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.
19 #include <kvm/iodev.h>
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>
50 #include <linux/sort.h>
51 #include <linux/bsearch.h>
53 #include <asm/processor.h>
55 #include <asm/ioctl.h>
56 #include <asm/uaccess.h>
57 #include <asm/pgtable.h>
59 #include "coalesced_mmio.h"
63 #define CREATE_TRACE_POINTS
64 #include <trace/events/kvm.h>
66 MODULE_AUTHOR("Qumranet");
67 MODULE_LICENSE("GPL");
69 static unsigned int halt_poll_ns;
70 module_param(halt_poll_ns, uint, S_IRUGO | S_IWUSR);
75 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
78 DEFINE_SPINLOCK(kvm_lock);
79 static DEFINE_RAW_SPINLOCK(kvm_count_lock);
82 static cpumask_var_t cpus_hardware_enabled;
83 static int kvm_usage_count;
84 static atomic_t hardware_enable_failed;
86 struct kmem_cache *kvm_vcpu_cache;
87 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
89 static __read_mostly struct preempt_ops kvm_preempt_ops;
91 struct dentry *kvm_debugfs_dir;
92 EXPORT_SYMBOL_GPL(kvm_debugfs_dir);
94 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
96 #ifdef CONFIG_KVM_COMPAT
97 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
100 static int hardware_enable_all(void);
101 static void hardware_disable_all(void);
103 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
105 static void kvm_release_pfn_dirty(pfn_t pfn);
106 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot, gfn_t gfn);
108 __visible bool kvm_rebooting;
109 EXPORT_SYMBOL_GPL(kvm_rebooting);
111 static bool largepages_enabled = true;
113 bool kvm_is_reserved_pfn(pfn_t pfn)
116 return PageReserved(pfn_to_page(pfn));
122 * Switches to specified vcpu, until a matching vcpu_put()
124 int vcpu_load(struct kvm_vcpu *vcpu)
128 if (mutex_lock_killable(&vcpu->mutex))
131 preempt_notifier_register(&vcpu->preempt_notifier);
132 kvm_arch_vcpu_load(vcpu, cpu);
137 void vcpu_put(struct kvm_vcpu *vcpu)
140 kvm_arch_vcpu_put(vcpu);
141 preempt_notifier_unregister(&vcpu->preempt_notifier);
143 mutex_unlock(&vcpu->mutex);
146 static void ack_flush(void *_completed)
150 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
155 struct kvm_vcpu *vcpu;
157 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
160 kvm_for_each_vcpu(i, vcpu, kvm) {
161 kvm_make_request(req, vcpu);
164 /* Set ->requests bit before we read ->mode */
167 if (cpus != NULL && cpu != -1 && cpu != me &&
168 kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
169 cpumask_set_cpu(cpu, cpus);
171 if (unlikely(cpus == NULL))
172 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
173 else if (!cpumask_empty(cpus))
174 smp_call_function_many(cpus, ack_flush, NULL, 1);
178 free_cpumask_var(cpus);
182 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
183 void kvm_flush_remote_tlbs(struct kvm *kvm)
185 long dirty_count = kvm->tlbs_dirty;
188 if (kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
189 ++kvm->stat.remote_tlb_flush;
190 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
192 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
195 void kvm_reload_remote_mmus(struct kvm *kvm)
197 kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
200 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
202 kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
205 void kvm_make_scan_ioapic_request(struct kvm *kvm)
207 kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
210 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
215 mutex_init(&vcpu->mutex);
220 init_waitqueue_head(&vcpu->wq);
221 kvm_async_pf_vcpu_init(vcpu);
223 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
228 vcpu->run = page_address(page);
230 kvm_vcpu_set_in_spin_loop(vcpu, false);
231 kvm_vcpu_set_dy_eligible(vcpu, false);
232 vcpu->preempted = false;
234 r = kvm_arch_vcpu_init(vcpu);
240 free_page((unsigned long)vcpu->run);
244 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
246 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
249 kvm_arch_vcpu_uninit(vcpu);
250 free_page((unsigned long)vcpu->run);
252 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
254 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
255 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
257 return container_of(mn, struct kvm, mmu_notifier);
260 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
261 struct mm_struct *mm,
262 unsigned long address)
264 struct kvm *kvm = mmu_notifier_to_kvm(mn);
265 int need_tlb_flush, idx;
268 * When ->invalidate_page runs, the linux pte has been zapped
269 * already but the page is still allocated until
270 * ->invalidate_page returns. So if we increase the sequence
271 * here the kvm page fault will notice if the spte can't be
272 * established because the page is going to be freed. If
273 * instead the kvm page fault establishes the spte before
274 * ->invalidate_page runs, kvm_unmap_hva will release it
277 * The sequence increase only need to be seen at spin_unlock
278 * time, and not at spin_lock time.
280 * Increasing the sequence after the spin_unlock would be
281 * unsafe because the kvm page fault could then establish the
282 * pte after kvm_unmap_hva returned, without noticing the page
283 * is going to be freed.
285 idx = srcu_read_lock(&kvm->srcu);
286 spin_lock(&kvm->mmu_lock);
288 kvm->mmu_notifier_seq++;
289 need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
290 /* we've to flush the tlb before the pages can be freed */
292 kvm_flush_remote_tlbs(kvm);
294 spin_unlock(&kvm->mmu_lock);
296 kvm_arch_mmu_notifier_invalidate_page(kvm, address);
298 srcu_read_unlock(&kvm->srcu, idx);
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 need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
334 need_tlb_flush |= kvm->tlbs_dirty;
335 /* we've to flush the tlb before the pages can be freed */
337 kvm_flush_remote_tlbs(kvm);
339 spin_unlock(&kvm->mmu_lock);
340 srcu_read_unlock(&kvm->srcu, idx);
343 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
344 struct mm_struct *mm,
348 struct kvm *kvm = mmu_notifier_to_kvm(mn);
350 spin_lock(&kvm->mmu_lock);
352 * This sequence increase will notify the kvm page fault that
353 * the page that is going to be mapped in the spte could have
356 kvm->mmu_notifier_seq++;
359 * The above sequence increase must be visible before the
360 * below count decrease, which is ensured by the smp_wmb above
361 * in conjunction with the smp_rmb in mmu_notifier_retry().
363 kvm->mmu_notifier_count--;
364 spin_unlock(&kvm->mmu_lock);
366 BUG_ON(kvm->mmu_notifier_count < 0);
369 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
370 struct mm_struct *mm,
374 struct kvm *kvm = mmu_notifier_to_kvm(mn);
377 idx = srcu_read_lock(&kvm->srcu);
378 spin_lock(&kvm->mmu_lock);
380 young = kvm_age_hva(kvm, start, end);
382 kvm_flush_remote_tlbs(kvm);
384 spin_unlock(&kvm->mmu_lock);
385 srcu_read_unlock(&kvm->srcu, idx);
390 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
391 struct mm_struct *mm,
392 unsigned long address)
394 struct kvm *kvm = mmu_notifier_to_kvm(mn);
397 idx = srcu_read_lock(&kvm->srcu);
398 spin_lock(&kvm->mmu_lock);
399 young = kvm_test_age_hva(kvm, address);
400 spin_unlock(&kvm->mmu_lock);
401 srcu_read_unlock(&kvm->srcu, idx);
406 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
407 struct mm_struct *mm)
409 struct kvm *kvm = mmu_notifier_to_kvm(mn);
412 idx = srcu_read_lock(&kvm->srcu);
413 kvm_arch_flush_shadow_all(kvm);
414 srcu_read_unlock(&kvm->srcu, idx);
417 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
418 .invalidate_page = kvm_mmu_notifier_invalidate_page,
419 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
420 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
421 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
422 .test_young = kvm_mmu_notifier_test_young,
423 .change_pte = kvm_mmu_notifier_change_pte,
424 .release = kvm_mmu_notifier_release,
427 static int kvm_init_mmu_notifier(struct kvm *kvm)
429 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
430 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
433 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
435 static int kvm_init_mmu_notifier(struct kvm *kvm)
440 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
442 static struct kvm_memslots *kvm_alloc_memslots(void)
445 struct kvm_memslots *slots;
447 slots = kvm_kvzalloc(sizeof(struct kvm_memslots));
452 * Init kvm generation close to the maximum to easily test the
453 * code of handling generation number wrap-around.
