1 // SPDX-License-Identifier: GPL-2.0-only
3 * Kernel-based Virtual Machine driver for Linux
5 * This module enables machines with Intel VT-x extensions to run virtual
6 * machines without emulation or binary translation.
8 * Copyright (C) 2006 Qumranet, Inc.
9 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
12 * Avi Kivity <avi@qumranet.com>
13 * Yaniv Kamay <yaniv@qumranet.com>
16 #include <kvm/iodev.h>
18 #include <linux/kvm_host.h>
19 #include <linux/kvm.h>
20 #include <linux/module.h>
21 #include <linux/errno.h>
22 #include <linux/percpu.h>
24 #include <linux/miscdevice.h>
25 #include <linux/vmalloc.h>
26 #include <linux/reboot.h>
27 #include <linux/debugfs.h>
28 #include <linux/highmem.h>
29 #include <linux/file.h>
30 #include <linux/syscore_ops.h>
31 #include <linux/cpu.h>
32 #include <linux/sched/signal.h>
33 #include <linux/sched/mm.h>
34 #include <linux/sched/stat.h>
35 #include <linux/cpumask.h>
36 #include <linux/smp.h>
37 #include <linux/anon_inodes.h>
38 #include <linux/profile.h>
39 #include <linux/kvm_para.h>
40 #include <linux/pagemap.h>
41 #include <linux/mman.h>
42 #include <linux/swap.h>
43 #include <linux/bitops.h>
44 #include <linux/spinlock.h>
45 #include <linux/compat.h>
46 #include <linux/srcu.h>
47 #include <linux/hugetlb.h>
48 #include <linux/slab.h>
49 #include <linux/sort.h>
50 #include <linux/bsearch.h>
52 #include <linux/lockdep.h>
53 #include <linux/kthread.h>
54 #include <linux/suspend.h>
56 #include <asm/processor.h>
57 #include <asm/ioctl.h>
58 #include <linux/uaccess.h>
60 #include "coalesced_mmio.h"
65 #define CREATE_TRACE_POINTS
66 #include <trace/events/kvm.h>
68 #include <linux/kvm_dirty_ring.h>
70 /* Worst case buffer size needed for holding an integer. */
71 #define ITOA_MAX_LEN 12
73 MODULE_AUTHOR("Qumranet");
74 MODULE_LICENSE("GPL");
76 /* Architectures should define their poll value according to the halt latency */
77 unsigned int halt_poll_ns = KVM_HALT_POLL_NS_DEFAULT;
78 module_param(halt_poll_ns, uint, 0644);
79 EXPORT_SYMBOL_GPL(halt_poll_ns);
81 /* Default doubles per-vcpu halt_poll_ns. */
82 unsigned int halt_poll_ns_grow = 2;
83 module_param(halt_poll_ns_grow, uint, 0644);
84 EXPORT_SYMBOL_GPL(halt_poll_ns_grow);
86 /* The start value to grow halt_poll_ns from */
87 unsigned int halt_poll_ns_grow_start = 10000; /* 10us */
88 module_param(halt_poll_ns_grow_start, uint, 0644);
89 EXPORT_SYMBOL_GPL(halt_poll_ns_grow_start);
91 /* Default resets per-vcpu halt_poll_ns . */
92 unsigned int halt_poll_ns_shrink;
93 module_param(halt_poll_ns_shrink, uint, 0644);
94 EXPORT_SYMBOL_GPL(halt_poll_ns_shrink);
99 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
102 DEFINE_MUTEX(kvm_lock);
103 static DEFINE_RAW_SPINLOCK(kvm_count_lock);
106 static cpumask_var_t cpus_hardware_enabled;
107 static int kvm_usage_count;
108 static atomic_t hardware_enable_failed;
110 static struct kmem_cache *kvm_vcpu_cache;
112 static __read_mostly struct preempt_ops kvm_preempt_ops;
113 static DEFINE_PER_CPU(struct kvm_vcpu *, kvm_running_vcpu);
115 struct dentry *kvm_debugfs_dir;
116 EXPORT_SYMBOL_GPL(kvm_debugfs_dir);
118 static const struct file_operations stat_fops_per_vm;
120 static struct file_operations kvm_chardev_ops;
122 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
124 #ifdef CONFIG_KVM_COMPAT
125 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
127 #define KVM_COMPAT(c) .compat_ioctl = (c)
130 * For architectures that don't implement a compat infrastructure,
131 * adopt a double line of defense:
132 * - Prevent a compat task from opening /dev/kvm
133 * - If the open has been done by a 64bit task, and the KVM fd
134 * passed to a compat task, let the ioctls fail.
136 static long kvm_no_compat_ioctl(struct file *file, unsigned int ioctl,
137 unsigned long arg) { return -EINVAL; }
139 static int kvm_no_compat_open(struct inode *inode, struct file *file)
141 return is_compat_task() ? -ENODEV : 0;
143 #define KVM_COMPAT(c) .compat_ioctl = kvm_no_compat_ioctl, \
144 .open = kvm_no_compat_open
146 static int hardware_enable_all(void);
147 static void hardware_disable_all(void);
149 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
151 __visible bool kvm_rebooting;
152 EXPORT_SYMBOL_GPL(kvm_rebooting);
154 #define KVM_EVENT_CREATE_VM 0
155 #define KVM_EVENT_DESTROY_VM 1
156 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm);
157 static unsigned long long kvm_createvm_count;
158 static unsigned long long kvm_active_vms;
160 static DEFINE_PER_CPU(cpumask_var_t, cpu_kick_mask);
162 __weak void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
163 unsigned long start, unsigned long end)
167 bool kvm_is_zone_device_pfn(kvm_pfn_t pfn)
170 * The metadata used by is_zone_device_page() to determine whether or
171 * not a page is ZONE_DEVICE is guaranteed to be valid if and only if
172 * the device has been pinned, e.g. by get_user_pages(). WARN if the
173 * page_count() is zero to help detect bad usage of this helper.
175 if (!pfn_valid(pfn) || WARN_ON_ONCE(!page_count(pfn_to_page(pfn))))
178 return is_zone_device_page(pfn_to_page(pfn));
181 bool kvm_is_reserved_pfn(kvm_pfn_t pfn)
184 * ZONE_DEVICE pages currently set PG_reserved, but from a refcounting
185 * perspective they are "normal" pages, albeit with slightly different
189 return PageReserved(pfn_to_page(pfn)) &&
191 !kvm_is_zone_device_pfn(pfn);
197 * Switches to specified vcpu, until a matching vcpu_put()
199 void vcpu_load(struct kvm_vcpu *vcpu)
203 __this_cpu_write(kvm_running_vcpu, vcpu);
204 preempt_notifier_register(&vcpu->preempt_notifier);
205 kvm_arch_vcpu_load(vcpu, cpu);
208 EXPORT_SYMBOL_GPL(vcpu_load);
210 void vcpu_put(struct kvm_vcpu *vcpu)
213 kvm_arch_vcpu_put(vcpu);
214 preempt_notifier_unregister(&vcpu->preempt_notifier);
215 __this_cpu_write(kvm_running_vcpu, NULL);
218 EXPORT_SYMBOL_GPL(vcpu_put);
220 /* TODO: merge with kvm_arch_vcpu_should_kick */
221 static bool kvm_request_needs_ipi(struct kvm_vcpu *vcpu, unsigned req)
223 int mode = kvm_vcpu_exiting_guest_mode(vcpu);
226 * We need to wait for the VCPU to reenable interrupts and get out of
227 * READING_SHADOW_PAGE_TABLES mode.
229 if (req & KVM_REQUEST_WAIT)
230 return mode != OUTSIDE_GUEST_MODE;
233 * Need to kick a running VCPU, but otherwise there is nothing to do.
235 return mode == IN_GUEST_MODE;
238 static void ack_flush(void *_completed)
242 static inline bool kvm_kick_many_cpus(struct cpumask *cpus, bool wait)
244 if (cpumask_empty(cpus))
247 smp_call_function_many(cpus, ack_flush, NULL, wait);
251 static void kvm_make_vcpu_request(struct kvm_vcpu *vcpu, unsigned int req,
252 struct cpumask *tmp, int current_cpu)
256 kvm_make_request(req, vcpu);
258 if (!(req & KVM_REQUEST_NO_WAKEUP) && kvm_vcpu_wake_up(vcpu))
262 * Note, the vCPU could get migrated to a different pCPU at any point
263 * after kvm_request_needs_ipi(), which could result in sending an IPI
264 * to the previous pCPU. But, that's OK because the purpose of the IPI
265 * is to ensure the vCPU returns to OUTSIDE_GUEST_MODE, which is
266 * satisfied if the vCPU migrates. Entering READING_SHADOW_PAGE_TABLES
267 * after this point is also OK, as the requirement is only that KVM wait
268 * for vCPUs that were reading SPTEs _before_ any changes were
269 * finalized. See kvm_vcpu_kick() for more details on handling requests.
271 if (kvm_request_needs_ipi(vcpu, req)) {
272 cpu = READ_ONCE(vcpu->cpu);
273 if (cpu != -1 && cpu != current_cpu)
274 __cpumask_set_cpu(cpu, tmp);
278 bool kvm_make_vcpus_request_mask(struct kvm *kvm, unsigned int req,
279 unsigned long *vcpu_bitmap)
281 struct kvm_vcpu *vcpu;
282 struct cpumask *cpus;
288 cpus = this_cpu_cpumask_var_ptr(cpu_kick_mask);
291 for_each_set_bit(i, vcpu_bitmap, KVM_MAX_VCPUS) {
292 vcpu = kvm_get_vcpu(kvm, i);
295 kvm_make_vcpu_request(vcpu, req, cpus, me);
298 called = kvm_kick_many_cpus(cpus, !!(req & KVM_REQUEST_WAIT));
304 bool kvm_make_all_cpus_request_except(struct kvm *kvm, unsigned int req,
305 struct kvm_vcpu *except)
307 struct kvm_vcpu *vcpu;
308 struct cpumask *cpus;
315 cpus = this_cpu_cpumask_var_ptr(cpu_kick_mask);
318 kvm_for_each_vcpu(i, vcpu, kvm) {
321 kvm_make_vcpu_request(vcpu, req, cpus, me);
324 called = kvm_kick_many_cpus(cpus, !!(req & KVM_REQUEST_WAIT));
330 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
332 return kvm_make_all_cpus_request_except(kvm, req, NULL);
334 EXPORT_SYMBOL_GPL(kvm_make_all_cpus_request);
336 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
337 void kvm_flush_remote_tlbs(struct kvm *kvm)
339 ++kvm->stat.generic.remote_tlb_flush_requests;
342 * We want to publish modifications to the page tables before reading
343 * mode. Pairs with a memory barrier in arch-specific code.
344 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
345 * and smp_mb in walk_shadow_page_lockless_begin/end.
346 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
348 * There is already an smp_mb__after_atomic() before
349 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
352 if (!kvm_arch_flush_remote_tlb(kvm)
353 || kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
354 ++kvm->stat.generic.remote_tlb_flush;
356 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
359 #ifdef KVM_ARCH_NR_OBJS_PER_MEMORY_CACHE
360 static inline void *mmu_memory_cache_alloc_obj(struct kvm_mmu_memory_cache *mc,
363 gfp_flags |= mc->gfp_zero;
366 return kmem_cache_alloc(mc->kmem_cache, gfp_flags);
368 return (void *)__get_free_page(gfp_flags);
371 int kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache *mc, int min)
375 if (mc->nobjs >= min)
377 while (mc->nobjs < ARRAY_SIZE(mc->objects)) {
378 obj = mmu_memory_cache_alloc_obj(mc, GFP_KERNEL_ACCOUNT);
380 return mc->nobjs >= min ? 0 : -ENOMEM;
381 mc->objects[mc->nobjs++] = obj;
386 int kvm_mmu_memory_cache_nr_free_objects(struct kvm_mmu_memory_cache *mc)
391 void kvm_mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc)
395 kmem_cache_free(mc->kmem_cache, mc->objects[--mc->nobjs]);
397 free_page((unsigned long)mc->objects[--mc->nobjs]);
401 void *kvm_mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc)
405 if (WARN_ON(!mc->nobjs))
406 p = mmu_memory_cache_alloc_obj(mc, GFP_ATOMIC | __GFP_ACCOUNT);
408 p = mc->objects[--mc->nobjs];
414 static void kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
416 mutex_init(&vcpu->mutex);
421 #ifndef __KVM_HAVE_ARCH_WQP
422 rcuwait_init(&vcpu->wait);
424 kvm_async_pf_vcpu_init(vcpu);
426 kvm_vcpu_set_in_spin_loop(vcpu, false);
427 kvm_vcpu_set_dy_eligible(vcpu, false);
428 vcpu->preempted = false;
430 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
431 vcpu->last_used_slot = NULL;
434 static void kvm_vcpu_destroy(struct kvm_vcpu *vcpu)
436 kvm_dirty_ring_free(&vcpu->dirty_ring);
437 kvm_arch_vcpu_destroy(vcpu);
440 * No need for rcu_read_lock as VCPU_RUN is the only place that changes
441 * the vcpu->pid pointer, and at destruction time all file descriptors
444 put_pid(rcu_dereference_protected(vcpu->pid, 1));
446 free_page((unsigned long)vcpu->run);
447 kmem_cache_free(kvm_vcpu_cache, vcpu);
450 void kvm_destroy_vcpus(struct kvm *kvm)
453 struct kvm_vcpu *vcpu;
455 kvm_for_each_vcpu(i, vcpu, kvm) {
456 kvm_vcpu_destroy(vcpu);
457 xa_erase(&kvm->vcpu_array, i);
460 atomic_set(&kvm->online_vcpus, 0);
462 EXPORT_SYMBOL_GPL(kvm_destroy_vcpus);
464 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
465 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
467 return container_of(mn, struct kvm, mmu_notifier);
470 static void kvm_mmu_notifier_invalidate_range(struct mmu_notifier *mn,
471 struct mm_struct *mm,
472 unsigned long start, unsigned long end)
474 struct kvm *kvm = mmu_notifier_to_kvm(mn);
477 idx = srcu_read_lock(&kvm->srcu);
478 kvm_arch_mmu_notifier_invalidate_range(kvm, start, end);
479 srcu_read_unlock(&kvm->srcu, idx);
482 typedef bool (*hva_handler_t)(struct kvm *kvm, struct kvm_gfn_range *range);
484 typedef void (*on_lock_fn_t)(struct kvm *kvm, unsigned long start,
487 struct kvm_hva_range {
491 hva_handler_t handler;
492 on_lock_fn_t on_lock;
498 * Use a dedicated stub instead of NULL to indicate that there is no callback
499 * function/handler. The compiler technically can't guarantee that a real
500 * function will have a non-zero address, and so it will generate code to
501 * check for !NULL, whereas comparing against a stub will be elided at compile
502 * time (unless the compiler is getting long in the tooth, e.g. gcc 4.9).
504 static void kvm_null_fn(void)
508 #define IS_KVM_NULL_FN(fn) ((fn) == (void *)kvm_null_fn)
510 /* Iterate over each memslot intersecting [start, last] (inclusive) range */
511 #define kvm_for_each_memslot_in_hva_range(node, slots, start, last) \
512 for (node = interval_tree_iter_first(&slots->hva_tree, start, last); \
514 node = interval_tree_iter_next(node, start, last)) \
516 static __always_inline int __kvm_handle_hva_range(struct kvm *kvm,
517 const struct kvm_hva_range *range)
519 bool ret = false, locked = false;
520 struct kvm_gfn_range gfn_range;
521 struct kvm_memory_slot *slot;
522 struct kvm_memslots *slots;
525 if (WARN_ON_ONCE(range->end <= range->start))
528 /* A null handler is allowed if and only if on_lock() is provided. */
529 if (WARN_ON_ONCE(IS_KVM_NULL_FN(range->on_lock) &&
530 IS_KVM_NULL_FN(range->handler)))
533 idx = srcu_read_lock(&kvm->srcu);
535 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
536 struct interval_tree_node *node;
538 slots = __kvm_memslots(kvm, i);
539 kvm_for_each_memslot_in_hva_range(node, slots,
540 range->start, range->end - 1) {
541 unsigned long hva_start, hva_end;
543 slot = container_of(node, struct kvm_memory_slot, hva_node[slots->node_idx]);
544 hva_start = max(range->start, slot->userspace_addr);
545 hva_end = min(range->end, slot->userspace_addr +
546 (slot->npages << PAGE_SHIFT));
549 * To optimize for the likely case where the address
550 * range is covered by zero or one memslots, don't
551 * bother making these conditional (to avoid writes on
552 * the second or later invocation of the handler).
554 gfn_range.pte = range->pte;
555 gfn_range.may_block = range->may_block;
558 * {gfn(page) | page intersects with [hva_start, hva_end)} =
559 * {gfn_start, gfn_start+1, ..., gfn_end-1}.
561 gfn_range.start = hva_to_gfn_memslot(hva_start, slot);
562 gfn_range.end = hva_to_gfn_memslot(hva_end + PAGE_SIZE - 1, slot);
563 gfn_range.slot = slot;
568 if (!IS_KVM_NULL_FN(range->on_lock))
569 range->on_lock(kvm, range->start, range->end);
570 if (IS_KVM_NULL_FN(range->handler))
573 ret |= range->handler(kvm, &gfn_range);
577 if (range->flush_on_ret && ret)
578 kvm_flush_remote_tlbs(kvm);
583 srcu_read_unlock(&kvm->srcu, idx);
585 /* The notifiers are averse to booleans. :-( */
589 static __always_inline int kvm_handle_hva_range(struct mmu_notifier *mn,
593 hva_handler_t handler)
595 struct kvm *kvm = mmu_notifier_to_kvm(mn);
596 const struct kvm_hva_range range = {
601 .on_lock = (void *)kvm_null_fn,
602 .flush_on_ret = true,
606 return __kvm_handle_hva_range(kvm, &range);
609 static __always_inline int kvm_handle_hva_range_no_flush(struct mmu_notifier *mn,
612 hva_handler_t handler)
614 struct kvm *kvm = mmu_notifier_to_kvm(mn);
615 const struct kvm_hva_range range = {
620 .on_lock = (void *)kvm_null_fn,
621 .flush_on_ret = false,
625 return __kvm_handle_hva_range(kvm, &range);
627 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
628 struct mm_struct *mm,
629 unsigned long address,
632 struct kvm *kvm = mmu_notifier_to_kvm(mn);
634 trace_kvm_set_spte_hva(address);
637 * .change_pte() must be surrounded by .invalidate_range_{start,end}().
