1 // SPDX-License-Identifier: GPL-2.0-only
3 * Kernel-based Virtual Machine driver for Linux
5 * This module enables machines with Intel VT-x extensions to run virtual
6 * machines without emulation or binary translation.
8 * Copyright (C) 2006 Qumranet, Inc.
9 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
12 * Avi Kivity <avi@qumranet.com>
13 * Yaniv Kamay <yaniv@qumranet.com>
16 #include <kvm/iodev.h>
18 #include <linux/kvm_host.h>
19 #include <linux/kvm.h>
20 #include <linux/module.h>
21 #include <linux/errno.h>
22 #include <linux/percpu.h>
24 #include <linux/miscdevice.h>
25 #include <linux/vmalloc.h>
26 #include <linux/reboot.h>
27 #include <linux/debugfs.h>
28 #include <linux/highmem.h>
29 #include <linux/file.h>
30 #include <linux/syscore_ops.h>
31 #include <linux/cpu.h>
32 #include <linux/sched/signal.h>
33 #include <linux/sched/mm.h>
34 #include <linux/sched/stat.h>
35 #include <linux/cpumask.h>
36 #include <linux/smp.h>
37 #include <linux/anon_inodes.h>
38 #include <linux/profile.h>
39 #include <linux/kvm_para.h>
40 #include <linux/pagemap.h>
41 #include <linux/mman.h>
42 #include <linux/swap.h>
43 #include <linux/bitops.h>
44 #include <linux/spinlock.h>
45 #include <linux/compat.h>
46 #include <linux/srcu.h>
47 #include <linux/hugetlb.h>
48 #include <linux/slab.h>
49 #include <linux/sort.h>
50 #include <linux/bsearch.h>
52 #include <linux/lockdep.h>
53 #include <linux/kthread.h>
54 #include <linux/suspend.h>
56 #include <asm/processor.h>
57 #include <asm/ioctl.h>
58 #include <linux/uaccess.h>
60 #include "coalesced_mmio.h"
65 #include <trace/events/ipi.h>
67 #define CREATE_TRACE_POINTS
68 #include <trace/events/kvm.h>
70 #include <linux/kvm_dirty_ring.h>
73 /* Worst case buffer size needed for holding an integer. */
74 #define ITOA_MAX_LEN 12
76 MODULE_AUTHOR("Qumranet");
77 MODULE_LICENSE("GPL");
79 /* Architectures should define their poll value according to the halt latency */
80 unsigned int halt_poll_ns = KVM_HALT_POLL_NS_DEFAULT;
81 module_param(halt_poll_ns, uint, 0644);
82 EXPORT_SYMBOL_GPL(halt_poll_ns);
84 /* Default doubles per-vcpu halt_poll_ns. */
85 unsigned int halt_poll_ns_grow = 2;
86 module_param(halt_poll_ns_grow, uint, 0644);
87 EXPORT_SYMBOL_GPL(halt_poll_ns_grow);
89 /* The start value to grow halt_poll_ns from */
90 unsigned int halt_poll_ns_grow_start = 10000; /* 10us */
91 module_param(halt_poll_ns_grow_start, uint, 0644);
92 EXPORT_SYMBOL_GPL(halt_poll_ns_grow_start);
94 /* Default resets per-vcpu halt_poll_ns . */
95 unsigned int halt_poll_ns_shrink;
96 module_param(halt_poll_ns_shrink, uint, 0644);
97 EXPORT_SYMBOL_GPL(halt_poll_ns_shrink);
102 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
105 DEFINE_MUTEX(kvm_lock);
108 static struct kmem_cache *kvm_vcpu_cache;
110 static __read_mostly struct preempt_ops kvm_preempt_ops;
111 static DEFINE_PER_CPU(struct kvm_vcpu *, kvm_running_vcpu);
113 struct dentry *kvm_debugfs_dir;
114 EXPORT_SYMBOL_GPL(kvm_debugfs_dir);
116 static const struct file_operations stat_fops_per_vm;
118 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
120 #ifdef CONFIG_KVM_COMPAT
121 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
123 #define KVM_COMPAT(c) .compat_ioctl = (c)
126 * For architectures that don't implement a compat infrastructure,
127 * adopt a double line of defense:
128 * - Prevent a compat task from opening /dev/kvm
129 * - If the open has been done by a 64bit task, and the KVM fd
130 * passed to a compat task, let the ioctls fail.
132 static long kvm_no_compat_ioctl(struct file *file, unsigned int ioctl,
133 unsigned long arg) { return -EINVAL; }
135 static int kvm_no_compat_open(struct inode *inode, struct file *file)
137 return is_compat_task() ? -ENODEV : 0;
139 #define KVM_COMPAT(c) .compat_ioctl = kvm_no_compat_ioctl, \
140 .open = kvm_no_compat_open
142 static int hardware_enable_all(void);
143 static void hardware_disable_all(void);
145 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
147 #define KVM_EVENT_CREATE_VM 0
148 #define KVM_EVENT_DESTROY_VM 1
149 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm);
150 static unsigned long long kvm_createvm_count;
151 static unsigned long long kvm_active_vms;
153 static DEFINE_PER_CPU(cpumask_var_t, cpu_kick_mask);
155 __weak void kvm_arch_guest_memory_reclaimed(struct kvm *kvm)
159 bool kvm_is_zone_device_page(struct page *page)
162 * The metadata used by is_zone_device_page() to determine whether or
163 * not a page is ZONE_DEVICE is guaranteed to be valid if and only if
164 * the device has been pinned, e.g. by get_user_pages(). WARN if the
165 * page_count() is zero to help detect bad usage of this helper.
167 if (WARN_ON_ONCE(!page_count(page)))
170 return is_zone_device_page(page);
174 * Returns a 'struct page' if the pfn is "valid" and backed by a refcounted
175 * page, NULL otherwise. Note, the list of refcounted PG_reserved page types
176 * is likely incomplete, it has been compiled purely through people wanting to
177 * back guest with a certain type of memory and encountering issues.
179 struct page *kvm_pfn_to_refcounted_page(kvm_pfn_t pfn)
186 page = pfn_to_page(pfn);
187 if (!PageReserved(page))
190 /* The ZERO_PAGE(s) is marked PG_reserved, but is refcounted. */
191 if (is_zero_pfn(pfn))
195 * ZONE_DEVICE pages currently set PG_reserved, but from a refcounting
196 * perspective they are "normal" pages, albeit with slightly different
199 if (kvm_is_zone_device_page(page))
206 * Switches to specified vcpu, until a matching vcpu_put()
208 void vcpu_load(struct kvm_vcpu *vcpu)
212 __this_cpu_write(kvm_running_vcpu, vcpu);
213 preempt_notifier_register(&vcpu->preempt_notifier);
214 kvm_arch_vcpu_load(vcpu, cpu);
217 EXPORT_SYMBOL_GPL(vcpu_load);
219 void vcpu_put(struct kvm_vcpu *vcpu)
222 kvm_arch_vcpu_put(vcpu);
223 preempt_notifier_unregister(&vcpu->preempt_notifier);
224 __this_cpu_write(kvm_running_vcpu, NULL);
227 EXPORT_SYMBOL_GPL(vcpu_put);
229 /* TODO: merge with kvm_arch_vcpu_should_kick */
230 static bool kvm_request_needs_ipi(struct kvm_vcpu *vcpu, unsigned req)
232 int mode = kvm_vcpu_exiting_guest_mode(vcpu);
235 * We need to wait for the VCPU to reenable interrupts and get out of
236 * READING_SHADOW_PAGE_TABLES mode.
238 if (req & KVM_REQUEST_WAIT)
239 return mode != OUTSIDE_GUEST_MODE;
242 * Need to kick a running VCPU, but otherwise there is nothing to do.
244 return mode == IN_GUEST_MODE;
247 static void ack_kick(void *_completed)
251 static inline bool kvm_kick_many_cpus(struct cpumask *cpus, bool wait)
253 if (cpumask_empty(cpus))
256 smp_call_function_many(cpus, ack_kick, NULL, wait);
260 static void kvm_make_vcpu_request(struct kvm_vcpu *vcpu, unsigned int req,
261 struct cpumask *tmp, int current_cpu)
265 if (likely(!(req & KVM_REQUEST_NO_ACTION)))
266 __kvm_make_request(req, vcpu);
268 if (!(req & KVM_REQUEST_NO_WAKEUP) && kvm_vcpu_wake_up(vcpu))
272 * Note, the vCPU could get migrated to a different pCPU at any point
273 * after kvm_request_needs_ipi(), which could result in sending an IPI
274 * to the previous pCPU. But, that's OK because the purpose of the IPI
275 * is to ensure the vCPU returns to OUTSIDE_GUEST_MODE, which is
276 * satisfied if the vCPU migrates. Entering READING_SHADOW_PAGE_TABLES
277 * after this point is also OK, as the requirement is only that KVM wait
278 * for vCPUs that were reading SPTEs _before_ any changes were
279 * finalized. See kvm_vcpu_kick() for more details on handling requests.
281 if (kvm_request_needs_ipi(vcpu, req)) {
282 cpu = READ_ONCE(vcpu->cpu);
283 if (cpu != -1 && cpu != current_cpu)
284 __cpumask_set_cpu(cpu, tmp);
288 bool kvm_make_vcpus_request_mask(struct kvm *kvm, unsigned int req,
289 unsigned long *vcpu_bitmap)
291 struct kvm_vcpu *vcpu;
292 struct cpumask *cpus;
298 cpus = this_cpu_cpumask_var_ptr(cpu_kick_mask);
301 for_each_set_bit(i, vcpu_bitmap, KVM_MAX_VCPUS) {
302 vcpu = kvm_get_vcpu(kvm, i);
305 kvm_make_vcpu_request(vcpu, req, cpus, me);
308 called = kvm_kick_many_cpus(cpus, !!(req & KVM_REQUEST_WAIT));
314 bool kvm_make_all_cpus_request_except(struct kvm *kvm, unsigned int req,
315 struct kvm_vcpu *except)
317 struct kvm_vcpu *vcpu;
318 struct cpumask *cpus;
325 cpus = this_cpu_cpumask_var_ptr(cpu_kick_mask);
328 kvm_for_each_vcpu(i, vcpu, kvm) {
331 kvm_make_vcpu_request(vcpu, req, cpus, me);
334 called = kvm_kick_many_cpus(cpus, !!(req & KVM_REQUEST_WAIT));
340 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
342 return kvm_make_all_cpus_request_except(kvm, req, NULL);
344 EXPORT_SYMBOL_GPL(kvm_make_all_cpus_request);
346 void kvm_flush_remote_tlbs(struct kvm *kvm)
348 ++kvm->stat.generic.remote_tlb_flush_requests;
351 * We want to publish modifications to the page tables before reading
352 * mode. Pairs with a memory barrier in arch-specific code.
353 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
354 * and smp_mb in walk_shadow_page_lockless_begin/end.
355 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
357 * There is already an smp_mb__after_atomic() before
358 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
361 if (!kvm_arch_flush_remote_tlbs(kvm)
362 || kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
363 ++kvm->stat.generic.remote_tlb_flush;
365 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
367 void kvm_flush_remote_tlbs_range(struct kvm *kvm, gfn_t gfn, u64 nr_pages)
369 if (!kvm_arch_flush_remote_tlbs_range(kvm, gfn, nr_pages))
373 * Fall back to a flushing entire TLBs if the architecture range-based
374 * TLB invalidation is unsupported or can't be performed for whatever
377 kvm_flush_remote_tlbs(kvm);
380 void kvm_flush_remote_tlbs_memslot(struct kvm *kvm,
381 const struct kvm_memory_slot *memslot)
384 * All current use cases for flushing the TLBs for a specific memslot
385 * are related to dirty logging, and many do the TLB flush out of
386 * mmu_lock. The interaction between the various operations on memslot
387 * must be serialized by slots_locks to ensure the TLB flush from one
388 * operation is observed by any other operation on the same memslot.
390 lockdep_assert_held(&kvm->slots_lock);
391 kvm_flush_remote_tlbs_range(kvm, memslot->base_gfn, memslot->npages);
394 static void kvm_flush_shadow_all(struct kvm *kvm)
396 kvm_arch_flush_shadow_all(kvm);
397 kvm_arch_guest_memory_reclaimed(kvm);
400 #ifdef KVM_ARCH_NR_OBJS_PER_MEMORY_CACHE
401 static inline void *mmu_memory_cache_alloc_obj(struct kvm_mmu_memory_cache *mc,
404 gfp_flags |= mc->gfp_zero;
407 return kmem_cache_alloc(mc->kmem_cache, gfp_flags);
409 return (void *)__get_free_page(gfp_flags);
412 int __kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache *mc, int capacity, int min)
414 gfp_t gfp = mc->gfp_custom ? mc->gfp_custom : GFP_KERNEL_ACCOUNT;
417 if (mc->nobjs >= min)
420 if (unlikely(!mc->objects)) {
421 if (WARN_ON_ONCE(!capacity))
424 mc->objects = kvmalloc_array(sizeof(void *), capacity, gfp);
428 mc->capacity = capacity;
431 /* It is illegal to request a different capacity across topups. */
432 if (WARN_ON_ONCE(mc->capacity != capacity))
435 while (mc->nobjs < mc->capacity) {
436 obj = mmu_memory_cache_alloc_obj(mc, gfp);
438 return mc->nobjs >= min ? 0 : -ENOMEM;
439 mc->objects[mc->nobjs++] = obj;
444 int kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache *mc, int min)
446 return __kvm_mmu_topup_memory_cache(mc, KVM_ARCH_NR_OBJS_PER_MEMORY_CACHE, min);
449 int kvm_mmu_memory_cache_nr_free_objects(struct kvm_mmu_memory_cache *mc)
454 void kvm_mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc)
458 kmem_cache_free(mc->kmem_cache, mc->objects[--mc->nobjs]);
460 free_page((unsigned long)mc->objects[--mc->nobjs]);
469 void *kvm_mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc)
473 if (WARN_ON(!mc->nobjs))
474 p = mmu_memory_cache_alloc_obj(mc, GFP_ATOMIC | __GFP_ACCOUNT);
476 p = mc->objects[--mc->nobjs];
482 static void kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
484 mutex_init(&vcpu->mutex);
489 #ifndef __KVM_HAVE_ARCH_WQP
490 rcuwait_init(&vcpu->wait);
492 kvm_async_pf_vcpu_init(vcpu);
494 kvm_vcpu_set_in_spin_loop(vcpu, false);
495 kvm_vcpu_set_dy_eligible(vcpu, false);
496 vcpu->preempted = false;
498 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
499 vcpu->last_used_slot = NULL;
501 /* Fill the stats id string for the vcpu */
502 snprintf(vcpu->stats_id, sizeof(vcpu->stats_id), "kvm-%d/vcpu-%d",
503 task_pid_nr(current), id);
506 static void kvm_vcpu_destroy(struct kvm_vcpu *vcpu)
508 kvm_arch_vcpu_destroy(vcpu);
509 kvm_dirty_ring_free(&vcpu->dirty_ring);
512 * No need for rcu_read_lock as VCPU_RUN is the only place that changes
513 * the vcpu->pid pointer, and at destruction time all file descriptors
516 put_pid(rcu_dereference_protected(vcpu->pid, 1));
518 free_page((unsigned long)vcpu->run);
519 kmem_cache_free(kvm_vcpu_cache, vcpu);
522 void kvm_destroy_vcpus(struct kvm *kvm)
525 struct kvm_vcpu *vcpu;
527 kvm_for_each_vcpu(i, vcpu, kvm) {
528 kvm_vcpu_destroy(vcpu);
529 xa_erase(&kvm->vcpu_array, i);
532 atomic_set(&kvm->online_vcpus, 0);
534 EXPORT_SYMBOL_GPL(kvm_destroy_vcpus);
536 #ifdef CONFIG_KVM_GENERIC_MMU_NOTIFIER
537 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
539 return container_of(mn, struct kvm, mmu_notifier);
542 typedef bool (*gfn_handler_t)(struct kvm *kvm, struct kvm_gfn_range *range);
544 typedef void (*on_lock_fn_t)(struct kvm *kvm);
546 struct kvm_mmu_notifier_range {
548 * 64-bit addresses, as KVM notifiers can operate on host virtual
549 * addresses (unsigned long) and guest physical addresses (64-bit).
553 union kvm_mmu_notifier_arg arg;
554 gfn_handler_t handler;
555 on_lock_fn_t on_lock;
561 * The inner-most helper returns a tuple containing the return value from the
562 * arch- and action-specific handler, plus a flag indicating whether or not at
563 * least one memslot was found, i.e. if the handler found guest memory.
565 * Note, most notifiers are averse to booleans, so even though KVM tracks the
566 * return from arch code as a bool, outer helpers will cast it to an int. :-(
568 typedef struct kvm_mmu_notifier_return {
574 * Use a dedicated stub instead of NULL to indicate that there is no callback
575 * function/handler. The compiler technically can't guarantee that a real
576 * function will have a non-zero address, and so it will generate code to
577 * check for !NULL, whereas comparing against a stub will be elided at compile
578 * time (unless the compiler is getting long in the tooth, e.g. gcc 4.9).
580 static void kvm_null_fn(void)
584 #define IS_KVM_NULL_FN(fn) ((fn) == (void *)kvm_null_fn)
586 static const union kvm_mmu_notifier_arg KVM_MMU_NOTIFIER_NO_ARG;
588 /* Iterate over each memslot intersecting [start, last] (inclusive) range */
589 #define kvm_for_each_memslot_in_hva_range(node, slots, start, last) \
590 for (node = interval_tree_iter_first(&slots->hva_tree, start, last); \
592 node = interval_tree_iter_next(node, start, last)) \
594 static __always_inline kvm_mn_ret_t __kvm_handle_hva_range(struct kvm *kvm,
595 const struct kvm_mmu_notifier_range *range)
597 struct kvm_mmu_notifier_return r = {
599 .found_memslot = false,
601 struct kvm_gfn_range gfn_range;
602 struct kvm_memory_slot *slot;
603 struct kvm_memslots *slots;
606 if (WARN_ON_ONCE(range->end <= range->start))
609 /* A null handler is allowed if and only if on_lock() is provided. */
610 if (WARN_ON_ONCE(IS_KVM_NULL_FN(range->on_lock) &&
611 IS_KVM_NULL_FN(range->handler)))
614 idx = srcu_read_lock(&kvm->srcu);
616 for (i = 0; i < kvm_arch_nr_memslot_as_ids(kvm); i++) {
617 struct interval_tree_node *node;
619 slots = __kvm_memslots(kvm, i);
620 kvm_for_each_memslot_in_hva_range(node, slots,
621 range->start, range->end - 1) {
622 unsigned long hva_start, hva_end;
624 slot = container_of(node, struct kvm_memory_slot, hva_node[slots->node_idx]);
625 hva_start = max_t(unsigned long, range->start, slot->userspace_addr);
626 hva_end = min_t(unsigned long, range->end,
627 slot->userspace_addr + (slot->npages << PAGE_SHIFT));
630 * To optimize for the likely case where the address
631 * range is covered by zero or one memslots, don't
632 * bother making these conditional (to avoid writes on
633 * the second or later invocation of the handler).
635 gfn_range.arg = range->arg;
636 gfn_range.may_block = range->may_block;
639 * {gfn(page) | page intersects with [hva_start, hva_end)} =
640 * {gfn_start, gfn_start+1, ..., gfn_end-1}.
642 gfn_range.start = hva_to_gfn_memslot(hva_start, slot);
643 gfn_range.end = hva_to_gfn_memslot(hva_end + PAGE_SIZE - 1, slot);
644 gfn_range.slot = slot;
646 if (!r.found_memslot) {
647 r.found_memslot = true;
649 if (!IS_KVM_NULL_FN(range->on_lock))
652 if (IS_KVM_NULL_FN(range->handler))
655 r.ret |= range->handler(kvm, &gfn_range);
659 if (range->flush_on_ret && r.ret)
660 kvm_flush_remote_tlbs(kvm);
665 srcu_read_unlock(&kvm->srcu, idx);
670 static __always_inline int kvm_handle_hva_range(struct mmu_notifier *mn,
673 union kvm_mmu_notifier_arg arg,
674 gfn_handler_t handler)
676 struct kvm *kvm = mmu_notifier_to_kvm(mn);
677 const struct kvm_mmu_notifier_range range = {
682 .on_lock = (void *)kvm_null_fn,
683 .flush_on_ret = true,
687 return __kvm_handle_hva_range(kvm, &range).ret;
690 static __always_inline int kvm_handle_hva_range_no_flush(struct mmu_notifier *mn,
693 gfn_handler_t handler)
695 struct kvm *kvm = mmu_notifier_to_kvm(mn);
696 const struct kvm_mmu_notifier_range range = {
700 .on_lock = (void *)kvm_null_fn,
701 .flush_on_ret = false,
705 return __kvm_handle_hva_range(kvm, &range).ret;
708 static bool kvm_change_spte_gfn(struct kvm *kvm, struct kvm_gfn_range *range)
711 * Skipping invalid memslots is correct if and only change_pte() is
712 * surrounded by invalidate_range_{start,end}(), which is currently
713 * guaranteed by the primary MMU. If that ever changes, KVM needs to
714 * unmap the memslot instead of skipping the memslot to ensure that KVM
715 * doesn't hold references to the old PFN.
717 WARN_ON_ONCE(!READ_ONCE(kvm->mn_active_invalidate_count));
719 if (range->slot->flags & KVM_MEMSLOT_INVALID)
722 return kvm_set_spte_gfn(kvm, range);
725 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
726 struct mm_struct *mm,
727 unsigned long address,
730 struct kvm *kvm = mmu_notifier_to_kvm(mn);
731 const union kvm_mmu_notifier_arg arg = { .pte = pte };
733 trace_kvm_set_spte_hva(address);
736 * .change_pte() must be surrounded by .invalidate_range_{start,end}().
737 * If mmu_invalidate_in_progress is zero, then no in-progress
738 * invalidations, including this one, found a relevant memslot at
739 * start(); rechecking memslots here is unnecessary. Note, a false
740 * positive (count elevated by a different invalidation) is sub-optimal
741 * but functionally ok.