455 slots->generation = -150;
456 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
457 slots->id_to_index[i] = slots->memslots[i].id = i;
462 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
464 if (!memslot->dirty_bitmap)
467 kvfree(memslot->dirty_bitmap);
468 memslot->dirty_bitmap = NULL;
472 * Free any memory in @free but not in @dont.
474 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
475 struct kvm_memory_slot *dont)
477 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
478 kvm_destroy_dirty_bitmap(free);
480 kvm_arch_free_memslot(kvm, free, dont);
485 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
487 struct kvm_memory_slot *memslot;
492 kvm_for_each_memslot(memslot, slots)
493 kvm_free_memslot(kvm, memslot, NULL);
498 static struct kvm *kvm_create_vm(unsigned long type)
501 struct kvm *kvm = kvm_arch_alloc_vm();
504 return ERR_PTR(-ENOMEM);
506 r = kvm_arch_init_vm(kvm, type);
508 goto out_err_no_disable;
510 r = hardware_enable_all();
512 goto out_err_no_disable;
514 #ifdef CONFIG_HAVE_KVM_IRQFD
515 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
518 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
521 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
522 kvm->memslots[i] = kvm_alloc_memslots();
523 if (!kvm->memslots[i])
524 goto out_err_no_srcu;
527 if (init_srcu_struct(&kvm->srcu))
528 goto out_err_no_srcu;
529 if (init_srcu_struct(&kvm->irq_srcu))
530 goto out_err_no_irq_srcu;
531 for (i = 0; i < KVM_NR_BUSES; i++) {
532 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
538 spin_lock_init(&kvm->mmu_lock);
539 kvm->mm = current->mm;
540 atomic_inc(&kvm->mm->mm_count);
541 kvm_eventfd_init(kvm);
542 mutex_init(&kvm->lock);
543 mutex_init(&kvm->irq_lock);
544 mutex_init(&kvm->slots_lock);
545 atomic_set(&kvm->users_count, 1);
546 INIT_LIST_HEAD(&kvm->devices);
548 r = kvm_init_mmu_notifier(kvm);
552 spin_lock(&kvm_lock);
553 list_add(&kvm->vm_list, &vm_list);
554 spin_unlock(&kvm_lock);
559 cleanup_srcu_struct(&kvm->irq_srcu);
561 cleanup_srcu_struct(&kvm->srcu);
563 hardware_disable_all();
565 for (i = 0; i < KVM_NR_BUSES; i++)
566 kfree(kvm->buses[i]);
567 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
568 kvm_free_memslots(kvm, kvm->memslots[i]);
569 kvm_arch_free_vm(kvm);
574 * Avoid using vmalloc for a small buffer.
575 * Should not be used when the size is statically known.
577 void *kvm_kvzalloc(unsigned long size)
579 if (size > PAGE_SIZE)
580 return vzalloc(size);
582 return kzalloc(size, GFP_KERNEL);
585 static void kvm_destroy_devices(struct kvm *kvm)
587 struct list_head *node, *tmp;
589 list_for_each_safe(node, tmp, &kvm->devices) {
590 struct kvm_device *dev =
591 list_entry(node, struct kvm_device, vm_node);
594 dev->ops->destroy(dev);
598 static void kvm_destroy_vm(struct kvm *kvm)
601 struct mm_struct *mm = kvm->mm;
603 kvm_arch_sync_events(kvm);
604 spin_lock(&kvm_lock);
605 list_del(&kvm->vm_list);
606 spin_unlock(&kvm_lock);
607 kvm_free_irq_routing(kvm);
608 for (i = 0; i < KVM_NR_BUSES; i++)
609 kvm_io_bus_destroy(kvm->buses[i]);
610 kvm_coalesced_mmio_free(kvm);
611 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
612 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
614 kvm_arch_flush_shadow_all(kvm);
616 kvm_arch_destroy_vm(kvm);
617 kvm_destroy_devices(kvm);
618 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
619 kvm_free_memslots(kvm, kvm->memslots[i]);
620 cleanup_srcu_struct(&kvm->irq_srcu);
621 cleanup_srcu_struct(&kvm->srcu);
622 kvm_arch_free_vm(kvm);
623 hardware_disable_all();
627 void kvm_get_kvm(struct kvm *kvm)
629 atomic_inc(&kvm->users_count);
631 EXPORT_SYMBOL_GPL(kvm_get_kvm);
633 void kvm_put_kvm(struct kvm *kvm)
635 if (atomic_dec_and_test(&kvm->users_count))
638 EXPORT_SYMBOL_GPL(kvm_put_kvm);
641 static int kvm_vm_release(struct inode *inode, struct file *filp)
643 struct kvm *kvm = filp->private_data;
645 kvm_irqfd_release(kvm);
652 * Allocation size is twice as large as the actual dirty bitmap size.
653 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
655 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
657 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
659 memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
660 if (!memslot->dirty_bitmap)
667 * Insert memslot and re-sort memslots based on their GFN,
668 * so binary search could be used to lookup GFN.
669 * Sorting algorithm takes advantage of having initially
670 * sorted array and known changed memslot position.
672 static void update_memslots(struct kvm_memslots *slots,
673 struct kvm_memory_slot *new)
676 int i = slots->id_to_index[id];
677 struct kvm_memory_slot *mslots = slots->memslots;
679 WARN_ON(mslots[i].id != id);
681 WARN_ON(!mslots[i].npages);
682 if (mslots[i].npages)
685 if (!mslots[i].npages)
689 while (i < KVM_MEM_SLOTS_NUM - 1 &&
690 new->base_gfn <= mslots[i + 1].base_gfn) {
691 if (!mslots[i + 1].npages)
693 mslots[i] = mslots[i + 1];
694 slots->id_to_index[mslots[i].id] = i;
699 * The ">=" is needed when creating a slot with base_gfn == 0,
700 * so that it moves before all those with base_gfn == npages == 0.
702 * On the other hand, if new->npages is zero, the above loop has
703 * already left i pointing to the beginning of the empty part of
704 * mslots, and the ">=" would move the hole backwards in this
705 * case---which is wrong. So skip the loop when deleting a slot.
709 new->base_gfn >= mslots[i - 1].base_gfn) {
710 mslots[i] = mslots[i - 1];
711 slots->id_to_index[mslots[i].id] = i;
715 WARN_ON_ONCE(i != slots->used_slots);
718 slots->id_to_index[mslots[i].id] = i;
721 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
723 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
725 #ifdef __KVM_HAVE_READONLY_MEM
726 valid_flags |= KVM_MEM_READONLY;
729 if (mem->flags & ~valid_flags)
735 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
736 int as_id, struct kvm_memslots *slots)
738 struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id);
741 * Set the low bit in the generation, which disables SPTE caching
742 * until the end of synchronize_srcu_expedited.
744 WARN_ON(old_memslots->generation & 1);
745 slots->generation = old_memslots->generation + 1;
747 rcu_assign_pointer(kvm->memslots[as_id], slots);
748 synchronize_srcu_expedited(&kvm->srcu);
751 * Increment the new memslot generation a second time. This prevents
752 * vm exits that race with memslot updates from caching a memslot
753 * generation that will (potentially) be valid forever.
757 kvm_arch_memslots_updated(kvm, slots);
763 * Allocate some memory and give it an address in the guest physical address
766 * Discontiguous memory is allowed, mostly for framebuffers.
768 * Must be called holding kvm->slots_lock for write.
