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 static void kvm_flush_shadow_all(struct kvm *kvm)
384 kvm_arch_flush_shadow_all(kvm);
385 kvm_arch_guest_memory_reclaimed(kvm);
388 #ifdef KVM_ARCH_NR_OBJS_PER_MEMORY_CACHE
389 static inline void *mmu_memory_cache_alloc_obj(struct kvm_mmu_memory_cache *mc,
392 gfp_flags |= mc->gfp_zero;
395 return kmem_cache_alloc(mc->kmem_cache, gfp_flags);
397 return (void *)__get_free_page(gfp_flags);
400 int __kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache *mc, int capacity, int min)
402 gfp_t gfp = mc->gfp_custom ? mc->gfp_custom : GFP_KERNEL_ACCOUNT;
405 if (mc->nobjs >= min)
408 if (unlikely(!mc->objects)) {
409 if (WARN_ON_ONCE(!capacity))
412 mc->objects = kvmalloc_array(sizeof(void *), capacity, gfp);
416 mc->capacity = capacity;
419 /* It is illegal to request a different capacity across topups. */
420 if (WARN_ON_ONCE(mc->capacity != capacity))
423 while (mc->nobjs < mc->capacity) {
424 obj = mmu_memory_cache_alloc_obj(mc, gfp);
426 return mc->nobjs >= min ? 0 : -ENOMEM;
427 mc->objects[mc->nobjs++] = obj;
432 int kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache *mc, int min)
434 return __kvm_mmu_topup_memory_cache(mc, KVM_ARCH_NR_OBJS_PER_MEMORY_CACHE, min);
437 int kvm_mmu_memory_cache_nr_free_objects(struct kvm_mmu_memory_cache *mc)
442 void kvm_mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc)
446 kmem_cache_free(mc->kmem_cache, mc->objects[--mc->nobjs]);
448 free_page((unsigned long)mc->objects[--mc->nobjs]);
457 void *kvm_mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc)
461 if (WARN_ON(!mc->nobjs))
462 p = mmu_memory_cache_alloc_obj(mc, GFP_ATOMIC | __GFP_ACCOUNT);
464 p = mc->objects[--mc->nobjs];
470 static void kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
472 mutex_init(&vcpu->mutex);
477 #ifndef __KVM_HAVE_ARCH_WQP
478 rcuwait_init(&vcpu->wait);
480 kvm_async_pf_vcpu_init(vcpu);
482 kvm_vcpu_set_in_spin_loop(vcpu, false);
483 kvm_vcpu_set_dy_eligible(vcpu, false);
484 vcpu->preempted = false;
486 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
487 vcpu->last_used_slot = NULL;
489 /* Fill the stats id string for the vcpu */
490 snprintf(vcpu->stats_id, sizeof(vcpu->stats_id), "kvm-%d/vcpu-%d",
491 task_pid_nr(current), id);
494 static void kvm_vcpu_destroy(struct kvm_vcpu *vcpu)
496 kvm_arch_vcpu_destroy(vcpu);
497 kvm_dirty_ring_free(&vcpu->dirty_ring);
500 * No need for rcu_read_lock as VCPU_RUN is the only place that changes
501 * the vcpu->pid pointer, and at destruction time all file descriptors
504 put_pid(rcu_dereference_protected(vcpu->pid, 1));
506 free_page((unsigned long)vcpu->run);
507 kmem_cache_free(kvm_vcpu_cache, vcpu);
510 void kvm_destroy_vcpus(struct kvm *kvm)
513 struct kvm_vcpu *vcpu;
515 kvm_for_each_vcpu(i, vcpu, kvm) {
516 kvm_vcpu_destroy(vcpu);
517 xa_erase(&kvm->vcpu_array, i);
520 atomic_set(&kvm->online_vcpus, 0);
522 EXPORT_SYMBOL_GPL(kvm_destroy_vcpus);
524 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
525 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
527 return container_of(mn, struct kvm, mmu_notifier);
530 typedef bool (*hva_handler_t)(struct kvm *kvm, struct kvm_gfn_range *range);
532 typedef void (*on_lock_fn_t)(struct kvm *kvm, unsigned long start,
535 typedef void (*on_unlock_fn_t)(struct kvm *kvm);
537 struct kvm_hva_range {
541 hva_handler_t handler;
542 on_lock_fn_t on_lock;
543 on_unlock_fn_t on_unlock;
549 * Use a dedicated stub instead of NULL to indicate that there is no callback
550 * function/handler. The compiler technically can't guarantee that a real
551 * function will have a non-zero address, and so it will generate code to
552 * check for !NULL, whereas comparing against a stub will be elided at compile
553 * time (unless the compiler is getting long in the tooth, e.g. gcc 4.9).
555 static void kvm_null_fn(void)
559 #define IS_KVM_NULL_FN(fn) ((fn) == (void *)kvm_null_fn)
561 /* Iterate over each memslot intersecting [start, last] (inclusive) range */
562 #define kvm_for_each_memslot_in_hva_range(node, slots, start, last) \
563 for (node = interval_tree_iter_first(&slots->hva_tree, start, last); \
565 node = interval_tree_iter_next(node, start, last)) \
567 static __always_inline int __kvm_handle_hva_range(struct kvm *kvm,
568 const struct kvm_hva_range *range)
570 bool ret = false, locked = false;
571 struct kvm_gfn_range gfn_range;
572 struct kvm_memory_slot *slot;
573 struct kvm_memslots *slots;
576 if (WARN_ON_ONCE(range->end <= range->start))
579 /* A null handler is allowed if and only if on_lock() is provided. */
580 if (WARN_ON_ONCE(IS_KVM_NULL_FN(range->on_lock) &&
581 IS_KVM_NULL_FN(range->handler)))
584 idx = srcu_read_lock(&kvm->srcu);
586 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
587 struct interval_tree_node *node;
589 slots = __kvm_memslots(kvm, i);
590 kvm_for_each_memslot_in_hva_range(node, slots,
591 range->start, range->end - 1) {
592 unsigned long hva_start, hva_end;
594 slot = container_of(node, struct kvm_memory_slot, hva_node[slots->node_idx]);
595 hva_start = max(range->start, slot->userspace_addr);
596 hva_end = min(range->end, slot->userspace_addr +
597 (slot->npages << PAGE_SHIFT));
600 * To optimize for the likely case where the address
601 * range is covered by zero or one memslots, don't
602 * bother making these conditional (to avoid writes on
603 * the second or later invocation of the handler).
605 gfn_range.pte = range->pte;
606 gfn_range.may_block = range->may_block;
609 * {gfn(page) | page intersects with [hva_start, hva_end)} =
610 * {gfn_start, gfn_start+1, ..., gfn_end-1}.
612 gfn_range.start = hva_to_gfn_memslot(hva_start, slot);
613 gfn_range.end = hva_to_gfn_memslot(hva_end + PAGE_SIZE - 1, slot);
614 gfn_range.slot = slot;
619 if (!IS_KVM_NULL_FN(range->on_lock))
620 range->on_lock(kvm, range->start, range->end);
621 if (IS_KVM_NULL_FN(range->handler))
624 ret |= range->handler(kvm, &gfn_range);
628 if (range->flush_on_ret && ret)
629 kvm_flush_remote_tlbs(kvm);
633 if (!IS_KVM_NULL_FN(range->on_unlock))
634 range->on_unlock(kvm);
637 srcu_read_unlock(&kvm->srcu, idx);
639 /* The notifiers are averse to booleans. :-( */
643 static __always_inline int kvm_handle_hva_range(struct mmu_notifier *mn,
647 hva_handler_t handler)
649 struct kvm *kvm = mmu_notifier_to_kvm(mn);
650 const struct kvm_hva_range range = {
655 .on_lock = (void *)kvm_null_fn,
656 .on_unlock = (void *)kvm_null_fn,
657 .flush_on_ret = true,
661 return __kvm_handle_hva_range(kvm, &range);
664 static __always_inline int kvm_handle_hva_range_no_flush(struct mmu_notifier *mn,
667 hva_handler_t handler)
669 struct kvm *kvm = mmu_notifier_to_kvm(mn);
670 const struct kvm_hva_range range = {
675 .on_lock = (void *)kvm_null_fn,
676 .on_unlock = (void *)kvm_null_fn,
677 .flush_on_ret = false,
681 return __kvm_handle_hva_range(kvm, &range);
684 static bool kvm_change_spte_gfn(struct kvm *kvm, struct kvm_gfn_range *range)
687 * Skipping invalid memslots is correct if and only change_pte() is
688 * surrounded by invalidate_range_{start,end}(), which is currently
689 * guaranteed by the primary MMU. If that ever changes, KVM needs to
690 * unmap the memslot instead of skipping the memslot to ensure that KVM
691 * doesn't hold references to the old PFN.
693 WARN_ON_ONCE(!READ_ONCE(kvm->mn_active_invalidate_count));
695 if (range->slot->flags & KVM_MEMSLOT_INVALID)
698 return kvm_set_spte_gfn(kvm, range);
701 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
702 struct mm_struct *mm,
703 unsigned long address,
706 struct kvm *kvm = mmu_notifier_to_kvm(mn);
708 trace_kvm_set_spte_hva(address);
711 * .change_pte() must be surrounded by .invalidate_range_{start,end}().
712 * If mmu_invalidate_in_progress is zero, then no in-progress
713 * invalidations, including this one, found a relevant memslot at
714 * start(); rechecking memslots here is unnecessary. Note, a false
715 * positive (count elevated by a different invalidation) is sub-optimal
716 * but functionally ok.
718 WARN_ON_ONCE(!READ_ONCE(kvm->mn_active_invalidate_count));
719 if (!READ_ONCE(kvm->mmu_invalidate_in_progress))
722 kvm_handle_hva_range(mn, address, address + 1, pte, kvm_change_spte_gfn);
725 void kvm_mmu_invalidate_begin(struct kvm *kvm, unsigned long start,
729 * The count increase must become visible at unlock time as no
730 * spte can be established without taking the mmu_lock and
731 * count is also read inside the mmu_lock critical section.
733 kvm->mmu_invalidate_in_progress++;
734 if (likely(kvm->mmu_invalidate_in_progress == 1)) {
735 kvm->mmu_invalidate_range_start = start;
736 kvm->mmu_invalidate_range_end = end;
739 * Fully tracking multiple concurrent ranges has diminishing
740 * returns. Keep things simple and just find the minimal range
741 * which includes the current and new ranges. As there won't be
742 * enough information to subtract a range after its invalidate
743 * completes, any ranges invalidated concurrently will
744 * accumulate and persist until all outstanding invalidates
747 kvm->mmu_invalidate_range_start =
748 min(kvm->mmu_invalidate_range_start, start);
749 kvm->mmu_invalidate_range_end =
750 max(kvm->mmu_invalidate_range_end, end);
754 static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
755 const struct mmu_notifier_range *range)
757 struct kvm *kvm = mmu_notifier_to_kvm(mn);
758 const struct kvm_hva_range hva_range = {
759 .start = range->start,
762 .handler = kvm_unmap_gfn_range,
763 .on_lock = kvm_mmu_invalidate_begin,
764 .on_unlock = kvm_arch_guest_memory_reclaimed,
765 .flush_on_ret = true,
766 .may_block = mmu_notifier_range_blockable(range),
769 trace_kvm_unmap_hva_range(range->start, range->end);
772 * Prevent memslot modification between range_start() and range_end()
773 * so that conditionally locking provides the same result in both
774 * functions. Without that guarantee, the mmu_invalidate_in_progress
775 * adjustments will be imbalanced.
777 * Pairs with the decrement in range_end().
779 spin_lock(&kvm->mn_invalidate_lock);
780 kvm->mn_active_invalidate_count++;
781 spin_unlock(&kvm->mn_invalidate_lock);
784 * Invalidate pfn caches _before_ invalidating the secondary MMUs, i.e.
785 * before acquiring mmu_lock, to avoid holding mmu_lock while acquiring
786 * each cache's lock. There are relatively few caches in existence at
787 * any given time, and the caches themselves can check for hva overlap,
788 * i.e. don't need to rely on memslot overlap checks for performance.
789 * Because this runs without holding mmu_lock, the pfn caches must use
790 * mn_active_invalidate_count (see above) instead of
791 * mmu_invalidate_in_progress.
793 gfn_to_pfn_cache_invalidate_start(kvm, range->start, range->end,
794 hva_range.may_block);
796 __kvm_handle_hva_range(kvm, &hva_range);
801 void kvm_mmu_invalidate_end(struct kvm *kvm, unsigned long start,
805 * This sequence increase will notify the kvm page fault that
806 * the page that is going to be mapped in the spte could have
809 kvm->mmu_invalidate_seq++;
812 * The above sequence increase must be visible before the
813 * below count decrease, which is ensured by the smp_wmb above
814 * in conjunction with the smp_rmb in mmu_invalidate_retry().
816 kvm->mmu_invalidate_in_progress--;
819 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
820 const struct mmu_notifier_range *range)
822 struct kvm *kvm = mmu_notifier_to_kvm(mn);
823 const struct kvm_hva_range hva_range = {
824 .start = range->start,
827 .handler = (void *)kvm_null_fn,
828 .on_lock = kvm_mmu_invalidate_end,
829 .on_unlock = (void *)kvm_null_fn,
830 .flush_on_ret = false,
831 .may_block = mmu_notifier_range_blockable(range),
835 __kvm_handle_hva_range(kvm, &hva_range);
837 /* Pairs with the increment in range_start(). */
838 spin_lock(&kvm->mn_invalidate_lock);
839 wake = (--kvm->mn_active_invalidate_count == 0);
840 spin_unlock(&kvm->mn_invalidate_lock);
843 * There can only be one waiter, since the wait happens under
847 rcuwait_wake_up(&kvm->mn_memslots_update_rcuwait);
849 BUG_ON(kvm->mmu_invalidate_in_progress < 0);
852 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
853 struct mm_struct *mm,
857 trace_kvm_age_hva(start, end);
859 return kvm_handle_hva_range(mn, start, end, __pte(0), kvm_age_gfn);
862 static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
863 struct mm_struct *mm,
867 trace_kvm_age_hva(start, end);
870 * Even though we do not flush TLB, this will still adversely
871 * affect performance on pre-Haswell Intel EPT, where there is
872 * no EPT Access Bit to clear so that we have to tear down EPT
873 * tables instead. If we find this unacceptable, we can always
874 * add a parameter to kvm_age_hva so that it effectively doesn't
875 * do anything on clear_young.
877 * Also note that currently we never issue secondary TLB flushes
878 * from clear_young, leaving this job up to the regular system
879 * cadence. If we find this inaccurate, we might come up with a
880 * more sophisticated heuristic later.
882 return kvm_handle_hva_range_no_flush(mn, start, end, kvm_age_gfn);
885 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
886 struct mm_struct *mm,
887 unsigned long address)
889 trace_kvm_test_age_hva(address);
891 return kvm_handle_hva_range_no_flush(mn, address, address + 1,
895 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
896 struct mm_struct *mm)
898 struct kvm *kvm = mmu_notifier_to_kvm(mn);
901 idx = srcu_read_lock(&kvm->srcu);
902 kvm_flush_shadow_all(kvm);
903 srcu_read_unlock(&kvm->srcu, idx);
906 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
907 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
908 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
909 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
910 .clear_young = kvm_mmu_notifier_clear_young,
911 .test_young = kvm_mmu_notifier_test_young,
912 .change_pte = kvm_mmu_notifier_change_pte,
913 .release = kvm_mmu_notifier_release,
916 static int kvm_init_mmu_notifier(struct kvm *kvm)
918 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
919 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
922 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
924 static int kvm_init_mmu_notifier(struct kvm *kvm)
929 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
931 #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER
932 static int kvm_pm_notifier_call(struct notifier_block *bl,
936 struct kvm *kvm = container_of(bl, struct kvm, pm_notifier);
938 return kvm_arch_pm_notifier(kvm, state);
941 static void kvm_init_pm_notifier(struct kvm *kvm)
943 kvm->pm_notifier.notifier_call = kvm_pm_notifier_call;
944 /* Suspend KVM before we suspend ftrace, RCU, etc. */
945 kvm->pm_notifier.priority = INT_MAX;
946 register_pm_notifier(&kvm->pm_notifier);
949 static void kvm_destroy_pm_notifier(struct kvm *kvm)
951 unregister_pm_notifier(&kvm->pm_notifier);
953 #else /* !CONFIG_HAVE_KVM_PM_NOTIFIER */
954 static void kvm_init_pm_notifier(struct kvm *kvm)
958 static void kvm_destroy_pm_notifier(struct kvm *kvm)
961 #endif /* CONFIG_HAVE_KVM_PM_NOTIFIER */
963 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
965 if (!memslot->dirty_bitmap)
968 kvfree(memslot->dirty_bitmap);
969 memslot->dirty_bitmap = NULL;
972 /* This does not remove the slot from struct kvm_memslots data structures */
973 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
975 kvm_destroy_dirty_bitmap(slot);
977 kvm_arch_free_memslot(kvm, slot);
982 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
984 struct hlist_node *idnode;
985 struct kvm_memory_slot *memslot;
989 * The same memslot objects live in both active and inactive sets,
990 * arbitrarily free using index '1' so the second invocation of this
991 * function isn't operating over a structure with dangling pointers
992 * (even though this function isn't actually touching them).
