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 __weak void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
161 unsigned long start, unsigned long end)
165 bool kvm_is_zone_device_pfn(kvm_pfn_t pfn)
168 * The metadata used by is_zone_device_page() to determine whether or
169 * not a page is ZONE_DEVICE is guaranteed to be valid if and only if
170 * the device has been pinned, e.g. by get_user_pages(). WARN if the
171 * page_count() is zero to help detect bad usage of this helper.
173 if (!pfn_valid(pfn) || WARN_ON_ONCE(!page_count(pfn_to_page(pfn))))
176 return is_zone_device_page(pfn_to_page(pfn));
179 bool kvm_is_reserved_pfn(kvm_pfn_t pfn)
182 * ZONE_DEVICE pages currently set PG_reserved, but from a refcounting
183 * perspective they are "normal" pages, albeit with slightly different
187 return PageReserved(pfn_to_page(pfn)) &&
189 !kvm_is_zone_device_pfn(pfn);
195 * Switches to specified vcpu, until a matching vcpu_put()
197 void vcpu_load(struct kvm_vcpu *vcpu)
201 __this_cpu_write(kvm_running_vcpu, vcpu);
202 preempt_notifier_register(&vcpu->preempt_notifier);
203 kvm_arch_vcpu_load(vcpu, cpu);
206 EXPORT_SYMBOL_GPL(vcpu_load);
208 void vcpu_put(struct kvm_vcpu *vcpu)
211 kvm_arch_vcpu_put(vcpu);
212 preempt_notifier_unregister(&vcpu->preempt_notifier);
213 __this_cpu_write(kvm_running_vcpu, NULL);
216 EXPORT_SYMBOL_GPL(vcpu_put);
218 /* TODO: merge with kvm_arch_vcpu_should_kick */
219 static bool kvm_request_needs_ipi(struct kvm_vcpu *vcpu, unsigned req)
221 int mode = kvm_vcpu_exiting_guest_mode(vcpu);
224 * We need to wait for the VCPU to reenable interrupts and get out of
225 * READING_SHADOW_PAGE_TABLES mode.
227 if (req & KVM_REQUEST_WAIT)
228 return mode != OUTSIDE_GUEST_MODE;
231 * Need to kick a running VCPU, but otherwise there is nothing to do.
233 return mode == IN_GUEST_MODE;
236 static void ack_flush(void *_completed)
240 static inline bool kvm_kick_many_cpus(cpumask_var_t tmp, bool wait)
242 const struct cpumask *cpus;
244 if (likely(cpumask_available(tmp)))
247 cpus = cpu_online_mask;
249 if (cpumask_empty(cpus))
252 smp_call_function_many(cpus, ack_flush, NULL, wait);
256 bool kvm_make_vcpus_request_mask(struct kvm *kvm, unsigned int req,
257 struct kvm_vcpu *except,
258 unsigned long *vcpu_bitmap, cpumask_var_t tmp)
261 struct kvm_vcpu *vcpu;
266 kvm_for_each_vcpu(i, vcpu, kvm) {
267 if ((vcpu_bitmap && !test_bit(i, vcpu_bitmap)) ||
271 kvm_make_request(req, vcpu);
273 if (!(req & KVM_REQUEST_NO_WAKEUP) && kvm_vcpu_wake_up(vcpu))
277 * tmp can be "unavailable" if cpumasks are allocated off stack
278 * as allocation of the mask is deliberately not fatal and is
279 * handled by falling back to kicking all online CPUs.
281 if (!cpumask_available(tmp))
285 * Note, the vCPU could get migrated to a different pCPU at any
286 * point after kvm_request_needs_ipi(), which could result in
287 * sending an IPI to the previous pCPU. But, that's ok because
288 * the purpose of the IPI is to ensure the vCPU returns to
289 * OUTSIDE_GUEST_MODE, which is satisfied if the vCPU migrates.
290 * Entering READING_SHADOW_PAGE_TABLES after this point is also
291 * ok, as the requirement is only that KVM wait for vCPUs that
292 * were reading SPTEs _before_ any changes were finalized. See
293 * kvm_vcpu_kick() for more details on handling requests.
295 if (kvm_request_needs_ipi(vcpu, req)) {
296 cpu = READ_ONCE(vcpu->cpu);
297 if (cpu != -1 && cpu != me)
298 __cpumask_set_cpu(cpu, tmp);
302 called = kvm_kick_many_cpus(tmp, !!(req & KVM_REQUEST_WAIT));
308 bool kvm_make_all_cpus_request_except(struct kvm *kvm, unsigned int req,
309 struct kvm_vcpu *except)
314 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
316 called = kvm_make_vcpus_request_mask(kvm, req, except, NULL, cpus);
318 free_cpumask_var(cpus);
322 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
324 return kvm_make_all_cpus_request_except(kvm, req, NULL);
326 EXPORT_SYMBOL_GPL(kvm_make_all_cpus_request);
328 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
329 void kvm_flush_remote_tlbs(struct kvm *kvm)
331 ++kvm->stat.generic.remote_tlb_flush_requests;
334 * We want to publish modifications to the page tables before reading
335 * mode. Pairs with a memory barrier in arch-specific code.
336 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
337 * and smp_mb in walk_shadow_page_lockless_begin/end.
338 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
340 * There is already an smp_mb__after_atomic() before
341 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
344 if (!kvm_arch_flush_remote_tlb(kvm)
345 || kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
346 ++kvm->stat.generic.remote_tlb_flush;
348 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
351 void kvm_reload_remote_mmus(struct kvm *kvm)
353 kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
356 #ifdef KVM_ARCH_NR_OBJS_PER_MEMORY_CACHE
357 static inline void *mmu_memory_cache_alloc_obj(struct kvm_mmu_memory_cache *mc,
360 gfp_flags |= mc->gfp_zero;
363 return kmem_cache_alloc(mc->kmem_cache, gfp_flags);
365 return (void *)__get_free_page(gfp_flags);
368 int kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache *mc, int min)
372 if (mc->nobjs >= min)
374 while (mc->nobjs < ARRAY_SIZE(mc->objects)) {
375 obj = mmu_memory_cache_alloc_obj(mc, GFP_KERNEL_ACCOUNT);
377 return mc->nobjs >= min ? 0 : -ENOMEM;
378 mc->objects[mc->nobjs++] = obj;
383 int kvm_mmu_memory_cache_nr_free_objects(struct kvm_mmu_memory_cache *mc)
388 void kvm_mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc)
392 kmem_cache_free(mc->kmem_cache, mc->objects[--mc->nobjs]);
394 free_page((unsigned long)mc->objects[--mc->nobjs]);
398 void *kvm_mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc)
402 if (WARN_ON(!mc->nobjs))
403 p = mmu_memory_cache_alloc_obj(mc, GFP_ATOMIC | __GFP_ACCOUNT);
405 p = mc->objects[--mc->nobjs];
411 static void kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
413 mutex_init(&vcpu->mutex);
418 rcuwait_init(&vcpu->wait);
419 kvm_async_pf_vcpu_init(vcpu);
422 INIT_LIST_HEAD(&vcpu->blocked_vcpu_list);
424 kvm_vcpu_set_in_spin_loop(vcpu, false);
425 kvm_vcpu_set_dy_eligible(vcpu, false);
426 vcpu->preempted = false;
428 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
429 vcpu->last_used_slot = 0;
432 void kvm_vcpu_destroy(struct kvm_vcpu *vcpu)
434 kvm_arch_vcpu_destroy(vcpu);
435 kvm_dirty_ring_free(&vcpu->dirty_ring);
438 * No need for rcu_read_lock as VCPU_RUN is the only place that changes
439 * the vcpu->pid pointer, and at destruction time all file descriptors
442 put_pid(rcu_dereference_protected(vcpu->pid, 1));
444 free_page((unsigned long)vcpu->run);
445 kmem_cache_free(kvm_vcpu_cache, vcpu);
447 EXPORT_SYMBOL_GPL(kvm_vcpu_destroy);
449 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
450 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
452 return container_of(mn, struct kvm, mmu_notifier);
455 static void kvm_mmu_notifier_invalidate_range(struct mmu_notifier *mn,
456 struct mm_struct *mm,
457 unsigned long start, unsigned long end)
459 struct kvm *kvm = mmu_notifier_to_kvm(mn);
462 idx = srcu_read_lock(&kvm->srcu);
463 kvm_arch_mmu_notifier_invalidate_range(kvm, start, end);
464 srcu_read_unlock(&kvm->srcu, idx);
467 typedef bool (*hva_handler_t)(struct kvm *kvm, struct kvm_gfn_range *range);
469 typedef void (*on_lock_fn_t)(struct kvm *kvm, unsigned long start,
472 struct kvm_hva_range {
476 hva_handler_t handler;
477 on_lock_fn_t on_lock;
483 * Use a dedicated stub instead of NULL to indicate that there is no callback
484 * function/handler. The compiler technically can't guarantee that a real
485 * function will have a non-zero address, and so it will generate code to
486 * check for !NULL, whereas comparing against a stub will be elided at compile
487 * time (unless the compiler is getting long in the tooth, e.g. gcc 4.9).
489 static void kvm_null_fn(void)
493 #define IS_KVM_NULL_FN(fn) ((fn) == (void *)kvm_null_fn)
495 static __always_inline int __kvm_handle_hva_range(struct kvm *kvm,
496 const struct kvm_hva_range *range)
498 bool ret = false, locked = false;
499 struct kvm_gfn_range gfn_range;
500 struct kvm_memory_slot *slot;
501 struct kvm_memslots *slots;
504 /* A null handler is allowed if and only if on_lock() is provided. */
505 if (WARN_ON_ONCE(IS_KVM_NULL_FN(range->on_lock) &&
506 IS_KVM_NULL_FN(range->handler)))
509 idx = srcu_read_lock(&kvm->srcu);
511 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
512 slots = __kvm_memslots(kvm, i);
513 kvm_for_each_memslot(slot, slots) {
514 unsigned long hva_start, hva_end;
516 hva_start = max(range->start, slot->userspace_addr);
517 hva_end = min(range->end, slot->userspace_addr +
518 (slot->npages << PAGE_SHIFT));
519 if (hva_start >= hva_end)
523 * To optimize for the likely case where the address
524 * range is covered by zero or one memslots, don't
525 * bother making these conditional (to avoid writes on
526 * the second or later invocation of the handler).
528 gfn_range.pte = range->pte;
529 gfn_range.may_block = range->may_block;
532 * {gfn(page) | page intersects with [hva_start, hva_end)} =
533 * {gfn_start, gfn_start+1, ..., gfn_end-1}.
535 gfn_range.start = hva_to_gfn_memslot(hva_start, slot);
536 gfn_range.end = hva_to_gfn_memslot(hva_end + PAGE_SIZE - 1, slot);
537 gfn_range.slot = slot;
542 if (!IS_KVM_NULL_FN(range->on_lock))
543 range->on_lock(kvm, range->start, range->end);
544 if (IS_KVM_NULL_FN(range->handler))
547 ret |= range->handler(kvm, &gfn_range);
551 if (range->flush_on_ret && ret)
552 kvm_flush_remote_tlbs(kvm);
557 srcu_read_unlock(&kvm->srcu, idx);
559 /* The notifiers are averse to booleans. :-( */
563 static __always_inline int kvm_handle_hva_range(struct mmu_notifier *mn,
567 hva_handler_t handler)
569 struct kvm *kvm = mmu_notifier_to_kvm(mn);
570 const struct kvm_hva_range range = {
575 .on_lock = (void *)kvm_null_fn,
576 .flush_on_ret = true,
580 return __kvm_handle_hva_range(kvm, &range);
583 static __always_inline int kvm_handle_hva_range_no_flush(struct mmu_notifier *mn,
586 hva_handler_t handler)
588 struct kvm *kvm = mmu_notifier_to_kvm(mn);
589 const struct kvm_hva_range range = {
594 .on_lock = (void *)kvm_null_fn,
595 .flush_on_ret = false,
599 return __kvm_handle_hva_range(kvm, &range);
601 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
602 struct mm_struct *mm,
603 unsigned long address,
606 struct kvm *kvm = mmu_notifier_to_kvm(mn);
608 trace_kvm_set_spte_hva(address);
611 * .change_pte() must be surrounded by .invalidate_range_{start,end}().
612 * If mmu_notifier_count is zero, then no in-progress invalidations,
613 * including this one, found a relevant memslot at start(); rechecking
614 * memslots here is unnecessary. Note, a false positive (count elevated
615 * by a different invalidation) is sub-optimal but functionally ok.
617 WARN_ON_ONCE(!READ_ONCE(kvm->mn_active_invalidate_count));
618 if (!READ_ONCE(kvm->mmu_notifier_count))
621 kvm_handle_hva_range(mn, address, address + 1, pte, kvm_set_spte_gfn);
624 void kvm_inc_notifier_count(struct kvm *kvm, unsigned long start,
628 * The count increase must become visible at unlock time as no
629 * spte can be established without taking the mmu_lock and
630 * count is also read inside the mmu_lock critical section.
632 kvm->mmu_notifier_count++;
633 if (likely(kvm->mmu_notifier_count == 1)) {
634 kvm->mmu_notifier_range_start = start;
635 kvm->mmu_notifier_range_end = end;
638 * Fully tracking multiple concurrent ranges has dimishing
639 * returns. Keep things simple and just find the minimal range
640 * which includes the current and new ranges. As there won't be
641 * enough information to subtract a range after its invalidate
642 * completes, any ranges invalidated concurrently will
643 * accumulate and persist until all outstanding invalidates
646 kvm->mmu_notifier_range_start =
647 min(kvm->mmu_notifier_range_start, start);
648 kvm->mmu_notifier_range_end =
649 max(kvm->mmu_notifier_range_end, end);
653 static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
654 const struct mmu_notifier_range *range)
656 struct kvm *kvm = mmu_notifier_to_kvm(mn);
657 const struct kvm_hva_range hva_range = {
658 .start = range->start,
661 .handler = kvm_unmap_gfn_range,
662 .on_lock = kvm_inc_notifier_count,
663 .flush_on_ret = true,
664 .may_block = mmu_notifier_range_blockable(range),
667 trace_kvm_unmap_hva_range(range->start, range->end);
670 * Prevent memslot modification between range_start() and range_end()
671 * so that conditionally locking provides the same result in both
672 * functions. Without that guarantee, the mmu_notifier_count
673 * adjustments will be imbalanced.
675 * Pairs with the decrement in range_end().
677 spin_lock(&kvm->mn_invalidate_lock);
678 kvm->mn_active_invalidate_count++;
679 spin_unlock(&kvm->mn_invalidate_lock);
681 __kvm_handle_hva_range(kvm, &hva_range);
686 void kvm_dec_notifier_count(struct kvm *kvm, unsigned long start,
690 * This sequence increase will notify the kvm page fault that
691 * the page that is going to be mapped in the spte could have
694 kvm->mmu_notifier_seq++;
697 * The above sequence increase must be visible before the
698 * below count decrease, which is ensured by the smp_wmb above
699 * in conjunction with the smp_rmb in mmu_notifier_retry().
701 kvm->mmu_notifier_count--;
704 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
705 const struct mmu_notifier_range *range)
707 struct kvm *kvm = mmu_notifier_to_kvm(mn);
708 const struct kvm_hva_range hva_range = {
709 .start = range->start,
712 .handler = (void *)kvm_null_fn,
713 .on_lock = kvm_dec_notifier_count,
714 .flush_on_ret = false,
715 .may_block = mmu_notifier_range_blockable(range),
719 __kvm_handle_hva_range(kvm, &hva_range);
721 /* Pairs with the increment in range_start(). */
722 spin_lock(&kvm->mn_invalidate_lock);
723 wake = (--kvm->mn_active_invalidate_count == 0);
724 spin_unlock(&kvm->mn_invalidate_lock);
727 * There can only be one waiter, since the wait happens under
731 rcuwait_wake_up(&kvm->mn_memslots_update_rcuwait);
733 BUG_ON(kvm->mmu_notifier_count < 0);
736 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
737 struct mm_struct *mm,
741 trace_kvm_age_hva(start, end);
743 return kvm_handle_hva_range(mn, start, end, __pte(0), kvm_age_gfn);
746 static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
747 struct mm_struct *mm,
751 trace_kvm_age_hva(start, end);
754 * Even though we do not flush TLB, this will still adversely
755 * affect performance on pre-Haswell Intel EPT, where there is
756 * no EPT Access Bit to clear so that we have to tear down EPT
757 * tables instead. If we find this unacceptable, we can always
758 * add a parameter to kvm_age_hva so that it effectively doesn't
759 * do anything on clear_young.
761 * Also note that currently we never issue secondary TLB flushes
762 * from clear_young, leaving this job up to the regular system
763 * cadence. If we find this inaccurate, we might come up with a
764 * more sophisticated heuristic later.
