2 * Kernel-based Virtual Machine driver for Linux
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
7 * Copyright (C) 2006 Qumranet, Inc.
8 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
11 * Avi Kivity <avi@qumranet.com>
12 * Yaniv Kamay <yaniv@qumranet.com>
14 * This work is licensed under the terms of the GNU GPL, version 2. See
15 * the COPYING file in the top-level directory.
21 #include <linux/kvm_host.h>
22 #include <linux/kvm.h>
23 #include <linux/module.h>
24 #include <linux/errno.h>
25 #include <linux/percpu.h>
27 #include <linux/miscdevice.h>
28 #include <linux/vmalloc.h>
29 #include <linux/reboot.h>
30 #include <linux/debugfs.h>
31 #include <linux/highmem.h>
32 #include <linux/file.h>
33 #include <linux/sysdev.h>
34 #include <linux/cpu.h>
35 #include <linux/sched.h>
36 #include <linux/cpumask.h>
37 #include <linux/smp.h>
38 #include <linux/anon_inodes.h>
39 #include <linux/profile.h>
40 #include <linux/kvm_para.h>
41 #include <linux/pagemap.h>
42 #include <linux/mman.h>
43 #include <linux/swap.h>
44 #include <linux/bitops.h>
45 #include <linux/spinlock.h>
46 #include <linux/compat.h>
47 #include <linux/srcu.h>
48 #include <linux/hugetlb.h>
49 #include <linux/slab.h>
51 #include <asm/processor.h>
53 #include <asm/uaccess.h>
54 #include <asm/pgtable.h>
55 #include <asm-generic/bitops/le.h>
57 #include "coalesced_mmio.h"
60 #define CREATE_TRACE_POINTS
61 #include <trace/events/kvm.h>
63 MODULE_AUTHOR("Qumranet");
64 MODULE_LICENSE("GPL");
69 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
72 DEFINE_SPINLOCK(kvm_lock);
75 static cpumask_var_t cpus_hardware_enabled;
76 static int kvm_usage_count = 0;
77 static atomic_t hardware_enable_failed;
79 struct kmem_cache *kvm_vcpu_cache;
80 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
82 static __read_mostly struct preempt_ops kvm_preempt_ops;
84 struct dentry *kvm_debugfs_dir;
86 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
88 static int hardware_enable_all(void);
89 static void hardware_disable_all(void);
91 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
93 static bool kvm_rebooting;
95 static bool largepages_enabled = true;
97 static struct page *hwpoison_page;
98 static pfn_t hwpoison_pfn;
100 static struct page *fault_page;
101 static pfn_t fault_pfn;
103 inline int kvm_is_mmio_pfn(pfn_t pfn)
105 if (pfn_valid(pfn)) {
106 struct page *page = compound_head(pfn_to_page(pfn));
107 return PageReserved(page);
114 * Switches to specified vcpu, until a matching vcpu_put()
116 void vcpu_load(struct kvm_vcpu *vcpu)
120 mutex_lock(&vcpu->mutex);
122 preempt_notifier_register(&vcpu->preempt_notifier);
123 kvm_arch_vcpu_load(vcpu, cpu);
127 void vcpu_put(struct kvm_vcpu *vcpu)
130 kvm_arch_vcpu_put(vcpu);
131 preempt_notifier_unregister(&vcpu->preempt_notifier);
133 mutex_unlock(&vcpu->mutex);
136 static void ack_flush(void *_completed)
140 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
145 struct kvm_vcpu *vcpu;
147 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
149 raw_spin_lock(&kvm->requests_lock);
150 me = smp_processor_id();
151 kvm_for_each_vcpu(i, vcpu, kvm) {
152 if (kvm_make_check_request(req, vcpu))
155 if (cpus != NULL && cpu != -1 && cpu != me)
156 cpumask_set_cpu(cpu, cpus);
158 if (unlikely(cpus == NULL))
159 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
160 else if (!cpumask_empty(cpus))
161 smp_call_function_many(cpus, ack_flush, NULL, 1);
164 raw_spin_unlock(&kvm->requests_lock);
165 free_cpumask_var(cpus);
169 void kvm_flush_remote_tlbs(struct kvm *kvm)
171 if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
172 ++kvm->stat.remote_tlb_flush;
175 void kvm_reload_remote_mmus(struct kvm *kvm)
177 make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
180 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
185 mutex_init(&vcpu->mutex);
189 init_waitqueue_head(&vcpu->wq);
190 kvm_async_pf_vcpu_init(vcpu);
192 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
197 vcpu->run = page_address(page);
199 r = kvm_arch_vcpu_init(vcpu);
205 free_page((unsigned long)vcpu->run);
209 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
211 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
213 kvm_arch_vcpu_uninit(vcpu);
214 free_page((unsigned long)vcpu->run);
216 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
218 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
219 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
221 return container_of(mn, struct kvm, mmu_notifier);
224 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
225 struct mm_struct *mm,
226 unsigned long address)
228 struct kvm *kvm = mmu_notifier_to_kvm(mn);
229 int need_tlb_flush, idx;
232 * When ->invalidate_page runs, the linux pte has been zapped
233 * already but the page is still allocated until
234 * ->invalidate_page returns. So if we increase the sequence
235 * here the kvm page fault will notice if the spte can't be
236 * established because the page is going to be freed. If
237 * instead the kvm page fault establishes the spte before
238 * ->invalidate_page runs, kvm_unmap_hva will release it
241 * The sequence increase only need to be seen at spin_unlock
242 * time, and not at spin_lock time.
244 * Increasing the sequence after the spin_unlock would be
245 * unsafe because the kvm page fault could then establish the
246 * pte after kvm_unmap_hva returned, without noticing the page
247 * is going to be freed.