994 if (!slots->node_idx)
997 hash_for_each_safe(slots->id_hash, bkt, idnode, memslot, id_node[1])
998 kvm_free_memslot(kvm, memslot);
1001 static umode_t kvm_stats_debugfs_mode(const struct _kvm_stats_desc *pdesc)
1003 switch (pdesc->desc.flags & KVM_STATS_TYPE_MASK) {
1004 case KVM_STATS_TYPE_INSTANT:
1006 case KVM_STATS_TYPE_CUMULATIVE:
1007 case KVM_STATS_TYPE_PEAK:
1014 static void kvm_destroy_vm_debugfs(struct kvm *kvm)
1017 int kvm_debugfs_num_entries = kvm_vm_stats_header.num_desc +
1018 kvm_vcpu_stats_header.num_desc;
1020 if (IS_ERR(kvm->debugfs_dentry))
1023 debugfs_remove_recursive(kvm->debugfs_dentry);
1025 if (kvm->debugfs_stat_data) {
1026 for (i = 0; i < kvm_debugfs_num_entries; i++)
1027 kfree(kvm->debugfs_stat_data[i]);
1028 kfree(kvm->debugfs_stat_data);
1032 static int kvm_create_vm_debugfs(struct kvm *kvm, const char *fdname)
1034 static DEFINE_MUTEX(kvm_debugfs_lock);
1035 struct dentry *dent;
1036 char dir_name[ITOA_MAX_LEN * 2];
1037 struct kvm_stat_data *stat_data;
1038 const struct _kvm_stats_desc *pdesc;
1039 int i, ret = -ENOMEM;
1040 int kvm_debugfs_num_entries = kvm_vm_stats_header.num_desc +
1041 kvm_vcpu_stats_header.num_desc;
1043 if (!debugfs_initialized())
1046 snprintf(dir_name, sizeof(dir_name), "%d-%s", task_pid_nr(current), fdname);
1047 mutex_lock(&kvm_debugfs_lock);
1048 dent = debugfs_lookup(dir_name, kvm_debugfs_dir);
1050 pr_warn_ratelimited("KVM: debugfs: duplicate directory %s\n", dir_name);
1052 mutex_unlock(&kvm_debugfs_lock);
1055 dent = debugfs_create_dir(dir_name, kvm_debugfs_dir);
1056 mutex_unlock(&kvm_debugfs_lock);
1060 kvm->debugfs_dentry = dent;
1061 kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries,
1062 sizeof(*kvm->debugfs_stat_data),
1063 GFP_KERNEL_ACCOUNT);
1064 if (!kvm->debugfs_stat_data)
1067 for (i = 0; i < kvm_vm_stats_header.num_desc; ++i) {
1068 pdesc = &kvm_vm_stats_desc[i];
1069 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT);
1073 stat_data->kvm = kvm;
1074 stat_data->desc = pdesc;
1075 stat_data->kind = KVM_STAT_VM;
1076 kvm->debugfs_stat_data[i] = stat_data;
1077 debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
1078 kvm->debugfs_dentry, stat_data,
1082 for (i = 0; i < kvm_vcpu_stats_header.num_desc; ++i) {
1083 pdesc = &kvm_vcpu_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_VCPU;
1091 kvm->debugfs_stat_data[i + kvm_vm_stats_header.num_desc] = stat_data;
1092 debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
1093 kvm->debugfs_dentry, stat_data,
1097 ret = kvm_arch_create_vm_debugfs(kvm);
1103 kvm_destroy_vm_debugfs(kvm);
1108 * Called after the VM is otherwise initialized, but just before adding it to
1111 int __weak kvm_arch_post_init_vm(struct kvm *kvm)
1117 * Called just after removing the VM from the vm_list, but before doing any
1118 * other destruction.
1120 void __weak kvm_arch_pre_destroy_vm(struct kvm *kvm)
1125 * Called after per-vm debugfs created. When called kvm->debugfs_dentry should
1126 * be setup already, so we can create arch-specific debugfs entries under it.
1127 * Cleanup should be automatic done in kvm_destroy_vm_debugfs() recursively, so
1128 * a per-arch destroy interface is not needed.
1130 int __weak kvm_arch_create_vm_debugfs(struct kvm *kvm)
1135 static struct kvm *kvm_create_vm(unsigned long type, const char *fdname)
1137 struct kvm *kvm = kvm_arch_alloc_vm();
1138 struct kvm_memslots *slots;
1143 return ERR_PTR(-ENOMEM);
1145 /* KVM is pinned via open("/dev/kvm"), the fd passed to this ioctl(). */
1146 __module_get(kvm_chardev_ops.owner);
1148 KVM_MMU_LOCK_INIT(kvm);
1149 mmgrab(current->mm);
1150 kvm->mm = current->mm;
1151 kvm_eventfd_init(kvm);
1152 mutex_init(&kvm->lock);
1153 mutex_init(&kvm->irq_lock);
1154 mutex_init(&kvm->slots_lock);
1155 mutex_init(&kvm->slots_arch_lock);
1156 spin_lock_init(&kvm->mn_invalidate_lock);
1157 rcuwait_init(&kvm->mn_memslots_update_rcuwait);
1158 xa_init(&kvm->vcpu_array);
1160 INIT_LIST_HEAD(&kvm->gpc_list);
1161 spin_lock_init(&kvm->gpc_lock);
1163 INIT_LIST_HEAD(&kvm->devices);
1164 kvm->max_vcpus = KVM_MAX_VCPUS;
1166 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
1169 * Force subsequent debugfs file creations to fail if the VM directory
1170 * is not created (by kvm_create_vm_debugfs()).
1172 kvm->debugfs_dentry = ERR_PTR(-ENOENT);
1174 snprintf(kvm->stats_id, sizeof(kvm->stats_id), "kvm-%d",
1175 task_pid_nr(current));
1177 if (init_srcu_struct(&kvm->srcu))
1178 goto out_err_no_srcu;
1179 if (init_srcu_struct(&kvm->irq_srcu))
1180 goto out_err_no_irq_srcu;
1182 refcount_set(&kvm->users_count, 1);
1183 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
1184 for (j = 0; j < 2; j++) {
1185 slots = &kvm->__memslots[i][j];
1187 atomic_long_set(&slots->last_used_slot, (unsigned long)NULL);
1188 slots->hva_tree = RB_ROOT_CACHED;
1189 slots->gfn_tree = RB_ROOT;
1190 hash_init(slots->id_hash);
1191 slots->node_idx = j;
1193 /* Generations must be different for each address space. */
1194 slots->generation = i;
1197 rcu_assign_pointer(kvm->memslots[i], &kvm->__memslots[i][0]);
1200 for (i = 0; i < KVM_NR_BUSES; i++) {
1201 rcu_assign_pointer(kvm->buses[i],
1202 kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL_ACCOUNT));
1204 goto out_err_no_arch_destroy_vm;
1207 r = kvm_arch_init_vm(kvm, type);
1209 goto out_err_no_arch_destroy_vm;
1211 r = hardware_enable_all();
1213 goto out_err_no_disable;
1215 #ifdef CONFIG_HAVE_KVM_IRQFD
1216 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
1219 r = kvm_init_mmu_notifier(kvm);
1221 goto out_err_no_mmu_notifier;
1223 r = kvm_coalesced_mmio_init(kvm);
1225 goto out_no_coalesced_mmio;
1227 r = kvm_create_vm_debugfs(kvm, fdname);
1229 goto out_err_no_debugfs;
1231 r = kvm_arch_post_init_vm(kvm);
1235 mutex_lock(&kvm_lock);
1236 list_add(&kvm->vm_list, &vm_list);
1237 mutex_unlock(&kvm_lock);
1239 preempt_notifier_inc();
1240 kvm_init_pm_notifier(kvm);
1245 kvm_destroy_vm_debugfs(kvm);
1247 kvm_coalesced_mmio_free(kvm);
1248 out_no_coalesced_mmio:
1249 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1250 if (kvm->mmu_notifier.ops)
1251 mmu_notifier_unregister(&kvm->mmu_notifier, current->mm);
1253 out_err_no_mmu_notifier:
1254 hardware_disable_all();
1256 kvm_arch_destroy_vm(kvm);
1257 out_err_no_arch_destroy_vm:
1258 WARN_ON_ONCE(!refcount_dec_and_test(&kvm->users_count));
1259 for (i = 0; i < KVM_NR_BUSES; i++)
1260 kfree(kvm_get_bus(kvm, i));
1261 cleanup_srcu_struct(&kvm->irq_srcu);
1262 out_err_no_irq_srcu:
1263 cleanup_srcu_struct(&kvm->srcu);
1265 kvm_arch_free_vm(kvm);
1266 mmdrop(current->mm);
1267 module_put(kvm_chardev_ops.owner);
1271 static void kvm_destroy_devices(struct kvm *kvm)
1273 struct kvm_device *dev, *tmp;
1276 * We do not need to take the kvm->lock here, because nobody else
1277 * has a reference to the struct kvm at this point and therefore
1278 * cannot access the devices list anyhow.
1280 list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
1281 list_del(&dev->vm_node);
1282 dev->ops->destroy(dev);
1286 static void kvm_destroy_vm(struct kvm *kvm)
1289 struct mm_struct *mm = kvm->mm;
1291 kvm_destroy_pm_notifier(kvm);
1292 kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm);
1293 kvm_destroy_vm_debugfs(kvm);
1294 kvm_arch_sync_events(kvm);
1295 mutex_lock(&kvm_lock);
1296 list_del(&kvm->vm_list);
1297 mutex_unlock(&kvm_lock);
1298 kvm_arch_pre_destroy_vm(kvm);
1300 kvm_free_irq_routing(kvm);
1301 for (i = 0; i < KVM_NR_BUSES; i++) {
1302 struct kvm_io_bus *bus = kvm_get_bus(kvm, i);
1305 kvm_io_bus_destroy(bus);
1306 kvm->buses[i] = NULL;
1308 kvm_coalesced_mmio_free(kvm);
1309 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1310 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
1312 * At this point, pending calls to invalidate_range_start()
1313 * have completed but no more MMU notifiers will run, so
1314 * mn_active_invalidate_count may remain unbalanced.
1315 * No threads can be waiting in kvm_swap_active_memslots() as the
1316 * last reference on KVM has been dropped, but freeing
1317 * memslots would deadlock without this manual intervention.
1319 WARN_ON(rcuwait_active(&kvm->mn_memslots_update_rcuwait));
1320 kvm->mn_active_invalidate_count = 0;
1322 kvm_flush_shadow_all(kvm);
1324 kvm_arch_destroy_vm(kvm);
1325 kvm_destroy_devices(kvm);
1326 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
1327 kvm_free_memslots(kvm, &kvm->__memslots[i][0]);
1328 kvm_free_memslots(kvm, &kvm->__memslots[i][1]);
1330 cleanup_srcu_struct(&kvm->irq_srcu);
1331 cleanup_srcu_struct(&kvm->srcu);
1332 kvm_arch_free_vm(kvm);
1333 preempt_notifier_dec();
1334 hardware_disable_all();
1336 module_put(kvm_chardev_ops.owner);
1339 void kvm_get_kvm(struct kvm *kvm)
1341 refcount_inc(&kvm->users_count);
1343 EXPORT_SYMBOL_GPL(kvm_get_kvm);
1346 * Make sure the vm is not during destruction, which is a safe version of
1347 * kvm_get_kvm(). Return true if kvm referenced successfully, false otherwise.
1349 bool kvm_get_kvm_safe(struct kvm *kvm)
1351 return refcount_inc_not_zero(&kvm->users_count);
1353 EXPORT_SYMBOL_GPL(kvm_get_kvm_safe);
1355 void kvm_put_kvm(struct kvm *kvm)
1357 if (refcount_dec_and_test(&kvm->users_count))
1358 kvm_destroy_vm(kvm);
1360 EXPORT_SYMBOL_GPL(kvm_put_kvm);
1363 * Used to put a reference that was taken on behalf of an object associated
1364 * with a user-visible file descriptor, e.g. a vcpu or device, if installation
1365 * of the new file descriptor fails and the reference cannot be transferred to
1366 * its final owner. In such cases, the caller is still actively using @kvm and
1367 * will fail miserably if the refcount unexpectedly hits zero.
1369 void kvm_put_kvm_no_destroy(struct kvm *kvm)
1371 WARN_ON(refcount_dec_and_test(&kvm->users_count));
1373 EXPORT_SYMBOL_GPL(kvm_put_kvm_no_destroy);
1375 static int kvm_vm_release(struct inode *inode, struct file *filp)
1377 struct kvm *kvm = filp->private_data;
1379 kvm_irqfd_release(kvm);
1386 * Allocation size is twice as large as the actual dirty bitmap size.
1387 * See kvm_vm_ioctl_get_dirty_log() why this is needed.
1389 static int kvm_alloc_dirty_bitmap(struct kvm_memory_slot *memslot)
1391 unsigned long dirty_bytes = kvm_dirty_bitmap_bytes(memslot);
1393 memslot->dirty_bitmap = __vcalloc(2, dirty_bytes, GFP_KERNEL_ACCOUNT);
1394 if (!memslot->dirty_bitmap)
1400 static struct kvm_memslots *kvm_get_inactive_memslots(struct kvm *kvm, int as_id)
1402 struct kvm_memslots *active = __kvm_memslots(kvm, as_id);
1403 int node_idx_inactive = active->node_idx ^ 1;
1405 return &kvm->__memslots[as_id][node_idx_inactive];
1409 * Helper to get the address space ID when one of memslot pointers may be NULL.
1410 * This also serves as a sanity that at least one of the pointers is non-NULL,
1411 * and that their address space IDs don't diverge.
1413 static int kvm_memslots_get_as_id(struct kvm_memory_slot *a,
1414 struct kvm_memory_slot *b)
1416 if (WARN_ON_ONCE(!a && !b))
1424 WARN_ON_ONCE(a->as_id != b->as_id);
1428 static void kvm_insert_gfn_node(struct kvm_memslots *slots,
1429 struct kvm_memory_slot *slot)
1431 struct rb_root *gfn_tree = &slots->gfn_tree;
1432 struct rb_node **node, *parent;
1433 int idx = slots->node_idx;
1436 for (node = &gfn_tree->rb_node; *node; ) {
1437 struct kvm_memory_slot *tmp;
1439 tmp = container_of(*node, struct kvm_memory_slot, gfn_node[idx]);
1441 if (slot->base_gfn < tmp->base_gfn)
1442 node = &(*node)->rb_left;
1443 else if (slot->base_gfn > tmp->base_gfn)
1444 node = &(*node)->rb_right;
1449 rb_link_node(&slot->gfn_node[idx], parent, node);
1450 rb_insert_color(&slot->gfn_node[idx], gfn_tree);
1453 static void kvm_erase_gfn_node(struct kvm_memslots *slots,
1454 struct kvm_memory_slot *slot)
1456 rb_erase(&slot->gfn_node[slots->node_idx], &slots->gfn_tree);
1459 static void kvm_replace_gfn_node(struct kvm_memslots *slots,
1460 struct kvm_memory_slot *old,
1461 struct kvm_memory_slot *new)
1463 int idx = slots->node_idx;
1465 WARN_ON_ONCE(old->base_gfn != new->base_gfn);
1467 rb_replace_node(&old->gfn_node[idx], &new->gfn_node[idx],
1472 * Replace @old with @new in the inactive memslots.
1474 * With NULL @old this simply adds @new.
1475 * With NULL @new this simply removes @old.
1477 * If @new is non-NULL its hva_node[slots_idx] range has to be set
1480 static void kvm_replace_memslot(struct kvm *kvm,
1481 struct kvm_memory_slot *old,
1482 struct kvm_memory_slot *new)
1484 int as_id = kvm_memslots_get_as_id(old, new);
1485 struct kvm_memslots *slots = kvm_get_inactive_memslots(kvm, as_id);
1486 int idx = slots->node_idx;
1489 hash_del(&old->id_node[idx]);
1490 interval_tree_remove(&old->hva_node[idx], &slots->hva_tree);
1492 if ((long)old == atomic_long_read(&slots->last_used_slot))
1493 atomic_long_set(&slots->last_used_slot, (long)new);
1496 kvm_erase_gfn_node(slots, old);
1502 * Initialize @new's hva range. Do this even when replacing an @old
1503 * slot, kvm_copy_memslot() deliberately does not touch node data.
1505 new->hva_node[idx].start = new->userspace_addr;
1506 new->hva_node[idx].last = new->userspace_addr +
1507 (new->npages << PAGE_SHIFT) - 1;
1510 * (Re)Add the new memslot. There is no O(1) interval_tree_replace(),
1511 * hva_node needs to be swapped with remove+insert even though hva can't
1512 * change when replacing an existing slot.
1514 hash_add(slots->id_hash, &new->id_node[idx], new->id);
1515 interval_tree_insert(&new->hva_node[idx], &slots->hva_tree);
1518 * If the memslot gfn is unchanged, rb_replace_node() can be used to
1519 * switch the node in the gfn tree instead of removing the old and
1520 * inserting the new as two separate operations. Replacement is a
1521 * single O(1) operation versus two O(log(n)) operations for
1524 if (old && old->base_gfn == new->base_gfn) {
1525 kvm_replace_gfn_node(slots, old, new);
1528 kvm_erase_gfn_node(slots, old);
1529 kvm_insert_gfn_node(slots, new);
1533 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
1535 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
1537 #ifdef __KVM_HAVE_READONLY_MEM
1538 valid_flags |= KVM_MEM_READONLY;
1541 if (mem->flags & ~valid_flags)
1547 static void kvm_swap_active_memslots(struct kvm *kvm, int as_id)
1549 struct kvm_memslots *slots = kvm_get_inactive_memslots(kvm, as_id);
1551 /* Grab the generation from the activate memslots. */
1552 u64 gen = __kvm_memslots(kvm, as_id)->generation;
1554 WARN_ON(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS);
1555 slots->generation = gen | KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
1558 * Do not store the new memslots while there are invalidations in
1559 * progress, otherwise the locking in invalidate_range_start and
1560 * invalidate_range_end will be unbalanced.
1562 spin_lock(&kvm->mn_invalidate_lock);
1563 prepare_to_rcuwait(&kvm->mn_memslots_update_rcuwait);
1564 while (kvm->mn_active_invalidate_count) {
1565 set_current_state(TASK_UNINTERRUPTIBLE);
1566 spin_unlock(&kvm->mn_invalidate_lock);
1568 spin_lock(&kvm->mn_invalidate_lock);
1570 finish_rcuwait(&kvm->mn_memslots_update_rcuwait);
1571 rcu_assign_pointer(kvm->memslots[as_id], slots);
1572 spin_unlock(&kvm->mn_invalidate_lock);
1575 * Acquired in kvm_set_memslot. Must be released before synchronize
1576 * SRCU below in order to avoid deadlock with another thread
1577 * acquiring the slots_arch_lock in an srcu critical section.
1579 mutex_unlock(&kvm->slots_arch_lock);
1581 synchronize_srcu_expedited(&kvm->srcu);
1584 * Increment the new memslot generation a second time, dropping the
1585 * update in-progress flag and incrementing the generation based on
1586 * the number of address spaces. This provides a unique and easily
1587 * identifiable generation number while the memslots are in flux.
1589 gen = slots->generation & ~KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
1592 * Generations must be unique even across address spaces. We do not need
1593 * a global counter for that, instead the generation space is evenly split
1594 * across address spaces. For example, with two address spaces, address
1595 * space 0 will use generations 0, 2, 4, ... while address space 1 will
1596 * use generations 1, 3, 5, ...
