2 * Kernel-based Virtual Machine driver for Linux
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
7 * Copyright (C) 2006 Qumranet, Inc.
8 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
11 * Avi Kivity <avi@qumranet.com>
12 * Yaniv Kamay <yaniv@qumranet.com>
14 * This work is licensed under the terms of the GNU GPL, version 2. See
15 * the COPYING file in the top-level directory.
21 #include <linux/kvm_host.h>
22 #include <linux/kvm.h>
23 #include <linux/module.h>
24 #include <linux/errno.h>
25 #include <linux/percpu.h>
27 #include <linux/miscdevice.h>
28 #include <linux/vmalloc.h>
29 #include <linux/reboot.h>
30 #include <linux/debugfs.h>
31 #include <linux/highmem.h>
32 #include <linux/file.h>
33 #include <linux/syscore_ops.h>
34 #include <linux/cpu.h>
35 #include <linux/sched.h>
36 #include <linux/cpumask.h>
37 #include <linux/smp.h>
38 #include <linux/anon_inodes.h>
39 #include <linux/profile.h>
40 #include <linux/kvm_para.h>
41 #include <linux/pagemap.h>
42 #include <linux/mman.h>
43 #include <linux/swap.h>
44 #include <linux/bitops.h>
45 #include <linux/spinlock.h>
46 #include <linux/compat.h>
47 #include <linux/srcu.h>
48 #include <linux/hugetlb.h>
49 #include <linux/slab.h>
50 #include <linux/sort.h>
51 #include <linux/bsearch.h>
53 #include <asm/processor.h>
55 #include <asm/uaccess.h>
56 #include <asm/pgtable.h>
58 #include "coalesced_mmio.h"
61 #define CREATE_TRACE_POINTS
62 #include <trace/events/kvm.h>
64 MODULE_AUTHOR("Qumranet");
65 MODULE_LICENSE("GPL");
70 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
73 DEFINE_SPINLOCK(kvm_lock);
74 static DEFINE_RAW_SPINLOCK(kvm_count_lock);
77 static cpumask_var_t cpus_hardware_enabled;
78 static int kvm_usage_count = 0;
79 static atomic_t hardware_enable_failed;
81 struct kmem_cache *kvm_vcpu_cache;
82 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
84 static __read_mostly struct preempt_ops kvm_preempt_ops;
86 struct dentry *kvm_debugfs_dir;
88 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
91 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
94 static int hardware_enable_all(void);
95 static void hardware_disable_all(void);
97 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
98 static void update_memslots(struct kvm_memslots *slots,
99 struct kvm_memory_slot *new, u64 last_generation);
101 static void kvm_release_pfn_dirty(pfn_t pfn);
102 static void mark_page_dirty_in_slot(struct kvm *kvm,
103 struct kvm_memory_slot *memslot, gfn_t gfn);
105 __visible bool kvm_rebooting;
106 EXPORT_SYMBOL_GPL(kvm_rebooting);
108 static bool largepages_enabled = true;
110 bool kvm_is_mmio_pfn(pfn_t pfn)
113 return PageReserved(pfn_to_page(pfn));
119 * Switches to specified vcpu, until a matching vcpu_put()
121 int vcpu_load(struct kvm_vcpu *vcpu)
125 if (mutex_lock_killable(&vcpu->mutex))
127 if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
128 /* The thread running this VCPU changed. */
129 struct pid *oldpid = vcpu->pid;
130 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
131 rcu_assign_pointer(vcpu->pid, newpid);
137 preempt_notifier_register(&vcpu->preempt_notifier);
138 kvm_arch_vcpu_load(vcpu, cpu);
143 void vcpu_put(struct kvm_vcpu *vcpu)
146 kvm_arch_vcpu_put(vcpu);
147 preempt_notifier_unregister(&vcpu->preempt_notifier);
149 mutex_unlock(&vcpu->mutex);
152 static void ack_flush(void *_completed)
156 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
161 struct kvm_vcpu *vcpu;
163 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
166 kvm_for_each_vcpu(i, vcpu, kvm) {
167 kvm_make_request(req, vcpu);
170 /* Set ->requests bit before we read ->mode */
173 if (cpus != NULL && cpu != -1 && cpu != me &&
174 kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
175 cpumask_set_cpu(cpu, cpus);
177 if (unlikely(cpus == NULL))
178 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
179 else if (!cpumask_empty(cpus))
180 smp_call_function_many(cpus, ack_flush, NULL, 1);
184 free_cpumask_var(cpus);
188 void kvm_flush_remote_tlbs(struct kvm *kvm)
190 long dirty_count = kvm->tlbs_dirty;
193 if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
194 ++kvm->stat.remote_tlb_flush;
195 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
198 void kvm_reload_remote_mmus(struct kvm *kvm)
200 make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
203 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
205 make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
208 void kvm_make_scan_ioapic_request(struct kvm *kvm)
210 make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
213 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
218 mutex_init(&vcpu->mutex);
223 init_waitqueue_head(&vcpu->wq);
224 kvm_async_pf_vcpu_init(vcpu);
226 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
231 vcpu->run = page_address(page);
233 kvm_vcpu_set_in_spin_loop(vcpu, false);
234 kvm_vcpu_set_dy_eligible(vcpu, false);
235 vcpu->preempted = false;
237 r = kvm_arch_vcpu_init(vcpu);
243 free_page((unsigned long)vcpu->run);
247 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
249 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
252 kvm_arch_vcpu_uninit(vcpu);
253 free_page((unsigned long)vcpu->run);
255 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
257 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
258 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
260 return container_of(mn, struct kvm, mmu_notifier);
263 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
264 struct mm_struct *mm,
265 unsigned long address)
267 struct kvm *kvm = mmu_notifier_to_kvm(mn);
268 int need_tlb_flush, idx;
271 * When ->invalidate_page runs, the linux pte has been zapped
272 * already but the page is still allocated until
273 * ->invalidate_page returns. So if we increase the sequence
274 * here the kvm page fault will notice if the spte can't be
275 * established because the page is going to be freed. If
276 * instead the kvm page fault establishes the spte before
277 * ->invalidate_page runs, kvm_unmap_hva will release it
280 * The sequence increase only need to be seen at spin_unlock
281 * time, and not at spin_lock time.
283 * Increasing the sequence after the spin_unlock would be
284 * unsafe because the kvm page fault could then establish the
285 * pte after kvm_unmap_hva returned, without noticing the page
286 * is going to be freed.
288 idx = srcu_read_lock(&kvm->srcu);
289 spin_lock(&kvm->mmu_lock);
291 kvm->mmu_notifier_seq++;
292 need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
293 /* we've to flush the tlb before the pages can be freed */
295 kvm_flush_remote_tlbs(kvm);
297 spin_unlock(&kvm->mmu_lock);
298 srcu_read_unlock(&kvm->srcu, idx);
301 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
302 struct mm_struct *mm,
303 unsigned long address,
306 struct kvm *kvm = mmu_notifier_to_kvm(mn);
309 idx = srcu_read_lock(&kvm->srcu);
310 spin_lock(&kvm->mmu_lock);
311 kvm->mmu_notifier_seq++;
312 kvm_set_spte_hva(kvm, address, pte);
313 spin_unlock(&kvm->mmu_lock);
314 srcu_read_unlock(&kvm->srcu, idx);
317 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
318 struct mm_struct *mm,
322 struct kvm *kvm = mmu_notifier_to_kvm(mn);
323 int need_tlb_flush = 0, idx;
325 idx = srcu_read_lock(&kvm->srcu);
326 spin_lock(&kvm->mmu_lock);
328 * The count increase must become visible at unlock time as no
329 * spte can be established without taking the mmu_lock and
330 * count is also read inside the mmu_lock critical section.
