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);
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;
479 kvm_init_memslots_id(kvm);
480 if (init_srcu_struct(&kvm->srcu))
481 goto out_err_no_srcu;
482 if (init_srcu_struct(&kvm->irq_srcu))
483 goto out_err_no_irq_srcu;
484 for (i = 0; i < KVM_NR_BUSES; i++) {
485 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
491 spin_lock_init(&kvm->mmu_lock);
492 kvm->mm = current->mm;
493 atomic_inc(&kvm->mm->mm_count);
494 kvm_eventfd_init(kvm);
495 mutex_init(&kvm->lock);
496 mutex_init(&kvm->irq_lock);
497 mutex_init(&kvm->slots_lock);
498 atomic_set(&kvm->users_count, 1);
499 INIT_LIST_HEAD(&kvm->devices);
501 r = kvm_init_mmu_notifier(kvm);
505 spin_lock(&kvm_lock);
506 list_add(&kvm->vm_list, &vm_list);
507 spin_unlock(&kvm_lock);
512 cleanup_srcu_struct(&kvm->irq_srcu);
514 cleanup_srcu_struct(&kvm->srcu);
516 hardware_disable_all();
518 for (i = 0; i < KVM_NR_BUSES; i++)
519 kfree(kvm->buses[i]);
520 kfree(kvm->memslots);
521 kvm_arch_free_vm(kvm);
526 * Avoid using vmalloc for a small buffer.
527 * Should not be used when the size is statically known.
529 void *kvm_kvzalloc(unsigned long size)
531 if (size > PAGE_SIZE)
532 return vzalloc(size);
534 return kzalloc(size, GFP_KERNEL);
537 void kvm_kvfree(const void *addr)
539 if (is_vmalloc_addr(addr))
545 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
547 if (!memslot->dirty_bitmap)
550 kvm_kvfree(memslot->dirty_bitmap);
551 memslot->dirty_bitmap = NULL;
555 * Free any memory in @free but not in @dont.
557 static void kvm_free_physmem_slot(struct kvm *kvm, struct kvm_memory_slot *free,
558 struct kvm_memory_slot *dont)
560 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
561 kvm_destroy_dirty_bitmap(free);
563 kvm_arch_free_memslot(kvm, free, dont);
568 static void kvm_free_physmem(struct kvm *kvm)
570 struct kvm_memslots *slots = kvm->memslots;
571 struct kvm_memory_slot *memslot;
573 kvm_for_each_memslot(memslot, slots)
574 kvm_free_physmem_slot(kvm, memslot, NULL);
576 kfree(kvm->memslots);
579 static void kvm_destroy_devices(struct kvm *kvm)
581 struct list_head *node, *tmp;
583 list_for_each_safe(node, tmp, &kvm->devices) {
584 struct kvm_device *dev =
585 list_entry(node, struct kvm_device, vm_node);
588 dev->ops->destroy(dev);
592 static void kvm_destroy_vm(struct kvm *kvm)
595 struct mm_struct *mm = kvm->mm;
597 kvm_arch_sync_events(kvm);
598 spin_lock(&kvm_lock);
599 list_del(&kvm->vm_list);
600 spin_unlock(&kvm_lock);
601 kvm_free_irq_routing(kvm);
602 for (i = 0; i < KVM_NR_BUSES; i++)
603 kvm_io_bus_destroy(kvm->buses[i]);
604 kvm_coalesced_mmio_free(kvm);
605 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
606 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
608 kvm_arch_flush_shadow_all(kvm);
610 kvm_arch_destroy_vm(kvm);
611 kvm_destroy_devices(kvm);
612 kvm_free_physmem(kvm);
613 cleanup_srcu_struct(&kvm->irq_srcu);
614 cleanup_srcu_struct(&kvm->srcu);
615 kvm_arch_free_vm(kvm);
616 hardware_disable_all();
620 void kvm_get_kvm(struct kvm *kvm)
622 atomic_inc(&kvm->users_count);
624 EXPORT_SYMBOL_GPL(kvm_get_kvm);
626 void kvm_put_kvm(struct kvm *kvm)
628 if (atomic_dec_and_test(&kvm->users_count))
631 EXPORT_SYMBOL_GPL(kvm_put_kvm);
634 static int kvm_vm_release(struct inode *inode, struct file *filp)
636 struct kvm *kvm = filp->private_data;
638 kvm_irqfd_release(kvm);
645 * Allocation size is twice as large as the actual dirty bitmap size.
646 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
648 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
650 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
652 memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
653 if (!memslot->dirty_bitmap)
659 static int cmp_memslot(const void *slot1, const void *slot2)
661 struct kvm_memory_slot *s1, *s2;
663 s1 = (struct kvm_memory_slot *)slot1;
664 s2 = (struct kvm_memory_slot *)slot2;
666 if (s1->npages < s2->npages)
668 if (s1->npages > s2->npages)
675 * Sort the memslots base on its size, so the larger slots
676 * will get better fit.
678 static void sort_memslots(struct kvm_memslots *slots)
682 sort(slots->memslots, KVM_MEM_SLOTS_NUM,
683 sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
685 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
686 slots->id_to_index[slots->memslots[i].id] = i;
689 static void update_memslots(struct kvm_memslots *slots,
690 struct kvm_memory_slot *new,
695 struct kvm_memory_slot *old = id_to_memslot(slots, id);
696 unsigned long npages = old->npages;
699 if (new->npages != npages)
700 sort_memslots(slots);
703 slots->generation = last_generation + 1;
706 static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
708 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
710 #ifdef KVM_CAP_READONLY_MEM
711 valid_flags |= KVM_MEM_READONLY;
714 if (mem->flags & ~valid_flags)
720 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
721 struct kvm_memslots *slots, struct kvm_memory_slot *new)
723 struct kvm_memslots *old_memslots = kvm->memslots;
725 update_memslots(slots, new, kvm->memslots->generation);
726 rcu_assign_pointer(kvm->memslots, slots);
727 synchronize_srcu_expedited(&kvm->srcu);
729 kvm_arch_memslots_updated(kvm);
735 * Allocate some memory and give it an address in the guest physical address
738 * Discontiguous memory is allowed, mostly for framebuffers.
740 * Must be called holding mmap_sem for write.
