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"
61 #define CREATE_TRACE_POINTS
62 #include <trace/events/kvm.h>
64 MODULE_AUTHOR("Qumranet");
65 MODULE_LICENSE("GPL");
70 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
73 DEFINE_SPINLOCK(kvm_lock);
74 static DEFINE_RAW_SPINLOCK(kvm_count_lock);
77 static cpumask_var_t cpus_hardware_enabled;
78 static int kvm_usage_count = 0;
79 static atomic_t hardware_enable_failed;
81 struct kmem_cache *kvm_vcpu_cache;
82 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
84 static __read_mostly struct preempt_ops kvm_preempt_ops;
86 struct dentry *kvm_debugfs_dir;
88 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
91 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
94 static int hardware_enable_all(void);
95 static void hardware_disable_all(void);
97 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
98 static void update_memslots(struct kvm_memslots *slots,
99 struct kvm_memory_slot *new, u64 last_generation);
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_mmio_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))
127 if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
128 /* The thread running this VCPU changed. */
129 struct pid *oldpid = vcpu->pid;
130 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
131 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);
468 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
471 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
474 kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
476 goto out_err_no_srcu;
477 kvm_init_memslots_id(kvm);
478 if (init_srcu_struct(&kvm->srcu))
479 goto out_err_no_srcu;
480 if (init_srcu_struct(&kvm->irq_srcu))
481 goto out_err_no_irq_srcu;
482 for (i = 0; i < KVM_NR_BUSES; i++) {
483 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
489 spin_lock_init(&kvm->mmu_lock);
490 kvm->mm = current->mm;
491 atomic_inc(&kvm->mm->mm_count);
492 kvm_eventfd_init(kvm);
493 mutex_init(&kvm->lock);
494 mutex_init(&kvm->irq_lock);
495 mutex_init(&kvm->slots_lock);
496 atomic_set(&kvm->users_count, 1);
497 INIT_LIST_HEAD(&kvm->devices);
499 r = kvm_init_mmu_notifier(kvm);
503 spin_lock(&kvm_lock);
504 list_add(&kvm->vm_list, &vm_list);
505 spin_unlock(&kvm_lock);
510 cleanup_srcu_struct(&kvm->irq_srcu);
512 cleanup_srcu_struct(&kvm->srcu);
514 hardware_disable_all();
516 for (i = 0; i < KVM_NR_BUSES; i++)
517 kfree(kvm->buses[i]);
518 kfree(kvm->memslots);
519 kvm_arch_free_vm(kvm);
524 * Avoid using vmalloc for a small buffer.
525 * Should not be used when the size is statically known.
527 void *kvm_kvzalloc(unsigned long size)
529 if (size > PAGE_SIZE)
530 return vzalloc(size);
532 return kzalloc(size, GFP_KERNEL);
535 void kvm_kvfree(const void *addr)
537 if (is_vmalloc_addr(addr))
543 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
545 if (!memslot->dirty_bitmap)
548 kvm_kvfree(memslot->dirty_bitmap);
549 memslot->dirty_bitmap = NULL;
553 * Free any memory in @free but not in @dont.
555 static void kvm_free_physmem_slot(struct kvm *kvm, struct kvm_memory_slot *free,
556 struct kvm_memory_slot *dont)
558 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
559 kvm_destroy_dirty_bitmap(free);
561 kvm_arch_free_memslot(kvm, free, dont);
566 static void kvm_free_physmem(struct kvm *kvm)
568 struct kvm_memslots *slots = kvm->memslots;
569 struct kvm_memory_slot *memslot;
571 kvm_for_each_memslot(memslot, slots)
572 kvm_free_physmem_slot(kvm, memslot, NULL);
574 kfree(kvm->memslots);
577 static void kvm_destroy_devices(struct kvm *kvm)
579 struct list_head *node, *tmp;
581 list_for_each_safe(node, tmp, &kvm->devices) {
582 struct kvm_device *dev =
583 list_entry(node, struct kvm_device, vm_node);
586 dev->ops->destroy(dev);
590 static void kvm_destroy_vm(struct kvm *kvm)
593 struct mm_struct *mm = kvm->mm;
595 kvm_arch_sync_events(kvm);
596 spin_lock(&kvm_lock);
597 list_del(&kvm->vm_list);
598 spin_unlock(&kvm_lock);
599 kvm_free_irq_routing(kvm);
600 for (i = 0; i < KVM_NR_BUSES; i++)
601 kvm_io_bus_destroy(kvm->buses[i]);
602 kvm_coalesced_mmio_free(kvm);
603 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
604 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
606 kvm_arch_flush_shadow_all(kvm);
608 kvm_arch_destroy_vm(kvm);
609 kvm_destroy_devices(kvm);
610 kvm_free_physmem(kvm);
611 cleanup_srcu_struct(&kvm->srcu);
612 kvm_arch_free_vm(kvm);
613 hardware_disable_all();
617 void kvm_get_kvm(struct kvm *kvm)
619 atomic_inc(&kvm->users_count);
621 EXPORT_SYMBOL_GPL(kvm_get_kvm);
623 void kvm_put_kvm(struct kvm *kvm)
625 if (atomic_dec_and_test(&kvm->users_count))
628 EXPORT_SYMBOL_GPL(kvm_put_kvm);
631 static int kvm_vm_release(struct inode *inode, struct file *filp)
633 struct kvm *kvm = filp->private_data;
635 kvm_irqfd_release(kvm);
642 * Allocation size is twice as large as the actual dirty bitmap size.
643 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
645 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
647 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
649 memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
650 if (!memslot->dirty_bitmap)
656 static int cmp_memslot(const void *slot1, const void *slot2)
658 struct kvm_memory_slot *s1, *s2;
660 s1 = (struct kvm_memory_slot *)slot1;
661 s2 = (struct kvm_memory_slot *)slot2;
663 if (s1->npages < s2->npages)
665 if (s1->npages > s2->npages)
672 * Sort the memslots base on its size, so the larger slots
673 * will get better fit.
675 static void sort_memslots(struct kvm_memslots *slots)
679 sort(slots->memslots, KVM_MEM_SLOTS_NUM,
680 sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
682 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
683 slots->id_to_index[slots->memslots[i].id] = i;
686 static void update_memslots(struct kvm_memslots *slots,
687 struct kvm_memory_slot *new,
692 struct kvm_memory_slot *old = id_to_memslot(slots, id);
693 unsigned long npages = old->npages;
696 if (new->npages != npages)
697 sort_memslots(slots);
700 slots->generation = last_generation + 1;
703 static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
705 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
707 #ifdef KVM_CAP_READONLY_MEM
708 valid_flags |= KVM_MEM_READONLY;
711 if (mem->flags & ~valid_flags)
717 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
718 struct kvm_memslots *slots, struct kvm_memory_slot *new)
720 struct kvm_memslots *old_memslots = kvm->memslots;
722 update_memslots(slots, new, kvm->memslots->generation);
723 rcu_assign_pointer(kvm->memslots, slots);
724 synchronize_srcu_expedited(&kvm->srcu);
726 kvm_arch_memslots_updated(kvm);
732 * Allocate some memory and give it an address in the guest physical address
735 * Discontiguous memory is allowed, mostly for framebuffers.