770 int __kvm_set_memory_region(struct kvm *kvm,
771 const struct kvm_userspace_memory_region *mem)
775 unsigned long npages;
776 struct kvm_memory_slot *slot;
777 struct kvm_memory_slot old, new;
778 struct kvm_memslots *slots = NULL, *old_memslots;
780 enum kvm_mr_change change;
782 r = check_memory_region_flags(mem);
787 as_id = mem->slot >> 16;
790 /* General sanity checks */
791 if (mem->memory_size & (PAGE_SIZE - 1))
793 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
795 /* We can read the guest memory with __xxx_user() later on. */
796 if ((id < KVM_USER_MEM_SLOTS) &&
797 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
798 !access_ok(VERIFY_WRITE,
799 (void __user *)(unsigned long)mem->userspace_addr,
802 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
804 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
807 slot = id_to_memslot(__kvm_memslots(kvm, as_id), id);
808 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
809 npages = mem->memory_size >> PAGE_SHIFT;
811 if (npages > KVM_MEM_MAX_NR_PAGES)
817 new.base_gfn = base_gfn;
819 new.flags = mem->flags;
823 change = KVM_MR_CREATE;
824 else { /* Modify an existing slot. */
825 if ((mem->userspace_addr != old.userspace_addr) ||
826 (npages != old.npages) ||
827 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
830 if (base_gfn != old.base_gfn)
831 change = KVM_MR_MOVE;
832 else if (new.flags != old.flags)
833 change = KVM_MR_FLAGS_ONLY;
834 else { /* Nothing to change. */
843 change = KVM_MR_DELETE;
848 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
849 /* Check for overlaps */
851 kvm_for_each_memslot(slot, __kvm_memslots(kvm, as_id)) {
852 if ((slot->id >= KVM_USER_MEM_SLOTS) ||
855 if (!((base_gfn + npages <= slot->base_gfn) ||
856 (base_gfn >= slot->base_gfn + slot->npages)))
861 /* Free page dirty bitmap if unneeded */
862 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
863 new.dirty_bitmap = NULL;
866 if (change == KVM_MR_CREATE) {
867 new.userspace_addr = mem->userspace_addr;
869 if (kvm_arch_create_memslot(kvm, &new, npages))
873 /* Allocate page dirty bitmap if needed */
874 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
875 if (kvm_create_dirty_bitmap(&new) < 0)
879 slots = kvm_kvzalloc(sizeof(struct kvm_memslots));
882 memcpy(slots, __kvm_memslots(kvm, as_id), sizeof(struct kvm_memslots));
884 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
885 slot = id_to_memslot(slots, id);
886 slot->flags |= KVM_MEMSLOT_INVALID;
888 old_memslots = install_new_memslots(kvm, as_id, slots);
890 /* slot was deleted or moved, clear iommu mapping */
891 kvm_iommu_unmap_pages(kvm, &old);
892 /* From this point no new shadow pages pointing to a deleted,
893 * or moved, memslot will be created.
895 * validation of sp->gfn happens in:
896 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
897 * - kvm_is_visible_gfn (mmu_check_roots)
899 kvm_arch_flush_shadow_memslot(kvm, slot);
902 * We can re-use the old_memslots from above, the only difference
903 * from the currently installed memslots is the invalid flag. This
904 * will get overwritten by update_memslots anyway.
906 slots = old_memslots;
909 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
913 /* actual memory is freed via old in kvm_free_memslot below */
914 if (change == KVM_MR_DELETE) {
915 new.dirty_bitmap = NULL;
916 memset(&new.arch, 0, sizeof(new.arch));
919 update_memslots(slots, &new);
920 old_memslots = install_new_memslots(kvm, as_id, slots);
922 kvm_arch_commit_memory_region(kvm, mem, &old, &new, change);
924 kvm_free_memslot(kvm, &old, &new);
925 kvfree(old_memslots);
928 * IOMMU mapping: New slots need to be mapped. Old slots need to be
929 * un-mapped and re-mapped if their base changes. Since base change
930 * unmapping is handled above with slot deletion, mapping alone is
931 * needed here. Anything else the iommu might care about for existing
932 * slots (size changes, userspace addr changes and read-only flag
933 * changes) is disallowed above, so any other attribute changes getting
934 * here can be skipped.
936 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
937 r = kvm_iommu_map_pages(kvm, &new);
946 kvm_free_memslot(kvm, &new, &old);
950 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
952 int kvm_set_memory_region(struct kvm *kvm,
953 const struct kvm_userspace_memory_region *mem)
957 mutex_lock(&kvm->slots_lock);
958 r = __kvm_set_memory_region(kvm, mem);
959 mutex_unlock(&kvm->slots_lock);
962 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
964 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
965 struct kvm_userspace_memory_region *mem)
967 if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
970 return kvm_set_memory_region(kvm, mem);
973 int kvm_get_dirty_log(struct kvm *kvm,
974 struct kvm_dirty_log *log, int *is_dirty)
976 struct kvm_memslots *slots;
977 struct kvm_memory_slot *memslot;
980 unsigned long any = 0;
983 as_id = log->slot >> 16;
985 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
988 slots = __kvm_memslots(kvm, as_id);
989 memslot = id_to_memslot(slots, id);
991 if (!memslot->dirty_bitmap)
994 n = kvm_dirty_bitmap_bytes(memslot);
996 for (i = 0; !any && i < n/sizeof(long); ++i)
997 any = memslot->dirty_bitmap[i];
1000 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1010 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1012 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1014 * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages
1015 * are dirty write protect them for next write.
1016 * @kvm: pointer to kvm instance
1017 * @log: slot id and address to which we copy the log
1018 * @is_dirty: flag set if any page is dirty
1020 * We need to keep it in mind that VCPU threads can write to the bitmap
1021 * concurrently. So, to avoid losing track of dirty pages we keep the
1024 * 1. Take a snapshot of the bit and clear it if needed.
1025 * 2. Write protect the corresponding page.
1026 * 3. Copy the snapshot to the userspace.
1027 * 4. Upon return caller flushes TLB's if needed.
1029 * Between 2 and 4, the guest may write to the page using the remaining TLB
1030 * entry. This is not a problem because the page is reported dirty using
1031 * the snapshot taken before and step 4 ensures that writes done after
1032 * exiting to userspace will be logged for the next call.
1035 int kvm_get_dirty_log_protect(struct kvm *kvm,
1036 struct kvm_dirty_log *log, bool *is_dirty)
1038 struct kvm_memslots *slots;
1039 struct kvm_memory_slot *memslot;
1040 int r, i, as_id, id;
1042 unsigned long *dirty_bitmap;
1043 unsigned long *dirty_bitmap_buffer;
1046 as_id = log->slot >> 16;
1047 id = (u16)log->slot;
1048 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1051 slots = __kvm_memslots(kvm, as_id);
1052 memslot = id_to_memslot(slots, id);
1054 dirty_bitmap = memslot->dirty_bitmap;
1059 n = kvm_dirty_bitmap_bytes(memslot);
1061 dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long);
1062 memset(dirty_bitmap_buffer, 0, n);
1064 spin_lock(&kvm->mmu_lock);
1066 for (i = 0; i < n / sizeof(long); i++) {
1070 if (!dirty_bitmap[i])
1075 mask = xchg(&dirty_bitmap[i], 0);
1076 dirty_bitmap_buffer[i] = mask;
1079 offset = i * BITS_PER_LONG;
1080 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1085 spin_unlock(&kvm->mmu_lock);
1088 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1095 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect);
1098 bool kvm_largepages_enabled(void)
1100 return largepages_enabled;
1103 void kvm_disable_largepages(void)
1105 largepages_enabled = false;
1107 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1109 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1111 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1113 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1115 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
1117 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn);
1120 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1122 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1124 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1125 memslot->flags & KVM_MEMSLOT_INVALID)
1130 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1132 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1134 struct vm_area_struct *vma;
1135 unsigned long addr, size;
1139 addr = gfn_to_hva(kvm, gfn);
1140 if (kvm_is_error_hva(addr))
1143 down_read(¤t->mm->mmap_sem);
1144 vma = find_vma(current->mm, addr);
1148 size = vma_kernel_pagesize(vma);
1151 up_read(¤t->mm->mmap_sem);
1156 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1158 return slot->flags & KVM_MEM_READONLY;
1161 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1162 gfn_t *nr_pages, bool write)
1164 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1165 return KVM_HVA_ERR_BAD;
1167 if (memslot_is_readonly(slot) && write)
1168 return KVM_HVA_ERR_RO_BAD;
1171 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1173 return __gfn_to_hva_memslot(slot, gfn);
1176 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1179 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1182 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1185 return gfn_to_hva_many(slot, gfn, NULL);
1187 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1189 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1191 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1193 EXPORT_SYMBOL_GPL(gfn_to_hva);
1195 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
1197 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
1199 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
1202 * If writable is set to false, the hva returned by this function is only
1203 * allowed to be read.
1205 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1206 gfn_t gfn, bool *writable)
1208 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1210 if (!kvm_is_error_hva(hva) && writable)
1211 *writable = !memslot_is_readonly(slot);
1216 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1218 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1220 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1223 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
1225 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1227 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1230 static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1231 unsigned long start, int write, struct page **page)
1233 int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1236 flags |= FOLL_WRITE;
1238 return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1241 static inline int check_user_page_hwpoison(unsigned long addr)
1243 int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1245 rc = __get_user_pages(current, current->mm, addr, 1,
1246 flags, NULL, NULL, NULL);
1247 return rc == -EHWPOISON;
1251 * The atomic path to get the writable pfn which will be stored in @pfn,
1252 * true indicates success, otherwise false is returned.
