2 * Kernel-based Virtual Machine driver for Linux
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
7 * Copyright (C) 2006 Qumranet, Inc.
8 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
11 * Avi Kivity <avi@qumranet.com>
12 * Yaniv Kamay <yaniv@qumranet.com>
14 * This work is licensed under the terms of the GNU GPL, version 2. See
15 * the COPYING file in the top-level directory.
19 #include <kvm/iodev.h>
21 #include <linux/kvm_host.h>
22 #include <linux/kvm.h>
23 #include <linux/module.h>
24 #include <linux/errno.h>
25 #include <linux/percpu.h>
27 #include <linux/miscdevice.h>
28 #include <linux/vmalloc.h>
29 #include <linux/reboot.h>
30 #include <linux/debugfs.h>
31 #include <linux/highmem.h>
32 #include <linux/file.h>
33 #include <linux/syscore_ops.h>
34 #include <linux/cpu.h>
35 #include <linux/sched.h>
36 #include <linux/cpumask.h>
37 #include <linux/smp.h>
38 #include <linux/anon_inodes.h>
39 #include <linux/profile.h>
40 #include <linux/kvm_para.h>
41 #include <linux/pagemap.h>
42 #include <linux/mman.h>
43 #include <linux/swap.h>
44 #include <linux/bitops.h>
45 #include <linux/spinlock.h>
46 #include <linux/compat.h>
47 #include <linux/srcu.h>
48 #include <linux/hugetlb.h>
49 #include <linux/slab.h>
50 #include <linux/sort.h>
51 #include <linux/bsearch.h>
53 #include <asm/processor.h>
55 #include <asm/ioctl.h>
56 #include <asm/uaccess.h>
57 #include <asm/pgtable.h>
59 #include "coalesced_mmio.h"
63 #define CREATE_TRACE_POINTS
64 #include <trace/events/kvm.h>
66 MODULE_AUTHOR("Qumranet");
67 MODULE_LICENSE("GPL");
69 /* Architectures should define their poll value according to the halt latency */
70 static unsigned int halt_poll_ns = KVM_HALT_POLL_NS_DEFAULT;
71 module_param(halt_poll_ns, uint, S_IRUGO | S_IWUSR);
73 /* Default doubles per-vcpu halt_poll_ns. */
74 static unsigned int halt_poll_ns_grow = 2;
75 module_param(halt_poll_ns_grow, uint, S_IRUGO | S_IWUSR);
77 /* Default resets per-vcpu halt_poll_ns . */
78 static unsigned int halt_poll_ns_shrink;
79 module_param(halt_poll_ns_shrink, uint, S_IRUGO | S_IWUSR);
84 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
87 DEFINE_SPINLOCK(kvm_lock);
88 static DEFINE_RAW_SPINLOCK(kvm_count_lock);
91 static cpumask_var_t cpus_hardware_enabled;
92 static int kvm_usage_count;
93 static atomic_t hardware_enable_failed;
95 struct kmem_cache *kvm_vcpu_cache;
96 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
98 static __read_mostly struct preempt_ops kvm_preempt_ops;
100 struct dentry *kvm_debugfs_dir;
101 EXPORT_SYMBOL_GPL(kvm_debugfs_dir);
103 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
105 #ifdef CONFIG_KVM_COMPAT
106 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
109 static int hardware_enable_all(void);
110 static void hardware_disable_all(void);
112 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
114 static void kvm_release_pfn_dirty(kvm_pfn_t pfn);
115 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot, gfn_t gfn);
117 __visible bool kvm_rebooting;
118 EXPORT_SYMBOL_GPL(kvm_rebooting);
120 static bool largepages_enabled = true;
122 bool kvm_is_reserved_pfn(kvm_pfn_t pfn)
125 return PageReserved(pfn_to_page(pfn));
131 * Switches to specified vcpu, until a matching vcpu_put()
133 int vcpu_load(struct kvm_vcpu *vcpu)
137 if (mutex_lock_killable(&vcpu->mutex))
140 preempt_notifier_register(&vcpu->preempt_notifier);
141 kvm_arch_vcpu_load(vcpu, cpu);
146 void vcpu_put(struct kvm_vcpu *vcpu)
149 kvm_arch_vcpu_put(vcpu);
150 preempt_notifier_unregister(&vcpu->preempt_notifier);
152 mutex_unlock(&vcpu->mutex);
155 static void ack_flush(void *_completed)
159 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
164 struct kvm_vcpu *vcpu;
166 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
169 kvm_for_each_vcpu(i, vcpu, kvm) {
170 kvm_make_request(req, vcpu);
173 /* Set ->requests bit before we read ->mode */
176 if (cpus != NULL && cpu != -1 && cpu != me &&
177 kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
178 cpumask_set_cpu(cpu, cpus);
180 if (unlikely(cpus == NULL))
181 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
182 else if (!cpumask_empty(cpus))
183 smp_call_function_many(cpus, ack_flush, NULL, 1);
187 free_cpumask_var(cpus);
191 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
192 void kvm_flush_remote_tlbs(struct kvm *kvm)
194 long dirty_count = kvm->tlbs_dirty;
197 if (kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
198 ++kvm->stat.remote_tlb_flush;
199 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
201 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
204 void kvm_reload_remote_mmus(struct kvm *kvm)
206 kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
209 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
214 mutex_init(&vcpu->mutex);
219 init_swait_queue_head(&vcpu->wq);
220 kvm_async_pf_vcpu_init(vcpu);
223 INIT_LIST_HEAD(&vcpu->blocked_vcpu_list);
225 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
230 vcpu->run = page_address(page);
232 kvm_vcpu_set_in_spin_loop(vcpu, false);
233 kvm_vcpu_set_dy_eligible(vcpu, false);
234 vcpu->preempted = false;
236 r = kvm_arch_vcpu_init(vcpu);
242 free_page((unsigned long)vcpu->run);
246 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
248 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
251 kvm_arch_vcpu_uninit(vcpu);
252 free_page((unsigned long)vcpu->run);
254 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
256 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
257 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
259 return container_of(mn, struct kvm, mmu_notifier);
262 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
263 struct mm_struct *mm,
264 unsigned long address)
266 struct kvm *kvm = mmu_notifier_to_kvm(mn);
267 int need_tlb_flush, idx;
270 * When ->invalidate_page runs, the linux pte has been zapped
271 * already but the page is still allocated until
272 * ->invalidate_page returns. So if we increase the sequence
273 * here the kvm page fault will notice if the spte can't be
274 * established because the page is going to be freed. If
275 * instead the kvm page fault establishes the spte before
276 * ->invalidate_page runs, kvm_unmap_hva will release it
279 * The sequence increase only need to be seen at spin_unlock
280 * time, and not at spin_lock time.
282 * Increasing the sequence after the spin_unlock would be
283 * unsafe because the kvm page fault could then establish the
284 * pte after kvm_unmap_hva returned, without noticing the page
285 * is going to be freed.
287 idx = srcu_read_lock(&kvm->srcu);
288 spin_lock(&kvm->mmu_lock);
290 kvm->mmu_notifier_seq++;
291 need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
292 /* we've to flush the tlb before the pages can be freed */
294 kvm_flush_remote_tlbs(kvm);
296 spin_unlock(&kvm->mmu_lock);
298 kvm_arch_mmu_notifier_invalidate_page(kvm, address);
300 srcu_read_unlock(&kvm->srcu, idx);
303 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
304 struct mm_struct *mm,
305 unsigned long address,
308 struct kvm *kvm = mmu_notifier_to_kvm(mn);
311 idx = srcu_read_lock(&kvm->srcu);
312 spin_lock(&kvm->mmu_lock);
313 kvm->mmu_notifier_seq++;
314 kvm_set_spte_hva(kvm, address, pte);
315 spin_unlock(&kvm->mmu_lock);
316 srcu_read_unlock(&kvm->srcu, idx);
319 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
320 struct mm_struct *mm,
324 struct kvm *kvm = mmu_notifier_to_kvm(mn);
325 int need_tlb_flush = 0, idx;
327 idx = srcu_read_lock(&kvm->srcu);
328 spin_lock(&kvm->mmu_lock);
330 * The count increase must become visible at unlock time as no
331 * spte can be established without taking the mmu_lock and
332 * count is also read inside the mmu_lock critical section.
334 kvm->mmu_notifier_count++;
335 need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
336 need_tlb_flush |= kvm->tlbs_dirty;
337 /* we've to flush the tlb before the pages can be freed */
339 kvm_flush_remote_tlbs(kvm);
341 spin_unlock(&kvm->mmu_lock);
342 srcu_read_unlock(&kvm->srcu, idx);
345 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
346 struct mm_struct *mm,
350 struct kvm *kvm = mmu_notifier_to_kvm(mn);
352 spin_lock(&kvm->mmu_lock);
354 * This sequence increase will notify the kvm page fault that
355 * the page that is going to be mapped in the spte could have
358 kvm->mmu_notifier_seq++;
361 * The above sequence increase must be visible before the
362 * below count decrease, which is ensured by the smp_wmb above
363 * in conjunction with the smp_rmb in mmu_notifier_retry().
