1 // SPDX-License-Identifier: GPL-2.0-only
3 * Kernel-based Virtual Machine driver for Linux
5 * This module enables machines with Intel VT-x extensions to run virtual
6 * machines without emulation or binary translation.
8 * Copyright (C) 2006 Qumranet, Inc.
9 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
12 * Avi Kivity <avi@qumranet.com>
13 * Yaniv Kamay <yaniv@qumranet.com>
16 #include <kvm/iodev.h>
18 #include <linux/kvm_host.h>
19 #include <linux/kvm.h>
20 #include <linux/module.h>
21 #include <linux/errno.h>
22 #include <linux/percpu.h>
24 #include <linux/miscdevice.h>
25 #include <linux/vmalloc.h>
26 #include <linux/reboot.h>
27 #include <linux/debugfs.h>
28 #include <linux/highmem.h>
29 #include <linux/file.h>
30 #include <linux/syscore_ops.h>
31 #include <linux/cpu.h>
32 #include <linux/sched/signal.h>
33 #include <linux/sched/mm.h>
34 #include <linux/sched/stat.h>
35 #include <linux/cpumask.h>
36 #include <linux/smp.h>
37 #include <linux/anon_inodes.h>
38 #include <linux/profile.h>
39 #include <linux/kvm_para.h>
40 #include <linux/pagemap.h>
41 #include <linux/mman.h>
42 #include <linux/swap.h>
43 #include <linux/bitops.h>
44 #include <linux/spinlock.h>
45 #include <linux/compat.h>
46 #include <linux/srcu.h>
47 #include <linux/hugetlb.h>
48 #include <linux/slab.h>
49 #include <linux/sort.h>
50 #include <linux/bsearch.h>
52 #include <linux/lockdep.h>
53 #include <linux/kthread.h>
54 #include <linux/suspend.h>
56 #include <asm/processor.h>
57 #include <asm/ioctl.h>
58 #include <linux/uaccess.h>
60 #include "coalesced_mmio.h"
65 #include <trace/events/ipi.h>
67 #define CREATE_TRACE_POINTS
68 #include <trace/events/kvm.h>
70 #include <linux/kvm_dirty_ring.h>
73 /* Worst case buffer size needed for holding an integer. */
74 #define ITOA_MAX_LEN 12
76 MODULE_AUTHOR("Qumranet");
77 MODULE_LICENSE("GPL");
79 /* Architectures should define their poll value according to the halt latency */
80 unsigned int halt_poll_ns = KVM_HALT_POLL_NS_DEFAULT;
81 module_param(halt_poll_ns, uint, 0644);
82 EXPORT_SYMBOL_GPL(halt_poll_ns);
84 /* Default doubles per-vcpu halt_poll_ns. */
85 unsigned int halt_poll_ns_grow = 2;
86 module_param(halt_poll_ns_grow, uint, 0644);
87 EXPORT_SYMBOL_GPL(halt_poll_ns_grow);
89 /* The start value to grow halt_poll_ns from */
90 unsigned int halt_poll_ns_grow_start = 10000; /* 10us */
91 module_param(halt_poll_ns_grow_start, uint, 0644);
92 EXPORT_SYMBOL_GPL(halt_poll_ns_grow_start);
94 /* Default resets per-vcpu halt_poll_ns . */
95 unsigned int halt_poll_ns_shrink;
96 module_param(halt_poll_ns_shrink, uint, 0644);
97 EXPORT_SYMBOL_GPL(halt_poll_ns_shrink);
102 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
105 DEFINE_MUTEX(kvm_lock);
108 static struct kmem_cache *kvm_vcpu_cache;
110 static __read_mostly struct preempt_ops kvm_preempt_ops;
111 static DEFINE_PER_CPU(struct kvm_vcpu *, kvm_running_vcpu);
113 struct dentry *kvm_debugfs_dir;
114 EXPORT_SYMBOL_GPL(kvm_debugfs_dir);
116 static const struct file_operations stat_fops_per_vm;
118 static struct file_operations kvm_chardev_ops;
120 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
122 #ifdef CONFIG_KVM_COMPAT
123 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
125 #define KVM_COMPAT(c) .compat_ioctl = (c)
128 * For architectures that don't implement a compat infrastructure,
129 * adopt a double line of defense:
130 * - Prevent a compat task from opening /dev/kvm
131 * - If the open has been done by a 64bit task, and the KVM fd
132 * passed to a compat task, let the ioctls fail.
134 static long kvm_no_compat_ioctl(struct file *file, unsigned int ioctl,
135 unsigned long arg) { return -EINVAL; }
137 static int kvm_no_compat_open(struct inode *inode, struct file *file)
139 return is_compat_task() ? -ENODEV : 0;
141 #define KVM_COMPAT(c) .compat_ioctl = kvm_no_compat_ioctl, \
142 .open = kvm_no_compat_open
144 static int hardware_enable_all(void);
145 static void hardware_disable_all(void);
147 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
149 #define KVM_EVENT_CREATE_VM 0
150 #define KVM_EVENT_DESTROY_VM 1
151 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm);
152 static unsigned long long kvm_createvm_count;
153 static unsigned long long kvm_active_vms;
155 static DEFINE_PER_CPU(cpumask_var_t, cpu_kick_mask);
157 __weak void kvm_arch_guest_memory_reclaimed(struct kvm *kvm)
161 bool kvm_is_zone_device_page(struct page *page)
164 * The metadata used by is_zone_device_page() to determine whether or
165 * not a page is ZONE_DEVICE is guaranteed to be valid if and only if
166 * the device has been pinned, e.g. by get_user_pages(). WARN if the
167 * page_count() is zero to help detect bad usage of this helper.
169 if (WARN_ON_ONCE(!page_count(page)))
172 return is_zone_device_page(page);
176 * Returns a 'struct page' if the pfn is "valid" and backed by a refcounted
177 * page, NULL otherwise. Note, the list of refcounted PG_reserved page types
178 * is likely incomplete, it has been compiled purely through people wanting to
179 * back guest with a certain type of memory and encountering issues.
181 struct page *kvm_pfn_to_refcounted_page(kvm_pfn_t pfn)
188 page = pfn_to_page(pfn);
189 if (!PageReserved(page))
192 /* The ZERO_PAGE(s) is marked PG_reserved, but is refcounted. */
193 if (is_zero_pfn(pfn))
197 * ZONE_DEVICE pages currently set PG_reserved, but from a refcounting
198 * perspective they are "normal" pages, albeit with slightly different
201 if (kvm_is_zone_device_page(page))
208 * Switches to specified vcpu, until a matching vcpu_put()
210 void vcpu_load(struct kvm_vcpu *vcpu)
214 __this_cpu_write(kvm_running_vcpu, vcpu);
215 preempt_notifier_register(&vcpu->preempt_notifier);
216 kvm_arch_vcpu_load(vcpu, cpu);
219 EXPORT_SYMBOL_GPL(vcpu_load);
221 void vcpu_put(struct kvm_vcpu *vcpu)
224 kvm_arch_vcpu_put(vcpu);
225 preempt_notifier_unregister(&vcpu->preempt_notifier);
226 __this_cpu_write(kvm_running_vcpu, NULL);
229 EXPORT_SYMBOL_GPL(vcpu_put);
231 /* TODO: merge with kvm_arch_vcpu_should_kick */
232 static bool kvm_request_needs_ipi(struct kvm_vcpu *vcpu, unsigned req)
234 int mode = kvm_vcpu_exiting_guest_mode(vcpu);
237 * We need to wait for the VCPU to reenable interrupts and get out of
238 * READING_SHADOW_PAGE_TABLES mode.
240 if (req & KVM_REQUEST_WAIT)
241 return mode != OUTSIDE_GUEST_MODE;
244 * Need to kick a running VCPU, but otherwise there is nothing to do.
246 return mode == IN_GUEST_MODE;
249 static void ack_kick(void *_completed)
253 static inline bool kvm_kick_many_cpus(struct cpumask *cpus, bool wait)
255 if (cpumask_empty(cpus))
258 smp_call_function_many(cpus, ack_kick, NULL, wait);
262 static void kvm_make_vcpu_request(struct kvm_vcpu *vcpu, unsigned int req,
263 struct cpumask *tmp, int current_cpu)
267 if (likely(!(req & KVM_REQUEST_NO_ACTION)))
268 __kvm_make_request(req, vcpu);
270 if (!(req & KVM_REQUEST_NO_WAKEUP) && kvm_vcpu_wake_up(vcpu))
274 * Note, the vCPU could get migrated to a different pCPU at any point
275 * after kvm_request_needs_ipi(), which could result in sending an IPI
276 * to the previous pCPU. But, that's OK because the purpose of the IPI
277 * is to ensure the vCPU returns to OUTSIDE_GUEST_MODE, which is
278 * satisfied if the vCPU migrates. Entering READING_SHADOW_PAGE_TABLES
279 * after this point is also OK, as the requirement is only that KVM wait
280 * for vCPUs that were reading SPTEs _before_ any changes were
281 * finalized. See kvm_vcpu_kick() for more details on handling requests.
283 if (kvm_request_needs_ipi(vcpu, req)) {
284 cpu = READ_ONCE(vcpu->cpu);
285 if (cpu != -1 && cpu != current_cpu)
286 __cpumask_set_cpu(cpu, tmp);
290 bool kvm_make_vcpus_request_mask(struct kvm *kvm, unsigned int req,
291 unsigned long *vcpu_bitmap)
293 struct kvm_vcpu *vcpu;
294 struct cpumask *cpus;
300 cpus = this_cpu_cpumask_var_ptr(cpu_kick_mask);
303 for_each_set_bit(i, vcpu_bitmap, KVM_MAX_VCPUS) {
304 vcpu = kvm_get_vcpu(kvm, i);
307 kvm_make_vcpu_request(vcpu, req, cpus, me);
310 called = kvm_kick_many_cpus(cpus, !!(req & KVM_REQUEST_WAIT));
316 bool kvm_make_all_cpus_request_except(struct kvm *kvm, unsigned int req,
317 struct kvm_vcpu *except)
319 struct kvm_vcpu *vcpu;
320 struct cpumask *cpus;
327 cpus = this_cpu_cpumask_var_ptr(cpu_kick_mask);
330 kvm_for_each_vcpu(i, vcpu, kvm) {
333 kvm_make_vcpu_request(vcpu, req, cpus, me);
336 called = kvm_kick_many_cpus(cpus, !!(req & KVM_REQUEST_WAIT));
342 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
344 return kvm_make_all_cpus_request_except(kvm, req, NULL);
346 EXPORT_SYMBOL_GPL(kvm_make_all_cpus_request);
348 void kvm_flush_remote_tlbs(struct kvm *kvm)
350 ++kvm->stat.generic.remote_tlb_flush_requests;
353 * We want to publish modifications to the page tables before reading
354 * mode. Pairs with a memory barrier in arch-specific code.
355 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
356 * and smp_mb in walk_shadow_page_lockless_begin/end.
357 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
359 * There is already an smp_mb__after_atomic() before
360 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
363 if (!kvm_arch_flush_remote_tlbs(kvm)
364 || kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
365 ++kvm->stat.generic.remote_tlb_flush;
367 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
369 void kvm_flush_remote_tlbs_range(struct kvm *kvm, gfn_t gfn, u64 nr_pages)
371 if (!kvm_arch_flush_remote_tlbs_range(kvm, gfn, nr_pages))
375 * Fall back to a flushing entire TLBs if the architecture range-based
376 * TLB invalidation is unsupported or can't be performed for whatever
379 kvm_flush_remote_tlbs(kvm);
382 void kvm_flush_remote_tlbs_memslot(struct kvm *kvm,
383 const struct kvm_memory_slot *memslot)
386 * All current use cases for flushing the TLBs for a specific memslot
387 * are related to dirty logging, and many do the TLB flush out of
388 * mmu_lock. The interaction between the various operations on memslot
389 * must be serialized by slots_locks to ensure the TLB flush from one
390 * operation is observed by any other operation on the same memslot.
392 lockdep_assert_held(&kvm->slots_lock);
393 kvm_flush_remote_tlbs_range(kvm, memslot->base_gfn, memslot->npages);
396 static void kvm_flush_shadow_all(struct kvm *kvm)
398 kvm_arch_flush_shadow_all(kvm);
399 kvm_arch_guest_memory_reclaimed(kvm);
402 #ifdef KVM_ARCH_NR_OBJS_PER_MEMORY_CACHE
403 static inline void *mmu_memory_cache_alloc_obj(struct kvm_mmu_memory_cache *mc,
406 gfp_flags |= mc->gfp_zero;
409 return kmem_cache_alloc(mc->kmem_cache, gfp_flags);
411 return (void *)__get_free_page(gfp_flags);
414 int __kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache *mc, int capacity, int min)
416 gfp_t gfp = mc->gfp_custom ? mc->gfp_custom : GFP_KERNEL_ACCOUNT;
419 if (mc->nobjs >= min)
422 if (unlikely(!mc->objects)) {
423 if (WARN_ON_ONCE(!capacity))
426 mc->objects = kvmalloc_array(sizeof(void *), capacity, gfp);
430 mc->capacity = capacity;
433 /* It is illegal to request a different capacity across topups. */
434 if (WARN_ON_ONCE(mc->capacity != capacity))
437 while (mc->nobjs < mc->capacity) {
438 obj = mmu_memory_cache_alloc_obj(mc, gfp);
440 return mc->nobjs >= min ? 0 : -ENOMEM;
441 mc->objects[mc->nobjs++] = obj;
446 int kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache *mc, int min)
448 return __kvm_mmu_topup_memory_cache(mc, KVM_ARCH_NR_OBJS_PER_MEMORY_CACHE, min);
451 int kvm_mmu_memory_cache_nr_free_objects(struct kvm_mmu_memory_cache *mc)
456 void kvm_mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc)
460 kmem_cache_free(mc->kmem_cache, mc->objects[--mc->nobjs]);
462 free_page((unsigned long)mc->objects[--mc->nobjs]);
471 void *kvm_mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc)
475 if (WARN_ON(!mc->nobjs))
476 p = mmu_memory_cache_alloc_obj(mc, GFP_ATOMIC | __GFP_ACCOUNT);
478 p = mc->objects[--mc->nobjs];
484 static void kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
486 mutex_init(&vcpu->mutex);
491 #ifndef __KVM_HAVE_ARCH_WQP
492 rcuwait_init(&vcpu->wait);
494 kvm_async_pf_vcpu_init(vcpu);
496 kvm_vcpu_set_in_spin_loop(vcpu, false);
497 kvm_vcpu_set_dy_eligible(vcpu, false);
498 vcpu->preempted = false;
500 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
501 vcpu->last_used_slot = NULL;
503 /* Fill the stats id string for the vcpu */
504 snprintf(vcpu->stats_id, sizeof(vcpu->stats_id), "kvm-%d/vcpu-%d",
505 task_pid_nr(current), id);
508 static void kvm_vcpu_destroy(struct kvm_vcpu *vcpu)
510 kvm_arch_vcpu_destroy(vcpu);
511 kvm_dirty_ring_free(&vcpu->dirty_ring);
514 * No need for rcu_read_lock as VCPU_RUN is the only place that changes
515 * the vcpu->pid pointer, and at destruction time all file descriptors
518 put_pid(rcu_dereference_protected(vcpu->pid, 1));
520 free_page((unsigned long)vcpu->run);
521 kmem_cache_free(kvm_vcpu_cache, vcpu);
524 void kvm_destroy_vcpus(struct kvm *kvm)
527 struct kvm_vcpu *vcpu;
529 kvm_for_each_vcpu(i, vcpu, kvm) {
530 kvm_vcpu_destroy(vcpu);
531 xa_erase(&kvm->vcpu_array, i);
534 atomic_set(&kvm->online_vcpus, 0);
536 EXPORT_SYMBOL_GPL(kvm_destroy_vcpus);
538 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
539 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
541 return container_of(mn, struct kvm, mmu_notifier);
544 typedef bool (*hva_handler_t)(struct kvm *kvm, struct kvm_gfn_range *range);
546 typedef void (*on_lock_fn_t)(struct kvm *kvm, unsigned long start,
549 typedef void (*on_unlock_fn_t)(struct kvm *kvm);
551 struct kvm_hva_range {
554 union kvm_mmu_notifier_arg arg;
555 hva_handler_t handler;
556 on_lock_fn_t on_lock;
557 on_unlock_fn_t on_unlock;
563 * Use a dedicated stub instead of NULL to indicate that there is no callback
564 * function/handler. The compiler technically can't guarantee that a real
565 * function will have a non-zero address, and so it will generate code to
566 * check for !NULL, whereas comparing against a stub will be elided at compile
567 * time (unless the compiler is getting long in the tooth, e.g. gcc 4.9).
569 static void kvm_null_fn(void)
573 #define IS_KVM_NULL_FN(fn) ((fn) == (void *)kvm_null_fn)
575 static const union kvm_mmu_notifier_arg KVM_MMU_NOTIFIER_NO_ARG;
577 /* Iterate over each memslot intersecting [start, last] (inclusive) range */
578 #define kvm_for_each_memslot_in_hva_range(node, slots, start, last) \
579 for (node = interval_tree_iter_first(&slots->hva_tree, start, last); \
581 node = interval_tree_iter_next(node, start, last)) \
583 static __always_inline int __kvm_handle_hva_range(struct kvm *kvm,
584 const struct kvm_hva_range *range)
586 bool ret = false, locked = false;
587 struct kvm_gfn_range gfn_range;
588 struct kvm_memory_slot *slot;
589 struct kvm_memslots *slots;
592 if (WARN_ON_ONCE(range->end <= range->start))
595 /* A null handler is allowed if and only if on_lock() is provided. */
596 if (WARN_ON_ONCE(IS_KVM_NULL_FN(range->on_lock) &&
597 IS_KVM_NULL_FN(range->handler)))
600 idx = srcu_read_lock(&kvm->srcu);
602 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
603 struct interval_tree_node *node;
605 slots = __kvm_memslots(kvm, i);
606 kvm_for_each_memslot_in_hva_range(node, slots,
607 range->start, range->end - 1) {
608 unsigned long hva_start, hva_end;
610 slot = container_of(node, struct kvm_memory_slot, hva_node[slots->node_idx]);
611 hva_start = max(range->start, slot->userspace_addr);
612 hva_end = min(range->end, slot->userspace_addr +
613 (slot->npages << PAGE_SHIFT));
616 * To optimize for the likely case where the address
617 * range is covered by zero or one memslots, don't
618 * bother making these conditional (to avoid writes on
619 * the second or later invocation of the handler).
621 gfn_range.arg = range->arg;
622 gfn_range.may_block = range->may_block;
625 * {gfn(page) | page intersects with [hva_start, hva_end)} =
626 * {gfn_start, gfn_start+1, ..., gfn_end-1}.
628 gfn_range.start = hva_to_gfn_memslot(hva_start, slot);
629 gfn_range.end = hva_to_gfn_memslot(hva_end + PAGE_SIZE - 1, slot);
630 gfn_range.slot = slot;
635 if (!IS_KVM_NULL_FN(range->on_lock))
636 range->on_lock(kvm, range->start, range->end);
637 if (IS_KVM_NULL_FN(range->handler))
640 ret |= range->handler(kvm, &gfn_range);
644 if (range->flush_on_ret && ret)
645 kvm_flush_remote_tlbs(kvm);
649 if (!IS_KVM_NULL_FN(range->on_unlock))
650 range->on_unlock(kvm);
653 srcu_read_unlock(&kvm->srcu, idx);
655 /* The notifiers are averse to booleans. :-( */
659 static __always_inline int kvm_handle_hva_range(struct mmu_notifier *mn,
662 union kvm_mmu_notifier_arg arg,
663 hva_handler_t handler)
665 struct kvm *kvm = mmu_notifier_to_kvm(mn);
666 const struct kvm_hva_range range = {
671 .on_lock = (void *)kvm_null_fn,
672 .on_unlock = (void *)kvm_null_fn,
673 .flush_on_ret = true,
677 return __kvm_handle_hva_range(kvm, &range);
680 static __always_inline int kvm_handle_hva_range_no_flush(struct mmu_notifier *mn,
683 hva_handler_t handler)
685 struct kvm *kvm = mmu_notifier_to_kvm(mn);
686 const struct kvm_hva_range range = {
690 .on_lock = (void *)kvm_null_fn,
691 .on_unlock = (void *)kvm_null_fn,
692 .flush_on_ret = false,
696 return __kvm_handle_hva_range(kvm, &range);
699 static bool kvm_change_spte_gfn(struct kvm *kvm, struct kvm_gfn_range *range)
702 * Skipping invalid memslots is correct if and only change_pte() is
703 * surrounded by invalidate_range_{start,end}(), which is currently
704 * guaranteed by the primary MMU. If that ever changes, KVM needs to
705 * unmap the memslot instead of skipping the memslot to ensure that KVM
706 * doesn't hold references to the old PFN.