249 idx = srcu_read_lock(&kvm->srcu);
250 spin_lock(&kvm->mmu_lock);
251 kvm->mmu_notifier_seq++;
252 need_tlb_flush = kvm_unmap_hva(kvm, address);
253 spin_unlock(&kvm->mmu_lock);
254 srcu_read_unlock(&kvm->srcu, idx);
256 /* we've to flush the tlb before the pages can be freed */
258 kvm_flush_remote_tlbs(kvm);
262 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
263 struct mm_struct *mm,
264 unsigned long address,
267 struct kvm *kvm = mmu_notifier_to_kvm(mn);
270 idx = srcu_read_lock(&kvm->srcu);
271 spin_lock(&kvm->mmu_lock);
272 kvm->mmu_notifier_seq++;
273 kvm_set_spte_hva(kvm, address, pte);
274 spin_unlock(&kvm->mmu_lock);
275 srcu_read_unlock(&kvm->srcu, idx);
278 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
279 struct mm_struct *mm,
283 struct kvm *kvm = mmu_notifier_to_kvm(mn);
284 int need_tlb_flush = 0, idx;
286 idx = srcu_read_lock(&kvm->srcu);
287 spin_lock(&kvm->mmu_lock);
289 * The count increase must become visible at unlock time as no
290 * spte can be established without taking the mmu_lock and
291 * count is also read inside the mmu_lock critical section.
293 kvm->mmu_notifier_count++;
294 for (; start < end; start += PAGE_SIZE)
295 need_tlb_flush |= kvm_unmap_hva(kvm, start);
296 spin_unlock(&kvm->mmu_lock);
297 srcu_read_unlock(&kvm->srcu, idx);
299 /* we've to flush the tlb before the pages can be freed */
301 kvm_flush_remote_tlbs(kvm);
304 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
305 struct mm_struct *mm,
309 struct kvm *kvm = mmu_notifier_to_kvm(mn);
311 spin_lock(&kvm->mmu_lock);
313 * This sequence increase will notify the kvm page fault that
314 * the page that is going to be mapped in the spte could have
317 kvm->mmu_notifier_seq++;
319 * The above sequence increase must be visible before the
320 * below count decrease but both values are read by the kvm
321 * page fault under mmu_lock spinlock so we don't need to add
322 * a smb_wmb() here in between the two.
324 kvm->mmu_notifier_count--;
325 spin_unlock(&kvm->mmu_lock);
327 BUG_ON(kvm->mmu_notifier_count < 0);
330 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
331 struct mm_struct *mm,
332 unsigned long address)
334 struct kvm *kvm = mmu_notifier_to_kvm(mn);
337 idx = srcu_read_lock(&kvm->srcu);
338 spin_lock(&kvm->mmu_lock);
339 young = kvm_age_hva(kvm, address);
340 spin_unlock(&kvm->mmu_lock);
341 srcu_read_unlock(&kvm->srcu, idx);
344 kvm_flush_remote_tlbs(kvm);
349 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
350 struct mm_struct *mm)
352 struct kvm *kvm = mmu_notifier_to_kvm(mn);
355 idx = srcu_read_lock(&kvm->srcu);
356 kvm_arch_flush_shadow(kvm);
357 srcu_read_unlock(&kvm->srcu, idx);
360 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
361 .invalidate_page = kvm_mmu_notifier_invalidate_page,
362 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
363 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
364 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
365 .change_pte = kvm_mmu_notifier_change_pte,
366 .release = kvm_mmu_notifier_release,
369 static int kvm_init_mmu_notifier(struct kvm *kvm)
371 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
372 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
375 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
377 static int kvm_init_mmu_notifier(struct kvm *kvm)
382 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
384 static struct kvm *kvm_create_vm(void)
387 struct kvm *kvm = kvm_arch_create_vm();
392 r = hardware_enable_all();
394 goto out_err_nodisable;
396 #ifdef CONFIG_HAVE_KVM_IRQCHIP
397 INIT_HLIST_HEAD(&kvm->mask_notifier_list);
398 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
402 kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
405 if (init_srcu_struct(&kvm->srcu))
407 for (i = 0; i < KVM_NR_BUSES; i++) {
408 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
410 if (!kvm->buses[i]) {
411 cleanup_srcu_struct(&kvm->srcu);
416 r = kvm_init_mmu_notifier(kvm);
418 cleanup_srcu_struct(&kvm->srcu);
422 kvm->mm = current->mm;
423 atomic_inc(&kvm->mm->mm_count);
424 spin_lock_init(&kvm->mmu_lock);
425 raw_spin_lock_init(&kvm->requests_lock);
426 kvm_eventfd_init(kvm);
427 mutex_init(&kvm->lock);
428 mutex_init(&kvm->irq_lock);
429 mutex_init(&kvm->slots_lock);
430 atomic_set(&kvm->users_count, 1);
431 spin_lock(&kvm_lock);
432 list_add(&kvm->vm_list, &vm_list);
433 spin_unlock(&kvm_lock);
438 hardware_disable_all();
440 for (i = 0; i < KVM_NR_BUSES; i++)
441 kfree(kvm->buses[i]);
442 kfree(kvm->memslots);
447 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
449 if (!memslot->dirty_bitmap)
452 if (2 * kvm_dirty_bitmap_bytes(memslot) > PAGE_SIZE)
453 vfree(memslot->dirty_bitmap_head);
455 kfree(memslot->dirty_bitmap_head);
457 memslot->dirty_bitmap = NULL;
458 memslot->dirty_bitmap_head = NULL;
462 * Free any memory in @free but not in @dont.
464 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
465 struct kvm_memory_slot *dont)
469 if (!dont || free->rmap != dont->rmap)
472 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
473 kvm_destroy_dirty_bitmap(free);
476 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
477 if (!dont || free->lpage_info[i] != dont->lpage_info[i]) {
478 vfree(free->lpage_info[i]);
479 free->lpage_info[i] = NULL;
487 void kvm_free_physmem(struct kvm *kvm)
490 struct kvm_memslots *slots = kvm->memslots;
492 for (i = 0; i < slots->nmemslots; ++i)
493 kvm_free_physmem_slot(&slots->memslots[i], NULL);
495 kfree(kvm->memslots);
498 static void kvm_destroy_vm(struct kvm *kvm)
501 struct mm_struct *mm = kvm->mm;
503 kvm_arch_sync_events(kvm);
504 spin_lock(&kvm_lock);
505 list_del(&kvm->vm_list);
506 spin_unlock(&kvm_lock);
507 kvm_free_irq_routing(kvm);
508 for (i = 0; i < KVM_NR_BUSES; i++)
509 kvm_io_bus_destroy(kvm->buses[i]);
510 kvm_coalesced_mmio_free(kvm);
511 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
512 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
514 kvm_arch_flush_shadow(kvm);
516 kvm_arch_destroy_vm(kvm);
517 hardware_disable_all();
521 void kvm_get_kvm(struct kvm *kvm)
523 atomic_inc(&kvm->users_count);
525 EXPORT_SYMBOL_GPL(kvm_get_kvm);
527 void kvm_put_kvm(struct kvm *kvm)
529 if (atomic_dec_and_test(&kvm->users_count))
532 EXPORT_SYMBOL_GPL(kvm_put_kvm);
535 static int kvm_vm_release(struct inode *inode, struct file *filp)
537 struct kvm *kvm = filp->private_data;
539 kvm_irqfd_release(kvm);
546 * Allocation size is twice as large as the actual dirty bitmap size.
