has one of the highest,
1229 * if not *the* highest, GFN, as the bulk of the guest's RAM is located in a
1230 * single memslot above the 4gb boundary. As the largest memslot is also the
1231 * most likely to be referenced, sorting it to the front of the array was
1232 * advantageous. The current binary search starts from the middle of the array
1233 * and uses an LRU pointer to improve performance for all memslots and GFNs.
1235 static void update_memslots(struct kvm_memslots *slots,
1236 struct kvm_memory_slot *memslot,
1237 enum kvm_mr_change change)
1241 if (change == KVM_MR_DELETE) {
1242 kvm_memslot_delete(slots, memslot);
1244 if (change == KVM_MR_CREATE)
1245 i = kvm_memslot_insert_back(slots);
1247 i = kvm_memslot_move_backward(slots, memslot);
1248 i = kvm_memslot_move_forward(slots, memslot, i);
1251 * Copy the memslot to its new position in memslots and update
1252 * its index accordingly.
1254 slots->memslots[i] = *memslot;
1255 slots->id_to_index[memslot->id] = i;
1259 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
1261 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
1263 #ifdef __KVM_HAVE_READONLY_MEM
1264 valid_flags |= KVM_MEM_READONLY;
1267 if (mem->flags & ~valid_flags)
1273 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
1274 int as_id, struct kvm_memslots *slots)
1276 struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id);
1277 u64 gen = old_memslots->generation;
1279 WARN_ON(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS);
1280 slots->generation = gen | KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
1282 rcu_assign_pointer(kvm->memslots[as_id], slots);
1283 synchronize_srcu_expedited(&kvm->srcu);
1286 * Increment the new memslot generation a second time, dropping the
1287 * update in-progress flag and incrementing the generation based on
1288 * the number of address spaces. This provides a unique and easily
1289 * identifiable generation number while the memslots are in flux.
1291 gen = slots->generation & ~KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
1294 * Generations must be unique even across address spaces. We do not need
1295 * a global counter for that, instead the generation space is evenly split
1296 * across address spaces. For example, with two address spaces, address
1297 * space 0 will use generations 0, 2, 4, ... while address space 1 will
1298 * use generations 1, 3, 5, ...
1300 gen += KVM_ADDRESS_SPACE_NUM;
1302 kvm_arch_memslots_updated(kvm, gen);
1304 slots->generation = gen;
1306 return old_memslots;
1310 * Note, at a minimum, the current number of used slots must be allocated, even
1311 * when deleting a memslot, as we need a complete duplicate of the memslots for
1312 * use when invalidating a memslot prior to deleting/moving the memslot.
1314 static struct kvm_memslots *kvm_dup_memslots(struct kvm_memslots *old,
1315 enum kvm_mr_change change)
1317 struct kvm_memslots *slots;
1318 size_t old_size, new_size;
1320 old_size = sizeof(struct kvm_memslots) +
1321 (sizeof(struct kvm_memory_slot) * old->used_slots);
1323 if (change == KVM_MR_CREATE)
1324 new_size = old_size + sizeof(struct kvm_memory_slot);
1326 new_size = old_size;
1328 slots = kvzalloc(new_size, GFP_KERNEL_ACCOUNT);
1330 memcpy(slots, old, old_size);
1335 static int kvm_set_memslot(struct kvm *kvm,
1336 const struct kvm_userspace_memory_region *mem,
1337 struct kvm_memory_slot *old,
1338 struct kvm_memory_slot *new, int as_id,
1339 enum kvm_mr_change change)
1341 struct kvm_memory_slot *slot;
1342 struct kvm_memslots *slots;
1345 slots = kvm_dup_memslots(__kvm_memslots(kvm, as_id), change);
1349 if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) {
1351 * Note, the INVALID flag needs to be in the appropriate entry
1352 * in the freshly allocated memslots, not in @old or @new.
1354 slot = id_to_memslot(slots, old->id);
1355 slot->flags |= KVM_MEMSLOT_INVALID;
1358 * We can re-use the old memslots, the only difference from the
1359 * newly installed memslots is the invalid flag, which will get
1360 * dropped by update_memslots anyway. We'll also revert to the
1361 * old memslots if preparing the new memory region fails.
1363 slots = install_new_memslots(kvm, as_id, slots);
1365 /* From this point no new shadow pages pointing to a deleted,
1366 * or moved, memslot will be created.
1368 * validation of sp->gfn happens in:
1369 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1370 * - kvm_is_visible_gfn (mmu_check_root)
1372 kvm_arch_flush_shadow_memslot(kvm, slot);
1375 r = kvm_arch_prepare_memory_region(kvm, new, mem, change);
1379 update_memslots(slots, new, change);
1380 slots = install_new_memslots(kvm, as_id, slots);
1382 kvm_arch_commit_memory_region(kvm, mem, old, new, change);
1388 if (change == KVM_MR_DELETE || change == KVM_MR_MOVE)
1389 slots = install_new_memslots(kvm, as_id, slots);
1394 static int kvm_delete_memslot(struct kvm *kvm,
1395 const struct kvm_userspace_memory_region *mem,
1396 struct kvm_memory_slot *old, int as_id)
1398 struct kvm_memory_slot new;
1404 memset(&new, 0, sizeof(new));
1407 * This is only for debugging purpose; it should never be referenced
1408 * for a removed memslot.
1412 r = kvm_set_memslot(kvm, mem, old, &new, as_id, KVM_MR_DELETE);
1416 kvm_free_memslot(kvm, old);
1421 * Allocate some memory and give it an address in the guest physical address
1424 * Discontiguous memory is allowed, mostly for framebuffers.
1426 * Must be called holding kvm->slots_lock for write.
1428 int __kvm_set_memory_region(struct kvm *kvm,
1429 const struct kvm_userspace_memory_region *mem)
1431 struct kvm_memory_slot old, new;
1432 struct kvm_memory_slot *tmp;
1433 enum kvm_mr_change change;
1437 r = check_memory_region_flags(mem);
1441 as_id = mem->slot >> 16;
1442 id = (u16)mem->slot;
1444 /* General sanity checks */
1445 if (mem->memory_size & (PAGE_SIZE - 1))
1447 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
1449 /* We can read the guest memory with __xxx_user() later on. */
1450 if ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
1451 (mem->userspace_addr != untagged_addr(mem->userspace_addr)) ||
1452 !access_ok((void __user *)(unsigned long)mem->userspace_addr,
1455 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
1457 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
1461 * Make a full copy of the old memslot, the pointer will become stale
1462 * when the memslots are re-sorted by update_memslots(), and the old
1463 * memslot needs to be referenced after calling update_memslots(), e.g.
1464 * to free its resources and for arch specific behavior.
1466 tmp = id_to_memslot(__kvm_memslots(kvm, as_id), id);
1471 memset(&old, 0, sizeof(old));
1475 if (!mem->memory_size)
1476 return kvm_delete_memslot(kvm, mem, &old, as_id);
1480 new.base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
1481 new.npages = mem->memory_size >> PAGE_SHIFT;
1482 new.flags = mem->flags;
1483 new.userspace_addr = mem->userspace_addr;
1485 if (new.npages > KVM_MEM_MAX_NR_PAGES)
1489 change = KVM_MR_CREATE;
1490 new.dirty_bitmap = NULL;
1491 memset(&new.arch, 0, sizeof(new.arch));
1492 } else { /* Modify an existing slot. */
1493 if ((new.userspace_addr != old.userspace_addr) ||
1494 (new.npages != old.npages) ||
1495 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
1498 if (new.base_gfn != old.base_gfn)
1499 change = KVM_MR_MOVE;
1500 else if (new.flags != old.flags)
1501 change = KVM_MR_FLAGS_ONLY;
1502 else /* Nothing to change. */
1505 /* Copy dirty_bitmap and arch from the current memslot. */
1506 new.dirty_bitmap = old.dirty_bitmap;
1507 memcpy(&new.arch, &old.arch, sizeof(new.arch));
1510 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
1511 /* Check for overlaps */
1512 kvm_for_each_memslot(tmp, __kvm_memslots(kvm, as_id)) {
1515 if (!((new.base_gfn + new.npages <= tmp->base_gfn) ||
1516 (new.base_gfn >= tmp->base_gfn + tmp->npages)))
1521 /* Allocate/free page dirty bitmap as needed */
1522 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
1523 new.dirty_bitmap = NULL;
1524 else if (!new.dirty_bitmap && !kvm->dirty_ring_size) {
1525 r = kvm_alloc_dirty_bitmap(&new);
1529 if (kvm_dirty_log_manual_protect_and_init_set(kvm))
1530 bitmap_set(new.dirty_bitmap, 0, new.npages);
1533 r = kvm_set_memslot(kvm, mem, &old, &new, as_id, change);
1537 if (old.dirty_bitmap && !new.dirty_bitmap)
1538 kvm_destroy_dirty_bitmap(&old);
1542 if (new.dirty_bitmap && !old.dirty_bitmap)
1543 kvm_destroy_dirty_bitmap(&new);
1546 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1548 int kvm_set_memory_region(struct kvm *kvm,
1549 const struct kvm_userspace_memory_region *mem)
1553 mutex_lock(&kvm->slots_lock);
1554 r = __kvm_set_memory_region(kvm, mem);
1555 mutex_unlock(&kvm->slots_lock);
1558 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1560 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1561 struct kvm_userspace_memory_region *mem)
1563 if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1566 return kvm_set_memory_region(kvm, mem);
1569 #ifndef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1571 * kvm_get_dirty_log - get a snapshot of dirty pages
1572 * @kvm: pointer to kvm instance
1573 * @log: slot id and address to which we copy the log
1574 * @is_dirty: set to '1' if any dirty pages were found
1575 * @memslot: set to the associated memslot, always valid on success
1577 int kvm_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log,
1578 int *is_dirty, struct kvm_memory_slot **memslot)
1580 struct kvm_memslots *slots;
1583 unsigned long any = 0;
1585 /* Dirty ring tracking is exclusive to dirty log tracking */
1586 if (kvm->dirty_ring_size)
1592 as_id = log->slot >> 16;
1593 id = (u16)log->slot;
1594 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1597 slots = __kvm_memslots(kvm, as_id);
1598 *memslot = id_to_memslot(slots, id);
1599 if (!(*memslot) || !(*memslot)->dirty_bitmap)
1602 kvm_arch_sync_dirty_log(kvm, *memslot);
1604 n = kvm_dirty_bitmap_bytes(*memslot);
1606 for (i = 0; !any && i < n/sizeof(long); ++i)
1607 any = (*memslot)->dirty_bitmap[i];
1609 if (copy_to_user(log->dirty_bitmap, (*memslot)->dirty_bitmap, n))
1616 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1618 #else /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
1620 * kvm_get_dirty_log_protect - get a snapshot of dirty pages
1621 * and reenable dirty page tracking for the corresponding pages.
1622 * @kvm: pointer to kvm instance
1623 * @log: slot id and address to which we copy the log
1625 * We need to keep it in mind that VCPU threads can write to the bitmap
1626 * concurrently. So, to avoid losing track of dirty pages we keep the
1629 * 1. Take a snapshot of the bit and clear it if needed.
1630 * 2. Write protect the corresponding page.
1631 * 3. Copy the snapshot to the userspace.
1632 * 4. Upon return caller flushes TLB's if needed.
