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 struct page *fault_page;
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);
835 * If the new memory slot is created, we need to clear all
838 if (npages && old.base_gfn != mem->guest_phys_addr >> PAGE_SHIFT)
839 kvm_arch_flush_shadow(kvm);
841 kvm_free_physmem_slot(&old, &new);
847 kvm_free_physmem_slot(&new, &old);
852 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
854 int kvm_set_memory_region(struct kvm *kvm,
855 struct kvm_userspace_memory_region *mem,
860 mutex_lock(&kvm->slots_lock);
861 r = __kvm_set_memory_region(kvm, mem, user_alloc);
862 mutex_unlock(&kvm->slots_lock);
865 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
867 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
869 kvm_userspace_memory_region *mem,
872 if (mem->slot >= KVM_MEMORY_SLOTS)
874 return kvm_set_memory_region(kvm, mem, user_alloc);
877 int kvm_get_dirty_log(struct kvm *kvm,
878 struct kvm_dirty_log *log, int *is_dirty)
880 struct kvm_memory_slot *memslot;
883 unsigned long any = 0;
886 if (log->slot >= KVM_MEMORY_SLOTS)
889 memslot = &kvm->memslots->memslots[log->slot];
891 if (!memslot->dirty_bitmap)
894 n = kvm_dirty_bitmap_bytes(memslot);
896 for (i = 0; !any && i < n/sizeof(long); ++i)
897 any = memslot->dirty_bitmap[i];
900 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
911 void kvm_disable_largepages(void)
913 largepages_enabled = false;
915 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
917 int is_error_page(struct page *page)
919 return page == bad_page || page == hwpoison_page || page == fault_page;
921 EXPORT_SYMBOL_GPL(is_error_page);
923 int is_error_pfn(pfn_t pfn)
925 return pfn == bad_pfn || pfn == hwpoison_pfn || pfn == fault_pfn;
927 EXPORT_SYMBOL_GPL(is_error_pfn);
929 int is_hwpoison_pfn(pfn_t pfn)
931 return pfn == hwpoison_pfn;
933 EXPORT_SYMBOL_GPL(is_hwpoison_pfn);
935 int is_fault_pfn(pfn_t pfn)
937 return pfn == fault_pfn;
939 EXPORT_SYMBOL_GPL(is_fault_pfn);
941 int is_noslot_pfn(pfn_t pfn)
943 return pfn == bad_pfn;
945 EXPORT_SYMBOL_GPL(is_noslot_pfn);
947 int is_invalid_pfn(pfn_t pfn)
949 return pfn == hwpoison_pfn || pfn == fault_pfn;
951 EXPORT_SYMBOL_GPL(is_invalid_pfn);
953 static inline unsigned long bad_hva(void)
958 int kvm_is_error_hva(unsigned long addr)
960 return addr == bad_hva();
962 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
964 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm_memslots *slots,
969 for (i = 0; i < slots->nmemslots; ++i) {
970 struct kvm_memory_slot *memslot = &slots->memslots[i];
972 if (gfn >= memslot->base_gfn
973 && gfn < memslot->base_gfn + memslot->npages)
979 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
981 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
983 EXPORT_SYMBOL_GPL(gfn_to_memslot);
985 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
988 struct kvm_memslots *slots = kvm_memslots(kvm);
990 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
991 struct kvm_memory_slot *memslot = &slots->memslots[i];
993 if (memslot->flags & KVM_MEMSLOT_INVALID)
996 if (gfn >= memslot->base_gfn
997 && gfn < memslot->base_gfn + memslot->npages)
1002 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1004 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1006 struct vm_area_struct *vma;
1007 unsigned long addr, size;
1011 addr = gfn_to_hva(kvm, gfn);
1012 if (kvm_is_error_hva(addr))
1015 down_read(¤t->mm->mmap_sem);
1016 vma = find_vma(current->mm, addr);
1020 size = vma_kernel_pagesize(vma);
1023 up_read(¤t->mm->mmap_sem);
1028 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1031 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1035 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1037 return gfn_to_hva_memslot(slot, gfn);
1040 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1042 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1044 EXPORT_SYMBOL_GPL(gfn_to_hva);
1046 static pfn_t get_fault_pfn(void)
1048 get_page(fault_page);
1052 int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1053 unsigned long start, int write, struct page **page)
1055 int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1058 flags |= FOLL_WRITE;
1060 return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1063 static inline int check_user_page_hwpoison(unsigned long addr)
1065 int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1067 rc = __get_user_pages(current, current->mm, addr, 1,
1068 flags, NULL, NULL, NULL);
1069 return rc == -EHWPOISON;
1072 static pfn_t hva_to_pfn(struct kvm *kvm, unsigned long addr, bool atomic,
1073 bool *async, bool write_fault, bool *writable)
