2 * Kernel-based Virtual Machine driver for Linux
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
7 * Copyright (C) 2006 Qumranet, Inc.
8 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
11 * Avi Kivity <avi@qumranet.com>
12 * Yaniv Kamay <yaniv@qumranet.com>
14 * This work is licensed under the terms of the GNU GPL, version 2. See
15 * the COPYING file in the top-level directory.
21 #include <linux/kvm_host.h>
22 #include <linux/kvm.h>
23 #include <linux/module.h>
24 #include <linux/errno.h>
25 #include <linux/percpu.h>
27 #include <linux/miscdevice.h>
28 #include <linux/vmalloc.h>
29 #include <linux/reboot.h>
30 #include <linux/debugfs.h>
31 #include <linux/highmem.h>
32 #include <linux/file.h>
33 #include <linux/syscore_ops.h>
34 #include <linux/cpu.h>
35 #include <linux/sched.h>
36 #include <linux/cpumask.h>
37 #include <linux/smp.h>
38 #include <linux/anon_inodes.h>
39 #include <linux/profile.h>
40 #include <linux/kvm_para.h>
41 #include <linux/pagemap.h>
42 #include <linux/mman.h>
43 #include <linux/swap.h>
44 #include <linux/bitops.h>
45 #include <linux/spinlock.h>
46 #include <linux/compat.h>
47 #include <linux/srcu.h>
48 #include <linux/hugetlb.h>
49 #include <linux/slab.h>
50 #include <linux/sort.h>
51 #include <linux/bsearch.h>
53 #include <asm/processor.h>
55 #include <asm/uaccess.h>
56 #include <asm/pgtable.h>
58 #include "coalesced_mmio.h"
61 #define CREATE_TRACE_POINTS
62 #include <trace/events/kvm.h>
64 MODULE_AUTHOR("Qumranet");
65 MODULE_LICENSE("GPL");
70 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
73 DEFINE_RAW_SPINLOCK(kvm_lock);
76 static cpumask_var_t cpus_hardware_enabled;
77 static int kvm_usage_count = 0;
78 static atomic_t hardware_enable_failed;
80 struct kmem_cache *kvm_vcpu_cache;
81 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
83 static __read_mostly struct preempt_ops kvm_preempt_ops;
85 struct dentry *kvm_debugfs_dir;
87 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
90 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
93 static int hardware_enable_all(void);
94 static void hardware_disable_all(void);
96 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
99 EXPORT_SYMBOL_GPL(kvm_rebooting);
101 static bool largepages_enabled = true;
103 bool kvm_is_mmio_pfn(pfn_t pfn)
106 return PageReserved(pfn_to_page(pfn));
112 * Switches to specified vcpu, until a matching vcpu_put()
114 int vcpu_load(struct kvm_vcpu *vcpu)
118 if (mutex_lock_killable(&vcpu->mutex))
120 if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
121 /* The thread running this VCPU changed. */
122 struct pid *oldpid = vcpu->pid;
123 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
124 rcu_assign_pointer(vcpu->pid, newpid);
129 preempt_notifier_register(&vcpu->preempt_notifier);
130 kvm_arch_vcpu_load(vcpu, cpu);
135 void vcpu_put(struct kvm_vcpu *vcpu)
138 kvm_arch_vcpu_put(vcpu);
139 preempt_notifier_unregister(&vcpu->preempt_notifier);
141 mutex_unlock(&vcpu->mutex);
144 static void ack_flush(void *_completed)
148 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
153 struct kvm_vcpu *vcpu;
155 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
158 kvm_for_each_vcpu(i, vcpu, kvm) {
159 kvm_make_request(req, vcpu);
162 /* Set ->requests bit before we read ->mode */
165 if (cpus != NULL && cpu != -1 && cpu != me &&
166 kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
167 cpumask_set_cpu(cpu, cpus);
169 if (unlikely(cpus == NULL))
170 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
171 else if (!cpumask_empty(cpus))
172 smp_call_function_many(cpus, ack_flush, NULL, 1);
176 free_cpumask_var(cpus);
180 void kvm_flush_remote_tlbs(struct kvm *kvm)
182 long dirty_count = kvm->tlbs_dirty;
185 if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
186 ++kvm->stat.remote_tlb_flush;
187 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
190 void kvm_reload_remote_mmus(struct kvm *kvm)
192 make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
195 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
197 make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
200 void kvm_make_scan_ioapic_request(struct kvm *kvm)
202 make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
205 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
210 mutex_init(&vcpu->mutex);
215 init_waitqueue_head(&vcpu->wq);
216 kvm_async_pf_vcpu_init(vcpu);
218 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
223 vcpu->run = page_address(page);
225 kvm_vcpu_set_in_spin_loop(vcpu, false);
226 kvm_vcpu_set_dy_eligible(vcpu, false);
227 vcpu->preempted = false;
229 r = kvm_arch_vcpu_init(vcpu);
235 free_page((unsigned long)vcpu->run);
239 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
241 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
244 kvm_arch_vcpu_uninit(vcpu);
245 free_page((unsigned long)vcpu->run);
247 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
249 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
250 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
252 return container_of(mn, struct kvm, mmu_notifier);
255 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
256 struct mm_struct *mm,
257 unsigned long address)
259 struct kvm *kvm = mmu_notifier_to_kvm(mn);
260 int need_tlb_flush, idx;
263 * When ->invalidate_page runs, the linux pte has been zapped
264 * already but the page is still allocated until
265 * ->invalidate_page returns. So if we increase the sequence
266 * here the kvm page fault will notice if the spte can't be
267 * established because the page is going to be freed. If
268 * instead the kvm page fault establishes the spte before
269 * ->invalidate_page runs, kvm_unmap_hva will release it
272 * The sequence increase only need to be seen at spin_unlock
273 * time, and not at spin_lock time.
275 * Increasing the sequence after the spin_unlock would be
276 * unsafe because the kvm page fault could then establish the
277 * pte after kvm_unmap_hva returned, without noticing the page
278 * is going to be freed.
280 idx = srcu_read_lock(&kvm->srcu);
281 spin_lock(&kvm->mmu_lock);
283 kvm->mmu_notifier_seq++;
284 need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
285 /* we've to flush the tlb before the pages can be freed */
287 kvm_flush_remote_tlbs(kvm);
289 spin_unlock(&kvm->mmu_lock);
290 srcu_read_unlock(&kvm->srcu, idx);
293 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
294 struct mm_struct *mm,
295 unsigned long address,
298 struct kvm *kvm = mmu_notifier_to_kvm(mn);
301 idx = srcu_read_lock(&kvm->srcu);
302 spin_lock(&kvm->mmu_lock);
303 kvm->mmu_notifier_seq++;
304 kvm_set_spte_hva(kvm, address, pte);
305 spin_unlock(&kvm->mmu_lock);
306 srcu_read_unlock(&kvm->srcu, idx);
309 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
310 struct mm_struct *mm,
314 struct kvm *kvm = mmu_notifier_to_kvm(mn);
315 int need_tlb_flush = 0, idx;
317 idx = srcu_read_lock(&kvm->srcu);
318 spin_lock(&kvm->mmu_lock);
320 * The count increase must become visible at unlock time as no
321 * spte can be established without taking the mmu_lock and
322 * count is also read inside the mmu_lock critical section.
324 kvm->mmu_notifier_count++;
325 need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
326 need_tlb_flush |= kvm->tlbs_dirty;
327 /* we've to flush the tlb before the pages can be freed */
329 kvm_flush_remote_tlbs(kvm);
331 spin_unlock(&kvm->mmu_lock);
332 srcu_read_unlock(&kvm->srcu, idx);
335 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
336 struct mm_struct *mm,
340 struct kvm *kvm = mmu_notifier_to_kvm(mn);
342 spin_lock(&kvm->mmu_lock);
344 * This sequence increase will notify the kvm page fault that
345 * the page that is going to be mapped in the spte could have
348 kvm->mmu_notifier_seq++;
351 * The above sequence increase must be visible before the
352 * below count decrease, which is ensured by the smp_wmb above
353 * in conjunction with the smp_rmb in mmu_notifier_retry().
