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/sysdev.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>
55 #include <asm-generic/bitops/le.h>
57 #include "coalesced_mmio.h"
60 #define CREATE_TRACE_POINTS
61 #include <trace/events/kvm.h>
63 MODULE_AUTHOR("Qumranet");
64 MODULE_LICENSE("GPL");
69 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
72 DEFINE_SPINLOCK(kvm_lock);
75 static cpumask_var_t cpus_hardware_enabled;
76 static int kvm_usage_count = 0;
77 static atomic_t hardware_enable_failed;
79 struct kmem_cache *kvm_vcpu_cache;
80 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
82 static __read_mostly struct preempt_ops kvm_preempt_ops;
84 struct dentry *kvm_debugfs_dir;
86 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
88 static int hardware_enable_all(void);
89 static void hardware_disable_all(void);
91 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
94 EXPORT_SYMBOL_GPL(kvm_rebooting);
96 static bool largepages_enabled = true;
98 static struct page *hwpoison_page;
99 static pfn_t hwpoison_pfn;
101 static struct page *fault_page;
102 static pfn_t fault_pfn;
104 inline int kvm_is_mmio_pfn(pfn_t pfn)
106 if (pfn_valid(pfn)) {
108 struct page *tail = pfn_to_page(pfn);
109 struct page *head = compound_trans_head(tail);
110 reserved = PageReserved(head);
113 * "head" is not a dangling pointer
114 * (compound_trans_head takes care of that)
115 * but the hugepage may have been splitted
116 * from under us (and we may not hold a
117 * reference count on the head page so it can
118 * be reused before we run PageReferenced), so
119 * we've to check PageTail before returning
126 return PageReserved(tail);
133 * Switches to specified vcpu, until a matching vcpu_put()
135 void vcpu_load(struct kvm_vcpu *vcpu)
139 mutex_lock(&vcpu->mutex);
141 preempt_notifier_register(&vcpu->preempt_notifier);
142 kvm_arch_vcpu_load(vcpu, cpu);
146 void vcpu_put(struct kvm_vcpu *vcpu)
149 kvm_arch_vcpu_put(vcpu);
150 preempt_notifier_unregister(&vcpu->preempt_notifier);
152 mutex_unlock(&vcpu->mutex);
155 static void ack_flush(void *_completed)
159 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
164 struct kvm_vcpu *vcpu;
166 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
168 raw_spin_lock(&kvm->requests_lock);
169 me = smp_processor_id();
170 kvm_for_each_vcpu(i, vcpu, kvm) {
171 if (kvm_make_check_request(req, vcpu))
174 if (cpus != NULL && cpu != -1 && cpu != me)
175 cpumask_set_cpu(cpu, cpus);
177 if (unlikely(cpus == NULL))
178 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
179 else if (!cpumask_empty(cpus))
180 smp_call_function_many(cpus, ack_flush, NULL, 1);
183 raw_spin_unlock(&kvm->requests_lock);
184 free_cpumask_var(cpus);
188 void kvm_flush_remote_tlbs(struct kvm *kvm)
190 int dirty_count = kvm->tlbs_dirty;
193 if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
194 ++kvm->stat.remote_tlb_flush;
195 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
198 void kvm_reload_remote_mmus(struct kvm *kvm)
200 make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
203 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
208 mutex_init(&vcpu->mutex);
212 init_waitqueue_head(&vcpu->wq);
213 kvm_async_pf_vcpu_init(vcpu);
215 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
220 vcpu->run = page_address(page);
222 r = kvm_arch_vcpu_init(vcpu);
228 free_page((unsigned long)vcpu->run);
232 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
234 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
236 kvm_arch_vcpu_uninit(vcpu);
237 free_page((unsigned long)vcpu->run);
239 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
241 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
242 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
244 return container_of(mn, struct kvm, mmu_notifier);
247 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
248 struct mm_struct *mm,
249 unsigned long address)
251 struct kvm *kvm = mmu_notifier_to_kvm(mn);
252 int need_tlb_flush, idx;
255 * When ->invalidate_page runs, the linux pte has been zapped
256 * already but the page is still allocated until
257 * ->invalidate_page returns. So if we increase the sequence
258 * here the kvm page fault will notice if the spte can't be
259 * established because the page is going to be freed. If
260 * instead the kvm page fault establishes the spte before
261 * ->invalidate_page runs, kvm_unmap_hva will release it
264 * The sequence increase only need to be seen at spin_unlock
265 * time, and not at spin_lock time.
267 * Increasing the sequence after the spin_unlock would be
268 * unsafe because the kvm page fault could then establish the
269 * pte after kvm_unmap_hva returned, without noticing the page
270 * is going to be freed.
272 idx = srcu_read_lock(&kvm->srcu);
273 spin_lock(&kvm->mmu_lock);
274 kvm->mmu_notifier_seq++;
275 need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
276 spin_unlock(&kvm->mmu_lock);
277 srcu_read_unlock(&kvm->srcu, idx);
279 /* we've to flush the tlb before the pages can be freed */
281 kvm_flush_remote_tlbs(kvm);
285 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
286 struct mm_struct *mm,
287 unsigned long address,
290 struct kvm *kvm = mmu_notifier_to_kvm(mn);
293 idx = srcu_read_lock(&kvm->srcu);
294 spin_lock(&kvm->mmu_lock);
295 kvm->mmu_notifier_seq++;
296 kvm_set_spte_hva(kvm, address, pte);
297 spin_unlock(&kvm->mmu_lock);
298 srcu_read_unlock(&kvm->srcu, idx);
301 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
302 struct mm_struct *mm,
306 struct kvm *kvm = mmu_notifier_to_kvm(mn);
307 int need_tlb_flush = 0, idx;
309 idx = srcu_read_lock(&kvm->srcu);
310 spin_lock(&kvm->mmu_lock);
312 * The count increase must become visible at unlock time as no
313 * spte can be established without taking the mmu_lock and
314 * count is also read inside the mmu_lock critical section.