766 return kvm_handle_hva_range_no_flush(mn, start, end, kvm_age_gfn);
769 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
770 struct mm_struct *mm,
771 unsigned long address)
773 trace_kvm_test_age_hva(address);
775 return kvm_handle_hva_range_no_flush(mn, address, address + 1,
779 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
780 struct mm_struct *mm)
782 struct kvm *kvm = mmu_notifier_to_kvm(mn);
785 idx = srcu_read_lock(&kvm->srcu);
786 kvm_arch_flush_shadow_all(kvm);
787 srcu_read_unlock(&kvm->srcu, idx);
790 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
791 .invalidate_range = kvm_mmu_notifier_invalidate_range,
792 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
793 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
794 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
795 .clear_young = kvm_mmu_notifier_clear_young,
796 .test_young = kvm_mmu_notifier_test_young,
797 .change_pte = kvm_mmu_notifier_change_pte,
798 .release = kvm_mmu_notifier_release,
801 static int kvm_init_mmu_notifier(struct kvm *kvm)
803 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
804 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
807 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
809 static int kvm_init_mmu_notifier(struct kvm *kvm)
814 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
816 #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER
817 static int kvm_pm_notifier_call(struct notifier_block *bl,
821 struct kvm *kvm = container_of(bl, struct kvm, pm_notifier);
823 return kvm_arch_pm_notifier(kvm, state);
826 static void kvm_init_pm_notifier(struct kvm *kvm)
828 kvm->pm_notifier.notifier_call = kvm_pm_notifier_call;
829 /* Suspend KVM before we suspend ftrace, RCU, etc. */
830 kvm->pm_notifier.priority = INT_MAX;
831 register_pm_notifier(&kvm->pm_notifier);
834 static void kvm_destroy_pm_notifier(struct kvm *kvm)
836 unregister_pm_notifier(&kvm->pm_notifier);
838 #else /* !CONFIG_HAVE_KVM_PM_NOTIFIER */
839 static void kvm_init_pm_notifier(struct kvm *kvm)
843 static void kvm_destroy_pm_notifier(struct kvm *kvm)
846 #endif /* CONFIG_HAVE_KVM_PM_NOTIFIER */
848 static struct kvm_memslots *kvm_alloc_memslots(void)
851 struct kvm_memslots *slots;
853 slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL_ACCOUNT);
857 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
858 slots->id_to_index[i] = -1;
863 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
865 if (!memslot->dirty_bitmap)
868 kvfree(memslot->dirty_bitmap);
869 memslot->dirty_bitmap = NULL;
872 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
874 kvm_destroy_dirty_bitmap(slot);
876 kvm_arch_free_memslot(kvm, slot);
882 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
884 struct kvm_memory_slot *memslot;
889 kvm_for_each_memslot(memslot, slots)
890 kvm_free_memslot(kvm, memslot);
895 static umode_t kvm_stats_debugfs_mode(const struct _kvm_stats_desc *pdesc)
897 switch (pdesc->desc.flags & KVM_STATS_TYPE_MASK) {
898 case KVM_STATS_TYPE_INSTANT:
900 case KVM_STATS_TYPE_CUMULATIVE:
901 case KVM_STATS_TYPE_PEAK:
908 static void kvm_destroy_vm_debugfs(struct kvm *kvm)
911 int kvm_debugfs_num_entries = kvm_vm_stats_header.num_desc +
912 kvm_vcpu_stats_header.num_desc;
914 if (IS_ERR(kvm->debugfs_dentry))
917 debugfs_remove_recursive(kvm->debugfs_dentry);
919 if (kvm->debugfs_stat_data) {
920 for (i = 0; i < kvm_debugfs_num_entries; i++)
921 kfree(kvm->debugfs_stat_data[i]);
922 kfree(kvm->debugfs_stat_data);
926 static int kvm_create_vm_debugfs(struct kvm *kvm, int fd)
928 static DEFINE_MUTEX(kvm_debugfs_lock);
930 char dir_name[ITOA_MAX_LEN * 2];
931 struct kvm_stat_data *stat_data;
932 const struct _kvm_stats_desc *pdesc;
934 int kvm_debugfs_num_entries = kvm_vm_stats_header.num_desc +
935 kvm_vcpu_stats_header.num_desc;
937 if (!debugfs_initialized())
940 snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd);
941 mutex_lock(&kvm_debugfs_lock);
942 dent = debugfs_lookup(dir_name, kvm_debugfs_dir);
944 pr_warn_ratelimited("KVM: debugfs: duplicate directory %s\n", dir_name);
946 mutex_unlock(&kvm_debugfs_lock);
949 dent = debugfs_create_dir(dir_name, kvm_debugfs_dir);
950 mutex_unlock(&kvm_debugfs_lock);
954 kvm->debugfs_dentry = dent;
955 kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries,
956 sizeof(*kvm->debugfs_stat_data),
958 if (!kvm->debugfs_stat_data)
961 for (i = 0; i < kvm_vm_stats_header.num_desc; ++i) {
962 pdesc = &kvm_vm_stats_desc[i];
963 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT);
967 stat_data->kvm = kvm;
968 stat_data->desc = pdesc;
969 stat_data->kind = KVM_STAT_VM;
970 kvm->debugfs_stat_data[i] = stat_data;
971 debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
972 kvm->debugfs_dentry, stat_data,
976 for (i = 0; i < kvm_vcpu_stats_header.num_desc; ++i) {
977 pdesc = &kvm_vcpu_stats_desc[i];
978 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT);
982 stat_data->kvm = kvm;
983 stat_data->desc = pdesc;
984 stat_data->kind = KVM_STAT_VCPU;
985 kvm->debugfs_stat_data[i + kvm_vm_stats_header.num_desc] = stat_data;
986 debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
987 kvm->debugfs_dentry, stat_data,
991 ret = kvm_arch_create_vm_debugfs(kvm);
993 kvm_destroy_vm_debugfs(kvm);
1001 * Called after the VM is otherwise initialized, but just before adding it to
1004 int __weak kvm_arch_post_init_vm(struct kvm *kvm)
1010 * Called just after removing the VM from the vm_list, but before doing any
1011 * other destruction.
1013 void __weak kvm_arch_pre_destroy_vm(struct kvm *kvm)
1018 * Called after per-vm debugfs created. When called kvm->debugfs_dentry should
1019 * be setup already, so we can create arch-specific debugfs entries under it.
1020 * Cleanup should be automatic done in kvm_destroy_vm_debugfs() recursively, so
1021 * a per-arch destroy interface is not needed.
1023 int __weak kvm_arch_create_vm_debugfs(struct kvm *kvm)
1028 static struct kvm *kvm_create_vm(unsigned long type)
1030 struct kvm *kvm = kvm_arch_alloc_vm();
1035 return ERR_PTR(-ENOMEM);
1037 KVM_MMU_LOCK_INIT(kvm);
1038 mmgrab(current->mm);
1039 kvm->mm = current->mm;
1040 kvm_eventfd_init(kvm);
1041 mutex_init(&kvm->lock);
1042 mutex_init(&kvm->irq_lock);
1043 mutex_init(&kvm->slots_lock);
1044 mutex_init(&kvm->slots_arch_lock);
1045 spin_lock_init(&kvm->mn_invalidate_lock);
1046 rcuwait_init(&kvm->mn_memslots_update_rcuwait);
1048 INIT_LIST_HEAD(&kvm->devices);
1050 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
1053 * Force subsequent debugfs file creations to fail if the VM directory
1054 * is not created (by kvm_create_vm_debugfs()).
1056 kvm->debugfs_dentry = ERR_PTR(-ENOENT);
1058 if (init_srcu_struct(&kvm->srcu))
1059 goto out_err_no_srcu;
1060 if (init_srcu_struct(&kvm->irq_srcu))
1061 goto out_err_no_irq_srcu;
1063 refcount_set(&kvm->users_count, 1);
1064 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
1065 struct kvm_memslots *slots = kvm_alloc_memslots();
1068 goto out_err_no_arch_destroy_vm;
1069 /* Generations must be different for each address space. */
1070 slots->generation = i;
1071 rcu_assign_pointer(kvm->memslots[i], slots);
1074 for (i = 0; i < KVM_NR_BUSES; i++) {
1075 rcu_assign_pointer(kvm->buses[i],
1076 kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL_ACCOUNT));
1078 goto out_err_no_arch_destroy_vm;
1081 kvm->max_halt_poll_ns = halt_poll_ns;
1083 r = kvm_arch_init_vm(kvm, type);
1085 goto out_err_no_arch_destroy_vm;
1087 r = hardware_enable_all();
1089 goto out_err_no_disable;
1091 #ifdef CONFIG_HAVE_KVM_IRQFD
1092 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
1095 r = kvm_init_mmu_notifier(kvm);
1097 goto out_err_no_mmu_notifier;
1099 r = kvm_arch_post_init_vm(kvm);
1103 mutex_lock(&kvm_lock);
1104 list_add(&kvm->vm_list, &vm_list);
1105 mutex_unlock(&kvm_lock);
1107 preempt_notifier_inc();
1108 kvm_init_pm_notifier(kvm);
1111 * When the fd passed to this ioctl() is opened it pins the module,
1112 * but try_module_get() also prevents getting a reference if the module
1113 * is in MODULE_STATE_GOING (e.g. if someone ran "rmmod --wait").
1115 if (!try_module_get(kvm_chardev_ops.owner)) {
1123 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1124 if (kvm->mmu_notifier.ops)
1125 mmu_notifier_unregister(&kvm->mmu_notifier, current->mm);
1127 out_err_no_mmu_notifier:
1128 hardware_disable_all();
1130 kvm_arch_destroy_vm(kvm);
1131 out_err_no_arch_destroy_vm:
1132 WARN_ON_ONCE(!refcount_dec_and_test(&kvm->users_count));
1133 for (i = 0; i < KVM_NR_BUSES; i++)
1134 kfree(kvm_get_bus(kvm, i));
1135 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
1136 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
1137 cleanup_srcu_struct(&kvm->irq_srcu);
1138 out_err_no_irq_srcu:
1139 cleanup_srcu_struct(&kvm->srcu);
1141 kvm_arch_free_vm(kvm);
1142 mmdrop(current->mm);
1146 static void kvm_destroy_devices(struct kvm *kvm)
1148 struct kvm_device *dev, *tmp;
1151 * We do not need to take the kvm->lock here, because nobody else
1152 * has a reference to the struct kvm at this point and therefore
1153 * cannot access the devices list anyhow.
1155 list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
1156 list_del(&dev->vm_node);
1157 dev->ops->destroy(dev);
1161 static void kvm_destroy_vm(struct kvm *kvm)
1164 struct mm_struct *mm = kvm->mm;
1166 kvm_destroy_pm_notifier(kvm);
1167 kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm);
1168 kvm_destroy_vm_debugfs(kvm);
1169 kvm_arch_sync_events(kvm);
1170 mutex_lock(&kvm_lock);
1171 list_del(&kvm->vm_list);
1172 mutex_unlock(&kvm_lock);
1173 kvm_arch_pre_destroy_vm(kvm);
1175 kvm_free_irq_routing(kvm);
1176 for (i = 0; i < KVM_NR_BUSES; i++) {
1177 struct kvm_io_bus *bus = kvm_get_bus(kvm, i);
1180 kvm_io_bus_destroy(bus);
1181 kvm->buses[i] = NULL;
1183 kvm_coalesced_mmio_free(kvm);
1184 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1185 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
1187 * At this point, pending calls to invalidate_range_start()
1188 * have completed but no more MMU notifiers will run, so
1189 * mn_active_invalidate_count may remain unbalanced.
1190 * No threads can be waiting in install_new_memslots as the
1191 * last reference on KVM has been dropped, but freeing
1192 * memslots would deadlock without this manual intervention.
1194 WARN_ON(rcuwait_active(&kvm->mn_memslots_update_rcuwait));
1195 kvm->mn_active_invalidate_count = 0;
1197 kvm_arch_flush_shadow_all(kvm);
1199 kvm_arch_destroy_vm(kvm);
1200 kvm_destroy_devices(kvm);
1201 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
1202 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
1203 cleanup_srcu_struct(&kvm->irq_srcu);
1204 cleanup_srcu_struct(&kvm->srcu);
1205 kvm_arch_free_vm(kvm);
1206 preempt_notifier_dec();
1207 hardware_disable_all();
1209 module_put(kvm_chardev_ops.owner);
1212 void kvm_get_kvm(struct kvm *kvm)
1214 refcount_inc(&kvm->users_count);
1216 EXPORT_SYMBOL_GPL(kvm_get_kvm);
1219 * Make sure the vm is not during destruction, which is a safe version of
1220 * kvm_get_kvm(). Return true if kvm referenced successfully, false otherwise.
1222 bool kvm_get_kvm_safe(struct kvm *kvm)
1224 return refcount_inc_not_zero(&kvm->users_count);
1226 EXPORT_SYMBOL_GPL(kvm_get_kvm_safe);
1228 void kvm_put_kvm(struct kvm *kvm)
1230 if (refcount_dec_and_test(&kvm->users_count))
1231 kvm_destroy_vm(kvm);
1233 EXPORT_SYMBOL_GPL(kvm_put_kvm);
1236 * Used to put a reference that was taken on behalf of an object associated
1237 * with a user-visible file descriptor, e.g. a vcpu or device, if installation
1238 * of the new file descriptor fails and the reference cannot be transferred to
1239 * its final owner. In such cases, the caller is still actively using @kvm and
1240 * will fail miserably if the refcount unexpectedly hits zero.
1242 void kvm_put_kvm_no_destroy(struct kvm *kvm)
1244 WARN_ON(refcount_dec_and_test(&kvm->users_count));
1246 EXPORT_SYMBOL_GPL(kvm_put_kvm_no_destroy);
1248 static int kvm_vm_release(struct inode *inode, struct file *filp)
1250 struct kvm *kvm = filp->private_data;
1252 kvm_irqfd_release(kvm);
1259 * Allocation size is twice as large as the actual dirty bitmap size.
1260 * See kvm_vm_ioctl_get_dirty_log() why this is needed.
1262 static int kvm_alloc_dirty_bitmap(struct kvm_memory_slot *memslot)
1264 unsigned long dirty_bytes = kvm_dirty_bitmap_bytes(memslot);
1266 memslot->dirty_bitmap = __vcalloc(2, dirty_bytes, GFP_KERNEL_ACCOUNT);
1267 if (!memslot->dirty_bitmap)
1274 * Delete a memslot by decrementing the number of used slots and shifting all
1275 * other entries in the array forward one spot.
1277 static inline void kvm_memslot_delete(struct kvm_memslots *slots,
1278 struct kvm_memory_slot *memslot)
1280 struct kvm_memory_slot *mslots = slots->memslots;
1283 if (WARN_ON(slots->id_to_index[memslot->id] == -1))
1286 slots->used_slots--;
1288 if (atomic_read(&slots->last_used_slot) >= slots->used_slots)
1289 atomic_set(&slots->last_used_slot, 0);
1291 for (i = slots->id_to_index[memslot->id]; i < slots->used_slots; i++) {
1292 mslots[i] = mslots[i + 1];
1293 slots->id_to_index[mslots[i].id] = i;
1295 mslots[i] = *memslot;
1296 slots->id_to_index[memslot->id] = -1;
1300 * "Insert" a new memslot by incrementing the number of used slots. Returns
1301 * the new slot's initial index into the memslots array.
1303 static inline int kvm_memslot_insert_back(struct kvm_memslots *slots)
1305 return slots->used_slots++;
1309 * Move a changed memslot backwards in the array by shifting existing slots
1310 * with a higher GFN toward the front of the array. Note, the changed memslot
1311 * itself is not preserved in the array, i.e. not swapped at this time, only
1312 * its new index into the array is tracked. Returns the changed memslot's
1313 * current index into the memslots array.
1315 static inline int kvm_memslot_move_backward(struct kvm_memslots *slots,
1316 struct kvm_memory_slot *memslot)
1318 struct kvm_memory_slot *mslots = slots->memslots;
1321 if (WARN_ON_ONCE(slots->id_to_index[memslot->id] == -1) ||
1322 WARN_ON_ONCE(!slots->used_slots))
1326 * Move the target memslot backward in the array by shifting existing
1327 * memslots with a higher GFN (than the target memslot) towards the
1328 * front of the array.
1330 for (i = slots->id_to_index[memslot->id]; i < slots->used_slots - 1; i++) {
1331 if (memslot->base_gfn > mslots[i + 1].base_gfn)
1334 WARN_ON_ONCE(memslot->base_gfn == mslots[i + 1].base_gfn);
1336 /* Shift the next memslot forward one and update its index. */
1337 mslots[i] = mslots[i + 1];
1338 slots->id_to_index[mslots[i].id] = i;
1344 * Move a changed memslot forwards in the array by shifting existing slots with
1345 * a lower GFN toward the back of the array. Note, the changed memslot itself
1346 * is not preserved in the array, i.e. not swapped at this time, only its new
1347 * index into the array is tracked. Returns the changed memslot's final index
1348 * into the memslots array.