355 kvm->mmu_notifier_count--;
356 spin_unlock(&kvm->mmu_lock);
358 BUG_ON(kvm->mmu_notifier_count < 0);
361 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
362 struct mm_struct *mm,
363 unsigned long address)
365 struct kvm *kvm = mmu_notifier_to_kvm(mn);
368 idx = srcu_read_lock(&kvm->srcu);
369 spin_lock(&kvm->mmu_lock);
371 young = kvm_age_hva(kvm, address);
373 kvm_flush_remote_tlbs(kvm);
375 spin_unlock(&kvm->mmu_lock);
376 srcu_read_unlock(&kvm->srcu, idx);
381 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
382 struct mm_struct *mm,
383 unsigned long address)
385 struct kvm *kvm = mmu_notifier_to_kvm(mn);
388 idx = srcu_read_lock(&kvm->srcu);
389 spin_lock(&kvm->mmu_lock);
390 young = kvm_test_age_hva(kvm, address);
391 spin_unlock(&kvm->mmu_lock);
392 srcu_read_unlock(&kvm->srcu, idx);
397 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
398 struct mm_struct *mm)
400 struct kvm *kvm = mmu_notifier_to_kvm(mn);
403 idx = srcu_read_lock(&kvm->srcu);
404 kvm_arch_flush_shadow_all(kvm);
405 srcu_read_unlock(&kvm->srcu, idx);
408 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
409 .invalidate_page = kvm_mmu_notifier_invalidate_page,
410 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
411 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
412 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
413 .test_young = kvm_mmu_notifier_test_young,
414 .change_pte = kvm_mmu_notifier_change_pte,
415 .release = kvm_mmu_notifier_release,
418 static int kvm_init_mmu_notifier(struct kvm *kvm)
420 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
421 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
424 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
426 static int kvm_init_mmu_notifier(struct kvm *kvm)
431 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
433 static void kvm_init_memslots_id(struct kvm *kvm)
436 struct kvm_memslots *slots = kvm->memslots;
438 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
439 slots->id_to_index[i] = slots->memslots[i].id = i;
442 static struct kvm *kvm_create_vm(unsigned long type)
445 struct kvm *kvm = kvm_arch_alloc_vm();
448 return ERR_PTR(-ENOMEM);
450 r = kvm_arch_init_vm(kvm, type);
452 goto out_err_nodisable;
454 r = hardware_enable_all();
456 goto out_err_nodisable;
458 #ifdef CONFIG_HAVE_KVM_IRQCHIP
459 INIT_HLIST_HEAD(&kvm->mask_notifier_list);
460 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
463 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
466 kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
469 kvm_init_memslots_id(kvm);
470 if (init_srcu_struct(&kvm->srcu))
472 for (i = 0; i < KVM_NR_BUSES; i++) {
473 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
479 spin_lock_init(&kvm->mmu_lock);
480 kvm->mm = current->mm;
481 atomic_inc(&kvm->mm->mm_count);
482 kvm_eventfd_init(kvm);
483 mutex_init(&kvm->lock);
484 mutex_init(&kvm->irq_lock);
485 mutex_init(&kvm->slots_lock);
486 atomic_set(&kvm->users_count, 1);
487 INIT_LIST_HEAD(&kvm->devices);
489 r = kvm_init_mmu_notifier(kvm);
493 raw_spin_lock(&kvm_lock);
494 list_add(&kvm->vm_list, &vm_list);
495 raw_spin_unlock(&kvm_lock);
500 cleanup_srcu_struct(&kvm->srcu);
502 hardware_disable_all();
504 for (i = 0; i < KVM_NR_BUSES; i++)
505 kfree(kvm->buses[i]);
506 kfree(kvm->memslots);
507 kvm_arch_free_vm(kvm);
512 * Avoid using vmalloc for a small buffer.
513 * Should not be used when the size is statically known.
515 void *kvm_kvzalloc(unsigned long size)
517 if (size > PAGE_SIZE)
518 return vzalloc(size);
520 return kzalloc(size, GFP_KERNEL);
523 void kvm_kvfree(const void *addr)
525 if (is_vmalloc_addr(addr))
531 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
533 if (!memslot->dirty_bitmap)
536 kvm_kvfree(memslot->dirty_bitmap);
537 memslot->dirty_bitmap = NULL;
541 * Free any memory in @free but not in @dont.
543 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
544 struct kvm_memory_slot *dont)
546 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
547 kvm_destroy_dirty_bitmap(free);
549 kvm_arch_free_memslot(free, dont);
554 void kvm_free_physmem(struct kvm *kvm)
556 struct kvm_memslots *slots = kvm->memslots;
557 struct kvm_memory_slot *memslot;
559 kvm_for_each_memslot(memslot, slots)
560 kvm_free_physmem_slot(memslot, NULL);
562 kfree(kvm->memslots);
565 static void kvm_destroy_devices(struct kvm *kvm)
567 struct list_head *node, *tmp;
569 list_for_each_safe(node, tmp, &kvm->devices) {
570 struct kvm_device *dev =
571 list_entry(node, struct kvm_device, vm_node);
574 dev->ops->destroy(dev);
578 static void kvm_destroy_vm(struct kvm *kvm)
581 struct mm_struct *mm = kvm->mm;
583 kvm_arch_sync_events(kvm);
584 raw_spin_lock(&kvm_lock);
585 list_del(&kvm->vm_list);
586 raw_spin_unlock(&kvm_lock);
587 kvm_free_irq_routing(kvm);
588 for (i = 0; i < KVM_NR_BUSES; i++)
589 kvm_io_bus_destroy(kvm->buses[i]);
590 kvm_coalesced_mmio_free(kvm);
591 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
592 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
594 kvm_arch_flush_shadow_all(kvm);
596 kvm_arch_destroy_vm(kvm);
597 kvm_destroy_devices(kvm);
598 kvm_free_physmem(kvm);
599 cleanup_srcu_struct(&kvm->srcu);
600 kvm_arch_free_vm(kvm);
601 hardware_disable_all();
605 void kvm_get_kvm(struct kvm *kvm)
607 atomic_inc(&kvm->users_count);
609 EXPORT_SYMBOL_GPL(kvm_get_kvm);
611 void kvm_put_kvm(struct kvm *kvm)
613 if (atomic_dec_and_test(&kvm->users_count))
616 EXPORT_SYMBOL_GPL(kvm_put_kvm);
619 static int kvm_vm_release(struct inode *inode, struct file *filp)
621 struct kvm *kvm = filp->private_data;
623 kvm_irqfd_release(kvm);
630 * Allocation size is twice as large as the actual dirty bitmap size.
631 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
633 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
636 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
638 memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
639 if (!memslot->dirty_bitmap)
642 #endif /* !CONFIG_S390 */
646 static int cmp_memslot(const void *slot1, const void *slot2)
648 struct kvm_memory_slot *s1, *s2;
650 s1 = (struct kvm_memory_slot *)slot1;
651 s2 = (struct kvm_memory_slot *)slot2;
653 if (s1->npages < s2->npages)
655 if (s1->npages > s2->npages)
662 * Sort the memslots base on its size, so the larger slots
663 * will get better fit.
665 static void sort_memslots(struct kvm_memslots *slots)
669 sort(slots->memslots, KVM_MEM_SLOTS_NUM,
670 sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
672 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
673 slots->id_to_index[slots->memslots[i].id] = i;
676 void update_memslots(struct kvm_memslots *slots, struct kvm_memory_slot *new,
681 struct kvm_memory_slot *old = id_to_memslot(slots, id);
682 unsigned long npages = old->npages;
685 if (new->npages != npages)
686 sort_memslots(slots);
689 slots->generation = last_generation + 1;
692 static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
694 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
696 #ifdef KVM_CAP_READONLY_MEM
697 valid_flags |= KVM_MEM_READONLY;
700 if (mem->flags & ~valid_flags)
706 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
707 struct kvm_memslots *slots, struct kvm_memory_slot *new)
709 struct kvm_memslots *old_memslots = kvm->memslots;
711 update_memslots(slots, new, kvm->memslots->generation);
712 rcu_assign_pointer(kvm->memslots, slots);
713 synchronize_srcu_expedited(&kvm->srcu);
715 kvm_arch_memslots_updated(kvm);
721 * Allocate some memory and give it an address in the guest physical address
724 * Discontiguous memory is allowed, mostly for framebuffers.
