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);
99 static void kvm_release_pfn_dirty(pfn_t pfn);
100 static void mark_page_dirty_in_slot(struct kvm *kvm,
101 struct kvm_memory_slot *memslot, gfn_t gfn);
103 __visible bool kvm_rebooting;
104 EXPORT_SYMBOL_GPL(kvm_rebooting);
106 static bool largepages_enabled = true;
108 bool kvm_is_mmio_pfn(pfn_t pfn)
111 return PageReserved(pfn_to_page(pfn));
117 * Switches to specified vcpu, until a matching vcpu_put()
119 int vcpu_load(struct kvm_vcpu *vcpu)
123 if (mutex_lock_killable(&vcpu->mutex))
125 if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
126 /* The thread running this VCPU changed. */
127 struct pid *oldpid = vcpu->pid;
128 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
129 rcu_assign_pointer(vcpu->pid, newpid);
135 preempt_notifier_register(&vcpu->preempt_notifier);
136 kvm_arch_vcpu_load(vcpu, cpu);
141 void vcpu_put(struct kvm_vcpu *vcpu)
144 kvm_arch_vcpu_put(vcpu);
145 preempt_notifier_unregister(&vcpu->preempt_notifier);
147 mutex_unlock(&vcpu->mutex);
150 static void ack_flush(void *_completed)
154 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
159 struct kvm_vcpu *vcpu;
161 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
164 kvm_for_each_vcpu(i, vcpu, kvm) {
165 kvm_make_request(req, vcpu);
168 /* Set ->requests bit before we read ->mode */
171 if (cpus != NULL && cpu != -1 && cpu != me &&
172 kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
173 cpumask_set_cpu(cpu, cpus);
175 if (unlikely(cpus == NULL))
176 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
177 else if (!cpumask_empty(cpus))
178 smp_call_function_many(cpus, ack_flush, NULL, 1);
182 free_cpumask_var(cpus);
186 void kvm_flush_remote_tlbs(struct kvm *kvm)
188 long dirty_count = kvm->tlbs_dirty;
191 if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
192 ++kvm->stat.remote_tlb_flush;
193 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
195 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
197 void kvm_reload_remote_mmus(struct kvm *kvm)
199 make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
202 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
204 make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
207 void kvm_make_scan_ioapic_request(struct kvm *kvm)
209 make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
212 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
217 mutex_init(&vcpu->mutex);
222 init_waitqueue_head(&vcpu->wq);
223 kvm_async_pf_vcpu_init(vcpu);
225 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
230 vcpu->run = page_address(page);
232 kvm_vcpu_set_in_spin_loop(vcpu, false);
233 kvm_vcpu_set_dy_eligible(vcpu, false);
234 vcpu->preempted = false;
236 r = kvm_arch_vcpu_init(vcpu);
242 free_page((unsigned long)vcpu->run);
246 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
248 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
251 kvm_arch_vcpu_uninit(vcpu);
252 free_page((unsigned long)vcpu->run);
254 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
256 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
257 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
259 return container_of(mn, struct kvm, mmu_notifier);
262 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
263 struct mm_struct *mm,
264 unsigned long address)
266 struct kvm *kvm = mmu_notifier_to_kvm(mn);
267 int need_tlb_flush, idx;
270 * When ->invalidate_page runs, the linux pte has been zapped
271 * already but the page is still allocated until
272 * ->invalidate_page returns. So if we increase the sequence
273 * here the kvm page fault will notice if the spte can't be
274 * established because the page is going to be freed. If
275 * instead the kvm page fault establishes the spte before
276 * ->invalidate_page runs, kvm_unmap_hva will release it
279 * The sequence increase only need to be seen at spin_unlock
280 * time, and not at spin_lock time.
282 * Increasing the sequence after the spin_unlock would be
283 * unsafe because the kvm page fault could then establish the
284 * pte after kvm_unmap_hva returned, without noticing the page
285 * is going to be freed.
287 idx = srcu_read_lock(&kvm->srcu);
288 spin_lock(&kvm->mmu_lock);
290 kvm->mmu_notifier_seq++;
291 need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
292 /* we've to flush the tlb before the pages can be freed */
294 kvm_flush_remote_tlbs(kvm);
296 spin_unlock(&kvm->mmu_lock);
297 srcu_read_unlock(&kvm->srcu, idx);
300 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
301 struct mm_struct *mm,
302 unsigned long address,
305 struct kvm *kvm = mmu_notifier_to_kvm(mn);
308 idx = srcu_read_lock(&kvm->srcu);
309 spin_lock(&kvm->mmu_lock);
310 kvm->mmu_notifier_seq++;
311 kvm_set_spte_hva(kvm, address, pte);
312 spin_unlock(&kvm->mmu_lock);
313 srcu_read_unlock(&kvm->srcu, idx);
316 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
317 struct mm_struct *mm,
321 struct kvm *kvm = mmu_notifier_to_kvm(mn);
322 int need_tlb_flush = 0, idx;
324 idx = srcu_read_lock(&kvm->srcu);
325 spin_lock(&kvm->mmu_lock);
327 * The count increase must become visible at unlock time as no
328 * spte can be established without taking the mmu_lock and
329 * count is also read inside the mmu_lock critical section.
331 kvm->mmu_notifier_count++;
332 need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
333 need_tlb_flush |= kvm->tlbs_dirty;
334 /* we've to flush the tlb before the pages can be freed */
336 kvm_flush_remote_tlbs(kvm);
338 spin_unlock(&kvm->mmu_lock);
339 srcu_read_unlock(&kvm->srcu, idx);
342 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
343 struct mm_struct *mm,
347 struct kvm *kvm = mmu_notifier_to_kvm(mn);
349 spin_lock(&kvm->mmu_lock);
351 * This sequence increase will notify the kvm page fault that
352 * the page that is going to be mapped in the spte could have
355 kvm->mmu_notifier_seq++;
358 * The above sequence increase must be visible before the
359 * below count decrease, which is ensured by the smp_wmb above
360 * in conjunction with the smp_rmb in mmu_notifier_retry().
362 kvm->mmu_notifier_count--;
363 spin_unlock(&kvm->mmu_lock);
365 BUG_ON(kvm->mmu_notifier_count < 0);
368 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
369 struct mm_struct *mm,
370 unsigned long address)
372 struct kvm *kvm = mmu_notifier_to_kvm(mn);
375 idx = srcu_read_lock(&kvm->srcu);
376 spin_lock(&kvm->mmu_lock);
378 young = kvm_age_hva(kvm, address);
380 kvm_flush_remote_tlbs(kvm);
382 spin_unlock(&kvm->mmu_lock);
383 srcu_read_unlock(&kvm->srcu, idx);
388 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
389 struct mm_struct *mm,
390 unsigned long address)
392 struct kvm *kvm = mmu_notifier_to_kvm(mn);
395 idx = srcu_read_lock(&kvm->srcu);
396 spin_lock(&kvm->mmu_lock);
397 young = kvm_test_age_hva(kvm, address);
398 spin_unlock(&kvm->mmu_lock);
399 srcu_read_unlock(&kvm->srcu, idx);
404 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
405 struct mm_struct *mm)
407 struct kvm *kvm = mmu_notifier_to_kvm(mn);
410 idx = srcu_read_lock(&kvm->srcu);
411 kvm_arch_flush_shadow_all(kvm);
412 srcu_read_unlock(&kvm->srcu, idx);
415 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
416 .invalidate_page = kvm_mmu_notifier_invalidate_page,
417 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
418 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
419 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
420 .test_young = kvm_mmu_notifier_test_young,
421 .change_pte = kvm_mmu_notifier_change_pte,
422 .release = kvm_mmu_notifier_release,
425 static int kvm_init_mmu_notifier(struct kvm *kvm)
427 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
428 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
431 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
433 static int kvm_init_mmu_notifier(struct kvm *kvm)
438 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
440 static void kvm_init_memslots_id(struct kvm *kvm)
443 struct kvm_memslots *slots = kvm->memslots;
445 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
446 slots->id_to_index[i] = slots->memslots[i].id = i;
449 static struct kvm *kvm_create_vm(unsigned long type)
452 struct kvm *kvm = kvm_arch_alloc_vm();
455 return ERR_PTR(-ENOMEM);
457 r = kvm_arch_init_vm(kvm, type);
459 goto out_err_no_disable;
461 r = hardware_enable_all();
463 goto out_err_no_disable;
465 #ifdef CONFIG_HAVE_KVM_IRQCHIP
466 INIT_HLIST_HEAD(&kvm->mask_notifier_list);
468 #ifdef CONFIG_HAVE_KVM_IRQFD
469 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
472 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
475 kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
477 goto out_err_no_srcu;
480 * Init kvm generation close to the maximum to easily test the
481 * code of handling generation number wrap-around.
