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,
87 static int hardware_enable_all(void);
88 static void hardware_disable_all(void);
90 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
93 EXPORT_SYMBOL_GPL(kvm_rebooting);
95 static bool largepages_enabled = true;
97 static struct page *hwpoison_page;
98 static pfn_t hwpoison_pfn;
100 static struct page *fault_page;
101 static pfn_t fault_pfn;
103 inline int kvm_is_mmio_pfn(pfn_t pfn)
105 if (pfn_valid(pfn)) {
107 struct page *tail = pfn_to_page(pfn);
108 struct page *head = compound_trans_head(tail);
109 reserved = PageReserved(head);
112 * "head" is not a dangling pointer
113 * (compound_trans_head takes care of that)
114 * but the hugepage may have been splitted
115 * from under us (and we may not hold a
116 * reference count on the head page so it can
117 * be reused before we run PageReferenced), so
118 * we've to check PageTail before returning
125 return PageReserved(tail);
132 * Switches to specified vcpu, until a matching vcpu_put()
134 void vcpu_load(struct kvm_vcpu *vcpu)
138 mutex_lock(&vcpu->mutex);
139 if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
140 /* The thread running this VCPU changed. */
141 struct pid *oldpid = vcpu->pid;
142 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
143 rcu_assign_pointer(vcpu->pid, newpid);
148 preempt_notifier_register(&vcpu->preempt_notifier);
149 kvm_arch_vcpu_load(vcpu, cpu);
153 void vcpu_put(struct kvm_vcpu *vcpu)
156 kvm_arch_vcpu_put(vcpu);
157 preempt_notifier_unregister(&vcpu->preempt_notifier);
159 mutex_unlock(&vcpu->mutex);
162 static void ack_flush(void *_completed)
166 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
171 struct kvm_vcpu *vcpu;
173 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
176 kvm_for_each_vcpu(i, vcpu, kvm) {
177 kvm_make_request(req, vcpu);
180 /* Set ->requests bit before we read ->mode */
183 if (cpus != NULL && cpu != -1 && cpu != me &&
184 kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
185 cpumask_set_cpu(cpu, cpus);
187 if (unlikely(cpus == NULL))
188 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
189 else if (!cpumask_empty(cpus))
190 smp_call_function_many(cpus, ack_flush, NULL, 1);
194 free_cpumask_var(cpus);
198 void kvm_flush_remote_tlbs(struct kvm *kvm)
200 int dirty_count = kvm->tlbs_dirty;
203 if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
204 ++kvm->stat.remote_tlb_flush;
205 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
208 void kvm_reload_remote_mmus(struct kvm *kvm)
210 make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
213 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
218 mutex_init(&vcpu->mutex);
223 init_waitqueue_head(&vcpu->wq);
224 kvm_async_pf_vcpu_init(vcpu);
226 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
231 vcpu->run = page_address(page);
233 r = kvm_arch_vcpu_init(vcpu);
239 free_page((unsigned long)vcpu->run);
243 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
245 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
248 kvm_arch_vcpu_uninit(vcpu);
249 free_page((unsigned long)vcpu->run);
251 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
253 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
254 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
256 return container_of(mn, struct kvm, mmu_notifier);
259 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
260 struct mm_struct *mm,
261 unsigned long address)
263 struct kvm *kvm = mmu_notifier_to_kvm(mn);
264 int need_tlb_flush, idx;
267 * When ->invalidate_page runs, the linux pte has been zapped
268 * already but the page is still allocated until
269 * ->invalidate_page returns. So if we increase the sequence
270 * here the kvm page fault will notice if the spte can't be
271 * established because the page is going to be freed. If
272 * instead the kvm page fault establishes the spte before
273 * ->invalidate_page runs, kvm_unmap_hva will release it
276 * The sequence increase only need to be seen at spin_unlock
277 * time, and not at spin_lock time.
279 * Increasing the sequence after the spin_unlock would be
280 * unsafe because the kvm page fault could then establish the
281 * pte after kvm_unmap_hva returned, without noticing the page
282 * is going to be freed.
284 idx = srcu_read_lock(&kvm->srcu);
285 spin_lock(&kvm->mmu_lock);
286 kvm->mmu_notifier_seq++;
287 need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
288 spin_unlock(&kvm->mmu_lock);
289 srcu_read_unlock(&kvm->srcu, idx);
291 /* we've to flush the tlb before the pages can be freed */
293 kvm_flush_remote_tlbs(kvm);
297 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
298 struct mm_struct *mm,
299 unsigned long address,
302 struct kvm *kvm = mmu_notifier_to_kvm(mn);
305 idx = srcu_read_lock(&kvm->srcu);
306 spin_lock(&kvm->mmu_lock);
307 kvm->mmu_notifier_seq++;
308 kvm_set_spte_hva(kvm, address, pte);
309 spin_unlock(&kvm->mmu_lock);
310 srcu_read_unlock(&kvm->srcu, idx);
313 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
314 struct mm_struct *mm,
318 struct kvm *kvm = mmu_notifier_to_kvm(mn);
319 int need_tlb_flush = 0, idx;
321 idx = srcu_read_lock(&kvm->srcu);
322 spin_lock(&kvm->mmu_lock);
324 * The count increase must become visible at unlock time as no
325 * spte can be established without taking the mmu_lock and
326 * count is also read inside the mmu_lock critical section.
328 kvm->mmu_notifier_count++;
329 for (; start < end; start += PAGE_SIZE)
330 need_tlb_flush |= kvm_unmap_hva(kvm, start);
331 need_tlb_flush |= kvm->tlbs_dirty;
332 spin_unlock(&kvm->mmu_lock);
333 srcu_read_unlock(&kvm->srcu, idx);
335 /* we've to flush the tlb before the pages can be freed */
337 kvm_flush_remote_tlbs(kvm);
340 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
341 struct mm_struct *mm,
345 struct kvm *kvm = mmu_notifier_to_kvm(mn);
347 spin_lock(&kvm->mmu_lock);
349 * This sequence increase will notify the kvm page fault that
350 * the page that is going to be mapped in the spte could have
353 kvm->mmu_notifier_seq++;
355 * The above sequence increase must be visible before the
356 * below count decrease but both values are read by the kvm
357 * page fault under mmu_lock spinlock so we don't need to add
358 * a smb_wmb() here in between the two.