1254 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1255 bool write_fault, bool *writable, pfn_t *pfn)
1257 struct page *page[1];
1260 if (!(async || atomic))
1264 * Fast pin a writable pfn only if it is a write fault request
1265 * or the caller allows to map a writable pfn for a read fault
1268 if (!(write_fault || writable))
1271 npages = __get_user_pages_fast(addr, 1, 1, page);
1273 *pfn = page_to_pfn(page[0]);
1284 * The slow path to get the pfn of the specified host virtual address,
1285 * 1 indicates success, -errno is returned if error is detected.
1287 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1288 bool *writable, pfn_t *pfn)
1290 struct page *page[1];
1296 *writable = write_fault;
1299 down_read(¤t->mm->mmap_sem);
1300 npages = get_user_page_nowait(current, current->mm,
1301 addr, write_fault, page);
1302 up_read(¤t->mm->mmap_sem);
1304 npages = __get_user_pages_unlocked(current, current->mm, addr, 1,
1305 write_fault, 0, page,
1306 FOLL_TOUCH|FOLL_HWPOISON);
1310 /* map read fault as writable if possible */
1311 if (unlikely(!write_fault) && writable) {
1312 struct page *wpage[1];
1314 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1323 *pfn = page_to_pfn(page[0]);
1327 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1329 if (unlikely(!(vma->vm_flags & VM_READ)))
1332 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1339 * Pin guest page in memory and return its pfn.
1340 * @addr: host virtual address which maps memory to the guest
1341 * @atomic: whether this function can sleep
1342 * @async: whether this function need to wait IO complete if the
1343 * host page is not in the memory
1344 * @write_fault: whether we should get a writable host page
1345 * @writable: whether it allows to map a writable host page for !@write_fault
1347 * The function will map a writable host page for these two cases:
1348 * 1): @write_fault = true
1349 * 2): @write_fault = false && @writable, @writable will tell the caller
1350 * whether the mapping is writable.
1352 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1353 bool write_fault, bool *writable)
1355 struct vm_area_struct *vma;
1359 /* we can do it either atomically or asynchronously, not both */
1360 BUG_ON(atomic && async);
1362 if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1366 return KVM_PFN_ERR_FAULT;
1368 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1372 down_read(¤t->mm->mmap_sem);
1373 if (npages == -EHWPOISON ||
1374 (!async && check_user_page_hwpoison(addr))) {
1375 pfn = KVM_PFN_ERR_HWPOISON;
1379 vma = find_vma_intersection(current->mm, addr, addr + 1);
1382 pfn = KVM_PFN_ERR_FAULT;
1383 else if ((vma->vm_flags & VM_PFNMAP)) {
1384 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1386 BUG_ON(!kvm_is_reserved_pfn(pfn));
1388 if (async && vma_is_valid(vma, write_fault))
1390 pfn = KVM_PFN_ERR_FAULT;
1393 up_read(¤t->mm->mmap_sem);
1397 pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1398 bool *async, bool write_fault, bool *writable)
1400 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1402 if (addr == KVM_HVA_ERR_RO_BAD)
1403 return KVM_PFN_ERR_RO_FAULT;
1405 if (kvm_is_error_hva(addr))
1406 return KVM_PFN_NOSLOT;
1408 /* Do not map writable pfn in the readonly memslot. */
1409 if (writable && memslot_is_readonly(slot)) {
1414 return hva_to_pfn(addr, atomic, async, write_fault,
1417 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
1419 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1422 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
1423 write_fault, writable);
1425 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1427 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1429 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1431 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
1433 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1435 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1437 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1439 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1441 return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm, gfn), gfn);
1443 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1445 pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
1447 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1449 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
1451 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1453 return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
1455 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1457 pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
1459 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1461 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
1463 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1464 struct page **pages, int nr_pages)
1469 addr = gfn_to_hva_many(slot, gfn, &entry);
1470 if (kvm_is_error_hva(addr))
1473 if (entry < nr_pages)
1476 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1478 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1480 static struct page *kvm_pfn_to_page(pfn_t pfn)
1482 if (is_error_noslot_pfn(pfn))
1483 return KVM_ERR_PTR_BAD_PAGE;
1485 if (kvm_is_reserved_pfn(pfn)) {
1487 return KVM_ERR_PTR_BAD_PAGE;
1490 return pfn_to_page(pfn);
1493 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1497 pfn = gfn_to_pfn(kvm, gfn);
1499 return kvm_pfn_to_page(pfn);
1501 EXPORT_SYMBOL_GPL(gfn_to_page);
1503 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
1507 pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
1509 return kvm_pfn_to_page(pfn);
1511 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
1513 void kvm_release_page_clean(struct page *page)
1515 WARN_ON(is_error_page(page));
1517 kvm_release_pfn_clean(page_to_pfn(page));
1519 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1521 void kvm_release_pfn_clean(pfn_t pfn)
1523 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
1524 put_page(pfn_to_page(pfn));
1526 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1528 void kvm_release_page_dirty(struct page *page)
1530 WARN_ON(is_error_page(page));
1532 kvm_release_pfn_dirty(page_to_pfn(page));
1534 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1536 static void kvm_release_pfn_dirty(pfn_t pfn)
1538 kvm_set_pfn_dirty(pfn);
1539 kvm_release_pfn_clean(pfn);
1542 void kvm_set_pfn_dirty(pfn_t pfn)
1544 if (!kvm_is_reserved_pfn(pfn)) {
1545 struct page *page = pfn_to_page(pfn);
1547 if (!PageReserved(page))
1551 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1553 void kvm_set_pfn_accessed(pfn_t pfn)
1555 if (!kvm_is_reserved_pfn(pfn))
1556 mark_page_accessed(pfn_to_page(pfn));
1558 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1560 void kvm_get_pfn(pfn_t pfn)
1562 if (!kvm_is_reserved_pfn(pfn))
1563 get_page(pfn_to_page(pfn));
1565 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1567 static int next_segment(unsigned long len, int offset)
1569 if (len > PAGE_SIZE - offset)
1570 return PAGE_SIZE - offset;
1575 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
1576 void *data, int offset, int len)
1581 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1582 if (kvm_is_error_hva(addr))
1584 r = __copy_from_user(data, (void __user *)addr + offset, len);
1590 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1593 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1595 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1597 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1599 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
1600 int offset, int len)
1602 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1604 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1606 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
1608 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1610 gfn_t gfn = gpa >> PAGE_SHIFT;
1612 int offset = offset_in_page(gpa);
1615 while ((seg = next_segment(len, offset)) != 0) {
1616 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1626 EXPORT_SYMBOL_GPL(kvm_read_guest);
1628 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
1630 gfn_t gfn = gpa >> PAGE_SHIFT;
1632 int offset = offset_in_page(gpa);
1635 while ((seg = next_segment(len, offset)) != 0) {
1636 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
1646 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
1648 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1649 void *data, int offset, unsigned long len)
1654 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1655 if (kvm_is_error_hva(addr))
1657 pagefault_disable();
1658 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1665 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1668 gfn_t gfn = gpa >> PAGE_SHIFT;
1669 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1670 int offset = offset_in_page(gpa);
1672 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1674 EXPORT_SYMBOL_GPL(kvm_read_guest_atomic);
1676 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
1677 void *data, unsigned long len)
1679 gfn_t gfn = gpa >> PAGE_SHIFT;
1680 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1681 int offset = offset_in_page(gpa);
1683 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1685 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
1687 static int __kvm_write_guest_page(struct kvm_memory_slot *memslot, gfn_t gfn,
1688 const void *data, int offset, int len)
1693 addr = gfn_to_hva_memslot(memslot, gfn);
1694 if (kvm_is_error_hva(addr))
1696 r = __copy_to_user((void __user *)addr + offset, data, len);
1699 mark_page_dirty_in_slot(memslot, gfn);
1703 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
1704 const void *data, int offset, int len)
1706 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1708 return __kvm_write_guest_page(slot, gfn, data, offset, len);
1710 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1712 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
1713 const void *data, int offset, int len)
1715 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1717 return __kvm_write_guest_page(slot, gfn, data, offset, len);
1719 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
1721 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1724 gfn_t gfn = gpa >> PAGE_SHIFT;
1726 int offset = offset_in_page(gpa);
1729 while ((seg = next_segment(len, offset)) != 0) {
1730 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1740 EXPORT_SYMBOL_GPL(kvm_write_guest);
1742 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
1745 gfn_t gfn = gpa >> PAGE_SHIFT;
1747 int offset = offset_in_page(gpa);
1750 while ((seg = next_segment(len, offset)) != 0) {
1751 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
1761 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
1763 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1764 gpa_t gpa, unsigned long len)
1766 struct kvm_memslots *slots = kvm_memslots(kvm);
1767 int offset = offset_in_page(gpa);
1768 gfn_t start_gfn = gpa >> PAGE_SHIFT;
1769 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1770 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1771 gfn_t nr_pages_avail;
1774 ghc->generation = slots->generation;
1776 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1777 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, NULL);
1778 if (!kvm_is_error_hva(ghc->hva) && nr_pages_needed <= 1) {
1782 * If the requested region crosses two memslots, we still
1783 * verify that the entire region is valid here.
