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 affilates.
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"
59 #define CREATE_TRACE_POINTS
60 #include <trace/events/kvm.h>
62 MODULE_AUTHOR("Qumranet");
63 MODULE_LICENSE("GPL");
68 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
71 DEFINE_SPINLOCK(kvm_lock);
74 static cpumask_var_t cpus_hardware_enabled;
75 static int kvm_usage_count = 0;
76 static atomic_t hardware_enable_failed;
78 struct kmem_cache *kvm_vcpu_cache;
79 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
81 static __read_mostly struct preempt_ops kvm_preempt_ops;
83 struct dentry *kvm_debugfs_dir;
85 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
87 static int hardware_enable_all(void);
88 static void hardware_disable_all(void);
90 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
92 static bool kvm_rebooting;
94 static bool largepages_enabled = true;
96 struct page *hwpoison_page;
99 inline int kvm_is_mmio_pfn(pfn_t pfn)
101 if (pfn_valid(pfn)) {
102 struct page *page = compound_head(pfn_to_page(pfn));
103 return PageReserved(page);
110 * Switches to specified vcpu, until a matching vcpu_put()
112 void vcpu_load(struct kvm_vcpu *vcpu)
116 mutex_lock(&vcpu->mutex);
118 preempt_notifier_register(&vcpu->preempt_notifier);
119 kvm_arch_vcpu_load(vcpu, cpu);
123 void vcpu_put(struct kvm_vcpu *vcpu)
126 kvm_arch_vcpu_put(vcpu);
127 preempt_notifier_unregister(&vcpu->preempt_notifier);
129 mutex_unlock(&vcpu->mutex);
132 static void ack_flush(void *_completed)
136 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
141 struct kvm_vcpu *vcpu;
143 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
145 raw_spin_lock(&kvm->requests_lock);
146 me = smp_processor_id();
147 kvm_for_each_vcpu(i, vcpu, kvm) {
148 if (kvm_make_check_request(req, vcpu))
151 if (cpus != NULL && cpu != -1 && cpu != me)
152 cpumask_set_cpu(cpu, cpus);
154 if (unlikely(cpus == NULL))
155 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
156 else if (!cpumask_empty(cpus))
157 smp_call_function_many(cpus, ack_flush, NULL, 1);
160 raw_spin_unlock(&kvm->requests_lock);
161 free_cpumask_var(cpus);
165 void kvm_flush_remote_tlbs(struct kvm *kvm)
167 if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
168 ++kvm->stat.remote_tlb_flush;
171 void kvm_reload_remote_mmus(struct kvm *kvm)
173 make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
176 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
181 mutex_init(&vcpu->mutex);
185 init_waitqueue_head(&vcpu->wq);
187 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
192 vcpu->run = page_address(page);
194 r = kvm_arch_vcpu_init(vcpu);
200 free_page((unsigned long)vcpu->run);
204 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
206 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
208 kvm_arch_vcpu_uninit(vcpu);
209 free_page((unsigned long)vcpu->run);
211 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
213 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
214 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
216 return container_of(mn, struct kvm, mmu_notifier);
219 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
220 struct mm_struct *mm,
221 unsigned long address)
223 struct kvm *kvm = mmu_notifier_to_kvm(mn);
224 int need_tlb_flush, idx;
227 * When ->invalidate_page runs, the linux pte has been zapped
228 * already but the page is still allocated until
229 * ->invalidate_page returns. So if we increase the sequence
230 * here the kvm page fault will notice if the spte can't be
231 * established because the page is going to be freed. If
232 * instead the kvm page fault establishes the spte before
233 * ->invalidate_page runs, kvm_unmap_hva will release it
236 * The sequence increase only need to be seen at spin_unlock
237 * time, and not at spin_lock time.
239 * Increasing the sequence after the spin_unlock would be
240 * unsafe because the kvm page fault could then establish the
241 * pte after kvm_unmap_hva returned, without noticing the page
242 * is going to be freed.
244 idx = srcu_read_lock(&kvm->srcu);
245 spin_lock(&kvm->mmu_lock);
246 kvm->mmu_notifier_seq++;
247 need_tlb_flush = kvm_unmap_hva(kvm, address);
248 spin_unlock(&kvm->mmu_lock);
249 srcu_read_unlock(&kvm->srcu, idx);
251 /* we've to flush the tlb before the pages can be freed */
253 kvm_flush_remote_tlbs(kvm);
257 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
258 struct mm_struct *mm,
259 unsigned long address,
262 struct kvm *kvm = mmu_notifier_to_kvm(mn);
265 idx = srcu_read_lock(&kvm->srcu);
266 spin_lock(&kvm->mmu_lock);
267 kvm->mmu_notifier_seq++;
268 kvm_set_spte_hva(kvm, address, pte);
269 spin_unlock(&kvm->mmu_lock);
270 srcu_read_unlock(&kvm->srcu, idx);
273 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
274 struct mm_struct *mm,
278 struct kvm *kvm = mmu_notifier_to_kvm(mn);
279 int need_tlb_flush = 0, idx;
281 idx = srcu_read_lock(&kvm->srcu);
282 spin_lock(&kvm->mmu_lock);
284 * The count increase must become visible at unlock time as no
285 * spte can be established without taking the mmu_lock and
286 * count is also read inside the mmu_lock critical section.
