2 * Kernel-based Virtual Machine driver for Linux
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
7 * Copyright (C) 2006 Qumranet, Inc.
8 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
11 * Avi Kivity <avi@qumranet.com>
12 * Yaniv Kamay <yaniv@qumranet.com>
14 * This work is licensed under the terms of the GNU GPL, version 2. See
15 * the COPYING file in the top-level directory.
21 #include <linux/kvm_host.h>
22 #include <linux/kvm.h>
23 #include <linux/module.h>
24 #include <linux/errno.h>
25 #include <linux/percpu.h>
27 #include <linux/miscdevice.h>
28 #include <linux/vmalloc.h>
29 #include <linux/reboot.h>
30 #include <linux/debugfs.h>
31 #include <linux/highmem.h>
32 #include <linux/file.h>
33 #include <linux/syscore_ops.h>
34 #include <linux/cpu.h>
35 #include <linux/sched.h>
36 #include <linux/cpumask.h>
37 #include <linux/smp.h>
38 #include <linux/anon_inodes.h>
39 #include <linux/profile.h>
40 #include <linux/kvm_para.h>
41 #include <linux/pagemap.h>
42 #include <linux/mman.h>
43 #include <linux/swap.h>
44 #include <linux/bitops.h>
45 #include <linux/spinlock.h>
46 #include <linux/compat.h>
47 #include <linux/srcu.h>
48 #include <linux/hugetlb.h>
49 #include <linux/slab.h>
50 #include <linux/sort.h>
51 #include <linux/bsearch.h>
53 #include <asm/processor.h>
55 #include <asm/uaccess.h>
56 #include <asm/pgtable.h>
58 #include "coalesced_mmio.h"
61 #define CREATE_TRACE_POINTS
62 #include <trace/events/kvm.h>
64 MODULE_AUTHOR("Qumranet");
65 MODULE_LICENSE("GPL");
70 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
73 DEFINE_SPINLOCK(kvm_lock);
74 static DEFINE_RAW_SPINLOCK(kvm_count_lock);
77 static cpumask_var_t cpus_hardware_enabled;
78 static int kvm_usage_count = 0;
79 static atomic_t hardware_enable_failed;
81 struct kmem_cache *kvm_vcpu_cache;
82 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
84 static __read_mostly struct preempt_ops kvm_preempt_ops;
86 struct dentry *kvm_debugfs_dir;
88 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
91 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
94 static int hardware_enable_all(void);
95 static void hardware_disable_all(void);
97 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
98 static void update_memslots(struct kvm_memslots *slots,
99 struct kvm_memory_slot *new, u64 last_generation);
101 static void kvm_release_pfn_dirty(pfn_t pfn);
102 static void mark_page_dirty_in_slot(struct kvm *kvm,
103 struct kvm_memory_slot *memslot, gfn_t gfn);
105 __visible bool kvm_rebooting;
106 EXPORT_SYMBOL_GPL(kvm_rebooting);
108 static bool largepages_enabled = true;
110 bool kvm_is_mmio_pfn(pfn_t pfn)
113 return PageReserved(pfn_to_page(pfn));
119 * Switches to specified vcpu, until a matching vcpu_put()
121 int vcpu_load(struct kvm_vcpu *vcpu)
125 if (mutex_lock_killable(&vcpu->mutex))
127 if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
128 /* The thread running this VCPU changed. */
129 struct pid *oldpid = vcpu->pid;
130 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
131 rcu_assign_pointer(vcpu->pid, newpid);
136 preempt_notifier_register(&vcpu->preempt_notifier);
137 kvm_arch_vcpu_load(vcpu, cpu);
142 void vcpu_put(struct kvm_vcpu *vcpu)
145 kvm_arch_vcpu_put(vcpu);
146 preempt_notifier_unregister(&vcpu->preempt_notifier);
148 mutex_unlock(&vcpu->mutex);
151 static void ack_flush(void *_completed)
155 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
160 struct kvm_vcpu *vcpu;
162 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
165 kvm_for_each_vcpu(i, vcpu, kvm) {
166 kvm_make_request(req, vcpu);
169 /* Set ->requests bit before we read ->mode */
172 if (cpus != NULL && cpu != -1 && cpu != me &&
173 kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
174 cpumask_set_cpu(cpu, cpus);
176 if (unlikely(cpus == NULL))
177 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
178 else if (!cpumask_empty(cpus))
179 smp_call_function_many(cpus, ack_flush, NULL, 1);
183 free_cpumask_var(cpus);
187 void kvm_flush_remote_tlbs(struct kvm *kvm)
189 if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
190 ++kvm->stat.remote_tlb_flush;
191 kvm->tlbs_dirty = false;
193 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
195 void kvm_reload_remote_mmus(struct kvm *kvm)
197 make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
200 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
202 make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
205 void kvm_make_scan_ioapic_request(struct kvm *kvm)
207 make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
210 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
215 mutex_init(&vcpu->mutex);
220 init_waitqueue_head(&vcpu->wq);
221 kvm_async_pf_vcpu_init(vcpu);
223 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
228 vcpu->run = page_address(page);
230 kvm_vcpu_set_in_spin_loop(vcpu, false);
231 kvm_vcpu_set_dy_eligible(vcpu, false);
232 vcpu->preempted = false;
234 r = kvm_arch_vcpu_init(vcpu);
240 free_page((unsigned long)vcpu->run);
244 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
246 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
249 kvm_arch_vcpu_uninit(vcpu);
250 free_page((unsigned long)vcpu->run);
252 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
254 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
255 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
257 return container_of(mn, struct kvm, mmu_notifier);
260 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
261 struct mm_struct *mm,
262 unsigned long address)
264 struct kvm *kvm = mmu_notifier_to_kvm(mn);
265 int need_tlb_flush, idx;
268 * When ->invalidate_page runs, the linux pte has been zapped
269 * already but the page is still allocated until
270 * ->invalidate_page returns. So if we increase the sequence
271 * here the kvm page fault will notice if the spte can't be
272 * established because the page is going to be freed. If
273 * instead the kvm page fault establishes the spte before
274 * ->invalidate_page runs, kvm_unmap_hva will release it
277 * The sequence increase only need to be seen at spin_unlock
278 * time, and not at spin_lock time.
280 * Increasing the sequence after the spin_unlock would be
281 * unsafe because the kvm page fault could then establish the
282 * pte after kvm_unmap_hva returned, without noticing the page
283 * is going to be freed.
285 idx = srcu_read_lock(&kvm->srcu);
286 spin_lock(&kvm->mmu_lock);
288 kvm->mmu_notifier_seq++;
289 need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
290 /* we've to flush the tlb before the pages can be freed */
292 kvm_flush_remote_tlbs(kvm);
294 spin_unlock(&kvm->mmu_lock);
295 srcu_read_unlock(&kvm->srcu, idx);
298 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
299 struct mm_struct *mm,
300 unsigned long address,
303 struct kvm *kvm = mmu_notifier_to_kvm(mn);
306 idx = srcu_read_lock(&kvm->srcu);
307 spin_lock(&kvm->mmu_lock);
308 kvm->mmu_notifier_seq++;
309 kvm_set_spte_hva(kvm, address, pte);
310 spin_unlock(&kvm->mmu_lock);
311 srcu_read_unlock(&kvm->srcu, idx);
314 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
315 struct mm_struct *mm,
319 struct kvm *kvm = mmu_notifier_to_kvm(mn);
320 int need_tlb_flush = 0, idx;
322 idx = srcu_read_lock(&kvm->srcu);
323 spin_lock(&kvm->mmu_lock);
325 * The count increase must become visible at unlock time as no
326 * spte can be established without taking the mmu_lock and
327 * count is also read inside the mmu_lock critical section.
