2 * Kernel-based Virtual Machine driver for Linux
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
7 * Copyright (C) 2006 Qumranet, Inc.
8 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
11 * Avi Kivity <avi@qumranet.com>
12 * Yaniv Kamay <yaniv@qumranet.com>
14 * This work is licensed under the terms of the GNU GPL, version 2. See
15 * the COPYING file in the top-level directory.
21 #include <linux/kvm_host.h>
22 #include <linux/kvm.h>
23 #include <linux/module.h>
24 #include <linux/errno.h>
25 #include <linux/percpu.h>
27 #include <linux/miscdevice.h>
28 #include <linux/vmalloc.h>
29 #include <linux/reboot.h>
30 #include <linux/debugfs.h>
31 #include <linux/highmem.h>
32 #include <linux/file.h>
33 #include <linux/syscore_ops.h>
34 #include <linux/cpu.h>
35 #include <linux/sched.h>
36 #include <linux/cpumask.h>
37 #include <linux/smp.h>
38 #include <linux/anon_inodes.h>
39 #include <linux/profile.h>
40 #include <linux/kvm_para.h>
41 #include <linux/pagemap.h>
42 #include <linux/mman.h>
43 #include <linux/swap.h>
44 #include <linux/bitops.h>
45 #include <linux/spinlock.h>
46 #include <linux/compat.h>
47 #include <linux/srcu.h>
48 #include <linux/hugetlb.h>
49 #include <linux/slab.h>
50 #include <linux/sort.h>
51 #include <linux/bsearch.h>
53 #include <asm/processor.h>
55 #include <asm/uaccess.h>
56 #include <asm/pgtable.h>
58 #include "coalesced_mmio.h"
61 #define CREATE_TRACE_POINTS
62 #include <trace/events/kvm.h>
64 MODULE_AUTHOR("Qumranet");
65 MODULE_LICENSE("GPL");
70 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
73 DEFINE_SPINLOCK(kvm_lock);
74 static DEFINE_RAW_SPINLOCK(kvm_count_lock);
77 static cpumask_var_t cpus_hardware_enabled;
78 static int kvm_usage_count = 0;
79 static atomic_t hardware_enable_failed;
81 struct kmem_cache *kvm_vcpu_cache;
82 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
84 static __read_mostly struct preempt_ops kvm_preempt_ops;
86 struct dentry *kvm_debugfs_dir;
88 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
91 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
94 static int hardware_enable_all(void);
95 static void hardware_disable_all(void);
97 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
98 static void update_memslots(struct kvm_memslots *slots,
99 struct kvm_memory_slot *new, u64 last_generation);
101 static void kvm_release_pfn_dirty(pfn_t pfn);
102 static void mark_page_dirty_in_slot(struct kvm *kvm,
103 struct kvm_memory_slot *memslot, gfn_t gfn);
105 __visible bool kvm_rebooting;
106 EXPORT_SYMBOL_GPL(kvm_rebooting);
108 static bool largepages_enabled = true;
110 bool kvm_is_mmio_pfn(pfn_t pfn)
113 return PageReserved(pfn_to_page(pfn));
119 * Switches to specified vcpu, until a matching vcpu_put()
121 int vcpu_load(struct kvm_vcpu *vcpu)
125 if (mutex_lock_killable(&vcpu->mutex))
127 if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
128 /* The thread running this VCPU changed. */
129 struct pid *oldpid = vcpu->pid;
130 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
131 rcu_assign_pointer(vcpu->pid, newpid);
137 preempt_notifier_register(&vcpu->preempt_notifier);
138 kvm_arch_vcpu_load(vcpu, cpu);
143 void vcpu_put(struct kvm_vcpu *vcpu)
146 kvm_arch_vcpu_put(vcpu);
147 preempt_notifier_unregister(&vcpu->preempt_notifier);
149 mutex_unlock(&vcpu->mutex);
152 static void ack_flush(void *_completed)
156 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
161 struct kvm_vcpu *vcpu;
163 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
166 kvm_for_each_vcpu(i, vcpu, kvm) {
167 kvm_make_request(req, vcpu);
170 /* Set ->requests bit before we read ->mode */
173 if (cpus != NULL && cpu != -1 && cpu != me &&
174 kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
175 cpumask_set_cpu(cpu, cpus);
177 if (unlikely(cpus == NULL))
178 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
179 else if (!cpumask_empty(cpus))
180 smp_call_function_many(cpus, ack_flush, NULL, 1);
184 free_cpumask_var(cpus);
188 void kvm_flush_remote_tlbs(struct kvm *kvm)
190 long dirty_count = kvm->tlbs_dirty;
193 if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
194 ++kvm->stat.remote_tlb_flush;