708 WARN_ON_ONCE(!READ_ONCE(kvm->mn_active_invalidate_count));
710 if (range->slot->flags & KVM_MEMSLOT_INVALID)
713 return kvm_set_spte_gfn(kvm, range);
716 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
717 struct mm_struct *mm,
718 unsigned long address,
721 struct kvm *kvm = mmu_notifier_to_kvm(mn);
722 const union kvm_mmu_notifier_arg arg = { .pte = pte };
724 trace_kvm_set_spte_hva(address);
727 * .change_pte() must be surrounded by .invalidate_range_{start,end}().
728 * If mmu_invalidate_in_progress is zero, then no in-progress
729 * invalidations, including this one, found a relevant memslot at
730 * start(); rechecking memslots here is unnecessary. Note, a false
731 * positive (count elevated by a different invalidation) is sub-optimal
732 * but functionally ok.
734 WARN_ON_ONCE(!READ_ONCE(kvm->mn_active_invalidate_count));
735 if (!READ_ONCE(kvm->mmu_invalidate_in_progress))
738 kvm_handle_hva_range(mn, address, address + 1, arg, kvm_change_spte_gfn);
741 void kvm_mmu_invalidate_begin(struct kvm *kvm, unsigned long start,
745 * The count increase must become visible at unlock time as no
746 * spte can be established without taking the mmu_lock and
747 * count is also read inside the mmu_lock critical section.
749 kvm->mmu_invalidate_in_progress++;
750 if (likely(kvm->mmu_invalidate_in_progress == 1)) {
751 kvm->mmu_invalidate_range_start = start;
752 kvm->mmu_invalidate_range_end = end;
755 * Fully tracking multiple concurrent ranges has diminishing
756 * returns. Keep things simple and just find the minimal range
757 * which includes the current and new ranges. As there won't be
758 * enough information to subtract a range after its invalidate
759 * completes, any ranges invalidated concurrently will
760 * accumulate and persist until all outstanding invalidates
763 kvm->mmu_invalidate_range_start =
764 min(kvm->mmu_invalidate_range_start, start);
765 kvm->mmu_invalidate_range_end =
766 max(kvm->mmu_invalidate_range_end, end);
770 static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
771 const struct mmu_notifier_range *range)
773 struct kvm *kvm = mmu_notifier_to_kvm(mn);
774 const struct kvm_hva_range hva_range = {
775 .start = range->start,
777 .handler = kvm_unmap_gfn_range,
778 .on_lock = kvm_mmu_invalidate_begin,
779 .on_unlock = kvm_arch_guest_memory_reclaimed,
780 .flush_on_ret = true,
781 .may_block = mmu_notifier_range_blockable(range),
784 trace_kvm_unmap_hva_range(range->start, range->end);
787 * Prevent memslot modification between range_start() and range_end()
788 * so that conditionally locking provides the same result in both
789 * functions. Without that guarantee, the mmu_invalidate_in_progress
790 * adjustments will be imbalanced.
792 * Pairs with the decrement in range_end().
794 spin_lock(&kvm->mn_invalidate_lock);
795 kvm->mn_active_invalidate_count++;
796 spin_unlock(&kvm->mn_invalidate_lock);
799 * Invalidate pfn caches _before_ invalidating the secondary MMUs, i.e.
800 * before acquiring mmu_lock, to avoid holding mmu_lock while acquiring
801 * each cache's lock. There are relatively few caches in existence at
802 * any given time, and the caches themselves can check for hva overlap,
803 * i.e. don't need to rely on memslot overlap checks for performance.
804 * Because this runs without holding mmu_lock, the pfn caches must use
805 * mn_active_invalidate_count (see above) instead of
806 * mmu_invalidate_in_progress.
808 gfn_to_pfn_cache_invalidate_start(kvm, range->start, range->end,
809 hva_range.may_block);
811 __kvm_handle_hva_range(kvm, &hva_range);
816 void kvm_mmu_invalidate_end(struct kvm *kvm, unsigned long start,
820 * This sequence increase will notify the kvm page fault that
821 * the page that is going to be mapped in the spte could have
824 kvm->mmu_invalidate_seq++;
827 * The above sequence increase must be visible before the
828 * below count decrease, which is ensured by the smp_wmb above
829 * in conjunction with the smp_rmb in mmu_invalidate_retry().
831 kvm->mmu_invalidate_in_progress--;
834 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
835 const struct mmu_notifier_range *range)
837 struct kvm *kvm = mmu_notifier_to_kvm(mn);
838 const struct kvm_hva_range hva_range = {
839 .start = range->start,
841 .handler = (void *)kvm_null_fn,
842 .on_lock = kvm_mmu_invalidate_end,
843 .on_unlock = (void *)kvm_null_fn,
844 .flush_on_ret = false,
845 .may_block = mmu_notifier_range_blockable(range),
849 __kvm_handle_hva_range(kvm, &hva_range);
851 /* Pairs with the increment in range_start(). */
852 spin_lock(&kvm->mn_invalidate_lock);
853 wake = (--kvm->mn_active_invalidate_count == 0);
854 spin_unlock(&kvm->mn_invalidate_lock);
857 * There can only be one waiter, since the wait happens under
861 rcuwait_wake_up(&kvm->mn_memslots_update_rcuwait);
863 BUG_ON(kvm->mmu_invalidate_in_progress < 0);
866 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
867 struct mm_struct *mm,
871 trace_kvm_age_hva(start, end);
873 return kvm_handle_hva_range(mn, start, end, KVM_MMU_NOTIFIER_NO_ARG,
877 static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
878 struct mm_struct *mm,
882 trace_kvm_age_hva(start, end);
885 * Even though we do not flush TLB, this will still adversely
886 * affect performance on pre-Haswell Intel EPT, where there is
887 * no EPT Access Bit to clear so that we have to tear down EPT
888 * tables instead. If we find this unacceptable, we can always
889 * add a parameter to kvm_age_hva so that it effectively doesn't
890 * do anything on clear_young.
892 * Also note that currently we never issue secondary TLB flushes
893 * from clear_young, leaving this job up to the regular system
894 * cadence. If we find this inaccurate, we might come up with a
895 * more sophisticated heuristic later.
897 return kvm_handle_hva_range_no_flush(mn, start, end, kvm_age_gfn);
900 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
901 struct mm_struct *mm,
902 unsigned long address)
904 trace_kvm_test_age_hva(address);
906 return kvm_handle_hva_range_no_flush(mn, address, address + 1,
910 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
911 struct mm_struct *mm)
913 struct kvm *kvm = mmu_notifier_to_kvm(mn);
916 idx = srcu_read_lock(&kvm->srcu);
917 kvm_flush_shadow_all(kvm);
918 srcu_read_unlock(&kvm->srcu, idx);
921 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
922 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
923 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
924 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
925 .clear_young = kvm_mmu_notifier_clear_young,
926 .test_young = kvm_mmu_notifier_test_young,
927 .change_pte = kvm_mmu_notifier_change_pte,
928 .release = kvm_mmu_notifier_release,
931 static int kvm_init_mmu_notifier(struct kvm *kvm)
933 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
934 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
937 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
939 static int kvm_init_mmu_notifier(struct kvm *kvm)
944 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
946 #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER
947 static int kvm_pm_notifier_call(struct notifier_block *bl,
951 struct kvm *kvm = container_of(bl, struct kvm, pm_notifier);
953 return kvm_arch_pm_notifier(kvm, state);
956 static void kvm_init_pm_notifier(struct kvm *kvm)
958 kvm->pm_notifier.notifier_call = kvm_pm_notifier_call;
959 /* Suspend KVM before we suspend ftrace, RCU, etc. */
960 kvm->pm_notifier.priority = INT_MAX;
961 register_pm_notifier(&kvm->pm_notifier);
964 static void kvm_destroy_pm_notifier(struct kvm *kvm)
966 unregister_pm_notifier(&kvm->pm_notifier);
968 #else /* !CONFIG_HAVE_KVM_PM_NOTIFIER */
969 static void kvm_init_pm_notifier(struct kvm *kvm)
973 static void kvm_destroy_pm_notifier(struct kvm *kvm)
976 #endif /* CONFIG_HAVE_KVM_PM_NOTIFIER */
978 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
980 if (!memslot->dirty_bitmap)
983 kvfree(memslot->dirty_bitmap);
984 memslot->dirty_bitmap = NULL;
987 /* This does not remove the slot from struct kvm_memslots data structures */
988 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
990 kvm_destroy_dirty_bitmap(slot);
992 kvm_arch_free_memslot(kvm, slot);
997 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
999 struct hlist_node *idnode;
1000 struct kvm_memory_slot *memslot;
1004 * The same memslot objects live in both active and inactive sets,
1005 * arbitrarily free using index '1' so the second invocation of this
1006 * function isn't operating over a structure with dangling pointers
1007 * (even though this function isn't actually touching them).
1009 if (!slots->node_idx)
1012 hash_for_each_safe(slots->id_hash, bkt, idnode, memslot, id_node[1])
1013 kvm_free_memslot(kvm, memslot);
1016 static umode_t kvm_stats_debugfs_mode(const struct _kvm_stats_desc *pdesc)
1018 switch (pdesc->desc.flags & KVM_STATS_TYPE_MASK) {
1019 case KVM_STATS_TYPE_INSTANT:
1021 case KVM_STATS_TYPE_CUMULATIVE:
1022 case KVM_STATS_TYPE_PEAK:
1029 static void kvm_destroy_vm_debugfs(struct kvm *kvm)
1032 int kvm_debugfs_num_entries = kvm_vm_stats_header.num_desc +
1033 kvm_vcpu_stats_header.num_desc;
1035 if (IS_ERR(kvm->debugfs_dentry))
1038 debugfs_remove_recursive(kvm->debugfs_dentry);
1040 if (kvm->debugfs_stat_data) {
1041 for (i = 0; i < kvm_debugfs_num_entries; i++)
1042 kfree(kvm->debugfs_stat_data[i]);
1043 kfree(kvm->debugfs_stat_data);
1047 static int kvm_create_vm_debugfs(struct kvm *kvm, const char *fdname)
1049 static DEFINE_MUTEX(kvm_debugfs_lock);
1050 struct dentry *dent;
1051 char dir_name[ITOA_MAX_LEN * 2];
1052 struct kvm_stat_data *stat_data;
1053 const struct _kvm_stats_desc *pdesc;
1054 int i, ret = -ENOMEM;
1055 int kvm_debugfs_num_entries = kvm_vm_stats_header.num_desc +
1056 kvm_vcpu_stats_header.num_desc;
1058 if (!debugfs_initialized())
1061 snprintf(dir_name, sizeof(dir_name), "%d-%s", task_pid_nr(current), fdname);
1062 mutex_lock(&kvm_debugfs_lock);
1063 dent = debugfs_lookup(dir_name, kvm_debugfs_dir);
1065 pr_warn_ratelimited("KVM: debugfs: duplicate directory %s\n", dir_name);
1067 mutex_unlock(&kvm_debugfs_lock);
1070 dent = debugfs_create_dir(dir_name, kvm_debugfs_dir);
1071 mutex_unlock(&kvm_debugfs_lock);
1075 kvm->debugfs_dentry = dent;
1076 kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries,
1077 sizeof(*kvm->debugfs_stat_data),
1078 GFP_KERNEL_ACCOUNT);
1079 if (!kvm->debugfs_stat_data)
1082 for (i = 0; i < kvm_vm_stats_header.num_desc; ++i) {
1083 pdesc = &kvm_vm_stats_desc[i];
1084 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT);
1088 stat_data->kvm = kvm;
1089 stat_data->desc = pdesc;
1090 stat_data->kind = KVM_STAT_VM;
1091 kvm->debugfs_stat_data[i] = stat_data;
1092 debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
1093 kvm->debugfs_dentry, stat_data,
1097 for (i = 0; i < kvm_vcpu_stats_header.num_desc; ++i) {
1098 pdesc = &kvm_vcpu_stats_desc[i];
1099 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT);
1103 stat_data->kvm = kvm;
1104 stat_data->desc = pdesc;
1105 stat_data->kind = KVM_STAT_VCPU;
1106 kvm->debugfs_stat_data[i + kvm_vm_stats_header.num_desc] = stat_data;
1107 debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
1108 kvm->debugfs_dentry, stat_data,
1112 ret = kvm_arch_create_vm_debugfs(kvm);
1118 kvm_destroy_vm_debugfs(kvm);
1123 * Called after the VM is otherwise initialized, but just before adding it to
1126 int __weak kvm_arch_post_init_vm(struct kvm *kvm)
1132 * Called just after removing the VM from the vm_list, but before doing any
1133 * other destruction.
1135 void __weak kvm_arch_pre_destroy_vm(struct kvm *kvm)
1140 * Called after per-vm debugfs created. When called kvm->debugfs_dentry should
1141 * be setup already, so we can create arch-specific debugfs entries under it.
1142 * Cleanup should be automatic done in kvm_destroy_vm_debugfs() recursively, so
1143 * a per-arch destroy interface is not needed.
1145 int __weak kvm_arch_create_vm_debugfs(struct kvm *kvm)
1150 static struct kvm *kvm_create_vm(unsigned long type, const char *fdname)
1152 struct kvm *kvm = kvm_arch_alloc_vm();
1153 struct kvm_memslots *slots;
1158 return ERR_PTR(-ENOMEM);
1160 /* KVM is pinned via open("/dev/kvm"), the fd passed to this ioctl(). */
1161 __module_get(kvm_chardev_ops.owner);
1163 KVM_MMU_LOCK_INIT(kvm);
1164 mmgrab(current->mm);
1165 kvm->mm = current->mm;
1166 kvm_eventfd_init(kvm);
1167 mutex_init(&kvm->lock);
1168 mutex_init(&kvm->irq_lock);
1169 mutex_init(&kvm->slots_lock);
1170 mutex_init(&kvm->slots_arch_lock);
1171 spin_lock_init(&kvm->mn_invalidate_lock);
1172 rcuwait_init(&kvm->mn_memslots_update_rcuwait);
1173 xa_init(&kvm->vcpu_array);
1175 INIT_LIST_HEAD(&kvm->gpc_list);
1176 spin_lock_init(&kvm->gpc_lock);
1178 INIT_LIST_HEAD(&kvm->devices);
1179 kvm->max_vcpus = KVM_MAX_VCPUS;
1181 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
1184 * Force subsequent debugfs file creations to fail if the VM directory
1185 * is not created (by kvm_create_vm_debugfs()).
1187 kvm->debugfs_dentry = ERR_PTR(-ENOENT);
1189 snprintf(kvm->stats_id, sizeof(kvm->stats_id), "kvm-%d",
1190 task_pid_nr(current));
1192 if (init_srcu_struct(&kvm->srcu))
1193 goto out_err_no_srcu;
1194 if (init_srcu_struct(&kvm->irq_srcu))
1195 goto out_err_no_irq_srcu;
1197 refcount_set(&kvm->users_count, 1);
1198 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
1199 for (j = 0; j < 2; j++) {
1200 slots = &kvm->__memslots[i][j];
1202 atomic_long_set(&slots->last_used_slot, (unsigned long)NULL);
1203 slots->hva_tree = RB_ROOT_CACHED;
1204 slots->gfn_tree = RB_ROOT;
1205 hash_init(slots->id_hash);
1206 slots->node_idx = j;
1208 /* Generations must be different for each address space. */
1209 slots->generation = i;
1212 rcu_assign_pointer(kvm->memslots[i], &kvm->__memslots[i][0]);
1215 for (i = 0; i < KVM_NR_BUSES; i++) {
1216 rcu_assign_pointer(kvm->buses[i],
1217 kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL_ACCOUNT));
1219 goto out_err_no_arch_destroy_vm;
1222 r = kvm_arch_init_vm(kvm, type);
1224 goto out_err_no_arch_destroy_vm;
1226 r = hardware_enable_all();
1228 goto out_err_no_disable;
1230 #ifdef CONFIG_HAVE_KVM_IRQFD
1231 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
1234 r = kvm_init_mmu_notifier(kvm);
1236 goto out_err_no_mmu_notifier;
1238 r = kvm_coalesced_mmio_init(kvm);
1240 goto out_no_coalesced_mmio;
1242 r = kvm_create_vm_debugfs(kvm, fdname);
1244 goto out_err_no_debugfs;
1246 r = kvm_arch_post_init_vm(kvm);
1250 mutex_lock(&kvm_lock);
1251 list_add(&kvm->vm_list, &vm_list);
1252 mutex_unlock(&kvm_lock);
1254 preempt_notifier_inc();
1255 kvm_init_pm_notifier(kvm);
1260 kvm_destroy_vm_debugfs(kvm);
1262 kvm_coalesced_mmio_free(kvm);
1263 out_no_coalesced_mmio:
1264 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1265 if (kvm->mmu_notifier.ops)
1266 mmu_notifier_unregister(&kvm->mmu_notifier, current->mm);
1268 out_err_no_mmu_notifier:
1269 hardware_disable_all();
1271 kvm_arch_destroy_vm(kvm);
1272 out_err_no_arch_destroy_vm:
1273 WARN_ON_ONCE(!refcount_dec_and_test(&kvm->users_count));
1274 for (i = 0; i < KVM_NR_BUSES; i++)
1275 kfree(kvm_get_bus(kvm, i));
1276 cleanup_srcu_struct(&kvm->irq_srcu);
1277 out_err_no_irq_srcu:
1278 cleanup_srcu_struct(&kvm->srcu);
1280 kvm_arch_free_vm(kvm);
1281 mmdrop(current->mm);
1282 module_put(kvm_chardev_ops.owner);
1286 static void kvm_destroy_devices(struct kvm *kvm)
1288 struct kvm_device *dev, *tmp;
1291 * We do not need to take the kvm->lock here, because nobody else
1292 * has a reference to the struct kvm at this point and therefore
1293 * cannot access the devices list anyhow.
1295 list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
1296 list_del(&dev->vm_node);
1297 dev->ops->destroy(dev);
1301 static void kvm_destroy_vm(struct kvm *kvm)
1304 struct mm_struct *mm = kvm->mm;
1306 kvm_destroy_pm_notifier(kvm);
1307 kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm);
1308 kvm_destroy_vm_debugfs(kvm);
1309 kvm_arch_sync_events(kvm);
1310 mutex_lock(&kvm_lock);
1311 list_del(&kvm->vm_list);
1312 mutex_unlock(&kvm_lock);
1313 kvm_arch_pre_destroy_vm(kvm);
1315 kvm_free_irq_routing(kvm);
1316 for (i = 0; i < KVM_NR_BUSES; i++) {
1317 struct kvm_io_bus *bus = kvm_get_bus(kvm, i);
1320 kvm_io_bus_destroy(bus);
1321 kvm->buses[i] = NULL;
1323 kvm_coalesced_mmio_free(kvm);
1324 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1325 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
1327 * At this point, pending calls to invalidate_range_start()
1328 * have completed but no more MMU notifiers will run, so
1329 * mn_active_invalidate_count may remain unbalanced.
1330 * No threads can be waiting in kvm_swap_active_memslots() as the
1331 * last reference on KVM has been dropped, but freeing
1332 * memslots would deadlock without this manual intervention.
