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 * KVM_SET_USER_MEMORY_REGION ioctl allows the following operations:
723 * - create a new memory slot
724 * - delete an existing memory slot
725 * - modify an existing memory slot
726 * -- move it in the guest physical memory space
727 * -- just change its flags
729 * Since flags can be changed by some of these operations, the following
730 * differentiation is the best we can do for __kvm_set_memory_region():
740 * Allocate some memory and give it an address in the guest physical address
743 * Discontiguous memory is allowed, mostly for framebuffers.
745 * Must be called holding mmap_sem for write.
747 int __kvm_set_memory_region(struct kvm *kvm,
748 struct kvm_userspace_memory_region *mem,
753 unsigned long npages;
754 struct kvm_memory_slot *slot;
755 struct kvm_memory_slot old, new;
756 struct kvm_memslots *slots = NULL, *old_memslots;
757 bool old_iommu_mapped;
758 enum kvm_mr_change change;
760 r = check_memory_region_flags(mem);
765 /* General sanity checks */
766 if (mem->memory_size & (PAGE_SIZE - 1))
768 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
770 /* We can read the guest memory with __xxx_user() later on. */
772 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
773 !access_ok(VERIFY_WRITE,
774 (void __user *)(unsigned long)mem->userspace_addr,
777 if (mem->slot >= KVM_MEM_SLOTS_NUM)
779 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
782 slot = id_to_memslot(kvm->memslots, mem->slot);
783 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
784 npages = mem->memory_size >> PAGE_SHIFT;
787 if (npages > KVM_MEM_MAX_NR_PAGES)
791 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
796 new.base_gfn = base_gfn;
798 new.flags = mem->flags;
800 old_iommu_mapped = old.npages;
805 change = KVM_MR_CREATE;
806 else { /* Modify an existing slot. */
807 if ((mem->userspace_addr != old.userspace_addr) ||
808 (npages != old.npages))
811 if (base_gfn != old.base_gfn)
812 change = KVM_MR_MOVE;
813 else if (new.flags != old.flags)
814 change = KVM_MR_FLAGS_ONLY;
815 else { /* Nothing to change. */
820 } else if (old.npages) {
821 change = KVM_MR_DELETE;
822 } else /* Modify a non-existent slot: disallowed. */
825 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
826 /* Check for overlaps */
828 kvm_for_each_memslot(slot, kvm->memslots) {
829 if ((slot->id >= KVM_USER_MEM_SLOTS) ||
830 (slot->id == mem->slot))
832 if (!((base_gfn + npages <= slot->base_gfn) ||
833 (base_gfn >= slot->base_gfn + slot->npages)))
838 /* Free page dirty bitmap if unneeded */
839 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
840 new.dirty_bitmap = NULL;
843 if (change == KVM_MR_CREATE) {
844 new.user_alloc = user_alloc;
845 new.userspace_addr = mem->userspace_addr;
847 if (kvm_arch_create_memslot(&new, npages))
851 /* Allocate page dirty bitmap if needed */
852 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
853 if (kvm_create_dirty_bitmap(&new) < 0)
857 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
859 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
863 slot = id_to_memslot(slots, mem->slot);
864 slot->flags |= KVM_MEMSLOT_INVALID;
866 old_memslots = install_new_memslots(kvm, slots, NULL);
868 /* slot was deleted or moved, clear iommu mapping */
869 kvm_iommu_unmap_pages(kvm, &old);
870 old_iommu_mapped = false;
871 /* From this point no new shadow pages pointing to a deleted,
872 * or moved, memslot will be created.
874 * validation of sp->gfn happens in:
875 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
876 * - kvm_is_visible_gfn (mmu_check_roots)
878 kvm_arch_flush_shadow_memslot(kvm, slot);
879 slots = old_memslots;
882 r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
888 * We can re-use the old_memslots from above, the only difference
889 * from the currently installed memslots is the invalid flag. This
890 * will get overwritten by update_memslots anyway.
893 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
900 * IOMMU mapping: New slots need to be mapped. Old slots need to be
901 * un-mapped and re-mapped if their base changes or if flags that the
902 * iommu cares about change (read-only). Base change unmapping is
903 * handled above with slot deletion, so we only unmap incompatible
904 * flags here. Anything else the iommu might care about for existing
905 * slots (size changes, userspace addr changes) is disallowed above,
906 * so any other attribute changes getting here can be skipped.