1634 * Between 2 and 4, the guest may write to the page using the remaining TLB
1635 * entry. This is not a problem because the page is reported dirty using
1636 * the snapshot taken before and step 4 ensures that writes done after
1637 * exiting to userspace will be logged for the next call.
1640 static int kvm_get_dirty_log_protect(struct kvm *kvm, struct kvm_dirty_log *log)
1642 struct kvm_memslots *slots;
1643 struct kvm_memory_slot *memslot;
1646 unsigned long *dirty_bitmap;
1647 unsigned long *dirty_bitmap_buffer;
1650 /* Dirty ring tracking is exclusive to dirty log tracking */
1651 if (kvm->dirty_ring_size)
1654 as_id = log->slot >> 16;
1655 id = (u16)log->slot;
1656 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1659 slots = __kvm_memslots(kvm, as_id);
1660 memslot = id_to_memslot(slots, id);
1661 if (!memslot || !memslot->dirty_bitmap)
1664 dirty_bitmap = memslot->dirty_bitmap;
1666 kvm_arch_sync_dirty_log(kvm, memslot);
1668 n = kvm_dirty_bitmap_bytes(memslot);
1670 if (kvm->manual_dirty_log_protect) {
1672 * Unlike kvm_get_dirty_log, we always return false in *flush,
1673 * because no flush is needed until KVM_CLEAR_DIRTY_LOG. There
1674 * is some code duplication between this function and
1675 * kvm_get_dirty_log, but hopefully all architecture
1676 * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log
1677 * can be eliminated.
1679 dirty_bitmap_buffer = dirty_bitmap;
1681 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
1682 memset(dirty_bitmap_buffer, 0, n);
1685 for (i = 0; i < n / sizeof(long); i++) {
1689 if (!dirty_bitmap[i])
1693 mask = xchg(&dirty_bitmap[i], 0);
1694 dirty_bitmap_buffer[i] = mask;
1696 offset = i * BITS_PER_LONG;
1697 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1700 KVM_MMU_UNLOCK(kvm);
1704 kvm_arch_flush_remote_tlbs_memslot(kvm, memslot);
1706 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1713 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
1714 * @kvm: kvm instance
1715 * @log: slot id and address to which we copy the log
1717 * Steps 1-4 below provide general overview of dirty page logging. See
1718 * kvm_get_dirty_log_protect() function description for additional details.
1720 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
1721 * always flush the TLB (step 4) even if previous step failed and the dirty
1722 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
1723 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
1724 * writes will be marked dirty for next log read.
1726 * 1. Take a snapshot of the bit and clear it if needed.
1727 * 2. Write protect the corresponding page.
1728 * 3. Copy the snapshot to the userspace.
1729 * 4. Flush TLB's if needed.
1731 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1732 struct kvm_dirty_log *log)
1736 mutex_lock(&kvm->slots_lock);
1738 r = kvm_get_dirty_log_protect(kvm, log);
1740 mutex_unlock(&kvm->slots_lock);
1745 * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap
1746 * and reenable dirty page tracking for the corresponding pages.
1747 * @kvm: pointer to kvm instance
1748 * @log: slot id and address from which to fetch the bitmap of dirty pages
1750 static int kvm_clear_dirty_log_protect(struct kvm *kvm,
1751 struct kvm_clear_dirty_log *log)
1753 struct kvm_memslots *slots;
1754 struct kvm_memory_slot *memslot;
1758 unsigned long *dirty_bitmap;
1759 unsigned long *dirty_bitmap_buffer;
1762 /* Dirty ring tracking is exclusive to dirty log tracking */
1763 if (kvm->dirty_ring_size)
1766 as_id = log->slot >> 16;
1767 id = (u16)log->slot;
1768 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1771 if (log->first_page & 63)
1774 slots = __kvm_memslots(kvm, as_id);
1775 memslot = id_to_memslot(slots, id);
1776 if (!memslot || !memslot->dirty_bitmap)
1779 dirty_bitmap = memslot->dirty_bitmap;
1781 n = ALIGN(log->num_pages, BITS_PER_LONG) / 8;
1783 if (log->first_page > memslot->npages ||
1784 log->num_pages > memslot->npages - log->first_page ||
1785 (log->num_pages < memslot->npages - log->first_page && (log->num_pages & 63)))
1788 kvm_arch_sync_dirty_log(kvm, memslot);
1791 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
1792 if (copy_from_user(dirty_bitmap_buffer, log->dirty_bitmap, n))
1796 for (offset = log->first_page, i = offset / BITS_PER_LONG,
1797 n = DIV_ROUND_UP(log->num_pages, BITS_PER_LONG); n--;
1798 i++, offset += BITS_PER_LONG) {
1799 unsigned long mask = *dirty_bitmap_buffer++;
1800 atomic_long_t *p = (atomic_long_t *) &dirty_bitmap[i];
1804 mask &= atomic_long_fetch_andnot(mask, p);
1807 * mask contains the bits that really have been cleared. This
1808 * never includes any bits beyond the length of the memslot (if
1809 * the length is not aligned to 64 pages), therefore it is not
1810 * a problem if userspace sets them in log->dirty_bitmap.
1814 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1818 KVM_MMU_UNLOCK(kvm);
1821 kvm_arch_flush_remote_tlbs_memslot(kvm, memslot);
1826 static int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm,
1827 struct kvm_clear_dirty_log *log)
1831 mutex_lock(&kvm->slots_lock);
1833 r = kvm_clear_dirty_log_protect(kvm, log);
1835 mutex_unlock(&kvm->slots_lock);
1838 #endif /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
1840 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1842 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1844 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1846 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
1848 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn);
1850 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_memslot);
1852 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1854 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1856 return kvm_is_visible_memslot(memslot);
1858 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1860 bool kvm_vcpu_is_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
1862 struct kvm_memory_slot *memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1864 return kvm_is_visible_memslot(memslot);
1866 EXPORT_SYMBOL_GPL(kvm_vcpu_is_visible_gfn);
1868 unsigned long kvm_host_page_size(struct kvm_vcpu *vcpu, gfn_t gfn)
1870 struct vm_area_struct *vma;
1871 unsigned long addr, size;
1875 addr = kvm_vcpu_gfn_to_hva_prot(vcpu, gfn, NULL);
1876 if (kvm_is_error_hva(addr))
1879 mmap_read_lock(current->mm);
1880 vma = find_vma(current->mm, addr);
1884 size = vma_kernel_pagesize(vma);
1887 mmap_read_unlock(current->mm);
1892 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1894 return slot->flags & KVM_MEM_READONLY;
1897 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1898 gfn_t *nr_pages, bool write)
1900 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1901 return KVM_HVA_ERR_BAD;
1903 if (memslot_is_readonly(slot) && write)
1904 return KVM_HVA_ERR_RO_BAD;
1907 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1909 return __gfn_to_hva_memslot(slot, gfn);
1912 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1915 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1918 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1921 return gfn_to_hva_many(slot, gfn, NULL);
1923 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1925 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1927 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1929 EXPORT_SYMBOL_GPL(gfn_to_hva);
1931 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
1933 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
1935 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
1938 * Return the hva of a @gfn and the R/W attribute if possible.
1940 * @slot: the kvm_memory_slot which contains @gfn
1941 * @gfn: the gfn to be translated
1942 * @writable: used to return the read/write attribute of the @slot if the hva
1943 * is valid and @writable is not NULL
1945 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1946 gfn_t gfn, bool *writable)
1948 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1950 if (!kvm_is_error_hva(hva) && writable)
1951 *writable = !memslot_is_readonly(slot);
1956 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1958 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1960 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1963 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
1965 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1967 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1970 static inline int check_user_page_hwpoison(unsigned long addr)
1972 int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
1974 rc = get_user_pages(addr, 1, flags, NULL, NULL);
1975 return rc == -EHWPOISON;
1979 * The fast path to get the writable pfn which will be stored in @pfn,
1980 * true indicates success, otherwise false is returned. It's also the
1981 * only part that runs if we can in atomic context.
1983 static bool hva_to_pfn_fast(unsigned long addr, bool write_fault,
1984 bool *writable, kvm_pfn_t *pfn)
1986 struct page *page[1];
1989 * Fast pin a writable pfn only if it is a write fault request
1990 * or the caller allows to map a writable pfn for a read fault
1993 if (!(write_fault || writable))
1996 if (get_user_page_fast_only(addr, FOLL_WRITE, page)) {
1997 *pfn = page_to_pfn(page[0]);
2008 * The slow path to get the pfn of the specified host virtual address,
2009 * 1 indicates success, -errno is returned if error is detected.
2011 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
2012 bool *writable, kvm_pfn_t *pfn)
2014 unsigned int flags = FOLL_HWPOISON;
2021 *writable = write_fault;
2024 flags |= FOLL_WRITE;
2026 flags |= FOLL_NOWAIT;
2028 npages = get_user_pages_unlocked(addr, 1, &page, flags);
2032 /* map read fault as writable if possible */
2033 if (unlikely(!write_fault) && writable) {
2036 if (get_user_page_fast_only(addr, FOLL_WRITE, &wpage)) {
2042 *pfn = page_to_pfn(page);
2046 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
2048 if (unlikely(!(vma->vm_flags & VM_READ)))
2051 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
2057 static int hva_to_pfn_remapped(struct vm_area_struct *vma,
2058 unsigned long addr, bool *async,
2059 bool write_fault, bool *writable,
2067 r = follow_pte(vma->vm_mm, addr, &ptep, &ptl);
2070 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
2071 * not call the fault handler, so do it here.
2073 bool unlocked = false;
2074 r = fixup_user_fault(current->mm, addr,
2075 (write_fault ? FAULT_FLAG_WRITE : 0),
2082 r = follow_pte(vma->vm_mm, addr, &ptep, &ptl);
2087 if (write_fault && !pte_write(*ptep)) {
2088 pfn = KVM_PFN_ERR_RO_FAULT;
2093 *writable = pte_write(*ptep);
2094 pfn = pte_pfn(*ptep);
2097 * Get a reference here because callers of *hva_to_pfn* and
2098 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
2099 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
2100 * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will
2101 * simply do nothing for reserved pfns.
2103 * Whoever called remap_pfn_range is also going to call e.g.
2104 * unmap_mapping_range before the underlying pages are freed,
2105 * causing a call to our MMU notifier.
2110 pte_unmap_unlock(ptep, ptl);
2116 * Pin guest page in memory and return its pfn.
2117 * @addr: host virtual address which maps memory to the guest
2118 * @atomic: whether this function can sleep
2119 * @async: whether this function need to wait IO complete if the
2120 * host page is not in the memory
2121 * @write_fault: whether we should get a writable host page
2122 * @writable: whether it allows to map a writable host page for !@write_fault
2124 * The function will map a writable host page for these two cases:
2125 * 1): @write_fault = true
2126 * 2): @write_fault = false && @writable, @writable will tell the caller
2127 * whether the mapping is writable.
