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);
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);
467 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
470 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
473 kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
475 goto out_err_no_srcu;
476 kvm_init_memslots_id(kvm);
477 if (init_srcu_struct(&kvm->srcu))
478 goto out_err_no_srcu;
479 if (init_srcu_struct(&kvm->irq_srcu))
480 goto out_err_no_irq_srcu;
481 for (i = 0; i < KVM_NR_BUSES; i++) {
482 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
488 spin_lock_init(&kvm->mmu_lock);
489 kvm->mm = current->mm;
490 atomic_inc(&kvm->mm->mm_count);
491 kvm_eventfd_init(kvm);
492 mutex_init(&kvm->lock);
493 mutex_init(&kvm->irq_lock);
494 mutex_init(&kvm->slots_lock);
495 atomic_set(&kvm->users_count, 1);
496 INIT_LIST_HEAD(&kvm->devices);
498 r = kvm_init_mmu_notifier(kvm);
502 spin_lock(&kvm_lock);
503 list_add(&kvm->vm_list, &vm_list);
504 spin_unlock(&kvm_lock);
509 cleanup_srcu_struct(&kvm->irq_srcu);
511 cleanup_srcu_struct(&kvm->srcu);
513 hardware_disable_all();
515 for (i = 0; i < KVM_NR_BUSES; i++)
516 kfree(kvm->buses[i]);
517 kfree(kvm->memslots);
518 kvm_arch_free_vm(kvm);
523 * Avoid using vmalloc for a small buffer.
524 * Should not be used when the size is statically known.
526 void *kvm_kvzalloc(unsigned long size)
528 if (size > PAGE_SIZE)
529 return vzalloc(size);
531 return kzalloc(size, GFP_KERNEL);
534 void kvm_kvfree(const void *addr)
536 if (is_vmalloc_addr(addr))
542 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
544 if (!memslot->dirty_bitmap)
547 kvm_kvfree(memslot->dirty_bitmap);
548 memslot->dirty_bitmap = NULL;
552 * Free any memory in @free but not in @dont.
554 static void kvm_free_physmem_slot(struct kvm *kvm, struct kvm_memory_slot *free,
555 struct kvm_memory_slot *dont)
557 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
558 kvm_destroy_dirty_bitmap(free);
560 kvm_arch_free_memslot(kvm, free, dont);
565 static void kvm_free_physmem(struct kvm *kvm)
567 struct kvm_memslots *slots = kvm->memslots;
568 struct kvm_memory_slot *memslot;
570 kvm_for_each_memslot(memslot, slots)
571 kvm_free_physmem_slot(kvm, memslot, NULL);
573 kfree(kvm->memslots);
576 static void kvm_destroy_devices(struct kvm *kvm)
578 struct list_head *node, *tmp;
580 list_for_each_safe(node, tmp, &kvm->devices) {
581 struct kvm_device *dev =
582 list_entry(node, struct kvm_device, vm_node);
585 dev->ops->destroy(dev);
589 static void kvm_destroy_vm(struct kvm *kvm)
592 struct mm_struct *mm = kvm->mm;
594 kvm_arch_sync_events(kvm);
595 spin_lock(&kvm_lock);
596 list_del(&kvm->vm_list);
597 spin_unlock(&kvm_lock);
598 kvm_free_irq_routing(kvm);
599 for (i = 0; i < KVM_NR_BUSES; i++)
600 kvm_io_bus_destroy(kvm->buses[i]);
601 kvm_coalesced_mmio_free(kvm);
602 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
603 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
605 kvm_arch_flush_shadow_all(kvm);
607 kvm_arch_destroy_vm(kvm);
608 kvm_destroy_devices(kvm);
609 kvm_free_physmem(kvm);
610 cleanup_srcu_struct(&kvm->irq_srcu);
611 cleanup_srcu_struct(&kvm->srcu);
612 kvm_arch_free_vm(kvm);
613 hardware_disable_all();
617 void kvm_get_kvm(struct kvm *kvm)
619 atomic_inc(&kvm->users_count);
621 EXPORT_SYMBOL_GPL(kvm_get_kvm);
623 void kvm_put_kvm(struct kvm *kvm)
625 if (atomic_dec_and_test(&kvm->users_count))
628 EXPORT_SYMBOL_GPL(kvm_put_kvm);
631 static int kvm_vm_release(struct inode *inode, struct file *filp)
633 struct kvm *kvm = filp->private_data;
635 kvm_irqfd_release(kvm);
642 * Allocation size is twice as large as the actual dirty bitmap size.
643 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
645 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
648 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
650 memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
651 if (!memslot->dirty_bitmap)
654 #endif /* !CONFIG_S390 */
658 static int cmp_memslot(const void *slot1, const void *slot2)
660 struct kvm_memory_slot *s1, *s2;
662 s1 = (struct kvm_memory_slot *)slot1;
663 s2 = (struct kvm_memory_slot *)slot2;
665 if (s1->npages < s2->npages)
667 if (s1->npages > s2->npages)
674 * Sort the memslots base on its size, so the larger slots
675 * will get better fit.
677 static void sort_memslots(struct kvm_memslots *slots)
681 sort(slots->memslots, KVM_MEM_SLOTS_NUM,
682 sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
684 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
685 slots->id_to_index[slots->memslots[i].id] = i;
688 static void update_memslots(struct kvm_memslots *slots,
689 struct kvm_memory_slot *new,
694 struct kvm_memory_slot *old = id_to_memslot(slots, id);
695 unsigned long npages = old->npages;
698 if (new->npages != npages)
699 sort_memslots(slots);
702 slots->generation = last_generation + 1;
705 static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
707 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
709 #ifdef KVM_CAP_READONLY_MEM
710 valid_flags |= KVM_MEM_READONLY;
713 if (mem->flags & ~valid_flags)
719 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
720 struct kvm_memslots *slots, struct kvm_memory_slot *new)
722 struct kvm_memslots *old_memslots = kvm->memslots;
724 update_memslots(slots, new, kvm->memslots->generation);
725 rcu_assign_pointer(kvm->memslots, slots);
726 synchronize_srcu_expedited(&kvm->srcu);
728 kvm_arch_memslots_updated(kvm);
734 * Allocate some memory and give it an address in the guest physical address
737 * Discontiguous memory is allowed, mostly for framebuffers.
739 * Must be called holding mmap_sem for write.