1075 struct page *page[1];
1079 /* we can do it either atomically or asynchronously, not both */
1080 BUG_ON(atomic && async);
1082 BUG_ON(!write_fault && !writable);
1087 if (atomic || async)
1088 npages = __get_user_pages_fast(addr, 1, 1, page);
1090 if (unlikely(npages != 1) && !atomic) {
1094 *writable = write_fault;
1097 down_read(¤t->mm->mmap_sem);
1098 npages = get_user_page_nowait(current, current->mm,
1099 addr, write_fault, page);
1100 up_read(¤t->mm->mmap_sem);
1102 npages = get_user_pages_fast(addr, 1, write_fault,
1105 /* map read fault as writable if possible */
1106 if (unlikely(!write_fault) && npages == 1) {
1107 struct page *wpage[1];
1109 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1119 if (unlikely(npages != 1)) {
1120 struct vm_area_struct *vma;
1123 return get_fault_pfn();
1125 down_read(¤t->mm->mmap_sem);
1126 if (npages == -EHWPOISON ||
1127 (!async && check_user_page_hwpoison(addr))) {
1128 up_read(¤t->mm->mmap_sem);
1129 get_page(hwpoison_page);
1130 return page_to_pfn(hwpoison_page);
1133 vma = find_vma_intersection(current->mm, addr, addr+1);
1136 pfn = get_fault_pfn();
1137 else if ((vma->vm_flags & VM_PFNMAP)) {
1138 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1140 BUG_ON(!kvm_is_mmio_pfn(pfn));
1142 if (async && (vma->vm_flags & VM_WRITE))
1144 pfn = get_fault_pfn();
1146 up_read(¤t->mm->mmap_sem);
1148 pfn = page_to_pfn(page[0]);
1153 pfn_t hva_to_pfn_atomic(struct kvm *kvm, unsigned long addr)
1155 return hva_to_pfn(kvm, addr, true, NULL, true, NULL);
1157 EXPORT_SYMBOL_GPL(hva_to_pfn_atomic);
1159 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1160 bool write_fault, bool *writable)
1167 addr = gfn_to_hva(kvm, gfn);
1168 if (kvm_is_error_hva(addr)) {
1170 return page_to_pfn(bad_page);
1173 return hva_to_pfn(kvm, addr, atomic, async, write_fault, writable);
1176 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1178 return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1180 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1182 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1183 bool write_fault, bool *writable)
1185 return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1187 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1189 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1191 return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1193 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1195 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1198 return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1200 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1202 pfn_t gfn_to_pfn_memslot(struct kvm *kvm,
1203 struct kvm_memory_slot *slot, gfn_t gfn)
1205 unsigned long addr = gfn_to_hva_memslot(slot, gfn);
1206 return hva_to_pfn(kvm, addr, false, NULL, true, NULL);
1209 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1215 addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1216 if (kvm_is_error_hva(addr))
1219 if (entry < nr_pages)
1222 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1224 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1226 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1230 pfn = gfn_to_pfn(kvm, gfn);
1231 if (!kvm_is_mmio_pfn(pfn))
1232 return pfn_to_page(pfn);
1234 WARN_ON(kvm_is_mmio_pfn(pfn));
1240 EXPORT_SYMBOL_GPL(gfn_to_page);
1242 void kvm_release_page_clean(struct page *page)
1244 kvm_release_pfn_clean(page_to_pfn(page));
1246 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1248 void kvm_release_pfn_clean(pfn_t pfn)
1250 if (!kvm_is_mmio_pfn(pfn))
1251 put_page(pfn_to_page(pfn));
1253 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1255 void kvm_release_page_dirty(struct page *page)
1257 kvm_release_pfn_dirty(page_to_pfn(page));
1259 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1261 void kvm_release_pfn_dirty(pfn_t pfn)
1263 kvm_set_pfn_dirty(pfn);
1264 kvm_release_pfn_clean(pfn);
1266 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1268 void kvm_set_page_dirty(struct page *page)
1270 kvm_set_pfn_dirty(page_to_pfn(page));
1272 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1274 void kvm_set_pfn_dirty(pfn_t pfn)
1276 if (!kvm_is_mmio_pfn(pfn)) {
1277 struct page *page = pfn_to_page(pfn);
1278 if (!PageReserved(page))
1282 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1284 void kvm_set_pfn_accessed(pfn_t pfn)
1286 if (!kvm_is_mmio_pfn(pfn))
1287 mark_page_accessed(pfn_to_page(pfn));
1289 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1291 void kvm_get_pfn(pfn_t pfn)
1293 if (!kvm_is_mmio_pfn(pfn))
1294 get_page(pfn_to_page(pfn));