365 kvm->mmu_notifier_count--;
366 spin_unlock(&kvm->mmu_lock);
368 BUG_ON(kvm->mmu_notifier_count < 0);
371 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
372 struct mm_struct *mm,
376 struct kvm *kvm = mmu_notifier_to_kvm(mn);
379 idx = srcu_read_lock(&kvm->srcu);
380 spin_lock(&kvm->mmu_lock);
382 young = kvm_age_hva(kvm, start, end);
384 kvm_flush_remote_tlbs(kvm);
386 spin_unlock(&kvm->mmu_lock);
387 srcu_read_unlock(&kvm->srcu, idx);
392 static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
393 struct mm_struct *mm,
397 struct kvm *kvm = mmu_notifier_to_kvm(mn);
400 idx = srcu_read_lock(&kvm->srcu);
401 spin_lock(&kvm->mmu_lock);
403 * Even though we do not flush TLB, this will still adversely
404 * affect performance on pre-Haswell Intel EPT, where there is
405 * no EPT Access Bit to clear so that we have to tear down EPT
406 * tables instead. If we find this unacceptable, we can always
407 * add a parameter to kvm_age_hva so that it effectively doesn't
408 * do anything on clear_young.
410 * Also note that currently we never issue secondary TLB flushes
411 * from clear_young, leaving this job up to the regular system
412 * cadence. If we find this inaccurate, we might come up with a
413 * more sophisticated heuristic later.
415 young = kvm_age_hva(kvm, start, end);
416 spin_unlock(&kvm->mmu_lock);
417 srcu_read_unlock(&kvm->srcu, idx);
422 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
423 struct mm_struct *mm,
424 unsigned long address)
426 struct kvm *kvm = mmu_notifier_to_kvm(mn);
429 idx = srcu_read_lock(&kvm->srcu);
430 spin_lock(&kvm->mmu_lock);
431 young = kvm_test_age_hva(kvm, address);
432 spin_unlock(&kvm->mmu_lock);
433 srcu_read_unlock(&kvm->srcu, idx);
438 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
439 struct mm_struct *mm)
441 struct kvm *kvm = mmu_notifier_to_kvm(mn);
444 idx = srcu_read_lock(&kvm->srcu);
445 kvm_arch_flush_shadow_all(kvm);
446 srcu_read_unlock(&kvm->srcu, idx);
449 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
450 .invalidate_page = kvm_mmu_notifier_invalidate_page,
451 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
452 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
453 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
454 .clear_young = kvm_mmu_notifier_clear_young,
455 .test_young = kvm_mmu_notifier_test_young,
456 .change_pte = kvm_mmu_notifier_change_pte,
457 .release = kvm_mmu_notifier_release,
460 static int kvm_init_mmu_notifier(struct kvm *kvm)
462 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
463 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
466 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
468 static int kvm_init_mmu_notifier(struct kvm *kvm)
473 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
475 static struct kvm_memslots *kvm_alloc_memslots(void)
478 struct kvm_memslots *slots;
480 slots = kvm_kvzalloc(sizeof(struct kvm_memslots));
485 * Init kvm generation close to the maximum to easily test the
486 * code of handling generation number wrap-around.
488 slots->generation = -150;
489 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
490 slots->id_to_index[i] = slots->memslots[i].id = i;
495 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
497 if (!memslot->dirty_bitmap)
500 kvfree(memslot->dirty_bitmap);
501 memslot->dirty_bitmap = NULL;
505 * Free any memory in @free but not in @dont.
507 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
508 struct kvm_memory_slot *dont)
510 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
511 kvm_destroy_dirty_bitmap(free);
513 kvm_arch_free_memslot(kvm, free, dont);
518 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
520 struct kvm_memory_slot *memslot;
525 kvm_for_each_memslot(memslot, slots)
526 kvm_free_memslot(kvm, memslot, NULL);
531 static struct kvm *kvm_create_vm(unsigned long type)
534 struct kvm *kvm = kvm_arch_alloc_vm();
537 return ERR_PTR(-ENOMEM);
539 r = kvm_arch_init_vm(kvm, type);
541 goto out_err_no_disable;
543 r = hardware_enable_all();
545 goto out_err_no_disable;
547 #ifdef CONFIG_HAVE_KVM_IRQFD
548 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
551 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
554 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
555 kvm->memslots[i] = kvm_alloc_memslots();
556 if (!kvm->memslots[i])
557 goto out_err_no_srcu;
560 if (init_srcu_struct(&kvm->srcu))
561 goto out_err_no_srcu;
562 if (init_srcu_struct(&kvm->irq_srcu))
563 goto out_err_no_irq_srcu;
564 for (i = 0; i < KVM_NR_BUSES; i++) {
565 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
571 spin_lock_init(&kvm->mmu_lock);
572 kvm->mm = current->mm;
573 atomic_inc(&kvm->mm->mm_count);
574 kvm_eventfd_init(kvm);
575 mutex_init(&kvm->lock);
576 mutex_init(&kvm->irq_lock);
577 mutex_init(&kvm->slots_lock);
578 atomic_set(&kvm->users_count, 1);
579 INIT_LIST_HEAD(&kvm->devices);
581 r = kvm_init_mmu_notifier(kvm);
585 spin_lock(&kvm_lock);
586 list_add(&kvm->vm_list, &vm_list);
587 spin_unlock(&kvm_lock);
589 preempt_notifier_inc();
594 cleanup_srcu_struct(&kvm->irq_srcu);
596 cleanup_srcu_struct(&kvm->srcu);
598 hardware_disable_all();
600 for (i = 0; i < KVM_NR_BUSES; i++)
601 kfree(kvm->buses[i]);
602 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
603 kvm_free_memslots(kvm, kvm->memslots[i]);
604 kvm_arch_free_vm(kvm);
609 * Avoid using vmalloc for a small buffer.
610 * Should not be used when the size is statically known.
612 void *kvm_kvzalloc(unsigned long size)
614 if (size > PAGE_SIZE)
615 return vzalloc(size);
617 return kzalloc(size, GFP_KERNEL);
620 static void kvm_destroy_devices(struct kvm *kvm)
622 struct kvm_device *dev, *tmp;
624 list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
625 list_del(&dev->vm_node);
626 dev->ops->destroy(dev);
630 static void kvm_destroy_vm(struct kvm *kvm)
633 struct mm_struct *mm = kvm->mm;
635 kvm_arch_sync_events(kvm);
636 spin_lock(&kvm_lock);
637 list_del(&kvm->vm_list);
638 spin_unlock(&kvm_lock);
639 kvm_free_irq_routing(kvm);
640 for (i = 0; i < KVM_NR_BUSES; i++)
641 kvm_io_bus_destroy(kvm->buses[i]);
642 kvm_coalesced_mmio_free(kvm);
643 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
644 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
646 kvm_arch_flush_shadow_all(kvm);
648 kvm_arch_destroy_vm(kvm);
649 kvm_destroy_devices(kvm);
650 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
651 kvm_free_memslots(kvm, kvm->memslots[i]);
652 cleanup_srcu_struct(&kvm->irq_srcu);
653 cleanup_srcu_struct(&kvm->srcu);
654 kvm_arch_free_vm(kvm);
655 preempt_notifier_dec();
656 hardware_disable_all();
660 void kvm_get_kvm(struct kvm *kvm)
662 atomic_inc(&kvm->users_count);
664 EXPORT_SYMBOL_GPL(kvm_get_kvm);
666 void kvm_put_kvm(struct kvm *kvm)
668 if (atomic_dec_and_test(&kvm->users_count))
671 EXPORT_SYMBOL_GPL(kvm_put_kvm);
674 static int kvm_vm_release(struct inode *inode, struct file *filp)
676 struct kvm *kvm = filp->private_data;
678 kvm_irqfd_release(kvm);
685 * Allocation size is twice as large as the actual dirty bitmap size.
686 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
688 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
690 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
692 memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
693 if (!memslot->dirty_bitmap)
700 * Insert memslot and re-sort memslots based on their GFN,
701 * so binary search could be used to lookup GFN.
702 * Sorting algorithm takes advantage of having initially
703 * sorted array and known changed memslot position.
705 static void update_memslots(struct kvm_memslots *slots,
706 struct kvm_memory_slot *new)
709 int i = slots->id_to_index[id];
710 struct kvm_memory_slot *mslots = slots->memslots;
712 WARN_ON(mslots[i].id != id);
714 WARN_ON(!mslots[i].npages);
715 if (mslots[i].npages)
718 if (!mslots[i].npages)
722 while (i < KVM_MEM_SLOTS_NUM - 1 &&
723 new->base_gfn <= mslots[i + 1].base_gfn) {
724 if (!mslots[i + 1].npages)
726 mslots[i] = mslots[i + 1];
727 slots->id_to_index[mslots[i].id] = i;
732 * The ">=" is needed when creating a slot with base_gfn == 0,
733 * so that it moves before all those with base_gfn == npages == 0.
