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;
1031 int i, ret = -ENOMEM;
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);
1095 kvm_destroy_vm_debugfs(kvm);
1100 * Called after the VM is otherwise initialized, but just before adding it to
1103 int __weak kvm_arch_post_init_vm(struct kvm *kvm)
1109 * Called just after removing the VM from the vm_list, but before doing any
1110 * other destruction.
1112 void __weak kvm_arch_pre_destroy_vm(struct kvm *kvm)
1117 * Called after per-vm debugfs created. When called kvm->debugfs_dentry should
1118 * be setup already, so we can create arch-specific debugfs entries under it.
1119 * Cleanup should be automatic done in kvm_destroy_vm_debugfs() recursively, so
1120 * a per-arch destroy interface is not needed.
1122 int __weak kvm_arch_create_vm_debugfs(struct kvm *kvm)
1127 static struct kvm *kvm_create_vm(unsigned long type, const char *fdname)
1129 struct kvm *kvm = kvm_arch_alloc_vm();
1130 struct kvm_memslots *slots;
1135 return ERR_PTR(-ENOMEM);
1137 KVM_MMU_LOCK_INIT(kvm);
1138 mmgrab(current->mm);
1139 kvm->mm = current->mm;
1140 kvm_eventfd_init(kvm);
1141 mutex_init(&kvm->lock);
1142 mutex_init(&kvm->irq_lock);
1143 mutex_init(&kvm->slots_lock);
1144 mutex_init(&kvm->slots_arch_lock);
1145 spin_lock_init(&kvm->mn_invalidate_lock);
1146 rcuwait_init(&kvm->mn_memslots_update_rcuwait);
1147 xa_init(&kvm->vcpu_array);
1149 INIT_LIST_HEAD(&kvm->gpc_list);
1150 spin_lock_init(&kvm->gpc_lock);
1152 INIT_LIST_HEAD(&kvm->devices);
1153 kvm->max_vcpus = KVM_MAX_VCPUS;
1155 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
1158 * Force subsequent debugfs file creations to fail if the VM directory
1159 * is not created (by kvm_create_vm_debugfs()).
1161 kvm->debugfs_dentry = ERR_PTR(-ENOENT);
1163 snprintf(kvm->stats_id, sizeof(kvm->stats_id), "kvm-%d",
1164 task_pid_nr(current));
1166 if (init_srcu_struct(&kvm->srcu))
1167 goto out_err_no_srcu;
1168 if (init_srcu_struct(&kvm->irq_srcu))
1169 goto out_err_no_irq_srcu;
1171 refcount_set(&kvm->users_count, 1);
1172 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
1173 for (j = 0; j < 2; j++) {
1174 slots = &kvm->__memslots[i][j];
1176 atomic_long_set(&slots->last_used_slot, (unsigned long)NULL);
1177 slots->hva_tree = RB_ROOT_CACHED;
1178 slots->gfn_tree = RB_ROOT;
1179 hash_init(slots->id_hash);
1180 slots->node_idx = j;
1182 /* Generations must be different for each address space. */
1183 slots->generation = i;
1186 rcu_assign_pointer(kvm->memslots[i], &kvm->__memslots[i][0]);
1189 for (i = 0; i < KVM_NR_BUSES; i++) {
1190 rcu_assign_pointer(kvm->buses[i],
1191 kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL_ACCOUNT));
1193 goto out_err_no_arch_destroy_vm;
1196 kvm->max_halt_poll_ns = halt_poll_ns;
1198 r = kvm_arch_init_vm(kvm, type);
1200 goto out_err_no_arch_destroy_vm;
1202 r = hardware_enable_all();
1204 goto out_err_no_disable;
1206 #ifdef CONFIG_HAVE_KVM_IRQFD
1207 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
1210 r = kvm_init_mmu_notifier(kvm);
1212 goto out_err_no_mmu_notifier;
1214 r = kvm_arch_post_init_vm(kvm);
1216 goto out_err_mmu_notifier;
1218 mutex_lock(&kvm_lock);
1219 list_add(&kvm->vm_list, &vm_list);
1220 mutex_unlock(&kvm_lock);
1222 preempt_notifier_inc();
1223 kvm_init_pm_notifier(kvm);
1226 * When the fd passed to this ioctl() is opened it pins the module,
1227 * but try_module_get() also prevents getting a reference if the module
1228 * is in MODULE_STATE_GOING (e.g. if someone ran "rmmod --wait").
1230 if (!try_module_get(kvm_chardev_ops.owner)) {
1232 goto out_err_mmu_notifier;
1235 r = kvm_create_vm_debugfs(kvm, fdname);
1242 module_put(kvm_chardev_ops.owner);
1243 out_err_mmu_notifier:
1244 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1245 if (kvm->mmu_notifier.ops)
1246 mmu_notifier_unregister(&kvm->mmu_notifier, current->mm);
1248 out_err_no_mmu_notifier:
1249 hardware_disable_all();
1251 kvm_arch_destroy_vm(kvm);
1252 out_err_no_arch_destroy_vm:
1253 WARN_ON_ONCE(!refcount_dec_and_test(&kvm->users_count));
1254 for (i = 0; i < KVM_NR_BUSES; i++)
1255 kfree(kvm_get_bus(kvm, i));
1256 cleanup_srcu_struct(&kvm->irq_srcu);
1257 out_err_no_irq_srcu:
1258 cleanup_srcu_struct(&kvm->srcu);
1260 kvm_arch_free_vm(kvm);
1261 mmdrop(current->mm);
1265 static void kvm_destroy_devices(struct kvm *kvm)
1267 struct kvm_device *dev, *tmp;
1270 * We do not need to take the kvm->lock here, because nobody else
1271 * has a reference to the struct kvm at this point and therefore
1272 * cannot access the devices list anyhow.
1274 list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
1275 list_del(&dev->vm_node);
1276 dev->ops->destroy(dev);
1280 static void kvm_destroy_vm(struct kvm *kvm)
1283 struct mm_struct *mm = kvm->mm;
1285 kvm_destroy_pm_notifier(kvm);
1286 kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm);
1287 kvm_destroy_vm_debugfs(kvm);
1288 kvm_arch_sync_events(kvm);
1289 mutex_lock(&kvm_lock);
1290 list_del(&kvm->vm_list);
1291 mutex_unlock(&kvm_lock);
1292 kvm_arch_pre_destroy_vm(kvm);
1294 kvm_free_irq_routing(kvm);
1295 for (i = 0; i < KVM_NR_BUSES; i++) {
1296 struct kvm_io_bus *bus = kvm_get_bus(kvm, i);
1299 kvm_io_bus_destroy(bus);
1300 kvm->buses[i] = NULL;
1302 kvm_coalesced_mmio_free(kvm);
1303 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1304 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
1306 * At this point, pending calls to invalidate_range_start()
1307 * have completed but no more MMU notifiers will run, so
1308 * mn_active_invalidate_count may remain unbalanced.
1309 * No threads can be waiting in install_new_memslots as the
1310 * last reference on KVM has been dropped, but freeing
1311 * memslots would deadlock without this manual intervention.
1313 WARN_ON(rcuwait_active(&kvm->mn_memslots_update_rcuwait));
1314 kvm->mn_active_invalidate_count = 0;
1316 kvm_flush_shadow_all(kvm);
1318 kvm_arch_destroy_vm(kvm);
1319 kvm_destroy_devices(kvm);
1320 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
1321 kvm_free_memslots(kvm, &kvm->__memslots[i][0]);
1322 kvm_free_memslots(kvm, &kvm->__memslots[i][1]);
1324 cleanup_srcu_struct(&kvm->irq_srcu);
1325 cleanup_srcu_struct(&kvm->srcu);
1326 kvm_arch_free_vm(kvm);
1327 preempt_notifier_dec();
1328 hardware_disable_all();
1330 module_put(kvm_chardev_ops.owner);
1333 void kvm_get_kvm(struct kvm *kvm)
1335 refcount_inc(&kvm->users_count);
1337 EXPORT_SYMBOL_GPL(kvm_get_kvm);
1340 * Make sure the vm is not during destruction, which is a safe version of
1341 * kvm_get_kvm(). Return true if kvm referenced successfully, false otherwise.
1343 bool kvm_get_kvm_safe(struct kvm *kvm)
1345 return refcount_inc_not_zero(&kvm->users_count);
1347 EXPORT_SYMBOL_GPL(kvm_get_kvm_safe);
1349 void kvm_put_kvm(struct kvm *kvm)
1351 if (refcount_dec_and_test(&kvm->users_count))
1352 kvm_destroy_vm(kvm);
1354 EXPORT_SYMBOL_GPL(kvm_put_kvm);
1357 * Used to put a reference that was taken on behalf of an object associated
1358 * with a user-visible file descriptor, e.g. a vcpu or device, if installation
1359 * of the new file descriptor fails and the reference cannot be transferred to
1360 * its final owner. In such cases, the caller is still actively using @kvm and
1361 * will fail miserably if the refcount unexpectedly hits zero.
1363 void kvm_put_kvm_no_destroy(struct kvm *kvm)
1365 WARN_ON(refcount_dec_and_test(&kvm->users_count));
1367 EXPORT_SYMBOL_GPL(kvm_put_kvm_no_destroy);
1369 static int kvm_vm_release(struct inode *inode, struct file *filp)
1371 struct kvm *kvm = filp->private_data;
1373 kvm_irqfd_release(kvm);
1380 * Allocation size is twice as large as the actual dirty bitmap size.
1381 * See kvm_vm_ioctl_get_dirty_log() why this is needed.
1383 static int kvm_alloc_dirty_bitmap(struct kvm_memory_slot *memslot)
1385 unsigned long dirty_bytes = kvm_dirty_bitmap_bytes(memslot);
1387 memslot->dirty_bitmap = __vcalloc(2, dirty_bytes, GFP_KERNEL_ACCOUNT);
1388 if (!memslot->dirty_bitmap)
1394 static struct kvm_memslots *kvm_get_inactive_memslots(struct kvm *kvm, int as_id)
1396 struct kvm_memslots *active = __kvm_memslots(kvm, as_id);
1397 int node_idx_inactive = active->node_idx ^ 1;
1399 return &kvm->__memslots[as_id][node_idx_inactive];
1403 * Helper to get the address space ID when one of memslot pointers may be NULL.
1404 * This also serves as a sanity that at least one of the pointers is non-NULL,
1405 * and that their address space IDs don't diverge.
1407 static int kvm_memslots_get_as_id(struct kvm_memory_slot *a,
1408 struct kvm_memory_slot *b)
1410 if (WARN_ON_ONCE(!a && !b))
1418 WARN_ON_ONCE(a->as_id != b->as_id);
1422 static void kvm_insert_gfn_node(struct kvm_memslots *slots,
1423 struct kvm_memory_slot *slot)
1425 struct rb_root *gfn_tree = &slots->gfn_tree;
1426 struct rb_node **node, *parent;
1427 int idx = slots->node_idx;
1430 for (node = &gfn_tree->rb_node; *node; ) {
1431 struct kvm_memory_slot *tmp;
1433 tmp = container_of(*node, struct kvm_memory_slot, gfn_node[idx]);
1435 if (slot->base_gfn < tmp->base_gfn)
1436 node = &(*node)->rb_left;
1437 else if (slot->base_gfn > tmp->base_gfn)
1438 node = &(*node)->rb_right;
1443 rb_link_node(&slot->gfn_node[idx], parent, node);
1444 rb_insert_color(&slot->gfn_node[idx], gfn_tree);
1447 static void kvm_erase_gfn_node(struct kvm_memslots *slots,
1448 struct kvm_memory_slot *slot)
1450 rb_erase(&slot->gfn_node[slots->node_idx], &slots->gfn_tree);
1453 static void kvm_replace_gfn_node(struct kvm_memslots *slots,
1454 struct kvm_memory_slot *old,
1455 struct kvm_memory_slot *new)
1457 int idx = slots->node_idx;
1459 WARN_ON_ONCE(old->base_gfn != new->base_gfn);
1461 rb_replace_node(&old->gfn_node[idx], &new->gfn_node[idx],
1466 * Replace @old with @new in the inactive memslots.
1468 * With NULL @old this simply adds @new.
1469 * With NULL @new this simply removes @old.
1471 * If @new is non-NULL its hva_node[slots_idx] range has to be set
1474 static void kvm_replace_memslot(struct kvm *kvm,
1475 struct kvm_memory_slot *old,
1476 struct kvm_memory_slot *new)
1478 int as_id = kvm_memslots_get_as_id(old, new);
1479 struct kvm_memslots *slots = kvm_get_inactive_memslots(kvm, as_id);
1480 int idx = slots->node_idx;
1483 hash_del(&old->id_node[idx]);
1484 interval_tree_remove(&old->hva_node[idx], &slots->hva_tree);
1486 if ((long)old == atomic_long_read(&slots->last_used_slot))
1487 atomic_long_set(&slots->last_used_slot, (long)new);
1490 kvm_erase_gfn_node(slots, old);
1496 * Initialize @new's hva range. Do this even when replacing an @old
1497 * slot, kvm_copy_memslot() deliberately does not touch node data.
1499 new->hva_node[idx].start = new->userspace_addr;
1500 new->hva_node[idx].last = new->userspace_addr +
1501 (new->npages << PAGE_SHIFT) - 1;
1504 * (Re)Add the new memslot. There is no O(1) interval_tree_replace(),
1505 * hva_node needs to be swapped with remove+insert even though hva can't
1506 * change when replacing an existing slot.
1508 hash_add(slots->id_hash, &new->id_node[idx], new->id);
1509 interval_tree_insert(&new->hva_node[idx], &slots->hva_tree);
1512 * If the memslot gfn is unchanged, rb_replace_node() can be used to
1513 * switch the node in the gfn tree instead of removing the old and
1514 * inserting the new as two separate operations. Replacement is a
1515 * single O(1) operation versus two O(log(n)) operations for
1518 if (old && old->base_gfn == new->base_gfn) {
1519 kvm_replace_gfn_node(slots, old, new);
1522 kvm_erase_gfn_node(slots, old);
1523 kvm_insert_gfn_node(slots, new);
1527 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
1529 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
1531 #ifdef __KVM_HAVE_READONLY_MEM
1532 valid_flags |= KVM_MEM_READONLY;
1535 if (mem->flags & ~valid_flags)
1541 static void kvm_swap_active_memslots(struct kvm *kvm, int as_id)
1543 struct kvm_memslots *slots = kvm_get_inactive_memslots(kvm, as_id);
1545 /* Grab the generation from the activate memslots. */
1546 u64 gen = __kvm_memslots(kvm, as_id)->generation;
1548 WARN_ON(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS);
1549 slots->generation = gen | KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
1552 * Do not store the new memslots while there are invalidations in
1553 * progress, otherwise the locking in invalidate_range_start and
1554 * invalidate_range_end will be unbalanced.
1556 spin_lock(&kvm->mn_invalidate_lock);
1557 prepare_to_rcuwait(&kvm->mn_memslots_update_rcuwait);
1558 while (kvm->mn_active_invalidate_count) {
1559 set_current_state(TASK_UNINTERRUPTIBLE);
1560 spin_unlock(&kvm->mn_invalidate_lock);
1562 spin_lock(&kvm->mn_invalidate_lock);
1564 finish_rcuwait(&kvm->mn_memslots_update_rcuwait);
1565 rcu_assign_pointer(kvm->memslots[as_id], slots);
1566 spin_unlock(&kvm->mn_invalidate_lock);
1569 * Acquired in kvm_set_memslot. Must be released before synchronize
1570 * SRCU below in order to avoid deadlock with another thread
1571 * acquiring the slots_arch_lock in an srcu critical section.