726 * Must be called holding mmap_sem for write.
728 int __kvm_set_memory_region(struct kvm *kvm,
729 struct kvm_userspace_memory_region *mem)
733 unsigned long npages;
734 struct kvm_memory_slot *slot;
735 struct kvm_memory_slot old, new;
736 struct kvm_memslots *slots = NULL, *old_memslots;
737 enum kvm_mr_change change;
739 r = check_memory_region_flags(mem);
744 /* General sanity checks */
745 if (mem->memory_size & (PAGE_SIZE - 1))
747 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
749 /* We can read the guest memory with __xxx_user() later on. */
750 if ((mem->slot < KVM_USER_MEM_SLOTS) &&
751 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
752 !access_ok(VERIFY_WRITE,
753 (void __user *)(unsigned long)mem->userspace_addr,
756 if (mem->slot >= KVM_MEM_SLOTS_NUM)
758 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
761 slot = id_to_memslot(kvm->memslots, mem->slot);
762 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
763 npages = mem->memory_size >> PAGE_SHIFT;
766 if (npages > KVM_MEM_MAX_NR_PAGES)
770 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
775 new.base_gfn = base_gfn;
777 new.flags = mem->flags;
782 change = KVM_MR_CREATE;
783 else { /* Modify an existing slot. */
784 if ((mem->userspace_addr != old.userspace_addr) ||
785 (npages != old.npages) ||
786 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
789 if (base_gfn != old.base_gfn)
790 change = KVM_MR_MOVE;
791 else if (new.flags != old.flags)
792 change = KVM_MR_FLAGS_ONLY;
793 else { /* Nothing to change. */
798 } else if (old.npages) {
799 change = KVM_MR_DELETE;
800 } else /* Modify a non-existent slot: disallowed. */
803 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
804 /* Check for overlaps */
806 kvm_for_each_memslot(slot, kvm->memslots) {
807 if ((slot->id >= KVM_USER_MEM_SLOTS) ||
808 (slot->id == mem->slot))
810 if (!((base_gfn + npages <= slot->base_gfn) ||
811 (base_gfn >= slot->base_gfn + slot->npages)))
816 /* Free page dirty bitmap if unneeded */
817 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
818 new.dirty_bitmap = NULL;
821 if (change == KVM_MR_CREATE) {
822 new.userspace_addr = mem->userspace_addr;
824 if (kvm_arch_create_memslot(&new, npages))
828 /* Allocate page dirty bitmap if needed */
829 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
830 if (kvm_create_dirty_bitmap(&new) < 0)
834 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
836 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
840 slot = id_to_memslot(slots, mem->slot);
841 slot->flags |= KVM_MEMSLOT_INVALID;
843 old_memslots = install_new_memslots(kvm, slots, NULL);
845 /* slot was deleted or moved, clear iommu mapping */
846 kvm_iommu_unmap_pages(kvm, &old);
847 /* From this point no new shadow pages pointing to a deleted,
848 * or moved, memslot will be created.
850 * validation of sp->gfn happens in:
851 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
852 * - kvm_is_visible_gfn (mmu_check_roots)
854 kvm_arch_flush_shadow_memslot(kvm, slot);
855 slots = old_memslots;
858 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
864 * We can re-use the old_memslots from above, the only difference
865 * from the currently installed memslots is the invalid flag. This
866 * will get overwritten by update_memslots anyway.
869 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
876 * IOMMU mapping: New slots need to be mapped. Old slots need to be
877 * un-mapped and re-mapped if their base changes. Since base change
878 * unmapping is handled above with slot deletion, mapping alone is
879 * needed here. Anything else the iommu might care about for existing
880 * slots (size changes, userspace addr changes and read-only flag
881 * changes) is disallowed above, so any other attribute changes getting
882 * here can be skipped.
884 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
885 r = kvm_iommu_map_pages(kvm, &new);
890 /* actual memory is freed via old in kvm_free_physmem_slot below */
891 if (change == KVM_MR_DELETE) {
892 new.dirty_bitmap = NULL;
893 memset(&new.arch, 0, sizeof(new.arch));
896 old_memslots = install_new_memslots(kvm, slots, &new);
898 kvm_arch_commit_memory_region(kvm, mem, &old, change);
900 kvm_free_physmem_slot(&old, &new);
908 kvm_free_physmem_slot(&new, &old);
912 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
914 int kvm_set_memory_region(struct kvm *kvm,
915 struct kvm_userspace_memory_region *mem)
919 mutex_lock(&kvm->slots_lock);
920 r = __kvm_set_memory_region(kvm, mem);
921 mutex_unlock(&kvm->slots_lock);
924 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
926 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
927 struct kvm_userspace_memory_region *mem)
929 if (mem->slot >= KVM_USER_MEM_SLOTS)
931 return kvm_set_memory_region(kvm, mem);
934 int kvm_get_dirty_log(struct kvm *kvm,
935 struct kvm_dirty_log *log, int *is_dirty)
937 struct kvm_memory_slot *memslot;
940 unsigned long any = 0;
943 if (log->slot >= KVM_USER_MEM_SLOTS)
946 memslot = id_to_memslot(kvm->memslots, log->slot);
948 if (!memslot->dirty_bitmap)
951 n = kvm_dirty_bitmap_bytes(memslot);
953 for (i = 0; !any && i < n/sizeof(long); ++i)
954 any = memslot->dirty_bitmap[i];
957 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
968 bool kvm_largepages_enabled(void)
970 return largepages_enabled;
973 void kvm_disable_largepages(void)
975 largepages_enabled = false;
977 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
979 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
981 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
983 EXPORT_SYMBOL_GPL(gfn_to_memslot);
985 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
987 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
989 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
990 memslot->flags & KVM_MEMSLOT_INVALID)
995 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
997 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
999 struct vm_area_struct *vma;
1000 unsigned long addr, size;
1004 addr = gfn_to_hva(kvm, gfn);
1005 if (kvm_is_error_hva(addr))
1008 down_read(¤t->mm->mmap_sem);
1009 vma = find_vma(current->mm, addr);
1013 size = vma_kernel_pagesize(vma);
1016 up_read(¤t->mm->mmap_sem);
1021 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1023 return slot->flags & KVM_MEM_READONLY;
1026 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1027 gfn_t *nr_pages, bool write)
1029 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1030 return KVM_HVA_ERR_BAD;
1032 if (memslot_is_readonly(slot) && write)
1033 return KVM_HVA_ERR_RO_BAD;
1036 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1038 return __gfn_to_hva_memslot(slot, gfn);
1041 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1044 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1047 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1050 return gfn_to_hva_many(slot, gfn, NULL);
1052 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1054 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1056 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1058 EXPORT_SYMBOL_GPL(gfn_to_hva);
1061 * The hva returned by this function is only allowed to be read.
1062 * It should pair with kvm_read_hva() or kvm_read_hva_atomic().
1064 static unsigned long gfn_to_hva_read(struct kvm *kvm, gfn_t gfn)
1066 return __gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL, false);
1069 static int kvm_read_hva(void *data, void __user *hva, int len)
1071 return __copy_from_user(data, hva, len);
1074 static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1076 return __copy_from_user_inatomic(data, hva, len);
1079 static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1080 unsigned long start, int write, struct page **page)
1082 int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1085 flags |= FOLL_WRITE;
1087 return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1090 static inline int check_user_page_hwpoison(unsigned long addr)
1092 int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1094 rc = __get_user_pages(current, current->mm, addr, 1,
1095 flags, NULL, NULL, NULL);
1096 return rc == -EHWPOISON;
1100 * The atomic path to get the writable pfn which will be stored in @pfn,
1101 * true indicates success, otherwise false is returned.
