1 // SPDX-License-Identifier: GPL-2.0-only
3 * Kernel-based Virtual Machine driver for Linux
5 * This module enables machines with Intel VT-x extensions to run virtual
6 * machines without emulation or binary translation.
8 * Copyright (C) 2006 Qumranet, Inc.
9 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
12 * Avi Kivity <avi@qumranet.com>
13 * Yaniv Kamay <yaniv@qumranet.com>
16 #include <kvm/iodev.h>
18 #include <linux/kvm_host.h>
19 #include <linux/kvm.h>
20 #include <linux/module.h>
21 #include <linux/errno.h>
22 #include <linux/percpu.h>
24 #include <linux/miscdevice.h>
25 #include <linux/vmalloc.h>
26 #include <linux/reboot.h>
27 #include <linux/debugfs.h>
28 #include <linux/highmem.h>
29 #include <linux/file.h>
30 #include <linux/syscore_ops.h>
31 #include <linux/cpu.h>
32 #include <linux/sched/signal.h>
33 #include <linux/sched/mm.h>
34 #include <linux/sched/stat.h>
35 #include <linux/cpumask.h>
36 #include <linux/smp.h>
37 #include <linux/anon_inodes.h>
38 #include <linux/profile.h>
39 #include <linux/kvm_para.h>
40 #include <linux/pagemap.h>
41 #include <linux/mman.h>
42 #include <linux/swap.h>
43 #include <linux/bitops.h>
44 #include <linux/spinlock.h>
45 #include <linux/compat.h>
46 #include <linux/srcu.h>
47 #include <linux/hugetlb.h>
48 #include <linux/slab.h>
49 #include <linux/sort.h>
50 #include <linux/bsearch.h>
52 #include <linux/lockdep.h>
53 #include <linux/kthread.h>
54 #include <linux/suspend.h>
56 #include <asm/processor.h>
57 #include <asm/ioctl.h>
58 #include <linux/uaccess.h>
60 #include "coalesced_mmio.h"
65 #include <trace/events/ipi.h>
67 #define CREATE_TRACE_POINTS
68 #include <trace/events/kvm.h>
70 #include <linux/kvm_dirty_ring.h>
73 /* Worst case buffer size needed for holding an integer. */
74 #define ITOA_MAX_LEN 12
76 MODULE_AUTHOR("Qumranet");
77 MODULE_LICENSE("GPL");
79 /* Architectures should define their poll value according to the halt latency */
80 unsigned int halt_poll_ns = KVM_HALT_POLL_NS_DEFAULT;
81 module_param(halt_poll_ns, uint, 0644);
82 EXPORT_SYMBOL_GPL(halt_poll_ns);
84 /* Default doubles per-vcpu halt_poll_ns. */
85 unsigned int halt_poll_ns_grow = 2;
86 module_param(halt_poll_ns_grow, uint, 0644);
87 EXPORT_SYMBOL_GPL(halt_poll_ns_grow);
89 /* The start value to grow halt_poll_ns from */
90 unsigned int halt_poll_ns_grow_start = 10000; /* 10us */
91 module_param(halt_poll_ns_grow_start, uint, 0644);
92 EXPORT_SYMBOL_GPL(halt_poll_ns_grow_start);
94 /* Default resets per-vcpu halt_poll_ns . */
95 unsigned int halt_poll_ns_shrink;
96 module_param(halt_poll_ns_shrink, uint, 0644);
97 EXPORT_SYMBOL_GPL(halt_poll_ns_shrink);
102 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
105 DEFINE_MUTEX(kvm_lock);
108 static struct kmem_cache *kvm_vcpu_cache;
110 static __read_mostly struct preempt_ops kvm_preempt_ops;
111 static DEFINE_PER_CPU(struct kvm_vcpu *, kvm_running_vcpu);
113 struct dentry *kvm_debugfs_dir;
114 EXPORT_SYMBOL_GPL(kvm_debugfs_dir);
116 static const struct file_operations stat_fops_per_vm;
118 static struct file_operations kvm_chardev_ops;
120 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
122 #ifdef CONFIG_KVM_COMPAT
123 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
125 #define KVM_COMPAT(c) .compat_ioctl = (c)
128 * For architectures that don't implement a compat infrastructure,
129 * adopt a double line of defense:
130 * - Prevent a compat task from opening /dev/kvm
131 * - If the open has been done by a 64bit task, and the KVM fd
132 * passed to a compat task, let the ioctls fail.
134 static long kvm_no_compat_ioctl(struct file *file, unsigned int ioctl,
135 unsigned long arg) { return -EINVAL; }
137 static int kvm_no_compat_open(struct inode *inode, struct file *file)
139 return is_compat_task() ? -ENODEV : 0;
141 #define KVM_COMPAT(c) .compat_ioctl = kvm_no_compat_ioctl, \
142 .open = kvm_no_compat_open
144 static int hardware_enable_all(void);
145 static void hardware_disable_all(void);
147 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
149 #define KVM_EVENT_CREATE_VM 0
150 #define KVM_EVENT_DESTROY_VM 1
151 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm);
152 static unsigned long long kvm_createvm_count;
153 static unsigned long long kvm_active_vms;
155 static DEFINE_PER_CPU(cpumask_var_t, cpu_kick_mask);
157 __weak void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
158 unsigned long start, unsigned long end)
162 __weak void kvm_arch_guest_memory_reclaimed(struct kvm *kvm)
166 bool kvm_is_zone_device_page(struct page *page)
169 * The metadata used by is_zone_device_page() to determine whether or
170 * not a page is ZONE_DEVICE is guaranteed to be valid if and only if
171 * the device has been pinned, e.g. by get_user_pages(). WARN if the
172 * page_count() is zero to help detect bad usage of this helper.
174 if (WARN_ON_ONCE(!page_count(page)))
177 return is_zone_device_page(page);
181 * Returns a 'struct page' if the pfn is "valid" and backed by a refcounted
182 * page, NULL otherwise. Note, the list of refcounted PG_reserved page types
183 * is likely incomplete, it has been compiled purely through people wanting to
184 * back guest with a certain type of memory and encountering issues.
186 struct page *kvm_pfn_to_refcounted_page(kvm_pfn_t pfn)
193 page = pfn_to_page(pfn);
194 if (!PageReserved(page))
197 /* The ZERO_PAGE(s) is marked PG_reserved, but is refcounted. */
198 if (is_zero_pfn(pfn))
202 * ZONE_DEVICE pages currently set PG_reserved, but from a refcounting
203 * perspective they are "normal" pages, albeit with slightly different
206 if (kvm_is_zone_device_page(page))
213 * Switches to specified vcpu, until a matching vcpu_put()
215 void vcpu_load(struct kvm_vcpu *vcpu)
219 __this_cpu_write(kvm_running_vcpu, vcpu);
220 preempt_notifier_register(&vcpu->preempt_notifier);
221 kvm_arch_vcpu_load(vcpu, cpu);
224 EXPORT_SYMBOL_GPL(vcpu_load);
226 void vcpu_put(struct kvm_vcpu *vcpu)
229 kvm_arch_vcpu_put(vcpu);
230 preempt_notifier_unregister(&vcpu->preempt_notifier);
231 __this_cpu_write(kvm_running_vcpu, NULL);
234 EXPORT_SYMBOL_GPL(vcpu_put);
236 /* TODO: merge with kvm_arch_vcpu_should_kick */
237 static bool kvm_request_needs_ipi(struct kvm_vcpu *vcpu, unsigned req)
239 int mode = kvm_vcpu_exiting_guest_mode(vcpu);
242 * We need to wait for the VCPU to reenable interrupts and get out of
243 * READING_SHADOW_PAGE_TABLES mode.
245 if (req & KVM_REQUEST_WAIT)
246 return mode != OUTSIDE_GUEST_MODE;
249 * Need to kick a running VCPU, but otherwise there is nothing to do.
251 return mode == IN_GUEST_MODE;
254 static void ack_kick(void *_completed)
258 static inline bool kvm_kick_many_cpus(struct cpumask *cpus, bool wait)
260 if (cpumask_empty(cpus))
263 smp_call_function_many(cpus, ack_kick, NULL, wait);
267 static void kvm_make_vcpu_request(struct kvm_vcpu *vcpu, unsigned int req,
268 struct cpumask *tmp, int current_cpu)
272 if (likely(!(req & KVM_REQUEST_NO_ACTION)))
273 __kvm_make_request(req, vcpu);
275 if (!(req & KVM_REQUEST_NO_WAKEUP) && kvm_vcpu_wake_up(vcpu))
279 * Note, the vCPU could get migrated to a different pCPU at any point
280 * after kvm_request_needs_ipi(), which could result in sending an IPI
281 * to the previous pCPU. But, that's OK because the purpose of the IPI
282 * is to ensure the vCPU returns to OUTSIDE_GUEST_MODE, which is
283 * satisfied if the vCPU migrates. Entering READING_SHADOW_PAGE_TABLES
284 * after this point is also OK, as the requirement is only that KVM wait
285 * for vCPUs that were reading SPTEs _before_ any changes were
286 * finalized. See kvm_vcpu_kick() for more details on handling requests.
288 if (kvm_request_needs_ipi(vcpu, req)) {
289 cpu = READ_ONCE(vcpu->cpu);
290 if (cpu != -1 && cpu != current_cpu)
291 __cpumask_set_cpu(cpu, tmp);
295 bool kvm_make_vcpus_request_mask(struct kvm *kvm, unsigned int req,
296 unsigned long *vcpu_bitmap)
298 struct kvm_vcpu *vcpu;
299 struct cpumask *cpus;
305 cpus = this_cpu_cpumask_var_ptr(cpu_kick_mask);
308 for_each_set_bit(i, vcpu_bitmap, KVM_MAX_VCPUS) {
309 vcpu = kvm_get_vcpu(kvm, i);
312 kvm_make_vcpu_request(vcpu, req, cpus, me);
315 called = kvm_kick_many_cpus(cpus, !!(req & KVM_REQUEST_WAIT));
321 bool kvm_make_all_cpus_request_except(struct kvm *kvm, unsigned int req,
322 struct kvm_vcpu *except)
324 struct kvm_vcpu *vcpu;
325 struct cpumask *cpus;
332 cpus = this_cpu_cpumask_var_ptr(cpu_kick_mask);
335 kvm_for_each_vcpu(i, vcpu, kvm) {
338 kvm_make_vcpu_request(vcpu, req, cpus, me);
341 called = kvm_kick_many_cpus(cpus, !!(req & KVM_REQUEST_WAIT));
347 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
349 return kvm_make_all_cpus_request_except(kvm, req, NULL);
351 EXPORT_SYMBOL_GPL(kvm_make_all_cpus_request);
353 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
354 void kvm_flush_remote_tlbs(struct kvm *kvm)
356 ++kvm->stat.generic.remote_tlb_flush_requests;
359 * We want to publish modifications to the page tables before reading
360 * mode. Pairs with a memory barrier in arch-specific code.
361 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
362 * and smp_mb in walk_shadow_page_lockless_begin/end.
363 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
365 * There is already an smp_mb__after_atomic() before
366 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
369 if (!kvm_arch_flush_remote_tlb(kvm)
370 || kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
371 ++kvm->stat.generic.remote_tlb_flush;
373 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
376 static void kvm_flush_shadow_all(struct kvm *kvm)
378 kvm_arch_flush_shadow_all(kvm);
379 kvm_arch_guest_memory_reclaimed(kvm);
382 #ifdef KVM_ARCH_NR_OBJS_PER_MEMORY_CACHE
383 static inline void *mmu_memory_cache_alloc_obj(struct kvm_mmu_memory_cache *mc,
386 gfp_flags |= mc->gfp_zero;
389 return kmem_cache_alloc(mc->kmem_cache, gfp_flags);
391 return (void *)__get_free_page(gfp_flags);
394 int __kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache *mc, int capacity, int min)
396 gfp_t gfp = mc->gfp_custom ? mc->gfp_custom : GFP_KERNEL_ACCOUNT;
399 if (mc->nobjs >= min)
402 if (unlikely(!mc->objects)) {
403 if (WARN_ON_ONCE(!capacity))
406 mc->objects = kvmalloc_array(sizeof(void *), capacity, gfp);
410 mc->capacity = capacity;
413 /* It is illegal to request a different capacity across topups. */
414 if (WARN_ON_ONCE(mc->capacity != capacity))
417 while (mc->nobjs < mc->capacity) {
418 obj = mmu_memory_cache_alloc_obj(mc, gfp);
420 return mc->nobjs >= min ? 0 : -ENOMEM;
421 mc->objects[mc->nobjs++] = obj;
426 int kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache *mc, int min)
428 return __kvm_mmu_topup_memory_cache(mc, KVM_ARCH_NR_OBJS_PER_MEMORY_CACHE, min);
431 int kvm_mmu_memory_cache_nr_free_objects(struct kvm_mmu_memory_cache *mc)
436 void kvm_mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc)
440 kmem_cache_free(mc->kmem_cache, mc->objects[--mc->nobjs]);
442 free_page((unsigned long)mc->objects[--mc->nobjs]);
451 void *kvm_mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc)
455 if (WARN_ON(!mc->nobjs))
456 p = mmu_memory_cache_alloc_obj(mc, GFP_ATOMIC | __GFP_ACCOUNT);
458 p = mc->objects[--mc->nobjs];
464 static void kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
466 mutex_init(&vcpu->mutex);
471 #ifndef __KVM_HAVE_ARCH_WQP
472 rcuwait_init(&vcpu->wait);
474 kvm_async_pf_vcpu_init(vcpu);
476 kvm_vcpu_set_in_spin_loop(vcpu, false);
477 kvm_vcpu_set_dy_eligible(vcpu, false);
478 vcpu->preempted = false;
480 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
481 vcpu->last_used_slot = NULL;
483 /* Fill the stats id string for the vcpu */
484 snprintf(vcpu->stats_id, sizeof(vcpu->stats_id), "kvm-%d/vcpu-%d",
485 task_pid_nr(current), id);
488 static void kvm_vcpu_destroy(struct kvm_vcpu *vcpu)
490 kvm_arch_vcpu_destroy(vcpu);
491 kvm_dirty_ring_free(&vcpu->dirty_ring);
494 * No need for rcu_read_lock as VCPU_RUN is the only place that changes
495 * the vcpu->pid pointer, and at destruction time all file descriptors
498 put_pid(rcu_dereference_protected(vcpu->pid, 1));
500 free_page((unsigned long)vcpu->run);
501 kmem_cache_free(kvm_vcpu_cache, vcpu);
504 void kvm_destroy_vcpus(struct kvm *kvm)
507 struct kvm_vcpu *vcpu;
509 kvm_for_each_vcpu(i, vcpu, kvm) {
510 kvm_vcpu_destroy(vcpu);
511 xa_erase(&kvm->vcpu_array, i);
514 atomic_set(&kvm->online_vcpus, 0);
516 EXPORT_SYMBOL_GPL(kvm_destroy_vcpus);
518 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
519 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
521 return container_of(mn, struct kvm, mmu_notifier);
524 static void kvm_mmu_notifier_invalidate_range(struct mmu_notifier *mn,
525 struct mm_struct *mm,
526 unsigned long start, unsigned long end)
528 struct kvm *kvm = mmu_notifier_to_kvm(mn);
531 idx = srcu_read_lock(&kvm->srcu);
532 kvm_arch_mmu_notifier_invalidate_range(kvm, start, end);
533 srcu_read_unlock(&kvm->srcu, idx);
536 typedef bool (*hva_handler_t)(struct kvm *kvm, struct kvm_gfn_range *range);
538 typedef void (*on_lock_fn_t)(struct kvm *kvm, unsigned long start,
541 typedef void (*on_unlock_fn_t)(struct kvm *kvm);
543 struct kvm_hva_range {
547 hva_handler_t handler;
548 on_lock_fn_t on_lock;
549 on_unlock_fn_t on_unlock;
555 * Use a dedicated stub instead of NULL to indicate that there is no callback
556 * function/handler. The compiler technically can't guarantee that a real
557 * function will have a non-zero address, and so it will generate code to
558 * check for !NULL, whereas comparing against a stub will be elided at compile
559 * time (unless the compiler is getting long in the tooth, e.g. gcc 4.9).
561 static void kvm_null_fn(void)
565 #define IS_KVM_NULL_FN(fn) ((fn) == (void *)kvm_null_fn)
567 /* Iterate over each memslot intersecting [start, last] (inclusive) range */
568 #define kvm_for_each_memslot_in_hva_range(node, slots, start, last) \
569 for (node = interval_tree_iter_first(&slots->hva_tree, start, last); \
571 node = interval_tree_iter_next(node, start, last)) \
573 static __always_inline int __kvm_handle_hva_range(struct kvm *kvm,
574 const struct kvm_hva_range *range)
576 bool ret = false, locked = false;
577 struct kvm_gfn_range gfn_range;
578 struct kvm_memory_slot *slot;
579 struct kvm_memslots *slots;
582 if (WARN_ON_ONCE(range->end <= range->start))
585 /* A null handler is allowed if and only if on_lock() is provided. */
586 if (WARN_ON_ONCE(IS_KVM_NULL_FN(range->on_lock) &&
587 IS_KVM_NULL_FN(range->handler)))
590 idx = srcu_read_lock(&kvm->srcu);
592 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
593 struct interval_tree_node *node;
595 slots = __kvm_memslots(kvm, i);
596 kvm_for_each_memslot_in_hva_range(node, slots,
597 range->start, range->end - 1) {
598 unsigned long hva_start, hva_end;
600 slot = container_of(node, struct kvm_memory_slot, hva_node[slots->node_idx]);
601 hva_start = max(range->start, slot->userspace_addr);
602 hva_end = min(range->end, slot->userspace_addr +
603 (slot->npages << PAGE_SHIFT));
606 * To optimize for the likely case where the address
607 * range is covered by zero or one memslots, don't
608 * bother making these conditional (to avoid writes on
609 * the second or later invocation of the handler).
611 gfn_range.pte = range->pte;
612 gfn_range.may_block = range->may_block;
615 * {gfn(page) | page intersects with [hva_start, hva_end)} =
616 * {gfn_start, gfn_start+1, ..., gfn_end-1}.
618 gfn_range.start = hva_to_gfn_memslot(hva_start, slot);
619 gfn_range.end = hva_to_gfn_memslot(hva_end + PAGE_SIZE - 1, slot);
620 gfn_range.slot = slot;
625 if (!IS_KVM_NULL_FN(range->on_lock))
626 range->on_lock(kvm, range->start, range->end);
627 if (IS_KVM_NULL_FN(range->handler))
630 ret |= range->handler(kvm, &gfn_range);
634 if (range->flush_on_ret && ret)
635 kvm_flush_remote_tlbs(kvm);
639 if (!IS_KVM_NULL_FN(range->on_unlock))
640 range->on_unlock(kvm);
643 srcu_read_unlock(&kvm->srcu, idx);
645 /* The notifiers are averse to booleans. :-( */
649 static __always_inline int kvm_handle_hva_range(struct mmu_notifier *mn,
653 hva_handler_t handler)
655 struct kvm *kvm = mmu_notifier_to_kvm(mn);
656 const struct kvm_hva_range range = {
661 .on_lock = (void *)kvm_null_fn,
662 .on_unlock = (void *)kvm_null_fn,
663 .flush_on_ret = true,
667 return __kvm_handle_hva_range(kvm, &range);
670 static __always_inline int kvm_handle_hva_range_no_flush(struct mmu_notifier *mn,
673 hva_handler_t handler)
675 struct kvm *kvm = mmu_notifier_to_kvm(mn);
676 const struct kvm_hva_range range = {
681 .on_lock = (void *)kvm_null_fn,
682 .on_unlock = (void *)kvm_null_fn,
683 .flush_on_ret = false,
687 return __kvm_handle_hva_range(kvm, &range);
690 static bool kvm_change_spte_gfn(struct kvm *kvm, struct kvm_gfn_range *range)
693 * Skipping invalid memslots is correct if and only change_pte() is
694 * surrounded by invalidate_range_{start,end}(), which is currently
695 * guaranteed by the primary MMU. If that ever changes, KVM needs to
696 * unmap the memslot instead of skipping the memslot to ensure that KVM
697 * doesn't hold references to the old PFN.
