1 // SPDX-License-Identifier: GPL-2.0-only
3 * Kernel-based Virtual Machine driver for Linux
5 * This module enables machines with Intel VT-x extensions to run virtual
6 * machines without emulation or binary translation.
8 * Copyright (C) 2006 Qumranet, Inc.
9 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
12 * Avi Kivity <avi@qumranet.com>
13 * Yaniv Kamay <yaniv@qumranet.com>
16 #include <kvm/iodev.h>
18 #include <linux/kvm_host.h>
19 #include <linux/kvm.h>
20 #include <linux/module.h>
21 #include <linux/errno.h>
22 #include <linux/percpu.h>
24 #include <linux/miscdevice.h>
25 #include <linux/vmalloc.h>
26 #include <linux/reboot.h>
27 #include <linux/debugfs.h>
28 #include <linux/highmem.h>
29 #include <linux/file.h>
30 #include <linux/syscore_ops.h>
31 #include <linux/cpu.h>
32 #include <linux/sched/signal.h>
33 #include <linux/sched/mm.h>
34 #include <linux/sched/stat.h>
35 #include <linux/cpumask.h>
36 #include <linux/smp.h>
37 #include <linux/anon_inodes.h>
38 #include <linux/profile.h>
39 #include <linux/kvm_para.h>
40 #include <linux/pagemap.h>
41 #include <linux/mman.h>
42 #include <linux/swap.h>
43 #include <linux/bitops.h>
44 #include <linux/spinlock.h>
45 #include <linux/compat.h>
46 #include <linux/srcu.h>
47 #include <linux/hugetlb.h>
48 #include <linux/slab.h>
49 #include <linux/sort.h>
50 #include <linux/bsearch.h>
52 #include <linux/lockdep.h>
53 #include <linux/kthread.h>
54 #include <linux/suspend.h>
56 #include <asm/processor.h>
57 #include <asm/ioctl.h>
58 #include <linux/uaccess.h>
60 #include "coalesced_mmio.h"
65 #define CREATE_TRACE_POINTS
66 #include <trace/events/kvm.h>
68 #include <linux/kvm_dirty_ring.h>
70 /* Worst case buffer size needed for holding an integer. */
71 #define ITOA_MAX_LEN 12
73 MODULE_AUTHOR("Qumranet");
74 MODULE_LICENSE("GPL");
76 /* Architectures should define their poll value according to the halt latency */
77 unsigned int halt_poll_ns = KVM_HALT_POLL_NS_DEFAULT;
78 module_param(halt_poll_ns, uint, 0644);
79 EXPORT_SYMBOL_GPL(halt_poll_ns);
81 /* Default doubles per-vcpu halt_poll_ns. */
82 unsigned int halt_poll_ns_grow = 2;
83 module_param(halt_poll_ns_grow, uint, 0644);
84 EXPORT_SYMBOL_GPL(halt_poll_ns_grow);
86 /* The start value to grow halt_poll_ns from */
87 unsigned int halt_poll_ns_grow_start = 10000; /* 10us */
88 module_param(halt_poll_ns_grow_start, uint, 0644);
89 EXPORT_SYMBOL_GPL(halt_poll_ns_grow_start);
91 /* Default resets per-vcpu halt_poll_ns . */
92 unsigned int halt_poll_ns_shrink;
93 module_param(halt_poll_ns_shrink, uint, 0644);
94 EXPORT_SYMBOL_GPL(halt_poll_ns_shrink);
99 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
102 DEFINE_MUTEX(kvm_lock);
103 static DEFINE_RAW_SPINLOCK(kvm_count_lock);
106 static cpumask_var_t cpus_hardware_enabled;
107 static int kvm_usage_count;
108 static atomic_t hardware_enable_failed;
110 static struct kmem_cache *kvm_vcpu_cache;
112 static __read_mostly struct preempt_ops kvm_preempt_ops;
113 static DEFINE_PER_CPU(struct kvm_vcpu *, kvm_running_vcpu);
115 struct dentry *kvm_debugfs_dir;
116 EXPORT_SYMBOL_GPL(kvm_debugfs_dir);
118 static const struct file_operations stat_fops_per_vm;
120 static struct file_operations kvm_chardev_ops;
122 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
124 #ifdef CONFIG_KVM_COMPAT
125 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
127 #define KVM_COMPAT(c) .compat_ioctl = (c)
130 * For architectures that don't implement a compat infrastructure,
131 * adopt a double line of defense:
132 * - Prevent a compat task from opening /dev/kvm
133 * - If the open has been done by a 64bit task, and the KVM fd
134 * passed to a compat task, let the ioctls fail.
136 static long kvm_no_compat_ioctl(struct file *file, unsigned int ioctl,
137 unsigned long arg) { return -EINVAL; }
139 static int kvm_no_compat_open(struct inode *inode, struct file *file)
141 return is_compat_task() ? -ENODEV : 0;
143 #define KVM_COMPAT(c) .compat_ioctl = kvm_no_compat_ioctl, \
144 .open = kvm_no_compat_open
146 static int hardware_enable_all(void);
147 static void hardware_disable_all(void);
149 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
151 __visible bool kvm_rebooting;
152 EXPORT_SYMBOL_GPL(kvm_rebooting);
154 #define KVM_EVENT_CREATE_VM 0
155 #define KVM_EVENT_DESTROY_VM 1
156 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm);
157 static unsigned long long kvm_createvm_count;
158 static unsigned long long kvm_active_vms;
160 static DEFINE_PER_CPU(cpumask_var_t, cpu_kick_mask);
162 __weak void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
163 unsigned long start, unsigned long end)
167 __weak void kvm_arch_guest_memory_reclaimed(struct kvm *kvm)
171 bool kvm_is_zone_device_page(struct page *page)
174 * The metadata used by is_zone_device_page() to determine whether or
175 * not a page is ZONE_DEVICE is guaranteed to be valid if and only if
176 * the device has been pinned, e.g. by get_user_pages(). WARN if the
177 * page_count() is zero to help detect bad usage of this helper.
179 if (WARN_ON_ONCE(!page_count(page)))
182 return is_zone_device_page(page);
186 * Returns a 'struct page' if the pfn is "valid" and backed by a refcounted
187 * page, NULL otherwise. Note, the list of refcounted PG_reserved page types
188 * is likely incomplete, it has been compiled purely through people wanting to
189 * back guest with a certain type of memory and encountering issues.
191 struct page *kvm_pfn_to_refcounted_page(kvm_pfn_t pfn)
198 page = pfn_to_page(pfn);
199 if (!PageReserved(page))
202 /* The ZERO_PAGE(s) is marked PG_reserved, but is refcounted. */
203 if (is_zero_pfn(pfn))
207 * ZONE_DEVICE pages currently set PG_reserved, but from a refcounting
208 * perspective they are "normal" pages, albeit with slightly different
211 if (kvm_is_zone_device_page(page))
218 * Switches to specified vcpu, until a matching vcpu_put()
220 void vcpu_load(struct kvm_vcpu *vcpu)
224 __this_cpu_write(kvm_running_vcpu, vcpu);
225 preempt_notifier_register(&vcpu->preempt_notifier);
226 kvm_arch_vcpu_load(vcpu, cpu);
229 EXPORT_SYMBOL_GPL(vcpu_load);
231 void vcpu_put(struct kvm_vcpu *vcpu)
234 kvm_arch_vcpu_put(vcpu);
235 preempt_notifier_unregister(&vcpu->preempt_notifier);
236 __this_cpu_write(kvm_running_vcpu, NULL);
239 EXPORT_SYMBOL_GPL(vcpu_put);
241 /* TODO: merge with kvm_arch_vcpu_should_kick */
242 static bool kvm_request_needs_ipi(struct kvm_vcpu *vcpu, unsigned req)
244 int mode = kvm_vcpu_exiting_guest_mode(vcpu);
247 * We need to wait for the VCPU to reenable interrupts and get out of
248 * READING_SHADOW_PAGE_TABLES mode.
250 if (req & KVM_REQUEST_WAIT)
251 return mode != OUTSIDE_GUEST_MODE;
254 * Need to kick a running VCPU, but otherwise there is nothing to do.
256 return mode == IN_GUEST_MODE;
259 static void ack_kick(void *_completed)
263 static inline bool kvm_kick_many_cpus(struct cpumask *cpus, bool wait)
265 if (cpumask_empty(cpus))
268 smp_call_function_many(cpus, ack_kick, NULL, wait);
272 static void kvm_make_vcpu_request(struct kvm_vcpu *vcpu, unsigned int req,
273 struct cpumask *tmp, int current_cpu)
277 if (likely(!(req & KVM_REQUEST_NO_ACTION)))
278 __kvm_make_request(req, vcpu);
280 if (!(req & KVM_REQUEST_NO_WAKEUP) && kvm_vcpu_wake_up(vcpu))
284 * Note, the vCPU could get migrated to a different pCPU at any point
285 * after kvm_request_needs_ipi(), which could result in sending an IPI
286 * to the previous pCPU. But, that's OK because the purpose of the IPI
287 * is to ensure the vCPU returns to OUTSIDE_GUEST_MODE, which is
288 * satisfied if the vCPU migrates. Entering READING_SHADOW_PAGE_TABLES
289 * after this point is also OK, as the requirement is only that KVM wait
290 * for vCPUs that were reading SPTEs _before_ any changes were
291 * finalized. See kvm_vcpu_kick() for more details on handling requests.
293 if (kvm_request_needs_ipi(vcpu, req)) {
294 cpu = READ_ONCE(vcpu->cpu);
295 if (cpu != -1 && cpu != current_cpu)
296 __cpumask_set_cpu(cpu, tmp);
300 bool kvm_make_vcpus_request_mask(struct kvm *kvm, unsigned int req,
301 unsigned long *vcpu_bitmap)
303 struct kvm_vcpu *vcpu;
304 struct cpumask *cpus;
310 cpus = this_cpu_cpumask_var_ptr(cpu_kick_mask);
313 for_each_set_bit(i, vcpu_bitmap, KVM_MAX_VCPUS) {
314 vcpu = kvm_get_vcpu(kvm, i);
317 kvm_make_vcpu_request(vcpu, req, cpus, me);
320 called = kvm_kick_many_cpus(cpus, !!(req & KVM_REQUEST_WAIT));
326 bool kvm_make_all_cpus_request_except(struct kvm *kvm, unsigned int req,
327 struct kvm_vcpu *except)
329 struct kvm_vcpu *vcpu;
330 struct cpumask *cpus;
337 cpus = this_cpu_cpumask_var_ptr(cpu_kick_mask);
340 kvm_for_each_vcpu(i, vcpu, kvm) {
343 kvm_make_vcpu_request(vcpu, req, cpus, me);
346 called = kvm_kick_many_cpus(cpus, !!(req & KVM_REQUEST_WAIT));
352 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
354 return kvm_make_all_cpus_request_except(kvm, req, NULL);
356 EXPORT_SYMBOL_GPL(kvm_make_all_cpus_request);
358 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
359 void kvm_flush_remote_tlbs(struct kvm *kvm)
361 ++kvm->stat.generic.remote_tlb_flush_requests;
364 * We want to publish modifications to the page tables before reading
365 * mode. Pairs with a memory barrier in arch-specific code.
366 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
367 * and smp_mb in walk_shadow_page_lockless_begin/end.
368 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
370 * There is already an smp_mb__after_atomic() before
371 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
374 if (!kvm_arch_flush_remote_tlb(kvm)
375 || kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
376 ++kvm->stat.generic.remote_tlb_flush;
378 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
381 static void kvm_flush_shadow_all(struct kvm *kvm)
383 kvm_arch_flush_shadow_all(kvm);
384 kvm_arch_guest_memory_reclaimed(kvm);
387 #ifdef KVM_ARCH_NR_OBJS_PER_MEMORY_CACHE
388 static inline void *mmu_memory_cache_alloc_obj(struct kvm_mmu_memory_cache *mc,
391 gfp_flags |= mc->gfp_zero;
394 return kmem_cache_alloc(mc->kmem_cache, gfp_flags);
396 return (void *)__get_free_page(gfp_flags);
399 int __kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache *mc, int capacity, int min)
401 gfp_t gfp = mc->gfp_custom ? mc->gfp_custom : GFP_KERNEL_ACCOUNT;
404 if (mc->nobjs >= min)
407 if (unlikely(!mc->objects)) {
408 if (WARN_ON_ONCE(!capacity))
411 mc->objects = kvmalloc_array(sizeof(void *), capacity, gfp);
415 mc->capacity = capacity;
418 /* It is illegal to request a different capacity across topups. */
419 if (WARN_ON_ONCE(mc->capacity != capacity))
422 while (mc->nobjs < mc->capacity) {
423 obj = mmu_memory_cache_alloc_obj(mc, gfp);
425 return mc->nobjs >= min ? 0 : -ENOMEM;
426 mc->objects[mc->nobjs++] = obj;
431 int kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache *mc, int min)
433 return __kvm_mmu_topup_memory_cache(mc, KVM_ARCH_NR_OBJS_PER_MEMORY_CACHE, min);
436 int kvm_mmu_memory_cache_nr_free_objects(struct kvm_mmu_memory_cache *mc)
441 void kvm_mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc)
445 kmem_cache_free(mc->kmem_cache, mc->objects[--mc->nobjs]);
447 free_page((unsigned long)mc->objects[--mc->nobjs]);
456 void *kvm_mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc)
460 if (WARN_ON(!mc->nobjs))
461 p = mmu_memory_cache_alloc_obj(mc, GFP_ATOMIC | __GFP_ACCOUNT);
463 p = mc->objects[--mc->nobjs];
469 static void kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
471 mutex_init(&vcpu->mutex);
476 #ifndef __KVM_HAVE_ARCH_WQP
477 rcuwait_init(&vcpu->wait);
479 kvm_async_pf_vcpu_init(vcpu);
481 kvm_vcpu_set_in_spin_loop(vcpu, false);
482 kvm_vcpu_set_dy_eligible(vcpu, false);
483 vcpu->preempted = false;
485 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
486 vcpu->last_used_slot = NULL;
488 /* Fill the stats id string for the vcpu */
489 snprintf(vcpu->stats_id, sizeof(vcpu->stats_id), "kvm-%d/vcpu-%d",
490 task_pid_nr(current), id);
493 static void kvm_vcpu_destroy(struct kvm_vcpu *vcpu)
495 kvm_arch_vcpu_destroy(vcpu);
496 kvm_dirty_ring_free(&vcpu->dirty_ring);
499 * No need for rcu_read_lock as VCPU_RUN is the only place that changes
500 * the vcpu->pid pointer, and at destruction time all file descriptors
503 put_pid(rcu_dereference_protected(vcpu->pid, 1));
505 free_page((unsigned long)vcpu->run);
506 kmem_cache_free(kvm_vcpu_cache, vcpu);
509 void kvm_destroy_vcpus(struct kvm *kvm)
512 struct kvm_vcpu *vcpu;
514 kvm_for_each_vcpu(i, vcpu, kvm) {
515 kvm_vcpu_destroy(vcpu);
516 xa_erase(&kvm->vcpu_array, i);
519 atomic_set(&kvm->online_vcpus, 0);
521 EXPORT_SYMBOL_GPL(kvm_destroy_vcpus);
523 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
524 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
526 return container_of(mn, struct kvm, mmu_notifier);
529 static void kvm_mmu_notifier_invalidate_range(struct mmu_notifier *mn,
530 struct mm_struct *mm,
531 unsigned long start, unsigned long end)
533 struct kvm *kvm = mmu_notifier_to_kvm(mn);
536 idx = srcu_read_lock(&kvm->srcu);
537 kvm_arch_mmu_notifier_invalidate_range(kvm, start, end);
538 srcu_read_unlock(&kvm->srcu, idx);
541 typedef bool (*hva_handler_t)(struct kvm *kvm, struct kvm_gfn_range *range);
543 typedef void (*on_lock_fn_t)(struct kvm *kvm, unsigned long start,
546 typedef void (*on_unlock_fn_t)(struct kvm *kvm);
548 struct kvm_hva_range {
552 hva_handler_t handler;
553 on_lock_fn_t on_lock;
554 on_unlock_fn_t on_unlock;
560 * Use a dedicated stub instead of NULL to indicate that there is no callback
561 * function/handler. The compiler technically can't guarantee that a real
562 * function will have a non-zero address, and so it will generate code to
563 * check for !NULL, whereas comparing against a stub will be elided at compile
564 * time (unless the compiler is getting long in the tooth, e.g. gcc 4.9).
566 static void kvm_null_fn(void)
570 #define IS_KVM_NULL_FN(fn) ((fn) == (void *)kvm_null_fn)
572 /* Iterate over each memslot intersecting [start, last] (inclusive) range */
573 #define kvm_for_each_memslot_in_hva_range(node, slots, start, last) \
574 for (node = interval_tree_iter_first(&slots->hva_tree, start, last); \
576 node = interval_tree_iter_next(node, start, last)) \
578 static __always_inline int __kvm_handle_hva_range(struct kvm *kvm,
579 const struct kvm_hva_range *range)
581 bool ret = false, locked = false;
582 struct kvm_gfn_range gfn_range;
583 struct kvm_memory_slot *slot;
584 struct kvm_memslots *slots;
587 if (WARN_ON_ONCE(range->end <= range->start))
590 /* A null handler is allowed if and only if on_lock() is provided. */
591 if (WARN_ON_ONCE(IS_KVM_NULL_FN(range->on_lock) &&
592 IS_KVM_NULL_FN(range->handler)))
595 idx = srcu_read_lock(&kvm->srcu);
597 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
598 struct interval_tree_node *node;
600 slots = __kvm_memslots(kvm, i);
601 kvm_for_each_memslot_in_hva_range(node, slots,
602 range->start, range->end - 1) {
603 unsigned long hva_start, hva_end;
605 slot = container_of(node, struct kvm_memory_slot, hva_node[slots->node_idx]);
606 hva_start = max(range->start, slot->userspace_addr);
607 hva_end = min(range->end, slot->userspace_addr +
608 (slot->npages << PAGE_SHIFT));
611 * To optimize for the likely case where the address
612 * range is covered by zero or one memslots, don't
613 * bother making these conditional (to avoid writes on
614 * the second or later invocation of the handler).
616 gfn_range.pte = range->pte;
617 gfn_range.may_block = range->may_block;
620 * {gfn(page) | page intersects with [hva_start, hva_end)} =
621 * {gfn_start, gfn_start+1, ..., gfn_end-1}.