737 * Must be called holding mmap_sem for write.
739 int __kvm_set_memory_region(struct kvm *kvm,
740 struct kvm_userspace_memory_region *mem)
744 unsigned long npages;
745 struct kvm_memory_slot *slot;
746 struct kvm_memory_slot old, new;
747 struct kvm_memslots *slots = NULL, *old_memslots;
748 enum kvm_mr_change change;
750 r = check_memory_region_flags(mem);
755 /* General sanity checks */
756 if (mem->memory_size & (PAGE_SIZE - 1))
758 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
760 /* We can read the guest memory with __xxx_user() later on. */
761 if ((mem->slot < KVM_USER_MEM_SLOTS) &&
762 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
763 !access_ok(VERIFY_WRITE,
764 (void __user *)(unsigned long)mem->userspace_addr,
767 if (mem->slot >= KVM_MEM_SLOTS_NUM)
769 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
772 slot = id_to_memslot(kvm->memslots, mem->slot);
773 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
774 npages = mem->memory_size >> PAGE_SHIFT;
777 if (npages > KVM_MEM_MAX_NR_PAGES)
781 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
786 new.base_gfn = base_gfn;
788 new.flags = mem->flags;
793 change = KVM_MR_CREATE;
794 else { /* Modify an existing slot. */
795 if ((mem->userspace_addr != old.userspace_addr) ||
796 (npages != old.npages) ||
797 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
800 if (base_gfn != old.base_gfn)
801 change = KVM_MR_MOVE;
802 else if (new.flags != old.flags)
803 change = KVM_MR_FLAGS_ONLY;
804 else { /* Nothing to change. */
809 } else if (old.npages) {
810 change = KVM_MR_DELETE;
811 } else /* Modify a non-existent slot: disallowed. */
814 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
815 /* Check for overlaps */
817 kvm_for_each_memslot(slot, kvm->memslots) {
818 if ((slot->id >= KVM_USER_MEM_SLOTS) ||
819 (slot->id == mem->slot))
821 if (!((base_gfn + npages <= slot->base_gfn) ||
822 (base_gfn >= slot->base_gfn + slot->npages)))
827 /* Free page dirty bitmap if unneeded */
828 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
829 new.dirty_bitmap = NULL;
832 if (change == KVM_MR_CREATE) {
833 new.userspace_addr = mem->userspace_addr;
835 if (kvm_arch_create_memslot(kvm, &new, npages))
839 /* Allocate page dirty bitmap if needed */
840 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
841 if (kvm_create_dirty_bitmap(&new) < 0)
845 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
847 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
851 slot = id_to_memslot(slots, mem->slot);
852 slot->flags |= KVM_MEMSLOT_INVALID;
854 old_memslots = install_new_memslots(kvm, slots, NULL);
856 /* slot was deleted or moved, clear iommu mapping */
857 kvm_iommu_unmap_pages(kvm, &old);
858 /* From this point no new shadow pages pointing to a deleted,
859 * or moved, memslot will be created.
861 * validation of sp->gfn happens in:
862 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
863 * - kvm_is_visible_gfn (mmu_check_roots)
865 kvm_arch_flush_shadow_memslot(kvm, slot);
866 slots = old_memslots;
869 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
875 * We can re-use the old_memslots from above, the only difference
876 * from the currently installed memslots is the invalid flag. This
877 * will get overwritten by update_memslots anyway.
880 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
886 /* actual memory is freed via old in kvm_free_physmem_slot below */
887 if (change == KVM_MR_DELETE) {
888 new.dirty_bitmap = NULL;
889 memset(&new.arch, 0, sizeof(new.arch));
892 old_memslots = install_new_memslots(kvm, slots, &new);
894 kvm_arch_commit_memory_region(kvm, mem, &old, change);
896 kvm_free_physmem_slot(kvm, &old, &new);
900 * IOMMU mapping: New slots need to be mapped. Old slots need to be
901 * un-mapped and re-mapped if their base changes. Since base change
902 * unmapping is handled above with slot deletion, mapping alone is
903 * needed here. Anything else the iommu might care about for existing
904 * slots (size changes, userspace addr changes and read-only flag
905 * changes) is disallowed above, so any other attribute changes getting
906 * here can be skipped.
908 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
909 r = kvm_iommu_map_pages(kvm, &new);
918 kvm_free_physmem_slot(kvm, &new, &old);
922 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
924 int kvm_set_memory_region(struct kvm *kvm,
925 struct kvm_userspace_memory_region *mem)
929 mutex_lock(&kvm->slots_lock);
930 r = __kvm_set_memory_region(kvm, mem);
931 mutex_unlock(&kvm->slots_lock);
934 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
936 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
937 struct kvm_userspace_memory_region *mem)
939 if (mem->slot >= KVM_USER_MEM_SLOTS)
941 return kvm_set_memory_region(kvm, mem);
944 int kvm_get_dirty_log(struct kvm *kvm,
945 struct kvm_dirty_log *log, int *is_dirty)
947 struct kvm_memory_slot *memslot;
950 unsigned long any = 0;
953 if (log->slot >= KVM_USER_MEM_SLOTS)
956 memslot = id_to_memslot(kvm->memslots, log->slot);
958 if (!memslot->dirty_bitmap)
961 n = kvm_dirty_bitmap_bytes(memslot);
963 for (i = 0; !any && i < n/sizeof(long); ++i)
964 any = memslot->dirty_bitmap[i];
967 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
977 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
979 bool kvm_largepages_enabled(void)
981 return largepages_enabled;
984 void kvm_disable_largepages(void)
986 largepages_enabled = false;
988 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
990 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
992 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
994 EXPORT_SYMBOL_GPL(gfn_to_memslot);
996 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
998 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1000 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1001 memslot->flags & KVM_MEMSLOT_INVALID)
1006 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1008 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1010 struct vm_area_struct *vma;
1011 unsigned long addr, size;
1015 addr = gfn_to_hva(kvm, gfn);
1016 if (kvm_is_error_hva(addr))
1019 down_read(¤t->mm->mmap_sem);
1020 vma = find_vma(current->mm, addr);
1024 size = vma_kernel_pagesize(vma);
1027 up_read(¤t->mm->mmap_sem);
1032 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1034 return slot->flags & KVM_MEM_READONLY;
1037 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1038 gfn_t *nr_pages, bool write)
1040 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1041 return KVM_HVA_ERR_BAD;
1043 if (memslot_is_readonly(slot) && write)
1044 return KVM_HVA_ERR_RO_BAD;
1047 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1049 return __gfn_to_hva_memslot(slot, gfn);
1052 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1055 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1058 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1061 return gfn_to_hva_many(slot, gfn, NULL);
1063 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1065 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1067 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1069 EXPORT_SYMBOL_GPL(gfn_to_hva);
1072 * If writable is set to false, the hva returned by this function is only
1073 * allowed to be read.
