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_RAW_SPINLOCK(kvm_lock);
76 static cpumask_var_t cpus_hardware_enabled;
77 static int kvm_usage_count = 0;
78 static atomic_t hardware_enable_failed;
80 struct kmem_cache *kvm_vcpu_cache;
81 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
83 static __read_mostly struct preempt_ops kvm_preempt_ops;
85 struct dentry *kvm_debugfs_dir;
87 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
90 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
93 static int hardware_enable_all(void);
94 static void hardware_disable_all(void);
96 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
99 EXPORT_SYMBOL_GPL(kvm_rebooting);
101 static bool largepages_enabled = true;
103 bool kvm_is_mmio_pfn(pfn_t pfn)
106 return PageReserved(pfn_to_page(pfn));
112 * Switches to specified vcpu, until a matching vcpu_put()
114 int vcpu_load(struct kvm_vcpu *vcpu)
118 if (mutex_lock_killable(&vcpu->mutex))
120 if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
121 /* The thread running this VCPU changed. */
122 struct pid *oldpid = vcpu->pid;
123 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
124 rcu_assign_pointer(vcpu->pid, newpid);
129 preempt_notifier_register(&vcpu->preempt_notifier);
130 kvm_arch_vcpu_load(vcpu, cpu);
135 void vcpu_put(struct kvm_vcpu *vcpu)
138 kvm_arch_vcpu_put(vcpu);
139 preempt_notifier_unregister(&vcpu->preempt_notifier);
141 mutex_unlock(&vcpu->mutex);
144 static void ack_flush(void *_completed)
148 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
153 struct kvm_vcpu *vcpu;
155 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
158 kvm_for_each_vcpu(i, vcpu, kvm) {
159 kvm_make_request(req, vcpu);
162 /* Set ->requests bit before we read ->mode */
165 if (cpus != NULL && cpu != -1 && cpu != me &&
166 kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
167 cpumask_set_cpu(cpu, cpus);
169 if (unlikely(cpus == NULL))
170 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
171 else if (!cpumask_empty(cpus))
172 smp_call_function_many(cpus, ack_flush, NULL, 1);
176 free_cpumask_var(cpus);
180 void kvm_flush_remote_tlbs(struct kvm *kvm)
182 long dirty_count = kvm->tlbs_dirty;
185 if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
186 ++kvm->stat.remote_tlb_flush;
187 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
190 void kvm_reload_remote_mmus(struct kvm *kvm)
192 make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
195 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
197 make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
200 void kvm_make_scan_ioapic_request(struct kvm *kvm)
202 make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
205 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
210 mutex_init(&vcpu->mutex);
215 init_waitqueue_head(&vcpu->wq);
216 kvm_async_pf_vcpu_init(vcpu);
218 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
223 vcpu->run = page_address(page);
225 kvm_vcpu_set_in_spin_loop(vcpu, false);
226 kvm_vcpu_set_dy_eligible(vcpu, false);
227 vcpu->preempted = false;
229 r = kvm_arch_vcpu_init(vcpu);
235 free_page((unsigned long)vcpu->run);
239 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
241 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
244 kvm_arch_vcpu_uninit(vcpu);
245 free_page((unsigned long)vcpu->run);
247 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
249 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
250 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
252 return container_of(mn, struct kvm, mmu_notifier);
255 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
256 struct mm_struct *mm,
257 unsigned long address)
259 struct kvm *kvm = mmu_notifier_to_kvm(mn);
260 int need_tlb_flush, idx;
263 * When ->invalidate_page runs, the linux pte has been zapped
264 * already but the page is still allocated until
265 * ->invalidate_page returns. So if we increase the sequence
266 * here the kvm page fault will notice if the spte can't be
267 * established because the page is going to be freed. If
268 * instead the kvm page fault establishes the spte before
269 * ->invalidate_page runs, kvm_unmap_hva will release it
272 * The sequence increase only need to be seen at spin_unlock
273 * time, and not at spin_lock time.
275 * Increasing the sequence after the spin_unlock would be
276 * unsafe because the kvm page fault could then establish the
277 * pte after kvm_unmap_hva returned, without noticing the page
278 * is going to be freed.
280 idx = srcu_read_lock(&kvm->srcu);
281 spin_lock(&kvm->mmu_lock);
283 kvm->mmu_notifier_seq++;
284 need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
285 /* we've to flush the tlb before the pages can be freed */
287 kvm_flush_remote_tlbs(kvm);
289 spin_unlock(&kvm->mmu_lock);
290 srcu_read_unlock(&kvm->srcu, idx);
293 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
294 struct mm_struct *mm,
295 unsigned long address,
298 struct kvm *kvm = mmu_notifier_to_kvm(mn);
301 idx = srcu_read_lock(&kvm->srcu);
302 spin_lock(&kvm->mmu_lock);
303 kvm->mmu_notifier_seq++;
304 kvm_set_spte_hva(kvm, address, pte);
305 spin_unlock(&kvm->mmu_lock);
306 srcu_read_unlock(&kvm->srcu, idx);
309 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
310 struct mm_struct *mm,
314 struct kvm *kvm = mmu_notifier_to_kvm(mn);
315 int need_tlb_flush = 0, idx;
317 idx = srcu_read_lock(&kvm->srcu);
318 spin_lock(&kvm->mmu_lock);
320 * The count increase must become visible at unlock time as no
321 * spte can be established without taking the mmu_lock and
322 * count is also read inside the mmu_lock critical section.
324 kvm->mmu_notifier_count++;
325 need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
326 need_tlb_flush |= kvm->tlbs_dirty;
327 /* we've to flush the tlb before the pages can be freed */
329 kvm_flush_remote_tlbs(kvm);
331 spin_unlock(&kvm->mmu_lock);
332 srcu_read_unlock(&kvm->srcu, idx);
335 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
336 struct mm_struct *mm,
340 struct kvm *kvm = mmu_notifier_to_kvm(mn);
342 spin_lock(&kvm->mmu_lock);
344 * This sequence increase will notify the kvm page fault that
345 * the page that is going to be mapped in the spte could have
348 kvm->mmu_notifier_seq++;
351 * The above sequence increase must be visible before the
352 * below count decrease, which is ensured by the smp_wmb above
353 * in conjunction with the smp_rmb in mmu_notifier_retry().
