2 * Kernel-based Virtual Machine driver for Linux
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
7 * Copyright (C) 2006 Qumranet, Inc.
8 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
11 * Avi Kivity <avi@qumranet.com>
12 * Yaniv Kamay <yaniv@qumranet.com>
14 * This work is licensed under the terms of the GNU GPL, version 2. See
15 * the COPYING file in the top-level directory.
21 #include <linux/kvm_host.h>
22 #include <linux/kvm.h>
23 #include <linux/module.h>
24 #include <linux/errno.h>
25 #include <linux/percpu.h>
27 #include <linux/miscdevice.h>
28 #include <linux/vmalloc.h>
29 #include <linux/reboot.h>
30 #include <linux/debugfs.h>
31 #include <linux/highmem.h>
32 #include <linux/file.h>
33 #include <linux/sysdev.h>
34 #include <linux/cpu.h>
35 #include <linux/sched.h>
36 #include <linux/cpumask.h>
37 #include <linux/smp.h>
38 #include <linux/anon_inodes.h>
39 #include <linux/profile.h>
40 #include <linux/kvm_para.h>
41 #include <linux/pagemap.h>
42 #include <linux/mman.h>
43 #include <linux/swap.h>
44 #include <linux/bitops.h>
45 #include <linux/spinlock.h>
46 #include <linux/compat.h>
47 #include <linux/srcu.h>
48 #include <linux/hugetlb.h>
49 #include <linux/slab.h>
51 #include <asm/processor.h>
53 #include <asm/uaccess.h>
54 #include <asm/pgtable.h>
55 #include <asm-generic/bitops/le.h>
57 #include "coalesced_mmio.h"
60 #define CREATE_TRACE_POINTS
61 #include <trace/events/kvm.h>
63 MODULE_AUTHOR("Qumranet");
64 MODULE_LICENSE("GPL");
69 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
72 DEFINE_SPINLOCK(kvm_lock);
75 static cpumask_var_t cpus_hardware_enabled;
76 static int kvm_usage_count = 0;
77 static atomic_t hardware_enable_failed;
79 struct kmem_cache *kvm_vcpu_cache;
80 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
82 static __read_mostly struct preempt_ops kvm_preempt_ops;
84 struct dentry *kvm_debugfs_dir;
86 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
88 static int hardware_enable_all(void);
89 static void hardware_disable_all(void);
91 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
94 EXPORT_SYMBOL_GPL(kvm_rebooting);
96 static bool largepages_enabled = true;
98 static struct page *hwpoison_page;
99 static pfn_t hwpoison_pfn;
101 static struct page *fault_page;
102 static pfn_t fault_pfn;
104 inline int kvm_is_mmio_pfn(pfn_t pfn)
106 if (pfn_valid(pfn)) {
107 struct page *page = compound_head(pfn_to_page(pfn));
108 return PageReserved(page);
115 * Switches to specified vcpu, until a matching vcpu_put()
117 void vcpu_load(struct kvm_vcpu *vcpu)
121 mutex_lock(&vcpu->mutex);
123 preempt_notifier_register(&vcpu->preempt_notifier);
124 kvm_arch_vcpu_load(vcpu, cpu);
128 void vcpu_put(struct kvm_vcpu *vcpu)
131 kvm_arch_vcpu_put(vcpu);
132 preempt_notifier_unregister(&vcpu->preempt_notifier);
134 mutex_unlock(&vcpu->mutex);
137 static void ack_flush(void *_completed)
141 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
146 struct kvm_vcpu *vcpu;
148 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
150 raw_spin_lock(&kvm->requests_lock);
151 me = smp_processor_id();
152 kvm_for_each_vcpu(i, vcpu, kvm) {
153 if (kvm_make_check_request(req, vcpu))
156 if (cpus != NULL && cpu != -1 && cpu != me)
157 cpumask_set_cpu(cpu, cpus);
159 if (unlikely(cpus == NULL))
160 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
161 else if (!cpumask_empty(cpus))
162 smp_call_function_many(cpus, ack_flush, NULL, 1);
165 raw_spin_unlock(&kvm->requests_lock);
166 free_cpumask_var(cpus);
170 void kvm_flush_remote_tlbs(struct kvm *kvm)
172 int dirty_count = kvm->tlbs_dirty;
175 if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
176 ++kvm->stat.remote_tlb_flush;
177 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
180 void kvm_reload_remote_mmus(struct kvm *kvm)
182 make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
185 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
190 mutex_init(&vcpu->mutex);
194 init_waitqueue_head(&vcpu->wq);
195 kvm_async_pf_vcpu_init(vcpu);
197 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
202 vcpu->run = page_address(page);
204 r = kvm_arch_vcpu_init(vcpu);
210 free_page((unsigned long)vcpu->run);
214 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
216 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
218 kvm_arch_vcpu_uninit(vcpu);
219 free_page((unsigned long)vcpu->run);
221 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
223 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
224 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
226 return container_of(mn, struct kvm, mmu_notifier);
229 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
230 struct mm_struct *mm,
231 unsigned long address)
233 struct kvm *kvm = mmu_notifier_to_kvm(mn);
234 int need_tlb_flush, idx;
237 * When ->invalidate_page runs, the linux pte has been zapped
238 * already but the page is still allocated until
239 * ->invalidate_page returns. So if we increase the sequence
240 * here the kvm page fault will notice if the spte can't be
241 * established because the page is going to be freed. If
242 * instead the kvm page fault establishes the spte before
243 * ->invalidate_page runs, kvm_unmap_hva will release it
246 * The sequence increase only need to be seen at spin_unlock
247 * time, and not at spin_lock time.
249 * Increasing the sequence after the spin_unlock would be
250 * unsafe because the kvm page fault could then establish the
251 * pte after kvm_unmap_hva returned, without noticing the page
252 * is going to be freed.