1075 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1077 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1078 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1080 if (!kvm_is_error_hva(hva) && writable)
1081 *writable = !memslot_is_readonly(slot);
1086 static int kvm_read_hva(void *data, void __user *hva, int len)
1088 return __copy_from_user(data, hva, len);
1091 static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1093 return __copy_from_user_inatomic(data, hva, len);
1096 static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1097 unsigned long start, int write, struct page **page)
1099 int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1102 flags |= FOLL_WRITE;
1104 return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1107 static inline int check_user_page_hwpoison(unsigned long addr)
1109 int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1111 rc = __get_user_pages(current, current->mm, addr, 1,
1112 flags, NULL, NULL, NULL);
1113 return rc == -EHWPOISON;
1117 * The atomic path to get the writable pfn which will be stored in @pfn,
1118 * true indicates success, otherwise false is returned.
1120 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1121 bool write_fault, bool *writable, pfn_t *pfn)
1123 struct page *page[1];
1126 if (!(async || atomic))
1130 * Fast pin a writable pfn only if it is a write fault request
1131 * or the caller allows to map a writable pfn for a read fault
1134 if (!(write_fault || writable))
1137 npages = __get_user_pages_fast(addr, 1, 1, page);
1139 *pfn = page_to_pfn(page[0]);
1150 * The slow path to get the pfn of the specified host virtual address,
1151 * 1 indicates success, -errno is returned if error is detected.
1153 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1154 bool *writable, pfn_t *pfn)
1156 struct page *page[1];
1162 *writable = write_fault;
1165 down_read(¤t->mm->mmap_sem);
1166 npages = get_user_page_nowait(current, current->mm,
1167 addr, write_fault, page);
1168 up_read(¤t->mm->mmap_sem);
1170 npages = get_user_pages_fast(addr, 1, write_fault,
1175 /* map read fault as writable if possible */
1176 if (unlikely(!write_fault) && writable) {
1177 struct page *wpage[1];
1179 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1188 *pfn = page_to_pfn(page[0]);
1192 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1194 if (unlikely(!(vma->vm_flags & VM_READ)))
1197 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1204 * Pin guest page in memory and return its pfn.
1205 * @addr: host virtual address which maps memory to the guest
1206 * @atomic: whether this function can sleep
1207 * @async: whether this function need to wait IO complete if the
1208 * host page is not in the memory
1209 * @write_fault: whether we should get a writable host page
1210 * @writable: whether it allows to map a writable host page for !@write_fault
1212 * The function will map a writable host page for these two cases:
1213 * 1): @write_fault = true
1214 * 2): @write_fault = false && @writable, @writable will tell the caller
1215 * whether the mapping is writable.
1217 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1218 bool write_fault, bool *writable)
1220 struct vm_area_struct *vma;
1224 /* we can do it either atomically or asynchronously, not both */
1225 BUG_ON(atomic && async);
1227 if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1231 return KVM_PFN_ERR_FAULT;
1233 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1237 down_read(¤t->mm->mmap_sem);
1238 if (npages == -EHWPOISON ||
1239 (!async && check_user_page_hwpoison(addr))) {
1240 pfn = KVM_PFN_ERR_HWPOISON;
1244 vma = find_vma_intersection(current->mm, addr, addr + 1);
1247 pfn = KVM_PFN_ERR_FAULT;
1248 else if ((vma->vm_flags & VM_PFNMAP)) {
1249 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1251 BUG_ON(!kvm_is_mmio_pfn(pfn));
1253 if (async && vma_is_valid(vma, write_fault))
1255 pfn = KVM_PFN_ERR_FAULT;
1258 up_read(¤t->mm->mmap_sem);
1263 __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1264 bool *async, bool write_fault, bool *writable)
1266 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1268 if (addr == KVM_HVA_ERR_RO_BAD)
1269 return KVM_PFN_ERR_RO_FAULT;
1271 if (kvm_is_error_hva(addr))
1272 return KVM_PFN_NOSLOT;
1274 /* Do not map writable pfn in the readonly memslot. */
1275 if (writable && memslot_is_readonly(slot)) {
1280 return hva_to_pfn(addr, atomic, async, write_fault,
1284 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1285 bool write_fault, bool *writable)
1287 struct kvm_memory_slot *slot;
1292 slot = gfn_to_memslot(kvm, gfn);
1294 return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1298 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1300 return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1302 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1304 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1305 bool write_fault, bool *writable)
1307 return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1309 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1311 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1313 return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1315 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1317 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1320 return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1322 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1324 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1326 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1329 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1331 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1333 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1335 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1341 addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1342 if (kvm_is_error_hva(addr))
1345 if (entry < nr_pages)
1348 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1350 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1352 static struct page *kvm_pfn_to_page(pfn_t pfn)
1354 if (is_error_noslot_pfn(pfn))
1355 return KVM_ERR_PTR_BAD_PAGE;
1357 if (kvm_is_mmio_pfn(pfn)) {
1359 return KVM_ERR_PTR_BAD_PAGE;
1362 return pfn_to_page(pfn);
1365 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1369 pfn = gfn_to_pfn(kvm, gfn);
1371 return kvm_pfn_to_page(pfn);
1374 EXPORT_SYMBOL_GPL(gfn_to_page);
1376 void kvm_release_page_clean(struct page *page)
1378 WARN_ON(is_error_page(page));
1380 kvm_release_pfn_clean(page_to_pfn(page));
1382 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1384 void kvm_release_pfn_clean(pfn_t pfn)
1386 if (!is_error_noslot_pfn(pfn) && !kvm_is_mmio_pfn(pfn))
1387 put_page(pfn_to_page(pfn));
1389 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1391 void kvm_release_page_dirty(struct page *page)
1393 WARN_ON(is_error_page(page));
1395 kvm_release_pfn_dirty(page_to_pfn(page));
1397 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1399 static void kvm_release_pfn_dirty(pfn_t pfn)
1401 kvm_set_pfn_dirty(pfn);
1402 kvm_release_pfn_clean(pfn);
1405 void kvm_set_pfn_dirty(pfn_t pfn)
1407 if (!kvm_is_mmio_pfn(pfn)) {
1408 struct page *page = pfn_to_page(pfn);
1409 if (!PageReserved(page))
1413 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1415 void kvm_set_pfn_accessed(pfn_t pfn)
1417 if (!kvm_is_mmio_pfn(pfn))
1418 mark_page_accessed(pfn_to_page(pfn));
1420 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1422 void kvm_get_pfn(pfn_t pfn)
1424 if (!kvm_is_mmio_pfn(pfn))
1425 get_page(pfn_to_page(pfn));
1427 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1429 static int next_segment(unsigned long len, int offset)
1431 if (len > PAGE_SIZE - offset)
1432 return PAGE_SIZE - offset;
