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_flush(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_flush, 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);
727 gfn_to_pfn_cache_invalidate_start(kvm, range->start, range->end,
728 hva_range.may_block);
730 __kvm_handle_hva_range(kvm, &hva_range);
735 void kvm_dec_notifier_count(struct kvm *kvm, unsigned long start,
739 * This sequence increase will notify the kvm page fault that
740 * the page that is going to be mapped in the spte could have
743 kvm->mmu_notifier_seq++;
746 * The above sequence increase must be visible before the
747 * below count decrease, which is ensured by the smp_wmb above
748 * in conjunction with the smp_rmb in mmu_notifier_retry().
750 kvm->mmu_notifier_count--;
753 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
754 const struct mmu_notifier_range *range)
756 struct kvm *kvm = mmu_notifier_to_kvm(mn);
757 const struct kvm_hva_range hva_range = {
758 .start = range->start,
761 .handler = (void *)kvm_null_fn,
762 .on_lock = kvm_dec_notifier_count,
763 .on_unlock = (void *)kvm_null_fn,
764 .flush_on_ret = false,
765 .may_block = mmu_notifier_range_blockable(range),
769 __kvm_handle_hva_range(kvm, &hva_range);
771 /* Pairs with the increment in range_start(). */
772 spin_lock(&kvm->mn_invalidate_lock);
773 wake = (--kvm->mn_active_invalidate_count == 0);
774 spin_unlock(&kvm->mn_invalidate_lock);
777 * There can only be one waiter, since the wait happens under
781 rcuwait_wake_up(&kvm->mn_memslots_update_rcuwait);
783 BUG_ON(kvm->mmu_notifier_count < 0);
786 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
787 struct mm_struct *mm,
791 trace_kvm_age_hva(start, end);
793 return kvm_handle_hva_range(mn, start, end, __pte(0), kvm_age_gfn);
796 static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
797 struct mm_struct *mm,
801 trace_kvm_age_hva(start, end);
804 * Even though we do not flush TLB, this will still adversely
805 * affect performance on pre-Haswell Intel EPT, where there is
806 * no EPT Access Bit to clear so that we have to tear down EPT
807 * tables instead. If we find this unacceptable, we can always
808 * add a parameter to kvm_age_hva so that it effectively doesn't
809 * do anything on clear_young.
811 * Also note that currently we never issue secondary TLB flushes
812 * from clear_young, leaving this job up to the regular system
813 * cadence. If we find this inaccurate, we might come up with a
814 * more sophisticated heuristic later.
816 return kvm_handle_hva_range_no_flush(mn, start, end, kvm_age_gfn);
819 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
820 struct mm_struct *mm,
821 unsigned long address)
823 trace_kvm_test_age_hva(address);
825 return kvm_handle_hva_range_no_flush(mn, address, address + 1,
829 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
830 struct mm_struct *mm)
832 struct kvm *kvm = mmu_notifier_to_kvm(mn);
835 idx = srcu_read_lock(&kvm->srcu);
836 kvm_flush_shadow_all(kvm);
837 srcu_read_unlock(&kvm->srcu, idx);
840 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
841 .invalidate_range = kvm_mmu_notifier_invalidate_range,
842 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
843 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
844 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
845 .clear_young = kvm_mmu_notifier_clear_young,
846 .test_young = kvm_mmu_notifier_test_young,
847 .change_pte = kvm_mmu_notifier_change_pte,
848 .release = kvm_mmu_notifier_release,
851 static int kvm_init_mmu_notifier(struct kvm *kvm)
853 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
854 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
857 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
859 static int kvm_init_mmu_notifier(struct kvm *kvm)
864 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
866 #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER
867 static int kvm_pm_notifier_call(struct notifier_block *bl,
871 struct kvm *kvm = container_of(bl, struct kvm, pm_notifier);
873 return kvm_arch_pm_notifier(kvm, state);
876 static void kvm_init_pm_notifier(struct kvm *kvm)
878 kvm->pm_notifier.notifier_call = kvm_pm_notifier_call;
879 /* Suspend KVM before we suspend ftrace, RCU, etc. */
880 kvm->pm_notifier.priority = INT_MAX;
881 register_pm_notifier(&kvm->pm_notifier);
884 static void kvm_destroy_pm_notifier(struct kvm *kvm)
886 unregister_pm_notifier(&kvm->pm_notifier);
888 #else /* !CONFIG_HAVE_KVM_PM_NOTIFIER */
889 static void kvm_init_pm_notifier(struct kvm *kvm)
893 static void kvm_destroy_pm_notifier(struct kvm *kvm)
896 #endif /* CONFIG_HAVE_KVM_PM_NOTIFIER */
898 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
900 if (!memslot->dirty_bitmap)
903 kvfree(memslot->dirty_bitmap);
904 memslot->dirty_bitmap = NULL;
907 /* This does not remove the slot from struct kvm_memslots data structures */
908 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
910 kvm_destroy_dirty_bitmap(slot);
912 kvm_arch_free_memslot(kvm, slot);
917 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
919 struct hlist_node *idnode;
920 struct kvm_memory_slot *memslot;
924 * The same memslot objects live in both active and inactive sets,
925 * arbitrarily free using index '1' so the second invocation of this
926 * function isn't operating over a structure with dangling pointers
927 * (even though this function isn't actually touching them).
929 if (!slots->node_idx)
932 hash_for_each_safe(slots->id_hash, bkt, idnode, memslot, id_node[1])
933 kvm_free_memslot(kvm, memslot);
936 static umode_t kvm_stats_debugfs_mode(const struct _kvm_stats_desc *pdesc)
938 switch (pdesc->desc.flags & KVM_STATS_TYPE_MASK) {
939 case KVM_STATS_TYPE_INSTANT:
941 case KVM_STATS_TYPE_CUMULATIVE:
942 case KVM_STATS_TYPE_PEAK:
949 static void kvm_destroy_vm_debugfs(struct kvm *kvm)
952 int kvm_debugfs_num_entries = kvm_vm_stats_header.num_desc +
953 kvm_vcpu_stats_header.num_desc;
955 if (IS_ERR(kvm->debugfs_dentry))
958 debugfs_remove_recursive(kvm->debugfs_dentry);
960 if (kvm->debugfs_stat_data) {
961 for (i = 0; i < kvm_debugfs_num_entries; i++)
962 kfree(kvm->debugfs_stat_data[i]);
963 kfree(kvm->debugfs_stat_data);
967 static int kvm_create_vm_debugfs(struct kvm *kvm, int fd)
969 static DEFINE_MUTEX(kvm_debugfs_lock);
971 char dir_name[ITOA_MAX_LEN * 2];
972 struct kvm_stat_data *stat_data;
973 const struct _kvm_stats_desc *pdesc;
975 int kvm_debugfs_num_entries = kvm_vm_stats_header.num_desc +
976 kvm_vcpu_stats_header.num_desc;
978 if (!debugfs_initialized())
981 snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd);
982 mutex_lock(&kvm_debugfs_lock);
983 dent = debugfs_lookup(dir_name, kvm_debugfs_dir);
985 pr_warn_ratelimited("KVM: debugfs: duplicate directory %s\n", dir_name);
987 mutex_unlock(&kvm_debugfs_lock);
990 dent = debugfs_create_dir(dir_name, kvm_debugfs_dir);
991 mutex_unlock(&kvm_debugfs_lock);
995 kvm->debugfs_dentry = dent;
996 kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries,
997 sizeof(*kvm->debugfs_stat_data),
999 if (!kvm->debugfs_stat_data)
1002 for (i = 0; i < kvm_vm_stats_header.num_desc; ++i) {
1003 pdesc = &kvm_vm_stats_desc[i];
1004 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT);
1008 stat_data->kvm = kvm;
1009 stat_data->desc = pdesc;
1010 stat_data->kind = KVM_STAT_VM;
1011 kvm->debugfs_stat_data[i] = stat_data;
1012 debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
1013 kvm->debugfs_dentry, stat_data,
1017 for (i = 0; i < kvm_vcpu_stats_header.num_desc; ++i) {
1018 pdesc = &kvm_vcpu_stats_desc[i];
1019 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT);
1023 stat_data->kvm = kvm;
1024 stat_data->desc = pdesc;
1025 stat_data->kind = KVM_STAT_VCPU;
1026 kvm->debugfs_stat_data[i + kvm_vm_stats_header.num_desc] = stat_data;
1027 debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
1028 kvm->debugfs_dentry, stat_data,
1032 ret = kvm_arch_create_vm_debugfs(kvm);
1034 kvm_destroy_vm_debugfs(kvm);
1042 * Called after the VM is otherwise initialized, but just before adding it to
1045 int __weak kvm_arch_post_init_vm(struct kvm *kvm)
1051 * Called just after removing the VM from the vm_list, but before doing any
1052 * other destruction.
1054 void __weak kvm_arch_pre_destroy_vm(struct kvm *kvm)
1059 * Called after per-vm debugfs created. When called kvm->debugfs_dentry should
1060 * be setup already, so we can create arch-specific debugfs entries under it.
1061 * Cleanup should be automatic done in kvm_destroy_vm_debugfs() recursively, so
1062 * a per-arch destroy interface is not needed.
1064 int __weak kvm_arch_create_vm_debugfs(struct kvm *kvm)
1069 static struct kvm *kvm_create_vm(unsigned long type)
1071 struct kvm *kvm = kvm_arch_alloc_vm();
1072 struct kvm_memslots *slots;
1077 return ERR_PTR(-ENOMEM);
1079 KVM_MMU_LOCK_INIT(kvm);
1080 mmgrab(current->mm);
1081 kvm->mm = current->mm;
1082 kvm_eventfd_init(kvm);
1083 mutex_init(&kvm->lock);
1084 mutex_init(&kvm->irq_lock);
1085 mutex_init(&kvm->slots_lock);
1086 mutex_init(&kvm->slots_arch_lock);
1087 spin_lock_init(&kvm->mn_invalidate_lock);
1088 rcuwait_init(&kvm->mn_memslots_update_rcuwait);
1089 xa_init(&kvm->vcpu_array);
1091 INIT_LIST_HEAD(&kvm->gpc_list);
1092 spin_lock_init(&kvm->gpc_lock);
1094 INIT_LIST_HEAD(&kvm->devices);
1096 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
1099 * Force subsequent debugfs file creations to fail if the VM directory
1100 * is not created (by kvm_create_vm_debugfs()).
1102 kvm->debugfs_dentry = ERR_PTR(-ENOENT);
1104 if (init_srcu_struct(&kvm->srcu))
1105 goto out_err_no_srcu;
1106 if (init_srcu_struct(&kvm->irq_srcu))
1107 goto out_err_no_irq_srcu;
1109 refcount_set(&kvm->users_count, 1);
1110 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
1111 for (j = 0; j < 2; j++) {
1112 slots = &kvm->__memslots[i][j];
1114 atomic_long_set(&slots->last_used_slot, (unsigned long)NULL);
1115 slots->hva_tree = RB_ROOT_CACHED;
1116 slots->gfn_tree = RB_ROOT;
1117 hash_init(slots->id_hash);
1118 slots->node_idx = j;
1120 /* Generations must be different for each address space. */
1121 slots->generation = i;
1124 rcu_assign_pointer(kvm->memslots[i], &kvm->__memslots[i][0]);
1127 for (i = 0; i < KVM_NR_BUSES; i++) {
1128 rcu_assign_pointer(kvm->buses[i],
1129 kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL_ACCOUNT));
1131 goto out_err_no_arch_destroy_vm;
1134 kvm->max_halt_poll_ns = halt_poll_ns;
1136 r = kvm_arch_init_vm(kvm, type);
1138 goto out_err_no_arch_destroy_vm;
1140 r = hardware_enable_all();
1142 goto out_err_no_disable;
1144 #ifdef CONFIG_HAVE_KVM_IRQFD
1145 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
1148 r = kvm_init_mmu_notifier(kvm);
1150 goto out_err_no_mmu_notifier;
1152 r = kvm_arch_post_init_vm(kvm);
1156 mutex_lock(&kvm_lock);
1157 list_add(&kvm->vm_list, &vm_list);
1158 mutex_unlock(&kvm_lock);
1160 preempt_notifier_inc();
1161 kvm_init_pm_notifier(kvm);
1164 * When the fd passed to this ioctl() is opened it pins the module,
1165 * but try_module_get() also prevents getting a reference if the module
1166 * is in MODULE_STATE_GOING (e.g. if someone ran "rmmod --wait").
1168 if (!try_module_get(kvm_chardev_ops.owner)) {
1176 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1177 if (kvm->mmu_notifier.ops)
1178 mmu_notifier_unregister(&kvm->mmu_notifier, current->mm);
1180 out_err_no_mmu_notifier:
1181 hardware_disable_all();
1183 kvm_arch_destroy_vm(kvm);
1184 out_err_no_arch_destroy_vm:
1185 WARN_ON_ONCE(!refcount_dec_and_test(&kvm->users_count));
1186 for (i = 0; i < KVM_NR_BUSES; i++)
1187 kfree(kvm_get_bus(kvm, i));
1188 cleanup_srcu_struct(&kvm->irq_srcu);
1189 out_err_no_irq_srcu:
1190 cleanup_srcu_struct(&kvm->srcu);
1192 kvm_arch_free_vm(kvm);
1193 mmdrop(current->mm);
1197 static void kvm_destroy_devices(struct kvm *kvm)
1199 struct kvm_device *dev, *tmp;
1202 * We do not need to take the kvm->lock here, because nobody else
1203 * has a reference to the struct kvm at this point and therefore
1204 * cannot access the devices list anyhow.
1206 list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
1207 list_del(&dev->vm_node);
1208 dev->ops->destroy(dev);
1212 static void kvm_destroy_vm(struct kvm *kvm)
1215 struct mm_struct *mm = kvm->mm;
1217 kvm_destroy_pm_notifier(kvm);
1218 kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm);
1219 kvm_destroy_vm_debugfs(kvm);
1220 kvm_arch_sync_events(kvm);
1221 mutex_lock(&kvm_lock);
1222 list_del(&kvm->vm_list);
1223 mutex_unlock(&kvm_lock);
1224 kvm_arch_pre_destroy_vm(kvm);
1226 kvm_free_irq_routing(kvm);
1227 for (i = 0; i < KVM_NR_BUSES; i++) {
1228 struct kvm_io_bus *bus = kvm_get_bus(kvm, i);
1231 kvm_io_bus_destroy(bus);
1232 kvm->buses[i] = NULL;
1234 kvm_coalesced_mmio_free(kvm);
1235 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1236 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
1238 * At this point, pending calls to invalidate_range_start()
1239 * have completed but no more MMU notifiers will run, so
1240 * mn_active_invalidate_count may remain unbalanced.
1241 * No threads can be waiting in install_new_memslots as the
1242 * last reference on KVM has been dropped, but freeing
1243 * memslots would deadlock without this manual intervention.
1245 WARN_ON(rcuwait_active(&kvm->mn_memslots_update_rcuwait));
1246 kvm->mn_active_invalidate_count = 0;
1248 kvm_flush_shadow_all(kvm);
1250 kvm_arch_destroy_vm(kvm);
1251 kvm_destroy_devices(kvm);
1252 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
1253 kvm_free_memslots(kvm, &kvm->__memslots[i][0]);
1254 kvm_free_memslots(kvm, &kvm->__memslots[i][1]);
1256 cleanup_srcu_struct(&kvm->irq_srcu);
1257 cleanup_srcu_struct(&kvm->srcu);
1258 kvm_arch_free_vm(kvm);
1259 preempt_notifier_dec();
1260 hardware_disable_all();
1262 module_put(kvm_chardev_ops.owner);
1265 void kvm_get_kvm(struct kvm *kvm)
1267 refcount_inc(&kvm->users_count);
1269 EXPORT_SYMBOL_GPL(kvm_get_kvm);
1272 * Make sure the vm is not during destruction, which is a safe version of
1273 * kvm_get_kvm(). Return true if kvm referenced successfully, false otherwise.