254 idx = srcu_read_lock(&kvm->srcu);
255 spin_lock(&kvm->mmu_lock);
256 kvm->mmu_notifier_seq++;
257 need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
258 spin_unlock(&kvm->mmu_lock);
259 srcu_read_unlock(&kvm->srcu, idx);
261 /* we've to flush the tlb before the pages can be freed */
263 kvm_flush_remote_tlbs(kvm);
267 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
268 struct mm_struct *mm,
269 unsigned long address,
272 struct kvm *kvm = mmu_notifier_to_kvm(mn);
275 idx = srcu_read_lock(&kvm->srcu);
276 spin_lock(&kvm->mmu_lock);
277 kvm->mmu_notifier_seq++;
278 kvm_set_spte_hva(kvm, address, pte);
279 spin_unlock(&kvm->mmu_lock);
280 srcu_read_unlock(&kvm->srcu, idx);
283 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
284 struct mm_struct *mm,
288 struct kvm *kvm = mmu_notifier_to_kvm(mn);
289 int need_tlb_flush = 0, idx;
291 idx = srcu_read_lock(&kvm->srcu);
292 spin_lock(&kvm->mmu_lock);
294 * The count increase must become visible at unlock time as no
295 * spte can be established without taking the mmu_lock and
296 * count is also read inside the mmu_lock critical section.
298 kvm->mmu_notifier_count++;
299 for (; start < end; start += PAGE_SIZE)
300 need_tlb_flush |= kvm_unmap_hva(kvm, start);
301 need_tlb_flush |= kvm->tlbs_dirty;
302 spin_unlock(&kvm->mmu_lock);
303 srcu_read_unlock(&kvm->srcu, idx);
305 /* we've to flush the tlb before the pages can be freed */
307 kvm_flush_remote_tlbs(kvm);
310 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
311 struct mm_struct *mm,
315 struct kvm *kvm = mmu_notifier_to_kvm(mn);
317 spin_lock(&kvm->mmu_lock);
319 * This sequence increase will notify the kvm page fault that
320 * the page that is going to be mapped in the spte could have
323 kvm->mmu_notifier_seq++;
325 * The above sequence increase must be visible before the
326 * below count decrease but both values are read by the kvm
327 * page fault under mmu_lock spinlock so we don't need to add
328 * a smb_wmb() here in between the two.
330 kvm->mmu_notifier_count--;
331 spin_unlock(&kvm->mmu_lock);
333 BUG_ON(kvm->mmu_notifier_count < 0);
336 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
337 struct mm_struct *mm,
338 unsigned long address)
340 struct kvm *kvm = mmu_notifier_to_kvm(mn);
343 idx = srcu_read_lock(&kvm->srcu);
344 spin_lock(&kvm->mmu_lock);
345 young = kvm_age_hva(kvm, address);
346 spin_unlock(&kvm->mmu_lock);
347 srcu_read_unlock(&kvm->srcu, idx);
350 kvm_flush_remote_tlbs(kvm);
355 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
356 struct mm_struct *mm)
358 struct kvm *kvm = mmu_notifier_to_kvm(mn);
361 idx = srcu_read_lock(&kvm->srcu);
362 kvm_arch_flush_shadow(kvm);
363 srcu_read_unlock(&kvm->srcu, idx);
366 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
367 .invalidate_page = kvm_mmu_notifier_invalidate_page,
368 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
369 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
370 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
371 .change_pte = kvm_mmu_notifier_change_pte,
372 .release = kvm_mmu_notifier_release,
375 static int kvm_init_mmu_notifier(struct kvm *kvm)
377 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
378 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
381 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
383 static int kvm_init_mmu_notifier(struct kvm *kvm)
388 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
390 static struct kvm *kvm_create_vm(void)
393 struct kvm *kvm = kvm_arch_alloc_vm();
396 return ERR_PTR(-ENOMEM);
398 r = kvm_arch_init_vm(kvm);
400 goto out_err_nodisable;
402 r = hardware_enable_all();
404 goto out_err_nodisable;
406 #ifdef CONFIG_HAVE_KVM_IRQCHIP
407 INIT_HLIST_HEAD(&kvm->mask_notifier_list);
408 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
412 kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
415 if (init_srcu_struct(&kvm->srcu))
417 for (i = 0; i < KVM_NR_BUSES; i++) {
418 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
424 r = kvm_init_mmu_notifier(kvm);
428 kvm->mm = current->mm;
429 atomic_inc(&kvm->mm->mm_count);
430 spin_lock_init(&kvm->mmu_lock);
431 raw_spin_lock_init(&kvm->requests_lock);
432 kvm_eventfd_init(kvm);
433 mutex_init(&kvm->lock);
434 mutex_init(&kvm->irq_lock);
435 mutex_init(&kvm->slots_lock);
436 atomic_set(&kvm->users_count, 1);
437 spin_lock(&kvm_lock);
438 list_add(&kvm->vm_list, &vm_list);
439 spin_unlock(&kvm_lock);
444 cleanup_srcu_struct(&kvm->srcu);
446 hardware_disable_all();
448 for (i = 0; i < KVM_NR_BUSES; i++)
449 kfree(kvm->buses[i]);
450 kfree(kvm->memslots);
451 kvm_arch_free_vm(kvm);
455 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
457 if (!memslot->dirty_bitmap)
460 if (2 * kvm_dirty_bitmap_bytes(memslot) > PAGE_SIZE)
461 vfree(memslot->dirty_bitmap_head);
463 kfree(memslot->dirty_bitmap_head);
465 memslot->dirty_bitmap = NULL;
466 memslot->dirty_bitmap_head = NULL;
470 * Free any memory in @free but not in @dont.
472 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
473 struct kvm_memory_slot *dont)
477 if (!dont || free->rmap != dont->rmap)
480 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
481 kvm_destroy_dirty_bitmap(free);
484 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
485 if (!dont || free->lpage_info[i] != dont->lpage_info[i]) {
486 vfree(free->lpage_info[i]);
487 free->lpage_info[i] = NULL;
495 void kvm_free_physmem(struct kvm *kvm)
498 struct kvm_memslots *slots = kvm->memslots;
500 for (i = 0; i < slots->nmemslots; ++i)
501 kvm_free_physmem_slot(&slots->memslots[i], NULL);
503 kfree(kvm->memslots);
506 static void kvm_destroy_vm(struct kvm *kvm)
509 struct mm_struct *mm = kvm->mm;
511 kvm_arch_sync_events(kvm);
512 spin_lock(&kvm_lock);
513 list_del(&kvm->vm_list);
514 spin_unlock(&kvm_lock);
515 kvm_free_irq_routing(kvm);
516 for (i = 0; i < KVM_NR_BUSES; i++)
517 kvm_io_bus_destroy(kvm->buses[i]);
518 kvm_coalesced_mmio_free(kvm);
519 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
520 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
522 kvm_arch_flush_shadow(kvm);
524 kvm_arch_destroy_vm(kvm);
525 kvm_free_physmem(kvm);
526 cleanup_srcu_struct(&kvm->srcu);
527 kvm_arch_free_vm(kvm);
528 hardware_disable_all();
532 void kvm_get_kvm(struct kvm *kvm)
534 atomic_inc(&kvm->users_count);
536 EXPORT_SYMBOL_GPL(kvm_get_kvm);
538 void kvm_put_kvm(struct kvm *kvm)
540 if (atomic_dec_and_test(&kvm->users_count))
543 EXPORT_SYMBOL_GPL(kvm_put_kvm);
546 static int kvm_vm_release(struct inode *inode, struct file *filp)
548 struct kvm *kvm = filp->private_data;
550 kvm_irqfd_release(kvm);
557 * Allocation size is twice as large as the actual dirty bitmap size.