638 * If mmu_notifier_count is zero, then no in-progress invalidations,
639 * including this one, found a relevant memslot at start(); rechecking
640 * memslots here is unnecessary. Note, a false positive (count elevated
641 * by a different invalidation) is sub-optimal but functionally ok.
643 WARN_ON_ONCE(!READ_ONCE(kvm->mn_active_invalidate_count));
644 if (!READ_ONCE(kvm->mmu_notifier_count))
647 kvm_handle_hva_range(mn, address, address + 1, pte, kvm_set_spte_gfn);
650 void kvm_inc_notifier_count(struct kvm *kvm, unsigned long start,
654 * The count increase must become visible at unlock time as no
655 * spte can be established without taking the mmu_lock and
656 * count is also read inside the mmu_lock critical section.
658 kvm->mmu_notifier_count++;
659 if (likely(kvm->mmu_notifier_count == 1)) {
660 kvm->mmu_notifier_range_start = start;
661 kvm->mmu_notifier_range_end = end;
664 * Fully tracking multiple concurrent ranges has dimishing
665 * returns. Keep things simple and just find the minimal range
666 * which includes the current and new ranges. As there won't be
667 * enough information to subtract a range after its invalidate
668 * completes, any ranges invalidated concurrently will
669 * accumulate and persist until all outstanding invalidates
672 kvm->mmu_notifier_range_start =
673 min(kvm->mmu_notifier_range_start, start);
674 kvm->mmu_notifier_range_end =
675 max(kvm->mmu_notifier_range_end, end);
679 static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
680 const struct mmu_notifier_range *range)
682 struct kvm *kvm = mmu_notifier_to_kvm(mn);
683 const struct kvm_hva_range hva_range = {
684 .start = range->start,
687 .handler = kvm_unmap_gfn_range,
688 .on_lock = kvm_inc_notifier_count,
689 .flush_on_ret = true,
690 .may_block = mmu_notifier_range_blockable(range),
693 trace_kvm_unmap_hva_range(range->start, range->end);
696 * Prevent memslot modification between range_start() and range_end()
697 * so that conditionally locking provides the same result in both
698 * functions. Without that guarantee, the mmu_notifier_count
699 * adjustments will be imbalanced.
701 * Pairs with the decrement in range_end().
703 spin_lock(&kvm->mn_invalidate_lock);
704 kvm->mn_active_invalidate_count++;
705 spin_unlock(&kvm->mn_invalidate_lock);
707 gfn_to_pfn_cache_invalidate_start(kvm, range->start, range->end,
708 hva_range.may_block);
710 __kvm_handle_hva_range(kvm, &hva_range);
715 void kvm_dec_notifier_count(struct kvm *kvm, unsigned long start,
719 * This sequence increase will notify the kvm page fault that
720 * the page that is going to be mapped in the spte could have
723 kvm->mmu_notifier_seq++;
726 * The above sequence increase must be visible before the
727 * below count decrease, which is ensured by the smp_wmb above
728 * in conjunction with the smp_rmb in mmu_notifier_retry().
730 kvm->mmu_notifier_count--;
733 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
734 const struct mmu_notifier_range *range)
736 struct kvm *kvm = mmu_notifier_to_kvm(mn);
737 const struct kvm_hva_range hva_range = {
738 .start = range->start,
741 .handler = (void *)kvm_null_fn,
742 .on_lock = kvm_dec_notifier_count,
743 .flush_on_ret = false,
744 .may_block = mmu_notifier_range_blockable(range),
748 __kvm_handle_hva_range(kvm, &hva_range);
750 /* Pairs with the increment in range_start(). */
751 spin_lock(&kvm->mn_invalidate_lock);
752 wake = (--kvm->mn_active_invalidate_count == 0);
753 spin_unlock(&kvm->mn_invalidate_lock);
756 * There can only be one waiter, since the wait happens under
760 rcuwait_wake_up(&kvm->mn_memslots_update_rcuwait);
762 BUG_ON(kvm->mmu_notifier_count < 0);
765 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
766 struct mm_struct *mm,
770 trace_kvm_age_hva(start, end);
772 return kvm_handle_hva_range(mn, start, end, __pte(0), kvm_age_gfn);
775 static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
776 struct mm_struct *mm,
780 trace_kvm_age_hva(start, end);
783 * Even though we do not flush TLB, this will still adversely
784 * affect performance on pre-Haswell Intel EPT, where there is
785 * no EPT Access Bit to clear so that we have to tear down EPT
786 * tables instead. If we find this unacceptable, we can always
787 * add a parameter to kvm_age_hva so that it effectively doesn't
788 * do anything on clear_young.
790 * Also note that currently we never issue secondary TLB flushes
791 * from clear_young, leaving this job up to the regular system
792 * cadence. If we find this inaccurate, we might come up with a
793 * more sophisticated heuristic later.
795 return kvm_handle_hva_range_no_flush(mn, start, end, kvm_age_gfn);
798 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
799 struct mm_struct *mm,
800 unsigned long address)
802 trace_kvm_test_age_hva(address);
804 return kvm_handle_hva_range_no_flush(mn, address, address + 1,
808 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
809 struct mm_struct *mm)
811 struct kvm *kvm = mmu_notifier_to_kvm(mn);
814 idx = srcu_read_lock(&kvm->srcu);
815 kvm_arch_flush_shadow_all(kvm);
816 srcu_read_unlock(&kvm->srcu, idx);
819 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
820 .invalidate_range = kvm_mmu_notifier_invalidate_range,
821 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
822 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
823 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
824 .clear_young = kvm_mmu_notifier_clear_young,
825 .test_young = kvm_mmu_notifier_test_young,
826 .change_pte = kvm_mmu_notifier_change_pte,
827 .release = kvm_mmu_notifier_release,
830 static int kvm_init_mmu_notifier(struct kvm *kvm)
832 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
833 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
836 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
838 static int kvm_init_mmu_notifier(struct kvm *kvm)
843 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
845 #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER
846 static int kvm_pm_notifier_call(struct notifier_block *bl,
850 struct kvm *kvm = container_of(bl, struct kvm, pm_notifier);
852 return kvm_arch_pm_notifier(kvm, state);
855 static void kvm_init_pm_notifier(struct kvm *kvm)
857 kvm->pm_notifier.notifier_call = kvm_pm_notifier_call;
858 /* Suspend KVM before we suspend ftrace, RCU, etc. */
859 kvm->pm_notifier.priority = INT_MAX;
860 register_pm_notifier(&kvm->pm_notifier);
863 static void kvm_destroy_pm_notifier(struct kvm *kvm)
865 unregister_pm_notifier(&kvm->pm_notifier);
867 #else /* !CONFIG_HAVE_KVM_PM_NOTIFIER */
868 static void kvm_init_pm_notifier(struct kvm *kvm)
872 static void kvm_destroy_pm_notifier(struct kvm *kvm)
875 #endif /* CONFIG_HAVE_KVM_PM_NOTIFIER */
877 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
879 if (!memslot->dirty_bitmap)
882 kvfree(memslot->dirty_bitmap);
883 memslot->dirty_bitmap = NULL;
886 /* This does not remove the slot from struct kvm_memslots data structures */
887 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
889 kvm_destroy_dirty_bitmap(slot);
891 kvm_arch_free_memslot(kvm, slot);
896 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
898 struct hlist_node *idnode;
899 struct kvm_memory_slot *memslot;
903 * The same memslot objects live in both active and inactive sets,
904 * arbitrarily free using index '1' so the second invocation of this
905 * function isn't operating over a structure with dangling pointers
906 * (even though this function isn't actually touching them).
908 if (!slots->node_idx)
911 hash_for_each_safe(slots->id_hash, bkt, idnode, memslot, id_node[1])
912 kvm_free_memslot(kvm, memslot);
915 static umode_t kvm_stats_debugfs_mode(const struct _kvm_stats_desc *pdesc)
917 switch (pdesc->desc.flags & KVM_STATS_TYPE_MASK) {
918 case KVM_STATS_TYPE_INSTANT:
920 case KVM_STATS_TYPE_CUMULATIVE:
921 case KVM_STATS_TYPE_PEAK:
928 static void kvm_destroy_vm_debugfs(struct kvm *kvm)
931 int kvm_debugfs_num_entries = kvm_vm_stats_header.num_desc +
932 kvm_vcpu_stats_header.num_desc;
934 if (!kvm->debugfs_dentry)
937 debugfs_remove_recursive(kvm->debugfs_dentry);
939 if (kvm->debugfs_stat_data) {
940 for (i = 0; i < kvm_debugfs_num_entries; i++)
941 kfree(kvm->debugfs_stat_data[i]);
942 kfree(kvm->debugfs_stat_data);
946 static int kvm_create_vm_debugfs(struct kvm *kvm, int fd)
948 static DEFINE_MUTEX(kvm_debugfs_lock);
950 char dir_name[ITOA_MAX_LEN * 2];
951 struct kvm_stat_data *stat_data;
952 const struct _kvm_stats_desc *pdesc;
954 int kvm_debugfs_num_entries = kvm_vm_stats_header.num_desc +
955 kvm_vcpu_stats_header.num_desc;
957 if (!debugfs_initialized())
960 snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd);
961 mutex_lock(&kvm_debugfs_lock);
962 dent = debugfs_lookup(dir_name, kvm_debugfs_dir);
964 pr_warn_ratelimited("KVM: debugfs: duplicate directory %s\n", dir_name);
966 mutex_unlock(&kvm_debugfs_lock);
969 dent = debugfs_create_dir(dir_name, kvm_debugfs_dir);
970 mutex_unlock(&kvm_debugfs_lock);
974 kvm->debugfs_dentry = dent;
975 kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries,
976 sizeof(*kvm->debugfs_stat_data),
978 if (!kvm->debugfs_stat_data)
981 for (i = 0; i < kvm_vm_stats_header.num_desc; ++i) {
982 pdesc = &kvm_vm_stats_desc[i];
983 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT);
987 stat_data->kvm = kvm;
988 stat_data->desc = pdesc;
989 stat_data->kind = KVM_STAT_VM;
990 kvm->debugfs_stat_data[i] = stat_data;
991 debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
992 kvm->debugfs_dentry, stat_data,
996 for (i = 0; i < kvm_vcpu_stats_header.num_desc; ++i) {
997 pdesc = &kvm_vcpu_stats_desc[i];
998 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT);
1002 stat_data->kvm = kvm;
1003 stat_data->desc = pdesc;
1004 stat_data->kind = KVM_STAT_VCPU;
1005 kvm->debugfs_stat_data[i + kvm_vm_stats_header.num_desc] = stat_data;
1006 debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
1007 kvm->debugfs_dentry, stat_data,
1011 ret = kvm_arch_create_vm_debugfs(kvm);
1013 kvm_destroy_vm_debugfs(kvm);
1021 * Called after the VM is otherwise initialized, but just before adding it to
1024 int __weak kvm_arch_post_init_vm(struct kvm *kvm)
1030 * Called just after removing the VM from the vm_list, but before doing any
1031 * other destruction.
1033 void __weak kvm_arch_pre_destroy_vm(struct kvm *kvm)
1038 * Called after per-vm debugfs created. When called kvm->debugfs_dentry should
1039 * be setup already, so we can create arch-specific debugfs entries under it.
1040 * Cleanup should be automatic done in kvm_destroy_vm_debugfs() recursively, so
1041 * a per-arch destroy interface is not needed.
1043 int __weak kvm_arch_create_vm_debugfs(struct kvm *kvm)
1048 static struct kvm *kvm_create_vm(unsigned long type)
1050 struct kvm *kvm = kvm_arch_alloc_vm();
1051 struct kvm_memslots *slots;
1056 return ERR_PTR(-ENOMEM);
1058 KVM_MMU_LOCK_INIT(kvm);
1059 mmgrab(current->mm);
1060 kvm->mm = current->mm;
1061 kvm_eventfd_init(kvm);
1062 mutex_init(&kvm->lock);
1063 mutex_init(&kvm->irq_lock);
1064 mutex_init(&kvm->slots_lock);
1065 mutex_init(&kvm->slots_arch_lock);
1066 spin_lock_init(&kvm->mn_invalidate_lock);
1067 rcuwait_init(&kvm->mn_memslots_update_rcuwait);
1068 xa_init(&kvm->vcpu_array);
1070 INIT_LIST_HEAD(&kvm->gpc_list);
1071 spin_lock_init(&kvm->gpc_lock);
1073 INIT_LIST_HEAD(&kvm->devices);
1075 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
1077 if (init_srcu_struct(&kvm->srcu))
1078 goto out_err_no_srcu;
1079 if (init_srcu_struct(&kvm->irq_srcu))
1080 goto out_err_no_irq_srcu;
1082 refcount_set(&kvm->users_count, 1);
1083 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
1084 for (j = 0; j < 2; j++) {
1085 slots = &kvm->__memslots[i][j];
1087 atomic_long_set(&slots->last_used_slot, (unsigned long)NULL);
1088 slots->hva_tree = RB_ROOT_CACHED;
1089 slots->gfn_tree = RB_ROOT;
1090 hash_init(slots->id_hash);
1091 slots->node_idx = j;
1093 /* Generations must be different for each address space. */
1094 slots->generation = i;
1097 rcu_assign_pointer(kvm->memslots[i], &kvm->__memslots[i][0]);
1100 for (i = 0; i < KVM_NR_BUSES; i++) {
1101 rcu_assign_pointer(kvm->buses[i],
1102 kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL_ACCOUNT));
1104 goto out_err_no_arch_destroy_vm;
1107 kvm->max_halt_poll_ns = halt_poll_ns;
1109 r = kvm_arch_init_vm(kvm, type);
1111 goto out_err_no_arch_destroy_vm;
1113 r = hardware_enable_all();
1115 goto out_err_no_disable;
1117 #ifdef CONFIG_HAVE_KVM_IRQFD
1118 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
1121 r = kvm_init_mmu_notifier(kvm);
1123 goto out_err_no_mmu_notifier;
1125 r = kvm_arch_post_init_vm(kvm);
1129 mutex_lock(&kvm_lock);
1130 list_add(&kvm->vm_list, &vm_list);
1131 mutex_unlock(&kvm_lock);
1133 preempt_notifier_inc();
1134 kvm_init_pm_notifier(kvm);
1137 * When the fd passed to this ioctl() is opened it pins the module,
1138 * but try_module_get() also prevents getting a reference if the module
1139 * is in MODULE_STATE_GOING (e.g. if someone ran "rmmod --wait").
1141 if (!try_module_get(kvm_chardev_ops.owner)) {
1149 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1150 if (kvm->mmu_notifier.ops)
1151 mmu_notifier_unregister(&kvm->mmu_notifier, current->mm);
1153 out_err_no_mmu_notifier:
1154 hardware_disable_all();
1156 kvm_arch_destroy_vm(kvm);
1157 out_err_no_arch_destroy_vm:
1158 WARN_ON_ONCE(!refcount_dec_and_test(&kvm->users_count));
1159 for (i = 0; i < KVM_NR_BUSES; i++)
1160 kfree(kvm_get_bus(kvm, i));
1161 cleanup_srcu_struct(&kvm->irq_srcu);
1162 out_err_no_irq_srcu:
1163 cleanup_srcu_struct(&kvm->srcu);
1165 kvm_arch_free_vm(kvm);
1166 mmdrop(current->mm);
1170 static void kvm_destroy_devices(struct kvm *kvm)
1172 struct kvm_device *dev, *tmp;
1175 * We do not need to take the kvm->lock here, because nobody else
1176 * has a reference to the struct kvm at this point and therefore
1177 * cannot access the devices list anyhow.
1179 list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
1180 list_del(&dev->vm_node);
1181 dev->ops->destroy(dev);
1185 static void kvm_destroy_vm(struct kvm *kvm)
1188 struct mm_struct *mm = kvm->mm;
1190 kvm_destroy_pm_notifier(kvm);
1191 kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm);
1192 kvm_destroy_vm_debugfs(kvm);
1193 kvm_arch_sync_events(kvm);
1194 mutex_lock(&kvm_lock);
1195 list_del(&kvm->vm_list);
1196 mutex_unlock(&kvm_lock);
1197 kvm_arch_pre_destroy_vm(kvm);
1199 kvm_free_irq_routing(kvm);
1200 for (i = 0; i < KVM_NR_BUSES; i++) {
1201 struct kvm_io_bus *bus = kvm_get_bus(kvm, i);
1204 kvm_io_bus_destroy(bus);
1205 kvm->buses[i] = NULL;
1207 kvm_coalesced_mmio_free(kvm);
1208 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1209 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
1211 * At this point, pending calls to invalidate_range_start()
1212 * have completed but no more MMU notifiers will run, so
1213 * mn_active_invalidate_count may remain unbalanced.
1214 * No threads can be waiting in install_new_memslots as the
1215 * last reference on KVM has been dropped, but freeing
1216 * memslots would deadlock without this manual intervention.
1218 WARN_ON(rcuwait_active(&kvm->mn_memslots_update_rcuwait));
1219 kvm->mn_active_invalidate_count = 0;
1221 kvm_arch_flush_shadow_all(kvm);
1223 kvm_arch_destroy_vm(kvm);
1224 kvm_destroy_devices(kvm);
1225 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
1226 kvm_free_memslots(kvm, &kvm->__memslots[i][0]);
1227 kvm_free_memslots(kvm, &kvm->__memslots[i][1]);
1229 cleanup_srcu_struct(&kvm->irq_srcu);
1230 cleanup_srcu_struct(&kvm->srcu);
1231 kvm_arch_free_vm(kvm);
1232 preempt_notifier_dec();
1233 hardware_disable_all();
1235 module_put(kvm_chardev_ops.owner);
1238 void kvm_get_kvm(struct kvm *kvm)
1240 refcount_inc(&kvm->users_count);
1242 EXPORT_SYMBOL_GPL(kvm_get_kvm);
1245 * Make sure the vm is not during destruction, which is a safe version of
1246 * kvm_get_kvm(). Return true if kvm referenced successfully, false otherwise.
1248 bool kvm_get_kvm_safe(struct kvm *kvm)
1250 return refcount_inc_not_zero(&kvm->users_count);
1252 EXPORT_SYMBOL_GPL(kvm_get_kvm_safe);
1254 void kvm_put_kvm(struct kvm *kvm)
1256 if (refcount_dec_and_test(&kvm->users_count))
1257 kvm_destroy_vm(kvm);
1259 EXPORT_SYMBOL_GPL(kvm_put_kvm);
1262 * Used to put a reference that was taken on behalf of an object associated
1263 * with a user-visible file descriptor, e.g. a vcpu or device, if installation
1264 * of the new file descriptor fails and the reference cannot be transferred to
1265 * its final owner. In such cases, the caller is still actively using @kvm and
1266 * will fail miserably if the refcount unexpectedly hits zero.