741 int __kvm_set_memory_region(struct kvm *kvm,
742 struct kvm_userspace_memory_region *mem)
746 unsigned long npages;
747 struct kvm_memory_slot *slot;
748 struct kvm_memory_slot old, new;
749 struct kvm_memslots *slots = NULL, *old_memslots;
750 enum kvm_mr_change change;
752 r = check_memory_region_flags(mem);
757 /* General sanity checks */
758 if (mem->memory_size & (PAGE_SIZE - 1))
760 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
762 /* We can read the guest memory with __xxx_user() later on. */
763 if ((mem->slot < KVM_USER_MEM_SLOTS) &&
764 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
765 !access_ok(VERIFY_WRITE,
766 (void __user *)(unsigned long)mem->userspace_addr,
769 if (mem->slot >= KVM_MEM_SLOTS_NUM)
771 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
774 slot = id_to_memslot(kvm->memslots, mem->slot);
775 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
776 npages = mem->memory_size >> PAGE_SHIFT;
779 if (npages > KVM_MEM_MAX_NR_PAGES)
783 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
788 new.base_gfn = base_gfn;
790 new.flags = mem->flags;
795 change = KVM_MR_CREATE;
796 else { /* Modify an existing slot. */
797 if ((mem->userspace_addr != old.userspace_addr) ||
798 (npages != old.npages) ||
799 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
802 if (base_gfn != old.base_gfn)
803 change = KVM_MR_MOVE;
804 else if (new.flags != old.flags)
805 change = KVM_MR_FLAGS_ONLY;
806 else { /* Nothing to change. */
811 } else if (old.npages) {
812 change = KVM_MR_DELETE;
813 } else /* Modify a non-existent slot: disallowed. */
816 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
817 /* Check for overlaps */
819 kvm_for_each_memslot(slot, kvm->memslots) {
820 if ((slot->id >= KVM_USER_MEM_SLOTS) ||
821 (slot->id == mem->slot))
823 if (!((base_gfn + npages <= slot->base_gfn) ||
824 (base_gfn >= slot->base_gfn + slot->npages)))
829 /* Free page dirty bitmap if unneeded */
830 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
831 new.dirty_bitmap = NULL;
834 if (change == KVM_MR_CREATE) {
835 new.userspace_addr = mem->userspace_addr;
837 if (kvm_arch_create_memslot(kvm, &new, npages))
841 /* Allocate page dirty bitmap if needed */
842 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
843 if (kvm_create_dirty_bitmap(&new) < 0)
847 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
849 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
853 slot = id_to_memslot(slots, mem->slot);
854 slot->flags |= KVM_MEMSLOT_INVALID;
856 old_memslots = install_new_memslots(kvm, slots, NULL);
858 /* slot was deleted or moved, clear iommu mapping */
859 kvm_iommu_unmap_pages(kvm, &old);
860 /* From this point no new shadow pages pointing to a deleted,
861 * or moved, memslot will be created.
863 * validation of sp->gfn happens in:
864 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
865 * - kvm_is_visible_gfn (mmu_check_roots)
867 kvm_arch_flush_shadow_memslot(kvm, slot);
868 slots = old_memslots;
871 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
877 * We can re-use the old_memslots from above, the only difference
878 * from the currently installed memslots is the invalid flag. This
879 * will get overwritten by update_memslots anyway.
882 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
888 /* actual memory is freed via old in kvm_free_physmem_slot below */
889 if (change == KVM_MR_DELETE) {
890 new.dirty_bitmap = NULL;
891 memset(&new.arch, 0, sizeof(new.arch));
894 old_memslots = install_new_memslots(kvm, slots, &new);
896 kvm_arch_commit_memory_region(kvm, mem, &old, change);
898 kvm_free_physmem_slot(kvm, &old, &new);
902 * IOMMU mapping: New slots need to be mapped. Old slots need to be
903 * un-mapped and re-mapped if their base changes. Since base change
904 * unmapping is handled above with slot deletion, mapping alone is
905 * needed here. Anything else the iommu might care about for existing
906 * slots (size changes, userspace addr changes and read-only flag
907 * changes) is disallowed above, so any other attribute changes getting
908 * here can be skipped.
910 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
911 r = kvm_iommu_map_pages(kvm, &new);
920 kvm_free_physmem_slot(kvm, &new, &old);
924 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
926 int kvm_set_memory_region(struct kvm *kvm,
927 struct kvm_userspace_memory_region *mem)
931 mutex_lock(&kvm->slots_lock);
932 r = __kvm_set_memory_region(kvm, mem);
933 mutex_unlock(&kvm->slots_lock);
936 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
938 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
939 struct kvm_userspace_memory_region *mem)
941 if (mem->slot >= KVM_USER_MEM_SLOTS)
943 return kvm_set_memory_region(kvm, mem);
946 int kvm_get_dirty_log(struct kvm *kvm,
947 struct kvm_dirty_log *log, int *is_dirty)
949 struct kvm_memory_slot *memslot;
952 unsigned long any = 0;
955 if (log->slot >= KVM_USER_MEM_SLOTS)
958 memslot = id_to_memslot(kvm->memslots, log->slot);
960 if (!memslot->dirty_bitmap)
963 n = kvm_dirty_bitmap_bytes(memslot);
965 for (i = 0; !any && i < n/sizeof(long); ++i)
966 any = memslot->dirty_bitmap[i];
969 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
980 bool kvm_largepages_enabled(void)
982 return largepages_enabled;
985 void kvm_disable_largepages(void)
987 largepages_enabled = false;
989 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
991 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
993 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
995 EXPORT_SYMBOL_GPL(gfn_to_memslot);
997 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
999 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1001 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1002 memslot->flags & KVM_MEMSLOT_INVALID)
1007 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1009 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1011 struct vm_area_struct *vma;
1012 unsigned long addr, size;
1016 addr = gfn_to_hva(kvm, gfn);
1017 if (kvm_is_error_hva(addr))
1020 down_read(¤t->mm->mmap_sem);
1021 vma = find_vma(current->mm, addr);
1025 size = vma_kernel_pagesize(vma);
1028 up_read(¤t->mm->mmap_sem);
1033 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1035 return slot->flags & KVM_MEM_READONLY;
1038 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1039 gfn_t *nr_pages, bool write)
1041 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1042 return KVM_HVA_ERR_BAD;
1044 if (memslot_is_readonly(slot) && write)
1045 return KVM_HVA_ERR_RO_BAD;
1048 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1050 return __gfn_to_hva_memslot(slot, gfn);
1053 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1056 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1059 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1062 return gfn_to_hva_many(slot, gfn, NULL);
1064 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1066 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1068 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1070 EXPORT_SYMBOL_GPL(gfn_to_hva);
1073 * If writable is set to false, the hva returned by this function is only
1074 * allowed to be read.