316 kvm->mmu_notifier_count++;
317 for (; start < end; start += PAGE_SIZE)
318 need_tlb_flush |= kvm_unmap_hva(kvm, start);
319 need_tlb_flush |= kvm->tlbs_dirty;
320 spin_unlock(&kvm->mmu_lock);
321 srcu_read_unlock(&kvm->srcu, idx);
323 /* we've to flush the tlb before the pages can be freed */
325 kvm_flush_remote_tlbs(kvm);
328 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
329 struct mm_struct *mm,
333 struct kvm *kvm = mmu_notifier_to_kvm(mn);
335 spin_lock(&kvm->mmu_lock);
337 * This sequence increase will notify the kvm page fault that
338 * the page that is going to be mapped in the spte could have
341 kvm->mmu_notifier_seq++;
343 * The above sequence increase must be visible before the
344 * below count decrease but both values are read by the kvm
345 * page fault under mmu_lock spinlock so we don't need to add
346 * a smb_wmb() here in between the two.
348 kvm->mmu_notifier_count--;
349 spin_unlock(&kvm->mmu_lock);
351 BUG_ON(kvm->mmu_notifier_count < 0);
354 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
355 struct mm_struct *mm,
356 unsigned long address)
358 struct kvm *kvm = mmu_notifier_to_kvm(mn);
361 idx = srcu_read_lock(&kvm->srcu);
362 spin_lock(&kvm->mmu_lock);
363 young = kvm_age_hva(kvm, address);
364 spin_unlock(&kvm->mmu_lock);
365 srcu_read_unlock(&kvm->srcu, idx);
368 kvm_flush_remote_tlbs(kvm);
373 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
374 struct mm_struct *mm,
375 unsigned long address)
377 struct kvm *kvm = mmu_notifier_to_kvm(mn);
380 idx = srcu_read_lock(&kvm->srcu);
381 spin_lock(&kvm->mmu_lock);
382 young = kvm_test_age_hva(kvm, address);
383 spin_unlock(&kvm->mmu_lock);
384 srcu_read_unlock(&kvm->srcu, idx);
389 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
390 struct mm_struct *mm)
392 struct kvm *kvm = mmu_notifier_to_kvm(mn);
395 idx = srcu_read_lock(&kvm->srcu);
396 kvm_arch_flush_shadow(kvm);
397 srcu_read_unlock(&kvm->srcu, idx);
400 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
401 .invalidate_page = kvm_mmu_notifier_invalidate_page,
402 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
403 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
404 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
405 .test_young = kvm_mmu_notifier_test_young,
406 .change_pte = kvm_mmu_notifier_change_pte,
407 .release = kvm_mmu_notifier_release,
410 static int kvm_init_mmu_notifier(struct kvm *kvm)
412 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
413 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
416 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
418 static int kvm_init_mmu_notifier(struct kvm *kvm)
423 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
425 static struct kvm *kvm_create_vm(void)
428 struct kvm *kvm = kvm_arch_alloc_vm();
431 return ERR_PTR(-ENOMEM);
433 r = kvm_arch_init_vm(kvm);
435 goto out_err_nodisable;
437 r = hardware_enable_all();
439 goto out_err_nodisable;
441 #ifdef CONFIG_HAVE_KVM_IRQCHIP
442 INIT_HLIST_HEAD(&kvm->mask_notifier_list);
443 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
447 kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
450 if (init_srcu_struct(&kvm->srcu))
452 for (i = 0; i < KVM_NR_BUSES; i++) {
453 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
459 r = kvm_init_mmu_notifier(kvm);
463 kvm->mm = current->mm;
464 atomic_inc(&kvm->mm->mm_count);
465 spin_lock_init(&kvm->mmu_lock);
466 raw_spin_lock_init(&kvm->requests_lock);
467 kvm_eventfd_init(kvm);
468 mutex_init(&kvm->lock);
469 mutex_init(&kvm->irq_lock);
470 mutex_init(&kvm->slots_lock);
471 atomic_set(&kvm->users_count, 1);
472 spin_lock(&kvm_lock);
473 list_add(&kvm->vm_list, &vm_list);
474 spin_unlock(&kvm_lock);
479 cleanup_srcu_struct(&kvm->srcu);
481 hardware_disable_all();
483 for (i = 0; i < KVM_NR_BUSES; i++)
484 kfree(kvm->buses[i]);
485 kfree(kvm->memslots);
486 kvm_arch_free_vm(kvm);
490 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
492 if (!memslot->dirty_bitmap)
495 if (2 * kvm_dirty_bitmap_bytes(memslot) > PAGE_SIZE)
496 vfree(memslot->dirty_bitmap_head);
498 kfree(memslot->dirty_bitmap_head);
500 memslot->dirty_bitmap = NULL;
501 memslot->dirty_bitmap_head = NULL;
505 * Free any memory in @free but not in @dont.
507 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
508 struct kvm_memory_slot *dont)
512 if (!dont || free->rmap != dont->rmap)
515 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
516 kvm_destroy_dirty_bitmap(free);
519 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
520 if (!dont || free->lpage_info[i] != dont->lpage_info[i]) {
521 vfree(free->lpage_info[i]);
522 free->lpage_info[i] = NULL;
530 void kvm_free_physmem(struct kvm *kvm)
533 struct kvm_memslots *slots = kvm->memslots;
535 for (i = 0; i < slots->nmemslots; ++i)
536 kvm_free_physmem_slot(&slots->memslots[i], NULL);
538 kfree(kvm->memslots);
541 static void kvm_destroy_vm(struct kvm *kvm)
544 struct mm_struct *mm = kvm->mm;
546 kvm_arch_sync_events(kvm);
547 spin_lock(&kvm_lock);
548 list_del(&kvm->vm_list);
549 spin_unlock(&kvm_lock);
550 kvm_free_irq_routing(kvm);
551 for (i = 0; i < KVM_NR_BUSES; i++)
552 kvm_io_bus_destroy(kvm->buses[i]);
553 kvm_coalesced_mmio_free(kvm);
554 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
555 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
557 kvm_arch_flush_shadow(kvm);
559 kvm_arch_destroy_vm(kvm);
560 kvm_free_physmem(kvm);
561 cleanup_srcu_struct(&kvm->srcu);
562 kvm_arch_free_vm(kvm);
563 hardware_disable_all();
567 void kvm_get_kvm(struct kvm *kvm)
569 atomic_inc(&kvm->users_count);
571 EXPORT_SYMBOL_GPL(kvm_get_kvm);
573 void kvm_put_kvm(struct kvm *kvm)
575 if (atomic_dec_and_test(&kvm->users_count))
578 EXPORT_SYMBOL_GPL(kvm_put_kvm);
581 static int kvm_vm_release(struct inode *inode, struct file *filp)
583 struct kvm *kvm = filp->private_data;
585 kvm_irqfd_release(kvm);
592 * Allocation size is twice as large as the actual dirty bitmap size.
