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 #ifdef KVM_ARCH_WANT_NEW_MMU_NOTIFIER_APIS
456 typedef bool (*hva_handler_t)(struct kvm *kvm, struct kvm_gfn_range *range);
458 struct kvm_hva_range {
462 hva_handler_t handler;
467 static __always_inline int __kvm_handle_hva_range(struct kvm *kvm,
468 const struct kvm_hva_range *range)
470 struct kvm_memory_slot *slot;
471 struct kvm_memslots *slots;
472 struct kvm_gfn_range gfn_range;
476 lockdep_assert_held_write(&kvm->mmu_lock);
478 idx = srcu_read_lock(&kvm->srcu);
480 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
481 slots = __kvm_memslots(kvm, i);
482 kvm_for_each_memslot(slot, slots) {
483 unsigned long hva_start, hva_end;
485 hva_start = max(range->start, slot->userspace_addr);
486 hva_end = min(range->end, slot->userspace_addr +
487 (slot->npages << PAGE_SHIFT));
488 if (hva_start >= hva_end)
492 * To optimize for the likely case where the address
493 * range is covered by zero or one memslots, don't
494 * bother making these conditional (to avoid writes on
495 * the second or later invocation of the handler).
497 gfn_range.pte = range->pte;
498 gfn_range.may_block = range->may_block;
501 * {gfn(page) | page intersects with [hva_start, hva_end)} =
502 * {gfn_start, gfn_start+1, ..., gfn_end-1}.
504 gfn_range.start = hva_to_gfn_memslot(hva_start, slot);
505 gfn_range.end = hva_to_gfn_memslot(hva_end + PAGE_SIZE - 1, slot);
506 gfn_range.slot = slot;
508 ret |= range->handler(kvm, &gfn_range);
512 if (range->flush_on_ret && (ret || kvm->tlbs_dirty))
513 kvm_flush_remote_tlbs(kvm);
515 srcu_read_unlock(&kvm->srcu, idx);
517 /* The notifiers are averse to booleans. :-( */
521 static __always_inline int kvm_handle_hva_range(struct mmu_notifier *mn,
525 hva_handler_t handler)
527 struct kvm *kvm = mmu_notifier_to_kvm(mn);
528 const struct kvm_hva_range range = {
533 .flush_on_ret = true,
539 ret = __kvm_handle_hva_range(kvm, &range);
545 static __always_inline int kvm_handle_hva_range_no_flush(struct mmu_notifier *mn,
548 hva_handler_t handler)
550 struct kvm *kvm = mmu_notifier_to_kvm(mn);
551 const struct kvm_hva_range range = {
556 .flush_on_ret = false,
562 ret = __kvm_handle_hva_range(kvm, &range);
567 #endif /* KVM_ARCH_WANT_NEW_MMU_NOTIFIER_APIS */
569 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
570 struct mm_struct *mm,
571 unsigned long address,
574 struct kvm *kvm = mmu_notifier_to_kvm(mn);
576 #ifndef KVM_ARCH_WANT_NEW_MMU_NOTIFIER_APIS
579 trace_kvm_set_spte_hva(address);
582 * .change_pte() must be surrounded by .invalidate_range_{start,end}(),
583 * and so always runs with an elevated notifier count. This obviates
584 * the need to bump the sequence count.
586 WARN_ON_ONCE(!kvm->mmu_notifier_count);
588 #ifdef KVM_ARCH_WANT_NEW_MMU_NOTIFIER_APIS
589 kvm_handle_hva_range(mn, address, address + 1, pte, kvm_set_spte_gfn);
591 idx = srcu_read_lock(&kvm->srcu);
595 if (kvm_set_spte_hva(kvm, address, pte))
596 kvm_flush_remote_tlbs(kvm);
599 srcu_read_unlock(&kvm->srcu, idx);
603 static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
604 const struct mmu_notifier_range *range)
606 struct kvm *kvm = mmu_notifier_to_kvm(mn);
607 #ifdef KVM_ARCH_WANT_NEW_MMU_NOTIFIER_APIS
608 const struct kvm_hva_range hva_range = {
609 .start = range->start,
612 .handler = kvm_unmap_gfn_range,
613 .flush_on_ret = true,
614 .may_block = mmu_notifier_range_blockable(range),
617 int need_tlb_flush = 0, idx;
620 trace_kvm_unmap_hva_range(range->start, range->end);
622 #ifndef KVM_ARCH_WANT_NEW_MMU_NOTIFIER_APIS
623 idx = srcu_read_lock(&kvm->srcu);
628 * The count increase must become visible at unlock time as no
629 * spte can be established without taking the mmu_lock and
630 * count is also read inside the mmu_lock critical section.
632 kvm->mmu_notifier_count++;
633 if (likely(kvm->mmu_notifier_count == 1)) {
634 kvm->mmu_notifier_range_start = range->start;
635 kvm->mmu_notifier_range_end = range->end;
638 * Fully tracking multiple concurrent ranges has dimishing
639 * returns. Keep things simple and just find the minimal range
640 * which includes the current and new ranges. As there won't be
641 * enough information to subtract a range after its invalidate
642 * completes, any ranges invalidated concurrently will
643 * accumulate and persist until all outstanding invalidates
646 kvm->mmu_notifier_range_start =
647 min(kvm->mmu_notifier_range_start, range->start);
648 kvm->mmu_notifier_range_end =
649 max(kvm->mmu_notifier_range_end, range->end);
652 #ifdef KVM_ARCH_WANT_NEW_MMU_NOTIFIER_APIS
653 __kvm_handle_hva_range(kvm, &hva_range);
655 need_tlb_flush = kvm_unmap_hva_range(kvm, range->start, range->end,
657 /* we've to flush the tlb before the pages can be freed */
658 if (need_tlb_flush || kvm->tlbs_dirty)
659 kvm_flush_remote_tlbs(kvm);
663 #ifndef KVM_ARCH_WANT_NEW_MMU_NOTIFIER_APIS
664 srcu_read_unlock(&kvm->srcu, idx);
670 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
671 const struct mmu_notifier_range *range)
673 struct kvm *kvm = mmu_notifier_to_kvm(mn);
677 * This sequence increase will notify the kvm page fault that
678 * the page that is going to be mapped in the spte could have
681 kvm->mmu_notifier_seq++;
684 * The above sequence increase must be visible before the
685 * below count decrease, which is ensured by the smp_wmb above
686 * in conjunction with the smp_rmb in mmu_notifier_retry().
688 kvm->mmu_notifier_count--;
691 BUG_ON(kvm->mmu_notifier_count < 0);
694 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
695 struct mm_struct *mm,
699 #ifndef KVM_ARCH_WANT_NEW_MMU_NOTIFIER_APIS
700 struct kvm *kvm = mmu_notifier_to_kvm(mn);
703 trace_kvm_age_hva(start, end);
705 #ifdef KVM_ARCH_WANT_NEW_MMU_NOTIFIER_APIS
706 return kvm_handle_hva_range(mn, start, end, __pte(0), kvm_age_gfn);
708 idx = srcu_read_lock(&kvm->srcu);
711 young = kvm_age_hva(kvm, start, end);
713 kvm_flush_remote_tlbs(kvm);
716 srcu_read_unlock(&kvm->srcu, idx);
722 static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
723 struct mm_struct *mm,
727 #ifndef KVM_ARCH_WANT_NEW_MMU_NOTIFIER_APIS
728 struct kvm *kvm = mmu_notifier_to_kvm(mn);
732 trace_kvm_age_hva(start, end);
735 * Even though we do not flush TLB, this will still adversely
736 * affect performance on pre-Haswell Intel EPT, where there is
737 * no EPT Access Bit to clear so that we have to tear down EPT
738 * tables instead. If we find this unacceptable, we can always
739 * add a parameter to kvm_age_hva so that it effectively doesn't
740 * do anything on clear_young.
742 * Also note that currently we never issue secondary TLB flushes
743 * from clear_young, leaving this job up to the regular system
744 * cadence. If we find this inaccurate, we might come up with a
745 * more sophisticated heuristic later.
747 #ifdef KVM_ARCH_WANT_NEW_MMU_NOTIFIER_APIS
748 return kvm_handle_hva_range_no_flush(mn, start, end, kvm_age_gfn);
750 idx = srcu_read_lock(&kvm->srcu);
752 young = kvm_age_hva(kvm, start, end);
754 srcu_read_unlock(&kvm->srcu, idx);
760 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
761 struct mm_struct *mm,
762 unsigned long address)
764 #ifndef KVM_ARCH_WANT_NEW_MMU_NOTIFIER_APIS
765 struct kvm *kvm = mmu_notifier_to_kvm(mn);
768 trace_kvm_test_age_hva(address);
770 #ifdef KVM_ARCH_WANT_NEW_MMU_NOTIFIER_APIS
771 return kvm_handle_hva_range_no_flush(mn, address, address + 1,
774 idx = srcu_read_lock(&kvm->srcu);
776 young = kvm_test_age_hva(kvm, address);
778 srcu_read_unlock(&kvm->srcu, idx);
784 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
785 struct mm_struct *mm)
787 struct kvm *kvm = mmu_notifier_to_kvm(mn);
790 idx = srcu_read_lock(&kvm->srcu);
791 kvm_arch_flush_shadow_all(kvm);
792 srcu_read_unlock(&kvm->srcu, idx);
795 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
796 .invalidate_range = kvm_mmu_notifier_invalidate_range,
797 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
798 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
799 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
800 .clear_young = kvm_mmu_notifier_clear_young,
801 .test_young = kvm_mmu_notifier_test_young,
802 .change_pte = kvm_mmu_notifier_change_pte,
803 .release = kvm_mmu_notifier_release,
806 static int kvm_init_mmu_notifier(struct kvm *kvm)
808 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
809 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
812 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
814 static int kvm_init_mmu_notifier(struct kvm *kvm)
819 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
821 static struct kvm_memslots *kvm_alloc_memslots(void)
824 struct kvm_memslots *slots;
826 slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL_ACCOUNT);
830 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
831 slots->id_to_index[i] = -1;
836 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
838 if (!memslot->dirty_bitmap)
841 kvfree(memslot->dirty_bitmap);
842 memslot->dirty_bitmap = NULL;
845 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
847 kvm_destroy_dirty_bitmap(slot);
849 kvm_arch_free_memslot(kvm, slot);
855 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
857 struct kvm_memory_slot *memslot;
862 kvm_for_each_memslot(memslot, slots)
863 kvm_free_memslot(kvm, memslot);
868 static void kvm_destroy_vm_debugfs(struct kvm *kvm)
872 if (!kvm->debugfs_dentry)
875 debugfs_remove_recursive(kvm->debugfs_dentry);
877 if (kvm->debugfs_stat_data) {
878 for (i = 0; i < kvm_debugfs_num_entries; i++)
879 kfree(kvm->debugfs_stat_data[i]);
880 kfree(kvm->debugfs_stat_data);
884 static int kvm_create_vm_debugfs(struct kvm *kvm, int fd)
886 char dir_name[ITOA_MAX_LEN * 2];
887 struct kvm_stat_data *stat_data;
888 struct kvm_stats_debugfs_item *p;
890 if (!debugfs_initialized())
893 snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd);
894 kvm->debugfs_dentry = debugfs_create_dir(dir_name, kvm_debugfs_dir);
896 kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries,
897 sizeof(*kvm->debugfs_stat_data),
899 if (!kvm->debugfs_stat_data)
902 for (p = debugfs_entries; p->name; p++) {
903 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT);
907 stat_data->kvm = kvm;
908 stat_data->dbgfs_item = p;
909 kvm->debugfs_stat_data[p - debugfs_entries] = stat_data;
910 debugfs_create_file(p->name, KVM_DBGFS_GET_MODE(p),
911 kvm->debugfs_dentry, stat_data,
918 * Called after the VM is otherwise initialized, but just before adding it to
921 int __weak kvm_arch_post_init_vm(struct kvm *kvm)
927 * Called just after removing the VM from the vm_list, but before doing any
930 void __weak kvm_arch_pre_destroy_vm(struct kvm *kvm)
934 static struct kvm *kvm_create_vm(unsigned long type)
936 struct kvm *kvm = kvm_arch_alloc_vm();
941 return ERR_PTR(-ENOMEM);
943 KVM_MMU_LOCK_INIT(kvm);
945 kvm->mm = current->mm;
946 kvm_eventfd_init(kvm);
947 mutex_init(&kvm->lock);
948 mutex_init(&kvm->irq_lock);
949 mutex_init(&kvm->slots_lock);
950 INIT_LIST_HEAD(&kvm->devices);
952 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
954 if (init_srcu_struct(&kvm->srcu))
955 goto out_err_no_srcu;
956 if (init_srcu_struct(&kvm->irq_srcu))
957 goto out_err_no_irq_srcu;
959 refcount_set(&kvm->users_count, 1);
960 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
961 struct kvm_memslots *slots = kvm_alloc_memslots();
964 goto out_err_no_arch_destroy_vm;
965 /* Generations must be different for each address space. */
966 slots->generation = i;
967 rcu_assign_pointer(kvm->memslots[i], slots);
970 for (i = 0; i < KVM_NR_BUSES; i++) {
971 rcu_assign_pointer(kvm->buses[i],
972 kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL_ACCOUNT));
974 goto out_err_no_arch_destroy_vm;
977 kvm->max_halt_poll_ns = halt_poll_ns;
979 r = kvm_arch_init_vm(kvm, type);
981 goto out_err_no_arch_destroy_vm;
983 r = hardware_enable_all();
985 goto out_err_no_disable;
987 #ifdef CONFIG_HAVE_KVM_IRQFD
988 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
991 r = kvm_init_mmu_notifier(kvm);
993 goto out_err_no_mmu_notifier;
995 r = kvm_arch_post_init_vm(kvm);
999 mutex_lock(&kvm_lock);
1000 list_add(&kvm->vm_list, &vm_list);
1001 mutex_unlock(&kvm_lock);
1003 preempt_notifier_inc();
1008 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1009 if (kvm->mmu_notifier.ops)
1010 mmu_notifier_unregister(&kvm->mmu_notifier, current->mm);
1012 out_err_no_mmu_notifier:
1013 hardware_disable_all();
1015 kvm_arch_destroy_vm(kvm);
1016 out_err_no_arch_destroy_vm:
1017 WARN_ON_ONCE(!refcount_dec_and_test(&kvm->users_count));
1018 for (i = 0; i < KVM_NR_BUSES; i++)
1019 kfree(kvm_get_bus(kvm, i));
1020 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
1021 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
1022 cleanup_srcu_struct(&kvm->irq_srcu);
1023 out_err_no_irq_srcu:
1024 cleanup_srcu_struct(&kvm->srcu);
1026 kvm_arch_free_vm(kvm);
1027 mmdrop(current->mm);
1031 static void kvm_destroy_devices(struct kvm *kvm)
1033 struct kvm_device *dev, *tmp;
1036 * We do not need to take the kvm->lock here, because nobody else
1037 * has a reference to the struct kvm at this point and therefore
1038 * cannot access the devices list anyhow.
