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);
675 if (mslots[i].npages)
678 if (!mslots[i].npages)
682 while (i < KVM_MEM_SLOTS_NUM - 1 &&
683 new->base_gfn <= mslots[i + 1].base_gfn) {
684 if (!mslots[i + 1].npages)
686 mslots[i] = mslots[i + 1];
687 slots->id_to_index[mslots[i].id] = i;
692 * The ">=" is needed when creating a slot with base_gfn == 0,
693 * so that it moves before all those with base_gfn == npages == 0.
695 * On the other hand, if new->npages is zero, the above loop has
696 * already left i pointing to the beginning of the empty part of
697 * mslots, and the ">=" would move the hole backwards in this
698 * case---which is wrong. So skip the loop when deleting a slot.
702 new->base_gfn >= mslots[i - 1].base_gfn) {
703 mslots[i] = mslots[i - 1];
704 slots->id_to_index[mslots[i].id] = i;
710 slots->id_to_index[mslots[i].id] = i;
713 static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
715 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
717 #ifdef __KVM_HAVE_READONLY_MEM
718 valid_flags |= KVM_MEM_READONLY;
721 if (mem->flags & ~valid_flags)
727 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
728 struct kvm_memslots *slots)
730 struct kvm_memslots *old_memslots = kvm->memslots;
733 * Set the low bit in the generation, which disables SPTE caching
734 * until the end of synchronize_srcu_expedited.
736 WARN_ON(old_memslots->generation & 1);
737 slots->generation = old_memslots->generation + 1;
739 rcu_assign_pointer(kvm->memslots, slots);
740 synchronize_srcu_expedited(&kvm->srcu);
743 * Increment the new memslot generation a second time. This prevents
744 * vm exits that race with memslot updates from caching a memslot
745 * generation that will (potentially) be valid forever.
749 kvm_arch_memslots_updated(kvm);
755 * Allocate some memory and give it an address in the guest physical address
758 * Discontiguous memory is allowed, mostly for framebuffers.
760 * Must be called holding kvm->slots_lock for write.
762 int __kvm_set_memory_region(struct kvm *kvm,
763 struct kvm_userspace_memory_region *mem)
767 unsigned long npages;
768 struct kvm_memory_slot *slot;
769 struct kvm_memory_slot old, new;
770 struct kvm_memslots *slots = NULL, *old_memslots;
771 enum kvm_mr_change change;
773 r = check_memory_region_flags(mem);
778 /* General sanity checks */
779 if (mem->memory_size & (PAGE_SIZE - 1))
781 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
783 /* We can read the guest memory with __xxx_user() later on. */
784 if ((mem->slot < KVM_USER_MEM_SLOTS) &&
785 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
786 !access_ok(VERIFY_WRITE,
787 (void __user *)(unsigned long)mem->userspace_addr,
790 if (mem->slot >= KVM_MEM_SLOTS_NUM)
792 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
795 slot = id_to_memslot(kvm->memslots, mem->slot);
796 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
797 npages = mem->memory_size >> PAGE_SHIFT;
799 if (npages > KVM_MEM_MAX_NR_PAGES)
803 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
808 new.base_gfn = base_gfn;
810 new.flags = mem->flags;
814 change = KVM_MR_CREATE;
815 else { /* Modify an existing slot. */
816 if ((mem->userspace_addr != old.userspace_addr) ||
817 (npages != old.npages) ||
818 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
821 if (base_gfn != old.base_gfn)
822 change = KVM_MR_MOVE;
823 else if (new.flags != old.flags)
824 change = KVM_MR_FLAGS_ONLY;
825 else { /* Nothing to change. */
830 } else if (old.npages) {
831 change = KVM_MR_DELETE;
832 } else /* Modify a non-existent slot: disallowed. */
835 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
836 /* Check for overlaps */
838 kvm_for_each_memslot(slot, kvm->memslots) {
839 if ((slot->id >= KVM_USER_MEM_SLOTS) ||
840 (slot->id == mem->slot))
842 if (!((base_gfn + npages <= slot->base_gfn) ||
843 (base_gfn >= slot->base_gfn + slot->npages)))
848 /* Free page dirty bitmap if unneeded */
849 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
850 new.dirty_bitmap = NULL;
853 if (change == KVM_MR_CREATE) {
854 new.userspace_addr = mem->userspace_addr;
856 if (kvm_arch_create_memslot(kvm, &new, npages))
860 /* Allocate page dirty bitmap if needed */
861 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
862 if (kvm_create_dirty_bitmap(&new) < 0)
866 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
871 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
872 slot = id_to_memslot(slots, mem->slot);
873 slot->flags |= KVM_MEMSLOT_INVALID;
875 old_memslots = install_new_memslots(kvm, slots);
877 /* slot was deleted or moved, clear iommu mapping */
878 kvm_iommu_unmap_pages(kvm, &old);
879 /* From this point no new shadow pages pointing to a deleted,
880 * or moved, memslot will be created.
882 * validation of sp->gfn happens in:
883 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
884 * - kvm_is_visible_gfn (mmu_check_roots)
886 kvm_arch_flush_shadow_memslot(kvm, slot);
889 * We can re-use the old_memslots from above, the only difference
890 * from the currently installed memslots is the invalid flag. This
891 * will get overwritten by update_memslots anyway.
893 slots = old_memslots;
896 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
900 /* actual memory is freed via old in kvm_free_physmem_slot below */
901 if (change == KVM_MR_DELETE) {
902 new.dirty_bitmap = NULL;
903 memset(&new.arch, 0, sizeof(new.arch));
906 update_memslots(slots, &new);
907 old_memslots = install_new_memslots(kvm, slots);
909 kvm_arch_commit_memory_region(kvm, mem, &old, change);
911 kvm_free_physmem_slot(kvm, &old, &new);
915 * IOMMU mapping: New slots need to be mapped. Old slots need to be
916 * un-mapped and re-mapped if their base changes. Since base change
917 * unmapping is handled above with slot deletion, mapping alone is
918 * needed here. Anything else the iommu might care about for existing
919 * slots (size changes, userspace addr changes and read-only flag
920 * changes) is disallowed above, so any other attribute changes getting
921 * here can be skipped.