360 kvm->mmu_notifier_count--;
361 spin_unlock(&kvm->mmu_lock);
363 BUG_ON(kvm->mmu_notifier_count < 0);
366 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
367 struct mm_struct *mm,
368 unsigned long address)
370 struct kvm *kvm = mmu_notifier_to_kvm(mn);
373 idx = srcu_read_lock(&kvm->srcu);
374 spin_lock(&kvm->mmu_lock);
375 young = kvm_age_hva(kvm, address);
376 spin_unlock(&kvm->mmu_lock);
377 srcu_read_unlock(&kvm->srcu, idx);
380 kvm_flush_remote_tlbs(kvm);
385 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
386 struct mm_struct *mm,
387 unsigned long address)
389 struct kvm *kvm = mmu_notifier_to_kvm(mn);
392 idx = srcu_read_lock(&kvm->srcu);
393 spin_lock(&kvm->mmu_lock);
394 young = kvm_test_age_hva(kvm, address);
395 spin_unlock(&kvm->mmu_lock);
396 srcu_read_unlock(&kvm->srcu, idx);
401 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
402 struct mm_struct *mm)
404 struct kvm *kvm = mmu_notifier_to_kvm(mn);
407 idx = srcu_read_lock(&kvm->srcu);
408 kvm_arch_flush_shadow(kvm);
409 srcu_read_unlock(&kvm->srcu, idx);
412 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
413 .invalidate_page = kvm_mmu_notifier_invalidate_page,
414 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
415 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
416 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
417 .test_young = kvm_mmu_notifier_test_young,
418 .change_pte = kvm_mmu_notifier_change_pte,
419 .release = kvm_mmu_notifier_release,
422 static int kvm_init_mmu_notifier(struct kvm *kvm)
424 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
425 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
428 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
430 static int kvm_init_mmu_notifier(struct kvm *kvm)
435 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
437 static struct kvm *kvm_create_vm(void)
440 struct kvm *kvm = kvm_arch_alloc_vm();
443 return ERR_PTR(-ENOMEM);
445 r = kvm_arch_init_vm(kvm);
447 goto out_err_nodisable;
449 r = hardware_enable_all();
451 goto out_err_nodisable;
453 #ifdef CONFIG_HAVE_KVM_IRQCHIP
454 INIT_HLIST_HEAD(&kvm->mask_notifier_list);
455 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
459 kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
462 if (init_srcu_struct(&kvm->srcu))
464 for (i = 0; i < KVM_NR_BUSES; i++) {
465 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
471 spin_lock_init(&kvm->mmu_lock);
472 kvm->mm = current->mm;
473 atomic_inc(&kvm->mm->mm_count);
474 kvm_eventfd_init(kvm);
475 mutex_init(&kvm->lock);
476 mutex_init(&kvm->irq_lock);
477 mutex_init(&kvm->slots_lock);
478 atomic_set(&kvm->users_count, 1);
480 r = kvm_init_mmu_notifier(kvm);
484 raw_spin_lock(&kvm_lock);
485 list_add(&kvm->vm_list, &vm_list);
486 raw_spin_unlock(&kvm_lock);
491 cleanup_srcu_struct(&kvm->srcu);
493 hardware_disable_all();
495 for (i = 0; i < KVM_NR_BUSES; i++)
496 kfree(kvm->buses[i]);
497 kfree(kvm->memslots);
498 kvm_arch_free_vm(kvm);
502 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
504 if (!memslot->dirty_bitmap)
507 if (2 * kvm_dirty_bitmap_bytes(memslot) > PAGE_SIZE)
508 vfree(memslot->dirty_bitmap_head);
510 kfree(memslot->dirty_bitmap_head);
512 memslot->dirty_bitmap = NULL;
513 memslot->dirty_bitmap_head = NULL;
517 * Free any memory in @free but not in @dont.
519 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
520 struct kvm_memory_slot *dont)
524 if (!dont || free->rmap != dont->rmap)
527 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
528 kvm_destroy_dirty_bitmap(free);
531 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
532 if (!dont || free->lpage_info[i] != dont->lpage_info[i]) {
533 vfree(free->lpage_info[i]);
534 free->lpage_info[i] = NULL;
542 void kvm_free_physmem(struct kvm *kvm)
545 struct kvm_memslots *slots = kvm->memslots;
547 for (i = 0; i < slots->nmemslots; ++i)
548 kvm_free_physmem_slot(&slots->memslots[i], NULL);
550 kfree(kvm->memslots);
553 static void kvm_destroy_vm(struct kvm *kvm)
556 struct mm_struct *mm = kvm->mm;
558 kvm_arch_sync_events(kvm);
559 raw_spin_lock(&kvm_lock);
560 list_del(&kvm->vm_list);
561 raw_spin_unlock(&kvm_lock);
562 kvm_free_irq_routing(kvm);
563 for (i = 0; i < KVM_NR_BUSES; i++)
564 kvm_io_bus_destroy(kvm->buses[i]);
565 kvm_coalesced_mmio_free(kvm);
566 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
567 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
569 kvm_arch_flush_shadow(kvm);
571 kvm_arch_destroy_vm(kvm);
572 kvm_free_physmem(kvm);
573 cleanup_srcu_struct(&kvm->srcu);
574 kvm_arch_free_vm(kvm);
575 hardware_disable_all();
579 void kvm_get_kvm(struct kvm *kvm)
581 atomic_inc(&kvm->users_count);
583 EXPORT_SYMBOL_GPL(kvm_get_kvm);
585 void kvm_put_kvm(struct kvm *kvm)
587 if (atomic_dec_and_test(&kvm->users_count))
590 EXPORT_SYMBOL_GPL(kvm_put_kvm);
593 static int kvm_vm_release(struct inode *inode, struct file *filp)
595 struct kvm *kvm = filp->private_data;
597 kvm_irqfd_release(kvm);
605 * Allocation size is twice as large as the actual dirty bitmap size.
606 * This makes it possible to do double buffering: see x86's
607 * kvm_vm_ioctl_get_dirty_log().
