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.
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/uaccess.h>
56 #include <asm/pgtable.h>
58 #include "coalesced_mmio.h"
61 #define CREATE_TRACE_POINTS
62 #include <trace/events/kvm.h>
64 MODULE_AUTHOR("Qumranet");
65 MODULE_LICENSE("GPL");
70 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
73 DEFINE_RAW_SPINLOCK(kvm_lock);
76 static cpumask_var_t cpus_hardware_enabled;
77 static int kvm_usage_count = 0;
78 static atomic_t hardware_enable_failed;
80 struct kmem_cache *kvm_vcpu_cache;
81 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
83 static __read_mostly struct preempt_ops kvm_preempt_ops;
85 struct dentry *kvm_debugfs_dir;
87 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
90 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
93 static int hardware_enable_all(void);
94 static void hardware_disable_all(void);
96 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
99 EXPORT_SYMBOL_GPL(kvm_rebooting);
101 static bool largepages_enabled = true;
103 bool kvm_is_mmio_pfn(pfn_t pfn)
105 if (pfn_valid(pfn)) {
107 struct page *tail = pfn_to_page(pfn);
108 struct page *head = compound_trans_head(tail);
109 reserved = PageReserved(head);
112 * "head" is not a dangling pointer
113 * (compound_trans_head takes care of that)
114 * but the hugepage may have been splitted
115 * from under us (and we may not hold a
116 * reference count on the head page so it can
117 * be reused before we run PageReferenced), so
118 * we've to check PageTail before returning
125 return PageReserved(tail);
132 * Switches to specified vcpu, until a matching vcpu_put()
134 int vcpu_load(struct kvm_vcpu *vcpu)
138 if (mutex_lock_killable(&vcpu->mutex))
140 if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
141 /* The thread running this VCPU changed. */
142 struct pid *oldpid = vcpu->pid;
143 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
144 rcu_assign_pointer(vcpu->pid, newpid);
149 preempt_notifier_register(&vcpu->preempt_notifier);
150 kvm_arch_vcpu_load(vcpu, cpu);
155 void vcpu_put(struct kvm_vcpu *vcpu)
158 kvm_arch_vcpu_put(vcpu);
159 preempt_notifier_unregister(&vcpu->preempt_notifier);
161 mutex_unlock(&vcpu->mutex);
164 static void ack_flush(void *_completed)
168 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
173 struct kvm_vcpu *vcpu;
175 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
178 kvm_for_each_vcpu(i, vcpu, kvm) {
179 kvm_make_request(req, vcpu);
182 /* Set ->requests bit before we read ->mode */
185 if (cpus != NULL && cpu != -1 && cpu != me &&
186 kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
187 cpumask_set_cpu(cpu, cpus);
189 if (unlikely(cpus == NULL))
190 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
191 else if (!cpumask_empty(cpus))
192 smp_call_function_many(cpus, ack_flush, NULL, 1);
196 free_cpumask_var(cpus);
200 void kvm_flush_remote_tlbs(struct kvm *kvm)
202 long dirty_count = kvm->tlbs_dirty;
205 if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
206 ++kvm->stat.remote_tlb_flush;
207 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
210 void kvm_reload_remote_mmus(struct kvm *kvm)
212 make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
215 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
217 make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
220 void kvm_make_update_eoibitmap_request(struct kvm *kvm)
222 make_all_cpus_request(kvm, KVM_REQ_EOIBITMAP);
225 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
230 mutex_init(&vcpu->mutex);
235 init_waitqueue_head(&vcpu->wq);
236 kvm_async_pf_vcpu_init(vcpu);
238 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
243 vcpu->run = page_address(page);
245 kvm_vcpu_set_in_spin_loop(vcpu, false);
246 kvm_vcpu_set_dy_eligible(vcpu, false);
248 r = kvm_arch_vcpu_init(vcpu);
254 free_page((unsigned long)vcpu->run);
258 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
260 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
263 kvm_arch_vcpu_uninit(vcpu);
264 free_page((unsigned long)vcpu->run);
266 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
268 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
269 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
271 return container_of(mn, struct kvm, mmu_notifier);
274 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
275 struct mm_struct *mm,
276 unsigned long address)
278 struct kvm *kvm = mmu_notifier_to_kvm(mn);
279 int need_tlb_flush, idx;
282 * When ->invalidate_page runs, the linux pte has been zapped
283 * already but the page is still allocated until
284 * ->invalidate_page returns. So if we increase the sequence
285 * here the kvm page fault will notice if the spte can't be
286 * established because the page is going to be freed. If
287 * instead the kvm page fault establishes the spte before
288 * ->invalidate_page runs, kvm_unmap_hva will release it
291 * The sequence increase only need to be seen at spin_unlock
292 * time, and not at spin_lock time.
294 * Increasing the sequence after the spin_unlock would be
295 * unsafe because the kvm page fault could then establish the
296 * pte after kvm_unmap_hva returned, without noticing the page
297 * is going to be freed.
299 idx = srcu_read_lock(&kvm->srcu);
300 spin_lock(&kvm->mmu_lock);
302 kvm->mmu_notifier_seq++;
303 need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
304 /* we've to flush the tlb before the pages can be freed */
306 kvm_flush_remote_tlbs(kvm);
308 spin_unlock(&kvm->mmu_lock);
309 srcu_read_unlock(&kvm->srcu, idx);
312 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
313 struct mm_struct *mm,
314 unsigned long address,
317 struct kvm *kvm = mmu_notifier_to_kvm(mn);
320 idx = srcu_read_lock(&kvm->srcu);
321 spin_lock(&kvm->mmu_lock);
322 kvm->mmu_notifier_seq++;
323 kvm_set_spte_hva(kvm, address, pte);
324 spin_unlock(&kvm->mmu_lock);
325 srcu_read_unlock(&kvm->srcu, idx);
328 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
329 struct mm_struct *mm,
333 struct kvm *kvm = mmu_notifier_to_kvm(mn);
334 int need_tlb_flush = 0, idx;
336 idx = srcu_read_lock(&kvm->srcu);
337 spin_lock(&kvm->mmu_lock);
339 * The count increase must become visible at unlock time as no
340 * spte can be established without taking the mmu_lock and
341 * count is also read inside the mmu_lock critical section.
343 kvm->mmu_notifier_count++;
344 need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
345 need_tlb_flush |= kvm->tlbs_dirty;
346 /* we've to flush the tlb before the pages can be freed */
348 kvm_flush_remote_tlbs(kvm);
350 spin_unlock(&kvm->mmu_lock);
351 srcu_read_unlock(&kvm->srcu, idx);
354 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
355 struct mm_struct *mm,
359 struct kvm *kvm = mmu_notifier_to_kvm(mn);
361 spin_lock(&kvm->mmu_lock);
363 * This sequence increase will notify the kvm page fault that
364 * the page that is going to be mapped in the spte could have
367 kvm->mmu_notifier_seq++;
370 * The above sequence increase must be visible before the
371 * below count decrease, which is ensured by the smp_wmb above
372 * in conjunction with the smp_rmb in mmu_notifier_retry().