1076 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1078 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1079 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1081 if (!kvm_is_error_hva(hva) && writable)
1082 *writable = !memslot_is_readonly(slot);
1087 static int kvm_read_hva(void *data, void __user *hva, int len)
1089 return __copy_from_user(data, hva, len);
1092 static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1094 return __copy_from_user_inatomic(data, hva, len);
1097 static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1098 unsigned long start, int write, struct page **page)
1100 int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1103 flags |= FOLL_WRITE;
1105 return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1108 static inline int check_user_page_hwpoison(unsigned long addr)
1110 int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1112 rc = __get_user_pages(current, current->mm, addr, 1,
1113 flags, NULL, NULL, NULL);
1114 return rc == -EHWPOISON;
1118 * The atomic path to get the writable pfn which will be stored in @pfn,
1119 * true indicates success, otherwise false is returned.
1121 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1122 bool write_fault, bool *writable, pfn_t *pfn)
1124 struct page *page[1];
1127 if (!(async || atomic))
1131 * Fast pin a writable pfn only if it is a write fault request
1132 * or the caller allows to map a writable pfn for a read fault
1135 if (!(write_fault || writable))
1138 npages = __get_user_pages_fast(addr, 1, 1, page);
1140 *pfn = page_to_pfn(page[0]);
1151 * The slow path to get the pfn of the specified host virtual address,
1152 * 1 indicates success, -errno is returned if error is detected.
1154 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1155 bool *writable, pfn_t *pfn)
1157 struct page *page[1];
1163 *writable = write_fault;
1166 down_read(¤t->mm->mmap_sem);
1167 npages = get_user_page_nowait(current, current->mm,
1168 addr, write_fault, page);
1169 up_read(¤t->mm->mmap_sem);
1171 npages = get_user_pages_fast(addr, 1, write_fault,
1176 /* map read fault as writable if possible */
1177 if (unlikely(!write_fault) && writable) {
1178 struct page *wpage[1];
1180 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1189 *pfn = page_to_pfn(page[0]);
1193 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1195 if (unlikely(!(vma->vm_flags & VM_READ)))
1198 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1205 * Pin guest page in memory and return its pfn.
1206 * @addr: host virtual address which maps memory to the guest
1207 * @atomic: whether this function can sleep
1208 * @async: whether this function need to wait IO complete if the
1209 * host page is not in the memory
1210 * @write_fault: whether we should get a writable host page
1211 * @writable: whether it allows to map a writable host page for !@write_fault
1213 * The function will map a writable host page for these two cases:
1214 * 1): @write_fault = true
1215 * 2): @write_fault = false && @writable, @writable will tell the caller
1216 * whether the mapping is writable.
1218 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1219 bool write_fault, bool *writable)
1221 struct vm_area_struct *vma;
1225 /* we can do it either atomically or asynchronously, not both */
1226 BUG_ON(atomic && async);
1228 if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1232 return KVM_PFN_ERR_FAULT;
1234 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1238 down_read(¤t->mm->mmap_sem);
1239 if (npages == -EHWPOISON ||
1240 (!async && check_user_page_hwpoison(addr))) {
1241 pfn = KVM_PFN_ERR_HWPOISON;
1245 vma = find_vma_intersection(current->mm, addr, addr + 1);
1248 pfn = KVM_PFN_ERR_FAULT;
1249 else if ((vma->vm_flags & VM_PFNMAP)) {
1250 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1252 BUG_ON(!kvm_is_mmio_pfn(pfn));
1254 if (async && vma_is_valid(vma, write_fault))
1256 pfn = KVM_PFN_ERR_FAULT;
1259 up_read(¤t->mm->mmap_sem);
1264 __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1265 bool *async, bool write_fault, bool *writable)
1267 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1269 if (addr == KVM_HVA_ERR_RO_BAD)
1270 return KVM_PFN_ERR_RO_FAULT;
1272 if (kvm_is_error_hva(addr))
1273 return KVM_PFN_NOSLOT;
1275 /* Do not map writable pfn in the readonly memslot. */
1276 if (writable && memslot_is_readonly(slot)) {
1281 return hva_to_pfn(addr, atomic, async, write_fault,
1285 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1286 bool write_fault, bool *writable)
1288 struct kvm_memory_slot *slot;
1293 slot = gfn_to_memslot(kvm, gfn);
1295 return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1299 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1301 return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1303 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1305 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1306 bool write_fault, bool *writable)
1308 return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1310 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1312 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1314 return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1316 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1318 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1321 return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1323 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1325 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1327 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1330 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1332 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1334 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1336 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1342 addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1343 if (kvm_is_error_hva(addr))
1346 if (entry < nr_pages)
1349 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1351 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1353 static struct page *kvm_pfn_to_page(pfn_t pfn)
1355 if (is_error_noslot_pfn(pfn))
1356 return KVM_ERR_PTR_BAD_PAGE;
1358 if (kvm_is_mmio_pfn(pfn)) {
1360 return KVM_ERR_PTR_BAD_PAGE;
1363 return pfn_to_page(pfn);
1366 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1370 pfn = gfn_to_pfn(kvm, gfn);
1372 return kvm_pfn_to_page(pfn);
1375 EXPORT_SYMBOL_GPL(gfn_to_page);
1377 void kvm_release_page_clean(struct page *page)
1379 WARN_ON(is_error_page(page));
1381 kvm_release_pfn_clean(page_to_pfn(page));
1383 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1385 void kvm_release_pfn_clean(pfn_t pfn)
1387 if (!is_error_noslot_pfn(pfn) && !kvm_is_mmio_pfn(pfn))
1388 put_page(pfn_to_page(pfn));
1390 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1392 void kvm_release_page_dirty(struct page *page)
1394 WARN_ON(is_error_page(page));
1396 kvm_release_pfn_dirty(page_to_pfn(page));
1398 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1400 static void kvm_release_pfn_dirty(pfn_t pfn)
1402 kvm_set_pfn_dirty(pfn);
1403 kvm_release_pfn_clean(pfn);
1406 void kvm_set_pfn_dirty(pfn_t pfn)
1408 if (!kvm_is_mmio_pfn(pfn)) {
1409 struct page *page = pfn_to_page(pfn);
1410 if (!PageReserved(page))
1414 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1416 void kvm_set_pfn_accessed(pfn_t pfn)
1418 if (!kvm_is_mmio_pfn(pfn))
1419 mark_page_accessed(pfn_to_page(pfn));
1421 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1423 void kvm_get_pfn(pfn_t pfn)
1425 if (!kvm_is_mmio_pfn(pfn))
1426 get_page(pfn_to_page(pfn));
1428 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1430 static int next_segment(unsigned long len, int offset)
1432 if (len > PAGE_SIZE - offset)
1433 return PAGE_SIZE - offset;
1438 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1444 addr = gfn_to_hva_prot(kvm, gfn, NULL);
1445 if (kvm_is_error_hva(addr))
1447 r = kvm_read_hva(data, (void __user *)addr + offset, len);
1452 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1454 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1456 gfn_t gfn = gpa >> PAGE_SHIFT;
1458 int offset = offset_in_page(gpa);
1461 while ((seg = next_segment(len, offset)) != 0) {
1462 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1472 EXPORT_SYMBOL_GPL(kvm_read_guest);
1474 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1479 gfn_t gfn = gpa >> PAGE_SHIFT;
1480 int offset = offset_in_page(gpa);
1482 addr = gfn_to_hva_prot(kvm, gfn, NULL);
1483 if (kvm_is_error_hva(addr))
1485 pagefault_disable();
1486 r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
1492 EXPORT_SYMBOL(kvm_read_guest_atomic);
1494 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1495 int offset, int len)
1500 addr = gfn_to_hva(kvm, gfn);
1501 if (kvm_is_error_hva(addr))
1503 r = __copy_to_user((void __user *)addr + offset, data, len);
1506 mark_page_dirty(kvm, gfn);
1509 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1511 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1514 gfn_t gfn = gpa >> PAGE_SHIFT;
1516 int offset = offset_in_page(gpa);
1519 while ((seg = next_segment(len, offset)) != 0) {
1520 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1531 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1532 gpa_t gpa, unsigned long len)
1534 struct kvm_memslots *slots = kvm_memslots(kvm);
1535 int offset = offset_in_page(gpa);
1536 gfn_t start_gfn = gpa >> PAGE_SHIFT;
1537 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1538 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1539 gfn_t nr_pages_avail;
1542 ghc->generation = slots->generation;
1544 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1545 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, &nr_pages_avail);
1546 if (!kvm_is_error_hva(ghc->hva) && nr_pages_avail >= nr_pages_needed) {
1550 * If the requested region crosses two memslots, we still
1551 * verify that the entire region is valid here.