329 kvm->mmu_notifier_count++;
330 need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
331 need_tlb_flush |= kvm->tlbs_dirty;
332 /* we've to flush the tlb before the pages can be freed */
334 kvm_flush_remote_tlbs(kvm);
336 spin_unlock(&kvm->mmu_lock);
337 srcu_read_unlock(&kvm->srcu, idx);
340 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
341 struct mm_struct *mm,
345 struct kvm *kvm = mmu_notifier_to_kvm(mn);
347 spin_lock(&kvm->mmu_lock);
349 * This sequence increase will notify the kvm page fault that
350 * the page that is going to be mapped in the spte could have
353 kvm->mmu_notifier_seq++;
356 * The above sequence increase must be visible before the
357 * below count decrease, which is ensured by the smp_wmb above
358 * in conjunction with the smp_rmb in mmu_notifier_retry().
360 kvm->mmu_notifier_count--;
361 spin_unlock(&kvm->mmu_lock);
363 BUG_ON(kvm->mmu_notifier_count < 0);
366 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
367 struct mm_struct *mm,
368 unsigned long address)
370 struct kvm *kvm = mmu_notifier_to_kvm(mn);
373 idx = srcu_read_lock(&kvm->srcu);
374 spin_lock(&kvm->mmu_lock);
376 young = kvm_age_hva(kvm, address);
378 kvm_flush_remote_tlbs(kvm);
380 spin_unlock(&kvm->mmu_lock);
381 srcu_read_unlock(&kvm->srcu, idx);
386 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
387 struct mm_struct *mm,
388 unsigned long address)
390 struct kvm *kvm = mmu_notifier_to_kvm(mn);
393 idx = srcu_read_lock(&kvm->srcu);
394 spin_lock(&kvm->mmu_lock);
395 young = kvm_test_age_hva(kvm, address);
396 spin_unlock(&kvm->mmu_lock);
397 srcu_read_unlock(&kvm->srcu, idx);
402 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
403 struct mm_struct *mm)
405 struct kvm *kvm = mmu_notifier_to_kvm(mn);
408 idx = srcu_read_lock(&kvm->srcu);
409 kvm_arch_flush_shadow_all(kvm);
410 srcu_read_unlock(&kvm->srcu, idx);
413 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
414 .invalidate_page = kvm_mmu_notifier_invalidate_page,
415 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
416 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
417 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
418 .test_young = kvm_mmu_notifier_test_young,
419 .change_pte = kvm_mmu_notifier_change_pte,
420 .release = kvm_mmu_notifier_release,
423 static int kvm_init_mmu_notifier(struct kvm *kvm)
425 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
426 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
429 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
431 static int kvm_init_mmu_notifier(struct kvm *kvm)
436 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
438 static void kvm_init_memslots_id(struct kvm *kvm)
441 struct kvm_memslots *slots = kvm->memslots;
443 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
444 slots->id_to_index[i] = slots->memslots[i].id = i;
447 static struct kvm *kvm_create_vm(unsigned long type)
450 struct kvm *kvm = kvm_arch_alloc_vm();
453 return ERR_PTR(-ENOMEM);
455 r = kvm_arch_init_vm(kvm, type);
457 goto out_err_nodisable;
459 r = hardware_enable_all();
461 goto out_err_nodisable;
463 #ifdef CONFIG_HAVE_KVM_IRQCHIP
464 INIT_HLIST_HEAD(&kvm->mask_notifier_list);
465 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
468 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
471 kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
474 kvm_init_memslots_id(kvm);
475 if (init_srcu_struct(&kvm->srcu))
477 for (i = 0; i < KVM_NR_BUSES; i++) {
478 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
484 spin_lock_init(&kvm->mmu_lock);
485 kvm->mm = current->mm;
486 atomic_inc(&kvm->mm->mm_count);
487 kvm_eventfd_init(kvm);
488 mutex_init(&kvm->lock);
489 mutex_init(&kvm->irq_lock);
490 mutex_init(&kvm->slots_lock);
491 atomic_set(&kvm->users_count, 1);
492 INIT_LIST_HEAD(&kvm->devices);
494 r = kvm_init_mmu_notifier(kvm);
498 spin_lock(&kvm_lock);
499 list_add(&kvm->vm_list, &vm_list);
500 spin_unlock(&kvm_lock);
505 cleanup_srcu_struct(&kvm->srcu);
507 hardware_disable_all();
509 for (i = 0; i < KVM_NR_BUSES; i++)
510 kfree(kvm->buses[i]);
511 kfree(kvm->memslots);
512 kvm_arch_free_vm(kvm);
517 * Avoid using vmalloc for a small buffer.
518 * Should not be used when the size is statically known.
520 void *kvm_kvzalloc(unsigned long size)
522 if (size > PAGE_SIZE)
523 return vzalloc(size);
525 return kzalloc(size, GFP_KERNEL);
528 void kvm_kvfree(const void *addr)
530 if (is_vmalloc_addr(addr))
536 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
538 if (!memslot->dirty_bitmap)
541 kvm_kvfree(memslot->dirty_bitmap);
542 memslot->dirty_bitmap = NULL;
546 * Free any memory in @free but not in @dont.
548 static void kvm_free_physmem_slot(struct kvm *kvm, struct kvm_memory_slot *free,
549 struct kvm_memory_slot *dont)
551 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
552 kvm_destroy_dirty_bitmap(free);
554 kvm_arch_free_memslot(kvm, free, dont);
559 static void kvm_free_physmem(struct kvm *kvm)
561 struct kvm_memslots *slots = kvm->memslots;
562 struct kvm_memory_slot *memslot;
564 kvm_for_each_memslot(memslot, slots)
565 kvm_free_physmem_slot(kvm, memslot, NULL);
567 kfree(kvm->memslots);
570 static void kvm_destroy_devices(struct kvm *kvm)
572 struct list_head *node, *tmp;
574 list_for_each_safe(node, tmp, &kvm->devices) {
575 struct kvm_device *dev =
576 list_entry(node, struct kvm_device, vm_node);
579 dev->ops->destroy(dev);
583 static void kvm_destroy_vm(struct kvm *kvm)
586 struct mm_struct *mm = kvm->mm;
588 kvm_arch_sync_events(kvm);
589 spin_lock(&kvm_lock);
590 list_del(&kvm->vm_list);
591 spin_unlock(&kvm_lock);
592 kvm_free_irq_routing(kvm);
593 for (i = 0; i < KVM_NR_BUSES; i++)
594 kvm_io_bus_destroy(kvm->buses[i]);
595 kvm_coalesced_mmio_free(kvm);
596 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
597 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
599 kvm_arch_flush_shadow_all(kvm);
601 kvm_arch_destroy_vm(kvm);
602 kvm_destroy_devices(kvm);
603 kvm_free_physmem(kvm);
604 cleanup_srcu_struct(&kvm->srcu);
605 kvm_arch_free_vm(kvm);
606 hardware_disable_all();
610 void kvm_get_kvm(struct kvm *kvm)
612 atomic_inc(&kvm->users_count);
614 EXPORT_SYMBOL_GPL(kvm_get_kvm);
616 void kvm_put_kvm(struct kvm *kvm)
618 if (atomic_dec_and_test(&kvm->users_count))
621 EXPORT_SYMBOL_GPL(kvm_put_kvm);
624 static int kvm_vm_release(struct inode *inode, struct file *filp)
626 struct kvm *kvm = filp->private_data;
628 kvm_irqfd_release(kvm);
635 * Allocation size is twice as large as the actual dirty bitmap size.
636 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
638 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
640 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
642 memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
643 if (!memslot->dirty_bitmap)
649 static int cmp_memslot(const void *slot1, const void *slot2)
651 struct kvm_memory_slot *s1, *s2;
653 s1 = (struct kvm_memory_slot *)slot1;
654 s2 = (struct kvm_memory_slot *)slot2;
656 if (s1->npages < s2->npages)
658 if (s1->npages > s2->npages)
665 * Sort the memslots base on its size, so the larger slots
666 * will get better fit.