742 int __kvm_set_memory_region(struct kvm *kvm,
743 struct kvm_userspace_memory_region *mem)
747 unsigned long npages;
748 struct kvm_memory_slot *slot;
749 struct kvm_memory_slot old, new;
750 struct kvm_memslots *slots = NULL, *old_memslots;
751 enum kvm_mr_change change;
753 r = check_memory_region_flags(mem);
758 /* General sanity checks */
759 if (mem->memory_size & (PAGE_SIZE - 1))
761 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
763 /* We can read the guest memory with __xxx_user() later on. */
764 if ((mem->slot < KVM_USER_MEM_SLOTS) &&
765 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
766 !access_ok(VERIFY_WRITE,
767 (void __user *)(unsigned long)mem->userspace_addr,
770 if (mem->slot >= KVM_MEM_SLOTS_NUM)
772 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
775 slot = id_to_memslot(kvm->memslots, mem->slot);
776 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
777 npages = mem->memory_size >> PAGE_SHIFT;
780 if (npages > KVM_MEM_MAX_NR_PAGES)
784 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
789 new.base_gfn = base_gfn;
791 new.flags = mem->flags;
796 change = KVM_MR_CREATE;
797 else { /* Modify an existing slot. */
798 if ((mem->userspace_addr != old.userspace_addr) ||
799 (npages != old.npages) ||
800 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
803 if (base_gfn != old.base_gfn)
804 change = KVM_MR_MOVE;
805 else if (new.flags != old.flags)
806 change = KVM_MR_FLAGS_ONLY;
807 else { /* Nothing to change. */
812 } else if (old.npages) {
813 change = KVM_MR_DELETE;
814 } else /* Modify a non-existent slot: disallowed. */
817 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
818 /* Check for overlaps */
820 kvm_for_each_memslot(slot, kvm->memslots) {
821 if ((slot->id >= KVM_USER_MEM_SLOTS) ||
822 (slot->id == mem->slot))
824 if (!((base_gfn + npages <= slot->base_gfn) ||
825 (base_gfn >= slot->base_gfn + slot->npages)))
830 /* Free page dirty bitmap if unneeded */
831 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
832 new.dirty_bitmap = NULL;
835 if (change == KVM_MR_CREATE) {
836 new.userspace_addr = mem->userspace_addr;
838 if (kvm_arch_create_memslot(kvm, &new, npages))
842 /* Allocate page dirty bitmap if needed */
843 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
844 if (kvm_create_dirty_bitmap(&new) < 0)
848 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
850 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
854 slot = id_to_memslot(slots, mem->slot);
855 slot->flags |= KVM_MEMSLOT_INVALID;
857 old_memslots = install_new_memslots(kvm, slots, NULL);
859 /* slot was deleted or moved, clear iommu mapping */
860 kvm_iommu_unmap_pages(kvm, &old);
861 /* From this point no new shadow pages pointing to a deleted,
862 * or moved, memslot will be created.
864 * validation of sp->gfn happens in:
865 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
866 * - kvm_is_visible_gfn (mmu_check_roots)
868 kvm_arch_flush_shadow_memslot(kvm, slot);
869 slots = old_memslots;
872 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
878 * We can re-use the old_memslots from above, the only difference
879 * from the currently installed memslots is the invalid flag. This
880 * will get overwritten by update_memslots anyway.
883 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
889 /* actual memory is freed via old in kvm_free_physmem_slot below */
890 if (change == KVM_MR_DELETE) {
891 new.dirty_bitmap = NULL;
892 memset(&new.arch, 0, sizeof(new.arch));
895 old_memslots = install_new_memslots(kvm, slots, &new);
897 kvm_arch_commit_memory_region(kvm, mem, &old, change);
899 kvm_free_physmem_slot(kvm, &old, &new);
903 * IOMMU mapping: New slots need to be mapped. Old slots need to be
904 * un-mapped and re-mapped if their base changes. Since base change
905 * unmapping is handled above with slot deletion, mapping alone is
906 * needed here. Anything else the iommu might care about for existing
907 * slots (size changes, userspace addr changes and read-only flag
908 * changes) is disallowed above, so any other attribute changes getting
909 * here can be skipped.
911 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
912 r = kvm_iommu_map_pages(kvm, &new);
921 kvm_free_physmem_slot(kvm, &new, &old);
925 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
927 int kvm_set_memory_region(struct kvm *kvm,
928 struct kvm_userspace_memory_region *mem)
932 mutex_lock(&kvm->slots_lock);
933 r = __kvm_set_memory_region(kvm, mem);
934 mutex_unlock(&kvm->slots_lock);
937 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
939 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
940 struct kvm_userspace_memory_region *mem)
942 if (mem->slot >= KVM_USER_MEM_SLOTS)
944 return kvm_set_memory_region(kvm, mem);
947 int kvm_get_dirty_log(struct kvm *kvm,
948 struct kvm_dirty_log *log, int *is_dirty)
950 struct kvm_memory_slot *memslot;
953 unsigned long any = 0;
956 if (log->slot >= KVM_USER_MEM_SLOTS)
959 memslot = id_to_memslot(kvm->memslots, log->slot);
961 if (!memslot->dirty_bitmap)
964 n = kvm_dirty_bitmap_bytes(memslot);
966 for (i = 0; !any && i < n/sizeof(long); ++i)
967 any = memslot->dirty_bitmap[i];
970 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
980 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
982 bool kvm_largepages_enabled(void)
984 return largepages_enabled;
987 void kvm_disable_largepages(void)
989 largepages_enabled = false;
991 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
993 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
995 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
997 EXPORT_SYMBOL_GPL(gfn_to_memslot);
999 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1001 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1003 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1004 memslot->flags & KVM_MEMSLOT_INVALID)
1009 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1011 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1013 struct vm_area_struct *vma;
1014 unsigned long addr, size;
1018 addr = gfn_to_hva(kvm, gfn);
1019 if (kvm_is_error_hva(addr))
1022 down_read(¤t->mm->mmap_sem);
1023 vma = find_vma(current->mm, addr);
1027 size = vma_kernel_pagesize(vma);
1030 up_read(¤t->mm->mmap_sem);
1035 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1037 return slot->flags & KVM_MEM_READONLY;
1040 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1041 gfn_t *nr_pages, bool write)
1043 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1044 return KVM_HVA_ERR_BAD;
1046 if (memslot_is_readonly(slot) && write)
1047 return KVM_HVA_ERR_RO_BAD;
1050 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1052 return __gfn_to_hva_memslot(slot, gfn);
1055 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1058 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1061 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1064 return gfn_to_hva_many(slot, gfn, NULL);
1066 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1068 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1070 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1072 EXPORT_SYMBOL_GPL(gfn_to_hva);
1075 * If writable is set to false, the hva returned by this function is only
1076 * allowed to be read.
1078 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1080 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1081 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1083 if (!kvm_is_error_hva(hva) && writable)
1084 *writable = !memslot_is_readonly(slot);
1089 static int kvm_read_hva(void *data, void __user *hva, int len)
1091 return __copy_from_user(data, hva, len);
1094 static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1096 return __copy_from_user_inatomic(data, hva, len);
1099 static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1100 unsigned long start, int write, struct page **page)
1102 int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1105 flags |= FOLL_WRITE;
1107 return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1110 static inline int check_user_page_hwpoison(unsigned long addr)
1112 int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1114 rc = __get_user_pages(current, current->mm, addr, 1,
1115 flags, NULL, NULL, NULL);
1116 return rc == -EHWPOISON;
1120 * The atomic path to get the writable pfn which will be stored in @pfn,
1121 * true indicates success, otherwise false is returned.