923 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
924 r = kvm_iommu_map_pages(kvm, &new);
933 kvm_free_physmem_slot(kvm, &new, &old);
937 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
939 int kvm_set_memory_region(struct kvm *kvm,
940 struct kvm_userspace_memory_region *mem)
944 mutex_lock(&kvm->slots_lock);
945 r = __kvm_set_memory_region(kvm, mem);
946 mutex_unlock(&kvm->slots_lock);
949 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
951 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
952 struct kvm_userspace_memory_region *mem)
954 if (mem->slot >= KVM_USER_MEM_SLOTS)
956 return kvm_set_memory_region(kvm, mem);
959 int kvm_get_dirty_log(struct kvm *kvm,
960 struct kvm_dirty_log *log, int *is_dirty)
962 struct kvm_memory_slot *memslot;
965 unsigned long any = 0;
968 if (log->slot >= KVM_USER_MEM_SLOTS)
971 memslot = id_to_memslot(kvm->memslots, log->slot);
973 if (!memslot->dirty_bitmap)
976 n = kvm_dirty_bitmap_bytes(memslot);
978 for (i = 0; !any && i < n/sizeof(long); ++i)
979 any = memslot->dirty_bitmap[i];
982 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
992 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
994 bool kvm_largepages_enabled(void)
996 return largepages_enabled;
999 void kvm_disable_largepages(void)
1001 largepages_enabled = false;
1003 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1005 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1007 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1009 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1011 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1013 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1015 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1016 memslot->flags & KVM_MEMSLOT_INVALID)
1021 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1023 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1025 struct vm_area_struct *vma;
1026 unsigned long addr, size;
1030 addr = gfn_to_hva(kvm, gfn);
1031 if (kvm_is_error_hva(addr))
1034 down_read(¤t->mm->mmap_sem);
1035 vma = find_vma(current->mm, addr);
1039 size = vma_kernel_pagesize(vma);
1042 up_read(¤t->mm->mmap_sem);
1047 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1049 return slot->flags & KVM_MEM_READONLY;
1052 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1053 gfn_t *nr_pages, bool write)
1055 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1056 return KVM_HVA_ERR_BAD;
1058 if (memslot_is_readonly(slot) && write)
1059 return KVM_HVA_ERR_RO_BAD;
1062 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1064 return __gfn_to_hva_memslot(slot, gfn);
1067 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1070 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1073 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1076 return gfn_to_hva_many(slot, gfn, NULL);
1078 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1080 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1082 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1084 EXPORT_SYMBOL_GPL(gfn_to_hva);
1087 * If writable is set to false, the hva returned by this function is only
1088 * allowed to be read.
1090 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1091 gfn_t gfn, bool *writable)
1093 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1095 if (!kvm_is_error_hva(hva) && writable)
1096 *writable = !memslot_is_readonly(slot);
1101 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1103 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1105 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1108 static int kvm_read_hva(void *data, void __user *hva, int len)
1110 return __copy_from_user(data, hva, len);
1113 static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1115 return __copy_from_user_inatomic(data, hva, len);
1118 static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1119 unsigned long start, int write, struct page **page)
1121 int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1124 flags |= FOLL_WRITE;
1126 return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1129 int kvm_get_user_page_io(struct task_struct *tsk, struct mm_struct *mm,
1130 unsigned long addr, bool write_fault,
1131 struct page **pagep)
1135 int flags = FOLL_TOUCH | FOLL_HWPOISON |
1136 (pagep ? FOLL_GET : 0) |
1137 (write_fault ? FOLL_WRITE : 0);
1140 * If retrying the fault, we get here *not* having allowed the filemap
1141 * to wait on the page lock. We should now allow waiting on the IO with
1142 * the mmap semaphore released.
1144 down_read(&mm->mmap_sem);
1145 npages = __get_user_pages(tsk, mm, addr, 1, flags, pagep, NULL,
1154 * The previous call has now waited on the IO. Now we can
1155 * retry and complete. Pass TRIED to ensure we do not re
1156 * schedule async IO (see e.g. filemap_fault).
1158 down_read(&mm->mmap_sem);
1159 npages = __get_user_pages(tsk, mm, addr, 1, flags | FOLL_TRIED,
1162 up_read(&mm->mmap_sem);
1166 static inline int check_user_page_hwpoison(unsigned long addr)
1168 int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1170 rc = __get_user_pages(current, current->mm, addr, 1,
1171 flags, NULL, NULL, NULL);
1172 return rc == -EHWPOISON;
1176 * The atomic path to get the writable pfn which will be stored in @pfn,
1177 * true indicates success, otherwise false is returned.
1179 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1180 bool write_fault, bool *writable, pfn_t *pfn)
1182 struct page *page[1];
1185 if (!(async || atomic))
1189 * Fast pin a writable pfn only if it is a write fault request
1190 * or the caller allows to map a writable pfn for a read fault
1193 if (!(write_fault || writable))
1196 npages = __get_user_pages_fast(addr, 1, 1, page);
1198 *pfn = page_to_pfn(page[0]);
1209 * The slow path to get the pfn of the specified host virtual address,
1210 * 1 indicates success, -errno is returned if error is detected.
1212 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1213 bool *writable, pfn_t *pfn)
1215 struct page *page[1];
1221 *writable = write_fault;
1224 down_read(¤t->mm->mmap_sem);
1225 npages = get_user_page_nowait(current, current->mm,
1226 addr, write_fault, page);
1227 up_read(¤t->mm->mmap_sem);
1230 * By now we have tried gup_fast, and possibly async_pf, and we
1231 * are certainly not atomic. Time to retry the gup, allowing
1232 * mmap semaphore to be relinquished in the case of IO.
1234 npages = kvm_get_user_page_io(current, current->mm, addr,
1240 /* map read fault as writable if possible */
1241 if (unlikely(!write_fault) && writable) {
1242 struct page *wpage[1];
1244 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1253 *pfn = page_to_pfn(page[0]);
1257 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1259 if (unlikely(!(vma->vm_flags & VM_READ)))
1262 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1269 * Pin guest page in memory and return its pfn.
1270 * @addr: host virtual address which maps memory to the guest
1271 * @atomic: whether this function can sleep
1272 * @async: whether this function need to wait IO complete if the
1273 * host page is not in the memory
1274 * @write_fault: whether we should get a writable host page
1275 * @writable: whether it allows to map a writable host page for !@write_fault
1277 * The function will map a writable host page for these two cases:
1278 * 1): @write_fault = true
1279 * 2): @write_fault = false && @writable, @writable will tell the caller
1280 * whether the mapping is writable.