699 WARN_ON_ONCE(!READ_ONCE(kvm->mn_active_invalidate_count));
701 if (range->slot->flags & KVM_MEMSLOT_INVALID)
704 return kvm_set_spte_gfn(kvm, range);
707 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
708 struct mm_struct *mm,
709 unsigned long address,
712 struct kvm *kvm = mmu_notifier_to_kvm(mn);
714 trace_kvm_set_spte_hva(address);
717 * .change_pte() must be surrounded by .invalidate_range_{start,end}().
718 * If mmu_invalidate_in_progress is zero, then no in-progress
719 * invalidations, including this one, found a relevant memslot at
720 * start(); rechecking memslots here is unnecessary. Note, a false
721 * positive (count elevated by a different invalidation) is sub-optimal
722 * but functionally ok.
724 WARN_ON_ONCE(!READ_ONCE(kvm->mn_active_invalidate_count));
725 if (!READ_ONCE(kvm->mmu_invalidate_in_progress))
728 kvm_handle_hva_range(mn, address, address + 1, pte, kvm_change_spte_gfn);
731 void kvm_mmu_invalidate_begin(struct kvm *kvm, unsigned long start,
735 * The count increase must become visible at unlock time as no
736 * spte can be established without taking the mmu_lock and
737 * count is also read inside the mmu_lock critical section.
739 kvm->mmu_invalidate_in_progress++;
740 if (likely(kvm->mmu_invalidate_in_progress == 1)) {
741 kvm->mmu_invalidate_range_start = start;
742 kvm->mmu_invalidate_range_end = end;
745 * Fully tracking multiple concurrent ranges has diminishing
746 * returns. Keep things simple and just find the minimal range
747 * which includes the current and new ranges. As there won't be
748 * enough information to subtract a range after its invalidate
749 * completes, any ranges invalidated concurrently will
750 * accumulate and persist until all outstanding invalidates
753 kvm->mmu_invalidate_range_start =
754 min(kvm->mmu_invalidate_range_start, start);
755 kvm->mmu_invalidate_range_end =
756 max(kvm->mmu_invalidate_range_end, end);
760 static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
761 const struct mmu_notifier_range *range)
763 struct kvm *kvm = mmu_notifier_to_kvm(mn);
764 const struct kvm_hva_range hva_range = {
765 .start = range->start,
768 .handler = kvm_unmap_gfn_range,
769 .on_lock = kvm_mmu_invalidate_begin,
770 .on_unlock = kvm_arch_guest_memory_reclaimed,
771 .flush_on_ret = true,
772 .may_block = mmu_notifier_range_blockable(range),
775 trace_kvm_unmap_hva_range(range->start, range->end);
778 * Prevent memslot modification between range_start() and range_end()
779 * so that conditionally locking provides the same result in both
780 * functions. Without that guarantee, the mmu_invalidate_in_progress
781 * adjustments will be imbalanced.
783 * Pairs with the decrement in range_end().
785 spin_lock(&kvm->mn_invalidate_lock);
786 kvm->mn_active_invalidate_count++;
787 spin_unlock(&kvm->mn_invalidate_lock);
790 * Invalidate pfn caches _before_ invalidating the secondary MMUs, i.e.
791 * before acquiring mmu_lock, to avoid holding mmu_lock while acquiring
792 * each cache's lock. There are relatively few caches in existence at
793 * any given time, and the caches themselves can check for hva overlap,
794 * i.e. don't need to rely on memslot overlap checks for performance.
795 * Because this runs without holding mmu_lock, the pfn caches must use
796 * mn_active_invalidate_count (see above) instead of
797 * mmu_invalidate_in_progress.
799 gfn_to_pfn_cache_invalidate_start(kvm, range->start, range->end,
800 hva_range.may_block);
802 __kvm_handle_hva_range(kvm, &hva_range);
807 void kvm_mmu_invalidate_end(struct kvm *kvm, unsigned long start,
811 * This sequence increase will notify the kvm page fault that
812 * the page that is going to be mapped in the spte could have
815 kvm->mmu_invalidate_seq++;
818 * The above sequence increase must be visible before the
819 * below count decrease, which is ensured by the smp_wmb above
820 * in conjunction with the smp_rmb in mmu_invalidate_retry().
822 kvm->mmu_invalidate_in_progress--;
825 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
826 const struct mmu_notifier_range *range)
828 struct kvm *kvm = mmu_notifier_to_kvm(mn);
829 const struct kvm_hva_range hva_range = {
830 .start = range->start,
833 .handler = (void *)kvm_null_fn,
834 .on_lock = kvm_mmu_invalidate_end,
835 .on_unlock = (void *)kvm_null_fn,
836 .flush_on_ret = false,
837 .may_block = mmu_notifier_range_blockable(range),
841 __kvm_handle_hva_range(kvm, &hva_range);
843 /* Pairs with the increment in range_start(). */
844 spin_lock(&kvm->mn_invalidate_lock);
845 wake = (--kvm->mn_active_invalidate_count == 0);
846 spin_unlock(&kvm->mn_invalidate_lock);
849 * There can only be one waiter, since the wait happens under
853 rcuwait_wake_up(&kvm->mn_memslots_update_rcuwait);
855 BUG_ON(kvm->mmu_invalidate_in_progress < 0);
858 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
859 struct mm_struct *mm,
863 trace_kvm_age_hva(start, end);
865 return kvm_handle_hva_range(mn, start, end, __pte(0), kvm_age_gfn);
868 static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
869 struct mm_struct *mm,
873 trace_kvm_age_hva(start, end);
876 * Even though we do not flush TLB, this will still adversely
877 * affect performance on pre-Haswell Intel EPT, where there is
878 * no EPT Access Bit to clear so that we have to tear down EPT
879 * tables instead. If we find this unacceptable, we can always
880 * add a parameter to kvm_age_hva so that it effectively doesn't
881 * do anything on clear_young.
883 * Also note that currently we never issue secondary TLB flushes
884 * from clear_young, leaving this job up to the regular system
885 * cadence. If we find this inaccurate, we might come up with a
886 * more sophisticated heuristic later.
888 return kvm_handle_hva_range_no_flush(mn, start, end, kvm_age_gfn);
891 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
892 struct mm_struct *mm,
893 unsigned long address)
895 trace_kvm_test_age_hva(address);
897 return kvm_handle_hva_range_no_flush(mn, address, address + 1,
901 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
902 struct mm_struct *mm)
904 struct kvm *kvm = mmu_notifier_to_kvm(mn);
907 idx = srcu_read_lock(&kvm->srcu);
908 kvm_flush_shadow_all(kvm);
909 srcu_read_unlock(&kvm->srcu, idx);
912 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
913 .invalidate_range = kvm_mmu_notifier_invalidate_range,
914 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
915 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
916 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
917 .clear_young = kvm_mmu_notifier_clear_young,
918 .test_young = kvm_mmu_notifier_test_young,
919 .change_pte = kvm_mmu_notifier_change_pte,
920 .release = kvm_mmu_notifier_release,
923 static int kvm_init_mmu_notifier(struct kvm *kvm)
925 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
926 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
929 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
931 static int kvm_init_mmu_notifier(struct kvm *kvm)
936 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
938 #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER
939 static int kvm_pm_notifier_call(struct notifier_block *bl,
943 struct kvm *kvm = container_of(bl, struct kvm, pm_notifier);
945 return kvm_arch_pm_notifier(kvm, state);
948 static void kvm_init_pm_notifier(struct kvm *kvm)
950 kvm->pm_notifier.notifier_call = kvm_pm_notifier_call;
951 /* Suspend KVM before we suspend ftrace, RCU, etc. */
952 kvm->pm_notifier.priority = INT_MAX;
953 register_pm_notifier(&kvm->pm_notifier);
956 static void kvm_destroy_pm_notifier(struct kvm *kvm)
958 unregister_pm_notifier(&kvm->pm_notifier);
960 #else /* !CONFIG_HAVE_KVM_PM_NOTIFIER */
961 static void kvm_init_pm_notifier(struct kvm *kvm)
965 static void kvm_destroy_pm_notifier(struct kvm *kvm)
968 #endif /* CONFIG_HAVE_KVM_PM_NOTIFIER */
970 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
972 if (!memslot->dirty_bitmap)
975 kvfree(memslot->dirty_bitmap);
976 memslot->dirty_bitmap = NULL;
979 /* This does not remove the slot from struct kvm_memslots data structures */
980 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
982 kvm_destroy_dirty_bitmap(slot);
984 kvm_arch_free_memslot(kvm, slot);
989 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
991 struct hlist_node *idnode;
992 struct kvm_memory_slot *memslot;
996 * The same memslot objects live in both active and inactive sets,
997 * arbitrarily free using index '1' so the second invocation of this
998 * function isn't operating over a structure with dangling pointers
999 * (even though this function isn't actually touching them).
1001 if (!slots->node_idx)
1004 hash_for_each_safe(slots->id_hash, bkt, idnode, memslot, id_node[1])
1005 kvm_free_memslot(kvm, memslot);
1008 static umode_t kvm_stats_debugfs_mode(const struct _kvm_stats_desc *pdesc)
1010 switch (pdesc->desc.flags & KVM_STATS_TYPE_MASK) {
1011 case KVM_STATS_TYPE_INSTANT:
1013 case KVM_STATS_TYPE_CUMULATIVE:
1014 case KVM_STATS_TYPE_PEAK:
1021 static void kvm_destroy_vm_debugfs(struct kvm *kvm)
1024 int kvm_debugfs_num_entries = kvm_vm_stats_header.num_desc +
1025 kvm_vcpu_stats_header.num_desc;
1027 if (IS_ERR(kvm->debugfs_dentry))
1030 debugfs_remove_recursive(kvm->debugfs_dentry);
1032 if (kvm->debugfs_stat_data) {
1033 for (i = 0; i < kvm_debugfs_num_entries; i++)
1034 kfree(kvm->debugfs_stat_data[i]);
1035 kfree(kvm->debugfs_stat_data);
1039 static int kvm_create_vm_debugfs(struct kvm *kvm, const char *fdname)
1041 static DEFINE_MUTEX(kvm_debugfs_lock);
1042 struct dentry *dent;
1043 char dir_name[ITOA_MAX_LEN * 2];
1044 struct kvm_stat_data *stat_data;
1045 const struct _kvm_stats_desc *pdesc;
1046 int i, ret = -ENOMEM;
1047 int kvm_debugfs_num_entries = kvm_vm_stats_header.num_desc +
1048 kvm_vcpu_stats_header.num_desc;
1050 if (!debugfs_initialized())
1053 snprintf(dir_name, sizeof(dir_name), "%d-%s", task_pid_nr(current), fdname);
1054 mutex_lock(&kvm_debugfs_lock);
1055 dent = debugfs_lookup(dir_name, kvm_debugfs_dir);
1057 pr_warn_ratelimited("KVM: debugfs: duplicate directory %s\n", dir_name);
1059 mutex_unlock(&kvm_debugfs_lock);
1062 dent = debugfs_create_dir(dir_name, kvm_debugfs_dir);
1063 mutex_unlock(&kvm_debugfs_lock);
1067 kvm->debugfs_dentry = dent;
1068 kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries,
1069 sizeof(*kvm->debugfs_stat_data),
1070 GFP_KERNEL_ACCOUNT);
1071 if (!kvm->debugfs_stat_data)
1074 for (i = 0; i < kvm_vm_stats_header.num_desc; ++i) {
1075 pdesc = &kvm_vm_stats_desc[i];
1076 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT);
1080 stat_data->kvm = kvm;
1081 stat_data->desc = pdesc;
1082 stat_data->kind = KVM_STAT_VM;
1083 kvm->debugfs_stat_data[i] = stat_data;
1084 debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
1085 kvm->debugfs_dentry, stat_data,
1089 for (i = 0; i < kvm_vcpu_stats_header.num_desc; ++i) {
1090 pdesc = &kvm_vcpu_stats_desc[i];
1091 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT);
1095 stat_data->kvm = kvm;
1096 stat_data->desc = pdesc;
1097 stat_data->kind = KVM_STAT_VCPU;
1098 kvm->debugfs_stat_data[i + kvm_vm_stats_header.num_desc] = stat_data;
1099 debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
1100 kvm->debugfs_dentry, stat_data,
1104 ret = kvm_arch_create_vm_debugfs(kvm);
1110 kvm_destroy_vm_debugfs(kvm);
1115 * Called after the VM is otherwise initialized, but just before adding it to
1118 int __weak kvm_arch_post_init_vm(struct kvm *kvm)
1124 * Called just after removing the VM from the vm_list, but before doing any
1125 * other destruction.
1127 void __weak kvm_arch_pre_destroy_vm(struct kvm *kvm)
1132 * Called after per-vm debugfs created. When called kvm->debugfs_dentry should
1133 * be setup already, so we can create arch-specific debugfs entries under it.
1134 * Cleanup should be automatic done in kvm_destroy_vm_debugfs() recursively, so
1135 * a per-arch destroy interface is not needed.
1137 int __weak kvm_arch_create_vm_debugfs(struct kvm *kvm)
1142 static struct kvm *kvm_create_vm(unsigned long type, const char *fdname)
1144 struct kvm *kvm = kvm_arch_alloc_vm();
1145 struct kvm_memslots *slots;
1150 return ERR_PTR(-ENOMEM);
1152 /* KVM is pinned via open("/dev/kvm"), the fd passed to this ioctl(). */
1153 __module_get(kvm_chardev_ops.owner);
1155 KVM_MMU_LOCK_INIT(kvm);
1156 mmgrab(current->mm);
1157 kvm->mm = current->mm;
1158 kvm_eventfd_init(kvm);
1159 mutex_init(&kvm->lock);
1160 mutex_init(&kvm->irq_lock);
1161 mutex_init(&kvm->slots_lock);
1162 mutex_init(&kvm->slots_arch_lock);
1163 spin_lock_init(&kvm->mn_invalidate_lock);
1164 rcuwait_init(&kvm->mn_memslots_update_rcuwait);
1165 xa_init(&kvm->vcpu_array);
1167 INIT_LIST_HEAD(&kvm->gpc_list);
1168 spin_lock_init(&kvm->gpc_lock);
1170 INIT_LIST_HEAD(&kvm->devices);
1171 kvm->max_vcpus = KVM_MAX_VCPUS;
1173 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
1176 * Force subsequent debugfs file creations to fail if the VM directory
1177 * is not created (by kvm_create_vm_debugfs()).
1179 kvm->debugfs_dentry = ERR_PTR(-ENOENT);
1181 snprintf(kvm->stats_id, sizeof(kvm->stats_id), "kvm-%d",
1182 task_pid_nr(current));
1184 if (init_srcu_struct(&kvm->srcu))
1185 goto out_err_no_srcu;
1186 if (init_srcu_struct(&kvm->irq_srcu))
1187 goto out_err_no_irq_srcu;
1189 refcount_set(&kvm->users_count, 1);
1190 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
1191 for (j = 0; j < 2; j++) {
1192 slots = &kvm->__memslots[i][j];
1194 atomic_long_set(&slots->last_used_slot, (unsigned long)NULL);
1195 slots->hva_tree = RB_ROOT_CACHED;
1196 slots->gfn_tree = RB_ROOT;
1197 hash_init(slots->id_hash);
1198 slots->node_idx = j;
1200 /* Generations must be different for each address space. */
1201 slots->generation = i;
1204 rcu_assign_pointer(kvm->memslots[i], &kvm->__memslots[i][0]);
1207 for (i = 0; i < KVM_NR_BUSES; i++) {
1208 rcu_assign_pointer(kvm->buses[i],
1209 kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL_ACCOUNT));
1211 goto out_err_no_arch_destroy_vm;
1214 r = kvm_arch_init_vm(kvm, type);
1216 goto out_err_no_arch_destroy_vm;
1218 r = hardware_enable_all();
1220 goto out_err_no_disable;
1222 #ifdef CONFIG_HAVE_KVM_IRQFD
1223 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
1226 r = kvm_init_mmu_notifier(kvm);
1228 goto out_err_no_mmu_notifier;
1230 r = kvm_coalesced_mmio_init(kvm);
1232 goto out_no_coalesced_mmio;
1234 r = kvm_create_vm_debugfs(kvm, fdname);
1236 goto out_err_no_debugfs;
1238 r = kvm_arch_post_init_vm(kvm);
1242 mutex_lock(&kvm_lock);
1243 list_add(&kvm->vm_list, &vm_list);
1244 mutex_unlock(&kvm_lock);
1246 preempt_notifier_inc();
1247 kvm_init_pm_notifier(kvm);
1252 kvm_destroy_vm_debugfs(kvm);
1254 kvm_coalesced_mmio_free(kvm);
1255 out_no_coalesced_mmio:
1256 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1257 if (kvm->mmu_notifier.ops)
1258 mmu_notifier_unregister(&kvm->mmu_notifier, current->mm);
1260 out_err_no_mmu_notifier:
1261 hardware_disable_all();
1263 kvm_arch_destroy_vm(kvm);
1264 out_err_no_arch_destroy_vm:
1265 WARN_ON_ONCE(!refcount_dec_and_test(&kvm->users_count));
1266 for (i = 0; i < KVM_NR_BUSES; i++)
1267 kfree(kvm_get_bus(kvm, i));
1268 cleanup_srcu_struct(&kvm->irq_srcu);
1269 out_err_no_irq_srcu:
1270 cleanup_srcu_struct(&kvm->srcu);
1272 kvm_arch_free_vm(kvm);
1273 mmdrop(current->mm);
1274 module_put(kvm_chardev_ops.owner);
1278 static void kvm_destroy_devices(struct kvm *kvm)
1280 struct kvm_device *dev, *tmp;
1283 * We do not need to take the kvm->lock here, because nobody else
1284 * has a reference to the struct kvm at this point and therefore
1285 * cannot access the devices list anyhow.
1287 list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
1288 list_del(&dev->vm_node);
1289 dev->ops->destroy(dev);
1293 static void kvm_destroy_vm(struct kvm *kvm)
1296 struct mm_struct *mm = kvm->mm;
1298 kvm_destroy_pm_notifier(kvm);
1299 kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm);
1300 kvm_destroy_vm_debugfs(kvm);
1301 kvm_arch_sync_events(kvm);
1302 mutex_lock(&kvm_lock);
1303 list_del(&kvm->vm_list);
1304 mutex_unlock(&kvm_lock);
1305 kvm_arch_pre_destroy_vm(kvm);
1307 kvm_free_irq_routing(kvm);
1308 for (i = 0; i < KVM_NR_BUSES; i++) {
1309 struct kvm_io_bus *bus = kvm_get_bus(kvm, i);
1312 kvm_io_bus_destroy(bus);
1313 kvm->buses[i] = NULL;
1315 kvm_coalesced_mmio_free(kvm);
1316 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1317 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
1319 * At this point, pending calls to invalidate_range_start()
1320 * have completed but no more MMU notifiers will run, so
1321 * mn_active_invalidate_count may remain unbalanced.
1322 * No threads can be waiting in kvm_swap_active_memslots() as the
1323 * last reference on KVM has been dropped, but freeing
1324 * memslots would deadlock without this manual intervention.
1326 WARN_ON(rcuwait_active(&kvm->mn_memslots_update_rcuwait));
1327 kvm->mn_active_invalidate_count = 0;
1329 kvm_flush_shadow_all(kvm);
1331 kvm_arch_destroy_vm(kvm);
1332 kvm_destroy_devices(kvm);
1333 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
1334 kvm_free_memslots(kvm, &kvm->__memslots[i][0]);
1335 kvm_free_memslots(kvm, &kvm->__memslots[i][1]);
1337 cleanup_srcu_struct(&kvm->irq_srcu);
1338 cleanup_srcu_struct(&kvm->srcu);
1339 kvm_arch_free_vm(kvm);
1340 preempt_notifier_dec();
1341 hardware_disable_all();
1343 module_put(kvm_chardev_ops.owner);
1346 void kvm_get_kvm(struct kvm *kvm)
1348 refcount_inc(&kvm->users_count);
1350 EXPORT_SYMBOL_GPL(kvm_get_kvm);
1353 * Make sure the vm is not during destruction, which is a safe version of
1354 * kvm_get_kvm(). Return true if kvm referenced successfully, false otherwise.