1350 static inline int kvm_memslot_move_forward(struct kvm_memslots *slots,
1351 struct kvm_memory_slot *memslot,
1354 struct kvm_memory_slot *mslots = slots->memslots;
1357 for (i = start; i > 0; i--) {
1358 if (memslot->base_gfn < mslots[i - 1].base_gfn)
1361 WARN_ON_ONCE(memslot->base_gfn == mslots[i - 1].base_gfn);
1363 /* Shift the next memslot back one and update its index. */
1364 mslots[i] = mslots[i - 1];
1365 slots->id_to_index[mslots[i].id] = i;
1371 * Re-sort memslots based on their GFN to account for an added, deleted, or
1372 * moved memslot. Sorting memslots by GFN allows using a binary search during
1375 * IMPORTANT: Slots are sorted from highest GFN to lowest GFN! I.e. the entry
1376 * at memslots[0] has the highest GFN.
1378 * The sorting algorithm takes advantage of having initially sorted memslots
1379 * and knowing the position of the changed memslot. Sorting is also optimized
1380 * by not swapping the updated memslot and instead only shifting other memslots
1381 * and tracking the new index for the update memslot. Only once its final
1382 * index is known is the updated memslot copied into its position in the array.
1384 * - When deleting a memslot, the deleted memslot simply needs to be moved to
1385 * the end of the array.
1387 * - When creating a memslot, the algorithm "inserts" the new memslot at the
1388 * end of the array and then it forward to its correct location.
1390 * - When moving a memslot, the algorithm first moves the updated memslot
1391 * backward to handle the scenario where the memslot's GFN was changed to a
1392 * lower value. update_memslots() then falls through and runs the same flow
1393 * as creating a memslot to move the memslot forward to handle the scenario
1394 * where its GFN was changed to a higher value.
1396 * Note, slots are sorted from highest->lowest instead of lowest->highest for
1397 * historical reasons. Originally, invalid memslots where denoted by having
1398 * GFN=0, thus sorting from highest->lowest naturally sorted invalid memslots
1399 * to the end of the array. The current algorithm uses dedicated logic to
1400 * delete a memslot and thus does not rely on invalid memslots having GFN=0.
1402 * The other historical motiviation for highest->lowest was to improve the
1403 * performance of memslot lookup. KVM originally used a linear search starting
1404 * at memslots[0]. On x86, the largest memslot usually has one of the highest,
1405 * if not *the* highest, GFN, as the bulk of the guest's RAM is located in a
1406 * single memslot above the 4gb boundary. As the largest memslot is also the
1407 * most likely to be referenced, sorting it to the front of the array was
1408 * advantageous. The current binary search starts from the middle of the array
1409 * and uses an LRU pointer to improve performance for all memslots and GFNs.
1411 static void update_memslots(struct kvm_memslots *slots,
1412 struct kvm_memory_slot *memslot,
1413 enum kvm_mr_change change)
1417 if (change == KVM_MR_DELETE) {
1418 kvm_memslot_delete(slots, memslot);
1420 if (change == KVM_MR_CREATE)
1421 i = kvm_memslot_insert_back(slots);
1423 i = kvm_memslot_move_backward(slots, memslot);
1424 i = kvm_memslot_move_forward(slots, memslot, i);
1427 * Copy the memslot to its new position in memslots and update
1428 * its index accordingly.
1430 slots->memslots[i] = *memslot;
1431 slots->id_to_index[memslot->id] = i;
1435 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
1437 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
1439 #ifdef __KVM_HAVE_READONLY_MEM
1440 valid_flags |= KVM_MEM_READONLY;
1443 if (mem->flags & ~valid_flags)
1449 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
1450 int as_id, struct kvm_memslots *slots)
1452 struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id);
1453 u64 gen = old_memslots->generation;
1455 WARN_ON(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS);
1456 slots->generation = gen | KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
1459 * Do not store the new memslots while there are invalidations in
1460 * progress, otherwise the locking in invalidate_range_start and
1461 * invalidate_range_end will be unbalanced.
1463 spin_lock(&kvm->mn_invalidate_lock);
1464 prepare_to_rcuwait(&kvm->mn_memslots_update_rcuwait);
1465 while (kvm->mn_active_invalidate_count) {
1466 set_current_state(TASK_UNINTERRUPTIBLE);
1467 spin_unlock(&kvm->mn_invalidate_lock);
1469 spin_lock(&kvm->mn_invalidate_lock);
1471 finish_rcuwait(&kvm->mn_memslots_update_rcuwait);
1472 rcu_assign_pointer(kvm->memslots[as_id], slots);
1473 spin_unlock(&kvm->mn_invalidate_lock);
1476 * Acquired in kvm_set_memslot. Must be released before synchronize
1477 * SRCU below in order to avoid deadlock with another thread
1478 * acquiring the slots_arch_lock in an srcu critical section.
1480 mutex_unlock(&kvm->slots_arch_lock);
1482 synchronize_srcu_expedited(&kvm->srcu);
1485 * Increment the new memslot generation a second time, dropping the
1486 * update in-progress flag and incrementing the generation based on
1487 * the number of address spaces. This provides a unique and easily
1488 * identifiable generation number while the memslots are in flux.
1490 gen = slots->generation & ~KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
1493 * Generations must be unique even across address spaces. We do not need
1494 * a global counter for that, instead the generation space is evenly split
1495 * across address spaces. For example, with two address spaces, address
1496 * space 0 will use generations 0, 2, 4, ... while address space 1 will
1497 * use generations 1, 3, 5, ...
1499 gen += KVM_ADDRESS_SPACE_NUM;
1501 kvm_arch_memslots_updated(kvm, gen);
1503 slots->generation = gen;
1505 return old_memslots;
1508 static size_t kvm_memslots_size(int slots)
1510 return sizeof(struct kvm_memslots) +
1511 (sizeof(struct kvm_memory_slot) * slots);
1514 static void kvm_copy_memslots(struct kvm_memslots *to,
1515 struct kvm_memslots *from)
1517 memcpy(to, from, kvm_memslots_size(from->used_slots));
1521 * Note, at a minimum, the current number of used slots must be allocated, even
1522 * when deleting a memslot, as we need a complete duplicate of the memslots for
1523 * use when invalidating a memslot prior to deleting/moving the memslot.
1525 static struct kvm_memslots *kvm_dup_memslots(struct kvm_memslots *old,
1526 enum kvm_mr_change change)
1528 struct kvm_memslots *slots;
1531 if (change == KVM_MR_CREATE)
1532 new_size = kvm_memslots_size(old->used_slots + 1);
1534 new_size = kvm_memslots_size(old->used_slots);
1536 slots = kvzalloc(new_size, GFP_KERNEL_ACCOUNT);
1538 kvm_copy_memslots(slots, old);
1543 static int kvm_set_memslot(struct kvm *kvm,
1544 const struct kvm_userspace_memory_region *mem,
1545 struct kvm_memory_slot *new, int as_id,
1546 enum kvm_mr_change change)
1548 struct kvm_memory_slot *slot, old;
1549 struct kvm_memslots *slots;
1553 * Released in install_new_memslots.
1555 * Must be held from before the current memslots are copied until
1556 * after the new memslots are installed with rcu_assign_pointer,
1557 * then released before the synchronize srcu in install_new_memslots.
1559 * When modifying memslots outside of the slots_lock, must be held
1560 * before reading the pointer to the current memslots until after all
1561 * changes to those memslots are complete.
1563 * These rules ensure that installing new memslots does not lose
1564 * changes made to the previous memslots.
1566 mutex_lock(&kvm->slots_arch_lock);
1568 slots = kvm_dup_memslots(__kvm_memslots(kvm, as_id), change);
1570 mutex_unlock(&kvm->slots_arch_lock);
1574 if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) {
1576 * Note, the INVALID flag needs to be in the appropriate entry
1577 * in the freshly allocated memslots, not in @old or @new.
1579 slot = id_to_memslot(slots, new->id);
1580 slot->flags |= KVM_MEMSLOT_INVALID;
1583 * We can re-use the memory from the old memslots.
1584 * It will be overwritten with a copy of the new memslots
1585 * after reacquiring the slots_arch_lock below.
1587 slots = install_new_memslots(kvm, as_id, slots);
1589 /* From this point no new shadow pages pointing to a deleted,
1590 * or moved, memslot will be created.
1592 * validation of sp->gfn happens in:
1593 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1594 * - kvm_is_visible_gfn (mmu_check_root)
1596 kvm_arch_flush_shadow_memslot(kvm, slot);
1598 /* Released in install_new_memslots. */
1599 mutex_lock(&kvm->slots_arch_lock);
1602 * The arch-specific fields of the memslots could have changed
1603 * between releasing the slots_arch_lock in
1604 * install_new_memslots and here, so get a fresh copy of the
1607 kvm_copy_memslots(slots, __kvm_memslots(kvm, as_id));
1611 * Make a full copy of the old memslot, the pointer will become stale
1612 * when the memslots are re-sorted by update_memslots(), and the old
1613 * memslot needs to be referenced after calling update_memslots(), e.g.
1614 * to free its resources and for arch specific behavior. This needs to
1615 * happen *after* (re)acquiring slots_arch_lock.
1617 slot = id_to_memslot(slots, new->id);
1621 WARN_ON_ONCE(change != KVM_MR_CREATE);
1622 memset(&old, 0, sizeof(old));
1627 /* Copy the arch-specific data, again after (re)acquiring slots_arch_lock. */
1628 memcpy(&new->arch, &old.arch, sizeof(old.arch));
1630 r = kvm_arch_prepare_memory_region(kvm, new, mem, change);
1634 update_memslots(slots, new, change);
1635 slots = install_new_memslots(kvm, as_id, slots);
1637 kvm_arch_commit_memory_region(kvm, mem, &old, new, change);
1639 /* Free the old memslot's metadata. Note, this is the full copy!!! */
1640 if (change == KVM_MR_DELETE)
1641 kvm_free_memslot(kvm, &old);
1647 if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) {
1648 slot = id_to_memslot(slots, new->id);
1649 slot->flags &= ~KVM_MEMSLOT_INVALID;
1650 slots = install_new_memslots(kvm, as_id, slots);
1652 mutex_unlock(&kvm->slots_arch_lock);
1658 static int kvm_delete_memslot(struct kvm *kvm,
1659 const struct kvm_userspace_memory_region *mem,
1660 struct kvm_memory_slot *old, int as_id)
1662 struct kvm_memory_slot new;
1667 memset(&new, 0, sizeof(new));
1670 * This is only for debugging purpose; it should never be referenced
1671 * for a removed memslot.
1675 return kvm_set_memslot(kvm, mem, &new, as_id, KVM_MR_DELETE);
1679 * Allocate some memory and give it an address in the guest physical address
1682 * Discontiguous memory is allowed, mostly for framebuffers.
1684 * Must be called holding kvm->slots_lock for write.
1686 int __kvm_set_memory_region(struct kvm *kvm,
1687 const struct kvm_userspace_memory_region *mem)
1689 struct kvm_memory_slot old, new;
1690 struct kvm_memory_slot *tmp;
1691 enum kvm_mr_change change;
1695 r = check_memory_region_flags(mem);
1699 as_id = mem->slot >> 16;
1700 id = (u16)mem->slot;
1702 /* General sanity checks */
1703 if ((mem->memory_size & (PAGE_SIZE - 1)) ||
1704 (mem->memory_size != (unsigned long)mem->memory_size))
1706 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
1708 /* We can read the guest memory with __xxx_user() later on. */
1709 if ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
1710 (mem->userspace_addr != untagged_addr(mem->userspace_addr)) ||
1711 !access_ok((void __user *)(unsigned long)mem->userspace_addr,
1714 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
1716 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
1720 * Make a full copy of the old memslot, the pointer will become stale
1721 * when the memslots are re-sorted by update_memslots(), and the old
1722 * memslot needs to be referenced after calling update_memslots(), e.g.
1723 * to free its resources and for arch specific behavior.
1725 tmp = id_to_memslot(__kvm_memslots(kvm, as_id), id);
1730 memset(&old, 0, sizeof(old));
1734 if (!mem->memory_size)
1735 return kvm_delete_memslot(kvm, mem, &old, as_id);
1739 new.base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
1740 new.npages = mem->memory_size >> PAGE_SHIFT;
1741 new.flags = mem->flags;
1742 new.userspace_addr = mem->userspace_addr;
1744 if (new.npages > KVM_MEM_MAX_NR_PAGES)
1748 change = KVM_MR_CREATE;
1749 new.dirty_bitmap = NULL;
1750 } else { /* Modify an existing slot. */
1751 if ((new.userspace_addr != old.userspace_addr) ||
1752 (new.npages != old.npages) ||
1753 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
1756 if (new.base_gfn != old.base_gfn)
1757 change = KVM_MR_MOVE;
1758 else if (new.flags != old.flags)
1759 change = KVM_MR_FLAGS_ONLY;
1760 else /* Nothing to change. */
1763 /* Copy dirty_bitmap from the current memslot. */
1764 new.dirty_bitmap = old.dirty_bitmap;
1767 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
1768 /* Check for overlaps */
1769 kvm_for_each_memslot(tmp, __kvm_memslots(kvm, as_id)) {
1772 if (!((new.base_gfn + new.npages <= tmp->base_gfn) ||
1773 (new.base_gfn >= tmp->base_gfn + tmp->npages)))
1778 /* Allocate/free page dirty bitmap as needed */
1779 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
1780 new.dirty_bitmap = NULL;
1781 else if (!new.dirty_bitmap && !kvm->dirty_ring_size) {
1782 r = kvm_alloc_dirty_bitmap(&new);
1786 if (kvm_dirty_log_manual_protect_and_init_set(kvm))
1787 bitmap_set(new.dirty_bitmap, 0, new.npages);
1790 r = kvm_set_memslot(kvm, mem, &new, as_id, change);
1794 if (old.dirty_bitmap && !new.dirty_bitmap)
1795 kvm_destroy_dirty_bitmap(&old);
1799 if (new.dirty_bitmap && !old.dirty_bitmap)
1800 kvm_destroy_dirty_bitmap(&new);
1803 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1805 int kvm_set_memory_region(struct kvm *kvm,
1806 const struct kvm_userspace_memory_region *mem)
1810 mutex_lock(&kvm->slots_lock);
1811 r = __kvm_set_memory_region(kvm, mem);
1812 mutex_unlock(&kvm->slots_lock);
1815 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1817 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1818 struct kvm_userspace_memory_region *mem)
1820 if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1823 return kvm_set_memory_region(kvm, mem);
1826 #ifndef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1828 * kvm_get_dirty_log - get a snapshot of dirty pages
1829 * @kvm: pointer to kvm instance
1830 * @log: slot id and address to which we copy the log
1831 * @is_dirty: set to '1' if any dirty pages were found
1832 * @memslot: set to the associated memslot, always valid on success
1834 int kvm_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log,
1835 int *is_dirty, struct kvm_memory_slot **memslot)
1837 struct kvm_memslots *slots;
1840 unsigned long any = 0;
1842 /* Dirty ring tracking is exclusive to dirty log tracking */
1843 if (kvm->dirty_ring_size)
1849 as_id = log->slot >> 16;
1850 id = (u16)log->slot;
1851 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1854 slots = __kvm_memslots(kvm, as_id);
1855 *memslot = id_to_memslot(slots, id);
1856 if (!(*memslot) || !(*memslot)->dirty_bitmap)
1859 kvm_arch_sync_dirty_log(kvm, *memslot);
1861 n = kvm_dirty_bitmap_bytes(*memslot);
1863 for (i = 0; !any && i < n/sizeof(long); ++i)
1864 any = (*memslot)->dirty_bitmap[i];
1866 if (copy_to_user(log->dirty_bitmap, (*memslot)->dirty_bitmap, n))
1873 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1875 #else /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
1877 * kvm_get_dirty_log_protect - get a snapshot of dirty pages
1878 * and reenable dirty page tracking for the corresponding pages.
1879 * @kvm: pointer to kvm instance
1880 * @log: slot id and address to which we copy the log
1882 * We need to keep it in mind that VCPU threads can write to the bitmap
1883 * concurrently. So, to avoid losing track of dirty pages we keep the
1886 * 1. Take a snapshot of the bit and clear it if needed.
1887 * 2. Write protect the corresponding page.
1888 * 3. Copy the snapshot to the userspace.
1889 * 4. Upon return caller flushes TLB's if needed.
1891 * Between 2 and 4, the guest may write to the page using the remaining TLB
1892 * entry. This is not a problem because the page is reported dirty using
1893 * the snapshot taken before and step 4 ensures that writes done after
1894 * exiting to userspace will be logged for the next call.