1040 list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
1041 list_del(&dev->vm_node);
1042 dev->ops->destroy(dev);
1046 static void kvm_destroy_vm(struct kvm *kvm)
1049 struct mm_struct *mm = kvm->mm;
1051 kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm);
1052 kvm_destroy_vm_debugfs(kvm);
1053 kvm_arch_sync_events(kvm);
1054 mutex_lock(&kvm_lock);
1055 list_del(&kvm->vm_list);
1056 mutex_unlock(&kvm_lock);
1057 kvm_arch_pre_destroy_vm(kvm);
1059 kvm_free_irq_routing(kvm);
1060 for (i = 0; i < KVM_NR_BUSES; i++) {
1061 struct kvm_io_bus *bus = kvm_get_bus(kvm, i);
1064 kvm_io_bus_destroy(bus);
1065 kvm->buses[i] = NULL;
1067 kvm_coalesced_mmio_free(kvm);
1068 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1069 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
1071 kvm_arch_flush_shadow_all(kvm);
1073 kvm_arch_destroy_vm(kvm);
1074 kvm_destroy_devices(kvm);
1075 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
1076 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
1077 cleanup_srcu_struct(&kvm->irq_srcu);
1078 cleanup_srcu_struct(&kvm->srcu);
1079 kvm_arch_free_vm(kvm);
1080 preempt_notifier_dec();
1081 hardware_disable_all();
1085 void kvm_get_kvm(struct kvm *kvm)
1087 refcount_inc(&kvm->users_count);
1089 EXPORT_SYMBOL_GPL(kvm_get_kvm);
1091 void kvm_put_kvm(struct kvm *kvm)
1093 if (refcount_dec_and_test(&kvm->users_count))
1094 kvm_destroy_vm(kvm);
1096 EXPORT_SYMBOL_GPL(kvm_put_kvm);
1099 * Used to put a reference that was taken on behalf of an object associated
1100 * with a user-visible file descriptor, e.g. a vcpu or device, if installation
1101 * of the new file descriptor fails and the reference cannot be transferred to
1102 * its final owner. In such cases, the caller is still actively using @kvm and
1103 * will fail miserably if the refcount unexpectedly hits zero.
1105 void kvm_put_kvm_no_destroy(struct kvm *kvm)
1107 WARN_ON(refcount_dec_and_test(&kvm->users_count));
1109 EXPORT_SYMBOL_GPL(kvm_put_kvm_no_destroy);
1111 static int kvm_vm_release(struct inode *inode, struct file *filp)
1113 struct kvm *kvm = filp->private_data;
1115 kvm_irqfd_release(kvm);
1122 * Allocation size is twice as large as the actual dirty bitmap size.
1123 * See kvm_vm_ioctl_get_dirty_log() why this is needed.
1125 static int kvm_alloc_dirty_bitmap(struct kvm_memory_slot *memslot)
1127 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
1129 memslot->dirty_bitmap = kvzalloc(dirty_bytes, GFP_KERNEL_ACCOUNT);
1130 if (!memslot->dirty_bitmap)
1137 * Delete a memslot by decrementing the number of used slots and shifting all
1138 * other entries in the array forward one spot.
1140 static inline void kvm_memslot_delete(struct kvm_memslots *slots,
1141 struct kvm_memory_slot *memslot)
1143 struct kvm_memory_slot *mslots = slots->memslots;
1146 if (WARN_ON(slots->id_to_index[memslot->id] == -1))
1149 slots->used_slots--;
1151 if (atomic_read(&slots->lru_slot) >= slots->used_slots)
1152 atomic_set(&slots->lru_slot, 0);
1154 for (i = slots->id_to_index[memslot->id]; i < slots->used_slots; i++) {
1155 mslots[i] = mslots[i + 1];
1156 slots->id_to_index[mslots[i].id] = i;
1158 mslots[i] = *memslot;
1159 slots->id_to_index[memslot->id] = -1;
1163 * "Insert" a new memslot by incrementing the number of used slots. Returns
1164 * the new slot's initial index into the memslots array.
1166 static inline int kvm_memslot_insert_back(struct kvm_memslots *slots)
1168 return slots->used_slots++;
1172 * Move a changed memslot backwards in the array by shifting existing slots
1173 * with a higher GFN toward the front of the array. Note, the changed memslot
1174 * itself is not preserved in the array, i.e. not swapped at this time, only
1175 * its new index into the array is tracked. Returns the changed memslot's
1176 * current index into the memslots array.
1178 static inline int kvm_memslot_move_backward(struct kvm_memslots *slots,
1179 struct kvm_memory_slot *memslot)
1181 struct kvm_memory_slot *mslots = slots->memslots;
1184 if (WARN_ON_ONCE(slots->id_to_index[memslot->id] == -1) ||
1185 WARN_ON_ONCE(!slots->used_slots))
1189 * Move the target memslot backward in the array by shifting existing
1190 * memslots with a higher GFN (than the target memslot) towards the
1191 * front of the array.
1193 for (i = slots->id_to_index[memslot->id]; i < slots->used_slots - 1; i++) {
1194 if (memslot->base_gfn > mslots[i + 1].base_gfn)
1197 WARN_ON_ONCE(memslot->base_gfn == mslots[i + 1].base_gfn);
1199 /* Shift the next memslot forward one and update its index. */
1200 mslots[i] = mslots[i + 1];
1201 slots->id_to_index[mslots[i].id] = i;
1207 * Move a changed memslot forwards in the array by shifting existing slots with
1208 * a lower GFN toward the back of the array. Note, the changed memslot itself
1209 * is not preserved in the array, i.e. not swapped at this time, only its new
1210 * index into the array is tracked. Returns the changed memslot's final index
1211 * into the memslots array.
1213 static inline int kvm_memslot_move_forward(struct kvm_memslots *slots,
1214 struct kvm_memory_slot *memslot,
1217 struct kvm_memory_slot *mslots = slots->memslots;
1220 for (i = start; i > 0; i--) {
1221 if (memslot->base_gfn < mslots[i - 1].base_gfn)
1224 WARN_ON_ONCE(memslot->base_gfn == mslots[i - 1].base_gfn);
1226 /* Shift the next memslot back one and update its index. */
1227 mslots[i] = mslots[i - 1];
1228 slots->id_to_index[mslots[i].id] = i;
1234 * Re-sort memslots based on their GFN to account for an added, deleted, or
1235 * moved memslot. Sorting memslots by GFN allows using a binary search during
1238 * IMPORTANT: Slots are sorted from highest GFN to lowest GFN! I.e. the entry
1239 * at memslots[0] has the highest GFN.
1241 * The sorting algorithm takes advantage of having initially sorted memslots
1242 * and knowing the position of the changed memslot. Sorting is also optimized
1243 * by not swapping the updated memslot and instead only shifting other memslots
1244 * and tracking the new index for the update memslot. Only once its final
1245 * index is known is the updated memslot copied into its position in the array.
1247 * - When deleting a memslot, the deleted memslot simply needs to be moved to
1248 * the end of the array.
1250 * - When creating a memslot, the algorithm "inserts" the new memslot at the
1251 * end of the array and then it forward to its correct location.
1253 * - When moving a memslot, the algorithm first moves the updated memslot
1254 * backward to handle the scenario where the memslot's GFN was changed to a
1255 * lower value. update_memslots() then falls through and runs the same flow
1256 * as creating a memslot to move the memslot forward to handle the scenario
1257 * where its GFN was changed to a higher value.
1259 * Note, slots are sorted from highest->lowest instead of lowest->highest for
1260 * historical reasons. Originally, invalid memslots where denoted by having
1261 * GFN=0, thus sorting from highest->lowest naturally sorted invalid memslots
1262 * to the end of the array. The current algorithm uses dedicated logic to
1263 * delete a memslot and thus does not rely on invalid memslots having GFN=0.
1265 * The other historical motiviation for highest->lowest was to improve the
1266 * performance of memslot lookup. KVM originally used a linear search starting
1267 * at memslots[0]. On x86, the largest memslot usually has one of the highest,
1268 * if not *the* highest, GFN, as the bulk of the guest's RAM is located in a
1269 * single memslot above the 4gb boundary. As the largest memslot is also the
1270 * most likely to be referenced, sorting it to the front of the array was
1271 * advantageous. The current binary search starts from the middle of the array
1272 * and uses an LRU pointer to improve performance for all memslots and GFNs.
1274 static void update_memslots(struct kvm_memslots *slots,
1275 struct kvm_memory_slot *memslot,
1276 enum kvm_mr_change change)
1280 if (change == KVM_MR_DELETE) {
1281 kvm_memslot_delete(slots, memslot);
1283 if (change == KVM_MR_CREATE)
1284 i = kvm_memslot_insert_back(slots);
1286 i = kvm_memslot_move_backward(slots, memslot);
1287 i = kvm_memslot_move_forward(slots, memslot, i);
1290 * Copy the memslot to its new position in memslots and update
1291 * its index accordingly.
1293 slots->memslots[i] = *memslot;
1294 slots->id_to_index[memslot->id] = i;
1298 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
1300 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
1302 #ifdef __KVM_HAVE_READONLY_MEM
1303 valid_flags |= KVM_MEM_READONLY;
1306 if (mem->flags & ~valid_flags)
1312 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
1313 int as_id, struct kvm_memslots *slots)
1315 struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id);
1316 u64 gen = old_memslots->generation;
1318 WARN_ON(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS);
1319 slots->generation = gen | KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
1321 rcu_assign_pointer(kvm->memslots[as_id], slots);
1322 synchronize_srcu_expedited(&kvm->srcu);
1325 * Increment the new memslot generation a second time, dropping the
1326 * update in-progress flag and incrementing the generation based on
1327 * the number of address spaces. This provides a unique and easily
1328 * identifiable generation number while the memslots are in flux.
1330 gen = slots->generation & ~KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
1333 * Generations must be unique even across address spaces. We do not need
1334 * a global counter for that, instead the generation space is evenly split
1335 * across address spaces. For example, with two address spaces, address
1336 * space 0 will use generations 0, 2, 4, ... while address space 1 will
1337 * use generations 1, 3, 5, ...
1339 gen += KVM_ADDRESS_SPACE_NUM;
1341 kvm_arch_memslots_updated(kvm, gen);
1343 slots->generation = gen;
1345 return old_memslots;
1349 * Note, at a minimum, the current number of used slots must be allocated, even
1350 * when deleting a memslot, as we need a complete duplicate of the memslots for
1351 * use when invalidating a memslot prior to deleting/moving the memslot.
1353 static struct kvm_memslots *kvm_dup_memslots(struct kvm_memslots *old,
1354 enum kvm_mr_change change)
1356 struct kvm_memslots *slots;
1357 size_t old_size, new_size;
1359 old_size = sizeof(struct kvm_memslots) +
1360 (sizeof(struct kvm_memory_slot) * old->used_slots);
1362 if (change == KVM_MR_CREATE)
1363 new_size = old_size + sizeof(struct kvm_memory_slot);
1365 new_size = old_size;
1367 slots = kvzalloc(new_size, GFP_KERNEL_ACCOUNT);
1369 memcpy(slots, old, old_size);
1374 static int kvm_set_memslot(struct kvm *kvm,
1375 const struct kvm_userspace_memory_region *mem,
1376 struct kvm_memory_slot *old,
1377 struct kvm_memory_slot *new, int as_id,
1378 enum kvm_mr_change change)
1380 struct kvm_memory_slot *slot;
1381 struct kvm_memslots *slots;
1384 slots = kvm_dup_memslots(__kvm_memslots(kvm, as_id), change);
1388 if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) {
1390 * Note, the INVALID flag needs to be in the appropriate entry
1391 * in the freshly allocated memslots, not in @old or @new.
1393 slot = id_to_memslot(slots, old->id);
1394 slot->flags |= KVM_MEMSLOT_INVALID;
1397 * We can re-use the old memslots, the only difference from the
1398 * newly installed memslots is the invalid flag, which will get
1399 * dropped by update_memslots anyway. We'll also revert to the
1400 * old memslots if preparing the new memory region fails.
1402 slots = install_new_memslots(kvm, as_id, slots);
1404 /* From this point no new shadow pages pointing to a deleted,
1405 * or moved, memslot will be created.
1407 * validation of sp->gfn happens in:
1408 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1409 * - kvm_is_visible_gfn (mmu_check_root)
1411 kvm_arch_flush_shadow_memslot(kvm, slot);
1414 r = kvm_arch_prepare_memory_region(kvm, new, mem, change);
1418 update_memslots(slots, new, change);
1419 slots = install_new_memslots(kvm, as_id, slots);
1421 kvm_arch_commit_memory_region(kvm, mem, old, new, change);
1427 if (change == KVM_MR_DELETE || change == KVM_MR_MOVE)
1428 slots = install_new_memslots(kvm, as_id, slots);
1433 static int kvm_delete_memslot(struct kvm *kvm,
1434 const struct kvm_userspace_memory_region *mem,
1435 struct kvm_memory_slot *old, int as_id)
1437 struct kvm_memory_slot new;
1443 memset(&new, 0, sizeof(new));
1446 * This is only for debugging purpose; it should never be referenced
1447 * for a removed memslot.