1275 bool kvm_get_kvm_safe(struct kvm *kvm)
1277 return refcount_inc_not_zero(&kvm->users_count);
1279 EXPORT_SYMBOL_GPL(kvm_get_kvm_safe);
1281 void kvm_put_kvm(struct kvm *kvm)
1283 if (refcount_dec_and_test(&kvm->users_count))
1284 kvm_destroy_vm(kvm);
1286 EXPORT_SYMBOL_GPL(kvm_put_kvm);
1289 * Used to put a reference that was taken on behalf of an object associated
1290 * with a user-visible file descriptor, e.g. a vcpu or device, if installation
1291 * of the new file descriptor fails and the reference cannot be transferred to
1292 * its final owner. In such cases, the caller is still actively using @kvm and
1293 * will fail miserably if the refcount unexpectedly hits zero.
1295 void kvm_put_kvm_no_destroy(struct kvm *kvm)
1297 WARN_ON(refcount_dec_and_test(&kvm->users_count));
1299 EXPORT_SYMBOL_GPL(kvm_put_kvm_no_destroy);
1301 static int kvm_vm_release(struct inode *inode, struct file *filp)
1303 struct kvm *kvm = filp->private_data;
1305 kvm_irqfd_release(kvm);
1312 * Allocation size is twice as large as the actual dirty bitmap size.
1313 * See kvm_vm_ioctl_get_dirty_log() why this is needed.
1315 static int kvm_alloc_dirty_bitmap(struct kvm_memory_slot *memslot)
1317 unsigned long dirty_bytes = kvm_dirty_bitmap_bytes(memslot);
1319 memslot->dirty_bitmap = __vcalloc(2, dirty_bytes, GFP_KERNEL_ACCOUNT);
1320 if (!memslot->dirty_bitmap)
1326 static struct kvm_memslots *kvm_get_inactive_memslots(struct kvm *kvm, int as_id)
1328 struct kvm_memslots *active = __kvm_memslots(kvm, as_id);
1329 int node_idx_inactive = active->node_idx ^ 1;
1331 return &kvm->__memslots[as_id][node_idx_inactive];
1335 * Helper to get the address space ID when one of memslot pointers may be NULL.
1336 * This also serves as a sanity that at least one of the pointers is non-NULL,
1337 * and that their address space IDs don't diverge.
1339 static int kvm_memslots_get_as_id(struct kvm_memory_slot *a,
1340 struct kvm_memory_slot *b)
1342 if (WARN_ON_ONCE(!a && !b))
1350 WARN_ON_ONCE(a->as_id != b->as_id);
1354 static void kvm_insert_gfn_node(struct kvm_memslots *slots,
1355 struct kvm_memory_slot *slot)
1357 struct rb_root *gfn_tree = &slots->gfn_tree;
1358 struct rb_node **node, *parent;
1359 int idx = slots->node_idx;
1362 for (node = &gfn_tree->rb_node; *node; ) {
1363 struct kvm_memory_slot *tmp;
1365 tmp = container_of(*node, struct kvm_memory_slot, gfn_node[idx]);
1367 if (slot->base_gfn < tmp->base_gfn)
1368 node = &(*node)->rb_left;
1369 else if (slot->base_gfn > tmp->base_gfn)
1370 node = &(*node)->rb_right;
1375 rb_link_node(&slot->gfn_node[idx], parent, node);
1376 rb_insert_color(&slot->gfn_node[idx], gfn_tree);
1379 static void kvm_erase_gfn_node(struct kvm_memslots *slots,
1380 struct kvm_memory_slot *slot)
1382 rb_erase(&slot->gfn_node[slots->node_idx], &slots->gfn_tree);
1385 static void kvm_replace_gfn_node(struct kvm_memslots *slots,
1386 struct kvm_memory_slot *old,
1387 struct kvm_memory_slot *new)
1389 int idx = slots->node_idx;
1391 WARN_ON_ONCE(old->base_gfn != new->base_gfn);
1393 rb_replace_node(&old->gfn_node[idx], &new->gfn_node[idx],
1398 * Replace @old with @new in the inactive memslots.
1400 * With NULL @old this simply adds @new.
1401 * With NULL @new this simply removes @old.
1403 * If @new is non-NULL its hva_node[slots_idx] range has to be set
1406 static void kvm_replace_memslot(struct kvm *kvm,
1407 struct kvm_memory_slot *old,
1408 struct kvm_memory_slot *new)
1410 int as_id = kvm_memslots_get_as_id(old, new);
1411 struct kvm_memslots *slots = kvm_get_inactive_memslots(kvm, as_id);
1412 int idx = slots->node_idx;
1415 hash_del(&old->id_node[idx]);
1416 interval_tree_remove(&old->hva_node[idx], &slots->hva_tree);
1418 if ((long)old == atomic_long_read(&slots->last_used_slot))
1419 atomic_long_set(&slots->last_used_slot, (long)new);
1422 kvm_erase_gfn_node(slots, old);
1428 * Initialize @new's hva range. Do this even when replacing an @old
1429 * slot, kvm_copy_memslot() deliberately does not touch node data.
1431 new->hva_node[idx].start = new->userspace_addr;
1432 new->hva_node[idx].last = new->userspace_addr +
1433 (new->npages << PAGE_SHIFT) - 1;
1436 * (Re)Add the new memslot. There is no O(1) interval_tree_replace(),
1437 * hva_node needs to be swapped with remove+insert even though hva can't
1438 * change when replacing an existing slot.
1440 hash_add(slots->id_hash, &new->id_node[idx], new->id);
1441 interval_tree_insert(&new->hva_node[idx], &slots->hva_tree);
1444 * If the memslot gfn is unchanged, rb_replace_node() can be used to
1445 * switch the node in the gfn tree instead of removing the old and
1446 * inserting the new as two separate operations. Replacement is a
1447 * single O(1) operation versus two O(log(n)) operations for
1450 if (old && old->base_gfn == new->base_gfn) {
1451 kvm_replace_gfn_node(slots, old, new);
1454 kvm_erase_gfn_node(slots, old);
1455 kvm_insert_gfn_node(slots, new);
1459 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
1461 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
1463 #ifdef __KVM_HAVE_READONLY_MEM
1464 valid_flags |= KVM_MEM_READONLY;
1467 if (mem->flags & ~valid_flags)
1473 static void kvm_swap_active_memslots(struct kvm *kvm, int as_id)
1475 struct kvm_memslots *slots = kvm_get_inactive_memslots(kvm, as_id);
1477 /* Grab the generation from the activate memslots. */
1478 u64 gen = __kvm_memslots(kvm, as_id)->generation;
1480 WARN_ON(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS);
1481 slots->generation = gen | KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
1484 * Do not store the new memslots while there are invalidations in
1485 * progress, otherwise the locking in invalidate_range_start and
1486 * invalidate_range_end will be unbalanced.
1488 spin_lock(&kvm->mn_invalidate_lock);
1489 prepare_to_rcuwait(&kvm->mn_memslots_update_rcuwait);
1490 while (kvm->mn_active_invalidate_count) {
1491 set_current_state(TASK_UNINTERRUPTIBLE);
1492 spin_unlock(&kvm->mn_invalidate_lock);
1494 spin_lock(&kvm->mn_invalidate_lock);
1496 finish_rcuwait(&kvm->mn_memslots_update_rcuwait);
1497 rcu_assign_pointer(kvm->memslots[as_id], slots);
1498 spin_unlock(&kvm->mn_invalidate_lock);
1501 * Acquired in kvm_set_memslot. Must be released before synchronize
1502 * SRCU below in order to avoid deadlock with another thread
1503 * acquiring the slots_arch_lock in an srcu critical section.
1505 mutex_unlock(&kvm->slots_arch_lock);
1507 synchronize_srcu_expedited(&kvm->srcu);
1510 * Increment the new memslot generation a second time, dropping the
1511 * update in-progress flag and incrementing the generation based on
1512 * the number of address spaces. This provides a unique and easily
1513 * identifiable generation number while the memslots are in flux.
1515 gen = slots->generation & ~KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
1518 * Generations must be unique even across address spaces. We do not need
1519 * a global counter for that, instead the generation space is evenly split
1520 * across address spaces. For example, with two address spaces, address
1521 * space 0 will use generations 0, 2, 4, ... while address space 1 will
1522 * use generations 1, 3, 5, ...
1524 gen += KVM_ADDRESS_SPACE_NUM;
1526 kvm_arch_memslots_updated(kvm, gen);
1528 slots->generation = gen;
1531 static int kvm_prepare_memory_region(struct kvm *kvm,
1532 const struct kvm_memory_slot *old,
1533 struct kvm_memory_slot *new,
1534 enum kvm_mr_change change)
1539 * If dirty logging is disabled, nullify the bitmap; the old bitmap
1540 * will be freed on "commit". If logging is enabled in both old and
1541 * new, reuse the existing bitmap. If logging is enabled only in the
1542 * new and KVM isn't using a ring buffer, allocate and initialize a
1545 if (change != KVM_MR_DELETE) {
1546 if (!(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
1547 new->dirty_bitmap = NULL;
1548 else if (old && old->dirty_bitmap)
1549 new->dirty_bitmap = old->dirty_bitmap;
1550 else if (!kvm->dirty_ring_size) {
1551 r = kvm_alloc_dirty_bitmap(new);
1555 if (kvm_dirty_log_manual_protect_and_init_set(kvm))
1556 bitmap_set(new->dirty_bitmap, 0, new->npages);
1560 r = kvm_arch_prepare_memory_region(kvm, old, new, change);
1562 /* Free the bitmap on failure if it was allocated above. */
1563 if (r && new && new->dirty_bitmap && old && !old->dirty_bitmap)
1564 kvm_destroy_dirty_bitmap(new);
1569 static void kvm_commit_memory_region(struct kvm *kvm,
1570 struct kvm_memory_slot *old,
1571 const struct kvm_memory_slot *new,
1572 enum kvm_mr_change change)
1575 * Update the total number of memslot pages before calling the arch
1576 * hook so that architectures can consume the result directly.
1578 if (change == KVM_MR_DELETE)
1579 kvm->nr_memslot_pages -= old->npages;
1580 else if (change == KVM_MR_CREATE)
1581 kvm->nr_memslot_pages += new->npages;
1583 kvm_arch_commit_memory_region(kvm, old, new, change);
1587 /* Nothing more to do. */
1590 /* Free the old memslot and all its metadata. */
1591 kvm_free_memslot(kvm, old);
1594 case KVM_MR_FLAGS_ONLY:
1596 * Free the dirty bitmap as needed; the below check encompasses
1597 * both the flags and whether a ring buffer is being used)
1599 if (old->dirty_bitmap && !new->dirty_bitmap)
1600 kvm_destroy_dirty_bitmap(old);
1603 * The final quirk. Free the detached, old slot, but only its
1604 * memory, not any metadata. Metadata, including arch specific
1605 * data, may be reused by @new.
1615 * Activate @new, which must be installed in the inactive slots by the caller,
1616 * by swapping the active slots and then propagating @new to @old once @old is
1617 * unreachable and can be safely modified.
1619 * With NULL @old this simply adds @new to @active (while swapping the sets).
1620 * With NULL @new this simply removes @old from @active and frees it
1621 * (while also swapping the sets).
1623 static void kvm_activate_memslot(struct kvm *kvm,
1624 struct kvm_memory_slot *old,
1625 struct kvm_memory_slot *new)
1627 int as_id = kvm_memslots_get_as_id(old, new);
1629 kvm_swap_active_memslots(kvm, as_id);
1631 /* Propagate the new memslot to the now inactive memslots. */
1632 kvm_replace_memslot(kvm, old, new);
1635 static void kvm_copy_memslot(struct kvm_memory_slot *dest,
1636 const struct kvm_memory_slot *src)
1638 dest->base_gfn = src->base_gfn;
1639 dest->npages = src->npages;
1640 dest->dirty_bitmap = src->dirty_bitmap;
1641 dest->arch = src->arch;
1642 dest->userspace_addr = src->userspace_addr;
1643 dest->flags = src->flags;
1645 dest->as_id = src->as_id;
1648 static void kvm_invalidate_memslot(struct kvm *kvm,
1649 struct kvm_memory_slot *old,
1650 struct kvm_memory_slot *invalid_slot)
1653 * Mark the current slot INVALID. As with all memslot modifications,
1654 * this must be done on an unreachable slot to avoid modifying the
1655 * current slot in the active tree.
1657 kvm_copy_memslot(invalid_slot, old);
1658 invalid_slot->flags |= KVM_MEMSLOT_INVALID;
1659 kvm_replace_memslot(kvm, old, invalid_slot);
1662 * Activate the slot that is now marked INVALID, but don't propagate
1663 * the slot to the now inactive slots. The slot is either going to be
1664 * deleted or recreated as a new slot.
1666 kvm_swap_active_memslots(kvm, old->as_id);
1669 * From this point no new shadow pages pointing to a deleted, or moved,
1670 * memslot will be created. Validation of sp->gfn happens in:
1671 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1672 * - kvm_is_visible_gfn (mmu_check_root)
1674 kvm_arch_flush_shadow_memslot(kvm, old);
1675 kvm_arch_guest_memory_reclaimed(kvm);
1677 /* Was released by kvm_swap_active_memslots, reacquire. */
1678 mutex_lock(&kvm->slots_arch_lock);
1681 * Copy the arch-specific field of the newly-installed slot back to the
1682 * old slot as the arch data could have changed between releasing
1683 * slots_arch_lock in install_new_memslots() and re-acquiring the lock
1684 * above. Writers are required to retrieve memslots *after* acquiring
1685 * slots_arch_lock, thus the active slot's data is guaranteed to be fresh.
1687 old->arch = invalid_slot->arch;
1690 static void kvm_create_memslot(struct kvm *kvm,
1691 struct kvm_memory_slot *new)
1693 /* Add the new memslot to the inactive set and activate. */
1694 kvm_replace_memslot(kvm, NULL, new);
1695 kvm_activate_memslot(kvm, NULL, new);
1698 static void kvm_delete_memslot(struct kvm *kvm,
1699 struct kvm_memory_slot *old,
1700 struct kvm_memory_slot *invalid_slot)
1703 * Remove the old memslot (in the inactive memslots) by passing NULL as
1704 * the "new" slot, and for the invalid version in the active slots.
1706 kvm_replace_memslot(kvm, old, NULL);
1707 kvm_activate_memslot(kvm, invalid_slot, NULL);
1710 static void kvm_move_memslot(struct kvm *kvm,
1711 struct kvm_memory_slot *old,
1712 struct kvm_memory_slot *new,
1713 struct kvm_memory_slot *invalid_slot)
1716 * Replace the old memslot in the inactive slots, and then swap slots
1717 * and replace the current INVALID with the new as well.
1719 kvm_replace_memslot(kvm, old, new);
1720 kvm_activate_memslot(kvm, invalid_slot, new);
1723 static void kvm_update_flags_memslot(struct kvm *kvm,
1724 struct kvm_memory_slot *old,
1725 struct kvm_memory_slot *new)
1728 * Similar to the MOVE case, but the slot doesn't need to be zapped as
1729 * an intermediate step. Instead, the old memslot is simply replaced
1730 * with a new, updated copy in both memslot sets.
1732 kvm_replace_memslot(kvm, old, new);
1733 kvm_activate_memslot(kvm, old, new);
1736 static int kvm_set_memslot(struct kvm *kvm,
1737 struct kvm_memory_slot *old,
1738 struct kvm_memory_slot *new,
1739 enum kvm_mr_change change)
1741 struct kvm_memory_slot *invalid_slot;
1745 * Released in kvm_swap_active_memslots.
1747 * Must be held from before the current memslots are copied until
1748 * after the new memslots are installed with rcu_assign_pointer,
1749 * then released before the synchronize srcu in kvm_swap_active_memslots.
1751 * When modifying memslots outside of the slots_lock, must be held
1752 * before reading the pointer to the current memslots until after all
1753 * changes to those memslots are complete.
1755 * These rules ensure that installing new memslots does not lose
1756 * changes made to the previous memslots.