743 WARN_ON_ONCE(!READ_ONCE(kvm->mn_active_invalidate_count));
744 if (!READ_ONCE(kvm->mmu_invalidate_in_progress))
747 kvm_handle_hva_range(mn, address, address + 1, arg, kvm_change_spte_gfn);
750 void kvm_mmu_invalidate_begin(struct kvm *kvm)
752 lockdep_assert_held_write(&kvm->mmu_lock);
754 * The count increase must become visible at unlock time as no
755 * spte can be established without taking the mmu_lock and
756 * count is also read inside the mmu_lock critical section.
758 kvm->mmu_invalidate_in_progress++;
760 if (likely(kvm->mmu_invalidate_in_progress == 1)) {
761 kvm->mmu_invalidate_range_start = INVALID_GPA;
762 kvm->mmu_invalidate_range_end = INVALID_GPA;
766 void kvm_mmu_invalidate_range_add(struct kvm *kvm, gfn_t start, gfn_t end)
768 lockdep_assert_held_write(&kvm->mmu_lock);
770 WARN_ON_ONCE(!kvm->mmu_invalidate_in_progress);
772 if (likely(kvm->mmu_invalidate_range_start == INVALID_GPA)) {
773 kvm->mmu_invalidate_range_start = start;
774 kvm->mmu_invalidate_range_end = end;
777 * Fully tracking multiple concurrent ranges has diminishing
778 * returns. Keep things simple and just find the minimal range
779 * which includes the current and new ranges. As there won't be
780 * enough information to subtract a range after its invalidate
781 * completes, any ranges invalidated concurrently will
782 * accumulate and persist until all outstanding invalidates
785 kvm->mmu_invalidate_range_start =
786 min(kvm->mmu_invalidate_range_start, start);
787 kvm->mmu_invalidate_range_end =
788 max(kvm->mmu_invalidate_range_end, end);
792 bool kvm_mmu_unmap_gfn_range(struct kvm *kvm, struct kvm_gfn_range *range)
794 kvm_mmu_invalidate_range_add(kvm, range->start, range->end);
795 return kvm_unmap_gfn_range(kvm, range);
798 static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
799 const struct mmu_notifier_range *range)
801 struct kvm *kvm = mmu_notifier_to_kvm(mn);
802 const struct kvm_mmu_notifier_range hva_range = {
803 .start = range->start,
805 .handler = kvm_mmu_unmap_gfn_range,
806 .on_lock = kvm_mmu_invalidate_begin,
807 .flush_on_ret = true,
808 .may_block = mmu_notifier_range_blockable(range),
811 trace_kvm_unmap_hva_range(range->start, range->end);
814 * Prevent memslot modification between range_start() and range_end()
815 * so that conditionally locking provides the same result in both
816 * functions. Without that guarantee, the mmu_invalidate_in_progress
817 * adjustments will be imbalanced.
819 * Pairs with the decrement in range_end().
821 spin_lock(&kvm->mn_invalidate_lock);
822 kvm->mn_active_invalidate_count++;
823 spin_unlock(&kvm->mn_invalidate_lock);
826 * Invalidate pfn caches _before_ invalidating the secondary MMUs, i.e.
827 * before acquiring mmu_lock, to avoid holding mmu_lock while acquiring
828 * each cache's lock. There are relatively few caches in existence at
829 * any given time, and the caches themselves can check for hva overlap,
830 * i.e. don't need to rely on memslot overlap checks for performance.
831 * Because this runs without holding mmu_lock, the pfn caches must use
832 * mn_active_invalidate_count (see above) instead of
833 * mmu_invalidate_in_progress.
835 gfn_to_pfn_cache_invalidate_start(kvm, range->start, range->end,
836 hva_range.may_block);
839 * If one or more memslots were found and thus zapped, notify arch code
840 * that guest memory has been reclaimed. This needs to be done *after*
841 * dropping mmu_lock, as x86's reclaim path is slooooow.
843 if (__kvm_handle_hva_range(kvm, &hva_range).found_memslot)
844 kvm_arch_guest_memory_reclaimed(kvm);
849 void kvm_mmu_invalidate_end(struct kvm *kvm)
851 lockdep_assert_held_write(&kvm->mmu_lock);
854 * This sequence increase will notify the kvm page fault that
855 * the page that is going to be mapped in the spte could have
858 kvm->mmu_invalidate_seq++;
861 * The above sequence increase must be visible before the
862 * below count decrease, which is ensured by the smp_wmb above
863 * in conjunction with the smp_rmb in mmu_invalidate_retry().
865 kvm->mmu_invalidate_in_progress--;
866 KVM_BUG_ON(kvm->mmu_invalidate_in_progress < 0, kvm);
869 * Assert that at least one range was added between start() and end().
870 * Not adding a range isn't fatal, but it is a KVM bug.
872 WARN_ON_ONCE(kvm->mmu_invalidate_range_start == INVALID_GPA);
875 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
876 const struct mmu_notifier_range *range)
878 struct kvm *kvm = mmu_notifier_to_kvm(mn);
879 const struct kvm_mmu_notifier_range hva_range = {
880 .start = range->start,
882 .handler = (void *)kvm_null_fn,
883 .on_lock = kvm_mmu_invalidate_end,
884 .flush_on_ret = false,
885 .may_block = mmu_notifier_range_blockable(range),
889 __kvm_handle_hva_range(kvm, &hva_range);
891 /* Pairs with the increment in range_start(). */
892 spin_lock(&kvm->mn_invalidate_lock);
893 wake = (--kvm->mn_active_invalidate_count == 0);
894 spin_unlock(&kvm->mn_invalidate_lock);
897 * There can only be one waiter, since the wait happens under
901 rcuwait_wake_up(&kvm->mn_memslots_update_rcuwait);
904 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
905 struct mm_struct *mm,
909 trace_kvm_age_hva(start, end);
911 return kvm_handle_hva_range(mn, start, end, KVM_MMU_NOTIFIER_NO_ARG,
915 static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
916 struct mm_struct *mm,
920 trace_kvm_age_hva(start, end);
923 * Even though we do not flush TLB, this will still adversely
924 * affect performance on pre-Haswell Intel EPT, where there is
925 * no EPT Access Bit to clear so that we have to tear down EPT
926 * tables instead. If we find this unacceptable, we can always
927 * add a parameter to kvm_age_hva so that it effectively doesn't
928 * do anything on clear_young.
930 * Also note that currently we never issue secondary TLB flushes
931 * from clear_young, leaving this job up to the regular system
932 * cadence. If we find this inaccurate, we might come up with a
933 * more sophisticated heuristic later.
935 return kvm_handle_hva_range_no_flush(mn, start, end, kvm_age_gfn);
938 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
939 struct mm_struct *mm,
940 unsigned long address)
942 trace_kvm_test_age_hva(address);
944 return kvm_handle_hva_range_no_flush(mn, address, address + 1,
948 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
949 struct mm_struct *mm)
951 struct kvm *kvm = mmu_notifier_to_kvm(mn);
954 idx = srcu_read_lock(&kvm->srcu);
955 kvm_flush_shadow_all(kvm);
956 srcu_read_unlock(&kvm->srcu, idx);
959 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
960 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
961 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
962 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
963 .clear_young = kvm_mmu_notifier_clear_young,
964 .test_young = kvm_mmu_notifier_test_young,
965 .change_pte = kvm_mmu_notifier_change_pte,
966 .release = kvm_mmu_notifier_release,
969 static int kvm_init_mmu_notifier(struct kvm *kvm)
971 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
972 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
975 #else /* !CONFIG_KVM_GENERIC_MMU_NOTIFIER */
977 static int kvm_init_mmu_notifier(struct kvm *kvm)
982 #endif /* CONFIG_KVM_GENERIC_MMU_NOTIFIER */
984 #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER
985 static int kvm_pm_notifier_call(struct notifier_block *bl,
989 struct kvm *kvm = container_of(bl, struct kvm, pm_notifier);
991 return kvm_arch_pm_notifier(kvm, state);
994 static void kvm_init_pm_notifier(struct kvm *kvm)
996 kvm->pm_notifier.notifier_call = kvm_pm_notifier_call;
997 /* Suspend KVM before we suspend ftrace, RCU, etc. */
998 kvm->pm_notifier.priority = INT_MAX;
999 register_pm_notifier(&kvm->pm_notifier);
1002 static void kvm_destroy_pm_notifier(struct kvm *kvm)
1004 unregister_pm_notifier(&kvm->pm_notifier);
1006 #else /* !CONFIG_HAVE_KVM_PM_NOTIFIER */
1007 static void kvm_init_pm_notifier(struct kvm *kvm)
1011 static void kvm_destroy_pm_notifier(struct kvm *kvm)
1014 #endif /* CONFIG_HAVE_KVM_PM_NOTIFIER */
1016 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
1018 if (!memslot->dirty_bitmap)
1021 kvfree(memslot->dirty_bitmap);
1022 memslot->dirty_bitmap = NULL;
1025 /* This does not remove the slot from struct kvm_memslots data structures */
1026 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
1028 if (slot->flags & KVM_MEM_GUEST_MEMFD)
1029 kvm_gmem_unbind(slot);
1031 kvm_destroy_dirty_bitmap(slot);
1033 kvm_arch_free_memslot(kvm, slot);
1038 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
1040 struct hlist_node *idnode;
1041 struct kvm_memory_slot *memslot;
1045 * The same memslot objects live in both active and inactive sets,
1046 * arbitrarily free using index '1' so the second invocation of this
1047 * function isn't operating over a structure with dangling pointers
1048 * (even though this function isn't actually touching them).
1050 if (!slots->node_idx)
1053 hash_for_each_safe(slots->id_hash, bkt, idnode, memslot, id_node[1])
1054 kvm_free_memslot(kvm, memslot);
1057 static umode_t kvm_stats_debugfs_mode(const struct _kvm_stats_desc *pdesc)
1059 switch (pdesc->desc.flags & KVM_STATS_TYPE_MASK) {
1060 case KVM_STATS_TYPE_INSTANT:
1062 case KVM_STATS_TYPE_CUMULATIVE:
1063 case KVM_STATS_TYPE_PEAK:
1070 static void kvm_destroy_vm_debugfs(struct kvm *kvm)
1073 int kvm_debugfs_num_entries = kvm_vm_stats_header.num_desc +
1074 kvm_vcpu_stats_header.num_desc;
1076 if (IS_ERR(kvm->debugfs_dentry))
1079 debugfs_remove_recursive(kvm->debugfs_dentry);
1081 if (kvm->debugfs_stat_data) {
1082 for (i = 0; i < kvm_debugfs_num_entries; i++)
1083 kfree(kvm->debugfs_stat_data[i]);
1084 kfree(kvm->debugfs_stat_data);
1088 static int kvm_create_vm_debugfs(struct kvm *kvm, const char *fdname)
1090 static DEFINE_MUTEX(kvm_debugfs_lock);
1091 struct dentry *dent;
1092 char dir_name[ITOA_MAX_LEN * 2];
1093 struct kvm_stat_data *stat_data;
1094 const struct _kvm_stats_desc *pdesc;
1095 int i, ret = -ENOMEM;
1096 int kvm_debugfs_num_entries = kvm_vm_stats_header.num_desc +
1097 kvm_vcpu_stats_header.num_desc;
1099 if (!debugfs_initialized())
1102 snprintf(dir_name, sizeof(dir_name), "%d-%s", task_pid_nr(current), fdname);
1103 mutex_lock(&kvm_debugfs_lock);
1104 dent = debugfs_lookup(dir_name, kvm_debugfs_dir);
1106 pr_warn_ratelimited("KVM: debugfs: duplicate directory %s\n", dir_name);
1108 mutex_unlock(&kvm_debugfs_lock);
1111 dent = debugfs_create_dir(dir_name, kvm_debugfs_dir);
1112 mutex_unlock(&kvm_debugfs_lock);
1116 kvm->debugfs_dentry = dent;
1117 kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries,
1118 sizeof(*kvm->debugfs_stat_data),
1119 GFP_KERNEL_ACCOUNT);
1120 if (!kvm->debugfs_stat_data)
1123 for (i = 0; i < kvm_vm_stats_header.num_desc; ++i) {
1124 pdesc = &kvm_vm_stats_desc[i];
1125 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT);
1129 stat_data->kvm = kvm;
1130 stat_data->desc = pdesc;
1131 stat_data->kind = KVM_STAT_VM;
1132 kvm->debugfs_stat_data[i] = stat_data;
1133 debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
1134 kvm->debugfs_dentry, stat_data,
1138 for (i = 0; i < kvm_vcpu_stats_header.num_desc; ++i) {
1139 pdesc = &kvm_vcpu_stats_desc[i];
1140 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT);
1144 stat_data->kvm = kvm;
1145 stat_data->desc = pdesc;
1146 stat_data->kind = KVM_STAT_VCPU;
1147 kvm->debugfs_stat_data[i + kvm_vm_stats_header.num_desc] = stat_data;
1148 debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
1149 kvm->debugfs_dentry, stat_data,
1153 ret = kvm_arch_create_vm_debugfs(kvm);
1159 kvm_destroy_vm_debugfs(kvm);
1164 * Called after the VM is otherwise initialized, but just before adding it to
1167 int __weak kvm_arch_post_init_vm(struct kvm *kvm)
1173 * Called just after removing the VM from the vm_list, but before doing any
1174 * other destruction.
1176 void __weak kvm_arch_pre_destroy_vm(struct kvm *kvm)
1181 * Called after per-vm debugfs created. When called kvm->debugfs_dentry should
1182 * be setup already, so we can create arch-specific debugfs entries under it.
1183 * Cleanup should be automatic done in kvm_destroy_vm_debugfs() recursively, so
1184 * a per-arch destroy interface is not needed.
1186 int __weak kvm_arch_create_vm_debugfs(struct kvm *kvm)
1191 static struct kvm *kvm_create_vm(unsigned long type, const char *fdname)
1193 struct kvm *kvm = kvm_arch_alloc_vm();
1194 struct kvm_memslots *slots;
1199 return ERR_PTR(-ENOMEM);
1201 KVM_MMU_LOCK_INIT(kvm);
1202 mmgrab(current->mm);
1203 kvm->mm = current->mm;
1204 kvm_eventfd_init(kvm);
1205 mutex_init(&kvm->lock);
1206 mutex_init(&kvm->irq_lock);
1207 mutex_init(&kvm->slots_lock);
1208 mutex_init(&kvm->slots_arch_lock);
1209 spin_lock_init(&kvm->mn_invalidate_lock);
1210 rcuwait_init(&kvm->mn_memslots_update_rcuwait);
1211 xa_init(&kvm->vcpu_array);
1212 #ifdef CONFIG_KVM_GENERIC_MEMORY_ATTRIBUTES
1213 xa_init(&kvm->mem_attr_array);
1216 INIT_LIST_HEAD(&kvm->gpc_list);
1217 spin_lock_init(&kvm->gpc_lock);
1219 INIT_LIST_HEAD(&kvm->devices);
1220 kvm->max_vcpus = KVM_MAX_VCPUS;
1222 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
1225 * Force subsequent debugfs file creations to fail if the VM directory
1226 * is not created (by kvm_create_vm_debugfs()).
1228 kvm->debugfs_dentry = ERR_PTR(-ENOENT);
1230 snprintf(kvm->stats_id, sizeof(kvm->stats_id), "kvm-%d",
1231 task_pid_nr(current));
1233 if (init_srcu_struct(&kvm->srcu))
1234 goto out_err_no_srcu;
1235 if (init_srcu_struct(&kvm->irq_srcu))
1236 goto out_err_no_irq_srcu;
1238 refcount_set(&kvm->users_count, 1);
1239 for (i = 0; i < kvm_arch_nr_memslot_as_ids(kvm); i++) {
1240 for (j = 0; j < 2; j++) {
1241 slots = &kvm->__memslots[i][j];
1243 atomic_long_set(&slots->last_used_slot, (unsigned long)NULL);
1244 slots->hva_tree = RB_ROOT_CACHED;
1245 slots->gfn_tree = RB_ROOT;
1246 hash_init(slots->id_hash);
1247 slots->node_idx = j;
1249 /* Generations must be different for each address space. */
1250 slots->generation = i;
1253 rcu_assign_pointer(kvm->memslots[i], &kvm->__memslots[i][0]);
1256 for (i = 0; i < KVM_NR_BUSES; i++) {
1257 rcu_assign_pointer(kvm->buses[i],
1258 kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL_ACCOUNT));
1260 goto out_err_no_arch_destroy_vm;
1263 r = kvm_arch_init_vm(kvm, type);
1265 goto out_err_no_arch_destroy_vm;
1267 r = hardware_enable_all();
1269 goto out_err_no_disable;
1271 #ifdef CONFIG_HAVE_KVM_IRQCHIP
1272 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
1275 r = kvm_init_mmu_notifier(kvm);
1277 goto out_err_no_mmu_notifier;
1279 r = kvm_coalesced_mmio_init(kvm);
1281 goto out_no_coalesced_mmio;
1283 r = kvm_create_vm_debugfs(kvm, fdname);
1285 goto out_err_no_debugfs;
1287 r = kvm_arch_post_init_vm(kvm);
1291 mutex_lock(&kvm_lock);
1292 list_add(&kvm->vm_list, &vm_list);
1293 mutex_unlock(&kvm_lock);
1295 preempt_notifier_inc();
1296 kvm_init_pm_notifier(kvm);
1301 kvm_destroy_vm_debugfs(kvm);
1303 kvm_coalesced_mmio_free(kvm);
1304 out_no_coalesced_mmio:
1305 #ifdef CONFIG_KVM_GENERIC_MMU_NOTIFIER
1306 if (kvm->mmu_notifier.ops)
1307 mmu_notifier_unregister(&kvm->mmu_notifier, current->mm);
1309 out_err_no_mmu_notifier:
1310 hardware_disable_all();
1312 kvm_arch_destroy_vm(kvm);
1313 out_err_no_arch_destroy_vm:
1314 WARN_ON_ONCE(!refcount_dec_and_test(&kvm->users_count));
1315 for (i = 0; i < KVM_NR_BUSES; i++)
1316 kfree(kvm_get_bus(kvm, i));
1317 cleanup_srcu_struct(&kvm->irq_srcu);
1318 out_err_no_irq_srcu:
1319 cleanup_srcu_struct(&kvm->srcu);
1321 kvm_arch_free_vm(kvm);
1322 mmdrop(current->mm);
1326 static void kvm_destroy_devices(struct kvm *kvm)
1328 struct kvm_device *dev, *tmp;
1331 * We do not need to take the kvm->lock here, because nobody else
1332 * has a reference to the struct kvm at this point and therefore
1333 * cannot access the devices list anyhow.
1335 list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
1336 list_del(&dev->vm_node);
1337 dev->ops->destroy(dev);
1341 static void kvm_destroy_vm(struct kvm *kvm)
1344 struct mm_struct *mm = kvm->mm;
1346 kvm_destroy_pm_notifier(kvm);
1347 kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm);
1348 kvm_destroy_vm_debugfs(kvm);
1349 kvm_arch_sync_events(kvm);
1350 mutex_lock(&kvm_lock);
1351 list_del(&kvm->vm_list);
1352 mutex_unlock(&kvm_lock);
1353 kvm_arch_pre_destroy_vm(kvm);
1355 kvm_free_irq_routing(kvm);
1356 for (i = 0; i < KVM_NR_BUSES; i++) {
1357 struct kvm_io_bus *bus = kvm_get_bus(kvm, i);
1360 kvm_io_bus_destroy(bus);
1361 kvm->buses[i] = NULL;
1363 kvm_coalesced_mmio_free(kvm);
1364 #ifdef CONFIG_KVM_GENERIC_MMU_NOTIFIER
1365 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
1367 * At this point, pending calls to invalidate_range_start()
1368 * have completed but no more MMU notifiers will run, so
1369 * mn_active_invalidate_count may remain unbalanced.
1370 * No threads can be waiting in kvm_swap_active_memslots() as the
1371 * last reference on KVM has been dropped, but freeing
1372 * memslots would deadlock without this manual intervention.
1374 * If the count isn't unbalanced, i.e. KVM did NOT unregister its MMU
1375 * notifier between a start() and end(), then there shouldn't be any
1376 * in-progress invalidations.
1378 WARN_ON(rcuwait_active(&kvm->mn_memslots_update_rcuwait));
1379 if (kvm->mn_active_invalidate_count)
1380 kvm->mn_active_invalidate_count = 0;
1382 WARN_ON(kvm->mmu_invalidate_in_progress);
1384 kvm_flush_shadow_all(kvm);
1386 kvm_arch_destroy_vm(kvm);
1387 kvm_destroy_devices(kvm);
1388 for (i = 0; i < kvm_arch_nr_memslot_as_ids(kvm); i++) {
1389 kvm_free_memslots(kvm, &kvm->__memslots[i][0]);
1390 kvm_free_memslots(kvm, &kvm->__memslots[i][1]);
1392 cleanup_srcu_struct(&kvm->irq_srcu);
1393 cleanup_srcu_struct(&kvm->srcu);
1394 #ifdef CONFIG_KVM_GENERIC_MEMORY_ATTRIBUTES
1395 xa_destroy(&kvm->mem_attr_array);
1397 kvm_arch_free_vm(kvm);
1398 preempt_notifier_dec();
1399 hardware_disable_all();
1403 void kvm_get_kvm(struct kvm *kvm)
1405 refcount_inc(&kvm->users_count);
1407 EXPORT_SYMBOL_GPL(kvm_get_kvm);
1410 * Make sure the vm is not during destruction, which is a safe version of
1411 * kvm_get_kvm(). Return true if kvm referenced successfully, false otherwise.