288 kvm->mmu_notifier_count++;
289 for (; start < end; start += PAGE_SIZE)
290 need_tlb_flush |= kvm_unmap_hva(kvm, start);
291 spin_unlock(&kvm->mmu_lock);
292 srcu_read_unlock(&kvm->srcu, idx);
294 /* we've to flush the tlb before the pages can be freed */
296 kvm_flush_remote_tlbs(kvm);
299 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
300 struct mm_struct *mm,
304 struct kvm *kvm = mmu_notifier_to_kvm(mn);
306 spin_lock(&kvm->mmu_lock);
308 * This sequence increase will notify the kvm page fault that
309 * the page that is going to be mapped in the spte could have
312 kvm->mmu_notifier_seq++;
314 * The above sequence increase must be visible before the
315 * below count decrease but both values are read by the kvm
316 * page fault under mmu_lock spinlock so we don't need to add
317 * a smb_wmb() here in between the two.
319 kvm->mmu_notifier_count--;
320 spin_unlock(&kvm->mmu_lock);
322 BUG_ON(kvm->mmu_notifier_count < 0);
325 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
326 struct mm_struct *mm,
327 unsigned long address)
329 struct kvm *kvm = mmu_notifier_to_kvm(mn);
332 idx = srcu_read_lock(&kvm->srcu);
333 spin_lock(&kvm->mmu_lock);
334 young = kvm_age_hva(kvm, address);
335 spin_unlock(&kvm->mmu_lock);
336 srcu_read_unlock(&kvm->srcu, idx);
339 kvm_flush_remote_tlbs(kvm);
344 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
345 struct mm_struct *mm)
347 struct kvm *kvm = mmu_notifier_to_kvm(mn);
350 idx = srcu_read_lock(&kvm->srcu);
351 kvm_arch_flush_shadow(kvm);
352 srcu_read_unlock(&kvm->srcu, idx);
355 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
356 .invalidate_page = kvm_mmu_notifier_invalidate_page,
357 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
358 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
359 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
360 .change_pte = kvm_mmu_notifier_change_pte,
361 .release = kvm_mmu_notifier_release,
364 static int kvm_init_mmu_notifier(struct kvm *kvm)
366 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
367 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
370 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
372 static int kvm_init_mmu_notifier(struct kvm *kvm)
377 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
379 static struct kvm *kvm_create_vm(void)
382 struct kvm *kvm = kvm_arch_create_vm();
387 r = hardware_enable_all();
389 goto out_err_nodisable;
391 #ifdef CONFIG_HAVE_KVM_IRQCHIP
392 INIT_HLIST_HEAD(&kvm->mask_notifier_list);
393 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
397 kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
400 if (init_srcu_struct(&kvm->srcu))
402 for (i = 0; i < KVM_NR_BUSES; i++) {
403 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
405 if (!kvm->buses[i]) {
406 cleanup_srcu_struct(&kvm->srcu);
411 r = kvm_init_mmu_notifier(kvm);
413 cleanup_srcu_struct(&kvm->srcu);
417 kvm->mm = current->mm;
418 atomic_inc(&kvm->mm->mm_count);
419 spin_lock_init(&kvm->mmu_lock);
420 raw_spin_lock_init(&kvm->requests_lock);
421 kvm_eventfd_init(kvm);
422 mutex_init(&kvm->lock);
423 mutex_init(&kvm->irq_lock);
424 mutex_init(&kvm->slots_lock);
425 atomic_set(&kvm->users_count, 1);
426 spin_lock(&kvm_lock);
427 list_add(&kvm->vm_list, &vm_list);
428 spin_unlock(&kvm_lock);
433 hardware_disable_all();
435 for (i = 0; i < KVM_NR_BUSES; i++)
436 kfree(kvm->buses[i]);
437 kfree(kvm->memslots);
443 * Free any memory in @free but not in @dont.
445 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
446 struct kvm_memory_slot *dont)
450 if (!dont || free->rmap != dont->rmap)
453 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
454 vfree(free->dirty_bitmap);
457 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
458 if (!dont || free->lpage_info[i] != dont->lpage_info[i]) {
459 vfree(free->lpage_info[i]);
460 free->lpage_info[i] = NULL;
465 free->dirty_bitmap = NULL;
469 void kvm_free_physmem(struct kvm *kvm)
472 struct kvm_memslots *slots = kvm->memslots;
474 for (i = 0; i < slots->nmemslots; ++i)
475 kvm_free_physmem_slot(&slots->memslots[i], NULL);
477 kfree(kvm->memslots);
480 static void kvm_destroy_vm(struct kvm *kvm)
483 struct mm_struct *mm = kvm->mm;
485 kvm_arch_sync_events(kvm);
486 spin_lock(&kvm_lock);
487 list_del(&kvm->vm_list);
488 spin_unlock(&kvm_lock);
489 kvm_free_irq_routing(kvm);
490 for (i = 0; i < KVM_NR_BUSES; i++)
491 kvm_io_bus_destroy(kvm->buses[i]);
492 kvm_coalesced_mmio_free(kvm);
493 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
494 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
496 kvm_arch_flush_shadow(kvm);
498 kvm_arch_destroy_vm(kvm);
499 hardware_disable_all();
503 void kvm_get_kvm(struct kvm *kvm)
505 atomic_inc(&kvm->users_count);
507 EXPORT_SYMBOL_GPL(kvm_get_kvm);
509 void kvm_put_kvm(struct kvm *kvm)
511 if (atomic_dec_and_test(&kvm->users_count))
514 EXPORT_SYMBOL_GPL(kvm_put_kvm);
517 static int kvm_vm_release(struct inode *inode, struct file *filp)
519 struct kvm *kvm = filp->private_data;
521 kvm_irqfd_release(kvm);
528 * Allocate some memory and give it an address in the guest physical address
531 * Discontiguous memory is allowed, mostly for framebuffers.
533 * Must be called holding mmap_sem for write.