668 static void sort_memslots(struct kvm_memslots *slots)
672 sort(slots->memslots, KVM_MEM_SLOTS_NUM,
673 sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
675 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
676 slots->id_to_index[slots->memslots[i].id] = i;
679 static void update_memslots(struct kvm_memslots *slots,
680 struct kvm_memory_slot *new,
685 struct kvm_memory_slot *old = id_to_memslot(slots, id);
686 unsigned long npages = old->npages;
689 if (new->npages != npages)
690 sort_memslots(slots);
693 slots->generation = last_generation + 1;
696 static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
698 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
700 #ifdef KVM_CAP_READONLY_MEM
701 valid_flags |= KVM_MEM_READONLY;
704 if (mem->flags & ~valid_flags)
710 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
711 struct kvm_memslots *slots, struct kvm_memory_slot *new)
713 struct kvm_memslots *old_memslots = kvm->memslots;
715 update_memslots(slots, new, kvm->memslots->generation);
716 rcu_assign_pointer(kvm->memslots, slots);
717 synchronize_srcu_expedited(&kvm->srcu);
719 kvm_arch_memslots_updated(kvm);
725 * Allocate some memory and give it an address in the guest physical address
728 * Discontiguous memory is allowed, mostly for framebuffers.
730 * Must be called holding mmap_sem for write.
732 int __kvm_set_memory_region(struct kvm *kvm,
733 struct kvm_userspace_memory_region *mem)
737 unsigned long npages;
738 struct kvm_memory_slot *slot;
739 struct kvm_memory_slot old, new;
740 struct kvm_memslots *slots = NULL, *old_memslots;
741 enum kvm_mr_change change;
743 r = check_memory_region_flags(mem);
748 /* General sanity checks */
749 if (mem->memory_size & (PAGE_SIZE - 1))
751 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
753 /* We can read the guest memory with __xxx_user() later on. */
754 if ((mem->slot < KVM_USER_MEM_SLOTS) &&
755 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
756 !access_ok(VERIFY_WRITE,
757 (void __user *)(unsigned long)mem->userspace_addr,
760 if (mem->slot >= KVM_MEM_SLOTS_NUM)
762 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
765 slot = id_to_memslot(kvm->memslots, mem->slot);
766 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
767 npages = mem->memory_size >> PAGE_SHIFT;
770 if (npages > KVM_MEM_MAX_NR_PAGES)
774 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
779 new.base_gfn = base_gfn;
781 new.flags = mem->flags;
786 change = KVM_MR_CREATE;
787 else { /* Modify an existing slot. */
788 if ((mem->userspace_addr != old.userspace_addr) ||
789 (npages != old.npages) ||
790 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
793 if (base_gfn != old.base_gfn)
794 change = KVM_MR_MOVE;
795 else if (new.flags != old.flags)
796 change = KVM_MR_FLAGS_ONLY;
797 else { /* Nothing to change. */
802 } else if (old.npages) {
803 change = KVM_MR_DELETE;
804 } else /* Modify a non-existent slot: disallowed. */
807 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
808 /* Check for overlaps */
810 kvm_for_each_memslot(slot, kvm->memslots) {
811 if ((slot->id >= KVM_USER_MEM_SLOTS) ||
812 (slot->id == mem->slot))
814 if (!((base_gfn + npages <= slot->base_gfn) ||
815 (base_gfn >= slot->base_gfn + slot->npages)))
820 /* Free page dirty bitmap if unneeded */
821 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
822 new.dirty_bitmap = NULL;
825 if (change == KVM_MR_CREATE) {
826 new.userspace_addr = mem->userspace_addr;
828 if (kvm_arch_create_memslot(kvm, &new, npages))
832 /* Allocate page dirty bitmap if needed */
833 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
834 if (kvm_create_dirty_bitmap(&new) < 0)
838 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
840 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
844 slot = id_to_memslot(slots, mem->slot);
845 slot->flags |= KVM_MEMSLOT_INVALID;
847 old_memslots = install_new_memslots(kvm, slots, NULL);
849 /* slot was deleted or moved, clear iommu mapping */
850 kvm_iommu_unmap_pages(kvm, &old);
851 /* From this point no new shadow pages pointing to a deleted,
852 * or moved, memslot will be created.
854 * validation of sp->gfn happens in:
855 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
856 * - kvm_is_visible_gfn (mmu_check_roots)
858 kvm_arch_flush_shadow_memslot(kvm, slot);
859 slots = old_memslots;
862 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
868 * We can re-use the old_memslots from above, the only difference
869 * from the currently installed memslots is the invalid flag. This
870 * will get overwritten by update_memslots anyway.
873 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
879 /* actual memory is freed via old in kvm_free_physmem_slot below */
880 if (change == KVM_MR_DELETE) {
881 new.dirty_bitmap = NULL;
882 memset(&new.arch, 0, sizeof(new.arch));
885 old_memslots = install_new_memslots(kvm, slots, &new);
887 kvm_arch_commit_memory_region(kvm, mem, &old, change);
889 kvm_free_physmem_slot(kvm, &old, &new);
893 * IOMMU mapping: New slots need to be mapped. Old slots need to be
894 * un-mapped and re-mapped if their base changes. Since base change
895 * unmapping is handled above with slot deletion, mapping alone is
896 * needed here. Anything else the iommu might care about for existing
897 * slots (size changes, userspace addr changes and read-only flag
898 * changes) is disallowed above, so any other attribute changes getting
899 * here can be skipped.
901 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
902 r = kvm_iommu_map_pages(kvm, &new);
911 kvm_free_physmem_slot(kvm, &new, &old);
915 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
917 int kvm_set_memory_region(struct kvm *kvm,
918 struct kvm_userspace_memory_region *mem)
922 mutex_lock(&kvm->slots_lock);
923 r = __kvm_set_memory_region(kvm, mem);
924 mutex_unlock(&kvm->slots_lock);
927 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
929 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
930 struct kvm_userspace_memory_region *mem)
932 if (mem->slot >= KVM_USER_MEM_SLOTS)
934 return kvm_set_memory_region(kvm, mem);
937 int kvm_get_dirty_log(struct kvm *kvm,
938 struct kvm_dirty_log *log, int *is_dirty)
940 struct kvm_memory_slot *memslot;
943 unsigned long any = 0;
946 if (log->slot >= KVM_USER_MEM_SLOTS)
949 memslot = id_to_memslot(kvm->memslots, log->slot);
951 if (!memslot->dirty_bitmap)
954 n = kvm_dirty_bitmap_bytes(memslot);
956 for (i = 0; !any && i < n/sizeof(long); ++i)
957 any = memslot->dirty_bitmap[i];
960 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
970 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
972 bool kvm_largepages_enabled(void)
974 return largepages_enabled;
977 void kvm_disable_largepages(void)
979 largepages_enabled = false;
981 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
983 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
985 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
987 EXPORT_SYMBOL_GPL(gfn_to_memslot);
989 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
991 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
993 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
994 memslot->flags & KVM_MEMSLOT_INVALID)
999 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1001 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1003 struct vm_area_struct *vma;
1004 unsigned long addr, size;
1008 addr = gfn_to_hva(kvm, gfn);
1009 if (kvm_is_error_hva(addr))
1012 down_read(¤t->mm->mmap_sem);
1013 vma = find_vma(current->mm, addr);
1017 size = vma_kernel_pagesize(vma);
1020 up_read(¤t->mm->mmap_sem);
1025 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1027 return slot->flags & KVM_MEM_READONLY;
1030 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1031 gfn_t *nr_pages, bool write)
1033 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1034 return KVM_HVA_ERR_BAD;
1036 if (memslot_is_readonly(slot) && write)
1037 return KVM_HVA_ERR_RO_BAD;
1040 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1042 return __gfn_to_hva_memslot(slot, gfn);
1045 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1048 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1051 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1054 return gfn_to_hva_many(slot, gfn, NULL);
1056 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1058 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1060 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1062 EXPORT_SYMBOL_GPL(gfn_to_hva);
1065 * If writable is set to false, the hva returned by this function is only
1066 * allowed to be read.
