1 // SPDX-License-Identifier: GPL-2.0-only
3 * Kernel-based Virtual Machine driver for Linux
5 * This module enables machines with Intel VT-x extensions to run virtual
6 * machines without emulation or binary translation.
8 * Copyright (C) 2006 Qumranet, Inc.
9 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
12 * Avi Kivity <avi@qumranet.com>
13 * Yaniv Kamay <yaniv@qumranet.com>
16 #include <kvm/iodev.h>
18 #include <linux/kvm_host.h>
19 #include <linux/kvm.h>
20 #include <linux/module.h>
21 #include <linux/errno.h>
22 #include <linux/percpu.h>
24 #include <linux/miscdevice.h>
25 #include <linux/vmalloc.h>
26 #include <linux/reboot.h>
27 #include <linux/debugfs.h>
28 #include <linux/highmem.h>
29 #include <linux/file.h>
30 #include <linux/syscore_ops.h>
31 #include <linux/cpu.h>
32 #include <linux/sched/signal.h>
33 #include <linux/sched/mm.h>
34 #include <linux/sched/stat.h>
35 #include <linux/cpumask.h>
36 #include <linux/smp.h>
37 #include <linux/anon_inodes.h>
38 #include <linux/profile.h>
39 #include <linux/kvm_para.h>
40 #include <linux/pagemap.h>
41 #include <linux/mman.h>
42 #include <linux/swap.h>
43 #include <linux/bitops.h>
44 #include <linux/spinlock.h>
45 #include <linux/compat.h>
46 #include <linux/srcu.h>
47 #include <linux/hugetlb.h>
48 #include <linux/slab.h>
49 #include <linux/sort.h>
50 #include <linux/bsearch.h>
52 #include <linux/lockdep.h>
53 #include <linux/kthread.h>
54 #include <linux/suspend.h>
56 #include <asm/processor.h>
57 #include <asm/ioctl.h>
58 #include <linux/uaccess.h>
60 #include "coalesced_mmio.h"
65 #define CREATE_TRACE_POINTS
66 #include <trace/events/kvm.h>
68 #include <linux/kvm_dirty_ring.h>
70 /* Worst case buffer size needed for holding an integer. */
71 #define ITOA_MAX_LEN 12
73 MODULE_AUTHOR("Qumranet");
74 MODULE_LICENSE("GPL");
76 /* Architectures should define their poll value according to the halt latency */
77 unsigned int halt_poll_ns = KVM_HALT_POLL_NS_DEFAULT;
78 module_param(halt_poll_ns, uint, 0644);
79 EXPORT_SYMBOL_GPL(halt_poll_ns);
81 /* Default doubles per-vcpu halt_poll_ns. */
82 unsigned int halt_poll_ns_grow = 2;
83 module_param(halt_poll_ns_grow, uint, 0644);
84 EXPORT_SYMBOL_GPL(halt_poll_ns_grow);
86 /* The start value to grow halt_poll_ns from */
87 unsigned int halt_poll_ns_grow_start = 10000; /* 10us */
88 module_param(halt_poll_ns_grow_start, uint, 0644);
89 EXPORT_SYMBOL_GPL(halt_poll_ns_grow_start);
91 /* Default resets per-vcpu halt_poll_ns . */
92 unsigned int halt_poll_ns_shrink;
93 module_param(halt_poll_ns_shrink, uint, 0644);
94 EXPORT_SYMBOL_GPL(halt_poll_ns_shrink);
99 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
102 DEFINE_MUTEX(kvm_lock);
105 static struct kmem_cache *kvm_vcpu_cache;
107 static __read_mostly struct preempt_ops kvm_preempt_ops;
108 static DEFINE_PER_CPU(struct kvm_vcpu *, kvm_running_vcpu);
110 struct dentry *kvm_debugfs_dir;
111 EXPORT_SYMBOL_GPL(kvm_debugfs_dir);
113 static const struct file_operations stat_fops_per_vm;
115 static struct file_operations kvm_chardev_ops;
117 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
119 #ifdef CONFIG_KVM_COMPAT
120 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
122 #define KVM_COMPAT(c) .compat_ioctl = (c)
125 * For architectures that don't implement a compat infrastructure,
126 * adopt a double line of defense:
127 * - Prevent a compat task from opening /dev/kvm
128 * - If the open has been done by a 64bit task, and the KVM fd
129 * passed to a compat task, let the ioctls fail.
131 static long kvm_no_compat_ioctl(struct file *file, unsigned int ioctl,
132 unsigned long arg) { return -EINVAL; }
134 static int kvm_no_compat_open(struct inode *inode, struct file *file)
136 return is_compat_task() ? -ENODEV : 0;
138 #define KVM_COMPAT(c) .compat_ioctl = kvm_no_compat_ioctl, \
139 .open = kvm_no_compat_open
141 static int hardware_enable_all(void);
142 static void hardware_disable_all(void);
144 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
146 #define KVM_EVENT_CREATE_VM 0
147 #define KVM_EVENT_DESTROY_VM 1
148 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm);
149 static unsigned long long kvm_createvm_count;
150 static unsigned long long kvm_active_vms;
152 static DEFINE_PER_CPU(cpumask_var_t, cpu_kick_mask);
154 __weak void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
155 unsigned long start, unsigned long end)
159 __weak void kvm_arch_guest_memory_reclaimed(struct kvm *kvm)
163 bool kvm_is_zone_device_page(struct page *page)
166 * The metadata used by is_zone_device_page() to determine whether or
167 * not a page is ZONE_DEVICE is guaranteed to be valid if and only if
168 * the device has been pinned, e.g. by get_user_pages(). WARN if the
169 * page_count() is zero to help detect bad usage of this helper.
171 if (WARN_ON_ONCE(!page_count(page)))
174 return is_zone_device_page(page);
178 * Returns a 'struct page' if the pfn is "valid" and backed by a refcounted
179 * page, NULL otherwise. Note, the list of refcounted PG_reserved page types
180 * is likely incomplete, it has been compiled purely through people wanting to
181 * back guest with a certain type of memory and encountering issues.
183 struct page *kvm_pfn_to_refcounted_page(kvm_pfn_t pfn)
190 page = pfn_to_page(pfn);
191 if (!PageReserved(page))
194 /* The ZERO_PAGE(s) is marked PG_reserved, but is refcounted. */
195 if (is_zero_pfn(pfn))
199 * ZONE_DEVICE pages currently set PG_reserved, but from a refcounting
200 * perspective they are "normal" pages, albeit with slightly different
203 if (kvm_is_zone_device_page(page))
210 * Switches to specified vcpu, until a matching vcpu_put()
212 void vcpu_load(struct kvm_vcpu *vcpu)
216 __this_cpu_write(kvm_running_vcpu, vcpu);
217 preempt_notifier_register(&vcpu->preempt_notifier);
218 kvm_arch_vcpu_load(vcpu, cpu);
221 EXPORT_SYMBOL_GPL(vcpu_load);
223 void vcpu_put(struct kvm_vcpu *vcpu)
226 kvm_arch_vcpu_put(vcpu);
227 preempt_notifier_unregister(&vcpu->preempt_notifier);
228 __this_cpu_write(kvm_running_vcpu, NULL);
231 EXPORT_SYMBOL_GPL(vcpu_put);
233 /* TODO: merge with kvm_arch_vcpu_should_kick */
234 static bool kvm_request_needs_ipi(struct kvm_vcpu *vcpu, unsigned req)
236 int mode = kvm_vcpu_exiting_guest_mode(vcpu);
239 * We need to wait for the VCPU to reenable interrupts and get out of
240 * READING_SHADOW_PAGE_TABLES mode.
242 if (req & KVM_REQUEST_WAIT)
243 return mode != OUTSIDE_GUEST_MODE;
246 * Need to kick a running VCPU, but otherwise there is nothing to do.
248 return mode == IN_GUEST_MODE;
251 static void ack_kick(void *_completed)
255 static inline bool kvm_kick_many_cpus(struct cpumask *cpus, bool wait)
257 if (cpumask_empty(cpus))
260 smp_call_function_many(cpus, ack_kick, NULL, wait);
264 static void kvm_make_vcpu_request(struct kvm_vcpu *vcpu, unsigned int req,
265 struct cpumask *tmp, int current_cpu)
269 if (likely(!(req & KVM_REQUEST_NO_ACTION)))
270 __kvm_make_request(req, vcpu);
272 if (!(req & KVM_REQUEST_NO_WAKEUP) && kvm_vcpu_wake_up(vcpu))
276 * Note, the vCPU could get migrated to a different pCPU at any point
277 * after kvm_request_needs_ipi(), which could result in sending an IPI
278 * to the previous pCPU. But, that's OK because the purpose of the IPI
279 * is to ensure the vCPU returns to OUTSIDE_GUEST_MODE, which is
280 * satisfied if the vCPU migrates. Entering READING_SHADOW_PAGE_TABLES
281 * after this point is also OK, as the requirement is only that KVM wait
282 * for vCPUs that were reading SPTEs _before_ any changes were
283 * finalized. See kvm_vcpu_kick() for more details on handling requests.
285 if (kvm_request_needs_ipi(vcpu, req)) {
286 cpu = READ_ONCE(vcpu->cpu);
287 if (cpu != -1 && cpu != current_cpu)
288 __cpumask_set_cpu(cpu, tmp);
292 bool kvm_make_vcpus_request_mask(struct kvm *kvm, unsigned int req,
293 unsigned long *vcpu_bitmap)
295 struct kvm_vcpu *vcpu;
296 struct cpumask *cpus;
302 cpus = this_cpu_cpumask_var_ptr(cpu_kick_mask);
305 for_each_set_bit(i, vcpu_bitmap, KVM_MAX_VCPUS) {
306 vcpu = kvm_get_vcpu(kvm, i);
309 kvm_make_vcpu_request(vcpu, req, cpus, me);
312 called = kvm_kick_many_cpus(cpus, !!(req & KVM_REQUEST_WAIT));
318 bool kvm_make_all_cpus_request_except(struct kvm *kvm, unsigned int req,
319 struct kvm_vcpu *except)
321 struct kvm_vcpu *vcpu;
322 struct cpumask *cpus;
329 cpus = this_cpu_cpumask_var_ptr(cpu_kick_mask);
332 kvm_for_each_vcpu(i, vcpu, kvm) {
335 kvm_make_vcpu_request(vcpu, req, cpus, me);
338 called = kvm_kick_many_cpus(cpus, !!(req & KVM_REQUEST_WAIT));
344 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
346 return kvm_make_all_cpus_request_except(kvm, req, NULL);
348 EXPORT_SYMBOL_GPL(kvm_make_all_cpus_request);
350 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
351 void kvm_flush_remote_tlbs(struct kvm *kvm)
353 ++kvm->stat.generic.remote_tlb_flush_requests;
356 * We want to publish modifications to the page tables before reading
357 * mode. Pairs with a memory barrier in arch-specific code.
358 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
359 * and smp_mb in walk_shadow_page_lockless_begin/end.
360 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
362 * There is already an smp_mb__after_atomic() before
363 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
366 if (!kvm_arch_flush_remote_tlb(kvm)
367 || kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
368 ++kvm->stat.generic.remote_tlb_flush;
370 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
373 static void kvm_flush_shadow_all(struct kvm *kvm)
375 kvm_arch_flush_shadow_all(kvm);
376 kvm_arch_guest_memory_reclaimed(kvm);
379 #ifdef KVM_ARCH_NR_OBJS_PER_MEMORY_CACHE
380 static inline void *mmu_memory_cache_alloc_obj(struct kvm_mmu_memory_cache *mc,
383 gfp_flags |= mc->gfp_zero;
386 return kmem_cache_alloc(mc->kmem_cache, gfp_flags);
388 return (void *)__get_free_page(gfp_flags);
391 int __kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache *mc, int capacity, int min)
393 gfp_t gfp = mc->gfp_custom ? mc->gfp_custom : GFP_KERNEL_ACCOUNT;
396 if (mc->nobjs >= min)
399 if (unlikely(!mc->objects)) {
400 if (WARN_ON_ONCE(!capacity))
403 mc->objects = kvmalloc_array(sizeof(void *), capacity, gfp);
407 mc->capacity = capacity;
410 /* It is illegal to request a different capacity across topups. */
411 if (WARN_ON_ONCE(mc->capacity != capacity))
414 while (mc->nobjs < mc->capacity) {
415 obj = mmu_memory_cache_alloc_obj(mc, gfp);
417 return mc->nobjs >= min ? 0 : -ENOMEM;
418 mc->objects[mc->nobjs++] = obj;
423 int kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache *mc, int min)
425 return __kvm_mmu_topup_memory_cache(mc, KVM_ARCH_NR_OBJS_PER_MEMORY_CACHE, min);
428 int kvm_mmu_memory_cache_nr_free_objects(struct kvm_mmu_memory_cache *mc)
433 void kvm_mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc)
437 kmem_cache_free(mc->kmem_cache, mc->objects[--mc->nobjs]);
439 free_page((unsigned long)mc->objects[--mc->nobjs]);
448 void *kvm_mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc)
452 if (WARN_ON(!mc->nobjs))
453 p = mmu_memory_cache_alloc_obj(mc, GFP_ATOMIC | __GFP_ACCOUNT);
455 p = mc->objects[--mc->nobjs];
461 static void kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
463 mutex_init(&vcpu->mutex);
468 #ifndef __KVM_HAVE_ARCH_WQP
469 rcuwait_init(&vcpu->wait);
471 kvm_async_pf_vcpu_init(vcpu);
473 kvm_vcpu_set_in_spin_loop(vcpu, false);
474 kvm_vcpu_set_dy_eligible(vcpu, false);
475 vcpu->preempted = false;
477 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
478 vcpu->last_used_slot = NULL;
480 /* Fill the stats id string for the vcpu */
481 snprintf(vcpu->stats_id, sizeof(vcpu->stats_id), "kvm-%d/vcpu-%d",
482 task_pid_nr(current), id);
485 static void kvm_vcpu_destroy(struct kvm_vcpu *vcpu)
487 kvm_arch_vcpu_destroy(vcpu);
488 kvm_dirty_ring_free(&vcpu->dirty_ring);
491 * No need for rcu_read_lock as VCPU_RUN is the only place that changes
492 * the vcpu->pid pointer, and at destruction time all file descriptors
495 put_pid(rcu_dereference_protected(vcpu->pid, 1));
497 free_page((unsigned long)vcpu->run);
498 kmem_cache_free(kvm_vcpu_cache, vcpu);
501 void kvm_destroy_vcpus(struct kvm *kvm)
504 struct kvm_vcpu *vcpu;
506 kvm_for_each_vcpu(i, vcpu, kvm) {
507 kvm_vcpu_destroy(vcpu);
508 xa_erase(&kvm->vcpu_array, i);
511 atomic_set(&kvm->online_vcpus, 0);
513 EXPORT_SYMBOL_GPL(kvm_destroy_vcpus);
515 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
516 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
518 return container_of(mn, struct kvm, mmu_notifier);
521 static void kvm_mmu_notifier_invalidate_range(struct mmu_notifier *mn,
522 struct mm_struct *mm,
523 unsigned long start, unsigned long end)
525 struct kvm *kvm = mmu_notifier_to_kvm(mn);
528 idx = srcu_read_lock(&kvm->srcu);
529 kvm_arch_mmu_notifier_invalidate_range(kvm, start, end);
530 srcu_read_unlock(&kvm->srcu, idx);
533 typedef bool (*hva_handler_t)(struct kvm *kvm, struct kvm_gfn_range *range);
535 typedef void (*on_lock_fn_t)(struct kvm *kvm, unsigned long start,
538 typedef void (*on_unlock_fn_t)(struct kvm *kvm);
540 struct kvm_hva_range {
544 hva_handler_t handler;
545 on_lock_fn_t on_lock;
546 on_unlock_fn_t on_unlock;
552 * Use a dedicated stub instead of NULL to indicate that there is no callback
553 * function/handler. The compiler technically can't guarantee that a real
554 * function will have a non-zero address, and so it will generate code to
555 * check for !NULL, whereas comparing against a stub will be elided at compile
556 * time (unless the compiler is getting long in the tooth, e.g. gcc 4.9).
558 static void kvm_null_fn(void)
562 #define IS_KVM_NULL_FN(fn) ((fn) == (void *)kvm_null_fn)
564 /* Iterate over each memslot intersecting [start, last] (inclusive) range */
565 #define kvm_for_each_memslot_in_hva_range(node, slots, start, last) \
566 for (node = interval_tree_iter_first(&slots->hva_tree, start, last); \
568 node = interval_tree_iter_next(node, start, last)) \
570 static __always_inline int __kvm_handle_hva_range(struct kvm *kvm,
571 const struct kvm_hva_range *range)
573 bool ret = false, locked = false;
574 struct kvm_gfn_range gfn_range;
575 struct kvm_memory_slot *slot;
576 struct kvm_memslots *slots;
579 if (WARN_ON_ONCE(range->end <= range->start))
582 /* A null handler is allowed if and only if on_lock() is provided. */
583 if (WARN_ON_ONCE(IS_KVM_NULL_FN(range->on_lock) &&
584 IS_KVM_NULL_FN(range->handler)))
587 idx = srcu_read_lock(&kvm->srcu);
589 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
590 struct interval_tree_node *node;
592 slots = __kvm_memslots(kvm, i);
593 kvm_for_each_memslot_in_hva_range(node, slots,
594 range->start, range->end - 1) {
595 unsigned long hva_start, hva_end;
597 slot = container_of(node, struct kvm_memory_slot, hva_node[slots->node_idx]);
598 hva_start = max(range->start, slot->userspace_addr);
599 hva_end = min(range->end, slot->userspace_addr +
600 (slot->npages << PAGE_SHIFT));
603 * To optimize for the likely case where the address
604 * range is covered by zero or one memslots, don't
605 * bother making these conditional (to avoid writes on
606 * the second or later invocation of the handler).
608 gfn_range.pte = range->pte;
609 gfn_range.may_block = range->may_block;
612 * {gfn(page) | page intersects with [hva_start, hva_end)} =
613 * {gfn_start, gfn_start+1, ..., gfn_end-1}.
615 gfn_range.start = hva_to_gfn_memslot(hva_start, slot);
616 gfn_range.end = hva_to_gfn_memslot(hva_end + PAGE_SIZE - 1, slot);
617 gfn_range.slot = slot;
622 if (!IS_KVM_NULL_FN(range->on_lock))
623 range->on_lock(kvm, range->start, range->end);
624 if (IS_KVM_NULL_FN(range->handler))
627 ret |= range->handler(kvm, &gfn_range);
631 if (range->flush_on_ret && ret)
632 kvm_flush_remote_tlbs(kvm);
636 if (!IS_KVM_NULL_FN(range->on_unlock))
637 range->on_unlock(kvm);
640 srcu_read_unlock(&kvm->srcu, idx);
642 /* The notifiers are averse to booleans. :-( */
646 static __always_inline int kvm_handle_hva_range(struct mmu_notifier *mn,
650 hva_handler_t handler)
652 struct kvm *kvm = mmu_notifier_to_kvm(mn);
653 const struct kvm_hva_range range = {
658 .on_lock = (void *)kvm_null_fn,
659 .on_unlock = (void *)kvm_null_fn,
660 .flush_on_ret = true,
664 return __kvm_handle_hva_range(kvm, &range);
667 static __always_inline int kvm_handle_hva_range_no_flush(struct mmu_notifier *mn,
670 hva_handler_t handler)
672 struct kvm *kvm = mmu_notifier_to_kvm(mn);
673 const struct kvm_hva_range range = {
678 .on_lock = (void *)kvm_null_fn,
679 .on_unlock = (void *)kvm_null_fn,
680 .flush_on_ret = false,
684 return __kvm_handle_hva_range(kvm, &range);
686 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
687 struct mm_struct *mm,
688 unsigned long address,
691 struct kvm *kvm = mmu_notifier_to_kvm(mn);
693 trace_kvm_set_spte_hva(address);
696 * .change_pte() must be surrounded by .invalidate_range_{start,end}().