2129 static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
2130 bool write_fault, bool *writable)
2132 struct vm_area_struct *vma;
2136 /* we can do it either atomically or asynchronously, not both */
2137 BUG_ON(atomic && async);
2139 if (hva_to_pfn_fast(addr, write_fault, writable, &pfn))
2143 return KVM_PFN_ERR_FAULT;
2145 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
2149 mmap_read_lock(current->mm);
2150 if (npages == -EHWPOISON ||
2151 (!async && check_user_page_hwpoison(addr))) {
2152 pfn = KVM_PFN_ERR_HWPOISON;
2157 vma = find_vma_intersection(current->mm, addr, addr + 1);
2160 pfn = KVM_PFN_ERR_FAULT;
2161 else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
2162 r = hva_to_pfn_remapped(vma, addr, async, write_fault, writable, &pfn);
2166 pfn = KVM_PFN_ERR_FAULT;
2168 if (async && vma_is_valid(vma, write_fault))
2170 pfn = KVM_PFN_ERR_FAULT;
2173 mmap_read_unlock(current->mm);
2177 kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn,
2178 bool atomic, bool *async, bool write_fault,
2179 bool *writable, hva_t *hva)
2181 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
2186 if (addr == KVM_HVA_ERR_RO_BAD) {
2189 return KVM_PFN_ERR_RO_FAULT;
2192 if (kvm_is_error_hva(addr)) {
2195 return KVM_PFN_NOSLOT;
2198 /* Do not map writable pfn in the readonly memslot. */
2199 if (writable && memslot_is_readonly(slot)) {
2204 return hva_to_pfn(addr, atomic, async, write_fault,
2207 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
2209 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
2212 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
2213 write_fault, writable, NULL);
2215 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
2217 kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
2219 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL, NULL);
2221 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
2223 kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
2225 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL, NULL);
2227 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
2229 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
2231 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
2233 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
2235 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
2237 return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
2239 EXPORT_SYMBOL_GPL(gfn_to_pfn);
2241 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
2243 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
2245 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
2247 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
2248 struct page **pages, int nr_pages)
2253 addr = gfn_to_hva_many(slot, gfn, &entry);
2254 if (kvm_is_error_hva(addr))
2257 if (entry < nr_pages)
2260 return get_user_pages_fast_only(addr, nr_pages, FOLL_WRITE, pages);
2262 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
2264 static struct page *kvm_pfn_to_page(kvm_pfn_t pfn)
2266 if (is_error_noslot_pfn(pfn))
2267 return KVM_ERR_PTR_BAD_PAGE;
2269 if (kvm_is_reserved_pfn(pfn)) {
2271 return KVM_ERR_PTR_BAD_PAGE;
2274 return pfn_to_page(pfn);
2277 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
2281 pfn = gfn_to_pfn(kvm, gfn);
2283 return kvm_pfn_to_page(pfn);
2285 EXPORT_SYMBOL_GPL(gfn_to_page);
2287 void kvm_release_pfn(kvm_pfn_t pfn, bool dirty, struct gfn_to_pfn_cache *cache)
2293 cache->pfn = cache->gfn = 0;
2296 kvm_release_pfn_dirty(pfn);
2298 kvm_release_pfn_clean(pfn);
2301 static void kvm_cache_gfn_to_pfn(struct kvm_memory_slot *slot, gfn_t gfn,
2302 struct gfn_to_pfn_cache *cache, u64 gen)
2304 kvm_release_pfn(cache->pfn, cache->dirty, cache);
2306 cache->pfn = gfn_to_pfn_memslot(slot, gfn);
2308 cache->dirty = false;
2309 cache->generation = gen;
2312 static int __kvm_map_gfn(struct kvm_memslots *slots, gfn_t gfn,
2313 struct kvm_host_map *map,
2314 struct gfn_to_pfn_cache *cache,
2319 struct page *page = KVM_UNMAPPED_PAGE;
2320 struct kvm_memory_slot *slot = __gfn_to_memslot(slots, gfn);
2321 u64 gen = slots->generation;
2327 if (!cache->pfn || cache->gfn != gfn ||
2328 cache->generation != gen) {
2331 kvm_cache_gfn_to_pfn(slot, gfn, cache, gen);
2337 pfn = gfn_to_pfn_memslot(slot, gfn);
2339 if (is_error_noslot_pfn(pfn))
2342 if (pfn_valid(pfn)) {
2343 page = pfn_to_page(pfn);
2345 hva = kmap_atomic(page);
2348 #ifdef CONFIG_HAS_IOMEM
2349 } else if (!atomic) {
2350 hva = memremap(pfn_to_hpa(pfn), PAGE_SIZE, MEMREMAP_WB);
2367 int kvm_map_gfn(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map,
2368 struct gfn_to_pfn_cache *cache, bool atomic)
2370 return __kvm_map_gfn(kvm_memslots(vcpu->kvm), gfn, map,
2373 EXPORT_SYMBOL_GPL(kvm_map_gfn);
2375 int kvm_vcpu_map(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map)
2377 return __kvm_map_gfn(kvm_vcpu_memslots(vcpu), gfn, map,
2380 EXPORT_SYMBOL_GPL(kvm_vcpu_map);
2382 static void __kvm_unmap_gfn(struct kvm *kvm,
2383 struct kvm_memory_slot *memslot,
2384 struct kvm_host_map *map,
2385 struct gfn_to_pfn_cache *cache,
2386 bool dirty, bool atomic)
2394 if (map->page != KVM_UNMAPPED_PAGE) {
2396 kunmap_atomic(map->hva);
2400 #ifdef CONFIG_HAS_IOMEM
2404 WARN_ONCE(1, "Unexpected unmapping in atomic context");
2408 mark_page_dirty_in_slot(kvm, memslot, map->gfn);
2411 cache->dirty |= dirty;
2413 kvm_release_pfn(map->pfn, dirty, NULL);
2419 int kvm_unmap_gfn(struct kvm_vcpu *vcpu, struct kvm_host_map *map,
2420 struct gfn_to_pfn_cache *cache, bool dirty, bool atomic)
2422 __kvm_unmap_gfn(vcpu->kvm, gfn_to_memslot(vcpu->kvm, map->gfn), map,
2423 cache, dirty, atomic);
2426 EXPORT_SYMBOL_GPL(kvm_unmap_gfn);
2428 void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map, bool dirty)
2430 __kvm_unmap_gfn(vcpu->kvm, kvm_vcpu_gfn_to_memslot(vcpu, map->gfn),
2431 map, NULL, dirty, false);
2433 EXPORT_SYMBOL_GPL(kvm_vcpu_unmap);
2435 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
2439 pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
2441 return kvm_pfn_to_page(pfn);
2443 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
2445 void kvm_release_page_clean(struct page *page)
2447 WARN_ON(is_error_page(page));
2449 kvm_release_pfn_clean(page_to_pfn(page));
2451 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
2453 void kvm_release_pfn_clean(kvm_pfn_t pfn)
2455 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
2456 put_page(pfn_to_page(pfn));
2458 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
2460 void kvm_release_page_dirty(struct page *page)
2462 WARN_ON(is_error_page(page));
2464 kvm_release_pfn_dirty(page_to_pfn(page));
2466 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
2468 void kvm_release_pfn_dirty(kvm_pfn_t pfn)
2470 kvm_set_pfn_dirty(pfn);
2471 kvm_release_pfn_clean(pfn);
2473 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
2475 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
2477 if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn))
2478 SetPageDirty(pfn_to_page(pfn));
2480 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
2482 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
2484 if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn))
2485 mark_page_accessed(pfn_to_page(pfn));
2487 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
2489 void kvm_get_pfn(kvm_pfn_t pfn)
2491 if (!kvm_is_reserved_pfn(pfn))
2492 get_page(pfn_to_page(pfn));
2494 EXPORT_SYMBOL_GPL(kvm_get_pfn);
2496 static int next_segment(unsigned long len, int offset)
2498 if (len > PAGE_SIZE - offset)
2499 return PAGE_SIZE - offset;
2504 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
2505 void *data, int offset, int len)
2510 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
2511 if (kvm_is_error_hva(addr))
2513 r = __copy_from_user(data, (void __user *)addr + offset, len);
2519 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
2522 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2524 return __kvm_read_guest_page(slot, gfn, data, offset, len);
2526 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
2528 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
2529 int offset, int len)
2531 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2533 return __kvm_read_guest_page(slot, gfn, data, offset, len);
2535 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
2537 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
2539 gfn_t gfn = gpa >> PAGE_SHIFT;
2541 int offset = offset_in_page(gpa);
2544 while ((seg = next_segment(len, offset)) != 0) {
2545 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
2555 EXPORT_SYMBOL_GPL(kvm_read_guest);
2557 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
2559 gfn_t gfn = gpa >> PAGE_SHIFT;
2561 int offset = offset_in_page(gpa);
2564 while ((seg = next_segment(len, offset)) != 0) {
2565 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
2575 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
2577 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
2578 void *data, int offset, unsigned long len)
2583 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
2584 if (kvm_is_error_hva(addr))
2586 pagefault_disable();
2587 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
2594 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
2595 void *data, unsigned long len)
2597 gfn_t gfn = gpa >> PAGE_SHIFT;
2598 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2599 int offset = offset_in_page(gpa);
2601 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
2603 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
2605 static int __kvm_write_guest_page(struct kvm *kvm,
2606 struct kvm_memory_slot *memslot, gfn_t gfn,
2607 const void *data, int offset, int len)
2612 addr = gfn_to_hva_memslot(memslot, gfn);
2613 if (kvm_is_error_hva(addr))
2615 r = __copy_to_user((void __user *)addr + offset, data, len);
2618 mark_page_dirty_in_slot(kvm, memslot, gfn);
2622 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
2623 const void *data, int offset, int len)
2625 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2627 return __kvm_write_guest_page(kvm, slot, gfn, data, offset, len);
2629 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
2631 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
2632 const void *data, int offset, int len)
2634 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2636 return __kvm_write_guest_page(vcpu->kvm, slot, gfn, data, offset, len);
2638 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
2640 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
2643 gfn_t gfn = gpa >> PAGE_SHIFT;
2645 int offset = offset_in_page(gpa);
2648 while ((seg = next_segment(len, offset)) != 0) {
2649 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
2659 EXPORT_SYMBOL_GPL(kvm_write_guest);
2661 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
2664 gfn_t gfn = gpa >> PAGE_SHIFT;
2666 int offset = offset_in_page(gpa);
2669 while ((seg = next_segment(len, offset)) != 0) {
2670 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
2680 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
2682 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
2683 struct gfn_to_hva_cache *ghc,
2684 gpa_t gpa, unsigned long len)
2686 int offset = offset_in_page(gpa);
2687 gfn_t start_gfn = gpa >> PAGE_SHIFT;
2688 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
2689 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
2690 gfn_t nr_pages_avail;
2692 /* Update ghc->generation before performing any error checks. */
2693 ghc->generation = slots->generation;
2695 if (start_gfn > end_gfn) {
2696 ghc->hva = KVM_HVA_ERR_BAD;
2701 * If the requested region crosses two memslots, we still
2702 * verify that the entire region is valid here.