535 int __kvm_set_memory_region(struct kvm *kvm,
536 struct kvm_userspace_memory_region *mem,
539 int r, flush_shadow = 0;
541 unsigned long npages;
543 struct kvm_memory_slot *memslot;
544 struct kvm_memory_slot old, new;
545 struct kvm_memslots *slots, *old_memslots;
548 /* General sanity checks */
549 if (mem->memory_size & (PAGE_SIZE - 1))
551 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
553 if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
555 if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
557 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
560 memslot = &kvm->memslots->memslots[mem->slot];
561 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
562 npages = mem->memory_size >> PAGE_SHIFT;
565 if (npages > KVM_MEM_MAX_NR_PAGES)
569 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
571 new = old = *memslot;
574 new.base_gfn = base_gfn;
576 new.flags = mem->flags;
578 /* Disallow changing a memory slot's size. */
580 if (npages && old.npages && npages != old.npages)
583 /* Check for overlaps */
585 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
586 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
588 if (s == memslot || !s->npages)
590 if (!((base_gfn + npages <= s->base_gfn) ||
591 (base_gfn >= s->base_gfn + s->npages)))
595 /* Free page dirty bitmap if unneeded */
596 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
597 new.dirty_bitmap = NULL;
601 /* Allocate if a slot is being created */
603 if (npages && !new.rmap) {
604 new.rmap = vmalloc(npages * sizeof(*new.rmap));
609 memset(new.rmap, 0, npages * sizeof(*new.rmap));
611 new.user_alloc = user_alloc;
612 new.userspace_addr = mem->userspace_addr;
617 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
623 /* Avoid unused variable warning if no large pages */
626 if (new.lpage_info[i])
629 lpages = 1 + (base_gfn + npages - 1) /
630 KVM_PAGES_PER_HPAGE(level);
631 lpages -= base_gfn / KVM_PAGES_PER_HPAGE(level);
633 new.lpage_info[i] = vmalloc(lpages * sizeof(*new.lpage_info[i]));
635 if (!new.lpage_info[i])
638 memset(new.lpage_info[i], 0,
639 lpages * sizeof(*new.lpage_info[i]));
641 if (base_gfn % KVM_PAGES_PER_HPAGE(level))
642 new.lpage_info[i][0].write_count = 1;
643 if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE(level))
644 new.lpage_info[i][lpages - 1].write_count = 1;
645 ugfn = new.userspace_addr >> PAGE_SHIFT;
647 * If the gfn and userspace address are not aligned wrt each
648 * other, or if explicitly asked to, disable large page
649 * support for this slot
651 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
653 for (j = 0; j < lpages; ++j)
654 new.lpage_info[i][j].write_count = 1;
659 /* Allocate page dirty bitmap if needed */
660 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
661 unsigned long dirty_bytes = kvm_dirty_bitmap_bytes(&new);
663 new.dirty_bitmap = vmalloc(dirty_bytes);
664 if (!new.dirty_bitmap)
666 memset(new.dirty_bitmap, 0, dirty_bytes);
667 /* destroy any largepage mappings for dirty tracking */
671 #else /* not defined CONFIG_S390 */
672 new.user_alloc = user_alloc;
674 new.userspace_addr = mem->userspace_addr;
675 #endif /* not defined CONFIG_S390 */
679 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
682 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
683 if (mem->slot >= slots->nmemslots)
684 slots->nmemslots = mem->slot + 1;
685 slots->memslots[mem->slot].flags |= KVM_MEMSLOT_INVALID;
687 old_memslots = kvm->memslots;
688 rcu_assign_pointer(kvm->memslots, slots);
689 synchronize_srcu_expedited(&kvm->srcu);
690 /* From this point no new shadow pages pointing to a deleted
691 * memslot will be created.
693 * validation of sp->gfn happens in:
694 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
695 * - kvm_is_visible_gfn (mmu_check_roots)
697 kvm_arch_flush_shadow(kvm);
701 r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
706 /* map the pages in iommu page table */
708 r = kvm_iommu_map_pages(kvm, &new);
715 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
718 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
719 if (mem->slot >= slots->nmemslots)
720 slots->nmemslots = mem->slot + 1;
722 /* actual memory is freed via old in kvm_free_physmem_slot below */
725 new.dirty_bitmap = NULL;
726 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i)
727 new.lpage_info[i] = NULL;
730 slots->memslots[mem->slot] = new;
731 old_memslots = kvm->memslots;
732 rcu_assign_pointer(kvm->memslots, slots);
733 synchronize_srcu_expedited(&kvm->srcu);
735 kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
737 kvm_free_physmem_slot(&old, &new);
741 kvm_arch_flush_shadow(kvm);
746 kvm_free_physmem_slot(&new, &old);
751 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
753 int kvm_set_memory_region(struct kvm *kvm,
754 struct kvm_userspace_memory_region *mem,
759 mutex_lock(&kvm->slots_lock);
760 r = __kvm_set_memory_region(kvm, mem, user_alloc);
761 mutex_unlock(&kvm->slots_lock);
764 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
766 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
768 kvm_userspace_memory_region *mem,
771 if (mem->slot >= KVM_MEMORY_SLOTS)
773 return kvm_set_memory_region(kvm, mem, user_alloc);
776 int kvm_get_dirty_log(struct kvm *kvm,
777 struct kvm_dirty_log *log, int *is_dirty)
779 struct kvm_memory_slot *memslot;
782 unsigned long any = 0;
785 if (log->slot >= KVM_MEMORY_SLOTS)
788 memslot = &kvm->memslots->memslots[log->slot];
790 if (!memslot->dirty_bitmap)
793 n = kvm_dirty_bitmap_bytes(memslot);
795 for (i = 0; !any && i < n/sizeof(long); ++i)
