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>
51 #include <asm/processor.h>
53 #include <asm/uaccess.h>
54 #include <asm/pgtable.h>
56 #include "coalesced_mmio.h"
59 #define CREATE_TRACE_POINTS
60 #include <trace/events/kvm.h>
62 MODULE_AUTHOR("Qumranet");
63 MODULE_LICENSE("GPL");
68 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
71 DEFINE_RAW_SPINLOCK(kvm_lock);
74 static cpumask_var_t cpus_hardware_enabled;
75 static int kvm_usage_count = 0;
76 static atomic_t hardware_enable_failed;
78 struct kmem_cache *kvm_vcpu_cache;
79 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
81 static __read_mostly struct preempt_ops kvm_preempt_ops;
83 struct dentry *kvm_debugfs_dir;
85 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
87 static int hardware_enable_all(void);
88 static void hardware_disable_all(void);
90 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
93 EXPORT_SYMBOL_GPL(kvm_rebooting);
95 static bool largepages_enabled = true;
97 static struct page *hwpoison_page;
98 static pfn_t hwpoison_pfn;
100 static struct page *fault_page;
101 static pfn_t fault_pfn;
103 inline int 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 void vcpu_load(struct kvm_vcpu *vcpu)
138 mutex_lock(&vcpu->mutex);
139 if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
140 /* The thread running this VCPU changed. */
141 struct pid *oldpid = vcpu->pid;
142 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
143 rcu_assign_pointer(vcpu->pid, newpid);
148 preempt_notifier_register(&vcpu->preempt_notifier);
149 kvm_arch_vcpu_load(vcpu, cpu);
153 void vcpu_put(struct kvm_vcpu *vcpu)
156 kvm_arch_vcpu_put(vcpu);
157 preempt_notifier_unregister(&vcpu->preempt_notifier);
159 mutex_unlock(&vcpu->mutex);
162 static void ack_flush(void *_completed)
166 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
171 struct kvm_vcpu *vcpu;
173 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
176 kvm_for_each_vcpu(i, vcpu, kvm) {
177 kvm_make_request(req, vcpu);
180 /* Set ->requests bit before we read ->mode */
183 if (cpus != NULL && cpu != -1 && cpu != me &&
184 kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
185 cpumask_set_cpu(cpu, cpus);
187 if (unlikely(cpus == NULL))
188 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
189 else if (!cpumask_empty(cpus))
190 smp_call_function_many(cpus, ack_flush, NULL, 1);
194 free_cpumask_var(cpus);
198 void kvm_flush_remote_tlbs(struct kvm *kvm)
200 int dirty_count = kvm->tlbs_dirty;
203 if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
204 ++kvm->stat.remote_tlb_flush;
205 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
208 void kvm_reload_remote_mmus(struct kvm *kvm)
210 make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
213 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
218 mutex_init(&vcpu->mutex);
223 init_waitqueue_head(&vcpu->wq);
224 kvm_async_pf_vcpu_init(vcpu);
226 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
231 vcpu->run = page_address(page);
233 r = kvm_arch_vcpu_init(vcpu);
239 free_page((unsigned long)vcpu->run);
243 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
245 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
248 kvm_arch_vcpu_uninit(vcpu);
249 free_page((unsigned long)vcpu->run);
251 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
253 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
254 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
256 return container_of(mn, struct kvm, mmu_notifier);
259 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
260 struct mm_struct *mm,
261 unsigned long address)
263 struct kvm *kvm = mmu_notifier_to_kvm(mn);
264 int need_tlb_flush, idx;
267 * When ->invalidate_page runs, the linux pte has been zapped
268 * already but the page is still allocated until
269 * ->invalidate_page returns. So if we increase the sequence
270 * here the kvm page fault will notice if the spte can't be
271 * established because the page is going to be freed. If
272 * instead the kvm page fault establishes the spte before
273 * ->invalidate_page runs, kvm_unmap_hva will release it
276 * The sequence increase only need to be seen at spin_unlock
277 * time, and not at spin_lock time.
279 * Increasing the sequence after the spin_unlock would be
280 * unsafe because the kvm page fault could then establish the
281 * pte after kvm_unmap_hva returned, without noticing the page
282 * is going to be freed.
284 idx = srcu_read_lock(&kvm->srcu);
285 spin_lock(&kvm->mmu_lock);
286 kvm->mmu_notifier_seq++;
287 need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
288 spin_unlock(&kvm->mmu_lock);
289 srcu_read_unlock(&kvm->srcu, idx);
291 /* we've to flush the tlb before the pages can be freed */
293 kvm_flush_remote_tlbs(kvm);
297 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
298 struct mm_struct *mm,
299 unsigned long address,
302 struct kvm *kvm = mmu_notifier_to_kvm(mn);
305 idx = srcu_read_lock(&kvm->srcu);
306 spin_lock(&kvm->mmu_lock);
307 kvm->mmu_notifier_seq++;
308 kvm_set_spte_hva(kvm, address, pte);
309 spin_unlock(&kvm->mmu_lock);
310 srcu_read_unlock(&kvm->srcu, idx);
313 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
314 struct mm_struct *mm,
318 struct kvm *kvm = mmu_notifier_to_kvm(mn);
319 int need_tlb_flush = 0, idx;
321 idx = srcu_read_lock(&kvm->srcu);
322 spin_lock(&kvm->mmu_lock);
324 * The count increase must become visible at unlock time as no
325 * spte can be established without taking the mmu_lock and
326 * count is also read inside the mmu_lock critical section.
328 kvm->mmu_notifier_count++;
329 for (; start < end; start += PAGE_SIZE)
330 need_tlb_flush |= kvm_unmap_hva(kvm, start);
331 need_tlb_flush |= kvm->tlbs_dirty;
332 spin_unlock(&kvm->mmu_lock);
333 srcu_read_unlock(&kvm->srcu, idx);
335 /* we've to flush the tlb before the pages can be freed */
337 kvm_flush_remote_tlbs(kvm);
340 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
341 struct mm_struct *mm,
345 struct kvm *kvm = mmu_notifier_to_kvm(mn);
347 spin_lock(&kvm->mmu_lock);
349 * This sequence increase will notify the kvm page fault that
350 * the page that is going to be mapped in the spte could have
353 kvm->mmu_notifier_seq++;
355 * The above sequence increase must be visible before the
356 * below count decrease but both values are read by the kvm
357 * page fault under mmu_lock spinlock so we don't need to add
358 * a smb_wmb() here in between the two.
