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>
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");
72 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
75 DEFINE_SPINLOCK(kvm_lock);
76 static DEFINE_RAW_SPINLOCK(kvm_count_lock);
79 static cpumask_var_t cpus_hardware_enabled;
80 static int kvm_usage_count = 0;
81 static atomic_t hardware_enable_failed;
83 struct kmem_cache *kvm_vcpu_cache;
84 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
86 static __read_mostly struct preempt_ops kvm_preempt_ops;
88 struct dentry *kvm_debugfs_dir;
90 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
93 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
96 static int hardware_enable_all(void);
97 static void hardware_disable_all(void);
99 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
101 static void kvm_release_pfn_dirty(pfn_t pfn);
102 static void mark_page_dirty_in_slot(struct kvm *kvm,
103 struct kvm_memory_slot *memslot, gfn_t gfn);
105 __visible bool kvm_rebooting;
106 EXPORT_SYMBOL_GPL(kvm_rebooting);
108 static bool largepages_enabled = true;
110 bool kvm_is_reserved_pfn(pfn_t pfn)
113 return PageReserved(pfn_to_page(pfn));
119 * Switches to specified vcpu, until a matching vcpu_put()
121 int vcpu_load(struct kvm_vcpu *vcpu)
125 if (mutex_lock_killable(&vcpu->mutex))
128 preempt_notifier_register(&vcpu->preempt_notifier);
129 kvm_arch_vcpu_load(vcpu, cpu);
134 void vcpu_put(struct kvm_vcpu *vcpu)
137 kvm_arch_vcpu_put(vcpu);
138 preempt_notifier_unregister(&vcpu->preempt_notifier);
140 mutex_unlock(&vcpu->mutex);
143 static void ack_flush(void *_completed)
147 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
152 struct kvm_vcpu *vcpu;
154 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
157 kvm_for_each_vcpu(i, vcpu, kvm) {
158 kvm_make_request(req, vcpu);
161 /* Set ->requests bit before we read ->mode */
164 if (cpus != NULL && cpu != -1 && cpu != me &&
165 kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
166 cpumask_set_cpu(cpu, cpus);
168 if (unlikely(cpus == NULL))
169 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
170 else if (!cpumask_empty(cpus))
171 smp_call_function_many(cpus, ack_flush, NULL, 1);
175 free_cpumask_var(cpus);
179 void kvm_flush_remote_tlbs(struct kvm *kvm)
181 long dirty_count = kvm->tlbs_dirty;
184 if (kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
185 ++kvm->stat.remote_tlb_flush;
186 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
188 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
190 void kvm_reload_remote_mmus(struct kvm *kvm)
192 kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
195 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
197 kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
200 void kvm_make_scan_ioapic_request(struct kvm *kvm)
202 kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
205 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
210 mutex_init(&vcpu->mutex);
215 init_waitqueue_head(&vcpu->wq);
216 kvm_async_pf_vcpu_init(vcpu);
218 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
223 vcpu->run = page_address(page);
225 kvm_vcpu_set_in_spin_loop(vcpu, false);
226 kvm_vcpu_set_dy_eligible(vcpu, false);
227 vcpu->preempted = false;
229 r = kvm_arch_vcpu_init(vcpu);
235 free_page((unsigned long)vcpu->run);
239 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
241 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
244 kvm_arch_vcpu_uninit(vcpu);
245 free_page((unsigned long)vcpu->run);
247 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
249 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
250 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
252 return container_of(mn, struct kvm, mmu_notifier);
255 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
256 struct mm_struct *mm,
257 unsigned long address)
259 struct kvm *kvm = mmu_notifier_to_kvm(mn);
260 int need_tlb_flush, idx;
263 * When ->invalidate_page runs, the linux pte has been zapped
264 * already but the page is still allocated until
265 * ->invalidate_page returns. So if we increase the sequence
266 * here the kvm page fault will notice if the spte can't be
267 * established because the page is going to be freed. If
268 * instead the kvm page fault establishes the spte before
269 * ->invalidate_page runs, kvm_unmap_hva will release it
272 * The sequence increase only need to be seen at spin_unlock
273 * time, and not at spin_lock time.
275 * Increasing the sequence after the spin_unlock would be
276 * unsafe because the kvm page fault could then establish the
277 * pte after kvm_unmap_hva returned, without noticing the page
278 * is going to be freed.
280 idx = srcu_read_lock(&kvm->srcu);
281 spin_lock(&kvm->mmu_lock);
283 kvm->mmu_notifier_seq++;
284 need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
285 /* we've to flush the tlb before the pages can be freed */
287 kvm_flush_remote_tlbs(kvm);
289 spin_unlock(&kvm->mmu_lock);
291 kvm_arch_mmu_notifier_invalidate_page(kvm, address);
293 srcu_read_unlock(&kvm->srcu, idx);
296 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
297 struct mm_struct *mm,
298 unsigned long address,
301 struct kvm *kvm = mmu_notifier_to_kvm(mn);
304 idx = srcu_read_lock(&kvm->srcu);
305 spin_lock(&kvm->mmu_lock);
306 kvm->mmu_notifier_seq++;
307 kvm_set_spte_hva(kvm, address, pte);
308 spin_unlock(&kvm->mmu_lock);
309 srcu_read_unlock(&kvm->srcu, idx);
312 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
313 struct mm_struct *mm,
317 struct kvm *kvm = mmu_notifier_to_kvm(mn);
318 int need_tlb_flush = 0, idx;
320 idx = srcu_read_lock(&kvm->srcu);
321 spin_lock(&kvm->mmu_lock);
323 * The count increase must become visible at unlock time as no
324 * spte can be established without taking the mmu_lock and
325 * count is also read inside the mmu_lock critical section.
327 kvm->mmu_notifier_count++;
328 need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
329 need_tlb_flush |= kvm->tlbs_dirty;
330 /* we've to flush the tlb before the pages can be freed */
332 kvm_flush_remote_tlbs(kvm);
334 spin_unlock(&kvm->mmu_lock);
335 srcu_read_unlock(&kvm->srcu, idx);
338 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
339 struct mm_struct *mm,
343 struct kvm *kvm = mmu_notifier_to_kvm(mn);
345 spin_lock(&kvm->mmu_lock);
347 * This sequence increase will notify the kvm page fault that
348 * the page that is going to be mapped in the spte could have
351 kvm->mmu_notifier_seq++;
354 * The above sequence increase must be visible before the
355 * below count decrease, which is ensured by the smp_wmb above
356 * in conjunction with the smp_rmb in mmu_notifier_retry().
358 kvm->mmu_notifier_count--;
359 spin_unlock(&kvm->mmu_lock);
361 BUG_ON(kvm->mmu_notifier_count < 0);
364 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
365 struct mm_struct *mm,
369 struct kvm *kvm = mmu_notifier_to_kvm(mn);
372 idx = srcu_read_lock(&kvm->srcu);
373 spin_lock(&kvm->mmu_lock);
375 young = kvm_age_hva(kvm, start, end);
377 kvm_flush_remote_tlbs(kvm);
379 spin_unlock(&kvm->mmu_lock);
380 srcu_read_unlock(&kvm->srcu, idx);
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_all(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 void kvm_init_memslots_id(struct kvm *kvm)
440 struct kvm_memslots *slots = kvm->memslots;
442 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
443 slots->id_to_index[i] = slots->memslots[i].id = i;
446 static struct kvm *kvm_create_vm(unsigned long type)
449 struct kvm *kvm = kvm_arch_alloc_vm();
452 return ERR_PTR(-ENOMEM);
454 r = kvm_arch_init_vm(kvm, type);
456 goto out_err_no_disable;
458 r = hardware_enable_all();
460 goto out_err_no_disable;
462 #ifdef CONFIG_HAVE_KVM_IRQFD
463 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
466 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
469 kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
471 goto out_err_no_srcu;
474 * Init kvm generation close to the maximum to easily test the
475 * code of handling generation number wrap-around.
