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_pfn(kvm_pfn_t pfn)
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 (!pfn_valid(pfn) || WARN_ON_ONCE(!page_count(pfn_to_page(pfn))))
182 return is_zone_device_page(pfn_to_page(pfn));
185 bool kvm_is_reserved_pfn(kvm_pfn_t pfn)
188 * ZONE_DEVICE pages currently set PG_reserved, but from a refcounting
189 * perspective they are "normal" pages, albeit with slightly different
193 return PageReserved(pfn_to_page(pfn)) &&
195 !kvm_is_zone_device_pfn(pfn);
201 * Switches to specified vcpu, until a matching vcpu_put()
203 void vcpu_load(struct kvm_vcpu *vcpu)
207 __this_cpu_write(kvm_running_vcpu, vcpu);
208 preempt_notifier_register(&vcpu->preempt_notifier);
209 kvm_arch_vcpu_load(vcpu, cpu);
212 EXPORT_SYMBOL_GPL(vcpu_load);
214 void vcpu_put(struct kvm_vcpu *vcpu)
217 kvm_arch_vcpu_put(vcpu);
218 preempt_notifier_unregister(&vcpu->preempt_notifier);
219 __this_cpu_write(kvm_running_vcpu, NULL);
222 EXPORT_SYMBOL_GPL(vcpu_put);
224 /* TODO: merge with kvm_arch_vcpu_should_kick */
225 static bool kvm_request_needs_ipi(struct kvm_vcpu *vcpu, unsigned req)
227 int mode = kvm_vcpu_exiting_guest_mode(vcpu);
230 * We need to wait for the VCPU to reenable interrupts and get out of
231 * READING_SHADOW_PAGE_TABLES mode.
233 if (req & KVM_REQUEST_WAIT)
234 return mode != OUTSIDE_GUEST_MODE;
237 * Need to kick a running VCPU, but otherwise there is nothing to do.
239 return mode == IN_GUEST_MODE;
242 static void ack_kick(void *_completed)
246 static inline bool kvm_kick_many_cpus(struct cpumask *cpus, bool wait)
248 if (cpumask_empty(cpus))
251 smp_call_function_many(cpus, ack_kick, NULL, wait);
255 static void kvm_make_vcpu_request(struct kvm_vcpu *vcpu, unsigned int req,
256 struct cpumask *tmp, int current_cpu)
260 if (likely(!(req & KVM_REQUEST_NO_ACTION)))
261 __kvm_make_request(req, vcpu);
263 if (!(req & KVM_REQUEST_NO_WAKEUP) && kvm_vcpu_wake_up(vcpu))
267 * Note, the vCPU could get migrated to a different pCPU at any point
268 * after kvm_request_needs_ipi(), which could result in sending an IPI
269 * to the previous pCPU. But, that's OK because the purpose of the IPI
270 * is to ensure the vCPU returns to OUTSIDE_GUEST_MODE, which is
271 * satisfied if the vCPU migrates. Entering READING_SHADOW_PAGE_TABLES
272 * after this point is also OK, as the requirement is only that KVM wait
273 * for vCPUs that were reading SPTEs _before_ any changes were
274 * finalized. See kvm_vcpu_kick() for more details on handling requests.
276 if (kvm_request_needs_ipi(vcpu, req)) {
277 cpu = READ_ONCE(vcpu->cpu);
278 if (cpu != -1 && cpu != current_cpu)
279 __cpumask_set_cpu(cpu, tmp);
283 bool kvm_make_vcpus_request_mask(struct kvm *kvm, unsigned int req,
284 unsigned long *vcpu_bitmap)
286 struct kvm_vcpu *vcpu;
287 struct cpumask *cpus;
293 cpus = this_cpu_cpumask_var_ptr(cpu_kick_mask);
296 for_each_set_bit(i, vcpu_bitmap, KVM_MAX_VCPUS) {
297 vcpu = kvm_get_vcpu(kvm, i);
300 kvm_make_vcpu_request(vcpu, req, cpus, me);
303 called = kvm_kick_many_cpus(cpus, !!(req & KVM_REQUEST_WAIT));
309 bool kvm_make_all_cpus_request_except(struct kvm *kvm, unsigned int req,
310 struct kvm_vcpu *except)
312 struct kvm_vcpu *vcpu;
313 struct cpumask *cpus;
320 cpus = this_cpu_cpumask_var_ptr(cpu_kick_mask);
323 kvm_for_each_vcpu(i, vcpu, kvm) {
326 kvm_make_vcpu_request(vcpu, req, cpus, me);
329 called = kvm_kick_many_cpus(cpus, !!(req & KVM_REQUEST_WAIT));
335 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
337 return kvm_make_all_cpus_request_except(kvm, req, NULL);
339 EXPORT_SYMBOL_GPL(kvm_make_all_cpus_request);
341 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
342 void kvm_flush_remote_tlbs(struct kvm *kvm)
344 ++kvm->stat.generic.remote_tlb_flush_requests;
347 * We want to publish modifications to the page tables before reading
348 * mode. Pairs with a memory barrier in arch-specific code.
349 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
350 * and smp_mb in walk_shadow_page_lockless_begin/end.
351 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
353 * There is already an smp_mb__after_atomic() before
354 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
357 if (!kvm_arch_flush_remote_tlb(kvm)
358 || kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
359 ++kvm->stat.generic.remote_tlb_flush;
361 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
364 static void kvm_flush_shadow_all(struct kvm *kvm)
366 kvm_arch_flush_shadow_all(kvm);
367 kvm_arch_guest_memory_reclaimed(kvm);
370 #ifdef KVM_ARCH_NR_OBJS_PER_MEMORY_CACHE
371 static inline void *mmu_memory_cache_alloc_obj(struct kvm_mmu_memory_cache *mc,
374 gfp_flags |= mc->gfp_zero;
377 return kmem_cache_alloc(mc->kmem_cache, gfp_flags);
379 return (void *)__get_free_page(gfp_flags);
382 int kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache *mc, int min)
386 if (mc->nobjs >= min)
388 while (mc->nobjs < ARRAY_SIZE(mc->objects)) {
389 obj = mmu_memory_cache_alloc_obj(mc, GFP_KERNEL_ACCOUNT);
391 return mc->nobjs >= min ? 0 : -ENOMEM;
392 mc->objects[mc->nobjs++] = obj;
397 int kvm_mmu_memory_cache_nr_free_objects(struct kvm_mmu_memory_cache *mc)
402 void kvm_mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc)
406 kmem_cache_free(mc->kmem_cache, mc->objects[--mc->nobjs]);
408 free_page((unsigned long)mc->objects[--mc->nobjs]);
412 void *kvm_mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc)
416 if (WARN_ON(!mc->nobjs))
417 p = mmu_memory_cache_alloc_obj(mc, GFP_ATOMIC | __GFP_ACCOUNT);
419 p = mc->objects[--mc->nobjs];
425 static void kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
427 mutex_init(&vcpu->mutex);
432 #ifndef __KVM_HAVE_ARCH_WQP
433 rcuwait_init(&vcpu->wait);
435 kvm_async_pf_vcpu_init(vcpu);
437 kvm_vcpu_set_in_spin_loop(vcpu, false);
438 kvm_vcpu_set_dy_eligible(vcpu, false);
439 vcpu->preempted = false;
441 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
442 vcpu->last_used_slot = NULL;
445 static void kvm_vcpu_destroy(struct kvm_vcpu *vcpu)
447 kvm_arch_vcpu_destroy(vcpu);
448 kvm_dirty_ring_free(&vcpu->dirty_ring);
451 * No need for rcu_read_lock as VCPU_RUN is the only place that changes
452 * the vcpu->pid pointer, and at destruction time all file descriptors
455 put_pid(rcu_dereference_protected(vcpu->pid, 1));
457 free_page((unsigned long)vcpu->run);
458 kmem_cache_free(kvm_vcpu_cache, vcpu);
461 void kvm_destroy_vcpus(struct kvm *kvm)
464 struct kvm_vcpu *vcpu;
466 kvm_for_each_vcpu(i, vcpu, kvm) {
467 kvm_vcpu_destroy(vcpu);
468 xa_erase(&kvm->vcpu_array, i);
471 atomic_set(&kvm->online_vcpus, 0);
473 EXPORT_SYMBOL_GPL(kvm_destroy_vcpus);
475 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
476 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
478 return container_of(mn, struct kvm, mmu_notifier);
481 static void kvm_mmu_notifier_invalidate_range(struct mmu_notifier *mn,
482 struct mm_struct *mm,
483 unsigned long start, unsigned long end)
485 struct kvm *kvm = mmu_notifier_to_kvm(mn);
488 idx = srcu_read_lock(&kvm->srcu);
489 kvm_arch_mmu_notifier_invalidate_range(kvm, start, end);
490 srcu_read_unlock(&kvm->srcu, idx);
493 typedef bool (*hva_handler_t)(struct kvm *kvm, struct kvm_gfn_range *range);
495 typedef void (*on_lock_fn_t)(struct kvm *kvm, unsigned long start,
498 typedef void (*on_unlock_fn_t)(struct kvm *kvm);
500 struct kvm_hva_range {
504 hva_handler_t handler;
505 on_lock_fn_t on_lock;
506 on_unlock_fn_t on_unlock;
512 * Use a dedicated stub instead of NULL to indicate that there is no callback
513 * function/handler. The compiler technically can't guarantee that a real
514 * function will have a non-zero address, and so it will generate code to
515 * check for !NULL, whereas comparing against a stub will be elided at compile
516 * time (unless the compiler is getting long in the tooth, e.g. gcc 4.9).
518 static void kvm_null_fn(void)
522 #define IS_KVM_NULL_FN(fn) ((fn) == (void *)kvm_null_fn)
524 /* Iterate over each memslot intersecting [start, last] (inclusive) range */
525 #define kvm_for_each_memslot_in_hva_range(node, slots, start, last) \
526 for (node = interval_tree_iter_first(&slots->hva_tree, start, last); \
528 node = interval_tree_iter_next(node, start, last)) \
530 static __always_inline int __kvm_handle_hva_range(struct kvm *kvm,
531 const struct kvm_hva_range *range)
533 bool ret = false, locked = false;
534 struct kvm_gfn_range gfn_range;
535 struct kvm_memory_slot *slot;
536 struct kvm_memslots *slots;
539 if (WARN_ON_ONCE(range->end <= range->start))
542 /* A null handler is allowed if and only if on_lock() is provided. */
543 if (WARN_ON_ONCE(IS_KVM_NULL_FN(range->on_lock) &&
544 IS_KVM_NULL_FN(range->handler)))
547 idx = srcu_read_lock(&kvm->srcu);
549 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
550 struct interval_tree_node *node;
552 slots = __kvm_memslots(kvm, i);
553 kvm_for_each_memslot_in_hva_range(node, slots,
554 range->start, range->end - 1) {
555 unsigned long hva_start, hva_end;
557 slot = container_of(node, struct kvm_memory_slot, hva_node[slots->node_idx]);
558 hva_start = max(range->start, slot->userspace_addr);
559 hva_end = min(range->end, slot->userspace_addr +
560 (slot->npages << PAGE_SHIFT));
563 * To optimize for the likely case where the address
564 * range is covered by zero or one memslots, don't
565 * bother making these conditional (to avoid writes on
566 * the second or later invocation of the handler).
568 gfn_range.pte = range->pte;
569 gfn_range.may_block = range->may_block;
572 * {gfn(page) | page intersects with [hva_start, hva_end)} =
573 * {gfn_start, gfn_start+1, ..., gfn_end-1}.
575 gfn_range.start = hva_to_gfn_memslot(hva_start, slot);
576 gfn_range.end = hva_to_gfn_memslot(hva_end + PAGE_SIZE - 1, slot);
577 gfn_range.slot = slot;
582 if (!IS_KVM_NULL_FN(range->on_lock))
583 range->on_lock(kvm, range->start, range->end);
584 if (IS_KVM_NULL_FN(range->handler))
587 ret |= range->handler(kvm, &gfn_range);
591 if (range->flush_on_ret && ret)
592 kvm_flush_remote_tlbs(kvm);
596 if (!IS_KVM_NULL_FN(range->on_unlock))
597 range->on_unlock(kvm);
600 srcu_read_unlock(&kvm->srcu, idx);
602 /* The notifiers are averse to booleans. :-( */
606 static __always_inline int kvm_handle_hva_range(struct mmu_notifier *mn,
610 hva_handler_t handler)
612 struct kvm *kvm = mmu_notifier_to_kvm(mn);
613 const struct kvm_hva_range range = {
618 .on_lock = (void *)kvm_null_fn,
619 .on_unlock = (void *)kvm_null_fn,
620 .flush_on_ret = true,
624 return __kvm_handle_hva_range(kvm, &range);
627 static __always_inline int kvm_handle_hva_range_no_flush(struct mmu_notifier *mn,
630 hva_handler_t handler)
632 struct kvm *kvm = mmu_notifier_to_kvm(mn);
633 const struct kvm_hva_range range = {
638 .on_lock = (void *)kvm_null_fn,
639 .on_unlock = (void *)kvm_null_fn,
640 .flush_on_ret = false,
644 return __kvm_handle_hva_range(kvm, &range);
646 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
647 struct mm_struct *mm,
648 unsigned long address,
651 struct kvm *kvm = mmu_notifier_to_kvm(mn);
653 trace_kvm_set_spte_hva(address);
656 * .change_pte() must be surrounded by .invalidate_range_{start,end}().
657 * If mmu_notifier_count is zero, then no in-progress invalidations,
658 * including this one, found a relevant memslot at start(); rechecking
659 * memslots here is unnecessary. Note, a false positive (count elevated
660 * by a different invalidation) is sub-optimal but functionally ok.
662 WARN_ON_ONCE(!READ_ONCE(kvm->mn_active_invalidate_count));
663 if (!READ_ONCE(kvm->mmu_notifier_count))
666 kvm_handle_hva_range(mn, address, address + 1, pte, kvm_set_spte_gfn);
669 void kvm_inc_notifier_count(struct kvm *kvm, unsigned long start,
673 * The count increase must become visible at unlock time as no
674 * spte can be established without taking the mmu_lock and
675 * count is also read inside the mmu_lock critical section.
677 kvm->mmu_notifier_count++;
678 if (likely(kvm->mmu_notifier_count == 1)) {
679 kvm->mmu_notifier_range_start = start;
680 kvm->mmu_notifier_range_end = end;
683 * Fully tracking multiple concurrent ranges has diminishing
684 * returns. Keep things simple and just find the minimal range
685 * which includes the current and new ranges. As there won't be
686 * enough information to subtract a range after its invalidate
687 * completes, any ranges invalidated concurrently will
688 * accumulate and persist until all outstanding invalidates
691 kvm->mmu_notifier_range_start =
692 min(kvm->mmu_notifier_range_start, start);
693 kvm->mmu_notifier_range_end =
694 max(kvm->mmu_notifier_range_end, end);
698 static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
699 const struct mmu_notifier_range *range)
701 struct kvm *kvm = mmu_notifier_to_kvm(mn);
702 const struct kvm_hva_range hva_range = {
703 .start = range->start,
706 .handler = kvm_unmap_gfn_range,
707 .on_lock = kvm_inc_notifier_count,
708 .on_unlock = kvm_arch_guest_memory_reclaimed,
709 .flush_on_ret = true,
710 .may_block = mmu_notifier_range_blockable(range),
713 trace_kvm_unmap_hva_range(range->start, range->end);
716 * Prevent memslot modification between range_start() and range_end()
717 * so that conditionally locking provides the same result in both
718 * functions. Without that guarantee, the mmu_notifier_count
719 * adjustments will be imbalanced.
721 * Pairs with the decrement in range_end().
723 spin_lock(&kvm->mn_invalidate_lock);
724 kvm->mn_active_invalidate_count++;
725 spin_unlock(&kvm->mn_invalidate_lock);
728 * Invalidate pfn caches _before_ invalidating the secondary MMUs, i.e.
729 * before acquiring mmu_lock, to avoid holding mmu_lock while acquiring
730 * each cache's lock. There are relatively few caches in existence at
731 * any given time, and the caches themselves can check for hva overlap,
732 * i.e. don't need to rely on memslot overlap checks for performance.
733 * Because this runs without holding mmu_lock, the pfn caches must use
734 * mn_active_invalidate_count (see above) instead of mmu_notifier_count.
736 gfn_to_pfn_cache_invalidate_start(kvm, range->start, range->end,
737 hva_range.may_block);
739 __kvm_handle_hva_range(kvm, &hva_range);
744 void kvm_dec_notifier_count(struct kvm *kvm, unsigned long start,
748 * This sequence increase will notify the kvm page fault that
749 * the page that is going to be mapped in the spte could have
752 kvm->mmu_notifier_seq++;
755 * The above sequence increase must be visible before the
756 * below count decrease, which is ensured by the smp_wmb above
757 * in conjunction with the smp_rmb in mmu_notifier_retry().
759 kvm->mmu_notifier_count--;
762 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
763 const struct mmu_notifier_range *range)
765 struct kvm *kvm = mmu_notifier_to_kvm(mn);
766 const struct kvm_hva_range hva_range = {
767 .start = range->start,
770 .handler = (void *)kvm_null_fn,
771 .on_lock = kvm_dec_notifier_count,
772 .on_unlock = (void *)kvm_null_fn,
773 .flush_on_ret = false,
774 .may_block = mmu_notifier_range_blockable(range),
778 __kvm_handle_hva_range(kvm, &hva_range);
780 /* Pairs with the increment in range_start(). */
781 spin_lock(&kvm->mn_invalidate_lock);
782 wake = (--kvm->mn_active_invalidate_count == 0);
783 spin_unlock(&kvm->mn_invalidate_lock);
786 * There can only be one waiter, since the wait happens under
790 rcuwait_wake_up(&kvm->mn_memslots_update_rcuwait);
792 BUG_ON(kvm->mmu_notifier_count < 0);
795 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
796 struct mm_struct *mm,
800 trace_kvm_age_hva(start, end);
802 return kvm_handle_hva_range(mn, start, end, __pte(0), kvm_age_gfn);
805 static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
806 struct mm_struct *mm,
810 trace_kvm_age_hva(start, end);
813 * Even though we do not flush TLB, this will still adversely
814 * affect performance on pre-Haswell Intel EPT, where there is
815 * no EPT Access Bit to clear so that we have to tear down EPT
816 * tables instead. If we find this unacceptable, we can always
817 * add a parameter to kvm_age_hva so that it effectively doesn't
818 * do anything on clear_young.