1785 while (start_gfn <= end_gfn) {
1786 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1787 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1789 if (kvm_is_error_hva(ghc->hva))
1791 start_gfn += nr_pages_avail;
1793 /* Use the slow path for cross page reads and writes. */
1794 ghc->memslot = NULL;
1798 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1800 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1801 void *data, unsigned long len)
1803 struct kvm_memslots *slots = kvm_memslots(kvm);
1806 BUG_ON(len > ghc->len);
1808 if (slots->generation != ghc->generation)
1809 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1811 if (unlikely(!ghc->memslot))
1812 return kvm_write_guest(kvm, ghc->gpa, data, len);
1814 if (kvm_is_error_hva(ghc->hva))
1817 r = __copy_to_user((void __user *)ghc->hva, data, len);
1820 mark_page_dirty_in_slot(ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1824 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1826 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1827 void *data, unsigned long len)
1829 struct kvm_memslots *slots = kvm_memslots(kvm);
1832 BUG_ON(len > ghc->len);
1834 if (slots->generation != ghc->generation)
1835 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1837 if (unlikely(!ghc->memslot))
1838 return kvm_read_guest(kvm, ghc->gpa, data, len);
1840 if (kvm_is_error_hva(ghc->hva))
1843 r = __copy_from_user(data, (void __user *)ghc->hva, len);
1849 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1851 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1853 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
1855 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
1857 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1859 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1861 gfn_t gfn = gpa >> PAGE_SHIFT;
1863 int offset = offset_in_page(gpa);
1866 while ((seg = next_segment(len, offset)) != 0) {
1867 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1876 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1878 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot,
1881 if (memslot && memslot->dirty_bitmap) {
1882 unsigned long rel_gfn = gfn - memslot->base_gfn;
1884 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1888 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1890 struct kvm_memory_slot *memslot;
1892 memslot = gfn_to_memslot(kvm, gfn);
1893 mark_page_dirty_in_slot(memslot, gfn);
1895 EXPORT_SYMBOL_GPL(mark_page_dirty);
1897 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
1899 struct kvm_memory_slot *memslot;
1901 memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1902 mark_page_dirty_in_slot(memslot, gfn);
1904 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
1906 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
1908 if (kvm_arch_vcpu_runnable(vcpu)) {
1909 kvm_make_request(KVM_REQ_UNHALT, vcpu);
1912 if (kvm_cpu_has_pending_timer(vcpu))
1914 if (signal_pending(current))
1921 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1923 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1927 bool waited = false;
1929 start = cur = ktime_get();
1931 ktime_t stop = ktime_add_ns(ktime_get(), halt_poll_ns);
1935 * This sets KVM_REQ_UNHALT if an interrupt
1938 if (kvm_vcpu_check_block(vcpu) < 0) {
1939 ++vcpu->stat.halt_successful_poll;
1943 } while (single_task_running() && ktime_before(cur, stop));
1947 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1949 if (kvm_vcpu_check_block(vcpu) < 0)
1956 finish_wait(&vcpu->wq, &wait);
1960 trace_kvm_vcpu_wakeup(ktime_to_ns(cur) - ktime_to_ns(start), waited);
1962 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
1966 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1968 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1971 int cpu = vcpu->cpu;
1972 wait_queue_head_t *wqp;
1974 wqp = kvm_arch_vcpu_wq(vcpu);
1975 if (waitqueue_active(wqp)) {
1976 wake_up_interruptible(wqp);
1977 ++vcpu->stat.halt_wakeup;
1981 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1982 if (kvm_arch_vcpu_should_kick(vcpu))
1983 smp_send_reschedule(cpu);
1986 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
1987 #endif /* !CONFIG_S390 */
1989 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
1992 struct task_struct *task = NULL;
1996 pid = rcu_dereference(target->pid);
1998 task = get_pid_task(pid, PIDTYPE_PID);
2002 ret = yield_to(task, 1);
2003 put_task_struct(task);
2007 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
2010 * Helper that checks whether a VCPU is eligible for directed yield.
2011 * Most eligible candidate to yield is decided by following heuristics:
2013 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2014 * (preempted lock holder), indicated by @in_spin_loop.
2015 * Set at the beiginning and cleared at the end of interception/PLE handler.
2017 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2018 * chance last time (mostly it has become eligible now since we have probably
2019 * yielded to lockholder in last iteration. This is done by toggling
2020 * @dy_eligible each time a VCPU checked for eligibility.)
2022 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2023 * to preempted lock-holder could result in wrong VCPU selection and CPU
2024 * burning. Giving priority for a potential lock-holder increases lock
2027 * Since algorithm is based on heuristics, accessing another VCPU data without
2028 * locking does not harm. It may result in trying to yield to same VCPU, fail
2029 * and continue with next VCPU and so on.
2031 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
2033 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2036 eligible = !vcpu->spin_loop.in_spin_loop ||
2037 vcpu->spin_loop.dy_eligible;
2039 if (vcpu->spin_loop.in_spin_loop)
2040 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
2048 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
2050 struct kvm *kvm = me->kvm;
2051 struct kvm_vcpu *vcpu;
2052 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
2058 kvm_vcpu_set_in_spin_loop(me, true);
2060 * We boost the priority of a VCPU that is runnable but not
2061 * currently running, because it got preempted by something
2062 * else and called schedule in __vcpu_run. Hopefully that
2063 * VCPU is holding the lock that we need and will release it.
2064 * We approximate round-robin by starting at the last boosted VCPU.
2066 for (pass = 0; pass < 2 && !yielded && try; pass++) {
2067 kvm_for_each_vcpu(i, vcpu, kvm) {
2068 if (!pass && i <= last_boosted_vcpu) {
2069 i = last_boosted_vcpu;
2071 } else if (pass && i > last_boosted_vcpu)
2073 if (!ACCESS_ONCE(vcpu->preempted))
2077 if (waitqueue_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
2079 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
2082 yielded = kvm_vcpu_yield_to(vcpu);
2084 kvm->last_boosted_vcpu = i;
2086 } else if (yielded < 0) {
2093 kvm_vcpu_set_in_spin_loop(me, false);
2095 /* Ensure vcpu is not eligible during next spinloop */
2096 kvm_vcpu_set_dy_eligible(me, false);
2098 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
2100 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2102 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
2105 if (vmf->pgoff == 0)
2106 page = virt_to_page(vcpu->run);
2108 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
2109 page = virt_to_page(vcpu->arch.pio_data);
2111 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2112 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
2113 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
2116 return kvm_arch_vcpu_fault(vcpu, vmf);
2122 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
2123 .fault = kvm_vcpu_fault,
2126 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2128 vma->vm_ops = &kvm_vcpu_vm_ops;
2132 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2134 struct kvm_vcpu *vcpu = filp->private_data;
2136 kvm_put_kvm(vcpu->kvm);
2140 static struct file_operations kvm_vcpu_fops = {
2141 .release = kvm_vcpu_release,
2142 .unlocked_ioctl = kvm_vcpu_ioctl,
2143 #ifdef CONFIG_KVM_COMPAT
2144 .compat_ioctl = kvm_vcpu_compat_ioctl,
2146 .mmap = kvm_vcpu_mmap,
2147 .llseek = noop_llseek,
2151 * Allocates an inode for the vcpu.
2153 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2155 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
2159 * Creates some virtual cpus. Good luck creating more than one.