1334 WARN_ON(rcuwait_active(&kvm->mn_memslots_update_rcuwait));
1335 kvm->mn_active_invalidate_count = 0;
1337 kvm_flush_shadow_all(kvm);
1339 kvm_arch_destroy_vm(kvm);
1340 kvm_destroy_devices(kvm);
1341 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
1342 kvm_free_memslots(kvm, &kvm->__memslots[i][0]);
1343 kvm_free_memslots(kvm, &kvm->__memslots[i][1]);
1345 cleanup_srcu_struct(&kvm->irq_srcu);
1346 cleanup_srcu_struct(&kvm->srcu);
1347 kvm_arch_free_vm(kvm);
1348 preempt_notifier_dec();
1349 hardware_disable_all();
1351 module_put(kvm_chardev_ops.owner);
1354 void kvm_get_kvm(struct kvm *kvm)
1356 refcount_inc(&kvm->users_count);
1358 EXPORT_SYMBOL_GPL(kvm_get_kvm);
1361 * Make sure the vm is not during destruction, which is a safe version of
1362 * kvm_get_kvm(). Return true if kvm referenced successfully, false otherwise.
1364 bool kvm_get_kvm_safe(struct kvm *kvm)
1366 return refcount_inc_not_zero(&kvm->users_count);
1368 EXPORT_SYMBOL_GPL(kvm_get_kvm_safe);
1370 void kvm_put_kvm(struct kvm *kvm)
1372 if (refcount_dec_and_test(&kvm->users_count))
1373 kvm_destroy_vm(kvm);
1375 EXPORT_SYMBOL_GPL(kvm_put_kvm);
1378 * Used to put a reference that was taken on behalf of an object associated
1379 * with a user-visible file descriptor, e.g. a vcpu or device, if installation
1380 * of the new file descriptor fails and the reference cannot be transferred to
1381 * its final owner. In such cases, the caller is still actively using @kvm and
1382 * will fail miserably if the refcount unexpectedly hits zero.
1384 void kvm_put_kvm_no_destroy(struct kvm *kvm)
1386 WARN_ON(refcount_dec_and_test(&kvm->users_count));
1388 EXPORT_SYMBOL_GPL(kvm_put_kvm_no_destroy);
1390 static int kvm_vm_release(struct inode *inode, struct file *filp)
1392 struct kvm *kvm = filp->private_data;
1394 kvm_irqfd_release(kvm);
1401 * Allocation size is twice as large as the actual dirty bitmap size.
1402 * See kvm_vm_ioctl_get_dirty_log() why this is needed.
1404 static int kvm_alloc_dirty_bitmap(struct kvm_memory_slot *memslot)
1406 unsigned long dirty_bytes = kvm_dirty_bitmap_bytes(memslot);
1408 memslot->dirty_bitmap = __vcalloc(2, dirty_bytes, GFP_KERNEL_ACCOUNT);
1409 if (!memslot->dirty_bitmap)
1415 static struct kvm_memslots *kvm_get_inactive_memslots(struct kvm *kvm, int as_id)
1417 struct kvm_memslots *active = __kvm_memslots(kvm, as_id);
1418 int node_idx_inactive = active->node_idx ^ 1;
1420 return &kvm->__memslots[as_id][node_idx_inactive];
1424 * Helper to get the address space ID when one of memslot pointers may be NULL.
1425 * This also serves as a sanity that at least one of the pointers is non-NULL,
1426 * and that their address space IDs don't diverge.
1428 static int kvm_memslots_get_as_id(struct kvm_memory_slot *a,
1429 struct kvm_memory_slot *b)
1431 if (WARN_ON_ONCE(!a && !b))
1439 WARN_ON_ONCE(a->as_id != b->as_id);
1443 static void kvm_insert_gfn_node(struct kvm_memslots *slots,
1444 struct kvm_memory_slot *slot)
1446 struct rb_root *gfn_tree = &slots->gfn_tree;
1447 struct rb_node **node, *parent;
1448 int idx = slots->node_idx;
1451 for (node = &gfn_tree->rb_node; *node; ) {
1452 struct kvm_memory_slot *tmp;
1454 tmp = container_of(*node, struct kvm_memory_slot, gfn_node[idx]);
1456 if (slot->base_gfn < tmp->base_gfn)
1457 node = &(*node)->rb_left;
1458 else if (slot->base_gfn > tmp->base_gfn)
1459 node = &(*node)->rb_right;
1464 rb_link_node(&slot->gfn_node[idx], parent, node);
1465 rb_insert_color(&slot->gfn_node[idx], gfn_tree);
1468 static void kvm_erase_gfn_node(struct kvm_memslots *slots,
1469 struct kvm_memory_slot *slot)
1471 rb_erase(&slot->gfn_node[slots->node_idx], &slots->gfn_tree);
1474 static void kvm_replace_gfn_node(struct kvm_memslots *slots,
1475 struct kvm_memory_slot *old,
1476 struct kvm_memory_slot *new)
1478 int idx = slots->node_idx;
1480 WARN_ON_ONCE(old->base_gfn != new->base_gfn);
1482 rb_replace_node(&old->gfn_node[idx], &new->gfn_node[idx],
1487 * Replace @old with @new in the inactive memslots.
1489 * With NULL @old this simply adds @new.
1490 * With NULL @new this simply removes @old.
1492 * If @new is non-NULL its hva_node[slots_idx] range has to be set
1495 static void kvm_replace_memslot(struct kvm *kvm,
1496 struct kvm_memory_slot *old,
1497 struct kvm_memory_slot *new)
1499 int as_id = kvm_memslots_get_as_id(old, new);
1500 struct kvm_memslots *slots = kvm_get_inactive_memslots(kvm, as_id);
1501 int idx = slots->node_idx;
1504 hash_del(&old->id_node[idx]);
1505 interval_tree_remove(&old->hva_node[idx], &slots->hva_tree);
1507 if ((long)old == atomic_long_read(&slots->last_used_slot))
1508 atomic_long_set(&slots->last_used_slot, (long)new);
1511 kvm_erase_gfn_node(slots, old);
1517 * Initialize @new's hva range. Do this even when replacing an @old
1518 * slot, kvm_copy_memslot() deliberately does not touch node data.
1520 new->hva_node[idx].start = new->userspace_addr;
1521 new->hva_node[idx].last = new->userspace_addr +
1522 (new->npages << PAGE_SHIFT) - 1;
1525 * (Re)Add the new memslot. There is no O(1) interval_tree_replace(),
1526 * hva_node needs to be swapped with remove+insert even though hva can't
1527 * change when replacing an existing slot.
1529 hash_add(slots->id_hash, &new->id_node[idx], new->id);
1530 interval_tree_insert(&new->hva_node[idx], &slots->hva_tree);
1533 * If the memslot gfn is unchanged, rb_replace_node() can be used to
1534 * switch the node in the gfn tree instead of removing the old and
1535 * inserting the new as two separate operations. Replacement is a
1536 * single O(1) operation versus two O(log(n)) operations for
1539 if (old && old->base_gfn == new->base_gfn) {
1540 kvm_replace_gfn_node(slots, old, new);
1543 kvm_erase_gfn_node(slots, old);
1544 kvm_insert_gfn_node(slots, new);
1548 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
1550 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
1552 #ifdef __KVM_HAVE_READONLY_MEM
1553 valid_flags |= KVM_MEM_READONLY;
1556 if (mem->flags & ~valid_flags)
1562 static void kvm_swap_active_memslots(struct kvm *kvm, int as_id)
1564 struct kvm_memslots *slots = kvm_get_inactive_memslots(kvm, as_id);
1566 /* Grab the generation from the activate memslots. */
1567 u64 gen = __kvm_memslots(kvm, as_id)->generation;
1569 WARN_ON(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS);
1570 slots->generation = gen | KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
1573 * Do not store the new memslots while there are invalidations in
1574 * progress, otherwise the locking in invalidate_range_start and
1575 * invalidate_range_end will be unbalanced.
1577 spin_lock(&kvm->mn_invalidate_lock);
1578 prepare_to_rcuwait(&kvm->mn_memslots_update_rcuwait);
1579 while (kvm->mn_active_invalidate_count) {
1580 set_current_state(TASK_UNINTERRUPTIBLE);
1581 spin_unlock(&kvm->mn_invalidate_lock);
1583 spin_lock(&kvm->mn_invalidate_lock);
1585 finish_rcuwait(&kvm->mn_memslots_update_rcuwait);
1586 rcu_assign_pointer(kvm->memslots[as_id], slots);
1587 spin_unlock(&kvm->mn_invalidate_lock);
1590 * Acquired in kvm_set_memslot. Must be released before synchronize
1591 * SRCU below in order to avoid deadlock with another thread
1592 * acquiring the slots_arch_lock in an srcu critical section.
1594 mutex_unlock(&kvm->slots_arch_lock);
1596 synchronize_srcu_expedited(&kvm->srcu);
1599 * Increment the new memslot generation a second time, dropping the
1600 * update in-progress flag and incrementing the generation based on
1601 * the number of address spaces. This provides a unique and easily
1602 * identifiable generation number while the memslots are in flux.
1604 gen = slots->generation & ~KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
1607 * Generations must be unique even across address spaces. We do not need
1608 * a global counter for that, instead the generation space is evenly split
1609 * across address spaces. For example, with two address spaces, address
1610 * space 0 will use generations 0, 2, 4, ... while address space 1 will
1611 * use generations 1, 3, 5, ...
1613 gen += KVM_ADDRESS_SPACE_NUM;
1615 kvm_arch_memslots_updated(kvm, gen);
1617 slots->generation = gen;
1620 static int kvm_prepare_memory_region(struct kvm *kvm,
1621 const struct kvm_memory_slot *old,
1622 struct kvm_memory_slot *new,
1623 enum kvm_mr_change change)
1628 * If dirty logging is disabled, nullify the bitmap; the old bitmap
1629 * will be freed on "commit". If logging is enabled in both old and
1630 * new, reuse the existing bitmap. If logging is enabled only in the
1631 * new and KVM isn't using a ring buffer, allocate and initialize a
1634 if (change != KVM_MR_DELETE) {
1635 if (!(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
1636 new->dirty_bitmap = NULL;
1637 else if (old && old->dirty_bitmap)
1638 new->dirty_bitmap = old->dirty_bitmap;
1639 else if (kvm_use_dirty_bitmap(kvm)) {
1640 r = kvm_alloc_dirty_bitmap(new);
1644 if (kvm_dirty_log_manual_protect_and_init_set(kvm))
1645 bitmap_set(new->dirty_bitmap, 0, new->npages);
1649 r = kvm_arch_prepare_memory_region(kvm, old, new, change);
1651 /* Free the bitmap on failure if it was allocated above. */
1652 if (r && new && new->dirty_bitmap && (!old || !old->dirty_bitmap))
1653 kvm_destroy_dirty_bitmap(new);
1658 static void kvm_commit_memory_region(struct kvm *kvm,
1659 struct kvm_memory_slot *old,
1660 const struct kvm_memory_slot *new,
1661 enum kvm_mr_change change)
1663 int old_flags = old ? old->flags : 0;
1664 int new_flags = new ? new->flags : 0;
1666 * Update the total number of memslot pages before calling the arch
1667 * hook so that architectures can consume the result directly.
1669 if (change == KVM_MR_DELETE)
1670 kvm->nr_memslot_pages -= old->npages;
1671 else if (change == KVM_MR_CREATE)
1672 kvm->nr_memslot_pages += new->npages;
1674 if ((old_flags ^ new_flags) & KVM_MEM_LOG_DIRTY_PAGES) {
1675 int change = (new_flags & KVM_MEM_LOG_DIRTY_PAGES) ? 1 : -1;
1676 atomic_set(&kvm->nr_memslots_dirty_logging,
1677 atomic_read(&kvm->nr_memslots_dirty_logging) + change);
1680 kvm_arch_commit_memory_region(kvm, old, new, change);
1684 /* Nothing more to do. */
1687 /* Free the old memslot and all its metadata. */
1688 kvm_free_memslot(kvm, old);
1691 case KVM_MR_FLAGS_ONLY:
1693 * Free the dirty bitmap as needed; the below check encompasses
1694 * both the flags and whether a ring buffer is being used)
1696 if (old->dirty_bitmap && !new->dirty_bitmap)
1697 kvm_destroy_dirty_bitmap(old);
1700 * The final quirk. Free the detached, old slot, but only its
1701 * memory, not any metadata. Metadata, including arch specific
1702 * data, may be reused by @new.
1712 * Activate @new, which must be installed in the inactive slots by the caller,
1713 * by swapping the active slots and then propagating @new to @old once @old is
1714 * unreachable and can be safely modified.
1716 * With NULL @old this simply adds @new to @active (while swapping the sets).
1717 * With NULL @new this simply removes @old from @active and frees it
1718 * (while also swapping the sets).
1720 static void kvm_activate_memslot(struct kvm *kvm,
1721 struct kvm_memory_slot *old,
1722 struct kvm_memory_slot *new)
1724 int as_id = kvm_memslots_get_as_id(old, new);
1726 kvm_swap_active_memslots(kvm, as_id);
1728 /* Propagate the new memslot to the now inactive memslots. */
1729 kvm_replace_memslot(kvm, old, new);
1732 static void kvm_copy_memslot(struct kvm_memory_slot *dest,
1733 const struct kvm_memory_slot *src)
1735 dest->base_gfn = src->base_gfn;
1736 dest->npages = src->npages;
1737 dest->dirty_bitmap = src->dirty_bitmap;
1738 dest->arch = src->arch;
1739 dest->userspace_addr = src->userspace_addr;
1740 dest->flags = src->flags;
1742 dest->as_id = src->as_id;
1745 static void kvm_invalidate_memslot(struct kvm *kvm,
1746 struct kvm_memory_slot *old,
1747 struct kvm_memory_slot *invalid_slot)
1750 * Mark the current slot INVALID. As with all memslot modifications,
1751 * this must be done on an unreachable slot to avoid modifying the
1752 * current slot in the active tree.
1754 kvm_copy_memslot(invalid_slot, old);
1755 invalid_slot->flags |= KVM_MEMSLOT_INVALID;
1756 kvm_replace_memslot(kvm, old, invalid_slot);
1759 * Activate the slot that is now marked INVALID, but don't propagate
1760 * the slot to the now inactive slots. The slot is either going to be
1761 * deleted or recreated as a new slot.
1763 kvm_swap_active_memslots(kvm, old->as_id);
1766 * From this point no new shadow pages pointing to a deleted, or moved,
1767 * memslot will be created. Validation of sp->gfn happens in:
1768 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1769 * - kvm_is_visible_gfn (mmu_check_root)
1771 kvm_arch_flush_shadow_memslot(kvm, old);
1772 kvm_arch_guest_memory_reclaimed(kvm);
1774 /* Was released by kvm_swap_active_memslots(), reacquire. */
1775 mutex_lock(&kvm->slots_arch_lock);
1778 * Copy the arch-specific field of the newly-installed slot back to the
1779 * old slot as the arch data could have changed between releasing
1780 * slots_arch_lock in kvm_swap_active_memslots() and re-acquiring the lock
1781 * above. Writers are required to retrieve memslots *after* acquiring
1782 * slots_arch_lock, thus the active slot's data is guaranteed to be fresh.
1784 old->arch = invalid_slot->arch;
1787 static void kvm_create_memslot(struct kvm *kvm,
1788 struct kvm_memory_slot *new)
1790 /* Add the new memslot to the inactive set and activate. */
1791 kvm_replace_memslot(kvm, NULL, new);
1792 kvm_activate_memslot(kvm, NULL, new);
1795 static void kvm_delete_memslot(struct kvm *kvm,
1796 struct kvm_memory_slot *old,
1797 struct kvm_memory_slot *invalid_slot)
1800 * Remove the old memslot (in the inactive memslots) by passing NULL as
1801 * the "new" slot, and for the invalid version in the active slots.
1803 kvm_replace_memslot(kvm, old, NULL);
1804 kvm_activate_memslot(kvm, invalid_slot, NULL);
1807 static void kvm_move_memslot(struct kvm *kvm,
1808 struct kvm_memory_slot *old,
1809 struct kvm_memory_slot *new,
1810 struct kvm_memory_slot *invalid_slot)
1813 * Replace the old memslot in the inactive slots, and then swap slots
1814 * and replace the current INVALID with the new as well.
1816 kvm_replace_memslot(kvm, old, new);
1817 kvm_activate_memslot(kvm, invalid_slot, new);
1820 static void kvm_update_flags_memslot(struct kvm *kvm,
1821 struct kvm_memory_slot *old,
1822 struct kvm_memory_slot *new)
1825 * Similar to the MOVE case, but the slot doesn't need to be zapped as
1826 * an intermediate step. Instead, the old memslot is simply replaced
1827 * with a new, updated copy in both memslot sets.
1829 kvm_replace_memslot(kvm, old, new);
1830 kvm_activate_memslot(kvm, old, new);
1833 static int kvm_set_memslot(struct kvm *kvm,
1834 struct kvm_memory_slot *old,
1835 struct kvm_memory_slot *new,
1836 enum kvm_mr_change change)
1838 struct kvm_memory_slot *invalid_slot;
1842 * Released in kvm_swap_active_memslots().
1844 * Must be held from before the current memslots are copied until after
1845 * the new memslots are installed with rcu_assign_pointer, then
1846 * released before the synchronize srcu in kvm_swap_active_memslots().
1848 * When modifying memslots outside of the slots_lock, must be held
1849 * before reading the pointer to the current memslots until after all
1850 * changes to those memslots are complete.
1852 * These rules ensure that installing new memslots does not lose
1853 * changes made to the previous memslots.
1855 mutex_lock(&kvm->slots_arch_lock);
1858 * Invalidate the old slot if it's being deleted or moved. This is
1859 * done prior to actually deleting/moving the memslot to allow vCPUs to
1860 * continue running by ensuring there are no mappings or shadow pages
1861 * for the memslot when it is deleted/moved. Without pre-invalidation
1862 * (and without a lock), a window would exist between effecting the
1863 * delete/move and committing the changes in arch code where KVM or a
1864 * guest could access a non-existent memslot.
1866 * Modifications are done on a temporary, unreachable slot. The old
1867 * slot needs to be preserved in case a later step fails and the
1868 * invalidation needs to be reverted.
1870 if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) {
1871 invalid_slot = kzalloc(sizeof(*invalid_slot), GFP_KERNEL_ACCOUNT);
1872 if (!invalid_slot) {
1873 mutex_unlock(&kvm->slots_arch_lock);
1876 kvm_invalidate_memslot(kvm, old, invalid_slot);
1879 r = kvm_prepare_memory_region(kvm, old, new, change);
1882 * For DELETE/MOVE, revert the above INVALID change. No
1883 * modifications required since the original slot was preserved
1884 * in the inactive slots. Changing the active memslots also
1885 * release slots_arch_lock.
1887 if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) {
1888 kvm_activate_memslot(kvm, invalid_slot, old);
1889 kfree(invalid_slot);
1891 mutex_unlock(&kvm->slots_arch_lock);
1897 * For DELETE and MOVE, the working slot is now active as the INVALID
1898 * version of the old slot. MOVE is particularly special as it reuses
1899 * the old slot and returns a copy of the old slot (in working_slot).
1900 * For CREATE, there is no old slot. For DELETE and FLAGS_ONLY, the
1901 * old slot is detached but otherwise preserved.
1903 if (change == KVM_MR_CREATE)
1904 kvm_create_memslot(kvm, new);
1905 else if (change == KVM_MR_DELETE)
1906 kvm_delete_memslot(kvm, old, invalid_slot);
1907 else if (change == KVM_MR_MOVE)
1908 kvm_move_memslot(kvm, old, new, invalid_slot);
1909 else if (change == KVM_MR_FLAGS_ONLY)
1910 kvm_update_flags_memslot(kvm, old, new);
1914 /* Free the temporary INVALID slot used for DELETE and MOVE. */
1915 if (change == KVM_MR_DELETE || change == KVM_MR_MOVE)
1916 kfree(invalid_slot);
1919 * No need to refresh new->arch, changes after dropping slots_arch_lock
1920 * will directly hit the final, active memslot. Architectures are
1921 * responsible for knowing that new->arch may be stale.
1923 kvm_commit_memory_region(kvm, old, new, change);
1928 static bool kvm_check_memslot_overlap(struct kvm_memslots *slots, int id,
1929 gfn_t start, gfn_t end)
1931 struct kvm_memslot_iter iter;
1933 kvm_for_each_memslot_in_gfn_range(&iter, slots, start, end) {
1934 if (iter.slot->id != id)
1942 * Allocate some memory and give it an address in the guest physical address
1945 * Discontiguous memory is allowed, mostly for framebuffers.
1947 * Must be called holding kvm->slots_lock for write.