374 kvm->mmu_notifier_count--;
375 spin_unlock(&kvm->mmu_lock);
377 BUG_ON(kvm->mmu_notifier_count < 0);
380 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
381 struct mm_struct *mm,
382 unsigned long address)
384 struct kvm *kvm = mmu_notifier_to_kvm(mn);
387 idx = srcu_read_lock(&kvm->srcu);
388 spin_lock(&kvm->mmu_lock);
390 young = kvm_age_hva(kvm, address);
392 kvm_flush_remote_tlbs(kvm);
394 spin_unlock(&kvm->mmu_lock);
395 srcu_read_unlock(&kvm->srcu, idx);
400 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
401 struct mm_struct *mm,
402 unsigned long address)
404 struct kvm *kvm = mmu_notifier_to_kvm(mn);
407 idx = srcu_read_lock(&kvm->srcu);
408 spin_lock(&kvm->mmu_lock);
409 young = kvm_test_age_hva(kvm, address);
410 spin_unlock(&kvm->mmu_lock);
411 srcu_read_unlock(&kvm->srcu, idx);
416 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
417 struct mm_struct *mm)
419 struct kvm *kvm = mmu_notifier_to_kvm(mn);
422 idx = srcu_read_lock(&kvm->srcu);
423 kvm_arch_flush_shadow_all(kvm);
424 srcu_read_unlock(&kvm->srcu, idx);
427 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
428 .invalidate_page = kvm_mmu_notifier_invalidate_page,
429 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
430 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
431 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
432 .test_young = kvm_mmu_notifier_test_young,
433 .change_pte = kvm_mmu_notifier_change_pte,
434 .release = kvm_mmu_notifier_release,
437 static int kvm_init_mmu_notifier(struct kvm *kvm)
439 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
440 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
443 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
445 static int kvm_init_mmu_notifier(struct kvm *kvm)
450 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
452 static void kvm_init_memslots_id(struct kvm *kvm)
455 struct kvm_memslots *slots = kvm->memslots;
457 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
458 slots->id_to_index[i] = slots->memslots[i].id = i;
461 static struct kvm *kvm_create_vm(unsigned long type)
464 struct kvm *kvm = kvm_arch_alloc_vm();
467 return ERR_PTR(-ENOMEM);
469 r = kvm_arch_init_vm(kvm, type);
471 goto out_err_nodisable;
473 r = hardware_enable_all();
475 goto out_err_nodisable;
477 #ifdef CONFIG_HAVE_KVM_IRQCHIP
478 INIT_HLIST_HEAD(&kvm->mask_notifier_list);
479 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
482 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
485 kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
488 kvm_init_memslots_id(kvm);
489 if (init_srcu_struct(&kvm->srcu))
491 for (i = 0; i < KVM_NR_BUSES; i++) {
492 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
498 spin_lock_init(&kvm->mmu_lock);
499 kvm->mm = current->mm;
500 atomic_inc(&kvm->mm->mm_count);
501 kvm_eventfd_init(kvm);
502 mutex_init(&kvm->lock);
503 mutex_init(&kvm->irq_lock);
504 mutex_init(&kvm->slots_lock);
505 atomic_set(&kvm->users_count, 1);
507 r = kvm_init_mmu_notifier(kvm);
511 raw_spin_lock(&kvm_lock);
512 list_add(&kvm->vm_list, &vm_list);
513 raw_spin_unlock(&kvm_lock);
518 cleanup_srcu_struct(&kvm->srcu);
520 hardware_disable_all();
522 for (i = 0; i < KVM_NR_BUSES; i++)
523 kfree(kvm->buses[i]);
524 kfree(kvm->memslots);
525 kvm_arch_free_vm(kvm);
530 * Avoid using vmalloc for a small buffer.
531 * Should not be used when the size is statically known.
533 void *kvm_kvzalloc(unsigned long size)
535 if (size > PAGE_SIZE)
536 return vzalloc(size);
538 return kzalloc(size, GFP_KERNEL);
541 void kvm_kvfree(const void *addr)
543 if (is_vmalloc_addr(addr))
549 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
551 if (!memslot->dirty_bitmap)
554 kvm_kvfree(memslot->dirty_bitmap);
555 memslot->dirty_bitmap = NULL;
559 * Free any memory in @free but not in @dont.
561 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
562 struct kvm_memory_slot *dont)
564 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
565 kvm_destroy_dirty_bitmap(free);
567 kvm_arch_free_memslot(free, dont);
572 void kvm_free_physmem(struct kvm *kvm)
574 struct kvm_memslots *slots = kvm->memslots;
575 struct kvm_memory_slot *memslot;
577 kvm_for_each_memslot(memslot, slots)
578 kvm_free_physmem_slot(memslot, NULL);
580 kfree(kvm->memslots);
583 static void kvm_destroy_vm(struct kvm *kvm)
586 struct mm_struct *mm = kvm->mm;
588 kvm_arch_sync_events(kvm);
589 raw_spin_lock(&kvm_lock);
590 list_del(&kvm->vm_list);
591 raw_spin_unlock(&kvm_lock);
592 kvm_free_irq_routing(kvm);
593 for (i = 0; i < KVM_NR_BUSES; i++)
594 kvm_io_bus_destroy(kvm->buses[i]);
595 kvm_coalesced_mmio_free(kvm);
596 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
597 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
599 kvm_arch_flush_shadow_all(kvm);
601 kvm_arch_destroy_vm(kvm);
602 kvm_free_physmem(kvm);
603 cleanup_srcu_struct(&kvm->srcu);
604 kvm_arch_free_vm(kvm);
605 hardware_disable_all();
609 void kvm_get_kvm(struct kvm *kvm)
611 atomic_inc(&kvm->users_count);
613 EXPORT_SYMBOL_GPL(kvm_get_kvm);
615 void kvm_put_kvm(struct kvm *kvm)
617 if (atomic_dec_and_test(&kvm->users_count))
620 EXPORT_SYMBOL_GPL(kvm_put_kvm);
623 static int kvm_vm_release(struct inode *inode, struct file *filp)
625 struct kvm *kvm = filp->private_data;
627 kvm_irqfd_release(kvm);
634 * Allocation size is twice as large as the actual dirty bitmap size.
635 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
637 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
640 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
642 memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
643 if (!memslot->dirty_bitmap)
646 #endif /* !CONFIG_S390 */
650 static int cmp_memslot(const void *slot1, const void *slot2)
652 struct kvm_memory_slot *s1, *s2;
654 s1 = (struct kvm_memory_slot *)slot1;
655 s2 = (struct kvm_memory_slot *)slot2;
657 if (s1->npages < s2->npages)
659 if (s1->npages > s2->npages)
666 * Sort the memslots base on its size, so the larger slots
667 * will get better fit.
669 static void sort_memslots(struct kvm_memslots *slots)
673 sort(slots->memslots, KVM_MEM_SLOTS_NUM,
674 sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
676 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
677 slots->id_to_index[slots->memslots[i].id] = i;
680 void update_memslots(struct kvm_memslots *slots, struct kvm_memory_slot *new,
685 struct kvm_memory_slot *old = id_to_memslot(slots, id);
686 unsigned long npages = old->npages;
689 if (new->npages != npages)
690 sort_memslots(slots);
693 slots->generation = last_generation + 1;
696 static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
698 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
700 #ifdef KVM_CAP_READONLY_MEM
701 valid_flags |= KVM_MEM_READONLY;
704 if (mem->flags & ~valid_flags)
710 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
711 struct kvm_memslots *slots, struct kvm_memory_slot *new)
713 struct kvm_memslots *old_memslots = kvm->memslots;
715 update_memslots(slots, new, kvm->memslots->generation);
716 rcu_assign_pointer(kvm->memslots, slots);
717 synchronize_srcu_expedited(&kvm->srcu);
722 * Allocate some memory and give it an address in the guest physical address
725 * Discontiguous memory is allowed, mostly for framebuffers.