355 kvm->mmu_notifier_count--;
356 spin_unlock(&kvm->mmu_lock);
358 BUG_ON(kvm->mmu_notifier_count < 0);
361 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
362 struct mm_struct *mm,
363 unsigned long address)
365 struct kvm *kvm = mmu_notifier_to_kvm(mn);
368 idx = srcu_read_lock(&kvm->srcu);
369 spin_lock(&kvm->mmu_lock);
371 young = kvm_age_hva(kvm, address);
373 kvm_flush_remote_tlbs(kvm);
375 spin_unlock(&kvm->mmu_lock);
376 srcu_read_unlock(&kvm->srcu, idx);
381 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
382 struct mm_struct *mm,
383 unsigned long address)
385 struct kvm *kvm = mmu_notifier_to_kvm(mn);
388 idx = srcu_read_lock(&kvm->srcu);
389 spin_lock(&kvm->mmu_lock);
390 young = kvm_test_age_hva(kvm, address);
391 spin_unlock(&kvm->mmu_lock);
392 srcu_read_unlock(&kvm->srcu, idx);
397 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
398 struct mm_struct *mm)
400 struct kvm *kvm = mmu_notifier_to_kvm(mn);
403 idx = srcu_read_lock(&kvm->srcu);
404 kvm_arch_flush_shadow_all(kvm);
405 srcu_read_unlock(&kvm->srcu, idx);
408 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
409 .invalidate_page = kvm_mmu_notifier_invalidate_page,
410 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
411 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
412 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
413 .test_young = kvm_mmu_notifier_test_young,
414 .change_pte = kvm_mmu_notifier_change_pte,
415 .release = kvm_mmu_notifier_release,
418 static int kvm_init_mmu_notifier(struct kvm *kvm)
420 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
421 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
424 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
426 static int kvm_init_mmu_notifier(struct kvm *kvm)
431 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
433 static void kvm_init_memslots_id(struct kvm *kvm)
436 struct kvm_memslots *slots = kvm->memslots;
438 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
439 slots->id_to_index[i] = slots->memslots[i].id = i;
442 static struct kvm *kvm_create_vm(unsigned long type)
445 struct kvm *kvm = kvm_arch_alloc_vm();
448 return ERR_PTR(-ENOMEM);
450 r = kvm_arch_init_vm(kvm, type);
452 goto out_err_nodisable;
454 r = hardware_enable_all();
456 goto out_err_nodisable;
458 #ifdef CONFIG_HAVE_KVM_IRQCHIP
459 INIT_HLIST_HEAD(&kvm->mask_notifier_list);
460 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
463 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
466 kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
469 kvm_init_memslots_id(kvm);
470 if (init_srcu_struct(&kvm->srcu))
472 for (i = 0; i < KVM_NR_BUSES; i++) {
473 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
479 spin_lock_init(&kvm->mmu_lock);
480 kvm->mm = current->mm;
481 atomic_inc(&kvm->mm->mm_count);
482 kvm_eventfd_init(kvm);
483 mutex_init(&kvm->lock);
484 mutex_init(&kvm->irq_lock);
485 mutex_init(&kvm->slots_lock);
486 atomic_set(&kvm->users_count, 1);
487 INIT_LIST_HEAD(&kvm->devices);
489 r = kvm_init_mmu_notifier(kvm);
493 raw_spin_lock(&kvm_lock);
494 list_add(&kvm->vm_list, &vm_list);
495 raw_spin_unlock(&kvm_lock);
500 cleanup_srcu_struct(&kvm->srcu);
502 hardware_disable_all();
504 for (i = 0; i < KVM_NR_BUSES; i++)
505 kfree(kvm->buses[i]);
506 kfree(kvm->memslots);
507 kvm_arch_free_vm(kvm);
512 * Avoid using vmalloc for a small buffer.
513 * Should not be used when the size is statically known.
515 void *kvm_kvzalloc(unsigned long size)
517 if (size > PAGE_SIZE)
518 return vzalloc(size);
520 return kzalloc(size, GFP_KERNEL);
523 void kvm_kvfree(const void *addr)
525 if (is_vmalloc_addr(addr))
531 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
533 if (!memslot->dirty_bitmap)
536 kvm_kvfree(memslot->dirty_bitmap);
537 memslot->dirty_bitmap = NULL;
541 * Free any memory in @free but not in @dont.
543 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
544 struct kvm_memory_slot *dont)
546 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
547 kvm_destroy_dirty_bitmap(free);
549 kvm_arch_free_memslot(free, dont);
554 void kvm_free_physmem(struct kvm *kvm)
556 struct kvm_memslots *slots = kvm->memslots;
557 struct kvm_memory_slot *memslot;
559 kvm_for_each_memslot(memslot, slots)
560 kvm_free_physmem_slot(memslot, NULL);
562 kfree(kvm->memslots);
565 static void kvm_destroy_devices(struct kvm *kvm)
567 struct list_head *node, *tmp;
569 list_for_each_safe(node, tmp, &kvm->devices) {
570 struct kvm_device *dev =
571 list_entry(node, struct kvm_device, vm_node);
574 dev->ops->destroy(dev);
578 static void kvm_destroy_vm(struct kvm *kvm)
581 struct mm_struct *mm = kvm->mm;
583 kvm_arch_sync_events(kvm);
584 raw_spin_lock(&kvm_lock);
585 list_del(&kvm->vm_list);
586 raw_spin_unlock(&kvm_lock);
587 kvm_free_irq_routing(kvm);
588 for (i = 0; i < KVM_NR_BUSES; i++)
589 kvm_io_bus_destroy(kvm->buses[i]);
590 kvm_coalesced_mmio_free(kvm);
591 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
592 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
594 kvm_arch_flush_shadow_all(kvm);
596 kvm_arch_destroy_vm(kvm);
597 kvm_destroy_devices(kvm);
598 kvm_free_physmem(kvm);
599 cleanup_srcu_struct(&kvm->srcu);
600 kvm_arch_free_vm(kvm);
601 hardware_disable_all();
605 void kvm_get_kvm(struct kvm *kvm)
607 atomic_inc(&kvm->users_count);
609 EXPORT_SYMBOL_GPL(kvm_get_kvm);
611 void kvm_put_kvm(struct kvm *kvm)
613 if (atomic_dec_and_test(&kvm->users_count))
616 EXPORT_SYMBOL_GPL(kvm_put_kvm);
619 static int kvm_vm_release(struct inode *inode, struct file *filp)
621 struct kvm *kvm = filp->private_data;
623 kvm_irqfd_release(kvm);
630 * Allocation size is twice as large as the actual dirty bitmap size.
631 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
633 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
636 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
638 memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
639 if (!memslot->dirty_bitmap)
642 #endif /* !CONFIG_S390 */
646 static int cmp_memslot(const void *slot1, const void *slot2)
648 struct kvm_memory_slot *s1, *s2;
650 s1 = (struct kvm_memory_slot *)slot1;
651 s2 = (struct kvm_memory_slot *)slot2;
653 if (s1->npages < s2->npages)
655 if (s1->npages > s2->npages)
662 * Sort the memslots base on its size, so the larger slots
663 * will get better fit.