623 gfn_range.start = hva_to_gfn_memslot(hva_start, slot);
624 gfn_range.end = hva_to_gfn_memslot(hva_end + PAGE_SIZE - 1, slot);
625 gfn_range.slot = slot;
630 if (!IS_KVM_NULL_FN(range->on_lock))
631 range->on_lock(kvm, range->start, range->end);
632 if (IS_KVM_NULL_FN(range->handler))
635 ret |= range->handler(kvm, &gfn_range);
639 if (range->flush_on_ret && ret)
640 kvm_flush_remote_tlbs(kvm);
644 if (!IS_KVM_NULL_FN(range->on_unlock))
645 range->on_unlock(kvm);
648 srcu_read_unlock(&kvm->srcu, idx);
650 /* The notifiers are averse to booleans. :-( */
654 static __always_inline int kvm_handle_hva_range(struct mmu_notifier *mn,
658 hva_handler_t handler)
660 struct kvm *kvm = mmu_notifier_to_kvm(mn);
661 const struct kvm_hva_range range = {
666 .on_lock = (void *)kvm_null_fn,
667 .on_unlock = (void *)kvm_null_fn,
668 .flush_on_ret = true,
672 return __kvm_handle_hva_range(kvm, &range);
675 static __always_inline int kvm_handle_hva_range_no_flush(struct mmu_notifier *mn,
678 hva_handler_t handler)
680 struct kvm *kvm = mmu_notifier_to_kvm(mn);
681 const struct kvm_hva_range range = {
686 .on_lock = (void *)kvm_null_fn,
687 .on_unlock = (void *)kvm_null_fn,
688 .flush_on_ret = false,
692 return __kvm_handle_hva_range(kvm, &range);
694 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
695 struct mm_struct *mm,
696 unsigned long address,
699 struct kvm *kvm = mmu_notifier_to_kvm(mn);
701 trace_kvm_set_spte_hva(address);
704 * .change_pte() must be surrounded by .invalidate_range_{start,end}().
705 * If mmu_notifier_count is zero, then no in-progress invalidations,
706 * including this one, found a relevant memslot at start(); rechecking
707 * memslots here is unnecessary. Note, a false positive (count elevated
708 * by a different invalidation) is sub-optimal but functionally ok.
710 WARN_ON_ONCE(!READ_ONCE(kvm->mn_active_invalidate_count));
711 if (!READ_ONCE(kvm->mmu_notifier_count))
714 kvm_handle_hva_range(mn, address, address + 1, pte, kvm_set_spte_gfn);
717 void kvm_inc_notifier_count(struct kvm *kvm, unsigned long start,
721 * The count increase must become visible at unlock time as no
722 * spte can be established without taking the mmu_lock and
723 * count is also read inside the mmu_lock critical section.
725 kvm->mmu_notifier_count++;
726 if (likely(kvm->mmu_notifier_count == 1)) {
727 kvm->mmu_notifier_range_start = start;
728 kvm->mmu_notifier_range_end = end;
731 * Fully tracking multiple concurrent ranges has diminishing
732 * returns. Keep things simple and just find the minimal range
733 * which includes the current and new ranges. As there won't be
734 * enough information to subtract a range after its invalidate
735 * completes, any ranges invalidated concurrently will
736 * accumulate and persist until all outstanding invalidates
739 kvm->mmu_notifier_range_start =
740 min(kvm->mmu_notifier_range_start, start);
741 kvm->mmu_notifier_range_end =
742 max(kvm->mmu_notifier_range_end, end);
746 static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
747 const struct mmu_notifier_range *range)
749 struct kvm *kvm = mmu_notifier_to_kvm(mn);
750 const struct kvm_hva_range hva_range = {
751 .start = range->start,
754 .handler = kvm_unmap_gfn_range,
755 .on_lock = kvm_inc_notifier_count,
756 .on_unlock = kvm_arch_guest_memory_reclaimed,
757 .flush_on_ret = true,
758 .may_block = mmu_notifier_range_blockable(range),
761 trace_kvm_unmap_hva_range(range->start, range->end);
764 * Prevent memslot modification between range_start() and range_end()
765 * so that conditionally locking provides the same result in both
766 * functions. Without that guarantee, the mmu_notifier_count
767 * adjustments will be imbalanced.
769 * Pairs with the decrement in range_end().
771 spin_lock(&kvm->mn_invalidate_lock);
772 kvm->mn_active_invalidate_count++;
773 spin_unlock(&kvm->mn_invalidate_lock);
776 * Invalidate pfn caches _before_ invalidating the secondary MMUs, i.e.
777 * before acquiring mmu_lock, to avoid holding mmu_lock while acquiring
778 * each cache's lock. There are relatively few caches in existence at
779 * any given time, and the caches themselves can check for hva overlap,
780 * i.e. don't need to rely on memslot overlap checks for performance.
781 * Because this runs without holding mmu_lock, the pfn caches must use
782 * mn_active_invalidate_count (see above) instead of mmu_notifier_count.
784 gfn_to_pfn_cache_invalidate_start(kvm, range->start, range->end,
785 hva_range.may_block);
787 __kvm_handle_hva_range(kvm, &hva_range);
792 void kvm_dec_notifier_count(struct kvm *kvm, unsigned long start,
796 * This sequence increase will notify the kvm page fault that
797 * the page that is going to be mapped in the spte could have
800 kvm->mmu_notifier_seq++;
803 * The above sequence increase must be visible before the
804 * below count decrease, which is ensured by the smp_wmb above
805 * in conjunction with the smp_rmb in mmu_notifier_retry().
807 kvm->mmu_notifier_count--;
810 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
811 const struct mmu_notifier_range *range)
813 struct kvm *kvm = mmu_notifier_to_kvm(mn);
814 const struct kvm_hva_range hva_range = {
815 .start = range->start,
818 .handler = (void *)kvm_null_fn,
819 .on_lock = kvm_dec_notifier_count,
820 .on_unlock = (void *)kvm_null_fn,
821 .flush_on_ret = false,
822 .may_block = mmu_notifier_range_blockable(range),
826 __kvm_handle_hva_range(kvm, &hva_range);
828 /* Pairs with the increment in range_start(). */
829 spin_lock(&kvm->mn_invalidate_lock);
830 wake = (--kvm->mn_active_invalidate_count == 0);
831 spin_unlock(&kvm->mn_invalidate_lock);
834 * There can only be one waiter, since the wait happens under
838 rcuwait_wake_up(&kvm->mn_memslots_update_rcuwait);
840 BUG_ON(kvm->mmu_notifier_count < 0);
843 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
844 struct mm_struct *mm,
848 trace_kvm_age_hva(start, end);
850 return kvm_handle_hva_range(mn, start, end, __pte(0), kvm_age_gfn);
853 static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
854 struct mm_struct *mm,
858 trace_kvm_age_hva(start, end);
861 * Even though we do not flush TLB, this will still adversely
862 * affect performance on pre-Haswell Intel EPT, where there is
863 * no EPT Access Bit to clear so that we have to tear down EPT
864 * tables instead. If we find this unacceptable, we can always
865 * add a parameter to kvm_age_hva so that it effectively doesn't
866 * do anything on clear_young.
868 * Also note that currently we never issue secondary TLB flushes
869 * from clear_young, leaving this job up to the regular system
870 * cadence. If we find this inaccurate, we might come up with a
871 * more sophisticated heuristic later.
873 return kvm_handle_hva_range_no_flush(mn, start, end, kvm_age_gfn);
876 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
877 struct mm_struct *mm,
878 unsigned long address)
880 trace_kvm_test_age_hva(address);
882 return kvm_handle_hva_range_no_flush(mn, address, address + 1,
886 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
887 struct mm_struct *mm)
889 struct kvm *kvm = mmu_notifier_to_kvm(mn);
892 idx = srcu_read_lock(&kvm->srcu);
893 kvm_flush_shadow_all(kvm);
894 srcu_read_unlock(&kvm->srcu, idx);
897 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
898 .invalidate_range = kvm_mmu_notifier_invalidate_range,
899 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
900 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
901 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
902 .clear_young = kvm_mmu_notifier_clear_young,
903 .test_young = kvm_mmu_notifier_test_young,
904 .change_pte = kvm_mmu_notifier_change_pte,
905 .release = kvm_mmu_notifier_release,
908 static int kvm_init_mmu_notifier(struct kvm *kvm)
910 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
911 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
914 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
916 static int kvm_init_mmu_notifier(struct kvm *kvm)
921 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
923 #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER
924 static int kvm_pm_notifier_call(struct notifier_block *bl,
928 struct kvm *kvm = container_of(bl, struct kvm, pm_notifier);
930 return kvm_arch_pm_notifier(kvm, state);
933 static void kvm_init_pm_notifier(struct kvm *kvm)
935 kvm->pm_notifier.notifier_call = kvm_pm_notifier_call;
936 /* Suspend KVM before we suspend ftrace, RCU, etc. */
937 kvm->pm_notifier.priority = INT_MAX;
938 register_pm_notifier(&kvm->pm_notifier);
941 static void kvm_destroy_pm_notifier(struct kvm *kvm)
943 unregister_pm_notifier(&kvm->pm_notifier);
945 #else /* !CONFIG_HAVE_KVM_PM_NOTIFIER */
946 static void kvm_init_pm_notifier(struct kvm *kvm)
950 static void kvm_destroy_pm_notifier(struct kvm *kvm)
953 #endif /* CONFIG_HAVE_KVM_PM_NOTIFIER */
955 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
957 if (!memslot->dirty_bitmap)
960 kvfree(memslot->dirty_bitmap);
961 memslot->dirty_bitmap = NULL;
964 /* This does not remove the slot from struct kvm_memslots data structures */
965 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
967 kvm_destroy_dirty_bitmap(slot);
969 kvm_arch_free_memslot(kvm, slot);
974 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
976 struct hlist_node *idnode;
977 struct kvm_memory_slot *memslot;
981 * The same memslot objects live in both active and inactive sets,
982 * arbitrarily free using index '1' so the second invocation of this
983 * function isn't operating over a structure with dangling pointers
984 * (even though this function isn't actually touching them).
986 if (!slots->node_idx)
989 hash_for_each_safe(slots->id_hash, bkt, idnode, memslot, id_node[1])
990 kvm_free_memslot(kvm, memslot);
993 static umode_t kvm_stats_debugfs_mode(const struct _kvm_stats_desc *pdesc)
995 switch (pdesc->desc.flags & KVM_STATS_TYPE_MASK) {
996 case KVM_STATS_TYPE_INSTANT:
998 case KVM_STATS_TYPE_CUMULATIVE:
999 case KVM_STATS_TYPE_PEAK:
1006 static void kvm_destroy_vm_debugfs(struct kvm *kvm)
1009 int kvm_debugfs_num_entries = kvm_vm_stats_header.num_desc +
1010 kvm_vcpu_stats_header.num_desc;
1012 if (IS_ERR(kvm->debugfs_dentry))
1015 debugfs_remove_recursive(kvm->debugfs_dentry);
1017 if (kvm->debugfs_stat_data) {
1018 for (i = 0; i < kvm_debugfs_num_entries; i++)
1019 kfree(kvm->debugfs_stat_data[i]);
1020 kfree(kvm->debugfs_stat_data);
1024 static int kvm_create_vm_debugfs(struct kvm *kvm, const char *fdname)
1026 static DEFINE_MUTEX(kvm_debugfs_lock);
1027 struct dentry *dent;
1028 char dir_name[ITOA_MAX_LEN * 2];
1029 struct kvm_stat_data *stat_data;
1030 const struct _kvm_stats_desc *pdesc;
1032 int kvm_debugfs_num_entries = kvm_vm_stats_header.num_desc +
1033 kvm_vcpu_stats_header.num_desc;
1035 if (!debugfs_initialized())
1038 snprintf(dir_name, sizeof(dir_name), "%d-%s", task_pid_nr(current), fdname);
1039 mutex_lock(&kvm_debugfs_lock);
1040 dent = debugfs_lookup(dir_name, kvm_debugfs_dir);
1042 pr_warn_ratelimited("KVM: debugfs: duplicate directory %s\n", dir_name);
1044 mutex_unlock(&kvm_debugfs_lock);
1047 dent = debugfs_create_dir(dir_name, kvm_debugfs_dir);
1048 mutex_unlock(&kvm_debugfs_lock);
1052 kvm->debugfs_dentry = dent;
1053 kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries,
1054 sizeof(*kvm->debugfs_stat_data),
1055 GFP_KERNEL_ACCOUNT);
1056 if (!kvm->debugfs_stat_data)
1059 for (i = 0; i < kvm_vm_stats_header.num_desc; ++i) {
1060 pdesc = &kvm_vm_stats_desc[i];
1061 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT);
1065 stat_data->kvm = kvm;
1066 stat_data->desc = pdesc;
1067 stat_data->kind = KVM_STAT_VM;
1068 kvm->debugfs_stat_data[i] = stat_data;
1069 debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
1070 kvm->debugfs_dentry, stat_data,
1074 for (i = 0; i < kvm_vcpu_stats_header.num_desc; ++i) {
1075 pdesc = &kvm_vcpu_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_VCPU;
1083 kvm->debugfs_stat_data[i + kvm_vm_stats_header.num_desc] = stat_data;
1084 debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
1085 kvm->debugfs_dentry, stat_data,
1089 ret = kvm_arch_create_vm_debugfs(kvm);
1091 kvm_destroy_vm_debugfs(kvm);
1099 * Called after the VM is otherwise initialized, but just before adding it to
1102 int __weak kvm_arch_post_init_vm(struct kvm *kvm)
1108 * Called just after removing the VM from the vm_list, but before doing any
1109 * other destruction.
1111 void __weak kvm_arch_pre_destroy_vm(struct kvm *kvm)
1116 * Called after per-vm debugfs created. When called kvm->debugfs_dentry should
1117 * be setup already, so we can create arch-specific debugfs entries under it.
1118 * Cleanup should be automatic done in kvm_destroy_vm_debugfs() recursively, so
1119 * a per-arch destroy interface is not needed.
1121 int __weak kvm_arch_create_vm_debugfs(struct kvm *kvm)
1126 static struct kvm *kvm_create_vm(unsigned long type)
1128 struct kvm *kvm = kvm_arch_alloc_vm();
1129 struct kvm_memslots *slots;
1134 return ERR_PTR(-ENOMEM);
1136 KVM_MMU_LOCK_INIT(kvm);
1137 mmgrab(current->mm);
1138 kvm->mm = current->mm;
1139 kvm_eventfd_init(kvm);
1140 mutex_init(&kvm->lock);
1141 mutex_init(&kvm->irq_lock);
1142 mutex_init(&kvm->slots_lock);
1143 mutex_init(&kvm->slots_arch_lock);
1144 spin_lock_init(&kvm->mn_invalidate_lock);
1145 rcuwait_init(&kvm->mn_memslots_update_rcuwait);
1146 xa_init(&kvm->vcpu_array);
1148 INIT_LIST_HEAD(&kvm->gpc_list);
1149 spin_lock_init(&kvm->gpc_lock);
1151 INIT_LIST_HEAD(&kvm->devices);
1152 kvm->max_vcpus = KVM_MAX_VCPUS;
1154 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
1157 * Force subsequent debugfs file creations to fail if the VM directory
1158 * is not created (by kvm_create_vm_debugfs()).
1160 kvm->debugfs_dentry = ERR_PTR(-ENOENT);
1162 snprintf(kvm->stats_id, sizeof(kvm->stats_id), "kvm-%d",
1163 task_pid_nr(current));
1165 if (init_srcu_struct(&kvm->srcu))
1166 goto out_err_no_srcu;
1167 if (init_srcu_struct(&kvm->irq_srcu))
1168 goto out_err_no_irq_srcu;
1170 refcount_set(&kvm->users_count, 1);
1171 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
1172 for (j = 0; j < 2; j++) {
1173 slots = &kvm->__memslots[i][j];
1175 atomic_long_set(&slots->last_used_slot, (unsigned long)NULL);
1176 slots->hva_tree = RB_ROOT_CACHED;
1177 slots->gfn_tree = RB_ROOT;
1178 hash_init(slots->id_hash);
1179 slots->node_idx = j;
1181 /* Generations must be different for each address space. */
1182 slots->generation = i;
1185 rcu_assign_pointer(kvm->memslots[i], &kvm->__memslots[i][0]);
1188 for (i = 0; i < KVM_NR_BUSES; i++) {
1189 rcu_assign_pointer(kvm->buses[i],
1190 kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL_ACCOUNT));
1192 goto out_err_no_arch_destroy_vm;
1195 kvm->max_halt_poll_ns = halt_poll_ns;
1197 r = kvm_arch_init_vm(kvm, type);
1199 goto out_err_no_arch_destroy_vm;
1201 r = hardware_enable_all();
1203 goto out_err_no_disable;
1205 #ifdef CONFIG_HAVE_KVM_IRQFD
1206 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
1209 r = kvm_init_mmu_notifier(kvm);
1211 goto out_err_no_mmu_notifier;
1213 r = kvm_arch_post_init_vm(kvm);
1217 mutex_lock(&kvm_lock);
1218 list_add(&kvm->vm_list, &vm_list);
1219 mutex_unlock(&kvm_lock);
1221 preempt_notifier_inc();
1222 kvm_init_pm_notifier(kvm);
1225 * When the fd passed to this ioctl() is opened it pins the module,
1226 * but try_module_get() also prevents getting a reference if the module
1227 * is in MODULE_STATE_GOING (e.g. if someone ran "rmmod --wait").
1229 if (!try_module_get(kvm_chardev_ops.owner)) {
1237 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1238 if (kvm->mmu_notifier.ops)
1239 mmu_notifier_unregister(&kvm->mmu_notifier, current->mm);
1241 out_err_no_mmu_notifier:
1242 hardware_disable_all();
1244 kvm_arch_destroy_vm(kvm);
1245 out_err_no_arch_destroy_vm:
1246 WARN_ON_ONCE(!refcount_dec_and_test(&kvm->users_count));
1247 for (i = 0; i < KVM_NR_BUSES; i++)
1248 kfree(kvm_get_bus(kvm, i));
1249 cleanup_srcu_struct(&kvm->irq_srcu);
1250 out_err_no_irq_srcu:
1251 cleanup_srcu_struct(&kvm->srcu);
1253 kvm_arch_free_vm(kvm);
1254 mmdrop(current->mm);
1258 static void kvm_destroy_devices(struct kvm *kvm)
1260 struct kvm_device *dev, *tmp;
1263 * We do not need to take the kvm->lock here, because nobody else
1264 * has a reference to the struct kvm at this point and therefore
1265 * cannot access the devices list anyhow.
1267 list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
1268 list_del(&dev->vm_node);
1269 dev->ops->destroy(dev);
1273 static void kvm_destroy_vm(struct kvm *kvm)
1276 struct mm_struct *mm = kvm->mm;
1278 kvm_destroy_pm_notifier(kvm);
1279 kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm);
1280 kvm_destroy_vm_debugfs(kvm);
1281 kvm_arch_sync_events(kvm);
1282 mutex_lock(&kvm_lock);
1283 list_del(&kvm->vm_list);
1284 mutex_unlock(&kvm_lock);
1285 kvm_arch_pre_destroy_vm(kvm);
1287 kvm_free_irq_routing(kvm);
1288 for (i = 0; i < KVM_NR_BUSES; i++) {
1289 struct kvm_io_bus *bus = kvm_get_bus(kvm, i);
1292 kvm_io_bus_destroy(bus);
1293 kvm->buses[i] = NULL;
1295 kvm_coalesced_mmio_free(kvm);
1296 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1297 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
1299 * At this point, pending calls to invalidate_range_start()
1300 * have completed but no more MMU notifiers will run, so
1301 * mn_active_invalidate_count may remain unbalanced.
1302 * No threads can be waiting in install_new_memslots as the
1303 * last reference on KVM has been dropped, but freeing
1304 * memslots would deadlock without this manual intervention.