1437 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1443 addr = gfn_to_hva_prot(kvm, gfn, NULL);
1444 if (kvm_is_error_hva(addr))
1446 r = kvm_read_hva(data, (void __user *)addr + offset, len);
1451 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1453 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1455 gfn_t gfn = gpa >> PAGE_SHIFT;
1457 int offset = offset_in_page(gpa);
1460 while ((seg = next_segment(len, offset)) != 0) {
1461 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1471 EXPORT_SYMBOL_GPL(kvm_read_guest);
1473 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1478 gfn_t gfn = gpa >> PAGE_SHIFT;
1479 int offset = offset_in_page(gpa);
1481 addr = gfn_to_hva_prot(kvm, gfn, NULL);
1482 if (kvm_is_error_hva(addr))
1484 pagefault_disable();
1485 r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
1491 EXPORT_SYMBOL(kvm_read_guest_atomic);
1493 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1494 int offset, int len)
1499 addr = gfn_to_hva(kvm, gfn);
1500 if (kvm_is_error_hva(addr))
1502 r = __copy_to_user((void __user *)addr + offset, data, len);
1505 mark_page_dirty(kvm, gfn);
1508 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1510 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1513 gfn_t gfn = gpa >> PAGE_SHIFT;
1515 int offset = offset_in_page(gpa);
1518 while ((seg = next_segment(len, offset)) != 0) {
1519 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1530 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1531 gpa_t gpa, unsigned long len)
1533 struct kvm_memslots *slots = kvm_memslots(kvm);
1534 int offset = offset_in_page(gpa);
1535 gfn_t start_gfn = gpa >> PAGE_SHIFT;
1536 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1537 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1538 gfn_t nr_pages_avail;
1541 ghc->generation = slots->generation;
1543 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1544 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, &nr_pages_avail);
1545 if (!kvm_is_error_hva(ghc->hva) && nr_pages_avail >= nr_pages_needed) {
1549 * If the requested region crosses two memslots, we still
1550 * verify that the entire region is valid here.
1552 while (start_gfn <= end_gfn) {
1553 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1554 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1556 if (kvm_is_error_hva(ghc->hva))
1558 start_gfn += nr_pages_avail;
1560 /* Use the slow path for cross page reads and writes. */
1561 ghc->memslot = NULL;
1565 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1567 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1568 void *data, unsigned long len)
1570 struct kvm_memslots *slots = kvm_memslots(kvm);
1573 BUG_ON(len > ghc->len);
1575 if (slots->generation != ghc->generation)
1576 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1578 if (unlikely(!ghc->memslot))
1579 return kvm_write_guest(kvm, ghc->gpa, data, len);
1581 if (kvm_is_error_hva(ghc->hva))
1584 r = __copy_to_user((void __user *)ghc->hva, data, len);
1587 mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1591 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1593 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1594 void *data, unsigned long len)
1596 struct kvm_memslots *slots = kvm_memslots(kvm);
1599 BUG_ON(len > ghc->len);
1601 if (slots->generation != ghc->generation)
1602 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1604 if (unlikely(!ghc->memslot))
1605 return kvm_read_guest(kvm, ghc->gpa, data, len);
1607 if (kvm_is_error_hva(ghc->hva))
1610 r = __copy_from_user(data, (void __user *)ghc->hva, len);
1616 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1618 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1620 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
1622 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
1624 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1626 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1628 gfn_t gfn = gpa >> PAGE_SHIFT;
1630 int offset = offset_in_page(gpa);
1633 while ((seg = next_segment(len, offset)) != 0) {
1634 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1643 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1645 static void mark_page_dirty_in_slot(struct kvm *kvm,
1646 struct kvm_memory_slot *memslot,
1649 if (memslot && memslot->dirty_bitmap) {
1650 unsigned long rel_gfn = gfn - memslot->base_gfn;
1652 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1656 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1658 struct kvm_memory_slot *memslot;
1660 memslot = gfn_to_memslot(kvm, gfn);
1661 mark_page_dirty_in_slot(kvm, memslot, gfn);
1663 EXPORT_SYMBOL_GPL(mark_page_dirty);
1666 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1668 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1673 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1675 if (kvm_arch_vcpu_runnable(vcpu)) {
1676 kvm_make_request(KVM_REQ_UNHALT, vcpu);
1679 if (kvm_cpu_has_pending_timer(vcpu))
1681 if (signal_pending(current))
1687 finish_wait(&vcpu->wq, &wait);
1689 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
1693 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1695 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1698 int cpu = vcpu->cpu;
1699 wait_queue_head_t *wqp;
1701 wqp = kvm_arch_vcpu_wq(vcpu);
1702 if (waitqueue_active(wqp)) {
1703 wake_up_interruptible(wqp);
1704 ++vcpu->stat.halt_wakeup;
1708 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1709 if (kvm_arch_vcpu_should_kick(vcpu))
1710 smp_send_reschedule(cpu);
1713 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
1714 #endif /* !CONFIG_S390 */
1716 bool kvm_vcpu_yield_to(struct kvm_vcpu *target)
1719 struct task_struct *task = NULL;
1723 pid = rcu_dereference(target->pid);
1725 task = get_pid_task(target->pid, PIDTYPE_PID);
1729 if (task->flags & PF_VCPU) {
1730 put_task_struct(task);
1733 ret = yield_to(task, 1);
1734 put_task_struct(task);
1738 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1741 * Helper that checks whether a VCPU is eligible for directed yield.
1742 * Most eligible candidate to yield is decided by following heuristics:
1744 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1745 * (preempted lock holder), indicated by @in_spin_loop.
1746 * Set at the beiginning and cleared at the end of interception/PLE handler.
1748 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1749 * chance last time (mostly it has become eligible now since we have probably
1750 * yielded to lockholder in last iteration. This is done by toggling
1751 * @dy_eligible each time a VCPU checked for eligibility.)
1753 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1754 * to preempted lock-holder could result in wrong VCPU selection and CPU
1755 * burning. Giving priority for a potential lock-holder increases lock
1758 * Since algorithm is based on heuristics, accessing another VCPU data without
1759 * locking does not harm. It may result in trying to yield to same VCPU, fail
1760 * and continue with next VCPU and so on.
1762 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1764 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1767 eligible = !vcpu->spin_loop.in_spin_loop ||
1768 (vcpu->spin_loop.in_spin_loop &&
1769 vcpu->spin_loop.dy_eligible);
1771 if (vcpu->spin_loop.in_spin_loop)
1772 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1780 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1782 struct kvm *kvm = me->kvm;
1783 struct kvm_vcpu *vcpu;
1784 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1790 kvm_vcpu_set_in_spin_loop(me, true);
1792 * We boost the priority of a VCPU that is runnable but not
1793 * currently running, because it got preempted by something
1794 * else and called schedule in __vcpu_run. Hopefully that
1795 * VCPU is holding the lock that we need and will release it.