697 * If mmu_invalidate_in_progress is zero, then no in-progress
698 * invalidations, including this one, found a relevant memslot at
699 * start(); rechecking memslots here is unnecessary. Note, a false
700 * positive (count elevated by a different invalidation) is sub-optimal
701 * but functionally ok.
703 WARN_ON_ONCE(!READ_ONCE(kvm->mn_active_invalidate_count));
704 if (!READ_ONCE(kvm->mmu_invalidate_in_progress))
707 kvm_handle_hva_range(mn, address, address + 1, pte, kvm_set_spte_gfn);
710 void kvm_mmu_invalidate_begin(struct kvm *kvm, unsigned long start,
714 * The count increase must become visible at unlock time as no
715 * spte can be established without taking the mmu_lock and
716 * count is also read inside the mmu_lock critical section.
718 kvm->mmu_invalidate_in_progress++;
719 if (likely(kvm->mmu_invalidate_in_progress == 1)) {
720 kvm->mmu_invalidate_range_start = start;
721 kvm->mmu_invalidate_range_end = end;
724 * Fully tracking multiple concurrent ranges has diminishing
725 * returns. Keep things simple and just find the minimal range
726 * which includes the current and new ranges. As there won't be
727 * enough information to subtract a range after its invalidate
728 * completes, any ranges invalidated concurrently will
729 * accumulate and persist until all outstanding invalidates
732 kvm->mmu_invalidate_range_start =
733 min(kvm->mmu_invalidate_range_start, start);
734 kvm->mmu_invalidate_range_end =
735 max(kvm->mmu_invalidate_range_end, end);
739 static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
740 const struct mmu_notifier_range *range)
742 struct kvm *kvm = mmu_notifier_to_kvm(mn);
743 const struct kvm_hva_range hva_range = {
744 .start = range->start,
747 .handler = kvm_unmap_gfn_range,
748 .on_lock = kvm_mmu_invalidate_begin,
749 .on_unlock = kvm_arch_guest_memory_reclaimed,
750 .flush_on_ret = true,
751 .may_block = mmu_notifier_range_blockable(range),
754 trace_kvm_unmap_hva_range(range->start, range->end);
757 * Prevent memslot modification between range_start() and range_end()
758 * so that conditionally locking provides the same result in both
759 * functions. Without that guarantee, the mmu_invalidate_in_progress
760 * adjustments will be imbalanced.
762 * Pairs with the decrement in range_end().
764 spin_lock(&kvm->mn_invalidate_lock);
765 kvm->mn_active_invalidate_count++;
766 spin_unlock(&kvm->mn_invalidate_lock);
769 * Invalidate pfn caches _before_ invalidating the secondary MMUs, i.e.
770 * before acquiring mmu_lock, to avoid holding mmu_lock while acquiring
771 * each cache's lock. There are relatively few caches in existence at
772 * any given time, and the caches themselves can check for hva overlap,
773 * i.e. don't need to rely on memslot overlap checks for performance.
774 * Because this runs without holding mmu_lock, the pfn caches must use
775 * mn_active_invalidate_count (see above) instead of
776 * mmu_invalidate_in_progress.
778 gfn_to_pfn_cache_invalidate_start(kvm, range->start, range->end,
779 hva_range.may_block);
781 __kvm_handle_hva_range(kvm, &hva_range);
786 void kvm_mmu_invalidate_end(struct kvm *kvm, unsigned long start,
790 * This sequence increase will notify the kvm page fault that
791 * the page that is going to be mapped in the spte could have
794 kvm->mmu_invalidate_seq++;
797 * The above sequence increase must be visible before the
798 * below count decrease, which is ensured by the smp_wmb above
799 * in conjunction with the smp_rmb in mmu_invalidate_retry().
801 kvm->mmu_invalidate_in_progress--;
804 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
805 const struct mmu_notifier_range *range)
807 struct kvm *kvm = mmu_notifier_to_kvm(mn);
808 const struct kvm_hva_range hva_range = {
809 .start = range->start,
812 .handler = (void *)kvm_null_fn,
813 .on_lock = kvm_mmu_invalidate_end,
814 .on_unlock = (void *)kvm_null_fn,
815 .flush_on_ret = false,
816 .may_block = mmu_notifier_range_blockable(range),
820 __kvm_handle_hva_range(kvm, &hva_range);
822 /* Pairs with the increment in range_start(). */
823 spin_lock(&kvm->mn_invalidate_lock);
824 wake = (--kvm->mn_active_invalidate_count == 0);
825 spin_unlock(&kvm->mn_invalidate_lock);
828 * There can only be one waiter, since the wait happens under
832 rcuwait_wake_up(&kvm->mn_memslots_update_rcuwait);
834 BUG_ON(kvm->mmu_invalidate_in_progress < 0);
837 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
838 struct mm_struct *mm,
842 trace_kvm_age_hva(start, end);
844 return kvm_handle_hva_range(mn, start, end, __pte(0), kvm_age_gfn);
847 static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
848 struct mm_struct *mm,
852 trace_kvm_age_hva(start, end);
855 * Even though we do not flush TLB, this will still adversely
856 * affect performance on pre-Haswell Intel EPT, where there is
857 * no EPT Access Bit to clear so that we have to tear down EPT
858 * tables instead. If we find this unacceptable, we can always
859 * add a parameter to kvm_age_hva so that it effectively doesn't
860 * do anything on clear_young.
862 * Also note that currently we never issue secondary TLB flushes
863 * from clear_young, leaving this job up to the regular system
864 * cadence. If we find this inaccurate, we might come up with a
865 * more sophisticated heuristic later.
867 return kvm_handle_hva_range_no_flush(mn, start, end, kvm_age_gfn);
870 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
871 struct mm_struct *mm,
872 unsigned long address)
874 trace_kvm_test_age_hva(address);
876 return kvm_handle_hva_range_no_flush(mn, address, address + 1,
880 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
881 struct mm_struct *mm)
883 struct kvm *kvm = mmu_notifier_to_kvm(mn);
886 idx = srcu_read_lock(&kvm->srcu);
887 kvm_flush_shadow_all(kvm);
888 srcu_read_unlock(&kvm->srcu, idx);
891 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
892 .invalidate_range = kvm_mmu_notifier_invalidate_range,
893 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
894 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
895 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
896 .clear_young = kvm_mmu_notifier_clear_young,
897 .test_young = kvm_mmu_notifier_test_young,
898 .change_pte = kvm_mmu_notifier_change_pte,
899 .release = kvm_mmu_notifier_release,
902 static int kvm_init_mmu_notifier(struct kvm *kvm)
904 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
905 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
908 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
910 static int kvm_init_mmu_notifier(struct kvm *kvm)
915 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
917 #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER
918 static int kvm_pm_notifier_call(struct notifier_block *bl,
922 struct kvm *kvm = container_of(bl, struct kvm, pm_notifier);
924 return kvm_arch_pm_notifier(kvm, state);
927 static void kvm_init_pm_notifier(struct kvm *kvm)
929 kvm->pm_notifier.notifier_call = kvm_pm_notifier_call;
930 /* Suspend KVM before we suspend ftrace, RCU, etc. */
931 kvm->pm_notifier.priority = INT_MAX;
932 register_pm_notifier(&kvm->pm_notifier);
935 static void kvm_destroy_pm_notifier(struct kvm *kvm)
937 unregister_pm_notifier(&kvm->pm_notifier);
939 #else /* !CONFIG_HAVE_KVM_PM_NOTIFIER */
940 static void kvm_init_pm_notifier(struct kvm *kvm)
944 static void kvm_destroy_pm_notifier(struct kvm *kvm)
947 #endif /* CONFIG_HAVE_KVM_PM_NOTIFIER */
949 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
951 if (!memslot->dirty_bitmap)
954 kvfree(memslot->dirty_bitmap);
955 memslot->dirty_bitmap = NULL;
958 /* This does not remove the slot from struct kvm_memslots data structures */
959 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
961 kvm_destroy_dirty_bitmap(slot);
963 kvm_arch_free_memslot(kvm, slot);
968 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
970 struct hlist_node *idnode;
971 struct kvm_memory_slot *memslot;
975 * The same memslot objects live in both active and inactive sets,
976 * arbitrarily free using index '1' so the second invocation of this
977 * function isn't operating over a structure with dangling pointers
978 * (even though this function isn't actually touching them).
980 if (!slots->node_idx)
983 hash_for_each_safe(slots->id_hash, bkt, idnode, memslot, id_node[1])
984 kvm_free_memslot(kvm, memslot);
987 static umode_t kvm_stats_debugfs_mode(const struct _kvm_stats_desc *pdesc)
989 switch (pdesc->desc.flags & KVM_STATS_TYPE_MASK) {
990 case KVM_STATS_TYPE_INSTANT:
992 case KVM_STATS_TYPE_CUMULATIVE:
993 case KVM_STATS_TYPE_PEAK:
1000 static void kvm_destroy_vm_debugfs(struct kvm *kvm)
1003 int kvm_debugfs_num_entries = kvm_vm_stats_header.num_desc +
1004 kvm_vcpu_stats_header.num_desc;
1006 if (IS_ERR(kvm->debugfs_dentry))
1009 debugfs_remove_recursive(kvm->debugfs_dentry);
1011 if (kvm->debugfs_stat_data) {
1012 for (i = 0; i < kvm_debugfs_num_entries; i++)
1013 kfree(kvm->debugfs_stat_data[i]);
1014 kfree(kvm->debugfs_stat_data);
1018 static int kvm_create_vm_debugfs(struct kvm *kvm, const char *fdname)
1020 static DEFINE_MUTEX(kvm_debugfs_lock);
1021 struct dentry *dent;
1022 char dir_name[ITOA_MAX_LEN * 2];
1023 struct kvm_stat_data *stat_data;
1024 const struct _kvm_stats_desc *pdesc;
1025 int i, ret = -ENOMEM;
1026 int kvm_debugfs_num_entries = kvm_vm_stats_header.num_desc +
1027 kvm_vcpu_stats_header.num_desc;
1029 if (!debugfs_initialized())
1032 snprintf(dir_name, sizeof(dir_name), "%d-%s", task_pid_nr(current), fdname);
1033 mutex_lock(&kvm_debugfs_lock);
1034 dent = debugfs_lookup(dir_name, kvm_debugfs_dir);
1036 pr_warn_ratelimited("KVM: debugfs: duplicate directory %s\n", dir_name);
1038 mutex_unlock(&kvm_debugfs_lock);
1041 dent = debugfs_create_dir(dir_name, kvm_debugfs_dir);
1042 mutex_unlock(&kvm_debugfs_lock);
1046 kvm->debugfs_dentry = dent;
1047 kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries,
1048 sizeof(*kvm->debugfs_stat_data),
1049 GFP_KERNEL_ACCOUNT);
1050 if (!kvm->debugfs_stat_data)
1053 for (i = 0; i < kvm_vm_stats_header.num_desc; ++i) {
1054 pdesc = &kvm_vm_stats_desc[i];
1055 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT);
1059 stat_data->kvm = kvm;
1060 stat_data->desc = pdesc;
1061 stat_data->kind = KVM_STAT_VM;
1062 kvm->debugfs_stat_data[i] = stat_data;
1063 debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
1064 kvm->debugfs_dentry, stat_data,
1068 for (i = 0; i < kvm_vcpu_stats_header.num_desc; ++i) {
1069 pdesc = &kvm_vcpu_stats_desc[i];
1070 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT);
1074 stat_data->kvm = kvm;
1075 stat_data->desc = pdesc;
1076 stat_data->kind = KVM_STAT_VCPU;
1077 kvm->debugfs_stat_data[i + kvm_vm_stats_header.num_desc] = stat_data;
1078 debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
1079 kvm->debugfs_dentry, stat_data,
1083 ret = kvm_arch_create_vm_debugfs(kvm);
1089 kvm_destroy_vm_debugfs(kvm);
1094 * Called after the VM is otherwise initialized, but just before adding it to
1097 int __weak kvm_arch_post_init_vm(struct kvm *kvm)
1103 * Called just after removing the VM from the vm_list, but before doing any
1104 * other destruction.
1106 void __weak kvm_arch_pre_destroy_vm(struct kvm *kvm)
1111 * Called after per-vm debugfs created. When called kvm->debugfs_dentry should
1112 * be setup already, so we can create arch-specific debugfs entries under it.
1113 * Cleanup should be automatic done in kvm_destroy_vm_debugfs() recursively, so
1114 * a per-arch destroy interface is not needed.
1116 int __weak kvm_arch_create_vm_debugfs(struct kvm *kvm)
1121 static struct kvm *kvm_create_vm(unsigned long type, const char *fdname)
1123 struct kvm *kvm = kvm_arch_alloc_vm();
1124 struct kvm_memslots *slots;
1129 return ERR_PTR(-ENOMEM);
1131 /* KVM is pinned via open("/dev/kvm"), the fd passed to this ioctl(). */
1132 __module_get(kvm_chardev_ops.owner);
1134 KVM_MMU_LOCK_INIT(kvm);
1135 mmgrab(current->mm);
1136 kvm->mm = current->mm;
1137 kvm_eventfd_init(kvm);
1138 mutex_init(&kvm->lock);
1139 mutex_init(&kvm->irq_lock);
1140 mutex_init(&kvm->slots_lock);
1141 mutex_init(&kvm->slots_arch_lock);
1142 spin_lock_init(&kvm->mn_invalidate_lock);
1143 rcuwait_init(&kvm->mn_memslots_update_rcuwait);
1144 xa_init(&kvm->vcpu_array);
1146 INIT_LIST_HEAD(&kvm->gpc_list);
1147 spin_lock_init(&kvm->gpc_lock);
1149 INIT_LIST_HEAD(&kvm->devices);
1150 kvm->max_vcpus = KVM_MAX_VCPUS;
1152 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
1155 * Force subsequent debugfs file creations to fail if the VM directory
1156 * is not created (by kvm_create_vm_debugfs()).
1158 kvm->debugfs_dentry = ERR_PTR(-ENOENT);
1160 snprintf(kvm->stats_id, sizeof(kvm->stats_id), "kvm-%d",
1161 task_pid_nr(current));
1163 if (init_srcu_struct(&kvm->srcu))
1164 goto out_err_no_srcu;
1165 if (init_srcu_struct(&kvm->irq_srcu))
1166 goto out_err_no_irq_srcu;
1168 refcount_set(&kvm->users_count, 1);
1169 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
1170 for (j = 0; j < 2; j++) {
1171 slots = &kvm->__memslots[i][j];
1173 atomic_long_set(&slots->last_used_slot, (unsigned long)NULL);
1174 slots->hva_tree = RB_ROOT_CACHED;
1175 slots->gfn_tree = RB_ROOT;
1176 hash_init(slots->id_hash);
1177 slots->node_idx = j;
1179 /* Generations must be different for each address space. */
1180 slots->generation = i;
1183 rcu_assign_pointer(kvm->memslots[i], &kvm->__memslots[i][0]);
1186 for (i = 0; i < KVM_NR_BUSES; i++) {
1187 rcu_assign_pointer(kvm->buses[i],
1188 kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL_ACCOUNT));
1190 goto out_err_no_arch_destroy_vm;
1193 r = kvm_arch_init_vm(kvm, type);
1195 goto out_err_no_arch_destroy_vm;
1197 r = hardware_enable_all();
1199 goto out_err_no_disable;
1201 #ifdef CONFIG_HAVE_KVM_IRQFD
1202 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
1205 r = kvm_init_mmu_notifier(kvm);
1207 goto out_err_no_mmu_notifier;
1209 r = kvm_coalesced_mmio_init(kvm);
1211 goto out_no_coalesced_mmio;
1213 r = kvm_create_vm_debugfs(kvm, fdname);
1215 goto out_err_no_debugfs;
1217 r = kvm_arch_post_init_vm(kvm);
1221 mutex_lock(&kvm_lock);
1222 list_add(&kvm->vm_list, &vm_list);
1223 mutex_unlock(&kvm_lock);
1225 preempt_notifier_inc();
1226 kvm_init_pm_notifier(kvm);
1231 kvm_destroy_vm_debugfs(kvm);
1233 kvm_coalesced_mmio_free(kvm);
1234 out_no_coalesced_mmio:
1235 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1236 if (kvm->mmu_notifier.ops)
1237 mmu_notifier_unregister(&kvm->mmu_notifier, current->mm);
1239 out_err_no_mmu_notifier:
1240 hardware_disable_all();
1242 kvm_arch_destroy_vm(kvm);
1243 out_err_no_arch_destroy_vm:
1244 WARN_ON_ONCE(!refcount_dec_and_test(&kvm->users_count));
1245 for (i = 0; i < KVM_NR_BUSES; i++)
1246 kfree(kvm_get_bus(kvm, i));
1247 cleanup_srcu_struct(&kvm->irq_srcu);
1248 out_err_no_irq_srcu:
1249 cleanup_srcu_struct(&kvm->srcu);
1251 kvm_arch_free_vm(kvm);
1252 mmdrop(current->mm);
1253 module_put(kvm_chardev_ops.owner);
1257 static void kvm_destroy_devices(struct kvm *kvm)
1259 struct kvm_device *dev, *tmp;
1262 * We do not need to take the kvm->lock here, because nobody else
1263 * has a reference to the struct kvm at this point and therefore
1264 * cannot access the devices list anyhow.
1266 list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
1267 list_del(&dev->vm_node);
1268 dev->ops->destroy(dev);
1272 static void kvm_destroy_vm(struct kvm *kvm)
1275 struct mm_struct *mm = kvm->mm;
1277 kvm_destroy_pm_notifier(kvm);
1278 kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm);
1279 kvm_destroy_vm_debugfs(kvm);
1280 kvm_arch_sync_events(kvm);
1281 mutex_lock(&kvm_lock);
1282 list_del(&kvm->vm_list);
1283 mutex_unlock(&kvm_lock);
1284 kvm_arch_pre_destroy_vm(kvm);
1286 kvm_free_irq_routing(kvm);
1287 for (i = 0; i < KVM_NR_BUSES; i++) {
1288 struct kvm_io_bus *bus = kvm_get_bus(kvm, i);
1291 kvm_io_bus_destroy(bus);
1292 kvm->buses[i] = NULL;
1294 kvm_coalesced_mmio_free(kvm);
1295 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1296 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
1298 * At this point, pending calls to invalidate_range_start()
1299 * have completed but no more MMU notifiers will run, so
1300 * mn_active_invalidate_count may remain unbalanced.
1301 * No threads can be waiting in install_new_memslots as the
1302 * last reference on KVM has been dropped, but freeing
1303 * memslots would deadlock without this manual intervention.
1305 WARN_ON(rcuwait_active(&kvm->mn_memslots_update_rcuwait));
1306 kvm->mn_active_invalidate_count = 0;
1308 kvm_flush_shadow_all(kvm);
1310 kvm_arch_destroy_vm(kvm);
1311 kvm_destroy_devices(kvm);
1312 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
1313 kvm_free_memslots(kvm, &kvm->__memslots[i][0]);
1314 kvm_free_memslots(kvm, &kvm->__memslots[i][1]);
1316 cleanup_srcu_struct(&kvm->irq_srcu);
1317 cleanup_srcu_struct(&kvm->srcu);
1318 kvm_arch_free_vm(kvm);
1319 preempt_notifier_dec();
1320 hardware_disable_all();
1322 module_put(kvm_chardev_ops.owner);
1325 void kvm_get_kvm(struct kvm *kvm)
1327 refcount_inc(&kvm->users_count);
1329 EXPORT_SYMBOL_GPL(kvm_get_kvm);
1332 * Make sure the vm is not during destruction, which is a safe version of
1333 * kvm_get_kvm(). Return true if kvm referenced successfully, false otherwise.