1103 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1104 bool write_fault, bool *writable, pfn_t *pfn)
1106 struct page *page[1];
1109 if (!(async || atomic))
1113 * Fast pin a writable pfn only if it is a write fault request
1114 * or the caller allows to map a writable pfn for a read fault
1117 if (!(write_fault || writable))
1120 npages = __get_user_pages_fast(addr, 1, 1, page);
1122 *pfn = page_to_pfn(page[0]);
1133 * The slow path to get the pfn of the specified host virtual address,
1134 * 1 indicates success, -errno is returned if error is detected.
1136 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1137 bool *writable, pfn_t *pfn)
1139 struct page *page[1];
1145 *writable = write_fault;
1148 down_read(¤t->mm->mmap_sem);
1149 npages = get_user_page_nowait(current, current->mm,
1150 addr, write_fault, page);
1151 up_read(¤t->mm->mmap_sem);
1153 npages = get_user_pages_fast(addr, 1, write_fault,
1158 /* map read fault as writable if possible */
1159 if (unlikely(!write_fault) && writable) {
1160 struct page *wpage[1];
1162 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1171 *pfn = page_to_pfn(page[0]);
1175 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1177 if (unlikely(!(vma->vm_flags & VM_READ)))
1180 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1187 * Pin guest page in memory and return its pfn.
1188 * @addr: host virtual address which maps memory to the guest
1189 * @atomic: whether this function can sleep
1190 * @async: whether this function need to wait IO complete if the
1191 * host page is not in the memory
1192 * @write_fault: whether we should get a writable host page
1193 * @writable: whether it allows to map a writable host page for !@write_fault
1195 * The function will map a writable host page for these two cases:
1196 * 1): @write_fault = true
1197 * 2): @write_fault = false && @writable, @writable will tell the caller
1198 * whether the mapping is writable.
1200 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1201 bool write_fault, bool *writable)
1203 struct vm_area_struct *vma;
1207 /* we can do it either atomically or asynchronously, not both */
1208 BUG_ON(atomic && async);
1210 if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1214 return KVM_PFN_ERR_FAULT;
1216 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1220 down_read(¤t->mm->mmap_sem);
1221 if (npages == -EHWPOISON ||
1222 (!async && check_user_page_hwpoison(addr))) {
1223 pfn = KVM_PFN_ERR_HWPOISON;
1227 vma = find_vma_intersection(current->mm, addr, addr + 1);
1230 pfn = KVM_PFN_ERR_FAULT;
1231 else if ((vma->vm_flags & VM_PFNMAP)) {
1232 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1234 BUG_ON(!kvm_is_mmio_pfn(pfn));
1236 if (async && vma_is_valid(vma, write_fault))
1238 pfn = KVM_PFN_ERR_FAULT;
1241 up_read(¤t->mm->mmap_sem);
1246 __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1247 bool *async, bool write_fault, bool *writable)
1249 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1251 if (addr == KVM_HVA_ERR_RO_BAD)
1252 return KVM_PFN_ERR_RO_FAULT;
1254 if (kvm_is_error_hva(addr))
1255 return KVM_PFN_NOSLOT;
1257 /* Do not map writable pfn in the readonly memslot. */
1258 if (writable && memslot_is_readonly(slot)) {
1263 return hva_to_pfn(addr, atomic, async, write_fault,
1267 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1268 bool write_fault, bool *writable)
1270 struct kvm_memory_slot *slot;
1275 slot = gfn_to_memslot(kvm, gfn);
1277 return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1281 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1283 return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1285 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1287 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1288 bool write_fault, bool *writable)
1290 return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1292 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1294 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1296 return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1298 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1300 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1303 return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1305 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1307 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1309 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1312 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1314 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1316 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1318 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1324 addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1325 if (kvm_is_error_hva(addr))
1328 if (entry < nr_pages)
1331 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1333 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1335 static struct page *kvm_pfn_to_page(pfn_t pfn)
1337 if (is_error_noslot_pfn(pfn))
1338 return KVM_ERR_PTR_BAD_PAGE;
1340 if (kvm_is_mmio_pfn(pfn)) {
1342 return KVM_ERR_PTR_BAD_PAGE;
1345 return pfn_to_page(pfn);
1348 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1352 pfn = gfn_to_pfn(kvm, gfn);
1354 return kvm_pfn_to_page(pfn);
1357 EXPORT_SYMBOL_GPL(gfn_to_page);
1359 void kvm_release_page_clean(struct page *page)
1361 WARN_ON(is_error_page(page));
1363 kvm_release_pfn_clean(page_to_pfn(page));
1365 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1367 void kvm_release_pfn_clean(pfn_t pfn)
1369 if (!is_error_noslot_pfn(pfn) && !kvm_is_mmio_pfn(pfn))
1370 put_page(pfn_to_page(pfn));
1372 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1374 void kvm_release_page_dirty(struct page *page)
1376 WARN_ON(is_error_page(page));
1378 kvm_release_pfn_dirty(page_to_pfn(page));
1380 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1382 void kvm_release_pfn_dirty(pfn_t pfn)
1384 kvm_set_pfn_dirty(pfn);
1385 kvm_release_pfn_clean(pfn);
1387 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1389 void kvm_set_page_dirty(struct page *page)
1391 kvm_set_pfn_dirty(page_to_pfn(page));
1393 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1395 void kvm_set_pfn_dirty(pfn_t pfn)
1397 if (!kvm_is_mmio_pfn(pfn)) {
1398 struct page *page = pfn_to_page(pfn);
1399 if (!PageReserved(page))
1403 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1405 void kvm_set_pfn_accessed(pfn_t pfn)
1407 if (!kvm_is_mmio_pfn(pfn))
1408 mark_page_accessed(pfn_to_page(pfn));
1410 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1412 void kvm_get_pfn(pfn_t pfn)
1414 if (!kvm_is_mmio_pfn(pfn))
1415 get_page(pfn_to_page(pfn));
1417 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1419 static int next_segment(unsigned long len, int offset)
1421 if (len > PAGE_SIZE - offset)
1422 return PAGE_SIZE - offset;
1427 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1433 addr = gfn_to_hva_read(kvm, gfn);
1434 if (kvm_is_error_hva(addr))
1436 r = kvm_read_hva(data, (void __user *)addr + offset, len);
1441 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1443 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1445 gfn_t gfn = gpa >> PAGE_SHIFT;
1447 int offset = offset_in_page(gpa);
1450 while ((seg = next_segment(len, offset)) != 0) {
1451 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1461 EXPORT_SYMBOL_GPL(kvm_read_guest);
1463 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1468 gfn_t gfn = gpa >> PAGE_SHIFT;
1469 int offset = offset_in_page(gpa);
1471 addr = gfn_to_hva_read(kvm, gfn);
1472 if (kvm_is_error_hva(addr))
1474 pagefault_disable();
1475 r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
1481 EXPORT_SYMBOL(kvm_read_guest_atomic);
1483 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1484 int offset, int len)
1489 addr = gfn_to_hva(kvm, gfn);
1490 if (kvm_is_error_hva(addr))
1492 r = __copy_to_user((void __user *)addr + offset, data, len);
1495 mark_page_dirty(kvm, gfn);
1498 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1500 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1503 gfn_t gfn = gpa >> PAGE_SHIFT;
1505 int offset = offset_in_page(gpa);
1508 while ((seg = next_segment(len, offset)) != 0) {
1509 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1520 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1521 gpa_t gpa, unsigned long len)
1523 struct kvm_memslots *slots = kvm_memslots(kvm);
1524 int offset = offset_in_page(gpa);
1525 gfn_t start_gfn = gpa >> PAGE_SHIFT;
1526 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1527 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1528 gfn_t nr_pages_avail;
1531 ghc->generation = slots->generation;
1533 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1534 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, &nr_pages_avail);
1535 if (!kvm_is_error_hva(ghc->hva) && nr_pages_avail >= nr_pages_needed) {
1539 * If the requested region crosses two memslots, we still
1540 * verify that the entire region is valid here.