2704 for ( ; start_gfn <= end_gfn; start_gfn += nr_pages_avail) {
2705 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
2706 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
2708 if (kvm_is_error_hva(ghc->hva))
2712 /* Use the slow path for cross page reads and writes. */
2713 if (nr_pages_needed == 1)
2716 ghc->memslot = NULL;
2723 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2724 gpa_t gpa, unsigned long len)
2726 struct kvm_memslots *slots = kvm_memslots(kvm);
2727 return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len);
2729 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
2731 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2732 void *data, unsigned int offset,
2735 struct kvm_memslots *slots = kvm_memslots(kvm);
2737 gpa_t gpa = ghc->gpa + offset;
2739 BUG_ON(len + offset > ghc->len);
2741 if (slots->generation != ghc->generation) {
2742 if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len))
2746 if (kvm_is_error_hva(ghc->hva))
2749 if (unlikely(!ghc->memslot))
2750 return kvm_write_guest(kvm, gpa, data, len);
2752 r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
2755 mark_page_dirty_in_slot(kvm, ghc->memslot, gpa >> PAGE_SHIFT);
2759 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached);
2761 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2762 void *data, unsigned long len)
2764 return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len);
2766 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
2768 int kvm_read_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2769 void *data, unsigned int offset,
2772 struct kvm_memslots *slots = kvm_memslots(kvm);
2774 gpa_t gpa = ghc->gpa + offset;
2776 BUG_ON(len + offset > ghc->len);
2778 if (slots->generation != ghc->generation) {
2779 if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len))
2783 if (kvm_is_error_hva(ghc->hva))
2786 if (unlikely(!ghc->memslot))
2787 return kvm_read_guest(kvm, gpa, data, len);
2789 r = __copy_from_user(data, (void __user *)ghc->hva + offset, len);
2795 EXPORT_SYMBOL_GPL(kvm_read_guest_offset_cached);
2797 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2798 void *data, unsigned long len)
2800 return kvm_read_guest_offset_cached(kvm, ghc, data, 0, len);
2802 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
2804 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
2806 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
2807 gfn_t gfn = gpa >> PAGE_SHIFT;
2809 int offset = offset_in_page(gpa);
2812 while ((seg = next_segment(len, offset)) != 0) {
2813 ret = kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
2822 EXPORT_SYMBOL_GPL(kvm_clear_guest);
2824 void mark_page_dirty_in_slot(struct kvm *kvm,
2825 struct kvm_memory_slot *memslot,
2828 if (memslot && kvm_slot_dirty_track_enabled(memslot)) {
2829 unsigned long rel_gfn = gfn - memslot->base_gfn;
2830 u32 slot = (memslot->as_id << 16) | memslot->id;
2832 if (kvm->dirty_ring_size)
2833 kvm_dirty_ring_push(kvm_dirty_ring_get(kvm),
2836 set_bit_le(rel_gfn, memslot->dirty_bitmap);
2839 EXPORT_SYMBOL_GPL(mark_page_dirty_in_slot);
2841 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
2843 struct kvm_memory_slot *memslot;
2845 memslot = gfn_to_memslot(kvm, gfn);
2846 mark_page_dirty_in_slot(kvm, memslot, gfn);
2848 EXPORT_SYMBOL_GPL(mark_page_dirty);
2850 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
2852 struct kvm_memory_slot *memslot;
2854 memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2855 mark_page_dirty_in_slot(vcpu->kvm, memslot, gfn);
2857 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
2859 void kvm_sigset_activate(struct kvm_vcpu *vcpu)
2861 if (!vcpu->sigset_active)
2865 * This does a lockless modification of ->real_blocked, which is fine
2866 * because, only current can change ->real_blocked and all readers of
2867 * ->real_blocked don't care as long ->real_blocked is always a subset
2870 sigprocmask(SIG_SETMASK, &vcpu->sigset, ¤t->real_blocked);
2873 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu)
2875 if (!vcpu->sigset_active)
2878 sigprocmask(SIG_SETMASK, ¤t->real_blocked, NULL);
2879 sigemptyset(¤t->real_blocked);
2882 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
2884 unsigned int old, val, grow, grow_start;
2886 old = val = vcpu->halt_poll_ns;
2887 grow_start = READ_ONCE(halt_poll_ns_grow_start);
2888 grow = READ_ONCE(halt_poll_ns_grow);
2893 if (val < grow_start)
2896 if (val > vcpu->kvm->max_halt_poll_ns)
2897 val = vcpu->kvm->max_halt_poll_ns;
2899 vcpu->halt_poll_ns = val;
2901 trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
2904 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
2906 unsigned int old, val, shrink;
2908 old = val = vcpu->halt_poll_ns;
2909 shrink = READ_ONCE(halt_poll_ns_shrink);
2915 vcpu->halt_poll_ns = val;
2916 trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
2919 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
2922 int idx = srcu_read_lock(&vcpu->kvm->srcu);
2924 if (kvm_arch_vcpu_runnable(vcpu)) {
2925 kvm_make_request(KVM_REQ_UNHALT, vcpu);
2928 if (kvm_cpu_has_pending_timer(vcpu))
2930 if (signal_pending(current))
2935 srcu_read_unlock(&vcpu->kvm->srcu, idx);
2940 update_halt_poll_stats(struct kvm_vcpu *vcpu, u64 poll_ns, bool waited)
2943 vcpu->stat.halt_poll_fail_ns += poll_ns;
2945 vcpu->stat.halt_poll_success_ns += poll_ns;
2949 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2951 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
2953 ktime_t start, cur, poll_end;
2954 bool waited = false;
2957 kvm_arch_vcpu_blocking(vcpu);
2959 start = cur = poll_end = ktime_get();
2960 if (vcpu->halt_poll_ns && !kvm_arch_no_poll(vcpu)) {
2961 ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns);
2963 ++vcpu->stat.halt_attempted_poll;
2966 * This sets KVM_REQ_UNHALT if an interrupt
2969 if (kvm_vcpu_check_block(vcpu) < 0) {
2970 ++vcpu->stat.halt_successful_poll;
2971 if (!vcpu_valid_wakeup(vcpu))
2972 ++vcpu->stat.halt_poll_invalid;
2975 poll_end = cur = ktime_get();
2976 } while (single_task_running() && !need_resched() &&
2977 ktime_before(cur, stop));
2980 prepare_to_rcuwait(&vcpu->wait);
2982 set_current_state(TASK_INTERRUPTIBLE);
2984 if (kvm_vcpu_check_block(vcpu) < 0)
2990 finish_rcuwait(&vcpu->wait);
2993 kvm_arch_vcpu_unblocking(vcpu);
2994 block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
2996 update_halt_poll_stats(
2997 vcpu, ktime_to_ns(ktime_sub(poll_end, start)), waited);
2999 if (!kvm_arch_no_poll(vcpu)) {
3000 if (!vcpu_valid_wakeup(vcpu)) {
3001 shrink_halt_poll_ns(vcpu);
3002 } else if (vcpu->kvm->max_halt_poll_ns) {
3003 if (block_ns <= vcpu->halt_poll_ns)
3005 /* we had a long block, shrink polling */
3006 else if (vcpu->halt_poll_ns &&
3007 block_ns > vcpu->kvm->max_halt_poll_ns)
3008 shrink_halt_poll_ns(vcpu);
3009 /* we had a short halt and our poll time is too small */
3010 else if (vcpu->halt_poll_ns < vcpu->kvm->max_halt_poll_ns &&
3011 block_ns < vcpu->kvm->max_halt_poll_ns)
3012 grow_halt_poll_ns(vcpu);
3014 vcpu->halt_poll_ns = 0;
3018 trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu));
3019 kvm_arch_vcpu_block_finish(vcpu);
3021 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
3023 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
3025 struct rcuwait *waitp;
3027 waitp = kvm_arch_vcpu_get_wait(vcpu);
3028 if (rcuwait_wake_up(waitp)) {
3029 WRITE_ONCE(vcpu->ready, true);
3030 ++vcpu->stat.halt_wakeup;
3036 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
3040 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
3042 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
3045 int cpu = vcpu->cpu;
3047 if (kvm_vcpu_wake_up(vcpu))
3051 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
3052 if (kvm_arch_vcpu_should_kick(vcpu))
3053 smp_send_reschedule(cpu);
3056 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
3057 #endif /* !CONFIG_S390 */
3059 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
3062 struct task_struct *task = NULL;
3066 pid = rcu_dereference(target->pid);
3068 task = get_pid_task(pid, PIDTYPE_PID);
3072 ret = yield_to(task, 1);
3073 put_task_struct(task);
3077 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
3080 * Helper that checks whether a VCPU is eligible for directed yield.
3081 * Most eligible candidate to yield is decided by following heuristics:
3083 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
3084 * (preempted lock holder), indicated by @in_spin_loop.
3085 * Set at the beginning and cleared at the end of interception/PLE handler.
3087 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
3088 * chance last time (mostly it has become eligible now since we have probably
3089 * yielded to lockholder in last iteration. This is done by toggling
3090 * @dy_eligible each time a VCPU checked for eligibility.)
3092 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
3093 * to preempted lock-holder could result in wrong VCPU selection and CPU
3094 * burning. Giving priority for a potential lock-holder increases lock
3097 * Since algorithm is based on heuristics, accessing another VCPU data without
3098 * locking does not harm. It may result in trying to yield to same VCPU, fail
3099 * and continue with next VCPU and so on.
3101 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
3103 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
3106 eligible = !vcpu->spin_loop.in_spin_loop ||
3107 vcpu->spin_loop.dy_eligible;
3109 if (vcpu->spin_loop.in_spin_loop)
3110 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
3119 * Unlike kvm_arch_vcpu_runnable, this function is called outside
3120 * a vcpu_load/vcpu_put pair. However, for most architectures
3121 * kvm_arch_vcpu_runnable does not require vcpu_load.
3123 bool __weak kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
3125 return kvm_arch_vcpu_runnable(vcpu);
3128 static bool vcpu_dy_runnable(struct kvm_vcpu *vcpu)
3130 if (kvm_arch_dy_runnable(vcpu))
3133 #ifdef CONFIG_KVM_ASYNC_PF
3134 if (!list_empty_careful(&vcpu->async_pf.done))
3141 bool __weak kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu)
3146 void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode)
3148 struct kvm *kvm = me->kvm;
3149 struct kvm_vcpu *vcpu;
3150 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
3156 kvm_vcpu_set_in_spin_loop(me, true);
3158 * We boost the priority of a VCPU that is runnable but not
3159 * currently running, because it got preempted by something
3160 * else and called schedule in __vcpu_run. Hopefully that
3161 * VCPU is holding the lock that we need and will release it.
3162 * We approximate round-robin by starting at the last boosted VCPU.