1335 bool kvm_get_kvm_safe(struct kvm *kvm)
1337 return refcount_inc_not_zero(&kvm->users_count);
1339 EXPORT_SYMBOL_GPL(kvm_get_kvm_safe);
1341 void kvm_put_kvm(struct kvm *kvm)
1343 if (refcount_dec_and_test(&kvm->users_count))
1344 kvm_destroy_vm(kvm);
1346 EXPORT_SYMBOL_GPL(kvm_put_kvm);
1349 * Used to put a reference that was taken on behalf of an object associated
1350 * with a user-visible file descriptor, e.g. a vcpu or device, if installation
1351 * of the new file descriptor fails and the reference cannot be transferred to
1352 * its final owner. In such cases, the caller is still actively using @kvm and
1353 * will fail miserably if the refcount unexpectedly hits zero.
1355 void kvm_put_kvm_no_destroy(struct kvm *kvm)
1357 WARN_ON(refcount_dec_and_test(&kvm->users_count));
1359 EXPORT_SYMBOL_GPL(kvm_put_kvm_no_destroy);
1361 static int kvm_vm_release(struct inode *inode, struct file *filp)
1363 struct kvm *kvm = filp->private_data;
1365 kvm_irqfd_release(kvm);
1372 * Allocation size is twice as large as the actual dirty bitmap size.
1373 * See kvm_vm_ioctl_get_dirty_log() why this is needed.
1375 static int kvm_alloc_dirty_bitmap(struct kvm_memory_slot *memslot)
1377 unsigned long dirty_bytes = kvm_dirty_bitmap_bytes(memslot);
1379 memslot->dirty_bitmap = __vcalloc(2, dirty_bytes, GFP_KERNEL_ACCOUNT);
1380 if (!memslot->dirty_bitmap)
1386 static struct kvm_memslots *kvm_get_inactive_memslots(struct kvm *kvm, int as_id)
1388 struct kvm_memslots *active = __kvm_memslots(kvm, as_id);
1389 int node_idx_inactive = active->node_idx ^ 1;
1391 return &kvm->__memslots[as_id][node_idx_inactive];
1395 * Helper to get the address space ID when one of memslot pointers may be NULL.
1396 * This also serves as a sanity that at least one of the pointers is non-NULL,
1397 * and that their address space IDs don't diverge.
1399 static int kvm_memslots_get_as_id(struct kvm_memory_slot *a,
1400 struct kvm_memory_slot *b)
1402 if (WARN_ON_ONCE(!a && !b))
1410 WARN_ON_ONCE(a->as_id != b->as_id);
1414 static void kvm_insert_gfn_node(struct kvm_memslots *slots,
1415 struct kvm_memory_slot *slot)
1417 struct rb_root *gfn_tree = &slots->gfn_tree;
1418 struct rb_node **node, *parent;
1419 int idx = slots->node_idx;
1422 for (node = &gfn_tree->rb_node; *node; ) {
1423 struct kvm_memory_slot *tmp;
1425 tmp = container_of(*node, struct kvm_memory_slot, gfn_node[idx]);
1427 if (slot->base_gfn < tmp->base_gfn)
1428 node = &(*node)->rb_left;
1429 else if (slot->base_gfn > tmp->base_gfn)
1430 node = &(*node)->rb_right;
1435 rb_link_node(&slot->gfn_node[idx], parent, node);
1436 rb_insert_color(&slot->gfn_node[idx], gfn_tree);
1439 static void kvm_erase_gfn_node(struct kvm_memslots *slots,
1440 struct kvm_memory_slot *slot)
1442 rb_erase(&slot->gfn_node[slots->node_idx], &slots->gfn_tree);
1445 static void kvm_replace_gfn_node(struct kvm_memslots *slots,
1446 struct kvm_memory_slot *old,
1447 struct kvm_memory_slot *new)
1449 int idx = slots->node_idx;
1451 WARN_ON_ONCE(old->base_gfn != new->base_gfn);
1453 rb_replace_node(&old->gfn_node[idx], &new->gfn_node[idx],
1458 * Replace @old with @new in the inactive memslots.
1460 * With NULL @old this simply adds @new.
1461 * With NULL @new this simply removes @old.
1463 * If @new is non-NULL its hva_node[slots_idx] range has to be set
1466 static void kvm_replace_memslot(struct kvm *kvm,
1467 struct kvm_memory_slot *old,
1468 struct kvm_memory_slot *new)
1470 int as_id = kvm_memslots_get_as_id(old, new);
1471 struct kvm_memslots *slots = kvm_get_inactive_memslots(kvm, as_id);
1472 int idx = slots->node_idx;
1475 hash_del(&old->id_node[idx]);
1476 interval_tree_remove(&old->hva_node[idx], &slots->hva_tree);
1478 if ((long)old == atomic_long_read(&slots->last_used_slot))
1479 atomic_long_set(&slots->last_used_slot, (long)new);
1482 kvm_erase_gfn_node(slots, old);
1488 * Initialize @new's hva range. Do this even when replacing an @old
1489 * slot, kvm_copy_memslot() deliberately does not touch node data.
1491 new->hva_node[idx].start = new->userspace_addr;
1492 new->hva_node[idx].last = new->userspace_addr +
1493 (new->npages << PAGE_SHIFT) - 1;
1496 * (Re)Add the new memslot. There is no O(1) interval_tree_replace(),
1497 * hva_node needs to be swapped with remove+insert even though hva can't
1498 * change when replacing an existing slot.
1500 hash_add(slots->id_hash, &new->id_node[idx], new->id);
1501 interval_tree_insert(&new->hva_node[idx], &slots->hva_tree);
1504 * If the memslot gfn is unchanged, rb_replace_node() can be used to
1505 * switch the node in the gfn tree instead of removing the old and
1506 * inserting the new as two separate operations. Replacement is a
1507 * single O(1) operation versus two O(log(n)) operations for
1510 if (old && old->base_gfn == new->base_gfn) {
1511 kvm_replace_gfn_node(slots, old, new);
1514 kvm_erase_gfn_node(slots, old);
1515 kvm_insert_gfn_node(slots, new);
1519 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
1521 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
1523 #ifdef __KVM_HAVE_READONLY_MEM
1524 valid_flags |= KVM_MEM_READONLY;
1527 if (mem->flags & ~valid_flags)
1533 static void kvm_swap_active_memslots(struct kvm *kvm, int as_id)
1535 struct kvm_memslots *slots = kvm_get_inactive_memslots(kvm, as_id);
1537 /* Grab the generation from the activate memslots. */
1538 u64 gen = __kvm_memslots(kvm, as_id)->generation;
1540 WARN_ON(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS);
1541 slots->generation = gen | KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
1544 * Do not store the new memslots while there are invalidations in
1545 * progress, otherwise the locking in invalidate_range_start and
1546 * invalidate_range_end will be unbalanced.
1548 spin_lock(&kvm->mn_invalidate_lock);
1549 prepare_to_rcuwait(&kvm->mn_memslots_update_rcuwait);
1550 while (kvm->mn_active_invalidate_count) {
1551 set_current_state(TASK_UNINTERRUPTIBLE);
1552 spin_unlock(&kvm->mn_invalidate_lock);
1554 spin_lock(&kvm->mn_invalidate_lock);
1556 finish_rcuwait(&kvm->mn_memslots_update_rcuwait);
1557 rcu_assign_pointer(kvm->memslots[as_id], slots);
1558 spin_unlock(&kvm->mn_invalidate_lock);
1561 * Acquired in kvm_set_memslot. Must be released before synchronize
1562 * SRCU below in order to avoid deadlock with another thread
1563 * acquiring the slots_arch_lock in an srcu critical section.
1565 mutex_unlock(&kvm->slots_arch_lock);
1567 synchronize_srcu_expedited(&kvm->srcu);
1570 * Increment the new memslot generation a second time, dropping the
1571 * update in-progress flag and incrementing the generation based on
1572 * the number of address spaces. This provides a unique and easily
1573 * identifiable generation number while the memslots are in flux.
1575 gen = slots->generation & ~KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
1578 * Generations must be unique even across address spaces. We do not need
1579 * a global counter for that, instead the generation space is evenly split
1580 * across address spaces. For example, with two address spaces, address
1581 * space 0 will use generations 0, 2, 4, ... while address space 1 will
1582 * use generations 1, 3, 5, ...
1584 gen += KVM_ADDRESS_SPACE_NUM;
1586 kvm_arch_memslots_updated(kvm, gen);
1588 slots->generation = gen;
1591 static int kvm_prepare_memory_region(struct kvm *kvm,
1592 const struct kvm_memory_slot *old,
1593 struct kvm_memory_slot *new,
1594 enum kvm_mr_change change)
1599 * If dirty logging is disabled, nullify the bitmap; the old bitmap
1600 * will be freed on "commit". If logging is enabled in both old and
1601 * new, reuse the existing bitmap. If logging is enabled only in the
1602 * new and KVM isn't using a ring buffer, allocate and initialize a
1605 if (change != KVM_MR_DELETE) {
1606 if (!(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
1607 new->dirty_bitmap = NULL;
1608 else if (old && old->dirty_bitmap)
1609 new->dirty_bitmap = old->dirty_bitmap;
1610 else if (kvm_use_dirty_bitmap(kvm)) {
1611 r = kvm_alloc_dirty_bitmap(new);
1615 if (kvm_dirty_log_manual_protect_and_init_set(kvm))
1616 bitmap_set(new->dirty_bitmap, 0, new->npages);
1620 r = kvm_arch_prepare_memory_region(kvm, old, new, change);
1622 /* Free the bitmap on failure if it was allocated above. */
1623 if (r && new && new->dirty_bitmap && (!old || !old->dirty_bitmap))
1624 kvm_destroy_dirty_bitmap(new);
1629 static void kvm_commit_memory_region(struct kvm *kvm,
1630 struct kvm_memory_slot *old,
1631 const struct kvm_memory_slot *new,
1632 enum kvm_mr_change change)
1634 int old_flags = old ? old->flags : 0;
1635 int new_flags = new ? new->flags : 0;
1637 * Update the total number of memslot pages before calling the arch
1638 * hook so that architectures can consume the result directly.
1640 if (change == KVM_MR_DELETE)
1641 kvm->nr_memslot_pages -= old->npages;
1642 else if (change == KVM_MR_CREATE)
1643 kvm->nr_memslot_pages += new->npages;
1645 if ((old_flags ^ new_flags) & KVM_MEM_LOG_DIRTY_PAGES) {
1646 int change = (new_flags & KVM_MEM_LOG_DIRTY_PAGES) ? 1 : -1;
1647 atomic_set(&kvm->nr_memslots_dirty_logging,
1648 atomic_read(&kvm->nr_memslots_dirty_logging) + change);
1651 kvm_arch_commit_memory_region(kvm, old, new, change);
1655 /* Nothing more to do. */
1658 /* Free the old memslot and all its metadata. */
1659 kvm_free_memslot(kvm, old);
1662 case KVM_MR_FLAGS_ONLY:
1664 * Free the dirty bitmap as needed; the below check encompasses
1665 * both the flags and whether a ring buffer is being used)
1667 if (old->dirty_bitmap && !new->dirty_bitmap)
1668 kvm_destroy_dirty_bitmap(old);
1671 * The final quirk. Free the detached, old slot, but only its
1672 * memory, not any metadata. Metadata, including arch specific
1673 * data, may be reused by @new.
1683 * Activate @new, which must be installed in the inactive slots by the caller,
1684 * by swapping the active slots and then propagating @new to @old once @old is
1685 * unreachable and can be safely modified.
1687 * With NULL @old this simply adds @new to @active (while swapping the sets).
1688 * With NULL @new this simply removes @old from @active and frees it
1689 * (while also swapping the sets).
1691 static void kvm_activate_memslot(struct kvm *kvm,
1692 struct kvm_memory_slot *old,
1693 struct kvm_memory_slot *new)
1695 int as_id = kvm_memslots_get_as_id(old, new);
1697 kvm_swap_active_memslots(kvm, as_id);
1699 /* Propagate the new memslot to the now inactive memslots. */
1700 kvm_replace_memslot(kvm, old, new);
1703 static void kvm_copy_memslot(struct kvm_memory_slot *dest,
1704 const struct kvm_memory_slot *src)
1706 dest->base_gfn = src->base_gfn;
1707 dest->npages = src->npages;
1708 dest->dirty_bitmap = src->dirty_bitmap;
1709 dest->arch = src->arch;
1710 dest->userspace_addr = src->userspace_addr;
1711 dest->flags = src->flags;
1713 dest->as_id = src->as_id;
1716 static void kvm_invalidate_memslot(struct kvm *kvm,
1717 struct kvm_memory_slot *old,
1718 struct kvm_memory_slot *invalid_slot)
1721 * Mark the current slot INVALID. As with all memslot modifications,
1722 * this must be done on an unreachable slot to avoid modifying the
1723 * current slot in the active tree.
1725 kvm_copy_memslot(invalid_slot, old);
1726 invalid_slot->flags |= KVM_MEMSLOT_INVALID;
1727 kvm_replace_memslot(kvm, old, invalid_slot);
1730 * Activate the slot that is now marked INVALID, but don't propagate
1731 * the slot to the now inactive slots. The slot is either going to be
1732 * deleted or recreated as a new slot.
1734 kvm_swap_active_memslots(kvm, old->as_id);
1737 * From this point no new shadow pages pointing to a deleted, or moved,
1738 * memslot will be created. Validation of sp->gfn happens in:
1739 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1740 * - kvm_is_visible_gfn (mmu_check_root)
1742 kvm_arch_flush_shadow_memslot(kvm, old);
1743 kvm_arch_guest_memory_reclaimed(kvm);
1745 /* Was released by kvm_swap_active_memslots, reacquire. */
1746 mutex_lock(&kvm->slots_arch_lock);
1749 * Copy the arch-specific field of the newly-installed slot back to the
1750 * old slot as the arch data could have changed between releasing
1751 * slots_arch_lock in install_new_memslots() and re-acquiring the lock
1752 * above. Writers are required to retrieve memslots *after* acquiring
1753 * slots_arch_lock, thus the active slot's data is guaranteed to be fresh.
1755 old->arch = invalid_slot->arch;
1758 static void kvm_create_memslot(struct kvm *kvm,
1759 struct kvm_memory_slot *new)
1761 /* Add the new memslot to the inactive set and activate. */
1762 kvm_replace_memslot(kvm, NULL, new);
1763 kvm_activate_memslot(kvm, NULL, new);
1766 static void kvm_delete_memslot(struct kvm *kvm,
1767 struct kvm_memory_slot *old,
1768 struct kvm_memory_slot *invalid_slot)
1771 * Remove the old memslot (in the inactive memslots) by passing NULL as
1772 * the "new" slot, and for the invalid version in the active slots.
1774 kvm_replace_memslot(kvm, old, NULL);
1775 kvm_activate_memslot(kvm, invalid_slot, NULL);
1778 static void kvm_move_memslot(struct kvm *kvm,
1779 struct kvm_memory_slot *old,
1780 struct kvm_memory_slot *new,
1781 struct kvm_memory_slot *invalid_slot)
1784 * Replace the old memslot in the inactive slots, and then swap slots
1785 * and replace the current INVALID with the new as well.
1787 kvm_replace_memslot(kvm, old, new);
1788 kvm_activate_memslot(kvm, invalid_slot, new);
1791 static void kvm_update_flags_memslot(struct kvm *kvm,
1792 struct kvm_memory_slot *old,
1793 struct kvm_memory_slot *new)
1796 * Similar to the MOVE case, but the slot doesn't need to be zapped as
1797 * an intermediate step. Instead, the old memslot is simply replaced
1798 * with a new, updated copy in both memslot sets.
1800 kvm_replace_memslot(kvm, old, new);
1801 kvm_activate_memslot(kvm, old, new);
1804 static int kvm_set_memslot(struct kvm *kvm,
1805 struct kvm_memory_slot *old,
1806 struct kvm_memory_slot *new,
1807 enum kvm_mr_change change)
1809 struct kvm_memory_slot *invalid_slot;
1813 * Released in kvm_swap_active_memslots.
1815 * Must be held from before the current memslots are copied until
1816 * after the new memslots are installed with rcu_assign_pointer,
1817 * then released before the synchronize srcu in kvm_swap_active_memslots.
1819 * When modifying memslots outside of the slots_lock, must be held
1820 * before reading the pointer to the current memslots until after all
1821 * changes to those memslots are complete.
1823 * These rules ensure that installing new memslots does not lose
1824 * changes made to the previous memslots.
1826 mutex_lock(&kvm->slots_arch_lock);
1829 * Invalidate the old slot if it's being deleted or moved. This is
1830 * done prior to actually deleting/moving the memslot to allow vCPUs to
1831 * continue running by ensuring there are no mappings or shadow pages
1832 * for the memslot when it is deleted/moved. Without pre-invalidation
1833 * (and without a lock), a window would exist between effecting the
1834 * delete/move and committing the changes in arch code where KVM or a
1835 * guest could access a non-existent memslot.
1837 * Modifications are done on a temporary, unreachable slot. The old
1838 * slot needs to be preserved in case a later step fails and the
1839 * invalidation needs to be reverted.
1841 if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) {
1842 invalid_slot = kzalloc(sizeof(*invalid_slot), GFP_KERNEL_ACCOUNT);
1843 if (!invalid_slot) {
1844 mutex_unlock(&kvm->slots_arch_lock);
1847 kvm_invalidate_memslot(kvm, old, invalid_slot);
1850 r = kvm_prepare_memory_region(kvm, old, new, change);
1853 * For DELETE/MOVE, revert the above INVALID change. No
1854 * modifications required since the original slot was preserved
1855 * in the inactive slots. Changing the active memslots also
1856 * release slots_arch_lock.
1858 if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) {
1859 kvm_activate_memslot(kvm, invalid_slot, old);
1860 kfree(invalid_slot);
1862 mutex_unlock(&kvm->slots_arch_lock);
1868 * For DELETE and MOVE, the working slot is now active as the INVALID
1869 * version of the old slot. MOVE is particularly special as it reuses
1870 * the old slot and returns a copy of the old slot (in working_slot).
1871 * For CREATE, there is no old slot. For DELETE and FLAGS_ONLY, the
1872 * old slot is detached but otherwise preserved.
1874 if (change == KVM_MR_CREATE)
1875 kvm_create_memslot(kvm, new);
1876 else if (change == KVM_MR_DELETE)
1877 kvm_delete_memslot(kvm, old, invalid_slot);
1878 else if (change == KVM_MR_MOVE)
1879 kvm_move_memslot(kvm, old, new, invalid_slot);
1880 else if (change == KVM_MR_FLAGS_ONLY)
1881 kvm_update_flags_memslot(kvm, old, new);
1885 /* Free the temporary INVALID slot used for DELETE and MOVE. */
1886 if (change == KVM_MR_DELETE || change == KVM_MR_MOVE)
1887 kfree(invalid_slot);
1890 * No need to refresh new->arch, changes after dropping slots_arch_lock
1891 * will directly hit the final, active memslot. Architectures are
1892 * responsible for knowing that new->arch may be stale.
1894 kvm_commit_memory_region(kvm, old, new, change);
1899 static bool kvm_check_memslot_overlap(struct kvm_memslots *slots, int id,
1900 gfn_t start, gfn_t end)
1902 struct kvm_memslot_iter iter;
1904 kvm_for_each_memslot_in_gfn_range(&iter, slots, start, end) {
1905 if (iter.slot->id != id)
1913 * Allocate some memory and give it an address in the guest physical address
1916 * Discontiguous memory is allowed, mostly for framebuffers.
1918 * Must be called holding kvm->slots_lock for write.