477 kvm->memslots->generation = -150;
479 kvm_init_memslots_id(kvm);
480 if (init_srcu_struct(&kvm->srcu))
481 goto out_err_no_srcu;
482 if (init_srcu_struct(&kvm->irq_srcu))
483 goto out_err_no_irq_srcu;
484 for (i = 0; i < KVM_NR_BUSES; i++) {
485 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
491 spin_lock_init(&kvm->mmu_lock);
492 kvm->mm = current->mm;
493 atomic_inc(&kvm->mm->mm_count);
494 kvm_eventfd_init(kvm);
495 mutex_init(&kvm->lock);
496 mutex_init(&kvm->irq_lock);
497 mutex_init(&kvm->slots_lock);
498 atomic_set(&kvm->users_count, 1);
499 INIT_LIST_HEAD(&kvm->devices);
501 r = kvm_init_mmu_notifier(kvm);
505 spin_lock(&kvm_lock);
506 list_add(&kvm->vm_list, &vm_list);
507 spin_unlock(&kvm_lock);
512 cleanup_srcu_struct(&kvm->irq_srcu);
514 cleanup_srcu_struct(&kvm->srcu);
516 hardware_disable_all();
518 for (i = 0; i < KVM_NR_BUSES; i++)
519 kfree(kvm->buses[i]);
520 kfree(kvm->memslots);
521 kvm_arch_free_vm(kvm);
526 * Avoid using vmalloc for a small buffer.
527 * Should not be used when the size is statically known.
529 void *kvm_kvzalloc(unsigned long size)
531 if (size > PAGE_SIZE)
532 return vzalloc(size);
534 return kzalloc(size, GFP_KERNEL);
537 void kvm_kvfree(const void *addr)
539 if (is_vmalloc_addr(addr))
545 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
547 if (!memslot->dirty_bitmap)
550 kvm_kvfree(memslot->dirty_bitmap);
551 memslot->dirty_bitmap = NULL;
555 * Free any memory in @free but not in @dont.
557 static void kvm_free_physmem_slot(struct kvm *kvm, struct kvm_memory_slot *free,
558 struct kvm_memory_slot *dont)
560 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
561 kvm_destroy_dirty_bitmap(free);
563 kvm_arch_free_memslot(kvm, free, dont);
568 static void kvm_free_physmem(struct kvm *kvm)
570 struct kvm_memslots *slots = kvm->memslots;
571 struct kvm_memory_slot *memslot;
573 kvm_for_each_memslot(memslot, slots)
574 kvm_free_physmem_slot(kvm, memslot, NULL);
576 kfree(kvm->memslots);
579 static void kvm_destroy_devices(struct kvm *kvm)
581 struct list_head *node, *tmp;
583 list_for_each_safe(node, tmp, &kvm->devices) {
584 struct kvm_device *dev =
585 list_entry(node, struct kvm_device, vm_node);
588 dev->ops->destroy(dev);
592 static void kvm_destroy_vm(struct kvm *kvm)
595 struct mm_struct *mm = kvm->mm;
597 kvm_arch_sync_events(kvm);
598 spin_lock(&kvm_lock);
599 list_del(&kvm->vm_list);
600 spin_unlock(&kvm_lock);
601 kvm_free_irq_routing(kvm);
602 for (i = 0; i < KVM_NR_BUSES; i++)
603 kvm_io_bus_destroy(kvm->buses[i]);
604 kvm_coalesced_mmio_free(kvm);
605 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
606 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
608 kvm_arch_flush_shadow_all(kvm);
610 kvm_arch_destroy_vm(kvm);
611 kvm_destroy_devices(kvm);
612 kvm_free_physmem(kvm);
613 cleanup_srcu_struct(&kvm->irq_srcu);
614 cleanup_srcu_struct(&kvm->srcu);
615 kvm_arch_free_vm(kvm);
616 hardware_disable_all();
620 void kvm_get_kvm(struct kvm *kvm)
622 atomic_inc(&kvm->users_count);
624 EXPORT_SYMBOL_GPL(kvm_get_kvm);
626 void kvm_put_kvm(struct kvm *kvm)
628 if (atomic_dec_and_test(&kvm->users_count))
631 EXPORT_SYMBOL_GPL(kvm_put_kvm);
634 static int kvm_vm_release(struct inode *inode, struct file *filp)
636 struct kvm *kvm = filp->private_data;
638 kvm_irqfd_release(kvm);
645 * Allocation size is twice as large as the actual dirty bitmap size.
646 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
648 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
650 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
652 memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
653 if (!memslot->dirty_bitmap)
660 * Insert memslot and re-sort memslots based on their GFN,
661 * so binary search could be used to lookup GFN.
662 * Sorting algorithm takes advantage of having initially
663 * sorted array and known changed memslot position.
665 static void update_memslots(struct kvm_memslots *slots,
666 struct kvm_memory_slot *new)
669 int i = slots->id_to_index[id];
670 struct kvm_memory_slot *mslots = slots->memslots;
672 WARN_ON(mslots[i].id != id);
674 WARN_ON(!mslots[i].npages);
676 if (mslots[i].npages)
679 if (!mslots[i].npages)
683 while (i < KVM_MEM_SLOTS_NUM - 1 &&
684 new->base_gfn <= mslots[i + 1].base_gfn) {
685 if (!mslots[i + 1].npages)
687 mslots[i] = mslots[i + 1];
688 slots->id_to_index[mslots[i].id] = i;
693 * The ">=" is needed when creating a slot with base_gfn == 0,
694 * so that it moves before all those with base_gfn == npages == 0.
696 * On the other hand, if new->npages is zero, the above loop has
697 * already left i pointing to the beginning of the empty part of
698 * mslots, and the ">=" would move the hole backwards in this
699 * case---which is wrong. So skip the loop when deleting a slot.
703 new->base_gfn >= mslots[i - 1].base_gfn) {
704 mslots[i] = mslots[i - 1];
705 slots->id_to_index[mslots[i].id] = i;
709 WARN_ON_ONCE(i != slots->used_slots);
712 slots->id_to_index[mslots[i].id] = i;
715 static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
717 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
719 #ifdef __KVM_HAVE_READONLY_MEM
720 valid_flags |= KVM_MEM_READONLY;
723 if (mem->flags & ~valid_flags)
729 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
730 struct kvm_memslots *slots)
732 struct kvm_memslots *old_memslots = kvm->memslots;
735 * Set the low bit in the generation, which disables SPTE caching
736 * until the end of synchronize_srcu_expedited.
738 WARN_ON(old_memslots->generation & 1);
739 slots->generation = old_memslots->generation + 1;
741 rcu_assign_pointer(kvm->memslots, slots);
742 synchronize_srcu_expedited(&kvm->srcu);
745 * Increment the new memslot generation a second time. This prevents
746 * vm exits that race with memslot updates from caching a memslot
747 * generation that will (potentially) be valid forever.
751 kvm_arch_memslots_updated(kvm);
757 * Allocate some memory and give it an address in the guest physical address
760 * Discontiguous memory is allowed, mostly for framebuffers.
762 * Must be called holding kvm->slots_lock for write.