1598 gen += KVM_ADDRESS_SPACE_NUM;
1600 kvm_arch_memslots_updated(kvm, gen);
1602 slots->generation = gen;
1605 static int kvm_prepare_memory_region(struct kvm *kvm,
1606 const struct kvm_memory_slot *old,
1607 struct kvm_memory_slot *new,
1608 enum kvm_mr_change change)
1613 * If dirty logging is disabled, nullify the bitmap; the old bitmap
1614 * will be freed on "commit". If logging is enabled in both old and
1615 * new, reuse the existing bitmap. If logging is enabled only in the
1616 * new and KVM isn't using a ring buffer, allocate and initialize a
1619 if (change != KVM_MR_DELETE) {
1620 if (!(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
1621 new->dirty_bitmap = NULL;
1622 else if (old && old->dirty_bitmap)
1623 new->dirty_bitmap = old->dirty_bitmap;
1624 else if (kvm_use_dirty_bitmap(kvm)) {
1625 r = kvm_alloc_dirty_bitmap(new);
1629 if (kvm_dirty_log_manual_protect_and_init_set(kvm))
1630 bitmap_set(new->dirty_bitmap, 0, new->npages);
1634 r = kvm_arch_prepare_memory_region(kvm, old, new, change);
1636 /* Free the bitmap on failure if it was allocated above. */
1637 if (r && new && new->dirty_bitmap && (!old || !old->dirty_bitmap))
1638 kvm_destroy_dirty_bitmap(new);
1643 static void kvm_commit_memory_region(struct kvm *kvm,
1644 struct kvm_memory_slot *old,
1645 const struct kvm_memory_slot *new,
1646 enum kvm_mr_change change)
1648 int old_flags = old ? old->flags : 0;
1649 int new_flags = new ? new->flags : 0;
1651 * Update the total number of memslot pages before calling the arch
1652 * hook so that architectures can consume the result directly.
1654 if (change == KVM_MR_DELETE)
1655 kvm->nr_memslot_pages -= old->npages;
1656 else if (change == KVM_MR_CREATE)
1657 kvm->nr_memslot_pages += new->npages;
1659 if ((old_flags ^ new_flags) & KVM_MEM_LOG_DIRTY_PAGES) {
1660 int change = (new_flags & KVM_MEM_LOG_DIRTY_PAGES) ? 1 : -1;
1661 atomic_set(&kvm->nr_memslots_dirty_logging,
1662 atomic_read(&kvm->nr_memslots_dirty_logging) + change);
1665 kvm_arch_commit_memory_region(kvm, old, new, change);
1669 /* Nothing more to do. */
1672 /* Free the old memslot and all its metadata. */
1673 kvm_free_memslot(kvm, old);
1676 case KVM_MR_FLAGS_ONLY:
1678 * Free the dirty bitmap as needed; the below check encompasses
1679 * both the flags and whether a ring buffer is being used)
1681 if (old->dirty_bitmap && !new->dirty_bitmap)
1682 kvm_destroy_dirty_bitmap(old);
1685 * The final quirk. Free the detached, old slot, but only its
1686 * memory, not any metadata. Metadata, including arch specific
1687 * data, may be reused by @new.
1697 * Activate @new, which must be installed in the inactive slots by the caller,
1698 * by swapping the active slots and then propagating @new to @old once @old is
1699 * unreachable and can be safely modified.
1701 * With NULL @old this simply adds @new to @active (while swapping the sets).
1702 * With NULL @new this simply removes @old from @active and frees it
1703 * (while also swapping the sets).
1705 static void kvm_activate_memslot(struct kvm *kvm,
1706 struct kvm_memory_slot *old,
1707 struct kvm_memory_slot *new)
1709 int as_id = kvm_memslots_get_as_id(old, new);
1711 kvm_swap_active_memslots(kvm, as_id);
1713 /* Propagate the new memslot to the now inactive memslots. */
1714 kvm_replace_memslot(kvm, old, new);
1717 static void kvm_copy_memslot(struct kvm_memory_slot *dest,
1718 const struct kvm_memory_slot *src)
1720 dest->base_gfn = src->base_gfn;
1721 dest->npages = src->npages;
1722 dest->dirty_bitmap = src->dirty_bitmap;
1723 dest->arch = src->arch;
1724 dest->userspace_addr = src->userspace_addr;
1725 dest->flags = src->flags;
1727 dest->as_id = src->as_id;
1730 static void kvm_invalidate_memslot(struct kvm *kvm,
1731 struct kvm_memory_slot *old,
1732 struct kvm_memory_slot *invalid_slot)
1735 * Mark the current slot INVALID. As with all memslot modifications,
1736 * this must be done on an unreachable slot to avoid modifying the
1737 * current slot in the active tree.
1739 kvm_copy_memslot(invalid_slot, old);
1740 invalid_slot->flags |= KVM_MEMSLOT_INVALID;
1741 kvm_replace_memslot(kvm, old, invalid_slot);
1744 * Activate the slot that is now marked INVALID, but don't propagate
1745 * the slot to the now inactive slots. The slot is either going to be
1746 * deleted or recreated as a new slot.
1748 kvm_swap_active_memslots(kvm, old->as_id);
1751 * From this point no new shadow pages pointing to a deleted, or moved,
1752 * memslot will be created. Validation of sp->gfn happens in:
1753 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1754 * - kvm_is_visible_gfn (mmu_check_root)
1756 kvm_arch_flush_shadow_memslot(kvm, old);
1757 kvm_arch_guest_memory_reclaimed(kvm);
1759 /* Was released by kvm_swap_active_memslots(), reacquire. */
1760 mutex_lock(&kvm->slots_arch_lock);
1763 * Copy the arch-specific field of the newly-installed slot back to the
1764 * old slot as the arch data could have changed between releasing
1765 * slots_arch_lock in kvm_swap_active_memslots() and re-acquiring the lock
1766 * above. Writers are required to retrieve memslots *after* acquiring
1767 * slots_arch_lock, thus the active slot's data is guaranteed to be fresh.
1769 old->arch = invalid_slot->arch;
1772 static void kvm_create_memslot(struct kvm *kvm,
1773 struct kvm_memory_slot *new)
1775 /* Add the new memslot to the inactive set and activate. */
1776 kvm_replace_memslot(kvm, NULL, new);
1777 kvm_activate_memslot(kvm, NULL, new);
1780 static void kvm_delete_memslot(struct kvm *kvm,
1781 struct kvm_memory_slot *old,
1782 struct kvm_memory_slot *invalid_slot)
1785 * Remove the old memslot (in the inactive memslots) by passing NULL as
1786 * the "new" slot, and for the invalid version in the active slots.
1788 kvm_replace_memslot(kvm, old, NULL);
1789 kvm_activate_memslot(kvm, invalid_slot, NULL);
1792 static void kvm_move_memslot(struct kvm *kvm,
1793 struct kvm_memory_slot *old,
1794 struct kvm_memory_slot *new,
1795 struct kvm_memory_slot *invalid_slot)
1798 * Replace the old memslot in the inactive slots, and then swap slots
1799 * and replace the current INVALID with the new as well.
1801 kvm_replace_memslot(kvm, old, new);
1802 kvm_activate_memslot(kvm, invalid_slot, new);
1805 static void kvm_update_flags_memslot(struct kvm *kvm,
1806 struct kvm_memory_slot *old,
1807 struct kvm_memory_slot *new)
1810 * Similar to the MOVE case, but the slot doesn't need to be zapped as
1811 * an intermediate step. Instead, the old memslot is simply replaced
1812 * with a new, updated copy in both memslot sets.
1814 kvm_replace_memslot(kvm, old, new);
1815 kvm_activate_memslot(kvm, old, new);
1818 static int kvm_set_memslot(struct kvm *kvm,
1819 struct kvm_memory_slot *old,
1820 struct kvm_memory_slot *new,
1821 enum kvm_mr_change change)
1823 struct kvm_memory_slot *invalid_slot;
1827 * Released in kvm_swap_active_memslots().
1829 * Must be held from before the current memslots are copied until after
1830 * the new memslots are installed with rcu_assign_pointer, then
1831 * released before the synchronize srcu in kvm_swap_active_memslots().
1833 * When modifying memslots outside of the slots_lock, must be held
1834 * before reading the pointer to the current memslots until after all
1835 * changes to those memslots are complete.
1837 * These rules ensure that installing new memslots does not lose
1838 * changes made to the previous memslots.
1840 mutex_lock(&kvm->slots_arch_lock);
1843 * Invalidate the old slot if it's being deleted or moved. This is
1844 * done prior to actually deleting/moving the memslot to allow vCPUs to
1845 * continue running by ensuring there are no mappings or shadow pages
1846 * for the memslot when it is deleted/moved. Without pre-invalidation
1847 * (and without a lock), a window would exist between effecting the
1848 * delete/move and committing the changes in arch code where KVM or a
1849 * guest could access a non-existent memslot.
1851 * Modifications are done on a temporary, unreachable slot. The old
1852 * slot needs to be preserved in case a later step fails and the
1853 * invalidation needs to be reverted.
1855 if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) {
1856 invalid_slot = kzalloc(sizeof(*invalid_slot), GFP_KERNEL_ACCOUNT);
1857 if (!invalid_slot) {
1858 mutex_unlock(&kvm->slots_arch_lock);
1861 kvm_invalidate_memslot(kvm, old, invalid_slot);
1864 r = kvm_prepare_memory_region(kvm, old, new, change);
1867 * For DELETE/MOVE, revert the above INVALID change. No
1868 * modifications required since the original slot was preserved
1869 * in the inactive slots. Changing the active memslots also
1870 * release slots_arch_lock.
1872 if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) {
1873 kvm_activate_memslot(kvm, invalid_slot, old);
1874 kfree(invalid_slot);
1876 mutex_unlock(&kvm->slots_arch_lock);
1882 * For DELETE and MOVE, the working slot is now active as the INVALID
1883 * version of the old slot. MOVE is particularly special as it reuses
1884 * the old slot and returns a copy of the old slot (in working_slot).
1885 * For CREATE, there is no old slot. For DELETE and FLAGS_ONLY, the
1886 * old slot is detached but otherwise preserved.
1888 if (change == KVM_MR_CREATE)
1889 kvm_create_memslot(kvm, new);
1890 else if (change == KVM_MR_DELETE)
1891 kvm_delete_memslot(kvm, old, invalid_slot);
1892 else if (change == KVM_MR_MOVE)
1893 kvm_move_memslot(kvm, old, new, invalid_slot);
1894 else if (change == KVM_MR_FLAGS_ONLY)
1895 kvm_update_flags_memslot(kvm, old, new);
1899 /* Free the temporary INVALID slot used for DELETE and MOVE. */
1900 if (change == KVM_MR_DELETE || change == KVM_MR_MOVE)
1901 kfree(invalid_slot);
1904 * No need to refresh new->arch, changes after dropping slots_arch_lock
1905 * will directly hit the final, active memslot. Architectures are
1906 * responsible for knowing that new->arch may be stale.
1908 kvm_commit_memory_region(kvm, old, new, change);
1913 static bool kvm_check_memslot_overlap(struct kvm_memslots *slots, int id,
1914 gfn_t start, gfn_t end)
1916 struct kvm_memslot_iter iter;
1918 kvm_for_each_memslot_in_gfn_range(&iter, slots, start, end) {
1919 if (iter.slot->id != id)
1927 * Allocate some memory and give it an address in the guest physical address
1930 * Discontiguous memory is allowed, mostly for framebuffers.
1932 * Must be called holding kvm->slots_lock for write.
1934 int __kvm_set_memory_region(struct kvm *kvm,
1935 const struct kvm_userspace_memory_region *mem)
1937 struct kvm_memory_slot *old, *new;
1938 struct kvm_memslots *slots;
1939 enum kvm_mr_change change;
1940 unsigned long npages;
1945 r = check_memory_region_flags(mem);
1949 as_id = mem->slot >> 16;
1950 id = (u16)mem->slot;
1952 /* General sanity checks */
1953 if ((mem->memory_size & (PAGE_SIZE - 1)) ||
1954 (mem->memory_size != (unsigned long)mem->memory_size))
1956 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
1958 /* We can read the guest memory with __xxx_user() later on. */
1959 if ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
1960 (mem->userspace_addr != untagged_addr(mem->userspace_addr)) ||
1961 !access_ok((void __user *)(unsigned long)mem->userspace_addr,
1964 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
1966 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
1968 if ((mem->memory_size >> PAGE_SHIFT) > KVM_MEM_MAX_NR_PAGES)
1971 slots = __kvm_memslots(kvm, as_id);
1974 * Note, the old memslot (and the pointer itself!) may be invalidated
1975 * and/or destroyed by kvm_set_memslot().
1977 old = id_to_memslot(slots, id);
1979 if (!mem->memory_size) {
1980 if (!old || !old->npages)
1983 if (WARN_ON_ONCE(kvm->nr_memslot_pages < old->npages))
1986 return kvm_set_memslot(kvm, old, NULL, KVM_MR_DELETE);
1989 base_gfn = (mem->guest_phys_addr >> PAGE_SHIFT);
1990 npages = (mem->memory_size >> PAGE_SHIFT);
1992 if (!old || !old->npages) {
1993 change = KVM_MR_CREATE;
1996 * To simplify KVM internals, the total number of pages across
1997 * all memslots must fit in an unsigned long.
1999 if ((kvm->nr_memslot_pages + npages) < kvm->nr_memslot_pages)
2001 } else { /* Modify an existing slot. */
2002 if ((mem->userspace_addr != old->userspace_addr) ||
2003 (npages != old->npages) ||
2004 ((mem->flags ^ old->flags) & KVM_MEM_READONLY))
2007 if (base_gfn != old->base_gfn)
2008 change = KVM_MR_MOVE;
2009 else if (mem->flags != old->flags)
2010 change = KVM_MR_FLAGS_ONLY;
2011 else /* Nothing to change. */
2015 if ((change == KVM_MR_CREATE || change == KVM_MR_MOVE) &&
2016 kvm_check_memslot_overlap(slots, id, base_gfn, base_gfn + npages))
2019 /* Allocate a slot that will persist in the memslot. */
2020 new = kzalloc(sizeof(*new), GFP_KERNEL_ACCOUNT);
2026 new->base_gfn = base_gfn;
2027 new->npages = npages;
2028 new->flags = mem->flags;
2029 new->userspace_addr = mem->userspace_addr;
2031 r = kvm_set_memslot(kvm, old, new, change);
2036 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
2038 int kvm_set_memory_region(struct kvm *kvm,
2039 const struct kvm_userspace_memory_region *mem)
2043 mutex_lock(&kvm->slots_lock);
2044 r = __kvm_set_memory_region(kvm, mem);
2045 mutex_unlock(&kvm->slots_lock);
2048 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
2050 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
2051 struct kvm_userspace_memory_region *mem)
2053 if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
2056 return kvm_set_memory_region(kvm, mem);
2059 #ifndef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
2061 * kvm_get_dirty_log - get a snapshot of dirty pages
2062 * @kvm: pointer to kvm instance
2063 * @log: slot id and address to which we copy the log
2064 * @is_dirty: set to '1' if any dirty pages were found
2065 * @memslot: set to the associated memslot, always valid on success
2067 int kvm_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log,
2068 int *is_dirty, struct kvm_memory_slot **memslot)
2070 struct kvm_memslots *slots;
2073 unsigned long any = 0;
2075 /* Dirty ring tracking may be exclusive to dirty log tracking */
2076 if (!kvm_use_dirty_bitmap(kvm))
2082 as_id = log->slot >> 16;
2083 id = (u16)log->slot;
2084 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
2087 slots = __kvm_memslots(kvm, as_id);
2088 *memslot = id_to_memslot(slots, id);
2089 if (!(*memslot) || !(*memslot)->dirty_bitmap)
2092 kvm_arch_sync_dirty_log(kvm, *memslot);
2094 n = kvm_dirty_bitmap_bytes(*memslot);
2096 for (i = 0; !any && i < n/sizeof(long); ++i)
2097 any = (*memslot)->dirty_bitmap[i];
2099 if (copy_to_user(log->dirty_bitmap, (*memslot)->dirty_bitmap, n))
2106 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
2108 #else /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
2110 * kvm_get_dirty_log_protect - get a snapshot of dirty pages
2111 * and reenable dirty page tracking for the corresponding pages.
2112 * @kvm: pointer to kvm instance
2113 * @log: slot id and address to which we copy the log
2115 * We need to keep it in mind that VCPU threads can write to the bitmap
2116 * concurrently. So, to avoid losing track of dirty pages we keep the
2119 * 1. Take a snapshot of the bit and clear it if needed.
2120 * 2. Write protect the corresponding page.
2121 * 3. Copy the snapshot to the userspace.
2122 * 4. Upon return caller flushes TLB's if needed.
2124 * Between 2 and 4, the guest may write to the page using the remaining TLB
2125 * entry. This is not a problem because the page is reported dirty using
2126 * the snapshot taken before and step 4 ensures that writes done after
2127 * exiting to userspace will be logged for the next call.
2130 static int kvm_get_dirty_log_protect(struct kvm *kvm, struct kvm_dirty_log *log)
2132 struct kvm_memslots *slots;
2133 struct kvm_memory_slot *memslot;
2136 unsigned long *dirty_bitmap;
2137 unsigned long *dirty_bitmap_buffer;
2140 /* Dirty ring tracking may be exclusive to dirty log tracking */
2141 if (!kvm_use_dirty_bitmap(kvm))
2144 as_id = log->slot >> 16;
2145 id = (u16)log->slot;
2146 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
2149 slots = __kvm_memslots(kvm, as_id);
2150 memslot = id_to_memslot(slots, id);
2151 if (!memslot || !memslot->dirty_bitmap)
2154 dirty_bitmap = memslot->dirty_bitmap;
2156 kvm_arch_sync_dirty_log(kvm, memslot);
2158 n = kvm_dirty_bitmap_bytes(memslot);
2160 if (kvm->manual_dirty_log_protect) {
2162 * Unlike kvm_get_dirty_log, we always return false in *flush,
2163 * because no flush is needed until KVM_CLEAR_DIRTY_LOG. There
2164 * is some code duplication between this function and
2165 * kvm_get_dirty_log, but hopefully all architecture
2166 * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log
2167 * can be eliminated.
2169 dirty_bitmap_buffer = dirty_bitmap;
2171 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
2172 memset(dirty_bitmap_buffer, 0, n);
2175 for (i = 0; i < n / sizeof(long); i++) {
2179 if (!dirty_bitmap[i])
2183 mask = xchg(&dirty_bitmap[i], 0);
2184 dirty_bitmap_buffer[i] = mask;
2186 offset = i * BITS_PER_LONG;
2187 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
2190 KVM_MMU_UNLOCK(kvm);
2194 kvm_arch_flush_remote_tlbs_memslot(kvm, memslot);
2196 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
2203 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
2204 * @kvm: kvm instance
2205 * @log: slot id and address to which we copy the log
2207 * Steps 1-4 below provide general overview of dirty page logging. See
2208 * kvm_get_dirty_log_protect() function description for additional details.
2210 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
2211 * always flush the TLB (step 4) even if previous step failed and the dirty
2212 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
2213 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
2214 * writes will be marked dirty for next log read.