1949 int __kvm_set_memory_region(struct kvm *kvm,
1950 const struct kvm_userspace_memory_region *mem)
1952 struct kvm_memory_slot *old, *new;
1953 struct kvm_memslots *slots;
1954 enum kvm_mr_change change;
1955 unsigned long npages;
1960 r = check_memory_region_flags(mem);
1964 as_id = mem->slot >> 16;
1965 id = (u16)mem->slot;
1967 /* General sanity checks */
1968 if ((mem->memory_size & (PAGE_SIZE - 1)) ||
1969 (mem->memory_size != (unsigned long)mem->memory_size))
1971 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
1973 /* We can read the guest memory with __xxx_user() later on. */
1974 if ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
1975 (mem->userspace_addr != untagged_addr(mem->userspace_addr)) ||
1976 !access_ok((void __user *)(unsigned long)mem->userspace_addr,
1979 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
1981 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
1983 if ((mem->memory_size >> PAGE_SHIFT) > KVM_MEM_MAX_NR_PAGES)
1986 slots = __kvm_memslots(kvm, as_id);
1989 * Note, the old memslot (and the pointer itself!) may be invalidated
1990 * and/or destroyed by kvm_set_memslot().
1992 old = id_to_memslot(slots, id);
1994 if (!mem->memory_size) {
1995 if (!old || !old->npages)
1998 if (WARN_ON_ONCE(kvm->nr_memslot_pages < old->npages))
2001 return kvm_set_memslot(kvm, old, NULL, KVM_MR_DELETE);
2004 base_gfn = (mem->guest_phys_addr >> PAGE_SHIFT);
2005 npages = (mem->memory_size >> PAGE_SHIFT);
2007 if (!old || !old->npages) {
2008 change = KVM_MR_CREATE;
2011 * To simplify KVM internals, the total number of pages across
2012 * all memslots must fit in an unsigned long.
2014 if ((kvm->nr_memslot_pages + npages) < kvm->nr_memslot_pages)
2016 } else { /* Modify an existing slot. */
2017 if ((mem->userspace_addr != old->userspace_addr) ||
2018 (npages != old->npages) ||
2019 ((mem->flags ^ old->flags) & KVM_MEM_READONLY))
2022 if (base_gfn != old->base_gfn)
2023 change = KVM_MR_MOVE;
2024 else if (mem->flags != old->flags)
2025 change = KVM_MR_FLAGS_ONLY;
2026 else /* Nothing to change. */
2030 if ((change == KVM_MR_CREATE || change == KVM_MR_MOVE) &&
2031 kvm_check_memslot_overlap(slots, id, base_gfn, base_gfn + npages))
2034 /* Allocate a slot that will persist in the memslot. */
2035 new = kzalloc(sizeof(*new), GFP_KERNEL_ACCOUNT);
2041 new->base_gfn = base_gfn;
2042 new->npages = npages;
2043 new->flags = mem->flags;
2044 new->userspace_addr = mem->userspace_addr;
2046 r = kvm_set_memslot(kvm, old, new, change);
2051 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
2053 int kvm_set_memory_region(struct kvm *kvm,
2054 const struct kvm_userspace_memory_region *mem)
2058 mutex_lock(&kvm->slots_lock);
2059 r = __kvm_set_memory_region(kvm, mem);
2060 mutex_unlock(&kvm->slots_lock);
2063 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
2065 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
2066 struct kvm_userspace_memory_region *mem)
2068 if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
2071 return kvm_set_memory_region(kvm, mem);
2074 #ifndef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
2076 * kvm_get_dirty_log - get a snapshot of dirty pages
2077 * @kvm: pointer to kvm instance
2078 * @log: slot id and address to which we copy the log
2079 * @is_dirty: set to '1' if any dirty pages were found
2080 * @memslot: set to the associated memslot, always valid on success
2082 int kvm_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log,
2083 int *is_dirty, struct kvm_memory_slot **memslot)
2085 struct kvm_memslots *slots;
2088 unsigned long any = 0;
2090 /* Dirty ring tracking may be exclusive to dirty log tracking */
2091 if (!kvm_use_dirty_bitmap(kvm))
2097 as_id = log->slot >> 16;
2098 id = (u16)log->slot;
2099 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
2102 slots = __kvm_memslots(kvm, as_id);
2103 *memslot = id_to_memslot(slots, id);
2104 if (!(*memslot) || !(*memslot)->dirty_bitmap)
2107 kvm_arch_sync_dirty_log(kvm, *memslot);
2109 n = kvm_dirty_bitmap_bytes(*memslot);
2111 for (i = 0; !any && i < n/sizeof(long); ++i)
2112 any = (*memslot)->dirty_bitmap[i];
2114 if (copy_to_user(log->dirty_bitmap, (*memslot)->dirty_bitmap, n))
2121 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
2123 #else /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
2125 * kvm_get_dirty_log_protect - get a snapshot of dirty pages
2126 * and reenable dirty page tracking for the corresponding pages.
2127 * @kvm: pointer to kvm instance
2128 * @log: slot id and address to which we copy the log
2130 * We need to keep it in mind that VCPU threads can write to the bitmap
2131 * concurrently. So, to avoid losing track of dirty pages we keep the
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. Upon return caller flushes TLB's if needed.
2139 * Between 2 and 4, the guest may write to the page using the remaining TLB
2140 * entry. This is not a problem because the page is reported dirty using
2141 * the snapshot taken before and step 4 ensures that writes done after
2142 * exiting to userspace will be logged for the next call.
2145 static int kvm_get_dirty_log_protect(struct kvm *kvm, struct kvm_dirty_log *log)
2147 struct kvm_memslots *slots;
2148 struct kvm_memory_slot *memslot;
2151 unsigned long *dirty_bitmap;
2152 unsigned long *dirty_bitmap_buffer;
2155 /* Dirty ring tracking may be exclusive to dirty log tracking */
2156 if (!kvm_use_dirty_bitmap(kvm))
2159 as_id = log->slot >> 16;
2160 id = (u16)log->slot;
2161 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
2164 slots = __kvm_memslots(kvm, as_id);
2165 memslot = id_to_memslot(slots, id);
2166 if (!memslot || !memslot->dirty_bitmap)
2169 dirty_bitmap = memslot->dirty_bitmap;
2171 kvm_arch_sync_dirty_log(kvm, memslot);
2173 n = kvm_dirty_bitmap_bytes(memslot);
2175 if (kvm->manual_dirty_log_protect) {
2177 * Unlike kvm_get_dirty_log, we always return false in *flush,
2178 * because no flush is needed until KVM_CLEAR_DIRTY_LOG. There
2179 * is some code duplication between this function and
2180 * kvm_get_dirty_log, but hopefully all architecture
2181 * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log
2182 * can be eliminated.
2184 dirty_bitmap_buffer = dirty_bitmap;
2186 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
2187 memset(dirty_bitmap_buffer, 0, n);
2190 for (i = 0; i < n / sizeof(long); i++) {
2194 if (!dirty_bitmap[i])
2198 mask = xchg(&dirty_bitmap[i], 0);
2199 dirty_bitmap_buffer[i] = mask;
2201 offset = i * BITS_PER_LONG;
2202 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
2205 KVM_MMU_UNLOCK(kvm);
2209 kvm_flush_remote_tlbs_memslot(kvm, memslot);
2211 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
2218 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
2219 * @kvm: kvm instance
2220 * @log: slot id and address to which we copy the log
2222 * Steps 1-4 below provide general overview of dirty page logging. See
2223 * kvm_get_dirty_log_protect() function description for additional details.
2225 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
2226 * always flush the TLB (step 4) even if previous step failed and the dirty
2227 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
2228 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
2229 * writes will be marked dirty for next log read.
2231 * 1. Take a snapshot of the bit and clear it if needed.
2232 * 2. Write protect the corresponding page.
2233 * 3. Copy the snapshot to the userspace.
2234 * 4. Flush TLB's if needed.
2236 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
2237 struct kvm_dirty_log *log)
2241 mutex_lock(&kvm->slots_lock);
2243 r = kvm_get_dirty_log_protect(kvm, log);
2245 mutex_unlock(&kvm->slots_lock);
2250 * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap
2251 * and reenable dirty page tracking for the corresponding pages.
2252 * @kvm: pointer to kvm instance
2253 * @log: slot id and address from which to fetch the bitmap of dirty pages
2255 static int kvm_clear_dirty_log_protect(struct kvm *kvm,
2256 struct kvm_clear_dirty_log *log)
2258 struct kvm_memslots *slots;
2259 struct kvm_memory_slot *memslot;
2263 unsigned long *dirty_bitmap;
2264 unsigned long *dirty_bitmap_buffer;
2267 /* Dirty ring tracking may be exclusive to dirty log tracking */
2268 if (!kvm_use_dirty_bitmap(kvm))
2271 as_id = log->slot >> 16;
2272 id = (u16)log->slot;
2273 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
2276 if (log->first_page & 63)
2279 slots = __kvm_memslots(kvm, as_id);
2280 memslot = id_to_memslot(slots, id);
2281 if (!memslot || !memslot->dirty_bitmap)
2284 dirty_bitmap = memslot->dirty_bitmap;
2286 n = ALIGN(log->num_pages, BITS_PER_LONG) / 8;
2288 if (log->first_page > memslot->npages ||
2289 log->num_pages > memslot->npages - log->first_page ||
2290 (log->num_pages < memslot->npages - log->first_page && (log->num_pages & 63)))
2293 kvm_arch_sync_dirty_log(kvm, memslot);
2296 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
2297 if (copy_from_user(dirty_bitmap_buffer, log->dirty_bitmap, n))
2301 for (offset = log->first_page, i = offset / BITS_PER_LONG,
2302 n = DIV_ROUND_UP(log->num_pages, BITS_PER_LONG); n--;
2303 i++, offset += BITS_PER_LONG) {
2304 unsigned long mask = *dirty_bitmap_buffer++;
2305 atomic_long_t *p = (atomic_long_t *) &dirty_bitmap[i];
2309 mask &= atomic_long_fetch_andnot(mask, p);
2312 * mask contains the bits that really have been cleared. This
2313 * never includes any bits beyond the length of the memslot (if
2314 * the length is not aligned to 64 pages), therefore it is not
2315 * a problem if userspace sets them in log->dirty_bitmap.
2319 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
2323 KVM_MMU_UNLOCK(kvm);
2326 kvm_flush_remote_tlbs_memslot(kvm, memslot);
2331 static int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm,
2332 struct kvm_clear_dirty_log *log)
2336 mutex_lock(&kvm->slots_lock);
2338 r = kvm_clear_dirty_log_protect(kvm, log);
2340 mutex_unlock(&kvm->slots_lock);
2343 #endif /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
2345 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
2347 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
2349 EXPORT_SYMBOL_GPL(gfn_to_memslot);
2351 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
2353 struct kvm_memslots *slots = kvm_vcpu_memslots(vcpu);
2354 u64 gen = slots->generation;
2355 struct kvm_memory_slot *slot;
2358 * This also protects against using a memslot from a different address space,
2359 * since different address spaces have different generation numbers.
2361 if (unlikely(gen != vcpu->last_used_slot_gen)) {
2362 vcpu->last_used_slot = NULL;
2363 vcpu->last_used_slot_gen = gen;
2366 slot = try_get_memslot(vcpu->last_used_slot, gfn);
2371 * Fall back to searching all memslots. We purposely use
2372 * search_memslots() instead of __gfn_to_memslot() to avoid
2373 * thrashing the VM-wide last_used_slot in kvm_memslots.
2375 slot = search_memslots(slots, gfn, false);
2377 vcpu->last_used_slot = slot;
2384 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
2386 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
2388 return kvm_is_visible_memslot(memslot);
2390 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
2392 bool kvm_vcpu_is_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
2394 struct kvm_memory_slot *memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2396 return kvm_is_visible_memslot(memslot);
2398 EXPORT_SYMBOL_GPL(kvm_vcpu_is_visible_gfn);
2400 unsigned long kvm_host_page_size(struct kvm_vcpu *vcpu, gfn_t gfn)
2402 struct vm_area_struct *vma;
2403 unsigned long addr, size;
2407 addr = kvm_vcpu_gfn_to_hva_prot(vcpu, gfn, NULL);
2408 if (kvm_is_error_hva(addr))
2411 mmap_read_lock(current->mm);
2412 vma = find_vma(current->mm, addr);
2416 size = vma_kernel_pagesize(vma);
2419 mmap_read_unlock(current->mm);
2424 static bool memslot_is_readonly(const struct kvm_memory_slot *slot)
2426 return slot->flags & KVM_MEM_READONLY;
2429 static unsigned long __gfn_to_hva_many(const struct kvm_memory_slot *slot, gfn_t gfn,
2430 gfn_t *nr_pages, bool write)
2432 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
2433 return KVM_HVA_ERR_BAD;
2435 if (memslot_is_readonly(slot) && write)
2436 return KVM_HVA_ERR_RO_BAD;
2439 *nr_pages = slot->npages - (gfn - slot->base_gfn);
2441 return __gfn_to_hva_memslot(slot, gfn);
2444 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
2447 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
2450 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
2453 return gfn_to_hva_many(slot, gfn, NULL);
2455 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
2457 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
2459 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
2461 EXPORT_SYMBOL_GPL(gfn_to_hva);
2463 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
2465 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
2467 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
2470 * Return the hva of a @gfn and the R/W attribute if possible.
2472 * @slot: the kvm_memory_slot which contains @gfn
2473 * @gfn: the gfn to be translated
2474 * @writable: used to return the read/write attribute of the @slot if the hva
2475 * is valid and @writable is not NULL
2477 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
2478 gfn_t gfn, bool *writable)
2480 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
2482 if (!kvm_is_error_hva(hva) && writable)
2483 *writable = !memslot_is_readonly(slot);
2488 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
2490 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2492 return gfn_to_hva_memslot_prot(slot, gfn, writable);
2495 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
2497 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2499 return gfn_to_hva_memslot_prot(slot, gfn, writable);
2502 static inline int check_user_page_hwpoison(unsigned long addr)
2504 int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
2506 rc = get_user_pages(addr, 1, flags, NULL);
2507 return rc == -EHWPOISON;
2511 * The fast path to get the writable pfn which will be stored in @pfn,
2512 * true indicates success, otherwise false is returned. It's also the
2513 * only part that runs if we can in atomic context.
2515 static bool hva_to_pfn_fast(unsigned long addr, bool write_fault,
2516 bool *writable, kvm_pfn_t *pfn)
2518 struct page *page[1];
2521 * Fast pin a writable pfn only if it is a write fault request
2522 * or the caller allows to map a writable pfn for a read fault
2525 if (!(write_fault || writable))
2528 if (get_user_page_fast_only(addr, FOLL_WRITE, page)) {
2529 *pfn = page_to_pfn(page[0]);
2540 * The slow path to get the pfn of the specified host virtual address,
2541 * 1 indicates success, -errno is returned if error is detected.
2543 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
2544 bool interruptible, bool *writable, kvm_pfn_t *pfn)
2546 unsigned int flags = FOLL_HWPOISON;
2553 *writable = write_fault;
2556 flags |= FOLL_WRITE;
2558 flags |= FOLL_NOWAIT;
2560 flags |= FOLL_INTERRUPTIBLE;
2562 npages = get_user_pages_unlocked(addr, 1, &page, flags);
2566 /* map read fault as writable if possible */
2567 if (unlikely(!write_fault) && writable) {
2570 if (get_user_page_fast_only(addr, FOLL_WRITE, &wpage)) {
2576 *pfn = page_to_pfn(page);
2580 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
2582 if (unlikely(!(vma->vm_flags & VM_READ)))
2585 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
2591 static int kvm_try_get_pfn(kvm_pfn_t pfn)
2593 struct page *page = kvm_pfn_to_refcounted_page(pfn);
2598 return get_page_unless_zero(page);
2601 static int hva_to_pfn_remapped(struct vm_area_struct *vma,
2602 unsigned long addr, bool write_fault,
2603 bool *writable, kvm_pfn_t *p_pfn)
2611 r = follow_pte(vma->vm_mm, addr, &ptep, &ptl);
2614 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
2615 * not call the fault handler, so do it here.
2617 bool unlocked = false;
2618 r = fixup_user_fault(current->mm, addr,
2619 (write_fault ? FAULT_FLAG_WRITE : 0),
2626 r = follow_pte(vma->vm_mm, addr, &ptep, &ptl);
2631 pte = ptep_get(ptep);
2633 if (write_fault && !pte_write(pte)) {
2634 pfn = KVM_PFN_ERR_RO_FAULT;
2639 *writable = pte_write(pte);
2643 * Get a reference here because callers of *hva_to_pfn* and
2644 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
2645 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
2646 * set, but the kvm_try_get_pfn/kvm_release_pfn_clean pair will
2647 * simply do nothing for reserved pfns.
2649 * Whoever called remap_pfn_range is also going to call e.g.
2650 * unmap_mapping_range before the underlying pages are freed,
2651 * causing a call to our MMU notifier.
2653 * Certain IO or PFNMAP mappings can be backed with valid
2654 * struct pages, but be allocated without refcounting e.g.,
2655 * tail pages of non-compound higher order allocations, which
2656 * would then underflow the refcount when the caller does the
2657 * required put_page. Don't allow those pages here.
2659 if (!kvm_try_get_pfn(pfn))
2663 pte_unmap_unlock(ptep, ptl);
2670 * Pin guest page in memory and return its pfn.
2671 * @addr: host virtual address which maps memory to the guest
2672 * @atomic: whether this function can sleep
2673 * @interruptible: whether the process can be interrupted by non-fatal signals
2674 * @async: whether this function need to wait IO complete if the
2675 * host page is not in the memory
2676 * @write_fault: whether we should get a writable host page
2677 * @writable: whether it allows to map a writable host page for !@write_fault
2679 * The function will map a writable host page for these two cases:
2680 * 1): @write_fault = true
2681 * 2): @write_fault = false && @writable, @writable will tell the caller
2682 * whether the mapping is writable.
2684 kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool interruptible,
2685 bool *async, bool write_fault, bool *writable)
2687 struct vm_area_struct *vma;
2691 /* we can do it either atomically or asynchronously, not both */
2692 BUG_ON(atomic && async);
2694 if (hva_to_pfn_fast(addr, write_fault, writable, &pfn))
2698 return KVM_PFN_ERR_FAULT;
2700 npages = hva_to_pfn_slow(addr, async, write_fault, interruptible,
2704 if (npages == -EINTR)
2705 return KVM_PFN_ERR_SIGPENDING;
2707 mmap_read_lock(current->mm);
2708 if (npages == -EHWPOISON ||
2709 (!async && check_user_page_hwpoison(addr))) {
2710 pfn = KVM_PFN_ERR_HWPOISON;
2715 vma = vma_lookup(current->mm, addr);
2718 pfn = KVM_PFN_ERR_FAULT;
2719 else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
2720 r = hva_to_pfn_remapped(vma, addr, write_fault, writable, &pfn);
2724 pfn = KVM_PFN_ERR_FAULT;
2726 if (async && vma_is_valid(vma, write_fault))
2728 pfn = KVM_PFN_ERR_FAULT;
2731 mmap_read_unlock(current->mm);
2735 kvm_pfn_t __gfn_to_pfn_memslot(const struct kvm_memory_slot *slot, gfn_t gfn,
2736 bool atomic, bool interruptible, bool *async,
2737 bool write_fault, bool *writable, hva_t *hva)
2739 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
2744 if (addr == KVM_HVA_ERR_RO_BAD) {
2747 return KVM_PFN_ERR_RO_FAULT;
2750 if (kvm_is_error_hva(addr)) {
2753 return KVM_PFN_NOSLOT;
2756 /* Do not map writable pfn in the readonly memslot. */
2757 if (writable && memslot_is_readonly(slot)) {
2762 return hva_to_pfn(addr, atomic, interruptible, async, write_fault,
2765 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
2767 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
2770 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, false,
2771 NULL, write_fault, writable, NULL);
2773 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
2775 kvm_pfn_t gfn_to_pfn_memslot(const struct kvm_memory_slot *slot, gfn_t gfn)
2777 return __gfn_to_pfn_memslot(slot, gfn, false, false, NULL, true,
2780 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
2782 kvm_pfn_t gfn_to_pfn_memslot_atomic(const struct kvm_memory_slot *slot, gfn_t gfn)
2784 return __gfn_to_pfn_memslot(slot, gfn, true, false, NULL, true,
2787 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
2789 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
2791 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
2793 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
2795 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
2797 return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
2799 EXPORT_SYMBOL_GPL(gfn_to_pfn);
2801 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
2803 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
2805 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
2807 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
2808 struct page **pages, int nr_pages)
2813 addr = gfn_to_hva_many(slot, gfn, &entry);
2814 if (kvm_is_error_hva(addr))
2817 if (entry < nr_pages)
2820 return get_user_pages_fast_only(addr, nr_pages, FOLL_WRITE, pages);
2822 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
2825 * Do not use this helper unless you are absolutely certain the gfn _must_ be
2826 * backed by 'struct page'. A valid example is if the backing memslot is
2827 * controlled by KVM. Note, if the returned page is valid, it's refcount has
2828 * been elevated by gfn_to_pfn().