796 any = memslot->dirty_bitmap[i];
799 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
810 void kvm_disable_largepages(void)
812 largepages_enabled = false;
814 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
816 int is_error_page(struct page *page)
818 return page == bad_page || page == hwpoison_page;
820 EXPORT_SYMBOL_GPL(is_error_page);
822 int is_error_pfn(pfn_t pfn)
824 return pfn == bad_pfn || pfn == hwpoison_pfn;
826 EXPORT_SYMBOL_GPL(is_error_pfn);
828 int is_hwpoison_pfn(pfn_t pfn)
830 return pfn == hwpoison_pfn;
832 EXPORT_SYMBOL_GPL(is_hwpoison_pfn);
834 static inline unsigned long bad_hva(void)
839 int kvm_is_error_hva(unsigned long addr)
841 return addr == bad_hva();
843 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
845 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
848 struct kvm_memslots *slots = kvm_memslots(kvm);
850 for (i = 0; i < slots->nmemslots; ++i) {
851 struct kvm_memory_slot *memslot = &slots->memslots[i];
853 if (gfn >= memslot->base_gfn
854 && gfn < memslot->base_gfn + memslot->npages)
859 EXPORT_SYMBOL_GPL(gfn_to_memslot);
861 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
864 struct kvm_memslots *slots = kvm_memslots(kvm);
866 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
867 struct kvm_memory_slot *memslot = &slots->memslots[i];
869 if (memslot->flags & KVM_MEMSLOT_INVALID)
872 if (gfn >= memslot->base_gfn
873 && gfn < memslot->base_gfn + memslot->npages)
878 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
880 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
882 struct vm_area_struct *vma;
883 unsigned long addr, size;
887 addr = gfn_to_hva(kvm, gfn);
888 if (kvm_is_error_hva(addr))
891 down_read(¤t->mm->mmap_sem);
892 vma = find_vma(current->mm, addr);
896 size = vma_kernel_pagesize(vma);
899 up_read(¤t->mm->mmap_sem);
904 int memslot_id(struct kvm *kvm, gfn_t gfn)
907 struct kvm_memslots *slots = kvm_memslots(kvm);
908 struct kvm_memory_slot *memslot = NULL;
910 for (i = 0; i < slots->nmemslots; ++i) {
911 memslot = &slots->memslots[i];
913 if (gfn >= memslot->base_gfn
914 && gfn < memslot->base_gfn + memslot->npages)
918 return memslot - slots->memslots;
921 static unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
923 return slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE;
926 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
928 struct kvm_memory_slot *slot;
930 slot = gfn_to_memslot(kvm, gfn);
931 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
933 return gfn_to_hva_memslot(slot, gfn);
935 EXPORT_SYMBOL_GPL(gfn_to_hva);
937 static pfn_t hva_to_pfn(struct kvm *kvm, unsigned long addr)
939 struct page *page[1];
945 npages = get_user_pages_fast(addr, 1, 1, page);
947 if (unlikely(npages != 1)) {
948 struct vm_area_struct *vma;
950 down_read(¤t->mm->mmap_sem);
951 if (is_hwpoison_address(addr)) {
952 up_read(¤t->mm->mmap_sem);
953 get_page(hwpoison_page);
954 return page_to_pfn(hwpoison_page);
957 vma = find_vma(current->mm, addr);
959 if (vma == NULL || addr < vma->vm_start ||
960 !(vma->vm_flags & VM_PFNMAP)) {
961 up_read(¤t->mm->mmap_sem);
963 return page_to_pfn(bad_page);
966 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
967 up_read(¤t->mm->mmap_sem);
968 BUG_ON(!kvm_is_mmio_pfn(pfn));
970 pfn = page_to_pfn(page[0]);
975 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
979 addr = gfn_to_hva(kvm, gfn);
980 if (kvm_is_error_hva(addr)) {
982 return page_to_pfn(bad_page);
985 return hva_to_pfn(kvm, addr);
987 EXPORT_SYMBOL_GPL(gfn_to_pfn);
989 pfn_t gfn_to_pfn_memslot(struct kvm *kvm,
990 struct kvm_memory_slot *slot, gfn_t gfn)
992 unsigned long addr = gfn_to_hva_memslot(slot, gfn);
993 return hva_to_pfn(kvm, addr);
996 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1000 pfn = gfn_to_pfn(kvm, gfn);
1001 if (!kvm_is_mmio_pfn(pfn))
1002 return pfn_to_page(pfn);
1004 WARN_ON(kvm_is_mmio_pfn(pfn));
1010 EXPORT_SYMBOL_GPL(gfn_to_page);
1012 void kvm_release_page_clean(struct page *page)
1014 kvm_release_pfn_clean(page_to_pfn(page));
1016 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1018 void kvm_release_pfn_clean(pfn_t pfn)
1020 if (!kvm_is_mmio_pfn(pfn))
1021 put_page(pfn_to_page(pfn));
1023 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1025 void kvm_release_page_dirty(struct page *page)
1027 kvm_release_pfn_dirty(page_to_pfn(page));
1029 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1031 void kvm_release_pfn_dirty(pfn_t pfn)
1033 kvm_set_pfn_dirty(pfn);
1034 kvm_release_pfn_clean(pfn);
1036 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1038 void kvm_set_page_dirty(struct page *page)
1040 kvm_set_pfn_dirty(page_to_pfn(page));
1042 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1044 void kvm_set_pfn_dirty(pfn_t pfn)
1046 if (!kvm_is_mmio_pfn(pfn)) {
1047 struct page *page = pfn_to_page(pfn);
1048 if (!PageReserved(page))
1052 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1054 void kvm_set_pfn_accessed(pfn_t pfn)
1056 if (!kvm_is_mmio_pfn(pfn))
1057 mark_page_accessed(pfn_to_page(pfn));
1059 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1061 void kvm_get_pfn(pfn_t pfn)
1063 if (!kvm_is_mmio_pfn(pfn))
1064 get_page(pfn_to_page(pfn));
1066 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1068 static int next_segment(unsigned long len, int offset)
1070 if (len > PAGE_SIZE - offset)
1071 return PAGE_SIZE - offset;
1076 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1082 addr = gfn_to_hva(kvm, gfn);
1083 if (kvm_is_error_hva(addr))
1085 r = copy_from_user(data, (void __user *)addr + offset, len);