1268 void kvm_put_kvm_no_destroy(struct kvm *kvm)
1270 WARN_ON(refcount_dec_and_test(&kvm->users_count));
1272 EXPORT_SYMBOL_GPL(kvm_put_kvm_no_destroy);
1274 static int kvm_vm_release(struct inode *inode, struct file *filp)
1276 struct kvm *kvm = filp->private_data;
1278 kvm_irqfd_release(kvm);
1285 * Allocation size is twice as large as the actual dirty bitmap size.
1286 * See kvm_vm_ioctl_get_dirty_log() why this is needed.
1288 static int kvm_alloc_dirty_bitmap(struct kvm_memory_slot *memslot)
1290 unsigned long dirty_bytes = kvm_dirty_bitmap_bytes(memslot);
1292 memslot->dirty_bitmap = __vcalloc(2, dirty_bytes, GFP_KERNEL_ACCOUNT);
1293 if (!memslot->dirty_bitmap)
1299 static struct kvm_memslots *kvm_get_inactive_memslots(struct kvm *kvm, int as_id)
1301 struct kvm_memslots *active = __kvm_memslots(kvm, as_id);
1302 int node_idx_inactive = active->node_idx ^ 1;
1304 return &kvm->__memslots[as_id][node_idx_inactive];
1308 * Helper to get the address space ID when one of memslot pointers may be NULL.
1309 * This also serves as a sanity that at least one of the pointers is non-NULL,
1310 * and that their address space IDs don't diverge.
1312 static int kvm_memslots_get_as_id(struct kvm_memory_slot *a,
1313 struct kvm_memory_slot *b)
1315 if (WARN_ON_ONCE(!a && !b))
1323 WARN_ON_ONCE(a->as_id != b->as_id);
1327 static void kvm_insert_gfn_node(struct kvm_memslots *slots,
1328 struct kvm_memory_slot *slot)
1330 struct rb_root *gfn_tree = &slots->gfn_tree;
1331 struct rb_node **node, *parent;
1332 int idx = slots->node_idx;
1335 for (node = &gfn_tree->rb_node; *node; ) {
1336 struct kvm_memory_slot *tmp;
1338 tmp = container_of(*node, struct kvm_memory_slot, gfn_node[idx]);
1340 if (slot->base_gfn < tmp->base_gfn)
1341 node = &(*node)->rb_left;
1342 else if (slot->base_gfn > tmp->base_gfn)
1343 node = &(*node)->rb_right;
1348 rb_link_node(&slot->gfn_node[idx], parent, node);
1349 rb_insert_color(&slot->gfn_node[idx], gfn_tree);
1352 static void kvm_erase_gfn_node(struct kvm_memslots *slots,
1353 struct kvm_memory_slot *slot)
1355 rb_erase(&slot->gfn_node[slots->node_idx], &slots->gfn_tree);
1358 static void kvm_replace_gfn_node(struct kvm_memslots *slots,
1359 struct kvm_memory_slot *old,
1360 struct kvm_memory_slot *new)
1362 int idx = slots->node_idx;
1364 WARN_ON_ONCE(old->base_gfn != new->base_gfn);
1366 rb_replace_node(&old->gfn_node[idx], &new->gfn_node[idx],
1371 * Replace @old with @new in the inactive memslots.
1373 * With NULL @old this simply adds @new.
1374 * With NULL @new this simply removes @old.
1376 * If @new is non-NULL its hva_node[slots_idx] range has to be set
1379 static void kvm_replace_memslot(struct kvm *kvm,
1380 struct kvm_memory_slot *old,
1381 struct kvm_memory_slot *new)
1383 int as_id = kvm_memslots_get_as_id(old, new);
1384 struct kvm_memslots *slots = kvm_get_inactive_memslots(kvm, as_id);
1385 int idx = slots->node_idx;
1388 hash_del(&old->id_node[idx]);
1389 interval_tree_remove(&old->hva_node[idx], &slots->hva_tree);
1391 if ((long)old == atomic_long_read(&slots->last_used_slot))
1392 atomic_long_set(&slots->last_used_slot, (long)new);
1395 kvm_erase_gfn_node(slots, old);
1401 * Initialize @new's hva range. Do this even when replacing an @old
1402 * slot, kvm_copy_memslot() deliberately does not touch node data.
1404 new->hva_node[idx].start = new->userspace_addr;
1405 new->hva_node[idx].last = new->userspace_addr +
1406 (new->npages << PAGE_SHIFT) - 1;
1409 * (Re)Add the new memslot. There is no O(1) interval_tree_replace(),
1410 * hva_node needs to be swapped with remove+insert even though hva can't
1411 * change when replacing an existing slot.
1413 hash_add(slots->id_hash, &new->id_node[idx], new->id);
1414 interval_tree_insert(&new->hva_node[idx], &slots->hva_tree);
1417 * If the memslot gfn is unchanged, rb_replace_node() can be used to
1418 * switch the node in the gfn tree instead of removing the old and
1419 * inserting the new as two separate operations. Replacement is a
1420 * single O(1) operation versus two O(log(n)) operations for
1423 if (old && old->base_gfn == new->base_gfn) {
1424 kvm_replace_gfn_node(slots, old, new);
1427 kvm_erase_gfn_node(slots, old);
1428 kvm_insert_gfn_node(slots, new);
1432 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
1434 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
1436 #ifdef __KVM_HAVE_READONLY_MEM
1437 valid_flags |= KVM_MEM_READONLY;
1440 if (mem->flags & ~valid_flags)
1446 static void kvm_swap_active_memslots(struct kvm *kvm, int as_id)
1448 struct kvm_memslots *slots = kvm_get_inactive_memslots(kvm, as_id);
1450 /* Grab the generation from the activate memslots. */
1451 u64 gen = __kvm_memslots(kvm, as_id)->generation;
1453 WARN_ON(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS);
1454 slots->generation = gen | KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
1457 * Do not store the new memslots while there are invalidations in
1458 * progress, otherwise the locking in invalidate_range_start and
1459 * invalidate_range_end will be unbalanced.
1461 spin_lock(&kvm->mn_invalidate_lock);
1462 prepare_to_rcuwait(&kvm->mn_memslots_update_rcuwait);
1463 while (kvm->mn_active_invalidate_count) {
1464 set_current_state(TASK_UNINTERRUPTIBLE);
1465 spin_unlock(&kvm->mn_invalidate_lock);
1467 spin_lock(&kvm->mn_invalidate_lock);
1469 finish_rcuwait(&kvm->mn_memslots_update_rcuwait);
1470 rcu_assign_pointer(kvm->memslots[as_id], slots);
1471 spin_unlock(&kvm->mn_invalidate_lock);
1474 * Acquired in kvm_set_memslot. Must be released before synchronize
1475 * SRCU below in order to avoid deadlock with another thread
1476 * acquiring the slots_arch_lock in an srcu critical section.
1478 mutex_unlock(&kvm->slots_arch_lock);
1480 synchronize_srcu_expedited(&kvm->srcu);
1483 * Increment the new memslot generation a second time, dropping the
1484 * update in-progress flag and incrementing the generation based on
1485 * the number of address spaces. This provides a unique and easily
1486 * identifiable generation number while the memslots are in flux.
1488 gen = slots->generation & ~KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
1491 * Generations must be unique even across address spaces. We do not need
1492 * a global counter for that, instead the generation space is evenly split
1493 * across address spaces. For example, with two address spaces, address
1494 * space 0 will use generations 0, 2, 4, ... while address space 1 will
1495 * use generations 1, 3, 5, ...
1497 gen += KVM_ADDRESS_SPACE_NUM;
1499 kvm_arch_memslots_updated(kvm, gen);
1501 slots->generation = gen;
1504 static int kvm_prepare_memory_region(struct kvm *kvm,
1505 const struct kvm_memory_slot *old,
1506 struct kvm_memory_slot *new,
1507 enum kvm_mr_change change)
1512 * If dirty logging is disabled, nullify the bitmap; the old bitmap
1513 * will be freed on "commit". If logging is enabled in both old and
1514 * new, reuse the existing bitmap. If logging is enabled only in the
1515 * new and KVM isn't using a ring buffer, allocate and initialize a
1518 if (change != KVM_MR_DELETE) {
1519 if (!(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
1520 new->dirty_bitmap = NULL;
1521 else if (old && old->dirty_bitmap)
1522 new->dirty_bitmap = old->dirty_bitmap;
1523 else if (!kvm->dirty_ring_size) {
1524 r = kvm_alloc_dirty_bitmap(new);
1528 if (kvm_dirty_log_manual_protect_and_init_set(kvm))
1529 bitmap_set(new->dirty_bitmap, 0, new->npages);
1533 r = kvm_arch_prepare_memory_region(kvm, old, new, change);
1535 /* Free the bitmap on failure if it was allocated above. */
1536 if (r && new && new->dirty_bitmap && old && !old->dirty_bitmap)
1537 kvm_destroy_dirty_bitmap(new);
1542 static void kvm_commit_memory_region(struct kvm *kvm,
1543 struct kvm_memory_slot *old,
1544 const struct kvm_memory_slot *new,
1545 enum kvm_mr_change change)
1548 * Update the total number of memslot pages before calling the arch
1549 * hook so that architectures can consume the result directly.
1551 if (change == KVM_MR_DELETE)
1552 kvm->nr_memslot_pages -= old->npages;
1553 else if (change == KVM_MR_CREATE)
1554 kvm->nr_memslot_pages += new->npages;
1556 kvm_arch_commit_memory_region(kvm, old, new, change);
1560 /* Nothing more to do. */
1563 /* Free the old memslot and all its metadata. */
1564 kvm_free_memslot(kvm, old);
1567 case KVM_MR_FLAGS_ONLY:
1569 * Free the dirty bitmap as needed; the below check encompasses
1570 * both the flags and whether a ring buffer is being used)
1572 if (old->dirty_bitmap && !new->dirty_bitmap)
1573 kvm_destroy_dirty_bitmap(old);
1576 * The final quirk. Free the detached, old slot, but only its
1577 * memory, not any metadata. Metadata, including arch specific
1578 * data, may be reused by @new.
1588 * Activate @new, which must be installed in the inactive slots by the caller,
1589 * by swapping the active slots and then propagating @new to @old once @old is
1590 * unreachable and can be safely modified.
1592 * With NULL @old this simply adds @new to @active (while swapping the sets).
1593 * With NULL @new this simply removes @old from @active and frees it
1594 * (while also swapping the sets).
1596 static void kvm_activate_memslot(struct kvm *kvm,
1597 struct kvm_memory_slot *old,
1598 struct kvm_memory_slot *new)
1600 int as_id = kvm_memslots_get_as_id(old, new);
1602 kvm_swap_active_memslots(kvm, as_id);
1604 /* Propagate the new memslot to the now inactive memslots. */
1605 kvm_replace_memslot(kvm, old, new);
1608 static void kvm_copy_memslot(struct kvm_memory_slot *dest,
1609 const struct kvm_memory_slot *src)
1611 dest->base_gfn = src->base_gfn;
1612 dest->npages = src->npages;
1613 dest->dirty_bitmap = src->dirty_bitmap;
1614 dest->arch = src->arch;
1615 dest->userspace_addr = src->userspace_addr;
1616 dest->flags = src->flags;
1618 dest->as_id = src->as_id;
1621 static void kvm_invalidate_memslot(struct kvm *kvm,
1622 struct kvm_memory_slot *old,
1623 struct kvm_memory_slot *invalid_slot)
1626 * Mark the current slot INVALID. As with all memslot modifications,
1627 * this must be done on an unreachable slot to avoid modifying the
1628 * current slot in the active tree.
1630 kvm_copy_memslot(invalid_slot, old);
1631 invalid_slot->flags |= KVM_MEMSLOT_INVALID;
1632 kvm_replace_memslot(kvm, old, invalid_slot);
1635 * Activate the slot that is now marked INVALID, but don't propagate
1636 * the slot to the now inactive slots. The slot is either going to be
1637 * deleted or recreated as a new slot.
1639 kvm_swap_active_memslots(kvm, old->as_id);
1642 * From this point no new shadow pages pointing to a deleted, or moved,
1643 * memslot will be created. Validation of sp->gfn happens in:
1644 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1645 * - kvm_is_visible_gfn (mmu_check_root)
1647 kvm_arch_flush_shadow_memslot(kvm, old);
1649 /* Was released by kvm_swap_active_memslots, reacquire. */
1650 mutex_lock(&kvm->slots_arch_lock);
1653 * Copy the arch-specific field of the newly-installed slot back to the
1654 * old slot as the arch data could have changed between releasing
1655 * slots_arch_lock in install_new_memslots() and re-acquiring the lock
1656 * above. Writers are required to retrieve memslots *after* acquiring
1657 * slots_arch_lock, thus the active slot's data is guaranteed to be fresh.
1659 old->arch = invalid_slot->arch;
1662 static void kvm_create_memslot(struct kvm *kvm,
1663 struct kvm_memory_slot *new)
1665 /* Add the new memslot to the inactive set and activate. */
1666 kvm_replace_memslot(kvm, NULL, new);
1667 kvm_activate_memslot(kvm, NULL, new);
1670 static void kvm_delete_memslot(struct kvm *kvm,
1671 struct kvm_memory_slot *old,
1672 struct kvm_memory_slot *invalid_slot)
1675 * Remove the old memslot (in the inactive memslots) by passing NULL as
1676 * the "new" slot, and for the invalid version in the active slots.
1678 kvm_replace_memslot(kvm, old, NULL);
1679 kvm_activate_memslot(kvm, invalid_slot, NULL);
1682 static void kvm_move_memslot(struct kvm *kvm,
1683 struct kvm_memory_slot *old,
1684 struct kvm_memory_slot *new,
1685 struct kvm_memory_slot *invalid_slot)
1688 * Replace the old memslot in the inactive slots, and then swap slots
1689 * and replace the current INVALID with the new as well.
1691 kvm_replace_memslot(kvm, old, new);
1692 kvm_activate_memslot(kvm, invalid_slot, new);
1695 static void kvm_update_flags_memslot(struct kvm *kvm,
1696 struct kvm_memory_slot *old,
1697 struct kvm_memory_slot *new)
1700 * Similar to the MOVE case, but the slot doesn't need to be zapped as
1701 * an intermediate step. Instead, the old memslot is simply replaced
1702 * with a new, updated copy in both memslot sets.
1704 kvm_replace_memslot(kvm, old, new);
1705 kvm_activate_memslot(kvm, old, new);
1708 static int kvm_set_memslot(struct kvm *kvm,
1709 struct kvm_memory_slot *old,
1710 struct kvm_memory_slot *new,
1711 enum kvm_mr_change change)
1713 struct kvm_memory_slot *invalid_slot;
1717 * Released in kvm_swap_active_memslots.
1719 * Must be held from before the current memslots are copied until
1720 * after the new memslots are installed with rcu_assign_pointer,
1721 * then released before the synchronize srcu in kvm_swap_active_memslots.
1723 * When modifying memslots outside of the slots_lock, must be held
1724 * before reading the pointer to the current memslots until after all
1725 * changes to those memslots are complete.
1727 * These rules ensure that installing new memslots does not lose
1728 * changes made to the previous memslots.
1730 mutex_lock(&kvm->slots_arch_lock);
1733 * Invalidate the old slot if it's being deleted or moved. This is
1734 * done prior to actually deleting/moving the memslot to allow vCPUs to
1735 * continue running by ensuring there are no mappings or shadow pages
1736 * for the memslot when it is deleted/moved. Without pre-invalidation
1737 * (and without a lock), a window would exist between effecting the
1738 * delete/move and committing the changes in arch code where KVM or a
1739 * guest could access a non-existent memslot.
1741 * Modifications are done on a temporary, unreachable slot. The old
1742 * slot needs to be preserved in case a later step fails and the
1743 * invalidation needs to be reverted.
1745 if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) {
1746 invalid_slot = kzalloc(sizeof(*invalid_slot), GFP_KERNEL_ACCOUNT);
1747 if (!invalid_slot) {
1748 mutex_unlock(&kvm->slots_arch_lock);
1751 kvm_invalidate_memslot(kvm, old, invalid_slot);
1754 r = kvm_prepare_memory_region(kvm, old, new, change);
1757 * For DELETE/MOVE, revert the above INVALID change. No
1758 * modifications required since the original slot was preserved
1759 * in the inactive slots. Changing the active memslots also
1760 * release slots_arch_lock.
1762 if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) {
1763 kvm_activate_memslot(kvm, invalid_slot, old);
1764 kfree(invalid_slot);
1766 mutex_unlock(&kvm->slots_arch_lock);
1772 * For DELETE and MOVE, the working slot is now active as the INVALID
1773 * version of the old slot. MOVE is particularly special as it reuses
1774 * the old slot and returns a copy of the old slot (in working_slot).
1775 * For CREATE, there is no old slot. For DELETE and FLAGS_ONLY, the
1776 * old slot is detached but otherwise preserved.
1778 if (change == KVM_MR_CREATE)
1779 kvm_create_memslot(kvm, new);
1780 else if (change == KVM_MR_DELETE)
1781 kvm_delete_memslot(kvm, old, invalid_slot);
1782 else if (change == KVM_MR_MOVE)
1783 kvm_move_memslot(kvm, old, new, invalid_slot);
1784 else if (change == KVM_MR_FLAGS_ONLY)
1785 kvm_update_flags_memslot(kvm, old, new);
1789 /* Free the temporary INVALID slot used for DELETE and MOVE. */
1790 if (change == KVM_MR_DELETE || change == KVM_MR_MOVE)
1791 kfree(invalid_slot);
1794 * No need to refresh new->arch, changes after dropping slots_arch_lock
1795 * will directly hit the final, active memsot. Architectures are
1796 * responsible for knowing that new->arch may be stale.
1798 kvm_commit_memory_region(kvm, old, new, change);
1803 static bool kvm_check_memslot_overlap(struct kvm_memslots *slots, int id,
1804 gfn_t start, gfn_t end)
1806 struct kvm_memslot_iter iter;
1808 kvm_for_each_memslot_in_gfn_range(&iter, slots, start, end) {
1809 if (iter.slot->id != id)
1817 * Allocate some memory and give it an address in the guest physical address
1820 * Discontiguous memory is allowed, mostly for framebuffers.
1822 * Must be called holding kvm->slots_lock for write.