1920 int __kvm_set_memory_region(struct kvm *kvm,
1921 const struct kvm_userspace_memory_region *mem)
1923 struct kvm_memory_slot *old, *new;
1924 struct kvm_memslots *slots;
1925 enum kvm_mr_change change;
1926 unsigned long npages;
1931 r = check_memory_region_flags(mem);
1935 as_id = mem->slot >> 16;
1936 id = (u16)mem->slot;
1938 /* General sanity checks */
1939 if ((mem->memory_size & (PAGE_SIZE - 1)) ||
1940 (mem->memory_size != (unsigned long)mem->memory_size))
1942 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
1944 /* We can read the guest memory with __xxx_user() later on. */
1945 if ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
1946 (mem->userspace_addr != untagged_addr(mem->userspace_addr)) ||
1947 !access_ok((void __user *)(unsigned long)mem->userspace_addr,
1950 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
1952 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
1954 if ((mem->memory_size >> PAGE_SHIFT) > KVM_MEM_MAX_NR_PAGES)
1957 slots = __kvm_memslots(kvm, as_id);
1960 * Note, the old memslot (and the pointer itself!) may be invalidated
1961 * and/or destroyed by kvm_set_memslot().
1963 old = id_to_memslot(slots, id);
1965 if (!mem->memory_size) {
1966 if (!old || !old->npages)
1969 if (WARN_ON_ONCE(kvm->nr_memslot_pages < old->npages))
1972 return kvm_set_memslot(kvm, old, NULL, KVM_MR_DELETE);
1975 base_gfn = (mem->guest_phys_addr >> PAGE_SHIFT);
1976 npages = (mem->memory_size >> PAGE_SHIFT);
1978 if (!old || !old->npages) {
1979 change = KVM_MR_CREATE;
1982 * To simplify KVM internals, the total number of pages across
1983 * all memslots must fit in an unsigned long.
1985 if ((kvm->nr_memslot_pages + npages) < kvm->nr_memslot_pages)
1987 } else { /* Modify an existing slot. */
1988 if ((mem->userspace_addr != old->userspace_addr) ||
1989 (npages != old->npages) ||
1990 ((mem->flags ^ old->flags) & KVM_MEM_READONLY))
1993 if (base_gfn != old->base_gfn)
1994 change = KVM_MR_MOVE;
1995 else if (mem->flags != old->flags)
1996 change = KVM_MR_FLAGS_ONLY;
1997 else /* Nothing to change. */
2001 if ((change == KVM_MR_CREATE || change == KVM_MR_MOVE) &&
2002 kvm_check_memslot_overlap(slots, id, base_gfn, base_gfn + npages))
2005 /* Allocate a slot that will persist in the memslot. */
2006 new = kzalloc(sizeof(*new), GFP_KERNEL_ACCOUNT);
2012 new->base_gfn = base_gfn;
2013 new->npages = npages;
2014 new->flags = mem->flags;
2015 new->userspace_addr = mem->userspace_addr;
2017 r = kvm_set_memslot(kvm, old, new, change);
2022 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
2024 int kvm_set_memory_region(struct kvm *kvm,
2025 const struct kvm_userspace_memory_region *mem)
2029 mutex_lock(&kvm->slots_lock);
2030 r = __kvm_set_memory_region(kvm, mem);
2031 mutex_unlock(&kvm->slots_lock);
2034 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
2036 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
2037 struct kvm_userspace_memory_region *mem)
2039 if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
2042 return kvm_set_memory_region(kvm, mem);
2045 #ifndef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
2047 * kvm_get_dirty_log - get a snapshot of dirty pages
2048 * @kvm: pointer to kvm instance
2049 * @log: slot id and address to which we copy the log
2050 * @is_dirty: set to '1' if any dirty pages were found
2051 * @memslot: set to the associated memslot, always valid on success
2053 int kvm_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log,
2054 int *is_dirty, struct kvm_memory_slot **memslot)
2056 struct kvm_memslots *slots;
2059 unsigned long any = 0;
2061 /* Dirty ring tracking may be exclusive to dirty log tracking */
2062 if (!kvm_use_dirty_bitmap(kvm))
2068 as_id = log->slot >> 16;
2069 id = (u16)log->slot;
2070 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
2073 slots = __kvm_memslots(kvm, as_id);
2074 *memslot = id_to_memslot(slots, id);
2075 if (!(*memslot) || !(*memslot)->dirty_bitmap)
2078 kvm_arch_sync_dirty_log(kvm, *memslot);
2080 n = kvm_dirty_bitmap_bytes(*memslot);
2082 for (i = 0; !any && i < n/sizeof(long); ++i)
2083 any = (*memslot)->dirty_bitmap[i];
2085 if (copy_to_user(log->dirty_bitmap, (*memslot)->dirty_bitmap, n))
2092 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
2094 #else /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
2096 * kvm_get_dirty_log_protect - get a snapshot of dirty pages
2097 * and reenable dirty page tracking for the corresponding pages.
2098 * @kvm: pointer to kvm instance
2099 * @log: slot id and address to which we copy the log
2101 * We need to keep it in mind that VCPU threads can write to the bitmap
2102 * concurrently. So, to avoid losing track of dirty pages we keep the
2105 * 1. Take a snapshot of the bit and clear it if needed.
2106 * 2. Write protect the corresponding page.
2107 * 3. Copy the snapshot to the userspace.
2108 * 4. Upon return caller flushes TLB's if needed.
2110 * Between 2 and 4, the guest may write to the page using the remaining TLB
2111 * entry. This is not a problem because the page is reported dirty using
2112 * the snapshot taken before and step 4 ensures that writes done after
2113 * exiting to userspace will be logged for the next call.
2116 static int kvm_get_dirty_log_protect(struct kvm *kvm, struct kvm_dirty_log *log)
2118 struct kvm_memslots *slots;
2119 struct kvm_memory_slot *memslot;
2122 unsigned long *dirty_bitmap;
2123 unsigned long *dirty_bitmap_buffer;
2126 /* Dirty ring tracking may be exclusive to dirty log tracking */
2127 if (!kvm_use_dirty_bitmap(kvm))
2130 as_id = log->slot >> 16;
2131 id = (u16)log->slot;
2132 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
2135 slots = __kvm_memslots(kvm, as_id);
2136 memslot = id_to_memslot(slots, id);
2137 if (!memslot || !memslot->dirty_bitmap)
2140 dirty_bitmap = memslot->dirty_bitmap;
2142 kvm_arch_sync_dirty_log(kvm, memslot);
2144 n = kvm_dirty_bitmap_bytes(memslot);
2146 if (kvm->manual_dirty_log_protect) {
2148 * Unlike kvm_get_dirty_log, we always return false in *flush,
2149 * because no flush is needed until KVM_CLEAR_DIRTY_LOG. There
2150 * is some code duplication between this function and
2151 * kvm_get_dirty_log, but hopefully all architecture
2152 * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log
2153 * can be eliminated.
2155 dirty_bitmap_buffer = dirty_bitmap;
2157 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
2158 memset(dirty_bitmap_buffer, 0, n);
2161 for (i = 0; i < n / sizeof(long); i++) {
2165 if (!dirty_bitmap[i])
2169 mask = xchg(&dirty_bitmap[i], 0);
2170 dirty_bitmap_buffer[i] = mask;
2172 offset = i * BITS_PER_LONG;
2173 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
2176 KVM_MMU_UNLOCK(kvm);
2180 kvm_arch_flush_remote_tlbs_memslot(kvm, memslot);
2182 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
2189 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
2190 * @kvm: kvm instance
2191 * @log: slot id and address to which we copy the log
2193 * Steps 1-4 below provide general overview of dirty page logging. See
2194 * kvm_get_dirty_log_protect() function description for additional details.
2196 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
2197 * always flush the TLB (step 4) even if previous step failed and the dirty
2198 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
2199 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
2200 * writes will be marked dirty for next log read.
2202 * 1. Take a snapshot of the bit and clear it if needed.
2203 * 2. Write protect the corresponding page.
2204 * 3. Copy the snapshot to the userspace.
2205 * 4. Flush TLB's if needed.
2207 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
2208 struct kvm_dirty_log *log)
2212 mutex_lock(&kvm->slots_lock);
2214 r = kvm_get_dirty_log_protect(kvm, log);
2216 mutex_unlock(&kvm->slots_lock);
2221 * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap
2222 * and reenable dirty page tracking for the corresponding pages.
2223 * @kvm: pointer to kvm instance
2224 * @log: slot id and address from which to fetch the bitmap of dirty pages
2226 static int kvm_clear_dirty_log_protect(struct kvm *kvm,
2227 struct kvm_clear_dirty_log *log)
2229 struct kvm_memslots *slots;
2230 struct kvm_memory_slot *memslot;
2234 unsigned long *dirty_bitmap;
2235 unsigned long *dirty_bitmap_buffer;
2238 /* Dirty ring tracking may be exclusive to dirty log tracking */
2239 if (!kvm_use_dirty_bitmap(kvm))
2242 as_id = log->slot >> 16;
2243 id = (u16)log->slot;
2244 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
2247 if (log->first_page & 63)
2250 slots = __kvm_memslots(kvm, as_id);
2251 memslot = id_to_memslot(slots, id);
2252 if (!memslot || !memslot->dirty_bitmap)
2255 dirty_bitmap = memslot->dirty_bitmap;
2257 n = ALIGN(log->num_pages, BITS_PER_LONG) / 8;
2259 if (log->first_page > memslot->npages ||
2260 log->num_pages > memslot->npages - log->first_page ||
2261 (log->num_pages < memslot->npages - log->first_page && (log->num_pages & 63)))
2264 kvm_arch_sync_dirty_log(kvm, memslot);
2267 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
2268 if (copy_from_user(dirty_bitmap_buffer, log->dirty_bitmap, n))
2272 for (offset = log->first_page, i = offset / BITS_PER_LONG,
2273 n = DIV_ROUND_UP(log->num_pages, BITS_PER_LONG); n--;
2274 i++, offset += BITS_PER_LONG) {
2275 unsigned long mask = *dirty_bitmap_buffer++;
2276 atomic_long_t *p = (atomic_long_t *) &dirty_bitmap[i];
2280 mask &= atomic_long_fetch_andnot(mask, p);
2283 * mask contains the bits that really have been cleared. This
2284 * never includes any bits beyond the length of the memslot (if
2285 * the length is not aligned to 64 pages), therefore it is not
2286 * a problem if userspace sets them in log->dirty_bitmap.
2290 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
2294 KVM_MMU_UNLOCK(kvm);
2297 kvm_arch_flush_remote_tlbs_memslot(kvm, memslot);
2302 static int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm,
2303 struct kvm_clear_dirty_log *log)
2307 mutex_lock(&kvm->slots_lock);
2309 r = kvm_clear_dirty_log_protect(kvm, log);
2311 mutex_unlock(&kvm->slots_lock);
2314 #endif /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
2316 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
2318 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
2320 EXPORT_SYMBOL_GPL(gfn_to_memslot);
2322 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
2324 struct kvm_memslots *slots = kvm_vcpu_memslots(vcpu);
2325 u64 gen = slots->generation;
2326 struct kvm_memory_slot *slot;
2329 * This also protects against using a memslot from a different address space,
2330 * since different address spaces have different generation numbers.
2332 if (unlikely(gen != vcpu->last_used_slot_gen)) {
2333 vcpu->last_used_slot = NULL;
2334 vcpu->last_used_slot_gen = gen;
2337 slot = try_get_memslot(vcpu->last_used_slot, gfn);
2342 * Fall back to searching all memslots. We purposely use
2343 * search_memslots() instead of __gfn_to_memslot() to avoid
2344 * thrashing the VM-wide last_used_slot in kvm_memslots.
2346 slot = search_memslots(slots, gfn, false);
2348 vcpu->last_used_slot = slot;
2355 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
2357 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
2359 return kvm_is_visible_memslot(memslot);
2361 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
2363 bool kvm_vcpu_is_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
2365 struct kvm_memory_slot *memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2367 return kvm_is_visible_memslot(memslot);
2369 EXPORT_SYMBOL_GPL(kvm_vcpu_is_visible_gfn);
2371 unsigned long kvm_host_page_size(struct kvm_vcpu *vcpu, gfn_t gfn)
2373 struct vm_area_struct *vma;
2374 unsigned long addr, size;
2378 addr = kvm_vcpu_gfn_to_hva_prot(vcpu, gfn, NULL);
2379 if (kvm_is_error_hva(addr))
2382 mmap_read_lock(current->mm);
2383 vma = find_vma(current->mm, addr);
2387 size = vma_kernel_pagesize(vma);
2390 mmap_read_unlock(current->mm);
2395 static bool memslot_is_readonly(const struct kvm_memory_slot *slot)
2397 return slot->flags & KVM_MEM_READONLY;
2400 static unsigned long __gfn_to_hva_many(const struct kvm_memory_slot *slot, gfn_t gfn,
2401 gfn_t *nr_pages, bool write)
2403 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
2404 return KVM_HVA_ERR_BAD;
2406 if (memslot_is_readonly(slot) && write)
2407 return KVM_HVA_ERR_RO_BAD;
2410 *nr_pages = slot->npages - (gfn - slot->base_gfn);
2412 return __gfn_to_hva_memslot(slot, gfn);
2415 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
2418 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
2421 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
2424 return gfn_to_hva_many(slot, gfn, NULL);
2426 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
2428 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
2430 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
2432 EXPORT_SYMBOL_GPL(gfn_to_hva);
2434 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
2436 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
2438 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
2441 * Return the hva of a @gfn and the R/W attribute if possible.
2443 * @slot: the kvm_memory_slot which contains @gfn
2444 * @gfn: the gfn to be translated
2445 * @writable: used to return the read/write attribute of the @slot if the hva
2446 * is valid and @writable is not NULL
2448 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
2449 gfn_t gfn, bool *writable)
2451 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
2453 if (!kvm_is_error_hva(hva) && writable)
2454 *writable = !memslot_is_readonly(slot);
2459 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
2461 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2463 return gfn_to_hva_memslot_prot(slot, gfn, writable);
2466 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
2468 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2470 return gfn_to_hva_memslot_prot(slot, gfn, writable);
2473 static inline int check_user_page_hwpoison(unsigned long addr)
2475 int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
2477 rc = get_user_pages(addr, 1, flags, NULL, NULL);
2478 return rc == -EHWPOISON;
2482 * The fast path to get the writable pfn which will be stored in @pfn,
2483 * true indicates success, otherwise false is returned. It's also the
2484 * only part that runs if we can in atomic context.
2486 static bool hva_to_pfn_fast(unsigned long addr, bool write_fault,
2487 bool *writable, kvm_pfn_t *pfn)
2489 struct page *page[1];
2492 * Fast pin a writable pfn only if it is a write fault request
2493 * or the caller allows to map a writable pfn for a read fault
2496 if (!(write_fault || writable))
2499 if (get_user_page_fast_only(addr, FOLL_WRITE, page)) {
2500 *pfn = page_to_pfn(page[0]);
2511 * The slow path to get the pfn of the specified host virtual address,
2512 * 1 indicates success, -errno is returned if error is detected.
2514 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
2515 bool interruptible, bool *writable, kvm_pfn_t *pfn)
2517 unsigned int flags = FOLL_HWPOISON;
2524 *writable = write_fault;
2527 flags |= FOLL_WRITE;
2529 flags |= FOLL_NOWAIT;
2531 flags |= FOLL_INTERRUPTIBLE;
2533 npages = get_user_pages_unlocked(addr, 1, &page, flags);
2537 /* map read fault as writable if possible */
2538 if (unlikely(!write_fault) && writable) {
2541 if (get_user_page_fast_only(addr, FOLL_WRITE, &wpage)) {
2547 *pfn = page_to_pfn(page);
2551 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
2553 if (unlikely(!(vma->vm_flags & VM_READ)))
2556 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
2562 static int kvm_try_get_pfn(kvm_pfn_t pfn)
2564 struct page *page = kvm_pfn_to_refcounted_page(pfn);
2569 return get_page_unless_zero(page);
2572 static int hva_to_pfn_remapped(struct vm_area_struct *vma,
2573 unsigned long addr, bool write_fault,
2574 bool *writable, kvm_pfn_t *p_pfn)
2581 r = follow_pte(vma->vm_mm, addr, &ptep, &ptl);
2584 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
2585 * not call the fault handler, so do it here.
2587 bool unlocked = false;
2588 r = fixup_user_fault(current->mm, addr,
2589 (write_fault ? FAULT_FLAG_WRITE : 0),
2596 r = follow_pte(vma->vm_mm, addr, &ptep, &ptl);
2601 if (write_fault && !pte_write(*ptep)) {
2602 pfn = KVM_PFN_ERR_RO_FAULT;
2607 *writable = pte_write(*ptep);
2608 pfn = pte_pfn(*ptep);
2611 * Get a reference here because callers of *hva_to_pfn* and
2612 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
2613 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
2614 * set, but the kvm_try_get_pfn/kvm_release_pfn_clean pair will
2615 * simply do nothing for reserved pfns.
2617 * Whoever called remap_pfn_range is also going to call e.g.
2618 * unmap_mapping_range before the underlying pages are freed,
2619 * causing a call to our MMU notifier.
2621 * Certain IO or PFNMAP mappings can be backed with valid
2622 * struct pages, but be allocated without refcounting e.g.,
2623 * tail pages of non-compound higher order allocations, which
2624 * would then underflow the refcount when the caller does the
2625 * required put_page. Don't allow those pages here.
2627 if (!kvm_try_get_pfn(pfn))
2631 pte_unmap_unlock(ptep, ptl);
2638 * Pin guest page in memory and return its pfn.
2639 * @addr: host virtual address which maps memory to the guest
2640 * @atomic: whether this function can sleep
2641 * @interruptible: whether the process can be interrupted by non-fatal signals
2642 * @async: whether this function need to wait IO complete if the
2643 * host page is not in the memory
2644 * @write_fault: whether we should get a writable host page
2645 * @writable: whether it allows to map a writable host page for !@write_fault
2647 * The function will map a writable host page for these two cases:
2648 * 1): @write_fault = true
2649 * 2): @write_fault = false && @writable, @writable will tell the caller
2650 * whether the mapping is writable.
2652 kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool interruptible,
2653 bool *async, bool write_fault, bool *writable)
2655 struct vm_area_struct *vma;
2659 /* we can do it either atomically or asynchronously, not both */
2660 BUG_ON(atomic && async);
2662 if (hva_to_pfn_fast(addr, write_fault, writable, &pfn))
2666 return KVM_PFN_ERR_FAULT;
2668 npages = hva_to_pfn_slow(addr, async, write_fault, interruptible,
2672 if (npages == -EINTR)
2673 return KVM_PFN_ERR_SIGPENDING;
2675 mmap_read_lock(current->mm);
2676 if (npages == -EHWPOISON ||
2677 (!async && check_user_page_hwpoison(addr))) {
2678 pfn = KVM_PFN_ERR_HWPOISON;
2683 vma = vma_lookup(current->mm, addr);
2686 pfn = KVM_PFN_ERR_FAULT;
2687 else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
2688 r = hva_to_pfn_remapped(vma, addr, write_fault, writable, &pfn);
2692 pfn = KVM_PFN_ERR_FAULT;
2694 if (async && vma_is_valid(vma, write_fault))
2696 pfn = KVM_PFN_ERR_FAULT;
2699 mmap_read_unlock(current->mm);
2703 kvm_pfn_t __gfn_to_pfn_memslot(const struct kvm_memory_slot *slot, gfn_t gfn,
2704 bool atomic, bool interruptible, bool *async,
2705 bool write_fault, bool *writable, hva_t *hva)
2707 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
2712 if (addr == KVM_HVA_ERR_RO_BAD) {
2715 return KVM_PFN_ERR_RO_FAULT;
2718 if (kvm_is_error_hva(addr)) {
2721 return KVM_PFN_NOSLOT;
2724 /* Do not map writable pfn in the readonly memslot. */
2725 if (writable && memslot_is_readonly(slot)) {
2730 return hva_to_pfn(addr, atomic, interruptible, async, write_fault,
2733 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
2735 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
2738 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, false,
2739 NULL, write_fault, writable, NULL);
2741 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
2743 kvm_pfn_t gfn_to_pfn_memslot(const struct kvm_memory_slot *slot, gfn_t gfn)
2745 return __gfn_to_pfn_memslot(slot, gfn, false, false, NULL, true,
2748 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
2750 kvm_pfn_t gfn_to_pfn_memslot_atomic(const struct kvm_memory_slot *slot, gfn_t gfn)
2752 return __gfn_to_pfn_memslot(slot, gfn, true, false, NULL, true,
2755 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
2757 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
2759 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
2761 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
2763 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
2765 return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
2767 EXPORT_SYMBOL_GPL(gfn_to_pfn);
2769 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
2771 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
2773 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
2775 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
2776 struct page **pages, int nr_pages)
2781 addr = gfn_to_hva_many(slot, gfn, &entry);
2782 if (kvm_is_error_hva(addr))
2785 if (entry < nr_pages)
2788 return get_user_pages_fast_only(addr, nr_pages, FOLL_WRITE, pages);
2790 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
2793 * Do not use this helper unless you are absolutely certain the gfn _must_ be
2794 * backed by 'struct page'. A valid example is if the backing memslot is
2795 * controlled by KVM. Note, if the returned page is valid, it's refcount has
2796 * been elevated by gfn_to_pfn().