1758 mutex_lock(&kvm->slots_arch_lock);
1761 * Invalidate the old slot if it's being deleted or moved. This is
1762 * done prior to actually deleting/moving the memslot to allow vCPUs to
1763 * continue running by ensuring there are no mappings or shadow pages
1764 * for the memslot when it is deleted/moved. Without pre-invalidation
1765 * (and without a lock), a window would exist between effecting the
1766 * delete/move and committing the changes in arch code where KVM or a
1767 * guest could access a non-existent memslot.
1769 * Modifications are done on a temporary, unreachable slot. The old
1770 * slot needs to be preserved in case a later step fails and the
1771 * invalidation needs to be reverted.
1773 if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) {
1774 invalid_slot = kzalloc(sizeof(*invalid_slot), GFP_KERNEL_ACCOUNT);
1775 if (!invalid_slot) {
1776 mutex_unlock(&kvm->slots_arch_lock);
1779 kvm_invalidate_memslot(kvm, old, invalid_slot);
1782 r = kvm_prepare_memory_region(kvm, old, new, change);
1785 * For DELETE/MOVE, revert the above INVALID change. No
1786 * modifications required since the original slot was preserved
1787 * in the inactive slots. Changing the active memslots also
1788 * release slots_arch_lock.
1790 if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) {
1791 kvm_activate_memslot(kvm, invalid_slot, old);
1792 kfree(invalid_slot);
1794 mutex_unlock(&kvm->slots_arch_lock);
1800 * For DELETE and MOVE, the working slot is now active as the INVALID
1801 * version of the old slot. MOVE is particularly special as it reuses
1802 * the old slot and returns a copy of the old slot (in working_slot).
1803 * For CREATE, there is no old slot. For DELETE and FLAGS_ONLY, the
1804 * old slot is detached but otherwise preserved.
1806 if (change == KVM_MR_CREATE)
1807 kvm_create_memslot(kvm, new);
1808 else if (change == KVM_MR_DELETE)
1809 kvm_delete_memslot(kvm, old, invalid_slot);
1810 else if (change == KVM_MR_MOVE)
1811 kvm_move_memslot(kvm, old, new, invalid_slot);
1812 else if (change == KVM_MR_FLAGS_ONLY)
1813 kvm_update_flags_memslot(kvm, old, new);
1817 /* Free the temporary INVALID slot used for DELETE and MOVE. */
1818 if (change == KVM_MR_DELETE || change == KVM_MR_MOVE)
1819 kfree(invalid_slot);
1822 * No need to refresh new->arch, changes after dropping slots_arch_lock
1823 * will directly hit the final, active memslot. Architectures are
1824 * responsible for knowing that new->arch may be stale.
1826 kvm_commit_memory_region(kvm, old, new, change);
1831 static bool kvm_check_memslot_overlap(struct kvm_memslots *slots, int id,
1832 gfn_t start, gfn_t end)
1834 struct kvm_memslot_iter iter;
1836 kvm_for_each_memslot_in_gfn_range(&iter, slots, start, end) {
1837 if (iter.slot->id != id)
1845 * Allocate some memory and give it an address in the guest physical address
1848 * Discontiguous memory is allowed, mostly for framebuffers.
1850 * Must be called holding kvm->slots_lock for write.
1852 int __kvm_set_memory_region(struct kvm *kvm,
1853 const struct kvm_userspace_memory_region *mem)
1855 struct kvm_memory_slot *old, *new;
1856 struct kvm_memslots *slots;
1857 enum kvm_mr_change change;
1858 unsigned long npages;
1863 r = check_memory_region_flags(mem);
1867 as_id = mem->slot >> 16;
1868 id = (u16)mem->slot;
1870 /* General sanity checks */
1871 if ((mem->memory_size & (PAGE_SIZE - 1)) ||
1872 (mem->memory_size != (unsigned long)mem->memory_size))
1874 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
1876 /* We can read the guest memory with __xxx_user() later on. */
1877 if ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
1878 (mem->userspace_addr != untagged_addr(mem->userspace_addr)) ||
1879 !access_ok((void __user *)(unsigned long)mem->userspace_addr,
1882 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
1884 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
1886 if ((mem->memory_size >> PAGE_SHIFT) > KVM_MEM_MAX_NR_PAGES)
1889 slots = __kvm_memslots(kvm, as_id);
1892 * Note, the old memslot (and the pointer itself!) may be invalidated
1893 * and/or destroyed by kvm_set_memslot().
1895 old = id_to_memslot(slots, id);
1897 if (!mem->memory_size) {
1898 if (!old || !old->npages)
1901 if (WARN_ON_ONCE(kvm->nr_memslot_pages < old->npages))
1904 return kvm_set_memslot(kvm, old, NULL, KVM_MR_DELETE);
1907 base_gfn = (mem->guest_phys_addr >> PAGE_SHIFT);
1908 npages = (mem->memory_size >> PAGE_SHIFT);
1910 if (!old || !old->npages) {
1911 change = KVM_MR_CREATE;
1914 * To simplify KVM internals, the total number of pages across
1915 * all memslots must fit in an unsigned long.
1917 if ((kvm->nr_memslot_pages + npages) < kvm->nr_memslot_pages)
1919 } else { /* Modify an existing slot. */
1920 if ((mem->userspace_addr != old->userspace_addr) ||
1921 (npages != old->npages) ||
1922 ((mem->flags ^ old->flags) & KVM_MEM_READONLY))
1925 if (base_gfn != old->base_gfn)
1926 change = KVM_MR_MOVE;
1927 else if (mem->flags != old->flags)
1928 change = KVM_MR_FLAGS_ONLY;
1929 else /* Nothing to change. */
1933 if ((change == KVM_MR_CREATE || change == KVM_MR_MOVE) &&
1934 kvm_check_memslot_overlap(slots, id, base_gfn, base_gfn + npages))
1937 /* Allocate a slot that will persist in the memslot. */
1938 new = kzalloc(sizeof(*new), GFP_KERNEL_ACCOUNT);
1944 new->base_gfn = base_gfn;
1945 new->npages = npages;
1946 new->flags = mem->flags;
1947 new->userspace_addr = mem->userspace_addr;
1949 r = kvm_set_memslot(kvm, old, new, change);
1954 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1956 int kvm_set_memory_region(struct kvm *kvm,
1957 const struct kvm_userspace_memory_region *mem)
1961 mutex_lock(&kvm->slots_lock);
1962 r = __kvm_set_memory_region(kvm, mem);
1963 mutex_unlock(&kvm->slots_lock);
1966 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1968 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1969 struct kvm_userspace_memory_region *mem)
1971 if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1974 return kvm_set_memory_region(kvm, mem);
1977 #ifndef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1979 * kvm_get_dirty_log - get a snapshot of dirty pages
1980 * @kvm: pointer to kvm instance
1981 * @log: slot id and address to which we copy the log
1982 * @is_dirty: set to '1' if any dirty pages were found
1983 * @memslot: set to the associated memslot, always valid on success
1985 int kvm_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log,
1986 int *is_dirty, struct kvm_memory_slot **memslot)
1988 struct kvm_memslots *slots;
1991 unsigned long any = 0;
1993 /* Dirty ring tracking is exclusive to dirty log tracking */
1994 if (kvm->dirty_ring_size)
2000 as_id = log->slot >> 16;
2001 id = (u16)log->slot;
2002 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
2005 slots = __kvm_memslots(kvm, as_id);
2006 *memslot = id_to_memslot(slots, id);
2007 if (!(*memslot) || !(*memslot)->dirty_bitmap)
2010 kvm_arch_sync_dirty_log(kvm, *memslot);
2012 n = kvm_dirty_bitmap_bytes(*memslot);
2014 for (i = 0; !any && i < n/sizeof(long); ++i)
2015 any = (*memslot)->dirty_bitmap[i];
2017 if (copy_to_user(log->dirty_bitmap, (*memslot)->dirty_bitmap, n))
2024 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
2026 #else /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
2028 * kvm_get_dirty_log_protect - get a snapshot of dirty pages
2029 * and reenable dirty page tracking for the corresponding pages.
2030 * @kvm: pointer to kvm instance
2031 * @log: slot id and address to which we copy the log
2033 * We need to keep it in mind that VCPU threads can write to the bitmap
2034 * concurrently. So, to avoid losing track of dirty pages we keep the
2037 * 1. Take a snapshot of the bit and clear it if needed.
2038 * 2. Write protect the corresponding page.
2039 * 3. Copy the snapshot to the userspace.
2040 * 4. Upon return caller flushes TLB's if needed.
2042 * Between 2 and 4, the guest may write to the page using the remaining TLB
2043 * entry. This is not a problem because the page is reported dirty using
2044 * the snapshot taken before and step 4 ensures that writes done after
2045 * exiting to userspace will be logged for the next call.
2048 static int kvm_get_dirty_log_protect(struct kvm *kvm, struct kvm_dirty_log *log)
2050 struct kvm_memslots *slots;
2051 struct kvm_memory_slot *memslot;
2054 unsigned long *dirty_bitmap;
2055 unsigned long *dirty_bitmap_buffer;
2058 /* Dirty ring tracking is exclusive to dirty log tracking */
2059 if (kvm->dirty_ring_size)
2062 as_id = log->slot >> 16;
2063 id = (u16)log->slot;
2064 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
2067 slots = __kvm_memslots(kvm, as_id);
2068 memslot = id_to_memslot(slots, id);
2069 if (!memslot || !memslot->dirty_bitmap)
2072 dirty_bitmap = memslot->dirty_bitmap;
2074 kvm_arch_sync_dirty_log(kvm, memslot);
2076 n = kvm_dirty_bitmap_bytes(memslot);
2078 if (kvm->manual_dirty_log_protect) {
2080 * Unlike kvm_get_dirty_log, we always return false in *flush,
2081 * because no flush is needed until KVM_CLEAR_DIRTY_LOG. There
2082 * is some code duplication between this function and
2083 * kvm_get_dirty_log, but hopefully all architecture
2084 * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log
2085 * can be eliminated.
2087 dirty_bitmap_buffer = dirty_bitmap;
2089 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
2090 memset(dirty_bitmap_buffer, 0, n);
2093 for (i = 0; i < n / sizeof(long); i++) {
2097 if (!dirty_bitmap[i])
2101 mask = xchg(&dirty_bitmap[i], 0);
2102 dirty_bitmap_buffer[i] = mask;
2104 offset = i * BITS_PER_LONG;
2105 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
2108 KVM_MMU_UNLOCK(kvm);
2112 kvm_arch_flush_remote_tlbs_memslot(kvm, memslot);
2114 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
2121 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
2122 * @kvm: kvm instance
2123 * @log: slot id and address to which we copy the log
2125 * Steps 1-4 below provide general overview of dirty page logging. See
2126 * kvm_get_dirty_log_protect() function description for additional details.
2128 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
2129 * always flush the TLB (step 4) even if previous step failed and the dirty
2130 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
2131 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
2132 * writes will be marked dirty for next log read.
2134 * 1. Take a snapshot of the bit and clear it if needed.
2135 * 2. Write protect the corresponding page.
2136 * 3. Copy the snapshot to the userspace.
2137 * 4. Flush TLB's if needed.
2139 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
2140 struct kvm_dirty_log *log)
2144 mutex_lock(&kvm->slots_lock);
2146 r = kvm_get_dirty_log_protect(kvm, log);
2148 mutex_unlock(&kvm->slots_lock);
2153 * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap
2154 * and reenable dirty page tracking for the corresponding pages.
2155 * @kvm: pointer to kvm instance
2156 * @log: slot id and address from which to fetch the bitmap of dirty pages
2158 static int kvm_clear_dirty_log_protect(struct kvm *kvm,
2159 struct kvm_clear_dirty_log *log)
2161 struct kvm_memslots *slots;
2162 struct kvm_memory_slot *memslot;
2166 unsigned long *dirty_bitmap;
2167 unsigned long *dirty_bitmap_buffer;
2170 /* Dirty ring tracking is exclusive to dirty log tracking */
2171 if (kvm->dirty_ring_size)
2174 as_id = log->slot >> 16;
2175 id = (u16)log->slot;
2176 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
2179 if (log->first_page & 63)
2182 slots = __kvm_memslots(kvm, as_id);
2183 memslot = id_to_memslot(slots, id);
2184 if (!memslot || !memslot->dirty_bitmap)
2187 dirty_bitmap = memslot->dirty_bitmap;
2189 n = ALIGN(log->num_pages, BITS_PER_LONG) / 8;
2191 if (log->first_page > memslot->npages ||
2192 log->num_pages > memslot->npages - log->first_page ||
2193 (log->num_pages < memslot->npages - log->first_page && (log->num_pages & 63)))
2196 kvm_arch_sync_dirty_log(kvm, memslot);
2199 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
2200 if (copy_from_user(dirty_bitmap_buffer, log->dirty_bitmap, n))
2204 for (offset = log->first_page, i = offset / BITS_PER_LONG,
2205 n = DIV_ROUND_UP(log->num_pages, BITS_PER_LONG); n--;
2206 i++, offset += BITS_PER_LONG) {
2207 unsigned long mask = *dirty_bitmap_buffer++;
2208 atomic_long_t *p = (atomic_long_t *) &dirty_bitmap[i];
2212 mask &= atomic_long_fetch_andnot(mask, p);
2215 * mask contains the bits that really have been cleared. This
2216 * never includes any bits beyond the length of the memslot (if
2217 * the length is not aligned to 64 pages), therefore it is not
2218 * a problem if userspace sets them in log->dirty_bitmap.
2222 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
2226 KVM_MMU_UNLOCK(kvm);
2229 kvm_arch_flush_remote_tlbs_memslot(kvm, memslot);
2234 static int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm,
2235 struct kvm_clear_dirty_log *log)
2239 mutex_lock(&kvm->slots_lock);
2241 r = kvm_clear_dirty_log_protect(kvm, log);
2243 mutex_unlock(&kvm->slots_lock);
2246 #endif /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
2248 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
2250 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
2252 EXPORT_SYMBOL_GPL(gfn_to_memslot);
2254 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
2256 struct kvm_memslots *slots = kvm_vcpu_memslots(vcpu);
2257 u64 gen = slots->generation;
2258 struct kvm_memory_slot *slot;
2261 * This also protects against using a memslot from a different address space,
2262 * since different address spaces have different generation numbers.
2264 if (unlikely(gen != vcpu->last_used_slot_gen)) {
2265 vcpu->last_used_slot = NULL;
2266 vcpu->last_used_slot_gen = gen;
2269 slot = try_get_memslot(vcpu->last_used_slot, gfn);
2274 * Fall back to searching all memslots. We purposely use
2275 * search_memslots() instead of __gfn_to_memslot() to avoid
2276 * thrashing the VM-wide last_used_slot in kvm_memslots.
2278 slot = search_memslots(slots, gfn, false);
2280 vcpu->last_used_slot = slot;
2287 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
2289 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
2291 return kvm_is_visible_memslot(memslot);
2293 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
2295 bool kvm_vcpu_is_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
2297 struct kvm_memory_slot *memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2299 return kvm_is_visible_memslot(memslot);
2301 EXPORT_SYMBOL_GPL(kvm_vcpu_is_visible_gfn);
2303 unsigned long kvm_host_page_size(struct kvm_vcpu *vcpu, gfn_t gfn)
2305 struct vm_area_struct *vma;
2306 unsigned long addr, size;
2310 addr = kvm_vcpu_gfn_to_hva_prot(vcpu, gfn, NULL);
2311 if (kvm_is_error_hva(addr))
2314 mmap_read_lock(current->mm);
2315 vma = find_vma(current->mm, addr);
2319 size = vma_kernel_pagesize(vma);
2322 mmap_read_unlock(current->mm);
2327 static bool memslot_is_readonly(const struct kvm_memory_slot *slot)
2329 return slot->flags & KVM_MEM_READONLY;
2332 static unsigned long __gfn_to_hva_many(const struct kvm_memory_slot *slot, gfn_t gfn,
2333 gfn_t *nr_pages, bool write)
2335 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
2336 return KVM_HVA_ERR_BAD;
2338 if (memslot_is_readonly(slot) && write)
2339 return KVM_HVA_ERR_RO_BAD;
2342 *nr_pages = slot->npages - (gfn - slot->base_gfn);
2344 return __gfn_to_hva_memslot(slot, gfn);
2347 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
2350 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
2353 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
2356 return gfn_to_hva_many(slot, gfn, NULL);
2358 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
2360 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
2362 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
2364 EXPORT_SYMBOL_GPL(gfn_to_hva);
2366 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
2368 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
2370 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
2373 * Return the hva of a @gfn and the R/W attribute if possible.