593 * This makes it possible to do double buffering: see x86's
594 * kvm_vm_ioctl_get_dirty_log().
596 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
598 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
600 if (dirty_bytes > PAGE_SIZE)
601 memslot->dirty_bitmap = vzalloc(dirty_bytes);
603 memslot->dirty_bitmap = kzalloc(dirty_bytes, GFP_KERNEL);
605 if (!memslot->dirty_bitmap)
608 memslot->dirty_bitmap_head = memslot->dirty_bitmap;
613 * Allocate some memory and give it an address in the guest physical address
616 * Discontiguous memory is allowed, mostly for framebuffers.
618 * Must be called holding mmap_sem for write.
620 int __kvm_set_memory_region(struct kvm *kvm,
621 struct kvm_userspace_memory_region *mem,
624 int r, flush_shadow = 0;
626 unsigned long npages;
628 struct kvm_memory_slot *memslot;
629 struct kvm_memory_slot old, new;
630 struct kvm_memslots *slots, *old_memslots;
633 /* General sanity checks */
634 if (mem->memory_size & (PAGE_SIZE - 1))
636 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
638 if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
640 if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
642 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
645 memslot = &kvm->memslots->memslots[mem->slot];
646 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
647 npages = mem->memory_size >> PAGE_SHIFT;
650 if (npages > KVM_MEM_MAX_NR_PAGES)
654 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
656 new = old = *memslot;
659 new.base_gfn = base_gfn;
661 new.flags = mem->flags;
663 /* Disallow changing a memory slot's size. */
665 if (npages && old.npages && npages != old.npages)
668 /* Check for overlaps */
670 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
671 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
673 if (s == memslot || !s->npages)
675 if (!((base_gfn + npages <= s->base_gfn) ||
676 (base_gfn >= s->base_gfn + s->npages)))
680 /* Free page dirty bitmap if unneeded */
681 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
682 new.dirty_bitmap = NULL;
686 /* Allocate if a slot is being created */
688 if (npages && !new.rmap) {
689 new.rmap = vzalloc(npages * sizeof(*new.rmap));
694 new.user_alloc = user_alloc;
695 new.userspace_addr = mem->userspace_addr;
700 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
706 /* Avoid unused variable warning if no large pages */
709 if (new.lpage_info[i])
712 lpages = 1 + ((base_gfn + npages - 1)
713 >> KVM_HPAGE_GFN_SHIFT(level));
714 lpages -= base_gfn >> KVM_HPAGE_GFN_SHIFT(level);
716 new.lpage_info[i] = vzalloc(lpages * sizeof(*new.lpage_info[i]));
718 if (!new.lpage_info[i])
721 if (base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
722 new.lpage_info[i][0].write_count = 1;
723 if ((base_gfn+npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
724 new.lpage_info[i][lpages - 1].write_count = 1;
725 ugfn = new.userspace_addr >> PAGE_SHIFT;
727 * If the gfn and userspace address are not aligned wrt each
728 * other, or if explicitly asked to, disable large page
729 * support for this slot
731 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
733 for (j = 0; j < lpages; ++j)
734 new.lpage_info[i][j].write_count = 1;
739 /* Allocate page dirty bitmap if needed */
740 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
741 if (kvm_create_dirty_bitmap(&new) < 0)
743 /* destroy any largepage mappings for dirty tracking */
747 #else /* not defined CONFIG_S390 */
748 new.user_alloc = user_alloc;
750 new.userspace_addr = mem->userspace_addr;
751 #endif /* not defined CONFIG_S390 */
755 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
758 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
759 if (mem->slot >= slots->nmemslots)
760 slots->nmemslots = mem->slot + 1;
762 slots->memslots[mem->slot].flags |= KVM_MEMSLOT_INVALID;
764 old_memslots = kvm->memslots;
765 rcu_assign_pointer(kvm->memslots, slots);
766 synchronize_srcu_expedited(&kvm->srcu);
767 /* From this point no new shadow pages pointing to a deleted
768 * memslot will be created.