1897 static int kvm_get_dirty_log_protect(struct kvm *kvm, struct kvm_dirty_log *log)
1899 struct kvm_memslots *slots;
1900 struct kvm_memory_slot *memslot;
1903 unsigned long *dirty_bitmap;
1904 unsigned long *dirty_bitmap_buffer;
1907 /* Dirty ring tracking is exclusive to dirty log tracking */
1908 if (kvm->dirty_ring_size)
1911 as_id = log->slot >> 16;
1912 id = (u16)log->slot;
1913 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1916 slots = __kvm_memslots(kvm, as_id);
1917 memslot = id_to_memslot(slots, id);
1918 if (!memslot || !memslot->dirty_bitmap)
1921 dirty_bitmap = memslot->dirty_bitmap;
1923 kvm_arch_sync_dirty_log(kvm, memslot);
1925 n = kvm_dirty_bitmap_bytes(memslot);
1927 if (kvm->manual_dirty_log_protect) {
1929 * Unlike kvm_get_dirty_log, we always return false in *flush,
1930 * because no flush is needed until KVM_CLEAR_DIRTY_LOG. There
1931 * is some code duplication between this function and
1932 * kvm_get_dirty_log, but hopefully all architecture
1933 * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log
1934 * can be eliminated.
1936 dirty_bitmap_buffer = dirty_bitmap;
1938 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
1939 memset(dirty_bitmap_buffer, 0, n);
1942 for (i = 0; i < n / sizeof(long); i++) {
1946 if (!dirty_bitmap[i])
1950 mask = xchg(&dirty_bitmap[i], 0);
1951 dirty_bitmap_buffer[i] = mask;
1953 offset = i * BITS_PER_LONG;
1954 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1957 KVM_MMU_UNLOCK(kvm);
1961 kvm_arch_flush_remote_tlbs_memslot(kvm, memslot);
1963 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1970 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
1971 * @kvm: kvm instance
1972 * @log: slot id and address to which we copy the log
1974 * Steps 1-4 below provide general overview of dirty page logging. See
1975 * kvm_get_dirty_log_protect() function description for additional details.
1977 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
1978 * always flush the TLB (step 4) even if previous step failed and the dirty
1979 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
1980 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
1981 * writes will be marked dirty for next log read.
1983 * 1. Take a snapshot of the bit and clear it if needed.
1984 * 2. Write protect the corresponding page.
1985 * 3. Copy the snapshot to the userspace.
1986 * 4. Flush TLB's if needed.
1988 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1989 struct kvm_dirty_log *log)
1993 mutex_lock(&kvm->slots_lock);
1995 r = kvm_get_dirty_log_protect(kvm, log);
1997 mutex_unlock(&kvm->slots_lock);
2002 * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap
2003 * and reenable dirty page tracking for the corresponding pages.
2004 * @kvm: pointer to kvm instance
2005 * @log: slot id and address from which to fetch the bitmap of dirty pages
2007 static int kvm_clear_dirty_log_protect(struct kvm *kvm,
2008 struct kvm_clear_dirty_log *log)
2010 struct kvm_memslots *slots;
2011 struct kvm_memory_slot *memslot;
2015 unsigned long *dirty_bitmap;
2016 unsigned long *dirty_bitmap_buffer;
2019 /* Dirty ring tracking is exclusive to dirty log tracking */
2020 if (kvm->dirty_ring_size)
2023 as_id = log->slot >> 16;
2024 id = (u16)log->slot;
2025 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
2028 if (log->first_page & 63)
2031 slots = __kvm_memslots(kvm, as_id);
2032 memslot = id_to_memslot(slots, id);
2033 if (!memslot || !memslot->dirty_bitmap)
2036 dirty_bitmap = memslot->dirty_bitmap;
2038 n = ALIGN(log->num_pages, BITS_PER_LONG) / 8;
2040 if (log->first_page > memslot->npages ||
2041 log->num_pages > memslot->npages - log->first_page ||
2042 (log->num_pages < memslot->npages - log->first_page && (log->num_pages & 63)))
2045 kvm_arch_sync_dirty_log(kvm, memslot);
2048 dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
2049 if (copy_from_user(dirty_bitmap_buffer, log->dirty_bitmap, n))
2053 for (offset = log->first_page, i = offset / BITS_PER_LONG,
2054 n = DIV_ROUND_UP(log->num_pages, BITS_PER_LONG); n--;
2055 i++, offset += BITS_PER_LONG) {
2056 unsigned long mask = *dirty_bitmap_buffer++;
2057 atomic_long_t *p = (atomic_long_t *) &dirty_bitmap[i];
2061 mask &= atomic_long_fetch_andnot(mask, p);
2064 * mask contains the bits that really have been cleared. This
2065 * never includes any bits beyond the length of the memslot (if
2066 * the length is not aligned to 64 pages), therefore it is not
2067 * a problem if userspace sets them in log->dirty_bitmap.
2071 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
2075 KVM_MMU_UNLOCK(kvm);
2078 kvm_arch_flush_remote_tlbs_memslot(kvm, memslot);
2083 static int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm,
2084 struct kvm_clear_dirty_log *log)
2088 mutex_lock(&kvm->slots_lock);
2090 r = kvm_clear_dirty_log_protect(kvm, log);
2092 mutex_unlock(&kvm->slots_lock);
2095 #endif /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
2097 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
2099 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
2101 EXPORT_SYMBOL_GPL(gfn_to_memslot);
2103 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
2105 struct kvm_memslots *slots = kvm_vcpu_memslots(vcpu);
2106 struct kvm_memory_slot *slot;
2109 slot = try_get_memslot(slots, vcpu->last_used_slot, gfn);
2114 * Fall back to searching all memslots. We purposely use
2115 * search_memslots() instead of __gfn_to_memslot() to avoid
2116 * thrashing the VM-wide last_used_index in kvm_memslots.
2118 slot = search_memslots(slots, gfn, &slot_index);
2120 vcpu->last_used_slot = slot_index;
2127 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
2129 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
2131 return kvm_is_visible_memslot(memslot);
2133 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
2135 bool kvm_vcpu_is_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
2137 struct kvm_memory_slot *memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2139 return kvm_is_visible_memslot(memslot);
2141 EXPORT_SYMBOL_GPL(kvm_vcpu_is_visible_gfn);
2143 unsigned long kvm_host_page_size(struct kvm_vcpu *vcpu, gfn_t gfn)
2145 struct vm_area_struct *vma;
2146 unsigned long addr, size;
2150 addr = kvm_vcpu_gfn_to_hva_prot(vcpu, gfn, NULL);
2151 if (kvm_is_error_hva(addr))
2154 mmap_read_lock(current->mm);
2155 vma = find_vma(current->mm, addr);
2159 size = vma_kernel_pagesize(vma);
2162 mmap_read_unlock(current->mm);
2167 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
2169 return slot->flags & KVM_MEM_READONLY;
2172 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
2173 gfn_t *nr_pages, bool write)
2175 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
2176 return KVM_HVA_ERR_BAD;
2178 if (memslot_is_readonly(slot) && write)
2179 return KVM_HVA_ERR_RO_BAD;
2182 *nr_pages = slot->npages - (gfn - slot->base_gfn);
2184 return __gfn_to_hva_memslot(slot, gfn);
2187 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
2190 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
2193 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
2196 return gfn_to_hva_many(slot, gfn, NULL);
2198 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
2200 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
2202 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
2204 EXPORT_SYMBOL_GPL(gfn_to_hva);
2206 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
2208 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
2210 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
2213 * Return the hva of a @gfn and the R/W attribute if possible.
2215 * @slot: the kvm_memory_slot which contains @gfn
2216 * @gfn: the gfn to be translated
2217 * @writable: used to return the read/write attribute of the @slot if the hva
2218 * is valid and @writable is not NULL
2220 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
2221 gfn_t gfn, bool *writable)
2223 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
2225 if (!kvm_is_error_hva(hva) && writable)
2226 *writable = !memslot_is_readonly(slot);
2231 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
2233 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2235 return gfn_to_hva_memslot_prot(slot, gfn, writable);
2238 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
2240 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2242 return gfn_to_hva_memslot_prot(slot, gfn, writable);
2245 static inline int check_user_page_hwpoison(unsigned long addr)
2247 int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
2249 rc = get_user_pages(addr, 1, flags, NULL, NULL);
2250 return rc == -EHWPOISON;
2254 * The fast path to get the writable pfn which will be stored in @pfn,
2255 * true indicates success, otherwise false is returned. It's also the
2256 * only part that runs if we can in atomic context.
2258 static bool hva_to_pfn_fast(unsigned long addr, bool write_fault,
2259 bool *writable, kvm_pfn_t *pfn)
2261 struct page *page[1];
2264 * Fast pin a writable pfn only if it is a write fault request
2265 * or the caller allows to map a writable pfn for a read fault
2268 if (!(write_fault || writable))
2271 if (get_user_page_fast_only(addr, FOLL_WRITE, page)) {
2272 *pfn = page_to_pfn(page[0]);
2283 * The slow path to get the pfn of the specified host virtual address,
2284 * 1 indicates success, -errno is returned if error is detected.
2286 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
2287 bool *writable, kvm_pfn_t *pfn)
2289 unsigned int flags = FOLL_HWPOISON;
2296 *writable = write_fault;
2299 flags |= FOLL_WRITE;
2301 flags |= FOLL_NOWAIT;
2303 npages = get_user_pages_unlocked(addr, 1, &page, flags);
2307 /* map read fault as writable if possible */
2308 if (unlikely(!write_fault) && writable) {
2311 if (get_user_page_fast_only(addr, FOLL_WRITE, &wpage)) {
2317 *pfn = page_to_pfn(page);
2321 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
2323 if (unlikely(!(vma->vm_flags & VM_READ)))
2326 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
2332 static int kvm_try_get_pfn(kvm_pfn_t pfn)
2334 if (kvm_is_reserved_pfn(pfn))
2336 return get_page_unless_zero(pfn_to_page(pfn));
2339 static int hva_to_pfn_remapped(struct vm_area_struct *vma,
2340 unsigned long addr, bool *async,
2341 bool write_fault, bool *writable,
2349 r = follow_pte(vma->vm_mm, addr, &ptep, &ptl);
2352 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
2353 * not call the fault handler, so do it here.
2355 bool unlocked = false;
2356 r = fixup_user_fault(current->mm, addr,
2357 (write_fault ? FAULT_FLAG_WRITE : 0),
2364 r = follow_pte(vma->vm_mm, addr, &ptep, &ptl);
2369 if (write_fault && !pte_write(*ptep)) {
2370 pfn = KVM_PFN_ERR_RO_FAULT;
2375 *writable = pte_write(*ptep);
2376 pfn = pte_pfn(*ptep);
2379 * Get a reference here because callers of *hva_to_pfn* and
2380 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
2381 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
2382 * set, but the kvm_try_get_pfn/kvm_release_pfn_clean pair will
2383 * simply do nothing for reserved pfns.
2385 * Whoever called remap_pfn_range is also going to call e.g.
2386 * unmap_mapping_range before the underlying pages are freed,
2387 * causing a call to our MMU notifier.
2389 * Certain IO or PFNMAP mappings can be backed with valid
2390 * struct pages, but be allocated without refcounting e.g.,
2391 * tail pages of non-compound higher order allocations, which
2392 * would then underflow the refcount when the caller does the
2393 * required put_page. Don't allow those pages here.
2395 if (!kvm_try_get_pfn(pfn))
2399 pte_unmap_unlock(ptep, ptl);
2406 * Pin guest page in memory and return its pfn.
2407 * @addr: host virtual address which maps memory to the guest
2408 * @atomic: whether this function can sleep
2409 * @async: whether this function need to wait IO complete if the
2410 * host page is not in the memory
2411 * @write_fault: whether we should get a writable host page
2412 * @writable: whether it allows to map a writable host page for !@write_fault
2414 * The function will map a writable host page for these two cases:
2415 * 1): @write_fault = true
2416 * 2): @write_fault = false && @writable, @writable will tell the caller
2417 * whether the mapping is writable.
2419 static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
2420 bool write_fault, bool *writable)
2422 struct vm_area_struct *vma;
2426 /* we can do it either atomically or asynchronously, not both */
2427 BUG_ON(atomic && async);
2429 if (hva_to_pfn_fast(addr, write_fault, writable, &pfn))
2433 return KVM_PFN_ERR_FAULT;
2435 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
2439 mmap_read_lock(current->mm);
2440 if (npages == -EHWPOISON ||
2441 (!async && check_user_page_hwpoison(addr))) {
2442 pfn = KVM_PFN_ERR_HWPOISON;
2447 vma = vma_lookup(current->mm, addr);
2450 pfn = KVM_PFN_ERR_FAULT;
2451 else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
2452 r = hva_to_pfn_remapped(vma, addr, async, write_fault, writable, &pfn);
2456 pfn = KVM_PFN_ERR_FAULT;
2458 if (async && vma_is_valid(vma, write_fault))
2460 pfn = KVM_PFN_ERR_FAULT;
2463 mmap_read_unlock(current->mm);
2467 kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn,
2468 bool atomic, bool *async, bool write_fault,
2469 bool *writable, hva_t *hva)
2471 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
2476 if (addr == KVM_HVA_ERR_RO_BAD) {
2479 return KVM_PFN_ERR_RO_FAULT;
2482 if (kvm_is_error_hva(addr)) {
2485 return KVM_PFN_NOSLOT;
2488 /* Do not map writable pfn in the readonly memslot. */
2489 if (writable && memslot_is_readonly(slot)) {
2494 return hva_to_pfn(addr, atomic, async, write_fault,
2497 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
2499 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
2502 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
2503 write_fault, writable, NULL);
2505 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
2507 kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
2509 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL, NULL);
2511 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
2513 kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
2515 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL, NULL);
2517 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
2519 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
2521 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
2523 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
2525 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
2527 return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
2529 EXPORT_SYMBOL_GPL(gfn_to_pfn);
2531 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
2533 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
2535 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
2537 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
2538 struct page **pages, int nr_pages)
2543 addr = gfn_to_hva_many(slot, gfn, &entry);
2544 if (kvm_is_error_hva(addr))
2547 if (entry < nr_pages)
2550 return get_user_pages_fast_only(addr, nr_pages, FOLL_WRITE, pages);
2552 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
2554 static struct page *kvm_pfn_to_page(kvm_pfn_t pfn)
2556 if (is_error_noslot_pfn(pfn))
2557 return KVM_ERR_PTR_BAD_PAGE;
2559 if (kvm_is_reserved_pfn(pfn)) {
2561 return KVM_ERR_PTR_BAD_PAGE;
2564 return pfn_to_page(pfn);
2567 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
2571 pfn = gfn_to_pfn(kvm, gfn);
2573 return kvm_pfn_to_page(pfn);
2575 EXPORT_SYMBOL_GPL(gfn_to_page);
2577 void kvm_release_pfn(kvm_pfn_t pfn, bool dirty, struct gfn_to_pfn_cache *cache)
2583 cache->pfn = cache->gfn = 0;
2586 kvm_release_pfn_dirty(pfn);
2588 kvm_release_pfn_clean(pfn);
2591 static void kvm_cache_gfn_to_pfn(struct kvm_memory_slot *slot, gfn_t gfn,
2592 struct gfn_to_pfn_cache *cache, u64 gen)
2594 kvm_release_pfn(cache->pfn, cache->dirty, cache);
2596 cache->pfn = gfn_to_pfn_memslot(slot, gfn);
2598 cache->dirty = false;
2599 cache->generation = gen;
2602 static int __kvm_map_gfn(struct kvm_memslots *slots, gfn_t gfn,
2603 struct kvm_host_map *map,
2604 struct gfn_to_pfn_cache *cache,
2609 struct page *page = KVM_UNMAPPED_PAGE;
2610 struct kvm_memory_slot *slot = __gfn_to_memslot(slots, gfn);
2611 u64 gen = slots->generation;
2617 if (!cache->pfn || cache->gfn != gfn ||
2618 cache->generation != gen) {
2621 kvm_cache_gfn_to_pfn(slot, gfn, cache, gen);
2627 pfn = gfn_to_pfn_memslot(slot, gfn);
2629 if (is_error_noslot_pfn(pfn))
2632 if (pfn_valid(pfn)) {
2633 page = pfn_to_page(pfn);
2635 hva = kmap_atomic(page);
2638 #ifdef CONFIG_HAS_IOMEM
2639 } else if (!atomic) {
2640 hva = memremap(pfn_to_hpa(pfn), PAGE_SIZE, MEMREMAP_WB);
2657 int kvm_map_gfn(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map,
2658 struct gfn_to_pfn_cache *cache, bool atomic)
2660 return __kvm_map_gfn(kvm_memslots(vcpu->kvm), gfn, map,
2663 EXPORT_SYMBOL_GPL(kvm_map_gfn);
2665 int kvm_vcpu_map(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map)
2667 return __kvm_map_gfn(kvm_vcpu_memslots(vcpu), gfn, map,
2670 EXPORT_SYMBOL_GPL(kvm_vcpu_map);
2672 static void __kvm_unmap_gfn(struct kvm *kvm,
2673 struct kvm_memory_slot *memslot,
2674 struct kvm_host_map *map,
2675 struct gfn_to_pfn_cache *cache,
2676 bool dirty, bool atomic)