332 kvm->mmu_notifier_count++;
333 need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
334 need_tlb_flush |= kvm->tlbs_dirty;
335 /* we've to flush the tlb before the pages can be freed */
337 kvm_flush_remote_tlbs(kvm);
339 spin_unlock(&kvm->mmu_lock);
340 srcu_read_unlock(&kvm->srcu, idx);
343 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
344 struct mm_struct *mm,
348 struct kvm *kvm = mmu_notifier_to_kvm(mn);
350 spin_lock(&kvm->mmu_lock);
352 * This sequence increase will notify the kvm page fault that
353 * the page that is going to be mapped in the spte could have
356 kvm->mmu_notifier_seq++;
359 * The above sequence increase must be visible before the
360 * below count decrease, which is ensured by the smp_wmb above
361 * in conjunction with the smp_rmb in mmu_notifier_retry().
363 kvm->mmu_notifier_count--;
364 spin_unlock(&kvm->mmu_lock);
366 BUG_ON(kvm->mmu_notifier_count < 0);
369 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
370 struct mm_struct *mm,
371 unsigned long address)
373 struct kvm *kvm = mmu_notifier_to_kvm(mn);
376 idx = srcu_read_lock(&kvm->srcu);
377 spin_lock(&kvm->mmu_lock);
379 young = kvm_age_hva(kvm, address);
381 kvm_flush_remote_tlbs(kvm);
383 spin_unlock(&kvm->mmu_lock);
384 srcu_read_unlock(&kvm->srcu, idx);
389 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
390 struct mm_struct *mm,
391 unsigned long address)
393 struct kvm *kvm = mmu_notifier_to_kvm(mn);
396 idx = srcu_read_lock(&kvm->srcu);
397 spin_lock(&kvm->mmu_lock);
398 young = kvm_test_age_hva(kvm, address);
399 spin_unlock(&kvm->mmu_lock);
400 srcu_read_unlock(&kvm->srcu, idx);
405 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
406 struct mm_struct *mm)
408 struct kvm *kvm = mmu_notifier_to_kvm(mn);
411 idx = srcu_read_lock(&kvm->srcu);
412 kvm_arch_flush_shadow_all(kvm);
413 srcu_read_unlock(&kvm->srcu, idx);
416 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
417 .invalidate_page = kvm_mmu_notifier_invalidate_page,
418 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
419 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
420 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
421 .test_young = kvm_mmu_notifier_test_young,
422 .change_pte = kvm_mmu_notifier_change_pte,
423 .release = kvm_mmu_notifier_release,
426 static int kvm_init_mmu_notifier(struct kvm *kvm)
428 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
429 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
432 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
434 static int kvm_init_mmu_notifier(struct kvm *kvm)
439 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
441 static void kvm_init_memslots_id(struct kvm *kvm)
444 struct kvm_memslots *slots = kvm->memslots;
446 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
447 slots->id_to_index[i] = slots->memslots[i].id = i;
450 static struct kvm *kvm_create_vm(unsigned long type)
453 struct kvm *kvm = kvm_arch_alloc_vm();
456 return ERR_PTR(-ENOMEM);
458 r = kvm_arch_init_vm(kvm, type);
460 goto out_err_no_disable;
462 r = hardware_enable_all();
464 goto out_err_no_disable;
466 #ifdef CONFIG_HAVE_KVM_IRQCHIP
467 INIT_HLIST_HEAD(&kvm->mask_notifier_list);
469 #ifdef CONFIG_HAVE_KVM_IRQFD
470 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
473 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
476 kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
478 goto out_err_no_srcu;
479 kvm_init_memslots_id(kvm);
480 if (init_srcu_struct(&kvm->srcu))
481 goto out_err_no_srcu;
482 if (init_srcu_struct(&kvm->irq_srcu))
483 goto out_err_no_irq_srcu;
484 for (i = 0; i < KVM_NR_BUSES; i++) {
485 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
491 spin_lock_init(&kvm->mmu_lock);
492 kvm->mm = current->mm;
493 atomic_inc(&kvm->mm->mm_count);
494 kvm_eventfd_init(kvm);
495 mutex_init(&kvm->lock);
496 mutex_init(&kvm->irq_lock);
497 mutex_init(&kvm->slots_lock);
498 atomic_set(&kvm->users_count, 1);
499 INIT_LIST_HEAD(&kvm->devices);
501 r = kvm_init_mmu_notifier(kvm);
505 spin_lock(&kvm_lock);
506 list_add(&kvm->vm_list, &vm_list);
507 spin_unlock(&kvm_lock);
512 cleanup_srcu_struct(&kvm->irq_srcu);
514 cleanup_srcu_struct(&kvm->srcu);
516 hardware_disable_all();
518 for (i = 0; i < KVM_NR_BUSES; i++)
519 kfree(kvm->buses[i]);
520 kfree(kvm->memslots);
521 kvm_arch_free_vm(kvm);
526 * Avoid using vmalloc for a small buffer.
527 * Should not be used when the size is statically known.
529 void *kvm_kvzalloc(unsigned long size)
531 if (size > PAGE_SIZE)
532 return vzalloc(size);
534 return kzalloc(size, GFP_KERNEL);
537 void kvm_kvfree(const void *addr)
539 if (is_vmalloc_addr(addr))
545 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
547 if (!memslot->dirty_bitmap)
550 kvm_kvfree(memslot->dirty_bitmap);
551 memslot->dirty_bitmap = NULL;
555 * Free any memory in @free but not in @dont.
557 static void kvm_free_physmem_slot(struct kvm *kvm, struct kvm_memory_slot *free,
558 struct kvm_memory_slot *dont)
560 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
561 kvm_destroy_dirty_bitmap(free);
563 kvm_arch_free_memslot(kvm, free, dont);
568 static void kvm_free_physmem(struct kvm *kvm)
570 struct kvm_memslots *slots = kvm->memslots;
571 struct kvm_memory_slot *memslot;
573 kvm_for_each_memslot(memslot, slots)
574 kvm_free_physmem_slot(kvm, memslot, NULL);
576 kfree(kvm->memslots);
579 static void kvm_destroy_devices(struct kvm *kvm)
581 struct list_head *node, *tmp;
583 list_for_each_safe(node, tmp, &kvm->devices) {
584 struct kvm_device *dev =
585 list_entry(node, struct kvm_device, vm_node);
588 dev->ops->destroy(dev);
592 static void kvm_destroy_vm(struct kvm *kvm)
595 struct mm_struct *mm = kvm->mm;
597 kvm_arch_sync_events(kvm);
598 spin_lock(&kvm_lock);
599 list_del(&kvm->vm_list);
600 spin_unlock(&kvm_lock);
601 kvm_free_irq_routing(kvm);
602 for (i = 0; i < KVM_NR_BUSES; i++)
603 kvm_io_bus_destroy(kvm->buses[i]);
604 kvm_coalesced_mmio_free(kvm);
605 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
606 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
608 kvm_arch_flush_shadow_all(kvm);
610 kvm_arch_destroy_vm(kvm);
611 kvm_destroy_devices(kvm);
612 kvm_free_physmem(kvm);
613 cleanup_srcu_struct(&kvm->irq_srcu);
614 cleanup_srcu_struct(&kvm->srcu);
615 kvm_arch_free_vm(kvm);
616 hardware_disable_all();
620 void kvm_get_kvm(struct kvm *kvm)
622 atomic_inc(&kvm->users_count);
624 EXPORT_SYMBOL_GPL(kvm_get_kvm);
626 void kvm_put_kvm(struct kvm *kvm)
628 if (atomic_dec_and_test(&kvm->users_count))
631 EXPORT_SYMBOL_GPL(kvm_put_kvm);
634 static int kvm_vm_release(struct inode *inode, struct file *filp)
636 struct kvm *kvm = filp->private_data;
638 kvm_irqfd_release(kvm);
645 * Allocation size is twice as large as the actual dirty bitmap size.
646 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
648 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
651 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
653 memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
654 if (!memslot->dirty_bitmap)
657 #endif /* !CONFIG_S390 */
661 static int cmp_memslot(const void *slot1, const void *slot2)
663 struct kvm_memory_slot *s1, *s2;
665 s1 = (struct kvm_memory_slot *)slot1;
666 s2 = (struct kvm_memory_slot *)slot2;
668 if (s1->npages < s2->npages)
670 if (s1->npages > s2->npages)
677 * Sort the memslots base on its size, so the larger slots
678 * will get better fit.
