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);
100 EXPORT_SYMBOL_GPL(kvm_rebooting);
102 static bool largepages_enabled = true;
104 bool kvm_is_mmio_pfn(pfn_t pfn)
107 return PageReserved(pfn_to_page(pfn));
113 * Switches to specified vcpu, until a matching vcpu_put()
115 int vcpu_load(struct kvm_vcpu *vcpu)
119 if (mutex_lock_killable(&vcpu->mutex))
121 if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
122 /* The thread running this VCPU changed. */
123 struct pid *oldpid = vcpu->pid;
124 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
125 rcu_assign_pointer(vcpu->pid, newpid);
130 preempt_notifier_register(&vcpu->preempt_notifier);
131 kvm_arch_vcpu_load(vcpu, cpu);
136 void vcpu_put(struct kvm_vcpu *vcpu)
139 kvm_arch_vcpu_put(vcpu);
140 preempt_notifier_unregister(&vcpu->preempt_notifier);
142 mutex_unlock(&vcpu->mutex);
145 static void ack_flush(void *_completed)
149 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
154 struct kvm_vcpu *vcpu;
156 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
159 kvm_for_each_vcpu(i, vcpu, kvm) {
160 kvm_make_request(req, vcpu);
163 /* Set ->requests bit before we read ->mode */
166 if (cpus != NULL && cpu != -1 && cpu != me &&
167 kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
168 cpumask_set_cpu(cpu, cpus);
170 if (unlikely(cpus == NULL))
171 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
172 else if (!cpumask_empty(cpus))
173 smp_call_function_many(cpus, ack_flush, NULL, 1);
177 free_cpumask_var(cpus);
181 void kvm_flush_remote_tlbs(struct kvm *kvm)
183 long dirty_count = kvm->tlbs_dirty;
186 if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
187 ++kvm->stat.remote_tlb_flush;
188 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
191 void kvm_reload_remote_mmus(struct kvm *kvm)
193 make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
196 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
198 make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
201 void kvm_make_scan_ioapic_request(struct kvm *kvm)
203 make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
206 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
211 mutex_init(&vcpu->mutex);
216 init_waitqueue_head(&vcpu->wq);
217 kvm_async_pf_vcpu_init(vcpu);
219 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
224 vcpu->run = page_address(page);
226 kvm_vcpu_set_in_spin_loop(vcpu, false);
227 kvm_vcpu_set_dy_eligible(vcpu, false);
228 vcpu->preempted = false;
230 r = kvm_arch_vcpu_init(vcpu);
236 free_page((unsigned long)vcpu->run);
240 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
242 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
245 kvm_arch_vcpu_uninit(vcpu);
246 free_page((unsigned long)vcpu->run);
248 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
250 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
251 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
253 return container_of(mn, struct kvm, mmu_notifier);
256 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
257 struct mm_struct *mm,
258 unsigned long address)
260 struct kvm *kvm = mmu_notifier_to_kvm(mn);
261 int need_tlb_flush, idx;
264 * When ->invalidate_page runs, the linux pte has been zapped
265 * already but the page is still allocated until
266 * ->invalidate_page returns. So if we increase the sequence
267 * here the kvm page fault will notice if the spte can't be
268 * established because the page is going to be freed. If
269 * instead the kvm page fault establishes the spte before
270 * ->invalidate_page runs, kvm_unmap_hva will release it
273 * The sequence increase only need to be seen at spin_unlock
274 * time, and not at spin_lock time.
276 * Increasing the sequence after the spin_unlock would be
277 * unsafe because the kvm page fault could then establish the
278 * pte after kvm_unmap_hva returned, without noticing the page
279 * is going to be freed.
281 idx = srcu_read_lock(&kvm->srcu);
282 spin_lock(&kvm->mmu_lock);
284 kvm->mmu_notifier_seq++;
285 need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
286 /* we've to flush the tlb before the pages can be freed */
288 kvm_flush_remote_tlbs(kvm);
290 spin_unlock(&kvm->mmu_lock);
291 srcu_read_unlock(&kvm->srcu, idx);
294 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
295 struct mm_struct *mm,
296 unsigned long address,
299 struct kvm *kvm = mmu_notifier_to_kvm(mn);
302 idx = srcu_read_lock(&kvm->srcu);
303 spin_lock(&kvm->mmu_lock);
304 kvm->mmu_notifier_seq++;
305 kvm_set_spte_hva(kvm, address, pte);
306 spin_unlock(&kvm->mmu_lock);
307 srcu_read_unlock(&kvm->srcu, idx);
310 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
311 struct mm_struct *mm,
315 struct kvm *kvm = mmu_notifier_to_kvm(mn);
316 int need_tlb_flush = 0, idx;
318 idx = srcu_read_lock(&kvm->srcu);
319 spin_lock(&kvm->mmu_lock);
321 * The count increase must become visible at unlock time as no
322 * spte can be established without taking the mmu_lock and
323 * count is also read inside the mmu_lock critical section.
325 kvm->mmu_notifier_count++;
326 need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
327 need_tlb_flush |= kvm->tlbs_dirty;
328 /* we've to flush the tlb before the pages can be freed */
330 kvm_flush_remote_tlbs(kvm);
332 spin_unlock(&kvm->mmu_lock);
333 srcu_read_unlock(&kvm->srcu, idx);
336 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
337 struct mm_struct *mm,
341 struct kvm *kvm = mmu_notifier_to_kvm(mn);
343 spin_lock(&kvm->mmu_lock);
345 * This sequence increase will notify the kvm page fault that
346 * the page that is going to be mapped in the spte could have
349 kvm->mmu_notifier_seq++;
352 * The above sequence increase must be visible before the
353 * below count decrease, which is ensured by the smp_wmb above
354 * in conjunction with the smp_rmb in mmu_notifier_retry().