1413 bool kvm_get_kvm_safe(struct kvm *kvm)
1415 return refcount_inc_not_zero(&kvm->users_count);
1417 EXPORT_SYMBOL_GPL(kvm_get_kvm_safe);
1419 void kvm_put_kvm(struct kvm *kvm)
1421 if (refcount_dec_and_test(&kvm->users_count))
1422 kvm_destroy_vm(kvm);
1424 EXPORT_SYMBOL_GPL(kvm_put_kvm);
1427 * Used to put a reference that was taken on behalf of an object associated
1428 * with a user-visible file descriptor, e.g. a vcpu or device, if installation
1429 * of the new file descriptor fails and the reference cannot be transferred to
1430 * its final owner. In such cases, the caller is still actively using @kvm and
1431 * will fail miserably if the refcount unexpectedly hits zero.
1433 void kvm_put_kvm_no_destroy(struct kvm *kvm)
1435 WARN_ON(refcount_dec_and_test(&kvm->users_count));
1437 EXPORT_SYMBOL_GPL(kvm_put_kvm_no_destroy);
1439 static int kvm_vm_release(struct inode *inode, struct file *filp)
1441 struct kvm *kvm = filp->private_data;
1443 kvm_irqfd_release(kvm);
1450 * Allocation size is twice as large as the actual dirty bitmap size.
1451 * See kvm_vm_ioctl_get_dirty_log() why this is needed.
1453 static int kvm_alloc_dirty_bitmap(struct kvm_memory_slot *memslot)
1455 unsigned long dirty_bytes = kvm_dirty_bitmap_bytes(memslot);
1457 memslot->dirty_bitmap = __vcalloc(2, dirty_bytes, GFP_KERNEL_ACCOUNT);
1458 if (!memslot->dirty_bitmap)
1464 static struct kvm_memslots *kvm_get_inactive_memslots(struct kvm *kvm, int as_id)
1466 struct kvm_memslots *active = __kvm_memslots(kvm, as_id);
1467 int node_idx_inactive = active->node_idx ^ 1;
1469 return &kvm->__memslots[as_id][node_idx_inactive];
1473 * Helper to get the address space ID when one of memslot pointers may be NULL.
1474 * This also serves as a sanity that at least one of the pointers is non-NULL,
1475 * and that their address space IDs don't diverge.
1477 static int kvm_memslots_get_as_id(struct kvm_memory_slot *a,
1478 struct kvm_memory_slot *b)
1480 if (WARN_ON_ONCE(!a && !b))
1488 WARN_ON_ONCE(a->as_id != b->as_id);
1492 static void kvm_insert_gfn_node(struct kvm_memslots *slots,
1493 struct kvm_memory_slot *slot)
1495 struct rb_root *gfn_tree = &slots->gfn_tree;
1496 struct rb_node **node, *parent;
1497 int idx = slots->node_idx;
1500 for (node = &gfn_tree->rb_node; *node; ) {
1501 struct kvm_memory_slot *tmp;
1503 tmp = container_of(*node, struct kvm_memory_slot, gfn_node[idx]);
1505 if (slot->base_gfn < tmp->base_gfn)
1506 node = &(*node)->rb_left;
1507 else if (slot->base_gfn > tmp->base_gfn)
1508 node = &(*node)->rb_right;
1513 rb_link_node(&slot->gfn_node[idx], parent, node);
1514 rb_insert_color(&slot->gfn_node[idx], gfn_tree);
1517 static void kvm_erase_gfn_node(struct kvm_memslots *slots,
1518 struct kvm_memory_slot *slot)
1520 rb_erase(&slot->gfn_node[slots->node_idx], &slots->gfn_tree);
1523 static void kvm_replace_gfn_node(struct kvm_memslots *slots,
1524 struct kvm_memory_slot *old,
1525 struct kvm_memory_slot *new)
1527 int idx = slots->node_idx;
1529 WARN_ON_ONCE(old->base_gfn != new->base_gfn);
1531 rb_replace_node(&old->gfn_node[idx], &new->gfn_node[idx],
1536 * Replace @old with @new in the inactive memslots.
1538 * With NULL @old this simply adds @new.
1539 * With NULL @new this simply removes @old.
1541 * If @new is non-NULL its hva_node[slots_idx] range has to be set
1544 static void kvm_replace_memslot(struct kvm *kvm,
1545 struct kvm_memory_slot *old,
1546 struct kvm_memory_slot *new)
1548 int as_id = kvm_memslots_get_as_id(old, new);
1549 struct kvm_memslots *slots = kvm_get_inactive_memslots(kvm, as_id);
1550 int idx = slots->node_idx;
1553 hash_del(&old->id_node[idx]);
1554 interval_tree_remove(&old->hva_node[idx], &slots->hva_tree);
1556 if ((long)old == atomic_long_read(&slots->last_used_slot))
1557 atomic_long_set(&slots->last_used_slot, (long)new);
1560 kvm_erase_gfn_node(slots, old);
1566 * Initialize @new's hva range. Do this even when replacing an @old
1567 * slot, kvm_copy_memslot() deliberately does not touch node data.
1569 new->hva_node[idx].start = new->userspace_addr;
1570 new->hva_node[idx].last = new->userspace_addr +
1571 (new->npages << PAGE_SHIFT) - 1;
1574 * (Re)Add the new memslot. There is no O(1) interval_tree_replace(),
1575 * hva_node needs to be swapped with remove+insert even though hva can't
1576 * change when replacing an existing slot.
1578 hash_add(slots->id_hash, &new->id_node[idx], new->id);
1579 interval_tree_insert(&new->hva_node[idx], &slots->hva_tree);
1582 * If the memslot gfn is unchanged, rb_replace_node() can be used to
1583 * switch the node in the gfn tree instead of removing the old and
1584 * inserting the new as two separate operations. Replacement is a
1585 * single O(1) operation versus two O(log(n)) operations for
1588 if (old && old->base_gfn == new->base_gfn) {
1589 kvm_replace_gfn_node(slots, old, new);
1592 kvm_erase_gfn_node(slots, old);
1593 kvm_insert_gfn_node(slots, new);
1598 * Flags that do not access any of the extra space of struct
1599 * kvm_userspace_memory_region2. KVM_SET_USER_MEMORY_REGION_V1_FLAGS
1600 * only allows these.
1602 #define KVM_SET_USER_MEMORY_REGION_V1_FLAGS \
1603 (KVM_MEM_LOG_DIRTY_PAGES | KVM_MEM_READONLY)
1605 static int check_memory_region_flags(struct kvm *kvm,
1606 const struct kvm_userspace_memory_region2 *mem)
1608 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
1610 if (kvm_arch_has_private_mem(kvm))
1611 valid_flags |= KVM_MEM_GUEST_MEMFD;
1613 /* Dirty logging private memory is not currently supported. */
1614 if (mem->flags & KVM_MEM_GUEST_MEMFD)
1615 valid_flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
1617 #ifdef __KVM_HAVE_READONLY_MEM
1619 * GUEST_MEMFD is incompatible with read-only memslots, as writes to
1620 * read-only memslots have emulated MMIO, not page fault, semantics,
1621 * and KVM doesn't allow emulated MMIO for private memory.
1623 if (!(mem->flags & KVM_MEM_GUEST_MEMFD))
1624 valid_flags |= KVM_MEM_READONLY;
1627 if (mem->flags & ~valid_flags)
1633 static void kvm_swap_active_memslots(struct kvm *kvm, int as_id)
1635 struct kvm_memslots *slots = kvm_get_inactive_memslots(kvm, as_id);
1637 /* Grab the generation from the activate memslots. */
1638 u64 gen = __kvm_memslots(kvm, as_id)->generation;
1640 WARN_ON(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS);
1641 slots->generation = gen | KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
1644 * Do not store the new memslots while there are invalidations in
1645 * progress, otherwise the locking in invalidate_range_start and
1646 * invalidate_range_end will be unbalanced.
1648 spin_lock(&kvm->mn_invalidate_lock);
1649 prepare_to_rcuwait(&kvm->mn_memslots_update_rcuwait);
1650 while (kvm->mn_active_invalidate_count) {
1651 set_current_state(TASK_UNINTERRUPTIBLE);
1652 spin_unlock(&kvm->mn_invalidate_lock);
1654 spin_lock(&kvm->mn_invalidate_lock);
1656 finish_rcuwait(&kvm->mn_memslots_update_rcuwait);
1657 rcu_assign_pointer(kvm->memslots[as_id], slots);
1658 spin_unlock(&kvm->mn_invalidate_lock);
1661 * Acquired in kvm_set_memslot. Must be released before synchronize
1662 * SRCU below in order to avoid deadlock with another thread
1663 * acquiring the slots_arch_lock in an srcu critical section.
1665 mutex_unlock(&kvm->slots_arch_lock);
1667 synchronize_srcu_expedited(&kvm->srcu);
1670 * Increment the new memslot generation a second time, dropping the
1671 * update in-progress flag and incrementing the generation based on
1672 * the number of address spaces. This provides a unique and easily
1673 * identifiable generation number while the memslots are in flux.
1675 gen = slots->generation & ~KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
1678 * Generations must be unique even across address spaces. We do not need
1679 * a global counter for that, instead the generation space is evenly split
1680 * across address spaces. For example, with two address spaces, address
1681 * space 0 will use generations 0, 2, 4, ... while address space 1 will
1682 * use generations 1, 3, 5, ...
1684 gen += kvm_arch_nr_memslot_as_ids(kvm);
1686 kvm_arch_memslots_updated(kvm, gen);
1688 slots->generation = gen;
1691 static int kvm_prepare_memory_region(struct kvm *kvm,
1692 const struct kvm_memory_slot *old,
1693 struct kvm_memory_slot *new,
1694 enum kvm_mr_change change)
1699 * If dirty logging is disabled, nullify the bitmap; the old bitmap
1700 * will be freed on "commit". If logging is enabled in both old and
1701 * new, reuse the existing bitmap. If logging is enabled only in the
1702 * new and KVM isn't using a ring buffer, allocate and initialize a
1705 if (change != KVM_MR_DELETE) {
1706 if (!(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
1707 new->dirty_bitmap = NULL;
1708 else if (old && old->dirty_bitmap)
1709 new->dirty_bitmap = old->dirty_bitmap;
1710 else if (kvm_use_dirty_bitmap(kvm)) {
1711 r = kvm_alloc_dirty_bitmap(new);
1715 if (kvm_dirty_log_manual_protect_and_init_set(kvm))
1716 bitmap_set(new->dirty_bitmap, 0, new->npages);
1720 r = kvm_arch_prepare_memory_region(kvm, old, new, change);
1722 /* Free the bitmap on failure if it was allocated above. */
1723 if (r && new && new->dirty_bitmap && (!old || !old->dirty_bitmap))
1724 kvm_destroy_dirty_bitmap(new);
1729 static void kvm_commit_memory_region(struct kvm *kvm,
1730 struct kvm_memory_slot *old,
1731 const struct kvm_memory_slot *new,
1732 enum kvm_mr_change change)
1734 int old_flags = old ? old->flags : 0;
1735 int new_flags = new ? new->flags : 0;
1737 * Update the total number of memslot pages before calling the arch
1738 * hook so that architectures can consume the result directly.
1740 if (change == KVM_MR_DELETE)
1741 kvm->nr_memslot_pages -= old->npages;
1742 else if (change == KVM_MR_CREATE)
1743 kvm->nr_memslot_pages += new->npages;
1745 if ((old_flags ^ new_flags) & KVM_MEM_LOG_DIRTY_PAGES) {
1746 int change = (new_flags & KVM_MEM_LOG_DIRTY_PAGES) ? 1 : -1;
1747 atomic_set(&kvm->nr_memslots_dirty_logging,
1748 atomic_read(&kvm->nr_memslots_dirty_logging) + change);
1751 kvm_arch_commit_memory_region(kvm, old, new, change);
1755 /* Nothing more to do. */
1758 /* Free the old memslot and all its metadata. */
1759 kvm_free_memslot(kvm, old);
1762 case KVM_MR_FLAGS_ONLY:
1764 * Free the dirty bitmap as needed; the below check encompasses
1765 * both the flags and whether a ring buffer is being used)
1767 if (old->dirty_bitmap && !new->dirty_bitmap)
1768 kvm_destroy_dirty_bitmap(old);
1771 * The final quirk. Free the detached, old slot, but only its
1772 * memory, not any metadata. Metadata, including arch specific
1773 * data, may be reused by @new.
1783 * Activate @new, which must be installed in the inactive slots by the caller,
1784 * by swapping the active slots and then propagating @new to @old once @old is
1785 * unreachable and can be safely modified.
1787 * With NULL @old this simply adds @new to @active (while swapping the sets).
1788 * With NULL @new this simply removes @old from @active and frees it
1789 * (while also swapping the sets).
1791 static void kvm_activate_memslot(struct kvm *kvm,
1792 struct kvm_memory_slot *old,
1793 struct kvm_memory_slot *new)
1795 int as_id = kvm_memslots_get_as_id(old, new);
1797 kvm_swap_active_memslots(kvm, as_id);
1799 /* Propagate the new memslot to the now inactive memslots. */
1800 kvm_replace_memslot(kvm, old, new);
1803 static void kvm_copy_memslot(struct kvm_memory_slot *dest,
1804 const struct kvm_memory_slot *src)
1806 dest->base_gfn = src->base_gfn;
1807 dest->npages = src->npages;
1808 dest->dirty_bitmap = src->dirty_bitmap;
1809 dest->arch = src->arch;
1810 dest->userspace_addr = src->userspace_addr;
1811 dest->flags = src->flags;
1813 dest->as_id = src->as_id;
1816 static void kvm_invalidate_memslot(struct kvm *kvm,
1817 struct kvm_memory_slot *old,
1818 struct kvm_memory_slot *invalid_slot)
1821 * Mark the current slot INVALID. As with all memslot modifications,
1822 * this must be done on an unreachable slot to avoid modifying the
1823 * current slot in the active tree.
1825 kvm_copy_memslot(invalid_slot, old);
1826 invalid_slot->flags |= KVM_MEMSLOT_INVALID;
1827 kvm_replace_memslot(kvm, old, invalid_slot);
1830 * Activate the slot that is now marked INVALID, but don't propagate
1831 * the slot to the now inactive slots. The slot is either going to be
1832 * deleted or recreated as a new slot.
1834 kvm_swap_active_memslots(kvm, old->as_id);
1837 * From this point no new shadow pages pointing to a deleted, or moved,
1838 * memslot will be created. Validation of sp->gfn happens in:
1839 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1840 * - kvm_is_visible_gfn (mmu_check_root)
1842 kvm_arch_flush_shadow_memslot(kvm, old);
1843 kvm_arch_guest_memory_reclaimed(kvm);
1845 /* Was released by kvm_swap_active_memslots(), reacquire. */
1846 mutex_lock(&kvm->slots_arch_lock);
1849 * Copy the arch-specific field of the newly-installed slot back to the
1850 * old slot as the arch data could have changed between releasing
1851 * slots_arch_lock in kvm_swap_active_memslots() and re-acquiring the lock
1852 * above. Writers are required to retrieve memslots *after* acquiring
1853 * slots_arch_lock, thus the active slot's data is guaranteed to be fresh.
1855 old->arch = invalid_slot->arch;
1858 static void kvm_create_memslot(struct kvm *kvm,
1859 struct kvm_memory_slot *new)
1861 /* Add the new memslot to the inactive set and activate. */
1862 kvm_replace_memslot(kvm, NULL, new);
1863 kvm_activate_memslot(kvm, NULL, new);
1866 static void kvm_delete_memslot(struct kvm *kvm,
1867 struct kvm_memory_slot *old,
1868 struct kvm_memory_slot *invalid_slot)
1871 * Remove the old memslot (in the inactive memslots) by passing NULL as
1872 * the "new" slot, and for the invalid version in the active slots.
1874 kvm_replace_memslot(kvm, old, NULL);
1875 kvm_activate_memslot(kvm, invalid_slot, NULL);
1878 static void kvm_move_memslot(struct kvm *kvm,
1879 struct kvm_memory_slot *old,
1880 struct kvm_memory_slot *new,
1881 struct kvm_memory_slot *invalid_slot)
1884 * Replace the old memslot in the inactive slots, and then swap slots
1885 * and replace the current INVALID with the new as well.
1887 kvm_replace_memslot(kvm, old, new);
1888 kvm_activate_memslot(kvm, invalid_slot, new);
1891 static void kvm_update_flags_memslot(struct kvm *kvm,
1892 struct kvm_memory_slot *old,
1893 struct kvm_memory_slot *new)
1896 * Similar to the MOVE case, but the slot doesn't need to be zapped as
1897 * an intermediate step. Instead, the old memslot is simply replaced
1898 * with a new, updated copy in both memslot sets.
1900 kvm_replace_memslot(kvm, old, new);
1901 kvm_activate_memslot(kvm, old, new);
1904 static int kvm_set_memslot(struct kvm *kvm,
1905 struct kvm_memory_slot *old,
1906 struct kvm_memory_slot *new,
1907 enum kvm_mr_change change)
1909 struct kvm_memory_slot *invalid_slot;
1913 * Released in kvm_swap_active_memslots().
1915 * Must be held from before the current memslots are copied until after
1916 * the new memslots are installed with rcu_assign_pointer, then
1917 * released before the synchronize srcu in kvm_swap_active_memslots().
1919 * When modifying memslots outside of the slots_lock, must be held
1920 * before reading the pointer to the current memslots until after all
1921 * changes to those memslots are complete.
1923 * These rules ensure that installing new memslots does not lose
1924 * changes made to the previous memslots.
1926 mutex_lock(&kvm->slots_arch_lock);
1929 * Invalidate the old slot if it's being deleted or moved. This is
1930 * done prior to actually deleting/moving the memslot to allow vCPUs to
1931 * continue running by ensuring there are no mappings or shadow pages
1932 * for the memslot when it is deleted/moved. Without pre-invalidation
1933 * (and without a lock), a window would exist between effecting the
1934 * delete/move and committing the changes in arch code where KVM or a
1935 * guest could access a non-existent memslot.
1937 * Modifications are done on a temporary, unreachable slot. The old
1938 * slot needs to be preserved in case a later step fails and the
1939 * invalidation needs to be reverted.
1941 if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) {
1942 invalid_slot = kzalloc(sizeof(*invalid_slot), GFP_KERNEL_ACCOUNT);
1943 if (!invalid_slot) {
1944 mutex_unlock(&kvm->slots_arch_lock);
1947 kvm_invalidate_memslot(kvm, old, invalid_slot);
1950 r = kvm_prepare_memory_region(kvm, old, new, change);
1953 * For DELETE/MOVE, revert the above INVALID change. No
1954 * modifications required since the original slot was preserved
1955 * in the inactive slots. Changing the active memslots also
1956 * release slots_arch_lock.
1958 if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) {
1959 kvm_activate_memslot(kvm, invalid_slot, old);
1960 kfree(invalid_slot);
1962 mutex_unlock(&kvm->slots_arch_lock);
1968 * For DELETE and MOVE, the working slot is now active as the INVALID
1969 * version of the old slot. MOVE is particularly special as it reuses
1970 * the old slot and returns a copy of the old slot (in working_slot).
1971 * For CREATE, there is no old slot. For DELETE and FLAGS_ONLY, the
1972 * old slot is detached but otherwise preserved.
1974 if (change == KVM_MR_CREATE)
1975 kvm_create_memslot(kvm, new);
1976 else if (change == KVM_MR_DELETE)
1977 kvm_delete_memslot(kvm, old, invalid_slot);
1978 else if (change == KVM_MR_MOVE)
1979 kvm_move_memslot(kvm, old, new, invalid_slot);
1980 else if (change == KVM_MR_FLAGS_ONLY)
1981 kvm_update_flags_memslot(kvm, old, new);
1985 /* Free the temporary INVALID slot used for DELETE and MOVE. */
1986 if (change == KVM_MR_DELETE || change == KVM_MR_MOVE)
1987 kfree(invalid_slot);
1990 * No need to refresh new->arch, changes after dropping slots_arch_lock
1991 * will directly hit the final, active memslot. Architectures are
1992 * responsible for knowing that new->arch may be stale.
1994 kvm_commit_memory_region(kvm, old, new, change);
1999 static bool kvm_check_memslot_overlap(struct kvm_memslots *slots, int id,
2000 gfn_t start, gfn_t end)
2002 struct kvm_memslot_iter iter;
2004 kvm_for_each_memslot_in_gfn_range(&iter, slots, start, end) {
2005 if (iter.slot->id != id)
2013 * Allocate some memory and give it an address in the guest physical address
2016 * Discontiguous memory is allowed, mostly for framebuffers.
2018 * Must be called holding kvm->slots_lock for write.
2020 int __kvm_set_memory_region(struct kvm *kvm,
2021 const struct kvm_userspace_memory_region2 *mem)
2023 struct kvm_memory_slot *old, *new;
2024 struct kvm_memslots *slots;
2025 enum kvm_mr_change change;
2026 unsigned long npages;
2031 r = check_memory_region_flags(kvm, mem);
2035 as_id = mem->slot >> 16;
2036 id = (u16)mem->slot;
2038 /* General sanity checks */
2039 if ((mem->memory_size & (PAGE_SIZE - 1)) ||
2040 (mem->memory_size != (unsigned long)mem->memory_size))
2042 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
2044 /* We can read the guest memory with __xxx_user() later on. */
2045 if ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
2046 (mem->userspace_addr != untagged_addr(mem->userspace_addr)) ||
2047 !access_ok((void __user *)(unsigned long)mem->userspace_addr,
2050 if (mem->flags & KVM_MEM_GUEST_MEMFD &&
2051 (mem->guest_memfd_offset & (PAGE_SIZE - 1) ||
2052 mem->guest_memfd_offset + mem->memory_size < mem->guest_memfd_offset))
2054 if (as_id >= kvm_arch_nr_memslot_as_ids(kvm) || id >= KVM_MEM_SLOTS_NUM)
2056 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
2058 if ((mem->memory_size >> PAGE_SHIFT) > KVM_MEM_MAX_NR_PAGES)
2061 slots = __kvm_memslots(kvm, as_id);
2064 * Note, the old memslot (and the pointer itself!) may be invalidated
2065 * and/or destroyed by kvm_set_memslot().