908 if (change != KVM_MR_DELETE) {
909 if (old_iommu_mapped &&
910 ((new.flags ^ old.flags) & KVM_MEM_READONLY)) {
911 kvm_iommu_unmap_pages(kvm, &old);
912 old_iommu_mapped = false;
915 if (!old_iommu_mapped) {
916 r = kvm_iommu_map_pages(kvm, &new);
922 /* actual memory is freed via old in kvm_free_physmem_slot below */
923 if (change == KVM_MR_DELETE) {
924 new.dirty_bitmap = NULL;
925 memset(&new.arch, 0, sizeof(new.arch));
928 old_memslots = install_new_memslots(kvm, slots, &new);
930 kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
932 kvm_free_physmem_slot(&old, &new);
940 kvm_free_physmem_slot(&new, &old);
944 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
946 int kvm_set_memory_region(struct kvm *kvm,
947 struct kvm_userspace_memory_region *mem,
952 mutex_lock(&kvm->slots_lock);
953 r = __kvm_set_memory_region(kvm, mem, user_alloc);
954 mutex_unlock(&kvm->slots_lock);
957 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
959 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
961 kvm_userspace_memory_region *mem,
964 if (mem->slot >= KVM_USER_MEM_SLOTS)
966 return kvm_set_memory_region(kvm, mem, user_alloc);
969 int kvm_get_dirty_log(struct kvm *kvm,
970 struct kvm_dirty_log *log, int *is_dirty)
972 struct kvm_memory_slot *memslot;
975 unsigned long any = 0;
978 if (log->slot >= KVM_USER_MEM_SLOTS)
981 memslot = id_to_memslot(kvm->memslots, log->slot);
983 if (!memslot->dirty_bitmap)
986 n = kvm_dirty_bitmap_bytes(memslot);
988 for (i = 0; !any && i < n/sizeof(long); ++i)
989 any = memslot->dirty_bitmap[i];
992 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1003 bool kvm_largepages_enabled(void)
1005 return largepages_enabled;
1008 void kvm_disable_largepages(void)
1010 largepages_enabled = false;
1012 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1014 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1016 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1018 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1020 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1022 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1024 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1025 memslot->flags & KVM_MEMSLOT_INVALID)
1030 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1032 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1034 struct vm_area_struct *vma;
1035 unsigned long addr, size;
1039 addr = gfn_to_hva(kvm, gfn);
1040 if (kvm_is_error_hva(addr))
1043 down_read(¤t->mm->mmap_sem);
1044 vma = find_vma(current->mm, addr);
1048 size = vma_kernel_pagesize(vma);
1051 up_read(¤t->mm->mmap_sem);
1056 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1058 return slot->flags & KVM_MEM_READONLY;
1061 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1062 gfn_t *nr_pages, bool write)
1064 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1065 return KVM_HVA_ERR_BAD;
1067 if (memslot_is_readonly(slot) && write)
1068 return KVM_HVA_ERR_RO_BAD;
1071 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1073 return __gfn_to_hva_memslot(slot, gfn);
1076 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1079 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1082 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1085 return gfn_to_hva_many(slot, gfn, NULL);
1087 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1089 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1091 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1093 EXPORT_SYMBOL_GPL(gfn_to_hva);
1096 * The hva returned by this function is only allowed to be read.
1097 * It should pair with kvm_read_hva() or kvm_read_hva_atomic().
1099 static unsigned long gfn_to_hva_read(struct kvm *kvm, gfn_t gfn)
1101 return __gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL, false);
1104 static int kvm_read_hva(void *data, void __user *hva, int len)
1106 return __copy_from_user(data, hva, len);
1109 static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1111 return __copy_from_user_inatomic(data, hva, len);
1114 int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1115 unsigned long start, int write, struct page **page)
1117 int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1120 flags |= FOLL_WRITE;
1122 return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1125 static inline int check_user_page_hwpoison(unsigned long addr)
1127 int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1129 rc = __get_user_pages(current, current->mm, addr, 1,
1130 flags, NULL, NULL, NULL);
1131 return rc == -EHWPOISON;
1135 * The atomic path to get the writable pfn which will be stored in @pfn,
1136 * true indicates success, otherwise false is returned.
1138 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1139 bool write_fault, bool *writable, pfn_t *pfn)
1141 struct page *page[1];
1144 if (!(async || atomic))
1148 * Fast pin a writable pfn only if it is a write fault request
1149 * or the caller allows to map a writable pfn for a read fault
1152 if (!(write_fault || writable))
1155 npages = __get_user_pages_fast(addr, 1, 1, page);
1157 *pfn = page_to_pfn(page[0]);
1168 * The slow path to get the pfn of the specified host virtual address,
1169 * 1 indicates success, -errno is returned if error is detected.
1171 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1172 bool *writable, pfn_t *pfn)
1174 struct page *page[1];
1180 *writable = write_fault;
1183 down_read(¤t->mm->mmap_sem);
1184 npages = get_user_page_nowait(current, current->mm,
1185 addr, write_fault, page);
1186 up_read(¤t->mm->mmap_sem);
1188 npages = get_user_pages_fast(addr, 1, write_fault,
1193 /* map read fault as writable if possible */
1194 if (unlikely(!write_fault) && writable) {
1195 struct page *wpage[1];
1197 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1206 *pfn = page_to_pfn(page[0]);
1210 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1212 if (unlikely(!(vma->vm_flags & VM_READ)))
1215 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1222 * Pin guest page in memory and return its pfn.
1223 * @addr: host virtual address which maps memory to the guest
1224 * @atomic: whether this function can sleep
1225 * @async: whether this function need to wait IO complete if the
1226 * host page is not in the memory
1227 * @write_fault: whether we should get a writable host page
1228 * @writable: whether it allows to map a writable host page for !@write_fault
1230 * The function will map a writable host page for these two cases:
1231 * 1): @write_fault = true
1232 * 2): @write_fault = false && @writable, @writable will tell the caller
1233 * whether the mapping is writable.