1306 WARN_ON(rcuwait_active(&kvm->mn_memslots_update_rcuwait));
1307 kvm->mn_active_invalidate_count = 0;
1309 kvm_flush_shadow_all(kvm);
1311 kvm_arch_destroy_vm(kvm);
1312 kvm_destroy_devices(kvm);
1313 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
1314 kvm_free_memslots(kvm, &kvm->__memslots[i][0]);
1315 kvm_free_memslots(kvm, &kvm->__memslots[i][1]);
1317 cleanup_srcu_struct(&kvm->irq_srcu);
1318 cleanup_srcu_struct(&kvm->srcu);
1319 kvm_arch_free_vm(kvm);
1320 preempt_notifier_dec();
1321 hardware_disable_all();
1323 module_put(kvm_chardev_ops.owner);
1326 void kvm_get_kvm(struct kvm *kvm)
1328 refcount_inc(&kvm->users_count);
1330 EXPORT_SYMBOL_GPL(kvm_get_kvm);
1333 * Make sure the vm is not during destruction, which is a safe version of
1334 * kvm_get_kvm(). Return true if kvm referenced successfully, false otherwise.
1336 bool kvm_get_kvm_safe(struct kvm *kvm)
1338 return refcount_inc_not_zero(&kvm->users_count);
1340 EXPORT_SYMBOL_GPL(kvm_get_kvm_safe);
1342 void kvm_put_kvm(struct kvm *kvm)
1344 if (refcount_dec_and_test(&kvm->users_count))
1345 kvm_destroy_vm(kvm);
1347 EXPORT_SYMBOL_GPL(kvm_put_kvm);
1350 * Used to put a reference that was taken on behalf of an object associated
1351 * with a user-visible file descriptor, e.g. a vcpu or device, if installation
1352 * of the new file descriptor fails and the reference cannot be transferred to
1353 * its final owner. In such cases, the caller is still actively using @kvm and
1354 * will fail miserably if the refcount unexpectedly hits zero.
1356 void kvm_put_kvm_no_destroy(struct kvm *kvm)
1358 WARN_ON(refcount_dec_and_test(&kvm->users_count));
1360 EXPORT_SYMBOL_GPL(kvm_put_kvm_no_destroy);
1362 static int kvm_vm_release(struct inode *inode, struct file *filp)
1364 struct kvm *kvm = filp->private_data;
1366 kvm_irqfd_release(kvm);
1373 * Allocation size is twice as large as the actual dirty bitmap size.
1374 * See kvm_vm_ioctl_get_dirty_log() why this is needed.
1376 static int kvm_alloc_dirty_bitmap(struct kvm_memory_slot *memslot)
1378 unsigned long dirty_bytes = kvm_dirty_bitmap_bytes(memslot);
1380 memslot->dirty_bitmap = __vcalloc(2, dirty_bytes, GFP_KERNEL_ACCOUNT);
1381 if (!memslot->dirty_bitmap)
1387 static struct kvm_memslots *kvm_get_inactive_memslots(struct kvm *kvm, int as_id)
1389 struct kvm_memslots *active = __kvm_memslots(kvm, as_id);
1390 int node_idx_inactive = active->node_idx ^ 1;
1392 return &kvm->__memslots[as_id][node_idx_inactive];
1396 * Helper to get the address space ID when one of memslot pointers may be NULL.
1397 * This also serves as a sanity that at least one of the pointers is non-NULL,
1398 * and that their address space IDs don't diverge.
1400 static int kvm_memslots_get_as_id(struct kvm_memory_slot *a,
1401 struct kvm_memory_slot *b)
1403 if (WARN_ON_ONCE(!a && !b))
1411 WARN_ON_ONCE(a->as_id != b->as_id);
1415 static void kvm_insert_gfn_node(struct kvm_memslots *slots,
1416 struct kvm_memory_slot *slot)
1418 struct rb_root *gfn_tree = &slots->gfn_tree;
1419 struct rb_node **node, *parent;
1420 int idx = slots->node_idx;
1423 for (node = &gfn_tree->rb_node; *node; ) {
1424 struct kvm_memory_slot *tmp;
1426 tmp = container_of(*node, struct kvm_memory_slot, gfn_node[idx]);
1428 if (slot->base_gfn < tmp->base_gfn)
1429 node = &(*node)->rb_left;
1430 else if (slot->base_gfn > tmp->base_gfn)
1431 node = &(*node)->rb_right;
1436 rb_link_node(&slot->gfn_node[idx], parent, node);
1437 rb_insert_color(&slot->gfn_node[idx], gfn_tree);
1440 static void kvm_erase_gfn_node(struct kvm_memslots *slots,
1441 struct kvm_memory_slot *slot)
1443 rb_erase(&slot->gfn_node[slots->node_idx], &slots->gfn_tree);
1446 static void kvm_replace_gfn_node(struct kvm_memslots *slots,
1447 struct kvm_memory_slot *old,
1448 struct kvm_memory_slot *new)
1450 int idx = slots->node_idx;
1452 WARN_ON_ONCE(old->base_gfn != new->base_gfn);
1454 rb_replace_node(&old->gfn_node[idx], &new->gfn_node[idx],
1459 * Replace @old with @new in the inactive memslots.
1461 * With NULL @old this simply adds @new.
1462 * With NULL @new this simply removes @old.
1464 * If @new is non-NULL its hva_node[slots_idx] range has to be set
1467 static void kvm_replace_memslot(struct kvm *kvm,
1468 struct kvm_memory_slot *old,
1469 struct kvm_memory_slot *new)
1471 int as_id = kvm_memslots_get_as_id(old, new);
1472 struct kvm_memslots *slots = kvm_get_inactive_memslots(kvm, as_id);
1473 int idx = slots->node_idx;
1476 hash_del(&old->id_node[idx]);
1477 interval_tree_remove(&old->hva_node[idx], &slots->hva_tree);
1479 if ((long)old == atomic_long_read(&slots->last_used_slot))
1480 atomic_long_set(&slots->last_used_slot, (long)new);
1483 kvm_erase_gfn_node(slots, old);
1489 * Initialize @new's hva range. Do this even when replacing an @old
1490 * slot, kvm_copy_memslot() deliberately does not touch node data.
1492 new->hva_node[idx].start = new->userspace_addr;
1493 new->hva_node[idx].last = new->userspace_addr +
1494 (new->npages << PAGE_SHIFT) - 1;
1497 * (Re)Add the new memslot. There is no O(1) interval_tree_replace(),
1498 * hva_node needs to be swapped with remove+insert even though hva can't
1499 * change when replacing an existing slot.
1501 hash_add(slots->id_hash, &new->id_node[idx], new->id);
1502 interval_tree_insert(&new->hva_node[idx], &slots->hva_tree);
1505 * If the memslot gfn is unchanged, rb_replace_node() can be used to
1506 * switch the node in the gfn tree instead of removing the old and
1507 * inserting the new as two separate operations. Replacement is a
1508 * single O(1) operation versus two O(log(n)) operations for
1511 if (old && old->base_gfn == new->base_gfn) {
1512 kvm_replace_gfn_node(slots, old, new);
1515 kvm_erase_gfn_node(slots, old);
1516 kvm_insert_gfn_node(slots, new);
1520 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
1522 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
1524 #ifdef __KVM_HAVE_READONLY_MEM
1525 valid_flags |= KVM_MEM_READONLY;
1528 if (mem->flags & ~valid_flags)
1534 static void kvm_swap_active_memslots(struct kvm *kvm, int as_id)
1536 struct kvm_memslots *slots = kvm_get_inactive_memslots(kvm, as_id);
1538 /* Grab the generation from the activate memslots. */
1539 u64 gen = __kvm_memslots(kvm, as_id)->generation;
1541 WARN_ON(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS);
1542 slots->generation = gen | KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
1545 * Do not store the new memslots while there are invalidations in
1546 * progress, otherwise the locking in invalidate_range_start and
1547 * invalidate_range_end will be unbalanced.
1549 spin_lock(&kvm->mn_invalidate_lock);
1550 prepare_to_rcuwait(&kvm->mn_memslots_update_rcuwait);
1551 while (kvm->mn_active_invalidate_count) {
1552 set_current_state(TASK_UNINTERRUPTIBLE);
1553 spin_unlock(&kvm->mn_invalidate_lock);
1555 spin_lock(&kvm->mn_invalidate_lock);
1557 finish_rcuwait(&kvm->mn_memslots_update_rcuwait);
1558 rcu_assign_pointer(kvm->memslots[as_id], slots);
1559 spin_unlock(&kvm->mn_invalidate_lock);
1562 * Acquired in kvm_set_memslot. Must be released before synchronize
1563 * SRCU below in order to avoid deadlock with another thread
1564 * acquiring the slots_arch_lock in an srcu critical section.
1566 mutex_unlock(&kvm->slots_arch_lock);
1568 synchronize_srcu_expedited(&kvm->srcu);
1571 * Increment the new memslot generation a second time, dropping the
1572 * update in-progress flag and incrementing the generation based on
1573 * the number of address spaces. This provides a unique and easily
1574 * identifiable generation number while the memslots are in flux.
1576 gen = slots->generation & ~KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
1579 * Generations must be unique even across address spaces. We do not need
1580 * a global counter for that, instead the generation space is evenly split
1581 * across address spaces. For example, with two address spaces, address
1582 * space 0 will use generations 0, 2, 4, ... while address space 1 will
1583 * use generations 1, 3, 5, ...
1585 gen += KVM_ADDRESS_SPACE_NUM;
1587 kvm_arch_memslots_updated(kvm, gen);
1589 slots->generation = gen;
1592 static int kvm_prepare_memory_region(struct kvm *kvm,
1593 const struct kvm_memory_slot *old,
1594 struct kvm_memory_slot *new,
1595 enum kvm_mr_change change)
1600 * If dirty logging is disabled, nullify the bitmap; the old bitmap
1601 * will be freed on "commit". If logging is enabled in both old and
1602 * new, reuse the existing bitmap. If logging is enabled only in the
1603 * new and KVM isn't using a ring buffer, allocate and initialize a
1606 if (change != KVM_MR_DELETE) {
1607 if (!(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
1608 new->dirty_bitmap = NULL;
1609 else if (old && old->dirty_bitmap)
1610 new->dirty_bitmap = old->dirty_bitmap;
1611 else if (!kvm->dirty_ring_size) {
1612 r = kvm_alloc_dirty_bitmap(new);
1616 if (kvm_dirty_log_manual_protect_and_init_set(kvm))
1617 bitmap_set(new->dirty_bitmap, 0, new->npages);
1621 r = kvm_arch_prepare_memory_region(kvm, old, new, change);
1623 /* Free the bitmap on failure if it was allocated above. */
1624 if (r && new && new->dirty_bitmap && (!old || !old->dirty_bitmap))
1625 kvm_destroy_dirty_bitmap(new);
1630 static void kvm_commit_memory_region(struct kvm *kvm,
1631 struct kvm_memory_slot *old,
1632 const struct kvm_memory_slot *new,
1633 enum kvm_mr_change change)
1636 * Update the total number of memslot pages before calling the arch
1637 * hook so that architectures can consume the result directly.
1639 if (change == KVM_MR_DELETE)
1640 kvm->nr_memslot_pages -= old->npages;
1641 else if (change == KVM_MR_CREATE)
1642 kvm->nr_memslot_pages += new->npages;
1644 kvm_arch_commit_memory_region(kvm, old, new, change);
1648 /* Nothing more to do. */
1651 /* Free the old memslot and all its metadata. */
1652 kvm_free_memslot(kvm, old);
1655 case KVM_MR_FLAGS_ONLY:
1657 * Free the dirty bitmap as needed; the below check encompasses
1658 * both the flags and whether a ring buffer is being used)
1660 if (old->dirty_bitmap && !new->dirty_bitmap)
1661 kvm_destroy_dirty_bitmap(old);
1664 * The final quirk. Free the detached, old slot, but only its
1665 * memory, not any metadata. Metadata, including arch specific
1666 * data, may be reused by @new.
1676 * Activate @new, which must be installed in the inactive slots by the caller,
1677 * by swapping the active slots and then propagating @new to @old once @old is
1678 * unreachable and can be safely modified.
1680 * With NULL @old this simply adds @new to @active (while swapping the sets).
1681 * With NULL @new this simply removes @old from @active and frees it
1682 * (while also swapping the sets).
1684 static void kvm_activate_memslot(struct kvm *kvm,
1685 struct kvm_memory_slot *old,
1686 struct kvm_memory_slot *new)
1688 int as_id = kvm_memslots_get_as_id(old, new);
1690 kvm_swap_active_memslots(kvm, as_id);
1692 /* Propagate the new memslot to the now inactive memslots. */
1693 kvm_replace_memslot(kvm, old, new);
1696 static void kvm_copy_memslot(struct kvm_memory_slot *dest,
1697 const struct kvm_memory_slot *src)
1699 dest->base_gfn = src->base_gfn;
1700 dest->npages = src->npages;
1701 dest->dirty_bitmap = src->dirty_bitmap;
1702 dest->arch = src->arch;
1703 dest->userspace_addr = src->userspace_addr;
1704 dest->flags = src->flags;
1706 dest->as_id = src->as_id;
1709 static void kvm_invalidate_memslot(struct kvm *kvm,
1710 struct kvm_memory_slot *old,
1711 struct kvm_memory_slot *invalid_slot)
1714 * Mark the current slot INVALID. As with all memslot modifications,
1715 * this must be done on an unreachable slot to avoid modifying the
1716 * current slot in the active tree.
1718 kvm_copy_memslot(invalid_slot, old);
1719 invalid_slot->flags |= KVM_MEMSLOT_INVALID;
1720 kvm_replace_memslot(kvm, old, invalid_slot);
1723 * Activate the slot that is now marked INVALID, but don't propagate
1724 * the slot to the now inactive slots. The slot is either going to be
1725 * deleted or recreated as a new slot.
1727 kvm_swap_active_memslots(kvm, old->as_id);
1730 * From this point no new shadow pages pointing to a deleted, or moved,
1731 * memslot will be created. Validation of sp->gfn happens in:
1732 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1733 * - kvm_is_visible_gfn (mmu_check_root)
1735 kvm_arch_flush_shadow_memslot(kvm, old);
1736 kvm_arch_guest_memory_reclaimed(kvm);
1738 /* Was released by kvm_swap_active_memslots, reacquire. */
1739 mutex_lock(&kvm->slots_arch_lock);
1742 * Copy the arch-specific field of the newly-installed slot back to the
1743 * old slot as the arch data could have changed between releasing
1744 * slots_arch_lock in install_new_memslots() and re-acquiring the lock
1745 * above. Writers are required to retrieve memslots *after* acquiring
1746 * slots_arch_lock, thus the active slot's data is guaranteed to be fresh.
1748 old->arch = invalid_slot->arch;
1751 static void kvm_create_memslot(struct kvm *kvm,
1752 struct kvm_memory_slot *new)
1754 /* Add the new memslot to the inactive set and activate. */
1755 kvm_replace_memslot(kvm, NULL, new);
1756 kvm_activate_memslot(kvm, NULL, new);
1759 static void kvm_delete_memslot(struct kvm *kvm,
1760 struct kvm_memory_slot *old,
1761 struct kvm_memory_slot *invalid_slot)
1764 * Remove the old memslot (in the inactive memslots) by passing NULL as
1765 * the "new" slot, and for the invalid version in the active slots.
1767 kvm_replace_memslot(kvm, old, NULL);
1768 kvm_activate_memslot(kvm, invalid_slot, NULL);
1771 static void kvm_move_memslot(struct kvm *kvm,
1772 struct kvm_memory_slot *old,
1773 struct kvm_memory_slot *new,
1774 struct kvm_memory_slot *invalid_slot)
1777 * Replace the old memslot in the inactive slots, and then swap slots
1778 * and replace the current INVALID with the new as well.
1780 kvm_replace_memslot(kvm, old, new);
1781 kvm_activate_memslot(kvm, invalid_slot, new);
1784 static void kvm_update_flags_memslot(struct kvm *kvm,
1785 struct kvm_memory_slot *old,
1786 struct kvm_memory_slot *new)
1789 * Similar to the MOVE case, but the slot doesn't need to be zapped as
1790 * an intermediate step. Instead, the old memslot is simply replaced
1791 * with a new, updated copy in both memslot sets.
1793 kvm_replace_memslot(kvm, old, new);
1794 kvm_activate_memslot(kvm, old, new);
1797 static int kvm_set_memslot(struct kvm *kvm,
1798 struct kvm_memory_slot *old,
1799 struct kvm_memory_slot *new,
1800 enum kvm_mr_change change)
1802 struct kvm_memory_slot *invalid_slot;
1806 * Released in kvm_swap_active_memslots.
1808 * Must be held from before the current memslots are copied until
1809 * after the new memslots are installed with rcu_assign_pointer,
1810 * then released before the synchronize srcu in kvm_swap_active_memslots.
1812 * When modifying memslots outside of the slots_lock, must be held
1813 * before reading the pointer to the current memslots until after all
1814 * changes to those memslots are complete.
1816 * These rules ensure that installing new memslots does not lose
1817 * changes made to the previous memslots.
1819 mutex_lock(&kvm->slots_arch_lock);
1822 * Invalidate the old slot if it's being deleted or moved. This is
1823 * done prior to actually deleting/moving the memslot to allow vCPUs to
1824 * continue running by ensuring there are no mappings or shadow pages
1825 * for the memslot when it is deleted/moved. Without pre-invalidation
1826 * (and without a lock), a window would exist between effecting the
1827 * delete/move and committing the changes in arch code where KVM or a
1828 * guest could access a non-existent memslot.
1830 * Modifications are done on a temporary, unreachable slot. The old
1831 * slot needs to be preserved in case a later step fails and the
1832 * invalidation needs to be reverted.
1834 if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) {
1835 invalid_slot = kzalloc(sizeof(*invalid_slot), GFP_KERNEL_ACCOUNT);
1836 if (!invalid_slot) {
1837 mutex_unlock(&kvm->slots_arch_lock);
1840 kvm_invalidate_memslot(kvm, old, invalid_slot);
1843 r = kvm_prepare_memory_region(kvm, old, new, change);
1846 * For DELETE/MOVE, revert the above INVALID change. No
1847 * modifications required since the original slot was preserved
1848 * in the inactive slots. Changing the active memslots also
1849 * release slots_arch_lock.
1851 if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) {
1852 kvm_activate_memslot(kvm, invalid_slot, old);
1853 kfree(invalid_slot);
1855 mutex_unlock(&kvm->slots_arch_lock);
1861 * For DELETE and MOVE, the working slot is now active as the INVALID
1862 * version of the old slot. MOVE is particularly special as it reuses
1863 * the old slot and returns a copy of the old slot (in working_slot).
1864 * For CREATE, there is no old slot. For DELETE and FLAGS_ONLY, the
1865 * old slot is detached but otherwise preserved.