665 static void sort_memslots(struct kvm_memslots *slots)
669 sort(slots->memslots, KVM_MEM_SLOTS_NUM,
670 sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
672 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
673 slots->id_to_index[slots->memslots[i].id] = i;
676 void update_memslots(struct kvm_memslots *slots, struct kvm_memory_slot *new,
681 struct kvm_memory_slot *old = id_to_memslot(slots, id);
682 unsigned long npages = old->npages;
685 if (new->npages != npages)
686 sort_memslots(slots);
689 slots->generation = last_generation + 1;
692 static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
694 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
696 #ifdef KVM_CAP_READONLY_MEM
697 valid_flags |= KVM_MEM_READONLY;
700 if (mem->flags & ~valid_flags)
706 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
707 struct kvm_memslots *slots, struct kvm_memory_slot *new)
709 struct kvm_memslots *old_memslots = kvm->memslots;
711 update_memslots(slots, new, kvm->memslots->generation);
712 rcu_assign_pointer(kvm->memslots, slots);
713 synchronize_srcu_expedited(&kvm->srcu);
715 kvm_arch_memslots_updated(kvm);
721 * Allocate some memory and give it an address in the guest physical address
724 * Discontiguous memory is allowed, mostly for framebuffers.
726 * Must be called holding mmap_sem for write.
728 int __kvm_set_memory_region(struct kvm *kvm,
729 struct kvm_userspace_memory_region *mem)
733 unsigned long npages;
734 struct kvm_memory_slot *slot;
735 struct kvm_memory_slot old, new;
736 struct kvm_memslots *slots = NULL, *old_memslots;
737 enum kvm_mr_change change;
739 r = check_memory_region_flags(mem);
744 /* General sanity checks */
745 if (mem->memory_size & (PAGE_SIZE - 1))
747 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
749 /* We can read the guest memory with __xxx_user() later on. */
750 if ((mem->slot < KVM_USER_MEM_SLOTS) &&
751 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
752 !access_ok(VERIFY_WRITE,
753 (void __user *)(unsigned long)mem->userspace_addr,
756 if (mem->slot >= KVM_MEM_SLOTS_NUM)
758 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
761 slot = id_to_memslot(kvm->memslots, mem->slot);
762 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
763 npages = mem->memory_size >> PAGE_SHIFT;
766 if (npages > KVM_MEM_MAX_NR_PAGES)
770 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
775 new.base_gfn = base_gfn;
777 new.flags = mem->flags;
782 change = KVM_MR_CREATE;
783 else { /* Modify an existing slot. */
784 if ((mem->userspace_addr != old.userspace_addr) ||
785 (npages != old.npages) ||
786 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
789 if (base_gfn != old.base_gfn)
790 change = KVM_MR_MOVE;
791 else if (new.flags != old.flags)
792 change = KVM_MR_FLAGS_ONLY;
793 else { /* Nothing to change. */
798 } else if (old.npages) {
799 change = KVM_MR_DELETE;
800 } else /* Modify a non-existent slot: disallowed. */
803 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
804 /* Check for overlaps */
806 kvm_for_each_memslot(slot, kvm->memslots) {
807 if ((slot->id >= KVM_USER_MEM_SLOTS) ||
808 (slot->id == mem->slot))
810 if (!((base_gfn + npages <= slot->base_gfn) ||
811 (base_gfn >= slot->base_gfn + slot->npages)))
816 /* Free page dirty bitmap if unneeded */
817 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
818 new.dirty_bitmap = NULL;
821 if (change == KVM_MR_CREATE) {
822 new.userspace_addr = mem->userspace_addr;
824 if (kvm_arch_create_memslot(&new, npages))
828 /* Allocate page dirty bitmap if needed */
829 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
830 if (kvm_create_dirty_bitmap(&new) < 0)
834 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
836 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
840 slot = id_to_memslot(slots, mem->slot);
841 slot->flags |= KVM_MEMSLOT_INVALID;
843 old_memslots = install_new_memslots(kvm, slots, NULL);
845 /* slot was deleted or moved, clear iommu mapping */
846 kvm_iommu_unmap_pages(kvm, &old);
847 /* From this point no new shadow pages pointing to a deleted,
848 * or moved, memslot will be created.
850 * validation of sp->gfn happens in:
851 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
852 * - kvm_is_visible_gfn (mmu_check_roots)
854 kvm_arch_flush_shadow_memslot(kvm, slot);
855 slots = old_memslots;
858 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
864 * We can re-use the old_memslots from above, the only difference
865 * from the currently installed memslots is the invalid flag. This
866 * will get overwritten by update_memslots anyway.
869 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
876 * IOMMU mapping: New slots need to be mapped. Old slots need to be
877 * un-mapped and re-mapped if their base changes. Since base change
878 * unmapping is handled above with slot deletion, mapping alone is
879 * needed here. Anything else the iommu might care about for existing
880 * slots (size changes, userspace addr changes and read-only flag
881 * changes) is disallowed above, so any other attribute changes getting
882 * here can be skipped.
884 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
885 r = kvm_iommu_map_pages(kvm, &new);
890 /* actual memory is freed via old in kvm_free_physmem_slot below */
891 if (change == KVM_MR_DELETE) {
892 new.dirty_bitmap = NULL;
893 memset(&new.arch, 0, sizeof(new.arch));
896 old_memslots = install_new_memslots(kvm, slots, &new);
898 kvm_arch_commit_memory_region(kvm, mem, &old, change);
900 kvm_free_physmem_slot(&old, &new);
908 kvm_free_physmem_slot(&new, &old);
912 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
914 int kvm_set_memory_region(struct kvm *kvm,
915 struct kvm_userspace_memory_region *mem)
919 mutex_lock(&kvm->slots_lock);
920 r = __kvm_set_memory_region(kvm, mem);
921 mutex_unlock(&kvm->slots_lock);
924 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
926 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
927 struct kvm_userspace_memory_region *mem)
929 if (mem->slot >= KVM_USER_MEM_SLOTS)
931 return kvm_set_memory_region(kvm, mem);
934 int kvm_get_dirty_log(struct kvm *kvm,
935 struct kvm_dirty_log *log, int *is_dirty)
937 struct kvm_memory_slot *memslot;
940 unsigned long any = 0;
943 if (log->slot >= KVM_USER_MEM_SLOTS)
946 memslot = id_to_memslot(kvm->memslots, log->slot);
948 if (!memslot->dirty_bitmap)
951 n = kvm_dirty_bitmap_bytes(memslot);
953 for (i = 0; !any && i < n/sizeof(long); ++i)
954 any = memslot->dirty_bitmap[i];
957 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
968 bool kvm_largepages_enabled(void)
970 return largepages_enabled;
973 void kvm_disable_largepages(void)
975 largepages_enabled = false;
977 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
979 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
981 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
983 EXPORT_SYMBOL_GPL(gfn_to_memslot);
985 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
987 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
989 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
990 memslot->flags & KVM_MEMSLOT_INVALID)
995 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
997 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
999 struct vm_area_struct *vma;
1000 unsigned long addr, size;
1004 addr = gfn_to_hva(kvm, gfn);
1005 if (kvm_is_error_hva(addr))
1008 down_read(¤t->mm->mmap_sem);
1009 vma = find_vma(current->mm, addr);
1013 size = vma_kernel_pagesize(vma);
1016 up_read(¤t->mm->mmap_sem);
1021 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1023 return slot->flags & KVM_MEM_READONLY;
1026 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1027 gfn_t *nr_pages, bool write)
1029 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1030 return KVM_HVA_ERR_BAD;
1032 if (memslot_is_readonly(slot) && write)
1033 return KVM_HVA_ERR_RO_BAD;
1036 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1038 return __gfn_to_hva_memslot(slot, gfn);
1041 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1044 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1047 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1050 return gfn_to_hva_many(slot, gfn, NULL);
1052 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1054 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1056 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1058 EXPORT_SYMBOL_GPL(gfn_to_hva);
1061 * If writable is set to false, the hva returned by this function is only
1062 * allowed to be read.