2798 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
2803 pfn = gfn_to_pfn(kvm, gfn);
2805 if (is_error_noslot_pfn(pfn))
2806 return KVM_ERR_PTR_BAD_PAGE;
2808 page = kvm_pfn_to_refcounted_page(pfn);
2810 return KVM_ERR_PTR_BAD_PAGE;
2814 EXPORT_SYMBOL_GPL(gfn_to_page);
2816 void kvm_release_pfn(kvm_pfn_t pfn, bool dirty)
2819 kvm_release_pfn_dirty(pfn);
2821 kvm_release_pfn_clean(pfn);
2824 int kvm_vcpu_map(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map)
2828 struct page *page = KVM_UNMAPPED_PAGE;
2833 pfn = gfn_to_pfn(vcpu->kvm, gfn);
2834 if (is_error_noslot_pfn(pfn))
2837 if (pfn_valid(pfn)) {
2838 page = pfn_to_page(pfn);
2840 #ifdef CONFIG_HAS_IOMEM
2842 hva = memremap(pfn_to_hpa(pfn), PAGE_SIZE, MEMREMAP_WB);
2856 EXPORT_SYMBOL_GPL(kvm_vcpu_map);
2858 void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map, bool dirty)
2866 if (map->page != KVM_UNMAPPED_PAGE)
2868 #ifdef CONFIG_HAS_IOMEM
2874 kvm_vcpu_mark_page_dirty(vcpu, map->gfn);
2876 kvm_release_pfn(map->pfn, dirty);
2881 EXPORT_SYMBOL_GPL(kvm_vcpu_unmap);
2883 static bool kvm_is_ad_tracked_page(struct page *page)
2886 * Per page-flags.h, pages tagged PG_reserved "should in general not be
2887 * touched (e.g. set dirty) except by its owner".
2889 return !PageReserved(page);
2892 static void kvm_set_page_dirty(struct page *page)
2894 if (kvm_is_ad_tracked_page(page))
2898 static void kvm_set_page_accessed(struct page *page)
2900 if (kvm_is_ad_tracked_page(page))
2901 mark_page_accessed(page);
2904 void kvm_release_page_clean(struct page *page)
2906 WARN_ON(is_error_page(page));
2908 kvm_set_page_accessed(page);
2911 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
2913 void kvm_release_pfn_clean(kvm_pfn_t pfn)
2917 if (is_error_noslot_pfn(pfn))
2920 page = kvm_pfn_to_refcounted_page(pfn);
2924 kvm_release_page_clean(page);
2926 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
2928 void kvm_release_page_dirty(struct page *page)
2930 WARN_ON(is_error_page(page));
2932 kvm_set_page_dirty(page);
2933 kvm_release_page_clean(page);
2935 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
2937 void kvm_release_pfn_dirty(kvm_pfn_t pfn)
2941 if (is_error_noslot_pfn(pfn))
2944 page = kvm_pfn_to_refcounted_page(pfn);
2948 kvm_release_page_dirty(page);
2950 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
2953 * Note, checking for an error/noslot pfn is the caller's responsibility when
2954 * directly marking a page dirty/accessed. Unlike the "release" helpers, the
2955 * "set" helpers are not to be used when the pfn might point at garbage.
2957 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
2959 if (WARN_ON(is_error_noslot_pfn(pfn)))
2963 kvm_set_page_dirty(pfn_to_page(pfn));
2965 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
2967 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
2969 if (WARN_ON(is_error_noslot_pfn(pfn)))
2973 kvm_set_page_accessed(pfn_to_page(pfn));
2975 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
2977 static int next_segment(unsigned long len, int offset)
2979 if (len > PAGE_SIZE - offset)
2980 return PAGE_SIZE - offset;
2985 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
2986 void *data, int offset, int len)
2991 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
2992 if (kvm_is_error_hva(addr))
2994 r = __copy_from_user(data, (void __user *)addr + offset, len);
3000 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
3003 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
3005 return __kvm_read_guest_page(slot, gfn, data, offset, len);
3007 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
3009 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
3010 int offset, int len)
3012 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
3014 return __kvm_read_guest_page(slot, gfn, data, offset, len);
3016 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
3018 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
3020 gfn_t gfn = gpa >> PAGE_SHIFT;
3022 int offset = offset_in_page(gpa);
3025 while ((seg = next_segment(len, offset)) != 0) {
3026 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
3036 EXPORT_SYMBOL_GPL(kvm_read_guest);
3038 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
3040 gfn_t gfn = gpa >> PAGE_SHIFT;
3042 int offset = offset_in_page(gpa);
3045 while ((seg = next_segment(len, offset)) != 0) {
3046 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
3056 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
3058 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
3059 void *data, int offset, unsigned long len)
3064 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
3065 if (kvm_is_error_hva(addr))
3067 pagefault_disable();
3068 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
3075 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
3076 void *data, unsigned long len)
3078 gfn_t gfn = gpa >> PAGE_SHIFT;
3079 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
3080 int offset = offset_in_page(gpa);
3082 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
3084 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
3086 static int __kvm_write_guest_page(struct kvm *kvm,
3087 struct kvm_memory_slot *memslot, gfn_t gfn,
3088 const void *data, int offset, int len)
3093 addr = gfn_to_hva_memslot(memslot, gfn);
3094 if (kvm_is_error_hva(addr))
3096 r = __copy_to_user((void __user *)addr + offset, data, len);
3099 mark_page_dirty_in_slot(kvm, memslot, gfn);
3103 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
3104 const void *data, int offset, int len)
3106 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
3108 return __kvm_write_guest_page(kvm, slot, gfn, data, offset, len);
3110 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
3112 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
3113 const void *data, int offset, int len)
3115 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
3117 return __kvm_write_guest_page(vcpu->kvm, slot, gfn, data, offset, len);
3119 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
3121 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
3124 gfn_t gfn = gpa >> PAGE_SHIFT;
3126 int offset = offset_in_page(gpa);
3129 while ((seg = next_segment(len, offset)) != 0) {
3130 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
3140 EXPORT_SYMBOL_GPL(kvm_write_guest);
3142 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
3145 gfn_t gfn = gpa >> PAGE_SHIFT;
3147 int offset = offset_in_page(gpa);
3150 while ((seg = next_segment(len, offset)) != 0) {
3151 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
3161 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
3163 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
3164 struct gfn_to_hva_cache *ghc,
3165 gpa_t gpa, unsigned long len)
3167 int offset = offset_in_page(gpa);
3168 gfn_t start_gfn = gpa >> PAGE_SHIFT;
3169 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
3170 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
3171 gfn_t nr_pages_avail;
3173 /* Update ghc->generation before performing any error checks. */
3174 ghc->generation = slots->generation;
3176 if (start_gfn > end_gfn) {
3177 ghc->hva = KVM_HVA_ERR_BAD;
3182 * If the requested region crosses two memslots, we still
3183 * verify that the entire region is valid here.
3185 for ( ; start_gfn <= end_gfn; start_gfn += nr_pages_avail) {
3186 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
3187 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
3189 if (kvm_is_error_hva(ghc->hva))
3193 /* Use the slow path for cross page reads and writes. */
3194 if (nr_pages_needed == 1)
3197 ghc->memslot = NULL;
3204 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3205 gpa_t gpa, unsigned long len)
3207 struct kvm_memslots *slots = kvm_memslots(kvm);
3208 return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len);
3210 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
3212 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3213 void *data, unsigned int offset,
3216 struct kvm_memslots *slots = kvm_memslots(kvm);
3218 gpa_t gpa = ghc->gpa + offset;
3220 if (WARN_ON_ONCE(len + offset > ghc->len))
3223 if (slots->generation != ghc->generation) {
3224 if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len))
3228 if (kvm_is_error_hva(ghc->hva))
3231 if (unlikely(!ghc->memslot))
3232 return kvm_write_guest(kvm, gpa, data, len);
3234 r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
3237 mark_page_dirty_in_slot(kvm, ghc->memslot, gpa >> PAGE_SHIFT);
3241 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached);
3243 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3244 void *data, unsigned long len)
3246 return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len);
3248 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
3250 int kvm_read_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3251 void *data, unsigned int offset,
3254 struct kvm_memslots *slots = kvm_memslots(kvm);
3256 gpa_t gpa = ghc->gpa + offset;
3258 if (WARN_ON_ONCE(len + offset > ghc->len))
3261 if (slots->generation != ghc->generation) {
3262 if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len))
3266 if (kvm_is_error_hva(ghc->hva))
3269 if (unlikely(!ghc->memslot))
3270 return kvm_read_guest(kvm, gpa, data, len);
3272 r = __copy_from_user(data, (void __user *)ghc->hva + offset, len);
3278 EXPORT_SYMBOL_GPL(kvm_read_guest_offset_cached);
3280 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3281 void *data, unsigned long len)
3283 return kvm_read_guest_offset_cached(kvm, ghc, data, 0, len);
3285 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
3287 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
3289 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
3290 gfn_t gfn = gpa >> PAGE_SHIFT;
3292 int offset = offset_in_page(gpa);
3295 while ((seg = next_segment(len, offset)) != 0) {
3296 ret = kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
3305 EXPORT_SYMBOL_GPL(kvm_clear_guest);
3307 void mark_page_dirty_in_slot(struct kvm *kvm,
3308 const struct kvm_memory_slot *memslot,
3311 struct kvm_vcpu *vcpu = kvm_get_running_vcpu();
3313 #ifdef CONFIG_HAVE_KVM_DIRTY_RING
3314 if (WARN_ON_ONCE(vcpu && vcpu->kvm != kvm))
3317 WARN_ON_ONCE(!vcpu && !kvm_arch_allow_write_without_running_vcpu(kvm));
3320 if (memslot && kvm_slot_dirty_track_enabled(memslot)) {
3321 unsigned long rel_gfn = gfn - memslot->base_gfn;
3322 u32 slot = (memslot->as_id << 16) | memslot->id;
3324 if (kvm->dirty_ring_size && vcpu)
3325 kvm_dirty_ring_push(vcpu, slot, rel_gfn);
3326 else if (memslot->dirty_bitmap)
3327 set_bit_le(rel_gfn, memslot->dirty_bitmap);
3330 EXPORT_SYMBOL_GPL(mark_page_dirty_in_slot);
3332 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
3334 struct kvm_memory_slot *memslot;
3336 memslot = gfn_to_memslot(kvm, gfn);
3337 mark_page_dirty_in_slot(kvm, memslot, gfn);
3339 EXPORT_SYMBOL_GPL(mark_page_dirty);
3341 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
3343 struct kvm_memory_slot *memslot;
3345 memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
3346 mark_page_dirty_in_slot(vcpu->kvm, memslot, gfn);
3348 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
3350 void kvm_sigset_activate(struct kvm_vcpu *vcpu)
3352 if (!vcpu->sigset_active)
3356 * This does a lockless modification of ->real_blocked, which is fine
3357 * because, only current can change ->real_blocked and all readers of
3358 * ->real_blocked don't care as long ->real_blocked is always a subset
3361 sigprocmask(SIG_SETMASK, &vcpu->sigset, ¤t->real_blocked);
3364 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu)
3366 if (!vcpu->sigset_active)
3369 sigprocmask(SIG_SETMASK, ¤t->real_blocked, NULL);
3370 sigemptyset(¤t->real_blocked);
3373 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
3375 unsigned int old, val, grow, grow_start;
3377 old = val = vcpu->halt_poll_ns;
3378 grow_start = READ_ONCE(halt_poll_ns_grow_start);
3379 grow = READ_ONCE(halt_poll_ns_grow);
3384 if (val < grow_start)
3387 vcpu->halt_poll_ns = val;
3389 trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
3392 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
3394 unsigned int old, val, shrink, grow_start;
3396 old = val = vcpu->halt_poll_ns;
3397 shrink = READ_ONCE(halt_poll_ns_shrink);
3398 grow_start = READ_ONCE(halt_poll_ns_grow_start);
3404 if (val < grow_start)
3407 vcpu->halt_poll_ns = val;
3408 trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
3411 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
3414 int idx = srcu_read_lock(&vcpu->kvm->srcu);
3416 if (kvm_arch_vcpu_runnable(vcpu))
3418 if (kvm_cpu_has_pending_timer(vcpu))
3420 if (signal_pending(current))
3422 if (kvm_check_request(KVM_REQ_UNBLOCK, vcpu))
3427 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3432 * Block the vCPU until the vCPU is runnable, an event arrives, or a signal is
3433 * pending. This is mostly used when halting a vCPU, but may also be used
3434 * directly for other vCPU non-runnable states, e.g. x86's Wait-For-SIPI.
3436 bool kvm_vcpu_block(struct kvm_vcpu *vcpu)
3438 struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu);
3439 bool waited = false;
3441 vcpu->stat.generic.blocking = 1;
3444 kvm_arch_vcpu_blocking(vcpu);
3445 prepare_to_rcuwait(wait);
3449 set_current_state(TASK_INTERRUPTIBLE);
3451 if (kvm_vcpu_check_block(vcpu) < 0)
3459 finish_rcuwait(wait);
3460 kvm_arch_vcpu_unblocking(vcpu);
3463 vcpu->stat.generic.blocking = 0;
3468 static inline void update_halt_poll_stats(struct kvm_vcpu *vcpu, ktime_t start,
3469 ktime_t end, bool success)
3471 struct kvm_vcpu_stat_generic *stats = &vcpu->stat.generic;
3472 u64 poll_ns = ktime_to_ns(ktime_sub(end, start));
3474 ++vcpu->stat.generic.halt_attempted_poll;
3477 ++vcpu->stat.generic.halt_successful_poll;
3479 if (!vcpu_valid_wakeup(vcpu))
3480 ++vcpu->stat.generic.halt_poll_invalid;
3482 stats->halt_poll_success_ns += poll_ns;
3483 KVM_STATS_LOG_HIST_UPDATE(stats->halt_poll_success_hist, poll_ns);
3485 stats->halt_poll_fail_ns += poll_ns;
3486 KVM_STATS_LOG_HIST_UPDATE(stats->halt_poll_fail_hist, poll_ns);
3490 static unsigned int kvm_vcpu_max_halt_poll_ns(struct kvm_vcpu *vcpu)
3492 struct kvm *kvm = vcpu->kvm;
3494 if (kvm->override_halt_poll_ns) {
3496 * Ensure kvm->max_halt_poll_ns is not read before
3497 * kvm->override_halt_poll_ns.
3499 * Pairs with the smp_wmb() when enabling KVM_CAP_HALT_POLL.
3502 return READ_ONCE(kvm->max_halt_poll_ns);
3505 return READ_ONCE(halt_poll_ns);
3509 * Emulate a vCPU halt condition, e.g. HLT on x86, WFI on arm, etc... If halt
3510 * polling is enabled, busy wait for a short time before blocking to avoid the
3511 * expensive block+unblock sequence if a wake event arrives soon after the vCPU
3514 void kvm_vcpu_halt(struct kvm_vcpu *vcpu)
3516 unsigned int max_halt_poll_ns = kvm_vcpu_max_halt_poll_ns(vcpu);
3517 bool halt_poll_allowed = !kvm_arch_no_poll(vcpu);
3518 ktime_t start, cur, poll_end;
3519 bool waited = false;
3523 if (vcpu->halt_poll_ns > max_halt_poll_ns)
3524 vcpu->halt_poll_ns = max_halt_poll_ns;
3526 do_halt_poll = halt_poll_allowed && vcpu->halt_poll_ns;
3528 start = cur = poll_end = ktime_get();
3530 ktime_t stop = ktime_add_ns(start, vcpu->halt_poll_ns);
3533 if (kvm_vcpu_check_block(vcpu) < 0)
3536 poll_end = cur = ktime_get();
3537 } while (kvm_vcpu_can_poll(cur, stop));
3540 waited = kvm_vcpu_block(vcpu);
3544 vcpu->stat.generic.halt_wait_ns +=
3545 ktime_to_ns(cur) - ktime_to_ns(poll_end);
3546 KVM_STATS_LOG_HIST_UPDATE(vcpu->stat.generic.halt_wait_hist,
3547 ktime_to_ns(cur) - ktime_to_ns(poll_end));
3550 /* The total time the vCPU was "halted", including polling time. */
3551 halt_ns = ktime_to_ns(cur) - ktime_to_ns(start);
3554 * Note, halt-polling is considered successful so long as the vCPU was
3555 * never actually scheduled out, i.e. even if the wake event arrived
3556 * after of the halt-polling loop itself, but before the full wait.
3559 update_halt_poll_stats(vcpu, start, poll_end, !waited);
3561 if (halt_poll_allowed) {
3562 /* Recompute the max halt poll time in case it changed. */
3563 max_halt_poll_ns = kvm_vcpu_max_halt_poll_ns(vcpu);
3565 if (!vcpu_valid_wakeup(vcpu)) {
3566 shrink_halt_poll_ns(vcpu);
3567 } else if (max_halt_poll_ns) {
3568 if (halt_ns <= vcpu->halt_poll_ns)
3570 /* we had a long block, shrink polling */
3571 else if (vcpu->halt_poll_ns &&
3572 halt_ns > max_halt_poll_ns)
3573 shrink_halt_poll_ns(vcpu);
3574 /* we had a short halt and our poll time is too small */
3575 else if (vcpu->halt_poll_ns < max_halt_poll_ns &&
3576 halt_ns < max_halt_poll_ns)
3577 grow_halt_poll_ns(vcpu);
3579 vcpu->halt_poll_ns = 0;
3583 trace_kvm_vcpu_wakeup(halt_ns, waited, vcpu_valid_wakeup(vcpu));
3585 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
3587 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
3589 if (__kvm_vcpu_wake_up(vcpu)) {
3590 WRITE_ONCE(vcpu->ready, true);
3591 ++vcpu->stat.generic.halt_wakeup;
3597 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
3601 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
3603 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
3607 if (kvm_vcpu_wake_up(vcpu))
3612 * The only state change done outside the vcpu mutex is IN_GUEST_MODE
3613 * to EXITING_GUEST_MODE. Therefore the moderately expensive "should
3614 * kick" check does not need atomic operations if kvm_vcpu_kick is used
3615 * within the vCPU thread itself.
3617 if (vcpu == __this_cpu_read(kvm_running_vcpu)) {
3618 if (vcpu->mode == IN_GUEST_MODE)
3619 WRITE_ONCE(vcpu->mode, EXITING_GUEST_MODE);
3624 * Note, the vCPU could get migrated to a different pCPU at any point
3625 * after kvm_arch_vcpu_should_kick(), which could result in sending an
3626 * IPI to the previous pCPU. But, that's ok because the purpose of the
3627 * IPI is to force the vCPU to leave IN_GUEST_MODE, and migrating the
3628 * vCPU also requires it to leave IN_GUEST_MODE.