1542 while (start_gfn <= end_gfn) {
1543 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1544 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1546 if (kvm_is_error_hva(ghc->hva))
1548 start_gfn += nr_pages_avail;
1550 /* Use the slow path for cross page reads and writes. */
1551 ghc->memslot = NULL;
1555 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1557 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1558 void *data, unsigned long len)
1560 struct kvm_memslots *slots = kvm_memslots(kvm);
1563 BUG_ON(len > ghc->len);
1565 if (slots->generation != ghc->generation)
1566 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1568 if (unlikely(!ghc->memslot))
1569 return kvm_write_guest(kvm, ghc->gpa, data, len);
1571 if (kvm_is_error_hva(ghc->hva))
1574 r = __copy_to_user((void __user *)ghc->hva, data, len);
1577 mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1581 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1583 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1584 void *data, unsigned long len)
1586 struct kvm_memslots *slots = kvm_memslots(kvm);
1589 BUG_ON(len > ghc->len);
1591 if (slots->generation != ghc->generation)
1592 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1594 if (unlikely(!ghc->memslot))
1595 return kvm_read_guest(kvm, ghc->gpa, data, len);
1597 if (kvm_is_error_hva(ghc->hva))
1600 r = __copy_from_user(data, (void __user *)ghc->hva, len);
1606 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1608 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1610 return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1613 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1615 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1617 gfn_t gfn = gpa >> PAGE_SHIFT;
1619 int offset = offset_in_page(gpa);
1622 while ((seg = next_segment(len, offset)) != 0) {
1623 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1632 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1634 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1637 if (memslot && memslot->dirty_bitmap) {
1638 unsigned long rel_gfn = gfn - memslot->base_gfn;
1640 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1644 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1646 struct kvm_memory_slot *memslot;
1648 memslot = gfn_to_memslot(kvm, gfn);
1649 mark_page_dirty_in_slot(kvm, memslot, gfn);
1653 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1655 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1660 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1662 if (kvm_arch_vcpu_runnable(vcpu)) {
1663 kvm_make_request(KVM_REQ_UNHALT, vcpu);
1666 if (kvm_cpu_has_pending_timer(vcpu))
1668 if (signal_pending(current))
1674 finish_wait(&vcpu->wq, &wait);
1679 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1681 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1684 int cpu = vcpu->cpu;
1685 wait_queue_head_t *wqp;
1687 wqp = kvm_arch_vcpu_wq(vcpu);
1688 if (waitqueue_active(wqp)) {
1689 wake_up_interruptible(wqp);
1690 ++vcpu->stat.halt_wakeup;
1694 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1695 if (kvm_arch_vcpu_should_kick(vcpu))
1696 smp_send_reschedule(cpu);
1699 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
1700 #endif /* !CONFIG_S390 */
1702 void kvm_resched(struct kvm_vcpu *vcpu)
1704 if (!need_resched())
1708 EXPORT_SYMBOL_GPL(kvm_resched);
1710 bool kvm_vcpu_yield_to(struct kvm_vcpu *target)
1713 struct task_struct *task = NULL;
1717 pid = rcu_dereference(target->pid);
1719 task = get_pid_task(target->pid, PIDTYPE_PID);
1723 if (task->flags & PF_VCPU) {
1724 put_task_struct(task);
1727 ret = yield_to(task, 1);
1728 put_task_struct(task);
1732 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1734 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1736 * Helper that checks whether a VCPU is eligible for directed yield.
1737 * Most eligible candidate to yield is decided by following heuristics:
1739 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1740 * (preempted lock holder), indicated by @in_spin_loop.
1741 * Set at the beiginning and cleared at the end of interception/PLE handler.
1743 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1744 * chance last time (mostly it has become eligible now since we have probably
1745 * yielded to lockholder in last iteration. This is done by toggling
1746 * @dy_eligible each time a VCPU checked for eligibility.)
1748 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1749 * to preempted lock-holder could result in wrong VCPU selection and CPU
1750 * burning. Giving priority for a potential lock-holder increases lock
1753 * Since algorithm is based on heuristics, accessing another VCPU data without
1754 * locking does not harm. It may result in trying to yield to same VCPU, fail
1755 * and continue with next VCPU and so on.
1757 bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1761 eligible = !vcpu->spin_loop.in_spin_loop ||
1762 (vcpu->spin_loop.in_spin_loop &&
1763 vcpu->spin_loop.dy_eligible);
1765 if (vcpu->spin_loop.in_spin_loop)
1766 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1772 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1774 struct kvm *kvm = me->kvm;
1775 struct kvm_vcpu *vcpu;
1776 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1782 kvm_vcpu_set_in_spin_loop(me, true);
1784 * We boost the priority of a VCPU that is runnable but not
1785 * currently running, because it got preempted by something
1786 * else and called schedule in __vcpu_run. Hopefully that
1787 * VCPU is holding the lock that we need and will release it.
1788 * We approximate round-robin by starting at the last boosted VCPU.
1790 for (pass = 0; pass < 2 && !yielded && try; pass++) {
1791 kvm_for_each_vcpu(i, vcpu, kvm) {
1792 if (!pass && i <= last_boosted_vcpu) {
1793 i = last_boosted_vcpu;
1795 } else if (pass && i > last_boosted_vcpu)
1797 if (!ACCESS_ONCE(vcpu->preempted))
1801 if (waitqueue_active(&vcpu->wq))
1803 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1806 yielded = kvm_vcpu_yield_to(vcpu);
1808 kvm->last_boosted_vcpu = i;
1810 } else if (yielded < 0) {
1817 kvm_vcpu_set_in_spin_loop(me, false);
1819 /* Ensure vcpu is not eligible during next spinloop */
1820 kvm_vcpu_set_dy_eligible(me, false);
1822 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1824 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1826 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1829 if (vmf->pgoff == 0)
1830 page = virt_to_page(vcpu->run);
1832 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1833 page = virt_to_page(vcpu->arch.pio_data);
1835 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1836 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1837 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1840 return kvm_arch_vcpu_fault(vcpu, vmf);
1846 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1847 .fault = kvm_vcpu_fault,
1850 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1852 vma->vm_ops = &kvm_vcpu_vm_ops;
1856 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1858 struct kvm_vcpu *vcpu = filp->private_data;
1860 kvm_put_kvm(vcpu->kvm);
1864 static struct file_operations kvm_vcpu_fops = {
1865 .release = kvm_vcpu_release,
1866 .unlocked_ioctl = kvm_vcpu_ioctl,
1867 #ifdef CONFIG_COMPAT
1868 .compat_ioctl = kvm_vcpu_compat_ioctl,
1870 .mmap = kvm_vcpu_mmap,
1871 .llseek = noop_llseek,
1875 * Allocates an inode for the vcpu.
1877 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1879 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
1883 * Creates some virtual cpus. Good luck creating more than one.
1885 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1888 struct kvm_vcpu *vcpu, *v;
1890 vcpu = kvm_arch_vcpu_create(kvm, id);
1892 return PTR_ERR(vcpu);
1894 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1896 r = kvm_arch_vcpu_setup(vcpu);
1900 mutex_lock(&kvm->lock);
1901 if (!kvm_vcpu_compatible(vcpu)) {
1903 goto unlock_vcpu_destroy;
1905 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1907 goto unlock_vcpu_destroy;
1910 kvm_for_each_vcpu(r, v, kvm)
1911 if (v->vcpu_id == id) {
1913 goto unlock_vcpu_destroy;
1916 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1918 /* Now it's all set up, let userspace reach it */
1920 r = create_vcpu_fd(vcpu);
1923 goto unlock_vcpu_destroy;
1926 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1928 atomic_inc(&kvm->online_vcpus);
1930 mutex_unlock(&kvm->lock);
1931 kvm_arch_vcpu_postcreate(vcpu);
1934 unlock_vcpu_destroy:
1935 mutex_unlock(&kvm->lock);
1937 kvm_arch_vcpu_destroy(vcpu);
1941 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1944 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1945 vcpu->sigset_active = 1;
1946 vcpu->sigset = *sigset;
1948 vcpu->sigset_active = 0;
1952 static long kvm_vcpu_ioctl(struct file *filp,
1953 unsigned int ioctl, unsigned long arg)
1955 struct kvm_vcpu *vcpu = filp->private_data;
1956 void __user *argp = (void __user *)arg;
1958 struct kvm_fpu *fpu = NULL;
1959 struct kvm_sregs *kvm_sregs = NULL;
1961 if (vcpu->kvm->mm != current->mm)
1964 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
1966 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1967 * so vcpu_load() would break it.
