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/uaccess.h>
56 #include <asm/pgtable.h>
58 #include "coalesced_mmio.h"
62 #define CREATE_TRACE_POINTS
63 #include <trace/events/kvm.h>
65 MODULE_AUTHOR("Qumranet");
66 MODULE_LICENSE("GPL");
71 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
74 DEFINE_SPINLOCK(kvm_lock);
75 static DEFINE_RAW_SPINLOCK(kvm_count_lock);
78 static cpumask_var_t cpus_hardware_enabled;
79 static int kvm_usage_count = 0;
80 static atomic_t hardware_enable_failed;
82 struct kmem_cache *kvm_vcpu_cache;
83 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
85 static __read_mostly struct preempt_ops kvm_preempt_ops;
87 struct dentry *kvm_debugfs_dir;
89 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
92 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
95 static int hardware_enable_all(void);
96 static void hardware_disable_all(void);
98 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
100 static void kvm_release_pfn_dirty(pfn_t pfn);
101 static void mark_page_dirty_in_slot(struct kvm *kvm,
102 struct kvm_memory_slot *memslot, gfn_t gfn);
104 __visible bool kvm_rebooting;
105 EXPORT_SYMBOL_GPL(kvm_rebooting);
107 static bool largepages_enabled = true;
109 bool kvm_is_mmio_pfn(pfn_t pfn)
112 return PageReserved(pfn_to_page(pfn));
118 * Switches to specified vcpu, until a matching vcpu_put()
120 int vcpu_load(struct kvm_vcpu *vcpu)
124 if (mutex_lock_killable(&vcpu->mutex))
126 if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
127 /* The thread running this VCPU changed. */
128 struct pid *oldpid = vcpu->pid;
129 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
130 rcu_assign_pointer(vcpu->pid, newpid);
136 preempt_notifier_register(&vcpu->preempt_notifier);
137 kvm_arch_vcpu_load(vcpu, cpu);
142 void vcpu_put(struct kvm_vcpu *vcpu)
145 kvm_arch_vcpu_put(vcpu);
146 preempt_notifier_unregister(&vcpu->preempt_notifier);
148 mutex_unlock(&vcpu->mutex);
151 static void ack_flush(void *_completed)
155 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
160 struct kvm_vcpu *vcpu;
162 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
165 kvm_for_each_vcpu(i, vcpu, kvm) {
166 kvm_make_request(req, vcpu);
169 /* Set ->requests bit before we read ->mode */
172 if (cpus != NULL && cpu != -1 && cpu != me &&
173 kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
174 cpumask_set_cpu(cpu, cpus);
176 if (unlikely(cpus == NULL))
177 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
178 else if (!cpumask_empty(cpus))
179 smp_call_function_many(cpus, ack_flush, NULL, 1);
183 free_cpumask_var(cpus);
187 void kvm_flush_remote_tlbs(struct kvm *kvm)
189 long dirty_count = kvm->tlbs_dirty;
192 if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
193 ++kvm->stat.remote_tlb_flush;
194 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
196 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
198 void kvm_reload_remote_mmus(struct kvm *kvm)
200 make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
203 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
205 make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
208 void kvm_make_scan_ioapic_request(struct kvm *kvm)
210 make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
213 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
218 mutex_init(&vcpu->mutex);
223 init_waitqueue_head(&vcpu->wq);
224 kvm_async_pf_vcpu_init(vcpu);
226 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
231 vcpu->run = page_address(page);
233 kvm_vcpu_set_in_spin_loop(vcpu, false);
234 kvm_vcpu_set_dy_eligible(vcpu, false);
235 vcpu->preempted = false;
237 r = kvm_arch_vcpu_init(vcpu);
243 free_page((unsigned long)vcpu->run);
247 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
249 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
252 kvm_arch_vcpu_uninit(vcpu);
253 free_page((unsigned long)vcpu->run);
255 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
257 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
258 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
260 return container_of(mn, struct kvm, mmu_notifier);
263 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
264 struct mm_struct *mm,
265 unsigned long address)
267 struct kvm *kvm = mmu_notifier_to_kvm(mn);
268 int need_tlb_flush, idx;
271 * When ->invalidate_page runs, the linux pte has been zapped
272 * already but the page is still allocated until
273 * ->invalidate_page returns. So if we increase the sequence
274 * here the kvm page fault will notice if the spte can't be
275 * established because the page is going to be freed. If
276 * instead the kvm page fault establishes the spte before
277 * ->invalidate_page runs, kvm_unmap_hva will release it
280 * The sequence increase only need to be seen at spin_unlock
281 * time, and not at spin_lock time.
283 * Increasing the sequence after the spin_unlock would be
284 * unsafe because the kvm page fault could then establish the
285 * pte after kvm_unmap_hva returned, without noticing the page
286 * is going to be freed.
288 idx = srcu_read_lock(&kvm->srcu);
289 spin_lock(&kvm->mmu_lock);
291 kvm->mmu_notifier_seq++;
292 need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
293 /* we've to flush the tlb before the pages can be freed */
295 kvm_flush_remote_tlbs(kvm);
297 spin_unlock(&kvm->mmu_lock);
298 srcu_read_unlock(&kvm->srcu, idx);
301 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
302 struct mm_struct *mm,
303 unsigned long address,
306 struct kvm *kvm = mmu_notifier_to_kvm(mn);
309 idx = srcu_read_lock(&kvm->srcu);
310 spin_lock(&kvm->mmu_lock);
311 kvm->mmu_notifier_seq++;
312 kvm_set_spte_hva(kvm, address, pte);
313 spin_unlock(&kvm->mmu_lock);
314 srcu_read_unlock(&kvm->srcu, idx);
317 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
318 struct mm_struct *mm,
322 struct kvm *kvm = mmu_notifier_to_kvm(mn);
323 int need_tlb_flush = 0, idx;
325 idx = srcu_read_lock(&kvm->srcu);
326 spin_lock(&kvm->mmu_lock);
328 * The count increase must become visible at unlock time as no
329 * spte can be established without taking the mmu_lock and
330 * count is also read inside the mmu_lock critical section.
332 kvm->mmu_notifier_count++;
333 need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
334 need_tlb_flush |= kvm->tlbs_dirty;
335 /* we've to flush the tlb before the pages can be freed */
337 kvm_flush_remote_tlbs(kvm);
339 spin_unlock(&kvm->mmu_lock);
340 srcu_read_unlock(&kvm->srcu, idx);
343 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
344 struct mm_struct *mm,
348 struct kvm *kvm = mmu_notifier_to_kvm(mn);
350 spin_lock(&kvm->mmu_lock);
352 * This sequence increase will notify the kvm page fault that
353 * the page that is going to be mapped in the spte could have
356 kvm->mmu_notifier_seq++;
359 * The above sequence increase must be visible before the
360 * below count decrease, which is ensured by the smp_wmb above
361 * in conjunction with the smp_rmb in mmu_notifier_retry().