2830 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
2835 pfn = gfn_to_pfn(kvm, gfn);
2837 if (is_error_noslot_pfn(pfn))
2838 return KVM_ERR_PTR_BAD_PAGE;
2840 page = kvm_pfn_to_refcounted_page(pfn);
2842 return KVM_ERR_PTR_BAD_PAGE;
2846 EXPORT_SYMBOL_GPL(gfn_to_page);
2848 void kvm_release_pfn(kvm_pfn_t pfn, bool dirty)
2851 kvm_release_pfn_dirty(pfn);
2853 kvm_release_pfn_clean(pfn);
2856 int kvm_vcpu_map(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map)
2860 struct page *page = KVM_UNMAPPED_PAGE;
2865 pfn = gfn_to_pfn(vcpu->kvm, gfn);
2866 if (is_error_noslot_pfn(pfn))
2869 if (pfn_valid(pfn)) {
2870 page = pfn_to_page(pfn);
2872 #ifdef CONFIG_HAS_IOMEM
2874 hva = memremap(pfn_to_hpa(pfn), PAGE_SIZE, MEMREMAP_WB);
2888 EXPORT_SYMBOL_GPL(kvm_vcpu_map);
2890 void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map, bool dirty)
2898 if (map->page != KVM_UNMAPPED_PAGE)
2900 #ifdef CONFIG_HAS_IOMEM
2906 kvm_vcpu_mark_page_dirty(vcpu, map->gfn);
2908 kvm_release_pfn(map->pfn, dirty);
2913 EXPORT_SYMBOL_GPL(kvm_vcpu_unmap);
2915 static bool kvm_is_ad_tracked_page(struct page *page)
2918 * Per page-flags.h, pages tagged PG_reserved "should in general not be
2919 * touched (e.g. set dirty) except by its owner".
2921 return !PageReserved(page);
2924 static void kvm_set_page_dirty(struct page *page)
2926 if (kvm_is_ad_tracked_page(page))
2930 static void kvm_set_page_accessed(struct page *page)
2932 if (kvm_is_ad_tracked_page(page))
2933 mark_page_accessed(page);
2936 void kvm_release_page_clean(struct page *page)
2938 WARN_ON(is_error_page(page));
2940 kvm_set_page_accessed(page);
2943 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
2945 void kvm_release_pfn_clean(kvm_pfn_t pfn)
2949 if (is_error_noslot_pfn(pfn))
2952 page = kvm_pfn_to_refcounted_page(pfn);
2956 kvm_release_page_clean(page);
2958 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
2960 void kvm_release_page_dirty(struct page *page)
2962 WARN_ON(is_error_page(page));
2964 kvm_set_page_dirty(page);
2965 kvm_release_page_clean(page);
2967 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
2969 void kvm_release_pfn_dirty(kvm_pfn_t pfn)
2973 if (is_error_noslot_pfn(pfn))
2976 page = kvm_pfn_to_refcounted_page(pfn);
2980 kvm_release_page_dirty(page);
2982 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
2985 * Note, checking for an error/noslot pfn is the caller's responsibility when
2986 * directly marking a page dirty/accessed. Unlike the "release" helpers, the
2987 * "set" helpers are not to be used when the pfn might point at garbage.
2989 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
2991 if (WARN_ON(is_error_noslot_pfn(pfn)))
2995 kvm_set_page_dirty(pfn_to_page(pfn));
2997 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
2999 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
3001 if (WARN_ON(is_error_noslot_pfn(pfn)))
3005 kvm_set_page_accessed(pfn_to_page(pfn));
3007 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
3009 static int next_segment(unsigned long len, int offset)
3011 if (len > PAGE_SIZE - offset)
3012 return PAGE_SIZE - offset;
3017 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
3018 void *data, int offset, int len)
3023 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
3024 if (kvm_is_error_hva(addr))
3026 r = __copy_from_user(data, (void __user *)addr + offset, len);
3032 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
3035 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
3037 return __kvm_read_guest_page(slot, gfn, data, offset, len);
3039 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
3041 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
3042 int offset, int len)
3044 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
3046 return __kvm_read_guest_page(slot, gfn, data, offset, len);
3048 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
3050 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
3052 gfn_t gfn = gpa >> PAGE_SHIFT;
3054 int offset = offset_in_page(gpa);
3057 while ((seg = next_segment(len, offset)) != 0) {
3058 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
3068 EXPORT_SYMBOL_GPL(kvm_read_guest);
3070 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
3072 gfn_t gfn = gpa >> PAGE_SHIFT;
3074 int offset = offset_in_page(gpa);
3077 while ((seg = next_segment(len, offset)) != 0) {
3078 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
3088 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
3090 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
3091 void *data, int offset, unsigned long len)
3096 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
3097 if (kvm_is_error_hva(addr))
3099 pagefault_disable();
3100 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
3107 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
3108 void *data, unsigned long len)
3110 gfn_t gfn = gpa >> PAGE_SHIFT;
3111 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
3112 int offset = offset_in_page(gpa);
3114 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
3116 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
3118 static int __kvm_write_guest_page(struct kvm *kvm,
3119 struct kvm_memory_slot *memslot, gfn_t gfn,
3120 const void *data, int offset, int len)
3125 addr = gfn_to_hva_memslot(memslot, gfn);
3126 if (kvm_is_error_hva(addr))
3128 r = __copy_to_user((void __user *)addr + offset, data, len);
3131 mark_page_dirty_in_slot(kvm, memslot, gfn);
3135 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
3136 const void *data, int offset, int len)
3138 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
3140 return __kvm_write_guest_page(kvm, slot, gfn, data, offset, len);
3142 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
3144 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
3145 const void *data, int offset, int len)
3147 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
3149 return __kvm_write_guest_page(vcpu->kvm, slot, gfn, data, offset, len);
3151 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
3153 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
3156 gfn_t gfn = gpa >> PAGE_SHIFT;
3158 int offset = offset_in_page(gpa);
3161 while ((seg = next_segment(len, offset)) != 0) {
3162 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
3172 EXPORT_SYMBOL_GPL(kvm_write_guest);
3174 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
3177 gfn_t gfn = gpa >> PAGE_SHIFT;
3179 int offset = offset_in_page(gpa);
3182 while ((seg = next_segment(len, offset)) != 0) {
3183 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
3193 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
3195 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
3196 struct gfn_to_hva_cache *ghc,
3197 gpa_t gpa, unsigned long len)
3199 int offset = offset_in_page(gpa);
3200 gfn_t start_gfn = gpa >> PAGE_SHIFT;
3201 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
3202 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
3203 gfn_t nr_pages_avail;
3205 /* Update ghc->generation before performing any error checks. */
3206 ghc->generation = slots->generation;
3208 if (start_gfn > end_gfn) {
3209 ghc->hva = KVM_HVA_ERR_BAD;
3214 * If the requested region crosses two memslots, we still
3215 * verify that the entire region is valid here.
3217 for ( ; start_gfn <= end_gfn; start_gfn += nr_pages_avail) {
3218 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
3219 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
3221 if (kvm_is_error_hva(ghc->hva))
3225 /* Use the slow path for cross page reads and writes. */
3226 if (nr_pages_needed == 1)
3229 ghc->memslot = NULL;
3236 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3237 gpa_t gpa, unsigned long len)
3239 struct kvm_memslots *slots = kvm_memslots(kvm);
3240 return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len);
3242 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
3244 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3245 void *data, unsigned int offset,
3248 struct kvm_memslots *slots = kvm_memslots(kvm);
3250 gpa_t gpa = ghc->gpa + offset;
3252 if (WARN_ON_ONCE(len + offset > ghc->len))
3255 if (slots->generation != ghc->generation) {
3256 if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len))
3260 if (kvm_is_error_hva(ghc->hva))
3263 if (unlikely(!ghc->memslot))
3264 return kvm_write_guest(kvm, gpa, data, len);
3266 r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
3269 mark_page_dirty_in_slot(kvm, ghc->memslot, gpa >> PAGE_SHIFT);
3273 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached);
3275 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3276 void *data, unsigned long len)
3278 return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len);
3280 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
3282 int kvm_read_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3283 void *data, unsigned int offset,
3286 struct kvm_memslots *slots = kvm_memslots(kvm);
3288 gpa_t gpa = ghc->gpa + offset;
3290 if (WARN_ON_ONCE(len + offset > ghc->len))
3293 if (slots->generation != ghc->generation) {
3294 if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len))
3298 if (kvm_is_error_hva(ghc->hva))
3301 if (unlikely(!ghc->memslot))
3302 return kvm_read_guest(kvm, gpa, data, len);
3304 r = __copy_from_user(data, (void __user *)ghc->hva + offset, len);
3310 EXPORT_SYMBOL_GPL(kvm_read_guest_offset_cached);
3312 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3313 void *data, unsigned long len)
3315 return kvm_read_guest_offset_cached(kvm, ghc, data, 0, len);
3317 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
3319 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
3321 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
3322 gfn_t gfn = gpa >> PAGE_SHIFT;
3324 int offset = offset_in_page(gpa);
3327 while ((seg = next_segment(len, offset)) != 0) {
3328 ret = kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
3337 EXPORT_SYMBOL_GPL(kvm_clear_guest);
3339 void mark_page_dirty_in_slot(struct kvm *kvm,
3340 const struct kvm_memory_slot *memslot,
3343 struct kvm_vcpu *vcpu = kvm_get_running_vcpu();
3345 #ifdef CONFIG_HAVE_KVM_DIRTY_RING
3346 if (WARN_ON_ONCE(vcpu && vcpu->kvm != kvm))
3349 WARN_ON_ONCE(!vcpu && !kvm_arch_allow_write_without_running_vcpu(kvm));
3352 if (memslot && kvm_slot_dirty_track_enabled(memslot)) {
3353 unsigned long rel_gfn = gfn - memslot->base_gfn;
3354 u32 slot = (memslot->as_id << 16) | memslot->id;
3356 if (kvm->dirty_ring_size && vcpu)
3357 kvm_dirty_ring_push(vcpu, slot, rel_gfn);
3358 else if (memslot->dirty_bitmap)
3359 set_bit_le(rel_gfn, memslot->dirty_bitmap);
3362 EXPORT_SYMBOL_GPL(mark_page_dirty_in_slot);
3364 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
3366 struct kvm_memory_slot *memslot;
3368 memslot = gfn_to_memslot(kvm, gfn);
3369 mark_page_dirty_in_slot(kvm, memslot, gfn);
3371 EXPORT_SYMBOL_GPL(mark_page_dirty);
3373 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
3375 struct kvm_memory_slot *memslot;
3377 memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
3378 mark_page_dirty_in_slot(vcpu->kvm, memslot, gfn);
3380 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
3382 void kvm_sigset_activate(struct kvm_vcpu *vcpu)
3384 if (!vcpu->sigset_active)
3388 * This does a lockless modification of ->real_blocked, which is fine
3389 * because, only current can change ->real_blocked and all readers of
3390 * ->real_blocked don't care as long ->real_blocked is always a subset
3393 sigprocmask(SIG_SETMASK, &vcpu->sigset, ¤t->real_blocked);
3396 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu)
3398 if (!vcpu->sigset_active)
3401 sigprocmask(SIG_SETMASK, ¤t->real_blocked, NULL);
3402 sigemptyset(¤t->real_blocked);
3405 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
3407 unsigned int old, val, grow, grow_start;
3409 old = val = vcpu->halt_poll_ns;
3410 grow_start = READ_ONCE(halt_poll_ns_grow_start);
3411 grow = READ_ONCE(halt_poll_ns_grow);
3416 if (val < grow_start)
3419 vcpu->halt_poll_ns = val;
3421 trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
3424 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
3426 unsigned int old, val, shrink, grow_start;
3428 old = val = vcpu->halt_poll_ns;
3429 shrink = READ_ONCE(halt_poll_ns_shrink);
3430 grow_start = READ_ONCE(halt_poll_ns_grow_start);
3436 if (val < grow_start)
3439 vcpu->halt_poll_ns = val;
3440 trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
3443 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
3446 int idx = srcu_read_lock(&vcpu->kvm->srcu);
3448 if (kvm_arch_vcpu_runnable(vcpu))
3450 if (kvm_cpu_has_pending_timer(vcpu))
3452 if (signal_pending(current))
3454 if (kvm_check_request(KVM_REQ_UNBLOCK, vcpu))
3459 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3464 * Block the vCPU until the vCPU is runnable, an event arrives, or a signal is
3465 * pending. This is mostly used when halting a vCPU, but may also be used
3466 * directly for other vCPU non-runnable states, e.g. x86's Wait-For-SIPI.
3468 bool kvm_vcpu_block(struct kvm_vcpu *vcpu)
3470 struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu);
3471 bool waited = false;
3473 vcpu->stat.generic.blocking = 1;
3476 kvm_arch_vcpu_blocking(vcpu);
3477 prepare_to_rcuwait(wait);
3481 set_current_state(TASK_INTERRUPTIBLE);
3483 if (kvm_vcpu_check_block(vcpu) < 0)
3491 finish_rcuwait(wait);
3492 kvm_arch_vcpu_unblocking(vcpu);
3495 vcpu->stat.generic.blocking = 0;
3500 static inline void update_halt_poll_stats(struct kvm_vcpu *vcpu, ktime_t start,
3501 ktime_t end, bool success)
3503 struct kvm_vcpu_stat_generic *stats = &vcpu->stat.generic;
3504 u64 poll_ns = ktime_to_ns(ktime_sub(end, start));
3506 ++vcpu->stat.generic.halt_attempted_poll;
3509 ++vcpu->stat.generic.halt_successful_poll;
3511 if (!vcpu_valid_wakeup(vcpu))
3512 ++vcpu->stat.generic.halt_poll_invalid;
3514 stats->halt_poll_success_ns += poll_ns;
3515 KVM_STATS_LOG_HIST_UPDATE(stats->halt_poll_success_hist, poll_ns);
3517 stats->halt_poll_fail_ns += poll_ns;
3518 KVM_STATS_LOG_HIST_UPDATE(stats->halt_poll_fail_hist, poll_ns);
3522 static unsigned int kvm_vcpu_max_halt_poll_ns(struct kvm_vcpu *vcpu)
3524 struct kvm *kvm = vcpu->kvm;
3526 if (kvm->override_halt_poll_ns) {
3528 * Ensure kvm->max_halt_poll_ns is not read before
3529 * kvm->override_halt_poll_ns.
3531 * Pairs with the smp_wmb() when enabling KVM_CAP_HALT_POLL.
3534 return READ_ONCE(kvm->max_halt_poll_ns);
3537 return READ_ONCE(halt_poll_ns);
3541 * Emulate a vCPU halt condition, e.g. HLT on x86, WFI on arm, etc... If halt
3542 * polling is enabled, busy wait for a short time before blocking to avoid the
3543 * expensive block+unblock sequence if a wake event arrives soon after the vCPU
3546 void kvm_vcpu_halt(struct kvm_vcpu *vcpu)
3548 unsigned int max_halt_poll_ns = kvm_vcpu_max_halt_poll_ns(vcpu);
3549 bool halt_poll_allowed = !kvm_arch_no_poll(vcpu);
3550 ktime_t start, cur, poll_end;
3551 bool waited = false;
3555 if (vcpu->halt_poll_ns > max_halt_poll_ns)
3556 vcpu->halt_poll_ns = max_halt_poll_ns;
3558 do_halt_poll = halt_poll_allowed && vcpu->halt_poll_ns;
3560 start = cur = poll_end = ktime_get();
3562 ktime_t stop = ktime_add_ns(start, vcpu->halt_poll_ns);
3565 if (kvm_vcpu_check_block(vcpu) < 0)
3568 poll_end = cur = ktime_get();
3569 } while (kvm_vcpu_can_poll(cur, stop));
3572 waited = kvm_vcpu_block(vcpu);
3576 vcpu->stat.generic.halt_wait_ns +=
3577 ktime_to_ns(cur) - ktime_to_ns(poll_end);
3578 KVM_STATS_LOG_HIST_UPDATE(vcpu->stat.generic.halt_wait_hist,
3579 ktime_to_ns(cur) - ktime_to_ns(poll_end));
3582 /* The total time the vCPU was "halted", including polling time. */
3583 halt_ns = ktime_to_ns(cur) - ktime_to_ns(start);
3586 * Note, halt-polling is considered successful so long as the vCPU was
3587 * never actually scheduled out, i.e. even if the wake event arrived
3588 * after of the halt-polling loop itself, but before the full wait.
3591 update_halt_poll_stats(vcpu, start, poll_end, !waited);
3593 if (halt_poll_allowed) {
3594 /* Recompute the max halt poll time in case it changed. */
3595 max_halt_poll_ns = kvm_vcpu_max_halt_poll_ns(vcpu);
3597 if (!vcpu_valid_wakeup(vcpu)) {
3598 shrink_halt_poll_ns(vcpu);
3599 } else if (max_halt_poll_ns) {
3600 if (halt_ns <= vcpu->halt_poll_ns)
3602 /* we had a long block, shrink polling */
3603 else if (vcpu->halt_poll_ns &&
3604 halt_ns > max_halt_poll_ns)
3605 shrink_halt_poll_ns(vcpu);
3606 /* we had a short halt and our poll time is too small */
3607 else if (vcpu->halt_poll_ns < max_halt_poll_ns &&
3608 halt_ns < max_halt_poll_ns)
3609 grow_halt_poll_ns(vcpu);
3611 vcpu->halt_poll_ns = 0;
3615 trace_kvm_vcpu_wakeup(halt_ns, waited, vcpu_valid_wakeup(vcpu));
3617 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
3619 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
3621 if (__kvm_vcpu_wake_up(vcpu)) {
3622 WRITE_ONCE(vcpu->ready, true);
3623 ++vcpu->stat.generic.halt_wakeup;
3629 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
3633 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
3635 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
3639 if (kvm_vcpu_wake_up(vcpu))
3644 * The only state change done outside the vcpu mutex is IN_GUEST_MODE
3645 * to EXITING_GUEST_MODE. Therefore the moderately expensive "should
3646 * kick" check does not need atomic operations if kvm_vcpu_kick is used
3647 * within the vCPU thread itself.
3649 if (vcpu == __this_cpu_read(kvm_running_vcpu)) {
3650 if (vcpu->mode == IN_GUEST_MODE)
3651 WRITE_ONCE(vcpu->mode, EXITING_GUEST_MODE);
3656 * Note, the vCPU could get migrated to a different pCPU at any point
3657 * after kvm_arch_vcpu_should_kick(), which could result in sending an
3658 * IPI to the previous pCPU. But, that's ok because the purpose of the
3659 * IPI is to force the vCPU to leave IN_GUEST_MODE, and migrating the
3660 * vCPU also requires it to leave IN_GUEST_MODE.
3662 if (kvm_arch_vcpu_should_kick(vcpu)) {
3663 cpu = READ_ONCE(vcpu->cpu);
3664 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
3665 smp_send_reschedule(cpu);
3670 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
3671 #endif /* !CONFIG_S390 */
3673 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
3676 struct task_struct *task = NULL;
3680 pid = rcu_dereference(target->pid);
3682 task = get_pid_task(pid, PIDTYPE_PID);
3686 ret = yield_to(task, 1);
3687 put_task_struct(task);
3691 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
3694 * Helper that checks whether a VCPU is eligible for directed yield.
3695 * Most eligible candidate to yield is decided by following heuristics:
3697 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
3698 * (preempted lock holder), indicated by @in_spin_loop.