195 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
197 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
199 void kvm_reload_remote_mmus(struct kvm *kvm)
201 make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
204 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
206 make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
209 void kvm_make_scan_ioapic_request(struct kvm *kvm)
211 make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
214 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
219 mutex_init(&vcpu->mutex);
224 init_waitqueue_head(&vcpu->wq);
225 kvm_async_pf_vcpu_init(vcpu);
227 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
232 vcpu->run = page_address(page);
234 kvm_vcpu_set_in_spin_loop(vcpu, false);
235 kvm_vcpu_set_dy_eligible(vcpu, false);
236 vcpu->preempted = false;
238 r = kvm_arch_vcpu_init(vcpu);
244 free_page((unsigned long)vcpu->run);
248 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
250 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
253 kvm_arch_vcpu_uninit(vcpu);
254 free_page((unsigned long)vcpu->run);
256 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
258 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
259 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
261 return container_of(mn, struct kvm, mmu_notifier);
264 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
265 struct mm_struct *mm,
266 unsigned long address)
268 struct kvm *kvm = mmu_notifier_to_kvm(mn);
269 int need_tlb_flush, idx;
272 * When ->invalidate_page runs, the linux pte has been zapped
273 * already but the page is still allocated until
274 * ->invalidate_page returns. So if we increase the sequence
275 * here the kvm page fault will notice if the spte can't be
276 * established because the page is going to be freed. If
277 * instead the kvm page fault establishes the spte before
278 * ->invalidate_page runs, kvm_unmap_hva will release it
281 * The sequence increase only need to be seen at spin_unlock
282 * time, and not at spin_lock time.
284 * Increasing the sequence after the spin_unlock would be
285 * unsafe because the kvm page fault could then establish the
286 * pte after kvm_unmap_hva returned, without noticing the page
287 * is going to be freed.
289 idx = srcu_read_lock(&kvm->srcu);
290 spin_lock(&kvm->mmu_lock);
292 kvm->mmu_notifier_seq++;
293 need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
294 /* we've to flush the tlb before the pages can be freed */
296 kvm_flush_remote_tlbs(kvm);
298 spin_unlock(&kvm->mmu_lock);
299 srcu_read_unlock(&kvm->srcu, idx);
302 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
303 struct mm_struct *mm,
304 unsigned long address,
307 struct kvm *kvm = mmu_notifier_to_kvm(mn);
310 idx = srcu_read_lock(&kvm->srcu);
311 spin_lock(&kvm->mmu_lock);
312 kvm->mmu_notifier_seq++;
313 kvm_set_spte_hva(kvm, address, pte);
314 spin_unlock(&kvm->mmu_lock);
315 srcu_read_unlock(&kvm->srcu, idx);
318 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
319 struct mm_struct *mm,
323 struct kvm *kvm = mmu_notifier_to_kvm(mn);
324 int need_tlb_flush = 0, idx;
326 idx = srcu_read_lock(&kvm->srcu);
327 spin_lock(&kvm->mmu_lock);
329 * The count increase must become visible at unlock time as no
330 * spte can be established without taking the mmu_lock and
331 * count is also read inside the mmu_lock critical section.
333 kvm->mmu_notifier_count++;
334 need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
335 need_tlb_flush |= kvm->tlbs_dirty;
336 /* we've to flush the tlb before the pages can be freed */
338 kvm_flush_remote_tlbs(kvm);
340 spin_unlock(&kvm->mmu_lock);
341 srcu_read_unlock(&kvm->srcu, idx);
344 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
345 struct mm_struct *mm,
349 struct kvm *kvm = mmu_notifier_to_kvm(mn);
351 spin_lock(&kvm->mmu_lock);
353 * This sequence increase will notify the kvm page fault that
354 * the page that is going to be mapped in the spte could have
357 kvm->mmu_notifier_seq++;
360 * The above sequence increase must be visible before the
361 * below count decrease, which is ensured by the smp_wmb above
362 * in conjunction with the smp_rmb in mmu_notifier_retry().