363 kvm->mmu_notifier_count--;
364 spin_unlock(&kvm->mmu_lock);
366 BUG_ON(kvm->mmu_notifier_count < 0);
369 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
370 struct mm_struct *mm,
371 unsigned long address)
373 struct kvm *kvm = mmu_notifier_to_kvm(mn);
376 idx = srcu_read_lock(&kvm->srcu);
377 spin_lock(&kvm->mmu_lock);
379 young = kvm_age_hva(kvm, address);
381 kvm_flush_remote_tlbs(kvm);
383 spin_unlock(&kvm->mmu_lock);
384 srcu_read_unlock(&kvm->srcu, idx);
389 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
390 struct mm_struct *mm,
391 unsigned long address)
393 struct kvm *kvm = mmu_notifier_to_kvm(mn);
396 idx = srcu_read_lock(&kvm->srcu);
397 spin_lock(&kvm->mmu_lock);
398 young = kvm_test_age_hva(kvm, address);
399 spin_unlock(&kvm->mmu_lock);
400 srcu_read_unlock(&kvm->srcu, idx);
405 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
406 struct mm_struct *mm)
408 struct kvm *kvm = mmu_notifier_to_kvm(mn);
411 idx = srcu_read_lock(&kvm->srcu);
412 kvm_arch_flush_shadow_all(kvm);
413 srcu_read_unlock(&kvm->srcu, idx);
416 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
417 .invalidate_page = kvm_mmu_notifier_invalidate_page,
418 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
419 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
420 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
421 .test_young = kvm_mmu_notifier_test_young,
422 .change_pte = kvm_mmu_notifier_change_pte,
423 .release = kvm_mmu_notifier_release,
426 static int kvm_init_mmu_notifier(struct kvm *kvm)
428 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
429 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
432 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
434 static int kvm_init_mmu_notifier(struct kvm *kvm)
439 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
441 static void kvm_init_memslots_id(struct kvm *kvm)
444 struct kvm_memslots *slots = kvm->memslots;
446 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
447 slots->id_to_index[i] = slots->memslots[i].id = i;
450 static struct kvm *kvm_create_vm(unsigned long type)
453 struct kvm *kvm = kvm_arch_alloc_vm();
456 return ERR_PTR(-ENOMEM);
458 r = kvm_arch_init_vm(kvm, type);
460 goto out_err_no_disable;
462 r = hardware_enable_all();
464 goto out_err_no_disable;
466 #ifdef CONFIG_HAVE_KVM_IRQCHIP
467 INIT_HLIST_HEAD(&kvm->mask_notifier_list);
469 #ifdef CONFIG_HAVE_KVM_IRQFD
470 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
473 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
476 kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
478 goto out_err_no_srcu;
481 * Init kvm generation close to the maximum to easily test the
482 * code of handling generation number wrap-around.
484 kvm->memslots->generation = -150;
486 kvm_init_memslots_id(kvm);
487 if (init_srcu_struct(&kvm->srcu))
488 goto out_err_no_srcu;
489 if (init_srcu_struct(&kvm->irq_srcu))
490 goto out_err_no_irq_srcu;
491 for (i = 0; i < KVM_NR_BUSES; i++) {
492 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
498 spin_lock_init(&kvm->mmu_lock);
499 kvm->mm = current->mm;
500 atomic_inc(&kvm->mm->mm_count);
501 kvm_eventfd_init(kvm);
502 mutex_init(&kvm->lock);
503 mutex_init(&kvm->irq_lock);
504 mutex_init(&kvm->slots_lock);
505 atomic_set(&kvm->users_count, 1);
506 INIT_LIST_HEAD(&kvm->devices);
508 r = kvm_init_mmu_notifier(kvm);
512 spin_lock(&kvm_lock);
513 list_add(&kvm->vm_list, &vm_list);
514 spin_unlock(&kvm_lock);
519 cleanup_srcu_struct(&kvm->irq_srcu);
521 cleanup_srcu_struct(&kvm->srcu);
523 hardware_disable_all();
525 for (i = 0; i < KVM_NR_BUSES; i++)
526 kfree(kvm->buses[i]);
527 kfree(kvm->memslots);
528 kvm_arch_free_vm(kvm);
533 * Avoid using vmalloc for a small buffer.
534 * Should not be used when the size is statically known.
536 void *kvm_kvzalloc(unsigned long size)
538 if (size > PAGE_SIZE)
539 return vzalloc(size);
541 return kzalloc(size, GFP_KERNEL);
544 void kvm_kvfree(const void *addr)
546 if (is_vmalloc_addr(addr))
552 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
554 if (!memslot->dirty_bitmap)
557 kvm_kvfree(memslot->dirty_bitmap);
558 memslot->dirty_bitmap = NULL;
562 * Free any memory in @free but not in @dont.
564 static void kvm_free_physmem_slot(struct kvm *kvm, struct kvm_memory_slot *free,
565 struct kvm_memory_slot *dont)
567 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
568 kvm_destroy_dirty_bitmap(free);
570 kvm_arch_free_memslot(kvm, free, dont);
575 static void kvm_free_physmem(struct kvm *kvm)
577 struct kvm_memslots *slots = kvm->memslots;
578 struct kvm_memory_slot *memslot;
580 kvm_for_each_memslot(memslot, slots)
581 kvm_free_physmem_slot(kvm, memslot, NULL);
583 kfree(kvm->memslots);
586 static void kvm_destroy_devices(struct kvm *kvm)
588 struct list_head *node, *tmp;
590 list_for_each_safe(node, tmp, &kvm->devices) {
591 struct kvm_device *dev =
592 list_entry(node, struct kvm_device, vm_node);
595 dev->ops->destroy(dev);
599 static void kvm_destroy_vm(struct kvm *kvm)
602 struct mm_struct *mm = kvm->mm;
604 kvm_arch_sync_events(kvm);
605 spin_lock(&kvm_lock);
606 list_del(&kvm->vm_list);
607 spin_unlock(&kvm_lock);
608 kvm_free_irq_routing(kvm);
609 for (i = 0; i < KVM_NR_BUSES; i++)
610 kvm_io_bus_destroy(kvm->buses[i]);
611 kvm_coalesced_mmio_free(kvm);
612 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
613 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
615 kvm_arch_flush_shadow_all(kvm);
617 kvm_arch_destroy_vm(kvm);
618 kvm_destroy_devices(kvm);
619 kvm_free_physmem(kvm);
620 cleanup_srcu_struct(&kvm->irq_srcu);
621 cleanup_srcu_struct(&kvm->srcu);
622 kvm_arch_free_vm(kvm);
623 hardware_disable_all();
627 void kvm_get_kvm(struct kvm *kvm)
629 atomic_inc(&kvm->users_count);
631 EXPORT_SYMBOL_GPL(kvm_get_kvm);
633 void kvm_put_kvm(struct kvm *kvm)
635 if (atomic_dec_and_test(&kvm->users_count))
638 EXPORT_SYMBOL_GPL(kvm_put_kvm);
641 static int kvm_vm_release(struct inode *inode, struct file *filp)
643 struct kvm *kvm = filp->private_data;
645 kvm_irqfd_release(kvm);
652 * Allocation size is twice as large as the actual dirty bitmap size.
653 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
655 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
657 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
659 memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
660 if (!memslot->dirty_bitmap)
666 static int cmp_memslot(const void *slot1, const void *slot2)
668 struct kvm_memory_slot *s1, *s2;
670 s1 = (struct kvm_memory_slot *)slot1;
671 s2 = (struct kvm_memory_slot *)slot2;
673 if (s1->npages < s2->npages)
675 if (s1->npages > s2->npages)
682 * Sort the memslots base on its size, so the larger slots
683 * will get better fit.