1282 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1283 bool write_fault, bool *writable)
1285 struct vm_area_struct *vma;
1289 /* we can do it either atomically or asynchronously, not both */
1290 BUG_ON(atomic && async);
1292 if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1296 return KVM_PFN_ERR_FAULT;
1298 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1302 down_read(¤t->mm->mmap_sem);
1303 if (npages == -EHWPOISON ||
1304 (!async && check_user_page_hwpoison(addr))) {
1305 pfn = KVM_PFN_ERR_HWPOISON;
1309 vma = find_vma_intersection(current->mm, addr, addr + 1);
1312 pfn = KVM_PFN_ERR_FAULT;
1313 else if ((vma->vm_flags & VM_PFNMAP)) {
1314 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1316 BUG_ON(!kvm_is_reserved_pfn(pfn));
1318 if (async && vma_is_valid(vma, write_fault))
1320 pfn = KVM_PFN_ERR_FAULT;
1323 up_read(¤t->mm->mmap_sem);
1328 __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1329 bool *async, bool write_fault, bool *writable)
1331 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1333 if (addr == KVM_HVA_ERR_RO_BAD)
1334 return KVM_PFN_ERR_RO_FAULT;
1336 if (kvm_is_error_hva(addr))
1337 return KVM_PFN_NOSLOT;
1339 /* Do not map writable pfn in the readonly memslot. */
1340 if (writable && memslot_is_readonly(slot)) {
1345 return hva_to_pfn(addr, atomic, async, write_fault,
1349 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1350 bool write_fault, bool *writable)
1352 struct kvm_memory_slot *slot;
1357 slot = gfn_to_memslot(kvm, gfn);
1359 return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1363 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1365 return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1367 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1369 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1370 bool write_fault, bool *writable)
1372 return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1374 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1376 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1378 return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1380 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1382 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1385 return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1387 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1389 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1391 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1394 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1396 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1398 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1400 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1406 addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1407 if (kvm_is_error_hva(addr))
1410 if (entry < nr_pages)
1413 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1415 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1417 static struct page *kvm_pfn_to_page(pfn_t pfn)
1419 if (is_error_noslot_pfn(pfn))
1420 return KVM_ERR_PTR_BAD_PAGE;
1422 if (kvm_is_reserved_pfn(pfn)) {
1424 return KVM_ERR_PTR_BAD_PAGE;
1427 return pfn_to_page(pfn);
1430 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1434 pfn = gfn_to_pfn(kvm, gfn);
1436 return kvm_pfn_to_page(pfn);
1439 EXPORT_SYMBOL_GPL(gfn_to_page);
1441 void kvm_release_page_clean(struct page *page)
1443 WARN_ON(is_error_page(page));
1445 kvm_release_pfn_clean(page_to_pfn(page));
1447 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1449 void kvm_release_pfn_clean(pfn_t pfn)
1451 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
1452 put_page(pfn_to_page(pfn));
1454 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1456 void kvm_release_page_dirty(struct page *page)
1458 WARN_ON(is_error_page(page));
1460 kvm_release_pfn_dirty(page_to_pfn(page));
1462 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1464 static void kvm_release_pfn_dirty(pfn_t pfn)
1466 kvm_set_pfn_dirty(pfn);
1467 kvm_release_pfn_clean(pfn);
1470 void kvm_set_pfn_dirty(pfn_t pfn)
1472 if (!kvm_is_reserved_pfn(pfn)) {
1473 struct page *page = pfn_to_page(pfn);
1474 if (!PageReserved(page))
1478 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1480 void kvm_set_pfn_accessed(pfn_t pfn)
1482 if (!kvm_is_reserved_pfn(pfn))
1483 mark_page_accessed(pfn_to_page(pfn));
1485 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1487 void kvm_get_pfn(pfn_t pfn)
1489 if (!kvm_is_reserved_pfn(pfn))
1490 get_page(pfn_to_page(pfn));
1492 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1494 static int next_segment(unsigned long len, int offset)
1496 if (len > PAGE_SIZE - offset)
1497 return PAGE_SIZE - offset;
1502 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1508 addr = gfn_to_hva_prot(kvm, gfn, NULL);
1509 if (kvm_is_error_hva(addr))
1511 r = kvm_read_hva(data, (void __user *)addr + offset, len);
1516 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1518 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1520 gfn_t gfn = gpa >> PAGE_SHIFT;
1522 int offset = offset_in_page(gpa);
1525 while ((seg = next_segment(len, offset)) != 0) {
1526 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1536 EXPORT_SYMBOL_GPL(kvm_read_guest);
1538 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1543 gfn_t gfn = gpa >> PAGE_SHIFT;
1544 int offset = offset_in_page(gpa);
1546 addr = gfn_to_hva_prot(kvm, gfn, NULL);
1547 if (kvm_is_error_hva(addr))
1549 pagefault_disable();
1550 r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
1556 EXPORT_SYMBOL(kvm_read_guest_atomic);
1558 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1559 int offset, int len)
1564 addr = gfn_to_hva(kvm, gfn);
1565 if (kvm_is_error_hva(addr))
1567 r = __copy_to_user((void __user *)addr + offset, data, len);
1570 mark_page_dirty(kvm, gfn);
1573 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1575 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1578 gfn_t gfn = gpa >> PAGE_SHIFT;
1580 int offset = offset_in_page(gpa);
1583 while ((seg = next_segment(len, offset)) != 0) {
1584 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1595 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1596 gpa_t gpa, unsigned long len)
1598 struct kvm_memslots *slots = kvm_memslots(kvm);
1599 int offset = offset_in_page(gpa);
1600 gfn_t start_gfn = gpa >> PAGE_SHIFT;
1601 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1602 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1603 gfn_t nr_pages_avail;
1606 ghc->generation = slots->generation;
1608 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1609 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, &nr_pages_avail);
1610 if (!kvm_is_error_hva(ghc->hva) && nr_pages_avail >= nr_pages_needed) {
1614 * If the requested region crosses two memslots, we still
1615 * verify that the entire region is valid here.