364 kvm->mmu_notifier_count--;
365 spin_unlock(&kvm->mmu_lock);
367 BUG_ON(kvm->mmu_notifier_count < 0);
370 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
371 struct mm_struct *mm,
372 unsigned long address)
374 struct kvm *kvm = mmu_notifier_to_kvm(mn);
377 idx = srcu_read_lock(&kvm->srcu);
378 spin_lock(&kvm->mmu_lock);
380 young = kvm_age_hva(kvm, address);
382 kvm_flush_remote_tlbs(kvm);
384 spin_unlock(&kvm->mmu_lock);
385 srcu_read_unlock(&kvm->srcu, idx);
390 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
391 struct mm_struct *mm,
392 unsigned long address)
394 struct kvm *kvm = mmu_notifier_to_kvm(mn);
397 idx = srcu_read_lock(&kvm->srcu);
398 spin_lock(&kvm->mmu_lock);
399 young = kvm_test_age_hva(kvm, address);
400 spin_unlock(&kvm->mmu_lock);
401 srcu_read_unlock(&kvm->srcu, idx);
406 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
407 struct mm_struct *mm)
409 struct kvm *kvm = mmu_notifier_to_kvm(mn);
412 idx = srcu_read_lock(&kvm->srcu);
413 kvm_arch_flush_shadow_all(kvm);
414 srcu_read_unlock(&kvm->srcu, idx);
417 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
418 .invalidate_page = kvm_mmu_notifier_invalidate_page,
419 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
420 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
421 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
422 .test_young = kvm_mmu_notifier_test_young,
423 .change_pte = kvm_mmu_notifier_change_pte,
424 .release = kvm_mmu_notifier_release,
427 static int kvm_init_mmu_notifier(struct kvm *kvm)
429 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
430 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
433 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
435 static int kvm_init_mmu_notifier(struct kvm *kvm)
440 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
442 static void kvm_init_memslots_id(struct kvm *kvm)
445 struct kvm_memslots *slots = kvm->memslots;
447 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
448 slots->id_to_index[i] = slots->memslots[i].id = i;
451 static struct kvm *kvm_create_vm(unsigned long type)
454 struct kvm *kvm = kvm_arch_alloc_vm();
457 return ERR_PTR(-ENOMEM);
459 r = kvm_arch_init_vm(kvm, type);
461 goto out_err_no_disable;
463 r = hardware_enable_all();
465 goto out_err_no_disable;
467 #ifdef CONFIG_HAVE_KVM_IRQCHIP
468 INIT_HLIST_HEAD(&kvm->mask_notifier_list);
470 #ifdef CONFIG_HAVE_KVM_IRQFD
471 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
474 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
477 kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
479 goto out_err_no_srcu;
482 * Init kvm generation close to the maximum to easily test the
483 * code of handling generation number wrap-around.
485 kvm->memslots->generation = -150;
487 kvm_init_memslots_id(kvm);
488 if (init_srcu_struct(&kvm->srcu))
489 goto out_err_no_srcu;
490 if (init_srcu_struct(&kvm->irq_srcu))
491 goto out_err_no_irq_srcu;
492 for (i = 0; i < KVM_NR_BUSES; i++) {
493 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
499 spin_lock_init(&kvm->mmu_lock);
500 kvm->mm = current->mm;
501 atomic_inc(&kvm->mm->mm_count);
502 kvm_eventfd_init(kvm);
503 mutex_init(&kvm->lock);
504 mutex_init(&kvm->irq_lock);
505 mutex_init(&kvm->slots_lock);
506 atomic_set(&kvm->users_count, 1);
507 INIT_LIST_HEAD(&kvm->devices);
509 r = kvm_init_mmu_notifier(kvm);
513 spin_lock(&kvm_lock);
514 list_add(&kvm->vm_list, &vm_list);
515 spin_unlock(&kvm_lock);
520 cleanup_srcu_struct(&kvm->irq_srcu);
522 cleanup_srcu_struct(&kvm->srcu);
524 hardware_disable_all();
526 for (i = 0; i < KVM_NR_BUSES; i++)
527 kfree(kvm->buses[i]);
528 kfree(kvm->memslots);
529 kvm_arch_free_vm(kvm);
534 * Avoid using vmalloc for a small buffer.
535 * Should not be used when the size is statically known.