764 int __kvm_set_memory_region(struct kvm *kvm,
765 struct kvm_userspace_memory_region *mem)
769 unsigned long npages;
770 struct kvm_memory_slot *slot;
771 struct kvm_memory_slot old, new;
772 struct kvm_memslots *slots = NULL, *old_memslots;
773 enum kvm_mr_change change;
775 r = check_memory_region_flags(mem);
780 /* General sanity checks */
781 if (mem->memory_size & (PAGE_SIZE - 1))
783 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
785 /* We can read the guest memory with __xxx_user() later on. */
786 if ((mem->slot < KVM_USER_MEM_SLOTS) &&
787 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
788 !access_ok(VERIFY_WRITE,
789 (void __user *)(unsigned long)mem->userspace_addr,
792 if (mem->slot >= KVM_MEM_SLOTS_NUM)
794 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
797 slot = id_to_memslot(kvm->memslots, mem->slot);
798 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
799 npages = mem->memory_size >> PAGE_SHIFT;
801 if (npages > KVM_MEM_MAX_NR_PAGES)
805 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
810 new.base_gfn = base_gfn;
812 new.flags = mem->flags;
816 change = KVM_MR_CREATE;
817 else { /* Modify an existing slot. */
818 if ((mem->userspace_addr != old.userspace_addr) ||
819 (npages != old.npages) ||
820 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
823 if (base_gfn != old.base_gfn)
824 change = KVM_MR_MOVE;
825 else if (new.flags != old.flags)
826 change = KVM_MR_FLAGS_ONLY;
827 else { /* Nothing to change. */
832 } else if (old.npages) {
833 change = KVM_MR_DELETE;
834 } else /* Modify a non-existent slot: disallowed. */
837 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
838 /* Check for overlaps */
840 kvm_for_each_memslot(slot, kvm->memslots) {
841 if ((slot->id >= KVM_USER_MEM_SLOTS) ||
842 (slot->id == mem->slot))
844 if (!((base_gfn + npages <= slot->base_gfn) ||
845 (base_gfn >= slot->base_gfn + slot->npages)))
850 /* Free page dirty bitmap if unneeded */
851 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
852 new.dirty_bitmap = NULL;
855 if (change == KVM_MR_CREATE) {
856 new.userspace_addr = mem->userspace_addr;
858 if (kvm_arch_create_memslot(kvm, &new, npages))
862 /* Allocate page dirty bitmap if needed */
863 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
864 if (kvm_create_dirty_bitmap(&new) < 0)
868 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
873 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
874 slot = id_to_memslot(slots, mem->slot);
875 slot->flags |= KVM_MEMSLOT_INVALID;
877 old_memslots = install_new_memslots(kvm, slots);
879 /* slot was deleted or moved, clear iommu mapping */
880 kvm_iommu_unmap_pages(kvm, &old);
881 /* From this point no new shadow pages pointing to a deleted,
882 * or moved, memslot will be created.
884 * validation of sp->gfn happens in:
885 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
886 * - kvm_is_visible_gfn (mmu_check_roots)
888 kvm_arch_flush_shadow_memslot(kvm, slot);
891 * We can re-use the old_memslots from above, the only difference
892 * from the currently installed memslots is the invalid flag. This
893 * will get overwritten by update_memslots anyway.
895 slots = old_memslots;
898 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
902 /* actual memory is freed via old in kvm_free_physmem_slot below */
903 if (change == KVM_MR_DELETE) {
904 new.dirty_bitmap = NULL;
905 memset(&new.arch, 0, sizeof(new.arch));
908 update_memslots(slots, &new);
909 old_memslots = install_new_memslots(kvm, slots);
911 kvm_arch_commit_memory_region(kvm, mem, &old, change);
913 kvm_free_physmem_slot(kvm, &old, &new);
917 * IOMMU mapping: New slots need to be mapped. Old slots need to be
918 * un-mapped and re-mapped if their base changes. Since base change
919 * unmapping is handled above with slot deletion, mapping alone is
920 * needed here. Anything else the iommu might care about for existing
921 * slots (size changes, userspace addr changes and read-only flag
922 * changes) is disallowed above, so any other attribute changes getting
923 * here can be skipped.
925 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
926 r = kvm_iommu_map_pages(kvm, &new);
935 kvm_free_physmem_slot(kvm, &new, &old);
939 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
941 int kvm_set_memory_region(struct kvm *kvm,
942 struct kvm_userspace_memory_region *mem)
946 mutex_lock(&kvm->slots_lock);
947 r = __kvm_set_memory_region(kvm, mem);
948 mutex_unlock(&kvm->slots_lock);
951 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
953 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
954 struct kvm_userspace_memory_region *mem)
956 if (mem->slot >= KVM_USER_MEM_SLOTS)
958 return kvm_set_memory_region(kvm, mem);
961 int kvm_get_dirty_log(struct kvm *kvm,
962 struct kvm_dirty_log *log, int *is_dirty)
964 struct kvm_memory_slot *memslot;
967 unsigned long any = 0;
970 if (log->slot >= KVM_USER_MEM_SLOTS)
973 memslot = id_to_memslot(kvm->memslots, log->slot);
975 if (!memslot->dirty_bitmap)
978 n = kvm_dirty_bitmap_bytes(memslot);
980 for (i = 0; !any && i < n/sizeof(long); ++i)
981 any = memslot->dirty_bitmap[i];
984 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
994 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
996 bool kvm_largepages_enabled(void)
998 return largepages_enabled;
1001 void kvm_disable_largepages(void)
1003 largepages_enabled = false;
1005 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1007 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1009 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1011 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1013 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1015 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1017 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1018 memslot->flags & KVM_MEMSLOT_INVALID)
1023 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1025 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1027 struct vm_area_struct *vma;
1028 unsigned long addr, size;
1032 addr = gfn_to_hva(kvm, gfn);
1033 if (kvm_is_error_hva(addr))
1036 down_read(¤t->mm->mmap_sem);
1037 vma = find_vma(current->mm, addr);
1041 size = vma_kernel_pagesize(vma);
1044 up_read(¤t->mm->mmap_sem);
1049 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1051 return slot->flags & KVM_MEM_READONLY;
1054 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1055 gfn_t *nr_pages, bool write)
1057 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1058 return KVM_HVA_ERR_BAD;
1060 if (memslot_is_readonly(slot) && write)
1061 return KVM_HVA_ERR_RO_BAD;
1064 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1066 return __gfn_to_hva_memslot(slot, gfn);
1069 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1072 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1075 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1078 return gfn_to_hva_many(slot, gfn, NULL);
1080 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1082 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1084 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1086 EXPORT_SYMBOL_GPL(gfn_to_hva);
1089 * If writable is set to false, the hva returned by this function is only
1090 * allowed to be read.
1092 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1093 gfn_t gfn, bool *writable)
1095 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1097 if (!kvm_is_error_hva(hva) && writable)
1098 *writable = !memslot_is_readonly(slot);
1103 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1105 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1107 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1110 static int kvm_read_hva(void *data, void __user *hva, int len)
1112 return __copy_from_user(data, hva, len);
1115 static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1117 return __copy_from_user_inatomic(data, hva, len);
1120 static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1121 unsigned long start, int write, struct page **page)
1123 int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1126 flags |= FOLL_WRITE;
1128 return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1131 int kvm_get_user_page_io(struct task_struct *tsk, struct mm_struct *mm,
1132 unsigned long addr, bool write_fault,
1133 struct page **pagep)
1137 int flags = FOLL_TOUCH | FOLL_HWPOISON |
1138 (pagep ? FOLL_GET : 0) |
1139 (write_fault ? FOLL_WRITE : 0);
1142 * If retrying the fault, we get here *not* having allowed the filemap
1143 * to wait on the page lock. We should now allow waiting on the IO with
1144 * the mmap semaphore released.
1146 down_read(&mm->mmap_sem);
1147 npages = __get_user_pages(tsk, mm, addr, 1, flags, pagep, NULL,
1156 * The previous call has now waited on the IO. Now we can
1157 * retry and complete. Pass TRIED to ensure we do not re
1158 * schedule async IO (see e.g. filemap_fault).
1160 down_read(&mm->mmap_sem);
1161 npages = __get_user_pages(tsk, mm, addr, 1, flags | FOLL_TRIED,
1164 up_read(&mm->mmap_sem);
1168 static inline int check_user_page_hwpoison(unsigned long addr)
1170 int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1172 rc = __get_user_pages(current, current->mm, addr, 1,
1173 flags, NULL, NULL, NULL);
1174 return rc == -EHWPOISON;
1178 * The atomic path to get the writable pfn which will be stored in @pfn,
1179 * true indicates success, otherwise false is returned.
1181 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1182 bool write_fault, bool *writable, pfn_t *pfn)
1184 struct page *page[1];
1187 if (!(async || atomic))
1191 * Fast pin a writable pfn only if it is a write fault request
1192 * or the caller allows to map a writable pfn for a read fault
1195 if (!(write_fault || writable))
1198 npages = __get_user_pages_fast(addr, 1, 1, page);
1200 *pfn = page_to_pfn(page[0]);
1211 * The slow path to get the pfn of the specified host virtual address,
1212 * 1 indicates success, -errno is returned if error is detected.