1356 bool kvm_get_kvm_safe(struct kvm *kvm)
1358 return refcount_inc_not_zero(&kvm->users_count);
1360 EXPORT_SYMBOL_GPL(kvm_get_kvm_safe);
1362 void kvm_put_kvm(struct kvm *kvm)
1364 if (refcount_dec_and_test(&kvm->users_count))
1365 kvm_destroy_vm(kvm);
1367 EXPORT_SYMBOL_GPL(kvm_put_kvm);
1370 * Used to put a reference that was taken on behalf of an object associated
1371 * with a user-visible file descriptor, e.g. a vcpu or device, if installation
1372 * of the new file descriptor fails and the reference cannot be transferred to
1373 * its final owner. In such cases, the caller is still actively using @kvm and
1374 * will fail miserably if the refcount unexpectedly hits zero.
1376 void kvm_put_kvm_no_destroy(struct kvm *kvm)
1378 WARN_ON(refcount_dec_and_test(&kvm->users_count));
1380 EXPORT_SYMBOL_GPL(kvm_put_kvm_no_destroy);
1382 static int kvm_vm_release(struct inode *inode, struct file *filp)
1384 struct kvm *kvm = filp->private_data;
1386 kvm_irqfd_release(kvm);
1393 * Allocation size is twice as large as the actual dirty bitmap size.
1394 * See kvm_vm_ioctl_get_dirty_log() why this is needed.
1396 static int kvm_alloc_dirty_bitmap(struct kvm_memory_slot *memslot)
1398 unsigned long dirty_bytes = kvm_dirty_bitmap_bytes(memslot);
1400 memslot->dirty_bitmap = __vcalloc(2, dirty_bytes, GFP_KERNEL_ACCOUNT);
1401 if (!memslot->dirty_bitmap)
1407 static struct kvm_memslots *kvm_get_inactive_memslots(struct kvm *kvm, int as_id)
1409 struct kvm_memslots *active = __kvm_memslots(kvm, as_id);
1410 int node_idx_inactive = active->node_idx ^ 1;
1412 return &kvm->__memslots[as_id][node_idx_inactive];
1416 * Helper to get the address space ID when one of memslot pointers may be NULL.
1417 * This also serves as a sanity that at least one of the pointers is non-NULL,
1418 * and that their address space IDs don't diverge.
1420 static int kvm_memslots_get_as_id(struct kvm_memory_slot *a,
1421 struct kvm_memory_slot *b)
1423 if (WARN_ON_ONCE(!a && !b))
1431 WARN_ON_ONCE(a->as_id != b->as_id);
1435 static void kvm_insert_gfn_node(struct kvm_memslots *slots,
1436 struct kvm_memory_slot *slot)
1438 struct rb_root *gfn_tree = &slots->gfn_tree;
1439 struct rb_node **node, *parent;
1440 int idx = slots->node_idx;
1443 for (node = &gfn_tree->rb_node; *node; ) {
1444 struct kvm_memory_slot *tmp;
1446 tmp = container_of(*node, struct kvm_memory_slot, gfn_node[idx]);
1448 if (slot->base_gfn < tmp->base_gfn)
1449 node = &(*node)->rb_left;
1450 else if (slot->base_gfn > tmp->base_gfn)
1451 node = &(*node)->rb_right;
1456 rb_link_node(&slot->gfn_node[idx], parent, node);
1457 rb_insert_color(&slot->gfn_node[idx], gfn_tree);
1460 static void kvm_erase_gfn_node(struct kvm_memslots *slots,
1461 struct kvm_memory_slot *slot)
1463 rb_erase(&slot->gfn_node[slots->node_idx], &slots->gfn_tree);
1466 static void kvm_replace_gfn_node(struct kvm_memslots *slots,
1467 struct kvm_memory_slot *old,
1468 struct kvm_memory_slot *new)
1470 int idx = slots->node_idx;
1472 WARN_ON_ONCE(old->base_gfn != new->base_gfn);
1474 rb_replace_node(&old->gfn_node[idx], &new->gfn_node[idx],
1479 * Replace @old with @new in the inactive memslots.
1481 * With NULL @old this simply adds @new.
1482 * With NULL @new this simply removes @old.
1484 * If @new is non-NULL its hva_node[slots_idx] range has to be set
1487 static void kvm_replace_memslot(struct kvm *kvm,
1488 struct kvm_memory_slot *old,
1489 struct kvm_memory_slot *new)
1491 int as_id = kvm_memslots_get_as_id(old, new);
1492 struct kvm_memslots *slots = kvm_get_inactive_memslots(kvm, as_id);
1493 int idx = slots->node_idx;
1496 hash_del(&old->id_node[idx]);
1497 interval_tree_remove(&old->hva_node[idx], &slots->hva_tree);
1499 if ((long)old == atomic_long_read(&slots->last_used_slot))
1500 atomic_long_set(&slots->last_used_slot, (long)new);
1503 kvm_erase_gfn_node(slots, old);
1509 * Initialize @new's hva range. Do this even when replacing an @old
1510 * slot, kvm_copy_memslot() deliberately does not touch node data.
1512 new->hva_node[idx].start = new->userspace_addr;
1513 new->hva_node[idx].last = new->userspace_addr +
1514 (new->npages << PAGE_SHIFT) - 1;
1517 * (Re)Add the new memslot. There is no O(1) interval_tree_replace(),
1518 * hva_node needs to be swapped with remove+insert even though hva can't
1519 * change when replacing an existing slot.
1521 hash_add(slots->id_hash, &new->id_node[idx], new->id);
1522 interval_tree_insert(&new->hva_node[idx], &slots->hva_tree);
1525 * If the memslot gfn is unchanged, rb_replace_node() can be used to
1526 * switch the node in the gfn tree instead of removing the old and
1527 * inserting the new as two separate operations. Replacement is a
1528 * single O(1) operation versus two O(log(n)) operations for
1531 if (old && old->base_gfn == new->base_gfn) {
1532 kvm_replace_gfn_node(slots, old, new);
1535 kvm_erase_gfn_node(slots, old);
1536 kvm_insert_gfn_node(slots, new);
1540 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
1542 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
1544 #ifdef __KVM_HAVE_READONLY_MEM
1545 valid_flags |= KVM_MEM_READONLY;
1548 if (mem->flags & ~valid_flags)
1554 static void kvm_swap_active_memslots(struct kvm *kvm, int as_id)
1556 struct kvm_memslots *slots = kvm_get_inactive_memslots(kvm, as_id);
1558 /* Grab the generation from the activate memslots. */
1559 u64 gen = __kvm_memslots(kvm, as_id)->generation;
1561 WARN_ON(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS);
1562 slots->generation = gen | KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
1565 * Do not store the new memslots while there are invalidations in
1566 * progress, otherwise the locking in invalidate_range_start and
1567 * invalidate_range_end will be unbalanced.
1569 spin_lock(&kvm->mn_invalidate_lock);
1570 prepare_to_rcuwait(&kvm->mn_memslots_update_rcuwait);
1571 while (kvm->mn_active_invalidate_count) {
1572 set_current_state(TASK_UNINTERRUPTIBLE);
1573 spin_unlock(&kvm->mn_invalidate_lock);
1575 spin_lock(&kvm->mn_invalidate_lock);
1577 finish_rcuwait(&kvm->mn_memslots_update_rcuwait);
1578 rcu_assign_pointer(kvm->memslots[as_id], slots);
1579 spin_unlock(&kvm->mn_invalidate_lock);
1582 * Acquired in kvm_set_memslot. Must be released before synchronize
1583 * SRCU below in order to avoid deadlock with another thread
1584 * acquiring the slots_arch_lock in an srcu critical section.
1586 mutex_unlock(&kvm->slots_arch_lock);
1588 synchronize_srcu_expedited(&kvm->srcu);
1591 * Increment the new memslot generation a second time, dropping the
1592 * update in-progress flag and incrementing the generation based on
1593 * the number of address spaces. This provides a unique and easily
1594 * identifiable generation number while the memslots are in flux.
1596 gen = slots->generation & ~KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
1599 * Generations must be unique even across address spaces. We do not need
1600 * a global counter for that, instead the generation space is evenly split
1601 * across address spaces. For example, with two address spaces, address
1602 * space 0 will use generations 0, 2, 4, ... while address space 1 will
1603 * use generations 1, 3, 5, ...
1605 gen += KVM_ADDRESS_SPACE_NUM;
1607 kvm_arch_memslots_updated(kvm, gen);
1609 slots->generation = gen;
1612 static int kvm_prepare_memory_region(struct kvm *kvm,
1613 const struct kvm_memory_slot *old,
1614 struct kvm_memory_slot *new,
1615 enum kvm_mr_change change)
1620 * If dirty logging is disabled, nullify the bitmap; the old bitmap
1621 * will be freed on "commit". If logging is enabled in both old and
1622 * new, reuse the existing bitmap. If logging is enabled only in the
1623 * new and KVM isn't using a ring buffer, allocate and initialize a
1626 if (change != KVM_MR_DELETE) {
1627 if (!(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
1628 new->dirty_bitmap = NULL;
1629 else if (old && old->dirty_bitmap)
1630 new->dirty_bitmap = old->dirty_bitmap;
1631 else if (kvm_use_dirty_bitmap(kvm)) {
1632 r = kvm_alloc_dirty_bitmap(new);
1636 if (kvm_dirty_log_manual_protect_and_init_set(kvm))
1637 bitmap_set(new->dirty_bitmap, 0, new->npages);
1641 r = kvm_arch_prepare_memory_region(kvm, old, new, change);
1643 /* Free the bitmap on failure if it was allocated above. */
1644 if (r && new && new->dirty_bitmap && (!old || !old->dirty_bitmap))
1645 kvm_destroy_dirty_bitmap(new);
1650 static void kvm_commit_memory_region(struct kvm *kvm,
1651 struct kvm_memory_slot *old,
1652 const struct kvm_memory_slot *new,
1653 enum kvm_mr_change change)
1655 int old_flags = old ? old->flags : 0;
1656 int new_flags = new ? new->flags : 0;
1658 * Update the total number of memslot pages before calling the arch
1659 * hook so that architectures can consume the result directly.
1661 if (change == KVM_MR_DELETE)
1662 kvm->nr_memslot_pages -= old->npages;
1663 else if (change == KVM_MR_CREATE)
1664 kvm->nr_memslot_pages += new->npages;
1666 if ((old_flags ^ new_flags) & KVM_MEM_LOG_DIRTY_PAGES) {
1667 int change = (new_flags & KVM_MEM_LOG_DIRTY_PAGES) ? 1 : -1;
1668 atomic_set(&kvm->nr_memslots_dirty_logging,
1669 atomic_read(&kvm->nr_memslots_dirty_logging) + change);
1672 kvm_arch_commit_memory_region(kvm, old, new, change);
1676 /* Nothing more to do. */
1679 /* Free the old memslot and all its metadata. */
1680 kvm_free_memslot(kvm, old);
1683 case KVM_MR_FLAGS_ONLY:
1685 * Free the dirty bitmap as needed; the below check encompasses
1686 * both the flags and whether a ring buffer is being used)
1688 if (old->dirty_bitmap && !new->dirty_bitmap)
1689 kvm_destroy_dirty_bitmap(old);
1692 * The final quirk. Free the detached, old slot, but only its
1693 * memory, not any metadata. Metadata, including arch specific
1694 * data, may be reused by @new.
1704 * Activate @new, which must be installed in the inactive slots by the caller,
1705 * by swapping the active slots and then propagating @new to @old once @old is
1706 * unreachable and can be safely modified.
1708 * With NULL @old this simply adds @new to @active (while swapping the sets).
1709 * With NULL @new this simply removes @old from @active and frees it
1710 * (while also swapping the sets).
1712 static void kvm_activate_memslot(struct kvm *kvm,
1713 struct kvm_memory_slot *old,
1714 struct kvm_memory_slot *new)
1716 int as_id = kvm_memslots_get_as_id(old, new);
1718 kvm_swap_active_memslots(kvm, as_id);
1720 /* Propagate the new memslot to the now inactive memslots. */
1721 kvm_replace_memslot(kvm, old, new);
1724 static void kvm_copy_memslot(struct kvm_memory_slot *dest,
1725 const struct kvm_memory_slot *src)
1727 dest->base_gfn = src->base_gfn;
1728 dest->npages = src->npages;
1729 dest->dirty_bitmap = src->dirty_bitmap;
1730 dest->arch = src->arch;
1731 dest->userspace_addr = src->userspace_addr;
1732 dest->flags = src->flags;
1734 dest->as_id = src->as_id;
1737 static void kvm_invalidate_memslot(struct kvm *kvm,
1738 struct kvm_memory_slot *old,
1739 struct kvm_memory_slot *invalid_slot)
1742 * Mark the current slot INVALID. As with all memslot modifications,
1743 * this must be done on an unreachable slot to avoid modifying the
1744 * current slot in the active tree.
1746 kvm_copy_memslot(invalid_slot, old);
1747 invalid_slot->flags |= KVM_MEMSLOT_INVALID;
1748 kvm_replace_memslot(kvm, old, invalid_slot);
1751 * Activate the slot that is now marked INVALID, but don't propagate
1752 * the slot to the now inactive slots. The slot is either going to be
1753 * deleted or recreated as a new slot.
1755 kvm_swap_active_memslots(kvm, old->as_id);
1758 * From this point no new shadow pages pointing to a deleted, or moved,
1759 * memslot will be created. Validation of sp->gfn happens in:
1760 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1761 * - kvm_is_visible_gfn (mmu_check_root)
1763 kvm_arch_flush_shadow_memslot(kvm, old);
1764 kvm_arch_guest_memory_reclaimed(kvm);
1766 /* Was released by kvm_swap_active_memslots(), reacquire. */
1767 mutex_lock(&kvm->slots_arch_lock);
1770 * Copy the arch-specific field of the newly-installed slot back to the
1771 * old slot as the arch data could have changed between releasing
1772 * slots_arch_lock in kvm_swap_active_memslots() and re-acquiring the lock
1773 * above. Writers are required to retrieve memslots *after* acquiring
1774 * slots_arch_lock, thus the active slot's data is guaranteed to be fresh.
1776 old->arch = invalid_slot->arch;
1779 static void kvm_create_memslot(struct kvm *kvm,
1780 struct kvm_memory_slot *new)
1782 /* Add the new memslot to the inactive set and activate. */
1783 kvm_replace_memslot(kvm, NULL, new);
1784 kvm_activate_memslot(kvm, NULL, new);
1787 static void kvm_delete_memslot(struct kvm *kvm,
1788 struct kvm_memory_slot *old,
1789 struct kvm_memory_slot *invalid_slot)
1792 * Remove the old memslot (in the inactive memslots) by passing NULL as
1793 * the "new" slot, and for the invalid version in the active slots.
1795 kvm_replace_memslot(kvm, old, NULL);
1796 kvm_activate_memslot(kvm, invalid_slot, NULL);
1799 static void kvm_move_memslot(struct kvm *kvm,
1800 struct kvm_memory_slot *old,
1801 struct kvm_memory_slot *new,
1802 struct kvm_memory_slot *invalid_slot)
1805 * Replace the old memslot in the inactive slots, and then swap slots
1806 * and replace the current INVALID with the new as well.
1808 kvm_replace_memslot(kvm, old, new);
1809 kvm_activate_memslot(kvm, invalid_slot, new);
1812 static void kvm_update_flags_memslot(struct kvm *kvm,
1813 struct kvm_memory_slot *old,
1814 struct kvm_memory_slot *new)
1817 * Similar to the MOVE case, but the slot doesn't need to be zapped as
1818 * an intermediate step. Instead, the old memslot is simply replaced
1819 * with a new, updated copy in both memslot sets.
1821 kvm_replace_memslot(kvm, old, new);
1822 kvm_activate_memslot(kvm, old, new);
1825 static int kvm_set_memslot(struct kvm *kvm,
1826 struct kvm_memory_slot *old,
1827 struct kvm_memory_slot *new,
1828 enum kvm_mr_change change)
1830 struct kvm_memory_slot *invalid_slot;
1834 * Released in kvm_swap_active_memslots().
1836 * Must be held from before the current memslots are copied until after
1837 * the new memslots are installed with rcu_assign_pointer, then
1838 * released before the synchronize srcu in kvm_swap_active_memslots().
1840 * When modifying memslots outside of the slots_lock, must be held
1841 * before reading the pointer to the current memslots until after all
1842 * changes to those memslots are complete.
1844 * These rules ensure that installing new memslots does not lose
1845 * changes made to the previous memslots.
1847 mutex_lock(&kvm->slots_arch_lock);
1850 * Invalidate the old slot if it's being deleted or moved. This is
1851 * done prior to actually deleting/moving the memslot to allow vCPUs to
1852 * continue running by ensuring there are no mappings or shadow pages
1853 * for the memslot when it is deleted/moved. Without pre-invalidation
1854 * (and without a lock), a window would exist between effecting the
1855 * delete/move and committing the changes in arch code where KVM or a
1856 * guest could access a non-existent memslot.
1858 * Modifications are done on a temporary, unreachable slot. The old
1859 * slot needs to be preserved in case a later step fails and the
1860 * invalidation needs to be reverted.
1862 if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) {
1863 invalid_slot = kzalloc(sizeof(*invalid_slot), GFP_KERNEL_ACCOUNT);
1864 if (!invalid_slot) {
1865 mutex_unlock(&kvm->slots_arch_lock);
1868 kvm_invalidate_memslot(kvm, old, invalid_slot);
1871 r = kvm_prepare_memory_region(kvm, old, new, change);
1874 * For DELETE/MOVE, revert the above INVALID change. No
1875 * modifications required since the original slot was preserved
1876 * in the inactive slots. Changing the active memslots also
1877 * release slots_arch_lock.
1879 if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) {
1880 kvm_activate_memslot(kvm, invalid_slot, old);
1881 kfree(invalid_slot);
1883 mutex_unlock(&kvm->slots_arch_lock);
1889 * For DELETE and MOVE, the working slot is now active as the INVALID
1890 * version of the old slot. MOVE is particularly special as it reuses
1891 * the old slot and returns a copy of the old slot (in working_slot).
1892 * For CREATE, there is no old slot. For DELETE and FLAGS_ONLY, the
1893 * old slot is detached but otherwise preserved.
1895 if (change == KVM_MR_CREATE)
1896 kvm_create_memslot(kvm, new);
1897 else if (change == KVM_MR_DELETE)
1898 kvm_delete_memslot(kvm, old, invalid_slot);
1899 else if (change == KVM_MR_MOVE)
1900 kvm_move_memslot(kvm, old, new, invalid_slot);
1901 else if (change == KVM_MR_FLAGS_ONLY)
1902 kvm_update_flags_memslot(kvm, old, new);
1906 /* Free the temporary INVALID slot used for DELETE and MOVE. */
1907 if (change == KVM_MR_DELETE || change == KVM_MR_MOVE)
1908 kfree(invalid_slot);
1911 * No need to refresh new->arch, changes after dropping slots_arch_lock
1912 * will directly hit the final, active memslot. Architectures are
1913 * responsible for knowing that new->arch may be stale.
1915 kvm_commit_memory_region(kvm, old, new, change);
1920 static bool kvm_check_memslot_overlap(struct kvm_memslots *slots, int id,
1921 gfn_t start, gfn_t end)
1923 struct kvm_memslot_iter iter;
1925 kvm_for_each_memslot_in_gfn_range(&iter, slots, start, end) {
1926 if (iter.slot->id != id)
1934 * Allocate some memory and give it an address in the guest physical address
1937 * Discontiguous memory is allowed, mostly for framebuffers.
1939 * Must be called holding kvm->slots_lock for write.