3164 for (pass = 0; pass < 2 && !yielded && try; pass++) {
3165 kvm_for_each_vcpu(i, vcpu, kvm) {
3166 if (!pass && i <= last_boosted_vcpu) {
3167 i = last_boosted_vcpu;
3169 } else if (pass && i > last_boosted_vcpu)
3171 if (!READ_ONCE(vcpu->ready))
3175 if (rcuwait_active(&vcpu->wait) &&
3176 !vcpu_dy_runnable(vcpu))
3178 if (READ_ONCE(vcpu->preempted) && yield_to_kernel_mode &&
3179 !kvm_arch_dy_has_pending_interrupt(vcpu) &&
3180 !kvm_arch_vcpu_in_kernel(vcpu))
3182 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
3185 yielded = kvm_vcpu_yield_to(vcpu);
3187 kvm->last_boosted_vcpu = i;
3189 } else if (yielded < 0) {
3196 kvm_vcpu_set_in_spin_loop(me, false);
3198 /* Ensure vcpu is not eligible during next spinloop */
3199 kvm_vcpu_set_dy_eligible(me, false);
3201 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
3203 static bool kvm_page_in_dirty_ring(struct kvm *kvm, unsigned long pgoff)
3205 #if KVM_DIRTY_LOG_PAGE_OFFSET > 0
3206 return (pgoff >= KVM_DIRTY_LOG_PAGE_OFFSET) &&
3207 (pgoff < KVM_DIRTY_LOG_PAGE_OFFSET +
3208 kvm->dirty_ring_size / PAGE_SIZE);
3214 static vm_fault_t kvm_vcpu_fault(struct vm_fault *vmf)
3216 struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data;
3219 if (vmf->pgoff == 0)
3220 page = virt_to_page(vcpu->run);
3222 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
3223 page = virt_to_page(vcpu->arch.pio_data);
3225 #ifdef CONFIG_KVM_MMIO
3226 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
3227 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
3229 else if (kvm_page_in_dirty_ring(vcpu->kvm, vmf->pgoff))
3230 page = kvm_dirty_ring_get_page(
3232 vmf->pgoff - KVM_DIRTY_LOG_PAGE_OFFSET);
3234 return kvm_arch_vcpu_fault(vcpu, vmf);
3240 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
3241 .fault = kvm_vcpu_fault,
3244 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
3246 struct kvm_vcpu *vcpu = file->private_data;
3247 unsigned long pages = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
3249 if ((kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff) ||
3250 kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff + pages - 1)) &&
3251 ((vma->vm_flags & VM_EXEC) || !(vma->vm_flags & VM_SHARED)))
3254 vma->vm_ops = &kvm_vcpu_vm_ops;
3258 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
3260 struct kvm_vcpu *vcpu = filp->private_data;
3262 kvm_put_kvm(vcpu->kvm);
3266 static struct file_operations kvm_vcpu_fops = {
3267 .release = kvm_vcpu_release,
3268 .unlocked_ioctl = kvm_vcpu_ioctl,
3269 .mmap = kvm_vcpu_mmap,
3270 .llseek = noop_llseek,
3271 KVM_COMPAT(kvm_vcpu_compat_ioctl),
3275 * Allocates an inode for the vcpu.
3277 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
3279 char name[8 + 1 + ITOA_MAX_LEN + 1];
3281 snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id);
3282 return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
3285 static void kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
3287 #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS
3288 struct dentry *debugfs_dentry;
3289 char dir_name[ITOA_MAX_LEN * 2];
3291 if (!debugfs_initialized())
3294 snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
3295 debugfs_dentry = debugfs_create_dir(dir_name,
3296 vcpu->kvm->debugfs_dentry);
3298 kvm_arch_create_vcpu_debugfs(vcpu, debugfs_dentry);
3303 * Creates some virtual cpus. Good luck creating more than one.
3305 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
3308 struct kvm_vcpu *vcpu;
3311 if (id >= KVM_MAX_VCPU_ID)
3314 mutex_lock(&kvm->lock);
3315 if (kvm->created_vcpus == KVM_MAX_VCPUS) {
3316 mutex_unlock(&kvm->lock);
3320 kvm->created_vcpus++;
3321 mutex_unlock(&kvm->lock);
3323 r = kvm_arch_vcpu_precreate(kvm, id);
3325 goto vcpu_decrement;
3327 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL_ACCOUNT);
3330 goto vcpu_decrement;
3333 BUILD_BUG_ON(sizeof(struct kvm_run) > PAGE_SIZE);
3334 page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
3339 vcpu->run = page_address(page);
3341 kvm_vcpu_init(vcpu, kvm, id);
3343 r = kvm_arch_vcpu_create(vcpu);
3345 goto vcpu_free_run_page;
3347 if (kvm->dirty_ring_size) {
3348 r = kvm_dirty_ring_alloc(&vcpu->dirty_ring,
3349 id, kvm->dirty_ring_size);
3351 goto arch_vcpu_destroy;
3354 mutex_lock(&kvm->lock);
3355 if (kvm_get_vcpu_by_id(kvm, id)) {
3357 goto unlock_vcpu_destroy;
3360 vcpu->vcpu_idx = atomic_read(&kvm->online_vcpus);
3361 BUG_ON(kvm->vcpus[vcpu->vcpu_idx]);
3363 /* Now it's all set up, let userspace reach it */
3365 r = create_vcpu_fd(vcpu);
3367 kvm_put_kvm_no_destroy(kvm);
3368 goto unlock_vcpu_destroy;
3371 kvm->vcpus[vcpu->vcpu_idx] = vcpu;
3374 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
3375 * before kvm->online_vcpu's incremented value.
3378 atomic_inc(&kvm->online_vcpus);
3380 mutex_unlock(&kvm->lock);
3381 kvm_arch_vcpu_postcreate(vcpu);
3382 kvm_create_vcpu_debugfs(vcpu);
3385 unlock_vcpu_destroy:
3386 mutex_unlock(&kvm->lock);
3387 kvm_dirty_ring_free(&vcpu->dirty_ring);
3389 kvm_arch_vcpu_destroy(vcpu);
3391 free_page((unsigned long)vcpu->run);
3393 kmem_cache_free(kvm_vcpu_cache, vcpu);
3395 mutex_lock(&kvm->lock);
3396 kvm->created_vcpus--;
3397 mutex_unlock(&kvm->lock);
3401 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
3404 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
3405 vcpu->sigset_active = 1;
3406 vcpu->sigset = *sigset;
3408 vcpu->sigset_active = 0;
3412 static long kvm_vcpu_ioctl(struct file *filp,
3413 unsigned int ioctl, unsigned long arg)
3415 struct kvm_vcpu *vcpu = filp->private_data;
3416 void __user *argp = (void __user *)arg;
3418 struct kvm_fpu *fpu = NULL;
3419 struct kvm_sregs *kvm_sregs = NULL;
3421 if (vcpu->kvm->mm != current->mm)
3424 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
3428 * Some architectures have vcpu ioctls that are asynchronous to vcpu
3429 * execution; mutex_lock() would break them.
3431 r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg);
3432 if (r != -ENOIOCTLCMD)
3435 if (mutex_lock_killable(&vcpu->mutex))
3443 oldpid = rcu_access_pointer(vcpu->pid);
3444 if (unlikely(oldpid != task_pid(current))) {
3445 /* The thread running this VCPU changed. */
3448 r = kvm_arch_vcpu_run_pid_change(vcpu);
3452 newpid = get_task_pid(current, PIDTYPE_PID);
3453 rcu_assign_pointer(vcpu->pid, newpid);
3458 r = kvm_arch_vcpu_ioctl_run(vcpu);
3459 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
3462 case KVM_GET_REGS: {
3463 struct kvm_regs *kvm_regs;
3466 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL_ACCOUNT);
3469 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
3473 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
3480 case KVM_SET_REGS: {
3481 struct kvm_regs *kvm_regs;
3483 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
3484 if (IS_ERR(kvm_regs)) {
3485 r = PTR_ERR(kvm_regs);
3488 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
3492 case KVM_GET_SREGS: {
3493 kvm_sregs = kzalloc(sizeof(struct kvm_sregs),
3494 GFP_KERNEL_ACCOUNT);
3498 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
3502 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
3507 case KVM_SET_SREGS: {
3508 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
3509 if (IS_ERR(kvm_sregs)) {
3510 r = PTR_ERR(kvm_sregs);
3514 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
3517 case KVM_GET_MP_STATE: {
3518 struct kvm_mp_state mp_state;
3520 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
3524 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
3529 case KVM_SET_MP_STATE: {
3530 struct kvm_mp_state mp_state;
3533 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
3535 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
3538 case KVM_TRANSLATE: {
3539 struct kvm_translation tr;
3542 if (copy_from_user(&tr, argp, sizeof(tr)))
3544 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
3548 if (copy_to_user(argp, &tr, sizeof(tr)))
3553 case KVM_SET_GUEST_DEBUG: {
3554 struct kvm_guest_debug dbg;
3557 if (copy_from_user(&dbg, argp, sizeof(dbg)))
3559 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
3562 case KVM_SET_SIGNAL_MASK: {
3563 struct kvm_signal_mask __user *sigmask_arg = argp;
3564 struct kvm_signal_mask kvm_sigmask;
3565 sigset_t sigset, *p;
3570 if (copy_from_user(&kvm_sigmask, argp,
3571 sizeof(kvm_sigmask)))
3574 if (kvm_sigmask.len != sizeof(sigset))
3577 if (copy_from_user(&sigset, sigmask_arg->sigset,
3582 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
3586 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL_ACCOUNT);
3590 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
3594 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
3600 fpu = memdup_user(argp, sizeof(*fpu));
3606 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
3610 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
3613 mutex_unlock(&vcpu->mutex);
3619 #ifdef CONFIG_KVM_COMPAT
3620 static long kvm_vcpu_compat_ioctl(struct file *filp,
3621 unsigned int ioctl, unsigned long arg)
3623 struct kvm_vcpu *vcpu = filp->private_data;
3624 void __user *argp = compat_ptr(arg);
3627 if (vcpu->kvm->mm != current->mm)
3631 case KVM_SET_SIGNAL_MASK: {
3632 struct kvm_signal_mask __user *sigmask_arg = argp;
3633 struct kvm_signal_mask kvm_sigmask;
3638 if (copy_from_user(&kvm_sigmask, argp,
3639 sizeof(kvm_sigmask)))
3642 if (kvm_sigmask.len != sizeof(compat_sigset_t))
3645 if (get_compat_sigset(&sigset,
3646 (compat_sigset_t __user *)sigmask_arg->sigset))
3648 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
3650 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
3654 r = kvm_vcpu_ioctl(filp, ioctl, arg);
3662 static int kvm_device_mmap(struct file *filp, struct vm_area_struct *vma)
3664 struct kvm_device *dev = filp->private_data;
3667 return dev->ops->mmap(dev, vma);
3672 static int kvm_device_ioctl_attr(struct kvm_device *dev,
3673 int (*accessor)(struct kvm_device *dev,
3674 struct kvm_device_attr *attr),
3677 struct kvm_device_attr attr;
3682 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
3685 return accessor(dev, &attr);
3688 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
3691 struct kvm_device *dev = filp->private_data;
3693 if (dev->kvm->mm != current->mm)
3697 case KVM_SET_DEVICE_ATTR:
3698 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
3699 case KVM_GET_DEVICE_ATTR:
3700 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
3701 case KVM_HAS_DEVICE_ATTR:
3702 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
3704 if (dev->ops->ioctl)
3705 return dev->ops->ioctl(dev, ioctl, arg);
3711 static int kvm_device_release(struct inode *inode, struct file *filp)
3713 struct kvm_device *dev = filp->private_data;
3714 struct kvm *kvm = dev->kvm;
3716 if (dev->ops->release) {
3717 mutex_lock(&kvm->lock);
3718 list_del(&dev->vm_node);
3719 dev->ops->release(dev);
3720 mutex_unlock(&kvm->lock);
3727 static const struct file_operations kvm_device_fops = {
3728 .unlocked_ioctl = kvm_device_ioctl,
3729 .release = kvm_device_release,
3730 KVM_COMPAT(kvm_device_ioctl),
3731 .mmap = kvm_device_mmap,
3734 struct kvm_device *kvm_device_from_filp(struct file *filp)
3736 if (filp->f_op != &kvm_device_fops)
3739 return filp->private_data;
3742 static const struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
3743 #ifdef CONFIG_KVM_MPIC
3744 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
3745 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
3749 int kvm_register_device_ops(const struct kvm_device_ops *ops, u32 type)
3751 if (type >= ARRAY_SIZE(kvm_device_ops_table))
3754 if (kvm_device_ops_table[type] != NULL)
3757 kvm_device_ops_table[type] = ops;
3761 void kvm_unregister_device_ops(u32 type)
3763 if (kvm_device_ops_table[type] != NULL)
3764 kvm_device_ops_table[type] = NULL;
3767 static int kvm_ioctl_create_device(struct kvm *kvm,
3768 struct kvm_create_device *cd)
3770 const struct kvm_device_ops *ops = NULL;
3771 struct kvm_device *dev;
3772 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
3776 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
3779 type = array_index_nospec(cd->type, ARRAY_SIZE(kvm_device_ops_table));
3780 ops = kvm_device_ops_table[type];
3787 dev = kzalloc(sizeof(*dev), GFP_KERNEL_ACCOUNT);
3794 mutex_lock(&kvm->lock);
3795 ret = ops->create(dev, type);
3797 mutex_unlock(&kvm->lock);
3801 list_add(&dev->vm_node, &kvm->devices);
3802 mutex_unlock(&kvm->lock);
3808 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
3810 kvm_put_kvm_no_destroy(kvm);
3811 mutex_lock(&kvm->lock);
3812 list_del(&dev->vm_node);
3813 mutex_unlock(&kvm->lock);
3822 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
3825 case KVM_CAP_USER_MEMORY:
3826 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
3827 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
3828 case KVM_CAP_INTERNAL_ERROR_DATA:
3829 #ifdef CONFIG_HAVE_KVM_MSI
3830 case KVM_CAP_SIGNAL_MSI:
3832 #ifdef CONFIG_HAVE_KVM_IRQFD
3834 case KVM_CAP_IRQFD_RESAMPLE:
3836 case KVM_CAP_IOEVENTFD_ANY_LENGTH:
3837 case KVM_CAP_CHECK_EXTENSION_VM:
3838 case KVM_CAP_ENABLE_CAP_VM:
3839 case KVM_CAP_HALT_POLL:
3841 #ifdef CONFIG_KVM_MMIO
3842 case KVM_CAP_COALESCED_MMIO:
3843 return KVM_COALESCED_MMIO_PAGE_OFFSET;
3844 case KVM_CAP_COALESCED_PIO:
3847 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3848 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2:
3849 return KVM_DIRTY_LOG_MANUAL_CAPS;
3851 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3852 case KVM_CAP_IRQ_ROUTING:
3853 return KVM_MAX_IRQ_ROUTES;
3855 #if KVM_ADDRESS_SPACE_NUM > 1
3856 case KVM_CAP_MULTI_ADDRESS_SPACE:
3857 return KVM_ADDRESS_SPACE_NUM;
3859 case KVM_CAP_NR_MEMSLOTS:
3860 return KVM_USER_MEM_SLOTS;
3861 case KVM_CAP_DIRTY_LOG_RING:
3862 #if KVM_DIRTY_LOG_PAGE_OFFSET > 0
3863 return KVM_DIRTY_RING_MAX_ENTRIES * sizeof(struct kvm_dirty_gfn);
3870 return kvm_vm_ioctl_check_extension(kvm, arg);
3873 static int kvm_vm_ioctl_enable_dirty_log_ring(struct kvm *kvm, u32 size)
3877 if (!KVM_DIRTY_LOG_PAGE_OFFSET)
3880 /* the size should be power of 2 */
3881 if (!size || (size & (size - 1)))
3884 /* Should be bigger to keep the reserved entries, or a page */
3885 if (size < kvm_dirty_ring_get_rsvd_entries() *
3886 sizeof(struct kvm_dirty_gfn) || size < PAGE_SIZE)
3889 if (size > KVM_DIRTY_RING_MAX_ENTRIES *
3890 sizeof(struct kvm_dirty_gfn))
3893 /* We only allow it to set once */
3894 if (kvm->dirty_ring_size)
3897 mutex_lock(&kvm->lock);
3899 if (kvm->created_vcpus) {
3900 /* We don't allow to change this value after vcpu created */
3903 kvm->dirty_ring_size = size;
3907 mutex_unlock(&kvm->lock);
3911 static int kvm_vm_ioctl_reset_dirty_pages(struct kvm *kvm)
3914 struct kvm_vcpu *vcpu;
3917 if (!kvm->dirty_ring_size)
3920 mutex_lock(&kvm->slots_lock);
3922 kvm_for_each_vcpu(i, vcpu, kvm)
3923 cleared += kvm_dirty_ring_reset(vcpu->kvm, &vcpu->dirty_ring);
3925 mutex_unlock(&kvm->slots_lock);
3928 kvm_flush_remote_tlbs(kvm);
3933 int __attribute__((weak)) kvm_vm_ioctl_enable_cap(struct kvm *kvm,
3934 struct kvm_enable_cap *cap)
3939 static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm,
3940 struct kvm_enable_cap *cap)
3943 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3944 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2: {
3945 u64 allowed_options = KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE;
3947 if (cap->args[0] & KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE)
3948 allowed_options = KVM_DIRTY_LOG_MANUAL_CAPS;
3950 if (cap->flags || (cap->args[0] & ~allowed_options))
3952 kvm->manual_dirty_log_protect = cap->args[0];
3956 case KVM_CAP_HALT_POLL: {
3957 if (cap->flags || cap->args[0] != (unsigned int)cap->args[0])
3960 kvm->max_halt_poll_ns = cap->args[0];
3963 case KVM_CAP_DIRTY_LOG_RING:
3964 return kvm_vm_ioctl_enable_dirty_log_ring(kvm, cap->args[0]);
3966 return kvm_vm_ioctl_enable_cap(kvm, cap);
3970 static long kvm_vm_ioctl(struct file *filp,
3971 unsigned int ioctl, unsigned long arg)
3973 struct kvm *kvm = filp->private_data;
3974 void __user *argp = (void __user *)arg;
3977 if (kvm->mm != current->mm)
3980 case KVM_CREATE_VCPU:
3981 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
3983 case KVM_ENABLE_CAP: {
3984 struct kvm_enable_cap cap;
3987 if (copy_from_user(&cap, argp, sizeof(cap)))
3989 r = kvm_vm_ioctl_enable_cap_generic(kvm, &cap);
3992 case KVM_SET_USER_MEMORY_REGION: {
3993 struct kvm_userspace_memory_region kvm_userspace_mem;
3996 if (copy_from_user(&kvm_userspace_mem, argp,
3997 sizeof(kvm_userspace_mem)))
4000 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
4003 case KVM_GET_DIRTY_LOG: {
4004 struct kvm_dirty_log log;
4007 if (copy_from_user(&log, argp, sizeof(log)))
4009 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
4012 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4013 case KVM_CLEAR_DIRTY_LOG: {
4014 struct kvm_clear_dirty_log log;
4017 if (copy_from_user(&log, argp, sizeof(log)))
4019 r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
4023 #ifdef CONFIG_KVM_MMIO
4024 case KVM_REGISTER_COALESCED_MMIO: {
4025 struct kvm_coalesced_mmio_zone zone;
4028 if (copy_from_user(&zone, argp, sizeof(zone)))
4030 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
4033 case KVM_UNREGISTER_COALESCED_MMIO: {
4034 struct kvm_coalesced_mmio_zone zone;
4037 if (copy_from_user(&zone, argp, sizeof(zone)))
4039 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
4044 struct kvm_irqfd data;
4047 if (copy_from_user(&data, argp, sizeof(data)))
4049 r = kvm_irqfd(kvm, &data);
4052 case KVM_IOEVENTFD: {
4053 struct kvm_ioeventfd data;
4056 if (copy_from_user(&data, argp, sizeof(data)))
4058 r = kvm_ioeventfd(kvm, &data);
4061 #ifdef CONFIG_HAVE_KVM_MSI
4062 case KVM_SIGNAL_MSI: {
4066 if (copy_from_user(&msi, argp, sizeof(msi)))
4068 r = kvm_send_userspace_msi(kvm, &msi);
4072 #ifdef __KVM_HAVE_IRQ_LINE
4073 case KVM_IRQ_LINE_STATUS:
4074 case KVM_IRQ_LINE: {
4075 struct kvm_irq_level irq_event;
4078 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
4081 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
4082 ioctl == KVM_IRQ_LINE_STATUS);
4087 if (ioctl == KVM_IRQ_LINE_STATUS) {
4088 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
4096 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
4097 case KVM_SET_GSI_ROUTING: {
4098 struct kvm_irq_routing routing;
4099 struct kvm_irq_routing __user *urouting;
4100 struct kvm_irq_routing_entry *entries = NULL;
4103 if (copy_from_user(&routing, argp, sizeof(routing)))
4106 if (!kvm_arch_can_set_irq_routing(kvm))
4108 if (routing.nr > KVM_MAX_IRQ_ROUTES)
4114 entries = vmemdup_user(urouting->entries,
4115 array_size(sizeof(*entries),
4117 if (IS_ERR(entries)) {
4118 r = PTR_ERR(entries);
4122 r = kvm_set_irq_routing(kvm, entries, routing.nr,
4127 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
4128 case KVM_CREATE_DEVICE: {
4129 struct kvm_create_device cd;
4132 if (copy_from_user(&cd, argp, sizeof(cd)))
4135 r = kvm_ioctl_create_device(kvm, &cd);
4140 if (copy_to_user(argp, &cd, sizeof(cd)))
4146 case KVM_CHECK_EXTENSION:
4147 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
4149 case KVM_RESET_DIRTY_RINGS:
4150 r = kvm_vm_ioctl_reset_dirty_pages(kvm);
4153 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
4159 #ifdef CONFIG_KVM_COMPAT
4160 struct compat_kvm_dirty_log {
4164 compat_uptr_t dirty_bitmap; /* one bit per page */
4169 static long kvm_vm_compat_ioctl(struct file *filp,
4170 unsigned int ioctl, unsigned long arg)
4172 struct kvm *kvm = filp->private_data;
4175 if (kvm->mm != current->mm)
4178 case KVM_GET_DIRTY_LOG: {
4179 struct compat_kvm_dirty_log compat_log;
4180 struct kvm_dirty_log log;
4182 if (copy_from_user(&compat_log, (void __user *)arg,
4183 sizeof(compat_log)))
4185 log.slot = compat_log.slot;
4186 log.padding1 = compat_log.padding1;
4187 log.padding2 = compat_log.padding2;
4188 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
4190 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
4194 r = kvm_vm_ioctl(filp, ioctl, arg);
4200 static struct file_operations kvm_vm_fops = {
4201 .release = kvm_vm_release,
4202 .unlocked_ioctl = kvm_vm_ioctl,
4203 .llseek = noop_llseek,
4204 KVM_COMPAT(kvm_vm_compat_ioctl),
4207 bool file_is_kvm(struct file *file)
4209 return file && file->f_op == &kvm_vm_fops;
4211 EXPORT_SYMBOL_GPL(file_is_kvm);
4213 static int kvm_dev_ioctl_create_vm(unsigned long type)
4219 kvm = kvm_create_vm(type);
4221 return PTR_ERR(kvm);
4222 #ifdef CONFIG_KVM_MMIO
4223 r = kvm_coalesced_mmio_init(kvm);
4227 r = get_unused_fd_flags(O_CLOEXEC);
4231 file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
4239 * Don't call kvm_put_kvm anymore at this point; file->f_op is
4240 * already set, with ->release() being kvm_vm_release(). In error
4241 * cases it will be called by the final fput(file) and will take
4242 * care of doing kvm_put_kvm(kvm).
