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/syscore_ops.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>
56 #include "coalesced_mmio.h"
59 #define CREATE_TRACE_POINTS
60 #include <trace/events/kvm.h>
62 MODULE_AUTHOR("Qumranet");
63 MODULE_LICENSE("GPL");
68 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
71 DEFINE_RAW_SPINLOCK(kvm_lock);
74 static cpumask_var_t cpus_hardware_enabled;
75 static int kvm_usage_count = 0;
76 static atomic_t hardware_enable_failed;
78 struct kmem_cache *kvm_vcpu_cache;
79 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
81 static __read_mostly struct preempt_ops kvm_preempt_ops;
83 struct dentry *kvm_debugfs_dir;
85 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
88 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
91 static int hardware_enable_all(void);
92 static void hardware_disable_all(void);
94 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
97 EXPORT_SYMBOL_GPL(kvm_rebooting);
99 static bool largepages_enabled = true;
101 static struct page *hwpoison_page;
102 static pfn_t hwpoison_pfn;
104 static struct page *fault_page;
105 static pfn_t fault_pfn;
107 inline int kvm_is_mmio_pfn(pfn_t pfn)
109 if (pfn_valid(pfn)) {
111 struct page *tail = pfn_to_page(pfn);
112 struct page *head = compound_trans_head(tail);
113 reserved = PageReserved(head);
116 * "head" is not a dangling pointer
117 * (compound_trans_head takes care of that)
118 * but the hugepage may have been splitted
119 * from under us (and we may not hold a
120 * reference count on the head page so it can
121 * be reused before we run PageReferenced), so
122 * we've to check PageTail before returning
129 return PageReserved(tail);
136 * Switches to specified vcpu, until a matching vcpu_put()
138 void vcpu_load(struct kvm_vcpu *vcpu)
142 mutex_lock(&vcpu->mutex);
143 if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
144 /* The thread running this VCPU changed. */
145 struct pid *oldpid = vcpu->pid;
146 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
147 rcu_assign_pointer(vcpu->pid, newpid);
152 preempt_notifier_register(&vcpu->preempt_notifier);
153 kvm_arch_vcpu_load(vcpu, cpu);
157 void vcpu_put(struct kvm_vcpu *vcpu)
160 kvm_arch_vcpu_put(vcpu);
161 preempt_notifier_unregister(&vcpu->preempt_notifier);
163 mutex_unlock(&vcpu->mutex);
166 static void ack_flush(void *_completed)
170 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
175 struct kvm_vcpu *vcpu;
177 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
180 kvm_for_each_vcpu(i, vcpu, kvm) {
181 kvm_make_request(req, vcpu);
184 /* Set ->requests bit before we read ->mode */
187 if (cpus != NULL && cpu != -1 && cpu != me &&
188 kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
189 cpumask_set_cpu(cpu, cpus);
191 if (unlikely(cpus == NULL))
192 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
193 else if (!cpumask_empty(cpus))
194 smp_call_function_many(cpus, ack_flush, NULL, 1);
198 free_cpumask_var(cpus);
202 void kvm_flush_remote_tlbs(struct kvm *kvm)
204 int dirty_count = kvm->tlbs_dirty;
207 if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
208 ++kvm->stat.remote_tlb_flush;
209 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
212 void kvm_reload_remote_mmus(struct kvm *kvm)
214 make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
217 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
222 mutex_init(&vcpu->mutex);
227 init_waitqueue_head(&vcpu->wq);
228 kvm_async_pf_vcpu_init(vcpu);
230 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
235 vcpu->run = page_address(page);
237 r = kvm_arch_vcpu_init(vcpu);
243 free_page((unsigned long)vcpu->run);
247 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
249 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
252 kvm_arch_vcpu_uninit(vcpu);
253 free_page((unsigned long)vcpu->run);
255 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
257 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
258 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
260 return container_of(mn, struct kvm, mmu_notifier);
263 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
264 struct mm_struct *mm,
265 unsigned long address)
267 struct kvm *kvm = mmu_notifier_to_kvm(mn);
268 int need_tlb_flush, idx;
271 * When ->invalidate_page runs, the linux pte has been zapped
272 * already but the page is still allocated until
273 * ->invalidate_page returns. So if we increase the sequence
274 * here the kvm page fault will notice if the spte can't be
275 * established because the page is going to be freed. If
276 * instead the kvm page fault establishes the spte before
277 * ->invalidate_page runs, kvm_unmap_hva will release it
280 * The sequence increase only need to be seen at spin_unlock
281 * time, and not at spin_lock time.
283 * Increasing the sequence after the spin_unlock would be
284 * unsafe because the kvm page fault could then establish the
285 * pte after kvm_unmap_hva returned, without noticing the page
286 * is going to be freed.
288 idx = srcu_read_lock(&kvm->srcu);
289 spin_lock(&kvm->mmu_lock);
290 kvm->mmu_notifier_seq++;
291 need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
292 spin_unlock(&kvm->mmu_lock);
293 srcu_read_unlock(&kvm->srcu, idx);
295 /* we've to flush the tlb before the pages can be freed */
297 kvm_flush_remote_tlbs(kvm);
301 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
302 struct mm_struct *mm,
303 unsigned long address,
306 struct kvm *kvm = mmu_notifier_to_kvm(mn);
309 idx = srcu_read_lock(&kvm->srcu);
310 spin_lock(&kvm->mmu_lock);
311 kvm->mmu_notifier_seq++;
312 kvm_set_spte_hva(kvm, address, pte);
313 spin_unlock(&kvm->mmu_lock);
314 srcu_read_unlock(&kvm->srcu, idx);
317 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
318 struct mm_struct *mm,
322 struct kvm *kvm = mmu_notifier_to_kvm(mn);
323 int need_tlb_flush = 0, idx;
325 idx = srcu_read_lock(&kvm->srcu);
326 spin_lock(&kvm->mmu_lock);
328 * The count increase must become visible at unlock time as no
329 * spte can be established without taking the mmu_lock and
330 * count is also read inside the mmu_lock critical section.