537 void *kvm_kvzalloc(unsigned long size)
539 if (size > PAGE_SIZE)
540 return vzalloc(size);
542 return kzalloc(size, GFP_KERNEL);
545 void kvm_kvfree(const void *addr)
547 if (is_vmalloc_addr(addr))
553 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
555 if (!memslot->dirty_bitmap)
558 kvm_kvfree(memslot->dirty_bitmap);
559 memslot->dirty_bitmap = NULL;
563 * Free any memory in @free but not in @dont.
565 static void kvm_free_physmem_slot(struct kvm *kvm, struct kvm_memory_slot *free,
566 struct kvm_memory_slot *dont)
568 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
569 kvm_destroy_dirty_bitmap(free);
571 kvm_arch_free_memslot(kvm, free, dont);
576 static void kvm_free_physmem(struct kvm *kvm)
578 struct kvm_memslots *slots = kvm->memslots;
579 struct kvm_memory_slot *memslot;
581 kvm_for_each_memslot(memslot, slots)
582 kvm_free_physmem_slot(kvm, memslot, NULL);
584 kfree(kvm->memslots);
587 static void kvm_destroy_devices(struct kvm *kvm)
589 struct list_head *node, *tmp;
591 list_for_each_safe(node, tmp, &kvm->devices) {
592 struct kvm_device *dev =
593 list_entry(node, struct kvm_device, vm_node);
596 dev->ops->destroy(dev);
600 static void kvm_destroy_vm(struct kvm *kvm)
603 struct mm_struct *mm = kvm->mm;
605 kvm_arch_sync_events(kvm);
606 spin_lock(&kvm_lock);
607 list_del(&kvm->vm_list);
608 spin_unlock(&kvm_lock);
609 kvm_free_irq_routing(kvm);
610 for (i = 0; i < KVM_NR_BUSES; i++)
611 kvm_io_bus_destroy(kvm->buses[i]);
612 kvm_coalesced_mmio_free(kvm);
613 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
614 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
616 kvm_arch_flush_shadow_all(kvm);
618 kvm_arch_destroy_vm(kvm);
619 kvm_destroy_devices(kvm);
620 kvm_free_physmem(kvm);
621 cleanup_srcu_struct(&kvm->irq_srcu);
622 cleanup_srcu_struct(&kvm->srcu);
623 kvm_arch_free_vm(kvm);
624 hardware_disable_all();
628 void kvm_get_kvm(struct kvm *kvm)
630 atomic_inc(&kvm->users_count);
632 EXPORT_SYMBOL_GPL(kvm_get_kvm);
634 void kvm_put_kvm(struct kvm *kvm)
636 if (atomic_dec_and_test(&kvm->users_count))
639 EXPORT_SYMBOL_GPL(kvm_put_kvm);
642 static int kvm_vm_release(struct inode *inode, struct file *filp)
644 struct kvm *kvm = filp->private_data;
646 kvm_irqfd_release(kvm);
653 * Allocation size is twice as large as the actual dirty bitmap size.
654 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
656 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
658 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
660 memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
661 if (!memslot->dirty_bitmap)
667 static int cmp_memslot(const void *slot1, const void *slot2)
669 struct kvm_memory_slot *s1, *s2;
671 s1 = (struct kvm_memory_slot *)slot1;
672 s2 = (struct kvm_memory_slot *)slot2;
674 if (s1->npages < s2->npages)
676 if (s1->npages > s2->npages)
683 * Sort the memslots base on its size, so the larger slots
684 * will get better fit.
686 static void sort_memslots(struct kvm_memslots *slots)
690 sort(slots->memslots, KVM_MEM_SLOTS_NUM,
691 sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
693 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
694 slots->id_to_index[slots->memslots[i].id] = i;
697 static void update_memslots(struct kvm_memslots *slots,
698 struct kvm_memory_slot *new,
703 struct kvm_memory_slot *old = id_to_memslot(slots, id);
704 unsigned long npages = old->npages;
707 if (new->npages != npages)
708 sort_memslots(slots);
711 slots->generation = last_generation + 1;
714 static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
716 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
718 #ifdef __KVM_HAVE_READONLY_MEM
719 valid_flags |= KVM_MEM_READONLY;
722 if (mem->flags & ~valid_flags)
728 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
729 struct kvm_memslots *slots, struct kvm_memory_slot *new)
731 struct kvm_memslots *old_memslots = kvm->memslots;
733 update_memslots(slots, new, kvm->memslots->generation);
734 rcu_assign_pointer(kvm->memslots, slots);
735 synchronize_srcu_expedited(&kvm->srcu);
737 kvm_arch_memslots_updated(kvm);
743 * Allocate some memory and give it an address in the guest physical address
746 * Discontiguous memory is allowed, mostly for framebuffers.