1824 int __kvm_set_memory_region(struct kvm *kvm,
1825 const struct kvm_userspace_memory_region *mem)
1827 struct kvm_memory_slot *old, *new;
1828 struct kvm_memslots *slots;
1829 enum kvm_mr_change change;
1830 unsigned long npages;
1835 r = check_memory_region_flags(mem);
1839 as_id = mem->slot >> 16;
1840 id = (u16)mem->slot;
1842 /* General sanity checks */
1843 if ((mem->memory_size & (PAGE_SIZE - 1)) ||
1844 (mem->memory_size != (unsigned long)mem->memory_size))
1846 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
1848 /* We can read the guest memory with __xxx_user() later on. */
1849 if ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
1850 (mem->userspace_addr != untagged_addr(mem->userspace_addr)) ||
1851 !access_ok((void __user *)(unsigned long)mem->userspace_addr,
1854 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
1856 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
1858 if ((mem->memory_size >> PAGE_SHIFT) > KVM_MEM_MAX_NR_PAGES)
1861 slots = __kvm_memslots(kvm, as_id);
1864 * Note, the old memslot (and the pointer itself!) may be invalidated
1865 * and/or destroyed by kvm_set_memslot().
1867 old = id_to_memslot(slots, id);
1869 if (!mem->memory_size) {
1870 if (!old || !old->npages)
1873 if (WARN_ON_ONCE(kvm->nr_memslot_pages < old->npages))
1876 return kvm_set_memslot(kvm, old, NULL, KVM_MR_DELETE);
1879 base_gfn = (mem->guest_phys_addr >> PAGE_SHIFT);
1880 npages = (mem->memory_size >> PAGE_SHIFT);
1882 if (!old || !old->npages) {
1883 change = KVM_MR_CREATE;
1886 * To simplify KVM internals, the total number of pages across
1887 * all memslots must fit in an unsigned long.
1889 if ((kvm->nr_memslot_pages + npages) < kvm->nr_memslot_pages)
1891 } else { /* Modify an existing slot. */
1892 if ((mem->userspace_addr != old->userspace_addr) ||
1893 (npages != old->npages) ||
1894 ((mem->flags ^ old->flags) & KVM_MEM_READONLY))
1897 if (base_gfn != old->base_gfn)
1898 change = KVM_MR_MOVE;
1899 else if (mem->flags != old->flags)
1900 change = KVM_MR_FLAGS_ONLY;
1901 else /* Nothing to change. */
1905 if ((change == KVM_MR_CREATE || change == KVM_MR_MOVE) &&
1906 kvm_check_memslot_overlap(slots, id, base_gfn, base_gfn + npages))
1909 /* Allocate a slot that will persist in the memslot. */
1910 new = kzalloc(sizeof(*new), GFP_KERNEL_ACCOUNT);
1916 new->base_gfn = base_gfn;
1917 new->npages = npages;
1918 new->flags = mem->flags;
1919 new->userspace_addr = mem->userspace_addr;
1921 r = kvm_set_memslot(kvm, old, new, change);
1926 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1928 int kvm_set_memory_region(struct kvm *kvm,
1929 const struct kvm_userspace_memory_region *mem)
1933 mutex_lock(&kvm->slots_lock);
1934 r = __kvm_set_memory_region(kvm, mem);
1935 mutex_unlock(&kvm->slots_lock);
1938 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1940 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1941 struct kvm_userspace_memory_region *mem)
1943 if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1946 return kvm_set_memory_region(kvm, mem);
1949 #ifndef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1951 * kvm_get_dirty_log - get a snapshot of dirty pages
1952 * @kvm: pointer to kvm instance
1953 * @log: slot id and address to which we copy the log
1954 * @is_dirty: set to '1' if any dirty pages were found
1955 * @memslot: set to the associated memslot, always valid on success
1957 int kvm_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log,
1958 int *is_dirty, struct kvm_memory_slot **memslot)
1960 struct kvm_memslots *slots;
1963 unsigned long any = 0;
1965 /* Dirty ring tracking is exclusive to dirty log tracking */
1966 if (kvm->dirty_ring_size)
1972 as_id = log->slot >> 16;
1973 id = (u16)log->slot;
1974 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1977 slots = __kvm_memslots(kvm, as_id);
1978 *memslot = id_to_memslot(slots, id);
1979 if (!(*memslot) || !(*memslot)->dirty_bitmap)
1982 kvm_arch_sync_dirty_log(kvm, *memslot);
1984 n = kvm_dirty_bitmap_bytes(*memslot);
1986 for (i = 0; !any && i < n/sizeof(long); ++i)
1987 any = (*memslot)->dirty_bitmap[i];
1989 if (copy_to_user(log->dirty_bitmap, (*memslot)->dirty_bitmap, n))
1996 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1998 #else /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
2000 * kvm_get_dirty_log_protect - get a snapshot of dirty pages
2001 * and reenable dirty page tracking for the corresponding pages.
2002 * @kvm: pointer to kvm instance
2003 * @log: slot id and address to which we copy the log
2005 * We need to keep it in mind that VCPU threads can write to the bitmap
2006 * concurrently. So, to avoid losing track of dirty pages we keep the
2009 * 1. Take a snapshot of the bit and clear it if needed.
2010 * 2. Write protect the corresponding page.
2011 * 3. Copy the snapshot to the userspace.
2012 * 4. Upon return caller flushes TLB's if needed.
2014 * Between 2 and 4, the guest may write to the page using the remaining TLB
2015 * entry. This is not a problem because the page is reported dirty using
2016 * the snapshot taken before and step 4 ensures that writes done after
2017 * exiting to userspace will be logged for the next call.
2020 static int kvm_get_dirty_log_protect(struct kvm *kvm, struct kvm_dirty_log *log)
2022 struct kvm_memslots *slots;
2023 struct kvm_memory_slot *memslot;
2026 unsigned long *dirty_bitmap;
2027 unsigned long *dirty_bitmap_buffer;
2030 /* Dirty ring tracking is exclusive to dirty log tracking */
2031 if (kvm->dirty_ring_size)
2034 as_id = log->slot >> 16;
2035 id = (u16)log->slot;
2036 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
2039 slots = __kvm_memslots(kvm, as_id);
2040 memslot = id_to_memslot(slots, id);
2041 if (!memslot || !memslot->dirty_bitmap)
2044 dirty_bitmap = memslot->dirty_bitmap;
2046 kvm_arch_sync_dirty_log(kvm, memslot);
2048 n = kvm_dirty_bitmap_bytes(memslot);
2050 if (kvm->manual_dirty_log_protect) {
2052 * Unlike kvm_get_dirty_log, we always return false in *flush,
2053 * because no flush is needed until KVM_CLEAR_DIRTY_LOG. There
2054 * is some code duplication between this function and
2055 * kvm_get_dirty_log, but hopefully all architecture
2056 * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log
2057 * can be eliminated.
2059 dirty_bitmap_buffer = dirty_bitmap;
2061 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
2062 memset(dirty_bitmap_buffer, 0, n);
2065 for (i = 0; i < n / sizeof(long); i++) {
2069 if (!dirty_bitmap[i])
2073 mask = xchg(&dirty_bitmap[i], 0);
2074 dirty_bitmap_buffer[i] = mask;
2076 offset = i * BITS_PER_LONG;
2077 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
2080 KVM_MMU_UNLOCK(kvm);
2084 kvm_arch_flush_remote_tlbs_memslot(kvm, memslot);
2086 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
2093 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
2094 * @kvm: kvm instance
2095 * @log: slot id and address to which we copy the log
2097 * Steps 1-4 below provide general overview of dirty page logging. See
2098 * kvm_get_dirty_log_protect() function description for additional details.
2100 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
2101 * always flush the TLB (step 4) even if previous step failed and the dirty
2102 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
2103 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
2104 * writes will be marked dirty for next log read.
2106 * 1. Take a snapshot of the bit and clear it if needed.
2107 * 2. Write protect the corresponding page.
2108 * 3. Copy the snapshot to the userspace.
2109 * 4. Flush TLB's if needed.
2111 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
2112 struct kvm_dirty_log *log)
2116 mutex_lock(&kvm->slots_lock);
2118 r = kvm_get_dirty_log_protect(kvm, log);
2120 mutex_unlock(&kvm->slots_lock);
2125 * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap
2126 * and reenable dirty page tracking for the corresponding pages.
2127 * @kvm: pointer to kvm instance
2128 * @log: slot id and address from which to fetch the bitmap of dirty pages
2130 static int kvm_clear_dirty_log_protect(struct kvm *kvm,
2131 struct kvm_clear_dirty_log *log)
2133 struct kvm_memslots *slots;
2134 struct kvm_memory_slot *memslot;
2138 unsigned long *dirty_bitmap;
2139 unsigned long *dirty_bitmap_buffer;
2142 /* Dirty ring tracking is exclusive to dirty log tracking */
2143 if (kvm->dirty_ring_size)
2146 as_id = log->slot >> 16;
2147 id = (u16)log->slot;
2148 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
2151 if (log->first_page & 63)
2154 slots = __kvm_memslots(kvm, as_id);
2155 memslot = id_to_memslot(slots, id);
2156 if (!memslot || !memslot->dirty_bitmap)
2159 dirty_bitmap = memslot->dirty_bitmap;
2161 n = ALIGN(log->num_pages, BITS_PER_LONG) / 8;
2163 if (log->first_page > memslot->npages ||
2164 log->num_pages > memslot->npages - log->first_page ||
2165 (log->num_pages < memslot->npages - log->first_page && (log->num_pages & 63)))
2168 kvm_arch_sync_dirty_log(kvm, memslot);
2171 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
2172 if (copy_from_user(dirty_bitmap_buffer, log->dirty_bitmap, n))
2176 for (offset = log->first_page, i = offset / BITS_PER_LONG,
2177 n = DIV_ROUND_UP(log->num_pages, BITS_PER_LONG); n--;
2178 i++, offset += BITS_PER_LONG) {
2179 unsigned long mask = *dirty_bitmap_buffer++;
2180 atomic_long_t *p = (atomic_long_t *) &dirty_bitmap[i];
2184 mask &= atomic_long_fetch_andnot(mask, p);
2187 * mask contains the bits that really have been cleared. This
2188 * never includes any bits beyond the length of the memslot (if
2189 * the length is not aligned to 64 pages), therefore it is not
2190 * a problem if userspace sets them in log->dirty_bitmap.
2194 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
2198 KVM_MMU_UNLOCK(kvm);
2201 kvm_arch_flush_remote_tlbs_memslot(kvm, memslot);
2206 static int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm,
2207 struct kvm_clear_dirty_log *log)
2211 mutex_lock(&kvm->slots_lock);
2213 r = kvm_clear_dirty_log_protect(kvm, log);
2215 mutex_unlock(&kvm->slots_lock);
2218 #endif /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
2220 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
2222 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
2224 EXPORT_SYMBOL_GPL(gfn_to_memslot);
2226 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
2228 struct kvm_memslots *slots = kvm_vcpu_memslots(vcpu);
2229 u64 gen = slots->generation;
2230 struct kvm_memory_slot *slot;
2233 * This also protects against using a memslot from a different address space,
2234 * since different address spaces have different generation numbers.
2236 if (unlikely(gen != vcpu->last_used_slot_gen)) {
2237 vcpu->last_used_slot = NULL;
2238 vcpu->last_used_slot_gen = gen;
2241 slot = try_get_memslot(vcpu->last_used_slot, gfn);
2246 * Fall back to searching all memslots. We purposely use
2247 * search_memslots() instead of __gfn_to_memslot() to avoid
2248 * thrashing the VM-wide last_used_slot in kvm_memslots.
2250 slot = search_memslots(slots, gfn, false);
2252 vcpu->last_used_slot = slot;
2259 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
2261 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
2263 return kvm_is_visible_memslot(memslot);
2265 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
2267 bool kvm_vcpu_is_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
2269 struct kvm_memory_slot *memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2271 return kvm_is_visible_memslot(memslot);
2273 EXPORT_SYMBOL_GPL(kvm_vcpu_is_visible_gfn);
2275 unsigned long kvm_host_page_size(struct kvm_vcpu *vcpu, gfn_t gfn)
2277 struct vm_area_struct *vma;
2278 unsigned long addr, size;
2282 addr = kvm_vcpu_gfn_to_hva_prot(vcpu, gfn, NULL);
2283 if (kvm_is_error_hva(addr))
2286 mmap_read_lock(current->mm);
2287 vma = find_vma(current->mm, addr);
2291 size = vma_kernel_pagesize(vma);
2294 mmap_read_unlock(current->mm);
2299 static bool memslot_is_readonly(const struct kvm_memory_slot *slot)
2301 return slot->flags & KVM_MEM_READONLY;
2304 static unsigned long __gfn_to_hva_many(const struct kvm_memory_slot *slot, gfn_t gfn,
2305 gfn_t *nr_pages, bool write)
2307 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
2308 return KVM_HVA_ERR_BAD;
2310 if (memslot_is_readonly(slot) && write)
2311 return KVM_HVA_ERR_RO_BAD;
2314 *nr_pages = slot->npages - (gfn - slot->base_gfn);
2316 return __gfn_to_hva_memslot(slot, gfn);
2319 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
2322 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
2325 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
2328 return gfn_to_hva_many(slot, gfn, NULL);
2330 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
2332 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
2334 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
2336 EXPORT_SYMBOL_GPL(gfn_to_hva);
2338 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
2340 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
2342 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
2345 * Return the hva of a @gfn and the R/W attribute if possible.
2347 * @slot: the kvm_memory_slot which contains @gfn
2348 * @gfn: the gfn to be translated
2349 * @writable: used to return the read/write attribute of the @slot if the hva
2350 * is valid and @writable is not NULL
2352 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
2353 gfn_t gfn, bool *writable)
2355 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
2357 if (!kvm_is_error_hva(hva) && writable)
2358 *writable = !memslot_is_readonly(slot);
2363 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
2365 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2367 return gfn_to_hva_memslot_prot(slot, gfn, writable);
2370 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
2372 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2374 return gfn_to_hva_memslot_prot(slot, gfn, writable);
2377 static inline int check_user_page_hwpoison(unsigned long addr)
2379 int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
2381 rc = get_user_pages(addr, 1, flags, NULL, NULL);
2382 return rc == -EHWPOISON;
2386 * The fast path to get the writable pfn which will be stored in @pfn,
2387 * true indicates success, otherwise false is returned. It's also the
2388 * only part that runs if we can in atomic context.
2390 static bool hva_to_pfn_fast(unsigned long addr, bool write_fault,
2391 bool *writable, kvm_pfn_t *pfn)
2393 struct page *page[1];
2396 * Fast pin a writable pfn only if it is a write fault request
2397 * or the caller allows to map a writable pfn for a read fault
2400 if (!(write_fault || writable))
2403 if (get_user_page_fast_only(addr, FOLL_WRITE, page)) {
2404 *pfn = page_to_pfn(page[0]);
2415 * The slow path to get the pfn of the specified host virtual address,
2416 * 1 indicates success, -errno is returned if error is detected.
2418 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
2419 bool *writable, kvm_pfn_t *pfn)
2421 unsigned int flags = FOLL_HWPOISON;
2428 *writable = write_fault;
2431 flags |= FOLL_WRITE;
2433 flags |= FOLL_NOWAIT;
2435 npages = get_user_pages_unlocked(addr, 1, &page, flags);
2439 /* map read fault as writable if possible */
2440 if (unlikely(!write_fault) && writable) {
2443 if (get_user_page_fast_only(addr, FOLL_WRITE, &wpage)) {
2449 *pfn = page_to_pfn(page);
2453 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
2455 if (unlikely(!(vma->vm_flags & VM_READ)))
2458 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
2464 static int kvm_try_get_pfn(kvm_pfn_t pfn)
2466 if (kvm_is_reserved_pfn(pfn))
2468 return get_page_unless_zero(pfn_to_page(pfn));
2471 static int hva_to_pfn_remapped(struct vm_area_struct *vma,
2472 unsigned long addr, bool write_fault,
2473 bool *writable, kvm_pfn_t *p_pfn)
2480 r = follow_pte(vma->vm_mm, addr, &ptep, &ptl);
2483 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
2484 * not call the fault handler, so do it here.
2486 bool unlocked = false;
2487 r = fixup_user_fault(current->mm, addr,
2488 (write_fault ? FAULT_FLAG_WRITE : 0),
2495 r = follow_pte(vma->vm_mm, addr, &ptep, &ptl);
2500 if (write_fault && !pte_write(*ptep)) {
2501 pfn = KVM_PFN_ERR_RO_FAULT;
2506 *writable = pte_write(*ptep);
2507 pfn = pte_pfn(*ptep);
2510 * Get a reference here because callers of *hva_to_pfn* and
2511 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
2512 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
2513 * set, but the kvm_try_get_pfn/kvm_release_pfn_clean pair will
2514 * simply do nothing for reserved pfns.
2516 * Whoever called remap_pfn_range is also going to call e.g.
2517 * unmap_mapping_range before the underlying pages are freed,
2518 * causing a call to our MMU notifier.
2520 * Certain IO or PFNMAP mappings can be backed with valid
2521 * struct pages, but be allocated without refcounting e.g.,
2522 * tail pages of non-compound higher order allocations, which
2523 * would then underflow the refcount when the caller does the
2524 * required put_page. Don't allow those pages here.
2526 if (!kvm_try_get_pfn(pfn))
2530 pte_unmap_unlock(ptep, ptl);
2537 * Pin guest page in memory and return its pfn.
2538 * @addr: host virtual address which maps memory to the guest
2539 * @atomic: whether this function can sleep
2540 * @async: whether this function need to wait IO complete if the
2541 * host page is not in the memory
2542 * @write_fault: whether we should get a writable host page
2543 * @writable: whether it allows to map a writable host page for !@write_fault
2545 * The function will map a writable host page for these two cases:
2546 * 1): @write_fault = true
2547 * 2): @write_fault = false && @writable, @writable will tell the caller
2548 * whether the mapping is writable.