356 kvm->mmu_notifier_count--;
357 spin_unlock(&kvm->mmu_lock);
359 BUG_ON(kvm->mmu_notifier_count < 0);
362 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
363 struct mm_struct *mm,
364 unsigned long address)
366 struct kvm *kvm = mmu_notifier_to_kvm(mn);
369 idx = srcu_read_lock(&kvm->srcu);
370 spin_lock(&kvm->mmu_lock);
372 young = kvm_age_hva(kvm, address);
374 kvm_flush_remote_tlbs(kvm);
376 spin_unlock(&kvm->mmu_lock);
377 srcu_read_unlock(&kvm->srcu, idx);
382 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
383 struct mm_struct *mm,
384 unsigned long address)
386 struct kvm *kvm = mmu_notifier_to_kvm(mn);
389 idx = srcu_read_lock(&kvm->srcu);
390 spin_lock(&kvm->mmu_lock);
391 young = kvm_test_age_hva(kvm, address);
392 spin_unlock(&kvm->mmu_lock);
393 srcu_read_unlock(&kvm->srcu, idx);
398 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
399 struct mm_struct *mm)
401 struct kvm *kvm = mmu_notifier_to_kvm(mn);
404 idx = srcu_read_lock(&kvm->srcu);
405 kvm_arch_flush_shadow_all(kvm);
406 srcu_read_unlock(&kvm->srcu, idx);
409 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
410 .invalidate_page = kvm_mmu_notifier_invalidate_page,
411 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
412 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
413 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
414 .test_young = kvm_mmu_notifier_test_young,
415 .change_pte = kvm_mmu_notifier_change_pte,
416 .release = kvm_mmu_notifier_release,
419 static int kvm_init_mmu_notifier(struct kvm *kvm)
421 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
422 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
425 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
427 static int kvm_init_mmu_notifier(struct kvm *kvm)
432 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
434 static void kvm_init_memslots_id(struct kvm *kvm)
437 struct kvm_memslots *slots = kvm->memslots;
439 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
440 slots->id_to_index[i] = slots->memslots[i].id = i;
443 static struct kvm *kvm_create_vm(unsigned long type)
446 struct kvm *kvm = kvm_arch_alloc_vm();
449 return ERR_PTR(-ENOMEM);
451 r = kvm_arch_init_vm(kvm, type);
453 goto out_err_nodisable;
455 r = hardware_enable_all();
457 goto out_err_nodisable;
459 #ifdef CONFIG_HAVE_KVM_IRQCHIP
460 INIT_HLIST_HEAD(&kvm->mask_notifier_list);
461 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
464 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
467 kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
470 kvm_init_memslots_id(kvm);
471 if (init_srcu_struct(&kvm->srcu))
473 for (i = 0; i < KVM_NR_BUSES; i++) {
474 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
480 spin_lock_init(&kvm->mmu_lock);
481 kvm->mm = current->mm;
482 atomic_inc(&kvm->mm->mm_count);
483 kvm_eventfd_init(kvm);
484 mutex_init(&kvm->lock);
485 mutex_init(&kvm->irq_lock);
486 mutex_init(&kvm->slots_lock);
487 atomic_set(&kvm->users_count, 1);
488 INIT_LIST_HEAD(&kvm->devices);
490 r = kvm_init_mmu_notifier(kvm);
494 spin_lock(&kvm_lock);
495 list_add(&kvm->vm_list, &vm_list);
496 spin_unlock(&kvm_lock);
501 cleanup_srcu_struct(&kvm->srcu);
503 hardware_disable_all();
505 for (i = 0; i < KVM_NR_BUSES; i++)
506 kfree(kvm->buses[i]);
507 kfree(kvm->memslots);
508 kvm_arch_free_vm(kvm);
513 * Avoid using vmalloc for a small buffer.
514 * Should not be used when the size is statically known.
516 void *kvm_kvzalloc(unsigned long size)
518 if (size > PAGE_SIZE)
519 return vzalloc(size);
521 return kzalloc(size, GFP_KERNEL);
524 void kvm_kvfree(const void *addr)
526 if (is_vmalloc_addr(addr))
532 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
534 if (!memslot->dirty_bitmap)
537 kvm_kvfree(memslot->dirty_bitmap);
538 memslot->dirty_bitmap = NULL;
542 * Free any memory in @free but not in @dont.
544 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
545 struct kvm_memory_slot *dont)
547 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
548 kvm_destroy_dirty_bitmap(free);
550 kvm_arch_free_memslot(free, dont);
555 void kvm_free_physmem(struct kvm *kvm)
557 struct kvm_memslots *slots = kvm->memslots;
558 struct kvm_memory_slot *memslot;
560 kvm_for_each_memslot(memslot, slots)
561 kvm_free_physmem_slot(memslot, NULL);
563 kfree(kvm->memslots);
566 static void kvm_destroy_devices(struct kvm *kvm)
568 struct list_head *node, *tmp;
570 list_for_each_safe(node, tmp, &kvm->devices) {
571 struct kvm_device *dev =
572 list_entry(node, struct kvm_device, vm_node);
575 dev->ops->destroy(dev);
579 static void kvm_destroy_vm(struct kvm *kvm)
582 struct mm_struct *mm = kvm->mm;
584 kvm_arch_sync_events(kvm);
585 spin_lock(&kvm_lock);
586 list_del(&kvm->vm_list);
587 spin_unlock(&kvm_lock);
588 kvm_free_irq_routing(kvm);
589 for (i = 0; i < KVM_NR_BUSES; i++)
590 kvm_io_bus_destroy(kvm->buses[i]);
591 kvm_coalesced_mmio_free(kvm);
592 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
593 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
595 kvm_arch_flush_shadow_all(kvm);
597 kvm_arch_destroy_vm(kvm);
598 kvm_destroy_devices(kvm);
599 kvm_free_physmem(kvm);
600 cleanup_srcu_struct(&kvm->srcu);
601 kvm_arch_free_vm(kvm);
602 hardware_disable_all();
606 void kvm_get_kvm(struct kvm *kvm)
608 atomic_inc(&kvm->users_count);
610 EXPORT_SYMBOL_GPL(kvm_get_kvm);
612 void kvm_put_kvm(struct kvm *kvm)
614 if (atomic_dec_and_test(&kvm->users_count))
617 EXPORT_SYMBOL_GPL(kvm_put_kvm);
620 static int kvm_vm_release(struct inode *inode, struct file *filp)
622 struct kvm *kvm = filp->private_data;
624 kvm_irqfd_release(kvm);
631 * Allocation size is twice as large as the actual dirty bitmap size.
632 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
634 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
637 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
639 memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
640 if (!memslot->dirty_bitmap)
643 #endif /* !CONFIG_S390 */
647 static int cmp_memslot(const void *slot1, const void *slot2)
649 struct kvm_memory_slot *s1, *s2;
651 s1 = (struct kvm_memory_slot *)slot1;
652 s2 = (struct kvm_memory_slot *)slot2;
654 if (s1->npages < s2->npages)
656 if (s1->npages > s2->npages)
663 * Sort the memslots base on its size, so the larger slots
664 * will get better fit.
666 static void sort_memslots(struct kvm_memslots *slots)
670 sort(slots->memslots, KVM_MEM_SLOTS_NUM,
671 sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
673 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
674 slots->id_to_index[slots->memslots[i].id] = i;
677 void update_memslots(struct kvm_memslots *slots, struct kvm_memory_slot *new,
682 struct kvm_memory_slot *old = id_to_memslot(slots, id);
683 unsigned long npages = old->npages;
686 if (new->npages != npages)
687 sort_memslots(slots);
690 slots->generation = last_generation + 1;
693 static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
695 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
697 #ifdef KVM_CAP_READONLY_MEM
698 valid_flags |= KVM_MEM_READONLY;
701 if (mem->flags & ~valid_flags)
707 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
708 struct kvm_memslots *slots, struct kvm_memory_slot *new)
710 struct kvm_memslots *old_memslots = kvm->memslots;
712 update_memslots(slots, new, kvm->memslots->generation);
713 rcu_assign_pointer(kvm->memslots, slots);
714 synchronize_srcu_expedited(&kvm->srcu);
716 kvm_arch_memslots_updated(kvm);
722 * Allocate some memory and give it an address in the guest physical address
725 * Discontiguous memory is allowed, mostly for framebuffers.
727 * Must be called holding mmap_sem for write.