2375 * @slot: the kvm_memory_slot which contains @gfn
2376 * @gfn: the gfn to be translated
2377 * @writable: used to return the read/write attribute of the @slot if the hva
2378 * is valid and @writable is not NULL
2380 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
2381 gfn_t gfn, bool *writable)
2383 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
2385 if (!kvm_is_error_hva(hva) && writable)
2386 *writable = !memslot_is_readonly(slot);
2391 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
2393 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2395 return gfn_to_hva_memslot_prot(slot, gfn, writable);
2398 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
2400 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2402 return gfn_to_hva_memslot_prot(slot, gfn, writable);
2405 static inline int check_user_page_hwpoison(unsigned long addr)
2407 int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
2409 rc = get_user_pages(addr, 1, flags, NULL, NULL);
2410 return rc == -EHWPOISON;
2414 * The fast path to get the writable pfn which will be stored in @pfn,
2415 * true indicates success, otherwise false is returned. It's also the
2416 * only part that runs if we can in atomic context.
2418 static bool hva_to_pfn_fast(unsigned long addr, bool write_fault,
2419 bool *writable, kvm_pfn_t *pfn)
2421 struct page *page[1];
2424 * Fast pin a writable pfn only if it is a write fault request
2425 * or the caller allows to map a writable pfn for a read fault
2428 if (!(write_fault || writable))
2431 if (get_user_page_fast_only(addr, FOLL_WRITE, page)) {
2432 *pfn = page_to_pfn(page[0]);
2443 * The slow path to get the pfn of the specified host virtual address,
2444 * 1 indicates success, -errno is returned if error is detected.
2446 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
2447 bool *writable, kvm_pfn_t *pfn)
2449 unsigned int flags = FOLL_HWPOISON;
2456 *writable = write_fault;
2459 flags |= FOLL_WRITE;
2461 flags |= FOLL_NOWAIT;
2463 npages = get_user_pages_unlocked(addr, 1, &page, flags);
2467 /* map read fault as writable if possible */
2468 if (unlikely(!write_fault) && writable) {
2471 if (get_user_page_fast_only(addr, FOLL_WRITE, &wpage)) {
2477 *pfn = page_to_pfn(page);
2481 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
2483 if (unlikely(!(vma->vm_flags & VM_READ)))
2486 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
2492 static int kvm_try_get_pfn(kvm_pfn_t pfn)
2494 if (kvm_is_reserved_pfn(pfn))
2496 return get_page_unless_zero(pfn_to_page(pfn));
2499 static int hva_to_pfn_remapped(struct vm_area_struct *vma,
2500 unsigned long addr, bool write_fault,
2501 bool *writable, kvm_pfn_t *p_pfn)
2508 r = follow_pte(vma->vm_mm, addr, &ptep, &ptl);
2511 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
2512 * not call the fault handler, so do it here.
2514 bool unlocked = false;
2515 r = fixup_user_fault(current->mm, addr,
2516 (write_fault ? FAULT_FLAG_WRITE : 0),
2523 r = follow_pte(vma->vm_mm, addr, &ptep, &ptl);
2528 if (write_fault && !pte_write(*ptep)) {
2529 pfn = KVM_PFN_ERR_RO_FAULT;
2534 *writable = pte_write(*ptep);
2535 pfn = pte_pfn(*ptep);
2538 * Get a reference here because callers of *hva_to_pfn* and
2539 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
2540 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
2541 * set, but the kvm_try_get_pfn/kvm_release_pfn_clean pair will
2542 * simply do nothing for reserved pfns.
2544 * Whoever called remap_pfn_range is also going to call e.g.
2545 * unmap_mapping_range before the underlying pages are freed,
2546 * causing a call to our MMU notifier.
2548 * Certain IO or PFNMAP mappings can be backed with valid
2549 * struct pages, but be allocated without refcounting e.g.,
2550 * tail pages of non-compound higher order allocations, which
2551 * would then underflow the refcount when the caller does the
2552 * required put_page. Don't allow those pages here.
2554 if (!kvm_try_get_pfn(pfn))
2558 pte_unmap_unlock(ptep, ptl);
2565 * Pin guest page in memory and return its pfn.
2566 * @addr: host virtual address which maps memory to the guest
2567 * @atomic: whether this function can sleep
2568 * @async: whether this function need to wait IO complete if the
2569 * host page is not in the memory
2570 * @write_fault: whether we should get a writable host page
2571 * @writable: whether it allows to map a writable host page for !@write_fault
2573 * The function will map a writable host page for these two cases:
2574 * 1): @write_fault = true
2575 * 2): @write_fault = false && @writable, @writable will tell the caller
2576 * whether the mapping is writable.
2578 kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
2579 bool write_fault, bool *writable)
2581 struct vm_area_struct *vma;
2585 /* we can do it either atomically or asynchronously, not both */
2586 BUG_ON(atomic && async);
2588 if (hva_to_pfn_fast(addr, write_fault, writable, &pfn))
2592 return KVM_PFN_ERR_FAULT;
2594 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
2598 mmap_read_lock(current->mm);
2599 if (npages == -EHWPOISON ||
2600 (!async && check_user_page_hwpoison(addr))) {
2601 pfn = KVM_PFN_ERR_HWPOISON;
2606 vma = vma_lookup(current->mm, addr);
2609 pfn = KVM_PFN_ERR_FAULT;
2610 else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
2611 r = hva_to_pfn_remapped(vma, addr, write_fault, writable, &pfn);
2615 pfn = KVM_PFN_ERR_FAULT;
2617 if (async && vma_is_valid(vma, write_fault))
2619 pfn = KVM_PFN_ERR_FAULT;
2622 mmap_read_unlock(current->mm);
2626 kvm_pfn_t __gfn_to_pfn_memslot(const struct kvm_memory_slot *slot, gfn_t gfn,
2627 bool atomic, bool *async, bool write_fault,
2628 bool *writable, hva_t *hva)
2630 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
2635 if (addr == KVM_HVA_ERR_RO_BAD) {
2638 return KVM_PFN_ERR_RO_FAULT;
2641 if (kvm_is_error_hva(addr)) {
2644 return KVM_PFN_NOSLOT;
2647 /* Do not map writable pfn in the readonly memslot. */
2648 if (writable && memslot_is_readonly(slot)) {
2653 return hva_to_pfn(addr, atomic, async, write_fault,
2656 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
2658 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
2661 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
2662 write_fault, writable, NULL);
2664 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
2666 kvm_pfn_t gfn_to_pfn_memslot(const struct kvm_memory_slot *slot, gfn_t gfn)
2668 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL, NULL);
2670 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
2672 kvm_pfn_t gfn_to_pfn_memslot_atomic(const struct kvm_memory_slot *slot, gfn_t gfn)
2674 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL, NULL);
2676 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
2678 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
2680 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
2682 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
2684 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
2686 return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
2688 EXPORT_SYMBOL_GPL(gfn_to_pfn);
2690 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
2692 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
2694 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
2696 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
2697 struct page **pages, int nr_pages)
2702 addr = gfn_to_hva_many(slot, gfn, &entry);
2703 if (kvm_is_error_hva(addr))
2706 if (entry < nr_pages)
2709 return get_user_pages_fast_only(addr, nr_pages, FOLL_WRITE, pages);
2711 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
2713 static struct page *kvm_pfn_to_page(kvm_pfn_t pfn)
2715 if (is_error_noslot_pfn(pfn))
2716 return KVM_ERR_PTR_BAD_PAGE;
2718 if (kvm_is_reserved_pfn(pfn)) {
2720 return KVM_ERR_PTR_BAD_PAGE;
2723 return pfn_to_page(pfn);
2726 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
2730 pfn = gfn_to_pfn(kvm, gfn);
2732 return kvm_pfn_to_page(pfn);
2734 EXPORT_SYMBOL_GPL(gfn_to_page);
2736 void kvm_release_pfn(kvm_pfn_t pfn, bool dirty)
2742 kvm_release_pfn_dirty(pfn);
2744 kvm_release_pfn_clean(pfn);
2747 int kvm_vcpu_map(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map)
2751 struct page *page = KVM_UNMAPPED_PAGE;
2756 pfn = gfn_to_pfn(vcpu->kvm, gfn);
2757 if (is_error_noslot_pfn(pfn))
2760 if (pfn_valid(pfn)) {
2761 page = pfn_to_page(pfn);
2763 #ifdef CONFIG_HAS_IOMEM
2765 hva = memremap(pfn_to_hpa(pfn), PAGE_SIZE, MEMREMAP_WB);
2779 EXPORT_SYMBOL_GPL(kvm_vcpu_map);
2781 void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map, bool dirty)
2789 if (map->page != KVM_UNMAPPED_PAGE)
2791 #ifdef CONFIG_HAS_IOMEM
2797 kvm_vcpu_mark_page_dirty(vcpu, map->gfn);
2799 kvm_release_pfn(map->pfn, dirty);
2804 EXPORT_SYMBOL_GPL(kvm_vcpu_unmap);
2806 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
2810 pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
2812 return kvm_pfn_to_page(pfn);
2814 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
2816 void kvm_release_page_clean(struct page *page)
2818 WARN_ON(is_error_page(page));
2820 kvm_release_pfn_clean(page_to_pfn(page));
2822 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
2824 void kvm_release_pfn_clean(kvm_pfn_t pfn)
2826 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
2827 put_page(pfn_to_page(pfn));
2829 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
2831 void kvm_release_page_dirty(struct page *page)
2833 WARN_ON(is_error_page(page));
2835 kvm_release_pfn_dirty(page_to_pfn(page));
2837 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
2839 void kvm_release_pfn_dirty(kvm_pfn_t pfn)
2841 kvm_set_pfn_dirty(pfn);
2842 kvm_release_pfn_clean(pfn);
2844 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
2846 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
2848 if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn))
2849 SetPageDirty(pfn_to_page(pfn));
2851 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
2853 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
2855 if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn))
2856 mark_page_accessed(pfn_to_page(pfn));
2858 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
2860 static int next_segment(unsigned long len, int offset)
2862 if (len > PAGE_SIZE - offset)
2863 return PAGE_SIZE - offset;
2868 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
2869 void *data, int offset, int len)
2874 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
2875 if (kvm_is_error_hva(addr))
2877 r = __copy_from_user(data, (void __user *)addr + offset, len);
2883 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
2886 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2888 return __kvm_read_guest_page(slot, gfn, data, offset, len);
2890 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
2892 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
2893 int offset, int len)
2895 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2897 return __kvm_read_guest_page(slot, gfn, data, offset, len);
2899 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
2901 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
2903 gfn_t gfn = gpa >> PAGE_SHIFT;
2905 int offset = offset_in_page(gpa);
2908 while ((seg = next_segment(len, offset)) != 0) {
2909 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
2919 EXPORT_SYMBOL_GPL(kvm_read_guest);
2921 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
2923 gfn_t gfn = gpa >> PAGE_SHIFT;
2925 int offset = offset_in_page(gpa);
2928 while ((seg = next_segment(len, offset)) != 0) {
2929 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
2939 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
2941 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
2942 void *data, int offset, unsigned long len)
2947 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
2948 if (kvm_is_error_hva(addr))
2950 pagefault_disable();
2951 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
2958 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
2959 void *data, unsigned long len)
2961 gfn_t gfn = gpa >> PAGE_SHIFT;
2962 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2963 int offset = offset_in_page(gpa);
2965 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
2967 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
2969 static int __kvm_write_guest_page(struct kvm *kvm,
2970 struct kvm_memory_slot *memslot, gfn_t gfn,
2971 const void *data, int offset, int len)
2976 addr = gfn_to_hva_memslot(memslot, gfn);
2977 if (kvm_is_error_hva(addr))
2979 r = __copy_to_user((void __user *)addr + offset, data, len);
2982 mark_page_dirty_in_slot(kvm, memslot, gfn);
2986 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
2987 const void *data, int offset, int len)
2989 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2991 return __kvm_write_guest_page(kvm, slot, gfn, data, offset, len);
2993 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
2995 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
2996 const void *data, int offset, int len)
2998 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
3000 return __kvm_write_guest_page(vcpu->kvm, slot, gfn, data, offset, len);
3002 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
3004 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
3007 gfn_t gfn = gpa >> PAGE_SHIFT;
3009 int offset = offset_in_page(gpa);
3012 while ((seg = next_segment(len, offset)) != 0) {
3013 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
3023 EXPORT_SYMBOL_GPL(kvm_write_guest);
3025 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
3028 gfn_t gfn = gpa >> PAGE_SHIFT;
3030 int offset = offset_in_page(gpa);
3033 while ((seg = next_segment(len, offset)) != 0) {
3034 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
3044 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
3046 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
3047 struct gfn_to_hva_cache *ghc,
3048 gpa_t gpa, unsigned long len)
3050 int offset = offset_in_page(gpa);
3051 gfn_t start_gfn = gpa >> PAGE_SHIFT;
3052 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
3053 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
3054 gfn_t nr_pages_avail;
3056 /* Update ghc->generation before performing any error checks. */
3057 ghc->generation = slots->generation;
3059 if (start_gfn > end_gfn) {
3060 ghc->hva = KVM_HVA_ERR_BAD;
3065 * If the requested region crosses two memslots, we still
3066 * verify that the entire region is valid here.