1573 mutex_unlock(&kvm->slots_arch_lock);
1575 synchronize_srcu_expedited(&kvm->srcu);
1578 * Increment the new memslot generation a second time, dropping the
1579 * update in-progress flag and incrementing the generation based on
1580 * the number of address spaces. This provides a unique and easily
1581 * identifiable generation number while the memslots are in flux.
1583 gen = slots->generation & ~KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
1586 * Generations must be unique even across address spaces. We do not need
1587 * a global counter for that, instead the generation space is evenly split
1588 * across address spaces. For example, with two address spaces, address
1589 * space 0 will use generations 0, 2, 4, ... while address space 1 will
1590 * use generations 1, 3, 5, ...
1592 gen += KVM_ADDRESS_SPACE_NUM;
1594 kvm_arch_memslots_updated(kvm, gen);
1596 slots->generation = gen;
1599 static int kvm_prepare_memory_region(struct kvm *kvm,
1600 const struct kvm_memory_slot *old,
1601 struct kvm_memory_slot *new,
1602 enum kvm_mr_change change)
1607 * If dirty logging is disabled, nullify the bitmap; the old bitmap
1608 * will be freed on "commit". If logging is enabled in both old and
1609 * new, reuse the existing bitmap. If logging is enabled only in the
1610 * new and KVM isn't using a ring buffer, allocate and initialize a
1613 if (change != KVM_MR_DELETE) {
1614 if (!(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
1615 new->dirty_bitmap = NULL;
1616 else if (old && old->dirty_bitmap)
1617 new->dirty_bitmap = old->dirty_bitmap;
1618 else if (!kvm->dirty_ring_size) {
1619 r = kvm_alloc_dirty_bitmap(new);
1623 if (kvm_dirty_log_manual_protect_and_init_set(kvm))
1624 bitmap_set(new->dirty_bitmap, 0, new->npages);
1628 r = kvm_arch_prepare_memory_region(kvm, old, new, change);
1630 /* Free the bitmap on failure if it was allocated above. */
1631 if (r && new && new->dirty_bitmap && (!old || !old->dirty_bitmap))
1632 kvm_destroy_dirty_bitmap(new);
1637 static void kvm_commit_memory_region(struct kvm *kvm,
1638 struct kvm_memory_slot *old,
1639 const struct kvm_memory_slot *new,
1640 enum kvm_mr_change change)
1643 * Update the total number of memslot pages before calling the arch
1644 * hook so that architectures can consume the result directly.
1646 if (change == KVM_MR_DELETE)
1647 kvm->nr_memslot_pages -= old->npages;
1648 else if (change == KVM_MR_CREATE)
1649 kvm->nr_memslot_pages += new->npages;
1651 kvm_arch_commit_memory_region(kvm, old, new, change);
1655 /* Nothing more to do. */
1658 /* Free the old memslot and all its metadata. */
1659 kvm_free_memslot(kvm, old);
1662 case KVM_MR_FLAGS_ONLY:
1664 * Free the dirty bitmap as needed; the below check encompasses
1665 * both the flags and whether a ring buffer is being used)
1667 if (old->dirty_bitmap && !new->dirty_bitmap)
1668 kvm_destroy_dirty_bitmap(old);
1671 * The final quirk. Free the detached, old slot, but only its
1672 * memory, not any metadata. Metadata, including arch specific
1673 * data, may be reused by @new.
1683 * Activate @new, which must be installed in the inactive slots by the caller,
1684 * by swapping the active slots and then propagating @new to @old once @old is
1685 * unreachable and can be safely modified.
1687 * With NULL @old this simply adds @new to @active (while swapping the sets).
1688 * With NULL @new this simply removes @old from @active and frees it
1689 * (while also swapping the sets).
1691 static void kvm_activate_memslot(struct kvm *kvm,
1692 struct kvm_memory_slot *old,
1693 struct kvm_memory_slot *new)
1695 int as_id = kvm_memslots_get_as_id(old, new);
1697 kvm_swap_active_memslots(kvm, as_id);
1699 /* Propagate the new memslot to the now inactive memslots. */
1700 kvm_replace_memslot(kvm, old, new);
1703 static void kvm_copy_memslot(struct kvm_memory_slot *dest,
1704 const struct kvm_memory_slot *src)
1706 dest->base_gfn = src->base_gfn;
1707 dest->npages = src->npages;
1708 dest->dirty_bitmap = src->dirty_bitmap;
1709 dest->arch = src->arch;
1710 dest->userspace_addr = src->userspace_addr;
1711 dest->flags = src->flags;
1713 dest->as_id = src->as_id;
1716 static void kvm_invalidate_memslot(struct kvm *kvm,
1717 struct kvm_memory_slot *old,
1718 struct kvm_memory_slot *invalid_slot)
1721 * Mark the current slot INVALID. As with all memslot modifications,
1722 * this must be done on an unreachable slot to avoid modifying the
1723 * current slot in the active tree.
1725 kvm_copy_memslot(invalid_slot, old);
1726 invalid_slot->flags |= KVM_MEMSLOT_INVALID;
1727 kvm_replace_memslot(kvm, old, invalid_slot);
1730 * Activate the slot that is now marked INVALID, but don't propagate
1731 * the slot to the now inactive slots. The slot is either going to be
1732 * deleted or recreated as a new slot.
1734 kvm_swap_active_memslots(kvm, old->as_id);
1737 * From this point no new shadow pages pointing to a deleted, or moved,
1738 * memslot will be created. Validation of sp->gfn happens in:
1739 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1740 * - kvm_is_visible_gfn (mmu_check_root)
1742 kvm_arch_flush_shadow_memslot(kvm, old);
1743 kvm_arch_guest_memory_reclaimed(kvm);
1745 /* Was released by kvm_swap_active_memslots, reacquire. */
1746 mutex_lock(&kvm->slots_arch_lock);
1749 * Copy the arch-specific field of the newly-installed slot back to the
1750 * old slot as the arch data could have changed between releasing
1751 * slots_arch_lock in install_new_memslots() and re-acquiring the lock
1752 * above. Writers are required to retrieve memslots *after* acquiring
1753 * slots_arch_lock, thus the active slot's data is guaranteed to be fresh.
1755 old->arch = invalid_slot->arch;
1758 static void kvm_create_memslot(struct kvm *kvm,
1759 struct kvm_memory_slot *new)
1761 /* Add the new memslot to the inactive set and activate. */
1762 kvm_replace_memslot(kvm, NULL, new);
1763 kvm_activate_memslot(kvm, NULL, new);
1766 static void kvm_delete_memslot(struct kvm *kvm,
1767 struct kvm_memory_slot *old,
1768 struct kvm_memory_slot *invalid_slot)
1771 * Remove the old memslot (in the inactive memslots) by passing NULL as
1772 * the "new" slot, and for the invalid version in the active slots.
1774 kvm_replace_memslot(kvm, old, NULL);
1775 kvm_activate_memslot(kvm, invalid_slot, NULL);
1778 static void kvm_move_memslot(struct kvm *kvm,
1779 struct kvm_memory_slot *old,
1780 struct kvm_memory_slot *new,
1781 struct kvm_memory_slot *invalid_slot)
1784 * Replace the old memslot in the inactive slots, and then swap slots
1785 * and replace the current INVALID with the new as well.
1787 kvm_replace_memslot(kvm, old, new);
1788 kvm_activate_memslot(kvm, invalid_slot, new);
1791 static void kvm_update_flags_memslot(struct kvm *kvm,
1792 struct kvm_memory_slot *old,
1793 struct kvm_memory_slot *new)
1796 * Similar to the MOVE case, but the slot doesn't need to be zapped as
1797 * an intermediate step. Instead, the old memslot is simply replaced
1798 * with a new, updated copy in both memslot sets.
1800 kvm_replace_memslot(kvm, old, new);
1801 kvm_activate_memslot(kvm, old, new);
1804 static int kvm_set_memslot(struct kvm *kvm,
1805 struct kvm_memory_slot *old,
1806 struct kvm_memory_slot *new,
1807 enum kvm_mr_change change)
1809 struct kvm_memory_slot *invalid_slot;
1813 * Released in kvm_swap_active_memslots.
1815 * Must be held from before the current memslots are copied until
1816 * after the new memslots are installed with rcu_assign_pointer,
1817 * then released before the synchronize srcu in kvm_swap_active_memslots.
1819 * When modifying memslots outside of the slots_lock, must be held
1820 * before reading the pointer to the current memslots until after all
1821 * changes to those memslots are complete.
1823 * These rules ensure that installing new memslots does not lose
1824 * changes made to the previous memslots.
1826 mutex_lock(&kvm->slots_arch_lock);
1829 * Invalidate the old slot if it's being deleted or moved. This is
1830 * done prior to actually deleting/moving the memslot to allow vCPUs to
1831 * continue running by ensuring there are no mappings or shadow pages
1832 * for the memslot when it is deleted/moved. Without pre-invalidation
1833 * (and without a lock), a window would exist between effecting the
1834 * delete/move and committing the changes in arch code where KVM or a
1835 * guest could access a non-existent memslot.
1837 * Modifications are done on a temporary, unreachable slot. The old
1838 * slot needs to be preserved in case a later step fails and the
1839 * invalidation needs to be reverted.
1841 if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) {
1842 invalid_slot = kzalloc(sizeof(*invalid_slot), GFP_KERNEL_ACCOUNT);
1843 if (!invalid_slot) {
1844 mutex_unlock(&kvm->slots_arch_lock);
1847 kvm_invalidate_memslot(kvm, old, invalid_slot);
1850 r = kvm_prepare_memory_region(kvm, old, new, change);
1853 * For DELETE/MOVE, revert the above INVALID change. No
1854 * modifications required since the original slot was preserved
1855 * in the inactive slots. Changing the active memslots also
1856 * release slots_arch_lock.
1858 if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) {
1859 kvm_activate_memslot(kvm, invalid_slot, old);
1860 kfree(invalid_slot);
1862 mutex_unlock(&kvm->slots_arch_lock);
1868 * For DELETE and MOVE, the working slot is now active as the INVALID
1869 * version of the old slot. MOVE is particularly special as it reuses
1870 * the old slot and returns a copy of the old slot (in working_slot).
1871 * For CREATE, there is no old slot. For DELETE and FLAGS_ONLY, the
1872 * old slot is detached but otherwise preserved.
1874 if (change == KVM_MR_CREATE)
1875 kvm_create_memslot(kvm, new);
1876 else if (change == KVM_MR_DELETE)
1877 kvm_delete_memslot(kvm, old, invalid_slot);
1878 else if (change == KVM_MR_MOVE)
1879 kvm_move_memslot(kvm, old, new, invalid_slot);
1880 else if (change == KVM_MR_FLAGS_ONLY)
1881 kvm_update_flags_memslot(kvm, old, new);
1885 /* Free the temporary INVALID slot used for DELETE and MOVE. */
1886 if (change == KVM_MR_DELETE || change == KVM_MR_MOVE)
1887 kfree(invalid_slot);
1890 * No need to refresh new->arch, changes after dropping slots_arch_lock
1891 * will directly hit the final, active memslot. Architectures are
1892 * responsible for knowing that new->arch may be stale.
1894 kvm_commit_memory_region(kvm, old, new, change);
1899 static bool kvm_check_memslot_overlap(struct kvm_memslots *slots, int id,
1900 gfn_t start, gfn_t end)
1902 struct kvm_memslot_iter iter;
1904 kvm_for_each_memslot_in_gfn_range(&iter, slots, start, end) {
1905 if (iter.slot->id != id)
1913 * Allocate some memory and give it an address in the guest physical address
1916 * Discontiguous memory is allowed, mostly for framebuffers.
1918 * Must be called holding kvm->slots_lock for write.
1920 int __kvm_set_memory_region(struct kvm *kvm,
1921 const struct kvm_userspace_memory_region *mem)
1923 struct kvm_memory_slot *old, *new;
1924 struct kvm_memslots *slots;
1925 enum kvm_mr_change change;
1926 unsigned long npages;
1931 r = check_memory_region_flags(mem);
1935 as_id = mem->slot >> 16;
1936 id = (u16)mem->slot;
1938 /* General sanity checks */
1939 if ((mem->memory_size & (PAGE_SIZE - 1)) ||
1940 (mem->memory_size != (unsigned long)mem->memory_size))
1942 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
1944 /* We can read the guest memory with __xxx_user() later on. */
1945 if ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
1946 (mem->userspace_addr != untagged_addr(mem->userspace_addr)) ||
1947 !access_ok((void __user *)(unsigned long)mem->userspace_addr,
1950 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
1952 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
1954 if ((mem->memory_size >> PAGE_SHIFT) > KVM_MEM_MAX_NR_PAGES)
1957 slots = __kvm_memslots(kvm, as_id);
1960 * Note, the old memslot (and the pointer itself!) may be invalidated
1961 * and/or destroyed by kvm_set_memslot().
1963 old = id_to_memslot(slots, id);
1965 if (!mem->memory_size) {
1966 if (!old || !old->npages)
1969 if (WARN_ON_ONCE(kvm->nr_memslot_pages < old->npages))
1972 return kvm_set_memslot(kvm, old, NULL, KVM_MR_DELETE);
1975 base_gfn = (mem->guest_phys_addr >> PAGE_SHIFT);
1976 npages = (mem->memory_size >> PAGE_SHIFT);
1978 if (!old || !old->npages) {
1979 change = KVM_MR_CREATE;
1982 * To simplify KVM internals, the total number of pages across
1983 * all memslots must fit in an unsigned long.