1796 * We approximate round-robin by starting at the last boosted VCPU.
1798 for (pass = 0; pass < 2 && !yielded && try; pass++) {
1799 kvm_for_each_vcpu(i, vcpu, kvm) {
1800 if (!pass && i <= last_boosted_vcpu) {
1801 i = last_boosted_vcpu;
1803 } else if (pass && i > last_boosted_vcpu)
1805 if (!ACCESS_ONCE(vcpu->preempted))
1809 if (waitqueue_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
1811 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1814 yielded = kvm_vcpu_yield_to(vcpu);
1816 kvm->last_boosted_vcpu = i;
1818 } else if (yielded < 0) {
1825 kvm_vcpu_set_in_spin_loop(me, false);
1827 /* Ensure vcpu is not eligible during next spinloop */
1828 kvm_vcpu_set_dy_eligible(me, false);
1830 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1832 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1834 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1837 if (vmf->pgoff == 0)
1838 page = virt_to_page(vcpu->run);
1840 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1841 page = virt_to_page(vcpu->arch.pio_data);
1843 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1844 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1845 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1848 return kvm_arch_vcpu_fault(vcpu, vmf);
1854 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1855 .fault = kvm_vcpu_fault,
1858 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1860 vma->vm_ops = &kvm_vcpu_vm_ops;
1864 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1866 struct kvm_vcpu *vcpu = filp->private_data;
1868 kvm_put_kvm(vcpu->kvm);
1872 static struct file_operations kvm_vcpu_fops = {
1873 .release = kvm_vcpu_release,
1874 .unlocked_ioctl = kvm_vcpu_ioctl,
1875 #ifdef CONFIG_COMPAT
1876 .compat_ioctl = kvm_vcpu_compat_ioctl,
1878 .mmap = kvm_vcpu_mmap,
1879 .llseek = noop_llseek,
1883 * Allocates an inode for the vcpu.
1885 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1887 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
1891 * Creates some virtual cpus. Good luck creating more than one.
1893 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1896 struct kvm_vcpu *vcpu, *v;
1898 if (id >= KVM_MAX_VCPUS)
1901 vcpu = kvm_arch_vcpu_create(kvm, id);
1903 return PTR_ERR(vcpu);
1905 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1907 r = kvm_arch_vcpu_setup(vcpu);
1911 mutex_lock(&kvm->lock);
1912 if (!kvm_vcpu_compatible(vcpu)) {
1914 goto unlock_vcpu_destroy;
1916 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1918 goto unlock_vcpu_destroy;
1921 kvm_for_each_vcpu(r, v, kvm)
1922 if (v->vcpu_id == id) {
1924 goto unlock_vcpu_destroy;
1927 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1929 /* Now it's all set up, let userspace reach it */
1931 r = create_vcpu_fd(vcpu);
1934 goto unlock_vcpu_destroy;
1937 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1939 atomic_inc(&kvm->online_vcpus);
1941 mutex_unlock(&kvm->lock);
1942 kvm_arch_vcpu_postcreate(vcpu);
1945 unlock_vcpu_destroy:
1946 mutex_unlock(&kvm->lock);
1948 kvm_arch_vcpu_destroy(vcpu);
1952 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1955 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1956 vcpu->sigset_active = 1;
1957 vcpu->sigset = *sigset;
1959 vcpu->sigset_active = 0;
1963 static long kvm_vcpu_ioctl(struct file *filp,
1964 unsigned int ioctl, unsigned long arg)
1966 struct kvm_vcpu *vcpu = filp->private_data;
1967 void __user *argp = (void __user *)arg;
1969 struct kvm_fpu *fpu = NULL;
1970 struct kvm_sregs *kvm_sregs = NULL;
1972 if (vcpu->kvm->mm != current->mm)
1975 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
1977 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1978 * so vcpu_load() would break it.
1980 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1981 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1985 r = vcpu_load(vcpu);
1993 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1994 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1996 case KVM_GET_REGS: {
1997 struct kvm_regs *kvm_regs;
2000 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2003 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2007 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2014 case KVM_SET_REGS: {
2015 struct kvm_regs *kvm_regs;
2018 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2019 if (IS_ERR(kvm_regs)) {
2020 r = PTR_ERR(kvm_regs);
2023 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2027 case KVM_GET_SREGS: {
2028 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2032 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2036 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2041 case KVM_SET_SREGS: {
2042 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2043 if (IS_ERR(kvm_sregs)) {
2044 r = PTR_ERR(kvm_sregs);
2048 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2051 case KVM_GET_MP_STATE: {
2052 struct kvm_mp_state mp_state;
2054 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2058 if (copy_to_user(argp, &mp_state, sizeof mp_state))
2063 case KVM_SET_MP_STATE: {
2064 struct kvm_mp_state mp_state;
2067 if (copy_from_user(&mp_state, argp, sizeof mp_state))
2069 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2072 case KVM_TRANSLATE: {
2073 struct kvm_translation tr;
2076 if (copy_from_user(&tr, argp, sizeof tr))
2078 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2082 if (copy_to_user(argp, &tr, sizeof tr))
2087 case KVM_SET_GUEST_DEBUG: {
2088 struct kvm_guest_debug dbg;
2091 if (copy_from_user(&dbg, argp, sizeof dbg))
2093 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2096 case KVM_SET_SIGNAL_MASK: {
2097 struct kvm_signal_mask __user *sigmask_arg = argp;
2098 struct kvm_signal_mask kvm_sigmask;
2099 sigset_t sigset, *p;
2104 if (copy_from_user(&kvm_sigmask, argp,
2105 sizeof kvm_sigmask))
2108 if (kvm_sigmask.len != sizeof sigset)
2111 if (copy_from_user(&sigset, sigmask_arg->sigset,
2116 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2120 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2124 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2128 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2134 fpu = memdup_user(argp, sizeof(*fpu));
2140 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2144 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2153 #ifdef CONFIG_COMPAT
2154 static long kvm_vcpu_compat_ioctl(struct file *filp,
2155 unsigned int ioctl, unsigned long arg)
2157 struct kvm_vcpu *vcpu = filp->private_data;
2158 void __user *argp = compat_ptr(arg);
2161 if (vcpu->kvm->mm != current->mm)
2165 case KVM_SET_SIGNAL_MASK: {