1214 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1215 bool *writable, pfn_t *pfn)
1217 struct page *page[1];
1223 *writable = write_fault;
1226 down_read(¤t->mm->mmap_sem);
1227 npages = get_user_page_nowait(current, current->mm,
1228 addr, write_fault, page);
1229 up_read(¤t->mm->mmap_sem);
1232 * By now we have tried gup_fast, and possibly async_pf, and we
1233 * are certainly not atomic. Time to retry the gup, allowing
1234 * mmap semaphore to be relinquished in the case of IO.
1236 npages = kvm_get_user_page_io(current, current->mm, addr,
1242 /* map read fault as writable if possible */
1243 if (unlikely(!write_fault) && writable) {
1244 struct page *wpage[1];
1246 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1255 *pfn = page_to_pfn(page[0]);
1259 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1261 if (unlikely(!(vma->vm_flags & VM_READ)))
1264 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1271 * Pin guest page in memory and return its pfn.
1272 * @addr: host virtual address which maps memory to the guest
1273 * @atomic: whether this function can sleep
1274 * @async: whether this function need to wait IO complete if the
1275 * host page is not in the memory
1276 * @write_fault: whether we should get a writable host page
1277 * @writable: whether it allows to map a writable host page for !@write_fault
1279 * The function will map a writable host page for these two cases:
1280 * 1): @write_fault = true
1281 * 2): @write_fault = false && @writable, @writable will tell the caller
1282 * whether the mapping is writable.
1284 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1285 bool write_fault, bool *writable)
1287 struct vm_area_struct *vma;
1291 /* we can do it either atomically or asynchronously, not both */
1292 BUG_ON(atomic && async);
1294 if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1298 return KVM_PFN_ERR_FAULT;
1300 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1304 down_read(¤t->mm->mmap_sem);
1305 if (npages == -EHWPOISON ||
1306 (!async && check_user_page_hwpoison(addr))) {
1307 pfn = KVM_PFN_ERR_HWPOISON;
1311 vma = find_vma_intersection(current->mm, addr, addr + 1);
1314 pfn = KVM_PFN_ERR_FAULT;
1315 else if ((vma->vm_flags & VM_PFNMAP)) {
1316 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1318 BUG_ON(!kvm_is_reserved_pfn(pfn));
1320 if (async && vma_is_valid(vma, write_fault))
1322 pfn = KVM_PFN_ERR_FAULT;
1325 up_read(¤t->mm->mmap_sem);
1330 __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1331 bool *async, bool write_fault, bool *writable)
1333 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1335 if (addr == KVM_HVA_ERR_RO_BAD)
1336 return KVM_PFN_ERR_RO_FAULT;
1338 if (kvm_is_error_hva(addr))
1339 return KVM_PFN_NOSLOT;
1341 /* Do not map writable pfn in the readonly memslot. */
1342 if (writable && memslot_is_readonly(slot)) {
1347 return hva_to_pfn(addr, atomic, async, write_fault,
1351 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1352 bool write_fault, bool *writable)
1354 struct kvm_memory_slot *slot;
1359 slot = gfn_to_memslot(kvm, gfn);
1361 return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1365 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1367 return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1369 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1371 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1372 bool write_fault, bool *writable)
1374 return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1376 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1378 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1380 return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1382 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1384 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1387 return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1389 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1391 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1393 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1396 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1398 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1400 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1402 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1408 addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1409 if (kvm_is_error_hva(addr))
1412 if (entry < nr_pages)
1415 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1417 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1419 static struct page *kvm_pfn_to_page(pfn_t pfn)
1421 if (is_error_noslot_pfn(pfn))
1422 return KVM_ERR_PTR_BAD_PAGE;
1424 if (kvm_is_reserved_pfn(pfn)) {
1426 return KVM_ERR_PTR_BAD_PAGE;
1429 return pfn_to_page(pfn);
1432 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1436 pfn = gfn_to_pfn(kvm, gfn);
1438 return kvm_pfn_to_page(pfn);
1441 EXPORT_SYMBOL_GPL(gfn_to_page);
1443 void kvm_release_page_clean(struct page *page)
1445 WARN_ON(is_error_page(page));
1447 kvm_release_pfn_clean(page_to_pfn(page));
1449 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1451 void kvm_release_pfn_clean(pfn_t pfn)
1453 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
1454 put_page(pfn_to_page(pfn));
1456 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1458 void kvm_release_page_dirty(struct page *page)
1460 WARN_ON(is_error_page(page));
1462 kvm_release_pfn_dirty(page_to_pfn(page));
1464 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1466 static void kvm_release_pfn_dirty(pfn_t pfn)
1468 kvm_set_pfn_dirty(pfn);
1469 kvm_release_pfn_clean(pfn);
1472 void kvm_set_pfn_dirty(pfn_t pfn)
1474 if (!kvm_is_reserved_pfn(pfn)) {
1475 struct page *page = pfn_to_page(pfn);
1476 if (!PageReserved(page))
1480 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1482 void kvm_set_pfn_accessed(pfn_t pfn)
1484 if (!kvm_is_reserved_pfn(pfn))
1485 mark_page_accessed(pfn_to_page(pfn));
1487 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1489 void kvm_get_pfn(pfn_t pfn)
1491 if (!kvm_is_reserved_pfn(pfn))
1492 get_page(pfn_to_page(pfn));
1494 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1496 static int next_segment(unsigned long len, int offset)
1498 if (len > PAGE_SIZE - offset)
1499 return PAGE_SIZE - offset;
1504 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1510 addr = gfn_to_hva_prot(kvm, gfn, NULL);
1511 if (kvm_is_error_hva(addr))
1513 r = kvm_read_hva(data, (void __user *)addr + offset, len);
1518 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1520 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1522 gfn_t gfn = gpa >> PAGE_SHIFT;
1524 int offset = offset_in_page(gpa);
1527 while ((seg = next_segment(len, offset)) != 0) {
1528 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1538 EXPORT_SYMBOL_GPL(kvm_read_guest);
1540 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1545 gfn_t gfn = gpa >> PAGE_SHIFT;
1546 int offset = offset_in_page(gpa);
1548 addr = gfn_to_hva_prot(kvm, gfn, NULL);
1549 if (kvm_is_error_hva(addr))
1551 pagefault_disable();
1552 r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
1558 EXPORT_SYMBOL(kvm_read_guest_atomic);
1560 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1561 int offset, int len)
1566 addr = gfn_to_hva(kvm, gfn);
1567 if (kvm_is_error_hva(addr))
1569 r = __copy_to_user((void __user *)addr + offset, data, len);
1572 mark_page_dirty(kvm, gfn);
1575 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1577 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1580 gfn_t gfn = gpa >> PAGE_SHIFT;
1582 int offset = offset_in_page(gpa);
1585 while ((seg = next_segment(len, offset)) != 0) {
1586 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1597 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1598 gpa_t gpa, unsigned long len)
1600 struct kvm_memslots *slots = kvm_memslots(kvm);
1601 int offset = offset_in_page(gpa);
1602 gfn_t start_gfn = gpa >> PAGE_SHIFT;
1603 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1604 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1605 gfn_t nr_pages_avail;
1608 ghc->generation = slots->generation;
1610 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1611 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, &nr_pages_avail);
1612 if (!kvm_is_error_hva(ghc->hva) && nr_pages_avail >= nr_pages_needed) {
1616 * If the requested region crosses two memslots, we still
1617 * verify that the entire region is valid here.