1985 if ((kvm->nr_memslot_pages + npages) < kvm->nr_memslot_pages)
1987 } else { /* Modify an existing slot. */
1988 if ((mem->userspace_addr != old->userspace_addr) ||
1989 (npages != old->npages) ||
1990 ((mem->flags ^ old->flags) & KVM_MEM_READONLY))
1993 if (base_gfn != old->base_gfn)
1994 change = KVM_MR_MOVE;
1995 else if (mem->flags != old->flags)
1996 change = KVM_MR_FLAGS_ONLY;
1997 else /* Nothing to change. */
2001 if ((change == KVM_MR_CREATE || change == KVM_MR_MOVE) &&
2002 kvm_check_memslot_overlap(slots, id, base_gfn, base_gfn + npages))
2005 /* Allocate a slot that will persist in the memslot. */
2006 new = kzalloc(sizeof(*new), GFP_KERNEL_ACCOUNT);
2012 new->base_gfn = base_gfn;
2013 new->npages = npages;
2014 new->flags = mem->flags;
2015 new->userspace_addr = mem->userspace_addr;
2017 r = kvm_set_memslot(kvm, old, new, change);
2022 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
2024 int kvm_set_memory_region(struct kvm *kvm,
2025 const struct kvm_userspace_memory_region *mem)
2029 mutex_lock(&kvm->slots_lock);
2030 r = __kvm_set_memory_region(kvm, mem);
2031 mutex_unlock(&kvm->slots_lock);
2034 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
2036 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
2037 struct kvm_userspace_memory_region *mem)
2039 if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
2042 return kvm_set_memory_region(kvm, mem);
2045 #ifndef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
2047 * kvm_get_dirty_log - get a snapshot of dirty pages
2048 * @kvm: pointer to kvm instance
2049 * @log: slot id and address to which we copy the log
2050 * @is_dirty: set to '1' if any dirty pages were found
2051 * @memslot: set to the associated memslot, always valid on success
2053 int kvm_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log,
2054 int *is_dirty, struct kvm_memory_slot **memslot)
2056 struct kvm_memslots *slots;
2059 unsigned long any = 0;
2061 /* Dirty ring tracking is exclusive to dirty log tracking */
2062 if (kvm->dirty_ring_size)
2068 as_id = log->slot >> 16;
2069 id = (u16)log->slot;
2070 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
2073 slots = __kvm_memslots(kvm, as_id);
2074 *memslot = id_to_memslot(slots, id);
2075 if (!(*memslot) || !(*memslot)->dirty_bitmap)
2078 kvm_arch_sync_dirty_log(kvm, *memslot);
2080 n = kvm_dirty_bitmap_bytes(*memslot);
2082 for (i = 0; !any && i < n/sizeof(long); ++i)
2083 any = (*memslot)->dirty_bitmap[i];
2085 if (copy_to_user(log->dirty_bitmap, (*memslot)->dirty_bitmap, n))
2092 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
2094 #else /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
2096 * kvm_get_dirty_log_protect - get a snapshot of dirty pages
2097 * and reenable dirty page tracking for the corresponding pages.
2098 * @kvm: pointer to kvm instance
2099 * @log: slot id and address to which we copy the log
2101 * We need to keep it in mind that VCPU threads can write to the bitmap
2102 * concurrently. So, to avoid losing track of dirty pages we keep the
2105 * 1. Take a snapshot of the bit and clear it if needed.
2106 * 2. Write protect the corresponding page.
2107 * 3. Copy the snapshot to the userspace.
2108 * 4. Upon return caller flushes TLB's if needed.
2110 * Between 2 and 4, the guest may write to the page using the remaining TLB
2111 * entry. This is not a problem because the page is reported dirty using
2112 * the snapshot taken before and step 4 ensures that writes done after
2113 * exiting to userspace will be logged for the next call.
2116 static int kvm_get_dirty_log_protect(struct kvm *kvm, struct kvm_dirty_log *log)
2118 struct kvm_memslots *slots;
2119 struct kvm_memory_slot *memslot;
2122 unsigned long *dirty_bitmap;
2123 unsigned long *dirty_bitmap_buffer;
2126 /* Dirty ring tracking is exclusive to dirty log tracking */
2127 if (kvm->dirty_ring_size)
2130 as_id = log->slot >> 16;
2131 id = (u16)log->slot;
2132 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
2135 slots = __kvm_memslots(kvm, as_id);
2136 memslot = id_to_memslot(slots, id);
2137 if (!memslot || !memslot->dirty_bitmap)
2140 dirty_bitmap = memslot->dirty_bitmap;
2142 kvm_arch_sync_dirty_log(kvm, memslot);
2144 n = kvm_dirty_bitmap_bytes(memslot);
2146 if (kvm->manual_dirty_log_protect) {
2148 * Unlike kvm_get_dirty_log, we always return false in *flush,
2149 * because no flush is needed until KVM_CLEAR_DIRTY_LOG. There
2150 * is some code duplication between this function and
2151 * kvm_get_dirty_log, but hopefully all architecture
2152 * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log
2153 * can be eliminated.
2155 dirty_bitmap_buffer = dirty_bitmap;
2157 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
2158 memset(dirty_bitmap_buffer, 0, n);
2161 for (i = 0; i < n / sizeof(long); i++) {
2165 if (!dirty_bitmap[i])
2169 mask = xchg(&dirty_bitmap[i], 0);
2170 dirty_bitmap_buffer[i] = mask;
2172 offset = i * BITS_PER_LONG;
2173 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
2176 KVM_MMU_UNLOCK(kvm);
2180 kvm_arch_flush_remote_tlbs_memslot(kvm, memslot);
2182 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
2189 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
2190 * @kvm: kvm instance
2191 * @log: slot id and address to which we copy the log
2193 * Steps 1-4 below provide general overview of dirty page logging. See
2194 * kvm_get_dirty_log_protect() function description for additional details.
2196 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
2197 * always flush the TLB (step 4) even if previous step failed and the dirty
2198 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
2199 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
2200 * writes will be marked dirty for next log read.
2202 * 1. Take a snapshot of the bit and clear it if needed.
2203 * 2. Write protect the corresponding page.
2204 * 3. Copy the snapshot to the userspace.
2205 * 4. Flush TLB's if needed.
2207 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
2208 struct kvm_dirty_log *log)
2212 mutex_lock(&kvm->slots_lock);
2214 r = kvm_get_dirty_log_protect(kvm, log);
2216 mutex_unlock(&kvm->slots_lock);
2221 * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap
2222 * and reenable dirty page tracking for the corresponding pages.
2223 * @kvm: pointer to kvm instance
2224 * @log: slot id and address from which to fetch the bitmap of dirty pages
2226 static int kvm_clear_dirty_log_protect(struct kvm *kvm,
2227 struct kvm_clear_dirty_log *log)
2229 struct kvm_memslots *slots;
2230 struct kvm_memory_slot *memslot;
2234 unsigned long *dirty_bitmap;
2235 unsigned long *dirty_bitmap_buffer;
2238 /* Dirty ring tracking is exclusive to dirty log tracking */
2239 if (kvm->dirty_ring_size)
2242 as_id = log->slot >> 16;
2243 id = (u16)log->slot;
2244 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
2247 if (log->first_page & 63)
2250 slots = __kvm_memslots(kvm, as_id);
2251 memslot = id_to_memslot(slots, id);
2252 if (!memslot || !memslot->dirty_bitmap)
2255 dirty_bitmap = memslot->dirty_bitmap;
2257 n = ALIGN(log->num_pages, BITS_PER_LONG) / 8;
2259 if (log->first_page > memslot->npages ||
2260 log->num_pages > memslot->npages - log->first_page ||
2261 (log->num_pages < memslot->npages - log->first_page && (log->num_pages & 63)))
2264 kvm_arch_sync_dirty_log(kvm, memslot);
2267 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
2268 if (copy_from_user(dirty_bitmap_buffer, log->dirty_bitmap, n))
2272 for (offset = log->first_page, i = offset / BITS_PER_LONG,
2273 n = DIV_ROUND_UP(log->num_pages, BITS_PER_LONG); n--;
2274 i++, offset += BITS_PER_LONG) {
2275 unsigned long mask = *dirty_bitmap_buffer++;
2276 atomic_long_t *p = (atomic_long_t *) &dirty_bitmap[i];
2280 mask &= atomic_long_fetch_andnot(mask, p);
2283 * mask contains the bits that really have been cleared. This
2284 * never includes any bits beyond the length of the memslot (if
2285 * the length is not aligned to 64 pages), therefore it is not
2286 * a problem if userspace sets them in log->dirty_bitmap.
2290 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
2294 KVM_MMU_UNLOCK(kvm);
2297 kvm_arch_flush_remote_tlbs_memslot(kvm, memslot);
2302 static int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm,
2303 struct kvm_clear_dirty_log *log)
2307 mutex_lock(&kvm->slots_lock);
2309 r = kvm_clear_dirty_log_protect(kvm, log);
2311 mutex_unlock(&kvm->slots_lock);
2314 #endif /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
2316 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
2318 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
2320 EXPORT_SYMBOL_GPL(gfn_to_memslot);
2322 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
2324 struct kvm_memslots *slots = kvm_vcpu_memslots(vcpu);
2325 u64 gen = slots->generation;
2326 struct kvm_memory_slot *slot;
2329 * This also protects against using a memslot from a different address space,
2330 * since different address spaces have different generation numbers.
2332 if (unlikely(gen != vcpu->last_used_slot_gen)) {
2333 vcpu->last_used_slot = NULL;
2334 vcpu->last_used_slot_gen = gen;
2337 slot = try_get_memslot(vcpu->last_used_slot, gfn);
2342 * Fall back to searching all memslots. We purposely use
2343 * search_memslots() instead of __gfn_to_memslot() to avoid
2344 * thrashing the VM-wide last_used_slot in kvm_memslots.
2346 slot = search_memslots(slots, gfn, false);
2348 vcpu->last_used_slot = slot;
2355 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
2357 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
2359 return kvm_is_visible_memslot(memslot);
2361 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
2363 bool kvm_vcpu_is_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
2365 struct kvm_memory_slot *memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2367 return kvm_is_visible_memslot(memslot);
2369 EXPORT_SYMBOL_GPL(kvm_vcpu_is_visible_gfn);
2371 unsigned long kvm_host_page_size(struct kvm_vcpu *vcpu, gfn_t gfn)
2373 struct vm_area_struct *vma;
2374 unsigned long addr, size;
2378 addr = kvm_vcpu_gfn_to_hva_prot(vcpu, gfn, NULL);
2379 if (kvm_is_error_hva(addr))
2382 mmap_read_lock(current->mm);
2383 vma = find_vma(current->mm, addr);
2387 size = vma_kernel_pagesize(vma);
2390 mmap_read_unlock(current->mm);
2395 static bool memslot_is_readonly(const struct kvm_memory_slot *slot)
2397 return slot->flags & KVM_MEM_READONLY;
2400 static unsigned long __gfn_to_hva_many(const struct kvm_memory_slot *slot, gfn_t gfn,
2401 gfn_t *nr_pages, bool write)
2403 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
2404 return KVM_HVA_ERR_BAD;
2406 if (memslot_is_readonly(slot) && write)
2407 return KVM_HVA_ERR_RO_BAD;
2410 *nr_pages = slot->npages - (gfn - slot->base_gfn);
2412 return __gfn_to_hva_memslot(slot, gfn);
2415 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
2418 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
2421 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
2424 return gfn_to_hva_many(slot, gfn, NULL);
2426 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
2428 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
2430 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
2432 EXPORT_SYMBOL_GPL(gfn_to_hva);
2434 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
2436 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
2438 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
2441 * Return the hva of a @gfn and the R/W attribute if possible.
2443 * @slot: the kvm_memory_slot which contains @gfn
2444 * @gfn: the gfn to be translated
2445 * @writable: used to return the read/write attribute of the @slot if the hva
2446 * is valid and @writable is not NULL
2448 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
2449 gfn_t gfn, bool *writable)
2451 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
2453 if (!kvm_is_error_hva(hva) && writable)
2454 *writable = !memslot_is_readonly(slot);
2459 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
2461 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2463 return gfn_to_hva_memslot_prot(slot, gfn, writable);
2466 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
2468 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2470 return gfn_to_hva_memslot_prot(slot, gfn, writable);
2473 static inline int check_user_page_hwpoison(unsigned long addr)
2475 int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
2477 rc = get_user_pages(addr, 1, flags, NULL, NULL);
2478 return rc == -EHWPOISON;
2482 * The fast path to get the writable pfn which will be stored in @pfn,
2483 * true indicates success, otherwise false is returned. It's also the
2484 * only part that runs if we can in atomic context.
2486 static bool hva_to_pfn_fast(unsigned long addr, bool write_fault,
2487 bool *writable, kvm_pfn_t *pfn)
2489 struct page *page[1];
2492 * Fast pin a writable pfn only if it is a write fault request
2493 * or the caller allows to map a writable pfn for a read fault
2496 if (!(write_fault || writable))
2499 if (get_user_page_fast_only(addr, FOLL_WRITE, page)) {
2500 *pfn = page_to_pfn(page[0]);
2511 * The slow path to get the pfn of the specified host virtual address,
2512 * 1 indicates success, -errno is returned if error is detected.
2514 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
2515 bool *writable, kvm_pfn_t *pfn)
2517 unsigned int flags = FOLL_HWPOISON;
2524 *writable = write_fault;
2527 flags |= FOLL_WRITE;
2529 flags |= FOLL_NOWAIT;
2531 npages = get_user_pages_unlocked(addr, 1, &page, flags);
2535 /* map read fault as writable if possible */
2536 if (unlikely(!write_fault) && writable) {
2539 if (get_user_page_fast_only(addr, FOLL_WRITE, &wpage)) {
2545 *pfn = page_to_pfn(page);
2549 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
2551 if (unlikely(!(vma->vm_flags & VM_READ)))
2554 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
2560 static int kvm_try_get_pfn(kvm_pfn_t pfn)
2562 struct page *page = kvm_pfn_to_refcounted_page(pfn);
2567 return get_page_unless_zero(page);
2570 static int hva_to_pfn_remapped(struct vm_area_struct *vma,
2571 unsigned long addr, bool write_fault,
2572 bool *writable, kvm_pfn_t *p_pfn)
2579 r = follow_pte(vma->vm_mm, addr, &ptep, &ptl);
2582 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
2583 * not call the fault handler, so do it here.
2585 bool unlocked = false;
2586 r = fixup_user_fault(current->mm, addr,
2587 (write_fault ? FAULT_FLAG_WRITE : 0),
2594 r = follow_pte(vma->vm_mm, addr, &ptep, &ptl);
2599 if (write_fault && !pte_write(*ptep)) {
2600 pfn = KVM_PFN_ERR_RO_FAULT;
2605 *writable = pte_write(*ptep);
2606 pfn = pte_pfn(*ptep);
2609 * Get a reference here because callers of *hva_to_pfn* and
2610 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
2611 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
2612 * set, but the kvm_try_get_pfn/kvm_release_pfn_clean pair will
2613 * simply do nothing for reserved pfns.
2615 * Whoever called remap_pfn_range is also going to call e.g.
2616 * unmap_mapping_range before the underlying pages are freed,
2617 * causing a call to our MMU notifier.
2619 * Certain IO or PFNMAP mappings can be backed with valid
2620 * struct pages, but be allocated without refcounting e.g.,
2621 * tail pages of non-compound higher order allocations, which
2622 * would then underflow the refcount when the caller does the
2623 * required put_page. Don't allow those pages here.
2625 if (!kvm_try_get_pfn(pfn))
2629 pte_unmap_unlock(ptep, ptl);
2636 * Pin guest page in memory and return its pfn.
2637 * @addr: host virtual address which maps memory to the guest
2638 * @atomic: whether this function can sleep
2639 * @async: whether this function need to wait IO complete if the
2640 * host page is not in the memory
2641 * @write_fault: whether we should get a writable host page
2642 * @writable: whether it allows to map a writable host page for !@write_fault
2644 * The function will map a writable host page for these two cases:
2645 * 1): @write_fault = true
2646 * 2): @write_fault = false && @writable, @writable will tell the caller
2647 * whether the mapping is writable.