1068 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1070 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1071 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1073 if (!kvm_is_error_hva(hva) && writable)
1074 *writable = !memslot_is_readonly(slot);
1079 static int kvm_read_hva(void *data, void __user *hva, int len)
1081 return __copy_from_user(data, hva, len);
1084 static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1086 return __copy_from_user_inatomic(data, hva, len);
1089 static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1090 unsigned long start, int write, struct page **page)
1092 int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1095 flags |= FOLL_WRITE;
1097 return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1100 static inline int check_user_page_hwpoison(unsigned long addr)
1102 int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1104 rc = __get_user_pages(current, current->mm, addr, 1,
1105 flags, NULL, NULL, NULL);
1106 return rc == -EHWPOISON;
1110 * The atomic path to get the writable pfn which will be stored in @pfn,
1111 * true indicates success, otherwise false is returned.
1113 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1114 bool write_fault, bool *writable, pfn_t *pfn)
1116 struct page *page[1];
1119 if (!(async || atomic))
1123 * Fast pin a writable pfn only if it is a write fault request
1124 * or the caller allows to map a writable pfn for a read fault
1127 if (!(write_fault || writable))
1130 npages = __get_user_pages_fast(addr, 1, 1, page);
1132 *pfn = page_to_pfn(page[0]);
1143 * The slow path to get the pfn of the specified host virtual address,
1144 * 1 indicates success, -errno is returned if error is detected.
1146 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1147 bool *writable, pfn_t *pfn)
1149 struct page *page[1];
1155 *writable = write_fault;
1158 down_read(¤t->mm->mmap_sem);
1159 npages = get_user_page_nowait(current, current->mm,
1160 addr, write_fault, page);
1161 up_read(¤t->mm->mmap_sem);
1163 npages = get_user_pages_fast(addr, 1, write_fault,
1168 /* map read fault as writable if possible */
1169 if (unlikely(!write_fault) && writable) {
1170 struct page *wpage[1];
1172 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1181 *pfn = page_to_pfn(page[0]);
1185 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1187 if (unlikely(!(vma->vm_flags & VM_READ)))
1190 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1197 * Pin guest page in memory and return its pfn.
1198 * @addr: host virtual address which maps memory to the guest
1199 * @atomic: whether this function can sleep
1200 * @async: whether this function need to wait IO complete if the
1201 * host page is not in the memory
1202 * @write_fault: whether we should get a writable host page
1203 * @writable: whether it allows to map a writable host page for !@write_fault
1205 * The function will map a writable host page for these two cases:
1206 * 1): @write_fault = true
1207 * 2): @write_fault = false && @writable, @writable will tell the caller
1208 * whether the mapping is writable.
1210 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1211 bool write_fault, bool *writable)
1213 struct vm_area_struct *vma;
1217 /* we can do it either atomically or asynchronously, not both */
1218 BUG_ON(atomic && async);
1220 if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1224 return KVM_PFN_ERR_FAULT;
1226 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1230 down_read(¤t->mm->mmap_sem);
1231 if (npages == -EHWPOISON ||
1232 (!async && check_user_page_hwpoison(addr))) {
1233 pfn = KVM_PFN_ERR_HWPOISON;
1237 vma = find_vma_intersection(current->mm, addr, addr + 1);
1240 pfn = KVM_PFN_ERR_FAULT;
1241 else if ((vma->vm_flags & VM_PFNMAP)) {
1242 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1244 BUG_ON(!kvm_is_mmio_pfn(pfn));
1246 if (async && vma_is_valid(vma, write_fault))
1248 pfn = KVM_PFN_ERR_FAULT;
1251 up_read(¤t->mm->mmap_sem);
1256 __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1257 bool *async, bool write_fault, bool *writable)
1259 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1261 if (addr == KVM_HVA_ERR_RO_BAD)
1262 return KVM_PFN_ERR_RO_FAULT;
1264 if (kvm_is_error_hva(addr))
1265 return KVM_PFN_NOSLOT;
1267 /* Do not map writable pfn in the readonly memslot. */
1268 if (writable && memslot_is_readonly(slot)) {
1273 return hva_to_pfn(addr, atomic, async, write_fault,
1277 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1278 bool write_fault, bool *writable)
1280 struct kvm_memory_slot *slot;
1285 slot = gfn_to_memslot(kvm, gfn);
1287 return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1291 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1293 return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1295 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1297 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1298 bool write_fault, bool *writable)
1300 return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1302 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1304 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1306 return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1308 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1310 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1313 return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1315 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1317 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1319 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1322 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1324 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1326 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1328 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1334 addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1335 if (kvm_is_error_hva(addr))
1338 if (entry < nr_pages)
1341 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1343 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1345 static struct page *kvm_pfn_to_page(pfn_t pfn)
1347 if (is_error_noslot_pfn(pfn))
1348 return KVM_ERR_PTR_BAD_PAGE;
1350 if (kvm_is_mmio_pfn(pfn)) {
1352 return KVM_ERR_PTR_BAD_PAGE;
1355 return pfn_to_page(pfn);
1358 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1362 pfn = gfn_to_pfn(kvm, gfn);
1364 return kvm_pfn_to_page(pfn);
1367 EXPORT_SYMBOL_GPL(gfn_to_page);
1369 void kvm_release_page_clean(struct page *page)
1371 WARN_ON(is_error_page(page));
1373 kvm_release_pfn_clean(page_to_pfn(page));
1375 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1377 void kvm_release_pfn_clean(pfn_t pfn)
1379 if (!is_error_noslot_pfn(pfn) && !kvm_is_mmio_pfn(pfn))
1380 put_page(pfn_to_page(pfn));
1382 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1384 void kvm_release_page_dirty(struct page *page)
1386 WARN_ON(is_error_page(page));
1388 kvm_release_pfn_dirty(page_to_pfn(page));
1390 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1392 static void kvm_release_pfn_dirty(pfn_t pfn)
1394 kvm_set_pfn_dirty(pfn);
1395 kvm_release_pfn_clean(pfn);
1398 void kvm_set_pfn_dirty(pfn_t pfn)
1400 if (!kvm_is_mmio_pfn(pfn)) {
1401 struct page *page = pfn_to_page(pfn);
1402 if (!PageReserved(page))
1406 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1408 void kvm_set_pfn_accessed(pfn_t pfn)
1410 if (!kvm_is_mmio_pfn(pfn))
1411 mark_page_accessed(pfn_to_page(pfn));
1413 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1415 void kvm_get_pfn(pfn_t pfn)
1417 if (!kvm_is_mmio_pfn(pfn))
1418 get_page(pfn_to_page(pfn));
1420 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1422 static int next_segment(unsigned long len, int offset)
1424 if (len > PAGE_SIZE - offset)
1425 return PAGE_SIZE - offset;
1430 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1436 addr = gfn_to_hva_prot(kvm, gfn, NULL);
1437 if (kvm_is_error_hva(addr))
1439 r = kvm_read_hva(data, (void __user *)addr + offset, len);
1444 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1446 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1448 gfn_t gfn = gpa >> PAGE_SHIFT;
1450 int offset = offset_in_page(gpa);
1453 while ((seg = next_segment(len, offset)) != 0) {
1454 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1464 EXPORT_SYMBOL_GPL(kvm_read_guest);
1466 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1471 gfn_t gfn = gpa >> PAGE_SHIFT;
1472 int offset = offset_in_page(gpa);
1474 addr = gfn_to_hva_prot(kvm, gfn, NULL);
1475 if (kvm_is_error_hva(addr))
1477 pagefault_disable();
1478 r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
1484 EXPORT_SYMBOL(kvm_read_guest_atomic);
1486 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1487 int offset, int len)
1492 addr = gfn_to_hva(kvm, gfn);
1493 if (kvm_is_error_hva(addr))
1495 r = __copy_to_user((void __user *)addr + offset, data, len);
1498 mark_page_dirty(kvm, gfn);
1501 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1503 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1506 gfn_t gfn = gpa >> PAGE_SHIFT;
1508 int offset = offset_in_page(gpa);
1511 while ((seg = next_segment(len, offset)) != 0) {
1512 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1523 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1524 gpa_t gpa, unsigned long len)
1526 struct kvm_memslots *slots = kvm_memslots(kvm);
1527 int offset = offset_in_page(gpa);
1528 gfn_t start_gfn = gpa >> PAGE_SHIFT;
1529 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1530 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1531 gfn_t nr_pages_avail;
1534 ghc->generation = slots->generation;
1536 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1537 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, &nr_pages_avail);
1538 if (!kvm_is_error_hva(ghc->hva) && nr_pages_avail >= nr_pages_needed) {
1542 * If the requested region crosses two memslots, we still
1543 * verify that the entire region is valid here.