1123 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1124 bool write_fault, bool *writable, pfn_t *pfn)
1126 struct page *page[1];
1129 if (!(async || atomic))
1133 * Fast pin a writable pfn only if it is a write fault request
1134 * or the caller allows to map a writable pfn for a read fault
1137 if (!(write_fault || writable))
1140 npages = __get_user_pages_fast(addr, 1, 1, page);
1142 *pfn = page_to_pfn(page[0]);
1153 * The slow path to get the pfn of the specified host virtual address,
1154 * 1 indicates success, -errno is returned if error is detected.
1156 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1157 bool *writable, pfn_t *pfn)
1159 struct page *page[1];
1165 *writable = write_fault;
1168 down_read(¤t->mm->mmap_sem);
1169 npages = get_user_page_nowait(current, current->mm,
1170 addr, write_fault, page);
1171 up_read(¤t->mm->mmap_sem);
1173 npages = get_user_pages_fast(addr, 1, write_fault,
1178 /* map read fault as writable if possible */
1179 if (unlikely(!write_fault) && writable) {
1180 struct page *wpage[1];
1182 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1191 *pfn = page_to_pfn(page[0]);
1195 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1197 if (unlikely(!(vma->vm_flags & VM_READ)))
1200 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1207 * Pin guest page in memory and return its pfn.
1208 * @addr: host virtual address which maps memory to the guest
1209 * @atomic: whether this function can sleep
1210 * @async: whether this function need to wait IO complete if the
1211 * host page is not in the memory
1212 * @write_fault: whether we should get a writable host page
1213 * @writable: whether it allows to map a writable host page for !@write_fault
1215 * The function will map a writable host page for these two cases:
1216 * 1): @write_fault = true
1217 * 2): @write_fault = false && @writable, @writable will tell the caller
1218 * whether the mapping is writable.
1220 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1221 bool write_fault, bool *writable)
1223 struct vm_area_struct *vma;
1227 /* we can do it either atomically or asynchronously, not both */
1228 BUG_ON(atomic && async);
1230 if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1234 return KVM_PFN_ERR_FAULT;
1236 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1240 down_read(¤t->mm->mmap_sem);
1241 if (npages == -EHWPOISON ||
1242 (!async && check_user_page_hwpoison(addr))) {
1243 pfn = KVM_PFN_ERR_HWPOISON;
1247 vma = find_vma_intersection(current->mm, addr, addr + 1);
1250 pfn = KVM_PFN_ERR_FAULT;
1251 else if ((vma->vm_flags & VM_PFNMAP)) {
1252 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1254 BUG_ON(!kvm_is_mmio_pfn(pfn));
1256 if (async && vma_is_valid(vma, write_fault))
1258 pfn = KVM_PFN_ERR_FAULT;
1261 up_read(¤t->mm->mmap_sem);
1266 __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1267 bool *async, bool write_fault, bool *writable)
1269 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1271 if (addr == KVM_HVA_ERR_RO_BAD)
1272 return KVM_PFN_ERR_RO_FAULT;
1274 if (kvm_is_error_hva(addr))
1275 return KVM_PFN_NOSLOT;
1277 /* Do not map writable pfn in the readonly memslot. */
1278 if (writable && memslot_is_readonly(slot)) {
1283 return hva_to_pfn(addr, atomic, async, write_fault,
1287 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1288 bool write_fault, bool *writable)
1290 struct kvm_memory_slot *slot;
1295 slot = gfn_to_memslot(kvm, gfn);
1297 return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1301 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1303 return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1305 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1307 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1308 bool write_fault, bool *writable)
1310 return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1312 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1314 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1316 return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1318 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1320 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1323 return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1325 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1327 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1329 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1332 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1334 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1336 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1338 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1344 addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1345 if (kvm_is_error_hva(addr))
1348 if (entry < nr_pages)
1351 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1353 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1355 static struct page *kvm_pfn_to_page(pfn_t pfn)
1357 if (is_error_noslot_pfn(pfn))
1358 return KVM_ERR_PTR_BAD_PAGE;
1360 if (kvm_is_mmio_pfn(pfn)) {
1362 return KVM_ERR_PTR_BAD_PAGE;
1365 return pfn_to_page(pfn);
1368 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1372 pfn = gfn_to_pfn(kvm, gfn);
1374 return kvm_pfn_to_page(pfn);
1377 EXPORT_SYMBOL_GPL(gfn_to_page);
1379 void kvm_release_page_clean(struct page *page)
1381 WARN_ON(is_error_page(page));
1383 kvm_release_pfn_clean(page_to_pfn(page));
1385 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1387 void kvm_release_pfn_clean(pfn_t pfn)
1389 if (!is_error_noslot_pfn(pfn) && !kvm_is_mmio_pfn(pfn))
1390 put_page(pfn_to_page(pfn));
1392 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1394 void kvm_release_page_dirty(struct page *page)
1396 WARN_ON(is_error_page(page));
1398 kvm_release_pfn_dirty(page_to_pfn(page));
1400 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1402 static void kvm_release_pfn_dirty(pfn_t pfn)
1404 kvm_set_pfn_dirty(pfn);
1405 kvm_release_pfn_clean(pfn);
1408 void kvm_set_pfn_dirty(pfn_t pfn)
1410 if (!kvm_is_mmio_pfn(pfn)) {
1411 struct page *page = pfn_to_page(pfn);
1412 if (!PageReserved(page))
1416 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1418 void kvm_set_pfn_accessed(pfn_t pfn)
1420 if (!kvm_is_mmio_pfn(pfn))
1421 mark_page_accessed(pfn_to_page(pfn));
1423 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1425 void kvm_get_pfn(pfn_t pfn)
1427 if (!kvm_is_mmio_pfn(pfn))
1428 get_page(pfn_to_page(pfn));
1430 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1432 static int next_segment(unsigned long len, int offset)
1434 if (len > PAGE_SIZE - offset)
1435 return PAGE_SIZE - offset;
1440 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1446 addr = gfn_to_hva_prot(kvm, gfn, NULL);
1447 if (kvm_is_error_hva(addr))
1449 r = kvm_read_hva(data, (void __user *)addr + offset, len);
1454 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1456 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1458 gfn_t gfn = gpa >> PAGE_SHIFT;
1460 int offset = offset_in_page(gpa);
1463 while ((seg = next_segment(len, offset)) != 0) {
1464 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1474 EXPORT_SYMBOL_GPL(kvm_read_guest);
1476 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1481 gfn_t gfn = gpa >> PAGE_SHIFT;
1482 int offset = offset_in_page(gpa);
1484 addr = gfn_to_hva_prot(kvm, gfn, NULL);
1485 if (kvm_is_error_hva(addr))
1487 pagefault_disable();
1488 r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
1494 EXPORT_SYMBOL(kvm_read_guest_atomic);
1496 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1497 int offset, int len)
1502 addr = gfn_to_hva(kvm, gfn);
1503 if (kvm_is_error_hva(addr))
1505 r = __copy_to_user((void __user *)addr + offset, data, len);
1508 mark_page_dirty(kvm, gfn);
1511 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1513 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
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_write_guest_page(kvm, gfn, data, offset, seg);
1533 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1534 gpa_t gpa, unsigned long len)
1536 struct kvm_memslots *slots = kvm_memslots(kvm);
1537 int offset = offset_in_page(gpa);
1538 gfn_t start_gfn = gpa >> PAGE_SHIFT;
1539 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1540 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1541 gfn_t nr_pages_avail;
1544 ghc->generation = slots->generation;
1546 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1547 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, &nr_pages_avail);
1548 if (!kvm_is_error_hva(ghc->hva) && nr_pages_avail >= nr_pages_needed) {
1552 * If the requested region crosses two memslots, we still
1553 * verify that the entire region is valid here.