2684 if (map->page != KVM_UNMAPPED_PAGE) {
2686 kunmap_atomic(map->hva);
2690 #ifdef CONFIG_HAS_IOMEM
2694 WARN_ONCE(1, "Unexpected unmapping in atomic context");
2698 mark_page_dirty_in_slot(kvm, memslot, map->gfn);
2701 cache->dirty |= dirty;
2703 kvm_release_pfn(map->pfn, dirty, NULL);
2709 int kvm_unmap_gfn(struct kvm_vcpu *vcpu, struct kvm_host_map *map,
2710 struct gfn_to_pfn_cache *cache, bool dirty, bool atomic)
2712 __kvm_unmap_gfn(vcpu->kvm, gfn_to_memslot(vcpu->kvm, map->gfn), map,
2713 cache, dirty, atomic);
2716 EXPORT_SYMBOL_GPL(kvm_unmap_gfn);
2718 void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map, bool dirty)
2720 __kvm_unmap_gfn(vcpu->kvm, kvm_vcpu_gfn_to_memslot(vcpu, map->gfn),
2721 map, NULL, dirty, false);
2723 EXPORT_SYMBOL_GPL(kvm_vcpu_unmap);
2725 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
2729 pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
2731 return kvm_pfn_to_page(pfn);
2733 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
2735 void kvm_release_page_clean(struct page *page)
2737 WARN_ON(is_error_page(page));
2739 kvm_release_pfn_clean(page_to_pfn(page));
2741 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
2743 void kvm_release_pfn_clean(kvm_pfn_t pfn)
2745 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
2746 put_page(pfn_to_page(pfn));
2748 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
2750 void kvm_release_page_dirty(struct page *page)
2752 WARN_ON(is_error_page(page));
2754 kvm_release_pfn_dirty(page_to_pfn(page));
2756 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
2758 void kvm_release_pfn_dirty(kvm_pfn_t pfn)
2760 kvm_set_pfn_dirty(pfn);
2761 kvm_release_pfn_clean(pfn);
2763 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
2765 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
2767 if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn))
2768 SetPageDirty(pfn_to_page(pfn));
2770 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
2772 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
2774 if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn))
2775 mark_page_accessed(pfn_to_page(pfn));
2777 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
2779 static int next_segment(unsigned long len, int offset)
2781 if (len > PAGE_SIZE - offset)
2782 return PAGE_SIZE - offset;
2787 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
2788 void *data, int offset, int len)
2793 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
2794 if (kvm_is_error_hva(addr))
2796 r = __copy_from_user(data, (void __user *)addr + offset, len);
2802 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
2805 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2807 return __kvm_read_guest_page(slot, gfn, data, offset, len);
2809 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
2811 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
2812 int offset, int len)
2814 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2816 return __kvm_read_guest_page(slot, gfn, data, offset, len);
2818 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
2820 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
2822 gfn_t gfn = gpa >> PAGE_SHIFT;
2824 int offset = offset_in_page(gpa);
2827 while ((seg = next_segment(len, offset)) != 0) {
2828 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
2838 EXPORT_SYMBOL_GPL(kvm_read_guest);
2840 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
2842 gfn_t gfn = gpa >> PAGE_SHIFT;
2844 int offset = offset_in_page(gpa);
2847 while ((seg = next_segment(len, offset)) != 0) {
2848 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
2858 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
2860 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
2861 void *data, int offset, unsigned long len)
2866 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
2867 if (kvm_is_error_hva(addr))
2869 pagefault_disable();
2870 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
2877 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
2878 void *data, unsigned long len)
2880 gfn_t gfn = gpa >> PAGE_SHIFT;
2881 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2882 int offset = offset_in_page(gpa);
2884 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
2886 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
2888 static int __kvm_write_guest_page(struct kvm *kvm,
2889 struct kvm_memory_slot *memslot, gfn_t gfn,
2890 const void *data, int offset, int len)
2895 addr = gfn_to_hva_memslot(memslot, gfn);
2896 if (kvm_is_error_hva(addr))
2898 r = __copy_to_user((void __user *)addr + offset, data, len);
2901 mark_page_dirty_in_slot(kvm, memslot, gfn);
2905 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
2906 const void *data, int offset, int len)
2908 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2910 return __kvm_write_guest_page(kvm, slot, gfn, data, offset, len);
2912 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
2914 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
2915 const void *data, int offset, int len)
2917 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2919 return __kvm_write_guest_page(vcpu->kvm, slot, gfn, data, offset, len);
2921 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
2923 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
2926 gfn_t gfn = gpa >> PAGE_SHIFT;
2928 int offset = offset_in_page(gpa);
2931 while ((seg = next_segment(len, offset)) != 0) {
2932 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
2942 EXPORT_SYMBOL_GPL(kvm_write_guest);
2944 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
2947 gfn_t gfn = gpa >> PAGE_SHIFT;
2949 int offset = offset_in_page(gpa);
2952 while ((seg = next_segment(len, offset)) != 0) {
2953 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
2963 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
2965 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
2966 struct gfn_to_hva_cache *ghc,
2967 gpa_t gpa, unsigned long len)
2969 int offset = offset_in_page(gpa);
2970 gfn_t start_gfn = gpa >> PAGE_SHIFT;
2971 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
2972 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
2973 gfn_t nr_pages_avail;
2975 /* Update ghc->generation before performing any error checks. */
2976 ghc->generation = slots->generation;
2978 if (start_gfn > end_gfn) {
2979 ghc->hva = KVM_HVA_ERR_BAD;
2984 * If the requested region crosses two memslots, we still
2985 * verify that the entire region is valid here.
2987 for ( ; start_gfn <= end_gfn; start_gfn += nr_pages_avail) {
2988 ghc->memslot = __gfn_to_memslot(slots, start_gfn);
2989 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
2991 if (kvm_is_error_hva(ghc->hva))
2995 /* Use the slow path for cross page reads and writes. */
2996 if (nr_pages_needed == 1)
2999 ghc->memslot = NULL;
3006 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3007 gpa_t gpa, unsigned long len)
3009 struct kvm_memslots *slots = kvm_memslots(kvm);
3010 return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len);
3012 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
3014 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3015 void *data, unsigned int offset,
3018 struct kvm_memslots *slots = kvm_memslots(kvm);
3020 gpa_t gpa = ghc->gpa + offset;
3022 if (WARN_ON_ONCE(len + offset > ghc->len))
3025 if (slots->generation != ghc->generation) {
3026 if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len))
3030 if (kvm_is_error_hva(ghc->hva))
3033 if (unlikely(!ghc->memslot))
3034 return kvm_write_guest(kvm, gpa, data, len);
3036 r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
3039 mark_page_dirty_in_slot(kvm, ghc->memslot, gpa >> PAGE_SHIFT);
3043 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached);
3045 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3046 void *data, unsigned long len)
3048 return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len);
3050 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
3052 int kvm_read_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3053 void *data, unsigned int offset,
3056 struct kvm_memslots *slots = kvm_memslots(kvm);
3058 gpa_t gpa = ghc->gpa + offset;
3060 if (WARN_ON_ONCE(len + offset > ghc->len))
3063 if (slots->generation != ghc->generation) {
3064 if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len))
3068 if (kvm_is_error_hva(ghc->hva))
3071 if (unlikely(!ghc->memslot))
3072 return kvm_read_guest(kvm, gpa, data, len);
3074 r = __copy_from_user(data, (void __user *)ghc->hva + offset, len);
3080 EXPORT_SYMBOL_GPL(kvm_read_guest_offset_cached);
3082 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3083 void *data, unsigned long len)
3085 return kvm_read_guest_offset_cached(kvm, ghc, data, 0, len);
3087 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
3089 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
3091 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
3092 gfn_t gfn = gpa >> PAGE_SHIFT;
3094 int offset = offset_in_page(gpa);
3097 while ((seg = next_segment(len, offset)) != 0) {
3098 ret = kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
3107 EXPORT_SYMBOL_GPL(kvm_clear_guest);
3109 void mark_page_dirty_in_slot(struct kvm *kvm,
3110 struct kvm_memory_slot *memslot,
3113 if (memslot && kvm_slot_dirty_track_enabled(memslot)) {
3114 unsigned long rel_gfn = gfn - memslot->base_gfn;
3115 u32 slot = (memslot->as_id << 16) | memslot->id;
3117 if (kvm->dirty_ring_size)
3118 kvm_dirty_ring_push(kvm_dirty_ring_get(kvm),
3121 set_bit_le(rel_gfn, memslot->dirty_bitmap);
3124 EXPORT_SYMBOL_GPL(mark_page_dirty_in_slot);
3126 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
3128 struct kvm_memory_slot *memslot;
3130 memslot = gfn_to_memslot(kvm, gfn);
3131 mark_page_dirty_in_slot(kvm, memslot, gfn);
3133 EXPORT_SYMBOL_GPL(mark_page_dirty);
3135 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
3137 struct kvm_memory_slot *memslot;
3139 memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
3140 mark_page_dirty_in_slot(vcpu->kvm, memslot, gfn);
3142 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
3144 void kvm_sigset_activate(struct kvm_vcpu *vcpu)
3146 if (!vcpu->sigset_active)
3150 * This does a lockless modification of ->real_blocked, which is fine
3151 * because, only current can change ->real_blocked and all readers of
3152 * ->real_blocked don't care as long ->real_blocked is always a subset
3155 sigprocmask(SIG_SETMASK, &vcpu->sigset, ¤t->real_blocked);
3158 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu)
3160 if (!vcpu->sigset_active)
3163 sigprocmask(SIG_SETMASK, ¤t->real_blocked, NULL);
3164 sigemptyset(¤t->real_blocked);
3167 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
3169 unsigned int old, val, grow, grow_start;
3171 old = val = vcpu->halt_poll_ns;
3172 grow_start = READ_ONCE(halt_poll_ns_grow_start);
3173 grow = READ_ONCE(halt_poll_ns_grow);
3178 if (val < grow_start)
3181 if (val > vcpu->kvm->max_halt_poll_ns)
3182 val = vcpu->kvm->max_halt_poll_ns;
3184 vcpu->halt_poll_ns = val;
3186 trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
3189 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
3191 unsigned int old, val, shrink, grow_start;
3193 old = val = vcpu->halt_poll_ns;
3194 shrink = READ_ONCE(halt_poll_ns_shrink);
3195 grow_start = READ_ONCE(halt_poll_ns_grow_start);
3201 if (val < grow_start)
3204 vcpu->halt_poll_ns = val;
3205 trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
3208 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
3211 int idx = srcu_read_lock(&vcpu->kvm->srcu);
3213 if (kvm_arch_vcpu_runnable(vcpu)) {
3214 kvm_make_request(KVM_REQ_UNHALT, vcpu);
3217 if (kvm_cpu_has_pending_timer(vcpu))
3219 if (signal_pending(current))
3221 if (kvm_check_request(KVM_REQ_UNBLOCK, vcpu))
3226 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3231 update_halt_poll_stats(struct kvm_vcpu *vcpu, u64 poll_ns, bool waited)
3234 vcpu->stat.generic.halt_poll_fail_ns += poll_ns;
3236 vcpu->stat.generic.halt_poll_success_ns += poll_ns;
3240 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
3242 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
3244 bool halt_poll_allowed = !kvm_arch_no_poll(vcpu);
3245 ktime_t start, cur, poll_end;
3246 bool waited = false;
3249 kvm_arch_vcpu_blocking(vcpu);
3251 start = cur = poll_end = ktime_get();
3252 if (vcpu->halt_poll_ns && halt_poll_allowed) {
3253 ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns);
3255 ++vcpu->stat.generic.halt_attempted_poll;
3258 * This sets KVM_REQ_UNHALT if an interrupt
3261 if (kvm_vcpu_check_block(vcpu) < 0) {
3262 ++vcpu->stat.generic.halt_successful_poll;
3263 if (!vcpu_valid_wakeup(vcpu))
3264 ++vcpu->stat.generic.halt_poll_invalid;
3266 KVM_STATS_LOG_HIST_UPDATE(
3267 vcpu->stat.generic.halt_poll_success_hist,
3268 ktime_to_ns(ktime_get()) -
3269 ktime_to_ns(start));
3273 poll_end = cur = ktime_get();
3274 } while (kvm_vcpu_can_poll(cur, stop));
3276 KVM_STATS_LOG_HIST_UPDATE(
3277 vcpu->stat.generic.halt_poll_fail_hist,
3278 ktime_to_ns(ktime_get()) - ktime_to_ns(start));
3282 prepare_to_rcuwait(&vcpu->wait);
3284 set_current_state(TASK_INTERRUPTIBLE);
3286 if (kvm_vcpu_check_block(vcpu) < 0)
3292 finish_rcuwait(&vcpu->wait);
3295 vcpu->stat.generic.halt_wait_ns +=
3296 ktime_to_ns(cur) - ktime_to_ns(poll_end);
3297 KVM_STATS_LOG_HIST_UPDATE(vcpu->stat.generic.halt_wait_hist,
3298 ktime_to_ns(cur) - ktime_to_ns(poll_end));
3301 kvm_arch_vcpu_unblocking(vcpu);
3302 block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
3304 update_halt_poll_stats(
3305 vcpu, ktime_to_ns(ktime_sub(poll_end, start)), waited);
3307 if (halt_poll_allowed) {
3308 if (!vcpu_valid_wakeup(vcpu)) {
3309 shrink_halt_poll_ns(vcpu);
3310 } else if (vcpu->kvm->max_halt_poll_ns) {
3311 if (block_ns <= vcpu->halt_poll_ns)
3313 /* we had a long block, shrink polling */
3314 else if (vcpu->halt_poll_ns &&
3315 block_ns > vcpu->kvm->max_halt_poll_ns)
3316 shrink_halt_poll_ns(vcpu);
3317 /* we had a short halt and our poll time is too small */
3318 else if (vcpu->halt_poll_ns < vcpu->kvm->max_halt_poll_ns &&
3319 block_ns < vcpu->kvm->max_halt_poll_ns)
3320 grow_halt_poll_ns(vcpu);
3322 vcpu->halt_poll_ns = 0;
3326 trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu));
3327 kvm_arch_vcpu_block_finish(vcpu);
3329 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
3331 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
3333 struct rcuwait *waitp;
3335 waitp = kvm_arch_vcpu_get_wait(vcpu);
3336 if (rcuwait_wake_up(waitp)) {
3337 WRITE_ONCE(vcpu->ready, true);
3338 ++vcpu->stat.generic.halt_wakeup;
3344 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
3348 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
3350 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
3354 if (kvm_vcpu_wake_up(vcpu))
3358 * Note, the vCPU could get migrated to a different pCPU at any point
3359 * after kvm_arch_vcpu_should_kick(), which could result in sending an
3360 * IPI to the previous pCPU. But, that's ok because the purpose of the
3361 * IPI is to force the vCPU to leave IN_GUEST_MODE, and migrating the
3362 * vCPU also requires it to leave IN_GUEST_MODE.
3365 if (kvm_arch_vcpu_should_kick(vcpu)) {
3366 cpu = READ_ONCE(vcpu->cpu);
3367 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
3368 smp_send_reschedule(cpu);
3372 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
3373 #endif /* !CONFIG_S390 */
3375 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
3378 struct task_struct *task = NULL;
3382 pid = rcu_dereference(target->pid);
3384 task = get_pid_task(pid, PIDTYPE_PID);
3388 ret = yield_to(task, 1);
3389 put_task_struct(task);
3393 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
3396 * Helper that checks whether a VCPU is eligible for directed yield.
3397 * Most eligible candidate to yield is decided by following heuristics:
3399 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
3400 * (preempted lock holder), indicated by @in_spin_loop.
3401 * Set at the beginning and cleared at the end of interception/PLE handler.
3403 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
3404 * chance last time (mostly it has become eligible now since we have probably
3405 * yielded to lockholder in last iteration. This is done by toggling
3406 * @dy_eligible each time a VCPU checked for eligibility.)
3408 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
3409 * to preempted lock-holder could result in wrong VCPU selection and CPU
3410 * burning. Giving priority for a potential lock-holder increases lock
3413 * Since algorithm is based on heuristics, accessing another VCPU data without
3414 * locking does not harm. It may result in trying to yield to same VCPU, fail
3415 * and continue with next VCPU and so on.
3417 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
3419 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
3422 eligible = !vcpu->spin_loop.in_spin_loop ||
3423 vcpu->spin_loop.dy_eligible;
3425 if (vcpu->spin_loop.in_spin_loop)
3426 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
3435 * Unlike kvm_arch_vcpu_runnable, this function is called outside
3436 * a vcpu_load/vcpu_put pair. However, for most architectures
3437 * kvm_arch_vcpu_runnable does not require vcpu_load.