547 * This makes it possible to do double buffering: see x86's
548 * kvm_vm_ioctl_get_dirty_log().
550 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
552 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
554 if (dirty_bytes > PAGE_SIZE)
555 memslot->dirty_bitmap = vzalloc(dirty_bytes);
557 memslot->dirty_bitmap = kzalloc(dirty_bytes, GFP_KERNEL);
559 if (!memslot->dirty_bitmap)
562 memslot->dirty_bitmap_head = memslot->dirty_bitmap;
567 * Allocate some memory and give it an address in the guest physical address
570 * Discontiguous memory is allowed, mostly for framebuffers.
572 * Must be called holding mmap_sem for write.
574 int __kvm_set_memory_region(struct kvm *kvm,
575 struct kvm_userspace_memory_region *mem,
578 int r, flush_shadow = 0;
580 unsigned long npages;
582 struct kvm_memory_slot *memslot;
583 struct kvm_memory_slot old, new;
584 struct kvm_memslots *slots, *old_memslots;
587 /* General sanity checks */
588 if (mem->memory_size & (PAGE_SIZE - 1))
590 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
592 if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
594 if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
596 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
599 memslot = &kvm->memslots->memslots[mem->slot];
600 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
601 npages = mem->memory_size >> PAGE_SHIFT;
604 if (npages > KVM_MEM_MAX_NR_PAGES)
608 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
610 new = old = *memslot;
613 new.base_gfn = base_gfn;
615 new.flags = mem->flags;
617 /* Disallow changing a memory slot's size. */
619 if (npages && old.npages && npages != old.npages)
622 /* Check for overlaps */
624 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
625 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
627 if (s == memslot || !s->npages)
629 if (!((base_gfn + npages <= s->base_gfn) ||
630 (base_gfn >= s->base_gfn + s->npages)))
634 /* Free page dirty bitmap if unneeded */
635 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
636 new.dirty_bitmap = NULL;
640 /* Allocate if a slot is being created */
642 if (npages && !new.rmap) {
643 new.rmap = vmalloc(npages * sizeof(*new.rmap));
648 memset(new.rmap, 0, npages * sizeof(*new.rmap));
650 new.user_alloc = user_alloc;
651 new.userspace_addr = mem->userspace_addr;
656 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
662 /* Avoid unused variable warning if no large pages */
665 if (new.lpage_info[i])
668 lpages = 1 + ((base_gfn + npages - 1)
669 >> KVM_HPAGE_GFN_SHIFT(level));
670 lpages -= base_gfn >> KVM_HPAGE_GFN_SHIFT(level);
672 new.lpage_info[i] = vmalloc(lpages * sizeof(*new.lpage_info[i]));
674 if (!new.lpage_info[i])
677 memset(new.lpage_info[i], 0,
678 lpages * sizeof(*new.lpage_info[i]));
680 if (base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
681 new.lpage_info[i][0].write_count = 1;
682 if ((base_gfn+npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
683 new.lpage_info[i][lpages - 1].write_count = 1;
684 ugfn = new.userspace_addr >> PAGE_SHIFT;
686 * If the gfn and userspace address are not aligned wrt each
687 * other, or if explicitly asked to, disable large page
688 * support for this slot
690 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
692 for (j = 0; j < lpages; ++j)
693 new.lpage_info[i][j].write_count = 1;
698 /* Allocate page dirty bitmap if needed */
699 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
700 if (kvm_create_dirty_bitmap(&new) < 0)
702 /* destroy any largepage mappings for dirty tracking */
706 #else /* not defined CONFIG_S390 */
707 new.user_alloc = user_alloc;
709 new.userspace_addr = mem->userspace_addr;
710 #endif /* not defined CONFIG_S390 */
714 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
717 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
718 if (mem->slot >= slots->nmemslots)
719 slots->nmemslots = mem->slot + 1;
721 slots->memslots[mem->slot].flags |= KVM_MEMSLOT_INVALID;
723 old_memslots = kvm->memslots;
724 rcu_assign_pointer(kvm->memslots, slots);
725 synchronize_srcu_expedited(&kvm->srcu);
726 /* From this point no new shadow pages pointing to a deleted
727 * memslot will be created.