680 static void sort_memslots(struct kvm_memslots *slots)
684 sort(slots->memslots, KVM_MEM_SLOTS_NUM,
685 sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
687 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
688 slots->id_to_index[slots->memslots[i].id] = i;
691 static void update_memslots(struct kvm_memslots *slots,
692 struct kvm_memory_slot *new,
697 struct kvm_memory_slot *old = id_to_memslot(slots, id);
698 unsigned long npages = old->npages;
701 if (new->npages != npages)
702 sort_memslots(slots);
705 slots->generation = last_generation + 1;
708 static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
710 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
712 #ifdef __KVM_HAVE_READONLY_MEM
713 valid_flags |= KVM_MEM_READONLY;
716 if (mem->flags & ~valid_flags)
722 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
723 struct kvm_memslots *slots, struct kvm_memory_slot *new)
725 struct kvm_memslots *old_memslots = kvm->memslots;
727 update_memslots(slots, new, kvm->memslots->generation);
728 rcu_assign_pointer(kvm->memslots, slots);
729 synchronize_srcu_expedited(&kvm->srcu);
731 kvm_arch_memslots_updated(kvm);
737 * Allocate some memory and give it an address in the guest physical address
740 * Discontiguous memory is allowed, mostly for framebuffers.
742 * Must be called holding mmap_sem for write.
744 int __kvm_set_memory_region(struct kvm *kvm,
745 struct kvm_userspace_memory_region *mem)
749 unsigned long npages;
750 struct kvm_memory_slot *slot;
751 struct kvm_memory_slot old, new;
752 struct kvm_memslots *slots = NULL, *old_memslots;
753 enum kvm_mr_change change;
755 r = check_memory_region_flags(mem);
760 /* General sanity checks */
761 if (mem->memory_size & (PAGE_SIZE - 1))
763 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
765 /* We can read the guest memory with __xxx_user() later on. */
766 if ((mem->slot < KVM_USER_MEM_SLOTS) &&
767 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
768 !access_ok(VERIFY_WRITE,
769 (void __user *)(unsigned long)mem->userspace_addr,
772 if (mem->slot >= KVM_MEM_SLOTS_NUM)
774 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
777 slot = id_to_memslot(kvm->memslots, mem->slot);
778 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
779 npages = mem->memory_size >> PAGE_SHIFT;
781 if (npages > KVM_MEM_MAX_NR_PAGES)
785 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
790 new.base_gfn = base_gfn;
792 new.flags = mem->flags;
796 change = KVM_MR_CREATE;
797 else { /* Modify an existing slot. */
798 if ((mem->userspace_addr != old.userspace_addr) ||
799 (npages != old.npages) ||
800 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
803 if (base_gfn != old.base_gfn)
804 change = KVM_MR_MOVE;
805 else if (new.flags != old.flags)
806 change = KVM_MR_FLAGS_ONLY;
807 else { /* Nothing to change. */
812 } else if (old.npages) {
813 change = KVM_MR_DELETE;
814 } else /* Modify a non-existent slot: disallowed. */
817 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
818 /* Check for overlaps */
820 kvm_for_each_memslot(slot, kvm->memslots) {
821 if ((slot->id >= KVM_USER_MEM_SLOTS) ||
822 (slot->id == mem->slot))
824 if (!((base_gfn + npages <= slot->base_gfn) ||
825 (base_gfn >= slot->base_gfn + slot->npages)))
830 /* Free page dirty bitmap if unneeded */
831 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
832 new.dirty_bitmap = NULL;
835 if (change == KVM_MR_CREATE) {
836 new.userspace_addr = mem->userspace_addr;
838 if (kvm_arch_create_memslot(kvm, &new, npages))
842 /* Allocate page dirty bitmap if needed */
843 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
844 if (kvm_create_dirty_bitmap(&new) < 0)
848 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
849 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
853 slot = id_to_memslot(slots, mem->slot);
854 slot->flags |= KVM_MEMSLOT_INVALID;
856 old_memslots = install_new_memslots(kvm, slots, NULL);
858 /* slot was deleted or moved, clear iommu mapping */
859 kvm_iommu_unmap_pages(kvm, &old);
860 /* From this point no new shadow pages pointing to a deleted,
861 * or moved, memslot will be created.
863 * validation of sp->gfn happens in:
864 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
865 * - kvm_is_visible_gfn (mmu_check_roots)
867 kvm_arch_flush_shadow_memslot(kvm, slot);
868 slots = old_memslots;
871 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
877 * We can re-use the old_memslots from above, the only difference
878 * from the currently installed memslots is the invalid flag. This
879 * will get overwritten by update_memslots anyway.
882 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
888 /* actual memory is freed via old in kvm_free_physmem_slot below */
889 if (change == KVM_MR_DELETE) {
890 new.dirty_bitmap = NULL;
891 memset(&new.arch, 0, sizeof(new.arch));
894 old_memslots = install_new_memslots(kvm, slots, &new);
896 kvm_arch_commit_memory_region(kvm, mem, &old, change);
898 kvm_free_physmem_slot(kvm, &old, &new);
902 * IOMMU mapping: New slots need to be mapped. Old slots need to be
903 * un-mapped and re-mapped if their base changes. Since base change
904 * unmapping is handled above with slot deletion, mapping alone is
905 * needed here. Anything else the iommu might care about for existing
906 * slots (size changes, userspace addr changes and read-only flag
907 * changes) is disallowed above, so any other attribute changes getting
908 * here can be skipped.
910 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
911 r = kvm_iommu_map_pages(kvm, &new);
920 kvm_free_physmem_slot(kvm, &new, &old);
924 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
926 int kvm_set_memory_region(struct kvm *kvm,
927 struct kvm_userspace_memory_region *mem)
931 mutex_lock(&kvm->slots_lock);
932 r = __kvm_set_memory_region(kvm, mem);
933 mutex_unlock(&kvm->slots_lock);
936 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
938 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
939 struct kvm_userspace_memory_region *mem)
941 if (mem->slot >= KVM_USER_MEM_SLOTS)
943 return kvm_set_memory_region(kvm, mem);
946 int kvm_get_dirty_log(struct kvm *kvm,
947 struct kvm_dirty_log *log, int *is_dirty)
949 struct kvm_memory_slot *memslot;
952 unsigned long any = 0;
955 if (log->slot >= KVM_USER_MEM_SLOTS)
958 memslot = id_to_memslot(kvm->memslots, log->slot);
960 if (!memslot->dirty_bitmap)
963 n = kvm_dirty_bitmap_bytes(memslot);
965 for (i = 0; !any && i < n/sizeof(long); ++i)
966 any = memslot->dirty_bitmap[i];
969 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
980 bool kvm_largepages_enabled(void)
982 return largepages_enabled;
985 void kvm_disable_largepages(void)
987 largepages_enabled = false;
989 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
991 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
993 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
995 EXPORT_SYMBOL_GPL(gfn_to_memslot);
997 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
999 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1001 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1002 memslot->flags & KVM_MEMSLOT_INVALID)
1007 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1009 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1011 struct vm_area_struct *vma;
1012 unsigned long addr, size;
1016 addr = gfn_to_hva(kvm, gfn);
1017 if (kvm_is_error_hva(addr))
1020 down_read(¤t->mm->mmap_sem);
1021 vma = find_vma(current->mm, addr);
1025 size = vma_kernel_pagesize(vma);
1028 up_read(¤t->mm->mmap_sem);
1033 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1035 return slot->flags & KVM_MEM_READONLY;
1038 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1039 gfn_t *nr_pages, bool write)
1041 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1042 return KVM_HVA_ERR_BAD;
1044 if (memslot_is_readonly(slot) && write)
1045 return KVM_HVA_ERR_RO_BAD;
1048 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1050 return __gfn_to_hva_memslot(slot, gfn);
1053 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1056 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1059 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1062 return gfn_to_hva_many(slot, gfn, NULL);
1064 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1066 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1068 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1070 EXPORT_SYMBOL_GPL(gfn_to_hva);
1073 * If writable is set to false, the hva returned by this function is only
1074 * allowed to be read.