735 * On the other hand, if new->npages is zero, the above loop has
736 * already left i pointing to the beginning of the empty part of
737 * mslots, and the ">=" would move the hole backwards in this
738 * case---which is wrong. So skip the loop when deleting a slot.
742 new->base_gfn >= mslots[i - 1].base_gfn) {
743 mslots[i] = mslots[i - 1];
744 slots->id_to_index[mslots[i].id] = i;
748 WARN_ON_ONCE(i != slots->used_slots);
751 slots->id_to_index[mslots[i].id] = i;
754 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
756 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
758 #ifdef __KVM_HAVE_READONLY_MEM
759 valid_flags |= KVM_MEM_READONLY;
762 if (mem->flags & ~valid_flags)
768 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
769 int as_id, struct kvm_memslots *slots)
771 struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id);
774 * Set the low bit in the generation, which disables SPTE caching
775 * until the end of synchronize_srcu_expedited.
777 WARN_ON(old_memslots->generation & 1);
778 slots->generation = old_memslots->generation + 1;
780 rcu_assign_pointer(kvm->memslots[as_id], slots);
781 synchronize_srcu_expedited(&kvm->srcu);
784 * Increment the new memslot generation a second time. This prevents
785 * vm exits that race with memslot updates from caching a memslot
786 * generation that will (potentially) be valid forever.
790 kvm_arch_memslots_updated(kvm, slots);
796 * Allocate some memory and give it an address in the guest physical address
799 * Discontiguous memory is allowed, mostly for framebuffers.
801 * Must be called holding kvm->slots_lock for write.
803 int __kvm_set_memory_region(struct kvm *kvm,
804 const struct kvm_userspace_memory_region *mem)
808 unsigned long npages;
809 struct kvm_memory_slot *slot;
810 struct kvm_memory_slot old, new;
811 struct kvm_memslots *slots = NULL, *old_memslots;
813 enum kvm_mr_change change;
815 r = check_memory_region_flags(mem);
820 as_id = mem->slot >> 16;
823 /* General sanity checks */
824 if (mem->memory_size & (PAGE_SIZE - 1))
826 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
828 /* We can read the guest memory with __xxx_user() later on. */
829 if ((id < KVM_USER_MEM_SLOTS) &&
830 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
831 !access_ok(VERIFY_WRITE,
832 (void __user *)(unsigned long)mem->userspace_addr,
835 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
837 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
840 slot = id_to_memslot(__kvm_memslots(kvm, as_id), id);
841 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
842 npages = mem->memory_size >> PAGE_SHIFT;
844 if (npages > KVM_MEM_MAX_NR_PAGES)
850 new.base_gfn = base_gfn;
852 new.flags = mem->flags;
856 change = KVM_MR_CREATE;
857 else { /* Modify an existing slot. */
858 if ((mem->userspace_addr != old.userspace_addr) ||
859 (npages != old.npages) ||
860 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
863 if (base_gfn != old.base_gfn)
864 change = KVM_MR_MOVE;
865 else if (new.flags != old.flags)
866 change = KVM_MR_FLAGS_ONLY;
867 else { /* Nothing to change. */
876 change = KVM_MR_DELETE;
881 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
882 /* Check for overlaps */
884 kvm_for_each_memslot(slot, __kvm_memslots(kvm, as_id)) {
885 if ((slot->id >= KVM_USER_MEM_SLOTS) ||
888 if (!((base_gfn + npages <= slot->base_gfn) ||
889 (base_gfn >= slot->base_gfn + slot->npages)))
894 /* Free page dirty bitmap if unneeded */
895 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
896 new.dirty_bitmap = NULL;
899 if (change == KVM_MR_CREATE) {
900 new.userspace_addr = mem->userspace_addr;
902 if (kvm_arch_create_memslot(kvm, &new, npages))
906 /* Allocate page dirty bitmap if needed */
907 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
908 if (kvm_create_dirty_bitmap(&new) < 0)
912 slots = kvm_kvzalloc(sizeof(struct kvm_memslots));
915 memcpy(slots, __kvm_memslots(kvm, as_id), sizeof(struct kvm_memslots));
917 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
918 slot = id_to_memslot(slots, id);
919 slot->flags |= KVM_MEMSLOT_INVALID;
921 old_memslots = install_new_memslots(kvm, as_id, slots);
923 /* slot was deleted or moved, clear iommu mapping */
924 kvm_iommu_unmap_pages(kvm, &old);
925 /* From this point no new shadow pages pointing to a deleted,
926 * or moved, memslot will be created.
928 * validation of sp->gfn happens in:
929 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
930 * - kvm_is_visible_gfn (mmu_check_roots)
932 kvm_arch_flush_shadow_memslot(kvm, slot);
935 * We can re-use the old_memslots from above, the only difference
936 * from the currently installed memslots is the invalid flag. This
937 * will get overwritten by update_memslots anyway.
939 slots = old_memslots;
942 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
946 /* actual memory is freed via old in kvm_free_memslot below */
947 if (change == KVM_MR_DELETE) {
948 new.dirty_bitmap = NULL;
949 memset(&new.arch, 0, sizeof(new.arch));
952 update_memslots(slots, &new);
953 old_memslots = install_new_memslots(kvm, as_id, slots);
955 kvm_arch_commit_memory_region(kvm, mem, &old, &new, change);
957 kvm_free_memslot(kvm, &old, &new);
958 kvfree(old_memslots);
961 * IOMMU mapping: New slots need to be mapped. Old slots need to be
962 * un-mapped and re-mapped if their base changes. Since base change
963 * unmapping is handled above with slot deletion, mapping alone is
964 * needed here. Anything else the iommu might care about for existing
965 * slots (size changes, userspace addr changes and read-only flag
966 * changes) is disallowed above, so any other attribute changes getting
967 * here can be skipped.
969 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
970 r = kvm_iommu_map_pages(kvm, &new);
979 kvm_free_memslot(kvm, &new, &old);
983 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
985 int kvm_set_memory_region(struct kvm *kvm,
986 const struct kvm_userspace_memory_region *mem)
990 mutex_lock(&kvm->slots_lock);
991 r = __kvm_set_memory_region(kvm, mem);
992 mutex_unlock(&kvm->slots_lock);
995 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
997 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
998 struct kvm_userspace_memory_region *mem)
1000 if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1003 return kvm_set_memory_region(kvm, mem);
1006 int kvm_get_dirty_log(struct kvm *kvm,
1007 struct kvm_dirty_log *log, int *is_dirty)
1009 struct kvm_memslots *slots;
1010 struct kvm_memory_slot *memslot;
1011 int r, i, as_id, id;
1013 unsigned long any = 0;
1016 as_id = log->slot >> 16;
1017 id = (u16)log->slot;
1018 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1021 slots = __kvm_memslots(kvm, as_id);
1022 memslot = id_to_memslot(slots, id);
1024 if (!memslot->dirty_bitmap)
1027 n = kvm_dirty_bitmap_bytes(memslot);
1029 for (i = 0; !any && i < n/sizeof(long); ++i)
1030 any = memslot->dirty_bitmap[i];
1033 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1043 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1045 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1047 * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages
1048 * are dirty write protect them for next write.
1049 * @kvm: pointer to kvm instance
1050 * @log: slot id and address to which we copy the log
1051 * @is_dirty: flag set if any page is dirty
1053 * We need to keep it in mind that VCPU threads can write to the bitmap
1054 * concurrently. So, to avoid losing track of dirty pages we keep the
1057 * 1. Take a snapshot of the bit and clear it if needed.
1058 * 2. Write protect the corresponding page.
1059 * 3. Copy the snapshot to the userspace.
1060 * 4. Upon return caller flushes TLB's if needed.
1062 * Between 2 and 4, the guest may write to the page using the remaining TLB
1063 * entry. This is not a problem because the page is reported dirty using
1064 * the snapshot taken before and step 4 ensures that writes done after
1065 * exiting to userspace will be logged for the next call.