3630 if (kvm_arch_vcpu_should_kick(vcpu)) {
3631 cpu = READ_ONCE(vcpu->cpu);
3632 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
3633 smp_send_reschedule(cpu);
3638 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
3639 #endif /* !CONFIG_S390 */
3641 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
3644 struct task_struct *task = NULL;
3648 pid = rcu_dereference(target->pid);
3650 task = get_pid_task(pid, PIDTYPE_PID);
3654 ret = yield_to(task, 1);
3655 put_task_struct(task);
3659 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
3662 * Helper that checks whether a VCPU is eligible for directed yield.
3663 * Most eligible candidate to yield is decided by following heuristics:
3665 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
3666 * (preempted lock holder), indicated by @in_spin_loop.
3667 * Set at the beginning and cleared at the end of interception/PLE handler.
3669 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
3670 * chance last time (mostly it has become eligible now since we have probably
3671 * yielded to lockholder in last iteration. This is done by toggling
3672 * @dy_eligible each time a VCPU checked for eligibility.)
3674 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
3675 * to preempted lock-holder could result in wrong VCPU selection and CPU
3676 * burning. Giving priority for a potential lock-holder increases lock
3679 * Since algorithm is based on heuristics, accessing another VCPU data without
3680 * locking does not harm. It may result in trying to yield to same VCPU, fail
3681 * and continue with next VCPU and so on.
3683 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
3685 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
3688 eligible = !vcpu->spin_loop.in_spin_loop ||
3689 vcpu->spin_loop.dy_eligible;
3691 if (vcpu->spin_loop.in_spin_loop)
3692 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
3701 * Unlike kvm_arch_vcpu_runnable, this function is called outside
3702 * a vcpu_load/vcpu_put pair. However, for most architectures
3703 * kvm_arch_vcpu_runnable does not require vcpu_load.
3705 bool __weak kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
3707 return kvm_arch_vcpu_runnable(vcpu);
3710 static bool vcpu_dy_runnable(struct kvm_vcpu *vcpu)
3712 if (kvm_arch_dy_runnable(vcpu))
3715 #ifdef CONFIG_KVM_ASYNC_PF
3716 if (!list_empty_careful(&vcpu->async_pf.done))
3723 bool __weak kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu)
3728 void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode)
3730 struct kvm *kvm = me->kvm;
3731 struct kvm_vcpu *vcpu;
3732 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
3738 kvm_vcpu_set_in_spin_loop(me, true);
3740 * We boost the priority of a VCPU that is runnable but not
3741 * currently running, because it got preempted by something
3742 * else and called schedule in __vcpu_run. Hopefully that
3743 * VCPU is holding the lock that we need and will release it.
3744 * We approximate round-robin by starting at the last boosted VCPU.
3746 for (pass = 0; pass < 2 && !yielded && try; pass++) {
3747 kvm_for_each_vcpu(i, vcpu, kvm) {
3748 if (!pass && i <= last_boosted_vcpu) {
3749 i = last_boosted_vcpu;
3751 } else if (pass && i > last_boosted_vcpu)
3753 if (!READ_ONCE(vcpu->ready))
3757 if (kvm_vcpu_is_blocking(vcpu) && !vcpu_dy_runnable(vcpu))
3759 if (READ_ONCE(vcpu->preempted) && yield_to_kernel_mode &&
3760 !kvm_arch_dy_has_pending_interrupt(vcpu) &&
3761 !kvm_arch_vcpu_in_kernel(vcpu))
3763 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
3766 yielded = kvm_vcpu_yield_to(vcpu);
3768 kvm->last_boosted_vcpu = i;
3770 } else if (yielded < 0) {
3777 kvm_vcpu_set_in_spin_loop(me, false);
3779 /* Ensure vcpu is not eligible during next spinloop */
3780 kvm_vcpu_set_dy_eligible(me, false);
3782 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
3784 static bool kvm_page_in_dirty_ring(struct kvm *kvm, unsigned long pgoff)
3786 #ifdef CONFIG_HAVE_KVM_DIRTY_RING
3787 return (pgoff >= KVM_DIRTY_LOG_PAGE_OFFSET) &&
3788 (pgoff < KVM_DIRTY_LOG_PAGE_OFFSET +
3789 kvm->dirty_ring_size / PAGE_SIZE);
3795 static vm_fault_t kvm_vcpu_fault(struct vm_fault *vmf)
3797 struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data;
3800 if (vmf->pgoff == 0)
3801 page = virt_to_page(vcpu->run);
3803 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
3804 page = virt_to_page(vcpu->arch.pio_data);
3806 #ifdef CONFIG_KVM_MMIO
3807 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
3808 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
3810 else if (kvm_page_in_dirty_ring(vcpu->kvm, vmf->pgoff))
3811 page = kvm_dirty_ring_get_page(
3813 vmf->pgoff - KVM_DIRTY_LOG_PAGE_OFFSET);
3815 return kvm_arch_vcpu_fault(vcpu, vmf);
3821 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
3822 .fault = kvm_vcpu_fault,
3825 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
3827 struct kvm_vcpu *vcpu = file->private_data;
3828 unsigned long pages = vma_pages(vma);
3830 if ((kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff) ||
3831 kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff + pages - 1)) &&
3832 ((vma->vm_flags & VM_EXEC) || !(vma->vm_flags & VM_SHARED)))
3835 vma->vm_ops = &kvm_vcpu_vm_ops;
3839 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
3841 struct kvm_vcpu *vcpu = filp->private_data;
3843 kvm_put_kvm(vcpu->kvm);
3847 static const struct file_operations kvm_vcpu_fops = {
3848 .release = kvm_vcpu_release,
3849 .unlocked_ioctl = kvm_vcpu_ioctl,
3850 .mmap = kvm_vcpu_mmap,
3851 .llseek = noop_llseek,
3852 KVM_COMPAT(kvm_vcpu_compat_ioctl),
3856 * Allocates an inode for the vcpu.
3858 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
3860 char name[8 + 1 + ITOA_MAX_LEN + 1];
3862 snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id);
3863 return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
3866 #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS
3867 static int vcpu_get_pid(void *data, u64 *val)
3869 struct kvm_vcpu *vcpu = (struct kvm_vcpu *) data;
3870 *val = pid_nr(rcu_access_pointer(vcpu->pid));
3874 DEFINE_SIMPLE_ATTRIBUTE(vcpu_get_pid_fops, vcpu_get_pid, NULL, "%llu\n");
3876 static void kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
3878 struct dentry *debugfs_dentry;
3879 char dir_name[ITOA_MAX_LEN * 2];
3881 if (!debugfs_initialized())
3884 snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
3885 debugfs_dentry = debugfs_create_dir(dir_name,
3886 vcpu->kvm->debugfs_dentry);
3887 debugfs_create_file("pid", 0444, debugfs_dentry, vcpu,
3888 &vcpu_get_pid_fops);
3890 kvm_arch_create_vcpu_debugfs(vcpu, debugfs_dentry);
3895 * Creates some virtual cpus. Good luck creating more than one.
3897 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
3900 struct kvm_vcpu *vcpu;
3903 if (id >= KVM_MAX_VCPU_IDS)
3906 mutex_lock(&kvm->lock);
3907 if (kvm->created_vcpus >= kvm->max_vcpus) {
3908 mutex_unlock(&kvm->lock);
3912 r = kvm_arch_vcpu_precreate(kvm, id);
3914 mutex_unlock(&kvm->lock);
3918 kvm->created_vcpus++;
3919 mutex_unlock(&kvm->lock);
3921 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL_ACCOUNT);
3924 goto vcpu_decrement;
3927 BUILD_BUG_ON(sizeof(struct kvm_run) > PAGE_SIZE);
3928 page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
3933 vcpu->run = page_address(page);
3935 kvm_vcpu_init(vcpu, kvm, id);
3937 r = kvm_arch_vcpu_create(vcpu);
3939 goto vcpu_free_run_page;
3941 if (kvm->dirty_ring_size) {
3942 r = kvm_dirty_ring_alloc(&vcpu->dirty_ring,
3943 id, kvm->dirty_ring_size);
3945 goto arch_vcpu_destroy;
3948 mutex_lock(&kvm->lock);
3950 #ifdef CONFIG_LOCKDEP
3951 /* Ensure that lockdep knows vcpu->mutex is taken *inside* kvm->lock */
3952 mutex_lock(&vcpu->mutex);
3953 mutex_unlock(&vcpu->mutex);
3956 if (kvm_get_vcpu_by_id(kvm, id)) {
3958 goto unlock_vcpu_destroy;
3961 vcpu->vcpu_idx = atomic_read(&kvm->online_vcpus);
3962 r = xa_insert(&kvm->vcpu_array, vcpu->vcpu_idx, vcpu, GFP_KERNEL_ACCOUNT);
3963 BUG_ON(r == -EBUSY);
3965 goto unlock_vcpu_destroy;
3967 /* Now it's all set up, let userspace reach it */
3969 r = create_vcpu_fd(vcpu);
3971 xa_erase(&kvm->vcpu_array, vcpu->vcpu_idx);
3972 kvm_put_kvm_no_destroy(kvm);
3973 goto unlock_vcpu_destroy;
3977 * Pairs with smp_rmb() in kvm_get_vcpu. Store the vcpu
3978 * pointer before kvm->online_vcpu's incremented value.
3981 atomic_inc(&kvm->online_vcpus);
3983 mutex_unlock(&kvm->lock);
3984 kvm_arch_vcpu_postcreate(vcpu);
3985 kvm_create_vcpu_debugfs(vcpu);
3988 unlock_vcpu_destroy:
3989 mutex_unlock(&kvm->lock);
3990 kvm_dirty_ring_free(&vcpu->dirty_ring);
3992 kvm_arch_vcpu_destroy(vcpu);
3994 free_page((unsigned long)vcpu->run);
3996 kmem_cache_free(kvm_vcpu_cache, vcpu);
3998 mutex_lock(&kvm->lock);
3999 kvm->created_vcpus--;
4000 mutex_unlock(&kvm->lock);
4004 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
4007 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
4008 vcpu->sigset_active = 1;
4009 vcpu->sigset = *sigset;
4011 vcpu->sigset_active = 0;
4015 static ssize_t kvm_vcpu_stats_read(struct file *file, char __user *user_buffer,
4016 size_t size, loff_t *offset)
4018 struct kvm_vcpu *vcpu = file->private_data;
4020 return kvm_stats_read(vcpu->stats_id, &kvm_vcpu_stats_header,
4021 &kvm_vcpu_stats_desc[0], &vcpu->stat,
4022 sizeof(vcpu->stat), user_buffer, size, offset);
4025 static const struct file_operations kvm_vcpu_stats_fops = {
4026 .read = kvm_vcpu_stats_read,
4027 .llseek = noop_llseek,
4030 static int kvm_vcpu_ioctl_get_stats_fd(struct kvm_vcpu *vcpu)
4034 char name[15 + ITOA_MAX_LEN + 1];
4036 snprintf(name, sizeof(name), "kvm-vcpu-stats:%d", vcpu->vcpu_id);
4038 fd = get_unused_fd_flags(O_CLOEXEC);
4042 file = anon_inode_getfile(name, &kvm_vcpu_stats_fops, vcpu, O_RDONLY);
4045 return PTR_ERR(file);
4047 file->f_mode |= FMODE_PREAD;
4048 fd_install(fd, file);
4053 static long kvm_vcpu_ioctl(struct file *filp,
4054 unsigned int ioctl, unsigned long arg)
4056 struct kvm_vcpu *vcpu = filp->private_data;
4057 void __user *argp = (void __user *)arg;
4059 struct kvm_fpu *fpu = NULL;
4060 struct kvm_sregs *kvm_sregs = NULL;
4062 if (vcpu->kvm->mm != current->mm || vcpu->kvm->vm_dead)
4065 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
4069 * Some architectures have vcpu ioctls that are asynchronous to vcpu
4070 * execution; mutex_lock() would break them.
4072 r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg);
4073 if (r != -ENOIOCTLCMD)
4076 if (mutex_lock_killable(&vcpu->mutex))
4084 oldpid = rcu_access_pointer(vcpu->pid);
4085 if (unlikely(oldpid != task_pid(current))) {
4086 /* The thread running this VCPU changed. */
4089 r = kvm_arch_vcpu_run_pid_change(vcpu);
4093 newpid = get_task_pid(current, PIDTYPE_PID);
4094 rcu_assign_pointer(vcpu->pid, newpid);
4099 r = kvm_arch_vcpu_ioctl_run(vcpu);
4100 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
4103 case KVM_GET_REGS: {
4104 struct kvm_regs *kvm_regs;
4107 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL_ACCOUNT);
4110 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
4114 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
4121 case KVM_SET_REGS: {
4122 struct kvm_regs *kvm_regs;
4124 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
4125 if (IS_ERR(kvm_regs)) {
4126 r = PTR_ERR(kvm_regs);
4129 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
4133 case KVM_GET_SREGS: {
4134 kvm_sregs = kzalloc(sizeof(struct kvm_sregs),
4135 GFP_KERNEL_ACCOUNT);
4139 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
4143 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
4148 case KVM_SET_SREGS: {
4149 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
4150 if (IS_ERR(kvm_sregs)) {
4151 r = PTR_ERR(kvm_sregs);
4155 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
4158 case KVM_GET_MP_STATE: {
4159 struct kvm_mp_state mp_state;
4161 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
4165 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
4170 case KVM_SET_MP_STATE: {
4171 struct kvm_mp_state mp_state;
4174 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
4176 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
4179 case KVM_TRANSLATE: {
4180 struct kvm_translation tr;
4183 if (copy_from_user(&tr, argp, sizeof(tr)))
4185 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
4189 if (copy_to_user(argp, &tr, sizeof(tr)))
4194 case KVM_SET_GUEST_DEBUG: {
4195 struct kvm_guest_debug dbg;
4198 if (copy_from_user(&dbg, argp, sizeof(dbg)))
4200 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
4203 case KVM_SET_SIGNAL_MASK: {
4204 struct kvm_signal_mask __user *sigmask_arg = argp;
4205 struct kvm_signal_mask kvm_sigmask;
4206 sigset_t sigset, *p;
4211 if (copy_from_user(&kvm_sigmask, argp,
4212 sizeof(kvm_sigmask)))
4215 if (kvm_sigmask.len != sizeof(sigset))
4218 if (copy_from_user(&sigset, sigmask_arg->sigset,
4223 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
4227 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL_ACCOUNT);
4231 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
4235 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
4241 fpu = memdup_user(argp, sizeof(*fpu));
4247 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
4250 case KVM_GET_STATS_FD: {
4251 r = kvm_vcpu_ioctl_get_stats_fd(vcpu);
4255 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
4258 mutex_unlock(&vcpu->mutex);
4264 #ifdef CONFIG_KVM_COMPAT
4265 static long kvm_vcpu_compat_ioctl(struct file *filp,
4266 unsigned int ioctl, unsigned long arg)
4268 struct kvm_vcpu *vcpu = filp->private_data;
4269 void __user *argp = compat_ptr(arg);
4272 if (vcpu->kvm->mm != current->mm || vcpu->kvm->vm_dead)
4276 case KVM_SET_SIGNAL_MASK: {
4277 struct kvm_signal_mask __user *sigmask_arg = argp;
4278 struct kvm_signal_mask kvm_sigmask;
4283 if (copy_from_user(&kvm_sigmask, argp,
4284 sizeof(kvm_sigmask)))
4287 if (kvm_sigmask.len != sizeof(compat_sigset_t))
4290 if (get_compat_sigset(&sigset,
4291 (compat_sigset_t __user *)sigmask_arg->sigset))
4293 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
4295 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
4299 r = kvm_vcpu_ioctl(filp, ioctl, arg);
4307 static int kvm_device_mmap(struct file *filp, struct vm_area_struct *vma)
4309 struct kvm_device *dev = filp->private_data;
4312 return dev->ops->mmap(dev, vma);
4317 static int kvm_device_ioctl_attr(struct kvm_device *dev,
4318 int (*accessor)(struct kvm_device *dev,
4319 struct kvm_device_attr *attr),
4322 struct kvm_device_attr attr;
4327 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
4330 return accessor(dev, &attr);
4333 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
4336 struct kvm_device *dev = filp->private_data;
4338 if (dev->kvm->mm != current->mm || dev->kvm->vm_dead)
4342 case KVM_SET_DEVICE_ATTR:
4343 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
4344 case KVM_GET_DEVICE_ATTR:
4345 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
4346 case KVM_HAS_DEVICE_ATTR:
4347 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
4349 if (dev->ops->ioctl)
4350 return dev->ops->ioctl(dev, ioctl, arg);
4356 static int kvm_device_release(struct inode *inode, struct file *filp)
4358 struct kvm_device *dev = filp->private_data;
4359 struct kvm *kvm = dev->kvm;
4361 if (dev->ops->release) {
4362 mutex_lock(&kvm->lock);
4363 list_del(&dev->vm_node);
4364 dev->ops->release(dev);
4365 mutex_unlock(&kvm->lock);
4372 static const struct file_operations kvm_device_fops = {
4373 .unlocked_ioctl = kvm_device_ioctl,
4374 .release = kvm_device_release,
4375 KVM_COMPAT(kvm_device_ioctl),
4376 .mmap = kvm_device_mmap,
4379 struct kvm_device *kvm_device_from_filp(struct file *filp)
4381 if (filp->f_op != &kvm_device_fops)
4384 return filp->private_data;
4387 static const struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
4388 #ifdef CONFIG_KVM_MPIC
4389 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
4390 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
4394 int kvm_register_device_ops(const struct kvm_device_ops *ops, u32 type)
4396 if (type >= ARRAY_SIZE(kvm_device_ops_table))
4399 if (kvm_device_ops_table[type] != NULL)
4402 kvm_device_ops_table[type] = ops;
4406 void kvm_unregister_device_ops(u32 type)
4408 if (kvm_device_ops_table[type] != NULL)
4409 kvm_device_ops_table[type] = NULL;
4412 static int kvm_ioctl_create_device(struct kvm *kvm,
4413 struct kvm_create_device *cd)
4415 const struct kvm_device_ops *ops;
4416 struct kvm_device *dev;
4417 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
4421 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
4424 type = array_index_nospec(cd->type, ARRAY_SIZE(kvm_device_ops_table));
4425 ops = kvm_device_ops_table[type];
4432 dev = kzalloc(sizeof(*dev), GFP_KERNEL_ACCOUNT);
4439 mutex_lock(&kvm->lock);
4440 ret = ops->create(dev, type);
4442 mutex_unlock(&kvm->lock);
4446 list_add(&dev->vm_node, &kvm->devices);
4447 mutex_unlock(&kvm->lock);
4453 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
4455 kvm_put_kvm_no_destroy(kvm);
4456 mutex_lock(&kvm->lock);
4457 list_del(&dev->vm_node);
4460 mutex_unlock(&kvm->lock);
4470 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
4473 case KVM_CAP_USER_MEMORY:
4474 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
4475 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
4476 case KVM_CAP_INTERNAL_ERROR_DATA:
4477 #ifdef CONFIG_HAVE_KVM_MSI
4478 case KVM_CAP_SIGNAL_MSI:
4480 #ifdef CONFIG_HAVE_KVM_IRQFD
4482 case KVM_CAP_IRQFD_RESAMPLE:
4484 case KVM_CAP_IOEVENTFD_ANY_LENGTH:
4485 case KVM_CAP_CHECK_EXTENSION_VM:
4486 case KVM_CAP_ENABLE_CAP_VM:
4487 case KVM_CAP_HALT_POLL:
4489 #ifdef CONFIG_KVM_MMIO
4490 case KVM_CAP_COALESCED_MMIO:
4491 return KVM_COALESCED_MMIO_PAGE_OFFSET;
4492 case KVM_CAP_COALESCED_PIO:
4495 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4496 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2:
4497 return KVM_DIRTY_LOG_MANUAL_CAPS;
4499 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
4500 case KVM_CAP_IRQ_ROUTING:
4501 return KVM_MAX_IRQ_ROUTES;
4503 #if KVM_ADDRESS_SPACE_NUM > 1
4504 case KVM_CAP_MULTI_ADDRESS_SPACE:
4505 return KVM_ADDRESS_SPACE_NUM;
4507 case KVM_CAP_NR_MEMSLOTS:
4508 return KVM_USER_MEM_SLOTS;
4509 case KVM_CAP_DIRTY_LOG_RING:
4510 #ifdef CONFIG_HAVE_KVM_DIRTY_RING_TSO
4511 return KVM_DIRTY_RING_MAX_ENTRIES * sizeof(struct kvm_dirty_gfn);
4515 case KVM_CAP_DIRTY_LOG_RING_ACQ_REL:
4516 #ifdef CONFIG_HAVE_KVM_DIRTY_RING_ACQ_REL
4517 return KVM_DIRTY_RING_MAX_ENTRIES * sizeof(struct kvm_dirty_gfn);
4521 #ifdef CONFIG_NEED_KVM_DIRTY_RING_WITH_BITMAP
4522 case KVM_CAP_DIRTY_LOG_RING_WITH_BITMAP:
4524 case KVM_CAP_BINARY_STATS_FD:
4525 case KVM_CAP_SYSTEM_EVENT_DATA:
4530 return kvm_vm_ioctl_check_extension(kvm, arg);
4533 static int kvm_vm_ioctl_enable_dirty_log_ring(struct kvm *kvm, u32 size)
4537 if (!KVM_DIRTY_LOG_PAGE_OFFSET)
4540 /* the size should be power of 2 */
4541 if (!size || (size & (size - 1)))
4544 /* Should be bigger to keep the reserved entries, or a page */
4545 if (size < kvm_dirty_ring_get_rsvd_entries() *
4546 sizeof(struct kvm_dirty_gfn) || size < PAGE_SIZE)
4549 if (size > KVM_DIRTY_RING_MAX_ENTRIES *
4550 sizeof(struct kvm_dirty_gfn))
4553 /* We only allow it to set once */
4554 if (kvm->dirty_ring_size)
4557 mutex_lock(&kvm->lock);
4559 if (kvm->created_vcpus) {
4560 /* We don't allow to change this value after vcpu created */
4563 kvm->dirty_ring_size = size;
4567 mutex_unlock(&kvm->lock);
4571 static int kvm_vm_ioctl_reset_dirty_pages(struct kvm *kvm)
4574 struct kvm_vcpu *vcpu;
4577 if (!kvm->dirty_ring_size)
4580 mutex_lock(&kvm->slots_lock);
4582 kvm_for_each_vcpu(i, vcpu, kvm)
4583 cleared += kvm_dirty_ring_reset(vcpu->kvm, &vcpu->dirty_ring);
4585 mutex_unlock(&kvm->slots_lock);
4588 kvm_flush_remote_tlbs(kvm);
4593 int __attribute__((weak)) kvm_vm_ioctl_enable_cap(struct kvm *kvm,
4594 struct kvm_enable_cap *cap)
4599 static bool kvm_are_all_memslots_empty(struct kvm *kvm)
4603 lockdep_assert_held(&kvm->slots_lock);
4605 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
4606 if (!kvm_memslots_empty(__kvm_memslots(kvm, i)))
4613 static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm,
4614 struct kvm_enable_cap *cap)
4617 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4618 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2: {
4619 u64 allowed_options = KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE;
4621 if (cap->args[0] & KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE)
4622 allowed_options = KVM_DIRTY_LOG_MANUAL_CAPS;
4624 if (cap->flags || (cap->args[0] & ~allowed_options))
4626 kvm->manual_dirty_log_protect = cap->args[0];
4630 case KVM_CAP_HALT_POLL: {
4631 if (cap->flags || cap->args[0] != (unsigned int)cap->args[0])
4634 kvm->max_halt_poll_ns = cap->args[0];
4637 * Ensure kvm->override_halt_poll_ns does not become visible
4638 * before kvm->max_halt_poll_ns.