2550 kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
2551 bool write_fault, bool *writable)
2553 struct vm_area_struct *vma;
2557 /* we can do it either atomically or asynchronously, not both */
2558 BUG_ON(atomic && async);
2560 if (hva_to_pfn_fast(addr, write_fault, writable, &pfn))
2564 return KVM_PFN_ERR_FAULT;
2566 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
2570 mmap_read_lock(current->mm);
2571 if (npages == -EHWPOISON ||
2572 (!async && check_user_page_hwpoison(addr))) {
2573 pfn = KVM_PFN_ERR_HWPOISON;
2578 vma = vma_lookup(current->mm, addr);
2581 pfn = KVM_PFN_ERR_FAULT;
2582 else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
2583 r = hva_to_pfn_remapped(vma, addr, write_fault, writable, &pfn);
2587 pfn = KVM_PFN_ERR_FAULT;
2589 if (async && vma_is_valid(vma, write_fault))
2591 pfn = KVM_PFN_ERR_FAULT;
2594 mmap_read_unlock(current->mm);
2598 kvm_pfn_t __gfn_to_pfn_memslot(const struct kvm_memory_slot *slot, gfn_t gfn,
2599 bool atomic, bool *async, bool write_fault,
2600 bool *writable, hva_t *hva)
2602 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
2607 if (addr == KVM_HVA_ERR_RO_BAD) {
2610 return KVM_PFN_ERR_RO_FAULT;
2613 if (kvm_is_error_hva(addr)) {
2616 return KVM_PFN_NOSLOT;
2619 /* Do not map writable pfn in the readonly memslot. */
2620 if (writable && memslot_is_readonly(slot)) {
2625 return hva_to_pfn(addr, atomic, async, write_fault,
2628 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
2630 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
2633 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
2634 write_fault, writable, NULL);
2636 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
2638 kvm_pfn_t gfn_to_pfn_memslot(const struct kvm_memory_slot *slot, gfn_t gfn)
2640 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL, NULL);
2642 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
2644 kvm_pfn_t gfn_to_pfn_memslot_atomic(const struct kvm_memory_slot *slot, gfn_t gfn)
2646 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL, NULL);
2648 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
2650 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
2652 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
2654 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
2656 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
2658 return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
2660 EXPORT_SYMBOL_GPL(gfn_to_pfn);
2662 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
2664 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
2666 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
2668 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
2669 struct page **pages, int nr_pages)
2674 addr = gfn_to_hva_many(slot, gfn, &entry);
2675 if (kvm_is_error_hva(addr))
2678 if (entry < nr_pages)
2681 return get_user_pages_fast_only(addr, nr_pages, FOLL_WRITE, pages);
2683 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
2685 static struct page *kvm_pfn_to_page(kvm_pfn_t pfn)
2687 if (is_error_noslot_pfn(pfn))
2688 return KVM_ERR_PTR_BAD_PAGE;
2690 if (kvm_is_reserved_pfn(pfn)) {
2692 return KVM_ERR_PTR_BAD_PAGE;
2695 return pfn_to_page(pfn);
2698 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
2702 pfn = gfn_to_pfn(kvm, gfn);
2704 return kvm_pfn_to_page(pfn);
2706 EXPORT_SYMBOL_GPL(gfn_to_page);
2708 void kvm_release_pfn(kvm_pfn_t pfn, bool dirty)
2714 kvm_release_pfn_dirty(pfn);
2716 kvm_release_pfn_clean(pfn);
2719 int kvm_vcpu_map(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map)
2723 struct page *page = KVM_UNMAPPED_PAGE;
2728 pfn = gfn_to_pfn(vcpu->kvm, gfn);
2729 if (is_error_noslot_pfn(pfn))
2732 if (pfn_valid(pfn)) {
2733 page = pfn_to_page(pfn);
2735 #ifdef CONFIG_HAS_IOMEM
2737 hva = memremap(pfn_to_hpa(pfn), PAGE_SIZE, MEMREMAP_WB);
2751 EXPORT_SYMBOL_GPL(kvm_vcpu_map);
2753 void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map, bool dirty)
2761 if (map->page != KVM_UNMAPPED_PAGE)
2763 #ifdef CONFIG_HAS_IOMEM
2769 kvm_vcpu_mark_page_dirty(vcpu, map->gfn);
2771 kvm_release_pfn(map->pfn, dirty);
2776 EXPORT_SYMBOL_GPL(kvm_vcpu_unmap);
2778 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
2782 pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
2784 return kvm_pfn_to_page(pfn);
2786 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
2788 void kvm_release_page_clean(struct page *page)
2790 WARN_ON(is_error_page(page));
2792 kvm_release_pfn_clean(page_to_pfn(page));
2794 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
2796 void kvm_release_pfn_clean(kvm_pfn_t pfn)
2798 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
2799 put_page(pfn_to_page(pfn));
2801 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
2803 void kvm_release_page_dirty(struct page *page)
2805 WARN_ON(is_error_page(page));
2807 kvm_release_pfn_dirty(page_to_pfn(page));
2809 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
2811 void kvm_release_pfn_dirty(kvm_pfn_t pfn)
2813 kvm_set_pfn_dirty(pfn);
2814 kvm_release_pfn_clean(pfn);
2816 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
2818 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
2820 if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn))
2821 SetPageDirty(pfn_to_page(pfn));
2823 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
2825 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
2827 if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn))
2828 mark_page_accessed(pfn_to_page(pfn));
2830 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
2832 static int next_segment(unsigned long len, int offset)
2834 if (len > PAGE_SIZE - offset)
2835 return PAGE_SIZE - offset;
2840 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
2841 void *data, int offset, int len)
2846 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
2847 if (kvm_is_error_hva(addr))
2849 r = __copy_from_user(data, (void __user *)addr + offset, len);
2855 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
2858 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2860 return __kvm_read_guest_page(slot, gfn, data, offset, len);
2862 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
2864 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
2865 int offset, int len)
2867 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2869 return __kvm_read_guest_page(slot, gfn, data, offset, len);
2871 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
2873 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
2875 gfn_t gfn = gpa >> PAGE_SHIFT;
2877 int offset = offset_in_page(gpa);
2880 while ((seg = next_segment(len, offset)) != 0) {
2881 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
2891 EXPORT_SYMBOL_GPL(kvm_read_guest);
2893 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
2895 gfn_t gfn = gpa >> PAGE_SHIFT;
2897 int offset = offset_in_page(gpa);
2900 while ((seg = next_segment(len, offset)) != 0) {
2901 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
2911 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
2913 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
2914 void *data, int offset, unsigned long len)
2919 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
2920 if (kvm_is_error_hva(addr))
2922 pagefault_disable();
2923 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
2930 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
2931 void *data, unsigned long len)
2933 gfn_t gfn = gpa >> PAGE_SHIFT;
2934 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2935 int offset = offset_in_page(gpa);
2937 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
2939 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
2941 static int __kvm_write_guest_page(struct kvm *kvm,
2942 struct kvm_memory_slot *memslot, gfn_t gfn,
2943 const void *data, int offset, int len)
2948 addr = gfn_to_hva_memslot(memslot, gfn);
2949 if (kvm_is_error_hva(addr))
2951 r = __copy_to_user((void __user *)addr + offset, data, len);
2954 mark_page_dirty_in_slot(kvm, memslot, gfn);
2958 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
2959 const void *data, int offset, int len)
2961 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2963 return __kvm_write_guest_page(kvm, slot, gfn, data, offset, len);
2965 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
2967 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
2968 const void *data, int offset, int len)
2970 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2972 return __kvm_write_guest_page(vcpu->kvm, slot, gfn, data, offset, len);
2974 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
2976 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
2979 gfn_t gfn = gpa >> PAGE_SHIFT;
2981 int offset = offset_in_page(gpa);
2984 while ((seg = next_segment(len, offset)) != 0) {
2985 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
2995 EXPORT_SYMBOL_GPL(kvm_write_guest);
2997 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
3000 gfn_t gfn = gpa >> PAGE_SHIFT;
3002 int offset = offset_in_page(gpa);
3005 while ((seg = next_segment(len, offset)) != 0) {
3006 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
3016 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
3018 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
3019 struct gfn_to_hva_cache *ghc,
3020 gpa_t gpa, unsigned long len)
3022 int offset = offset_in_page(gpa);
3023 gfn_t start_gfn = gpa >> PAGE_SHIFT;
3024 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
3025 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
3026 gfn_t nr_pages_avail;
3028 /* Update ghc->generation before performing any error checks. */
3029 ghc->generation = slots->generation;
3031 if (start_gfn > end_gfn) {
3032 ghc->hva = KVM_HVA_ERR_BAD;
3037 * If the requested region crosses two memslots, we still
3038 * verify that the entire region is valid here.
3040 for ( ; start_gfn <= end_gfn; start_gfn += nr_pages_avail) {
3041 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
3042 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
3044 if (kvm_is_error_hva(ghc->hva))
3048 /* Use the slow path for cross page reads and writes. */
3049 if (nr_pages_needed == 1)
3052 ghc->memslot = NULL;
3059 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3060 gpa_t gpa, unsigned long len)
3062 struct kvm_memslots *slots = kvm_memslots(kvm);
3063 return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len);
3065 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
3067 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3068 void *data, unsigned int offset,
3071 struct kvm_memslots *slots = kvm_memslots(kvm);
3073 gpa_t gpa = ghc->gpa + offset;
3075 if (WARN_ON_ONCE(len + offset > ghc->len))
3078 if (slots->generation != ghc->generation) {
3079 if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len))
3083 if (kvm_is_error_hva(ghc->hva))
3086 if (unlikely(!ghc->memslot))
3087 return kvm_write_guest(kvm, gpa, data, len);
3089 r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
3092 mark_page_dirty_in_slot(kvm, ghc->memslot, gpa >> PAGE_SHIFT);
3096 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached);
3098 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3099 void *data, unsigned long len)
3101 return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len);
3103 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
3105 int kvm_read_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3106 void *data, unsigned int offset,
3109 struct kvm_memslots *slots = kvm_memslots(kvm);
3111 gpa_t gpa = ghc->gpa + offset;
3113 if (WARN_ON_ONCE(len + offset > ghc->len))
3116 if (slots->generation != ghc->generation) {
3117 if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len))
3121 if (kvm_is_error_hva(ghc->hva))
3124 if (unlikely(!ghc->memslot))
3125 return kvm_read_guest(kvm, gpa, data, len);
3127 r = __copy_from_user(data, (void __user *)ghc->hva + offset, len);
3133 EXPORT_SYMBOL_GPL(kvm_read_guest_offset_cached);
3135 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3136 void *data, unsigned long len)
3138 return kvm_read_guest_offset_cached(kvm, ghc, data, 0, len);
3140 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
3142 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
3144 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
3145 gfn_t gfn = gpa >> PAGE_SHIFT;
3147 int offset = offset_in_page(gpa);
3150 while ((seg = next_segment(len, offset)) != 0) {
3151 ret = kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
3160 EXPORT_SYMBOL_GPL(kvm_clear_guest);
3162 void mark_page_dirty_in_slot(struct kvm *kvm,
3163 const struct kvm_memory_slot *memslot,
3166 struct kvm_vcpu *vcpu = kvm_get_running_vcpu();
3168 #ifdef CONFIG_HAVE_KVM_DIRTY_RING
3169 if (WARN_ON_ONCE(!vcpu) || WARN_ON_ONCE(vcpu->kvm != kvm))
3173 if (memslot && kvm_slot_dirty_track_enabled(memslot)) {
3174 unsigned long rel_gfn = gfn - memslot->base_gfn;
3175 u32 slot = (memslot->as_id << 16) | memslot->id;
3177 if (kvm->dirty_ring_size)
3178 kvm_dirty_ring_push(&vcpu->dirty_ring,
3181 set_bit_le(rel_gfn, memslot->dirty_bitmap);
3184 EXPORT_SYMBOL_GPL(mark_page_dirty_in_slot);
3186 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
3188 struct kvm_memory_slot *memslot;
3190 memslot = gfn_to_memslot(kvm, gfn);
3191 mark_page_dirty_in_slot(kvm, memslot, gfn);
3193 EXPORT_SYMBOL_GPL(mark_page_dirty);
3195 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
3197 struct kvm_memory_slot *memslot;
3199 memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
3200 mark_page_dirty_in_slot(vcpu->kvm, memslot, gfn);
3202 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
3204 void kvm_sigset_activate(struct kvm_vcpu *vcpu)
3206 if (!vcpu->sigset_active)
3210 * This does a lockless modification of ->real_blocked, which is fine
3211 * because, only current can change ->real_blocked and all readers of
3212 * ->real_blocked don't care as long ->real_blocked is always a subset
3215 sigprocmask(SIG_SETMASK, &vcpu->sigset, ¤t->real_blocked);
3218 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu)
3220 if (!vcpu->sigset_active)
3223 sigprocmask(SIG_SETMASK, ¤t->real_blocked, NULL);
3224 sigemptyset(¤t->real_blocked);
3227 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
3229 unsigned int old, val, grow, grow_start;
3231 old = val = vcpu->halt_poll_ns;
3232 grow_start = READ_ONCE(halt_poll_ns_grow_start);
3233 grow = READ_ONCE(halt_poll_ns_grow);
3238 if (val < grow_start)
3241 if (val > vcpu->kvm->max_halt_poll_ns)
3242 val = vcpu->kvm->max_halt_poll_ns;
3244 vcpu->halt_poll_ns = val;
3246 trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
3249 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
3251 unsigned int old, val, shrink, grow_start;
3253 old = val = vcpu->halt_poll_ns;
3254 shrink = READ_ONCE(halt_poll_ns_shrink);
3255 grow_start = READ_ONCE(halt_poll_ns_grow_start);
3261 if (val < grow_start)
3264 vcpu->halt_poll_ns = val;
3265 trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
3268 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
3271 int idx = srcu_read_lock(&vcpu->kvm->srcu);
3273 if (kvm_arch_vcpu_runnable(vcpu)) {
3274 kvm_make_request(KVM_REQ_UNHALT, vcpu);
3277 if (kvm_cpu_has_pending_timer(vcpu))
3279 if (signal_pending(current))
3281 if (kvm_check_request(KVM_REQ_UNBLOCK, vcpu))
3286 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3291 * Block the vCPU until the vCPU is runnable, an event arrives, or a signal is
3292 * pending. This is mostly used when halting a vCPU, but may also be used
3293 * directly for other vCPU non-runnable states, e.g. x86's Wait-For-SIPI.
3295 bool kvm_vcpu_block(struct kvm_vcpu *vcpu)
3297 struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu);
3298 bool waited = false;
3300 vcpu->stat.generic.blocking = 1;
3302 kvm_arch_vcpu_blocking(vcpu);
3304 prepare_to_rcuwait(wait);
3306 set_current_state(TASK_INTERRUPTIBLE);
3308 if (kvm_vcpu_check_block(vcpu) < 0)
3314 finish_rcuwait(wait);
3316 kvm_arch_vcpu_unblocking(vcpu);
3318 vcpu->stat.generic.blocking = 0;
3323 static inline void update_halt_poll_stats(struct kvm_vcpu *vcpu, ktime_t start,
3324 ktime_t end, bool success)
3326 struct kvm_vcpu_stat_generic *stats = &vcpu->stat.generic;
3327 u64 poll_ns = ktime_to_ns(ktime_sub(end, start));
3329 ++vcpu->stat.generic.halt_attempted_poll;
3332 ++vcpu->stat.generic.halt_successful_poll;
3334 if (!vcpu_valid_wakeup(vcpu))
3335 ++vcpu->stat.generic.halt_poll_invalid;
3337 stats->halt_poll_success_ns += poll_ns;
3338 KVM_STATS_LOG_HIST_UPDATE(stats->halt_poll_success_hist, poll_ns);
3340 stats->halt_poll_fail_ns += poll_ns;
3341 KVM_STATS_LOG_HIST_UPDATE(stats->halt_poll_fail_hist, poll_ns);
3346 * Emulate a vCPU halt condition, e.g. HLT on x86, WFI on arm, etc... If halt
3347 * polling is enabled, busy wait for a short time before blocking to avoid the
3348 * expensive block+unblock sequence if a wake event arrives soon after the vCPU
3351 void kvm_vcpu_halt(struct kvm_vcpu *vcpu)
3353 bool halt_poll_allowed = !kvm_arch_no_poll(vcpu);
3354 bool do_halt_poll = halt_poll_allowed && vcpu->halt_poll_ns;
3355 ktime_t start, cur, poll_end;
3356 bool waited = false;
3359 start = cur = poll_end = ktime_get();
3361 ktime_t stop = ktime_add_ns(start, vcpu->halt_poll_ns);
3365 * This sets KVM_REQ_UNHALT if an interrupt
3368 if (kvm_vcpu_check_block(vcpu) < 0)
3371 poll_end = cur = ktime_get();
3372 } while (kvm_vcpu_can_poll(cur, stop));
3375 waited = kvm_vcpu_block(vcpu);
3379 vcpu->stat.generic.halt_wait_ns +=
3380 ktime_to_ns(cur) - ktime_to_ns(poll_end);
3381 KVM_STATS_LOG_HIST_UPDATE(vcpu->stat.generic.halt_wait_hist,
3382 ktime_to_ns(cur) - ktime_to_ns(poll_end));
3385 /* The total time the vCPU was "halted", including polling time. */
3386 halt_ns = ktime_to_ns(cur) - ktime_to_ns(start);
3389 * Note, halt-polling is considered successful so long as the vCPU was
3390 * never actually scheduled out, i.e. even if the wake event arrived
3391 * after of the halt-polling loop itself, but before the full wait.
3394 update_halt_poll_stats(vcpu, start, poll_end, !waited);
3396 if (halt_poll_allowed) {
3397 if (!vcpu_valid_wakeup(vcpu)) {
3398 shrink_halt_poll_ns(vcpu);
3399 } else if (vcpu->kvm->max_halt_poll_ns) {
3400 if (halt_ns <= vcpu->halt_poll_ns)
3402 /* we had a long block, shrink polling */
3403 else if (vcpu->halt_poll_ns &&
3404 halt_ns > vcpu->kvm->max_halt_poll_ns)
3405 shrink_halt_poll_ns(vcpu);
3406 /* we had a short halt and our poll time is too small */
3407 else if (vcpu->halt_poll_ns < vcpu->kvm->max_halt_poll_ns &&
3408 halt_ns < vcpu->kvm->max_halt_poll_ns)
3409 grow_halt_poll_ns(vcpu);
3411 vcpu->halt_poll_ns = 0;
3415 trace_kvm_vcpu_wakeup(halt_ns, waited, vcpu_valid_wakeup(vcpu));
3417 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
3419 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
3421 if (__kvm_vcpu_wake_up(vcpu)) {
3422 WRITE_ONCE(vcpu->ready, true);
3423 ++vcpu->stat.generic.halt_wakeup;
3429 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
3433 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
3435 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
3439 if (kvm_vcpu_wake_up(vcpu))
3444 * The only state change done outside the vcpu mutex is IN_GUEST_MODE
3445 * to EXITING_GUEST_MODE. Therefore the moderately expensive "should
3446 * kick" check does not need atomic operations if kvm_vcpu_kick is used
3447 * within the vCPU thread itself.