558 * This makes it possible to do double buffering: see x86's
559 * kvm_vm_ioctl_get_dirty_log().
561 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
563 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
565 if (dirty_bytes > PAGE_SIZE)
566 memslot->dirty_bitmap = vzalloc(dirty_bytes);
568 memslot->dirty_bitmap = kzalloc(dirty_bytes, GFP_KERNEL);
570 if (!memslot->dirty_bitmap)
573 memslot->dirty_bitmap_head = memslot->dirty_bitmap;
578 * Allocate some memory and give it an address in the guest physical address
581 * Discontiguous memory is allowed, mostly for framebuffers.
583 * Must be called holding mmap_sem for write.
585 int __kvm_set_memory_region(struct kvm *kvm,
586 struct kvm_userspace_memory_region *mem,
589 int r, flush_shadow = 0;
591 unsigned long npages;
593 struct kvm_memory_slot *memslot;
594 struct kvm_memory_slot old, new;
595 struct kvm_memslots *slots, *old_memslots;
598 /* General sanity checks */
599 if (mem->memory_size & (PAGE_SIZE - 1))
601 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
603 if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
605 if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
607 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
610 memslot = &kvm->memslots->memslots[mem->slot];
611 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
612 npages = mem->memory_size >> PAGE_SHIFT;
615 if (npages > KVM_MEM_MAX_NR_PAGES)
619 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
621 new = old = *memslot;
624 new.base_gfn = base_gfn;
626 new.flags = mem->flags;
628 /* Disallow changing a memory slot's size. */
630 if (npages && old.npages && npages != old.npages)
633 /* Check for overlaps */
635 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
636 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
638 if (s == memslot || !s->npages)
640 if (!((base_gfn + npages <= s->base_gfn) ||
641 (base_gfn >= s->base_gfn + s->npages)))
645 /* Free page dirty bitmap if unneeded */
646 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
647 new.dirty_bitmap = NULL;
651 /* Allocate if a slot is being created */
653 if (npages && !new.rmap) {
654 new.rmap = vzalloc(npages * sizeof(*new.rmap));
659 new.user_alloc = user_alloc;
660 new.userspace_addr = mem->userspace_addr;
665 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
671 /* Avoid unused variable warning if no large pages */
674 if (new.lpage_info[i])
677 lpages = 1 + ((base_gfn + npages - 1)
678 >> KVM_HPAGE_GFN_SHIFT(level));
679 lpages -= base_gfn >> KVM_HPAGE_GFN_SHIFT(level);
681 new.lpage_info[i] = vzalloc(lpages * sizeof(*new.lpage_info[i]));
683 if (!new.lpage_info[i])
686 if (base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
687 new.lpage_info[i][0].write_count = 1;
688 if ((base_gfn+npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
689 new.lpage_info[i][lpages - 1].write_count = 1;
690 ugfn = new.userspace_addr >> PAGE_SHIFT;
692 * If the gfn and userspace address are not aligned wrt each
693 * other, or if explicitly asked to, disable large page
694 * support for this slot
696 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
698 for (j = 0; j < lpages; ++j)
699 new.lpage_info[i][j].write_count = 1;
704 /* Allocate page dirty bitmap if needed */
705 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
706 if (kvm_create_dirty_bitmap(&new) < 0)
708 /* destroy any largepage mappings for dirty tracking */
712 #else /* not defined CONFIG_S390 */
713 new.user_alloc = user_alloc;
715 new.userspace_addr = mem->userspace_addr;
716 #endif /* not defined CONFIG_S390 */
720 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
723 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
724 if (mem->slot >= slots->nmemslots)
725 slots->nmemslots = mem->slot + 1;
727 slots->memslots[mem->slot].flags |= KVM_MEMSLOT_INVALID;
729 old_memslots = kvm->memslots;
730 rcu_assign_pointer(kvm->memslots, slots);
731 synchronize_srcu_expedited(&kvm->srcu);
732 /* From this point no new shadow pages pointing to a deleted
733 * memslot will be created.