748 * Must be called holding mmap_sem for write.
750 int __kvm_set_memory_region(struct kvm *kvm,
751 struct kvm_userspace_memory_region *mem)
755 unsigned long npages;
756 struct kvm_memory_slot *slot;
757 struct kvm_memory_slot old, new;
758 struct kvm_memslots *slots = NULL, *old_memslots;
759 enum kvm_mr_change change;
761 r = check_memory_region_flags(mem);
766 /* General sanity checks */
767 if (mem->memory_size & (PAGE_SIZE - 1))
769 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
771 /* We can read the guest memory with __xxx_user() later on. */
772 if ((mem->slot < KVM_USER_MEM_SLOTS) &&
773 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
774 !access_ok(VERIFY_WRITE,
775 (void __user *)(unsigned long)mem->userspace_addr,
778 if (mem->slot >= KVM_MEM_SLOTS_NUM)
780 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
783 slot = id_to_memslot(kvm->memslots, mem->slot);
784 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
785 npages = mem->memory_size >> PAGE_SHIFT;
788 if (npages > KVM_MEM_MAX_NR_PAGES)
792 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
797 new.base_gfn = base_gfn;
799 new.flags = mem->flags;
804 change = KVM_MR_CREATE;
805 else { /* Modify an existing slot. */
806 if ((mem->userspace_addr != old.userspace_addr) ||
807 (npages != old.npages) ||
808 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
811 if (base_gfn != old.base_gfn)
812 change = KVM_MR_MOVE;
813 else if (new.flags != old.flags)
814 change = KVM_MR_FLAGS_ONLY;
815 else { /* Nothing to change. */
820 } else if (old.npages) {
821 change = KVM_MR_DELETE;
822 } else /* Modify a non-existent slot: disallowed. */
825 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
826 /* Check for overlaps */
828 kvm_for_each_memslot(slot, kvm->memslots) {
829 if ((slot->id >= KVM_USER_MEM_SLOTS) ||
830 (slot->id == mem->slot))
832 if (!((base_gfn + npages <= slot->base_gfn) ||
833 (base_gfn >= slot->base_gfn + slot->npages)))
838 /* Free page dirty bitmap if unneeded */
839 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
840 new.dirty_bitmap = NULL;
843 if (change == KVM_MR_CREATE) {
844 new.userspace_addr = mem->userspace_addr;
846 if (kvm_arch_create_memslot(kvm, &new, npages))
850 /* Allocate page dirty bitmap if needed */
851 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
852 if (kvm_create_dirty_bitmap(&new) < 0)
856 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
858 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
862 slot = id_to_memslot(slots, mem->slot);
863 slot->flags |= KVM_MEMSLOT_INVALID;
865 old_memslots = install_new_memslots(kvm, slots, NULL);
867 /* slot was deleted or moved, clear iommu mapping */
868 kvm_iommu_unmap_pages(kvm, &old);
869 /* From this point no new shadow pages pointing to a deleted,
870 * or moved, memslot will be created.
872 * validation of sp->gfn happens in:
873 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
874 * - kvm_is_visible_gfn (mmu_check_roots)
876 kvm_arch_flush_shadow_memslot(kvm, slot);
877 slots = old_memslots;
880 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
886 * We can re-use the old_memslots from above, the only difference
887 * from the currently installed memslots is the invalid flag. This
888 * will get overwritten by update_memslots anyway.
891 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
897 /* actual memory is freed via old in kvm_free_physmem_slot below */
898 if (change == KVM_MR_DELETE) {
899 new.dirty_bitmap = NULL;
900 memset(&new.arch, 0, sizeof(new.arch));
903 old_memslots = install_new_memslots(kvm, slots, &new);
905 kvm_arch_commit_memory_region(kvm, mem, &old, change);
907 kvm_free_physmem_slot(kvm, &old, &new);
911 * IOMMU mapping: New slots need to be mapped. Old slots need to be
912 * un-mapped and re-mapped if their base changes. Since base change
913 * unmapping is handled above with slot deletion, mapping alone is
914 * needed here. Anything else the iommu might care about for existing
915 * slots (size changes, userspace addr changes and read-only flag
916 * changes) is disallowed above, so any other attribute changes getting
917 * here can be skipped.