1076 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1077 gfn_t gfn, bool *writable)
1079 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1081 if (!kvm_is_error_hva(hva) && writable)
1082 *writable = !memslot_is_readonly(slot);
1087 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1089 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1091 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1094 static int kvm_read_hva(void *data, void __user *hva, int len)
1096 return __copy_from_user(data, hva, len);
1099 static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1101 return __copy_from_user_inatomic(data, hva, len);
1104 static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1105 unsigned long start, int write, struct page **page)
1107 int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1110 flags |= FOLL_WRITE;
1112 return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1115 static inline int check_user_page_hwpoison(unsigned long addr)
1117 int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1119 rc = __get_user_pages(current, current->mm, addr, 1,
1120 flags, NULL, NULL, NULL);
1121 return rc == -EHWPOISON;
1125 * The atomic path to get the writable pfn which will be stored in @pfn,
1126 * true indicates success, otherwise false is returned.
1128 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1129 bool write_fault, bool *writable, pfn_t *pfn)
1131 struct page *page[1];
1134 if (!(async || atomic))
1138 * Fast pin a writable pfn only if it is a write fault request
1139 * or the caller allows to map a writable pfn for a read fault
1142 if (!(write_fault || writable))
1145 npages = __get_user_pages_fast(addr, 1, 1, page);
1147 *pfn = page_to_pfn(page[0]);
1158 * The slow path to get the pfn of the specified host virtual address,
1159 * 1 indicates success, -errno is returned if error is detected.
1161 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1162 bool *writable, pfn_t *pfn)
1164 struct page *page[1];
1170 *writable = write_fault;
1173 down_read(¤t->mm->mmap_sem);
1174 npages = get_user_page_nowait(current, current->mm,
1175 addr, write_fault, page);
1176 up_read(¤t->mm->mmap_sem);
1178 npages = get_user_pages_fast(addr, 1, write_fault,
1183 /* map read fault as writable if possible */
1184 if (unlikely(!write_fault) && writable) {
1185 struct page *wpage[1];
1187 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1196 *pfn = page_to_pfn(page[0]);
1200 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1202 if (unlikely(!(vma->vm_flags & VM_READ)))
1205 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1212 * Pin guest page in memory and return its pfn.
1213 * @addr: host virtual address which maps memory to the guest
1214 * @atomic: whether this function can sleep
1215 * @async: whether this function need to wait IO complete if the
1216 * host page is not in the memory
1217 * @write_fault: whether we should get a writable host page
1218 * @writable: whether it allows to map a writable host page for !@write_fault
1220 * The function will map a writable host page for these two cases:
1221 * 1): @write_fault = true
1222 * 2): @write_fault = false && @writable, @writable will tell the caller
1223 * whether the mapping is writable.
1225 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1226 bool write_fault, bool *writable)
1228 struct vm_area_struct *vma;
1232 /* we can do it either atomically or asynchronously, not both */
1233 BUG_ON(atomic && async);
1235 if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1239 return KVM_PFN_ERR_FAULT;
1241 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1245 down_read(¤t->mm->mmap_sem);
1246 if (npages == -EHWPOISON ||
1247 (!async && check_user_page_hwpoison(addr))) {
1248 pfn = KVM_PFN_ERR_HWPOISON;
1252 vma = find_vma_intersection(current->mm, addr, addr + 1);
1255 pfn = KVM_PFN_ERR_FAULT;
1256 else if ((vma->vm_flags & VM_PFNMAP)) {
1257 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1259 BUG_ON(!kvm_is_mmio_pfn(pfn));
1261 if (async && vma_is_valid(vma, write_fault))
1263 pfn = KVM_PFN_ERR_FAULT;
1266 up_read(¤t->mm->mmap_sem);
1271 __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1272 bool *async, bool write_fault, bool *writable)
1274 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1276 if (addr == KVM_HVA_ERR_RO_BAD)
1277 return KVM_PFN_ERR_RO_FAULT;
1279 if (kvm_is_error_hva(addr))
1280 return KVM_PFN_NOSLOT;
1282 /* Do not map writable pfn in the readonly memslot. */
1283 if (writable && memslot_is_readonly(slot)) {
1288 return hva_to_pfn(addr, atomic, async, write_fault,
1292 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1293 bool write_fault, bool *writable)
1295 struct kvm_memory_slot *slot;
1300 slot = gfn_to_memslot(kvm, gfn);
1302 return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1306 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1308 return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1310 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1312 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1313 bool write_fault, bool *writable)
1315 return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1317 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1319 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1321 return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1323 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1325 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1328 return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1330 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1332 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1334 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1337 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1339 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1341 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1343 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1349 addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1350 if (kvm_is_error_hva(addr))
1353 if (entry < nr_pages)
1356 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1358 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1360 static struct page *kvm_pfn_to_page(pfn_t pfn)
1362 if (is_error_noslot_pfn(pfn))
1363 return KVM_ERR_PTR_BAD_PAGE;
1365 if (kvm_is_mmio_pfn(pfn)) {
1367 return KVM_ERR_PTR_BAD_PAGE;
1370 return pfn_to_page(pfn);
1373 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1377 pfn = gfn_to_pfn(kvm, gfn);
1379 return kvm_pfn_to_page(pfn);
1382 EXPORT_SYMBOL_GPL(gfn_to_page);
1384 void kvm_release_page_clean(struct page *page)
1386 WARN_ON(is_error_page(page));
1388 kvm_release_pfn_clean(page_to_pfn(page));
1390 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1392 void kvm_release_pfn_clean(pfn_t pfn)
1394 if (!is_error_noslot_pfn(pfn) && !kvm_is_mmio_pfn(pfn))
1395 put_page(pfn_to_page(pfn));
1397 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1399 void kvm_release_page_dirty(struct page *page)
1401 WARN_ON(is_error_page(page));
1403 kvm_release_pfn_dirty(page_to_pfn(page));
1405 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1407 static void kvm_release_pfn_dirty(pfn_t pfn)
1409 kvm_set_pfn_dirty(pfn);
1410 kvm_release_pfn_clean(pfn);
1413 void kvm_set_pfn_dirty(pfn_t pfn)
1415 if (!kvm_is_mmio_pfn(pfn)) {
1416 struct page *page = pfn_to_page(pfn);
1417 if (!PageReserved(page))
1421 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1423 void kvm_set_pfn_accessed(pfn_t pfn)
1425 if (!kvm_is_mmio_pfn(pfn))
1426 mark_page_accessed(pfn_to_page(pfn));
1428 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1430 void kvm_get_pfn(pfn_t pfn)
1432 if (!kvm_is_mmio_pfn(pfn))
1433 get_page(pfn_to_page(pfn));
1435 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1437 static int next_segment(unsigned long len, int offset)
1439 if (len > PAGE_SIZE - offset)
1440 return PAGE_SIZE - offset;
1445 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1451 addr = gfn_to_hva_prot(kvm, gfn, NULL);
1452 if (kvm_is_error_hva(addr))
1454 r = kvm_read_hva(data, (void __user *)addr + offset, len);
1459 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1461 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1463 gfn_t gfn = gpa >> PAGE_SHIFT;
1465 int offset = offset_in_page(gpa);
1468 while ((seg = next_segment(len, offset)) != 0) {
1469 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1479 EXPORT_SYMBOL_GPL(kvm_read_guest);
1481 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1486 gfn_t gfn = gpa >> PAGE_SHIFT;
1487 int offset = offset_in_page(gpa);
1489 addr = gfn_to_hva_prot(kvm, gfn, NULL);
1490 if (kvm_is_error_hva(addr))
1492 pagefault_disable();
1493 r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
1499 EXPORT_SYMBOL(kvm_read_guest_atomic);
1501 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1502 int offset, int len)
1507 addr = gfn_to_hva(kvm, gfn);
1508 if (kvm_is_error_hva(addr))
1510 r = __copy_to_user((void __user *)addr + offset, data, len);
1513 mark_page_dirty(kvm, gfn);
1516 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1518 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1521 gfn_t gfn = gpa >> PAGE_SHIFT;
1523 int offset = offset_in_page(gpa);
1526 while ((seg = next_segment(len, offset)) != 0) {
1527 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1538 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1539 gpa_t gpa, unsigned long len)
1541 struct kvm_memslots *slots = kvm_memslots(kvm);
1542 int offset = offset_in_page(gpa);
1543 gfn_t start_gfn = gpa >> PAGE_SHIFT;
1544 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1545 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1546 gfn_t nr_pages_avail;
1549 ghc->generation = slots->generation;
1551 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1552 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, &nr_pages_avail);
1553 if (!kvm_is_error_hva(ghc->hva) && nr_pages_avail >= nr_pages_needed) {
1557 * If the requested region crosses two memslots, we still
1558 * verify that the entire region is valid here.