1090 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1092 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1094 gfn_t gfn = gpa >> PAGE_SHIFT;
1096 int offset = offset_in_page(gpa);
1099 while ((seg = next_segment(len, offset)) != 0) {
1100 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1110 EXPORT_SYMBOL_GPL(kvm_read_guest);
1112 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1117 gfn_t gfn = gpa >> PAGE_SHIFT;
1118 int offset = offset_in_page(gpa);
1120 addr = gfn_to_hva(kvm, gfn);
1121 if (kvm_is_error_hva(addr))
1123 pagefault_disable();
1124 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1130 EXPORT_SYMBOL(kvm_read_guest_atomic);
1132 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1133 int offset, int len)
1138 addr = gfn_to_hva(kvm, gfn);
1139 if (kvm_is_error_hva(addr))
1141 r = copy_to_user((void __user *)addr + offset, data, len);
1144 mark_page_dirty(kvm, gfn);
1147 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1149 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1152 gfn_t gfn = gpa >> PAGE_SHIFT;
1154 int offset = offset_in_page(gpa);
1157 while ((seg = next_segment(len, offset)) != 0) {
1158 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1169 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1171 return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
1173 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1175 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1177 gfn_t gfn = gpa >> PAGE_SHIFT;
1179 int offset = offset_in_page(gpa);
1182 while ((seg = next_segment(len, offset)) != 0) {
1183 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1192 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1194 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1196 struct kvm_memory_slot *memslot;
1198 memslot = gfn_to_memslot(kvm, gfn);
1199 if (memslot && memslot->dirty_bitmap) {
1200 unsigned long rel_gfn = gfn - memslot->base_gfn;
1202 generic___set_le_bit(rel_gfn, memslot->dirty_bitmap);
1207 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1209 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1214 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1216 if (kvm_arch_vcpu_runnable(vcpu)) {
1217 kvm_make_request(KVM_REQ_UNHALT, vcpu);
1220 if (kvm_cpu_has_pending_timer(vcpu))
1222 if (signal_pending(current))
1228 finish_wait(&vcpu->wq, &wait);
1231 void kvm_resched(struct kvm_vcpu *vcpu)
1233 if (!need_resched())
1237 EXPORT_SYMBOL_GPL(kvm_resched);
1239 void kvm_vcpu_on_spin(struct kvm_vcpu *vcpu)
1244 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1246 /* Sleep for 100 us, and hope lock-holder got scheduled */
1247 expires = ktime_add_ns(ktime_get(), 100000UL);
1248 schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
1250 finish_wait(&vcpu->wq, &wait);
1252 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1254 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1256 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1259 if (vmf->pgoff == 0)
1260 page = virt_to_page(vcpu->run);
1262 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1263 page = virt_to_page(vcpu->arch.pio_data);
1265 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1266 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1267 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1270 return VM_FAULT_SIGBUS;
1276 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1277 .fault = kvm_vcpu_fault,
1280 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1282 vma->vm_ops = &kvm_vcpu_vm_ops;
1286 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1288 struct kvm_vcpu *vcpu = filp->private_data;
1290 kvm_put_kvm(vcpu->kvm);
1294 static struct file_operations kvm_vcpu_fops = {
1295 .release = kvm_vcpu_release,
1296 .unlocked_ioctl = kvm_vcpu_ioctl,
1297 .compat_ioctl = kvm_vcpu_ioctl,
1298 .mmap = kvm_vcpu_mmap,
1302 * Allocates an inode for the vcpu.
1304 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1306 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1310 * Creates some virtual cpus. Good luck creating more than one.
1312 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1315 struct kvm_vcpu *vcpu, *v;
1317 vcpu = kvm_arch_vcpu_create(kvm, id);
1319 return PTR_ERR(vcpu);
1321 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1323 r = kvm_arch_vcpu_setup(vcpu);
1327 mutex_lock(&kvm->lock);
1328 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1333 kvm_for_each_vcpu(r, v, kvm)
1334 if (v->vcpu_id == id) {
1339 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1341 /* Now it's all set up, let userspace reach it */
1343 r = create_vcpu_fd(vcpu);
1349 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1351 atomic_inc(&kvm->online_vcpus);
1353 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1354 if (kvm->bsp_vcpu_id == id)
1355 kvm->bsp_vcpu = vcpu;
1357 mutex_unlock(&kvm->lock);
1361 mutex_unlock(&kvm->lock);
1362 kvm_arch_vcpu_destroy(vcpu);
1366 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1369 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1370 vcpu->sigset_active = 1;
1371 vcpu->sigset = *sigset;
1373 vcpu->sigset_active = 0;
1377 static long kvm_vcpu_ioctl(struct file *filp,
1378 unsigned int ioctl, unsigned long arg)
1380 struct kvm_vcpu *vcpu = filp->private_data;
1381 void __user *argp = (void __user *)arg;
1383 struct kvm_fpu *fpu = NULL;
1384 struct kvm_sregs *kvm_sregs = NULL;
1386 if (vcpu->kvm->mm != current->mm)
1389 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1391 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1392 * so vcpu_load() would break it.
