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);
93 static bool 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)) {
106 struct page *page = compound_head(pfn_to_page(pfn));
107 return PageReserved(page);
114 * Switches to specified vcpu, until a matching vcpu_put()
116 void vcpu_load(struct kvm_vcpu *vcpu)
120 mutex_lock(&vcpu->mutex);
122 preempt_notifier_register(&vcpu->preempt_notifier);
123 kvm_arch_vcpu_load(vcpu, cpu);
127 void vcpu_put(struct kvm_vcpu *vcpu)
130 kvm_arch_vcpu_put(vcpu);
131 preempt_notifier_unregister(&vcpu->preempt_notifier);
133 mutex_unlock(&vcpu->mutex);
136 static void ack_flush(void *_completed)
140 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
145 struct kvm_vcpu *vcpu;
147 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
149 raw_spin_lock(&kvm->requests_lock);
150 me = smp_processor_id();
151 kvm_for_each_vcpu(i, vcpu, kvm) {
152 if (kvm_make_check_request(req, vcpu))
155 if (cpus != NULL && cpu != -1 && cpu != me)
156 cpumask_set_cpu(cpu, cpus);
158 if (unlikely(cpus == NULL))
159 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
160 else if (!cpumask_empty(cpus))
161 smp_call_function_many(cpus, ack_flush, NULL, 1);
164 raw_spin_unlock(&kvm->requests_lock);
165 free_cpumask_var(cpus);
169 void kvm_flush_remote_tlbs(struct kvm *kvm)
171 if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
172 ++kvm->stat.remote_tlb_flush;
175 void kvm_reload_remote_mmus(struct kvm *kvm)
177 make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
180 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
185 mutex_init(&vcpu->mutex);
189 init_waitqueue_head(&vcpu->wq);
190 kvm_async_pf_vcpu_init(vcpu);
192 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
197 vcpu->run = page_address(page);
199 r = kvm_arch_vcpu_init(vcpu);
205 free_page((unsigned long)vcpu->run);
209 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
211 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
213 kvm_arch_vcpu_uninit(vcpu);
214 free_page((unsigned long)vcpu->run);
216 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
218 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
219 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
221 return container_of(mn, struct kvm, mmu_notifier);
224 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
225 struct mm_struct *mm,
226 unsigned long address)
228 struct kvm *kvm = mmu_notifier_to_kvm(mn);
229 int need_tlb_flush, idx;
232 * When ->invalidate_page runs, the linux pte has been zapped
233 * already but the page is still allocated until
234 * ->invalidate_page returns. So if we increase the sequence
235 * here the kvm page fault will notice if the spte can't be
236 * established because the page is going to be freed. If
237 * instead the kvm page fault establishes the spte before
238 * ->invalidate_page runs, kvm_unmap_hva will release it
241 * The sequence increase only need to be seen at spin_unlock
242 * time, and not at spin_lock time.
244 * Increasing the sequence after the spin_unlock would be
245 * unsafe because the kvm page fault could then establish the
246 * pte after kvm_unmap_hva returned, without noticing the page
247 * is going to be freed.
249 idx = srcu_read_lock(&kvm->srcu);
250 spin_lock(&kvm->mmu_lock);
251 kvm->mmu_notifier_seq++;
252 need_tlb_flush = kvm_unmap_hva(kvm, address);
253 spin_unlock(&kvm->mmu_lock);
254 srcu_read_unlock(&kvm->srcu, idx);
256 /* we've to flush the tlb before the pages can be freed */
258 kvm_flush_remote_tlbs(kvm);
262 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
263 struct mm_struct *mm,
264 unsigned long address,
267 struct kvm *kvm = mmu_notifier_to_kvm(mn);
270 idx = srcu_read_lock(&kvm->srcu);
271 spin_lock(&kvm->mmu_lock);
272 kvm->mmu_notifier_seq++;
273 kvm_set_spte_hva(kvm, address, pte);
274 spin_unlock(&kvm->mmu_lock);
275 srcu_read_unlock(&kvm->srcu, idx);
278 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
279 struct mm_struct *mm,
283 struct kvm *kvm = mmu_notifier_to_kvm(mn);
284 int need_tlb_flush = 0, idx;
286 idx = srcu_read_lock(&kvm->srcu);
287 spin_lock(&kvm->mmu_lock);
289 * The count increase must become visible at unlock time as no
290 * spte can be established without taking the mmu_lock and
291 * count is also read inside the mmu_lock critical section.
293 kvm->mmu_notifier_count++;
294 for (; start < end; start += PAGE_SIZE)
295 need_tlb_flush |= kvm_unmap_hva(kvm, start);
296 spin_unlock(&kvm->mmu_lock);
297 srcu_read_unlock(&kvm->srcu, idx);
299 /* we've to flush the tlb before the pages can be freed */
301 kvm_flush_remote_tlbs(kvm);
304 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
305 struct mm_struct *mm,
309 struct kvm *kvm = mmu_notifier_to_kvm(mn);
311 spin_lock(&kvm->mmu_lock);
313 * This sequence increase will notify the kvm page fault that
314 * the page that is going to be mapped in the spte could have
317 kvm->mmu_notifier_seq++;
319 * The above sequence increase must be visible before the
320 * below count decrease but both values are read by the kvm
321 * page fault under mmu_lock spinlock so we don't need to add
322 * a smb_wmb() here in between the two.