729 * validation of sp->gfn happens in:
730 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
731 * - kvm_is_visible_gfn (mmu_check_roots)
733 kvm_arch_flush_shadow(kvm);
737 r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
741 /* map the pages in iommu page table */
743 r = kvm_iommu_map_pages(kvm, &new);
749 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
752 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
753 if (mem->slot >= slots->nmemslots)
754 slots->nmemslots = mem->slot + 1;
757 /* actual memory is freed via old in kvm_free_physmem_slot below */
760 new.dirty_bitmap = NULL;
761 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i)
762 new.lpage_info[i] = NULL;
765 slots->memslots[mem->slot] = new;
766 old_memslots = kvm->memslots;
767 rcu_assign_pointer(kvm->memslots, slots);
768 synchronize_srcu_expedited(&kvm->srcu);
770 kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
772 kvm_free_physmem_slot(&old, &new);
776 kvm_arch_flush_shadow(kvm);
781 kvm_free_physmem_slot(&new, &old);
786 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
788 int kvm_set_memory_region(struct kvm *kvm,
789 struct kvm_userspace_memory_region *mem,
794 mutex_lock(&kvm->slots_lock);
795 r = __kvm_set_memory_region(kvm, mem, user_alloc);
796 mutex_unlock(&kvm->slots_lock);
799 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
801 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
803 kvm_userspace_memory_region *mem,
806 if (mem->slot >= KVM_MEMORY_SLOTS)
808 return kvm_set_memory_region(kvm, mem, user_alloc);
811 int kvm_get_dirty_log(struct kvm *kvm,
812 struct kvm_dirty_log *log, int *is_dirty)
814 struct kvm_memory_slot *memslot;
817 unsigned long any = 0;
820 if (log->slot >= KVM_MEMORY_SLOTS)
823 memslot = &kvm->memslots->memslots[log->slot];
825 if (!memslot->dirty_bitmap)
828 n = kvm_dirty_bitmap_bytes(memslot);
830 for (i = 0; !any && i < n/sizeof(long); ++i)
831 any = memslot->dirty_bitmap[i];
834 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
845 void kvm_disable_largepages(void)
847 largepages_enabled = false;
849 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
851 int is_error_page(struct page *page)
853 return page == bad_page || page == hwpoison_page || page == fault_page;
855 EXPORT_SYMBOL_GPL(is_error_page);
857 int is_error_pfn(pfn_t pfn)
859 return pfn == bad_pfn || pfn == hwpoison_pfn || pfn == fault_pfn;
861 EXPORT_SYMBOL_GPL(is_error_pfn);
863 int is_hwpoison_pfn(pfn_t pfn)
865 return pfn == hwpoison_pfn;
867 EXPORT_SYMBOL_GPL(is_hwpoison_pfn);
869 int is_fault_pfn(pfn_t pfn)
871 return pfn == fault_pfn;
873 EXPORT_SYMBOL_GPL(is_fault_pfn);
875 static inline unsigned long bad_hva(void)
880 int kvm_is_error_hva(unsigned long addr)
882 return addr == bad_hva();
884 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
886 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm_memslots *slots,
891 for (i = 0; i < slots->nmemslots; ++i) {
892 struct kvm_memory_slot *memslot = &slots->memslots[i];
894 if (gfn >= memslot->base_gfn
895 && gfn < memslot->base_gfn + memslot->npages)
901 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
903 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
905 EXPORT_SYMBOL_GPL(gfn_to_memslot);
907 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
910 struct kvm_memslots *slots = kvm_memslots(kvm);
912 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
913 struct kvm_memory_slot *memslot = &slots->memslots[i];
915 if (memslot->flags & KVM_MEMSLOT_INVALID)
918 if (gfn >= memslot->base_gfn
919 && gfn < memslot->base_gfn + memslot->npages)
924 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
926 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
928 struct vm_area_struct *vma;
929 unsigned long addr, size;
933 addr = gfn_to_hva(kvm, gfn);
934 if (kvm_is_error_hva(addr))
937 down_read(¤t->mm->mmap_sem);
938 vma = find_vma(current->mm, addr);
942 size = vma_kernel_pagesize(vma);
945 up_read(¤t->mm->mmap_sem);
950 int memslot_id(struct kvm *kvm, gfn_t gfn)
953 struct kvm_memslots *slots = kvm_memslots(kvm);
954 struct kvm_memory_slot *memslot = NULL;
956 for (i = 0; i < slots->nmemslots; ++i) {
957 memslot = &slots->memslots[i];
959 if (gfn >= memslot->base_gfn
960 && gfn < memslot->base_gfn + memslot->npages)
964 return memslot - slots->memslots;
967 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
970 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
974 *nr_pages = slot->npages - (gfn - slot->base_gfn);
976 return gfn_to_hva_memslot(slot, gfn);
979 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
981 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
983 EXPORT_SYMBOL_GPL(gfn_to_hva);
985 static pfn_t get_fault_pfn(void)
987 get_page(fault_page);
991 static pfn_t hva_to_pfn(struct kvm *kvm, unsigned long addr, bool atomic,
992 bool *async, bool write_fault, bool *writable)
994 struct page *page[1];
998 /* we can do it either atomically or asynchronously, not both */
999 BUG_ON(atomic && async);
1001 BUG_ON(!write_fault && !writable);
1006 if (atomic || async)
1007 npages = __get_user_pages_fast(addr, 1, 1, page);
1009 if (unlikely(npages != 1) && !atomic) {
1013 *writable = write_fault;
1015 npages = get_user_pages_fast(addr, 1, write_fault, page);
1017 /* map read fault as writable if possible */
1018 if (unlikely(!write_fault) && npages == 1) {
1019 struct page *wpage[1];
1021 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1031 if (unlikely(npages != 1)) {
1032 struct vm_area_struct *vma;
1035 return get_fault_pfn();
1037 down_read(¤t->mm->mmap_sem);
1038 if (is_hwpoison_address(addr)) {
1039 up_read(¤t->mm->mmap_sem);
1040 get_page(hwpoison_page);
1041 return page_to_pfn(hwpoison_page);
1044 vma = find_vma_intersection(current->mm, addr, addr+1);
1047 pfn = get_fault_pfn();
1048 else if ((vma->vm_flags & VM_PFNMAP)) {
1049 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1051 BUG_ON(!kvm_is_mmio_pfn(pfn));
1053 if (async && (vma->vm_flags & VM_WRITE))
1055 pfn = get_fault_pfn();
1057 up_read(¤t->mm->mmap_sem);
1059 pfn = page_to_pfn(page[0]);
1064 pfn_t hva_to_pfn_atomic(struct kvm *kvm, unsigned long addr)
1066 return hva_to_pfn(kvm, addr, true, NULL, true, NULL);
1068 EXPORT_SYMBOL_GPL(hva_to_pfn_atomic);
1070 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1071 bool write_fault, bool *writable)
1078 addr = gfn_to_hva(kvm, gfn);
1079 if (kvm_is_error_hva(addr)) {
1081 return page_to_pfn(bad_page);
1084 return hva_to_pfn(kvm, addr, atomic, async, write_fault, writable);
1087 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1089 return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1091 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1093 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1094 bool write_fault, bool *writable)
1096 return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1098 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1100 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1102 return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1104 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1106 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1109 return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1111 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1113 pfn_t gfn_to_pfn_memslot(struct kvm *kvm,