609 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
611 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
613 if (dirty_bytes > PAGE_SIZE)
614 memslot->dirty_bitmap = vzalloc(dirty_bytes);
616 memslot->dirty_bitmap = kzalloc(dirty_bytes, GFP_KERNEL);
618 if (!memslot->dirty_bitmap)
621 memslot->dirty_bitmap_head = memslot->dirty_bitmap;
624 #endif /* !CONFIG_S390 */
627 * Allocate some memory and give it an address in the guest physical address
630 * Discontiguous memory is allowed, mostly for framebuffers.
632 * Must be called holding mmap_sem for write.
634 int __kvm_set_memory_region(struct kvm *kvm,
635 struct kvm_userspace_memory_region *mem,
640 unsigned long npages;
642 struct kvm_memory_slot *memslot;
643 struct kvm_memory_slot old, new;
644 struct kvm_memslots *slots, *old_memslots;
647 /* General sanity checks */
648 if (mem->memory_size & (PAGE_SIZE - 1))
650 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
652 /* We can read the guest memory with __xxx_user() later on. */
654 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
655 !access_ok(VERIFY_WRITE,
656 (void __user *)(unsigned long)mem->userspace_addr,
659 if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
661 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
664 memslot = &kvm->memslots->memslots[mem->slot];
665 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
666 npages = mem->memory_size >> PAGE_SHIFT;
669 if (npages > KVM_MEM_MAX_NR_PAGES)
673 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
675 new = old = *memslot;
678 new.base_gfn = base_gfn;
680 new.flags = mem->flags;
682 /* Disallow changing a memory slot's size. */
684 if (npages && old.npages && npages != old.npages)
687 /* Check for overlaps */
689 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
690 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
692 if (s == memslot || !s->npages)
694 if (!((base_gfn + npages <= s->base_gfn) ||
695 (base_gfn >= s->base_gfn + s->npages)))
699 /* Free page dirty bitmap if unneeded */
700 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
701 new.dirty_bitmap = NULL;
705 /* Allocate if a slot is being created */
707 if (npages && !new.rmap) {
708 new.rmap = vzalloc(npages * sizeof(*new.rmap));
713 new.user_alloc = user_alloc;
714 new.userspace_addr = mem->userspace_addr;
719 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
725 /* Avoid unused variable warning if no large pages */
728 if (new.lpage_info[i])
731 lpages = 1 + ((base_gfn + npages - 1)
732 >> KVM_HPAGE_GFN_SHIFT(level));
733 lpages -= base_gfn >> KVM_HPAGE_GFN_SHIFT(level);
735 new.lpage_info[i] = vzalloc(lpages * sizeof(*new.lpage_info[i]));
737 if (!new.lpage_info[i])
740 if (base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
741 new.lpage_info[i][0].write_count = 1;
742 if ((base_gfn+npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
743 new.lpage_info[i][lpages - 1].write_count = 1;
744 ugfn = new.userspace_addr >> PAGE_SHIFT;
746 * If the gfn and userspace address are not aligned wrt each
747 * other, or if explicitly asked to, disable large page
748 * support for this slot
750 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
752 for (j = 0; j < lpages; ++j)
753 new.lpage_info[i][j].write_count = 1;
758 /* Allocate page dirty bitmap if needed */
759 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
760 if (kvm_create_dirty_bitmap(&new) < 0)
762 /* destroy any largepage mappings for dirty tracking */
764 #else /* not defined CONFIG_S390 */
765 new.user_alloc = user_alloc;
767 new.userspace_addr = mem->userspace_addr;
768 #endif /* not defined CONFIG_S390 */
772 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
775 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
776 if (mem->slot >= slots->nmemslots)
777 slots->nmemslots = mem->slot + 1;
779 slots->memslots[mem->slot].flags |= KVM_MEMSLOT_INVALID;
781 old_memslots = kvm->memslots;
782 rcu_assign_pointer(kvm->memslots, slots);
783 synchronize_srcu_expedited(&kvm->srcu);
784 /* From this point no new shadow pages pointing to a deleted
785 * memslot will be created.
787 * validation of sp->gfn happens in:
788 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
789 * - kvm_is_visible_gfn (mmu_check_roots)
791 kvm_arch_flush_shadow(kvm);
795 r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
799 /* map the pages in iommu page table */
801 r = kvm_iommu_map_pages(kvm, &new);
807 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
810 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
811 if (mem->slot >= slots->nmemslots)
812 slots->nmemslots = mem->slot + 1;
815 /* actual memory is freed via old in kvm_free_physmem_slot below */
818 new.dirty_bitmap = NULL;
819 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i)
820 new.lpage_info[i] = NULL;
823 slots->memslots[mem->slot] = new;
824 old_memslots = kvm->memslots;
825 rcu_assign_pointer(kvm->memslots, slots);
826 synchronize_srcu_expedited(&kvm->srcu);
828 kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
830 kvm_free_physmem_slot(&old, &new);
836 kvm_free_physmem_slot(&new, &old);
841 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
843 int kvm_set_memory_region(struct kvm *kvm,
844 struct kvm_userspace_memory_region *mem,
849 mutex_lock(&kvm->slots_lock);
850 r = __kvm_set_memory_region(kvm, mem, user_alloc);
851 mutex_unlock(&kvm->slots_lock);
854 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
856 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
858 kvm_userspace_memory_region *mem,
861 if (mem->slot >= KVM_MEMORY_SLOTS)
863 return kvm_set_memory_region(kvm, mem, user_alloc);
866 int kvm_get_dirty_log(struct kvm *kvm,
867 struct kvm_dirty_log *log, int *is_dirty)
869 struct kvm_memory_slot *memslot;
872 unsigned long any = 0;
875 if (log->slot >= KVM_MEMORY_SLOTS)
878 memslot = &kvm->memslots->memslots[log->slot];
880 if (!memslot->dirty_bitmap)
883 n = kvm_dirty_bitmap_bytes(memslot);
885 for (i = 0; !any && i < n/sizeof(long); ++i)
886 any = memslot->dirty_bitmap[i];
889 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
900 void kvm_disable_largepages(void)
902 largepages_enabled = false;
904 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
906 int is_error_page(struct page *page)
908 return page == bad_page || page == hwpoison_page || page == fault_page;
910 EXPORT_SYMBOL_GPL(is_error_page);
912 int is_error_pfn(pfn_t pfn)
914 return pfn == bad_pfn || pfn == hwpoison_pfn || pfn == fault_pfn;
916 EXPORT_SYMBOL_GPL(is_error_pfn);
918 int is_hwpoison_pfn(pfn_t pfn)
920 return pfn == hwpoison_pfn;