770 * validation of sp->gfn happens in:
771 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
772 * - kvm_is_visible_gfn (mmu_check_roots)
774 kvm_arch_flush_shadow(kvm);
778 r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
782 /* map the pages in iommu page table */
784 r = kvm_iommu_map_pages(kvm, &new);
790 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
793 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
794 if (mem->slot >= slots->nmemslots)
795 slots->nmemslots = mem->slot + 1;
798 /* actual memory is freed via old in kvm_free_physmem_slot below */
801 new.dirty_bitmap = NULL;
802 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i)
803 new.lpage_info[i] = NULL;
806 slots->memslots[mem->slot] = new;
807 old_memslots = kvm->memslots;
808 rcu_assign_pointer(kvm->memslots, slots);
809 synchronize_srcu_expedited(&kvm->srcu);
811 kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
813 kvm_free_physmem_slot(&old, &new);
817 kvm_arch_flush_shadow(kvm);
822 kvm_free_physmem_slot(&new, &old);
827 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
829 int kvm_set_memory_region(struct kvm *kvm,
830 struct kvm_userspace_memory_region *mem,
835 mutex_lock(&kvm->slots_lock);
836 r = __kvm_set_memory_region(kvm, mem, user_alloc);
837 mutex_unlock(&kvm->slots_lock);
840 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
842 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
844 kvm_userspace_memory_region *mem,
847 if (mem->slot >= KVM_MEMORY_SLOTS)
849 return kvm_set_memory_region(kvm, mem, user_alloc);
852 int kvm_get_dirty_log(struct kvm *kvm,
853 struct kvm_dirty_log *log, int *is_dirty)
855 struct kvm_memory_slot *memslot;
858 unsigned long any = 0;
861 if (log->slot >= KVM_MEMORY_SLOTS)
864 memslot = &kvm->memslots->memslots[log->slot];
866 if (!memslot->dirty_bitmap)
869 n = kvm_dirty_bitmap_bytes(memslot);
871 for (i = 0; !any && i < n/sizeof(long); ++i)
872 any = memslot->dirty_bitmap[i];
875 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
886 void kvm_disable_largepages(void)
888 largepages_enabled = false;
890 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
892 int is_error_page(struct page *page)
894 return page == bad_page || page == hwpoison_page || page == fault_page;
896 EXPORT_SYMBOL_GPL(is_error_page);
898 int is_error_pfn(pfn_t pfn)
900 return pfn == bad_pfn || pfn == hwpoison_pfn || pfn == fault_pfn;
902 EXPORT_SYMBOL_GPL(is_error_pfn);
904 int is_hwpoison_pfn(pfn_t pfn)
906 return pfn == hwpoison_pfn;
908 EXPORT_SYMBOL_GPL(is_hwpoison_pfn);
910 int is_fault_pfn(pfn_t pfn)
912 return pfn == fault_pfn;
914 EXPORT_SYMBOL_GPL(is_fault_pfn);
916 static inline unsigned long bad_hva(void)
921 int kvm_is_error_hva(unsigned long addr)
923 return addr == bad_hva();
925 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
927 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm_memslots *slots,
932 for (i = 0; i < slots->nmemslots; ++i) {
933 struct kvm_memory_slot *memslot = &slots->memslots[i];
935 if (gfn >= memslot->base_gfn
936 && gfn < memslot->base_gfn + memslot->npages)
942 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
944 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
946 EXPORT_SYMBOL_GPL(gfn_to_memslot);
948 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
951 struct kvm_memslots *slots = kvm_memslots(kvm);
953 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
954 struct kvm_memory_slot *memslot = &slots->memslots[i];
956 if (memslot->flags & KVM_MEMSLOT_INVALID)
959 if (gfn >= memslot->base_gfn
960 && gfn < memslot->base_gfn + memslot->npages)
965 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
967 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
969 struct vm_area_struct *vma;
970 unsigned long addr, size;
974 addr = gfn_to_hva(kvm, gfn);
975 if (kvm_is_error_hva(addr))
978 down_read(¤t->mm->mmap_sem);
979 vma = find_vma(current->mm, addr);
983 size = vma_kernel_pagesize(vma);
986 up_read(¤t->mm->mmap_sem);
991 int memslot_id(struct kvm *kvm, gfn_t gfn)
994 struct kvm_memslots *slots = kvm_memslots(kvm);
995 struct kvm_memory_slot *memslot = NULL;
997 for (i = 0; i < slots->nmemslots; ++i) {
998 memslot = &slots->memslots[i];
1000 if (gfn >= memslot->base_gfn
1001 && gfn < memslot->base_gfn + memslot->npages)
1005 return memslot - slots->memslots;
1008 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1011 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1015 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1017 return gfn_to_hva_memslot(slot, gfn);
1020 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1022 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1024 EXPORT_SYMBOL_GPL(gfn_to_hva);
1026 static pfn_t get_fault_pfn(void)
1028 get_page(fault_page);
1032 static pfn_t hva_to_pfn(struct kvm *kvm, unsigned long addr, bool atomic,
1033 bool *async, bool write_fault, bool *writable)
1035 struct page *page[1];
1039 /* we can do it either atomically or asynchronously, not both */
1040 BUG_ON(atomic && async);
1042 BUG_ON(!write_fault && !writable);
1047 if (atomic || async)
1048 npages = __get_user_pages_fast(addr, 1, 1, page);
1050 if (unlikely(npages != 1) && !atomic) {
1054 *writable = write_fault;
1056 npages = get_user_pages_fast(addr, 1, write_fault, page);
1058 /* map read fault as writable if possible */
1059 if (unlikely(!write_fault) && npages == 1) {
1060 struct page *wpage[1];
1062 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1072 if (unlikely(npages != 1)) {
1073 struct vm_area_struct *vma;
1076 return get_fault_pfn();
1078 down_read(¤t->mm->mmap_sem);
1079 if (is_hwpoison_address(addr)) {
1080 up_read(¤t->mm->mmap_sem);
1081 get_page(hwpoison_page);
1082 return page_to_pfn(hwpoison_page);
1085 vma = find_vma_intersection(current->mm, addr, addr+1);
1088 pfn = get_fault_pfn();
1089 else if ((vma->vm_flags & VM_PFNMAP)) {
1090 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1092 BUG_ON(!kvm_is_mmio_pfn(pfn));
1094 if (async && (vma->vm_flags & VM_WRITE))
1096 pfn = get_fault_pfn();
1098 up_read(¤t->mm->mmap_sem);
1100 pfn = page_to_pfn(page[0]);
1105 pfn_t hva_to_pfn_atomic(struct kvm *kvm, unsigned long addr)
1107 return hva_to_pfn(kvm, addr, true, NULL, true, NULL);
1109 EXPORT_SYMBOL_GPL(hva_to_pfn_atomic);
1111 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1112 bool write_fault, bool *writable)
1119 addr = gfn_to_hva(kvm, gfn);
1120 if (kvm_is_error_hva(addr)) {
1122 return page_to_pfn(bad_page);
1125 return hva_to_pfn(kvm, addr, atomic, async, write_fault, writable);
1128 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1130 return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1132 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1134 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1135 bool write_fault, bool *writable)
1137 return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1139 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1141 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1143 return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1145 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1147 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1150 return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1152 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1154 pfn_t gfn_to_pfn_memslot(struct kvm *kvm,
1155 struct kvm_memory_slot *slot, gfn_t gfn)