1296 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1298 static int next_segment(unsigned long len, int offset)
1300 if (len > PAGE_SIZE - offset)
1301 return PAGE_SIZE - offset;
1306 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1312 addr = gfn_to_hva(kvm, gfn);
1313 if (kvm_is_error_hva(addr))
1315 r = __copy_from_user(data, (void __user *)addr + offset, len);
1320 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1322 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1324 gfn_t gfn = gpa >> PAGE_SHIFT;
1326 int offset = offset_in_page(gpa);
1329 while ((seg = next_segment(len, offset)) != 0) {
1330 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1340 EXPORT_SYMBOL_GPL(kvm_read_guest);
1342 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1347 gfn_t gfn = gpa >> PAGE_SHIFT;
1348 int offset = offset_in_page(gpa);
1350 addr = gfn_to_hva(kvm, gfn);
1351 if (kvm_is_error_hva(addr))
1353 pagefault_disable();
1354 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1360 EXPORT_SYMBOL(kvm_read_guest_atomic);
1362 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1363 int offset, int len)
1368 addr = gfn_to_hva(kvm, gfn);
1369 if (kvm_is_error_hva(addr))
1371 r = __copy_to_user((void __user *)addr + offset, data, len);
1374 mark_page_dirty(kvm, gfn);
1377 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1379 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1382 gfn_t gfn = gpa >> PAGE_SHIFT;
1384 int offset = offset_in_page(gpa);
1387 while ((seg = next_segment(len, offset)) != 0) {
1388 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1399 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1402 struct kvm_memslots *slots = kvm_memslots(kvm);
1403 int offset = offset_in_page(gpa);
1404 gfn_t gfn = gpa >> PAGE_SHIFT;
1407 ghc->generation = slots->generation;
1408 ghc->memslot = __gfn_to_memslot(slots, gfn);
1409 ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1410 if (!kvm_is_error_hva(ghc->hva))
1417 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1419 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1420 void *data, unsigned long len)
1422 struct kvm_memslots *slots = kvm_memslots(kvm);
1425 if (slots->generation != ghc->generation)
1426 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1428 if (kvm_is_error_hva(ghc->hva))
1431 r = __copy_to_user((void __user *)ghc->hva, data, len);
1434 mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1438 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1440 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1441 void *data, unsigned long len)
1443 struct kvm_memslots *slots = kvm_memslots(kvm);
1446 if (slots->generation != ghc->generation)
1447 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1449 if (kvm_is_error_hva(ghc->hva))
1452 r = __copy_from_user(data, (void __user *)ghc->hva, len);
1458 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1460 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1462 return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1465 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1467 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1469 gfn_t gfn = gpa >> PAGE_SHIFT;
1471 int offset = offset_in_page(gpa);
1474 while ((seg = next_segment(len, offset)) != 0) {
1475 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1484 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1486 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1489 if (memslot && memslot->dirty_bitmap) {
1490 unsigned long rel_gfn = gfn - memslot->base_gfn;
1492 __set_bit_le(rel_gfn, memslot->dirty_bitmap);
1496 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1498 struct kvm_memory_slot *memslot;
1500 memslot = gfn_to_memslot(kvm, gfn);
1501 mark_page_dirty_in_slot(kvm, memslot, gfn);
1505 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1507 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1512 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1514 if (kvm_arch_vcpu_runnable(vcpu)) {
1515 kvm_make_request(KVM_REQ_UNHALT, vcpu);
1518 if (kvm_cpu_has_pending_timer(vcpu))
1520 if (signal_pending(current))
1526 finish_wait(&vcpu->wq, &wait);
1529 void kvm_resched(struct kvm_vcpu *vcpu)
1531 if (!need_resched())
1535 EXPORT_SYMBOL_GPL(kvm_resched);
1537 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1539 struct kvm *kvm = me->kvm;
1540 struct kvm_vcpu *vcpu;
1541 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1547 * We boost the priority of a VCPU that is runnable but not
1548 * currently running, because it got preempted by something
1549 * else and called schedule in __vcpu_run. Hopefully that
1550 * VCPU is holding the lock that we need and will release it.