483 kvm->memslots->generation = -150;
485 kvm_init_memslots_id(kvm);
486 if (init_srcu_struct(&kvm->srcu))
487 goto out_err_no_srcu;
488 if (init_srcu_struct(&kvm->irq_srcu))
489 goto out_err_no_irq_srcu;
490 for (i = 0; i < KVM_NR_BUSES; i++) {
491 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
497 spin_lock_init(&kvm->mmu_lock);
498 kvm->mm = current->mm;
499 atomic_inc(&kvm->mm->mm_count);
500 kvm_eventfd_init(kvm);
501 mutex_init(&kvm->lock);
502 mutex_init(&kvm->irq_lock);
503 mutex_init(&kvm->slots_lock);
504 atomic_set(&kvm->users_count, 1);
505 INIT_LIST_HEAD(&kvm->devices);
507 r = kvm_init_mmu_notifier(kvm);
511 spin_lock(&kvm_lock);
512 list_add(&kvm->vm_list, &vm_list);
513 spin_unlock(&kvm_lock);
518 cleanup_srcu_struct(&kvm->irq_srcu);
520 cleanup_srcu_struct(&kvm->srcu);
522 hardware_disable_all();
524 for (i = 0; i < KVM_NR_BUSES; i++)
525 kfree(kvm->buses[i]);
526 kfree(kvm->memslots);
527 kvm_arch_free_vm(kvm);
532 * Avoid using vmalloc for a small buffer.
533 * Should not be used when the size is statically known.
535 void *kvm_kvzalloc(unsigned long size)
537 if (size > PAGE_SIZE)
538 return vzalloc(size);
540 return kzalloc(size, GFP_KERNEL);
543 void kvm_kvfree(const void *addr)
545 if (is_vmalloc_addr(addr))
551 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
553 if (!memslot->dirty_bitmap)
556 kvm_kvfree(memslot->dirty_bitmap);
557 memslot->dirty_bitmap = NULL;
561 * Free any memory in @free but not in @dont.
563 static void kvm_free_physmem_slot(struct kvm *kvm, struct kvm_memory_slot *free,
564 struct kvm_memory_slot *dont)
566 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
567 kvm_destroy_dirty_bitmap(free);
569 kvm_arch_free_memslot(kvm, free, dont);
574 static void kvm_free_physmem(struct kvm *kvm)
576 struct kvm_memslots *slots = kvm->memslots;
577 struct kvm_memory_slot *memslot;
579 kvm_for_each_memslot(memslot, slots)
580 kvm_free_physmem_slot(kvm, memslot, NULL);
582 kfree(kvm->memslots);
585 static void kvm_destroy_devices(struct kvm *kvm)
587 struct list_head *node, *tmp;
589 list_for_each_safe(node, tmp, &kvm->devices) {
590 struct kvm_device *dev =
591 list_entry(node, struct kvm_device, vm_node);
594 dev->ops->destroy(dev);
598 static void kvm_destroy_vm(struct kvm *kvm)
601 struct mm_struct *mm = kvm->mm;
603 kvm_arch_sync_events(kvm);
604 spin_lock(&kvm_lock);
605 list_del(&kvm->vm_list);
606 spin_unlock(&kvm_lock);
607 kvm_free_irq_routing(kvm);
608 for (i = 0; i < KVM_NR_BUSES; i++)
609 kvm_io_bus_destroy(kvm->buses[i]);
610 kvm_coalesced_mmio_free(kvm);
611 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
612 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
614 kvm_arch_flush_shadow_all(kvm);
616 kvm_arch_destroy_vm(kvm);
617 kvm_destroy_devices(kvm);
618 kvm_free_physmem(kvm);
619 cleanup_srcu_struct(&kvm->irq_srcu);
620 cleanup_srcu_struct(&kvm->srcu);
621 kvm_arch_free_vm(kvm);
622 hardware_disable_all();
626 void kvm_get_kvm(struct kvm *kvm)
628 atomic_inc(&kvm->users_count);
630 EXPORT_SYMBOL_GPL(kvm_get_kvm);
632 void kvm_put_kvm(struct kvm *kvm)
634 if (atomic_dec_and_test(&kvm->users_count))
637 EXPORT_SYMBOL_GPL(kvm_put_kvm);
640 static int kvm_vm_release(struct inode *inode, struct file *filp)
642 struct kvm *kvm = filp->private_data;
644 kvm_irqfd_release(kvm);
651 * Allocation size is twice as large as the actual dirty bitmap size.
652 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
654 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
656 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
658 memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
659 if (!memslot->dirty_bitmap)
665 static int cmp_memslot(const void *slot1, const void *slot2)
667 struct kvm_memory_slot *s1, *s2;
669 s1 = (struct kvm_memory_slot *)slot1;
670 s2 = (struct kvm_memory_slot *)slot2;
672 if (s1->npages < s2->npages)
674 if (s1->npages > s2->npages)
681 * Sort the memslots base on its size, so the larger slots
682 * will get better fit.
684 static void sort_memslots(struct kvm_memslots *slots)
688 sort(slots->memslots, KVM_MEM_SLOTS_NUM,
689 sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
691 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
692 slots->id_to_index[slots->memslots[i].id] = i;
695 static void update_memslots(struct kvm_memslots *slots,
696 struct kvm_memory_slot *new)
700 struct kvm_memory_slot *old = id_to_memslot(slots, id);
701 unsigned long npages = old->npages;
704 if (new->npages != npages)
705 sort_memslots(slots);
709 static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
711 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
713 #ifdef __KVM_HAVE_READONLY_MEM
714 valid_flags |= KVM_MEM_READONLY;
717 if (mem->flags & ~valid_flags)
723 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
724 struct kvm_memslots *slots, struct kvm_memory_slot *new)
726 struct kvm_memslots *old_memslots = kvm->memslots;
729 * Set the low bit in the generation, which disables SPTE caching
730 * until the end of synchronize_srcu_expedited.
732 WARN_ON(old_memslots->generation & 1);
733 slots->generation = old_memslots->generation + 1;
735 update_memslots(slots, new);
736 rcu_assign_pointer(kvm->memslots, slots);
737 synchronize_srcu_expedited(&kvm->srcu);
740 * Increment the new memslot generation a second time. This prevents
741 * vm exits that race with memslot updates from caching a memslot
742 * generation that will (potentially) be valid forever.
746 kvm_arch_memslots_updated(kvm);
752 * Allocate some memory and give it an address in the guest physical address
755 * Discontiguous memory is allowed, mostly for framebuffers.
757 * Must be called holding mmap_sem for write.