1545 while (start_gfn <= end_gfn) {
1546 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1547 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1549 if (kvm_is_error_hva(ghc->hva))
1551 start_gfn += nr_pages_avail;
1553 /* Use the slow path for cross page reads and writes. */
1554 ghc->memslot = NULL;
1558 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1560 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1561 void *data, unsigned long len)
1563 struct kvm_memslots *slots = kvm_memslots(kvm);
1566 BUG_ON(len > ghc->len);
1568 if (slots->generation != ghc->generation)
1569 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1571 if (unlikely(!ghc->memslot))
1572 return kvm_write_guest(kvm, ghc->gpa, data, len);
1574 if (kvm_is_error_hva(ghc->hva))
1577 r = __copy_to_user((void __user *)ghc->hva, data, len);
1580 mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1584 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1586 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1587 void *data, unsigned long len)
1589 struct kvm_memslots *slots = kvm_memslots(kvm);
1592 BUG_ON(len > ghc->len);
1594 if (slots->generation != ghc->generation)
1595 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1597 if (unlikely(!ghc->memslot))
1598 return kvm_read_guest(kvm, ghc->gpa, data, len);
1600 if (kvm_is_error_hva(ghc->hva))
1603 r = __copy_from_user(data, (void __user *)ghc->hva, len);
1609 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1611 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1613 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
1615 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
1617 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1619 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1621 gfn_t gfn = gpa >> PAGE_SHIFT;
1623 int offset = offset_in_page(gpa);
1626 while ((seg = next_segment(len, offset)) != 0) {
1627 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1636 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1638 static void mark_page_dirty_in_slot(struct kvm *kvm,
1639 struct kvm_memory_slot *memslot,
1642 if (memslot && memslot->dirty_bitmap) {
1643 unsigned long rel_gfn = gfn - memslot->base_gfn;
1645 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1649 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1651 struct kvm_memory_slot *memslot;
1653 memslot = gfn_to_memslot(kvm, gfn);
1654 mark_page_dirty_in_slot(kvm, memslot, gfn);
1656 EXPORT_SYMBOL_GPL(mark_page_dirty);
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);
1682 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
1686 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1688 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1691 int cpu = vcpu->cpu;
1692 wait_queue_head_t *wqp;
1694 wqp = kvm_arch_vcpu_wq(vcpu);
1695 if (waitqueue_active(wqp)) {
1696 wake_up_interruptible(wqp);
1697 ++vcpu->stat.halt_wakeup;
1701 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1702 if (kvm_arch_vcpu_should_kick(vcpu))
1703 smp_send_reschedule(cpu);
1706 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
1707 #endif /* !CONFIG_S390 */
1709 bool kvm_vcpu_yield_to(struct kvm_vcpu *target)
1712 struct task_struct *task = NULL;
1716 pid = rcu_dereference(target->pid);
1718 task = get_pid_task(target->pid, PIDTYPE_PID);
1722 if (task->flags & PF_VCPU) {
1723 put_task_struct(task);
1726 ret = yield_to(task, 1);
1727 put_task_struct(task);
1731 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1734 * Helper that checks whether a VCPU is eligible for directed yield.
1735 * Most eligible candidate to yield is decided by following heuristics:
1737 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1738 * (preempted lock holder), indicated by @in_spin_loop.
1739 * Set at the beiginning and cleared at the end of interception/PLE handler.
1741 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1742 * chance last time (mostly it has become eligible now since we have probably
1743 * yielded to lockholder in last iteration. This is done by toggling
1744 * @dy_eligible each time a VCPU checked for eligibility.)
1746 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1747 * to preempted lock-holder could result in wrong VCPU selection and CPU
1748 * burning. Giving priority for a potential lock-holder increases lock
1751 * Since algorithm is based on heuristics, accessing another VCPU data without
1752 * locking does not harm. It may result in trying to yield to same VCPU, fail
1753 * and continue with next VCPU and so on.
1755 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1757 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1760 eligible = !vcpu->spin_loop.in_spin_loop ||
1761 (vcpu->spin_loop.in_spin_loop &&
1762 vcpu->spin_loop.dy_eligible);
1764 if (vcpu->spin_loop.in_spin_loop)
1765 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1773 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1775 struct kvm *kvm = me->kvm;
1776 struct kvm_vcpu *vcpu;
1777 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1783 kvm_vcpu_set_in_spin_loop(me, true);
1785 * We boost the priority of a VCPU that is runnable but not
1786 * currently running, because it got preempted by something
1787 * else and called schedule in __vcpu_run. Hopefully that
1788 * VCPU is holding the lock that we need and will release it.
1789 * We approximate round-robin by starting at the last boosted VCPU.
1791 for (pass = 0; pass < 2 && !yielded && try; pass++) {
1792 kvm_for_each_vcpu(i, vcpu, kvm) {
1793 if (!pass && i <= last_boosted_vcpu) {
1794 i = last_boosted_vcpu;
1796 } else if (pass && i > last_boosted_vcpu)
1798 if (!ACCESS_ONCE(vcpu->preempted))
1802 if (waitqueue_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
1804 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1807 yielded = kvm_vcpu_yield_to(vcpu);
1809 kvm->last_boosted_vcpu = i;
1811 } else if (yielded < 0) {
1818 kvm_vcpu_set_in_spin_loop(me, false);
1820 /* Ensure vcpu is not eligible during next spinloop */
1821 kvm_vcpu_set_dy_eligible(me, false);
1823 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1825 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1827 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1830 if (vmf->pgoff == 0)
1831 page = virt_to_page(vcpu->run);
1833 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1834 page = virt_to_page(vcpu->arch.pio_data);
1836 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1837 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1838 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1841 return kvm_arch_vcpu_fault(vcpu, vmf);
1847 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1848 .fault = kvm_vcpu_fault,
1851 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1853 vma->vm_ops = &kvm_vcpu_vm_ops;
1857 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1859 struct kvm_vcpu *vcpu = filp->private_data;
1861 kvm_put_kvm(vcpu->kvm);
1865 static struct file_operations kvm_vcpu_fops = {
1866 .release = kvm_vcpu_release,
1867 .unlocked_ioctl = kvm_vcpu_ioctl,
1868 #ifdef CONFIG_COMPAT
1869 .compat_ioctl = kvm_vcpu_compat_ioctl,
1871 .mmap = kvm_vcpu_mmap,
1872 .llseek = noop_llseek,
1876 * Allocates an inode for the vcpu.
1878 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1880 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
1884 * Creates some virtual cpus. Good luck creating more than one.
1886 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1889 struct kvm_vcpu *vcpu, *v;
1891 if (id >= KVM_MAX_VCPUS)
1894 vcpu = kvm_arch_vcpu_create(kvm, id);
1896 return PTR_ERR(vcpu);
1898 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1900 r = kvm_arch_vcpu_setup(vcpu);
1904 mutex_lock(&kvm->lock);
1905 if (!kvm_vcpu_compatible(vcpu)) {
1907 goto unlock_vcpu_destroy;
1909 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1911 goto unlock_vcpu_destroy;
1914 kvm_for_each_vcpu(r, v, kvm)
1915 if (v->vcpu_id == id) {
1917 goto unlock_vcpu_destroy;
1920 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1922 /* Now it's all set up, let userspace reach it */
1924 r = create_vcpu_fd(vcpu);
1927 goto unlock_vcpu_destroy;
1930 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1932 atomic_inc(&kvm->online_vcpus);
1934 mutex_unlock(&kvm->lock);
1935 kvm_arch_vcpu_postcreate(vcpu);
1938 unlock_vcpu_destroy:
1939 mutex_unlock(&kvm->lock);
1941 kvm_arch_vcpu_destroy(vcpu);
1945 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1948 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1949 vcpu->sigset_active = 1;
1950 vcpu->sigset = *sigset;
1952 vcpu->sigset_active = 0;
1956 static long kvm_vcpu_ioctl(struct file *filp,
1957 unsigned int ioctl, unsigned long arg)
1959 struct kvm_vcpu *vcpu = filp->private_data;
1960 void __user *argp = (void __user *)arg;
1962 struct kvm_fpu *fpu = NULL;
1963 struct kvm_sregs *kvm_sregs = NULL;
1965 if (vcpu->kvm->mm != current->mm)
1968 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
1970 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1971 * so vcpu_load() would break it.