1114 struct kvm_memory_slot *slot, gfn_t gfn)
1116 unsigned long addr = gfn_to_hva_memslot(slot, gfn);
1117 return hva_to_pfn(kvm, addr, false, NULL, true, NULL);
1120 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1126 addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1127 if (kvm_is_error_hva(addr))
1130 if (entry < nr_pages)
1133 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1135 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1137 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1141 pfn = gfn_to_pfn(kvm, gfn);
1142 if (!kvm_is_mmio_pfn(pfn))
1143 return pfn_to_page(pfn);
1145 WARN_ON(kvm_is_mmio_pfn(pfn));
1151 EXPORT_SYMBOL_GPL(gfn_to_page);
1153 void kvm_release_page_clean(struct page *page)
1155 kvm_release_pfn_clean(page_to_pfn(page));
1157 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1159 void kvm_release_pfn_clean(pfn_t pfn)
1161 if (!kvm_is_mmio_pfn(pfn))
1162 put_page(pfn_to_page(pfn));
1164 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1166 void kvm_release_page_dirty(struct page *page)
1168 kvm_release_pfn_dirty(page_to_pfn(page));
1170 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1172 void kvm_release_pfn_dirty(pfn_t pfn)
1174 kvm_set_pfn_dirty(pfn);
1175 kvm_release_pfn_clean(pfn);
1177 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1179 void kvm_set_page_dirty(struct page *page)
1181 kvm_set_pfn_dirty(page_to_pfn(page));
1183 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1185 void kvm_set_pfn_dirty(pfn_t pfn)
1187 if (!kvm_is_mmio_pfn(pfn)) {
1188 struct page *page = pfn_to_page(pfn);
1189 if (!PageReserved(page))
1193 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1195 void kvm_set_pfn_accessed(pfn_t pfn)
1197 if (!kvm_is_mmio_pfn(pfn))
1198 mark_page_accessed(pfn_to_page(pfn));
1200 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1202 void kvm_get_pfn(pfn_t pfn)
1204 if (!kvm_is_mmio_pfn(pfn))
1205 get_page(pfn_to_page(pfn));
1207 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1209 static int next_segment(unsigned long len, int offset)
1211 if (len > PAGE_SIZE - offset)
1212 return PAGE_SIZE - offset;
1217 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1223 addr = gfn_to_hva(kvm, gfn);
1224 if (kvm_is_error_hva(addr))
1226 r = copy_from_user(data, (void __user *)addr + offset, len);
1231 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1233 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1235 gfn_t gfn = gpa >> PAGE_SHIFT;
1237 int offset = offset_in_page(gpa);
1240 while ((seg = next_segment(len, offset)) != 0) {
1241 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1251 EXPORT_SYMBOL_GPL(kvm_read_guest);
1253 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1258 gfn_t gfn = gpa >> PAGE_SHIFT;
1259 int offset = offset_in_page(gpa);
1261 addr = gfn_to_hva(kvm, gfn);
1262 if (kvm_is_error_hva(addr))
1264 pagefault_disable();
1265 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1271 EXPORT_SYMBOL(kvm_read_guest_atomic);
1273 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1274 int offset, int len)
1279 addr = gfn_to_hva(kvm, gfn);
1280 if (kvm_is_error_hva(addr))
1282 r = copy_to_user((void __user *)addr + offset, data, len);
1285 mark_page_dirty(kvm, gfn);
1288 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1290 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1293 gfn_t gfn = gpa >> PAGE_SHIFT;
1295 int offset = offset_in_page(gpa);
1298 while ((seg = next_segment(len, offset)) != 0) {
1299 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1310 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1313 struct kvm_memslots *slots = kvm_memslots(kvm);
1314 int offset = offset_in_page(gpa);
1315 gfn_t gfn = gpa >> PAGE_SHIFT;
1318 ghc->generation = slots->generation;
1319 ghc->memslot = __gfn_to_memslot(slots, gfn);
1320 ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1321 if (!kvm_is_error_hva(ghc->hva))
1328 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1330 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1331 void *data, unsigned long len)
1333 struct kvm_memslots *slots = kvm_memslots(kvm);
1336 if (slots->generation != ghc->generation)
1337 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1339 if (kvm_is_error_hva(ghc->hva))
1342 r = copy_to_user((void __user *)ghc->hva, data, len);
1345 mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1349 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1351 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1353 return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1356 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1358 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1360 gfn_t gfn = gpa >> PAGE_SHIFT;
1362 int offset = offset_in_page(gpa);
1365 while ((seg = next_segment(len, offset)) != 0) {
1366 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1375 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1377 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1380 if (memslot && memslot->dirty_bitmap) {
1381 unsigned long rel_gfn = gfn - memslot->base_gfn;
1383 generic___set_le_bit(rel_gfn, memslot->dirty_bitmap);
1387 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1389 struct kvm_memory_slot *memslot;
1391 memslot = gfn_to_memslot(kvm, gfn);
1392 mark_page_dirty_in_slot(kvm, memslot, gfn);
1396 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1398 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1403 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1405 if (kvm_arch_vcpu_runnable(vcpu)) {
1406 kvm_make_request(KVM_REQ_UNHALT, vcpu);
1409 if (kvm_cpu_has_pending_timer(vcpu))
1411 if (signal_pending(current))
1417 finish_wait(&vcpu->wq, &wait);
1420 void kvm_resched(struct kvm_vcpu *vcpu)
1422 if (!need_resched())
1426 EXPORT_SYMBOL_GPL(kvm_resched);
1428 void kvm_vcpu_on_spin(struct kvm_vcpu *vcpu)
1433 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1435 /* Sleep for 100 us, and hope lock-holder got scheduled */
1436 expires = ktime_add_ns(ktime_get(), 100000UL);
1437 schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
1439 finish_wait(&vcpu->wq, &wait);
1441 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1443 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1445 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1448 if (vmf->pgoff == 0)
1449 page = virt_to_page(vcpu->run);
1451 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1452 page = virt_to_page(vcpu->arch.pio_data);
1454 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1455 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1456 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1459 return VM_FAULT_SIGBUS;
1465 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1466 .fault = kvm_vcpu_fault,
1469 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1471 vma->vm_ops = &kvm_vcpu_vm_ops;
1475 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1477 struct kvm_vcpu *vcpu = filp->private_data;
1479 kvm_put_kvm(vcpu->kvm);
1483 static struct file_operations kvm_vcpu_fops = {
1484 .release = kvm_vcpu_release,
1485 .unlocked_ioctl = kvm_vcpu_ioctl,
1486 .compat_ioctl = kvm_vcpu_ioctl,
1487 .mmap = kvm_vcpu_mmap,
1488 .llseek = noop_llseek,
1492 * Allocates an inode for the vcpu.