4640 * Pairs with the smp_rmb() in kvm_vcpu_max_halt_poll_ns().
4643 kvm->override_halt_poll_ns = true;
4647 case KVM_CAP_DIRTY_LOG_RING:
4648 case KVM_CAP_DIRTY_LOG_RING_ACQ_REL:
4649 if (!kvm_vm_ioctl_check_extension_generic(kvm, cap->cap))
4652 return kvm_vm_ioctl_enable_dirty_log_ring(kvm, cap->args[0]);
4653 case KVM_CAP_DIRTY_LOG_RING_WITH_BITMAP: {
4656 if (!IS_ENABLED(CONFIG_NEED_KVM_DIRTY_RING_WITH_BITMAP) ||
4657 !kvm->dirty_ring_size || cap->flags)
4660 mutex_lock(&kvm->slots_lock);
4663 * For simplicity, allow enabling ring+bitmap if and only if
4664 * there are no memslots, e.g. to ensure all memslots allocate
4665 * a bitmap after the capability is enabled.
4667 if (kvm_are_all_memslots_empty(kvm)) {
4668 kvm->dirty_ring_with_bitmap = true;
4672 mutex_unlock(&kvm->slots_lock);
4677 return kvm_vm_ioctl_enable_cap(kvm, cap);
4681 static ssize_t kvm_vm_stats_read(struct file *file, char __user *user_buffer,
4682 size_t size, loff_t *offset)
4684 struct kvm *kvm = file->private_data;
4686 return kvm_stats_read(kvm->stats_id, &kvm_vm_stats_header,
4687 &kvm_vm_stats_desc[0], &kvm->stat,
4688 sizeof(kvm->stat), user_buffer, size, offset);
4691 static const struct file_operations kvm_vm_stats_fops = {
4692 .read = kvm_vm_stats_read,
4693 .llseek = noop_llseek,
4696 static int kvm_vm_ioctl_get_stats_fd(struct kvm *kvm)
4701 fd = get_unused_fd_flags(O_CLOEXEC);
4705 file = anon_inode_getfile("kvm-vm-stats",
4706 &kvm_vm_stats_fops, kvm, O_RDONLY);
4709 return PTR_ERR(file);
4711 file->f_mode |= FMODE_PREAD;
4712 fd_install(fd, file);
4717 static long kvm_vm_ioctl(struct file *filp,
4718 unsigned int ioctl, unsigned long arg)
4720 struct kvm *kvm = filp->private_data;
4721 void __user *argp = (void __user *)arg;
4724 if (kvm->mm != current->mm || kvm->vm_dead)
4727 case KVM_CREATE_VCPU:
4728 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
4730 case KVM_ENABLE_CAP: {
4731 struct kvm_enable_cap cap;
4734 if (copy_from_user(&cap, argp, sizeof(cap)))
4736 r = kvm_vm_ioctl_enable_cap_generic(kvm, &cap);
4739 case KVM_SET_USER_MEMORY_REGION: {
4740 struct kvm_userspace_memory_region kvm_userspace_mem;
4743 if (copy_from_user(&kvm_userspace_mem, argp,
4744 sizeof(kvm_userspace_mem)))
4747 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
4750 case KVM_GET_DIRTY_LOG: {
4751 struct kvm_dirty_log log;
4754 if (copy_from_user(&log, argp, sizeof(log)))
4756 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
4759 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4760 case KVM_CLEAR_DIRTY_LOG: {
4761 struct kvm_clear_dirty_log log;
4764 if (copy_from_user(&log, argp, sizeof(log)))
4766 r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
4770 #ifdef CONFIG_KVM_MMIO
4771 case KVM_REGISTER_COALESCED_MMIO: {
4772 struct kvm_coalesced_mmio_zone zone;
4775 if (copy_from_user(&zone, argp, sizeof(zone)))
4777 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
4780 case KVM_UNREGISTER_COALESCED_MMIO: {
4781 struct kvm_coalesced_mmio_zone zone;
4784 if (copy_from_user(&zone, argp, sizeof(zone)))
4786 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
4791 struct kvm_irqfd data;
4794 if (copy_from_user(&data, argp, sizeof(data)))
4796 r = kvm_irqfd(kvm, &data);
4799 case KVM_IOEVENTFD: {
4800 struct kvm_ioeventfd data;
4803 if (copy_from_user(&data, argp, sizeof(data)))
4805 r = kvm_ioeventfd(kvm, &data);
4808 #ifdef CONFIG_HAVE_KVM_MSI
4809 case KVM_SIGNAL_MSI: {
4813 if (copy_from_user(&msi, argp, sizeof(msi)))
4815 r = kvm_send_userspace_msi(kvm, &msi);
4819 #ifdef __KVM_HAVE_IRQ_LINE
4820 case KVM_IRQ_LINE_STATUS:
4821 case KVM_IRQ_LINE: {
4822 struct kvm_irq_level irq_event;
4825 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
4828 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
4829 ioctl == KVM_IRQ_LINE_STATUS);
4834 if (ioctl == KVM_IRQ_LINE_STATUS) {
4835 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
4843 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
4844 case KVM_SET_GSI_ROUTING: {
4845 struct kvm_irq_routing routing;
4846 struct kvm_irq_routing __user *urouting;
4847 struct kvm_irq_routing_entry *entries = NULL;
4850 if (copy_from_user(&routing, argp, sizeof(routing)))
4853 if (!kvm_arch_can_set_irq_routing(kvm))
4855 if (routing.nr > KVM_MAX_IRQ_ROUTES)
4861 entries = vmemdup_user(urouting->entries,
4862 array_size(sizeof(*entries),
4864 if (IS_ERR(entries)) {
4865 r = PTR_ERR(entries);
4869 r = kvm_set_irq_routing(kvm, entries, routing.nr,
4874 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
4875 case KVM_CREATE_DEVICE: {
4876 struct kvm_create_device cd;
4879 if (copy_from_user(&cd, argp, sizeof(cd)))
4882 r = kvm_ioctl_create_device(kvm, &cd);
4887 if (copy_to_user(argp, &cd, sizeof(cd)))
4893 case KVM_CHECK_EXTENSION:
4894 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
4896 case KVM_RESET_DIRTY_RINGS:
4897 r = kvm_vm_ioctl_reset_dirty_pages(kvm);
4899 case KVM_GET_STATS_FD:
4900 r = kvm_vm_ioctl_get_stats_fd(kvm);
4903 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
4909 #ifdef CONFIG_KVM_COMPAT
4910 struct compat_kvm_dirty_log {
4914 compat_uptr_t dirty_bitmap; /* one bit per page */
4919 struct compat_kvm_clear_dirty_log {
4924 compat_uptr_t dirty_bitmap; /* one bit per page */
4929 long __weak kvm_arch_vm_compat_ioctl(struct file *filp, unsigned int ioctl,
4935 static long kvm_vm_compat_ioctl(struct file *filp,
4936 unsigned int ioctl, unsigned long arg)
4938 struct kvm *kvm = filp->private_data;
4941 if (kvm->mm != current->mm || kvm->vm_dead)
4944 r = kvm_arch_vm_compat_ioctl(filp, ioctl, arg);
4949 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4950 case KVM_CLEAR_DIRTY_LOG: {
4951 struct compat_kvm_clear_dirty_log compat_log;
4952 struct kvm_clear_dirty_log log;
4954 if (copy_from_user(&compat_log, (void __user *)arg,
4955 sizeof(compat_log)))
4957 log.slot = compat_log.slot;
4958 log.num_pages = compat_log.num_pages;
4959 log.first_page = compat_log.first_page;
4960 log.padding2 = compat_log.padding2;
4961 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
4963 r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
4967 case KVM_GET_DIRTY_LOG: {
4968 struct compat_kvm_dirty_log compat_log;
4969 struct kvm_dirty_log log;
4971 if (copy_from_user(&compat_log, (void __user *)arg,
4972 sizeof(compat_log)))
4974 log.slot = compat_log.slot;
4975 log.padding1 = compat_log.padding1;
4976 log.padding2 = compat_log.padding2;
4977 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
4979 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
4983 r = kvm_vm_ioctl(filp, ioctl, arg);
4989 static const struct file_operations kvm_vm_fops = {
4990 .release = kvm_vm_release,
4991 .unlocked_ioctl = kvm_vm_ioctl,
4992 .llseek = noop_llseek,
4993 KVM_COMPAT(kvm_vm_compat_ioctl),
4996 bool file_is_kvm(struct file *file)
4998 return file && file->f_op == &kvm_vm_fops;
5000 EXPORT_SYMBOL_GPL(file_is_kvm);
5002 static int kvm_dev_ioctl_create_vm(unsigned long type)
5004 char fdname[ITOA_MAX_LEN + 1];
5009 fd = get_unused_fd_flags(O_CLOEXEC);
5013 snprintf(fdname, sizeof(fdname), "%d", fd);
5015 kvm = kvm_create_vm(type, fdname);
5021 file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
5028 * Don't call kvm_put_kvm anymore at this point; file->f_op is
5029 * already set, with ->release() being kvm_vm_release(). In error
5030 * cases it will be called by the final fput(file) and will take
5031 * care of doing kvm_put_kvm(kvm).
5033 kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm);
5035 fd_install(fd, file);
5045 static long kvm_dev_ioctl(struct file *filp,
5046 unsigned int ioctl, unsigned long arg)
5051 case KVM_GET_API_VERSION:
5054 r = KVM_API_VERSION;
5057 r = kvm_dev_ioctl_create_vm(arg);
5059 case KVM_CHECK_EXTENSION:
5060 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
5062 case KVM_GET_VCPU_MMAP_SIZE:
5065 r = PAGE_SIZE; /* struct kvm_run */
5067 r += PAGE_SIZE; /* pio data page */
5069 #ifdef CONFIG_KVM_MMIO
5070 r += PAGE_SIZE; /* coalesced mmio ring page */
5073 case KVM_TRACE_ENABLE:
5074 case KVM_TRACE_PAUSE:
5075 case KVM_TRACE_DISABLE:
5079 return kvm_arch_dev_ioctl(filp, ioctl, arg);
5085 static struct file_operations kvm_chardev_ops = {
5086 .unlocked_ioctl = kvm_dev_ioctl,
5087 .llseek = noop_llseek,
5088 KVM_COMPAT(kvm_dev_ioctl),
5091 static struct miscdevice kvm_dev = {
5097 #ifdef CONFIG_KVM_GENERIC_HARDWARE_ENABLING
5098 __visible bool kvm_rebooting;
5099 EXPORT_SYMBOL_GPL(kvm_rebooting);
5101 static DEFINE_PER_CPU(bool, hardware_enabled);
5102 static int kvm_usage_count;
5104 static int __hardware_enable_nolock(void)
5106 if (__this_cpu_read(hardware_enabled))
5109 if (kvm_arch_hardware_enable()) {
5110 pr_info("kvm: enabling virtualization on CPU%d failed\n",
5111 raw_smp_processor_id());
5115 __this_cpu_write(hardware_enabled, true);
5119 static void hardware_enable_nolock(void *failed)
5121 if (__hardware_enable_nolock())
5125 static int kvm_online_cpu(unsigned int cpu)
5130 * Abort the CPU online process if hardware virtualization cannot
5131 * be enabled. Otherwise running VMs would encounter unrecoverable
5132 * errors when scheduled to this CPU.
5134 mutex_lock(&kvm_lock);
5135 if (kvm_usage_count)
5136 ret = __hardware_enable_nolock();
5137 mutex_unlock(&kvm_lock);
5141 static void hardware_disable_nolock(void *junk)
5144 * Note, hardware_disable_all_nolock() tells all online CPUs to disable
5145 * hardware, not just CPUs that successfully enabled hardware!
5147 if (!__this_cpu_read(hardware_enabled))
5150 kvm_arch_hardware_disable();
5152 __this_cpu_write(hardware_enabled, false);
5155 static int kvm_offline_cpu(unsigned int cpu)
5157 mutex_lock(&kvm_lock);
5158 if (kvm_usage_count)
5159 hardware_disable_nolock(NULL);
5160 mutex_unlock(&kvm_lock);
5164 static void hardware_disable_all_nolock(void)
5166 BUG_ON(!kvm_usage_count);
5169 if (!kvm_usage_count)
5170 on_each_cpu(hardware_disable_nolock, NULL, 1);
5173 static void hardware_disable_all(void)
5176 mutex_lock(&kvm_lock);
5177 hardware_disable_all_nolock();
5178 mutex_unlock(&kvm_lock);
5182 static int hardware_enable_all(void)
5184 atomic_t failed = ATOMIC_INIT(0);
5188 * When onlining a CPU, cpu_online_mask is set before kvm_online_cpu()
5189 * is called, and so on_each_cpu() between them includes the CPU that
5190 * is being onlined. As a result, hardware_enable_nolock() may get
5191 * invoked before kvm_online_cpu(), which also enables hardware if the
5192 * usage count is non-zero. Disable CPU hotplug to avoid attempting to
5193 * enable hardware multiple times.
5196 mutex_lock(&kvm_lock);
5199 if (kvm_usage_count == 1) {
5200 on_each_cpu(hardware_enable_nolock, &failed, 1);
5202 if (atomic_read(&failed)) {
5203 hardware_disable_all_nolock();
5208 mutex_unlock(&kvm_lock);
5214 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
5218 * Some (well, at least mine) BIOSes hang on reboot if
5221 * And Intel TXT required VMX off for all cpu when system shutdown.
5223 pr_info("kvm: exiting hardware virtualization\n");
5224 kvm_rebooting = true;
5225 on_each_cpu(hardware_disable_nolock, NULL, 1);
5229 static struct notifier_block kvm_reboot_notifier = {
5230 .notifier_call = kvm_reboot,
5234 static int kvm_suspend(void)
5237 * Secondary CPUs and CPU hotplug are disabled across the suspend/resume
5238 * callbacks, i.e. no need to acquire kvm_lock to ensure the usage count
5239 * is stable. Assert that kvm_lock is not held to ensure the system
5240 * isn't suspended while KVM is enabling hardware. Hardware enabling
5241 * can be preempted, but the task cannot be frozen until it has dropped
5242 * all locks (userspace tasks are frozen via a fake signal).
5244 lockdep_assert_not_held(&kvm_lock);
5245 lockdep_assert_irqs_disabled();
5247 if (kvm_usage_count)
5248 hardware_disable_nolock(NULL);
5252 static void kvm_resume(void)
5254 lockdep_assert_not_held(&kvm_lock);
5255 lockdep_assert_irqs_disabled();
5257 if (kvm_usage_count)
5258 WARN_ON_ONCE(__hardware_enable_nolock());
5261 static struct syscore_ops kvm_syscore_ops = {
5262 .suspend = kvm_suspend,
5263 .resume = kvm_resume,
5265 #else /* CONFIG_KVM_GENERIC_HARDWARE_ENABLING */
5266 static int hardware_enable_all(void)
5271 static void hardware_disable_all(void)
5275 #endif /* CONFIG_KVM_GENERIC_HARDWARE_ENABLING */
5277 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
5281 for (i = 0; i < bus->dev_count; i++) {
5282 struct kvm_io_device *pos = bus->range[i].dev;
5284 kvm_iodevice_destructor(pos);
5289 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
5290 const struct kvm_io_range *r2)
5292 gpa_t addr1 = r1->addr;
5293 gpa_t addr2 = r2->addr;
5298 /* If r2->len == 0, match the exact address. If r2->len != 0,
5299 * accept any overlapping write. Any order is acceptable for
5300 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
5301 * we process all of them.