1553 while (start_gfn <= end_gfn) {
1554 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1555 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1557 if (kvm_is_error_hva(ghc->hva))
1559 start_gfn += nr_pages_avail;
1561 /* Use the slow path for cross page reads and writes. */
1562 ghc->memslot = NULL;
1566 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1568 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1569 void *data, unsigned long len)
1571 struct kvm_memslots *slots = kvm_memslots(kvm);
1574 BUG_ON(len > ghc->len);
1576 if (slots->generation != ghc->generation)
1577 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1579 if (unlikely(!ghc->memslot))
1580 return kvm_write_guest(kvm, ghc->gpa, data, len);
1582 if (kvm_is_error_hva(ghc->hva))
1585 r = __copy_to_user((void __user *)ghc->hva, data, len);
1588 mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1592 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1594 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1595 void *data, unsigned long len)
1597 struct kvm_memslots *slots = kvm_memslots(kvm);
1600 BUG_ON(len > ghc->len);
1602 if (slots->generation != ghc->generation)
1603 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1605 if (unlikely(!ghc->memslot))
1606 return kvm_read_guest(kvm, ghc->gpa, data, len);
1608 if (kvm_is_error_hva(ghc->hva))
1611 r = __copy_from_user(data, (void __user *)ghc->hva, len);
1617 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1619 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1621 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
1623 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
1625 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1627 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1629 gfn_t gfn = gpa >> PAGE_SHIFT;
1631 int offset = offset_in_page(gpa);
1634 while ((seg = next_segment(len, offset)) != 0) {
1635 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1644 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1646 static void mark_page_dirty_in_slot(struct kvm *kvm,
1647 struct kvm_memory_slot *memslot,
1650 if (memslot && memslot->dirty_bitmap) {
1651 unsigned long rel_gfn = gfn - memslot->base_gfn;
1653 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1657 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1659 struct kvm_memory_slot *memslot;
1661 memslot = gfn_to_memslot(kvm, gfn);
1662 mark_page_dirty_in_slot(kvm, memslot, gfn);
1666 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1668 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1673 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1675 if (kvm_arch_vcpu_runnable(vcpu)) {
1676 kvm_make_request(KVM_REQ_UNHALT, vcpu);
1679 if (kvm_cpu_has_pending_timer(vcpu))
1681 if (signal_pending(current))
1687 finish_wait(&vcpu->wq, &wait);
1692 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1694 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1697 int cpu = vcpu->cpu;
1698 wait_queue_head_t *wqp;
1700 wqp = kvm_arch_vcpu_wq(vcpu);
1701 if (waitqueue_active(wqp)) {
1702 wake_up_interruptible(wqp);
1703 ++vcpu->stat.halt_wakeup;
1707 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1708 if (kvm_arch_vcpu_should_kick(vcpu))
1709 smp_send_reschedule(cpu);
1712 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
1713 #endif /* !CONFIG_S390 */
1715 bool kvm_vcpu_yield_to(struct kvm_vcpu *target)
1718 struct task_struct *task = NULL;
1722 pid = rcu_dereference(target->pid);
1724 task = get_pid_task(target->pid, PIDTYPE_PID);
1728 if (task->flags & PF_VCPU) {
1729 put_task_struct(task);
1732 ret = yield_to(task, 1);
1733 put_task_struct(task);
1737 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1740 * Helper that checks whether a VCPU is eligible for directed yield.
1741 * Most eligible candidate to yield is decided by following heuristics:
1743 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1744 * (preempted lock holder), indicated by @in_spin_loop.
1745 * Set at the beiginning and cleared at the end of interception/PLE handler.
1747 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1748 * chance last time (mostly it has become eligible now since we have probably
1749 * yielded to lockholder in last iteration. This is done by toggling
1750 * @dy_eligible each time a VCPU checked for eligibility.)
1752 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1753 * to preempted lock-holder could result in wrong VCPU selection and CPU
1754 * burning. Giving priority for a potential lock-holder increases lock
1757 * Since algorithm is based on heuristics, accessing another VCPU data without
1758 * locking does not harm. It may result in trying to yield to same VCPU, fail
1759 * and continue with next VCPU and so on.
1761 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1763 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1766 eligible = !vcpu->spin_loop.in_spin_loop ||
1767 (vcpu->spin_loop.in_spin_loop &&
1768 vcpu->spin_loop.dy_eligible);
1770 if (vcpu->spin_loop.in_spin_loop)
1771 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1779 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1781 struct kvm *kvm = me->kvm;
1782 struct kvm_vcpu *vcpu;
1783 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1789 kvm_vcpu_set_in_spin_loop(me, true);
1791 * We boost the priority of a VCPU that is runnable but not
1792 * currently running, because it got preempted by something
1793 * else and called schedule in __vcpu_run. Hopefully that
1794 * VCPU is holding the lock that we need and will release it.