2649 kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
2650 bool write_fault, bool *writable)
2652 struct vm_area_struct *vma;
2656 /* we can do it either atomically or asynchronously, not both */
2657 BUG_ON(atomic && async);
2659 if (hva_to_pfn_fast(addr, write_fault, writable, &pfn))
2663 return KVM_PFN_ERR_FAULT;
2665 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
2669 mmap_read_lock(current->mm);
2670 if (npages == -EHWPOISON ||
2671 (!async && check_user_page_hwpoison(addr))) {
2672 pfn = KVM_PFN_ERR_HWPOISON;
2677 vma = vma_lookup(current->mm, addr);
2680 pfn = KVM_PFN_ERR_FAULT;
2681 else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
2682 r = hva_to_pfn_remapped(vma, addr, write_fault, writable, &pfn);
2686 pfn = KVM_PFN_ERR_FAULT;
2688 if (async && vma_is_valid(vma, write_fault))
2690 pfn = KVM_PFN_ERR_FAULT;
2693 mmap_read_unlock(current->mm);
2697 kvm_pfn_t __gfn_to_pfn_memslot(const struct kvm_memory_slot *slot, gfn_t gfn,
2698 bool atomic, bool *async, bool write_fault,
2699 bool *writable, hva_t *hva)
2701 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
2706 if (addr == KVM_HVA_ERR_RO_BAD) {
2709 return KVM_PFN_ERR_RO_FAULT;
2712 if (kvm_is_error_hva(addr)) {
2715 return KVM_PFN_NOSLOT;
2718 /* Do not map writable pfn in the readonly memslot. */
2719 if (writable && memslot_is_readonly(slot)) {
2724 return hva_to_pfn(addr, atomic, async, write_fault,
2727 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
2729 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
2732 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
2733 write_fault, writable, NULL);
2735 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
2737 kvm_pfn_t gfn_to_pfn_memslot(const struct kvm_memory_slot *slot, gfn_t gfn)
2739 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL, NULL);
2741 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
2743 kvm_pfn_t gfn_to_pfn_memslot_atomic(const struct kvm_memory_slot *slot, gfn_t gfn)
2745 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL, NULL);
2747 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
2749 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
2751 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
2753 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
2755 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
2757 return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
2759 EXPORT_SYMBOL_GPL(gfn_to_pfn);
2761 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
2763 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
2765 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
2767 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
2768 struct page **pages, int nr_pages)
2773 addr = gfn_to_hva_many(slot, gfn, &entry);
2774 if (kvm_is_error_hva(addr))
2777 if (entry < nr_pages)
2780 return get_user_pages_fast_only(addr, nr_pages, FOLL_WRITE, pages);
2782 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
2785 * Do not use this helper unless you are absolutely certain the gfn _must_ be
2786 * backed by 'struct page'. A valid example is if the backing memslot is
2787 * controlled by KVM. Note, if the returned page is valid, it's refcount has
2788 * been elevated by gfn_to_pfn().
2790 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
2795 pfn = gfn_to_pfn(kvm, gfn);
2797 if (is_error_noslot_pfn(pfn))
2798 return KVM_ERR_PTR_BAD_PAGE;
2800 page = kvm_pfn_to_refcounted_page(pfn);
2802 return KVM_ERR_PTR_BAD_PAGE;
2806 EXPORT_SYMBOL_GPL(gfn_to_page);
2808 void kvm_release_pfn(kvm_pfn_t pfn, bool dirty)
2811 kvm_release_pfn_dirty(pfn);
2813 kvm_release_pfn_clean(pfn);
2816 int kvm_vcpu_map(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map)
2820 struct page *page = KVM_UNMAPPED_PAGE;
2825 pfn = gfn_to_pfn(vcpu->kvm, gfn);
2826 if (is_error_noslot_pfn(pfn))
2829 if (pfn_valid(pfn)) {
2830 page = pfn_to_page(pfn);
2832 #ifdef CONFIG_HAS_IOMEM
2834 hva = memremap(pfn_to_hpa(pfn), PAGE_SIZE, MEMREMAP_WB);
2848 EXPORT_SYMBOL_GPL(kvm_vcpu_map);
2850 void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map, bool dirty)
2858 if (map->page != KVM_UNMAPPED_PAGE)
2860 #ifdef CONFIG_HAS_IOMEM
2866 kvm_vcpu_mark_page_dirty(vcpu, map->gfn);
2868 kvm_release_pfn(map->pfn, dirty);
2873 EXPORT_SYMBOL_GPL(kvm_vcpu_unmap);
2875 static bool kvm_is_ad_tracked_page(struct page *page)
2878 * Per page-flags.h, pages tagged PG_reserved "should in general not be
2879 * touched (e.g. set dirty) except by its owner".
2881 return !PageReserved(page);
2884 static void kvm_set_page_dirty(struct page *page)
2886 if (kvm_is_ad_tracked_page(page))
2890 static void kvm_set_page_accessed(struct page *page)
2892 if (kvm_is_ad_tracked_page(page))
2893 mark_page_accessed(page);
2896 void kvm_release_page_clean(struct page *page)
2898 WARN_ON(is_error_page(page));
2900 kvm_set_page_accessed(page);
2903 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
2905 void kvm_release_pfn_clean(kvm_pfn_t pfn)
2909 if (is_error_noslot_pfn(pfn))
2912 page = kvm_pfn_to_refcounted_page(pfn);
2916 kvm_release_page_clean(page);
2918 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
2920 void kvm_release_page_dirty(struct page *page)
2922 WARN_ON(is_error_page(page));
2924 kvm_set_page_dirty(page);
2925 kvm_release_page_clean(page);
2927 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
2929 void kvm_release_pfn_dirty(kvm_pfn_t pfn)
2933 if (is_error_noslot_pfn(pfn))
2936 page = kvm_pfn_to_refcounted_page(pfn);
2940 kvm_release_page_dirty(page);
2942 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
2945 * Note, checking for an error/noslot pfn is the caller's responsibility when
2946 * directly marking a page dirty/accessed. Unlike the "release" helpers, the
2947 * "set" helpers are not to be used when the pfn might point at garbage.
2949 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
2951 if (WARN_ON(is_error_noslot_pfn(pfn)))
2955 kvm_set_page_dirty(pfn_to_page(pfn));
2957 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
2959 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
2961 if (WARN_ON(is_error_noslot_pfn(pfn)))
2965 kvm_set_page_accessed(pfn_to_page(pfn));
2967 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
2969 static int next_segment(unsigned long len, int offset)
2971 if (len > PAGE_SIZE - offset)
2972 return PAGE_SIZE - offset;
2977 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
2978 void *data, int offset, int len)
2983 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
2984 if (kvm_is_error_hva(addr))
2986 r = __copy_from_user(data, (void __user *)addr + offset, len);
2992 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
2995 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2997 return __kvm_read_guest_page(slot, gfn, data, offset, len);
2999 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
3001 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
3002 int offset, int len)
3004 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
3006 return __kvm_read_guest_page(slot, gfn, data, offset, len);
3008 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
3010 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
3012 gfn_t gfn = gpa >> PAGE_SHIFT;
3014 int offset = offset_in_page(gpa);
3017 while ((seg = next_segment(len, offset)) != 0) {
3018 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
3028 EXPORT_SYMBOL_GPL(kvm_read_guest);
3030 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
3032 gfn_t gfn = gpa >> PAGE_SHIFT;
3034 int offset = offset_in_page(gpa);
3037 while ((seg = next_segment(len, offset)) != 0) {
3038 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
3048 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
3050 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
3051 void *data, int offset, unsigned long len)
3056 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
3057 if (kvm_is_error_hva(addr))
3059 pagefault_disable();
3060 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
3067 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
3068 void *data, unsigned long len)
3070 gfn_t gfn = gpa >> PAGE_SHIFT;
3071 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
3072 int offset = offset_in_page(gpa);
3074 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
3076 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
3078 static int __kvm_write_guest_page(struct kvm *kvm,
3079 struct kvm_memory_slot *memslot, gfn_t gfn,
3080 const void *data, int offset, int len)
3085 addr = gfn_to_hva_memslot(memslot, gfn);
3086 if (kvm_is_error_hva(addr))
3088 r = __copy_to_user((void __user *)addr + offset, data, len);
3091 mark_page_dirty_in_slot(kvm, memslot, gfn);
3095 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
3096 const void *data, int offset, int len)
3098 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
3100 return __kvm_write_guest_page(kvm, slot, gfn, data, offset, len);
3102 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
3104 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
3105 const void *data, int offset, int len)
3107 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
3109 return __kvm_write_guest_page(vcpu->kvm, slot, gfn, data, offset, len);
3111 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
3113 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
3116 gfn_t gfn = gpa >> PAGE_SHIFT;
3118 int offset = offset_in_page(gpa);
3121 while ((seg = next_segment(len, offset)) != 0) {
3122 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
3132 EXPORT_SYMBOL_GPL(kvm_write_guest);
3134 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
3137 gfn_t gfn = gpa >> PAGE_SHIFT;
3139 int offset = offset_in_page(gpa);
3142 while ((seg = next_segment(len, offset)) != 0) {
3143 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
3153 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
3155 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
3156 struct gfn_to_hva_cache *ghc,
3157 gpa_t gpa, unsigned long len)
3159 int offset = offset_in_page(gpa);
3160 gfn_t start_gfn = gpa >> PAGE_SHIFT;
3161 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
3162 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
3163 gfn_t nr_pages_avail;
3165 /* Update ghc->generation before performing any error checks. */
3166 ghc->generation = slots->generation;
3168 if (start_gfn > end_gfn) {
3169 ghc->hva = KVM_HVA_ERR_BAD;
3174 * If the requested region crosses two memslots, we still
3175 * verify that the entire region is valid here.
3177 for ( ; start_gfn <= end_gfn; start_gfn += nr_pages_avail) {
3178 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
3179 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
3181 if (kvm_is_error_hva(ghc->hva))
3185 /* Use the slow path for cross page reads and writes. */
3186 if (nr_pages_needed == 1)
3189 ghc->memslot = NULL;
3196 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3197 gpa_t gpa, unsigned long len)
3199 struct kvm_memslots *slots = kvm_memslots(kvm);
3200 return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len);
3202 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
3204 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3205 void *data, unsigned int offset,
3208 struct kvm_memslots *slots = kvm_memslots(kvm);
3210 gpa_t gpa = ghc->gpa + offset;
3212 if (WARN_ON_ONCE(len + offset > ghc->len))
3215 if (slots->generation != ghc->generation) {
3216 if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len))
3220 if (kvm_is_error_hva(ghc->hva))
3223 if (unlikely(!ghc->memslot))
3224 return kvm_write_guest(kvm, gpa, data, len);
3226 r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
3229 mark_page_dirty_in_slot(kvm, ghc->memslot, gpa >> PAGE_SHIFT);
3233 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached);
3235 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3236 void *data, unsigned long len)
3238 return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len);
3240 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
3242 int kvm_read_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3243 void *data, unsigned int offset,
3246 struct kvm_memslots *slots = kvm_memslots(kvm);
3248 gpa_t gpa = ghc->gpa + offset;
3250 if (WARN_ON_ONCE(len + offset > ghc->len))
3253 if (slots->generation != ghc->generation) {
3254 if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len))
3258 if (kvm_is_error_hva(ghc->hva))
3261 if (unlikely(!ghc->memslot))
3262 return kvm_read_guest(kvm, gpa, data, len);
3264 r = __copy_from_user(data, (void __user *)ghc->hva + offset, len);
3270 EXPORT_SYMBOL_GPL(kvm_read_guest_offset_cached);
3272 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3273 void *data, unsigned long len)
3275 return kvm_read_guest_offset_cached(kvm, ghc, data, 0, len);
3277 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
3279 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
3281 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
3282 gfn_t gfn = gpa >> PAGE_SHIFT;
3284 int offset = offset_in_page(gpa);
3287 while ((seg = next_segment(len, offset)) != 0) {
3288 ret = kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
3297 EXPORT_SYMBOL_GPL(kvm_clear_guest);
3299 void mark_page_dirty_in_slot(struct kvm *kvm,
3300 const struct kvm_memory_slot *memslot,
3303 struct kvm_vcpu *vcpu = kvm_get_running_vcpu();
3305 #ifdef CONFIG_HAVE_KVM_DIRTY_RING
3306 if (WARN_ON_ONCE(!vcpu) || WARN_ON_ONCE(vcpu->kvm != kvm))
3310 if (memslot && kvm_slot_dirty_track_enabled(memslot)) {
3311 unsigned long rel_gfn = gfn - memslot->base_gfn;
3312 u32 slot = (memslot->as_id << 16) | memslot->id;
3314 if (kvm->dirty_ring_size)
3315 kvm_dirty_ring_push(&vcpu->dirty_ring,
3318 set_bit_le(rel_gfn, memslot->dirty_bitmap);
3321 EXPORT_SYMBOL_GPL(mark_page_dirty_in_slot);
3323 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
3325 struct kvm_memory_slot *memslot;
3327 memslot = gfn_to_memslot(kvm, gfn);
3328 mark_page_dirty_in_slot(kvm, memslot, gfn);
3330 EXPORT_SYMBOL_GPL(mark_page_dirty);
3332 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
3334 struct kvm_memory_slot *memslot;
3336 memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
3337 mark_page_dirty_in_slot(vcpu->kvm, memslot, gfn);
3339 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
3341 void kvm_sigset_activate(struct kvm_vcpu *vcpu)
3343 if (!vcpu->sigset_active)
3347 * This does a lockless modification of ->real_blocked, which is fine
3348 * because, only current can change ->real_blocked and all readers of
3349 * ->real_blocked don't care as long ->real_blocked is always a subset
3352 sigprocmask(SIG_SETMASK, &vcpu->sigset, ¤t->real_blocked);
3355 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu)
3357 if (!vcpu->sigset_active)
3360 sigprocmask(SIG_SETMASK, ¤t->real_blocked, NULL);
3361 sigemptyset(¤t->real_blocked);
3364 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
3366 unsigned int old, val, grow, grow_start;
3368 old = val = vcpu->halt_poll_ns;
3369 grow_start = READ_ONCE(halt_poll_ns_grow_start);
3370 grow = READ_ONCE(halt_poll_ns_grow);
3375 if (val < grow_start)
3378 if (val > vcpu->kvm->max_halt_poll_ns)
3379 val = vcpu->kvm->max_halt_poll_ns;
3381 vcpu->halt_poll_ns = val;
3383 trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
3386 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
3388 unsigned int old, val, shrink, grow_start;
3390 old = val = vcpu->halt_poll_ns;
3391 shrink = READ_ONCE(halt_poll_ns_shrink);
3392 grow_start = READ_ONCE(halt_poll_ns_grow_start);
3398 if (val < grow_start)
3401 vcpu->halt_poll_ns = val;
3402 trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
3405 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
3408 int idx = srcu_read_lock(&vcpu->kvm->srcu);
3410 if (kvm_arch_vcpu_runnable(vcpu)) {
3411 kvm_make_request(KVM_REQ_UNHALT, vcpu);
3414 if (kvm_cpu_has_pending_timer(vcpu))
3416 if (signal_pending(current))
3418 if (kvm_check_request(KVM_REQ_UNBLOCK, vcpu))
3423 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3428 * Block the vCPU until the vCPU is runnable, an event arrives, or a signal is
3429 * pending. This is mostly used when halting a vCPU, but may also be used
3430 * directly for other vCPU non-runnable states, e.g. x86's Wait-For-SIPI.