1617 while (start_gfn <= end_gfn) {
1618 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1619 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1621 if (kvm_is_error_hva(ghc->hva))
1623 start_gfn += nr_pages_avail;
1625 /* Use the slow path for cross page reads and writes. */
1626 ghc->memslot = NULL;
1630 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1632 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1633 void *data, unsigned long len)
1635 struct kvm_memslots *slots = kvm_memslots(kvm);
1638 BUG_ON(len > ghc->len);
1640 if (slots->generation != ghc->generation)
1641 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1643 if (unlikely(!ghc->memslot))
1644 return kvm_write_guest(kvm, ghc->gpa, data, len);
1646 if (kvm_is_error_hva(ghc->hva))
1649 r = __copy_to_user((void __user *)ghc->hva, data, len);
1652 mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1656 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1658 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1659 void *data, unsigned long len)
1661 struct kvm_memslots *slots = kvm_memslots(kvm);
1664 BUG_ON(len > ghc->len);
1666 if (slots->generation != ghc->generation)
1667 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1669 if (unlikely(!ghc->memslot))
1670 return kvm_read_guest(kvm, ghc->gpa, data, len);
1672 if (kvm_is_error_hva(ghc->hva))
1675 r = __copy_from_user(data, (void __user *)ghc->hva, len);
1681 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1683 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1685 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
1687 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
1689 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1691 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1693 gfn_t gfn = gpa >> PAGE_SHIFT;
1695 int offset = offset_in_page(gpa);
1698 while ((seg = next_segment(len, offset)) != 0) {
1699 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1708 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1710 static void mark_page_dirty_in_slot(struct kvm *kvm,
1711 struct kvm_memory_slot *memslot,
1714 if (memslot && memslot->dirty_bitmap) {
1715 unsigned long rel_gfn = gfn - memslot->base_gfn;
1717 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1721 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1723 struct kvm_memory_slot *memslot;
1725 memslot = gfn_to_memslot(kvm, gfn);
1726 mark_page_dirty_in_slot(kvm, memslot, gfn);
1728 EXPORT_SYMBOL_GPL(mark_page_dirty);
1731 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1733 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1738 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1740 if (kvm_arch_vcpu_runnable(vcpu)) {
1741 kvm_make_request(KVM_REQ_UNHALT, vcpu);
1744 if (kvm_cpu_has_pending_timer(vcpu))
1746 if (signal_pending(current))
1752 finish_wait(&vcpu->wq, &wait);
1754 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
1758 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1760 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1763 int cpu = vcpu->cpu;
1764 wait_queue_head_t *wqp;
1766 wqp = kvm_arch_vcpu_wq(vcpu);
1767 if (waitqueue_active(wqp)) {
1768 wake_up_interruptible(wqp);
1769 ++vcpu->stat.halt_wakeup;
1773 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1774 if (kvm_arch_vcpu_should_kick(vcpu))
1775 smp_send_reschedule(cpu);
1778 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
1779 #endif /* !CONFIG_S390 */
1781 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
1784 struct task_struct *task = NULL;
1788 pid = rcu_dereference(target->pid);
1790 task = get_pid_task(pid, PIDTYPE_PID);
1794 ret = yield_to(task, 1);
1795 put_task_struct(task);
1799 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1802 * Helper that checks whether a VCPU is eligible for directed yield.
1803 * Most eligible candidate to yield is decided by following heuristics:
1805 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1806 * (preempted lock holder), indicated by @in_spin_loop.
1807 * Set at the beiginning and cleared at the end of interception/PLE handler.
1809 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1810 * chance last time (mostly it has become eligible now since we have probably
1811 * yielded to lockholder in last iteration. This is done by toggling
1812 * @dy_eligible each time a VCPU checked for eligibility.)
1814 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1815 * to preempted lock-holder could result in wrong VCPU selection and CPU
1816 * burning. Giving priority for a potential lock-holder increases lock
1819 * Since algorithm is based on heuristics, accessing another VCPU data without
1820 * locking does not harm. It may result in trying to yield to same VCPU, fail
1821 * and continue with next VCPU and so on.
1823 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1825 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1828 eligible = !vcpu->spin_loop.in_spin_loop ||
1829 vcpu->spin_loop.dy_eligible;
1831 if (vcpu->spin_loop.in_spin_loop)
1832 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1840 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1842 struct kvm *kvm = me->kvm;
1843 struct kvm_vcpu *vcpu;
1844 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1850 kvm_vcpu_set_in_spin_loop(me, true);
1852 * We boost the priority of a VCPU that is runnable but not
1853 * currently running, because it got preempted by something
1854 * else and called schedule in __vcpu_run. Hopefully that
1855 * VCPU is holding the lock that we need and will release it.
1856 * We approximate round-robin by starting at the last boosted VCPU.
1858 for (pass = 0; pass < 2 && !yielded && try; pass++) {
1859 kvm_for_each_vcpu(i, vcpu, kvm) {
1860 if (!pass && i <= last_boosted_vcpu) {
1861 i = last_boosted_vcpu;
1863 } else if (pass && i > last_boosted_vcpu)
1865 if (!ACCESS_ONCE(vcpu->preempted))
1869 if (waitqueue_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
1871 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1874 yielded = kvm_vcpu_yield_to(vcpu);
1876 kvm->last_boosted_vcpu = i;
1878 } else if (yielded < 0) {
1885 kvm_vcpu_set_in_spin_loop(me, false);
1887 /* Ensure vcpu is not eligible during next spinloop */
1888 kvm_vcpu_set_dy_eligible(me, false);
1890 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1892 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1894 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1897 if (vmf->pgoff == 0)
1898 page = virt_to_page(vcpu->run);
1900 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1901 page = virt_to_page(vcpu->arch.pio_data);
1903 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1904 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1905 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1908 return kvm_arch_vcpu_fault(vcpu, vmf);
1914 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1915 .fault = kvm_vcpu_fault,
1918 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1920 vma->vm_ops = &kvm_vcpu_vm_ops;
1924 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1926 struct kvm_vcpu *vcpu = filp->private_data;
1928 kvm_put_kvm(vcpu->kvm);
1932 static struct file_operations kvm_vcpu_fops = {
1933 .release = kvm_vcpu_release,
1934 .unlocked_ioctl = kvm_vcpu_ioctl,
1935 #ifdef CONFIG_COMPAT
1936 .compat_ioctl = kvm_vcpu_compat_ioctl,
1938 .mmap = kvm_vcpu_mmap,
1939 .llseek = noop_llseek,
1943 * Allocates an inode for the vcpu.
1945 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1947 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
1951 * Creates some virtual cpus. Good luck creating more than one.