685 static void sort_memslots(struct kvm_memslots *slots)
689 sort(slots->memslots, KVM_MEM_SLOTS_NUM,
690 sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
692 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
693 slots->id_to_index[slots->memslots[i].id] = i;
696 static void update_memslots(struct kvm_memslots *slots,
697 struct kvm_memory_slot *new)
701 struct kvm_memory_slot *old = id_to_memslot(slots, id);
702 unsigned long npages = old->npages;
705 if (new->npages != npages)
706 sort_memslots(slots);
710 static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
712 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
714 #ifdef __KVM_HAVE_READONLY_MEM
715 valid_flags |= KVM_MEM_READONLY;
718 if (mem->flags & ~valid_flags)
724 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
725 struct kvm_memslots *slots, struct kvm_memory_slot *new)
727 struct kvm_memslots *old_memslots = kvm->memslots;
730 * Set the low bit in the generation, which disables SPTE caching
731 * until the end of synchronize_srcu_expedited.
733 WARN_ON(old_memslots->generation & 1);
734 slots->generation = old_memslots->generation + 1;
736 update_memslots(slots, new);
737 rcu_assign_pointer(kvm->memslots, slots);
738 synchronize_srcu_expedited(&kvm->srcu);
741 * Increment the new memslot generation a second time. This prevents
742 * vm exits that race with memslot updates from caching a memslot
743 * generation that will (potentially) be valid forever.
747 kvm_arch_memslots_updated(kvm);
753 * Allocate some memory and give it an address in the guest physical address
756 * Discontiguous memory is allowed, mostly for framebuffers.
758 * Must be called holding mmap_sem for write.
760 int __kvm_set_memory_region(struct kvm *kvm,
761 struct kvm_userspace_memory_region *mem)
765 unsigned long npages;
766 struct kvm_memory_slot *slot;
767 struct kvm_memory_slot old, new;
768 struct kvm_memslots *slots = NULL, *old_memslots;
769 enum kvm_mr_change change;
771 r = check_memory_region_flags(mem);
776 /* General sanity checks */
777 if (mem->memory_size & (PAGE_SIZE - 1))
779 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
781 /* We can read the guest memory with __xxx_user() later on. */
782 if ((mem->slot < KVM_USER_MEM_SLOTS) &&
783 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
784 !access_ok(VERIFY_WRITE,
785 (void __user *)(unsigned long)mem->userspace_addr,
788 if (mem->slot >= KVM_MEM_SLOTS_NUM)
790 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
793 slot = id_to_memslot(kvm->memslots, mem->slot);
794 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
795 npages = mem->memory_size >> PAGE_SHIFT;
797 if (npages > KVM_MEM_MAX_NR_PAGES)
801 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
806 new.base_gfn = base_gfn;
808 new.flags = mem->flags;
812 change = KVM_MR_CREATE;
813 else { /* Modify an existing slot. */
814 if ((mem->userspace_addr != old.userspace_addr) ||
815 (npages != old.npages) ||
816 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
819 if (base_gfn != old.base_gfn)
820 change = KVM_MR_MOVE;
821 else if (new.flags != old.flags)
822 change = KVM_MR_FLAGS_ONLY;
823 else { /* Nothing to change. */
828 } else if (old.npages) {
829 change = KVM_MR_DELETE;
830 } else /* Modify a non-existent slot: disallowed. */
833 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
834 /* Check for overlaps */
836 kvm_for_each_memslot(slot, kvm->memslots) {
837 if ((slot->id >= KVM_USER_MEM_SLOTS) ||
838 (slot->id == mem->slot))
840 if (!((base_gfn + npages <= slot->base_gfn) ||
841 (base_gfn >= slot->base_gfn + slot->npages)))
846 /* Free page dirty bitmap if unneeded */
847 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
848 new.dirty_bitmap = NULL;
851 if (change == KVM_MR_CREATE) {
852 new.userspace_addr = mem->userspace_addr;
854 if (kvm_arch_create_memslot(kvm, &new, npages))
858 /* Allocate page dirty bitmap if needed */
859 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
860 if (kvm_create_dirty_bitmap(&new) < 0)
864 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
865 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
869 slot = id_to_memslot(slots, mem->slot);
870 slot->flags |= KVM_MEMSLOT_INVALID;
872 old_memslots = install_new_memslots(kvm, slots, NULL);
874 /* slot was deleted or moved, clear iommu mapping */
875 kvm_iommu_unmap_pages(kvm, &old);
876 /* From this point no new shadow pages pointing to a deleted,
877 * or moved, memslot will be created.
879 * validation of sp->gfn happens in:
880 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
881 * - kvm_is_visible_gfn (mmu_check_roots)
883 kvm_arch_flush_shadow_memslot(kvm, slot);
884 slots = old_memslots;
887 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
893 * We can re-use the old_memslots from above, the only difference
894 * from the currently installed memslots is the invalid flag. This
895 * will get overwritten by update_memslots anyway.
898 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
904 /* actual memory is freed via old in kvm_free_physmem_slot below */
905 if (change == KVM_MR_DELETE) {
906 new.dirty_bitmap = NULL;
907 memset(&new.arch, 0, sizeof(new.arch));
910 old_memslots = install_new_memslots(kvm, slots, &new);
912 kvm_arch_commit_memory_region(kvm, mem, &old, change);
914 kvm_free_physmem_slot(kvm, &old, &new);
918 * IOMMU mapping: New slots need to be mapped. Old slots need to be
919 * un-mapped and re-mapped if their base changes. Since base change
920 * unmapping is handled above with slot deletion, mapping alone is
921 * needed here. Anything else the iommu might care about for existing
922 * slots (size changes, userspace addr changes and read-only flag
923 * changes) is disallowed above, so any other attribute changes getting
924 * here can be skipped.
926 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
927 r = kvm_iommu_map_pages(kvm, &new);
936 kvm_free_physmem_slot(kvm, &new, &old);
940 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
942 int kvm_set_memory_region(struct kvm *kvm,
943 struct kvm_userspace_memory_region *mem)
947 mutex_lock(&kvm->slots_lock);
948 r = __kvm_set_memory_region(kvm, mem);
949 mutex_unlock(&kvm->slots_lock);
952 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
954 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
955 struct kvm_userspace_memory_region *mem)
957 if (mem->slot >= KVM_USER_MEM_SLOTS)
959 return kvm_set_memory_region(kvm, mem);
962 int kvm_get_dirty_log(struct kvm *kvm,
963 struct kvm_dirty_log *log, int *is_dirty)
965 struct kvm_memory_slot *memslot;
968 unsigned long any = 0;
971 if (log->slot >= KVM_USER_MEM_SLOTS)
974 memslot = id_to_memslot(kvm->memslots, log->slot);
976 if (!memslot->dirty_bitmap)
979 n = kvm_dirty_bitmap_bytes(memslot);
981 for (i = 0; !any && i < n/sizeof(long); ++i)
982 any = memslot->dirty_bitmap[i];
985 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
995 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
997 bool kvm_largepages_enabled(void)
999 return largepages_enabled;
1002 void kvm_disable_largepages(void)
1004 largepages_enabled = false;
1006 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1008 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1010 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1012 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1014 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1016 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1018 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1019 memslot->flags & KVM_MEMSLOT_INVALID)
1024 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1026 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1028 struct vm_area_struct *vma;
1029 unsigned long addr, size;
1033 addr = gfn_to_hva(kvm, gfn);
1034 if (kvm_is_error_hva(addr))
1037 down_read(¤t->mm->mmap_sem);
1038 vma = find_vma(current->mm, addr);
1042 size = vma_kernel_pagesize(vma);
1045 up_read(¤t->mm->mmap_sem);
1050 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1052 return slot->flags & KVM_MEM_READONLY;
1055 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1056 gfn_t *nr_pages, bool write)
1058 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1059 return KVM_HVA_ERR_BAD;
1061 if (memslot_is_readonly(slot) && write)
1062 return KVM_HVA_ERR_RO_BAD;
1065 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1067 return __gfn_to_hva_memslot(slot, gfn);
1070 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1073 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1076 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1079 return gfn_to_hva_many(slot, gfn, NULL);
1081 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1083 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1085 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1087 EXPORT_SYMBOL_GPL(gfn_to_hva);
1090 * If writable is set to false, the hva returned by this function is only
1091 * allowed to be read.