2067 old = id_to_memslot(slots, id);
2069 if (!mem->memory_size) {
2070 if (!old || !old->npages)
2073 if (WARN_ON_ONCE(kvm->nr_memslot_pages < old->npages))
2076 return kvm_set_memslot(kvm, old, NULL, KVM_MR_DELETE);
2079 base_gfn = (mem->guest_phys_addr >> PAGE_SHIFT);
2080 npages = (mem->memory_size >> PAGE_SHIFT);
2082 if (!old || !old->npages) {
2083 change = KVM_MR_CREATE;
2086 * To simplify KVM internals, the total number of pages across
2087 * all memslots must fit in an unsigned long.
2089 if ((kvm->nr_memslot_pages + npages) < kvm->nr_memslot_pages)
2091 } else { /* Modify an existing slot. */
2092 /* Private memslots are immutable, they can only be deleted. */
2093 if (mem->flags & KVM_MEM_GUEST_MEMFD)
2095 if ((mem->userspace_addr != old->userspace_addr) ||
2096 (npages != old->npages) ||
2097 ((mem->flags ^ old->flags) & KVM_MEM_READONLY))
2100 if (base_gfn != old->base_gfn)
2101 change = KVM_MR_MOVE;
2102 else if (mem->flags != old->flags)
2103 change = KVM_MR_FLAGS_ONLY;
2104 else /* Nothing to change. */
2108 if ((change == KVM_MR_CREATE || change == KVM_MR_MOVE) &&
2109 kvm_check_memslot_overlap(slots, id, base_gfn, base_gfn + npages))
2112 /* Allocate a slot that will persist in the memslot. */
2113 new = kzalloc(sizeof(*new), GFP_KERNEL_ACCOUNT);
2119 new->base_gfn = base_gfn;
2120 new->npages = npages;
2121 new->flags = mem->flags;
2122 new->userspace_addr = mem->userspace_addr;
2123 if (mem->flags & KVM_MEM_GUEST_MEMFD) {
2124 r = kvm_gmem_bind(kvm, new, mem->guest_memfd, mem->guest_memfd_offset);
2129 r = kvm_set_memslot(kvm, old, new, change);
2136 if (mem->flags & KVM_MEM_GUEST_MEMFD)
2137 kvm_gmem_unbind(new);
2142 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
2144 int kvm_set_memory_region(struct kvm *kvm,
2145 const struct kvm_userspace_memory_region2 *mem)
2149 mutex_lock(&kvm->slots_lock);
2150 r = __kvm_set_memory_region(kvm, mem);
2151 mutex_unlock(&kvm->slots_lock);
2154 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
2156 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
2157 struct kvm_userspace_memory_region2 *mem)
2159 if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
2162 return kvm_set_memory_region(kvm, mem);
2165 #ifndef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
2167 * kvm_get_dirty_log - get a snapshot of dirty pages
2168 * @kvm: pointer to kvm instance
2169 * @log: slot id and address to which we copy the log
2170 * @is_dirty: set to '1' if any dirty pages were found
2171 * @memslot: set to the associated memslot, always valid on success
2173 int kvm_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log,
2174 int *is_dirty, struct kvm_memory_slot **memslot)
2176 struct kvm_memslots *slots;
2179 unsigned long any = 0;
2181 /* Dirty ring tracking may be exclusive to dirty log tracking */
2182 if (!kvm_use_dirty_bitmap(kvm))
2188 as_id = log->slot >> 16;
2189 id = (u16)log->slot;
2190 if (as_id >= kvm_arch_nr_memslot_as_ids(kvm) || id >= KVM_USER_MEM_SLOTS)
2193 slots = __kvm_memslots(kvm, as_id);
2194 *memslot = id_to_memslot(slots, id);
2195 if (!(*memslot) || !(*memslot)->dirty_bitmap)
2198 kvm_arch_sync_dirty_log(kvm, *memslot);
2200 n = kvm_dirty_bitmap_bytes(*memslot);
2202 for (i = 0; !any && i < n/sizeof(long); ++i)
2203 any = (*memslot)->dirty_bitmap[i];
2205 if (copy_to_user(log->dirty_bitmap, (*memslot)->dirty_bitmap, n))
2212 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
2214 #else /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
2216 * kvm_get_dirty_log_protect - get a snapshot of dirty pages
2217 * and reenable dirty page tracking for the corresponding pages.
2218 * @kvm: pointer to kvm instance
2219 * @log: slot id and address to which we copy the log
2221 * We need to keep it in mind that VCPU threads can write to the bitmap
2222 * concurrently. So, to avoid losing track of dirty pages we keep the
2225 * 1. Take a snapshot of the bit and clear it if needed.
2226 * 2. Write protect the corresponding page.
2227 * 3. Copy the snapshot to the userspace.
2228 * 4. Upon return caller flushes TLB's if needed.
2230 * Between 2 and 4, the guest may write to the page using the remaining TLB
2231 * entry. This is not a problem because the page is reported dirty using
2232 * the snapshot taken before and step 4 ensures that writes done after
2233 * exiting to userspace will be logged for the next call.
2236 static int kvm_get_dirty_log_protect(struct kvm *kvm, struct kvm_dirty_log *log)
2238 struct kvm_memslots *slots;
2239 struct kvm_memory_slot *memslot;
2242 unsigned long *dirty_bitmap;
2243 unsigned long *dirty_bitmap_buffer;
2246 /* Dirty ring tracking may be exclusive to dirty log tracking */
2247 if (!kvm_use_dirty_bitmap(kvm))
2250 as_id = log->slot >> 16;
2251 id = (u16)log->slot;
2252 if (as_id >= kvm_arch_nr_memslot_as_ids(kvm) || id >= KVM_USER_MEM_SLOTS)
2255 slots = __kvm_memslots(kvm, as_id);
2256 memslot = id_to_memslot(slots, id);
2257 if (!memslot || !memslot->dirty_bitmap)
2260 dirty_bitmap = memslot->dirty_bitmap;
2262 kvm_arch_sync_dirty_log(kvm, memslot);
2264 n = kvm_dirty_bitmap_bytes(memslot);
2266 if (kvm->manual_dirty_log_protect) {
2268 * Unlike kvm_get_dirty_log, we always return false in *flush,
2269 * because no flush is needed until KVM_CLEAR_DIRTY_LOG. There
2270 * is some code duplication between this function and
2271 * kvm_get_dirty_log, but hopefully all architecture
2272 * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log
2273 * can be eliminated.
2275 dirty_bitmap_buffer = dirty_bitmap;
2277 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
2278 memset(dirty_bitmap_buffer, 0, n);
2281 for (i = 0; i < n / sizeof(long); i++) {
2285 if (!dirty_bitmap[i])
2289 mask = xchg(&dirty_bitmap[i], 0);
2290 dirty_bitmap_buffer[i] = mask;
2292 offset = i * BITS_PER_LONG;
2293 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
2296 KVM_MMU_UNLOCK(kvm);
2300 kvm_flush_remote_tlbs_memslot(kvm, memslot);
2302 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
2309 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
2310 * @kvm: kvm instance
2311 * @log: slot id and address to which we copy the log
2313 * Steps 1-4 below provide general overview of dirty page logging. See
2314 * kvm_get_dirty_log_protect() function description for additional details.
2316 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
2317 * always flush the TLB (step 4) even if previous step failed and the dirty
2318 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
2319 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
2320 * writes will be marked dirty for next log read.
2322 * 1. Take a snapshot of the bit and clear it if needed.
2323 * 2. Write protect the corresponding page.
2324 * 3. Copy the snapshot to the userspace.
2325 * 4. Flush TLB's if needed.
2327 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
2328 struct kvm_dirty_log *log)
2332 mutex_lock(&kvm->slots_lock);
2334 r = kvm_get_dirty_log_protect(kvm, log);
2336 mutex_unlock(&kvm->slots_lock);
2341 * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap
2342 * and reenable dirty page tracking for the corresponding pages.
2343 * @kvm: pointer to kvm instance
2344 * @log: slot id and address from which to fetch the bitmap of dirty pages
2346 static int kvm_clear_dirty_log_protect(struct kvm *kvm,
2347 struct kvm_clear_dirty_log *log)
2349 struct kvm_memslots *slots;
2350 struct kvm_memory_slot *memslot;
2354 unsigned long *dirty_bitmap;
2355 unsigned long *dirty_bitmap_buffer;
2358 /* Dirty ring tracking may be exclusive to dirty log tracking */
2359 if (!kvm_use_dirty_bitmap(kvm))
2362 as_id = log->slot >> 16;
2363 id = (u16)log->slot;
2364 if (as_id >= kvm_arch_nr_memslot_as_ids(kvm) || id >= KVM_USER_MEM_SLOTS)
2367 if (log->first_page & 63)
2370 slots = __kvm_memslots(kvm, as_id);
2371 memslot = id_to_memslot(slots, id);
2372 if (!memslot || !memslot->dirty_bitmap)
2375 dirty_bitmap = memslot->dirty_bitmap;
2377 n = ALIGN(log->num_pages, BITS_PER_LONG) / 8;
2379 if (log->first_page > memslot->npages ||
2380 log->num_pages > memslot->npages - log->first_page ||
2381 (log->num_pages < memslot->npages - log->first_page && (log->num_pages & 63)))
2384 kvm_arch_sync_dirty_log(kvm, memslot);
2387 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
2388 if (copy_from_user(dirty_bitmap_buffer, log->dirty_bitmap, n))
2392 for (offset = log->first_page, i = offset / BITS_PER_LONG,
2393 n = DIV_ROUND_UP(log->num_pages, BITS_PER_LONG); n--;
2394 i++, offset += BITS_PER_LONG) {
2395 unsigned long mask = *dirty_bitmap_buffer++;
2396 atomic_long_t *p = (atomic_long_t *) &dirty_bitmap[i];
2400 mask &= atomic_long_fetch_andnot(mask, p);
2403 * mask contains the bits that really have been cleared. This
2404 * never includes any bits beyond the length of the memslot (if
2405 * the length is not aligned to 64 pages), therefore it is not
2406 * a problem if userspace sets them in log->dirty_bitmap.
2410 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
2414 KVM_MMU_UNLOCK(kvm);
2417 kvm_flush_remote_tlbs_memslot(kvm, memslot);
2422 static int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm,
2423 struct kvm_clear_dirty_log *log)
2427 mutex_lock(&kvm->slots_lock);
2429 r = kvm_clear_dirty_log_protect(kvm, log);
2431 mutex_unlock(&kvm->slots_lock);
2434 #endif /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
2436 #ifdef CONFIG_KVM_GENERIC_MEMORY_ATTRIBUTES
2438 * Returns true if _all_ gfns in the range [@start, @end) have attributes
2441 bool kvm_range_has_memory_attributes(struct kvm *kvm, gfn_t start, gfn_t end,
2442 unsigned long attrs)
2444 XA_STATE(xas, &kvm->mem_attr_array, start);
2445 unsigned long index;
2452 has_attrs = !xas_find(&xas, end - 1);
2457 for (index = start; index < end; index++) {
2459 entry = xas_next(&xas);
2460 } while (xas_retry(&xas, entry));
2462 if (xas.xa_index != index || xa_to_value(entry) != attrs) {
2473 static u64 kvm_supported_mem_attributes(struct kvm *kvm)
2475 if (!kvm || kvm_arch_has_private_mem(kvm))
2476 return KVM_MEMORY_ATTRIBUTE_PRIVATE;
2481 static __always_inline void kvm_handle_gfn_range(struct kvm *kvm,
2482 struct kvm_mmu_notifier_range *range)
2484 struct kvm_gfn_range gfn_range;
2485 struct kvm_memory_slot *slot;
2486 struct kvm_memslots *slots;
2487 struct kvm_memslot_iter iter;
2488 bool found_memslot = false;
2492 gfn_range.arg = range->arg;
2493 gfn_range.may_block = range->may_block;
2495 for (i = 0; i < kvm_arch_nr_memslot_as_ids(kvm); i++) {
2496 slots = __kvm_memslots(kvm, i);
2498 kvm_for_each_memslot_in_gfn_range(&iter, slots, range->start, range->end) {
2500 gfn_range.slot = slot;
2502 gfn_range.start = max(range->start, slot->base_gfn);
2503 gfn_range.end = min(range->end, slot->base_gfn + slot->npages);
2504 if (gfn_range.start >= gfn_range.end)
2507 if (!found_memslot) {
2508 found_memslot = true;
2510 if (!IS_KVM_NULL_FN(range->on_lock))
2511 range->on_lock(kvm);
2514 ret |= range->handler(kvm, &gfn_range);
2518 if (range->flush_on_ret && ret)
2519 kvm_flush_remote_tlbs(kvm);
2522 KVM_MMU_UNLOCK(kvm);
2525 static bool kvm_pre_set_memory_attributes(struct kvm *kvm,
2526 struct kvm_gfn_range *range)
2529 * Unconditionally add the range to the invalidation set, regardless of
2530 * whether or not the arch callback actually needs to zap SPTEs. E.g.
2531 * if KVM supports RWX attributes in the future and the attributes are
2532 * going from R=>RW, zapping isn't strictly necessary. Unconditionally
2533 * adding the range allows KVM to require that MMU invalidations add at
2534 * least one range between begin() and end(), e.g. allows KVM to detect
2535 * bugs where the add() is missed. Relaxing the rule *might* be safe,
2536 * but it's not obvious that allowing new mappings while the attributes
2537 * are in flux is desirable or worth the complexity.
2539 kvm_mmu_invalidate_range_add(kvm, range->start, range->end);
2541 return kvm_arch_pre_set_memory_attributes(kvm, range);
2544 /* Set @attributes for the gfn range [@start, @end). */
2545 static int kvm_vm_set_mem_attributes(struct kvm *kvm, gfn_t start, gfn_t end,
2546 unsigned long attributes)
2548 struct kvm_mmu_notifier_range pre_set_range = {
2551 .handler = kvm_pre_set_memory_attributes,
2552 .on_lock = kvm_mmu_invalidate_begin,
2553 .flush_on_ret = true,
2556 struct kvm_mmu_notifier_range post_set_range = {
2559 .arg.attributes = attributes,
2560 .handler = kvm_arch_post_set_memory_attributes,
2561 .on_lock = kvm_mmu_invalidate_end,
2568 entry = attributes ? xa_mk_value(attributes) : NULL;
2570 mutex_lock(&kvm->slots_lock);
2572 /* Nothing to do if the entire range as the desired attributes. */
2573 if (kvm_range_has_memory_attributes(kvm, start, end, attributes))
2577 * Reserve memory ahead of time to avoid having to deal with failures
2578 * partway through setting the new attributes.
2580 for (i = start; i < end; i++) {
2581 r = xa_reserve(&kvm->mem_attr_array, i, GFP_KERNEL_ACCOUNT);
2586 kvm_handle_gfn_range(kvm, &pre_set_range);
2588 for (i = start; i < end; i++) {
2589 r = xa_err(xa_store(&kvm->mem_attr_array, i, entry,
2590 GFP_KERNEL_ACCOUNT));
2594 kvm_handle_gfn_range(kvm, &post_set_range);
2597 mutex_unlock(&kvm->slots_lock);
2601 static int kvm_vm_ioctl_set_mem_attributes(struct kvm *kvm,
2602 struct kvm_memory_attributes *attrs)
2606 /* flags is currently not used. */
2609 if (attrs->attributes & ~kvm_supported_mem_attributes(kvm))
2611 if (attrs->size == 0 || attrs->address + attrs->size < attrs->address)
2613 if (!PAGE_ALIGNED(attrs->address) || !PAGE_ALIGNED(attrs->size))
2616 start = attrs->address >> PAGE_SHIFT;
2617 end = (attrs->address + attrs->size) >> PAGE_SHIFT;
2620 * xarray tracks data using "unsigned long", and as a result so does
2621 * KVM. For simplicity, supports generic attributes only on 64-bit
2624 BUILD_BUG_ON(sizeof(attrs->attributes) != sizeof(unsigned long));
2626 return kvm_vm_set_mem_attributes(kvm, start, end, attrs->attributes);
2628 #endif /* CONFIG_KVM_GENERIC_MEMORY_ATTRIBUTES */
2630 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
2632 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
2634 EXPORT_SYMBOL_GPL(gfn_to_memslot);
2636 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
2638 struct kvm_memslots *slots = kvm_vcpu_memslots(vcpu);
2639 u64 gen = slots->generation;
2640 struct kvm_memory_slot *slot;
2643 * This also protects against using a memslot from a different address space,
2644 * since different address spaces have different generation numbers.
2646 if (unlikely(gen != vcpu->last_used_slot_gen)) {
2647 vcpu->last_used_slot = NULL;
2648 vcpu->last_used_slot_gen = gen;
2651 slot = try_get_memslot(vcpu->last_used_slot, gfn);
2656 * Fall back to searching all memslots. We purposely use
2657 * search_memslots() instead of __gfn_to_memslot() to avoid
2658 * thrashing the VM-wide last_used_slot in kvm_memslots.
2660 slot = search_memslots(slots, gfn, false);
2662 vcpu->last_used_slot = slot;
2669 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
2671 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
2673 return kvm_is_visible_memslot(memslot);
2675 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
2677 bool kvm_vcpu_is_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
2679 struct kvm_memory_slot *memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2681 return kvm_is_visible_memslot(memslot);
2683 EXPORT_SYMBOL_GPL(kvm_vcpu_is_visible_gfn);
2685 unsigned long kvm_host_page_size(struct kvm_vcpu *vcpu, gfn_t gfn)
2687 struct vm_area_struct *vma;
2688 unsigned long addr, size;
2692 addr = kvm_vcpu_gfn_to_hva_prot(vcpu, gfn, NULL);
2693 if (kvm_is_error_hva(addr))
2696 mmap_read_lock(current->mm);
2697 vma = find_vma(current->mm, addr);
2701 size = vma_kernel_pagesize(vma);
2704 mmap_read_unlock(current->mm);
2709 static bool memslot_is_readonly(const struct kvm_memory_slot *slot)
2711 return slot->flags & KVM_MEM_READONLY;
2714 static unsigned long __gfn_to_hva_many(const struct kvm_memory_slot *slot, gfn_t gfn,
2715 gfn_t *nr_pages, bool write)
2717 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
2718 return KVM_HVA_ERR_BAD;
2720 if (memslot_is_readonly(slot) && write)
2721 return KVM_HVA_ERR_RO_BAD;
2724 *nr_pages = slot->npages - (gfn - slot->base_gfn);
2726 return __gfn_to_hva_memslot(slot, gfn);
2729 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
2732 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
2735 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
2738 return gfn_to_hva_many(slot, gfn, NULL);
2740 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
2742 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
2744 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
2746 EXPORT_SYMBOL_GPL(gfn_to_hva);
2748 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
2750 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
2752 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
2755 * Return the hva of a @gfn and the R/W attribute if possible.
2757 * @slot: the kvm_memory_slot which contains @gfn
2758 * @gfn: the gfn to be translated
2759 * @writable: used to return the read/write attribute of the @slot if the hva
2760 * is valid and @writable is not NULL
2762 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
2763 gfn_t gfn, bool *writable)
2765 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
2767 if (!kvm_is_error_hva(hva) && writable)
2768 *writable = !memslot_is_readonly(slot);
2773 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
2775 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2777 return gfn_to_hva_memslot_prot(slot, gfn, writable);
2780 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
2782 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2784 return gfn_to_hva_memslot_prot(slot, gfn, writable);
2787 static inline int check_user_page_hwpoison(unsigned long addr)
2789 int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
2791 rc = get_user_pages(addr, 1, flags, NULL);
2792 return rc == -EHWPOISON;
2796 * The fast path to get the writable pfn which will be stored in @pfn,
2797 * true indicates success, otherwise false is returned. It's also the
2798 * only part that runs if we can in atomic context.
2800 static bool hva_to_pfn_fast(unsigned long addr, bool write_fault,
2801 bool *writable, kvm_pfn_t *pfn)
2803 struct page *page[1];
2806 * Fast pin a writable pfn only if it is a write fault request
2807 * or the caller allows to map a writable pfn for a read fault
2810 if (!(write_fault || writable))
2813 if (get_user_page_fast_only(addr, FOLL_WRITE, page)) {
2814 *pfn = page_to_pfn(page[0]);
2825 * The slow path to get the pfn of the specified host virtual address,
2826 * 1 indicates success, -errno is returned if error is detected.
2828 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
2829 bool interruptible, bool *writable, kvm_pfn_t *pfn)
2832 * When a VCPU accesses a page that is not mapped into the secondary
2833 * MMU, we lookup the page using GUP to map it, so the guest VCPU can
2834 * make progress. We always want to honor NUMA hinting faults in that
2835 * case, because GUP usage corresponds to memory accesses from the VCPU.
2836 * Otherwise, we'd not trigger NUMA hinting faults once a page is
2837 * mapped into the secondary MMU and gets accessed by a VCPU.
2839 * Note that get_user_page_fast_only() and FOLL_WRITE for now
2840 * implicitly honor NUMA hinting faults and don't need this flag.
2842 unsigned int flags = FOLL_HWPOISON | FOLL_HONOR_NUMA_FAULT;
2849 *writable = write_fault;
2852 flags |= FOLL_WRITE;
2854 flags |= FOLL_NOWAIT;
2856 flags |= FOLL_INTERRUPTIBLE;
2858 npages = get_user_pages_unlocked(addr, 1, &page, flags);
2862 /* map read fault as writable if possible */
2863 if (unlikely(!write_fault) && writable) {
2866 if (get_user_page_fast_only(addr, FOLL_WRITE, &wpage)) {
2872 *pfn = page_to_pfn(page);
2876 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
2878 if (unlikely(!(vma->vm_flags & VM_READ)))
2881 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
2887 static int kvm_try_get_pfn(kvm_pfn_t pfn)
2889 struct page *page = kvm_pfn_to_refcounted_page(pfn);
2894 return get_page_unless_zero(page);
2897 static int hva_to_pfn_remapped(struct vm_area_struct *vma,
2898 unsigned long addr, bool write_fault,
2899 bool *writable, kvm_pfn_t *p_pfn)
2907 r = follow_pte(vma->vm_mm, addr, &ptep, &ptl);
2910 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
2911 * not call the fault handler, so do it here.