4244 if (kvm_create_vm_debugfs(kvm, r) < 0) {
4249 kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm);
4251 fd_install(r, file);
4259 static long kvm_dev_ioctl(struct file *filp,
4260 unsigned int ioctl, unsigned long arg)
4265 case KVM_GET_API_VERSION:
4268 r = KVM_API_VERSION;
4271 r = kvm_dev_ioctl_create_vm(arg);
4273 case KVM_CHECK_EXTENSION:
4274 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
4276 case KVM_GET_VCPU_MMAP_SIZE:
4279 r = PAGE_SIZE; /* struct kvm_run */
4281 r += PAGE_SIZE; /* pio data page */
4283 #ifdef CONFIG_KVM_MMIO
4284 r += PAGE_SIZE; /* coalesced mmio ring page */
4287 case KVM_TRACE_ENABLE:
4288 case KVM_TRACE_PAUSE:
4289 case KVM_TRACE_DISABLE:
4293 return kvm_arch_dev_ioctl(filp, ioctl, arg);
4299 static struct file_operations kvm_chardev_ops = {
4300 .unlocked_ioctl = kvm_dev_ioctl,
4301 .llseek = noop_llseek,
4302 KVM_COMPAT(kvm_dev_ioctl),
4305 static struct miscdevice kvm_dev = {
4311 static void hardware_enable_nolock(void *junk)
4313 int cpu = raw_smp_processor_id();
4316 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
4319 cpumask_set_cpu(cpu, cpus_hardware_enabled);
4321 r = kvm_arch_hardware_enable();
4324 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
4325 atomic_inc(&hardware_enable_failed);
4326 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
4330 static int kvm_starting_cpu(unsigned int cpu)
4332 raw_spin_lock(&kvm_count_lock);
4333 if (kvm_usage_count)
4334 hardware_enable_nolock(NULL);
4335 raw_spin_unlock(&kvm_count_lock);
4339 static void hardware_disable_nolock(void *junk)
4341 int cpu = raw_smp_processor_id();
4343 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
4345 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
4346 kvm_arch_hardware_disable();
4349 static int kvm_dying_cpu(unsigned int cpu)
4351 raw_spin_lock(&kvm_count_lock);
4352 if (kvm_usage_count)
4353 hardware_disable_nolock(NULL);
4354 raw_spin_unlock(&kvm_count_lock);
4358 static void hardware_disable_all_nolock(void)
4360 BUG_ON(!kvm_usage_count);
4363 if (!kvm_usage_count)
4364 on_each_cpu(hardware_disable_nolock, NULL, 1);
4367 static void hardware_disable_all(void)
4369 raw_spin_lock(&kvm_count_lock);
4370 hardware_disable_all_nolock();
4371 raw_spin_unlock(&kvm_count_lock);
4374 static int hardware_enable_all(void)
4378 raw_spin_lock(&kvm_count_lock);
4381 if (kvm_usage_count == 1) {
4382 atomic_set(&hardware_enable_failed, 0);
4383 on_each_cpu(hardware_enable_nolock, NULL, 1);
4385 if (atomic_read(&hardware_enable_failed)) {
4386 hardware_disable_all_nolock();
4391 raw_spin_unlock(&kvm_count_lock);
4396 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
4400 * Some (well, at least mine) BIOSes hang on reboot if
4403 * And Intel TXT required VMX off for all cpu when system shutdown.
4405 pr_info("kvm: exiting hardware virtualization\n");
4406 kvm_rebooting = true;
4407 on_each_cpu(hardware_disable_nolock, NULL, 1);
4411 static struct notifier_block kvm_reboot_notifier = {
4412 .notifier_call = kvm_reboot,
4416 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
4420 for (i = 0; i < bus->dev_count; i++) {
4421 struct kvm_io_device *pos = bus->range[i].dev;
4423 kvm_iodevice_destructor(pos);
4428 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
4429 const struct kvm_io_range *r2)
4431 gpa_t addr1 = r1->addr;
4432 gpa_t addr2 = r2->addr;
4437 /* If r2->len == 0, match the exact address. If r2->len != 0,
4438 * accept any overlapping write. Any order is acceptable for
4439 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
4440 * we process all of them.
4453 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
4455 return kvm_io_bus_cmp(p1, p2);
4458 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
4459 gpa_t addr, int len)
4461 struct kvm_io_range *range, key;
4464 key = (struct kvm_io_range) {
4469 range = bsearch(&key, bus->range, bus->dev_count,
4470 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
4474 off = range - bus->range;
4476 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
4482 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
4483 struct kvm_io_range *range, const void *val)
4487 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
4491 while (idx < bus->dev_count &&
4492 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
4493 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
4502 /* kvm_io_bus_write - called under kvm->slots_lock */
4503 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
4504 int len, const void *val)
4506 struct kvm_io_bus *bus;
4507 struct kvm_io_range range;
4510 range = (struct kvm_io_range) {
4515 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
4518 r = __kvm_io_bus_write(vcpu, bus, &range, val);
4519 return r < 0 ? r : 0;
4521 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
4523 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
4524 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
4525 gpa_t addr, int len, const void *val, long cookie)
4527 struct kvm_io_bus *bus;
4528 struct kvm_io_range range;
4530 range = (struct kvm_io_range) {
4535 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
4539 /* First try the device referenced by cookie. */
4540 if ((cookie >= 0) && (cookie < bus->dev_count) &&
4541 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
4542 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
4547 * cookie contained garbage; fall back to search and return the
4548 * correct cookie value.
4550 return __kvm_io_bus_write(vcpu, bus, &range, val);
4553 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
4554 struct kvm_io_range *range, void *val)
4558 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
4562 while (idx < bus->dev_count &&
4563 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
4564 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
4573 /* kvm_io_bus_read - called under kvm->slots_lock */
4574 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
4577 struct kvm_io_bus *bus;
4578 struct kvm_io_range range;
4581 range = (struct kvm_io_range) {
4586 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
4589 r = __kvm_io_bus_read(vcpu, bus, &range, val);
4590 return r < 0 ? r : 0;
4593 /* Caller must hold slots_lock. */
4594 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
4595 int len, struct kvm_io_device *dev)
4598 struct kvm_io_bus *new_bus, *bus;
4599 struct kvm_io_range range;
4601 bus = kvm_get_bus(kvm, bus_idx);
4605 /* exclude ioeventfd which is limited by maximum fd */
4606 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
4609 new_bus = kmalloc(struct_size(bus, range, bus->dev_count + 1),
4610 GFP_KERNEL_ACCOUNT);
4614 range = (struct kvm_io_range) {
4620 for (i = 0; i < bus->dev_count; i++)
4621 if (kvm_io_bus_cmp(&bus->range[i], &range) > 0)
4624 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
4625 new_bus->dev_count++;
4626 new_bus->range[i] = range;
4627 memcpy(new_bus->range + i + 1, bus->range + i,
4628 (bus->dev_count - i) * sizeof(struct kvm_io_range));
4629 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
4630 synchronize_srcu_expedited(&kvm->srcu);
4636 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
4637 struct kvm_io_device *dev)
4640 struct kvm_io_bus *new_bus, *bus;
4642 lockdep_assert_held(&kvm->slots_lock);
4644 bus = kvm_get_bus(kvm, bus_idx);
4648 for (i = 0; i < bus->dev_count; i++) {
4649 if (bus->range[i].dev == dev) {
4654 if (i == bus->dev_count)
4657 new_bus = kmalloc(struct_size(bus, range, bus->dev_count - 1),
4658 GFP_KERNEL_ACCOUNT);
4660 memcpy(new_bus, bus, struct_size(bus, range, i));
4661 new_bus->dev_count--;
4662 memcpy(new_bus->range + i, bus->range + i + 1,
4663 flex_array_size(new_bus, range, new_bus->dev_count - i));
4666 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
4667 synchronize_srcu_expedited(&kvm->srcu);
4669 /* Destroy the old bus _after_ installing the (null) bus. */
4671 pr_err("kvm: failed to shrink bus, removing it completely\n");
4672 for (j = 0; j < bus->dev_count; j++) {
4675 kvm_iodevice_destructor(bus->range[j].dev);
4680 return new_bus ? 0 : -ENOMEM;
4683 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
4686 struct kvm_io_bus *bus;
4687 int dev_idx, srcu_idx;
4688 struct kvm_io_device *iodev = NULL;
4690 srcu_idx = srcu_read_lock(&kvm->srcu);
4692 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
4696 dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
4700 iodev = bus->range[dev_idx].dev;
4703 srcu_read_unlock(&kvm->srcu, srcu_idx);
4707 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
4709 static int kvm_debugfs_open(struct inode *inode, struct file *file,
4710 int (*get)(void *, u64 *), int (*set)(void *, u64),
4713 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
4716 /* The debugfs files are a reference to the kvm struct which
4717 * is still valid when kvm_destroy_vm is called.
4718 * To avoid the race between open and the removal of the debugfs
4719 * directory we test against the users count.