2216 * 1. Take a snapshot of the bit and clear it if needed.
2217 * 2. Write protect the corresponding page.
2218 * 3. Copy the snapshot to the userspace.
2219 * 4. Flush TLB's if needed.
2221 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
2222 struct kvm_dirty_log *log)
2226 mutex_lock(&kvm->slots_lock);
2228 r = kvm_get_dirty_log_protect(kvm, log);
2230 mutex_unlock(&kvm->slots_lock);
2235 * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap
2236 * and reenable dirty page tracking for the corresponding pages.
2237 * @kvm: pointer to kvm instance
2238 * @log: slot id and address from which to fetch the bitmap of dirty pages
2240 static int kvm_clear_dirty_log_protect(struct kvm *kvm,
2241 struct kvm_clear_dirty_log *log)
2243 struct kvm_memslots *slots;
2244 struct kvm_memory_slot *memslot;
2248 unsigned long *dirty_bitmap;
2249 unsigned long *dirty_bitmap_buffer;
2252 /* Dirty ring tracking may be exclusive to dirty log tracking */
2253 if (!kvm_use_dirty_bitmap(kvm))
2256 as_id = log->slot >> 16;
2257 id = (u16)log->slot;
2258 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
2261 if (log->first_page & 63)
2264 slots = __kvm_memslots(kvm, as_id);
2265 memslot = id_to_memslot(slots, id);
2266 if (!memslot || !memslot->dirty_bitmap)
2269 dirty_bitmap = memslot->dirty_bitmap;
2271 n = ALIGN(log->num_pages, BITS_PER_LONG) / 8;
2273 if (log->first_page > memslot->npages ||
2274 log->num_pages > memslot->npages - log->first_page ||
2275 (log->num_pages < memslot->npages - log->first_page && (log->num_pages & 63)))
2278 kvm_arch_sync_dirty_log(kvm, memslot);
2281 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
2282 if (copy_from_user(dirty_bitmap_buffer, log->dirty_bitmap, n))
2286 for (offset = log->first_page, i = offset / BITS_PER_LONG,
2287 n = DIV_ROUND_UP(log->num_pages, BITS_PER_LONG); n--;
2288 i++, offset += BITS_PER_LONG) {
2289 unsigned long mask = *dirty_bitmap_buffer++;
2290 atomic_long_t *p = (atomic_long_t *) &dirty_bitmap[i];
2294 mask &= atomic_long_fetch_andnot(mask, p);
2297 * mask contains the bits that really have been cleared. This
2298 * never includes any bits beyond the length of the memslot (if
2299 * the length is not aligned to 64 pages), therefore it is not
2300 * a problem if userspace sets them in log->dirty_bitmap.
2304 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
2308 KVM_MMU_UNLOCK(kvm);
2311 kvm_arch_flush_remote_tlbs_memslot(kvm, memslot);
2316 static int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm,
2317 struct kvm_clear_dirty_log *log)
2321 mutex_lock(&kvm->slots_lock);
2323 r = kvm_clear_dirty_log_protect(kvm, log);
2325 mutex_unlock(&kvm->slots_lock);
2328 #endif /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
2330 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
2332 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
2334 EXPORT_SYMBOL_GPL(gfn_to_memslot);
2336 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
2338 struct kvm_memslots *slots = kvm_vcpu_memslots(vcpu);
2339 u64 gen = slots->generation;
2340 struct kvm_memory_slot *slot;
2343 * This also protects against using a memslot from a different address space,
2344 * since different address spaces have different generation numbers.
2346 if (unlikely(gen != vcpu->last_used_slot_gen)) {
2347 vcpu->last_used_slot = NULL;
2348 vcpu->last_used_slot_gen = gen;
2351 slot = try_get_memslot(vcpu->last_used_slot, gfn);
2356 * Fall back to searching all memslots. We purposely use
2357 * search_memslots() instead of __gfn_to_memslot() to avoid
2358 * thrashing the VM-wide last_used_slot in kvm_memslots.
2360 slot = search_memslots(slots, gfn, false);
2362 vcpu->last_used_slot = slot;
2369 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
2371 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
2373 return kvm_is_visible_memslot(memslot);
2375 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
2377 bool kvm_vcpu_is_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
2379 struct kvm_memory_slot *memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2381 return kvm_is_visible_memslot(memslot);
2383 EXPORT_SYMBOL_GPL(kvm_vcpu_is_visible_gfn);
2385 unsigned long kvm_host_page_size(struct kvm_vcpu *vcpu, gfn_t gfn)
2387 struct vm_area_struct *vma;
2388 unsigned long addr, size;
2392 addr = kvm_vcpu_gfn_to_hva_prot(vcpu, gfn, NULL);
2393 if (kvm_is_error_hva(addr))
2396 mmap_read_lock(current->mm);
2397 vma = find_vma(current->mm, addr);
2401 size = vma_kernel_pagesize(vma);
2404 mmap_read_unlock(current->mm);
2409 static bool memslot_is_readonly(const struct kvm_memory_slot *slot)
2411 return slot->flags & KVM_MEM_READONLY;
2414 static unsigned long __gfn_to_hva_many(const struct kvm_memory_slot *slot, gfn_t gfn,
2415 gfn_t *nr_pages, bool write)
2417 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
2418 return KVM_HVA_ERR_BAD;
2420 if (memslot_is_readonly(slot) && write)
2421 return KVM_HVA_ERR_RO_BAD;
2424 *nr_pages = slot->npages - (gfn - slot->base_gfn);
2426 return __gfn_to_hva_memslot(slot, gfn);
2429 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
2432 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
2435 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
2438 return gfn_to_hva_many(slot, gfn, NULL);
2440 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
2442 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
2444 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
2446 EXPORT_SYMBOL_GPL(gfn_to_hva);
2448 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
2450 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
2452 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
2455 * Return the hva of a @gfn and the R/W attribute if possible.
2457 * @slot: the kvm_memory_slot which contains @gfn
2458 * @gfn: the gfn to be translated
2459 * @writable: used to return the read/write attribute of the @slot if the hva
2460 * is valid and @writable is not NULL
2462 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
2463 gfn_t gfn, bool *writable)
2465 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
2467 if (!kvm_is_error_hva(hva) && writable)
2468 *writable = !memslot_is_readonly(slot);
2473 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
2475 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2477 return gfn_to_hva_memslot_prot(slot, gfn, writable);
2480 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
2482 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2484 return gfn_to_hva_memslot_prot(slot, gfn, writable);
2487 static inline int check_user_page_hwpoison(unsigned long addr)
2489 int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
2491 rc = get_user_pages(addr, 1, flags, NULL);
2492 return rc == -EHWPOISON;
2496 * The fast path to get the writable pfn which will be stored in @pfn,
2497 * true indicates success, otherwise false is returned. It's also the
2498 * only part that runs if we can in atomic context.
2500 static bool hva_to_pfn_fast(unsigned long addr, bool write_fault,
2501 bool *writable, kvm_pfn_t *pfn)
2503 struct page *page[1];
2506 * Fast pin a writable pfn only if it is a write fault request
2507 * or the caller allows to map a writable pfn for a read fault
2510 if (!(write_fault || writable))
2513 if (get_user_page_fast_only(addr, FOLL_WRITE, page)) {
2514 *pfn = page_to_pfn(page[0]);
2525 * The slow path to get the pfn of the specified host virtual address,
2526 * 1 indicates success, -errno is returned if error is detected.
2528 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
2529 bool interruptible, bool *writable, kvm_pfn_t *pfn)
2531 unsigned int flags = FOLL_HWPOISON;
2538 *writable = write_fault;
2541 flags |= FOLL_WRITE;
2543 flags |= FOLL_NOWAIT;
2545 flags |= FOLL_INTERRUPTIBLE;
2547 npages = get_user_pages_unlocked(addr, 1, &page, flags);
2551 /* map read fault as writable if possible */
2552 if (unlikely(!write_fault) && writable) {
2555 if (get_user_page_fast_only(addr, FOLL_WRITE, &wpage)) {
2561 *pfn = page_to_pfn(page);
2565 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
2567 if (unlikely(!(vma->vm_flags & VM_READ)))
2570 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
2576 static int kvm_try_get_pfn(kvm_pfn_t pfn)
2578 struct page *page = kvm_pfn_to_refcounted_page(pfn);
2583 return get_page_unless_zero(page);
2586 static int hva_to_pfn_remapped(struct vm_area_struct *vma,
2587 unsigned long addr, bool write_fault,
2588 bool *writable, kvm_pfn_t *p_pfn)
2596 r = follow_pte(vma->vm_mm, addr, &ptep, &ptl);
2599 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
2600 * not call the fault handler, so do it here.
2602 bool unlocked = false;
2603 r = fixup_user_fault(current->mm, addr,
2604 (write_fault ? FAULT_FLAG_WRITE : 0),
2611 r = follow_pte(vma->vm_mm, addr, &ptep, &ptl);
2616 pte = ptep_get(ptep);
2618 if (write_fault && !pte_write(pte)) {
2619 pfn = KVM_PFN_ERR_RO_FAULT;
2624 *writable = pte_write(pte);
2628 * Get a reference here because callers of *hva_to_pfn* and
2629 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
2630 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
2631 * set, but the kvm_try_get_pfn/kvm_release_pfn_clean pair will
2632 * simply do nothing for reserved pfns.
2634 * Whoever called remap_pfn_range is also going to call e.g.
2635 * unmap_mapping_range before the underlying pages are freed,
2636 * causing a call to our MMU notifier.
2638 * Certain IO or PFNMAP mappings can be backed with valid
2639 * struct pages, but be allocated without refcounting e.g.,
2640 * tail pages of non-compound higher order allocations, which
2641 * would then underflow the refcount when the caller does the
2642 * required put_page. Don't allow those pages here.
2644 if (!kvm_try_get_pfn(pfn))
2648 pte_unmap_unlock(ptep, ptl);
2655 * Pin guest page in memory and return its pfn.
2656 * @addr: host virtual address which maps memory to the guest
2657 * @atomic: whether this function can sleep
2658 * @interruptible: whether the process can be interrupted by non-fatal signals
2659 * @async: whether this function need to wait IO complete if the
2660 * host page is not in the memory
2661 * @write_fault: whether we should get a writable host page
2662 * @writable: whether it allows to map a writable host page for !@write_fault
2664 * The function will map a writable host page for these two cases:
2665 * 1): @write_fault = true
2666 * 2): @write_fault = false && @writable, @writable will tell the caller
2667 * whether the mapping is writable.
2669 kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool interruptible,
2670 bool *async, bool write_fault, bool *writable)
2672 struct vm_area_struct *vma;
2676 /* we can do it either atomically or asynchronously, not both */
2677 BUG_ON(atomic && async);
2679 if (hva_to_pfn_fast(addr, write_fault, writable, &pfn))
2683 return KVM_PFN_ERR_FAULT;
2685 npages = hva_to_pfn_slow(addr, async, write_fault, interruptible,
2689 if (npages == -EINTR)
2690 return KVM_PFN_ERR_SIGPENDING;
2692 mmap_read_lock(current->mm);
2693 if (npages == -EHWPOISON ||
2694 (!async && check_user_page_hwpoison(addr))) {
2695 pfn = KVM_PFN_ERR_HWPOISON;
2700 vma = vma_lookup(current->mm, addr);
2703 pfn = KVM_PFN_ERR_FAULT;
2704 else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
2705 r = hva_to_pfn_remapped(vma, addr, write_fault, writable, &pfn);
2709 pfn = KVM_PFN_ERR_FAULT;
2711 if (async && vma_is_valid(vma, write_fault))
2713 pfn = KVM_PFN_ERR_FAULT;
2716 mmap_read_unlock(current->mm);
2720 kvm_pfn_t __gfn_to_pfn_memslot(const struct kvm_memory_slot *slot, gfn_t gfn,
2721 bool atomic, bool interruptible, bool *async,
2722 bool write_fault, bool *writable, hva_t *hva)
2724 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
2729 if (addr == KVM_HVA_ERR_RO_BAD) {
2732 return KVM_PFN_ERR_RO_FAULT;
2735 if (kvm_is_error_hva(addr)) {
2738 return KVM_PFN_NOSLOT;
2741 /* Do not map writable pfn in the readonly memslot. */
2742 if (writable && memslot_is_readonly(slot)) {
2747 return hva_to_pfn(addr, atomic, interruptible, async, write_fault,
2750 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
2752 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
2755 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, false,
2756 NULL, write_fault, writable, NULL);
2758 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
2760 kvm_pfn_t gfn_to_pfn_memslot(const struct kvm_memory_slot *slot, gfn_t gfn)
2762 return __gfn_to_pfn_memslot(slot, gfn, false, false, NULL, true,
2765 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
2767 kvm_pfn_t gfn_to_pfn_memslot_atomic(const struct kvm_memory_slot *slot, gfn_t gfn)
2769 return __gfn_to_pfn_memslot(slot, gfn, true, false, NULL, true,
2772 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
2774 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
2776 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
2778 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
2780 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
2782 return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
2784 EXPORT_SYMBOL_GPL(gfn_to_pfn);
2786 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
2788 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
2790 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
2792 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
2793 struct page **pages, int nr_pages)
2798 addr = gfn_to_hva_many(slot, gfn, &entry);
2799 if (kvm_is_error_hva(addr))
2802 if (entry < nr_pages)
2805 return get_user_pages_fast_only(addr, nr_pages, FOLL_WRITE, pages);
2807 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
2810 * Do not use this helper unless you are absolutely certain the gfn _must_ be
2811 * backed by 'struct page'. A valid example is if the backing memslot is
2812 * controlled by KVM. Note, if the returned page is valid, it's refcount has
2813 * been elevated by gfn_to_pfn().
2815 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
2820 pfn = gfn_to_pfn(kvm, gfn);
2822 if (is_error_noslot_pfn(pfn))
2823 return KVM_ERR_PTR_BAD_PAGE;
2825 page = kvm_pfn_to_refcounted_page(pfn);
2827 return KVM_ERR_PTR_BAD_PAGE;
2831 EXPORT_SYMBOL_GPL(gfn_to_page);
2833 void kvm_release_pfn(kvm_pfn_t pfn, bool dirty)
2836 kvm_release_pfn_dirty(pfn);
2838 kvm_release_pfn_clean(pfn);
2841 int kvm_vcpu_map(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map)
2845 struct page *page = KVM_UNMAPPED_PAGE;
2850 pfn = gfn_to_pfn(vcpu->kvm, gfn);
2851 if (is_error_noslot_pfn(pfn))
2854 if (pfn_valid(pfn)) {
2855 page = pfn_to_page(pfn);
2857 #ifdef CONFIG_HAS_IOMEM
2859 hva = memremap(pfn_to_hpa(pfn), PAGE_SIZE, MEMREMAP_WB);
2873 EXPORT_SYMBOL_GPL(kvm_vcpu_map);
2875 void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map, bool dirty)
2883 if (map->page != KVM_UNMAPPED_PAGE)
2885 #ifdef CONFIG_HAS_IOMEM
2891 kvm_vcpu_mark_page_dirty(vcpu, map->gfn);
2893 kvm_release_pfn(map->pfn, dirty);
2898 EXPORT_SYMBOL_GPL(kvm_vcpu_unmap);
2900 static bool kvm_is_ad_tracked_page(struct page *page)
2903 * Per page-flags.h, pages tagged PG_reserved "should in general not be
2904 * touched (e.g. set dirty) except by its owner".
2906 return !PageReserved(page);
2909 static void kvm_set_page_dirty(struct page *page)
2911 if (kvm_is_ad_tracked_page(page))
2915 static void kvm_set_page_accessed(struct page *page)
2917 if (kvm_is_ad_tracked_page(page))
2918 mark_page_accessed(page);
2921 void kvm_release_page_clean(struct page *page)
2923 WARN_ON(is_error_page(page));
2925 kvm_set_page_accessed(page);
2928 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
2930 void kvm_release_pfn_clean(kvm_pfn_t pfn)
2934 if (is_error_noslot_pfn(pfn))
2937 page = kvm_pfn_to_refcounted_page(pfn);
2941 kvm_release_page_clean(page);
2943 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
2945 void kvm_release_page_dirty(struct page *page)
2947 WARN_ON(is_error_page(page));
2949 kvm_set_page_dirty(page);
2950 kvm_release_page_clean(page);
2952 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
2954 void kvm_release_pfn_dirty(kvm_pfn_t pfn)
2958 if (is_error_noslot_pfn(pfn))
2961 page = kvm_pfn_to_refcounted_page(pfn);
2965 kvm_release_page_dirty(page);
2967 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
2970 * Note, checking for an error/noslot pfn is the caller's responsibility when
2971 * directly marking a page dirty/accessed. Unlike the "release" helpers, the
2972 * "set" helpers are not to be used when the pfn might point at garbage.
2974 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
2976 if (WARN_ON(is_error_noslot_pfn(pfn)))
2980 kvm_set_page_dirty(pfn_to_page(pfn));
2982 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
2984 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
2986 if (WARN_ON(is_error_noslot_pfn(pfn)))
2990 kvm_set_page_accessed(pfn_to_page(pfn));
2992 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
2994 static int next_segment(unsigned long len, int offset)
2996 if (len > PAGE_SIZE - offset)
2997 return PAGE_SIZE - offset;
3002 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
3003 void *data, int offset, int len)
3008 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
3009 if (kvm_is_error_hva(addr))
3011 r = __copy_from_user(data, (void __user *)addr + offset, len);
3017 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
3020 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
3022 return __kvm_read_guest_page(slot, gfn, data, offset, len);
3024 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
3026 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
3027 int offset, int len)
3029 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
3031 return __kvm_read_guest_page(slot, gfn, data, offset, len);
3033 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
3035 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
3037 gfn_t gfn = gpa >> PAGE_SHIFT;
3039 int offset = offset_in_page(gpa);
3042 while ((seg = next_segment(len, offset)) != 0) {
3043 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
3053 EXPORT_SYMBOL_GPL(kvm_read_guest);
3055 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
3057 gfn_t gfn = gpa >> PAGE_SHIFT;
3059 int offset = offset_in_page(gpa);
3062 while ((seg = next_segment(len, offset)) != 0) {
3063 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
3073 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
3075 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
3076 void *data, int offset, unsigned long len)
3081 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
3082 if (kvm_is_error_hva(addr))
3084 pagefault_disable();
3085 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
3092 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
3093 void *data, unsigned long len)
3095 gfn_t gfn = gpa >> PAGE_SHIFT;
3096 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
3097 int offset = offset_in_page(gpa);
3099 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
3101 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
3103 static int __kvm_write_guest_page(struct kvm *kvm,
3104 struct kvm_memory_slot *memslot, gfn_t gfn,
3105 const void *data, int offset, int len)
3110 addr = gfn_to_hva_memslot(memslot, gfn);
3111 if (kvm_is_error_hva(addr))
3113 r = __copy_to_user((void __user *)addr + offset, data, len);
3116 mark_page_dirty_in_slot(kvm, memslot, gfn);
3120 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
3121 const void *data, int offset, int len)
3123 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
3125 return __kvm_write_guest_page(kvm, slot, gfn, data, offset, len);
3127 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
3129 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
3130 const void *data, int offset, int len)
3132 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
3134 return __kvm_write_guest_page(vcpu->kvm, slot, gfn, data, offset, len);
3136 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
3138 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
3141 gfn_t gfn = gpa >> PAGE_SHIFT;
3143 int offset = offset_in_page(gpa);
3146 while ((seg = next_segment(len, offset)) != 0) {
3147 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
3157 EXPORT_SYMBOL_GPL(kvm_write_guest);
3159 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
3162 gfn_t gfn = gpa >> PAGE_SHIFT;
3164 int offset = offset_in_page(gpa);
3167 while ((seg = next_segment(len, offset)) != 0) {
3168 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
3178 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
3180 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
3181 struct gfn_to_hva_cache *ghc,
3182 gpa_t gpa, unsigned long len)
3184 int offset = offset_in_page(gpa);
3185 gfn_t start_gfn = gpa >> PAGE_SHIFT;
3186 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
3187 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
3188 gfn_t nr_pages_avail;
3190 /* Update ghc->generation before performing any error checks. */
3191 ghc->generation = slots->generation;
3193 if (start_gfn > end_gfn) {
3194 ghc->hva = KVM_HVA_ERR_BAD;
3199 * If the requested region crosses two memslots, we still
3200 * verify that the entire region is valid here.