1235 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1236 bool write_fault, bool *writable)
1238 struct vm_area_struct *vma;
1242 /* we can do it either atomically or asynchronously, not both */
1243 BUG_ON(atomic && async);
1245 if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1249 return KVM_PFN_ERR_FAULT;
1251 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1255 down_read(¤t->mm->mmap_sem);
1256 if (npages == -EHWPOISON ||
1257 (!async && check_user_page_hwpoison(addr))) {
1258 pfn = KVM_PFN_ERR_HWPOISON;
1262 vma = find_vma_intersection(current->mm, addr, addr + 1);
1265 pfn = KVM_PFN_ERR_FAULT;
1266 else if ((vma->vm_flags & VM_PFNMAP)) {
1267 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1269 BUG_ON(!kvm_is_mmio_pfn(pfn));
1271 if (async && vma_is_valid(vma, write_fault))
1273 pfn = KVM_PFN_ERR_FAULT;
1276 up_read(¤t->mm->mmap_sem);
1281 __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1282 bool *async, bool write_fault, bool *writable)
1284 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1286 if (addr == KVM_HVA_ERR_RO_BAD)
1287 return KVM_PFN_ERR_RO_FAULT;
1289 if (kvm_is_error_hva(addr))
1290 return KVM_PFN_NOSLOT;
1292 /* Do not map writable pfn in the readonly memslot. */
1293 if (writable && memslot_is_readonly(slot)) {
1298 return hva_to_pfn(addr, atomic, async, write_fault,
1302 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1303 bool write_fault, bool *writable)
1305 struct kvm_memory_slot *slot;
1310 slot = gfn_to_memslot(kvm, gfn);
1312 return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1316 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1318 return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1320 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1322 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1323 bool write_fault, bool *writable)
1325 return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1327 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1329 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1331 return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1333 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1335 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1338 return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1340 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1342 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1344 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1347 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1349 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1351 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1353 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1359 addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1360 if (kvm_is_error_hva(addr))
1363 if (entry < nr_pages)
1366 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1368 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1370 static struct page *kvm_pfn_to_page(pfn_t pfn)
1372 if (is_error_noslot_pfn(pfn))
1373 return KVM_ERR_PTR_BAD_PAGE;
1375 if (kvm_is_mmio_pfn(pfn)) {
1377 return KVM_ERR_PTR_BAD_PAGE;
1380 return pfn_to_page(pfn);
1383 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1387 pfn = gfn_to_pfn(kvm, gfn);
1389 return kvm_pfn_to_page(pfn);
1392 EXPORT_SYMBOL_GPL(gfn_to_page);
1394 void kvm_release_page_clean(struct page *page)
1396 WARN_ON(is_error_page(page));
1398 kvm_release_pfn_clean(page_to_pfn(page));
1400 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1402 void kvm_release_pfn_clean(pfn_t pfn)
1404 if (!is_error_noslot_pfn(pfn) && !kvm_is_mmio_pfn(pfn))
1405 put_page(pfn_to_page(pfn));
1407 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1409 void kvm_release_page_dirty(struct page *page)
1411 WARN_ON(is_error_page(page));
1413 kvm_release_pfn_dirty(page_to_pfn(page));
1415 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1417 void kvm_release_pfn_dirty(pfn_t pfn)
1419 kvm_set_pfn_dirty(pfn);
1420 kvm_release_pfn_clean(pfn);
1422 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1424 void kvm_set_page_dirty(struct page *page)
1426 kvm_set_pfn_dirty(page_to_pfn(page));
1428 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1430 void kvm_set_pfn_dirty(pfn_t pfn)
1432 if (!kvm_is_mmio_pfn(pfn)) {
1433 struct page *page = pfn_to_page(pfn);
1434 if (!PageReserved(page))
1438 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1440 void kvm_set_pfn_accessed(pfn_t pfn)
1442 if (!kvm_is_mmio_pfn(pfn))
1443 mark_page_accessed(pfn_to_page(pfn));
1445 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1447 void kvm_get_pfn(pfn_t pfn)
1449 if (!kvm_is_mmio_pfn(pfn))
1450 get_page(pfn_to_page(pfn));
1452 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1454 static int next_segment(unsigned long len, int offset)
1456 if (len > PAGE_SIZE - offset)
1457 return PAGE_SIZE - offset;
1462 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1468 addr = gfn_to_hva_read(kvm, gfn);
1469 if (kvm_is_error_hva(addr))
1471 r = kvm_read_hva(data, (void __user *)addr + offset, len);
1476 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1478 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1480 gfn_t gfn = gpa >> PAGE_SHIFT;
1482 int offset = offset_in_page(gpa);
1485 while ((seg = next_segment(len, offset)) != 0) {
1486 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1496 EXPORT_SYMBOL_GPL(kvm_read_guest);
1498 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1503 gfn_t gfn = gpa >> PAGE_SHIFT;
1504 int offset = offset_in_page(gpa);
1506 addr = gfn_to_hva_read(kvm, gfn);
1507 if (kvm_is_error_hva(addr))
1509 pagefault_disable();
1510 r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
1516 EXPORT_SYMBOL(kvm_read_guest_atomic);
1518 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1519 int offset, int len)
1524 addr = gfn_to_hva(kvm, gfn);
1525 if (kvm_is_error_hva(addr))
1527 r = __copy_to_user((void __user *)addr + offset, data, len);
1530 mark_page_dirty(kvm, gfn);
1533 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1535 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1538 gfn_t gfn = gpa >> PAGE_SHIFT;
1540 int offset = offset_in_page(gpa);
1543 while ((seg = next_segment(len, offset)) != 0) {
1544 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1555 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1558 struct kvm_memslots *slots = kvm_memslots(kvm);
1559 int offset = offset_in_page(gpa);
1560 gfn_t gfn = gpa >> PAGE_SHIFT;
1563 ghc->generation = slots->generation;
1564 ghc->memslot = gfn_to_memslot(kvm, gfn);
1565 ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1566 if (!kvm_is_error_hva(ghc->hva))