1068 int kvm_get_dirty_log_protect(struct kvm *kvm,
1069 struct kvm_dirty_log *log, bool *is_dirty)
1071 struct kvm_memslots *slots;
1072 struct kvm_memory_slot *memslot;
1073 int r, i, as_id, id;
1075 unsigned long *dirty_bitmap;
1076 unsigned long *dirty_bitmap_buffer;
1079 as_id = log->slot >> 16;
1080 id = (u16)log->slot;
1081 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1084 slots = __kvm_memslots(kvm, as_id);
1085 memslot = id_to_memslot(slots, id);
1087 dirty_bitmap = memslot->dirty_bitmap;
1092 n = kvm_dirty_bitmap_bytes(memslot);
1094 dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long);
1095 memset(dirty_bitmap_buffer, 0, n);
1097 spin_lock(&kvm->mmu_lock);
1099 for (i = 0; i < n / sizeof(long); i++) {
1103 if (!dirty_bitmap[i])
1108 mask = xchg(&dirty_bitmap[i], 0);
1109 dirty_bitmap_buffer[i] = mask;
1112 offset = i * BITS_PER_LONG;
1113 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1118 spin_unlock(&kvm->mmu_lock);
1121 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1128 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect);
1131 bool kvm_largepages_enabled(void)
1133 return largepages_enabled;
1136 void kvm_disable_largepages(void)
1138 largepages_enabled = false;
1140 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1142 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1144 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1146 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1148 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
1150 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn);
1153 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1155 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1157 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1158 memslot->flags & KVM_MEMSLOT_INVALID)
1163 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1165 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1167 struct vm_area_struct *vma;
1168 unsigned long addr, size;
1172 addr = gfn_to_hva(kvm, gfn);
1173 if (kvm_is_error_hva(addr))
1176 down_read(¤t->mm->mmap_sem);
1177 vma = find_vma(current->mm, addr);
1181 size = vma_kernel_pagesize(vma);
1184 up_read(¤t->mm->mmap_sem);
1189 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1191 return slot->flags & KVM_MEM_READONLY;
1194 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1195 gfn_t *nr_pages, bool write)
1197 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1198 return KVM_HVA_ERR_BAD;
1200 if (memslot_is_readonly(slot) && write)
1201 return KVM_HVA_ERR_RO_BAD;
1204 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1206 return __gfn_to_hva_memslot(slot, gfn);
1209 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1212 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1215 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1218 return gfn_to_hva_many(slot, gfn, NULL);
1220 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1222 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1224 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1226 EXPORT_SYMBOL_GPL(gfn_to_hva);
1228 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
1230 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
1232 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
1235 * If writable is set to false, the hva returned by this function is only
1236 * allowed to be read.
1238 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1239 gfn_t gfn, bool *writable)
1241 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1243 if (!kvm_is_error_hva(hva) && writable)
1244 *writable = !memslot_is_readonly(slot);
1249 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1251 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1253 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1256 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
1258 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1260 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1263 static int get_user_page_nowait(unsigned long start, int write,
1266 int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1269 flags |= FOLL_WRITE;
1271 return __get_user_pages(current, current->mm, start, 1, flags, page,
1275 static inline int check_user_page_hwpoison(unsigned long addr)
1277 int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1279 rc = __get_user_pages(current, current->mm, addr, 1,
1280 flags, NULL, NULL, NULL);
1281 return rc == -EHWPOISON;
1285 * The atomic path to get the writable pfn which will be stored in @pfn,
1286 * true indicates success, otherwise false is returned.
1288 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1289 bool write_fault, bool *writable, kvm_pfn_t *pfn)
1291 struct page *page[1];
1294 if (!(async || atomic))
1298 * Fast pin a writable pfn only if it is a write fault request
1299 * or the caller allows to map a writable pfn for a read fault
1302 if (!(write_fault || writable))
1305 npages = __get_user_pages_fast(addr, 1, 1, page);
1307 *pfn = page_to_pfn(page[0]);
1318 * The slow path to get the pfn of the specified host virtual address,
1319 * 1 indicates success, -errno is returned if error is detected.
1321 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1322 bool *writable, kvm_pfn_t *pfn)
1324 struct page *page[1];
1330 *writable = write_fault;
1333 down_read(¤t->mm->mmap_sem);
1334 npages = get_user_page_nowait(addr, write_fault, page);
1335 up_read(¤t->mm->mmap_sem);
1337 npages = __get_user_pages_unlocked(current, current->mm, addr, 1,
1338 write_fault, 0, page,
1339 FOLL_TOUCH|FOLL_HWPOISON);
1343 /* map read fault as writable if possible */
1344 if (unlikely(!write_fault) && writable) {
1345 struct page *wpage[1];
1347 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1356 *pfn = page_to_pfn(page[0]);
1360 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1362 if (unlikely(!(vma->vm_flags & VM_READ)))
1365 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1372 * Pin guest page in memory and return its pfn.
1373 * @addr: host virtual address which maps memory to the guest
1374 * @atomic: whether this function can sleep
1375 * @async: whether this function need to wait IO complete if the
1376 * host page is not in the memory
1377 * @write_fault: whether we should get a writable host page
1378 * @writable: whether it allows to map a writable host page for !@write_fault
1380 * The function will map a writable host page for these two cases:
1381 * 1): @write_fault = true
1382 * 2): @write_fault = false && @writable, @writable will tell the caller
1383 * whether the mapping is writable.
1385 static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1386 bool write_fault, bool *writable)
1388 struct vm_area_struct *vma;
1392 /* we can do it either atomically or asynchronously, not both */
1393 BUG_ON(atomic && async);
1395 if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1399 return KVM_PFN_ERR_FAULT;
1401 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1405 down_read(¤t->mm->mmap_sem);
1406 if (npages == -EHWPOISON ||
1407 (!async && check_user_page_hwpoison(addr))) {
1408 pfn = KVM_PFN_ERR_HWPOISON;
1412 vma = find_vma_intersection(current->mm, addr, addr + 1);
1415 pfn = KVM_PFN_ERR_FAULT;
1416 else if ((vma->vm_flags & VM_PFNMAP)) {
1417 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1419 BUG_ON(!kvm_is_reserved_pfn(pfn));
1421 if (async && vma_is_valid(vma, write_fault))
1423 pfn = KVM_PFN_ERR_FAULT;
1426 up_read(¤t->mm->mmap_sem);
1430 kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn,
1431 bool atomic, bool *async, bool write_fault,
1434 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1436 if (addr == KVM_HVA_ERR_RO_BAD) {
1439 return KVM_PFN_ERR_RO_FAULT;
1442 if (kvm_is_error_hva(addr)) {
1445 return KVM_PFN_NOSLOT;
1448 /* Do not map writable pfn in the readonly memslot. */
1449 if (writable && memslot_is_readonly(slot)) {
1454 return hva_to_pfn(addr, atomic, async, write_fault,
1457 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
1459 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1462 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
1463 write_fault, writable);
1465 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1467 kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1469 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1471 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
1473 kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1475 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1477 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1479 kvm_pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1481 return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm, gfn), gfn);
1483 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1485 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
1487 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1489 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
1491 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1493 return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
1495 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1497 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
1499 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1501 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
1503 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1504 struct page **pages, int nr_pages)
1509 addr = gfn_to_hva_many(slot, gfn, &entry);
1510 if (kvm_is_error_hva(addr))
1513 if (entry < nr_pages)
1516 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1518 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1520 static struct page *kvm_pfn_to_page(kvm_pfn_t pfn)
1522 if (is_error_noslot_pfn(pfn))
1523 return KVM_ERR_PTR_BAD_PAGE;
1525 if (kvm_is_reserved_pfn(pfn)) {
1527 return KVM_ERR_PTR_BAD_PAGE;
1530 return pfn_to_page(pfn);
1533 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1537 pfn = gfn_to_pfn(kvm, gfn);
1539 return kvm_pfn_to_page(pfn);
1541 EXPORT_SYMBOL_GPL(gfn_to_page);
1543 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
1547 pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
1549 return kvm_pfn_to_page(pfn);
1551 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
1553 void kvm_release_page_clean(struct page *page)
1555 WARN_ON(is_error_page(page));
1557 kvm_release_pfn_clean(page_to_pfn(page));
1559 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1561 void kvm_release_pfn_clean(kvm_pfn_t pfn)
1563 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
1564 put_page(pfn_to_page(pfn));
1566 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1568 void kvm_release_page_dirty(struct page *page)
1570 WARN_ON(is_error_page(page));
1572 kvm_release_pfn_dirty(page_to_pfn(page));
1574 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1576 static void kvm_release_pfn_dirty(kvm_pfn_t pfn)
1578 kvm_set_pfn_dirty(pfn);
1579 kvm_release_pfn_clean(pfn);
1582 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
1584 if (!kvm_is_reserved_pfn(pfn)) {
1585 struct page *page = pfn_to_page(pfn);
1587 if (!PageReserved(page))
1591 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1593 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
1595 if (!kvm_is_reserved_pfn(pfn))
1596 mark_page_accessed(pfn_to_page(pfn));
1598 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1600 void kvm_get_pfn(kvm_pfn_t pfn)
1602 if (!kvm_is_reserved_pfn(pfn))
1603 get_page(pfn_to_page(pfn));
1605 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1607 static int next_segment(unsigned long len, int offset)
1609 if (len > PAGE_SIZE - offset)
1610 return PAGE_SIZE - offset;
1615 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
1616 void *data, int offset, int len)
1621 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1622 if (kvm_is_error_hva(addr))
1624 r = __copy_from_user(data, (void __user *)addr + offset, len);
1630 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1633 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1635 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1637 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1639 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
1640 int offset, int len)
1642 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1644 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1646 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
1648 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1650 gfn_t gfn = gpa >> PAGE_SHIFT;
1652 int offset = offset_in_page(gpa);
1655 while ((seg = next_segment(len, offset)) != 0) {
1656 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1666 EXPORT_SYMBOL_GPL(kvm_read_guest);
1668 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
1670 gfn_t gfn = gpa >> PAGE_SHIFT;
1672 int offset = offset_in_page(gpa);
1675 while ((seg = next_segment(len, offset)) != 0) {
1676 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
1686 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
1688 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1689 void *data, int offset, unsigned long len)
1694 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1695 if (kvm_is_error_hva(addr))
1697 pagefault_disable();
1698 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1705 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1708 gfn_t gfn = gpa >> PAGE_SHIFT;
1709 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1710 int offset = offset_in_page(gpa);
1712 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1714 EXPORT_SYMBOL_GPL(kvm_read_guest_atomic);
1716 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
1717 void *data, unsigned long len)
1719 gfn_t gfn = gpa >> PAGE_SHIFT;
1720 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1721 int offset = offset_in_page(gpa);
1723 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1725 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
1727 static int __kvm_write_guest_page(struct kvm_memory_slot *memslot, gfn_t gfn,
1728 const void *data, int offset, int len)
1733 addr = gfn_to_hva_memslot(memslot, gfn);
1734 if (kvm_is_error_hva(addr))
1736 r = __copy_to_user((void __user *)addr + offset, data, len);
1739 mark_page_dirty_in_slot(memslot, gfn);
1743 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
1744 const void *data, int offset, int len)
1746 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1748 return __kvm_write_guest_page(slot, gfn, data, offset, len);
1750 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1752 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
1753 const void *data, int offset, int len)
1755 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1757 return __kvm_write_guest_page(slot, gfn, data, offset, len);
1759 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
1761 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1764 gfn_t gfn = gpa >> PAGE_SHIFT;
1766 int offset = offset_in_page(gpa);
1769 while ((seg = next_segment(len, offset)) != 0) {
1770 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1780 EXPORT_SYMBOL_GPL(kvm_write_guest);
1782 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
1785 gfn_t gfn = gpa >> PAGE_SHIFT;
1787 int offset = offset_in_page(gpa);
1790 while ((seg = next_segment(len, offset)) != 0) {
1791 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
1801 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
1803 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1804 gpa_t gpa, unsigned long len)
1806 struct kvm_memslots *slots = kvm_memslots(kvm);
1807 int offset = offset_in_page(gpa);
1808 gfn_t start_gfn = gpa >> PAGE_SHIFT;
1809 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1810 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1811 gfn_t nr_pages_avail;
1814 ghc->generation = slots->generation;
1816 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1817 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, NULL);
1818 if (!kvm_is_error_hva(ghc->hva) && nr_pages_needed <= 1) {
1822 * If the requested region crosses two memslots, we still
1823 * verify that the entire region is valid here.