3699 * Set at the beginning and cleared at the end of interception/PLE handler.
3701 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
3702 * chance last time (mostly it has become eligible now since we have probably
3703 * yielded to lockholder in last iteration. This is done by toggling
3704 * @dy_eligible each time a VCPU checked for eligibility.)
3706 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
3707 * to preempted lock-holder could result in wrong VCPU selection and CPU
3708 * burning. Giving priority for a potential lock-holder increases lock
3711 * Since algorithm is based on heuristics, accessing another VCPU data without
3712 * locking does not harm. It may result in trying to yield to same VCPU, fail
3713 * and continue with next VCPU and so on.
3715 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
3717 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
3720 eligible = !vcpu->spin_loop.in_spin_loop ||
3721 vcpu->spin_loop.dy_eligible;
3723 if (vcpu->spin_loop.in_spin_loop)
3724 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
3733 * Unlike kvm_arch_vcpu_runnable, this function is called outside
3734 * a vcpu_load/vcpu_put pair. However, for most architectures
3735 * kvm_arch_vcpu_runnable does not require vcpu_load.
3737 bool __weak kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
3739 return kvm_arch_vcpu_runnable(vcpu);
3742 static bool vcpu_dy_runnable(struct kvm_vcpu *vcpu)
3744 if (kvm_arch_dy_runnable(vcpu))
3747 #ifdef CONFIG_KVM_ASYNC_PF
3748 if (!list_empty_careful(&vcpu->async_pf.done))
3755 bool __weak kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu)
3760 void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode)
3762 struct kvm *kvm = me->kvm;
3763 struct kvm_vcpu *vcpu;
3764 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
3770 kvm_vcpu_set_in_spin_loop(me, true);
3772 * We boost the priority of a VCPU that is runnable but not
3773 * currently running, because it got preempted by something
3774 * else and called schedule in __vcpu_run. Hopefully that
3775 * VCPU is holding the lock that we need and will release it.
3776 * We approximate round-robin by starting at the last boosted VCPU.
3778 for (pass = 0; pass < 2 && !yielded && try; pass++) {
3779 kvm_for_each_vcpu(i, vcpu, kvm) {
3780 if (!pass && i <= last_boosted_vcpu) {
3781 i = last_boosted_vcpu;
3783 } else if (pass && i > last_boosted_vcpu)
3785 if (!READ_ONCE(vcpu->ready))
3789 if (kvm_vcpu_is_blocking(vcpu) && !vcpu_dy_runnable(vcpu))
3791 if (READ_ONCE(vcpu->preempted) && yield_to_kernel_mode &&
3792 !kvm_arch_dy_has_pending_interrupt(vcpu) &&
3793 !kvm_arch_vcpu_in_kernel(vcpu))
3795 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
3798 yielded = kvm_vcpu_yield_to(vcpu);
3800 kvm->last_boosted_vcpu = i;
3802 } else if (yielded < 0) {
3809 kvm_vcpu_set_in_spin_loop(me, false);
3811 /* Ensure vcpu is not eligible during next spinloop */
3812 kvm_vcpu_set_dy_eligible(me, false);
3814 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
3816 static bool kvm_page_in_dirty_ring(struct kvm *kvm, unsigned long pgoff)
3818 #ifdef CONFIG_HAVE_KVM_DIRTY_RING
3819 return (pgoff >= KVM_DIRTY_LOG_PAGE_OFFSET) &&
3820 (pgoff < KVM_DIRTY_LOG_PAGE_OFFSET +
3821 kvm->dirty_ring_size / PAGE_SIZE);
3827 static vm_fault_t kvm_vcpu_fault(struct vm_fault *vmf)
3829 struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data;
3832 if (vmf->pgoff == 0)
3833 page = virt_to_page(vcpu->run);
3835 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
3836 page = virt_to_page(vcpu->arch.pio_data);
3838 #ifdef CONFIG_KVM_MMIO
3839 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
3840 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
3842 else if (kvm_page_in_dirty_ring(vcpu->kvm, vmf->pgoff))
3843 page = kvm_dirty_ring_get_page(
3845 vmf->pgoff - KVM_DIRTY_LOG_PAGE_OFFSET);
3847 return kvm_arch_vcpu_fault(vcpu, vmf);
3853 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
3854 .fault = kvm_vcpu_fault,
3857 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
3859 struct kvm_vcpu *vcpu = file->private_data;
3860 unsigned long pages = vma_pages(vma);
3862 if ((kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff) ||
3863 kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff + pages - 1)) &&
3864 ((vma->vm_flags & VM_EXEC) || !(vma->vm_flags & VM_SHARED)))
3867 vma->vm_ops = &kvm_vcpu_vm_ops;
3871 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
3873 struct kvm_vcpu *vcpu = filp->private_data;
3875 kvm_put_kvm(vcpu->kvm);
3879 static const struct file_operations kvm_vcpu_fops = {
3880 .release = kvm_vcpu_release,
3881 .unlocked_ioctl = kvm_vcpu_ioctl,
3882 .mmap = kvm_vcpu_mmap,
3883 .llseek = noop_llseek,
3884 KVM_COMPAT(kvm_vcpu_compat_ioctl),
3888 * Allocates an inode for the vcpu.
3890 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
3892 char name[8 + 1 + ITOA_MAX_LEN + 1];
3894 snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id);
3895 return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
3898 #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS
3899 static int vcpu_get_pid(void *data, u64 *val)
3901 struct kvm_vcpu *vcpu = data;
3904 *val = pid_nr(rcu_dereference(vcpu->pid));
3909 DEFINE_SIMPLE_ATTRIBUTE(vcpu_get_pid_fops, vcpu_get_pid, NULL, "%llu\n");
3911 static void kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
3913 struct dentry *debugfs_dentry;
3914 char dir_name[ITOA_MAX_LEN * 2];
3916 if (!debugfs_initialized())
3919 snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
3920 debugfs_dentry = debugfs_create_dir(dir_name,
3921 vcpu->kvm->debugfs_dentry);
3922 debugfs_create_file("pid", 0444, debugfs_dentry, vcpu,
3923 &vcpu_get_pid_fops);
3925 kvm_arch_create_vcpu_debugfs(vcpu, debugfs_dentry);
3930 * Creates some virtual cpus. Good luck creating more than one.
3932 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
3935 struct kvm_vcpu *vcpu;
3938 if (id >= KVM_MAX_VCPU_IDS)
3941 mutex_lock(&kvm->lock);
3942 if (kvm->created_vcpus >= kvm->max_vcpus) {
3943 mutex_unlock(&kvm->lock);
3947 r = kvm_arch_vcpu_precreate(kvm, id);
3949 mutex_unlock(&kvm->lock);
3953 kvm->created_vcpus++;
3954 mutex_unlock(&kvm->lock);
3956 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL_ACCOUNT);
3959 goto vcpu_decrement;
3962 BUILD_BUG_ON(sizeof(struct kvm_run) > PAGE_SIZE);
3963 page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
3968 vcpu->run = page_address(page);
3970 kvm_vcpu_init(vcpu, kvm, id);
3972 r = kvm_arch_vcpu_create(vcpu);
3974 goto vcpu_free_run_page;
3976 if (kvm->dirty_ring_size) {
3977 r = kvm_dirty_ring_alloc(&vcpu->dirty_ring,
3978 id, kvm->dirty_ring_size);
3980 goto arch_vcpu_destroy;
3983 mutex_lock(&kvm->lock);
3985 #ifdef CONFIG_LOCKDEP
3986 /* Ensure that lockdep knows vcpu->mutex is taken *inside* kvm->lock */
3987 mutex_lock(&vcpu->mutex);
3988 mutex_unlock(&vcpu->mutex);
3991 if (kvm_get_vcpu_by_id(kvm, id)) {
3993 goto unlock_vcpu_destroy;
3996 vcpu->vcpu_idx = atomic_read(&kvm->online_vcpus);
3997 r = xa_reserve(&kvm->vcpu_array, vcpu->vcpu_idx, GFP_KERNEL_ACCOUNT);
3999 goto unlock_vcpu_destroy;
4001 /* Now it's all set up, let userspace reach it */
4003 r = create_vcpu_fd(vcpu);
4005 goto kvm_put_xa_release;
4007 if (KVM_BUG_ON(xa_store(&kvm->vcpu_array, vcpu->vcpu_idx, vcpu, 0), kvm)) {
4009 goto kvm_put_xa_release;
4013 * Pairs with smp_rmb() in kvm_get_vcpu. Store the vcpu
4014 * pointer before kvm->online_vcpu's incremented value.
4017 atomic_inc(&kvm->online_vcpus);
4019 mutex_unlock(&kvm->lock);
4020 kvm_arch_vcpu_postcreate(vcpu);
4021 kvm_create_vcpu_debugfs(vcpu);
4025 kvm_put_kvm_no_destroy(kvm);
4026 xa_release(&kvm->vcpu_array, vcpu->vcpu_idx);
4027 unlock_vcpu_destroy:
4028 mutex_unlock(&kvm->lock);
4029 kvm_dirty_ring_free(&vcpu->dirty_ring);
4031 kvm_arch_vcpu_destroy(vcpu);
4033 free_page((unsigned long)vcpu->run);
4035 kmem_cache_free(kvm_vcpu_cache, vcpu);
4037 mutex_lock(&kvm->lock);
4038 kvm->created_vcpus--;
4039 mutex_unlock(&kvm->lock);
4043 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
4046 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
4047 vcpu->sigset_active = 1;
4048 vcpu->sigset = *sigset;
4050 vcpu->sigset_active = 0;
4054 static ssize_t kvm_vcpu_stats_read(struct file *file, char __user *user_buffer,
4055 size_t size, loff_t *offset)
4057 struct kvm_vcpu *vcpu = file->private_data;
4059 return kvm_stats_read(vcpu->stats_id, &kvm_vcpu_stats_header,
4060 &kvm_vcpu_stats_desc[0], &vcpu->stat,
4061 sizeof(vcpu->stat), user_buffer, size, offset);
4064 static int kvm_vcpu_stats_release(struct inode *inode, struct file *file)
4066 struct kvm_vcpu *vcpu = file->private_data;
4068 kvm_put_kvm(vcpu->kvm);
4072 static const struct file_operations kvm_vcpu_stats_fops = {
4073 .read = kvm_vcpu_stats_read,
4074 .release = kvm_vcpu_stats_release,
4075 .llseek = noop_llseek,
4078 static int kvm_vcpu_ioctl_get_stats_fd(struct kvm_vcpu *vcpu)
4082 char name[15 + ITOA_MAX_LEN + 1];
4084 snprintf(name, sizeof(name), "kvm-vcpu-stats:%d", vcpu->vcpu_id);
4086 fd = get_unused_fd_flags(O_CLOEXEC);
4090 file = anon_inode_getfile(name, &kvm_vcpu_stats_fops, vcpu, O_RDONLY);
4093 return PTR_ERR(file);
4096 kvm_get_kvm(vcpu->kvm);
4098 file->f_mode |= FMODE_PREAD;
4099 fd_install(fd, file);
4104 static long kvm_vcpu_ioctl(struct file *filp,
4105 unsigned int ioctl, unsigned long arg)
4107 struct kvm_vcpu *vcpu = filp->private_data;
4108 void __user *argp = (void __user *)arg;
4110 struct kvm_fpu *fpu = NULL;
4111 struct kvm_sregs *kvm_sregs = NULL;
4113 if (vcpu->kvm->mm != current->mm || vcpu->kvm->vm_dead)
4116 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
4120 * Some architectures have vcpu ioctls that are asynchronous to vcpu
4121 * execution; mutex_lock() would break them.
4123 r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg);
4124 if (r != -ENOIOCTLCMD)
4127 if (mutex_lock_killable(&vcpu->mutex))
4135 oldpid = rcu_access_pointer(vcpu->pid);
4136 if (unlikely(oldpid != task_pid(current))) {
4137 /* The thread running this VCPU changed. */
4140 r = kvm_arch_vcpu_run_pid_change(vcpu);
4144 newpid = get_task_pid(current, PIDTYPE_PID);
4145 rcu_assign_pointer(vcpu->pid, newpid);
4150 r = kvm_arch_vcpu_ioctl_run(vcpu);
4151 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
4154 case KVM_GET_REGS: {
4155 struct kvm_regs *kvm_regs;
4158 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL_ACCOUNT);
4161 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
4165 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
4172 case KVM_SET_REGS: {
4173 struct kvm_regs *kvm_regs;
4175 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
4176 if (IS_ERR(kvm_regs)) {
4177 r = PTR_ERR(kvm_regs);
4180 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
4184 case KVM_GET_SREGS: {
4185 kvm_sregs = kzalloc(sizeof(struct kvm_sregs),
4186 GFP_KERNEL_ACCOUNT);
4190 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
4194 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
4199 case KVM_SET_SREGS: {
4200 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
4201 if (IS_ERR(kvm_sregs)) {
4202 r = PTR_ERR(kvm_sregs);
4206 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
4209 case KVM_GET_MP_STATE: {
4210 struct kvm_mp_state mp_state;
4212 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
4216 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
4221 case KVM_SET_MP_STATE: {
4222 struct kvm_mp_state mp_state;
4225 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
4227 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
4230 case KVM_TRANSLATE: {
4231 struct kvm_translation tr;
4234 if (copy_from_user(&tr, argp, sizeof(tr)))
4236 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
4240 if (copy_to_user(argp, &tr, sizeof(tr)))
4245 case KVM_SET_GUEST_DEBUG: {
4246 struct kvm_guest_debug dbg;
4249 if (copy_from_user(&dbg, argp, sizeof(dbg)))
4251 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
4254 case KVM_SET_SIGNAL_MASK: {
4255 struct kvm_signal_mask __user *sigmask_arg = argp;
4256 struct kvm_signal_mask kvm_sigmask;
4257 sigset_t sigset, *p;
4262 if (copy_from_user(&kvm_sigmask, argp,
4263 sizeof(kvm_sigmask)))
4266 if (kvm_sigmask.len != sizeof(sigset))
4269 if (copy_from_user(&sigset, sigmask_arg->sigset,
4274 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
4278 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL_ACCOUNT);
4282 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
4286 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
4292 fpu = memdup_user(argp, sizeof(*fpu));
4298 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
4301 case KVM_GET_STATS_FD: {
4302 r = kvm_vcpu_ioctl_get_stats_fd(vcpu);
4306 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
4309 mutex_unlock(&vcpu->mutex);
4315 #ifdef CONFIG_KVM_COMPAT
4316 static long kvm_vcpu_compat_ioctl(struct file *filp,
4317 unsigned int ioctl, unsigned long arg)
4319 struct kvm_vcpu *vcpu = filp->private_data;
4320 void __user *argp = compat_ptr(arg);
4323 if (vcpu->kvm->mm != current->mm || vcpu->kvm->vm_dead)
4327 case KVM_SET_SIGNAL_MASK: {
4328 struct kvm_signal_mask __user *sigmask_arg = argp;
4329 struct kvm_signal_mask kvm_sigmask;
4334 if (copy_from_user(&kvm_sigmask, argp,
4335 sizeof(kvm_sigmask)))
4338 if (kvm_sigmask.len != sizeof(compat_sigset_t))
4341 if (get_compat_sigset(&sigset,
4342 (compat_sigset_t __user *)sigmask_arg->sigset))
4344 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
4346 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
4350 r = kvm_vcpu_ioctl(filp, ioctl, arg);
4358 static int kvm_device_mmap(struct file *filp, struct vm_area_struct *vma)
4360 struct kvm_device *dev = filp->private_data;
4363 return dev->ops->mmap(dev, vma);
4368 static int kvm_device_ioctl_attr(struct kvm_device *dev,
4369 int (*accessor)(struct kvm_device *dev,
4370 struct kvm_device_attr *attr),
4373 struct kvm_device_attr attr;
4378 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
4381 return accessor(dev, &attr);
4384 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
4387 struct kvm_device *dev = filp->private_data;
4389 if (dev->kvm->mm != current->mm || dev->kvm->vm_dead)
4393 case KVM_SET_DEVICE_ATTR:
4394 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
4395 case KVM_GET_DEVICE_ATTR:
4396 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
4397 case KVM_HAS_DEVICE_ATTR:
4398 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
4400 if (dev->ops->ioctl)
4401 return dev->ops->ioctl(dev, ioctl, arg);
4407 static int kvm_device_release(struct inode *inode, struct file *filp)
4409 struct kvm_device *dev = filp->private_data;
4410 struct kvm *kvm = dev->kvm;
4412 if (dev->ops->release) {
4413 mutex_lock(&kvm->lock);
4414 list_del(&dev->vm_node);
4415 dev->ops->release(dev);
4416 mutex_unlock(&kvm->lock);
4423 static const struct file_operations kvm_device_fops = {
4424 .unlocked_ioctl = kvm_device_ioctl,
4425 .release = kvm_device_release,
4426 KVM_COMPAT(kvm_device_ioctl),
4427 .mmap = kvm_device_mmap,
4430 struct kvm_device *kvm_device_from_filp(struct file *filp)
4432 if (filp->f_op != &kvm_device_fops)
4435 return filp->private_data;
4438 static const struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
4439 #ifdef CONFIG_KVM_MPIC
4440 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
4441 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
4445 int kvm_register_device_ops(const struct kvm_device_ops *ops, u32 type)
4447 if (type >= ARRAY_SIZE(kvm_device_ops_table))
4450 if (kvm_device_ops_table[type] != NULL)
4453 kvm_device_ops_table[type] = ops;
4457 void kvm_unregister_device_ops(u32 type)
4459 if (kvm_device_ops_table[type] != NULL)
4460 kvm_device_ops_table[type] = NULL;
4463 static int kvm_ioctl_create_device(struct kvm *kvm,
4464 struct kvm_create_device *cd)
4466 const struct kvm_device_ops *ops;
4467 struct kvm_device *dev;
4468 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
4472 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
4475 type = array_index_nospec(cd->type, ARRAY_SIZE(kvm_device_ops_table));
4476 ops = kvm_device_ops_table[type];
4483 dev = kzalloc(sizeof(*dev), GFP_KERNEL_ACCOUNT);
4490 mutex_lock(&kvm->lock);
4491 ret = ops->create(dev, type);
4493 mutex_unlock(&kvm->lock);
4497 list_add(&dev->vm_node, &kvm->devices);
4498 mutex_unlock(&kvm->lock);
4504 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
4506 kvm_put_kvm_no_destroy(kvm);
4507 mutex_lock(&kvm->lock);
4508 list_del(&dev->vm_node);
4511 mutex_unlock(&kvm->lock);
4521 static int kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
4524 case KVM_CAP_USER_MEMORY:
4525 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
4526 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
4527 case KVM_CAP_INTERNAL_ERROR_DATA:
4528 #ifdef CONFIG_HAVE_KVM_MSI
4529 case KVM_CAP_SIGNAL_MSI:
4531 #ifdef CONFIG_HAVE_KVM_IRQFD
4534 case KVM_CAP_IOEVENTFD_ANY_LENGTH:
4535 case KVM_CAP_CHECK_EXTENSION_VM:
4536 case KVM_CAP_ENABLE_CAP_VM:
4537 case KVM_CAP_HALT_POLL:
4539 #ifdef CONFIG_KVM_MMIO
4540 case KVM_CAP_COALESCED_MMIO:
4541 return KVM_COALESCED_MMIO_PAGE_OFFSET;
4542 case KVM_CAP_COALESCED_PIO:
4545 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4546 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2:
4547 return KVM_DIRTY_LOG_MANUAL_CAPS;
4549 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
4550 case KVM_CAP_IRQ_ROUTING:
4551 return KVM_MAX_IRQ_ROUTES;
4553 #if KVM_ADDRESS_SPACE_NUM > 1
4554 case KVM_CAP_MULTI_ADDRESS_SPACE:
4555 return KVM_ADDRESS_SPACE_NUM;
4557 case KVM_CAP_NR_MEMSLOTS:
4558 return KVM_USER_MEM_SLOTS;
4559 case KVM_CAP_DIRTY_LOG_RING:
4560 #ifdef CONFIG_HAVE_KVM_DIRTY_RING_TSO
4561 return KVM_DIRTY_RING_MAX_ENTRIES * sizeof(struct kvm_dirty_gfn);
4565 case KVM_CAP_DIRTY_LOG_RING_ACQ_REL:
4566 #ifdef CONFIG_HAVE_KVM_DIRTY_RING_ACQ_REL
4567 return KVM_DIRTY_RING_MAX_ENTRIES * sizeof(struct kvm_dirty_gfn);
4571 #ifdef CONFIG_NEED_KVM_DIRTY_RING_WITH_BITMAP
4572 case KVM_CAP_DIRTY_LOG_RING_WITH_BITMAP:
4574 case KVM_CAP_BINARY_STATS_FD:
4575 case KVM_CAP_SYSTEM_EVENT_DATA:
4580 return kvm_vm_ioctl_check_extension(kvm, arg);
4583 static int kvm_vm_ioctl_enable_dirty_log_ring(struct kvm *kvm, u32 size)
4587 if (!KVM_DIRTY_LOG_PAGE_OFFSET)
4590 /* the size should be power of 2 */
4591 if (!size || (size & (size - 1)))
4594 /* Should be bigger to keep the reserved entries, or a page */
4595 if (size < kvm_dirty_ring_get_rsvd_entries() *
4596 sizeof(struct kvm_dirty_gfn) || size < PAGE_SIZE)
4599 if (size > KVM_DIRTY_RING_MAX_ENTRIES *
4600 sizeof(struct kvm_dirty_gfn))
4603 /* We only allow it to set once */
4604 if (kvm->dirty_ring_size)
4607 mutex_lock(&kvm->lock);
4609 if (kvm->created_vcpus) {
4610 /* We don't allow to change this value after vcpu created */
4613 kvm->dirty_ring_size = size;
4617 mutex_unlock(&kvm->lock);
4621 static int kvm_vm_ioctl_reset_dirty_pages(struct kvm *kvm)
4624 struct kvm_vcpu *vcpu;
4627 if (!kvm->dirty_ring_size)
4630 mutex_lock(&kvm->slots_lock);
4632 kvm_for_each_vcpu(i, vcpu, kvm)
4633 cleared += kvm_dirty_ring_reset(vcpu->kvm, &vcpu->dirty_ring);
4635 mutex_unlock(&kvm->slots_lock);
4638 kvm_flush_remote_tlbs(kvm);
4643 int __attribute__((weak)) kvm_vm_ioctl_enable_cap(struct kvm *kvm,
4644 struct kvm_enable_cap *cap)
4649 bool kvm_are_all_memslots_empty(struct kvm *kvm)
4653 lockdep_assert_held(&kvm->slots_lock);
4655 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
4656 if (!kvm_memslots_empty(__kvm_memslots(kvm, i)))
4662 EXPORT_SYMBOL_GPL(kvm_are_all_memslots_empty);
4664 static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm,
4665 struct kvm_enable_cap *cap)
4668 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4669 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2: {
4670 u64 allowed_options = KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE;
4672 if (cap->args[0] & KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE)
4673 allowed_options = KVM_DIRTY_LOG_MANUAL_CAPS;
4675 if (cap->flags || (cap->args[0] & ~allowed_options))
4677 kvm->manual_dirty_log_protect = cap->args[0];
4681 case KVM_CAP_HALT_POLL: {
4682 if (cap->flags || cap->args[0] != (unsigned int)cap->args[0])
4685 kvm->max_halt_poll_ns = cap->args[0];
4688 * Ensure kvm->override_halt_poll_ns does not become visible
4689 * before kvm->max_halt_poll_ns.