1973 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1974 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1978 r = vcpu_load(vcpu);
1986 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1987 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1989 case KVM_GET_REGS: {
1990 struct kvm_regs *kvm_regs;
1993 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1996 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2000 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2007 case KVM_SET_REGS: {
2008 struct kvm_regs *kvm_regs;
2011 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2012 if (IS_ERR(kvm_regs)) {
2013 r = PTR_ERR(kvm_regs);
2016 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2020 case KVM_GET_SREGS: {
2021 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2025 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2029 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2034 case KVM_SET_SREGS: {
2035 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2036 if (IS_ERR(kvm_sregs)) {
2037 r = PTR_ERR(kvm_sregs);
2041 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2044 case KVM_GET_MP_STATE: {
2045 struct kvm_mp_state mp_state;
2047 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2051 if (copy_to_user(argp, &mp_state, sizeof mp_state))
2056 case KVM_SET_MP_STATE: {
2057 struct kvm_mp_state mp_state;
2060 if (copy_from_user(&mp_state, argp, sizeof mp_state))
2062 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2065 case KVM_TRANSLATE: {
2066 struct kvm_translation tr;
2069 if (copy_from_user(&tr, argp, sizeof tr))
2071 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2075 if (copy_to_user(argp, &tr, sizeof tr))
2080 case KVM_SET_GUEST_DEBUG: {
2081 struct kvm_guest_debug dbg;
2084 if (copy_from_user(&dbg, argp, sizeof dbg))
2086 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2089 case KVM_SET_SIGNAL_MASK: {
2090 struct kvm_signal_mask __user *sigmask_arg = argp;
2091 struct kvm_signal_mask kvm_sigmask;
2092 sigset_t sigset, *p;
2097 if (copy_from_user(&kvm_sigmask, argp,
2098 sizeof kvm_sigmask))
2101 if (kvm_sigmask.len != sizeof sigset)
2104 if (copy_from_user(&sigset, sigmask_arg->sigset,
2109 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2113 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2117 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2121 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2127 fpu = memdup_user(argp, sizeof(*fpu));
2133 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2137 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2146 #ifdef CONFIG_COMPAT
2147 static long kvm_vcpu_compat_ioctl(struct file *filp,
2148 unsigned int ioctl, unsigned long arg)
2150 struct kvm_vcpu *vcpu = filp->private_data;
2151 void __user *argp = compat_ptr(arg);
2154 if (vcpu->kvm->mm != current->mm)
2158 case KVM_SET_SIGNAL_MASK: {
2159 struct kvm_signal_mask __user *sigmask_arg = argp;
2160 struct kvm_signal_mask kvm_sigmask;
2161 compat_sigset_t csigset;
2166 if (copy_from_user(&kvm_sigmask, argp,
2167 sizeof kvm_sigmask))
2170 if (kvm_sigmask.len != sizeof csigset)
2173 if (copy_from_user(&csigset, sigmask_arg->sigset,
2176 sigset_from_compat(&sigset, &csigset);
2177 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2179 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2183 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2191 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2192 int (*accessor)(struct kvm_device *dev,
2193 struct kvm_device_attr *attr),
2196 struct kvm_device_attr attr;
2201 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2204 return accessor(dev, &attr);
2207 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2210 struct kvm_device *dev = filp->private_data;
2213 case KVM_SET_DEVICE_ATTR:
2214 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2215 case KVM_GET_DEVICE_ATTR:
2216 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2217 case KVM_HAS_DEVICE_ATTR:
2218 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2220 if (dev->ops->ioctl)
2221 return dev->ops->ioctl(dev, ioctl, arg);
2227 static int kvm_device_release(struct inode *inode, struct file *filp)
2229 struct kvm_device *dev = filp->private_data;
2230 struct kvm *kvm = dev->kvm;
2236 static const struct file_operations kvm_device_fops = {
2237 .unlocked_ioctl = kvm_device_ioctl,
2238 #ifdef CONFIG_COMPAT
2239 .compat_ioctl = kvm_device_ioctl,
2241 .release = kvm_device_release,
2244 struct kvm_device *kvm_device_from_filp(struct file *filp)
2246 if (filp->f_op != &kvm_device_fops)
2249 return filp->private_data;
2252 static int kvm_ioctl_create_device(struct kvm *kvm,
2253 struct kvm_create_device *cd)
2255 struct kvm_device_ops *ops = NULL;
2256 struct kvm_device *dev;
2257 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2261 #ifdef CONFIG_KVM_MPIC
2262 case KVM_DEV_TYPE_FSL_MPIC_20:
2263 case KVM_DEV_TYPE_FSL_MPIC_42:
2264 ops = &kvm_mpic_ops;
2267 #ifdef CONFIG_KVM_XICS
2268 case KVM_DEV_TYPE_XICS:
2269 ops = &kvm_xics_ops;
2272 #ifdef CONFIG_KVM_VFIO
2273 case KVM_DEV_TYPE_VFIO:
2274 ops = &kvm_vfio_ops;
2277 #ifdef CONFIG_KVM_ARM_VGIC
2278 case KVM_DEV_TYPE_ARM_VGIC_V2:
2279 ops = &kvm_arm_vgic_v2_ops;
2283 case KVM_DEV_TYPE_FLIC:
2284 ops = &kvm_flic_ops;
2294 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2301 ret = ops->create(dev, cd->type);
2307 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2313 list_add(&dev->vm_node, &kvm->devices);
2319 static long kvm_vm_ioctl(struct file *filp,
2320 unsigned int ioctl, unsigned long arg)
2322 struct kvm *kvm = filp->private_data;
2323 void __user *argp = (void __user *)arg;
2326 if (kvm->mm != current->mm)
2329 case KVM_CREATE_VCPU:
2330 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2332 case KVM_SET_USER_MEMORY_REGION: {
2333 struct kvm_userspace_memory_region kvm_userspace_mem;
2336 if (copy_from_user(&kvm_userspace_mem, argp,
2337 sizeof kvm_userspace_mem))
2340 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2343 case KVM_GET_DIRTY_LOG: {
2344 struct kvm_dirty_log log;
2347 if (copy_from_user(&log, argp, sizeof log))
2349 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2352 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2353 case KVM_REGISTER_COALESCED_MMIO: {
2354 struct kvm_coalesced_mmio_zone zone;
2356 if (copy_from_user(&zone, argp, sizeof zone))
2358 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2361 case KVM_UNREGISTER_COALESCED_MMIO: {
2362 struct kvm_coalesced_mmio_zone zone;
2364 if (copy_from_user(&zone, argp, sizeof zone))
2366 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2371 struct kvm_irqfd data;
2374 if (copy_from_user(&data, argp, sizeof data))
2376 r = kvm_irqfd(kvm, &data);
2379 case KVM_IOEVENTFD: {
2380 struct kvm_ioeventfd data;
2383 if (copy_from_user(&data, argp, sizeof data))
2385 r = kvm_ioeventfd(kvm, &data);
2388 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2389 case KVM_SET_BOOT_CPU_ID:
2391 mutex_lock(&kvm->lock);
2392 if (atomic_read(&kvm->online_vcpus) != 0)
2395 kvm->bsp_vcpu_id = arg;
2396 mutex_unlock(&kvm->lock);
2399 #ifdef CONFIG_HAVE_KVM_MSI
2400 case KVM_SIGNAL_MSI: {
2404 if (copy_from_user(&msi, argp, sizeof msi))
2406 r = kvm_send_userspace_msi(kvm, &msi);
2410 #ifdef __KVM_HAVE_IRQ_LINE
2411 case KVM_IRQ_LINE_STATUS:
2412 case KVM_IRQ_LINE: {
2413 struct kvm_irq_level irq_event;
2416 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2419 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
2420 ioctl == KVM_IRQ_LINE_STATUS);
2425 if (ioctl == KVM_IRQ_LINE_STATUS) {
2426 if (copy_to_user(argp, &irq_event, sizeof irq_event))
2434 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2435 case KVM_SET_GSI_ROUTING: {
2436 struct kvm_irq_routing routing;
2437 struct kvm_irq_routing __user *urouting;
2438 struct kvm_irq_routing_entry *entries;
2441 if (copy_from_user(&routing, argp, sizeof(routing)))
2444 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2449 entries = vmalloc(routing.nr * sizeof(*entries));