1867 if (change == KVM_MR_CREATE)
1868 kvm_create_memslot(kvm, new);
1869 else if (change == KVM_MR_DELETE)
1870 kvm_delete_memslot(kvm, old, invalid_slot);
1871 else if (change == KVM_MR_MOVE)
1872 kvm_move_memslot(kvm, old, new, invalid_slot);
1873 else if (change == KVM_MR_FLAGS_ONLY)
1874 kvm_update_flags_memslot(kvm, old, new);
1878 /* Free the temporary INVALID slot used for DELETE and MOVE. */
1879 if (change == KVM_MR_DELETE || change == KVM_MR_MOVE)
1880 kfree(invalid_slot);
1883 * No need to refresh new->arch, changes after dropping slots_arch_lock
1884 * will directly hit the final, active memslot. Architectures are
1885 * responsible for knowing that new->arch may be stale.
1887 kvm_commit_memory_region(kvm, old, new, change);
1892 static bool kvm_check_memslot_overlap(struct kvm_memslots *slots, int id,
1893 gfn_t start, gfn_t end)
1895 struct kvm_memslot_iter iter;
1897 kvm_for_each_memslot_in_gfn_range(&iter, slots, start, end) {
1898 if (iter.slot->id != id)
1906 * Allocate some memory and give it an address in the guest physical address
1909 * Discontiguous memory is allowed, mostly for framebuffers.
1911 * Must be called holding kvm->slots_lock for write.
1913 int __kvm_set_memory_region(struct kvm *kvm,
1914 const struct kvm_userspace_memory_region *mem)
1916 struct kvm_memory_slot *old, *new;
1917 struct kvm_memslots *slots;
1918 enum kvm_mr_change change;
1919 unsigned long npages;
1924 r = check_memory_region_flags(mem);
1928 as_id = mem->slot >> 16;
1929 id = (u16)mem->slot;
1931 /* General sanity checks */
1932 if ((mem->memory_size & (PAGE_SIZE - 1)) ||
1933 (mem->memory_size != (unsigned long)mem->memory_size))
1935 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
1937 /* We can read the guest memory with __xxx_user() later on. */
1938 if ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
1939 (mem->userspace_addr != untagged_addr(mem->userspace_addr)) ||
1940 !access_ok((void __user *)(unsigned long)mem->userspace_addr,
1943 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
1945 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
1947 if ((mem->memory_size >> PAGE_SHIFT) > KVM_MEM_MAX_NR_PAGES)
1950 slots = __kvm_memslots(kvm, as_id);
1953 * Note, the old memslot (and the pointer itself!) may be invalidated
1954 * and/or destroyed by kvm_set_memslot().
1956 old = id_to_memslot(slots, id);
1958 if (!mem->memory_size) {
1959 if (!old || !old->npages)
1962 if (WARN_ON_ONCE(kvm->nr_memslot_pages < old->npages))
1965 return kvm_set_memslot(kvm, old, NULL, KVM_MR_DELETE);
1968 base_gfn = (mem->guest_phys_addr >> PAGE_SHIFT);
1969 npages = (mem->memory_size >> PAGE_SHIFT);
1971 if (!old || !old->npages) {
1972 change = KVM_MR_CREATE;
1975 * To simplify KVM internals, the total number of pages across
1976 * all memslots must fit in an unsigned long.
1978 if ((kvm->nr_memslot_pages + npages) < kvm->nr_memslot_pages)
1980 } else { /* Modify an existing slot. */
1981 if ((mem->userspace_addr != old->userspace_addr) ||
1982 (npages != old->npages) ||
1983 ((mem->flags ^ old->flags) & KVM_MEM_READONLY))
1986 if (base_gfn != old->base_gfn)
1987 change = KVM_MR_MOVE;
1988 else if (mem->flags != old->flags)
1989 change = KVM_MR_FLAGS_ONLY;
1990 else /* Nothing to change. */
1994 if ((change == KVM_MR_CREATE || change == KVM_MR_MOVE) &&
1995 kvm_check_memslot_overlap(slots, id, base_gfn, base_gfn + npages))
1998 /* Allocate a slot that will persist in the memslot. */
1999 new = kzalloc(sizeof(*new), GFP_KERNEL_ACCOUNT);
2005 new->base_gfn = base_gfn;
2006 new->npages = npages;
2007 new->flags = mem->flags;
2008 new->userspace_addr = mem->userspace_addr;
2010 r = kvm_set_memslot(kvm, old, new, change);
2015 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
2017 int kvm_set_memory_region(struct kvm *kvm,
2018 const struct kvm_userspace_memory_region *mem)
2022 mutex_lock(&kvm->slots_lock);
2023 r = __kvm_set_memory_region(kvm, mem);
2024 mutex_unlock(&kvm->slots_lock);
2027 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
2029 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
2030 struct kvm_userspace_memory_region *mem)
2032 if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
2035 return kvm_set_memory_region(kvm, mem);
2038 #ifndef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
2040 * kvm_get_dirty_log - get a snapshot of dirty pages
2041 * @kvm: pointer to kvm instance
2042 * @log: slot id and address to which we copy the log
2043 * @is_dirty: set to '1' if any dirty pages were found
2044 * @memslot: set to the associated memslot, always valid on success
2046 int kvm_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log,
2047 int *is_dirty, struct kvm_memory_slot **memslot)
2049 struct kvm_memslots *slots;
2052 unsigned long any = 0;
2054 /* Dirty ring tracking is exclusive to dirty log tracking */
2055 if (kvm->dirty_ring_size)
2061 as_id = log->slot >> 16;
2062 id = (u16)log->slot;
2063 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
2066 slots = __kvm_memslots(kvm, as_id);
2067 *memslot = id_to_memslot(slots, id);
2068 if (!(*memslot) || !(*memslot)->dirty_bitmap)
2071 kvm_arch_sync_dirty_log(kvm, *memslot);
2073 n = kvm_dirty_bitmap_bytes(*memslot);
2075 for (i = 0; !any && i < n/sizeof(long); ++i)
2076 any = (*memslot)->dirty_bitmap[i];
2078 if (copy_to_user(log->dirty_bitmap, (*memslot)->dirty_bitmap, n))
2085 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
2087 #else /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
2089 * kvm_get_dirty_log_protect - get a snapshot of dirty pages
2090 * and reenable dirty page tracking for the corresponding pages.
2091 * @kvm: pointer to kvm instance
2092 * @log: slot id and address to which we copy the log
2094 * We need to keep it in mind that VCPU threads can write to the bitmap
2095 * concurrently. So, to avoid losing track of dirty pages we keep the
2098 * 1. Take a snapshot of the bit and clear it if needed.
2099 * 2. Write protect the corresponding page.
2100 * 3. Copy the snapshot to the userspace.
2101 * 4. Upon return caller flushes TLB's if needed.
2103 * Between 2 and 4, the guest may write to the page using the remaining TLB
2104 * entry. This is not a problem because the page is reported dirty using
2105 * the snapshot taken before and step 4 ensures that writes done after
2106 * exiting to userspace will be logged for the next call.
2109 static int kvm_get_dirty_log_protect(struct kvm *kvm, struct kvm_dirty_log *log)
2111 struct kvm_memslots *slots;
2112 struct kvm_memory_slot *memslot;
2115 unsigned long *dirty_bitmap;
2116 unsigned long *dirty_bitmap_buffer;
2119 /* Dirty ring tracking is exclusive to dirty log tracking */
2120 if (kvm->dirty_ring_size)
2123 as_id = log->slot >> 16;
2124 id = (u16)log->slot;
2125 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
2128 slots = __kvm_memslots(kvm, as_id);
2129 memslot = id_to_memslot(slots, id);
2130 if (!memslot || !memslot->dirty_bitmap)
2133 dirty_bitmap = memslot->dirty_bitmap;
2135 kvm_arch_sync_dirty_log(kvm, memslot);
2137 n = kvm_dirty_bitmap_bytes(memslot);
2139 if (kvm->manual_dirty_log_protect) {
2141 * Unlike kvm_get_dirty_log, we always return false in *flush,
2142 * because no flush is needed until KVM_CLEAR_DIRTY_LOG. There
2143 * is some code duplication between this function and
2144 * kvm_get_dirty_log, but hopefully all architecture
2145 * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log
2146 * can be eliminated.
2148 dirty_bitmap_buffer = dirty_bitmap;
2150 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
2151 memset(dirty_bitmap_buffer, 0, n);
2154 for (i = 0; i < n / sizeof(long); i++) {
2158 if (!dirty_bitmap[i])
2162 mask = xchg(&dirty_bitmap[i], 0);
2163 dirty_bitmap_buffer[i] = mask;
2165 offset = i * BITS_PER_LONG;
2166 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
2169 KVM_MMU_UNLOCK(kvm);
2173 kvm_arch_flush_remote_tlbs_memslot(kvm, memslot);
2175 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
2182 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
2183 * @kvm: kvm instance
2184 * @log: slot id and address to which we copy the log
2186 * Steps 1-4 below provide general overview of dirty page logging. See
2187 * kvm_get_dirty_log_protect() function description for additional details.
2189 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
2190 * always flush the TLB (step 4) even if previous step failed and the dirty
2191 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
2192 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
2193 * writes will be marked dirty for next log read.
2195 * 1. Take a snapshot of the bit and clear it if needed.
2196 * 2. Write protect the corresponding page.
2197 * 3. Copy the snapshot to the userspace.
2198 * 4. Flush TLB's if needed.
2200 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
2201 struct kvm_dirty_log *log)
2205 mutex_lock(&kvm->slots_lock);
2207 r = kvm_get_dirty_log_protect(kvm, log);
2209 mutex_unlock(&kvm->slots_lock);
2214 * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap
2215 * and reenable dirty page tracking for the corresponding pages.
2216 * @kvm: pointer to kvm instance
2217 * @log: slot id and address from which to fetch the bitmap of dirty pages
2219 static int kvm_clear_dirty_log_protect(struct kvm *kvm,
2220 struct kvm_clear_dirty_log *log)
2222 struct kvm_memslots *slots;
2223 struct kvm_memory_slot *memslot;
2227 unsigned long *dirty_bitmap;
2228 unsigned long *dirty_bitmap_buffer;
2231 /* Dirty ring tracking is exclusive to dirty log tracking */
2232 if (kvm->dirty_ring_size)
2235 as_id = log->slot >> 16;
2236 id = (u16)log->slot;
2237 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
2240 if (log->first_page & 63)
2243 slots = __kvm_memslots(kvm, as_id);
2244 memslot = id_to_memslot(slots, id);
2245 if (!memslot || !memslot->dirty_bitmap)
2248 dirty_bitmap = memslot->dirty_bitmap;
2250 n = ALIGN(log->num_pages, BITS_PER_LONG) / 8;
2252 if (log->first_page > memslot->npages ||
2253 log->num_pages > memslot->npages - log->first_page ||
2254 (log->num_pages < memslot->npages - log->first_page && (log->num_pages & 63)))
2257 kvm_arch_sync_dirty_log(kvm, memslot);
2260 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
2261 if (copy_from_user(dirty_bitmap_buffer, log->dirty_bitmap, n))
2265 for (offset = log->first_page, i = offset / BITS_PER_LONG,
2266 n = DIV_ROUND_UP(log->num_pages, BITS_PER_LONG); n--;
2267 i++, offset += BITS_PER_LONG) {
2268 unsigned long mask = *dirty_bitmap_buffer++;
2269 atomic_long_t *p = (atomic_long_t *) &dirty_bitmap[i];
2273 mask &= atomic_long_fetch_andnot(mask, p);
2276 * mask contains the bits that really have been cleared. This
2277 * never includes any bits beyond the length of the memslot (if
2278 * the length is not aligned to 64 pages), therefore it is not
2279 * a problem if userspace sets them in log->dirty_bitmap.
2283 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
2287 KVM_MMU_UNLOCK(kvm);
2290 kvm_arch_flush_remote_tlbs_memslot(kvm, memslot);
2295 static int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm,
2296 struct kvm_clear_dirty_log *log)
2300 mutex_lock(&kvm->slots_lock);
2302 r = kvm_clear_dirty_log_protect(kvm, log);
2304 mutex_unlock(&kvm->slots_lock);
2307 #endif /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
2309 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
2311 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
2313 EXPORT_SYMBOL_GPL(gfn_to_memslot);
2315 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
2317 struct kvm_memslots *slots = kvm_vcpu_memslots(vcpu);
2318 u64 gen = slots->generation;
2319 struct kvm_memory_slot *slot;
2322 * This also protects against using a memslot from a different address space,
2323 * since different address spaces have different generation numbers.
2325 if (unlikely(gen != vcpu->last_used_slot_gen)) {
2326 vcpu->last_used_slot = NULL;
2327 vcpu->last_used_slot_gen = gen;
2330 slot = try_get_memslot(vcpu->last_used_slot, gfn);
2335 * Fall back to searching all memslots. We purposely use
2336 * search_memslots() instead of __gfn_to_memslot() to avoid
2337 * thrashing the VM-wide last_used_slot in kvm_memslots.
2339 slot = search_memslots(slots, gfn, false);
2341 vcpu->last_used_slot = slot;
2348 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
2350 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
2352 return kvm_is_visible_memslot(memslot);
2354 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
2356 bool kvm_vcpu_is_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
2358 struct kvm_memory_slot *memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2360 return kvm_is_visible_memslot(memslot);
2362 EXPORT_SYMBOL_GPL(kvm_vcpu_is_visible_gfn);
2364 unsigned long kvm_host_page_size(struct kvm_vcpu *vcpu, gfn_t gfn)
2366 struct vm_area_struct *vma;
2367 unsigned long addr, size;
2371 addr = kvm_vcpu_gfn_to_hva_prot(vcpu, gfn, NULL);
2372 if (kvm_is_error_hva(addr))
2375 mmap_read_lock(current->mm);
2376 vma = find_vma(current->mm, addr);
2380 size = vma_kernel_pagesize(vma);
2383 mmap_read_unlock(current->mm);
2388 static bool memslot_is_readonly(const struct kvm_memory_slot *slot)
2390 return slot->flags & KVM_MEM_READONLY;
2393 static unsigned long __gfn_to_hva_many(const struct kvm_memory_slot *slot, gfn_t gfn,
2394 gfn_t *nr_pages, bool write)
2396 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
2397 return KVM_HVA_ERR_BAD;
2399 if (memslot_is_readonly(slot) && write)
2400 return KVM_HVA_ERR_RO_BAD;
2403 *nr_pages = slot->npages - (gfn - slot->base_gfn);
2405 return __gfn_to_hva_memslot(slot, gfn);
2408 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
2411 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
2414 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
2417 return gfn_to_hva_many(slot, gfn, NULL);
2419 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
2421 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
2423 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
2425 EXPORT_SYMBOL_GPL(gfn_to_hva);
2427 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
2429 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
2431 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
2434 * Return the hva of a @gfn and the R/W attribute if possible.
2436 * @slot: the kvm_memory_slot which contains @gfn
2437 * @gfn: the gfn to be translated
2438 * @writable: used to return the read/write attribute of the @slot if the hva
2439 * is valid and @writable is not NULL
2441 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
2442 gfn_t gfn, bool *writable)
2444 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
2446 if (!kvm_is_error_hva(hva) && writable)
2447 *writable = !memslot_is_readonly(slot);
2452 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
2454 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2456 return gfn_to_hva_memslot_prot(slot, gfn, writable);
2459 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
2461 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2463 return gfn_to_hva_memslot_prot(slot, gfn, writable);
2466 static inline int check_user_page_hwpoison(unsigned long addr)
2468 int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
2470 rc = get_user_pages(addr, 1, flags, NULL, NULL);
2471 return rc == -EHWPOISON;
2475 * The fast path to get the writable pfn which will be stored in @pfn,
2476 * true indicates success, otherwise false is returned. It's also the
2477 * only part that runs if we can in atomic context.
2479 static bool hva_to_pfn_fast(unsigned long addr, bool write_fault,
2480 bool *writable, kvm_pfn_t *pfn)
2482 struct page *page[1];
2485 * Fast pin a writable pfn only if it is a write fault request
2486 * or the caller allows to map a writable pfn for a read fault
2489 if (!(write_fault || writable))
2492 if (get_user_page_fast_only(addr, FOLL_WRITE, page)) {
2493 *pfn = page_to_pfn(page[0]);
2504 * The slow path to get the pfn of the specified host virtual address,
2505 * 1 indicates success, -errno is returned if error is detected.
2507 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
2508 bool *writable, kvm_pfn_t *pfn)
2510 unsigned int flags = FOLL_HWPOISON;
2517 *writable = write_fault;
2520 flags |= FOLL_WRITE;
2522 flags |= FOLL_NOWAIT;
2524 npages = get_user_pages_unlocked(addr, 1, &page, flags);
2528 /* map read fault as writable if possible */
2529 if (unlikely(!write_fault) && writable) {
2532 if (get_user_page_fast_only(addr, FOLL_WRITE, &wpage)) {
2538 *pfn = page_to_pfn(page);
2542 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
2544 if (unlikely(!(vma->vm_flags & VM_READ)))
2547 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
2553 static int kvm_try_get_pfn(kvm_pfn_t pfn)
2555 struct page *page = kvm_pfn_to_refcounted_page(pfn);
2560 return get_page_unless_zero(page);
2563 static int hva_to_pfn_remapped(struct vm_area_struct *vma,
2564 unsigned long addr, bool write_fault,
2565 bool *writable, kvm_pfn_t *p_pfn)
2572 r = follow_pte(vma->vm_mm, addr, &ptep, &ptl);
2575 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
2576 * not call the fault handler, so do it here.
2578 bool unlocked = false;
2579 r = fixup_user_fault(current->mm, addr,
2580 (write_fault ? FAULT_FLAG_WRITE : 0),
2587 r = follow_pte(vma->vm_mm, addr, &ptep, &ptl);
2592 if (write_fault && !pte_write(*ptep)) {
2593 pfn = KVM_PFN_ERR_RO_FAULT;
2598 *writable = pte_write(*ptep);
2599 pfn = pte_pfn(*ptep);
2602 * Get a reference here because callers of *hva_to_pfn* and
2603 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
2604 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
2605 * set, but the kvm_try_get_pfn/kvm_release_pfn_clean pair will
2606 * simply do nothing for reserved pfns.
2608 * Whoever called remap_pfn_range is also going to call e.g.
2609 * unmap_mapping_range before the underlying pages are freed,
2610 * causing a call to our MMU notifier.
2612 * Certain IO or PFNMAP mappings can be backed with valid
2613 * struct pages, but be allocated without refcounting e.g.,
2614 * tail pages of non-compound higher order allocations, which
2615 * would then underflow the refcount when the caller does the
2616 * required put_page. Don't allow those pages here.
2618 if (!kvm_try_get_pfn(pfn))
2622 pte_unmap_unlock(ptep, ptl);
2629 * Pin guest page in memory and return its pfn.
2630 * @addr: host virtual address which maps memory to the guest
2631 * @atomic: whether this function can sleep
2632 * @async: whether this function need to wait IO complete if the
2633 * host page is not in the memory
2634 * @write_fault: whether we should get a writable host page
2635 * @writable: whether it allows to map a writable host page for !@write_fault
2637 * The function will map a writable host page for these two cases:
2638 * 1): @write_fault = true
2639 * 2): @write_fault = false && @writable, @writable will tell the caller
2640 * whether the mapping is writable.