1825 while (start_gfn <= end_gfn) {
1826 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1827 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1829 if (kvm_is_error_hva(ghc->hva))
1831 start_gfn += nr_pages_avail;
1833 /* Use the slow path for cross page reads and writes. */
1834 ghc->memslot = NULL;
1838 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1840 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1841 void *data, unsigned long len)
1843 struct kvm_memslots *slots = kvm_memslots(kvm);
1846 BUG_ON(len > ghc->len);
1848 if (slots->generation != ghc->generation)
1849 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1851 if (unlikely(!ghc->memslot))
1852 return kvm_write_guest(kvm, ghc->gpa, data, len);
1854 if (kvm_is_error_hva(ghc->hva))
1857 r = __copy_to_user((void __user *)ghc->hva, data, len);
1860 mark_page_dirty_in_slot(ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1864 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1866 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1867 void *data, unsigned long len)
1869 struct kvm_memslots *slots = kvm_memslots(kvm);
1872 BUG_ON(len > ghc->len);
1874 if (slots->generation != ghc->generation)
1875 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1877 if (unlikely(!ghc->memslot))
1878 return kvm_read_guest(kvm, ghc->gpa, data, len);
1880 if (kvm_is_error_hva(ghc->hva))
1883 r = __copy_from_user(data, (void __user *)ghc->hva, len);
1889 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1891 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1893 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
1895 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
1897 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1899 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1901 gfn_t gfn = gpa >> PAGE_SHIFT;
1903 int offset = offset_in_page(gpa);
1906 while ((seg = next_segment(len, offset)) != 0) {
1907 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1916 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1918 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot,
1921 if (memslot && memslot->dirty_bitmap) {
1922 unsigned long rel_gfn = gfn - memslot->base_gfn;
1924 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1928 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1930 struct kvm_memory_slot *memslot;
1932 memslot = gfn_to_memslot(kvm, gfn);
1933 mark_page_dirty_in_slot(memslot, gfn);
1935 EXPORT_SYMBOL_GPL(mark_page_dirty);
1937 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
1939 struct kvm_memory_slot *memslot;
1941 memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1942 mark_page_dirty_in_slot(memslot, gfn);
1944 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
1946 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
1948 unsigned int old, val, grow;
1950 old = val = vcpu->halt_poll_ns;
1951 grow = READ_ONCE(halt_poll_ns_grow);
1953 if (val == 0 && grow)
1958 if (val > halt_poll_ns)
1961 vcpu->halt_poll_ns = val;
1962 trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
1965 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
1967 unsigned int old, val, shrink;
1969 old = val = vcpu->halt_poll_ns;
1970 shrink = READ_ONCE(halt_poll_ns_shrink);
1976 vcpu->halt_poll_ns = val;
1977 trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
1980 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
1982 if (kvm_arch_vcpu_runnable(vcpu)) {
1983 kvm_make_request(KVM_REQ_UNHALT, vcpu);
1986 if (kvm_cpu_has_pending_timer(vcpu))
1988 if (signal_pending(current))
1995 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1997 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
2000 DECLARE_SWAITQUEUE(wait);
2001 bool waited = false;
2004 start = cur = ktime_get();
2005 if (vcpu->halt_poll_ns) {
2006 ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns);
2008 ++vcpu->stat.halt_attempted_poll;
2011 * This sets KVM_REQ_UNHALT if an interrupt
2014 if (kvm_vcpu_check_block(vcpu) < 0) {
2015 ++vcpu->stat.halt_successful_poll;
2019 } while (single_task_running() && ktime_before(cur, stop));
2022 kvm_arch_vcpu_blocking(vcpu);
2025 prepare_to_swait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
2027 if (kvm_vcpu_check_block(vcpu) < 0)
2034 finish_swait(&vcpu->wq, &wait);
2037 kvm_arch_vcpu_unblocking(vcpu);
2039 block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
2042 if (block_ns <= vcpu->halt_poll_ns)
2044 /* we had a long block, shrink polling */
2045 else if (vcpu->halt_poll_ns && block_ns > halt_poll_ns)
2046 shrink_halt_poll_ns(vcpu);
2047 /* we had a short halt and our poll time is too small */
2048 else if (vcpu->halt_poll_ns < halt_poll_ns &&
2049 block_ns < halt_poll_ns)
2050 grow_halt_poll_ns(vcpu);
2052 vcpu->halt_poll_ns = 0;
2054 trace_kvm_vcpu_wakeup(block_ns, waited);
2056 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
2060 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2062 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
2065 int cpu = vcpu->cpu;
2066 struct swait_queue_head *wqp;
2068 wqp = kvm_arch_vcpu_wq(vcpu);
2069 if (swait_active(wqp)) {
2071 ++vcpu->stat.halt_wakeup;
2075 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
2076 if (kvm_arch_vcpu_should_kick(vcpu))
2077 smp_send_reschedule(cpu);
2080 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
2081 #endif /* !CONFIG_S390 */
2083 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
2086 struct task_struct *task = NULL;
2090 pid = rcu_dereference(target->pid);
2092 task = get_pid_task(pid, PIDTYPE_PID);
2096 ret = yield_to(task, 1);
2097 put_task_struct(task);
2101 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
2104 * Helper that checks whether a VCPU is eligible for directed yield.
2105 * Most eligible candidate to yield is decided by following heuristics:
2107 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2108 * (preempted lock holder), indicated by @in_spin_loop.
2109 * Set at the beiginning and cleared at the end of interception/PLE handler.
2111 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2112 * chance last time (mostly it has become eligible now since we have probably
2113 * yielded to lockholder in last iteration. This is done by toggling
2114 * @dy_eligible each time a VCPU checked for eligibility.)
2116 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2117 * to preempted lock-holder could result in wrong VCPU selection and CPU
2118 * burning. Giving priority for a potential lock-holder increases lock
2121 * Since algorithm is based on heuristics, accessing another VCPU data without
2122 * locking does not harm. It may result in trying to yield to same VCPU, fail
2123 * and continue with next VCPU and so on.
2125 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
2127 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2130 eligible = !vcpu->spin_loop.in_spin_loop ||
2131 vcpu->spin_loop.dy_eligible;
2133 if (vcpu->spin_loop.in_spin_loop)
2134 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
2142 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
2144 struct kvm *kvm = me->kvm;
2145 struct kvm_vcpu *vcpu;
2146 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
2152 kvm_vcpu_set_in_spin_loop(me, true);
2154 * We boost the priority of a VCPU that is runnable but not
2155 * currently running, because it got preempted by something
2156 * else and called schedule in __vcpu_run. Hopefully that
2157 * VCPU is holding the lock that we need and will release it.
2158 * We approximate round-robin by starting at the last boosted VCPU.