759 int __kvm_set_memory_region(struct kvm *kvm,
760 struct kvm_userspace_memory_region *mem)
764 unsigned long npages;
765 struct kvm_memory_slot *slot;
766 struct kvm_memory_slot old, new;
767 struct kvm_memslots *slots = NULL, *old_memslots;
768 enum kvm_mr_change change;
770 r = check_memory_region_flags(mem);
775 /* General sanity checks */
776 if (mem->memory_size & (PAGE_SIZE - 1))
778 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
780 /* We can read the guest memory with __xxx_user() later on. */
781 if ((mem->slot < KVM_USER_MEM_SLOTS) &&
782 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
783 !access_ok(VERIFY_WRITE,
784 (void __user *)(unsigned long)mem->userspace_addr,
787 if (mem->slot >= KVM_MEM_SLOTS_NUM)
789 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
792 slot = id_to_memslot(kvm->memslots, mem->slot);
793 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
794 npages = mem->memory_size >> PAGE_SHIFT;
797 if (npages > KVM_MEM_MAX_NR_PAGES)
801 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
806 new.base_gfn = base_gfn;
808 new.flags = mem->flags;
813 change = KVM_MR_CREATE;
814 else { /* Modify an existing slot. */
815 if ((mem->userspace_addr != old.userspace_addr) ||
816 (npages != old.npages) ||
817 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
820 if (base_gfn != old.base_gfn)
821 change = KVM_MR_MOVE;
822 else if (new.flags != old.flags)
823 change = KVM_MR_FLAGS_ONLY;
824 else { /* Nothing to change. */
829 } else if (old.npages) {
830 change = KVM_MR_DELETE;
831 } else /* Modify a non-existent slot: disallowed. */
834 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
835 /* Check for overlaps */
837 kvm_for_each_memslot(slot, kvm->memslots) {
838 if ((slot->id >= KVM_USER_MEM_SLOTS) ||
839 (slot->id == mem->slot))
841 if (!((base_gfn + npages <= slot->base_gfn) ||
842 (base_gfn >= slot->base_gfn + slot->npages)))
847 /* Free page dirty bitmap if unneeded */
848 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
849 new.dirty_bitmap = NULL;
852 if (change == KVM_MR_CREATE) {
853 new.userspace_addr = mem->userspace_addr;
855 if (kvm_arch_create_memslot(kvm, &new, npages))
859 /* Allocate page dirty bitmap if needed */
860 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
861 if (kvm_create_dirty_bitmap(&new) < 0)
865 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
867 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
871 slot = id_to_memslot(slots, mem->slot);
872 slot->flags |= KVM_MEMSLOT_INVALID;
874 old_memslots = install_new_memslots(kvm, slots, NULL);
876 /* slot was deleted or moved, clear iommu mapping */
877 kvm_iommu_unmap_pages(kvm, &old);
878 /* From this point no new shadow pages pointing to a deleted,
879 * or moved, memslot will be created.
881 * validation of sp->gfn happens in:
882 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
883 * - kvm_is_visible_gfn (mmu_check_roots)
885 kvm_arch_flush_shadow_memslot(kvm, slot);
886 slots = old_memslots;
889 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
895 * We can re-use the old_memslots from above, the only difference
896 * from the currently installed memslots is the invalid flag. This
897 * will get overwritten by update_memslots anyway.
900 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
906 /* actual memory is freed via old in kvm_free_physmem_slot below */
907 if (change == KVM_MR_DELETE) {
908 new.dirty_bitmap = NULL;
909 memset(&new.arch, 0, sizeof(new.arch));
912 old_memslots = install_new_memslots(kvm, slots, &new);
914 kvm_arch_commit_memory_region(kvm, mem, &old, change);
916 kvm_free_physmem_slot(kvm, &old, &new);
920 * IOMMU mapping: New slots need to be mapped. Old slots need to be
921 * un-mapped and re-mapped if their base changes. Since base change
922 * unmapping is handled above with slot deletion, mapping alone is
923 * needed here. Anything else the iommu might care about for existing
924 * slots (size changes, userspace addr changes and read-only flag
925 * changes) is disallowed above, so any other attribute changes getting
926 * here can be skipped.
928 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
929 r = kvm_iommu_map_pages(kvm, &new);
938 kvm_free_physmem_slot(kvm, &new, &old);
942 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
944 int kvm_set_memory_region(struct kvm *kvm,
945 struct kvm_userspace_memory_region *mem)
949 mutex_lock(&kvm->slots_lock);
950 r = __kvm_set_memory_region(kvm, mem);
951 mutex_unlock(&kvm->slots_lock);
954 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
956 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
957 struct kvm_userspace_memory_region *mem)
959 if (mem->slot >= KVM_USER_MEM_SLOTS)
961 return kvm_set_memory_region(kvm, mem);
964 int kvm_get_dirty_log(struct kvm *kvm,
965 struct kvm_dirty_log *log, int *is_dirty)
967 struct kvm_memory_slot *memslot;
970 unsigned long any = 0;
973 if (log->slot >= KVM_USER_MEM_SLOTS)
976 memslot = id_to_memslot(kvm->memslots, log->slot);
978 if (!memslot->dirty_bitmap)
981 n = kvm_dirty_bitmap_bytes(memslot);
983 for (i = 0; !any && i < n/sizeof(long); ++i)
984 any = memslot->dirty_bitmap[i];
987 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
997 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
999 bool kvm_largepages_enabled(void)
1001 return largepages_enabled;
1004 void kvm_disable_largepages(void)
1006 largepages_enabled = false;
1008 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1010 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1012 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1014 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1016 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1018 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1020 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1021 memslot->flags & KVM_MEMSLOT_INVALID)
1026 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1028 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1030 struct vm_area_struct *vma;
1031 unsigned long addr, size;
1035 addr = gfn_to_hva(kvm, gfn);
1036 if (kvm_is_error_hva(addr))
1039 down_read(¤t->mm->mmap_sem);
1040 vma = find_vma(current->mm, addr);
1044 size = vma_kernel_pagesize(vma);
1047 up_read(¤t->mm->mmap_sem);
1052 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1054 return slot->flags & KVM_MEM_READONLY;
1057 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1058 gfn_t *nr_pages, bool write)
1060 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1061 return KVM_HVA_ERR_BAD;
1063 if (memslot_is_readonly(slot) && write)
1064 return KVM_HVA_ERR_RO_BAD;
1067 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1069 return __gfn_to_hva_memslot(slot, gfn);
1072 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1075 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1078 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1081 return gfn_to_hva_many(slot, gfn, NULL);
1083 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1085 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1087 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1089 EXPORT_SYMBOL_GPL(gfn_to_hva);
1092 * If writable is set to false, the hva returned by this function is only
1093 * allowed to be read.
1095 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1097 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1098 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1100 if (!kvm_is_error_hva(hva) && writable)
1101 *writable = !memslot_is_readonly(slot);
1106 static int kvm_read_hva(void *data, void __user *hva, int len)
1108 return __copy_from_user(data, hva, len);
1111 static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1113 return __copy_from_user_inatomic(data, hva, len);
1116 static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1117 unsigned long start, int write, struct page **page)
1119 int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1122 flags |= FOLL_WRITE;
1124 return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1127 static inline int check_user_page_hwpoison(unsigned long addr)
1129 int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1131 rc = __get_user_pages(current, current->mm, addr, 1,
1132 flags, NULL, NULL, NULL);
1133 return rc == -EHWPOISON;
1137 * The atomic path to get the writable pfn which will be stored in @pfn,
1138 * true indicates success, otherwise false is returned.
1140 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1141 bool write_fault, bool *writable, pfn_t *pfn)
1143 struct page *page[1];
1146 if (!(async || atomic))
1150 * Fast pin a writable pfn only if it is a write fault request
1151 * or the caller allows to map a writable pfn for a read fault
1154 if (!(write_fault || writable))
1157 npages = __get_user_pages_fast(addr, 1, 1, page);
1159 *pfn = page_to_pfn(page[0]);
1170 * The slow path to get the pfn of the specified host virtual address,
1171 * 1 indicates success, -errno is returned if error is detected.