729 int __kvm_set_memory_region(struct kvm *kvm,
730 struct kvm_userspace_memory_region *mem)
734 unsigned long npages;
735 struct kvm_memory_slot *slot;
736 struct kvm_memory_slot old, new;
737 struct kvm_memslots *slots = NULL, *old_memslots;
738 enum kvm_mr_change change;
740 r = check_memory_region_flags(mem);
745 /* General sanity checks */
746 if (mem->memory_size & (PAGE_SIZE - 1))
748 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
750 /* We can read the guest memory with __xxx_user() later on. */
751 if ((mem->slot < KVM_USER_MEM_SLOTS) &&
752 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
753 !access_ok(VERIFY_WRITE,
754 (void __user *)(unsigned long)mem->userspace_addr,
757 if (mem->slot >= KVM_MEM_SLOTS_NUM)
759 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
762 slot = id_to_memslot(kvm->memslots, mem->slot);
763 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
764 npages = mem->memory_size >> PAGE_SHIFT;
767 if (npages > KVM_MEM_MAX_NR_PAGES)
771 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
776 new.base_gfn = base_gfn;
778 new.flags = mem->flags;
783 change = KVM_MR_CREATE;
784 else { /* Modify an existing slot. */
785 if ((mem->userspace_addr != old.userspace_addr) ||
786 (npages != old.npages) ||
787 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
790 if (base_gfn != old.base_gfn)
791 change = KVM_MR_MOVE;
792 else if (new.flags != old.flags)
793 change = KVM_MR_FLAGS_ONLY;
794 else { /* Nothing to change. */
799 } else if (old.npages) {
800 change = KVM_MR_DELETE;
801 } else /* Modify a non-existent slot: disallowed. */
804 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
805 /* Check for overlaps */
807 kvm_for_each_memslot(slot, kvm->memslots) {
808 if ((slot->id >= KVM_USER_MEM_SLOTS) ||
809 (slot->id == mem->slot))
811 if (!((base_gfn + npages <= slot->base_gfn) ||
812 (base_gfn >= slot->base_gfn + slot->npages)))
817 /* Free page dirty bitmap if unneeded */
818 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
819 new.dirty_bitmap = NULL;
822 if (change == KVM_MR_CREATE) {
823 new.userspace_addr = mem->userspace_addr;
825 if (kvm_arch_create_memslot(&new, npages))
829 /* Allocate page dirty bitmap if needed */
830 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
831 if (kvm_create_dirty_bitmap(&new) < 0)
835 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
837 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
841 slot = id_to_memslot(slots, mem->slot);
842 slot->flags |= KVM_MEMSLOT_INVALID;
844 old_memslots = install_new_memslots(kvm, slots, NULL);
846 /* slot was deleted or moved, clear iommu mapping */
847 kvm_iommu_unmap_pages(kvm, &old);
848 /* From this point no new shadow pages pointing to a deleted,
849 * or moved, memslot will be created.
851 * validation of sp->gfn happens in:
852 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
853 * - kvm_is_visible_gfn (mmu_check_roots)
855 kvm_arch_flush_shadow_memslot(kvm, slot);
856 slots = old_memslots;
859 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
865 * We can re-use the old_memslots from above, the only difference
866 * from the currently installed memslots is the invalid flag. This
867 * will get overwritten by update_memslots anyway.
870 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
877 * IOMMU mapping: New slots need to be mapped. Old slots need to be
878 * un-mapped and re-mapped if their base changes. Since base change
879 * unmapping is handled above with slot deletion, mapping alone is
880 * needed here. Anything else the iommu might care about for existing
881 * slots (size changes, userspace addr changes and read-only flag
882 * changes) is disallowed above, so any other attribute changes getting
883 * here can be skipped.
885 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
886 r = kvm_iommu_map_pages(kvm, &new);
891 /* actual memory is freed via old in kvm_free_physmem_slot below */
892 if (change == KVM_MR_DELETE) {
893 new.dirty_bitmap = NULL;
894 memset(&new.arch, 0, sizeof(new.arch));
897 old_memslots = install_new_memslots(kvm, slots, &new);
899 kvm_arch_commit_memory_region(kvm, mem, &old, change);
901 kvm_free_physmem_slot(&old, &new);
909 kvm_free_physmem_slot(&new, &old);
913 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
915 int kvm_set_memory_region(struct kvm *kvm,
916 struct kvm_userspace_memory_region *mem)
920 mutex_lock(&kvm->slots_lock);
921 r = __kvm_set_memory_region(kvm, mem);
922 mutex_unlock(&kvm->slots_lock);
925 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
927 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
928 struct kvm_userspace_memory_region *mem)
930 if (mem->slot >= KVM_USER_MEM_SLOTS)
932 return kvm_set_memory_region(kvm, mem);
935 int kvm_get_dirty_log(struct kvm *kvm,
936 struct kvm_dirty_log *log, int *is_dirty)
938 struct kvm_memory_slot *memslot;
941 unsigned long any = 0;
944 if (log->slot >= KVM_USER_MEM_SLOTS)
947 memslot = id_to_memslot(kvm->memslots, log->slot);
949 if (!memslot->dirty_bitmap)
952 n = kvm_dirty_bitmap_bytes(memslot);
954 for (i = 0; !any && i < n/sizeof(long); ++i)
955 any = memslot->dirty_bitmap[i];
958 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
969 bool kvm_largepages_enabled(void)
971 return largepages_enabled;
974 void kvm_disable_largepages(void)
976 largepages_enabled = false;
978 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
980 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
982 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
984 EXPORT_SYMBOL_GPL(gfn_to_memslot);
986 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
988 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
990 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
991 memslot->flags & KVM_MEMSLOT_INVALID)
996 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
998 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1000 struct vm_area_struct *vma;
1001 unsigned long addr, size;
1005 addr = gfn_to_hva(kvm, gfn);
1006 if (kvm_is_error_hva(addr))
1009 down_read(¤t->mm->mmap_sem);
1010 vma = find_vma(current->mm, addr);
1014 size = vma_kernel_pagesize(vma);
1017 up_read(¤t->mm->mmap_sem);
1022 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1024 return slot->flags & KVM_MEM_READONLY;
1027 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1028 gfn_t *nr_pages, bool write)
1030 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1031 return KVM_HVA_ERR_BAD;
1033 if (memslot_is_readonly(slot) && write)
1034 return KVM_HVA_ERR_RO_BAD;
1037 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1039 return __gfn_to_hva_memslot(slot, gfn);
1042 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1045 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1048 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1051 return gfn_to_hva_many(slot, gfn, NULL);
1053 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1055 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1057 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1059 EXPORT_SYMBOL_GPL(gfn_to_hva);
1062 * If writable is set to false, the hva returned by this function is only
1063 * allowed to be read.
1065 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1067 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1068 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1070 if (!kvm_is_error_hva(hva) && writable)
1071 *writable = !memslot_is_readonly(slot);
1076 static int kvm_read_hva(void *data, void __user *hva, int len)
1078 return __copy_from_user(data, hva, len);
1081 static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1083 return __copy_from_user_inatomic(data, hva, len);
1086 static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1087 unsigned long start, int write, struct page **page)
1089 int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1092 flags |= FOLL_WRITE;
1094 return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1097 static inline int check_user_page_hwpoison(unsigned long addr)
1099 int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1101 rc = __get_user_pages(current, current->mm, addr, 1,
1102 flags, NULL, NULL, NULL);
1103 return rc == -EHWPOISON;
1107 * The atomic path to get the writable pfn which will be stored in @pfn,
1108 * true indicates success, otherwise false is returned.
1110 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1111 bool write_fault, bool *writable, pfn_t *pfn)
1113 struct page *page[1];
1116 if (!(async || atomic))
1120 * Fast pin a writable pfn only if it is a write fault request
1121 * or the caller allows to map a writable pfn for a read fault
1124 if (!(write_fault || writable))
1127 npages = __get_user_pages_fast(addr, 1, 1, page);
1129 *pfn = page_to_pfn(page[0]);
1140 * The slow path to get the pfn of the specified host virtual address,
1141 * 1 indicates success, -errno is returned if error is detected.
1143 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1144 bool *writable, pfn_t *pfn)
1146 struct page *page[1];
1152 *writable = write_fault;
1155 down_read(¤t->mm->mmap_sem);
1156 npages = get_user_page_nowait(current, current->mm,
1157 addr, write_fault, page);
1158 up_read(¤t->mm->mmap_sem);
1160 npages = get_user_pages_fast(addr, 1, write_fault,
1165 /* map read fault as writable if possible */
1166 if (unlikely(!write_fault) && writable) {
1167 struct page *wpage[1];
1169 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1178 *pfn = page_to_pfn(page[0]);
1182 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1184 if (unlikely(!(vma->vm_flags & VM_READ)))
1187 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1194 * Pin guest page in memory and return its pfn.