3432 bool kvm_vcpu_block(struct kvm_vcpu *vcpu)
3434 struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu);
3435 bool waited = false;
3437 vcpu->stat.generic.blocking = 1;
3440 kvm_arch_vcpu_blocking(vcpu);
3441 prepare_to_rcuwait(wait);
3445 set_current_state(TASK_INTERRUPTIBLE);
3447 if (kvm_vcpu_check_block(vcpu) < 0)
3455 finish_rcuwait(wait);
3456 kvm_arch_vcpu_unblocking(vcpu);
3459 vcpu->stat.generic.blocking = 0;
3464 static inline void update_halt_poll_stats(struct kvm_vcpu *vcpu, ktime_t start,
3465 ktime_t end, bool success)
3467 struct kvm_vcpu_stat_generic *stats = &vcpu->stat.generic;
3468 u64 poll_ns = ktime_to_ns(ktime_sub(end, start));
3470 ++vcpu->stat.generic.halt_attempted_poll;
3473 ++vcpu->stat.generic.halt_successful_poll;
3475 if (!vcpu_valid_wakeup(vcpu))
3476 ++vcpu->stat.generic.halt_poll_invalid;
3478 stats->halt_poll_success_ns += poll_ns;
3479 KVM_STATS_LOG_HIST_UPDATE(stats->halt_poll_success_hist, poll_ns);
3481 stats->halt_poll_fail_ns += poll_ns;
3482 KVM_STATS_LOG_HIST_UPDATE(stats->halt_poll_fail_hist, poll_ns);
3487 * Emulate a vCPU halt condition, e.g. HLT on x86, WFI on arm, etc... If halt
3488 * polling is enabled, busy wait for a short time before blocking to avoid the
3489 * expensive block+unblock sequence if a wake event arrives soon after the vCPU
3492 void kvm_vcpu_halt(struct kvm_vcpu *vcpu)
3494 bool halt_poll_allowed = !kvm_arch_no_poll(vcpu);
3495 bool do_halt_poll = halt_poll_allowed && vcpu->halt_poll_ns;
3496 ktime_t start, cur, poll_end;
3497 bool waited = false;
3500 start = cur = poll_end = ktime_get();
3502 ktime_t stop = ktime_add_ns(start, vcpu->halt_poll_ns);
3506 * This sets KVM_REQ_UNHALT if an interrupt
3509 if (kvm_vcpu_check_block(vcpu) < 0)
3512 poll_end = cur = ktime_get();
3513 } while (kvm_vcpu_can_poll(cur, stop));
3516 waited = kvm_vcpu_block(vcpu);
3520 vcpu->stat.generic.halt_wait_ns +=
3521 ktime_to_ns(cur) - ktime_to_ns(poll_end);
3522 KVM_STATS_LOG_HIST_UPDATE(vcpu->stat.generic.halt_wait_hist,
3523 ktime_to_ns(cur) - ktime_to_ns(poll_end));
3526 /* The total time the vCPU was "halted", including polling time. */
3527 halt_ns = ktime_to_ns(cur) - ktime_to_ns(start);
3530 * Note, halt-polling is considered successful so long as the vCPU was
3531 * never actually scheduled out, i.e. even if the wake event arrived
3532 * after of the halt-polling loop itself, but before the full wait.
3535 update_halt_poll_stats(vcpu, start, poll_end, !waited);
3537 if (halt_poll_allowed) {
3538 if (!vcpu_valid_wakeup(vcpu)) {
3539 shrink_halt_poll_ns(vcpu);
3540 } else if (vcpu->kvm->max_halt_poll_ns) {
3541 if (halt_ns <= vcpu->halt_poll_ns)
3543 /* we had a long block, shrink polling */
3544 else if (vcpu->halt_poll_ns &&
3545 halt_ns > vcpu->kvm->max_halt_poll_ns)
3546 shrink_halt_poll_ns(vcpu);
3547 /* we had a short halt and our poll time is too small */
3548 else if (vcpu->halt_poll_ns < vcpu->kvm->max_halt_poll_ns &&
3549 halt_ns < vcpu->kvm->max_halt_poll_ns)
3550 grow_halt_poll_ns(vcpu);
3552 vcpu->halt_poll_ns = 0;
3556 trace_kvm_vcpu_wakeup(halt_ns, waited, vcpu_valid_wakeup(vcpu));
3558 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
3560 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
3562 if (__kvm_vcpu_wake_up(vcpu)) {
3563 WRITE_ONCE(vcpu->ready, true);
3564 ++vcpu->stat.generic.halt_wakeup;
3570 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
3574 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
3576 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
3580 if (kvm_vcpu_wake_up(vcpu))
3585 * The only state change done outside the vcpu mutex is IN_GUEST_MODE
3586 * to EXITING_GUEST_MODE. Therefore the moderately expensive "should
3587 * kick" check does not need atomic operations if kvm_vcpu_kick is used
3588 * within the vCPU thread itself.
3590 if (vcpu == __this_cpu_read(kvm_running_vcpu)) {
3591 if (vcpu->mode == IN_GUEST_MODE)
3592 WRITE_ONCE(vcpu->mode, EXITING_GUEST_MODE);
3597 * Note, the vCPU could get migrated to a different pCPU at any point
3598 * after kvm_arch_vcpu_should_kick(), which could result in sending an
3599 * IPI to the previous pCPU. But, that's ok because the purpose of the
3600 * IPI is to force the vCPU to leave IN_GUEST_MODE, and migrating the
3601 * vCPU also requires it to leave IN_GUEST_MODE.
3603 if (kvm_arch_vcpu_should_kick(vcpu)) {
3604 cpu = READ_ONCE(vcpu->cpu);
3605 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
3606 smp_send_reschedule(cpu);
3611 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
3612 #endif /* !CONFIG_S390 */
3614 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
3617 struct task_struct *task = NULL;
3621 pid = rcu_dereference(target->pid);
3623 task = get_pid_task(pid, PIDTYPE_PID);
3627 ret = yield_to(task, 1);
3628 put_task_struct(task);
3632 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
3635 * Helper that checks whether a VCPU is eligible for directed yield.
3636 * Most eligible candidate to yield is decided by following heuristics:
3638 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
3639 * (preempted lock holder), indicated by @in_spin_loop.
3640 * Set at the beginning and cleared at the end of interception/PLE handler.
3642 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
3643 * chance last time (mostly it has become eligible now since we have probably
3644 * yielded to lockholder in last iteration. This is done by toggling
3645 * @dy_eligible each time a VCPU checked for eligibility.)
3647 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
3648 * to preempted lock-holder could result in wrong VCPU selection and CPU
3649 * burning. Giving priority for a potential lock-holder increases lock
3652 * Since algorithm is based on heuristics, accessing another VCPU data without
3653 * locking does not harm. It may result in trying to yield to same VCPU, fail
3654 * and continue with next VCPU and so on.
3656 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
3658 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
3661 eligible = !vcpu->spin_loop.in_spin_loop ||
3662 vcpu->spin_loop.dy_eligible;
3664 if (vcpu->spin_loop.in_spin_loop)
3665 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
3674 * Unlike kvm_arch_vcpu_runnable, this function is called outside
3675 * a vcpu_load/vcpu_put pair. However, for most architectures
3676 * kvm_arch_vcpu_runnable does not require vcpu_load.
3678 bool __weak kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
3680 return kvm_arch_vcpu_runnable(vcpu);
3683 static bool vcpu_dy_runnable(struct kvm_vcpu *vcpu)
3685 if (kvm_arch_dy_runnable(vcpu))
3688 #ifdef CONFIG_KVM_ASYNC_PF
3689 if (!list_empty_careful(&vcpu->async_pf.done))
3696 bool __weak kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu)
3701 void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode)
3703 struct kvm *kvm = me->kvm;
3704 struct kvm_vcpu *vcpu;
3705 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
3711 kvm_vcpu_set_in_spin_loop(me, true);
3713 * We boost the priority of a VCPU that is runnable but not
3714 * currently running, because it got preempted by something
3715 * else and called schedule in __vcpu_run. Hopefully that
3716 * VCPU is holding the lock that we need and will release it.
3717 * We approximate round-robin by starting at the last boosted VCPU.
3719 for (pass = 0; pass < 2 && !yielded && try; pass++) {
3720 kvm_for_each_vcpu(i, vcpu, kvm) {
3721 if (!pass && i <= last_boosted_vcpu) {
3722 i = last_boosted_vcpu;
3724 } else if (pass && i > last_boosted_vcpu)
3726 if (!READ_ONCE(vcpu->ready))
3730 if (kvm_vcpu_is_blocking(vcpu) && !vcpu_dy_runnable(vcpu))
3732 if (READ_ONCE(vcpu->preempted) && yield_to_kernel_mode &&
3733 !kvm_arch_dy_has_pending_interrupt(vcpu) &&
3734 !kvm_arch_vcpu_in_kernel(vcpu))
3736 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
3739 yielded = kvm_vcpu_yield_to(vcpu);
3741 kvm->last_boosted_vcpu = i;
3743 } else if (yielded < 0) {
3750 kvm_vcpu_set_in_spin_loop(me, false);
3752 /* Ensure vcpu is not eligible during next spinloop */
3753 kvm_vcpu_set_dy_eligible(me, false);
3755 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
3757 static bool kvm_page_in_dirty_ring(struct kvm *kvm, unsigned long pgoff)
3759 #ifdef CONFIG_HAVE_KVM_DIRTY_RING
3760 return (pgoff >= KVM_DIRTY_LOG_PAGE_OFFSET) &&
3761 (pgoff < KVM_DIRTY_LOG_PAGE_OFFSET +
3762 kvm->dirty_ring_size / PAGE_SIZE);
3768 static vm_fault_t kvm_vcpu_fault(struct vm_fault *vmf)
3770 struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data;
3773 if (vmf->pgoff == 0)
3774 page = virt_to_page(vcpu->run);
3776 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
3777 page = virt_to_page(vcpu->arch.pio_data);
3779 #ifdef CONFIG_KVM_MMIO
3780 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
3781 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
3783 else if (kvm_page_in_dirty_ring(vcpu->kvm, vmf->pgoff))
3784 page = kvm_dirty_ring_get_page(
3786 vmf->pgoff - KVM_DIRTY_LOG_PAGE_OFFSET);
3788 return kvm_arch_vcpu_fault(vcpu, vmf);
3794 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
3795 .fault = kvm_vcpu_fault,
3798 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
3800 struct kvm_vcpu *vcpu = file->private_data;
3801 unsigned long pages = vma_pages(vma);
3803 if ((kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff) ||
3804 kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff + pages - 1)) &&
3805 ((vma->vm_flags & VM_EXEC) || !(vma->vm_flags & VM_SHARED)))
3808 vma->vm_ops = &kvm_vcpu_vm_ops;
3812 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
3814 struct kvm_vcpu *vcpu = filp->private_data;
3816 kvm_put_kvm(vcpu->kvm);
3820 static const struct file_operations kvm_vcpu_fops = {
3821 .release = kvm_vcpu_release,
3822 .unlocked_ioctl = kvm_vcpu_ioctl,
3823 .mmap = kvm_vcpu_mmap,
3824 .llseek = noop_llseek,
3825 KVM_COMPAT(kvm_vcpu_compat_ioctl),
3829 * Allocates an inode for the vcpu.
3831 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
3833 char name[8 + 1 + ITOA_MAX_LEN + 1];
3835 snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id);
3836 return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
3839 #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS
3840 static int vcpu_get_pid(void *data, u64 *val)
3842 struct kvm_vcpu *vcpu = (struct kvm_vcpu *) data;
3843 *val = pid_nr(rcu_access_pointer(vcpu->pid));
3847 DEFINE_SIMPLE_ATTRIBUTE(vcpu_get_pid_fops, vcpu_get_pid, NULL, "%llu\n");
3849 static void kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
3851 struct dentry *debugfs_dentry;
3852 char dir_name[ITOA_MAX_LEN * 2];
3854 if (!debugfs_initialized())
3857 snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
3858 debugfs_dentry = debugfs_create_dir(dir_name,
3859 vcpu->kvm->debugfs_dentry);
3860 debugfs_create_file("pid", 0444, debugfs_dentry, vcpu,
3861 &vcpu_get_pid_fops);
3863 kvm_arch_create_vcpu_debugfs(vcpu, debugfs_dentry);
3868 * Creates some virtual cpus. Good luck creating more than one.
3870 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
3873 struct kvm_vcpu *vcpu;
3876 if (id >= KVM_MAX_VCPU_IDS)
3879 mutex_lock(&kvm->lock);
3880 if (kvm->created_vcpus >= kvm->max_vcpus) {
3881 mutex_unlock(&kvm->lock);
3885 r = kvm_arch_vcpu_precreate(kvm, id);
3887 mutex_unlock(&kvm->lock);
3891 kvm->created_vcpus++;
3892 mutex_unlock(&kvm->lock);
3894 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL_ACCOUNT);
3897 goto vcpu_decrement;
3900 BUILD_BUG_ON(sizeof(struct kvm_run) > PAGE_SIZE);
3901 page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
3906 vcpu->run = page_address(page);
3908 kvm_vcpu_init(vcpu, kvm, id);
3910 r = kvm_arch_vcpu_create(vcpu);
3912 goto vcpu_free_run_page;
3914 if (kvm->dirty_ring_size) {
3915 r = kvm_dirty_ring_alloc(&vcpu->dirty_ring,
3916 id, kvm->dirty_ring_size);
3918 goto arch_vcpu_destroy;
3921 mutex_lock(&kvm->lock);
3922 if (kvm_get_vcpu_by_id(kvm, id)) {
3924 goto unlock_vcpu_destroy;
3927 vcpu->vcpu_idx = atomic_read(&kvm->online_vcpus);
3928 r = xa_insert(&kvm->vcpu_array, vcpu->vcpu_idx, vcpu, GFP_KERNEL_ACCOUNT);
3929 BUG_ON(r == -EBUSY);
3931 goto unlock_vcpu_destroy;
3933 /* Now it's all set up, let userspace reach it */
3935 r = create_vcpu_fd(vcpu);
3937 xa_erase(&kvm->vcpu_array, vcpu->vcpu_idx);
3938 kvm_put_kvm_no_destroy(kvm);
3939 goto unlock_vcpu_destroy;
3943 * Pairs with smp_rmb() in kvm_get_vcpu. Store the vcpu
3944 * pointer before kvm->online_vcpu's incremented value.