3068 for ( ; start_gfn <= end_gfn; start_gfn += nr_pages_avail) {
3069 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
3070 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
3072 if (kvm_is_error_hva(ghc->hva))
3076 /* Use the slow path for cross page reads and writes. */
3077 if (nr_pages_needed == 1)
3080 ghc->memslot = NULL;
3087 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3088 gpa_t gpa, unsigned long len)
3090 struct kvm_memslots *slots = kvm_memslots(kvm);
3091 return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len);
3093 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
3095 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3096 void *data, unsigned int offset,
3099 struct kvm_memslots *slots = kvm_memslots(kvm);
3101 gpa_t gpa = ghc->gpa + offset;
3103 if (WARN_ON_ONCE(len + offset > ghc->len))
3106 if (slots->generation != ghc->generation) {
3107 if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len))
3111 if (kvm_is_error_hva(ghc->hva))
3114 if (unlikely(!ghc->memslot))
3115 return kvm_write_guest(kvm, gpa, data, len);
3117 r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
3120 mark_page_dirty_in_slot(kvm, ghc->memslot, gpa >> PAGE_SHIFT);
3124 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached);
3126 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3127 void *data, unsigned long len)
3129 return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len);
3131 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
3133 int kvm_read_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3134 void *data, unsigned int offset,
3137 struct kvm_memslots *slots = kvm_memslots(kvm);
3139 gpa_t gpa = ghc->gpa + offset;
3141 if (WARN_ON_ONCE(len + offset > ghc->len))
3144 if (slots->generation != ghc->generation) {
3145 if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len))
3149 if (kvm_is_error_hva(ghc->hva))
3152 if (unlikely(!ghc->memslot))
3153 return kvm_read_guest(kvm, gpa, data, len);
3155 r = __copy_from_user(data, (void __user *)ghc->hva + offset, len);
3161 EXPORT_SYMBOL_GPL(kvm_read_guest_offset_cached);
3163 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3164 void *data, unsigned long len)
3166 return kvm_read_guest_offset_cached(kvm, ghc, data, 0, len);
3168 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
3170 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
3172 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
3173 gfn_t gfn = gpa >> PAGE_SHIFT;
3175 int offset = offset_in_page(gpa);
3178 while ((seg = next_segment(len, offset)) != 0) {
3179 ret = kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
3188 EXPORT_SYMBOL_GPL(kvm_clear_guest);
3190 void mark_page_dirty_in_slot(struct kvm *kvm,
3191 const struct kvm_memory_slot *memslot,
3194 struct kvm_vcpu *vcpu = kvm_get_running_vcpu();
3196 #ifdef CONFIG_HAVE_KVM_DIRTY_RING
3197 if (WARN_ON_ONCE(!vcpu) || WARN_ON_ONCE(vcpu->kvm != kvm))
3201 if (memslot && kvm_slot_dirty_track_enabled(memslot)) {
3202 unsigned long rel_gfn = gfn - memslot->base_gfn;
3203 u32 slot = (memslot->as_id << 16) | memslot->id;
3205 if (kvm->dirty_ring_size)
3206 kvm_dirty_ring_push(&vcpu->dirty_ring,
3209 set_bit_le(rel_gfn, memslot->dirty_bitmap);
3212 EXPORT_SYMBOL_GPL(mark_page_dirty_in_slot);
3214 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
3216 struct kvm_memory_slot *memslot;
3218 memslot = gfn_to_memslot(kvm, gfn);
3219 mark_page_dirty_in_slot(kvm, memslot, gfn);
3221 EXPORT_SYMBOL_GPL(mark_page_dirty);
3223 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
3225 struct kvm_memory_slot *memslot;
3227 memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
3228 mark_page_dirty_in_slot(vcpu->kvm, memslot, gfn);
3230 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
3232 void kvm_sigset_activate(struct kvm_vcpu *vcpu)
3234 if (!vcpu->sigset_active)
3238 * This does a lockless modification of ->real_blocked, which is fine
3239 * because, only current can change ->real_blocked and all readers of
3240 * ->real_blocked don't care as long ->real_blocked is always a subset
3243 sigprocmask(SIG_SETMASK, &vcpu->sigset, ¤t->real_blocked);
3246 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu)
3248 if (!vcpu->sigset_active)
3251 sigprocmask(SIG_SETMASK, ¤t->real_blocked, NULL);
3252 sigemptyset(¤t->real_blocked);
3255 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
3257 unsigned int old, val, grow, grow_start;
3259 old = val = vcpu->halt_poll_ns;
3260 grow_start = READ_ONCE(halt_poll_ns_grow_start);
3261 grow = READ_ONCE(halt_poll_ns_grow);
3266 if (val < grow_start)
3269 if (val > vcpu->kvm->max_halt_poll_ns)
3270 val = vcpu->kvm->max_halt_poll_ns;
3272 vcpu->halt_poll_ns = val;
3274 trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
3277 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
3279 unsigned int old, val, shrink, grow_start;
3281 old = val = vcpu->halt_poll_ns;
3282 shrink = READ_ONCE(halt_poll_ns_shrink);
3283 grow_start = READ_ONCE(halt_poll_ns_grow_start);
3289 if (val < grow_start)
3292 vcpu->halt_poll_ns = val;
3293 trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
3296 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
3299 int idx = srcu_read_lock(&vcpu->kvm->srcu);
3301 if (kvm_arch_vcpu_runnable(vcpu)) {
3302 kvm_make_request(KVM_REQ_UNHALT, vcpu);
3305 if (kvm_cpu_has_pending_timer(vcpu))
3307 if (signal_pending(current))
3309 if (kvm_check_request(KVM_REQ_UNBLOCK, vcpu))
3314 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3319 * Block the vCPU until the vCPU is runnable, an event arrives, or a signal is
3320 * pending. This is mostly used when halting a vCPU, but may also be used
3321 * directly for other vCPU non-runnable states, e.g. x86's Wait-For-SIPI.
3323 bool kvm_vcpu_block(struct kvm_vcpu *vcpu)
3325 struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu);
3326 bool waited = false;
3328 vcpu->stat.generic.blocking = 1;
3330 kvm_arch_vcpu_blocking(vcpu);
3332 prepare_to_rcuwait(wait);
3334 set_current_state(TASK_INTERRUPTIBLE);
3336 if (kvm_vcpu_check_block(vcpu) < 0)
3342 finish_rcuwait(wait);
3344 kvm_arch_vcpu_unblocking(vcpu);
3346 vcpu->stat.generic.blocking = 0;
3351 static inline void update_halt_poll_stats(struct kvm_vcpu *vcpu, ktime_t start,
3352 ktime_t end, bool success)
3354 struct kvm_vcpu_stat_generic *stats = &vcpu->stat.generic;
3355 u64 poll_ns = ktime_to_ns(ktime_sub(end, start));
3357 ++vcpu->stat.generic.halt_attempted_poll;
3360 ++vcpu->stat.generic.halt_successful_poll;
3362 if (!vcpu_valid_wakeup(vcpu))
3363 ++vcpu->stat.generic.halt_poll_invalid;
3365 stats->halt_poll_success_ns += poll_ns;
3366 KVM_STATS_LOG_HIST_UPDATE(stats->halt_poll_success_hist, poll_ns);
3368 stats->halt_poll_fail_ns += poll_ns;
3369 KVM_STATS_LOG_HIST_UPDATE(stats->halt_poll_fail_hist, poll_ns);
3374 * Emulate a vCPU halt condition, e.g. HLT on x86, WFI on arm, etc... If halt
3375 * polling is enabled, busy wait for a short time before blocking to avoid the
3376 * expensive block+unblock sequence if a wake event arrives soon after the vCPU
3379 void kvm_vcpu_halt(struct kvm_vcpu *vcpu)
3381 bool halt_poll_allowed = !kvm_arch_no_poll(vcpu);
3382 bool do_halt_poll = halt_poll_allowed && vcpu->halt_poll_ns;
3383 ktime_t start, cur, poll_end;
3384 bool waited = false;
3387 start = cur = poll_end = ktime_get();
3389 ktime_t stop = ktime_add_ns(start, vcpu->halt_poll_ns);
3393 * This sets KVM_REQ_UNHALT if an interrupt
3396 if (kvm_vcpu_check_block(vcpu) < 0)
3399 poll_end = cur = ktime_get();
3400 } while (kvm_vcpu_can_poll(cur, stop));
3403 waited = kvm_vcpu_block(vcpu);
3407 vcpu->stat.generic.halt_wait_ns +=
3408 ktime_to_ns(cur) - ktime_to_ns(poll_end);
3409 KVM_STATS_LOG_HIST_UPDATE(vcpu->stat.generic.halt_wait_hist,
3410 ktime_to_ns(cur) - ktime_to_ns(poll_end));
3413 /* The total time the vCPU was "halted", including polling time. */
3414 halt_ns = ktime_to_ns(cur) - ktime_to_ns(start);
3417 * Note, halt-polling is considered successful so long as the vCPU was
3418 * never actually scheduled out, i.e. even if the wake event arrived
3419 * after of the halt-polling loop itself, but before the full wait.
3422 update_halt_poll_stats(vcpu, start, poll_end, !waited);
3424 if (halt_poll_allowed) {
3425 if (!vcpu_valid_wakeup(vcpu)) {
3426 shrink_halt_poll_ns(vcpu);
3427 } else if (vcpu->kvm->max_halt_poll_ns) {
3428 if (halt_ns <= vcpu->halt_poll_ns)
3430 /* we had a long block, shrink polling */
3431 else if (vcpu->halt_poll_ns &&
3432 halt_ns > vcpu->kvm->max_halt_poll_ns)
3433 shrink_halt_poll_ns(vcpu);
3434 /* we had a short halt and our poll time is too small */
3435 else if (vcpu->halt_poll_ns < vcpu->kvm->max_halt_poll_ns &&
3436 halt_ns < vcpu->kvm->max_halt_poll_ns)
3437 grow_halt_poll_ns(vcpu);
3439 vcpu->halt_poll_ns = 0;
3443 trace_kvm_vcpu_wakeup(halt_ns, waited, vcpu_valid_wakeup(vcpu));
3445 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
3447 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
3449 if (__kvm_vcpu_wake_up(vcpu)) {
3450 WRITE_ONCE(vcpu->ready, true);
3451 ++vcpu->stat.generic.halt_wakeup;
3457 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
3461 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
3463 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
3467 if (kvm_vcpu_wake_up(vcpu))
3472 * The only state change done outside the vcpu mutex is IN_GUEST_MODE
3473 * to EXITING_GUEST_MODE. Therefore the moderately expensive "should
3474 * kick" check does not need atomic operations if kvm_vcpu_kick is used
3475 * within the vCPU thread itself.
3477 if (vcpu == __this_cpu_read(kvm_running_vcpu)) {
3478 if (vcpu->mode == IN_GUEST_MODE)
3479 WRITE_ONCE(vcpu->mode, EXITING_GUEST_MODE);
3484 * Note, the vCPU could get migrated to a different pCPU at any point
3485 * after kvm_arch_vcpu_should_kick(), which could result in sending an
3486 * IPI to the previous pCPU. But, that's ok because the purpose of the
3487 * IPI is to force the vCPU to leave IN_GUEST_MODE, and migrating the
3488 * vCPU also requires it to leave IN_GUEST_MODE.
3490 if (kvm_arch_vcpu_should_kick(vcpu)) {
3491 cpu = READ_ONCE(vcpu->cpu);
3492 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
3493 smp_send_reschedule(cpu);
3498 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
3499 #endif /* !CONFIG_S390 */
3501 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
3504 struct task_struct *task = NULL;
3508 pid = rcu_dereference(target->pid);
3510 task = get_pid_task(pid, PIDTYPE_PID);
3514 ret = yield_to(task, 1);
3515 put_task_struct(task);
3519 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
3522 * Helper that checks whether a VCPU is eligible for directed yield.
3523 * Most eligible candidate to yield is decided by following heuristics:
3525 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
3526 * (preempted lock holder), indicated by @in_spin_loop.
3527 * Set at the beginning and cleared at the end of interception/PLE handler.
3529 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
3530 * chance last time (mostly it has become eligible now since we have probably
3531 * yielded to lockholder in last iteration. This is done by toggling
3532 * @dy_eligible each time a VCPU checked for eligibility.)
3534 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
3535 * to preempted lock-holder could result in wrong VCPU selection and CPU
3536 * burning. Giving priority for a potential lock-holder increases lock
3539 * Since algorithm is based on heuristics, accessing another VCPU data without
3540 * locking does not harm. It may result in trying to yield to same VCPU, fail
3541 * and continue with next VCPU and so on.
3543 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
3545 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
3548 eligible = !vcpu->spin_loop.in_spin_loop ||
3549 vcpu->spin_loop.dy_eligible;
3551 if (vcpu->spin_loop.in_spin_loop)
3552 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
3561 * Unlike kvm_arch_vcpu_runnable, this function is called outside
3562 * a vcpu_load/vcpu_put pair. However, for most architectures
3563 * kvm_arch_vcpu_runnable does not require vcpu_load.
3565 bool __weak kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
3567 return kvm_arch_vcpu_runnable(vcpu);
3570 static bool vcpu_dy_runnable(struct kvm_vcpu *vcpu)
3572 if (kvm_arch_dy_runnable(vcpu))
3575 #ifdef CONFIG_KVM_ASYNC_PF
3576 if (!list_empty_careful(&vcpu->async_pf.done))
3583 bool __weak kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu)
3588 void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode)
3590 struct kvm *kvm = me->kvm;
3591 struct kvm_vcpu *vcpu;
3592 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
3598 kvm_vcpu_set_in_spin_loop(me, true);
3600 * We boost the priority of a VCPU that is runnable but not
3601 * currently running, because it got preempted by something
3602 * else and called schedule in __vcpu_run. Hopefully that
3603 * VCPU is holding the lock that we need and will release it.
3604 * We approximate round-robin by starting at the last boosted VCPU.
3606 for (pass = 0; pass < 2 && !yielded && try; pass++) {
3607 kvm_for_each_vcpu(i, vcpu, kvm) {
3608 if (!pass && i <= last_boosted_vcpu) {
3609 i = last_boosted_vcpu;
3611 } else if (pass && i > last_boosted_vcpu)
3613 if (!READ_ONCE(vcpu->ready))
3617 if (kvm_vcpu_is_blocking(vcpu) && !vcpu_dy_runnable(vcpu))
3619 if (READ_ONCE(vcpu->preempted) && yield_to_kernel_mode &&
3620 !kvm_arch_dy_has_pending_interrupt(vcpu) &&
3621 !kvm_arch_vcpu_in_kernel(vcpu))
3623 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
3626 yielded = kvm_vcpu_yield_to(vcpu);
3628 kvm->last_boosted_vcpu = i;
3630 } else if (yielded < 0) {
3637 kvm_vcpu_set_in_spin_loop(me, false);
3639 /* Ensure vcpu is not eligible during next spinloop */
3640 kvm_vcpu_set_dy_eligible(me, false);
3642 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
3644 static bool kvm_page_in_dirty_ring(struct kvm *kvm, unsigned long pgoff)
3646 #ifdef CONFIG_HAVE_KVM_DIRTY_RING
3647 return (pgoff >= KVM_DIRTY_LOG_PAGE_OFFSET) &&
3648 (pgoff < KVM_DIRTY_LOG_PAGE_OFFSET +
3649 kvm->dirty_ring_size / PAGE_SIZE);
3655 static vm_fault_t kvm_vcpu_fault(struct vm_fault *vmf)
3657 struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data;
3660 if (vmf->pgoff == 0)
3661 page = virt_to_page(vcpu->run);
3663 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
3664 page = virt_to_page(vcpu->arch.pio_data);
3666 #ifdef CONFIG_KVM_MMIO
3667 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
3668 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
3670 else if (kvm_page_in_dirty_ring(vcpu->kvm, vmf->pgoff))
3671 page = kvm_dirty_ring_get_page(
3673 vmf->pgoff - KVM_DIRTY_LOG_PAGE_OFFSET);
3675 return kvm_arch_vcpu_fault(vcpu, vmf);
3681 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
3682 .fault = kvm_vcpu_fault,
3685 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
3687 struct kvm_vcpu *vcpu = file->private_data;
3688 unsigned long pages = vma_pages(vma);
3690 if ((kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff) ||
3691 kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff + pages - 1)) &&
3692 ((vma->vm_flags & VM_EXEC) || !(vma->vm_flags & VM_SHARED)))
3695 vma->vm_ops = &kvm_vcpu_vm_ops;
3699 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
3701 struct kvm_vcpu *vcpu = filp->private_data;
3703 kvm_put_kvm(vcpu->kvm);
3707 static const struct file_operations kvm_vcpu_fops = {
3708 .release = kvm_vcpu_release,
3709 .unlocked_ioctl = kvm_vcpu_ioctl,
3710 .mmap = kvm_vcpu_mmap,
3711 .llseek = noop_llseek,
3712 KVM_COMPAT(kvm_vcpu_compat_ioctl),
3716 * Allocates an inode for the vcpu.
3718 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
3720 char name[8 + 1 + ITOA_MAX_LEN + 1];
3722 snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id);
3723 return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
3726 static void kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
3728 #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS
3729 struct dentry *debugfs_dentry;
3730 char dir_name[ITOA_MAX_LEN * 2];
3732 if (!debugfs_initialized())
3735 snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
3736 debugfs_dentry = debugfs_create_dir(dir_name,
3737 vcpu->kvm->debugfs_dentry);
3739 kvm_arch_create_vcpu_debugfs(vcpu, debugfs_dentry);
3744 * Creates some virtual cpus. Good luck creating more than one.