727 * Must be called holding mmap_sem for write.
729 int __kvm_set_memory_region(struct kvm *kvm,
730 struct kvm_userspace_memory_region *mem,
735 unsigned long npages;
736 struct kvm_memory_slot *slot;
737 struct kvm_memory_slot old, new;
738 struct kvm_memslots *slots = NULL, *old_memslots;
739 bool old_iommu_mapped;
741 r = check_memory_region_flags(mem);
746 /* General sanity checks */
747 if (mem->memory_size & (PAGE_SIZE - 1))
749 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
751 /* We can read the guest memory with __xxx_user() later on. */
753 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
754 !access_ok(VERIFY_WRITE,
755 (void __user *)(unsigned long)mem->userspace_addr,
758 if (mem->slot >= KVM_MEM_SLOTS_NUM)
760 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
763 slot = id_to_memslot(kvm->memslots, mem->slot);
764 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
765 npages = mem->memory_size >> PAGE_SHIFT;
768 if (npages > KVM_MEM_MAX_NR_PAGES)
772 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
777 new.base_gfn = base_gfn;
779 new.flags = mem->flags;
781 old_iommu_mapped = old.npages;
784 * Disallow changing a memory slot's size or changing anything about
785 * zero sized slots that doesn't involve making them non-zero.
788 if (npages && old.npages && npages != old.npages)
790 if (!npages && !old.npages)
793 if ((npages && !old.npages) || (base_gfn != old.base_gfn)) {
794 /* Check for overlaps */
796 kvm_for_each_memslot(slot, kvm->memslots) {
797 if ((slot->id >= KVM_USER_MEM_SLOTS) ||
798 (slot->id == mem->slot))
800 if (!((base_gfn + npages <= slot->base_gfn) ||
801 (base_gfn >= slot->base_gfn + slot->npages)))
806 /* Free page dirty bitmap if unneeded */
807 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
808 new.dirty_bitmap = NULL;
813 * Allocate if a slot is being created. If modifying a slot,
814 * the userspace_addr cannot change.
817 new.user_alloc = user_alloc;
818 new.userspace_addr = mem->userspace_addr;
820 if (kvm_arch_create_memslot(&new, npages))
822 } else if (npages && mem->userspace_addr != old.userspace_addr) {
827 /* Allocate page dirty bitmap if needed */
828 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
829 if (kvm_create_dirty_bitmap(&new) < 0)
833 if (!npages || base_gfn != old.base_gfn) {
835 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
839 slot = id_to_memslot(slots, mem->slot);
840 slot->flags |= KVM_MEMSLOT_INVALID;
842 old_memslots = install_new_memslots(kvm, slots, NULL);
844 /* slot was deleted or moved, clear iommu mapping */
845 kvm_iommu_unmap_pages(kvm, &old);
846 old_iommu_mapped = false;
847 /* From this point no new shadow pages pointing to a deleted,
848 * or moved, memslot will be created.
850 * validation of sp->gfn happens in:
851 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
852 * - kvm_is_visible_gfn (mmu_check_roots)
854 kvm_arch_flush_shadow_memslot(kvm, slot);
855 slots = old_memslots;
858 r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
864 * We can re-use the old_memslots from above, the only difference
865 * from the currently installed memslots is the invalid flag. This
866 * will get overwritten by update_memslots anyway.
869 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
876 * IOMMU mapping: New slots need to be mapped. Old slots need to be
877 * un-mapped and re-mapped if their base changes or if flags that the
878 * iommu cares about change (read-only). Base change unmapping is
879 * handled above with slot deletion, so we only unmap incompatible
880 * flags here. Anything else the iommu might care about for existing
881 * slots (size changes, userspace addr changes) is disallowed above,
882 * so any other attribute changes getting here can be skipped.
885 if (old_iommu_mapped &&
886 ((new.flags ^ old.flags) & KVM_MEM_READONLY)) {
887 kvm_iommu_unmap_pages(kvm, &old);
888 old_iommu_mapped = false;
891 if (!old_iommu_mapped) {
892 r = kvm_iommu_map_pages(kvm, &new);
898 /* actual memory is freed via old in kvm_free_physmem_slot below */
900 new.dirty_bitmap = NULL;
901 memset(&new.arch, 0, sizeof(new.arch));
904 old_memslots = install_new_memslots(kvm, slots, &new);
906 kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
908 kvm_free_physmem_slot(&old, &new);
916 kvm_free_physmem_slot(&new, &old);
920 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
922 int kvm_set_memory_region(struct kvm *kvm,
923 struct kvm_userspace_memory_region *mem,
928 mutex_lock(&kvm->slots_lock);
929 r = __kvm_set_memory_region(kvm, mem, user_alloc);
930 mutex_unlock(&kvm->slots_lock);
933 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
935 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
937 kvm_userspace_memory_region *mem,
940 if (mem->slot >= KVM_USER_MEM_SLOTS)
942 return kvm_set_memory_region(kvm, mem, user_alloc);
945 int kvm_get_dirty_log(struct kvm *kvm,
946 struct kvm_dirty_log *log, int *is_dirty)
948 struct kvm_memory_slot *memslot;
951 unsigned long any = 0;
954 if (log->slot >= KVM_USER_MEM_SLOTS)
957 memslot = id_to_memslot(kvm->memslots, log->slot);
959 if (!memslot->dirty_bitmap)
962 n = kvm_dirty_bitmap_bytes(memslot);
964 for (i = 0; !any && i < n/sizeof(long); ++i)
965 any = memslot->dirty_bitmap[i];
968 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
979 bool kvm_largepages_enabled(void)
981 return largepages_enabled;
984 void kvm_disable_largepages(void)
986 largepages_enabled = false;
988 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
990 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
992 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
994 EXPORT_SYMBOL_GPL(gfn_to_memslot);
996 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
998 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1000 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1001 memslot->flags & KVM_MEMSLOT_INVALID)
1006 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1008 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1010 struct vm_area_struct *vma;
1011 unsigned long addr, size;
1015 addr = gfn_to_hva(kvm, gfn);
1016 if (kvm_is_error_hva(addr))
1019 down_read(¤t->mm->mmap_sem);
1020 vma = find_vma(current->mm, addr);
1024 size = vma_kernel_pagesize(vma);
1027 up_read(¤t->mm->mmap_sem);
1032 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1034 return slot->flags & KVM_MEM_READONLY;
1037 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1038 gfn_t *nr_pages, bool write)
1040 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1041 return KVM_HVA_ERR_BAD;
1043 if (memslot_is_readonly(slot) && write)
1044 return KVM_HVA_ERR_RO_BAD;
1047 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1049 return __gfn_to_hva_memslot(slot, gfn);
1052 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1055 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1058 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1061 return gfn_to_hva_many(slot, gfn, NULL);
1063 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1065 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1067 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1069 EXPORT_SYMBOL_GPL(gfn_to_hva);
1072 * The hva returned by this function is only allowed to be read.
1073 * It should pair with kvm_read_hva() or kvm_read_hva_atomic().