1173 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1174 bool *writable, pfn_t *pfn)
1176 struct page *page[1];
1182 *writable = write_fault;
1185 down_read(¤t->mm->mmap_sem);
1186 npages = get_user_page_nowait(current, current->mm,
1187 addr, write_fault, page);
1188 up_read(¤t->mm->mmap_sem);
1190 npages = get_user_pages_fast(addr, 1, write_fault,
1195 /* map read fault as writable if possible */
1196 if (unlikely(!write_fault) && writable) {
1197 struct page *wpage[1];
1199 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1208 *pfn = page_to_pfn(page[0]);
1212 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1214 if (unlikely(!(vma->vm_flags & VM_READ)))
1217 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1224 * Pin guest page in memory and return its pfn.
1225 * @addr: host virtual address which maps memory to the guest
1226 * @atomic: whether this function can sleep
1227 * @async: whether this function need to wait IO complete if the
1228 * host page is not in the memory
1229 * @write_fault: whether we should get a writable host page
1230 * @writable: whether it allows to map a writable host page for !@write_fault
1232 * The function will map a writable host page for these two cases:
1233 * 1): @write_fault = true
1234 * 2): @write_fault = false && @writable, @writable will tell the caller
1235 * whether the mapping is writable.
1237 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1238 bool write_fault, bool *writable)
1240 struct vm_area_struct *vma;
1244 /* we can do it either atomically or asynchronously, not both */
1245 BUG_ON(atomic && async);
1247 if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1251 return KVM_PFN_ERR_FAULT;
1253 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1257 down_read(¤t->mm->mmap_sem);
1258 if (npages == -EHWPOISON ||
1259 (!async && check_user_page_hwpoison(addr))) {
1260 pfn = KVM_PFN_ERR_HWPOISON;
1264 vma = find_vma_intersection(current->mm, addr, addr + 1);
1267 pfn = KVM_PFN_ERR_FAULT;
1268 else if ((vma->vm_flags & VM_PFNMAP)) {
1269 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1271 BUG_ON(!kvm_is_mmio_pfn(pfn));
1273 if (async && vma_is_valid(vma, write_fault))
1275 pfn = KVM_PFN_ERR_FAULT;
1278 up_read(¤t->mm->mmap_sem);
1283 __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1284 bool *async, bool write_fault, bool *writable)
1286 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1288 if (addr == KVM_HVA_ERR_RO_BAD)
1289 return KVM_PFN_ERR_RO_FAULT;
1291 if (kvm_is_error_hva(addr))
1292 return KVM_PFN_NOSLOT;
1294 /* Do not map writable pfn in the readonly memslot. */
1295 if (writable && memslot_is_readonly(slot)) {
1300 return hva_to_pfn(addr, atomic, async, write_fault,
1304 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1305 bool write_fault, bool *writable)
1307 struct kvm_memory_slot *slot;
1312 slot = gfn_to_memslot(kvm, gfn);
1314 return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1318 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1320 return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1322 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1324 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1325 bool write_fault, bool *writable)
1327 return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1329 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1331 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1333 return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1335 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1337 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1340 return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1342 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1344 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1346 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1349 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1351 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1353 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1355 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1361 addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1362 if (kvm_is_error_hva(addr))
1365 if (entry < nr_pages)
1368 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1370 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1372 static struct page *kvm_pfn_to_page(pfn_t pfn)
1374 if (is_error_noslot_pfn(pfn))
1375 return KVM_ERR_PTR_BAD_PAGE;
1377 if (kvm_is_mmio_pfn(pfn)) {
1379 return KVM_ERR_PTR_BAD_PAGE;
1382 return pfn_to_page(pfn);
1385 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1389 pfn = gfn_to_pfn(kvm, gfn);
1391 return kvm_pfn_to_page(pfn);
1394 EXPORT_SYMBOL_GPL(gfn_to_page);
1396 void kvm_release_page_clean(struct page *page)
1398 WARN_ON(is_error_page(page));
1400 kvm_release_pfn_clean(page_to_pfn(page));
1402 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1404 void kvm_release_pfn_clean(pfn_t pfn)
1406 if (!is_error_noslot_pfn(pfn) && !kvm_is_mmio_pfn(pfn))
1407 put_page(pfn_to_page(pfn));
1409 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1411 void kvm_release_page_dirty(struct page *page)
1413 WARN_ON(is_error_page(page));
1415 kvm_release_pfn_dirty(page_to_pfn(page));
1417 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1419 static void kvm_release_pfn_dirty(pfn_t pfn)
1421 kvm_set_pfn_dirty(pfn);
1422 kvm_release_pfn_clean(pfn);
1425 void kvm_set_pfn_dirty(pfn_t pfn)
1427 if (!kvm_is_mmio_pfn(pfn)) {
1428 struct page *page = pfn_to_page(pfn);
1429 if (!PageReserved(page))
1433 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1435 void kvm_set_pfn_accessed(pfn_t pfn)
1437 if (!kvm_is_mmio_pfn(pfn))
1438 mark_page_accessed(pfn_to_page(pfn));
1440 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1442 void kvm_get_pfn(pfn_t pfn)
1444 if (!kvm_is_mmio_pfn(pfn))
1445 get_page(pfn_to_page(pfn));
1447 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1449 static int next_segment(unsigned long len, int offset)
1451 if (len > PAGE_SIZE - offset)
1452 return PAGE_SIZE - offset;
1457 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1463 addr = gfn_to_hva_prot(kvm, gfn, NULL);
1464 if (kvm_is_error_hva(addr))
1466 r = kvm_read_hva(data, (void __user *)addr + offset, len);
1471 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1473 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1475 gfn_t gfn = gpa >> PAGE_SHIFT;
1477 int offset = offset_in_page(gpa);
1480 while ((seg = next_segment(len, offset)) != 0) {
1481 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1491 EXPORT_SYMBOL_GPL(kvm_read_guest);
1493 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1498 gfn_t gfn = gpa >> PAGE_SHIFT;
1499 int offset = offset_in_page(gpa);
1501 addr = gfn_to_hva_prot(kvm, gfn, NULL);
1502 if (kvm_is_error_hva(addr))
1504 pagefault_disable();
1505 r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
1511 EXPORT_SYMBOL(kvm_read_guest_atomic);
1513 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1514 int offset, int len)
1519 addr = gfn_to_hva(kvm, gfn);
1520 if (kvm_is_error_hva(addr))
1522 r = __copy_to_user((void __user *)addr + offset, data, len);
1525 mark_page_dirty(kvm, gfn);
1528 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1530 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1533 gfn_t gfn = gpa >> PAGE_SHIFT;
1535 int offset = offset_in_page(gpa);
1538 while ((seg = next_segment(len, offset)) != 0) {
1539 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1550 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1551 gpa_t gpa, unsigned long len)
1553 struct kvm_memslots *slots = kvm_memslots(kvm);
1554 int offset = offset_in_page(gpa);
1555 gfn_t start_gfn = gpa >> PAGE_SHIFT;
1556 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1557 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1558 gfn_t nr_pages_avail;
1561 ghc->generation = slots->generation;
1563 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1564 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, &nr_pages_avail);
1565 if (!kvm_is_error_hva(ghc->hva) && nr_pages_avail >= nr_pages_needed) {
1569 * If the requested region crosses two memslots, we still
1570 * verify that the entire region is valid here.