1573 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1575 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1576 void *data, unsigned long len)
1578 struct kvm_memslots *slots = kvm_memslots(kvm);
1581 if (slots->generation != ghc->generation)
1582 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1584 if (kvm_is_error_hva(ghc->hva))
1587 r = __copy_to_user((void __user *)ghc->hva, data, len);
1590 mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1594 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1596 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1597 void *data, unsigned long len)
1599 struct kvm_memslots *slots = kvm_memslots(kvm);
1602 if (slots->generation != ghc->generation)
1603 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1605 if (kvm_is_error_hva(ghc->hva))
1608 r = __copy_from_user(data, (void __user *)ghc->hva, len);
1614 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1616 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1618 return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1621 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1623 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1625 gfn_t gfn = gpa >> PAGE_SHIFT;
1627 int offset = offset_in_page(gpa);
1630 while ((seg = next_segment(len, offset)) != 0) {
1631 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1640 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1642 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1645 if (memslot && memslot->dirty_bitmap) {
1646 unsigned long rel_gfn = gfn - memslot->base_gfn;
1648 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1652 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1654 struct kvm_memory_slot *memslot;
1656 memslot = gfn_to_memslot(kvm, gfn);
1657 mark_page_dirty_in_slot(kvm, memslot, gfn);
1661 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1663 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1668 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1670 if (kvm_arch_vcpu_runnable(vcpu)) {
1671 kvm_make_request(KVM_REQ_UNHALT, vcpu);
1674 if (kvm_cpu_has_pending_timer(vcpu))
1676 if (signal_pending(current))
1682 finish_wait(&vcpu->wq, &wait);
1687 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1689 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1692 int cpu = vcpu->cpu;
1693 wait_queue_head_t *wqp;
1695 wqp = kvm_arch_vcpu_wq(vcpu);
1696 if (waitqueue_active(wqp)) {
1697 wake_up_interruptible(wqp);
1698 ++vcpu->stat.halt_wakeup;
1702 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1703 if (kvm_arch_vcpu_should_kick(vcpu))
1704 smp_send_reschedule(cpu);
1707 #endif /* !CONFIG_S390 */
1709 void kvm_resched(struct kvm_vcpu *vcpu)
1711 if (!need_resched())
1715 EXPORT_SYMBOL_GPL(kvm_resched);
1717 bool kvm_vcpu_yield_to(struct kvm_vcpu *target)
1720 struct task_struct *task = NULL;
1724 pid = rcu_dereference(target->pid);
1726 task = get_pid_task(target->pid, PIDTYPE_PID);
1730 if (task->flags & PF_VCPU) {
1731 put_task_struct(task);
1734 ret = yield_to(task, 1);
1735 put_task_struct(task);
1739 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1741 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1743 * Helper that checks whether a VCPU is eligible for directed yield.
1744 * Most eligible candidate to yield is decided by following heuristics:
1746 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1747 * (preempted lock holder), indicated by @in_spin_loop.
1748 * Set at the beiginning and cleared at the end of interception/PLE handler.
1750 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1751 * chance last time (mostly it has become eligible now since we have probably
1752 * yielded to lockholder in last iteration. This is done by toggling
1753 * @dy_eligible each time a VCPU checked for eligibility.)
1755 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1756 * to preempted lock-holder could result in wrong VCPU selection and CPU
1757 * burning. Giving priority for a potential lock-holder increases lock
1760 * Since algorithm is based on heuristics, accessing another VCPU data without
1761 * locking does not harm. It may result in trying to yield to same VCPU, fail
1762 * and continue with next VCPU and so on.
1764 bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1768 eligible = !vcpu->spin_loop.in_spin_loop ||
1769 (vcpu->spin_loop.in_spin_loop &&
1770 vcpu->spin_loop.dy_eligible);
1772 if (vcpu->spin_loop.in_spin_loop)
1773 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1779 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1781 struct kvm *kvm = me->kvm;
1782 struct kvm_vcpu *vcpu;
1783 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1789 kvm_vcpu_set_in_spin_loop(me, true);
1791 * We boost the priority of a VCPU that is runnable but not
1792 * currently running, because it got preempted by something
1793 * else and called schedule in __vcpu_run. Hopefully that
1794 * VCPU is holding the lock that we need and will release it.
1795 * We approximate round-robin by starting at the last boosted VCPU.
1797 for (pass = 0; pass < 2 && !yielded && try; pass++) {
1798 kvm_for_each_vcpu(i, vcpu, kvm) {
1799 if (!pass && i <= last_boosted_vcpu) {
1800 i = last_boosted_vcpu;
1802 } else if (pass && i > last_boosted_vcpu)
1806 if (waitqueue_active(&vcpu->wq))
1808 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1811 yielded = kvm_vcpu_yield_to(vcpu);
1813 kvm->last_boosted_vcpu = i;
1815 } else if (yielded < 0) {
1822 kvm_vcpu_set_in_spin_loop(me, false);
1824 /* Ensure vcpu is not eligible during next spinloop */
1825 kvm_vcpu_set_dy_eligible(me, false);
1827 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1829 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1831 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1834 if (vmf->pgoff == 0)
1835 page = virt_to_page(vcpu->run);
1837 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1838 page = virt_to_page(vcpu->arch.pio_data);
1840 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1841 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1842 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1845 return kvm_arch_vcpu_fault(vcpu, vmf);
1851 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1852 .fault = kvm_vcpu_fault,
1855 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1857 vma->vm_ops = &kvm_vcpu_vm_ops;
1861 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1863 struct kvm_vcpu *vcpu = filp->private_data;
1865 kvm_put_kvm(vcpu->kvm);
1869 static struct file_operations kvm_vcpu_fops = {
1870 .release = kvm_vcpu_release,
1871 .unlocked_ioctl = kvm_vcpu_ioctl,
1872 #ifdef CONFIG_COMPAT
1873 .compat_ioctl = kvm_vcpu_compat_ioctl,
1875 .mmap = kvm_vcpu_mmap,
1876 .llseek = noop_llseek,
1880 * Allocates an inode for the vcpu.