324 kvm->mmu_notifier_count--;
325 spin_unlock(&kvm->mmu_lock);
327 BUG_ON(kvm->mmu_notifier_count < 0);
330 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
331 struct mm_struct *mm,
332 unsigned long address)
334 struct kvm *kvm = mmu_notifier_to_kvm(mn);
337 idx = srcu_read_lock(&kvm->srcu);
338 spin_lock(&kvm->mmu_lock);
339 young = kvm_age_hva(kvm, address);
340 spin_unlock(&kvm->mmu_lock);
341 srcu_read_unlock(&kvm->srcu, idx);
344 kvm_flush_remote_tlbs(kvm);
349 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
350 struct mm_struct *mm)
352 struct kvm *kvm = mmu_notifier_to_kvm(mn);
355 idx = srcu_read_lock(&kvm->srcu);
356 kvm_arch_flush_shadow(kvm);
357 srcu_read_unlock(&kvm->srcu, idx);
360 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
361 .invalidate_page = kvm_mmu_notifier_invalidate_page,
362 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
363 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
364 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
365 .change_pte = kvm_mmu_notifier_change_pte,
366 .release = kvm_mmu_notifier_release,
369 static int kvm_init_mmu_notifier(struct kvm *kvm)
371 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
372 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
375 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
377 static int kvm_init_mmu_notifier(struct kvm *kvm)
382 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
384 static struct kvm *kvm_create_vm(void)
387 struct kvm *kvm = kvm_arch_create_vm();
392 r = hardware_enable_all();
394 goto out_err_nodisable;
396 #ifdef CONFIG_HAVE_KVM_IRQCHIP
397 INIT_HLIST_HEAD(&kvm->mask_notifier_list);
398 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
402 kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
405 if (init_srcu_struct(&kvm->srcu))
407 for (i = 0; i < KVM_NR_BUSES; i++) {
408 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
410 if (!kvm->buses[i]) {
411 cleanup_srcu_struct(&kvm->srcu);
416 r = kvm_init_mmu_notifier(kvm);
418 cleanup_srcu_struct(&kvm->srcu);
422 kvm->mm = current->mm;
423 atomic_inc(&kvm->mm->mm_count);
424 spin_lock_init(&kvm->mmu_lock);
425 raw_spin_lock_init(&kvm->requests_lock);
426 kvm_eventfd_init(kvm);
427 mutex_init(&kvm->lock);
428 mutex_init(&kvm->irq_lock);
429 mutex_init(&kvm->slots_lock);
430 atomic_set(&kvm->users_count, 1);
431 spin_lock(&kvm_lock);
432 list_add(&kvm->vm_list, &vm_list);
433 spin_unlock(&kvm_lock);
438 hardware_disable_all();
440 for (i = 0; i < KVM_NR_BUSES; i++)
441 kfree(kvm->buses[i]);
442 kfree(kvm->memslots);
447 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
449 if (!memslot->dirty_bitmap)
452 if (2 * kvm_dirty_bitmap_bytes(memslot) > PAGE_SIZE)
453 vfree(memslot->dirty_bitmap_head);
455 kfree(memslot->dirty_bitmap_head);
457 memslot->dirty_bitmap = NULL;
458 memslot->dirty_bitmap_head = NULL;
462 * Free any memory in @free but not in @dont.
464 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
465 struct kvm_memory_slot *dont)
469 if (!dont || free->rmap != dont->rmap)
472 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
473 kvm_destroy_dirty_bitmap(free);
476 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
477 if (!dont || free->lpage_info[i] != dont->lpage_info[i]) {
478 vfree(free->lpage_info[i]);
479 free->lpage_info[i] = NULL;
487 void kvm_free_physmem(struct kvm *kvm)
490 struct kvm_memslots *slots = kvm->memslots;
492 for (i = 0; i < slots->nmemslots; ++i)
493 kvm_free_physmem_slot(&slots->memslots[i], NULL);
495 kfree(kvm->memslots);
498 static void kvm_destroy_vm(struct kvm *kvm)
501 struct mm_struct *mm = kvm->mm;
503 kvm_arch_sync_events(kvm);
504 spin_lock(&kvm_lock);
505 list_del(&kvm->vm_list);
506 spin_unlock(&kvm_lock);
507 kvm_free_irq_routing(kvm);
508 for (i = 0; i < KVM_NR_BUSES; i++)
509 kvm_io_bus_destroy(kvm->buses[i]);
510 kvm_coalesced_mmio_free(kvm);
511 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
512 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
514 kvm_arch_flush_shadow(kvm);
516 kvm_arch_destroy_vm(kvm);
517 hardware_disable_all();
521 void kvm_get_kvm(struct kvm *kvm)
523 atomic_inc(&kvm->users_count);
525 EXPORT_SYMBOL_GPL(kvm_get_kvm);
527 void kvm_put_kvm(struct kvm *kvm)
529 if (atomic_dec_and_test(&kvm->users_count))
532 EXPORT_SYMBOL_GPL(kvm_put_kvm);
535 static int kvm_vm_release(struct inode *inode, struct file *filp)
537 struct kvm *kvm = filp->private_data;
539 kvm_irqfd_release(kvm);
546 * Allocation size is twice as large as the actual dirty bitmap size.
547 * This makes it possible to do double buffering: see x86's
548 * kvm_vm_ioctl_get_dirty_log().
550 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
552 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
554 if (dirty_bytes > PAGE_SIZE)
555 memslot->dirty_bitmap = vzalloc(dirty_bytes);
557 memslot->dirty_bitmap = kzalloc(dirty_bytes, GFP_KERNEL);
559 if (!memslot->dirty_bitmap)
562 memslot->dirty_bitmap_head = memslot->dirty_bitmap;
567 * Allocate some memory and give it an address in the guest physical address
570 * Discontiguous memory is allowed, mostly for framebuffers.
572 * Must be called holding mmap_sem for write.