1075 static unsigned long gfn_to_hva_read(struct kvm *kvm, gfn_t gfn)
1077 return __gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL, false);
1080 static int kvm_read_hva(void *data, void __user *hva, int len)
1082 return __copy_from_user(data, hva, len);
1085 static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1087 return __copy_from_user_inatomic(data, hva, len);
1090 int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1091 unsigned long start, int write, struct page **page)
1093 int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1096 flags |= FOLL_WRITE;
1098 return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1101 static inline int check_user_page_hwpoison(unsigned long addr)
1103 int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1105 rc = __get_user_pages(current, current->mm, addr, 1,
1106 flags, NULL, NULL, NULL);
1107 return rc == -EHWPOISON;
1111 * The atomic path to get the writable pfn which will be stored in @pfn,
1112 * true indicates success, otherwise false is returned.
1114 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1115 bool write_fault, bool *writable, pfn_t *pfn)
1117 struct page *page[1];
1120 if (!(async || atomic))
1124 * Fast pin a writable pfn only if it is a write fault request
1125 * or the caller allows to map a writable pfn for a read fault
1128 if (!(write_fault || writable))
1131 npages = __get_user_pages_fast(addr, 1, 1, page);
1133 *pfn = page_to_pfn(page[0]);
1144 * The slow path to get the pfn of the specified host virtual address,
1145 * 1 indicates success, -errno is returned if error is detected.
1147 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1148 bool *writable, pfn_t *pfn)
1150 struct page *page[1];
1156 *writable = write_fault;
1159 down_read(¤t->mm->mmap_sem);
1160 npages = get_user_page_nowait(current, current->mm,
1161 addr, write_fault, page);
1162 up_read(¤t->mm->mmap_sem);
1164 npages = get_user_pages_fast(addr, 1, write_fault,
1169 /* map read fault as writable if possible */
1170 if (unlikely(!write_fault) && writable) {
1171 struct page *wpage[1];
1173 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1182 *pfn = page_to_pfn(page[0]);
1186 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1188 if (unlikely(!(vma->vm_flags & VM_READ)))
1191 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1198 * Pin guest page in memory and return its pfn.
1199 * @addr: host virtual address which maps memory to the guest
1200 * @atomic: whether this function can sleep
1201 * @async: whether this function need to wait IO complete if the
1202 * host page is not in the memory
1203 * @write_fault: whether we should get a writable host page
1204 * @writable: whether it allows to map a writable host page for !@write_fault
1206 * The function will map a writable host page for these two cases:
1207 * 1): @write_fault = true
1208 * 2): @write_fault = false && @writable, @writable will tell the caller
1209 * whether the mapping is writable.
1211 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1212 bool write_fault, bool *writable)
1214 struct vm_area_struct *vma;
1218 /* we can do it either atomically or asynchronously, not both */
1219 BUG_ON(atomic && async);
1221 if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1225 return KVM_PFN_ERR_FAULT;
1227 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1231 down_read(¤t->mm->mmap_sem);
1232 if (npages == -EHWPOISON ||
1233 (!async && check_user_page_hwpoison(addr))) {
1234 pfn = KVM_PFN_ERR_HWPOISON;
1238 vma = find_vma_intersection(current->mm, addr, addr + 1);
1241 pfn = KVM_PFN_ERR_FAULT;
1242 else if ((vma->vm_flags & VM_PFNMAP)) {
1243 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1245 BUG_ON(!kvm_is_mmio_pfn(pfn));
1247 if (async && vma_is_valid(vma, write_fault))
1249 pfn = KVM_PFN_ERR_FAULT;
1252 up_read(¤t->mm->mmap_sem);
1257 __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1258 bool *async, bool write_fault, bool *writable)
1260 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1262 if (addr == KVM_HVA_ERR_RO_BAD)
1263 return KVM_PFN_ERR_RO_FAULT;
1265 if (kvm_is_error_hva(addr))
1266 return KVM_PFN_NOSLOT;
1268 /* Do not map writable pfn in the readonly memslot. */
1269 if (writable && memslot_is_readonly(slot)) {
1274 return hva_to_pfn(addr, atomic, async, write_fault,
1278 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1279 bool write_fault, bool *writable)
1281 struct kvm_memory_slot *slot;
1286 slot = gfn_to_memslot(kvm, gfn);
1288 return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1292 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1294 return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1296 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1298 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1299 bool write_fault, bool *writable)
1301 return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1303 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1305 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1307 return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1309 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1311 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1314 return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1316 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1318 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1320 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1323 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1325 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1327 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1329 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1335 addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1336 if (kvm_is_error_hva(addr))
1339 if (entry < nr_pages)
1342 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1344 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1346 static struct page *kvm_pfn_to_page(pfn_t pfn)
1348 if (is_error_noslot_pfn(pfn))
1349 return KVM_ERR_PTR_BAD_PAGE;
1351 if (kvm_is_mmio_pfn(pfn)) {
1353 return KVM_ERR_PTR_BAD_PAGE;
1356 return pfn_to_page(pfn);
1359 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1363 pfn = gfn_to_pfn(kvm, gfn);
1365 return kvm_pfn_to_page(pfn);
1368 EXPORT_SYMBOL_GPL(gfn_to_page);
1370 void kvm_release_page_clean(struct page *page)
1372 WARN_ON(is_error_page(page));
1374 kvm_release_pfn_clean(page_to_pfn(page));
1376 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1378 void kvm_release_pfn_clean(pfn_t pfn)
1380 if (!is_error_noslot_pfn(pfn) && !kvm_is_mmio_pfn(pfn))
1381 put_page(pfn_to_page(pfn));
1383 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1385 void kvm_release_page_dirty(struct page *page)
1387 WARN_ON(is_error_page(page));
1389 kvm_release_pfn_dirty(page_to_pfn(page));
1391 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1393 void kvm_release_pfn_dirty(pfn_t pfn)
1395 kvm_set_pfn_dirty(pfn);
1396 kvm_release_pfn_clean(pfn);
1398 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1400 void kvm_set_page_dirty(struct page *page)
1402 kvm_set_pfn_dirty(page_to_pfn(page));
1404 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1406 void kvm_set_pfn_dirty(pfn_t pfn)
1408 if (!kvm_is_mmio_pfn(pfn)) {
1409 struct page *page = pfn_to_page(pfn);
1410 if (!PageReserved(page))
1414 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1416 void kvm_set_pfn_accessed(pfn_t pfn)
1418 if (!kvm_is_mmio_pfn(pfn))
1419 mark_page_accessed(pfn_to_page(pfn));
1421 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1423 void kvm_get_pfn(pfn_t pfn)
1425 if (!kvm_is_mmio_pfn(pfn))
1426 get_page(pfn_to_page(pfn));
1428 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1430 static int next_segment(unsigned long len, int offset)
1432 if (len > PAGE_SIZE - offset)
1433 return PAGE_SIZE - offset;
1438 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1444 addr = gfn_to_hva_read(kvm, gfn);
1445 if (kvm_is_error_hva(addr))
1447 r = kvm_read_hva(data, (void __user *)addr + offset, len);
1452 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1454 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1456 gfn_t gfn = gpa >> PAGE_SHIFT;
1458 int offset = offset_in_page(gpa);