1195 * @addr: host virtual address which maps memory to the guest
1196 * @atomic: whether this function can sleep
1197 * @async: whether this function need to wait IO complete if the
1198 * host page is not in the memory
1199 * @write_fault: whether we should get a writable host page
1200 * @writable: whether it allows to map a writable host page for !@write_fault
1202 * The function will map a writable host page for these two cases:
1203 * 1): @write_fault = true
1204 * 2): @write_fault = false && @writable, @writable will tell the caller
1205 * whether the mapping is writable.
1207 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1208 bool write_fault, bool *writable)
1210 struct vm_area_struct *vma;
1214 /* we can do it either atomically or asynchronously, not both */
1215 BUG_ON(atomic && async);
1217 if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1221 return KVM_PFN_ERR_FAULT;
1223 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1227 down_read(¤t->mm->mmap_sem);
1228 if (npages == -EHWPOISON ||
1229 (!async && check_user_page_hwpoison(addr))) {
1230 pfn = KVM_PFN_ERR_HWPOISON;
1234 vma = find_vma_intersection(current->mm, addr, addr + 1);
1237 pfn = KVM_PFN_ERR_FAULT;
1238 else if ((vma->vm_flags & VM_PFNMAP)) {
1239 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1241 BUG_ON(!kvm_is_mmio_pfn(pfn));
1243 if (async && vma_is_valid(vma, write_fault))
1245 pfn = KVM_PFN_ERR_FAULT;
1248 up_read(¤t->mm->mmap_sem);
1253 __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1254 bool *async, bool write_fault, bool *writable)
1256 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1258 if (addr == KVM_HVA_ERR_RO_BAD)
1259 return KVM_PFN_ERR_RO_FAULT;
1261 if (kvm_is_error_hva(addr))
1262 return KVM_PFN_NOSLOT;
1264 /* Do not map writable pfn in the readonly memslot. */
1265 if (writable && memslot_is_readonly(slot)) {
1270 return hva_to_pfn(addr, atomic, async, write_fault,
1274 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1275 bool write_fault, bool *writable)
1277 struct kvm_memory_slot *slot;
1282 slot = gfn_to_memslot(kvm, gfn);
1284 return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1288 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1290 return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1292 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1294 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1295 bool write_fault, bool *writable)
1297 return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1299 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1301 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1303 return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1305 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1307 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1310 return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1312 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1314 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1316 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1319 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1321 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1323 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1325 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1331 addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1332 if (kvm_is_error_hva(addr))
1335 if (entry < nr_pages)
1338 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1340 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1342 static struct page *kvm_pfn_to_page(pfn_t pfn)
1344 if (is_error_noslot_pfn(pfn))
1345 return KVM_ERR_PTR_BAD_PAGE;
1347 if (kvm_is_mmio_pfn(pfn)) {
1349 return KVM_ERR_PTR_BAD_PAGE;
1352 return pfn_to_page(pfn);
1355 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1359 pfn = gfn_to_pfn(kvm, gfn);
1361 return kvm_pfn_to_page(pfn);
1364 EXPORT_SYMBOL_GPL(gfn_to_page);
1366 void kvm_release_page_clean(struct page *page)
1368 WARN_ON(is_error_page(page));
1370 kvm_release_pfn_clean(page_to_pfn(page));
1372 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1374 void kvm_release_pfn_clean(pfn_t pfn)
1376 if (!is_error_noslot_pfn(pfn) && !kvm_is_mmio_pfn(pfn))
1377 put_page(pfn_to_page(pfn));
1379 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1381 void kvm_release_page_dirty(struct page *page)
1383 WARN_ON(is_error_page(page));
1385 kvm_release_pfn_dirty(page_to_pfn(page));
1387 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1389 void kvm_release_pfn_dirty(pfn_t pfn)
1391 kvm_set_pfn_dirty(pfn);
1392 kvm_release_pfn_clean(pfn);
1394 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1396 void kvm_set_page_dirty(struct page *page)
1398 kvm_set_pfn_dirty(page_to_pfn(page));
1400 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1402 void kvm_set_pfn_dirty(pfn_t pfn)
1404 if (!kvm_is_mmio_pfn(pfn)) {
1405 struct page *page = pfn_to_page(pfn);
1406 if (!PageReserved(page))
1410 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1412 void kvm_set_pfn_accessed(pfn_t pfn)
1414 if (!kvm_is_mmio_pfn(pfn))
1415 mark_page_accessed(pfn_to_page(pfn));
1417 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1419 void kvm_get_pfn(pfn_t pfn)
1421 if (!kvm_is_mmio_pfn(pfn))
1422 get_page(pfn_to_page(pfn));
1424 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1426 static int next_segment(unsigned long len, int offset)
1428 if (len > PAGE_SIZE - offset)
1429 return PAGE_SIZE - offset;
1434 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1440 addr = gfn_to_hva_prot(kvm, gfn, NULL);
1441 if (kvm_is_error_hva(addr))
1443 r = kvm_read_hva(data, (void __user *)addr + offset, len);
1448 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1450 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1452 gfn_t gfn = gpa >> PAGE_SHIFT;
1454 int offset = offset_in_page(gpa);
1457 while ((seg = next_segment(len, offset)) != 0) {
1458 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1468 EXPORT_SYMBOL_GPL(kvm_read_guest);
1470 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1475 gfn_t gfn = gpa >> PAGE_SHIFT;
1476 int offset = offset_in_page(gpa);
1478 addr = gfn_to_hva_prot(kvm, gfn, NULL);
1479 if (kvm_is_error_hva(addr))
1481 pagefault_disable();
1482 r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
1488 EXPORT_SYMBOL(kvm_read_guest_atomic);
1490 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1491 int offset, int len)
1496 addr = gfn_to_hva(kvm, gfn);
1497 if (kvm_is_error_hva(addr))
1499 r = __copy_to_user((void __user *)addr + offset, data, len);
1502 mark_page_dirty(kvm, gfn);
1505 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1507 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1510 gfn_t gfn = gpa >> PAGE_SHIFT;
1512 int offset = offset_in_page(gpa);
1515 while ((seg = next_segment(len, offset)) != 0) {
1516 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1527 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1528 gpa_t gpa, unsigned long len)
1530 struct kvm_memslots *slots = kvm_memslots(kvm);
1531 int offset = offset_in_page(gpa);
1532 gfn_t start_gfn = gpa >> PAGE_SHIFT;
1533 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1534 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1535 gfn_t nr_pages_avail;
1538 ghc->generation = slots->generation;
1540 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1541 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, &nr_pages_avail);
1542 if (!kvm_is_error_hva(ghc->hva) && nr_pages_avail >= nr_pages_needed) {
1546 * If the requested region crosses two memslots, we still
1547 * verify that the entire region is valid here.