2454 if (copy_from_user(entries, urouting->entries,
2455 routing.nr * sizeof(*entries)))
2456 goto out_free_irq_routing;
2457 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2459 out_free_irq_routing:
2463 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
2464 case KVM_CREATE_DEVICE: {
2465 struct kvm_create_device cd;
2468 if (copy_from_user(&cd, argp, sizeof(cd)))
2471 r = kvm_ioctl_create_device(kvm, &cd);
2476 if (copy_to_user(argp, &cd, sizeof(cd)))
2483 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2485 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2491 #ifdef CONFIG_COMPAT
2492 struct compat_kvm_dirty_log {
2496 compat_uptr_t dirty_bitmap; /* one bit per page */
2501 static long kvm_vm_compat_ioctl(struct file *filp,
2502 unsigned int ioctl, unsigned long arg)
2504 struct kvm *kvm = filp->private_data;
2507 if (kvm->mm != current->mm)
2510 case KVM_GET_DIRTY_LOG: {
2511 struct compat_kvm_dirty_log compat_log;
2512 struct kvm_dirty_log log;
2515 if (copy_from_user(&compat_log, (void __user *)arg,
2516 sizeof(compat_log)))
2518 log.slot = compat_log.slot;
2519 log.padding1 = compat_log.padding1;
2520 log.padding2 = compat_log.padding2;
2521 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2523 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2527 r = kvm_vm_ioctl(filp, ioctl, arg);
2535 static struct file_operations kvm_vm_fops = {
2536 .release = kvm_vm_release,
2537 .unlocked_ioctl = kvm_vm_ioctl,
2538 #ifdef CONFIG_COMPAT
2539 .compat_ioctl = kvm_vm_compat_ioctl,
2541 .llseek = noop_llseek,
2544 static int kvm_dev_ioctl_create_vm(unsigned long type)
2549 kvm = kvm_create_vm(type);
2551 return PTR_ERR(kvm);
2552 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2553 r = kvm_coalesced_mmio_init(kvm);
2559 r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC);
2566 static long kvm_dev_ioctl_check_extension_generic(long arg)
2569 case KVM_CAP_USER_MEMORY:
2570 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2571 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2572 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2573 case KVM_CAP_SET_BOOT_CPU_ID:
2575 case KVM_CAP_INTERNAL_ERROR_DATA:
2576 #ifdef CONFIG_HAVE_KVM_MSI
2577 case KVM_CAP_SIGNAL_MSI:
2579 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2580 case KVM_CAP_IRQFD_RESAMPLE:
2583 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2584 case KVM_CAP_IRQ_ROUTING:
2585 return KVM_MAX_IRQ_ROUTES;
2590 return kvm_dev_ioctl_check_extension(arg);
2593 static long kvm_dev_ioctl(struct file *filp,
2594 unsigned int ioctl, unsigned long arg)
2599 case KVM_GET_API_VERSION:
2603 r = KVM_API_VERSION;
2606 r = kvm_dev_ioctl_create_vm(arg);
2608 case KVM_CHECK_EXTENSION:
2609 r = kvm_dev_ioctl_check_extension_generic(arg);
2611 case KVM_GET_VCPU_MMAP_SIZE:
2615 r = PAGE_SIZE; /* struct kvm_run */
2617 r += PAGE_SIZE; /* pio data page */
2619 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2620 r += PAGE_SIZE; /* coalesced mmio ring page */
2623 case KVM_TRACE_ENABLE:
2624 case KVM_TRACE_PAUSE:
2625 case KVM_TRACE_DISABLE:
2629 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2635 static struct file_operations kvm_chardev_ops = {
2636 .unlocked_ioctl = kvm_dev_ioctl,
2637 .compat_ioctl = kvm_dev_ioctl,
2638 .llseek = noop_llseek,
2641 static struct miscdevice kvm_dev = {
2647 static void hardware_enable_nolock(void *junk)
2649 int cpu = raw_smp_processor_id();
2652 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2655 cpumask_set_cpu(cpu, cpus_hardware_enabled);
2657 r = kvm_arch_hardware_enable(NULL);
2660 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2661 atomic_inc(&hardware_enable_failed);
2662 printk(KERN_INFO "kvm: enabling virtualization on "
2663 "CPU%d failed\n", cpu);
2667 static void hardware_enable(void)
2669 raw_spin_lock(&kvm_count_lock);
2670 if (kvm_usage_count)
2671 hardware_enable_nolock(NULL);
2672 raw_spin_unlock(&kvm_count_lock);
2675 static void hardware_disable_nolock(void *junk)
2677 int cpu = raw_smp_processor_id();
2679 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2681 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2682 kvm_arch_hardware_disable(NULL);
2685 static void hardware_disable(void)
2687 raw_spin_lock(&kvm_count_lock);
2688 if (kvm_usage_count)
2689 hardware_disable_nolock(NULL);
2690 raw_spin_unlock(&kvm_count_lock);
2693 static void hardware_disable_all_nolock(void)
2695 BUG_ON(!kvm_usage_count);
2698 if (!kvm_usage_count)
2699 on_each_cpu(hardware_disable_nolock, NULL, 1);
2702 static void hardware_disable_all(void)
2704 raw_spin_lock(&kvm_count_lock);
2705 hardware_disable_all_nolock();
2706 raw_spin_unlock(&kvm_count_lock);
2709 static int hardware_enable_all(void)
2713 raw_spin_lock(&kvm_count_lock);
2716 if (kvm_usage_count == 1) {
2717 atomic_set(&hardware_enable_failed, 0);
2718 on_each_cpu(hardware_enable_nolock, NULL, 1);
2720 if (atomic_read(&hardware_enable_failed)) {
2721 hardware_disable_all_nolock();
2726 raw_spin_unlock(&kvm_count_lock);
2731 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2736 val &= ~CPU_TASKS_FROZEN;
2739 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2744 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2752 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2756 * Some (well, at least mine) BIOSes hang on reboot if
2759 * And Intel TXT required VMX off for all cpu when system shutdown.
2761 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2762 kvm_rebooting = true;
2763 on_each_cpu(hardware_disable_nolock, NULL, 1);
2767 static struct notifier_block kvm_reboot_notifier = {
2768 .notifier_call = kvm_reboot,
2772 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2776 for (i = 0; i < bus->dev_count; i++) {
2777 struct kvm_io_device *pos = bus->range[i].dev;
2779 kvm_iodevice_destructor(pos);
2784 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
2785 const struct kvm_io_range *r2)
2787 if (r1->addr < r2->addr)
2789 if (r1->addr + r1->len > r2->addr + r2->len)
2794 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2796 return kvm_io_bus_cmp(p1, p2);
2799 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2800 gpa_t addr, int len)
2802 bus->range[bus->dev_count++] = (struct kvm_io_range) {
2808 sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2809 kvm_io_bus_sort_cmp, NULL);
2814 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2815 gpa_t addr, int len)
2817 struct kvm_io_range *range, key;
2820 key = (struct kvm_io_range) {
2825 range = bsearch(&key, bus->range, bus->dev_count,
2826 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2830 off = range - bus->range;
2832 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
2838 static int __kvm_io_bus_write(struct kvm_io_bus *bus,
2839 struct kvm_io_range *range, const void *val)
2843 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2847 while (idx < bus->dev_count &&
2848 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
2849 if (!kvm_iodevice_write(bus->range[idx].dev, range->addr,
2858 /* kvm_io_bus_write - called under kvm->slots_lock */
2859 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2860 int len, const void *val)
2862 struct kvm_io_bus *bus;
2863 struct kvm_io_range range;
2866 range = (struct kvm_io_range) {
2871 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2872 r = __kvm_io_bus_write(bus, &range, val);
2873 return r < 0 ? r : 0;
2876 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
2877 int kvm_io_bus_write_cookie(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2878 int len, const void *val, long cookie)
2880 struct kvm_io_bus *bus;
2881 struct kvm_io_range range;
2883 range = (struct kvm_io_range) {
2888 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2890 /* First try the device referenced by cookie. */
2891 if ((cookie >= 0) && (cookie < bus->dev_count) &&
2892 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
2893 if (!kvm_iodevice_write(bus->range[cookie].dev, addr, len,
2898 * cookie contained garbage; fall back to search and return the
2899 * correct cookie value.