332 kvm->mmu_notifier_count++;
333 for (; start < end; start += PAGE_SIZE)
334 need_tlb_flush |= kvm_unmap_hva(kvm, start);
335 need_tlb_flush |= kvm->tlbs_dirty;
336 spin_unlock(&kvm->mmu_lock);
337 srcu_read_unlock(&kvm->srcu, idx);
339 /* we've to flush the tlb before the pages can be freed */
341 kvm_flush_remote_tlbs(kvm);
344 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
345 struct mm_struct *mm,
349 struct kvm *kvm = mmu_notifier_to_kvm(mn);
351 spin_lock(&kvm->mmu_lock);
353 * This sequence increase will notify the kvm page fault that
354 * the page that is going to be mapped in the spte could have
357 kvm->mmu_notifier_seq++;
359 * The above sequence increase must be visible before the
360 * below count decrease but both values are read by the kvm
361 * page fault under mmu_lock spinlock so we don't need to add
362 * a smb_wmb() here in between the two.
364 kvm->mmu_notifier_count--;
365 spin_unlock(&kvm->mmu_lock);
367 BUG_ON(kvm->mmu_notifier_count < 0);
370 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
371 struct mm_struct *mm,
372 unsigned long address)
374 struct kvm *kvm = mmu_notifier_to_kvm(mn);
377 idx = srcu_read_lock(&kvm->srcu);
378 spin_lock(&kvm->mmu_lock);
379 young = kvm_age_hva(kvm, address);
380 spin_unlock(&kvm->mmu_lock);
381 srcu_read_unlock(&kvm->srcu, idx);
384 kvm_flush_remote_tlbs(kvm);
389 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
390 struct mm_struct *mm,
391 unsigned long address)
393 struct kvm *kvm = mmu_notifier_to_kvm(mn);
396 idx = srcu_read_lock(&kvm->srcu);
397 spin_lock(&kvm->mmu_lock);
398 young = kvm_test_age_hva(kvm, address);
399 spin_unlock(&kvm->mmu_lock);
400 srcu_read_unlock(&kvm->srcu, idx);
405 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
406 struct mm_struct *mm)
408 struct kvm *kvm = mmu_notifier_to_kvm(mn);
411 idx = srcu_read_lock(&kvm->srcu);
412 kvm_arch_flush_shadow(kvm);
413 srcu_read_unlock(&kvm->srcu, idx);
416 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
417 .invalidate_page = kvm_mmu_notifier_invalidate_page,
418 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
419 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
420 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
421 .test_young = kvm_mmu_notifier_test_young,
422 .change_pte = kvm_mmu_notifier_change_pte,
423 .release = kvm_mmu_notifier_release,
426 static int kvm_init_mmu_notifier(struct kvm *kvm)
428 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
429 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
432 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
434 static int kvm_init_mmu_notifier(struct kvm *kvm)
439 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
441 static struct kvm *kvm_create_vm(void)
444 struct kvm *kvm = kvm_arch_alloc_vm();
447 return ERR_PTR(-ENOMEM);
449 r = kvm_arch_init_vm(kvm);
451 goto out_err_nodisable;
453 r = hardware_enable_all();
455 goto out_err_nodisable;
457 #ifdef CONFIG_HAVE_KVM_IRQCHIP
458 INIT_HLIST_HEAD(&kvm->mask_notifier_list);
459 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
463 kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
466 if (init_srcu_struct(&kvm->srcu))
468 for (i = 0; i < KVM_NR_BUSES; i++) {
469 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
475 spin_lock_init(&kvm->mmu_lock);
476 kvm->mm = current->mm;
477 atomic_inc(&kvm->mm->mm_count);
478 kvm_eventfd_init(kvm);
479 mutex_init(&kvm->lock);
480 mutex_init(&kvm->irq_lock);
481 mutex_init(&kvm->slots_lock);
482 atomic_set(&kvm->users_count, 1);
484 r = kvm_init_mmu_notifier(kvm);
488 raw_spin_lock(&kvm_lock);
489 list_add(&kvm->vm_list, &vm_list);
490 raw_spin_unlock(&kvm_lock);
495 cleanup_srcu_struct(&kvm->srcu);
497 hardware_disable_all();
499 for (i = 0; i < KVM_NR_BUSES; i++)
500 kfree(kvm->buses[i]);
501 kfree(kvm->memslots);
502 kvm_arch_free_vm(kvm);
506 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
508 if (!memslot->dirty_bitmap)
511 if (2 * kvm_dirty_bitmap_bytes(memslot) > PAGE_SIZE)
512 vfree(memslot->dirty_bitmap_head);
514 kfree(memslot->dirty_bitmap_head);
516 memslot->dirty_bitmap = NULL;
517 memslot->dirty_bitmap_head = NULL;
521 * Free any memory in @free but not in @dont.
523 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
524 struct kvm_memory_slot *dont)
528 if (!dont || free->rmap != dont->rmap)
531 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
532 kvm_destroy_dirty_bitmap(free);
535 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
536 if (!dont || free->lpage_info[i] != dont->lpage_info[i]) {
537 vfree(free->lpage_info[i]);
538 free->lpage_info[i] = NULL;
546 void kvm_free_physmem(struct kvm *kvm)
549 struct kvm_memslots *slots = kvm->memslots;
551 for (i = 0; i < slots->nmemslots; ++i)
552 kvm_free_physmem_slot(&slots->memslots[i], NULL);
554 kfree(kvm->memslots);
557 static void kvm_destroy_vm(struct kvm *kvm)
560 struct mm_struct *mm = kvm->mm;
562 kvm_arch_sync_events(kvm);
563 raw_spin_lock(&kvm_lock);
564 list_del(&kvm->vm_list);
565 raw_spin_unlock(&kvm_lock);
566 kvm_free_irq_routing(kvm);
567 for (i = 0; i < KVM_NR_BUSES; i++)
568 kvm_io_bus_destroy(kvm->buses[i]);
569 kvm_coalesced_mmio_free(kvm);
570 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
571 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
573 kvm_arch_flush_shadow(kvm);
575 kvm_arch_destroy_vm(kvm);
576 kvm_free_physmem(kvm);
577 cleanup_srcu_struct(&kvm->srcu);
578 kvm_arch_free_vm(kvm);
579 hardware_disable_all();
583 void kvm_get_kvm(struct kvm *kvm)
585 atomic_inc(&kvm->users_count);
587 EXPORT_SYMBOL_GPL(kvm_get_kvm);
589 void kvm_put_kvm(struct kvm *kvm)
591 if (atomic_dec_and_test(&kvm->users_count))
594 EXPORT_SYMBOL_GPL(kvm_put_kvm);
597 static int kvm_vm_release(struct inode *inode, struct file *filp)
599 struct kvm *kvm = filp->private_data;
601 kvm_irqfd_release(kvm);
609 * Allocation size is twice as large as the actual dirty bitmap size.