3202 for ( ; start_gfn <= end_gfn; start_gfn += nr_pages_avail) {
3203 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
3204 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
3206 if (kvm_is_error_hva(ghc->hva))
3210 /* Use the slow path for cross page reads and writes. */
3211 if (nr_pages_needed == 1)
3214 ghc->memslot = NULL;
3221 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3222 gpa_t gpa, unsigned long len)
3224 struct kvm_memslots *slots = kvm_memslots(kvm);
3225 return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len);
3227 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
3229 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3230 void *data, unsigned int offset,
3233 struct kvm_memslots *slots = kvm_memslots(kvm);
3235 gpa_t gpa = ghc->gpa + offset;
3237 if (WARN_ON_ONCE(len + offset > ghc->len))
3240 if (slots->generation != ghc->generation) {
3241 if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len))
3245 if (kvm_is_error_hva(ghc->hva))
3248 if (unlikely(!ghc->memslot))
3249 return kvm_write_guest(kvm, gpa, data, len);
3251 r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
3254 mark_page_dirty_in_slot(kvm, ghc->memslot, gpa >> PAGE_SHIFT);
3258 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached);
3260 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3261 void *data, unsigned long len)
3263 return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len);
3265 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
3267 int kvm_read_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3268 void *data, unsigned int offset,
3271 struct kvm_memslots *slots = kvm_memslots(kvm);
3273 gpa_t gpa = ghc->gpa + offset;
3275 if (WARN_ON_ONCE(len + offset > ghc->len))
3278 if (slots->generation != ghc->generation) {
3279 if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len))
3283 if (kvm_is_error_hva(ghc->hva))
3286 if (unlikely(!ghc->memslot))
3287 return kvm_read_guest(kvm, gpa, data, len);
3289 r = __copy_from_user(data, (void __user *)ghc->hva + offset, len);
3295 EXPORT_SYMBOL_GPL(kvm_read_guest_offset_cached);
3297 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3298 void *data, unsigned long len)
3300 return kvm_read_guest_offset_cached(kvm, ghc, data, 0, len);
3302 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
3304 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
3306 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
3307 gfn_t gfn = gpa >> PAGE_SHIFT;
3309 int offset = offset_in_page(gpa);
3312 while ((seg = next_segment(len, offset)) != 0) {
3313 ret = kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
3322 EXPORT_SYMBOL_GPL(kvm_clear_guest);
3324 void mark_page_dirty_in_slot(struct kvm *kvm,
3325 const struct kvm_memory_slot *memslot,
3328 struct kvm_vcpu *vcpu = kvm_get_running_vcpu();
3330 #ifdef CONFIG_HAVE_KVM_DIRTY_RING
3331 if (WARN_ON_ONCE(vcpu && vcpu->kvm != kvm))
3334 WARN_ON_ONCE(!vcpu && !kvm_arch_allow_write_without_running_vcpu(kvm));
3337 if (memslot && kvm_slot_dirty_track_enabled(memslot)) {
3338 unsigned long rel_gfn = gfn - memslot->base_gfn;
3339 u32 slot = (memslot->as_id << 16) | memslot->id;
3341 if (kvm->dirty_ring_size && vcpu)
3342 kvm_dirty_ring_push(vcpu, slot, rel_gfn);
3343 else if (memslot->dirty_bitmap)
3344 set_bit_le(rel_gfn, memslot->dirty_bitmap);
3347 EXPORT_SYMBOL_GPL(mark_page_dirty_in_slot);
3349 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
3351 struct kvm_memory_slot *memslot;
3353 memslot = gfn_to_memslot(kvm, gfn);
3354 mark_page_dirty_in_slot(kvm, memslot, gfn);
3356 EXPORT_SYMBOL_GPL(mark_page_dirty);
3358 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
3360 struct kvm_memory_slot *memslot;
3362 memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
3363 mark_page_dirty_in_slot(vcpu->kvm, memslot, gfn);
3365 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
3367 void kvm_sigset_activate(struct kvm_vcpu *vcpu)
3369 if (!vcpu->sigset_active)
3373 * This does a lockless modification of ->real_blocked, which is fine
3374 * because, only current can change ->real_blocked and all readers of
3375 * ->real_blocked don't care as long ->real_blocked is always a subset
3378 sigprocmask(SIG_SETMASK, &vcpu->sigset, ¤t->real_blocked);
3381 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu)
3383 if (!vcpu->sigset_active)
3386 sigprocmask(SIG_SETMASK, ¤t->real_blocked, NULL);
3387 sigemptyset(¤t->real_blocked);
3390 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
3392 unsigned int old, val, grow, grow_start;
3394 old = val = vcpu->halt_poll_ns;
3395 grow_start = READ_ONCE(halt_poll_ns_grow_start);
3396 grow = READ_ONCE(halt_poll_ns_grow);
3401 if (val < grow_start)
3404 vcpu->halt_poll_ns = val;
3406 trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
3409 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
3411 unsigned int old, val, shrink, grow_start;
3413 old = val = vcpu->halt_poll_ns;
3414 shrink = READ_ONCE(halt_poll_ns_shrink);
3415 grow_start = READ_ONCE(halt_poll_ns_grow_start);
3421 if (val < grow_start)
3424 vcpu->halt_poll_ns = val;
3425 trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
3428 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
3431 int idx = srcu_read_lock(&vcpu->kvm->srcu);
3433 if (kvm_arch_vcpu_runnable(vcpu))
3435 if (kvm_cpu_has_pending_timer(vcpu))
3437 if (signal_pending(current))
3439 if (kvm_check_request(KVM_REQ_UNBLOCK, vcpu))
3444 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3449 * Block the vCPU until the vCPU is runnable, an event arrives, or a signal is
3450 * pending. This is mostly used when halting a vCPU, but may also be used
3451 * directly for other vCPU non-runnable states, e.g. x86's Wait-For-SIPI.
3453 bool kvm_vcpu_block(struct kvm_vcpu *vcpu)
3455 struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu);
3456 bool waited = false;
3458 vcpu->stat.generic.blocking = 1;
3461 kvm_arch_vcpu_blocking(vcpu);
3462 prepare_to_rcuwait(wait);
3466 set_current_state(TASK_INTERRUPTIBLE);
3468 if (kvm_vcpu_check_block(vcpu) < 0)
3476 finish_rcuwait(wait);
3477 kvm_arch_vcpu_unblocking(vcpu);
3480 vcpu->stat.generic.blocking = 0;
3485 static inline void update_halt_poll_stats(struct kvm_vcpu *vcpu, ktime_t start,
3486 ktime_t end, bool success)
3488 struct kvm_vcpu_stat_generic *stats = &vcpu->stat.generic;
3489 u64 poll_ns = ktime_to_ns(ktime_sub(end, start));
3491 ++vcpu->stat.generic.halt_attempted_poll;
3494 ++vcpu->stat.generic.halt_successful_poll;
3496 if (!vcpu_valid_wakeup(vcpu))
3497 ++vcpu->stat.generic.halt_poll_invalid;
3499 stats->halt_poll_success_ns += poll_ns;
3500 KVM_STATS_LOG_HIST_UPDATE(stats->halt_poll_success_hist, poll_ns);
3502 stats->halt_poll_fail_ns += poll_ns;
3503 KVM_STATS_LOG_HIST_UPDATE(stats->halt_poll_fail_hist, poll_ns);
3507 static unsigned int kvm_vcpu_max_halt_poll_ns(struct kvm_vcpu *vcpu)
3509 struct kvm *kvm = vcpu->kvm;
3511 if (kvm->override_halt_poll_ns) {
3513 * Ensure kvm->max_halt_poll_ns is not read before
3514 * kvm->override_halt_poll_ns.
3516 * Pairs with the smp_wmb() when enabling KVM_CAP_HALT_POLL.
3519 return READ_ONCE(kvm->max_halt_poll_ns);
3522 return READ_ONCE(halt_poll_ns);
3526 * Emulate a vCPU halt condition, e.g. HLT on x86, WFI on arm, etc... If halt
3527 * polling is enabled, busy wait for a short time before blocking to avoid the
3528 * expensive block+unblock sequence if a wake event arrives soon after the vCPU
3531 void kvm_vcpu_halt(struct kvm_vcpu *vcpu)
3533 unsigned int max_halt_poll_ns = kvm_vcpu_max_halt_poll_ns(vcpu);
3534 bool halt_poll_allowed = !kvm_arch_no_poll(vcpu);
3535 ktime_t start, cur, poll_end;
3536 bool waited = false;
3540 if (vcpu->halt_poll_ns > max_halt_poll_ns)
3541 vcpu->halt_poll_ns = max_halt_poll_ns;
3543 do_halt_poll = halt_poll_allowed && vcpu->halt_poll_ns;
3545 start = cur = poll_end = ktime_get();
3547 ktime_t stop = ktime_add_ns(start, vcpu->halt_poll_ns);
3550 if (kvm_vcpu_check_block(vcpu) < 0)
3553 poll_end = cur = ktime_get();
3554 } while (kvm_vcpu_can_poll(cur, stop));
3557 waited = kvm_vcpu_block(vcpu);
3561 vcpu->stat.generic.halt_wait_ns +=
3562 ktime_to_ns(cur) - ktime_to_ns(poll_end);
3563 KVM_STATS_LOG_HIST_UPDATE(vcpu->stat.generic.halt_wait_hist,
3564 ktime_to_ns(cur) - ktime_to_ns(poll_end));
3567 /* The total time the vCPU was "halted", including polling time. */
3568 halt_ns = ktime_to_ns(cur) - ktime_to_ns(start);
3571 * Note, halt-polling is considered successful so long as the vCPU was
3572 * never actually scheduled out, i.e. even if the wake event arrived
3573 * after of the halt-polling loop itself, but before the full wait.
3576 update_halt_poll_stats(vcpu, start, poll_end, !waited);
3578 if (halt_poll_allowed) {
3579 /* Recompute the max halt poll time in case it changed. */
3580 max_halt_poll_ns = kvm_vcpu_max_halt_poll_ns(vcpu);
3582 if (!vcpu_valid_wakeup(vcpu)) {
3583 shrink_halt_poll_ns(vcpu);
3584 } else if (max_halt_poll_ns) {
3585 if (halt_ns <= vcpu->halt_poll_ns)
3587 /* we had a long block, shrink polling */
3588 else if (vcpu->halt_poll_ns &&
3589 halt_ns > max_halt_poll_ns)
3590 shrink_halt_poll_ns(vcpu);
3591 /* we had a short halt and our poll time is too small */
3592 else if (vcpu->halt_poll_ns < max_halt_poll_ns &&
3593 halt_ns < max_halt_poll_ns)
3594 grow_halt_poll_ns(vcpu);
3596 vcpu->halt_poll_ns = 0;
3600 trace_kvm_vcpu_wakeup(halt_ns, waited, vcpu_valid_wakeup(vcpu));
3602 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
3604 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
3606 if (__kvm_vcpu_wake_up(vcpu)) {
3607 WRITE_ONCE(vcpu->ready, true);
3608 ++vcpu->stat.generic.halt_wakeup;
3614 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
3618 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
3620 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
3624 if (kvm_vcpu_wake_up(vcpu))
3629 * The only state change done outside the vcpu mutex is IN_GUEST_MODE
3630 * to EXITING_GUEST_MODE. Therefore the moderately expensive "should
3631 * kick" check does not need atomic operations if kvm_vcpu_kick is used
3632 * within the vCPU thread itself.
3634 if (vcpu == __this_cpu_read(kvm_running_vcpu)) {
3635 if (vcpu->mode == IN_GUEST_MODE)
3636 WRITE_ONCE(vcpu->mode, EXITING_GUEST_MODE);
3641 * Note, the vCPU could get migrated to a different pCPU at any point
3642 * after kvm_arch_vcpu_should_kick(), which could result in sending an
3643 * IPI to the previous pCPU. But, that's ok because the purpose of the
3644 * IPI is to force the vCPU to leave IN_GUEST_MODE, and migrating the
3645 * vCPU also requires it to leave IN_GUEST_MODE.
3647 if (kvm_arch_vcpu_should_kick(vcpu)) {
3648 cpu = READ_ONCE(vcpu->cpu);
3649 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
3650 smp_send_reschedule(cpu);
3655 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
3656 #endif /* !CONFIG_S390 */
3658 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
3661 struct task_struct *task = NULL;
3665 pid = rcu_dereference(target->pid);
3667 task = get_pid_task(pid, PIDTYPE_PID);
3671 ret = yield_to(task, 1);
3672 put_task_struct(task);
3676 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
3679 * Helper that checks whether a VCPU is eligible for directed yield.
3680 * Most eligible candidate to yield is decided by following heuristics:
3682 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
3683 * (preempted lock holder), indicated by @in_spin_loop.
3684 * Set at the beginning and cleared at the end of interception/PLE handler.
3686 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
3687 * chance last time (mostly it has become eligible now since we have probably
3688 * yielded to lockholder in last iteration. This is done by toggling
3689 * @dy_eligible each time a VCPU checked for eligibility.)
3691 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
3692 * to preempted lock-holder could result in wrong VCPU selection and CPU
3693 * burning. Giving priority for a potential lock-holder increases lock
3696 * Since algorithm is based on heuristics, accessing another VCPU data without
3697 * locking does not harm. It may result in trying to yield to same VCPU, fail
3698 * and continue with next VCPU and so on.
3700 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
3702 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
3705 eligible = !vcpu->spin_loop.in_spin_loop ||
3706 vcpu->spin_loop.dy_eligible;
3708 if (vcpu->spin_loop.in_spin_loop)
3709 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
3718 * Unlike kvm_arch_vcpu_runnable, this function is called outside
3719 * a vcpu_load/vcpu_put pair. However, for most architectures
3720 * kvm_arch_vcpu_runnable does not require vcpu_load.
3722 bool __weak kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
3724 return kvm_arch_vcpu_runnable(vcpu);
3727 static bool vcpu_dy_runnable(struct kvm_vcpu *vcpu)
3729 if (kvm_arch_dy_runnable(vcpu))
3732 #ifdef CONFIG_KVM_ASYNC_PF
3733 if (!list_empty_careful(&vcpu->async_pf.done))
3740 bool __weak kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu)
3745 void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode)
3747 struct kvm *kvm = me->kvm;
3748 struct kvm_vcpu *vcpu;
3749 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
3755 kvm_vcpu_set_in_spin_loop(me, true);
3757 * We boost the priority of a VCPU that is runnable but not
3758 * currently running, because it got preempted by something
3759 * else and called schedule in __vcpu_run. Hopefully that
3760 * VCPU is holding the lock that we need and will release it.
3761 * We approximate round-robin by starting at the last boosted VCPU.
3763 for (pass = 0; pass < 2 && !yielded && try; pass++) {
3764 kvm_for_each_vcpu(i, vcpu, kvm) {
3765 if (!pass && i <= last_boosted_vcpu) {
3766 i = last_boosted_vcpu;
3768 } else if (pass && i > last_boosted_vcpu)
3770 if (!READ_ONCE(vcpu->ready))
3774 if (kvm_vcpu_is_blocking(vcpu) && !vcpu_dy_runnable(vcpu))
3776 if (READ_ONCE(vcpu->preempted) && yield_to_kernel_mode &&
3777 !kvm_arch_dy_has_pending_interrupt(vcpu) &&
3778 !kvm_arch_vcpu_in_kernel(vcpu))
3780 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
3783 yielded = kvm_vcpu_yield_to(vcpu);
3785 kvm->last_boosted_vcpu = i;
3787 } else if (yielded < 0) {
3794 kvm_vcpu_set_in_spin_loop(me, false);
3796 /* Ensure vcpu is not eligible during next spinloop */
3797 kvm_vcpu_set_dy_eligible(me, false);
3799 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
3801 static bool kvm_page_in_dirty_ring(struct kvm *kvm, unsigned long pgoff)
3803 #ifdef CONFIG_HAVE_KVM_DIRTY_RING
3804 return (pgoff >= KVM_DIRTY_LOG_PAGE_OFFSET) &&
3805 (pgoff < KVM_DIRTY_LOG_PAGE_OFFSET +
3806 kvm->dirty_ring_size / PAGE_SIZE);
3812 static vm_fault_t kvm_vcpu_fault(struct vm_fault *vmf)
3814 struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data;
3817 if (vmf->pgoff == 0)
3818 page = virt_to_page(vcpu->run);
3820 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
3821 page = virt_to_page(vcpu->arch.pio_data);
3823 #ifdef CONFIG_KVM_MMIO
3824 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
3825 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
3827 else if (kvm_page_in_dirty_ring(vcpu->kvm, vmf->pgoff))
3828 page = kvm_dirty_ring_get_page(
3830 vmf->pgoff - KVM_DIRTY_LOG_PAGE_OFFSET);
3832 return kvm_arch_vcpu_fault(vcpu, vmf);
3838 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
3839 .fault = kvm_vcpu_fault,
3842 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
3844 struct kvm_vcpu *vcpu = file->private_data;
3845 unsigned long pages = vma_pages(vma);
3847 if ((kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff) ||
3848 kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff + pages - 1)) &&
3849 ((vma->vm_flags & VM_EXEC) || !(vma->vm_flags & VM_SHARED)))
3852 vma->vm_ops = &kvm_vcpu_vm_ops;
3856 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
3858 struct kvm_vcpu *vcpu = filp->private_data;
3860 kvm_put_kvm(vcpu->kvm);
3864 static const struct file_operations kvm_vcpu_fops = {
3865 .release = kvm_vcpu_release,
3866 .unlocked_ioctl = kvm_vcpu_ioctl,
3867 .mmap = kvm_vcpu_mmap,
3868 .llseek = noop_llseek,
3869 KVM_COMPAT(kvm_vcpu_compat_ioctl),
3873 * Allocates an inode for the vcpu.