1064 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1066 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1067 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1069 if (!kvm_is_error_hva(hva) && writable)
1070 *writable = !memslot_is_readonly(slot);
1075 static int kvm_read_hva(void *data, void __user *hva, int len)
1077 return __copy_from_user(data, hva, len);
1080 static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1082 return __copy_from_user_inatomic(data, hva, len);
1085 static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1086 unsigned long start, int write, struct page **page)
1088 int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1091 flags |= FOLL_WRITE;
1093 return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1096 static inline int check_user_page_hwpoison(unsigned long addr)
1098 int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1100 rc = __get_user_pages(current, current->mm, addr, 1,
1101 flags, NULL, NULL, NULL);
1102 return rc == -EHWPOISON;
1106 * The atomic path to get the writable pfn which will be stored in @pfn,
1107 * true indicates success, otherwise false is returned.
1109 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1110 bool write_fault, bool *writable, pfn_t *pfn)
1112 struct page *page[1];
1115 if (!(async || atomic))
1119 * Fast pin a writable pfn only if it is a write fault request
1120 * or the caller allows to map a writable pfn for a read fault
1123 if (!(write_fault || writable))
1126 npages = __get_user_pages_fast(addr, 1, 1, page);
1128 *pfn = page_to_pfn(page[0]);
1139 * The slow path to get the pfn of the specified host virtual address,
1140 * 1 indicates success, -errno is returned if error is detected.
1142 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1143 bool *writable, pfn_t *pfn)
1145 struct page *page[1];
1151 *writable = write_fault;
1154 down_read(¤t->mm->mmap_sem);
1155 npages = get_user_page_nowait(current, current->mm,
1156 addr, write_fault, page);
1157 up_read(¤t->mm->mmap_sem);
1159 npages = get_user_pages_fast(addr, 1, write_fault,
1164 /* map read fault as writable if possible */
1165 if (unlikely(!write_fault) && writable) {
1166 struct page *wpage[1];
1168 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1177 *pfn = page_to_pfn(page[0]);
1181 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1183 if (unlikely(!(vma->vm_flags & VM_READ)))
1186 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1193 * Pin guest page in memory and return its pfn.
1194 * @addr: host virtual address which maps memory to the guest
1195 * @atomic: whether this function can sleep
1196 * @async: whether this function need to wait IO complete if the
1197 * host page is not in the memory
1198 * @write_fault: whether we should get a writable host page
1199 * @writable: whether it allows to map a writable host page for !@write_fault
1201 * The function will map a writable host page for these two cases:
1202 * 1): @write_fault = true
1203 * 2): @write_fault = false && @writable, @writable will tell the caller
1204 * whether the mapping is writable.
1206 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1207 bool write_fault, bool *writable)
1209 struct vm_area_struct *vma;
1213 /* we can do it either atomically or asynchronously, not both */
1214 BUG_ON(atomic && async);
1216 if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1220 return KVM_PFN_ERR_FAULT;
1222 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1226 down_read(¤t->mm->mmap_sem);
1227 if (npages == -EHWPOISON ||
1228 (!async && check_user_page_hwpoison(addr))) {
1229 pfn = KVM_PFN_ERR_HWPOISON;
1233 vma = find_vma_intersection(current->mm, addr, addr + 1);
1236 pfn = KVM_PFN_ERR_FAULT;
1237 else if ((vma->vm_flags & VM_PFNMAP)) {
1238 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1240 BUG_ON(!kvm_is_mmio_pfn(pfn));
1242 if (async && vma_is_valid(vma, write_fault))
1244 pfn = KVM_PFN_ERR_FAULT;
1247 up_read(¤t->mm->mmap_sem);
1252 __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1253 bool *async, bool write_fault, bool *writable)
1255 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1257 if (addr == KVM_HVA_ERR_RO_BAD)
1258 return KVM_PFN_ERR_RO_FAULT;
1260 if (kvm_is_error_hva(addr))
1261 return KVM_PFN_NOSLOT;
1263 /* Do not map writable pfn in the readonly memslot. */
1264 if (writable && memslot_is_readonly(slot)) {
1269 return hva_to_pfn(addr, atomic, async, write_fault,
1273 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1274 bool write_fault, bool *writable)
1276 struct kvm_memory_slot *slot;
1281 slot = gfn_to_memslot(kvm, gfn);
1283 return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1287 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1289 return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1291 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1293 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1294 bool write_fault, bool *writable)
1296 return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1298 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1300 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1302 return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1304 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1306 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1309 return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1311 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1313 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1315 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1318 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1320 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1322 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1324 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1330 addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1331 if (kvm_is_error_hva(addr))
1334 if (entry < nr_pages)
1337 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1339 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1341 static struct page *kvm_pfn_to_page(pfn_t pfn)
1343 if (is_error_noslot_pfn(pfn))
1344 return KVM_ERR_PTR_BAD_PAGE;
1346 if (kvm_is_mmio_pfn(pfn)) {
1348 return KVM_ERR_PTR_BAD_PAGE;
1351 return pfn_to_page(pfn);
1354 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1358 pfn = gfn_to_pfn(kvm, gfn);
1360 return kvm_pfn_to_page(pfn);
1363 EXPORT_SYMBOL_GPL(gfn_to_page);
1365 void kvm_release_page_clean(struct page *page)
1367 WARN_ON(is_error_page(page));
1369 kvm_release_pfn_clean(page_to_pfn(page));
1371 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1373 void kvm_release_pfn_clean(pfn_t pfn)
1375 if (!is_error_noslot_pfn(pfn) && !kvm_is_mmio_pfn(pfn))
1376 put_page(pfn_to_page(pfn));
1378 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1380 void kvm_release_page_dirty(struct page *page)
1382 WARN_ON(is_error_page(page));
1384 kvm_release_pfn_dirty(page_to_pfn(page));
1386 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1388 void kvm_release_pfn_dirty(pfn_t pfn)