1394 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1395 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1405 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1407 case KVM_GET_REGS: {
1408 struct kvm_regs *kvm_regs;
1411 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1414 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1418 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1425 case KVM_SET_REGS: {
1426 struct kvm_regs *kvm_regs;
1429 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1433 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1435 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1443 case KVM_GET_SREGS: {
1444 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1448 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1452 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1457 case KVM_SET_SREGS: {
1458 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1463 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1465 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1471 case KVM_GET_MP_STATE: {
1472 struct kvm_mp_state mp_state;
1474 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1478 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1483 case KVM_SET_MP_STATE: {
1484 struct kvm_mp_state mp_state;
1487 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1489 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1495 case KVM_TRANSLATE: {
1496 struct kvm_translation tr;
1499 if (copy_from_user(&tr, argp, sizeof tr))
1501 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1505 if (copy_to_user(argp, &tr, sizeof tr))
1510 case KVM_SET_GUEST_DEBUG: {
1511 struct kvm_guest_debug dbg;
1514 if (copy_from_user(&dbg, argp, sizeof dbg))
1516 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1522 case KVM_SET_SIGNAL_MASK: {
1523 struct kvm_signal_mask __user *sigmask_arg = argp;
1524 struct kvm_signal_mask kvm_sigmask;
1525 sigset_t sigset, *p;
1530 if (copy_from_user(&kvm_sigmask, argp,
1531 sizeof kvm_sigmask))
1534 if (kvm_sigmask.len != sizeof sigset)
1537 if (copy_from_user(&sigset, sigmask_arg->sigset,
1542 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
1546 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1550 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1554 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1560 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1565 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1567 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1574 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1583 static long kvm_vm_ioctl(struct file *filp,
1584 unsigned int ioctl, unsigned long arg)
1586 struct kvm *kvm = filp->private_data;
1587 void __user *argp = (void __user *)arg;
1590 if (kvm->mm != current->mm)
1593 case KVM_CREATE_VCPU:
1594 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1598 case KVM_SET_USER_MEMORY_REGION: {
1599 struct kvm_userspace_memory_region kvm_userspace_mem;
1602 if (copy_from_user(&kvm_userspace_mem, argp,
1603 sizeof kvm_userspace_mem))
1606 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1611 case KVM_GET_DIRTY_LOG: {
1612 struct kvm_dirty_log log;
1615 if (copy_from_user(&log, argp, sizeof log))
1617 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1622 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1623 case KVM_REGISTER_COALESCED_MMIO: {
1624 struct kvm_coalesced_mmio_zone zone;
1626 if (copy_from_user(&zone, argp, sizeof zone))
1628 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1634 case KVM_UNREGISTER_COALESCED_MMIO: {
1635 struct kvm_coalesced_mmio_zone zone;
1637 if (copy_from_user(&zone, argp, sizeof zone))
1639 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1647 struct kvm_irqfd data;
1650 if (copy_from_user(&data, argp, sizeof data))
1652 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
1655 case KVM_IOEVENTFD: {
1656 struct kvm_ioeventfd data;
1659 if (copy_from_user(&data, argp, sizeof data))
1661 r = kvm_ioeventfd(kvm, &data);
1664 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1665 case KVM_SET_BOOT_CPU_ID:
1667 mutex_lock(&kvm->lock);
1668 if (atomic_read(&kvm->online_vcpus) != 0)
1671 kvm->bsp_vcpu_id = arg;
1672 mutex_unlock(&kvm->lock);
1676 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1678 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
1684 #ifdef CONFIG_COMPAT
1685 struct compat_kvm_dirty_log {
1689 compat_uptr_t dirty_bitmap; /* one bit per page */
1694 static long kvm_vm_compat_ioctl(struct file *filp,
1695 unsigned int ioctl, unsigned long arg)
1697 struct kvm *kvm = filp->private_data;
1700 if (kvm->mm != current->mm)
1703 case KVM_GET_DIRTY_LOG: {
1704 struct compat_kvm_dirty_log compat_log;