2642 kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
2643 bool write_fault, bool *writable)
2645 struct vm_area_struct *vma;
2649 /* we can do it either atomically or asynchronously, not both */
2650 BUG_ON(atomic && async);
2652 if (hva_to_pfn_fast(addr, write_fault, writable, &pfn))
2656 return KVM_PFN_ERR_FAULT;
2658 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
2662 mmap_read_lock(current->mm);
2663 if (npages == -EHWPOISON ||
2664 (!async && check_user_page_hwpoison(addr))) {
2665 pfn = KVM_PFN_ERR_HWPOISON;
2670 vma = vma_lookup(current->mm, addr);
2673 pfn = KVM_PFN_ERR_FAULT;
2674 else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
2675 r = hva_to_pfn_remapped(vma, addr, write_fault, writable, &pfn);
2679 pfn = KVM_PFN_ERR_FAULT;
2681 if (async && vma_is_valid(vma, write_fault))
2683 pfn = KVM_PFN_ERR_FAULT;
2686 mmap_read_unlock(current->mm);
2690 kvm_pfn_t __gfn_to_pfn_memslot(const struct kvm_memory_slot *slot, gfn_t gfn,
2691 bool atomic, bool *async, bool write_fault,
2692 bool *writable, hva_t *hva)
2694 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
2699 if (addr == KVM_HVA_ERR_RO_BAD) {
2702 return KVM_PFN_ERR_RO_FAULT;
2705 if (kvm_is_error_hva(addr)) {
2708 return KVM_PFN_NOSLOT;
2711 /* Do not map writable pfn in the readonly memslot. */
2712 if (writable && memslot_is_readonly(slot)) {
2717 return hva_to_pfn(addr, atomic, async, write_fault,
2720 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
2722 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
2725 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
2726 write_fault, writable, NULL);
2728 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
2730 kvm_pfn_t gfn_to_pfn_memslot(const struct kvm_memory_slot *slot, gfn_t gfn)
2732 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL, NULL);
2734 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
2736 kvm_pfn_t gfn_to_pfn_memslot_atomic(const struct kvm_memory_slot *slot, gfn_t gfn)
2738 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL, NULL);
2740 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
2742 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
2744 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
2746 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
2748 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
2750 return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
2752 EXPORT_SYMBOL_GPL(gfn_to_pfn);
2754 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
2756 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
2758 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
2760 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
2761 struct page **pages, int nr_pages)
2766 addr = gfn_to_hva_many(slot, gfn, &entry);
2767 if (kvm_is_error_hva(addr))
2770 if (entry < nr_pages)
2773 return get_user_pages_fast_only(addr, nr_pages, FOLL_WRITE, pages);
2775 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
2778 * Do not use this helper unless you are absolutely certain the gfn _must_ be
2779 * backed by 'struct page'. A valid example is if the backing memslot is
2780 * controlled by KVM. Note, if the returned page is valid, it's refcount has
2781 * been elevated by gfn_to_pfn().
2783 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
2788 pfn = gfn_to_pfn(kvm, gfn);
2790 if (is_error_noslot_pfn(pfn))
2791 return KVM_ERR_PTR_BAD_PAGE;
2793 page = kvm_pfn_to_refcounted_page(pfn);
2795 return KVM_ERR_PTR_BAD_PAGE;
2799 EXPORT_SYMBOL_GPL(gfn_to_page);
2801 void kvm_release_pfn(kvm_pfn_t pfn, bool dirty)
2804 kvm_release_pfn_dirty(pfn);
2806 kvm_release_pfn_clean(pfn);
2809 int kvm_vcpu_map(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map)
2813 struct page *page = KVM_UNMAPPED_PAGE;
2818 pfn = gfn_to_pfn(vcpu->kvm, gfn);
2819 if (is_error_noslot_pfn(pfn))
2822 if (pfn_valid(pfn)) {
2823 page = pfn_to_page(pfn);
2825 #ifdef CONFIG_HAS_IOMEM
2827 hva = memremap(pfn_to_hpa(pfn), PAGE_SIZE, MEMREMAP_WB);
2841 EXPORT_SYMBOL_GPL(kvm_vcpu_map);
2843 void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map, bool dirty)
2851 if (map->page != KVM_UNMAPPED_PAGE)
2853 #ifdef CONFIG_HAS_IOMEM
2859 kvm_vcpu_mark_page_dirty(vcpu, map->gfn);
2861 kvm_release_pfn(map->pfn, dirty);
2866 EXPORT_SYMBOL_GPL(kvm_vcpu_unmap);
2868 static bool kvm_is_ad_tracked_page(struct page *page)
2871 * Per page-flags.h, pages tagged PG_reserved "should in general not be
2872 * touched (e.g. set dirty) except by its owner".
2874 return !PageReserved(page);
2877 static void kvm_set_page_dirty(struct page *page)
2879 if (kvm_is_ad_tracked_page(page))
2883 static void kvm_set_page_accessed(struct page *page)
2885 if (kvm_is_ad_tracked_page(page))
2886 mark_page_accessed(page);
2889 void kvm_release_page_clean(struct page *page)
2891 WARN_ON(is_error_page(page));
2893 kvm_set_page_accessed(page);
2896 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
2898 void kvm_release_pfn_clean(kvm_pfn_t pfn)
2902 if (is_error_noslot_pfn(pfn))
2905 page = kvm_pfn_to_refcounted_page(pfn);
2909 kvm_release_page_clean(page);
2911 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
2913 void kvm_release_page_dirty(struct page *page)
2915 WARN_ON(is_error_page(page));
2917 kvm_set_page_dirty(page);
2918 kvm_release_page_clean(page);
2920 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
2922 void kvm_release_pfn_dirty(kvm_pfn_t pfn)
2926 if (is_error_noslot_pfn(pfn))
2929 page = kvm_pfn_to_refcounted_page(pfn);
2933 kvm_release_page_dirty(page);
2935 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
2938 * Note, checking for an error/noslot pfn is the caller's responsibility when
2939 * directly marking a page dirty/accessed. Unlike the "release" helpers, the
2940 * "set" helpers are not to be used when the pfn might point at garbage.
2942 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
2944 if (WARN_ON(is_error_noslot_pfn(pfn)))
2948 kvm_set_page_dirty(pfn_to_page(pfn));
2950 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
2952 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
2954 if (WARN_ON(is_error_noslot_pfn(pfn)))
2958 kvm_set_page_accessed(pfn_to_page(pfn));
2960 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
2962 static int next_segment(unsigned long len, int offset)
2964 if (len > PAGE_SIZE - offset)
2965 return PAGE_SIZE - offset;
2970 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
2971 void *data, int offset, int len)
2976 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
2977 if (kvm_is_error_hva(addr))
2979 r = __copy_from_user(data, (void __user *)addr + offset, len);
2985 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
2988 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2990 return __kvm_read_guest_page(slot, gfn, data, offset, len);
2992 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
2994 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
2995 int offset, int len)
2997 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2999 return __kvm_read_guest_page(slot, gfn, data, offset, len);
3001 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
3003 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
3005 gfn_t gfn = gpa >> PAGE_SHIFT;
3007 int offset = offset_in_page(gpa);
3010 while ((seg = next_segment(len, offset)) != 0) {
3011 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
3021 EXPORT_SYMBOL_GPL(kvm_read_guest);
3023 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
3025 gfn_t gfn = gpa >> PAGE_SHIFT;
3027 int offset = offset_in_page(gpa);
3030 while ((seg = next_segment(len, offset)) != 0) {
3031 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
3041 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
3043 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
3044 void *data, int offset, unsigned long len)
3049 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
3050 if (kvm_is_error_hva(addr))
3052 pagefault_disable();
3053 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
3060 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
3061 void *data, unsigned long len)
3063 gfn_t gfn = gpa >> PAGE_SHIFT;
3064 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
3065 int offset = offset_in_page(gpa);
3067 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
3069 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
3071 static int __kvm_write_guest_page(struct kvm *kvm,
3072 struct kvm_memory_slot *memslot, gfn_t gfn,
3073 const void *data, int offset, int len)
3078 addr = gfn_to_hva_memslot(memslot, gfn);
3079 if (kvm_is_error_hva(addr))
3081 r = __copy_to_user((void __user *)addr + offset, data, len);
3084 mark_page_dirty_in_slot(kvm, memslot, gfn);
3088 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
3089 const void *data, int offset, int len)
3091 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
3093 return __kvm_write_guest_page(kvm, slot, gfn, data, offset, len);
3095 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
3097 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
3098 const void *data, int offset, int len)
3100 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
3102 return __kvm_write_guest_page(vcpu->kvm, slot, gfn, data, offset, len);
3104 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
3106 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
3109 gfn_t gfn = gpa >> PAGE_SHIFT;
3111 int offset = offset_in_page(gpa);
3114 while ((seg = next_segment(len, offset)) != 0) {
3115 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
3125 EXPORT_SYMBOL_GPL(kvm_write_guest);
3127 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
3130 gfn_t gfn = gpa >> PAGE_SHIFT;
3132 int offset = offset_in_page(gpa);
3135 while ((seg = next_segment(len, offset)) != 0) {
3136 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
3146 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
3148 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
3149 struct gfn_to_hva_cache *ghc,
3150 gpa_t gpa, unsigned long len)
3152 int offset = offset_in_page(gpa);
3153 gfn_t start_gfn = gpa >> PAGE_SHIFT;
3154 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
3155 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
3156 gfn_t nr_pages_avail;
3158 /* Update ghc->generation before performing any error checks. */
3159 ghc->generation = slots->generation;
3161 if (start_gfn > end_gfn) {
3162 ghc->hva = KVM_HVA_ERR_BAD;
3167 * If the requested region crosses two memslots, we still
3168 * verify that the entire region is valid here.
3170 for ( ; start_gfn <= end_gfn; start_gfn += nr_pages_avail) {
3171 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
3172 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
3174 if (kvm_is_error_hva(ghc->hva))
3178 /* Use the slow path for cross page reads and writes. */
3179 if (nr_pages_needed == 1)
3182 ghc->memslot = NULL;
3189 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3190 gpa_t gpa, unsigned long len)
3192 struct kvm_memslots *slots = kvm_memslots(kvm);
3193 return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len);
3195 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
3197 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3198 void *data, unsigned int offset,
3201 struct kvm_memslots *slots = kvm_memslots(kvm);
3203 gpa_t gpa = ghc->gpa + offset;
3205 if (WARN_ON_ONCE(len + offset > ghc->len))
3208 if (slots->generation != ghc->generation) {
3209 if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len))
3213 if (kvm_is_error_hva(ghc->hva))
3216 if (unlikely(!ghc->memslot))
3217 return kvm_write_guest(kvm, gpa, data, len);
3219 r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
3222 mark_page_dirty_in_slot(kvm, ghc->memslot, gpa >> PAGE_SHIFT);
3226 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached);
3228 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3229 void *data, unsigned long len)
3231 return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len);
3233 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
3235 int kvm_read_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3236 void *data, unsigned int offset,
3239 struct kvm_memslots *slots = kvm_memslots(kvm);
3241 gpa_t gpa = ghc->gpa + offset;
3243 if (WARN_ON_ONCE(len + offset > ghc->len))
3246 if (slots->generation != ghc->generation) {
3247 if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len))
3251 if (kvm_is_error_hva(ghc->hva))
3254 if (unlikely(!ghc->memslot))
3255 return kvm_read_guest(kvm, gpa, data, len);
3257 r = __copy_from_user(data, (void __user *)ghc->hva + offset, len);
3263 EXPORT_SYMBOL_GPL(kvm_read_guest_offset_cached);
3265 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3266 void *data, unsigned long len)
3268 return kvm_read_guest_offset_cached(kvm, ghc, data, 0, len);
3270 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
3272 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
3274 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
3275 gfn_t gfn = gpa >> PAGE_SHIFT;
3277 int offset = offset_in_page(gpa);
3280 while ((seg = next_segment(len, offset)) != 0) {
3281 ret = kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
3290 EXPORT_SYMBOL_GPL(kvm_clear_guest);
3292 void mark_page_dirty_in_slot(struct kvm *kvm,
3293 const struct kvm_memory_slot *memslot,
3296 struct kvm_vcpu *vcpu = kvm_get_running_vcpu();
3298 #ifdef CONFIG_HAVE_KVM_DIRTY_RING
3299 if (WARN_ON_ONCE(!vcpu) || WARN_ON_ONCE(vcpu->kvm != kvm))
3303 if (memslot && kvm_slot_dirty_track_enabled(memslot)) {
3304 unsigned long rel_gfn = gfn - memslot->base_gfn;
3305 u32 slot = (memslot->as_id << 16) | memslot->id;
3307 if (kvm->dirty_ring_size)
3308 kvm_dirty_ring_push(&vcpu->dirty_ring,
3311 set_bit_le(rel_gfn, memslot->dirty_bitmap);
3314 EXPORT_SYMBOL_GPL(mark_page_dirty_in_slot);
3316 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
3318 struct kvm_memory_slot *memslot;
3320 memslot = gfn_to_memslot(kvm, gfn);
3321 mark_page_dirty_in_slot(kvm, memslot, gfn);
3323 EXPORT_SYMBOL_GPL(mark_page_dirty);
3325 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
3327 struct kvm_memory_slot *memslot;
3329 memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
3330 mark_page_dirty_in_slot(vcpu->kvm, memslot, gfn);
3332 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
3334 void kvm_sigset_activate(struct kvm_vcpu *vcpu)
3336 if (!vcpu->sigset_active)
3340 * This does a lockless modification of ->real_blocked, which is fine
3341 * because, only current can change ->real_blocked and all readers of
3342 * ->real_blocked don't care as long ->real_blocked is always a subset
3345 sigprocmask(SIG_SETMASK, &vcpu->sigset, ¤t->real_blocked);
3348 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu)
3350 if (!vcpu->sigset_active)
3353 sigprocmask(SIG_SETMASK, ¤t->real_blocked, NULL);
3354 sigemptyset(¤t->real_blocked);
3357 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
3359 unsigned int old, val, grow, grow_start;
3361 old = val = vcpu->halt_poll_ns;
3362 grow_start = READ_ONCE(halt_poll_ns_grow_start);
3363 grow = READ_ONCE(halt_poll_ns_grow);
3368 if (val < grow_start)
3371 if (val > vcpu->kvm->max_halt_poll_ns)
3372 val = vcpu->kvm->max_halt_poll_ns;
3374 vcpu->halt_poll_ns = val;
3376 trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
3379 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
3381 unsigned int old, val, shrink, grow_start;
3383 old = val = vcpu->halt_poll_ns;
3384 shrink = READ_ONCE(halt_poll_ns_shrink);
3385 grow_start = READ_ONCE(halt_poll_ns_grow_start);
3391 if (val < grow_start)
3394 vcpu->halt_poll_ns = val;
3395 trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
3398 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
3401 int idx = srcu_read_lock(&vcpu->kvm->srcu);
3403 if (kvm_arch_vcpu_runnable(vcpu)) {
3404 kvm_make_request(KVM_REQ_UNHALT, vcpu);
3407 if (kvm_cpu_has_pending_timer(vcpu))
3409 if (signal_pending(current))
3411 if (kvm_check_request(KVM_REQ_UNBLOCK, vcpu))
3416 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3421 * Block the vCPU until the vCPU is runnable, an event arrives, or a signal is
3422 * pending. This is mostly used when halting a vCPU, but may also be used
3423 * directly for other vCPU non-runnable states, e.g. x86's Wait-For-SIPI.
3425 bool kvm_vcpu_block(struct kvm_vcpu *vcpu)
3427 struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu);
3428 bool waited = false;
3430 vcpu->stat.generic.blocking = 1;
3433 kvm_arch_vcpu_blocking(vcpu);
3434 prepare_to_rcuwait(wait);
3438 set_current_state(TASK_INTERRUPTIBLE);
3440 if (kvm_vcpu_check_block(vcpu) < 0)
3448 finish_rcuwait(wait);
3449 kvm_arch_vcpu_unblocking(vcpu);
3452 vcpu->stat.generic.blocking = 0;
3457 static inline void update_halt_poll_stats(struct kvm_vcpu *vcpu, ktime_t start,
3458 ktime_t end, bool success)
3460 struct kvm_vcpu_stat_generic *stats = &vcpu->stat.generic;
3461 u64 poll_ns = ktime_to_ns(ktime_sub(end, start));
3463 ++vcpu->stat.generic.halt_attempted_poll;
3466 ++vcpu->stat.generic.halt_successful_poll;
3468 if (!vcpu_valid_wakeup(vcpu))
3469 ++vcpu->stat.generic.halt_poll_invalid;
3471 stats->halt_poll_success_ns += poll_ns;
3472 KVM_STATS_LOG_HIST_UPDATE(stats->halt_poll_success_hist, poll_ns);
3474 stats->halt_poll_fail_ns += poll_ns;
3475 KVM_STATS_LOG_HIST_UPDATE(stats->halt_poll_fail_hist, poll_ns);
3480 * Emulate a vCPU halt condition, e.g. HLT on x86, WFI on arm, etc... If halt
3481 * polling is enabled, busy wait for a short time before blocking to avoid the
3482 * expensive block+unblock sequence if a wake event arrives soon after the vCPU
3485 void kvm_vcpu_halt(struct kvm_vcpu *vcpu)
3487 bool halt_poll_allowed = !kvm_arch_no_poll(vcpu);
3488 bool do_halt_poll = halt_poll_allowed && vcpu->halt_poll_ns;
3489 ktime_t start, cur, poll_end;
3490 bool waited = false;
3493 start = cur = poll_end = ktime_get();
3495 ktime_t stop = ktime_add_ns(start, vcpu->halt_poll_ns);
3499 * This sets KVM_REQ_UNHALT if an interrupt
3502 if (kvm_vcpu_check_block(vcpu) < 0)
3505 poll_end = cur = ktime_get();
3506 } while (kvm_vcpu_can_poll(cur, stop));
3509 waited = kvm_vcpu_block(vcpu);
3513 vcpu->stat.generic.halt_wait_ns +=
3514 ktime_to_ns(cur) - ktime_to_ns(poll_end);
3515 KVM_STATS_LOG_HIST_UPDATE(vcpu->stat.generic.halt_wait_hist,
3516 ktime_to_ns(cur) - ktime_to_ns(poll_end));
3519 /* The total time the vCPU was "halted", including polling time. */
3520 halt_ns = ktime_to_ns(cur) - ktime_to_ns(start);
3523 * Note, halt-polling is considered successful so long as the vCPU was
3524 * never actually scheduled out, i.e. even if the wake event arrived
3525 * after of the halt-polling loop itself, but before the full wait.