574 int __kvm_set_memory_region(struct kvm *kvm,
575 struct kvm_userspace_memory_region *mem,
578 int r, flush_shadow = 0;
580 unsigned long npages;
582 struct kvm_memory_slot *memslot;
583 struct kvm_memory_slot old, new;
584 struct kvm_memslots *slots, *old_memslots;
587 /* General sanity checks */
588 if (mem->memory_size & (PAGE_SIZE - 1))
590 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
592 if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
594 if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
596 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
599 memslot = &kvm->memslots->memslots[mem->slot];
600 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
601 npages = mem->memory_size >> PAGE_SHIFT;
604 if (npages > KVM_MEM_MAX_NR_PAGES)
608 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
610 new = old = *memslot;
613 new.base_gfn = base_gfn;
615 new.flags = mem->flags;
617 /* Disallow changing a memory slot's size. */
619 if (npages && old.npages && npages != old.npages)
622 /* Check for overlaps */
624 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
625 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
627 if (s == memslot || !s->npages)
629 if (!((base_gfn + npages <= s->base_gfn) ||
630 (base_gfn >= s->base_gfn + s->npages)))
634 /* Free page dirty bitmap if unneeded */
635 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
636 new.dirty_bitmap = NULL;
640 /* Allocate if a slot is being created */
642 if (npages && !new.rmap) {
643 new.rmap = vzalloc(npages * sizeof(*new.rmap));
648 new.user_alloc = user_alloc;
649 new.userspace_addr = mem->userspace_addr;
654 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
660 /* Avoid unused variable warning if no large pages */
663 if (new.lpage_info[i])
666 lpages = 1 + ((base_gfn + npages - 1)
667 >> KVM_HPAGE_GFN_SHIFT(level));
668 lpages -= base_gfn >> KVM_HPAGE_GFN_SHIFT(level);
670 new.lpage_info[i] = vzalloc(lpages * sizeof(*new.lpage_info[i]));
672 if (!new.lpage_info[i])
675 if (base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
676 new.lpage_info[i][0].write_count = 1;
677 if ((base_gfn+npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
678 new.lpage_info[i][lpages - 1].write_count = 1;
679 ugfn = new.userspace_addr >> PAGE_SHIFT;
681 * If the gfn and userspace address are not aligned wrt each
682 * other, or if explicitly asked to, disable large page
683 * support for this slot
685 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
687 for (j = 0; j < lpages; ++j)
688 new.lpage_info[i][j].write_count = 1;
693 /* Allocate page dirty bitmap if needed */
694 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
695 if (kvm_create_dirty_bitmap(&new) < 0)
697 /* destroy any largepage mappings for dirty tracking */
701 #else /* not defined CONFIG_S390 */
702 new.user_alloc = user_alloc;
704 new.userspace_addr = mem->userspace_addr;
705 #endif /* not defined CONFIG_S390 */
709 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
712 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
713 if (mem->slot >= slots->nmemslots)
714 slots->nmemslots = mem->slot + 1;
716 slots->memslots[mem->slot].flags |= KVM_MEMSLOT_INVALID;
718 old_memslots = kvm->memslots;
719 rcu_assign_pointer(kvm->memslots, slots);
720 synchronize_srcu_expedited(&kvm->srcu);
721 /* From this point no new shadow pages pointing to a deleted
722 * memslot will be created.
724 * validation of sp->gfn happens in:
725 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
726 * - kvm_is_visible_gfn (mmu_check_roots)
728 kvm_arch_flush_shadow(kvm);
732 r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
736 /* map the pages in iommu page table */
738 r = kvm_iommu_map_pages(kvm, &new);
744 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
747 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
748 if (mem->slot >= slots->nmemslots)
749 slots->nmemslots = mem->slot + 1;
752 /* actual memory is freed via old in kvm_free_physmem_slot below */
755 new.dirty_bitmap = NULL;
756 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i)
757 new.lpage_info[i] = NULL;
760 slots->memslots[mem->slot] = new;
761 old_memslots = kvm->memslots;
762 rcu_assign_pointer(kvm->memslots, slots);
763 synchronize_srcu_expedited(&kvm->srcu);
765 kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
767 kvm_free_physmem_slot(&old, &new);
771 kvm_arch_flush_shadow(kvm);
776 kvm_free_physmem_slot(&new, &old);
781 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
783 int kvm_set_memory_region(struct kvm *kvm,
784 struct kvm_userspace_memory_region *mem,
789 mutex_lock(&kvm->slots_lock);
790 r = __kvm_set_memory_region(kvm, mem, user_alloc);
791 mutex_unlock(&kvm->slots_lock);
794 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
796 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
798 kvm_userspace_memory_region *mem,
801 if (mem->slot >= KVM_MEMORY_SLOTS)
803 return kvm_set_memory_region(kvm, mem, user_alloc);
806 int kvm_get_dirty_log(struct kvm *kvm,
807 struct kvm_dirty_log *log, int *is_dirty)
809 struct kvm_memory_slot *memslot;
812 unsigned long any = 0;
815 if (log->slot >= KVM_MEMORY_SLOTS)
818 memslot = &kvm->memslots->memslots[log->slot];
820 if (!memslot->dirty_bitmap)
823 n = kvm_dirty_bitmap_bytes(memslot);
825 for (i = 0; !any && i < n/sizeof(long); ++i)
826 any = memslot->dirty_bitmap[i];
829 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
840 void kvm_disable_largepages(void)
842 largepages_enabled = false;
844 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
846 int is_error_page(struct page *page)
848 return page == bad_page || page == hwpoison_page || page == fault_page;
850 EXPORT_SYMBOL_GPL(is_error_page);
852 int is_error_pfn(pfn_t pfn)
854 return pfn == bad_pfn || pfn == hwpoison_pfn || pfn == fault_pfn;
856 EXPORT_SYMBOL_GPL(is_error_pfn);
858 int is_hwpoison_pfn(pfn_t pfn)
860 return pfn == hwpoison_pfn;
862 EXPORT_SYMBOL_GPL(is_hwpoison_pfn);
864 int is_fault_pfn(pfn_t pfn)
866 return pfn == fault_pfn;
868 EXPORT_SYMBOL_GPL(is_fault_pfn);