2161 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
2164 struct kvm_vcpu *vcpu, *v;
2166 if (id >= KVM_MAX_VCPUS)
2169 vcpu = kvm_arch_vcpu_create(kvm, id);
2171 return PTR_ERR(vcpu);
2173 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2175 r = kvm_arch_vcpu_setup(vcpu);
2179 mutex_lock(&kvm->lock);
2180 if (!kvm_vcpu_compatible(vcpu)) {
2182 goto unlock_vcpu_destroy;
2184 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
2186 goto unlock_vcpu_destroy;
2189 kvm_for_each_vcpu(r, v, kvm)
2190 if (v->vcpu_id == id) {
2192 goto unlock_vcpu_destroy;
2195 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
2197 /* Now it's all set up, let userspace reach it */
2199 r = create_vcpu_fd(vcpu);
2202 goto unlock_vcpu_destroy;
2205 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
2207 atomic_inc(&kvm->online_vcpus);
2209 mutex_unlock(&kvm->lock);
2210 kvm_arch_vcpu_postcreate(vcpu);
2213 unlock_vcpu_destroy:
2214 mutex_unlock(&kvm->lock);
2216 kvm_arch_vcpu_destroy(vcpu);
2220 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2223 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2224 vcpu->sigset_active = 1;
2225 vcpu->sigset = *sigset;
2227 vcpu->sigset_active = 0;
2231 static long kvm_vcpu_ioctl(struct file *filp,
2232 unsigned int ioctl, unsigned long arg)
2234 struct kvm_vcpu *vcpu = filp->private_data;
2235 void __user *argp = (void __user *)arg;
2237 struct kvm_fpu *fpu = NULL;
2238 struct kvm_sregs *kvm_sregs = NULL;
2240 if (vcpu->kvm->mm != current->mm)
2243 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2246 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
2248 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
2249 * so vcpu_load() would break it.
2251 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_S390_IRQ || ioctl == KVM_INTERRUPT)
2252 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2256 r = vcpu_load(vcpu);
2264 if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
2265 /* The thread running this VCPU changed. */
2266 struct pid *oldpid = vcpu->pid;
2267 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
2269 rcu_assign_pointer(vcpu->pid, newpid);
2274 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2275 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2277 case KVM_GET_REGS: {
2278 struct kvm_regs *kvm_regs;
2281 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2284 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2288 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2295 case KVM_SET_REGS: {
2296 struct kvm_regs *kvm_regs;
2299 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2300 if (IS_ERR(kvm_regs)) {
2301 r = PTR_ERR(kvm_regs);
2304 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2308 case KVM_GET_SREGS: {
2309 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2313 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2317 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2322 case KVM_SET_SREGS: {
2323 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2324 if (IS_ERR(kvm_sregs)) {
2325 r = PTR_ERR(kvm_sregs);
2329 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2332 case KVM_GET_MP_STATE: {
2333 struct kvm_mp_state mp_state;
2335 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2339 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
2344 case KVM_SET_MP_STATE: {
2345 struct kvm_mp_state mp_state;
2348 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
2350 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2353 case KVM_TRANSLATE: {
2354 struct kvm_translation tr;
2357 if (copy_from_user(&tr, argp, sizeof(tr)))
2359 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2363 if (copy_to_user(argp, &tr, sizeof(tr)))
2368 case KVM_SET_GUEST_DEBUG: {
2369 struct kvm_guest_debug dbg;
2372 if (copy_from_user(&dbg, argp, sizeof(dbg)))
2374 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2377 case KVM_SET_SIGNAL_MASK: {
2378 struct kvm_signal_mask __user *sigmask_arg = argp;
2379 struct kvm_signal_mask kvm_sigmask;
2380 sigset_t sigset, *p;
2385 if (copy_from_user(&kvm_sigmask, argp,
2386 sizeof(kvm_sigmask)))
2389 if (kvm_sigmask.len != sizeof(sigset))
2392 if (copy_from_user(&sigset, sigmask_arg->sigset,
2397 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2401 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2405 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2409 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2415 fpu = memdup_user(argp, sizeof(*fpu));
2421 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2425 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2434 #ifdef CONFIG_KVM_COMPAT
2435 static long kvm_vcpu_compat_ioctl(struct file *filp,
2436 unsigned int ioctl, unsigned long arg)
2438 struct kvm_vcpu *vcpu = filp->private_data;
2439 void __user *argp = compat_ptr(arg);
2442 if (vcpu->kvm->mm != current->mm)
2446 case KVM_SET_SIGNAL_MASK: {
2447 struct kvm_signal_mask __user *sigmask_arg = argp;
2448 struct kvm_signal_mask kvm_sigmask;
2449 compat_sigset_t csigset;
2454 if (copy_from_user(&kvm_sigmask, argp,
2455 sizeof(kvm_sigmask)))
2458 if (kvm_sigmask.len != sizeof(csigset))
2461 if (copy_from_user(&csigset, sigmask_arg->sigset,
2464 sigset_from_compat(&sigset, &csigset);
2465 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2467 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2471 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2479 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2480 int (*accessor)(struct kvm_device *dev,
2481 struct kvm_device_attr *attr),
2484 struct kvm_device_attr attr;
2489 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2492 return accessor(dev, &attr);
2495 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2498 struct kvm_device *dev = filp->private_data;
2501 case KVM_SET_DEVICE_ATTR:
2502 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2503 case KVM_GET_DEVICE_ATTR:
2504 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2505 case KVM_HAS_DEVICE_ATTR:
2506 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2508 if (dev->ops->ioctl)
2509 return dev->ops->ioctl(dev, ioctl, arg);
2515 static int kvm_device_release(struct inode *inode, struct file *filp)
2517 struct kvm_device *dev = filp->private_data;
2518 struct kvm *kvm = dev->kvm;
2524 static const struct file_operations kvm_device_fops = {
2525 .unlocked_ioctl = kvm_device_ioctl,
2526 #ifdef CONFIG_KVM_COMPAT
2527 .compat_ioctl = kvm_device_ioctl,
2529 .release = kvm_device_release,
2532 struct kvm_device *kvm_device_from_filp(struct file *filp)
2534 if (filp->f_op != &kvm_device_fops)
2537 return filp->private_data;
2540 static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
2541 #ifdef CONFIG_KVM_MPIC
2542 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
2543 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
2546 #ifdef CONFIG_KVM_XICS
2547 [KVM_DEV_TYPE_XICS] = &kvm_xics_ops,
2551 int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
2553 if (type >= ARRAY_SIZE(kvm_device_ops_table))
2556 if (kvm_device_ops_table[type] != NULL)
2559 kvm_device_ops_table[type] = ops;
2563 void kvm_unregister_device_ops(u32 type)
2565 if (kvm_device_ops_table[type] != NULL)
2566 kvm_device_ops_table[type] = NULL;
2569 static int kvm_ioctl_create_device(struct kvm *kvm,
2570 struct kvm_create_device *cd)
2572 struct kvm_device_ops *ops = NULL;
2573 struct kvm_device *dev;
2574 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2577 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
2580 ops = kvm_device_ops_table[cd->type];
2587 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2594 ret = ops->create(dev, cd->type);
2600 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2606 list_add(&dev->vm_node, &kvm->devices);
2612 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2615 case KVM_CAP_USER_MEMORY:
2616 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2617 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2618 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2619 case KVM_CAP_SET_BOOT_CPU_ID:
2621 case KVM_CAP_INTERNAL_ERROR_DATA:
2622 #ifdef CONFIG_HAVE_KVM_MSI
2623 case KVM_CAP_SIGNAL_MSI:
2625 #ifdef CONFIG_HAVE_KVM_IRQFD
2627 case KVM_CAP_IRQFD_RESAMPLE:
2629 case KVM_CAP_CHECK_EXTENSION_VM:
2631 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2632 case KVM_CAP_IRQ_ROUTING:
2633 return KVM_MAX_IRQ_ROUTES;
2635 #if KVM_ADDRESS_SPACE_NUM > 1
2636 case KVM_CAP_MULTI_ADDRESS_SPACE:
2637 return KVM_ADDRESS_SPACE_NUM;
2642 return kvm_vm_ioctl_check_extension(kvm, arg);
2645 static long kvm_vm_ioctl(struct file *filp,
2646 unsigned int ioctl, unsigned long arg)
2648 struct kvm *kvm = filp->private_data;
2649 void __user *argp = (void __user *)arg;
2652 if (kvm->mm != current->mm)
2655 case KVM_CREATE_VCPU:
2656 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2658 case KVM_SET_USER_MEMORY_REGION: {
2659 struct kvm_userspace_memory_region kvm_userspace_mem;
2662 if (copy_from_user(&kvm_userspace_mem, argp,
2663 sizeof(kvm_userspace_mem)))
2666 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2669 case KVM_GET_DIRTY_LOG: {
2670 struct kvm_dirty_log log;
2673 if (copy_from_user(&log, argp, sizeof(log)))
2675 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2678 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2679 case KVM_REGISTER_COALESCED_MMIO: {
2680 struct kvm_coalesced_mmio_zone zone;
2683 if (copy_from_user(&zone, argp, sizeof(zone)))
2685 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2688 case KVM_UNREGISTER_COALESCED_MMIO: {
2689 struct kvm_coalesced_mmio_zone zone;
2692 if (copy_from_user(&zone, argp, sizeof(zone)))
2694 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2699 struct kvm_irqfd data;
2702 if (copy_from_user(&data, argp, sizeof(data)))
2704 r = kvm_irqfd(kvm, &data);
2707 case KVM_IOEVENTFD: {
2708 struct kvm_ioeventfd data;
2711 if (copy_from_user(&data, argp, sizeof(data)))
2713 r = kvm_ioeventfd(kvm, &data);
2716 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2717 case KVM_SET_BOOT_CPU_ID:
2719 mutex_lock(&kvm->lock);
2720 if (atomic_read(&kvm->online_vcpus) != 0)
2723 kvm->bsp_vcpu_id = arg;
2724 mutex_unlock(&kvm->lock);
2727 #ifdef CONFIG_HAVE_KVM_MSI
2728 case KVM_SIGNAL_MSI: {
2732 if (copy_from_user(&msi, argp, sizeof(msi)))
2734 r = kvm_send_userspace_msi(kvm, &msi);
2738 #ifdef __KVM_HAVE_IRQ_LINE
2739 case KVM_IRQ_LINE_STATUS:
2740 case KVM_IRQ_LINE: {
2741 struct kvm_irq_level irq_event;
2744 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
2747 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
2748 ioctl == KVM_IRQ_LINE_STATUS);
2753 if (ioctl == KVM_IRQ_LINE_STATUS) {
2754 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
2762 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2763 case KVM_SET_GSI_ROUTING: {
2764 struct kvm_irq_routing routing;
2765 struct kvm_irq_routing __user *urouting;
2766 struct kvm_irq_routing_entry *entries;
2769 if (copy_from_user(&routing, argp, sizeof(routing)))
2772 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2777 entries = vmalloc(routing.nr * sizeof(*entries));
2782 if (copy_from_user(entries, urouting->entries,
2783 routing.nr * sizeof(*entries)))
2784 goto out_free_irq_routing;
2785 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2787 out_free_irq_routing:
2791 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
2792 case KVM_CREATE_DEVICE: {
2793 struct kvm_create_device cd;
2796 if (copy_from_user(&cd, argp, sizeof(cd)))
2799 r = kvm_ioctl_create_device(kvm, &cd);
2804 if (copy_to_user(argp, &cd, sizeof(cd)))
2810 case KVM_CHECK_EXTENSION:
2811 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
2814 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2820 #ifdef CONFIG_KVM_COMPAT
2821 struct compat_kvm_dirty_log {
2825 compat_uptr_t dirty_bitmap; /* one bit per page */
2830 static long kvm_vm_compat_ioctl(struct file *filp,
2831 unsigned int ioctl, unsigned long arg)
2833 struct kvm *kvm = filp->private_data;
2836 if (kvm->mm != current->mm)
2839 case KVM_GET_DIRTY_LOG: {
2840 struct compat_kvm_dirty_log compat_log;
2841 struct kvm_dirty_log log;
2844 if (copy_from_user(&compat_log, (void __user *)arg,
2845 sizeof(compat_log)))
2847 log.slot = compat_log.slot;
2848 log.padding1 = compat_log.padding1;
2849 log.padding2 = compat_log.padding2;
2850 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2852 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2856 r = kvm_vm_ioctl(filp, ioctl, arg);
2864 static struct file_operations kvm_vm_fops = {
2865 .release = kvm_vm_release,
2866 .unlocked_ioctl = kvm_vm_ioctl,
2867 #ifdef CONFIG_KVM_COMPAT
2868 .compat_ioctl = kvm_vm_compat_ioctl,
2870 .llseek = noop_llseek,
2873 static int kvm_dev_ioctl_create_vm(unsigned long type)
2878 kvm = kvm_create_vm(type);
2880 return PTR_ERR(kvm);
2881 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2882 r = kvm_coalesced_mmio_init(kvm);
2888 r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC);
2895 static long kvm_dev_ioctl(struct file *filp,
2896 unsigned int ioctl, unsigned long arg)
2901 case KVM_GET_API_VERSION:
2904 r = KVM_API_VERSION;
2907 r = kvm_dev_ioctl_create_vm(arg);
2909 case KVM_CHECK_EXTENSION:
2910 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
2912 case KVM_GET_VCPU_MMAP_SIZE:
2915 r = PAGE_SIZE; /* struct kvm_run */
2917 r += PAGE_SIZE; /* pio data page */
2919 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2920 r += PAGE_SIZE; /* coalesced mmio ring page */
2923 case KVM_TRACE_ENABLE:
2924 case KVM_TRACE_PAUSE:
2925 case KVM_TRACE_DISABLE:
2929 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2935 static struct file_operations kvm_chardev_ops = {
2936 .unlocked_ioctl = kvm_dev_ioctl,
2937 .compat_ioctl = kvm_dev_ioctl,
2938 .llseek = noop_llseek,
2941 static struct miscdevice kvm_dev = {
2947 static void hardware_enable_nolock(void *junk)
2949 int cpu = raw_smp_processor_id();
2952 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2955 cpumask_set_cpu(cpu, cpus_hardware_enabled);
2957 r = kvm_arch_hardware_enable();
2960 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2961 atomic_inc(&hardware_enable_failed);
2962 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
2966 static void hardware_enable(void)
2968 raw_spin_lock(&kvm_count_lock);
2969 if (kvm_usage_count)
2970 hardware_enable_nolock(NULL);
2971 raw_spin_unlock(&kvm_count_lock);
2974 static void hardware_disable_nolock(void *junk)
2976 int cpu = raw_smp_processor_id();
2978 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2980 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2981 kvm_arch_hardware_disable();
2984 static void hardware_disable(void)
2986 raw_spin_lock(&kvm_count_lock);
2987 if (kvm_usage_count)
2988 hardware_disable_nolock(NULL);
2989 raw_spin_unlock(&kvm_count_lock);
2992 static void hardware_disable_all_nolock(void)
2994 BUG_ON(!kvm_usage_count);
2997 if (!kvm_usage_count)
2998 on_each_cpu(hardware_disable_nolock, NULL, 1);
3001 static void hardware_disable_all(void)
3003 raw_spin_lock(&kvm_count_lock);
3004 hardware_disable_all_nolock();
3005 raw_spin_unlock(&kvm_count_lock);
3008 static int hardware_enable_all(void)
3012 raw_spin_lock(&kvm_count_lock);
3015 if (kvm_usage_count == 1) {
3016 atomic_set(&hardware_enable_failed, 0);
3017 on_each_cpu(hardware_enable_nolock, NULL, 1);
3019 if (atomic_read(&hardware_enable_failed)) {
3020 hardware_disable_all_nolock();
3025 raw_spin_unlock(&kvm_count_lock);
3030 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
3033 val &= ~CPU_TASKS_FROZEN;
3045 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3049 * Some (well, at least mine) BIOSes hang on reboot if
3052 * And Intel TXT required VMX off for all cpu when system shutdown.