3746 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
3749 struct kvm_vcpu *vcpu;
3752 if (id >= KVM_MAX_VCPU_IDS)
3755 mutex_lock(&kvm->lock);
3756 if (kvm->created_vcpus == KVM_MAX_VCPUS) {
3757 mutex_unlock(&kvm->lock);
3761 kvm->created_vcpus++;
3762 mutex_unlock(&kvm->lock);
3764 r = kvm_arch_vcpu_precreate(kvm, id);
3766 goto vcpu_decrement;
3768 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL_ACCOUNT);
3771 goto vcpu_decrement;
3774 BUILD_BUG_ON(sizeof(struct kvm_run) > PAGE_SIZE);
3775 page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
3780 vcpu->run = page_address(page);
3782 kvm_vcpu_init(vcpu, kvm, id);
3784 r = kvm_arch_vcpu_create(vcpu);
3786 goto vcpu_free_run_page;
3788 if (kvm->dirty_ring_size) {
3789 r = kvm_dirty_ring_alloc(&vcpu->dirty_ring,
3790 id, kvm->dirty_ring_size);
3792 goto arch_vcpu_destroy;
3795 mutex_lock(&kvm->lock);
3796 if (kvm_get_vcpu_by_id(kvm, id)) {
3798 goto unlock_vcpu_destroy;
3801 vcpu->vcpu_idx = atomic_read(&kvm->online_vcpus);
3802 r = xa_insert(&kvm->vcpu_array, vcpu->vcpu_idx, vcpu, GFP_KERNEL_ACCOUNT);
3803 BUG_ON(r == -EBUSY);
3805 goto unlock_vcpu_destroy;
3807 /* Fill the stats id string for the vcpu */
3808 snprintf(vcpu->stats_id, sizeof(vcpu->stats_id), "kvm-%d/vcpu-%d",
3809 task_pid_nr(current), id);
3811 /* Now it's all set up, let userspace reach it */
3813 r = create_vcpu_fd(vcpu);
3815 xa_erase(&kvm->vcpu_array, vcpu->vcpu_idx);
3816 kvm_put_kvm_no_destroy(kvm);
3817 goto unlock_vcpu_destroy;
3821 * Pairs with smp_rmb() in kvm_get_vcpu. Store the vcpu
3822 * pointer before kvm->online_vcpu's incremented value.
3825 atomic_inc(&kvm->online_vcpus);
3827 mutex_unlock(&kvm->lock);
3828 kvm_arch_vcpu_postcreate(vcpu);
3829 kvm_create_vcpu_debugfs(vcpu);
3832 unlock_vcpu_destroy:
3833 mutex_unlock(&kvm->lock);
3834 kvm_dirty_ring_free(&vcpu->dirty_ring);
3836 kvm_arch_vcpu_destroy(vcpu);
3838 free_page((unsigned long)vcpu->run);
3840 kmem_cache_free(kvm_vcpu_cache, vcpu);
3842 mutex_lock(&kvm->lock);
3843 kvm->created_vcpus--;
3844 mutex_unlock(&kvm->lock);
3848 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
3851 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
3852 vcpu->sigset_active = 1;
3853 vcpu->sigset = *sigset;
3855 vcpu->sigset_active = 0;
3859 static ssize_t kvm_vcpu_stats_read(struct file *file, char __user *user_buffer,
3860 size_t size, loff_t *offset)
3862 struct kvm_vcpu *vcpu = file->private_data;
3864 return kvm_stats_read(vcpu->stats_id, &kvm_vcpu_stats_header,
3865 &kvm_vcpu_stats_desc[0], &vcpu->stat,
3866 sizeof(vcpu->stat), user_buffer, size, offset);
3869 static const struct file_operations kvm_vcpu_stats_fops = {
3870 .read = kvm_vcpu_stats_read,
3871 .llseek = noop_llseek,
3874 static int kvm_vcpu_ioctl_get_stats_fd(struct kvm_vcpu *vcpu)
3878 char name[15 + ITOA_MAX_LEN + 1];
3880 snprintf(name, sizeof(name), "kvm-vcpu-stats:%d", vcpu->vcpu_id);
3882 fd = get_unused_fd_flags(O_CLOEXEC);
3886 file = anon_inode_getfile(name, &kvm_vcpu_stats_fops, vcpu, O_RDONLY);
3889 return PTR_ERR(file);
3891 file->f_mode |= FMODE_PREAD;
3892 fd_install(fd, file);
3897 static long kvm_vcpu_ioctl(struct file *filp,
3898 unsigned int ioctl, unsigned long arg)
3900 struct kvm_vcpu *vcpu = filp->private_data;
3901 void __user *argp = (void __user *)arg;
3903 struct kvm_fpu *fpu = NULL;
3904 struct kvm_sregs *kvm_sregs = NULL;
3906 if (vcpu->kvm->mm != current->mm || vcpu->kvm->vm_dead)
3909 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
3913 * Some architectures have vcpu ioctls that are asynchronous to vcpu
3914 * execution; mutex_lock() would break them.
3916 r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg);
3917 if (r != -ENOIOCTLCMD)
3920 if (mutex_lock_killable(&vcpu->mutex))
3928 oldpid = rcu_access_pointer(vcpu->pid);
3929 if (unlikely(oldpid != task_pid(current))) {
3930 /* The thread running this VCPU changed. */
3933 r = kvm_arch_vcpu_run_pid_change(vcpu);
3937 newpid = get_task_pid(current, PIDTYPE_PID);
3938 rcu_assign_pointer(vcpu->pid, newpid);
3943 r = kvm_arch_vcpu_ioctl_run(vcpu);
3944 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
3947 case KVM_GET_REGS: {
3948 struct kvm_regs *kvm_regs;
3951 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL_ACCOUNT);
3954 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
3958 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
3965 case KVM_SET_REGS: {
3966 struct kvm_regs *kvm_regs;
3968 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
3969 if (IS_ERR(kvm_regs)) {
3970 r = PTR_ERR(kvm_regs);
3973 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
3977 case KVM_GET_SREGS: {
3978 kvm_sregs = kzalloc(sizeof(struct kvm_sregs),
3979 GFP_KERNEL_ACCOUNT);
3983 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
3987 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
3992 case KVM_SET_SREGS: {
3993 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
3994 if (IS_ERR(kvm_sregs)) {
3995 r = PTR_ERR(kvm_sregs);
3999 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
4002 case KVM_GET_MP_STATE: {
4003 struct kvm_mp_state mp_state;
4005 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
4009 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
4014 case KVM_SET_MP_STATE: {
4015 struct kvm_mp_state mp_state;
4018 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
4020 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
4023 case KVM_TRANSLATE: {
4024 struct kvm_translation tr;
4027 if (copy_from_user(&tr, argp, sizeof(tr)))
4029 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
4033 if (copy_to_user(argp, &tr, sizeof(tr)))
4038 case KVM_SET_GUEST_DEBUG: {
4039 struct kvm_guest_debug dbg;
4042 if (copy_from_user(&dbg, argp, sizeof(dbg)))
4044 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
4047 case KVM_SET_SIGNAL_MASK: {
4048 struct kvm_signal_mask __user *sigmask_arg = argp;
4049 struct kvm_signal_mask kvm_sigmask;
4050 sigset_t sigset, *p;
4055 if (copy_from_user(&kvm_sigmask, argp,
4056 sizeof(kvm_sigmask)))
4059 if (kvm_sigmask.len != sizeof(sigset))
4062 if (copy_from_user(&sigset, sigmask_arg->sigset,
4067 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
4071 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL_ACCOUNT);
4075 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
4079 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
4085 fpu = memdup_user(argp, sizeof(*fpu));
4091 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
4094 case KVM_GET_STATS_FD: {
4095 r = kvm_vcpu_ioctl_get_stats_fd(vcpu);
4099 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
4102 mutex_unlock(&vcpu->mutex);
4108 #ifdef CONFIG_KVM_COMPAT
4109 static long kvm_vcpu_compat_ioctl(struct file *filp,
4110 unsigned int ioctl, unsigned long arg)
4112 struct kvm_vcpu *vcpu = filp->private_data;
4113 void __user *argp = compat_ptr(arg);
4116 if (vcpu->kvm->mm != current->mm || vcpu->kvm->vm_dead)
4120 case KVM_SET_SIGNAL_MASK: {
4121 struct kvm_signal_mask __user *sigmask_arg = argp;
4122 struct kvm_signal_mask kvm_sigmask;
4127 if (copy_from_user(&kvm_sigmask, argp,
4128 sizeof(kvm_sigmask)))
4131 if (kvm_sigmask.len != sizeof(compat_sigset_t))
4134 if (get_compat_sigset(&sigset,
4135 (compat_sigset_t __user *)sigmask_arg->sigset))
4137 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
4139 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
4143 r = kvm_vcpu_ioctl(filp, ioctl, arg);
4151 static int kvm_device_mmap(struct file *filp, struct vm_area_struct *vma)
4153 struct kvm_device *dev = filp->private_data;
4156 return dev->ops->mmap(dev, vma);
4161 static int kvm_device_ioctl_attr(struct kvm_device *dev,
4162 int (*accessor)(struct kvm_device *dev,
4163 struct kvm_device_attr *attr),
4166 struct kvm_device_attr attr;
4171 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
4174 return accessor(dev, &attr);
4177 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
4180 struct kvm_device *dev = filp->private_data;
4182 if (dev->kvm->mm != current->mm || dev->kvm->vm_dead)
4186 case KVM_SET_DEVICE_ATTR:
4187 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
4188 case KVM_GET_DEVICE_ATTR:
4189 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
4190 case KVM_HAS_DEVICE_ATTR:
4191 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
4193 if (dev->ops->ioctl)
4194 return dev->ops->ioctl(dev, ioctl, arg);
4200 static int kvm_device_release(struct inode *inode, struct file *filp)
4202 struct kvm_device *dev = filp->private_data;
4203 struct kvm *kvm = dev->kvm;
4205 if (dev->ops->release) {
4206 mutex_lock(&kvm->lock);
4207 list_del(&dev->vm_node);
4208 dev->ops->release(dev);
4209 mutex_unlock(&kvm->lock);
4216 static const struct file_operations kvm_device_fops = {
4217 .unlocked_ioctl = kvm_device_ioctl,
4218 .release = kvm_device_release,
4219 KVM_COMPAT(kvm_device_ioctl),
4220 .mmap = kvm_device_mmap,
4223 struct kvm_device *kvm_device_from_filp(struct file *filp)
4225 if (filp->f_op != &kvm_device_fops)
4228 return filp->private_data;
4231 static const struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
4232 #ifdef CONFIG_KVM_MPIC
4233 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
4234 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
4238 int kvm_register_device_ops(const struct kvm_device_ops *ops, u32 type)
4240 if (type >= ARRAY_SIZE(kvm_device_ops_table))
4243 if (kvm_device_ops_table[type] != NULL)
4246 kvm_device_ops_table[type] = ops;
4250 void kvm_unregister_device_ops(u32 type)
4252 if (kvm_device_ops_table[type] != NULL)
4253 kvm_device_ops_table[type] = NULL;
4256 static int kvm_ioctl_create_device(struct kvm *kvm,
4257 struct kvm_create_device *cd)
4259 const struct kvm_device_ops *ops = NULL;
4260 struct kvm_device *dev;
4261 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
4265 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
4268 type = array_index_nospec(cd->type, ARRAY_SIZE(kvm_device_ops_table));
4269 ops = kvm_device_ops_table[type];
4276 dev = kzalloc(sizeof(*dev), GFP_KERNEL_ACCOUNT);
4283 mutex_lock(&kvm->lock);
4284 ret = ops->create(dev, type);
4286 mutex_unlock(&kvm->lock);
4290 list_add(&dev->vm_node, &kvm->devices);
4291 mutex_unlock(&kvm->lock);
4297 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
4299 kvm_put_kvm_no_destroy(kvm);
4300 mutex_lock(&kvm->lock);
4301 list_del(&dev->vm_node);
4302 mutex_unlock(&kvm->lock);
4311 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
4314 case KVM_CAP_USER_MEMORY:
4315 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
4316 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
4317 case KVM_CAP_INTERNAL_ERROR_DATA:
4318 #ifdef CONFIG_HAVE_KVM_MSI
4319 case KVM_CAP_SIGNAL_MSI:
4321 #ifdef CONFIG_HAVE_KVM_IRQFD
4323 case KVM_CAP_IRQFD_RESAMPLE:
4325 case KVM_CAP_IOEVENTFD_ANY_LENGTH:
4326 case KVM_CAP_CHECK_EXTENSION_VM:
4327 case KVM_CAP_ENABLE_CAP_VM:
4328 case KVM_CAP_HALT_POLL:
4330 #ifdef CONFIG_KVM_MMIO
4331 case KVM_CAP_COALESCED_MMIO:
4332 return KVM_COALESCED_MMIO_PAGE_OFFSET;
4333 case KVM_CAP_COALESCED_PIO:
4336 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4337 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2:
4338 return KVM_DIRTY_LOG_MANUAL_CAPS;
4340 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
4341 case KVM_CAP_IRQ_ROUTING:
4342 return KVM_MAX_IRQ_ROUTES;
4344 #if KVM_ADDRESS_SPACE_NUM > 1
4345 case KVM_CAP_MULTI_ADDRESS_SPACE:
4346 return KVM_ADDRESS_SPACE_NUM;
4348 case KVM_CAP_NR_MEMSLOTS:
4349 return KVM_USER_MEM_SLOTS;
4350 case KVM_CAP_DIRTY_LOG_RING:
4351 #ifdef CONFIG_HAVE_KVM_DIRTY_RING
4352 return KVM_DIRTY_RING_MAX_ENTRIES * sizeof(struct kvm_dirty_gfn);
4356 case KVM_CAP_BINARY_STATS_FD:
4357 case KVM_CAP_SYSTEM_EVENT_DATA:
4362 return kvm_vm_ioctl_check_extension(kvm, arg);
4365 static int kvm_vm_ioctl_enable_dirty_log_ring(struct kvm *kvm, u32 size)
4369 if (!KVM_DIRTY_LOG_PAGE_OFFSET)
4372 /* the size should be power of 2 */
4373 if (!size || (size & (size - 1)))
4376 /* Should be bigger to keep the reserved entries, or a page */
4377 if (size < kvm_dirty_ring_get_rsvd_entries() *
4378 sizeof(struct kvm_dirty_gfn) || size < PAGE_SIZE)
4381 if (size > KVM_DIRTY_RING_MAX_ENTRIES *
4382 sizeof(struct kvm_dirty_gfn))
4385 /* We only allow it to set once */
4386 if (kvm->dirty_ring_size)
4389 mutex_lock(&kvm->lock);
4391 if (kvm->created_vcpus) {
4392 /* We don't allow to change this value after vcpu created */
4395 kvm->dirty_ring_size = size;
4399 mutex_unlock(&kvm->lock);
4403 static int kvm_vm_ioctl_reset_dirty_pages(struct kvm *kvm)
4406 struct kvm_vcpu *vcpu;
4409 if (!kvm->dirty_ring_size)
4412 mutex_lock(&kvm->slots_lock);
4414 kvm_for_each_vcpu(i, vcpu, kvm)
4415 cleared += kvm_dirty_ring_reset(vcpu->kvm, &vcpu->dirty_ring);
4417 mutex_unlock(&kvm->slots_lock);
4420 kvm_flush_remote_tlbs(kvm);
4425 int __attribute__((weak)) kvm_vm_ioctl_enable_cap(struct kvm *kvm,
4426 struct kvm_enable_cap *cap)
4431 static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm,
4432 struct kvm_enable_cap *cap)
4435 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4436 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2: {
4437 u64 allowed_options = KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE;
4439 if (cap->args[0] & KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE)
4440 allowed_options = KVM_DIRTY_LOG_MANUAL_CAPS;
4442 if (cap->flags || (cap->args[0] & ~allowed_options))
4444 kvm->manual_dirty_log_protect = cap->args[0];
4448 case KVM_CAP_HALT_POLL: {
4449 if (cap->flags || cap->args[0] != (unsigned int)cap->args[0])
4452 kvm->max_halt_poll_ns = cap->args[0];
4455 case KVM_CAP_DIRTY_LOG_RING:
4456 return kvm_vm_ioctl_enable_dirty_log_ring(kvm, cap->args[0]);
4458 return kvm_vm_ioctl_enable_cap(kvm, cap);
4462 static ssize_t kvm_vm_stats_read(struct file *file, char __user *user_buffer,
4463 size_t size, loff_t *offset)
4465 struct kvm *kvm = file->private_data;
4467 return kvm_stats_read(kvm->stats_id, &kvm_vm_stats_header,
4468 &kvm_vm_stats_desc[0], &kvm->stat,
4469 sizeof(kvm->stat), user_buffer, size, offset);
4472 static const struct file_operations kvm_vm_stats_fops = {
4473 .read = kvm_vm_stats_read,
4474 .llseek = noop_llseek,
4477 static int kvm_vm_ioctl_get_stats_fd(struct kvm *kvm)
4482 fd = get_unused_fd_flags(O_CLOEXEC);
4486 file = anon_inode_getfile("kvm-vm-stats",
4487 &kvm_vm_stats_fops, kvm, O_RDONLY);
4490 return PTR_ERR(file);
4492 file->f_mode |= FMODE_PREAD;
4493 fd_install(fd, file);
4498 static long kvm_vm_ioctl(struct file *filp,
4499 unsigned int ioctl, unsigned long arg)
4501 struct kvm *kvm = filp->private_data;