3875 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
3877 char name[8 + 1 + ITOA_MAX_LEN + 1];
3879 snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id);
3880 return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
3883 #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS
3884 static int vcpu_get_pid(void *data, u64 *val)
3886 struct kvm_vcpu *vcpu = data;
3889 *val = pid_nr(rcu_dereference(vcpu->pid));
3894 DEFINE_SIMPLE_ATTRIBUTE(vcpu_get_pid_fops, vcpu_get_pid, NULL, "%llu\n");
3896 static void kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
3898 struct dentry *debugfs_dentry;
3899 char dir_name[ITOA_MAX_LEN * 2];
3901 if (!debugfs_initialized())
3904 snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
3905 debugfs_dentry = debugfs_create_dir(dir_name,
3906 vcpu->kvm->debugfs_dentry);
3907 debugfs_create_file("pid", 0444, debugfs_dentry, vcpu,
3908 &vcpu_get_pid_fops);
3910 kvm_arch_create_vcpu_debugfs(vcpu, debugfs_dentry);
3915 * Creates some virtual cpus. Good luck creating more than one.
3917 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
3920 struct kvm_vcpu *vcpu;
3923 if (id >= KVM_MAX_VCPU_IDS)
3926 mutex_lock(&kvm->lock);
3927 if (kvm->created_vcpus >= kvm->max_vcpus) {
3928 mutex_unlock(&kvm->lock);
3932 r = kvm_arch_vcpu_precreate(kvm, id);
3934 mutex_unlock(&kvm->lock);
3938 kvm->created_vcpus++;
3939 mutex_unlock(&kvm->lock);
3941 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL_ACCOUNT);
3944 goto vcpu_decrement;
3947 BUILD_BUG_ON(sizeof(struct kvm_run) > PAGE_SIZE);
3948 page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
3953 vcpu->run = page_address(page);
3955 kvm_vcpu_init(vcpu, kvm, id);
3957 r = kvm_arch_vcpu_create(vcpu);
3959 goto vcpu_free_run_page;
3961 if (kvm->dirty_ring_size) {
3962 r = kvm_dirty_ring_alloc(&vcpu->dirty_ring,
3963 id, kvm->dirty_ring_size);
3965 goto arch_vcpu_destroy;
3968 mutex_lock(&kvm->lock);
3970 #ifdef CONFIG_LOCKDEP
3971 /* Ensure that lockdep knows vcpu->mutex is taken *inside* kvm->lock */
3972 mutex_lock(&vcpu->mutex);
3973 mutex_unlock(&vcpu->mutex);
3976 if (kvm_get_vcpu_by_id(kvm, id)) {
3978 goto unlock_vcpu_destroy;
3981 vcpu->vcpu_idx = atomic_read(&kvm->online_vcpus);
3982 r = xa_reserve(&kvm->vcpu_array, vcpu->vcpu_idx, GFP_KERNEL_ACCOUNT);
3984 goto unlock_vcpu_destroy;
3986 /* Now it's all set up, let userspace reach it */
3988 r = create_vcpu_fd(vcpu);
3990 goto kvm_put_xa_release;
3992 if (KVM_BUG_ON(xa_store(&kvm->vcpu_array, vcpu->vcpu_idx, vcpu, 0), kvm)) {
3994 goto kvm_put_xa_release;
3998 * Pairs with smp_rmb() in kvm_get_vcpu. Store the vcpu
3999 * pointer before kvm->online_vcpu's incremented value.
4002 atomic_inc(&kvm->online_vcpus);
4004 mutex_unlock(&kvm->lock);
4005 kvm_arch_vcpu_postcreate(vcpu);
4006 kvm_create_vcpu_debugfs(vcpu);
4010 kvm_put_kvm_no_destroy(kvm);
4011 xa_release(&kvm->vcpu_array, vcpu->vcpu_idx);
4012 unlock_vcpu_destroy:
4013 mutex_unlock(&kvm->lock);
4014 kvm_dirty_ring_free(&vcpu->dirty_ring);
4016 kvm_arch_vcpu_destroy(vcpu);
4018 free_page((unsigned long)vcpu->run);
4020 kmem_cache_free(kvm_vcpu_cache, vcpu);
4022 mutex_lock(&kvm->lock);
4023 kvm->created_vcpus--;
4024 mutex_unlock(&kvm->lock);
4028 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
4031 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
4032 vcpu->sigset_active = 1;
4033 vcpu->sigset = *sigset;
4035 vcpu->sigset_active = 0;
4039 static ssize_t kvm_vcpu_stats_read(struct file *file, char __user *user_buffer,
4040 size_t size, loff_t *offset)
4042 struct kvm_vcpu *vcpu = file->private_data;
4044 return kvm_stats_read(vcpu->stats_id, &kvm_vcpu_stats_header,
4045 &kvm_vcpu_stats_desc[0], &vcpu->stat,
4046 sizeof(vcpu->stat), user_buffer, size, offset);
4049 static const struct file_operations kvm_vcpu_stats_fops = {
4050 .read = kvm_vcpu_stats_read,
4051 .llseek = noop_llseek,
4054 static int kvm_vcpu_ioctl_get_stats_fd(struct kvm_vcpu *vcpu)
4058 char name[15 + ITOA_MAX_LEN + 1];
4060 snprintf(name, sizeof(name), "kvm-vcpu-stats:%d", vcpu->vcpu_id);
4062 fd = get_unused_fd_flags(O_CLOEXEC);
4066 file = anon_inode_getfile(name, &kvm_vcpu_stats_fops, vcpu, O_RDONLY);
4069 return PTR_ERR(file);
4071 file->f_mode |= FMODE_PREAD;
4072 fd_install(fd, file);
4077 static long kvm_vcpu_ioctl(struct file *filp,
4078 unsigned int ioctl, unsigned long arg)
4080 struct kvm_vcpu *vcpu = filp->private_data;
4081 void __user *argp = (void __user *)arg;
4083 struct kvm_fpu *fpu = NULL;
4084 struct kvm_sregs *kvm_sregs = NULL;
4086 if (vcpu->kvm->mm != current->mm || vcpu->kvm->vm_dead)
4089 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
4093 * Some architectures have vcpu ioctls that are asynchronous to vcpu
4094 * execution; mutex_lock() would break them.
4096 r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg);
4097 if (r != -ENOIOCTLCMD)
4100 if (mutex_lock_killable(&vcpu->mutex))
4108 oldpid = rcu_access_pointer(vcpu->pid);
4109 if (unlikely(oldpid != task_pid(current))) {
4110 /* The thread running this VCPU changed. */
4113 r = kvm_arch_vcpu_run_pid_change(vcpu);
4117 newpid = get_task_pid(current, PIDTYPE_PID);
4118 rcu_assign_pointer(vcpu->pid, newpid);
4123 r = kvm_arch_vcpu_ioctl_run(vcpu);
4124 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
4127 case KVM_GET_REGS: {
4128 struct kvm_regs *kvm_regs;
4131 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL_ACCOUNT);
4134 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
4138 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
4145 case KVM_SET_REGS: {
4146 struct kvm_regs *kvm_regs;
4148 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
4149 if (IS_ERR(kvm_regs)) {
4150 r = PTR_ERR(kvm_regs);
4153 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
4157 case KVM_GET_SREGS: {
4158 kvm_sregs = kzalloc(sizeof(struct kvm_sregs),
4159 GFP_KERNEL_ACCOUNT);
4163 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
4167 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
4172 case KVM_SET_SREGS: {
4173 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
4174 if (IS_ERR(kvm_sregs)) {
4175 r = PTR_ERR(kvm_sregs);
4179 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
4182 case KVM_GET_MP_STATE: {
4183 struct kvm_mp_state mp_state;
4185 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
4189 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
4194 case KVM_SET_MP_STATE: {
4195 struct kvm_mp_state mp_state;
4198 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
4200 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
4203 case KVM_TRANSLATE: {
4204 struct kvm_translation tr;
4207 if (copy_from_user(&tr, argp, sizeof(tr)))
4209 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
4213 if (copy_to_user(argp, &tr, sizeof(tr)))
4218 case KVM_SET_GUEST_DEBUG: {
4219 struct kvm_guest_debug dbg;
4222 if (copy_from_user(&dbg, argp, sizeof(dbg)))
4224 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
4227 case KVM_SET_SIGNAL_MASK: {
4228 struct kvm_signal_mask __user *sigmask_arg = argp;
4229 struct kvm_signal_mask kvm_sigmask;
4230 sigset_t sigset, *p;
4235 if (copy_from_user(&kvm_sigmask, argp,
4236 sizeof(kvm_sigmask)))
4239 if (kvm_sigmask.len != sizeof(sigset))
4242 if (copy_from_user(&sigset, sigmask_arg->sigset,
4247 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
4251 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL_ACCOUNT);
4255 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
4259 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
4265 fpu = memdup_user(argp, sizeof(*fpu));
4271 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
4274 case KVM_GET_STATS_FD: {
4275 r = kvm_vcpu_ioctl_get_stats_fd(vcpu);
4279 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
4282 mutex_unlock(&vcpu->mutex);
4288 #ifdef CONFIG_KVM_COMPAT
4289 static long kvm_vcpu_compat_ioctl(struct file *filp,
4290 unsigned int ioctl, unsigned long arg)
4292 struct kvm_vcpu *vcpu = filp->private_data;
4293 void __user *argp = compat_ptr(arg);
4296 if (vcpu->kvm->mm != current->mm || vcpu->kvm->vm_dead)
4300 case KVM_SET_SIGNAL_MASK: {
4301 struct kvm_signal_mask __user *sigmask_arg = argp;
4302 struct kvm_signal_mask kvm_sigmask;
4307 if (copy_from_user(&kvm_sigmask, argp,
4308 sizeof(kvm_sigmask)))
4311 if (kvm_sigmask.len != sizeof(compat_sigset_t))
4314 if (get_compat_sigset(&sigset,
4315 (compat_sigset_t __user *)sigmask_arg->sigset))
4317 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
4319 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
4323 r = kvm_vcpu_ioctl(filp, ioctl, arg);
4331 static int kvm_device_mmap(struct file *filp, struct vm_area_struct *vma)
4333 struct kvm_device *dev = filp->private_data;
4336 return dev->ops->mmap(dev, vma);
4341 static int kvm_device_ioctl_attr(struct kvm_device *dev,
4342 int (*accessor)(struct kvm_device *dev,
4343 struct kvm_device_attr *attr),
4346 struct kvm_device_attr attr;
4351 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
4354 return accessor(dev, &attr);
4357 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
4360 struct kvm_device *dev = filp->private_data;
4362 if (dev->kvm->mm != current->mm || dev->kvm->vm_dead)
4366 case KVM_SET_DEVICE_ATTR:
4367 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
4368 case KVM_GET_DEVICE_ATTR:
4369 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
4370 case KVM_HAS_DEVICE_ATTR:
4371 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
4373 if (dev->ops->ioctl)
4374 return dev->ops->ioctl(dev, ioctl, arg);
4380 static int kvm_device_release(struct inode *inode, struct file *filp)
4382 struct kvm_device *dev = filp->private_data;
4383 struct kvm *kvm = dev->kvm;
4385 if (dev->ops->release) {
4386 mutex_lock(&kvm->lock);
4387 list_del(&dev->vm_node);
4388 dev->ops->release(dev);
4389 mutex_unlock(&kvm->lock);
4396 static const struct file_operations kvm_device_fops = {
4397 .unlocked_ioctl = kvm_device_ioctl,
4398 .release = kvm_device_release,
4399 KVM_COMPAT(kvm_device_ioctl),
4400 .mmap = kvm_device_mmap,
4403 struct kvm_device *kvm_device_from_filp(struct file *filp)
4405 if (filp->f_op != &kvm_device_fops)
4408 return filp->private_data;
4411 static const struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
4412 #ifdef CONFIG_KVM_MPIC
4413 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
4414 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
4418 int kvm_register_device_ops(const struct kvm_device_ops *ops, u32 type)
4420 if (type >= ARRAY_SIZE(kvm_device_ops_table))
4423 if (kvm_device_ops_table[type] != NULL)
4426 kvm_device_ops_table[type] = ops;
4430 void kvm_unregister_device_ops(u32 type)
4432 if (kvm_device_ops_table[type] != NULL)
4433 kvm_device_ops_table[type] = NULL;
4436 static int kvm_ioctl_create_device(struct kvm *kvm,
4437 struct kvm_create_device *cd)
4439 const struct kvm_device_ops *ops;
4440 struct kvm_device *dev;
4441 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
4445 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
4448 type = array_index_nospec(cd->type, ARRAY_SIZE(kvm_device_ops_table));
4449 ops = kvm_device_ops_table[type];
4456 dev = kzalloc(sizeof(*dev), GFP_KERNEL_ACCOUNT);
4463 mutex_lock(&kvm->lock);
4464 ret = ops->create(dev, type);
4466 mutex_unlock(&kvm->lock);
4470 list_add(&dev->vm_node, &kvm->devices);
4471 mutex_unlock(&kvm->lock);
4477 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
4479 kvm_put_kvm_no_destroy(kvm);
4480 mutex_lock(&kvm->lock);
4481 list_del(&dev->vm_node);
4484 mutex_unlock(&kvm->lock);
4494 static int kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
4497 case KVM_CAP_USER_MEMORY:
4498 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
4499 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
4500 case KVM_CAP_INTERNAL_ERROR_DATA:
4501 #ifdef CONFIG_HAVE_KVM_MSI
4502 case KVM_CAP_SIGNAL_MSI:
4504 #ifdef CONFIG_HAVE_KVM_IRQFD
4507 case KVM_CAP_IOEVENTFD_ANY_LENGTH:
4508 case KVM_CAP_CHECK_EXTENSION_VM:
4509 case KVM_CAP_ENABLE_CAP_VM:
4510 case KVM_CAP_HALT_POLL:
4512 #ifdef CONFIG_KVM_MMIO
4513 case KVM_CAP_COALESCED_MMIO:
4514 return KVM_COALESCED_MMIO_PAGE_OFFSET;
4515 case KVM_CAP_COALESCED_PIO:
4518 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4519 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2:
4520 return KVM_DIRTY_LOG_MANUAL_CAPS;
4522 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
4523 case KVM_CAP_IRQ_ROUTING:
4524 return KVM_MAX_IRQ_ROUTES;
4526 #if KVM_ADDRESS_SPACE_NUM > 1
4527 case KVM_CAP_MULTI_ADDRESS_SPACE:
4528 return KVM_ADDRESS_SPACE_NUM;
4530 case KVM_CAP_NR_MEMSLOTS:
4531 return KVM_USER_MEM_SLOTS;
4532 case KVM_CAP_DIRTY_LOG_RING:
4533 #ifdef CONFIG_HAVE_KVM_DIRTY_RING_TSO
4534 return KVM_DIRTY_RING_MAX_ENTRIES * sizeof(struct kvm_dirty_gfn);
4538 case KVM_CAP_DIRTY_LOG_RING_ACQ_REL:
4539 #ifdef CONFIG_HAVE_KVM_DIRTY_RING_ACQ_REL
4540 return KVM_DIRTY_RING_MAX_ENTRIES * sizeof(struct kvm_dirty_gfn);
4544 #ifdef CONFIG_NEED_KVM_DIRTY_RING_WITH_BITMAP
4545 case KVM_CAP_DIRTY_LOG_RING_WITH_BITMAP:
4547 case KVM_CAP_BINARY_STATS_FD:
4548 case KVM_CAP_SYSTEM_EVENT_DATA:
4553 return kvm_vm_ioctl_check_extension(kvm, arg);
4556 static int kvm_vm_ioctl_enable_dirty_log_ring(struct kvm *kvm, u32 size)
4560 if (!KVM_DIRTY_LOG_PAGE_OFFSET)
4563 /* the size should be power of 2 */
4564 if (!size || (size & (size - 1)))
4567 /* Should be bigger to keep the reserved entries, or a page */
4568 if (size < kvm_dirty_ring_get_rsvd_entries() *
4569 sizeof(struct kvm_dirty_gfn) || size < PAGE_SIZE)
4572 if (size > KVM_DIRTY_RING_MAX_ENTRIES *
4573 sizeof(struct kvm_dirty_gfn))
4576 /* We only allow it to set once */
4577 if (kvm->dirty_ring_size)
4580 mutex_lock(&kvm->lock);
4582 if (kvm->created_vcpus) {
4583 /* We don't allow to change this value after vcpu created */
4586 kvm->dirty_ring_size = size;
4590 mutex_unlock(&kvm->lock);
4594 static int kvm_vm_ioctl_reset_dirty_pages(struct kvm *kvm)
4597 struct kvm_vcpu *vcpu;
4600 if (!kvm->dirty_ring_size)
4603 mutex_lock(&kvm->slots_lock);
4605 kvm_for_each_vcpu(i, vcpu, kvm)
4606 cleared += kvm_dirty_ring_reset(vcpu->kvm, &vcpu->dirty_ring);
4608 mutex_unlock(&kvm->slots_lock);
4611 kvm_flush_remote_tlbs(kvm);
4616 int __attribute__((weak)) kvm_vm_ioctl_enable_cap(struct kvm *kvm,
4617 struct kvm_enable_cap *cap)
4622 bool kvm_are_all_memslots_empty(struct kvm *kvm)
4626 lockdep_assert_held(&kvm->slots_lock);
4628 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
4629 if (!kvm_memslots_empty(__kvm_memslots(kvm, i)))
4635 EXPORT_SYMBOL_GPL(kvm_are_all_memslots_empty);
4637 static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm,
4638 struct kvm_enable_cap *cap)
4641 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4642 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2: {
4643 u64 allowed_options = KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE;
4645 if (cap->args[0] & KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE)
4646 allowed_options = KVM_DIRTY_LOG_MANUAL_CAPS;
4648 if (cap->flags || (cap->args[0] & ~allowed_options))
4650 kvm->manual_dirty_log_protect = cap->args[0];
4654 case KVM_CAP_HALT_POLL: {
4655 if (cap->flags || cap->args[0] != (unsigned int)cap->args[0])
4658 kvm->max_halt_poll_ns = cap->args[0];
4661 * Ensure kvm->override_halt_poll_ns does not become visible
4662 * before kvm->max_halt_poll_ns.