870 static inline unsigned long bad_hva(void)
875 int kvm_is_error_hva(unsigned long addr)
877 return addr == bad_hva();
879 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
881 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm_memslots *slots,
886 for (i = 0; i < slots->nmemslots; ++i) {
887 struct kvm_memory_slot *memslot = &slots->memslots[i];
889 if (gfn >= memslot->base_gfn
890 && gfn < memslot->base_gfn + memslot->npages)
896 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
898 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
900 EXPORT_SYMBOL_GPL(gfn_to_memslot);
902 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
905 struct kvm_memslots *slots = kvm_memslots(kvm);
907 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
908 struct kvm_memory_slot *memslot = &slots->memslots[i];
910 if (memslot->flags & KVM_MEMSLOT_INVALID)
913 if (gfn >= memslot->base_gfn
914 && gfn < memslot->base_gfn + memslot->npages)
919 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
921 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
923 struct vm_area_struct *vma;
924 unsigned long addr, size;
928 addr = gfn_to_hva(kvm, gfn);
929 if (kvm_is_error_hva(addr))
932 down_read(¤t->mm->mmap_sem);
933 vma = find_vma(current->mm, addr);
937 size = vma_kernel_pagesize(vma);
940 up_read(¤t->mm->mmap_sem);
945 int memslot_id(struct kvm *kvm, gfn_t gfn)
948 struct kvm_memslots *slots = kvm_memslots(kvm);
949 struct kvm_memory_slot *memslot = NULL;
951 for (i = 0; i < slots->nmemslots; ++i) {
952 memslot = &slots->memslots[i];
954 if (gfn >= memslot->base_gfn
955 && gfn < memslot->base_gfn + memslot->npages)
959 return memslot - slots->memslots;
962 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
965 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
969 *nr_pages = slot->npages - (gfn - slot->base_gfn);
971 return gfn_to_hva_memslot(slot, gfn);
974 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
976 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
978 EXPORT_SYMBOL_GPL(gfn_to_hva);
980 static pfn_t get_fault_pfn(void)
982 get_page(fault_page);
986 static pfn_t hva_to_pfn(struct kvm *kvm, unsigned long addr, bool atomic,
987 bool *async, bool write_fault, bool *writable)
989 struct page *page[1];
993 /* we can do it either atomically or asynchronously, not both */
994 BUG_ON(atomic && async);
996 BUG_ON(!write_fault && !writable);
1001 if (atomic || async)
1002 npages = __get_user_pages_fast(addr, 1, 1, page);
1004 if (unlikely(npages != 1) && !atomic) {
1008 *writable = write_fault;
1010 npages = get_user_pages_fast(addr, 1, write_fault, page);
1012 /* map read fault as writable if possible */
1013 if (unlikely(!write_fault) && npages == 1) {
1014 struct page *wpage[1];
1016 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1026 if (unlikely(npages != 1)) {
1027 struct vm_area_struct *vma;
1030 return get_fault_pfn();
1032 down_read(¤t->mm->mmap_sem);
1033 if (is_hwpoison_address(addr)) {
1034 up_read(¤t->mm->mmap_sem);
1035 get_page(hwpoison_page);
1036 return page_to_pfn(hwpoison_page);
1039 vma = find_vma_intersection(current->mm, addr, addr+1);
1042 pfn = get_fault_pfn();
1043 else if ((vma->vm_flags & VM_PFNMAP)) {
1044 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1046 BUG_ON(!kvm_is_mmio_pfn(pfn));
1048 if (async && (vma->vm_flags & VM_WRITE))
1050 pfn = get_fault_pfn();
1052 up_read(¤t->mm->mmap_sem);
1054 pfn = page_to_pfn(page[0]);
1059 pfn_t hva_to_pfn_atomic(struct kvm *kvm, unsigned long addr)
1061 return hva_to_pfn(kvm, addr, true, NULL, true, NULL);
1063 EXPORT_SYMBOL_GPL(hva_to_pfn_atomic);
1065 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1066 bool write_fault, bool *writable)
1073 addr = gfn_to_hva(kvm, gfn);
1074 if (kvm_is_error_hva(addr)) {
1076 return page_to_pfn(bad_page);
1079 return hva_to_pfn(kvm, addr, atomic, async, write_fault, writable);
1082 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1084 return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1086 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1088 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1089 bool write_fault, bool *writable)
1091 return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1093 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1095 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1097 return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1099 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1101 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1104 return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1106 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1108 pfn_t gfn_to_pfn_memslot(struct kvm *kvm,
1109 struct kvm_memory_slot *slot, gfn_t gfn)
1111 unsigned long addr = gfn_to_hva_memslot(slot, gfn);
1112 return hva_to_pfn(kvm, addr, false, NULL, true, NULL);
1115 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1121 addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1122 if (kvm_is_error_hva(addr))
1125 if (entry < nr_pages)
1128 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1130 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1132 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1136 pfn = gfn_to_pfn(kvm, gfn);
1137 if (!kvm_is_mmio_pfn(pfn))
1138 return pfn_to_page(pfn);
1140 WARN_ON(kvm_is_mmio_pfn(pfn));
1146 EXPORT_SYMBOL_GPL(gfn_to_page);
1148 void kvm_release_page_clean(struct page *page)
1150 kvm_release_pfn_clean(page_to_pfn(page));
1152 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1154 void kvm_release_pfn_clean(pfn_t pfn)
1156 if (!kvm_is_mmio_pfn(pfn))
1157 put_page(pfn_to_page(pfn));
1159 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1161 void kvm_release_page_dirty(struct page *page)
1163 kvm_release_pfn_dirty(page_to_pfn(page));
1165 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1167 void kvm_release_pfn_dirty(pfn_t pfn)
1169 kvm_set_pfn_dirty(pfn);
1170 kvm_release_pfn_clean(pfn);
1172 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1174 void kvm_set_page_dirty(struct page *page)