1953 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1956 struct kvm_vcpu *vcpu, *v;
1958 if (id >= KVM_MAX_VCPUS)
1961 vcpu = kvm_arch_vcpu_create(kvm, id);
1963 return PTR_ERR(vcpu);
1965 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1967 r = kvm_arch_vcpu_setup(vcpu);
1971 mutex_lock(&kvm->lock);
1972 if (!kvm_vcpu_compatible(vcpu)) {
1974 goto unlock_vcpu_destroy;
1976 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1978 goto unlock_vcpu_destroy;
1981 kvm_for_each_vcpu(r, v, kvm)
1982 if (v->vcpu_id == id) {
1984 goto unlock_vcpu_destroy;
1987 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1989 /* Now it's all set up, let userspace reach it */
1991 r = create_vcpu_fd(vcpu);
1994 goto unlock_vcpu_destroy;
1997 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1999 atomic_inc(&kvm->online_vcpus);
2001 mutex_unlock(&kvm->lock);
2002 kvm_arch_vcpu_postcreate(vcpu);
2005 unlock_vcpu_destroy:
2006 mutex_unlock(&kvm->lock);
2008 kvm_arch_vcpu_destroy(vcpu);
2012 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2015 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2016 vcpu->sigset_active = 1;
2017 vcpu->sigset = *sigset;
2019 vcpu->sigset_active = 0;
2023 static long kvm_vcpu_ioctl(struct file *filp,
2024 unsigned int ioctl, unsigned long arg)
2026 struct kvm_vcpu *vcpu = filp->private_data;
2027 void __user *argp = (void __user *)arg;
2029 struct kvm_fpu *fpu = NULL;
2030 struct kvm_sregs *kvm_sregs = NULL;
2032 if (vcpu->kvm->mm != current->mm)
2035 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2038 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
2040 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
2041 * so vcpu_load() would break it.
2043 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
2044 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2048 r = vcpu_load(vcpu);
2056 if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
2057 /* The thread running this VCPU changed. */
2058 struct pid *oldpid = vcpu->pid;
2059 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
2060 rcu_assign_pointer(vcpu->pid, newpid);
2065 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2066 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2068 case KVM_GET_REGS: {
2069 struct kvm_regs *kvm_regs;
2072 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2075 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2079 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2086 case KVM_SET_REGS: {
2087 struct kvm_regs *kvm_regs;
2090 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2091 if (IS_ERR(kvm_regs)) {
2092 r = PTR_ERR(kvm_regs);
2095 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2099 case KVM_GET_SREGS: {
2100 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2104 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2108 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2113 case KVM_SET_SREGS: {
2114 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2115 if (IS_ERR(kvm_sregs)) {
2116 r = PTR_ERR(kvm_sregs);
2120 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2123 case KVM_GET_MP_STATE: {
2124 struct kvm_mp_state mp_state;
2126 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2130 if (copy_to_user(argp, &mp_state, sizeof mp_state))
2135 case KVM_SET_MP_STATE: {
2136 struct kvm_mp_state mp_state;
2139 if (copy_from_user(&mp_state, argp, sizeof mp_state))
2141 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2144 case KVM_TRANSLATE: {
2145 struct kvm_translation tr;
2148 if (copy_from_user(&tr, argp, sizeof tr))
2150 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2154 if (copy_to_user(argp, &tr, sizeof tr))
2159 case KVM_SET_GUEST_DEBUG: {
2160 struct kvm_guest_debug dbg;
2163 if (copy_from_user(&dbg, argp, sizeof dbg))
2165 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2168 case KVM_SET_SIGNAL_MASK: {
2169 struct kvm_signal_mask __user *sigmask_arg = argp;
2170 struct kvm_signal_mask kvm_sigmask;
2171 sigset_t sigset, *p;
2176 if (copy_from_user(&kvm_sigmask, argp,
2177 sizeof kvm_sigmask))
2180 if (kvm_sigmask.len != sizeof sigset)
2183 if (copy_from_user(&sigset, sigmask_arg->sigset,
2188 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2192 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2196 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2200 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2206 fpu = memdup_user(argp, sizeof(*fpu));
2212 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2216 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2225 #ifdef CONFIG_COMPAT
2226 static long kvm_vcpu_compat_ioctl(struct file *filp,
2227 unsigned int ioctl, unsigned long arg)
2229 struct kvm_vcpu *vcpu = filp->private_data;
2230 void __user *argp = compat_ptr(arg);
2233 if (vcpu->kvm->mm != current->mm)
2237 case KVM_SET_SIGNAL_MASK: {
2238 struct kvm_signal_mask __user *sigmask_arg = argp;
2239 struct kvm_signal_mask kvm_sigmask;
2240 compat_sigset_t csigset;
2245 if (copy_from_user(&kvm_sigmask, argp,
2246 sizeof kvm_sigmask))
2249 if (kvm_sigmask.len != sizeof csigset)
2252 if (copy_from_user(&csigset, sigmask_arg->sigset,
2255 sigset_from_compat(&sigset, &csigset);
2256 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2258 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2262 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2270 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2271 int (*accessor)(struct kvm_device *dev,
2272 struct kvm_device_attr *attr),
2275 struct kvm_device_attr attr;
2280 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2283 return accessor(dev, &attr);
2286 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2289 struct kvm_device *dev = filp->private_data;
2292 case KVM_SET_DEVICE_ATTR:
2293 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2294 case KVM_GET_DEVICE_ATTR:
2295 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2296 case KVM_HAS_DEVICE_ATTR:
2297 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2299 if (dev->ops->ioctl)
2300 return dev->ops->ioctl(dev, ioctl, arg);
2306 static int kvm_device_release(struct inode *inode, struct file *filp)