1451 r = kvm_set_memslot(kvm, mem, old, &new, as_id, KVM_MR_DELETE);
1455 kvm_free_memslot(kvm, old);
1460 * Allocate some memory and give it an address in the guest physical address
1463 * Discontiguous memory is allowed, mostly for framebuffers.
1465 * Must be called holding kvm->slots_lock for write.
1467 int __kvm_set_memory_region(struct kvm *kvm,
1468 const struct kvm_userspace_memory_region *mem)
1470 struct kvm_memory_slot old, new;
1471 struct kvm_memory_slot *tmp;
1472 enum kvm_mr_change change;
1476 r = check_memory_region_flags(mem);
1480 as_id = mem->slot >> 16;
1481 id = (u16)mem->slot;
1483 /* General sanity checks */
1484 if (mem->memory_size & (PAGE_SIZE - 1))
1486 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
1488 /* We can read the guest memory with __xxx_user() later on. */
1489 if ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
1490 (mem->userspace_addr != untagged_addr(mem->userspace_addr)) ||
1491 !access_ok((void __user *)(unsigned long)mem->userspace_addr,
1494 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
1496 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
1500 * Make a full copy of the old memslot, the pointer will become stale
1501 * when the memslots are re-sorted by update_memslots(), and the old
1502 * memslot needs to be referenced after calling update_memslots(), e.g.
1503 * to free its resources and for arch specific behavior.
1505 tmp = id_to_memslot(__kvm_memslots(kvm, as_id), id);
1510 memset(&old, 0, sizeof(old));
1514 if (!mem->memory_size)
1515 return kvm_delete_memslot(kvm, mem, &old, as_id);
1519 new.base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
1520 new.npages = mem->memory_size >> PAGE_SHIFT;
1521 new.flags = mem->flags;
1522 new.userspace_addr = mem->userspace_addr;
1524 if (new.npages > KVM_MEM_MAX_NR_PAGES)
1528 change = KVM_MR_CREATE;
1529 new.dirty_bitmap = NULL;
1530 memset(&new.arch, 0, sizeof(new.arch));
1531 } else { /* Modify an existing slot. */
1532 if ((new.userspace_addr != old.userspace_addr) ||
1533 (new.npages != old.npages) ||
1534 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
1537 if (new.base_gfn != old.base_gfn)
1538 change = KVM_MR_MOVE;
1539 else if (new.flags != old.flags)
1540 change = KVM_MR_FLAGS_ONLY;
1541 else /* Nothing to change. */
1544 /* Copy dirty_bitmap and arch from the current memslot. */
1545 new.dirty_bitmap = old.dirty_bitmap;
1546 memcpy(&new.arch, &old.arch, sizeof(new.arch));
1549 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
1550 /* Check for overlaps */
1551 kvm_for_each_memslot(tmp, __kvm_memslots(kvm, as_id)) {
1554 if (!((new.base_gfn + new.npages <= tmp->base_gfn) ||
1555 (new.base_gfn >= tmp->base_gfn + tmp->npages)))
1560 /* Allocate/free page dirty bitmap as needed */
1561 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
1562 new.dirty_bitmap = NULL;
1563 else if (!new.dirty_bitmap && !kvm->dirty_ring_size) {
1564 r = kvm_alloc_dirty_bitmap(&new);
1568 if (kvm_dirty_log_manual_protect_and_init_set(kvm))
1569 bitmap_set(new.dirty_bitmap, 0, new.npages);
1572 r = kvm_set_memslot(kvm, mem, &old, &new, as_id, change);
1576 if (old.dirty_bitmap && !new.dirty_bitmap)
1577 kvm_destroy_dirty_bitmap(&old);
1581 if (new.dirty_bitmap && !old.dirty_bitmap)
1582 kvm_destroy_dirty_bitmap(&new);
1585 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1587 int kvm_set_memory_region(struct kvm *kvm,
1588 const struct kvm_userspace_memory_region *mem)
1592 mutex_lock(&kvm->slots_lock);
1593 r = __kvm_set_memory_region(kvm, mem);
1594 mutex_unlock(&kvm->slots_lock);
1597 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1599 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1600 struct kvm_userspace_memory_region *mem)
1602 if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1605 return kvm_set_memory_region(kvm, mem);
1608 #ifndef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1610 * kvm_get_dirty_log - get a snapshot of dirty pages
1611 * @kvm: pointer to kvm instance
1612 * @log: slot id and address to which we copy the log
1613 * @is_dirty: set to '1' if any dirty pages were found
1614 * @memslot: set to the associated memslot, always valid on success
1616 int kvm_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log,
1617 int *is_dirty, struct kvm_memory_slot **memslot)
1619 struct kvm_memslots *slots;
1622 unsigned long any = 0;
1624 /* Dirty ring tracking is exclusive to dirty log tracking */
1625 if (kvm->dirty_ring_size)
1631 as_id = log->slot >> 16;
1632 id = (u16)log->slot;
1633 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1636 slots = __kvm_memslots(kvm, as_id);
1637 *memslot = id_to_memslot(slots, id);
1638 if (!(*memslot) || !(*memslot)->dirty_bitmap)
1641 kvm_arch_sync_dirty_log(kvm, *memslot);
1643 n = kvm_dirty_bitmap_bytes(*memslot);
1645 for (i = 0; !any && i < n/sizeof(long); ++i)
1646 any = (*memslot)->dirty_bitmap[i];
1648 if (copy_to_user(log->dirty_bitmap, (*memslot)->dirty_bitmap, n))
1655 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1657 #else /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
1659 * kvm_get_dirty_log_protect - get a snapshot of dirty pages
1660 * and reenable dirty page tracking for the corresponding pages.
1661 * @kvm: pointer to kvm instance
1662 * @log: slot id and address to which we copy the log
1664 * We need to keep it in mind that VCPU threads can write to the bitmap
1665 * concurrently. So, to avoid losing track of dirty pages we keep the
1668 * 1. Take a snapshot of the bit and clear it if needed.
1669 * 2. Write protect the corresponding page.
1670 * 3. Copy the snapshot to the userspace.
1671 * 4. Upon return caller flushes TLB's if needed.
1673 * Between 2 and 4, the guest may write to the page using the remaining TLB
1674 * entry. This is not a problem because the page is reported dirty using
1675 * the snapshot taken before and step 4 ensures that writes done after
1676 * exiting to userspace will be logged for the next call.
1679 static int kvm_get_dirty_log_protect(struct kvm *kvm, struct kvm_dirty_log *log)
1681 struct kvm_memslots *slots;
1682 struct kvm_memory_slot *memslot;
1685 unsigned long *dirty_bitmap;
1686 unsigned long *dirty_bitmap_buffer;
1689 /* Dirty ring tracking is exclusive to dirty log tracking */
1690 if (kvm->dirty_ring_size)
1693 as_id = log->slot >> 16;
1694 id = (u16)log->slot;
1695 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1698 slots = __kvm_memslots(kvm, as_id);
1699 memslot = id_to_memslot(slots, id);
1700 if (!memslot || !memslot->dirty_bitmap)
1703 dirty_bitmap = memslot->dirty_bitmap;
1705 kvm_arch_sync_dirty_log(kvm, memslot);
1707 n = kvm_dirty_bitmap_bytes(memslot);
1709 if (kvm->manual_dirty_log_protect) {
1711 * Unlike kvm_get_dirty_log, we always return false in *flush,
1712 * because no flush is needed until KVM_CLEAR_DIRTY_LOG. There
1713 * is some code duplication between this function and
1714 * kvm_get_dirty_log, but hopefully all architecture
1715 * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log
1716 * can be eliminated.
1718 dirty_bitmap_buffer = dirty_bitmap;
1720 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
1721 memset(dirty_bitmap_buffer, 0, n);
1724 for (i = 0; i < n / sizeof(long); i++) {
1728 if (!dirty_bitmap[i])
1732 mask = xchg(&dirty_bitmap[i], 0);
1733 dirty_bitmap_buffer[i] = mask;
1735 offset = i * BITS_PER_LONG;
1736 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1739 KVM_MMU_UNLOCK(kvm);
1743 kvm_arch_flush_remote_tlbs_memslot(kvm, memslot);
1745 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1752 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
1753 * @kvm: kvm instance
1754 * @log: slot id and address to which we copy the log
1756 * Steps 1-4 below provide general overview of dirty page logging. See
1757 * kvm_get_dirty_log_protect() function description for additional details.
1759 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
1760 * always flush the TLB (step 4) even if previous step failed and the dirty
1761 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
1762 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
1763 * writes will be marked dirty for next log read.
1765 * 1. Take a snapshot of the bit and clear it if needed.
1766 * 2. Write protect the corresponding page.
1767 * 3. Copy the snapshot to the userspace.
1768 * 4. Flush TLB's if needed.
1770 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1771 struct kvm_dirty_log *log)
1775 mutex_lock(&kvm->slots_lock);
1777 r = kvm_get_dirty_log_protect(kvm, log);
1779 mutex_unlock(&kvm->slots_lock);
1784 * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap
1785 * and reenable dirty page tracking for the corresponding pages.
1786 * @kvm: pointer to kvm instance
1787 * @log: slot id and address from which to fetch the bitmap of dirty pages
1789 static int kvm_clear_dirty_log_protect(struct kvm *kvm,
1790 struct kvm_clear_dirty_log *log)
1792 struct kvm_memslots *slots;
1793 struct kvm_memory_slot *memslot;
1797 unsigned long *dirty_bitmap;
1798 unsigned long *dirty_bitmap_buffer;
1801 /* Dirty ring tracking is exclusive to dirty log tracking */
1802 if (kvm->dirty_ring_size)
1805 as_id = log->slot >> 16;
1806 id = (u16)log->slot;
1807 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1810 if (log->first_page & 63)
1813 slots = __kvm_memslots(kvm, as_id);
1814 memslot = id_to_memslot(slots, id);
1815 if (!memslot || !memslot->dirty_bitmap)
1818 dirty_bitmap = memslot->dirty_bitmap;
1820 n = ALIGN(log->num_pages, BITS_PER_LONG) / 8;
1822 if (log->first_page > memslot->npages ||
1823 log->num_pages > memslot->npages - log->first_page ||
1824 (log->num_pages < memslot->npages - log->first_page && (log->num_pages & 63)))
1827 kvm_arch_sync_dirty_log(kvm, memslot);
1830 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
1831 if (copy_from_user(dirty_bitmap_buffer, log->dirty_bitmap, n))
1835 for (offset = log->first_page, i = offset / BITS_PER_LONG,
1836 n = DIV_ROUND_UP(log->num_pages, BITS_PER_LONG); n--;
1837 i++, offset += BITS_PER_LONG) {
1838 unsigned long mask = *dirty_bitmap_buffer++;
1839 atomic_long_t *p = (atomic_long_t *) &dirty_bitmap[i];
1843 mask &= atomic_long_fetch_andnot(mask, p);
1846 * mask contains the bits that really have been cleared. This
1847 * never includes any bits beyond the length of the memslot (if
1848 * the length is not aligned to 64 pages), therefore it is not
1849 * a problem if userspace sets them in log->dirty_bitmap.
1853 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1857 KVM_MMU_UNLOCK(kvm);
1860 kvm_arch_flush_remote_tlbs_memslot(kvm, memslot);
1865 static int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm,
1866 struct kvm_clear_dirty_log *log)
1870 mutex_lock(&kvm->slots_lock);
1872 r = kvm_clear_dirty_log_protect(kvm, log);
1874 mutex_unlock(&kvm->slots_lock);
1877 #endif /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
1879 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1881 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1883 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1885 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
1887 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn);
1889 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_memslot);
1891 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1893 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1895 return kvm_is_visible_memslot(memslot);
1897 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1899 bool kvm_vcpu_is_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
1901 struct kvm_memory_slot *memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1903 return kvm_is_visible_memslot(memslot);
1905 EXPORT_SYMBOL_GPL(kvm_vcpu_is_visible_gfn);
1907 unsigned long kvm_host_page_size(struct kvm_vcpu *vcpu, gfn_t gfn)
1909 struct vm_area_struct *vma;
1910 unsigned long addr, size;
1914 addr = kvm_vcpu_gfn_to_hva_prot(vcpu, gfn, NULL);
1915 if (kvm_is_error_hva(addr))
1918 mmap_read_lock(current->mm);
1919 vma = find_vma(current->mm, addr);
1923 size = vma_kernel_pagesize(vma);
1926 mmap_read_unlock(current->mm);
1931 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1933 return slot->flags & KVM_MEM_READONLY;
1936 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1937 gfn_t *nr_pages, bool write)
1939 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1940 return KVM_HVA_ERR_BAD;
1942 if (memslot_is_readonly(slot) && write)
1943 return KVM_HVA_ERR_RO_BAD;
1946 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1948 return __gfn_to_hva_memslot(slot, gfn);
1951 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1954 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1957 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1960 return gfn_to_hva_many(slot, gfn, NULL);
1962 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1964 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1966 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1968 EXPORT_SYMBOL_GPL(gfn_to_hva);
1970 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
1972 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
1974 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
1977 * Return the hva of a @gfn and the R/W attribute if possible.
1979 * @slot: the kvm_memory_slot which contains @gfn
1980 * @gfn: the gfn to be translated
1981 * @writable: used to return the read/write attribute of the @slot if the hva
1982 * is valid and @writable is not NULL
1984 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1985 gfn_t gfn, bool *writable)
1987 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1989 if (!kvm_is_error_hva(hva) && writable)
1990 *writable = !memslot_is_readonly(slot);
1995 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1997 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1999 return gfn_to_hva_memslot_prot(slot, gfn, writable);
2002 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
2004 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2006 return gfn_to_hva_memslot_prot(slot, gfn, writable);
2009 static inline int check_user_page_hwpoison(unsigned long addr)
2011 int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
2013 rc = get_user_pages(addr, 1, flags, NULL, NULL);
2014 return rc == -EHWPOISON;
2018 * The fast path to get the writable pfn which will be stored in @pfn,
2019 * true indicates success, otherwise false is returned. It's also the
2020 * only part that runs if we can in atomic context.