1882 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1884 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1888 * Creates some virtual cpus. Good luck creating more than one.
1890 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1893 struct kvm_vcpu *vcpu, *v;
1895 vcpu = kvm_arch_vcpu_create(kvm, id);
1897 return PTR_ERR(vcpu);
1899 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1901 r = kvm_arch_vcpu_setup(vcpu);
1905 mutex_lock(&kvm->lock);
1906 if (!kvm_vcpu_compatible(vcpu)) {
1908 goto unlock_vcpu_destroy;
1910 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1912 goto unlock_vcpu_destroy;
1915 kvm_for_each_vcpu(r, v, kvm)
1916 if (v->vcpu_id == id) {
1918 goto unlock_vcpu_destroy;
1921 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1923 /* Now it's all set up, let userspace reach it */
1925 r = create_vcpu_fd(vcpu);
1928 goto unlock_vcpu_destroy;
1931 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1933 atomic_inc(&kvm->online_vcpus);
1935 mutex_unlock(&kvm->lock);
1936 kvm_arch_vcpu_postcreate(vcpu);
1939 unlock_vcpu_destroy:
1940 mutex_unlock(&kvm->lock);
1942 kvm_arch_vcpu_destroy(vcpu);
1946 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1949 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1950 vcpu->sigset_active = 1;
1951 vcpu->sigset = *sigset;
1953 vcpu->sigset_active = 0;
1957 static long kvm_vcpu_ioctl(struct file *filp,
1958 unsigned int ioctl, unsigned long arg)
1960 struct kvm_vcpu *vcpu = filp->private_data;
1961 void __user *argp = (void __user *)arg;
1963 struct kvm_fpu *fpu = NULL;
1964 struct kvm_sregs *kvm_sregs = NULL;
1966 if (vcpu->kvm->mm != current->mm)
1969 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1971 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1972 * so vcpu_load() would break it.
1974 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1975 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1979 r = vcpu_load(vcpu);
1987 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1988 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1990 case KVM_GET_REGS: {
1991 struct kvm_regs *kvm_regs;
1994 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1997 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2001 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2008 case KVM_SET_REGS: {
2009 struct kvm_regs *kvm_regs;
2012 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2013 if (IS_ERR(kvm_regs)) {
2014 r = PTR_ERR(kvm_regs);
2017 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2021 case KVM_GET_SREGS: {
2022 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2026 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2030 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2035 case KVM_SET_SREGS: {
2036 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2037 if (IS_ERR(kvm_sregs)) {
2038 r = PTR_ERR(kvm_sregs);
2042 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2045 case KVM_GET_MP_STATE: {
2046 struct kvm_mp_state mp_state;
2048 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2052 if (copy_to_user(argp, &mp_state, sizeof mp_state))
2057 case KVM_SET_MP_STATE: {
2058 struct kvm_mp_state mp_state;
2061 if (copy_from_user(&mp_state, argp, sizeof mp_state))
2063 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2066 case KVM_TRANSLATE: {
2067 struct kvm_translation tr;
2070 if (copy_from_user(&tr, argp, sizeof tr))
2072 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2076 if (copy_to_user(argp, &tr, sizeof tr))
2081 case KVM_SET_GUEST_DEBUG: {
2082 struct kvm_guest_debug dbg;
2085 if (copy_from_user(&dbg, argp, sizeof dbg))
2087 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2090 case KVM_SET_SIGNAL_MASK: {
2091 struct kvm_signal_mask __user *sigmask_arg = argp;
2092 struct kvm_signal_mask kvm_sigmask;
2093 sigset_t sigset, *p;
2098 if (copy_from_user(&kvm_sigmask, argp,
2099 sizeof kvm_sigmask))
2102 if (kvm_sigmask.len != sizeof sigset)
2105 if (copy_from_user(&sigset, sigmask_arg->sigset,
2110 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2114 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2118 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2122 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2128 fpu = memdup_user(argp, sizeof(*fpu));
2134 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2138 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2147 #ifdef CONFIG_COMPAT
2148 static long kvm_vcpu_compat_ioctl(struct file *filp,
2149 unsigned int ioctl, unsigned long arg)
2151 struct kvm_vcpu *vcpu = filp->private_data;
2152 void __user *argp = compat_ptr(arg);
2155 if (vcpu->kvm->mm != current->mm)
2159 case KVM_SET_SIGNAL_MASK: {
2160 struct kvm_signal_mask __user *sigmask_arg = argp;
2161 struct kvm_signal_mask kvm_sigmask;
2162 compat_sigset_t csigset;
2167 if (copy_from_user(&kvm_sigmask, argp,
2168 sizeof kvm_sigmask))
2171 if (kvm_sigmask.len != sizeof csigset)
2174 if (copy_from_user(&csigset, sigmask_arg->sigset,
2177 sigset_from_compat(&sigset, &csigset);
2178 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2180 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2184 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2192 static long kvm_vm_ioctl(struct file *filp,