1560 while (start_gfn <= end_gfn) {
1561 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1562 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1564 if (kvm_is_error_hva(ghc->hva))
1566 start_gfn += nr_pages_avail;
1568 /* Use the slow path for cross page reads and writes. */
1569 ghc->memslot = NULL;
1573 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1575 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1576 void *data, unsigned long len)
1578 struct kvm_memslots *slots = kvm_memslots(kvm);
1581 BUG_ON(len > ghc->len);
1583 if (slots->generation != ghc->generation)
1584 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1586 if (unlikely(!ghc->memslot))
1587 return kvm_write_guest(kvm, ghc->gpa, data, len);
1589 if (kvm_is_error_hva(ghc->hva))
1592 r = __copy_to_user((void __user *)ghc->hva, data, len);
1595 mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1599 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1601 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1602 void *data, unsigned long len)
1604 struct kvm_memslots *slots = kvm_memslots(kvm);
1607 BUG_ON(len > ghc->len);
1609 if (slots->generation != ghc->generation)
1610 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1612 if (unlikely(!ghc->memslot))
1613 return kvm_read_guest(kvm, ghc->gpa, data, len);
1615 if (kvm_is_error_hva(ghc->hva))
1618 r = __copy_from_user(data, (void __user *)ghc->hva, len);
1624 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1626 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1628 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
1630 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
1632 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1634 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1636 gfn_t gfn = gpa >> PAGE_SHIFT;
1638 int offset = offset_in_page(gpa);
1641 while ((seg = next_segment(len, offset)) != 0) {
1642 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1651 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1653 static void mark_page_dirty_in_slot(struct kvm *kvm,
1654 struct kvm_memory_slot *memslot,
1657 if (memslot && memslot->dirty_bitmap) {
1658 unsigned long rel_gfn = gfn - memslot->base_gfn;
1660 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1664 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1666 struct kvm_memory_slot *memslot;
1668 memslot = gfn_to_memslot(kvm, gfn);
1669 mark_page_dirty_in_slot(kvm, memslot, gfn);
1673 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1675 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1680 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1682 if (kvm_arch_vcpu_runnable(vcpu)) {
1683 kvm_make_request(KVM_REQ_UNHALT, vcpu);
1686 if (kvm_cpu_has_pending_timer(vcpu))
1688 if (signal_pending(current))
1694 finish_wait(&vcpu->wq, &wait);
1699 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1701 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1704 int cpu = vcpu->cpu;
1705 wait_queue_head_t *wqp;
1707 wqp = kvm_arch_vcpu_wq(vcpu);
1708 if (waitqueue_active(wqp)) {
1709 wake_up_interruptible(wqp);
1710 ++vcpu->stat.halt_wakeup;
1714 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1715 if (kvm_arch_vcpu_should_kick(vcpu))
1716 smp_send_reschedule(cpu);
1719 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
1720 #endif /* !CONFIG_S390 */
1722 bool kvm_vcpu_yield_to(struct kvm_vcpu *target)
1725 struct task_struct *task = NULL;
1729 pid = rcu_dereference(target->pid);
1731 task = get_pid_task(target->pid, PIDTYPE_PID);
1735 if (task->flags & PF_VCPU) {
1736 put_task_struct(task);
1739 ret = yield_to(task, 1);
1740 put_task_struct(task);
1744 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1747 * Helper that checks whether a VCPU is eligible for directed yield.
1748 * Most eligible candidate to yield is decided by following heuristics:
1750 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1751 * (preempted lock holder), indicated by @in_spin_loop.
1752 * Set at the beiginning and cleared at the end of interception/PLE handler.
1754 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1755 * chance last time (mostly it has become eligible now since we have probably
1756 * yielded to lockholder in last iteration. This is done by toggling
1757 * @dy_eligible each time a VCPU checked for eligibility.)
1759 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1760 * to preempted lock-holder could result in wrong VCPU selection and CPU
1761 * burning. Giving priority for a potential lock-holder increases lock
1764 * Since algorithm is based on heuristics, accessing another VCPU data without
1765 * locking does not harm. It may result in trying to yield to same VCPU, fail
1766 * and continue with next VCPU and so on.
1768 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1770 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1773 eligible = !vcpu->spin_loop.in_spin_loop ||
1774 vcpu->spin_loop.dy_eligible;
1776 if (vcpu->spin_loop.in_spin_loop)
1777 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1785 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1787 struct kvm *kvm = me->kvm;
1788 struct kvm_vcpu *vcpu;
1789 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1795 kvm_vcpu_set_in_spin_loop(me, true);
1797 * We boost the priority of a VCPU that is runnable but not
1798 * currently running, because it got preempted by something
1799 * else and called schedule in __vcpu_run. Hopefully that
1800 * VCPU is holding the lock that we need and will release it.
1801 * We approximate round-robin by starting at the last boosted VCPU.
1803 for (pass = 0; pass < 2 && !yielded && try; pass++) {
1804 kvm_for_each_vcpu(i, vcpu, kvm) {
1805 if (!pass && i <= last_boosted_vcpu) {
1806 i = last_boosted_vcpu;
1808 } else if (pass && i > last_boosted_vcpu)
1810 if (!ACCESS_ONCE(vcpu->preempted))
1814 if (waitqueue_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
1816 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1819 yielded = kvm_vcpu_yield_to(vcpu);
1821 kvm->last_boosted_vcpu = i;
1823 } else if (yielded < 0) {
1830 kvm_vcpu_set_in_spin_loop(me, false);
1832 /* Ensure vcpu is not eligible during next spinloop */
1833 kvm_vcpu_set_dy_eligible(me, false);
1835 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1837 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1839 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1842 if (vmf->pgoff == 0)
1843 page = virt_to_page(vcpu->run);
1845 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1846 page = virt_to_page(vcpu->arch.pio_data);
1848 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1849 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1850 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1853 return kvm_arch_vcpu_fault(vcpu, vmf);
1859 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1860 .fault = kvm_vcpu_fault,
1863 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1865 vma->vm_ops = &kvm_vcpu_vm_ops;
1869 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1871 struct kvm_vcpu *vcpu = filp->private_data;
1873 kvm_put_kvm(vcpu->kvm);
1877 static struct file_operations kvm_vcpu_fops = {
1878 .release = kvm_vcpu_release,
1879 .unlocked_ioctl = kvm_vcpu_ioctl,
1880 #ifdef CONFIG_COMPAT
1881 .compat_ioctl = kvm_vcpu_compat_ioctl,
1883 .mmap = kvm_vcpu_mmap,
1884 .llseek = noop_llseek,
1888 * Allocates an inode for the vcpu.
1890 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1892 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
1896 * Creates some virtual cpus. Good luck creating more than one.