2901 return __kvm_io_bus_write(bus, &range, val);
2904 static int __kvm_io_bus_read(struct kvm_io_bus *bus, struct kvm_io_range *range,
2909 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2913 while (idx < bus->dev_count &&
2914 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
2915 if (!kvm_iodevice_read(bus->range[idx].dev, range->addr,
2924 /* kvm_io_bus_read - called under kvm->slots_lock */
2925 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2928 struct kvm_io_bus *bus;
2929 struct kvm_io_range range;
2932 range = (struct kvm_io_range) {
2937 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2938 r = __kvm_io_bus_read(bus, &range, val);
2939 return r < 0 ? r : 0;
2943 /* Caller must hold slots_lock. */
2944 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2945 int len, struct kvm_io_device *dev)
2947 struct kvm_io_bus *new_bus, *bus;
2949 bus = kvm->buses[bus_idx];
2950 /* exclude ioeventfd which is limited by maximum fd */
2951 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
2954 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
2955 sizeof(struct kvm_io_range)), GFP_KERNEL);
2958 memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
2959 sizeof(struct kvm_io_range)));
2960 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2961 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2962 synchronize_srcu_expedited(&kvm->srcu);
2968 /* Caller must hold slots_lock. */
2969 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2970 struct kvm_io_device *dev)
2973 struct kvm_io_bus *new_bus, *bus;
2975 bus = kvm->buses[bus_idx];
2977 for (i = 0; i < bus->dev_count; i++)
2978 if (bus->range[i].dev == dev) {
2986 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
2987 sizeof(struct kvm_io_range)), GFP_KERNEL);
2991 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
2992 new_bus->dev_count--;
2993 memcpy(new_bus->range + i, bus->range + i + 1,
2994 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
2996 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2997 synchronize_srcu_expedited(&kvm->srcu);
3002 static struct notifier_block kvm_cpu_notifier = {
3003 .notifier_call = kvm_cpu_hotplug,
3006 static int vm_stat_get(void *_offset, u64 *val)
3008 unsigned offset = (long)_offset;
3012 spin_lock(&kvm_lock);
3013 list_for_each_entry(kvm, &vm_list, vm_list)
3014 *val += *(u32 *)((void *)kvm + offset);
3015 spin_unlock(&kvm_lock);
3019 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
3021 static int vcpu_stat_get(void *_offset, u64 *val)
3023 unsigned offset = (long)_offset;
3025 struct kvm_vcpu *vcpu;
3029 spin_lock(&kvm_lock);
3030 list_for_each_entry(kvm, &vm_list, vm_list)
3031 kvm_for_each_vcpu(i, vcpu, kvm)
3032 *val += *(u32 *)((void *)vcpu + offset);
3034 spin_unlock(&kvm_lock);
3038 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
3040 static const struct file_operations *stat_fops[] = {
3041 [KVM_STAT_VCPU] = &vcpu_stat_fops,
3042 [KVM_STAT_VM] = &vm_stat_fops,
3045 static int kvm_init_debug(void)
3048 struct kvm_stats_debugfs_item *p;
3050 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3051 if (kvm_debugfs_dir == NULL)
3054 for (p = debugfs_entries; p->name; ++p) {
3055 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
3056 (void *)(long)p->offset,
3057 stat_fops[p->kind]);
3058 if (p->dentry == NULL)
3065 debugfs_remove_recursive(kvm_debugfs_dir);
3070 static void kvm_exit_debug(void)
3072 struct kvm_stats_debugfs_item *p;
3074 for (p = debugfs_entries; p->name; ++p)
3075 debugfs_remove(p->dentry);
3076 debugfs_remove(kvm_debugfs_dir);
3079 static int kvm_suspend(void)
3081 if (kvm_usage_count)
3082 hardware_disable_nolock(NULL);
3086 static void kvm_resume(void)
3088 if (kvm_usage_count) {
3089 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3090 hardware_enable_nolock(NULL);
3094 static struct syscore_ops kvm_syscore_ops = {
3095 .suspend = kvm_suspend,
3096 .resume = kvm_resume,
3100 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3102 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3105 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3107 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3108 if (vcpu->preempted)
3109 vcpu->preempted = false;
3111 kvm_arch_vcpu_load(vcpu, cpu);
3114 static void kvm_sched_out(struct preempt_notifier *pn,
3115 struct task_struct *next)
3117 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3119 if (current->state == TASK_RUNNING)
3120 vcpu->preempted = true;
3121 kvm_arch_vcpu_put(vcpu);
3124 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3125 struct module *module)
3130 r = kvm_arch_init(opaque);
3135 * kvm_arch_init makes sure there's at most one caller
3136 * for architectures that support multiple implementations,
3137 * like intel and amd on x86.
3138 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3139 * conflicts in case kvm is already setup for another implementation.
3141 r = kvm_irqfd_init();
3145 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
3150 r = kvm_arch_hardware_setup();
3154 for_each_online_cpu(cpu) {
3155 smp_call_function_single(cpu,
3156 kvm_arch_check_processor_compat,
3162 r = register_cpu_notifier(&kvm_cpu_notifier);
3165 register_reboot_notifier(&kvm_reboot_notifier);
3167 /* A kmem cache lets us meet the alignment requirements of fx_save. */
3169 vcpu_align = __alignof__(struct kvm_vcpu);
3170 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
3172 if (!kvm_vcpu_cache) {
3177 r = kvm_async_pf_init();
3181 kvm_chardev_ops.owner = module;
3182 kvm_vm_fops.owner = module;
3183 kvm_vcpu_fops.owner = module;
3185 r = misc_register(&kvm_dev);
3187 printk(KERN_ERR "kvm: misc device register failed\n");
3191 register_syscore_ops(&kvm_syscore_ops);
3193 kvm_preempt_ops.sched_in = kvm_sched_in;
3194 kvm_preempt_ops.sched_out = kvm_sched_out;
3196 r = kvm_init_debug();
3198 printk(KERN_ERR "kvm: create debugfs files failed\n");
3205 unregister_syscore_ops(&kvm_syscore_ops);
3206 misc_deregister(&kvm_dev);
3208 kvm_async_pf_deinit();
3210 kmem_cache_destroy(kvm_vcpu_cache);
3212 unregister_reboot_notifier(&kvm_reboot_notifier);
3213 unregister_cpu_notifier(&kvm_cpu_notifier);
3216 kvm_arch_hardware_unsetup();
3218 free_cpumask_var(cpus_hardware_enabled);
3226 EXPORT_SYMBOL_GPL(kvm_init);
3231 misc_deregister(&kvm_dev);
3232 kmem_cache_destroy(kvm_vcpu_cache);
3233 kvm_async_pf_deinit();
3234 unregister_syscore_ops(&kvm_syscore_ops);
3235 unregister_reboot_notifier(&kvm_reboot_notifier);
3236 unregister_cpu_notifier(&kvm_cpu_notifier);
3237 on_each_cpu(hardware_disable_nolock, NULL, 1);
3238 kvm_arch_hardware_unsetup();
3241 free_cpumask_var(cpus_hardware_enabled);
3243 EXPORT_SYMBOL_GPL(kvm_exit);