2308 struct kvm_device *dev = filp->private_data;
2309 struct kvm *kvm = dev->kvm;
2315 static const struct file_operations kvm_device_fops = {
2316 .unlocked_ioctl = kvm_device_ioctl,
2317 #ifdef CONFIG_COMPAT
2318 .compat_ioctl = kvm_device_ioctl,
2320 .release = kvm_device_release,
2323 struct kvm_device *kvm_device_from_filp(struct file *filp)
2325 if (filp->f_op != &kvm_device_fops)
2328 return filp->private_data;
2331 static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
2332 #ifdef CONFIG_KVM_MPIC
2333 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
2334 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
2337 #ifdef CONFIG_KVM_XICS
2338 [KVM_DEV_TYPE_XICS] = &kvm_xics_ops,
2342 int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
2344 if (type >= ARRAY_SIZE(kvm_device_ops_table))
2347 if (kvm_device_ops_table[type] != NULL)
2350 kvm_device_ops_table[type] = ops;
2354 void kvm_unregister_device_ops(u32 type)
2356 if (kvm_device_ops_table[type] != NULL)
2357 kvm_device_ops_table[type] = NULL;
2360 static int kvm_ioctl_create_device(struct kvm *kvm,
2361 struct kvm_create_device *cd)
2363 struct kvm_device_ops *ops = NULL;
2364 struct kvm_device *dev;
2365 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2368 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
2371 ops = kvm_device_ops_table[cd->type];
2378 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2385 ret = ops->create(dev, cd->type);
2391 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2397 list_add(&dev->vm_node, &kvm->devices);
2403 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2406 case KVM_CAP_USER_MEMORY:
2407 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2408 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2409 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2410 case KVM_CAP_SET_BOOT_CPU_ID:
2412 case KVM_CAP_INTERNAL_ERROR_DATA:
2413 #ifdef CONFIG_HAVE_KVM_MSI
2414 case KVM_CAP_SIGNAL_MSI:
2416 #ifdef CONFIG_HAVE_KVM_IRQFD
2417 case KVM_CAP_IRQFD_RESAMPLE:
2419 case KVM_CAP_CHECK_EXTENSION_VM:
2421 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2422 case KVM_CAP_IRQ_ROUTING:
2423 return KVM_MAX_IRQ_ROUTES;
2428 return kvm_vm_ioctl_check_extension(kvm, arg);
2431 static long kvm_vm_ioctl(struct file *filp,
2432 unsigned int ioctl, unsigned long arg)
2434 struct kvm *kvm = filp->private_data;
2435 void __user *argp = (void __user *)arg;
2438 if (kvm->mm != current->mm)
2441 case KVM_CREATE_VCPU:
2442 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2444 case KVM_SET_USER_MEMORY_REGION: {
2445 struct kvm_userspace_memory_region kvm_userspace_mem;
2448 if (copy_from_user(&kvm_userspace_mem, argp,
2449 sizeof kvm_userspace_mem))
2452 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2455 case KVM_GET_DIRTY_LOG: {
2456 struct kvm_dirty_log log;
2459 if (copy_from_user(&log, argp, sizeof log))
2461 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2464 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2465 case KVM_REGISTER_COALESCED_MMIO: {
2466 struct kvm_coalesced_mmio_zone zone;
2468 if (copy_from_user(&zone, argp, sizeof zone))
2470 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2473 case KVM_UNREGISTER_COALESCED_MMIO: {
2474 struct kvm_coalesced_mmio_zone zone;
2476 if (copy_from_user(&zone, argp, sizeof zone))
2478 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2483 struct kvm_irqfd data;
2486 if (copy_from_user(&data, argp, sizeof data))
2488 r = kvm_irqfd(kvm, &data);
2491 case KVM_IOEVENTFD: {
2492 struct kvm_ioeventfd data;
2495 if (copy_from_user(&data, argp, sizeof data))
2497 r = kvm_ioeventfd(kvm, &data);
2500 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2501 case KVM_SET_BOOT_CPU_ID:
2503 mutex_lock(&kvm->lock);
2504 if (atomic_read(&kvm->online_vcpus) != 0)
2507 kvm->bsp_vcpu_id = arg;
2508 mutex_unlock(&kvm->lock);
2511 #ifdef CONFIG_HAVE_KVM_MSI
2512 case KVM_SIGNAL_MSI: {
2516 if (copy_from_user(&msi, argp, sizeof msi))
2518 r = kvm_send_userspace_msi(kvm, &msi);
2522 #ifdef __KVM_HAVE_IRQ_LINE
2523 case KVM_IRQ_LINE_STATUS:
2524 case KVM_IRQ_LINE: {
2525 struct kvm_irq_level irq_event;
2528 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2531 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
2532 ioctl == KVM_IRQ_LINE_STATUS);
2537 if (ioctl == KVM_IRQ_LINE_STATUS) {
2538 if (copy_to_user(argp, &irq_event, sizeof irq_event))
2546 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2547 case KVM_SET_GSI_ROUTING: {
2548 struct kvm_irq_routing routing;
2549 struct kvm_irq_routing __user *urouting;
2550 struct kvm_irq_routing_entry *entries;
2553 if (copy_from_user(&routing, argp, sizeof(routing)))
2556 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2561 entries = vmalloc(routing.nr * sizeof(*entries));
2566 if (copy_from_user(entries, urouting->entries,
2567 routing.nr * sizeof(*entries)))
2568 goto out_free_irq_routing;
2569 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2571 out_free_irq_routing:
2575 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
2576 case KVM_CREATE_DEVICE: {
2577 struct kvm_create_device cd;
2580 if (copy_from_user(&cd, argp, sizeof(cd)))
2583 r = kvm_ioctl_create_device(kvm, &cd);
2588 if (copy_to_user(argp, &cd, sizeof(cd)))
2594 case KVM_CHECK_EXTENSION:
2595 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
2598 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2604 #ifdef CONFIG_COMPAT
2605 struct compat_kvm_dirty_log {
2609 compat_uptr_t dirty_bitmap; /* one bit per page */
2614 static long kvm_vm_compat_ioctl(struct file *filp,
2615 unsigned int ioctl, unsigned long arg)
2617 struct kvm *kvm = filp->private_data;
2620 if (kvm->mm != current->mm)
2623 case KVM_GET_DIRTY_LOG: {
2624 struct compat_kvm_dirty_log compat_log;
2625 struct kvm_dirty_log log;
2628 if (copy_from_user(&compat_log, (void __user *)arg,
2629 sizeof(compat_log)))
2631 log.slot = compat_log.slot;
2632 log.padding1 = compat_log.padding1;
2633 log.padding2 = compat_log.padding2;
2634 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2636 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2640 r = kvm_vm_ioctl(filp, ioctl, arg);