1898 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1901 struct kvm_vcpu *vcpu, *v;
1903 if (id >= KVM_MAX_VCPUS)
1906 vcpu = kvm_arch_vcpu_create(kvm, id);
1908 return PTR_ERR(vcpu);
1910 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1912 r = kvm_arch_vcpu_setup(vcpu);
1916 mutex_lock(&kvm->lock);
1917 if (!kvm_vcpu_compatible(vcpu)) {
1919 goto unlock_vcpu_destroy;
1921 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1923 goto unlock_vcpu_destroy;
1926 kvm_for_each_vcpu(r, v, kvm)
1927 if (v->vcpu_id == id) {
1929 goto unlock_vcpu_destroy;
1932 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1934 /* Now it's all set up, let userspace reach it */
1936 r = create_vcpu_fd(vcpu);
1939 goto unlock_vcpu_destroy;
1942 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1944 atomic_inc(&kvm->online_vcpus);
1946 mutex_unlock(&kvm->lock);
1947 kvm_arch_vcpu_postcreate(vcpu);
1950 unlock_vcpu_destroy:
1951 mutex_unlock(&kvm->lock);
1953 kvm_arch_vcpu_destroy(vcpu);
1957 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1960 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1961 vcpu->sigset_active = 1;
1962 vcpu->sigset = *sigset;
1964 vcpu->sigset_active = 0;
1968 static long kvm_vcpu_ioctl(struct file *filp,
1969 unsigned int ioctl, unsigned long arg)
1971 struct kvm_vcpu *vcpu = filp->private_data;
1972 void __user *argp = (void __user *)arg;
1974 struct kvm_fpu *fpu = NULL;
1975 struct kvm_sregs *kvm_sregs = NULL;
1977 if (vcpu->kvm->mm != current->mm)
1980 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
1982 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1983 * so vcpu_load() would break it.
1985 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1986 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1990 r = vcpu_load(vcpu);
1998 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1999 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2001 case KVM_GET_REGS: {
2002 struct kvm_regs *kvm_regs;
2005 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2008 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2012 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2019 case KVM_SET_REGS: {
2020 struct kvm_regs *kvm_regs;
2023 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2024 if (IS_ERR(kvm_regs)) {
2025 r = PTR_ERR(kvm_regs);
2028 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2032 case KVM_GET_SREGS: {
2033 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2037 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2041 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2046 case KVM_SET_SREGS: {
2047 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2048 if (IS_ERR(kvm_sregs)) {
2049 r = PTR_ERR(kvm_sregs);
2053 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2056 case KVM_GET_MP_STATE: {
2057 struct kvm_mp_state mp_state;
2059 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2063 if (copy_to_user(argp, &mp_state, sizeof mp_state))
2068 case KVM_SET_MP_STATE: {
2069 struct kvm_mp_state mp_state;
2072 if (copy_from_user(&mp_state, argp, sizeof mp_state))
2074 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2077 case KVM_TRANSLATE: {
2078 struct kvm_translation tr;
2081 if (copy_from_user(&tr, argp, sizeof tr))
2083 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2087 if (copy_to_user(argp, &tr, sizeof tr))
2092 case KVM_SET_GUEST_DEBUG: {
2093 struct kvm_guest_debug dbg;
2096 if (copy_from_user(&dbg, argp, sizeof dbg))
2098 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2101 case KVM_SET_SIGNAL_MASK: {
2102 struct kvm_signal_mask __user *sigmask_arg = argp;
2103 struct kvm_signal_mask kvm_sigmask;
2104 sigset_t sigset, *p;
2109 if (copy_from_user(&kvm_sigmask, argp,
2110 sizeof kvm_sigmask))
2113 if (kvm_sigmask.len != sizeof sigset)
2116 if (copy_from_user(&sigset, sigmask_arg->sigset,
2121 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2125 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2129 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2133 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2139 fpu = memdup_user(argp, sizeof(*fpu));
2145 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2149 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2158 #ifdef CONFIG_COMPAT
2159 static long kvm_vcpu_compat_ioctl(struct file *filp,
2160 unsigned int ioctl, unsigned long arg)
2162 struct kvm_vcpu *vcpu = filp->private_data;
2163 void __user *argp = compat_ptr(arg);
2166 if (vcpu->kvm->mm != current->mm)
2170 case KVM_SET_SIGNAL_MASK: {
2171 struct kvm_signal_mask __user *sigmask_arg = argp;
2172 struct kvm_signal_mask kvm_sigmask;
2173 compat_sigset_t csigset;
2178 if (copy_from_user(&kvm_sigmask, argp,
2179 sizeof kvm_sigmask))
2182 if (kvm_sigmask.len != sizeof csigset)
2185 if (copy_from_user(&csigset, sigmask_arg->sigset,
2188 sigset_from_compat(&sigset, &csigset);
2189 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2191 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2195 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2203 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2204 int (*accessor)(struct kvm_device *dev,
2205 struct kvm_device_attr *attr),
2208 struct kvm_device_attr attr;
2213 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2216 return accessor(dev, &attr);
2219 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2222 struct kvm_device *dev = filp->private_data;
2225 case KVM_SET_DEVICE_ATTR:
2226 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2227 case KVM_GET_DEVICE_ATTR:
2228 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2229 case KVM_HAS_DEVICE_ATTR:
2230 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2232 if (dev->ops->ioctl)
2233 return dev->ops->ioctl(dev, ioctl, arg);
2239 static int kvm_device_release(struct inode *inode, struct file *filp)
2241 struct kvm_device *dev = filp->private_data;
2242 struct kvm *kvm = dev->kvm;
2248 static const struct file_operations kvm_device_fops = {
2249 .unlocked_ioctl = kvm_device_ioctl,
2250 #ifdef CONFIG_COMPAT
2251 .compat_ioctl = kvm_device_ioctl,
2253 .release = kvm_device_release,
2256 struct kvm_device *kvm_device_from_filp(struct file *filp)
2258 if (filp->f_op != &kvm_device_fops)
2261 return filp->private_data;
2264 static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
2265 #ifdef CONFIG_KVM_MPIC
2266 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
2267 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
2270 #ifdef CONFIG_KVM_XICS
2271 [KVM_DEV_TYPE_XICS] = &kvm_xics_ops,
2275 int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
2277 if (type >= ARRAY_SIZE(kvm_device_ops_table))
2280 if (kvm_device_ops_table[type] != NULL)
2283 kvm_device_ops_table[type] = ops;
2287 static int kvm_ioctl_create_device(struct kvm *kvm,
2288 struct kvm_create_device *cd)
2290 struct kvm_device_ops *ops = NULL;
2291 struct kvm_device *dev;
2292 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2295 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
2298 ops = kvm_device_ops_table[cd->type];
2305 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2312 ret = ops->create(dev, cd->type);
2318 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2324 list_add(&dev->vm_node, &kvm->devices);
2330 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2333 case KVM_CAP_USER_MEMORY:
2334 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2335 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2336 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2337 case KVM_CAP_SET_BOOT_CPU_ID:
2339 case KVM_CAP_INTERNAL_ERROR_DATA:
2340 #ifdef CONFIG_HAVE_KVM_MSI
2341 case KVM_CAP_SIGNAL_MSI:
2343 #ifdef CONFIG_HAVE_KVM_IRQFD
2344 case KVM_CAP_IRQFD_RESAMPLE:
2346 case KVM_CAP_CHECK_EXTENSION_VM:
2348 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2349 case KVM_CAP_IRQ_ROUTING:
2350 return KVM_MAX_IRQ_ROUTES;
2355 return kvm_vm_ioctl_check_extension(kvm, arg);
2358 static long kvm_vm_ioctl(struct file *filp,
2359 unsigned int ioctl, unsigned long arg)
2361 struct kvm *kvm = filp->private_data;
2362 void __user *argp = (void __user *)arg;
2365 if (kvm->mm != current->mm)
2368 case KVM_CREATE_VCPU:
2369 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2371 case KVM_SET_USER_MEMORY_REGION: {
2372 struct kvm_userspace_memory_region kvm_userspace_mem;
2375 if (copy_from_user(&kvm_userspace_mem, argp,
2376 sizeof kvm_userspace_mem))
2379 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2382 case KVM_GET_DIRTY_LOG: {
2383 struct kvm_dirty_log log;
2386 if (copy_from_user(&log, argp, sizeof log))
2388 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2391 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2392 case KVM_REGISTER_COALESCED_MMIO: {
2393 struct kvm_coalesced_mmio_zone zone;
2395 if (copy_from_user(&zone, argp, sizeof zone))
2397 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2400 case KVM_UNREGISTER_COALESCED_MMIO: {
2401 struct kvm_coalesced_mmio_zone zone;
2403 if (copy_from_user(&zone, argp, sizeof zone))
2405 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2410 struct kvm_irqfd data;
2413 if (copy_from_user(&data, argp, sizeof data))
2415 r = kvm_irqfd(kvm, &data);
2418 case KVM_IOEVENTFD: {
2419 struct kvm_ioeventfd data;
2422 if (copy_from_user(&data, argp, sizeof data))
2424 r = kvm_ioeventfd(kvm, &data);
2427 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2428 case KVM_SET_BOOT_CPU_ID:
2430 mutex_lock(&kvm->lock);
2431 if (atomic_read(&kvm->online_vcpus) != 0)
2434 kvm->bsp_vcpu_id = arg;
2435 mutex_unlock(&kvm->lock);
2438 #ifdef CONFIG_HAVE_KVM_MSI
2439 case KVM_SIGNAL_MSI: {
2443 if (copy_from_user(&msi, argp, sizeof msi))
2445 r = kvm_send_userspace_msi(kvm, &msi);
2449 #ifdef __KVM_HAVE_IRQ_LINE
2450 case KVM_IRQ_LINE_STATUS:
2451 case KVM_IRQ_LINE: {
2452 struct kvm_irq_level irq_event;
2455 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2458 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
2459 ioctl == KVM_IRQ_LINE_STATUS);
2464 if (ioctl == KVM_IRQ_LINE_STATUS) {
2465 if (copy_to_user(argp, &irq_event, sizeof irq_event))
2473 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2474 case KVM_SET_GSI_ROUTING: {
2475 struct kvm_irq_routing routing;
2476 struct kvm_irq_routing __user *urouting;
2477 struct kvm_irq_routing_entry *entries;
2480 if (copy_from_user(&routing, argp, sizeof(routing)))
2483 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2488 entries = vmalloc(routing.nr * sizeof(*entries));
2493 if (copy_from_user(entries, urouting->entries,
2494 routing.nr * sizeof(*entries)))
2495 goto out_free_irq_routing;
2496 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2498 out_free_irq_routing:
2502 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
2503 case KVM_CREATE_DEVICE: {
2504 struct kvm_create_device cd;
2507 if (copy_from_user(&cd, argp, sizeof(cd)))
2510 r = kvm_ioctl_create_device(kvm, &cd);
2515 if (copy_to_user(argp, &cd, sizeof(cd)))
2521 case KVM_CHECK_EXTENSION:
2522 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
2525 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2527 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2533 #ifdef CONFIG_COMPAT
2534 struct compat_kvm_dirty_log {
2538 compat_uptr_t dirty_bitmap; /* one bit per page */
2543 static long kvm_vm_compat_ioctl(struct file *filp,
2544 unsigned int ioctl, unsigned long arg)
2546 struct kvm *kvm = filp->private_data;
2549 if (kvm->mm != current->mm)
2552 case KVM_GET_DIRTY_LOG: {
2553 struct compat_kvm_dirty_log compat_log;
2554 struct kvm_dirty_log log;
2557 if (copy_from_user(&compat_log, (void __user *)arg,
2558 sizeof(compat_log)))
2560 log.slot = compat_log.slot;
2561 log.padding1 = compat_log.padding1;
2562 log.padding2 = compat_log.padding2;
2563 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2565 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2569 r = kvm_vm_ioctl(filp, ioctl, arg);
2577 static struct file_operations kvm_vm_fops = {
2578 .release = kvm_vm_release,
2579 .unlocked_ioctl = kvm_vm_ioctl,
2580 #ifdef CONFIG_COMPAT
2581 .compat_ioctl = kvm_vm_compat_ioctl,
2583 .llseek = noop_llseek,
2586 static int kvm_dev_ioctl_create_vm(unsigned long type)
2591 kvm = kvm_create_vm(type);
2593 return PTR_ERR(kvm);
2594 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2595 r = kvm_coalesced_mmio_init(kvm);
2601 r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC);
2608 static long kvm_dev_ioctl(struct file *filp,
2609 unsigned int ioctl, unsigned long arg)
2614 case KVM_GET_API_VERSION:
2617 r = KVM_API_VERSION;
2620 r = kvm_dev_ioctl_create_vm(arg);
2622 case KVM_CHECK_EXTENSION:
2623 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
2625 case KVM_GET_VCPU_MMAP_SIZE:
2628 r = PAGE_SIZE; /* struct kvm_run */
2630 r += PAGE_SIZE; /* pio data page */
2632 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2633 r += PAGE_SIZE; /* coalesced mmio ring page */
2636 case KVM_TRACE_ENABLE:
2637 case KVM_TRACE_PAUSE:
2638 case KVM_TRACE_DISABLE:
2642 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2648 static struct file_operations kvm_chardev_ops = {
2649 .unlocked_ioctl = kvm_dev_ioctl,
2650 .compat_ioctl = kvm_dev_ioctl,
2651 .llseek = noop_llseek,
2654 static struct miscdevice kvm_dev = {
2660 static void hardware_enable_nolock(void *junk)
2662 int cpu = raw_smp_processor_id();
2665 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2668 cpumask_set_cpu(cpu, cpus_hardware_enabled);
2670 r = kvm_arch_hardware_enable();
2673 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2674 atomic_inc(&hardware_enable_failed);
2675 printk(KERN_INFO "kvm: enabling virtualization on "
2676 "CPU%d failed\n", cpu);
2680 static void hardware_enable(void)
2682 raw_spin_lock(&kvm_count_lock);
2683 if (kvm_usage_count)
2684 hardware_enable_nolock(NULL);
2685 raw_spin_unlock(&kvm_count_lock);
2688 static void hardware_disable_nolock(void *junk)
2690 int cpu = raw_smp_processor_id();
2692 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2694 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2695 kvm_arch_hardware_disable();
2698 static void hardware_disable(void)
2700 raw_spin_lock(&kvm_count_lock);
2701 if (kvm_usage_count)
2702 hardware_disable_nolock(NULL);
2703 raw_spin_unlock(&kvm_count_lock);
2706 static void hardware_disable_all_nolock(void)
2708 BUG_ON(!kvm_usage_count);
2711 if (!kvm_usage_count)
2712 on_each_cpu(hardware_disable_nolock, NULL, 1);
2715 static void hardware_disable_all(void)
2717 raw_spin_lock(&kvm_count_lock);
2718 hardware_disable_all_nolock();
2719 raw_spin_unlock(&kvm_count_lock);
2722 static int hardware_enable_all(void)
2726 raw_spin_lock(&kvm_count_lock);
2729 if (kvm_usage_count == 1) {
2730 atomic_set(&hardware_enable_failed, 0);
2731 on_each_cpu(hardware_enable_nolock, NULL, 1);
2733 if (atomic_read(&hardware_enable_failed)) {
2734 hardware_disable_all_nolock();
2739 raw_spin_unlock(&kvm_count_lock);
2744 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2749 val &= ~CPU_TASKS_FROZEN;
2752 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2757 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2765 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2769 * Some (well, at least mine) BIOSes hang on reboot if
2772 * And Intel TXT required VMX off for all cpu when system shutdown.