922 EXPORT_SYMBOL_GPL(is_hwpoison_pfn);
924 int is_fault_pfn(pfn_t pfn)
926 return pfn == fault_pfn;
928 EXPORT_SYMBOL_GPL(is_fault_pfn);
930 static inline unsigned long bad_hva(void)
935 int kvm_is_error_hva(unsigned long addr)
937 return addr == bad_hva();
939 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
941 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm_memslots *slots,
946 for (i = 0; i < slots->nmemslots; ++i) {
947 struct kvm_memory_slot *memslot = &slots->memslots[i];
949 if (gfn >= memslot->base_gfn
950 && gfn < memslot->base_gfn + memslot->npages)
956 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
958 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
960 EXPORT_SYMBOL_GPL(gfn_to_memslot);
962 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
965 struct kvm_memslots *slots = kvm_memslots(kvm);
967 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
968 struct kvm_memory_slot *memslot = &slots->memslots[i];
970 if (memslot->flags & KVM_MEMSLOT_INVALID)
973 if (gfn >= memslot->base_gfn
974 && gfn < memslot->base_gfn + memslot->npages)
979 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
981 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
983 struct vm_area_struct *vma;
984 unsigned long addr, size;
988 addr = gfn_to_hva(kvm, gfn);
989 if (kvm_is_error_hva(addr))
992 down_read(¤t->mm->mmap_sem);
993 vma = find_vma(current->mm, addr);
997 size = vma_kernel_pagesize(vma);
1000 up_read(¤t->mm->mmap_sem);
1005 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1008 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1012 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1014 return gfn_to_hva_memslot(slot, gfn);
1017 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1019 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1021 EXPORT_SYMBOL_GPL(gfn_to_hva);
1023 static pfn_t get_fault_pfn(void)
1025 get_page(fault_page);
1029 int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1030 unsigned long start, int write, struct page **page)
1032 int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1035 flags |= FOLL_WRITE;
1037 return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1040 static inline int check_user_page_hwpoison(unsigned long addr)
1042 int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1044 rc = __get_user_pages(current, current->mm, addr, 1,
1045 flags, NULL, NULL, NULL);
1046 return rc == -EHWPOISON;
1049 static pfn_t hva_to_pfn(struct kvm *kvm, unsigned long addr, bool atomic,
1050 bool *async, bool write_fault, bool *writable)
1052 struct page *page[1];
1056 /* we can do it either atomically or asynchronously, not both */
1057 BUG_ON(atomic && async);
1059 BUG_ON(!write_fault && !writable);
1064 if (atomic || async)
1065 npages = __get_user_pages_fast(addr, 1, 1, page);
1067 if (unlikely(npages != 1) && !atomic) {
1071 *writable = write_fault;
1074 down_read(¤t->mm->mmap_sem);
1075 npages = get_user_page_nowait(current, current->mm,
1076 addr, write_fault, page);
1077 up_read(¤t->mm->mmap_sem);
1079 npages = get_user_pages_fast(addr, 1, write_fault,
1082 /* map read fault as writable if possible */
1083 if (unlikely(!write_fault) && npages == 1) {
1084 struct page *wpage[1];
1086 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1096 if (unlikely(npages != 1)) {
1097 struct vm_area_struct *vma;
1100 return get_fault_pfn();
1102 down_read(¤t->mm->mmap_sem);
1103 if (npages == -EHWPOISON ||
1104 (!async && check_user_page_hwpoison(addr))) {
1105 up_read(¤t->mm->mmap_sem);
1106 get_page(hwpoison_page);
1107 return page_to_pfn(hwpoison_page);
1110 vma = find_vma_intersection(current->mm, addr, addr+1);
1113 pfn = get_fault_pfn();
1114 else if ((vma->vm_flags & VM_PFNMAP)) {
1115 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1117 BUG_ON(!kvm_is_mmio_pfn(pfn));
1119 if (async && (vma->vm_flags & VM_WRITE))
1121 pfn = get_fault_pfn();
1123 up_read(¤t->mm->mmap_sem);
1125 pfn = page_to_pfn(page[0]);
1130 pfn_t hva_to_pfn_atomic(struct kvm *kvm, unsigned long addr)
1132 return hva_to_pfn(kvm, addr, true, NULL, true, NULL);
1134 EXPORT_SYMBOL_GPL(hva_to_pfn_atomic);
1136 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1137 bool write_fault, bool *writable)
1144 addr = gfn_to_hva(kvm, gfn);
1145 if (kvm_is_error_hva(addr)) {
1147 return page_to_pfn(bad_page);
1150 return hva_to_pfn(kvm, addr, atomic, async, write_fault, writable);
1153 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1155 return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1157 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1159 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1160 bool write_fault, bool *writable)
1162 return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1164 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1166 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1168 return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1170 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1172 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1175 return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1177 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1179 pfn_t gfn_to_pfn_memslot(struct kvm *kvm,
1180 struct kvm_memory_slot *slot, gfn_t gfn)
1182 unsigned long addr = gfn_to_hva_memslot(slot, gfn);
1183 return hva_to_pfn(kvm, addr, false, NULL, true, NULL);
1186 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1192 addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1193 if (kvm_is_error_hva(addr))
1196 if (entry < nr_pages)
1199 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1201 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1203 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1207 pfn = gfn_to_pfn(kvm, gfn);
1208 if (!kvm_is_mmio_pfn(pfn))
1209 return pfn_to_page(pfn);
1211 WARN_ON(kvm_is_mmio_pfn(pfn));
1217 EXPORT_SYMBOL_GPL(gfn_to_page);
1219 void kvm_release_page_clean(struct page *page)
1221 kvm_release_pfn_clean(page_to_pfn(page));
1223 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1225 void kvm_release_pfn_clean(pfn_t pfn)
1227 if (!kvm_is_mmio_pfn(pfn))
1228 put_page(pfn_to_page(pfn));
1230 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1232 void kvm_release_page_dirty(struct page *page)
1234 kvm_release_pfn_dirty(page_to_pfn(page));
1236 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1238 void kvm_release_pfn_dirty(pfn_t pfn)
1240 kvm_set_pfn_dirty(pfn);
1241 kvm_release_pfn_clean(pfn);
1243 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1245 void kvm_set_page_dirty(struct page *page)
1247 kvm_set_pfn_dirty(page_to_pfn(page));