1941 int __kvm_set_memory_region(struct kvm *kvm,
1942 const struct kvm_userspace_memory_region *mem)
1944 struct kvm_memory_slot *old, *new;
1945 struct kvm_memslots *slots;
1946 enum kvm_mr_change change;
1947 unsigned long npages;
1952 r = check_memory_region_flags(mem);
1956 as_id = mem->slot >> 16;
1957 id = (u16)mem->slot;
1959 /* General sanity checks */
1960 if ((mem->memory_size & (PAGE_SIZE - 1)) ||
1961 (mem->memory_size != (unsigned long)mem->memory_size))
1963 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
1965 /* We can read the guest memory with __xxx_user() later on. */
1966 if ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
1967 (mem->userspace_addr != untagged_addr(mem->userspace_addr)) ||
1968 !access_ok((void __user *)(unsigned long)mem->userspace_addr,
1971 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
1973 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
1975 if ((mem->memory_size >> PAGE_SHIFT) > KVM_MEM_MAX_NR_PAGES)
1978 slots = __kvm_memslots(kvm, as_id);
1981 * Note, the old memslot (and the pointer itself!) may be invalidated
1982 * and/or destroyed by kvm_set_memslot().
1984 old = id_to_memslot(slots, id);
1986 if (!mem->memory_size) {
1987 if (!old || !old->npages)
1990 if (WARN_ON_ONCE(kvm->nr_memslot_pages < old->npages))
1993 return kvm_set_memslot(kvm, old, NULL, KVM_MR_DELETE);
1996 base_gfn = (mem->guest_phys_addr >> PAGE_SHIFT);
1997 npages = (mem->memory_size >> PAGE_SHIFT);
1999 if (!old || !old->npages) {
2000 change = KVM_MR_CREATE;
2003 * To simplify KVM internals, the total number of pages across
2004 * all memslots must fit in an unsigned long.
2006 if ((kvm->nr_memslot_pages + npages) < kvm->nr_memslot_pages)
2008 } else { /* Modify an existing slot. */
2009 if ((mem->userspace_addr != old->userspace_addr) ||
2010 (npages != old->npages) ||
2011 ((mem->flags ^ old->flags) & KVM_MEM_READONLY))
2014 if (base_gfn != old->base_gfn)
2015 change = KVM_MR_MOVE;
2016 else if (mem->flags != old->flags)
2017 change = KVM_MR_FLAGS_ONLY;
2018 else /* Nothing to change. */
2022 if ((change == KVM_MR_CREATE || change == KVM_MR_MOVE) &&
2023 kvm_check_memslot_overlap(slots, id, base_gfn, base_gfn + npages))
2026 /* Allocate a slot that will persist in the memslot. */
2027 new = kzalloc(sizeof(*new), GFP_KERNEL_ACCOUNT);
2033 new->base_gfn = base_gfn;
2034 new->npages = npages;
2035 new->flags = mem->flags;
2036 new->userspace_addr = mem->userspace_addr;
2038 r = kvm_set_memslot(kvm, old, new, change);
2043 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
2045 int kvm_set_memory_region(struct kvm *kvm,
2046 const struct kvm_userspace_memory_region *mem)
2050 mutex_lock(&kvm->slots_lock);
2051 r = __kvm_set_memory_region(kvm, mem);
2052 mutex_unlock(&kvm->slots_lock);
2055 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
2057 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
2058 struct kvm_userspace_memory_region *mem)
2060 if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
2063 return kvm_set_memory_region(kvm, mem);
2066 #ifndef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
2068 * kvm_get_dirty_log - get a snapshot of dirty pages
2069 * @kvm: pointer to kvm instance
2070 * @log: slot id and address to which we copy the log
2071 * @is_dirty: set to '1' if any dirty pages were found
2072 * @memslot: set to the associated memslot, always valid on success
2074 int kvm_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log,
2075 int *is_dirty, struct kvm_memory_slot **memslot)
2077 struct kvm_memslots *slots;
2080 unsigned long any = 0;
2082 /* Dirty ring tracking may be exclusive to dirty log tracking */
2083 if (!kvm_use_dirty_bitmap(kvm))
2089 as_id = log->slot >> 16;
2090 id = (u16)log->slot;
2091 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
2094 slots = __kvm_memslots(kvm, as_id);
2095 *memslot = id_to_memslot(slots, id);
2096 if (!(*memslot) || !(*memslot)->dirty_bitmap)
2099 kvm_arch_sync_dirty_log(kvm, *memslot);
2101 n = kvm_dirty_bitmap_bytes(*memslot);
2103 for (i = 0; !any && i < n/sizeof(long); ++i)
2104 any = (*memslot)->dirty_bitmap[i];
2106 if (copy_to_user(log->dirty_bitmap, (*memslot)->dirty_bitmap, n))
2113 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
2115 #else /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
2117 * kvm_get_dirty_log_protect - get a snapshot of dirty pages
2118 * and reenable dirty page tracking for the corresponding pages.
2119 * @kvm: pointer to kvm instance
2120 * @log: slot id and address to which we copy the log
2122 * We need to keep it in mind that VCPU threads can write to the bitmap
2123 * concurrently. So, to avoid losing track of dirty pages we keep the
2126 * 1. Take a snapshot of the bit and clear it if needed.
2127 * 2. Write protect the corresponding page.
2128 * 3. Copy the snapshot to the userspace.
2129 * 4. Upon return caller flushes TLB's if needed.
2131 * Between 2 and 4, the guest may write to the page using the remaining TLB
2132 * entry. This is not a problem because the page is reported dirty using
2133 * the snapshot taken before and step 4 ensures that writes done after
2134 * exiting to userspace will be logged for the next call.
2137 static int kvm_get_dirty_log_protect(struct kvm *kvm, struct kvm_dirty_log *log)
2139 struct kvm_memslots *slots;
2140 struct kvm_memory_slot *memslot;
2143 unsigned long *dirty_bitmap;
2144 unsigned long *dirty_bitmap_buffer;
2147 /* Dirty ring tracking may be exclusive to dirty log tracking */
2148 if (!kvm_use_dirty_bitmap(kvm))
2151 as_id = log->slot >> 16;
2152 id = (u16)log->slot;
2153 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
2156 slots = __kvm_memslots(kvm, as_id);
2157 memslot = id_to_memslot(slots, id);
2158 if (!memslot || !memslot->dirty_bitmap)
2161 dirty_bitmap = memslot->dirty_bitmap;
2163 kvm_arch_sync_dirty_log(kvm, memslot);
2165 n = kvm_dirty_bitmap_bytes(memslot);
2167 if (kvm->manual_dirty_log_protect) {
2169 * Unlike kvm_get_dirty_log, we always return false in *flush,
2170 * because no flush is needed until KVM_CLEAR_DIRTY_LOG. There
2171 * is some code duplication between this function and
2172 * kvm_get_dirty_log, but hopefully all architecture
2173 * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log
2174 * can be eliminated.
2176 dirty_bitmap_buffer = dirty_bitmap;
2178 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
2179 memset(dirty_bitmap_buffer, 0, n);
2182 for (i = 0; i < n / sizeof(long); i++) {
2186 if (!dirty_bitmap[i])
2190 mask = xchg(&dirty_bitmap[i], 0);
2191 dirty_bitmap_buffer[i] = mask;
2193 offset = i * BITS_PER_LONG;
2194 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
2197 KVM_MMU_UNLOCK(kvm);
2201 kvm_arch_flush_remote_tlbs_memslot(kvm, memslot);
2203 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
2210 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
2211 * @kvm: kvm instance
2212 * @log: slot id and address to which we copy the log
2214 * Steps 1-4 below provide general overview of dirty page logging. See
2215 * kvm_get_dirty_log_protect() function description for additional details.
2217 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
2218 * always flush the TLB (step 4) even if previous step failed and the dirty
2219 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
2220 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
2221 * writes will be marked dirty for next log read.
2223 * 1. Take a snapshot of the bit and clear it if needed.
2224 * 2. Write protect the corresponding page.
2225 * 3. Copy the snapshot to the userspace.
2226 * 4. Flush TLB's if needed.
2228 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
2229 struct kvm_dirty_log *log)
2233 mutex_lock(&kvm->slots_lock);
2235 r = kvm_get_dirty_log_protect(kvm, log);
2237 mutex_unlock(&kvm->slots_lock);
2242 * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap
2243 * and reenable dirty page tracking for the corresponding pages.
2244 * @kvm: pointer to kvm instance
2245 * @log: slot id and address from which to fetch the bitmap of dirty pages
2247 static int kvm_clear_dirty_log_protect(struct kvm *kvm,
2248 struct kvm_clear_dirty_log *log)
2250 struct kvm_memslots *slots;
2251 struct kvm_memory_slot *memslot;
2255 unsigned long *dirty_bitmap;
2256 unsigned long *dirty_bitmap_buffer;
2259 /* Dirty ring tracking may be exclusive to dirty log tracking */
2260 if (!kvm_use_dirty_bitmap(kvm))
2263 as_id = log->slot >> 16;
2264 id = (u16)log->slot;
2265 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
2268 if (log->first_page & 63)
2271 slots = __kvm_memslots(kvm, as_id);
2272 memslot = id_to_memslot(slots, id);
2273 if (!memslot || !memslot->dirty_bitmap)
2276 dirty_bitmap = memslot->dirty_bitmap;
2278 n = ALIGN(log->num_pages, BITS_PER_LONG) / 8;
2280 if (log->first_page > memslot->npages ||
2281 log->num_pages > memslot->npages - log->first_page ||
2282 (log->num_pages < memslot->npages - log->first_page && (log->num_pages & 63)))
2285 kvm_arch_sync_dirty_log(kvm, memslot);
2288 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
2289 if (copy_from_user(dirty_bitmap_buffer, log->dirty_bitmap, n))
2293 for (offset = log->first_page, i = offset / BITS_PER_LONG,
2294 n = DIV_ROUND_UP(log->num_pages, BITS_PER_LONG); n--;
2295 i++, offset += BITS_PER_LONG) {
2296 unsigned long mask = *dirty_bitmap_buffer++;
2297 atomic_long_t *p = (atomic_long_t *) &dirty_bitmap[i];
2301 mask &= atomic_long_fetch_andnot(mask, p);
2304 * mask contains the bits that really have been cleared. This
2305 * never includes any bits beyond the length of the memslot (if
2306 * the length is not aligned to 64 pages), therefore it is not
2307 * a problem if userspace sets them in log->dirty_bitmap.
2311 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
2315 KVM_MMU_UNLOCK(kvm);
2318 kvm_arch_flush_remote_tlbs_memslot(kvm, memslot);
2323 static int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm,
2324 struct kvm_clear_dirty_log *log)
2328 mutex_lock(&kvm->slots_lock);
2330 r = kvm_clear_dirty_log_protect(kvm, log);
2332 mutex_unlock(&kvm->slots_lock);
2335 #endif /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
2337 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
2339 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
2341 EXPORT_SYMBOL_GPL(gfn_to_memslot);
2343 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
2345 struct kvm_memslots *slots = kvm_vcpu_memslots(vcpu);
2346 u64 gen = slots->generation;
2347 struct kvm_memory_slot *slot;
2350 * This also protects against using a memslot from a different address space,
2351 * since different address spaces have different generation numbers.
2353 if (unlikely(gen != vcpu->last_used_slot_gen)) {
2354 vcpu->last_used_slot = NULL;
2355 vcpu->last_used_slot_gen = gen;
2358 slot = try_get_memslot(vcpu->last_used_slot, gfn);
2363 * Fall back to searching all memslots. We purposely use
2364 * search_memslots() instead of __gfn_to_memslot() to avoid
2365 * thrashing the VM-wide last_used_slot in kvm_memslots.
2367 slot = search_memslots(slots, gfn, false);
2369 vcpu->last_used_slot = slot;
2376 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
2378 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
2380 return kvm_is_visible_memslot(memslot);
2382 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
2384 bool kvm_vcpu_is_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
2386 struct kvm_memory_slot *memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2388 return kvm_is_visible_memslot(memslot);
2390 EXPORT_SYMBOL_GPL(kvm_vcpu_is_visible_gfn);
2392 unsigned long kvm_host_page_size(struct kvm_vcpu *vcpu, gfn_t gfn)
2394 struct vm_area_struct *vma;
2395 unsigned long addr, size;
2399 addr = kvm_vcpu_gfn_to_hva_prot(vcpu, gfn, NULL);
2400 if (kvm_is_error_hva(addr))
2403 mmap_read_lock(current->mm);
2404 vma = find_vma(current->mm, addr);
2408 size = vma_kernel_pagesize(vma);
2411 mmap_read_unlock(current->mm);
2416 static bool memslot_is_readonly(const struct kvm_memory_slot *slot)
2418 return slot->flags & KVM_MEM_READONLY;
2421 static unsigned long __gfn_to_hva_many(const struct kvm_memory_slot *slot, gfn_t gfn,
2422 gfn_t *nr_pages, bool write)
2424 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
2425 return KVM_HVA_ERR_BAD;
2427 if (memslot_is_readonly(slot) && write)
2428 return KVM_HVA_ERR_RO_BAD;
2431 *nr_pages = slot->npages - (gfn - slot->base_gfn);
2433 return __gfn_to_hva_memslot(slot, gfn);
2436 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
2439 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
2442 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
2445 return gfn_to_hva_many(slot, gfn, NULL);
2447 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
2449 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
2451 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
2453 EXPORT_SYMBOL_GPL(gfn_to_hva);
2455 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
2457 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
2459 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
2462 * Return the hva of a @gfn and the R/W attribute if possible.
2464 * @slot: the kvm_memory_slot which contains @gfn
2465 * @gfn: the gfn to be translated
2466 * @writable: used to return the read/write attribute of the @slot if the hva
2467 * is valid and @writable is not NULL
2469 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
2470 gfn_t gfn, bool *writable)
2472 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
2474 if (!kvm_is_error_hva(hva) && writable)
2475 *writable = !memslot_is_readonly(slot);
2480 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
2482 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2484 return gfn_to_hva_memslot_prot(slot, gfn, writable);
2487 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
2489 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2491 return gfn_to_hva_memslot_prot(slot, gfn, writable);
2494 static inline int check_user_page_hwpoison(unsigned long addr)
2496 int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
2498 rc = get_user_pages(addr, 1, flags, NULL, NULL);
2499 return rc == -EHWPOISON;
2503 * The fast path to get the writable pfn which will be stored in @pfn,
2504 * true indicates success, otherwise false is returned. It's also the
2505 * only part that runs if we can in atomic context.
2507 static bool hva_to_pfn_fast(unsigned long addr, bool write_fault,
2508 bool *writable, kvm_pfn_t *pfn)
2510 struct page *page[1];
2513 * Fast pin a writable pfn only if it is a write fault request
2514 * or the caller allows to map a writable pfn for a read fault
2517 if (!(write_fault || writable))
2520 if (get_user_page_fast_only(addr, FOLL_WRITE, page)) {
2521 *pfn = page_to_pfn(page[0]);
2532 * The slow path to get the pfn of the specified host virtual address,
2533 * 1 indicates success, -errno is returned if error is detected.
2535 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
2536 bool interruptible, bool *writable, kvm_pfn_t *pfn)
2538 unsigned int flags = FOLL_HWPOISON;
2545 *writable = write_fault;
2548 flags |= FOLL_WRITE;
2550 flags |= FOLL_NOWAIT;
2552 flags |= FOLL_INTERRUPTIBLE;
2554 npages = get_user_pages_unlocked(addr, 1, &page, flags);
2558 /* map read fault as writable if possible */
2559 if (unlikely(!write_fault) && writable) {
2562 if (get_user_page_fast_only(addr, FOLL_WRITE, &wpage)) {
2568 *pfn = page_to_pfn(page);
2572 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
2574 if (unlikely(!(vma->vm_flags & VM_READ)))
2577 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
2583 static int kvm_try_get_pfn(kvm_pfn_t pfn)
2585 struct page *page = kvm_pfn_to_refcounted_page(pfn);
2590 return get_page_unless_zero(page);
2593 static int hva_to_pfn_remapped(struct vm_area_struct *vma,
2594 unsigned long addr, bool write_fault,
2595 bool *writable, kvm_pfn_t *p_pfn)
2602 r = follow_pte(vma->vm_mm, addr, &ptep, &ptl);
2605 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
2606 * not call the fault handler, so do it here.
2608 bool unlocked = false;
2609 r = fixup_user_fault(current->mm, addr,
2610 (write_fault ? FAULT_FLAG_WRITE : 0),
2617 r = follow_pte(vma->vm_mm, addr, &ptep, &ptl);
2622 if (write_fault && !pte_write(*ptep)) {
2623 pfn = KVM_PFN_ERR_RO_FAULT;
2628 *writable = pte_write(*ptep);
2629 pfn = pte_pfn(*ptep);
2632 * Get a reference here because callers of *hva_to_pfn* and
2633 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
2634 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
2635 * set, but the kvm_try_get_pfn/kvm_release_pfn_clean pair will
2636 * simply do nothing for reserved pfns.
2638 * Whoever called remap_pfn_range is also going to call e.g.
2639 * unmap_mapping_range before the underlying pages are freed,
2640 * causing a call to our MMU notifier.
2642 * Certain IO or PFNMAP mappings can be backed with valid
2643 * struct pages, but be allocated without refcounting e.g.,
2644 * tail pages of non-compound higher order allocations, which
2645 * would then underflow the refcount when the caller does the
2646 * required put_page. Don't allow those pages here.
2648 if (!kvm_try_get_pfn(pfn))
2652 pte_unmap_unlock(ptep, ptl);
2659 * Pin guest page in memory and return its pfn.
2660 * @addr: host virtual address which maps memory to the guest
2661 * @atomic: whether this function can sleep
2662 * @interruptible: whether the process can be interrupted by non-fatal signals
2663 * @async: whether this function need to wait IO complete if the
2664 * host page is not in the memory
2665 * @write_fault: whether we should get a writable host page
2666 * @writable: whether it allows to map a writable host page for !@write_fault
2668 * The function will map a writable host page for these two cases:
2669 * 1): @write_fault = true
2670 * 2): @write_fault = false && @writable, @writable will tell the caller
2671 * whether the mapping is writable.
2673 kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool interruptible,
2674 bool *async, bool write_fault, bool *writable)
2676 struct vm_area_struct *vma;
2680 /* we can do it either atomically or asynchronously, not both */
2681 BUG_ON(atomic && async);
2683 if (hva_to_pfn_fast(addr, write_fault, writable, &pfn))
2687 return KVM_PFN_ERR_FAULT;
2689 npages = hva_to_pfn_slow(addr, async, write_fault, interruptible,
2693 if (npages == -EINTR)
2694 return KVM_PFN_ERR_SIGPENDING;
2696 mmap_read_lock(current->mm);
2697 if (npages == -EHWPOISON ||
2698 (!async && check_user_page_hwpoison(addr))) {
2699 pfn = KVM_PFN_ERR_HWPOISON;
2704 vma = vma_lookup(current->mm, addr);
2707 pfn = KVM_PFN_ERR_FAULT;
2708 else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
2709 r = hva_to_pfn_remapped(vma, addr, write_fault, writable, &pfn);
2713 pfn = KVM_PFN_ERR_FAULT;
2715 if (async && vma_is_valid(vma, write_fault))
2717 pfn = KVM_PFN_ERR_FAULT;
2720 mmap_read_unlock(current->mm);
2724 kvm_pfn_t __gfn_to_pfn_memslot(const struct kvm_memory_slot *slot, gfn_t gfn,
2725 bool atomic, bool interruptible, bool *async,
2726 bool write_fault, bool *writable, hva_t *hva)
2728 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
2733 if (addr == KVM_HVA_ERR_RO_BAD) {
2736 return KVM_PFN_ERR_RO_FAULT;
2739 if (kvm_is_error_hva(addr)) {
2742 return KVM_PFN_NOSLOT;
2745 /* Do not map writable pfn in the readonly memslot. */
2746 if (writable && memslot_is_readonly(slot)) {
2751 return hva_to_pfn(addr, atomic, interruptible, async, write_fault,
2754 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
2756 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
2759 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, false,
2760 NULL, write_fault, writable, NULL);
2762 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
2764 kvm_pfn_t gfn_to_pfn_memslot(const struct kvm_memory_slot *slot, gfn_t gfn)
2766 return __gfn_to_pfn_memslot(slot, gfn, false, false, NULL, true,
2769 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
2771 kvm_pfn_t gfn_to_pfn_memslot_atomic(const struct kvm_memory_slot *slot, gfn_t gfn)
2773 return __gfn_to_pfn_memslot(slot, gfn, true, false, NULL, true,
2776 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
2778 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
2780 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
2782 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
2784 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
2786 return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
2788 EXPORT_SYMBOL_GPL(gfn_to_pfn);
2790 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
2792 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
2794 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
2796 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
2797 struct page **pages, int nr_pages)
2802 addr = gfn_to_hva_many(slot, gfn, &entry);
2803 if (kvm_is_error_hva(addr))
2806 if (entry < nr_pages)
2809 return get_user_pages_fast_only(addr, nr_pages, FOLL_WRITE, pages);
2811 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
2814 * Do not use this helper unless you are absolutely certain the gfn _must_ be
2815 * backed by 'struct page'. A valid example is if the backing memslot is
2816 * controlled by KVM. Note, if the returned page is valid, it's refcount has
2817 * been elevated by gfn_to_pfn().