3449 if (vcpu == __this_cpu_read(kvm_running_vcpu)) {
3450 if (vcpu->mode == IN_GUEST_MODE)
3451 WRITE_ONCE(vcpu->mode, EXITING_GUEST_MODE);
3456 * Note, the vCPU could get migrated to a different pCPU at any point
3457 * after kvm_arch_vcpu_should_kick(), which could result in sending an
3458 * IPI to the previous pCPU. But, that's ok because the purpose of the
3459 * IPI is to force the vCPU to leave IN_GUEST_MODE, and migrating the
3460 * vCPU also requires it to leave IN_GUEST_MODE.
3462 if (kvm_arch_vcpu_should_kick(vcpu)) {
3463 cpu = READ_ONCE(vcpu->cpu);
3464 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
3465 smp_send_reschedule(cpu);
3470 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
3471 #endif /* !CONFIG_S390 */
3473 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
3476 struct task_struct *task = NULL;
3480 pid = rcu_dereference(target->pid);
3482 task = get_pid_task(pid, PIDTYPE_PID);
3486 ret = yield_to(task, 1);
3487 put_task_struct(task);
3491 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
3494 * Helper that checks whether a VCPU is eligible for directed yield.
3495 * Most eligible candidate to yield is decided by following heuristics:
3497 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
3498 * (preempted lock holder), indicated by @in_spin_loop.
3499 * Set at the beginning and cleared at the end of interception/PLE handler.
3501 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
3502 * chance last time (mostly it has become eligible now since we have probably
3503 * yielded to lockholder in last iteration. This is done by toggling
3504 * @dy_eligible each time a VCPU checked for eligibility.)
3506 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
3507 * to preempted lock-holder could result in wrong VCPU selection and CPU
3508 * burning. Giving priority for a potential lock-holder increases lock
3511 * Since algorithm is based on heuristics, accessing another VCPU data without
3512 * locking does not harm. It may result in trying to yield to same VCPU, fail
3513 * and continue with next VCPU and so on.
3515 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
3517 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
3520 eligible = !vcpu->spin_loop.in_spin_loop ||
3521 vcpu->spin_loop.dy_eligible;
3523 if (vcpu->spin_loop.in_spin_loop)
3524 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
3533 * Unlike kvm_arch_vcpu_runnable, this function is called outside
3534 * a vcpu_load/vcpu_put pair. However, for most architectures
3535 * kvm_arch_vcpu_runnable does not require vcpu_load.
3537 bool __weak kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
3539 return kvm_arch_vcpu_runnable(vcpu);
3542 static bool vcpu_dy_runnable(struct kvm_vcpu *vcpu)
3544 if (kvm_arch_dy_runnable(vcpu))
3547 #ifdef CONFIG_KVM_ASYNC_PF
3548 if (!list_empty_careful(&vcpu->async_pf.done))
3555 bool __weak kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu)
3560 void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode)
3562 struct kvm *kvm = me->kvm;
3563 struct kvm_vcpu *vcpu;
3564 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
3570 kvm_vcpu_set_in_spin_loop(me, true);
3572 * We boost the priority of a VCPU that is runnable but not
3573 * currently running, because it got preempted by something
3574 * else and called schedule in __vcpu_run. Hopefully that
3575 * VCPU is holding the lock that we need and will release it.
3576 * We approximate round-robin by starting at the last boosted VCPU.
3578 for (pass = 0; pass < 2 && !yielded && try; pass++) {
3579 kvm_for_each_vcpu(i, vcpu, kvm) {
3580 if (!pass && i <= last_boosted_vcpu) {
3581 i = last_boosted_vcpu;
3583 } else if (pass && i > last_boosted_vcpu)
3585 if (!READ_ONCE(vcpu->ready))
3589 if (kvm_vcpu_is_blocking(vcpu) && !vcpu_dy_runnable(vcpu))
3591 if (READ_ONCE(vcpu->preempted) && yield_to_kernel_mode &&
3592 !kvm_arch_dy_has_pending_interrupt(vcpu) &&
3593 !kvm_arch_vcpu_in_kernel(vcpu))
3595 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
3598 yielded = kvm_vcpu_yield_to(vcpu);
3600 kvm->last_boosted_vcpu = i;
3602 } else if (yielded < 0) {
3609 kvm_vcpu_set_in_spin_loop(me, false);
3611 /* Ensure vcpu is not eligible during next spinloop */
3612 kvm_vcpu_set_dy_eligible(me, false);
3614 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
3616 static bool kvm_page_in_dirty_ring(struct kvm *kvm, unsigned long pgoff)
3618 #ifdef CONFIG_HAVE_KVM_DIRTY_RING
3619 return (pgoff >= KVM_DIRTY_LOG_PAGE_OFFSET) &&
3620 (pgoff < KVM_DIRTY_LOG_PAGE_OFFSET +
3621 kvm->dirty_ring_size / PAGE_SIZE);
3627 static vm_fault_t kvm_vcpu_fault(struct vm_fault *vmf)
3629 struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data;
3632 if (vmf->pgoff == 0)
3633 page = virt_to_page(vcpu->run);
3635 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
3636 page = virt_to_page(vcpu->arch.pio_data);
3638 #ifdef CONFIG_KVM_MMIO
3639 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
3640 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
3642 else if (kvm_page_in_dirty_ring(vcpu->kvm, vmf->pgoff))
3643 page = kvm_dirty_ring_get_page(
3645 vmf->pgoff - KVM_DIRTY_LOG_PAGE_OFFSET);
3647 return kvm_arch_vcpu_fault(vcpu, vmf);
3653 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
3654 .fault = kvm_vcpu_fault,
3657 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
3659 struct kvm_vcpu *vcpu = file->private_data;
3660 unsigned long pages = vma_pages(vma);
3662 if ((kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff) ||
3663 kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff + pages - 1)) &&
3664 ((vma->vm_flags & VM_EXEC) || !(vma->vm_flags & VM_SHARED)))
3667 vma->vm_ops = &kvm_vcpu_vm_ops;
3671 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
3673 struct kvm_vcpu *vcpu = filp->private_data;
3675 kvm_put_kvm(vcpu->kvm);
3679 static struct file_operations kvm_vcpu_fops = {
3680 .release = kvm_vcpu_release,
3681 .unlocked_ioctl = kvm_vcpu_ioctl,
3682 .mmap = kvm_vcpu_mmap,
3683 .llseek = noop_llseek,
3684 KVM_COMPAT(kvm_vcpu_compat_ioctl),
3688 * Allocates an inode for the vcpu.
3690 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
3692 char name[8 + 1 + ITOA_MAX_LEN + 1];
3694 snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id);
3695 return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
3698 static void kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
3700 #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS
3701 struct dentry *debugfs_dentry;
3702 char dir_name[ITOA_MAX_LEN * 2];
3704 if (!debugfs_initialized())
3707 snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
3708 debugfs_dentry = debugfs_create_dir(dir_name,
3709 vcpu->kvm->debugfs_dentry);
3711 kvm_arch_create_vcpu_debugfs(vcpu, debugfs_dentry);
3716 * Creates some virtual cpus. Good luck creating more than one.
3718 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
3721 struct kvm_vcpu *vcpu;
3724 if (id >= KVM_MAX_VCPU_IDS)
3727 mutex_lock(&kvm->lock);
3728 if (kvm->created_vcpus == KVM_MAX_VCPUS) {
3729 mutex_unlock(&kvm->lock);
3733 kvm->created_vcpus++;
3734 mutex_unlock(&kvm->lock);
3736 r = kvm_arch_vcpu_precreate(kvm, id);
3738 goto vcpu_decrement;
3740 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL_ACCOUNT);
3743 goto vcpu_decrement;
3746 BUILD_BUG_ON(sizeof(struct kvm_run) > PAGE_SIZE);
3747 page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
3752 vcpu->run = page_address(page);
3754 kvm_vcpu_init(vcpu, kvm, id);
3756 r = kvm_arch_vcpu_create(vcpu);
3758 goto vcpu_free_run_page;
3760 if (kvm->dirty_ring_size) {
3761 r = kvm_dirty_ring_alloc(&vcpu->dirty_ring,
3762 id, kvm->dirty_ring_size);
3764 goto arch_vcpu_destroy;
3767 mutex_lock(&kvm->lock);
3768 if (kvm_get_vcpu_by_id(kvm, id)) {
3770 goto unlock_vcpu_destroy;
3773 vcpu->vcpu_idx = atomic_read(&kvm->online_vcpus);
3774 r = xa_insert(&kvm->vcpu_array, vcpu->vcpu_idx, vcpu, GFP_KERNEL_ACCOUNT);
3775 BUG_ON(r == -EBUSY);
3777 goto unlock_vcpu_destroy;
3779 /* Fill the stats id string for the vcpu */
3780 snprintf(vcpu->stats_id, sizeof(vcpu->stats_id), "kvm-%d/vcpu-%d",
3781 task_pid_nr(current), id);
3783 /* Now it's all set up, let userspace reach it */
3785 r = create_vcpu_fd(vcpu);
3787 xa_erase(&kvm->vcpu_array, vcpu->vcpu_idx);
3788 kvm_put_kvm_no_destroy(kvm);
3789 goto unlock_vcpu_destroy;
3793 * Pairs with smp_rmb() in kvm_get_vcpu. Store the vcpu
3794 * pointer before kvm->online_vcpu's incremented value.
3797 atomic_inc(&kvm->online_vcpus);
3799 mutex_unlock(&kvm->lock);
3800 kvm_arch_vcpu_postcreate(vcpu);
3801 kvm_create_vcpu_debugfs(vcpu);
3804 unlock_vcpu_destroy:
3805 mutex_unlock(&kvm->lock);
3806 kvm_dirty_ring_free(&vcpu->dirty_ring);
3808 kvm_arch_vcpu_destroy(vcpu);
3810 free_page((unsigned long)vcpu->run);
3812 kmem_cache_free(kvm_vcpu_cache, vcpu);
3814 mutex_lock(&kvm->lock);
3815 kvm->created_vcpus--;
3816 mutex_unlock(&kvm->lock);
3820 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
3823 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
3824 vcpu->sigset_active = 1;
3825 vcpu->sigset = *sigset;
3827 vcpu->sigset_active = 0;
3831 static ssize_t kvm_vcpu_stats_read(struct file *file, char __user *user_buffer,
3832 size_t size, loff_t *offset)
3834 struct kvm_vcpu *vcpu = file->private_data;
3836 return kvm_stats_read(vcpu->stats_id, &kvm_vcpu_stats_header,
3837 &kvm_vcpu_stats_desc[0], &vcpu->stat,
3838 sizeof(vcpu->stat), user_buffer, size, offset);
3841 static const struct file_operations kvm_vcpu_stats_fops = {
3842 .read = kvm_vcpu_stats_read,
3843 .llseek = noop_llseek,
3846 static int kvm_vcpu_ioctl_get_stats_fd(struct kvm_vcpu *vcpu)
3850 char name[15 + ITOA_MAX_LEN + 1];
3852 snprintf(name, sizeof(name), "kvm-vcpu-stats:%d", vcpu->vcpu_id);
3854 fd = get_unused_fd_flags(O_CLOEXEC);
3858 file = anon_inode_getfile(name, &kvm_vcpu_stats_fops, vcpu, O_RDONLY);
3861 return PTR_ERR(file);
3863 file->f_mode |= FMODE_PREAD;
3864 fd_install(fd, file);
3869 static long kvm_vcpu_ioctl(struct file *filp,
3870 unsigned int ioctl, unsigned long arg)
3872 struct kvm_vcpu *vcpu = filp->private_data;
3873 void __user *argp = (void __user *)arg;
3875 struct kvm_fpu *fpu = NULL;
3876 struct kvm_sregs *kvm_sregs = NULL;
3878 if (vcpu->kvm->mm != current->mm || vcpu->kvm->vm_dead)
3881 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
3885 * Some architectures have vcpu ioctls that are asynchronous to vcpu
3886 * execution; mutex_lock() would break them.
3888 r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg);
3889 if (r != -ENOIOCTLCMD)
3892 if (mutex_lock_killable(&vcpu->mutex))
3900 oldpid = rcu_access_pointer(vcpu->pid);
3901 if (unlikely(oldpid != task_pid(current))) {
3902 /* The thread running this VCPU changed. */
3905 r = kvm_arch_vcpu_run_pid_change(vcpu);
3909 newpid = get_task_pid(current, PIDTYPE_PID);
3910 rcu_assign_pointer(vcpu->pid, newpid);
3915 r = kvm_arch_vcpu_ioctl_run(vcpu);
3916 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
3919 case KVM_GET_REGS: {
3920 struct kvm_regs *kvm_regs;
3923 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL_ACCOUNT);
3926 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
3930 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
3937 case KVM_SET_REGS: {
3938 struct kvm_regs *kvm_regs;
3940 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
3941 if (IS_ERR(kvm_regs)) {
3942 r = PTR_ERR(kvm_regs);
3945 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
3949 case KVM_GET_SREGS: {
3950 kvm_sregs = kzalloc(sizeof(struct kvm_sregs),
3951 GFP_KERNEL_ACCOUNT);
3955 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
3959 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
3964 case KVM_SET_SREGS: {
3965 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
3966 if (IS_ERR(kvm_sregs)) {
3967 r = PTR_ERR(kvm_sregs);
3971 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
3974 case KVM_GET_MP_STATE: {
3975 struct kvm_mp_state mp_state;
3977 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
3981 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
3986 case KVM_SET_MP_STATE: {
3987 struct kvm_mp_state mp_state;
3990 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
3992 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
3995 case KVM_TRANSLATE: {
3996 struct kvm_translation tr;
3999 if (copy_from_user(&tr, argp, sizeof(tr)))
4001 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
4005 if (copy_to_user(argp, &tr, sizeof(tr)))
4010 case KVM_SET_GUEST_DEBUG: {
4011 struct kvm_guest_debug dbg;
4014 if (copy_from_user(&dbg, argp, sizeof(dbg)))
4016 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
4019 case KVM_SET_SIGNAL_MASK: {
4020 struct kvm_signal_mask __user *sigmask_arg = argp;
4021 struct kvm_signal_mask kvm_sigmask;
4022 sigset_t sigset, *p;
4027 if (copy_from_user(&kvm_sigmask, argp,
4028 sizeof(kvm_sigmask)))
4031 if (kvm_sigmask.len != sizeof(sigset))
4034 if (copy_from_user(&sigset, sigmask_arg->sigset,
4039 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
4043 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL_ACCOUNT);
4047 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
4051 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
4057 fpu = memdup_user(argp, sizeof(*fpu));
4063 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
4066 case KVM_GET_STATS_FD: {
4067 r = kvm_vcpu_ioctl_get_stats_fd(vcpu);
4071 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
4074 mutex_unlock(&vcpu->mutex);
4080 #ifdef CONFIG_KVM_COMPAT
4081 static long kvm_vcpu_compat_ioctl(struct file *filp,
4082 unsigned int ioctl, unsigned long arg)
4084 struct kvm_vcpu *vcpu = filp->private_data;
4085 void __user *argp = compat_ptr(arg);
4088 if (vcpu->kvm->mm != current->mm || vcpu->kvm->vm_dead)
4092 case KVM_SET_SIGNAL_MASK: {
4093 struct kvm_signal_mask __user *sigmask_arg = argp;
4094 struct kvm_signal_mask kvm_sigmask;
4099 if (copy_from_user(&kvm_sigmask, argp,
4100 sizeof(kvm_sigmask)))
4103 if (kvm_sigmask.len != sizeof(compat_sigset_t))
4106 if (get_compat_sigset(&sigset,
4107 (compat_sigset_t __user *)sigmask_arg->sigset))
4109 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
4111 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
4115 r = kvm_vcpu_ioctl(filp, ioctl, arg);
4123 static int kvm_device_mmap(struct file *filp, struct vm_area_struct *vma)
4125 struct kvm_device *dev = filp->private_data;
4128 return dev->ops->mmap(dev, vma);
4133 static int kvm_device_ioctl_attr(struct kvm_device *dev,
4134 int (*accessor)(struct kvm_device *dev,
4135 struct kvm_device_attr *attr),
4138 struct kvm_device_attr attr;
4143 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
4146 return accessor(dev, &attr);
4149 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
4152 struct kvm_device *dev = filp->private_data;
4154 if (dev->kvm->mm != current->mm || dev->kvm->vm_dead)
4158 case KVM_SET_DEVICE_ATTR:
4159 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
4160 case KVM_GET_DEVICE_ATTR:
4161 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
4162 case KVM_HAS_DEVICE_ATTR:
4163 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
4165 if (dev->ops->ioctl)
4166 return dev->ops->ioctl(dev, ioctl, arg);
4172 static int kvm_device_release(struct inode *inode, struct file *filp)
4174 struct kvm_device *dev = filp->private_data;
4175 struct kvm *kvm = dev->kvm;
4177 if (dev->ops->release) {
4178 mutex_lock(&kvm->lock);
4179 list_del(&dev->vm_node);
4180 dev->ops->release(dev);
4181 mutex_unlock(&kvm->lock);
4188 static const struct file_operations kvm_device_fops = {
4189 .unlocked_ioctl = kvm_device_ioctl,
4190 .release = kvm_device_release,
4191 KVM_COMPAT(kvm_device_ioctl),
4192 .mmap = kvm_device_mmap,
4195 struct kvm_device *kvm_device_from_filp(struct file *filp)
4197 if (filp->f_op != &kvm_device_fops)
4200 return filp->private_data;
4203 static const struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
4204 #ifdef CONFIG_KVM_MPIC
4205 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
4206 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
4210 int kvm_register_device_ops(const struct kvm_device_ops *ops, u32 type)
4212 if (type >= ARRAY_SIZE(kvm_device_ops_table))
4215 if (kvm_device_ops_table[type] != NULL)