1619 while (start_gfn <= end_gfn) {
1620 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1621 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1623 if (kvm_is_error_hva(ghc->hva))
1625 start_gfn += nr_pages_avail;
1627 /* Use the slow path for cross page reads and writes. */
1628 ghc->memslot = NULL;
1632 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1634 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1635 void *data, unsigned long len)
1637 struct kvm_memslots *slots = kvm_memslots(kvm);
1640 BUG_ON(len > ghc->len);
1642 if (slots->generation != ghc->generation)
1643 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1645 if (unlikely(!ghc->memslot))
1646 return kvm_write_guest(kvm, ghc->gpa, data, len);
1648 if (kvm_is_error_hva(ghc->hva))
1651 r = __copy_to_user((void __user *)ghc->hva, data, len);
1654 mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1658 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1660 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1661 void *data, unsigned long len)
1663 struct kvm_memslots *slots = kvm_memslots(kvm);
1666 BUG_ON(len > ghc->len);
1668 if (slots->generation != ghc->generation)
1669 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1671 if (unlikely(!ghc->memslot))
1672 return kvm_read_guest(kvm, ghc->gpa, data, len);
1674 if (kvm_is_error_hva(ghc->hva))
1677 r = __copy_from_user(data, (void __user *)ghc->hva, len);
1683 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1685 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1687 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
1689 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
1691 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1693 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1695 gfn_t gfn = gpa >> PAGE_SHIFT;
1697 int offset = offset_in_page(gpa);
1700 while ((seg = next_segment(len, offset)) != 0) {
1701 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1710 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1712 static void mark_page_dirty_in_slot(struct kvm *kvm,
1713 struct kvm_memory_slot *memslot,
1716 if (memslot && memslot->dirty_bitmap) {
1717 unsigned long rel_gfn = gfn - memslot->base_gfn;
1719 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1723 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1725 struct kvm_memory_slot *memslot;
1727 memslot = gfn_to_memslot(kvm, gfn);
1728 mark_page_dirty_in_slot(kvm, memslot, gfn);
1730 EXPORT_SYMBOL_GPL(mark_page_dirty);
1733 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1735 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1740 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1742 if (kvm_arch_vcpu_runnable(vcpu)) {
1743 kvm_make_request(KVM_REQ_UNHALT, vcpu);
1746 if (kvm_cpu_has_pending_timer(vcpu))
1748 if (signal_pending(current))
1754 finish_wait(&vcpu->wq, &wait);
1756 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
1760 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1762 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1765 int cpu = vcpu->cpu;
1766 wait_queue_head_t *wqp;
1768 wqp = kvm_arch_vcpu_wq(vcpu);
1769 if (waitqueue_active(wqp)) {
1770 wake_up_interruptible(wqp);
1771 ++vcpu->stat.halt_wakeup;
1775 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1776 if (kvm_arch_vcpu_should_kick(vcpu))
1777 smp_send_reschedule(cpu);
1780 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
1781 #endif /* !CONFIG_S390 */
1783 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
1786 struct task_struct *task = NULL;
1790 pid = rcu_dereference(target->pid);
1792 task = get_pid_task(pid, PIDTYPE_PID);
1796 ret = yield_to(task, 1);
1797 put_task_struct(task);
1801 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1804 * Helper that checks whether a VCPU is eligible for directed yield.
1805 * Most eligible candidate to yield is decided by following heuristics:
1807 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1808 * (preempted lock holder), indicated by @in_spin_loop.
1809 * Set at the beiginning and cleared at the end of interception/PLE handler.
1811 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1812 * chance last time (mostly it has become eligible now since we have probably
1813 * yielded to lockholder in last iteration. This is done by toggling
1814 * @dy_eligible each time a VCPU checked for eligibility.)
1816 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1817 * to preempted lock-holder could result in wrong VCPU selection and CPU
1818 * burning. Giving priority for a potential lock-holder increases lock
1821 * Since algorithm is based on heuristics, accessing another VCPU data without
1822 * locking does not harm. It may result in trying to yield to same VCPU, fail
1823 * and continue with next VCPU and so on.
1825 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1827 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1830 eligible = !vcpu->spin_loop.in_spin_loop ||
1831 vcpu->spin_loop.dy_eligible;
1833 if (vcpu->spin_loop.in_spin_loop)
1834 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1842 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1844 struct kvm *kvm = me->kvm;
1845 struct kvm_vcpu *vcpu;
1846 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1852 kvm_vcpu_set_in_spin_loop(me, true);
1854 * We boost the priority of a VCPU that is runnable but not
1855 * currently running, because it got preempted by something
1856 * else and called schedule in __vcpu_run. Hopefully that
1857 * VCPU is holding the lock that we need and will release it.
1858 * We approximate round-robin by starting at the last boosted VCPU.
1860 for (pass = 0; pass < 2 && !yielded && try; pass++) {
1861 kvm_for_each_vcpu(i, vcpu, kvm) {
1862 if (!pass && i <= last_boosted_vcpu) {
1863 i = last_boosted_vcpu;
1865 } else if (pass && i > last_boosted_vcpu)
1867 if (!ACCESS_ONCE(vcpu->preempted))
1871 if (waitqueue_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
1873 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1876 yielded = kvm_vcpu_yield_to(vcpu);
1878 kvm->last_boosted_vcpu = i;
1880 } else if (yielded < 0) {
1887 kvm_vcpu_set_in_spin_loop(me, false);
1889 /* Ensure vcpu is not eligible during next spinloop */
1890 kvm_vcpu_set_dy_eligible(me, false);
1892 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1894 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1896 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1899 if (vmf->pgoff == 0)
1900 page = virt_to_page(vcpu->run);
1902 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1903 page = virt_to_page(vcpu->arch.pio_data);
1905 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1906 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1907 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1910 return kvm_arch_vcpu_fault(vcpu, vmf);
1916 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1917 .fault = kvm_vcpu_fault,
1920 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1922 vma->vm_ops = &kvm_vcpu_vm_ops;
1926 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1928 struct kvm_vcpu *vcpu = filp->private_data;
1930 kvm_put_kvm(vcpu->kvm);
1934 static struct file_operations kvm_vcpu_fops = {
1935 .release = kvm_vcpu_release,
1936 .unlocked_ioctl = kvm_vcpu_ioctl,
1937 #ifdef CONFIG_COMPAT
1938 .compat_ioctl = kvm_vcpu_compat_ioctl,
1940 .mmap = kvm_vcpu_mmap,
1941 .llseek = noop_llseek,
1945 * Allocates an inode for the vcpu.
1947 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1949 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
1953 * Creates some virtual cpus. Good luck creating more than one.
1955 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1958 struct kvm_vcpu *vcpu, *v;
1960 if (id >= KVM_MAX_VCPUS)
1963 vcpu = kvm_arch_vcpu_create(kvm, id);
1965 return PTR_ERR(vcpu);
1967 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1969 r = kvm_arch_vcpu_setup(vcpu);
1973 mutex_lock(&kvm->lock);
1974 if (!kvm_vcpu_compatible(vcpu)) {
1976 goto unlock_vcpu_destroy;
1978 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1980 goto unlock_vcpu_destroy;
1983 kvm_for_each_vcpu(r, v, kvm)
1984 if (v->vcpu_id == id) {
1986 goto unlock_vcpu_destroy;
1989 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1991 /* Now it's all set up, let userspace reach it */
1993 r = create_vcpu_fd(vcpu);
1996 goto unlock_vcpu_destroy;
1999 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
2001 atomic_inc(&kvm->online_vcpus);
2003 mutex_unlock(&kvm->lock);
2004 kvm_arch_vcpu_postcreate(vcpu);
2007 unlock_vcpu_destroy:
2008 mutex_unlock(&kvm->lock);
2010 kvm_arch_vcpu_destroy(vcpu);
2014 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2017 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2018 vcpu->sigset_active = 1;
2019 vcpu->sigset = *sigset;
2021 vcpu->sigset_active = 0;
2025 static long kvm_vcpu_ioctl(struct file *filp,
2026 unsigned int ioctl, unsigned long arg)
2028 struct kvm_vcpu *vcpu = filp->private_data;
2029 void __user *argp = (void __user *)arg;
2031 struct kvm_fpu *fpu = NULL;
2032 struct kvm_sregs *kvm_sregs = NULL;
2034 if (vcpu->kvm->mm != current->mm)
2037 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2040 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
2042 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
2043 * so vcpu_load() would break it.