820 * Also note that currently we never issue secondary TLB flushes
821 * from clear_young, leaving this job up to the regular system
822 * cadence. If we find this inaccurate, we might come up with a
823 * more sophisticated heuristic later.
825 return kvm_handle_hva_range_no_flush(mn, start, end, kvm_age_gfn);
828 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
829 struct mm_struct *mm,
830 unsigned long address)
832 trace_kvm_test_age_hva(address);
834 return kvm_handle_hva_range_no_flush(mn, address, address + 1,
838 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
839 struct mm_struct *mm)
841 struct kvm *kvm = mmu_notifier_to_kvm(mn);
844 idx = srcu_read_lock(&kvm->srcu);
845 kvm_flush_shadow_all(kvm);
846 srcu_read_unlock(&kvm->srcu, idx);
849 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
850 .invalidate_range = kvm_mmu_notifier_invalidate_range,
851 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
852 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
853 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
854 .clear_young = kvm_mmu_notifier_clear_young,
855 .test_young = kvm_mmu_notifier_test_young,
856 .change_pte = kvm_mmu_notifier_change_pte,
857 .release = kvm_mmu_notifier_release,
860 static int kvm_init_mmu_notifier(struct kvm *kvm)
862 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
863 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
866 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
868 static int kvm_init_mmu_notifier(struct kvm *kvm)
873 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
875 #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER
876 static int kvm_pm_notifier_call(struct notifier_block *bl,
880 struct kvm *kvm = container_of(bl, struct kvm, pm_notifier);
882 return kvm_arch_pm_notifier(kvm, state);
885 static void kvm_init_pm_notifier(struct kvm *kvm)
887 kvm->pm_notifier.notifier_call = kvm_pm_notifier_call;
888 /* Suspend KVM before we suspend ftrace, RCU, etc. */
889 kvm->pm_notifier.priority = INT_MAX;
890 register_pm_notifier(&kvm->pm_notifier);
893 static void kvm_destroy_pm_notifier(struct kvm *kvm)
895 unregister_pm_notifier(&kvm->pm_notifier);
897 #else /* !CONFIG_HAVE_KVM_PM_NOTIFIER */
898 static void kvm_init_pm_notifier(struct kvm *kvm)
902 static void kvm_destroy_pm_notifier(struct kvm *kvm)
905 #endif /* CONFIG_HAVE_KVM_PM_NOTIFIER */
907 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
909 if (!memslot->dirty_bitmap)
912 kvfree(memslot->dirty_bitmap);
913 memslot->dirty_bitmap = NULL;
916 /* This does not remove the slot from struct kvm_memslots data structures */
917 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
919 kvm_destroy_dirty_bitmap(slot);
921 kvm_arch_free_memslot(kvm, slot);
926 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
928 struct hlist_node *idnode;
929 struct kvm_memory_slot *memslot;
933 * The same memslot objects live in both active and inactive sets,
934 * arbitrarily free using index '1' so the second invocation of this
935 * function isn't operating over a structure with dangling pointers
936 * (even though this function isn't actually touching them).
938 if (!slots->node_idx)
941 hash_for_each_safe(slots->id_hash, bkt, idnode, memslot, id_node[1])
942 kvm_free_memslot(kvm, memslot);
945 static umode_t kvm_stats_debugfs_mode(const struct _kvm_stats_desc *pdesc)
947 switch (pdesc->desc.flags & KVM_STATS_TYPE_MASK) {
948 case KVM_STATS_TYPE_INSTANT:
950 case KVM_STATS_TYPE_CUMULATIVE:
951 case KVM_STATS_TYPE_PEAK:
958 static void kvm_destroy_vm_debugfs(struct kvm *kvm)
961 int kvm_debugfs_num_entries = kvm_vm_stats_header.num_desc +
962 kvm_vcpu_stats_header.num_desc;
964 if (IS_ERR(kvm->debugfs_dentry))
967 debugfs_remove_recursive(kvm->debugfs_dentry);
969 if (kvm->debugfs_stat_data) {
970 for (i = 0; i < kvm_debugfs_num_entries; i++)
971 kfree(kvm->debugfs_stat_data[i]);
972 kfree(kvm->debugfs_stat_data);
976 static int kvm_create_vm_debugfs(struct kvm *kvm, int fd)
978 static DEFINE_MUTEX(kvm_debugfs_lock);
980 char dir_name[ITOA_MAX_LEN * 2];
981 struct kvm_stat_data *stat_data;
982 const struct _kvm_stats_desc *pdesc;
984 int kvm_debugfs_num_entries = kvm_vm_stats_header.num_desc +
985 kvm_vcpu_stats_header.num_desc;
987 if (!debugfs_initialized())
990 snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd);
991 mutex_lock(&kvm_debugfs_lock);
992 dent = debugfs_lookup(dir_name, kvm_debugfs_dir);
994 pr_warn_ratelimited("KVM: debugfs: duplicate directory %s\n", dir_name);
996 mutex_unlock(&kvm_debugfs_lock);
999 dent = debugfs_create_dir(dir_name, kvm_debugfs_dir);
1000 mutex_unlock(&kvm_debugfs_lock);
1004 kvm->debugfs_dentry = dent;
1005 kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries,
1006 sizeof(*kvm->debugfs_stat_data),
1007 GFP_KERNEL_ACCOUNT);
1008 if (!kvm->debugfs_stat_data)
1011 for (i = 0; i < kvm_vm_stats_header.num_desc; ++i) {
1012 pdesc = &kvm_vm_stats_desc[i];
1013 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT);
1017 stat_data->kvm = kvm;
1018 stat_data->desc = pdesc;
1019 stat_data->kind = KVM_STAT_VM;
1020 kvm->debugfs_stat_data[i] = stat_data;
1021 debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
1022 kvm->debugfs_dentry, stat_data,
1026 for (i = 0; i < kvm_vcpu_stats_header.num_desc; ++i) {
1027 pdesc = &kvm_vcpu_stats_desc[i];
1028 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT);
1032 stat_data->kvm = kvm;
1033 stat_data->desc = pdesc;
1034 stat_data->kind = KVM_STAT_VCPU;
1035 kvm->debugfs_stat_data[i + kvm_vm_stats_header.num_desc] = stat_data;
1036 debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
1037 kvm->debugfs_dentry, stat_data,
1041 ret = kvm_arch_create_vm_debugfs(kvm);
1043 kvm_destroy_vm_debugfs(kvm);
1051 * Called after the VM is otherwise initialized, but just before adding it to
1054 int __weak kvm_arch_post_init_vm(struct kvm *kvm)
1060 * Called just after removing the VM from the vm_list, but before doing any
1061 * other destruction.
1063 void __weak kvm_arch_pre_destroy_vm(struct kvm *kvm)
1068 * Called after per-vm debugfs created. When called kvm->debugfs_dentry should
1069 * be setup already, so we can create arch-specific debugfs entries under it.
1070 * Cleanup should be automatic done in kvm_destroy_vm_debugfs() recursively, so
1071 * a per-arch destroy interface is not needed.
1073 int __weak kvm_arch_create_vm_debugfs(struct kvm *kvm)
1078 static struct kvm *kvm_create_vm(unsigned long type)
1080 struct kvm *kvm = kvm_arch_alloc_vm();
1081 struct kvm_memslots *slots;
1086 return ERR_PTR(-ENOMEM);
1088 KVM_MMU_LOCK_INIT(kvm);
1089 mmgrab(current->mm);
1090 kvm->mm = current->mm;
1091 kvm_eventfd_init(kvm);
1092 mutex_init(&kvm->lock);
1093 mutex_init(&kvm->irq_lock);
1094 mutex_init(&kvm->slots_lock);
1095 mutex_init(&kvm->slots_arch_lock);
1096 spin_lock_init(&kvm->mn_invalidate_lock);
1097 rcuwait_init(&kvm->mn_memslots_update_rcuwait);
1098 xa_init(&kvm->vcpu_array);
1100 INIT_LIST_HEAD(&kvm->gpc_list);
1101 spin_lock_init(&kvm->gpc_lock);
1103 INIT_LIST_HEAD(&kvm->devices);
1104 kvm->max_vcpus = KVM_MAX_VCPUS;
1106 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
1109 * Force subsequent debugfs file creations to fail if the VM directory
1110 * is not created (by kvm_create_vm_debugfs()).
1112 kvm->debugfs_dentry = ERR_PTR(-ENOENT);
1114 if (init_srcu_struct(&kvm->srcu))
1115 goto out_err_no_srcu;
1116 if (init_srcu_struct(&kvm->irq_srcu))
1117 goto out_err_no_irq_srcu;
1119 refcount_set(&kvm->users_count, 1);
1120 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
1121 for (j = 0; j < 2; j++) {
1122 slots = &kvm->__memslots[i][j];
1124 atomic_long_set(&slots->last_used_slot, (unsigned long)NULL);
1125 slots->hva_tree = RB_ROOT_CACHED;
1126 slots->gfn_tree = RB_ROOT;
1127 hash_init(slots->id_hash);
1128 slots->node_idx = j;
1130 /* Generations must be different for each address space. */
1131 slots->generation = i;
1134 rcu_assign_pointer(kvm->memslots[i], &kvm->__memslots[i][0]);
1137 for (i = 0; i < KVM_NR_BUSES; i++) {
1138 rcu_assign_pointer(kvm->buses[i],
1139 kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL_ACCOUNT));
1141 goto out_err_no_arch_destroy_vm;
1144 kvm->max_halt_poll_ns = halt_poll_ns;
1146 r = kvm_arch_init_vm(kvm, type);
1148 goto out_err_no_arch_destroy_vm;
1150 r = hardware_enable_all();
1152 goto out_err_no_disable;
1154 #ifdef CONFIG_HAVE_KVM_IRQFD
1155 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
1158 r = kvm_init_mmu_notifier(kvm);
1160 goto out_err_no_mmu_notifier;
1162 r = kvm_arch_post_init_vm(kvm);
1166 mutex_lock(&kvm_lock);
1167 list_add(&kvm->vm_list, &vm_list);
1168 mutex_unlock(&kvm_lock);
1170 preempt_notifier_inc();
1171 kvm_init_pm_notifier(kvm);
1174 * When the fd passed to this ioctl() is opened it pins the module,
1175 * but try_module_get() also prevents getting a reference if the module
1176 * is in MODULE_STATE_GOING (e.g. if someone ran "rmmod --wait").
1178 if (!try_module_get(kvm_chardev_ops.owner)) {
1186 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1187 if (kvm->mmu_notifier.ops)
1188 mmu_notifier_unregister(&kvm->mmu_notifier, current->mm);
1190 out_err_no_mmu_notifier:
1191 hardware_disable_all();
1193 kvm_arch_destroy_vm(kvm);
1194 out_err_no_arch_destroy_vm:
1195 WARN_ON_ONCE(!refcount_dec_and_test(&kvm->users_count));
1196 for (i = 0; i < KVM_NR_BUSES; i++)
1197 kfree(kvm_get_bus(kvm, i));
1198 cleanup_srcu_struct(&kvm->irq_srcu);
1199 out_err_no_irq_srcu:
1200 cleanup_srcu_struct(&kvm->srcu);
1202 kvm_arch_free_vm(kvm);
1203 mmdrop(current->mm);
1207 static void kvm_destroy_devices(struct kvm *kvm)
1209 struct kvm_device *dev, *tmp;
1212 * We do not need to take the kvm->lock here, because nobody else
1213 * has a reference to the struct kvm at this point and therefore
1214 * cannot access the devices list anyhow.
1216 list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
1217 list_del(&dev->vm_node);
1218 dev->ops->destroy(dev);
1222 static void kvm_destroy_vm(struct kvm *kvm)
1225 struct mm_struct *mm = kvm->mm;
1227 kvm_destroy_pm_notifier(kvm);
1228 kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm);
1229 kvm_destroy_vm_debugfs(kvm);
1230 kvm_arch_sync_events(kvm);
1231 mutex_lock(&kvm_lock);
1232 list_del(&kvm->vm_list);
1233 mutex_unlock(&kvm_lock);
1234 kvm_arch_pre_destroy_vm(kvm);
1236 kvm_free_irq_routing(kvm);
1237 for (i = 0; i < KVM_NR_BUSES; i++) {
1238 struct kvm_io_bus *bus = kvm_get_bus(kvm, i);
1241 kvm_io_bus_destroy(bus);
1242 kvm->buses[i] = NULL;
1244 kvm_coalesced_mmio_free(kvm);
1245 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1246 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
1248 * At this point, pending calls to invalidate_range_start()
1249 * have completed but no more MMU notifiers will run, so
1250 * mn_active_invalidate_count may remain unbalanced.
1251 * No threads can be waiting in install_new_memslots as the
1252 * last reference on KVM has been dropped, but freeing
1253 * memslots would deadlock without this manual intervention.
1255 WARN_ON(rcuwait_active(&kvm->mn_memslots_update_rcuwait));
1256 kvm->mn_active_invalidate_count = 0;
1258 kvm_flush_shadow_all(kvm);
1260 kvm_arch_destroy_vm(kvm);
1261 kvm_destroy_devices(kvm);
1262 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
1263 kvm_free_memslots(kvm, &kvm->__memslots[i][0]);
1264 kvm_free_memslots(kvm, &kvm->__memslots[i][1]);
1266 cleanup_srcu_struct(&kvm->irq_srcu);
1267 cleanup_srcu_struct(&kvm->srcu);
1268 kvm_arch_free_vm(kvm);
1269 preempt_notifier_dec();
1270 hardware_disable_all();
1272 module_put(kvm_chardev_ops.owner);
1275 void kvm_get_kvm(struct kvm *kvm)
1277 refcount_inc(&kvm->users_count);
1279 EXPORT_SYMBOL_GPL(kvm_get_kvm);
1282 * Make sure the vm is not during destruction, which is a safe version of
1283 * kvm_get_kvm(). Return true if kvm referenced successfully, false otherwise.
1285 bool kvm_get_kvm_safe(struct kvm *kvm)
1287 return refcount_inc_not_zero(&kvm->users_count);
1289 EXPORT_SYMBOL_GPL(kvm_get_kvm_safe);
1291 void kvm_put_kvm(struct kvm *kvm)
1293 if (refcount_dec_and_test(&kvm->users_count))
1294 kvm_destroy_vm(kvm);
1296 EXPORT_SYMBOL_GPL(kvm_put_kvm);
1299 * Used to put a reference that was taken on behalf of an object associated
1300 * with a user-visible file descriptor, e.g. a vcpu or device, if installation
1301 * of the new file descriptor fails and the reference cannot be transferred to
1302 * its final owner. In such cases, the caller is still actively using @kvm and
1303 * will fail miserably if the refcount unexpectedly hits zero.
1305 void kvm_put_kvm_no_destroy(struct kvm *kvm)
1307 WARN_ON(refcount_dec_and_test(&kvm->users_count));
1309 EXPORT_SYMBOL_GPL(kvm_put_kvm_no_destroy);
1311 static int kvm_vm_release(struct inode *inode, struct file *filp)
1313 struct kvm *kvm = filp->private_data;
1315 kvm_irqfd_release(kvm);
1322 * Allocation size is twice as large as the actual dirty bitmap size.
1323 * See kvm_vm_ioctl_get_dirty_log() why this is needed.
1325 static int kvm_alloc_dirty_bitmap(struct kvm_memory_slot *memslot)
1327 unsigned long dirty_bytes = kvm_dirty_bitmap_bytes(memslot);
1329 memslot->dirty_bitmap = __vcalloc(2, dirty_bytes, GFP_KERNEL_ACCOUNT);
1330 if (!memslot->dirty_bitmap)
1336 static struct kvm_memslots *kvm_get_inactive_memslots(struct kvm *kvm, int as_id)
1338 struct kvm_memslots *active = __kvm_memslots(kvm, as_id);
1339 int node_idx_inactive = active->node_idx ^ 1;
1341 return &kvm->__memslots[as_id][node_idx_inactive];
1345 * Helper to get the address space ID when one of memslot pointers may be NULL.
1346 * This also serves as a sanity that at least one of the pointers is non-NULL,
1347 * and that their address space IDs don't diverge.
1349 static int kvm_memslots_get_as_id(struct kvm_memory_slot *a,
1350 struct kvm_memory_slot *b)
1352 if (WARN_ON_ONCE(!a && !b))
1360 WARN_ON_ONCE(a->as_id != b->as_id);
1364 static void kvm_insert_gfn_node(struct kvm_memslots *slots,
1365 struct kvm_memory_slot *slot)
1367 struct rb_root *gfn_tree = &slots->gfn_tree;
1368 struct rb_node **node, *parent;
1369 int idx = slots->node_idx;
1372 for (node = &gfn_tree->rb_node; *node; ) {
1373 struct kvm_memory_slot *tmp;
1375 tmp = container_of(*node, struct kvm_memory_slot, gfn_node[idx]);
1377 if (slot->base_gfn < tmp->base_gfn)
1378 node = &(*node)->rb_left;
1379 else if (slot->base_gfn > tmp->base_gfn)
1380 node = &(*node)->rb_right;
1385 rb_link_node(&slot->gfn_node[idx], parent, node);
1386 rb_insert_color(&slot->gfn_node[idx], gfn_tree);
1389 static void kvm_erase_gfn_node(struct kvm_memslots *slots,
1390 struct kvm_memory_slot *slot)
1392 rb_erase(&slot->gfn_node[slots->node_idx], &slots->gfn_tree);
1395 static void kvm_replace_gfn_node(struct kvm_memslots *slots,
1396 struct kvm_memory_slot *old,
1397 struct kvm_memory_slot *new)
1399 int idx = slots->node_idx;
1401 WARN_ON_ONCE(old->base_gfn != new->base_gfn);
1403 rb_replace_node(&old->gfn_node[idx], &new->gfn_node[idx],
1408 * Replace @old with @new in the inactive memslots.
1410 * With NULL @old this simply adds @new.
1411 * With NULL @new this simply removes @old.