2913 bool unlocked = false;
2914 r = fixup_user_fault(current->mm, addr,
2915 (write_fault ? FAULT_FLAG_WRITE : 0),
2922 r = follow_pte(vma->vm_mm, addr, &ptep, &ptl);
2927 pte = ptep_get(ptep);
2929 if (write_fault && !pte_write(pte)) {
2930 pfn = KVM_PFN_ERR_RO_FAULT;
2935 *writable = pte_write(pte);
2939 * Get a reference here because callers of *hva_to_pfn* and
2940 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
2941 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
2942 * set, but the kvm_try_get_pfn/kvm_release_pfn_clean pair will
2943 * simply do nothing for reserved pfns.
2945 * Whoever called remap_pfn_range is also going to call e.g.
2946 * unmap_mapping_range before the underlying pages are freed,
2947 * causing a call to our MMU notifier.
2949 * Certain IO or PFNMAP mappings can be backed with valid
2950 * struct pages, but be allocated without refcounting e.g.,
2951 * tail pages of non-compound higher order allocations, which
2952 * would then underflow the refcount when the caller does the
2953 * required put_page. Don't allow those pages here.
2955 if (!kvm_try_get_pfn(pfn))
2959 pte_unmap_unlock(ptep, ptl);
2966 * Pin guest page in memory and return its pfn.
2967 * @addr: host virtual address which maps memory to the guest
2968 * @atomic: whether this function can sleep
2969 * @interruptible: whether the process can be interrupted by non-fatal signals
2970 * @async: whether this function need to wait IO complete if the
2971 * host page is not in the memory
2972 * @write_fault: whether we should get a writable host page
2973 * @writable: whether it allows to map a writable host page for !@write_fault
2975 * The function will map a writable host page for these two cases:
2976 * 1): @write_fault = true
2977 * 2): @write_fault = false && @writable, @writable will tell the caller
2978 * whether the mapping is writable.
2980 kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool interruptible,
2981 bool *async, bool write_fault, bool *writable)
2983 struct vm_area_struct *vma;
2987 /* we can do it either atomically or asynchronously, not both */
2988 BUG_ON(atomic && async);
2990 if (hva_to_pfn_fast(addr, write_fault, writable, &pfn))
2994 return KVM_PFN_ERR_FAULT;
2996 npages = hva_to_pfn_slow(addr, async, write_fault, interruptible,
3000 if (npages == -EINTR)
3001 return KVM_PFN_ERR_SIGPENDING;
3003 mmap_read_lock(current->mm);
3004 if (npages == -EHWPOISON ||
3005 (!async && check_user_page_hwpoison(addr))) {
3006 pfn = KVM_PFN_ERR_HWPOISON;
3011 vma = vma_lookup(current->mm, addr);
3014 pfn = KVM_PFN_ERR_FAULT;
3015 else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
3016 r = hva_to_pfn_remapped(vma, addr, write_fault, writable, &pfn);
3020 pfn = KVM_PFN_ERR_FAULT;
3022 if (async && vma_is_valid(vma, write_fault))
3024 pfn = KVM_PFN_ERR_FAULT;
3027 mmap_read_unlock(current->mm);
3031 kvm_pfn_t __gfn_to_pfn_memslot(const struct kvm_memory_slot *slot, gfn_t gfn,
3032 bool atomic, bool interruptible, bool *async,
3033 bool write_fault, bool *writable, hva_t *hva)
3035 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
3040 if (addr == KVM_HVA_ERR_RO_BAD) {
3043 return KVM_PFN_ERR_RO_FAULT;
3046 if (kvm_is_error_hva(addr)) {
3049 return KVM_PFN_NOSLOT;
3052 /* Do not map writable pfn in the readonly memslot. */
3053 if (writable && memslot_is_readonly(slot)) {
3058 return hva_to_pfn(addr, atomic, interruptible, async, write_fault,
3061 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
3063 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
3066 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, false,
3067 NULL, write_fault, writable, NULL);
3069 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
3071 kvm_pfn_t gfn_to_pfn_memslot(const struct kvm_memory_slot *slot, gfn_t gfn)
3073 return __gfn_to_pfn_memslot(slot, gfn, false, false, NULL, true,
3076 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
3078 kvm_pfn_t gfn_to_pfn_memslot_atomic(const struct kvm_memory_slot *slot, gfn_t gfn)
3080 return __gfn_to_pfn_memslot(slot, gfn, true, false, NULL, true,
3083 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
3085 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
3087 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
3089 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
3091 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
3093 return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
3095 EXPORT_SYMBOL_GPL(gfn_to_pfn);
3097 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
3099 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
3101 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
3103 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
3104 struct page **pages, int nr_pages)
3109 addr = gfn_to_hva_many(slot, gfn, &entry);
3110 if (kvm_is_error_hva(addr))
3113 if (entry < nr_pages)
3116 return get_user_pages_fast_only(addr, nr_pages, FOLL_WRITE, pages);
3118 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
3121 * Do not use this helper unless you are absolutely certain the gfn _must_ be
3122 * backed by 'struct page'. A valid example is if the backing memslot is
3123 * controlled by KVM. Note, if the returned page is valid, it's refcount has
3124 * been elevated by gfn_to_pfn().
3126 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
3131 pfn = gfn_to_pfn(kvm, gfn);
3133 if (is_error_noslot_pfn(pfn))
3134 return KVM_ERR_PTR_BAD_PAGE;
3136 page = kvm_pfn_to_refcounted_page(pfn);
3138 return KVM_ERR_PTR_BAD_PAGE;
3142 EXPORT_SYMBOL_GPL(gfn_to_page);
3144 void kvm_release_pfn(kvm_pfn_t pfn, bool dirty)
3147 kvm_release_pfn_dirty(pfn);
3149 kvm_release_pfn_clean(pfn);
3152 int kvm_vcpu_map(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map)
3156 struct page *page = KVM_UNMAPPED_PAGE;
3161 pfn = gfn_to_pfn(vcpu->kvm, gfn);
3162 if (is_error_noslot_pfn(pfn))
3165 if (pfn_valid(pfn)) {
3166 page = pfn_to_page(pfn);
3168 #ifdef CONFIG_HAS_IOMEM
3170 hva = memremap(pfn_to_hpa(pfn), PAGE_SIZE, MEMREMAP_WB);
3184 EXPORT_SYMBOL_GPL(kvm_vcpu_map);
3186 void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map, bool dirty)
3194 if (map->page != KVM_UNMAPPED_PAGE)
3196 #ifdef CONFIG_HAS_IOMEM
3202 kvm_vcpu_mark_page_dirty(vcpu, map->gfn);
3204 kvm_release_pfn(map->pfn, dirty);
3209 EXPORT_SYMBOL_GPL(kvm_vcpu_unmap);
3211 static bool kvm_is_ad_tracked_page(struct page *page)
3214 * Per page-flags.h, pages tagged PG_reserved "should in general not be
3215 * touched (e.g. set dirty) except by its owner".
3217 return !PageReserved(page);
3220 static void kvm_set_page_dirty(struct page *page)
3222 if (kvm_is_ad_tracked_page(page))
3226 static void kvm_set_page_accessed(struct page *page)
3228 if (kvm_is_ad_tracked_page(page))
3229 mark_page_accessed(page);
3232 void kvm_release_page_clean(struct page *page)
3234 WARN_ON(is_error_page(page));
3236 kvm_set_page_accessed(page);
3239 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
3241 void kvm_release_pfn_clean(kvm_pfn_t pfn)
3245 if (is_error_noslot_pfn(pfn))
3248 page = kvm_pfn_to_refcounted_page(pfn);
3252 kvm_release_page_clean(page);
3254 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
3256 void kvm_release_page_dirty(struct page *page)
3258 WARN_ON(is_error_page(page));
3260 kvm_set_page_dirty(page);
3261 kvm_release_page_clean(page);
3263 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
3265 void kvm_release_pfn_dirty(kvm_pfn_t pfn)
3269 if (is_error_noslot_pfn(pfn))
3272 page = kvm_pfn_to_refcounted_page(pfn);
3276 kvm_release_page_dirty(page);
3278 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
3281 * Note, checking for an error/noslot pfn is the caller's responsibility when
3282 * directly marking a page dirty/accessed. Unlike the "release" helpers, the
3283 * "set" helpers are not to be used when the pfn might point at garbage.
3285 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
3287 if (WARN_ON(is_error_noslot_pfn(pfn)))
3291 kvm_set_page_dirty(pfn_to_page(pfn));
3293 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
3295 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
3297 if (WARN_ON(is_error_noslot_pfn(pfn)))
3301 kvm_set_page_accessed(pfn_to_page(pfn));
3303 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
3305 static int next_segment(unsigned long len, int offset)
3307 if (len > PAGE_SIZE - offset)
3308 return PAGE_SIZE - offset;
3313 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
3314 void *data, int offset, int len)
3319 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
3320 if (kvm_is_error_hva(addr))
3322 r = __copy_from_user(data, (void __user *)addr + offset, len);
3328 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
3331 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
3333 return __kvm_read_guest_page(slot, gfn, data, offset, len);
3335 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
3337 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
3338 int offset, int len)
3340 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
3342 return __kvm_read_guest_page(slot, gfn, data, offset, len);
3344 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
3346 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
3348 gfn_t gfn = gpa >> PAGE_SHIFT;
3350 int offset = offset_in_page(gpa);
3353 while ((seg = next_segment(len, offset)) != 0) {
3354 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
3364 EXPORT_SYMBOL_GPL(kvm_read_guest);
3366 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
3368 gfn_t gfn = gpa >> PAGE_SHIFT;
3370 int offset = offset_in_page(gpa);
3373 while ((seg = next_segment(len, offset)) != 0) {
3374 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
3384 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
3386 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
3387 void *data, int offset, unsigned long len)
3392 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
3393 if (kvm_is_error_hva(addr))
3395 pagefault_disable();
3396 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
3403 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
3404 void *data, unsigned long len)
3406 gfn_t gfn = gpa >> PAGE_SHIFT;
3407 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
3408 int offset = offset_in_page(gpa);
3410 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
3412 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
3414 static int __kvm_write_guest_page(struct kvm *kvm,
3415 struct kvm_memory_slot *memslot, gfn_t gfn,
3416 const void *data, int offset, int len)
3421 addr = gfn_to_hva_memslot(memslot, gfn);
3422 if (kvm_is_error_hva(addr))
3424 r = __copy_to_user((void __user *)addr + offset, data, len);
3427 mark_page_dirty_in_slot(kvm, memslot, gfn);
3431 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
3432 const void *data, int offset, int len)
3434 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
3436 return __kvm_write_guest_page(kvm, slot, gfn, data, offset, len);
3438 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
3440 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
3441 const void *data, int offset, int len)
3443 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
3445 return __kvm_write_guest_page(vcpu->kvm, slot, gfn, data, offset, len);
3447 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
3449 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
3452 gfn_t gfn = gpa >> PAGE_SHIFT;
3454 int offset = offset_in_page(gpa);
3457 while ((seg = next_segment(len, offset)) != 0) {
3458 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
3468 EXPORT_SYMBOL_GPL(kvm_write_guest);
3470 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
3473 gfn_t gfn = gpa >> PAGE_SHIFT;
3475 int offset = offset_in_page(gpa);
3478 while ((seg = next_segment(len, offset)) != 0) {
3479 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
3489 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
3491 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
3492 struct gfn_to_hva_cache *ghc,
3493 gpa_t gpa, unsigned long len)
3495 int offset = offset_in_page(gpa);
3496 gfn_t start_gfn = gpa >> PAGE_SHIFT;
3497 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
3498 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
3499 gfn_t nr_pages_avail;
3501 /* Update ghc->generation before performing any error checks. */
3502 ghc->generation = slots->generation;
3504 if (start_gfn > end_gfn) {
3505 ghc->hva = KVM_HVA_ERR_BAD;
3510 * If the requested region crosses two memslots, we still
3511 * verify that the entire region is valid here.
3513 for ( ; start_gfn <= end_gfn; start_gfn += nr_pages_avail) {
3514 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
3515 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
3517 if (kvm_is_error_hva(ghc->hva))
3521 /* Use the slow path for cross page reads and writes. */
3522 if (nr_pages_needed == 1)
3525 ghc->memslot = NULL;
3532 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3533 gpa_t gpa, unsigned long len)
3535 struct kvm_memslots *slots = kvm_memslots(kvm);
3536 return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len);
3538 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
3540 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3541 void *data, unsigned int offset,
3544 struct kvm_memslots *slots = kvm_memslots(kvm);
3546 gpa_t gpa = ghc->gpa + offset;
3548 if (WARN_ON_ONCE(len + offset > ghc->len))
3551 if (slots->generation != ghc->generation) {
3552 if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len))
3556 if (kvm_is_error_hva(ghc->hva))
3559 if (unlikely(!ghc->memslot))
3560 return kvm_write_guest(kvm, gpa, data, len);
3562 r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
3565 mark_page_dirty_in_slot(kvm, ghc->memslot, gpa >> PAGE_SHIFT);
3569 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached);
3571 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3572 void *data, unsigned long len)
3574 return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len);
3576 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
3578 int kvm_read_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3579 void *data, unsigned int offset,
3582 struct kvm_memslots *slots = kvm_memslots(kvm);
3584 gpa_t gpa = ghc->gpa + offset;
3586 if (WARN_ON_ONCE(len + offset > ghc->len))
3589 if (slots->generation != ghc->generation) {
3590 if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len))
3594 if (kvm_is_error_hva(ghc->hva))
3597 if (unlikely(!ghc->memslot))
3598 return kvm_read_guest(kvm, gpa, data, len);
3600 r = __copy_from_user(data, (void __user *)ghc->hva + offset, len);
3606 EXPORT_SYMBOL_GPL(kvm_read_guest_offset_cached);
3608 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3609 void *data, unsigned long len)
3611 return kvm_read_guest_offset_cached(kvm, ghc, data, 0, len);
3613 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
3615 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
3617 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
3618 gfn_t gfn = gpa >> PAGE_SHIFT;
3620 int offset = offset_in_page(gpa);
3623 while ((seg = next_segment(len, offset)) != 0) {
3624 ret = kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
3633 EXPORT_SYMBOL_GPL(kvm_clear_guest);
3635 void mark_page_dirty_in_slot(struct kvm *kvm,
3636 const struct kvm_memory_slot *memslot,
3639 struct kvm_vcpu *vcpu = kvm_get_running_vcpu();
3641 #ifdef CONFIG_HAVE_KVM_DIRTY_RING
3642 if (WARN_ON_ONCE(vcpu && vcpu->kvm != kvm))
3645 WARN_ON_ONCE(!vcpu && !kvm_arch_allow_write_without_running_vcpu(kvm));
3648 if (memslot && kvm_slot_dirty_track_enabled(memslot)) {
3649 unsigned long rel_gfn = gfn - memslot->base_gfn;
3650 u32 slot = (memslot->as_id << 16) | memslot->id;
3652 if (kvm->dirty_ring_size && vcpu)
3653 kvm_dirty_ring_push(vcpu, slot, rel_gfn);
3654 else if (memslot->dirty_bitmap)
3655 set_bit_le(rel_gfn, memslot->dirty_bitmap);
3658 EXPORT_SYMBOL_GPL(mark_page_dirty_in_slot);
3660 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
3662 struct kvm_memory_slot *memslot;
3664 memslot = gfn_to_memslot(kvm, gfn);
3665 mark_page_dirty_in_slot(kvm, memslot, gfn);
3667 EXPORT_SYMBOL_GPL(mark_page_dirty);
3669 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
3671 struct kvm_memory_slot *memslot;
3673 memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
3674 mark_page_dirty_in_slot(vcpu->kvm, memslot, gfn);
3676 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
3678 void kvm_sigset_activate(struct kvm_vcpu *vcpu)
3680 if (!vcpu->sigset_active)
3684 * This does a lockless modification of ->real_blocked, which is fine
3685 * because, only current can change ->real_blocked and all readers of
3686 * ->real_blocked don't care as long ->real_blocked is always a subset
3689 sigprocmask(SIG_SETMASK, &vcpu->sigset, ¤t->real_blocked);
3692 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu)
3694 if (!vcpu->sigset_active)
3697 sigprocmask(SIG_SETMASK, ¤t->real_blocked, NULL);
3698 sigemptyset(¤t->real_blocked);
3701 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
3703 unsigned int old, val, grow, grow_start;
3705 old = val = vcpu->halt_poll_ns;
3706 grow_start = READ_ONCE(halt_poll_ns_grow_start);
3707 grow = READ_ONCE(halt_poll_ns_grow);
3712 if (val < grow_start)
3715 vcpu->halt_poll_ns = val;
3717 trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
3720 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
3722 unsigned int old, val, shrink, grow_start;
3724 old = val = vcpu->halt_poll_ns;
3725 shrink = READ_ONCE(halt_poll_ns_shrink);
3726 grow_start = READ_ONCE(halt_poll_ns_grow_start);
3732 if (val < grow_start)
3735 vcpu->halt_poll_ns = val;
3736 trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
3739 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
3742 int idx = srcu_read_lock(&vcpu->kvm->srcu);
3744 if (kvm_arch_vcpu_runnable(vcpu))
3746 if (kvm_cpu_has_pending_timer(vcpu))
3748 if (signal_pending(current))
3750 if (kvm_check_request(KVM_REQ_UNBLOCK, vcpu))
3755 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3760 * Block the vCPU until the vCPU is runnable, an event arrives, or a signal is
3761 * pending. This is mostly used when halting a vCPU, but may also be used
3762 * directly for other vCPU non-runnable states, e.g. x86's Wait-For-SIPI.
3764 bool kvm_vcpu_block(struct kvm_vcpu *vcpu)
3766 struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu);
3767 bool waited = false;
3769 vcpu->stat.generic.blocking = 1;
3772 kvm_arch_vcpu_blocking(vcpu);
3773 prepare_to_rcuwait(wait);
3777 set_current_state(TASK_INTERRUPTIBLE);
3779 if (kvm_vcpu_check_block(vcpu) < 0)
3787 finish_rcuwait(wait);
3788 kvm_arch_vcpu_unblocking(vcpu);
3791 vcpu->stat.generic.blocking = 0;
3796 static inline void update_halt_poll_stats(struct kvm_vcpu *vcpu, ktime_t start,
3797 ktime_t end, bool success)
3799 struct kvm_vcpu_stat_generic *stats = &vcpu->stat.generic;
3800 u64 poll_ns = ktime_to_ns(ktime_sub(end, start));
3802 ++vcpu->stat.generic.halt_attempted_poll;
3805 ++vcpu->stat.generic.halt_successful_poll;
3807 if (!vcpu_valid_wakeup(vcpu))
3808 ++vcpu->stat.generic.halt_poll_invalid;
3810 stats->halt_poll_success_ns += poll_ns;
3811 KVM_STATS_LOG_HIST_UPDATE(stats->halt_poll_success_hist, poll_ns);
3813 stats->halt_poll_fail_ns += poll_ns;
3814 KVM_STATS_LOG_HIST_UPDATE(stats->halt_poll_fail_hist, poll_ns);
3818 static unsigned int kvm_vcpu_max_halt_poll_ns(struct kvm_vcpu *vcpu)
3820 struct kvm *kvm = vcpu->kvm;
3822 if (kvm->override_halt_poll_ns) {
3824 * Ensure kvm->max_halt_poll_ns is not read before
3825 * kvm->override_halt_poll_ns.
3827 * Pairs with the smp_wmb() when enabling KVM_CAP_HALT_POLL.
3830 return READ_ONCE(kvm->max_halt_poll_ns);
3833 return READ_ONCE(halt_poll_ns);
3837 * Emulate a vCPU halt condition, e.g. HLT on x86, WFI on arm, etc... If halt
3838 * polling is enabled, busy wait for a short time before blocking to avoid the
3839 * expensive block+unblock sequence if a wake event arrives soon after the vCPU
3842 void kvm_vcpu_halt(struct kvm_vcpu *vcpu)
3844 unsigned int max_halt_poll_ns = kvm_vcpu_max_halt_poll_ns(vcpu);
3845 bool halt_poll_allowed = !kvm_arch_no_poll(vcpu);
3846 ktime_t start, cur, poll_end;
3847 bool waited = false;
3851 if (vcpu->halt_poll_ns > max_halt_poll_ns)
3852 vcpu->halt_poll_ns = max_halt_poll_ns;
3854 do_halt_poll = halt_poll_allowed && vcpu->halt_poll_ns;
3856 start = cur = poll_end = ktime_get();
3858 ktime_t stop = ktime_add_ns(start, vcpu->halt_poll_ns);
3861 if (kvm_vcpu_check_block(vcpu) < 0)
3864 poll_end = cur = ktime_get();
3865 } while (kvm_vcpu_can_poll(cur, stop));
3868 waited = kvm_vcpu_block(vcpu);
3872 vcpu->stat.generic.halt_wait_ns +=
3873 ktime_to_ns(cur) - ktime_to_ns(poll_end);
3874 KVM_STATS_LOG_HIST_UPDATE(vcpu->stat.generic.halt_wait_hist,
3875 ktime_to_ns(cur) - ktime_to_ns(poll_end));
3878 /* The total time the vCPU was "halted", including polling time. */
3879 halt_ns = ktime_to_ns(cur) - ktime_to_ns(start);
3882 * Note, halt-polling is considered successful so long as the vCPU was
3883 * never actually scheduled out, i.e. even if the wake event arrived
3884 * after of the halt-polling loop itself, but before the full wait.