610 * This makes it possible to do double buffering: see x86's
611 * kvm_vm_ioctl_get_dirty_log().
613 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
615 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
617 if (dirty_bytes > PAGE_SIZE)
618 memslot->dirty_bitmap = vzalloc(dirty_bytes);
620 memslot->dirty_bitmap = kzalloc(dirty_bytes, GFP_KERNEL);
622 if (!memslot->dirty_bitmap)
625 memslot->dirty_bitmap_head = memslot->dirty_bitmap;
628 #endif /* !CONFIG_S390 */
631 * Allocate some memory and give it an address in the guest physical address
634 * Discontiguous memory is allowed, mostly for framebuffers.
636 * Must be called holding mmap_sem for write.
638 int __kvm_set_memory_region(struct kvm *kvm,
639 struct kvm_userspace_memory_region *mem,
644 unsigned long npages;
646 struct kvm_memory_slot *memslot;
647 struct kvm_memory_slot old, new;
648 struct kvm_memslots *slots, *old_memslots;
651 /* General sanity checks */
652 if (mem->memory_size & (PAGE_SIZE - 1))
654 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
656 /* We can read the guest memory with __xxx_user() later on. */
658 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
659 !access_ok(VERIFY_WRITE,
660 (void __user *)(unsigned long)mem->userspace_addr,
663 if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
665 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
668 memslot = &kvm->memslots->memslots[mem->slot];
669 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
670 npages = mem->memory_size >> PAGE_SHIFT;
673 if (npages > KVM_MEM_MAX_NR_PAGES)
677 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
679 new = old = *memslot;
682 new.base_gfn = base_gfn;
684 new.flags = mem->flags;
686 /* Disallow changing a memory slot's size. */
688 if (npages && old.npages && npages != old.npages)
691 /* Check for overlaps */
693 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
694 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
696 if (s == memslot || !s->npages)
698 if (!((base_gfn + npages <= s->base_gfn) ||
699 (base_gfn >= s->base_gfn + s->npages)))
703 /* Free page dirty bitmap if unneeded */
704 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
705 new.dirty_bitmap = NULL;
709 /* Allocate if a slot is being created */
711 if (npages && !new.rmap) {
712 new.rmap = vzalloc(npages * sizeof(*new.rmap));
717 new.user_alloc = user_alloc;
718 new.userspace_addr = mem->userspace_addr;
723 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
729 /* Avoid unused variable warning if no large pages */
732 if (new.lpage_info[i])
735 lpages = 1 + ((base_gfn + npages - 1)
736 >> KVM_HPAGE_GFN_SHIFT(level));
737 lpages -= base_gfn >> KVM_HPAGE_GFN_SHIFT(level);
739 new.lpage_info[i] = vzalloc(lpages * sizeof(*new.lpage_info[i]));
741 if (!new.lpage_info[i])
744 if (base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
745 new.lpage_info[i][0].write_count = 1;
746 if ((base_gfn+npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
747 new.lpage_info[i][lpages - 1].write_count = 1;
748 ugfn = new.userspace_addr >> PAGE_SHIFT;
750 * If the gfn and userspace address are not aligned wrt each
751 * other, or if explicitly asked to, disable large page
752 * support for this slot
754 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
756 for (j = 0; j < lpages; ++j)
757 new.lpage_info[i][j].write_count = 1;
762 /* Allocate page dirty bitmap if needed */
763 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
764 if (kvm_create_dirty_bitmap(&new) < 0)
766 /* destroy any largepage mappings for dirty tracking */
768 #else /* not defined CONFIG_S390 */
769 new.user_alloc = user_alloc;
771 new.userspace_addr = mem->userspace_addr;
772 #endif /* not defined CONFIG_S390 */
776 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
779 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
780 if (mem->slot >= slots->nmemslots)
781 slots->nmemslots = mem->slot + 1;
783 slots->memslots[mem->slot].flags |= KVM_MEMSLOT_INVALID;
785 old_memslots = kvm->memslots;
786 rcu_assign_pointer(kvm->memslots, slots);
787 synchronize_srcu_expedited(&kvm->srcu);
788 /* From this point no new shadow pages pointing to a deleted
789 * memslot will be created.
791 * validation of sp->gfn happens in:
792 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
793 * - kvm_is_visible_gfn (mmu_check_roots)
795 kvm_arch_flush_shadow(kvm);
799 r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
803 /* map the pages in iommu page table */
805 r = kvm_iommu_map_pages(kvm, &new);
811 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
814 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
815 if (mem->slot >= slots->nmemslots)
816 slots->nmemslots = mem->slot + 1;
819 /* actual memory is freed via old in kvm_free_physmem_slot below */
822 new.dirty_bitmap = NULL;
823 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i)
824 new.lpage_info[i] = NULL;
827 slots->memslots[mem->slot] = new;
828 old_memslots = kvm->memslots;
829 rcu_assign_pointer(kvm->memslots, slots);
830 synchronize_srcu_expedited(&kvm->srcu);
832 kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
834 kvm_free_physmem_slot(&old, &new);
840 kvm_free_physmem_slot(&new, &old);
845 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
847 int kvm_set_memory_region(struct kvm *kvm,
848 struct kvm_userspace_memory_region *mem,
853 mutex_lock(&kvm->slots_lock);
854 r = __kvm_set_memory_region(kvm, mem, user_alloc);
855 mutex_unlock(&kvm->slots_lock);
858 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
860 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
862 kvm_userspace_memory_region *mem,
865 if (mem->slot >= KVM_MEMORY_SLOTS)
867 return kvm_set_memory_region(kvm, mem, user_alloc);
870 int kvm_get_dirty_log(struct kvm *kvm,
871 struct kvm_dirty_log *log, int *is_dirty)
873 struct kvm_memory_slot *memslot;
876 unsigned long any = 0;
879 if (log->slot >= KVM_MEMORY_SLOTS)
882 memslot = &kvm->memslots->memslots[log->slot];
884 if (!memslot->dirty_bitmap)
887 n = kvm_dirty_bitmap_bytes(memslot);
889 for (i = 0; !any && i < n/sizeof(long); ++i)
890 any = memslot->dirty_bitmap[i];
893 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
904 void kvm_disable_largepages(void)
906 largepages_enabled = false;
908 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
910 int is_error_page(struct page *page)
912 return page == bad_page || page == hwpoison_page || page == fault_page;
914 EXPORT_SYMBOL_GPL(is_error_page);
916 int is_error_pfn(pfn_t pfn)
918 return pfn == bad_pfn || pfn == hwpoison_pfn || pfn == fault_pfn;
920 EXPORT_SYMBOL_GPL(is_error_pfn);
922 int is_hwpoison_pfn(pfn_t pfn)
924 return pfn == hwpoison_pfn;
926 EXPORT_SYMBOL_GPL(is_hwpoison_pfn);