3439 bool __weak kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
3441 return kvm_arch_vcpu_runnable(vcpu);
3444 static bool vcpu_dy_runnable(struct kvm_vcpu *vcpu)
3446 if (kvm_arch_dy_runnable(vcpu))
3449 #ifdef CONFIG_KVM_ASYNC_PF
3450 if (!list_empty_careful(&vcpu->async_pf.done))
3457 bool __weak kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu)
3462 void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode)
3464 struct kvm *kvm = me->kvm;
3465 struct kvm_vcpu *vcpu;
3466 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
3472 kvm_vcpu_set_in_spin_loop(me, true);
3474 * We boost the priority of a VCPU that is runnable but not
3475 * currently running, because it got preempted by something
3476 * else and called schedule in __vcpu_run. Hopefully that
3477 * VCPU is holding the lock that we need and will release it.
3478 * We approximate round-robin by starting at the last boosted VCPU.
3480 for (pass = 0; pass < 2 && !yielded && try; pass++) {
3481 kvm_for_each_vcpu(i, vcpu, kvm) {
3482 if (!pass && i <= last_boosted_vcpu) {
3483 i = last_boosted_vcpu;
3485 } else if (pass && i > last_boosted_vcpu)
3487 if (!READ_ONCE(vcpu->ready))
3491 if (rcuwait_active(&vcpu->wait) &&
3492 !vcpu_dy_runnable(vcpu))
3494 if (READ_ONCE(vcpu->preempted) && yield_to_kernel_mode &&
3495 !kvm_arch_dy_has_pending_interrupt(vcpu) &&
3496 !kvm_arch_vcpu_in_kernel(vcpu))
3498 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
3501 yielded = kvm_vcpu_yield_to(vcpu);
3503 kvm->last_boosted_vcpu = i;
3505 } else if (yielded < 0) {
3512 kvm_vcpu_set_in_spin_loop(me, false);
3514 /* Ensure vcpu is not eligible during next spinloop */
3515 kvm_vcpu_set_dy_eligible(me, false);
3517 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
3519 static bool kvm_page_in_dirty_ring(struct kvm *kvm, unsigned long pgoff)
3521 #if KVM_DIRTY_LOG_PAGE_OFFSET > 0
3522 return (pgoff >= KVM_DIRTY_LOG_PAGE_OFFSET) &&
3523 (pgoff < KVM_DIRTY_LOG_PAGE_OFFSET +
3524 kvm->dirty_ring_size / PAGE_SIZE);
3530 static vm_fault_t kvm_vcpu_fault(struct vm_fault *vmf)
3532 struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data;
3535 if (vmf->pgoff == 0)
3536 page = virt_to_page(vcpu->run);
3538 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
3539 page = virt_to_page(vcpu->arch.pio_data);
3541 #ifdef CONFIG_KVM_MMIO
3542 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
3543 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
3545 else if (kvm_page_in_dirty_ring(vcpu->kvm, vmf->pgoff))
3546 page = kvm_dirty_ring_get_page(
3548 vmf->pgoff - KVM_DIRTY_LOG_PAGE_OFFSET);
3550 return kvm_arch_vcpu_fault(vcpu, vmf);
3556 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
3557 .fault = kvm_vcpu_fault,
3560 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
3562 struct kvm_vcpu *vcpu = file->private_data;
3563 unsigned long pages = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
3565 if ((kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff) ||
3566 kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff + pages - 1)) &&
3567 ((vma->vm_flags & VM_EXEC) || !(vma->vm_flags & VM_SHARED)))
3570 vma->vm_ops = &kvm_vcpu_vm_ops;
3574 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
3576 struct kvm_vcpu *vcpu = filp->private_data;
3578 kvm_put_kvm(vcpu->kvm);
3582 static struct file_operations kvm_vcpu_fops = {
3583 .release = kvm_vcpu_release,
3584 .unlocked_ioctl = kvm_vcpu_ioctl,
3585 .mmap = kvm_vcpu_mmap,
3586 .llseek = noop_llseek,
3587 KVM_COMPAT(kvm_vcpu_compat_ioctl),
3591 * Allocates an inode for the vcpu.
3593 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
3595 char name[8 + 1 + ITOA_MAX_LEN + 1];
3597 snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id);
3598 return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
3601 static void kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
3603 #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS
3604 struct dentry *debugfs_dentry;
3605 char dir_name[ITOA_MAX_LEN * 2];
3607 if (!debugfs_initialized())
3610 snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
3611 debugfs_dentry = debugfs_create_dir(dir_name,
3612 vcpu->kvm->debugfs_dentry);
3614 kvm_arch_create_vcpu_debugfs(vcpu, debugfs_dentry);
3619 * Creates some virtual cpus. Good luck creating more than one.
3621 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
3624 struct kvm_vcpu *vcpu;
3627 if (id >= KVM_MAX_VCPU_ID)
3630 mutex_lock(&kvm->lock);
3631 if (kvm->created_vcpus == KVM_MAX_VCPUS) {
3632 mutex_unlock(&kvm->lock);
3636 kvm->created_vcpus++;
3637 mutex_unlock(&kvm->lock);
3639 r = kvm_arch_vcpu_precreate(kvm, id);
3641 goto vcpu_decrement;
3643 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL_ACCOUNT);
3646 goto vcpu_decrement;
3649 BUILD_BUG_ON(sizeof(struct kvm_run) > PAGE_SIZE);
3650 page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
3655 vcpu->run = page_address(page);
3657 kvm_vcpu_init(vcpu, kvm, id);
3659 r = kvm_arch_vcpu_create(vcpu);
3661 goto vcpu_free_run_page;
3663 if (kvm->dirty_ring_size) {
3664 r = kvm_dirty_ring_alloc(&vcpu->dirty_ring,
3665 id, kvm->dirty_ring_size);
3667 goto arch_vcpu_destroy;
3670 mutex_lock(&kvm->lock);
3671 if (kvm_get_vcpu_by_id(kvm, id)) {
3673 goto unlock_vcpu_destroy;
3676 vcpu->vcpu_idx = atomic_read(&kvm->online_vcpus);
3677 BUG_ON(kvm->vcpus[vcpu->vcpu_idx]);
3679 /* Fill the stats id string for the vcpu */
3680 snprintf(vcpu->stats_id, sizeof(vcpu->stats_id), "kvm-%d/vcpu-%d",
3681 task_pid_nr(current), id);
3683 /* Now it's all set up, let userspace reach it */
3685 r = create_vcpu_fd(vcpu);
3687 kvm_put_kvm_no_destroy(kvm);
3688 goto unlock_vcpu_destroy;
3691 kvm->vcpus[vcpu->vcpu_idx] = vcpu;
3694 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
3695 * before kvm->online_vcpu's incremented value.
3698 atomic_inc(&kvm->online_vcpus);
3700 mutex_unlock(&kvm->lock);
3701 kvm_arch_vcpu_postcreate(vcpu);
3702 kvm_create_vcpu_debugfs(vcpu);
3705 unlock_vcpu_destroy:
3706 mutex_unlock(&kvm->lock);
3707 kvm_dirty_ring_free(&vcpu->dirty_ring);
3709 kvm_arch_vcpu_destroy(vcpu);
3711 free_page((unsigned long)vcpu->run);
3713 kmem_cache_free(kvm_vcpu_cache, vcpu);
3715 mutex_lock(&kvm->lock);
3716 kvm->created_vcpus--;
3717 mutex_unlock(&kvm->lock);
3721 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
3724 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
3725 vcpu->sigset_active = 1;
3726 vcpu->sigset = *sigset;
3728 vcpu->sigset_active = 0;
3732 static ssize_t kvm_vcpu_stats_read(struct file *file, char __user *user_buffer,
3733 size_t size, loff_t *offset)
3735 struct kvm_vcpu *vcpu = file->private_data;
3737 return kvm_stats_read(vcpu->stats_id, &kvm_vcpu_stats_header,
3738 &kvm_vcpu_stats_desc[0], &vcpu->stat,
3739 sizeof(vcpu->stat), user_buffer, size, offset);
3742 static const struct file_operations kvm_vcpu_stats_fops = {
3743 .read = kvm_vcpu_stats_read,
3744 .llseek = noop_llseek,
3747 static int kvm_vcpu_ioctl_get_stats_fd(struct kvm_vcpu *vcpu)
3751 char name[15 + ITOA_MAX_LEN + 1];
3753 snprintf(name, sizeof(name), "kvm-vcpu-stats:%d", vcpu->vcpu_id);
3755 fd = get_unused_fd_flags(O_CLOEXEC);
3759 file = anon_inode_getfile(name, &kvm_vcpu_stats_fops, vcpu, O_RDONLY);
3762 return PTR_ERR(file);
3764 file->f_mode |= FMODE_PREAD;
3765 fd_install(fd, file);
3770 static long kvm_vcpu_ioctl(struct file *filp,
3771 unsigned int ioctl, unsigned long arg)
3773 struct kvm_vcpu *vcpu = filp->private_data;
3774 void __user *argp = (void __user *)arg;
3776 struct kvm_fpu *fpu = NULL;
3777 struct kvm_sregs *kvm_sregs = NULL;
3779 if (vcpu->kvm->mm != current->mm || vcpu->kvm->vm_bugged)
3782 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
3786 * Some architectures have vcpu ioctls that are asynchronous to vcpu
3787 * execution; mutex_lock() would break them.
3789 r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg);
3790 if (r != -ENOIOCTLCMD)
3793 if (mutex_lock_killable(&vcpu->mutex))
3801 oldpid = rcu_access_pointer(vcpu->pid);
3802 if (unlikely(oldpid != task_pid(current))) {
3803 /* The thread running this VCPU changed. */
3806 r = kvm_arch_vcpu_run_pid_change(vcpu);
3810 newpid = get_task_pid(current, PIDTYPE_PID);
3811 rcu_assign_pointer(vcpu->pid, newpid);
3816 r = kvm_arch_vcpu_ioctl_run(vcpu);
3817 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
3820 case KVM_GET_REGS: {
3821 struct kvm_regs *kvm_regs;
3824 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL_ACCOUNT);
3827 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
3831 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
3838 case KVM_SET_REGS: {
3839 struct kvm_regs *kvm_regs;
3841 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
3842 if (IS_ERR(kvm_regs)) {
3843 r = PTR_ERR(kvm_regs);
3846 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
3850 case KVM_GET_SREGS: {
3851 kvm_sregs = kzalloc(sizeof(struct kvm_sregs),
3852 GFP_KERNEL_ACCOUNT);
3856 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
3860 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
3865 case KVM_SET_SREGS: {
3866 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
3867 if (IS_ERR(kvm_sregs)) {
3868 r = PTR_ERR(kvm_sregs);
3872 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
3875 case KVM_GET_MP_STATE: {
3876 struct kvm_mp_state mp_state;
3878 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
3882 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
3887 case KVM_SET_MP_STATE: {
3888 struct kvm_mp_state mp_state;
3891 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
3893 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
3896 case KVM_TRANSLATE: {
3897 struct kvm_translation tr;
3900 if (copy_from_user(&tr, argp, sizeof(tr)))
3902 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
3906 if (copy_to_user(argp, &tr, sizeof(tr)))
3911 case KVM_SET_GUEST_DEBUG: {
3912 struct kvm_guest_debug dbg;
3915 if (copy_from_user(&dbg, argp, sizeof(dbg)))
3917 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
3920 case KVM_SET_SIGNAL_MASK: {
3921 struct kvm_signal_mask __user *sigmask_arg = argp;
3922 struct kvm_signal_mask kvm_sigmask;
3923 sigset_t sigset, *p;
3928 if (copy_from_user(&kvm_sigmask, argp,
3929 sizeof(kvm_sigmask)))
3932 if (kvm_sigmask.len != sizeof(sigset))
3935 if (copy_from_user(&sigset, sigmask_arg->sigset,
3940 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
3944 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL_ACCOUNT);
3948 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
3952 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
3958 fpu = memdup_user(argp, sizeof(*fpu));
3964 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
3967 case KVM_GET_STATS_FD: {
3968 r = kvm_vcpu_ioctl_get_stats_fd(vcpu);
3972 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
3975 mutex_unlock(&vcpu->mutex);
3981 #ifdef CONFIG_KVM_COMPAT
3982 static long kvm_vcpu_compat_ioctl(struct file *filp,
3983 unsigned int ioctl, unsigned long arg)
3985 struct kvm_vcpu *vcpu = filp->private_data;
3986 void __user *argp = compat_ptr(arg);
3989 if (vcpu->kvm->mm != current->mm || vcpu->kvm->vm_bugged)
3993 case KVM_SET_SIGNAL_MASK: {
3994 struct kvm_signal_mask __user *sigmask_arg = argp;
3995 struct kvm_signal_mask kvm_sigmask;
4000 if (copy_from_user(&kvm_sigmask, argp,
4001 sizeof(kvm_sigmask)))
4004 if (kvm_sigmask.len != sizeof(compat_sigset_t))
4007 if (get_compat_sigset(&sigset,
4008 (compat_sigset_t __user *)sigmask_arg->sigset))
4010 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
4012 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
4016 r = kvm_vcpu_ioctl(filp, ioctl, arg);
4024 static int kvm_device_mmap(struct file *filp, struct vm_area_struct *vma)
4026 struct kvm_device *dev = filp->private_data;
4029 return dev->ops->mmap(dev, vma);
4034 static int kvm_device_ioctl_attr(struct kvm_device *dev,
4035 int (*accessor)(struct kvm_device *dev,
4036 struct kvm_device_attr *attr),
4039 struct kvm_device_attr attr;
4044 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
4047 return accessor(dev, &attr);
4050 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
4053 struct kvm_device *dev = filp->private_data;
4055 if (dev->kvm->mm != current->mm || dev->kvm->vm_bugged)
4059 case KVM_SET_DEVICE_ATTR:
4060 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
4061 case KVM_GET_DEVICE_ATTR:
4062 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
4063 case KVM_HAS_DEVICE_ATTR:
4064 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
4066 if (dev->ops->ioctl)
4067 return dev->ops->ioctl(dev, ioctl, arg);
4073 static int kvm_device_release(struct inode *inode, struct file *filp)
4075 struct kvm_device *dev = filp->private_data;
4076 struct kvm *kvm = dev->kvm;
4078 if (dev->ops->release) {
4079 mutex_lock(&kvm->lock);
4080 list_del(&dev->vm_node);
4081 dev->ops->release(dev);
4082 mutex_unlock(&kvm->lock);
4089 static const struct file_operations kvm_device_fops = {
4090 .unlocked_ioctl = kvm_device_ioctl,
4091 .release = kvm_device_release,
4092 KVM_COMPAT(kvm_device_ioctl),
4093 .mmap = kvm_device_mmap,
4096 struct kvm_device *kvm_device_from_filp(struct file *filp)
4098 if (filp->f_op != &kvm_device_fops)
4101 return filp->private_data;
4104 static const struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
4105 #ifdef CONFIG_KVM_MPIC
4106 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
4107 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
4111 int kvm_register_device_ops(const struct kvm_device_ops *ops, u32 type)
4113 if (type >= ARRAY_SIZE(kvm_device_ops_table))
4116 if (kvm_device_ops_table[type] != NULL)
4119 kvm_device_ops_table[type] = ops;
4123 void kvm_unregister_device_ops(u32 type)
4125 if (kvm_device_ops_table[type] != NULL)
4126 kvm_device_ops_table[type] = NULL;
4129 static int kvm_ioctl_create_device(struct kvm *kvm,
4130 struct kvm_create_device *cd)
4132 const struct kvm_device_ops *ops = NULL;
4133 struct kvm_device *dev;
4134 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
4138 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
4141 type = array_index_nospec(cd->type, ARRAY_SIZE(kvm_device_ops_table));
4142 ops = kvm_device_ops_table[type];
4149 dev = kzalloc(sizeof(*dev), GFP_KERNEL_ACCOUNT);
4156 mutex_lock(&kvm->lock);
4157 ret = ops->create(dev, type);
4159 mutex_unlock(&kvm->lock);
4163 list_add(&dev->vm_node, &kvm->devices);
4164 mutex_unlock(&kvm->lock);
4170 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
4172 kvm_put_kvm_no_destroy(kvm);
4173 mutex_lock(&kvm->lock);
4174 list_del(&dev->vm_node);
4175 mutex_unlock(&kvm->lock);
4184 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
4187 case KVM_CAP_USER_MEMORY:
4188 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
4189 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
4190 case KVM_CAP_INTERNAL_ERROR_DATA:
4191 #ifdef CONFIG_HAVE_KVM_MSI
4192 case KVM_CAP_SIGNAL_MSI:
4194 #ifdef CONFIG_HAVE_KVM_IRQFD
4196 case KVM_CAP_IRQFD_RESAMPLE:
4198 case KVM_CAP_IOEVENTFD_ANY_LENGTH:
4199 case KVM_CAP_CHECK_EXTENSION_VM:
4200 case KVM_CAP_ENABLE_CAP_VM:
4201 case KVM_CAP_HALT_POLL:
4203 #ifdef CONFIG_KVM_MMIO
4204 case KVM_CAP_COALESCED_MMIO:
4205 return KVM_COALESCED_MMIO_PAGE_OFFSET;
4206 case KVM_CAP_COALESCED_PIO:
4209 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4210 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2:
4211 return KVM_DIRTY_LOG_MANUAL_CAPS;
4213 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
4214 case KVM_CAP_IRQ_ROUTING:
4215 return KVM_MAX_IRQ_ROUTES;
4217 #if KVM_ADDRESS_SPACE_NUM > 1
4218 case KVM_CAP_MULTI_ADDRESS_SPACE:
4219 return KVM_ADDRESS_SPACE_NUM;
4221 case KVM_CAP_NR_MEMSLOTS:
4222 return KVM_USER_MEM_SLOTS;
4223 case KVM_CAP_DIRTY_LOG_RING:
4224 #if KVM_DIRTY_LOG_PAGE_OFFSET > 0
4225 return KVM_DIRTY_RING_MAX_ENTRIES * sizeof(struct kvm_dirty_gfn);
4229 case KVM_CAP_BINARY_STATS_FD:
4234 return kvm_vm_ioctl_check_extension(kvm, arg);
4237 static int kvm_vm_ioctl_enable_dirty_log_ring(struct kvm *kvm, u32 size)
4241 if (!KVM_DIRTY_LOG_PAGE_OFFSET)
4244 /* the size should be power of 2 */
4245 if (!size || (size & (size - 1)))
4248 /* Should be bigger to keep the reserved entries, or a page */
4249 if (size < kvm_dirty_ring_get_rsvd_entries() *
4250 sizeof(struct kvm_dirty_gfn) || size < PAGE_SIZE)
4253 if (size > KVM_DIRTY_RING_MAX_ENTRIES *
4254 sizeof(struct kvm_dirty_gfn))
4257 /* We only allow it to set once */
4258 if (kvm->dirty_ring_size)
4261 mutex_lock(&kvm->lock);
4263 if (kvm->created_vcpus) {
4264 /* We don't allow to change this value after vcpu created */
4267 kvm->dirty_ring_size = size;
4271 mutex_unlock(&kvm->lock);
4275 static int kvm_vm_ioctl_reset_dirty_pages(struct kvm *kvm)
4278 struct kvm_vcpu *vcpu;
4281 if (!kvm->dirty_ring_size)
4284 mutex_lock(&kvm->slots_lock);
4286 kvm_for_each_vcpu(i, vcpu, kvm)
4287 cleared += kvm_dirty_ring_reset(vcpu->kvm, &vcpu->dirty_ring);
4289 mutex_unlock(&kvm->slots_lock);
4292 kvm_flush_remote_tlbs(kvm);
4297 int __attribute__((weak)) kvm_vm_ioctl_enable_cap(struct kvm *kvm,
4298 struct kvm_enable_cap *cap)
4303 static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm,
4304 struct kvm_enable_cap *cap)
4307 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4308 case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2: {
4309 u64 allowed_options = KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE;
4311 if (cap->args[0] & KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE)
4312 allowed_options = KVM_DIRTY_LOG_MANUAL_CAPS;
4314 if (cap->flags || (cap->args[0] & ~allowed_options))
4316 kvm->manual_dirty_log_protect = cap->args[0];
4320 case KVM_CAP_HALT_POLL: {
4321 if (cap->flags || cap->args[0] != (unsigned int)cap->args[0])
4324 kvm->max_halt_poll_ns = cap->args[0];
4327 case KVM_CAP_DIRTY_LOG_RING:
4328 return kvm_vm_ioctl_enable_dirty_log_ring(kvm, cap->args[0]);
4330 return kvm_vm_ioctl_enable_cap(kvm, cap);
4334 static ssize_t kvm_vm_stats_read(struct file *file, char __user *user_buffer,
4335 size_t size, loff_t *offset)
4337 struct kvm *kvm = file->private_data;
4339 return kvm_stats_read(kvm->stats_id, &kvm_vm_stats_header,
4340 &kvm_vm_stats_desc[0], &kvm->stat,
4341 sizeof(kvm->stat), user_buffer, size, offset);
4344 static const struct file_operations kvm_vm_stats_fops = {
4345 .read = kvm_vm_stats_read,
4346 .llseek = noop_llseek,
4349 static int kvm_vm_ioctl_get_stats_fd(struct kvm *kvm)
4354 fd = get_unused_fd_flags(O_CLOEXEC);
4358 file = anon_inode_getfile("kvm-vm-stats",
4359 &kvm_vm_stats_fops, kvm, O_RDONLY);
4362 return PTR_ERR(file);
4364 file->f_mode |= FMODE_PREAD;
4365 fd_install(fd, file);
4370 static long kvm_vm_ioctl(struct file *filp,
4371 unsigned int ioctl, unsigned long arg)
4373 struct kvm *kvm = filp->private_data;
4374 void __user *argp = (void __user *)arg;
4377 if (kvm->mm != current->mm || kvm->vm_bugged)
4380 case KVM_CREATE_VCPU:
4381 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
4383 case KVM_ENABLE_CAP: {
4384 struct kvm_enable_cap cap;
4387 if (copy_from_user(&cap, argp, sizeof(cap)))
4389 r = kvm_vm_ioctl_enable_cap_generic(kvm, &cap);
4392 case KVM_SET_USER_MEMORY_REGION: {
4393 struct kvm_userspace_memory_region kvm_userspace_mem;
4396 if (copy_from_user(&kvm_userspace_mem, argp,
4397 sizeof(kvm_userspace_mem)))
4400 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
4403 case KVM_GET_DIRTY_LOG: {
4404 struct kvm_dirty_log log;
4407 if (copy_from_user(&log, argp, sizeof(log)))
4409 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
4412 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4413 case KVM_CLEAR_DIRTY_LOG: {
4414 struct kvm_clear_dirty_log log;
4417 if (copy_from_user(&log, argp, sizeof(log)))
4419 r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
4423 #ifdef CONFIG_KVM_MMIO
4424 case KVM_REGISTER_COALESCED_MMIO: {
4425 struct kvm_coalesced_mmio_zone zone;
4428 if (copy_from_user(&zone, argp, sizeof(zone)))
4430 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
4433 case KVM_UNREGISTER_COALESCED_MMIO: {
4434 struct kvm_coalesced_mmio_zone zone;
4437 if (copy_from_user(&zone, argp, sizeof(zone)))
4439 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
4444 struct kvm_irqfd data;
4447 if (copy_from_user(&data, argp, sizeof(data)))
4449 r = kvm_irqfd(kvm, &data);
4452 case KVM_IOEVENTFD: {
4453 struct kvm_ioeventfd data;
4456 if (copy_from_user(&data, argp, sizeof(data)))
4458 r = kvm_ioeventfd(kvm, &data);
4461 #ifdef CONFIG_HAVE_KVM_MSI
4462 case KVM_SIGNAL_MSI: {
4466 if (copy_from_user(&msi, argp, sizeof(msi)))
4468 r = kvm_send_userspace_msi(kvm, &msi);
4472 #ifdef __KVM_HAVE_IRQ_LINE
4473 case KVM_IRQ_LINE_STATUS:
4474 case KVM_IRQ_LINE: {
4475 struct kvm_irq_level irq_event;
4478 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
4481 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
4482 ioctl == KVM_IRQ_LINE_STATUS);
4487 if (ioctl == KVM_IRQ_LINE_STATUS) {
4488 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
4496 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
4497 case KVM_SET_GSI_ROUTING: {
4498 struct kvm_irq_routing routing;
4499 struct kvm_irq_routing __user *urouting;
4500 struct kvm_irq_routing_entry *entries = NULL;
4503 if (copy_from_user(&routing, argp, sizeof(routing)))
4506 if (!kvm_arch_can_set_irq_routing(kvm))
4508 if (routing.nr > KVM_MAX_IRQ_ROUTES)
4514 entries = vmemdup_user(urouting->entries,
4515 array_size(sizeof(*entries),
4517 if (IS_ERR(entries)) {
4518 r = PTR_ERR(entries);
4522 r = kvm_set_irq_routing(kvm, entries, routing.nr,
4527 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
4528 case KVM_CREATE_DEVICE: {
4529 struct kvm_create_device cd;
4532 if (copy_from_user(&cd, argp, sizeof(cd)))
4535 r = kvm_ioctl_create_device(kvm, &cd);
4540 if (copy_to_user(argp, &cd, sizeof(cd)))
4546 case KVM_CHECK_EXTENSION:
4547 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
4549 case KVM_RESET_DIRTY_RINGS:
4550 r = kvm_vm_ioctl_reset_dirty_pages(kvm);
4552 case KVM_GET_STATS_FD:
4553 r = kvm_vm_ioctl_get_stats_fd(kvm);
4556 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
4562 #ifdef CONFIG_KVM_COMPAT
4563 struct compat_kvm_dirty_log {
4567 compat_uptr_t dirty_bitmap; /* one bit per page */
4572 struct compat_kvm_clear_dirty_log {
4577 compat_uptr_t dirty_bitmap; /* one bit per page */
4582 static long kvm_vm_compat_ioctl(struct file *filp,
4583 unsigned int ioctl, unsigned long arg)
4585 struct kvm *kvm = filp->private_data;
4588 if (kvm->mm != current->mm || kvm->vm_bugged)
4591 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4592 case KVM_CLEAR_DIRTY_LOG: {
4593 struct compat_kvm_clear_dirty_log compat_log;
4594 struct kvm_clear_dirty_log log;
4596 if (copy_from_user(&compat_log, (void __user *)arg,
4597 sizeof(compat_log)))
4599 log.slot = compat_log.slot;
4600 log.num_pages = compat_log.num_pages;
4601 log.first_page = compat_log.first_page;
4602 log.padding2 = compat_log.padding2;
4603 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
4605 r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
4609 case KVM_GET_DIRTY_LOG: {
4610 struct compat_kvm_dirty_log compat_log;
4611 struct kvm_dirty_log log;
4613 if (copy_from_user(&compat_log, (void __user *)arg,
4614 sizeof(compat_log)))
4616 log.slot = compat_log.slot;
4617 log.padding1 = compat_log.padding1;
4618 log.padding2 = compat_log.padding2;
4619 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
4621 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
4625 r = kvm_vm_ioctl(filp, ioctl, arg);
4631 static struct file_operations kvm_vm_fops = {
4632 .release = kvm_vm_release,
4633 .unlocked_ioctl = kvm_vm_ioctl,
4634 .llseek = noop_llseek,
4635 KVM_COMPAT(kvm_vm_compat_ioctl),
4638 bool file_is_kvm(struct file *file)
4640 return file && file->f_op == &kvm_vm_fops;
4642 EXPORT_SYMBOL_GPL(file_is_kvm);
4644 static int kvm_dev_ioctl_create_vm(unsigned long type)
4650 kvm = kvm_create_vm(type);
4652 return PTR_ERR(kvm);
4653 #ifdef CONFIG_KVM_MMIO
4654 r = kvm_coalesced_mmio_init(kvm);
4658 r = get_unused_fd_flags(O_CLOEXEC);
4662 snprintf(kvm->stats_id, sizeof(kvm->stats_id),
4663 "kvm-%d", task_pid_nr(current));
4665 file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
4673 * Don't call kvm_put_kvm anymore at this point; file->f_op is
4674 * already set, with ->release() being kvm_vm_release(). In error
4675 * cases it will be called by the final fput(file) and will take
4676 * care of doing kvm_put_kvm(kvm).
4678 if (kvm_create_vm_debugfs(kvm, r) < 0) {
4683 kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm);
4685 fd_install(r, file);
4693 static long kvm_dev_ioctl(struct file *filp,
4694 unsigned int ioctl, unsigned long arg)
4699 case KVM_GET_API_VERSION:
4702 r = KVM_API_VERSION;
4705 r = kvm_dev_ioctl_create_vm(arg);
4707 case KVM_CHECK_EXTENSION:
4708 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
4710 case KVM_GET_VCPU_MMAP_SIZE:
4713 r = PAGE_SIZE; /* struct kvm_run */
4715 r += PAGE_SIZE; /* pio data page */
4717 #ifdef CONFIG_KVM_MMIO
4718 r += PAGE_SIZE; /* coalesced mmio ring page */
4721 case KVM_TRACE_ENABLE:
4722 case KVM_TRACE_PAUSE:
4723 case KVM_TRACE_DISABLE:
4727 return kvm_arch_dev_ioctl(filp, ioctl, arg);
4733 static struct file_operations kvm_chardev_ops = {
4734 .unlocked_ioctl = kvm_dev_ioctl,
4735 .llseek = noop_llseek,
4736 KVM_COMPAT(kvm_dev_ioctl),
4739 static struct miscdevice kvm_dev = {
4745 static void hardware_enable_nolock(void *junk)
4747 int cpu = raw_smp_processor_id();
4750 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
4753 cpumask_set_cpu(cpu, cpus_hardware_enabled);
4755 r = kvm_arch_hardware_enable();
4758 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
4759 atomic_inc(&hardware_enable_failed);
4760 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
4764 static int kvm_starting_cpu(unsigned int cpu)
4766 raw_spin_lock(&kvm_count_lock);
4767 if (kvm_usage_count)
4768 hardware_enable_nolock(NULL);
4769 raw_spin_unlock(&kvm_count_lock);
4773 static void hardware_disable_nolock(void *junk)
4775 int cpu = raw_smp_processor_id();
4777 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
4779 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
4780 kvm_arch_hardware_disable();
4783 static int kvm_dying_cpu(unsigned int cpu)
4785 raw_spin_lock(&kvm_count_lock);
4786 if (kvm_usage_count)
4787 hardware_disable_nolock(NULL);
4788 raw_spin_unlock(&kvm_count_lock);
4792 static void hardware_disable_all_nolock(void)
4794 BUG_ON(!kvm_usage_count);
4797 if (!kvm_usage_count)
4798 on_each_cpu(hardware_disable_nolock, NULL, 1);
4801 static void hardware_disable_all(void)
4803 raw_spin_lock(&kvm_count_lock);
4804 hardware_disable_all_nolock();
4805 raw_spin_unlock(&kvm_count_lock);
4808 static int hardware_enable_all(void)
4812 raw_spin_lock(&kvm_count_lock);
4815 if (kvm_usage_count == 1) {
4816 atomic_set(&hardware_enable_failed, 0);
4817 on_each_cpu(hardware_enable_nolock, NULL, 1);
4819 if (atomic_read(&hardware_enable_failed)) {
4820 hardware_disable_all_nolock();
4825 raw_spin_unlock(&kvm_count_lock);
4830 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
4834 * Some (well, at least mine) BIOSes hang on reboot if
4837 * And Intel TXT required VMX off for all cpu when system shutdown.
4839 pr_info("kvm: exiting hardware virtualization\n");
4840 kvm_rebooting = true;
4841 on_each_cpu(hardware_disable_nolock, NULL, 1);
4845 static struct notifier_block kvm_reboot_notifier = {
4846 .notifier_call = kvm_reboot,
4850 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
4854 for (i = 0; i < bus->dev_count; i++) {
4855 struct kvm_io_device *pos = bus->range[i].dev;
4857 kvm_iodevice_destructor(pos);
4862 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
4863 const struct kvm_io_range *r2)
4865 gpa_t addr1 = r1->addr;
4866 gpa_t addr2 = r2->addr;
4871 /* If r2->len == 0, match the exact address. If r2->len != 0,
4872 * accept any overlapping write. Any order is acceptable for
4873 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
4874 * we process all of them.
4887 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
4889 return kvm_io_bus_cmp(p1, p2);
4892 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
4893 gpa_t addr, int len)
4895 struct kvm_io_range *range, key;
4898 key = (struct kvm_io_range) {
4903 range = bsearch(&key, bus->range, bus->dev_count,
4904 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
4908 off = range - bus->range;
4910 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
4916 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
4917 struct kvm_io_range *range, const void *val)
4921 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
4925 while (idx < bus->dev_count &&
4926 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
4927 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
4936 /* kvm_io_bus_write - called under kvm->slots_lock */
4937 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
4938 int len, const void *val)
4940 struct kvm_io_bus *bus;
4941 struct kvm_io_range range;
4944 range = (struct kvm_io_range) {
4949 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
4952 r = __kvm_io_bus_write(vcpu, bus, &range, val);
4953 return r < 0 ? r : 0;
4955 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
4957 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
4958 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
4959 gpa_t addr, int len, const void *val, long cookie)
4961 struct kvm_io_bus *bus;
4962 struct kvm_io_range range;
4964 range = (struct kvm_io_range) {
4969 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
4973 /* First try the device referenced by cookie. */
4974 if ((cookie >= 0) && (cookie < bus->dev_count) &&
4975 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
4976 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
4981 * cookie contained garbage; fall back to search and return the
4982 * correct cookie value.