2648 static struct file_operations kvm_vm_fops = {
2649 .release = kvm_vm_release,
2650 .unlocked_ioctl = kvm_vm_ioctl,
2651 #ifdef CONFIG_COMPAT
2652 .compat_ioctl = kvm_vm_compat_ioctl,
2654 .llseek = noop_llseek,
2657 static int kvm_dev_ioctl_create_vm(unsigned long type)
2662 kvm = kvm_create_vm(type);
2664 return PTR_ERR(kvm);
2665 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2666 r = kvm_coalesced_mmio_init(kvm);
2672 r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC);
2679 static long kvm_dev_ioctl(struct file *filp,
2680 unsigned int ioctl, unsigned long arg)
2685 case KVM_GET_API_VERSION:
2688 r = KVM_API_VERSION;
2691 r = kvm_dev_ioctl_create_vm(arg);
2693 case KVM_CHECK_EXTENSION:
2694 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
2696 case KVM_GET_VCPU_MMAP_SIZE:
2699 r = PAGE_SIZE; /* struct kvm_run */
2701 r += PAGE_SIZE; /* pio data page */
2703 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2704 r += PAGE_SIZE; /* coalesced mmio ring page */
2707 case KVM_TRACE_ENABLE:
2708 case KVM_TRACE_PAUSE:
2709 case KVM_TRACE_DISABLE:
2713 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2719 static struct file_operations kvm_chardev_ops = {
2720 .unlocked_ioctl = kvm_dev_ioctl,
2721 .compat_ioctl = kvm_dev_ioctl,
2722 .llseek = noop_llseek,
2725 static struct miscdevice kvm_dev = {
2731 static void hardware_enable_nolock(void *junk)
2733 int cpu = raw_smp_processor_id();
2736 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2739 cpumask_set_cpu(cpu, cpus_hardware_enabled);
2741 r = kvm_arch_hardware_enable();
2744 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2745 atomic_inc(&hardware_enable_failed);
2746 printk(KERN_INFO "kvm: enabling virtualization on "
2747 "CPU%d failed\n", cpu);
2751 static void hardware_enable(void)
2753 raw_spin_lock(&kvm_count_lock);
2754 if (kvm_usage_count)
2755 hardware_enable_nolock(NULL);
2756 raw_spin_unlock(&kvm_count_lock);
2759 static void hardware_disable_nolock(void *junk)
2761 int cpu = raw_smp_processor_id();
2763 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2765 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2766 kvm_arch_hardware_disable();
2769 static void hardware_disable(void)
2771 raw_spin_lock(&kvm_count_lock);
2772 if (kvm_usage_count)
2773 hardware_disable_nolock(NULL);
2774 raw_spin_unlock(&kvm_count_lock);
2777 static void hardware_disable_all_nolock(void)
2779 BUG_ON(!kvm_usage_count);
2782 if (!kvm_usage_count)
2783 on_each_cpu(hardware_disable_nolock, NULL, 1);
2786 static void hardware_disable_all(void)
2788 raw_spin_lock(&kvm_count_lock);
2789 hardware_disable_all_nolock();
2790 raw_spin_unlock(&kvm_count_lock);
2793 static int hardware_enable_all(void)
2797 raw_spin_lock(&kvm_count_lock);
2800 if (kvm_usage_count == 1) {
2801 atomic_set(&hardware_enable_failed, 0);
2802 on_each_cpu(hardware_enable_nolock, NULL, 1);
2804 if (atomic_read(&hardware_enable_failed)) {
2805 hardware_disable_all_nolock();
2810 raw_spin_unlock(&kvm_count_lock);
2815 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2820 val &= ~CPU_TASKS_FROZEN;
2823 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2828 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2836 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2840 * Some (well, at least mine) BIOSes hang on reboot if
2843 * And Intel TXT required VMX off for all cpu when system shutdown.
2845 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2846 kvm_rebooting = true;
2847 on_each_cpu(hardware_disable_nolock, NULL, 1);
2851 static struct notifier_block kvm_reboot_notifier = {
2852 .notifier_call = kvm_reboot,
2856 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2860 for (i = 0; i < bus->dev_count; i++) {
2861 struct kvm_io_device *pos = bus->range[i].dev;
2863 kvm_iodevice_destructor(pos);
2868 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
2869 const struct kvm_io_range *r2)
2871 if (r1->addr < r2->addr)
2873 if (r1->addr + r1->len > r2->addr + r2->len)
2878 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2880 return kvm_io_bus_cmp(p1, p2);
2883 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2884 gpa_t addr, int len)
2886 bus->range[bus->dev_count++] = (struct kvm_io_range) {
2892 sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2893 kvm_io_bus_sort_cmp, NULL);
2898 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2899 gpa_t addr, int len)
2901 struct kvm_io_range *range, key;
2904 key = (struct kvm_io_range) {
2909 range = bsearch(&key, bus->range, bus->dev_count,
2910 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2914 off = range - bus->range;
2916 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
2922 static int __kvm_io_bus_write(struct kvm_io_bus *bus,
2923 struct kvm_io_range *range, const void *val)
2927 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2931 while (idx < bus->dev_count &&
2932 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
2933 if (!kvm_iodevice_write(bus->range[idx].dev, range->addr,
2942 /* kvm_io_bus_write - called under kvm->slots_lock */
2943 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2944 int len, const void *val)
2946 struct kvm_io_bus *bus;
2947 struct kvm_io_range range;
2950 range = (struct kvm_io_range) {
2955 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2956 r = __kvm_io_bus_write(bus, &range, val);
2957 return r < 0 ? r : 0;
2960 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
2961 int kvm_io_bus_write_cookie(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2962 int len, const void *val, long cookie)
2964 struct kvm_io_bus *bus;
2965 struct kvm_io_range range;
2967 range = (struct kvm_io_range) {
2972 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2974 /* First try the device referenced by cookie. */
2975 if ((cookie >= 0) && (cookie < bus->dev_count) &&
2976 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
2977 if (!kvm_iodevice_write(bus->range[cookie].dev, addr, len,
2982 * cookie contained garbage; fall back to search and return the
2983 * correct cookie value.