1413 * If @new is non-NULL its hva_node[slots_idx] range has to be set
1416 static void kvm_replace_memslot(struct kvm *kvm,
1417 struct kvm_memory_slot *old,
1418 struct kvm_memory_slot *new)
1420 int as_id = kvm_memslots_get_as_id(old, new);
1421 struct kvm_memslots *slots = kvm_get_inactive_memslots(kvm, as_id);
1422 int idx = slots->node_idx;
1425 hash_del(&old->id_node[idx]);
1426 interval_tree_remove(&old->hva_node[idx], &slots->hva_tree);
1428 if ((long)old == atomic_long_read(&slots->last_used_slot))
1429 atomic_long_set(&slots->last_used_slot, (long)new);
1432 kvm_erase_gfn_node(slots, old);
1438 * Initialize @new's hva range. Do this even when replacing an @old
1439 * slot, kvm_copy_memslot() deliberately does not touch node data.
1441 new->hva_node[idx].start = new->userspace_addr;
1442 new->hva_node[idx].last = new->userspace_addr +
1443 (new->npages << PAGE_SHIFT) - 1;
1446 * (Re)Add the new memslot. There is no O(1) interval_tree_replace(),
1447 * hva_node needs to be swapped with remove+insert even though hva can't
1448 * change when replacing an existing slot.
1450 hash_add(slots->id_hash, &new->id_node[idx], new->id);
1451 interval_tree_insert(&new->hva_node[idx], &slots->hva_tree);
1454 * If the memslot gfn is unchanged, rb_replace_node() can be used to
1455 * switch the node in the gfn tree instead of removing the old and
1456 * inserting the new as two separate operations. Replacement is a
1457 * single O(1) operation versus two O(log(n)) operations for
1460 if (old && old->base_gfn == new->base_gfn) {
1461 kvm_replace_gfn_node(slots, old, new);
1464 kvm_erase_gfn_node(slots, old);
1465 kvm_insert_gfn_node(slots, new);
1469 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
1471 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
1473 #ifdef __KVM_HAVE_READONLY_MEM
1474 valid_flags |= KVM_MEM_READONLY;
1477 if (mem->flags & ~valid_flags)
1483 static void kvm_swap_active_memslots(struct kvm *kvm, int as_id)
1485 struct kvm_memslots *slots = kvm_get_inactive_memslots(kvm, as_id);
1487 /* Grab the generation from the activate memslots. */
1488 u64 gen = __kvm_memslots(kvm, as_id)->generation;
1490 WARN_ON(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS);
1491 slots->generation = gen | KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
1494 * Do not store the new memslots while there are invalidations in
1495 * progress, otherwise the locking in invalidate_range_start and
1496 * invalidate_range_end will be unbalanced.
1498 spin_lock(&kvm->mn_invalidate_lock);
1499 prepare_to_rcuwait(&kvm->mn_memslots_update_rcuwait);
1500 while (kvm->mn_active_invalidate_count) {
1501 set_current_state(TASK_UNINTERRUPTIBLE);
1502 spin_unlock(&kvm->mn_invalidate_lock);
1504 spin_lock(&kvm->mn_invalidate_lock);
1506 finish_rcuwait(&kvm->mn_memslots_update_rcuwait);
1507 rcu_assign_pointer(kvm->memslots[as_id], slots);
1508 spin_unlock(&kvm->mn_invalidate_lock);
1511 * Acquired in kvm_set_memslot. Must be released before synchronize
1512 * SRCU below in order to avoid deadlock with another thread
1513 * acquiring the slots_arch_lock in an srcu critical section.
1515 mutex_unlock(&kvm->slots_arch_lock);
1517 synchronize_srcu_expedited(&kvm->srcu);
1520 * Increment the new memslot generation a second time, dropping the
1521 * update in-progress flag and incrementing the generation based on
1522 * the number of address spaces. This provides a unique and easily
1523 * identifiable generation number while the memslots are in flux.
1525 gen = slots->generation & ~KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
1528 * Generations must be unique even across address spaces. We do not need
1529 * a global counter for that, instead the generation space is evenly split
1530 * across address spaces. For example, with two address spaces, address
1531 * space 0 will use generations 0, 2, 4, ... while address space 1 will
1532 * use generations 1, 3, 5, ...
1534 gen += KVM_ADDRESS_SPACE_NUM;
1536 kvm_arch_memslots_updated(kvm, gen);
1538 slots->generation = gen;
1541 static int kvm_prepare_memory_region(struct kvm *kvm,
1542 const struct kvm_memory_slot *old,
1543 struct kvm_memory_slot *new,
1544 enum kvm_mr_change change)
1549 * If dirty logging is disabled, nullify the bitmap; the old bitmap
1550 * will be freed on "commit". If logging is enabled in both old and
1551 * new, reuse the existing bitmap. If logging is enabled only in the
1552 * new and KVM isn't using a ring buffer, allocate and initialize a
1555 if (change != KVM_MR_DELETE) {
1556 if (!(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
1557 new->dirty_bitmap = NULL;
1558 else if (old && old->dirty_bitmap)
1559 new->dirty_bitmap = old->dirty_bitmap;
1560 else if (!kvm->dirty_ring_size) {
1561 r = kvm_alloc_dirty_bitmap(new);
1565 if (kvm_dirty_log_manual_protect_and_init_set(kvm))
1566 bitmap_set(new->dirty_bitmap, 0, new->npages);
1570 r = kvm_arch_prepare_memory_region(kvm, old, new, change);
1572 /* Free the bitmap on failure if it was allocated above. */
1573 if (r && new && new->dirty_bitmap && (!old || !old->dirty_bitmap))
1574 kvm_destroy_dirty_bitmap(new);
1579 static void kvm_commit_memory_region(struct kvm *kvm,
1580 struct kvm_memory_slot *old,
1581 const struct kvm_memory_slot *new,
1582 enum kvm_mr_change change)
1585 * Update the total number of memslot pages before calling the arch
1586 * hook so that architectures can consume the result directly.
1588 if (change == KVM_MR_DELETE)
1589 kvm->nr_memslot_pages -= old->npages;
1590 else if (change == KVM_MR_CREATE)
1591 kvm->nr_memslot_pages += new->npages;
1593 kvm_arch_commit_memory_region(kvm, old, new, change);
1597 /* Nothing more to do. */
1600 /* Free the old memslot and all its metadata. */
1601 kvm_free_memslot(kvm, old);
1604 case KVM_MR_FLAGS_ONLY:
1606 * Free the dirty bitmap as needed; the below check encompasses
1607 * both the flags and whether a ring buffer is being used)
1609 if (old->dirty_bitmap && !new->dirty_bitmap)
1610 kvm_destroy_dirty_bitmap(old);
1613 * The final quirk. Free the detached, old slot, but only its
1614 * memory, not any metadata. Metadata, including arch specific
1615 * data, may be reused by @new.
1625 * Activate @new, which must be installed in the inactive slots by the caller,
1626 * by swapping the active slots and then propagating @new to @old once @old is
1627 * unreachable and can be safely modified.
1629 * With NULL @old this simply adds @new to @active (while swapping the sets).
1630 * With NULL @new this simply removes @old from @active and frees it
1631 * (while also swapping the sets).
1633 static void kvm_activate_memslot(struct kvm *kvm,
1634 struct kvm_memory_slot *old,
1635 struct kvm_memory_slot *new)
1637 int as_id = kvm_memslots_get_as_id(old, new);
1639 kvm_swap_active_memslots(kvm, as_id);
1641 /* Propagate the new memslot to the now inactive memslots. */
1642 kvm_replace_memslot(kvm, old, new);
1645 static void kvm_copy_memslot(struct kvm_memory_slot *dest,
1646 const struct kvm_memory_slot *src)
1648 dest->base_gfn = src->base_gfn;
1649 dest->npages = src->npages;
1650 dest->dirty_bitmap = src->dirty_bitmap;
1651 dest->arch = src->arch;
1652 dest->userspace_addr = src->userspace_addr;
1653 dest->flags = src->flags;
1655 dest->as_id = src->as_id;
1658 static void kvm_invalidate_memslot(struct kvm *kvm,
1659 struct kvm_memory_slot *old,
1660 struct kvm_memory_slot *invalid_slot)
1663 * Mark the current slot INVALID. As with all memslot modifications,
1664 * this must be done on an unreachable slot to avoid modifying the
1665 * current slot in the active tree.
1667 kvm_copy_memslot(invalid_slot, old);
1668 invalid_slot->flags |= KVM_MEMSLOT_INVALID;
1669 kvm_replace_memslot(kvm, old, invalid_slot);
1672 * Activate the slot that is now marked INVALID, but don't propagate
1673 * the slot to the now inactive slots. The slot is either going to be
1674 * deleted or recreated as a new slot.
1676 kvm_swap_active_memslots(kvm, old->as_id);
1679 * From this point no new shadow pages pointing to a deleted, or moved,
1680 * memslot will be created. Validation of sp->gfn happens in:
1681 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1682 * - kvm_is_visible_gfn (mmu_check_root)
1684 kvm_arch_flush_shadow_memslot(kvm, old);
1685 kvm_arch_guest_memory_reclaimed(kvm);
1687 /* Was released by kvm_swap_active_memslots, reacquire. */
1688 mutex_lock(&kvm->slots_arch_lock);
1691 * Copy the arch-specific field of the newly-installed slot back to the
1692 * old slot as the arch data could have changed between releasing
1693 * slots_arch_lock in install_new_memslots() and re-acquiring the lock
1694 * above. Writers are required to retrieve memslots *after* acquiring
1695 * slots_arch_lock, thus the active slot's data is guaranteed to be fresh.
1697 old->arch = invalid_slot->arch;
1700 static void kvm_create_memslot(struct kvm *kvm,
1701 struct kvm_memory_slot *new)
1703 /* Add the new memslot to the inactive set and activate. */
1704 kvm_replace_memslot(kvm, NULL, new);
1705 kvm_activate_memslot(kvm, NULL, new);
1708 static void kvm_delete_memslot(struct kvm *kvm,
1709 struct kvm_memory_slot *old,
1710 struct kvm_memory_slot *invalid_slot)
1713 * Remove the old memslot (in the inactive memslots) by passing NULL as
1714 * the "new" slot, and for the invalid version in the active slots.
1716 kvm_replace_memslot(kvm, old, NULL);
1717 kvm_activate_memslot(kvm, invalid_slot, NULL);
1720 static void kvm_move_memslot(struct kvm *kvm,
1721 struct kvm_memory_slot *old,
1722 struct kvm_memory_slot *new,
1723 struct kvm_memory_slot *invalid_slot)
1726 * Replace the old memslot in the inactive slots, and then swap slots
1727 * and replace the current INVALID with the new as well.
1729 kvm_replace_memslot(kvm, old, new);
1730 kvm_activate_memslot(kvm, invalid_slot, new);
1733 static void kvm_update_flags_memslot(struct kvm *kvm,
1734 struct kvm_memory_slot *old,
1735 struct kvm_memory_slot *new)
1738 * Similar to the MOVE case, but the slot doesn't need to be zapped as
1739 * an intermediate step. Instead, the old memslot is simply replaced
1740 * with a new, updated copy in both memslot sets.
1742 kvm_replace_memslot(kvm, old, new);
1743 kvm_activate_memslot(kvm, old, new);
1746 static int kvm_set_memslot(struct kvm *kvm,
1747 struct kvm_memory_slot *old,
1748 struct kvm_memory_slot *new,
1749 enum kvm_mr_change change)
1751 struct kvm_memory_slot *invalid_slot;
1755 * Released in kvm_swap_active_memslots.
1757 * Must be held from before the current memslots are copied until
1758 * after the new memslots are installed with rcu_assign_pointer,
1759 * then released before the synchronize srcu in kvm_swap_active_memslots.
1761 * When modifying memslots outside of the slots_lock, must be held
1762 * before reading the pointer to the current memslots until after all
1763 * changes to those memslots are complete.
1765 * These rules ensure that installing new memslots does not lose
1766 * changes made to the previous memslots.
1768 mutex_lock(&kvm->slots_arch_lock);
1771 * Invalidate the old slot if it's being deleted or moved. This is
1772 * done prior to actually deleting/moving the memslot to allow vCPUs to
1773 * continue running by ensuring there are no mappings or shadow pages
1774 * for the memslot when it is deleted/moved. Without pre-invalidation
1775 * (and without a lock), a window would exist between effecting the
1776 * delete/move and committing the changes in arch code where KVM or a
1777 * guest could access a non-existent memslot.
1779 * Modifications are done on a temporary, unreachable slot. The old
1780 * slot needs to be preserved in case a later step fails and the
1781 * invalidation needs to be reverted.
1783 if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) {
1784 invalid_slot = kzalloc(sizeof(*invalid_slot), GFP_KERNEL_ACCOUNT);
1785 if (!invalid_slot) {
1786 mutex_unlock(&kvm->slots_arch_lock);
1789 kvm_invalidate_memslot(kvm, old, invalid_slot);
1792 r = kvm_prepare_memory_region(kvm, old, new, change);
1795 * For DELETE/MOVE, revert the above INVALID change. No
1796 * modifications required since the original slot was preserved
1797 * in the inactive slots. Changing the active memslots also
1798 * release slots_arch_lock.
1800 if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) {
1801 kvm_activate_memslot(kvm, invalid_slot, old);
1802 kfree(invalid_slot);
1804 mutex_unlock(&kvm->slots_arch_lock);
1810 * For DELETE and MOVE, the working slot is now active as the INVALID
1811 * version of the old slot. MOVE is particularly special as it reuses
1812 * the old slot and returns a copy of the old slot (in working_slot).
1813 * For CREATE, there is no old slot. For DELETE and FLAGS_ONLY, the
1814 * old slot is detached but otherwise preserved.
1816 if (change == KVM_MR_CREATE)
1817 kvm_create_memslot(kvm, new);
1818 else if (change == KVM_MR_DELETE)
1819 kvm_delete_memslot(kvm, old, invalid_slot);
1820 else if (change == KVM_MR_MOVE)
1821 kvm_move_memslot(kvm, old, new, invalid_slot);
1822 else if (change == KVM_MR_FLAGS_ONLY)
1823 kvm_update_flags_memslot(kvm, old, new);
1827 /* Free the temporary INVALID slot used for DELETE and MOVE. */
1828 if (change == KVM_MR_DELETE || change == KVM_MR_MOVE)
1829 kfree(invalid_slot);
1832 * No need to refresh new->arch, changes after dropping slots_arch_lock
1833 * will directly hit the final, active memslot. Architectures are
1834 * responsible for knowing that new->arch may be stale.
1836 kvm_commit_memory_region(kvm, old, new, change);
1841 static bool kvm_check_memslot_overlap(struct kvm_memslots *slots, int id,
1842 gfn_t start, gfn_t end)
1844 struct kvm_memslot_iter iter;
1846 kvm_for_each_memslot_in_gfn_range(&iter, slots, start, end) {
1847 if (iter.slot->id != id)
1855 * Allocate some memory and give it an address in the guest physical address
1858 * Discontiguous memory is allowed, mostly for framebuffers.
1860 * Must be called holding kvm->slots_lock for write.
1862 int __kvm_set_memory_region(struct kvm *kvm,
1863 const struct kvm_userspace_memory_region *mem)
1865 struct kvm_memory_slot *old, *new;
1866 struct kvm_memslots *slots;
1867 enum kvm_mr_change change;
1868 unsigned long npages;
1873 r = check_memory_region_flags(mem);
1877 as_id = mem->slot >> 16;
1878 id = (u16)mem->slot;
1880 /* General sanity checks */
1881 if ((mem->memory_size & (PAGE_SIZE - 1)) ||
1882 (mem->memory_size != (unsigned long)mem->memory_size))
1884 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
1886 /* We can read the guest memory with __xxx_user() later on. */
1887 if ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
1888 (mem->userspace_addr != untagged_addr(mem->userspace_addr)) ||
1889 !access_ok((void __user *)(unsigned long)mem->userspace_addr,
1892 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
1894 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
1896 if ((mem->memory_size >> PAGE_SHIFT) > KVM_MEM_MAX_NR_PAGES)
1899 slots = __kvm_memslots(kvm, as_id);
1902 * Note, the old memslot (and the pointer itself!) may be invalidated
1903 * and/or destroyed by kvm_set_memslot().
1905 old = id_to_memslot(slots, id);
1907 if (!mem->memory_size) {
1908 if (!old || !old->npages)
1911 if (WARN_ON_ONCE(kvm->nr_memslot_pages < old->npages))
1914 return kvm_set_memslot(kvm, old, NULL, KVM_MR_DELETE);
1917 base_gfn = (mem->guest_phys_addr >> PAGE_SHIFT);
1918 npages = (mem->memory_size >> PAGE_SHIFT);
1920 if (!old || !old->npages) {
1921 change = KVM_MR_CREATE;
1924 * To simplify KVM internals, the total number of pages across
1925 * all memslots must fit in an unsigned long.