4218 kvm_device_ops_table[type] = ops;
4222 void kvm_unregister_device_ops(u32 type)
4224 if (kvm_device_ops_table[type] != NULL)
4225 kvm_device_ops_table[type] = NULL;
4228 static int kvm_ioctl_create_device(struct kvm *kvm,
4229 struct kvm_create_device *cd)
4231 const struct kvm_device_ops *ops = NULL;
4232 struct kvm_device *dev;
4233 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
4237 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
4240 type = array_index_nospec(cd->type, ARRAY_SIZE(kvm_device_ops_table));
4241 ops = kvm_device_ops_table[type];
4248 dev = kzalloc(sizeof(*dev), GFP_KERNEL_ACCOUNT);
4255 mutex_lock(&kvm->lock);
4256 ret = ops->create(dev, type);
4258 mutex_unlock(&kvm->lock);
4262 list_add(&dev->vm_node, &kvm->devices);
4263 mutex_unlock(&kvm->lock);
4269 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
4271 kvm_put_kvm_no_destroy(kvm);
4272 mutex_lock(&kvm->lock);
4273 list_del(&dev->vm_node);
4274 mutex_unlock(&kvm->lock);
4283 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
4286 case KVM_CAP_USER_MEMORY:
4287 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
4288 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
4289 case KVM_CAP_INTERNAL_ERROR_DATA:
4290 #ifdef CONFIG_HAVE_KVM_MSI
4291 case KVM_CAP_SIGNAL_MSI:
4293 #ifdef CONFIG_HAVE_KVM_IRQFD
4295 case KVM_CAP_IRQFD_RESAMPLE:
4297 case KVM_CAP_IOEVENTFD_ANY_LENGTH:
4298 case KVM_CAP_CHECK_EXTENSION_VM:
4299 case KVM_CAP_ENABLE_CAP_VM:
4300 case KVM_CAP_HALT_POLL:
4302 #ifdef CONFIG_KVM_MMIO
4303 case KVM_CAP_COALESCED_MMIO:
4304 return KVM_COALESCED_MMIO_PAGE_OFFSET;
4305 case KVM_CAP_COALESCED_PIO:
4308 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4309 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2:
4310 return KVM_DIRTY_LOG_MANUAL_CAPS;
4312 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
4313 case KVM_CAP_IRQ_ROUTING:
4314 return KVM_MAX_IRQ_ROUTES;
4316 #if KVM_ADDRESS_SPACE_NUM > 1
4317 case KVM_CAP_MULTI_ADDRESS_SPACE:
4318 return KVM_ADDRESS_SPACE_NUM;
4320 case KVM_CAP_NR_MEMSLOTS:
4321 return KVM_USER_MEM_SLOTS;
4322 case KVM_CAP_DIRTY_LOG_RING:
4323 #ifdef CONFIG_HAVE_KVM_DIRTY_RING
4324 return KVM_DIRTY_RING_MAX_ENTRIES * sizeof(struct kvm_dirty_gfn);
4328 case KVM_CAP_BINARY_STATS_FD:
4333 return kvm_vm_ioctl_check_extension(kvm, arg);
4336 static int kvm_vm_ioctl_enable_dirty_log_ring(struct kvm *kvm, u32 size)
4340 if (!KVM_DIRTY_LOG_PAGE_OFFSET)
4343 /* the size should be power of 2 */
4344 if (!size || (size & (size - 1)))
4347 /* Should be bigger to keep the reserved entries, or a page */
4348 if (size < kvm_dirty_ring_get_rsvd_entries() *
4349 sizeof(struct kvm_dirty_gfn) || size < PAGE_SIZE)
4352 if (size > KVM_DIRTY_RING_MAX_ENTRIES *
4353 sizeof(struct kvm_dirty_gfn))
4356 /* We only allow it to set once */
4357 if (kvm->dirty_ring_size)
4360 mutex_lock(&kvm->lock);
4362 if (kvm->created_vcpus) {
4363 /* We don't allow to change this value after vcpu created */
4366 kvm->dirty_ring_size = size;
4370 mutex_unlock(&kvm->lock);
4374 static int kvm_vm_ioctl_reset_dirty_pages(struct kvm *kvm)
4377 struct kvm_vcpu *vcpu;
4380 if (!kvm->dirty_ring_size)
4383 mutex_lock(&kvm->slots_lock);
4385 kvm_for_each_vcpu(i, vcpu, kvm)
4386 cleared += kvm_dirty_ring_reset(vcpu->kvm, &vcpu->dirty_ring);
4388 mutex_unlock(&kvm->slots_lock);
4391 kvm_flush_remote_tlbs(kvm);
4396 int __attribute__((weak)) kvm_vm_ioctl_enable_cap(struct kvm *kvm,
4397 struct kvm_enable_cap *cap)
4402 static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm,
4403 struct kvm_enable_cap *cap)
4406 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4407 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2: {
4408 u64 allowed_options = KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE;
4410 if (cap->args[0] & KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE)
4411 allowed_options = KVM_DIRTY_LOG_MANUAL_CAPS;
4413 if (cap->flags || (cap->args[0] & ~allowed_options))
4415 kvm->manual_dirty_log_protect = cap->args[0];
4419 case KVM_CAP_HALT_POLL: {
4420 if (cap->flags || cap->args[0] != (unsigned int)cap->args[0])
4423 kvm->max_halt_poll_ns = cap->args[0];
4426 case KVM_CAP_DIRTY_LOG_RING:
4427 return kvm_vm_ioctl_enable_dirty_log_ring(kvm, cap->args[0]);
4429 return kvm_vm_ioctl_enable_cap(kvm, cap);
4433 static ssize_t kvm_vm_stats_read(struct file *file, char __user *user_buffer,
4434 size_t size, loff_t *offset)
4436 struct kvm *kvm = file->private_data;
4438 return kvm_stats_read(kvm->stats_id, &kvm_vm_stats_header,
4439 &kvm_vm_stats_desc[0], &kvm->stat,
4440 sizeof(kvm->stat), user_buffer, size, offset);
4443 static const struct file_operations kvm_vm_stats_fops = {
4444 .read = kvm_vm_stats_read,
4445 .llseek = noop_llseek,
4448 static int kvm_vm_ioctl_get_stats_fd(struct kvm *kvm)
4453 fd = get_unused_fd_flags(O_CLOEXEC);
4457 file = anon_inode_getfile("kvm-vm-stats",
4458 &kvm_vm_stats_fops, kvm, O_RDONLY);
4461 return PTR_ERR(file);
4463 file->f_mode |= FMODE_PREAD;
4464 fd_install(fd, file);
4469 static long kvm_vm_ioctl(struct file *filp,
4470 unsigned int ioctl, unsigned long arg)
4472 struct kvm *kvm = filp->private_data;
4473 void __user *argp = (void __user *)arg;
4476 if (kvm->mm != current->mm || kvm->vm_dead)
4479 case KVM_CREATE_VCPU:
4480 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
4482 case KVM_ENABLE_CAP: {
4483 struct kvm_enable_cap cap;
4486 if (copy_from_user(&cap, argp, sizeof(cap)))
4488 r = kvm_vm_ioctl_enable_cap_generic(kvm, &cap);
4491 case KVM_SET_USER_MEMORY_REGION: {
4492 struct kvm_userspace_memory_region kvm_userspace_mem;
4495 if (copy_from_user(&kvm_userspace_mem, argp,
4496 sizeof(kvm_userspace_mem)))
4499 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
4502 case KVM_GET_DIRTY_LOG: {
4503 struct kvm_dirty_log log;
4506 if (copy_from_user(&log, argp, sizeof(log)))
4508 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
4511 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4512 case KVM_CLEAR_DIRTY_LOG: {
4513 struct kvm_clear_dirty_log log;
4516 if (copy_from_user(&log, argp, sizeof(log)))
4518 r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
4522 #ifdef CONFIG_KVM_MMIO
4523 case KVM_REGISTER_COALESCED_MMIO: {
4524 struct kvm_coalesced_mmio_zone zone;
4527 if (copy_from_user(&zone, argp, sizeof(zone)))
4529 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
4532 case KVM_UNREGISTER_COALESCED_MMIO: {
4533 struct kvm_coalesced_mmio_zone zone;
4536 if (copy_from_user(&zone, argp, sizeof(zone)))
4538 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
4543 struct kvm_irqfd data;
4546 if (copy_from_user(&data, argp, sizeof(data)))
4548 r = kvm_irqfd(kvm, &data);
4551 case KVM_IOEVENTFD: {
4552 struct kvm_ioeventfd data;
4555 if (copy_from_user(&data, argp, sizeof(data)))
4557 r = kvm_ioeventfd(kvm, &data);
4560 #ifdef CONFIG_HAVE_KVM_MSI
4561 case KVM_SIGNAL_MSI: {
4565 if (copy_from_user(&msi, argp, sizeof(msi)))
4567 r = kvm_send_userspace_msi(kvm, &msi);
4571 #ifdef __KVM_HAVE_IRQ_LINE
4572 case KVM_IRQ_LINE_STATUS:
4573 case KVM_IRQ_LINE: {
4574 struct kvm_irq_level irq_event;
4577 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
4580 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
4581 ioctl == KVM_IRQ_LINE_STATUS);
4586 if (ioctl == KVM_IRQ_LINE_STATUS) {
4587 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
4595 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
4596 case KVM_SET_GSI_ROUTING: {
4597 struct kvm_irq_routing routing;
4598 struct kvm_irq_routing __user *urouting;
4599 struct kvm_irq_routing_entry *entries = NULL;
4602 if (copy_from_user(&routing, argp, sizeof(routing)))
4605 if (!kvm_arch_can_set_irq_routing(kvm))
4607 if (routing.nr > KVM_MAX_IRQ_ROUTES)
4613 entries = vmemdup_user(urouting->entries,
4614 array_size(sizeof(*entries),
4616 if (IS_ERR(entries)) {
4617 r = PTR_ERR(entries);
4621 r = kvm_set_irq_routing(kvm, entries, routing.nr,
4626 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
4627 case KVM_CREATE_DEVICE: {
4628 struct kvm_create_device cd;
4631 if (copy_from_user(&cd, argp, sizeof(cd)))
4634 r = kvm_ioctl_create_device(kvm, &cd);
4639 if (copy_to_user(argp, &cd, sizeof(cd)))
4645 case KVM_CHECK_EXTENSION:
4646 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
4648 case KVM_RESET_DIRTY_RINGS:
4649 r = kvm_vm_ioctl_reset_dirty_pages(kvm);
4651 case KVM_GET_STATS_FD:
4652 r = kvm_vm_ioctl_get_stats_fd(kvm);
4655 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
4661 #ifdef CONFIG_KVM_COMPAT
4662 struct compat_kvm_dirty_log {
4666 compat_uptr_t dirty_bitmap; /* one bit per page */
4671 struct compat_kvm_clear_dirty_log {
4676 compat_uptr_t dirty_bitmap; /* one bit per page */
4681 static long kvm_vm_compat_ioctl(struct file *filp,
4682 unsigned int ioctl, unsigned long arg)
4684 struct kvm *kvm = filp->private_data;
4687 if (kvm->mm != current->mm || kvm->vm_dead)
4690 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4691 case KVM_CLEAR_DIRTY_LOG: {
4692 struct compat_kvm_clear_dirty_log compat_log;
4693 struct kvm_clear_dirty_log log;
4695 if (copy_from_user(&compat_log, (void __user *)arg,
4696 sizeof(compat_log)))
4698 log.slot = compat_log.slot;
4699 log.num_pages = compat_log.num_pages;
4700 log.first_page = compat_log.first_page;
4701 log.padding2 = compat_log.padding2;
4702 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
4704 r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
4708 case KVM_GET_DIRTY_LOG: {
4709 struct compat_kvm_dirty_log compat_log;
4710 struct kvm_dirty_log log;
4712 if (copy_from_user(&compat_log, (void __user *)arg,
4713 sizeof(compat_log)))
4715 log.slot = compat_log.slot;
4716 log.padding1 = compat_log.padding1;
4717 log.padding2 = compat_log.padding2;
4718 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
4720 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
4724 r = kvm_vm_ioctl(filp, ioctl, arg);
4730 static struct file_operations kvm_vm_fops = {
4731 .release = kvm_vm_release,
4732 .unlocked_ioctl = kvm_vm_ioctl,
4733 .llseek = noop_llseek,
4734 KVM_COMPAT(kvm_vm_compat_ioctl),
4737 bool file_is_kvm(struct file *file)
4739 return file && file->f_op == &kvm_vm_fops;
4741 EXPORT_SYMBOL_GPL(file_is_kvm);
4743 static int kvm_dev_ioctl_create_vm(unsigned long type)
4749 kvm = kvm_create_vm(type);
4751 return PTR_ERR(kvm);
4752 #ifdef CONFIG_KVM_MMIO
4753 r = kvm_coalesced_mmio_init(kvm);
4757 r = get_unused_fd_flags(O_CLOEXEC);
4761 snprintf(kvm->stats_id, sizeof(kvm->stats_id),
4762 "kvm-%d", task_pid_nr(current));
4764 file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
4772 * Don't call kvm_put_kvm anymore at this point; file->f_op is
4773 * already set, with ->release() being kvm_vm_release(). In error
4774 * cases it will be called by the final fput(file) and will take
4775 * care of doing kvm_put_kvm(kvm).
4777 if (kvm_create_vm_debugfs(kvm, r) < 0) {
4782 kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm);
4784 fd_install(r, file);
4792 static long kvm_dev_ioctl(struct file *filp,
4793 unsigned int ioctl, unsigned long arg)
4798 case KVM_GET_API_VERSION:
4801 r = KVM_API_VERSION;
4804 r = kvm_dev_ioctl_create_vm(arg);
4806 case KVM_CHECK_EXTENSION:
4807 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
4809 case KVM_GET_VCPU_MMAP_SIZE:
4812 r = PAGE_SIZE; /* struct kvm_run */
4814 r += PAGE_SIZE; /* pio data page */
4816 #ifdef CONFIG_KVM_MMIO
4817 r += PAGE_SIZE; /* coalesced mmio ring page */
4820 case KVM_TRACE_ENABLE:
4821 case KVM_TRACE_PAUSE:
4822 case KVM_TRACE_DISABLE:
4826 return kvm_arch_dev_ioctl(filp, ioctl, arg);
4832 static struct file_operations kvm_chardev_ops = {
4833 .unlocked_ioctl = kvm_dev_ioctl,
4834 .llseek = noop_llseek,
4835 KVM_COMPAT(kvm_dev_ioctl),
4838 static struct miscdevice kvm_dev = {
4844 static void hardware_enable_nolock(void *junk)
4846 int cpu = raw_smp_processor_id();
4849 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
4852 cpumask_set_cpu(cpu, cpus_hardware_enabled);
4854 r = kvm_arch_hardware_enable();
4857 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
4858 atomic_inc(&hardware_enable_failed);
4859 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
4863 static int kvm_starting_cpu(unsigned int cpu)
4865 raw_spin_lock(&kvm_count_lock);
4866 if (kvm_usage_count)
4867 hardware_enable_nolock(NULL);
4868 raw_spin_unlock(&kvm_count_lock);
4872 static void hardware_disable_nolock(void *junk)
4874 int cpu = raw_smp_processor_id();
4876 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
4878 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
4879 kvm_arch_hardware_disable();
4882 static int kvm_dying_cpu(unsigned int cpu)
4884 raw_spin_lock(&kvm_count_lock);
4885 if (kvm_usage_count)
4886 hardware_disable_nolock(NULL);
4887 raw_spin_unlock(&kvm_count_lock);
4891 static void hardware_disable_all_nolock(void)
4893 BUG_ON(!kvm_usage_count);
4896 if (!kvm_usage_count)
4897 on_each_cpu(hardware_disable_nolock, NULL, 1);
4900 static void hardware_disable_all(void)
4902 raw_spin_lock(&kvm_count_lock);
4903 hardware_disable_all_nolock();
4904 raw_spin_unlock(&kvm_count_lock);
4907 static int hardware_enable_all(void)
4911 raw_spin_lock(&kvm_count_lock);
4914 if (kvm_usage_count == 1) {
4915 atomic_set(&hardware_enable_failed, 0);
4916 on_each_cpu(hardware_enable_nolock, NULL, 1);
4918 if (atomic_read(&hardware_enable_failed)) {
4919 hardware_disable_all_nolock();
4924 raw_spin_unlock(&kvm_count_lock);
4929 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
4933 * Some (well, at least mine) BIOSes hang on reboot if
4936 * And Intel TXT required VMX off for all cpu when system shutdown.
4938 pr_info("kvm: exiting hardware virtualization\n");
4939 kvm_rebooting = true;
4940 on_each_cpu(hardware_disable_nolock, NULL, 1);
4944 static struct notifier_block kvm_reboot_notifier = {
4945 .notifier_call = kvm_reboot,
4949 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
4953 for (i = 0; i < bus->dev_count; i++) {
4954 struct kvm_io_device *pos = bus->range[i].dev;
4956 kvm_iodevice_destructor(pos);
4961 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
4962 const struct kvm_io_range *r2)
4964 gpa_t addr1 = r1->addr;
4965 gpa_t addr2 = r2->addr;
4970 /* If r2->len == 0, match the exact address. If r2->len != 0,
4971 * accept any overlapping write. Any order is acceptable for
4972 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
4973 * we process all of them.
4986 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
4988 return kvm_io_bus_cmp(p1, p2);
4991 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
4992 gpa_t addr, int len)
4994 struct kvm_io_range *range, key;
4997 key = (struct kvm_io_range) {
5002 range = bsearch(&key, bus->range, bus->dev_count,
5003 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
5007 off = range - bus->range;
5009 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
5015 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
5016 struct kvm_io_range *range, const void *val)
5020 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
5024 while (idx < bus->dev_count &&
5025 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
5026 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
5035 /* kvm_io_bus_write - called under kvm->slots_lock */
5036 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
5037 int len, const void *val)
5039 struct kvm_io_bus *bus;
5040 struct kvm_io_range range;
5043 range = (struct kvm_io_range) {
5048 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
5051 r = __kvm_io_bus_write(vcpu, bus, &range, val);
5052 return r < 0 ? r : 0;
5054 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
5056 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
5057 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
5058 gpa_t addr, int len, const void *val, long cookie)
5060 struct kvm_io_bus *bus;
5061 struct kvm_io_range range;
5063 range = (struct kvm_io_range) {
5068 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
5072 /* First try the device referenced by cookie. */
5073 if ((cookie >= 0) && (cookie < bus->dev_count) &&
5074 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
5075 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
5080 * cookie contained garbage; fall back to search and return the
5081 * correct cookie value.