1494 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1496 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1500 * Creates some virtual cpus. Good luck creating more than one.
1502 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1505 struct kvm_vcpu *vcpu, *v;
1507 vcpu = kvm_arch_vcpu_create(kvm, id);
1509 return PTR_ERR(vcpu);
1511 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1513 r = kvm_arch_vcpu_setup(vcpu);
1517 mutex_lock(&kvm->lock);
1518 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1523 kvm_for_each_vcpu(r, v, kvm)
1524 if (v->vcpu_id == id) {
1529 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1531 /* Now it's all set up, let userspace reach it */
1533 r = create_vcpu_fd(vcpu);
1539 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1541 atomic_inc(&kvm->online_vcpus);
1543 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1544 if (kvm->bsp_vcpu_id == id)
1545 kvm->bsp_vcpu = vcpu;
1547 mutex_unlock(&kvm->lock);
1551 mutex_unlock(&kvm->lock);
1552 kvm_arch_vcpu_destroy(vcpu);
1556 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1559 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1560 vcpu->sigset_active = 1;
1561 vcpu->sigset = *sigset;
1563 vcpu->sigset_active = 0;
1567 static long kvm_vcpu_ioctl(struct file *filp,
1568 unsigned int ioctl, unsigned long arg)
1570 struct kvm_vcpu *vcpu = filp->private_data;
1571 void __user *argp = (void __user *)arg;
1573 struct kvm_fpu *fpu = NULL;
1574 struct kvm_sregs *kvm_sregs = NULL;
1576 if (vcpu->kvm->mm != current->mm)
1579 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1581 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1582 * so vcpu_load() would break it.
1584 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1585 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1595 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1596 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1598 case KVM_GET_REGS: {
1599 struct kvm_regs *kvm_regs;
1602 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1605 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1609 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1616 case KVM_SET_REGS: {
1617 struct kvm_regs *kvm_regs;
1620 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1624 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1626 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1634 case KVM_GET_SREGS: {
1635 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1639 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1643 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1648 case KVM_SET_SREGS: {
1649 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1654 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1656 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1662 case KVM_GET_MP_STATE: {
1663 struct kvm_mp_state mp_state;
1665 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1669 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1674 case KVM_SET_MP_STATE: {
1675 struct kvm_mp_state mp_state;
1678 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1680 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1686 case KVM_TRANSLATE: {
1687 struct kvm_translation tr;
1690 if (copy_from_user(&tr, argp, sizeof tr))
1692 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1696 if (copy_to_user(argp, &tr, sizeof tr))
1701 case KVM_SET_GUEST_DEBUG: {
1702 struct kvm_guest_debug dbg;
1705 if (copy_from_user(&dbg, argp, sizeof dbg))
1707 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1713 case KVM_SET_SIGNAL_MASK: {
1714 struct kvm_signal_mask __user *sigmask_arg = argp;
1715 struct kvm_signal_mask kvm_sigmask;
1716 sigset_t sigset, *p;
1721 if (copy_from_user(&kvm_sigmask, argp,
1722 sizeof kvm_sigmask))
1725 if (kvm_sigmask.len != sizeof sigset)
1728 if (copy_from_user(&sigset, sigmask_arg->sigset,
1733 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
1737 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1741 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1745 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1751 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1756 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1758 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1765 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1774 static long kvm_vm_ioctl(struct file *filp,
1775 unsigned int ioctl, unsigned long arg)
1777 struct kvm *kvm = filp->private_data;
1778 void __user *argp = (void __user *)arg;
1781 if (kvm->mm != current->mm)
1784 case KVM_CREATE_VCPU:
1785 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1789 case KVM_SET_USER_MEMORY_REGION: {
1790 struct kvm_userspace_memory_region kvm_userspace_mem;
1793 if (copy_from_user(&kvm_userspace_mem, argp,
1794 sizeof kvm_userspace_mem))
1797 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1802 case KVM_GET_DIRTY_LOG: {
1803 struct kvm_dirty_log log;
1806 if (copy_from_user(&log, argp, sizeof log))
1808 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1813 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1814 case KVM_REGISTER_COALESCED_MMIO: {
1815 struct kvm_coalesced_mmio_zone zone;
1817 if (copy_from_user(&zone, argp, sizeof zone))
1819 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1825 case KVM_UNREGISTER_COALESCED_MMIO: {
1826 struct kvm_coalesced_mmio_zone zone;
1828 if (copy_from_user(&zone, argp, sizeof zone))
1830 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1838 struct kvm_irqfd data;
1841 if (copy_from_user(&data, argp, sizeof data))
1843 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
1846 case KVM_IOEVENTFD: {