360 kvm->mmu_notifier_count--;
361 spin_unlock(&kvm->mmu_lock);
363 BUG_ON(kvm->mmu_notifier_count < 0);
366 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
367 struct mm_struct *mm,
368 unsigned long address)
370 struct kvm *kvm = mmu_notifier_to_kvm(mn);
373 idx = srcu_read_lock(&kvm->srcu);
374 spin_lock(&kvm->mmu_lock);
375 young = kvm_age_hva(kvm, address);
376 spin_unlock(&kvm->mmu_lock);
377 srcu_read_unlock(&kvm->srcu, idx);
380 kvm_flush_remote_tlbs(kvm);
385 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
386 struct mm_struct *mm,
387 unsigned long address)
389 struct kvm *kvm = mmu_notifier_to_kvm(mn);
392 idx = srcu_read_lock(&kvm->srcu);
393 spin_lock(&kvm->mmu_lock);
394 young = kvm_test_age_hva(kvm, address);
395 spin_unlock(&kvm->mmu_lock);
396 srcu_read_unlock(&kvm->srcu, idx);
401 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
402 struct mm_struct *mm)
404 struct kvm *kvm = mmu_notifier_to_kvm(mn);
407 idx = srcu_read_lock(&kvm->srcu);
408 kvm_arch_flush_shadow(kvm);
409 srcu_read_unlock(&kvm->srcu, idx);
412 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
413 .invalidate_page = kvm_mmu_notifier_invalidate_page,
414 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
415 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
416 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
417 .test_young = kvm_mmu_notifier_test_young,
418 .change_pte = kvm_mmu_notifier_change_pte,
419 .release = kvm_mmu_notifier_release,
422 static int kvm_init_mmu_notifier(struct kvm *kvm)
424 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
425 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
428 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
430 static int kvm_init_mmu_notifier(struct kvm *kvm)
435 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
437 static struct kvm *kvm_create_vm(void)
440 struct kvm *kvm = kvm_arch_alloc_vm();
443 return ERR_PTR(-ENOMEM);
445 r = kvm_arch_init_vm(kvm);
447 goto out_err_nodisable;
449 r = hardware_enable_all();
451 goto out_err_nodisable;
453 #ifdef CONFIG_HAVE_KVM_IRQCHIP
454 INIT_HLIST_HEAD(&kvm->mask_notifier_list);
455 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
459 kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
462 if (init_srcu_struct(&kvm->srcu))
464 for (i = 0; i < KVM_NR_BUSES; i++) {
465 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
471 r = kvm_init_mmu_notifier(kvm);
475 kvm->mm = current->mm;
476 atomic_inc(&kvm->mm->mm_count);
477 spin_lock_init(&kvm->mmu_lock);
478 kvm_eventfd_init(kvm);
479 mutex_init(&kvm->lock);
480 mutex_init(&kvm->irq_lock);
481 mutex_init(&kvm->slots_lock);
482 atomic_set(&kvm->users_count, 1);
483 raw_spin_lock(&kvm_lock);
484 list_add(&kvm->vm_list, &vm_list);
485 raw_spin_unlock(&kvm_lock);
490 cleanup_srcu_struct(&kvm->srcu);
492 hardware_disable_all();
494 for (i = 0; i < KVM_NR_BUSES; i++)
495 kfree(kvm->buses[i]);
496 kfree(kvm->memslots);
497 kvm_arch_free_vm(kvm);
501 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
503 if (!memslot->dirty_bitmap)
506 if (2 * kvm_dirty_bitmap_bytes(memslot) > PAGE_SIZE)
507 vfree(memslot->dirty_bitmap_head);
509 kfree(memslot->dirty_bitmap_head);
511 memslot->dirty_bitmap = NULL;
512 memslot->dirty_bitmap_head = NULL;
516 * Free any memory in @free but not in @dont.
518 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
519 struct kvm_memory_slot *dont)
523 if (!dont || free->rmap != dont->rmap)
526 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
527 kvm_destroy_dirty_bitmap(free);
530 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
531 if (!dont || free->lpage_info[i] != dont->lpage_info[i]) {
532 vfree(free->lpage_info[i]);
533 free->lpage_info[i] = NULL;
541 void kvm_free_physmem(struct kvm *kvm)
544 struct kvm_memslots *slots = kvm->memslots;
546 for (i = 0; i < slots->nmemslots; ++i)
547 kvm_free_physmem_slot(&slots->memslots[i], NULL);
549 kfree(kvm->memslots);
552 static void kvm_destroy_vm(struct kvm *kvm)
555 struct mm_struct *mm = kvm->mm;
557 kvm_arch_sync_events(kvm);
558 raw_spin_lock(&kvm_lock);
559 list_del(&kvm->vm_list);
560 raw_spin_unlock(&kvm_lock);
561 kvm_free_irq_routing(kvm);
562 for (i = 0; i < KVM_NR_BUSES; i++)
563 kvm_io_bus_destroy(kvm->buses[i]);
564 kvm_coalesced_mmio_free(kvm);
565 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
566 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
568 kvm_arch_flush_shadow(kvm);
570 kvm_arch_destroy_vm(kvm);
571 kvm_free_physmem(kvm);
572 cleanup_srcu_struct(&kvm->srcu);
573 kvm_arch_free_vm(kvm);
574 hardware_disable_all();
578 void kvm_get_kvm(struct kvm *kvm)
580 atomic_inc(&kvm->users_count);
582 EXPORT_SYMBOL_GPL(kvm_get_kvm);
584 void kvm_put_kvm(struct kvm *kvm)
586 if (atomic_dec_and_test(&kvm->users_count))
589 EXPORT_SYMBOL_GPL(kvm_put_kvm);
592 static int kvm_vm_release(struct inode *inode, struct file *filp)
594 struct kvm *kvm = filp->private_data;
596 kvm_irqfd_release(kvm);
604 * Allocation size is twice as large as the actual dirty bitmap size.
605 * This makes it possible to do double buffering: see x86's
606 * kvm_vm_ioctl_get_dirty_log().