2160 for (pass = 0; pass < 2 && !yielded && try; pass++) {
2161 kvm_for_each_vcpu(i, vcpu, kvm) {
2162 if (!pass && i <= last_boosted_vcpu) {
2163 i = last_boosted_vcpu;
2165 } else if (pass && i > last_boosted_vcpu)
2167 if (!ACCESS_ONCE(vcpu->preempted))
2171 if (swait_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
2173 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
2176 yielded = kvm_vcpu_yield_to(vcpu);
2178 kvm->last_boosted_vcpu = i;
2180 } else if (yielded < 0) {
2187 kvm_vcpu_set_in_spin_loop(me, false);
2189 /* Ensure vcpu is not eligible during next spinloop */
2190 kvm_vcpu_set_dy_eligible(me, false);
2192 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
2194 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2196 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
2199 if (vmf->pgoff == 0)
2200 page = virt_to_page(vcpu->run);
2202 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
2203 page = virt_to_page(vcpu->arch.pio_data);
2205 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2206 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
2207 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
2210 return kvm_arch_vcpu_fault(vcpu, vmf);
2216 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
2217 .fault = kvm_vcpu_fault,
2220 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2222 vma->vm_ops = &kvm_vcpu_vm_ops;
2226 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2228 struct kvm_vcpu *vcpu = filp->private_data;
2230 kvm_put_kvm(vcpu->kvm);
2234 static struct file_operations kvm_vcpu_fops = {
2235 .release = kvm_vcpu_release,
2236 .unlocked_ioctl = kvm_vcpu_ioctl,
2237 #ifdef CONFIG_KVM_COMPAT
2238 .compat_ioctl = kvm_vcpu_compat_ioctl,
2240 .mmap = kvm_vcpu_mmap,
2241 .llseek = noop_llseek,
2245 * Allocates an inode for the vcpu.
2247 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2249 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
2253 * Creates some virtual cpus. Good luck creating more than one.
2255 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
2258 struct kvm_vcpu *vcpu;
2260 if (id >= KVM_MAX_VCPUS)
2263 vcpu = kvm_arch_vcpu_create(kvm, id);
2265 return PTR_ERR(vcpu);
2267 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2269 r = kvm_arch_vcpu_setup(vcpu);
2273 mutex_lock(&kvm->lock);
2274 if (!kvm_vcpu_compatible(vcpu)) {
2276 goto unlock_vcpu_destroy;
2278 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
2280 goto unlock_vcpu_destroy;
2282 if (kvm_get_vcpu_by_id(kvm, id)) {
2284 goto unlock_vcpu_destroy;
2287 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
2289 /* Now it's all set up, let userspace reach it */
2291 r = create_vcpu_fd(vcpu);
2294 goto unlock_vcpu_destroy;
2297 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
2300 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
2301 * before kvm->online_vcpu's incremented value.
2304 atomic_inc(&kvm->online_vcpus);
2306 mutex_unlock(&kvm->lock);
2307 kvm_arch_vcpu_postcreate(vcpu);
2310 unlock_vcpu_destroy:
2311 mutex_unlock(&kvm->lock);
2313 kvm_arch_vcpu_destroy(vcpu);
2317 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2320 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2321 vcpu->sigset_active = 1;
2322 vcpu->sigset = *sigset;
2324 vcpu->sigset_active = 0;
2328 static long kvm_vcpu_ioctl(struct file *filp,
2329 unsigned int ioctl, unsigned long arg)
2331 struct kvm_vcpu *vcpu = filp->private_data;
2332 void __user *argp = (void __user *)arg;
2334 struct kvm_fpu *fpu = NULL;
2335 struct kvm_sregs *kvm_sregs = NULL;
2337 if (vcpu->kvm->mm != current->mm)
2340 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2343 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
2345 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
2346 * so vcpu_load() would break it.
2348 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_S390_IRQ || ioctl == KVM_INTERRUPT)
2349 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2353 r = vcpu_load(vcpu);
2361 if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
2362 /* The thread running this VCPU changed. */
2363 struct pid *oldpid = vcpu->pid;
2364 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
2366 rcu_assign_pointer(vcpu->pid, newpid);
2371 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2372 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2374 case KVM_GET_REGS: {
2375 struct kvm_regs *kvm_regs;
2378 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2381 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2385 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2392 case KVM_SET_REGS: {
2393 struct kvm_regs *kvm_regs;
2396 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2397 if (IS_ERR(kvm_regs)) {
2398 r = PTR_ERR(kvm_regs);
2401 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2405 case KVM_GET_SREGS: {
2406 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2410 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2414 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2419 case KVM_SET_SREGS: {
2420 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2421 if (IS_ERR(kvm_sregs)) {
2422 r = PTR_ERR(kvm_sregs);
2426 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2429 case KVM_GET_MP_STATE: {
2430 struct kvm_mp_state mp_state;
2432 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2436 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
2441 case KVM_SET_MP_STATE: {
2442 struct kvm_mp_state mp_state;
2445 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
2447 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2450 case KVM_TRANSLATE: {
2451 struct kvm_translation tr;
2454 if (copy_from_user(&tr, argp, sizeof(tr)))
2456 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2460 if (copy_to_user(argp, &tr, sizeof(tr)))
2465 case KVM_SET_GUEST_DEBUG: {
2466 struct kvm_guest_debug dbg;
2469 if (copy_from_user(&dbg, argp, sizeof(dbg)))
2471 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2474 case KVM_SET_SIGNAL_MASK: {
2475 struct kvm_signal_mask __user *sigmask_arg = argp;
2476 struct kvm_signal_mask kvm_sigmask;
2477 sigset_t sigset, *p;
2482 if (copy_from_user(&kvm_sigmask, argp,
2483 sizeof(kvm_sigmask)))
2486 if (kvm_sigmask.len != sizeof(sigset))
2489 if (copy_from_user(&sigset, sigmask_arg->sigset,
2494 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2498 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2502 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2506 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2512 fpu = memdup_user(argp, sizeof(*fpu));
2518 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2522 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2531 #ifdef CONFIG_KVM_COMPAT
2532 static long kvm_vcpu_compat_ioctl(struct file *filp,
2533 unsigned int ioctl, unsigned long arg)
2535 struct kvm_vcpu *vcpu = filp->private_data;
2536 void __user *argp = compat_ptr(arg);
2539 if (vcpu->kvm->mm != current->mm)
2543 case KVM_SET_SIGNAL_MASK: {
2544 struct kvm_signal_mask __user *sigmask_arg = argp;
2545 struct kvm_signal_mask kvm_sigmask;
2546 compat_sigset_t csigset;
2551 if (copy_from_user(&kvm_sigmask, argp,
2552 sizeof(kvm_sigmask)))
2555 if (kvm_sigmask.len != sizeof(csigset))
2558 if (copy_from_user(&csigset, sigmask_arg->sigset,
2561 sigset_from_compat(&sigset, &csigset);
2562 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2564 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2568 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2576 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2577 int (*accessor)(struct kvm_device *dev,
2578 struct kvm_device_attr *attr),
2581 struct kvm_device_attr attr;
2586 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2589 return accessor(dev, &attr);
2592 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2595 struct kvm_device *dev = filp->private_data;
2598 case KVM_SET_DEVICE_ATTR:
2599 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2600 case KVM_GET_DEVICE_ATTR:
2601 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2602 case KVM_HAS_DEVICE_ATTR:
2603 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2605 if (dev->ops->ioctl)
2606 return dev->ops->ioctl(dev, ioctl, arg);
2612 static int kvm_device_release(struct inode *inode, struct file *filp)
2614 struct kvm_device *dev = filp->private_data;
2615 struct kvm *kvm = dev->kvm;
2621 static const struct file_operations kvm_device_fops = {
2622 .unlocked_ioctl = kvm_device_ioctl,
2623 #ifdef CONFIG_KVM_COMPAT
2624 .compat_ioctl = kvm_device_ioctl,
2626 .release = kvm_device_release,
2629 struct kvm_device *kvm_device_from_filp(struct file *filp)
2631 if (filp->f_op != &kvm_device_fops)
2634 return filp->private_data;
2637 static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