1927 if ((kvm->nr_memslot_pages + npages) < kvm->nr_memslot_pages)
1929 } else { /* Modify an existing slot. */
1930 if ((mem->userspace_addr != old->userspace_addr) ||
1931 (npages != old->npages) ||
1932 ((mem->flags ^ old->flags) & KVM_MEM_READONLY))
1935 if (base_gfn != old->base_gfn)
1936 change = KVM_MR_MOVE;
1937 else if (mem->flags != old->flags)
1938 change = KVM_MR_FLAGS_ONLY;
1939 else /* Nothing to change. */
1943 if ((change == KVM_MR_CREATE || change == KVM_MR_MOVE) &&
1944 kvm_check_memslot_overlap(slots, id, base_gfn, base_gfn + npages))
1947 /* Allocate a slot that will persist in the memslot. */
1948 new = kzalloc(sizeof(*new), GFP_KERNEL_ACCOUNT);
1954 new->base_gfn = base_gfn;
1955 new->npages = npages;
1956 new->flags = mem->flags;
1957 new->userspace_addr = mem->userspace_addr;
1959 r = kvm_set_memslot(kvm, old, new, change);
1964 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1966 int kvm_set_memory_region(struct kvm *kvm,
1967 const struct kvm_userspace_memory_region *mem)
1971 mutex_lock(&kvm->slots_lock);
1972 r = __kvm_set_memory_region(kvm, mem);
1973 mutex_unlock(&kvm->slots_lock);
1976 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1978 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1979 struct kvm_userspace_memory_region *mem)
1981 if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1984 return kvm_set_memory_region(kvm, mem);
1987 #ifndef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1989 * kvm_get_dirty_log - get a snapshot of dirty pages
1990 * @kvm: pointer to kvm instance
1991 * @log: slot id and address to which we copy the log
1992 * @is_dirty: set to '1' if any dirty pages were found
1993 * @memslot: set to the associated memslot, always valid on success
1995 int kvm_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log,
1996 int *is_dirty, struct kvm_memory_slot **memslot)
1998 struct kvm_memslots *slots;
2001 unsigned long any = 0;
2003 /* Dirty ring tracking is exclusive to dirty log tracking */
2004 if (kvm->dirty_ring_size)
2010 as_id = log->slot >> 16;
2011 id = (u16)log->slot;
2012 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
2015 slots = __kvm_memslots(kvm, as_id);
2016 *memslot = id_to_memslot(slots, id);
2017 if (!(*memslot) || !(*memslot)->dirty_bitmap)
2020 kvm_arch_sync_dirty_log(kvm, *memslot);
2022 n = kvm_dirty_bitmap_bytes(*memslot);
2024 for (i = 0; !any && i < n/sizeof(long); ++i)
2025 any = (*memslot)->dirty_bitmap[i];
2027 if (copy_to_user(log->dirty_bitmap, (*memslot)->dirty_bitmap, n))
2034 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
2036 #else /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
2038 * kvm_get_dirty_log_protect - get a snapshot of dirty pages
2039 * and reenable dirty page tracking for the corresponding pages.
2040 * @kvm: pointer to kvm instance
2041 * @log: slot id and address to which we copy the log
2043 * We need to keep it in mind that VCPU threads can write to the bitmap
2044 * concurrently. So, to avoid losing track of dirty pages we keep the
2047 * 1. Take a snapshot of the bit and clear it if needed.
2048 * 2. Write protect the corresponding page.
2049 * 3. Copy the snapshot to the userspace.
2050 * 4. Upon return caller flushes TLB's if needed.
2052 * Between 2 and 4, the guest may write to the page using the remaining TLB
2053 * entry. This is not a problem because the page is reported dirty using
2054 * the snapshot taken before and step 4 ensures that writes done after
2055 * exiting to userspace will be logged for the next call.
2058 static int kvm_get_dirty_log_protect(struct kvm *kvm, struct kvm_dirty_log *log)
2060 struct kvm_memslots *slots;
2061 struct kvm_memory_slot *memslot;
2064 unsigned long *dirty_bitmap;
2065 unsigned long *dirty_bitmap_buffer;
2068 /* Dirty ring tracking is exclusive to dirty log tracking */
2069 if (kvm->dirty_ring_size)
2072 as_id = log->slot >> 16;
2073 id = (u16)log->slot;
2074 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
2077 slots = __kvm_memslots(kvm, as_id);
2078 memslot = id_to_memslot(slots, id);
2079 if (!memslot || !memslot->dirty_bitmap)
2082 dirty_bitmap = memslot->dirty_bitmap;
2084 kvm_arch_sync_dirty_log(kvm, memslot);
2086 n = kvm_dirty_bitmap_bytes(memslot);
2088 if (kvm->manual_dirty_log_protect) {
2090 * Unlike kvm_get_dirty_log, we always return false in *flush,
2091 * because no flush is needed until KVM_CLEAR_DIRTY_LOG. There
2092 * is some code duplication between this function and
2093 * kvm_get_dirty_log, but hopefully all architecture
2094 * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log
2095 * can be eliminated.
2097 dirty_bitmap_buffer = dirty_bitmap;
2099 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
2100 memset(dirty_bitmap_buffer, 0, n);
2103 for (i = 0; i < n / sizeof(long); i++) {
2107 if (!dirty_bitmap[i])
2111 mask = xchg(&dirty_bitmap[i], 0);
2112 dirty_bitmap_buffer[i] = mask;
2114 offset = i * BITS_PER_LONG;
2115 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
2118 KVM_MMU_UNLOCK(kvm);
2122 kvm_arch_flush_remote_tlbs_memslot(kvm, memslot);
2124 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
2131 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
2132 * @kvm: kvm instance
2133 * @log: slot id and address to which we copy the log
2135 * Steps 1-4 below provide general overview of dirty page logging. See
2136 * kvm_get_dirty_log_protect() function description for additional details.
2138 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
2139 * always flush the TLB (step 4) even if previous step failed and the dirty
2140 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
2141 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
2142 * writes will be marked dirty for next log read.
2144 * 1. Take a snapshot of the bit and clear it if needed.
2145 * 2. Write protect the corresponding page.
2146 * 3. Copy the snapshot to the userspace.
2147 * 4. Flush TLB's if needed.
2149 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
2150 struct kvm_dirty_log *log)
2154 mutex_lock(&kvm->slots_lock);
2156 r = kvm_get_dirty_log_protect(kvm, log);
2158 mutex_unlock(&kvm->slots_lock);
2163 * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap
2164 * and reenable dirty page tracking for the corresponding pages.
2165 * @kvm: pointer to kvm instance
2166 * @log: slot id and address from which to fetch the bitmap of dirty pages
2168 static int kvm_clear_dirty_log_protect(struct kvm *kvm,
2169 struct kvm_clear_dirty_log *log)
2171 struct kvm_memslots *slots;
2172 struct kvm_memory_slot *memslot;
2176 unsigned long *dirty_bitmap;
2177 unsigned long *dirty_bitmap_buffer;
2180 /* Dirty ring tracking is exclusive to dirty log tracking */
2181 if (kvm->dirty_ring_size)
2184 as_id = log->slot >> 16;
2185 id = (u16)log->slot;
2186 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
2189 if (log->first_page & 63)
2192 slots = __kvm_memslots(kvm, as_id);
2193 memslot = id_to_memslot(slots, id);
2194 if (!memslot || !memslot->dirty_bitmap)
2197 dirty_bitmap = memslot->dirty_bitmap;
2199 n = ALIGN(log->num_pages, BITS_PER_LONG) / 8;
2201 if (log->first_page > memslot->npages ||
2202 log->num_pages > memslot->npages - log->first_page ||
2203 (log->num_pages < memslot->npages - log->first_page && (log->num_pages & 63)))
2206 kvm_arch_sync_dirty_log(kvm, memslot);
2209 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
2210 if (copy_from_user(dirty_bitmap_buffer, log->dirty_bitmap, n))
2214 for (offset = log->first_page, i = offset / BITS_PER_LONG,
2215 n = DIV_ROUND_UP(log->num_pages, BITS_PER_LONG); n--;
2216 i++, offset += BITS_PER_LONG) {
2217 unsigned long mask = *dirty_bitmap_buffer++;
2218 atomic_long_t *p = (atomic_long_t *) &dirty_bitmap[i];
2222 mask &= atomic_long_fetch_andnot(mask, p);
2225 * mask contains the bits that really have been cleared. This
2226 * never includes any bits beyond the length of the memslot (if
2227 * the length is not aligned to 64 pages), therefore it is not
2228 * a problem if userspace sets them in log->dirty_bitmap.
2232 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
2236 KVM_MMU_UNLOCK(kvm);
2239 kvm_arch_flush_remote_tlbs_memslot(kvm, memslot);
2244 static int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm,
2245 struct kvm_clear_dirty_log *log)
2249 mutex_lock(&kvm->slots_lock);
2251 r = kvm_clear_dirty_log_protect(kvm, log);
2253 mutex_unlock(&kvm->slots_lock);
2256 #endif /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
2258 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
2260 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
2262 EXPORT_SYMBOL_GPL(gfn_to_memslot);
2264 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
2266 struct kvm_memslots *slots = kvm_vcpu_memslots(vcpu);
2267 u64 gen = slots->generation;
2268 struct kvm_memory_slot *slot;
2271 * This also protects against using a memslot from a different address space,
2272 * since different address spaces have different generation numbers.
2274 if (unlikely(gen != vcpu->last_used_slot_gen)) {
2275 vcpu->last_used_slot = NULL;
2276 vcpu->last_used_slot_gen = gen;
2279 slot = try_get_memslot(vcpu->last_used_slot, gfn);
2284 * Fall back to searching all memslots. We purposely use
2285 * search_memslots() instead of __gfn_to_memslot() to avoid
2286 * thrashing the VM-wide last_used_slot in kvm_memslots.
2288 slot = search_memslots(slots, gfn, false);
2290 vcpu->last_used_slot = slot;
2297 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
2299 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
2301 return kvm_is_visible_memslot(memslot);
2303 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
2305 bool kvm_vcpu_is_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
2307 struct kvm_memory_slot *memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2309 return kvm_is_visible_memslot(memslot);
2311 EXPORT_SYMBOL_GPL(kvm_vcpu_is_visible_gfn);
2313 unsigned long kvm_host_page_size(struct kvm_vcpu *vcpu, gfn_t gfn)
2315 struct vm_area_struct *vma;
2316 unsigned long addr, size;
2320 addr = kvm_vcpu_gfn_to_hva_prot(vcpu, gfn, NULL);
2321 if (kvm_is_error_hva(addr))
2324 mmap_read_lock(current->mm);
2325 vma = find_vma(current->mm, addr);
2329 size = vma_kernel_pagesize(vma);
2332 mmap_read_unlock(current->mm);
2337 static bool memslot_is_readonly(const struct kvm_memory_slot *slot)
2339 return slot->flags & KVM_MEM_READONLY;
2342 static unsigned long __gfn_to_hva_many(const struct kvm_memory_slot *slot, gfn_t gfn,
2343 gfn_t *nr_pages, bool write)
2345 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
2346 return KVM_HVA_ERR_BAD;
2348 if (memslot_is_readonly(slot) && write)
2349 return KVM_HVA_ERR_RO_BAD;
2352 *nr_pages = slot->npages - (gfn - slot->base_gfn);
2354 return __gfn_to_hva_memslot(slot, gfn);
2357 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
2360 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
2363 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
2366 return gfn_to_hva_many(slot, gfn, NULL);
2368 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
2370 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
2372 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
2374 EXPORT_SYMBOL_GPL(gfn_to_hva);
2376 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
2378 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
2380 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
2383 * Return the hva of a @gfn and the R/W attribute if possible.
2385 * @slot: the kvm_memory_slot which contains @gfn
2386 * @gfn: the gfn to be translated
2387 * @writable: used to return the read/write attribute of the @slot if the hva
2388 * is valid and @writable is not NULL
2390 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
2391 gfn_t gfn, bool *writable)
2393 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
2395 if (!kvm_is_error_hva(hva) && writable)
2396 *writable = !memslot_is_readonly(slot);
2401 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
2403 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2405 return gfn_to_hva_memslot_prot(slot, gfn, writable);
2408 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
2410 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2412 return gfn_to_hva_memslot_prot(slot, gfn, writable);
2415 static inline int check_user_page_hwpoison(unsigned long addr)
2417 int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
2419 rc = get_user_pages(addr, 1, flags, NULL, NULL);
2420 return rc == -EHWPOISON;
2424 * The fast path to get the writable pfn which will be stored in @pfn,
2425 * true indicates success, otherwise false is returned. It's also the
2426 * only part that runs if we can in atomic context.
2428 static bool hva_to_pfn_fast(unsigned long addr, bool write_fault,
2429 bool *writable, kvm_pfn_t *pfn)
2431 struct page *page[1];
2434 * Fast pin a writable pfn only if it is a write fault request
2435 * or the caller allows to map a writable pfn for a read fault
2438 if (!(write_fault || writable))
2441 if (get_user_page_fast_only(addr, FOLL_WRITE, page)) {
2442 *pfn = page_to_pfn(page[0]);
2453 * The slow path to get the pfn of the specified host virtual address,
2454 * 1 indicates success, -errno is returned if error is detected.
2456 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
2457 bool *writable, kvm_pfn_t *pfn)
2459 unsigned int flags = FOLL_HWPOISON;
2466 *writable = write_fault;
2469 flags |= FOLL_WRITE;
2471 flags |= FOLL_NOWAIT;
2473 npages = get_user_pages_unlocked(addr, 1, &page, flags);
2477 /* map read fault as writable if possible */
2478 if (unlikely(!write_fault) && writable) {
2481 if (get_user_page_fast_only(addr, FOLL_WRITE, &wpage)) {
2487 *pfn = page_to_pfn(page);
2491 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
2493 if (unlikely(!(vma->vm_flags & VM_READ)))
2496 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
2502 static int kvm_try_get_pfn(kvm_pfn_t pfn)
2504 if (kvm_is_reserved_pfn(pfn))
2506 return get_page_unless_zero(pfn_to_page(pfn));
2509 static int hva_to_pfn_remapped(struct vm_area_struct *vma,
2510 unsigned long addr, bool write_fault,
2511 bool *writable, kvm_pfn_t *p_pfn)
2518 r = follow_pte(vma->vm_mm, addr, &ptep, &ptl);
2521 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
2522 * not call the fault handler, so do it here.
2524 bool unlocked = false;
2525 r = fixup_user_fault(current->mm, addr,
2526 (write_fault ? FAULT_FLAG_WRITE : 0),
2533 r = follow_pte(vma->vm_mm, addr, &ptep, &ptl);
2538 if (write_fault && !pte_write(*ptep)) {
2539 pfn = KVM_PFN_ERR_RO_FAULT;
2544 *writable = pte_write(*ptep);
2545 pfn = pte_pfn(*ptep);
2548 * Get a reference here because callers of *hva_to_pfn* and
2549 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
2550 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
2551 * set, but the kvm_try_get_pfn/kvm_release_pfn_clean pair will
2552 * simply do nothing for reserved pfns.
2554 * Whoever called remap_pfn_range is also going to call e.g.
2555 * unmap_mapping_range before the underlying pages are freed,
2556 * causing a call to our MMU notifier.
2558 * Certain IO or PFNMAP mappings can be backed with valid
2559 * struct pages, but be allocated without refcounting e.g.,
2560 * tail pages of non-compound higher order allocations, which
2561 * would then underflow the refcount when the caller does the
2562 * required put_page. Don't allow those pages here.
2564 if (!kvm_try_get_pfn(pfn))
2568 pte_unmap_unlock(ptep, ptl);
2575 * Pin guest page in memory and return its pfn.
2576 * @addr: host virtual address which maps memory to the guest
2577 * @atomic: whether this function can sleep
2578 * @async: whether this function need to wait IO complete if the
2579 * host page is not in the memory
2580 * @write_fault: whether we should get a writable host page
2581 * @writable: whether it allows to map a writable host page for !@write_fault
2583 * The function will map a writable host page for these two cases:
2584 * 1): @write_fault = true
2585 * 2): @write_fault = false && @writable, @writable will tell the caller
2586 * whether the mapping is writable.
2588 kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
2589 bool write_fault, bool *writable)
2591 struct vm_area_struct *vma;
2595 /* we can do it either atomically or asynchronously, not both */
2596 BUG_ON(atomic && async);
2598 if (hva_to_pfn_fast(addr, write_fault, writable, &pfn))
2602 return KVM_PFN_ERR_FAULT;
2604 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
2608 mmap_read_lock(current->mm);
2609 if (npages == -EHWPOISON ||
2610 (!async && check_user_page_hwpoison(addr))) {
2611 pfn = KVM_PFN_ERR_HWPOISON;
2616 vma = vma_lookup(current->mm, addr);
2619 pfn = KVM_PFN_ERR_FAULT;
2620 else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
2621 r = hva_to_pfn_remapped(vma, addr, write_fault, writable, &pfn);
2625 pfn = KVM_PFN_ERR_FAULT;
2627 if (async && vma_is_valid(vma, write_fault))
2629 pfn = KVM_PFN_ERR_FAULT;
2632 mmap_read_unlock(current->mm);
2636 kvm_pfn_t __gfn_to_pfn_memslot(const struct kvm_memory_slot *slot, gfn_t gfn,
2637 bool atomic, bool *async, bool write_fault,
2638 bool *writable, hva_t *hva)
2640 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
2645 if (addr == KVM_HVA_ERR_RO_BAD) {
2648 return KVM_PFN_ERR_RO_FAULT;
2651 if (kvm_is_error_hva(addr)) {
2654 return KVM_PFN_NOSLOT;
2657 /* Do not map writable pfn in the readonly memslot. */
2658 if (writable && memslot_is_readonly(slot)) {
2663 return hva_to_pfn(addr, atomic, async, write_fault,
2666 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
2668 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
2671 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
2672 write_fault, writable, NULL);
2674 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
2676 kvm_pfn_t gfn_to_pfn_memslot(const struct kvm_memory_slot *slot, gfn_t gfn)
2678 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL, NULL);
2680 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
2682 kvm_pfn_t gfn_to_pfn_memslot_atomic(const struct kvm_memory_slot *slot, gfn_t gfn)
2684 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL, NULL);
2686 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
2688 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
2690 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
2692 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
2694 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
2696 return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
2698 EXPORT_SYMBOL_GPL(gfn_to_pfn);
2700 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
2702 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
2704 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
2706 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
2707 struct page **pages, int nr_pages)
2712 addr = gfn_to_hva_many(slot, gfn, &entry);
2713 if (kvm_is_error_hva(addr))
2716 if (entry < nr_pages)
2719 return get_user_pages_fast_only(addr, nr_pages, FOLL_WRITE, pages);
2721 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
2723 static struct page *kvm_pfn_to_page(kvm_pfn_t pfn)
2725 if (is_error_noslot_pfn(pfn))
2726 return KVM_ERR_PTR_BAD_PAGE;
2728 if (kvm_is_reserved_pfn(pfn))
2729 return KVM_ERR_PTR_BAD_PAGE;
2731 return pfn_to_page(pfn);
2734 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
2738 pfn = gfn_to_pfn(kvm, gfn);
2740 return kvm_pfn_to_page(pfn);
2742 EXPORT_SYMBOL_GPL(gfn_to_page);
2744 void kvm_release_pfn(kvm_pfn_t pfn, bool dirty)
2747 kvm_release_pfn_dirty(pfn);
2749 kvm_release_pfn_clean(pfn);
2752 int kvm_vcpu_map(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map)
2756 struct page *page = KVM_UNMAPPED_PAGE;
2761 pfn = gfn_to_pfn(vcpu->kvm, gfn);
2762 if (is_error_noslot_pfn(pfn))
2765 if (pfn_valid(pfn)) {
2766 page = pfn_to_page(pfn);
2768 #ifdef CONFIG_HAS_IOMEM
2770 hva = memremap(pfn_to_hpa(pfn), PAGE_SIZE, MEMREMAP_WB);
2784 EXPORT_SYMBOL_GPL(kvm_vcpu_map);
2786 void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map, bool dirty)
2794 if (map->page != KVM_UNMAPPED_PAGE)
2796 #ifdef CONFIG_HAS_IOMEM
2802 kvm_vcpu_mark_page_dirty(vcpu, map->gfn);
2804 kvm_release_pfn(map->pfn, dirty);
2809 EXPORT_SYMBOL_GPL(kvm_vcpu_unmap);
2811 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
2815 pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
2817 return kvm_pfn_to_page(pfn);
2819 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
2821 static bool kvm_is_ad_tracked_page(struct page *page)
2824 * Per page-flags.h, pages tagged PG_reserved "should in general not be
2825 * touched (e.g. set dirty) except by its owner".