3947 atomic_inc(&kvm->online_vcpus);
3949 mutex_unlock(&kvm->lock);
3950 kvm_arch_vcpu_postcreate(vcpu);
3951 kvm_create_vcpu_debugfs(vcpu);
3954 unlock_vcpu_destroy:
3955 mutex_unlock(&kvm->lock);
3956 kvm_dirty_ring_free(&vcpu->dirty_ring);
3958 kvm_arch_vcpu_destroy(vcpu);
3960 free_page((unsigned long)vcpu->run);
3962 kmem_cache_free(kvm_vcpu_cache, vcpu);
3964 mutex_lock(&kvm->lock);
3965 kvm->created_vcpus--;
3966 mutex_unlock(&kvm->lock);
3970 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
3973 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
3974 vcpu->sigset_active = 1;
3975 vcpu->sigset = *sigset;
3977 vcpu->sigset_active = 0;
3981 static ssize_t kvm_vcpu_stats_read(struct file *file, char __user *user_buffer,
3982 size_t size, loff_t *offset)
3984 struct kvm_vcpu *vcpu = file->private_data;
3986 return kvm_stats_read(vcpu->stats_id, &kvm_vcpu_stats_header,
3987 &kvm_vcpu_stats_desc[0], &vcpu->stat,
3988 sizeof(vcpu->stat), user_buffer, size, offset);
3991 static const struct file_operations kvm_vcpu_stats_fops = {
3992 .read = kvm_vcpu_stats_read,
3993 .llseek = noop_llseek,
3996 static int kvm_vcpu_ioctl_get_stats_fd(struct kvm_vcpu *vcpu)
4000 char name[15 + ITOA_MAX_LEN + 1];
4002 snprintf(name, sizeof(name), "kvm-vcpu-stats:%d", vcpu->vcpu_id);
4004 fd = get_unused_fd_flags(O_CLOEXEC);
4008 file = anon_inode_getfile(name, &kvm_vcpu_stats_fops, vcpu, O_RDONLY);
4011 return PTR_ERR(file);
4013 file->f_mode |= FMODE_PREAD;
4014 fd_install(fd, file);
4019 static long kvm_vcpu_ioctl(struct file *filp,
4020 unsigned int ioctl, unsigned long arg)
4022 struct kvm_vcpu *vcpu = filp->private_data;
4023 void __user *argp = (void __user *)arg;
4025 struct kvm_fpu *fpu = NULL;
4026 struct kvm_sregs *kvm_sregs = NULL;
4028 if (vcpu->kvm->mm != current->mm || vcpu->kvm->vm_dead)
4031 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
4035 * Some architectures have vcpu ioctls that are asynchronous to vcpu
4036 * execution; mutex_lock() would break them.
4038 r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg);
4039 if (r != -ENOIOCTLCMD)
4042 if (mutex_lock_killable(&vcpu->mutex))
4050 oldpid = rcu_access_pointer(vcpu->pid);
4051 if (unlikely(oldpid != task_pid(current))) {
4052 /* The thread running this VCPU changed. */
4055 r = kvm_arch_vcpu_run_pid_change(vcpu);
4059 newpid = get_task_pid(current, PIDTYPE_PID);
4060 rcu_assign_pointer(vcpu->pid, newpid);
4065 r = kvm_arch_vcpu_ioctl_run(vcpu);
4066 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
4069 case KVM_GET_REGS: {
4070 struct kvm_regs *kvm_regs;
4073 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL_ACCOUNT);
4076 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
4080 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
4087 case KVM_SET_REGS: {
4088 struct kvm_regs *kvm_regs;
4090 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
4091 if (IS_ERR(kvm_regs)) {
4092 r = PTR_ERR(kvm_regs);
4095 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
4099 case KVM_GET_SREGS: {
4100 kvm_sregs = kzalloc(sizeof(struct kvm_sregs),
4101 GFP_KERNEL_ACCOUNT);
4105 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
4109 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
4114 case KVM_SET_SREGS: {
4115 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
4116 if (IS_ERR(kvm_sregs)) {
4117 r = PTR_ERR(kvm_sregs);
4121 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
4124 case KVM_GET_MP_STATE: {
4125 struct kvm_mp_state mp_state;
4127 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
4131 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
4136 case KVM_SET_MP_STATE: {
4137 struct kvm_mp_state mp_state;
4140 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
4142 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
4145 case KVM_TRANSLATE: {
4146 struct kvm_translation tr;
4149 if (copy_from_user(&tr, argp, sizeof(tr)))
4151 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
4155 if (copy_to_user(argp, &tr, sizeof(tr)))
4160 case KVM_SET_GUEST_DEBUG: {
4161 struct kvm_guest_debug dbg;
4164 if (copy_from_user(&dbg, argp, sizeof(dbg)))
4166 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
4169 case KVM_SET_SIGNAL_MASK: {
4170 struct kvm_signal_mask __user *sigmask_arg = argp;
4171 struct kvm_signal_mask kvm_sigmask;
4172 sigset_t sigset, *p;
4177 if (copy_from_user(&kvm_sigmask, argp,
4178 sizeof(kvm_sigmask)))
4181 if (kvm_sigmask.len != sizeof(sigset))
4184 if (copy_from_user(&sigset, sigmask_arg->sigset,
4189 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
4193 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL_ACCOUNT);
4197 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
4201 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
4207 fpu = memdup_user(argp, sizeof(*fpu));
4213 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
4216 case KVM_GET_STATS_FD: {
4217 r = kvm_vcpu_ioctl_get_stats_fd(vcpu);
4221 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
4224 mutex_unlock(&vcpu->mutex);
4230 #ifdef CONFIG_KVM_COMPAT
4231 static long kvm_vcpu_compat_ioctl(struct file *filp,
4232 unsigned int ioctl, unsigned long arg)
4234 struct kvm_vcpu *vcpu = filp->private_data;
4235 void __user *argp = compat_ptr(arg);
4238 if (vcpu->kvm->mm != current->mm || vcpu->kvm->vm_dead)
4242 case KVM_SET_SIGNAL_MASK: {
4243 struct kvm_signal_mask __user *sigmask_arg = argp;
4244 struct kvm_signal_mask kvm_sigmask;
4249 if (copy_from_user(&kvm_sigmask, argp,
4250 sizeof(kvm_sigmask)))
4253 if (kvm_sigmask.len != sizeof(compat_sigset_t))
4256 if (get_compat_sigset(&sigset,
4257 (compat_sigset_t __user *)sigmask_arg->sigset))
4259 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
4261 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
4265 r = kvm_vcpu_ioctl(filp, ioctl, arg);
4273 static int kvm_device_mmap(struct file *filp, struct vm_area_struct *vma)
4275 struct kvm_device *dev = filp->private_data;
4278 return dev->ops->mmap(dev, vma);
4283 static int kvm_device_ioctl_attr(struct kvm_device *dev,
4284 int (*accessor)(struct kvm_device *dev,
4285 struct kvm_device_attr *attr),
4288 struct kvm_device_attr attr;
4293 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
4296 return accessor(dev, &attr);
4299 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
4302 struct kvm_device *dev = filp->private_data;
4304 if (dev->kvm->mm != current->mm || dev->kvm->vm_dead)
4308 case KVM_SET_DEVICE_ATTR:
4309 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
4310 case KVM_GET_DEVICE_ATTR:
4311 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
4312 case KVM_HAS_DEVICE_ATTR:
4313 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
4315 if (dev->ops->ioctl)
4316 return dev->ops->ioctl(dev, ioctl, arg);
4322 static int kvm_device_release(struct inode *inode, struct file *filp)
4324 struct kvm_device *dev = filp->private_data;
4325 struct kvm *kvm = dev->kvm;
4327 if (dev->ops->release) {
4328 mutex_lock(&kvm->lock);
4329 list_del(&dev->vm_node);
4330 dev->ops->release(dev);
4331 mutex_unlock(&kvm->lock);
4338 static const struct file_operations kvm_device_fops = {
4339 .unlocked_ioctl = kvm_device_ioctl,
4340 .release = kvm_device_release,
4341 KVM_COMPAT(kvm_device_ioctl),
4342 .mmap = kvm_device_mmap,
4345 struct kvm_device *kvm_device_from_filp(struct file *filp)
4347 if (filp->f_op != &kvm_device_fops)
4350 return filp->private_data;
4353 static const struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
4354 #ifdef CONFIG_KVM_MPIC
4355 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
4356 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
4360 int kvm_register_device_ops(const struct kvm_device_ops *ops, u32 type)
4362 if (type >= ARRAY_SIZE(kvm_device_ops_table))
4365 if (kvm_device_ops_table[type] != NULL)
4368 kvm_device_ops_table[type] = ops;
4372 void kvm_unregister_device_ops(u32 type)
4374 if (kvm_device_ops_table[type] != NULL)
4375 kvm_device_ops_table[type] = NULL;
4378 static int kvm_ioctl_create_device(struct kvm *kvm,
4379 struct kvm_create_device *cd)
4381 const struct kvm_device_ops *ops = NULL;
4382 struct kvm_device *dev;
4383 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
4387 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
4390 type = array_index_nospec(cd->type, ARRAY_SIZE(kvm_device_ops_table));
4391 ops = kvm_device_ops_table[type];
4398 dev = kzalloc(sizeof(*dev), GFP_KERNEL_ACCOUNT);
4405 mutex_lock(&kvm->lock);
4406 ret = ops->create(dev, type);
4408 mutex_unlock(&kvm->lock);
4412 list_add(&dev->vm_node, &kvm->devices);
4413 mutex_unlock(&kvm->lock);
4419 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
4421 kvm_put_kvm_no_destroy(kvm);
4422 mutex_lock(&kvm->lock);
4423 list_del(&dev->vm_node);
4426 mutex_unlock(&kvm->lock);
4436 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
4439 case KVM_CAP_USER_MEMORY:
4440 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
4441 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
4442 case KVM_CAP_INTERNAL_ERROR_DATA:
4443 #ifdef CONFIG_HAVE_KVM_MSI
4444 case KVM_CAP_SIGNAL_MSI:
4446 #ifdef CONFIG_HAVE_KVM_IRQFD
4448 case KVM_CAP_IRQFD_RESAMPLE:
4450 case KVM_CAP_IOEVENTFD_ANY_LENGTH:
4451 case KVM_CAP_CHECK_EXTENSION_VM:
4452 case KVM_CAP_ENABLE_CAP_VM:
4453 case KVM_CAP_HALT_POLL:
4455 #ifdef CONFIG_KVM_MMIO
4456 case KVM_CAP_COALESCED_MMIO:
4457 return KVM_COALESCED_MMIO_PAGE_OFFSET;
4458 case KVM_CAP_COALESCED_PIO:
4461 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4462 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2:
4463 return KVM_DIRTY_LOG_MANUAL_CAPS;
4465 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
4466 case KVM_CAP_IRQ_ROUTING:
4467 return KVM_MAX_IRQ_ROUTES;
4469 #if KVM_ADDRESS_SPACE_NUM > 1
4470 case KVM_CAP_MULTI_ADDRESS_SPACE:
4471 return KVM_ADDRESS_SPACE_NUM;
4473 case KVM_CAP_NR_MEMSLOTS:
4474 return KVM_USER_MEM_SLOTS;
4475 case KVM_CAP_DIRTY_LOG_RING:
4476 #ifdef CONFIG_HAVE_KVM_DIRTY_RING
4477 return KVM_DIRTY_RING_MAX_ENTRIES * sizeof(struct kvm_dirty_gfn);
4481 case KVM_CAP_BINARY_STATS_FD:
4482 case KVM_CAP_SYSTEM_EVENT_DATA:
4487 return kvm_vm_ioctl_check_extension(kvm, arg);
4490 static int kvm_vm_ioctl_enable_dirty_log_ring(struct kvm *kvm, u32 size)
4494 if (!KVM_DIRTY_LOG_PAGE_OFFSET)
4497 /* the size should be power of 2 */
4498 if (!size || (size & (size - 1)))
4501 /* Should be bigger to keep the reserved entries, or a page */
4502 if (size < kvm_dirty_ring_get_rsvd_entries() *
4503 sizeof(struct kvm_dirty_gfn) || size < PAGE_SIZE)
4506 if (size > KVM_DIRTY_RING_MAX_ENTRIES *
4507 sizeof(struct kvm_dirty_gfn))
4510 /* We only allow it to set once */
4511 if (kvm->dirty_ring_size)
4514 mutex_lock(&kvm->lock);
4516 if (kvm->created_vcpus) {
4517 /* We don't allow to change this value after vcpu created */
4520 kvm->dirty_ring_size = size;
4524 mutex_unlock(&kvm->lock);
4528 static int kvm_vm_ioctl_reset_dirty_pages(struct kvm *kvm)
4531 struct kvm_vcpu *vcpu;
4534 if (!kvm->dirty_ring_size)
4537 mutex_lock(&kvm->slots_lock);
4539 kvm_for_each_vcpu(i, vcpu, kvm)
4540 cleared += kvm_dirty_ring_reset(vcpu->kvm, &vcpu->dirty_ring);
4542 mutex_unlock(&kvm->slots_lock);
4545 kvm_flush_remote_tlbs(kvm);
4550 int __attribute__((weak)) kvm_vm_ioctl_enable_cap(struct kvm *kvm,
4551 struct kvm_enable_cap *cap)
4556 static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm,
4557 struct kvm_enable_cap *cap)
4560 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4561 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2: {
4562 u64 allowed_options = KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE;
4564 if (cap->args[0] & KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE)
4565 allowed_options = KVM_DIRTY_LOG_MANUAL_CAPS;
4567 if (cap->flags || (cap->args[0] & ~allowed_options))
4569 kvm->manual_dirty_log_protect = cap->args[0];
4573 case KVM_CAP_HALT_POLL: {
4574 if (cap->flags || cap->args[0] != (unsigned int)cap->args[0])
4577 kvm->max_halt_poll_ns = cap->args[0];
4580 case KVM_CAP_DIRTY_LOG_RING:
4581 return kvm_vm_ioctl_enable_dirty_log_ring(kvm, cap->args[0]);
4583 return kvm_vm_ioctl_enable_cap(kvm, cap);
4587 static ssize_t kvm_vm_stats_read(struct file *file, char __user *user_buffer,
4588 size_t size, loff_t *offset)
4590 struct kvm *kvm = file->private_data;
4592 return kvm_stats_read(kvm->stats_id, &kvm_vm_stats_header,
4593 &kvm_vm_stats_desc[0], &kvm->stat,
4594 sizeof(kvm->stat), user_buffer, size, offset);
4597 static const struct file_operations kvm_vm_stats_fops = {
4598 .read = kvm_vm_stats_read,
4599 .llseek = noop_llseek,
4602 static int kvm_vm_ioctl_get_stats_fd(struct kvm *kvm)
4607 fd = get_unused_fd_flags(O_CLOEXEC);
4611 file = anon_inode_getfile("kvm-vm-stats",
4612 &kvm_vm_stats_fops, kvm, O_RDONLY);
4615 return PTR_ERR(file);
4617 file->f_mode |= FMODE_PREAD;
4618 fd_install(fd, file);
4623 static long kvm_vm_ioctl(struct file *filp,
4624 unsigned int ioctl, unsigned long arg)
4626 struct kvm *kvm = filp->private_data;
4627 void __user *argp = (void __user *)arg;
4630 if (kvm->mm != current->mm || kvm->vm_dead)
4633 case KVM_CREATE_VCPU:
4634 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
4636 case KVM_ENABLE_CAP: {
4637 struct kvm_enable_cap cap;
4640 if (copy_from_user(&cap, argp, sizeof(cap)))
4642 r = kvm_vm_ioctl_enable_cap_generic(kvm, &cap);
4645 case KVM_SET_USER_MEMORY_REGION: {
4646 struct kvm_userspace_memory_region kvm_userspace_mem;
4649 if (copy_from_user(&kvm_userspace_mem, argp,
4650 sizeof(kvm_userspace_mem)))
4653 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
4656 case KVM_GET_DIRTY_LOG: {
4657 struct kvm_dirty_log log;
4660 if (copy_from_user(&log, argp, sizeof(log)))
4662 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