3887 update_halt_poll_stats(vcpu, start, poll_end, !waited);
3889 if (halt_poll_allowed) {
3890 /* Recompute the max halt poll time in case it changed. */
3891 max_halt_poll_ns = kvm_vcpu_max_halt_poll_ns(vcpu);
3893 if (!vcpu_valid_wakeup(vcpu)) {
3894 shrink_halt_poll_ns(vcpu);
3895 } else if (max_halt_poll_ns) {
3896 if (halt_ns <= vcpu->halt_poll_ns)
3898 /* we had a long block, shrink polling */
3899 else if (vcpu->halt_poll_ns &&
3900 halt_ns > max_halt_poll_ns)
3901 shrink_halt_poll_ns(vcpu);
3902 /* we had a short halt and our poll time is too small */
3903 else if (vcpu->halt_poll_ns < max_halt_poll_ns &&
3904 halt_ns < max_halt_poll_ns)
3905 grow_halt_poll_ns(vcpu);
3907 vcpu->halt_poll_ns = 0;
3911 trace_kvm_vcpu_wakeup(halt_ns, waited, vcpu_valid_wakeup(vcpu));
3913 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
3915 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
3917 if (__kvm_vcpu_wake_up(vcpu)) {
3918 WRITE_ONCE(vcpu->ready, true);
3919 ++vcpu->stat.generic.halt_wakeup;
3925 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
3929 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
3931 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
3935 if (kvm_vcpu_wake_up(vcpu))
3940 * The only state change done outside the vcpu mutex is IN_GUEST_MODE
3941 * to EXITING_GUEST_MODE. Therefore the moderately expensive "should
3942 * kick" check does not need atomic operations if kvm_vcpu_kick is used
3943 * within the vCPU thread itself.
3945 if (vcpu == __this_cpu_read(kvm_running_vcpu)) {
3946 if (vcpu->mode == IN_GUEST_MODE)
3947 WRITE_ONCE(vcpu->mode, EXITING_GUEST_MODE);
3952 * Note, the vCPU could get migrated to a different pCPU at any point
3953 * after kvm_arch_vcpu_should_kick(), which could result in sending an
3954 * IPI to the previous pCPU. But, that's ok because the purpose of the
3955 * IPI is to force the vCPU to leave IN_GUEST_MODE, and migrating the
3956 * vCPU also requires it to leave IN_GUEST_MODE.
3958 if (kvm_arch_vcpu_should_kick(vcpu)) {
3959 cpu = READ_ONCE(vcpu->cpu);
3960 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
3961 smp_send_reschedule(cpu);
3966 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
3967 #endif /* !CONFIG_S390 */
3969 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
3972 struct task_struct *task = NULL;
3976 pid = rcu_dereference(target->pid);
3978 task = get_pid_task(pid, PIDTYPE_PID);
3982 ret = yield_to(task, 1);
3983 put_task_struct(task);
3987 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
3990 * Helper that checks whether a VCPU is eligible for directed yield.
3991 * Most eligible candidate to yield is decided by following heuristics:
3993 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
3994 * (preempted lock holder), indicated by @in_spin_loop.
3995 * Set at the beginning and cleared at the end of interception/PLE handler.
3997 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
3998 * chance last time (mostly it has become eligible now since we have probably
3999 * yielded to lockholder in last iteration. This is done by toggling
4000 * @dy_eligible each time a VCPU checked for eligibility.)
4002 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
4003 * to preempted lock-holder could result in wrong VCPU selection and CPU
4004 * burning. Giving priority for a potential lock-holder increases lock
4007 * Since algorithm is based on heuristics, accessing another VCPU data without
4008 * locking does not harm. It may result in trying to yield to same VCPU, fail
4009 * and continue with next VCPU and so on.
4011 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
4013 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
4016 eligible = !vcpu->spin_loop.in_spin_loop ||
4017 vcpu->spin_loop.dy_eligible;
4019 if (vcpu->spin_loop.in_spin_loop)
4020 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
4029 * Unlike kvm_arch_vcpu_runnable, this function is called outside
4030 * a vcpu_load/vcpu_put pair. However, for most architectures
4031 * kvm_arch_vcpu_runnable does not require vcpu_load.
4033 bool __weak kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
4035 return kvm_arch_vcpu_runnable(vcpu);
4038 static bool vcpu_dy_runnable(struct kvm_vcpu *vcpu)
4040 if (kvm_arch_dy_runnable(vcpu))
4043 #ifdef CONFIG_KVM_ASYNC_PF
4044 if (!list_empty_careful(&vcpu->async_pf.done))
4051 bool __weak kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu)
4056 void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode)
4058 struct kvm *kvm = me->kvm;
4059 struct kvm_vcpu *vcpu;
4060 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
4066 kvm_vcpu_set_in_spin_loop(me, true);
4068 * We boost the priority of a VCPU that is runnable but not
4069 * currently running, because it got preempted by something
4070 * else and called schedule in __vcpu_run. Hopefully that
4071 * VCPU is holding the lock that we need and will release it.
4072 * We approximate round-robin by starting at the last boosted VCPU.
4074 for (pass = 0; pass < 2 && !yielded && try; pass++) {
4075 kvm_for_each_vcpu(i, vcpu, kvm) {
4076 if (!pass && i <= last_boosted_vcpu) {
4077 i = last_boosted_vcpu;
4079 } else if (pass && i > last_boosted_vcpu)
4081 if (!READ_ONCE(vcpu->ready))
4085 if (kvm_vcpu_is_blocking(vcpu) && !vcpu_dy_runnable(vcpu))
4087 if (READ_ONCE(vcpu->preempted) && yield_to_kernel_mode &&
4088 !kvm_arch_dy_has_pending_interrupt(vcpu) &&
4089 !kvm_arch_vcpu_in_kernel(vcpu))
4091 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
4094 yielded = kvm_vcpu_yield_to(vcpu);
4096 kvm->last_boosted_vcpu = i;
4098 } else if (yielded < 0) {
4105 kvm_vcpu_set_in_spin_loop(me, false);
4107 /* Ensure vcpu is not eligible during next spinloop */
4108 kvm_vcpu_set_dy_eligible(me, false);
4110 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
4112 static bool kvm_page_in_dirty_ring(struct kvm *kvm, unsigned long pgoff)
4114 #ifdef CONFIG_HAVE_KVM_DIRTY_RING
4115 return (pgoff >= KVM_DIRTY_LOG_PAGE_OFFSET) &&
4116 (pgoff < KVM_DIRTY_LOG_PAGE_OFFSET +
4117 kvm->dirty_ring_size / PAGE_SIZE);
4123 static vm_fault_t kvm_vcpu_fault(struct vm_fault *vmf)
4125 struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data;
4128 if (vmf->pgoff == 0)
4129 page = virt_to_page(vcpu->run);
4131 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
4132 page = virt_to_page(vcpu->arch.pio_data);
4134 #ifdef CONFIG_KVM_MMIO
4135 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
4136 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
4138 else if (kvm_page_in_dirty_ring(vcpu->kvm, vmf->pgoff))
4139 page = kvm_dirty_ring_get_page(
4141 vmf->pgoff - KVM_DIRTY_LOG_PAGE_OFFSET);
4143 return kvm_arch_vcpu_fault(vcpu, vmf);
4149 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
4150 .fault = kvm_vcpu_fault,
4153 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
4155 struct kvm_vcpu *vcpu = file->private_data;
4156 unsigned long pages = vma_pages(vma);
4158 if ((kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff) ||
4159 kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff + pages - 1)) &&
4160 ((vma->vm_flags & VM_EXEC) || !(vma->vm_flags & VM_SHARED)))
4163 vma->vm_ops = &kvm_vcpu_vm_ops;
4167 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
4169 struct kvm_vcpu *vcpu = filp->private_data;
4171 kvm_put_kvm(vcpu->kvm);
4175 static struct file_operations kvm_vcpu_fops = {
4176 .release = kvm_vcpu_release,
4177 .unlocked_ioctl = kvm_vcpu_ioctl,
4178 .mmap = kvm_vcpu_mmap,
4179 .llseek = noop_llseek,
4180 KVM_COMPAT(kvm_vcpu_compat_ioctl),
4184 * Allocates an inode for the vcpu.
4186 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
4188 char name[8 + 1 + ITOA_MAX_LEN + 1];
4190 snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id);
4191 return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
4194 #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS
4195 static int vcpu_get_pid(void *data, u64 *val)
4197 struct kvm_vcpu *vcpu = data;
4200 *val = pid_nr(rcu_dereference(vcpu->pid));
4205 DEFINE_SIMPLE_ATTRIBUTE(vcpu_get_pid_fops, vcpu_get_pid, NULL, "%llu\n");
4207 static void kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
4209 struct dentry *debugfs_dentry;
4210 char dir_name[ITOA_MAX_LEN * 2];
4212 if (!debugfs_initialized())
4215 snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
4216 debugfs_dentry = debugfs_create_dir(dir_name,
4217 vcpu->kvm->debugfs_dentry);
4218 debugfs_create_file("pid", 0444, debugfs_dentry, vcpu,
4219 &vcpu_get_pid_fops);
4221 kvm_arch_create_vcpu_debugfs(vcpu, debugfs_dentry);
4226 * Creates some virtual cpus. Good luck creating more than one.
4228 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
4231 struct kvm_vcpu *vcpu;
4234 if (id >= KVM_MAX_VCPU_IDS)
4237 mutex_lock(&kvm->lock);
4238 if (kvm->created_vcpus >= kvm->max_vcpus) {
4239 mutex_unlock(&kvm->lock);
4243 r = kvm_arch_vcpu_precreate(kvm, id);
4245 mutex_unlock(&kvm->lock);
4249 kvm->created_vcpus++;
4250 mutex_unlock(&kvm->lock);
4252 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL_ACCOUNT);
4255 goto vcpu_decrement;
4258 BUILD_BUG_ON(sizeof(struct kvm_run) > PAGE_SIZE);
4259 page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
4264 vcpu->run = page_address(page);
4266 kvm_vcpu_init(vcpu, kvm, id);
4268 r = kvm_arch_vcpu_create(vcpu);
4270 goto vcpu_free_run_page;
4272 if (kvm->dirty_ring_size) {
4273 r = kvm_dirty_ring_alloc(&vcpu->dirty_ring,
4274 id, kvm->dirty_ring_size);
4276 goto arch_vcpu_destroy;
4279 mutex_lock(&kvm->lock);
4281 #ifdef CONFIG_LOCKDEP
4282 /* Ensure that lockdep knows vcpu->mutex is taken *inside* kvm->lock */
4283 mutex_lock(&vcpu->mutex);
4284 mutex_unlock(&vcpu->mutex);
4287 if (kvm_get_vcpu_by_id(kvm, id)) {
4289 goto unlock_vcpu_destroy;
4292 vcpu->vcpu_idx = atomic_read(&kvm->online_vcpus);
4293 r = xa_reserve(&kvm->vcpu_array, vcpu->vcpu_idx, GFP_KERNEL_ACCOUNT);
4295 goto unlock_vcpu_destroy;
4297 /* Now it's all set up, let userspace reach it */
4299 r = create_vcpu_fd(vcpu);
4301 goto kvm_put_xa_release;
4303 if (KVM_BUG_ON(xa_store(&kvm->vcpu_array, vcpu->vcpu_idx, vcpu, 0), kvm)) {
4305 goto kvm_put_xa_release;
4309 * Pairs with smp_rmb() in kvm_get_vcpu. Store the vcpu
4310 * pointer before kvm->online_vcpu's incremented value.
4313 atomic_inc(&kvm->online_vcpus);
4315 mutex_unlock(&kvm->lock);
4316 kvm_arch_vcpu_postcreate(vcpu);
4317 kvm_create_vcpu_debugfs(vcpu);
4321 kvm_put_kvm_no_destroy(kvm);
4322 xa_release(&kvm->vcpu_array, vcpu->vcpu_idx);
4323 unlock_vcpu_destroy:
4324 mutex_unlock(&kvm->lock);
4325 kvm_dirty_ring_free(&vcpu->dirty_ring);
4327 kvm_arch_vcpu_destroy(vcpu);
4329 free_page((unsigned long)vcpu->run);
4331 kmem_cache_free(kvm_vcpu_cache, vcpu);
4333 mutex_lock(&kvm->lock);
4334 kvm->created_vcpus--;
4335 mutex_unlock(&kvm->lock);
4339 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
4342 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
4343 vcpu->sigset_active = 1;
4344 vcpu->sigset = *sigset;
4346 vcpu->sigset_active = 0;
4350 static ssize_t kvm_vcpu_stats_read(struct file *file, char __user *user_buffer,
4351 size_t size, loff_t *offset)
4353 struct kvm_vcpu *vcpu = file->private_data;
4355 return kvm_stats_read(vcpu->stats_id, &kvm_vcpu_stats_header,
4356 &kvm_vcpu_stats_desc[0], &vcpu->stat,
4357 sizeof(vcpu->stat), user_buffer, size, offset);
4360 static int kvm_vcpu_stats_release(struct inode *inode, struct file *file)
4362 struct kvm_vcpu *vcpu = file->private_data;
4364 kvm_put_kvm(vcpu->kvm);
4368 static const struct file_operations kvm_vcpu_stats_fops = {
4369 .owner = THIS_MODULE,
4370 .read = kvm_vcpu_stats_read,
4371 .release = kvm_vcpu_stats_release,
4372 .llseek = noop_llseek,
4375 static int kvm_vcpu_ioctl_get_stats_fd(struct kvm_vcpu *vcpu)
4379 char name[15 + ITOA_MAX_LEN + 1];
4381 snprintf(name, sizeof(name), "kvm-vcpu-stats:%d", vcpu->vcpu_id);
4383 fd = get_unused_fd_flags(O_CLOEXEC);
4387 file = anon_inode_getfile(name, &kvm_vcpu_stats_fops, vcpu, O_RDONLY);
4390 return PTR_ERR(file);
4393 kvm_get_kvm(vcpu->kvm);
4395 file->f_mode |= FMODE_PREAD;
4396 fd_install(fd, file);
4401 static long kvm_vcpu_ioctl(struct file *filp,
4402 unsigned int ioctl, unsigned long arg)
4404 struct kvm_vcpu *vcpu = filp->private_data;
4405 void __user *argp = (void __user *)arg;
4407 struct kvm_fpu *fpu = NULL;
4408 struct kvm_sregs *kvm_sregs = NULL;
4410 if (vcpu->kvm->mm != current->mm || vcpu->kvm->vm_dead)
4413 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
4417 * Some architectures have vcpu ioctls that are asynchronous to vcpu
4418 * execution; mutex_lock() would break them.
4420 r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg);
4421 if (r != -ENOIOCTLCMD)
4424 if (mutex_lock_killable(&vcpu->mutex))
4432 oldpid = rcu_access_pointer(vcpu->pid);
4433 if (unlikely(oldpid != task_pid(current))) {
4434 /* The thread running this VCPU changed. */
4437 r = kvm_arch_vcpu_run_pid_change(vcpu);
4441 newpid = get_task_pid(current, PIDTYPE_PID);
4442 rcu_assign_pointer(vcpu->pid, newpid);
4447 r = kvm_arch_vcpu_ioctl_run(vcpu);
4448 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
4451 case KVM_GET_REGS: {
4452 struct kvm_regs *kvm_regs;
4455 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL_ACCOUNT);
4458 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
4462 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
4469 case KVM_SET_REGS: {
4470 struct kvm_regs *kvm_regs;
4472 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
4473 if (IS_ERR(kvm_regs)) {
4474 r = PTR_ERR(kvm_regs);
4477 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
4481 case KVM_GET_SREGS: {
4482 kvm_sregs = kzalloc(sizeof(struct kvm_sregs),
4483 GFP_KERNEL_ACCOUNT);
4487 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
4491 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
4496 case KVM_SET_SREGS: {
4497 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
4498 if (IS_ERR(kvm_sregs)) {
4499 r = PTR_ERR(kvm_sregs);
4503 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
4506 case KVM_GET_MP_STATE: {
4507 struct kvm_mp_state mp_state;
4509 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
4513 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
4518 case KVM_SET_MP_STATE: {
4519 struct kvm_mp_state mp_state;
4522 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
4524 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
4527 case KVM_TRANSLATE: {
4528 struct kvm_translation tr;
4531 if (copy_from_user(&tr, argp, sizeof(tr)))
4533 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
4537 if (copy_to_user(argp, &tr, sizeof(tr)))
4542 case KVM_SET_GUEST_DEBUG: {
4543 struct kvm_guest_debug dbg;
4546 if (copy_from_user(&dbg, argp, sizeof(dbg)))
4548 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
4551 case KVM_SET_SIGNAL_MASK: {
4552 struct kvm_signal_mask __user *sigmask_arg = argp;
4553 struct kvm_signal_mask kvm_sigmask;
4554 sigset_t sigset, *p;
4559 if (copy_from_user(&kvm_sigmask, argp,
4560 sizeof(kvm_sigmask)))
4563 if (kvm_sigmask.len != sizeof(sigset))
4566 if (copy_from_user(&sigset, sigmask_arg->sigset,
4571 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
4575 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL_ACCOUNT);
4579 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
4583 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
4589 fpu = memdup_user(argp, sizeof(*fpu));
4595 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
4598 case KVM_GET_STATS_FD: {
4599 r = kvm_vcpu_ioctl_get_stats_fd(vcpu);
4603 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
4606 mutex_unlock(&vcpu->mutex);
4612 #ifdef CONFIG_KVM_COMPAT
4613 static long kvm_vcpu_compat_ioctl(struct file *filp,
4614 unsigned int ioctl, unsigned long arg)
4616 struct kvm_vcpu *vcpu = filp->private_data;
4617 void __user *argp = compat_ptr(arg);
4620 if (vcpu->kvm->mm != current->mm || vcpu->kvm->vm_dead)
4624 case KVM_SET_SIGNAL_MASK: {
4625 struct kvm_signal_mask __user *sigmask_arg = argp;
4626 struct kvm_signal_mask kvm_sigmask;
4631 if (copy_from_user(&kvm_sigmask, argp,
4632 sizeof(kvm_sigmask)))
4635 if (kvm_sigmask.len != sizeof(compat_sigset_t))
4638 if (get_compat_sigset(&sigset,
4639 (compat_sigset_t __user *)sigmask_arg->sigset))
4641 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
4643 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
4647 r = kvm_vcpu_ioctl(filp, ioctl, arg);
4655 static int kvm_device_mmap(struct file *filp, struct vm_area_struct *vma)
4657 struct kvm_device *dev = filp->private_data;
4660 return dev->ops->mmap(dev, vma);
4665 static int kvm_device_ioctl_attr(struct kvm_device *dev,
4666 int (*accessor)(struct kvm_device *dev,
4667 struct kvm_device_attr *attr),
4670 struct kvm_device_attr attr;
4675 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
4678 return accessor(dev, &attr);
4681 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
4684 struct kvm_device *dev = filp->private_data;
4686 if (dev->kvm->mm != current->mm || dev->kvm->vm_dead)
4690 case KVM_SET_DEVICE_ATTR:
4691 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
4692 case KVM_GET_DEVICE_ATTR:
4693 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
4694 case KVM_HAS_DEVICE_ATTR:
4695 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
4697 if (dev->ops->ioctl)
4698 return dev->ops->ioctl(dev, ioctl, arg);
4704 static int kvm_device_release(struct inode *inode, struct file *filp)
4706 struct kvm_device *dev = filp->private_data;
4707 struct kvm *kvm = dev->kvm;
4709 if (dev->ops->release) {
4710 mutex_lock(&kvm->lock);
4711 list_del(&dev->vm_node);
4712 dev->ops->release(dev);
4713 mutex_unlock(&kvm->lock);
4720 static struct file_operations kvm_device_fops = {
4721 .unlocked_ioctl = kvm_device_ioctl,
4722 .release = kvm_device_release,
4723 KVM_COMPAT(kvm_device_ioctl),
4724 .mmap = kvm_device_mmap,
4727 struct kvm_device *kvm_device_from_filp(struct file *filp)
4729 if (filp->f_op != &kvm_device_fops)
4732 return filp->private_data;
4735 static const struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
4736 #ifdef CONFIG_KVM_MPIC
4737 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
4738 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
4742 int kvm_register_device_ops(const struct kvm_device_ops *ops, u32 type)
4744 if (type >= ARRAY_SIZE(kvm_device_ops_table))
4747 if (kvm_device_ops_table[type] != NULL)
4750 kvm_device_ops_table[type] = ops;
4754 void kvm_unregister_device_ops(u32 type)
4756 if (kvm_device_ops_table[type] != NULL)
4757 kvm_device_ops_table[type] = NULL;
4760 static int kvm_ioctl_create_device(struct kvm *kvm,
4761 struct kvm_create_device *cd)
4763 const struct kvm_device_ops *ops;
4764 struct kvm_device *dev;
4765 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
4769 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
4772 type = array_index_nospec(cd->type, ARRAY_SIZE(kvm_device_ops_table));
4773 ops = kvm_device_ops_table[type];
4780 dev = kzalloc(sizeof(*dev), GFP_KERNEL_ACCOUNT);
4787 mutex_lock(&kvm->lock);
4788 ret = ops->create(dev, type);
4790 mutex_unlock(&kvm->lock);
4794 list_add(&dev->vm_node, &kvm->devices);
4795 mutex_unlock(&kvm->lock);
4801 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
4803 kvm_put_kvm_no_destroy(kvm);
4804 mutex_lock(&kvm->lock);
4805 list_del(&dev->vm_node);
4808 mutex_unlock(&kvm->lock);
4818 static int kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
4821 case KVM_CAP_USER_MEMORY:
4822 case KVM_CAP_USER_MEMORY2:
4823 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
4824 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
4825 case KVM_CAP_INTERNAL_ERROR_DATA:
4826 #ifdef CONFIG_HAVE_KVM_MSI
4827 case KVM_CAP_SIGNAL_MSI:
4829 #ifdef CONFIG_HAVE_KVM_IRQCHIP
4832 case KVM_CAP_IOEVENTFD_ANY_LENGTH:
4833 case KVM_CAP_CHECK_EXTENSION_VM:
4834 case KVM_CAP_ENABLE_CAP_VM:
4835 case KVM_CAP_HALT_POLL:
4837 #ifdef CONFIG_KVM_MMIO
4838 case KVM_CAP_COALESCED_MMIO:
4839 return KVM_COALESCED_MMIO_PAGE_OFFSET;
4840 case KVM_CAP_COALESCED_PIO:
4843 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4844 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2:
4845 return KVM_DIRTY_LOG_MANUAL_CAPS;
4847 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
4848 case KVM_CAP_IRQ_ROUTING:
4849 return KVM_MAX_IRQ_ROUTES;
4851 #if KVM_MAX_NR_ADDRESS_SPACES > 1
4852 case KVM_CAP_MULTI_ADDRESS_SPACE:
4854 return kvm_arch_nr_memslot_as_ids(kvm);
4855 return KVM_MAX_NR_ADDRESS_SPACES;
4857 case KVM_CAP_NR_MEMSLOTS:
4858 return KVM_USER_MEM_SLOTS;
4859 case KVM_CAP_DIRTY_LOG_RING:
4860 #ifdef CONFIG_HAVE_KVM_DIRTY_RING_TSO
4861 return KVM_DIRTY_RING_MAX_ENTRIES * sizeof(struct kvm_dirty_gfn);
4865 case KVM_CAP_DIRTY_LOG_RING_ACQ_REL:
4866 #ifdef CONFIG_HAVE_KVM_DIRTY_RING_ACQ_REL
4867 return KVM_DIRTY_RING_MAX_ENTRIES * sizeof(struct kvm_dirty_gfn);
4871 #ifdef CONFIG_NEED_KVM_DIRTY_RING_WITH_BITMAP
4872 case KVM_CAP_DIRTY_LOG_RING_WITH_BITMAP:
4874 case KVM_CAP_BINARY_STATS_FD:
4875 case KVM_CAP_SYSTEM_EVENT_DATA:
4876 case KVM_CAP_DEVICE_CTRL:
4878 #ifdef CONFIG_KVM_GENERIC_MEMORY_ATTRIBUTES
4879 case KVM_CAP_MEMORY_ATTRIBUTES:
4880 return kvm_supported_mem_attributes(kvm);
4882 #ifdef CONFIG_KVM_PRIVATE_MEM
4883 case KVM_CAP_GUEST_MEMFD:
4884 return !kvm || kvm_arch_has_private_mem(kvm);
4889 return kvm_vm_ioctl_check_extension(kvm, arg);
4892 static int kvm_vm_ioctl_enable_dirty_log_ring(struct kvm *kvm, u32 size)
4896 if (!KVM_DIRTY_LOG_PAGE_OFFSET)
4899 /* the size should be power of 2 */
4900 if (!size || (size & (size - 1)))
4903 /* Should be bigger to keep the reserved entries, or a page */
4904 if (size < kvm_dirty_ring_get_rsvd_entries() *
4905 sizeof(struct kvm_dirty_gfn) || size < PAGE_SIZE)
4908 if (size > KVM_DIRTY_RING_MAX_ENTRIES *
4909 sizeof(struct kvm_dirty_gfn))
4912 /* We only allow it to set once */
4913 if (kvm->dirty_ring_size)
4916 mutex_lock(&kvm->lock);
4918 if (kvm->created_vcpus) {
4919 /* We don't allow to change this value after vcpu created */
4922 kvm->dirty_ring_size = size;
4926 mutex_unlock(&kvm->lock);
4930 static int kvm_vm_ioctl_reset_dirty_pages(struct kvm *kvm)
4933 struct kvm_vcpu *vcpu;
4936 if (!kvm->dirty_ring_size)
4939 mutex_lock(&kvm->slots_lock);
4941 kvm_for_each_vcpu(i, vcpu, kvm)
4942 cleared += kvm_dirty_ring_reset(vcpu->kvm, &vcpu->dirty_ring);
4944 mutex_unlock(&kvm->slots_lock);
4947 kvm_flush_remote_tlbs(kvm);
4952 int __attribute__((weak)) kvm_vm_ioctl_enable_cap(struct kvm *kvm,
4953 struct kvm_enable_cap *cap)
4958 bool kvm_are_all_memslots_empty(struct kvm *kvm)
4962 lockdep_assert_held(&kvm->slots_lock);
4964 for (i = 0; i < kvm_arch_nr_memslot_as_ids(kvm); i++) {
4965 if (!kvm_memslots_empty(__kvm_memslots(kvm, i)))
4971 EXPORT_SYMBOL_GPL(kvm_are_all_memslots_empty);
4973 static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm,
4974 struct kvm_enable_cap *cap)
4977 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4978 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2: {
4979 u64 allowed_options = KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE;
4981 if (cap->args[0] & KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE)
4982 allowed_options = KVM_DIRTY_LOG_MANUAL_CAPS;
4984 if (cap->flags || (cap->args[0] & ~allowed_options))
4986 kvm->manual_dirty_log_protect = cap->args[0];
4990 case KVM_CAP_HALT_POLL: {
4991 if (cap->flags || cap->args[0] != (unsigned int)cap->args[0])
4994 kvm->max_halt_poll_ns = cap->args[0];
4997 * Ensure kvm->override_halt_poll_ns does not become visible
4998 * before kvm->max_halt_poll_ns.