4502 void __user *argp = (void __user *)arg;
4505 if (kvm->mm != current->mm || kvm->vm_dead)
4508 case KVM_CREATE_VCPU:
4509 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
4511 case KVM_ENABLE_CAP: {
4512 struct kvm_enable_cap cap;
4515 if (copy_from_user(&cap, argp, sizeof(cap)))
4517 r = kvm_vm_ioctl_enable_cap_generic(kvm, &cap);
4520 case KVM_SET_USER_MEMORY_REGION: {
4521 struct kvm_userspace_memory_region kvm_userspace_mem;
4524 if (copy_from_user(&kvm_userspace_mem, argp,
4525 sizeof(kvm_userspace_mem)))
4528 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
4531 case KVM_GET_DIRTY_LOG: {
4532 struct kvm_dirty_log log;
4535 if (copy_from_user(&log, argp, sizeof(log)))
4537 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
4540 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4541 case KVM_CLEAR_DIRTY_LOG: {
4542 struct kvm_clear_dirty_log log;
4545 if (copy_from_user(&log, argp, sizeof(log)))
4547 r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
4551 #ifdef CONFIG_KVM_MMIO
4552 case KVM_REGISTER_COALESCED_MMIO: {
4553 struct kvm_coalesced_mmio_zone zone;
4556 if (copy_from_user(&zone, argp, sizeof(zone)))
4558 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
4561 case KVM_UNREGISTER_COALESCED_MMIO: {
4562 struct kvm_coalesced_mmio_zone zone;
4565 if (copy_from_user(&zone, argp, sizeof(zone)))
4567 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
4572 struct kvm_irqfd data;
4575 if (copy_from_user(&data, argp, sizeof(data)))
4577 r = kvm_irqfd(kvm, &data);
4580 case KVM_IOEVENTFD: {
4581 struct kvm_ioeventfd data;
4584 if (copy_from_user(&data, argp, sizeof(data)))
4586 r = kvm_ioeventfd(kvm, &data);
4589 #ifdef CONFIG_HAVE_KVM_MSI
4590 case KVM_SIGNAL_MSI: {
4594 if (copy_from_user(&msi, argp, sizeof(msi)))
4596 r = kvm_send_userspace_msi(kvm, &msi);
4600 #ifdef __KVM_HAVE_IRQ_LINE
4601 case KVM_IRQ_LINE_STATUS:
4602 case KVM_IRQ_LINE: {
4603 struct kvm_irq_level irq_event;
4606 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
4609 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
4610 ioctl == KVM_IRQ_LINE_STATUS);
4615 if (ioctl == KVM_IRQ_LINE_STATUS) {
4616 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
4624 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
4625 case KVM_SET_GSI_ROUTING: {
4626 struct kvm_irq_routing routing;
4627 struct kvm_irq_routing __user *urouting;
4628 struct kvm_irq_routing_entry *entries = NULL;
4631 if (copy_from_user(&routing, argp, sizeof(routing)))
4634 if (!kvm_arch_can_set_irq_routing(kvm))
4636 if (routing.nr > KVM_MAX_IRQ_ROUTES)
4642 entries = vmemdup_user(urouting->entries,
4643 array_size(sizeof(*entries),
4645 if (IS_ERR(entries)) {
4646 r = PTR_ERR(entries);
4650 r = kvm_set_irq_routing(kvm, entries, routing.nr,
4655 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
4656 case KVM_CREATE_DEVICE: {
4657 struct kvm_create_device cd;
4660 if (copy_from_user(&cd, argp, sizeof(cd)))
4663 r = kvm_ioctl_create_device(kvm, &cd);
4668 if (copy_to_user(argp, &cd, sizeof(cd)))
4674 case KVM_CHECK_EXTENSION:
4675 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
4677 case KVM_RESET_DIRTY_RINGS:
4678 r = kvm_vm_ioctl_reset_dirty_pages(kvm);
4680 case KVM_GET_STATS_FD:
4681 r = kvm_vm_ioctl_get_stats_fd(kvm);
4684 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
4690 #ifdef CONFIG_KVM_COMPAT
4691 struct compat_kvm_dirty_log {
4695 compat_uptr_t dirty_bitmap; /* one bit per page */
4700 struct compat_kvm_clear_dirty_log {
4705 compat_uptr_t dirty_bitmap; /* one bit per page */
4710 static long kvm_vm_compat_ioctl(struct file *filp,
4711 unsigned int ioctl, unsigned long arg)
4713 struct kvm *kvm = filp->private_data;
4716 if (kvm->mm != current->mm || kvm->vm_dead)
4719 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4720 case KVM_CLEAR_DIRTY_LOG: {
4721 struct compat_kvm_clear_dirty_log compat_log;
4722 struct kvm_clear_dirty_log log;
4724 if (copy_from_user(&compat_log, (void __user *)arg,
4725 sizeof(compat_log)))
4727 log.slot = compat_log.slot;
4728 log.num_pages = compat_log.num_pages;
4729 log.first_page = compat_log.first_page;
4730 log.padding2 = compat_log.padding2;
4731 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
4733 r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
4737 case KVM_GET_DIRTY_LOG: {
4738 struct compat_kvm_dirty_log compat_log;
4739 struct kvm_dirty_log log;
4741 if (copy_from_user(&compat_log, (void __user *)arg,
4742 sizeof(compat_log)))
4744 log.slot = compat_log.slot;
4745 log.padding1 = compat_log.padding1;
4746 log.padding2 = compat_log.padding2;
4747 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
4749 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
4753 r = kvm_vm_ioctl(filp, ioctl, arg);
4759 static const struct file_operations kvm_vm_fops = {
4760 .release = kvm_vm_release,
4761 .unlocked_ioctl = kvm_vm_ioctl,
4762 .llseek = noop_llseek,
4763 KVM_COMPAT(kvm_vm_compat_ioctl),
4766 bool file_is_kvm(struct file *file)
4768 return file && file->f_op == &kvm_vm_fops;
4770 EXPORT_SYMBOL_GPL(file_is_kvm);
4772 static int kvm_dev_ioctl_create_vm(unsigned long type)
4778 kvm = kvm_create_vm(type);
4780 return PTR_ERR(kvm);
4781 #ifdef CONFIG_KVM_MMIO
4782 r = kvm_coalesced_mmio_init(kvm);
4786 r = get_unused_fd_flags(O_CLOEXEC);
4790 snprintf(kvm->stats_id, sizeof(kvm->stats_id),
4791 "kvm-%d", task_pid_nr(current));
4793 file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
4801 * Don't call kvm_put_kvm anymore at this point; file->f_op is
4802 * already set, with ->release() being kvm_vm_release(). In error
4803 * cases it will be called by the final fput(file) and will take
4804 * care of doing kvm_put_kvm(kvm).
4806 if (kvm_create_vm_debugfs(kvm, r) < 0) {
4811 kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm);
4813 fd_install(r, file);
4821 static long kvm_dev_ioctl(struct file *filp,
4822 unsigned int ioctl, unsigned long arg)
4827 case KVM_GET_API_VERSION:
4830 r = KVM_API_VERSION;
4833 r = kvm_dev_ioctl_create_vm(arg);
4835 case KVM_CHECK_EXTENSION:
4836 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
4838 case KVM_GET_VCPU_MMAP_SIZE:
4841 r = PAGE_SIZE; /* struct kvm_run */
4843 r += PAGE_SIZE; /* pio data page */
4845 #ifdef CONFIG_KVM_MMIO
4846 r += PAGE_SIZE; /* coalesced mmio ring page */
4849 case KVM_TRACE_ENABLE:
4850 case KVM_TRACE_PAUSE:
4851 case KVM_TRACE_DISABLE:
4855 return kvm_arch_dev_ioctl(filp, ioctl, arg);
4861 static struct file_operations kvm_chardev_ops = {
4862 .unlocked_ioctl = kvm_dev_ioctl,
4863 .llseek = noop_llseek,
4864 KVM_COMPAT(kvm_dev_ioctl),
4867 static struct miscdevice kvm_dev = {
4873 static void hardware_enable_nolock(void *junk)
4875 int cpu = raw_smp_processor_id();
4878 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
4881 cpumask_set_cpu(cpu, cpus_hardware_enabled);
4883 r = kvm_arch_hardware_enable();
4886 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
4887 atomic_inc(&hardware_enable_failed);
4888 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
4892 static int kvm_starting_cpu(unsigned int cpu)
4894 raw_spin_lock(&kvm_count_lock);
4895 if (kvm_usage_count)
4896 hardware_enable_nolock(NULL);
4897 raw_spin_unlock(&kvm_count_lock);
4901 static void hardware_disable_nolock(void *junk)
4903 int cpu = raw_smp_processor_id();
4905 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
4907 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
4908 kvm_arch_hardware_disable();
4911 static int kvm_dying_cpu(unsigned int cpu)
4913 raw_spin_lock(&kvm_count_lock);
4914 if (kvm_usage_count)
4915 hardware_disable_nolock(NULL);
4916 raw_spin_unlock(&kvm_count_lock);
4920 static void hardware_disable_all_nolock(void)
4922 BUG_ON(!kvm_usage_count);
4925 if (!kvm_usage_count)
4926 on_each_cpu(hardware_disable_nolock, NULL, 1);
4929 static void hardware_disable_all(void)
4931 raw_spin_lock(&kvm_count_lock);
4932 hardware_disable_all_nolock();
4933 raw_spin_unlock(&kvm_count_lock);
4936 static int hardware_enable_all(void)
4940 raw_spin_lock(&kvm_count_lock);
4943 if (kvm_usage_count == 1) {
4944 atomic_set(&hardware_enable_failed, 0);
4945 on_each_cpu(hardware_enable_nolock, NULL, 1);
4947 if (atomic_read(&hardware_enable_failed)) {
4948 hardware_disable_all_nolock();
4953 raw_spin_unlock(&kvm_count_lock);
4958 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
4962 * Some (well, at least mine) BIOSes hang on reboot if
4965 * And Intel TXT required VMX off for all cpu when system shutdown.
4967 pr_info("kvm: exiting hardware virtualization\n");
4968 kvm_rebooting = true;
4969 on_each_cpu(hardware_disable_nolock, NULL, 1);
4973 static struct notifier_block kvm_reboot_notifier = {
4974 .notifier_call = kvm_reboot,
4978 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
4982 for (i = 0; i < bus->dev_count; i++) {
4983 struct kvm_io_device *pos = bus->range[i].dev;
4985 kvm_iodevice_destructor(pos);
4990 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
4991 const struct kvm_io_range *r2)
4993 gpa_t addr1 = r1->addr;
4994 gpa_t addr2 = r2->addr;
4999 /* If r2->len == 0, match the exact address. If r2->len != 0,
5000 * accept any overlapping write. Any order is acceptable for
5001 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
5002 * we process all of them.
5015 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
5017 return kvm_io_bus_cmp(p1, p2);
5020 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
5021 gpa_t addr, int len)
5023 struct kvm_io_range *range, key;
5026 key = (struct kvm_io_range) {
5031 range = bsearch(&key, bus->range, bus->dev_count,
5032 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
5036 off = range - bus->range;
5038 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
5044 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
5045 struct kvm_io_range *range, const void *val)
5049 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
5053 while (idx < bus->dev_count &&
5054 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
5055 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
5064 /* kvm_io_bus_write - called under kvm->slots_lock */
5065 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
5066 int len, const void *val)
5068 struct kvm_io_bus *bus;
5069 struct kvm_io_range range;
5072 range = (struct kvm_io_range) {
5077 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
5080 r = __kvm_io_bus_write(vcpu, bus, &range, val);
5081 return r < 0 ? r : 0;
5083 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
5085 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
5086 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
5087 gpa_t addr, int len, const void *val, long cookie)
5089 struct kvm_io_bus *bus;
5090 struct kvm_io_range range;
5092 range = (struct kvm_io_range) {
5097 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
5101 /* First try the device referenced by cookie. */
5102 if ((cookie >= 0) && (cookie < bus->dev_count) &&
5103 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
5104 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
5109 * cookie contained garbage; fall back to search and return the
5110 * correct cookie value.
5112 return __kvm_io_bus_write(vcpu, bus, &range, val);
5115 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
5116 struct kvm_io_range *range, void *val)
5120 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
5124 while (idx < bus->dev_count &&
5125 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
5126 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
5135 /* kvm_io_bus_read - called under kvm->slots_lock */
5136 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
5139 struct kvm_io_bus *bus;
5140 struct kvm_io_range range;
5143 range = (struct kvm_io_range) {
5148 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
5151 r = __kvm_io_bus_read(vcpu, bus, &range, val);
5152 return r < 0 ? r : 0;
5155 /* Caller must hold slots_lock. */
5156 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
5157 int len, struct kvm_io_device *dev)
5160 struct kvm_io_bus *new_bus, *bus;
5161 struct kvm_io_range range;
5163 bus = kvm_get_bus(kvm, bus_idx);
5167 /* exclude ioeventfd which is limited by maximum fd */
5168 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
5171 new_bus = kmalloc(struct_size(bus, range, bus->dev_count + 1),
5172 GFP_KERNEL_ACCOUNT);
5176 range = (struct kvm_io_range) {
5182 for (i = 0; i < bus->dev_count; i++)
5183 if (kvm_io_bus_cmp(&bus->range[i], &range) > 0)
5186 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
5187 new_bus->dev_count++;
5188 new_bus->range[i] = range;
5189 memcpy(new_bus->range + i + 1, bus->range + i,
5190 (bus->dev_count - i) * sizeof(struct kvm_io_range));
5191 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
5192 synchronize_srcu_expedited(&kvm->srcu);
5198 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
5199 struct kvm_io_device *dev)
5202 struct kvm_io_bus *new_bus, *bus;
5204 lockdep_assert_held(&kvm->slots_lock);
5206 bus = kvm_get_bus(kvm, bus_idx);
5210 for (i = 0; i < bus->dev_count; i++) {
5211 if (bus->range[i].dev == dev) {
5216 if (i == bus->dev_count)
5219 new_bus = kmalloc(struct_size(bus, range, bus->dev_count - 1),
5220 GFP_KERNEL_ACCOUNT);
5222 memcpy(new_bus, bus, struct_size(bus, range, i));
5223 new_bus->dev_count--;
5224 memcpy(new_bus->range + i, bus->range + i + 1,
5225 flex_array_size(new_bus, range, new_bus->dev_count - i));
5228 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
5229 synchronize_srcu_expedited(&kvm->srcu);
5231 /* Destroy the old bus _after_ installing the (null) bus. */
5233 pr_err("kvm: failed to shrink bus, removing it completely\n");
5234 for (j = 0; j < bus->dev_count; j++) {
5237 kvm_iodevice_destructor(bus->range[j].dev);
5242 return new_bus ? 0 : -ENOMEM;
5245 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
5248 struct kvm_io_bus *bus;
5249 int dev_idx, srcu_idx;
5250 struct kvm_io_device *iodev = NULL;
5252 srcu_idx = srcu_read_lock(&kvm->srcu);
5254 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
5258 dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
5262 iodev = bus->range[dev_idx].dev;
5265 srcu_read_unlock(&kvm->srcu, srcu_idx);
5269 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
5271 static int kvm_debugfs_open(struct inode *inode, struct file *file,
5272 int (*get)(void *, u64 *), int (*set)(void *, u64),
5275 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
5279 * The debugfs files are a reference to the kvm struct which
5280 * is still valid when kvm_destroy_vm is called. kvm_get_kvm_safe
5281 * avoids the race between open and the removal of the debugfs directory.