2827 return !PageReserved(page);
2830 static void kvm_set_page_dirty(struct page *page)
2832 if (kvm_is_ad_tracked_page(page))
2836 static void kvm_set_page_accessed(struct page *page)
2838 if (kvm_is_ad_tracked_page(page))
2839 mark_page_accessed(page);
2842 void kvm_release_page_clean(struct page *page)
2844 WARN_ON(is_error_page(page));
2846 kvm_set_page_accessed(page);
2849 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
2851 void kvm_release_pfn_clean(kvm_pfn_t pfn)
2853 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
2854 kvm_release_page_clean(pfn_to_page(pfn));
2856 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
2858 void kvm_release_page_dirty(struct page *page)
2860 WARN_ON(is_error_page(page));
2862 kvm_set_page_dirty(page);
2863 kvm_release_page_clean(page);
2865 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
2867 void kvm_release_pfn_dirty(kvm_pfn_t pfn)
2869 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
2870 kvm_release_page_dirty(pfn_to_page(pfn));
2872 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
2875 * Note, checking for an error/noslot pfn is the caller's responsibility when
2876 * directly marking a page dirty/accessed. Unlike the "release" helpers, the
2877 * "set" helpers are not to be used when the pfn might point at garbage.
2879 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
2881 if (WARN_ON(is_error_noslot_pfn(pfn)))
2885 kvm_set_page_dirty(pfn_to_page(pfn));
2887 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
2889 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
2891 if (WARN_ON(is_error_noslot_pfn(pfn)))
2895 kvm_set_page_accessed(pfn_to_page(pfn));
2897 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
2899 static int next_segment(unsigned long len, int offset)
2901 if (len > PAGE_SIZE - offset)
2902 return PAGE_SIZE - offset;
2907 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
2908 void *data, int offset, int len)
2913 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
2914 if (kvm_is_error_hva(addr))
2916 r = __copy_from_user(data, (void __user *)addr + offset, len);
2922 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
2925 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2927 return __kvm_read_guest_page(slot, gfn, data, offset, len);
2929 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
2931 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
2932 int offset, int len)
2934 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2936 return __kvm_read_guest_page(slot, gfn, data, offset, len);
2938 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
2940 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
2942 gfn_t gfn = gpa >> PAGE_SHIFT;
2944 int offset = offset_in_page(gpa);
2947 while ((seg = next_segment(len, offset)) != 0) {
2948 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
2958 EXPORT_SYMBOL_GPL(kvm_read_guest);
2960 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
2962 gfn_t gfn = gpa >> PAGE_SHIFT;
2964 int offset = offset_in_page(gpa);
2967 while ((seg = next_segment(len, offset)) != 0) {
2968 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
2978 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
2980 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
2981 void *data, int offset, unsigned long len)
2986 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
2987 if (kvm_is_error_hva(addr))
2989 pagefault_disable();
2990 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
2997 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
2998 void *data, unsigned long len)
3000 gfn_t gfn = gpa >> PAGE_SHIFT;
3001 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
3002 int offset = offset_in_page(gpa);
3004 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
3006 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
3008 static int __kvm_write_guest_page(struct kvm *kvm,
3009 struct kvm_memory_slot *memslot, gfn_t gfn,
3010 const void *data, int offset, int len)
3015 addr = gfn_to_hva_memslot(memslot, gfn);
3016 if (kvm_is_error_hva(addr))
3018 r = __copy_to_user((void __user *)addr + offset, data, len);
3021 mark_page_dirty_in_slot(kvm, memslot, gfn);
3025 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
3026 const void *data, int offset, int len)
3028 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
3030 return __kvm_write_guest_page(kvm, slot, gfn, data, offset, len);
3032 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
3034 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
3035 const void *data, int offset, int len)
3037 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
3039 return __kvm_write_guest_page(vcpu->kvm, slot, gfn, data, offset, len);
3041 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
3043 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
3046 gfn_t gfn = gpa >> PAGE_SHIFT;
3048 int offset = offset_in_page(gpa);
3051 while ((seg = next_segment(len, offset)) != 0) {
3052 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
3062 EXPORT_SYMBOL_GPL(kvm_write_guest);
3064 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
3067 gfn_t gfn = gpa >> PAGE_SHIFT;
3069 int offset = offset_in_page(gpa);
3072 while ((seg = next_segment(len, offset)) != 0) {
3073 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
3083 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
3085 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
3086 struct gfn_to_hva_cache *ghc,
3087 gpa_t gpa, unsigned long len)
3089 int offset = offset_in_page(gpa);
3090 gfn_t start_gfn = gpa >> PAGE_SHIFT;
3091 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
3092 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
3093 gfn_t nr_pages_avail;
3095 /* Update ghc->generation before performing any error checks. */
3096 ghc->generation = slots->generation;
3098 if (start_gfn > end_gfn) {
3099 ghc->hva = KVM_HVA_ERR_BAD;
3104 * If the requested region crosses two memslots, we still
3105 * verify that the entire region is valid here.
3107 for ( ; start_gfn <= end_gfn; start_gfn += nr_pages_avail) {
3108 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
3109 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
3111 if (kvm_is_error_hva(ghc->hva))
3115 /* Use the slow path for cross page reads and writes. */
3116 if (nr_pages_needed == 1)
3119 ghc->memslot = NULL;
3126 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3127 gpa_t gpa, unsigned long len)
3129 struct kvm_memslots *slots = kvm_memslots(kvm);
3130 return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len);
3132 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
3134 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3135 void *data, unsigned int offset,
3138 struct kvm_memslots *slots = kvm_memslots(kvm);
3140 gpa_t gpa = ghc->gpa + offset;
3142 if (WARN_ON_ONCE(len + offset > ghc->len))
3145 if (slots->generation != ghc->generation) {
3146 if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len))
3150 if (kvm_is_error_hva(ghc->hva))
3153 if (unlikely(!ghc->memslot))
3154 return kvm_write_guest(kvm, gpa, data, len);
3156 r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
3159 mark_page_dirty_in_slot(kvm, ghc->memslot, gpa >> PAGE_SHIFT);
3163 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached);
3165 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3166 void *data, unsigned long len)
3168 return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len);
3170 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
3172 int kvm_read_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3173 void *data, unsigned int offset,
3176 struct kvm_memslots *slots = kvm_memslots(kvm);
3178 gpa_t gpa = ghc->gpa + offset;
3180 if (WARN_ON_ONCE(len + offset > ghc->len))
3183 if (slots->generation != ghc->generation) {
3184 if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len))
3188 if (kvm_is_error_hva(ghc->hva))
3191 if (unlikely(!ghc->memslot))
3192 return kvm_read_guest(kvm, gpa, data, len);
3194 r = __copy_from_user(data, (void __user *)ghc->hva + offset, len);
3200 EXPORT_SYMBOL_GPL(kvm_read_guest_offset_cached);
3202 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3203 void *data, unsigned long len)
3205 return kvm_read_guest_offset_cached(kvm, ghc, data, 0, len);
3207 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
3209 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
3211 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
3212 gfn_t gfn = gpa >> PAGE_SHIFT;
3214 int offset = offset_in_page(gpa);
3217 while ((seg = next_segment(len, offset)) != 0) {
3218 ret = kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
3227 EXPORT_SYMBOL_GPL(kvm_clear_guest);
3229 void mark_page_dirty_in_slot(struct kvm *kvm,
3230 const struct kvm_memory_slot *memslot,
3233 struct kvm_vcpu *vcpu = kvm_get_running_vcpu();
3235 #ifdef CONFIG_HAVE_KVM_DIRTY_RING
3236 if (WARN_ON_ONCE(!vcpu) || WARN_ON_ONCE(vcpu->kvm != kvm))
3240 if (memslot && kvm_slot_dirty_track_enabled(memslot)) {
3241 unsigned long rel_gfn = gfn - memslot->base_gfn;
3242 u32 slot = (memslot->as_id << 16) | memslot->id;
3244 if (kvm->dirty_ring_size)
3245 kvm_dirty_ring_push(&vcpu->dirty_ring,
3248 set_bit_le(rel_gfn, memslot->dirty_bitmap);
3251 EXPORT_SYMBOL_GPL(mark_page_dirty_in_slot);
3253 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
3255 struct kvm_memory_slot *memslot;
3257 memslot = gfn_to_memslot(kvm, gfn);
3258 mark_page_dirty_in_slot(kvm, memslot, gfn);
3260 EXPORT_SYMBOL_GPL(mark_page_dirty);
3262 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
3264 struct kvm_memory_slot *memslot;
3266 memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
3267 mark_page_dirty_in_slot(vcpu->kvm, memslot, gfn);
3269 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
3271 void kvm_sigset_activate(struct kvm_vcpu *vcpu)
3273 if (!vcpu->sigset_active)
3277 * This does a lockless modification of ->real_blocked, which is fine
3278 * because, only current can change ->real_blocked and all readers of
3279 * ->real_blocked don't care as long ->real_blocked is always a subset
3282 sigprocmask(SIG_SETMASK, &vcpu->sigset, ¤t->real_blocked);
3285 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu)
3287 if (!vcpu->sigset_active)
3290 sigprocmask(SIG_SETMASK, ¤t->real_blocked, NULL);
3291 sigemptyset(¤t->real_blocked);
3294 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
3296 unsigned int old, val, grow, grow_start;
3298 old = val = vcpu->halt_poll_ns;
3299 grow_start = READ_ONCE(halt_poll_ns_grow_start);
3300 grow = READ_ONCE(halt_poll_ns_grow);
3305 if (val < grow_start)
3308 if (val > vcpu->kvm->max_halt_poll_ns)
3309 val = vcpu->kvm->max_halt_poll_ns;
3311 vcpu->halt_poll_ns = val;
3313 trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
3316 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
3318 unsigned int old, val, shrink, grow_start;
3320 old = val = vcpu->halt_poll_ns;
3321 shrink = READ_ONCE(halt_poll_ns_shrink);
3322 grow_start = READ_ONCE(halt_poll_ns_grow_start);
3328 if (val < grow_start)
3331 vcpu->halt_poll_ns = val;
3332 trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
3335 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
3338 int idx = srcu_read_lock(&vcpu->kvm->srcu);
3340 if (kvm_arch_vcpu_runnable(vcpu)) {
3341 kvm_make_request(KVM_REQ_UNHALT, vcpu);
3344 if (kvm_cpu_has_pending_timer(vcpu))
3346 if (signal_pending(current))
3348 if (kvm_check_request(KVM_REQ_UNBLOCK, vcpu))
3353 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3358 * Block the vCPU until the vCPU is runnable, an event arrives, or a signal is
3359 * pending. This is mostly used when halting a vCPU, but may also be used
3360 * directly for other vCPU non-runnable states, e.g. x86's Wait-For-SIPI.
3362 bool kvm_vcpu_block(struct kvm_vcpu *vcpu)
3364 struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu);
3365 bool waited = false;
3367 vcpu->stat.generic.blocking = 1;
3370 kvm_arch_vcpu_blocking(vcpu);
3371 prepare_to_rcuwait(wait);
3375 set_current_state(TASK_INTERRUPTIBLE);
3377 if (kvm_vcpu_check_block(vcpu) < 0)
3385 finish_rcuwait(wait);
3386 kvm_arch_vcpu_unblocking(vcpu);
3389 vcpu->stat.generic.blocking = 0;
3394 static inline void update_halt_poll_stats(struct kvm_vcpu *vcpu, ktime_t start,
3395 ktime_t end, bool success)
3397 struct kvm_vcpu_stat_generic *stats = &vcpu->stat.generic;
3398 u64 poll_ns = ktime_to_ns(ktime_sub(end, start));
3400 ++vcpu->stat.generic.halt_attempted_poll;
3403 ++vcpu->stat.generic.halt_successful_poll;
3405 if (!vcpu_valid_wakeup(vcpu))
3406 ++vcpu->stat.generic.halt_poll_invalid;
3408 stats->halt_poll_success_ns += poll_ns;
3409 KVM_STATS_LOG_HIST_UPDATE(stats->halt_poll_success_hist, poll_ns);
3411 stats->halt_poll_fail_ns += poll_ns;
3412 KVM_STATS_LOG_HIST_UPDATE(stats->halt_poll_fail_hist, poll_ns);
3417 * Emulate a vCPU halt condition, e.g. HLT on x86, WFI on arm, etc... If halt
3418 * polling is enabled, busy wait for a short time before blocking to avoid the
3419 * expensive block+unblock sequence if a wake event arrives soon after the vCPU
3422 void kvm_vcpu_halt(struct kvm_vcpu *vcpu)
3424 bool halt_poll_allowed = !kvm_arch_no_poll(vcpu);
3425 bool do_halt_poll = halt_poll_allowed && vcpu->halt_poll_ns;
3426 ktime_t start, cur, poll_end;
3427 bool waited = false;
3430 start = cur = poll_end = ktime_get();
3432 ktime_t stop = ktime_add_ns(start, vcpu->halt_poll_ns);
3436 * This sets KVM_REQ_UNHALT if an interrupt
3439 if (kvm_vcpu_check_block(vcpu) < 0)
3442 poll_end = cur = ktime_get();
3443 } while (kvm_vcpu_can_poll(cur, stop));
3446 waited = kvm_vcpu_block(vcpu);
3450 vcpu->stat.generic.halt_wait_ns +=
3451 ktime_to_ns(cur) - ktime_to_ns(poll_end);
3452 KVM_STATS_LOG_HIST_UPDATE(vcpu->stat.generic.halt_wait_hist,
3453 ktime_to_ns(cur) - ktime_to_ns(poll_end));
3456 /* The total time the vCPU was "halted", including polling time. */
3457 halt_ns = ktime_to_ns(cur) - ktime_to_ns(start);
3460 * Note, halt-polling is considered successful so long as the vCPU was
3461 * never actually scheduled out, i.e. even if the wake event arrived
3462 * after of the halt-polling loop itself, but before the full wait.
3465 update_halt_poll_stats(vcpu, start, poll_end, !waited);
3467 if (halt_poll_allowed) {
3468 if (!vcpu_valid_wakeup(vcpu)) {
3469 shrink_halt_poll_ns(vcpu);
3470 } else if (vcpu->kvm->max_halt_poll_ns) {
3471 if (halt_ns <= vcpu->halt_poll_ns)
3473 /* we had a long block, shrink polling */
3474 else if (vcpu->halt_poll_ns &&
3475 halt_ns > vcpu->kvm->max_halt_poll_ns)
3476 shrink_halt_poll_ns(vcpu);
3477 /* we had a short halt and our poll time is too small */
3478 else if (vcpu->halt_poll_ns < vcpu->kvm->max_halt_poll_ns &&
3479 halt_ns < vcpu->kvm->max_halt_poll_ns)
3480 grow_halt_poll_ns(vcpu);
3482 vcpu->halt_poll_ns = 0;
3486 trace_kvm_vcpu_wakeup(halt_ns, waited, vcpu_valid_wakeup(vcpu));
3488 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
3490 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
3492 if (__kvm_vcpu_wake_up(vcpu)) {
3493 WRITE_ONCE(vcpu->ready, true);
3494 ++vcpu->stat.generic.halt_wakeup;
3500 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
3504 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
3506 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
3510 if (kvm_vcpu_wake_up(vcpu))
3515 * The only state change done outside the vcpu mutex is IN_GUEST_MODE
3516 * to EXITING_GUEST_MODE. Therefore the moderately expensive "should
3517 * kick" check does not need atomic operations if kvm_vcpu_kick is used
3518 * within the vCPU thread itself.
3520 if (vcpu == __this_cpu_read(kvm_running_vcpu)) {
3521 if (vcpu->mode == IN_GUEST_MODE)
3522 WRITE_ONCE(vcpu->mode, EXITING_GUEST_MODE);
3527 * Note, the vCPU could get migrated to a different pCPU at any point
3528 * after kvm_arch_vcpu_should_kick(), which could result in sending an
3529 * IPI to the previous pCPU. But, that's ok because the purpose of the
3530 * IPI is to force the vCPU to leave IN_GUEST_MODE, and migrating the
3531 * vCPU also requires it to leave IN_GUEST_MODE.
3533 if (kvm_arch_vcpu_should_kick(vcpu)) {
3534 cpu = READ_ONCE(vcpu->cpu);
3535 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
3536 smp_send_reschedule(cpu);
3541 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
3542 #endif /* !CONFIG_S390 */
3544 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
3547 struct task_struct *task = NULL;
3551 pid = rcu_dereference(target->pid);
3553 task = get_pid_task(pid, PIDTYPE_PID);
3557 ret = yield_to(task, 1);
3558 put_task_struct(task);
3562 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
3565 * Helper that checks whether a VCPU is eligible for directed yield.
3566 * Most eligible candidate to yield is decided by following heuristics:
3568 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
3569 * (preempted lock holder), indicated by @in_spin_loop.
3570 * Set at the beginning and cleared at the end of interception/PLE handler.