5000 * Pairs with the smp_rmb() in kvm_vcpu_max_halt_poll_ns().
5003 kvm->override_halt_poll_ns = true;
5007 case KVM_CAP_DIRTY_LOG_RING:
5008 case KVM_CAP_DIRTY_LOG_RING_ACQ_REL:
5009 if (!kvm_vm_ioctl_check_extension_generic(kvm, cap->cap))
5012 return kvm_vm_ioctl_enable_dirty_log_ring(kvm, cap->args[0]);
5013 case KVM_CAP_DIRTY_LOG_RING_WITH_BITMAP: {
5016 if (!IS_ENABLED(CONFIG_NEED_KVM_DIRTY_RING_WITH_BITMAP) ||
5017 !kvm->dirty_ring_size || cap->flags)
5020 mutex_lock(&kvm->slots_lock);
5023 * For simplicity, allow enabling ring+bitmap if and only if
5024 * there are no memslots, e.g. to ensure all memslots allocate
5025 * a bitmap after the capability is enabled.
5027 if (kvm_are_all_memslots_empty(kvm)) {
5028 kvm->dirty_ring_with_bitmap = true;
5032 mutex_unlock(&kvm->slots_lock);
5037 return kvm_vm_ioctl_enable_cap(kvm, cap);
5041 static ssize_t kvm_vm_stats_read(struct file *file, char __user *user_buffer,
5042 size_t size, loff_t *offset)
5044 struct kvm *kvm = file->private_data;
5046 return kvm_stats_read(kvm->stats_id, &kvm_vm_stats_header,
5047 &kvm_vm_stats_desc[0], &kvm->stat,
5048 sizeof(kvm->stat), user_buffer, size, offset);
5051 static int kvm_vm_stats_release(struct inode *inode, struct file *file)
5053 struct kvm *kvm = file->private_data;
5059 static const struct file_operations kvm_vm_stats_fops = {
5060 .owner = THIS_MODULE,
5061 .read = kvm_vm_stats_read,
5062 .release = kvm_vm_stats_release,
5063 .llseek = noop_llseek,
5066 static int kvm_vm_ioctl_get_stats_fd(struct kvm *kvm)
5071 fd = get_unused_fd_flags(O_CLOEXEC);
5075 file = anon_inode_getfile("kvm-vm-stats",
5076 &kvm_vm_stats_fops, kvm, O_RDONLY);
5079 return PTR_ERR(file);
5084 file->f_mode |= FMODE_PREAD;
5085 fd_install(fd, file);
5090 #define SANITY_CHECK_MEM_REGION_FIELD(field) \
5092 BUILD_BUG_ON(offsetof(struct kvm_userspace_memory_region, field) != \
5093 offsetof(struct kvm_userspace_memory_region2, field)); \
5094 BUILD_BUG_ON(sizeof_field(struct kvm_userspace_memory_region, field) != \
5095 sizeof_field(struct kvm_userspace_memory_region2, field)); \
5098 static long kvm_vm_ioctl(struct file *filp,
5099 unsigned int ioctl, unsigned long arg)
5101 struct kvm *kvm = filp->private_data;
5102 void __user *argp = (void __user *)arg;
5105 if (kvm->mm != current->mm || kvm->vm_dead)
5108 case KVM_CREATE_VCPU:
5109 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
5111 case KVM_ENABLE_CAP: {
5112 struct kvm_enable_cap cap;
5115 if (copy_from_user(&cap, argp, sizeof(cap)))
5117 r = kvm_vm_ioctl_enable_cap_generic(kvm, &cap);
5120 case KVM_SET_USER_MEMORY_REGION2:
5121 case KVM_SET_USER_MEMORY_REGION: {
5122 struct kvm_userspace_memory_region2 mem;
5125 if (ioctl == KVM_SET_USER_MEMORY_REGION) {
5127 * Fields beyond struct kvm_userspace_memory_region shouldn't be
5128 * accessed, but avoid leaking kernel memory in case of a bug.
5130 memset(&mem, 0, sizeof(mem));
5131 size = sizeof(struct kvm_userspace_memory_region);
5133 size = sizeof(struct kvm_userspace_memory_region2);
5136 /* Ensure the common parts of the two structs are identical. */
5137 SANITY_CHECK_MEM_REGION_FIELD(slot);
5138 SANITY_CHECK_MEM_REGION_FIELD(flags);
5139 SANITY_CHECK_MEM_REGION_FIELD(guest_phys_addr);
5140 SANITY_CHECK_MEM_REGION_FIELD(memory_size);
5141 SANITY_CHECK_MEM_REGION_FIELD(userspace_addr);
5144 if (copy_from_user(&mem, argp, size))
5148 if (ioctl == KVM_SET_USER_MEMORY_REGION &&
5149 (mem.flags & ~KVM_SET_USER_MEMORY_REGION_V1_FLAGS))
5152 r = kvm_vm_ioctl_set_memory_region(kvm, &mem);
5155 case KVM_GET_DIRTY_LOG: {
5156 struct kvm_dirty_log log;
5159 if (copy_from_user(&log, argp, sizeof(log)))
5161 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
5164 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
5165 case KVM_CLEAR_DIRTY_LOG: {
5166 struct kvm_clear_dirty_log log;
5169 if (copy_from_user(&log, argp, sizeof(log)))
5171 r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
5175 #ifdef CONFIG_KVM_MMIO
5176 case KVM_REGISTER_COALESCED_MMIO: {
5177 struct kvm_coalesced_mmio_zone zone;
5180 if (copy_from_user(&zone, argp, sizeof(zone)))
5182 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
5185 case KVM_UNREGISTER_COALESCED_MMIO: {
5186 struct kvm_coalesced_mmio_zone zone;
5189 if (copy_from_user(&zone, argp, sizeof(zone)))
5191 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
5196 struct kvm_irqfd data;
5199 if (copy_from_user(&data, argp, sizeof(data)))
5201 r = kvm_irqfd(kvm, &data);
5204 case KVM_IOEVENTFD: {
5205 struct kvm_ioeventfd data;
5208 if (copy_from_user(&data, argp, sizeof(data)))
5210 r = kvm_ioeventfd(kvm, &data);
5213 #ifdef CONFIG_HAVE_KVM_MSI
5214 case KVM_SIGNAL_MSI: {
5218 if (copy_from_user(&msi, argp, sizeof(msi)))
5220 r = kvm_send_userspace_msi(kvm, &msi);
5224 #ifdef __KVM_HAVE_IRQ_LINE
5225 case KVM_IRQ_LINE_STATUS:
5226 case KVM_IRQ_LINE: {
5227 struct kvm_irq_level irq_event;
5230 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
5233 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
5234 ioctl == KVM_IRQ_LINE_STATUS);
5239 if (ioctl == KVM_IRQ_LINE_STATUS) {
5240 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
5248 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
5249 case KVM_SET_GSI_ROUTING: {
5250 struct kvm_irq_routing routing;
5251 struct kvm_irq_routing __user *urouting;
5252 struct kvm_irq_routing_entry *entries = NULL;
5255 if (copy_from_user(&routing, argp, sizeof(routing)))
5258 if (!kvm_arch_can_set_irq_routing(kvm))
5260 if (routing.nr > KVM_MAX_IRQ_ROUTES)
5266 entries = vmemdup_array_user(urouting->entries,
5267 routing.nr, sizeof(*entries));
5268 if (IS_ERR(entries)) {
5269 r = PTR_ERR(entries);
5273 r = kvm_set_irq_routing(kvm, entries, routing.nr,
5278 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
5279 #ifdef CONFIG_KVM_GENERIC_MEMORY_ATTRIBUTES
5280 case KVM_SET_MEMORY_ATTRIBUTES: {
5281 struct kvm_memory_attributes attrs;
5284 if (copy_from_user(&attrs, argp, sizeof(attrs)))
5287 r = kvm_vm_ioctl_set_mem_attributes(kvm, &attrs);
5290 #endif /* CONFIG_KVM_GENERIC_MEMORY_ATTRIBUTES */
5291 case KVM_CREATE_DEVICE: {
5292 struct kvm_create_device cd;
5295 if (copy_from_user(&cd, argp, sizeof(cd)))
5298 r = kvm_ioctl_create_device(kvm, &cd);
5303 if (copy_to_user(argp, &cd, sizeof(cd)))
5309 case KVM_CHECK_EXTENSION:
5310 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
5312 case KVM_RESET_DIRTY_RINGS:
5313 r = kvm_vm_ioctl_reset_dirty_pages(kvm);
5315 case KVM_GET_STATS_FD:
5316 r = kvm_vm_ioctl_get_stats_fd(kvm);
5318 #ifdef CONFIG_KVM_PRIVATE_MEM
5319 case KVM_CREATE_GUEST_MEMFD: {
5320 struct kvm_create_guest_memfd guest_memfd;
5323 if (copy_from_user(&guest_memfd, argp, sizeof(guest_memfd)))
5326 r = kvm_gmem_create(kvm, &guest_memfd);
5331 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
5337 #ifdef CONFIG_KVM_COMPAT
5338 struct compat_kvm_dirty_log {
5342 compat_uptr_t dirty_bitmap; /* one bit per page */
5347 struct compat_kvm_clear_dirty_log {
5352 compat_uptr_t dirty_bitmap; /* one bit per page */
5357 long __weak kvm_arch_vm_compat_ioctl(struct file *filp, unsigned int ioctl,
5363 static long kvm_vm_compat_ioctl(struct file *filp,
5364 unsigned int ioctl, unsigned long arg)
5366 struct kvm *kvm = filp->private_data;
5369 if (kvm->mm != current->mm || kvm->vm_dead)
5372 r = kvm_arch_vm_compat_ioctl(filp, ioctl, arg);
5377 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
5378 case KVM_CLEAR_DIRTY_LOG: {
5379 struct compat_kvm_clear_dirty_log compat_log;
5380 struct kvm_clear_dirty_log log;
5382 if (copy_from_user(&compat_log, (void __user *)arg,
5383 sizeof(compat_log)))
5385 log.slot = compat_log.slot;
5386 log.num_pages = compat_log.num_pages;
5387 log.first_page = compat_log.first_page;
5388 log.padding2 = compat_log.padding2;
5389 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
5391 r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
5395 case KVM_GET_DIRTY_LOG: {
5396 struct compat_kvm_dirty_log compat_log;
5397 struct kvm_dirty_log log;
5399 if (copy_from_user(&compat_log, (void __user *)arg,
5400 sizeof(compat_log)))
5402 log.slot = compat_log.slot;
5403 log.padding1 = compat_log.padding1;
5404 log.padding2 = compat_log.padding2;
5405 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
5407 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
5411 r = kvm_vm_ioctl(filp, ioctl, arg);
5417 static struct file_operations kvm_vm_fops = {
5418 .release = kvm_vm_release,
5419 .unlocked_ioctl = kvm_vm_ioctl,
5420 .llseek = noop_llseek,
5421 KVM_COMPAT(kvm_vm_compat_ioctl),
5424 bool file_is_kvm(struct file *file)
5426 return file && file->f_op == &kvm_vm_fops;
5428 EXPORT_SYMBOL_GPL(file_is_kvm);
5430 static int kvm_dev_ioctl_create_vm(unsigned long type)
5432 char fdname[ITOA_MAX_LEN + 1];
5437 fd = get_unused_fd_flags(O_CLOEXEC);
5441 snprintf(fdname, sizeof(fdname), "%d", fd);
5443 kvm = kvm_create_vm(type, fdname);
5449 file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
5456 * Don't call kvm_put_kvm anymore at this point; file->f_op is
5457 * already set, with ->release() being kvm_vm_release(). In error
5458 * cases it will be called by the final fput(file) and will take
5459 * care of doing kvm_put_kvm(kvm).
5461 kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm);
5463 fd_install(fd, file);
5473 static long kvm_dev_ioctl(struct file *filp,
5474 unsigned int ioctl, unsigned long arg)
5479 case KVM_GET_API_VERSION:
5482 r = KVM_API_VERSION;
5485 r = kvm_dev_ioctl_create_vm(arg);
5487 case KVM_CHECK_EXTENSION:
5488 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
5490 case KVM_GET_VCPU_MMAP_SIZE:
5493 r = PAGE_SIZE; /* struct kvm_run */
5495 r += PAGE_SIZE; /* pio data page */
5497 #ifdef CONFIG_KVM_MMIO
5498 r += PAGE_SIZE; /* coalesced mmio ring page */
5502 return kvm_arch_dev_ioctl(filp, ioctl, arg);
5508 static struct file_operations kvm_chardev_ops = {
5509 .unlocked_ioctl = kvm_dev_ioctl,
5510 .llseek = noop_llseek,
5511 KVM_COMPAT(kvm_dev_ioctl),
5514 static struct miscdevice kvm_dev = {
5520 #ifdef CONFIG_KVM_GENERIC_HARDWARE_ENABLING
5521 __visible bool kvm_rebooting;
5522 EXPORT_SYMBOL_GPL(kvm_rebooting);
5524 static DEFINE_PER_CPU(bool, hardware_enabled);
5525 static int kvm_usage_count;
5527 static int __hardware_enable_nolock(void)
5529 if (__this_cpu_read(hardware_enabled))
5532 if (kvm_arch_hardware_enable()) {
5533 pr_info("kvm: enabling virtualization on CPU%d failed\n",
5534 raw_smp_processor_id());
5538 __this_cpu_write(hardware_enabled, true);
5542 static void hardware_enable_nolock(void *failed)
5544 if (__hardware_enable_nolock())
5548 static int kvm_online_cpu(unsigned int cpu)
5553 * Abort the CPU online process if hardware virtualization cannot
5554 * be enabled. Otherwise running VMs would encounter unrecoverable
5555 * errors when scheduled to this CPU.
5557 mutex_lock(&kvm_lock);
5558 if (kvm_usage_count)
5559 ret = __hardware_enable_nolock();
5560 mutex_unlock(&kvm_lock);
5564 static void hardware_disable_nolock(void *junk)
5567 * Note, hardware_disable_all_nolock() tells all online CPUs to disable
5568 * hardware, not just CPUs that successfully enabled hardware!
5570 if (!__this_cpu_read(hardware_enabled))
5573 kvm_arch_hardware_disable();
5575 __this_cpu_write(hardware_enabled, false);
5578 static int kvm_offline_cpu(unsigned int cpu)
5580 mutex_lock(&kvm_lock);
5581 if (kvm_usage_count)
5582 hardware_disable_nolock(NULL);
5583 mutex_unlock(&kvm_lock);
5587 static void hardware_disable_all_nolock(void)
5589 BUG_ON(!kvm_usage_count);
5592 if (!kvm_usage_count)
5593 on_each_cpu(hardware_disable_nolock, NULL, 1);
5596 static void hardware_disable_all(void)
5599 mutex_lock(&kvm_lock);
5600 hardware_disable_all_nolock();
5601 mutex_unlock(&kvm_lock);
5605 static int hardware_enable_all(void)
5607 atomic_t failed = ATOMIC_INIT(0);
5611 * Do not enable hardware virtualization if the system is going down.
5612 * If userspace initiated a forced reboot, e.g. reboot -f, then it's
5613 * possible for an in-flight KVM_CREATE_VM to trigger hardware enabling
5614 * after kvm_reboot() is called. Note, this relies on system_state
5615 * being set _before_ kvm_reboot(), which is why KVM uses a syscore ops
5616 * hook instead of registering a dedicated reboot notifier (the latter
5617 * runs before system_state is updated).
5619 if (system_state == SYSTEM_HALT || system_state == SYSTEM_POWER_OFF ||
5620 system_state == SYSTEM_RESTART)
5624 * When onlining a CPU, cpu_online_mask is set before kvm_online_cpu()
5625 * is called, and so on_each_cpu() between them includes the CPU that
5626 * is being onlined. As a result, hardware_enable_nolock() may get
5627 * invoked before kvm_online_cpu(), which also enables hardware if the
5628 * usage count is non-zero. Disable CPU hotplug to avoid attempting to
5629 * enable hardware multiple times.
5632 mutex_lock(&kvm_lock);
5637 if (kvm_usage_count == 1) {
5638 on_each_cpu(hardware_enable_nolock, &failed, 1);
5640 if (atomic_read(&failed)) {
5641 hardware_disable_all_nolock();
5646 mutex_unlock(&kvm_lock);
5652 static void kvm_shutdown(void)
5655 * Disable hardware virtualization and set kvm_rebooting to indicate
5656 * that KVM has asynchronously disabled hardware virtualization, i.e.
5657 * that relevant errors and exceptions aren't entirely unexpected.
5658 * Some flavors of hardware virtualization need to be disabled before
5659 * transferring control to firmware (to perform shutdown/reboot), e.g.
5660 * on x86, virtualization can block INIT interrupts, which are used by
5661 * firmware to pull APs back under firmware control. Note, this path
5662 * is used for both shutdown and reboot scenarios, i.e. neither name is
5663 * 100% comprehensive.
5665 pr_info("kvm: exiting hardware virtualization\n");
5666 kvm_rebooting = true;
5667 on_each_cpu(hardware_disable_nolock, NULL, 1);
5670 static int kvm_suspend(void)
5673 * Secondary CPUs and CPU hotplug are disabled across the suspend/resume
5674 * callbacks, i.e. no need to acquire kvm_lock to ensure the usage count
5675 * is stable. Assert that kvm_lock is not held to ensure the system
5676 * isn't suspended while KVM is enabling hardware. Hardware enabling
5677 * can be preempted, but the task cannot be frozen until it has dropped
5678 * all locks (userspace tasks are frozen via a fake signal).