1176 kvm_set_pfn_dirty(page_to_pfn(page));
1178 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1180 void kvm_set_pfn_dirty(pfn_t pfn)
1182 if (!kvm_is_mmio_pfn(pfn)) {
1183 struct page *page = pfn_to_page(pfn);
1184 if (!PageReserved(page))
1188 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1190 void kvm_set_pfn_accessed(pfn_t pfn)
1192 if (!kvm_is_mmio_pfn(pfn))
1193 mark_page_accessed(pfn_to_page(pfn));
1195 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1197 void kvm_get_pfn(pfn_t pfn)
1199 if (!kvm_is_mmio_pfn(pfn))
1200 get_page(pfn_to_page(pfn));
1202 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1204 static int next_segment(unsigned long len, int offset)
1206 if (len > PAGE_SIZE - offset)
1207 return PAGE_SIZE - offset;
1212 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1218 addr = gfn_to_hva(kvm, gfn);
1219 if (kvm_is_error_hva(addr))
1221 r = copy_from_user(data, (void __user *)addr + offset, len);
1226 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1228 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1230 gfn_t gfn = gpa >> PAGE_SHIFT;
1232 int offset = offset_in_page(gpa);
1235 while ((seg = next_segment(len, offset)) != 0) {
1236 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1246 EXPORT_SYMBOL_GPL(kvm_read_guest);
1248 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1253 gfn_t gfn = gpa >> PAGE_SHIFT;
1254 int offset = offset_in_page(gpa);
1256 addr = gfn_to_hva(kvm, gfn);
1257 if (kvm_is_error_hva(addr))
1259 pagefault_disable();
1260 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1266 EXPORT_SYMBOL(kvm_read_guest_atomic);
1268 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1269 int offset, int len)
1274 addr = gfn_to_hva(kvm, gfn);
1275 if (kvm_is_error_hva(addr))
1277 r = copy_to_user((void __user *)addr + offset, data, len);
1280 mark_page_dirty(kvm, gfn);
1283 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1285 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1288 gfn_t gfn = gpa >> PAGE_SHIFT;
1290 int offset = offset_in_page(gpa);
1293 while ((seg = next_segment(len, offset)) != 0) {
1294 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1305 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1308 struct kvm_memslots *slots = kvm_memslots(kvm);
1309 int offset = offset_in_page(gpa);
1310 gfn_t gfn = gpa >> PAGE_SHIFT;
1313 ghc->generation = slots->generation;
1314 ghc->memslot = __gfn_to_memslot(slots, gfn);
1315 ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1316 if (!kvm_is_error_hva(ghc->hva))
1323 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1325 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1326 void *data, unsigned long len)
1328 struct kvm_memslots *slots = kvm_memslots(kvm);
1331 if (slots->generation != ghc->generation)
1332 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1334 if (kvm_is_error_hva(ghc->hva))
1337 r = copy_to_user((void __user *)ghc->hva, data, len);
1340 mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1344 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1346 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1348 return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1351 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1353 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1355 gfn_t gfn = gpa >> PAGE_SHIFT;
1357 int offset = offset_in_page(gpa);
1360 while ((seg = next_segment(len, offset)) != 0) {
1361 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1370 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1372 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1375 if (memslot && memslot->dirty_bitmap) {
1376 unsigned long rel_gfn = gfn - memslot->base_gfn;
1378 generic___set_le_bit(rel_gfn, memslot->dirty_bitmap);
1382 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1384 struct kvm_memory_slot *memslot;
1386 memslot = gfn_to_memslot(kvm, gfn);
1387 mark_page_dirty_in_slot(kvm, memslot, gfn);
1391 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1393 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1398 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1400 if (kvm_arch_vcpu_runnable(vcpu)) {
1401 kvm_make_request(KVM_REQ_UNHALT, vcpu);
1404 if (kvm_cpu_has_pending_timer(vcpu))
1406 if (signal_pending(current))
1412 finish_wait(&vcpu->wq, &wait);
1415 void kvm_resched(struct kvm_vcpu *vcpu)
1417 if (!need_resched())
1421 EXPORT_SYMBOL_GPL(kvm_resched);
1423 void kvm_vcpu_on_spin(struct kvm_vcpu *vcpu)
1428 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1430 /* Sleep for 100 us, and hope lock-holder got scheduled */
1431 expires = ktime_add_ns(ktime_get(), 100000UL);
1432 schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
1434 finish_wait(&vcpu->wq, &wait);
1436 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1438 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1440 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1443 if (vmf->pgoff == 0)
1444 page = virt_to_page(vcpu->run);
1446 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1447 page = virt_to_page(vcpu->arch.pio_data);
1449 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1450 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1451 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1454 return VM_FAULT_SIGBUS;
1460 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1461 .fault = kvm_vcpu_fault,
1464 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1466 vma->vm_ops = &kvm_vcpu_vm_ops;
1470 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1472 struct kvm_vcpu *vcpu = filp->private_data;
1474 kvm_put_kvm(vcpu->kvm);
1478 static struct file_operations kvm_vcpu_fops = {
1479 .release = kvm_vcpu_release,
1480 .unlocked_ioctl = kvm_vcpu_ioctl,
1481 .compat_ioctl = kvm_vcpu_ioctl,
1482 .mmap = kvm_vcpu_mmap,
1483 .llseek = noop_llseek,
1487 * Allocates an inode for the vcpu.
1489 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1491 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1495 * Creates some virtual cpus. Good luck creating more than one.