1551 * We approximate round-robin by starting at the last boosted VCPU.
1553 for (pass = 0; pass < 2 && !yielded; pass++) {
1554 kvm_for_each_vcpu(i, vcpu, kvm) {
1555 struct task_struct *task = NULL;
1557 if (!pass && i < last_boosted_vcpu) {
1558 i = last_boosted_vcpu;
1560 } else if (pass && i > last_boosted_vcpu)
1564 if (waitqueue_active(&vcpu->wq))
1567 pid = rcu_dereference(vcpu->pid);
1569 task = get_pid_task(vcpu->pid, PIDTYPE_PID);
1573 if (task->flags & PF_VCPU) {
1574 put_task_struct(task);
1577 if (yield_to(task, 1)) {
1578 put_task_struct(task);
1579 kvm->last_boosted_vcpu = i;
1583 put_task_struct(task);
1587 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1589 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1591 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1594 if (vmf->pgoff == 0)
1595 page = virt_to_page(vcpu->run);
1597 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1598 page = virt_to_page(vcpu->arch.pio_data);
1600 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1601 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1602 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1605 return VM_FAULT_SIGBUS;
1611 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1612 .fault = kvm_vcpu_fault,
1615 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1617 vma->vm_ops = &kvm_vcpu_vm_ops;
1621 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1623 struct kvm_vcpu *vcpu = filp->private_data;
1625 kvm_put_kvm(vcpu->kvm);
1629 static struct file_operations kvm_vcpu_fops = {
1630 .release = kvm_vcpu_release,
1631 .unlocked_ioctl = kvm_vcpu_ioctl,
1632 #ifdef CONFIG_COMPAT
1633 .compat_ioctl = kvm_vcpu_compat_ioctl,
1635 .mmap = kvm_vcpu_mmap,
1636 .llseek = noop_llseek,
1640 * Allocates an inode for the vcpu.
1642 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1644 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1648 * Creates some virtual cpus. Good luck creating more than one.
1650 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1653 struct kvm_vcpu *vcpu, *v;
1655 vcpu = kvm_arch_vcpu_create(kvm, id);
1657 return PTR_ERR(vcpu);
1659 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1661 r = kvm_arch_vcpu_setup(vcpu);
1665 mutex_lock(&kvm->lock);
1666 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1668 goto unlock_vcpu_destroy;
1671 kvm_for_each_vcpu(r, v, kvm)
1672 if (v->vcpu_id == id) {
1674 goto unlock_vcpu_destroy;
1677 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1679 /* Now it's all set up, let userspace reach it */
1681 r = create_vcpu_fd(vcpu);
1684 goto unlock_vcpu_destroy;
1687 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1689 atomic_inc(&kvm->online_vcpus);
1691 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1692 if (kvm->bsp_vcpu_id == id)
1693 kvm->bsp_vcpu = vcpu;
1695 mutex_unlock(&kvm->lock);
1698 unlock_vcpu_destroy:
1699 mutex_unlock(&kvm->lock);
1701 kvm_arch_vcpu_destroy(vcpu);
1705 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1708 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1709 vcpu->sigset_active = 1;
1710 vcpu->sigset = *sigset;
1712 vcpu->sigset_active = 0;
1716 static long kvm_vcpu_ioctl(struct file *filp,
1717 unsigned int ioctl, unsigned long arg)
1719 struct kvm_vcpu *vcpu = filp->private_data;
1720 void __user *argp = (void __user *)arg;
1722 struct kvm_fpu *fpu = NULL;
1723 struct kvm_sregs *kvm_sregs = NULL;
1725 if (vcpu->kvm->mm != current->mm)
1728 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1730 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1731 * so vcpu_load() would break it.