1969 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1970 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1974 r = vcpu_load(vcpu);
1982 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1983 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1985 case KVM_GET_REGS: {
1986 struct kvm_regs *kvm_regs;
1989 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1992 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1996 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2003 case KVM_SET_REGS: {
2004 struct kvm_regs *kvm_regs;
2007 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2008 if (IS_ERR(kvm_regs)) {
2009 r = PTR_ERR(kvm_regs);
2012 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2016 case KVM_GET_SREGS: {
2017 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2021 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2025 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2030 case KVM_SET_SREGS: {
2031 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2032 if (IS_ERR(kvm_sregs)) {
2033 r = PTR_ERR(kvm_sregs);
2037 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2040 case KVM_GET_MP_STATE: {
2041 struct kvm_mp_state mp_state;
2043 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2047 if (copy_to_user(argp, &mp_state, sizeof mp_state))
2052 case KVM_SET_MP_STATE: {
2053 struct kvm_mp_state mp_state;
2056 if (copy_from_user(&mp_state, argp, sizeof mp_state))
2058 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2061 case KVM_TRANSLATE: {
2062 struct kvm_translation tr;
2065 if (copy_from_user(&tr, argp, sizeof tr))
2067 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2071 if (copy_to_user(argp, &tr, sizeof tr))
2076 case KVM_SET_GUEST_DEBUG: {
2077 struct kvm_guest_debug dbg;
2080 if (copy_from_user(&dbg, argp, sizeof dbg))
2082 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2085 case KVM_SET_SIGNAL_MASK: {
2086 struct kvm_signal_mask __user *sigmask_arg = argp;
2087 struct kvm_signal_mask kvm_sigmask;
2088 sigset_t sigset, *p;
2093 if (copy_from_user(&kvm_sigmask, argp,
2094 sizeof kvm_sigmask))
2097 if (kvm_sigmask.len != sizeof sigset)
2100 if (copy_from_user(&sigset, sigmask_arg->sigset,
2105 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2109 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2113 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2117 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2123 fpu = memdup_user(argp, sizeof(*fpu));
2129 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2133 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2142 #ifdef CONFIG_COMPAT
2143 static long kvm_vcpu_compat_ioctl(struct file *filp,
2144 unsigned int ioctl, unsigned long arg)
2146 struct kvm_vcpu *vcpu = filp->private_data;
2147 void __user *argp = compat_ptr(arg);
2150 if (vcpu->kvm->mm != current->mm)
2154 case KVM_SET_SIGNAL_MASK: {
2155 struct kvm_signal_mask __user *sigmask_arg = argp;
2156 struct kvm_signal_mask kvm_sigmask;
2157 compat_sigset_t csigset;
2162 if (copy_from_user(&kvm_sigmask, argp,
2163 sizeof kvm_sigmask))
2166 if (kvm_sigmask.len != sizeof csigset)
2169 if (copy_from_user(&csigset, sigmask_arg->sigset,
2172 sigset_from_compat(&sigset, &csigset);
2173 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2175 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2179 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2187 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2188 int (*accessor)(struct kvm_device *dev,
2189 struct kvm_device_attr *attr),
2192 struct kvm_device_attr attr;
2197 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2200 return accessor(dev, &attr);
2203 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2206 struct kvm_device *dev = filp->private_data;
2209 case KVM_SET_DEVICE_ATTR:
2210 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2211 case KVM_GET_DEVICE_ATTR:
2212 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2213 case KVM_HAS_DEVICE_ATTR:
2214 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2216 if (dev->ops->ioctl)
2217 return dev->ops->ioctl(dev, ioctl, arg);
2223 static int kvm_device_release(struct inode *inode, struct file *filp)
2225 struct kvm_device *dev = filp->private_data;
2226 struct kvm *kvm = dev->kvm;
2232 static const struct file_operations kvm_device_fops = {
2233 .unlocked_ioctl = kvm_device_ioctl,
2234 #ifdef CONFIG_COMPAT
2235 .compat_ioctl = kvm_device_ioctl,
2237 .release = kvm_device_release,
2240 struct kvm_device *kvm_device_from_filp(struct file *filp)
2242 if (filp->f_op != &kvm_device_fops)
2245 return filp->private_data;
2248 static int kvm_ioctl_create_device(struct kvm *kvm,
2249 struct kvm_create_device *cd)
2251 struct kvm_device_ops *ops = NULL;
2252 struct kvm_device *dev;
2253 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2257 #ifdef CONFIG_KVM_MPIC
2258 case KVM_DEV_TYPE_FSL_MPIC_20:
2259 case KVM_DEV_TYPE_FSL_MPIC_42:
2260 ops = &kvm_mpic_ops;
2263 #ifdef CONFIG_KVM_XICS
2264 case KVM_DEV_TYPE_XICS:
2265 ops = &kvm_xics_ops;
2275 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2282 ret = ops->create(dev, cd->type);
2288 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2294 list_add(&dev->vm_node, &kvm->devices);
2300 static long kvm_vm_ioctl(struct file *filp,
2301 unsigned int ioctl, unsigned long arg)
2303 struct kvm *kvm = filp->private_data;
2304 void __user *argp = (void __user *)arg;
2307 if (kvm->mm != current->mm)
2310 case KVM_CREATE_VCPU:
2311 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2313 case KVM_SET_USER_MEMORY_REGION: {
2314 struct kvm_userspace_memory_region kvm_userspace_mem;
2317 if (copy_from_user(&kvm_userspace_mem, argp,
2318 sizeof kvm_userspace_mem))
2321 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2324 case KVM_GET_DIRTY_LOG: {
2325 struct kvm_dirty_log log;
2328 if (copy_from_user(&log, argp, sizeof log))
2330 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2333 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2334 case KVM_REGISTER_COALESCED_MMIO: {
2335 struct kvm_coalesced_mmio_zone zone;
2337 if (copy_from_user(&zone, argp, sizeof zone))
2339 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2342 case KVM_UNREGISTER_COALESCED_MMIO: {
2343 struct kvm_coalesced_mmio_zone zone;
2345 if (copy_from_user(&zone, argp, sizeof zone))
2347 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2352 struct kvm_irqfd data;
2355 if (copy_from_user(&data, argp, sizeof data))
2357 r = kvm_irqfd(kvm, &data);
2360 case KVM_IOEVENTFD: {
2361 struct kvm_ioeventfd data;
2364 if (copy_from_user(&data, argp, sizeof data))
2366 r = kvm_ioeventfd(kvm, &data);
2369 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2370 case KVM_SET_BOOT_CPU_ID:
2372 mutex_lock(&kvm->lock);
2373 if (atomic_read(&kvm->online_vcpus) != 0)
2376 kvm->bsp_vcpu_id = arg;
2377 mutex_unlock(&kvm->lock);
2380 #ifdef CONFIG_HAVE_KVM_MSI
2381 case KVM_SIGNAL_MSI: {
2385 if (copy_from_user(&msi, argp, sizeof msi))
2387 r = kvm_send_userspace_msi(kvm, &msi);
2391 #ifdef __KVM_HAVE_IRQ_LINE
2392 case KVM_IRQ_LINE_STATUS:
2393 case KVM_IRQ_LINE: {
2394 struct kvm_irq_level irq_event;
2397 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2400 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
2401 ioctl == KVM_IRQ_LINE_STATUS);
2406 if (ioctl == KVM_IRQ_LINE_STATUS) {
2407 if (copy_to_user(argp, &irq_event, sizeof irq_event))
2415 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2416 case KVM_SET_GSI_ROUTING: {
2417 struct kvm_irq_routing routing;
2418 struct kvm_irq_routing __user *urouting;
2419 struct kvm_irq_routing_entry *entries;
2422 if (copy_from_user(&routing, argp, sizeof(routing)))
2425 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2430 entries = vmalloc(routing.nr * sizeof(*entries));