1795 * We approximate round-robin by starting at the last boosted VCPU.
1797 for (pass = 0; pass < 2 && !yielded && try; pass++) {
1798 kvm_for_each_vcpu(i, vcpu, kvm) {
1799 if (!pass && i <= last_boosted_vcpu) {
1800 i = last_boosted_vcpu;
1802 } else if (pass && i > last_boosted_vcpu)
1804 if (!ACCESS_ONCE(vcpu->preempted))
1808 if (waitqueue_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
1810 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1813 yielded = kvm_vcpu_yield_to(vcpu);
1815 kvm->last_boosted_vcpu = i;
1817 } else if (yielded < 0) {
1824 kvm_vcpu_set_in_spin_loop(me, false);
1826 /* Ensure vcpu is not eligible during next spinloop */
1827 kvm_vcpu_set_dy_eligible(me, false);
1829 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1831 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1833 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1836 if (vmf->pgoff == 0)
1837 page = virt_to_page(vcpu->run);
1839 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1840 page = virt_to_page(vcpu->arch.pio_data);
1842 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1843 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1844 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1847 return kvm_arch_vcpu_fault(vcpu, vmf);
1853 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1854 .fault = kvm_vcpu_fault,
1857 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1859 vma->vm_ops = &kvm_vcpu_vm_ops;
1863 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1865 struct kvm_vcpu *vcpu = filp->private_data;
1867 kvm_put_kvm(vcpu->kvm);
1871 static struct file_operations kvm_vcpu_fops = {
1872 .release = kvm_vcpu_release,
1873 .unlocked_ioctl = kvm_vcpu_ioctl,
1874 #ifdef CONFIG_COMPAT
1875 .compat_ioctl = kvm_vcpu_compat_ioctl,
1877 .mmap = kvm_vcpu_mmap,
1878 .llseek = noop_llseek,
1882 * Allocates an inode for the vcpu.
1884 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1886 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
1890 * Creates some virtual cpus. Good luck creating more than one.
1892 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1895 struct kvm_vcpu *vcpu, *v;
1897 if (id >= KVM_MAX_VCPUS)
1900 vcpu = kvm_arch_vcpu_create(kvm, id);
1902 return PTR_ERR(vcpu);
1904 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1906 r = kvm_arch_vcpu_setup(vcpu);
1910 mutex_lock(&kvm->lock);
1911 if (!kvm_vcpu_compatible(vcpu)) {
1913 goto unlock_vcpu_destroy;
1915 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1917 goto unlock_vcpu_destroy;
1920 kvm_for_each_vcpu(r, v, kvm)
1921 if (v->vcpu_id == id) {
1923 goto unlock_vcpu_destroy;
1926 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1928 /* Now it's all set up, let userspace reach it */
1930 r = create_vcpu_fd(vcpu);
1933 goto unlock_vcpu_destroy;
1936 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1938 atomic_inc(&kvm->online_vcpus);
1940 mutex_unlock(&kvm->lock);
1941 kvm_arch_vcpu_postcreate(vcpu);
1944 unlock_vcpu_destroy:
1945 mutex_unlock(&kvm->lock);
1947 kvm_arch_vcpu_destroy(vcpu);
1951 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1954 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1955 vcpu->sigset_active = 1;
1956 vcpu->sigset = *sigset;
1958 vcpu->sigset_active = 0;
1962 static long kvm_vcpu_ioctl(struct file *filp,
1963 unsigned int ioctl, unsigned long arg)
1965 struct kvm_vcpu *vcpu = filp->private_data;
1966 void __user *argp = (void __user *)arg;
1968 struct kvm_fpu *fpu = NULL;
1969 struct kvm_sregs *kvm_sregs = NULL;
1971 if (vcpu->kvm->mm != current->mm)
1974 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
1976 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1977 * so vcpu_load() would break it.
1979 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1980 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1984 r = vcpu_load(vcpu);
1992 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1993 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1995 case KVM_GET_REGS: {
1996 struct kvm_regs *kvm_regs;
1999 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2002 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2006 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2013 case KVM_SET_REGS: {
2014 struct kvm_regs *kvm_regs;
2017 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2018 if (IS_ERR(kvm_regs)) {
2019 r = PTR_ERR(kvm_regs);
2022 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2026 case KVM_GET_SREGS: {
2027 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2031 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2035 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2040 case KVM_SET_SREGS: {
2041 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2042 if (IS_ERR(kvm_sregs)) {
2043 r = PTR_ERR(kvm_sregs);
2047 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2050 case KVM_GET_MP_STATE: {
2051 struct kvm_mp_state mp_state;
2053 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2057 if (copy_to_user(argp, &mp_state, sizeof mp_state))
2062 case KVM_SET_MP_STATE: {
2063 struct kvm_mp_state mp_state;
2066 if (copy_from_user(&mp_state, argp, sizeof mp_state))
2068 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2071 case KVM_TRANSLATE: {
2072 struct kvm_translation tr;
2075 if (copy_from_user(&tr, argp, sizeof tr))
2077 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2081 if (copy_to_user(argp, &tr, sizeof tr))
2086 case KVM_SET_GUEST_DEBUG: {
2087 struct kvm_guest_debug dbg;
2090 if (copy_from_user(&dbg, argp, sizeof dbg))
2092 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2095 case KVM_SET_SIGNAL_MASK: {
2096 struct kvm_signal_mask __user *sigmask_arg = argp;