1249 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1251 void kvm_set_pfn_dirty(pfn_t pfn)
1253 if (!kvm_is_mmio_pfn(pfn)) {
1254 struct page *page = pfn_to_page(pfn);
1255 if (!PageReserved(page))
1259 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1261 void kvm_set_pfn_accessed(pfn_t pfn)
1263 if (!kvm_is_mmio_pfn(pfn))
1264 mark_page_accessed(pfn_to_page(pfn));
1266 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1268 void kvm_get_pfn(pfn_t pfn)
1270 if (!kvm_is_mmio_pfn(pfn))
1271 get_page(pfn_to_page(pfn));
1273 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1275 static int next_segment(unsigned long len, int offset)
1277 if (len > PAGE_SIZE - offset)
1278 return PAGE_SIZE - offset;
1283 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1289 addr = gfn_to_hva(kvm, gfn);
1290 if (kvm_is_error_hva(addr))
1292 r = __copy_from_user(data, (void __user *)addr + offset, len);
1297 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1299 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1301 gfn_t gfn = gpa >> PAGE_SHIFT;
1303 int offset = offset_in_page(gpa);
1306 while ((seg = next_segment(len, offset)) != 0) {
1307 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1317 EXPORT_SYMBOL_GPL(kvm_read_guest);
1319 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1324 gfn_t gfn = gpa >> PAGE_SHIFT;
1325 int offset = offset_in_page(gpa);
1327 addr = gfn_to_hva(kvm, gfn);
1328 if (kvm_is_error_hva(addr))
1330 pagefault_disable();
1331 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1337 EXPORT_SYMBOL(kvm_read_guest_atomic);
1339 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1340 int offset, int len)
1345 addr = gfn_to_hva(kvm, gfn);
1346 if (kvm_is_error_hva(addr))
1348 r = __copy_to_user((void __user *)addr + offset, data, len);
1351 mark_page_dirty(kvm, gfn);
1354 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1356 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1359 gfn_t gfn = gpa >> PAGE_SHIFT;
1361 int offset = offset_in_page(gpa);
1364 while ((seg = next_segment(len, offset)) != 0) {
1365 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1376 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1379 struct kvm_memslots *slots = kvm_memslots(kvm);
1380 int offset = offset_in_page(gpa);
1381 gfn_t gfn = gpa >> PAGE_SHIFT;
1384 ghc->generation = slots->generation;
1385 ghc->memslot = __gfn_to_memslot(slots, gfn);
1386 ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1387 if (!kvm_is_error_hva(ghc->hva))
1394 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1396 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1397 void *data, unsigned long len)
1399 struct kvm_memslots *slots = kvm_memslots(kvm);
1402 if (slots->generation != ghc->generation)
1403 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1405 if (kvm_is_error_hva(ghc->hva))
1408 r = __copy_to_user((void __user *)ghc->hva, data, len);
1411 mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1415 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1417 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1419 return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1422 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1424 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1426 gfn_t gfn = gpa >> PAGE_SHIFT;
1428 int offset = offset_in_page(gpa);
1431 while ((seg = next_segment(len, offset)) != 0) {
1432 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1441 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1443 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1446 if (memslot && memslot->dirty_bitmap) {
1447 unsigned long rel_gfn = gfn - memslot->base_gfn;
1449 __set_bit_le(rel_gfn, memslot->dirty_bitmap);
1453 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1455 struct kvm_memory_slot *memslot;
1457 memslot = gfn_to_memslot(kvm, gfn);
1458 mark_page_dirty_in_slot(kvm, memslot, gfn);
1462 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1464 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1469 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1471 if (kvm_arch_vcpu_runnable(vcpu)) {
1472 kvm_make_request(KVM_REQ_UNHALT, vcpu);
1475 if (kvm_cpu_has_pending_timer(vcpu))
1477 if (signal_pending(current))
1483 finish_wait(&vcpu->wq, &wait);
1486 void kvm_resched(struct kvm_vcpu *vcpu)
1488 if (!need_resched())
1492 EXPORT_SYMBOL_GPL(kvm_resched);
1494 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1496 struct kvm *kvm = me->kvm;
1497 struct kvm_vcpu *vcpu;
1498 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1504 * We boost the priority of a VCPU that is runnable but not
1505 * currently running, because it got preempted by something
1506 * else and called schedule in __vcpu_run. Hopefully that
1507 * VCPU is holding the lock that we need and will release it.
1508 * We approximate round-robin by starting at the last boosted VCPU.
1510 for (pass = 0; pass < 2 && !yielded; pass++) {
1511 kvm_for_each_vcpu(i, vcpu, kvm) {
1512 struct task_struct *task = NULL;
1514 if (!pass && i < last_boosted_vcpu) {
1515 i = last_boosted_vcpu;
1517 } else if (pass && i > last_boosted_vcpu)
1521 if (waitqueue_active(&vcpu->wq))
1524 pid = rcu_dereference(vcpu->pid);
1526 task = get_pid_task(vcpu->pid, PIDTYPE_PID);
1530 if (task->flags & PF_VCPU) {
1531 put_task_struct(task);
1534 if (yield_to(task, 1)) {
1535 put_task_struct(task);
1536 kvm->last_boosted_vcpu = i;
1540 put_task_struct(task);
1544 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1546 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1548 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1551 if (vmf->pgoff == 0)
1552 page = virt_to_page(vcpu->run);
1554 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1555 page = virt_to_page(vcpu->arch.pio_data);
1557 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1558 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1559 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1562 return VM_FAULT_SIGBUS;
1568 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1569 .fault = kvm_vcpu_fault,
1572 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1574 vma->vm_ops = &kvm_vcpu_vm_ops;
1578 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1580 struct kvm_vcpu *vcpu = filp->private_data;
1582 kvm_put_kvm(vcpu->kvm);
1586 static struct file_operations kvm_vcpu_fops = {
1587 .release = kvm_vcpu_release,
1588 .unlocked_ioctl = kvm_vcpu_ioctl,
1589 .compat_ioctl = kvm_vcpu_ioctl,
1590 .mmap = kvm_vcpu_mmap,
1591 .llseek = noop_llseek,
1595 * Allocates an inode for the vcpu.