2638 #ifdef CONFIG_KVM_MPIC
2639 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
2640 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
2643 #ifdef CONFIG_KVM_XICS
2644 [KVM_DEV_TYPE_XICS] = &kvm_xics_ops,
2648 int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
2650 if (type >= ARRAY_SIZE(kvm_device_ops_table))
2653 if (kvm_device_ops_table[type] != NULL)
2656 kvm_device_ops_table[type] = ops;
2660 void kvm_unregister_device_ops(u32 type)
2662 if (kvm_device_ops_table[type] != NULL)
2663 kvm_device_ops_table[type] = NULL;
2666 static int kvm_ioctl_create_device(struct kvm *kvm,
2667 struct kvm_create_device *cd)
2669 struct kvm_device_ops *ops = NULL;
2670 struct kvm_device *dev;
2671 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2674 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
2677 ops = kvm_device_ops_table[cd->type];
2684 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2691 ret = ops->create(dev, cd->type);
2697 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2703 list_add(&dev->vm_node, &kvm->devices);
2709 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2712 case KVM_CAP_USER_MEMORY:
2713 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2714 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2715 case KVM_CAP_INTERNAL_ERROR_DATA:
2716 #ifdef CONFIG_HAVE_KVM_MSI
2717 case KVM_CAP_SIGNAL_MSI:
2719 #ifdef CONFIG_HAVE_KVM_IRQFD
2721 case KVM_CAP_IRQFD_RESAMPLE:
2723 case KVM_CAP_IOEVENTFD_ANY_LENGTH:
2724 case KVM_CAP_CHECK_EXTENSION_VM:
2726 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2727 case KVM_CAP_IRQ_ROUTING:
2728 return KVM_MAX_IRQ_ROUTES;
2730 #if KVM_ADDRESS_SPACE_NUM > 1
2731 case KVM_CAP_MULTI_ADDRESS_SPACE:
2732 return KVM_ADDRESS_SPACE_NUM;
2737 return kvm_vm_ioctl_check_extension(kvm, arg);
2740 static long kvm_vm_ioctl(struct file *filp,
2741 unsigned int ioctl, unsigned long arg)
2743 struct kvm *kvm = filp->private_data;
2744 void __user *argp = (void __user *)arg;
2747 if (kvm->mm != current->mm)
2750 case KVM_CREATE_VCPU:
2751 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2753 case KVM_SET_USER_MEMORY_REGION: {
2754 struct kvm_userspace_memory_region kvm_userspace_mem;
2757 if (copy_from_user(&kvm_userspace_mem, argp,
2758 sizeof(kvm_userspace_mem)))
2761 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2764 case KVM_GET_DIRTY_LOG: {
2765 struct kvm_dirty_log log;
2768 if (copy_from_user(&log, argp, sizeof(log)))
2770 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2773 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2774 case KVM_REGISTER_COALESCED_MMIO: {
2775 struct kvm_coalesced_mmio_zone zone;
2778 if (copy_from_user(&zone, argp, sizeof(zone)))
2780 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2783 case KVM_UNREGISTER_COALESCED_MMIO: {
2784 struct kvm_coalesced_mmio_zone zone;
2787 if (copy_from_user(&zone, argp, sizeof(zone)))
2789 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2794 struct kvm_irqfd data;
2797 if (copy_from_user(&data, argp, sizeof(data)))
2799 r = kvm_irqfd(kvm, &data);
2802 case KVM_IOEVENTFD: {
2803 struct kvm_ioeventfd data;
2806 if (copy_from_user(&data, argp, sizeof(data)))
2808 r = kvm_ioeventfd(kvm, &data);
2811 #ifdef CONFIG_HAVE_KVM_MSI
2812 case KVM_SIGNAL_MSI: {
2816 if (copy_from_user(&msi, argp, sizeof(msi)))
2818 r = kvm_send_userspace_msi(kvm, &msi);
2822 #ifdef __KVM_HAVE_IRQ_LINE
2823 case KVM_IRQ_LINE_STATUS:
2824 case KVM_IRQ_LINE: {
2825 struct kvm_irq_level irq_event;
2828 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
2831 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
2832 ioctl == KVM_IRQ_LINE_STATUS);
2837 if (ioctl == KVM_IRQ_LINE_STATUS) {
2838 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
2846 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2847 case KVM_SET_GSI_ROUTING: {
2848 struct kvm_irq_routing routing;
2849 struct kvm_irq_routing __user *urouting;
2850 struct kvm_irq_routing_entry *entries;
2853 if (copy_from_user(&routing, argp, sizeof(routing)))
2856 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2861 entries = vmalloc(routing.nr * sizeof(*entries));
2866 if (copy_from_user(entries, urouting->entries,
2867 routing.nr * sizeof(*entries)))
2868 goto out_free_irq_routing;
2869 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2871 out_free_irq_routing:
2875 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
2876 case KVM_CREATE_DEVICE: {
2877 struct kvm_create_device cd;
2880 if (copy_from_user(&cd, argp, sizeof(cd)))
2883 r = kvm_ioctl_create_device(kvm, &cd);
2888 if (copy_to_user(argp, &cd, sizeof(cd)))
2894 case KVM_CHECK_EXTENSION:
2895 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
2898 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2904 #ifdef CONFIG_KVM_COMPAT
2905 struct compat_kvm_dirty_log {
2909 compat_uptr_t dirty_bitmap; /* one bit per page */
2914 static long kvm_vm_compat_ioctl(struct file *filp,
2915 unsigned int ioctl, unsigned long arg)
2917 struct kvm *kvm = filp->private_data;
2920 if (kvm->mm != current->mm)
2923 case KVM_GET_DIRTY_LOG: {
2924 struct compat_kvm_dirty_log compat_log;
2925 struct kvm_dirty_log log;
2928 if (copy_from_user(&compat_log, (void __user *)arg,
2929 sizeof(compat_log)))
2931 log.slot = compat_log.slot;
2932 log.padding1 = compat_log.padding1;
2933 log.padding2 = compat_log.padding2;
2934 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2936 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2940 r = kvm_vm_ioctl(filp, ioctl, arg);
2948 static struct file_operations kvm_vm_fops = {
2949 .release = kvm_vm_release,
2950 .unlocked_ioctl = kvm_vm_ioctl,
2951 #ifdef CONFIG_KVM_COMPAT
2952 .compat_ioctl = kvm_vm_compat_ioctl,
2954 .llseek = noop_llseek,
2957 static int kvm_dev_ioctl_create_vm(unsigned long type)
2962 kvm = kvm_create_vm(type);
2964 return PTR_ERR(kvm);
2965 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2966 r = kvm_coalesced_mmio_init(kvm);
2972 r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC);
2979 static long kvm_dev_ioctl(struct file *filp,
2980 unsigned int ioctl, unsigned long arg)
2985 case KVM_GET_API_VERSION:
2988 r = KVM_API_VERSION;
2991 r = kvm_dev_ioctl_create_vm(arg);
2993 case KVM_CHECK_EXTENSION:
2994 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
2996 case KVM_GET_VCPU_MMAP_SIZE:
2999 r = PAGE_SIZE; /* struct kvm_run */
3001 r += PAGE_SIZE; /* pio data page */
3003 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
3004 r += PAGE_SIZE; /* coalesced mmio ring page */
3007 case KVM_TRACE_ENABLE:
3008 case KVM_TRACE_PAUSE:
3009 case KVM_TRACE_DISABLE:
3013 return kvm_arch_dev_ioctl(filp, ioctl, arg);
3019 static struct file_operations kvm_chardev_ops = {
3020 .unlocked_ioctl = kvm_dev_ioctl,
3021 .compat_ioctl = kvm_dev_ioctl,
3022 .llseek = noop_llseek,
3025 static struct miscdevice kvm_dev = {
3031 static void hardware_enable_nolock(void *junk)
3033 int cpu = raw_smp_processor_id();
3036 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
3039 cpumask_set_cpu(cpu, cpus_hardware_enabled);
3041 r = kvm_arch_hardware_enable();
3044 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3045 atomic_inc(&hardware_enable_failed);
3046 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
3050 static void hardware_enable(void)
3052 raw_spin_lock(&kvm_count_lock);
3053 if (kvm_usage_count)
3054 hardware_enable_nolock(NULL);
3055 raw_spin_unlock(&kvm_count_lock);
3058 static void hardware_disable_nolock(void *junk)
3060 int cpu = raw_smp_processor_id();
3062 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
3064 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3065 kvm_arch_hardware_disable();
3068 static void hardware_disable(void)
3070 raw_spin_lock(&kvm_count_lock);
3071 if (kvm_usage_count)
3072 hardware_disable_nolock(NULL);
3073 raw_spin_unlock(&kvm_count_lock);
3076 static void hardware_disable_all_nolock(void)
3078 BUG_ON(!kvm_usage_count);
3081 if (!kvm_usage_count)
3082 on_each_cpu(hardware_disable_nolock, NULL, 1);
3085 static void hardware_disable_all(void)
3087 raw_spin_lock(&kvm_count_lock);
3088 hardware_disable_all_nolock();
3089 raw_spin_unlock(&kvm_count_lock);
3092 static int hardware_enable_all(void)
3096 raw_spin_lock(&kvm_count_lock);
3099 if (kvm_usage_count == 1) {
3100 atomic_set(&hardware_enable_failed, 0);
3101 on_each_cpu(hardware_enable_nolock, NULL, 1);
3103 if (atomic_read(&hardware_enable_failed)) {
3104 hardware_disable_all_nolock();
3109 raw_spin_unlock(&kvm_count_lock);
3114 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
3117 val &= ~CPU_TASKS_FROZEN;
3129 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3133 * Some (well, at least mine) BIOSes hang on reboot if
3136 * And Intel TXT required VMX off for all cpu when system shutdown.