608 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
610 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
612 if (dirty_bytes > PAGE_SIZE)
613 memslot->dirty_bitmap = vzalloc(dirty_bytes);
615 memslot->dirty_bitmap = kzalloc(dirty_bytes, GFP_KERNEL);
617 if (!memslot->dirty_bitmap)
620 memslot->dirty_bitmap_head = memslot->dirty_bitmap;
623 #endif /* !CONFIG_S390 */
626 * Allocate some memory and give it an address in the guest physical address
629 * Discontiguous memory is allowed, mostly for framebuffers.
631 * Must be called holding mmap_sem for write.
633 int __kvm_set_memory_region(struct kvm *kvm,
634 struct kvm_userspace_memory_region *mem,
639 unsigned long npages;
641 struct kvm_memory_slot *memslot;
642 struct kvm_memory_slot old, new;
643 struct kvm_memslots *slots, *old_memslots;
646 /* General sanity checks */
647 if (mem->memory_size & (PAGE_SIZE - 1))
649 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
651 if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
653 if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
655 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
658 memslot = &kvm->memslots->memslots[mem->slot];
659 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
660 npages = mem->memory_size >> PAGE_SHIFT;
663 if (npages > KVM_MEM_MAX_NR_PAGES)
667 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
669 new = old = *memslot;
672 new.base_gfn = base_gfn;
674 new.flags = mem->flags;
676 /* Disallow changing a memory slot's size. */
678 if (npages && old.npages && npages != old.npages)
681 /* Check for overlaps */
683 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
684 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
686 if (s == memslot || !s->npages)
688 if (!((base_gfn + npages <= s->base_gfn) ||
689 (base_gfn >= s->base_gfn + s->npages)))
693 /* Free page dirty bitmap if unneeded */
694 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
695 new.dirty_bitmap = NULL;
699 /* Allocate if a slot is being created */
701 if (npages && !new.rmap) {
702 new.rmap = vzalloc(npages * sizeof(*new.rmap));
707 new.user_alloc = user_alloc;
708 new.userspace_addr = mem->userspace_addr;
713 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
719 /* Avoid unused variable warning if no large pages */
722 if (new.lpage_info[i])
725 lpages = 1 + ((base_gfn + npages - 1)
726 >> KVM_HPAGE_GFN_SHIFT(level));
727 lpages -= base_gfn >> KVM_HPAGE_GFN_SHIFT(level);
729 new.lpage_info[i] = vzalloc(lpages * sizeof(*new.lpage_info[i]));
731 if (!new.lpage_info[i])
734 if (base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
735 new.lpage_info[i][0].write_count = 1;
736 if ((base_gfn+npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
737 new.lpage_info[i][lpages - 1].write_count = 1;
738 ugfn = new.userspace_addr >> PAGE_SHIFT;
740 * If the gfn and userspace address are not aligned wrt each
741 * other, or if explicitly asked to, disable large page
742 * support for this slot
744 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
746 for (j = 0; j < lpages; ++j)
747 new.lpage_info[i][j].write_count = 1;
752 /* Allocate page dirty bitmap if needed */
753 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
754 if (kvm_create_dirty_bitmap(&new) < 0)
756 /* destroy any largepage mappings for dirty tracking */
758 #else /* not defined CONFIG_S390 */
759 new.user_alloc = user_alloc;
761 new.userspace_addr = mem->userspace_addr;
762 #endif /* not defined CONFIG_S390 */
766 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
769 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
770 if (mem->slot >= slots->nmemslots)
771 slots->nmemslots = mem->slot + 1;
773 slots->memslots[mem->slot].flags |= KVM_MEMSLOT_INVALID;
775 old_memslots = kvm->memslots;
776 rcu_assign_pointer(kvm->memslots, slots);
777 synchronize_srcu_expedited(&kvm->srcu);
778 /* From this point no new shadow pages pointing to a deleted
779 * memslot will be created.
781 * validation of sp->gfn happens in:
782 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
783 * - kvm_is_visible_gfn (mmu_check_roots)
785 kvm_arch_flush_shadow(kvm);
789 r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
793 /* map the pages in iommu page table */
795 r = kvm_iommu_map_pages(kvm, &new);
801 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
804 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
805 if (mem->slot >= slots->nmemslots)
806 slots->nmemslots = mem->slot + 1;
809 /* actual memory is freed via old in kvm_free_physmem_slot below */
812 new.dirty_bitmap = NULL;
813 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i)
814 new.lpage_info[i] = NULL;
817 slots->memslots[mem->slot] = new;
818 old_memslots = kvm->memslots;
819 rcu_assign_pointer(kvm->memslots, slots);
820 synchronize_srcu_expedited(&kvm->srcu);
822 kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
824 kvm_free_physmem_slot(&old, &new);
830 kvm_free_physmem_slot(&new, &old);
835 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
837 int kvm_set_memory_region(struct kvm *kvm,
838 struct kvm_userspace_memory_region *mem,
843 mutex_lock(&kvm->slots_lock);
844 r = __kvm_set_memory_region(kvm, mem, user_alloc);
845 mutex_unlock(&kvm->slots_lock);
848 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
850 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
852 kvm_userspace_memory_region *mem,
855 if (mem->slot >= KVM_MEMORY_SLOTS)
857 return kvm_set_memory_region(kvm, mem, user_alloc);
860 int kvm_get_dirty_log(struct kvm *kvm,
861 struct kvm_dirty_log *log, int *is_dirty)
863 struct kvm_memory_slot *memslot;
866 unsigned long any = 0;
869 if (log->slot >= KVM_MEMORY_SLOTS)
872 memslot = &kvm->memslots->memslots[log->slot];
874 if (!memslot->dirty_bitmap)
877 n = kvm_dirty_bitmap_bytes(memslot);
879 for (i = 0; !any && i < n/sizeof(long); ++i)
880 any = memslot->dirty_bitmap[i];
883 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
894 void kvm_disable_largepages(void)
896 largepages_enabled = false;
898 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
900 int is_error_page(struct page *page)
902 return page == bad_page || page == hwpoison_page || page == fault_page;
904 EXPORT_SYMBOL_GPL(is_error_page);
906 int is_error_pfn(pfn_t pfn)
908 return pfn == bad_pfn || pfn == hwpoison_pfn || pfn == fault_pfn;
910 EXPORT_SYMBOL_GPL(is_error_pfn);
912 int is_hwpoison_pfn(pfn_t pfn)
914 return pfn == hwpoison_pfn;
916 EXPORT_SYMBOL_GPL(is_hwpoison_pfn);
918 int is_fault_pfn(pfn_t pfn)
920 return pfn == fault_pfn;
922 EXPORT_SYMBOL_GPL(is_fault_pfn);
924 static inline unsigned long bad_hva(void)
929 int kvm_is_error_hva(unsigned long addr)