1572 while (start_gfn <= end_gfn) {
1573 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1574 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1576 if (kvm_is_error_hva(ghc->hva))
1578 start_gfn += nr_pages_avail;
1580 /* Use the slow path for cross page reads and writes. */
1581 ghc->memslot = NULL;
1585 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1587 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1588 void *data, unsigned long len)
1590 struct kvm_memslots *slots = kvm_memslots(kvm);
1593 BUG_ON(len > ghc->len);
1595 if (slots->generation != ghc->generation)
1596 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1598 if (unlikely(!ghc->memslot))
1599 return kvm_write_guest(kvm, ghc->gpa, data, len);
1601 if (kvm_is_error_hva(ghc->hva))
1604 r = __copy_to_user((void __user *)ghc->hva, data, len);
1607 mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1611 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1613 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1614 void *data, unsigned long len)
1616 struct kvm_memslots *slots = kvm_memslots(kvm);
1619 BUG_ON(len > ghc->len);
1621 if (slots->generation != ghc->generation)
1622 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1624 if (unlikely(!ghc->memslot))
1625 return kvm_read_guest(kvm, ghc->gpa, data, len);
1627 if (kvm_is_error_hva(ghc->hva))
1630 r = __copy_from_user(data, (void __user *)ghc->hva, len);
1636 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1638 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1640 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
1642 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
1644 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1646 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1648 gfn_t gfn = gpa >> PAGE_SHIFT;
1650 int offset = offset_in_page(gpa);
1653 while ((seg = next_segment(len, offset)) != 0) {
1654 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1663 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1665 static void mark_page_dirty_in_slot(struct kvm *kvm,
1666 struct kvm_memory_slot *memslot,
1669 if (memslot && memslot->dirty_bitmap) {
1670 unsigned long rel_gfn = gfn - memslot->base_gfn;
1672 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1676 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1678 struct kvm_memory_slot *memslot;
1680 memslot = gfn_to_memslot(kvm, gfn);
1681 mark_page_dirty_in_slot(kvm, memslot, gfn);
1683 EXPORT_SYMBOL_GPL(mark_page_dirty);
1686 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1688 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1693 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1695 if (kvm_arch_vcpu_runnable(vcpu)) {
1696 kvm_make_request(KVM_REQ_UNHALT, vcpu);
1699 if (kvm_cpu_has_pending_timer(vcpu))
1701 if (signal_pending(current))
1707 finish_wait(&vcpu->wq, &wait);
1709 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
1713 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1715 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1718 int cpu = vcpu->cpu;
1719 wait_queue_head_t *wqp;
1721 wqp = kvm_arch_vcpu_wq(vcpu);
1722 if (waitqueue_active(wqp)) {
1723 wake_up_interruptible(wqp);
1724 ++vcpu->stat.halt_wakeup;
1728 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1729 if (kvm_arch_vcpu_should_kick(vcpu))
1730 smp_send_reschedule(cpu);
1733 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
1734 #endif /* !CONFIG_S390 */
1736 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
1739 struct task_struct *task = NULL;
1743 pid = rcu_dereference(target->pid);
1745 task = get_pid_task(target->pid, PIDTYPE_PID);
1749 if (task->flags & PF_VCPU) {
1750 put_task_struct(task);
1753 ret = yield_to(task, 1);
1754 put_task_struct(task);
1758 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1761 * Helper that checks whether a VCPU is eligible for directed yield.
1762 * Most eligible candidate to yield is decided by following heuristics:
1764 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1765 * (preempted lock holder), indicated by @in_spin_loop.
1766 * Set at the beiginning and cleared at the end of interception/PLE handler.
1768 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1769 * chance last time (mostly it has become eligible now since we have probably
1770 * yielded to lockholder in last iteration. This is done by toggling
1771 * @dy_eligible each time a VCPU checked for eligibility.)
1773 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1774 * to preempted lock-holder could result in wrong VCPU selection and CPU
1775 * burning. Giving priority for a potential lock-holder increases lock
1778 * Since algorithm is based on heuristics, accessing another VCPU data without
1779 * locking does not harm. It may result in trying to yield to same VCPU, fail
1780 * and continue with next VCPU and so on.
1782 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1784 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1787 eligible = !vcpu->spin_loop.in_spin_loop ||
1788 vcpu->spin_loop.dy_eligible;
1790 if (vcpu->spin_loop.in_spin_loop)
1791 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1799 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1801 struct kvm *kvm = me->kvm;
1802 struct kvm_vcpu *vcpu;
1803 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1809 kvm_vcpu_set_in_spin_loop(me, true);
1811 * We boost the priority of a VCPU that is runnable but not
1812 * currently running, because it got preempted by something
1813 * else and called schedule in __vcpu_run. Hopefully that
1814 * VCPU is holding the lock that we need and will release it.
1815 * We approximate round-robin by starting at the last boosted VCPU.
1817 for (pass = 0; pass < 2 && !yielded && try; pass++) {
1818 kvm_for_each_vcpu(i, vcpu, kvm) {
1819 if (!pass && i <= last_boosted_vcpu) {
1820 i = last_boosted_vcpu;
1822 } else if (pass && i > last_boosted_vcpu)
1824 if (!ACCESS_ONCE(vcpu->preempted))
1828 if (waitqueue_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
1830 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1833 yielded = kvm_vcpu_yield_to(vcpu);
1835 kvm->last_boosted_vcpu = i;
1837 } else if (yielded < 0) {
1844 kvm_vcpu_set_in_spin_loop(me, false);
1846 /* Ensure vcpu is not eligible during next spinloop */
1847 kvm_vcpu_set_dy_eligible(me, false);
1849 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1851 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1853 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1856 if (vmf->pgoff == 0)
1857 page = virt_to_page(vcpu->run);
1859 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1860 page = virt_to_page(vcpu->arch.pio_data);
1862 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1863 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1864 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1867 return kvm_arch_vcpu_fault(vcpu, vmf);
1873 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1874 .fault = kvm_vcpu_fault,
1877 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1879 vma->vm_ops = &kvm_vcpu_vm_ops;
1883 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1885 struct kvm_vcpu *vcpu = filp->private_data;
1887 kvm_put_kvm(vcpu->kvm);
1891 static struct file_operations kvm_vcpu_fops = {
1892 .release = kvm_vcpu_release,
1893 .unlocked_ioctl = kvm_vcpu_ioctl,
1894 #ifdef CONFIG_COMPAT
1895 .compat_ioctl = kvm_vcpu_compat_ioctl,
1897 .mmap = kvm_vcpu_mmap,
1898 .llseek = noop_llseek,
1902 * Allocates an inode for the vcpu.
1904 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1906 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
1910 * Creates some virtual cpus. Good luck creating more than one.
1912 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1915 struct kvm_vcpu *vcpu, *v;
1917 if (id >= KVM_MAX_VCPUS)
1920 vcpu = kvm_arch_vcpu_create(kvm, id);
1922 return PTR_ERR(vcpu);
1924 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1926 r = kvm_arch_vcpu_setup(vcpu);
1930 mutex_lock(&kvm->lock);
1931 if (!kvm_vcpu_compatible(vcpu)) {
1933 goto unlock_vcpu_destroy;
1935 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1937 goto unlock_vcpu_destroy;
1940 kvm_for_each_vcpu(r, v, kvm)
1941 if (v->vcpu_id == id) {
1943 goto unlock_vcpu_destroy;
1946 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1948 /* Now it's all set up, let userspace reach it */
1950 r = create_vcpu_fd(vcpu);
1953 goto unlock_vcpu_destroy;
1956 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1958 atomic_inc(&kvm->online_vcpus);
1960 mutex_unlock(&kvm->lock);
1961 kvm_arch_vcpu_postcreate(vcpu);
1964 unlock_vcpu_destroy:
1965 mutex_unlock(&kvm->lock);
1967 kvm_arch_vcpu_destroy(vcpu);
1971 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1974 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1975 vcpu->sigset_active = 1;
1976 vcpu->sigset = *sigset;
1978 vcpu->sigset_active = 0;
1982 static long kvm_vcpu_ioctl(struct file *filp,
1983 unsigned int ioctl, unsigned long arg)
1985 struct kvm_vcpu *vcpu = filp->private_data;
1986 void __user *argp = (void __user *)arg;
1988 struct kvm_fpu *fpu = NULL;
1989 struct kvm_sregs *kvm_sregs = NULL;
1991 if (vcpu->kvm->mm != current->mm)
1994 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
1996 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1997 * so vcpu_load() would break it.