3572 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
3573 * chance last time (mostly it has become eligible now since we have probably
3574 * yielded to lockholder in last iteration. This is done by toggling
3575 * @dy_eligible each time a VCPU checked for eligibility.)
3577 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
3578 * to preempted lock-holder could result in wrong VCPU selection and CPU
3579 * burning. Giving priority for a potential lock-holder increases lock
3582 * Since algorithm is based on heuristics, accessing another VCPU data without
3583 * locking does not harm. It may result in trying to yield to same VCPU, fail
3584 * and continue with next VCPU and so on.
3586 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
3588 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
3591 eligible = !vcpu->spin_loop.in_spin_loop ||
3592 vcpu->spin_loop.dy_eligible;
3594 if (vcpu->spin_loop.in_spin_loop)
3595 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
3604 * Unlike kvm_arch_vcpu_runnable, this function is called outside
3605 * a vcpu_load/vcpu_put pair. However, for most architectures
3606 * kvm_arch_vcpu_runnable does not require vcpu_load.
3608 bool __weak kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
3610 return kvm_arch_vcpu_runnable(vcpu);
3613 static bool vcpu_dy_runnable(struct kvm_vcpu *vcpu)
3615 if (kvm_arch_dy_runnable(vcpu))
3618 #ifdef CONFIG_KVM_ASYNC_PF
3619 if (!list_empty_careful(&vcpu->async_pf.done))
3626 bool __weak kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu)
3631 void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode)
3633 struct kvm *kvm = me->kvm;
3634 struct kvm_vcpu *vcpu;
3635 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
3641 kvm_vcpu_set_in_spin_loop(me, true);
3643 * We boost the priority of a VCPU that is runnable but not
3644 * currently running, because it got preempted by something
3645 * else and called schedule in __vcpu_run. Hopefully that
3646 * VCPU is holding the lock that we need and will release it.
3647 * We approximate round-robin by starting at the last boosted VCPU.
3649 for (pass = 0; pass < 2 && !yielded && try; pass++) {
3650 kvm_for_each_vcpu(i, vcpu, kvm) {
3651 if (!pass && i <= last_boosted_vcpu) {
3652 i = last_boosted_vcpu;
3654 } else if (pass && i > last_boosted_vcpu)
3656 if (!READ_ONCE(vcpu->ready))
3660 if (kvm_vcpu_is_blocking(vcpu) && !vcpu_dy_runnable(vcpu))
3662 if (READ_ONCE(vcpu->preempted) && yield_to_kernel_mode &&
3663 !kvm_arch_dy_has_pending_interrupt(vcpu) &&
3664 !kvm_arch_vcpu_in_kernel(vcpu))
3666 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
3669 yielded = kvm_vcpu_yield_to(vcpu);
3671 kvm->last_boosted_vcpu = i;
3673 } else if (yielded < 0) {
3680 kvm_vcpu_set_in_spin_loop(me, false);
3682 /* Ensure vcpu is not eligible during next spinloop */
3683 kvm_vcpu_set_dy_eligible(me, false);
3685 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
3687 static bool kvm_page_in_dirty_ring(struct kvm *kvm, unsigned long pgoff)
3689 #ifdef CONFIG_HAVE_KVM_DIRTY_RING
3690 return (pgoff >= KVM_DIRTY_LOG_PAGE_OFFSET) &&
3691 (pgoff < KVM_DIRTY_LOG_PAGE_OFFSET +
3692 kvm->dirty_ring_size / PAGE_SIZE);
3698 static vm_fault_t kvm_vcpu_fault(struct vm_fault *vmf)
3700 struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data;
3703 if (vmf->pgoff == 0)
3704 page = virt_to_page(vcpu->run);
3706 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
3707 page = virt_to_page(vcpu->arch.pio_data);
3709 #ifdef CONFIG_KVM_MMIO
3710 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
3711 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
3713 else if (kvm_page_in_dirty_ring(vcpu->kvm, vmf->pgoff))
3714 page = kvm_dirty_ring_get_page(
3716 vmf->pgoff - KVM_DIRTY_LOG_PAGE_OFFSET);
3718 return kvm_arch_vcpu_fault(vcpu, vmf);
3724 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
3725 .fault = kvm_vcpu_fault,
3728 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
3730 struct kvm_vcpu *vcpu = file->private_data;
3731 unsigned long pages = vma_pages(vma);
3733 if ((kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff) ||
3734 kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff + pages - 1)) &&
3735 ((vma->vm_flags & VM_EXEC) || !(vma->vm_flags & VM_SHARED)))
3738 vma->vm_ops = &kvm_vcpu_vm_ops;
3742 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
3744 struct kvm_vcpu *vcpu = filp->private_data;
3746 kvm_put_kvm(vcpu->kvm);
3750 static const struct file_operations kvm_vcpu_fops = {
3751 .release = kvm_vcpu_release,
3752 .unlocked_ioctl = kvm_vcpu_ioctl,
3753 .mmap = kvm_vcpu_mmap,
3754 .llseek = noop_llseek,
3755 KVM_COMPAT(kvm_vcpu_compat_ioctl),
3759 * Allocates an inode for the vcpu.
3761 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
3763 char name[8 + 1 + ITOA_MAX_LEN + 1];
3765 snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id);
3766 return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
3769 static void kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
3771 #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS
3772 struct dentry *debugfs_dentry;
3773 char dir_name[ITOA_MAX_LEN * 2];
3775 if (!debugfs_initialized())
3778 snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
3779 debugfs_dentry = debugfs_create_dir(dir_name,
3780 vcpu->kvm->debugfs_dentry);
3782 kvm_arch_create_vcpu_debugfs(vcpu, debugfs_dentry);
3787 * Creates some virtual cpus. Good luck creating more than one.
3789 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
3792 struct kvm_vcpu *vcpu;
3795 if (id >= KVM_MAX_VCPU_IDS)
3798 mutex_lock(&kvm->lock);
3799 if (kvm->created_vcpus >= kvm->max_vcpus) {
3800 mutex_unlock(&kvm->lock);
3804 r = kvm_arch_vcpu_precreate(kvm, id);
3806 mutex_unlock(&kvm->lock);
3810 kvm->created_vcpus++;
3811 mutex_unlock(&kvm->lock);
3813 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL_ACCOUNT);
3816 goto vcpu_decrement;
3819 BUILD_BUG_ON(sizeof(struct kvm_run) > PAGE_SIZE);
3820 page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
3825 vcpu->run = page_address(page);
3827 kvm_vcpu_init(vcpu, kvm, id);
3829 r = kvm_arch_vcpu_create(vcpu);
3831 goto vcpu_free_run_page;
3833 if (kvm->dirty_ring_size) {
3834 r = kvm_dirty_ring_alloc(&vcpu->dirty_ring,
3835 id, kvm->dirty_ring_size);
3837 goto arch_vcpu_destroy;
3840 mutex_lock(&kvm->lock);
3841 if (kvm_get_vcpu_by_id(kvm, id)) {
3843 goto unlock_vcpu_destroy;
3846 vcpu->vcpu_idx = atomic_read(&kvm->online_vcpus);
3847 r = xa_insert(&kvm->vcpu_array, vcpu->vcpu_idx, vcpu, GFP_KERNEL_ACCOUNT);
3848 BUG_ON(r == -EBUSY);
3850 goto unlock_vcpu_destroy;
3852 /* Fill the stats id string for the vcpu */
3853 snprintf(vcpu->stats_id, sizeof(vcpu->stats_id), "kvm-%d/vcpu-%d",
3854 task_pid_nr(current), id);
3856 /* Now it's all set up, let userspace reach it */
3858 r = create_vcpu_fd(vcpu);
3860 xa_erase(&kvm->vcpu_array, vcpu->vcpu_idx);
3861 kvm_put_kvm_no_destroy(kvm);
3862 goto unlock_vcpu_destroy;
3866 * Pairs with smp_rmb() in kvm_get_vcpu. Store the vcpu
3867 * pointer before kvm->online_vcpu's incremented value.
3870 atomic_inc(&kvm->online_vcpus);
3872 mutex_unlock(&kvm->lock);
3873 kvm_arch_vcpu_postcreate(vcpu);
3874 kvm_create_vcpu_debugfs(vcpu);
3877 unlock_vcpu_destroy:
3878 mutex_unlock(&kvm->lock);
3879 kvm_dirty_ring_free(&vcpu->dirty_ring);
3881 kvm_arch_vcpu_destroy(vcpu);
3883 free_page((unsigned long)vcpu->run);
3885 kmem_cache_free(kvm_vcpu_cache, vcpu);
3887 mutex_lock(&kvm->lock);
3888 kvm->created_vcpus--;
3889 mutex_unlock(&kvm->lock);
3893 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
3896 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
3897 vcpu->sigset_active = 1;
3898 vcpu->sigset = *sigset;
3900 vcpu->sigset_active = 0;
3904 static ssize_t kvm_vcpu_stats_read(struct file *file, char __user *user_buffer,
3905 size_t size, loff_t *offset)
3907 struct kvm_vcpu *vcpu = file->private_data;
3909 return kvm_stats_read(vcpu->stats_id, &kvm_vcpu_stats_header,
3910 &kvm_vcpu_stats_desc[0], &vcpu->stat,
3911 sizeof(vcpu->stat), user_buffer, size, offset);
3914 static const struct file_operations kvm_vcpu_stats_fops = {
3915 .read = kvm_vcpu_stats_read,
3916 .llseek = noop_llseek,
3919 static int kvm_vcpu_ioctl_get_stats_fd(struct kvm_vcpu *vcpu)
3923 char name[15 + ITOA_MAX_LEN + 1];
3925 snprintf(name, sizeof(name), "kvm-vcpu-stats:%d", vcpu->vcpu_id);
3927 fd = get_unused_fd_flags(O_CLOEXEC);
3931 file = anon_inode_getfile(name, &kvm_vcpu_stats_fops, vcpu, O_RDONLY);
3934 return PTR_ERR(file);
3936 file->f_mode |= FMODE_PREAD;
3937 fd_install(fd, file);
3942 static long kvm_vcpu_ioctl(struct file *filp,
3943 unsigned int ioctl, unsigned long arg)
3945 struct kvm_vcpu *vcpu = filp->private_data;
3946 void __user *argp = (void __user *)arg;
3948 struct kvm_fpu *fpu = NULL;
3949 struct kvm_sregs *kvm_sregs = NULL;
3951 if (vcpu->kvm->mm != current->mm || vcpu->kvm->vm_dead)
3954 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
3958 * Some architectures have vcpu ioctls that are asynchronous to vcpu
3959 * execution; mutex_lock() would break them.
3961 r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg);
3962 if (r != -ENOIOCTLCMD)
3965 if (mutex_lock_killable(&vcpu->mutex))
3973 oldpid = rcu_access_pointer(vcpu->pid);
3974 if (unlikely(oldpid != task_pid(current))) {
3975 /* The thread running this VCPU changed. */
3978 r = kvm_arch_vcpu_run_pid_change(vcpu);
3982 newpid = get_task_pid(current, PIDTYPE_PID);
3983 rcu_assign_pointer(vcpu->pid, newpid);
3988 r = kvm_arch_vcpu_ioctl_run(vcpu);
3989 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
3992 case KVM_GET_REGS: {
3993 struct kvm_regs *kvm_regs;
3996 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL_ACCOUNT);
3999 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
4003 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
4010 case KVM_SET_REGS: {
4011 struct kvm_regs *kvm_regs;
4013 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
4014 if (IS_ERR(kvm_regs)) {
4015 r = PTR_ERR(kvm_regs);
4018 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
4022 case KVM_GET_SREGS: {
4023 kvm_sregs = kzalloc(sizeof(struct kvm_sregs),
4024 GFP_KERNEL_ACCOUNT);
4028 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
4032 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
4037 case KVM_SET_SREGS: {
4038 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
4039 if (IS_ERR(kvm_sregs)) {
4040 r = PTR_ERR(kvm_sregs);
4044 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
4047 case KVM_GET_MP_STATE: {
4048 struct kvm_mp_state mp_state;
4050 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
4054 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
4059 case KVM_SET_MP_STATE: {
4060 struct kvm_mp_state mp_state;
4063 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
4065 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
4068 case KVM_TRANSLATE: {
4069 struct kvm_translation tr;
4072 if (copy_from_user(&tr, argp, sizeof(tr)))
4074 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
4078 if (copy_to_user(argp, &tr, sizeof(tr)))
4083 case KVM_SET_GUEST_DEBUG: {
4084 struct kvm_guest_debug dbg;
4087 if (copy_from_user(&dbg, argp, sizeof(dbg)))
4089 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
4092 case KVM_SET_SIGNAL_MASK: {
4093 struct kvm_signal_mask __user *sigmask_arg = argp;
4094 struct kvm_signal_mask kvm_sigmask;
4095 sigset_t sigset, *p;
4100 if (copy_from_user(&kvm_sigmask, argp,
4101 sizeof(kvm_sigmask)))
4104 if (kvm_sigmask.len != sizeof(sigset))
4107 if (copy_from_user(&sigset, sigmask_arg->sigset,
4112 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
4116 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL_ACCOUNT);
4120 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
4124 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
4130 fpu = memdup_user(argp, sizeof(*fpu));
4136 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
4139 case KVM_GET_STATS_FD: {
4140 r = kvm_vcpu_ioctl_get_stats_fd(vcpu);
4144 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
4147 mutex_unlock(&vcpu->mutex);
4153 #ifdef CONFIG_KVM_COMPAT
4154 static long kvm_vcpu_compat_ioctl(struct file *filp,
4155 unsigned int ioctl, unsigned long arg)
4157 struct kvm_vcpu *vcpu = filp->private_data;
4158 void __user *argp = compat_ptr(arg);
4161 if (vcpu->kvm->mm != current->mm || vcpu->kvm->vm_dead)
4165 case KVM_SET_SIGNAL_MASK: {
4166 struct kvm_signal_mask __user *sigmask_arg = argp;
4167 struct kvm_signal_mask kvm_sigmask;
4172 if (copy_from_user(&kvm_sigmask, argp,
4173 sizeof(kvm_sigmask)))
4176 if (kvm_sigmask.len != sizeof(compat_sigset_t))
4179 if (get_compat_sigset(&sigset,
4180 (compat_sigset_t __user *)sigmask_arg->sigset))
4182 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
4184 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
4188 r = kvm_vcpu_ioctl(filp, ioctl, arg);
4196 static int kvm_device_mmap(struct file *filp, struct vm_area_struct *vma)
4198 struct kvm_device *dev = filp->private_data;
4201 return dev->ops->mmap(dev, vma);
4206 static int kvm_device_ioctl_attr(struct kvm_device *dev,
4207 int (*accessor)(struct kvm_device *dev,
4208 struct kvm_device_attr *attr),
4211 struct kvm_device_attr attr;
4216 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
4219 return accessor(dev, &attr);
4222 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
4225 struct kvm_device *dev = filp->private_data;
4227 if (dev->kvm->mm != current->mm || dev->kvm->vm_dead)
4231 case KVM_SET_DEVICE_ATTR:
4232 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
4233 case KVM_GET_DEVICE_ATTR:
4234 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
4235 case KVM_HAS_DEVICE_ATTR:
4236 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
4238 if (dev->ops->ioctl)
4239 return dev->ops->ioctl(dev, ioctl, arg);
4245 static int kvm_device_release(struct inode *inode, struct file *filp)
4247 struct kvm_device *dev = filp->private_data;
4248 struct kvm *kvm = dev->kvm;
4250 if (dev->ops->release) {
4251 mutex_lock(&kvm->lock);
4252 list_del(&dev->vm_node);
4253 dev->ops->release(dev);
4254 mutex_unlock(&kvm->lock);
4261 static const struct file_operations kvm_device_fops = {
4262 .unlocked_ioctl = kvm_device_ioctl,
4263 .release = kvm_device_release,
4264 KVM_COMPAT(kvm_device_ioctl),
4265 .mmap = kvm_device_mmap,
4268 struct kvm_device *kvm_device_from_filp(struct file *filp)
4270 if (filp->f_op != &kvm_device_fops)
4273 return filp->private_data;
4276 static const struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
4277 #ifdef CONFIG_KVM_MPIC
4278 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
4279 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
4283 int kvm_register_device_ops(const struct kvm_device_ops *ops, u32 type)
4285 if (type >= ARRAY_SIZE(kvm_device_ops_table))
4288 if (kvm_device_ops_table[type] != NULL)
4291 kvm_device_ops_table[type] = ops;
4295 void kvm_unregister_device_ops(u32 type)
4297 if (kvm_device_ops_table[type] != NULL)
4298 kvm_device_ops_table[type] = NULL;
4301 static int kvm_ioctl_create_device(struct kvm *kvm,
4302 struct kvm_create_device *cd)
4304 const struct kvm_device_ops *ops = NULL;
4305 struct kvm_device *dev;
4306 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
4310 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
4313 type = array_index_nospec(cd->type, ARRAY_SIZE(kvm_device_ops_table));
4314 ops = kvm_device_ops_table[type];
4321 dev = kzalloc(sizeof(*dev), GFP_KERNEL_ACCOUNT);
4328 mutex_lock(&kvm->lock);
4329 ret = ops->create(dev, type);
4331 mutex_unlock(&kvm->lock);