1157 unsigned long addr = gfn_to_hva_memslot(slot, gfn);
1158 return hva_to_pfn(kvm, addr, false, NULL, true, NULL);
1161 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1167 addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1168 if (kvm_is_error_hva(addr))
1171 if (entry < nr_pages)
1174 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1176 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1178 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1182 pfn = gfn_to_pfn(kvm, gfn);
1183 if (!kvm_is_mmio_pfn(pfn))
1184 return pfn_to_page(pfn);
1186 WARN_ON(kvm_is_mmio_pfn(pfn));
1192 EXPORT_SYMBOL_GPL(gfn_to_page);
1194 void kvm_release_page_clean(struct page *page)
1196 kvm_release_pfn_clean(page_to_pfn(page));
1198 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1200 void kvm_release_pfn_clean(pfn_t pfn)
1202 if (!kvm_is_mmio_pfn(pfn))
1203 put_page(pfn_to_page(pfn));
1205 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1207 void kvm_release_page_dirty(struct page *page)
1209 kvm_release_pfn_dirty(page_to_pfn(page));
1211 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1213 void kvm_release_pfn_dirty(pfn_t pfn)
1215 kvm_set_pfn_dirty(pfn);
1216 kvm_release_pfn_clean(pfn);
1218 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1220 void kvm_set_page_dirty(struct page *page)
1222 kvm_set_pfn_dirty(page_to_pfn(page));
1224 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1226 void kvm_set_pfn_dirty(pfn_t pfn)
1228 if (!kvm_is_mmio_pfn(pfn)) {
1229 struct page *page = pfn_to_page(pfn);
1230 if (!PageReserved(page))
1234 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1236 void kvm_set_pfn_accessed(pfn_t pfn)
1238 if (!kvm_is_mmio_pfn(pfn))
1239 mark_page_accessed(pfn_to_page(pfn));
1241 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1243 void kvm_get_pfn(pfn_t pfn)
1245 if (!kvm_is_mmio_pfn(pfn))
1246 get_page(pfn_to_page(pfn));
1248 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1250 static int next_segment(unsigned long len, int offset)
1252 if (len > PAGE_SIZE - offset)
1253 return PAGE_SIZE - offset;
1258 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1264 addr = gfn_to_hva(kvm, gfn);
1265 if (kvm_is_error_hva(addr))
1267 r = copy_from_user(data, (void __user *)addr + offset, len);
1272 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1274 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1276 gfn_t gfn = gpa >> PAGE_SHIFT;
1278 int offset = offset_in_page(gpa);
1281 while ((seg = next_segment(len, offset)) != 0) {
1282 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1292 EXPORT_SYMBOL_GPL(kvm_read_guest);
1294 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1299 gfn_t gfn = gpa >> PAGE_SHIFT;
1300 int offset = offset_in_page(gpa);
1302 addr = gfn_to_hva(kvm, gfn);
1303 if (kvm_is_error_hva(addr))
1305 pagefault_disable();
1306 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1312 EXPORT_SYMBOL(kvm_read_guest_atomic);
1314 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1315 int offset, int len)
1320 addr = gfn_to_hva(kvm, gfn);
1321 if (kvm_is_error_hva(addr))
1323 r = copy_to_user((void __user *)addr + offset, data, len);
1326 mark_page_dirty(kvm, gfn);
1329 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1331 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1334 gfn_t gfn = gpa >> PAGE_SHIFT;
1336 int offset = offset_in_page(gpa);
1339 while ((seg = next_segment(len, offset)) != 0) {
1340 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1351 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1354 struct kvm_memslots *slots = kvm_memslots(kvm);
1355 int offset = offset_in_page(gpa);
1356 gfn_t gfn = gpa >> PAGE_SHIFT;
1359 ghc->generation = slots->generation;
1360 ghc->memslot = __gfn_to_memslot(slots, gfn);
1361 ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1362 if (!kvm_is_error_hva(ghc->hva))
1369 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1371 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1372 void *data, unsigned long len)
1374 struct kvm_memslots *slots = kvm_memslots(kvm);
1377 if (slots->generation != ghc->generation)
1378 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1380 if (kvm_is_error_hva(ghc->hva))
1383 r = copy_to_user((void __user *)ghc->hva, data, len);
1386 mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1390 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1392 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1394 return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1397 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1399 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1401 gfn_t gfn = gpa >> PAGE_SHIFT;
1403 int offset = offset_in_page(gpa);
1406 while ((seg = next_segment(len, offset)) != 0) {
1407 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1416 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1418 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1421 if (memslot && memslot->dirty_bitmap) {
1422 unsigned long rel_gfn = gfn - memslot->base_gfn;
1424 generic___set_le_bit(rel_gfn, memslot->dirty_bitmap);
1428 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1430 struct kvm_memory_slot *memslot;
1432 memslot = gfn_to_memslot(kvm, gfn);
1433 mark_page_dirty_in_slot(kvm, memslot, gfn);
1437 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1439 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1444 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1446 if (kvm_arch_vcpu_runnable(vcpu)) {
1447 kvm_make_request(KVM_REQ_UNHALT, vcpu);
1450 if (kvm_cpu_has_pending_timer(vcpu))
1452 if (signal_pending(current))
1458 finish_wait(&vcpu->wq, &wait);
1461 void kvm_resched(struct kvm_vcpu *vcpu)
1463 if (!need_resched())
1467 EXPORT_SYMBOL_GPL(kvm_resched);
1469 void kvm_vcpu_on_spin(struct kvm_vcpu *vcpu)
1474 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1476 /* Sleep for 100 us, and hope lock-holder got scheduled */
1477 expires = ktime_add_ns(ktime_get(), 100000UL);
1478 schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
1480 finish_wait(&vcpu->wq, &wait);
1482 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1484 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1486 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1489 if (vmf->pgoff == 0)
1490 page = virt_to_page(vcpu->run);
1492 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1493 page = virt_to_page(vcpu->arch.pio_data);
1495 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1496 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1497 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1500 return VM_FAULT_SIGBUS;
1506 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1507 .fault = kvm_vcpu_fault,
1510 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1512 vma->vm_ops = &kvm_vcpu_vm_ops;
1516 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1518 struct kvm_vcpu *vcpu = filp->private_data;
1520 kvm_put_kvm(vcpu->kvm);
1524 static struct file_operations kvm_vcpu_fops = {
1525 .release = kvm_vcpu_release,
1526 .unlocked_ioctl = kvm_vcpu_ioctl,
1527 .compat_ioctl = kvm_vcpu_ioctl,
1528 .mmap = kvm_vcpu_mmap,
1529 .llseek = noop_llseek,
1533 * Allocates an inode for the vcpu.
1535 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1537 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1541 * Creates some virtual cpus. Good luck creating more than one.