4691 * Pairs with the smp_rmb() in kvm_vcpu_max_halt_poll_ns().
4694 kvm->override_halt_poll_ns = true;
4698 case KVM_CAP_DIRTY_LOG_RING:
4699 case KVM_CAP_DIRTY_LOG_RING_ACQ_REL:
4700 if (!kvm_vm_ioctl_check_extension_generic(kvm, cap->cap))
4703 return kvm_vm_ioctl_enable_dirty_log_ring(kvm, cap->args[0]);
4704 case KVM_CAP_DIRTY_LOG_RING_WITH_BITMAP: {
4707 if (!IS_ENABLED(CONFIG_NEED_KVM_DIRTY_RING_WITH_BITMAP) ||
4708 !kvm->dirty_ring_size || cap->flags)
4711 mutex_lock(&kvm->slots_lock);
4714 * For simplicity, allow enabling ring+bitmap if and only if
4715 * there are no memslots, e.g. to ensure all memslots allocate
4716 * a bitmap after the capability is enabled.
4718 if (kvm_are_all_memslots_empty(kvm)) {
4719 kvm->dirty_ring_with_bitmap = true;
4723 mutex_unlock(&kvm->slots_lock);
4728 return kvm_vm_ioctl_enable_cap(kvm, cap);
4732 static ssize_t kvm_vm_stats_read(struct file *file, char __user *user_buffer,
4733 size_t size, loff_t *offset)
4735 struct kvm *kvm = file->private_data;
4737 return kvm_stats_read(kvm->stats_id, &kvm_vm_stats_header,
4738 &kvm_vm_stats_desc[0], &kvm->stat,
4739 sizeof(kvm->stat), user_buffer, size, offset);
4742 static int kvm_vm_stats_release(struct inode *inode, struct file *file)
4744 struct kvm *kvm = file->private_data;
4750 static const struct file_operations kvm_vm_stats_fops = {
4751 .read = kvm_vm_stats_read,
4752 .release = kvm_vm_stats_release,
4753 .llseek = noop_llseek,
4756 static int kvm_vm_ioctl_get_stats_fd(struct kvm *kvm)
4761 fd = get_unused_fd_flags(O_CLOEXEC);
4765 file = anon_inode_getfile("kvm-vm-stats",
4766 &kvm_vm_stats_fops, kvm, O_RDONLY);
4769 return PTR_ERR(file);
4774 file->f_mode |= FMODE_PREAD;
4775 fd_install(fd, file);
4780 static long kvm_vm_ioctl(struct file *filp,
4781 unsigned int ioctl, unsigned long arg)
4783 struct kvm *kvm = filp->private_data;
4784 void __user *argp = (void __user *)arg;
4787 if (kvm->mm != current->mm || kvm->vm_dead)
4790 case KVM_CREATE_VCPU:
4791 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
4793 case KVM_ENABLE_CAP: {
4794 struct kvm_enable_cap cap;
4797 if (copy_from_user(&cap, argp, sizeof(cap)))
4799 r = kvm_vm_ioctl_enable_cap_generic(kvm, &cap);
4802 case KVM_SET_USER_MEMORY_REGION: {
4803 struct kvm_userspace_memory_region kvm_userspace_mem;
4806 if (copy_from_user(&kvm_userspace_mem, argp,
4807 sizeof(kvm_userspace_mem)))
4810 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
4813 case KVM_GET_DIRTY_LOG: {
4814 struct kvm_dirty_log log;
4817 if (copy_from_user(&log, argp, sizeof(log)))
4819 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
4822 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4823 case KVM_CLEAR_DIRTY_LOG: {
4824 struct kvm_clear_dirty_log log;
4827 if (copy_from_user(&log, argp, sizeof(log)))
4829 r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
4833 #ifdef CONFIG_KVM_MMIO
4834 case KVM_REGISTER_COALESCED_MMIO: {
4835 struct kvm_coalesced_mmio_zone zone;
4838 if (copy_from_user(&zone, argp, sizeof(zone)))
4840 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
4843 case KVM_UNREGISTER_COALESCED_MMIO: {
4844 struct kvm_coalesced_mmio_zone zone;
4847 if (copy_from_user(&zone, argp, sizeof(zone)))
4849 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
4854 struct kvm_irqfd data;
4857 if (copy_from_user(&data, argp, sizeof(data)))
4859 r = kvm_irqfd(kvm, &data);
4862 case KVM_IOEVENTFD: {
4863 struct kvm_ioeventfd data;
4866 if (copy_from_user(&data, argp, sizeof(data)))
4868 r = kvm_ioeventfd(kvm, &data);
4871 #ifdef CONFIG_HAVE_KVM_MSI
4872 case KVM_SIGNAL_MSI: {
4876 if (copy_from_user(&msi, argp, sizeof(msi)))
4878 r = kvm_send_userspace_msi(kvm, &msi);
4882 #ifdef __KVM_HAVE_IRQ_LINE
4883 case KVM_IRQ_LINE_STATUS:
4884 case KVM_IRQ_LINE: {
4885 struct kvm_irq_level irq_event;
4888 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
4891 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
4892 ioctl == KVM_IRQ_LINE_STATUS);
4897 if (ioctl == KVM_IRQ_LINE_STATUS) {
4898 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
4906 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
4907 case KVM_SET_GSI_ROUTING: {
4908 struct kvm_irq_routing routing;
4909 struct kvm_irq_routing __user *urouting;
4910 struct kvm_irq_routing_entry *entries = NULL;
4913 if (copy_from_user(&routing, argp, sizeof(routing)))
4916 if (!kvm_arch_can_set_irq_routing(kvm))
4918 if (routing.nr > KVM_MAX_IRQ_ROUTES)
4924 entries = vmemdup_user(urouting->entries,
4925 array_size(sizeof(*entries),
4927 if (IS_ERR(entries)) {
4928 r = PTR_ERR(entries);
4932 r = kvm_set_irq_routing(kvm, entries, routing.nr,
4937 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
4938 case KVM_CREATE_DEVICE: {
4939 struct kvm_create_device cd;
4942 if (copy_from_user(&cd, argp, sizeof(cd)))
4945 r = kvm_ioctl_create_device(kvm, &cd);
4950 if (copy_to_user(argp, &cd, sizeof(cd)))
4956 case KVM_CHECK_EXTENSION:
4957 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
4959 case KVM_RESET_DIRTY_RINGS:
4960 r = kvm_vm_ioctl_reset_dirty_pages(kvm);
4962 case KVM_GET_STATS_FD:
4963 r = kvm_vm_ioctl_get_stats_fd(kvm);
4966 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
4972 #ifdef CONFIG_KVM_COMPAT
4973 struct compat_kvm_dirty_log {
4977 compat_uptr_t dirty_bitmap; /* one bit per page */
4982 struct compat_kvm_clear_dirty_log {
4987 compat_uptr_t dirty_bitmap; /* one bit per page */
4992 long __weak kvm_arch_vm_compat_ioctl(struct file *filp, unsigned int ioctl,
4998 static long kvm_vm_compat_ioctl(struct file *filp,
4999 unsigned int ioctl, unsigned long arg)
5001 struct kvm *kvm = filp->private_data;
5004 if (kvm->mm != current->mm || kvm->vm_dead)
5007 r = kvm_arch_vm_compat_ioctl(filp, ioctl, arg);
5012 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
5013 case KVM_CLEAR_DIRTY_LOG: {
5014 struct compat_kvm_clear_dirty_log compat_log;
5015 struct kvm_clear_dirty_log log;
5017 if (copy_from_user(&compat_log, (void __user *)arg,
5018 sizeof(compat_log)))
5020 log.slot = compat_log.slot;
5021 log.num_pages = compat_log.num_pages;
5022 log.first_page = compat_log.first_page;
5023 log.padding2 = compat_log.padding2;
5024 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
5026 r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
5030 case KVM_GET_DIRTY_LOG: {
5031 struct compat_kvm_dirty_log compat_log;
5032 struct kvm_dirty_log log;
5034 if (copy_from_user(&compat_log, (void __user *)arg,
5035 sizeof(compat_log)))
5037 log.slot = compat_log.slot;
5038 log.padding1 = compat_log.padding1;
5039 log.padding2 = compat_log.padding2;
5040 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
5042 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
5046 r = kvm_vm_ioctl(filp, ioctl, arg);
5052 static const struct file_operations kvm_vm_fops = {
5053 .release = kvm_vm_release,
5054 .unlocked_ioctl = kvm_vm_ioctl,
5055 .llseek = noop_llseek,
5056 KVM_COMPAT(kvm_vm_compat_ioctl),
5059 bool file_is_kvm(struct file *file)
5061 return file && file->f_op == &kvm_vm_fops;
5063 EXPORT_SYMBOL_GPL(file_is_kvm);
5065 static int kvm_dev_ioctl_create_vm(unsigned long type)
5067 char fdname[ITOA_MAX_LEN + 1];
5072 fd = get_unused_fd_flags(O_CLOEXEC);
5076 snprintf(fdname, sizeof(fdname), "%d", fd);
5078 kvm = kvm_create_vm(type, fdname);
5084 file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
5091 * Don't call kvm_put_kvm anymore at this point; file->f_op is
5092 * already set, with ->release() being kvm_vm_release(). In error
5093 * cases it will be called by the final fput(file) and will take
5094 * care of doing kvm_put_kvm(kvm).
5096 kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm);
5098 fd_install(fd, file);
5108 static long kvm_dev_ioctl(struct file *filp,
5109 unsigned int ioctl, unsigned long arg)
5114 case KVM_GET_API_VERSION:
5117 r = KVM_API_VERSION;
5120 r = kvm_dev_ioctl_create_vm(arg);
5122 case KVM_CHECK_EXTENSION:
5123 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
5125 case KVM_GET_VCPU_MMAP_SIZE:
5128 r = PAGE_SIZE; /* struct kvm_run */
5130 r += PAGE_SIZE; /* pio data page */
5132 #ifdef CONFIG_KVM_MMIO
5133 r += PAGE_SIZE; /* coalesced mmio ring page */
5136 case KVM_TRACE_ENABLE:
5137 case KVM_TRACE_PAUSE:
5138 case KVM_TRACE_DISABLE:
5142 return kvm_arch_dev_ioctl(filp, ioctl, arg);
5148 static struct file_operations kvm_chardev_ops = {
5149 .unlocked_ioctl = kvm_dev_ioctl,
5150 .llseek = noop_llseek,
5151 KVM_COMPAT(kvm_dev_ioctl),
5154 static struct miscdevice kvm_dev = {
5160 #ifdef CONFIG_KVM_GENERIC_HARDWARE_ENABLING
5161 __visible bool kvm_rebooting;
5162 EXPORT_SYMBOL_GPL(kvm_rebooting);
5164 static DEFINE_PER_CPU(bool, hardware_enabled);
5165 static int kvm_usage_count;
5167 static int __hardware_enable_nolock(void)
5169 if (__this_cpu_read(hardware_enabled))
5172 if (kvm_arch_hardware_enable()) {
5173 pr_info("kvm: enabling virtualization on CPU%d failed\n",
5174 raw_smp_processor_id());
5178 __this_cpu_write(hardware_enabled, true);
5182 static void hardware_enable_nolock(void *failed)
5184 if (__hardware_enable_nolock())
5188 static int kvm_online_cpu(unsigned int cpu)
5193 * Abort the CPU online process if hardware virtualization cannot
5194 * be enabled. Otherwise running VMs would encounter unrecoverable
5195 * errors when scheduled to this CPU.
5197 mutex_lock(&kvm_lock);
5198 if (kvm_usage_count)
5199 ret = __hardware_enable_nolock();
5200 mutex_unlock(&kvm_lock);
5204 static void hardware_disable_nolock(void *junk)
5207 * Note, hardware_disable_all_nolock() tells all online CPUs to disable
5208 * hardware, not just CPUs that successfully enabled hardware!
5210 if (!__this_cpu_read(hardware_enabled))
5213 kvm_arch_hardware_disable();
5215 __this_cpu_write(hardware_enabled, false);
5218 static int kvm_offline_cpu(unsigned int cpu)
5220 mutex_lock(&kvm_lock);
5221 if (kvm_usage_count)
5222 hardware_disable_nolock(NULL);
5223 mutex_unlock(&kvm_lock);
5227 static void hardware_disable_all_nolock(void)
5229 BUG_ON(!kvm_usage_count);
5232 if (!kvm_usage_count)
5233 on_each_cpu(hardware_disable_nolock, NULL, 1);
5236 static void hardware_disable_all(void)
5239 mutex_lock(&kvm_lock);
5240 hardware_disable_all_nolock();
5241 mutex_unlock(&kvm_lock);
5245 static int hardware_enable_all(void)
5247 atomic_t failed = ATOMIC_INIT(0);
5251 * Do not enable hardware virtualization if the system is going down.
5252 * If userspace initiated a forced reboot, e.g. reboot -f, then it's
5253 * possible for an in-flight KVM_CREATE_VM to trigger hardware enabling
5254 * after kvm_reboot() is called. Note, this relies on system_state
5255 * being set _before_ kvm_reboot(), which is why KVM uses a syscore ops
5256 * hook instead of registering a dedicated reboot notifier (the latter
5257 * runs before system_state is updated).
5259 if (system_state == SYSTEM_HALT || system_state == SYSTEM_POWER_OFF ||
5260 system_state == SYSTEM_RESTART)
5264 * When onlining a CPU, cpu_online_mask is set before kvm_online_cpu()
5265 * is called, and so on_each_cpu() between them includes the CPU that
5266 * is being onlined. As a result, hardware_enable_nolock() may get
5267 * invoked before kvm_online_cpu(), which also enables hardware if the
5268 * usage count is non-zero. Disable CPU hotplug to avoid attempting to
5269 * enable hardware multiple times.
5272 mutex_lock(&kvm_lock);
5277 if (kvm_usage_count == 1) {
5278 on_each_cpu(hardware_enable_nolock, &failed, 1);
5280 if (atomic_read(&failed)) {
5281 hardware_disable_all_nolock();
5286 mutex_unlock(&kvm_lock);
5292 static void kvm_shutdown(void)
5295 * Disable hardware virtualization and set kvm_rebooting to indicate
5296 * that KVM has asynchronously disabled hardware virtualization, i.e.
5297 * that relevant errors and exceptions aren't entirely unexpected.
5298 * Some flavors of hardware virtualization need to be disabled before
5299 * transferring control to firmware (to perform shutdown/reboot), e.g.
5300 * on x86, virtualization can block INIT interrupts, which are used by
5301 * firmware to pull APs back under firmware control. Note, this path
5302 * is used for both shutdown and reboot scenarios, i.e. neither name is
5303 * 100% comprehensive.
5305 pr_info("kvm: exiting hardware virtualization\n");
5306 kvm_rebooting = true;
5307 on_each_cpu(hardware_disable_nolock, NULL, 1);
5310 static int kvm_suspend(void)
5313 * Secondary CPUs and CPU hotplug are disabled across the suspend/resume
5314 * callbacks, i.e. no need to acquire kvm_lock to ensure the usage count
5315 * is stable. Assert that kvm_lock is not held to ensure the system
5316 * isn't suspended while KVM is enabling hardware. Hardware enabling
5317 * can be preempted, but the task cannot be frozen until it has dropped
5318 * all locks (userspace tasks are frozen via a fake signal).
5320 lockdep_assert_not_held(&kvm_lock);
5321 lockdep_assert_irqs_disabled();
5323 if (kvm_usage_count)
5324 hardware_disable_nolock(NULL);
5328 static void kvm_resume(void)
5330 lockdep_assert_not_held(&kvm_lock);
5331 lockdep_assert_irqs_disabled();
5333 if (kvm_usage_count)
5334 WARN_ON_ONCE(__hardware_enable_nolock());
5337 static struct syscore_ops kvm_syscore_ops = {
5338 .suspend = kvm_suspend,
5339 .resume = kvm_resume,
5340 .shutdown = kvm_shutdown,
5342 #else /* CONFIG_KVM_GENERIC_HARDWARE_ENABLING */
5343 static int hardware_enable_all(void)
5348 static void hardware_disable_all(void)
5352 #endif /* CONFIG_KVM_GENERIC_HARDWARE_ENABLING */
5354 static void kvm_iodevice_destructor(struct kvm_io_device *dev)
5356 if (dev->ops->destructor)
5357 dev->ops->destructor(dev);
5360 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
5364 for (i = 0; i < bus->dev_count; i++) {
5365 struct kvm_io_device *pos = bus->range[i].dev;
5367 kvm_iodevice_destructor(pos);
5372 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
5373 const struct kvm_io_range *r2)
5375 gpa_t addr1 = r1->addr;
5376 gpa_t addr2 = r2->addr;
5381 /* If r2->len == 0, match the exact address. If r2->len != 0,
5382 * accept any overlapping write. Any order is acceptable for
5383 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
5384 * we process all of them.