1733 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1734 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1744 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1745 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1747 case KVM_GET_REGS: {
1748 struct kvm_regs *kvm_regs;
1751 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1754 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1758 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1765 case KVM_SET_REGS: {
1766 struct kvm_regs *kvm_regs;
1769 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1773 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1775 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1783 case KVM_GET_SREGS: {
1784 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1788 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1792 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1797 case KVM_SET_SREGS: {
1798 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1803 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1805 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1811 case KVM_GET_MP_STATE: {
1812 struct kvm_mp_state mp_state;
1814 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1818 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1823 case KVM_SET_MP_STATE: {
1824 struct kvm_mp_state mp_state;
1827 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1829 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1835 case KVM_TRANSLATE: {
1836 struct kvm_translation tr;
1839 if (copy_from_user(&tr, argp, sizeof tr))
1841 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1845 if (copy_to_user(argp, &tr, sizeof tr))
1850 case KVM_SET_GUEST_DEBUG: {
1851 struct kvm_guest_debug dbg;
1854 if (copy_from_user(&dbg, argp, sizeof dbg))
1856 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1862 case KVM_SET_SIGNAL_MASK: {
1863 struct kvm_signal_mask __user *sigmask_arg = argp;
1864 struct kvm_signal_mask kvm_sigmask;
1865 sigset_t sigset, *p;
1870 if (copy_from_user(&kvm_sigmask, argp,
1871 sizeof kvm_sigmask))
1874 if (kvm_sigmask.len != sizeof sigset)
1877 if (copy_from_user(&sigset, sigmask_arg->sigset,
1882 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
1886 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1890 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1894 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1900 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1905 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1907 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1914 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1923 #ifdef CONFIG_COMPAT
1924 static long kvm_vcpu_compat_ioctl(struct file *filp,
1925 unsigned int ioctl, unsigned long arg)
1927 struct kvm_vcpu *vcpu = filp->private_data;
1928 void __user *argp = compat_ptr(arg);
1931 if (vcpu->kvm->mm != current->mm)
1935 case KVM_SET_SIGNAL_MASK: {
1936 struct kvm_signal_mask __user *sigmask_arg = argp;
1937 struct kvm_signal_mask kvm_sigmask;
1938 compat_sigset_t csigset;
1943 if (copy_from_user(&kvm_sigmask, argp,
1944 sizeof kvm_sigmask))
1947 if (kvm_sigmask.len != sizeof csigset)
1950 if (copy_from_user(&csigset, sigmask_arg->sigset,
1954 sigset_from_compat(&sigset, &csigset);
1955 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1959 r = kvm_vcpu_ioctl(filp, ioctl, arg);
1967 static long kvm_vm_ioctl(struct file *filp,
1968 unsigned int ioctl, unsigned long arg)
1970 struct kvm *kvm = filp->private_data;
1971 void __user *argp = (void __user *)arg;
1974 if (kvm->mm != current->mm)
1977 case KVM_CREATE_VCPU:
1978 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1982 case KVM_SET_USER_MEMORY_REGION: {
1983 struct kvm_userspace_memory_region kvm_userspace_mem;
1986 if (copy_from_user(&kvm_userspace_mem, argp,
1987 sizeof kvm_userspace_mem))
1990 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1995 case KVM_GET_DIRTY_LOG: {
1996 struct kvm_dirty_log log;
1999 if (copy_from_user(&log, argp, sizeof log))
2001 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2006 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2007 case KVM_REGISTER_COALESCED_MMIO: {
2008 struct kvm_coalesced_mmio_zone zone;
2010 if (copy_from_user(&zone, argp, sizeof zone))