1461 while ((seg = next_segment(len, offset)) != 0) {
1462 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1472 EXPORT_SYMBOL_GPL(kvm_read_guest);
1474 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1479 gfn_t gfn = gpa >> PAGE_SHIFT;
1480 int offset = offset_in_page(gpa);
1482 addr = gfn_to_hva_read(kvm, gfn);
1483 if (kvm_is_error_hva(addr))
1485 pagefault_disable();
1486 r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
1492 EXPORT_SYMBOL(kvm_read_guest_atomic);
1494 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1495 int offset, int len)
1500 addr = gfn_to_hva(kvm, gfn);
1501 if (kvm_is_error_hva(addr))
1503 r = __copy_to_user((void __user *)addr + offset, data, len);
1506 mark_page_dirty(kvm, gfn);
1509 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1511 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1514 gfn_t gfn = gpa >> PAGE_SHIFT;
1516 int offset = offset_in_page(gpa);
1519 while ((seg = next_segment(len, offset)) != 0) {
1520 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1531 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1534 struct kvm_memslots *slots = kvm_memslots(kvm);
1535 int offset = offset_in_page(gpa);
1536 gfn_t gfn = gpa >> PAGE_SHIFT;
1539 ghc->generation = slots->generation;
1540 ghc->memslot = gfn_to_memslot(kvm, gfn);
1541 ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1542 if (!kvm_is_error_hva(ghc->hva))
1549 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1551 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1552 void *data, unsigned long len)
1554 struct kvm_memslots *slots = kvm_memslots(kvm);
1557 if (slots->generation != ghc->generation)
1558 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1560 if (kvm_is_error_hva(ghc->hva))
1563 r = __copy_to_user((void __user *)ghc->hva, data, len);
1566 mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1570 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1572 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1573 void *data, unsigned long len)
1575 struct kvm_memslots *slots = kvm_memslots(kvm);
1578 if (slots->generation != ghc->generation)
1579 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1581 if (kvm_is_error_hva(ghc->hva))
1584 r = __copy_from_user(data, (void __user *)ghc->hva, len);
1590 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1592 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1594 return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1597 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1599 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1601 gfn_t gfn = gpa >> PAGE_SHIFT;
1603 int offset = offset_in_page(gpa);
1606 while ((seg = next_segment(len, offset)) != 0) {
1607 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1616 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1618 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1621 if (memslot && memslot->dirty_bitmap) {
1622 unsigned long rel_gfn = gfn - memslot->base_gfn;
1624 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1628 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1630 struct kvm_memory_slot *memslot;
1632 memslot = gfn_to_memslot(kvm, gfn);
1633 mark_page_dirty_in_slot(kvm, memslot, gfn);
1637 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1639 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1644 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1646 if (kvm_arch_vcpu_runnable(vcpu)) {
1647 kvm_make_request(KVM_REQ_UNHALT, vcpu);
1650 if (kvm_cpu_has_pending_timer(vcpu))
1652 if (signal_pending(current))
1658 finish_wait(&vcpu->wq, &wait);
1663 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1665 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1668 int cpu = vcpu->cpu;
1669 wait_queue_head_t *wqp;
1671 wqp = kvm_arch_vcpu_wq(vcpu);
1672 if (waitqueue_active(wqp)) {
1673 wake_up_interruptible(wqp);
1674 ++vcpu->stat.halt_wakeup;
1678 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1679 if (kvm_arch_vcpu_should_kick(vcpu))
1680 smp_send_reschedule(cpu);
1683 #endif /* !CONFIG_S390 */
1685 void kvm_resched(struct kvm_vcpu *vcpu)
1687 if (!need_resched())
1691 EXPORT_SYMBOL_GPL(kvm_resched);
1693 bool kvm_vcpu_yield_to(struct kvm_vcpu *target)
1696 struct task_struct *task = NULL;
1700 pid = rcu_dereference(target->pid);
1702 task = get_pid_task(target->pid, PIDTYPE_PID);
1706 if (task->flags & PF_VCPU) {
1707 put_task_struct(task);
1710 ret = yield_to(task, 1);
1711 put_task_struct(task);
1715 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1717 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1719 * Helper that checks whether a VCPU is eligible for directed yield.
1720 * Most eligible candidate to yield is decided by following heuristics:
1722 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1723 * (preempted lock holder), indicated by @in_spin_loop.
1724 * Set at the beiginning and cleared at the end of interception/PLE handler.
1726 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1727 * chance last time (mostly it has become eligible now since we have probably
1728 * yielded to lockholder in last iteration. This is done by toggling
1729 * @dy_eligible each time a VCPU checked for eligibility.)
1731 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1732 * to preempted lock-holder could result in wrong VCPU selection and CPU
1733 * burning. Giving priority for a potential lock-holder increases lock
1736 * Since algorithm is based on heuristics, accessing another VCPU data without
1737 * locking does not harm. It may result in trying to yield to same VCPU, fail
1738 * and continue with next VCPU and so on.
1740 bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1744 eligible = !vcpu->spin_loop.in_spin_loop ||
1745 (vcpu->spin_loop.in_spin_loop &&
1746 vcpu->spin_loop.dy_eligible);
1748 if (vcpu->spin_loop.in_spin_loop)
1749 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1755 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1757 struct kvm *kvm = me->kvm;
1758 struct kvm_vcpu *vcpu;
1759 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1765 kvm_vcpu_set_in_spin_loop(me, true);
1767 * We boost the priority of a VCPU that is runnable but not
1768 * currently running, because it got preempted by something
1769 * else and called schedule in __vcpu_run. Hopefully that
1770 * VCPU is holding the lock that we need and will release it.
1771 * We approximate round-robin by starting at the last boosted VCPU.
1773 for (pass = 0; pass < 2 && !yielded && try; pass++) {
1774 kvm_for_each_vcpu(i, vcpu, kvm) {
1775 if (!pass && i <= last_boosted_vcpu) {
1776 i = last_boosted_vcpu;
1778 } else if (pass && i > last_boosted_vcpu)
1782 if (waitqueue_active(&vcpu->wq))
1784 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1787 yielded = kvm_vcpu_yield_to(vcpu);
1789 kvm->last_boosted_vcpu = i;
1791 } else if (yielded < 0) {
1798 kvm_vcpu_set_in_spin_loop(me, false);
1800 /* Ensure vcpu is not eligible during next spinloop */
1801 kvm_vcpu_set_dy_eligible(me, false);
1803 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1805 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1807 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1810 if (vmf->pgoff == 0)
1811 page = virt_to_page(vcpu->run);
1813 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1814 page = virt_to_page(vcpu->arch.pio_data);
1816 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1817 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1818 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1821 return kvm_arch_vcpu_fault(vcpu, vmf);
1827 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1828 .fault = kvm_vcpu_fault,
1831 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1833 vma->vm_ops = &kvm_vcpu_vm_ops;
1837 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1839 struct kvm_vcpu *vcpu = filp->private_data;
1841 kvm_put_kvm(vcpu->kvm);
1845 static struct file_operations kvm_vcpu_fops = {
1846 .release = kvm_vcpu_release,
1847 .unlocked_ioctl = kvm_vcpu_ioctl,
1848 #ifdef CONFIG_COMPAT
1849 .compat_ioctl = kvm_vcpu_compat_ioctl,
1851 .mmap = kvm_vcpu_mmap,
1852 .llseek = noop_llseek,
1856 * Allocates an inode for the vcpu.