1549 while (start_gfn <= end_gfn) {
1550 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1551 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1553 if (kvm_is_error_hva(ghc->hva))
1555 start_gfn += nr_pages_avail;
1557 /* Use the slow path for cross page reads and writes. */
1558 ghc->memslot = NULL;
1562 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1564 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1565 void *data, unsigned long len)
1567 struct kvm_memslots *slots = kvm_memslots(kvm);
1570 BUG_ON(len > ghc->len);
1572 if (slots->generation != ghc->generation)
1573 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1575 if (unlikely(!ghc->memslot))
1576 return kvm_write_guest(kvm, ghc->gpa, data, len);
1578 if (kvm_is_error_hva(ghc->hva))
1581 r = __copy_to_user((void __user *)ghc->hva, data, len);
1584 mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1588 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1590 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1591 void *data, unsigned long len)
1593 struct kvm_memslots *slots = kvm_memslots(kvm);
1596 BUG_ON(len > ghc->len);
1598 if (slots->generation != ghc->generation)
1599 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1601 if (unlikely(!ghc->memslot))
1602 return kvm_read_guest(kvm, ghc->gpa, data, len);
1604 if (kvm_is_error_hva(ghc->hva))
1607 r = __copy_from_user(data, (void __user *)ghc->hva, len);
1613 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1615 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1617 return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1620 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1622 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1624 gfn_t gfn = gpa >> PAGE_SHIFT;
1626 int offset = offset_in_page(gpa);
1629 while ((seg = next_segment(len, offset)) != 0) {
1630 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1639 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1641 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1644 if (memslot && memslot->dirty_bitmap) {
1645 unsigned long rel_gfn = gfn - memslot->base_gfn;
1647 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1651 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1653 struct kvm_memory_slot *memslot;
1655 memslot = gfn_to_memslot(kvm, gfn);
1656 mark_page_dirty_in_slot(kvm, memslot, gfn);
1660 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1662 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1667 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1669 if (kvm_arch_vcpu_runnable(vcpu)) {
1670 kvm_make_request(KVM_REQ_UNHALT, vcpu);
1673 if (kvm_cpu_has_pending_timer(vcpu))
1675 if (signal_pending(current))
1681 finish_wait(&vcpu->wq, &wait);
1686 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1688 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1691 int cpu = vcpu->cpu;
1692 wait_queue_head_t *wqp;
1694 wqp = kvm_arch_vcpu_wq(vcpu);
1695 if (waitqueue_active(wqp)) {
1696 wake_up_interruptible(wqp);
1697 ++vcpu->stat.halt_wakeup;
1701 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1702 if (kvm_arch_vcpu_should_kick(vcpu))
1703 smp_send_reschedule(cpu);
1706 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
1707 #endif /* !CONFIG_S390 */
1709 void kvm_resched(struct kvm_vcpu *vcpu)
1711 if (!need_resched())
1715 EXPORT_SYMBOL_GPL(kvm_resched);
1717 bool kvm_vcpu_yield_to(struct kvm_vcpu *target)
1720 struct task_struct *task = NULL;
1724 pid = rcu_dereference(target->pid);
1726 task = get_pid_task(target->pid, PIDTYPE_PID);
1730 if (task->flags & PF_VCPU) {
1731 put_task_struct(task);
1734 ret = yield_to(task, 1);
1735 put_task_struct(task);
1739 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1741 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1743 * Helper that checks whether a VCPU is eligible for directed yield.
1744 * Most eligible candidate to yield is decided by following heuristics:
1746 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1747 * (preempted lock holder), indicated by @in_spin_loop.
1748 * Set at the beiginning and cleared at the end of interception/PLE handler.
1750 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1751 * chance last time (mostly it has become eligible now since we have probably
1752 * yielded to lockholder in last iteration. This is done by toggling
1753 * @dy_eligible each time a VCPU checked for eligibility.)
1755 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1756 * to preempted lock-holder could result in wrong VCPU selection and CPU
1757 * burning. Giving priority for a potential lock-holder increases lock
1760 * Since algorithm is based on heuristics, accessing another VCPU data without
1761 * locking does not harm. It may result in trying to yield to same VCPU, fail
1762 * and continue with next VCPU and so on.
1764 bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1768 eligible = !vcpu->spin_loop.in_spin_loop ||
1769 (vcpu->spin_loop.in_spin_loop &&
1770 vcpu->spin_loop.dy_eligible);
1772 if (vcpu->spin_loop.in_spin_loop)
1773 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))
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 vcpu = kvm_arch_vcpu_create(kvm, id);
1899 return PTR_ERR(vcpu);
1901 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1903 r = kvm_arch_vcpu_setup(vcpu);
1907 mutex_lock(&kvm->lock);
1908 if (!kvm_vcpu_compatible(vcpu)) {
1910 goto unlock_vcpu_destroy;
1912 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1914 goto unlock_vcpu_destroy;
1917 kvm_for_each_vcpu(r, v, kvm)
1918 if (v->vcpu_id == id) {
1920 goto unlock_vcpu_destroy;
1923 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1925 /* Now it's all set up, let userspace reach it */
1927 r = create_vcpu_fd(vcpu);
1930 goto unlock_vcpu_destroy;
1933 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1935 atomic_inc(&kvm->online_vcpus);
1937 mutex_unlock(&kvm->lock);
1938 kvm_arch_vcpu_postcreate(vcpu);
1941 unlock_vcpu_destroy:
1942 mutex_unlock(&kvm->lock);
1944 kvm_arch_vcpu_destroy(vcpu);
1948 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1951 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1952 vcpu->sigset_active = 1;
1953 vcpu->sigset = *sigset;
1955 vcpu->sigset_active = 0;
1959 static long kvm_vcpu_ioctl(struct file *filp,
1960 unsigned int ioctl, unsigned long arg)
1962 struct kvm_vcpu *vcpu = filp->private_data;
1963 void __user *argp = (void __user *)arg;
1965 struct kvm_fpu *fpu = NULL;
1966 struct kvm_sregs *kvm_sregs = NULL;
1968 if (vcpu->kvm->mm != current->mm)
1971 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
1973 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1974 * so vcpu_load() would break it.