2819 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
2824 pfn = gfn_to_pfn(kvm, gfn);
2826 if (is_error_noslot_pfn(pfn))
2827 return KVM_ERR_PTR_BAD_PAGE;
2829 page = kvm_pfn_to_refcounted_page(pfn);
2831 return KVM_ERR_PTR_BAD_PAGE;
2835 EXPORT_SYMBOL_GPL(gfn_to_page);
2837 void kvm_release_pfn(kvm_pfn_t pfn, bool dirty)
2840 kvm_release_pfn_dirty(pfn);
2842 kvm_release_pfn_clean(pfn);
2845 int kvm_vcpu_map(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map)
2849 struct page *page = KVM_UNMAPPED_PAGE;
2854 pfn = gfn_to_pfn(vcpu->kvm, gfn);
2855 if (is_error_noslot_pfn(pfn))
2858 if (pfn_valid(pfn)) {
2859 page = pfn_to_page(pfn);
2861 #ifdef CONFIG_HAS_IOMEM
2863 hva = memremap(pfn_to_hpa(pfn), PAGE_SIZE, MEMREMAP_WB);
2877 EXPORT_SYMBOL_GPL(kvm_vcpu_map);
2879 void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map, bool dirty)
2887 if (map->page != KVM_UNMAPPED_PAGE)
2889 #ifdef CONFIG_HAS_IOMEM
2895 kvm_vcpu_mark_page_dirty(vcpu, map->gfn);
2897 kvm_release_pfn(map->pfn, dirty);
2902 EXPORT_SYMBOL_GPL(kvm_vcpu_unmap);
2904 static bool kvm_is_ad_tracked_page(struct page *page)
2907 * Per page-flags.h, pages tagged PG_reserved "should in general not be
2908 * touched (e.g. set dirty) except by its owner".
2910 return !PageReserved(page);
2913 static void kvm_set_page_dirty(struct page *page)
2915 if (kvm_is_ad_tracked_page(page))
2919 static void kvm_set_page_accessed(struct page *page)
2921 if (kvm_is_ad_tracked_page(page))
2922 mark_page_accessed(page);
2925 void kvm_release_page_clean(struct page *page)
2927 WARN_ON(is_error_page(page));
2929 kvm_set_page_accessed(page);
2932 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
2934 void kvm_release_pfn_clean(kvm_pfn_t pfn)
2938 if (is_error_noslot_pfn(pfn))
2941 page = kvm_pfn_to_refcounted_page(pfn);
2945 kvm_release_page_clean(page);
2947 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
2949 void kvm_release_page_dirty(struct page *page)
2951 WARN_ON(is_error_page(page));
2953 kvm_set_page_dirty(page);
2954 kvm_release_page_clean(page);
2956 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
2958 void kvm_release_pfn_dirty(kvm_pfn_t pfn)
2962 if (is_error_noslot_pfn(pfn))
2965 page = kvm_pfn_to_refcounted_page(pfn);
2969 kvm_release_page_dirty(page);
2971 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
2974 * Note, checking for an error/noslot pfn is the caller's responsibility when
2975 * directly marking a page dirty/accessed. Unlike the "release" helpers, the
2976 * "set" helpers are not to be used when the pfn might point at garbage.
2978 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
2980 if (WARN_ON(is_error_noslot_pfn(pfn)))
2984 kvm_set_page_dirty(pfn_to_page(pfn));
2986 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
2988 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
2990 if (WARN_ON(is_error_noslot_pfn(pfn)))
2994 kvm_set_page_accessed(pfn_to_page(pfn));
2996 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
2998 static int next_segment(unsigned long len, int offset)
3000 if (len > PAGE_SIZE - offset)
3001 return PAGE_SIZE - offset;
3006 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
3007 void *data, int offset, int len)
3012 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
3013 if (kvm_is_error_hva(addr))
3015 r = __copy_from_user(data, (void __user *)addr + offset, len);
3021 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
3024 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
3026 return __kvm_read_guest_page(slot, gfn, data, offset, len);
3028 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
3030 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
3031 int offset, int len)
3033 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
3035 return __kvm_read_guest_page(slot, gfn, data, offset, len);
3037 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
3039 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
3041 gfn_t gfn = gpa >> PAGE_SHIFT;
3043 int offset = offset_in_page(gpa);
3046 while ((seg = next_segment(len, offset)) != 0) {
3047 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
3057 EXPORT_SYMBOL_GPL(kvm_read_guest);
3059 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
3061 gfn_t gfn = gpa >> PAGE_SHIFT;
3063 int offset = offset_in_page(gpa);
3066 while ((seg = next_segment(len, offset)) != 0) {
3067 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
3077 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
3079 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
3080 void *data, int offset, unsigned long len)
3085 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
3086 if (kvm_is_error_hva(addr))
3088 pagefault_disable();
3089 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
3096 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
3097 void *data, unsigned long len)
3099 gfn_t gfn = gpa >> PAGE_SHIFT;
3100 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
3101 int offset = offset_in_page(gpa);
3103 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
3105 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
3107 static int __kvm_write_guest_page(struct kvm *kvm,
3108 struct kvm_memory_slot *memslot, gfn_t gfn,
3109 const void *data, int offset, int len)
3114 addr = gfn_to_hva_memslot(memslot, gfn);
3115 if (kvm_is_error_hva(addr))
3117 r = __copy_to_user((void __user *)addr + offset, data, len);
3120 mark_page_dirty_in_slot(kvm, memslot, gfn);
3124 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
3125 const void *data, int offset, int len)
3127 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
3129 return __kvm_write_guest_page(kvm, slot, gfn, data, offset, len);
3131 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
3133 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
3134 const void *data, int offset, int len)
3136 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
3138 return __kvm_write_guest_page(vcpu->kvm, slot, gfn, data, offset, len);
3140 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
3142 int kvm_write_guest(struct kvm *kvm, 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_write_guest_page(kvm, gfn, data, offset, seg);
3161 EXPORT_SYMBOL_GPL(kvm_write_guest);
3163 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
3166 gfn_t gfn = gpa >> PAGE_SHIFT;
3168 int offset = offset_in_page(gpa);
3171 while ((seg = next_segment(len, offset)) != 0) {
3172 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
3182 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
3184 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
3185 struct gfn_to_hva_cache *ghc,
3186 gpa_t gpa, unsigned long len)
3188 int offset = offset_in_page(gpa);
3189 gfn_t start_gfn = gpa >> PAGE_SHIFT;
3190 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
3191 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
3192 gfn_t nr_pages_avail;
3194 /* Update ghc->generation before performing any error checks. */
3195 ghc->generation = slots->generation;
3197 if (start_gfn > end_gfn) {
3198 ghc->hva = KVM_HVA_ERR_BAD;
3203 * If the requested region crosses two memslots, we still
3204 * verify that the entire region is valid here.
3206 for ( ; start_gfn <= end_gfn; start_gfn += nr_pages_avail) {
3207 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
3208 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
3210 if (kvm_is_error_hva(ghc->hva))
3214 /* Use the slow path for cross page reads and writes. */
3215 if (nr_pages_needed == 1)
3218 ghc->memslot = NULL;
3225 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3226 gpa_t gpa, unsigned long len)
3228 struct kvm_memslots *slots = kvm_memslots(kvm);
3229 return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len);
3231 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
3233 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3234 void *data, unsigned int offset,
3237 struct kvm_memslots *slots = kvm_memslots(kvm);
3239 gpa_t gpa = ghc->gpa + offset;
3241 if (WARN_ON_ONCE(len + offset > ghc->len))
3244 if (slots->generation != ghc->generation) {
3245 if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len))
3249 if (kvm_is_error_hva(ghc->hva))
3252 if (unlikely(!ghc->memslot))
3253 return kvm_write_guest(kvm, gpa, data, len);
3255 r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
3258 mark_page_dirty_in_slot(kvm, ghc->memslot, gpa >> PAGE_SHIFT);
3262 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached);
3264 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3265 void *data, unsigned long len)
3267 return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len);
3269 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
3271 int kvm_read_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3272 void *data, unsigned int offset,
3275 struct kvm_memslots *slots = kvm_memslots(kvm);
3277 gpa_t gpa = ghc->gpa + offset;
3279 if (WARN_ON_ONCE(len + offset > ghc->len))
3282 if (slots->generation != ghc->generation) {
3283 if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len))
3287 if (kvm_is_error_hva(ghc->hva))
3290 if (unlikely(!ghc->memslot))
3291 return kvm_read_guest(kvm, gpa, data, len);
3293 r = __copy_from_user(data, (void __user *)ghc->hva + offset, len);
3299 EXPORT_SYMBOL_GPL(kvm_read_guest_offset_cached);
3301 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3302 void *data, unsigned long len)
3304 return kvm_read_guest_offset_cached(kvm, ghc, data, 0, len);
3306 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
3308 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
3310 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
3311 gfn_t gfn = gpa >> PAGE_SHIFT;
3313 int offset = offset_in_page(gpa);
3316 while ((seg = next_segment(len, offset)) != 0) {
3317 ret = kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
3326 EXPORT_SYMBOL_GPL(kvm_clear_guest);
3328 void mark_page_dirty_in_slot(struct kvm *kvm,
3329 const struct kvm_memory_slot *memslot,
3332 struct kvm_vcpu *vcpu = kvm_get_running_vcpu();
3334 #ifdef CONFIG_HAVE_KVM_DIRTY_RING
3335 if (WARN_ON_ONCE(vcpu && vcpu->kvm != kvm))
3338 WARN_ON_ONCE(!vcpu && !kvm_arch_allow_write_without_running_vcpu(kvm));
3341 if (memslot && kvm_slot_dirty_track_enabled(memslot)) {
3342 unsigned long rel_gfn = gfn - memslot->base_gfn;
3343 u32 slot = (memslot->as_id << 16) | memslot->id;
3345 if (kvm->dirty_ring_size && vcpu)
3346 kvm_dirty_ring_push(vcpu, slot, rel_gfn);
3347 else if (memslot->dirty_bitmap)
3348 set_bit_le(rel_gfn, memslot->dirty_bitmap);
3351 EXPORT_SYMBOL_GPL(mark_page_dirty_in_slot);
3353 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
3355 struct kvm_memory_slot *memslot;
3357 memslot = gfn_to_memslot(kvm, gfn);
3358 mark_page_dirty_in_slot(kvm, memslot, gfn);
3360 EXPORT_SYMBOL_GPL(mark_page_dirty);
3362 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
3364 struct kvm_memory_slot *memslot;
3366 memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
3367 mark_page_dirty_in_slot(vcpu->kvm, memslot, gfn);
3369 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
3371 void kvm_sigset_activate(struct kvm_vcpu *vcpu)
3373 if (!vcpu->sigset_active)
3377 * This does a lockless modification of ->real_blocked, which is fine
3378 * because, only current can change ->real_blocked and all readers of
3379 * ->real_blocked don't care as long ->real_blocked is always a subset
3382 sigprocmask(SIG_SETMASK, &vcpu->sigset, ¤t->real_blocked);
3385 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu)
3387 if (!vcpu->sigset_active)
3390 sigprocmask(SIG_SETMASK, ¤t->real_blocked, NULL);
3391 sigemptyset(¤t->real_blocked);
3394 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
3396 unsigned int old, val, grow, grow_start;
3398 old = val = vcpu->halt_poll_ns;
3399 grow_start = READ_ONCE(halt_poll_ns_grow_start);
3400 grow = READ_ONCE(halt_poll_ns_grow);
3405 if (val < grow_start)
3408 vcpu->halt_poll_ns = val;
3410 trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
3413 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
3415 unsigned int old, val, shrink, grow_start;
3417 old = val = vcpu->halt_poll_ns;
3418 shrink = READ_ONCE(halt_poll_ns_shrink);
3419 grow_start = READ_ONCE(halt_poll_ns_grow_start);
3425 if (val < grow_start)
3428 vcpu->halt_poll_ns = val;
3429 trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
3432 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
3435 int idx = srcu_read_lock(&vcpu->kvm->srcu);
3437 if (kvm_arch_vcpu_runnable(vcpu))
3439 if (kvm_cpu_has_pending_timer(vcpu))
3441 if (signal_pending(current))
3443 if (kvm_check_request(KVM_REQ_UNBLOCK, vcpu))
3448 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3453 * Block the vCPU until the vCPU is runnable, an event arrives, or a signal is
3454 * pending. This is mostly used when halting a vCPU, but may also be used
3455 * directly for other vCPU non-runnable states, e.g. x86's Wait-For-SIPI.
3457 bool kvm_vcpu_block(struct kvm_vcpu *vcpu)
3459 struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu);
3460 bool waited = false;
3462 vcpu->stat.generic.blocking = 1;
3465 kvm_arch_vcpu_blocking(vcpu);
3466 prepare_to_rcuwait(wait);
3470 set_current_state(TASK_INTERRUPTIBLE);
3472 if (kvm_vcpu_check_block(vcpu) < 0)
3480 finish_rcuwait(wait);
3481 kvm_arch_vcpu_unblocking(vcpu);
3484 vcpu->stat.generic.blocking = 0;
3489 static inline void update_halt_poll_stats(struct kvm_vcpu *vcpu, ktime_t start,
3490 ktime_t end, bool success)
3492 struct kvm_vcpu_stat_generic *stats = &vcpu->stat.generic;
3493 u64 poll_ns = ktime_to_ns(ktime_sub(end, start));
3495 ++vcpu->stat.generic.halt_attempted_poll;
3498 ++vcpu->stat.generic.halt_successful_poll;
3500 if (!vcpu_valid_wakeup(vcpu))
3501 ++vcpu->stat.generic.halt_poll_invalid;
3503 stats->halt_poll_success_ns += poll_ns;
3504 KVM_STATS_LOG_HIST_UPDATE(stats->halt_poll_success_hist, poll_ns);
3506 stats->halt_poll_fail_ns += poll_ns;
3507 KVM_STATS_LOG_HIST_UPDATE(stats->halt_poll_fail_hist, poll_ns);
3511 static unsigned int kvm_vcpu_max_halt_poll_ns(struct kvm_vcpu *vcpu)
3513 struct kvm *kvm = vcpu->kvm;
3515 if (kvm->override_halt_poll_ns) {
3517 * Ensure kvm->max_halt_poll_ns is not read before
3518 * kvm->override_halt_poll_ns.
3520 * Pairs with the smp_wmb() when enabling KVM_CAP_HALT_POLL.
3523 return READ_ONCE(kvm->max_halt_poll_ns);
3526 return READ_ONCE(halt_poll_ns);
3530 * Emulate a vCPU halt condition, e.g. HLT on x86, WFI on arm, etc... If halt
3531 * polling is enabled, busy wait for a short time before blocking to avoid the
3532 * expensive block+unblock sequence if a wake event arrives soon after the vCPU
3535 void kvm_vcpu_halt(struct kvm_vcpu *vcpu)
3537 unsigned int max_halt_poll_ns = kvm_vcpu_max_halt_poll_ns(vcpu);
3538 bool halt_poll_allowed = !kvm_arch_no_poll(vcpu);
3539 ktime_t start, cur, poll_end;
3540 bool waited = false;
3544 if (vcpu->halt_poll_ns > max_halt_poll_ns)
3545 vcpu->halt_poll_ns = max_halt_poll_ns;
3547 do_halt_poll = halt_poll_allowed && vcpu->halt_poll_ns;
3549 start = cur = poll_end = ktime_get();
3551 ktime_t stop = ktime_add_ns(start, vcpu->halt_poll_ns);
3554 if (kvm_vcpu_check_block(vcpu) < 0)
3557 poll_end = cur = ktime_get();
3558 } while (kvm_vcpu_can_poll(cur, stop));
3561 waited = kvm_vcpu_block(vcpu);
3565 vcpu->stat.generic.halt_wait_ns +=
3566 ktime_to_ns(cur) - ktime_to_ns(poll_end);
3567 KVM_STATS_LOG_HIST_UPDATE(vcpu->stat.generic.halt_wait_hist,
3568 ktime_to_ns(cur) - ktime_to_ns(poll_end));
3571 /* The total time the vCPU was "halted", including polling time. */
3572 halt_ns = ktime_to_ns(cur) - ktime_to_ns(start);
3575 * Note, halt-polling is considered successful so long as the vCPU was
3576 * never actually scheduled out, i.e. even if the wake event arrived
3577 * after of the halt-polling loop itself, but before the full wait.
3580 update_halt_poll_stats(vcpu, start, poll_end, !waited);
3582 if (halt_poll_allowed) {
3583 /* Recompute the max halt poll time in case it changed. */
3584 max_halt_poll_ns = kvm_vcpu_max_halt_poll_ns(vcpu);
3586 if (!vcpu_valid_wakeup(vcpu)) {
3587 shrink_halt_poll_ns(vcpu);
3588 } else if (max_halt_poll_ns) {
3589 if (halt_ns <= vcpu->halt_poll_ns)
3591 /* we had a long block, shrink polling */
3592 else if (vcpu->halt_poll_ns &&
3593 halt_ns > max_halt_poll_ns)
3594 shrink_halt_poll_ns(vcpu);
3595 /* we had a short halt and our poll time is too small */
3596 else if (vcpu->halt_poll_ns < max_halt_poll_ns &&
3597 halt_ns < max_halt_poll_ns)
3598 grow_halt_poll_ns(vcpu);
3600 vcpu->halt_poll_ns = 0;
3604 trace_kvm_vcpu_wakeup(halt_ns, waited, vcpu_valid_wakeup(vcpu));
3606 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
3608 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
3610 if (__kvm_vcpu_wake_up(vcpu)) {
3611 WRITE_ONCE(vcpu->ready, true);
3612 ++vcpu->stat.generic.halt_wakeup;
3618 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
3622 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
3624 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
3628 if (kvm_vcpu_wake_up(vcpu))
3633 * The only state change done outside the vcpu mutex is IN_GUEST_MODE
3634 * to EXITING_GUEST_MODE. Therefore the moderately expensive "should
3635 * kick" check does not need atomic operations if kvm_vcpu_kick is used
3636 * within the vCPU thread itself.
3638 if (vcpu == __this_cpu_read(kvm_running_vcpu)) {
3639 if (vcpu->mode == IN_GUEST_MODE)
3640 WRITE_ONCE(vcpu->mode, EXITING_GUEST_MODE);
3645 * Note, the vCPU could get migrated to a different pCPU at any point
3646 * after kvm_arch_vcpu_should_kick(), which could result in sending an
3647 * IPI to the previous pCPU. But, that's ok because the purpose of the
3648 * IPI is to force the vCPU to leave IN_GUEST_MODE, and migrating the
3649 * vCPU also requires it to leave IN_GUEST_MODE.
3651 if (kvm_arch_vcpu_should_kick(vcpu)) {
3652 cpu = READ_ONCE(vcpu->cpu);
3653 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
3654 smp_send_reschedule(cpu);
3659 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
3660 #endif /* !CONFIG_S390 */
3662 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
3665 struct task_struct *task = NULL;
3669 pid = rcu_dereference(target->pid);
3671 task = get_pid_task(pid, PIDTYPE_PID);
3675 ret = yield_to(task, 1);
3676 put_task_struct(task);
3680 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
3683 * Helper that checks whether a VCPU is eligible for directed yield.