5397 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
5399 return kvm_io_bus_cmp(p1, p2);
5402 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
5403 gpa_t addr, int len)
5405 struct kvm_io_range *range, key;
5408 key = (struct kvm_io_range) {
5413 range = bsearch(&key, bus->range, bus->dev_count,
5414 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
5418 off = range - bus->range;
5420 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
5426 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
5427 struct kvm_io_range *range, const void *val)
5431 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
5435 while (idx < bus->dev_count &&
5436 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
5437 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
5446 /* kvm_io_bus_write - called under kvm->slots_lock */
5447 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
5448 int len, const void *val)
5450 struct kvm_io_bus *bus;
5451 struct kvm_io_range range;
5454 range = (struct kvm_io_range) {
5459 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
5462 r = __kvm_io_bus_write(vcpu, bus, &range, val);
5463 return r < 0 ? r : 0;
5465 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
5467 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
5468 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
5469 gpa_t addr, int len, const void *val, long cookie)
5471 struct kvm_io_bus *bus;
5472 struct kvm_io_range range;
5474 range = (struct kvm_io_range) {
5479 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
5483 /* First try the device referenced by cookie. */
5484 if ((cookie >= 0) && (cookie < bus->dev_count) &&
5485 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
5486 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
5491 * cookie contained garbage; fall back to search and return the
5492 * correct cookie value.
5494 return __kvm_io_bus_write(vcpu, bus, &range, val);
5497 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
5498 struct kvm_io_range *range, void *val)
5502 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
5506 while (idx < bus->dev_count &&
5507 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
5508 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
5517 /* kvm_io_bus_read - called under kvm->slots_lock */
5518 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
5521 struct kvm_io_bus *bus;
5522 struct kvm_io_range range;
5525 range = (struct kvm_io_range) {
5530 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
5533 r = __kvm_io_bus_read(vcpu, bus, &range, val);
5534 return r < 0 ? r : 0;
5537 /* Caller must hold slots_lock. */
5538 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
5539 int len, struct kvm_io_device *dev)
5542 struct kvm_io_bus *new_bus, *bus;
5543 struct kvm_io_range range;
5545 bus = kvm_get_bus(kvm, bus_idx);
5549 /* exclude ioeventfd which is limited by maximum fd */
5550 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
5553 new_bus = kmalloc(struct_size(bus, range, bus->dev_count + 1),
5554 GFP_KERNEL_ACCOUNT);
5558 range = (struct kvm_io_range) {
5564 for (i = 0; i < bus->dev_count; i++)
5565 if (kvm_io_bus_cmp(&bus->range[i], &range) > 0)
5568 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
5569 new_bus->dev_count++;
5570 new_bus->range[i] = range;
5571 memcpy(new_bus->range + i + 1, bus->range + i,
5572 (bus->dev_count - i) * sizeof(struct kvm_io_range));
5573 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
5574 synchronize_srcu_expedited(&kvm->srcu);
5580 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
5581 struct kvm_io_device *dev)
5584 struct kvm_io_bus *new_bus, *bus;
5586 lockdep_assert_held(&kvm->slots_lock);
5588 bus = kvm_get_bus(kvm, bus_idx);
5592 for (i = 0; i < bus->dev_count; i++) {
5593 if (bus->range[i].dev == dev) {
5598 if (i == bus->dev_count)
5601 new_bus = kmalloc(struct_size(bus, range, bus->dev_count - 1),
5602 GFP_KERNEL_ACCOUNT);
5604 memcpy(new_bus, bus, struct_size(bus, range, i));
5605 new_bus->dev_count--;
5606 memcpy(new_bus->range + i, bus->range + i + 1,
5607 flex_array_size(new_bus, range, new_bus->dev_count - i));
5610 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
5611 synchronize_srcu_expedited(&kvm->srcu);
5614 * If NULL bus is installed, destroy the old bus, including all the
5615 * attached devices. Otherwise, destroy the caller's device only.
5618 pr_err("kvm: failed to shrink bus, removing it completely\n");
5619 kvm_io_bus_destroy(bus);
5623 kvm_iodevice_destructor(dev);
5628 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
5631 struct kvm_io_bus *bus;
5632 int dev_idx, srcu_idx;
5633 struct kvm_io_device *iodev = NULL;
5635 srcu_idx = srcu_read_lock(&kvm->srcu);
5637 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
5641 dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
5645 iodev = bus->range[dev_idx].dev;
5648 srcu_read_unlock(&kvm->srcu, srcu_idx);
5652 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
5654 static int kvm_debugfs_open(struct inode *inode, struct file *file,
5655 int (*get)(void *, u64 *), int (*set)(void *, u64),
5659 struct kvm_stat_data *stat_data = inode->i_private;
5662 * The debugfs files are a reference to the kvm struct which
5663 * is still valid when kvm_destroy_vm is called. kvm_get_kvm_safe
5664 * avoids the race between open and the removal of the debugfs directory.
5666 if (!kvm_get_kvm_safe(stat_data->kvm))
5669 ret = simple_attr_open(inode, file, get,
5670 kvm_stats_debugfs_mode(stat_data->desc) & 0222
5673 kvm_put_kvm(stat_data->kvm);
5678 static int kvm_debugfs_release(struct inode *inode, struct file *file)
5680 struct kvm_stat_data *stat_data = inode->i_private;
5682 simple_attr_release(inode, file);
5683 kvm_put_kvm(stat_data->kvm);
5688 static int kvm_get_stat_per_vm(struct kvm *kvm, size_t offset, u64 *val)
5690 *val = *(u64 *)((void *)(&kvm->stat) + offset);
5695 static int kvm_clear_stat_per_vm(struct kvm *kvm, size_t offset)
5697 *(u64 *)((void *)(&kvm->stat) + offset) = 0;
5702 static int kvm_get_stat_per_vcpu(struct kvm *kvm, size_t offset, u64 *val)
5705 struct kvm_vcpu *vcpu;
5709 kvm_for_each_vcpu(i, vcpu, kvm)
5710 *val += *(u64 *)((void *)(&vcpu->stat) + offset);
5715 static int kvm_clear_stat_per_vcpu(struct kvm *kvm, size_t offset)
5718 struct kvm_vcpu *vcpu;
5720 kvm_for_each_vcpu(i, vcpu, kvm)
5721 *(u64 *)((void *)(&vcpu->stat) + offset) = 0;
5726 static int kvm_stat_data_get(void *data, u64 *val)
5729 struct kvm_stat_data *stat_data = data;
5731 switch (stat_data->kind) {
5733 r = kvm_get_stat_per_vm(stat_data->kvm,
5734 stat_data->desc->desc.offset, val);
5737 r = kvm_get_stat_per_vcpu(stat_data->kvm,
5738 stat_data->desc->desc.offset, val);
5745 static int kvm_stat_data_clear(void *data, u64 val)
5748 struct kvm_stat_data *stat_data = data;
5753 switch (stat_data->kind) {
5755 r = kvm_clear_stat_per_vm(stat_data->kvm,
5756 stat_data->desc->desc.offset);
5759 r = kvm_clear_stat_per_vcpu(stat_data->kvm,
5760 stat_data->desc->desc.offset);
5767 static int kvm_stat_data_open(struct inode *inode, struct file *file)
5769 __simple_attr_check_format("%llu\n", 0ull);
5770 return kvm_debugfs_open(inode, file, kvm_stat_data_get,
5771 kvm_stat_data_clear, "%llu\n");
5774 static const struct file_operations stat_fops_per_vm = {
5775 .owner = THIS_MODULE,
5776 .open = kvm_stat_data_open,
5777 .release = kvm_debugfs_release,
5778 .read = simple_attr_read,
5779 .write = simple_attr_write,
5780 .llseek = no_llseek,
5783 static int vm_stat_get(void *_offset, u64 *val)
5785 unsigned offset = (long)_offset;
5790 mutex_lock(&kvm_lock);
5791 list_for_each_entry(kvm, &vm_list, vm_list) {
5792 kvm_get_stat_per_vm(kvm, offset, &tmp_val);
5795 mutex_unlock(&kvm_lock);
5799 static int vm_stat_clear(void *_offset, u64 val)
5801 unsigned offset = (long)_offset;
5807 mutex_lock(&kvm_lock);
5808 list_for_each_entry(kvm, &vm_list, vm_list) {
5809 kvm_clear_stat_per_vm(kvm, offset);
5811 mutex_unlock(&kvm_lock);
5816 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n");
5817 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_readonly_fops, vm_stat_get, NULL, "%llu\n");
5819 static int vcpu_stat_get(void *_offset, u64 *val)
5821 unsigned offset = (long)_offset;
5826 mutex_lock(&kvm_lock);
5827 list_for_each_entry(kvm, &vm_list, vm_list) {
5828 kvm_get_stat_per_vcpu(kvm, offset, &tmp_val);
5831 mutex_unlock(&kvm_lock);
5835 static int vcpu_stat_clear(void *_offset, u64 val)
5837 unsigned offset = (long)_offset;
5843 mutex_lock(&kvm_lock);
5844 list_for_each_entry(kvm, &vm_list, vm_list) {
5845 kvm_clear_stat_per_vcpu(kvm, offset);
5847 mutex_unlock(&kvm_lock);
5852 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear,
5854 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_readonly_fops, vcpu_stat_get, NULL, "%llu\n");
5856 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm)
5858 struct kobj_uevent_env *env;
5859 unsigned long long created, active;
5861 if (!kvm_dev.this_device || !kvm)
5864 mutex_lock(&kvm_lock);
5865 if (type == KVM_EVENT_CREATE_VM) {
5866 kvm_createvm_count++;
5868 } else if (type == KVM_EVENT_DESTROY_VM) {
5871 created = kvm_createvm_count;
5872 active = kvm_active_vms;
5873 mutex_unlock(&kvm_lock);
5875 env = kzalloc(sizeof(*env), GFP_KERNEL_ACCOUNT);
5879 add_uevent_var(env, "CREATED=%llu", created);
5880 add_uevent_var(env, "COUNT=%llu", active);
5882 if (type == KVM_EVENT_CREATE_VM) {
5883 add_uevent_var(env, "EVENT=create");
5884 kvm->userspace_pid = task_pid_nr(current);
5885 } else if (type == KVM_EVENT_DESTROY_VM) {
5886 add_uevent_var(env, "EVENT=destroy");
5888 add_uevent_var(env, "PID=%d", kvm->userspace_pid);
5890 if (!IS_ERR(kvm->debugfs_dentry)) {
5891 char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL_ACCOUNT);
5894 tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX);
5896 add_uevent_var(env, "STATS_PATH=%s", tmp);
5900 /* no need for checks, since we are adding at most only 5 keys */
5901 env->envp[env->envp_idx++] = NULL;
5902 kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp);
5906 static void kvm_init_debug(void)
5908 const struct file_operations *fops;
5909 const struct _kvm_stats_desc *pdesc;
5912 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
5914 for (i = 0; i < kvm_vm_stats_header.num_desc; ++i) {
5915 pdesc = &kvm_vm_stats_desc[i];
5916 if (kvm_stats_debugfs_mode(pdesc) & 0222)
5917 fops = &vm_stat_fops;
5919 fops = &vm_stat_readonly_fops;
5920 debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
5922 (void *)(long)pdesc->desc.offset, fops);
5925 for (i = 0; i < kvm_vcpu_stats_header.num_desc; ++i) {
5926 pdesc = &kvm_vcpu_stats_desc[i];
5927 if (kvm_stats_debugfs_mode(pdesc) & 0222)
5928 fops = &vcpu_stat_fops;
5930 fops = &vcpu_stat_readonly_fops;
5931 debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
5933 (void *)(long)pdesc->desc.offset, fops);
5938 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
5940 return container_of(pn, struct kvm_vcpu, preempt_notifier);
5943 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
5945 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
5947 WRITE_ONCE(vcpu->preempted, false);
5948 WRITE_ONCE(vcpu->ready, false);
5950 __this_cpu_write(kvm_running_vcpu, vcpu);
5951 kvm_arch_sched_in(vcpu, cpu);
5952 kvm_arch_vcpu_load(vcpu, cpu);
5955 static void kvm_sched_out(struct preempt_notifier *pn,
5956 struct task_struct *next)
5958 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
5960 if (current->on_rq) {
5961 WRITE_ONCE(vcpu->preempted, true);
5962 WRITE_ONCE(vcpu->ready, true);
5964 kvm_arch_vcpu_put(vcpu);
5965 __this_cpu_write(kvm_running_vcpu, NULL);
5969 * kvm_get_running_vcpu - get the vcpu running on the current CPU.
5971 * We can disable preemption locally around accessing the per-CPU variable,
5972 * and use the resolved vcpu pointer after enabling preemption again,
5973 * because even if the current thread is migrated to another CPU, reading
5974 * the per-CPU value later will give us the same value as we update the
5975 * per-CPU variable in the preempt notifier handlers.
5977 struct kvm_vcpu *kvm_get_running_vcpu(void)
5979 struct kvm_vcpu *vcpu;
5982 vcpu = __this_cpu_read(kvm_running_vcpu);
5987 EXPORT_SYMBOL_GPL(kvm_get_running_vcpu);
5990 * kvm_get_running_vcpus - get the per-CPU array of currently running vcpus.
5992 struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void)
5994 return &kvm_running_vcpu;
5997 #ifdef CONFIG_GUEST_PERF_EVENTS
5998 static unsigned int kvm_guest_state(void)
6000 struct kvm_vcpu *vcpu = kvm_get_running_vcpu();
6003 if (!kvm_arch_pmi_in_guest(vcpu))
6006 state = PERF_GUEST_ACTIVE;
6007 if (!kvm_arch_vcpu_in_kernel(vcpu))
6008 state |= PERF_GUEST_USER;
6013 static unsigned long kvm_guest_get_ip(void)
6015 struct kvm_vcpu *vcpu = kvm_get_running_vcpu();
6017 /* Retrieving the IP must be guarded by a call to kvm_guest_state(). */
6018 if (WARN_ON_ONCE(!kvm_arch_pmi_in_guest(vcpu)))
6021 return kvm_arch_vcpu_get_ip(vcpu);
6024 static struct perf_guest_info_callbacks kvm_guest_cbs = {
6025 .state = kvm_guest_state,
6026 .get_ip = kvm_guest_get_ip,
6027 .handle_intel_pt_intr = NULL,
6030 void kvm_register_perf_callbacks(unsigned int (*pt_intr_handler)(void))
6032 kvm_guest_cbs.handle_intel_pt_intr = pt_intr_handler;
6033 perf_register_guest_info_callbacks(&kvm_guest_cbs);
6035 void kvm_unregister_perf_callbacks(void)
6037 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
6041 int kvm_init(unsigned vcpu_size, unsigned vcpu_align, struct module *module)
6046 #ifdef CONFIG_KVM_GENERIC_HARDWARE_ENABLING
6047 r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_ONLINE, "kvm/cpu:online",
6048 kvm_online_cpu, kvm_offline_cpu);
6052 register_syscore_ops(&kvm_syscore_ops);
6055 /* A kmem cache lets us meet the alignment requirements of fx_save. */
6057 vcpu_align = __alignof__(struct kvm_vcpu);
6059 kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align,
6061 offsetof(struct kvm_vcpu, arch),
6062 offsetofend(struct kvm_vcpu, stats_id)
6063 - offsetof(struct kvm_vcpu, arch),
6065 if (!kvm_vcpu_cache) {
6067 goto err_vcpu_cache;
6070 for_each_possible_cpu(cpu) {
6071 if (!alloc_cpumask_var_node(&per_cpu(cpu_kick_mask, cpu),
6072 GFP_KERNEL, cpu_to_node(cpu))) {
6074 goto err_cpu_kick_mask;
6078 r = kvm_irqfd_init();
6082 r = kvm_async_pf_init();
6086 kvm_chardev_ops.owner = module;
6088 kvm_preempt_ops.sched_in = kvm_sched_in;
6089 kvm_preempt_ops.sched_out = kvm_sched_out;
6093 r = kvm_vfio_ops_init();
6094 if (WARN_ON_ONCE(r))
6098 * Registration _must_ be the very last thing done, as this exposes
6099 * /dev/kvm to userspace, i.e. all infrastructure must be setup!
6101 r = misc_register(&kvm_dev);
6103 pr_err("kvm: misc device register failed\n");
6110 kvm_vfio_ops_exit();
6112 kvm_async_pf_deinit();
6117 for_each_possible_cpu(cpu)
6118 free_cpumask_var(per_cpu(cpu_kick_mask, cpu));
6119 kmem_cache_destroy(kvm_vcpu_cache);
6121 #ifdef CONFIG_KVM_GENERIC_HARDWARE_ENABLING
6122 unregister_syscore_ops(&kvm_syscore_ops);
6123 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_ONLINE);
6127 EXPORT_SYMBOL_GPL(kvm_init);
6134 * Note, unregistering /dev/kvm doesn't strictly need to come first,
6135 * fops_get(), a.k.a. try_module_get(), prevents acquiring references
6136 * to KVM while the module is being stopped.
6138 misc_deregister(&kvm_dev);
6140 debugfs_remove_recursive(kvm_debugfs_dir);
6141 for_each_possible_cpu(cpu)
6142 free_cpumask_var(per_cpu(cpu_kick_mask, cpu));
6143 kmem_cache_destroy(kvm_vcpu_cache);
6144 kvm_vfio_ops_exit();
6145 kvm_async_pf_deinit();
6146 #ifdef CONFIG_KVM_GENERIC_HARDWARE_ENABLING
6147 unregister_syscore_ops(&kvm_syscore_ops);
6148 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_ONLINE);
6152 EXPORT_SYMBOL_GPL(kvm_exit);
6154 struct kvm_vm_worker_thread_context {
6156 struct task_struct *parent;
6157 struct completion init_done;
6158 kvm_vm_thread_fn_t thread_fn;
6163 static int kvm_vm_worker_thread(void *context)
6166 * The init_context is allocated on the stack of the parent thread, so
6167 * we have to locally copy anything that is needed beyond initialization
6169 struct kvm_vm_worker_thread_context *init_context = context;
6170 struct task_struct *parent;
6171 struct kvm *kvm = init_context->kvm;
6172 kvm_vm_thread_fn_t thread_fn = init_context->thread_fn;
6173 uintptr_t data = init_context->data;
6176 err = kthread_park(current);
6177 /* kthread_park(current) is never supposed to return an error */
6182 err = cgroup_attach_task_all(init_context->parent, current);
6184 kvm_err("%s: cgroup_attach_task_all failed with err %d\n",
6189 set_user_nice(current, task_nice(init_context->parent));
6192 init_context->err = err;
6193 complete(&init_context->init_done);
6194 init_context = NULL;
6199 /* Wait to be woken up by the spawner before proceeding. */
6202 if (!kthread_should_stop())
6203 err = thread_fn(kvm, data);
6207 * Move kthread back to its original cgroup to prevent it lingering in
6208 * the cgroup of the VM process, after the latter finishes its
6211 * kthread_stop() waits on the 'exited' completion condition which is
6212 * set in exit_mm(), via mm_release(), in do_exit(). However, the
6213 * kthread is removed from the cgroup in the cgroup_exit() which is
6214 * called after the exit_mm(). This causes the kthread_stop() to return
6215 * before the kthread actually quits the cgroup.
6218 parent = rcu_dereference(current->real_parent);
6219 get_task_struct(parent);
6221 cgroup_attach_task_all(parent, current);
6222 put_task_struct(parent);
6227 int kvm_vm_create_worker_thread(struct kvm *kvm, kvm_vm_thread_fn_t thread_fn,
6228 uintptr_t data, const char *name,
6229 struct task_struct **thread_ptr)
6231 struct kvm_vm_worker_thread_context init_context = {};
6232 struct task_struct *thread;
6235 init_context.kvm = kvm;
6236 init_context.parent = current;
6237 init_context.thread_fn = thread_fn;
6238 init_context.data = data;
6239 init_completion(&init_context.init_done);
6241 thread = kthread_run(kvm_vm_worker_thread, &init_context,
6242 "%s-%d", name, task_pid_nr(current));
6244 return PTR_ERR(thread);
6246 /* kthread_run is never supposed to return NULL */
6247 WARN_ON(thread == NULL);
6249 wait_for_completion(&init_context.init_done);
6251 if (!init_context.err)
6252 *thread_ptr = thread;
6254 return init_context.err;