735 * validation of sp->gfn happens in:
736 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
737 * - kvm_is_visible_gfn (mmu_check_roots)
739 kvm_arch_flush_shadow(kvm);
743 r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
747 /* map the pages in iommu page table */
749 r = kvm_iommu_map_pages(kvm, &new);
755 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
758 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
759 if (mem->slot >= slots->nmemslots)
760 slots->nmemslots = mem->slot + 1;
763 /* actual memory is freed via old in kvm_free_physmem_slot below */
766 new.dirty_bitmap = NULL;
767 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i)
768 new.lpage_info[i] = NULL;
771 slots->memslots[mem->slot] = new;
772 old_memslots = kvm->memslots;
773 rcu_assign_pointer(kvm->memslots, slots);
774 synchronize_srcu_expedited(&kvm->srcu);
776 kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
778 kvm_free_physmem_slot(&old, &new);
782 kvm_arch_flush_shadow(kvm);
787 kvm_free_physmem_slot(&new, &old);
792 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
794 int kvm_set_memory_region(struct kvm *kvm,
795 struct kvm_userspace_memory_region *mem,
800 mutex_lock(&kvm->slots_lock);
801 r = __kvm_set_memory_region(kvm, mem, user_alloc);
802 mutex_unlock(&kvm->slots_lock);
805 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
807 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
809 kvm_userspace_memory_region *mem,
812 if (mem->slot >= KVM_MEMORY_SLOTS)
814 return kvm_set_memory_region(kvm, mem, user_alloc);
817 int kvm_get_dirty_log(struct kvm *kvm,
818 struct kvm_dirty_log *log, int *is_dirty)
820 struct kvm_memory_slot *memslot;
823 unsigned long any = 0;
826 if (log->slot >= KVM_MEMORY_SLOTS)
829 memslot = &kvm->memslots->memslots[log->slot];
831 if (!memslot->dirty_bitmap)
834 n = kvm_dirty_bitmap_bytes(memslot);
836 for (i = 0; !any && i < n/sizeof(long); ++i)
837 any = memslot->dirty_bitmap[i];
840 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
851 void kvm_disable_largepages(void)
853 largepages_enabled = false;
855 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
857 int is_error_page(struct page *page)
859 return page == bad_page || page == hwpoison_page || page == fault_page;
861 EXPORT_SYMBOL_GPL(is_error_page);
863 int is_error_pfn(pfn_t pfn)
865 return pfn == bad_pfn || pfn == hwpoison_pfn || pfn == fault_pfn;
867 EXPORT_SYMBOL_GPL(is_error_pfn);
869 int is_hwpoison_pfn(pfn_t pfn)
871 return pfn == hwpoison_pfn;
873 EXPORT_SYMBOL_GPL(is_hwpoison_pfn);
875 int is_fault_pfn(pfn_t pfn)
877 return pfn == fault_pfn;
879 EXPORT_SYMBOL_GPL(is_fault_pfn);
881 static inline unsigned long bad_hva(void)
886 int kvm_is_error_hva(unsigned long addr)
888 return addr == bad_hva();
890 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
892 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm_memslots *slots,
897 for (i = 0; i < slots->nmemslots; ++i) {
898 struct kvm_memory_slot *memslot = &slots->memslots[i];
900 if (gfn >= memslot->base_gfn
901 && gfn < memslot->base_gfn + memslot->npages)
907 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
909 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
911 EXPORT_SYMBOL_GPL(gfn_to_memslot);
913 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
916 struct kvm_memslots *slots = kvm_memslots(kvm);
918 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
919 struct kvm_memory_slot *memslot = &slots->memslots[i];
921 if (memslot->flags & KVM_MEMSLOT_INVALID)
924 if (gfn >= memslot->base_gfn
925 && gfn < memslot->base_gfn + memslot->npages)
930 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
932 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
934 struct vm_area_struct *vma;
935 unsigned long addr, size;
939 addr = gfn_to_hva(kvm, gfn);
940 if (kvm_is_error_hva(addr))
943 down_read(¤t->mm->mmap_sem);
944 vma = find_vma(current->mm, addr);
948 size = vma_kernel_pagesize(vma);
951 up_read(¤t->mm->mmap_sem);
956 int memslot_id(struct kvm *kvm, gfn_t gfn)
959 struct kvm_memslots *slots = kvm_memslots(kvm);
960 struct kvm_memory_slot *memslot = NULL;
962 for (i = 0; i < slots->nmemslots; ++i) {
963 memslot = &slots->memslots[i];
965 if (gfn >= memslot->base_gfn
966 && gfn < memslot->base_gfn + memslot->npages)
970 return memslot - slots->memslots;
973 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
976 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
980 *nr_pages = slot->npages - (gfn - slot->base_gfn);
982 return gfn_to_hva_memslot(slot, gfn);
985 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
987 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
989 EXPORT_SYMBOL_GPL(gfn_to_hva);
991 static pfn_t get_fault_pfn(void)
993 get_page(fault_page);
997 static pfn_t hva_to_pfn(struct kvm *kvm, unsigned long addr, bool atomic,
998 bool *async, bool write_fault, bool *writable)
1000 struct page *page[1];
1004 /* we can do it either atomically or asynchronously, not both */
1005 BUG_ON(atomic && async);
1007 BUG_ON(!write_fault && !writable);
1012 if (atomic || async)
1013 npages = __get_user_pages_fast(addr, 1, 1, page);
1015 if (unlikely(npages != 1) && !atomic) {
1019 *writable = write_fault;
1021 npages = get_user_pages_fast(addr, 1, write_fault, page);
1023 /* map read fault as writable if possible */
1024 if (unlikely(!write_fault) && npages == 1) {
1025 struct page *wpage[1];
1027 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1037 if (unlikely(npages != 1)) {
1038 struct vm_area_struct *vma;
1041 return get_fault_pfn();
1043 down_read(¤t->mm->mmap_sem);
1044 if (is_hwpoison_address(addr)) {
1045 up_read(¤t->mm->mmap_sem);
1046 get_page(hwpoison_page);
1047 return page_to_pfn(hwpoison_page);
1050 vma = find_vma_intersection(current->mm, addr, addr+1);
1053 pfn = get_fault_pfn();
1054 else if ((vma->vm_flags & VM_PFNMAP)) {
1055 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1057 BUG_ON(!kvm_is_mmio_pfn(pfn));
1059 if (async && (vma->vm_flags & VM_WRITE))
1061 pfn = get_fault_pfn();
1063 up_read(¤t->mm->mmap_sem);
1065 pfn = page_to_pfn(page[0]);
1070 pfn_t hva_to_pfn_atomic(struct kvm *kvm, unsigned long addr)
1072 return hva_to_pfn(kvm, addr, true, NULL, true, NULL);
1074 EXPORT_SYMBOL_GPL(hva_to_pfn_atomic);
1076 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1077 bool write_fault, bool *writable)
1084 addr = gfn_to_hva(kvm, gfn);
1085 if (kvm_is_error_hva(addr)) {
1087 return page_to_pfn(bad_page);
1090 return hva_to_pfn(kvm, addr, atomic, async, write_fault, writable);
1093 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1095 return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1097 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1099 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1100 bool write_fault, bool *writable)
1102 return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1104 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1106 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1108 return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1110 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1112 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1115 return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1117 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1119 pfn_t gfn_to_pfn_memslot(struct kvm *kvm,