1858 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1860 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1864 * Creates some virtual cpus. Good luck creating more than one.
1866 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1869 struct kvm_vcpu *vcpu, *v;
1871 vcpu = kvm_arch_vcpu_create(kvm, id);
1873 return PTR_ERR(vcpu);
1875 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1877 r = kvm_arch_vcpu_setup(vcpu);
1881 mutex_lock(&kvm->lock);
1882 if (!kvm_vcpu_compatible(vcpu)) {
1884 goto unlock_vcpu_destroy;
1886 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1888 goto unlock_vcpu_destroy;
1891 kvm_for_each_vcpu(r, v, kvm)
1892 if (v->vcpu_id == id) {
1894 goto unlock_vcpu_destroy;
1897 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1899 /* Now it's all set up, let userspace reach it */
1901 r = create_vcpu_fd(vcpu);
1904 goto unlock_vcpu_destroy;
1907 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1909 atomic_inc(&kvm->online_vcpus);
1911 mutex_unlock(&kvm->lock);
1912 kvm_arch_vcpu_postcreate(vcpu);
1915 unlock_vcpu_destroy:
1916 mutex_unlock(&kvm->lock);
1918 kvm_arch_vcpu_destroy(vcpu);
1922 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1925 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1926 vcpu->sigset_active = 1;
1927 vcpu->sigset = *sigset;
1929 vcpu->sigset_active = 0;
1933 static long kvm_vcpu_ioctl(struct file *filp,
1934 unsigned int ioctl, unsigned long arg)
1936 struct kvm_vcpu *vcpu = filp->private_data;
1937 void __user *argp = (void __user *)arg;
1939 struct kvm_fpu *fpu = NULL;
1940 struct kvm_sregs *kvm_sregs = NULL;
1942 if (vcpu->kvm->mm != current->mm)
1945 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1947 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1948 * so vcpu_load() would break it.
1950 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1951 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1955 r = vcpu_load(vcpu);
1963 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1964 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1966 case KVM_GET_REGS: {
1967 struct kvm_regs *kvm_regs;
1970 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1973 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1977 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1984 case KVM_SET_REGS: {
1985 struct kvm_regs *kvm_regs;
1988 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
1989 if (IS_ERR(kvm_regs)) {
1990 r = PTR_ERR(kvm_regs);
1993 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1997 case KVM_GET_SREGS: {
1998 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2002 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2006 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2011 case KVM_SET_SREGS: {
2012 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2013 if (IS_ERR(kvm_sregs)) {
2014 r = PTR_ERR(kvm_sregs);
2018 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2021 case KVM_GET_MP_STATE: {
2022 struct kvm_mp_state mp_state;
2024 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2028 if (copy_to_user(argp, &mp_state, sizeof mp_state))
2033 case KVM_SET_MP_STATE: {
2034 struct kvm_mp_state mp_state;
2037 if (copy_from_user(&mp_state, argp, sizeof mp_state))
2039 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2042 case KVM_TRANSLATE: {
2043 struct kvm_translation tr;
2046 if (copy_from_user(&tr, argp, sizeof tr))
2048 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2052 if (copy_to_user(argp, &tr, sizeof tr))
2057 case KVM_SET_GUEST_DEBUG: {
2058 struct kvm_guest_debug dbg;
2061 if (copy_from_user(&dbg, argp, sizeof dbg))
2063 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2066 case KVM_SET_SIGNAL_MASK: {
2067 struct kvm_signal_mask __user *sigmask_arg = argp;
2068 struct kvm_signal_mask kvm_sigmask;
2069 sigset_t sigset, *p;
2074 if (copy_from_user(&kvm_sigmask, argp,
2075 sizeof kvm_sigmask))
2078 if (kvm_sigmask.len != sizeof sigset)
2081 if (copy_from_user(&sigset, sigmask_arg->sigset,
2086 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2090 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2094 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2098 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2104 fpu = memdup_user(argp, sizeof(*fpu));
2110 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2114 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2123 #ifdef CONFIG_COMPAT
2124 static long kvm_vcpu_compat_ioctl(struct file *filp,
2125 unsigned int ioctl, unsigned long arg)
2127 struct kvm_vcpu *vcpu = filp->private_data;
2128 void __user *argp = compat_ptr(arg);
2131 if (vcpu->kvm->mm != current->mm)
2135 case KVM_SET_SIGNAL_MASK: {
2136 struct kvm_signal_mask __user *sigmask_arg = argp;
2137 struct kvm_signal_mask kvm_sigmask;
2138 compat_sigset_t csigset;
2143 if (copy_from_user(&kvm_sigmask, argp,
2144 sizeof kvm_sigmask))
2147 if (kvm_sigmask.len != sizeof csigset)
2150 if (copy_from_user(&csigset, sigmask_arg->sigset,
2153 sigset_from_compat(&sigset, &csigset);
2154 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2156 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2160 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2168 static long kvm_vm_ioctl(struct file *filp,
2169 unsigned int ioctl, unsigned long arg)
2171 struct kvm *kvm = filp->private_data;
2172 void __user *argp = (void __user *)arg;
2175 if (kvm->mm != current->mm)
2178 case KVM_CREATE_VCPU:
2179 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2181 case KVM_SET_USER_MEMORY_REGION: {
2182 struct kvm_userspace_memory_region kvm_userspace_mem;
2185 if (copy_from_user(&kvm_userspace_mem, argp,
2186 sizeof kvm_userspace_mem))
2189 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, true);
2192 case KVM_GET_DIRTY_LOG: {
2193 struct kvm_dirty_log log;
2196 if (copy_from_user(&log, argp, sizeof log))
2198 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2201 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2202 case KVM_REGISTER_COALESCED_MMIO: {
2203 struct kvm_coalesced_mmio_zone zone;
2205 if (copy_from_user(&zone, argp, sizeof zone))
2207 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2210 case KVM_UNREGISTER_COALESCED_MMIO: {
2211 struct kvm_coalesced_mmio_zone zone;
2213 if (copy_from_user(&zone, argp, sizeof zone))
2215 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2220 struct kvm_irqfd data;
2223 if (copy_from_user(&data, argp, sizeof data))
2225 r = kvm_irqfd(kvm, &data);