928 int is_fault_pfn(pfn_t pfn)
930 return pfn == fault_pfn;
932 EXPORT_SYMBOL_GPL(is_fault_pfn);
934 static inline unsigned long bad_hva(void)
939 int kvm_is_error_hva(unsigned long addr)
941 return addr == bad_hva();
943 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
945 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm_memslots *slots,
950 for (i = 0; i < slots->nmemslots; ++i) {
951 struct kvm_memory_slot *memslot = &slots->memslots[i];
953 if (gfn >= memslot->base_gfn
954 && gfn < memslot->base_gfn + memslot->npages)
960 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
962 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
964 EXPORT_SYMBOL_GPL(gfn_to_memslot);
966 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
969 struct kvm_memslots *slots = kvm_memslots(kvm);
971 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
972 struct kvm_memory_slot *memslot = &slots->memslots[i];
974 if (memslot->flags & KVM_MEMSLOT_INVALID)
977 if (gfn >= memslot->base_gfn
978 && gfn < memslot->base_gfn + memslot->npages)
983 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
985 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
987 struct vm_area_struct *vma;
988 unsigned long addr, size;
992 addr = gfn_to_hva(kvm, gfn);
993 if (kvm_is_error_hva(addr))
996 down_read(¤t->mm->mmap_sem);
997 vma = find_vma(current->mm, addr);
1001 size = vma_kernel_pagesize(vma);
1004 up_read(¤t->mm->mmap_sem);
1009 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1012 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1016 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1018 return gfn_to_hva_memslot(slot, gfn);
1021 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1023 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1025 EXPORT_SYMBOL_GPL(gfn_to_hva);
1027 static pfn_t get_fault_pfn(void)
1029 get_page(fault_page);
1033 int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1034 unsigned long start, int write, struct page **page)
1036 int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1039 flags |= FOLL_WRITE;
1041 return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1044 static inline int check_user_page_hwpoison(unsigned long addr)
1046 int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1048 rc = __get_user_pages(current, current->mm, addr, 1,
1049 flags, NULL, NULL, NULL);
1050 return rc == -EHWPOISON;
1053 static pfn_t hva_to_pfn(struct kvm *kvm, unsigned long addr, bool atomic,
1054 bool *async, bool write_fault, bool *writable)
1056 struct page *page[1];
1060 /* we can do it either atomically or asynchronously, not both */
1061 BUG_ON(atomic && async);
1063 BUG_ON(!write_fault && !writable);
1068 if (atomic || async)
1069 npages = __get_user_pages_fast(addr, 1, 1, page);
1071 if (unlikely(npages != 1) && !atomic) {
1075 *writable = write_fault;
1078 down_read(¤t->mm->mmap_sem);
1079 npages = get_user_page_nowait(current, current->mm,
1080 addr, write_fault, page);
1081 up_read(¤t->mm->mmap_sem);
1083 npages = get_user_pages_fast(addr, 1, write_fault,
1086 /* map read fault as writable if possible */
1087 if (unlikely(!write_fault) && npages == 1) {
1088 struct page *wpage[1];
1090 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1100 if (unlikely(npages != 1)) {
1101 struct vm_area_struct *vma;
1104 return get_fault_pfn();
1106 down_read(¤t->mm->mmap_sem);
1107 if (npages == -EHWPOISON ||
1108 (!async && check_user_page_hwpoison(addr))) {
1109 up_read(¤t->mm->mmap_sem);
1110 get_page(hwpoison_page);
1111 return page_to_pfn(hwpoison_page);
1114 vma = find_vma_intersection(current->mm, addr, addr+1);
1117 pfn = get_fault_pfn();
1118 else if ((vma->vm_flags & VM_PFNMAP)) {
1119 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1121 BUG_ON(!kvm_is_mmio_pfn(pfn));
1123 if (async && (vma->vm_flags & VM_WRITE))
1125 pfn = get_fault_pfn();
1127 up_read(¤t->mm->mmap_sem);
1129 pfn = page_to_pfn(page[0]);
1134 pfn_t hva_to_pfn_atomic(struct kvm *kvm, unsigned long addr)
1136 return hva_to_pfn(kvm, addr, true, NULL, true, NULL);
1138 EXPORT_SYMBOL_GPL(hva_to_pfn_atomic);
1140 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1141 bool write_fault, bool *writable)
1148 addr = gfn_to_hva(kvm, gfn);
1149 if (kvm_is_error_hva(addr)) {
1151 return page_to_pfn(bad_page);
1154 return hva_to_pfn(kvm, addr, atomic, async, write_fault, writable);
1157 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1159 return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1161 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1163 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1164 bool write_fault, bool *writable)
1166 return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1168 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1170 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1172 return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1174 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1176 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1179 return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1181 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1183 pfn_t gfn_to_pfn_memslot(struct kvm *kvm,
1184 struct kvm_memory_slot *slot, gfn_t gfn)
1186 unsigned long addr = gfn_to_hva_memslot(slot, gfn);
1187 return hva_to_pfn(kvm, addr, false, NULL, true, NULL);
1190 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1196 addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1197 if (kvm_is_error_hva(addr))
1200 if (entry < nr_pages)
1203 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1205 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1207 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1211 pfn = gfn_to_pfn(kvm, gfn);
1212 if (!kvm_is_mmio_pfn(pfn))
1213 return pfn_to_page(pfn);
1215 WARN_ON(kvm_is_mmio_pfn(pfn));
1221 EXPORT_SYMBOL_GPL(gfn_to_page);
1223 void kvm_release_page_clean(struct page *page)
1225 kvm_release_pfn_clean(page_to_pfn(page));
1227 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1229 void kvm_release_pfn_clean(pfn_t pfn)
1231 if (!kvm_is_mmio_pfn(pfn))
1232 put_page(pfn_to_page(pfn));
1234 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1236 void kvm_release_page_dirty(struct page *page)
1238 kvm_release_pfn_dirty(page_to_pfn(page));
1240 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1242 void kvm_release_pfn_dirty(pfn_t pfn)
1244 kvm_set_pfn_dirty(pfn);
1245 kvm_release_pfn_clean(pfn);
1247 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1249 void kvm_set_page_dirty(struct page *page)
1251 kvm_set_pfn_dirty(page_to_pfn(page));
1253 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1255 void kvm_set_pfn_dirty(pfn_t pfn)