2985 return __kvm_io_bus_write(bus, &range, val);
2988 static int __kvm_io_bus_read(struct kvm_io_bus *bus, struct kvm_io_range *range,
2993 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2997 while (idx < bus->dev_count &&
2998 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
2999 if (!kvm_iodevice_read(bus->range[idx].dev, range->addr,
3007 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
3009 /* kvm_io_bus_read - called under kvm->slots_lock */
3010 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3013 struct kvm_io_bus *bus;
3014 struct kvm_io_range range;
3017 range = (struct kvm_io_range) {
3022 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
3023 r = __kvm_io_bus_read(bus, &range, val);
3024 return r < 0 ? r : 0;
3028 /* Caller must hold slots_lock. */
3029 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3030 int len, struct kvm_io_device *dev)
3032 struct kvm_io_bus *new_bus, *bus;
3034 bus = kvm->buses[bus_idx];
3035 /* exclude ioeventfd which is limited by maximum fd */
3036 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3039 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
3040 sizeof(struct kvm_io_range)), GFP_KERNEL);
3043 memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
3044 sizeof(struct kvm_io_range)));
3045 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
3046 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3047 synchronize_srcu_expedited(&kvm->srcu);
3053 /* Caller must hold slots_lock. */
3054 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3055 struct kvm_io_device *dev)
3058 struct kvm_io_bus *new_bus, *bus;
3060 bus = kvm->buses[bus_idx];
3062 for (i = 0; i < bus->dev_count; i++)
3063 if (bus->range[i].dev == dev) {
3071 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3072 sizeof(struct kvm_io_range)), GFP_KERNEL);
3076 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3077 new_bus->dev_count--;
3078 memcpy(new_bus->range + i, bus->range + i + 1,
3079 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3081 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3082 synchronize_srcu_expedited(&kvm->srcu);
3087 static struct notifier_block kvm_cpu_notifier = {
3088 .notifier_call = kvm_cpu_hotplug,
3091 static int vm_stat_get(void *_offset, u64 *val)
3093 unsigned offset = (long)_offset;
3097 spin_lock(&kvm_lock);
3098 list_for_each_entry(kvm, &vm_list, vm_list)
3099 *val += *(u32 *)((void *)kvm + offset);
3100 spin_unlock(&kvm_lock);
3104 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
3106 static int vcpu_stat_get(void *_offset, u64 *val)
3108 unsigned offset = (long)_offset;
3110 struct kvm_vcpu *vcpu;
3114 spin_lock(&kvm_lock);
3115 list_for_each_entry(kvm, &vm_list, vm_list)
3116 kvm_for_each_vcpu(i, vcpu, kvm)
3117 *val += *(u32 *)((void *)vcpu + offset);
3119 spin_unlock(&kvm_lock);
3123 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
3125 static const struct file_operations *stat_fops[] = {
3126 [KVM_STAT_VCPU] = &vcpu_stat_fops,
3127 [KVM_STAT_VM] = &vm_stat_fops,
3130 static int kvm_init_debug(void)
3133 struct kvm_stats_debugfs_item *p;
3135 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3136 if (kvm_debugfs_dir == NULL)
3139 for (p = debugfs_entries; p->name; ++p) {
3140 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
3141 (void *)(long)p->offset,
3142 stat_fops[p->kind]);
3143 if (p->dentry == NULL)
3150 debugfs_remove_recursive(kvm_debugfs_dir);
3155 static void kvm_exit_debug(void)
3157 struct kvm_stats_debugfs_item *p;
3159 for (p = debugfs_entries; p->name; ++p)
3160 debugfs_remove(p->dentry);
3161 debugfs_remove(kvm_debugfs_dir);
3164 static int kvm_suspend(void)
3166 if (kvm_usage_count)
3167 hardware_disable_nolock(NULL);
3171 static void kvm_resume(void)
3173 if (kvm_usage_count) {
3174 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3175 hardware_enable_nolock(NULL);
3179 static struct syscore_ops kvm_syscore_ops = {
3180 .suspend = kvm_suspend,
3181 .resume = kvm_resume,
3185 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3187 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3190 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3192 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3193 if (vcpu->preempted)
3194 vcpu->preempted = false;
3196 kvm_arch_sched_in(vcpu, cpu);
3198 kvm_arch_vcpu_load(vcpu, cpu);
3201 static void kvm_sched_out(struct preempt_notifier *pn,
3202 struct task_struct *next)
3204 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3206 if (current->state == TASK_RUNNING)
3207 vcpu->preempted = true;
3208 kvm_arch_vcpu_put(vcpu);
3211 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3212 struct module *module)
3217 r = kvm_arch_init(opaque);
3222 * kvm_arch_init makes sure there's at most one caller
3223 * for architectures that support multiple implementations,
3224 * like intel and amd on x86.
3225 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3226 * conflicts in case kvm is already setup for another implementation.
3228 r = kvm_irqfd_init();
3232 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
3237 r = kvm_arch_hardware_setup();
3241 for_each_online_cpu(cpu) {
3242 smp_call_function_single(cpu,
3243 kvm_arch_check_processor_compat,
3249 r = register_cpu_notifier(&kvm_cpu_notifier);
3252 register_reboot_notifier(&kvm_reboot_notifier);
3254 /* A kmem cache lets us meet the alignment requirements of fx_save. */
3256 vcpu_align = __alignof__(struct kvm_vcpu);
3257 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
3259 if (!kvm_vcpu_cache) {
3264 r = kvm_async_pf_init();
3268 kvm_chardev_ops.owner = module;
3269 kvm_vm_fops.owner = module;
3270 kvm_vcpu_fops.owner = module;
3272 r = misc_register(&kvm_dev);
3274 printk(KERN_ERR "kvm: misc device register failed\n");
3278 register_syscore_ops(&kvm_syscore_ops);
3280 kvm_preempt_ops.sched_in = kvm_sched_in;
3281 kvm_preempt_ops.sched_out = kvm_sched_out;
3283 r = kvm_init_debug();
3285 printk(KERN_ERR "kvm: create debugfs files failed\n");
3289 r = kvm_vfio_ops_init();
3295 unregister_syscore_ops(&kvm_syscore_ops);
3296 misc_deregister(&kvm_dev);
3298 kvm_async_pf_deinit();
3300 kmem_cache_destroy(kvm_vcpu_cache);
3302 unregister_reboot_notifier(&kvm_reboot_notifier);
3303 unregister_cpu_notifier(&kvm_cpu_notifier);
3306 kvm_arch_hardware_unsetup();
3308 free_cpumask_var(cpus_hardware_enabled);
3316 EXPORT_SYMBOL_GPL(kvm_init);
3321 misc_deregister(&kvm_dev);
3322 kmem_cache_destroy(kvm_vcpu_cache);
3323 kvm_async_pf_deinit();
3324 unregister_syscore_ops(&kvm_syscore_ops);
3325 unregister_reboot_notifier(&kvm_reboot_notifier);
3326 unregister_cpu_notifier(&kvm_cpu_notifier);
3327 on_each_cpu(hardware_disable_nolock, NULL, 1);
3328 kvm_arch_hardware_unsetup();
3331 free_cpumask_var(cpus_hardware_enabled);
3332 kvm_vfio_ops_exit();
3334 EXPORT_SYMBOL_GPL(kvm_exit);