2166 struct kvm_signal_mask __user *sigmask_arg = argp;
2167 struct kvm_signal_mask kvm_sigmask;
2168 compat_sigset_t csigset;
2173 if (copy_from_user(&kvm_sigmask, argp,
2174 sizeof kvm_sigmask))
2177 if (kvm_sigmask.len != sizeof csigset)
2180 if (copy_from_user(&csigset, sigmask_arg->sigset,
2183 sigset_from_compat(&sigset, &csigset);
2184 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2186 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2190 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2198 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2199 int (*accessor)(struct kvm_device *dev,
2200 struct kvm_device_attr *attr),
2203 struct kvm_device_attr attr;
2208 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2211 return accessor(dev, &attr);
2214 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2217 struct kvm_device *dev = filp->private_data;
2220 case KVM_SET_DEVICE_ATTR:
2221 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2222 case KVM_GET_DEVICE_ATTR:
2223 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2224 case KVM_HAS_DEVICE_ATTR:
2225 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2227 if (dev->ops->ioctl)
2228 return dev->ops->ioctl(dev, ioctl, arg);
2234 static int kvm_device_release(struct inode *inode, struct file *filp)
2236 struct kvm_device *dev = filp->private_data;
2237 struct kvm *kvm = dev->kvm;
2243 static const struct file_operations kvm_device_fops = {
2244 .unlocked_ioctl = kvm_device_ioctl,
2245 #ifdef CONFIG_COMPAT
2246 .compat_ioctl = kvm_device_ioctl,
2248 .release = kvm_device_release,
2251 struct kvm_device *kvm_device_from_filp(struct file *filp)
2253 if (filp->f_op != &kvm_device_fops)
2256 return filp->private_data;
2259 static int kvm_ioctl_create_device(struct kvm *kvm,
2260 struct kvm_create_device *cd)
2262 struct kvm_device_ops *ops = NULL;
2263 struct kvm_device *dev;
2264 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2268 #ifdef CONFIG_KVM_MPIC
2269 case KVM_DEV_TYPE_FSL_MPIC_20:
2270 case KVM_DEV_TYPE_FSL_MPIC_42:
2271 ops = &kvm_mpic_ops;
2274 #ifdef CONFIG_KVM_XICS
2275 case KVM_DEV_TYPE_XICS:
2276 ops = &kvm_xics_ops;
2279 #ifdef CONFIG_KVM_VFIO
2280 case KVM_DEV_TYPE_VFIO:
2281 ops = &kvm_vfio_ops;
2284 #ifdef CONFIG_KVM_ARM_VGIC
2285 case KVM_DEV_TYPE_ARM_VGIC_V2:
2286 ops = &kvm_arm_vgic_v2_ops;
2290 case KVM_DEV_TYPE_FLIC:
2291 ops = &kvm_flic_ops;
2301 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2308 ret = ops->create(dev, cd->type);
2314 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2320 list_add(&dev->vm_node, &kvm->devices);
2326 static long kvm_vm_ioctl(struct file *filp,
2327 unsigned int ioctl, unsigned long arg)
2329 struct kvm *kvm = filp->private_data;
2330 void __user *argp = (void __user *)arg;
2333 if (kvm->mm != current->mm)
2336 case KVM_CREATE_VCPU:
2337 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2339 case KVM_SET_USER_MEMORY_REGION: {
2340 struct kvm_userspace_memory_region kvm_userspace_mem;
2343 if (copy_from_user(&kvm_userspace_mem, argp,
2344 sizeof kvm_userspace_mem))
2347 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2350 case KVM_GET_DIRTY_LOG: {
2351 struct kvm_dirty_log log;
2354 if (copy_from_user(&log, argp, sizeof log))
2356 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2359 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2360 case KVM_REGISTER_COALESCED_MMIO: {
2361 struct kvm_coalesced_mmio_zone zone;
2363 if (copy_from_user(&zone, argp, sizeof zone))
2365 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2368 case KVM_UNREGISTER_COALESCED_MMIO: {
2369 struct kvm_coalesced_mmio_zone zone;
2371 if (copy_from_user(&zone, argp, sizeof zone))
2373 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2378 struct kvm_irqfd data;
2381 if (copy_from_user(&data, argp, sizeof data))
2383 r = kvm_irqfd(kvm, &data);
2386 case KVM_IOEVENTFD: {
2387 struct kvm_ioeventfd data;
2390 if (copy_from_user(&data, argp, sizeof data))
2392 r = kvm_ioeventfd(kvm, &data);
2395 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2396 case KVM_SET_BOOT_CPU_ID:
2398 mutex_lock(&kvm->lock);
2399 if (atomic_read(&kvm->online_vcpus) != 0)
2402 kvm->bsp_vcpu_id = arg;
2403 mutex_unlock(&kvm->lock);
2406 #ifdef CONFIG_HAVE_KVM_MSI
2407 case KVM_SIGNAL_MSI: {
2411 if (copy_from_user(&msi, argp, sizeof msi))
2413 r = kvm_send_userspace_msi(kvm, &msi);
2417 #ifdef __KVM_HAVE_IRQ_LINE
2418 case KVM_IRQ_LINE_STATUS:
2419 case KVM_IRQ_LINE: {
2420 struct kvm_irq_level irq_event;
2423 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2426 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
2427 ioctl == KVM_IRQ_LINE_STATUS);
2432 if (ioctl == KVM_IRQ_LINE_STATUS) {
2433 if (copy_to_user(argp, &irq_event, sizeof irq_event))
2441 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2442 case KVM_SET_GSI_ROUTING: {
2443 struct kvm_irq_routing routing;
2444 struct kvm_irq_routing __user *urouting;
2445 struct kvm_irq_routing_entry *entries;
2448 if (copy_from_user(&routing, argp, sizeof(routing)))
2451 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2456 entries = vmalloc(routing.nr * sizeof(*entries));
2461 if (copy_from_user(entries, urouting->entries,
2462 routing.nr * sizeof(*entries)))
2463 goto out_free_irq_routing;
2464 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2466 out_free_irq_routing:
2470 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
2471 case KVM_CREATE_DEVICE: {
2472 struct kvm_create_device cd;
2475 if (copy_from_user(&cd, argp, sizeof(cd)))
2478 r = kvm_ioctl_create_device(kvm, &cd);
2483 if (copy_to_user(argp, &cd, sizeof(cd)))
2490 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2492 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2498 #ifdef CONFIG_COMPAT
2499 struct compat_kvm_dirty_log {
2503 compat_uptr_t dirty_bitmap; /* one bit per page */
2508 static long kvm_vm_compat_ioctl(struct file *filp,
2509 unsigned int ioctl, unsigned long arg)
2511 struct kvm *kvm = filp->private_data;
2514 if (kvm->mm != current->mm)
2517 case KVM_GET_DIRTY_LOG: {
2518 struct compat_kvm_dirty_log compat_log;
2519 struct kvm_dirty_log log;
2522 if (copy_from_user(&compat_log, (void __user *)arg,
2523 sizeof(compat_log)))
2525 log.slot = compat_log.slot;
2526 log.padding1 = compat_log.padding1;
2527 log.padding2 = compat_log.padding2;
2528 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2530 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2534 r = kvm_vm_ioctl(filp, ioctl, arg);
2542 static struct file_operations kvm_vm_fops = {
2543 .release = kvm_vm_release,
2544 .unlocked_ioctl = kvm_vm_ioctl,