1999 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
2000 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2004 r = vcpu_load(vcpu);
2012 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2013 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2015 case KVM_GET_REGS: {
2016 struct kvm_regs *kvm_regs;
2019 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2022 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2026 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2033 case KVM_SET_REGS: {
2034 struct kvm_regs *kvm_regs;
2037 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2038 if (IS_ERR(kvm_regs)) {
2039 r = PTR_ERR(kvm_regs);
2042 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2046 case KVM_GET_SREGS: {
2047 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2051 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2055 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2060 case KVM_SET_SREGS: {
2061 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2062 if (IS_ERR(kvm_sregs)) {
2063 r = PTR_ERR(kvm_sregs);
2067 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2070 case KVM_GET_MP_STATE: {
2071 struct kvm_mp_state mp_state;
2073 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2077 if (copy_to_user(argp, &mp_state, sizeof mp_state))
2082 case KVM_SET_MP_STATE: {
2083 struct kvm_mp_state mp_state;
2086 if (copy_from_user(&mp_state, argp, sizeof mp_state))
2088 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2091 case KVM_TRANSLATE: {
2092 struct kvm_translation tr;
2095 if (copy_from_user(&tr, argp, sizeof tr))
2097 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2101 if (copy_to_user(argp, &tr, sizeof tr))
2106 case KVM_SET_GUEST_DEBUG: {
2107 struct kvm_guest_debug dbg;
2110 if (copy_from_user(&dbg, argp, sizeof dbg))
2112 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2115 case KVM_SET_SIGNAL_MASK: {
2116 struct kvm_signal_mask __user *sigmask_arg = argp;
2117 struct kvm_signal_mask kvm_sigmask;
2118 sigset_t sigset, *p;
2123 if (copy_from_user(&kvm_sigmask, argp,
2124 sizeof kvm_sigmask))
2127 if (kvm_sigmask.len != sizeof sigset)
2130 if (copy_from_user(&sigset, sigmask_arg->sigset,
2135 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2139 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2143 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2147 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2153 fpu = memdup_user(argp, sizeof(*fpu));
2159 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2163 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2172 #ifdef CONFIG_COMPAT
2173 static long kvm_vcpu_compat_ioctl(struct file *filp,
2174 unsigned int ioctl, unsigned long arg)
2176 struct kvm_vcpu *vcpu = filp->private_data;
2177 void __user *argp = compat_ptr(arg);
2180 if (vcpu->kvm->mm != current->mm)
2184 case KVM_SET_SIGNAL_MASK: {
2185 struct kvm_signal_mask __user *sigmask_arg = argp;
2186 struct kvm_signal_mask kvm_sigmask;
2187 compat_sigset_t csigset;
2192 if (copy_from_user(&kvm_sigmask, argp,
2193 sizeof kvm_sigmask))
2196 if (kvm_sigmask.len != sizeof csigset)
2199 if (copy_from_user(&csigset, sigmask_arg->sigset,
2202 sigset_from_compat(&sigset, &csigset);
2203 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2205 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2209 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2217 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2218 int (*accessor)(struct kvm_device *dev,
2219 struct kvm_device_attr *attr),
2222 struct kvm_device_attr attr;
2227 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2230 return accessor(dev, &attr);
2233 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2236 struct kvm_device *dev = filp->private_data;
2239 case KVM_SET_DEVICE_ATTR:
2240 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2241 case KVM_GET_DEVICE_ATTR:
2242 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2243 case KVM_HAS_DEVICE_ATTR:
2244 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2246 if (dev->ops->ioctl)
2247 return dev->ops->ioctl(dev, ioctl, arg);
2253 static int kvm_device_release(struct inode *inode, struct file *filp)
2255 struct kvm_device *dev = filp->private_data;
2256 struct kvm *kvm = dev->kvm;
2262 static const struct file_operations kvm_device_fops = {
2263 .unlocked_ioctl = kvm_device_ioctl,
2264 #ifdef CONFIG_COMPAT
2265 .compat_ioctl = kvm_device_ioctl,
2267 .release = kvm_device_release,
2270 struct kvm_device *kvm_device_from_filp(struct file *filp)
2272 if (filp->f_op != &kvm_device_fops)
2275 return filp->private_data;
2278 static int kvm_ioctl_create_device(struct kvm *kvm,
2279 struct kvm_create_device *cd)
2281 struct kvm_device_ops *ops = NULL;
2282 struct kvm_device *dev;
2283 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2287 #ifdef CONFIG_KVM_MPIC
2288 case KVM_DEV_TYPE_FSL_MPIC_20:
2289 case KVM_DEV_TYPE_FSL_MPIC_42:
2290 ops = &kvm_mpic_ops;
2293 #ifdef CONFIG_KVM_XICS
2294 case KVM_DEV_TYPE_XICS:
2295 ops = &kvm_xics_ops;
2298 #ifdef CONFIG_KVM_VFIO
2299 case KVM_DEV_TYPE_VFIO:
2300 ops = &kvm_vfio_ops;
2303 #ifdef CONFIG_KVM_ARM_VGIC
2304 case KVM_DEV_TYPE_ARM_VGIC_V2:
2305 ops = &kvm_arm_vgic_v2_ops;
2309 case KVM_DEV_TYPE_FLIC:
2310 ops = &kvm_flic_ops;
2320 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2327 ret = ops->create(dev, cd->type);
2333 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2339 list_add(&dev->vm_node, &kvm->devices);
2345 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2348 case KVM_CAP_USER_MEMORY:
2349 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2350 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2351 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2352 case KVM_CAP_SET_BOOT_CPU_ID:
2354 case KVM_CAP_INTERNAL_ERROR_DATA:
2355 #ifdef CONFIG_HAVE_KVM_MSI
2356 case KVM_CAP_SIGNAL_MSI:
2358 #ifdef CONFIG_HAVE_KVM_IRQFD
2359 case KVM_CAP_IRQFD_RESAMPLE:
2361 case KVM_CAP_CHECK_EXTENSION_VM:
2363 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2364 case KVM_CAP_IRQ_ROUTING:
2365 return KVM_MAX_IRQ_ROUTES;
2370 return kvm_vm_ioctl_check_extension(kvm, arg);
2373 static long kvm_vm_ioctl(struct file *filp,
2374 unsigned int ioctl, unsigned long arg)
2376 struct kvm *kvm = filp->private_data;
2377 void __user *argp = (void __user *)arg;
2380 if (kvm->mm != current->mm)
2383 case KVM_CREATE_VCPU:
2384 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2386 case KVM_SET_USER_MEMORY_REGION: {
2387 struct kvm_userspace_memory_region kvm_userspace_mem;
2390 if (copy_from_user(&kvm_userspace_mem, argp,
2391 sizeof kvm_userspace_mem))
2394 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2397 case KVM_GET_DIRTY_LOG: {
2398 struct kvm_dirty_log log;
2401 if (copy_from_user(&log, argp, sizeof log))
2403 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2406 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2407 case KVM_REGISTER_COALESCED_MMIO: {
2408 struct kvm_coalesced_mmio_zone zone;
2410 if (copy_from_user(&zone, argp, sizeof zone))
2412 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2415 case KVM_UNREGISTER_COALESCED_MMIO: {
2416 struct kvm_coalesced_mmio_zone zone;
2418 if (copy_from_user(&zone, argp, sizeof zone))
2420 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2425 struct kvm_irqfd data;
2428 if (copy_from_user(&data, argp, sizeof data))
2430 r = kvm_irqfd(kvm, &data);
2433 case KVM_IOEVENTFD: {
2434 struct kvm_ioeventfd data;
2437 if (copy_from_user(&data, argp, sizeof data))
2439 r = kvm_ioeventfd(kvm, &data);
2442 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2443 case KVM_SET_BOOT_CPU_ID:
2445 mutex_lock(&kvm->lock);
2446 if (atomic_read(&kvm->online_vcpus) != 0)
2449 kvm->bsp_vcpu_id = arg;
2450 mutex_unlock(&kvm->lock);
2453 #ifdef CONFIG_HAVE_KVM_MSI
2454 case KVM_SIGNAL_MSI: {
2458 if (copy_from_user(&msi, argp, sizeof msi))
2460 r = kvm_send_userspace_msi(kvm, &msi);
2464 #ifdef __KVM_HAVE_IRQ_LINE
2465 case KVM_IRQ_LINE_STATUS:
2466 case KVM_IRQ_LINE: {
2467 struct kvm_irq_level irq_event;