2012 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2018 case KVM_UNREGISTER_COALESCED_MMIO: {
2019 struct kvm_coalesced_mmio_zone zone;
2021 if (copy_from_user(&zone, argp, sizeof zone))
2023 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2031 struct kvm_irqfd data;
2034 if (copy_from_user(&data, argp, sizeof data))
2036 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
2039 case KVM_IOEVENTFD: {
2040 struct kvm_ioeventfd data;
2043 if (copy_from_user(&data, argp, sizeof data))
2045 r = kvm_ioeventfd(kvm, &data);
2048 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2049 case KVM_SET_BOOT_CPU_ID:
2051 mutex_lock(&kvm->lock);
2052 if (atomic_read(&kvm->online_vcpus) != 0)
2055 kvm->bsp_vcpu_id = arg;
2056 mutex_unlock(&kvm->lock);
2060 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2062 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2068 #ifdef CONFIG_COMPAT
2069 struct compat_kvm_dirty_log {
2073 compat_uptr_t dirty_bitmap; /* one bit per page */
2078 static long kvm_vm_compat_ioctl(struct file *filp,
2079 unsigned int ioctl, unsigned long arg)
2081 struct kvm *kvm = filp->private_data;
2084 if (kvm->mm != current->mm)
2087 case KVM_GET_DIRTY_LOG: {
2088 struct compat_kvm_dirty_log compat_log;
2089 struct kvm_dirty_log log;
2092 if (copy_from_user(&compat_log, (void __user *)arg,
2093 sizeof(compat_log)))
2095 log.slot = compat_log.slot;
2096 log.padding1 = compat_log.padding1;
2097 log.padding2 = compat_log.padding2;
2098 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2100 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2106 r = kvm_vm_ioctl(filp, ioctl, arg);
2114 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2116 struct page *page[1];
2119 gfn_t gfn = vmf->pgoff;
2120 struct kvm *kvm = vma->vm_file->private_data;
2122 addr = gfn_to_hva(kvm, gfn);
2123 if (kvm_is_error_hva(addr))
2124 return VM_FAULT_SIGBUS;
2126 npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2128 if (unlikely(npages != 1))
2129 return VM_FAULT_SIGBUS;
2131 vmf->page = page[0];
2135 static const struct vm_operations_struct kvm_vm_vm_ops = {
2136 .fault = kvm_vm_fault,
2139 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2141 vma->vm_ops = &kvm_vm_vm_ops;
2145 static struct file_operations kvm_vm_fops = {
2146 .release = kvm_vm_release,
2147 .unlocked_ioctl = kvm_vm_ioctl,
2148 #ifdef CONFIG_COMPAT
2149 .compat_ioctl = kvm_vm_compat_ioctl,
2151 .mmap = kvm_vm_mmap,
2152 .llseek = noop_llseek,
2155 static int kvm_dev_ioctl_create_vm(void)
2160 kvm = kvm_create_vm();
2162 return PTR_ERR(kvm);
2163 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2164 r = kvm_coalesced_mmio_init(kvm);
2170 r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2177 static long kvm_dev_ioctl_check_extension_generic(long arg)
2180 case KVM_CAP_USER_MEMORY:
2181 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2182 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2183 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2184 case KVM_CAP_SET_BOOT_CPU_ID:
2186 case KVM_CAP_INTERNAL_ERROR_DATA:
2188 #ifdef CONFIG_HAVE_KVM_IRQCHIP
2189 case KVM_CAP_IRQ_ROUTING:
2190 return KVM_MAX_IRQ_ROUTES;
2195 return kvm_dev_ioctl_check_extension(arg);
2198 static long kvm_dev_ioctl(struct file *filp,
2199 unsigned int ioctl, unsigned long arg)
2204 case KVM_GET_API_VERSION:
2208 r = KVM_API_VERSION;
2214 r = kvm_dev_ioctl_create_vm();
2216 case KVM_CHECK_EXTENSION:
2217 r = kvm_dev_ioctl_check_extension_generic(arg);
2219 case KVM_GET_VCPU_MMAP_SIZE:
2223 r = PAGE_SIZE; /* struct kvm_run */
2225 r += PAGE_SIZE; /* pio data page */
2227 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2228 r += PAGE_SIZE; /* coalesced mmio ring page */
2231 case KVM_TRACE_ENABLE:
2232 case KVM_TRACE_PAUSE:
2233 case KVM_TRACE_DISABLE:
2237 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2243 static struct file_operations kvm_chardev_ops = {
2244 .unlocked_ioctl = kvm_dev_ioctl,
2245 .compat_ioctl = kvm_dev_ioctl,
2246 .llseek = noop_llseek,
2249 static struct miscdevice kvm_dev = {
2255 static void hardware_enable_nolock(void *junk)
2257 int cpu = raw_smp_processor_id();
2260 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2263 cpumask_set_cpu(cpu, cpus_hardware_enabled);
2265 r = kvm_arch_hardware_enable(NULL);