1257 if (!kvm_is_mmio_pfn(pfn)) {
1258 struct page *page = pfn_to_page(pfn);
1259 if (!PageReserved(page))
1263 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1265 void kvm_set_pfn_accessed(pfn_t pfn)
1267 if (!kvm_is_mmio_pfn(pfn))
1268 mark_page_accessed(pfn_to_page(pfn));
1270 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1272 void kvm_get_pfn(pfn_t pfn)
1274 if (!kvm_is_mmio_pfn(pfn))
1275 get_page(pfn_to_page(pfn));
1277 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1279 static int next_segment(unsigned long len, int offset)
1281 if (len > PAGE_SIZE - offset)
1282 return PAGE_SIZE - offset;
1287 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1293 addr = gfn_to_hva(kvm, gfn);
1294 if (kvm_is_error_hva(addr))
1296 r = __copy_from_user(data, (void __user *)addr + offset, len);
1301 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1303 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1305 gfn_t gfn = gpa >> PAGE_SHIFT;
1307 int offset = offset_in_page(gpa);
1310 while ((seg = next_segment(len, offset)) != 0) {
1311 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1321 EXPORT_SYMBOL_GPL(kvm_read_guest);
1323 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1328 gfn_t gfn = gpa >> PAGE_SHIFT;
1329 int offset = offset_in_page(gpa);
1331 addr = gfn_to_hva(kvm, gfn);
1332 if (kvm_is_error_hva(addr))
1334 pagefault_disable();
1335 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1341 EXPORT_SYMBOL(kvm_read_guest_atomic);
1343 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1344 int offset, int len)
1349 addr = gfn_to_hva(kvm, gfn);
1350 if (kvm_is_error_hva(addr))
1352 r = __copy_to_user((void __user *)addr + offset, data, len);
1355 mark_page_dirty(kvm, gfn);
1358 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1360 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1363 gfn_t gfn = gpa >> PAGE_SHIFT;
1365 int offset = offset_in_page(gpa);
1368 while ((seg = next_segment(len, offset)) != 0) {
1369 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1380 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1383 struct kvm_memslots *slots = kvm_memslots(kvm);
1384 int offset = offset_in_page(gpa);
1385 gfn_t gfn = gpa >> PAGE_SHIFT;
1388 ghc->generation = slots->generation;
1389 ghc->memslot = __gfn_to_memslot(slots, gfn);
1390 ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1391 if (!kvm_is_error_hva(ghc->hva))
1398 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1400 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1401 void *data, unsigned long len)
1403 struct kvm_memslots *slots = kvm_memslots(kvm);
1406 if (slots->generation != ghc->generation)
1407 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1409 if (kvm_is_error_hva(ghc->hva))
1412 r = __copy_to_user((void __user *)ghc->hva, data, len);
1415 mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1419 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1421 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1423 return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1426 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1428 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1430 gfn_t gfn = gpa >> PAGE_SHIFT;
1432 int offset = offset_in_page(gpa);
1435 while ((seg = next_segment(len, offset)) != 0) {
1436 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1445 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1447 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1450 if (memslot && memslot->dirty_bitmap) {
1451 unsigned long rel_gfn = gfn - memslot->base_gfn;
1453 __set_bit_le(rel_gfn, memslot->dirty_bitmap);
1457 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1459 struct kvm_memory_slot *memslot;
1461 memslot = gfn_to_memslot(kvm, gfn);
1462 mark_page_dirty_in_slot(kvm, memslot, gfn);
1466 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1468 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1473 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1475 if (kvm_arch_vcpu_runnable(vcpu)) {
1476 kvm_make_request(KVM_REQ_UNHALT, vcpu);
1479 if (kvm_cpu_has_pending_timer(vcpu))
1481 if (signal_pending(current))
1487 finish_wait(&vcpu->wq, &wait);
1490 void kvm_resched(struct kvm_vcpu *vcpu)
1492 if (!need_resched())
1496 EXPORT_SYMBOL_GPL(kvm_resched);
1498 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1500 struct kvm *kvm = me->kvm;
1501 struct kvm_vcpu *vcpu;
1502 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1508 * We boost the priority of a VCPU that is runnable but not
1509 * currently running, because it got preempted by something
1510 * else and called schedule in __vcpu_run. Hopefully that
1511 * VCPU is holding the lock that we need and will release it.
1512 * We approximate round-robin by starting at the last boosted VCPU.
1514 for (pass = 0; pass < 2 && !yielded; pass++) {
1515 kvm_for_each_vcpu(i, vcpu, kvm) {
1516 struct task_struct *task = NULL;
1518 if (!pass && i < last_boosted_vcpu) {
1519 i = last_boosted_vcpu;
1521 } else if (pass && i > last_boosted_vcpu)
1525 if (waitqueue_active(&vcpu->wq))
1528 pid = rcu_dereference(vcpu->pid);
1530 task = get_pid_task(vcpu->pid, PIDTYPE_PID);
1534 if (task->flags & PF_VCPU) {
1535 put_task_struct(task);
1538 if (yield_to(task, 1)) {
1539 put_task_struct(task);
1540 kvm->last_boosted_vcpu = i;
1544 put_task_struct(task);
1548 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1550 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1552 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1555 if (vmf->pgoff == 0)
1556 page = virt_to_page(vcpu->run);
1558 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1559 page = virt_to_page(vcpu->arch.pio_data);
1561 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1562 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1563 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1566 return VM_FAULT_SIGBUS;
1572 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1573 .fault = kvm_vcpu_fault,
1576 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1578 vma->vm_ops = &kvm_vcpu_vm_ops;
1582 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1584 struct kvm_vcpu *vcpu = filp->private_data;
1586 kvm_put_kvm(vcpu->kvm);
1590 static struct file_operations kvm_vcpu_fops = {
1591 .release = kvm_vcpu_release,
1592 .unlocked_ioctl = kvm_vcpu_ioctl,
1593 #ifdef CONFIG_COMPAT
1594 .compat_ioctl = kvm_vcpu_compat_ioctl,
1596 .mmap = kvm_vcpu_mmap,
1597 .llseek = noop_llseek,
1601 * Allocates an inode for the vcpu.
1603 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1605 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1609 * Creates some virtual cpus. Good luck creating more than one.