2097 struct kvm_signal_mask kvm_sigmask;
2098 sigset_t sigset, *p;
2103 if (copy_from_user(&kvm_sigmask, argp,
2104 sizeof kvm_sigmask))
2107 if (kvm_sigmask.len != sizeof sigset)
2110 if (copy_from_user(&sigset, sigmask_arg->sigset,
2115 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2119 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2123 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2127 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2133 fpu = memdup_user(argp, sizeof(*fpu));
2139 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2143 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2152 #ifdef CONFIG_COMPAT
2153 static long kvm_vcpu_compat_ioctl(struct file *filp,
2154 unsigned int ioctl, unsigned long arg)
2156 struct kvm_vcpu *vcpu = filp->private_data;
2157 void __user *argp = compat_ptr(arg);
2160 if (vcpu->kvm->mm != current->mm)
2164 case KVM_SET_SIGNAL_MASK: {
2165 struct kvm_signal_mask __user *sigmask_arg = argp;
2166 struct kvm_signal_mask kvm_sigmask;
2167 compat_sigset_t csigset;
2172 if (copy_from_user(&kvm_sigmask, argp,
2173 sizeof kvm_sigmask))
2176 if (kvm_sigmask.len != sizeof csigset)
2179 if (copy_from_user(&csigset, sigmask_arg->sigset,
2182 sigset_from_compat(&sigset, &csigset);
2183 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2185 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2189 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2197 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2198 int (*accessor)(struct kvm_device *dev,
2199 struct kvm_device_attr *attr),
2202 struct kvm_device_attr attr;
2207 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2210 return accessor(dev, &attr);
2213 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2216 struct kvm_device *dev = filp->private_data;
2219 case KVM_SET_DEVICE_ATTR:
2220 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2221 case KVM_GET_DEVICE_ATTR:
2222 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2223 case KVM_HAS_DEVICE_ATTR:
2224 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2226 if (dev->ops->ioctl)
2227 return dev->ops->ioctl(dev, ioctl, arg);
2233 static int kvm_device_release(struct inode *inode, struct file *filp)
2235 struct kvm_device *dev = filp->private_data;
2236 struct kvm *kvm = dev->kvm;
2242 static const struct file_operations kvm_device_fops = {
2243 .unlocked_ioctl = kvm_device_ioctl,
2244 #ifdef CONFIG_COMPAT
2245 .compat_ioctl = kvm_device_ioctl,
2247 .release = kvm_device_release,
2250 struct kvm_device *kvm_device_from_filp(struct file *filp)
2252 if (filp->f_op != &kvm_device_fops)
2255 return filp->private_data;
2258 static int kvm_ioctl_create_device(struct kvm *kvm,
2259 struct kvm_create_device *cd)
2261 struct kvm_device_ops *ops = NULL;
2262 struct kvm_device *dev;
2263 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2267 #ifdef CONFIG_KVM_MPIC
2268 case KVM_DEV_TYPE_FSL_MPIC_20:
2269 case KVM_DEV_TYPE_FSL_MPIC_42:
2270 ops = &kvm_mpic_ops;
2273 #ifdef CONFIG_KVM_XICS
2274 case KVM_DEV_TYPE_XICS:
2275 ops = &kvm_xics_ops;
2278 #ifdef CONFIG_KVM_VFIO
2279 case KVM_DEV_TYPE_VFIO:
2280 ops = &kvm_vfio_ops;
2283 #ifdef CONFIG_KVM_ARM_VGIC
2284 case KVM_DEV_TYPE_ARM_VGIC_V2:
2285 ops = &kvm_arm_vgic_v2_ops;
2295 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2302 ret = ops->create(dev, cd->type);
2308 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2314 list_add(&dev->vm_node, &kvm->devices);
2320 static long kvm_vm_ioctl(struct file *filp,
2321 unsigned int ioctl, unsigned long arg)
2323 struct kvm *kvm = filp->private_data;
2324 void __user *argp = (void __user *)arg;
2327 if (kvm->mm != current->mm)
2330 case KVM_CREATE_VCPU:
2331 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2333 case KVM_SET_USER_MEMORY_REGION: {
2334 struct kvm_userspace_memory_region kvm_userspace_mem;
2337 if (copy_from_user(&kvm_userspace_mem, argp,
2338 sizeof kvm_userspace_mem))
2341 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2344 case KVM_GET_DIRTY_LOG: {
2345 struct kvm_dirty_log log;
2348 if (copy_from_user(&log, argp, sizeof log))
2350 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2353 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2354 case KVM_REGISTER_COALESCED_MMIO: {
2355 struct kvm_coalesced_mmio_zone zone;
2357 if (copy_from_user(&zone, argp, sizeof zone))
2359 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2362 case KVM_UNREGISTER_COALESCED_MMIO: {
2363 struct kvm_coalesced_mmio_zone zone;
2365 if (copy_from_user(&zone, argp, sizeof zone))
2367 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2372 struct kvm_irqfd data;
2375 if (copy_from_user(&data, argp, sizeof data))
2377 r = kvm_irqfd(kvm, &data);
2380 case KVM_IOEVENTFD: {
2381 struct kvm_ioeventfd data;
2384 if (copy_from_user(&data, argp, sizeof data))
2386 r = kvm_ioeventfd(kvm, &data);
2389 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2390 case KVM_SET_BOOT_CPU_ID:
2392 mutex_lock(&kvm->lock);
2393 if (atomic_read(&kvm->online_vcpus) != 0)
2396 kvm->bsp_vcpu_id = arg;
2397 mutex_unlock(&kvm->lock);
2400 #ifdef CONFIG_HAVE_KVM_MSI
2401 case KVM_SIGNAL_MSI: {
2405 if (copy_from_user(&msi, argp, sizeof msi))
2407 r = kvm_send_userspace_msi(kvm, &msi);
2411 #ifdef __KVM_HAVE_IRQ_LINE
2412 case KVM_IRQ_LINE_STATUS:
2413 case KVM_IRQ_LINE: {
2414 struct kvm_irq_level irq_event;
2417 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2420 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
2421 ioctl == KVM_IRQ_LINE_STATUS);
2426 if (ioctl == KVM_IRQ_LINE_STATUS) {
2427 if (copy_to_user(argp, &irq_event, sizeof irq_event))
2435 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2436 case KVM_SET_GSI_ROUTING: {