4665 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4666 case KVM_CLEAR_DIRTY_LOG: {
4667 struct kvm_clear_dirty_log log;
4670 if (copy_from_user(&log, argp, sizeof(log)))
4672 r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
4676 #ifdef CONFIG_KVM_MMIO
4677 case KVM_REGISTER_COALESCED_MMIO: {
4678 struct kvm_coalesced_mmio_zone zone;
4681 if (copy_from_user(&zone, argp, sizeof(zone)))
4683 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
4686 case KVM_UNREGISTER_COALESCED_MMIO: {
4687 struct kvm_coalesced_mmio_zone zone;
4690 if (copy_from_user(&zone, argp, sizeof(zone)))
4692 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
4697 struct kvm_irqfd data;
4700 if (copy_from_user(&data, argp, sizeof(data)))
4702 r = kvm_irqfd(kvm, &data);
4705 case KVM_IOEVENTFD: {
4706 struct kvm_ioeventfd data;
4709 if (copy_from_user(&data, argp, sizeof(data)))
4711 r = kvm_ioeventfd(kvm, &data);
4714 #ifdef CONFIG_HAVE_KVM_MSI
4715 case KVM_SIGNAL_MSI: {
4719 if (copy_from_user(&msi, argp, sizeof(msi)))
4721 r = kvm_send_userspace_msi(kvm, &msi);
4725 #ifdef __KVM_HAVE_IRQ_LINE
4726 case KVM_IRQ_LINE_STATUS:
4727 case KVM_IRQ_LINE: {
4728 struct kvm_irq_level irq_event;
4731 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
4734 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
4735 ioctl == KVM_IRQ_LINE_STATUS);
4740 if (ioctl == KVM_IRQ_LINE_STATUS) {
4741 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
4749 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
4750 case KVM_SET_GSI_ROUTING: {
4751 struct kvm_irq_routing routing;
4752 struct kvm_irq_routing __user *urouting;
4753 struct kvm_irq_routing_entry *entries = NULL;
4756 if (copy_from_user(&routing, argp, sizeof(routing)))
4759 if (!kvm_arch_can_set_irq_routing(kvm))
4761 if (routing.nr > KVM_MAX_IRQ_ROUTES)
4767 entries = vmemdup_user(urouting->entries,
4768 array_size(sizeof(*entries),
4770 if (IS_ERR(entries)) {
4771 r = PTR_ERR(entries);
4775 r = kvm_set_irq_routing(kvm, entries, routing.nr,
4780 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
4781 case KVM_CREATE_DEVICE: {
4782 struct kvm_create_device cd;
4785 if (copy_from_user(&cd, argp, sizeof(cd)))
4788 r = kvm_ioctl_create_device(kvm, &cd);
4793 if (copy_to_user(argp, &cd, sizeof(cd)))
4799 case KVM_CHECK_EXTENSION:
4800 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
4802 case KVM_RESET_DIRTY_RINGS:
4803 r = kvm_vm_ioctl_reset_dirty_pages(kvm);
4805 case KVM_GET_STATS_FD:
4806 r = kvm_vm_ioctl_get_stats_fd(kvm);
4809 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
4815 #ifdef CONFIG_KVM_COMPAT
4816 struct compat_kvm_dirty_log {
4820 compat_uptr_t dirty_bitmap; /* one bit per page */
4825 struct compat_kvm_clear_dirty_log {
4830 compat_uptr_t dirty_bitmap; /* one bit per page */
4835 static long kvm_vm_compat_ioctl(struct file *filp,
4836 unsigned int ioctl, unsigned long arg)
4838 struct kvm *kvm = filp->private_data;
4841 if (kvm->mm != current->mm || kvm->vm_dead)
4844 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4845 case KVM_CLEAR_DIRTY_LOG: {
4846 struct compat_kvm_clear_dirty_log compat_log;
4847 struct kvm_clear_dirty_log log;
4849 if (copy_from_user(&compat_log, (void __user *)arg,
4850 sizeof(compat_log)))
4852 log.slot = compat_log.slot;
4853 log.num_pages = compat_log.num_pages;
4854 log.first_page = compat_log.first_page;
4855 log.padding2 = compat_log.padding2;
4856 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
4858 r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
4862 case KVM_GET_DIRTY_LOG: {
4863 struct compat_kvm_dirty_log compat_log;
4864 struct kvm_dirty_log log;
4866 if (copy_from_user(&compat_log, (void __user *)arg,
4867 sizeof(compat_log)))
4869 log.slot = compat_log.slot;
4870 log.padding1 = compat_log.padding1;
4871 log.padding2 = compat_log.padding2;
4872 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
4874 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
4878 r = kvm_vm_ioctl(filp, ioctl, arg);
4884 static const struct file_operations kvm_vm_fops = {
4885 .release = kvm_vm_release,
4886 .unlocked_ioctl = kvm_vm_ioctl,
4887 .llseek = noop_llseek,
4888 KVM_COMPAT(kvm_vm_compat_ioctl),
4891 bool file_is_kvm(struct file *file)
4893 return file && file->f_op == &kvm_vm_fops;
4895 EXPORT_SYMBOL_GPL(file_is_kvm);
4897 static int kvm_dev_ioctl_create_vm(unsigned long type)
4899 char fdname[ITOA_MAX_LEN + 1];
4904 fd = get_unused_fd_flags(O_CLOEXEC);
4908 snprintf(fdname, sizeof(fdname), "%d", fd);
4910 kvm = kvm_create_vm(type, fdname);
4916 #ifdef CONFIG_KVM_MMIO
4917 r = kvm_coalesced_mmio_init(kvm);
4921 file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
4928 * Don't call kvm_put_kvm anymore at this point; file->f_op is
4929 * already set, with ->release() being kvm_vm_release(). In error
4930 * cases it will be called by the final fput(file) and will take
4931 * care of doing kvm_put_kvm(kvm).
4933 kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm);
4935 fd_install(fd, file);
4945 static long kvm_dev_ioctl(struct file *filp,
4946 unsigned int ioctl, unsigned long arg)
4951 case KVM_GET_API_VERSION:
4954 r = KVM_API_VERSION;
4957 r = kvm_dev_ioctl_create_vm(arg);
4959 case KVM_CHECK_EXTENSION:
4960 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
4962 case KVM_GET_VCPU_MMAP_SIZE:
4965 r = PAGE_SIZE; /* struct kvm_run */
4967 r += PAGE_SIZE; /* pio data page */
4969 #ifdef CONFIG_KVM_MMIO
4970 r += PAGE_SIZE; /* coalesced mmio ring page */
4973 case KVM_TRACE_ENABLE:
4974 case KVM_TRACE_PAUSE:
4975 case KVM_TRACE_DISABLE:
4979 return kvm_arch_dev_ioctl(filp, ioctl, arg);
4985 static struct file_operations kvm_chardev_ops = {
4986 .unlocked_ioctl = kvm_dev_ioctl,
4987 .llseek = noop_llseek,
4988 KVM_COMPAT(kvm_dev_ioctl),
4991 static struct miscdevice kvm_dev = {
4997 static void hardware_enable_nolock(void *junk)
4999 int cpu = raw_smp_processor_id();
5002 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
5005 cpumask_set_cpu(cpu, cpus_hardware_enabled);
5007 r = kvm_arch_hardware_enable();
5010 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
5011 atomic_inc(&hardware_enable_failed);
5012 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
5016 static int kvm_starting_cpu(unsigned int cpu)
5018 raw_spin_lock(&kvm_count_lock);
5019 if (kvm_usage_count)
5020 hardware_enable_nolock(NULL);
5021 raw_spin_unlock(&kvm_count_lock);
5025 static void hardware_disable_nolock(void *junk)
5027 int cpu = raw_smp_processor_id();
5029 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
5031 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
5032 kvm_arch_hardware_disable();
5035 static int kvm_dying_cpu(unsigned int cpu)
5037 raw_spin_lock(&kvm_count_lock);
5038 if (kvm_usage_count)
5039 hardware_disable_nolock(NULL);
5040 raw_spin_unlock(&kvm_count_lock);
5044 static void hardware_disable_all_nolock(void)
5046 BUG_ON(!kvm_usage_count);
5049 if (!kvm_usage_count)
5050 on_each_cpu(hardware_disable_nolock, NULL, 1);
5053 static void hardware_disable_all(void)
5055 raw_spin_lock(&kvm_count_lock);
5056 hardware_disable_all_nolock();
5057 raw_spin_unlock(&kvm_count_lock);
5060 static int hardware_enable_all(void)
5064 raw_spin_lock(&kvm_count_lock);
5067 if (kvm_usage_count == 1) {
5068 atomic_set(&hardware_enable_failed, 0);
5069 on_each_cpu(hardware_enable_nolock, NULL, 1);
5071 if (atomic_read(&hardware_enable_failed)) {
5072 hardware_disable_all_nolock();
5077 raw_spin_unlock(&kvm_count_lock);
5082 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
5086 * Some (well, at least mine) BIOSes hang on reboot if
5089 * And Intel TXT required VMX off for all cpu when system shutdown.
5091 pr_info("kvm: exiting hardware virtualization\n");
5092 kvm_rebooting = true;
5093 on_each_cpu(hardware_disable_nolock, NULL, 1);
5097 static struct notifier_block kvm_reboot_notifier = {
5098 .notifier_call = kvm_reboot,
5102 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
5106 for (i = 0; i < bus->dev_count; i++) {
5107 struct kvm_io_device *pos = bus->range[i].dev;
5109 kvm_iodevice_destructor(pos);
5114 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
5115 const struct kvm_io_range *r2)
5117 gpa_t addr1 = r1->addr;
5118 gpa_t addr2 = r2->addr;
5123 /* If r2->len == 0, match the exact address. If r2->len != 0,
5124 * accept any overlapping write. Any order is acceptable for
5125 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
5126 * we process all of them.
5139 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
5141 return kvm_io_bus_cmp(p1, p2);
5144 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
5145 gpa_t addr, int len)
5147 struct kvm_io_range *range, key;
5150 key = (struct kvm_io_range) {
5155 range = bsearch(&key, bus->range, bus->dev_count,
5156 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
5160 off = range - bus->range;
5162 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
5168 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
5169 struct kvm_io_range *range, const void *val)
5173 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
5177 while (idx < bus->dev_count &&
5178 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
5179 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
5188 /* kvm_io_bus_write - called under kvm->slots_lock */
5189 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
5190 int len, const void *val)
5192 struct kvm_io_bus *bus;
5193 struct kvm_io_range range;
5196 range = (struct kvm_io_range) {
5201 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
5204 r = __kvm_io_bus_write(vcpu, bus, &range, val);
5205 return r < 0 ? r : 0;
5207 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
5209 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
5210 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
5211 gpa_t addr, int len, const void *val, long cookie)
5213 struct kvm_io_bus *bus;
5214 struct kvm_io_range range;
5216 range = (struct kvm_io_range) {
5221 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
5225 /* First try the device referenced by cookie. */
5226 if ((cookie >= 0) && (cookie < bus->dev_count) &&
5227 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
5228 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
5233 * cookie contained garbage; fall back to search and return the
5234 * correct cookie value.
5236 return __kvm_io_bus_write(vcpu, bus, &range, val);
5239 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
5240 struct kvm_io_range *range, void *val)
5244 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
5248 while (idx < bus->dev_count &&
5249 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
5250 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
5259 /* kvm_io_bus_read - called under kvm->slots_lock */
5260 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
5263 struct kvm_io_bus *bus;
5264 struct kvm_io_range range;
5267 range = (struct kvm_io_range) {
5272 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
5275 r = __kvm_io_bus_read(vcpu, bus, &range, val);
5276 return r < 0 ? r : 0;
5279 /* Caller must hold slots_lock. */
5280 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
5281 int len, struct kvm_io_device *dev)
5284 struct kvm_io_bus *new_bus, *bus;
5285 struct kvm_io_range range;
5287 bus = kvm_get_bus(kvm, bus_idx);
5291 /* exclude ioeventfd which is limited by maximum fd */
5292 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
5295 new_bus = kmalloc(struct_size(bus, range, bus->dev_count + 1),
5296 GFP_KERNEL_ACCOUNT);
5300 range = (struct kvm_io_range) {
5306 for (i = 0; i < bus->dev_count; i++)
5307 if (kvm_io_bus_cmp(&bus->range[i], &range) > 0)
5310 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
5311 new_bus->dev_count++;
5312 new_bus->range[i] = range;
5313 memcpy(new_bus->range + i + 1, bus->range + i,
5314 (bus->dev_count - i) * sizeof(struct kvm_io_range));
5315 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
5316 synchronize_srcu_expedited(&kvm->srcu);
5322 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
5323 struct kvm_io_device *dev)
5326 struct kvm_io_bus *new_bus, *bus;
5328 lockdep_assert_held(&kvm->slots_lock);
5330 bus = kvm_get_bus(kvm, bus_idx);
5334 for (i = 0; i < bus->dev_count; i++) {
5335 if (bus->range[i].dev == dev) {
5340 if (i == bus->dev_count)
5343 new_bus = kmalloc(struct_size(bus, range, bus->dev_count - 1),
5344 GFP_KERNEL_ACCOUNT);
5346 memcpy(new_bus, bus, struct_size(bus, range, i));
5347 new_bus->dev_count--;
5348 memcpy(new_bus->range + i, bus->range + i + 1,
5349 flex_array_size(new_bus, range, new_bus->dev_count - i));
5352 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
5353 synchronize_srcu_expedited(&kvm->srcu);
5355 /* Destroy the old bus _after_ installing the (null) bus. */
5357 pr_err("kvm: failed to shrink bus, removing it completely\n");
5358 for (j = 0; j < bus->dev_count; j++) {
5361 kvm_iodevice_destructor(bus->range[j].dev);
5366 return new_bus ? 0 : -ENOMEM;
5369 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
5372 struct kvm_io_bus *bus;
5373 int dev_idx, srcu_idx;
5374 struct kvm_io_device *iodev = NULL;
5376 srcu_idx = srcu_read_lock(&kvm->srcu);
5378 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
5382 dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
5386 iodev = bus->range[dev_idx].dev;
5389 srcu_read_unlock(&kvm->srcu, srcu_idx);
5393 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
5395 static int kvm_debugfs_open(struct inode *inode, struct file *file,
5396 int (*get)(void *, u64 *), int (*set)(void *, u64),
5399 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
5403 * The debugfs files are a reference to the kvm struct which
5404 * is still valid when kvm_destroy_vm is called. kvm_get_kvm_safe
5405 * avoids the race between open and the removal of the debugfs directory.