1597 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1599 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1603 * Creates some virtual cpus. Good luck creating more than one.
1605 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1608 struct kvm_vcpu *vcpu, *v;
1610 vcpu = kvm_arch_vcpu_create(kvm, id);
1612 return PTR_ERR(vcpu);
1614 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1616 r = kvm_arch_vcpu_setup(vcpu);
1620 mutex_lock(&kvm->lock);
1621 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1623 goto unlock_vcpu_destroy;
1626 kvm_for_each_vcpu(r, v, kvm)
1627 if (v->vcpu_id == id) {
1629 goto unlock_vcpu_destroy;
1632 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1634 /* Now it's all set up, let userspace reach it */
1636 r = create_vcpu_fd(vcpu);
1639 goto unlock_vcpu_destroy;
1642 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1644 atomic_inc(&kvm->online_vcpus);
1646 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1647 if (kvm->bsp_vcpu_id == id)
1648 kvm->bsp_vcpu = vcpu;
1650 mutex_unlock(&kvm->lock);
1653 unlock_vcpu_destroy:
1654 mutex_unlock(&kvm->lock);
1656 kvm_arch_vcpu_destroy(vcpu);
1660 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1663 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1664 vcpu->sigset_active = 1;
1665 vcpu->sigset = *sigset;
1667 vcpu->sigset_active = 0;
1671 static long kvm_vcpu_ioctl(struct file *filp,
1672 unsigned int ioctl, unsigned long arg)
1674 struct kvm_vcpu *vcpu = filp->private_data;
1675 void __user *argp = (void __user *)arg;
1677 struct kvm_fpu *fpu = NULL;
1678 struct kvm_sregs *kvm_sregs = NULL;
1680 if (vcpu->kvm->mm != current->mm)
1683 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1685 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1686 * so vcpu_load() would break it.
1688 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1689 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1699 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1700 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1702 case KVM_GET_REGS: {
1703 struct kvm_regs *kvm_regs;
1706 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1709 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1713 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1720 case KVM_SET_REGS: {
1721 struct kvm_regs *kvm_regs;
1724 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1728 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1730 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1738 case KVM_GET_SREGS: {
1739 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1743 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1747 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1752 case KVM_SET_SREGS: {
1753 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1758 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1760 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1766 case KVM_GET_MP_STATE: {
1767 struct kvm_mp_state mp_state;
1769 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1773 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1778 case KVM_SET_MP_STATE: {
1779 struct kvm_mp_state mp_state;
1782 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1784 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1790 case KVM_TRANSLATE: {
1791 struct kvm_translation tr;
1794 if (copy_from_user(&tr, argp, sizeof tr))
1796 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1800 if (copy_to_user(argp, &tr, sizeof tr))
1805 case KVM_SET_GUEST_DEBUG: {
1806 struct kvm_guest_debug dbg;
1809 if (copy_from_user(&dbg, argp, sizeof dbg))
1811 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1817 case KVM_SET_SIGNAL_MASK: {
1818 struct kvm_signal_mask __user *sigmask_arg = argp;
1819 struct kvm_signal_mask kvm_sigmask;
1820 sigset_t sigset, *p;
1825 if (copy_from_user(&kvm_sigmask, argp,
1826 sizeof kvm_sigmask))
1829 if (kvm_sigmask.len != sizeof sigset)
1832 if (copy_from_user(&sigset, sigmask_arg->sigset,
1837 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
1841 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1845 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1849 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1855 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1860 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1862 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1869 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1878 static long kvm_vm_ioctl(struct file *filp,
1879 unsigned int ioctl, unsigned long arg)
1881 struct kvm *kvm = filp->private_data;
1882 void __user *argp = (void __user *)arg;
1885 if (kvm->mm != current->mm)
1888 case KVM_CREATE_VCPU:
1889 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1893 case KVM_SET_USER_MEMORY_REGION: {
1894 struct kvm_userspace_memory_region kvm_userspace_mem;
1897 if (copy_from_user(&kvm_userspace_mem, argp,
1898 sizeof kvm_userspace_mem))
1901 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1906 case KVM_GET_DIRTY_LOG: {
1907 struct kvm_dirty_log log;
1910 if (copy_from_user(&log, argp, sizeof log))
1912 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1917 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1918 case KVM_REGISTER_COALESCED_MMIO: {
1919 struct kvm_coalesced_mmio_zone zone;
1921 if (copy_from_user(&zone, argp, sizeof zone))
1923 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1929 case KVM_UNREGISTER_COALESCED_MMIO: {
1930 struct kvm_coalesced_mmio_zone zone;
1932 if (copy_from_user(&zone, argp, sizeof zone))
1934 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1942 struct kvm_irqfd data;
1945 if (copy_from_user(&data, argp, sizeof data))
1947 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
1950 case KVM_IOEVENTFD: {
1951 struct kvm_ioeventfd data;