5283 if (!kvm_get_kvm_safe(stat_data->kvm))
5286 if (simple_attr_open(inode, file, get,
5287 kvm_stats_debugfs_mode(stat_data->desc) & 0222
5290 kvm_put_kvm(stat_data->kvm);
5297 static int kvm_debugfs_release(struct inode *inode, struct file *file)
5299 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
5302 simple_attr_release(inode, file);
5303 kvm_put_kvm(stat_data->kvm);
5308 static int kvm_get_stat_per_vm(struct kvm *kvm, size_t offset, u64 *val)
5310 *val = *(u64 *)((void *)(&kvm->stat) + offset);
5315 static int kvm_clear_stat_per_vm(struct kvm *kvm, size_t offset)
5317 *(u64 *)((void *)(&kvm->stat) + offset) = 0;
5322 static int kvm_get_stat_per_vcpu(struct kvm *kvm, size_t offset, u64 *val)
5325 struct kvm_vcpu *vcpu;
5329 kvm_for_each_vcpu(i, vcpu, kvm)
5330 *val += *(u64 *)((void *)(&vcpu->stat) + offset);
5335 static int kvm_clear_stat_per_vcpu(struct kvm *kvm, size_t offset)
5338 struct kvm_vcpu *vcpu;
5340 kvm_for_each_vcpu(i, vcpu, kvm)
5341 *(u64 *)((void *)(&vcpu->stat) + offset) = 0;
5346 static int kvm_stat_data_get(void *data, u64 *val)
5349 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
5351 switch (stat_data->kind) {
5353 r = kvm_get_stat_per_vm(stat_data->kvm,
5354 stat_data->desc->desc.offset, val);
5357 r = kvm_get_stat_per_vcpu(stat_data->kvm,
5358 stat_data->desc->desc.offset, val);
5365 static int kvm_stat_data_clear(void *data, u64 val)
5368 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
5373 switch (stat_data->kind) {
5375 r = kvm_clear_stat_per_vm(stat_data->kvm,
5376 stat_data->desc->desc.offset);
5379 r = kvm_clear_stat_per_vcpu(stat_data->kvm,
5380 stat_data->desc->desc.offset);
5387 static int kvm_stat_data_open(struct inode *inode, struct file *file)
5389 __simple_attr_check_format("%llu\n", 0ull);
5390 return kvm_debugfs_open(inode, file, kvm_stat_data_get,
5391 kvm_stat_data_clear, "%llu\n");
5394 static const struct file_operations stat_fops_per_vm = {
5395 .owner = THIS_MODULE,
5396 .open = kvm_stat_data_open,
5397 .release = kvm_debugfs_release,
5398 .read = simple_attr_read,
5399 .write = simple_attr_write,
5400 .llseek = no_llseek,
5403 static int vm_stat_get(void *_offset, u64 *val)
5405 unsigned offset = (long)_offset;
5410 mutex_lock(&kvm_lock);
5411 list_for_each_entry(kvm, &vm_list, vm_list) {
5412 kvm_get_stat_per_vm(kvm, offset, &tmp_val);
5415 mutex_unlock(&kvm_lock);
5419 static int vm_stat_clear(void *_offset, u64 val)
5421 unsigned offset = (long)_offset;
5427 mutex_lock(&kvm_lock);
5428 list_for_each_entry(kvm, &vm_list, vm_list) {
5429 kvm_clear_stat_per_vm(kvm, offset);
5431 mutex_unlock(&kvm_lock);
5436 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n");
5437 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_readonly_fops, vm_stat_get, NULL, "%llu\n");
5439 static int vcpu_stat_get(void *_offset, u64 *val)
5441 unsigned offset = (long)_offset;
5446 mutex_lock(&kvm_lock);
5447 list_for_each_entry(kvm, &vm_list, vm_list) {
5448 kvm_get_stat_per_vcpu(kvm, offset, &tmp_val);
5451 mutex_unlock(&kvm_lock);
5455 static int vcpu_stat_clear(void *_offset, u64 val)
5457 unsigned offset = (long)_offset;
5463 mutex_lock(&kvm_lock);
5464 list_for_each_entry(kvm, &vm_list, vm_list) {
5465 kvm_clear_stat_per_vcpu(kvm, offset);
5467 mutex_unlock(&kvm_lock);
5472 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear,
5474 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_readonly_fops, vcpu_stat_get, NULL, "%llu\n");
5476 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm)
5478 struct kobj_uevent_env *env;
5479 unsigned long long created, active;
5481 if (!kvm_dev.this_device || !kvm)
5484 mutex_lock(&kvm_lock);
5485 if (type == KVM_EVENT_CREATE_VM) {
5486 kvm_createvm_count++;
5488 } else if (type == KVM_EVENT_DESTROY_VM) {
5491 created = kvm_createvm_count;
5492 active = kvm_active_vms;
5493 mutex_unlock(&kvm_lock);
5495 env = kzalloc(sizeof(*env), GFP_KERNEL_ACCOUNT);
5499 add_uevent_var(env, "CREATED=%llu", created);
5500 add_uevent_var(env, "COUNT=%llu", active);
5502 if (type == KVM_EVENT_CREATE_VM) {
5503 add_uevent_var(env, "EVENT=create");
5504 kvm->userspace_pid = task_pid_nr(current);
5505 } else if (type == KVM_EVENT_DESTROY_VM) {
5506 add_uevent_var(env, "EVENT=destroy");
5508 add_uevent_var(env, "PID=%d", kvm->userspace_pid);
5510 if (!IS_ERR(kvm->debugfs_dentry)) {
5511 char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL_ACCOUNT);
5514 tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX);
5516 add_uevent_var(env, "STATS_PATH=%s", tmp);
5520 /* no need for checks, since we are adding at most only 5 keys */
5521 env->envp[env->envp_idx++] = NULL;
5522 kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp);
5526 static void kvm_init_debug(void)
5528 const struct file_operations *fops;
5529 const struct _kvm_stats_desc *pdesc;
5532 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
5534 for (i = 0; i < kvm_vm_stats_header.num_desc; ++i) {
5535 pdesc = &kvm_vm_stats_desc[i];
5536 if (kvm_stats_debugfs_mode(pdesc) & 0222)
5537 fops = &vm_stat_fops;
5539 fops = &vm_stat_readonly_fops;
5540 debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
5542 (void *)(long)pdesc->desc.offset, fops);
5545 for (i = 0; i < kvm_vcpu_stats_header.num_desc; ++i) {
5546 pdesc = &kvm_vcpu_stats_desc[i];
5547 if (kvm_stats_debugfs_mode(pdesc) & 0222)
5548 fops = &vcpu_stat_fops;
5550 fops = &vcpu_stat_readonly_fops;
5551 debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
5553 (void *)(long)pdesc->desc.offset, fops);
5557 static int kvm_suspend(void)
5559 if (kvm_usage_count)
5560 hardware_disable_nolock(NULL);
5564 static void kvm_resume(void)
5566 if (kvm_usage_count) {
5567 lockdep_assert_not_held(&kvm_count_lock);
5568 hardware_enable_nolock(NULL);
5572 static struct syscore_ops kvm_syscore_ops = {
5573 .suspend = kvm_suspend,
5574 .resume = kvm_resume,
5578 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
5580 return container_of(pn, struct kvm_vcpu, preempt_notifier);
5583 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
5585 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
5587 WRITE_ONCE(vcpu->preempted, false);
5588 WRITE_ONCE(vcpu->ready, false);
5590 __this_cpu_write(kvm_running_vcpu, vcpu);
5591 kvm_arch_sched_in(vcpu, cpu);
5592 kvm_arch_vcpu_load(vcpu, cpu);
5595 static void kvm_sched_out(struct preempt_notifier *pn,
5596 struct task_struct *next)
5598 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
5600 if (current->on_rq) {
5601 WRITE_ONCE(vcpu->preempted, true);
5602 WRITE_ONCE(vcpu->ready, true);
5604 kvm_arch_vcpu_put(vcpu);
5605 __this_cpu_write(kvm_running_vcpu, NULL);
5609 * kvm_get_running_vcpu - get the vcpu running on the current CPU.
5611 * We can disable preemption locally around accessing the per-CPU variable,
5612 * and use the resolved vcpu pointer after enabling preemption again,
5613 * because even if the current thread is migrated to another CPU, reading
5614 * the per-CPU value later will give us the same value as we update the
5615 * per-CPU variable in the preempt notifier handlers.
5617 struct kvm_vcpu *kvm_get_running_vcpu(void)
5619 struct kvm_vcpu *vcpu;
5622 vcpu = __this_cpu_read(kvm_running_vcpu);
5627 EXPORT_SYMBOL_GPL(kvm_get_running_vcpu);
5630 * kvm_get_running_vcpus - get the per-CPU array of currently running vcpus.
5632 struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void)
5634 return &kvm_running_vcpu;
5637 #ifdef CONFIG_GUEST_PERF_EVENTS
5638 static unsigned int kvm_guest_state(void)
5640 struct kvm_vcpu *vcpu = kvm_get_running_vcpu();
5643 if (!kvm_arch_pmi_in_guest(vcpu))
5646 state = PERF_GUEST_ACTIVE;
5647 if (!kvm_arch_vcpu_in_kernel(vcpu))
5648 state |= PERF_GUEST_USER;
5653 static unsigned long kvm_guest_get_ip(void)
5655 struct kvm_vcpu *vcpu = kvm_get_running_vcpu();
5657 /* Retrieving the IP must be guarded by a call to kvm_guest_state(). */
5658 if (WARN_ON_ONCE(!kvm_arch_pmi_in_guest(vcpu)))
5661 return kvm_arch_vcpu_get_ip(vcpu);
5664 static struct perf_guest_info_callbacks kvm_guest_cbs = {
5665 .state = kvm_guest_state,
5666 .get_ip = kvm_guest_get_ip,
5667 .handle_intel_pt_intr = NULL,
5670 void kvm_register_perf_callbacks(unsigned int (*pt_intr_handler)(void))
5672 kvm_guest_cbs.handle_intel_pt_intr = pt_intr_handler;
5673 perf_register_guest_info_callbacks(&kvm_guest_cbs);
5675 void kvm_unregister_perf_callbacks(void)
5677 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
5681 struct kvm_cpu_compat_check {
5686 static void check_processor_compat(void *data)
5688 struct kvm_cpu_compat_check *c = data;
5690 *c->ret = kvm_arch_check_processor_compat(c->opaque);
5693 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
5694 struct module *module)
5696 struct kvm_cpu_compat_check c;
5700 r = kvm_arch_init(opaque);
5705 * kvm_arch_init makes sure there's at most one caller
5706 * for architectures that support multiple implementations,
5707 * like intel and amd on x86.
5708 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
5709 * conflicts in case kvm is already setup for another implementation.
5711 r = kvm_irqfd_init();
5715 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
5720 r = kvm_arch_hardware_setup(opaque);
5726 for_each_online_cpu(cpu) {
5727 smp_call_function_single(cpu, check_processor_compat, &c, 1);
5732 r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting",
5733 kvm_starting_cpu, kvm_dying_cpu);
5736 register_reboot_notifier(&kvm_reboot_notifier);
5738 /* A kmem cache lets us meet the alignment requirements of fx_save. */
5740 vcpu_align = __alignof__(struct kvm_vcpu);
5742 kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align,
5744 offsetof(struct kvm_vcpu, arch),
5745 offsetofend(struct kvm_vcpu, stats_id)
5746 - offsetof(struct kvm_vcpu, arch),
5748 if (!kvm_vcpu_cache) {
5753 for_each_possible_cpu(cpu) {
5754 if (!alloc_cpumask_var_node(&per_cpu(cpu_kick_mask, cpu),
5755 GFP_KERNEL, cpu_to_node(cpu))) {
5761 r = kvm_async_pf_init();
5765 kvm_chardev_ops.owner = module;
5767 r = misc_register(&kvm_dev);
5769 pr_err("kvm: misc device register failed\n");
5773 register_syscore_ops(&kvm_syscore_ops);
5775 kvm_preempt_ops.sched_in = kvm_sched_in;
5776 kvm_preempt_ops.sched_out = kvm_sched_out;
5780 r = kvm_vfio_ops_init();
5786 kvm_async_pf_deinit();
5788 for_each_possible_cpu(cpu)
5789 free_cpumask_var(per_cpu(cpu_kick_mask, cpu));
5791 kmem_cache_destroy(kvm_vcpu_cache);
5793 unregister_reboot_notifier(&kvm_reboot_notifier);
5794 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
5796 kvm_arch_hardware_unsetup();
5798 free_cpumask_var(cpus_hardware_enabled);
5806 EXPORT_SYMBOL_GPL(kvm_init);
5812 debugfs_remove_recursive(kvm_debugfs_dir);
5813 misc_deregister(&kvm_dev);
5814 for_each_possible_cpu(cpu)
5815 free_cpumask_var(per_cpu(cpu_kick_mask, cpu));
5816 kmem_cache_destroy(kvm_vcpu_cache);
5817 kvm_async_pf_deinit();
5818 unregister_syscore_ops(&kvm_syscore_ops);
5819 unregister_reboot_notifier(&kvm_reboot_notifier);
5820 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
5821 on_each_cpu(hardware_disable_nolock, NULL, 1);
5822 kvm_arch_hardware_unsetup();
5825 free_cpumask_var(cpus_hardware_enabled);
5826 kvm_vfio_ops_exit();
5828 EXPORT_SYMBOL_GPL(kvm_exit);
5830 struct kvm_vm_worker_thread_context {
5832 struct task_struct *parent;
5833 struct completion init_done;
5834 kvm_vm_thread_fn_t thread_fn;
5839 static int kvm_vm_worker_thread(void *context)
5842 * The init_context is allocated on the stack of the parent thread, so
5843 * we have to locally copy anything that is needed beyond initialization
5845 struct kvm_vm_worker_thread_context *init_context = context;
5846 struct task_struct *parent;
5847 struct kvm *kvm = init_context->kvm;
5848 kvm_vm_thread_fn_t thread_fn = init_context->thread_fn;
5849 uintptr_t data = init_context->data;
5852 err = kthread_park(current);
5853 /* kthread_park(current) is never supposed to return an error */
5858 err = cgroup_attach_task_all(init_context->parent, current);
5860 kvm_err("%s: cgroup_attach_task_all failed with err %d\n",
5865 set_user_nice(current, task_nice(init_context->parent));
5868 init_context->err = err;
5869 complete(&init_context->init_done);
5870 init_context = NULL;
5875 /* Wait to be woken up by the spawner before proceeding. */
5878 if (!kthread_should_stop())
5879 err = thread_fn(kvm, data);
5883 * Move kthread back to its original cgroup to prevent it lingering in
5884 * the cgroup of the VM process, after the latter finishes its
5887 * kthread_stop() waits on the 'exited' completion condition which is
5888 * set in exit_mm(), via mm_release(), in do_exit(). However, the
5889 * kthread is removed from the cgroup in the cgroup_exit() which is
5890 * called after the exit_mm(). This causes the kthread_stop() to return
5891 * before the kthread actually quits the cgroup.
5894 parent = rcu_dereference(current->real_parent);
5895 get_task_struct(parent);
5897 cgroup_attach_task_all(parent, current);
5898 put_task_struct(parent);
5903 int kvm_vm_create_worker_thread(struct kvm *kvm, kvm_vm_thread_fn_t thread_fn,
5904 uintptr_t data, const char *name,
5905 struct task_struct **thread_ptr)
5907 struct kvm_vm_worker_thread_context init_context = {};
5908 struct task_struct *thread;
5911 init_context.kvm = kvm;
5912 init_context.parent = current;
5913 init_context.thread_fn = thread_fn;
5914 init_context.data = data;
5915 init_completion(&init_context.init_done);
5917 thread = kthread_run(kvm_vm_worker_thread, &init_context,
5918 "%s-%d", name, task_pid_nr(current));
5920 return PTR_ERR(thread);
5922 /* kthread_run is never supposed to return NULL */
5923 WARN_ON(thread == NULL);
5925 wait_for_completion(&init_context.init_done);
5927 if (!init_context.err)
5928 *thread_ptr = thread;
5930 return init_context.err;