2435 if (copy_from_user(entries, urouting->entries,
2436 routing.nr * sizeof(*entries)))
2437 goto out_free_irq_routing;
2438 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2440 out_free_irq_routing:
2444 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
2445 case KVM_CREATE_DEVICE: {
2446 struct kvm_create_device cd;
2449 if (copy_from_user(&cd, argp, sizeof(cd)))
2452 r = kvm_ioctl_create_device(kvm, &cd);
2457 if (copy_to_user(argp, &cd, sizeof(cd)))
2464 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2466 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2472 #ifdef CONFIG_COMPAT
2473 struct compat_kvm_dirty_log {
2477 compat_uptr_t dirty_bitmap; /* one bit per page */
2482 static long kvm_vm_compat_ioctl(struct file *filp,
2483 unsigned int ioctl, unsigned long arg)
2485 struct kvm *kvm = filp->private_data;
2488 if (kvm->mm != current->mm)
2491 case KVM_GET_DIRTY_LOG: {
2492 struct compat_kvm_dirty_log compat_log;
2493 struct kvm_dirty_log log;
2496 if (copy_from_user(&compat_log, (void __user *)arg,
2497 sizeof(compat_log)))
2499 log.slot = compat_log.slot;
2500 log.padding1 = compat_log.padding1;
2501 log.padding2 = compat_log.padding2;
2502 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2504 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2508 r = kvm_vm_ioctl(filp, ioctl, arg);
2516 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2518 struct page *page[1];
2521 gfn_t gfn = vmf->pgoff;
2522 struct kvm *kvm = vma->vm_file->private_data;
2524 addr = gfn_to_hva(kvm, gfn);
2525 if (kvm_is_error_hva(addr))
2526 return VM_FAULT_SIGBUS;
2528 npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2530 if (unlikely(npages != 1))
2531 return VM_FAULT_SIGBUS;
2533 vmf->page = page[0];
2537 static const struct vm_operations_struct kvm_vm_vm_ops = {
2538 .fault = kvm_vm_fault,
2541 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2543 vma->vm_ops = &kvm_vm_vm_ops;
2547 static struct file_operations kvm_vm_fops = {
2548 .release = kvm_vm_release,
2549 .unlocked_ioctl = kvm_vm_ioctl,
2550 #ifdef CONFIG_COMPAT
2551 .compat_ioctl = kvm_vm_compat_ioctl,
2553 .mmap = kvm_vm_mmap,
2554 .llseek = noop_llseek,
2557 static int kvm_dev_ioctl_create_vm(unsigned long type)
2562 kvm = kvm_create_vm(type);
2564 return PTR_ERR(kvm);
2565 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2566 r = kvm_coalesced_mmio_init(kvm);
2572 r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC);
2579 static long kvm_dev_ioctl_check_extension_generic(long arg)
2582 case KVM_CAP_USER_MEMORY:
2583 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2584 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2585 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2586 case KVM_CAP_SET_BOOT_CPU_ID:
2588 case KVM_CAP_INTERNAL_ERROR_DATA:
2589 #ifdef CONFIG_HAVE_KVM_MSI
2590 case KVM_CAP_SIGNAL_MSI:
2592 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2593 case KVM_CAP_IRQFD_RESAMPLE:
2596 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2597 case KVM_CAP_IRQ_ROUTING:
2598 return KVM_MAX_IRQ_ROUTES;
2603 return kvm_dev_ioctl_check_extension(arg);
2606 static long kvm_dev_ioctl(struct file *filp,
2607 unsigned int ioctl, unsigned long arg)
2612 case KVM_GET_API_VERSION:
2616 r = KVM_API_VERSION;
2619 r = kvm_dev_ioctl_create_vm(arg);
2621 case KVM_CHECK_EXTENSION:
2622 r = kvm_dev_ioctl_check_extension_generic(arg);
2624 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(NULL);
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 *junk)
2682 raw_spin_lock(&kvm_lock);
2683 hardware_enable_nolock(junk);
2684 raw_spin_unlock(&kvm_lock);
2687 static void hardware_disable_nolock(void *junk)
2689 int cpu = raw_smp_processor_id();
2691 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2693 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2694 kvm_arch_hardware_disable(NULL);
2697 static void hardware_disable(void *junk)
2699 raw_spin_lock(&kvm_lock);
2700 hardware_disable_nolock(junk);
2701 raw_spin_unlock(&kvm_lock);
2704 static void hardware_disable_all_nolock(void)
2706 BUG_ON(!kvm_usage_count);
2709 if (!kvm_usage_count)
2710 on_each_cpu(hardware_disable_nolock, NULL, 1);
2713 static void hardware_disable_all(void)
2715 raw_spin_lock(&kvm_lock);
2716 hardware_disable_all_nolock();
2717 raw_spin_unlock(&kvm_lock);
2720 static int hardware_enable_all(void)
2724 raw_spin_lock(&kvm_lock);
2727 if (kvm_usage_count == 1) {
2728 atomic_set(&hardware_enable_failed, 0);
2729 on_each_cpu(hardware_enable_nolock, NULL, 1);
2731 if (atomic_read(&hardware_enable_failed)) {
2732 hardware_disable_all_nolock();
2737 raw_spin_unlock(&kvm_lock);
2742 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2747 if (!kvm_usage_count)
2750 val &= ~CPU_TASKS_FROZEN;
2753 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2755 hardware_disable(NULL);
2758 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2760 hardware_enable(NULL);
2766 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2770 * Some (well, at least mine) BIOSes hang on reboot if
2773 * And Intel TXT required VMX off for all cpu when system shutdown.
2775 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2776 kvm_rebooting = true;
2777 on_each_cpu(hardware_disable_nolock, NULL, 1);
2781 static struct notifier_block kvm_reboot_notifier = {
2782 .notifier_call = kvm_reboot,
2786 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2790 for (i = 0; i < bus->dev_count; i++) {
2791 struct kvm_io_device *pos = bus->range[i].dev;
2793 kvm_iodevice_destructor(pos);
2798 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
2799 const struct kvm_io_range *r2)
2801 if (r1->addr < r2->addr)
2803 if (r1->addr + r1->len > r2->addr + r2->len)
2808 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2810 return kvm_io_bus_cmp(p1, p2);
2813 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2814 gpa_t addr, int len)
2816 bus->range[bus->dev_count++] = (struct kvm_io_range) {
2822 sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2823 kvm_io_bus_sort_cmp, NULL);
2828 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2829 gpa_t addr, int len)
2831 struct kvm_io_range *range, key;
2834 key = (struct kvm_io_range) {
2839 range = bsearch(&key, bus->range, bus->dev_count,
2840 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2844 off = range - bus->range;
2846 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
2852 static int __kvm_io_bus_write(struct kvm_io_bus *bus,
2853 struct kvm_io_range *range, const void *val)
2857 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2861 while (idx < bus->dev_count &&
2862 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
2863 if (!kvm_iodevice_write(bus->range[idx].dev, range->addr,
2872 /* kvm_io_bus_write - called under kvm->slots_lock */
2873 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2874 int len, const void *val)
2876 struct kvm_io_bus *bus;
2877 struct kvm_io_range range;
2880 range = (struct kvm_io_range) {
2885 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2886 r = __kvm_io_bus_write(bus, &range, val);
2887 return r < 0 ? r : 0;
2890 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
2891 int kvm_io_bus_write_cookie(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2892 int len, const void *val, long cookie)
2894 struct kvm_io_bus *bus;
2895 struct kvm_io_range range;
2897 range = (struct kvm_io_range) {
2902 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2904 /* First try the device referenced by cookie. */
2905 if ((cookie >= 0) && (cookie < bus->dev_count) &&
2906 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
2907 if (!kvm_iodevice_write(bus->range[cookie].dev, addr, len,
2912 * cookie contained garbage; fall back to search and return the
2913 * correct cookie value.