2193 unsigned int ioctl, unsigned long arg)
2195 struct kvm *kvm = filp->private_data;
2196 void __user *argp = (void __user *)arg;
2199 if (kvm->mm != current->mm)
2202 case KVM_CREATE_VCPU:
2203 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2205 case KVM_SET_USER_MEMORY_REGION: {
2206 struct kvm_userspace_memory_region kvm_userspace_mem;
2209 if (copy_from_user(&kvm_userspace_mem, argp,
2210 sizeof kvm_userspace_mem))
2213 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, true);
2216 case KVM_GET_DIRTY_LOG: {
2217 struct kvm_dirty_log log;
2220 if (copy_from_user(&log, argp, sizeof log))
2222 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2225 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2226 case KVM_REGISTER_COALESCED_MMIO: {
2227 struct kvm_coalesced_mmio_zone zone;
2229 if (copy_from_user(&zone, argp, sizeof zone))
2231 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2234 case KVM_UNREGISTER_COALESCED_MMIO: {
2235 struct kvm_coalesced_mmio_zone zone;
2237 if (copy_from_user(&zone, argp, sizeof zone))
2239 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2244 struct kvm_irqfd data;
2247 if (copy_from_user(&data, argp, sizeof data))
2249 r = kvm_irqfd(kvm, &data);
2252 case KVM_IOEVENTFD: {
2253 struct kvm_ioeventfd data;
2256 if (copy_from_user(&data, argp, sizeof data))
2258 r = kvm_ioeventfd(kvm, &data);
2261 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2262 case KVM_SET_BOOT_CPU_ID:
2264 mutex_lock(&kvm->lock);
2265 if (atomic_read(&kvm->online_vcpus) != 0)
2268 kvm->bsp_vcpu_id = arg;
2269 mutex_unlock(&kvm->lock);
2272 #ifdef CONFIG_HAVE_KVM_MSI
2273 case KVM_SIGNAL_MSI: {
2277 if (copy_from_user(&msi, argp, sizeof msi))
2279 r = kvm_send_userspace_msi(kvm, &msi);
2283 #ifdef __KVM_HAVE_IRQ_LINE
2284 case KVM_IRQ_LINE_STATUS:
2285 case KVM_IRQ_LINE: {
2286 struct kvm_irq_level irq_event;
2289 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2292 r = kvm_vm_ioctl_irq_line(kvm, &irq_event);
2297 if (ioctl == KVM_IRQ_LINE_STATUS) {
2298 if (copy_to_user(argp, &irq_event, sizeof irq_event))
2307 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2309 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2315 #ifdef CONFIG_COMPAT
2316 struct compat_kvm_dirty_log {
2320 compat_uptr_t dirty_bitmap; /* one bit per page */
2325 static long kvm_vm_compat_ioctl(struct file *filp,
2326 unsigned int ioctl, unsigned long arg)
2328 struct kvm *kvm = filp->private_data;
2331 if (kvm->mm != current->mm)
2334 case KVM_GET_DIRTY_LOG: {
2335 struct compat_kvm_dirty_log compat_log;
2336 struct kvm_dirty_log log;
2339 if (copy_from_user(&compat_log, (void __user *)arg,
2340 sizeof(compat_log)))
2342 log.slot = compat_log.slot;
2343 log.padding1 = compat_log.padding1;
2344 log.padding2 = compat_log.padding2;
2345 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2347 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2351 r = kvm_vm_ioctl(filp, ioctl, arg);
2359 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2361 struct page *page[1];
2364 gfn_t gfn = vmf->pgoff;
2365 struct kvm *kvm = vma->vm_file->private_data;
2367 addr = gfn_to_hva(kvm, gfn);
2368 if (kvm_is_error_hva(addr))
2369 return VM_FAULT_SIGBUS;
2371 npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2373 if (unlikely(npages != 1))
2374 return VM_FAULT_SIGBUS;
2376 vmf->page = page[0];
2380 static const struct vm_operations_struct kvm_vm_vm_ops = {
2381 .fault = kvm_vm_fault,
2384 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2386 vma->vm_ops = &kvm_vm_vm_ops;
2390 static struct file_operations kvm_vm_fops = {
2391 .release = kvm_vm_release,
2392 .unlocked_ioctl = kvm_vm_ioctl,
2393 #ifdef CONFIG_COMPAT
2394 .compat_ioctl = kvm_vm_compat_ioctl,
2396 .mmap = kvm_vm_mmap,
2397 .llseek = noop_llseek,
2400 static int kvm_dev_ioctl_create_vm(unsigned long type)
2405 kvm = kvm_create_vm(type);
2407 return PTR_ERR(kvm);
2408 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2409 r = kvm_coalesced_mmio_init(kvm);
2415 r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2422 static long kvm_dev_ioctl_check_extension_generic(long arg)
2425 case KVM_CAP_USER_MEMORY:
2426 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2427 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2428 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2429 case KVM_CAP_SET_BOOT_CPU_ID:
2431 case KVM_CAP_INTERNAL_ERROR_DATA:
2432 #ifdef CONFIG_HAVE_KVM_MSI
2433 case KVM_CAP_SIGNAL_MSI:
2436 #ifdef KVM_CAP_IRQ_ROUTING
2437 case KVM_CAP_IRQ_ROUTING:
2438 return KVM_MAX_IRQ_ROUTES;
2443 return kvm_dev_ioctl_check_extension(arg);
2446 static long kvm_dev_ioctl(struct file *filp,
2447 unsigned int ioctl, unsigned long arg)
2452 case KVM_GET_API_VERSION:
2456 r = KVM_API_VERSION;
2459 r = kvm_dev_ioctl_create_vm(arg);
2461 case KVM_CHECK_EXTENSION:
2462 r = kvm_dev_ioctl_check_extension_generic(arg);
2464 case KVM_GET_VCPU_MMAP_SIZE:
2468 r = PAGE_SIZE; /* struct kvm_run */
2470 r += PAGE_SIZE; /* pio data page */
2472 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2473 r += PAGE_SIZE; /* coalesced mmio ring page */