2437 struct kvm_irq_routing routing;
2438 struct kvm_irq_routing __user *urouting;
2439 struct kvm_irq_routing_entry *entries;
2442 if (copy_from_user(&routing, argp, sizeof(routing)))
2445 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2450 entries = vmalloc(routing.nr * sizeof(*entries));
2455 if (copy_from_user(entries, urouting->entries,
2456 routing.nr * sizeof(*entries)))
2457 goto out_free_irq_routing;
2458 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2460 out_free_irq_routing:
2464 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
2465 case KVM_CREATE_DEVICE: {
2466 struct kvm_create_device cd;
2469 if (copy_from_user(&cd, argp, sizeof(cd)))
2472 r = kvm_ioctl_create_device(kvm, &cd);
2477 if (copy_to_user(argp, &cd, sizeof(cd)))
2484 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2486 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2492 #ifdef CONFIG_COMPAT
2493 struct compat_kvm_dirty_log {
2497 compat_uptr_t dirty_bitmap; /* one bit per page */
2502 static long kvm_vm_compat_ioctl(struct file *filp,
2503 unsigned int ioctl, unsigned long arg)
2505 struct kvm *kvm = filp->private_data;
2508 if (kvm->mm != current->mm)
2511 case KVM_GET_DIRTY_LOG: {
2512 struct compat_kvm_dirty_log compat_log;
2513 struct kvm_dirty_log log;
2516 if (copy_from_user(&compat_log, (void __user *)arg,
2517 sizeof(compat_log)))
2519 log.slot = compat_log.slot;
2520 log.padding1 = compat_log.padding1;
2521 log.padding2 = compat_log.padding2;
2522 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2524 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2528 r = kvm_vm_ioctl(filp, ioctl, arg);
2536 static struct file_operations kvm_vm_fops = {
2537 .release = kvm_vm_release,
2538 .unlocked_ioctl = kvm_vm_ioctl,
2539 #ifdef CONFIG_COMPAT
2540 .compat_ioctl = kvm_vm_compat_ioctl,
2542 .llseek = noop_llseek,
2545 static int kvm_dev_ioctl_create_vm(unsigned long type)
2550 kvm = kvm_create_vm(type);
2552 return PTR_ERR(kvm);
2553 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2554 r = kvm_coalesced_mmio_init(kvm);
2560 r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC);
2567 static long kvm_dev_ioctl_check_extension_generic(long arg)
2570 case KVM_CAP_USER_MEMORY:
2571 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2572 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2573 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2574 case KVM_CAP_SET_BOOT_CPU_ID:
2576 case KVM_CAP_INTERNAL_ERROR_DATA:
2577 #ifdef CONFIG_HAVE_KVM_MSI
2578 case KVM_CAP_SIGNAL_MSI:
2580 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2581 case KVM_CAP_IRQFD_RESAMPLE:
2584 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2585 case KVM_CAP_IRQ_ROUTING:
2586 return KVM_MAX_IRQ_ROUTES;
2591 return kvm_dev_ioctl_check_extension(arg);
2594 static long kvm_dev_ioctl(struct file *filp,
2595 unsigned int ioctl, unsigned long arg)
2600 case KVM_GET_API_VERSION:
2604 r = KVM_API_VERSION;
2607 r = kvm_dev_ioctl_create_vm(arg);
2609 case KVM_CHECK_EXTENSION:
2610 r = kvm_dev_ioctl_check_extension_generic(arg);
2612 case KVM_GET_VCPU_MMAP_SIZE:
2616 r = PAGE_SIZE; /* struct kvm_run */
2618 r += PAGE_SIZE; /* pio data page */
2620 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2621 r += PAGE_SIZE; /* coalesced mmio ring page */
2624 case KVM_TRACE_ENABLE:
2625 case KVM_TRACE_PAUSE:
2626 case KVM_TRACE_DISABLE:
2630 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2636 static struct file_operations kvm_chardev_ops = {
2637 .unlocked_ioctl = kvm_dev_ioctl,
2638 .compat_ioctl = kvm_dev_ioctl,
2639 .llseek = noop_llseek,
2642 static struct miscdevice kvm_dev = {
2648 static void hardware_enable_nolock(void *junk)
2650 int cpu = raw_smp_processor_id();
2653 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2656 cpumask_set_cpu(cpu, cpus_hardware_enabled);
2658 r = kvm_arch_hardware_enable(NULL);
2661 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2662 atomic_inc(&hardware_enable_failed);
2663 printk(KERN_INFO "kvm: enabling virtualization on "
2664 "CPU%d failed\n", cpu);
2668 static void hardware_enable(void)
2670 raw_spin_lock(&kvm_count_lock);
2671 if (kvm_usage_count)
2672 hardware_enable_nolock(NULL);
2673 raw_spin_unlock(&kvm_count_lock);
2676 static void hardware_disable_nolock(void *junk)
2678 int cpu = raw_smp_processor_id();
2680 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2682 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2683 kvm_arch_hardware_disable(NULL);
2686 static void hardware_disable(void)
2688 raw_spin_lock(&kvm_count_lock);
2689 if (kvm_usage_count)
2690 hardware_disable_nolock(NULL);
2691 raw_spin_unlock(&kvm_count_lock);
2694 static void hardware_disable_all_nolock(void)
2696 BUG_ON(!kvm_usage_count);
2699 if (!kvm_usage_count)
2700 on_each_cpu(hardware_disable_nolock, NULL, 1);
2703 static void hardware_disable_all(void)
2705 raw_spin_lock(&kvm_count_lock);
2706 hardware_disable_all_nolock();
2707 raw_spin_unlock(&kvm_count_lock);
2710 static int hardware_enable_all(void)
2714 raw_spin_lock(&kvm_count_lock);
2717 if (kvm_usage_count == 1) {
2718 atomic_set(&hardware_enable_failed, 0);
2719 on_each_cpu(hardware_enable_nolock, NULL, 1);
2721 if (atomic_read(&hardware_enable_failed)) {
2722 hardware_disable_all_nolock();
2727 raw_spin_unlock(&kvm_count_lock);
2732 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2737 val &= ~CPU_TASKS_FROZEN;
2740 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2745 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2753 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2757 * Some (well, at least mine) BIOSes hang on reboot if
2760 * And Intel TXT required VMX off for all cpu when system shutdown.