1847 struct kvm_ioeventfd data;
1850 if (copy_from_user(&data, argp, sizeof data))
1852 r = kvm_ioeventfd(kvm, &data);
1855 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1856 case KVM_SET_BOOT_CPU_ID:
1858 mutex_lock(&kvm->lock);
1859 if (atomic_read(&kvm->online_vcpus) != 0)
1862 kvm->bsp_vcpu_id = arg;
1863 mutex_unlock(&kvm->lock);
1867 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1869 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
1875 #ifdef CONFIG_COMPAT
1876 struct compat_kvm_dirty_log {
1880 compat_uptr_t dirty_bitmap; /* one bit per page */
1885 static long kvm_vm_compat_ioctl(struct file *filp,
1886 unsigned int ioctl, unsigned long arg)
1888 struct kvm *kvm = filp->private_data;
1891 if (kvm->mm != current->mm)
1894 case KVM_GET_DIRTY_LOG: {
1895 struct compat_kvm_dirty_log compat_log;
1896 struct kvm_dirty_log log;
1899 if (copy_from_user(&compat_log, (void __user *)arg,
1900 sizeof(compat_log)))
1902 log.slot = compat_log.slot;
1903 log.padding1 = compat_log.padding1;
1904 log.padding2 = compat_log.padding2;
1905 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
1907 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1913 r = kvm_vm_ioctl(filp, ioctl, arg);
1921 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1923 struct page *page[1];
1926 gfn_t gfn = vmf->pgoff;
1927 struct kvm *kvm = vma->vm_file->private_data;
1929 addr = gfn_to_hva(kvm, gfn);
1930 if (kvm_is_error_hva(addr))
1931 return VM_FAULT_SIGBUS;
1933 npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
1935 if (unlikely(npages != 1))
1936 return VM_FAULT_SIGBUS;
1938 vmf->page = page[0];
1942 static const struct vm_operations_struct kvm_vm_vm_ops = {
1943 .fault = kvm_vm_fault,
1946 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1948 vma->vm_ops = &kvm_vm_vm_ops;
1952 static struct file_operations kvm_vm_fops = {
1953 .release = kvm_vm_release,
1954 .unlocked_ioctl = kvm_vm_ioctl,
1955 #ifdef CONFIG_COMPAT
1956 .compat_ioctl = kvm_vm_compat_ioctl,
1958 .mmap = kvm_vm_mmap,
1959 .llseek = noop_llseek,
1962 static int kvm_dev_ioctl_create_vm(void)
1967 kvm = kvm_create_vm();
1969 return PTR_ERR(kvm);
1970 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1971 r = kvm_coalesced_mmio_init(kvm);
1977 fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
1984 static long kvm_dev_ioctl_check_extension_generic(long arg)
1987 case KVM_CAP_USER_MEMORY:
1988 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
1989 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
1990 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1991 case KVM_CAP_SET_BOOT_CPU_ID:
1993 case KVM_CAP_INTERNAL_ERROR_DATA:
1995 #ifdef CONFIG_HAVE_KVM_IRQCHIP
1996 case KVM_CAP_IRQ_ROUTING:
1997 return KVM_MAX_IRQ_ROUTES;
2002 return kvm_dev_ioctl_check_extension(arg);
2005 static long kvm_dev_ioctl(struct file *filp,
2006 unsigned int ioctl, unsigned long arg)
2011 case KVM_GET_API_VERSION:
2015 r = KVM_API_VERSION;
2021 r = kvm_dev_ioctl_create_vm();
2023 case KVM_CHECK_EXTENSION:
2024 r = kvm_dev_ioctl_check_extension_generic(arg);
2026 case KVM_GET_VCPU_MMAP_SIZE:
2030 r = PAGE_SIZE; /* struct kvm_run */
2032 r += PAGE_SIZE; /* pio data page */
2034 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2035 r += PAGE_SIZE; /* coalesced mmio ring page */
2038 case KVM_TRACE_ENABLE:
2039 case KVM_TRACE_PAUSE:
2040 case KVM_TRACE_DISABLE:
2044 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2050 static struct file_operations kvm_chardev_ops = {
2051 .unlocked_ioctl = kvm_dev_ioctl,
2052 .compat_ioctl = kvm_dev_ioctl,
2053 .llseek = noop_llseek,
2056 static struct miscdevice kvm_dev = {
2062 static void hardware_enable(void *junk)
2064 int cpu = raw_smp_processor_id();
2067 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2070 cpumask_set_cpu(cpu, cpus_hardware_enabled);
2072 r = kvm_arch_hardware_enable(NULL);
2075 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2076 atomic_inc(&hardware_enable_failed);
2077 printk(KERN_INFO "kvm: enabling virtualization on "
2078 "CPU%d failed\n", cpu);
2082 static void hardware_disable(void *junk)
2084 int cpu = raw_smp_processor_id();
2086 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2088 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2089 kvm_arch_hardware_disable(NULL);
2092 static void hardware_disable_all_nolock(void)
2094 BUG_ON(!kvm_usage_count);
2097 if (!kvm_usage_count)
2098 on_each_cpu(hardware_disable, NULL, 1);
2101 static void hardware_disable_all(void)
2103 spin_lock(&kvm_lock);
2104 hardware_disable_all_nolock();
2105 spin_unlock(&kvm_lock);
2108 static int hardware_enable_all(void)
2112 spin_lock(&kvm_lock);
2115 if (kvm_usage_count == 1) {
2116 atomic_set(&hardware_enable_failed, 0);
2117 on_each_cpu(hardware_enable, NULL, 1);
2119 if (atomic_read(&hardware_enable_failed)) {
2120 hardware_disable_all_nolock();
2125 spin_unlock(&kvm_lock);
2130 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2135 if (!kvm_usage_count)
2138 val &= ~CPU_TASKS_FROZEN;
2141 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2143 hardware_disable(NULL);
2146 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2148 spin_lock(&kvm_lock);
2149 hardware_enable(NULL);
2150 spin_unlock(&kvm_lock);
2157 asmlinkage void kvm_handle_fault_on_reboot(void)
2159 if (kvm_rebooting) {
2160 /* spin while reset goes on */
2165 /* Fault while not rebooting. We want the trace. */
2168 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
2170 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2174 * Some (well, at least mine) BIOSes hang on reboot if
2177 * And Intel TXT required VMX off for all cpu when system shutdown.