1611 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1614 struct kvm_vcpu *vcpu, *v;
1616 vcpu = kvm_arch_vcpu_create(kvm, id);
1618 return PTR_ERR(vcpu);
1620 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1622 r = kvm_arch_vcpu_setup(vcpu);
1626 mutex_lock(&kvm->lock);
1627 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1629 goto unlock_vcpu_destroy;
1632 kvm_for_each_vcpu(r, v, kvm)
1633 if (v->vcpu_id == id) {
1635 goto unlock_vcpu_destroy;
1638 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1640 /* Now it's all set up, let userspace reach it */
1642 r = create_vcpu_fd(vcpu);
1645 goto unlock_vcpu_destroy;
1648 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1650 atomic_inc(&kvm->online_vcpus);
1652 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1653 if (kvm->bsp_vcpu_id == id)
1654 kvm->bsp_vcpu = vcpu;
1656 mutex_unlock(&kvm->lock);
1659 unlock_vcpu_destroy:
1660 mutex_unlock(&kvm->lock);
1662 kvm_arch_vcpu_destroy(vcpu);
1666 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1669 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1670 vcpu->sigset_active = 1;
1671 vcpu->sigset = *sigset;
1673 vcpu->sigset_active = 0;
1677 static long kvm_vcpu_ioctl(struct file *filp,
1678 unsigned int ioctl, unsigned long arg)
1680 struct kvm_vcpu *vcpu = filp->private_data;
1681 void __user *argp = (void __user *)arg;
1683 struct kvm_fpu *fpu = NULL;
1684 struct kvm_sregs *kvm_sregs = NULL;
1686 if (vcpu->kvm->mm != current->mm)
1689 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1691 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1692 * so vcpu_load() would break it.
1694 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1695 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1705 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1706 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1708 case KVM_GET_REGS: {
1709 struct kvm_regs *kvm_regs;
1712 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1715 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1719 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1726 case KVM_SET_REGS: {
1727 struct kvm_regs *kvm_regs;
1730 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1734 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1736 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1744 case KVM_GET_SREGS: {
1745 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1749 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1753 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1758 case KVM_SET_SREGS: {
1759 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1764 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1766 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1772 case KVM_GET_MP_STATE: {
1773 struct kvm_mp_state mp_state;
1775 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1779 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1784 case KVM_SET_MP_STATE: {
1785 struct kvm_mp_state mp_state;
1788 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1790 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1796 case KVM_TRANSLATE: {
1797 struct kvm_translation tr;
1800 if (copy_from_user(&tr, argp, sizeof tr))
1802 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1806 if (copy_to_user(argp, &tr, sizeof tr))
1811 case KVM_SET_GUEST_DEBUG: {
1812 struct kvm_guest_debug dbg;
1815 if (copy_from_user(&dbg, argp, sizeof dbg))
1817 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1823 case KVM_SET_SIGNAL_MASK: {
1824 struct kvm_signal_mask __user *sigmask_arg = argp;
1825 struct kvm_signal_mask kvm_sigmask;
1826 sigset_t sigset, *p;
1831 if (copy_from_user(&kvm_sigmask, argp,
1832 sizeof kvm_sigmask))
1835 if (kvm_sigmask.len != sizeof sigset)
1838 if (copy_from_user(&sigset, sigmask_arg->sigset,
1843 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
1847 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1851 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1855 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1861 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1866 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1868 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1875 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1884 #ifdef CONFIG_COMPAT
1885 static long kvm_vcpu_compat_ioctl(struct file *filp,
1886 unsigned int ioctl, unsigned long arg)
1888 struct kvm_vcpu *vcpu = filp->private_data;
1889 void __user *argp = compat_ptr(arg);
1892 if (vcpu->kvm->mm != current->mm)
1896 case KVM_SET_SIGNAL_MASK: {
1897 struct kvm_signal_mask __user *sigmask_arg = argp;
1898 struct kvm_signal_mask kvm_sigmask;
1899 compat_sigset_t csigset;
1904 if (copy_from_user(&kvm_sigmask, argp,
1905 sizeof kvm_sigmask))
1908 if (kvm_sigmask.len != sizeof csigset)
1911 if (copy_from_user(&csigset, sigmask_arg->sigset,
1915 sigset_from_compat(&sigset, &csigset);
1916 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1920 r = kvm_vcpu_ioctl(filp, ioctl, arg);
1928 static long kvm_vm_ioctl(struct file *filp,
1929 unsigned int ioctl, unsigned long arg)
1931 struct kvm *kvm = filp->private_data;
1932 void __user *argp = (void __user *)arg;
1935 if (kvm->mm != current->mm)
1938 case KVM_CREATE_VCPU:
1939 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1943 case KVM_SET_USER_MEMORY_REGION: {
1944 struct kvm_userspace_memory_region kvm_userspace_mem;
1947 if (copy_from_user(&kvm_userspace_mem, argp,
1948 sizeof kvm_userspace_mem))
1951 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1956 case KVM_GET_DIRTY_LOG: {
1957 struct kvm_dirty_log log;
1960 if (copy_from_user(&log, argp, sizeof log))
1962 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1967 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1968 case KVM_REGISTER_COALESCED_MMIO: {
1969 struct kvm_coalesced_mmio_zone zone;
1971 if (copy_from_user(&zone, argp, sizeof zone))
1973 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1979 case KVM_UNREGISTER_COALESCED_MMIO: {
1980 struct kvm_coalesced_mmio_zone zone;
1982 if (copy_from_user(&zone, argp, sizeof zone))