2470 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2473 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
2474 ioctl == KVM_IRQ_LINE_STATUS);
2479 if (ioctl == KVM_IRQ_LINE_STATUS) {
2480 if (copy_to_user(argp, &irq_event, sizeof irq_event))
2488 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2489 case KVM_SET_GSI_ROUTING: {
2490 struct kvm_irq_routing routing;
2491 struct kvm_irq_routing __user *urouting;
2492 struct kvm_irq_routing_entry *entries;
2495 if (copy_from_user(&routing, argp, sizeof(routing)))
2498 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2503 entries = vmalloc(routing.nr * sizeof(*entries));
2508 if (copy_from_user(entries, urouting->entries,
2509 routing.nr * sizeof(*entries)))
2510 goto out_free_irq_routing;
2511 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2513 out_free_irq_routing:
2517 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
2518 case KVM_CREATE_DEVICE: {
2519 struct kvm_create_device cd;
2522 if (copy_from_user(&cd, argp, sizeof(cd)))
2525 r = kvm_ioctl_create_device(kvm, &cd);
2530 if (copy_to_user(argp, &cd, sizeof(cd)))
2536 case KVM_CHECK_EXTENSION:
2537 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
2540 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2542 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2548 #ifdef CONFIG_COMPAT
2549 struct compat_kvm_dirty_log {
2553 compat_uptr_t dirty_bitmap; /* one bit per page */
2558 static long kvm_vm_compat_ioctl(struct file *filp,
2559 unsigned int ioctl, unsigned long arg)
2561 struct kvm *kvm = filp->private_data;
2564 if (kvm->mm != current->mm)
2567 case KVM_GET_DIRTY_LOG: {
2568 struct compat_kvm_dirty_log compat_log;
2569 struct kvm_dirty_log log;
2572 if (copy_from_user(&compat_log, (void __user *)arg,
2573 sizeof(compat_log)))
2575 log.slot = compat_log.slot;
2576 log.padding1 = compat_log.padding1;
2577 log.padding2 = compat_log.padding2;
2578 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2580 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2584 r = kvm_vm_ioctl(filp, ioctl, arg);
2592 static struct file_operations kvm_vm_fops = {
2593 .release = kvm_vm_release,
2594 .unlocked_ioctl = kvm_vm_ioctl,
2595 #ifdef CONFIG_COMPAT
2596 .compat_ioctl = kvm_vm_compat_ioctl,
2598 .llseek = noop_llseek,
2601 static int kvm_dev_ioctl_create_vm(unsigned long type)
2606 kvm = kvm_create_vm(type);
2608 return PTR_ERR(kvm);
2609 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2610 r = kvm_coalesced_mmio_init(kvm);
2616 r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC);
2623 static long kvm_dev_ioctl(struct file *filp,
2624 unsigned int ioctl, unsigned long arg)
2629 case KVM_GET_API_VERSION:
2632 r = KVM_API_VERSION;
2635 r = kvm_dev_ioctl_create_vm(arg);
2637 case KVM_CHECK_EXTENSION:
2638 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
2640 case KVM_GET_VCPU_MMAP_SIZE:
2643 r = PAGE_SIZE; /* struct kvm_run */
2645 r += PAGE_SIZE; /* pio data page */
2647 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2648 r += PAGE_SIZE; /* coalesced mmio ring page */
2651 case KVM_TRACE_ENABLE:
2652 case KVM_TRACE_PAUSE:
2653 case KVM_TRACE_DISABLE:
2657 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2663 static struct file_operations kvm_chardev_ops = {
2664 .unlocked_ioctl = kvm_dev_ioctl,
2665 .compat_ioctl = kvm_dev_ioctl,
2666 .llseek = noop_llseek,
2669 static struct miscdevice kvm_dev = {
2675 static void hardware_enable_nolock(void *junk)
2677 int cpu = raw_smp_processor_id();
2680 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2683 cpumask_set_cpu(cpu, cpus_hardware_enabled);
2685 r = kvm_arch_hardware_enable();
2688 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2689 atomic_inc(&hardware_enable_failed);
2690 printk(KERN_INFO "kvm: enabling virtualization on "
2691 "CPU%d failed\n", cpu);
2695 static void hardware_enable(void)
2697 raw_spin_lock(&kvm_count_lock);
2698 if (kvm_usage_count)
2699 hardware_enable_nolock(NULL);
2700 raw_spin_unlock(&kvm_count_lock);
2703 static void hardware_disable_nolock(void *junk)
2705 int cpu = raw_smp_processor_id();
2707 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2709 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2710 kvm_arch_hardware_disable();
2713 static void hardware_disable(void)
2715 raw_spin_lock(&kvm_count_lock);
2716 if (kvm_usage_count)
2717 hardware_disable_nolock(NULL);
2718 raw_spin_unlock(&kvm_count_lock);
2721 static void hardware_disable_all_nolock(void)
2723 BUG_ON(!kvm_usage_count);
2726 if (!kvm_usage_count)
2727 on_each_cpu(hardware_disable_nolock, NULL, 1);
2730 static void hardware_disable_all(void)
2732 raw_spin_lock(&kvm_count_lock);
2733 hardware_disable_all_nolock();
2734 raw_spin_unlock(&kvm_count_lock);
2737 static int hardware_enable_all(void)
2741 raw_spin_lock(&kvm_count_lock);
2744 if (kvm_usage_count == 1) {
2745 atomic_set(&hardware_enable_failed, 0);
2746 on_each_cpu(hardware_enable_nolock, NULL, 1);
2748 if (atomic_read(&hardware_enable_failed)) {
2749 hardware_disable_all_nolock();
2754 raw_spin_unlock(&kvm_count_lock);
2759 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2764 val &= ~CPU_TASKS_FROZEN;
2767 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2772 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2780 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2784 * Some (well, at least mine) BIOSes hang on reboot if
2787 * And Intel TXT required VMX off for all cpu when system shutdown.
2789 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2790 kvm_rebooting = true;
2791 on_each_cpu(hardware_disable_nolock, NULL, 1);
2795 static struct notifier_block kvm_reboot_notifier = {
2796 .notifier_call = kvm_reboot,
2800 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2804 for (i = 0; i < bus->dev_count; i++) {
2805 struct kvm_io_device *pos = bus->range[i].dev;
2807 kvm_iodevice_destructor(pos);
2812 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
2813 const struct kvm_io_range *r2)
2815 if (r1->addr < r2->addr)
2817 if (r1->addr + r1->len > r2->addr + r2->len)
2822 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2824 return kvm_io_bus_cmp(p1, p2);
2827 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2828 gpa_t addr, int len)
2830 bus->range[bus->dev_count++] = (struct kvm_io_range) {
2836 sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2837 kvm_io_bus_sort_cmp, NULL);
2842 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2843 gpa_t addr, int len)
2845 struct kvm_io_range *range, key;
2848 key = (struct kvm_io_range) {
2853 range = bsearch(&key, bus->range, bus->dev_count,
2854 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2858 off = range - bus->range;
2860 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
2866 static int __kvm_io_bus_write(struct kvm_io_bus *bus,
2867 struct kvm_io_range *range, const void *val)
2871 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2875 while (idx < bus->dev_count &&
2876 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
2877 if (!kvm_iodevice_write(bus->range[idx].dev, range->addr,
2886 /* kvm_io_bus_write - called under kvm->slots_lock */
2887 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2888 int len, const void *val)
2890 struct kvm_io_bus *bus;
2891 struct kvm_io_range range;
2894 range = (struct kvm_io_range) {
2899 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2900 r = __kvm_io_bus_write(bus, &range, val);
2901 return r < 0 ? r : 0;
2904 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
2905 int kvm_io_bus_write_cookie(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2906 int len, const void *val, long cookie)
2908 struct kvm_io_bus *bus;
2909 struct kvm_io_range range;
2911 range = (struct kvm_io_range) {
2916 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2918 /* First try the device referenced by cookie. */
2919 if ((cookie >= 0) && (cookie < bus->dev_count) &&
2920 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
2921 if (!kvm_iodevice_write(bus->range[cookie].dev, addr, len,
2926 * cookie contained garbage; fall back to search and return the
2927 * correct cookie value.