1390 kvm_set_pfn_dirty(pfn);
1391 kvm_release_pfn_clean(pfn);
1393 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1395 void kvm_set_page_dirty(struct page *page)
1397 kvm_set_pfn_dirty(page_to_pfn(page));
1399 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1401 void kvm_set_pfn_dirty(pfn_t pfn)
1403 if (!kvm_is_mmio_pfn(pfn)) {
1404 struct page *page = pfn_to_page(pfn);
1405 if (!PageReserved(page))
1409 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1411 void kvm_set_pfn_accessed(pfn_t pfn)
1413 if (!kvm_is_mmio_pfn(pfn))
1414 mark_page_accessed(pfn_to_page(pfn));
1416 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1418 void kvm_get_pfn(pfn_t pfn)
1420 if (!kvm_is_mmio_pfn(pfn))
1421 get_page(pfn_to_page(pfn));
1423 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1425 static int next_segment(unsigned long len, int offset)
1427 if (len > PAGE_SIZE - offset)
1428 return PAGE_SIZE - offset;
1433 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1439 addr = gfn_to_hva_prot(kvm, gfn, NULL);
1440 if (kvm_is_error_hva(addr))
1442 r = kvm_read_hva(data, (void __user *)addr + offset, len);
1447 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1449 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1451 gfn_t gfn = gpa >> PAGE_SHIFT;
1453 int offset = offset_in_page(gpa);
1456 while ((seg = next_segment(len, offset)) != 0) {
1457 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1467 EXPORT_SYMBOL_GPL(kvm_read_guest);
1469 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1474 gfn_t gfn = gpa >> PAGE_SHIFT;
1475 int offset = offset_in_page(gpa);
1477 addr = gfn_to_hva_prot(kvm, gfn, NULL);
1478 if (kvm_is_error_hva(addr))
1480 pagefault_disable();
1481 r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
1487 EXPORT_SYMBOL(kvm_read_guest_atomic);
1489 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1490 int offset, int len)
1495 addr = gfn_to_hva(kvm, gfn);
1496 if (kvm_is_error_hva(addr))
1498 r = __copy_to_user((void __user *)addr + offset, data, len);
1501 mark_page_dirty(kvm, gfn);
1504 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1506 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1509 gfn_t gfn = gpa >> PAGE_SHIFT;
1511 int offset = offset_in_page(gpa);
1514 while ((seg = next_segment(len, offset)) != 0) {
1515 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1526 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1527 gpa_t gpa, unsigned long len)
1529 struct kvm_memslots *slots = kvm_memslots(kvm);
1530 int offset = offset_in_page(gpa);
1531 gfn_t start_gfn = gpa >> PAGE_SHIFT;
1532 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1533 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1534 gfn_t nr_pages_avail;
1537 ghc->generation = slots->generation;
1539 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1540 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, &nr_pages_avail);
1541 if (!kvm_is_error_hva(ghc->hva) && nr_pages_avail >= nr_pages_needed) {
1545 * If the requested region crosses two memslots, we still
1546 * verify that the entire region is valid here.
1548 while (start_gfn <= end_gfn) {
1549 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1550 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1552 if (kvm_is_error_hva(ghc->hva))
1554 start_gfn += nr_pages_avail;
1556 /* Use the slow path for cross page reads and writes. */
1557 ghc->memslot = NULL;
1561 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1563 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1564 void *data, unsigned long len)
1566 struct kvm_memslots *slots = kvm_memslots(kvm);
1569 BUG_ON(len > ghc->len);
1571 if (slots->generation != ghc->generation)
1572 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1574 if (unlikely(!ghc->memslot))
1575 return kvm_write_guest(kvm, ghc->gpa, data, len);
1577 if (kvm_is_error_hva(ghc->hva))
1580 r = __copy_to_user((void __user *)ghc->hva, data, len);
1583 mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1587 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1589 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1590 void *data, unsigned long len)
1592 struct kvm_memslots *slots = kvm_memslots(kvm);
1595 BUG_ON(len > ghc->len);
1597 if (slots->generation != ghc->generation)
1598 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1600 if (unlikely(!ghc->memslot))
1601 return kvm_read_guest(kvm, ghc->gpa, data, len);
1603 if (kvm_is_error_hva(ghc->hva))
1606 r = __copy_from_user(data, (void __user *)ghc->hva, len);
1612 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1614 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1616 return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1619 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1621 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1623 gfn_t gfn = gpa >> PAGE_SHIFT;
1625 int offset = offset_in_page(gpa);
1628 while ((seg = next_segment(len, offset)) != 0) {
1629 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1638 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1640 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1643 if (memslot && memslot->dirty_bitmap) {
1644 unsigned long rel_gfn = gfn - memslot->base_gfn;
1646 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1650 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1652 struct kvm_memory_slot *memslot;
1654 memslot = gfn_to_memslot(kvm, gfn);
1655 mark_page_dirty_in_slot(kvm, memslot, gfn);
1659 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1661 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1666 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1668 if (kvm_arch_vcpu_runnable(vcpu)) {
1669 kvm_make_request(KVM_REQ_UNHALT, vcpu);
1672 if (kvm_cpu_has_pending_timer(vcpu))
1674 if (signal_pending(current))
1680 finish_wait(&vcpu->wq, &wait);
1685 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1687 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1690 int cpu = vcpu->cpu;
1691 wait_queue_head_t *wqp;
1693 wqp = kvm_arch_vcpu_wq(vcpu);
1694 if (waitqueue_active(wqp)) {
1695 wake_up_interruptible(wqp);
1696 ++vcpu->stat.halt_wakeup;
1700 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1701 if (kvm_arch_vcpu_should_kick(vcpu))
1702 smp_send_reschedule(cpu);
1705 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
1706 #endif /* !CONFIG_S390 */
1708 void kvm_resched(struct kvm_vcpu *vcpu)
1710 if (!need_resched())
1714 EXPORT_SYMBOL_GPL(kvm_resched);
1716 bool kvm_vcpu_yield_to(struct kvm_vcpu *target)
1719 struct task_struct *task = NULL;
1723 pid = rcu_dereference(target->pid);
1725 task = get_pid_task(target->pid, PIDTYPE_PID);
1729 if (task->flags & PF_VCPU) {
1730 put_task_struct(task);
1733 ret = yield_to(task, 1);
1734 put_task_struct(task);
1738 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1740 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1742 * Helper that checks whether a VCPU is eligible for directed yield.