3054 pr_info("kvm: exiting hardware virtualization\n");
3055 kvm_rebooting = true;
3056 on_each_cpu(hardware_disable_nolock, NULL, 1);
3060 static struct notifier_block kvm_reboot_notifier = {
3061 .notifier_call = kvm_reboot,
3065 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3069 for (i = 0; i < bus->dev_count; i++) {
3070 struct kvm_io_device *pos = bus->range[i].dev;
3072 kvm_iodevice_destructor(pos);
3077 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
3078 const struct kvm_io_range *r2)
3080 if (r1->addr < r2->addr)
3082 if (r1->addr + r1->len > r2->addr + r2->len)
3087 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
3089 return kvm_io_bus_cmp(p1, p2);
3092 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
3093 gpa_t addr, int len)
3095 bus->range[bus->dev_count++] = (struct kvm_io_range) {
3101 sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
3102 kvm_io_bus_sort_cmp, NULL);
3107 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
3108 gpa_t addr, int len)
3110 struct kvm_io_range *range, key;
3113 key = (struct kvm_io_range) {
3118 range = bsearch(&key, bus->range, bus->dev_count,
3119 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
3123 off = range - bus->range;
3125 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
3131 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3132 struct kvm_io_range *range, const void *val)
3136 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3140 while (idx < bus->dev_count &&
3141 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3142 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
3151 /* kvm_io_bus_write - called under kvm->slots_lock */
3152 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3153 int len, const void *val)
3155 struct kvm_io_bus *bus;
3156 struct kvm_io_range range;
3159 range = (struct kvm_io_range) {
3164 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3165 r = __kvm_io_bus_write(vcpu, bus, &range, val);
3166 return r < 0 ? r : 0;
3169 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3170 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
3171 gpa_t addr, int len, const void *val, long cookie)
3173 struct kvm_io_bus *bus;
3174 struct kvm_io_range range;
3176 range = (struct kvm_io_range) {
3181 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3183 /* First try the device referenced by cookie. */
3184 if ((cookie >= 0) && (cookie < bus->dev_count) &&
3185 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
3186 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
3191 * cookie contained garbage; fall back to search and return the
3192 * correct cookie value.
3194 return __kvm_io_bus_write(vcpu, bus, &range, val);
3197 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3198 struct kvm_io_range *range, void *val)
3202 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3206 while (idx < bus->dev_count &&
3207 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3208 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
3216 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
3218 /* kvm_io_bus_read - called under kvm->slots_lock */
3219 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3222 struct kvm_io_bus *bus;
3223 struct kvm_io_range range;
3226 range = (struct kvm_io_range) {
3231 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3232 r = __kvm_io_bus_read(vcpu, bus, &range, val);
3233 return r < 0 ? r : 0;
3237 /* Caller must hold slots_lock. */
3238 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3239 int len, struct kvm_io_device *dev)
3241 struct kvm_io_bus *new_bus, *bus;
3243 bus = kvm->buses[bus_idx];
3244 /* exclude ioeventfd which is limited by maximum fd */
3245 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3248 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
3249 sizeof(struct kvm_io_range)), GFP_KERNEL);
3252 memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
3253 sizeof(struct kvm_io_range)));
3254 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
3255 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3256 synchronize_srcu_expedited(&kvm->srcu);
3262 /* Caller must hold slots_lock. */
3263 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3264 struct kvm_io_device *dev)
3267 struct kvm_io_bus *new_bus, *bus;
3269 bus = kvm->buses[bus_idx];
3271 for (i = 0; i < bus->dev_count; i++)
3272 if (bus->range[i].dev == dev) {
3280 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3281 sizeof(struct kvm_io_range)), GFP_KERNEL);
3285 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3286 new_bus->dev_count--;
3287 memcpy(new_bus->range + i, bus->range + i + 1,
3288 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3290 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3291 synchronize_srcu_expedited(&kvm->srcu);
3296 static struct notifier_block kvm_cpu_notifier = {
3297 .notifier_call = kvm_cpu_hotplug,
3300 static int vm_stat_get(void *_offset, u64 *val)
3302 unsigned offset = (long)_offset;
3306 spin_lock(&kvm_lock);
3307 list_for_each_entry(kvm, &vm_list, vm_list)
3308 *val += *(u32 *)((void *)kvm + offset);
3309 spin_unlock(&kvm_lock);
3313 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
3315 static int vcpu_stat_get(void *_offset, u64 *val)
3317 unsigned offset = (long)_offset;
3319 struct kvm_vcpu *vcpu;
3323 spin_lock(&kvm_lock);
3324 list_for_each_entry(kvm, &vm_list, vm_list)
3325 kvm_for_each_vcpu(i, vcpu, kvm)
3326 *val += *(u32 *)((void *)vcpu + offset);
3328 spin_unlock(&kvm_lock);
3332 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
3334 static const struct file_operations *stat_fops[] = {
3335 [KVM_STAT_VCPU] = &vcpu_stat_fops,
3336 [KVM_STAT_VM] = &vm_stat_fops,
3339 static int kvm_init_debug(void)
3342 struct kvm_stats_debugfs_item *p;
3344 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3345 if (kvm_debugfs_dir == NULL)
3348 for (p = debugfs_entries; p->name; ++p) {
3349 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
3350 (void *)(long)p->offset,
3351 stat_fops[p->kind]);
3352 if (p->dentry == NULL)
3359 debugfs_remove_recursive(kvm_debugfs_dir);
3364 static void kvm_exit_debug(void)
3366 struct kvm_stats_debugfs_item *p;
3368 for (p = debugfs_entries; p->name; ++p)
3369 debugfs_remove(p->dentry);
3370 debugfs_remove(kvm_debugfs_dir);
3373 static int kvm_suspend(void)
3375 if (kvm_usage_count)
3376 hardware_disable_nolock(NULL);
3380 static void kvm_resume(void)
3382 if (kvm_usage_count) {
3383 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3384 hardware_enable_nolock(NULL);
3388 static struct syscore_ops kvm_syscore_ops = {
3389 .suspend = kvm_suspend,
3390 .resume = kvm_resume,
3394 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3396 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3399 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3401 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3403 if (vcpu->preempted)
3404 vcpu->preempted = false;
3406 kvm_arch_sched_in(vcpu, cpu);
3408 kvm_arch_vcpu_load(vcpu, cpu);
3411 static void kvm_sched_out(struct preempt_notifier *pn,
3412 struct task_struct *next)
3414 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3416 if (current->state == TASK_RUNNING)
3417 vcpu->preempted = true;
3418 kvm_arch_vcpu_put(vcpu);
3421 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3422 struct module *module)
3427 r = kvm_arch_init(opaque);
3432 * kvm_arch_init makes sure there's at most one caller
3433 * for architectures that support multiple implementations,
3434 * like intel and amd on x86.
3435 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3436 * conflicts in case kvm is already setup for another implementation.
3438 r = kvm_irqfd_init();
3442 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
3447 r = kvm_arch_hardware_setup();
3451 for_each_online_cpu(cpu) {
3452 smp_call_function_single(cpu,
3453 kvm_arch_check_processor_compat,
3459 r = register_cpu_notifier(&kvm_cpu_notifier);
3462 register_reboot_notifier(&kvm_reboot_notifier);
3464 /* A kmem cache lets us meet the alignment requirements of fx_save. */
3466 vcpu_align = __alignof__(struct kvm_vcpu);
3467 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
3469 if (!kvm_vcpu_cache) {
3474 r = kvm_async_pf_init();
3478 kvm_chardev_ops.owner = module;
3479 kvm_vm_fops.owner = module;
3480 kvm_vcpu_fops.owner = module;
3482 r = misc_register(&kvm_dev);
3484 pr_err("kvm: misc device register failed\n");
3488 register_syscore_ops(&kvm_syscore_ops);
3490 kvm_preempt_ops.sched_in = kvm_sched_in;
3491 kvm_preempt_ops.sched_out = kvm_sched_out;
3493 r = kvm_init_debug();
3495 pr_err("kvm: create debugfs files failed\n");
3499 r = kvm_vfio_ops_init();
3505 unregister_syscore_ops(&kvm_syscore_ops);
3506 misc_deregister(&kvm_dev);
3508 kvm_async_pf_deinit();
3510 kmem_cache_destroy(kvm_vcpu_cache);
3512 unregister_reboot_notifier(&kvm_reboot_notifier);
3513 unregister_cpu_notifier(&kvm_cpu_notifier);
3516 kvm_arch_hardware_unsetup();
3518 free_cpumask_var(cpus_hardware_enabled);
3526 EXPORT_SYMBOL_GPL(kvm_init);
3531 misc_deregister(&kvm_dev);
3532 kmem_cache_destroy(kvm_vcpu_cache);
3533 kvm_async_pf_deinit();
3534 unregister_syscore_ops(&kvm_syscore_ops);
3535 unregister_reboot_notifier(&kvm_reboot_notifier);
3536 unregister_cpu_notifier(&kvm_cpu_notifier);
3537 on_each_cpu(hardware_disable_nolock, NULL, 1);
3538 kvm_arch_hardware_unsetup();
3541 free_cpumask_var(cpus_hardware_enabled);
3542 kvm_vfio_ops_exit();
3544 EXPORT_SYMBOL_GPL(kvm_exit);