4664 * Pairs with the smp_rmb() in kvm_vcpu_max_halt_poll_ns().
4667 kvm->override_halt_poll_ns = true;
4671 case KVM_CAP_DIRTY_LOG_RING:
4672 case KVM_CAP_DIRTY_LOG_RING_ACQ_REL:
4673 if (!kvm_vm_ioctl_check_extension_generic(kvm, cap->cap))
4676 return kvm_vm_ioctl_enable_dirty_log_ring(kvm, cap->args[0]);
4677 case KVM_CAP_DIRTY_LOG_RING_WITH_BITMAP: {
4680 if (!IS_ENABLED(CONFIG_NEED_KVM_DIRTY_RING_WITH_BITMAP) ||
4681 !kvm->dirty_ring_size || cap->flags)
4684 mutex_lock(&kvm->slots_lock);
4687 * For simplicity, allow enabling ring+bitmap if and only if
4688 * there are no memslots, e.g. to ensure all memslots allocate
4689 * a bitmap after the capability is enabled.
4691 if (kvm_are_all_memslots_empty(kvm)) {
4692 kvm->dirty_ring_with_bitmap = true;
4696 mutex_unlock(&kvm->slots_lock);
4701 return kvm_vm_ioctl_enable_cap(kvm, cap);
4705 static ssize_t kvm_vm_stats_read(struct file *file, char __user *user_buffer,
4706 size_t size, loff_t *offset)
4708 struct kvm *kvm = file->private_data;
4710 return kvm_stats_read(kvm->stats_id, &kvm_vm_stats_header,
4711 &kvm_vm_stats_desc[0], &kvm->stat,
4712 sizeof(kvm->stat), user_buffer, size, offset);
4715 static const struct file_operations kvm_vm_stats_fops = {
4716 .read = kvm_vm_stats_read,
4717 .llseek = noop_llseek,
4720 static int kvm_vm_ioctl_get_stats_fd(struct kvm *kvm)
4725 fd = get_unused_fd_flags(O_CLOEXEC);
4729 file = anon_inode_getfile("kvm-vm-stats",
4730 &kvm_vm_stats_fops, kvm, O_RDONLY);
4733 return PTR_ERR(file);
4735 file->f_mode |= FMODE_PREAD;
4736 fd_install(fd, file);
4741 static long kvm_vm_ioctl(struct file *filp,
4742 unsigned int ioctl, unsigned long arg)
4744 struct kvm *kvm = filp->private_data;
4745 void __user *argp = (void __user *)arg;
4748 if (kvm->mm != current->mm || kvm->vm_dead)
4751 case KVM_CREATE_VCPU:
4752 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
4754 case KVM_ENABLE_CAP: {
4755 struct kvm_enable_cap cap;
4758 if (copy_from_user(&cap, argp, sizeof(cap)))
4760 r = kvm_vm_ioctl_enable_cap_generic(kvm, &cap);
4763 case KVM_SET_USER_MEMORY_REGION: {
4764 struct kvm_userspace_memory_region kvm_userspace_mem;
4767 if (copy_from_user(&kvm_userspace_mem, argp,
4768 sizeof(kvm_userspace_mem)))
4771 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
4774 case KVM_GET_DIRTY_LOG: {
4775 struct kvm_dirty_log log;
4778 if (copy_from_user(&log, argp, sizeof(log)))
4780 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
4783 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4784 case KVM_CLEAR_DIRTY_LOG: {
4785 struct kvm_clear_dirty_log log;
4788 if (copy_from_user(&log, argp, sizeof(log)))
4790 r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
4794 #ifdef CONFIG_KVM_MMIO
4795 case KVM_REGISTER_COALESCED_MMIO: {
4796 struct kvm_coalesced_mmio_zone zone;
4799 if (copy_from_user(&zone, argp, sizeof(zone)))
4801 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
4804 case KVM_UNREGISTER_COALESCED_MMIO: {
4805 struct kvm_coalesced_mmio_zone zone;
4808 if (copy_from_user(&zone, argp, sizeof(zone)))
4810 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
4815 struct kvm_irqfd data;
4818 if (copy_from_user(&data, argp, sizeof(data)))
4820 r = kvm_irqfd(kvm, &data);
4823 case KVM_IOEVENTFD: {
4824 struct kvm_ioeventfd data;
4827 if (copy_from_user(&data, argp, sizeof(data)))
4829 r = kvm_ioeventfd(kvm, &data);
4832 #ifdef CONFIG_HAVE_KVM_MSI
4833 case KVM_SIGNAL_MSI: {
4837 if (copy_from_user(&msi, argp, sizeof(msi)))
4839 r = kvm_send_userspace_msi(kvm, &msi);
4843 #ifdef __KVM_HAVE_IRQ_LINE
4844 case KVM_IRQ_LINE_STATUS:
4845 case KVM_IRQ_LINE: {
4846 struct kvm_irq_level irq_event;
4849 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
4852 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
4853 ioctl == KVM_IRQ_LINE_STATUS);
4858 if (ioctl == KVM_IRQ_LINE_STATUS) {
4859 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
4867 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
4868 case KVM_SET_GSI_ROUTING: {
4869 struct kvm_irq_routing routing;
4870 struct kvm_irq_routing __user *urouting;
4871 struct kvm_irq_routing_entry *entries = NULL;
4874 if (copy_from_user(&routing, argp, sizeof(routing)))
4877 if (!kvm_arch_can_set_irq_routing(kvm))
4879 if (routing.nr > KVM_MAX_IRQ_ROUTES)
4885 entries = vmemdup_user(urouting->entries,
4886 array_size(sizeof(*entries),
4888 if (IS_ERR(entries)) {
4889 r = PTR_ERR(entries);
4893 r = kvm_set_irq_routing(kvm, entries, routing.nr,
4898 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
4899 case KVM_CREATE_DEVICE: {
4900 struct kvm_create_device cd;
4903 if (copy_from_user(&cd, argp, sizeof(cd)))
4906 r = kvm_ioctl_create_device(kvm, &cd);
4911 if (copy_to_user(argp, &cd, sizeof(cd)))
4917 case KVM_CHECK_EXTENSION:
4918 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
4920 case KVM_RESET_DIRTY_RINGS:
4921 r = kvm_vm_ioctl_reset_dirty_pages(kvm);
4923 case KVM_GET_STATS_FD:
4924 r = kvm_vm_ioctl_get_stats_fd(kvm);
4927 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
4933 #ifdef CONFIG_KVM_COMPAT
4934 struct compat_kvm_dirty_log {
4938 compat_uptr_t dirty_bitmap; /* one bit per page */
4943 struct compat_kvm_clear_dirty_log {
4948 compat_uptr_t dirty_bitmap; /* one bit per page */
4953 long __weak kvm_arch_vm_compat_ioctl(struct file *filp, unsigned int ioctl,
4959 static long kvm_vm_compat_ioctl(struct file *filp,
4960 unsigned int ioctl, unsigned long arg)
4962 struct kvm *kvm = filp->private_data;
4965 if (kvm->mm != current->mm || kvm->vm_dead)
4968 r = kvm_arch_vm_compat_ioctl(filp, ioctl, arg);
4973 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4974 case KVM_CLEAR_DIRTY_LOG: {
4975 struct compat_kvm_clear_dirty_log compat_log;
4976 struct kvm_clear_dirty_log log;
4978 if (copy_from_user(&compat_log, (void __user *)arg,
4979 sizeof(compat_log)))
4981 log.slot = compat_log.slot;
4982 log.num_pages = compat_log.num_pages;
4983 log.first_page = compat_log.first_page;
4984 log.padding2 = compat_log.padding2;
4985 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
4987 r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
4991 case KVM_GET_DIRTY_LOG: {
4992 struct compat_kvm_dirty_log compat_log;
4993 struct kvm_dirty_log log;
4995 if (copy_from_user(&compat_log, (void __user *)arg,
4996 sizeof(compat_log)))
4998 log.slot = compat_log.slot;
4999 log.padding1 = compat_log.padding1;
5000 log.padding2 = compat_log.padding2;
5001 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
5003 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
5007 r = kvm_vm_ioctl(filp, ioctl, arg);
5013 static const struct file_operations kvm_vm_fops = {
5014 .release = kvm_vm_release,
5015 .unlocked_ioctl = kvm_vm_ioctl,
5016 .llseek = noop_llseek,
5017 KVM_COMPAT(kvm_vm_compat_ioctl),
5020 bool file_is_kvm(struct file *file)
5022 return file && file->f_op == &kvm_vm_fops;
5024 EXPORT_SYMBOL_GPL(file_is_kvm);
5026 static int kvm_dev_ioctl_create_vm(unsigned long type)
5028 char fdname[ITOA_MAX_LEN + 1];
5033 fd = get_unused_fd_flags(O_CLOEXEC);
5037 snprintf(fdname, sizeof(fdname), "%d", fd);
5039 kvm = kvm_create_vm(type, fdname);
5045 file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
5052 * Don't call kvm_put_kvm anymore at this point; file->f_op is
5053 * already set, with ->release() being kvm_vm_release(). In error
5054 * cases it will be called by the final fput(file) and will take
5055 * care of doing kvm_put_kvm(kvm).
5057 kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm);
5059 fd_install(fd, file);
5069 static long kvm_dev_ioctl(struct file *filp,
5070 unsigned int ioctl, unsigned long arg)
5075 case KVM_GET_API_VERSION:
5078 r = KVM_API_VERSION;
5081 r = kvm_dev_ioctl_create_vm(arg);
5083 case KVM_CHECK_EXTENSION:
5084 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
5086 case KVM_GET_VCPU_MMAP_SIZE:
5089 r = PAGE_SIZE; /* struct kvm_run */
5091 r += PAGE_SIZE; /* pio data page */
5093 #ifdef CONFIG_KVM_MMIO
5094 r += PAGE_SIZE; /* coalesced mmio ring page */
5097 case KVM_TRACE_ENABLE:
5098 case KVM_TRACE_PAUSE:
5099 case KVM_TRACE_DISABLE:
5103 return kvm_arch_dev_ioctl(filp, ioctl, arg);
5109 static struct file_operations kvm_chardev_ops = {
5110 .unlocked_ioctl = kvm_dev_ioctl,
5111 .llseek = noop_llseek,
5112 KVM_COMPAT(kvm_dev_ioctl),
5115 static struct miscdevice kvm_dev = {
5121 #ifdef CONFIG_KVM_GENERIC_HARDWARE_ENABLING
5122 __visible bool kvm_rebooting;
5123 EXPORT_SYMBOL_GPL(kvm_rebooting);
5125 static DEFINE_PER_CPU(bool, hardware_enabled);
5126 static int kvm_usage_count;
5128 static int __hardware_enable_nolock(void)
5130 if (__this_cpu_read(hardware_enabled))
5133 if (kvm_arch_hardware_enable()) {
5134 pr_info("kvm: enabling virtualization on CPU%d failed\n",
5135 raw_smp_processor_id());
5139 __this_cpu_write(hardware_enabled, true);
5143 static void hardware_enable_nolock(void *failed)
5145 if (__hardware_enable_nolock())
5149 static int kvm_online_cpu(unsigned int cpu)
5154 * Abort the CPU online process if hardware virtualization cannot
5155 * be enabled. Otherwise running VMs would encounter unrecoverable
5156 * errors when scheduled to this CPU.
5158 mutex_lock(&kvm_lock);
5159 if (kvm_usage_count)
5160 ret = __hardware_enable_nolock();
5161 mutex_unlock(&kvm_lock);
5165 static void hardware_disable_nolock(void *junk)
5168 * Note, hardware_disable_all_nolock() tells all online CPUs to disable
5169 * hardware, not just CPUs that successfully enabled hardware!
5171 if (!__this_cpu_read(hardware_enabled))
5174 kvm_arch_hardware_disable();
5176 __this_cpu_write(hardware_enabled, false);
5179 static int kvm_offline_cpu(unsigned int cpu)
5181 mutex_lock(&kvm_lock);
5182 if (kvm_usage_count)
5183 hardware_disable_nolock(NULL);
5184 mutex_unlock(&kvm_lock);
5188 static void hardware_disable_all_nolock(void)
5190 BUG_ON(!kvm_usage_count);
5193 if (!kvm_usage_count)
5194 on_each_cpu(hardware_disable_nolock, NULL, 1);
5197 static void hardware_disable_all(void)
5200 mutex_lock(&kvm_lock);
5201 hardware_disable_all_nolock();
5202 mutex_unlock(&kvm_lock);
5206 static int hardware_enable_all(void)
5208 atomic_t failed = ATOMIC_INIT(0);
5212 * Do not enable hardware virtualization if the system is going down.
5213 * If userspace initiated a forced reboot, e.g. reboot -f, then it's
5214 * possible for an in-flight KVM_CREATE_VM to trigger hardware enabling
5215 * after kvm_reboot() is called. Note, this relies on system_state
5216 * being set _before_ kvm_reboot(), which is why KVM uses a syscore ops
5217 * hook instead of registering a dedicated reboot notifier (the latter
5218 * runs before system_state is updated).
5220 if (system_state == SYSTEM_HALT || system_state == SYSTEM_POWER_OFF ||
5221 system_state == SYSTEM_RESTART)
5225 * When onlining a CPU, cpu_online_mask is set before kvm_online_cpu()
5226 * is called, and so on_each_cpu() between them includes the CPU that
5227 * is being onlined. As a result, hardware_enable_nolock() may get
5228 * invoked before kvm_online_cpu(), which also enables hardware if the
5229 * usage count is non-zero. Disable CPU hotplug to avoid attempting to
5230 * enable hardware multiple times.
5233 mutex_lock(&kvm_lock);
5238 if (kvm_usage_count == 1) {
5239 on_each_cpu(hardware_enable_nolock, &failed, 1);
5241 if (atomic_read(&failed)) {
5242 hardware_disable_all_nolock();
5247 mutex_unlock(&kvm_lock);
5253 static void kvm_shutdown(void)
5256 * Disable hardware virtualization and set kvm_rebooting to indicate
5257 * that KVM has asynchronously disabled hardware virtualization, i.e.
5258 * that relevant errors and exceptions aren't entirely unexpected.
5259 * Some flavors of hardware virtualization need to be disabled before
5260 * transferring control to firmware (to perform shutdown/reboot), e.g.
5261 * on x86, virtualization can block INIT interrupts, which are used by
5262 * firmware to pull APs back under firmware control. Note, this path
5263 * is used for both shutdown and reboot scenarios, i.e. neither name is
5264 * 100% comprehensive.
5266 pr_info("kvm: exiting hardware virtualization\n");
5267 kvm_rebooting = true;
5268 on_each_cpu(hardware_disable_nolock, NULL, 1);
5271 static int kvm_suspend(void)
5274 * Secondary CPUs and CPU hotplug are disabled across the suspend/resume
5275 * callbacks, i.e. no need to acquire kvm_lock to ensure the usage count
5276 * is stable. Assert that kvm_lock is not held to ensure the system
5277 * isn't suspended while KVM is enabling hardware. Hardware enabling
5278 * can be preempted, but the task cannot be frozen until it has dropped
5279 * all locks (userspace tasks are frozen via a fake signal).
5281 lockdep_assert_not_held(&kvm_lock);
5282 lockdep_assert_irqs_disabled();
5284 if (kvm_usage_count)
5285 hardware_disable_nolock(NULL);
5289 static void kvm_resume(void)
5291 lockdep_assert_not_held(&kvm_lock);
5292 lockdep_assert_irqs_disabled();
5294 if (kvm_usage_count)
5295 WARN_ON_ONCE(__hardware_enable_nolock());
5298 static struct syscore_ops kvm_syscore_ops = {
5299 .suspend = kvm_suspend,
5300 .resume = kvm_resume,
5301 .shutdown = kvm_shutdown,
5303 #else /* CONFIG_KVM_GENERIC_HARDWARE_ENABLING */
5304 static int hardware_enable_all(void)
5309 static void hardware_disable_all(void)
5313 #endif /* CONFIG_KVM_GENERIC_HARDWARE_ENABLING */
5315 static void kvm_iodevice_destructor(struct kvm_io_device *dev)
5317 if (dev->ops->destructor)
5318 dev->ops->destructor(dev);
5321 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
5325 for (i = 0; i < bus->dev_count; i++) {
5326 struct kvm_io_device *pos = bus->range[i].dev;
5328 kvm_iodevice_destructor(pos);
5333 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
5334 const struct kvm_io_range *r2)
5336 gpa_t addr1 = r1->addr;
5337 gpa_t addr2 = r2->addr;
5342 /* If r2->len == 0, match the exact address. If r2->len != 0,
5343 * accept any overlapping write. Any order is acceptable for
5344 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
5345 * we process all of them.
5358 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
5360 return kvm_io_bus_cmp(p1, p2);
5363 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
5364 gpa_t addr, int len)
5366 struct kvm_io_range *range, key;
5369 key = (struct kvm_io_range) {
5374 range = bsearch(&key, bus->range, bus->dev_count,
5375 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
5379 off = range - bus->range;
5381 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
5387 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
5388 struct kvm_io_range *range, const void *val)
5392 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
5396 while (idx < bus->dev_count &&
5397 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
5398 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
5407 /* kvm_io_bus_write - called under kvm->slots_lock */
5408 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
5409 int len, const void *val)
5411 struct kvm_io_bus *bus;
5412 struct kvm_io_range range;
5415 range = (struct kvm_io_range) {
5420 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
5423 r = __kvm_io_bus_write(vcpu, bus, &range, val);
5424 return r < 0 ? r : 0;
5426 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
5428 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
5429 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
5430 gpa_t addr, int len, const void *val, long cookie)
5432 struct kvm_io_bus *bus;
5433 struct kvm_io_range range;
5435 range = (struct kvm_io_range) {
5440 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
5444 /* First try the device referenced by cookie. */
5445 if ((cookie >= 0) && (cookie < bus->dev_count) &&
5446 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
5447 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
5452 * cookie contained garbage; fall back to search and return the
5453 * correct cookie value.