919 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
920 r = kvm_iommu_map_pages(kvm, &new);
929 kvm_free_physmem_slot(kvm, &new, &old);
933 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
935 int kvm_set_memory_region(struct kvm *kvm,
936 struct kvm_userspace_memory_region *mem)
940 mutex_lock(&kvm->slots_lock);
941 r = __kvm_set_memory_region(kvm, mem);
942 mutex_unlock(&kvm->slots_lock);
945 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
947 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
948 struct kvm_userspace_memory_region *mem)
950 if (mem->slot >= KVM_USER_MEM_SLOTS)
952 return kvm_set_memory_region(kvm, mem);
955 int kvm_get_dirty_log(struct kvm *kvm,
956 struct kvm_dirty_log *log, int *is_dirty)
958 struct kvm_memory_slot *memslot;
961 unsigned long any = 0;
964 if (log->slot >= KVM_USER_MEM_SLOTS)
967 memslot = id_to_memslot(kvm->memslots, log->slot);
969 if (!memslot->dirty_bitmap)
972 n = kvm_dirty_bitmap_bytes(memslot);
974 for (i = 0; !any && i < n/sizeof(long); ++i)
975 any = memslot->dirty_bitmap[i];
978 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
988 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
990 bool kvm_largepages_enabled(void)
992 return largepages_enabled;
995 void kvm_disable_largepages(void)
997 largepages_enabled = false;
999 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1001 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1003 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1005 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1007 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1009 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1011 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1012 memslot->flags & KVM_MEMSLOT_INVALID)
1017 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1019 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1021 struct vm_area_struct *vma;
1022 unsigned long addr, size;
1026 addr = gfn_to_hva(kvm, gfn);
1027 if (kvm_is_error_hva(addr))
1030 down_read(¤t->mm->mmap_sem);
1031 vma = find_vma(current->mm, addr);
1035 size = vma_kernel_pagesize(vma);
1038 up_read(¤t->mm->mmap_sem);
1043 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1045 return slot->flags & KVM_MEM_READONLY;
1048 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1049 gfn_t *nr_pages, bool write)
1051 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1052 return KVM_HVA_ERR_BAD;
1054 if (memslot_is_readonly(slot) && write)
1055 return KVM_HVA_ERR_RO_BAD;
1058 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1060 return __gfn_to_hva_memslot(slot, gfn);
1063 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1066 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1069 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1072 return gfn_to_hva_many(slot, gfn, NULL);
1074 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1076 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1078 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1080 EXPORT_SYMBOL_GPL(gfn_to_hva);
1083 * If writable is set to false, the hva returned by this function is only
1084 * allowed to be read.
1086 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1088 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1089 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1091 if (!kvm_is_error_hva(hva) && writable)
1092 *writable = !memslot_is_readonly(slot);
1097 static int kvm_read_hva(void *data, void __user *hva, int len)
1099 return __copy_from_user(data, hva, len);
1102 static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1104 return __copy_from_user_inatomic(data, hva, len);
1107 static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1108 unsigned long start, int write, struct page **page)
1110 int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1113 flags |= FOLL_WRITE;
1115 return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1118 static inline int check_user_page_hwpoison(unsigned long addr)
1120 int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1122 rc = __get_user_pages(current, current->mm, addr, 1,
1123 flags, NULL, NULL, NULL);
1124 return rc == -EHWPOISON;
1128 * The atomic path to get the writable pfn which will be stored in @pfn,
1129 * true indicates success, otherwise false is returned.
1131 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1132 bool write_fault, bool *writable, pfn_t *pfn)
1134 struct page *page[1];
1137 if (!(async || atomic))
1141 * Fast pin a writable pfn only if it is a write fault request
1142 * or the caller allows to map a writable pfn for a read fault
1145 if (!(write_fault || writable))
1148 npages = __get_user_pages_fast(addr, 1, 1, page);
1150 *pfn = page_to_pfn(page[0]);
1161 * The slow path to get the pfn of the specified host virtual address,
1162 * 1 indicates success, -errno is returned if error is detected.