1093 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1095 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1096 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1098 if (!kvm_is_error_hva(hva) && writable)
1099 *writable = !memslot_is_readonly(slot);
1104 static int kvm_read_hva(void *data, void __user *hva, int len)
1106 return __copy_from_user(data, hva, len);
1109 static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1111 return __copy_from_user_inatomic(data, hva, len);
1114 static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1115 unsigned long start, int write, struct page **page)
1117 int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1120 flags |= FOLL_WRITE;
1122 return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1125 int kvm_get_user_page_io(struct task_struct *tsk, struct mm_struct *mm,
1126 unsigned long addr, bool write_fault,
1127 struct page **pagep)
1131 int flags = FOLL_TOUCH | FOLL_HWPOISON |
1132 (pagep ? FOLL_GET : 0) |
1133 (write_fault ? FOLL_WRITE : 0);
1136 * If retrying the fault, we get here *not* having allowed the filemap
1137 * to wait on the page lock. We should now allow waiting on the IO with
1138 * the mmap semaphore released.
1140 down_read(&mm->mmap_sem);
1141 npages = __get_user_pages(tsk, mm, addr, 1, flags, pagep, NULL,
1144 VM_BUG_ON(npages != -EBUSY);
1150 * The previous call has now waited on the IO. Now we can
1151 * retry and complete. Pass TRIED to ensure we do not re
1152 * schedule async IO (see e.g. filemap_fault).
1154 down_read(&mm->mmap_sem);
1155 npages = __get_user_pages(tsk, mm, addr, 1, flags | FOLL_TRIED,
1158 up_read(&mm->mmap_sem);
1162 static inline int check_user_page_hwpoison(unsigned long addr)
1164 int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1166 rc = __get_user_pages(current, current->mm, addr, 1,
1167 flags, NULL, NULL, NULL);
1168 return rc == -EHWPOISON;
1172 * The atomic path to get the writable pfn which will be stored in @pfn,
1173 * true indicates success, otherwise false is returned.
1175 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1176 bool write_fault, bool *writable, pfn_t *pfn)
1178 struct page *page[1];
1181 if (!(async || atomic))
1185 * Fast pin a writable pfn only if it is a write fault request
1186 * or the caller allows to map a writable pfn for a read fault
1189 if (!(write_fault || writable))
1192 npages = __get_user_pages_fast(addr, 1, 1, page);
1194 *pfn = page_to_pfn(page[0]);
1205 * The slow path to get the pfn of the specified host virtual address,
1206 * 1 indicates success, -errno is returned if error is detected.
1208 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1209 bool *writable, pfn_t *pfn)
1211 struct page *page[1];
1217 *writable = write_fault;
1220 down_read(¤t->mm->mmap_sem);
1221 npages = get_user_page_nowait(current, current->mm,
1222 addr, write_fault, page);
1223 up_read(¤t->mm->mmap_sem);
1226 * By now we have tried gup_fast, and possibly async_pf, and we
1227 * are certainly not atomic. Time to retry the gup, allowing
1228 * mmap semaphore to be relinquished in the case of IO.
1230 npages = kvm_get_user_page_io(current, current->mm, addr,
1236 /* map read fault as writable if possible */
1237 if (unlikely(!write_fault) && writable) {
1238 struct page *wpage[1];
1240 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1249 *pfn = page_to_pfn(page[0]);
1253 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1255 if (unlikely(!(vma->vm_flags & VM_READ)))
1258 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1265 * Pin guest page in memory and return its pfn.
1266 * @addr: host virtual address which maps memory to the guest
1267 * @atomic: whether this function can sleep
1268 * @async: whether this function need to wait IO complete if the
1269 * host page is not in the memory
1270 * @write_fault: whether we should get a writable host page
1271 * @writable: whether it allows to map a writable host page for !@write_fault
1273 * The function will map a writable host page for these two cases:
1274 * 1): @write_fault = true
1275 * 2): @write_fault = false && @writable, @writable will tell the caller
1276 * whether the mapping is writable.
1278 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1279 bool write_fault, bool *writable)
1281 struct vm_area_struct *vma;
1285 /* we can do it either atomically or asynchronously, not both */
1286 BUG_ON(atomic && async);
1288 if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1292 return KVM_PFN_ERR_FAULT;
1294 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1298 down_read(¤t->mm->mmap_sem);
1299 if (npages == -EHWPOISON ||
1300 (!async && check_user_page_hwpoison(addr))) {
1301 pfn = KVM_PFN_ERR_HWPOISON;
1305 vma = find_vma_intersection(current->mm, addr, addr + 1);
1308 pfn = KVM_PFN_ERR_FAULT;
1309 else if ((vma->vm_flags & VM_PFNMAP)) {
1310 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1312 BUG_ON(!kvm_is_mmio_pfn(pfn));
1314 if (async && vma_is_valid(vma, write_fault))
1316 pfn = KVM_PFN_ERR_FAULT;
1319 up_read(¤t->mm->mmap_sem);
1324 __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1325 bool *async, bool write_fault, bool *writable)
1327 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1329 if (addr == KVM_HVA_ERR_RO_BAD)
1330 return KVM_PFN_ERR_RO_FAULT;
1332 if (kvm_is_error_hva(addr))
1333 return KVM_PFN_NOSLOT;
1335 /* Do not map writable pfn in the readonly memslot. */
1336 if (writable && memslot_is_readonly(slot)) {
1341 return hva_to_pfn(addr, atomic, async, write_fault,
1345 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1346 bool write_fault, bool *writable)
1348 struct kvm_memory_slot *slot;
1353 slot = gfn_to_memslot(kvm, gfn);
1355 return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1359 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1361 return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1363 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1365 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1366 bool write_fault, bool *writable)
1368 return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1370 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1372 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1374 return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1376 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1378 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1381 return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1383 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1385 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1387 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1390 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1392 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1394 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1396 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1402 addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1403 if (kvm_is_error_hva(addr))
1406 if (entry < nr_pages)
1409 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1411 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1413 static struct page *kvm_pfn_to_page(pfn_t pfn)
1415 if (is_error_noslot_pfn(pfn))
1416 return KVM_ERR_PTR_BAD_PAGE;
1418 if (kvm_is_mmio_pfn(pfn)) {
1420 return KVM_ERR_PTR_BAD_PAGE;
1423 return pfn_to_page(pfn);
1426 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1430 pfn = gfn_to_pfn(kvm, gfn);
1432 return kvm_pfn_to_page(pfn);