1555 while (start_gfn <= end_gfn) {
1556 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1557 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1559 if (kvm_is_error_hva(ghc->hva))
1561 start_gfn += nr_pages_avail;
1563 /* Use the slow path for cross page reads and writes. */
1564 ghc->memslot = NULL;
1568 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1570 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1571 void *data, unsigned long len)
1573 struct kvm_memslots *slots = kvm_memslots(kvm);
1576 BUG_ON(len > ghc->len);
1578 if (slots->generation != ghc->generation)
1579 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1581 if (unlikely(!ghc->memslot))
1582 return kvm_write_guest(kvm, ghc->gpa, data, len);
1584 if (kvm_is_error_hva(ghc->hva))
1587 r = __copy_to_user((void __user *)ghc->hva, data, len);
1590 mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1594 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1596 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1597 void *data, unsigned long len)
1599 struct kvm_memslots *slots = kvm_memslots(kvm);
1602 BUG_ON(len > ghc->len);
1604 if (slots->generation != ghc->generation)
1605 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1607 if (unlikely(!ghc->memslot))
1608 return kvm_read_guest(kvm, ghc->gpa, data, len);
1610 if (kvm_is_error_hva(ghc->hva))
1613 r = __copy_from_user(data, (void __user *)ghc->hva, len);
1619 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1621 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1623 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
1625 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
1627 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1629 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1631 gfn_t gfn = gpa >> PAGE_SHIFT;
1633 int offset = offset_in_page(gpa);
1636 while ((seg = next_segment(len, offset)) != 0) {
1637 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1646 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1648 static void mark_page_dirty_in_slot(struct kvm *kvm,
1649 struct kvm_memory_slot *memslot,
1652 if (memslot && memslot->dirty_bitmap) {
1653 unsigned long rel_gfn = gfn - memslot->base_gfn;
1655 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1659 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1661 struct kvm_memory_slot *memslot;
1663 memslot = gfn_to_memslot(kvm, gfn);
1664 mark_page_dirty_in_slot(kvm, memslot, gfn);
1666 EXPORT_SYMBOL_GPL(mark_page_dirty);
1669 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1671 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1676 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1678 if (kvm_arch_vcpu_runnable(vcpu)) {
1679 kvm_make_request(KVM_REQ_UNHALT, vcpu);
1682 if (kvm_cpu_has_pending_timer(vcpu))
1684 if (signal_pending(current))
1690 finish_wait(&vcpu->wq, &wait);
1692 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
1696 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1698 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1701 int cpu = vcpu->cpu;
1702 wait_queue_head_t *wqp;
1704 wqp = kvm_arch_vcpu_wq(vcpu);
1705 if (waitqueue_active(wqp)) {
1706 wake_up_interruptible(wqp);
1707 ++vcpu->stat.halt_wakeup;
1711 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1712 if (kvm_arch_vcpu_should_kick(vcpu))
1713 smp_send_reschedule(cpu);
1716 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
1717 #endif /* !CONFIG_S390 */
1719 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
1722 struct task_struct *task = NULL;
1726 pid = rcu_dereference(target->pid);
1728 task = get_pid_task(target->pid, PIDTYPE_PID);
1732 if (task->flags & PF_VCPU) {
1733 put_task_struct(task);
1736 ret = yield_to(task, 1);
1737 put_task_struct(task);
1741 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1744 * Helper that checks whether a VCPU is eligible for directed yield.
1745 * Most eligible candidate to yield is decided by following heuristics:
1747 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1748 * (preempted lock holder), indicated by @in_spin_loop.
1749 * Set at the beiginning and cleared at the end of interception/PLE handler.
1751 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1752 * chance last time (mostly it has become eligible now since we have probably
1753 * yielded to lockholder in last iteration. This is done by toggling
1754 * @dy_eligible each time a VCPU checked for eligibility.)
1756 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1757 * to preempted lock-holder could result in wrong VCPU selection and CPU
1758 * burning. Giving priority for a potential lock-holder increases lock
1761 * Since algorithm is based on heuristics, accessing another VCPU data without
1762 * locking does not harm. It may result in trying to yield to same VCPU, fail
1763 * and continue with next VCPU and so on.
1765 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1767 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1770 eligible = !vcpu->spin_loop.in_spin_loop ||
1771 (vcpu->spin_loop.in_spin_loop &&
1772 vcpu->spin_loop.dy_eligible);
1774 if (vcpu->spin_loop.in_spin_loop)
1775 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1783 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1785 struct kvm *kvm = me->kvm;
1786 struct kvm_vcpu *vcpu;
1787 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1793 kvm_vcpu_set_in_spin_loop(me, true);
1795 * We boost the priority of a VCPU that is runnable but not
1796 * currently running, because it got preempted by something
1797 * else and called schedule in __vcpu_run. Hopefully that
1798 * VCPU is holding the lock that we need and will release it.
1799 * We approximate round-robin by starting at the last boosted VCPU.