1120 struct kvm_memory_slot *slot, gfn_t gfn)
1122 unsigned long addr = gfn_to_hva_memslot(slot, gfn);
1123 return hva_to_pfn(kvm, addr, false, NULL, true, NULL);
1126 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1132 addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1133 if (kvm_is_error_hva(addr))
1136 if (entry < nr_pages)
1139 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1141 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1143 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1147 pfn = gfn_to_pfn(kvm, gfn);
1148 if (!kvm_is_mmio_pfn(pfn))
1149 return pfn_to_page(pfn);
1151 WARN_ON(kvm_is_mmio_pfn(pfn));
1157 EXPORT_SYMBOL_GPL(gfn_to_page);
1159 void kvm_release_page_clean(struct page *page)
1161 kvm_release_pfn_clean(page_to_pfn(page));
1163 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1165 void kvm_release_pfn_clean(pfn_t pfn)
1167 if (!kvm_is_mmio_pfn(pfn))
1168 put_page(pfn_to_page(pfn));
1170 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1172 void kvm_release_page_dirty(struct page *page)
1174 kvm_release_pfn_dirty(page_to_pfn(page));
1176 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1178 void kvm_release_pfn_dirty(pfn_t pfn)
1180 kvm_set_pfn_dirty(pfn);
1181 kvm_release_pfn_clean(pfn);
1183 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1185 void kvm_set_page_dirty(struct page *page)
1187 kvm_set_pfn_dirty(page_to_pfn(page));
1189 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1191 void kvm_set_pfn_dirty(pfn_t pfn)
1193 if (!kvm_is_mmio_pfn(pfn)) {
1194 struct page *page = pfn_to_page(pfn);
1195 if (!PageReserved(page))
1199 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1201 void kvm_set_pfn_accessed(pfn_t pfn)
1203 if (!kvm_is_mmio_pfn(pfn))
1204 mark_page_accessed(pfn_to_page(pfn));
1206 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1208 void kvm_get_pfn(pfn_t pfn)
1210 if (!kvm_is_mmio_pfn(pfn))
1211 get_page(pfn_to_page(pfn));
1213 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1215 static int next_segment(unsigned long len, int offset)
1217 if (len > PAGE_SIZE - offset)
1218 return PAGE_SIZE - offset;
1223 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1229 addr = gfn_to_hva(kvm, gfn);
1230 if (kvm_is_error_hva(addr))
1232 r = copy_from_user(data, (void __user *)addr + offset, len);
1237 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1239 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1241 gfn_t gfn = gpa >> PAGE_SHIFT;
1243 int offset = offset_in_page(gpa);
1246 while ((seg = next_segment(len, offset)) != 0) {
1247 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1257 EXPORT_SYMBOL_GPL(kvm_read_guest);
1259 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1264 gfn_t gfn = gpa >> PAGE_SHIFT;
1265 int offset = offset_in_page(gpa);
1267 addr = gfn_to_hva(kvm, gfn);
1268 if (kvm_is_error_hva(addr))
1270 pagefault_disable();
1271 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1277 EXPORT_SYMBOL(kvm_read_guest_atomic);
1279 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1280 int offset, int len)
1285 addr = gfn_to_hva(kvm, gfn);
1286 if (kvm_is_error_hva(addr))
1288 r = copy_to_user((void __user *)addr + offset, data, len);
1291 mark_page_dirty(kvm, gfn);
1294 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1296 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1299 gfn_t gfn = gpa >> PAGE_SHIFT;
1301 int offset = offset_in_page(gpa);
1304 while ((seg = next_segment(len, offset)) != 0) {
1305 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1316 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1319 struct kvm_memslots *slots = kvm_memslots(kvm);
1320 int offset = offset_in_page(gpa);
1321 gfn_t gfn = gpa >> PAGE_SHIFT;
1324 ghc->generation = slots->generation;
1325 ghc->memslot = __gfn_to_memslot(slots, gfn);
1326 ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1327 if (!kvm_is_error_hva(ghc->hva))
1334 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1336 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1337 void *data, unsigned long len)
1339 struct kvm_memslots *slots = kvm_memslots(kvm);
1342 if (slots->generation != ghc->generation)
1343 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1345 if (kvm_is_error_hva(ghc->hva))
1348 r = copy_to_user((void __user *)ghc->hva, data, len);
1351 mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1355 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1357 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1359 return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1362 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1364 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1366 gfn_t gfn = gpa >> PAGE_SHIFT;
1368 int offset = offset_in_page(gpa);
1371 while ((seg = next_segment(len, offset)) != 0) {
1372 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1381 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1383 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1386 if (memslot && memslot->dirty_bitmap) {
1387 unsigned long rel_gfn = gfn - memslot->base_gfn;
1389 generic___set_le_bit(rel_gfn, memslot->dirty_bitmap);
1393 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1395 struct kvm_memory_slot *memslot;
1397 memslot = gfn_to_memslot(kvm, gfn);
1398 mark_page_dirty_in_slot(kvm, memslot, gfn);
1402 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1404 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1409 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1411 if (kvm_arch_vcpu_runnable(vcpu)) {
1412 kvm_make_request(KVM_REQ_UNHALT, vcpu);
1415 if (kvm_cpu_has_pending_timer(vcpu))
1417 if (signal_pending(current))
1423 finish_wait(&vcpu->wq, &wait);
1426 void kvm_resched(struct kvm_vcpu *vcpu)
1428 if (!need_resched())
1432 EXPORT_SYMBOL_GPL(kvm_resched);
1434 void kvm_vcpu_on_spin(struct kvm_vcpu *vcpu)
1439 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1441 /* Sleep for 100 us, and hope lock-holder got scheduled */
1442 expires = ktime_add_ns(ktime_get(), 100000UL);
1443 schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
1445 finish_wait(&vcpu->wq, &wait);
1447 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1449 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1451 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1454 if (vmf->pgoff == 0)
1455 page = virt_to_page(vcpu->run);
1457 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1458 page = virt_to_page(vcpu->arch.pio_data);
1460 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1461 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1462 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1465 return VM_FAULT_SIGBUS;
1471 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1472 .fault = kvm_vcpu_fault,
1475 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1477 vma->vm_ops = &kvm_vcpu_vm_ops;
1481 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1483 struct kvm_vcpu *vcpu = filp->private_data;
1485 kvm_put_kvm(vcpu->kvm);
1489 static struct file_operations kvm_vcpu_fops = {
1490 .release = kvm_vcpu_release,
1491 .unlocked_ioctl = kvm_vcpu_ioctl,
1492 .compat_ioctl = kvm_vcpu_ioctl,
1493 .mmap = kvm_vcpu_mmap,
1494 .llseek = noop_llseek,
1498 * Allocates an inode for the vcpu.