2545 #ifdef CONFIG_COMPAT
2546 .compat_ioctl = kvm_vm_compat_ioctl,
2548 .llseek = noop_llseek,
2551 static int kvm_dev_ioctl_create_vm(unsigned long type)
2556 kvm = kvm_create_vm(type);
2558 return PTR_ERR(kvm);
2559 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2560 r = kvm_coalesced_mmio_init(kvm);
2566 r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC);
2573 static long kvm_dev_ioctl_check_extension_generic(long arg)
2576 case KVM_CAP_USER_MEMORY:
2577 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2578 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2579 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2580 case KVM_CAP_SET_BOOT_CPU_ID:
2582 case KVM_CAP_INTERNAL_ERROR_DATA:
2583 #ifdef CONFIG_HAVE_KVM_MSI
2584 case KVM_CAP_SIGNAL_MSI:
2586 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2587 case KVM_CAP_IRQFD_RESAMPLE:
2590 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2591 case KVM_CAP_IRQ_ROUTING:
2592 return KVM_MAX_IRQ_ROUTES;
2597 return kvm_dev_ioctl_check_extension(arg);
2600 static long kvm_dev_ioctl(struct file *filp,
2601 unsigned int ioctl, unsigned long arg)
2606 case KVM_GET_API_VERSION:
2610 r = KVM_API_VERSION;
2613 r = kvm_dev_ioctl_create_vm(arg);
2615 case KVM_CHECK_EXTENSION:
2616 r = kvm_dev_ioctl_check_extension_generic(arg);
2618 case KVM_GET_VCPU_MMAP_SIZE:
2622 r = PAGE_SIZE; /* struct kvm_run */
2624 r += PAGE_SIZE; /* pio data page */
2626 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2627 r += PAGE_SIZE; /* coalesced mmio ring page */
2630 case KVM_TRACE_ENABLE:
2631 case KVM_TRACE_PAUSE:
2632 case KVM_TRACE_DISABLE:
2636 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2642 static struct file_operations kvm_chardev_ops = {
2643 .unlocked_ioctl = kvm_dev_ioctl,
2644 .compat_ioctl = kvm_dev_ioctl,
2645 .llseek = noop_llseek,
2648 static struct miscdevice kvm_dev = {
2654 static void hardware_enable_nolock(void *junk)
2656 int cpu = raw_smp_processor_id();
2659 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2662 cpumask_set_cpu(cpu, cpus_hardware_enabled);
2664 r = kvm_arch_hardware_enable(NULL);
2667 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2668 atomic_inc(&hardware_enable_failed);
2669 printk(KERN_INFO "kvm: enabling virtualization on "
2670 "CPU%d failed\n", cpu);
2674 static void hardware_enable(void)
2676 raw_spin_lock(&kvm_count_lock);
2677 if (kvm_usage_count)
2678 hardware_enable_nolock(NULL);
2679 raw_spin_unlock(&kvm_count_lock);
2682 static void hardware_disable_nolock(void *junk)
2684 int cpu = raw_smp_processor_id();
2686 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2688 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2689 kvm_arch_hardware_disable(NULL);
2692 static void hardware_disable(void)
2694 raw_spin_lock(&kvm_count_lock);
2695 if (kvm_usage_count)
2696 hardware_disable_nolock(NULL);
2697 raw_spin_unlock(&kvm_count_lock);
2700 static void hardware_disable_all_nolock(void)
2702 BUG_ON(!kvm_usage_count);
2705 if (!kvm_usage_count)
2706 on_each_cpu(hardware_disable_nolock, NULL, 1);
2709 static void hardware_disable_all(void)
2711 raw_spin_lock(&kvm_count_lock);
2712 hardware_disable_all_nolock();
2713 raw_spin_unlock(&kvm_count_lock);
2716 static int hardware_enable_all(void)
2720 raw_spin_lock(&kvm_count_lock);
2723 if (kvm_usage_count == 1) {
2724 atomic_set(&hardware_enable_failed, 0);
2725 on_each_cpu(hardware_enable_nolock, NULL, 1);
2727 if (atomic_read(&hardware_enable_failed)) {
2728 hardware_disable_all_nolock();
2733 raw_spin_unlock(&kvm_count_lock);
2738 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2743 val &= ~CPU_TASKS_FROZEN;
2746 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2751 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2759 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2763 * Some (well, at least mine) BIOSes hang on reboot if
2766 * And Intel TXT required VMX off for all cpu when system shutdown.
2768 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2769 kvm_rebooting = true;
2770 on_each_cpu(hardware_disable_nolock, NULL, 1);
2774 static struct notifier_block kvm_reboot_notifier = {
2775 .notifier_call = kvm_reboot,
2779 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2783 for (i = 0; i < bus->dev_count; i++) {
2784 struct kvm_io_device *pos = bus->range[i].dev;
2786 kvm_iodevice_destructor(pos);
2791 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
2792 const struct kvm_io_range *r2)
2794 if (r1->addr < r2->addr)
2796 if (r1->addr + r1->len > r2->addr + r2->len)
2801 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2803 return kvm_io_bus_cmp(p1, p2);
2806 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2807 gpa_t addr, int len)
2809 bus->range[bus->dev_count++] = (struct kvm_io_range) {
2815 sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2816 kvm_io_bus_sort_cmp, NULL);
2821 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2822 gpa_t addr, int len)
2824 struct kvm_io_range *range, key;
2827 key = (struct kvm_io_range) {
2832 range = bsearch(&key, bus->range, bus->dev_count,
2833 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2837 off = range - bus->range;
2839 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
2845 static int __kvm_io_bus_write(struct kvm_io_bus *bus,
2846 struct kvm_io_range *range, const void *val)
2850 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2854 while (idx < bus->dev_count &&
2855 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
2856 if (!kvm_iodevice_write(bus->range[idx].dev, range->addr,
2865 /* kvm_io_bus_write - called under kvm->slots_lock */
2866 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2867 int len, const void *val)
2869 struct kvm_io_bus *bus;
2870 struct kvm_io_range range;
2873 range = (struct kvm_io_range) {
2878 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2879 r = __kvm_io_bus_write(bus, &range, val);
2880 return r < 0 ? r : 0;
2883 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
2884 int kvm_io_bus_write_cookie(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2885 int len, const void *val, long cookie)
2887 struct kvm_io_bus *bus;
2888 struct kvm_io_range range;
2890 range = (struct kvm_io_range) {
2895 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2897 /* First try the device referenced by cookie. */
2898 if ((cookie >= 0) && (cookie < bus->dev_count) &&
2899 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
2900 if (!kvm_iodevice_write(bus->range[cookie].dev, addr, len,
2905 * cookie contained garbage; fall back to search and return the
2906 * correct cookie value.