2045 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
2046 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2050 r = vcpu_load(vcpu);
2058 if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
2059 /* The thread running this VCPU changed. */
2060 struct pid *oldpid = vcpu->pid;
2061 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
2062 rcu_assign_pointer(vcpu->pid, newpid);
2067 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2068 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2070 case KVM_GET_REGS: {
2071 struct kvm_regs *kvm_regs;
2074 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2077 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2081 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2088 case KVM_SET_REGS: {
2089 struct kvm_regs *kvm_regs;
2092 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2093 if (IS_ERR(kvm_regs)) {
2094 r = PTR_ERR(kvm_regs);
2097 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2101 case KVM_GET_SREGS: {
2102 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2106 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2110 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2115 case KVM_SET_SREGS: {
2116 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2117 if (IS_ERR(kvm_sregs)) {
2118 r = PTR_ERR(kvm_sregs);
2122 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2125 case KVM_GET_MP_STATE: {
2126 struct kvm_mp_state mp_state;
2128 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2132 if (copy_to_user(argp, &mp_state, sizeof mp_state))
2137 case KVM_SET_MP_STATE: {
2138 struct kvm_mp_state mp_state;
2141 if (copy_from_user(&mp_state, argp, sizeof mp_state))
2143 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2146 case KVM_TRANSLATE: {
2147 struct kvm_translation tr;
2150 if (copy_from_user(&tr, argp, sizeof tr))
2152 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2156 if (copy_to_user(argp, &tr, sizeof tr))
2161 case KVM_SET_GUEST_DEBUG: {
2162 struct kvm_guest_debug dbg;
2165 if (copy_from_user(&dbg, argp, sizeof dbg))
2167 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2170 case KVM_SET_SIGNAL_MASK: {
2171 struct kvm_signal_mask __user *sigmask_arg = argp;
2172 struct kvm_signal_mask kvm_sigmask;
2173 sigset_t sigset, *p;
2178 if (copy_from_user(&kvm_sigmask, argp,
2179 sizeof kvm_sigmask))
2182 if (kvm_sigmask.len != sizeof sigset)
2185 if (copy_from_user(&sigset, sigmask_arg->sigset,
2190 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2194 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2198 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2202 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2208 fpu = memdup_user(argp, sizeof(*fpu));
2214 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2218 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2227 #ifdef CONFIG_COMPAT
2228 static long kvm_vcpu_compat_ioctl(struct file *filp,
2229 unsigned int ioctl, unsigned long arg)
2231 struct kvm_vcpu *vcpu = filp->private_data;
2232 void __user *argp = compat_ptr(arg);
2235 if (vcpu->kvm->mm != current->mm)
2239 case KVM_SET_SIGNAL_MASK: {
2240 struct kvm_signal_mask __user *sigmask_arg = argp;
2241 struct kvm_signal_mask kvm_sigmask;
2242 compat_sigset_t csigset;
2247 if (copy_from_user(&kvm_sigmask, argp,
2248 sizeof kvm_sigmask))
2251 if (kvm_sigmask.len != sizeof csigset)
2254 if (copy_from_user(&csigset, sigmask_arg->sigset,
2257 sigset_from_compat(&sigset, &csigset);
2258 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2260 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2264 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2272 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2273 int (*accessor)(struct kvm_device *dev,
2274 struct kvm_device_attr *attr),
2277 struct kvm_device_attr attr;
2282 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2285 return accessor(dev, &attr);
2288 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2291 struct kvm_device *dev = filp->private_data;
2294 case KVM_SET_DEVICE_ATTR:
2295 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2296 case KVM_GET_DEVICE_ATTR:
2297 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2298 case KVM_HAS_DEVICE_ATTR:
2299 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2301 if (dev->ops->ioctl)
2302 return dev->ops->ioctl(dev, ioctl, arg);
2308 static int kvm_device_release(struct inode *inode, struct file *filp)
2310 struct kvm_device *dev = filp->private_data;
2311 struct kvm *kvm = dev->kvm;
2317 static const struct file_operations kvm_device_fops = {
2318 .unlocked_ioctl = kvm_device_ioctl,
2319 #ifdef CONFIG_COMPAT
2320 .compat_ioctl = kvm_device_ioctl,
2322 .release = kvm_device_release,
2325 struct kvm_device *kvm_device_from_filp(struct file *filp)
2327 if (filp->f_op != &kvm_device_fops)
2330 return filp->private_data;
2333 static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
2334 #ifdef CONFIG_KVM_MPIC
2335 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
2336 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
2339 #ifdef CONFIG_KVM_XICS
2340 [KVM_DEV_TYPE_XICS] = &kvm_xics_ops,
2344 int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
2346 if (type >= ARRAY_SIZE(kvm_device_ops_table))
2349 if (kvm_device_ops_table[type] != NULL)
2352 kvm_device_ops_table[type] = ops;
2356 void kvm_unregister_device_ops(u32 type)
2358 if (kvm_device_ops_table[type] != NULL)
2359 kvm_device_ops_table[type] = NULL;
2362 static int kvm_ioctl_create_device(struct kvm *kvm,
2363 struct kvm_create_device *cd)
2365 struct kvm_device_ops *ops = NULL;
2366 struct kvm_device *dev;
2367 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2370 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
2373 ops = kvm_device_ops_table[cd->type];
2380 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2387 ret = ops->create(dev, cd->type);
2393 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2399 list_add(&dev->vm_node, &kvm->devices);
2405 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2408 case KVM_CAP_USER_MEMORY:
2409 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2410 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2411 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2412 case KVM_CAP_SET_BOOT_CPU_ID:
2414 case KVM_CAP_INTERNAL_ERROR_DATA:
2415 #ifdef CONFIG_HAVE_KVM_MSI
2416 case KVM_CAP_SIGNAL_MSI:
2418 #ifdef CONFIG_HAVE_KVM_IRQFD
2419 case KVM_CAP_IRQFD_RESAMPLE:
2421 case KVM_CAP_CHECK_EXTENSION_VM:
2423 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2424 case KVM_CAP_IRQ_ROUTING:
2425 return KVM_MAX_IRQ_ROUTES;
2430 return kvm_vm_ioctl_check_extension(kvm, arg);
2433 static long kvm_vm_ioctl(struct file *filp,
2434 unsigned int ioctl, unsigned long arg)
2436 struct kvm *kvm = filp->private_data;
2437 void __user *argp = (void __user *)arg;
2440 if (kvm->mm != current->mm)
2443 case KVM_CREATE_VCPU:
2444 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2446 case KVM_SET_USER_MEMORY_REGION: {
2447 struct kvm_userspace_memory_region kvm_userspace_mem;
2450 if (copy_from_user(&kvm_userspace_mem, argp,
2451 sizeof kvm_userspace_mem))
2454 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2457 case KVM_GET_DIRTY_LOG: {
2458 struct kvm_dirty_log log;
2461 if (copy_from_user(&log, argp, sizeof log))
2463 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2466 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2467 case KVM_REGISTER_COALESCED_MMIO: {
2468 struct kvm_coalesced_mmio_zone zone;
2470 if (copy_from_user(&zone, argp, sizeof zone))
2472 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2475 case KVM_UNREGISTER_COALESCED_MMIO: {
2476 struct kvm_coalesced_mmio_zone zone;
2478 if (copy_from_user(&zone, argp, sizeof zone))
2480 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2485 struct kvm_irqfd data;
2488 if (copy_from_user(&data, argp, sizeof data))
2490 r = kvm_irqfd(kvm, &data);
2493 case KVM_IOEVENTFD: {
2494 struct kvm_ioeventfd data;
2497 if (copy_from_user(&data, argp, sizeof data))
2499 r = kvm_ioeventfd(kvm, &data);
2502 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2503 case KVM_SET_BOOT_CPU_ID:
2505 mutex_lock(&kvm->lock);
2506 if (atomic_read(&kvm->online_vcpus) != 0)
2509 kvm->bsp_vcpu_id = arg;
2510 mutex_unlock(&kvm->lock);
2513 #ifdef CONFIG_HAVE_KVM_MSI