1976 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1977 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1981 r = vcpu_load(vcpu);
1989 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1990 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1992 case KVM_GET_REGS: {
1993 struct kvm_regs *kvm_regs;
1996 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1999 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2003 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2010 case KVM_SET_REGS: {
2011 struct kvm_regs *kvm_regs;
2014 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2015 if (IS_ERR(kvm_regs)) {
2016 r = PTR_ERR(kvm_regs);
2019 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2023 case KVM_GET_SREGS: {
2024 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2028 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2032 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2037 case KVM_SET_SREGS: {
2038 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2039 if (IS_ERR(kvm_sregs)) {
2040 r = PTR_ERR(kvm_sregs);
2044 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2047 case KVM_GET_MP_STATE: {
2048 struct kvm_mp_state mp_state;
2050 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2054 if (copy_to_user(argp, &mp_state, sizeof mp_state))
2059 case KVM_SET_MP_STATE: {
2060 struct kvm_mp_state mp_state;
2063 if (copy_from_user(&mp_state, argp, sizeof mp_state))
2065 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2068 case KVM_TRANSLATE: {
2069 struct kvm_translation tr;
2072 if (copy_from_user(&tr, argp, sizeof tr))
2074 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2078 if (copy_to_user(argp, &tr, sizeof tr))
2083 case KVM_SET_GUEST_DEBUG: {
2084 struct kvm_guest_debug dbg;
2087 if (copy_from_user(&dbg, argp, sizeof dbg))
2089 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2092 case KVM_SET_SIGNAL_MASK: {
2093 struct kvm_signal_mask __user *sigmask_arg = argp;
2094 struct kvm_signal_mask kvm_sigmask;
2095 sigset_t sigset, *p;
2100 if (copy_from_user(&kvm_sigmask, argp,
2101 sizeof kvm_sigmask))
2104 if (kvm_sigmask.len != sizeof sigset)
2107 if (copy_from_user(&sigset, sigmask_arg->sigset,
2112 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2116 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2120 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2124 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2130 fpu = memdup_user(argp, sizeof(*fpu));
2136 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2140 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2149 #ifdef CONFIG_COMPAT
2150 static long kvm_vcpu_compat_ioctl(struct file *filp,
2151 unsigned int ioctl, unsigned long arg)
2153 struct kvm_vcpu *vcpu = filp->private_data;
2154 void __user *argp = compat_ptr(arg);
2157 if (vcpu->kvm->mm != current->mm)
2161 case KVM_SET_SIGNAL_MASK: {
2162 struct kvm_signal_mask __user *sigmask_arg = argp;
2163 struct kvm_signal_mask kvm_sigmask;
2164 compat_sigset_t csigset;
2169 if (copy_from_user(&kvm_sigmask, argp,
2170 sizeof kvm_sigmask))
2173 if (kvm_sigmask.len != sizeof csigset)
2176 if (copy_from_user(&csigset, sigmask_arg->sigset,
2179 sigset_from_compat(&sigset, &csigset);
2180 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2182 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2186 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2194 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2195 int (*accessor)(struct kvm_device *dev,
2196 struct kvm_device_attr *attr),
2199 struct kvm_device_attr attr;
2204 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2207 return accessor(dev, &attr);
2210 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2213 struct kvm_device *dev = filp->private_data;
2216 case KVM_SET_DEVICE_ATTR:
2217 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2218 case KVM_GET_DEVICE_ATTR:
2219 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2220 case KVM_HAS_DEVICE_ATTR:
2221 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2223 if (dev->ops->ioctl)
2224 return dev->ops->ioctl(dev, ioctl, arg);
2230 static int kvm_device_release(struct inode *inode, struct file *filp)
2232 struct kvm_device *dev = filp->private_data;
2233 struct kvm *kvm = dev->kvm;
2239 static const struct file_operations kvm_device_fops = {
2240 .unlocked_ioctl = kvm_device_ioctl,
2241 #ifdef CONFIG_COMPAT
2242 .compat_ioctl = kvm_device_ioctl,
2244 .release = kvm_device_release,
2247 struct kvm_device *kvm_device_from_filp(struct file *filp)
2249 if (filp->f_op != &kvm_device_fops)
2252 return filp->private_data;
2255 static int kvm_ioctl_create_device(struct kvm *kvm,
2256 struct kvm_create_device *cd)
2258 struct kvm_device_ops *ops = NULL;
2259 struct kvm_device *dev;
2260 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2264 #ifdef CONFIG_KVM_MPIC
2265 case KVM_DEV_TYPE_FSL_MPIC_20:
2266 case KVM_DEV_TYPE_FSL_MPIC_42:
2267 ops = &kvm_mpic_ops;
2270 #ifdef CONFIG_KVM_XICS
2271 case KVM_DEV_TYPE_XICS:
2272 ops = &kvm_xics_ops;
2282 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2289 ret = ops->create(dev, cd->type);
2295 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2301 list_add(&dev->vm_node, &kvm->devices);
2307 static long kvm_vm_ioctl(struct file *filp,
2308 unsigned int ioctl, unsigned long arg)
2310 struct kvm *kvm = filp->private_data;
2311 void __user *argp = (void __user *)arg;
2314 if (kvm->mm != current->mm)
2317 case KVM_CREATE_VCPU:
2318 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2320 case KVM_SET_USER_MEMORY_REGION: {
2321 struct kvm_userspace_memory_region kvm_userspace_mem;
2324 if (copy_from_user(&kvm_userspace_mem, argp,
2325 sizeof kvm_userspace_mem))
2328 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2331 case KVM_GET_DIRTY_LOG: {
2332 struct kvm_dirty_log log;
2335 if (copy_from_user(&log, argp, sizeof log))
2337 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2340 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2341 case KVM_REGISTER_COALESCED_MMIO: {
2342 struct kvm_coalesced_mmio_zone zone;
2344 if (copy_from_user(&zone, argp, sizeof zone))
2346 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2349 case KVM_UNREGISTER_COALESCED_MMIO: {
2350 struct kvm_coalesced_mmio_zone zone;
2352 if (copy_from_user(&zone, argp, sizeof zone))
2354 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2359 struct kvm_irqfd data;
2362 if (copy_from_user(&data, argp, sizeof data))
2364 r = kvm_irqfd(kvm, &data);
2367 case KVM_IOEVENTFD: {
2368 struct kvm_ioeventfd data;
2371 if (copy_from_user(&data, argp, sizeof data))
2373 r = kvm_ioeventfd(kvm, &data);
2376 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2377 case KVM_SET_BOOT_CPU_ID:
2379 mutex_lock(&kvm->lock);
2380 if (atomic_read(&kvm->online_vcpus) != 0)
2383 kvm->bsp_vcpu_id = arg;
2384 mutex_unlock(&kvm->lock);
2387 #ifdef CONFIG_HAVE_KVM_MSI
2388 case KVM_SIGNAL_MSI: {
2392 if (copy_from_user(&msi, argp, sizeof msi))
2394 r = kvm_send_userspace_msi(kvm, &msi);
2398 #ifdef __KVM_HAVE_IRQ_LINE
2399 case KVM_IRQ_LINE_STATUS:
2400 case KVM_IRQ_LINE: {
2401 struct kvm_irq_level irq_event;
2404 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2407 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
2408 ioctl == KVM_IRQ_LINE_STATUS);
2413 if (ioctl == KVM_IRQ_LINE_STATUS) {
2414 if (copy_to_user(argp, &irq_event, sizeof irq_event))
2422 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2423 case KVM_SET_GSI_ROUTING: {
2424 struct kvm_irq_routing routing;
2425 struct kvm_irq_routing __user *urouting;
2426 struct kvm_irq_routing_entry *entries;
2429 if (copy_from_user(&routing, argp, sizeof(routing)))
2432 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2437 entries = vmalloc(routing.nr * sizeof(*entries));
2442 if (copy_from_user(entries, urouting->entries,
2443 routing.nr * sizeof(*entries)))
2444 goto out_free_irq_routing;
2445 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2447 out_free_irq_routing:
2451 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
2452 case KVM_CREATE_DEVICE: {
2453 struct kvm_create_device cd;
2456 if (copy_from_user(&cd, argp, sizeof(cd)))
2459 r = kvm_ioctl_create_device(kvm, &cd);
2464 if (copy_to_user(argp, &cd, sizeof(cd)))