1497 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1500 struct kvm_vcpu *vcpu, *v;
1502 vcpu = kvm_arch_vcpu_create(kvm, id);
1504 return PTR_ERR(vcpu);
1506 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1508 r = kvm_arch_vcpu_setup(vcpu);
1512 mutex_lock(&kvm->lock);
1513 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1518 kvm_for_each_vcpu(r, v, kvm)
1519 if (v->vcpu_id == id) {
1524 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1526 /* Now it's all set up, let userspace reach it */
1528 r = create_vcpu_fd(vcpu);
1534 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1536 atomic_inc(&kvm->online_vcpus);
1538 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1539 if (kvm->bsp_vcpu_id == id)
1540 kvm->bsp_vcpu = vcpu;
1542 mutex_unlock(&kvm->lock);
1546 mutex_unlock(&kvm->lock);
1547 kvm_arch_vcpu_destroy(vcpu);
1551 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1554 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1555 vcpu->sigset_active = 1;
1556 vcpu->sigset = *sigset;
1558 vcpu->sigset_active = 0;
1562 static long kvm_vcpu_ioctl(struct file *filp,
1563 unsigned int ioctl, unsigned long arg)
1565 struct kvm_vcpu *vcpu = filp->private_data;
1566 void __user *argp = (void __user *)arg;
1568 struct kvm_fpu *fpu = NULL;
1569 struct kvm_sregs *kvm_sregs = NULL;
1571 if (vcpu->kvm->mm != current->mm)
1574 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1576 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1577 * so vcpu_load() would break it.
1579 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1580 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1590 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1591 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1593 case KVM_GET_REGS: {
1594 struct kvm_regs *kvm_regs;
1597 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1600 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1604 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1611 case KVM_SET_REGS: {
1612 struct kvm_regs *kvm_regs;
1615 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1619 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1621 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1629 case KVM_GET_SREGS: {
1630 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1634 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1638 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1643 case KVM_SET_SREGS: {
1644 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1649 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1651 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1657 case KVM_GET_MP_STATE: {
1658 struct kvm_mp_state mp_state;
1660 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1664 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1669 case KVM_SET_MP_STATE: {
1670 struct kvm_mp_state mp_state;
1673 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1675 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1681 case KVM_TRANSLATE: {
1682 struct kvm_translation tr;
1685 if (copy_from_user(&tr, argp, sizeof tr))
1687 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1691 if (copy_to_user(argp, &tr, sizeof tr))
1696 case KVM_SET_GUEST_DEBUG: {
1697 struct kvm_guest_debug dbg;
1700 if (copy_from_user(&dbg, argp, sizeof dbg))
1702 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1708 case KVM_SET_SIGNAL_MASK: {
1709 struct kvm_signal_mask __user *sigmask_arg = argp;
1710 struct kvm_signal_mask kvm_sigmask;
1711 sigset_t sigset, *p;
1716 if (copy_from_user(&kvm_sigmask, argp,
1717 sizeof kvm_sigmask))
1720 if (kvm_sigmask.len != sizeof sigset)
1723 if (copy_from_user(&sigset, sigmask_arg->sigset,
1728 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
1732 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1736 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1740 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1746 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1751 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1753 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1760 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1769 static long kvm_vm_ioctl(struct file *filp,
1770 unsigned int ioctl, unsigned long arg)
1772 struct kvm *kvm = filp->private_data;
1773 void __user *argp = (void __user *)arg;
1776 if (kvm->mm != current->mm)
1779 case KVM_CREATE_VCPU:
1780 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1784 case KVM_SET_USER_MEMORY_REGION: {
1785 struct kvm_userspace_memory_region kvm_userspace_mem;
1788 if (copy_from_user(&kvm_userspace_mem, argp,
1789 sizeof kvm_userspace_mem))
1792 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1797 case KVM_GET_DIRTY_LOG: {
1798 struct kvm_dirty_log log;
1801 if (copy_from_user(&log, argp, sizeof log))
1803 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1808 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1809 case KVM_REGISTER_COALESCED_MMIO: {
1810 struct kvm_coalesced_mmio_zone zone;
1812 if (copy_from_user(&zone, argp, sizeof zone))
1814 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1820 case KVM_UNREGISTER_COALESCED_MMIO: {
1821 struct kvm_coalesced_mmio_zone zone;
1823 if (copy_from_user(&zone, argp, sizeof zone))
1825 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1833 struct kvm_irqfd data;
1836 if (copy_from_user(&data, argp, sizeof data))
1838 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
1841 case KVM_IOEVENTFD: {
1842 struct kvm_ioeventfd data;
1845 if (copy_from_user(&data, argp, sizeof data))