1543 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1546 struct kvm_vcpu *vcpu, *v;
1548 vcpu = kvm_arch_vcpu_create(kvm, id);
1550 return PTR_ERR(vcpu);
1552 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1554 r = kvm_arch_vcpu_setup(vcpu);
1558 mutex_lock(&kvm->lock);
1559 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1564 kvm_for_each_vcpu(r, v, kvm)
1565 if (v->vcpu_id == id) {
1570 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1572 /* Now it's all set up, let userspace reach it */
1574 r = create_vcpu_fd(vcpu);
1580 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1582 atomic_inc(&kvm->online_vcpus);
1584 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1585 if (kvm->bsp_vcpu_id == id)
1586 kvm->bsp_vcpu = vcpu;
1588 mutex_unlock(&kvm->lock);
1592 mutex_unlock(&kvm->lock);
1593 kvm_arch_vcpu_destroy(vcpu);
1597 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1600 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1601 vcpu->sigset_active = 1;
1602 vcpu->sigset = *sigset;
1604 vcpu->sigset_active = 0;
1608 static long kvm_vcpu_ioctl(struct file *filp,
1609 unsigned int ioctl, unsigned long arg)
1611 struct kvm_vcpu *vcpu = filp->private_data;
1612 void __user *argp = (void __user *)arg;
1614 struct kvm_fpu *fpu = NULL;
1615 struct kvm_sregs *kvm_sregs = NULL;
1617 if (vcpu->kvm->mm != current->mm)
1620 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1622 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1623 * so vcpu_load() would break it.
1625 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1626 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1636 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1637 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1639 case KVM_GET_REGS: {
1640 struct kvm_regs *kvm_regs;
1643 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1646 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1650 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1657 case KVM_SET_REGS: {
1658 struct kvm_regs *kvm_regs;
1661 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1665 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1667 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1675 case KVM_GET_SREGS: {
1676 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1680 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1684 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1689 case KVM_SET_SREGS: {
1690 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1695 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1697 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1703 case KVM_GET_MP_STATE: {
1704 struct kvm_mp_state mp_state;
1706 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1710 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1715 case KVM_SET_MP_STATE: {
1716 struct kvm_mp_state mp_state;
1719 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1721 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1727 case KVM_TRANSLATE: {
1728 struct kvm_translation tr;
1731 if (copy_from_user(&tr, argp, sizeof tr))
1733 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1737 if (copy_to_user(argp, &tr, sizeof tr))
1742 case KVM_SET_GUEST_DEBUG: {
1743 struct kvm_guest_debug dbg;
1746 if (copy_from_user(&dbg, argp, sizeof dbg))
1748 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1754 case KVM_SET_SIGNAL_MASK: {
1755 struct kvm_signal_mask __user *sigmask_arg = argp;
1756 struct kvm_signal_mask kvm_sigmask;
1757 sigset_t sigset, *p;
1762 if (copy_from_user(&kvm_sigmask, argp,
1763 sizeof kvm_sigmask))
1766 if (kvm_sigmask.len != sizeof sigset)
1769 if (copy_from_user(&sigset, sigmask_arg->sigset,
1774 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
1778 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1782 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1786 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1792 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1797 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1799 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1806 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1815 static long kvm_vm_ioctl(struct file *filp,
1816 unsigned int ioctl, unsigned long arg)
1818 struct kvm *kvm = filp->private_data;
1819 void __user *argp = (void __user *)arg;
1822 if (kvm->mm != current->mm)
1825 case KVM_CREATE_VCPU:
1826 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1830 case KVM_SET_USER_MEMORY_REGION: {
1831 struct kvm_userspace_memory_region kvm_userspace_mem;
1834 if (copy_from_user(&kvm_userspace_mem, argp,
1835 sizeof kvm_userspace_mem))
1838 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1843 case KVM_GET_DIRTY_LOG: {
1844 struct kvm_dirty_log log;
1847 if (copy_from_user(&log, argp, sizeof log))
1849 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1854 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1855 case KVM_REGISTER_COALESCED_MMIO: {
1856 struct kvm_coalesced_mmio_zone zone;
1858 if (copy_from_user(&zone, argp, sizeof zone))
1860 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1866 case KVM_UNREGISTER_COALESCED_MMIO: {
1867 struct kvm_coalesced_mmio_zone zone;
1869 if (copy_from_user(&zone, argp, sizeof zone))
1871 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1879 struct kvm_irqfd data;
1882 if (copy_from_user(&data, argp, sizeof data))
1884 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
1887 case KVM_IOEVENTFD: {
1888 struct kvm_ioeventfd data;
1891 if (copy_from_user(&data, argp, sizeof data))
1893 r = kvm_ioeventfd(kvm, &data);
1896 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1897 case KVM_SET_BOOT_CPU_ID:
1899 mutex_lock(&kvm->lock);
1900 if (atomic_read(&kvm->online_vcpus) != 0)
1903 kvm->bsp_vcpu_id = arg;
1904 mutex_unlock(&kvm->lock);
1908 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1910 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
1916 #ifdef CONFIG_COMPAT
1917 struct compat_kvm_dirty_log {
1921 compat_uptr_t dirty_bitmap; /* one bit per page */
1926 static long kvm_vm_compat_ioctl(struct file *filp,
1927 unsigned int ioctl, unsigned long arg)
1929 struct kvm *kvm = filp->private_data;
1932 if (kvm->mm != current->mm)
1935 case KVM_GET_DIRTY_LOG: {
1936 struct compat_kvm_dirty_log compat_log;
1937 struct kvm_dirty_log log;
1940 if (copy_from_user(&compat_log, (void __user *)arg,
1941 sizeof(compat_log)))
1943 log.slot = compat_log.slot;
1944 log.padding1 = compat_log.padding1;
1945 log.padding2 = compat_log.padding2;
1946 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
1948 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1954 r = kvm_vm_ioctl(filp, ioctl, arg);
1962 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1964 struct page *page[1];
1967 gfn_t gfn = vmf->pgoff;
1968 struct kvm *kvm = vma->vm_file->private_data;
1970 addr = gfn_to_hva(kvm, gfn);
1971 if (kvm_is_error_hva(addr))
1972 return VM_FAULT_SIGBUS;
1974 npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
1976 if (unlikely(npages != 1))
1977 return VM_FAULT_SIGBUS;
1979 vmf->page = page[0];
1983 static const struct vm_operations_struct kvm_vm_vm_ops = {
1984 .fault = kvm_vm_fault,
1987 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1989 vma->vm_ops = &kvm_vm_vm_ops;
1993 static struct file_operations kvm_vm_fops = {
1994 .release = kvm_vm_release,
1995 .unlocked_ioctl = kvm_vm_ioctl,
1996 #ifdef CONFIG_COMPAT
1997 .compat_ioctl = kvm_vm_compat_ioctl,
1999 .mmap = kvm_vm_mmap,
2000 .llseek = noop_llseek,
2003 static int kvm_dev_ioctl_create_vm(void)
2008 kvm = kvm_create_vm();
2010 return PTR_ERR(kvm);
2011 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2012 r = kvm_coalesced_mmio_init(kvm);
2018 r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2025 static long kvm_dev_ioctl_check_extension_generic(long arg)
2028 case KVM_CAP_USER_MEMORY:
2029 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2030 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2031 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2032 case KVM_CAP_SET_BOOT_CPU_ID:
2034 case KVM_CAP_INTERNAL_ERROR_DATA:
2036 #ifdef CONFIG_HAVE_KVM_IRQCHIP
2037 case KVM_CAP_IRQ_ROUTING:
2038 return KVM_MAX_IRQ_ROUTES;
2043 return kvm_dev_ioctl_check_extension(arg);
2046 static long kvm_dev_ioctl(struct file *filp,
2047 unsigned int ioctl, unsigned long arg)
2052 case KVM_GET_API_VERSION:
2056 r = KVM_API_VERSION;
2062 r = kvm_dev_ioctl_create_vm();
2064 case KVM_CHECK_EXTENSION:
2065 r = kvm_dev_ioctl_check_extension_generic(arg);
2067 case KVM_GET_VCPU_MMAP_SIZE:
2071 r = PAGE_SIZE; /* struct kvm_run */
2073 r += PAGE_SIZE; /* pio data page */
2075 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2076 r += PAGE_SIZE; /* coalesced mmio ring page */
2079 case KVM_TRACE_ENABLE:
2080 case KVM_TRACE_PAUSE:
2081 case KVM_TRACE_DISABLE:
2085 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2091 static struct file_operations kvm_chardev_ops = {
2092 .unlocked_ioctl = kvm_dev_ioctl,
2093 .compat_ioctl = kvm_dev_ioctl,
2094 .llseek = noop_llseek,
2097 static struct miscdevice kvm_dev = {
2103 static void hardware_enable_nolock(void *junk)
2105 int cpu = raw_smp_processor_id();
2108 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2111 cpumask_set_cpu(cpu, cpus_hardware_enabled);
2113 r = kvm_arch_hardware_enable(NULL);
2116 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2117 atomic_inc(&hardware_enable_failed);
2118 printk(KERN_INFO "kvm: enabling virtualization on "
2119 "CPU%d failed\n", cpu);
2123 static void hardware_enable(void *junk)
2125 spin_lock(&kvm_lock);
2126 hardware_enable_nolock(junk);
2127 spin_unlock(&kvm_lock);
2130 static void hardware_disable_nolock(void *junk)
2132 int cpu = raw_smp_processor_id();
2134 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2136 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2137 kvm_arch_hardware_disable(NULL);
2140 static void hardware_disable(void *junk)
2142 spin_lock(&kvm_lock);
2143 hardware_disable_nolock(junk);
2144 spin_unlock(&kvm_lock);
2147 static void hardware_disable_all_nolock(void)
2149 BUG_ON(!kvm_usage_count);
2152 if (!kvm_usage_count)
2153 on_each_cpu(hardware_disable_nolock, NULL, 1);
2156 static void hardware_disable_all(void)
2158 spin_lock(&kvm_lock);
2159 hardware_disable_all_nolock();
2160 spin_unlock(&kvm_lock);
2163 static int hardware_enable_all(void)
2167 spin_lock(&kvm_lock);
2170 if (kvm_usage_count == 1) {
2171 atomic_set(&hardware_enable_failed, 0);
2172 on_each_cpu(hardware_enable_nolock, NULL, 1);
2174 if (atomic_read(&hardware_enable_failed)) {
2175 hardware_disable_all_nolock();
2180 spin_unlock(&kvm_lock);
2185 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2190 if (!kvm_usage_count)
2193 val &= ~CPU_TASKS_FROZEN;
2196 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2198 hardware_disable(NULL);
2201 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2203 hardware_enable(NULL);
2210 asmlinkage void kvm_spurious_fault(void)
2212 /* Fault while not rebooting. We want the trace. */
2215 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
2217 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2221 * Some (well, at least mine) BIOSes hang on reboot if
2224 * And Intel TXT required VMX off for all cpu when system shutdown.