2022 static bool hva_to_pfn_fast(unsigned long addr, bool write_fault,
2023 bool *writable, kvm_pfn_t *pfn)
2025 struct page *page[1];
2028 * Fast pin a writable pfn only if it is a write fault request
2029 * or the caller allows to map a writable pfn for a read fault
2032 if (!(write_fault || writable))
2035 if (get_user_page_fast_only(addr, FOLL_WRITE, page)) {
2036 *pfn = page_to_pfn(page[0]);
2047 * The slow path to get the pfn of the specified host virtual address,
2048 * 1 indicates success, -errno is returned if error is detected.
2050 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
2051 bool *writable, kvm_pfn_t *pfn)
2053 unsigned int flags = FOLL_HWPOISON;
2060 *writable = write_fault;
2063 flags |= FOLL_WRITE;
2065 flags |= FOLL_NOWAIT;
2067 npages = get_user_pages_unlocked(addr, 1, &page, flags);
2071 /* map read fault as writable if possible */
2072 if (unlikely(!write_fault) && writable) {
2075 if (get_user_page_fast_only(addr, FOLL_WRITE, &wpage)) {
2081 *pfn = page_to_pfn(page);
2085 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
2087 if (unlikely(!(vma->vm_flags & VM_READ)))
2090 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
2096 static int hva_to_pfn_remapped(struct vm_area_struct *vma,
2097 unsigned long addr, bool *async,
2098 bool write_fault, bool *writable,
2106 r = follow_pte(vma->vm_mm, addr, &ptep, &ptl);
2109 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
2110 * not call the fault handler, so do it here.
2112 bool unlocked = false;
2113 r = fixup_user_fault(current->mm, addr,
2114 (write_fault ? FAULT_FLAG_WRITE : 0),
2121 r = follow_pte(vma->vm_mm, addr, &ptep, &ptl);
2126 if (write_fault && !pte_write(*ptep)) {
2127 pfn = KVM_PFN_ERR_RO_FAULT;
2132 *writable = pte_write(*ptep);
2133 pfn = pte_pfn(*ptep);
2136 * Get a reference here because callers of *hva_to_pfn* and
2137 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
2138 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
2139 * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will
2140 * simply do nothing for reserved pfns.
2142 * Whoever called remap_pfn_range is also going to call e.g.
2143 * unmap_mapping_range before the underlying pages are freed,
2144 * causing a call to our MMU notifier.
2149 pte_unmap_unlock(ptep, ptl);
2155 * Pin guest page in memory and return its pfn.
2156 * @addr: host virtual address which maps memory to the guest
2157 * @atomic: whether this function can sleep
2158 * @async: whether this function need to wait IO complete if the
2159 * host page is not in the memory
2160 * @write_fault: whether we should get a writable host page
2161 * @writable: whether it allows to map a writable host page for !@write_fault
2163 * The function will map a writable host page for these two cases:
2164 * 1): @write_fault = true
2165 * 2): @write_fault = false && @writable, @writable will tell the caller
2166 * whether the mapping is writable.
2168 static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
2169 bool write_fault, bool *writable)
2171 struct vm_area_struct *vma;
2175 /* we can do it either atomically or asynchronously, not both */
2176 BUG_ON(atomic && async);
2178 if (hva_to_pfn_fast(addr, write_fault, writable, &pfn))
2182 return KVM_PFN_ERR_FAULT;
2184 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
2188 mmap_read_lock(current->mm);
2189 if (npages == -EHWPOISON ||
2190 (!async && check_user_page_hwpoison(addr))) {
2191 pfn = KVM_PFN_ERR_HWPOISON;
2196 vma = find_vma_intersection(current->mm, addr, addr + 1);
2199 pfn = KVM_PFN_ERR_FAULT;
2200 else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
2201 r = hva_to_pfn_remapped(vma, addr, async, write_fault, writable, &pfn);
2205 pfn = KVM_PFN_ERR_FAULT;
2207 if (async && vma_is_valid(vma, write_fault))
2209 pfn = KVM_PFN_ERR_FAULT;
2212 mmap_read_unlock(current->mm);
2216 kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn,
2217 bool atomic, bool *async, bool write_fault,
2218 bool *writable, hva_t *hva)
2220 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
2225 if (addr == KVM_HVA_ERR_RO_BAD) {
2228 return KVM_PFN_ERR_RO_FAULT;
2231 if (kvm_is_error_hva(addr)) {
2234 return KVM_PFN_NOSLOT;
2237 /* Do not map writable pfn in the readonly memslot. */
2238 if (writable && memslot_is_readonly(slot)) {
2243 return hva_to_pfn(addr, atomic, async, write_fault,
2246 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
2248 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
2251 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
2252 write_fault, writable, NULL);
2254 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
2256 kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
2258 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL, NULL);
2260 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
2262 kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
2264 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL, NULL);
2266 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
2268 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
2270 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
2272 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
2274 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
2276 return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
2278 EXPORT_SYMBOL_GPL(gfn_to_pfn);
2280 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
2282 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
2284 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
2286 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
2287 struct page **pages, int nr_pages)
2292 addr = gfn_to_hva_many(slot, gfn, &entry);
2293 if (kvm_is_error_hva(addr))
2296 if (entry < nr_pages)
2299 return get_user_pages_fast_only(addr, nr_pages, FOLL_WRITE, pages);
2301 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
2303 static struct page *kvm_pfn_to_page(kvm_pfn_t pfn)
2305 if (is_error_noslot_pfn(pfn))
2306 return KVM_ERR_PTR_BAD_PAGE;
2308 if (kvm_is_reserved_pfn(pfn)) {
2310 return KVM_ERR_PTR_BAD_PAGE;
2313 return pfn_to_page(pfn);
2316 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
2320 pfn = gfn_to_pfn(kvm, gfn);
2322 return kvm_pfn_to_page(pfn);
2324 EXPORT_SYMBOL_GPL(gfn_to_page);
2326 void kvm_release_pfn(kvm_pfn_t pfn, bool dirty, struct gfn_to_pfn_cache *cache)
2332 cache->pfn = cache->gfn = 0;
2335 kvm_release_pfn_dirty(pfn);
2337 kvm_release_pfn_clean(pfn);
2340 static void kvm_cache_gfn_to_pfn(struct kvm_memory_slot *slot, gfn_t gfn,
2341 struct gfn_to_pfn_cache *cache, u64 gen)
2343 kvm_release_pfn(cache->pfn, cache->dirty, cache);
2345 cache->pfn = gfn_to_pfn_memslot(slot, gfn);
2347 cache->dirty = false;
2348 cache->generation = gen;
2351 static int __kvm_map_gfn(struct kvm_memslots *slots, gfn_t gfn,
2352 struct kvm_host_map *map,
2353 struct gfn_to_pfn_cache *cache,
2358 struct page *page = KVM_UNMAPPED_PAGE;
2359 struct kvm_memory_slot *slot = __gfn_to_memslot(slots, gfn);
2360 u64 gen = slots->generation;
2366 if (!cache->pfn || cache->gfn != gfn ||
2367 cache->generation != gen) {
2370 kvm_cache_gfn_to_pfn(slot, gfn, cache, gen);
2376 pfn = gfn_to_pfn_memslot(slot, gfn);
2378 if (is_error_noslot_pfn(pfn))
2381 if (pfn_valid(pfn)) {
2382 page = pfn_to_page(pfn);
2384 hva = kmap_atomic(page);
2387 #ifdef CONFIG_HAS_IOMEM
2388 } else if (!atomic) {
2389 hva = memremap(pfn_to_hpa(pfn), PAGE_SIZE, MEMREMAP_WB);
2406 int kvm_map_gfn(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map,
2407 struct gfn_to_pfn_cache *cache, bool atomic)
2409 return __kvm_map_gfn(kvm_memslots(vcpu->kvm), gfn, map,
2412 EXPORT_SYMBOL_GPL(kvm_map_gfn);
2414 int kvm_vcpu_map(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map)
2416 return __kvm_map_gfn(kvm_vcpu_memslots(vcpu), gfn, map,
2419 EXPORT_SYMBOL_GPL(kvm_vcpu_map);
2421 static void __kvm_unmap_gfn(struct kvm *kvm,
2422 struct kvm_memory_slot *memslot,
2423 struct kvm_host_map *map,
2424 struct gfn_to_pfn_cache *cache,
2425 bool dirty, bool atomic)
2433 if (map->page != KVM_UNMAPPED_PAGE) {
2435 kunmap_atomic(map->hva);
2439 #ifdef CONFIG_HAS_IOMEM
2443 WARN_ONCE(1, "Unexpected unmapping in atomic context");
2447 mark_page_dirty_in_slot(kvm, memslot, map->gfn);
2450 cache->dirty |= dirty;
2452 kvm_release_pfn(map->pfn, dirty, NULL);
2458 int kvm_unmap_gfn(struct kvm_vcpu *vcpu, struct kvm_host_map *map,
2459 struct gfn_to_pfn_cache *cache, bool dirty, bool atomic)
2461 __kvm_unmap_gfn(vcpu->kvm, gfn_to_memslot(vcpu->kvm, map->gfn), map,
2462 cache, dirty, atomic);
2465 EXPORT_SYMBOL_GPL(kvm_unmap_gfn);
2467 void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map, bool dirty)
2469 __kvm_unmap_gfn(vcpu->kvm, kvm_vcpu_gfn_to_memslot(vcpu, map->gfn),
2470 map, NULL, dirty, false);
2472 EXPORT_SYMBOL_GPL(kvm_vcpu_unmap);
2474 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
2478 pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
2480 return kvm_pfn_to_page(pfn);
2482 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
2484 void kvm_release_page_clean(struct page *page)
2486 WARN_ON(is_error_page(page));
2488 kvm_release_pfn_clean(page_to_pfn(page));
2490 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
2492 void kvm_release_pfn_clean(kvm_pfn_t pfn)
2494 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
2495 put_page(pfn_to_page(pfn));
2497 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
2499 void kvm_release_page_dirty(struct page *page)
2501 WARN_ON(is_error_page(page));
2503 kvm_release_pfn_dirty(page_to_pfn(page));
2505 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
2507 void kvm_release_pfn_dirty(kvm_pfn_t pfn)
2509 kvm_set_pfn_dirty(pfn);
2510 kvm_release_pfn_clean(pfn);
2512 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
2514 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
2516 if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn))
2517 SetPageDirty(pfn_to_page(pfn));
2519 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
2521 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
2523 if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn))
2524 mark_page_accessed(pfn_to_page(pfn));
2526 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
2528 void kvm_get_pfn(kvm_pfn_t pfn)
2530 if (!kvm_is_reserved_pfn(pfn))
2531 get_page(pfn_to_page(pfn));
2533 EXPORT_SYMBOL_GPL(kvm_get_pfn);
2535 static int next_segment(unsigned long len, int offset)
2537 if (len > PAGE_SIZE - offset)
2538 return PAGE_SIZE - offset;
2543 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
2544 void *data, int offset, int len)
2549 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
2550 if (kvm_is_error_hva(addr))
2552 r = __copy_from_user(data, (void __user *)addr + offset, len);
2558 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
2561 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2563 return __kvm_read_guest_page(slot, gfn, data, offset, len);
2565 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
2567 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
2568 int offset, int len)
2570 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2572 return __kvm_read_guest_page(slot, gfn, data, offset, len);
2574 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
2576 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
2578 gfn_t gfn = gpa >> PAGE_SHIFT;
2580 int offset = offset_in_page(gpa);
2583 while ((seg = next_segment(len, offset)) != 0) {
2584 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
2594 EXPORT_SYMBOL_GPL(kvm_read_guest);
2596 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
2598 gfn_t gfn = gpa >> PAGE_SHIFT;
2600 int offset = offset_in_page(gpa);
2603 while ((seg = next_segment(len, offset)) != 0) {
2604 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
2614 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
2616 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
2617 void *data, int offset, unsigned long len)
2622 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
2623 if (kvm_is_error_hva(addr))
2625 pagefault_disable();
2626 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
2633 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
2634 void *data, unsigned long len)
2636 gfn_t gfn = gpa >> PAGE_SHIFT;
2637 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2638 int offset = offset_in_page(gpa);
2640 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
2642 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
2644 static int __kvm_write_guest_page(struct kvm *kvm,
2645 struct kvm_memory_slot *memslot, gfn_t gfn,
2646 const void *data, int offset, int len)
2651 addr = gfn_to_hva_memslot(memslot, gfn);
2652 if (kvm_is_error_hva(addr))
2654 r = __copy_to_user((void __user *)addr + offset, data, len);
2657 mark_page_dirty_in_slot(kvm, memslot, gfn);
2661 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
2662 const void *data, int offset, int len)
2664 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2666 return __kvm_write_guest_page(kvm, slot, gfn, data, offset, len);
2668 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
2670 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
2671 const void *data, int offset, int len)
2673 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2675 return __kvm_write_guest_page(vcpu->kvm, slot, gfn, data, offset, len);
2677 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
2679 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
2682 gfn_t gfn = gpa >> PAGE_SHIFT;
2684 int offset = offset_in_page(gpa);
2687 while ((seg = next_segment(len, offset)) != 0) {
2688 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
2698 EXPORT_SYMBOL_GPL(kvm_write_guest);
2700 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
2703 gfn_t gfn = gpa >> PAGE_SHIFT;
2705 int offset = offset_in_page(gpa);
2708 while ((seg = next_segment(len, offset)) != 0) {
2709 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
2719 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
2721 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
2722 struct gfn_to_hva_cache *ghc,
2723 gpa_t gpa, unsigned long len)
2725 int offset = offset_in_page(gpa);
2726 gfn_t start_gfn = gpa >> PAGE_SHIFT;
2727 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
2728 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
2729 gfn_t nr_pages_avail;
2731 /* Update ghc->generation before performing any error checks. */
2732 ghc->generation = slots->generation;
2734 if (start_gfn > end_gfn) {
2735 ghc->hva = KVM_HVA_ERR_BAD;
2740 * If the requested region crosses two memslots, we still
2741 * verify that the entire region is valid here.