1847 r = kvm_ioeventfd(kvm, &data);
1850 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1851 case KVM_SET_BOOT_CPU_ID:
1853 mutex_lock(&kvm->lock);
1854 if (atomic_read(&kvm->online_vcpus) != 0)
1857 kvm->bsp_vcpu_id = arg;
1858 mutex_unlock(&kvm->lock);
1862 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1864 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
1870 #ifdef CONFIG_COMPAT
1871 struct compat_kvm_dirty_log {
1875 compat_uptr_t dirty_bitmap; /* one bit per page */
1880 static long kvm_vm_compat_ioctl(struct file *filp,
1881 unsigned int ioctl, unsigned long arg)
1883 struct kvm *kvm = filp->private_data;
1886 if (kvm->mm != current->mm)
1889 case KVM_GET_DIRTY_LOG: {
1890 struct compat_kvm_dirty_log compat_log;
1891 struct kvm_dirty_log log;
1894 if (copy_from_user(&compat_log, (void __user *)arg,
1895 sizeof(compat_log)))
1897 log.slot = compat_log.slot;
1898 log.padding1 = compat_log.padding1;
1899 log.padding2 = compat_log.padding2;
1900 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
1902 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1908 r = kvm_vm_ioctl(filp, ioctl, arg);
1916 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1918 struct page *page[1];
1921 gfn_t gfn = vmf->pgoff;
1922 struct kvm *kvm = vma->vm_file->private_data;
1924 addr = gfn_to_hva(kvm, gfn);
1925 if (kvm_is_error_hva(addr))
1926 return VM_FAULT_SIGBUS;
1928 npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
1930 if (unlikely(npages != 1))
1931 return VM_FAULT_SIGBUS;
1933 vmf->page = page[0];
1937 static const struct vm_operations_struct kvm_vm_vm_ops = {
1938 .fault = kvm_vm_fault,
1941 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1943 vma->vm_ops = &kvm_vm_vm_ops;
1947 static struct file_operations kvm_vm_fops = {
1948 .release = kvm_vm_release,
1949 .unlocked_ioctl = kvm_vm_ioctl,
1950 #ifdef CONFIG_COMPAT
1951 .compat_ioctl = kvm_vm_compat_ioctl,
1953 .mmap = kvm_vm_mmap,
1954 .llseek = noop_llseek,
1957 static int kvm_dev_ioctl_create_vm(void)
1962 kvm = kvm_create_vm();
1964 return PTR_ERR(kvm);
1965 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1966 r = kvm_coalesced_mmio_init(kvm);
1972 r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
1979 static long kvm_dev_ioctl_check_extension_generic(long arg)
1982 case KVM_CAP_USER_MEMORY:
1983 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
1984 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
1985 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1986 case KVM_CAP_SET_BOOT_CPU_ID:
1988 case KVM_CAP_INTERNAL_ERROR_DATA:
1990 #ifdef CONFIG_HAVE_KVM_IRQCHIP
1991 case KVM_CAP_IRQ_ROUTING:
1992 return KVM_MAX_IRQ_ROUTES;
1997 return kvm_dev_ioctl_check_extension(arg);
2000 static long kvm_dev_ioctl(struct file *filp,
2001 unsigned int ioctl, unsigned long arg)
2006 case KVM_GET_API_VERSION:
2010 r = KVM_API_VERSION;
2016 r = kvm_dev_ioctl_create_vm();
2018 case KVM_CHECK_EXTENSION:
2019 r = kvm_dev_ioctl_check_extension_generic(arg);
2021 case KVM_GET_VCPU_MMAP_SIZE:
2025 r = PAGE_SIZE; /* struct kvm_run */
2027 r += PAGE_SIZE; /* pio data page */
2029 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2030 r += PAGE_SIZE; /* coalesced mmio ring page */
2033 case KVM_TRACE_ENABLE:
2034 case KVM_TRACE_PAUSE:
2035 case KVM_TRACE_DISABLE:
2039 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2045 static struct file_operations kvm_chardev_ops = {
2046 .unlocked_ioctl = kvm_dev_ioctl,
2047 .compat_ioctl = kvm_dev_ioctl,
2048 .llseek = noop_llseek,
2051 static struct miscdevice kvm_dev = {
2057 static void hardware_enable(void *junk)
2059 int cpu = raw_smp_processor_id();
2062 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2065 cpumask_set_cpu(cpu, cpus_hardware_enabled);
2067 r = kvm_arch_hardware_enable(NULL);
2070 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2071 atomic_inc(&hardware_enable_failed);
2072 printk(KERN_INFO "kvm: enabling virtualization on "
2073 "CPU%d failed\n", cpu);
2077 static void hardware_disable(void *junk)
2079 int cpu = raw_smp_processor_id();
2081 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2083 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2084 kvm_arch_hardware_disable(NULL);
2087 static void hardware_disable_all_nolock(void)
2089 BUG_ON(!kvm_usage_count);
2092 if (!kvm_usage_count)
2093 on_each_cpu(hardware_disable, NULL, 1);
2096 static void hardware_disable_all(void)
2098 spin_lock(&kvm_lock);
2099 hardware_disable_all_nolock();
2100 spin_unlock(&kvm_lock);
2103 static int hardware_enable_all(void)
2107 spin_lock(&kvm_lock);
2110 if (kvm_usage_count == 1) {
2111 atomic_set(&hardware_enable_failed, 0);
2112 on_each_cpu(hardware_enable, NULL, 1);
2114 if (atomic_read(&hardware_enable_failed)) {
2115 hardware_disable_all_nolock();
2120 spin_unlock(&kvm_lock);
2125 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2130 if (!kvm_usage_count)
2133 val &= ~CPU_TASKS_FROZEN;
2136 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2138 hardware_disable(NULL);
2141 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2143 spin_lock(&kvm_lock);
2144 hardware_enable(NULL);
2145 spin_unlock(&kvm_lock);
2152 asmlinkage void kvm_handle_fault_on_reboot(void)
2154 if (kvm_rebooting) {
2155 /* spin while reset goes on */
2160 /* Fault while not rebooting. We want the trace. */
2163 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
2165 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2169 * Some (well, at least mine) BIOSes hang on reboot if
2172 * And Intel TXT required VMX off for all cpu when system shutdown.