1435 EXPORT_SYMBOL_GPL(gfn_to_page);
1437 void kvm_release_page_clean(struct page *page)
1439 WARN_ON(is_error_page(page));
1441 kvm_release_pfn_clean(page_to_pfn(page));
1443 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1445 void kvm_release_pfn_clean(pfn_t pfn)
1447 if (!is_error_noslot_pfn(pfn) && !kvm_is_mmio_pfn(pfn))
1448 put_page(pfn_to_page(pfn));
1450 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1452 void kvm_release_page_dirty(struct page *page)
1454 WARN_ON(is_error_page(page));
1456 kvm_release_pfn_dirty(page_to_pfn(page));
1458 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1460 static void kvm_release_pfn_dirty(pfn_t pfn)
1462 kvm_set_pfn_dirty(pfn);
1463 kvm_release_pfn_clean(pfn);
1466 void kvm_set_pfn_dirty(pfn_t pfn)
1468 if (!kvm_is_mmio_pfn(pfn)) {
1469 struct page *page = pfn_to_page(pfn);
1470 if (!PageReserved(page))
1474 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1476 void kvm_set_pfn_accessed(pfn_t pfn)
1478 if (!kvm_is_mmio_pfn(pfn))
1479 mark_page_accessed(pfn_to_page(pfn));
1481 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1483 void kvm_get_pfn(pfn_t pfn)
1485 if (!kvm_is_mmio_pfn(pfn))
1486 get_page(pfn_to_page(pfn));
1488 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1490 static int next_segment(unsigned long len, int offset)
1492 if (len > PAGE_SIZE - offset)
1493 return PAGE_SIZE - offset;
1498 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1504 addr = gfn_to_hva_prot(kvm, gfn, NULL);
1505 if (kvm_is_error_hva(addr))
1507 r = kvm_read_hva(data, (void __user *)addr + offset, len);
1512 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1514 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1516 gfn_t gfn = gpa >> PAGE_SHIFT;
1518 int offset = offset_in_page(gpa);
1521 while ((seg = next_segment(len, offset)) != 0) {
1522 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1532 EXPORT_SYMBOL_GPL(kvm_read_guest);
1534 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1539 gfn_t gfn = gpa >> PAGE_SHIFT;
1540 int offset = offset_in_page(gpa);
1542 addr = gfn_to_hva_prot(kvm, gfn, NULL);
1543 if (kvm_is_error_hva(addr))
1545 pagefault_disable();
1546 r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
1552 EXPORT_SYMBOL(kvm_read_guest_atomic);
1554 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1555 int offset, int len)
1560 addr = gfn_to_hva(kvm, gfn);
1561 if (kvm_is_error_hva(addr))
1563 r = __copy_to_user((void __user *)addr + offset, data, len);
1566 mark_page_dirty(kvm, gfn);
1569 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1571 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1574 gfn_t gfn = gpa >> PAGE_SHIFT;
1576 int offset = offset_in_page(gpa);
1579 while ((seg = next_segment(len, offset)) != 0) {
1580 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1591 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1592 gpa_t gpa, unsigned long len)
1594 struct kvm_memslots *slots = kvm_memslots(kvm);
1595 int offset = offset_in_page(gpa);
1596 gfn_t start_gfn = gpa >> PAGE_SHIFT;
1597 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1598 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1599 gfn_t nr_pages_avail;
1602 ghc->generation = slots->generation;
1604 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1605 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, &nr_pages_avail);
1606 if (!kvm_is_error_hva(ghc->hva) && nr_pages_avail >= nr_pages_needed) {
1610 * If the requested region crosses two memslots, we still
1611 * verify that the entire region is valid here.
1613 while (start_gfn <= end_gfn) {
1614 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1615 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1617 if (kvm_is_error_hva(ghc->hva))
1619 start_gfn += nr_pages_avail;
1621 /* Use the slow path for cross page reads and writes. */
1622 ghc->memslot = NULL;
1626 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1628 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1629 void *data, unsigned long len)
1631 struct kvm_memslots *slots = kvm_memslots(kvm);
1634 BUG_ON(len > ghc->len);
1636 if (slots->generation != ghc->generation)
1637 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1639 if (unlikely(!ghc->memslot))
1640 return kvm_write_guest(kvm, ghc->gpa, data, len);
1642 if (kvm_is_error_hva(ghc->hva))
1645 r = __copy_to_user((void __user *)ghc->hva, data, len);
1648 mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1652 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1654 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1655 void *data, unsigned long len)
1657 struct kvm_memslots *slots = kvm_memslots(kvm);
1660 BUG_ON(len > ghc->len);
1662 if (slots->generation != ghc->generation)
1663 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1665 if (unlikely(!ghc->memslot))
1666 return kvm_read_guest(kvm, ghc->gpa, data, len);
1668 if (kvm_is_error_hva(ghc->hva))
1671 r = __copy_from_user(data, (void __user *)ghc->hva, len);
1677 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1679 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1681 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
1683 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
1685 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1687 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1689 gfn_t gfn = gpa >> PAGE_SHIFT;
1691 int offset = offset_in_page(gpa);
1694 while ((seg = next_segment(len, offset)) != 0) {
1695 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1704 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1706 static void mark_page_dirty_in_slot(struct kvm *kvm,
1707 struct kvm_memory_slot *memslot,
1710 if (memslot && memslot->dirty_bitmap) {
1711 unsigned long rel_gfn = gfn - memslot->base_gfn;
1713 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1717 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1719 struct kvm_memory_slot *memslot;
1721 memslot = gfn_to_memslot(kvm, gfn);
1722 mark_page_dirty_in_slot(kvm, memslot, gfn);
1724 EXPORT_SYMBOL_GPL(mark_page_dirty);
1727 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1729 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1734 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1736 if (kvm_arch_vcpu_runnable(vcpu)) {
1737 kvm_make_request(KVM_REQ_UNHALT, vcpu);
1740 if (kvm_cpu_has_pending_timer(vcpu))
1742 if (signal_pending(current))
1748 finish_wait(&vcpu->wq, &wait);
1750 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
1754 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1756 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1759 int cpu = vcpu->cpu;
1760 wait_queue_head_t *wqp;
1762 wqp = kvm_arch_vcpu_wq(vcpu);
1763 if (waitqueue_active(wqp)) {
1764 wake_up_interruptible(wqp);
1765 ++vcpu->stat.halt_wakeup;
1769 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1770 if (kvm_arch_vcpu_should_kick(vcpu))
1771 smp_send_reschedule(cpu);
1774 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
1775 #endif /* !CONFIG_S390 */
1777 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
1780 struct task_struct *task = NULL;
1784 pid = rcu_dereference(target->pid);
1786 task = get_pid_task(target->pid, PIDTYPE_PID);
1790 if (task->flags & PF_VCPU) {
1791 put_task_struct(task);
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 defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
2037 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
2038 * so vcpu_load() would break it.