3528 update_halt_poll_stats(vcpu, start, poll_end, !waited);
3530 if (halt_poll_allowed) {
3531 if (!vcpu_valid_wakeup(vcpu)) {
3532 shrink_halt_poll_ns(vcpu);
3533 } else if (vcpu->kvm->max_halt_poll_ns) {
3534 if (halt_ns <= vcpu->halt_poll_ns)
3536 /* we had a long block, shrink polling */
3537 else if (vcpu->halt_poll_ns &&
3538 halt_ns > vcpu->kvm->max_halt_poll_ns)
3539 shrink_halt_poll_ns(vcpu);
3540 /* we had a short halt and our poll time is too small */
3541 else if (vcpu->halt_poll_ns < vcpu->kvm->max_halt_poll_ns &&
3542 halt_ns < vcpu->kvm->max_halt_poll_ns)
3543 grow_halt_poll_ns(vcpu);
3545 vcpu->halt_poll_ns = 0;
3549 trace_kvm_vcpu_wakeup(halt_ns, waited, vcpu_valid_wakeup(vcpu));
3551 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
3553 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
3555 if (__kvm_vcpu_wake_up(vcpu)) {
3556 WRITE_ONCE(vcpu->ready, true);
3557 ++vcpu->stat.generic.halt_wakeup;
3563 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
3567 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
3569 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
3573 if (kvm_vcpu_wake_up(vcpu))
3578 * The only state change done outside the vcpu mutex is IN_GUEST_MODE
3579 * to EXITING_GUEST_MODE. Therefore the moderately expensive "should
3580 * kick" check does not need atomic operations if kvm_vcpu_kick is used
3581 * within the vCPU thread itself.
3583 if (vcpu == __this_cpu_read(kvm_running_vcpu)) {
3584 if (vcpu->mode == IN_GUEST_MODE)
3585 WRITE_ONCE(vcpu->mode, EXITING_GUEST_MODE);
3590 * Note, the vCPU could get migrated to a different pCPU at any point
3591 * after kvm_arch_vcpu_should_kick(), which could result in sending an
3592 * IPI to the previous pCPU. But, that's ok because the purpose of the
3593 * IPI is to force the vCPU to leave IN_GUEST_MODE, and migrating the
3594 * vCPU also requires it to leave IN_GUEST_MODE.
3596 if (kvm_arch_vcpu_should_kick(vcpu)) {
3597 cpu = READ_ONCE(vcpu->cpu);
3598 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
3599 smp_send_reschedule(cpu);
3604 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
3605 #endif /* !CONFIG_S390 */
3607 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
3610 struct task_struct *task = NULL;
3614 pid = rcu_dereference(target->pid);
3616 task = get_pid_task(pid, PIDTYPE_PID);
3620 ret = yield_to(task, 1);
3621 put_task_struct(task);
3625 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
3628 * Helper that checks whether a VCPU is eligible for directed yield.
3629 * Most eligible candidate to yield is decided by following heuristics:
3631 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
3632 * (preempted lock holder), indicated by @in_spin_loop.
3633 * Set at the beginning and cleared at the end of interception/PLE handler.
3635 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
3636 * chance last time (mostly it has become eligible now since we have probably
3637 * yielded to lockholder in last iteration. This is done by toggling
3638 * @dy_eligible each time a VCPU checked for eligibility.)
3640 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
3641 * to preempted lock-holder could result in wrong VCPU selection and CPU
3642 * burning. Giving priority for a potential lock-holder increases lock
3645 * Since algorithm is based on heuristics, accessing another VCPU data without
3646 * locking does not harm. It may result in trying to yield to same VCPU, fail
3647 * and continue with next VCPU and so on.
3649 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
3651 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
3654 eligible = !vcpu->spin_loop.in_spin_loop ||
3655 vcpu->spin_loop.dy_eligible;
3657 if (vcpu->spin_loop.in_spin_loop)
3658 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
3667 * Unlike kvm_arch_vcpu_runnable, this function is called outside
3668 * a vcpu_load/vcpu_put pair. However, for most architectures
3669 * kvm_arch_vcpu_runnable does not require vcpu_load.
3671 bool __weak kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
3673 return kvm_arch_vcpu_runnable(vcpu);
3676 static bool vcpu_dy_runnable(struct kvm_vcpu *vcpu)
3678 if (kvm_arch_dy_runnable(vcpu))
3681 #ifdef CONFIG_KVM_ASYNC_PF
3682 if (!list_empty_careful(&vcpu->async_pf.done))
3689 bool __weak kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu)
3694 void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode)
3696 struct kvm *kvm = me->kvm;
3697 struct kvm_vcpu *vcpu;
3698 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
3704 kvm_vcpu_set_in_spin_loop(me, true);
3706 * We boost the priority of a VCPU that is runnable but not
3707 * currently running, because it got preempted by something
3708 * else and called schedule in __vcpu_run. Hopefully that
3709 * VCPU is holding the lock that we need and will release it.
3710 * We approximate round-robin by starting at the last boosted VCPU.
3712 for (pass = 0; pass < 2 && !yielded && try; pass++) {
3713 kvm_for_each_vcpu(i, vcpu, kvm) {
3714 if (!pass && i <= last_boosted_vcpu) {
3715 i = last_boosted_vcpu;
3717 } else if (pass && i > last_boosted_vcpu)
3719 if (!READ_ONCE(vcpu->ready))
3723 if (kvm_vcpu_is_blocking(vcpu) && !vcpu_dy_runnable(vcpu))
3725 if (READ_ONCE(vcpu->preempted) && yield_to_kernel_mode &&
3726 !kvm_arch_dy_has_pending_interrupt(vcpu) &&
3727 !kvm_arch_vcpu_in_kernel(vcpu))
3729 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
3732 yielded = kvm_vcpu_yield_to(vcpu);
3734 kvm->last_boosted_vcpu = i;
3736 } else if (yielded < 0) {
3743 kvm_vcpu_set_in_spin_loop(me, false);
3745 /* Ensure vcpu is not eligible during next spinloop */
3746 kvm_vcpu_set_dy_eligible(me, false);
3748 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
3750 static bool kvm_page_in_dirty_ring(struct kvm *kvm, unsigned long pgoff)
3752 #ifdef CONFIG_HAVE_KVM_DIRTY_RING
3753 return (pgoff >= KVM_DIRTY_LOG_PAGE_OFFSET) &&
3754 (pgoff < KVM_DIRTY_LOG_PAGE_OFFSET +
3755 kvm->dirty_ring_size / PAGE_SIZE);
3761 static vm_fault_t kvm_vcpu_fault(struct vm_fault *vmf)
3763 struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data;
3766 if (vmf->pgoff == 0)
3767 page = virt_to_page(vcpu->run);
3769 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
3770 page = virt_to_page(vcpu->arch.pio_data);
3772 #ifdef CONFIG_KVM_MMIO
3773 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
3774 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
3776 else if (kvm_page_in_dirty_ring(vcpu->kvm, vmf->pgoff))
3777 page = kvm_dirty_ring_get_page(
3779 vmf->pgoff - KVM_DIRTY_LOG_PAGE_OFFSET);
3781 return kvm_arch_vcpu_fault(vcpu, vmf);
3787 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
3788 .fault = kvm_vcpu_fault,
3791 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
3793 struct kvm_vcpu *vcpu = file->private_data;
3794 unsigned long pages = vma_pages(vma);
3796 if ((kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff) ||
3797 kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff + pages - 1)) &&
3798 ((vma->vm_flags & VM_EXEC) || !(vma->vm_flags & VM_SHARED)))
3801 vma->vm_ops = &kvm_vcpu_vm_ops;
3805 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
3807 struct kvm_vcpu *vcpu = filp->private_data;
3809 kvm_put_kvm(vcpu->kvm);
3813 static const struct file_operations kvm_vcpu_fops = {
3814 .release = kvm_vcpu_release,
3815 .unlocked_ioctl = kvm_vcpu_ioctl,
3816 .mmap = kvm_vcpu_mmap,
3817 .llseek = noop_llseek,
3818 KVM_COMPAT(kvm_vcpu_compat_ioctl),
3822 * Allocates an inode for the vcpu.
3824 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
3826 char name[8 + 1 + ITOA_MAX_LEN + 1];
3828 snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id);
3829 return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
3832 #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS
3833 static int vcpu_get_pid(void *data, u64 *val)
3835 struct kvm_vcpu *vcpu = (struct kvm_vcpu *) data;
3836 *val = pid_nr(rcu_access_pointer(vcpu->pid));
3840 DEFINE_SIMPLE_ATTRIBUTE(vcpu_get_pid_fops, vcpu_get_pid, NULL, "%llu\n");
3842 static void kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
3844 struct dentry *debugfs_dentry;
3845 char dir_name[ITOA_MAX_LEN * 2];
3847 if (!debugfs_initialized())
3850 snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
3851 debugfs_dentry = debugfs_create_dir(dir_name,
3852 vcpu->kvm->debugfs_dentry);
3853 debugfs_create_file("pid", 0444, debugfs_dentry, vcpu,
3854 &vcpu_get_pid_fops);
3856 kvm_arch_create_vcpu_debugfs(vcpu, debugfs_dentry);
3861 * Creates some virtual cpus. Good luck creating more than one.
3863 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
3866 struct kvm_vcpu *vcpu;
3869 if (id >= KVM_MAX_VCPU_IDS)
3872 mutex_lock(&kvm->lock);
3873 if (kvm->created_vcpus >= kvm->max_vcpus) {
3874 mutex_unlock(&kvm->lock);
3878 r = kvm_arch_vcpu_precreate(kvm, id);
3880 mutex_unlock(&kvm->lock);
3884 kvm->created_vcpus++;
3885 mutex_unlock(&kvm->lock);
3887 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL_ACCOUNT);
3890 goto vcpu_decrement;
3893 BUILD_BUG_ON(sizeof(struct kvm_run) > PAGE_SIZE);
3894 page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
3899 vcpu->run = page_address(page);
3901 kvm_vcpu_init(vcpu, kvm, id);
3903 r = kvm_arch_vcpu_create(vcpu);
3905 goto vcpu_free_run_page;
3907 if (kvm->dirty_ring_size) {
3908 r = kvm_dirty_ring_alloc(&vcpu->dirty_ring,
3909 id, kvm->dirty_ring_size);
3911 goto arch_vcpu_destroy;
3914 mutex_lock(&kvm->lock);
3915 if (kvm_get_vcpu_by_id(kvm, id)) {
3917 goto unlock_vcpu_destroy;
3920 vcpu->vcpu_idx = atomic_read(&kvm->online_vcpus);
3921 r = xa_insert(&kvm->vcpu_array, vcpu->vcpu_idx, vcpu, GFP_KERNEL_ACCOUNT);
3922 BUG_ON(r == -EBUSY);
3924 goto unlock_vcpu_destroy;
3926 /* Now it's all set up, let userspace reach it */
3928 r = create_vcpu_fd(vcpu);
3930 xa_erase(&kvm->vcpu_array, vcpu->vcpu_idx);
3931 kvm_put_kvm_no_destroy(kvm);
3932 goto unlock_vcpu_destroy;
3936 * Pairs with smp_rmb() in kvm_get_vcpu. Store the vcpu
3937 * pointer before kvm->online_vcpu's incremented value.
3940 atomic_inc(&kvm->online_vcpus);
3942 mutex_unlock(&kvm->lock);
3943 kvm_arch_vcpu_postcreate(vcpu);
3944 kvm_create_vcpu_debugfs(vcpu);
3947 unlock_vcpu_destroy:
3948 mutex_unlock(&kvm->lock);
3949 kvm_dirty_ring_free(&vcpu->dirty_ring);
3951 kvm_arch_vcpu_destroy(vcpu);
3953 free_page((unsigned long)vcpu->run);
3955 kmem_cache_free(kvm_vcpu_cache, vcpu);
3957 mutex_lock(&kvm->lock);
3958 kvm->created_vcpus--;
3959 mutex_unlock(&kvm->lock);
3963 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
3966 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
3967 vcpu->sigset_active = 1;
3968 vcpu->sigset = *sigset;
3970 vcpu->sigset_active = 0;
3974 static ssize_t kvm_vcpu_stats_read(struct file *file, char __user *user_buffer,
3975 size_t size, loff_t *offset)
3977 struct kvm_vcpu *vcpu = file->private_data;
3979 return kvm_stats_read(vcpu->stats_id, &kvm_vcpu_stats_header,
3980 &kvm_vcpu_stats_desc[0], &vcpu->stat,
3981 sizeof(vcpu->stat), user_buffer, size, offset);
3984 static const struct file_operations kvm_vcpu_stats_fops = {
3985 .read = kvm_vcpu_stats_read,
3986 .llseek = noop_llseek,
3989 static int kvm_vcpu_ioctl_get_stats_fd(struct kvm_vcpu *vcpu)
3993 char name[15 + ITOA_MAX_LEN + 1];
3995 snprintf(name, sizeof(name), "kvm-vcpu-stats:%d", vcpu->vcpu_id);
3997 fd = get_unused_fd_flags(O_CLOEXEC);
4001 file = anon_inode_getfile(name, &kvm_vcpu_stats_fops, vcpu, O_RDONLY);
4004 return PTR_ERR(file);
4006 file->f_mode |= FMODE_PREAD;
4007 fd_install(fd, file);
4012 static long kvm_vcpu_ioctl(struct file *filp,
4013 unsigned int ioctl, unsigned long arg)
4015 struct kvm_vcpu *vcpu = filp->private_data;
4016 void __user *argp = (void __user *)arg;
4018 struct kvm_fpu *fpu = NULL;
4019 struct kvm_sregs *kvm_sregs = NULL;
4021 if (vcpu->kvm->mm != current->mm || vcpu->kvm->vm_dead)
4024 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
4028 * Some architectures have vcpu ioctls that are asynchronous to vcpu
4029 * execution; mutex_lock() would break them.
4031 r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg);
4032 if (r != -ENOIOCTLCMD)
4035 if (mutex_lock_killable(&vcpu->mutex))
4043 oldpid = rcu_access_pointer(vcpu->pid);
4044 if (unlikely(oldpid != task_pid(current))) {
4045 /* The thread running this VCPU changed. */
4048 r = kvm_arch_vcpu_run_pid_change(vcpu);
4052 newpid = get_task_pid(current, PIDTYPE_PID);
4053 rcu_assign_pointer(vcpu->pid, newpid);
4058 r = kvm_arch_vcpu_ioctl_run(vcpu);
4059 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
4062 case KVM_GET_REGS: {
4063 struct kvm_regs *kvm_regs;
4066 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL_ACCOUNT);
4069 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
4073 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
4080 case KVM_SET_REGS: {
4081 struct kvm_regs *kvm_regs;
4083 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
4084 if (IS_ERR(kvm_regs)) {
4085 r = PTR_ERR(kvm_regs);
4088 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
4092 case KVM_GET_SREGS: {
4093 kvm_sregs = kzalloc(sizeof(struct kvm_sregs),
4094 GFP_KERNEL_ACCOUNT);
4098 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
4102 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
4107 case KVM_SET_SREGS: {
4108 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
4109 if (IS_ERR(kvm_sregs)) {
4110 r = PTR_ERR(kvm_sregs);
4114 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
4117 case KVM_GET_MP_STATE: {
4118 struct kvm_mp_state mp_state;
4120 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
4124 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
4129 case KVM_SET_MP_STATE: {
4130 struct kvm_mp_state mp_state;
4133 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
4135 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
4138 case KVM_TRANSLATE: {
4139 struct kvm_translation tr;
4142 if (copy_from_user(&tr, argp, sizeof(tr)))
4144 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
4148 if (copy_to_user(argp, &tr, sizeof(tr)))
4153 case KVM_SET_GUEST_DEBUG: {
4154 struct kvm_guest_debug dbg;
4157 if (copy_from_user(&dbg, argp, sizeof(dbg)))
4159 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
4162 case KVM_SET_SIGNAL_MASK: {
4163 struct kvm_signal_mask __user *sigmask_arg = argp;
4164 struct kvm_signal_mask kvm_sigmask;
4165 sigset_t sigset, *p;
4170 if (copy_from_user(&kvm_sigmask, argp,
4171 sizeof(kvm_sigmask)))
4174 if (kvm_sigmask.len != sizeof(sigset))
4177 if (copy_from_user(&sigset, sigmask_arg->sigset,
4182 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
4186 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL_ACCOUNT);
4190 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
4194 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
4200 fpu = memdup_user(argp, sizeof(*fpu));
4206 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
4209 case KVM_GET_STATS_FD: {
4210 r = kvm_vcpu_ioctl_get_stats_fd(vcpu);
4214 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
4217 mutex_unlock(&vcpu->mutex);
4223 #ifdef CONFIG_KVM_COMPAT
4224 static long kvm_vcpu_compat_ioctl(struct file *filp,
4225 unsigned int ioctl, unsigned long arg)
4227 struct kvm_vcpu *vcpu = filp->private_data;
4228 void __user *argp = compat_ptr(arg);
4231 if (vcpu->kvm->mm != current->mm || vcpu->kvm->vm_dead)
4235 case KVM_SET_SIGNAL_MASK: {
4236 struct kvm_signal_mask __user *sigmask_arg = argp;
4237 struct kvm_signal_mask kvm_sigmask;
4242 if (copy_from_user(&kvm_sigmask, argp,
4243 sizeof(kvm_sigmask)))
4246 if (kvm_sigmask.len != sizeof(compat_sigset_t))
4249 if (get_compat_sigset(&sigset,
4250 (compat_sigset_t __user *)sigmask_arg->sigset))
4252 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
4254 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
4258 r = kvm_vcpu_ioctl(filp, ioctl, arg);
4266 static int kvm_device_mmap(struct file *filp, struct vm_area_struct *vma)
4268 struct kvm_device *dev = filp->private_data;
4271 return dev->ops->mmap(dev, vma);
4276 static int kvm_device_ioctl_attr(struct kvm_device *dev,
4277 int (*accessor)(struct kvm_device *dev,
4278 struct kvm_device_attr *attr),
4281 struct kvm_device_attr attr;
4286 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
4289 return accessor(dev, &attr);
4292 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
4295 struct kvm_device *dev = filp->private_data;
4297 if (dev->kvm->mm != current->mm || dev->kvm->vm_dead)
4301 case KVM_SET_DEVICE_ATTR:
4302 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
4303 case KVM_GET_DEVICE_ATTR:
4304 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
4305 case KVM_HAS_DEVICE_ATTR:
4306 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
4308 if (dev->ops->ioctl)
4309 return dev->ops->ioctl(dev, ioctl, arg);
4315 static int kvm_device_release(struct inode *inode, struct file *filp)
4317 struct kvm_device *dev = filp->private_data;
4318 struct kvm *kvm = dev->kvm;
4320 if (dev->ops->release) {
4321 mutex_lock(&kvm->lock);
4322 list_del(&dev->vm_node);
4323 dev->ops->release(dev);