1984 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1992 struct kvm_irqfd data;
1995 if (copy_from_user(&data, argp, sizeof data))
1997 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
2000 case KVM_IOEVENTFD: {
2001 struct kvm_ioeventfd data;
2004 if (copy_from_user(&data, argp, sizeof data))
2006 r = kvm_ioeventfd(kvm, &data);
2009 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2010 case KVM_SET_BOOT_CPU_ID:
2012 mutex_lock(&kvm->lock);
2013 if (atomic_read(&kvm->online_vcpus) != 0)
2016 kvm->bsp_vcpu_id = arg;
2017 mutex_unlock(&kvm->lock);
2021 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2023 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2029 #ifdef CONFIG_COMPAT
2030 struct compat_kvm_dirty_log {
2034 compat_uptr_t dirty_bitmap; /* one bit per page */
2039 static long kvm_vm_compat_ioctl(struct file *filp,
2040 unsigned int ioctl, unsigned long arg)
2042 struct kvm *kvm = filp->private_data;
2045 if (kvm->mm != current->mm)
2048 case KVM_GET_DIRTY_LOG: {
2049 struct compat_kvm_dirty_log compat_log;
2050 struct kvm_dirty_log log;
2053 if (copy_from_user(&compat_log, (void __user *)arg,
2054 sizeof(compat_log)))
2056 log.slot = compat_log.slot;
2057 log.padding1 = compat_log.padding1;
2058 log.padding2 = compat_log.padding2;
2059 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2061 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2067 r = kvm_vm_ioctl(filp, ioctl, arg);
2075 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2077 struct page *page[1];
2080 gfn_t gfn = vmf->pgoff;
2081 struct kvm *kvm = vma->vm_file->private_data;
2083 addr = gfn_to_hva(kvm, gfn);
2084 if (kvm_is_error_hva(addr))
2085 return VM_FAULT_SIGBUS;
2087 npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2089 if (unlikely(npages != 1))
2090 return VM_FAULT_SIGBUS;
2092 vmf->page = page[0];
2096 static const struct vm_operations_struct kvm_vm_vm_ops = {
2097 .fault = kvm_vm_fault,
2100 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2102 vma->vm_ops = &kvm_vm_vm_ops;
2106 static struct file_operations kvm_vm_fops = {
2107 .release = kvm_vm_release,
2108 .unlocked_ioctl = kvm_vm_ioctl,
2109 #ifdef CONFIG_COMPAT
2110 .compat_ioctl = kvm_vm_compat_ioctl,
2112 .mmap = kvm_vm_mmap,
2113 .llseek = noop_llseek,
2116 static int kvm_dev_ioctl_create_vm(void)
2121 kvm = kvm_create_vm();
2123 return PTR_ERR(kvm);
2124 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2125 r = kvm_coalesced_mmio_init(kvm);
2131 r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2138 static long kvm_dev_ioctl_check_extension_generic(long arg)
2141 case KVM_CAP_USER_MEMORY:
2142 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2143 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2144 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2145 case KVM_CAP_SET_BOOT_CPU_ID:
2147 case KVM_CAP_INTERNAL_ERROR_DATA:
2149 #ifdef CONFIG_HAVE_KVM_IRQCHIP
2150 case KVM_CAP_IRQ_ROUTING:
2151 return KVM_MAX_IRQ_ROUTES;
2156 return kvm_dev_ioctl_check_extension(arg);
2159 static long kvm_dev_ioctl(struct file *filp,
2160 unsigned int ioctl, unsigned long arg)
2165 case KVM_GET_API_VERSION:
2169 r = KVM_API_VERSION;
2175 r = kvm_dev_ioctl_create_vm();
2177 case KVM_CHECK_EXTENSION:
2178 r = kvm_dev_ioctl_check_extension_generic(arg);
2180 case KVM_GET_VCPU_MMAP_SIZE:
2184 r = PAGE_SIZE; /* struct kvm_run */
2186 r += PAGE_SIZE; /* pio data page */
2188 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2189 r += PAGE_SIZE; /* coalesced mmio ring page */
2192 case KVM_TRACE_ENABLE:
2193 case KVM_TRACE_PAUSE:
2194 case KVM_TRACE_DISABLE:
2198 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2204 static struct file_operations kvm_chardev_ops = {
2205 .unlocked_ioctl = kvm_dev_ioctl,
2206 .compat_ioctl = kvm_dev_ioctl,
2207 .llseek = noop_llseek,
2210 static struct miscdevice kvm_dev = {
2216 static void hardware_enable_nolock(void *junk)
2218 int cpu = raw_smp_processor_id();
2221 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2224 cpumask_set_cpu(cpu, cpus_hardware_enabled);
2226 r = kvm_arch_hardware_enable(NULL);
2229 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2230 atomic_inc(&hardware_enable_failed);
2231 printk(KERN_INFO "kvm: enabling virtualization on "
2232 "CPU%d failed\n", cpu);
2236 static void hardware_enable(void *junk)
2238 raw_spin_lock(&kvm_lock);
2239 hardware_enable_nolock(junk);
2240 raw_spin_unlock(&kvm_lock);
2243 static void hardware_disable_nolock(void *junk)
2245 int cpu = raw_smp_processor_id();
2247 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2249 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2250 kvm_arch_hardware_disable(NULL);
2253 static void hardware_disable(void *junk)
2255 raw_spin_lock(&kvm_lock);
2256 hardware_disable_nolock(junk);
2257 raw_spin_unlock(&kvm_lock);
2260 static void hardware_disable_all_nolock(void)
2262 BUG_ON(!kvm_usage_count);
2265 if (!kvm_usage_count)
2266 on_each_cpu(hardware_disable_nolock, NULL, 1);
2269 static void hardware_disable_all(void)
2271 raw_spin_lock(&kvm_lock);
2272 hardware_disable_all_nolock();
2273 raw_spin_unlock(&kvm_lock);
2276 static int hardware_enable_all(void)
2280 raw_spin_lock(&kvm_lock);
2283 if (kvm_usage_count == 1) {
2284 atomic_set(&hardware_enable_failed, 0);
2285 on_each_cpu(hardware_enable_nolock, NULL, 1);
2287 if (atomic_read(&hardware_enable_failed)) {
2288 hardware_disable_all_nolock();
2293 raw_spin_unlock(&kvm_lock);
2298 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2303 if (!kvm_usage_count)
2306 val &= ~CPU_TASKS_FROZEN;
2309 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2311 hardware_disable(NULL);
2314 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2316 hardware_enable(NULL);
2323 asmlinkage void kvm_spurious_fault(void)
2325 /* Fault while not rebooting. We want the trace. */
2328 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
2330 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2334 * Some (well, at least mine) BIOSes hang on reboot if
2337 * And Intel TXT required VMX off for all cpu when system shutdown.