5407 if (!kvm_get_kvm_safe(stat_data->kvm))
5410 if (simple_attr_open(inode, file, get,
5411 kvm_stats_debugfs_mode(stat_data->desc) & 0222
5414 kvm_put_kvm(stat_data->kvm);
5421 static int kvm_debugfs_release(struct inode *inode, struct file *file)
5423 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
5426 simple_attr_release(inode, file);
5427 kvm_put_kvm(stat_data->kvm);
5432 static int kvm_get_stat_per_vm(struct kvm *kvm, size_t offset, u64 *val)
5434 *val = *(u64 *)((void *)(&kvm->stat) + offset);
5439 static int kvm_clear_stat_per_vm(struct kvm *kvm, size_t offset)
5441 *(u64 *)((void *)(&kvm->stat) + offset) = 0;
5446 static int kvm_get_stat_per_vcpu(struct kvm *kvm, size_t offset, u64 *val)
5449 struct kvm_vcpu *vcpu;
5453 kvm_for_each_vcpu(i, vcpu, kvm)
5454 *val += *(u64 *)((void *)(&vcpu->stat) + offset);
5459 static int kvm_clear_stat_per_vcpu(struct kvm *kvm, size_t offset)
5462 struct kvm_vcpu *vcpu;
5464 kvm_for_each_vcpu(i, vcpu, kvm)
5465 *(u64 *)((void *)(&vcpu->stat) + offset) = 0;
5470 static int kvm_stat_data_get(void *data, u64 *val)
5473 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
5475 switch (stat_data->kind) {
5477 r = kvm_get_stat_per_vm(stat_data->kvm,
5478 stat_data->desc->desc.offset, val);
5481 r = kvm_get_stat_per_vcpu(stat_data->kvm,
5482 stat_data->desc->desc.offset, val);
5489 static int kvm_stat_data_clear(void *data, u64 val)
5492 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
5497 switch (stat_data->kind) {
5499 r = kvm_clear_stat_per_vm(stat_data->kvm,
5500 stat_data->desc->desc.offset);
5503 r = kvm_clear_stat_per_vcpu(stat_data->kvm,
5504 stat_data->desc->desc.offset);
5511 static int kvm_stat_data_open(struct inode *inode, struct file *file)
5513 __simple_attr_check_format("%llu\n", 0ull);
5514 return kvm_debugfs_open(inode, file, kvm_stat_data_get,
5515 kvm_stat_data_clear, "%llu\n");
5518 static const struct file_operations stat_fops_per_vm = {
5519 .owner = THIS_MODULE,
5520 .open = kvm_stat_data_open,
5521 .release = kvm_debugfs_release,
5522 .read = simple_attr_read,
5523 .write = simple_attr_write,
5524 .llseek = no_llseek,
5527 static int vm_stat_get(void *_offset, u64 *val)
5529 unsigned offset = (long)_offset;
5534 mutex_lock(&kvm_lock);
5535 list_for_each_entry(kvm, &vm_list, vm_list) {
5536 kvm_get_stat_per_vm(kvm, offset, &tmp_val);
5539 mutex_unlock(&kvm_lock);
5543 static int vm_stat_clear(void *_offset, u64 val)
5545 unsigned offset = (long)_offset;
5551 mutex_lock(&kvm_lock);
5552 list_for_each_entry(kvm, &vm_list, vm_list) {
5553 kvm_clear_stat_per_vm(kvm, offset);
5555 mutex_unlock(&kvm_lock);
5560 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n");
5561 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_readonly_fops, vm_stat_get, NULL, "%llu\n");
5563 static int vcpu_stat_get(void *_offset, u64 *val)
5565 unsigned offset = (long)_offset;
5570 mutex_lock(&kvm_lock);
5571 list_for_each_entry(kvm, &vm_list, vm_list) {
5572 kvm_get_stat_per_vcpu(kvm, offset, &tmp_val);
5575 mutex_unlock(&kvm_lock);
5579 static int vcpu_stat_clear(void *_offset, u64 val)
5581 unsigned offset = (long)_offset;
5587 mutex_lock(&kvm_lock);
5588 list_for_each_entry(kvm, &vm_list, vm_list) {
5589 kvm_clear_stat_per_vcpu(kvm, offset);
5591 mutex_unlock(&kvm_lock);
5596 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear,
5598 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_readonly_fops, vcpu_stat_get, NULL, "%llu\n");
5600 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm)
5602 struct kobj_uevent_env *env;
5603 unsigned long long created, active;
5605 if (!kvm_dev.this_device || !kvm)
5608 mutex_lock(&kvm_lock);
5609 if (type == KVM_EVENT_CREATE_VM) {
5610 kvm_createvm_count++;
5612 } else if (type == KVM_EVENT_DESTROY_VM) {
5615 created = kvm_createvm_count;
5616 active = kvm_active_vms;
5617 mutex_unlock(&kvm_lock);
5619 env = kzalloc(sizeof(*env), GFP_KERNEL_ACCOUNT);
5623 add_uevent_var(env, "CREATED=%llu", created);
5624 add_uevent_var(env, "COUNT=%llu", active);
5626 if (type == KVM_EVENT_CREATE_VM) {
5627 add_uevent_var(env, "EVENT=create");
5628 kvm->userspace_pid = task_pid_nr(current);
5629 } else if (type == KVM_EVENT_DESTROY_VM) {
5630 add_uevent_var(env, "EVENT=destroy");
5632 add_uevent_var(env, "PID=%d", kvm->userspace_pid);
5634 if (!IS_ERR(kvm->debugfs_dentry)) {
5635 char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL_ACCOUNT);
5638 tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX);
5640 add_uevent_var(env, "STATS_PATH=%s", tmp);
5644 /* no need for checks, since we are adding at most only 5 keys */
5645 env->envp[env->envp_idx++] = NULL;
5646 kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp);
5650 static void kvm_init_debug(void)
5652 const struct file_operations *fops;
5653 const struct _kvm_stats_desc *pdesc;
5656 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
5658 for (i = 0; i < kvm_vm_stats_header.num_desc; ++i) {
5659 pdesc = &kvm_vm_stats_desc[i];
5660 if (kvm_stats_debugfs_mode(pdesc) & 0222)
5661 fops = &vm_stat_fops;
5663 fops = &vm_stat_readonly_fops;
5664 debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
5666 (void *)(long)pdesc->desc.offset, fops);
5669 for (i = 0; i < kvm_vcpu_stats_header.num_desc; ++i) {
5670 pdesc = &kvm_vcpu_stats_desc[i];
5671 if (kvm_stats_debugfs_mode(pdesc) & 0222)
5672 fops = &vcpu_stat_fops;
5674 fops = &vcpu_stat_readonly_fops;
5675 debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
5677 (void *)(long)pdesc->desc.offset, fops);
5681 static int kvm_suspend(void)
5683 if (kvm_usage_count)
5684 hardware_disable_nolock(NULL);
5688 static void kvm_resume(void)
5690 if (kvm_usage_count) {
5691 lockdep_assert_not_held(&kvm_count_lock);
5692 hardware_enable_nolock(NULL);
5696 static struct syscore_ops kvm_syscore_ops = {
5697 .suspend = kvm_suspend,
5698 .resume = kvm_resume,
5702 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
5704 return container_of(pn, struct kvm_vcpu, preempt_notifier);
5707 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
5709 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
5711 WRITE_ONCE(vcpu->preempted, false);
5712 WRITE_ONCE(vcpu->ready, false);
5714 __this_cpu_write(kvm_running_vcpu, vcpu);
5715 kvm_arch_sched_in(vcpu, cpu);
5716 kvm_arch_vcpu_load(vcpu, cpu);
5719 static void kvm_sched_out(struct preempt_notifier *pn,
5720 struct task_struct *next)
5722 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
5724 if (current->on_rq) {
5725 WRITE_ONCE(vcpu->preempted, true);
5726 WRITE_ONCE(vcpu->ready, true);
5728 kvm_arch_vcpu_put(vcpu);
5729 __this_cpu_write(kvm_running_vcpu, NULL);
5733 * kvm_get_running_vcpu - get the vcpu running on the current CPU.
5735 * We can disable preemption locally around accessing the per-CPU variable,
5736 * and use the resolved vcpu pointer after enabling preemption again,
5737 * because even if the current thread is migrated to another CPU, reading
5738 * the per-CPU value later will give us the same value as we update the
5739 * per-CPU variable in the preempt notifier handlers.
5741 struct kvm_vcpu *kvm_get_running_vcpu(void)
5743 struct kvm_vcpu *vcpu;
5746 vcpu = __this_cpu_read(kvm_running_vcpu);
5751 EXPORT_SYMBOL_GPL(kvm_get_running_vcpu);
5754 * kvm_get_running_vcpus - get the per-CPU array of currently running vcpus.
5756 struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void)
5758 return &kvm_running_vcpu;
5761 #ifdef CONFIG_GUEST_PERF_EVENTS
5762 static unsigned int kvm_guest_state(void)
5764 struct kvm_vcpu *vcpu = kvm_get_running_vcpu();
5767 if (!kvm_arch_pmi_in_guest(vcpu))
5770 state = PERF_GUEST_ACTIVE;
5771 if (!kvm_arch_vcpu_in_kernel(vcpu))
5772 state |= PERF_GUEST_USER;
5777 static unsigned long kvm_guest_get_ip(void)
5779 struct kvm_vcpu *vcpu = kvm_get_running_vcpu();
5781 /* Retrieving the IP must be guarded by a call to kvm_guest_state(). */
5782 if (WARN_ON_ONCE(!kvm_arch_pmi_in_guest(vcpu)))
5785 return kvm_arch_vcpu_get_ip(vcpu);
5788 static struct perf_guest_info_callbacks kvm_guest_cbs = {
5789 .state = kvm_guest_state,
5790 .get_ip = kvm_guest_get_ip,
5791 .handle_intel_pt_intr = NULL,
5794 void kvm_register_perf_callbacks(unsigned int (*pt_intr_handler)(void))
5796 kvm_guest_cbs.handle_intel_pt_intr = pt_intr_handler;
5797 perf_register_guest_info_callbacks(&kvm_guest_cbs);
5799 void kvm_unregister_perf_callbacks(void)
5801 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
5805 struct kvm_cpu_compat_check {
5810 static void check_processor_compat(void *data)
5812 struct kvm_cpu_compat_check *c = data;
5814 *c->ret = kvm_arch_check_processor_compat(c->opaque);
5817 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
5818 struct module *module)
5820 struct kvm_cpu_compat_check c;
5824 r = kvm_arch_init(opaque);
5829 * kvm_arch_init makes sure there's at most one caller
5830 * for architectures that support multiple implementations,
5831 * like intel and amd on x86.
5832 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
5833 * conflicts in case kvm is already setup for another implementation.
5835 r = kvm_irqfd_init();
5839 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
5844 r = kvm_arch_hardware_setup(opaque);
5850 for_each_online_cpu(cpu) {
5851 smp_call_function_single(cpu, check_processor_compat, &c, 1);
5856 r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting",
5857 kvm_starting_cpu, kvm_dying_cpu);
5860 register_reboot_notifier(&kvm_reboot_notifier);
5862 /* A kmem cache lets us meet the alignment requirements of fx_save. */
5864 vcpu_align = __alignof__(struct kvm_vcpu);
5866 kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align,
5868 offsetof(struct kvm_vcpu, arch),
5869 offsetofend(struct kvm_vcpu, stats_id)
5870 - offsetof(struct kvm_vcpu, arch),
5872 if (!kvm_vcpu_cache) {
5877 for_each_possible_cpu(cpu) {
5878 if (!alloc_cpumask_var_node(&per_cpu(cpu_kick_mask, cpu),
5879 GFP_KERNEL, cpu_to_node(cpu))) {
5885 r = kvm_async_pf_init();
5889 kvm_chardev_ops.owner = module;
5891 r = misc_register(&kvm_dev);
5893 pr_err("kvm: misc device register failed\n");
5897 register_syscore_ops(&kvm_syscore_ops);
5899 kvm_preempt_ops.sched_in = kvm_sched_in;
5900 kvm_preempt_ops.sched_out = kvm_sched_out;
5904 r = kvm_vfio_ops_init();
5910 kvm_async_pf_deinit();
5912 for_each_possible_cpu(cpu)
5913 free_cpumask_var(per_cpu(cpu_kick_mask, cpu));
5915 kmem_cache_destroy(kvm_vcpu_cache);
5917 unregister_reboot_notifier(&kvm_reboot_notifier);
5918 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
5920 kvm_arch_hardware_unsetup();
5922 free_cpumask_var(cpus_hardware_enabled);
5930 EXPORT_SYMBOL_GPL(kvm_init);
5936 debugfs_remove_recursive(kvm_debugfs_dir);
5937 misc_deregister(&kvm_dev);
5938 for_each_possible_cpu(cpu)
5939 free_cpumask_var(per_cpu(cpu_kick_mask, cpu));
5940 kmem_cache_destroy(kvm_vcpu_cache);
5941 kvm_async_pf_deinit();
5942 unregister_syscore_ops(&kvm_syscore_ops);
5943 unregister_reboot_notifier(&kvm_reboot_notifier);
5944 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
5945 on_each_cpu(hardware_disable_nolock, NULL, 1);
5946 kvm_arch_hardware_unsetup();
5949 free_cpumask_var(cpus_hardware_enabled);
5950 kvm_vfio_ops_exit();
5952 EXPORT_SYMBOL_GPL(kvm_exit);
5954 struct kvm_vm_worker_thread_context {
5956 struct task_struct *parent;
5957 struct completion init_done;
5958 kvm_vm_thread_fn_t thread_fn;
5963 static int kvm_vm_worker_thread(void *context)
5966 * The init_context is allocated on the stack of the parent thread, so
5967 * we have to locally copy anything that is needed beyond initialization
5969 struct kvm_vm_worker_thread_context *init_context = context;
5970 struct task_struct *parent;
5971 struct kvm *kvm = init_context->kvm;
5972 kvm_vm_thread_fn_t thread_fn = init_context->thread_fn;
5973 uintptr_t data = init_context->data;
5976 err = kthread_park(current);
5977 /* kthread_park(current) is never supposed to return an error */
5982 err = cgroup_attach_task_all(init_context->parent, current);
5984 kvm_err("%s: cgroup_attach_task_all failed with err %d\n",
5989 set_user_nice(current, task_nice(init_context->parent));
5992 init_context->err = err;
5993 complete(&init_context->init_done);
5994 init_context = NULL;
5999 /* Wait to be woken up by the spawner before proceeding. */
6002 if (!kthread_should_stop())
6003 err = thread_fn(kvm, data);
6007 * Move kthread back to its original cgroup to prevent it lingering in
6008 * the cgroup of the VM process, after the latter finishes its
6011 * kthread_stop() waits on the 'exited' completion condition which is
6012 * set in exit_mm(), via mm_release(), in do_exit(). However, the
6013 * kthread is removed from the cgroup in the cgroup_exit() which is
6014 * called after the exit_mm(). This causes the kthread_stop() to return
6015 * before the kthread actually quits the cgroup.
6018 parent = rcu_dereference(current->real_parent);
6019 get_task_struct(parent);
6021 cgroup_attach_task_all(parent, current);
6022 put_task_struct(parent);
6027 int kvm_vm_create_worker_thread(struct kvm *kvm, kvm_vm_thread_fn_t thread_fn,
6028 uintptr_t data, const char *name,
6029 struct task_struct **thread_ptr)
6031 struct kvm_vm_worker_thread_context init_context = {};
6032 struct task_struct *thread;
6035 init_context.kvm = kvm;
6036 init_context.parent = current;
6037 init_context.thread_fn = thread_fn;
6038 init_context.data = data;
6039 init_completion(&init_context.init_done);
6041 thread = kthread_run(kvm_vm_worker_thread, &init_context,
6042 "%s-%d", name, task_pid_nr(current));
6044 return PTR_ERR(thread);
6046 /* kthread_run is never supposed to return NULL */
6047 WARN_ON(thread == NULL);
6049 wait_for_completion(&init_context.init_done);
6051 if (!init_context.err)
6052 *thread_ptr = thread;
6054 return init_context.err;