2228 case KVM_IOEVENTFD: {
2229 struct kvm_ioeventfd data;
2232 if (copy_from_user(&data, argp, sizeof data))
2234 r = kvm_ioeventfd(kvm, &data);
2237 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2238 case KVM_SET_BOOT_CPU_ID:
2240 mutex_lock(&kvm->lock);
2241 if (atomic_read(&kvm->online_vcpus) != 0)
2244 kvm->bsp_vcpu_id = arg;
2245 mutex_unlock(&kvm->lock);
2248 #ifdef CONFIG_HAVE_KVM_MSI
2249 case KVM_SIGNAL_MSI: {
2253 if (copy_from_user(&msi, argp, sizeof msi))
2255 r = kvm_send_userspace_msi(kvm, &msi);
2259 #ifdef __KVM_HAVE_IRQ_LINE
2260 case KVM_IRQ_LINE_STATUS:
2261 case KVM_IRQ_LINE: {
2262 struct kvm_irq_level irq_event;
2265 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2268 r = kvm_vm_ioctl_irq_line(kvm, &irq_event);
2273 if (ioctl == KVM_IRQ_LINE_STATUS) {
2274 if (copy_to_user(argp, &irq_event, sizeof irq_event))
2283 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2285 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2291 #ifdef CONFIG_COMPAT
2292 struct compat_kvm_dirty_log {
2296 compat_uptr_t dirty_bitmap; /* one bit per page */
2301 static long kvm_vm_compat_ioctl(struct file *filp,
2302 unsigned int ioctl, unsigned long arg)
2304 struct kvm *kvm = filp->private_data;
2307 if (kvm->mm != current->mm)
2310 case KVM_GET_DIRTY_LOG: {
2311 struct compat_kvm_dirty_log compat_log;
2312 struct kvm_dirty_log log;
2315 if (copy_from_user(&compat_log, (void __user *)arg,
2316 sizeof(compat_log)))
2318 log.slot = compat_log.slot;
2319 log.padding1 = compat_log.padding1;
2320 log.padding2 = compat_log.padding2;
2321 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2323 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2327 r = kvm_vm_ioctl(filp, ioctl, arg);
2335 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2337 struct page *page[1];
2340 gfn_t gfn = vmf->pgoff;
2341 struct kvm *kvm = vma->vm_file->private_data;
2343 addr = gfn_to_hva(kvm, gfn);
2344 if (kvm_is_error_hva(addr))
2345 return VM_FAULT_SIGBUS;
2347 npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2349 if (unlikely(npages != 1))
2350 return VM_FAULT_SIGBUS;
2352 vmf->page = page[0];
2356 static const struct vm_operations_struct kvm_vm_vm_ops = {
2357 .fault = kvm_vm_fault,
2360 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2362 vma->vm_ops = &kvm_vm_vm_ops;
2366 static struct file_operations kvm_vm_fops = {
2367 .release = kvm_vm_release,
2368 .unlocked_ioctl = kvm_vm_ioctl,
2369 #ifdef CONFIG_COMPAT
2370 .compat_ioctl = kvm_vm_compat_ioctl,
2372 .mmap = kvm_vm_mmap,
2373 .llseek = noop_llseek,
2376 static int kvm_dev_ioctl_create_vm(unsigned long type)
2381 kvm = kvm_create_vm(type);
2383 return PTR_ERR(kvm);
2384 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2385 r = kvm_coalesced_mmio_init(kvm);
2391 r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2398 static long kvm_dev_ioctl_check_extension_generic(long arg)
2401 case KVM_CAP_USER_MEMORY:
2402 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2403 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2404 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2405 case KVM_CAP_SET_BOOT_CPU_ID:
2407 case KVM_CAP_INTERNAL_ERROR_DATA:
2408 #ifdef CONFIG_HAVE_KVM_MSI
2409 case KVM_CAP_SIGNAL_MSI:
2412 #ifdef KVM_CAP_IRQ_ROUTING
2413 case KVM_CAP_IRQ_ROUTING:
2414 return KVM_MAX_IRQ_ROUTES;
2419 return kvm_dev_ioctl_check_extension(arg);
2422 static long kvm_dev_ioctl(struct file *filp,
2423 unsigned int ioctl, unsigned long arg)
2428 case KVM_GET_API_VERSION:
2432 r = KVM_API_VERSION;
2435 r = kvm_dev_ioctl_create_vm(arg);
2437 case KVM_CHECK_EXTENSION:
2438 r = kvm_dev_ioctl_check_extension_generic(arg);
2440 case KVM_GET_VCPU_MMAP_SIZE:
2444 r = PAGE_SIZE; /* struct kvm_run */
2446 r += PAGE_SIZE; /* pio data page */
2448 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2449 r += PAGE_SIZE; /* coalesced mmio ring page */
2452 case KVM_TRACE_ENABLE:
2453 case KVM_TRACE_PAUSE:
2454 case KVM_TRACE_DISABLE:
2458 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2464 static struct file_operations kvm_chardev_ops = {
2465 .unlocked_ioctl = kvm_dev_ioctl,
2466 .compat_ioctl = kvm_dev_ioctl,
2467 .llseek = noop_llseek,
2470 static struct miscdevice kvm_dev = {
2476 static void hardware_enable_nolock(void *junk)
2478 int cpu = raw_smp_processor_id();
2481 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2484 cpumask_set_cpu(cpu, cpus_hardware_enabled);
2486 r = kvm_arch_hardware_enable(NULL);
2489 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2490 atomic_inc(&hardware_enable_failed);
2491 printk(KERN_INFO "kvm: enabling virtualization on "
2492 "CPU%d failed\n", cpu);
2496 static void hardware_enable(void *junk)
2498 raw_spin_lock(&kvm_lock);
2499 hardware_enable_nolock(junk);
2500 raw_spin_unlock(&kvm_lock);
2503 static void hardware_disable_nolock(void *junk)
2505 int cpu = raw_smp_processor_id();
2507 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2509 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2510 kvm_arch_hardware_disable(NULL);
2513 static void hardware_disable(void *junk)
2515 raw_spin_lock(&kvm_lock);
2516 hardware_disable_nolock(junk);
2517 raw_spin_unlock(&kvm_lock);
2520 static void hardware_disable_all_nolock(void)
2522 BUG_ON(!kvm_usage_count);
2525 if (!kvm_usage_count)
2526 on_each_cpu(hardware_disable_nolock, NULL, 1);
2529 static void hardware_disable_all(void)
2531 raw_spin_lock(&kvm_lock);
2532 hardware_disable_all_nolock();
2533 raw_spin_unlock(&kvm_lock);
2536 static int hardware_enable_all(void)
2540 raw_spin_lock(&kvm_lock);
2543 if (kvm_usage_count == 1) {
2544 atomic_set(&hardware_enable_failed, 0);
2545 on_each_cpu(hardware_enable_nolock, NULL, 1);
2547 if (atomic_read(&hardware_enable_failed)) {
2548 hardware_disable_all_nolock();
2553 raw_spin_unlock(&kvm_lock);
2558 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2563 if (!kvm_usage_count)
2566 val &= ~CPU_TASKS_FROZEN;
2569 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2571 hardware_disable(NULL);
2574 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2576 hardware_enable(NULL);
2583 asmlinkage void kvm_spurious_fault(void)
2585 /* Fault while not rebooting. We want the trace. */
2588 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
2590 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2594 * Some (well, at least mine) BIOSes hang on reboot if
2597 * And Intel TXT required VMX off for all cpu when system shutdown.