3684 * Most eligible candidate to yield is decided by following heuristics:
3686 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
3687 * (preempted lock holder), indicated by @in_spin_loop.
3688 * Set at the beginning and cleared at the end of interception/PLE handler.
3690 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
3691 * chance last time (mostly it has become eligible now since we have probably
3692 * yielded to lockholder in last iteration. This is done by toggling
3693 * @dy_eligible each time a VCPU checked for eligibility.)
3695 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
3696 * to preempted lock-holder could result in wrong VCPU selection and CPU
3697 * burning. Giving priority for a potential lock-holder increases lock
3700 * Since algorithm is based on heuristics, accessing another VCPU data without
3701 * locking does not harm. It may result in trying to yield to same VCPU, fail
3702 * and continue with next VCPU and so on.
3704 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
3706 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
3709 eligible = !vcpu->spin_loop.in_spin_loop ||
3710 vcpu->spin_loop.dy_eligible;
3712 if (vcpu->spin_loop.in_spin_loop)
3713 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
3722 * Unlike kvm_arch_vcpu_runnable, this function is called outside
3723 * a vcpu_load/vcpu_put pair. However, for most architectures
3724 * kvm_arch_vcpu_runnable does not require vcpu_load.
3726 bool __weak kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
3728 return kvm_arch_vcpu_runnable(vcpu);
3731 static bool vcpu_dy_runnable(struct kvm_vcpu *vcpu)
3733 if (kvm_arch_dy_runnable(vcpu))
3736 #ifdef CONFIG_KVM_ASYNC_PF
3737 if (!list_empty_careful(&vcpu->async_pf.done))
3744 bool __weak kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu)
3749 void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode)
3751 struct kvm *kvm = me->kvm;
3752 struct kvm_vcpu *vcpu;
3753 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
3759 kvm_vcpu_set_in_spin_loop(me, true);
3761 * We boost the priority of a VCPU that is runnable but not
3762 * currently running, because it got preempted by something
3763 * else and called schedule in __vcpu_run. Hopefully that
3764 * VCPU is holding the lock that we need and will release it.
3765 * We approximate round-robin by starting at the last boosted VCPU.
3767 for (pass = 0; pass < 2 && !yielded && try; pass++) {
3768 kvm_for_each_vcpu(i, vcpu, kvm) {
3769 if (!pass && i <= last_boosted_vcpu) {
3770 i = last_boosted_vcpu;
3772 } else if (pass && i > last_boosted_vcpu)
3774 if (!READ_ONCE(vcpu->ready))
3778 if (kvm_vcpu_is_blocking(vcpu) && !vcpu_dy_runnable(vcpu))
3780 if (READ_ONCE(vcpu->preempted) && yield_to_kernel_mode &&
3781 !kvm_arch_dy_has_pending_interrupt(vcpu) &&
3782 !kvm_arch_vcpu_in_kernel(vcpu))
3784 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
3787 yielded = kvm_vcpu_yield_to(vcpu);
3789 kvm->last_boosted_vcpu = i;
3791 } else if (yielded < 0) {
3798 kvm_vcpu_set_in_spin_loop(me, false);
3800 /* Ensure vcpu is not eligible during next spinloop */
3801 kvm_vcpu_set_dy_eligible(me, false);
3803 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
3805 static bool kvm_page_in_dirty_ring(struct kvm *kvm, unsigned long pgoff)
3807 #ifdef CONFIG_HAVE_KVM_DIRTY_RING
3808 return (pgoff >= KVM_DIRTY_LOG_PAGE_OFFSET) &&
3809 (pgoff < KVM_DIRTY_LOG_PAGE_OFFSET +
3810 kvm->dirty_ring_size / PAGE_SIZE);
3816 static vm_fault_t kvm_vcpu_fault(struct vm_fault *vmf)
3818 struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data;
3821 if (vmf->pgoff == 0)
3822 page = virt_to_page(vcpu->run);
3824 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
3825 page = virt_to_page(vcpu->arch.pio_data);
3827 #ifdef CONFIG_KVM_MMIO
3828 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
3829 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
3831 else if (kvm_page_in_dirty_ring(vcpu->kvm, vmf->pgoff))
3832 page = kvm_dirty_ring_get_page(
3834 vmf->pgoff - KVM_DIRTY_LOG_PAGE_OFFSET);
3836 return kvm_arch_vcpu_fault(vcpu, vmf);
3842 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
3843 .fault = kvm_vcpu_fault,
3846 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
3848 struct kvm_vcpu *vcpu = file->private_data;
3849 unsigned long pages = vma_pages(vma);
3851 if ((kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff) ||
3852 kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff + pages - 1)) &&
3853 ((vma->vm_flags & VM_EXEC) || !(vma->vm_flags & VM_SHARED)))
3856 vma->vm_ops = &kvm_vcpu_vm_ops;
3860 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
3862 struct kvm_vcpu *vcpu = filp->private_data;
3864 kvm_put_kvm(vcpu->kvm);
3868 static const struct file_operations kvm_vcpu_fops = {
3869 .release = kvm_vcpu_release,
3870 .unlocked_ioctl = kvm_vcpu_ioctl,
3871 .mmap = kvm_vcpu_mmap,
3872 .llseek = noop_llseek,
3873 KVM_COMPAT(kvm_vcpu_compat_ioctl),
3877 * Allocates an inode for the vcpu.
3879 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
3881 char name[8 + 1 + ITOA_MAX_LEN + 1];
3883 snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id);
3884 return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
3887 #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS
3888 static int vcpu_get_pid(void *data, u64 *val)
3890 struct kvm_vcpu *vcpu = data;
3891 *val = pid_nr(rcu_access_pointer(vcpu->pid));
3895 DEFINE_SIMPLE_ATTRIBUTE(vcpu_get_pid_fops, vcpu_get_pid, NULL, "%llu\n");
3897 static void kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
3899 struct dentry *debugfs_dentry;
3900 char dir_name[ITOA_MAX_LEN * 2];
3902 if (!debugfs_initialized())
3905 snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
3906 debugfs_dentry = debugfs_create_dir(dir_name,
3907 vcpu->kvm->debugfs_dentry);
3908 debugfs_create_file("pid", 0444, debugfs_dentry, vcpu,
3909 &vcpu_get_pid_fops);
3911 kvm_arch_create_vcpu_debugfs(vcpu, debugfs_dentry);
3916 * Creates some virtual cpus. Good luck creating more than one.
3918 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
3921 struct kvm_vcpu *vcpu;
3924 if (id >= KVM_MAX_VCPU_IDS)
3927 mutex_lock(&kvm->lock);
3928 if (kvm->created_vcpus >= kvm->max_vcpus) {
3929 mutex_unlock(&kvm->lock);
3933 r = kvm_arch_vcpu_precreate(kvm, id);
3935 mutex_unlock(&kvm->lock);
3939 kvm->created_vcpus++;
3940 mutex_unlock(&kvm->lock);
3942 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL_ACCOUNT);
3945 goto vcpu_decrement;
3948 BUILD_BUG_ON(sizeof(struct kvm_run) > PAGE_SIZE);
3949 page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
3954 vcpu->run = page_address(page);
3956 kvm_vcpu_init(vcpu, kvm, id);
3958 r = kvm_arch_vcpu_create(vcpu);
3960 goto vcpu_free_run_page;
3962 if (kvm->dirty_ring_size) {
3963 r = kvm_dirty_ring_alloc(&vcpu->dirty_ring,
3964 id, kvm->dirty_ring_size);
3966 goto arch_vcpu_destroy;
3969 mutex_lock(&kvm->lock);
3971 #ifdef CONFIG_LOCKDEP
3972 /* Ensure that lockdep knows vcpu->mutex is taken *inside* kvm->lock */
3973 mutex_lock(&vcpu->mutex);
3974 mutex_unlock(&vcpu->mutex);
3977 if (kvm_get_vcpu_by_id(kvm, id)) {
3979 goto unlock_vcpu_destroy;
3982 vcpu->vcpu_idx = atomic_read(&kvm->online_vcpus);
3983 r = xa_reserve(&kvm->vcpu_array, vcpu->vcpu_idx, GFP_KERNEL_ACCOUNT);
3985 goto unlock_vcpu_destroy;
3987 /* Now it's all set up, let userspace reach it */
3989 r = create_vcpu_fd(vcpu);
3991 goto kvm_put_xa_release;
3993 if (KVM_BUG_ON(!!xa_store(&kvm->vcpu_array, vcpu->vcpu_idx, vcpu, 0), kvm)) {
3995 goto kvm_put_xa_release;
3999 * Pairs with smp_rmb() in kvm_get_vcpu. Store the vcpu
4000 * pointer before kvm->online_vcpu's incremented value.
4003 atomic_inc(&kvm->online_vcpus);
4005 mutex_unlock(&kvm->lock);
4006 kvm_arch_vcpu_postcreate(vcpu);
4007 kvm_create_vcpu_debugfs(vcpu);
4011 kvm_put_kvm_no_destroy(kvm);
4012 xa_release(&kvm->vcpu_array, vcpu->vcpu_idx);
4013 unlock_vcpu_destroy:
4014 mutex_unlock(&kvm->lock);
4015 kvm_dirty_ring_free(&vcpu->dirty_ring);
4017 kvm_arch_vcpu_destroy(vcpu);
4019 free_page((unsigned long)vcpu->run);
4021 kmem_cache_free(kvm_vcpu_cache, vcpu);
4023 mutex_lock(&kvm->lock);
4024 kvm->created_vcpus--;
4025 mutex_unlock(&kvm->lock);
4029 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
4032 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
4033 vcpu->sigset_active = 1;
4034 vcpu->sigset = *sigset;
4036 vcpu->sigset_active = 0;
4040 static ssize_t kvm_vcpu_stats_read(struct file *file, char __user *user_buffer,
4041 size_t size, loff_t *offset)
4043 struct kvm_vcpu *vcpu = file->private_data;
4045 return kvm_stats_read(vcpu->stats_id, &kvm_vcpu_stats_header,
4046 &kvm_vcpu_stats_desc[0], &vcpu->stat,
4047 sizeof(vcpu->stat), user_buffer, size, offset);
4050 static const struct file_operations kvm_vcpu_stats_fops = {
4051 .read = kvm_vcpu_stats_read,
4052 .llseek = noop_llseek,
4055 static int kvm_vcpu_ioctl_get_stats_fd(struct kvm_vcpu *vcpu)
4059 char name[15 + ITOA_MAX_LEN + 1];
4061 snprintf(name, sizeof(name), "kvm-vcpu-stats:%d", vcpu->vcpu_id);
4063 fd = get_unused_fd_flags(O_CLOEXEC);
4067 file = anon_inode_getfile(name, &kvm_vcpu_stats_fops, vcpu, O_RDONLY);
4070 return PTR_ERR(file);
4072 file->f_mode |= FMODE_PREAD;
4073 fd_install(fd, file);
4078 static long kvm_vcpu_ioctl(struct file *filp,
4079 unsigned int ioctl, unsigned long arg)
4081 struct kvm_vcpu *vcpu = filp->private_data;
4082 void __user *argp = (void __user *)arg;
4084 struct kvm_fpu *fpu = NULL;
4085 struct kvm_sregs *kvm_sregs = NULL;
4087 if (vcpu->kvm->mm != current->mm || vcpu->kvm->vm_dead)
4090 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
4094 * Some architectures have vcpu ioctls that are asynchronous to vcpu
4095 * execution; mutex_lock() would break them.
4097 r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg);
4098 if (r != -ENOIOCTLCMD)
4101 if (mutex_lock_killable(&vcpu->mutex))
4109 oldpid = rcu_access_pointer(vcpu->pid);
4110 if (unlikely(oldpid != task_pid(current))) {
4111 /* The thread running this VCPU changed. */
4114 r = kvm_arch_vcpu_run_pid_change(vcpu);
4118 newpid = get_task_pid(current, PIDTYPE_PID);
4119 rcu_assign_pointer(vcpu->pid, newpid);
4124 r = kvm_arch_vcpu_ioctl_run(vcpu);
4125 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
4128 case KVM_GET_REGS: {
4129 struct kvm_regs *kvm_regs;
4132 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL_ACCOUNT);
4135 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
4139 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
4146 case KVM_SET_REGS: {
4147 struct kvm_regs *kvm_regs;
4149 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
4150 if (IS_ERR(kvm_regs)) {
4151 r = PTR_ERR(kvm_regs);
4154 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
4158 case KVM_GET_SREGS: {
4159 kvm_sregs = kzalloc(sizeof(struct kvm_sregs),
4160 GFP_KERNEL_ACCOUNT);
4164 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
4168 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
4173 case KVM_SET_SREGS: {
4174 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
4175 if (IS_ERR(kvm_sregs)) {
4176 r = PTR_ERR(kvm_sregs);
4180 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
4183 case KVM_GET_MP_STATE: {
4184 struct kvm_mp_state mp_state;
4186 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
4190 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
4195 case KVM_SET_MP_STATE: {
4196 struct kvm_mp_state mp_state;
4199 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
4201 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
4204 case KVM_TRANSLATE: {
4205 struct kvm_translation tr;
4208 if (copy_from_user(&tr, argp, sizeof(tr)))
4210 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
4214 if (copy_to_user(argp, &tr, sizeof(tr)))
4219 case KVM_SET_GUEST_DEBUG: {
4220 struct kvm_guest_debug dbg;
4223 if (copy_from_user(&dbg, argp, sizeof(dbg)))
4225 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
4228 case KVM_SET_SIGNAL_MASK: {
4229 struct kvm_signal_mask __user *sigmask_arg = argp;
4230 struct kvm_signal_mask kvm_sigmask;
4231 sigset_t sigset, *p;
4236 if (copy_from_user(&kvm_sigmask, argp,
4237 sizeof(kvm_sigmask)))
4240 if (kvm_sigmask.len != sizeof(sigset))
4243 if (copy_from_user(&sigset, sigmask_arg->sigset,
4248 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
4252 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL_ACCOUNT);
4256 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
4260 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
4266 fpu = memdup_user(argp, sizeof(*fpu));
4272 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
4275 case KVM_GET_STATS_FD: {
4276 r = kvm_vcpu_ioctl_get_stats_fd(vcpu);
4280 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
4283 mutex_unlock(&vcpu->mutex);
4289 #ifdef CONFIG_KVM_COMPAT
4290 static long kvm_vcpu_compat_ioctl(struct file *filp,
4291 unsigned int ioctl, unsigned long arg)
4293 struct kvm_vcpu *vcpu = filp->private_data;
4294 void __user *argp = compat_ptr(arg);
4297 if (vcpu->kvm->mm != current->mm || vcpu->kvm->vm_dead)
4301 case KVM_SET_SIGNAL_MASK: {
4302 struct kvm_signal_mask __user *sigmask_arg = argp;
4303 struct kvm_signal_mask kvm_sigmask;
4308 if (copy_from_user(&kvm_sigmask, argp,
4309 sizeof(kvm_sigmask)))
4312 if (kvm_sigmask.len != sizeof(compat_sigset_t))
4315 if (get_compat_sigset(&sigset,
4316 (compat_sigset_t __user *)sigmask_arg->sigset))
4318 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
4320 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
4324 r = kvm_vcpu_ioctl(filp, ioctl, arg);
4332 static int kvm_device_mmap(struct file *filp, struct vm_area_struct *vma)
4334 struct kvm_device *dev = filp->private_data;
4337 return dev->ops->mmap(dev, vma);
4342 static int kvm_device_ioctl_attr(struct kvm_device *dev,
4343 int (*accessor)(struct kvm_device *dev,
4344 struct kvm_device_attr *attr),
4347 struct kvm_device_attr attr;
4352 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
4355 return accessor(dev, &attr);
4358 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
4361 struct kvm_device *dev = filp->private_data;
4363 if (dev->kvm->mm != current->mm || dev->kvm->vm_dead)
4367 case KVM_SET_DEVICE_ATTR:
4368 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
4369 case KVM_GET_DEVICE_ATTR:
4370 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
4371 case KVM_HAS_DEVICE_ATTR:
4372 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
4374 if (dev->ops->ioctl)
4375 return dev->ops->ioctl(dev, ioctl, arg);
4381 static int kvm_device_release(struct inode *inode, struct file *filp)
4383 struct kvm_device *dev = filp->private_data;
4384 struct kvm *kvm = dev->kvm;
4386 if (dev->ops->release) {
4387 mutex_lock(&kvm->lock);
4388 list_del(&dev->vm_node);
4389 dev->ops->release(dev);
4390 mutex_unlock(&kvm->lock);
4397 static const struct file_operations kvm_device_fops = {
4398 .unlocked_ioctl = kvm_device_ioctl,
4399 .release = kvm_device_release,
4400 KVM_COMPAT(kvm_device_ioctl),
4401 .mmap = kvm_device_mmap,
4404 struct kvm_device *kvm_device_from_filp(struct file *filp)
4406 if (filp->f_op != &kvm_device_fops)
4409 return filp->private_data;
4412 static const struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
4413 #ifdef CONFIG_KVM_MPIC
4414 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
4415 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
4419 int kvm_register_device_ops(const struct kvm_device_ops *ops, u32 type)
4421 if (type >= ARRAY_SIZE(kvm_device_ops_table))
4424 if (kvm_device_ops_table[type] != NULL)
4427 kvm_device_ops_table[type] = ops;
4431 void kvm_unregister_device_ops(u32 type)
4433 if (kvm_device_ops_table[type] != NULL)
4434 kvm_device_ops_table[type] = NULL;
4437 static int kvm_ioctl_create_device(struct kvm *kvm,
4438 struct kvm_create_device *cd)
4440 const struct kvm_device_ops *ops;
4441 struct kvm_device *dev;
4442 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
4446 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
4449 type = array_index_nospec(cd->type, ARRAY_SIZE(kvm_device_ops_table));
4450 ops = kvm_device_ops_table[type];
4457 dev = kzalloc(sizeof(*dev), GFP_KERNEL_ACCOUNT);
4464 mutex_lock(&kvm->lock);
4465 ret = ops->create(dev, type);
4467 mutex_unlock(&kvm->lock);
4471 list_add(&dev->vm_node, &kvm->devices);
4472 mutex_unlock(&kvm->lock);
4478 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
4480 kvm_put_kvm_no_destroy(kvm);
4481 mutex_lock(&kvm->lock);
4482 list_del(&dev->vm_node);
4485 mutex_unlock(&kvm->lock);
4495 static int kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
4498 case KVM_CAP_USER_MEMORY:
4499 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
4500 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
4501 case KVM_CAP_INTERNAL_ERROR_DATA:
4502 #ifdef CONFIG_HAVE_KVM_MSI
4503 case KVM_CAP_SIGNAL_MSI:
4505 #ifdef CONFIG_HAVE_KVM_IRQFD
4508 case KVM_CAP_IOEVENTFD_ANY_LENGTH:
4509 case KVM_CAP_CHECK_EXTENSION_VM:
4510 case KVM_CAP_ENABLE_CAP_VM:
4511 case KVM_CAP_HALT_POLL:
4513 #ifdef CONFIG_KVM_MMIO
4514 case KVM_CAP_COALESCED_MMIO:
4515 return KVM_COALESCED_MMIO_PAGE_OFFSET;
4516 case KVM_CAP_COALESCED_PIO:
4519 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4520 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2:
4521 return KVM_DIRTY_LOG_MANUAL_CAPS;
4523 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
4524 case KVM_CAP_IRQ_ROUTING:
4525 return KVM_MAX_IRQ_ROUTES;
4527 #if KVM_ADDRESS_SPACE_NUM > 1
4528 case KVM_CAP_MULTI_ADDRESS_SPACE:
4529 return KVM_ADDRESS_SPACE_NUM;
4531 case KVM_CAP_NR_MEMSLOTS:
4532 return KVM_USER_MEM_SLOTS;
4533 case KVM_CAP_DIRTY_LOG_RING:
4534 #ifdef CONFIG_HAVE_KVM_DIRTY_RING_TSO
4535 return KVM_DIRTY_RING_MAX_ENTRIES * sizeof(struct kvm_dirty_gfn);
4539 case KVM_CAP_DIRTY_LOG_RING_ACQ_REL:
4540 #ifdef CONFIG_HAVE_KVM_DIRTY_RING_ACQ_REL
4541 return KVM_DIRTY_RING_MAX_ENTRIES * sizeof(struct kvm_dirty_gfn);
4545 #ifdef CONFIG_NEED_KVM_DIRTY_RING_WITH_BITMAP
4546 case KVM_CAP_DIRTY_LOG_RING_WITH_BITMAP:
4548 case KVM_CAP_BINARY_STATS_FD:
4549 case KVM_CAP_SYSTEM_EVENT_DATA:
4554 return kvm_vm_ioctl_check_extension(kvm, arg);
4557 static int kvm_vm_ioctl_enable_dirty_log_ring(struct kvm *kvm, u32 size)
4561 if (!KVM_DIRTY_LOG_PAGE_OFFSET)
4564 /* the size should be power of 2 */
4565 if (!size || (size & (size - 1)))
4568 /* Should be bigger to keep the reserved entries, or a page */
4569 if (size < kvm_dirty_ring_get_rsvd_entries() *
4570 sizeof(struct kvm_dirty_gfn) || size < PAGE_SIZE)
4573 if (size > KVM_DIRTY_RING_MAX_ENTRIES *
4574 sizeof(struct kvm_dirty_gfn))
4577 /* We only allow it to set once */
4578 if (kvm->dirty_ring_size)
4581 mutex_lock(&kvm->lock);
4583 if (kvm->created_vcpus) {
4584 /* We don't allow to change this value after vcpu created */
4587 kvm->dirty_ring_size = size;
4591 mutex_unlock(&kvm->lock);
4595 static int kvm_vm_ioctl_reset_dirty_pages(struct kvm *kvm)
4598 struct kvm_vcpu *vcpu;
4601 if (!kvm->dirty_ring_size)
4604 mutex_lock(&kvm->slots_lock);
4606 kvm_for_each_vcpu(i, vcpu, kvm)
4607 cleared += kvm_dirty_ring_reset(vcpu->kvm, &vcpu->dirty_ring);
4609 mutex_unlock(&kvm->slots_lock);
4612 kvm_flush_remote_tlbs(kvm);
4617 int __attribute__((weak)) kvm_vm_ioctl_enable_cap(struct kvm *kvm,
4618 struct kvm_enable_cap *cap)
4623 bool kvm_are_all_memslots_empty(struct kvm *kvm)
4627 lockdep_assert_held(&kvm->slots_lock);
4629 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
4630 if (!kvm_memslots_empty(__kvm_memslots(kvm, i)))
4636 EXPORT_SYMBOL_GPL(kvm_are_all_memslots_empty);
4638 static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm,
4639 struct kvm_enable_cap *cap)
4642 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4643 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2: {
4644 u64 allowed_options = KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE;
4646 if (cap->args[0] & KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE)
4647 allowed_options = KVM_DIRTY_LOG_MANUAL_CAPS;
4649 if (cap->flags || (cap->args[0] & ~allowed_options))
4651 kvm->manual_dirty_log_protect = cap->args[0];
4655 case KVM_CAP_HALT_POLL: {
4656 if (cap->flags || cap->args[0] != (unsigned int)cap->args[0])
4659 kvm->max_halt_poll_ns = cap->args[0];
4662 * Ensure kvm->override_halt_poll_ns does not become visible
4663 * before kvm->max_halt_poll_ns.