4324 mutex_unlock(&kvm->lock);
4331 static const struct file_operations kvm_device_fops = {
4332 .unlocked_ioctl = kvm_device_ioctl,
4333 .release = kvm_device_release,
4334 KVM_COMPAT(kvm_device_ioctl),
4335 .mmap = kvm_device_mmap,
4338 struct kvm_device *kvm_device_from_filp(struct file *filp)
4340 if (filp->f_op != &kvm_device_fops)
4343 return filp->private_data;
4346 static const struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
4347 #ifdef CONFIG_KVM_MPIC
4348 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
4349 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
4353 int kvm_register_device_ops(const struct kvm_device_ops *ops, u32 type)
4355 if (type >= ARRAY_SIZE(kvm_device_ops_table))
4358 if (kvm_device_ops_table[type] != NULL)
4361 kvm_device_ops_table[type] = ops;
4365 void kvm_unregister_device_ops(u32 type)
4367 if (kvm_device_ops_table[type] != NULL)
4368 kvm_device_ops_table[type] = NULL;
4371 static int kvm_ioctl_create_device(struct kvm *kvm,
4372 struct kvm_create_device *cd)
4374 const struct kvm_device_ops *ops = NULL;
4375 struct kvm_device *dev;
4376 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
4380 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
4383 type = array_index_nospec(cd->type, ARRAY_SIZE(kvm_device_ops_table));
4384 ops = kvm_device_ops_table[type];
4391 dev = kzalloc(sizeof(*dev), GFP_KERNEL_ACCOUNT);
4398 mutex_lock(&kvm->lock);
4399 ret = ops->create(dev, type);
4401 mutex_unlock(&kvm->lock);
4405 list_add(&dev->vm_node, &kvm->devices);
4406 mutex_unlock(&kvm->lock);
4412 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
4414 kvm_put_kvm_no_destroy(kvm);
4415 mutex_lock(&kvm->lock);
4416 list_del(&dev->vm_node);
4419 mutex_unlock(&kvm->lock);
4429 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
4432 case KVM_CAP_USER_MEMORY:
4433 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
4434 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
4435 case KVM_CAP_INTERNAL_ERROR_DATA:
4436 #ifdef CONFIG_HAVE_KVM_MSI
4437 case KVM_CAP_SIGNAL_MSI:
4439 #ifdef CONFIG_HAVE_KVM_IRQFD
4441 case KVM_CAP_IRQFD_RESAMPLE:
4443 case KVM_CAP_IOEVENTFD_ANY_LENGTH:
4444 case KVM_CAP_CHECK_EXTENSION_VM:
4445 case KVM_CAP_ENABLE_CAP_VM:
4446 case KVM_CAP_HALT_POLL:
4448 #ifdef CONFIG_KVM_MMIO
4449 case KVM_CAP_COALESCED_MMIO:
4450 return KVM_COALESCED_MMIO_PAGE_OFFSET;
4451 case KVM_CAP_COALESCED_PIO:
4454 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4455 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2:
4456 return KVM_DIRTY_LOG_MANUAL_CAPS;
4458 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
4459 case KVM_CAP_IRQ_ROUTING:
4460 return KVM_MAX_IRQ_ROUTES;
4462 #if KVM_ADDRESS_SPACE_NUM > 1
4463 case KVM_CAP_MULTI_ADDRESS_SPACE:
4464 return KVM_ADDRESS_SPACE_NUM;
4466 case KVM_CAP_NR_MEMSLOTS:
4467 return KVM_USER_MEM_SLOTS;
4468 case KVM_CAP_DIRTY_LOG_RING:
4469 #ifdef CONFIG_HAVE_KVM_DIRTY_RING
4470 return KVM_DIRTY_RING_MAX_ENTRIES * sizeof(struct kvm_dirty_gfn);
4474 case KVM_CAP_BINARY_STATS_FD:
4475 case KVM_CAP_SYSTEM_EVENT_DATA:
4480 return kvm_vm_ioctl_check_extension(kvm, arg);
4483 static int kvm_vm_ioctl_enable_dirty_log_ring(struct kvm *kvm, u32 size)
4487 if (!KVM_DIRTY_LOG_PAGE_OFFSET)
4490 /* the size should be power of 2 */
4491 if (!size || (size & (size - 1)))
4494 /* Should be bigger to keep the reserved entries, or a page */
4495 if (size < kvm_dirty_ring_get_rsvd_entries() *
4496 sizeof(struct kvm_dirty_gfn) || size < PAGE_SIZE)
4499 if (size > KVM_DIRTY_RING_MAX_ENTRIES *
4500 sizeof(struct kvm_dirty_gfn))
4503 /* We only allow it to set once */
4504 if (kvm->dirty_ring_size)
4507 mutex_lock(&kvm->lock);
4509 if (kvm->created_vcpus) {
4510 /* We don't allow to change this value after vcpu created */
4513 kvm->dirty_ring_size = size;
4517 mutex_unlock(&kvm->lock);
4521 static int kvm_vm_ioctl_reset_dirty_pages(struct kvm *kvm)
4524 struct kvm_vcpu *vcpu;
4527 if (!kvm->dirty_ring_size)
4530 mutex_lock(&kvm->slots_lock);
4532 kvm_for_each_vcpu(i, vcpu, kvm)
4533 cleared += kvm_dirty_ring_reset(vcpu->kvm, &vcpu->dirty_ring);
4535 mutex_unlock(&kvm->slots_lock);
4538 kvm_flush_remote_tlbs(kvm);
4543 int __attribute__((weak)) kvm_vm_ioctl_enable_cap(struct kvm *kvm,
4544 struct kvm_enable_cap *cap)
4549 static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm,
4550 struct kvm_enable_cap *cap)
4553 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4554 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2: {
4555 u64 allowed_options = KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE;
4557 if (cap->args[0] & KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE)
4558 allowed_options = KVM_DIRTY_LOG_MANUAL_CAPS;
4560 if (cap->flags || (cap->args[0] & ~allowed_options))
4562 kvm->manual_dirty_log_protect = cap->args[0];
4566 case KVM_CAP_HALT_POLL: {
4567 if (cap->flags || cap->args[0] != (unsigned int)cap->args[0])
4570 kvm->max_halt_poll_ns = cap->args[0];
4573 case KVM_CAP_DIRTY_LOG_RING:
4574 return kvm_vm_ioctl_enable_dirty_log_ring(kvm, cap->args[0]);
4576 return kvm_vm_ioctl_enable_cap(kvm, cap);
4580 static ssize_t kvm_vm_stats_read(struct file *file, char __user *user_buffer,
4581 size_t size, loff_t *offset)
4583 struct kvm *kvm = file->private_data;
4585 return kvm_stats_read(kvm->stats_id, &kvm_vm_stats_header,
4586 &kvm_vm_stats_desc[0], &kvm->stat,
4587 sizeof(kvm->stat), user_buffer, size, offset);
4590 static const struct file_operations kvm_vm_stats_fops = {
4591 .read = kvm_vm_stats_read,
4592 .llseek = noop_llseek,
4595 static int kvm_vm_ioctl_get_stats_fd(struct kvm *kvm)
4600 fd = get_unused_fd_flags(O_CLOEXEC);
4604 file = anon_inode_getfile("kvm-vm-stats",
4605 &kvm_vm_stats_fops, kvm, O_RDONLY);
4608 return PTR_ERR(file);
4610 file->f_mode |= FMODE_PREAD;
4611 fd_install(fd, file);
4616 static long kvm_vm_ioctl(struct file *filp,
4617 unsigned int ioctl, unsigned long arg)
4619 struct kvm *kvm = filp->private_data;
4620 void __user *argp = (void __user *)arg;
4623 if (kvm->mm != current->mm || kvm->vm_dead)
4626 case KVM_CREATE_VCPU:
4627 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
4629 case KVM_ENABLE_CAP: {
4630 struct kvm_enable_cap cap;
4633 if (copy_from_user(&cap, argp, sizeof(cap)))
4635 r = kvm_vm_ioctl_enable_cap_generic(kvm, &cap);
4638 case KVM_SET_USER_MEMORY_REGION: {
4639 struct kvm_userspace_memory_region kvm_userspace_mem;
4642 if (copy_from_user(&kvm_userspace_mem, argp,
4643 sizeof(kvm_userspace_mem)))
4646 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
4649 case KVM_GET_DIRTY_LOG: {
4650 struct kvm_dirty_log log;
4653 if (copy_from_user(&log, argp, sizeof(log)))
4655 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
4658 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4659 case KVM_CLEAR_DIRTY_LOG: {
4660 struct kvm_clear_dirty_log log;
4663 if (copy_from_user(&log, argp, sizeof(log)))
4665 r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
4669 #ifdef CONFIG_KVM_MMIO
4670 case KVM_REGISTER_COALESCED_MMIO: {
4671 struct kvm_coalesced_mmio_zone zone;
4674 if (copy_from_user(&zone, argp, sizeof(zone)))
4676 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
4679 case KVM_UNREGISTER_COALESCED_MMIO: {
4680 struct kvm_coalesced_mmio_zone zone;
4683 if (copy_from_user(&zone, argp, sizeof(zone)))
4685 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
4690 struct kvm_irqfd data;
4693 if (copy_from_user(&data, argp, sizeof(data)))
4695 r = kvm_irqfd(kvm, &data);
4698 case KVM_IOEVENTFD: {
4699 struct kvm_ioeventfd data;
4702 if (copy_from_user(&data, argp, sizeof(data)))
4704 r = kvm_ioeventfd(kvm, &data);
4707 #ifdef CONFIG_HAVE_KVM_MSI
4708 case KVM_SIGNAL_MSI: {
4712 if (copy_from_user(&msi, argp, sizeof(msi)))
4714 r = kvm_send_userspace_msi(kvm, &msi);
4718 #ifdef __KVM_HAVE_IRQ_LINE
4719 case KVM_IRQ_LINE_STATUS:
4720 case KVM_IRQ_LINE: {
4721 struct kvm_irq_level irq_event;
4724 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
4727 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
4728 ioctl == KVM_IRQ_LINE_STATUS);
4733 if (ioctl == KVM_IRQ_LINE_STATUS) {
4734 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
4742 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
4743 case KVM_SET_GSI_ROUTING: {
4744 struct kvm_irq_routing routing;
4745 struct kvm_irq_routing __user *urouting;
4746 struct kvm_irq_routing_entry *entries = NULL;
4749 if (copy_from_user(&routing, argp, sizeof(routing)))
4752 if (!kvm_arch_can_set_irq_routing(kvm))
4754 if (routing.nr > KVM_MAX_IRQ_ROUTES)
4760 entries = vmemdup_user(urouting->entries,
4761 array_size(sizeof(*entries),
4763 if (IS_ERR(entries)) {
4764 r = PTR_ERR(entries);
4768 r = kvm_set_irq_routing(kvm, entries, routing.nr,
4773 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
4774 case KVM_CREATE_DEVICE: {
4775 struct kvm_create_device cd;
4778 if (copy_from_user(&cd, argp, sizeof(cd)))
4781 r = kvm_ioctl_create_device(kvm, &cd);
4786 if (copy_to_user(argp, &cd, sizeof(cd)))
4792 case KVM_CHECK_EXTENSION:
4793 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
4795 case KVM_RESET_DIRTY_RINGS:
4796 r = kvm_vm_ioctl_reset_dirty_pages(kvm);
4798 case KVM_GET_STATS_FD:
4799 r = kvm_vm_ioctl_get_stats_fd(kvm);
4802 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
4808 #ifdef CONFIG_KVM_COMPAT
4809 struct compat_kvm_dirty_log {
4813 compat_uptr_t dirty_bitmap; /* one bit per page */
4818 struct compat_kvm_clear_dirty_log {
4823 compat_uptr_t dirty_bitmap; /* one bit per page */
4828 static long kvm_vm_compat_ioctl(struct file *filp,
4829 unsigned int ioctl, unsigned long arg)
4831 struct kvm *kvm = filp->private_data;
4834 if (kvm->mm != current->mm || kvm->vm_dead)
4837 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4838 case KVM_CLEAR_DIRTY_LOG: {
4839 struct compat_kvm_clear_dirty_log compat_log;
4840 struct kvm_clear_dirty_log log;
4842 if (copy_from_user(&compat_log, (void __user *)arg,
4843 sizeof(compat_log)))
4845 log.slot = compat_log.slot;
4846 log.num_pages = compat_log.num_pages;
4847 log.first_page = compat_log.first_page;
4848 log.padding2 = compat_log.padding2;
4849 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
4851 r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
4855 case KVM_GET_DIRTY_LOG: {
4856 struct compat_kvm_dirty_log compat_log;
4857 struct kvm_dirty_log log;
4859 if (copy_from_user(&compat_log, (void __user *)arg,
4860 sizeof(compat_log)))
4862 log.slot = compat_log.slot;
4863 log.padding1 = compat_log.padding1;
4864 log.padding2 = compat_log.padding2;
4865 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
4867 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
4871 r = kvm_vm_ioctl(filp, ioctl, arg);
4877 static const struct file_operations kvm_vm_fops = {
4878 .release = kvm_vm_release,
4879 .unlocked_ioctl = kvm_vm_ioctl,
4880 .llseek = noop_llseek,
4881 KVM_COMPAT(kvm_vm_compat_ioctl),
4884 bool file_is_kvm(struct file *file)
4886 return file && file->f_op == &kvm_vm_fops;
4888 EXPORT_SYMBOL_GPL(file_is_kvm);
4890 static int kvm_dev_ioctl_create_vm(unsigned long type)
4892 char fdname[ITOA_MAX_LEN + 1];
4897 fd = get_unused_fd_flags(O_CLOEXEC);
4901 snprintf(fdname, sizeof(fdname), "%d", fd);
4903 kvm = kvm_create_vm(type);
4909 #ifdef CONFIG_KVM_MMIO
4910 r = kvm_coalesced_mmio_init(kvm);
4914 file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
4921 * Don't call kvm_put_kvm anymore at this point; file->f_op is
4922 * already set, with ->release() being kvm_vm_release(). In error
4923 * cases it will be called by the final fput(file) and will take
4924 * care of doing kvm_put_kvm(kvm).
4926 if (kvm_create_vm_debugfs(kvm, fdname) < 0) {
4931 kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm);
4933 fd_install(fd, file);
4943 static long kvm_dev_ioctl(struct file *filp,
4944 unsigned int ioctl, unsigned long arg)
4949 case KVM_GET_API_VERSION:
4952 r = KVM_API_VERSION;
4955 r = kvm_dev_ioctl_create_vm(arg);
4957 case KVM_CHECK_EXTENSION:
4958 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
4960 case KVM_GET_VCPU_MMAP_SIZE:
4963 r = PAGE_SIZE; /* struct kvm_run */
4965 r += PAGE_SIZE; /* pio data page */
4967 #ifdef CONFIG_KVM_MMIO
4968 r += PAGE_SIZE; /* coalesced mmio ring page */
4971 case KVM_TRACE_ENABLE:
4972 case KVM_TRACE_PAUSE:
4973 case KVM_TRACE_DISABLE:
4977 return kvm_arch_dev_ioctl(filp, ioctl, arg);
4983 static struct file_operations kvm_chardev_ops = {
4984 .unlocked_ioctl = kvm_dev_ioctl,
4985 .llseek = noop_llseek,
4986 KVM_COMPAT(kvm_dev_ioctl),
4989 static struct miscdevice kvm_dev = {
4995 static void hardware_enable_nolock(void *junk)
4997 int cpu = raw_smp_processor_id();
5000 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
5003 cpumask_set_cpu(cpu, cpus_hardware_enabled);
5005 r = kvm_arch_hardware_enable();
5008 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
5009 atomic_inc(&hardware_enable_failed);
5010 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
5014 static int kvm_starting_cpu(unsigned int cpu)
5016 raw_spin_lock(&kvm_count_lock);
5017 if (kvm_usage_count)
5018 hardware_enable_nolock(NULL);
5019 raw_spin_unlock(&kvm_count_lock);
5023 static void hardware_disable_nolock(void *junk)
5025 int cpu = raw_smp_processor_id();
5027 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
5029 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
5030 kvm_arch_hardware_disable();
5033 static int kvm_dying_cpu(unsigned int cpu)
5035 raw_spin_lock(&kvm_count_lock);
5036 if (kvm_usage_count)
5037 hardware_disable_nolock(NULL);
5038 raw_spin_unlock(&kvm_count_lock);
5042 static void hardware_disable_all_nolock(void)
5044 BUG_ON(!kvm_usage_count);
5047 if (!kvm_usage_count)
5048 on_each_cpu(hardware_disable_nolock, NULL, 1);
5051 static void hardware_disable_all(void)
5053 raw_spin_lock(&kvm_count_lock);
5054 hardware_disable_all_nolock();
5055 raw_spin_unlock(&kvm_count_lock);
5058 static int hardware_enable_all(void)
5062 raw_spin_lock(&kvm_count_lock);
5065 if (kvm_usage_count == 1) {
5066 atomic_set(&hardware_enable_failed, 0);
5067 on_each_cpu(hardware_enable_nolock, NULL, 1);
5069 if (atomic_read(&hardware_enable_failed)) {
5070 hardware_disable_all_nolock();
5075 raw_spin_unlock(&kvm_count_lock);
5080 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
5084 * Some (well, at least mine) BIOSes hang on reboot if
5087 * And Intel TXT required VMX off for all cpu when system shutdown.
5089 pr_info("kvm: exiting hardware virtualization\n");
5090 kvm_rebooting = true;
5091 on_each_cpu(hardware_disable_nolock, NULL, 1);
5095 static struct notifier_block kvm_reboot_notifier = {
5096 .notifier_call = kvm_reboot,
5100 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
5104 for (i = 0; i < bus->dev_count; i++) {
5105 struct kvm_io_device *pos = bus->range[i].dev;
5107 kvm_iodevice_destructor(pos);
5112 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
5113 const struct kvm_io_range *r2)
5115 gpa_t addr1 = r1->addr;
5116 gpa_t addr2 = r2->addr;
5121 /* If r2->len == 0, match the exact address. If r2->len != 0,
5122 * accept any overlapping write. Any order is acceptable for
5123 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
5124 * we process all of them.
5137 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
5139 return kvm_io_bus_cmp(p1, p2);
5142 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
5143 gpa_t addr, int len)
5145 struct kvm_io_range *range, key;
5148 key = (struct kvm_io_range) {
5153 range = bsearch(&key, bus->range, bus->dev_count,
5154 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
5158 off = range - bus->range;
5160 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
5166 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
5167 struct kvm_io_range *range, const void *val)
5171 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
5175 while (idx < bus->dev_count &&
5176 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
5177 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
5186 /* kvm_io_bus_write - called under kvm->slots_lock */
5187 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
5188 int len, const void *val)
5190 struct kvm_io_bus *bus;
5191 struct kvm_io_range range;
5194 range = (struct kvm_io_range) {
5199 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
5202 r = __kvm_io_bus_write(vcpu, bus, &range, val);
5203 return r < 0 ? r : 0;
5205 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
5207 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
5208 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
5209 gpa_t addr, int len, const void *val, long cookie)
5211 struct kvm_io_bus *bus;
5212 struct kvm_io_range range;
5214 range = (struct kvm_io_range) {
5219 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
5223 /* First try the device referenced by cookie. */
5224 if ((cookie >= 0) && (cookie < bus->dev_count) &&
5225 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
5226 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
5231 * cookie contained garbage; fall back to search and return the
5232 * correct cookie value.