1743 * Most eligible candidate to yield is decided by following heuristics:
1745 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1746 * (preempted lock holder), indicated by @in_spin_loop.
1747 * Set at the beiginning and cleared at the end of interception/PLE handler.
1749 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1750 * chance last time (mostly it has become eligible now since we have probably
1751 * yielded to lockholder in last iteration. This is done by toggling
1752 * @dy_eligible each time a VCPU checked for eligibility.)
1754 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1755 * to preempted lock-holder could result in wrong VCPU selection and CPU
1756 * burning. Giving priority for a potential lock-holder increases lock
1759 * Since algorithm is based on heuristics, accessing another VCPU data without
1760 * locking does not harm. It may result in trying to yield to same VCPU, fail
1761 * and continue with next VCPU and so on.
1763 bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1767 eligible = !vcpu->spin_loop.in_spin_loop ||
1768 (vcpu->spin_loop.in_spin_loop &&
1769 vcpu->spin_loop.dy_eligible);
1771 if (vcpu->spin_loop.in_spin_loop)
1772 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1778 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1780 struct kvm *kvm = me->kvm;
1781 struct kvm_vcpu *vcpu;
1782 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1788 kvm_vcpu_set_in_spin_loop(me, true);
1790 * We boost the priority of a VCPU that is runnable but not
1791 * currently running, because it got preempted by something
1792 * else and called schedule in __vcpu_run. Hopefully that
1793 * VCPU is holding the lock that we need and will release it.
1794 * We approximate round-robin by starting at the last boosted VCPU.
1796 for (pass = 0; pass < 2 && !yielded && try; pass++) {
1797 kvm_for_each_vcpu(i, vcpu, kvm) {
1798 if (!pass && i <= last_boosted_vcpu) {
1799 i = last_boosted_vcpu;
1801 } else if (pass && i > last_boosted_vcpu)
1803 if (!ACCESS_ONCE(vcpu->preempted))
1807 if (waitqueue_active(&vcpu->wq))
1809 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1812 yielded = kvm_vcpu_yield_to(vcpu);
1814 kvm->last_boosted_vcpu = i;
1816 } else if (yielded < 0) {
1823 kvm_vcpu_set_in_spin_loop(me, false);
1825 /* Ensure vcpu is not eligible during next spinloop */
1826 kvm_vcpu_set_dy_eligible(me, false);
1828 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1830 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1832 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1835 if (vmf->pgoff == 0)
1836 page = virt_to_page(vcpu->run);
1838 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1839 page = virt_to_page(vcpu->arch.pio_data);
1841 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1842 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1843 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1846 return kvm_arch_vcpu_fault(vcpu, vmf);
1852 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1853 .fault = kvm_vcpu_fault,
1856 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1858 vma->vm_ops = &kvm_vcpu_vm_ops;
1862 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1864 struct kvm_vcpu *vcpu = filp->private_data;
1866 kvm_put_kvm(vcpu->kvm);
1870 static struct file_operations kvm_vcpu_fops = {
1871 .release = kvm_vcpu_release,
1872 .unlocked_ioctl = kvm_vcpu_ioctl,
1873 #ifdef CONFIG_COMPAT
1874 .compat_ioctl = kvm_vcpu_compat_ioctl,
1876 .mmap = kvm_vcpu_mmap,
1877 .llseek = noop_llseek,
1881 * Allocates an inode for the vcpu.
1883 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1885 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
1889 * Creates some virtual cpus. Good luck creating more than one.
1891 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1894 struct kvm_vcpu *vcpu, *v;
1896 vcpu = kvm_arch_vcpu_create(kvm, id);
1898 return PTR_ERR(vcpu);
1900 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1902 r = kvm_arch_vcpu_setup(vcpu);
1906 mutex_lock(&kvm->lock);
1907 if (!kvm_vcpu_compatible(vcpu)) {
1909 goto unlock_vcpu_destroy;
1911 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1913 goto unlock_vcpu_destroy;
1916 kvm_for_each_vcpu(r, v, kvm)
1917 if (v->vcpu_id == id) {
1919 goto unlock_vcpu_destroy;
1922 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1924 /* Now it's all set up, let userspace reach it */
1926 r = create_vcpu_fd(vcpu);
1929 goto unlock_vcpu_destroy;
1932 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1934 atomic_inc(&kvm->online_vcpus);
1936 mutex_unlock(&kvm->lock);
1937 kvm_arch_vcpu_postcreate(vcpu);
1940 unlock_vcpu_destroy:
1941 mutex_unlock(&kvm->lock);
1943 kvm_arch_vcpu_destroy(vcpu);
1947 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1950 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1951 vcpu->sigset_active = 1;
1952 vcpu->sigset = *sigset;
1954 vcpu->sigset_active = 0;
1958 static long kvm_vcpu_ioctl(struct file *filp,
1959 unsigned int ioctl, unsigned long arg)
1961 struct kvm_vcpu *vcpu = filp->private_data;
1962 void __user *argp = (void __user *)arg;
1964 struct kvm_fpu *fpu = NULL;
1965 struct kvm_sregs *kvm_sregs = NULL;
1967 if (vcpu->kvm->mm != current->mm)
1970 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
1972 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1973 * so vcpu_load() would break it.