2040 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
2041 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2045 r = vcpu_load(vcpu);
2053 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2054 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2056 case KVM_GET_REGS: {
2057 struct kvm_regs *kvm_regs;
2060 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2063 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2067 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2074 case KVM_SET_REGS: {
2075 struct kvm_regs *kvm_regs;
2078 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2079 if (IS_ERR(kvm_regs)) {
2080 r = PTR_ERR(kvm_regs);
2083 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2087 case KVM_GET_SREGS: {
2088 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2092 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2096 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2101 case KVM_SET_SREGS: {
2102 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2103 if (IS_ERR(kvm_sregs)) {
2104 r = PTR_ERR(kvm_sregs);
2108 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2111 case KVM_GET_MP_STATE: {
2112 struct kvm_mp_state mp_state;
2114 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2118 if (copy_to_user(argp, &mp_state, sizeof mp_state))
2123 case KVM_SET_MP_STATE: {
2124 struct kvm_mp_state mp_state;
2127 if (copy_from_user(&mp_state, argp, sizeof mp_state))
2129 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2132 case KVM_TRANSLATE: {
2133 struct kvm_translation tr;
2136 if (copy_from_user(&tr, argp, sizeof tr))
2138 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2142 if (copy_to_user(argp, &tr, sizeof tr))
2147 case KVM_SET_GUEST_DEBUG: {
2148 struct kvm_guest_debug dbg;
2151 if (copy_from_user(&dbg, argp, sizeof dbg))
2153 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2156 case KVM_SET_SIGNAL_MASK: {
2157 struct kvm_signal_mask __user *sigmask_arg = argp;
2158 struct kvm_signal_mask kvm_sigmask;
2159 sigset_t sigset, *p;
2164 if (copy_from_user(&kvm_sigmask, argp,
2165 sizeof kvm_sigmask))
2168 if (kvm_sigmask.len != sizeof sigset)
2171 if (copy_from_user(&sigset, sigmask_arg->sigset,
2176 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2180 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2184 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2188 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2194 fpu = memdup_user(argp, sizeof(*fpu));
2200 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2204 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2213 #ifdef CONFIG_COMPAT
2214 static long kvm_vcpu_compat_ioctl(struct file *filp,
2215 unsigned int ioctl, unsigned long arg)
2217 struct kvm_vcpu *vcpu = filp->private_data;
2218 void __user *argp = compat_ptr(arg);
2221 if (vcpu->kvm->mm != current->mm)
2225 case KVM_SET_SIGNAL_MASK: {
2226 struct kvm_signal_mask __user *sigmask_arg = argp;
2227 struct kvm_signal_mask kvm_sigmask;
2228 compat_sigset_t csigset;
2233 if (copy_from_user(&kvm_sigmask, argp,
2234 sizeof kvm_sigmask))
2237 if (kvm_sigmask.len != sizeof csigset)
2240 if (copy_from_user(&csigset, sigmask_arg->sigset,
2243 sigset_from_compat(&sigset, &csigset);
2244 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2246 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2250 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2258 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2259 int (*accessor)(struct kvm_device *dev,
2260 struct kvm_device_attr *attr),
2263 struct kvm_device_attr attr;
2268 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2271 return accessor(dev, &attr);
2274 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2277 struct kvm_device *dev = filp->private_data;
2280 case KVM_SET_DEVICE_ATTR:
2281 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2282 case KVM_GET_DEVICE_ATTR:
2283 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2284 case KVM_HAS_DEVICE_ATTR:
2285 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2287 if (dev->ops->ioctl)
2288 return dev->ops->ioctl(dev, ioctl, arg);
2294 static int kvm_device_release(struct inode *inode, struct file *filp)
2296 struct kvm_device *dev = filp->private_data;
2297 struct kvm *kvm = dev->kvm;
2303 static const struct file_operations kvm_device_fops = {
2304 .unlocked_ioctl = kvm_device_ioctl,
2305 #ifdef CONFIG_COMPAT
2306 .compat_ioctl = kvm_device_ioctl,
2308 .release = kvm_device_release,
2311 struct kvm_device *kvm_device_from_filp(struct file *filp)
2313 if (filp->f_op != &kvm_device_fops)
2316 return filp->private_data;
2319 static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
2320 #ifdef CONFIG_KVM_MPIC
2321 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
2322 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
2325 #ifdef CONFIG_KVM_XICS
2326 [KVM_DEV_TYPE_XICS] = &kvm_xics_ops,
2330 int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
2332 if (type >= ARRAY_SIZE(kvm_device_ops_table))
2335 if (kvm_device_ops_table[type] != NULL)
2338 kvm_device_ops_table[type] = ops;
2342 static int kvm_ioctl_create_device(struct kvm *kvm,
2343 struct kvm_create_device *cd)
2345 struct kvm_device_ops *ops = NULL;
2346 struct kvm_device *dev;
2347 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2350 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
2353 ops = kvm_device_ops_table[cd->type];
2360 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2367 ret = ops->create(dev, cd->type);
2373 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2379 list_add(&dev->vm_node, &kvm->devices);
2385 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2388 case KVM_CAP_USER_MEMORY:
2389 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2390 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2391 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2392 case KVM_CAP_SET_BOOT_CPU_ID:
2394 case KVM_CAP_INTERNAL_ERROR_DATA:
2395 #ifdef CONFIG_HAVE_KVM_MSI
2396 case KVM_CAP_SIGNAL_MSI:
2398 #ifdef CONFIG_HAVE_KVM_IRQFD
2399 case KVM_CAP_IRQFD_RESAMPLE:
2401 case KVM_CAP_CHECK_EXTENSION_VM:
2403 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2404 case KVM_CAP_IRQ_ROUTING:
2405 return KVM_MAX_IRQ_ROUTES;
2410 return kvm_vm_ioctl_check_extension(kvm, arg);
2413 static long kvm_vm_ioctl(struct file *filp,
2414 unsigned int ioctl, unsigned long arg)
2416 struct kvm *kvm = filp->private_data;
2417 void __user *argp = (void __user *)arg;
2420 if (kvm->mm != current->mm)
2423 case KVM_CREATE_VCPU:
2424 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2426 case KVM_SET_USER_MEMORY_REGION: {
2427 struct kvm_userspace_memory_region kvm_userspace_mem;
2430 if (copy_from_user(&kvm_userspace_mem, argp,
2431 sizeof kvm_userspace_mem))
2434 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2437 case KVM_GET_DIRTY_LOG: {
2438 struct kvm_dirty_log log;
2441 if (copy_from_user(&log, argp, sizeof log))
2443 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2446 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2447 case KVM_REGISTER_COALESCED_MMIO: {
2448 struct kvm_coalesced_mmio_zone zone;
2450 if (copy_from_user(&zone, argp, sizeof zone))
2452 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2455 case KVM_UNREGISTER_COALESCED_MMIO: {
2456 struct kvm_coalesced_mmio_zone zone;
2458 if (copy_from_user(&zone, argp, sizeof zone))
2460 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2465 struct kvm_irqfd data;
2468 if (copy_from_user(&data, argp, sizeof data))
2470 r = kvm_irqfd(kvm, &data);
2473 case KVM_IOEVENTFD: {
2474 struct kvm_ioeventfd data;
2477 if (copy_from_user(&data, argp, sizeof data))
2479 r = kvm_ioeventfd(kvm, &data);
2482 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2483 case KVM_SET_BOOT_CPU_ID:
2485 mutex_lock(&kvm->lock);
2486 if (atomic_read(&kvm->online_vcpus) != 0)
2489 kvm->bsp_vcpu_id = arg;
2490 mutex_unlock(&kvm->lock);
2493 #ifdef CONFIG_HAVE_KVM_MSI
2494 case KVM_SIGNAL_MSI: {
2498 if (copy_from_user(&msi, argp, sizeof msi))
2500 r = kvm_send_userspace_msi(kvm, &msi);
2504 #ifdef __KVM_HAVE_IRQ_LINE
2505 case KVM_IRQ_LINE_STATUS:
2506 case KVM_IRQ_LINE: {