2743 for ( ; start_gfn <= end_gfn; start_gfn += nr_pages_avail) {
2744 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
2745 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
2747 if (kvm_is_error_hva(ghc->hva))
2751 /* Use the slow path for cross page reads and writes. */
2752 if (nr_pages_needed == 1)
2755 ghc->memslot = NULL;
2762 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2763 gpa_t gpa, unsigned long len)
2765 struct kvm_memslots *slots = kvm_memslots(kvm);
2766 return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len);
2768 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
2770 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2771 void *data, unsigned int offset,
2774 struct kvm_memslots *slots = kvm_memslots(kvm);
2776 gpa_t gpa = ghc->gpa + offset;
2778 BUG_ON(len + offset > ghc->len);
2780 if (slots->generation != ghc->generation) {
2781 if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len))
2785 if (kvm_is_error_hva(ghc->hva))
2788 if (unlikely(!ghc->memslot))
2789 return kvm_write_guest(kvm, gpa, data, len);
2791 r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
2794 mark_page_dirty_in_slot(kvm, ghc->memslot, gpa >> PAGE_SHIFT);
2798 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached);
2800 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2801 void *data, unsigned long len)
2803 return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len);
2805 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
2807 int kvm_read_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2808 void *data, unsigned int offset,
2811 struct kvm_memslots *slots = kvm_memslots(kvm);
2813 gpa_t gpa = ghc->gpa + offset;
2815 BUG_ON(len + offset > ghc->len);
2817 if (slots->generation != ghc->generation) {
2818 if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len))
2822 if (kvm_is_error_hva(ghc->hva))
2825 if (unlikely(!ghc->memslot))
2826 return kvm_read_guest(kvm, gpa, data, len);
2828 r = __copy_from_user(data, (void __user *)ghc->hva + offset, len);
2834 EXPORT_SYMBOL_GPL(kvm_read_guest_offset_cached);
2836 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2837 void *data, unsigned long len)
2839 return kvm_read_guest_offset_cached(kvm, ghc, data, 0, len);
2841 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
2843 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
2845 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
2846 gfn_t gfn = gpa >> PAGE_SHIFT;
2848 int offset = offset_in_page(gpa);
2851 while ((seg = next_segment(len, offset)) != 0) {
2852 ret = kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
2861 EXPORT_SYMBOL_GPL(kvm_clear_guest);
2863 void mark_page_dirty_in_slot(struct kvm *kvm,
2864 struct kvm_memory_slot *memslot,
2867 if (memslot && kvm_slot_dirty_track_enabled(memslot)) {
2868 unsigned long rel_gfn = gfn - memslot->base_gfn;
2869 u32 slot = (memslot->as_id << 16) | memslot->id;
2871 if (kvm->dirty_ring_size)
2872 kvm_dirty_ring_push(kvm_dirty_ring_get(kvm),
2875 set_bit_le(rel_gfn, memslot->dirty_bitmap);
2878 EXPORT_SYMBOL_GPL(mark_page_dirty_in_slot);
2880 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
2882 struct kvm_memory_slot *memslot;
2884 memslot = gfn_to_memslot(kvm, gfn);
2885 mark_page_dirty_in_slot(kvm, memslot, gfn);
2887 EXPORT_SYMBOL_GPL(mark_page_dirty);
2889 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
2891 struct kvm_memory_slot *memslot;
2893 memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2894 mark_page_dirty_in_slot(vcpu->kvm, memslot, gfn);
2896 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
2898 void kvm_sigset_activate(struct kvm_vcpu *vcpu)
2900 if (!vcpu->sigset_active)
2904 * This does a lockless modification of ->real_blocked, which is fine
2905 * because, only current can change ->real_blocked and all readers of
2906 * ->real_blocked don't care as long ->real_blocked is always a subset
2909 sigprocmask(SIG_SETMASK, &vcpu->sigset, ¤t->real_blocked);
2912 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu)
2914 if (!vcpu->sigset_active)
2917 sigprocmask(SIG_SETMASK, ¤t->real_blocked, NULL);
2918 sigemptyset(¤t->real_blocked);
2921 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
2923 unsigned int old, val, grow, grow_start;
2925 old = val = vcpu->halt_poll_ns;
2926 grow_start = READ_ONCE(halt_poll_ns_grow_start);
2927 grow = READ_ONCE(halt_poll_ns_grow);
2932 if (val < grow_start)
2935 if (val > halt_poll_ns)
2938 vcpu->halt_poll_ns = val;
2940 trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
2943 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
2945 unsigned int old, val, shrink;
2947 old = val = vcpu->halt_poll_ns;
2948 shrink = READ_ONCE(halt_poll_ns_shrink);
2954 vcpu->halt_poll_ns = val;
2955 trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
2958 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
2961 int idx = srcu_read_lock(&vcpu->kvm->srcu);
2963 if (kvm_arch_vcpu_runnable(vcpu)) {
2964 kvm_make_request(KVM_REQ_UNHALT, vcpu);
2967 if (kvm_cpu_has_pending_timer(vcpu))
2969 if (signal_pending(current))
2974 srcu_read_unlock(&vcpu->kvm->srcu, idx);
2979 update_halt_poll_stats(struct kvm_vcpu *vcpu, u64 poll_ns, bool waited)
2982 vcpu->stat.halt_poll_fail_ns += poll_ns;
2984 vcpu->stat.halt_poll_success_ns += poll_ns;
2988 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2990 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
2992 ktime_t start, cur, poll_end;
2993 bool waited = false;
2996 kvm_arch_vcpu_blocking(vcpu);
2998 start = cur = poll_end = ktime_get();
2999 if (vcpu->halt_poll_ns && !kvm_arch_no_poll(vcpu)) {
3000 ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns);
3002 ++vcpu->stat.halt_attempted_poll;
3005 * This sets KVM_REQ_UNHALT if an interrupt
3008 if (kvm_vcpu_check_block(vcpu) < 0) {
3009 ++vcpu->stat.halt_successful_poll;
3010 if (!vcpu_valid_wakeup(vcpu))
3011 ++vcpu->stat.halt_poll_invalid;
3014 poll_end = cur = ktime_get();
3015 } while (single_task_running() && ktime_before(cur, stop));
3018 prepare_to_rcuwait(&vcpu->wait);
3020 set_current_state(TASK_INTERRUPTIBLE);
3022 if (kvm_vcpu_check_block(vcpu) < 0)
3028 finish_rcuwait(&vcpu->wait);
3031 kvm_arch_vcpu_unblocking(vcpu);
3032 block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
3034 update_halt_poll_stats(
3035 vcpu, ktime_to_ns(ktime_sub(poll_end, start)), waited);
3037 if (!kvm_arch_no_poll(vcpu)) {
3038 if (!vcpu_valid_wakeup(vcpu)) {
3039 shrink_halt_poll_ns(vcpu);
3040 } else if (vcpu->kvm->max_halt_poll_ns) {
3041 if (block_ns <= vcpu->halt_poll_ns)
3043 /* we had a long block, shrink polling */
3044 else if (vcpu->halt_poll_ns &&
3045 block_ns > vcpu->kvm->max_halt_poll_ns)
3046 shrink_halt_poll_ns(vcpu);
3047 /* we had a short halt and our poll time is too small */
3048 else if (vcpu->halt_poll_ns < vcpu->kvm->max_halt_poll_ns &&
3049 block_ns < vcpu->kvm->max_halt_poll_ns)
3050 grow_halt_poll_ns(vcpu);
3052 vcpu->halt_poll_ns = 0;
3056 trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu));
3057 kvm_arch_vcpu_block_finish(vcpu);
3059 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
3061 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
3063 struct rcuwait *waitp;
3065 waitp = kvm_arch_vcpu_get_wait(vcpu);
3066 if (rcuwait_wake_up(waitp)) {
3067 WRITE_ONCE(vcpu->ready, true);
3068 ++vcpu->stat.halt_wakeup;
3074 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
3078 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
3080 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
3083 int cpu = vcpu->cpu;
3085 if (kvm_vcpu_wake_up(vcpu))
3089 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
3090 if (kvm_arch_vcpu_should_kick(vcpu))
3091 smp_send_reschedule(cpu);
3094 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
3095 #endif /* !CONFIG_S390 */
3097 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
3100 struct task_struct *task = NULL;
3104 pid = rcu_dereference(target->pid);
3106 task = get_pid_task(pid, PIDTYPE_PID);
3110 ret = yield_to(task, 1);
3111 put_task_struct(task);
3115 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
3118 * Helper that checks whether a VCPU is eligible for directed yield.
3119 * Most eligible candidate to yield is decided by following heuristics:
3121 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
3122 * (preempted lock holder), indicated by @in_spin_loop.
3123 * Set at the beginning and cleared at the end of interception/PLE handler.
3125 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
3126 * chance last time (mostly it has become eligible now since we have probably
3127 * yielded to lockholder in last iteration. This is done by toggling
3128 * @dy_eligible each time a VCPU checked for eligibility.)
3130 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
3131 * to preempted lock-holder could result in wrong VCPU selection and CPU
3132 * burning. Giving priority for a potential lock-holder increases lock
3135 * Since algorithm is based on heuristics, accessing another VCPU data without
3136 * locking does not harm. It may result in trying to yield to same VCPU, fail
3137 * and continue with next VCPU and so on.
3139 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
3141 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
3144 eligible = !vcpu->spin_loop.in_spin_loop ||
3145 vcpu->spin_loop.dy_eligible;
3147 if (vcpu->spin_loop.in_spin_loop)
3148 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
3157 * Unlike kvm_arch_vcpu_runnable, this function is called outside
3158 * a vcpu_load/vcpu_put pair. However, for most architectures
3159 * kvm_arch_vcpu_runnable does not require vcpu_load.
3161 bool __weak kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
3163 return kvm_arch_vcpu_runnable(vcpu);
3166 static bool vcpu_dy_runnable(struct kvm_vcpu *vcpu)
3168 if (kvm_arch_dy_runnable(vcpu))
3171 #ifdef CONFIG_KVM_ASYNC_PF
3172 if (!list_empty_careful(&vcpu->async_pf.done))
3179 void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode)
3181 struct kvm *kvm = me->kvm;
3182 struct kvm_vcpu *vcpu;
3183 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
3189 kvm_vcpu_set_in_spin_loop(me, true);
3191 * We boost the priority of a VCPU that is runnable but not
3192 * currently running, because it got preempted by something
3193 * else and called schedule in __vcpu_run. Hopefully that
3194 * VCPU is holding the lock that we need and will release it.
3195 * We approximate round-robin by starting at the last boosted VCPU.
3197 for (pass = 0; pass < 2 && !yielded && try; pass++) {
3198 kvm_for_each_vcpu(i, vcpu, kvm) {
3199 if (!pass && i <= last_boosted_vcpu) {
3200 i = last_boosted_vcpu;
3202 } else if (pass && i > last_boosted_vcpu)
3204 if (!READ_ONCE(vcpu->ready))
3208 if (rcuwait_active(&vcpu->wait) &&
3209 !vcpu_dy_runnable(vcpu))
3211 if (READ_ONCE(vcpu->preempted) && yield_to_kernel_mode &&
3212 !kvm_arch_vcpu_in_kernel(vcpu))
3214 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
3217 yielded = kvm_vcpu_yield_to(vcpu);
3219 kvm->last_boosted_vcpu = i;
3221 } else if (yielded < 0) {
3228 kvm_vcpu_set_in_spin_loop(me, false);
3230 /* Ensure vcpu is not eligible during next spinloop */
3231 kvm_vcpu_set_dy_eligible(me, false);
3233 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
3235 static bool kvm_page_in_dirty_ring(struct kvm *kvm, unsigned long pgoff)
3237 #if KVM_DIRTY_LOG_PAGE_OFFSET > 0
3238 return (pgoff >= KVM_DIRTY_LOG_PAGE_OFFSET) &&
3239 (pgoff < KVM_DIRTY_LOG_PAGE_OFFSET +
3240 kvm->dirty_ring_size / PAGE_SIZE);
3246 static vm_fault_t kvm_vcpu_fault(struct vm_fault *vmf)
3248 struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data;
3251 if (vmf->pgoff == 0)
3252 page = virt_to_page(vcpu->run);
3254 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
3255 page = virt_to_page(vcpu->arch.pio_data);
3257 #ifdef CONFIG_KVM_MMIO
3258 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
3259 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
3261 else if (kvm_page_in_dirty_ring(vcpu->kvm, vmf->pgoff))
3262 page = kvm_dirty_ring_get_page(
3264 vmf->pgoff - KVM_DIRTY_LOG_PAGE_OFFSET);
3266 return kvm_arch_vcpu_fault(vcpu, vmf);
3272 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
3273 .fault = kvm_vcpu_fault,
3276 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
3278 struct kvm_vcpu *vcpu = file->private_data;
3279 unsigned long pages = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
3281 if ((kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff) ||
3282 kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff + pages - 1)) &&
3283 ((vma->vm_flags & VM_EXEC) || !(vma->vm_flags & VM_SHARED)))
3286 vma->vm_ops = &kvm_vcpu_vm_ops;
3290 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
3292 struct kvm_vcpu *vcpu = filp->private_data;
3294 kvm_put_kvm(vcpu->kvm);
3298 static struct file_operations kvm_vcpu_fops = {
3299 .release = kvm_vcpu_release,
3300 .unlocked_ioctl = kvm_vcpu_ioctl,
3301 .mmap = kvm_vcpu_mmap,
3302 .llseek = noop_llseek,
3303 KVM_COMPAT(kvm_vcpu_compat_ioctl),
3307 * Allocates an inode for the vcpu.
3309 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
3311 char name[8 + 1 + ITOA_MAX_LEN + 1];
3313 snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id);
3314 return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
3317 static void kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
3319 #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS
3320 struct dentry *debugfs_dentry;
3321 char dir_name[ITOA_MAX_LEN * 2];
3323 if (!debugfs_initialized())
3326 snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
3327 debugfs_dentry = debugfs_create_dir(dir_name,
3328 vcpu->kvm->debugfs_dentry);
3330 kvm_arch_create_vcpu_debugfs(vcpu, debugfs_dentry);
3335 * Creates some virtual cpus. Good luck creating more than one.