2268 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2269 atomic_inc(&hardware_enable_failed);
2270 printk(KERN_INFO "kvm: enabling virtualization on "
2271 "CPU%d failed\n", cpu);
2275 static void hardware_enable(void *junk)
2277 raw_spin_lock(&kvm_lock);
2278 hardware_enable_nolock(junk);
2279 raw_spin_unlock(&kvm_lock);
2282 static void hardware_disable_nolock(void *junk)
2284 int cpu = raw_smp_processor_id();
2286 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2288 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2289 kvm_arch_hardware_disable(NULL);
2292 static void hardware_disable(void *junk)
2294 raw_spin_lock(&kvm_lock);
2295 hardware_disable_nolock(junk);
2296 raw_spin_unlock(&kvm_lock);
2299 static void hardware_disable_all_nolock(void)
2301 BUG_ON(!kvm_usage_count);
2304 if (!kvm_usage_count)
2305 on_each_cpu(hardware_disable_nolock, NULL, 1);
2308 static void hardware_disable_all(void)
2310 raw_spin_lock(&kvm_lock);
2311 hardware_disable_all_nolock();
2312 raw_spin_unlock(&kvm_lock);
2315 static int hardware_enable_all(void)
2319 raw_spin_lock(&kvm_lock);
2322 if (kvm_usage_count == 1) {
2323 atomic_set(&hardware_enable_failed, 0);
2324 on_each_cpu(hardware_enable_nolock, NULL, 1);
2326 if (atomic_read(&hardware_enable_failed)) {
2327 hardware_disable_all_nolock();
2332 raw_spin_unlock(&kvm_lock);
2337 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2342 if (!kvm_usage_count)
2345 val &= ~CPU_TASKS_FROZEN;
2348 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2350 hardware_disable(NULL);
2353 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2355 hardware_enable(NULL);
2362 asmlinkage void kvm_spurious_fault(void)
2364 /* Fault while not rebooting. We want the trace. */
2367 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
2369 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2373 * Some (well, at least mine) BIOSes hang on reboot if
2376 * And Intel TXT required VMX off for all cpu when system shutdown.
2378 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2379 kvm_rebooting = true;
2380 on_each_cpu(hardware_disable_nolock, NULL, 1);
2384 static struct notifier_block kvm_reboot_notifier = {
2385 .notifier_call = kvm_reboot,
2389 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2393 for (i = 0; i < bus->dev_count; i++) {
2394 struct kvm_io_device *pos = bus->devs[i];
2396 kvm_iodevice_destructor(pos);
2401 /* kvm_io_bus_write - called under kvm->slots_lock */
2402 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2403 int len, const void *val)
2406 struct kvm_io_bus *bus;
2408 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2409 for (i = 0; i < bus->dev_count; i++)
2410 if (!kvm_iodevice_write(bus->devs[i], addr, len, val))
2415 /* kvm_io_bus_read - called under kvm->slots_lock */
2416 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2420 struct kvm_io_bus *bus;
2422 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2423 for (i = 0; i < bus->dev_count; i++)
2424 if (!kvm_iodevice_read(bus->devs[i], addr, len, val))
2429 /* Caller must hold slots_lock. */
2430 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2431 struct kvm_io_device *dev)
2433 struct kvm_io_bus *new_bus, *bus;
2435 bus = kvm->buses[bus_idx];
2436 if (bus->dev_count > NR_IOBUS_DEVS-1)
2439 new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2442 memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2443 new_bus->devs[new_bus->dev_count++] = dev;
2444 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2445 synchronize_srcu_expedited(&kvm->srcu);
2451 /* Caller must hold slots_lock. */
2452 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2453 struct kvm_io_device *dev)
2456 struct kvm_io_bus *new_bus, *bus;
2458 new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2462 bus = kvm->buses[bus_idx];
2463 memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2466 for (i = 0; i < new_bus->dev_count; i++)
2467 if (new_bus->devs[i] == dev) {
2469 new_bus->devs[i] = new_bus->devs[--new_bus->dev_count];
2478 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2479 synchronize_srcu_expedited(&kvm->srcu);
2484 static struct notifier_block kvm_cpu_notifier = {
2485 .notifier_call = kvm_cpu_hotplug,
2488 static int vm_stat_get(void *_offset, u64 *val)
2490 unsigned offset = (long)_offset;