2507 struct kvm_irq_level irq_event;
2510 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2513 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
2514 ioctl == KVM_IRQ_LINE_STATUS);
2519 if (ioctl == KVM_IRQ_LINE_STATUS) {
2520 if (copy_to_user(argp, &irq_event, sizeof irq_event))
2528 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2529 case KVM_SET_GSI_ROUTING: {
2530 struct kvm_irq_routing routing;
2531 struct kvm_irq_routing __user *urouting;
2532 struct kvm_irq_routing_entry *entries;
2535 if (copy_from_user(&routing, argp, sizeof(routing)))
2538 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2543 entries = vmalloc(routing.nr * sizeof(*entries));
2548 if (copy_from_user(entries, urouting->entries,
2549 routing.nr * sizeof(*entries)))
2550 goto out_free_irq_routing;
2551 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2553 out_free_irq_routing:
2557 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
2558 case KVM_CREATE_DEVICE: {
2559 struct kvm_create_device cd;
2562 if (copy_from_user(&cd, argp, sizeof(cd)))
2565 r = kvm_ioctl_create_device(kvm, &cd);
2570 if (copy_to_user(argp, &cd, sizeof(cd)))
2576 case KVM_CHECK_EXTENSION:
2577 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
2580 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2582 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2588 #ifdef CONFIG_COMPAT
2589 struct compat_kvm_dirty_log {
2593 compat_uptr_t dirty_bitmap; /* one bit per page */
2598 static long kvm_vm_compat_ioctl(struct file *filp,
2599 unsigned int ioctl, unsigned long arg)
2601 struct kvm *kvm = filp->private_data;
2604 if (kvm->mm != current->mm)
2607 case KVM_GET_DIRTY_LOG: {
2608 struct compat_kvm_dirty_log compat_log;
2609 struct kvm_dirty_log log;
2612 if (copy_from_user(&compat_log, (void __user *)arg,
2613 sizeof(compat_log)))
2615 log.slot = compat_log.slot;
2616 log.padding1 = compat_log.padding1;
2617 log.padding2 = compat_log.padding2;
2618 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2620 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2624 r = kvm_vm_ioctl(filp, ioctl, arg);
2632 static struct file_operations kvm_vm_fops = {
2633 .release = kvm_vm_release,
2634 .unlocked_ioctl = kvm_vm_ioctl,
2635 #ifdef CONFIG_COMPAT
2636 .compat_ioctl = kvm_vm_compat_ioctl,
2638 .llseek = noop_llseek,
2641 static int kvm_dev_ioctl_create_vm(unsigned long type)
2646 kvm = kvm_create_vm(type);
2648 return PTR_ERR(kvm);
2649 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2650 r = kvm_coalesced_mmio_init(kvm);
2656 r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC);
2663 static long kvm_dev_ioctl(struct file *filp,
2664 unsigned int ioctl, unsigned long arg)
2669 case KVM_GET_API_VERSION:
2672 r = KVM_API_VERSION;
2675 r = kvm_dev_ioctl_create_vm(arg);
2677 case KVM_CHECK_EXTENSION:
2678 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
2680 case KVM_GET_VCPU_MMAP_SIZE:
2683 r = PAGE_SIZE; /* struct kvm_run */
2685 r += PAGE_SIZE; /* pio data page */
2687 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2688 r += PAGE_SIZE; /* coalesced mmio ring page */
2691 case KVM_TRACE_ENABLE:
2692 case KVM_TRACE_PAUSE:
2693 case KVM_TRACE_DISABLE:
2697 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2703 static struct file_operations kvm_chardev_ops = {
2704 .unlocked_ioctl = kvm_dev_ioctl,
2705 .compat_ioctl = kvm_dev_ioctl,
2706 .llseek = noop_llseek,
2709 static struct miscdevice kvm_dev = {
2715 static void hardware_enable_nolock(void *junk)
2717 int cpu = raw_smp_processor_id();
2720 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2723 cpumask_set_cpu(cpu, cpus_hardware_enabled);
2725 r = kvm_arch_hardware_enable();
2728 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2729 atomic_inc(&hardware_enable_failed);
2730 printk(KERN_INFO "kvm: enabling virtualization on "
2731 "CPU%d failed\n", cpu);
2735 static void hardware_enable(void)
2737 raw_spin_lock(&kvm_count_lock);
2738 if (kvm_usage_count)
2739 hardware_enable_nolock(NULL);
2740 raw_spin_unlock(&kvm_count_lock);
2743 static void hardware_disable_nolock(void *junk)
2745 int cpu = raw_smp_processor_id();
2747 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2749 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2750 kvm_arch_hardware_disable();
2753 static void hardware_disable(void)
2755 raw_spin_lock(&kvm_count_lock);
2756 if (kvm_usage_count)
2757 hardware_disable_nolock(NULL);
2758 raw_spin_unlock(&kvm_count_lock);
2761 static void hardware_disable_all_nolock(void)
2763 BUG_ON(!kvm_usage_count);
2766 if (!kvm_usage_count)
2767 on_each_cpu(hardware_disable_nolock, NULL, 1);
2770 static void hardware_disable_all(void)
2772 raw_spin_lock(&kvm_count_lock);
2773 hardware_disable_all_nolock();
2774 raw_spin_unlock(&kvm_count_lock);
2777 static int hardware_enable_all(void)
2781 raw_spin_lock(&kvm_count_lock);
2784 if (kvm_usage_count == 1) {
2785 atomic_set(&hardware_enable_failed, 0);
2786 on_each_cpu(hardware_enable_nolock, NULL, 1);
2788 if (atomic_read(&hardware_enable_failed)) {
2789 hardware_disable_all_nolock();
2794 raw_spin_unlock(&kvm_count_lock);
2799 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2804 val &= ~CPU_TASKS_FROZEN;
2807 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2812 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2820 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2824 * Some (well, at least mine) BIOSes hang on reboot if
2827 * And Intel TXT required VMX off for all cpu when system shutdown.
2829 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2830 kvm_rebooting = true;
2831 on_each_cpu(hardware_disable_nolock, NULL, 1);
2835 static struct notifier_block kvm_reboot_notifier = {
2836 .notifier_call = kvm_reboot,
2840 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2844 for (i = 0; i < bus->dev_count; i++) {
2845 struct kvm_io_device *pos = bus->range[i].dev;
2847 kvm_iodevice_destructor(pos);
2852 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
2853 const struct kvm_io_range *r2)
2855 if (r1->addr < r2->addr)
2857 if (r1->addr + r1->len > r2->addr + r2->len)
2862 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2864 return kvm_io_bus_cmp(p1, p2);
2867 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2868 gpa_t addr, int len)
2870 bus->range[bus->dev_count++] = (struct kvm_io_range) {
2876 sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2877 kvm_io_bus_sort_cmp, NULL);
2882 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2883 gpa_t addr, int len)
2885 struct kvm_io_range *range, key;
2888 key = (struct kvm_io_range) {
2893 range = bsearch(&key, bus->range, bus->dev_count,
2894 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2898 off = range - bus->range;
2900 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
2906 static int __kvm_io_bus_write(struct kvm_io_bus *bus,
2907 struct kvm_io_range *range, const void *val)
2911 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2915 while (idx < bus->dev_count &&
2916 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
2917 if (!kvm_iodevice_write(bus->range[idx].dev, range->addr,
2926 /* kvm_io_bus_write - called under kvm->slots_lock */
2927 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2928 int len, const void *val)
2930 struct kvm_io_bus *bus;
2931 struct kvm_io_range range;
2934 range = (struct kvm_io_range) {
2939 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2940 r = __kvm_io_bus_write(bus, &range, val);
2941 return r < 0 ? r : 0;
2944 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
2945 int kvm_io_bus_write_cookie(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2946 int len, const void *val, long cookie)
2948 struct kvm_io_bus *bus;
2949 struct kvm_io_range range;
2951 range = (struct kvm_io_range) {
2956 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2958 /* First try the device referenced by cookie. */
2959 if ((cookie >= 0) && (cookie < bus->dev_count) &&
2960 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
2961 if (!kvm_iodevice_write(bus->range[cookie].dev, addr, len,
2966 * cookie contained garbage; fall back to search and return the
2967 * correct cookie value.