1705 struct kvm_dirty_log log;
1708 if (copy_from_user(&compat_log, (void __user *)arg,
1709 sizeof(compat_log)))
1711 log.slot = compat_log.slot;
1712 log.padding1 = compat_log.padding1;
1713 log.padding2 = compat_log.padding2;
1714 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
1716 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1722 r = kvm_vm_ioctl(filp, ioctl, arg);
1730 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1732 struct page *page[1];
1735 gfn_t gfn = vmf->pgoff;
1736 struct kvm *kvm = vma->vm_file->private_data;
1738 addr = gfn_to_hva(kvm, gfn);
1739 if (kvm_is_error_hva(addr))
1740 return VM_FAULT_SIGBUS;
1742 npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
1744 if (unlikely(npages != 1))
1745 return VM_FAULT_SIGBUS;
1747 vmf->page = page[0];
1751 static const struct vm_operations_struct kvm_vm_vm_ops = {
1752 .fault = kvm_vm_fault,
1755 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1757 vma->vm_ops = &kvm_vm_vm_ops;
1761 static struct file_operations kvm_vm_fops = {
1762 .release = kvm_vm_release,
1763 .unlocked_ioctl = kvm_vm_ioctl,
1764 #ifdef CONFIG_COMPAT
1765 .compat_ioctl = kvm_vm_compat_ioctl,
1767 .mmap = kvm_vm_mmap,
1770 static int kvm_dev_ioctl_create_vm(void)
1775 kvm = kvm_create_vm();
1777 return PTR_ERR(kvm);
1778 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1779 r = kvm_coalesced_mmio_init(kvm);
1785 fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
1792 static long kvm_dev_ioctl_check_extension_generic(long arg)
1795 case KVM_CAP_USER_MEMORY:
1796 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
1797 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
1798 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1799 case KVM_CAP_SET_BOOT_CPU_ID:
1801 case KVM_CAP_INTERNAL_ERROR_DATA:
1803 #ifdef CONFIG_HAVE_KVM_IRQCHIP
1804 case KVM_CAP_IRQ_ROUTING:
1805 return KVM_MAX_IRQ_ROUTES;
1810 return kvm_dev_ioctl_check_extension(arg);
1813 static long kvm_dev_ioctl(struct file *filp,
1814 unsigned int ioctl, unsigned long arg)
1819 case KVM_GET_API_VERSION:
1823 r = KVM_API_VERSION;
1829 r = kvm_dev_ioctl_create_vm();
1831 case KVM_CHECK_EXTENSION:
1832 r = kvm_dev_ioctl_check_extension_generic(arg);
1834 case KVM_GET_VCPU_MMAP_SIZE:
1838 r = PAGE_SIZE; /* struct kvm_run */
1840 r += PAGE_SIZE; /* pio data page */
1842 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1843 r += PAGE_SIZE; /* coalesced mmio ring page */
1846 case KVM_TRACE_ENABLE:
1847 case KVM_TRACE_PAUSE:
1848 case KVM_TRACE_DISABLE:
1852 return kvm_arch_dev_ioctl(filp, ioctl, arg);
1858 static struct file_operations kvm_chardev_ops = {
1859 .unlocked_ioctl = kvm_dev_ioctl,
1860 .compat_ioctl = kvm_dev_ioctl,
1863 static struct miscdevice kvm_dev = {
1869 static void hardware_enable(void *junk)
1871 int cpu = raw_smp_processor_id();
1874 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
1877 cpumask_set_cpu(cpu, cpus_hardware_enabled);
1879 r = kvm_arch_hardware_enable(NULL);
1882 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
1883 atomic_inc(&hardware_enable_failed);
1884 printk(KERN_INFO "kvm: enabling virtualization on "
1885 "CPU%d failed\n", cpu);
1889 static void hardware_disable(void *junk)
1891 int cpu = raw_smp_processor_id();
1893 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
1895 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
1896 kvm_arch_hardware_disable(NULL);
1899 static void hardware_disable_all_nolock(void)
1901 BUG_ON(!kvm_usage_count);
1904 if (!kvm_usage_count)
1905 on_each_cpu(hardware_disable, NULL, 1);
1908 static void hardware_disable_all(void)
1910 spin_lock(&kvm_lock);
1911 hardware_disable_all_nolock();
1912 spin_unlock(&kvm_lock);
1915 static int hardware_enable_all(void)
1919 spin_lock(&kvm_lock);
1922 if (kvm_usage_count == 1) {
1923 atomic_set(&hardware_enable_failed, 0);
1924 on_each_cpu(hardware_enable, NULL, 1);
1926 if (atomic_read(&hardware_enable_failed)) {
1927 hardware_disable_all_nolock();
1932 spin_unlock(&kvm_lock);
1937 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
1942 if (!kvm_usage_count)
1945 val &= ~CPU_TASKS_FROZEN;
1948 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1950 hardware_disable(NULL);
1953 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
1955 smp_call_function_single(cpu, hardware_enable, NULL, 1);
1962 asmlinkage void kvm_handle_fault_on_reboot(void)
1965 /* spin while reset goes on */
1968 /* Fault while not rebooting. We want the trace. */
1971 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
1973 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
1977 * Some (well, at least mine) BIOSes hang on reboot if
1980 * And Intel TXT required VMX off for all cpu when system shutdown.