2774 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2775 kvm_rebooting = true;
2776 on_each_cpu(hardware_disable_nolock, NULL, 1);
2780 static struct notifier_block kvm_reboot_notifier = {
2781 .notifier_call = kvm_reboot,
2785 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2789 for (i = 0; i < bus->dev_count; i++) {
2790 struct kvm_io_device *pos = bus->range[i].dev;
2792 kvm_iodevice_destructor(pos);
2797 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2799 const struct kvm_io_range *r1 = p1;
2800 const struct kvm_io_range *r2 = p2;
2802 if (r1->addr < r2->addr)
2804 if (r1->addr + r1->len > r2->addr + r2->len)
2809 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2810 gpa_t addr, int len)
2812 bus->range[bus->dev_count++] = (struct kvm_io_range) {
2818 sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2819 kvm_io_bus_sort_cmp, NULL);
2824 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2825 gpa_t addr, int len)
2827 struct kvm_io_range *range, key;
2830 key = (struct kvm_io_range) {
2835 range = bsearch(&key, bus->range, bus->dev_count,
2836 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2840 off = range - bus->range;
2842 while (off > 0 && kvm_io_bus_sort_cmp(&key, &bus->range[off-1]) == 0)
2848 /* kvm_io_bus_write - called under kvm->slots_lock */
2849 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2850 int len, const void *val)
2853 struct kvm_io_bus *bus;
2854 struct kvm_io_range range;
2856 range = (struct kvm_io_range) {
2861 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2862 idx = kvm_io_bus_get_first_dev(bus, addr, len);
2866 while (idx < bus->dev_count &&
2867 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2868 if (!kvm_iodevice_write(bus->range[idx].dev, addr, len, val))
2876 /* kvm_io_bus_read - called under kvm->slots_lock */
2877 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2881 struct kvm_io_bus *bus;
2882 struct kvm_io_range range;
2884 range = (struct kvm_io_range) {
2889 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2890 idx = kvm_io_bus_get_first_dev(bus, addr, len);
2894 while (idx < bus->dev_count &&
2895 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2896 if (!kvm_iodevice_read(bus->range[idx].dev, addr, len, val))
2904 /* Caller must hold slots_lock. */
2905 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2906 int len, struct kvm_io_device *dev)
2908 struct kvm_io_bus *new_bus, *bus;
2910 bus = kvm->buses[bus_idx];
2911 if (bus->dev_count > NR_IOBUS_DEVS - 1)
2914 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
2915 sizeof(struct kvm_io_range)), GFP_KERNEL);
2918 memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
2919 sizeof(struct kvm_io_range)));
2920 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2921 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2922 synchronize_srcu_expedited(&kvm->srcu);
2928 /* Caller must hold slots_lock. */
2929 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2930 struct kvm_io_device *dev)
2933 struct kvm_io_bus *new_bus, *bus;
2935 bus = kvm->buses[bus_idx];
2937 for (i = 0; i < bus->dev_count; i++)
2938 if (bus->range[i].dev == dev) {
2946 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
2947 sizeof(struct kvm_io_range)), GFP_KERNEL);
2951 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
2952 new_bus->dev_count--;
2953 memcpy(new_bus->range + i, bus->range + i + 1,
2954 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
2956 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2957 synchronize_srcu_expedited(&kvm->srcu);
2962 static struct notifier_block kvm_cpu_notifier = {
2963 .notifier_call = kvm_cpu_hotplug,
2966 static int vm_stat_get(void *_offset, u64 *val)
2968 unsigned offset = (long)_offset;
2972 spin_lock(&kvm_lock);
2973 list_for_each_entry(kvm, &vm_list, vm_list)
2974 *val += *(u32 *)((void *)kvm + offset);
2975 spin_unlock(&kvm_lock);
2979 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2981 static int vcpu_stat_get(void *_offset, u64 *val)
2983 unsigned offset = (long)_offset;
2985 struct kvm_vcpu *vcpu;
2989 spin_lock(&kvm_lock);
2990 list_for_each_entry(kvm, &vm_list, vm_list)
2991 kvm_for_each_vcpu(i, vcpu, kvm)
2992 *val += *(u32 *)((void *)vcpu + offset);
2994 spin_unlock(&kvm_lock);
2998 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
3000 static const struct file_operations *stat_fops[] = {
3001 [KVM_STAT_VCPU] = &vcpu_stat_fops,
3002 [KVM_STAT_VM] = &vm_stat_fops,
3005 static int kvm_init_debug(void)
3008 struct kvm_stats_debugfs_item *p;
3010 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3011 if (kvm_debugfs_dir == NULL)
3014 for (p = debugfs_entries; p->name; ++p) {
3015 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
3016 (void *)(long)p->offset,
3017 stat_fops[p->kind]);
3018 if (p->dentry == NULL)
3025 debugfs_remove_recursive(kvm_debugfs_dir);
3030 static void kvm_exit_debug(void)
3032 struct kvm_stats_debugfs_item *p;
3034 for (p = debugfs_entries; p->name; ++p)
3035 debugfs_remove(p->dentry);
3036 debugfs_remove(kvm_debugfs_dir);
3039 static int kvm_suspend(void)
3041 if (kvm_usage_count)
3042 hardware_disable_nolock(NULL);
3046 static void kvm_resume(void)
3048 if (kvm_usage_count) {
3049 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3050 hardware_enable_nolock(NULL);
3054 static struct syscore_ops kvm_syscore_ops = {
3055 .suspend = kvm_suspend,
3056 .resume = kvm_resume,
3060 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3062 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3065 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3067 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3068 if (vcpu->preempted)
3069 vcpu->preempted = false;
3071 kvm_arch_sched_in(vcpu, cpu);
3073 kvm_arch_vcpu_load(vcpu, cpu);
3076 static void kvm_sched_out(struct preempt_notifier *pn,
3077 struct task_struct *next)
3079 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3081 if (current->state == TASK_RUNNING)
3082 vcpu->preempted = true;
3083 kvm_arch_vcpu_put(vcpu);
3086 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3087 struct module *module)
3092 r = kvm_arch_init(opaque);
3097 * kvm_arch_init makes sure there's at most one caller
3098 * for architectures that support multiple implementations,
3099 * like intel and amd on x86.
3100 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3101 * conflicts in case kvm is already setup for another implementation.
3103 r = kvm_irqfd_init();
3107 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
3112 r = kvm_arch_hardware_setup();
3116 for_each_online_cpu(cpu) {
3117 smp_call_function_single(cpu,
3118 kvm_arch_check_processor_compat,
3124 r = register_cpu_notifier(&kvm_cpu_notifier);
3127 register_reboot_notifier(&kvm_reboot_notifier);
3129 /* A kmem cache lets us meet the alignment requirements of fx_save. */
3131 vcpu_align = __alignof__(struct kvm_vcpu);
3132 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
3134 if (!kvm_vcpu_cache) {
3139 r = kvm_async_pf_init();
3143 kvm_chardev_ops.owner = module;
3144 kvm_vm_fops.owner = module;
3145 kvm_vcpu_fops.owner = module;
3147 r = misc_register(&kvm_dev);
3149 printk(KERN_ERR "kvm: misc device register failed\n");
3153 register_syscore_ops(&kvm_syscore_ops);
3155 kvm_preempt_ops.sched_in = kvm_sched_in;
3156 kvm_preempt_ops.sched_out = kvm_sched_out;
3158 r = kvm_init_debug();
3160 printk(KERN_ERR "kvm: create debugfs files failed\n");
3167 unregister_syscore_ops(&kvm_syscore_ops);
3168 misc_deregister(&kvm_dev);
3170 kvm_async_pf_deinit();
3172 kmem_cache_destroy(kvm_vcpu_cache);
3174 unregister_reboot_notifier(&kvm_reboot_notifier);
3175 unregister_cpu_notifier(&kvm_cpu_notifier);
3178 kvm_arch_hardware_unsetup();
3180 free_cpumask_var(cpus_hardware_enabled);
3188 EXPORT_SYMBOL_GPL(kvm_init);
3193 misc_deregister(&kvm_dev);
3194 kmem_cache_destroy(kvm_vcpu_cache);
3195 kvm_async_pf_deinit();
3196 unregister_syscore_ops(&kvm_syscore_ops);
3197 unregister_reboot_notifier(&kvm_reboot_notifier);
3198 unregister_cpu_notifier(&kvm_cpu_notifier);
3199 on_each_cpu(hardware_disable_nolock, NULL, 1);
3200 kvm_arch_hardware_unsetup();
3203 free_cpumask_var(cpus_hardware_enabled);
3205 EXPORT_SYMBOL_GPL(kvm_exit);