2599 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2600 kvm_rebooting = true;
2601 on_each_cpu(hardware_disable_nolock, NULL, 1);
2605 static struct notifier_block kvm_reboot_notifier = {
2606 .notifier_call = kvm_reboot,
2610 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2614 for (i = 0; i < bus->dev_count; i++) {
2615 struct kvm_io_device *pos = bus->range[i].dev;
2617 kvm_iodevice_destructor(pos);
2622 int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2624 const struct kvm_io_range *r1 = p1;
2625 const struct kvm_io_range *r2 = p2;
2627 if (r1->addr < r2->addr)
2629 if (r1->addr + r1->len > r2->addr + r2->len)
2634 int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2635 gpa_t addr, int len)
2637 bus->range[bus->dev_count++] = (struct kvm_io_range) {
2643 sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2644 kvm_io_bus_sort_cmp, NULL);
2649 int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2650 gpa_t addr, int len)
2652 struct kvm_io_range *range, key;
2655 key = (struct kvm_io_range) {
2660 range = bsearch(&key, bus->range, bus->dev_count,
2661 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2665 off = range - bus->range;
2667 while (off > 0 && kvm_io_bus_sort_cmp(&key, &bus->range[off-1]) == 0)
2673 /* kvm_io_bus_write - called under kvm->slots_lock */
2674 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2675 int len, const void *val)
2678 struct kvm_io_bus *bus;
2679 struct kvm_io_range range;
2681 range = (struct kvm_io_range) {
2686 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2687 idx = kvm_io_bus_get_first_dev(bus, addr, len);
2691 while (idx < bus->dev_count &&
2692 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2693 if (!kvm_iodevice_write(bus->range[idx].dev, addr, len, val))
2701 /* kvm_io_bus_read - called under kvm->slots_lock */
2702 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2706 struct kvm_io_bus *bus;
2707 struct kvm_io_range range;
2709 range = (struct kvm_io_range) {
2714 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2715 idx = kvm_io_bus_get_first_dev(bus, addr, len);
2719 while (idx < bus->dev_count &&
2720 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2721 if (!kvm_iodevice_read(bus->range[idx].dev, addr, len, val))
2729 /* Caller must hold slots_lock. */
2730 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2731 int len, struct kvm_io_device *dev)
2733 struct kvm_io_bus *new_bus, *bus;
2735 bus = kvm->buses[bus_idx];
2736 if (bus->dev_count > NR_IOBUS_DEVS - 1)
2739 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
2740 sizeof(struct kvm_io_range)), GFP_KERNEL);
2743 memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
2744 sizeof(struct kvm_io_range)));
2745 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2746 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2747 synchronize_srcu_expedited(&kvm->srcu);
2753 /* Caller must hold slots_lock. */
2754 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2755 struct kvm_io_device *dev)
2758 struct kvm_io_bus *new_bus, *bus;
2760 bus = kvm->buses[bus_idx];
2762 for (i = 0; i < bus->dev_count; i++)
2763 if (bus->range[i].dev == dev) {
2771 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
2772 sizeof(struct kvm_io_range)), GFP_KERNEL);
2776 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
2777 new_bus->dev_count--;
2778 memcpy(new_bus->range + i, bus->range + i + 1,
2779 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
2781 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2782 synchronize_srcu_expedited(&kvm->srcu);
2787 static struct notifier_block kvm_cpu_notifier = {
2788 .notifier_call = kvm_cpu_hotplug,
2791 static int vm_stat_get(void *_offset, u64 *val)
2793 unsigned offset = (long)_offset;
2797 raw_spin_lock(&kvm_lock);
2798 list_for_each_entry(kvm, &vm_list, vm_list)
2799 *val += *(u32 *)((void *)kvm + offset);
2800 raw_spin_unlock(&kvm_lock);
2804 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2806 static int vcpu_stat_get(void *_offset, u64 *val)
2808 unsigned offset = (long)_offset;
2810 struct kvm_vcpu *vcpu;
2814 raw_spin_lock(&kvm_lock);
2815 list_for_each_entry(kvm, &vm_list, vm_list)
2816 kvm_for_each_vcpu(i, vcpu, kvm)
2817 *val += *(u32 *)((void *)vcpu + offset);
2819 raw_spin_unlock(&kvm_lock);
2823 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2825 static const struct file_operations *stat_fops[] = {
2826 [KVM_STAT_VCPU] = &vcpu_stat_fops,
2827 [KVM_STAT_VM] = &vm_stat_fops,
2830 static int kvm_init_debug(void)
2833 struct kvm_stats_debugfs_item *p;
2835 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2836 if (kvm_debugfs_dir == NULL)
2839 for (p = debugfs_entries; p->name; ++p) {
2840 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2841 (void *)(long)p->offset,
2842 stat_fops[p->kind]);
2843 if (p->dentry == NULL)
2850 debugfs_remove_recursive(kvm_debugfs_dir);
2855 static void kvm_exit_debug(void)
2857 struct kvm_stats_debugfs_item *p;
2859 for (p = debugfs_entries; p->name; ++p)
2860 debugfs_remove(p->dentry);
2861 debugfs_remove(kvm_debugfs_dir);
2864 static int kvm_suspend(void)
2866 if (kvm_usage_count)
2867 hardware_disable_nolock(NULL);
2871 static void kvm_resume(void)
2873 if (kvm_usage_count) {
2874 WARN_ON(raw_spin_is_locked(&kvm_lock));
2875 hardware_enable_nolock(NULL);
2879 static struct syscore_ops kvm_syscore_ops = {
2880 .suspend = kvm_suspend,
2881 .resume = kvm_resume,
2885 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2887 return container_of(pn, struct kvm_vcpu, preempt_notifier);
2890 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2892 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2894 kvm_arch_vcpu_load(vcpu, cpu);
2897 static void kvm_sched_out(struct preempt_notifier *pn,
2898 struct task_struct *next)
2900 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2902 kvm_arch_vcpu_put(vcpu);
2905 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2906 struct module *module)
2911 r = kvm_arch_init(opaque);
2915 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2920 r = kvm_arch_hardware_setup();
2924 for_each_online_cpu(cpu) {
2925 smp_call_function_single(cpu,
2926 kvm_arch_check_processor_compat,
2932 r = register_cpu_notifier(&kvm_cpu_notifier);
2935 register_reboot_notifier(&kvm_reboot_notifier);
2937 /* A kmem cache lets us meet the alignment requirements of fx_save. */
2939 vcpu_align = __alignof__(struct kvm_vcpu);
2940 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2942 if (!kvm_vcpu_cache) {
2947 r = kvm_async_pf_init();
2951 kvm_chardev_ops.owner = module;
2952 kvm_vm_fops.owner = module;
2953 kvm_vcpu_fops.owner = module;
2955 r = misc_register(&kvm_dev);
2957 printk(KERN_ERR "kvm: misc device register failed\n");
2961 register_syscore_ops(&kvm_syscore_ops);
2963 kvm_preempt_ops.sched_in = kvm_sched_in;
2964 kvm_preempt_ops.sched_out = kvm_sched_out;
2966 r = kvm_init_debug();
2968 printk(KERN_ERR "kvm: create debugfs files failed\n");
2975 unregister_syscore_ops(&kvm_syscore_ops);
2977 kvm_async_pf_deinit();
2979 kmem_cache_destroy(kvm_vcpu_cache);
2981 unregister_reboot_notifier(&kvm_reboot_notifier);
2982 unregister_cpu_notifier(&kvm_cpu_notifier);
2985 kvm_arch_hardware_unsetup();
2987 free_cpumask_var(cpus_hardware_enabled);
2993 EXPORT_SYMBOL_GPL(kvm_init);
2998 misc_deregister(&kvm_dev);
2999 kmem_cache_destroy(kvm_vcpu_cache);
3000 kvm_async_pf_deinit();
3001 unregister_syscore_ops(&kvm_syscore_ops);
3002 unregister_reboot_notifier(&kvm_reboot_notifier);
3003 unregister_cpu_notifier(&kvm_cpu_notifier);
3004 on_each_cpu(hardware_disable_nolock, NULL, 1);
3005 kvm_arch_hardware_unsetup();
3007 free_cpumask_var(cpus_hardware_enabled);
3009 EXPORT_SYMBOL_GPL(kvm_exit);