1500 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1502 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1506 * Creates some virtual cpus. Good luck creating more than one.
1508 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1511 struct kvm_vcpu *vcpu, *v;
1513 vcpu = kvm_arch_vcpu_create(kvm, id);
1515 return PTR_ERR(vcpu);
1517 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1519 r = kvm_arch_vcpu_setup(vcpu);
1523 mutex_lock(&kvm->lock);
1524 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1529 kvm_for_each_vcpu(r, v, kvm)
1530 if (v->vcpu_id == id) {
1535 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1537 /* Now it's all set up, let userspace reach it */
1539 r = create_vcpu_fd(vcpu);
1545 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1547 atomic_inc(&kvm->online_vcpus);
1549 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1550 if (kvm->bsp_vcpu_id == id)
1551 kvm->bsp_vcpu = vcpu;
1553 mutex_unlock(&kvm->lock);
1557 mutex_unlock(&kvm->lock);
1558 kvm_arch_vcpu_destroy(vcpu);
1562 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1565 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1566 vcpu->sigset_active = 1;
1567 vcpu->sigset = *sigset;
1569 vcpu->sigset_active = 0;
1573 static long kvm_vcpu_ioctl(struct file *filp,
1574 unsigned int ioctl, unsigned long arg)
1576 struct kvm_vcpu *vcpu = filp->private_data;
1577 void __user *argp = (void __user *)arg;
1579 struct kvm_fpu *fpu = NULL;
1580 struct kvm_sregs *kvm_sregs = NULL;
1582 if (vcpu->kvm->mm != current->mm)
1585 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1587 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1588 * so vcpu_load() would break it.
1590 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1591 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1601 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1602 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1604 case KVM_GET_REGS: {
1605 struct kvm_regs *kvm_regs;
1608 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1611 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1615 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1622 case KVM_SET_REGS: {
1623 struct kvm_regs *kvm_regs;
1626 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1630 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1632 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1640 case KVM_GET_SREGS: {
1641 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1645 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1649 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1654 case KVM_SET_SREGS: {
1655 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1660 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1662 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1668 case KVM_GET_MP_STATE: {
1669 struct kvm_mp_state mp_state;
1671 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1675 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1680 case KVM_SET_MP_STATE: {
1681 struct kvm_mp_state mp_state;
1684 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1686 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1692 case KVM_TRANSLATE: {
1693 struct kvm_translation tr;
1696 if (copy_from_user(&tr, argp, sizeof tr))
1698 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1702 if (copy_to_user(argp, &tr, sizeof tr))
1707 case KVM_SET_GUEST_DEBUG: {
1708 struct kvm_guest_debug dbg;
1711 if (copy_from_user(&dbg, argp, sizeof dbg))
1713 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1719 case KVM_SET_SIGNAL_MASK: {
1720 struct kvm_signal_mask __user *sigmask_arg = argp;
1721 struct kvm_signal_mask kvm_sigmask;
1722 sigset_t sigset, *p;
1727 if (copy_from_user(&kvm_sigmask, argp,
1728 sizeof kvm_sigmask))
1731 if (kvm_sigmask.len != sizeof sigset)
1734 if (copy_from_user(&sigset, sigmask_arg->sigset,
1739 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
1743 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1747 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1751 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1757 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1762 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1764 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1771 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1780 static long kvm_vm_ioctl(struct file *filp,
1781 unsigned int ioctl, unsigned long arg)
1783 struct kvm *kvm = filp->private_data;
1784 void __user *argp = (void __user *)arg;
1787 if (kvm->mm != current->mm)
1790 case KVM_CREATE_VCPU:
1791 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1795 case KVM_SET_USER_MEMORY_REGION: {
1796 struct kvm_userspace_memory_region kvm_userspace_mem;
1799 if (copy_from_user(&kvm_userspace_mem, argp,
1800 sizeof kvm_userspace_mem))
1803 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1808 case KVM_GET_DIRTY_LOG: {
1809 struct kvm_dirty_log log;
1812 if (copy_from_user(&log, argp, sizeof log))
1814 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1819 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1820 case KVM_REGISTER_COALESCED_MMIO: {
1821 struct kvm_coalesced_mmio_zone zone;
1823 if (copy_from_user(&zone, argp, sizeof zone))
1825 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1831 case KVM_UNREGISTER_COALESCED_MMIO: {
1832 struct kvm_coalesced_mmio_zone zone;
1834 if (copy_from_user(&zone, argp, sizeof zone))
1836 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1844 struct kvm_irqfd data;
1847 if (copy_from_user(&data, argp, sizeof data))
1849 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
1852 case KVM_IOEVENTFD: {
1853 struct kvm_ioeventfd data;
1856 if (copy_from_user(&data, argp, sizeof data))