3337 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
3340 struct kvm_vcpu *vcpu;
3343 if (id >= KVM_MAX_VCPU_ID)
3346 mutex_lock(&kvm->lock);
3347 if (kvm->created_vcpus == KVM_MAX_VCPUS) {
3348 mutex_unlock(&kvm->lock);
3352 kvm->created_vcpus++;
3353 mutex_unlock(&kvm->lock);
3355 r = kvm_arch_vcpu_precreate(kvm, id);
3357 goto vcpu_decrement;
3359 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL_ACCOUNT);
3362 goto vcpu_decrement;
3365 BUILD_BUG_ON(sizeof(struct kvm_run) > PAGE_SIZE);
3366 page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
3371 vcpu->run = page_address(page);
3373 kvm_vcpu_init(vcpu, kvm, id);
3375 r = kvm_arch_vcpu_create(vcpu);
3377 goto vcpu_free_run_page;
3379 if (kvm->dirty_ring_size) {
3380 r = kvm_dirty_ring_alloc(&vcpu->dirty_ring,
3381 id, kvm->dirty_ring_size);
3383 goto arch_vcpu_destroy;
3386 mutex_lock(&kvm->lock);
3387 if (kvm_get_vcpu_by_id(kvm, id)) {
3389 goto unlock_vcpu_destroy;
3392 vcpu->vcpu_idx = atomic_read(&kvm->online_vcpus);
3393 BUG_ON(kvm->vcpus[vcpu->vcpu_idx]);
3395 /* Now it's all set up, let userspace reach it */
3397 r = create_vcpu_fd(vcpu);
3399 kvm_put_kvm_no_destroy(kvm);
3400 goto unlock_vcpu_destroy;
3403 kvm->vcpus[vcpu->vcpu_idx] = vcpu;
3406 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
3407 * before kvm->online_vcpu's incremented value.
3410 atomic_inc(&kvm->online_vcpus);
3412 mutex_unlock(&kvm->lock);
3413 kvm_arch_vcpu_postcreate(vcpu);
3414 kvm_create_vcpu_debugfs(vcpu);
3417 unlock_vcpu_destroy:
3418 mutex_unlock(&kvm->lock);
3419 kvm_dirty_ring_free(&vcpu->dirty_ring);
3421 kvm_arch_vcpu_destroy(vcpu);
3423 free_page((unsigned long)vcpu->run);
3425 kmem_cache_free(kvm_vcpu_cache, vcpu);
3427 mutex_lock(&kvm->lock);
3428 kvm->created_vcpus--;
3429 mutex_unlock(&kvm->lock);
3433 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
3436 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
3437 vcpu->sigset_active = 1;
3438 vcpu->sigset = *sigset;
3440 vcpu->sigset_active = 0;
3444 static long kvm_vcpu_ioctl(struct file *filp,
3445 unsigned int ioctl, unsigned long arg)
3447 struct kvm_vcpu *vcpu = filp->private_data;
3448 void __user *argp = (void __user *)arg;
3450 struct kvm_fpu *fpu = NULL;
3451 struct kvm_sregs *kvm_sregs = NULL;
3453 if (vcpu->kvm->mm != current->mm)
3456 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
3460 * Some architectures have vcpu ioctls that are asynchronous to vcpu
3461 * execution; mutex_lock() would break them.
3463 r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg);
3464 if (r != -ENOIOCTLCMD)
3467 if (mutex_lock_killable(&vcpu->mutex))
3475 oldpid = rcu_access_pointer(vcpu->pid);
3476 if (unlikely(oldpid != task_pid(current))) {
3477 /* The thread running this VCPU changed. */
3480 r = kvm_arch_vcpu_run_pid_change(vcpu);
3484 newpid = get_task_pid(current, PIDTYPE_PID);
3485 rcu_assign_pointer(vcpu->pid, newpid);
3490 r = kvm_arch_vcpu_ioctl_run(vcpu);
3491 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
3494 case KVM_GET_REGS: {
3495 struct kvm_regs *kvm_regs;
3498 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL_ACCOUNT);
3501 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
3505 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
3512 case KVM_SET_REGS: {
3513 struct kvm_regs *kvm_regs;
3515 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
3516 if (IS_ERR(kvm_regs)) {
3517 r = PTR_ERR(kvm_regs);
3520 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
3524 case KVM_GET_SREGS: {
3525 kvm_sregs = kzalloc(sizeof(struct kvm_sregs),
3526 GFP_KERNEL_ACCOUNT);
3530 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
3534 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
3539 case KVM_SET_SREGS: {
3540 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
3541 if (IS_ERR(kvm_sregs)) {
3542 r = PTR_ERR(kvm_sregs);
3546 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
3549 case KVM_GET_MP_STATE: {
3550 struct kvm_mp_state mp_state;
3552 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
3556 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
3561 case KVM_SET_MP_STATE: {
3562 struct kvm_mp_state mp_state;
3565 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
3567 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
3570 case KVM_TRANSLATE: {
3571 struct kvm_translation tr;
3574 if (copy_from_user(&tr, argp, sizeof(tr)))
3576 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
3580 if (copy_to_user(argp, &tr, sizeof(tr)))
3585 case KVM_SET_GUEST_DEBUG: {
3586 struct kvm_guest_debug dbg;
3589 if (copy_from_user(&dbg, argp, sizeof(dbg)))
3591 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
3594 case KVM_SET_SIGNAL_MASK: {
3595 struct kvm_signal_mask __user *sigmask_arg = argp;
3596 struct kvm_signal_mask kvm_sigmask;
3597 sigset_t sigset, *p;
3602 if (copy_from_user(&kvm_sigmask, argp,
3603 sizeof(kvm_sigmask)))
3606 if (kvm_sigmask.len != sizeof(sigset))
3609 if (copy_from_user(&sigset, sigmask_arg->sigset,
3614 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
3618 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL_ACCOUNT);
3622 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
3626 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
3632 fpu = memdup_user(argp, sizeof(*fpu));
3638 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
3642 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
3645 mutex_unlock(&vcpu->mutex);
3651 #ifdef CONFIG_KVM_COMPAT
3652 static long kvm_vcpu_compat_ioctl(struct file *filp,
3653 unsigned int ioctl, unsigned long arg)
3655 struct kvm_vcpu *vcpu = filp->private_data;
3656 void __user *argp = compat_ptr(arg);
3659 if (vcpu->kvm->mm != current->mm)
3663 case KVM_SET_SIGNAL_MASK: {
3664 struct kvm_signal_mask __user *sigmask_arg = argp;
3665 struct kvm_signal_mask kvm_sigmask;
3670 if (copy_from_user(&kvm_sigmask, argp,
3671 sizeof(kvm_sigmask)))
3674 if (kvm_sigmask.len != sizeof(compat_sigset_t))
3677 if (get_compat_sigset(&sigset,
3678 (compat_sigset_t __user *)sigmask_arg->sigset))
3680 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
3682 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
3686 r = kvm_vcpu_ioctl(filp, ioctl, arg);
3694 static int kvm_device_mmap(struct file *filp, struct vm_area_struct *vma)
3696 struct kvm_device *dev = filp->private_data;
3699 return dev->ops->mmap(dev, vma);
3704 static int kvm_device_ioctl_attr(struct kvm_device *dev,
3705 int (*accessor)(struct kvm_device *dev,
3706 struct kvm_device_attr *attr),
3709 struct kvm_device_attr attr;
3714 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
3717 return accessor(dev, &attr);
3720 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
3723 struct kvm_device *dev = filp->private_data;
3725 if (dev->kvm->mm != current->mm)
3729 case KVM_SET_DEVICE_ATTR:
3730 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
3731 case KVM_GET_DEVICE_ATTR:
3732 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
3733 case KVM_HAS_DEVICE_ATTR:
3734 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
3736 if (dev->ops->ioctl)
3737 return dev->ops->ioctl(dev, ioctl, arg);
3743 static int kvm_device_release(struct inode *inode, struct file *filp)
3745 struct kvm_device *dev = filp->private_data;
3746 struct kvm *kvm = dev->kvm;
3748 if (dev->ops->release) {
3749 mutex_lock(&kvm->lock);
3750 list_del(&dev->vm_node);
3751 dev->ops->release(dev);
3752 mutex_unlock(&kvm->lock);
3759 static const struct file_operations kvm_device_fops = {
3760 .unlocked_ioctl = kvm_device_ioctl,
3761 .release = kvm_device_release,
3762 KVM_COMPAT(kvm_device_ioctl),
3763 .mmap = kvm_device_mmap,
3766 struct kvm_device *kvm_device_from_filp(struct file *filp)
3768 if (filp->f_op != &kvm_device_fops)
3771 return filp->private_data;
3774 static const struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
3775 #ifdef CONFIG_KVM_MPIC
3776 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
3777 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
3781 int kvm_register_device_ops(const struct kvm_device_ops *ops, u32 type)
3783 if (type >= ARRAY_SIZE(kvm_device_ops_table))
3786 if (kvm_device_ops_table[type] != NULL)
3789 kvm_device_ops_table[type] = ops;
3793 void kvm_unregister_device_ops(u32 type)
3795 if (kvm_device_ops_table[type] != NULL)
3796 kvm_device_ops_table[type] = NULL;
3799 static int kvm_ioctl_create_device(struct kvm *kvm,
3800 struct kvm_create_device *cd)
3802 const struct kvm_device_ops *ops = NULL;
3803 struct kvm_device *dev;
3804 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
3808 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
3811 type = array_index_nospec(cd->type, ARRAY_SIZE(kvm_device_ops_table));
3812 ops = kvm_device_ops_table[type];
3819 dev = kzalloc(sizeof(*dev), GFP_KERNEL_ACCOUNT);
3826 mutex_lock(&kvm->lock);
3827 ret = ops->create(dev, type);
3829 mutex_unlock(&kvm->lock);
3833 list_add(&dev->vm_node, &kvm->devices);
3834 mutex_unlock(&kvm->lock);
3840 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
3842 kvm_put_kvm_no_destroy(kvm);
3843 mutex_lock(&kvm->lock);
3844 list_del(&dev->vm_node);
3845 mutex_unlock(&kvm->lock);
3854 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
3857 case KVM_CAP_USER_MEMORY:
3858 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
3859 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
3860 case KVM_CAP_INTERNAL_ERROR_DATA:
3861 #ifdef CONFIG_HAVE_KVM_MSI
3862 case KVM_CAP_SIGNAL_MSI:
3864 #ifdef CONFIG_HAVE_KVM_IRQFD
3866 case KVM_CAP_IRQFD_RESAMPLE:
3868 case KVM_CAP_IOEVENTFD_ANY_LENGTH:
3869 case KVM_CAP_CHECK_EXTENSION_VM:
3870 case KVM_CAP_ENABLE_CAP_VM:
3871 case KVM_CAP_HALT_POLL:
3873 #ifdef CONFIG_KVM_MMIO
3874 case KVM_CAP_COALESCED_MMIO:
3875 return KVM_COALESCED_MMIO_PAGE_OFFSET;
3876 case KVM_CAP_COALESCED_PIO:
3879 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3880 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2:
3881 return KVM_DIRTY_LOG_MANUAL_CAPS;
3883 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3884 case KVM_CAP_IRQ_ROUTING:
3885 return KVM_MAX_IRQ_ROUTES;
3887 #if KVM_ADDRESS_SPACE_NUM > 1
3888 case KVM_CAP_MULTI_ADDRESS_SPACE:
3889 return KVM_ADDRESS_SPACE_NUM;
3891 case KVM_CAP_NR_MEMSLOTS:
3892 return KVM_USER_MEM_SLOTS;
3893 case KVM_CAP_DIRTY_LOG_RING:
3894 #if KVM_DIRTY_LOG_PAGE_OFFSET > 0
3895 return KVM_DIRTY_RING_MAX_ENTRIES * sizeof(struct kvm_dirty_gfn);
3902 return kvm_vm_ioctl_check_extension(kvm, arg);
3905 static int kvm_vm_ioctl_enable_dirty_log_ring(struct kvm *kvm, u32 size)
3909 if (!KVM_DIRTY_LOG_PAGE_OFFSET)
3912 /* the size should be power of 2 */
3913 if (!size || (size & (size - 1)))
3916 /* Should be bigger to keep the reserved entries, or a page */
3917 if (size < kvm_dirty_ring_get_rsvd_entries() *
3918 sizeof(struct kvm_dirty_gfn) || size < PAGE_SIZE)
3921 if (size > KVM_DIRTY_RING_MAX_ENTRIES *
3922 sizeof(struct kvm_dirty_gfn))
3925 /* We only allow it to set once */
3926 if (kvm->dirty_ring_size)
3929 mutex_lock(&kvm->lock);
3931 if (kvm->created_vcpus) {
3932 /* We don't allow to change this value after vcpu created */
3935 kvm->dirty_ring_size = size;
3939 mutex_unlock(&kvm->lock);
3943 static int kvm_vm_ioctl_reset_dirty_pages(struct kvm *kvm)
3946 struct kvm_vcpu *vcpu;
3949 if (!kvm->dirty_ring_size)
3952 mutex_lock(&kvm->slots_lock);
3954 kvm_for_each_vcpu(i, vcpu, kvm)
3955 cleared += kvm_dirty_ring_reset(vcpu->kvm, &vcpu->dirty_ring);
3957 mutex_unlock(&kvm->slots_lock);
3960 kvm_flush_remote_tlbs(kvm);
3965 int __attribute__((weak)) kvm_vm_ioctl_enable_cap(struct kvm *kvm,
3966 struct kvm_enable_cap *cap)
3971 static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm,
3972 struct kvm_enable_cap *cap)
3975 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3976 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2: {
3977 u64 allowed_options = KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE;
3979 if (cap->args[0] & KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE)
3980 allowed_options = KVM_DIRTY_LOG_MANUAL_CAPS;
3982 if (cap->flags || (cap->args[0] & ~allowed_options))
3984 kvm->manual_dirty_log_protect = cap->args[0];
3988 case KVM_CAP_HALT_POLL: {
3989 if (cap->flags || cap->args[0] != (unsigned int)cap->args[0])
3992 kvm->max_halt_poll_ns = cap->args[0];
3995 case KVM_CAP_DIRTY_LOG_RING:
3996 return kvm_vm_ioctl_enable_dirty_log_ring(kvm, cap->args[0]);
3998 return kvm_vm_ioctl_enable_cap(kvm, cap);
4002 static long kvm_vm_ioctl(struct file *filp,
4003 unsigned int ioctl, unsigned long arg)
4005 struct kvm *kvm = filp->private_data;
4006 void __user *argp = (void __user *)arg;
4009 if (kvm->mm != current->mm)
4012 case KVM_CREATE_VCPU:
4013 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
4015 case KVM_ENABLE_CAP: {
4016 struct kvm_enable_cap cap;
4019 if (copy_from_user(&cap, argp, sizeof(cap)))
4021 r = kvm_vm_ioctl_enable_cap_generic(kvm, &cap);
4024 case KVM_SET_USER_MEMORY_REGION: {
4025 struct kvm_userspace_memory_region kvm_userspace_mem;
4028 if (copy_from_user(&kvm_userspace_mem, argp,