2476 case KVM_TRACE_ENABLE:
2477 case KVM_TRACE_PAUSE:
2478 case KVM_TRACE_DISABLE:
2482 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2488 static struct file_operations kvm_chardev_ops = {
2489 .unlocked_ioctl = kvm_dev_ioctl,
2490 .compat_ioctl = kvm_dev_ioctl,
2491 .llseek = noop_llseek,
2494 static struct miscdevice kvm_dev = {
2500 static void hardware_enable_nolock(void *junk)
2502 int cpu = raw_smp_processor_id();
2505 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2508 cpumask_set_cpu(cpu, cpus_hardware_enabled);
2510 r = kvm_arch_hardware_enable(NULL);
2513 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2514 atomic_inc(&hardware_enable_failed);
2515 printk(KERN_INFO "kvm: enabling virtualization on "
2516 "CPU%d failed\n", cpu);
2520 static void hardware_enable(void *junk)
2522 raw_spin_lock(&kvm_lock);
2523 hardware_enable_nolock(junk);
2524 raw_spin_unlock(&kvm_lock);
2527 static void hardware_disable_nolock(void *junk)
2529 int cpu = raw_smp_processor_id();
2531 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2533 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2534 kvm_arch_hardware_disable(NULL);
2537 static void hardware_disable(void *junk)
2539 raw_spin_lock(&kvm_lock);
2540 hardware_disable_nolock(junk);
2541 raw_spin_unlock(&kvm_lock);
2544 static void hardware_disable_all_nolock(void)
2546 BUG_ON(!kvm_usage_count);
2549 if (!kvm_usage_count)
2550 on_each_cpu(hardware_disable_nolock, NULL, 1);
2553 static void hardware_disable_all(void)
2555 raw_spin_lock(&kvm_lock);
2556 hardware_disable_all_nolock();
2557 raw_spin_unlock(&kvm_lock);
2560 static int hardware_enable_all(void)
2564 raw_spin_lock(&kvm_lock);
2567 if (kvm_usage_count == 1) {
2568 atomic_set(&hardware_enable_failed, 0);
2569 on_each_cpu(hardware_enable_nolock, NULL, 1);
2571 if (atomic_read(&hardware_enable_failed)) {
2572 hardware_disable_all_nolock();
2577 raw_spin_unlock(&kvm_lock);
2582 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2587 if (!kvm_usage_count)
2590 val &= ~CPU_TASKS_FROZEN;
2593 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2595 hardware_disable(NULL);
2598 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2600 hardware_enable(NULL);
2607 asmlinkage void kvm_spurious_fault(void)
2609 /* Fault while not rebooting. We want the trace. */
2612 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
2614 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2618 * Some (well, at least mine) BIOSes hang on reboot if
2621 * And Intel TXT required VMX off for all cpu when system shutdown.
2623 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2624 kvm_rebooting = true;
2625 on_each_cpu(hardware_disable_nolock, NULL, 1);
2629 static struct notifier_block kvm_reboot_notifier = {
2630 .notifier_call = kvm_reboot,
2634 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2638 for (i = 0; i < bus->dev_count; i++) {
2639 struct kvm_io_device *pos = bus->range[i].dev;
2641 kvm_iodevice_destructor(pos);
2646 int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2648 const struct kvm_io_range *r1 = p1;
2649 const struct kvm_io_range *r2 = p2;
2651 if (r1->addr < r2->addr)
2653 if (r1->addr + r1->len > r2->addr + r2->len)
2658 int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2659 gpa_t addr, int len)
2661 bus->range[bus->dev_count++] = (struct kvm_io_range) {
2667 sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2668 kvm_io_bus_sort_cmp, NULL);
2673 int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2674 gpa_t addr, int len)
2676 struct kvm_io_range *range, key;
2679 key = (struct kvm_io_range) {
2684 range = bsearch(&key, bus->range, bus->dev_count,
2685 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2689 off = range - bus->range;
2691 while (off > 0 && kvm_io_bus_sort_cmp(&key, &bus->range[off-1]) == 0)
2697 /* kvm_io_bus_write - called under kvm->slots_lock */
2698 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2699 int len, const void *val)
2702 struct kvm_io_bus *bus;
2703 struct kvm_io_range range;
2705 range = (struct kvm_io_range) {
2710 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2711 idx = kvm_io_bus_get_first_dev(bus, addr, len);
2715 while (idx < bus->dev_count &&
2716 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2717 if (!kvm_iodevice_write(bus->range[idx].dev, addr, len, val))
2725 /* kvm_io_bus_read - called under kvm->slots_lock */
2726 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2730 struct kvm_io_bus *bus;
2731 struct kvm_io_range range;
2733 range = (struct kvm_io_range) {
2738 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2739 idx = kvm_io_bus_get_first_dev(bus, addr, len);
2743 while (idx < bus->dev_count &&
2744 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2745 if (!kvm_iodevice_read(bus->range[idx].dev, addr, len, val))
2753 /* Caller must hold slots_lock. */
2754 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2755 int len, struct kvm_io_device *dev)
2757 struct kvm_io_bus *new_bus, *bus;
2759 bus = kvm->buses[bus_idx];
2760 if (bus->dev_count > NR_IOBUS_DEVS - 1)
2763 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