2226 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2227 kvm_rebooting = true;
2228 on_each_cpu(hardware_disable_nolock, NULL, 1);
2232 static struct notifier_block kvm_reboot_notifier = {
2233 .notifier_call = kvm_reboot,
2237 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2241 for (i = 0; i < bus->dev_count; i++) {
2242 struct kvm_io_device *pos = bus->devs[i];
2244 kvm_iodevice_destructor(pos);
2249 /* kvm_io_bus_write - called under kvm->slots_lock */
2250 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2251 int len, const void *val)
2254 struct kvm_io_bus *bus;
2256 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2257 for (i = 0; i < bus->dev_count; i++)
2258 if (!kvm_iodevice_write(bus->devs[i], addr, len, val))
2263 /* kvm_io_bus_read - called under kvm->slots_lock */
2264 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2268 struct kvm_io_bus *bus;
2270 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2271 for (i = 0; i < bus->dev_count; i++)
2272 if (!kvm_iodevice_read(bus->devs[i], addr, len, val))
2277 /* Caller must hold slots_lock. */
2278 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2279 struct kvm_io_device *dev)
2281 struct kvm_io_bus *new_bus, *bus;
2283 bus = kvm->buses[bus_idx];
2284 if (bus->dev_count > NR_IOBUS_DEVS-1)
2287 new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2290 memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2291 new_bus->devs[new_bus->dev_count++] = dev;
2292 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2293 synchronize_srcu_expedited(&kvm->srcu);
2299 /* Caller must hold slots_lock. */
2300 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2301 struct kvm_io_device *dev)
2304 struct kvm_io_bus *new_bus, *bus;
2306 new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2310 bus = kvm->buses[bus_idx];
2311 memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2314 for (i = 0; i < new_bus->dev_count; i++)
2315 if (new_bus->devs[i] == dev) {
2317 new_bus->devs[i] = new_bus->devs[--new_bus->dev_count];
2326 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2327 synchronize_srcu_expedited(&kvm->srcu);
2332 static struct notifier_block kvm_cpu_notifier = {
2333 .notifier_call = kvm_cpu_hotplug,
2336 static int vm_stat_get(void *_offset, u64 *val)
2338 unsigned offset = (long)_offset;
2342 spin_lock(&kvm_lock);
2343 list_for_each_entry(kvm, &vm_list, vm_list)
2344 *val += *(u32 *)((void *)kvm + offset);
2345 spin_unlock(&kvm_lock);
2349 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2351 static int vcpu_stat_get(void *_offset, u64 *val)
2353 unsigned offset = (long)_offset;
2355 struct kvm_vcpu *vcpu;
2359 spin_lock(&kvm_lock);
2360 list_for_each_entry(kvm, &vm_list, vm_list)
2361 kvm_for_each_vcpu(i, vcpu, kvm)
2362 *val += *(u32 *)((void *)vcpu + offset);
2364 spin_unlock(&kvm_lock);
2368 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2370 static const struct file_operations *stat_fops[] = {
2371 [KVM_STAT_VCPU] = &vcpu_stat_fops,
2372 [KVM_STAT_VM] = &vm_stat_fops,
2375 static void kvm_init_debug(void)
2377 struct kvm_stats_debugfs_item *p;
2379 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2380 for (p = debugfs_entries; p->name; ++p)
2381 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2382 (void *)(long)p->offset,
2383 stat_fops[p->kind]);
2386 static void kvm_exit_debug(void)
2388 struct kvm_stats_debugfs_item *p;
2390 for (p = debugfs_entries; p->name; ++p)
2391 debugfs_remove(p->dentry);
2392 debugfs_remove(kvm_debugfs_dir);
2395 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2397 if (kvm_usage_count)
2398 hardware_disable_nolock(NULL);
2402 static int kvm_resume(struct sys_device *dev)
2404 if (kvm_usage_count) {
2405 WARN_ON(spin_is_locked(&kvm_lock));
2406 hardware_enable_nolock(NULL);
2411 static struct sysdev_class kvm_sysdev_class = {
2413 .suspend = kvm_suspend,
2414 .resume = kvm_resume,
2417 static struct sys_device kvm_sysdev = {
2419 .cls = &kvm_sysdev_class,
2422 struct page *bad_page;
2426 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2428 return container_of(pn, struct kvm_vcpu, preempt_notifier);
2431 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2433 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2435 kvm_arch_vcpu_load(vcpu, cpu);
2438 static void kvm_sched_out(struct preempt_notifier *pn,
2439 struct task_struct *next)
2441 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2443 kvm_arch_vcpu_put(vcpu);
2446 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2447 struct module *module)
2452 r = kvm_arch_init(opaque);
2456 bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2458 if (bad_page == NULL) {
2463 bad_pfn = page_to_pfn(bad_page);
2465 hwpoison_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2467 if (hwpoison_page == NULL) {
2472 hwpoison_pfn = page_to_pfn(hwpoison_page);
2474 fault_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2476 if (fault_page == NULL) {
2481 fault_pfn = page_to_pfn(fault_page);
2483 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2488 r = kvm_arch_hardware_setup();
2492 for_each_online_cpu(cpu) {
2493 smp_call_function_single(cpu,
2494 kvm_arch_check_processor_compat,
2500 r = register_cpu_notifier(&kvm_cpu_notifier);
2503 register_reboot_notifier(&kvm_reboot_notifier);
2505 r = sysdev_class_register(&kvm_sysdev_class);
2509 r = sysdev_register(&kvm_sysdev);
2513 /* A kmem cache lets us meet the alignment requirements of fx_save. */
2515 vcpu_align = __alignof__(struct kvm_vcpu);
2516 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2518 if (!kvm_vcpu_cache) {
2523 r = kvm_async_pf_init();
2527 kvm_chardev_ops.owner = module;
2528 kvm_vm_fops.owner = module;
2529 kvm_vcpu_fops.owner = module;
2531 r = misc_register(&kvm_dev);
2533 printk(KERN_ERR "kvm: misc device register failed\n");
2537 kvm_preempt_ops.sched_in = kvm_sched_in;
2538 kvm_preempt_ops.sched_out = kvm_sched_out;
2545 kvm_async_pf_deinit();
2547 kmem_cache_destroy(kvm_vcpu_cache);
2549 sysdev_unregister(&kvm_sysdev);
2551 sysdev_class_unregister(&kvm_sysdev_class);
2553 unregister_reboot_notifier(&kvm_reboot_notifier);
2554 unregister_cpu_notifier(&kvm_cpu_notifier);
2557 kvm_arch_hardware_unsetup();
2559 free_cpumask_var(cpus_hardware_enabled);
2562 __free_page(fault_page);
2564 __free_page(hwpoison_page);
2565 __free_page(bad_page);
2571 EXPORT_SYMBOL_GPL(kvm_init);
2576 misc_deregister(&kvm_dev);
2577 kmem_cache_destroy(kvm_vcpu_cache);
2578 kvm_async_pf_deinit();
2579 sysdev_unregister(&kvm_sysdev);
2580 sysdev_class_unregister(&kvm_sysdev_class);
2581 unregister_reboot_notifier(&kvm_reboot_notifier);
2582 unregister_cpu_notifier(&kvm_cpu_notifier);
2583 on_each_cpu(hardware_disable_nolock, NULL, 1);
2584 kvm_arch_hardware_unsetup();
2586 free_cpumask_var(cpus_hardware_enabled);
2587 __free_page(hwpoison_page);
2588 __free_page(bad_page);
2590 EXPORT_SYMBOL_GPL(kvm_exit);