1954 if (copy_from_user(&data, argp, sizeof data))
1956 r = kvm_ioeventfd(kvm, &data);
1959 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1960 case KVM_SET_BOOT_CPU_ID:
1962 mutex_lock(&kvm->lock);
1963 if (atomic_read(&kvm->online_vcpus) != 0)
1966 kvm->bsp_vcpu_id = arg;
1967 mutex_unlock(&kvm->lock);
1971 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1973 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
1979 #ifdef CONFIG_COMPAT
1980 struct compat_kvm_dirty_log {
1984 compat_uptr_t dirty_bitmap; /* one bit per page */
1989 static long kvm_vm_compat_ioctl(struct file *filp,
1990 unsigned int ioctl, unsigned long arg)
1992 struct kvm *kvm = filp->private_data;
1995 if (kvm->mm != current->mm)
1998 case KVM_GET_DIRTY_LOG: {
1999 struct compat_kvm_dirty_log compat_log;
2000 struct kvm_dirty_log log;
2003 if (copy_from_user(&compat_log, (void __user *)arg,
2004 sizeof(compat_log)))
2006 log.slot = compat_log.slot;
2007 log.padding1 = compat_log.padding1;
2008 log.padding2 = compat_log.padding2;
2009 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2011 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2017 r = kvm_vm_ioctl(filp, ioctl, arg);
2025 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2027 struct page *page[1];
2030 gfn_t gfn = vmf->pgoff;
2031 struct kvm *kvm = vma->vm_file->private_data;
2033 addr = gfn_to_hva(kvm, gfn);
2034 if (kvm_is_error_hva(addr))
2035 return VM_FAULT_SIGBUS;
2037 npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2039 if (unlikely(npages != 1))
2040 return VM_FAULT_SIGBUS;
2042 vmf->page = page[0];
2046 static const struct vm_operations_struct kvm_vm_vm_ops = {
2047 .fault = kvm_vm_fault,
2050 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2052 vma->vm_ops = &kvm_vm_vm_ops;
2056 static struct file_operations kvm_vm_fops = {
2057 .release = kvm_vm_release,
2058 .unlocked_ioctl = kvm_vm_ioctl,
2059 #ifdef CONFIG_COMPAT
2060 .compat_ioctl = kvm_vm_compat_ioctl,
2062 .mmap = kvm_vm_mmap,
2063 .llseek = noop_llseek,
2066 static int kvm_dev_ioctl_create_vm(void)
2071 kvm = kvm_create_vm();
2073 return PTR_ERR(kvm);
2074 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2075 r = kvm_coalesced_mmio_init(kvm);
2081 r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2088 static long kvm_dev_ioctl_check_extension_generic(long arg)
2091 case KVM_CAP_USER_MEMORY:
2092 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2093 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2094 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2095 case KVM_CAP_SET_BOOT_CPU_ID:
2097 case KVM_CAP_INTERNAL_ERROR_DATA:
2099 #ifdef CONFIG_HAVE_KVM_IRQCHIP
2100 case KVM_CAP_IRQ_ROUTING:
2101 return KVM_MAX_IRQ_ROUTES;
2106 return kvm_dev_ioctl_check_extension(arg);
2109 static long kvm_dev_ioctl(struct file *filp,
2110 unsigned int ioctl, unsigned long arg)
2115 case KVM_GET_API_VERSION:
2119 r = KVM_API_VERSION;
2125 r = kvm_dev_ioctl_create_vm();
2127 case KVM_CHECK_EXTENSION:
2128 r = kvm_dev_ioctl_check_extension_generic(arg);
2130 case KVM_GET_VCPU_MMAP_SIZE:
2134 r = PAGE_SIZE; /* struct kvm_run */
2136 r += PAGE_SIZE; /* pio data page */
2138 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2139 r += PAGE_SIZE; /* coalesced mmio ring page */
2142 case KVM_TRACE_ENABLE:
2143 case KVM_TRACE_PAUSE:
2144 case KVM_TRACE_DISABLE:
2148 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2154 static struct file_operations kvm_chardev_ops = {
2155 .unlocked_ioctl = kvm_dev_ioctl,
2156 .compat_ioctl = kvm_dev_ioctl,
2157 .llseek = noop_llseek,
2160 static struct miscdevice kvm_dev = {
2166 static void hardware_enable_nolock(void *junk)
2168 int cpu = raw_smp_processor_id();
2171 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2174 cpumask_set_cpu(cpu, cpus_hardware_enabled);
2176 r = kvm_arch_hardware_enable(NULL);
2179 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2180 atomic_inc(&hardware_enable_failed);
2181 printk(KERN_INFO "kvm: enabling virtualization on "
2182 "CPU%d failed\n", cpu);
2186 static void hardware_enable(void *junk)
2188 raw_spin_lock(&kvm_lock);
2189 hardware_enable_nolock(junk);
2190 raw_spin_unlock(&kvm_lock);
2193 static void hardware_disable_nolock(void *junk)
2195 int cpu = raw_smp_processor_id();
2197 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2199 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2200 kvm_arch_hardware_disable(NULL);
2203 static void hardware_disable(void *junk)
2205 raw_spin_lock(&kvm_lock);
2206 hardware_disable_nolock(junk);
2207 raw_spin_unlock(&kvm_lock);
2210 static void hardware_disable_all_nolock(void)
2212 BUG_ON(!kvm_usage_count);
2215 if (!kvm_usage_count)
2216 on_each_cpu(hardware_disable_nolock, NULL, 1);
2219 static void hardware_disable_all(void)
2221 raw_spin_lock(&kvm_lock);
2222 hardware_disable_all_nolock();
2223 raw_spin_unlock(&kvm_lock);
2226 static int hardware_enable_all(void)
2230 raw_spin_lock(&kvm_lock);
2233 if (kvm_usage_count == 1) {
2234 atomic_set(&hardware_enable_failed, 0);
2235 on_each_cpu(hardware_enable_nolock, NULL, 1);
2237 if (atomic_read(&hardware_enable_failed)) {
2238 hardware_disable_all_nolock();
2243 raw_spin_unlock(&kvm_lock);
2248 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2253 if (!kvm_usage_count)
2256 val &= ~CPU_TASKS_FROZEN;
2259 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2261 hardware_disable(NULL);
2264 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2266 hardware_enable(NULL);
2273 asmlinkage void kvm_spurious_fault(void)
2275 /* Fault while not rebooting. We want the trace. */
2278 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
2280 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2284 * Some (well, at least mine) BIOSes hang on reboot if
2287 * And Intel TXT required VMX off for all cpu when system shutdown.