1801 for (pass = 0; pass < 2 && !yielded && try; pass++) {
1802 kvm_for_each_vcpu(i, vcpu, kvm) {
1803 if (!pass && i <= last_boosted_vcpu) {
1804 i = last_boosted_vcpu;
1806 } else if (pass && i > last_boosted_vcpu)
1808 if (!ACCESS_ONCE(vcpu->preempted))
1812 if (waitqueue_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
1814 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1817 yielded = kvm_vcpu_yield_to(vcpu);
1819 kvm->last_boosted_vcpu = i;
1821 } else if (yielded < 0) {
1828 kvm_vcpu_set_in_spin_loop(me, false);
1830 /* Ensure vcpu is not eligible during next spinloop */
1831 kvm_vcpu_set_dy_eligible(me, false);
1833 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1835 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1837 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1840 if (vmf->pgoff == 0)
1841 page = virt_to_page(vcpu->run);
1843 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1844 page = virt_to_page(vcpu->arch.pio_data);
1846 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1847 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1848 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1851 return kvm_arch_vcpu_fault(vcpu, vmf);
1857 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1858 .fault = kvm_vcpu_fault,
1861 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1863 vma->vm_ops = &kvm_vcpu_vm_ops;
1867 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1869 struct kvm_vcpu *vcpu = filp->private_data;
1871 kvm_put_kvm(vcpu->kvm);
1875 static struct file_operations kvm_vcpu_fops = {
1876 .release = kvm_vcpu_release,
1877 .unlocked_ioctl = kvm_vcpu_ioctl,
1878 #ifdef CONFIG_COMPAT
1879 .compat_ioctl = kvm_vcpu_compat_ioctl,
1881 .mmap = kvm_vcpu_mmap,
1882 .llseek = noop_llseek,
1886 * Allocates an inode for the vcpu.
1888 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1890 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
1894 * Creates some virtual cpus. Good luck creating more than one.
1896 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1899 struct kvm_vcpu *vcpu, *v;
1901 if (id >= KVM_MAX_VCPUS)
1904 vcpu = kvm_arch_vcpu_create(kvm, id);
1906 return PTR_ERR(vcpu);
1908 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1910 r = kvm_arch_vcpu_setup(vcpu);
1914 mutex_lock(&kvm->lock);
1915 if (!kvm_vcpu_compatible(vcpu)) {
1917 goto unlock_vcpu_destroy;
1919 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1921 goto unlock_vcpu_destroy;
1924 kvm_for_each_vcpu(r, v, kvm)
1925 if (v->vcpu_id == id) {
1927 goto unlock_vcpu_destroy;
1930 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1932 /* Now it's all set up, let userspace reach it */
1934 r = create_vcpu_fd(vcpu);
1937 goto unlock_vcpu_destroy;
1940 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1942 atomic_inc(&kvm->online_vcpus);
1944 mutex_unlock(&kvm->lock);
1945 kvm_arch_vcpu_postcreate(vcpu);
1948 unlock_vcpu_destroy:
1949 mutex_unlock(&kvm->lock);
1951 kvm_arch_vcpu_destroy(vcpu);
1955 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1958 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1959 vcpu->sigset_active = 1;
1960 vcpu->sigset = *sigset;
1962 vcpu->sigset_active = 0;
1966 static long kvm_vcpu_ioctl(struct file *filp,
1967 unsigned int ioctl, unsigned long arg)
1969 struct kvm_vcpu *vcpu = filp->private_data;
1970 void __user *argp = (void __user *)arg;
1972 struct kvm_fpu *fpu = NULL;
1973 struct kvm_sregs *kvm_sregs = NULL;
1975 if (vcpu->kvm->mm != current->mm)
1978 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
1980 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1981 * so vcpu_load() would break it.
1983 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1984 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1988 r = vcpu_load(vcpu);
1996 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1997 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1999 case KVM_GET_REGS: {
2000 struct kvm_regs *kvm_regs;
2003 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2006 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2010 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2017 case KVM_SET_REGS: {
2018 struct kvm_regs *kvm_regs;
2021 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2022 if (IS_ERR(kvm_regs)) {
2023 r = PTR_ERR(kvm_regs);
2026 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2030 case KVM_GET_SREGS: {
2031 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2035 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2039 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2044 case KVM_SET_SREGS: {
2045 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2046 if (IS_ERR(kvm_sregs)) {
2047 r = PTR_ERR(kvm_sregs);
2051 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2054 case KVM_GET_MP_STATE: {
2055 struct kvm_mp_state mp_state;
2057 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2061 if (copy_to_user(argp, &mp_state, sizeof mp_state))
2066 case KVM_SET_MP_STATE: {
2067 struct kvm_mp_state mp_state;
2070 if (copy_from_user(&mp_state, argp, sizeof mp_state))
2072 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2075 case KVM_TRANSLATE: {
2076 struct kvm_translation tr;
2079 if (copy_from_user(&tr, argp, sizeof tr))
2081 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2085 if (copy_to_user(argp, &tr, sizeof tr))
2090 case KVM_SET_GUEST_DEBUG: {
2091 struct kvm_guest_debug dbg;
2094 if (copy_from_user(&dbg, argp, sizeof dbg))
2096 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2099 case KVM_SET_SIGNAL_MASK: {
2100 struct kvm_signal_mask __user *sigmask_arg = argp;
2101 struct kvm_signal_mask kvm_sigmask;
2102 sigset_t sigset, *p;
2107 if (copy_from_user(&kvm_sigmask, argp,
2108 sizeof kvm_sigmask))
2111 if (kvm_sigmask.len != sizeof sigset)
2114 if (copy_from_user(&sigset, sigmask_arg->sigset,
2119 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2123 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2127 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2131 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2137 fpu = memdup_user(argp, sizeof(*fpu));
2143 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2147 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2156 #ifdef CONFIG_COMPAT
2157 static long kvm_vcpu_compat_ioctl(struct file *filp,
2158 unsigned int ioctl, unsigned long arg)
2160 struct kvm_vcpu *vcpu = filp->private_data;
2161 void __user *argp = compat_ptr(arg);
2164 if (vcpu->kvm->mm != current->mm)
2168 case KVM_SET_SIGNAL_MASK: {
2169 struct kvm_signal_mask __user *sigmask_arg = argp;
2170 struct kvm_signal_mask kvm_sigmask;