5083 return __kvm_io_bus_write(vcpu, bus, &range, val);
5086 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
5087 struct kvm_io_range *range, void *val)
5091 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
5095 while (idx < bus->dev_count &&
5096 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
5097 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
5106 /* kvm_io_bus_read - called under kvm->slots_lock */
5107 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
5110 struct kvm_io_bus *bus;
5111 struct kvm_io_range range;
5114 range = (struct kvm_io_range) {
5119 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
5122 r = __kvm_io_bus_read(vcpu, bus, &range, val);
5123 return r < 0 ? r : 0;
5126 /* Caller must hold slots_lock. */
5127 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
5128 int len, struct kvm_io_device *dev)
5131 struct kvm_io_bus *new_bus, *bus;
5132 struct kvm_io_range range;
5134 bus = kvm_get_bus(kvm, bus_idx);
5138 /* exclude ioeventfd which is limited by maximum fd */
5139 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
5142 new_bus = kmalloc(struct_size(bus, range, bus->dev_count + 1),
5143 GFP_KERNEL_ACCOUNT);
5147 range = (struct kvm_io_range) {
5153 for (i = 0; i < bus->dev_count; i++)
5154 if (kvm_io_bus_cmp(&bus->range[i], &range) > 0)
5157 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
5158 new_bus->dev_count++;
5159 new_bus->range[i] = range;
5160 memcpy(new_bus->range + i + 1, bus->range + i,
5161 (bus->dev_count - i) * sizeof(struct kvm_io_range));
5162 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
5163 synchronize_srcu_expedited(&kvm->srcu);
5169 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
5170 struct kvm_io_device *dev)
5173 struct kvm_io_bus *new_bus, *bus;
5175 lockdep_assert_held(&kvm->slots_lock);
5177 bus = kvm_get_bus(kvm, bus_idx);
5181 for (i = 0; i < bus->dev_count; i++) {
5182 if (bus->range[i].dev == dev) {
5187 if (i == bus->dev_count)
5190 new_bus = kmalloc(struct_size(bus, range, bus->dev_count - 1),
5191 GFP_KERNEL_ACCOUNT);
5193 memcpy(new_bus, bus, struct_size(bus, range, i));
5194 new_bus->dev_count--;
5195 memcpy(new_bus->range + i, bus->range + i + 1,
5196 flex_array_size(new_bus, range, new_bus->dev_count - i));
5199 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
5200 synchronize_srcu_expedited(&kvm->srcu);
5202 /* Destroy the old bus _after_ installing the (null) bus. */
5204 pr_err("kvm: failed to shrink bus, removing it completely\n");
5205 for (j = 0; j < bus->dev_count; j++) {
5208 kvm_iodevice_destructor(bus->range[j].dev);
5213 return new_bus ? 0 : -ENOMEM;
5216 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
5219 struct kvm_io_bus *bus;
5220 int dev_idx, srcu_idx;
5221 struct kvm_io_device *iodev = NULL;
5223 srcu_idx = srcu_read_lock(&kvm->srcu);
5225 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
5229 dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
5233 iodev = bus->range[dev_idx].dev;
5236 srcu_read_unlock(&kvm->srcu, srcu_idx);
5240 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
5242 static int kvm_debugfs_open(struct inode *inode, struct file *file,
5243 int (*get)(void *, u64 *), int (*set)(void *, u64),
5246 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
5250 * The debugfs files are a reference to the kvm struct which
5251 * is still valid when kvm_destroy_vm is called. kvm_get_kvm_safe
5252 * avoids the race between open and the removal of the debugfs directory.
5254 if (!kvm_get_kvm_safe(stat_data->kvm))
5257 if (simple_attr_open(inode, file, get,
5258 kvm_stats_debugfs_mode(stat_data->desc) & 0222
5261 kvm_put_kvm(stat_data->kvm);
5268 static int kvm_debugfs_release(struct inode *inode, struct file *file)
5270 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
5273 simple_attr_release(inode, file);
5274 kvm_put_kvm(stat_data->kvm);
5279 static int kvm_get_stat_per_vm(struct kvm *kvm, size_t offset, u64 *val)
5281 *val = *(u64 *)((void *)(&kvm->stat) + offset);
5286 static int kvm_clear_stat_per_vm(struct kvm *kvm, size_t offset)
5288 *(u64 *)((void *)(&kvm->stat) + offset) = 0;
5293 static int kvm_get_stat_per_vcpu(struct kvm *kvm, size_t offset, u64 *val)
5296 struct kvm_vcpu *vcpu;
5300 kvm_for_each_vcpu(i, vcpu, kvm)
5301 *val += *(u64 *)((void *)(&vcpu->stat) + offset);
5306 static int kvm_clear_stat_per_vcpu(struct kvm *kvm, size_t offset)
5309 struct kvm_vcpu *vcpu;
5311 kvm_for_each_vcpu(i, vcpu, kvm)
5312 *(u64 *)((void *)(&vcpu->stat) + offset) = 0;
5317 static int kvm_stat_data_get(void *data, u64 *val)
5320 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
5322 switch (stat_data->kind) {
5324 r = kvm_get_stat_per_vm(stat_data->kvm,
5325 stat_data->desc->desc.offset, val);
5328 r = kvm_get_stat_per_vcpu(stat_data->kvm,
5329 stat_data->desc->desc.offset, val);
5336 static int kvm_stat_data_clear(void *data, u64 val)
5339 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
5344 switch (stat_data->kind) {
5346 r = kvm_clear_stat_per_vm(stat_data->kvm,
5347 stat_data->desc->desc.offset);
5350 r = kvm_clear_stat_per_vcpu(stat_data->kvm,
5351 stat_data->desc->desc.offset);
5358 static int kvm_stat_data_open(struct inode *inode, struct file *file)
5360 __simple_attr_check_format("%llu\n", 0ull);
5361 return kvm_debugfs_open(inode, file, kvm_stat_data_get,
5362 kvm_stat_data_clear, "%llu\n");
5365 static const struct file_operations stat_fops_per_vm = {
5366 .owner = THIS_MODULE,
5367 .open = kvm_stat_data_open,
5368 .release = kvm_debugfs_release,
5369 .read = simple_attr_read,
5370 .write = simple_attr_write,
5371 .llseek = no_llseek,
5374 static int vm_stat_get(void *_offset, u64 *val)
5376 unsigned offset = (long)_offset;
5381 mutex_lock(&kvm_lock);
5382 list_for_each_entry(kvm, &vm_list, vm_list) {
5383 kvm_get_stat_per_vm(kvm, offset, &tmp_val);
5386 mutex_unlock(&kvm_lock);
5390 static int vm_stat_clear(void *_offset, u64 val)
5392 unsigned offset = (long)_offset;
5398 mutex_lock(&kvm_lock);
5399 list_for_each_entry(kvm, &vm_list, vm_list) {
5400 kvm_clear_stat_per_vm(kvm, offset);
5402 mutex_unlock(&kvm_lock);
5407 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n");
5408 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_readonly_fops, vm_stat_get, NULL, "%llu\n");
5410 static int vcpu_stat_get(void *_offset, u64 *val)
5412 unsigned offset = (long)_offset;
5417 mutex_lock(&kvm_lock);
5418 list_for_each_entry(kvm, &vm_list, vm_list) {
5419 kvm_get_stat_per_vcpu(kvm, offset, &tmp_val);
5422 mutex_unlock(&kvm_lock);
5426 static int vcpu_stat_clear(void *_offset, u64 val)
5428 unsigned offset = (long)_offset;
5434 mutex_lock(&kvm_lock);
5435 list_for_each_entry(kvm, &vm_list, vm_list) {
5436 kvm_clear_stat_per_vcpu(kvm, offset);
5438 mutex_unlock(&kvm_lock);
5443 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear,
5445 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_readonly_fops, vcpu_stat_get, NULL, "%llu\n");
5447 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm)
5449 struct kobj_uevent_env *env;
5450 unsigned long long created, active;
5452 if (!kvm_dev.this_device || !kvm)
5455 mutex_lock(&kvm_lock);
5456 if (type == KVM_EVENT_CREATE_VM) {
5457 kvm_createvm_count++;
5459 } else if (type == KVM_EVENT_DESTROY_VM) {
5462 created = kvm_createvm_count;
5463 active = kvm_active_vms;
5464 mutex_unlock(&kvm_lock);
5466 env = kzalloc(sizeof(*env), GFP_KERNEL_ACCOUNT);
5470 add_uevent_var(env, "CREATED=%llu", created);
5471 add_uevent_var(env, "COUNT=%llu", active);
5473 if (type == KVM_EVENT_CREATE_VM) {
5474 add_uevent_var(env, "EVENT=create");
5475 kvm->userspace_pid = task_pid_nr(current);
5476 } else if (type == KVM_EVENT_DESTROY_VM) {
5477 add_uevent_var(env, "EVENT=destroy");
5479 add_uevent_var(env, "PID=%d", kvm->userspace_pid);
5481 if (kvm->debugfs_dentry) {
5482 char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL_ACCOUNT);
5485 tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX);
5487 add_uevent_var(env, "STATS_PATH=%s", tmp);
5491 /* no need for checks, since we are adding at most only 5 keys */
5492 env->envp[env->envp_idx++] = NULL;
5493 kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp);
5497 static void kvm_init_debug(void)
5499 const struct file_operations *fops;
5500 const struct _kvm_stats_desc *pdesc;
5503 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
5505 for (i = 0; i < kvm_vm_stats_header.num_desc; ++i) {
5506 pdesc = &kvm_vm_stats_desc[i];
5507 if (kvm_stats_debugfs_mode(pdesc) & 0222)
5508 fops = &vm_stat_fops;
5510 fops = &vm_stat_readonly_fops;
5511 debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
5513 (void *)(long)pdesc->desc.offset, fops);
5516 for (i = 0; i < kvm_vcpu_stats_header.num_desc; ++i) {
5517 pdesc = &kvm_vcpu_stats_desc[i];
5518 if (kvm_stats_debugfs_mode(pdesc) & 0222)
5519 fops = &vcpu_stat_fops;
5521 fops = &vcpu_stat_readonly_fops;
5522 debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
5524 (void *)(long)pdesc->desc.offset, fops);
5528 static int kvm_suspend(void)
5530 if (kvm_usage_count)
5531 hardware_disable_nolock(NULL);
5535 static void kvm_resume(void)
5537 if (kvm_usage_count) {
5538 lockdep_assert_not_held(&kvm_count_lock);
5539 hardware_enable_nolock(NULL);
5543 static struct syscore_ops kvm_syscore_ops = {
5544 .suspend = kvm_suspend,
5545 .resume = kvm_resume,
5549 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
5551 return container_of(pn, struct kvm_vcpu, preempt_notifier);
5554 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
5556 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
5558 WRITE_ONCE(vcpu->preempted, false);
5559 WRITE_ONCE(vcpu->ready, false);
5561 __this_cpu_write(kvm_running_vcpu, vcpu);
5562 kvm_arch_sched_in(vcpu, cpu);
5563 kvm_arch_vcpu_load(vcpu, cpu);
5566 static void kvm_sched_out(struct preempt_notifier *pn,
5567 struct task_struct *next)
5569 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
5571 if (current->on_rq) {
5572 WRITE_ONCE(vcpu->preempted, true);
5573 WRITE_ONCE(vcpu->ready, true);
5575 kvm_arch_vcpu_put(vcpu);
5576 __this_cpu_write(kvm_running_vcpu, NULL);
5580 * kvm_get_running_vcpu - get the vcpu running on the current CPU.
5582 * We can disable preemption locally around accessing the per-CPU variable,
5583 * and use the resolved vcpu pointer after enabling preemption again,
5584 * because even if the current thread is migrated to another CPU, reading
5585 * the per-CPU value later will give us the same value as we update the
5586 * per-CPU variable in the preempt notifier handlers.
5588 struct kvm_vcpu *kvm_get_running_vcpu(void)
5590 struct kvm_vcpu *vcpu;
5593 vcpu = __this_cpu_read(kvm_running_vcpu);
5598 EXPORT_SYMBOL_GPL(kvm_get_running_vcpu);
5601 * kvm_get_running_vcpus - get the per-CPU array of currently running vcpus.
5603 struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void)
5605 return &kvm_running_vcpu;
5608 #ifdef CONFIG_GUEST_PERF_EVENTS
5609 static unsigned int kvm_guest_state(void)
5611 struct kvm_vcpu *vcpu = kvm_get_running_vcpu();
5614 if (!kvm_arch_pmi_in_guest(vcpu))
5617 state = PERF_GUEST_ACTIVE;
5618 if (!kvm_arch_vcpu_in_kernel(vcpu))
5619 state |= PERF_GUEST_USER;
5624 static unsigned long kvm_guest_get_ip(void)
5626 struct kvm_vcpu *vcpu = kvm_get_running_vcpu();
5628 /* Retrieving the IP must be guarded by a call to kvm_guest_state(). */
5629 if (WARN_ON_ONCE(!kvm_arch_pmi_in_guest(vcpu)))
5632 return kvm_arch_vcpu_get_ip(vcpu);
5635 static struct perf_guest_info_callbacks kvm_guest_cbs = {
5636 .state = kvm_guest_state,
5637 .get_ip = kvm_guest_get_ip,
5638 .handle_intel_pt_intr = NULL,
5641 void kvm_register_perf_callbacks(unsigned int (*pt_intr_handler)(void))
5643 kvm_guest_cbs.handle_intel_pt_intr = pt_intr_handler;
5644 perf_register_guest_info_callbacks(&kvm_guest_cbs);
5646 void kvm_unregister_perf_callbacks(void)
5648 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
5652 struct kvm_cpu_compat_check {
5657 static void check_processor_compat(void *data)
5659 struct kvm_cpu_compat_check *c = data;
5661 *c->ret = kvm_arch_check_processor_compat(c->opaque);
5664 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
5665 struct module *module)
5667 struct kvm_cpu_compat_check c;
5671 r = kvm_arch_init(opaque);
5676 * kvm_arch_init makes sure there's at most one caller
5677 * for architectures that support multiple implementations,
5678 * like intel and amd on x86.
5679 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
5680 * conflicts in case kvm is already setup for another implementation.
5682 r = kvm_irqfd_init();
5686 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
5691 r = kvm_arch_hardware_setup(opaque);
5697 for_each_online_cpu(cpu) {
5698 smp_call_function_single(cpu, check_processor_compat, &c, 1);
5703 r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting",
5704 kvm_starting_cpu, kvm_dying_cpu);
5707 register_reboot_notifier(&kvm_reboot_notifier);
5709 /* A kmem cache lets us meet the alignment requirements of fx_save. */
5711 vcpu_align = __alignof__(struct kvm_vcpu);
5713 kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align,
5715 offsetof(struct kvm_vcpu, arch),
5716 offsetofend(struct kvm_vcpu, stats_id)
5717 - offsetof(struct kvm_vcpu, arch),
5719 if (!kvm_vcpu_cache) {
5724 for_each_possible_cpu(cpu) {
5725 if (!alloc_cpumask_var_node(&per_cpu(cpu_kick_mask, cpu),
5726 GFP_KERNEL, cpu_to_node(cpu))) {
5732 r = kvm_async_pf_init();
5736 kvm_chardev_ops.owner = module;
5737 kvm_vm_fops.owner = module;
5738 kvm_vcpu_fops.owner = module;
5740 r = misc_register(&kvm_dev);
5742 pr_err("kvm: misc device register failed\n");
5746 register_syscore_ops(&kvm_syscore_ops);
5748 kvm_preempt_ops.sched_in = kvm_sched_in;
5749 kvm_preempt_ops.sched_out = kvm_sched_out;
5753 r = kvm_vfio_ops_init();
5759 kvm_async_pf_deinit();
5761 for_each_possible_cpu(cpu)
5762 free_cpumask_var(per_cpu(cpu_kick_mask, cpu));
5764 kmem_cache_destroy(kvm_vcpu_cache);
5766 unregister_reboot_notifier(&kvm_reboot_notifier);
5767 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
5769 kvm_arch_hardware_unsetup();
5771 free_cpumask_var(cpus_hardware_enabled);
5779 EXPORT_SYMBOL_GPL(kvm_init);
5785 debugfs_remove_recursive(kvm_debugfs_dir);
5786 misc_deregister(&kvm_dev);
5787 for_each_possible_cpu(cpu)
5788 free_cpumask_var(per_cpu(cpu_kick_mask, cpu));
5789 kmem_cache_destroy(kvm_vcpu_cache);
5790 kvm_async_pf_deinit();
5791 unregister_syscore_ops(&kvm_syscore_ops);
5792 unregister_reboot_notifier(&kvm_reboot_notifier);
5793 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
5794 on_each_cpu(hardware_disable_nolock, NULL, 1);
5795 kvm_arch_hardware_unsetup();
5798 free_cpumask_var(cpus_hardware_enabled);
5799 kvm_vfio_ops_exit();
5801 EXPORT_SYMBOL_GPL(kvm_exit);
5803 struct kvm_vm_worker_thread_context {
5805 struct task_struct *parent;
5806 struct completion init_done;
5807 kvm_vm_thread_fn_t thread_fn;
5812 static int kvm_vm_worker_thread(void *context)
5815 * The init_context is allocated on the stack of the parent thread, so
5816 * we have to locally copy anything that is needed beyond initialization
5818 struct kvm_vm_worker_thread_context *init_context = context;
5819 struct task_struct *parent;
5820 struct kvm *kvm = init_context->kvm;
5821 kvm_vm_thread_fn_t thread_fn = init_context->thread_fn;
5822 uintptr_t data = init_context->data;
5825 err = kthread_park(current);
5826 /* kthread_park(current) is never supposed to return an error */
5831 err = cgroup_attach_task_all(init_context->parent, current);
5833 kvm_err("%s: cgroup_attach_task_all failed with err %d\n",
5838 set_user_nice(current, task_nice(init_context->parent));
5841 init_context->err = err;
5842 complete(&init_context->init_done);
5843 init_context = NULL;
5848 /* Wait to be woken up by the spawner before proceeding. */
5851 if (!kthread_should_stop())
5852 err = thread_fn(kvm, data);
5856 * Move kthread back to its original cgroup to prevent it lingering in
5857 * the cgroup of the VM process, after the latter finishes its
5860 * kthread_stop() waits on the 'exited' completion condition which is
5861 * set in exit_mm(), via mm_release(), in do_exit(). However, the
5862 * kthread is removed from the cgroup in the cgroup_exit() which is
5863 * called after the exit_mm(). This causes the kthread_stop() to return
5864 * before the kthread actually quits the cgroup.
5867 parent = rcu_dereference(current->real_parent);
5868 get_task_struct(parent);
5870 cgroup_attach_task_all(parent, current);
5871 put_task_struct(parent);
5876 int kvm_vm_create_worker_thread(struct kvm *kvm, kvm_vm_thread_fn_t thread_fn,
5877 uintptr_t data, const char *name,
5878 struct task_struct **thread_ptr)
5880 struct kvm_vm_worker_thread_context init_context = {};
5881 struct task_struct *thread;
5884 init_context.kvm = kvm;
5885 init_context.parent = current;
5886 init_context.thread_fn = thread_fn;
5887 init_context.data = data;
5888 init_completion(&init_context.init_done);
5890 thread = kthread_run(kvm_vm_worker_thread, &init_context,
5891 "%s-%d", name, task_pid_nr(current));
5893 return PTR_ERR(thread);
5895 /* kthread_run is never supposed to return NULL */
5896 WARN_ON(thread == NULL);
5898 wait_for_completion(&init_context.init_done);
5900 if (!init_context.err)
5901 *thread_ptr = thread;
5903 return init_context.err;