4029 sizeof(kvm_userspace_mem)))
4032 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
4035 case KVM_GET_DIRTY_LOG: {
4036 struct kvm_dirty_log log;
4039 if (copy_from_user(&log, argp, sizeof(log)))
4041 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
4044 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4045 case KVM_CLEAR_DIRTY_LOG: {
4046 struct kvm_clear_dirty_log log;
4049 if (copy_from_user(&log, argp, sizeof(log)))
4051 r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
4055 #ifdef CONFIG_KVM_MMIO
4056 case KVM_REGISTER_COALESCED_MMIO: {
4057 struct kvm_coalesced_mmio_zone zone;
4060 if (copy_from_user(&zone, argp, sizeof(zone)))
4062 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
4065 case KVM_UNREGISTER_COALESCED_MMIO: {
4066 struct kvm_coalesced_mmio_zone zone;
4069 if (copy_from_user(&zone, argp, sizeof(zone)))
4071 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
4076 struct kvm_irqfd data;
4079 if (copy_from_user(&data, argp, sizeof(data)))
4081 r = kvm_irqfd(kvm, &data);
4084 case KVM_IOEVENTFD: {
4085 struct kvm_ioeventfd data;
4088 if (copy_from_user(&data, argp, sizeof(data)))
4090 r = kvm_ioeventfd(kvm, &data);
4093 #ifdef CONFIG_HAVE_KVM_MSI
4094 case KVM_SIGNAL_MSI: {
4098 if (copy_from_user(&msi, argp, sizeof(msi)))
4100 r = kvm_send_userspace_msi(kvm, &msi);
4104 #ifdef __KVM_HAVE_IRQ_LINE
4105 case KVM_IRQ_LINE_STATUS:
4106 case KVM_IRQ_LINE: {
4107 struct kvm_irq_level irq_event;
4110 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
4113 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
4114 ioctl == KVM_IRQ_LINE_STATUS);
4119 if (ioctl == KVM_IRQ_LINE_STATUS) {
4120 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
4128 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
4129 case KVM_SET_GSI_ROUTING: {
4130 struct kvm_irq_routing routing;
4131 struct kvm_irq_routing __user *urouting;
4132 struct kvm_irq_routing_entry *entries = NULL;
4135 if (copy_from_user(&routing, argp, sizeof(routing)))
4138 if (!kvm_arch_can_set_irq_routing(kvm))
4140 if (routing.nr > KVM_MAX_IRQ_ROUTES)
4146 entries = vmemdup_user(urouting->entries,
4147 array_size(sizeof(*entries),
4149 if (IS_ERR(entries)) {
4150 r = PTR_ERR(entries);
4154 r = kvm_set_irq_routing(kvm, entries, routing.nr,
4159 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
4160 case KVM_CREATE_DEVICE: {
4161 struct kvm_create_device cd;
4164 if (copy_from_user(&cd, argp, sizeof(cd)))
4167 r = kvm_ioctl_create_device(kvm, &cd);
4172 if (copy_to_user(argp, &cd, sizeof(cd)))
4178 case KVM_CHECK_EXTENSION:
4179 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
4181 case KVM_RESET_DIRTY_RINGS:
4182 r = kvm_vm_ioctl_reset_dirty_pages(kvm);
4185 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
4191 #ifdef CONFIG_KVM_COMPAT
4192 struct compat_kvm_dirty_log {
4196 compat_uptr_t dirty_bitmap; /* one bit per page */
4201 static long kvm_vm_compat_ioctl(struct file *filp,
4202 unsigned int ioctl, unsigned long arg)
4204 struct kvm *kvm = filp->private_data;
4207 if (kvm->mm != current->mm)
4210 case KVM_GET_DIRTY_LOG: {
4211 struct compat_kvm_dirty_log compat_log;
4212 struct kvm_dirty_log log;
4214 if (copy_from_user(&compat_log, (void __user *)arg,
4215 sizeof(compat_log)))
4217 log.slot = compat_log.slot;
4218 log.padding1 = compat_log.padding1;
4219 log.padding2 = compat_log.padding2;
4220 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
4222 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
4226 r = kvm_vm_ioctl(filp, ioctl, arg);
4232 static struct file_operations kvm_vm_fops = {
4233 .release = kvm_vm_release,
4234 .unlocked_ioctl = kvm_vm_ioctl,
4235 .llseek = noop_llseek,
4236 KVM_COMPAT(kvm_vm_compat_ioctl),
4239 static int kvm_dev_ioctl_create_vm(unsigned long type)
4245 kvm = kvm_create_vm(type);
4247 return PTR_ERR(kvm);
4248 #ifdef CONFIG_KVM_MMIO
4249 r = kvm_coalesced_mmio_init(kvm);
4253 r = get_unused_fd_flags(O_CLOEXEC);
4257 file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
4265 * Don't call kvm_put_kvm anymore at this point; file->f_op is
4266 * already set, with ->release() being kvm_vm_release(). In error
4267 * cases it will be called by the final fput(file) and will take
4268 * care of doing kvm_put_kvm(kvm).
4270 if (kvm_create_vm_debugfs(kvm, r) < 0) {
4275 kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm);
4277 fd_install(r, file);
4285 static long kvm_dev_ioctl(struct file *filp,
4286 unsigned int ioctl, unsigned long arg)
4291 case KVM_GET_API_VERSION:
4294 r = KVM_API_VERSION;
4297 r = kvm_dev_ioctl_create_vm(arg);
4299 case KVM_CHECK_EXTENSION:
4300 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
4302 case KVM_GET_VCPU_MMAP_SIZE:
4305 r = PAGE_SIZE; /* struct kvm_run */
4307 r += PAGE_SIZE; /* pio data page */
4309 #ifdef CONFIG_KVM_MMIO
4310 r += PAGE_SIZE; /* coalesced mmio ring page */
4313 case KVM_TRACE_ENABLE:
4314 case KVM_TRACE_PAUSE:
4315 case KVM_TRACE_DISABLE:
4319 return kvm_arch_dev_ioctl(filp, ioctl, arg);
4325 static struct file_operations kvm_chardev_ops = {
4326 .unlocked_ioctl = kvm_dev_ioctl,
4327 .llseek = noop_llseek,
4328 KVM_COMPAT(kvm_dev_ioctl),
4331 static struct miscdevice kvm_dev = {
4337 static void hardware_enable_nolock(void *junk)
4339 int cpu = raw_smp_processor_id();
4342 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
4345 cpumask_set_cpu(cpu, cpus_hardware_enabled);
4347 r = kvm_arch_hardware_enable();
4350 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
4351 atomic_inc(&hardware_enable_failed);
4352 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
4356 static int kvm_starting_cpu(unsigned int cpu)
4358 raw_spin_lock(&kvm_count_lock);
4359 if (kvm_usage_count)
4360 hardware_enable_nolock(NULL);
4361 raw_spin_unlock(&kvm_count_lock);
4365 static void hardware_disable_nolock(void *junk)
4367 int cpu = raw_smp_processor_id();
4369 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
4371 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
4372 kvm_arch_hardware_disable();
4375 static int kvm_dying_cpu(unsigned int cpu)
4377 raw_spin_lock(&kvm_count_lock);
4378 if (kvm_usage_count)
4379 hardware_disable_nolock(NULL);
4380 raw_spin_unlock(&kvm_count_lock);
4384 static void hardware_disable_all_nolock(void)
4386 BUG_ON(!kvm_usage_count);
4389 if (!kvm_usage_count)
4390 on_each_cpu(hardware_disable_nolock, NULL, 1);
4393 static void hardware_disable_all(void)
4395 raw_spin_lock(&kvm_count_lock);
4396 hardware_disable_all_nolock();
4397 raw_spin_unlock(&kvm_count_lock);
4400 static int hardware_enable_all(void)
4404 raw_spin_lock(&kvm_count_lock);
4407 if (kvm_usage_count == 1) {
4408 atomic_set(&hardware_enable_failed, 0);
4409 on_each_cpu(hardware_enable_nolock, NULL, 1);
4411 if (atomic_read(&hardware_enable_failed)) {
4412 hardware_disable_all_nolock();
4417 raw_spin_unlock(&kvm_count_lock);
4422 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
4426 * Some (well, at least mine) BIOSes hang on reboot if
4429 * And Intel TXT required VMX off for all cpu when system shutdown.
4431 pr_info("kvm: exiting hardware virtualization\n");
4432 kvm_rebooting = true;
4433 on_each_cpu(hardware_disable_nolock, NULL, 1);
4437 static struct notifier_block kvm_reboot_notifier = {
4438 .notifier_call = kvm_reboot,
4442 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
4446 for (i = 0; i < bus->dev_count; i++) {
4447 struct kvm_io_device *pos = bus->range[i].dev;
4449 kvm_iodevice_destructor(pos);
4454 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
4455 const struct kvm_io_range *r2)
4457 gpa_t addr1 = r1->addr;
4458 gpa_t addr2 = r2->addr;
4463 /* If r2->len == 0, match the exact address. If r2->len != 0,
4464 * accept any overlapping write. Any order is acceptable for
4465 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
4466 * we process all of them.
4479 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
4481 return kvm_io_bus_cmp(p1, p2);
4484 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
4485 gpa_t addr, int len)
4487 struct kvm_io_range *range, key;
4490 key = (struct kvm_io_range) {
4495 range = bsearch(&key, bus->range, bus->dev_count,
4496 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
4500 off = range - bus->range;
4502 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
4508 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
4509 struct kvm_io_range *range, const void *val)
4513 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
4517 while (idx < bus->dev_count &&
4518 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
4519 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
4528 /* kvm_io_bus_write - called under kvm->slots_lock */
4529 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
4530 int len, const void *val)
4532 struct kvm_io_bus *bus;
4533 struct kvm_io_range range;
4536 range = (struct kvm_io_range) {
4541 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
4544 r = __kvm_io_bus_write(vcpu, bus, &range, val);
4545 return r < 0 ? r : 0;
4547 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
4549 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
4550 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
4551 gpa_t addr, int len, const void *val, long cookie)
4553 struct kvm_io_bus *bus;
4554 struct kvm_io_range range;
4556 range = (struct kvm_io_range) {
4561 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
4565 /* First try the device referenced by cookie. */
4566 if ((cookie >= 0) && (cookie < bus->dev_count) &&
4567 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
4568 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
4573 * cookie contained garbage; fall back to search and return the
4574 * correct cookie value.
4576 return __kvm_io_bus_write(vcpu, bus, &range, val);
4579 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
4580 struct kvm_io_range *range, void *val)
4584 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
4588 while (idx < bus->dev_count &&
4589 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
4590 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
4599 /* kvm_io_bus_read - called under kvm->slots_lock */
4600 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
4603 struct kvm_io_bus *bus;
4604 struct kvm_io_range range;
4607 range = (struct kvm_io_range) {
4612 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
4615 r = __kvm_io_bus_read(vcpu, bus, &range, val);
4616 return r < 0 ? r : 0;
4619 /* Caller must hold slots_lock. */
4620 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
4621 int len, struct kvm_io_device *dev)
4624 struct kvm_io_bus *new_bus, *bus;
4625 struct kvm_io_range range;
4627 bus = kvm_get_bus(kvm, bus_idx);
4631 /* exclude ioeventfd which is limited by maximum fd */
4632 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
4635 new_bus = kmalloc(struct_size(bus, range, bus->dev_count + 1),
4636 GFP_KERNEL_ACCOUNT);
4640 range = (struct kvm_io_range) {
4646 for (i = 0; i < bus->dev_count; i++)
4647 if (kvm_io_bus_cmp(&bus->range[i], &range) > 0)
4650 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
4651 new_bus->dev_count++;
4652 new_bus->range[i] = range;
4653 memcpy(new_bus->range + i + 1, bus->range + i,
4654 (bus->dev_count - i) * sizeof(struct kvm_io_range));
4655 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
4656 synchronize_srcu_expedited(&kvm->srcu);
4662 /* Caller must hold slots_lock. */
4663 void kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
4664 struct kvm_io_device *dev)
4667 struct kvm_io_bus *new_bus, *bus;
4669 bus = kvm_get_bus(kvm, bus_idx);
4673 for (i = 0; i < bus->dev_count; i++)
4674 if (bus->range[i].dev == dev) {
4678 if (i == bus->dev_count)
4681 new_bus = kmalloc(struct_size(bus, range, bus->dev_count - 1),
4682 GFP_KERNEL_ACCOUNT);
4684 memcpy(new_bus, bus, struct_size(bus, range, i));
4685 new_bus->dev_count--;
4686 memcpy(new_bus->range + i, bus->range + i + 1,
4687 flex_array_size(new_bus, range, new_bus->dev_count - i));
4689 pr_err("kvm: failed to shrink bus, removing it completely\n");
4690 for (j = 0; j < bus->dev_count; j++) {
4693 kvm_iodevice_destructor(bus->range[j].dev);
4697 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
4698 synchronize_srcu_expedited(&kvm->srcu);
4703 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
4706 struct kvm_io_bus *bus;
4707 int dev_idx, srcu_idx;
4708 struct kvm_io_device *iodev = NULL;
4710 srcu_idx = srcu_read_lock(&kvm->srcu);
4712 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
4716 dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
4720 iodev = bus->range[dev_idx].dev;
4723 srcu_read_unlock(&kvm->srcu, srcu_idx);
4727 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
4729 static int kvm_debugfs_open(struct inode *inode, struct file *file,
4730 int (*get)(void *, u64 *), int (*set)(void *, u64),
4733 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
4736 /* The debugfs files are a reference to the kvm struct which
4737 * is still valid when kvm_destroy_vm is called.