1858 r = kvm_ioeventfd(kvm, &data);
1861 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1862 case KVM_SET_BOOT_CPU_ID:
1864 mutex_lock(&kvm->lock);
1865 if (atomic_read(&kvm->online_vcpus) != 0)
1868 kvm->bsp_vcpu_id = arg;
1869 mutex_unlock(&kvm->lock);
1873 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1875 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
1881 #ifdef CONFIG_COMPAT
1882 struct compat_kvm_dirty_log {
1886 compat_uptr_t dirty_bitmap; /* one bit per page */
1891 static long kvm_vm_compat_ioctl(struct file *filp,
1892 unsigned int ioctl, unsigned long arg)
1894 struct kvm *kvm = filp->private_data;
1897 if (kvm->mm != current->mm)
1900 case KVM_GET_DIRTY_LOG: {
1901 struct compat_kvm_dirty_log compat_log;
1902 struct kvm_dirty_log log;
1905 if (copy_from_user(&compat_log, (void __user *)arg,
1906 sizeof(compat_log)))
1908 log.slot = compat_log.slot;
1909 log.padding1 = compat_log.padding1;
1910 log.padding2 = compat_log.padding2;
1911 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
1913 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1919 r = kvm_vm_ioctl(filp, ioctl, arg);
1927 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1929 struct page *page[1];
1932 gfn_t gfn = vmf->pgoff;
1933 struct kvm *kvm = vma->vm_file->private_data;
1935 addr = gfn_to_hva(kvm, gfn);
1936 if (kvm_is_error_hva(addr))
1937 return VM_FAULT_SIGBUS;
1939 npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
1941 if (unlikely(npages != 1))
1942 return VM_FAULT_SIGBUS;
1944 vmf->page = page[0];
1948 static const struct vm_operations_struct kvm_vm_vm_ops = {
1949 .fault = kvm_vm_fault,
1952 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1954 vma->vm_ops = &kvm_vm_vm_ops;
1958 static struct file_operations kvm_vm_fops = {
1959 .release = kvm_vm_release,
1960 .unlocked_ioctl = kvm_vm_ioctl,
1961 #ifdef CONFIG_COMPAT
1962 .compat_ioctl = kvm_vm_compat_ioctl,
1964 .mmap = kvm_vm_mmap,
1965 .llseek = noop_llseek,
1968 static int kvm_dev_ioctl_create_vm(void)
1973 kvm = kvm_create_vm();
1975 return PTR_ERR(kvm);
1976 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1977 r = kvm_coalesced_mmio_init(kvm);
1983 r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
1990 static long kvm_dev_ioctl_check_extension_generic(long arg)
1993 case KVM_CAP_USER_MEMORY:
1994 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
1995 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
1996 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1997 case KVM_CAP_SET_BOOT_CPU_ID:
1999 case KVM_CAP_INTERNAL_ERROR_DATA:
2001 #ifdef CONFIG_HAVE_KVM_IRQCHIP
2002 case KVM_CAP_IRQ_ROUTING:
2003 return KVM_MAX_IRQ_ROUTES;
2008 return kvm_dev_ioctl_check_extension(arg);
2011 static long kvm_dev_ioctl(struct file *filp,
2012 unsigned int ioctl, unsigned long arg)
2017 case KVM_GET_API_VERSION:
2021 r = KVM_API_VERSION;
2027 r = kvm_dev_ioctl_create_vm();
2029 case KVM_CHECK_EXTENSION:
2030 r = kvm_dev_ioctl_check_extension_generic(arg);
2032 case KVM_GET_VCPU_MMAP_SIZE:
2036 r = PAGE_SIZE; /* struct kvm_run */
2038 r += PAGE_SIZE; /* pio data page */
2040 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2041 r += PAGE_SIZE; /* coalesced mmio ring page */
2044 case KVM_TRACE_ENABLE:
2045 case KVM_TRACE_PAUSE:
2046 case KVM_TRACE_DISABLE:
2050 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2056 static struct file_operations kvm_chardev_ops = {
2057 .unlocked_ioctl = kvm_dev_ioctl,
2058 .compat_ioctl = kvm_dev_ioctl,
2059 .llseek = noop_llseek,
2062 static struct miscdevice kvm_dev = {
2068 static void hardware_enable_nolock(void *junk)
2070 int cpu = raw_smp_processor_id();
2073 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2076 cpumask_set_cpu(cpu, cpus_hardware_enabled);
2078 r = kvm_arch_hardware_enable(NULL);
2081 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2082 atomic_inc(&hardware_enable_failed);
2083 printk(KERN_INFO "kvm: enabling virtualization on "
2084 "CPU%d failed\n", cpu);
2088 static void hardware_enable(void *junk)
2090 spin_lock(&kvm_lock);
2091 hardware_enable_nolock(junk);
2092 spin_unlock(&kvm_lock);
2095 static void hardware_disable_nolock(void *junk)
2097 int cpu = raw_smp_processor_id();
2099 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2101 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2102 kvm_arch_hardware_disable(NULL);
2105 static void hardware_disable(void *junk)
2107 spin_lock(&kvm_lock);
2108 hardware_disable_nolock(junk);
2109 spin_unlock(&kvm_lock);
2112 static void hardware_disable_all_nolock(void)
2114 BUG_ON(!kvm_usage_count);
2117 if (!kvm_usage_count)
2118 on_each_cpu(hardware_disable_nolock, NULL, 1);
2121 static void hardware_disable_all(void)
2123 spin_lock(&kvm_lock);
2124 hardware_disable_all_nolock();
2125 spin_unlock(&kvm_lock);
2128 static int hardware_enable_all(void)
2132 spin_lock(&kvm_lock);
2135 if (kvm_usage_count == 1) {
2136 atomic_set(&hardware_enable_failed, 0);
2137 on_each_cpu(hardware_enable_nolock, NULL, 1);
2139 if (atomic_read(&hardware_enable_failed)) {
2140 hardware_disable_all_nolock();
2145 spin_unlock(&kvm_lock);
2150 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2155 if (!kvm_usage_count)
2158 val &= ~CPU_TASKS_FROZEN;
2161 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2163 hardware_disable(NULL);
2166 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2168 hardware_enable(NULL);
2175 asmlinkage void kvm_spurious_fault(void)
2177 /* Fault while not rebooting. We want the trace. */
2180 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
2182 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2186 * Some (well, at least mine) BIOSes hang on reboot if
2189 * And Intel TXT required VMX off for all cpu when system shutdown.