1975 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1976 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1980 r = vcpu_load(vcpu);
1988 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1989 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1991 case KVM_GET_REGS: {
1992 struct kvm_regs *kvm_regs;
1995 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1998 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2002 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2009 case KVM_SET_REGS: {
2010 struct kvm_regs *kvm_regs;
2013 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2014 if (IS_ERR(kvm_regs)) {
2015 r = PTR_ERR(kvm_regs);
2018 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2022 case KVM_GET_SREGS: {
2023 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2027 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2031 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2036 case KVM_SET_SREGS: {
2037 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2038 if (IS_ERR(kvm_sregs)) {
2039 r = PTR_ERR(kvm_sregs);
2043 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2046 case KVM_GET_MP_STATE: {
2047 struct kvm_mp_state mp_state;
2049 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2053 if (copy_to_user(argp, &mp_state, sizeof mp_state))
2058 case KVM_SET_MP_STATE: {
2059 struct kvm_mp_state mp_state;
2062 if (copy_from_user(&mp_state, argp, sizeof mp_state))
2064 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2067 case KVM_TRANSLATE: {
2068 struct kvm_translation tr;
2071 if (copy_from_user(&tr, argp, sizeof tr))
2073 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2077 if (copy_to_user(argp, &tr, sizeof tr))
2082 case KVM_SET_GUEST_DEBUG: {
2083 struct kvm_guest_debug dbg;
2086 if (copy_from_user(&dbg, argp, sizeof dbg))
2088 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2091 case KVM_SET_SIGNAL_MASK: {
2092 struct kvm_signal_mask __user *sigmask_arg = argp;
2093 struct kvm_signal_mask kvm_sigmask;
2094 sigset_t sigset, *p;
2099 if (copy_from_user(&kvm_sigmask, argp,
2100 sizeof kvm_sigmask))
2103 if (kvm_sigmask.len != sizeof sigset)
2106 if (copy_from_user(&sigset, sigmask_arg->sigset,
2111 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2115 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2119 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2123 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2129 fpu = memdup_user(argp, sizeof(*fpu));
2135 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2139 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2148 #ifdef CONFIG_COMPAT
2149 static long kvm_vcpu_compat_ioctl(struct file *filp,
2150 unsigned int ioctl, unsigned long arg)
2152 struct kvm_vcpu *vcpu = filp->private_data;
2153 void __user *argp = compat_ptr(arg);
2156 if (vcpu->kvm->mm != current->mm)
2160 case KVM_SET_SIGNAL_MASK: {
2161 struct kvm_signal_mask __user *sigmask_arg = argp;
2162 struct kvm_signal_mask kvm_sigmask;
2163 compat_sigset_t csigset;
2168 if (copy_from_user(&kvm_sigmask, argp,
2169 sizeof kvm_sigmask))
2172 if (kvm_sigmask.len != sizeof csigset)
2175 if (copy_from_user(&csigset, sigmask_arg->sigset,
2178 sigset_from_compat(&sigset, &csigset);
2179 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2181 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2185 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2193 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2194 int (*accessor)(struct kvm_device *dev,
2195 struct kvm_device_attr *attr),
2198 struct kvm_device_attr attr;
2203 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2206 return accessor(dev, &attr);
2209 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2212 struct kvm_device *dev = filp->private_data;
2215 case KVM_SET_DEVICE_ATTR:
2216 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2217 case KVM_GET_DEVICE_ATTR:
2218 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2219 case KVM_HAS_DEVICE_ATTR:
2220 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2222 if (dev->ops->ioctl)
2223 return dev->ops->ioctl(dev, ioctl, arg);
2229 static int kvm_device_release(struct inode *inode, struct file *filp)
2231 struct kvm_device *dev = filp->private_data;
2232 struct kvm *kvm = dev->kvm;
2238 static const struct file_operations kvm_device_fops = {
2239 .unlocked_ioctl = kvm_device_ioctl,
2240 #ifdef CONFIG_COMPAT
2241 .compat_ioctl = kvm_device_ioctl,
2243 .release = kvm_device_release,
2246 struct kvm_device *kvm_device_from_filp(struct file *filp)
2248 if (filp->f_op != &kvm_device_fops)
2251 return filp->private_data;
2254 static int kvm_ioctl_create_device(struct kvm *kvm,
2255 struct kvm_create_device *cd)
2257 struct kvm_device_ops *ops = NULL;
2258 struct kvm_device *dev;
2259 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2263 #ifdef CONFIG_KVM_MPIC
2264 case KVM_DEV_TYPE_FSL_MPIC_20:
2265 case KVM_DEV_TYPE_FSL_MPIC_42:
2266 ops = &kvm_mpic_ops;
2269 #ifdef CONFIG_KVM_XICS
2270 case KVM_DEV_TYPE_XICS:
2271 ops = &kvm_xics_ops;
2281 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2288 ret = ops->create(dev, cd->type);
2294 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2300 list_add(&dev->vm_node, &kvm->devices);
2306 static long kvm_vm_ioctl(struct file *filp,
2307 unsigned int ioctl, unsigned long arg)
2309 struct kvm *kvm = filp->private_data;
2310 void __user *argp = (void __user *)arg;
2313 if (kvm->mm != current->mm)
2316 case KVM_CREATE_VCPU:
2317 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2319 case KVM_SET_USER_MEMORY_REGION: {
2320 struct kvm_userspace_memory_region kvm_userspace_mem;
2323 if (copy_from_user(&kvm_userspace_mem, argp,
2324 sizeof kvm_userspace_mem))
2327 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2330 case KVM_GET_DIRTY_LOG: {
2331 struct kvm_dirty_log log;
2334 if (copy_from_user(&log, argp, sizeof log))
2336 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2339 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2340 case KVM_REGISTER_COALESCED_MMIO: {
2341 struct kvm_coalesced_mmio_zone zone;
2343 if (copy_from_user(&zone, argp, sizeof zone))
2345 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2348 case KVM_UNREGISTER_COALESCED_MMIO: {
2349 struct kvm_coalesced_mmio_zone zone;
2351 if (copy_from_user(&zone, argp, sizeof zone))
2353 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2358 struct kvm_irqfd data;
2361 if (copy_from_user(&data, argp, sizeof data))
2363 r = kvm_irqfd(kvm, &data);
2366 case KVM_IOEVENTFD: {
2367 struct kvm_ioeventfd data;
2370 if (copy_from_user(&data, argp, sizeof data))
2372 r = kvm_ioeventfd(kvm, &data);
2375 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2376 case KVM_SET_BOOT_CPU_ID:
2378 mutex_lock(&kvm->lock);
2379 if (atomic_read(&kvm->online_vcpus) != 0)
2382 kvm->bsp_vcpu_id = arg;
2383 mutex_unlock(&kvm->lock);
2386 #ifdef CONFIG_HAVE_KVM_MSI
2387 case KVM_SIGNAL_MSI: {
2391 if (copy_from_user(&msi, argp, sizeof msi))
2393 r = kvm_send_userspace_msi(kvm, &msi);