5314 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
5316 return kvm_io_bus_cmp(p1, p2);
5319 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
5320 gpa_t addr, int len)
5322 struct kvm_io_range *range, key;
5325 key = (struct kvm_io_range) {
5330 range = bsearch(&key, bus->range, bus->dev_count,
5331 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
5335 off = range - bus->range;
5337 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
5343 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
5344 struct kvm_io_range *range, const void *val)
5348 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
5352 while (idx < bus->dev_count &&
5353 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
5354 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
5363 /* kvm_io_bus_write - called under kvm->slots_lock */
5364 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
5365 int len, const void *val)
5367 struct kvm_io_bus *bus;
5368 struct kvm_io_range range;
5371 range = (struct kvm_io_range) {
5376 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
5379 r = __kvm_io_bus_write(vcpu, bus, &range, val);
5380 return r < 0 ? r : 0;
5382 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
5384 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
5385 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
5386 gpa_t addr, int len, const void *val, long cookie)
5388 struct kvm_io_bus *bus;
5389 struct kvm_io_range range;
5391 range = (struct kvm_io_range) {
5396 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
5400 /* First try the device referenced by cookie. */
5401 if ((cookie >= 0) && (cookie < bus->dev_count) &&
5402 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
5403 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
5408 * cookie contained garbage; fall back to search and return the
5409 * correct cookie value.
5411 return __kvm_io_bus_write(vcpu, bus, &range, val);
5414 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
5415 struct kvm_io_range *range, void *val)
5419 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
5423 while (idx < bus->dev_count &&
5424 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
5425 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
5434 /* kvm_io_bus_read - called under kvm->slots_lock */
5435 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
5438 struct kvm_io_bus *bus;
5439 struct kvm_io_range range;
5442 range = (struct kvm_io_range) {
5447 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
5450 r = __kvm_io_bus_read(vcpu, bus, &range, val);
5451 return r < 0 ? r : 0;
5454 /* Caller must hold slots_lock. */
5455 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
5456 int len, struct kvm_io_device *dev)
5459 struct kvm_io_bus *new_bus, *bus;
5460 struct kvm_io_range range;
5462 bus = kvm_get_bus(kvm, bus_idx);
5466 /* exclude ioeventfd which is limited by maximum fd */
5467 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
5470 new_bus = kmalloc(struct_size(bus, range, bus->dev_count + 1),
5471 GFP_KERNEL_ACCOUNT);
5475 range = (struct kvm_io_range) {
5481 for (i = 0; i < bus->dev_count; i++)
5482 if (kvm_io_bus_cmp(&bus->range[i], &range) > 0)
5485 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
5486 new_bus->dev_count++;
5487 new_bus->range[i] = range;
5488 memcpy(new_bus->range + i + 1, bus->range + i,
5489 (bus->dev_count - i) * sizeof(struct kvm_io_range));
5490 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
5491 synchronize_srcu_expedited(&kvm->srcu);
5497 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
5498 struct kvm_io_device *dev)
5501 struct kvm_io_bus *new_bus, *bus;
5503 lockdep_assert_held(&kvm->slots_lock);
5505 bus = kvm_get_bus(kvm, bus_idx);
5509 for (i = 0; i < bus->dev_count; i++) {
5510 if (bus->range[i].dev == dev) {
5515 if (i == bus->dev_count)
5518 new_bus = kmalloc(struct_size(bus, range, bus->dev_count - 1),
5519 GFP_KERNEL_ACCOUNT);
5521 memcpy(new_bus, bus, struct_size(bus, range, i));
5522 new_bus->dev_count--;
5523 memcpy(new_bus->range + i, bus->range + i + 1,
5524 flex_array_size(new_bus, range, new_bus->dev_count - i));
5527 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
5528 synchronize_srcu_expedited(&kvm->srcu);
5530 /* Destroy the old bus _after_ installing the (null) bus. */
5532 pr_err("kvm: failed to shrink bus, removing it completely\n");
5533 for (j = 0; j < bus->dev_count; j++) {
5536 kvm_iodevice_destructor(bus->range[j].dev);
5541 return new_bus ? 0 : -ENOMEM;
5544 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
5547 struct kvm_io_bus *bus;
5548 int dev_idx, srcu_idx;
5549 struct kvm_io_device *iodev = NULL;
5551 srcu_idx = srcu_read_lock(&kvm->srcu);
5553 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
5557 dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
5561 iodev = bus->range[dev_idx].dev;
5564 srcu_read_unlock(&kvm->srcu, srcu_idx);
5568 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
5570 static int kvm_debugfs_open(struct inode *inode, struct file *file,
5571 int (*get)(void *, u64 *), int (*set)(void *, u64),
5575 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
5579 * The debugfs files are a reference to the kvm struct which
5580 * is still valid when kvm_destroy_vm is called. kvm_get_kvm_safe
5581 * avoids the race between open and the removal of the debugfs directory.
5583 if (!kvm_get_kvm_safe(stat_data->kvm))
5586 ret = simple_attr_open(inode, file, get,
5587 kvm_stats_debugfs_mode(stat_data->desc) & 0222
5590 kvm_put_kvm(stat_data->kvm);
5595 static int kvm_debugfs_release(struct inode *inode, struct file *file)
5597 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
5600 simple_attr_release(inode, file);
5601 kvm_put_kvm(stat_data->kvm);
5606 static int kvm_get_stat_per_vm(struct kvm *kvm, size_t offset, u64 *val)
5608 *val = *(u64 *)((void *)(&kvm->stat) + offset);
5613 static int kvm_clear_stat_per_vm(struct kvm *kvm, size_t offset)
5615 *(u64 *)((void *)(&kvm->stat) + offset) = 0;
5620 static int kvm_get_stat_per_vcpu(struct kvm *kvm, size_t offset, u64 *val)
5623 struct kvm_vcpu *vcpu;
5627 kvm_for_each_vcpu(i, vcpu, kvm)
5628 *val += *(u64 *)((void *)(&vcpu->stat) + offset);
5633 static int kvm_clear_stat_per_vcpu(struct kvm *kvm, size_t offset)
5636 struct kvm_vcpu *vcpu;
5638 kvm_for_each_vcpu(i, vcpu, kvm)
5639 *(u64 *)((void *)(&vcpu->stat) + offset) = 0;
5644 static int kvm_stat_data_get(void *data, u64 *val)
5647 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
5649 switch (stat_data->kind) {
5651 r = kvm_get_stat_per_vm(stat_data->kvm,
5652 stat_data->desc->desc.offset, val);
5655 r = kvm_get_stat_per_vcpu(stat_data->kvm,
5656 stat_data->desc->desc.offset, val);
5663 static int kvm_stat_data_clear(void *data, u64 val)
5666 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
5671 switch (stat_data->kind) {
5673 r = kvm_clear_stat_per_vm(stat_data->kvm,
5674 stat_data->desc->desc.offset);
5677 r = kvm_clear_stat_per_vcpu(stat_data->kvm,
5678 stat_data->desc->desc.offset);
5685 static int kvm_stat_data_open(struct inode *inode, struct file *file)
5687 __simple_attr_check_format("%llu\n", 0ull);
5688 return kvm_debugfs_open(inode, file, kvm_stat_data_get,
5689 kvm_stat_data_clear, "%llu\n");
5692 static const struct file_operations stat_fops_per_vm = {
5693 .owner = THIS_MODULE,
5694 .open = kvm_stat_data_open,
5695 .release = kvm_debugfs_release,
5696 .read = simple_attr_read,
5697 .write = simple_attr_write,
5698 .llseek = no_llseek,
5701 static int vm_stat_get(void *_offset, u64 *val)
5703 unsigned offset = (long)_offset;
5708 mutex_lock(&kvm_lock);
5709 list_for_each_entry(kvm, &vm_list, vm_list) {
5710 kvm_get_stat_per_vm(kvm, offset, &tmp_val);
5713 mutex_unlock(&kvm_lock);
5717 static int vm_stat_clear(void *_offset, u64 val)
5719 unsigned offset = (long)_offset;
5725 mutex_lock(&kvm_lock);
5726 list_for_each_entry(kvm, &vm_list, vm_list) {
5727 kvm_clear_stat_per_vm(kvm, offset);
5729 mutex_unlock(&kvm_lock);
5734 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n");
5735 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_readonly_fops, vm_stat_get, NULL, "%llu\n");
5737 static int vcpu_stat_get(void *_offset, u64 *val)
5739 unsigned offset = (long)_offset;
5744 mutex_lock(&kvm_lock);
5745 list_for_each_entry(kvm, &vm_list, vm_list) {
5746 kvm_get_stat_per_vcpu(kvm, offset, &tmp_val);
5749 mutex_unlock(&kvm_lock);
5753 static int vcpu_stat_clear(void *_offset, u64 val)
5755 unsigned offset = (long)_offset;
5761 mutex_lock(&kvm_lock);
5762 list_for_each_entry(kvm, &vm_list, vm_list) {
5763 kvm_clear_stat_per_vcpu(kvm, offset);
5765 mutex_unlock(&kvm_lock);
5770 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear,
5772 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_readonly_fops, vcpu_stat_get, NULL, "%llu\n");
5774 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm)
5776 struct kobj_uevent_env *env;
5777 unsigned long long created, active;
5779 if (!kvm_dev.this_device || !kvm)
5782 mutex_lock(&kvm_lock);
5783 if (type == KVM_EVENT_CREATE_VM) {
5784 kvm_createvm_count++;
5786 } else if (type == KVM_EVENT_DESTROY_VM) {
5789 created = kvm_createvm_count;
5790 active = kvm_active_vms;
5791 mutex_unlock(&kvm_lock);
5793 env = kzalloc(sizeof(*env), GFP_KERNEL_ACCOUNT);
5797 add_uevent_var(env, "CREATED=%llu", created);
5798 add_uevent_var(env, "COUNT=%llu", active);
5800 if (type == KVM_EVENT_CREATE_VM) {
5801 add_uevent_var(env, "EVENT=create");
5802 kvm->userspace_pid = task_pid_nr(current);
5803 } else if (type == KVM_EVENT_DESTROY_VM) {
5804 add_uevent_var(env, "EVENT=destroy");
5806 add_uevent_var(env, "PID=%d", kvm->userspace_pid);
5808 if (!IS_ERR(kvm->debugfs_dentry)) {
5809 char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL_ACCOUNT);
5812 tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX);
5814 add_uevent_var(env, "STATS_PATH=%s", tmp);
5818 /* no need for checks, since we are adding at most only 5 keys */
5819 env->envp[env->envp_idx++] = NULL;
5820 kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp);
5824 static void kvm_init_debug(void)
5826 const struct file_operations *fops;
5827 const struct _kvm_stats_desc *pdesc;
5830 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
5832 for (i = 0; i < kvm_vm_stats_header.num_desc; ++i) {
5833 pdesc = &kvm_vm_stats_desc[i];
5834 if (kvm_stats_debugfs_mode(pdesc) & 0222)
5835 fops = &vm_stat_fops;
5837 fops = &vm_stat_readonly_fops;
5838 debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
5840 (void *)(long)pdesc->desc.offset, fops);
5843 for (i = 0; i < kvm_vcpu_stats_header.num_desc; ++i) {
5844 pdesc = &kvm_vcpu_stats_desc[i];
5845 if (kvm_stats_debugfs_mode(pdesc) & 0222)
5846 fops = &vcpu_stat_fops;
5848 fops = &vcpu_stat_readonly_fops;
5849 debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
5851 (void *)(long)pdesc->desc.offset, fops);
5856 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
5858 return container_of(pn, struct kvm_vcpu, preempt_notifier);
5861 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
5863 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
5865 WRITE_ONCE(vcpu->preempted, false);
5866 WRITE_ONCE(vcpu->ready, false);
5868 __this_cpu_write(kvm_running_vcpu, vcpu);
5869 kvm_arch_sched_in(vcpu, cpu);
5870 kvm_arch_vcpu_load(vcpu, cpu);
5873 static void kvm_sched_out(struct preempt_notifier *pn,
5874 struct task_struct *next)
5876 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
5878 if (current->on_rq) {
5879 WRITE_ONCE(vcpu->preempted, true);
5880 WRITE_ONCE(vcpu->ready, true);
5882 kvm_arch_vcpu_put(vcpu);
5883 __this_cpu_write(kvm_running_vcpu, NULL);
5887 * kvm_get_running_vcpu - get the vcpu running on the current CPU.
5889 * We can disable preemption locally around accessing the per-CPU variable,
5890 * and use the resolved vcpu pointer after enabling preemption again,
5891 * because even if the current thread is migrated to another CPU, reading
5892 * the per-CPU value later will give us the same value as we update the
5893 * per-CPU variable in the preempt notifier handlers.
5895 struct kvm_vcpu *kvm_get_running_vcpu(void)
5897 struct kvm_vcpu *vcpu;
5900 vcpu = __this_cpu_read(kvm_running_vcpu);
5905 EXPORT_SYMBOL_GPL(kvm_get_running_vcpu);
5908 * kvm_get_running_vcpus - get the per-CPU array of currently running vcpus.
5910 struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void)
5912 return &kvm_running_vcpu;
5915 #ifdef CONFIG_GUEST_PERF_EVENTS
5916 static unsigned int kvm_guest_state(void)
5918 struct kvm_vcpu *vcpu = kvm_get_running_vcpu();
5921 if (!kvm_arch_pmi_in_guest(vcpu))
5924 state = PERF_GUEST_ACTIVE;
5925 if (!kvm_arch_vcpu_in_kernel(vcpu))
5926 state |= PERF_GUEST_USER;
5931 static unsigned long kvm_guest_get_ip(void)
5933 struct kvm_vcpu *vcpu = kvm_get_running_vcpu();
5935 /* Retrieving the IP must be guarded by a call to kvm_guest_state(). */
5936 if (WARN_ON_ONCE(!kvm_arch_pmi_in_guest(vcpu)))
5939 return kvm_arch_vcpu_get_ip(vcpu);
5942 static struct perf_guest_info_callbacks kvm_guest_cbs = {
5943 .state = kvm_guest_state,
5944 .get_ip = kvm_guest_get_ip,
5945 .handle_intel_pt_intr = NULL,
5948 void kvm_register_perf_callbacks(unsigned int (*pt_intr_handler)(void))
5950 kvm_guest_cbs.handle_intel_pt_intr = pt_intr_handler;
5951 perf_register_guest_info_callbacks(&kvm_guest_cbs);
5953 void kvm_unregister_perf_callbacks(void)
5955 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
5959 int kvm_init(unsigned vcpu_size, unsigned vcpu_align, struct module *module)
5964 #ifdef CONFIG_KVM_GENERIC_HARDWARE_ENABLING
5965 r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_ONLINE, "kvm/cpu:online",
5966 kvm_online_cpu, kvm_offline_cpu);
5970 register_reboot_notifier(&kvm_reboot_notifier);
5971 register_syscore_ops(&kvm_syscore_ops);
5974 /* A kmem cache lets us meet the alignment requirements of fx_save. */
5976 vcpu_align = __alignof__(struct kvm_vcpu);
5978 kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align,
5980 offsetof(struct kvm_vcpu, arch),
5981 offsetofend(struct kvm_vcpu, stats_id)
5982 - offsetof(struct kvm_vcpu, arch),
5984 if (!kvm_vcpu_cache) {
5986 goto err_vcpu_cache;
5989 for_each_possible_cpu(cpu) {
5990 if (!alloc_cpumask_var_node(&per_cpu(cpu_kick_mask, cpu),
5991 GFP_KERNEL, cpu_to_node(cpu))) {
5993 goto err_cpu_kick_mask;
5997 r = kvm_irqfd_init();
6001 r = kvm_async_pf_init();
6005 kvm_chardev_ops.owner = module;
6007 kvm_preempt_ops.sched_in = kvm_sched_in;
6008 kvm_preempt_ops.sched_out = kvm_sched_out;
6012 r = kvm_vfio_ops_init();
6013 if (WARN_ON_ONCE(r))
6017 * Registration _must_ be the very last thing done, as this exposes
6018 * /dev/kvm to userspace, i.e. all infrastructure must be setup!
6020 r = misc_register(&kvm_dev);
6022 pr_err("kvm: misc device register failed\n");
6029 kvm_vfio_ops_exit();
6031 kvm_async_pf_deinit();
6036 for_each_possible_cpu(cpu)
6037 free_cpumask_var(per_cpu(cpu_kick_mask, cpu));
6038 kmem_cache_destroy(kvm_vcpu_cache);
6040 #ifdef CONFIG_KVM_GENERIC_HARDWARE_ENABLING
6041 unregister_syscore_ops(&kvm_syscore_ops);
6042 unregister_reboot_notifier(&kvm_reboot_notifier);
6043 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_ONLINE);
6047 EXPORT_SYMBOL_GPL(kvm_init);
6054 * Note, unregistering /dev/kvm doesn't strictly need to come first,
6055 * fops_get(), a.k.a. try_module_get(), prevents acquiring references
6056 * to KVM while the module is being stopped.
6058 misc_deregister(&kvm_dev);
6060 debugfs_remove_recursive(kvm_debugfs_dir);
6061 for_each_possible_cpu(cpu)
6062 free_cpumask_var(per_cpu(cpu_kick_mask, cpu));
6063 kmem_cache_destroy(kvm_vcpu_cache);
6064 kvm_vfio_ops_exit();
6065 kvm_async_pf_deinit();
6066 #ifdef CONFIG_KVM_GENERIC_HARDWARE_ENABLING
6067 unregister_syscore_ops(&kvm_syscore_ops);
6068 unregister_reboot_notifier(&kvm_reboot_notifier);
6069 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_ONLINE);
6073 EXPORT_SYMBOL_GPL(kvm_exit);
6075 struct kvm_vm_worker_thread_context {
6077 struct task_struct *parent;
6078 struct completion init_done;
6079 kvm_vm_thread_fn_t thread_fn;
6084 static int kvm_vm_worker_thread(void *context)
6087 * The init_context is allocated on the stack of the parent thread, so
6088 * we have to locally copy anything that is needed beyond initialization
6090 struct kvm_vm_worker_thread_context *init_context = context;
6091 struct task_struct *parent;
6092 struct kvm *kvm = init_context->kvm;
6093 kvm_vm_thread_fn_t thread_fn = init_context->thread_fn;
6094 uintptr_t data = init_context->data;
6097 err = kthread_park(current);
6098 /* kthread_park(current) is never supposed to return an error */
6103 err = cgroup_attach_task_all(init_context->parent, current);
6105 kvm_err("%s: cgroup_attach_task_all failed with err %d\n",
6110 set_user_nice(current, task_nice(init_context->parent));
6113 init_context->err = err;
6114 complete(&init_context->init_done);
6115 init_context = NULL;
6120 /* Wait to be woken up by the spawner before proceeding. */
6123 if (!kthread_should_stop())
6124 err = thread_fn(kvm, data);
6128 * Move kthread back to its original cgroup to prevent it lingering in
6129 * the cgroup of the VM process, after the latter finishes its
6132 * kthread_stop() waits on the 'exited' completion condition which is
6133 * set in exit_mm(), via mm_release(), in do_exit(). However, the
6134 * kthread is removed from the cgroup in the cgroup_exit() which is
6135 * called after the exit_mm(). This causes the kthread_stop() to return
6136 * before the kthread actually quits the cgroup.
6139 parent = rcu_dereference(current->real_parent);
6140 get_task_struct(parent);
6142 cgroup_attach_task_all(parent, current);
6143 put_task_struct(parent);
6148 int kvm_vm_create_worker_thread(struct kvm *kvm, kvm_vm_thread_fn_t thread_fn,
6149 uintptr_t data, const char *name,
6150 struct task_struct **thread_ptr)
6152 struct kvm_vm_worker_thread_context init_context = {};
6153 struct task_struct *thread;
6156 init_context.kvm = kvm;
6157 init_context.parent = current;
6158 init_context.thread_fn = thread_fn;
6159 init_context.data = data;
6160 init_completion(&init_context.init_done);
6162 thread = kthread_run(kvm_vm_worker_thread, &init_context,
6163 "%s-%d", name, task_pid_nr(current));
6165 return PTR_ERR(thread);
6167 /* kthread_run is never supposed to return NULL */
6168 WARN_ON(thread == NULL);
6170 wait_for_completion(&init_context.init_done);
6172 if (!init_context.err)
6173 *thread_ptr = thread;
6175 return init_context.err;