931 return addr == bad_hva();
933 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
935 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm_memslots *slots,
940 for (i = 0; i < slots->nmemslots; ++i) {
941 struct kvm_memory_slot *memslot = &slots->memslots[i];
943 if (gfn >= memslot->base_gfn
944 && gfn < memslot->base_gfn + memslot->npages)
950 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
952 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
954 EXPORT_SYMBOL_GPL(gfn_to_memslot);
956 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
959 struct kvm_memslots *slots = kvm_memslots(kvm);
961 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
962 struct kvm_memory_slot *memslot = &slots->memslots[i];
964 if (memslot->flags & KVM_MEMSLOT_INVALID)
967 if (gfn >= memslot->base_gfn
968 && gfn < memslot->base_gfn + memslot->npages)
973 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
975 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
977 struct vm_area_struct *vma;
978 unsigned long addr, size;
982 addr = gfn_to_hva(kvm, gfn);
983 if (kvm_is_error_hva(addr))
986 down_read(¤t->mm->mmap_sem);
987 vma = find_vma(current->mm, addr);
991 size = vma_kernel_pagesize(vma);
994 up_read(¤t->mm->mmap_sem);
999 int memslot_id(struct kvm *kvm, gfn_t gfn)
1002 struct kvm_memslots *slots = kvm_memslots(kvm);
1003 struct kvm_memory_slot *memslot = NULL;
1005 for (i = 0; i < slots->nmemslots; ++i) {
1006 memslot = &slots->memslots[i];
1008 if (gfn >= memslot->base_gfn
1009 && gfn < memslot->base_gfn + memslot->npages)
1013 return memslot - slots->memslots;
1016 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1019 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1023 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1025 return gfn_to_hva_memslot(slot, gfn);
1028 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1030 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1032 EXPORT_SYMBOL_GPL(gfn_to_hva);
1034 static pfn_t get_fault_pfn(void)
1036 get_page(fault_page);
1040 int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1041 unsigned long start, int write, struct page **page)
1043 int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1046 flags |= FOLL_WRITE;
1048 return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1051 static inline int check_user_page_hwpoison(unsigned long addr)
1053 int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1055 rc = __get_user_pages(current, current->mm, addr, 1,
1056 flags, NULL, NULL, NULL);
1057 return rc == -EHWPOISON;
1060 static pfn_t hva_to_pfn(struct kvm *kvm, unsigned long addr, bool atomic,
1061 bool *async, bool write_fault, bool *writable)
1063 struct page *page[1];
1067 /* we can do it either atomically or asynchronously, not both */
1068 BUG_ON(atomic && async);
1070 BUG_ON(!write_fault && !writable);
1075 if (atomic || async)
1076 npages = __get_user_pages_fast(addr, 1, 1, page);
1078 if (unlikely(npages != 1) && !atomic) {
1082 *writable = write_fault;
1085 down_read(¤t->mm->mmap_sem);
1086 npages = get_user_page_nowait(current, current->mm,
1087 addr, write_fault, page);
1088 up_read(¤t->mm->mmap_sem);
1090 npages = get_user_pages_fast(addr, 1, write_fault,
1093 /* map read fault as writable if possible */
1094 if (unlikely(!write_fault) && npages == 1) {
1095 struct page *wpage[1];
1097 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1107 if (unlikely(npages != 1)) {
1108 struct vm_area_struct *vma;
1111 return get_fault_pfn();
1113 down_read(¤t->mm->mmap_sem);
1114 if (npages == -EHWPOISON ||
1115 (!async && check_user_page_hwpoison(addr))) {
1116 up_read(¤t->mm->mmap_sem);
1117 get_page(hwpoison_page);
1118 return page_to_pfn(hwpoison_page);
1121 vma = find_vma_intersection(current->mm, addr, addr+1);
1124 pfn = get_fault_pfn();
1125 else if ((vma->vm_flags & VM_PFNMAP)) {
1126 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1128 BUG_ON(!kvm_is_mmio_pfn(pfn));
1130 if (async && (vma->vm_flags & VM_WRITE))
1132 pfn = get_fault_pfn();
1134 up_read(¤t->mm->mmap_sem);
1136 pfn = page_to_pfn(page[0]);
1141 pfn_t hva_to_pfn_atomic(struct kvm *kvm, unsigned long addr)
1143 return hva_to_pfn(kvm, addr, true, NULL, true, NULL);
1145 EXPORT_SYMBOL_GPL(hva_to_pfn_atomic);
1147 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1148 bool write_fault, bool *writable)
1155 addr = gfn_to_hva(kvm, gfn);
1156 if (kvm_is_error_hva(addr)) {
1158 return page_to_pfn(bad_page);
1161 return hva_to_pfn(kvm, addr, atomic, async, write_fault, writable);
1164 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1166 return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1168 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1170 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1171 bool write_fault, bool *writable)
1173 return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1175 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1177 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1179 return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1181 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1183 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1186 return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1188 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1190 pfn_t gfn_to_pfn_memslot(struct kvm *kvm,
1191 struct kvm_memory_slot *slot, gfn_t gfn)
1193 unsigned long addr = gfn_to_hva_memslot(slot, gfn);
1194 return hva_to_pfn(kvm, addr, false, NULL, true, NULL);
1197 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1203 addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1204 if (kvm_is_error_hva(addr))
1207 if (entry < nr_pages)
1210 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1212 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1214 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1218 pfn = gfn_to_pfn(kvm, gfn);
1219 if (!kvm_is_mmio_pfn(pfn))
1220 return pfn_to_page(pfn);
1222 WARN_ON(kvm_is_mmio_pfn(pfn));
1228 EXPORT_SYMBOL_GPL(gfn_to_page);
1230 void kvm_release_page_clean(struct page *page)
1232 kvm_release_pfn_clean(page_to_pfn(page));
1234 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1236 void kvm_release_pfn_clean(pfn_t pfn)
1238 if (!kvm_is_mmio_pfn(pfn))
1239 put_page(pfn_to_page(pfn));
1241 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1243 void kvm_release_page_dirty(struct page *page)
1245 kvm_release_pfn_dirty(page_to_pfn(page));
1247 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1249 void kvm_release_pfn_dirty(pfn_t pfn)
1251 kvm_set_pfn_dirty(pfn);
1252 kvm_release_pfn_clean(pfn);
1254 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1256 void kvm_set_page_dirty(struct page *page)