2514 case KVM_SIGNAL_MSI: {
2518 if (copy_from_user(&msi, argp, sizeof msi))
2520 r = kvm_send_userspace_msi(kvm, &msi);
2524 #ifdef __KVM_HAVE_IRQ_LINE
2525 case KVM_IRQ_LINE_STATUS:
2526 case KVM_IRQ_LINE: {
2527 struct kvm_irq_level irq_event;
2530 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2533 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
2534 ioctl == KVM_IRQ_LINE_STATUS);
2539 if (ioctl == KVM_IRQ_LINE_STATUS) {
2540 if (copy_to_user(argp, &irq_event, sizeof irq_event))
2548 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2549 case KVM_SET_GSI_ROUTING: {
2550 struct kvm_irq_routing routing;
2551 struct kvm_irq_routing __user *urouting;
2552 struct kvm_irq_routing_entry *entries;
2555 if (copy_from_user(&routing, argp, sizeof(routing)))
2558 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2563 entries = vmalloc(routing.nr * sizeof(*entries));
2568 if (copy_from_user(entries, urouting->entries,
2569 routing.nr * sizeof(*entries)))
2570 goto out_free_irq_routing;
2571 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2573 out_free_irq_routing:
2577 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
2578 case KVM_CREATE_DEVICE: {
2579 struct kvm_create_device cd;
2582 if (copy_from_user(&cd, argp, sizeof(cd)))
2585 r = kvm_ioctl_create_device(kvm, &cd);
2590 if (copy_to_user(argp, &cd, sizeof(cd)))
2596 case KVM_CHECK_EXTENSION:
2597 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
2600 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2606 #ifdef CONFIG_COMPAT
2607 struct compat_kvm_dirty_log {
2611 compat_uptr_t dirty_bitmap; /* one bit per page */
2616 static long kvm_vm_compat_ioctl(struct file *filp,
2617 unsigned int ioctl, unsigned long arg)
2619 struct kvm *kvm = filp->private_data;
2622 if (kvm->mm != current->mm)
2625 case KVM_GET_DIRTY_LOG: {
2626 struct compat_kvm_dirty_log compat_log;
2627 struct kvm_dirty_log log;
2630 if (copy_from_user(&compat_log, (void __user *)arg,
2631 sizeof(compat_log)))
2633 log.slot = compat_log.slot;
2634 log.padding1 = compat_log.padding1;
2635 log.padding2 = compat_log.padding2;
2636 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2638 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2642 r = kvm_vm_ioctl(filp, ioctl, arg);
2650 static struct file_operations kvm_vm_fops = {
2651 .release = kvm_vm_release,
2652 .unlocked_ioctl = kvm_vm_ioctl,
2653 #ifdef CONFIG_COMPAT
2654 .compat_ioctl = kvm_vm_compat_ioctl,
2656 .llseek = noop_llseek,
2659 static int kvm_dev_ioctl_create_vm(unsigned long type)
2664 kvm = kvm_create_vm(type);
2666 return PTR_ERR(kvm);
2667 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2668 r = kvm_coalesced_mmio_init(kvm);
2674 r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC);
2681 static long kvm_dev_ioctl(struct file *filp,
2682 unsigned int ioctl, unsigned long arg)
2687 case KVM_GET_API_VERSION:
2690 r = KVM_API_VERSION;
2693 r = kvm_dev_ioctl_create_vm(arg);
2695 case KVM_CHECK_EXTENSION:
2696 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
2698 case KVM_GET_VCPU_MMAP_SIZE:
2701 r = PAGE_SIZE; /* struct kvm_run */
2703 r += PAGE_SIZE; /* pio data page */
2705 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2706 r += PAGE_SIZE; /* coalesced mmio ring page */
2709 case KVM_TRACE_ENABLE:
2710 case KVM_TRACE_PAUSE:
2711 case KVM_TRACE_DISABLE:
2715 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2721 static struct file_operations kvm_chardev_ops = {
2722 .unlocked_ioctl = kvm_dev_ioctl,
2723 .compat_ioctl = kvm_dev_ioctl,
2724 .llseek = noop_llseek,
2727 static struct miscdevice kvm_dev = {
2733 static void hardware_enable_nolock(void *junk)
2735 int cpu = raw_smp_processor_id();
2738 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2741 cpumask_set_cpu(cpu, cpus_hardware_enabled);
2743 r = kvm_arch_hardware_enable();
2746 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2747 atomic_inc(&hardware_enable_failed);
2748 printk(KERN_INFO "kvm: enabling virtualization on "
2749 "CPU%d failed\n", cpu);
2753 static void hardware_enable(void)
2755 raw_spin_lock(&kvm_count_lock);
2756 if (kvm_usage_count)
2757 hardware_enable_nolock(NULL);
2758 raw_spin_unlock(&kvm_count_lock);
2761 static void hardware_disable_nolock(void *junk)
2763 int cpu = raw_smp_processor_id();
2765 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2767 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2768 kvm_arch_hardware_disable();
2771 static void hardware_disable(void)
2773 raw_spin_lock(&kvm_count_lock);
2774 if (kvm_usage_count)
2775 hardware_disable_nolock(NULL);
2776 raw_spin_unlock(&kvm_count_lock);
2779 static void hardware_disable_all_nolock(void)
2781 BUG_ON(!kvm_usage_count);
2784 if (!kvm_usage_count)
2785 on_each_cpu(hardware_disable_nolock, NULL, 1);
2788 static void hardware_disable_all(void)
2790 raw_spin_lock(&kvm_count_lock);
2791 hardware_disable_all_nolock();
2792 raw_spin_unlock(&kvm_count_lock);
2795 static int hardware_enable_all(void)
2799 raw_spin_lock(&kvm_count_lock);
2802 if (kvm_usage_count == 1) {
2803 atomic_set(&hardware_enable_failed, 0);
2804 on_each_cpu(hardware_enable_nolock, NULL, 1);
2806 if (atomic_read(&hardware_enable_failed)) {
2807 hardware_disable_all_nolock();
2812 raw_spin_unlock(&kvm_count_lock);
2817 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2822 val &= ~CPU_TASKS_FROZEN;
2825 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2830 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2838 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2842 * Some (well, at least mine) BIOSes hang on reboot if
2845 * And Intel TXT required VMX off for all cpu when system shutdown.
2847 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2848 kvm_rebooting = true;
2849 on_each_cpu(hardware_disable_nolock, NULL, 1);
2853 static struct notifier_block kvm_reboot_notifier = {
2854 .notifier_call = kvm_reboot,
2858 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2862 for (i = 0; i < bus->dev_count; i++) {
2863 struct kvm_io_device *pos = bus->range[i].dev;
2865 kvm_iodevice_destructor(pos);
2870 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
2871 const struct kvm_io_range *r2)
2873 if (r1->addr < r2->addr)
2875 if (r1->addr + r1->len > r2->addr + r2->len)
2880 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2882 return kvm_io_bus_cmp(p1, p2);
2885 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2886 gpa_t addr, int len)
2888 bus->range[bus->dev_count++] = (struct kvm_io_range) {
2894 sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2895 kvm_io_bus_sort_cmp, NULL);
2900 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2901 gpa_t addr, int len)
2903 struct kvm_io_range *range, key;
2906 key = (struct kvm_io_range) {
2911 range = bsearch(&key, bus->range, bus->dev_count,
2912 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2916 off = range - bus->range;
2918 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
2924 static int __kvm_io_bus_write(struct kvm_io_bus *bus,
2925 struct kvm_io_range *range, const void *val)
2929 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2933 while (idx < bus->dev_count &&
2934 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
2935 if (!kvm_iodevice_write(bus->range[idx].dev, range->addr,
2944 /* kvm_io_bus_write - called under kvm->slots_lock */
2945 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2946 int len, const void *val)
2948 struct kvm_io_bus *bus;
2949 struct kvm_io_range range;
2952 range = (struct kvm_io_range) {
2957 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2958 r = __kvm_io_bus_write(bus, &range, val);
2959 return r < 0 ? r : 0;
2962 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
2963 int kvm_io_bus_write_cookie(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2964 int len, const void *val, long cookie)
2966 struct kvm_io_bus *bus;
2967 struct kvm_io_range range;
2969 range = (struct kvm_io_range) {
2974 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2976 /* First try the device referenced by cookie. */
2977 if ((cookie >= 0) && (cookie < bus->dev_count) &&
2978 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
2979 if (!kvm_iodevice_write(bus->range[cookie].dev, addr, len,
2984 * cookie contained garbage; fall back to search and return the
2985 * correct cookie value.