4335 list_add(&dev->vm_node, &kvm->devices);
4336 mutex_unlock(&kvm->lock);
4342 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
4344 kvm_put_kvm_no_destroy(kvm);
4345 mutex_lock(&kvm->lock);
4346 list_del(&dev->vm_node);
4349 mutex_unlock(&kvm->lock);
4359 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
4362 case KVM_CAP_USER_MEMORY:
4363 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
4364 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
4365 case KVM_CAP_INTERNAL_ERROR_DATA:
4366 #ifdef CONFIG_HAVE_KVM_MSI
4367 case KVM_CAP_SIGNAL_MSI:
4369 #ifdef CONFIG_HAVE_KVM_IRQFD
4371 case KVM_CAP_IRQFD_RESAMPLE:
4373 case KVM_CAP_IOEVENTFD_ANY_LENGTH:
4374 case KVM_CAP_CHECK_EXTENSION_VM:
4375 case KVM_CAP_ENABLE_CAP_VM:
4376 case KVM_CAP_HALT_POLL:
4378 #ifdef CONFIG_KVM_MMIO
4379 case KVM_CAP_COALESCED_MMIO:
4380 return KVM_COALESCED_MMIO_PAGE_OFFSET;
4381 case KVM_CAP_COALESCED_PIO:
4384 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4385 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2:
4386 return KVM_DIRTY_LOG_MANUAL_CAPS;
4388 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
4389 case KVM_CAP_IRQ_ROUTING:
4390 return KVM_MAX_IRQ_ROUTES;
4392 #if KVM_ADDRESS_SPACE_NUM > 1
4393 case KVM_CAP_MULTI_ADDRESS_SPACE:
4394 return KVM_ADDRESS_SPACE_NUM;
4396 case KVM_CAP_NR_MEMSLOTS:
4397 return KVM_USER_MEM_SLOTS;
4398 case KVM_CAP_DIRTY_LOG_RING:
4399 #ifdef CONFIG_HAVE_KVM_DIRTY_RING
4400 return KVM_DIRTY_RING_MAX_ENTRIES * sizeof(struct kvm_dirty_gfn);
4404 case KVM_CAP_BINARY_STATS_FD:
4405 case KVM_CAP_SYSTEM_EVENT_DATA:
4410 return kvm_vm_ioctl_check_extension(kvm, arg);
4413 static int kvm_vm_ioctl_enable_dirty_log_ring(struct kvm *kvm, u32 size)
4417 if (!KVM_DIRTY_LOG_PAGE_OFFSET)
4420 /* the size should be power of 2 */
4421 if (!size || (size & (size - 1)))
4424 /* Should be bigger to keep the reserved entries, or a page */
4425 if (size < kvm_dirty_ring_get_rsvd_entries() *
4426 sizeof(struct kvm_dirty_gfn) || size < PAGE_SIZE)
4429 if (size > KVM_DIRTY_RING_MAX_ENTRIES *
4430 sizeof(struct kvm_dirty_gfn))
4433 /* We only allow it to set once */
4434 if (kvm->dirty_ring_size)
4437 mutex_lock(&kvm->lock);
4439 if (kvm->created_vcpus) {
4440 /* We don't allow to change this value after vcpu created */
4443 kvm->dirty_ring_size = size;
4447 mutex_unlock(&kvm->lock);
4451 static int kvm_vm_ioctl_reset_dirty_pages(struct kvm *kvm)
4454 struct kvm_vcpu *vcpu;
4457 if (!kvm->dirty_ring_size)
4460 mutex_lock(&kvm->slots_lock);
4462 kvm_for_each_vcpu(i, vcpu, kvm)
4463 cleared += kvm_dirty_ring_reset(vcpu->kvm, &vcpu->dirty_ring);
4465 mutex_unlock(&kvm->slots_lock);
4468 kvm_flush_remote_tlbs(kvm);
4473 int __attribute__((weak)) kvm_vm_ioctl_enable_cap(struct kvm *kvm,
4474 struct kvm_enable_cap *cap)
4479 static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm,
4480 struct kvm_enable_cap *cap)
4483 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4484 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2: {
4485 u64 allowed_options = KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE;
4487 if (cap->args[0] & KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE)
4488 allowed_options = KVM_DIRTY_LOG_MANUAL_CAPS;
4490 if (cap->flags || (cap->args[0] & ~allowed_options))
4492 kvm->manual_dirty_log_protect = cap->args[0];
4496 case KVM_CAP_HALT_POLL: {
4497 if (cap->flags || cap->args[0] != (unsigned int)cap->args[0])
4500 kvm->max_halt_poll_ns = cap->args[0];
4503 case KVM_CAP_DIRTY_LOG_RING:
4504 return kvm_vm_ioctl_enable_dirty_log_ring(kvm, cap->args[0]);
4506 return kvm_vm_ioctl_enable_cap(kvm, cap);
4510 static ssize_t kvm_vm_stats_read(struct file *file, char __user *user_buffer,
4511 size_t size, loff_t *offset)
4513 struct kvm *kvm = file->private_data;
4515 return kvm_stats_read(kvm->stats_id, &kvm_vm_stats_header,
4516 &kvm_vm_stats_desc[0], &kvm->stat,
4517 sizeof(kvm->stat), user_buffer, size, offset);
4520 static const struct file_operations kvm_vm_stats_fops = {
4521 .read = kvm_vm_stats_read,
4522 .llseek = noop_llseek,
4525 static int kvm_vm_ioctl_get_stats_fd(struct kvm *kvm)
4530 fd = get_unused_fd_flags(O_CLOEXEC);
4534 file = anon_inode_getfile("kvm-vm-stats",
4535 &kvm_vm_stats_fops, kvm, O_RDONLY);
4538 return PTR_ERR(file);
4540 file->f_mode |= FMODE_PREAD;
4541 fd_install(fd, file);
4546 static long kvm_vm_ioctl(struct file *filp,
4547 unsigned int ioctl, unsigned long arg)
4549 struct kvm *kvm = filp->private_data;
4550 void __user *argp = (void __user *)arg;
4553 if (kvm->mm != current->mm || kvm->vm_dead)
4556 case KVM_CREATE_VCPU:
4557 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
4559 case KVM_ENABLE_CAP: {
4560 struct kvm_enable_cap cap;
4563 if (copy_from_user(&cap, argp, sizeof(cap)))
4565 r = kvm_vm_ioctl_enable_cap_generic(kvm, &cap);
4568 case KVM_SET_USER_MEMORY_REGION: {
4569 struct kvm_userspace_memory_region kvm_userspace_mem;
4572 if (copy_from_user(&kvm_userspace_mem, argp,
4573 sizeof(kvm_userspace_mem)))
4576 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
4579 case KVM_GET_DIRTY_LOG: {
4580 struct kvm_dirty_log log;
4583 if (copy_from_user(&log, argp, sizeof(log)))
4585 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
4588 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4589 case KVM_CLEAR_DIRTY_LOG: {
4590 struct kvm_clear_dirty_log log;
4593 if (copy_from_user(&log, argp, sizeof(log)))
4595 r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
4599 #ifdef CONFIG_KVM_MMIO
4600 case KVM_REGISTER_COALESCED_MMIO: {
4601 struct kvm_coalesced_mmio_zone zone;
4604 if (copy_from_user(&zone, argp, sizeof(zone)))
4606 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
4609 case KVM_UNREGISTER_COALESCED_MMIO: {
4610 struct kvm_coalesced_mmio_zone zone;
4613 if (copy_from_user(&zone, argp, sizeof(zone)))
4615 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
4620 struct kvm_irqfd data;
4623 if (copy_from_user(&data, argp, sizeof(data)))
4625 r = kvm_irqfd(kvm, &data);
4628 case KVM_IOEVENTFD: {
4629 struct kvm_ioeventfd data;
4632 if (copy_from_user(&data, argp, sizeof(data)))
4634 r = kvm_ioeventfd(kvm, &data);
4637 #ifdef CONFIG_HAVE_KVM_MSI
4638 case KVM_SIGNAL_MSI: {
4642 if (copy_from_user(&msi, argp, sizeof(msi)))
4644 r = kvm_send_userspace_msi(kvm, &msi);
4648 #ifdef __KVM_HAVE_IRQ_LINE
4649 case KVM_IRQ_LINE_STATUS:
4650 case KVM_IRQ_LINE: {
4651 struct kvm_irq_level irq_event;
4654 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
4657 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
4658 ioctl == KVM_IRQ_LINE_STATUS);
4663 if (ioctl == KVM_IRQ_LINE_STATUS) {
4664 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
4672 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
4673 case KVM_SET_GSI_ROUTING: {
4674 struct kvm_irq_routing routing;
4675 struct kvm_irq_routing __user *urouting;
4676 struct kvm_irq_routing_entry *entries = NULL;
4679 if (copy_from_user(&routing, argp, sizeof(routing)))
4682 if (!kvm_arch_can_set_irq_routing(kvm))
4684 if (routing.nr > KVM_MAX_IRQ_ROUTES)
4690 entries = vmemdup_user(urouting->entries,
4691 array_size(sizeof(*entries),
4693 if (IS_ERR(entries)) {
4694 r = PTR_ERR(entries);
4698 r = kvm_set_irq_routing(kvm, entries, routing.nr,
4703 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
4704 case KVM_CREATE_DEVICE: {
4705 struct kvm_create_device cd;
4708 if (copy_from_user(&cd, argp, sizeof(cd)))
4711 r = kvm_ioctl_create_device(kvm, &cd);
4716 if (copy_to_user(argp, &cd, sizeof(cd)))
4722 case KVM_CHECK_EXTENSION:
4723 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
4725 case KVM_RESET_DIRTY_RINGS:
4726 r = kvm_vm_ioctl_reset_dirty_pages(kvm);
4728 case KVM_GET_STATS_FD:
4729 r = kvm_vm_ioctl_get_stats_fd(kvm);
4732 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
4738 #ifdef CONFIG_KVM_COMPAT
4739 struct compat_kvm_dirty_log {
4743 compat_uptr_t dirty_bitmap; /* one bit per page */
4748 struct compat_kvm_clear_dirty_log {
4753 compat_uptr_t dirty_bitmap; /* one bit per page */
4758 static long kvm_vm_compat_ioctl(struct file *filp,
4759 unsigned int ioctl, unsigned long arg)
4761 struct kvm *kvm = filp->private_data;
4764 if (kvm->mm != current->mm || kvm->vm_dead)
4767 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4768 case KVM_CLEAR_DIRTY_LOG: {
4769 struct compat_kvm_clear_dirty_log compat_log;
4770 struct kvm_clear_dirty_log log;
4772 if (copy_from_user(&compat_log, (void __user *)arg,
4773 sizeof(compat_log)))
4775 log.slot = compat_log.slot;
4776 log.num_pages = compat_log.num_pages;
4777 log.first_page = compat_log.first_page;
4778 log.padding2 = compat_log.padding2;
4779 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
4781 r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
4785 case KVM_GET_DIRTY_LOG: {
4786 struct compat_kvm_dirty_log compat_log;
4787 struct kvm_dirty_log log;
4789 if (copy_from_user(&compat_log, (void __user *)arg,
4790 sizeof(compat_log)))
4792 log.slot = compat_log.slot;
4793 log.padding1 = compat_log.padding1;
4794 log.padding2 = compat_log.padding2;
4795 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
4797 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
4801 r = kvm_vm_ioctl(filp, ioctl, arg);
4807 static const struct file_operations kvm_vm_fops = {
4808 .release = kvm_vm_release,
4809 .unlocked_ioctl = kvm_vm_ioctl,
4810 .llseek = noop_llseek,
4811 KVM_COMPAT(kvm_vm_compat_ioctl),
4814 bool file_is_kvm(struct file *file)
4816 return file && file->f_op == &kvm_vm_fops;
4818 EXPORT_SYMBOL_GPL(file_is_kvm);
4820 static int kvm_dev_ioctl_create_vm(unsigned long type)
4826 kvm = kvm_create_vm(type);
4828 return PTR_ERR(kvm);
4829 #ifdef CONFIG_KVM_MMIO
4830 r = kvm_coalesced_mmio_init(kvm);
4834 r = get_unused_fd_flags(O_CLOEXEC);
4838 snprintf(kvm->stats_id, sizeof(kvm->stats_id),
4839 "kvm-%d", task_pid_nr(current));
4841 file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
4849 * Don't call kvm_put_kvm anymore at this point; file->f_op is
4850 * already set, with ->release() being kvm_vm_release(). In error
4851 * cases it will be called by the final fput(file) and will take
4852 * care of doing kvm_put_kvm(kvm).
4854 if (kvm_create_vm_debugfs(kvm, r) < 0) {
4859 kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm);
4861 fd_install(r, file);
4869 static long kvm_dev_ioctl(struct file *filp,
4870 unsigned int ioctl, unsigned long arg)
4875 case KVM_GET_API_VERSION:
4878 r = KVM_API_VERSION;
4881 r = kvm_dev_ioctl_create_vm(arg);
4883 case KVM_CHECK_EXTENSION:
4884 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
4886 case KVM_GET_VCPU_MMAP_SIZE:
4889 r = PAGE_SIZE; /* struct kvm_run */
4891 r += PAGE_SIZE; /* pio data page */
4893 #ifdef CONFIG_KVM_MMIO
4894 r += PAGE_SIZE; /* coalesced mmio ring page */
4897 case KVM_TRACE_ENABLE:
4898 case KVM_TRACE_PAUSE:
4899 case KVM_TRACE_DISABLE:
4903 return kvm_arch_dev_ioctl(filp, ioctl, arg);
4909 static struct file_operations kvm_chardev_ops = {
4910 .unlocked_ioctl = kvm_dev_ioctl,
4911 .llseek = noop_llseek,
4912 KVM_COMPAT(kvm_dev_ioctl),
4915 static struct miscdevice kvm_dev = {
4921 static void hardware_enable_nolock(void *junk)
4923 int cpu = raw_smp_processor_id();
4926 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
4929 cpumask_set_cpu(cpu, cpus_hardware_enabled);
4931 r = kvm_arch_hardware_enable();
4934 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
4935 atomic_inc(&hardware_enable_failed);
4936 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
4940 static int kvm_starting_cpu(unsigned int cpu)
4942 raw_spin_lock(&kvm_count_lock);
4943 if (kvm_usage_count)
4944 hardware_enable_nolock(NULL);
4945 raw_spin_unlock(&kvm_count_lock);
4949 static void hardware_disable_nolock(void *junk)
4951 int cpu = raw_smp_processor_id();
4953 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
4955 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
4956 kvm_arch_hardware_disable();
4959 static int kvm_dying_cpu(unsigned int cpu)
4961 raw_spin_lock(&kvm_count_lock);
4962 if (kvm_usage_count)
4963 hardware_disable_nolock(NULL);
4964 raw_spin_unlock(&kvm_count_lock);
4968 static void hardware_disable_all_nolock(void)
4970 BUG_ON(!kvm_usage_count);
4973 if (!kvm_usage_count)
4974 on_each_cpu(hardware_disable_nolock, NULL, 1);
4977 static void hardware_disable_all(void)
4979 raw_spin_lock(&kvm_count_lock);
4980 hardware_disable_all_nolock();
4981 raw_spin_unlock(&kvm_count_lock);
4984 static int hardware_enable_all(void)
4988 raw_spin_lock(&kvm_count_lock);
4991 if (kvm_usage_count == 1) {
4992 atomic_set(&hardware_enable_failed, 0);
4993 on_each_cpu(hardware_enable_nolock, NULL, 1);
4995 if (atomic_read(&hardware_enable_failed)) {
4996 hardware_disable_all_nolock();
5001 raw_spin_unlock(&kvm_count_lock);
5006 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
5010 * Some (well, at least mine) BIOSes hang on reboot if
5013 * And Intel TXT required VMX off for all cpu when system shutdown.
5015 pr_info("kvm: exiting hardware virtualization\n");
5016 kvm_rebooting = true;
5017 on_each_cpu(hardware_disable_nolock, NULL, 1);
5021 static struct notifier_block kvm_reboot_notifier = {
5022 .notifier_call = kvm_reboot,
5026 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
5030 for (i = 0; i < bus->dev_count; i++) {
5031 struct kvm_io_device *pos = bus->range[i].dev;
5033 kvm_iodevice_destructor(pos);
5038 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
5039 const struct kvm_io_range *r2)
5041 gpa_t addr1 = r1->addr;
5042 gpa_t addr2 = r2->addr;
5047 /* If r2->len == 0, match the exact address. If r2->len != 0,
5048 * accept any overlapping write. Any order is acceptable for
5049 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
5050 * we process all of them.
5063 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
5065 return kvm_io_bus_cmp(p1, p2);
5068 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
5069 gpa_t addr, int len)
5071 struct kvm_io_range *range, key;
5074 key = (struct kvm_io_range) {
5079 range = bsearch(&key, bus->range, bus->dev_count,
5080 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
5084 off = range - bus->range;
5086 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
5092 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
5093 struct kvm_io_range *range, const void *val)
5097 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
5101 while (idx < bus->dev_count &&
5102 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
5103 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
5112 /* kvm_io_bus_write - called under kvm->slots_lock */
5113 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
5114 int len, const void *val)
5116 struct kvm_io_bus *bus;
5117 struct kvm_io_range range;
5120 range = (struct kvm_io_range) {
5125 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
5128 r = __kvm_io_bus_write(vcpu, bus, &range, val);
5129 return r < 0 ? r : 0;
5131 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
5133 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
5134 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
5135 gpa_t addr, int len, const void *val, long cookie)
5137 struct kvm_io_bus *bus;
5138 struct kvm_io_range range;
5140 range = (struct kvm_io_range) {
5145 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
5149 /* First try the device referenced by cookie. */
5150 if ((cookie >= 0) && (cookie < bus->dev_count) &&
5151 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
5152 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
5157 * cookie contained garbage; fall back to search and return the
5158 * correct cookie value.