4665 * Pairs with the smp_rmb() in kvm_vcpu_max_halt_poll_ns().
4668 kvm->override_halt_poll_ns = true;
4672 case KVM_CAP_DIRTY_LOG_RING:
4673 case KVM_CAP_DIRTY_LOG_RING_ACQ_REL:
4674 if (!kvm_vm_ioctl_check_extension_generic(kvm, cap->cap))
4677 return kvm_vm_ioctl_enable_dirty_log_ring(kvm, cap->args[0]);
4678 case KVM_CAP_DIRTY_LOG_RING_WITH_BITMAP: {
4681 if (!IS_ENABLED(CONFIG_NEED_KVM_DIRTY_RING_WITH_BITMAP) ||
4682 !kvm->dirty_ring_size || cap->flags)
4685 mutex_lock(&kvm->slots_lock);
4688 * For simplicity, allow enabling ring+bitmap if and only if
4689 * there are no memslots, e.g. to ensure all memslots allocate
4690 * a bitmap after the capability is enabled.
4692 if (kvm_are_all_memslots_empty(kvm)) {
4693 kvm->dirty_ring_with_bitmap = true;
4697 mutex_unlock(&kvm->slots_lock);
4702 return kvm_vm_ioctl_enable_cap(kvm, cap);
4706 static ssize_t kvm_vm_stats_read(struct file *file, char __user *user_buffer,
4707 size_t size, loff_t *offset)
4709 struct kvm *kvm = file->private_data;
4711 return kvm_stats_read(kvm->stats_id, &kvm_vm_stats_header,
4712 &kvm_vm_stats_desc[0], &kvm->stat,
4713 sizeof(kvm->stat), user_buffer, size, offset);
4716 static const struct file_operations kvm_vm_stats_fops = {
4717 .read = kvm_vm_stats_read,
4718 .llseek = noop_llseek,
4721 static int kvm_vm_ioctl_get_stats_fd(struct kvm *kvm)
4726 fd = get_unused_fd_flags(O_CLOEXEC);
4730 file = anon_inode_getfile("kvm-vm-stats",
4731 &kvm_vm_stats_fops, kvm, O_RDONLY);
4734 return PTR_ERR(file);
4736 file->f_mode |= FMODE_PREAD;
4737 fd_install(fd, file);
4742 static long kvm_vm_ioctl(struct file *filp,
4743 unsigned int ioctl, unsigned long arg)
4745 struct kvm *kvm = filp->private_data;
4746 void __user *argp = (void __user *)arg;
4749 if (kvm->mm != current->mm || kvm->vm_dead)
4752 case KVM_CREATE_VCPU:
4753 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
4755 case KVM_ENABLE_CAP: {
4756 struct kvm_enable_cap cap;
4759 if (copy_from_user(&cap, argp, sizeof(cap)))
4761 r = kvm_vm_ioctl_enable_cap_generic(kvm, &cap);
4764 case KVM_SET_USER_MEMORY_REGION: {
4765 struct kvm_userspace_memory_region kvm_userspace_mem;
4768 if (copy_from_user(&kvm_userspace_mem, argp,
4769 sizeof(kvm_userspace_mem)))
4772 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
4775 case KVM_GET_DIRTY_LOG: {
4776 struct kvm_dirty_log log;
4779 if (copy_from_user(&log, argp, sizeof(log)))
4781 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
4784 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4785 case KVM_CLEAR_DIRTY_LOG: {
4786 struct kvm_clear_dirty_log log;
4789 if (copy_from_user(&log, argp, sizeof(log)))
4791 r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
4795 #ifdef CONFIG_KVM_MMIO
4796 case KVM_REGISTER_COALESCED_MMIO: {
4797 struct kvm_coalesced_mmio_zone zone;
4800 if (copy_from_user(&zone, argp, sizeof(zone)))
4802 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
4805 case KVM_UNREGISTER_COALESCED_MMIO: {
4806 struct kvm_coalesced_mmio_zone zone;
4809 if (copy_from_user(&zone, argp, sizeof(zone)))
4811 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
4816 struct kvm_irqfd data;
4819 if (copy_from_user(&data, argp, sizeof(data)))
4821 r = kvm_irqfd(kvm, &data);
4824 case KVM_IOEVENTFD: {
4825 struct kvm_ioeventfd data;
4828 if (copy_from_user(&data, argp, sizeof(data)))
4830 r = kvm_ioeventfd(kvm, &data);
4833 #ifdef CONFIG_HAVE_KVM_MSI
4834 case KVM_SIGNAL_MSI: {
4838 if (copy_from_user(&msi, argp, sizeof(msi)))
4840 r = kvm_send_userspace_msi(kvm, &msi);
4844 #ifdef __KVM_HAVE_IRQ_LINE
4845 case KVM_IRQ_LINE_STATUS:
4846 case KVM_IRQ_LINE: {
4847 struct kvm_irq_level irq_event;
4850 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
4853 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
4854 ioctl == KVM_IRQ_LINE_STATUS);
4859 if (ioctl == KVM_IRQ_LINE_STATUS) {
4860 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
4868 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
4869 case KVM_SET_GSI_ROUTING: {
4870 struct kvm_irq_routing routing;
4871 struct kvm_irq_routing __user *urouting;
4872 struct kvm_irq_routing_entry *entries = NULL;
4875 if (copy_from_user(&routing, argp, sizeof(routing)))
4878 if (!kvm_arch_can_set_irq_routing(kvm))
4880 if (routing.nr > KVM_MAX_IRQ_ROUTES)
4886 entries = vmemdup_user(urouting->entries,
4887 array_size(sizeof(*entries),
4889 if (IS_ERR(entries)) {
4890 r = PTR_ERR(entries);
4894 r = kvm_set_irq_routing(kvm, entries, routing.nr,
4899 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
4900 case KVM_CREATE_DEVICE: {
4901 struct kvm_create_device cd;
4904 if (copy_from_user(&cd, argp, sizeof(cd)))
4907 r = kvm_ioctl_create_device(kvm, &cd);
4912 if (copy_to_user(argp, &cd, sizeof(cd)))
4918 case KVM_CHECK_EXTENSION:
4919 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
4921 case KVM_RESET_DIRTY_RINGS:
4922 r = kvm_vm_ioctl_reset_dirty_pages(kvm);
4924 case KVM_GET_STATS_FD:
4925 r = kvm_vm_ioctl_get_stats_fd(kvm);
4928 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
4934 #ifdef CONFIG_KVM_COMPAT
4935 struct compat_kvm_dirty_log {
4939 compat_uptr_t dirty_bitmap; /* one bit per page */
4944 struct compat_kvm_clear_dirty_log {
4949 compat_uptr_t dirty_bitmap; /* one bit per page */
4954 long __weak kvm_arch_vm_compat_ioctl(struct file *filp, unsigned int ioctl,
4960 static long kvm_vm_compat_ioctl(struct file *filp,
4961 unsigned int ioctl, unsigned long arg)
4963 struct kvm *kvm = filp->private_data;
4966 if (kvm->mm != current->mm || kvm->vm_dead)
4969 r = kvm_arch_vm_compat_ioctl(filp, ioctl, arg);
4974 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4975 case KVM_CLEAR_DIRTY_LOG: {
4976 struct compat_kvm_clear_dirty_log compat_log;
4977 struct kvm_clear_dirty_log log;
4979 if (copy_from_user(&compat_log, (void __user *)arg,
4980 sizeof(compat_log)))
4982 log.slot = compat_log.slot;
4983 log.num_pages = compat_log.num_pages;
4984 log.first_page = compat_log.first_page;
4985 log.padding2 = compat_log.padding2;
4986 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
4988 r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
4992 case KVM_GET_DIRTY_LOG: {
4993 struct compat_kvm_dirty_log compat_log;
4994 struct kvm_dirty_log log;
4996 if (copy_from_user(&compat_log, (void __user *)arg,
4997 sizeof(compat_log)))
4999 log.slot = compat_log.slot;
5000 log.padding1 = compat_log.padding1;
5001 log.padding2 = compat_log.padding2;
5002 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
5004 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
5008 r = kvm_vm_ioctl(filp, ioctl, arg);
5014 static const struct file_operations kvm_vm_fops = {
5015 .release = kvm_vm_release,
5016 .unlocked_ioctl = kvm_vm_ioctl,
5017 .llseek = noop_llseek,
5018 KVM_COMPAT(kvm_vm_compat_ioctl),
5021 bool file_is_kvm(struct file *file)
5023 return file && file->f_op == &kvm_vm_fops;
5025 EXPORT_SYMBOL_GPL(file_is_kvm);
5027 static int kvm_dev_ioctl_create_vm(unsigned long type)
5029 char fdname[ITOA_MAX_LEN + 1];
5034 fd = get_unused_fd_flags(O_CLOEXEC);
5038 snprintf(fdname, sizeof(fdname), "%d", fd);
5040 kvm = kvm_create_vm(type, fdname);
5046 file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
5053 * Don't call kvm_put_kvm anymore at this point; file->f_op is
5054 * already set, with ->release() being kvm_vm_release(). In error
5055 * cases it will be called by the final fput(file) and will take
5056 * care of doing kvm_put_kvm(kvm).
5058 kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm);
5060 fd_install(fd, file);
5070 static long kvm_dev_ioctl(struct file *filp,
5071 unsigned int ioctl, unsigned long arg)
5076 case KVM_GET_API_VERSION:
5079 r = KVM_API_VERSION;
5082 r = kvm_dev_ioctl_create_vm(arg);
5084 case KVM_CHECK_EXTENSION:
5085 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
5087 case KVM_GET_VCPU_MMAP_SIZE:
5090 r = PAGE_SIZE; /* struct kvm_run */
5092 r += PAGE_SIZE; /* pio data page */
5094 #ifdef CONFIG_KVM_MMIO
5095 r += PAGE_SIZE; /* coalesced mmio ring page */
5098 case KVM_TRACE_ENABLE:
5099 case KVM_TRACE_PAUSE:
5100 case KVM_TRACE_DISABLE:
5104 return kvm_arch_dev_ioctl(filp, ioctl, arg);
5110 static struct file_operations kvm_chardev_ops = {
5111 .unlocked_ioctl = kvm_dev_ioctl,
5112 .llseek = noop_llseek,
5113 KVM_COMPAT(kvm_dev_ioctl),
5116 static struct miscdevice kvm_dev = {
5122 #ifdef CONFIG_KVM_GENERIC_HARDWARE_ENABLING
5123 __visible bool kvm_rebooting;
5124 EXPORT_SYMBOL_GPL(kvm_rebooting);
5126 static DEFINE_PER_CPU(bool, hardware_enabled);
5127 static int kvm_usage_count;
5129 static int __hardware_enable_nolock(void)
5131 if (__this_cpu_read(hardware_enabled))
5134 if (kvm_arch_hardware_enable()) {
5135 pr_info("kvm: enabling virtualization on CPU%d failed\n",
5136 raw_smp_processor_id());
5140 __this_cpu_write(hardware_enabled, true);
5144 static void hardware_enable_nolock(void *failed)
5146 if (__hardware_enable_nolock())
5150 static int kvm_online_cpu(unsigned int cpu)
5155 * Abort the CPU online process if hardware virtualization cannot
5156 * be enabled. Otherwise running VMs would encounter unrecoverable
5157 * errors when scheduled to this CPU.
5159 mutex_lock(&kvm_lock);
5160 if (kvm_usage_count)
5161 ret = __hardware_enable_nolock();
5162 mutex_unlock(&kvm_lock);
5166 static void hardware_disable_nolock(void *junk)
5169 * Note, hardware_disable_all_nolock() tells all online CPUs to disable
5170 * hardware, not just CPUs that successfully enabled hardware!
5172 if (!__this_cpu_read(hardware_enabled))
5175 kvm_arch_hardware_disable();
5177 __this_cpu_write(hardware_enabled, false);
5180 static int kvm_offline_cpu(unsigned int cpu)
5182 mutex_lock(&kvm_lock);
5183 if (kvm_usage_count)
5184 hardware_disable_nolock(NULL);
5185 mutex_unlock(&kvm_lock);
5189 static void hardware_disable_all_nolock(void)
5191 BUG_ON(!kvm_usage_count);
5194 if (!kvm_usage_count)
5195 on_each_cpu(hardware_disable_nolock, NULL, 1);
5198 static void hardware_disable_all(void)
5201 mutex_lock(&kvm_lock);
5202 hardware_disable_all_nolock();
5203 mutex_unlock(&kvm_lock);
5207 static int hardware_enable_all(void)
5209 atomic_t failed = ATOMIC_INIT(0);
5213 * Do not enable hardware virtualization if the system is going down.
5214 * If userspace initiated a forced reboot, e.g. reboot -f, then it's
5215 * possible for an in-flight KVM_CREATE_VM to trigger hardware enabling
5216 * after kvm_reboot() is called. Note, this relies on system_state
5217 * being set _before_ kvm_reboot(), which is why KVM uses a syscore ops
5218 * hook instead of registering a dedicated reboot notifier (the latter
5219 * runs before system_state is updated).
5221 if (system_state == SYSTEM_HALT || system_state == SYSTEM_POWER_OFF ||
5222 system_state == SYSTEM_RESTART)
5226 * When onlining a CPU, cpu_online_mask is set before kvm_online_cpu()
5227 * is called, and so on_each_cpu() between them includes the CPU that
5228 * is being onlined. As a result, hardware_enable_nolock() may get
5229 * invoked before kvm_online_cpu(), which also enables hardware if the
5230 * usage count is non-zero. Disable CPU hotplug to avoid attempting to
5231 * enable hardware multiple times.
5234 mutex_lock(&kvm_lock);
5239 if (kvm_usage_count == 1) {
5240 on_each_cpu(hardware_enable_nolock, &failed, 1);
5242 if (atomic_read(&failed)) {
5243 hardware_disable_all_nolock();
5248 mutex_unlock(&kvm_lock);
5254 static void kvm_shutdown(void)
5257 * Disable hardware virtualization and set kvm_rebooting to indicate
5258 * that KVM has asynchronously disabled hardware virtualization, i.e.
5259 * that relevant errors and exceptions aren't entirely unexpected.
5260 * Some flavors of hardware virtualization need to be disabled before
5261 * transferring control to firmware (to perform shutdown/reboot), e.g.
5262 * on x86, virtualization can block INIT interrupts, which are used by
5263 * firmware to pull APs back under firmware control. Note, this path
5264 * is used for both shutdown and reboot scenarios, i.e. neither name is
5265 * 100% comprehensive.
5267 pr_info("kvm: exiting hardware virtualization\n");
5268 kvm_rebooting = true;
5269 on_each_cpu(hardware_disable_nolock, NULL, 1);
5272 static int kvm_suspend(void)
5275 * Secondary CPUs and CPU hotplug are disabled across the suspend/resume
5276 * callbacks, i.e. no need to acquire kvm_lock to ensure the usage count
5277 * is stable. Assert that kvm_lock is not held to ensure the system
5278 * isn't suspended while KVM is enabling hardware. Hardware enabling
5279 * can be preempted, but the task cannot be frozen until it has dropped
5280 * all locks (userspace tasks are frozen via a fake signal).
5282 lockdep_assert_not_held(&kvm_lock);
5283 lockdep_assert_irqs_disabled();
5285 if (kvm_usage_count)
5286 hardware_disable_nolock(NULL);
5290 static void kvm_resume(void)
5292 lockdep_assert_not_held(&kvm_lock);
5293 lockdep_assert_irqs_disabled();
5295 if (kvm_usage_count)
5296 WARN_ON_ONCE(__hardware_enable_nolock());
5299 static struct syscore_ops kvm_syscore_ops = {
5300 .suspend = kvm_suspend,
5301 .resume = kvm_resume,
5302 .shutdown = kvm_shutdown,
5304 #else /* CONFIG_KVM_GENERIC_HARDWARE_ENABLING */
5305 static int hardware_enable_all(void)
5310 static void hardware_disable_all(void)
5314 #endif /* CONFIG_KVM_GENERIC_HARDWARE_ENABLING */
5316 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
5320 for (i = 0; i < bus->dev_count; i++) {
5321 struct kvm_io_device *pos = bus->range[i].dev;
5323 kvm_iodevice_destructor(pos);
5328 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
5329 const struct kvm_io_range *r2)
5331 gpa_t addr1 = r1->addr;
5332 gpa_t addr2 = r2->addr;
5337 /* If r2->len == 0, match the exact address. If r2->len != 0,
5338 * accept any overlapping write. Any order is acceptable for
5339 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
5340 * we process all of them.