1164 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1165 bool *writable, pfn_t *pfn)
1167 struct page *page[1];
1173 *writable = write_fault;
1176 down_read(¤t->mm->mmap_sem);
1177 npages = get_user_page_nowait(current, current->mm,
1178 addr, write_fault, page);
1179 up_read(¤t->mm->mmap_sem);
1181 npages = get_user_pages_fast(addr, 1, write_fault,
1186 /* map read fault as writable if possible */
1187 if (unlikely(!write_fault) && writable) {
1188 struct page *wpage[1];
1190 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1199 *pfn = page_to_pfn(page[0]);
1203 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1205 if (unlikely(!(vma->vm_flags & VM_READ)))
1208 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1215 * Pin guest page in memory and return its pfn.
1216 * @addr: host virtual address which maps memory to the guest
1217 * @atomic: whether this function can sleep
1218 * @async: whether this function need to wait IO complete if the
1219 * host page is not in the memory
1220 * @write_fault: whether we should get a writable host page
1221 * @writable: whether it allows to map a writable host page for !@write_fault
1223 * The function will map a writable host page for these two cases:
1224 * 1): @write_fault = true
1225 * 2): @write_fault = false && @writable, @writable will tell the caller
1226 * whether the mapping is writable.
1228 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1229 bool write_fault, bool *writable)
1231 struct vm_area_struct *vma;
1235 /* we can do it either atomically or asynchronously, not both */
1236 BUG_ON(atomic && async);
1238 if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1242 return KVM_PFN_ERR_FAULT;
1244 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1248 down_read(¤t->mm->mmap_sem);
1249 if (npages == -EHWPOISON ||
1250 (!async && check_user_page_hwpoison(addr))) {
1251 pfn = KVM_PFN_ERR_HWPOISON;
1255 vma = find_vma_intersection(current->mm, addr, addr + 1);
1258 pfn = KVM_PFN_ERR_FAULT;
1259 else if ((vma->vm_flags & VM_PFNMAP)) {
1260 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1262 BUG_ON(!kvm_is_mmio_pfn(pfn));
1264 if (async && vma_is_valid(vma, write_fault))
1266 pfn = KVM_PFN_ERR_FAULT;
1269 up_read(¤t->mm->mmap_sem);
1274 __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1275 bool *async, bool write_fault, bool *writable)
1277 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1279 if (addr == KVM_HVA_ERR_RO_BAD)
1280 return KVM_PFN_ERR_RO_FAULT;
1282 if (kvm_is_error_hva(addr))
1283 return KVM_PFN_NOSLOT;
1285 /* Do not map writable pfn in the readonly memslot. */
1286 if (writable && memslot_is_readonly(slot)) {
1291 return hva_to_pfn(addr, atomic, async, write_fault,
1295 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1296 bool write_fault, bool *writable)
1298 struct kvm_memory_slot *slot;
1303 slot = gfn_to_memslot(kvm, gfn);
1305 return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1309 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1311 return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1313 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1315 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1316 bool write_fault, bool *writable)
1318 return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1320 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1322 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1324 return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1326 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1328 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1331 return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1333 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1335 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1337 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1340 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1342 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1344 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1346 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1352 addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1353 if (kvm_is_error_hva(addr))
1356 if (entry < nr_pages)
1359 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1361 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1363 static struct page *kvm_pfn_to_page(pfn_t pfn)
1365 if (is_error_noslot_pfn(pfn))
1366 return KVM_ERR_PTR_BAD_PAGE;
1368 if (kvm_is_mmio_pfn(pfn)) {
1370 return KVM_ERR_PTR_BAD_PAGE;
1373 return pfn_to_page(pfn);
1376 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1380 pfn = gfn_to_pfn(kvm, gfn);
1382 return kvm_pfn_to_page(pfn);
1385 EXPORT_SYMBOL_GPL(gfn_to_page);
1387 void kvm_release_page_clean(struct page *page)
1389 WARN_ON(is_error_page(page));