2339 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2340 kvm_rebooting = true;
2341 on_each_cpu(hardware_disable_nolock, NULL, 1);
2345 static struct notifier_block kvm_reboot_notifier = {
2346 .notifier_call = kvm_reboot,
2350 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2354 for (i = 0; i < bus->dev_count; i++) {
2355 struct kvm_io_device *pos = bus->devs[i];
2357 kvm_iodevice_destructor(pos);
2362 /* kvm_io_bus_write - called under kvm->slots_lock */
2363 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2364 int len, const void *val)
2367 struct kvm_io_bus *bus;
2369 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2370 for (i = 0; i < bus->dev_count; i++)
2371 if (!kvm_iodevice_write(bus->devs[i], addr, len, val))
2376 /* kvm_io_bus_read - called under kvm->slots_lock */
2377 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2381 struct kvm_io_bus *bus;
2383 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2384 for (i = 0; i < bus->dev_count; i++)
2385 if (!kvm_iodevice_read(bus->devs[i], addr, len, val))
2390 /* Caller must hold slots_lock. */
2391 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2392 struct kvm_io_device *dev)
2394 struct kvm_io_bus *new_bus, *bus;
2396 bus = kvm->buses[bus_idx];
2397 if (bus->dev_count > NR_IOBUS_DEVS-1)
2400 new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2403 memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2404 new_bus->devs[new_bus->dev_count++] = dev;
2405 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2406 synchronize_srcu_expedited(&kvm->srcu);
2412 /* Caller must hold slots_lock. */
2413 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2414 struct kvm_io_device *dev)
2417 struct kvm_io_bus *new_bus, *bus;
2419 new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2423 bus = kvm->buses[bus_idx];
2424 memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2427 for (i = 0; i < new_bus->dev_count; i++)
2428 if (new_bus->devs[i] == dev) {
2430 new_bus->devs[i] = new_bus->devs[--new_bus->dev_count];
2439 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2440 synchronize_srcu_expedited(&kvm->srcu);
2445 static struct notifier_block kvm_cpu_notifier = {
2446 .notifier_call = kvm_cpu_hotplug,
2449 static int vm_stat_get(void *_offset, u64 *val)
2451 unsigned offset = (long)_offset;
2455 raw_spin_lock(&kvm_lock);
2456 list_for_each_entry(kvm, &vm_list, vm_list)
2457 *val += *(u32 *)((void *)kvm + offset);
2458 raw_spin_unlock(&kvm_lock);
2462 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2464 static int vcpu_stat_get(void *_offset, u64 *val)
2466 unsigned offset = (long)_offset;
2468 struct kvm_vcpu *vcpu;
2472 raw_spin_lock(&kvm_lock);
2473 list_for_each_entry(kvm, &vm_list, vm_list)
2474 kvm_for_each_vcpu(i, vcpu, kvm)
2475 *val += *(u32 *)((void *)vcpu + offset);
2477 raw_spin_unlock(&kvm_lock);
2481 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2483 static const struct file_operations *stat_fops[] = {
2484 [KVM_STAT_VCPU] = &vcpu_stat_fops,
2485 [KVM_STAT_VM] = &vm_stat_fops,
2488 static void kvm_init_debug(void)
2490 struct kvm_stats_debugfs_item *p;
2492 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2493 for (p = debugfs_entries; p->name; ++p)
2494 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2495 (void *)(long)p->offset,
2496 stat_fops[p->kind]);
2499 static void kvm_exit_debug(void)
2501 struct kvm_stats_debugfs_item *p;
2503 for (p = debugfs_entries; p->name; ++p)
2504 debugfs_remove(p->dentry);
2505 debugfs_remove(kvm_debugfs_dir);
2508 static int kvm_suspend(void)
2510 if (kvm_usage_count)
2511 hardware_disable_nolock(NULL);
2515 static void kvm_resume(void)
2517 if (kvm_usage_count) {
2518 WARN_ON(raw_spin_is_locked(&kvm_lock));
2519 hardware_enable_nolock(NULL);
2523 static struct syscore_ops kvm_syscore_ops = {
2524 .suspend = kvm_suspend,
2525 .resume = kvm_resume,
2528 struct page *bad_page;
2532 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2534 return container_of(pn, struct kvm_vcpu, preempt_notifier);
2537 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2539 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2541 kvm_arch_vcpu_load(vcpu, cpu);
2544 static void kvm_sched_out(struct preempt_notifier *pn,
2545 struct task_struct *next)
2547 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2549 kvm_arch_vcpu_put(vcpu);
2552 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2553 struct module *module)
2558 r = kvm_arch_init(opaque);
2562 bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2564 if (bad_page == NULL) {
2569 bad_pfn = page_to_pfn(bad_page);
2571 hwpoison_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2573 if (hwpoison_page == NULL) {
2578 hwpoison_pfn = page_to_pfn(hwpoison_page);
2580 fault_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2582 if (fault_page == NULL) {
2587 fault_pfn = page_to_pfn(fault_page);
2589 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2594 r = kvm_arch_hardware_setup();
2598 for_each_online_cpu(cpu) {
2599 smp_call_function_single(cpu,
2600 kvm_arch_check_processor_compat,
2606 r = register_cpu_notifier(&kvm_cpu_notifier);
2609 register_reboot_notifier(&kvm_reboot_notifier);
2611 /* A kmem cache lets us meet the alignment requirements of fx_save. */
2613 vcpu_align = __alignof__(struct kvm_vcpu);
2614 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2616 if (!kvm_vcpu_cache) {
2621 r = kvm_async_pf_init();
2625 kvm_chardev_ops.owner = module;
2626 kvm_vm_fops.owner = module;
2627 kvm_vcpu_fops.owner = module;
2629 r = misc_register(&kvm_dev);
2631 printk(KERN_ERR "kvm: misc device register failed\n");
2635 register_syscore_ops(&kvm_syscore_ops);
2637 kvm_preempt_ops.sched_in = kvm_sched_in;
2638 kvm_preempt_ops.sched_out = kvm_sched_out;
2645 kvm_async_pf_deinit();
2647 kmem_cache_destroy(kvm_vcpu_cache);
2649 unregister_reboot_notifier(&kvm_reboot_notifier);
2650 unregister_cpu_notifier(&kvm_cpu_notifier);
2653 kvm_arch_hardware_unsetup();
2655 free_cpumask_var(cpus_hardware_enabled);
2658 __free_page(fault_page);
2660 __free_page(hwpoison_page);
2661 __free_page(bad_page);
2667 EXPORT_SYMBOL_GPL(kvm_init);
2672 misc_deregister(&kvm_dev);
2673 kmem_cache_destroy(kvm_vcpu_cache);
2674 kvm_async_pf_deinit();
2675 unregister_syscore_ops(&kvm_syscore_ops);
2676 unregister_reboot_notifier(&kvm_reboot_notifier);
2677 unregister_cpu_notifier(&kvm_cpu_notifier);
2678 on_each_cpu(hardware_disable_nolock, NULL, 1);
2679 kvm_arch_hardware_unsetup();
2681 free_cpumask_var(cpus_hardware_enabled);
2682 __free_page(hwpoison_page);
2683 __free_page(bad_page);
2685 EXPORT_SYMBOL_GPL(kvm_exit);