2171 compat_sigset_t csigset;
2176 if (copy_from_user(&kvm_sigmask, argp,
2177 sizeof kvm_sigmask))
2180 if (kvm_sigmask.len != sizeof csigset)
2183 if (copy_from_user(&csigset, sigmask_arg->sigset,
2186 sigset_from_compat(&sigset, &csigset);
2187 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2189 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2193 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2201 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2202 int (*accessor)(struct kvm_device *dev,
2203 struct kvm_device_attr *attr),
2206 struct kvm_device_attr attr;
2211 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2214 return accessor(dev, &attr);
2217 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2220 struct kvm_device *dev = filp->private_data;
2223 case KVM_SET_DEVICE_ATTR:
2224 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2225 case KVM_GET_DEVICE_ATTR:
2226 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2227 case KVM_HAS_DEVICE_ATTR:
2228 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2230 if (dev->ops->ioctl)
2231 return dev->ops->ioctl(dev, ioctl, arg);
2237 static int kvm_device_release(struct inode *inode, struct file *filp)
2239 struct kvm_device *dev = filp->private_data;
2240 struct kvm *kvm = dev->kvm;
2246 static const struct file_operations kvm_device_fops = {
2247 .unlocked_ioctl = kvm_device_ioctl,
2248 #ifdef CONFIG_COMPAT
2249 .compat_ioctl = kvm_device_ioctl,
2251 .release = kvm_device_release,
2254 struct kvm_device *kvm_device_from_filp(struct file *filp)
2256 if (filp->f_op != &kvm_device_fops)
2259 return filp->private_data;
2262 static int kvm_ioctl_create_device(struct kvm *kvm,
2263 struct kvm_create_device *cd)
2265 struct kvm_device_ops *ops = NULL;
2266 struct kvm_device *dev;
2267 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2271 #ifdef CONFIG_KVM_MPIC
2272 case KVM_DEV_TYPE_FSL_MPIC_20:
2273 case KVM_DEV_TYPE_FSL_MPIC_42:
2274 ops = &kvm_mpic_ops;
2277 #ifdef CONFIG_KVM_XICS
2278 case KVM_DEV_TYPE_XICS:
2279 ops = &kvm_xics_ops;
2282 #ifdef CONFIG_KVM_VFIO
2283 case KVM_DEV_TYPE_VFIO:
2284 ops = &kvm_vfio_ops;
2287 #ifdef CONFIG_KVM_ARM_VGIC
2288 case KVM_DEV_TYPE_ARM_VGIC_V2:
2289 ops = &kvm_arm_vgic_v2_ops;
2293 case KVM_DEV_TYPE_FLIC:
2294 ops = &kvm_flic_ops;
2304 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2311 ret = ops->create(dev, cd->type);
2317 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2323 list_add(&dev->vm_node, &kvm->devices);
2329 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2332 case KVM_CAP_USER_MEMORY:
2333 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2334 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2335 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2336 case KVM_CAP_SET_BOOT_CPU_ID:
2338 case KVM_CAP_INTERNAL_ERROR_DATA:
2339 #ifdef CONFIG_HAVE_KVM_MSI
2340 case KVM_CAP_SIGNAL_MSI:
2342 #ifdef CONFIG_HAVE_KVM_IRQFD
2343 case KVM_CAP_IRQFD_RESAMPLE:
2345 case KVM_CAP_CHECK_EXTENSION_VM:
2347 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2348 case KVM_CAP_IRQ_ROUTING:
2349 return KVM_MAX_IRQ_ROUTES;
2354 return kvm_vm_ioctl_check_extension(kvm, arg);
2357 static long kvm_vm_ioctl(struct file *filp,
2358 unsigned int ioctl, unsigned long arg)
2360 struct kvm *kvm = filp->private_data;
2361 void __user *argp = (void __user *)arg;
2364 if (kvm->mm != current->mm)
2367 case KVM_CREATE_VCPU:
2368 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2370 case KVM_SET_USER_MEMORY_REGION: {
2371 struct kvm_userspace_memory_region kvm_userspace_mem;
2374 if (copy_from_user(&kvm_userspace_mem, argp,
2375 sizeof kvm_userspace_mem))
2378 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2381 case KVM_GET_DIRTY_LOG: {
2382 struct kvm_dirty_log log;
2385 if (copy_from_user(&log, argp, sizeof log))
2387 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2390 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2391 case KVM_REGISTER_COALESCED_MMIO: {
2392 struct kvm_coalesced_mmio_zone zone;
2394 if (copy_from_user(&zone, argp, sizeof zone))
2396 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2399 case KVM_UNREGISTER_COALESCED_MMIO: {
2400 struct kvm_coalesced_mmio_zone zone;
2402 if (copy_from_user(&zone, argp, sizeof zone))
2404 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2409 struct kvm_irqfd data;
2412 if (copy_from_user(&data, argp, sizeof data))
2414 r = kvm_irqfd(kvm, &data);
2417 case KVM_IOEVENTFD: {
2418 struct kvm_ioeventfd data;
2421 if (copy_from_user(&data, argp, sizeof data))
2423 r = kvm_ioeventfd(kvm, &data);
2426 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2427 case KVM_SET_BOOT_CPU_ID:
2429 mutex_lock(&kvm->lock);
2430 if (atomic_read(&kvm->online_vcpus) != 0)
2433 kvm->bsp_vcpu_id = arg;
2434 mutex_unlock(&kvm->lock);
2437 #ifdef CONFIG_HAVE_KVM_MSI
2438 case KVM_SIGNAL_MSI: {
2442 if (copy_from_user(&msi, argp, sizeof msi))
2444 r = kvm_send_userspace_msi(kvm, &msi);
2448 #ifdef __KVM_HAVE_IRQ_LINE
2449 case KVM_IRQ_LINE_STATUS:
2450 case KVM_IRQ_LINE: {
2451 struct kvm_irq_level irq_event;
2454 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2457 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
2458 ioctl == KVM_IRQ_LINE_STATUS);
2463 if (ioctl == KVM_IRQ_LINE_STATUS) {
2464 if (copy_to_user(argp, &irq_event, sizeof irq_event))
2472 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2473 case KVM_SET_GSI_ROUTING: {
2474 struct kvm_irq_routing routing;
2475 struct kvm_irq_routing __user *urouting;
2476 struct kvm_irq_routing_entry *entries;
2479 if (copy_from_user(&routing, argp, sizeof(routing)))
2482 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2487 entries = vmalloc(routing.nr * sizeof(*entries));
2492 if (copy_from_user(entries, urouting->entries,
2493 routing.nr * sizeof(*entries)))
2494 goto out_free_irq_routing;
2495 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2497 out_free_irq_routing:
2501 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
2502 case KVM_CREATE_DEVICE: {
2503 struct kvm_create_device cd;
2506 if (copy_from_user(&cd, argp, sizeof(cd)))
2509 r = kvm_ioctl_create_device(kvm, &cd);
2514 if (copy_to_user(argp, &cd, sizeof(cd)))
2520 case KVM_CHECK_EXTENSION:
2521 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
2524 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2526 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2532 #ifdef CONFIG_COMPAT
2533 struct compat_kvm_dirty_log {
2537 compat_uptr_t dirty_bitmap; /* one bit per page */
2542 static long kvm_vm_compat_ioctl(struct file *filp,
2543 unsigned int ioctl, unsigned long arg)