2764 sizeof(struct kvm_io_range)), GFP_KERNEL);
2767 memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
2768 sizeof(struct kvm_io_range)));
2769 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2770 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2771 synchronize_srcu_expedited(&kvm->srcu);
2777 /* Caller must hold slots_lock. */
2778 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2779 struct kvm_io_device *dev)
2782 struct kvm_io_bus *new_bus, *bus;
2784 bus = kvm->buses[bus_idx];
2786 for (i = 0; i < bus->dev_count; i++)
2787 if (bus->range[i].dev == dev) {
2795 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
2796 sizeof(struct kvm_io_range)), GFP_KERNEL);
2800 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
2801 new_bus->dev_count--;
2802 memcpy(new_bus->range + i, bus->range + i + 1,
2803 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
2805 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2806 synchronize_srcu_expedited(&kvm->srcu);
2811 static struct notifier_block kvm_cpu_notifier = {
2812 .notifier_call = kvm_cpu_hotplug,
2815 static int vm_stat_get(void *_offset, u64 *val)
2817 unsigned offset = (long)_offset;
2821 raw_spin_lock(&kvm_lock);
2822 list_for_each_entry(kvm, &vm_list, vm_list)
2823 *val += *(u32 *)((void *)kvm + offset);
2824 raw_spin_unlock(&kvm_lock);
2828 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2830 static int vcpu_stat_get(void *_offset, u64 *val)
2832 unsigned offset = (long)_offset;
2834 struct kvm_vcpu *vcpu;
2838 raw_spin_lock(&kvm_lock);
2839 list_for_each_entry(kvm, &vm_list, vm_list)
2840 kvm_for_each_vcpu(i, vcpu, kvm)
2841 *val += *(u32 *)((void *)vcpu + offset);
2843 raw_spin_unlock(&kvm_lock);
2847 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2849 static const struct file_operations *stat_fops[] = {
2850 [KVM_STAT_VCPU] = &vcpu_stat_fops,
2851 [KVM_STAT_VM] = &vm_stat_fops,
2854 static int kvm_init_debug(void)
2857 struct kvm_stats_debugfs_item *p;
2859 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2860 if (kvm_debugfs_dir == NULL)
2863 for (p = debugfs_entries; p->name; ++p) {
2864 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2865 (void *)(long)p->offset,
2866 stat_fops[p->kind]);
2867 if (p->dentry == NULL)
2874 debugfs_remove_recursive(kvm_debugfs_dir);
2879 static void kvm_exit_debug(void)
2881 struct kvm_stats_debugfs_item *p;
2883 for (p = debugfs_entries; p->name; ++p)
2884 debugfs_remove(p->dentry);
2885 debugfs_remove(kvm_debugfs_dir);
2888 static int kvm_suspend(void)
2890 if (kvm_usage_count)
2891 hardware_disable_nolock(NULL);
2895 static void kvm_resume(void)
2897 if (kvm_usage_count) {
2898 WARN_ON(raw_spin_is_locked(&kvm_lock));
2899 hardware_enable_nolock(NULL);
2903 static struct syscore_ops kvm_syscore_ops = {
2904 .suspend = kvm_suspend,
2905 .resume = kvm_resume,
2909 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2911 return container_of(pn, struct kvm_vcpu, preempt_notifier);
2914 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2916 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2918 kvm_arch_vcpu_load(vcpu, cpu);
2921 static void kvm_sched_out(struct preempt_notifier *pn,
2922 struct task_struct *next)
2924 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2926 kvm_arch_vcpu_put(vcpu);
2929 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2930 struct module *module)
2935 r = kvm_arch_init(opaque);
2939 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2944 r = kvm_arch_hardware_setup();
2948 for_each_online_cpu(cpu) {
2949 smp_call_function_single(cpu,
2950 kvm_arch_check_processor_compat,
2956 r = register_cpu_notifier(&kvm_cpu_notifier);
2959 register_reboot_notifier(&kvm_reboot_notifier);
2961 /* A kmem cache lets us meet the alignment requirements of fx_save. */
2963 vcpu_align = __alignof__(struct kvm_vcpu);
2964 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2966 if (!kvm_vcpu_cache) {
2971 r = kvm_async_pf_init();
2975 kvm_chardev_ops.owner = module;
2976 kvm_vm_fops.owner = module;
2977 kvm_vcpu_fops.owner = module;
2979 r = misc_register(&kvm_dev);
2981 printk(KERN_ERR "kvm: misc device register failed\n");
2985 register_syscore_ops(&kvm_syscore_ops);
2987 kvm_preempt_ops.sched_in = kvm_sched_in;
2988 kvm_preempt_ops.sched_out = kvm_sched_out;
2990 r = kvm_init_debug();
2992 printk(KERN_ERR "kvm: create debugfs files failed\n");
2999 unregister_syscore_ops(&kvm_syscore_ops);
3001 kvm_async_pf_deinit();
3003 kmem_cache_destroy(kvm_vcpu_cache);
3005 unregister_reboot_notifier(&kvm_reboot_notifier);
3006 unregister_cpu_notifier(&kvm_cpu_notifier);
3009 kvm_arch_hardware_unsetup();
3011 free_cpumask_var(cpus_hardware_enabled);
3017 EXPORT_SYMBOL_GPL(kvm_init);
3022 misc_deregister(&kvm_dev);
3023 kmem_cache_destroy(kvm_vcpu_cache);
3024 kvm_async_pf_deinit();
3025 unregister_syscore_ops(&kvm_syscore_ops);
3026 unregister_reboot_notifier(&kvm_reboot_notifier);
3027 unregister_cpu_notifier(&kvm_cpu_notifier);
3028 on_each_cpu(hardware_disable_nolock, NULL, 1);
3029 kvm_arch_hardware_unsetup();
3031 free_cpumask_var(cpus_hardware_enabled);
3033 EXPORT_SYMBOL_GPL(kvm_exit);