3138 pr_info("kvm: exiting hardware virtualization\n");
3139 kvm_rebooting = true;
3140 on_each_cpu(hardware_disable_nolock, NULL, 1);
3144 static struct notifier_block kvm_reboot_notifier = {
3145 .notifier_call = kvm_reboot,
3149 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3153 for (i = 0; i < bus->dev_count; i++) {
3154 struct kvm_io_device *pos = bus->range[i].dev;
3156 kvm_iodevice_destructor(pos);
3161 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
3162 const struct kvm_io_range *r2)
3164 gpa_t addr1 = r1->addr;
3165 gpa_t addr2 = r2->addr;
3170 /* If r2->len == 0, match the exact address. If r2->len != 0,
3171 * accept any overlapping write. Any order is acceptable for
3172 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
3173 * we process all of them.
3186 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
3188 return kvm_io_bus_cmp(p1, p2);
3191 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
3192 gpa_t addr, int len)
3194 bus->range[bus->dev_count++] = (struct kvm_io_range) {
3200 sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
3201 kvm_io_bus_sort_cmp, NULL);
3206 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
3207 gpa_t addr, int len)
3209 struct kvm_io_range *range, key;
3212 key = (struct kvm_io_range) {
3217 range = bsearch(&key, bus->range, bus->dev_count,
3218 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
3222 off = range - bus->range;
3224 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
3230 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3231 struct kvm_io_range *range, const void *val)
3235 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3239 while (idx < bus->dev_count &&
3240 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3241 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
3250 /* kvm_io_bus_write - called under kvm->slots_lock */
3251 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3252 int len, const void *val)
3254 struct kvm_io_bus *bus;
3255 struct kvm_io_range range;
3258 range = (struct kvm_io_range) {
3263 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3264 r = __kvm_io_bus_write(vcpu, bus, &range, val);
3265 return r < 0 ? r : 0;
3268 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3269 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
3270 gpa_t addr, int len, const void *val, long cookie)
3272 struct kvm_io_bus *bus;
3273 struct kvm_io_range range;
3275 range = (struct kvm_io_range) {
3280 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3282 /* First try the device referenced by cookie. */
3283 if ((cookie >= 0) && (cookie < bus->dev_count) &&
3284 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
3285 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
3290 * cookie contained garbage; fall back to search and return the
3291 * correct cookie value.
3293 return __kvm_io_bus_write(vcpu, bus, &range, val);
3296 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3297 struct kvm_io_range *range, void *val)
3301 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3305 while (idx < bus->dev_count &&
3306 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3307 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
3315 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
3317 /* kvm_io_bus_read - called under kvm->slots_lock */
3318 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3321 struct kvm_io_bus *bus;
3322 struct kvm_io_range range;
3325 range = (struct kvm_io_range) {
3330 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3331 r = __kvm_io_bus_read(vcpu, bus, &range, val);
3332 return r < 0 ? r : 0;
3336 /* Caller must hold slots_lock. */
3337 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3338 int len, struct kvm_io_device *dev)
3340 struct kvm_io_bus *new_bus, *bus;
3342 bus = kvm->buses[bus_idx];
3343 /* exclude ioeventfd which is limited by maximum fd */
3344 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3347 new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count + 1) *
3348 sizeof(struct kvm_io_range)), GFP_KERNEL);
3351 memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
3352 sizeof(struct kvm_io_range)));
3353 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
3354 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3355 synchronize_srcu_expedited(&kvm->srcu);
3361 /* Caller must hold slots_lock. */
3362 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3363 struct kvm_io_device *dev)
3366 struct kvm_io_bus *new_bus, *bus;
3368 bus = kvm->buses[bus_idx];
3370 for (i = 0; i < bus->dev_count; i++)
3371 if (bus->range[i].dev == dev) {
3379 new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3380 sizeof(struct kvm_io_range)), GFP_KERNEL);
3384 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3385 new_bus->dev_count--;
3386 memcpy(new_bus->range + i, bus->range + i + 1,
3387 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3389 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3390 synchronize_srcu_expedited(&kvm->srcu);
3395 static struct notifier_block kvm_cpu_notifier = {
3396 .notifier_call = kvm_cpu_hotplug,
3399 static int vm_stat_get(void *_offset, u64 *val)
3401 unsigned offset = (long)_offset;
3405 spin_lock(&kvm_lock);
3406 list_for_each_entry(kvm, &vm_list, vm_list)
3407 *val += *(u32 *)((void *)kvm + offset);
3408 spin_unlock(&kvm_lock);
3412 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
3414 static int vcpu_stat_get(void *_offset, u64 *val)
3416 unsigned offset = (long)_offset;
3418 struct kvm_vcpu *vcpu;
3422 spin_lock(&kvm_lock);
3423 list_for_each_entry(kvm, &vm_list, vm_list)
3424 kvm_for_each_vcpu(i, vcpu, kvm)
3425 *val += *(u32 *)((void *)vcpu + offset);
3427 spin_unlock(&kvm_lock);
3431 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
3433 static const struct file_operations *stat_fops[] = {
3434 [KVM_STAT_VCPU] = &vcpu_stat_fops,
3435 [KVM_STAT_VM] = &vm_stat_fops,
3438 static int kvm_init_debug(void)
3441 struct kvm_stats_debugfs_item *p;
3443 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3444 if (kvm_debugfs_dir == NULL)
3447 for (p = debugfs_entries; p->name; ++p) {
3448 if (!debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
3449 (void *)(long)p->offset,
3450 stat_fops[p->kind]))
3457 debugfs_remove_recursive(kvm_debugfs_dir);
3462 static int kvm_suspend(void)
3464 if (kvm_usage_count)
3465 hardware_disable_nolock(NULL);
3469 static void kvm_resume(void)
3471 if (kvm_usage_count) {
3472 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3473 hardware_enable_nolock(NULL);
3477 static struct syscore_ops kvm_syscore_ops = {
3478 .suspend = kvm_suspend,
3479 .resume = kvm_resume,
3483 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3485 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3488 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3490 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3492 if (vcpu->preempted)
3493 vcpu->preempted = false;
3495 kvm_arch_sched_in(vcpu, cpu);
3497 kvm_arch_vcpu_load(vcpu, cpu);
3500 static void kvm_sched_out(struct preempt_notifier *pn,
3501 struct task_struct *next)
3503 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3505 if (current->state == TASK_RUNNING)
3506 vcpu->preempted = true;
3507 kvm_arch_vcpu_put(vcpu);
3510 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3511 struct module *module)
3516 r = kvm_arch_init(opaque);
3521 * kvm_arch_init makes sure there's at most one caller
3522 * for architectures that support multiple implementations,
3523 * like intel and amd on x86.
3524 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3525 * conflicts in case kvm is already setup for another implementation.
3527 r = kvm_irqfd_init();
3531 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
3536 r = kvm_arch_hardware_setup();
3540 for_each_online_cpu(cpu) {
3541 smp_call_function_single(cpu,
3542 kvm_arch_check_processor_compat,
3548 r = register_cpu_notifier(&kvm_cpu_notifier);
3551 register_reboot_notifier(&kvm_reboot_notifier);
3553 /* A kmem cache lets us meet the alignment requirements of fx_save. */
3555 vcpu_align = __alignof__(struct kvm_vcpu);
3556 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
3558 if (!kvm_vcpu_cache) {
3563 r = kvm_async_pf_init();
3567 kvm_chardev_ops.owner = module;
3568 kvm_vm_fops.owner = module;
3569 kvm_vcpu_fops.owner = module;
3571 r = misc_register(&kvm_dev);
3573 pr_err("kvm: misc device register failed\n");
3577 register_syscore_ops(&kvm_syscore_ops);
3579 kvm_preempt_ops.sched_in = kvm_sched_in;
3580 kvm_preempt_ops.sched_out = kvm_sched_out;
3582 r = kvm_init_debug();
3584 pr_err("kvm: create debugfs files failed\n");
3588 r = kvm_vfio_ops_init();
3594 unregister_syscore_ops(&kvm_syscore_ops);
3595 misc_deregister(&kvm_dev);
3597 kvm_async_pf_deinit();
3599 kmem_cache_destroy(kvm_vcpu_cache);
3601 unregister_reboot_notifier(&kvm_reboot_notifier);
3602 unregister_cpu_notifier(&kvm_cpu_notifier);
3605 kvm_arch_hardware_unsetup();
3607 free_cpumask_var(cpus_hardware_enabled);
3615 EXPORT_SYMBOL_GPL(kvm_init);
3619 debugfs_remove_recursive(kvm_debugfs_dir);
3620 misc_deregister(&kvm_dev);
3621 kmem_cache_destroy(kvm_vcpu_cache);
3622 kvm_async_pf_deinit();
3623 unregister_syscore_ops(&kvm_syscore_ops);
3624 unregister_reboot_notifier(&kvm_reboot_notifier);
3625 unregister_cpu_notifier(&kvm_cpu_notifier);
3626 on_each_cpu(hardware_disable_nolock, NULL, 1);
3627 kvm_arch_hardware_unsetup();
3630 free_cpumask_var(cpus_hardware_enabled);
3631 kvm_vfio_ops_exit();
3633 EXPORT_SYMBOL_GPL(kvm_exit);