1982 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
1983 kvm_rebooting = true;
1984 on_each_cpu(hardware_disable, NULL, 1);
1988 static struct notifier_block kvm_reboot_notifier = {
1989 .notifier_call = kvm_reboot,
1993 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
1997 for (i = 0; i < bus->dev_count; i++) {
1998 struct kvm_io_device *pos = bus->devs[i];
2000 kvm_iodevice_destructor(pos);
2005 /* kvm_io_bus_write - called under kvm->slots_lock */
2006 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2007 int len, const void *val)
2010 struct kvm_io_bus *bus;
2012 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2013 for (i = 0; i < bus->dev_count; i++)
2014 if (!kvm_iodevice_write(bus->devs[i], addr, len, val))
2019 /* kvm_io_bus_read - called under kvm->slots_lock */
2020 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2024 struct kvm_io_bus *bus;
2026 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2027 for (i = 0; i < bus->dev_count; i++)
2028 if (!kvm_iodevice_read(bus->devs[i], addr, len, val))
2033 /* Caller must hold slots_lock. */
2034 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2035 struct kvm_io_device *dev)
2037 struct kvm_io_bus *new_bus, *bus;
2039 bus = kvm->buses[bus_idx];
2040 if (bus->dev_count > NR_IOBUS_DEVS-1)
2043 new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2046 memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2047 new_bus->devs[new_bus->dev_count++] = dev;
2048 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2049 synchronize_srcu_expedited(&kvm->srcu);
2055 /* Caller must hold slots_lock. */
2056 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2057 struct kvm_io_device *dev)
2060 struct kvm_io_bus *new_bus, *bus;
2062 new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2066 bus = kvm->buses[bus_idx];
2067 memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2070 for (i = 0; i < new_bus->dev_count; i++)
2071 if (new_bus->devs[i] == dev) {
2073 new_bus->devs[i] = new_bus->devs[--new_bus->dev_count];
2082 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2083 synchronize_srcu_expedited(&kvm->srcu);
2088 static struct notifier_block kvm_cpu_notifier = {
2089 .notifier_call = kvm_cpu_hotplug,
2090 .priority = 20, /* must be > scheduler priority */
2093 static int vm_stat_get(void *_offset, u64 *val)
2095 unsigned offset = (long)_offset;
2099 spin_lock(&kvm_lock);
2100 list_for_each_entry(kvm, &vm_list, vm_list)
2101 *val += *(u32 *)((void *)kvm + offset);
2102 spin_unlock(&kvm_lock);
2106 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2108 static int vcpu_stat_get(void *_offset, u64 *val)
2110 unsigned offset = (long)_offset;
2112 struct kvm_vcpu *vcpu;
2116 spin_lock(&kvm_lock);
2117 list_for_each_entry(kvm, &vm_list, vm_list)
2118 kvm_for_each_vcpu(i, vcpu, kvm)
2119 *val += *(u32 *)((void *)vcpu + offset);
2121 spin_unlock(&kvm_lock);
2125 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2127 static const struct file_operations *stat_fops[] = {
2128 [KVM_STAT_VCPU] = &vcpu_stat_fops,
2129 [KVM_STAT_VM] = &vm_stat_fops,
2132 static void kvm_init_debug(void)
2134 struct kvm_stats_debugfs_item *p;
2136 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2137 for (p = debugfs_entries; p->name; ++p)
2138 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2139 (void *)(long)p->offset,
2140 stat_fops[p->kind]);
2143 static void kvm_exit_debug(void)
2145 struct kvm_stats_debugfs_item *p;
2147 for (p = debugfs_entries; p->name; ++p)
2148 debugfs_remove(p->dentry);
2149 debugfs_remove(kvm_debugfs_dir);
2152 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2154 if (kvm_usage_count)
2155 hardware_disable(NULL);
2159 static int kvm_resume(struct sys_device *dev)
2161 if (kvm_usage_count)
2162 hardware_enable(NULL);
2166 static struct sysdev_class kvm_sysdev_class = {
2168 .suspend = kvm_suspend,
2169 .resume = kvm_resume,
2172 static struct sys_device kvm_sysdev = {
2174 .cls = &kvm_sysdev_class,
2177 struct page *bad_page;
2181 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2183 return container_of(pn, struct kvm_vcpu, preempt_notifier);
2186 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2188 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2190 kvm_arch_vcpu_load(vcpu, cpu);
2193 static void kvm_sched_out(struct preempt_notifier *pn,
2194 struct task_struct *next)
2196 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2198 kvm_arch_vcpu_put(vcpu);
2201 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2202 struct module *module)
2207 r = kvm_arch_init(opaque);
2211 bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2213 if (bad_page == NULL) {
2218 bad_pfn = page_to_pfn(bad_page);
2220 hwpoison_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2222 if (hwpoison_page == NULL) {
2227 hwpoison_pfn = page_to_pfn(hwpoison_page);
2229 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2234 r = kvm_arch_hardware_setup();
2238 for_each_online_cpu(cpu) {
2239 smp_call_function_single(cpu,
2240 kvm_arch_check_processor_compat,
2246 r = register_cpu_notifier(&kvm_cpu_notifier);
2249 register_reboot_notifier(&kvm_reboot_notifier);
2251 r = sysdev_class_register(&kvm_sysdev_class);
2255 r = sysdev_register(&kvm_sysdev);
2259 /* A kmem cache lets us meet the alignment requirements of fx_save. */
2261 vcpu_align = __alignof__(struct kvm_vcpu);
2262 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2264 if (!kvm_vcpu_cache) {
2269 kvm_chardev_ops.owner = module;
2270 kvm_vm_fops.owner = module;
2271 kvm_vcpu_fops.owner = module;
2273 r = misc_register(&kvm_dev);
2275 printk(KERN_ERR "kvm: misc device register failed\n");
2279 kvm_preempt_ops.sched_in = kvm_sched_in;
2280 kvm_preempt_ops.sched_out = kvm_sched_out;
2287 kmem_cache_destroy(kvm_vcpu_cache);
2289 sysdev_unregister(&kvm_sysdev);
2291 sysdev_class_unregister(&kvm_sysdev_class);
2293 unregister_reboot_notifier(&kvm_reboot_notifier);
2294 unregister_cpu_notifier(&kvm_cpu_notifier);
2297 kvm_arch_hardware_unsetup();
2299 free_cpumask_var(cpus_hardware_enabled);
2302 __free_page(hwpoison_page);
2303 __free_page(bad_page);
2309 EXPORT_SYMBOL_GPL(kvm_init);
2314 misc_deregister(&kvm_dev);
2315 kmem_cache_destroy(kvm_vcpu_cache);
2316 sysdev_unregister(&kvm_sysdev);
2317 sysdev_class_unregister(&kvm_sysdev_class);
2318 unregister_reboot_notifier(&kvm_reboot_notifier);
2319 unregister_cpu_notifier(&kvm_cpu_notifier);
2320 on_each_cpu(hardware_disable, NULL, 1);
2321 kvm_arch_hardware_unsetup();
2323 free_cpumask_var(cpus_hardware_enabled);
2324 __free_page(hwpoison_page);
2325 __free_page(bad_page);
2327 EXPORT_SYMBOL_GPL(kvm_exit);