2915 return __kvm_io_bus_write(bus, &range, val);
2918 static int __kvm_io_bus_read(struct kvm_io_bus *bus, struct kvm_io_range *range,
2923 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2927 while (idx < bus->dev_count &&
2928 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
2929 if (!kvm_iodevice_read(bus->range[idx].dev, range->addr,
2938 /* kvm_io_bus_read - called under kvm->slots_lock */
2939 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2942 struct kvm_io_bus *bus;
2943 struct kvm_io_range range;
2946 range = (struct kvm_io_range) {
2951 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2952 r = __kvm_io_bus_read(bus, &range, val);
2953 return r < 0 ? r : 0;
2956 /* kvm_io_bus_read_cookie - called under kvm->slots_lock */
2957 int kvm_io_bus_read_cookie(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2958 int len, void *val, long cookie)
2960 struct kvm_io_bus *bus;
2961 struct kvm_io_range range;
2963 range = (struct kvm_io_range) {
2968 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2970 /* First try the device referenced by cookie. */
2971 if ((cookie >= 0) && (cookie < bus->dev_count) &&
2972 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
2973 if (!kvm_iodevice_read(bus->range[cookie].dev, addr, len,
2978 * cookie contained garbage; fall back to search and return the
2979 * correct cookie value.
2981 return __kvm_io_bus_read(bus, &range, val);
2984 /* Caller must hold slots_lock. */
2985 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2986 int len, struct kvm_io_device *dev)
2988 struct kvm_io_bus *new_bus, *bus;
2990 bus = kvm->buses[bus_idx];
2991 /* exclude ioeventfd which is limited by maximum fd */
2992 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
2995 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
2996 sizeof(struct kvm_io_range)), GFP_KERNEL);
2999 memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
3000 sizeof(struct kvm_io_range)));
3001 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
3002 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3003 synchronize_srcu_expedited(&kvm->srcu);
3009 /* Caller must hold slots_lock. */
3010 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3011 struct kvm_io_device *dev)
3014 struct kvm_io_bus *new_bus, *bus;
3016 bus = kvm->buses[bus_idx];
3018 for (i = 0; i < bus->dev_count; i++)
3019 if (bus->range[i].dev == dev) {
3027 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3028 sizeof(struct kvm_io_range)), GFP_KERNEL);
3032 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3033 new_bus->dev_count--;
3034 memcpy(new_bus->range + i, bus->range + i + 1,
3035 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3037 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3038 synchronize_srcu_expedited(&kvm->srcu);
3043 static struct notifier_block kvm_cpu_notifier = {
3044 .notifier_call = kvm_cpu_hotplug,
3047 static int vm_stat_get(void *_offset, u64 *val)
3049 unsigned offset = (long)_offset;
3053 raw_spin_lock(&kvm_lock);
3054 list_for_each_entry(kvm, &vm_list, vm_list)
3055 *val += *(u32 *)((void *)kvm + offset);
3056 raw_spin_unlock(&kvm_lock);
3060 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
3062 static int vcpu_stat_get(void *_offset, u64 *val)
3064 unsigned offset = (long)_offset;
3066 struct kvm_vcpu *vcpu;
3070 raw_spin_lock(&kvm_lock);
3071 list_for_each_entry(kvm, &vm_list, vm_list)
3072 kvm_for_each_vcpu(i, vcpu, kvm)
3073 *val += *(u32 *)((void *)vcpu + offset);
3075 raw_spin_unlock(&kvm_lock);
3079 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
3081 static const struct file_operations *stat_fops[] = {
3082 [KVM_STAT_VCPU] = &vcpu_stat_fops,
3083 [KVM_STAT_VM] = &vm_stat_fops,
3086 static int kvm_init_debug(void)
3089 struct kvm_stats_debugfs_item *p;
3091 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3092 if (kvm_debugfs_dir == NULL)
3095 for (p = debugfs_entries; p->name; ++p) {
3096 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
3097 (void *)(long)p->offset,
3098 stat_fops[p->kind]);
3099 if (p->dentry == NULL)
3106 debugfs_remove_recursive(kvm_debugfs_dir);
3111 static void kvm_exit_debug(void)
3113 struct kvm_stats_debugfs_item *p;
3115 for (p = debugfs_entries; p->name; ++p)
3116 debugfs_remove(p->dentry);
3117 debugfs_remove(kvm_debugfs_dir);
3120 static int kvm_suspend(void)
3122 if (kvm_usage_count)
3123 hardware_disable_nolock(NULL);
3127 static void kvm_resume(void)
3129 if (kvm_usage_count) {
3130 WARN_ON(raw_spin_is_locked(&kvm_lock));
3131 hardware_enable_nolock(NULL);
3135 static struct syscore_ops kvm_syscore_ops = {
3136 .suspend = kvm_suspend,
3137 .resume = kvm_resume,
3141 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3143 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3146 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3148 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3149 if (vcpu->preempted)
3150 vcpu->preempted = false;
3152 kvm_arch_vcpu_load(vcpu, cpu);
3155 static void kvm_sched_out(struct preempt_notifier *pn,
3156 struct task_struct *next)
3158 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3160 if (current->state == TASK_RUNNING)
3161 vcpu->preempted = true;
3162 kvm_arch_vcpu_put(vcpu);
3165 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3166 struct module *module)
3171 r = kvm_arch_init(opaque);
3176 * kvm_arch_init makes sure there's at most one caller
3177 * for architectures that support multiple implementations,
3178 * like intel and amd on x86.
3179 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3180 * conflicts in case kvm is already setup for another implementation.
3182 r = kvm_irqfd_init();
3186 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
3191 r = kvm_arch_hardware_setup();
3195 for_each_online_cpu(cpu) {
3196 smp_call_function_single(cpu,
3197 kvm_arch_check_processor_compat,
3203 r = register_cpu_notifier(&kvm_cpu_notifier);
3206 register_reboot_notifier(&kvm_reboot_notifier);
3208 /* A kmem cache lets us meet the alignment requirements of fx_save. */
3210 vcpu_align = __alignof__(struct kvm_vcpu);
3211 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
3213 if (!kvm_vcpu_cache) {
3218 r = kvm_async_pf_init();
3222 kvm_chardev_ops.owner = module;
3223 kvm_vm_fops.owner = module;
3224 kvm_vcpu_fops.owner = module;
3226 r = misc_register(&kvm_dev);
3228 printk(KERN_ERR "kvm: misc device register failed\n");
3232 register_syscore_ops(&kvm_syscore_ops);
3234 kvm_preempt_ops.sched_in = kvm_sched_in;
3235 kvm_preempt_ops.sched_out = kvm_sched_out;
3237 r = kvm_init_debug();
3239 printk(KERN_ERR "kvm: create debugfs files failed\n");
3246 unregister_syscore_ops(&kvm_syscore_ops);
3247 misc_deregister(&kvm_dev);
3249 kvm_async_pf_deinit();
3251 kmem_cache_destroy(kvm_vcpu_cache);
3253 unregister_reboot_notifier(&kvm_reboot_notifier);
3254 unregister_cpu_notifier(&kvm_cpu_notifier);
3257 kvm_arch_hardware_unsetup();
3259 free_cpumask_var(cpus_hardware_enabled);
3267 EXPORT_SYMBOL_GPL(kvm_init);
3272 misc_deregister(&kvm_dev);
3273 kmem_cache_destroy(kvm_vcpu_cache);
3274 kvm_async_pf_deinit();
3275 unregister_syscore_ops(&kvm_syscore_ops);
3276 unregister_reboot_notifier(&kvm_reboot_notifier);
3277 unregister_cpu_notifier(&kvm_cpu_notifier);
3278 on_each_cpu(hardware_disable_nolock, NULL, 1);
3279 kvm_arch_hardware_unsetup();
3282 free_cpumask_var(cpus_hardware_enabled);
3284 EXPORT_SYMBOL_GPL(kvm_exit);