2545 struct kvm *kvm = filp->private_data;
2548 if (kvm->mm != current->mm)
2551 case KVM_GET_DIRTY_LOG: {
2552 struct compat_kvm_dirty_log compat_log;
2553 struct kvm_dirty_log log;
2556 if (copy_from_user(&compat_log, (void __user *)arg,
2557 sizeof(compat_log)))
2559 log.slot = compat_log.slot;
2560 log.padding1 = compat_log.padding1;
2561 log.padding2 = compat_log.padding2;
2562 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2564 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2568 r = kvm_vm_ioctl(filp, ioctl, arg);
2576 static struct file_operations kvm_vm_fops = {
2577 .release = kvm_vm_release,
2578 .unlocked_ioctl = kvm_vm_ioctl,
2579 #ifdef CONFIG_COMPAT
2580 .compat_ioctl = kvm_vm_compat_ioctl,
2582 .llseek = noop_llseek,
2585 static int kvm_dev_ioctl_create_vm(unsigned long type)
2590 kvm = kvm_create_vm(type);
2592 return PTR_ERR(kvm);
2593 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2594 r = kvm_coalesced_mmio_init(kvm);
2600 r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC);
2607 static long kvm_dev_ioctl(struct file *filp,
2608 unsigned int ioctl, unsigned long arg)
2613 case KVM_GET_API_VERSION:
2617 r = KVM_API_VERSION;
2620 r = kvm_dev_ioctl_create_vm(arg);
2622 case KVM_CHECK_EXTENSION:
2623 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
2625 case KVM_GET_VCPU_MMAP_SIZE:
2629 r = PAGE_SIZE; /* struct kvm_run */
2631 r += PAGE_SIZE; /* pio data page */
2633 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2634 r += PAGE_SIZE; /* coalesced mmio ring page */
2637 case KVM_TRACE_ENABLE:
2638 case KVM_TRACE_PAUSE:
2639 case KVM_TRACE_DISABLE:
2643 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2649 static struct file_operations kvm_chardev_ops = {
2650 .unlocked_ioctl = kvm_dev_ioctl,
2651 .compat_ioctl = kvm_dev_ioctl,
2652 .llseek = noop_llseek,
2655 static struct miscdevice kvm_dev = {
2661 static void hardware_enable_nolock(void *junk)
2663 int cpu = raw_smp_processor_id();
2666 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2669 cpumask_set_cpu(cpu, cpus_hardware_enabled);
2671 r = kvm_arch_hardware_enable(NULL);
2674 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2675 atomic_inc(&hardware_enable_failed);
2676 printk(KERN_INFO "kvm: enabling virtualization on "
2677 "CPU%d failed\n", cpu);
2681 static void hardware_enable(void)
2683 raw_spin_lock(&kvm_count_lock);
2684 if (kvm_usage_count)
2685 hardware_enable_nolock(NULL);
2686 raw_spin_unlock(&kvm_count_lock);
2689 static void hardware_disable_nolock(void *junk)
2691 int cpu = raw_smp_processor_id();
2693 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2695 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2696 kvm_arch_hardware_disable(NULL);
2699 static void hardware_disable(void)
2701 raw_spin_lock(&kvm_count_lock);
2702 if (kvm_usage_count)
2703 hardware_disable_nolock(NULL);
2704 raw_spin_unlock(&kvm_count_lock);
2707 static void hardware_disable_all_nolock(void)
2709 BUG_ON(!kvm_usage_count);
2712 if (!kvm_usage_count)
2713 on_each_cpu(hardware_disable_nolock, NULL, 1);
2716 static void hardware_disable_all(void)
2718 raw_spin_lock(&kvm_count_lock);
2719 hardware_disable_all_nolock();
2720 raw_spin_unlock(&kvm_count_lock);
2723 static int hardware_enable_all(void)
2727 raw_spin_lock(&kvm_count_lock);
2730 if (kvm_usage_count == 1) {
2731 atomic_set(&hardware_enable_failed, 0);
2732 on_each_cpu(hardware_enable_nolock, NULL, 1);
2734 if (atomic_read(&hardware_enable_failed)) {
2735 hardware_disable_all_nolock();
2740 raw_spin_unlock(&kvm_count_lock);
2745 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2750 val &= ~CPU_TASKS_FROZEN;
2753 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2758 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2766 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2770 * Some (well, at least mine) BIOSes hang on reboot if
2773 * And Intel TXT required VMX off for all cpu when system shutdown.
2775 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2776 kvm_rebooting = true;
2777 on_each_cpu(hardware_disable_nolock, NULL, 1);
2781 static struct notifier_block kvm_reboot_notifier = {
2782 .notifier_call = kvm_reboot,
2786 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2790 for (i = 0; i < bus->dev_count; i++) {
2791 struct kvm_io_device *pos = bus->range[i].dev;
2793 kvm_iodevice_destructor(pos);
2798 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
2799 const struct kvm_io_range *r2)
2801 if (r1->addr < r2->addr)
2803 if (r1->addr + r1->len > r2->addr + r2->len)
2808 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2810 return kvm_io_bus_cmp(p1, p2);
2813 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2814 gpa_t addr, int len)
2816 bus->range[bus->dev_count++] = (struct kvm_io_range) {
2822 sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2823 kvm_io_bus_sort_cmp, NULL);
2828 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2829 gpa_t addr, int len)
2831 struct kvm_io_range *range, key;
2834 key = (struct kvm_io_range) {
2839 range = bsearch(&key, bus->range, bus->dev_count,
2840 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2844 off = range - bus->range;
2846 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
2852 static int __kvm_io_bus_write(struct kvm_io_bus *bus,
2853 struct kvm_io_range *range, const void *val)
2857 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2861 while (idx < bus->dev_count &&
2862 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
2863 if (!kvm_iodevice_write(bus->range[idx].dev, range->addr,
2872 /* kvm_io_bus_write - called under kvm->slots_lock */
2873 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2874 int len, const void *val)
2876 struct kvm_io_bus *bus;
2877 struct kvm_io_range range;
2880 range = (struct kvm_io_range) {
2885 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2886 r = __kvm_io_bus_write(bus, &range, val);
2887 return r < 0 ? r : 0;
2890 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
2891 int kvm_io_bus_write_cookie(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2892 int len, const void *val, long cookie)
2894 struct kvm_io_bus *bus;
2895 struct kvm_io_range range;
2897 range = (struct kvm_io_range) {
2902 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2904 /* First try the device referenced by cookie. */
2905 if ((cookie >= 0) && (cookie < bus->dev_count) &&
2906 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
2907 if (!kvm_iodevice_write(bus->range[cookie].dev, addr, len,
2912 * cookie contained garbage; fall back to search and return the
2913 * correct cookie value.