2289 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2290 kvm_rebooting = true;
2291 on_each_cpu(hardware_disable_nolock, NULL, 1);
2295 static struct notifier_block kvm_reboot_notifier = {
2296 .notifier_call = kvm_reboot,
2300 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2304 for (i = 0; i < bus->dev_count; i++) {
2305 struct kvm_io_device *pos = bus->devs[i];
2307 kvm_iodevice_destructor(pos);
2312 /* kvm_io_bus_write - called under kvm->slots_lock */
2313 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2314 int len, const void *val)
2317 struct kvm_io_bus *bus;
2319 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2320 for (i = 0; i < bus->dev_count; i++)
2321 if (!kvm_iodevice_write(bus->devs[i], addr, len, val))
2326 /* kvm_io_bus_read - called under kvm->slots_lock */
2327 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2331 struct kvm_io_bus *bus;
2333 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2334 for (i = 0; i < bus->dev_count; i++)
2335 if (!kvm_iodevice_read(bus->devs[i], addr, len, val))
2340 /* Caller must hold slots_lock. */
2341 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2342 struct kvm_io_device *dev)
2344 struct kvm_io_bus *new_bus, *bus;
2346 bus = kvm->buses[bus_idx];
2347 if (bus->dev_count > NR_IOBUS_DEVS-1)
2350 new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2353 memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2354 new_bus->devs[new_bus->dev_count++] = dev;
2355 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2356 synchronize_srcu_expedited(&kvm->srcu);
2362 /* Caller must hold slots_lock. */
2363 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2364 struct kvm_io_device *dev)
2367 struct kvm_io_bus *new_bus, *bus;
2369 new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2373 bus = kvm->buses[bus_idx];
2374 memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2377 for (i = 0; i < new_bus->dev_count; i++)
2378 if (new_bus->devs[i] == dev) {
2380 new_bus->devs[i] = new_bus->devs[--new_bus->dev_count];
2389 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2390 synchronize_srcu_expedited(&kvm->srcu);
2395 static struct notifier_block kvm_cpu_notifier = {
2396 .notifier_call = kvm_cpu_hotplug,
2399 static int vm_stat_get(void *_offset, u64 *val)
2401 unsigned offset = (long)_offset;
2405 raw_spin_lock(&kvm_lock);
2406 list_for_each_entry(kvm, &vm_list, vm_list)
2407 *val += *(u32 *)((void *)kvm + offset);
2408 raw_spin_unlock(&kvm_lock);
2412 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2414 static int vcpu_stat_get(void *_offset, u64 *val)
2416 unsigned offset = (long)_offset;
2418 struct kvm_vcpu *vcpu;
2422 raw_spin_lock(&kvm_lock);
2423 list_for_each_entry(kvm, &vm_list, vm_list)
2424 kvm_for_each_vcpu(i, vcpu, kvm)
2425 *val += *(u32 *)((void *)vcpu + offset);
2427 raw_spin_unlock(&kvm_lock);
2431 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2433 static const struct file_operations *stat_fops[] = {
2434 [KVM_STAT_VCPU] = &vcpu_stat_fops,
2435 [KVM_STAT_VM] = &vm_stat_fops,
2438 static void kvm_init_debug(void)
2440 struct kvm_stats_debugfs_item *p;
2442 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2443 for (p = debugfs_entries; p->name; ++p)
2444 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2445 (void *)(long)p->offset,
2446 stat_fops[p->kind]);
2449 static void kvm_exit_debug(void)
2451 struct kvm_stats_debugfs_item *p;
2453 for (p = debugfs_entries; p->name; ++p)
2454 debugfs_remove(p->dentry);
2455 debugfs_remove(kvm_debugfs_dir);
2458 static int kvm_suspend(void)
2460 if (kvm_usage_count)
2461 hardware_disable_nolock(NULL);
2465 static void kvm_resume(void)
2467 if (kvm_usage_count) {
2468 WARN_ON(raw_spin_is_locked(&kvm_lock));
2469 hardware_enable_nolock(NULL);
2473 static struct syscore_ops kvm_syscore_ops = {
2474 .suspend = kvm_suspend,
2475 .resume = kvm_resume,
2478 struct page *bad_page;
2482 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2484 return container_of(pn, struct kvm_vcpu, preempt_notifier);
2487 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2489 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2491 kvm_arch_vcpu_load(vcpu, cpu);
2494 static void kvm_sched_out(struct preempt_notifier *pn,
2495 struct task_struct *next)
2497 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2499 kvm_arch_vcpu_put(vcpu);
2502 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2503 struct module *module)
2508 r = kvm_arch_init(opaque);
2512 bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2514 if (bad_page == NULL) {
2519 bad_pfn = page_to_pfn(bad_page);
2521 hwpoison_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2523 if (hwpoison_page == NULL) {
2528 hwpoison_pfn = page_to_pfn(hwpoison_page);
2530 fault_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2532 if (fault_page == NULL) {
2537 fault_pfn = page_to_pfn(fault_page);
2539 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2544 r = kvm_arch_hardware_setup();
2548 for_each_online_cpu(cpu) {
2549 smp_call_function_single(cpu,
2550 kvm_arch_check_processor_compat,
2556 r = register_cpu_notifier(&kvm_cpu_notifier);
2559 register_reboot_notifier(&kvm_reboot_notifier);
2561 /* A kmem cache lets us meet the alignment requirements of fx_save. */
2563 vcpu_align = __alignof__(struct kvm_vcpu);
2564 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2566 if (!kvm_vcpu_cache) {
2571 r = kvm_async_pf_init();
2575 kvm_chardev_ops.owner = module;
2576 kvm_vm_fops.owner = module;
2577 kvm_vcpu_fops.owner = module;
2579 r = misc_register(&kvm_dev);
2581 printk(KERN_ERR "kvm: misc device register failed\n");
2585 register_syscore_ops(&kvm_syscore_ops);
2587 kvm_preempt_ops.sched_in = kvm_sched_in;
2588 kvm_preempt_ops.sched_out = kvm_sched_out;
2595 kvm_async_pf_deinit();
2597 kmem_cache_destroy(kvm_vcpu_cache);
2599 unregister_reboot_notifier(&kvm_reboot_notifier);
2600 unregister_cpu_notifier(&kvm_cpu_notifier);
2603 kvm_arch_hardware_unsetup();
2605 free_cpumask_var(cpus_hardware_enabled);
2608 __free_page(fault_page);
2610 __free_page(hwpoison_page);
2611 __free_page(bad_page);
2617 EXPORT_SYMBOL_GPL(kvm_init);
2622 misc_deregister(&kvm_dev);
2623 kmem_cache_destroy(kvm_vcpu_cache);
2624 kvm_async_pf_deinit();
2625 unregister_syscore_ops(&kvm_syscore_ops);
2626 unregister_reboot_notifier(&kvm_reboot_notifier);
2627 unregister_cpu_notifier(&kvm_cpu_notifier);
2628 on_each_cpu(hardware_disable_nolock, NULL, 1);
2629 kvm_arch_hardware_unsetup();
2631 free_cpumask_var(cpus_hardware_enabled);
2632 __free_page(hwpoison_page);
2633 __free_page(bad_page);
2635 EXPORT_SYMBOL_GPL(kvm_exit);