1258 kvm_set_pfn_dirty(page_to_pfn(page));
1260 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1262 void kvm_set_pfn_dirty(pfn_t pfn)
1264 if (!kvm_is_mmio_pfn(pfn)) {
1265 struct page *page = pfn_to_page(pfn);
1266 if (!PageReserved(page))
1270 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1272 void kvm_set_pfn_accessed(pfn_t pfn)
1274 if (!kvm_is_mmio_pfn(pfn))
1275 mark_page_accessed(pfn_to_page(pfn));
1277 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1279 void kvm_get_pfn(pfn_t pfn)
1281 if (!kvm_is_mmio_pfn(pfn))
1282 get_page(pfn_to_page(pfn));
1284 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1286 static int next_segment(unsigned long len, int offset)
1288 if (len > PAGE_SIZE - offset)
1289 return PAGE_SIZE - offset;
1294 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1300 addr = gfn_to_hva(kvm, gfn);
1301 if (kvm_is_error_hva(addr))
1303 r = copy_from_user(data, (void __user *)addr + offset, len);
1308 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1310 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1312 gfn_t gfn = gpa >> PAGE_SHIFT;
1314 int offset = offset_in_page(gpa);
1317 while ((seg = next_segment(len, offset)) != 0) {
1318 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1328 EXPORT_SYMBOL_GPL(kvm_read_guest);
1330 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1335 gfn_t gfn = gpa >> PAGE_SHIFT;
1336 int offset = offset_in_page(gpa);
1338 addr = gfn_to_hva(kvm, gfn);
1339 if (kvm_is_error_hva(addr))
1341 pagefault_disable();
1342 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1348 EXPORT_SYMBOL(kvm_read_guest_atomic);
1350 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1351 int offset, int len)
1356 addr = gfn_to_hva(kvm, gfn);
1357 if (kvm_is_error_hva(addr))
1359 r = copy_to_user((void __user *)addr + offset, data, len);
1362 mark_page_dirty(kvm, gfn);
1365 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1367 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1370 gfn_t gfn = gpa >> PAGE_SHIFT;
1372 int offset = offset_in_page(gpa);
1375 while ((seg = next_segment(len, offset)) != 0) {
1376 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1387 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1390 struct kvm_memslots *slots = kvm_memslots(kvm);
1391 int offset = offset_in_page(gpa);
1392 gfn_t gfn = gpa >> PAGE_SHIFT;
1395 ghc->generation = slots->generation;
1396 ghc->memslot = __gfn_to_memslot(slots, gfn);
1397 ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1398 if (!kvm_is_error_hva(ghc->hva))
1405 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1407 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1408 void *data, unsigned long len)
1410 struct kvm_memslots *slots = kvm_memslots(kvm);
1413 if (slots->generation != ghc->generation)
1414 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1416 if (kvm_is_error_hva(ghc->hva))
1419 r = copy_to_user((void __user *)ghc->hva, data, len);
1422 mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1426 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1428 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1430 return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1433 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1435 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1437 gfn_t gfn = gpa >> PAGE_SHIFT;
1439 int offset = offset_in_page(gpa);
1442 while ((seg = next_segment(len, offset)) != 0) {
1443 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1452 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1454 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1457 if (memslot && memslot->dirty_bitmap) {
1458 unsigned long rel_gfn = gfn - memslot->base_gfn;
1460 __set_bit_le(rel_gfn, memslot->dirty_bitmap);
1464 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1466 struct kvm_memory_slot *memslot;
1468 memslot = gfn_to_memslot(kvm, gfn);
1469 mark_page_dirty_in_slot(kvm, memslot, gfn);
1473 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1475 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1480 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1482 if (kvm_arch_vcpu_runnable(vcpu)) {
1483 kvm_make_request(KVM_REQ_UNHALT, vcpu);
1486 if (kvm_cpu_has_pending_timer(vcpu))
1488 if (signal_pending(current))
1494 finish_wait(&vcpu->wq, &wait);
1497 void kvm_resched(struct kvm_vcpu *vcpu)
1499 if (!need_resched())
1503 EXPORT_SYMBOL_GPL(kvm_resched);
1505 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1507 struct kvm *kvm = me->kvm;
1508 struct kvm_vcpu *vcpu;
1509 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1515 * We boost the priority of a VCPU that is runnable but not
1516 * currently running, because it got preempted by something
1517 * else and called schedule in __vcpu_run. Hopefully that
1518 * VCPU is holding the lock that we need and will release it.
1519 * We approximate round-robin by starting at the last boosted VCPU.
1521 for (pass = 0; pass < 2 && !yielded; pass++) {
1522 kvm_for_each_vcpu(i, vcpu, kvm) {
1523 struct task_struct *task = NULL;
1525 if (!pass && i < last_boosted_vcpu) {
1526 i = last_boosted_vcpu;
1528 } else if (pass && i > last_boosted_vcpu)
1532 if (waitqueue_active(&vcpu->wq))
1535 pid = rcu_dereference(vcpu->pid);
1537 task = get_pid_task(vcpu->pid, PIDTYPE_PID);
1541 if (task->flags & PF_VCPU) {
1542 put_task_struct(task);
1545 if (yield_to(task, 1)) {
1546 put_task_struct(task);
1547 kvm->last_boosted_vcpu = i;
1551 put_task_struct(task);
1555 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1557 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1559 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1562 if (vmf->pgoff == 0)
1563 page = virt_to_page(vcpu->run);
1565 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1566 page = virt_to_page(vcpu->arch.pio_data);
1568 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1569 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1570 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1573 return VM_FAULT_SIGBUS;
1579 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1580 .fault = kvm_vcpu_fault,
1583 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1585 vma->vm_ops = &kvm_vcpu_vm_ops;
1589 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1591 struct kvm_vcpu *vcpu = filp->private_data;
1593 kvm_put_kvm(vcpu->kvm);
1597 static struct file_operations kvm_vcpu_fops = {
1598 .release = kvm_vcpu_release,
1599 .unlocked_ioctl = kvm_vcpu_ioctl,
1600 .compat_ioctl = kvm_vcpu_ioctl,
1601 .mmap = kvm_vcpu_mmap,
1602 .llseek = noop_llseek,
1606 * Allocates an inode for the vcpu.