4721 if (!refcount_inc_not_zero(&stat_data->kvm->users_count))
4724 if (simple_attr_open(inode, file, get,
4725 KVM_DBGFS_GET_MODE(stat_data->dbgfs_item) & 0222
4728 kvm_put_kvm(stat_data->kvm);
4735 static int kvm_debugfs_release(struct inode *inode, struct file *file)
4737 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
4740 simple_attr_release(inode, file);
4741 kvm_put_kvm(stat_data->kvm);
4746 static int kvm_get_stat_per_vm(struct kvm *kvm, size_t offset, u64 *val)
4748 *val = *(ulong *)((void *)kvm + offset);
4753 static int kvm_clear_stat_per_vm(struct kvm *kvm, size_t offset)
4755 *(ulong *)((void *)kvm + offset) = 0;
4760 static int kvm_get_stat_per_vcpu(struct kvm *kvm, size_t offset, u64 *val)
4763 struct kvm_vcpu *vcpu;
4767 kvm_for_each_vcpu(i, vcpu, kvm)
4768 *val += *(u64 *)((void *)vcpu + offset);
4773 static int kvm_clear_stat_per_vcpu(struct kvm *kvm, size_t offset)
4776 struct kvm_vcpu *vcpu;
4778 kvm_for_each_vcpu(i, vcpu, kvm)
4779 *(u64 *)((void *)vcpu + offset) = 0;
4784 static int kvm_stat_data_get(void *data, u64 *val)
4787 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
4789 switch (stat_data->dbgfs_item->kind) {
4791 r = kvm_get_stat_per_vm(stat_data->kvm,
4792 stat_data->dbgfs_item->offset, val);
4795 r = kvm_get_stat_per_vcpu(stat_data->kvm,
4796 stat_data->dbgfs_item->offset, val);
4803 static int kvm_stat_data_clear(void *data, u64 val)
4806 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
4811 switch (stat_data->dbgfs_item->kind) {
4813 r = kvm_clear_stat_per_vm(stat_data->kvm,
4814 stat_data->dbgfs_item->offset);
4817 r = kvm_clear_stat_per_vcpu(stat_data->kvm,
4818 stat_data->dbgfs_item->offset);
4825 static int kvm_stat_data_open(struct inode *inode, struct file *file)
4827 __simple_attr_check_format("%llu\n", 0ull);
4828 return kvm_debugfs_open(inode, file, kvm_stat_data_get,
4829 kvm_stat_data_clear, "%llu\n");
4832 static const struct file_operations stat_fops_per_vm = {
4833 .owner = THIS_MODULE,
4834 .open = kvm_stat_data_open,
4835 .release = kvm_debugfs_release,
4836 .read = simple_attr_read,
4837 .write = simple_attr_write,
4838 .llseek = no_llseek,
4841 static int vm_stat_get(void *_offset, u64 *val)
4843 unsigned offset = (long)_offset;
4848 mutex_lock(&kvm_lock);
4849 list_for_each_entry(kvm, &vm_list, vm_list) {
4850 kvm_get_stat_per_vm(kvm, offset, &tmp_val);
4853 mutex_unlock(&kvm_lock);
4857 static int vm_stat_clear(void *_offset, u64 val)
4859 unsigned offset = (long)_offset;
4865 mutex_lock(&kvm_lock);
4866 list_for_each_entry(kvm, &vm_list, vm_list) {
4867 kvm_clear_stat_per_vm(kvm, offset);
4869 mutex_unlock(&kvm_lock);
4874 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n");
4876 static int vcpu_stat_get(void *_offset, u64 *val)
4878 unsigned offset = (long)_offset;
4883 mutex_lock(&kvm_lock);
4884 list_for_each_entry(kvm, &vm_list, vm_list) {
4885 kvm_get_stat_per_vcpu(kvm, offset, &tmp_val);
4888 mutex_unlock(&kvm_lock);
4892 static int vcpu_stat_clear(void *_offset, u64 val)
4894 unsigned offset = (long)_offset;
4900 mutex_lock(&kvm_lock);
4901 list_for_each_entry(kvm, &vm_list, vm_list) {
4902 kvm_clear_stat_per_vcpu(kvm, offset);
4904 mutex_unlock(&kvm_lock);
4909 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear,
4912 static const struct file_operations *stat_fops[] = {
4913 [KVM_STAT_VCPU] = &vcpu_stat_fops,
4914 [KVM_STAT_VM] = &vm_stat_fops,
4917 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm)
4919 struct kobj_uevent_env *env;
4920 unsigned long long created, active;
4922 if (!kvm_dev.this_device || !kvm)
4925 mutex_lock(&kvm_lock);
4926 if (type == KVM_EVENT_CREATE_VM) {
4927 kvm_createvm_count++;
4929 } else if (type == KVM_EVENT_DESTROY_VM) {
4932 created = kvm_createvm_count;
4933 active = kvm_active_vms;
4934 mutex_unlock(&kvm_lock);
4936 env = kzalloc(sizeof(*env), GFP_KERNEL_ACCOUNT);
4940 add_uevent_var(env, "CREATED=%llu", created);
4941 add_uevent_var(env, "COUNT=%llu", active);
4943 if (type == KVM_EVENT_CREATE_VM) {
4944 add_uevent_var(env, "EVENT=create");
4945 kvm->userspace_pid = task_pid_nr(current);
4946 } else if (type == KVM_EVENT_DESTROY_VM) {
4947 add_uevent_var(env, "EVENT=destroy");
4949 add_uevent_var(env, "PID=%d", kvm->userspace_pid);
4951 if (!IS_ERR_OR_NULL(kvm->debugfs_dentry)) {
4952 char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL_ACCOUNT);
4955 tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX);
4957 add_uevent_var(env, "STATS_PATH=%s", tmp);
4961 /* no need for checks, since we are adding at most only 5 keys */
4962 env->envp[env->envp_idx++] = NULL;
4963 kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp);
4967 static void kvm_init_debug(void)
4969 struct kvm_stats_debugfs_item *p;
4971 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
4973 kvm_debugfs_num_entries = 0;
4974 for (p = debugfs_entries; p->name; ++p, kvm_debugfs_num_entries++) {
4975 debugfs_create_file(p->name, KVM_DBGFS_GET_MODE(p),
4976 kvm_debugfs_dir, (void *)(long)p->offset,
4977 stat_fops[p->kind]);
4981 static int kvm_suspend(void)
4983 if (kvm_usage_count)
4984 hardware_disable_nolock(NULL);
4988 static void kvm_resume(void)
4990 if (kvm_usage_count) {
4991 #ifdef CONFIG_LOCKDEP
4992 WARN_ON(lockdep_is_held(&kvm_count_lock));
4994 hardware_enable_nolock(NULL);
4998 static struct syscore_ops kvm_syscore_ops = {
4999 .suspend = kvm_suspend,
5000 .resume = kvm_resume,
5004 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
5006 return container_of(pn, struct kvm_vcpu, preempt_notifier);
5009 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
5011 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
5013 WRITE_ONCE(vcpu->preempted, false);
5014 WRITE_ONCE(vcpu->ready, false);
5016 __this_cpu_write(kvm_running_vcpu, vcpu);
5017 kvm_arch_sched_in(vcpu, cpu);
5018 kvm_arch_vcpu_load(vcpu, cpu);
5021 static void kvm_sched_out(struct preempt_notifier *pn,
5022 struct task_struct *next)
5024 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
5026 if (current->state == TASK_RUNNING) {
5027 WRITE_ONCE(vcpu->preempted, true);
5028 WRITE_ONCE(vcpu->ready, true);
5030 kvm_arch_vcpu_put(vcpu);
5031 __this_cpu_write(kvm_running_vcpu, NULL);
5035 * kvm_get_running_vcpu - get the vcpu running on the current CPU.
5037 * We can disable preemption locally around accessing the per-CPU variable,
5038 * and use the resolved vcpu pointer after enabling preemption again,
5039 * because even if the current thread is migrated to another CPU, reading
5040 * the per-CPU value later will give us the same value as we update the
5041 * per-CPU variable in the preempt notifier handlers.
5043 struct kvm_vcpu *kvm_get_running_vcpu(void)
5045 struct kvm_vcpu *vcpu;
5048 vcpu = __this_cpu_read(kvm_running_vcpu);
5053 EXPORT_SYMBOL_GPL(kvm_get_running_vcpu);
5056 * kvm_get_running_vcpus - get the per-CPU array of currently running vcpus.
5058 struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void)
5060 return &kvm_running_vcpu;
5063 struct kvm_cpu_compat_check {
5068 static void check_processor_compat(void *data)
5070 struct kvm_cpu_compat_check *c = data;
5072 *c->ret = kvm_arch_check_processor_compat(c->opaque);
5075 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
5076 struct module *module)
5078 struct kvm_cpu_compat_check c;
5082 r = kvm_arch_init(opaque);
5087 * kvm_arch_init makes sure there's at most one caller
5088 * for architectures that support multiple implementations,
5089 * like intel and amd on x86.
5090 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
5091 * conflicts in case kvm is already setup for another implementation.
5093 r = kvm_irqfd_init();
5097 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
5102 r = kvm_arch_hardware_setup(opaque);
5108 for_each_online_cpu(cpu) {
5109 smp_call_function_single(cpu, check_processor_compat, &c, 1);
5114 r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting",
5115 kvm_starting_cpu, kvm_dying_cpu);
5118 register_reboot_notifier(&kvm_reboot_notifier);
5120 /* A kmem cache lets us meet the alignment requirements of fx_save. */
5122 vcpu_align = __alignof__(struct kvm_vcpu);
5124 kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align,
5126 offsetof(struct kvm_vcpu, arch),
5127 sizeof_field(struct kvm_vcpu, arch),
5129 if (!kvm_vcpu_cache) {
5134 r = kvm_async_pf_init();
5138 kvm_chardev_ops.owner = module;
5139 kvm_vm_fops.owner = module;
5140 kvm_vcpu_fops.owner = module;
5142 r = misc_register(&kvm_dev);
5144 pr_err("kvm: misc device register failed\n");
5148 register_syscore_ops(&kvm_syscore_ops);
5150 kvm_preempt_ops.sched_in = kvm_sched_in;
5151 kvm_preempt_ops.sched_out = kvm_sched_out;
5155 r = kvm_vfio_ops_init();
5161 kvm_async_pf_deinit();
5163 kmem_cache_destroy(kvm_vcpu_cache);
5165 unregister_reboot_notifier(&kvm_reboot_notifier);
5166 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
5168 kvm_arch_hardware_unsetup();
5170 free_cpumask_var(cpus_hardware_enabled);
5178 EXPORT_SYMBOL_GPL(kvm_init);
5182 debugfs_remove_recursive(kvm_debugfs_dir);
5183 misc_deregister(&kvm_dev);
5184 kmem_cache_destroy(kvm_vcpu_cache);
5185 kvm_async_pf_deinit();
5186 unregister_syscore_ops(&kvm_syscore_ops);
5187 unregister_reboot_notifier(&kvm_reboot_notifier);
5188 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
5189 on_each_cpu(hardware_disable_nolock, NULL, 1);
5190 kvm_arch_hardware_unsetup();
5193 free_cpumask_var(cpus_hardware_enabled);
5194 kvm_vfio_ops_exit();
5196 EXPORT_SYMBOL_GPL(kvm_exit);
5198 struct kvm_vm_worker_thread_context {
5200 struct task_struct *parent;
5201 struct completion init_done;
5202 kvm_vm_thread_fn_t thread_fn;
5207 static int kvm_vm_worker_thread(void *context)
5210 * The init_context is allocated on the stack of the parent thread, so
5211 * we have to locally copy anything that is needed beyond initialization
5213 struct kvm_vm_worker_thread_context *init_context = context;
5214 struct kvm *kvm = init_context->kvm;
5215 kvm_vm_thread_fn_t thread_fn = init_context->thread_fn;
5216 uintptr_t data = init_context->data;
5219 err = kthread_park(current);
5220 /* kthread_park(current) is never supposed to return an error */
5225 err = cgroup_attach_task_all(init_context->parent, current);
5227 kvm_err("%s: cgroup_attach_task_all failed with err %d\n",
5232 set_user_nice(current, task_nice(init_context->parent));
5235 init_context->err = err;
5236 complete(&init_context->init_done);
5237 init_context = NULL;
5242 /* Wait to be woken up by the spawner before proceeding. */
5245 if (!kthread_should_stop())
5246 err = thread_fn(kvm, data);
5251 int kvm_vm_create_worker_thread(struct kvm *kvm, kvm_vm_thread_fn_t thread_fn,
5252 uintptr_t data, const char *name,
5253 struct task_struct **thread_ptr)
5255 struct kvm_vm_worker_thread_context init_context = {};
5256 struct task_struct *thread;
5259 init_context.kvm = kvm;
5260 init_context.parent = current;
5261 init_context.thread_fn = thread_fn;
5262 init_context.data = data;
5263 init_completion(&init_context.init_done);
5265 thread = kthread_run(kvm_vm_worker_thread, &init_context,
5266 "%s-%d", name, task_pid_nr(current));
5268 return PTR_ERR(thread);
5270 /* kthread_run is never supposed to return NULL */
5271 WARN_ON(thread == NULL);
5273 wait_for_completion(&init_context.init_done);
5275 if (!init_context.err)
5276 *thread_ptr = thread;
5278 return init_context.err;