2987 return __kvm_io_bus_write(bus, &range, val);
2990 static int __kvm_io_bus_read(struct kvm_io_bus *bus, struct kvm_io_range *range,
2995 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2999 while (idx < bus->dev_count &&
3000 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3001 if (!kvm_iodevice_read(bus->range[idx].dev, range->addr,
3009 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
3011 /* kvm_io_bus_read - called under kvm->slots_lock */
3012 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3015 struct kvm_io_bus *bus;
3016 struct kvm_io_range range;
3019 range = (struct kvm_io_range) {
3024 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
3025 r = __kvm_io_bus_read(bus, &range, val);
3026 return r < 0 ? r : 0;
3030 /* Caller must hold slots_lock. */
3031 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3032 int len, struct kvm_io_device *dev)
3034 struct kvm_io_bus *new_bus, *bus;
3036 bus = kvm->buses[bus_idx];
3037 /* exclude ioeventfd which is limited by maximum fd */
3038 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3041 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
3042 sizeof(struct kvm_io_range)), GFP_KERNEL);
3045 memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
3046 sizeof(struct kvm_io_range)));
3047 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
3048 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3049 synchronize_srcu_expedited(&kvm->srcu);
3055 /* Caller must hold slots_lock. */
3056 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3057 struct kvm_io_device *dev)
3060 struct kvm_io_bus *new_bus, *bus;
3062 bus = kvm->buses[bus_idx];
3064 for (i = 0; i < bus->dev_count; i++)
3065 if (bus->range[i].dev == dev) {
3073 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3074 sizeof(struct kvm_io_range)), GFP_KERNEL);
3078 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3079 new_bus->dev_count--;
3080 memcpy(new_bus->range + i, bus->range + i + 1,
3081 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3083 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3084 synchronize_srcu_expedited(&kvm->srcu);
3089 static struct notifier_block kvm_cpu_notifier = {
3090 .notifier_call = kvm_cpu_hotplug,
3093 static int vm_stat_get(void *_offset, u64 *val)
3095 unsigned offset = (long)_offset;
3099 spin_lock(&kvm_lock);
3100 list_for_each_entry(kvm, &vm_list, vm_list)
3101 *val += *(u32 *)((void *)kvm + offset);
3102 spin_unlock(&kvm_lock);
3106 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
3108 static int vcpu_stat_get(void *_offset, u64 *val)
3110 unsigned offset = (long)_offset;
3112 struct kvm_vcpu *vcpu;
3116 spin_lock(&kvm_lock);
3117 list_for_each_entry(kvm, &vm_list, vm_list)
3118 kvm_for_each_vcpu(i, vcpu, kvm)
3119 *val += *(u32 *)((void *)vcpu + offset);
3121 spin_unlock(&kvm_lock);
3125 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
3127 static const struct file_operations *stat_fops[] = {
3128 [KVM_STAT_VCPU] = &vcpu_stat_fops,
3129 [KVM_STAT_VM] = &vm_stat_fops,
3132 static int kvm_init_debug(void)
3135 struct kvm_stats_debugfs_item *p;
3137 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3138 if (kvm_debugfs_dir == NULL)
3141 for (p = debugfs_entries; p->name; ++p) {
3142 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
3143 (void *)(long)p->offset,
3144 stat_fops[p->kind]);
3145 if (p->dentry == NULL)
3152 debugfs_remove_recursive(kvm_debugfs_dir);
3157 static void kvm_exit_debug(void)
3159 struct kvm_stats_debugfs_item *p;
3161 for (p = debugfs_entries; p->name; ++p)
3162 debugfs_remove(p->dentry);
3163 debugfs_remove(kvm_debugfs_dir);
3166 static int kvm_suspend(void)
3168 if (kvm_usage_count)
3169 hardware_disable_nolock(NULL);
3173 static void kvm_resume(void)
3175 if (kvm_usage_count) {
3176 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3177 hardware_enable_nolock(NULL);
3181 static struct syscore_ops kvm_syscore_ops = {
3182 .suspend = kvm_suspend,
3183 .resume = kvm_resume,
3187 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3189 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3192 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3194 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3195 if (vcpu->preempted)
3196 vcpu->preempted = false;
3198 kvm_arch_sched_in(vcpu, cpu);
3200 kvm_arch_vcpu_load(vcpu, cpu);
3203 static void kvm_sched_out(struct preempt_notifier *pn,
3204 struct task_struct *next)
3206 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3208 if (current->state == TASK_RUNNING)
3209 vcpu->preempted = true;
3210 kvm_arch_vcpu_put(vcpu);
3213 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3214 struct module *module)
3219 r = kvm_arch_init(opaque);
3224 * kvm_arch_init makes sure there's at most one caller
3225 * for architectures that support multiple implementations,
3226 * like intel and amd on x86.
3227 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3228 * conflicts in case kvm is already setup for another implementation.
3230 r = kvm_irqfd_init();
3234 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
3239 r = kvm_arch_hardware_setup();
3243 for_each_online_cpu(cpu) {
3244 smp_call_function_single(cpu,
3245 kvm_arch_check_processor_compat,
3251 r = register_cpu_notifier(&kvm_cpu_notifier);
3254 register_reboot_notifier(&kvm_reboot_notifier);
3256 /* A kmem cache lets us meet the alignment requirements of fx_save. */
3258 vcpu_align = __alignof__(struct kvm_vcpu);
3259 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
3261 if (!kvm_vcpu_cache) {
3266 r = kvm_async_pf_init();
3270 kvm_chardev_ops.owner = module;
3271 kvm_vm_fops.owner = module;
3272 kvm_vcpu_fops.owner = module;
3274 r = misc_register(&kvm_dev);
3276 printk(KERN_ERR "kvm: misc device register failed\n");
3280 register_syscore_ops(&kvm_syscore_ops);
3282 kvm_preempt_ops.sched_in = kvm_sched_in;
3283 kvm_preempt_ops.sched_out = kvm_sched_out;
3285 r = kvm_init_debug();
3287 printk(KERN_ERR "kvm: create debugfs files failed\n");
3291 r = kvm_vfio_ops_init();
3297 unregister_syscore_ops(&kvm_syscore_ops);
3298 misc_deregister(&kvm_dev);
3300 kvm_async_pf_deinit();
3302 kmem_cache_destroy(kvm_vcpu_cache);
3304 unregister_reboot_notifier(&kvm_reboot_notifier);
3305 unregister_cpu_notifier(&kvm_cpu_notifier);
3308 kvm_arch_hardware_unsetup();
3310 free_cpumask_var(cpus_hardware_enabled);
3318 EXPORT_SYMBOL_GPL(kvm_init);
3323 misc_deregister(&kvm_dev);
3324 kmem_cache_destroy(kvm_vcpu_cache);
3325 kvm_async_pf_deinit();
3326 unregister_syscore_ops(&kvm_syscore_ops);
3327 unregister_reboot_notifier(&kvm_reboot_notifier);
3328 unregister_cpu_notifier(&kvm_cpu_notifier);
3329 on_each_cpu(hardware_disable_nolock, NULL, 1);
3330 kvm_arch_hardware_unsetup();
3333 free_cpumask_var(cpus_hardware_enabled);
3334 kvm_vfio_ops_exit();
3336 EXPORT_SYMBOL_GPL(kvm_exit);