2397 #ifdef __KVM_HAVE_IRQ_LINE
2398 case KVM_IRQ_LINE_STATUS:
2399 case KVM_IRQ_LINE: {
2400 struct kvm_irq_level irq_event;
2403 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2406 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
2407 ioctl == KVM_IRQ_LINE_STATUS);
2412 if (ioctl == KVM_IRQ_LINE_STATUS) {
2413 if (copy_to_user(argp, &irq_event, sizeof irq_event))
2421 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2422 case KVM_SET_GSI_ROUTING: {
2423 struct kvm_irq_routing routing;
2424 struct kvm_irq_routing __user *urouting;
2425 struct kvm_irq_routing_entry *entries;
2428 if (copy_from_user(&routing, argp, sizeof(routing)))
2431 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2436 entries = vmalloc(routing.nr * sizeof(*entries));
2441 if (copy_from_user(entries, urouting->entries,
2442 routing.nr * sizeof(*entries)))
2443 goto out_free_irq_routing;
2444 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2446 out_free_irq_routing:
2450 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
2451 case KVM_CREATE_DEVICE: {
2452 struct kvm_create_device cd;
2455 if (copy_from_user(&cd, argp, sizeof(cd)))
2458 r = kvm_ioctl_create_device(kvm, &cd);
2463 if (copy_to_user(argp, &cd, sizeof(cd)))
2470 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2472 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2478 #ifdef CONFIG_COMPAT
2479 struct compat_kvm_dirty_log {
2483 compat_uptr_t dirty_bitmap; /* one bit per page */
2488 static long kvm_vm_compat_ioctl(struct file *filp,
2489 unsigned int ioctl, unsigned long arg)
2491 struct kvm *kvm = filp->private_data;
2494 if (kvm->mm != current->mm)
2497 case KVM_GET_DIRTY_LOG: {
2498 struct compat_kvm_dirty_log compat_log;
2499 struct kvm_dirty_log log;
2502 if (copy_from_user(&compat_log, (void __user *)arg,
2503 sizeof(compat_log)))
2505 log.slot = compat_log.slot;
2506 log.padding1 = compat_log.padding1;
2507 log.padding2 = compat_log.padding2;
2508 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2510 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2514 r = kvm_vm_ioctl(filp, ioctl, arg);
2522 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2524 struct page *page[1];
2527 gfn_t gfn = vmf->pgoff;
2528 struct kvm *kvm = vma->vm_file->private_data;
2530 addr = gfn_to_hva(kvm, gfn);
2531 if (kvm_is_error_hva(addr))
2532 return VM_FAULT_SIGBUS;
2534 npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2536 if (unlikely(npages != 1))
2537 return VM_FAULT_SIGBUS;
2539 vmf->page = page[0];
2543 static const struct vm_operations_struct kvm_vm_vm_ops = {
2544 .fault = kvm_vm_fault,
2547 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2549 vma->vm_ops = &kvm_vm_vm_ops;
2553 static struct file_operations kvm_vm_fops = {
2554 .release = kvm_vm_release,
2555 .unlocked_ioctl = kvm_vm_ioctl,
2556 #ifdef CONFIG_COMPAT
2557 .compat_ioctl = kvm_vm_compat_ioctl,
2559 .mmap = kvm_vm_mmap,
2560 .llseek = noop_llseek,
2563 static int kvm_dev_ioctl_create_vm(unsigned long type)
2568 kvm = kvm_create_vm(type);
2570 return PTR_ERR(kvm);
2571 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2572 r = kvm_coalesced_mmio_init(kvm);
2578 r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC);
2585 static long kvm_dev_ioctl_check_extension_generic(long arg)
2588 case KVM_CAP_USER_MEMORY:
2589 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2590 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2591 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2592 case KVM_CAP_SET_BOOT_CPU_ID:
2594 case KVM_CAP_INTERNAL_ERROR_DATA:
2595 #ifdef CONFIG_HAVE_KVM_MSI
2596 case KVM_CAP_SIGNAL_MSI:
2598 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2599 case KVM_CAP_IRQFD_RESAMPLE:
2602 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2603 case KVM_CAP_IRQ_ROUTING:
2604 return KVM_MAX_IRQ_ROUTES;
2609 return kvm_dev_ioctl_check_extension(arg);
2612 static long kvm_dev_ioctl(struct file *filp,
2613 unsigned int ioctl, unsigned long arg)
2618 case KVM_GET_API_VERSION:
2622 r = KVM_API_VERSION;
2625 r = kvm_dev_ioctl_create_vm(arg);
2627 case KVM_CHECK_EXTENSION:
2628 r = kvm_dev_ioctl_check_extension_generic(arg);
2630 case KVM_GET_VCPU_MMAP_SIZE:
2634 r = PAGE_SIZE; /* struct kvm_run */
2636 r += PAGE_SIZE; /* pio data page */
2638 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2639 r += PAGE_SIZE; /* coalesced mmio ring page */
2642 case KVM_TRACE_ENABLE:
2643 case KVM_TRACE_PAUSE:
2644 case KVM_TRACE_DISABLE:
2648 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2654 static struct file_operations kvm_chardev_ops = {
2655 .unlocked_ioctl = kvm_dev_ioctl,
2656 .compat_ioctl = kvm_dev_ioctl,
2657 .llseek = noop_llseek,
2660 static struct miscdevice kvm_dev = {
2666 static void hardware_enable_nolock(void *junk)
2668 int cpu = raw_smp_processor_id();
2671 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2674 cpumask_set_cpu(cpu, cpus_hardware_enabled);
2676 r = kvm_arch_hardware_enable(NULL);
2679 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2680 atomic_inc(&hardware_enable_failed);
2681 printk(KERN_INFO "kvm: enabling virtualization on "
2682 "CPU%d failed\n", cpu);
2686 static void hardware_enable(void *junk)
2688 raw_spin_lock(&kvm_lock);
2689 hardware_enable_nolock(junk);
2690 raw_spin_unlock(&kvm_lock);
2693 static void hardware_disable_nolock(void *junk)
2695 int cpu = raw_smp_processor_id();
2697 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2699 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2700 kvm_arch_hardware_disable(NULL);
2703 static void hardware_disable(void *junk)
2705 raw_spin_lock(&kvm_lock);
2706 hardware_disable_nolock(junk);
2707 raw_spin_unlock(&kvm_lock);
2710 static void hardware_disable_all_nolock(void)
2712 BUG_ON(!kvm_usage_count);
2715 if (!kvm_usage_count)
2716 on_each_cpu(hardware_disable_nolock, NULL, 1);
2719 static void hardware_disable_all(void)
2721 raw_spin_lock(&kvm_lock);
2722 hardware_disable_all_nolock();
2723 raw_spin_unlock(&kvm_lock);
2726 static int hardware_enable_all(void)
2730 raw_spin_lock(&kvm_lock);
2733 if (kvm_usage_count == 1) {
2734 atomic_set(&hardware_enable_failed, 0);
2735 on_each_cpu(hardware_enable_nolock, NULL, 1);
2737 if (atomic_read(&hardware_enable_failed)) {
2738 hardware_disable_all_nolock();
2743 raw_spin_unlock(&kvm_lock);
2748 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2753 if (!kvm_usage_count)
2756 val &= ~CPU_TASKS_FROZEN;
2759 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2761 hardware_disable(NULL);
2764 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2766 hardware_enable(NULL);
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 raw_spin_lock(&kvm_lock);
3060 list_for_each_entry(kvm, &vm_list, vm_list)
3061 *val += *(u32 *)((void *)kvm + offset);
3062 raw_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 raw_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 raw_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_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);