1608 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1610 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1614 * Creates some virtual cpus. Good luck creating more than one.
1616 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1619 struct kvm_vcpu *vcpu, *v;
1621 vcpu = kvm_arch_vcpu_create(kvm, id);
1623 return PTR_ERR(vcpu);
1625 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1627 r = kvm_arch_vcpu_setup(vcpu);
1631 mutex_lock(&kvm->lock);
1632 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1637 kvm_for_each_vcpu(r, v, kvm)
1638 if (v->vcpu_id == id) {
1643 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1645 /* Now it's all set up, let userspace reach it */
1647 r = create_vcpu_fd(vcpu);
1653 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1655 atomic_inc(&kvm->online_vcpus);
1657 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1658 if (kvm->bsp_vcpu_id == id)
1659 kvm->bsp_vcpu = vcpu;
1661 mutex_unlock(&kvm->lock);
1665 mutex_unlock(&kvm->lock);
1666 kvm_arch_vcpu_destroy(vcpu);
1670 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1673 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1674 vcpu->sigset_active = 1;
1675 vcpu->sigset = *sigset;
1677 vcpu->sigset_active = 0;
1681 static long kvm_vcpu_ioctl(struct file *filp,
1682 unsigned int ioctl, unsigned long arg)
1684 struct kvm_vcpu *vcpu = filp->private_data;
1685 void __user *argp = (void __user *)arg;
1687 struct kvm_fpu *fpu = NULL;
1688 struct kvm_sregs *kvm_sregs = NULL;
1690 if (vcpu->kvm->mm != current->mm)
1693 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1695 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1696 * so vcpu_load() would break it.
1698 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1699 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1709 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1710 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1712 case KVM_GET_REGS: {
1713 struct kvm_regs *kvm_regs;
1716 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1719 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1723 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1730 case KVM_SET_REGS: {
1731 struct kvm_regs *kvm_regs;
1734 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1738 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1740 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1748 case KVM_GET_SREGS: {
1749 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1753 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1757 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1762 case KVM_SET_SREGS: {
1763 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1768 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1770 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1776 case KVM_GET_MP_STATE: {
1777 struct kvm_mp_state mp_state;
1779 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1783 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1788 case KVM_SET_MP_STATE: {
1789 struct kvm_mp_state mp_state;
1792 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1794 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1800 case KVM_TRANSLATE: {
1801 struct kvm_translation tr;
1804 if (copy_from_user(&tr, argp, sizeof tr))
1806 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1810 if (copy_to_user(argp, &tr, sizeof tr))
1815 case KVM_SET_GUEST_DEBUG: {
1816 struct kvm_guest_debug dbg;
1819 if (copy_from_user(&dbg, argp, sizeof dbg))
1821 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1827 case KVM_SET_SIGNAL_MASK: {
1828 struct kvm_signal_mask __user *sigmask_arg = argp;
1829 struct kvm_signal_mask kvm_sigmask;
1830 sigset_t sigset, *p;
1835 if (copy_from_user(&kvm_sigmask, argp,
1836 sizeof kvm_sigmask))
1839 if (kvm_sigmask.len != sizeof sigset)
1842 if (copy_from_user(&sigset, sigmask_arg->sigset,
1847 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
1851 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1855 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1859 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1865 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1870 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1872 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1879 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1888 static long kvm_vm_ioctl(struct file *filp,
1889 unsigned int ioctl, unsigned long arg)
1891 struct kvm *kvm = filp->private_data;
1892 void __user *argp = (void __user *)arg;
1895 if (kvm->mm != current->mm)
1898 case KVM_CREATE_VCPU:
1899 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1903 case KVM_SET_USER_MEMORY_REGION: {
1904 struct kvm_userspace_memory_region kvm_userspace_mem;
1907 if (copy_from_user(&kvm_userspace_mem, argp,
1908 sizeof kvm_userspace_mem))
1911 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1916 case KVM_GET_DIRTY_LOG: {
1917 struct kvm_dirty_log log;
1920 if (copy_from_user(&log, argp, sizeof log))
1922 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1927 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1928 case KVM_REGISTER_COALESCED_MMIO: {
1929 struct kvm_coalesced_mmio_zone zone;
1931 if (copy_from_user(&zone, argp, sizeof zone))
1933 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1939 case KVM_UNREGISTER_COALESCED_MMIO: {
1940 struct kvm_coalesced_mmio_zone zone;
1942 if (copy_from_user(&zone, argp, sizeof zone))
1944 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1952 struct kvm_irqfd data;
1955 if (copy_from_user(&data, argp, sizeof data))
1957 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
1960 case KVM_IOEVENTFD: {
1961 struct kvm_ioeventfd data;
1964 if (copy_from_user(&data, argp, sizeof data))
1966 r = kvm_ioeventfd(kvm, &data);