2471 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2473 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2479 #ifdef CONFIG_COMPAT
2480 struct compat_kvm_dirty_log {
2484 compat_uptr_t dirty_bitmap; /* one bit per page */
2489 static long kvm_vm_compat_ioctl(struct file *filp,
2490 unsigned int ioctl, unsigned long arg)
2492 struct kvm *kvm = filp->private_data;
2495 if (kvm->mm != current->mm)
2498 case KVM_GET_DIRTY_LOG: {
2499 struct compat_kvm_dirty_log compat_log;
2500 struct kvm_dirty_log log;
2503 if (copy_from_user(&compat_log, (void __user *)arg,
2504 sizeof(compat_log)))
2506 log.slot = compat_log.slot;
2507 log.padding1 = compat_log.padding1;
2508 log.padding2 = compat_log.padding2;
2509 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2511 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2515 r = kvm_vm_ioctl(filp, ioctl, arg);
2523 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2525 struct page *page[1];
2528 gfn_t gfn = vmf->pgoff;
2529 struct kvm *kvm = vma->vm_file->private_data;
2531 addr = gfn_to_hva(kvm, gfn);
2532 if (kvm_is_error_hva(addr))
2533 return VM_FAULT_SIGBUS;
2535 npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2537 if (unlikely(npages != 1))
2538 return VM_FAULT_SIGBUS;
2540 vmf->page = page[0];
2544 static const struct vm_operations_struct kvm_vm_vm_ops = {
2545 .fault = kvm_vm_fault,
2548 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2550 vma->vm_ops = &kvm_vm_vm_ops;
2554 static struct file_operations kvm_vm_fops = {
2555 .release = kvm_vm_release,
2556 .unlocked_ioctl = kvm_vm_ioctl,
2557 #ifdef CONFIG_COMPAT
2558 .compat_ioctl = kvm_vm_compat_ioctl,
2560 .mmap = kvm_vm_mmap,
2561 .llseek = noop_llseek,
2564 static int kvm_dev_ioctl_create_vm(unsigned long type)
2569 kvm = kvm_create_vm(type);
2571 return PTR_ERR(kvm);
2572 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2573 r = kvm_coalesced_mmio_init(kvm);
2579 r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC);
2586 static long kvm_dev_ioctl_check_extension_generic(long arg)
2589 case KVM_CAP_USER_MEMORY:
2590 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2591 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2592 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2593 case KVM_CAP_SET_BOOT_CPU_ID:
2595 case KVM_CAP_INTERNAL_ERROR_DATA:
2596 #ifdef CONFIG_HAVE_KVM_MSI
2597 case KVM_CAP_SIGNAL_MSI:
2599 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2600 case KVM_CAP_IRQFD_RESAMPLE:
2603 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2604 case KVM_CAP_IRQ_ROUTING:
2605 return KVM_MAX_IRQ_ROUTES;
2610 return kvm_dev_ioctl_check_extension(arg);
2613 static long kvm_dev_ioctl(struct file *filp,
2614 unsigned int ioctl, unsigned long arg)
2619 case KVM_GET_API_VERSION:
2623 r = KVM_API_VERSION;
2626 r = kvm_dev_ioctl_create_vm(arg);
2628 case KVM_CHECK_EXTENSION:
2629 r = kvm_dev_ioctl_check_extension_generic(arg);
2631 case KVM_GET_VCPU_MMAP_SIZE:
2635 r = PAGE_SIZE; /* struct kvm_run */
2637 r += PAGE_SIZE; /* pio data page */
2639 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2640 r += PAGE_SIZE; /* coalesced mmio ring page */
2643 case KVM_TRACE_ENABLE:
2644 case KVM_TRACE_PAUSE:
2645 case KVM_TRACE_DISABLE:
2649 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2655 static struct file_operations kvm_chardev_ops = {
2656 .unlocked_ioctl = kvm_dev_ioctl,
2657 .compat_ioctl = kvm_dev_ioctl,
2658 .llseek = noop_llseek,
2661 static struct miscdevice kvm_dev = {
2667 static void hardware_enable_nolock(void *junk)
2669 int cpu = raw_smp_processor_id();
2672 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2675 cpumask_set_cpu(cpu, cpus_hardware_enabled);
2677 r = kvm_arch_hardware_enable(NULL);
2680 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2681 atomic_inc(&hardware_enable_failed);
2682 printk(KERN_INFO "kvm: enabling virtualization on "
2683 "CPU%d failed\n", cpu);
2687 static void hardware_enable(void)
2689 raw_spin_lock(&kvm_count_lock);
2690 if (kvm_usage_count)
2691 hardware_enable_nolock(NULL);
2692 raw_spin_unlock(&kvm_count_lock);
2695 static void hardware_disable_nolock(void *junk)
2697 int cpu = raw_smp_processor_id();
2699 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2701 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2702 kvm_arch_hardware_disable(NULL);
2705 static void hardware_disable(void)
2707 raw_spin_lock(&kvm_count_lock);
2708 if (kvm_usage_count)
2709 hardware_disable_nolock(NULL);
2710 raw_spin_unlock(&kvm_count_lock);
2713 static void hardware_disable_all_nolock(void)
2715 BUG_ON(!kvm_usage_count);
2718 if (!kvm_usage_count)
2719 on_each_cpu(hardware_disable_nolock, NULL, 1);
2722 static void hardware_disable_all(void)
2724 raw_spin_lock(&kvm_count_lock);
2725 hardware_disable_all_nolock();
2726 raw_spin_unlock(&kvm_count_lock);
2729 static int hardware_enable_all(void)
2733 raw_spin_lock(&kvm_count_lock);
2736 if (kvm_usage_count == 1) {
2737 atomic_set(&hardware_enable_failed, 0);
2738 on_each_cpu(hardware_enable_nolock, NULL, 1);
2740 if (atomic_read(&hardware_enable_failed)) {
2741 hardware_disable_all_nolock();
2746 raw_spin_unlock(&kvm_count_lock);
2751 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2756 val &= ~CPU_TASKS_FROZEN;
2759 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2764 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2772 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2776 * Some (well, at least mine) BIOSes hang on reboot if
2779 * And Intel TXT required VMX off for all cpu when system shutdown.
2781 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2782 kvm_rebooting = true;
2783 on_each_cpu(hardware_disable_nolock, NULL, 1);
2787 static struct notifier_block kvm_reboot_notifier = {
2788 .notifier_call = kvm_reboot,
2792 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2796 for (i = 0; i < bus->dev_count; i++) {
2797 struct kvm_io_device *pos = bus->range[i].dev;
2799 kvm_iodevice_destructor(pos);
2804 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
2805 const struct kvm_io_range *r2)
2807 if (r1->addr < r2->addr)
2809 if (r1->addr + r1->len > r2->addr + r2->len)
2814 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2816 return kvm_io_bus_cmp(p1, p2);
2819 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2820 gpa_t addr, int len)
2822 bus->range[bus->dev_count++] = (struct kvm_io_range) {
2828 sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2829 kvm_io_bus_sort_cmp, NULL);
2834 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2835 gpa_t addr, int len)
2837 struct kvm_io_range *range, key;
2840 key = (struct kvm_io_range) {
2845 range = bsearch(&key, bus->range, bus->dev_count,
2846 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2850 off = range - bus->range;
2852 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
2858 static int __kvm_io_bus_write(struct kvm_io_bus *bus,
2859 struct kvm_io_range *range, const void *val)
2863 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2867 while (idx < bus->dev_count &&
2868 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
2869 if (!kvm_iodevice_write(bus->range[idx].dev, range->addr,
2878 /* kvm_io_bus_write - called under kvm->slots_lock */
2879 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2880 int len, const void *val)
2882 struct kvm_io_bus *bus;
2883 struct kvm_io_range range;
2886 range = (struct kvm_io_range) {
2891 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2892 r = __kvm_io_bus_write(bus, &range, val);
2893 return r < 0 ? r : 0;
2896 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
2897 int kvm_io_bus_write_cookie(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2898 int len, const void *val, long cookie)
2900 struct kvm_io_bus *bus;
2901 struct kvm_io_range range;
2903 range = (struct kvm_io_range) {
2908 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2910 /* First try the device referenced by cookie. */
2911 if ((cookie >= 0) && (cookie < bus->dev_count) &&
2912 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
2913 if (!kvm_iodevice_write(bus->range[cookie].dev, addr, len,
2918 * cookie contained garbage; fall back to search and return the
2919 * correct cookie value.