2179 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2180 kvm_rebooting = true;
2181 on_each_cpu(hardware_disable, NULL, 1);
2185 static struct notifier_block kvm_reboot_notifier = {
2186 .notifier_call = kvm_reboot,
2190 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2194 for (i = 0; i < bus->dev_count; i++) {
2195 struct kvm_io_device *pos = bus->devs[i];
2197 kvm_iodevice_destructor(pos);
2202 /* kvm_io_bus_write - called under kvm->slots_lock */
2203 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2204 int len, const void *val)
2207 struct kvm_io_bus *bus;
2209 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2210 for (i = 0; i < bus->dev_count; i++)
2211 if (!kvm_iodevice_write(bus->devs[i], addr, len, val))
2216 /* kvm_io_bus_read - called under kvm->slots_lock */
2217 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2221 struct kvm_io_bus *bus;
2223 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2224 for (i = 0; i < bus->dev_count; i++)
2225 if (!kvm_iodevice_read(bus->devs[i], addr, len, val))
2230 /* Caller must hold slots_lock. */
2231 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2232 struct kvm_io_device *dev)
2234 struct kvm_io_bus *new_bus, *bus;
2236 bus = kvm->buses[bus_idx];
2237 if (bus->dev_count > NR_IOBUS_DEVS-1)
2240 new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2243 memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2244 new_bus->devs[new_bus->dev_count++] = dev;
2245 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2246 synchronize_srcu_expedited(&kvm->srcu);
2252 /* Caller must hold slots_lock. */
2253 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2254 struct kvm_io_device *dev)
2257 struct kvm_io_bus *new_bus, *bus;
2259 new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2263 bus = kvm->buses[bus_idx];
2264 memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2267 for (i = 0; i < new_bus->dev_count; i++)
2268 if (new_bus->devs[i] == dev) {
2270 new_bus->devs[i] = new_bus->devs[--new_bus->dev_count];
2279 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2280 synchronize_srcu_expedited(&kvm->srcu);
2285 static struct notifier_block kvm_cpu_notifier = {
2286 .notifier_call = kvm_cpu_hotplug,
2289 static int vm_stat_get(void *_offset, u64 *val)
2291 unsigned offset = (long)_offset;
2295 spin_lock(&kvm_lock);
2296 list_for_each_entry(kvm, &vm_list, vm_list)
2297 *val += *(u32 *)((void *)kvm + offset);
2298 spin_unlock(&kvm_lock);
2302 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2304 static int vcpu_stat_get(void *_offset, u64 *val)
2306 unsigned offset = (long)_offset;
2308 struct kvm_vcpu *vcpu;
2312 spin_lock(&kvm_lock);
2313 list_for_each_entry(kvm, &vm_list, vm_list)
2314 kvm_for_each_vcpu(i, vcpu, kvm)
2315 *val += *(u32 *)((void *)vcpu + offset);
2317 spin_unlock(&kvm_lock);
2321 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2323 static const struct file_operations *stat_fops[] = {
2324 [KVM_STAT_VCPU] = &vcpu_stat_fops,
2325 [KVM_STAT_VM] = &vm_stat_fops,
2328 static void kvm_init_debug(void)
2330 struct kvm_stats_debugfs_item *p;
2332 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2333 for (p = debugfs_entries; p->name; ++p)
2334 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2335 (void *)(long)p->offset,
2336 stat_fops[p->kind]);
2339 static void kvm_exit_debug(void)
2341 struct kvm_stats_debugfs_item *p;
2343 for (p = debugfs_entries; p->name; ++p)
2344 debugfs_remove(p->dentry);
2345 debugfs_remove(kvm_debugfs_dir);
2348 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2350 if (kvm_usage_count)
2351 hardware_disable(NULL);
2355 static int kvm_resume(struct sys_device *dev)
2357 if (kvm_usage_count) {
2358 WARN_ON(spin_is_locked(&kvm_lock));
2359 hardware_enable(NULL);
2364 static struct sysdev_class kvm_sysdev_class = {
2366 .suspend = kvm_suspend,
2367 .resume = kvm_resume,
2370 static struct sys_device kvm_sysdev = {
2372 .cls = &kvm_sysdev_class,
2375 struct page *bad_page;
2379 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2381 return container_of(pn, struct kvm_vcpu, preempt_notifier);
2384 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2386 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2388 kvm_arch_vcpu_load(vcpu, cpu);
2391 static void kvm_sched_out(struct preempt_notifier *pn,
2392 struct task_struct *next)
2394 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2396 kvm_arch_vcpu_put(vcpu);
2399 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2400 struct module *module)
2405 r = kvm_arch_init(opaque);
2409 bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2411 if (bad_page == NULL) {
2416 bad_pfn = page_to_pfn(bad_page);
2418 hwpoison_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2420 if (hwpoison_page == NULL) {
2425 hwpoison_pfn = page_to_pfn(hwpoison_page);
2427 fault_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2429 if (fault_page == NULL) {
2434 fault_pfn = page_to_pfn(fault_page);
2436 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2441 r = kvm_arch_hardware_setup();
2445 for_each_online_cpu(cpu) {
2446 smp_call_function_single(cpu,
2447 kvm_arch_check_processor_compat,
2453 r = register_cpu_notifier(&kvm_cpu_notifier);
2456 register_reboot_notifier(&kvm_reboot_notifier);
2458 r = sysdev_class_register(&kvm_sysdev_class);
2462 r = sysdev_register(&kvm_sysdev);
2466 /* A kmem cache lets us meet the alignment requirements of fx_save. */
2468 vcpu_align = __alignof__(struct kvm_vcpu);
2469 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2471 if (!kvm_vcpu_cache) {
2476 r = kvm_async_pf_init();
2480 kvm_chardev_ops.owner = module;
2481 kvm_vm_fops.owner = module;
2482 kvm_vcpu_fops.owner = module;
2484 r = misc_register(&kvm_dev);
2486 printk(KERN_ERR "kvm: misc device register failed\n");
2490 kvm_preempt_ops.sched_in = kvm_sched_in;
2491 kvm_preempt_ops.sched_out = kvm_sched_out;
2498 kvm_async_pf_deinit();
2500 kmem_cache_destroy(kvm_vcpu_cache);
2502 sysdev_unregister(&kvm_sysdev);
2504 sysdev_class_unregister(&kvm_sysdev_class);
2506 unregister_reboot_notifier(&kvm_reboot_notifier);
2507 unregister_cpu_notifier(&kvm_cpu_notifier);
2510 kvm_arch_hardware_unsetup();
2512 free_cpumask_var(cpus_hardware_enabled);
2515 __free_page(fault_page);
2517 __free_page(hwpoison_page);
2518 __free_page(bad_page);
2524 EXPORT_SYMBOL_GPL(kvm_init);
2529 misc_deregister(&kvm_dev);
2530 kmem_cache_destroy(kvm_vcpu_cache);
2531 kvm_async_pf_deinit();
2532 sysdev_unregister(&kvm_sysdev);
2533 sysdev_class_unregister(&kvm_sysdev_class);
2534 unregister_reboot_notifier(&kvm_reboot_notifier);
2535 unregister_cpu_notifier(&kvm_cpu_notifier);
2536 on_each_cpu(hardware_disable, NULL, 1);
2537 kvm_arch_hardware_unsetup();
2539 free_cpumask_var(cpus_hardware_enabled);
2540 __free_page(hwpoison_page);
2541 __free_page(bad_page);
2543 EXPORT_SYMBOL_GPL(kvm_exit);