4984 return __kvm_io_bus_write(vcpu, bus, &range, val);
4987 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
4988 struct kvm_io_range *range, void *val)
4992 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
4996 while (idx < bus->dev_count &&
4997 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
4998 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
5007 /* kvm_io_bus_read - called under kvm->slots_lock */
5008 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
5011 struct kvm_io_bus *bus;
5012 struct kvm_io_range range;
5015 range = (struct kvm_io_range) {
5020 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
5023 r = __kvm_io_bus_read(vcpu, bus, &range, val);
5024 return r < 0 ? r : 0;
5027 /* Caller must hold slots_lock. */
5028 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
5029 int len, struct kvm_io_device *dev)
5032 struct kvm_io_bus *new_bus, *bus;
5033 struct kvm_io_range range;
5035 bus = kvm_get_bus(kvm, bus_idx);
5039 /* exclude ioeventfd which is limited by maximum fd */
5040 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
5043 new_bus = kmalloc(struct_size(bus, range, bus->dev_count + 1),
5044 GFP_KERNEL_ACCOUNT);
5048 range = (struct kvm_io_range) {
5054 for (i = 0; i < bus->dev_count; i++)
5055 if (kvm_io_bus_cmp(&bus->range[i], &range) > 0)
5058 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
5059 new_bus->dev_count++;
5060 new_bus->range[i] = range;
5061 memcpy(new_bus->range + i + 1, bus->range + i,
5062 (bus->dev_count - i) * sizeof(struct kvm_io_range));
5063 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
5064 synchronize_srcu_expedited(&kvm->srcu);
5070 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
5071 struct kvm_io_device *dev)
5074 struct kvm_io_bus *new_bus, *bus;
5076 lockdep_assert_held(&kvm->slots_lock);
5078 bus = kvm_get_bus(kvm, bus_idx);
5082 for (i = 0; i < bus->dev_count; i++) {
5083 if (bus->range[i].dev == dev) {
5088 if (i == bus->dev_count)
5091 new_bus = kmalloc(struct_size(bus, range, bus->dev_count - 1),
5092 GFP_KERNEL_ACCOUNT);
5094 memcpy(new_bus, bus, struct_size(bus, range, i));
5095 new_bus->dev_count--;
5096 memcpy(new_bus->range + i, bus->range + i + 1,
5097 flex_array_size(new_bus, range, new_bus->dev_count - i));
5100 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
5101 synchronize_srcu_expedited(&kvm->srcu);
5103 /* Destroy the old bus _after_ installing the (null) bus. */
5105 pr_err("kvm: failed to shrink bus, removing it completely\n");
5106 for (j = 0; j < bus->dev_count; j++) {
5109 kvm_iodevice_destructor(bus->range[j].dev);
5114 return new_bus ? 0 : -ENOMEM;
5117 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
5120 struct kvm_io_bus *bus;
5121 int dev_idx, srcu_idx;
5122 struct kvm_io_device *iodev = NULL;
5124 srcu_idx = srcu_read_lock(&kvm->srcu);
5126 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
5130 dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
5134 iodev = bus->range[dev_idx].dev;
5137 srcu_read_unlock(&kvm->srcu, srcu_idx);
5141 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
5143 static int kvm_debugfs_open(struct inode *inode, struct file *file,
5144 int (*get)(void *, u64 *), int (*set)(void *, u64),
5147 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
5151 * The debugfs files are a reference to the kvm struct which
5152 * is still valid when kvm_destroy_vm is called. kvm_get_kvm_safe
5153 * avoids the race between open and the removal of the debugfs directory.
5155 if (!kvm_get_kvm_safe(stat_data->kvm))
5158 if (simple_attr_open(inode, file, get,
5159 kvm_stats_debugfs_mode(stat_data->desc) & 0222
5162 kvm_put_kvm(stat_data->kvm);
5169 static int kvm_debugfs_release(struct inode *inode, struct file *file)
5171 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
5174 simple_attr_release(inode, file);
5175 kvm_put_kvm(stat_data->kvm);
5180 static int kvm_get_stat_per_vm(struct kvm *kvm, size_t offset, u64 *val)
5182 *val = *(u64 *)((void *)(&kvm->stat) + offset);
5187 static int kvm_clear_stat_per_vm(struct kvm *kvm, size_t offset)
5189 *(u64 *)((void *)(&kvm->stat) + offset) = 0;
5194 static int kvm_get_stat_per_vcpu(struct kvm *kvm, size_t offset, u64 *val)
5197 struct kvm_vcpu *vcpu;
5201 kvm_for_each_vcpu(i, vcpu, kvm)
5202 *val += *(u64 *)((void *)(&vcpu->stat) + offset);
5207 static int kvm_clear_stat_per_vcpu(struct kvm *kvm, size_t offset)
5210 struct kvm_vcpu *vcpu;
5212 kvm_for_each_vcpu(i, vcpu, kvm)
5213 *(u64 *)((void *)(&vcpu->stat) + offset) = 0;
5218 static int kvm_stat_data_get(void *data, u64 *val)
5221 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
5223 switch (stat_data->kind) {
5225 r = kvm_get_stat_per_vm(stat_data->kvm,
5226 stat_data->desc->desc.offset, val);
5229 r = kvm_get_stat_per_vcpu(stat_data->kvm,
5230 stat_data->desc->desc.offset, val);
5237 static int kvm_stat_data_clear(void *data, u64 val)
5240 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
5245 switch (stat_data->kind) {
5247 r = kvm_clear_stat_per_vm(stat_data->kvm,
5248 stat_data->desc->desc.offset);
5251 r = kvm_clear_stat_per_vcpu(stat_data->kvm,
5252 stat_data->desc->desc.offset);
5259 static int kvm_stat_data_open(struct inode *inode, struct file *file)
5261 __simple_attr_check_format("%llu\n", 0ull);
5262 return kvm_debugfs_open(inode, file, kvm_stat_data_get,
5263 kvm_stat_data_clear, "%llu\n");
5266 static const struct file_operations stat_fops_per_vm = {
5267 .owner = THIS_MODULE,
5268 .open = kvm_stat_data_open,
5269 .release = kvm_debugfs_release,
5270 .read = simple_attr_read,
5271 .write = simple_attr_write,
5272 .llseek = no_llseek,
5275 static int vm_stat_get(void *_offset, u64 *val)
5277 unsigned offset = (long)_offset;
5282 mutex_lock(&kvm_lock);
5283 list_for_each_entry(kvm, &vm_list, vm_list) {
5284 kvm_get_stat_per_vm(kvm, offset, &tmp_val);
5287 mutex_unlock(&kvm_lock);
5291 static int vm_stat_clear(void *_offset, u64 val)
5293 unsigned offset = (long)_offset;
5299 mutex_lock(&kvm_lock);
5300 list_for_each_entry(kvm, &vm_list, vm_list) {
5301 kvm_clear_stat_per_vm(kvm, offset);
5303 mutex_unlock(&kvm_lock);
5308 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n");
5309 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_readonly_fops, vm_stat_get, NULL, "%llu\n");
5311 static int vcpu_stat_get(void *_offset, u64 *val)
5313 unsigned offset = (long)_offset;
5318 mutex_lock(&kvm_lock);
5319 list_for_each_entry(kvm, &vm_list, vm_list) {
5320 kvm_get_stat_per_vcpu(kvm, offset, &tmp_val);
5323 mutex_unlock(&kvm_lock);
5327 static int vcpu_stat_clear(void *_offset, u64 val)
5329 unsigned offset = (long)_offset;
5335 mutex_lock(&kvm_lock);
5336 list_for_each_entry(kvm, &vm_list, vm_list) {
5337 kvm_clear_stat_per_vcpu(kvm, offset);
5339 mutex_unlock(&kvm_lock);
5344 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear,
5346 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_readonly_fops, vcpu_stat_get, NULL, "%llu\n");
5348 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm)
5350 struct kobj_uevent_env *env;
5351 unsigned long long created, active;
5353 if (!kvm_dev.this_device || !kvm)
5356 mutex_lock(&kvm_lock);
5357 if (type == KVM_EVENT_CREATE_VM) {
5358 kvm_createvm_count++;
5360 } else if (type == KVM_EVENT_DESTROY_VM) {
5363 created = kvm_createvm_count;
5364 active = kvm_active_vms;
5365 mutex_unlock(&kvm_lock);
5367 env = kzalloc(sizeof(*env), GFP_KERNEL_ACCOUNT);
5371 add_uevent_var(env, "CREATED=%llu", created);
5372 add_uevent_var(env, "COUNT=%llu", active);
5374 if (type == KVM_EVENT_CREATE_VM) {
5375 add_uevent_var(env, "EVENT=create");
5376 kvm->userspace_pid = task_pid_nr(current);
5377 } else if (type == KVM_EVENT_DESTROY_VM) {
5378 add_uevent_var(env, "EVENT=destroy");
5380 add_uevent_var(env, "PID=%d", kvm->userspace_pid);
5382 if (!IS_ERR(kvm->debugfs_dentry)) {
5383 char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL_ACCOUNT);
5386 tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX);
5388 add_uevent_var(env, "STATS_PATH=%s", tmp);
5392 /* no need for checks, since we are adding at most only 5 keys */
5393 env->envp[env->envp_idx++] = NULL;
5394 kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp);
5398 static void kvm_init_debug(void)
5400 const struct file_operations *fops;
5401 const struct _kvm_stats_desc *pdesc;
5404 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
5406 for (i = 0; i < kvm_vm_stats_header.num_desc; ++i) {
5407 pdesc = &kvm_vm_stats_desc[i];
5408 if (kvm_stats_debugfs_mode(pdesc) & 0222)
5409 fops = &vm_stat_fops;
5411 fops = &vm_stat_readonly_fops;
5412 debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
5414 (void *)(long)pdesc->desc.offset, fops);
5417 for (i = 0; i < kvm_vcpu_stats_header.num_desc; ++i) {
5418 pdesc = &kvm_vcpu_stats_desc[i];
5419 if (kvm_stats_debugfs_mode(pdesc) & 0222)
5420 fops = &vcpu_stat_fops;
5422 fops = &vcpu_stat_readonly_fops;
5423 debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
5425 (void *)(long)pdesc->desc.offset, fops);
5429 static int kvm_suspend(void)
5431 if (kvm_usage_count)
5432 hardware_disable_nolock(NULL);
5436 static void kvm_resume(void)
5438 if (kvm_usage_count) {
5439 lockdep_assert_not_held(&kvm_count_lock);
5440 hardware_enable_nolock(NULL);
5444 static struct syscore_ops kvm_syscore_ops = {
5445 .suspend = kvm_suspend,
5446 .resume = kvm_resume,
5450 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
5452 return container_of(pn, struct kvm_vcpu, preempt_notifier);
5455 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
5457 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
5459 WRITE_ONCE(vcpu->preempted, false);
5460 WRITE_ONCE(vcpu->ready, false);
5462 __this_cpu_write(kvm_running_vcpu, vcpu);
5463 kvm_arch_sched_in(vcpu, cpu);
5464 kvm_arch_vcpu_load(vcpu, cpu);
5467 static void kvm_sched_out(struct preempt_notifier *pn,
5468 struct task_struct *next)
5470 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
5472 if (current->on_rq) {
5473 WRITE_ONCE(vcpu->preempted, true);
5474 WRITE_ONCE(vcpu->ready, true);
5476 kvm_arch_vcpu_put(vcpu);
5477 __this_cpu_write(kvm_running_vcpu, NULL);
5481 * kvm_get_running_vcpu - get the vcpu running on the current CPU.
5483 * We can disable preemption locally around accessing the per-CPU variable,
5484 * and use the resolved vcpu pointer after enabling preemption again,
5485 * because even if the current thread is migrated to another CPU, reading
5486 * the per-CPU value later will give us the same value as we update the
5487 * per-CPU variable in the preempt notifier handlers.
5489 struct kvm_vcpu *kvm_get_running_vcpu(void)
5491 struct kvm_vcpu *vcpu;
5494 vcpu = __this_cpu_read(kvm_running_vcpu);
5499 EXPORT_SYMBOL_GPL(kvm_get_running_vcpu);
5502 * kvm_get_running_vcpus - get the per-CPU array of currently running vcpus.
5504 struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void)
5506 return &kvm_running_vcpu;
5509 struct kvm_cpu_compat_check {
5514 static void check_processor_compat(void *data)
5516 struct kvm_cpu_compat_check *c = data;
5518 *c->ret = kvm_arch_check_processor_compat(c->opaque);
5521 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
5522 struct module *module)
5524 struct kvm_cpu_compat_check c;
5528 r = kvm_arch_init(opaque);
5533 * kvm_arch_init makes sure there's at most one caller
5534 * for architectures that support multiple implementations,
5535 * like intel and amd on x86.
5536 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
5537 * conflicts in case kvm is already setup for another implementation.
5539 r = kvm_irqfd_init();
5543 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
5548 r = kvm_arch_hardware_setup(opaque);
5554 for_each_online_cpu(cpu) {
5555 smp_call_function_single(cpu, check_processor_compat, &c, 1);
5560 r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting",
5561 kvm_starting_cpu, kvm_dying_cpu);
5564 register_reboot_notifier(&kvm_reboot_notifier);
5566 /* A kmem cache lets us meet the alignment requirements of fx_save. */
5568 vcpu_align = __alignof__(struct kvm_vcpu);
5570 kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align,
5572 offsetof(struct kvm_vcpu, arch),
5573 offsetofend(struct kvm_vcpu, stats_id)
5574 - offsetof(struct kvm_vcpu, arch),
5576 if (!kvm_vcpu_cache) {
5581 r = kvm_async_pf_init();
5585 kvm_chardev_ops.owner = module;
5586 kvm_vm_fops.owner = module;
5587 kvm_vcpu_fops.owner = module;
5589 r = misc_register(&kvm_dev);
5591 pr_err("kvm: misc device register failed\n");
5595 register_syscore_ops(&kvm_syscore_ops);
5597 kvm_preempt_ops.sched_in = kvm_sched_in;
5598 kvm_preempt_ops.sched_out = kvm_sched_out;
5602 r = kvm_vfio_ops_init();
5608 kvm_async_pf_deinit();
5610 kmem_cache_destroy(kvm_vcpu_cache);
5612 unregister_reboot_notifier(&kvm_reboot_notifier);
5613 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
5615 kvm_arch_hardware_unsetup();
5617 free_cpumask_var(cpus_hardware_enabled);
5625 EXPORT_SYMBOL_GPL(kvm_init);
5629 debugfs_remove_recursive(kvm_debugfs_dir);
5630 misc_deregister(&kvm_dev);
5631 kmem_cache_destroy(kvm_vcpu_cache);
5632 kvm_async_pf_deinit();
5633 unregister_syscore_ops(&kvm_syscore_ops);
5634 unregister_reboot_notifier(&kvm_reboot_notifier);
5635 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
5636 on_each_cpu(hardware_disable_nolock, NULL, 1);
5637 kvm_arch_hardware_unsetup();
5640 free_cpumask_var(cpus_hardware_enabled);
5641 kvm_vfio_ops_exit();
5643 EXPORT_SYMBOL_GPL(kvm_exit);
5645 struct kvm_vm_worker_thread_context {
5647 struct task_struct *parent;
5648 struct completion init_done;
5649 kvm_vm_thread_fn_t thread_fn;
5654 static int kvm_vm_worker_thread(void *context)
5657 * The init_context is allocated on the stack of the parent thread, so
5658 * we have to locally copy anything that is needed beyond initialization
5660 struct kvm_vm_worker_thread_context *init_context = context;
5661 struct kvm *kvm = init_context->kvm;
5662 kvm_vm_thread_fn_t thread_fn = init_context->thread_fn;
5663 uintptr_t data = init_context->data;
5666 err = kthread_park(current);
5667 /* kthread_park(current) is never supposed to return an error */
5672 err = cgroup_attach_task_all(init_context->parent, current);
5674 kvm_err("%s: cgroup_attach_task_all failed with err %d\n",
5679 set_user_nice(current, task_nice(init_context->parent));
5682 init_context->err = err;
5683 complete(&init_context->init_done);
5684 init_context = NULL;
5689 /* Wait to be woken up by the spawner before proceeding. */
5692 if (!kthread_should_stop())
5693 err = thread_fn(kvm, data);
5698 int kvm_vm_create_worker_thread(struct kvm *kvm, kvm_vm_thread_fn_t thread_fn,
5699 uintptr_t data, const char *name,
5700 struct task_struct **thread_ptr)
5702 struct kvm_vm_worker_thread_context init_context = {};
5703 struct task_struct *thread;
5706 init_context.kvm = kvm;
5707 init_context.parent = current;
5708 init_context.thread_fn = thread_fn;
5709 init_context.data = data;
5710 init_completion(&init_context.init_done);
5712 thread = kthread_run(kvm_vm_worker_thread, &init_context,
5713 "%s-%d", name, task_pid_nr(current));
5715 return PTR_ERR(thread);
5717 /* kthread_run is never supposed to return NULL */
5718 WARN_ON(thread == NULL);
5720 wait_for_completion(&init_context.init_done);
5722 if (!init_context.err)
5723 *thread_ptr = thread;
5725 return init_context.err;