2908 return __kvm_io_bus_write(bus, &range, val);
2911 static int __kvm_io_bus_read(struct kvm_io_bus *bus, struct kvm_io_range *range,
2916 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2920 while (idx < bus->dev_count &&
2921 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
2922 if (!kvm_iodevice_read(bus->range[idx].dev, range->addr,
2930 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
2932 /* kvm_io_bus_read - called under kvm->slots_lock */
2933 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2936 struct kvm_io_bus *bus;
2937 struct kvm_io_range range;
2940 range = (struct kvm_io_range) {
2945 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2946 r = __kvm_io_bus_read(bus, &range, val);
2947 return r < 0 ? r : 0;
2951 /* Caller must hold slots_lock. */
2952 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2953 int len, struct kvm_io_device *dev)
2955 struct kvm_io_bus *new_bus, *bus;
2957 bus = kvm->buses[bus_idx];
2958 /* exclude ioeventfd which is limited by maximum fd */
2959 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
2962 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
2963 sizeof(struct kvm_io_range)), GFP_KERNEL);
2966 memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
2967 sizeof(struct kvm_io_range)));
2968 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2969 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2970 synchronize_srcu_expedited(&kvm->srcu);
2976 /* Caller must hold slots_lock. */
2977 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2978 struct kvm_io_device *dev)
2981 struct kvm_io_bus *new_bus, *bus;
2983 bus = kvm->buses[bus_idx];
2985 for (i = 0; i < bus->dev_count; i++)
2986 if (bus->range[i].dev == dev) {
2994 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
2995 sizeof(struct kvm_io_range)), GFP_KERNEL);
2999 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3000 new_bus->dev_count--;
3001 memcpy(new_bus->range + i, bus->range + i + 1,
3002 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3004 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3005 synchronize_srcu_expedited(&kvm->srcu);
3010 static struct notifier_block kvm_cpu_notifier = {
3011 .notifier_call = kvm_cpu_hotplug,
3014 static int vm_stat_get(void *_offset, u64 *val)
3016 unsigned offset = (long)_offset;
3020 spin_lock(&kvm_lock);
3021 list_for_each_entry(kvm, &vm_list, vm_list)
3022 *val += *(u32 *)((void *)kvm + offset);
3023 spin_unlock(&kvm_lock);
3027 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
3029 static int vcpu_stat_get(void *_offset, u64 *val)
3031 unsigned offset = (long)_offset;
3033 struct kvm_vcpu *vcpu;
3037 spin_lock(&kvm_lock);
3038 list_for_each_entry(kvm, &vm_list, vm_list)
3039 kvm_for_each_vcpu(i, vcpu, kvm)
3040 *val += *(u32 *)((void *)vcpu + offset);
3042 spin_unlock(&kvm_lock);
3046 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
3048 static const struct file_operations *stat_fops[] = {
3049 [KVM_STAT_VCPU] = &vcpu_stat_fops,
3050 [KVM_STAT_VM] = &vm_stat_fops,
3053 static int kvm_init_debug(void)
3056 struct kvm_stats_debugfs_item *p;
3058 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3059 if (kvm_debugfs_dir == NULL)
3062 for (p = debugfs_entries; p->name; ++p) {
3063 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
3064 (void *)(long)p->offset,
3065 stat_fops[p->kind]);
3066 if (p->dentry == NULL)
3073 debugfs_remove_recursive(kvm_debugfs_dir);
3078 static void kvm_exit_debug(void)
3080 struct kvm_stats_debugfs_item *p;
3082 for (p = debugfs_entries; p->name; ++p)
3083 debugfs_remove(p->dentry);
3084 debugfs_remove(kvm_debugfs_dir);
3087 static int kvm_suspend(void)
3089 if (kvm_usage_count)
3090 hardware_disable_nolock(NULL);
3094 static void kvm_resume(void)
3096 if (kvm_usage_count) {
3097 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3098 hardware_enable_nolock(NULL);
3102 static struct syscore_ops kvm_syscore_ops = {
3103 .suspend = kvm_suspend,
3104 .resume = kvm_resume,
3108 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3110 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3113 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3115 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3116 if (vcpu->preempted)
3117 vcpu->preempted = false;
3119 kvm_arch_vcpu_load(vcpu, cpu);
3122 static void kvm_sched_out(struct preempt_notifier *pn,
3123 struct task_struct *next)
3125 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3127 if (current->state == TASK_RUNNING)
3128 vcpu->preempted = true;
3129 kvm_arch_vcpu_put(vcpu);
3132 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3133 struct module *module)
3138 r = kvm_arch_init(opaque);
3143 * kvm_arch_init makes sure there's at most one caller
3144 * for architectures that support multiple implementations,
3145 * like intel and amd on x86.
3146 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3147 * conflicts in case kvm is already setup for another implementation.
3149 r = kvm_irqfd_init();
3153 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
3158 r = kvm_arch_hardware_setup();
3162 for_each_online_cpu(cpu) {
3163 smp_call_function_single(cpu,
3164 kvm_arch_check_processor_compat,
3170 r = register_cpu_notifier(&kvm_cpu_notifier);
3173 register_reboot_notifier(&kvm_reboot_notifier);
3175 /* A kmem cache lets us meet the alignment requirements of fx_save. */
3177 vcpu_align = __alignof__(struct kvm_vcpu);
3178 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
3180 if (!kvm_vcpu_cache) {
3185 r = kvm_async_pf_init();
3189 kvm_chardev_ops.owner = module;
3190 kvm_vm_fops.owner = module;
3191 kvm_vcpu_fops.owner = module;
3193 r = misc_register(&kvm_dev);
3195 printk(KERN_ERR "kvm: misc device register failed\n");
3199 register_syscore_ops(&kvm_syscore_ops);
3201 kvm_preempt_ops.sched_in = kvm_sched_in;
3202 kvm_preempt_ops.sched_out = kvm_sched_out;
3204 r = kvm_init_debug();
3206 printk(KERN_ERR "kvm: create debugfs files failed\n");
3213 unregister_syscore_ops(&kvm_syscore_ops);
3214 misc_deregister(&kvm_dev);
3216 kvm_async_pf_deinit();
3218 kmem_cache_destroy(kvm_vcpu_cache);
3220 unregister_reboot_notifier(&kvm_reboot_notifier);
3221 unregister_cpu_notifier(&kvm_cpu_notifier);
3224 kvm_arch_hardware_unsetup();
3226 free_cpumask_var(cpus_hardware_enabled);
3234 EXPORT_SYMBOL_GPL(kvm_init);
3239 misc_deregister(&kvm_dev);
3240 kmem_cache_destroy(kvm_vcpu_cache);
3241 kvm_async_pf_deinit();
3242 unregister_syscore_ops(&kvm_syscore_ops);
3243 unregister_reboot_notifier(&kvm_reboot_notifier);
3244 unregister_cpu_notifier(&kvm_cpu_notifier);
3245 on_each_cpu(hardware_disable_nolock, NULL, 1);
3246 kvm_arch_hardware_unsetup();
3249 free_cpumask_var(cpus_hardware_enabled);
3251 EXPORT_SYMBOL_GPL(kvm_exit);