5160 return __kvm_io_bus_write(vcpu, bus, &range, val);
5163 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
5164 struct kvm_io_range *range, void *val)
5168 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
5172 while (idx < bus->dev_count &&
5173 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
5174 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
5183 /* kvm_io_bus_read - called under kvm->slots_lock */
5184 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
5187 struct kvm_io_bus *bus;
5188 struct kvm_io_range range;
5191 range = (struct kvm_io_range) {
5196 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
5199 r = __kvm_io_bus_read(vcpu, bus, &range, val);
5200 return r < 0 ? r : 0;
5203 /* Caller must hold slots_lock. */
5204 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
5205 int len, struct kvm_io_device *dev)
5208 struct kvm_io_bus *new_bus, *bus;
5209 struct kvm_io_range range;
5211 bus = kvm_get_bus(kvm, bus_idx);
5215 /* exclude ioeventfd which is limited by maximum fd */
5216 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
5219 new_bus = kmalloc(struct_size(bus, range, bus->dev_count + 1),
5220 GFP_KERNEL_ACCOUNT);
5224 range = (struct kvm_io_range) {
5230 for (i = 0; i < bus->dev_count; i++)
5231 if (kvm_io_bus_cmp(&bus->range[i], &range) > 0)
5234 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
5235 new_bus->dev_count++;
5236 new_bus->range[i] = range;
5237 memcpy(new_bus->range + i + 1, bus->range + i,
5238 (bus->dev_count - i) * sizeof(struct kvm_io_range));
5239 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
5240 synchronize_srcu_expedited(&kvm->srcu);
5246 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
5247 struct kvm_io_device *dev)
5250 struct kvm_io_bus *new_bus, *bus;
5252 lockdep_assert_held(&kvm->slots_lock);
5254 bus = kvm_get_bus(kvm, bus_idx);
5258 for (i = 0; i < bus->dev_count; i++) {
5259 if (bus->range[i].dev == dev) {
5264 if (i == bus->dev_count)
5267 new_bus = kmalloc(struct_size(bus, range, bus->dev_count - 1),
5268 GFP_KERNEL_ACCOUNT);
5270 memcpy(new_bus, bus, struct_size(bus, range, i));
5271 new_bus->dev_count--;
5272 memcpy(new_bus->range + i, bus->range + i + 1,
5273 flex_array_size(new_bus, range, new_bus->dev_count - i));
5276 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
5277 synchronize_srcu_expedited(&kvm->srcu);
5279 /* Destroy the old bus _after_ installing the (null) bus. */
5281 pr_err("kvm: failed to shrink bus, removing it completely\n");
5282 for (j = 0; j < bus->dev_count; j++) {
5285 kvm_iodevice_destructor(bus->range[j].dev);
5290 return new_bus ? 0 : -ENOMEM;
5293 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
5296 struct kvm_io_bus *bus;
5297 int dev_idx, srcu_idx;
5298 struct kvm_io_device *iodev = NULL;
5300 srcu_idx = srcu_read_lock(&kvm->srcu);
5302 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
5306 dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
5310 iodev = bus->range[dev_idx].dev;
5313 srcu_read_unlock(&kvm->srcu, srcu_idx);
5317 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
5319 static int kvm_debugfs_open(struct inode *inode, struct file *file,
5320 int (*get)(void *, u64 *), int (*set)(void *, u64),
5323 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
5327 * The debugfs files are a reference to the kvm struct which
5328 * is still valid when kvm_destroy_vm is called. kvm_get_kvm_safe
5329 * avoids the race between open and the removal of the debugfs directory.
5331 if (!kvm_get_kvm_safe(stat_data->kvm))
5334 if (simple_attr_open(inode, file, get,
5335 kvm_stats_debugfs_mode(stat_data->desc) & 0222
5338 kvm_put_kvm(stat_data->kvm);
5345 static int kvm_debugfs_release(struct inode *inode, struct file *file)
5347 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
5350 simple_attr_release(inode, file);
5351 kvm_put_kvm(stat_data->kvm);
5356 static int kvm_get_stat_per_vm(struct kvm *kvm, size_t offset, u64 *val)
5358 *val = *(u64 *)((void *)(&kvm->stat) + offset);
5363 static int kvm_clear_stat_per_vm(struct kvm *kvm, size_t offset)
5365 *(u64 *)((void *)(&kvm->stat) + offset) = 0;
5370 static int kvm_get_stat_per_vcpu(struct kvm *kvm, size_t offset, u64 *val)
5373 struct kvm_vcpu *vcpu;
5377 kvm_for_each_vcpu(i, vcpu, kvm)
5378 *val += *(u64 *)((void *)(&vcpu->stat) + offset);
5383 static int kvm_clear_stat_per_vcpu(struct kvm *kvm, size_t offset)
5386 struct kvm_vcpu *vcpu;
5388 kvm_for_each_vcpu(i, vcpu, kvm)
5389 *(u64 *)((void *)(&vcpu->stat) + offset) = 0;
5394 static int kvm_stat_data_get(void *data, u64 *val)
5397 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
5399 switch (stat_data->kind) {
5401 r = kvm_get_stat_per_vm(stat_data->kvm,
5402 stat_data->desc->desc.offset, val);
5405 r = kvm_get_stat_per_vcpu(stat_data->kvm,
5406 stat_data->desc->desc.offset, val);
5413 static int kvm_stat_data_clear(void *data, u64 val)
5416 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
5421 switch (stat_data->kind) {
5423 r = kvm_clear_stat_per_vm(stat_data->kvm,
5424 stat_data->desc->desc.offset);
5427 r = kvm_clear_stat_per_vcpu(stat_data->kvm,
5428 stat_data->desc->desc.offset);
5435 static int kvm_stat_data_open(struct inode *inode, struct file *file)
5437 __simple_attr_check_format("%llu\n", 0ull);
5438 return kvm_debugfs_open(inode, file, kvm_stat_data_get,
5439 kvm_stat_data_clear, "%llu\n");
5442 static const struct file_operations stat_fops_per_vm = {
5443 .owner = THIS_MODULE,
5444 .open = kvm_stat_data_open,
5445 .release = kvm_debugfs_release,
5446 .read = simple_attr_read,
5447 .write = simple_attr_write,
5448 .llseek = no_llseek,
5451 static int vm_stat_get(void *_offset, u64 *val)
5453 unsigned offset = (long)_offset;
5458 mutex_lock(&kvm_lock);
5459 list_for_each_entry(kvm, &vm_list, vm_list) {
5460 kvm_get_stat_per_vm(kvm, offset, &tmp_val);
5463 mutex_unlock(&kvm_lock);
5467 static int vm_stat_clear(void *_offset, u64 val)
5469 unsigned offset = (long)_offset;
5475 mutex_lock(&kvm_lock);
5476 list_for_each_entry(kvm, &vm_list, vm_list) {
5477 kvm_clear_stat_per_vm(kvm, offset);
5479 mutex_unlock(&kvm_lock);
5484 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n");
5485 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_readonly_fops, vm_stat_get, NULL, "%llu\n");
5487 static int vcpu_stat_get(void *_offset, u64 *val)
5489 unsigned offset = (long)_offset;
5494 mutex_lock(&kvm_lock);
5495 list_for_each_entry(kvm, &vm_list, vm_list) {
5496 kvm_get_stat_per_vcpu(kvm, offset, &tmp_val);
5499 mutex_unlock(&kvm_lock);
5503 static int vcpu_stat_clear(void *_offset, u64 val)
5505 unsigned offset = (long)_offset;
5511 mutex_lock(&kvm_lock);
5512 list_for_each_entry(kvm, &vm_list, vm_list) {
5513 kvm_clear_stat_per_vcpu(kvm, offset);
5515 mutex_unlock(&kvm_lock);
5520 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear,
5522 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_readonly_fops, vcpu_stat_get, NULL, "%llu\n");
5524 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm)
5526 struct kobj_uevent_env *env;
5527 unsigned long long created, active;
5529 if (!kvm_dev.this_device || !kvm)
5532 mutex_lock(&kvm_lock);
5533 if (type == KVM_EVENT_CREATE_VM) {
5534 kvm_createvm_count++;
5536 } else if (type == KVM_EVENT_DESTROY_VM) {
5539 created = kvm_createvm_count;
5540 active = kvm_active_vms;
5541 mutex_unlock(&kvm_lock);
5543 env = kzalloc(sizeof(*env), GFP_KERNEL_ACCOUNT);
5547 add_uevent_var(env, "CREATED=%llu", created);
5548 add_uevent_var(env, "COUNT=%llu", active);
5550 if (type == KVM_EVENT_CREATE_VM) {
5551 add_uevent_var(env, "EVENT=create");
5552 kvm->userspace_pid = task_pid_nr(current);
5553 } else if (type == KVM_EVENT_DESTROY_VM) {
5554 add_uevent_var(env, "EVENT=destroy");
5556 add_uevent_var(env, "PID=%d", kvm->userspace_pid);
5558 if (!IS_ERR(kvm->debugfs_dentry)) {
5559 char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL_ACCOUNT);
5562 tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX);
5564 add_uevent_var(env, "STATS_PATH=%s", tmp);
5568 /* no need for checks, since we are adding at most only 5 keys */
5569 env->envp[env->envp_idx++] = NULL;
5570 kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp);
5574 static void kvm_init_debug(void)
5576 const struct file_operations *fops;
5577 const struct _kvm_stats_desc *pdesc;
5580 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
5582 for (i = 0; i < kvm_vm_stats_header.num_desc; ++i) {
5583 pdesc = &kvm_vm_stats_desc[i];
5584 if (kvm_stats_debugfs_mode(pdesc) & 0222)
5585 fops = &vm_stat_fops;
5587 fops = &vm_stat_readonly_fops;
5588 debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
5590 (void *)(long)pdesc->desc.offset, fops);
5593 for (i = 0; i < kvm_vcpu_stats_header.num_desc; ++i) {
5594 pdesc = &kvm_vcpu_stats_desc[i];
5595 if (kvm_stats_debugfs_mode(pdesc) & 0222)
5596 fops = &vcpu_stat_fops;
5598 fops = &vcpu_stat_readonly_fops;
5599 debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
5601 (void *)(long)pdesc->desc.offset, fops);
5605 static int kvm_suspend(void)
5607 if (kvm_usage_count)
5608 hardware_disable_nolock(NULL);
5612 static void kvm_resume(void)
5614 if (kvm_usage_count) {
5615 lockdep_assert_not_held(&kvm_count_lock);
5616 hardware_enable_nolock(NULL);
5620 static struct syscore_ops kvm_syscore_ops = {
5621 .suspend = kvm_suspend,
5622 .resume = kvm_resume,
5626 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
5628 return container_of(pn, struct kvm_vcpu, preempt_notifier);
5631 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
5633 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
5635 WRITE_ONCE(vcpu->preempted, false);
5636 WRITE_ONCE(vcpu->ready, false);
5638 __this_cpu_write(kvm_running_vcpu, vcpu);
5639 kvm_arch_sched_in(vcpu, cpu);
5640 kvm_arch_vcpu_load(vcpu, cpu);
5643 static void kvm_sched_out(struct preempt_notifier *pn,
5644 struct task_struct *next)
5646 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
5648 if (current->on_rq) {
5649 WRITE_ONCE(vcpu->preempted, true);
5650 WRITE_ONCE(vcpu->ready, true);
5652 kvm_arch_vcpu_put(vcpu);
5653 __this_cpu_write(kvm_running_vcpu, NULL);
5657 * kvm_get_running_vcpu - get the vcpu running on the current CPU.
5659 * We can disable preemption locally around accessing the per-CPU variable,
5660 * and use the resolved vcpu pointer after enabling preemption again,
5661 * because even if the current thread is migrated to another CPU, reading
5662 * the per-CPU value later will give us the same value as we update the
5663 * per-CPU variable in the preempt notifier handlers.
5665 struct kvm_vcpu *kvm_get_running_vcpu(void)
5667 struct kvm_vcpu *vcpu;
5670 vcpu = __this_cpu_read(kvm_running_vcpu);
5675 EXPORT_SYMBOL_GPL(kvm_get_running_vcpu);
5678 * kvm_get_running_vcpus - get the per-CPU array of currently running vcpus.
5680 struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void)
5682 return &kvm_running_vcpu;
5685 #ifdef CONFIG_GUEST_PERF_EVENTS
5686 static unsigned int kvm_guest_state(void)
5688 struct kvm_vcpu *vcpu = kvm_get_running_vcpu();
5691 if (!kvm_arch_pmi_in_guest(vcpu))
5694 state = PERF_GUEST_ACTIVE;
5695 if (!kvm_arch_vcpu_in_kernel(vcpu))
5696 state |= PERF_GUEST_USER;
5701 static unsigned long kvm_guest_get_ip(void)
5703 struct kvm_vcpu *vcpu = kvm_get_running_vcpu();
5705 /* Retrieving the IP must be guarded by a call to kvm_guest_state(). */
5706 if (WARN_ON_ONCE(!kvm_arch_pmi_in_guest(vcpu)))
5709 return kvm_arch_vcpu_get_ip(vcpu);
5712 static struct perf_guest_info_callbacks kvm_guest_cbs = {
5713 .state = kvm_guest_state,
5714 .get_ip = kvm_guest_get_ip,
5715 .handle_intel_pt_intr = NULL,
5718 void kvm_register_perf_callbacks(unsigned int (*pt_intr_handler)(void))
5720 kvm_guest_cbs.handle_intel_pt_intr = pt_intr_handler;
5721 perf_register_guest_info_callbacks(&kvm_guest_cbs);
5723 void kvm_unregister_perf_callbacks(void)
5725 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
5729 struct kvm_cpu_compat_check {
5734 static void check_processor_compat(void *data)
5736 struct kvm_cpu_compat_check *c = data;
5738 *c->ret = kvm_arch_check_processor_compat(c->opaque);
5741 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
5742 struct module *module)
5744 struct kvm_cpu_compat_check c;
5748 r = kvm_arch_init(opaque);
5753 * kvm_arch_init makes sure there's at most one caller
5754 * for architectures that support multiple implementations,
5755 * like intel and amd on x86.
5756 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
5757 * conflicts in case kvm is already setup for another implementation.
5759 r = kvm_irqfd_init();
5763 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
5768 r = kvm_arch_hardware_setup(opaque);
5774 for_each_online_cpu(cpu) {
5775 smp_call_function_single(cpu, check_processor_compat, &c, 1);
5780 r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting",
5781 kvm_starting_cpu, kvm_dying_cpu);
5784 register_reboot_notifier(&kvm_reboot_notifier);
5786 /* A kmem cache lets us meet the alignment requirements of fx_save. */
5788 vcpu_align = __alignof__(struct kvm_vcpu);
5790 kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align,
5792 offsetof(struct kvm_vcpu, arch),
5793 offsetofend(struct kvm_vcpu, stats_id)
5794 - offsetof(struct kvm_vcpu, arch),
5796 if (!kvm_vcpu_cache) {
5801 for_each_possible_cpu(cpu) {
5802 if (!alloc_cpumask_var_node(&per_cpu(cpu_kick_mask, cpu),
5803 GFP_KERNEL, cpu_to_node(cpu))) {
5809 r = kvm_async_pf_init();
5813 kvm_chardev_ops.owner = module;
5815 r = misc_register(&kvm_dev);
5817 pr_err("kvm: misc device register failed\n");
5821 register_syscore_ops(&kvm_syscore_ops);
5823 kvm_preempt_ops.sched_in = kvm_sched_in;
5824 kvm_preempt_ops.sched_out = kvm_sched_out;
5828 r = kvm_vfio_ops_init();
5834 kvm_async_pf_deinit();
5836 for_each_possible_cpu(cpu)
5837 free_cpumask_var(per_cpu(cpu_kick_mask, cpu));
5839 kmem_cache_destroy(kvm_vcpu_cache);
5841 unregister_reboot_notifier(&kvm_reboot_notifier);
5842 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
5844 kvm_arch_hardware_unsetup();
5846 free_cpumask_var(cpus_hardware_enabled);
5854 EXPORT_SYMBOL_GPL(kvm_init);
5860 debugfs_remove_recursive(kvm_debugfs_dir);
5861 misc_deregister(&kvm_dev);
5862 for_each_possible_cpu(cpu)
5863 free_cpumask_var(per_cpu(cpu_kick_mask, cpu));
5864 kmem_cache_destroy(kvm_vcpu_cache);
5865 kvm_async_pf_deinit();
5866 unregister_syscore_ops(&kvm_syscore_ops);
5867 unregister_reboot_notifier(&kvm_reboot_notifier);
5868 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
5869 on_each_cpu(hardware_disable_nolock, NULL, 1);
5870 kvm_arch_hardware_unsetup();
5873 free_cpumask_var(cpus_hardware_enabled);
5874 kvm_vfio_ops_exit();
5876 EXPORT_SYMBOL_GPL(kvm_exit);
5878 struct kvm_vm_worker_thread_context {
5880 struct task_struct *parent;
5881 struct completion init_done;
5882 kvm_vm_thread_fn_t thread_fn;
5887 static int kvm_vm_worker_thread(void *context)
5890 * The init_context is allocated on the stack of the parent thread, so
5891 * we have to locally copy anything that is needed beyond initialization
5893 struct kvm_vm_worker_thread_context *init_context = context;
5894 struct task_struct *parent;
5895 struct kvm *kvm = init_context->kvm;
5896 kvm_vm_thread_fn_t thread_fn = init_context->thread_fn;
5897 uintptr_t data = init_context->data;
5900 err = kthread_park(current);
5901 /* kthread_park(current) is never supposed to return an error */
5906 err = cgroup_attach_task_all(init_context->parent, current);
5908 kvm_err("%s: cgroup_attach_task_all failed with err %d\n",
5913 set_user_nice(current, task_nice(init_context->parent));
5916 init_context->err = err;
5917 complete(&init_context->init_done);
5918 init_context = NULL;
5923 /* Wait to be woken up by the spawner before proceeding. */
5926 if (!kthread_should_stop())
5927 err = thread_fn(kvm, data);
5931 * Move kthread back to its original cgroup to prevent it lingering in
5932 * the cgroup of the VM process, after the latter finishes its
5935 * kthread_stop() waits on the 'exited' completion condition which is
5936 * set in exit_mm(), via mm_release(), in do_exit(). However, the
5937 * kthread is removed from the cgroup in the cgroup_exit() which is
5938 * called after the exit_mm(). This causes the kthread_stop() to return
5939 * before the kthread actually quits the cgroup.
5942 parent = rcu_dereference(current->real_parent);
5943 get_task_struct(parent);
5945 cgroup_attach_task_all(parent, current);
5946 put_task_struct(parent);
5951 int kvm_vm_create_worker_thread(struct kvm *kvm, kvm_vm_thread_fn_t thread_fn,
5952 uintptr_t data, const char *name,
5953 struct task_struct **thread_ptr)
5955 struct kvm_vm_worker_thread_context init_context = {};
5956 struct task_struct *thread;
5959 init_context.kvm = kvm;
5960 init_context.parent = current;
5961 init_context.thread_fn = thread_fn;
5962 init_context.data = data;
5963 init_completion(&init_context.init_done);
5965 thread = kthread_run(kvm_vm_worker_thread, &init_context,
5966 "%s-%d", name, task_pid_nr(current));
5968 return PTR_ERR(thread);
5970 /* kthread_run is never supposed to return NULL */
5971 WARN_ON(thread == NULL);
5973 wait_for_completion(&init_context.init_done);
5975 if (!init_context.err)
5976 *thread_ptr = thread;
5978 return init_context.err;