1969 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1970 case KVM_SET_BOOT_CPU_ID:
1972 mutex_lock(&kvm->lock);
1973 if (atomic_read(&kvm->online_vcpus) != 0)
1976 kvm->bsp_vcpu_id = arg;
1977 mutex_unlock(&kvm->lock);
1981 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1983 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
1989 #ifdef CONFIG_COMPAT
1990 struct compat_kvm_dirty_log {
1994 compat_uptr_t dirty_bitmap; /* one bit per page */
1999 static long kvm_vm_compat_ioctl(struct file *filp,
2000 unsigned int ioctl, unsigned long arg)
2002 struct kvm *kvm = filp->private_data;
2005 if (kvm->mm != current->mm)
2008 case KVM_GET_DIRTY_LOG: {
2009 struct compat_kvm_dirty_log compat_log;
2010 struct kvm_dirty_log log;
2013 if (copy_from_user(&compat_log, (void __user *)arg,
2014 sizeof(compat_log)))
2016 log.slot = compat_log.slot;
2017 log.padding1 = compat_log.padding1;
2018 log.padding2 = compat_log.padding2;
2019 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2021 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2027 r = kvm_vm_ioctl(filp, ioctl, arg);
2035 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2037 struct page *page[1];
2040 gfn_t gfn = vmf->pgoff;
2041 struct kvm *kvm = vma->vm_file->private_data;
2043 addr = gfn_to_hva(kvm, gfn);
2044 if (kvm_is_error_hva(addr))
2045 return VM_FAULT_SIGBUS;
2047 npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2049 if (unlikely(npages != 1))
2050 return VM_FAULT_SIGBUS;
2052 vmf->page = page[0];
2056 static const struct vm_operations_struct kvm_vm_vm_ops = {
2057 .fault = kvm_vm_fault,
2060 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2062 vma->vm_ops = &kvm_vm_vm_ops;
2066 static struct file_operations kvm_vm_fops = {
2067 .release = kvm_vm_release,
2068 .unlocked_ioctl = kvm_vm_ioctl,
2069 #ifdef CONFIG_COMPAT
2070 .compat_ioctl = kvm_vm_compat_ioctl,
2072 .mmap = kvm_vm_mmap,
2073 .llseek = noop_llseek,
2076 static int kvm_dev_ioctl_create_vm(void)
2081 kvm = kvm_create_vm();
2083 return PTR_ERR(kvm);
2084 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2085 r = kvm_coalesced_mmio_init(kvm);
2091 r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2098 static long kvm_dev_ioctl_check_extension_generic(long arg)
2101 case KVM_CAP_USER_MEMORY:
2102 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2103 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2104 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2105 case KVM_CAP_SET_BOOT_CPU_ID:
2107 case KVM_CAP_INTERNAL_ERROR_DATA:
2109 #ifdef CONFIG_HAVE_KVM_IRQCHIP
2110 case KVM_CAP_IRQ_ROUTING:
2111 return KVM_MAX_IRQ_ROUTES;
2116 return kvm_dev_ioctl_check_extension(arg);
2119 static long kvm_dev_ioctl(struct file *filp,
2120 unsigned int ioctl, unsigned long arg)
2125 case KVM_GET_API_VERSION:
2129 r = KVM_API_VERSION;
2135 r = kvm_dev_ioctl_create_vm();
2137 case KVM_CHECK_EXTENSION:
2138 r = kvm_dev_ioctl_check_extension_generic(arg);
2140 case KVM_GET_VCPU_MMAP_SIZE:
2144 r = PAGE_SIZE; /* struct kvm_run */
2146 r += PAGE_SIZE; /* pio data page */
2148 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2149 r += PAGE_SIZE; /* coalesced mmio ring page */
2152 case KVM_TRACE_ENABLE:
2153 case KVM_TRACE_PAUSE:
2154 case KVM_TRACE_DISABLE:
2158 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2164 static struct file_operations kvm_chardev_ops = {
2165 .unlocked_ioctl = kvm_dev_ioctl,
2166 .compat_ioctl = kvm_dev_ioctl,
2167 .llseek = noop_llseek,
2170 static struct miscdevice kvm_dev = {
2176 static void hardware_enable_nolock(void *junk)
2178 int cpu = raw_smp_processor_id();
2181 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2184 cpumask_set_cpu(cpu, cpus_hardware_enabled);
2186 r = kvm_arch_hardware_enable(NULL);
2189 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2190 atomic_inc(&hardware_enable_failed);
2191 printk(KERN_INFO "kvm: enabling virtualization on "
2192 "CPU%d failed\n", cpu);
2196 static void hardware_enable(void *junk)
2198 raw_spin_lock(&kvm_lock);
2199 hardware_enable_nolock(junk);
2200 raw_spin_unlock(&kvm_lock);
2203 static void hardware_disable_nolock(void *junk)
2205 int cpu = raw_smp_processor_id();
2207 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2209 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2210 kvm_arch_hardware_disable(NULL);
2213 static void hardware_disable(void *junk)
2215 raw_spin_lock(&kvm_lock);
2216 hardware_disable_nolock(junk);
2217 raw_spin_unlock(&kvm_lock);
2220 static void hardware_disable_all_nolock(void)
2222 BUG_ON(!kvm_usage_count);
2225 if (!kvm_usage_count)
2226 on_each_cpu(hardware_disable_nolock, NULL, 1);
2229 static void hardware_disable_all(void)
2231 raw_spin_lock(&kvm_lock);
2232 hardware_disable_all_nolock();
2233 raw_spin_unlock(&kvm_lock);
2236 static int hardware_enable_all(void)
2240 raw_spin_lock(&kvm_lock);
2243 if (kvm_usage_count == 1) {
2244 atomic_set(&hardware_enable_failed, 0);
2245 on_each_cpu(hardware_enable_nolock, NULL, 1);
2247 if (atomic_read(&hardware_enable_failed)) {
2248 hardware_disable_all_nolock();
2253 raw_spin_unlock(&kvm_lock);
2258 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2263 if (!kvm_usage_count)
2266 val &= ~CPU_TASKS_FROZEN;
2269 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2271 hardware_disable(NULL);
2274 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2276 hardware_enable(NULL);
2283 asmlinkage void kvm_spurious_fault(void)
2285 /* Fault while not rebooting. We want the trace. */
2288 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
2290 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2294 * Some (well, at least mine) BIOSes hang on reboot if
2297 * And Intel TXT required VMX off for all cpu when system shutdown.