2762 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2763 kvm_rebooting = true;
2764 on_each_cpu(hardware_disable_nolock, NULL, 1);
2768 static struct notifier_block kvm_reboot_notifier = {
2769 .notifier_call = kvm_reboot,
2773 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2777 for (i = 0; i < bus->dev_count; i++) {
2778 struct kvm_io_device *pos = bus->range[i].dev;
2780 kvm_iodevice_destructor(pos);
2785 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2787 const struct kvm_io_range *r1 = p1;
2788 const struct kvm_io_range *r2 = p2;
2790 if (r1->addr < r2->addr)
2792 if (r1->addr + r1->len > r2->addr + r2->len)
2797 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2798 gpa_t addr, int len)
2800 bus->range[bus->dev_count++] = (struct kvm_io_range) {
2806 sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2807 kvm_io_bus_sort_cmp, NULL);
2812 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2813 gpa_t addr, int len)
2815 struct kvm_io_range *range, key;
2818 key = (struct kvm_io_range) {
2823 range = bsearch(&key, bus->range, bus->dev_count,
2824 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2828 off = range - bus->range;
2830 while (off > 0 && kvm_io_bus_sort_cmp(&key, &bus->range[off-1]) == 0)
2836 /* kvm_io_bus_write - called under kvm->slots_lock */
2837 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2838 int len, const void *val)
2841 struct kvm_io_bus *bus;
2842 struct kvm_io_range range;
2844 range = (struct kvm_io_range) {
2849 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2850 idx = kvm_io_bus_get_first_dev(bus, addr, len);
2854 while (idx < bus->dev_count &&
2855 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2856 if (!kvm_iodevice_write(bus->range[idx].dev, addr, len, val))
2864 /* kvm_io_bus_read - called under kvm->slots_lock */
2865 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2869 struct kvm_io_bus *bus;
2870 struct kvm_io_range range;
2872 range = (struct kvm_io_range) {
2877 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2878 idx = kvm_io_bus_get_first_dev(bus, addr, len);
2882 while (idx < bus->dev_count &&
2883 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2884 if (!kvm_iodevice_read(bus->range[idx].dev, addr, len, val))
2892 /* Caller must hold slots_lock. */
2893 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2894 int len, struct kvm_io_device *dev)
2896 struct kvm_io_bus *new_bus, *bus;
2898 bus = kvm->buses[bus_idx];
2899 if (bus->dev_count > NR_IOBUS_DEVS - 1)
2902 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
2903 sizeof(struct kvm_io_range)), GFP_KERNEL);
2906 memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
2907 sizeof(struct kvm_io_range)));
2908 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2909 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2910 synchronize_srcu_expedited(&kvm->srcu);
2916 /* Caller must hold slots_lock. */
2917 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2918 struct kvm_io_device *dev)
2921 struct kvm_io_bus *new_bus, *bus;
2923 bus = kvm->buses[bus_idx];
2925 for (i = 0; i < bus->dev_count; i++)
2926 if (bus->range[i].dev == dev) {
2934 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
2935 sizeof(struct kvm_io_range)), GFP_KERNEL);
2939 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
2940 new_bus->dev_count--;
2941 memcpy(new_bus->range + i, bus->range + i + 1,
2942 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
2944 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2945 synchronize_srcu_expedited(&kvm->srcu);
2950 static struct notifier_block kvm_cpu_notifier = {
2951 .notifier_call = kvm_cpu_hotplug,
2954 static int vm_stat_get(void *_offset, u64 *val)
2956 unsigned offset = (long)_offset;
2960 spin_lock(&kvm_lock);
2961 list_for_each_entry(kvm, &vm_list, vm_list)
2962 *val += *(u32 *)((void *)kvm + offset);
2963 spin_unlock(&kvm_lock);
2967 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2969 static int vcpu_stat_get(void *_offset, u64 *val)
2971 unsigned offset = (long)_offset;
2973 struct kvm_vcpu *vcpu;
2977 spin_lock(&kvm_lock);
2978 list_for_each_entry(kvm, &vm_list, vm_list)
2979 kvm_for_each_vcpu(i, vcpu, kvm)
2980 *val += *(u32 *)((void *)vcpu + offset);
2982 spin_unlock(&kvm_lock);
2986 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2988 static const struct file_operations *stat_fops[] = {
2989 [KVM_STAT_VCPU] = &vcpu_stat_fops,
2990 [KVM_STAT_VM] = &vm_stat_fops,
2993 static int kvm_init_debug(void)
2996 struct kvm_stats_debugfs_item *p;
2998 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2999 if (kvm_debugfs_dir == NULL)
3002 for (p = debugfs_entries; p->name; ++p) {
3003 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
3004 (void *)(long)p->offset,
3005 stat_fops[p->kind]);
3006 if (p->dentry == NULL)
3013 debugfs_remove_recursive(kvm_debugfs_dir);
3018 static void kvm_exit_debug(void)
3020 struct kvm_stats_debugfs_item *p;
3022 for (p = debugfs_entries; p->name; ++p)
3023 debugfs_remove(p->dentry);
3024 debugfs_remove(kvm_debugfs_dir);
3027 static int kvm_suspend(void)
3029 if (kvm_usage_count)
3030 hardware_disable_nolock(NULL);
3034 static void kvm_resume(void)
3036 if (kvm_usage_count) {
3037 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3038 hardware_enable_nolock(NULL);
3042 static struct syscore_ops kvm_syscore_ops = {
3043 .suspend = kvm_suspend,
3044 .resume = kvm_resume,
3048 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3050 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3053 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3055 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3056 if (vcpu->preempted)
3057 vcpu->preempted = false;
3059 kvm_arch_vcpu_load(vcpu, cpu);
3062 static void kvm_sched_out(struct preempt_notifier *pn,
3063 struct task_struct *next)
3065 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3067 if (current->state == TASK_RUNNING)
3068 vcpu->preempted = true;
3069 kvm_arch_vcpu_put(vcpu);
3072 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3073 struct module *module)
3078 r = kvm_arch_init(opaque);
3083 * kvm_arch_init makes sure there's at most one caller
3084 * for architectures that support multiple implementations,
3085 * like intel and amd on x86.
3086 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3087 * conflicts in case kvm is already setup for another implementation.
3089 r = kvm_irqfd_init();
3093 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
3098 r = kvm_arch_hardware_setup();
3102 for_each_online_cpu(cpu) {
3103 smp_call_function_single(cpu,
3104 kvm_arch_check_processor_compat,
3110 r = register_cpu_notifier(&kvm_cpu_notifier);
3113 register_reboot_notifier(&kvm_reboot_notifier);
3115 /* A kmem cache lets us meet the alignment requirements of fx_save. */
3117 vcpu_align = __alignof__(struct kvm_vcpu);
3118 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
3120 if (!kvm_vcpu_cache) {
3125 r = kvm_async_pf_init();
3129 kvm_chardev_ops.owner = module;
3130 kvm_vm_fops.owner = module;
3131 kvm_vcpu_fops.owner = module;
3133 r = misc_register(&kvm_dev);
3135 printk(KERN_ERR "kvm: misc device register failed\n");
3139 register_syscore_ops(&kvm_syscore_ops);
3141 kvm_preempt_ops.sched_in = kvm_sched_in;
3142 kvm_preempt_ops.sched_out = kvm_sched_out;
3144 r = kvm_init_debug();
3146 printk(KERN_ERR "kvm: create debugfs files failed\n");
3153 unregister_syscore_ops(&kvm_syscore_ops);
3154 misc_deregister(&kvm_dev);
3156 kvm_async_pf_deinit();
3158 kmem_cache_destroy(kvm_vcpu_cache);
3160 unregister_reboot_notifier(&kvm_reboot_notifier);
3161 unregister_cpu_notifier(&kvm_cpu_notifier);
3164 kvm_arch_hardware_unsetup();
3166 free_cpumask_var(cpus_hardware_enabled);
3174 EXPORT_SYMBOL_GPL(kvm_init);
3179 misc_deregister(&kvm_dev);
3180 kmem_cache_destroy(kvm_vcpu_cache);
3181 kvm_async_pf_deinit();
3182 unregister_syscore_ops(&kvm_syscore_ops);
3183 unregister_reboot_notifier(&kvm_reboot_notifier);
3184 unregister_cpu_notifier(&kvm_cpu_notifier);
3185 on_each_cpu(hardware_disable_nolock, NULL, 1);
3186 kvm_arch_hardware_unsetup();
3189 free_cpumask_var(cpus_hardware_enabled);
3191 EXPORT_SYMBOL_GPL(kvm_exit);