2494 raw_spin_lock(&kvm_lock);
2495 list_for_each_entry(kvm, &vm_list, vm_list)
2496 *val += *(u32 *)((void *)kvm + offset);
2497 raw_spin_unlock(&kvm_lock);
2501 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2503 static int vcpu_stat_get(void *_offset, u64 *val)
2505 unsigned offset = (long)_offset;
2507 struct kvm_vcpu *vcpu;
2511 raw_spin_lock(&kvm_lock);
2512 list_for_each_entry(kvm, &vm_list, vm_list)
2513 kvm_for_each_vcpu(i, vcpu, kvm)
2514 *val += *(u32 *)((void *)vcpu + offset);
2516 raw_spin_unlock(&kvm_lock);
2520 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2522 static const struct file_operations *stat_fops[] = {
2523 [KVM_STAT_VCPU] = &vcpu_stat_fops,
2524 [KVM_STAT_VM] = &vm_stat_fops,
2527 static void kvm_init_debug(void)
2529 struct kvm_stats_debugfs_item *p;
2531 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2532 for (p = debugfs_entries; p->name; ++p)
2533 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2534 (void *)(long)p->offset,
2535 stat_fops[p->kind]);
2538 static void kvm_exit_debug(void)
2540 struct kvm_stats_debugfs_item *p;
2542 for (p = debugfs_entries; p->name; ++p)
2543 debugfs_remove(p->dentry);
2544 debugfs_remove(kvm_debugfs_dir);
2547 static int kvm_suspend(void)
2549 if (kvm_usage_count)
2550 hardware_disable_nolock(NULL);
2554 static void kvm_resume(void)
2556 if (kvm_usage_count) {
2557 WARN_ON(raw_spin_is_locked(&kvm_lock));
2558 hardware_enable_nolock(NULL);
2562 static struct syscore_ops kvm_syscore_ops = {
2563 .suspend = kvm_suspend,
2564 .resume = kvm_resume,
2567 struct page *bad_page;
2571 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2573 return container_of(pn, struct kvm_vcpu, preempt_notifier);
2576 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2578 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2580 kvm_arch_vcpu_load(vcpu, cpu);
2583 static void kvm_sched_out(struct preempt_notifier *pn,
2584 struct task_struct *next)
2586 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2588 kvm_arch_vcpu_put(vcpu);
2591 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2592 struct module *module)
2597 r = kvm_arch_init(opaque);
2601 bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2603 if (bad_page == NULL) {
2608 bad_pfn = page_to_pfn(bad_page);
2610 hwpoison_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2612 if (hwpoison_page == NULL) {
2617 hwpoison_pfn = page_to_pfn(hwpoison_page);
2619 fault_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2621 if (fault_page == NULL) {
2626 fault_pfn = page_to_pfn(fault_page);
2628 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2633 r = kvm_arch_hardware_setup();
2637 for_each_online_cpu(cpu) {
2638 smp_call_function_single(cpu,
2639 kvm_arch_check_processor_compat,
2645 r = register_cpu_notifier(&kvm_cpu_notifier);
2648 register_reboot_notifier(&kvm_reboot_notifier);
2650 /* A kmem cache lets us meet the alignment requirements of fx_save. */
2652 vcpu_align = __alignof__(struct kvm_vcpu);
2653 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2655 if (!kvm_vcpu_cache) {
2660 r = kvm_async_pf_init();
2664 kvm_chardev_ops.owner = module;
2665 kvm_vm_fops.owner = module;
2666 kvm_vcpu_fops.owner = module;
2668 r = misc_register(&kvm_dev);
2670 printk(KERN_ERR "kvm: misc device register failed\n");
2674 register_syscore_ops(&kvm_syscore_ops);
2676 kvm_preempt_ops.sched_in = kvm_sched_in;
2677 kvm_preempt_ops.sched_out = kvm_sched_out;
2684 kvm_async_pf_deinit();
2686 kmem_cache_destroy(kvm_vcpu_cache);
2688 unregister_reboot_notifier(&kvm_reboot_notifier);
2689 unregister_cpu_notifier(&kvm_cpu_notifier);
2692 kvm_arch_hardware_unsetup();
2694 free_cpumask_var(cpus_hardware_enabled);
2697 __free_page(fault_page);
2699 __free_page(hwpoison_page);
2700 __free_page(bad_page);
2706 EXPORT_SYMBOL_GPL(kvm_init);
2711 misc_deregister(&kvm_dev);
2712 kmem_cache_destroy(kvm_vcpu_cache);
2713 kvm_async_pf_deinit();
2714 unregister_syscore_ops(&kvm_syscore_ops);
2715 unregister_reboot_notifier(&kvm_reboot_notifier);
2716 unregister_cpu_notifier(&kvm_cpu_notifier);
2717 on_each_cpu(hardware_disable_nolock, NULL, 1);
2718 kvm_arch_hardware_unsetup();
2720 free_cpumask_var(cpus_hardware_enabled);
2721 __free_page(hwpoison_page);
2722 __free_page(bad_page);
2724 EXPORT_SYMBOL_GPL(kvm_exit);