2299 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2300 kvm_rebooting = true;
2301 on_each_cpu(hardware_disable_nolock, NULL, 1);
2305 static struct notifier_block kvm_reboot_notifier = {
2306 .notifier_call = kvm_reboot,
2310 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2314 for (i = 0; i < bus->dev_count; i++) {
2315 struct kvm_io_device *pos = bus->devs[i];
2317 kvm_iodevice_destructor(pos);
2322 /* kvm_io_bus_write - called under kvm->slots_lock */
2323 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2324 int len, const void *val)
2327 struct kvm_io_bus *bus;
2329 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2330 for (i = 0; i < bus->dev_count; i++)
2331 if (!kvm_iodevice_write(bus->devs[i], addr, len, val))
2336 /* kvm_io_bus_read - called under kvm->slots_lock */
2337 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2341 struct kvm_io_bus *bus;
2343 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2344 for (i = 0; i < bus->dev_count; i++)
2345 if (!kvm_iodevice_read(bus->devs[i], addr, len, val))
2350 /* Caller must hold slots_lock. */
2351 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2352 struct kvm_io_device *dev)
2354 struct kvm_io_bus *new_bus, *bus;
2356 bus = kvm->buses[bus_idx];
2357 if (bus->dev_count > NR_IOBUS_DEVS-1)
2360 new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2363 memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2364 new_bus->devs[new_bus->dev_count++] = dev;
2365 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2366 synchronize_srcu_expedited(&kvm->srcu);
2372 /* Caller must hold slots_lock. */
2373 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2374 struct kvm_io_device *dev)
2377 struct kvm_io_bus *new_bus, *bus;
2379 new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2383 bus = kvm->buses[bus_idx];
2384 memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2387 for (i = 0; i < new_bus->dev_count; i++)
2388 if (new_bus->devs[i] == dev) {
2390 new_bus->devs[i] = new_bus->devs[--new_bus->dev_count];
2399 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2400 synchronize_srcu_expedited(&kvm->srcu);
2405 static struct notifier_block kvm_cpu_notifier = {
2406 .notifier_call = kvm_cpu_hotplug,
2409 static int vm_stat_get(void *_offset, u64 *val)
2411 unsigned offset = (long)_offset;
2415 raw_spin_lock(&kvm_lock);
2416 list_for_each_entry(kvm, &vm_list, vm_list)
2417 *val += *(u32 *)((void *)kvm + offset);
2418 raw_spin_unlock(&kvm_lock);
2422 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2424 static int vcpu_stat_get(void *_offset, u64 *val)
2426 unsigned offset = (long)_offset;
2428 struct kvm_vcpu *vcpu;
2432 raw_spin_lock(&kvm_lock);
2433 list_for_each_entry(kvm, &vm_list, vm_list)
2434 kvm_for_each_vcpu(i, vcpu, kvm)
2435 *val += *(u32 *)((void *)vcpu + offset);
2437 raw_spin_unlock(&kvm_lock);
2441 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2443 static const struct file_operations *stat_fops[] = {
2444 [KVM_STAT_VCPU] = &vcpu_stat_fops,
2445 [KVM_STAT_VM] = &vm_stat_fops,
2448 static void kvm_init_debug(void)
2450 struct kvm_stats_debugfs_item *p;
2452 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2453 for (p = debugfs_entries; p->name; ++p)
2454 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2455 (void *)(long)p->offset,
2456 stat_fops[p->kind]);
2459 static void kvm_exit_debug(void)
2461 struct kvm_stats_debugfs_item *p;
2463 for (p = debugfs_entries; p->name; ++p)
2464 debugfs_remove(p->dentry);
2465 debugfs_remove(kvm_debugfs_dir);
2468 static int kvm_suspend(void)
2470 if (kvm_usage_count)
2471 hardware_disable_nolock(NULL);
2475 static void kvm_resume(void)
2477 if (kvm_usage_count) {
2478 WARN_ON(raw_spin_is_locked(&kvm_lock));
2479 hardware_enable_nolock(NULL);
2483 static struct syscore_ops kvm_syscore_ops = {
2484 .suspend = kvm_suspend,
2485 .resume = kvm_resume,
2488 struct page *bad_page;
2492 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2494 return container_of(pn, struct kvm_vcpu, preempt_notifier);
2497 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2499 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2501 kvm_arch_vcpu_load(vcpu, cpu);
2504 static void kvm_sched_out(struct preempt_notifier *pn,
2505 struct task_struct *next)
2507 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2509 kvm_arch_vcpu_put(vcpu);
2512 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2513 struct module *module)
2518 r = kvm_arch_init(opaque);
2522 bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2524 if (bad_page == NULL) {
2529 bad_pfn = page_to_pfn(bad_page);
2531 hwpoison_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2533 if (hwpoison_page == NULL) {
2538 hwpoison_pfn = page_to_pfn(hwpoison_page);
2540 fault_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2542 if (fault_page == NULL) {
2547 fault_pfn = page_to_pfn(fault_page);
2549 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2554 r = kvm_arch_hardware_setup();
2558 for_each_online_cpu(cpu) {
2559 smp_call_function_single(cpu,
2560 kvm_arch_check_processor_compat,
2566 r = register_cpu_notifier(&kvm_cpu_notifier);
2569 register_reboot_notifier(&kvm_reboot_notifier);
2571 /* A kmem cache lets us meet the alignment requirements of fx_save. */
2573 vcpu_align = __alignof__(struct kvm_vcpu);
2574 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2576 if (!kvm_vcpu_cache) {
2581 r = kvm_async_pf_init();
2585 kvm_chardev_ops.owner = module;
2586 kvm_vm_fops.owner = module;
2587 kvm_vcpu_fops.owner = module;
2589 r = misc_register(&kvm_dev);
2591 printk(KERN_ERR "kvm: misc device register failed\n");
2595 register_syscore_ops(&kvm_syscore_ops);
2597 kvm_preempt_ops.sched_in = kvm_sched_in;
2598 kvm_preempt_ops.sched_out = kvm_sched_out;
2605 kvm_async_pf_deinit();
2607 kmem_cache_destroy(kvm_vcpu_cache);
2609 unregister_reboot_notifier(&kvm_reboot_notifier);
2610 unregister_cpu_notifier(&kvm_cpu_notifier);
2613 kvm_arch_hardware_unsetup();
2615 free_cpumask_var(cpus_hardware_enabled);
2618 __free_page(fault_page);
2620 __free_page(hwpoison_page);
2621 __free_page(bad_page);
2627 EXPORT_SYMBOL_GPL(kvm_init);
2632 misc_deregister(&kvm_dev);
2633 kmem_cache_destroy(kvm_vcpu_cache);
2634 kvm_async_pf_deinit();
2635 unregister_syscore_ops(&kvm_syscore_ops);
2636 unregister_reboot_notifier(&kvm_reboot_notifier);
2637 unregister_cpu_notifier(&kvm_cpu_notifier);
2638 on_each_cpu(hardware_disable_nolock, NULL, 1);
2639 kvm_arch_hardware_unsetup();
2641 free_cpumask_var(cpus_hardware_enabled);
2642 __free_page(hwpoison_page);
2643 __free_page(bad_page);
2645 EXPORT_SYMBOL_GPL(kvm_exit);