4738 * To avoid the race between open and the removal of the debugfs
4739 * directory we test against the users count.
4741 if (!refcount_inc_not_zero(&stat_data->kvm->users_count))
4744 if (simple_attr_open(inode, file, get,
4745 KVM_DBGFS_GET_MODE(stat_data->dbgfs_item) & 0222
4748 kvm_put_kvm(stat_data->kvm);
4755 static int kvm_debugfs_release(struct inode *inode, struct file *file)
4757 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
4760 simple_attr_release(inode, file);
4761 kvm_put_kvm(stat_data->kvm);
4766 static int kvm_get_stat_per_vm(struct kvm *kvm, size_t offset, u64 *val)
4768 *val = *(ulong *)((void *)kvm + offset);
4773 static int kvm_clear_stat_per_vm(struct kvm *kvm, size_t offset)
4775 *(ulong *)((void *)kvm + offset) = 0;
4780 static int kvm_get_stat_per_vcpu(struct kvm *kvm, size_t offset, u64 *val)
4783 struct kvm_vcpu *vcpu;
4787 kvm_for_each_vcpu(i, vcpu, kvm)
4788 *val += *(u64 *)((void *)vcpu + offset);
4793 static int kvm_clear_stat_per_vcpu(struct kvm *kvm, size_t offset)
4796 struct kvm_vcpu *vcpu;
4798 kvm_for_each_vcpu(i, vcpu, kvm)
4799 *(u64 *)((void *)vcpu + offset) = 0;
4804 static int kvm_stat_data_get(void *data, u64 *val)
4807 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
4809 switch (stat_data->dbgfs_item->kind) {
4811 r = kvm_get_stat_per_vm(stat_data->kvm,
4812 stat_data->dbgfs_item->offset, val);
4815 r = kvm_get_stat_per_vcpu(stat_data->kvm,
4816 stat_data->dbgfs_item->offset, val);
4823 static int kvm_stat_data_clear(void *data, u64 val)
4826 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
4831 switch (stat_data->dbgfs_item->kind) {
4833 r = kvm_clear_stat_per_vm(stat_data->kvm,
4834 stat_data->dbgfs_item->offset);
4837 r = kvm_clear_stat_per_vcpu(stat_data->kvm,
4838 stat_data->dbgfs_item->offset);
4845 static int kvm_stat_data_open(struct inode *inode, struct file *file)
4847 __simple_attr_check_format("%llu\n", 0ull);
4848 return kvm_debugfs_open(inode, file, kvm_stat_data_get,
4849 kvm_stat_data_clear, "%llu\n");
4852 static const struct file_operations stat_fops_per_vm = {
4853 .owner = THIS_MODULE,
4854 .open = kvm_stat_data_open,
4855 .release = kvm_debugfs_release,
4856 .read = simple_attr_read,
4857 .write = simple_attr_write,
4858 .llseek = no_llseek,
4861 static int vm_stat_get(void *_offset, u64 *val)
4863 unsigned offset = (long)_offset;
4868 mutex_lock(&kvm_lock);
4869 list_for_each_entry(kvm, &vm_list, vm_list) {
4870 kvm_get_stat_per_vm(kvm, offset, &tmp_val);
4873 mutex_unlock(&kvm_lock);
4877 static int vm_stat_clear(void *_offset, u64 val)
4879 unsigned offset = (long)_offset;
4885 mutex_lock(&kvm_lock);
4886 list_for_each_entry(kvm, &vm_list, vm_list) {
4887 kvm_clear_stat_per_vm(kvm, offset);
4889 mutex_unlock(&kvm_lock);
4894 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n");
4896 static int vcpu_stat_get(void *_offset, u64 *val)
4898 unsigned offset = (long)_offset;
4903 mutex_lock(&kvm_lock);
4904 list_for_each_entry(kvm, &vm_list, vm_list) {
4905 kvm_get_stat_per_vcpu(kvm, offset, &tmp_val);
4908 mutex_unlock(&kvm_lock);
4912 static int vcpu_stat_clear(void *_offset, u64 val)
4914 unsigned offset = (long)_offset;
4920 mutex_lock(&kvm_lock);
4921 list_for_each_entry(kvm, &vm_list, vm_list) {
4922 kvm_clear_stat_per_vcpu(kvm, offset);
4924 mutex_unlock(&kvm_lock);
4929 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear,
4932 static const struct file_operations *stat_fops[] = {
4933 [KVM_STAT_VCPU] = &vcpu_stat_fops,
4934 [KVM_STAT_VM] = &vm_stat_fops,
4937 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm)
4939 struct kobj_uevent_env *env;
4940 unsigned long long created, active;
4942 if (!kvm_dev.this_device || !kvm)
4945 mutex_lock(&kvm_lock);
4946 if (type == KVM_EVENT_CREATE_VM) {
4947 kvm_createvm_count++;
4949 } else if (type == KVM_EVENT_DESTROY_VM) {
4952 created = kvm_createvm_count;
4953 active = kvm_active_vms;
4954 mutex_unlock(&kvm_lock);
4956 env = kzalloc(sizeof(*env), GFP_KERNEL_ACCOUNT);
4960 add_uevent_var(env, "CREATED=%llu", created);
4961 add_uevent_var(env, "COUNT=%llu", active);
4963 if (type == KVM_EVENT_CREATE_VM) {
4964 add_uevent_var(env, "EVENT=create");
4965 kvm->userspace_pid = task_pid_nr(current);
4966 } else if (type == KVM_EVENT_DESTROY_VM) {
4967 add_uevent_var(env, "EVENT=destroy");
4969 add_uevent_var(env, "PID=%d", kvm->userspace_pid);
4971 if (!IS_ERR_OR_NULL(kvm->debugfs_dentry)) {
4972 char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL_ACCOUNT);
4975 tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX);
4977 add_uevent_var(env, "STATS_PATH=%s", tmp);
4981 /* no need for checks, since we are adding at most only 5 keys */
4982 env->envp[env->envp_idx++] = NULL;
4983 kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp);
4987 static void kvm_init_debug(void)
4989 struct kvm_stats_debugfs_item *p;
4991 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
4993 kvm_debugfs_num_entries = 0;
4994 for (p = debugfs_entries; p->name; ++p, kvm_debugfs_num_entries++) {
4995 debugfs_create_file(p->name, KVM_DBGFS_GET_MODE(p),
4996 kvm_debugfs_dir, (void *)(long)p->offset,
4997 stat_fops[p->kind]);
5001 static int kvm_suspend(void)
5003 if (kvm_usage_count)
5004 hardware_disable_nolock(NULL);
5008 static void kvm_resume(void)
5010 if (kvm_usage_count) {
5011 #ifdef CONFIG_LOCKDEP
5012 WARN_ON(lockdep_is_held(&kvm_count_lock));
5014 hardware_enable_nolock(NULL);
5018 static struct syscore_ops kvm_syscore_ops = {
5019 .suspend = kvm_suspend,
5020 .resume = kvm_resume,
5024 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
5026 return container_of(pn, struct kvm_vcpu, preempt_notifier);
5029 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
5031 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
5033 WRITE_ONCE(vcpu->preempted, false);
5034 WRITE_ONCE(vcpu->ready, false);
5036 __this_cpu_write(kvm_running_vcpu, vcpu);
5037 kvm_arch_sched_in(vcpu, cpu);
5038 kvm_arch_vcpu_load(vcpu, cpu);
5041 static void kvm_sched_out(struct preempt_notifier *pn,
5042 struct task_struct *next)
5044 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
5046 if (current->state == TASK_RUNNING) {
5047 WRITE_ONCE(vcpu->preempted, true);
5048 WRITE_ONCE(vcpu->ready, true);
5050 kvm_arch_vcpu_put(vcpu);
5051 __this_cpu_write(kvm_running_vcpu, NULL);
5055 * kvm_get_running_vcpu - get the vcpu running on the current CPU.
5057 * We can disable preemption locally around accessing the per-CPU variable,
5058 * and use the resolved vcpu pointer after enabling preemption again,
5059 * because even if the current thread is migrated to another CPU, reading
5060 * the per-CPU value later will give us the same value as we update the
5061 * per-CPU variable in the preempt notifier handlers.
5063 struct kvm_vcpu *kvm_get_running_vcpu(void)
5065 struct kvm_vcpu *vcpu;
5068 vcpu = __this_cpu_read(kvm_running_vcpu);
5073 EXPORT_SYMBOL_GPL(kvm_get_running_vcpu);
5076 * kvm_get_running_vcpus - get the per-CPU array of currently running vcpus.
5078 struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void)
5080 return &kvm_running_vcpu;
5083 struct kvm_cpu_compat_check {
5088 static void check_processor_compat(void *data)
5090 struct kvm_cpu_compat_check *c = data;
5092 *c->ret = kvm_arch_check_processor_compat(c->opaque);
5095 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
5096 struct module *module)
5098 struct kvm_cpu_compat_check c;
5102 r = kvm_arch_init(opaque);
5107 * kvm_arch_init makes sure there's at most one caller
5108 * for architectures that support multiple implementations,
5109 * like intel and amd on x86.
5110 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
5111 * conflicts in case kvm is already setup for another implementation.
5113 r = kvm_irqfd_init();
5117 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
5122 r = kvm_arch_hardware_setup(opaque);
5128 for_each_online_cpu(cpu) {
5129 smp_call_function_single(cpu, check_processor_compat, &c, 1);
5134 r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting",
5135 kvm_starting_cpu, kvm_dying_cpu);
5138 register_reboot_notifier(&kvm_reboot_notifier);
5140 /* A kmem cache lets us meet the alignment requirements of fx_save. */
5142 vcpu_align = __alignof__(struct kvm_vcpu);
5144 kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align,
5146 offsetof(struct kvm_vcpu, arch),
5147 sizeof_field(struct kvm_vcpu, arch),
5149 if (!kvm_vcpu_cache) {
5154 r = kvm_async_pf_init();
5158 kvm_chardev_ops.owner = module;
5159 kvm_vm_fops.owner = module;
5160 kvm_vcpu_fops.owner = module;
5162 r = misc_register(&kvm_dev);
5164 pr_err("kvm: misc device register failed\n");
5168 register_syscore_ops(&kvm_syscore_ops);
5170 kvm_preempt_ops.sched_in = kvm_sched_in;
5171 kvm_preempt_ops.sched_out = kvm_sched_out;
5175 r = kvm_vfio_ops_init();
5181 kvm_async_pf_deinit();
5183 kmem_cache_destroy(kvm_vcpu_cache);
5185 unregister_reboot_notifier(&kvm_reboot_notifier);
5186 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
5188 kvm_arch_hardware_unsetup();
5190 free_cpumask_var(cpus_hardware_enabled);
5198 EXPORT_SYMBOL_GPL(kvm_init);
5202 debugfs_remove_recursive(kvm_debugfs_dir);
5203 misc_deregister(&kvm_dev);
5204 kmem_cache_destroy(kvm_vcpu_cache);
5205 kvm_async_pf_deinit();
5206 unregister_syscore_ops(&kvm_syscore_ops);
5207 unregister_reboot_notifier(&kvm_reboot_notifier);
5208 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
5209 on_each_cpu(hardware_disable_nolock, NULL, 1);
5210 kvm_arch_hardware_unsetup();
5213 free_cpumask_var(cpus_hardware_enabled);
5214 kvm_vfio_ops_exit();
5216 EXPORT_SYMBOL_GPL(kvm_exit);
5218 struct kvm_vm_worker_thread_context {
5220 struct task_struct *parent;
5221 struct completion init_done;
5222 kvm_vm_thread_fn_t thread_fn;
5227 static int kvm_vm_worker_thread(void *context)
5230 * The init_context is allocated on the stack of the parent thread, so
5231 * we have to locally copy anything that is needed beyond initialization
5233 struct kvm_vm_worker_thread_context *init_context = context;
5234 struct kvm *kvm = init_context->kvm;
5235 kvm_vm_thread_fn_t thread_fn = init_context->thread_fn;
5236 uintptr_t data = init_context->data;
5239 err = kthread_park(current);
5240 /* kthread_park(current) is never supposed to return an error */
5245 err = cgroup_attach_task_all(init_context->parent, current);
5247 kvm_err("%s: cgroup_attach_task_all failed with err %d\n",
5252 set_user_nice(current, task_nice(init_context->parent));
5255 init_context->err = err;
5256 complete(&init_context->init_done);
5257 init_context = NULL;
5262 /* Wait to be woken up by the spawner before proceeding. */
5265 if (!kthread_should_stop())
5266 err = thread_fn(kvm, data);
5271 int kvm_vm_create_worker_thread(struct kvm *kvm, kvm_vm_thread_fn_t thread_fn,
5272 uintptr_t data, const char *name,
5273 struct task_struct **thread_ptr)
5275 struct kvm_vm_worker_thread_context init_context = {};
5276 struct task_struct *thread;
5279 init_context.kvm = kvm;
5280 init_context.parent = current;
5281 init_context.thread_fn = thread_fn;
5282 init_context.data = data;
5283 init_completion(&init_context.init_done);
5285 thread = kthread_run(kvm_vm_worker_thread, &init_context,
5286 "%s-%d", name, task_pid_nr(current));
5288 return PTR_ERR(thread);
5290 /* kthread_run is never supposed to return NULL */
5291 WARN_ON(thread == NULL);
5293 wait_for_completion(&init_context.init_done);
5295 if (!init_context.err)
5296 *thread_ptr = thread;
5298 return init_context.err;