5353 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
5355 return kvm_io_bus_cmp(p1, p2);
5358 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
5359 gpa_t addr, int len)
5361 struct kvm_io_range *range, key;
5364 key = (struct kvm_io_range) {
5369 range = bsearch(&key, bus->range, bus->dev_count,
5370 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
5374 off = range - bus->range;
5376 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
5382 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
5383 struct kvm_io_range *range, const void *val)
5387 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
5391 while (idx < bus->dev_count &&
5392 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
5393 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
5402 /* kvm_io_bus_write - called under kvm->slots_lock */
5403 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
5404 int len, const void *val)
5406 struct kvm_io_bus *bus;
5407 struct kvm_io_range range;
5410 range = (struct kvm_io_range) {
5415 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
5418 r = __kvm_io_bus_write(vcpu, bus, &range, val);
5419 return r < 0 ? r : 0;
5421 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
5423 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
5424 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
5425 gpa_t addr, int len, const void *val, long cookie)
5427 struct kvm_io_bus *bus;
5428 struct kvm_io_range range;
5430 range = (struct kvm_io_range) {
5435 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
5439 /* First try the device referenced by cookie. */
5440 if ((cookie >= 0) && (cookie < bus->dev_count) &&
5441 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
5442 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
5447 * cookie contained garbage; fall back to search and return the
5448 * correct cookie value.
5450 return __kvm_io_bus_write(vcpu, bus, &range, val);
5453 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
5454 struct kvm_io_range *range, void *val)
5458 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
5462 while (idx < bus->dev_count &&
5463 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
5464 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
5473 /* kvm_io_bus_read - called under kvm->slots_lock */
5474 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
5477 struct kvm_io_bus *bus;
5478 struct kvm_io_range range;
5481 range = (struct kvm_io_range) {
5486 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
5489 r = __kvm_io_bus_read(vcpu, bus, &range, val);
5490 return r < 0 ? r : 0;
5493 /* Caller must hold slots_lock. */
5494 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
5495 int len, struct kvm_io_device *dev)
5498 struct kvm_io_bus *new_bus, *bus;
5499 struct kvm_io_range range;
5501 bus = kvm_get_bus(kvm, bus_idx);
5505 /* exclude ioeventfd which is limited by maximum fd */
5506 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
5509 new_bus = kmalloc(struct_size(bus, range, bus->dev_count + 1),
5510 GFP_KERNEL_ACCOUNT);
5514 range = (struct kvm_io_range) {
5520 for (i = 0; i < bus->dev_count; i++)
5521 if (kvm_io_bus_cmp(&bus->range[i], &range) > 0)
5524 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
5525 new_bus->dev_count++;
5526 new_bus->range[i] = range;
5527 memcpy(new_bus->range + i + 1, bus->range + i,
5528 (bus->dev_count - i) * sizeof(struct kvm_io_range));
5529 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
5530 synchronize_srcu_expedited(&kvm->srcu);
5536 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
5537 struct kvm_io_device *dev)
5540 struct kvm_io_bus *new_bus, *bus;
5542 lockdep_assert_held(&kvm->slots_lock);
5544 bus = kvm_get_bus(kvm, bus_idx);
5548 for (i = 0; i < bus->dev_count; i++) {
5549 if (bus->range[i].dev == dev) {
5554 if (i == bus->dev_count)
5557 new_bus = kmalloc(struct_size(bus, range, bus->dev_count - 1),
5558 GFP_KERNEL_ACCOUNT);
5560 memcpy(new_bus, bus, struct_size(bus, range, i));
5561 new_bus->dev_count--;
5562 memcpy(new_bus->range + i, bus->range + i + 1,
5563 flex_array_size(new_bus, range, new_bus->dev_count - i));
5566 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
5567 synchronize_srcu_expedited(&kvm->srcu);
5569 /* Destroy the old bus _after_ installing the (null) bus. */
5571 pr_err("kvm: failed to shrink bus, removing it completely\n");
5572 for (j = 0; j < bus->dev_count; j++) {
5575 kvm_iodevice_destructor(bus->range[j].dev);
5580 return new_bus ? 0 : -ENOMEM;
5583 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
5586 struct kvm_io_bus *bus;
5587 int dev_idx, srcu_idx;
5588 struct kvm_io_device *iodev = NULL;
5590 srcu_idx = srcu_read_lock(&kvm->srcu);
5592 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
5596 dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
5600 iodev = bus->range[dev_idx].dev;
5603 srcu_read_unlock(&kvm->srcu, srcu_idx);
5607 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
5609 static int kvm_debugfs_open(struct inode *inode, struct file *file,
5610 int (*get)(void *, u64 *), int (*set)(void *, u64),
5614 struct kvm_stat_data *stat_data = inode->i_private;
5617 * The debugfs files are a reference to the kvm struct which
5618 * is still valid when kvm_destroy_vm is called. kvm_get_kvm_safe
5619 * avoids the race between open and the removal of the debugfs directory.
5621 if (!kvm_get_kvm_safe(stat_data->kvm))
5624 ret = simple_attr_open(inode, file, get,
5625 kvm_stats_debugfs_mode(stat_data->desc) & 0222
5628 kvm_put_kvm(stat_data->kvm);
5633 static int kvm_debugfs_release(struct inode *inode, struct file *file)
5635 struct kvm_stat_data *stat_data = inode->i_private;
5637 simple_attr_release(inode, file);
5638 kvm_put_kvm(stat_data->kvm);
5643 static int kvm_get_stat_per_vm(struct kvm *kvm, size_t offset, u64 *val)
5645 *val = *(u64 *)((void *)(&kvm->stat) + offset);
5650 static int kvm_clear_stat_per_vm(struct kvm *kvm, size_t offset)
5652 *(u64 *)((void *)(&kvm->stat) + offset) = 0;
5657 static int kvm_get_stat_per_vcpu(struct kvm *kvm, size_t offset, u64 *val)
5660 struct kvm_vcpu *vcpu;
5664 kvm_for_each_vcpu(i, vcpu, kvm)
5665 *val += *(u64 *)((void *)(&vcpu->stat) + offset);
5670 static int kvm_clear_stat_per_vcpu(struct kvm *kvm, size_t offset)
5673 struct kvm_vcpu *vcpu;
5675 kvm_for_each_vcpu(i, vcpu, kvm)
5676 *(u64 *)((void *)(&vcpu->stat) + offset) = 0;
5681 static int kvm_stat_data_get(void *data, u64 *val)
5684 struct kvm_stat_data *stat_data = data;
5686 switch (stat_data->kind) {
5688 r = kvm_get_stat_per_vm(stat_data->kvm,
5689 stat_data->desc->desc.offset, val);
5692 r = kvm_get_stat_per_vcpu(stat_data->kvm,
5693 stat_data->desc->desc.offset, val);
5700 static int kvm_stat_data_clear(void *data, u64 val)
5703 struct kvm_stat_data *stat_data = data;
5708 switch (stat_data->kind) {
5710 r = kvm_clear_stat_per_vm(stat_data->kvm,
5711 stat_data->desc->desc.offset);
5714 r = kvm_clear_stat_per_vcpu(stat_data->kvm,
5715 stat_data->desc->desc.offset);
5722 static int kvm_stat_data_open(struct inode *inode, struct file *file)
5724 __simple_attr_check_format("%llu\n", 0ull);
5725 return kvm_debugfs_open(inode, file, kvm_stat_data_get,
5726 kvm_stat_data_clear, "%llu\n");
5729 static const struct file_operations stat_fops_per_vm = {
5730 .owner = THIS_MODULE,
5731 .open = kvm_stat_data_open,
5732 .release = kvm_debugfs_release,
5733 .read = simple_attr_read,
5734 .write = simple_attr_write,
5735 .llseek = no_llseek,
5738 static int vm_stat_get(void *_offset, u64 *val)
5740 unsigned offset = (long)_offset;
5745 mutex_lock(&kvm_lock);
5746 list_for_each_entry(kvm, &vm_list, vm_list) {
5747 kvm_get_stat_per_vm(kvm, offset, &tmp_val);
5750 mutex_unlock(&kvm_lock);
5754 static int vm_stat_clear(void *_offset, u64 val)
5756 unsigned offset = (long)_offset;
5762 mutex_lock(&kvm_lock);
5763 list_for_each_entry(kvm, &vm_list, vm_list) {
5764 kvm_clear_stat_per_vm(kvm, offset);
5766 mutex_unlock(&kvm_lock);
5771 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n");
5772 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_readonly_fops, vm_stat_get, NULL, "%llu\n");
5774 static int vcpu_stat_get(void *_offset, u64 *val)
5776 unsigned offset = (long)_offset;
5781 mutex_lock(&kvm_lock);
5782 list_for_each_entry(kvm, &vm_list, vm_list) {
5783 kvm_get_stat_per_vcpu(kvm, offset, &tmp_val);
5786 mutex_unlock(&kvm_lock);
5790 static int vcpu_stat_clear(void *_offset, u64 val)
5792 unsigned offset = (long)_offset;
5798 mutex_lock(&kvm_lock);
5799 list_for_each_entry(kvm, &vm_list, vm_list) {
5800 kvm_clear_stat_per_vcpu(kvm, offset);
5802 mutex_unlock(&kvm_lock);
5807 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear,
5809 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_readonly_fops, vcpu_stat_get, NULL, "%llu\n");
5811 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm)
5813 struct kobj_uevent_env *env;
5814 unsigned long long created, active;
5816 if (!kvm_dev.this_device || !kvm)
5819 mutex_lock(&kvm_lock);
5820 if (type == KVM_EVENT_CREATE_VM) {
5821 kvm_createvm_count++;
5823 } else if (type == KVM_EVENT_DESTROY_VM) {
5826 created = kvm_createvm_count;
5827 active = kvm_active_vms;
5828 mutex_unlock(&kvm_lock);
5830 env = kzalloc(sizeof(*env), GFP_KERNEL_ACCOUNT);
5834 add_uevent_var(env, "CREATED=%llu", created);
5835 add_uevent_var(env, "COUNT=%llu", active);
5837 if (type == KVM_EVENT_CREATE_VM) {
5838 add_uevent_var(env, "EVENT=create");
5839 kvm->userspace_pid = task_pid_nr(current);
5840 } else if (type == KVM_EVENT_DESTROY_VM) {
5841 add_uevent_var(env, "EVENT=destroy");
5843 add_uevent_var(env, "PID=%d", kvm->userspace_pid);
5845 if (!IS_ERR(kvm->debugfs_dentry)) {
5846 char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL_ACCOUNT);
5849 tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX);
5851 add_uevent_var(env, "STATS_PATH=%s", tmp);
5855 /* no need for checks, since we are adding at most only 5 keys */
5856 env->envp[env->envp_idx++] = NULL;
5857 kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp);
5861 static void kvm_init_debug(void)
5863 const struct file_operations *fops;
5864 const struct _kvm_stats_desc *pdesc;
5867 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
5869 for (i = 0; i < kvm_vm_stats_header.num_desc; ++i) {
5870 pdesc = &kvm_vm_stats_desc[i];
5871 if (kvm_stats_debugfs_mode(pdesc) & 0222)
5872 fops = &vm_stat_fops;
5874 fops = &vm_stat_readonly_fops;
5875 debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
5877 (void *)(long)pdesc->desc.offset, fops);
5880 for (i = 0; i < kvm_vcpu_stats_header.num_desc; ++i) {
5881 pdesc = &kvm_vcpu_stats_desc[i];
5882 if (kvm_stats_debugfs_mode(pdesc) & 0222)
5883 fops = &vcpu_stat_fops;
5885 fops = &vcpu_stat_readonly_fops;
5886 debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
5888 (void *)(long)pdesc->desc.offset, fops);
5893 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
5895 return container_of(pn, struct kvm_vcpu, preempt_notifier);
5898 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
5900 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
5902 WRITE_ONCE(vcpu->preempted, false);
5903 WRITE_ONCE(vcpu->ready, false);
5905 __this_cpu_write(kvm_running_vcpu, vcpu);
5906 kvm_arch_sched_in(vcpu, cpu);
5907 kvm_arch_vcpu_load(vcpu, cpu);
5910 static void kvm_sched_out(struct preempt_notifier *pn,
5911 struct task_struct *next)
5913 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
5915 if (current->on_rq) {
5916 WRITE_ONCE(vcpu->preempted, true);
5917 WRITE_ONCE(vcpu->ready, true);
5919 kvm_arch_vcpu_put(vcpu);
5920 __this_cpu_write(kvm_running_vcpu, NULL);
5924 * kvm_get_running_vcpu - get the vcpu running on the current CPU.
5926 * We can disable preemption locally around accessing the per-CPU variable,
5927 * and use the resolved vcpu pointer after enabling preemption again,
5928 * because even if the current thread is migrated to another CPU, reading
5929 * the per-CPU value later will give us the same value as we update the
5930 * per-CPU variable in the preempt notifier handlers.
5932 struct kvm_vcpu *kvm_get_running_vcpu(void)
5934 struct kvm_vcpu *vcpu;
5937 vcpu = __this_cpu_read(kvm_running_vcpu);
5942 EXPORT_SYMBOL_GPL(kvm_get_running_vcpu);
5945 * kvm_get_running_vcpus - get the per-CPU array of currently running vcpus.
5947 struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void)
5949 return &kvm_running_vcpu;
5952 #ifdef CONFIG_GUEST_PERF_EVENTS
5953 static unsigned int kvm_guest_state(void)
5955 struct kvm_vcpu *vcpu = kvm_get_running_vcpu();
5958 if (!kvm_arch_pmi_in_guest(vcpu))
5961 state = PERF_GUEST_ACTIVE;
5962 if (!kvm_arch_vcpu_in_kernel(vcpu))
5963 state |= PERF_GUEST_USER;
5968 static unsigned long kvm_guest_get_ip(void)
5970 struct kvm_vcpu *vcpu = kvm_get_running_vcpu();
5972 /* Retrieving the IP must be guarded by a call to kvm_guest_state(). */
5973 if (WARN_ON_ONCE(!kvm_arch_pmi_in_guest(vcpu)))
5976 return kvm_arch_vcpu_get_ip(vcpu);
5979 static struct perf_guest_info_callbacks kvm_guest_cbs = {
5980 .state = kvm_guest_state,
5981 .get_ip = kvm_guest_get_ip,
5982 .handle_intel_pt_intr = NULL,
5985 void kvm_register_perf_callbacks(unsigned int (*pt_intr_handler)(void))
5987 kvm_guest_cbs.handle_intel_pt_intr = pt_intr_handler;
5988 perf_register_guest_info_callbacks(&kvm_guest_cbs);
5990 void kvm_unregister_perf_callbacks(void)
5992 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
5996 int kvm_init(unsigned vcpu_size, unsigned vcpu_align, struct module *module)
6001 #ifdef CONFIG_KVM_GENERIC_HARDWARE_ENABLING
6002 r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_ONLINE, "kvm/cpu:online",
6003 kvm_online_cpu, kvm_offline_cpu);
6007 register_syscore_ops(&kvm_syscore_ops);
6010 /* A kmem cache lets us meet the alignment requirements of fx_save. */
6012 vcpu_align = __alignof__(struct kvm_vcpu);
6014 kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align,
6016 offsetof(struct kvm_vcpu, arch),
6017 offsetofend(struct kvm_vcpu, stats_id)
6018 - offsetof(struct kvm_vcpu, arch),
6020 if (!kvm_vcpu_cache) {
6022 goto err_vcpu_cache;
6025 for_each_possible_cpu(cpu) {
6026 if (!alloc_cpumask_var_node(&per_cpu(cpu_kick_mask, cpu),
6027 GFP_KERNEL, cpu_to_node(cpu))) {
6029 goto err_cpu_kick_mask;
6033 r = kvm_irqfd_init();
6037 r = kvm_async_pf_init();
6041 kvm_chardev_ops.owner = module;
6043 kvm_preempt_ops.sched_in = kvm_sched_in;
6044 kvm_preempt_ops.sched_out = kvm_sched_out;
6048 r = kvm_vfio_ops_init();
6049 if (WARN_ON_ONCE(r))
6053 * Registration _must_ be the very last thing done, as this exposes
6054 * /dev/kvm to userspace, i.e. all infrastructure must be setup!
6056 r = misc_register(&kvm_dev);
6058 pr_err("kvm: misc device register failed\n");
6065 kvm_vfio_ops_exit();
6067 kvm_async_pf_deinit();
6072 for_each_possible_cpu(cpu)
6073 free_cpumask_var(per_cpu(cpu_kick_mask, cpu));
6074 kmem_cache_destroy(kvm_vcpu_cache);
6076 #ifdef CONFIG_KVM_GENERIC_HARDWARE_ENABLING
6077 unregister_syscore_ops(&kvm_syscore_ops);
6078 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_ONLINE);
6082 EXPORT_SYMBOL_GPL(kvm_init);
6089 * Note, unregistering /dev/kvm doesn't strictly need to come first,
6090 * fops_get(), a.k.a. try_module_get(), prevents acquiring references
6091 * to KVM while the module is being stopped.
6093 misc_deregister(&kvm_dev);
6095 debugfs_remove_recursive(kvm_debugfs_dir);
6096 for_each_possible_cpu(cpu)
6097 free_cpumask_var(per_cpu(cpu_kick_mask, cpu));
6098 kmem_cache_destroy(kvm_vcpu_cache);
6099 kvm_vfio_ops_exit();
6100 kvm_async_pf_deinit();
6101 #ifdef CONFIG_KVM_GENERIC_HARDWARE_ENABLING
6102 unregister_syscore_ops(&kvm_syscore_ops);
6103 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_ONLINE);
6107 EXPORT_SYMBOL_GPL(kvm_exit);
6109 struct kvm_vm_worker_thread_context {
6111 struct task_struct *parent;
6112 struct completion init_done;
6113 kvm_vm_thread_fn_t thread_fn;
6118 static int kvm_vm_worker_thread(void *context)
6121 * The init_context is allocated on the stack of the parent thread, so
6122 * we have to locally copy anything that is needed beyond initialization
6124 struct kvm_vm_worker_thread_context *init_context = context;
6125 struct task_struct *parent;
6126 struct kvm *kvm = init_context->kvm;
6127 kvm_vm_thread_fn_t thread_fn = init_context->thread_fn;
6128 uintptr_t data = init_context->data;
6131 err = kthread_park(current);
6132 /* kthread_park(current) is never supposed to return an error */
6137 err = cgroup_attach_task_all(init_context->parent, current);
6139 kvm_err("%s: cgroup_attach_task_all failed with err %d\n",
6144 set_user_nice(current, task_nice(init_context->parent));
6147 init_context->err = err;
6148 complete(&init_context->init_done);
6149 init_context = NULL;
6154 /* Wait to be woken up by the spawner before proceeding. */
6157 if (!kthread_should_stop())
6158 err = thread_fn(kvm, data);
6162 * Move kthread back to its original cgroup to prevent it lingering in
6163 * the cgroup of the VM process, after the latter finishes its
6166 * kthread_stop() waits on the 'exited' completion condition which is
6167 * set in exit_mm(), via mm_release(), in do_exit(). However, the
6168 * kthread is removed from the cgroup in the cgroup_exit() which is
6169 * called after the exit_mm(). This causes the kthread_stop() to return
6170 * before the kthread actually quits the cgroup.
6173 parent = rcu_dereference(current->real_parent);
6174 get_task_struct(parent);
6176 cgroup_attach_task_all(parent, current);
6177 put_task_struct(parent);
6182 int kvm_vm_create_worker_thread(struct kvm *kvm, kvm_vm_thread_fn_t thread_fn,
6183 uintptr_t data, const char *name,
6184 struct task_struct **thread_ptr)
6186 struct kvm_vm_worker_thread_context init_context = {};
6187 struct task_struct *thread;
6190 init_context.kvm = kvm;
6191 init_context.parent = current;
6192 init_context.thread_fn = thread_fn;
6193 init_context.data = data;
6194 init_completion(&init_context.init_done);
6196 thread = kthread_run(kvm_vm_worker_thread, &init_context,
6197 "%s-%d", name, task_pid_nr(current));
6199 return PTR_ERR(thread);
6201 /* kthread_run is never supposed to return NULL */
6202 WARN_ON(thread == NULL);
6204 wait_for_completion(&init_context.init_done);
6206 if (!init_context.err)
6207 *thread_ptr = thread;
6209 return init_context.err;