2191 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2192 kvm_rebooting = true;
2193 on_each_cpu(hardware_disable_nolock, NULL, 1);
2197 static struct notifier_block kvm_reboot_notifier = {
2198 .notifier_call = kvm_reboot,
2202 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2206 for (i = 0; i < bus->dev_count; i++) {
2207 struct kvm_io_device *pos = bus->devs[i];
2209 kvm_iodevice_destructor(pos);
2214 /* kvm_io_bus_write - called under kvm->slots_lock */
2215 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2216 int len, const void *val)
2219 struct kvm_io_bus *bus;
2221 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2222 for (i = 0; i < bus->dev_count; i++)
2223 if (!kvm_iodevice_write(bus->devs[i], addr, len, val))
2228 /* kvm_io_bus_read - called under kvm->slots_lock */
2229 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2233 struct kvm_io_bus *bus;
2235 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2236 for (i = 0; i < bus->dev_count; i++)
2237 if (!kvm_iodevice_read(bus->devs[i], addr, len, val))
2242 /* Caller must hold slots_lock. */
2243 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2244 struct kvm_io_device *dev)
2246 struct kvm_io_bus *new_bus, *bus;
2248 bus = kvm->buses[bus_idx];
2249 if (bus->dev_count > NR_IOBUS_DEVS-1)
2252 new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2255 memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2256 new_bus->devs[new_bus->dev_count++] = dev;
2257 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2258 synchronize_srcu_expedited(&kvm->srcu);
2264 /* Caller must hold slots_lock. */
2265 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2266 struct kvm_io_device *dev)
2269 struct kvm_io_bus *new_bus, *bus;
2271 new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2275 bus = kvm->buses[bus_idx];
2276 memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2279 for (i = 0; i < new_bus->dev_count; i++)
2280 if (new_bus->devs[i] == dev) {
2282 new_bus->devs[i] = new_bus->devs[--new_bus->dev_count];
2291 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2292 synchronize_srcu_expedited(&kvm->srcu);
2297 static struct notifier_block kvm_cpu_notifier = {
2298 .notifier_call = kvm_cpu_hotplug,
2301 static int vm_stat_get(void *_offset, u64 *val)
2303 unsigned offset = (long)_offset;
2307 spin_lock(&kvm_lock);
2308 list_for_each_entry(kvm, &vm_list, vm_list)
2309 *val += *(u32 *)((void *)kvm + offset);
2310 spin_unlock(&kvm_lock);
2314 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2316 static int vcpu_stat_get(void *_offset, u64 *val)
2318 unsigned offset = (long)_offset;
2320 struct kvm_vcpu *vcpu;
2324 spin_lock(&kvm_lock);
2325 list_for_each_entry(kvm, &vm_list, vm_list)
2326 kvm_for_each_vcpu(i, vcpu, kvm)
2327 *val += *(u32 *)((void *)vcpu + offset);
2329 spin_unlock(&kvm_lock);
2333 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2335 static const struct file_operations *stat_fops[] = {
2336 [KVM_STAT_VCPU] = &vcpu_stat_fops,
2337 [KVM_STAT_VM] = &vm_stat_fops,
2340 static void kvm_init_debug(void)
2342 struct kvm_stats_debugfs_item *p;
2344 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2345 for (p = debugfs_entries; p->name; ++p)
2346 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2347 (void *)(long)p->offset,
2348 stat_fops[p->kind]);
2351 static void kvm_exit_debug(void)
2353 struct kvm_stats_debugfs_item *p;
2355 for (p = debugfs_entries; p->name; ++p)
2356 debugfs_remove(p->dentry);
2357 debugfs_remove(kvm_debugfs_dir);
2360 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2362 if (kvm_usage_count)
2363 hardware_disable_nolock(NULL);
2367 static int kvm_resume(struct sys_device *dev)
2369 if (kvm_usage_count) {
2370 WARN_ON(spin_is_locked(&kvm_lock));
2371 hardware_enable_nolock(NULL);
2376 static struct sysdev_class kvm_sysdev_class = {
2378 .suspend = kvm_suspend,
2379 .resume = kvm_resume,
2382 static struct sys_device kvm_sysdev = {
2384 .cls = &kvm_sysdev_class,
2387 struct page *bad_page;
2391 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2393 return container_of(pn, struct kvm_vcpu, preempt_notifier);
2396 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2398 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2400 kvm_arch_vcpu_load(vcpu, cpu);
2403 static void kvm_sched_out(struct preempt_notifier *pn,
2404 struct task_struct *next)
2406 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2408 kvm_arch_vcpu_put(vcpu);
2411 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2412 struct module *module)
2417 r = kvm_arch_init(opaque);
2421 bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2423 if (bad_page == NULL) {
2428 bad_pfn = page_to_pfn(bad_page);
2430 hwpoison_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2432 if (hwpoison_page == NULL) {
2437 hwpoison_pfn = page_to_pfn(hwpoison_page);
2439 fault_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2441 if (fault_page == NULL) {
2446 fault_pfn = page_to_pfn(fault_page);
2448 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2453 r = kvm_arch_hardware_setup();
2457 for_each_online_cpu(cpu) {
2458 smp_call_function_single(cpu,
2459 kvm_arch_check_processor_compat,
2465 r = register_cpu_notifier(&kvm_cpu_notifier);
2468 register_reboot_notifier(&kvm_reboot_notifier);
2470 r = sysdev_class_register(&kvm_sysdev_class);
2474 r = sysdev_register(&kvm_sysdev);
2478 /* A kmem cache lets us meet the alignment requirements of fx_save. */
2480 vcpu_align = __alignof__(struct kvm_vcpu);
2481 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2483 if (!kvm_vcpu_cache) {
2488 r = kvm_async_pf_init();
2492 kvm_chardev_ops.owner = module;
2493 kvm_vm_fops.owner = module;
2494 kvm_vcpu_fops.owner = module;
2496 r = misc_register(&kvm_dev);
2498 printk(KERN_ERR "kvm: misc device register failed\n");
2502 kvm_preempt_ops.sched_in = kvm_sched_in;
2503 kvm_preempt_ops.sched_out = kvm_sched_out;
2510 kvm_async_pf_deinit();
2512 kmem_cache_destroy(kvm_vcpu_cache);
2514 sysdev_unregister(&kvm_sysdev);
2516 sysdev_class_unregister(&kvm_sysdev_class);
2518 unregister_reboot_notifier(&kvm_reboot_notifier);
2519 unregister_cpu_notifier(&kvm_cpu_notifier);
2522 kvm_arch_hardware_unsetup();
2524 free_cpumask_var(cpus_hardware_enabled);
2527 __free_page(fault_page);
2529 __free_page(hwpoison_page);
2530 __free_page(bad_page);
2536 EXPORT_SYMBOL_GPL(kvm_init);
2541 misc_deregister(&kvm_dev);
2542 kmem_cache_destroy(kvm_vcpu_cache);
2543 kvm_async_pf_deinit();
2544 sysdev_unregister(&kvm_sysdev);
2545 sysdev_class_unregister(&kvm_sysdev_class);
2546 unregister_reboot_notifier(&kvm_reboot_notifier);
2547 unregister_cpu_notifier(&kvm_cpu_notifier);
2548 on_each_cpu(hardware_disable_nolock, NULL, 1);
2549 kvm_arch_hardware_unsetup();
2551 free_cpumask_var(cpus_hardware_enabled);
2552 __free_page(hwpoison_page);
2553 __free_page(bad_page);
2555 EXPORT_SYMBOL_GPL(kvm_exit);