2174 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2175 kvm_rebooting = true;
2176 on_each_cpu(hardware_disable, NULL, 1);
2180 static struct notifier_block kvm_reboot_notifier = {
2181 .notifier_call = kvm_reboot,
2185 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2189 for (i = 0; i < bus->dev_count; i++) {
2190 struct kvm_io_device *pos = bus->devs[i];
2192 kvm_iodevice_destructor(pos);
2197 /* kvm_io_bus_write - called under kvm->slots_lock */
2198 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2199 int len, const void *val)
2202 struct kvm_io_bus *bus;
2204 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2205 for (i = 0; i < bus->dev_count; i++)
2206 if (!kvm_iodevice_write(bus->devs[i], addr, len, val))
2211 /* kvm_io_bus_read - called under kvm->slots_lock */
2212 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2216 struct kvm_io_bus *bus;
2218 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2219 for (i = 0; i < bus->dev_count; i++)
2220 if (!kvm_iodevice_read(bus->devs[i], addr, len, val))
2225 /* Caller must hold slots_lock. */
2226 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2227 struct kvm_io_device *dev)
2229 struct kvm_io_bus *new_bus, *bus;
2231 bus = kvm->buses[bus_idx];
2232 if (bus->dev_count > NR_IOBUS_DEVS-1)
2235 new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2238 memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2239 new_bus->devs[new_bus->dev_count++] = dev;
2240 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2241 synchronize_srcu_expedited(&kvm->srcu);
2247 /* Caller must hold slots_lock. */
2248 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2249 struct kvm_io_device *dev)
2252 struct kvm_io_bus *new_bus, *bus;
2254 new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2258 bus = kvm->buses[bus_idx];
2259 memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2262 for (i = 0; i < new_bus->dev_count; i++)
2263 if (new_bus->devs[i] == dev) {
2265 new_bus->devs[i] = new_bus->devs[--new_bus->dev_count];
2274 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2275 synchronize_srcu_expedited(&kvm->srcu);
2280 static struct notifier_block kvm_cpu_notifier = {
2281 .notifier_call = kvm_cpu_hotplug,
2284 static int vm_stat_get(void *_offset, u64 *val)
2286 unsigned offset = (long)_offset;
2290 spin_lock(&kvm_lock);
2291 list_for_each_entry(kvm, &vm_list, vm_list)
2292 *val += *(u32 *)((void *)kvm + offset);
2293 spin_unlock(&kvm_lock);
2297 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2299 static int vcpu_stat_get(void *_offset, u64 *val)
2301 unsigned offset = (long)_offset;
2303 struct kvm_vcpu *vcpu;
2307 spin_lock(&kvm_lock);
2308 list_for_each_entry(kvm, &vm_list, vm_list)
2309 kvm_for_each_vcpu(i, vcpu, kvm)
2310 *val += *(u32 *)((void *)vcpu + offset);
2312 spin_unlock(&kvm_lock);
2316 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2318 static const struct file_operations *stat_fops[] = {
2319 [KVM_STAT_VCPU] = &vcpu_stat_fops,
2320 [KVM_STAT_VM] = &vm_stat_fops,
2323 static void kvm_init_debug(void)
2325 struct kvm_stats_debugfs_item *p;
2327 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2328 for (p = debugfs_entries; p->name; ++p)
2329 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2330 (void *)(long)p->offset,
2331 stat_fops[p->kind]);
2334 static void kvm_exit_debug(void)
2336 struct kvm_stats_debugfs_item *p;
2338 for (p = debugfs_entries; p->name; ++p)
2339 debugfs_remove(p->dentry);
2340 debugfs_remove(kvm_debugfs_dir);
2343 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2345 if (kvm_usage_count)
2346 hardware_disable(NULL);
2350 static int kvm_resume(struct sys_device *dev)
2352 if (kvm_usage_count) {
2353 WARN_ON(spin_is_locked(&kvm_lock));
2354 hardware_enable(NULL);
2359 static struct sysdev_class kvm_sysdev_class = {
2361 .suspend = kvm_suspend,
2362 .resume = kvm_resume,
2365 static struct sys_device kvm_sysdev = {
2367 .cls = &kvm_sysdev_class,
2370 struct page *bad_page;
2374 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2376 return container_of(pn, struct kvm_vcpu, preempt_notifier);
2379 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2381 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2383 kvm_arch_vcpu_load(vcpu, cpu);
2386 static void kvm_sched_out(struct preempt_notifier *pn,
2387 struct task_struct *next)
2389 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2391 kvm_arch_vcpu_put(vcpu);
2394 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2395 struct module *module)
2400 r = kvm_arch_init(opaque);
2404 bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2406 if (bad_page == NULL) {
2411 bad_pfn = page_to_pfn(bad_page);
2413 hwpoison_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2415 if (hwpoison_page == NULL) {
2420 hwpoison_pfn = page_to_pfn(hwpoison_page);
2422 fault_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2424 if (fault_page == NULL) {
2429 fault_pfn = page_to_pfn(fault_page);
2431 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2436 r = kvm_arch_hardware_setup();
2440 for_each_online_cpu(cpu) {
2441 smp_call_function_single(cpu,
2442 kvm_arch_check_processor_compat,
2448 r = register_cpu_notifier(&kvm_cpu_notifier);
2451 register_reboot_notifier(&kvm_reboot_notifier);
2453 r = sysdev_class_register(&kvm_sysdev_class);
2457 r = sysdev_register(&kvm_sysdev);
2461 /* A kmem cache lets us meet the alignment requirements of fx_save. */
2463 vcpu_align = __alignof__(struct kvm_vcpu);
2464 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2466 if (!kvm_vcpu_cache) {
2471 r = kvm_async_pf_init();
2475 kvm_chardev_ops.owner = module;
2476 kvm_vm_fops.owner = module;
2477 kvm_vcpu_fops.owner = module;
2479 r = misc_register(&kvm_dev);
2481 printk(KERN_ERR "kvm: misc device register failed\n");
2485 kvm_preempt_ops.sched_in = kvm_sched_in;
2486 kvm_preempt_ops.sched_out = kvm_sched_out;
2493 kvm_async_pf_deinit();
2495 kmem_cache_destroy(kvm_vcpu_cache);
2497 sysdev_unregister(&kvm_sysdev);
2499 sysdev_class_unregister(&kvm_sysdev_class);
2501 unregister_reboot_notifier(&kvm_reboot_notifier);
2502 unregister_cpu_notifier(&kvm_cpu_notifier);
2505 kvm_arch_hardware_unsetup();
2507 free_cpumask_var(cpus_hardware_enabled);
2510 __free_page(fault_page);
2512 __free_page(hwpoison_page);
2513 __free_page(bad_page);
2519 EXPORT_SYMBOL_GPL(kvm_init);
2524 misc_deregister(&kvm_dev);
2525 kmem_cache_destroy(kvm_vcpu_cache);
2526 kvm_async_pf_deinit();
2527 sysdev_unregister(&kvm_sysdev);
2528 sysdev_class_unregister(&kvm_sysdev_class);
2529 unregister_reboot_notifier(&kvm_reboot_notifier);
2530 unregister_cpu_notifier(&kvm_cpu_notifier);
2531 on_each_cpu(hardware_disable, NULL, 1);
2532 kvm_arch_hardware_unsetup();
2534 free_cpumask_var(cpus_hardware_enabled);
2535 __free_page(hwpoison_page);
2536 __free_page(bad_page);
2538 EXPORT_SYMBOL_GPL(kvm_exit);