5234 return __kvm_io_bus_write(vcpu, bus, &range, val);
5237 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
5238 struct kvm_io_range *range, void *val)
5242 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
5246 while (idx < bus->dev_count &&
5247 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
5248 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
5257 /* kvm_io_bus_read - called under kvm->slots_lock */
5258 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
5261 struct kvm_io_bus *bus;
5262 struct kvm_io_range range;
5265 range = (struct kvm_io_range) {
5270 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
5273 r = __kvm_io_bus_read(vcpu, bus, &range, val);
5274 return r < 0 ? r : 0;
5277 /* Caller must hold slots_lock. */
5278 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
5279 int len, struct kvm_io_device *dev)
5282 struct kvm_io_bus *new_bus, *bus;
5283 struct kvm_io_range range;
5285 bus = kvm_get_bus(kvm, bus_idx);
5289 /* exclude ioeventfd which is limited by maximum fd */
5290 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
5293 new_bus = kmalloc(struct_size(bus, range, bus->dev_count + 1),
5294 GFP_KERNEL_ACCOUNT);
5298 range = (struct kvm_io_range) {
5304 for (i = 0; i < bus->dev_count; i++)
5305 if (kvm_io_bus_cmp(&bus->range[i], &range) > 0)
5308 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
5309 new_bus->dev_count++;
5310 new_bus->range[i] = range;
5311 memcpy(new_bus->range + i + 1, bus->range + i,
5312 (bus->dev_count - i) * sizeof(struct kvm_io_range));
5313 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
5314 synchronize_srcu_expedited(&kvm->srcu);
5320 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
5321 struct kvm_io_device *dev)
5324 struct kvm_io_bus *new_bus, *bus;
5326 lockdep_assert_held(&kvm->slots_lock);
5328 bus = kvm_get_bus(kvm, bus_idx);
5332 for (i = 0; i < bus->dev_count; i++) {
5333 if (bus->range[i].dev == dev) {
5338 if (i == bus->dev_count)
5341 new_bus = kmalloc(struct_size(bus, range, bus->dev_count - 1),
5342 GFP_KERNEL_ACCOUNT);
5344 memcpy(new_bus, bus, struct_size(bus, range, i));
5345 new_bus->dev_count--;
5346 memcpy(new_bus->range + i, bus->range + i + 1,
5347 flex_array_size(new_bus, range, new_bus->dev_count - i));
5350 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
5351 synchronize_srcu_expedited(&kvm->srcu);
5353 /* Destroy the old bus _after_ installing the (null) bus. */
5355 pr_err("kvm: failed to shrink bus, removing it completely\n");
5356 for (j = 0; j < bus->dev_count; j++) {
5359 kvm_iodevice_destructor(bus->range[j].dev);
5364 return new_bus ? 0 : -ENOMEM;
5367 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
5370 struct kvm_io_bus *bus;
5371 int dev_idx, srcu_idx;
5372 struct kvm_io_device *iodev = NULL;
5374 srcu_idx = srcu_read_lock(&kvm->srcu);
5376 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
5380 dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
5384 iodev = bus->range[dev_idx].dev;
5387 srcu_read_unlock(&kvm->srcu, srcu_idx);
5391 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
5393 static int kvm_debugfs_open(struct inode *inode, struct file *file,
5394 int (*get)(void *, u64 *), int (*set)(void *, u64),
5397 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
5401 * The debugfs files are a reference to the kvm struct which
5402 * is still valid when kvm_destroy_vm is called. kvm_get_kvm_safe
5403 * avoids the race between open and the removal of the debugfs directory.
5405 if (!kvm_get_kvm_safe(stat_data->kvm))
5408 if (simple_attr_open(inode, file, get,
5409 kvm_stats_debugfs_mode(stat_data->desc) & 0222
5412 kvm_put_kvm(stat_data->kvm);
5419 static int kvm_debugfs_release(struct inode *inode, struct file *file)
5421 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
5424 simple_attr_release(inode, file);
5425 kvm_put_kvm(stat_data->kvm);
5430 static int kvm_get_stat_per_vm(struct kvm *kvm, size_t offset, u64 *val)
5432 *val = *(u64 *)((void *)(&kvm->stat) + offset);
5437 static int kvm_clear_stat_per_vm(struct kvm *kvm, size_t offset)
5439 *(u64 *)((void *)(&kvm->stat) + offset) = 0;
5444 static int kvm_get_stat_per_vcpu(struct kvm *kvm, size_t offset, u64 *val)
5447 struct kvm_vcpu *vcpu;
5451 kvm_for_each_vcpu(i, vcpu, kvm)
5452 *val += *(u64 *)((void *)(&vcpu->stat) + offset);
5457 static int kvm_clear_stat_per_vcpu(struct kvm *kvm, size_t offset)
5460 struct kvm_vcpu *vcpu;
5462 kvm_for_each_vcpu(i, vcpu, kvm)
5463 *(u64 *)((void *)(&vcpu->stat) + offset) = 0;
5468 static int kvm_stat_data_get(void *data, u64 *val)
5471 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
5473 switch (stat_data->kind) {
5475 r = kvm_get_stat_per_vm(stat_data->kvm,
5476 stat_data->desc->desc.offset, val);
5479 r = kvm_get_stat_per_vcpu(stat_data->kvm,
5480 stat_data->desc->desc.offset, val);
5487 static int kvm_stat_data_clear(void *data, u64 val)
5490 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
5495 switch (stat_data->kind) {
5497 r = kvm_clear_stat_per_vm(stat_data->kvm,
5498 stat_data->desc->desc.offset);
5501 r = kvm_clear_stat_per_vcpu(stat_data->kvm,
5502 stat_data->desc->desc.offset);
5509 static int kvm_stat_data_open(struct inode *inode, struct file *file)
5511 __simple_attr_check_format("%llu\n", 0ull);
5512 return kvm_debugfs_open(inode, file, kvm_stat_data_get,
5513 kvm_stat_data_clear, "%llu\n");
5516 static const struct file_operations stat_fops_per_vm = {
5517 .owner = THIS_MODULE,
5518 .open = kvm_stat_data_open,
5519 .release = kvm_debugfs_release,
5520 .read = simple_attr_read,
5521 .write = simple_attr_write,
5522 .llseek = no_llseek,
5525 static int vm_stat_get(void *_offset, u64 *val)
5527 unsigned offset = (long)_offset;
5532 mutex_lock(&kvm_lock);
5533 list_for_each_entry(kvm, &vm_list, vm_list) {
5534 kvm_get_stat_per_vm(kvm, offset, &tmp_val);
5537 mutex_unlock(&kvm_lock);
5541 static int vm_stat_clear(void *_offset, u64 val)
5543 unsigned offset = (long)_offset;
5549 mutex_lock(&kvm_lock);
5550 list_for_each_entry(kvm, &vm_list, vm_list) {
5551 kvm_clear_stat_per_vm(kvm, offset);
5553 mutex_unlock(&kvm_lock);
5558 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n");
5559 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_readonly_fops, vm_stat_get, NULL, "%llu\n");
5561 static int vcpu_stat_get(void *_offset, u64 *val)
5563 unsigned offset = (long)_offset;
5568 mutex_lock(&kvm_lock);
5569 list_for_each_entry(kvm, &vm_list, vm_list) {
5570 kvm_get_stat_per_vcpu(kvm, offset, &tmp_val);
5573 mutex_unlock(&kvm_lock);
5577 static int vcpu_stat_clear(void *_offset, u64 val)
5579 unsigned offset = (long)_offset;
5585 mutex_lock(&kvm_lock);
5586 list_for_each_entry(kvm, &vm_list, vm_list) {
5587 kvm_clear_stat_per_vcpu(kvm, offset);
5589 mutex_unlock(&kvm_lock);
5594 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear,
5596 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_readonly_fops, vcpu_stat_get, NULL, "%llu\n");
5598 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm)
5600 struct kobj_uevent_env *env;
5601 unsigned long long created, active;
5603 if (!kvm_dev.this_device || !kvm)
5606 mutex_lock(&kvm_lock);
5607 if (type == KVM_EVENT_CREATE_VM) {
5608 kvm_createvm_count++;
5610 } else if (type == KVM_EVENT_DESTROY_VM) {
5613 created = kvm_createvm_count;
5614 active = kvm_active_vms;
5615 mutex_unlock(&kvm_lock);
5617 env = kzalloc(sizeof(*env), GFP_KERNEL_ACCOUNT);
5621 add_uevent_var(env, "CREATED=%llu", created);
5622 add_uevent_var(env, "COUNT=%llu", active);
5624 if (type == KVM_EVENT_CREATE_VM) {
5625 add_uevent_var(env, "EVENT=create");
5626 kvm->userspace_pid = task_pid_nr(current);
5627 } else if (type == KVM_EVENT_DESTROY_VM) {
5628 add_uevent_var(env, "EVENT=destroy");
5630 add_uevent_var(env, "PID=%d", kvm->userspace_pid);
5632 if (!IS_ERR(kvm->debugfs_dentry)) {
5633 char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL_ACCOUNT);
5636 tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX);
5638 add_uevent_var(env, "STATS_PATH=%s", tmp);
5642 /* no need for checks, since we are adding at most only 5 keys */
5643 env->envp[env->envp_idx++] = NULL;
5644 kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp);
5648 static void kvm_init_debug(void)
5650 const struct file_operations *fops;
5651 const struct _kvm_stats_desc *pdesc;
5654 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
5656 for (i = 0; i < kvm_vm_stats_header.num_desc; ++i) {
5657 pdesc = &kvm_vm_stats_desc[i];
5658 if (kvm_stats_debugfs_mode(pdesc) & 0222)
5659 fops = &vm_stat_fops;
5661 fops = &vm_stat_readonly_fops;
5662 debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
5664 (void *)(long)pdesc->desc.offset, fops);
5667 for (i = 0; i < kvm_vcpu_stats_header.num_desc; ++i) {
5668 pdesc = &kvm_vcpu_stats_desc[i];
5669 if (kvm_stats_debugfs_mode(pdesc) & 0222)
5670 fops = &vcpu_stat_fops;
5672 fops = &vcpu_stat_readonly_fops;
5673 debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
5675 (void *)(long)pdesc->desc.offset, fops);
5679 static int kvm_suspend(void)
5681 if (kvm_usage_count)
5682 hardware_disable_nolock(NULL);
5686 static void kvm_resume(void)
5688 if (kvm_usage_count) {
5689 lockdep_assert_not_held(&kvm_count_lock);
5690 hardware_enable_nolock(NULL);
5694 static struct syscore_ops kvm_syscore_ops = {
5695 .suspend = kvm_suspend,
5696 .resume = kvm_resume,
5700 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
5702 return container_of(pn, struct kvm_vcpu, preempt_notifier);
5705 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
5707 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
5709 WRITE_ONCE(vcpu->preempted, false);
5710 WRITE_ONCE(vcpu->ready, false);
5712 __this_cpu_write(kvm_running_vcpu, vcpu);
5713 kvm_arch_sched_in(vcpu, cpu);
5714 kvm_arch_vcpu_load(vcpu, cpu);
5717 static void kvm_sched_out(struct preempt_notifier *pn,
5718 struct task_struct *next)
5720 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
5722 if (current->on_rq) {
5723 WRITE_ONCE(vcpu->preempted, true);
5724 WRITE_ONCE(vcpu->ready, true);
5726 kvm_arch_vcpu_put(vcpu);
5727 __this_cpu_write(kvm_running_vcpu, NULL);
5731 * kvm_get_running_vcpu - get the vcpu running on the current CPU.
5733 * We can disable preemption locally around accessing the per-CPU variable,
5734 * and use the resolved vcpu pointer after enabling preemption again,
5735 * because even if the current thread is migrated to another CPU, reading
5736 * the per-CPU value later will give us the same value as we update the
5737 * per-CPU variable in the preempt notifier handlers.
5739 struct kvm_vcpu *kvm_get_running_vcpu(void)
5741 struct kvm_vcpu *vcpu;
5744 vcpu = __this_cpu_read(kvm_running_vcpu);
5749 EXPORT_SYMBOL_GPL(kvm_get_running_vcpu);
5752 * kvm_get_running_vcpus - get the per-CPU array of currently running vcpus.
5754 struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void)
5756 return &kvm_running_vcpu;
5759 #ifdef CONFIG_GUEST_PERF_EVENTS
5760 static unsigned int kvm_guest_state(void)
5762 struct kvm_vcpu *vcpu = kvm_get_running_vcpu();
5765 if (!kvm_arch_pmi_in_guest(vcpu))
5768 state = PERF_GUEST_ACTIVE;
5769 if (!kvm_arch_vcpu_in_kernel(vcpu))
5770 state |= PERF_GUEST_USER;
5775 static unsigned long kvm_guest_get_ip(void)
5777 struct kvm_vcpu *vcpu = kvm_get_running_vcpu();
5779 /* Retrieving the IP must be guarded by a call to kvm_guest_state(). */
5780 if (WARN_ON_ONCE(!kvm_arch_pmi_in_guest(vcpu)))
5783 return kvm_arch_vcpu_get_ip(vcpu);
5786 static struct perf_guest_info_callbacks kvm_guest_cbs = {
5787 .state = kvm_guest_state,
5788 .get_ip = kvm_guest_get_ip,
5789 .handle_intel_pt_intr = NULL,
5792 void kvm_register_perf_callbacks(unsigned int (*pt_intr_handler)(void))
5794 kvm_guest_cbs.handle_intel_pt_intr = pt_intr_handler;
5795 perf_register_guest_info_callbacks(&kvm_guest_cbs);
5797 void kvm_unregister_perf_callbacks(void)
5799 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
5803 struct kvm_cpu_compat_check {
5808 static void check_processor_compat(void *data)
5810 struct kvm_cpu_compat_check *c = data;
5812 *c->ret = kvm_arch_check_processor_compat(c->opaque);
5815 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
5816 struct module *module)
5818 struct kvm_cpu_compat_check c;
5822 r = kvm_arch_init(opaque);
5827 * kvm_arch_init makes sure there's at most one caller
5828 * for architectures that support multiple implementations,
5829 * like intel and amd on x86.
5830 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
5831 * conflicts in case kvm is already setup for another implementation.
5833 r = kvm_irqfd_init();
5837 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
5842 r = kvm_arch_hardware_setup(opaque);
5848 for_each_online_cpu(cpu) {
5849 smp_call_function_single(cpu, check_processor_compat, &c, 1);
5854 r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting",
5855 kvm_starting_cpu, kvm_dying_cpu);
5858 register_reboot_notifier(&kvm_reboot_notifier);
5860 /* A kmem cache lets us meet the alignment requirements of fx_save. */
5862 vcpu_align = __alignof__(struct kvm_vcpu);
5864 kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align,
5866 offsetof(struct kvm_vcpu, arch),
5867 offsetofend(struct kvm_vcpu, stats_id)
5868 - offsetof(struct kvm_vcpu, arch),
5870 if (!kvm_vcpu_cache) {
5875 for_each_possible_cpu(cpu) {
5876 if (!alloc_cpumask_var_node(&per_cpu(cpu_kick_mask, cpu),
5877 GFP_KERNEL, cpu_to_node(cpu))) {
5883 r = kvm_async_pf_init();
5887 kvm_chardev_ops.owner = module;
5889 r = misc_register(&kvm_dev);
5891 pr_err("kvm: misc device register failed\n");
5895 register_syscore_ops(&kvm_syscore_ops);
5897 kvm_preempt_ops.sched_in = kvm_sched_in;
5898 kvm_preempt_ops.sched_out = kvm_sched_out;
5902 r = kvm_vfio_ops_init();
5908 kvm_async_pf_deinit();
5910 for_each_possible_cpu(cpu)
5911 free_cpumask_var(per_cpu(cpu_kick_mask, cpu));
5913 kmem_cache_destroy(kvm_vcpu_cache);
5915 unregister_reboot_notifier(&kvm_reboot_notifier);
5916 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
5918 kvm_arch_hardware_unsetup();
5920 free_cpumask_var(cpus_hardware_enabled);
5928 EXPORT_SYMBOL_GPL(kvm_init);
5934 debugfs_remove_recursive(kvm_debugfs_dir);
5935 misc_deregister(&kvm_dev);
5936 for_each_possible_cpu(cpu)
5937 free_cpumask_var(per_cpu(cpu_kick_mask, cpu));
5938 kmem_cache_destroy(kvm_vcpu_cache);
5939 kvm_async_pf_deinit();
5940 unregister_syscore_ops(&kvm_syscore_ops);
5941 unregister_reboot_notifier(&kvm_reboot_notifier);
5942 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
5943 on_each_cpu(hardware_disable_nolock, NULL, 1);
5944 kvm_arch_hardware_unsetup();
5947 free_cpumask_var(cpus_hardware_enabled);
5948 kvm_vfio_ops_exit();
5950 EXPORT_SYMBOL_GPL(kvm_exit);
5952 struct kvm_vm_worker_thread_context {
5954 struct task_struct *parent;
5955 struct completion init_done;
5956 kvm_vm_thread_fn_t thread_fn;
5961 static int kvm_vm_worker_thread(void *context)
5964 * The init_context is allocated on the stack of the parent thread, so
5965 * we have to locally copy anything that is needed beyond initialization
5967 struct kvm_vm_worker_thread_context *init_context = context;
5968 struct task_struct *parent;
5969 struct kvm *kvm = init_context->kvm;
5970 kvm_vm_thread_fn_t thread_fn = init_context->thread_fn;
5971 uintptr_t data = init_context->data;
5974 err = kthread_park(current);
5975 /* kthread_park(current) is never supposed to return an error */
5980 err = cgroup_attach_task_all(init_context->parent, current);
5982 kvm_err("%s: cgroup_attach_task_all failed with err %d\n",
5987 set_user_nice(current, task_nice(init_context->parent));
5990 init_context->err = err;
5991 complete(&init_context->init_done);
5992 init_context = NULL;
5997 /* Wait to be woken up by the spawner before proceeding. */
6000 if (!kthread_should_stop())
6001 err = thread_fn(kvm, data);
6005 * Move kthread back to its original cgroup to prevent it lingering in
6006 * the cgroup of the VM process, after the latter finishes its
6009 * kthread_stop() waits on the 'exited' completion condition which is
6010 * set in exit_mm(), via mm_release(), in do_exit(). However, the
6011 * kthread is removed from the cgroup in the cgroup_exit() which is
6012 * called after the exit_mm(). This causes the kthread_stop() to return
6013 * before the kthread actually quits the cgroup.
6016 parent = rcu_dereference(current->real_parent);
6017 get_task_struct(parent);
6019 cgroup_attach_task_all(parent, current);
6020 put_task_struct(parent);
6025 int kvm_vm_create_worker_thread(struct kvm *kvm, kvm_vm_thread_fn_t thread_fn,
6026 uintptr_t data, const char *name,
6027 struct task_struct **thread_ptr)
6029 struct kvm_vm_worker_thread_context init_context = {};
6030 struct task_struct *thread;
6033 init_context.kvm = kvm;
6034 init_context.parent = current;
6035 init_context.thread_fn = thread_fn;
6036 init_context.data = data;
6037 init_completion(&init_context.init_done);
6039 thread = kthread_run(kvm_vm_worker_thread, &init_context,
6040 "%s-%d", name, task_pid_nr(current));
6042 return PTR_ERR(thread);
6044 /* kthread_run is never supposed to return NULL */
6045 WARN_ON(thread == NULL);
6047 wait_for_completion(&init_context.init_done);
6049 if (!init_context.err)
6050 *thread_ptr = thread;
6052 return init_context.err;