5455 return __kvm_io_bus_write(vcpu, bus, &range, val);
5458 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
5459 struct kvm_io_range *range, void *val)
5463 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
5467 while (idx < bus->dev_count &&
5468 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
5469 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
5478 /* kvm_io_bus_read - called under kvm->slots_lock */
5479 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
5482 struct kvm_io_bus *bus;
5483 struct kvm_io_range range;
5486 range = (struct kvm_io_range) {
5491 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
5494 r = __kvm_io_bus_read(vcpu, bus, &range, val);
5495 return r < 0 ? r : 0;
5498 /* Caller must hold slots_lock. */
5499 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
5500 int len, struct kvm_io_device *dev)
5503 struct kvm_io_bus *new_bus, *bus;
5504 struct kvm_io_range range;
5506 bus = kvm_get_bus(kvm, bus_idx);
5510 /* exclude ioeventfd which is limited by maximum fd */
5511 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
5514 new_bus = kmalloc(struct_size(bus, range, bus->dev_count + 1),
5515 GFP_KERNEL_ACCOUNT);
5519 range = (struct kvm_io_range) {
5525 for (i = 0; i < bus->dev_count; i++)
5526 if (kvm_io_bus_cmp(&bus->range[i], &range) > 0)
5529 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
5530 new_bus->dev_count++;
5531 new_bus->range[i] = range;
5532 memcpy(new_bus->range + i + 1, bus->range + i,
5533 (bus->dev_count - i) * sizeof(struct kvm_io_range));
5534 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
5535 synchronize_srcu_expedited(&kvm->srcu);
5541 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
5542 struct kvm_io_device *dev)
5545 struct kvm_io_bus *new_bus, *bus;
5547 lockdep_assert_held(&kvm->slots_lock);
5549 bus = kvm_get_bus(kvm, bus_idx);
5553 for (i = 0; i < bus->dev_count; i++) {
5554 if (bus->range[i].dev == dev) {
5559 if (i == bus->dev_count)
5562 new_bus = kmalloc(struct_size(bus, range, bus->dev_count - 1),
5563 GFP_KERNEL_ACCOUNT);
5565 memcpy(new_bus, bus, struct_size(bus, range, i));
5566 new_bus->dev_count--;
5567 memcpy(new_bus->range + i, bus->range + i + 1,
5568 flex_array_size(new_bus, range, new_bus->dev_count - i));
5571 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
5572 synchronize_srcu_expedited(&kvm->srcu);
5575 * If NULL bus is installed, destroy the old bus, including all the
5576 * attached devices. Otherwise, destroy the caller's device only.
5579 pr_err("kvm: failed to shrink bus, removing it completely\n");
5580 kvm_io_bus_destroy(bus);
5584 kvm_iodevice_destructor(dev);
5589 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
5592 struct kvm_io_bus *bus;
5593 int dev_idx, srcu_idx;
5594 struct kvm_io_device *iodev = NULL;
5596 srcu_idx = srcu_read_lock(&kvm->srcu);
5598 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
5602 dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
5606 iodev = bus->range[dev_idx].dev;
5609 srcu_read_unlock(&kvm->srcu, srcu_idx);
5613 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
5615 static int kvm_debugfs_open(struct inode *inode, struct file *file,
5616 int (*get)(void *, u64 *), int (*set)(void *, u64),
5620 struct kvm_stat_data *stat_data = inode->i_private;
5623 * The debugfs files are a reference to the kvm struct which
5624 * is still valid when kvm_destroy_vm is called. kvm_get_kvm_safe
5625 * avoids the race between open and the removal of the debugfs directory.
5627 if (!kvm_get_kvm_safe(stat_data->kvm))
5630 ret = simple_attr_open(inode, file, get,
5631 kvm_stats_debugfs_mode(stat_data->desc) & 0222
5634 kvm_put_kvm(stat_data->kvm);
5639 static int kvm_debugfs_release(struct inode *inode, struct file *file)
5641 struct kvm_stat_data *stat_data = inode->i_private;
5643 simple_attr_release(inode, file);
5644 kvm_put_kvm(stat_data->kvm);
5649 static int kvm_get_stat_per_vm(struct kvm *kvm, size_t offset, u64 *val)
5651 *val = *(u64 *)((void *)(&kvm->stat) + offset);
5656 static int kvm_clear_stat_per_vm(struct kvm *kvm, size_t offset)
5658 *(u64 *)((void *)(&kvm->stat) + offset) = 0;
5663 static int kvm_get_stat_per_vcpu(struct kvm *kvm, size_t offset, u64 *val)
5666 struct kvm_vcpu *vcpu;
5670 kvm_for_each_vcpu(i, vcpu, kvm)
5671 *val += *(u64 *)((void *)(&vcpu->stat) + offset);
5676 static int kvm_clear_stat_per_vcpu(struct kvm *kvm, size_t offset)
5679 struct kvm_vcpu *vcpu;
5681 kvm_for_each_vcpu(i, vcpu, kvm)
5682 *(u64 *)((void *)(&vcpu->stat) + offset) = 0;
5687 static int kvm_stat_data_get(void *data, u64 *val)
5690 struct kvm_stat_data *stat_data = data;
5692 switch (stat_data->kind) {
5694 r = kvm_get_stat_per_vm(stat_data->kvm,
5695 stat_data->desc->desc.offset, val);
5698 r = kvm_get_stat_per_vcpu(stat_data->kvm,
5699 stat_data->desc->desc.offset, val);
5706 static int kvm_stat_data_clear(void *data, u64 val)
5709 struct kvm_stat_data *stat_data = data;
5714 switch (stat_data->kind) {
5716 r = kvm_clear_stat_per_vm(stat_data->kvm,
5717 stat_data->desc->desc.offset);
5720 r = kvm_clear_stat_per_vcpu(stat_data->kvm,
5721 stat_data->desc->desc.offset);
5728 static int kvm_stat_data_open(struct inode *inode, struct file *file)
5730 __simple_attr_check_format("%llu\n", 0ull);
5731 return kvm_debugfs_open(inode, file, kvm_stat_data_get,
5732 kvm_stat_data_clear, "%llu\n");
5735 static const struct file_operations stat_fops_per_vm = {
5736 .owner = THIS_MODULE,
5737 .open = kvm_stat_data_open,
5738 .release = kvm_debugfs_release,
5739 .read = simple_attr_read,
5740 .write = simple_attr_write,
5741 .llseek = no_llseek,
5744 static int vm_stat_get(void *_offset, u64 *val)
5746 unsigned offset = (long)_offset;
5751 mutex_lock(&kvm_lock);
5752 list_for_each_entry(kvm, &vm_list, vm_list) {
5753 kvm_get_stat_per_vm(kvm, offset, &tmp_val);
5756 mutex_unlock(&kvm_lock);
5760 static int vm_stat_clear(void *_offset, u64 val)
5762 unsigned offset = (long)_offset;
5768 mutex_lock(&kvm_lock);
5769 list_for_each_entry(kvm, &vm_list, vm_list) {
5770 kvm_clear_stat_per_vm(kvm, offset);
5772 mutex_unlock(&kvm_lock);
5777 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n");
5778 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_readonly_fops, vm_stat_get, NULL, "%llu\n");
5780 static int vcpu_stat_get(void *_offset, u64 *val)
5782 unsigned offset = (long)_offset;
5787 mutex_lock(&kvm_lock);
5788 list_for_each_entry(kvm, &vm_list, vm_list) {
5789 kvm_get_stat_per_vcpu(kvm, offset, &tmp_val);
5792 mutex_unlock(&kvm_lock);
5796 static int vcpu_stat_clear(void *_offset, u64 val)
5798 unsigned offset = (long)_offset;
5804 mutex_lock(&kvm_lock);
5805 list_for_each_entry(kvm, &vm_list, vm_list) {
5806 kvm_clear_stat_per_vcpu(kvm, offset);
5808 mutex_unlock(&kvm_lock);
5813 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear,
5815 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_readonly_fops, vcpu_stat_get, NULL, "%llu\n");
5817 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm)
5819 struct kobj_uevent_env *env;
5820 unsigned long long created, active;
5822 if (!kvm_dev.this_device || !kvm)
5825 mutex_lock(&kvm_lock);
5826 if (type == KVM_EVENT_CREATE_VM) {
5827 kvm_createvm_count++;
5829 } else if (type == KVM_EVENT_DESTROY_VM) {
5832 created = kvm_createvm_count;
5833 active = kvm_active_vms;
5834 mutex_unlock(&kvm_lock);
5836 env = kzalloc(sizeof(*env), GFP_KERNEL_ACCOUNT);
5840 add_uevent_var(env, "CREATED=%llu", created);
5841 add_uevent_var(env, "COUNT=%llu", active);
5843 if (type == KVM_EVENT_CREATE_VM) {
5844 add_uevent_var(env, "EVENT=create");
5845 kvm->userspace_pid = task_pid_nr(current);
5846 } else if (type == KVM_EVENT_DESTROY_VM) {
5847 add_uevent_var(env, "EVENT=destroy");
5849 add_uevent_var(env, "PID=%d", kvm->userspace_pid);
5851 if (!IS_ERR(kvm->debugfs_dentry)) {
5852 char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL_ACCOUNT);
5855 tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX);
5857 add_uevent_var(env, "STATS_PATH=%s", tmp);
5861 /* no need for checks, since we are adding at most only 5 keys */
5862 env->envp[env->envp_idx++] = NULL;
5863 kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp);
5867 static void kvm_init_debug(void)
5869 const struct file_operations *fops;
5870 const struct _kvm_stats_desc *pdesc;
5873 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
5875 for (i = 0; i < kvm_vm_stats_header.num_desc; ++i) {
5876 pdesc = &kvm_vm_stats_desc[i];
5877 if (kvm_stats_debugfs_mode(pdesc) & 0222)
5878 fops = &vm_stat_fops;
5880 fops = &vm_stat_readonly_fops;
5881 debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
5883 (void *)(long)pdesc->desc.offset, fops);
5886 for (i = 0; i < kvm_vcpu_stats_header.num_desc; ++i) {
5887 pdesc = &kvm_vcpu_stats_desc[i];
5888 if (kvm_stats_debugfs_mode(pdesc) & 0222)
5889 fops = &vcpu_stat_fops;
5891 fops = &vcpu_stat_readonly_fops;
5892 debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
5894 (void *)(long)pdesc->desc.offset, fops);
5899 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
5901 return container_of(pn, struct kvm_vcpu, preempt_notifier);
5904 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
5906 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
5908 WRITE_ONCE(vcpu->preempted, false);
5909 WRITE_ONCE(vcpu->ready, false);
5911 __this_cpu_write(kvm_running_vcpu, vcpu);
5912 kvm_arch_sched_in(vcpu, cpu);
5913 kvm_arch_vcpu_load(vcpu, cpu);
5916 static void kvm_sched_out(struct preempt_notifier *pn,
5917 struct task_struct *next)
5919 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
5921 if (current->on_rq) {
5922 WRITE_ONCE(vcpu->preempted, true);
5923 WRITE_ONCE(vcpu->ready, true);
5925 kvm_arch_vcpu_put(vcpu);
5926 __this_cpu_write(kvm_running_vcpu, NULL);
5930 * kvm_get_running_vcpu - get the vcpu running on the current CPU.
5932 * We can disable preemption locally around accessing the per-CPU variable,
5933 * and use the resolved vcpu pointer after enabling preemption again,
5934 * because even if the current thread is migrated to another CPU, reading
5935 * the per-CPU value later will give us the same value as we update the
5936 * per-CPU variable in the preempt notifier handlers.
5938 struct kvm_vcpu *kvm_get_running_vcpu(void)
5940 struct kvm_vcpu *vcpu;
5943 vcpu = __this_cpu_read(kvm_running_vcpu);
5948 EXPORT_SYMBOL_GPL(kvm_get_running_vcpu);
5951 * kvm_get_running_vcpus - get the per-CPU array of currently running vcpus.
5953 struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void)
5955 return &kvm_running_vcpu;
5958 #ifdef CONFIG_GUEST_PERF_EVENTS
5959 static unsigned int kvm_guest_state(void)
5961 struct kvm_vcpu *vcpu = kvm_get_running_vcpu();
5964 if (!kvm_arch_pmi_in_guest(vcpu))
5967 state = PERF_GUEST_ACTIVE;
5968 if (!kvm_arch_vcpu_in_kernel(vcpu))
5969 state |= PERF_GUEST_USER;
5974 static unsigned long kvm_guest_get_ip(void)
5976 struct kvm_vcpu *vcpu = kvm_get_running_vcpu();
5978 /* Retrieving the IP must be guarded by a call to kvm_guest_state(). */
5979 if (WARN_ON_ONCE(!kvm_arch_pmi_in_guest(vcpu)))
5982 return kvm_arch_vcpu_get_ip(vcpu);
5985 static struct perf_guest_info_callbacks kvm_guest_cbs = {
5986 .state = kvm_guest_state,
5987 .get_ip = kvm_guest_get_ip,
5988 .handle_intel_pt_intr = NULL,
5991 void kvm_register_perf_callbacks(unsigned int (*pt_intr_handler)(void))
5993 kvm_guest_cbs.handle_intel_pt_intr = pt_intr_handler;
5994 perf_register_guest_info_callbacks(&kvm_guest_cbs);
5996 void kvm_unregister_perf_callbacks(void)
5998 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
6002 int kvm_init(unsigned vcpu_size, unsigned vcpu_align, struct module *module)
6007 #ifdef CONFIG_KVM_GENERIC_HARDWARE_ENABLING
6008 r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_ONLINE, "kvm/cpu:online",
6009 kvm_online_cpu, kvm_offline_cpu);
6013 register_syscore_ops(&kvm_syscore_ops);
6016 /* A kmem cache lets us meet the alignment requirements of fx_save. */
6018 vcpu_align = __alignof__(struct kvm_vcpu);
6020 kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align,
6022 offsetof(struct kvm_vcpu, arch),
6023 offsetofend(struct kvm_vcpu, stats_id)
6024 - offsetof(struct kvm_vcpu, arch),
6026 if (!kvm_vcpu_cache) {
6028 goto err_vcpu_cache;
6031 for_each_possible_cpu(cpu) {
6032 if (!alloc_cpumask_var_node(&per_cpu(cpu_kick_mask, cpu),
6033 GFP_KERNEL, cpu_to_node(cpu))) {
6035 goto err_cpu_kick_mask;
6039 r = kvm_irqfd_init();
6043 r = kvm_async_pf_init();
6047 kvm_chardev_ops.owner = module;
6049 kvm_preempt_ops.sched_in = kvm_sched_in;
6050 kvm_preempt_ops.sched_out = kvm_sched_out;
6054 r = kvm_vfio_ops_init();
6055 if (WARN_ON_ONCE(r))
6059 * Registration _must_ be the very last thing done, as this exposes
6060 * /dev/kvm to userspace, i.e. all infrastructure must be setup!
6062 r = misc_register(&kvm_dev);
6064 pr_err("kvm: misc device register failed\n");
6071 kvm_vfio_ops_exit();
6073 kvm_async_pf_deinit();
6078 for_each_possible_cpu(cpu)
6079 free_cpumask_var(per_cpu(cpu_kick_mask, cpu));
6080 kmem_cache_destroy(kvm_vcpu_cache);
6082 #ifdef CONFIG_KVM_GENERIC_HARDWARE_ENABLING
6083 unregister_syscore_ops(&kvm_syscore_ops);
6084 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_ONLINE);
6088 EXPORT_SYMBOL_GPL(kvm_init);
6095 * Note, unregistering /dev/kvm doesn't strictly need to come first,
6096 * fops_get(), a.k.a. try_module_get(), prevents acquiring references
6097 * to KVM while the module is being stopped.
6099 misc_deregister(&kvm_dev);
6101 debugfs_remove_recursive(kvm_debugfs_dir);
6102 for_each_possible_cpu(cpu)
6103 free_cpumask_var(per_cpu(cpu_kick_mask, cpu));
6104 kmem_cache_destroy(kvm_vcpu_cache);
6105 kvm_vfio_ops_exit();
6106 kvm_async_pf_deinit();
6107 #ifdef CONFIG_KVM_GENERIC_HARDWARE_ENABLING
6108 unregister_syscore_ops(&kvm_syscore_ops);
6109 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_ONLINE);
6113 EXPORT_SYMBOL_GPL(kvm_exit);
6115 struct kvm_vm_worker_thread_context {
6117 struct task_struct *parent;
6118 struct completion init_done;
6119 kvm_vm_thread_fn_t thread_fn;
6124 static int kvm_vm_worker_thread(void *context)
6127 * The init_context is allocated on the stack of the parent thread, so
6128 * we have to locally copy anything that is needed beyond initialization
6130 struct kvm_vm_worker_thread_context *init_context = context;
6131 struct task_struct *parent;
6132 struct kvm *kvm = init_context->kvm;
6133 kvm_vm_thread_fn_t thread_fn = init_context->thread_fn;
6134 uintptr_t data = init_context->data;
6137 err = kthread_park(current);
6138 /* kthread_park(current) is never supposed to return an error */
6143 err = cgroup_attach_task_all(init_context->parent, current);
6145 kvm_err("%s: cgroup_attach_task_all failed with err %d\n",
6150 set_user_nice(current, task_nice(init_context->parent));
6153 init_context->err = err;
6154 complete(&init_context->init_done);
6155 init_context = NULL;
6160 /* Wait to be woken up by the spawner before proceeding. */
6163 if (!kthread_should_stop())
6164 err = thread_fn(kvm, data);
6168 * Move kthread back to its original cgroup to prevent it lingering in
6169 * the cgroup of the VM process, after the latter finishes its
6172 * kthread_stop() waits on the 'exited' completion condition which is
6173 * set in exit_mm(), via mm_release(), in do_exit(). However, the
6174 * kthread is removed from the cgroup in the cgroup_exit() which is
6175 * called after the exit_mm(). This causes the kthread_stop() to return
6176 * before the kthread actually quits the cgroup.
6179 parent = rcu_dereference(current->real_parent);
6180 get_task_struct(parent);
6182 cgroup_attach_task_all(parent, current);
6183 put_task_struct(parent);
6188 int kvm_vm_create_worker_thread(struct kvm *kvm, kvm_vm_thread_fn_t thread_fn,
6189 uintptr_t data, const char *name,
6190 struct task_struct **thread_ptr)
6192 struct kvm_vm_worker_thread_context init_context = {};
6193 struct task_struct *thread;
6196 init_context.kvm = kvm;
6197 init_context.parent = current;
6198 init_context.thread_fn = thread_fn;
6199 init_context.data = data;
6200 init_completion(&init_context.init_done);
6202 thread = kthread_run(kvm_vm_worker_thread, &init_context,
6203 "%s-%d", name, task_pid_nr(current));
6205 return PTR_ERR(thread);
6207 /* kthread_run is never supposed to return NULL */
6208 WARN_ON(thread == NULL);
6210 wait_for_completion(&init_context.init_done);
6212 if (!init_context.err)
6213 *thread_ptr = thread;
6215 return init_context.err;