2969 return __kvm_io_bus_write(bus, &range, val);
2972 static int __kvm_io_bus_read(struct kvm_io_bus *bus, struct kvm_io_range *range,
2977 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2981 while (idx < bus->dev_count &&
2982 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
2983 if (!kvm_iodevice_read(bus->range[idx].dev, range->addr,
2991 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
2993 /* kvm_io_bus_read - called under kvm->slots_lock */
2994 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2997 struct kvm_io_bus *bus;
2998 struct kvm_io_range range;
3001 range = (struct kvm_io_range) {
3006 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
3007 r = __kvm_io_bus_read(bus, &range, val);
3008 return r < 0 ? r : 0;
3012 /* Caller must hold slots_lock. */
3013 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3014 int len, struct kvm_io_device *dev)
3016 struct kvm_io_bus *new_bus, *bus;
3018 bus = kvm->buses[bus_idx];
3019 /* exclude ioeventfd which is limited by maximum fd */
3020 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3023 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
3024 sizeof(struct kvm_io_range)), GFP_KERNEL);
3027 memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
3028 sizeof(struct kvm_io_range)));
3029 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
3030 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3031 synchronize_srcu_expedited(&kvm->srcu);
3037 /* Caller must hold slots_lock. */
3038 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3039 struct kvm_io_device *dev)
3042 struct kvm_io_bus *new_bus, *bus;
3044 bus = kvm->buses[bus_idx];
3046 for (i = 0; i < bus->dev_count; i++)
3047 if (bus->range[i].dev == dev) {
3055 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3056 sizeof(struct kvm_io_range)), GFP_KERNEL);
3060 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3061 new_bus->dev_count--;
3062 memcpy(new_bus->range + i, bus->range + i + 1,
3063 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3065 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3066 synchronize_srcu_expedited(&kvm->srcu);
3071 static struct notifier_block kvm_cpu_notifier = {
3072 .notifier_call = kvm_cpu_hotplug,
3075 static int vm_stat_get(void *_offset, u64 *val)
3077 unsigned offset = (long)_offset;
3081 spin_lock(&kvm_lock);
3082 list_for_each_entry(kvm, &vm_list, vm_list)
3083 *val += *(u32 *)((void *)kvm + offset);
3084 spin_unlock(&kvm_lock);
3088 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
3090 static int vcpu_stat_get(void *_offset, u64 *val)
3092 unsigned offset = (long)_offset;
3094 struct kvm_vcpu *vcpu;
3098 spin_lock(&kvm_lock);
3099 list_for_each_entry(kvm, &vm_list, vm_list)
3100 kvm_for_each_vcpu(i, vcpu, kvm)
3101 *val += *(u32 *)((void *)vcpu + offset);
3103 spin_unlock(&kvm_lock);
3107 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
3109 static const struct file_operations *stat_fops[] = {
3110 [KVM_STAT_VCPU] = &vcpu_stat_fops,
3111 [KVM_STAT_VM] = &vm_stat_fops,
3114 static int kvm_init_debug(void)
3117 struct kvm_stats_debugfs_item *p;
3119 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3120 if (kvm_debugfs_dir == NULL)
3123 for (p = debugfs_entries; p->name; ++p) {
3124 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
3125 (void *)(long)p->offset,
3126 stat_fops[p->kind]);
3127 if (p->dentry == NULL)
3134 debugfs_remove_recursive(kvm_debugfs_dir);
3139 static void kvm_exit_debug(void)
3141 struct kvm_stats_debugfs_item *p;
3143 for (p = debugfs_entries; p->name; ++p)
3144 debugfs_remove(p->dentry);
3145 debugfs_remove(kvm_debugfs_dir);
3148 static int kvm_suspend(void)
3150 if (kvm_usage_count)
3151 hardware_disable_nolock(NULL);
3155 static void kvm_resume(void)
3157 if (kvm_usage_count) {
3158 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3159 hardware_enable_nolock(NULL);
3163 static struct syscore_ops kvm_syscore_ops = {
3164 .suspend = kvm_suspend,
3165 .resume = kvm_resume,
3169 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3171 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3174 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3176 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3177 if (vcpu->preempted)
3178 vcpu->preempted = false;
3180 kvm_arch_sched_in(vcpu, cpu);
3182 kvm_arch_vcpu_load(vcpu, cpu);
3185 static void kvm_sched_out(struct preempt_notifier *pn,
3186 struct task_struct *next)
3188 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3190 if (current->state == TASK_RUNNING)
3191 vcpu->preempted = true;
3192 kvm_arch_vcpu_put(vcpu);
3195 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3196 struct module *module)
3201 r = kvm_arch_init(opaque);
3206 * kvm_arch_init makes sure there's at most one caller
3207 * for architectures that support multiple implementations,
3208 * like intel and amd on x86.
3209 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3210 * conflicts in case kvm is already setup for another implementation.
3212 r = kvm_irqfd_init();
3216 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
3221 r = kvm_arch_hardware_setup();
3225 for_each_online_cpu(cpu) {
3226 smp_call_function_single(cpu,
3227 kvm_arch_check_processor_compat,
3233 r = register_cpu_notifier(&kvm_cpu_notifier);
3236 register_reboot_notifier(&kvm_reboot_notifier);
3238 /* A kmem cache lets us meet the alignment requirements of fx_save. */
3240 vcpu_align = __alignof__(struct kvm_vcpu);
3241 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
3243 if (!kvm_vcpu_cache) {
3248 r = kvm_async_pf_init();
3252 kvm_chardev_ops.owner = module;
3253 kvm_vm_fops.owner = module;
3254 kvm_vcpu_fops.owner = module;
3256 r = misc_register(&kvm_dev);
3258 printk(KERN_ERR "kvm: misc device register failed\n");
3262 register_syscore_ops(&kvm_syscore_ops);
3264 kvm_preempt_ops.sched_in = kvm_sched_in;
3265 kvm_preempt_ops.sched_out = kvm_sched_out;
3267 r = kvm_init_debug();
3269 printk(KERN_ERR "kvm: create debugfs files failed\n");
3273 r = kvm_vfio_ops_init();
3279 unregister_syscore_ops(&kvm_syscore_ops);
3280 misc_deregister(&kvm_dev);
3282 kvm_async_pf_deinit();
3284 kmem_cache_destroy(kvm_vcpu_cache);
3286 unregister_reboot_notifier(&kvm_reboot_notifier);
3287 unregister_cpu_notifier(&kvm_cpu_notifier);
3290 kvm_arch_hardware_unsetup();
3292 free_cpumask_var(cpus_hardware_enabled);
3300 EXPORT_SYMBOL_GPL(kvm_init);
3305 misc_deregister(&kvm_dev);
3306 kmem_cache_destroy(kvm_vcpu_cache);
3307 kvm_async_pf_deinit();
3308 unregister_syscore_ops(&kvm_syscore_ops);
3309 unregister_reboot_notifier(&kvm_reboot_notifier);
3310 unregister_cpu_notifier(&kvm_cpu_notifier);
3311 on_each_cpu(hardware_disable_nolock, NULL, 1);
3312 kvm_arch_hardware_unsetup();
3315 free_cpumask_var(cpus_hardware_enabled);
3317 EXPORT_SYMBOL_GPL(kvm_exit);