2929 return __kvm_io_bus_write(bus, &range, val);
2932 static int __kvm_io_bus_read(struct kvm_io_bus *bus, struct kvm_io_range *range,
2937 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2941 while (idx < bus->dev_count &&
2942 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
2943 if (!kvm_iodevice_read(bus->range[idx].dev, range->addr,
2951 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
2953 /* kvm_io_bus_read - called under kvm->slots_lock */
2954 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2957 struct kvm_io_bus *bus;
2958 struct kvm_io_range range;
2961 range = (struct kvm_io_range) {
2966 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2967 r = __kvm_io_bus_read(bus, &range, val);
2968 return r < 0 ? r : 0;
2972 /* Caller must hold slots_lock. */
2973 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2974 int len, struct kvm_io_device *dev)
2976 struct kvm_io_bus *new_bus, *bus;
2978 bus = kvm->buses[bus_idx];
2979 /* exclude ioeventfd which is limited by maximum fd */
2980 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
2983 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
2984 sizeof(struct kvm_io_range)), GFP_KERNEL);
2987 memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
2988 sizeof(struct kvm_io_range)));
2989 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2990 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2991 synchronize_srcu_expedited(&kvm->srcu);
2997 /* Caller must hold slots_lock. */
2998 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2999 struct kvm_io_device *dev)
3002 struct kvm_io_bus *new_bus, *bus;
3004 bus = kvm->buses[bus_idx];
3006 for (i = 0; i < bus->dev_count; i++)
3007 if (bus->range[i].dev == dev) {
3015 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3016 sizeof(struct kvm_io_range)), GFP_KERNEL);
3020 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3021 new_bus->dev_count--;
3022 memcpy(new_bus->range + i, bus->range + i + 1,
3023 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3025 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3026 synchronize_srcu_expedited(&kvm->srcu);
3031 static struct notifier_block kvm_cpu_notifier = {
3032 .notifier_call = kvm_cpu_hotplug,
3035 static int vm_stat_get(void *_offset, u64 *val)
3037 unsigned offset = (long)_offset;
3041 spin_lock(&kvm_lock);
3042 list_for_each_entry(kvm, &vm_list, vm_list)
3043 *val += *(u32 *)((void *)kvm + offset);
3044 spin_unlock(&kvm_lock);
3048 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
3050 static int vcpu_stat_get(void *_offset, u64 *val)
3052 unsigned offset = (long)_offset;
3054 struct kvm_vcpu *vcpu;
3058 spin_lock(&kvm_lock);
3059 list_for_each_entry(kvm, &vm_list, vm_list)
3060 kvm_for_each_vcpu(i, vcpu, kvm)
3061 *val += *(u32 *)((void *)vcpu + offset);
3063 spin_unlock(&kvm_lock);
3067 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
3069 static const struct file_operations *stat_fops[] = {
3070 [KVM_STAT_VCPU] = &vcpu_stat_fops,
3071 [KVM_STAT_VM] = &vm_stat_fops,
3074 static int kvm_init_debug(void)
3077 struct kvm_stats_debugfs_item *p;
3079 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3080 if (kvm_debugfs_dir == NULL)
3083 for (p = debugfs_entries; p->name; ++p) {
3084 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
3085 (void *)(long)p->offset,
3086 stat_fops[p->kind]);
3087 if (p->dentry == NULL)
3094 debugfs_remove_recursive(kvm_debugfs_dir);
3099 static void kvm_exit_debug(void)
3101 struct kvm_stats_debugfs_item *p;
3103 for (p = debugfs_entries; p->name; ++p)
3104 debugfs_remove(p->dentry);
3105 debugfs_remove(kvm_debugfs_dir);
3108 static int kvm_suspend(void)
3110 if (kvm_usage_count)
3111 hardware_disable_nolock(NULL);
3115 static void kvm_resume(void)
3117 if (kvm_usage_count) {
3118 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3119 hardware_enable_nolock(NULL);
3123 static struct syscore_ops kvm_syscore_ops = {
3124 .suspend = kvm_suspend,
3125 .resume = kvm_resume,
3129 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3131 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3134 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3136 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3137 if (vcpu->preempted)
3138 vcpu->preempted = false;
3140 kvm_arch_sched_in(vcpu, cpu);
3142 kvm_arch_vcpu_load(vcpu, cpu);
3145 static void kvm_sched_out(struct preempt_notifier *pn,
3146 struct task_struct *next)
3148 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3150 if (current->state == TASK_RUNNING)
3151 vcpu->preempted = true;
3152 kvm_arch_vcpu_put(vcpu);
3155 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3156 struct module *module)
3161 r = kvm_arch_init(opaque);
3166 * kvm_arch_init makes sure there's at most one caller
3167 * for architectures that support multiple implementations,
3168 * like intel and amd on x86.
3169 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3170 * conflicts in case kvm is already setup for another implementation.
3172 r = kvm_irqfd_init();
3176 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
3181 r = kvm_arch_hardware_setup();
3185 for_each_online_cpu(cpu) {
3186 smp_call_function_single(cpu,
3187 kvm_arch_check_processor_compat,
3193 r = register_cpu_notifier(&kvm_cpu_notifier);
3196 register_reboot_notifier(&kvm_reboot_notifier);
3198 /* A kmem cache lets us meet the alignment requirements of fx_save. */
3200 vcpu_align = __alignof__(struct kvm_vcpu);
3201 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
3203 if (!kvm_vcpu_cache) {
3208 r = kvm_async_pf_init();
3212 kvm_chardev_ops.owner = module;
3213 kvm_vm_fops.owner = module;
3214 kvm_vcpu_fops.owner = module;
3216 r = misc_register(&kvm_dev);
3218 printk(KERN_ERR "kvm: misc device register failed\n");
3222 register_syscore_ops(&kvm_syscore_ops);
3224 kvm_preempt_ops.sched_in = kvm_sched_in;
3225 kvm_preempt_ops.sched_out = kvm_sched_out;
3227 r = kvm_init_debug();
3229 printk(KERN_ERR "kvm: create debugfs files failed\n");
3236 unregister_syscore_ops(&kvm_syscore_ops);
3237 misc_deregister(&kvm_dev);
3239 kvm_async_pf_deinit();
3241 kmem_cache_destroy(kvm_vcpu_cache);
3243 unregister_reboot_notifier(&kvm_reboot_notifier);
3244 unregister_cpu_notifier(&kvm_cpu_notifier);
3247 kvm_arch_hardware_unsetup();
3249 free_cpumask_var(cpus_hardware_enabled);
3257 EXPORT_SYMBOL_GPL(kvm_init);
3262 misc_deregister(&kvm_dev);
3263 kmem_cache_destroy(kvm_vcpu_cache);
3264 kvm_async_pf_deinit();
3265 unregister_syscore_ops(&kvm_syscore_ops);
3266 unregister_reboot_notifier(&kvm_reboot_notifier);
3267 unregister_cpu_notifier(&kvm_cpu_notifier);
3268 on_each_cpu(hardware_disable_nolock, NULL, 1);
3269 kvm_arch_hardware_unsetup();
3272 free_cpumask_var(cpus_hardware_enabled);
3274 EXPORT_SYMBOL_GPL(kvm_exit);