5680 lockdep_assert_not_held(&kvm_lock);
5681 lockdep_assert_irqs_disabled();
5683 if (kvm_usage_count)
5684 hardware_disable_nolock(NULL);
5688 static void kvm_resume(void)
5690 lockdep_assert_not_held(&kvm_lock);
5691 lockdep_assert_irqs_disabled();
5693 if (kvm_usage_count)
5694 WARN_ON_ONCE(__hardware_enable_nolock());
5697 static struct syscore_ops kvm_syscore_ops = {
5698 .suspend = kvm_suspend,
5699 .resume = kvm_resume,
5700 .shutdown = kvm_shutdown,
5702 #else /* CONFIG_KVM_GENERIC_HARDWARE_ENABLING */
5703 static int hardware_enable_all(void)
5708 static void hardware_disable_all(void)
5712 #endif /* CONFIG_KVM_GENERIC_HARDWARE_ENABLING */
5714 static void kvm_iodevice_destructor(struct kvm_io_device *dev)
5716 if (dev->ops->destructor)
5717 dev->ops->destructor(dev);
5720 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
5724 for (i = 0; i < bus->dev_count; i++) {
5725 struct kvm_io_device *pos = bus->range[i].dev;
5727 kvm_iodevice_destructor(pos);
5732 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
5733 const struct kvm_io_range *r2)
5735 gpa_t addr1 = r1->addr;
5736 gpa_t addr2 = r2->addr;
5741 /* If r2->len == 0, match the exact address. If r2->len != 0,
5742 * accept any overlapping write. Any order is acceptable for
5743 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
5744 * we process all of them.
5757 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
5759 return kvm_io_bus_cmp(p1, p2);
5762 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
5763 gpa_t addr, int len)
5765 struct kvm_io_range *range, key;
5768 key = (struct kvm_io_range) {
5773 range = bsearch(&key, bus->range, bus->dev_count,
5774 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
5778 off = range - bus->range;
5780 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
5786 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
5787 struct kvm_io_range *range, const void *val)
5791 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
5795 while (idx < bus->dev_count &&
5796 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
5797 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
5806 /* kvm_io_bus_write - called under kvm->slots_lock */
5807 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
5808 int len, const void *val)
5810 struct kvm_io_bus *bus;
5811 struct kvm_io_range range;
5814 range = (struct kvm_io_range) {
5819 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
5822 r = __kvm_io_bus_write(vcpu, bus, &range, val);
5823 return r < 0 ? r : 0;
5825 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
5827 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
5828 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
5829 gpa_t addr, int len, const void *val, long cookie)
5831 struct kvm_io_bus *bus;
5832 struct kvm_io_range range;
5834 range = (struct kvm_io_range) {
5839 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
5843 /* First try the device referenced by cookie. */
5844 if ((cookie >= 0) && (cookie < bus->dev_count) &&
5845 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
5846 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
5851 * cookie contained garbage; fall back to search and return the
5852 * correct cookie value.
5854 return __kvm_io_bus_write(vcpu, bus, &range, val);
5857 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
5858 struct kvm_io_range *range, void *val)
5862 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
5866 while (idx < bus->dev_count &&
5867 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
5868 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
5877 /* kvm_io_bus_read - called under kvm->slots_lock */
5878 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
5881 struct kvm_io_bus *bus;
5882 struct kvm_io_range range;
5885 range = (struct kvm_io_range) {
5890 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
5893 r = __kvm_io_bus_read(vcpu, bus, &range, val);
5894 return r < 0 ? r : 0;
5897 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
5898 int len, struct kvm_io_device *dev)
5901 struct kvm_io_bus *new_bus, *bus;
5902 struct kvm_io_range range;
5904 lockdep_assert_held(&kvm->slots_lock);
5906 bus = kvm_get_bus(kvm, bus_idx);
5910 /* exclude ioeventfd which is limited by maximum fd */
5911 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
5914 new_bus = kmalloc(struct_size(bus, range, bus->dev_count + 1),
5915 GFP_KERNEL_ACCOUNT);
5919 range = (struct kvm_io_range) {
5925 for (i = 0; i < bus->dev_count; i++)
5926 if (kvm_io_bus_cmp(&bus->range[i], &range) > 0)
5929 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
5930 new_bus->dev_count++;
5931 new_bus->range[i] = range;
5932 memcpy(new_bus->range + i + 1, bus->range + i,
5933 (bus->dev_count - i) * sizeof(struct kvm_io_range));
5934 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
5935 synchronize_srcu_expedited(&kvm->srcu);
5941 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
5942 struct kvm_io_device *dev)
5945 struct kvm_io_bus *new_bus, *bus;
5947 lockdep_assert_held(&kvm->slots_lock);
5949 bus = kvm_get_bus(kvm, bus_idx);
5953 for (i = 0; i < bus->dev_count; i++) {
5954 if (bus->range[i].dev == dev) {
5959 if (i == bus->dev_count)
5962 new_bus = kmalloc(struct_size(bus, range, bus->dev_count - 1),
5963 GFP_KERNEL_ACCOUNT);
5965 memcpy(new_bus, bus, struct_size(bus, range, i));
5966 new_bus->dev_count--;
5967 memcpy(new_bus->range + i, bus->range + i + 1,
5968 flex_array_size(new_bus, range, new_bus->dev_count - i));
5971 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
5972 synchronize_srcu_expedited(&kvm->srcu);
5975 * If NULL bus is installed, destroy the old bus, including all the
5976 * attached devices. Otherwise, destroy the caller's device only.
5979 pr_err("kvm: failed to shrink bus, removing it completely\n");
5980 kvm_io_bus_destroy(bus);
5984 kvm_iodevice_destructor(dev);
5989 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
5992 struct kvm_io_bus *bus;
5993 int dev_idx, srcu_idx;
5994 struct kvm_io_device *iodev = NULL;
5996 srcu_idx = srcu_read_lock(&kvm->srcu);
5998 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
6002 dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
6006 iodev = bus->range[dev_idx].dev;
6009 srcu_read_unlock(&kvm->srcu, srcu_idx);
6013 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
6015 static int kvm_debugfs_open(struct inode *inode, struct file *file,
6016 int (*get)(void *, u64 *), int (*set)(void *, u64),
6020 struct kvm_stat_data *stat_data = inode->i_private;
6023 * The debugfs files are a reference to the kvm struct which
6024 * is still valid when kvm_destroy_vm is called. kvm_get_kvm_safe
6025 * avoids the race between open and the removal of the debugfs directory.
6027 if (!kvm_get_kvm_safe(stat_data->kvm))
6030 ret = simple_attr_open(inode, file, get,
6031 kvm_stats_debugfs_mode(stat_data->desc) & 0222
6034 kvm_put_kvm(stat_data->kvm);
6039 static int kvm_debugfs_release(struct inode *inode, struct file *file)
6041 struct kvm_stat_data *stat_data = inode->i_private;
6043 simple_attr_release(inode, file);
6044 kvm_put_kvm(stat_data->kvm);
6049 static int kvm_get_stat_per_vm(struct kvm *kvm, size_t offset, u64 *val)
6051 *val = *(u64 *)((void *)(&kvm->stat) + offset);
6056 static int kvm_clear_stat_per_vm(struct kvm *kvm, size_t offset)
6058 *(u64 *)((void *)(&kvm->stat) + offset) = 0;
6063 static int kvm_get_stat_per_vcpu(struct kvm *kvm, size_t offset, u64 *val)
6066 struct kvm_vcpu *vcpu;
6070 kvm_for_each_vcpu(i, vcpu, kvm)
6071 *val += *(u64 *)((void *)(&vcpu->stat) + offset);
6076 static int kvm_clear_stat_per_vcpu(struct kvm *kvm, size_t offset)
6079 struct kvm_vcpu *vcpu;
6081 kvm_for_each_vcpu(i, vcpu, kvm)
6082 *(u64 *)((void *)(&vcpu->stat) + offset) = 0;
6087 static int kvm_stat_data_get(void *data, u64 *val)
6090 struct kvm_stat_data *stat_data = data;
6092 switch (stat_data->kind) {
6094 r = kvm_get_stat_per_vm(stat_data->kvm,
6095 stat_data->desc->desc.offset, val);
6098 r = kvm_get_stat_per_vcpu(stat_data->kvm,
6099 stat_data->desc->desc.offset, val);
6106 static int kvm_stat_data_clear(void *data, u64 val)
6109 struct kvm_stat_data *stat_data = data;
6114 switch (stat_data->kind) {
6116 r = kvm_clear_stat_per_vm(stat_data->kvm,
6117 stat_data->desc->desc.offset);
6120 r = kvm_clear_stat_per_vcpu(stat_data->kvm,
6121 stat_data->desc->desc.offset);
6128 static int kvm_stat_data_open(struct inode *inode, struct file *file)
6130 __simple_attr_check_format("%llu\n", 0ull);
6131 return kvm_debugfs_open(inode, file, kvm_stat_data_get,
6132 kvm_stat_data_clear, "%llu\n");
6135 static const struct file_operations stat_fops_per_vm = {
6136 .owner = THIS_MODULE,
6137 .open = kvm_stat_data_open,
6138 .release = kvm_debugfs_release,
6139 .read = simple_attr_read,
6140 .write = simple_attr_write,
6141 .llseek = no_llseek,
6144 static int vm_stat_get(void *_offset, u64 *val)
6146 unsigned offset = (long)_offset;
6151 mutex_lock(&kvm_lock);
6152 list_for_each_entry(kvm, &vm_list, vm_list) {
6153 kvm_get_stat_per_vm(kvm, offset, &tmp_val);
6156 mutex_unlock(&kvm_lock);
6160 static int vm_stat_clear(void *_offset, u64 val)
6162 unsigned offset = (long)_offset;
6168 mutex_lock(&kvm_lock);
6169 list_for_each_entry(kvm, &vm_list, vm_list) {
6170 kvm_clear_stat_per_vm(kvm, offset);
6172 mutex_unlock(&kvm_lock);
6177 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n");
6178 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_readonly_fops, vm_stat_get, NULL, "%llu\n");
6180 static int vcpu_stat_get(void *_offset, u64 *val)
6182 unsigned offset = (long)_offset;
6187 mutex_lock(&kvm_lock);
6188 list_for_each_entry(kvm, &vm_list, vm_list) {
6189 kvm_get_stat_per_vcpu(kvm, offset, &tmp_val);
6192 mutex_unlock(&kvm_lock);
6196 static int vcpu_stat_clear(void *_offset, u64 val)
6198 unsigned offset = (long)_offset;
6204 mutex_lock(&kvm_lock);
6205 list_for_each_entry(kvm, &vm_list, vm_list) {
6206 kvm_clear_stat_per_vcpu(kvm, offset);
6208 mutex_unlock(&kvm_lock);
6213 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear,
6215 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_readonly_fops, vcpu_stat_get, NULL, "%llu\n");
6217 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm)
6219 struct kobj_uevent_env *env;
6220 unsigned long long created, active;
6222 if (!kvm_dev.this_device || !kvm)
6225 mutex_lock(&kvm_lock);
6226 if (type == KVM_EVENT_CREATE_VM) {
6227 kvm_createvm_count++;
6229 } else if (type == KVM_EVENT_DESTROY_VM) {
6232 created = kvm_createvm_count;
6233 active = kvm_active_vms;
6234 mutex_unlock(&kvm_lock);
6236 env = kzalloc(sizeof(*env), GFP_KERNEL_ACCOUNT);
6240 add_uevent_var(env, "CREATED=%llu", created);
6241 add_uevent_var(env, "COUNT=%llu", active);
6243 if (type == KVM_EVENT_CREATE_VM) {
6244 add_uevent_var(env, "EVENT=create");
6245 kvm->userspace_pid = task_pid_nr(current);
6246 } else if (type == KVM_EVENT_DESTROY_VM) {
6247 add_uevent_var(env, "EVENT=destroy");
6249 add_uevent_var(env, "PID=%d", kvm->userspace_pid);
6251 if (!IS_ERR(kvm->debugfs_dentry)) {
6252 char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL_ACCOUNT);
6255 tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX);
6257 add_uevent_var(env, "STATS_PATH=%s", tmp);
6261 /* no need for checks, since we are adding at most only 5 keys */
6262 env->envp[env->envp_idx++] = NULL;
6263 kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp);
6267 static void kvm_init_debug(void)
6269 const struct file_operations *fops;
6270 const struct _kvm_stats_desc *pdesc;
6273 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
6275 for (i = 0; i < kvm_vm_stats_header.num_desc; ++i) {
6276 pdesc = &kvm_vm_stats_desc[i];
6277 if (kvm_stats_debugfs_mode(pdesc) & 0222)
6278 fops = &vm_stat_fops;
6280 fops = &vm_stat_readonly_fops;
6281 debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
6283 (void *)(long)pdesc->desc.offset, fops);
6286 for (i = 0; i < kvm_vcpu_stats_header.num_desc; ++i) {
6287 pdesc = &kvm_vcpu_stats_desc[i];
6288 if (kvm_stats_debugfs_mode(pdesc) & 0222)
6289 fops = &vcpu_stat_fops;
6291 fops = &vcpu_stat_readonly_fops;
6292 debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
6294 (void *)(long)pdesc->desc.offset, fops);
6299 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
6301 return container_of(pn, struct kvm_vcpu, preempt_notifier);
6304 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
6306 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
6308 WRITE_ONCE(vcpu->preempted, false);
6309 WRITE_ONCE(vcpu->ready, false);
6311 __this_cpu_write(kvm_running_vcpu, vcpu);
6312 kvm_arch_sched_in(vcpu, cpu);
6313 kvm_arch_vcpu_load(vcpu, cpu);
6316 static void kvm_sched_out(struct preempt_notifier *pn,
6317 struct task_struct *next)
6319 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
6321 if (current->on_rq) {
6322 WRITE_ONCE(vcpu->preempted, true);
6323 WRITE_ONCE(vcpu->ready, true);
6325 kvm_arch_vcpu_put(vcpu);
6326 __this_cpu_write(kvm_running_vcpu, NULL);
6330 * kvm_get_running_vcpu - get the vcpu running on the current CPU.
6332 * We can disable preemption locally around accessing the per-CPU variable,
6333 * and use the resolved vcpu pointer after enabling preemption again,
6334 * because even if the current thread is migrated to another CPU, reading
6335 * the per-CPU value later will give us the same value as we update the
6336 * per-CPU variable in the preempt notifier handlers.
6338 struct kvm_vcpu *kvm_get_running_vcpu(void)
6340 struct kvm_vcpu *vcpu;
6343 vcpu = __this_cpu_read(kvm_running_vcpu);
6348 EXPORT_SYMBOL_GPL(kvm_get_running_vcpu);
6351 * kvm_get_running_vcpus - get the per-CPU array of currently running vcpus.
6353 struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void)
6355 return &kvm_running_vcpu;
6358 #ifdef CONFIG_GUEST_PERF_EVENTS
6359 static unsigned int kvm_guest_state(void)
6361 struct kvm_vcpu *vcpu = kvm_get_running_vcpu();
6364 if (!kvm_arch_pmi_in_guest(vcpu))
6367 state = PERF_GUEST_ACTIVE;
6368 if (!kvm_arch_vcpu_in_kernel(vcpu))
6369 state |= PERF_GUEST_USER;
6374 static unsigned long kvm_guest_get_ip(void)
6376 struct kvm_vcpu *vcpu = kvm_get_running_vcpu();
6378 /* Retrieving the IP must be guarded by a call to kvm_guest_state(). */
6379 if (WARN_ON_ONCE(!kvm_arch_pmi_in_guest(vcpu)))
6382 return kvm_arch_vcpu_get_ip(vcpu);
6385 static struct perf_guest_info_callbacks kvm_guest_cbs = {
6386 .state = kvm_guest_state,
6387 .get_ip = kvm_guest_get_ip,
6388 .handle_intel_pt_intr = NULL,
6391 void kvm_register_perf_callbacks(unsigned int (*pt_intr_handler)(void))
6393 kvm_guest_cbs.handle_intel_pt_intr = pt_intr_handler;
6394 perf_register_guest_info_callbacks(&kvm_guest_cbs);
6396 void kvm_unregister_perf_callbacks(void)
6398 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
6402 int kvm_init(unsigned vcpu_size, unsigned vcpu_align, struct module *module)
6407 #ifdef CONFIG_KVM_GENERIC_HARDWARE_ENABLING
6408 r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_ONLINE, "kvm/cpu:online",
6409 kvm_online_cpu, kvm_offline_cpu);
6413 register_syscore_ops(&kvm_syscore_ops);
6416 /* A kmem cache lets us meet the alignment requirements of fx_save. */
6418 vcpu_align = __alignof__(struct kvm_vcpu);
6420 kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align,
6422 offsetof(struct kvm_vcpu, arch),
6423 offsetofend(struct kvm_vcpu, stats_id)
6424 - offsetof(struct kvm_vcpu, arch),
6426 if (!kvm_vcpu_cache) {
6428 goto err_vcpu_cache;
6431 for_each_possible_cpu(cpu) {
6432 if (!alloc_cpumask_var_node(&per_cpu(cpu_kick_mask, cpu),
6433 GFP_KERNEL, cpu_to_node(cpu))) {
6435 goto err_cpu_kick_mask;
6439 r = kvm_irqfd_init();
6443 r = kvm_async_pf_init();
6447 kvm_chardev_ops.owner = module;
6448 kvm_vm_fops.owner = module;
6449 kvm_vcpu_fops.owner = module;
6450 kvm_device_fops.owner = module;
6452 kvm_preempt_ops.sched_in = kvm_sched_in;
6453 kvm_preempt_ops.sched_out = kvm_sched_out;
6457 r = kvm_vfio_ops_init();
6458 if (WARN_ON_ONCE(r))
6461 kvm_gmem_init(module);
6464 * Registration _must_ be the very last thing done, as this exposes
6465 * /dev/kvm to userspace, i.e. all infrastructure must be setup!
6467 r = misc_register(&kvm_dev);
6469 pr_err("kvm: misc device register failed\n");
6476 kvm_vfio_ops_exit();
6478 kvm_async_pf_deinit();
6483 for_each_possible_cpu(cpu)
6484 free_cpumask_var(per_cpu(cpu_kick_mask, cpu));
6485 kmem_cache_destroy(kvm_vcpu_cache);
6487 #ifdef CONFIG_KVM_GENERIC_HARDWARE_ENABLING
6488 unregister_syscore_ops(&kvm_syscore_ops);
6489 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_ONLINE);
6493 EXPORT_SYMBOL_GPL(kvm_init);
6500 * Note, unregistering /dev/kvm doesn't strictly need to come first,
6501 * fops_get(), a.k.a. try_module_get(), prevents acquiring references
6502 * to KVM while the module is being stopped.
6504 misc_deregister(&kvm_dev);
6506 debugfs_remove_recursive(kvm_debugfs_dir);
6507 for_each_possible_cpu(cpu)
6508 free_cpumask_var(per_cpu(cpu_kick_mask, cpu));
6509 kmem_cache_destroy(kvm_vcpu_cache);
6510 kvm_vfio_ops_exit();
6511 kvm_async_pf_deinit();
6512 #ifdef CONFIG_KVM_GENERIC_HARDWARE_ENABLING
6513 unregister_syscore_ops(&kvm_syscore_ops);
6514 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_ONLINE);
6518 EXPORT_SYMBOL_GPL(kvm_exit);
6520 struct kvm_vm_worker_thread_context {
6522 struct task_struct *parent;
6523 struct completion init_done;
6524 kvm_vm_thread_fn_t thread_fn;
6529 static int kvm_vm_worker_thread(void *context)
6532 * The init_context is allocated on the stack of the parent thread, so
6533 * we have to locally copy anything that is needed beyond initialization
6535 struct kvm_vm_worker_thread_context *init_context = context;
6536 struct task_struct *parent;
6537 struct kvm *kvm = init_context->kvm;
6538 kvm_vm_thread_fn_t thread_fn = init_context->thread_fn;
6539 uintptr_t data = init_context->data;
6542 err = kthread_park(current);
6543 /* kthread_park(current) is never supposed to return an error */
6548 err = cgroup_attach_task_all(init_context->parent, current);
6550 kvm_err("%s: cgroup_attach_task_all failed with err %d\n",
6555 set_user_nice(current, task_nice(init_context->parent));
6558 init_context->err = err;
6559 complete(&init_context->init_done);
6560 init_context = NULL;
6565 /* Wait to be woken up by the spawner before proceeding. */
6568 if (!kthread_should_stop())
6569 err = thread_fn(kvm, data);
6573 * Move kthread back to its original cgroup to prevent it lingering in
6574 * the cgroup of the VM process, after the latter finishes its
6577 * kthread_stop() waits on the 'exited' completion condition which is
6578 * set in exit_mm(), via mm_release(), in do_exit(). However, the
6579 * kthread is removed from the cgroup in the cgroup_exit() which is
6580 * called after the exit_mm(). This causes the kthread_stop() to return
6581 * before the kthread actually quits the cgroup.
6584 parent = rcu_dereference(current->real_parent);
6585 get_task_struct(parent);
6587 cgroup_attach_task_all(parent, current);
6588 put_task_struct(parent);
6593 int kvm_vm_create_worker_thread(struct kvm *kvm, kvm_vm_thread_fn_t thread_fn,
6594 uintptr_t data, const char *name,
6595 struct task_struct **thread_ptr)
6597 struct kvm_vm_worker_thread_context init_context = {};
6598 struct task_struct *thread;
6601 init_context.kvm = kvm;
6602 init_context.parent = current;
6603 init_context.thread_fn = thread_fn;
6604 init_context.data = data;
6605 init_completion(&init_context.init_done);
6607 thread = kthread_run(kvm_vm_worker_thread, &init_context,
6608 "%s-%d", name, task_pid_nr(current));
6610 return PTR_ERR(thread);
6612 /* kthread_run is never supposed to return NULL */
6613 WARN_ON(thread == NULL);
6615 wait_for_completion(&init_context.init_done);
6617 if (!init_context.err)
6618 *thread_ptr = thread;
6620 return init_context.err;