2915 return __kvm_io_bus_write(bus, &range, val);
2918 static int __kvm_io_bus_read(struct kvm_io_bus *bus, struct kvm_io_range *range,
2923 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2927 while (idx < bus->dev_count &&
2928 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
2929 if (!kvm_iodevice_read(bus->range[idx].dev, range->addr,
2937 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
2939 /* kvm_io_bus_read - called under kvm->slots_lock */
2940 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2943 struct kvm_io_bus *bus;
2944 struct kvm_io_range range;
2947 range = (struct kvm_io_range) {
2952 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2953 r = __kvm_io_bus_read(bus, &range, val);
2954 return r < 0 ? r : 0;
2958 /* Caller must hold slots_lock. */
2959 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2960 int len, struct kvm_io_device *dev)
2962 struct kvm_io_bus *new_bus, *bus;
2964 bus = kvm->buses[bus_idx];
2965 /* exclude ioeventfd which is limited by maximum fd */
2966 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
2969 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
2970 sizeof(struct kvm_io_range)), GFP_KERNEL);
2973 memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
2974 sizeof(struct kvm_io_range)));
2975 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2976 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2977 synchronize_srcu_expedited(&kvm->srcu);
2983 /* Caller must hold slots_lock. */
2984 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2985 struct kvm_io_device *dev)
2988 struct kvm_io_bus *new_bus, *bus;
2990 bus = kvm->buses[bus_idx];
2992 for (i = 0; i < bus->dev_count; i++)
2993 if (bus->range[i].dev == dev) {
3001 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3002 sizeof(struct kvm_io_range)), GFP_KERNEL);
3006 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3007 new_bus->dev_count--;
3008 memcpy(new_bus->range + i, bus->range + i + 1,
3009 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3011 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3012 synchronize_srcu_expedited(&kvm->srcu);
3017 static struct notifier_block kvm_cpu_notifier = {
3018 .notifier_call = kvm_cpu_hotplug,
3021 static int vm_stat_get(void *_offset, u64 *val)
3023 unsigned offset = (long)_offset;
3027 spin_lock(&kvm_lock);
3028 list_for_each_entry(kvm, &vm_list, vm_list)
3029 *val += *(u32 *)((void *)kvm + offset);
3030 spin_unlock(&kvm_lock);
3034 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
3036 static int vcpu_stat_get(void *_offset, u64 *val)
3038 unsigned offset = (long)_offset;
3040 struct kvm_vcpu *vcpu;
3044 spin_lock(&kvm_lock);
3045 list_for_each_entry(kvm, &vm_list, vm_list)
3046 kvm_for_each_vcpu(i, vcpu, kvm)
3047 *val += *(u32 *)((void *)vcpu + offset);
3049 spin_unlock(&kvm_lock);
3053 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
3055 static const struct file_operations *stat_fops[] = {
3056 [KVM_STAT_VCPU] = &vcpu_stat_fops,
3057 [KVM_STAT_VM] = &vm_stat_fops,
3060 static int kvm_init_debug(void)
3063 struct kvm_stats_debugfs_item *p;
3065 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3066 if (kvm_debugfs_dir == NULL)
3069 for (p = debugfs_entries; p->name; ++p) {
3070 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
3071 (void *)(long)p->offset,
3072 stat_fops[p->kind]);
3073 if (p->dentry == NULL)
3080 debugfs_remove_recursive(kvm_debugfs_dir);
3085 static void kvm_exit_debug(void)
3087 struct kvm_stats_debugfs_item *p;
3089 for (p = debugfs_entries; p->name; ++p)
3090 debugfs_remove(p->dentry);
3091 debugfs_remove(kvm_debugfs_dir);
3094 static int kvm_suspend(void)
3096 if (kvm_usage_count)
3097 hardware_disable_nolock(NULL);
3101 static void kvm_resume(void)
3103 if (kvm_usage_count) {
3104 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3105 hardware_enable_nolock(NULL);
3109 static struct syscore_ops kvm_syscore_ops = {
3110 .suspend = kvm_suspend,
3111 .resume = kvm_resume,
3115 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3117 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3120 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3122 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3123 if (vcpu->preempted)
3124 vcpu->preempted = false;
3126 kvm_arch_vcpu_load(vcpu, cpu);
3129 static void kvm_sched_out(struct preempt_notifier *pn,
3130 struct task_struct *next)
3132 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3134 if (current->state == TASK_RUNNING)
3135 vcpu->preempted = true;
3136 kvm_arch_vcpu_put(vcpu);
3139 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3140 struct module *module)
3145 r = kvm_arch_init(opaque);
3150 * kvm_arch_init makes sure there's at most one caller
3151 * for architectures that support multiple implementations,
3152 * like intel and amd on x86.
3153 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3154 * conflicts in case kvm is already setup for another implementation.
3156 r = kvm_irqfd_init();
3160 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
3165 r = kvm_arch_hardware_setup();
3169 for_each_online_cpu(cpu) {
3170 smp_call_function_single(cpu,
3171 kvm_arch_check_processor_compat,
3177 r = register_cpu_notifier(&kvm_cpu_notifier);
3180 register_reboot_notifier(&kvm_reboot_notifier);
3182 /* A kmem cache lets us meet the alignment requirements of fx_save. */
3184 vcpu_align = __alignof__(struct kvm_vcpu);
3185 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
3187 if (!kvm_vcpu_cache) {
3192 r = kvm_async_pf_init();
3196 kvm_chardev_ops.owner = module;
3197 kvm_vm_fops.owner = module;
3198 kvm_vcpu_fops.owner = module;
3200 r = misc_register(&kvm_dev);
3202 printk(KERN_ERR "kvm: misc device register failed\n");
3206 register_syscore_ops(&kvm_syscore_ops);
3208 kvm_preempt_ops.sched_in = kvm_sched_in;
3209 kvm_preempt_ops.sched_out = kvm_sched_out;
3211 r = kvm_init_debug();
3213 printk(KERN_ERR "kvm: create debugfs files failed\n");
3220 unregister_syscore_ops(&kvm_syscore_ops);
3221 misc_deregister(&kvm_dev);
3223 kvm_async_pf_deinit();
3225 kmem_cache_destroy(kvm_vcpu_cache);
3227 unregister_reboot_notifier(&kvm_reboot_notifier);
3228 unregister_cpu_notifier(&kvm_cpu_notifier);
3231 kvm_arch_hardware_unsetup();
3233 free_cpumask_var(cpus_hardware_enabled);
3241 EXPORT_SYMBOL_GPL(kvm_init);
3246 misc_deregister(&kvm_dev);
3247 kmem_cache_destroy(kvm_vcpu_cache);
3248 kvm_async_pf_deinit();
3249 unregister_syscore_ops(&kvm_syscore_ops);
3250 unregister_reboot_notifier(&kvm_reboot_notifier);
3251 unregister_cpu_notifier(&kvm_cpu_notifier);
3252 on_each_cpu(hardware_disable_nolock, NULL, 1);
3253 kvm_arch_hardware_unsetup();
3256 free_cpumask_var(cpus_hardware_enabled);
3258 EXPORT_SYMBOL_GPL(kvm_exit);