2921 return __kvm_io_bus_write(bus, &range, val);
2924 static int __kvm_io_bus_read(struct kvm_io_bus *bus, struct kvm_io_range *range,
2929 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2933 while (idx < bus->dev_count &&
2934 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
2935 if (!kvm_iodevice_read(bus->range[idx].dev, range->addr,
2944 /* kvm_io_bus_read - called under kvm->slots_lock */
2945 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2948 struct kvm_io_bus *bus;
2949 struct kvm_io_range range;
2952 range = (struct kvm_io_range) {
2957 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2958 r = __kvm_io_bus_read(bus, &range, val);
2959 return r < 0 ? r : 0;
2962 /* kvm_io_bus_read_cookie - called under kvm->slots_lock */
2963 int kvm_io_bus_read_cookie(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2964 int len, void *val, long cookie)
2966 struct kvm_io_bus *bus;
2967 struct kvm_io_range range;
2969 range = (struct kvm_io_range) {
2974 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2976 /* First try the device referenced by cookie. */
2977 if ((cookie >= 0) && (cookie < bus->dev_count) &&
2978 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
2979 if (!kvm_iodevice_read(bus->range[cookie].dev, addr, len,
2984 * cookie contained garbage; fall back to search and return the
2985 * correct cookie value.
2987 return __kvm_io_bus_read(bus, &range, val);
2990 /* Caller must hold slots_lock. */
2991 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2992 int len, struct kvm_io_device *dev)
2994 struct kvm_io_bus *new_bus, *bus;
2996 bus = kvm->buses[bus_idx];
2997 /* exclude ioeventfd which is limited by maximum fd */
2998 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3001 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
3002 sizeof(struct kvm_io_range)), GFP_KERNEL);
3005 memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
3006 sizeof(struct kvm_io_range)));
3007 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
3008 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3009 synchronize_srcu_expedited(&kvm->srcu);
3015 /* Caller must hold slots_lock. */
3016 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3017 struct kvm_io_device *dev)
3020 struct kvm_io_bus *new_bus, *bus;
3022 bus = kvm->buses[bus_idx];
3024 for (i = 0; i < bus->dev_count; i++)
3025 if (bus->range[i].dev == dev) {
3033 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3034 sizeof(struct kvm_io_range)), GFP_KERNEL);
3038 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3039 new_bus->dev_count--;
3040 memcpy(new_bus->range + i, bus->range + i + 1,
3041 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3043 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3044 synchronize_srcu_expedited(&kvm->srcu);
3049 static struct notifier_block kvm_cpu_notifier = {
3050 .notifier_call = kvm_cpu_hotplug,
3053 static int vm_stat_get(void *_offset, u64 *val)
3055 unsigned offset = (long)_offset;
3059 spin_lock(&kvm_lock);
3060 list_for_each_entry(kvm, &vm_list, vm_list)
3061 *val += *(u32 *)((void *)kvm + offset);
3062 spin_unlock(&kvm_lock);
3066 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
3068 static int vcpu_stat_get(void *_offset, u64 *val)
3070 unsigned offset = (long)_offset;
3072 struct kvm_vcpu *vcpu;
3076 spin_lock(&kvm_lock);
3077 list_for_each_entry(kvm, &vm_list, vm_list)
3078 kvm_for_each_vcpu(i, vcpu, kvm)
3079 *val += *(u32 *)((void *)vcpu + offset);
3081 spin_unlock(&kvm_lock);
3085 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
3087 static const struct file_operations *stat_fops[] = {
3088 [KVM_STAT_VCPU] = &vcpu_stat_fops,
3089 [KVM_STAT_VM] = &vm_stat_fops,
3092 static int kvm_init_debug(void)
3095 struct kvm_stats_debugfs_item *p;
3097 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3098 if (kvm_debugfs_dir == NULL)
3101 for (p = debugfs_entries; p->name; ++p) {
3102 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
3103 (void *)(long)p->offset,
3104 stat_fops[p->kind]);
3105 if (p->dentry == NULL)
3112 debugfs_remove_recursive(kvm_debugfs_dir);
3117 static void kvm_exit_debug(void)
3119 struct kvm_stats_debugfs_item *p;
3121 for (p = debugfs_entries; p->name; ++p)
3122 debugfs_remove(p->dentry);
3123 debugfs_remove(kvm_debugfs_dir);
3126 static int kvm_suspend(void)
3128 if (kvm_usage_count)
3129 hardware_disable_nolock(NULL);
3133 static void kvm_resume(void)
3135 if (kvm_usage_count) {
3136 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3137 hardware_enable_nolock(NULL);
3141 static struct syscore_ops kvm_syscore_ops = {
3142 .suspend = kvm_suspend,
3143 .resume = kvm_resume,
3147 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3149 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3152 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3154 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3155 if (vcpu->preempted)
3156 vcpu->preempted = false;
3158 kvm_arch_vcpu_load(vcpu, cpu);
3161 static void kvm_sched_out(struct preempt_notifier *pn,
3162 struct task_struct *next)
3164 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3166 if (current->state == TASK_RUNNING)
3167 vcpu->preempted = true;
3168 kvm_arch_vcpu_put(vcpu);
3171 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3172 struct module *module)
3177 r = kvm_arch_init(opaque);
3182 * kvm_arch_init makes sure there's at most one caller
3183 * for architectures that support multiple implementations,
3184 * like intel and amd on x86.
3185 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3186 * conflicts in case kvm is already setup for another implementation.
3188 r = kvm_irqfd_init();
3192 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
3197 r = kvm_arch_hardware_setup();
3201 for_each_online_cpu(cpu) {
3202 smp_call_function_single(cpu,
3203 kvm_arch_check_processor_compat,
3209 r = register_cpu_notifier(&kvm_cpu_notifier);
3212 register_reboot_notifier(&kvm_reboot_notifier);
3214 /* A kmem cache lets us meet the alignment requirements of fx_save. */
3216 vcpu_align = __alignof__(struct kvm_vcpu);
3217 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
3219 if (!kvm_vcpu_cache) {
3224 r = kvm_async_pf_init();
3228 kvm_chardev_ops.owner = module;
3229 kvm_vm_fops.owner = module;
3230 kvm_vcpu_fops.owner = module;
3232 r = misc_register(&kvm_dev);
3234 printk(KERN_ERR "kvm: misc device register failed\n");
3238 register_syscore_ops(&kvm_syscore_ops);
3240 kvm_preempt_ops.sched_in = kvm_sched_in;
3241 kvm_preempt_ops.sched_out = kvm_sched_out;
3243 r = kvm_init_debug();
3245 printk(KERN_ERR "kvm: create debugfs files failed\n");
3252 unregister_syscore_ops(&kvm_syscore_ops);
3253 misc_deregister(&kvm_dev);
3255 kvm_async_pf_deinit();
3257 kmem_cache_destroy(kvm_vcpu_cache);
3259 unregister_reboot_notifier(&kvm_reboot_notifier);
3260 unregister_cpu_notifier(&kvm_cpu_notifier);
3263 kvm_arch_hardware_unsetup();
3265 free_cpumask_var(cpus_hardware_enabled);
3273 EXPORT_SYMBOL_GPL(kvm_init);
3278 misc_deregister(&kvm_dev);
3279 kmem_cache_destroy(kvm_vcpu_cache);
3280 kvm_async_pf_deinit();
3281 unregister_syscore_ops(&kvm_syscore_ops);
3282 unregister_reboot_notifier(&kvm_reboot_notifier);
3283 unregister_cpu_notifier(&kvm_cpu_notifier);
3284 on_each_cpu(hardware_disable_nolock, NULL, 1);
3285 kvm_arch_hardware_unsetup();
3288 free_cpumask_var(cpus_hardware_enabled);
3290 EXPORT_SYMBOL_GPL(kvm_exit);