1391 kvm_release_pfn_clean(page_to_pfn(page));
1393 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1395 void kvm_release_pfn_clean(pfn_t pfn)
1397 if (!is_error_noslot_pfn(pfn) && !kvm_is_mmio_pfn(pfn))
1398 put_page(pfn_to_page(pfn));
1400 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1402 void kvm_release_page_dirty(struct page *page)
1404 WARN_ON(is_error_page(page));
1406 kvm_release_pfn_dirty(page_to_pfn(page));
1408 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1410 static void kvm_release_pfn_dirty(pfn_t pfn)
1412 kvm_set_pfn_dirty(pfn);
1413 kvm_release_pfn_clean(pfn);
1416 void kvm_set_pfn_dirty(pfn_t pfn)
1418 if (!kvm_is_mmio_pfn(pfn)) {
1419 struct page *page = pfn_to_page(pfn);
1420 if (!PageReserved(page))
1424 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1426 void kvm_set_pfn_accessed(pfn_t pfn)
1428 if (!kvm_is_mmio_pfn(pfn))
1429 mark_page_accessed(pfn_to_page(pfn));
1431 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1433 void kvm_get_pfn(pfn_t pfn)
1435 if (!kvm_is_mmio_pfn(pfn))
1436 get_page(pfn_to_page(pfn));
1438 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1440 static int next_segment(unsigned long len, int offset)
1442 if (len > PAGE_SIZE - offset)
1443 return PAGE_SIZE - offset;
1448 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1454 addr = gfn_to_hva_prot(kvm, gfn, NULL);
1455 if (kvm_is_error_hva(addr))
1457 r = kvm_read_hva(data, (void __user *)addr + offset, len);
1462 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1464 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1466 gfn_t gfn = gpa >> PAGE_SHIFT;
1468 int offset = offset_in_page(gpa);
1471 while ((seg = next_segment(len, offset)) != 0) {
1472 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1482 EXPORT_SYMBOL_GPL(kvm_read_guest);
1484 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1489 gfn_t gfn = gpa >> PAGE_SHIFT;
1490 int offset = offset_in_page(gpa);
1492 addr = gfn_to_hva_prot(kvm, gfn, NULL);
1493 if (kvm_is_error_hva(addr))
1495 pagefault_disable();
1496 r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
1502 EXPORT_SYMBOL(kvm_read_guest_atomic);
1504 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1505 int offset, int len)
1510 addr = gfn_to_hva(kvm, gfn);
1511 if (kvm_is_error_hva(addr))
1513 r = __copy_to_user((void __user *)addr + offset, data, len);
1516 mark_page_dirty(kvm, gfn);
1519 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1521 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1524 gfn_t gfn = gpa >> PAGE_SHIFT;
1526 int offset = offset_in_page(gpa);
1529 while ((seg = next_segment(len, offset)) != 0) {
1530 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1541 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1542 gpa_t gpa, unsigned long len)
1544 struct kvm_memslots *slots = kvm_memslots(kvm);
1545 int offset = offset_in_page(gpa);
1546 gfn_t start_gfn = gpa >> PAGE_SHIFT;
1547 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1548 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1549 gfn_t nr_pages_avail;
1552 ghc->generation = slots->generation;
1554 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1555 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, &nr_pages_avail);
1556 if (!kvm_is_error_hva(ghc->hva) && nr_pages_avail >= nr_pages_needed) {
1560 * If the requested region crosses two memslots, we still
1561 * verify that the entire region is valid here.
1563 while (start_gfn <= end_gfn) {
1564 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1565 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1567 if (kvm_is_error_hva(ghc->hva))
1569 start_gfn += nr_pages_avail;
1571 /* Use the slow path for cross page reads and writes. */
1572 ghc->memslot = NULL;
1576 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1578 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1579 void *data, unsigned long len)
1581 struct kvm_memslots *slots = kvm_memslots(kvm);
1584 BUG_ON(len > ghc->len);
1586 if (slots->generation != ghc->generation)
1587 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1589 if (unlikely(!ghc->memslot))
1590 return kvm_write_guest(kvm, ghc->gpa, data, len);
1592 if (kvm_is_error_hva(ghc->hva))
1595 r = __copy_to_user((void __user *)ghc->hva, data, len);
1598 mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1602 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1604 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1605 void *data, unsigned long len)
1607 struct kvm_memslots *slots = kvm_memslots(kvm);
1610 BUG_ON(len > ghc->len);
1612 if (slots->generation != ghc->generation)
1613 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1615 if (unlikely(!ghc->memslot))
1616 return kvm_read_guest(kvm, ghc->gpa, data, len);
1618 if (kvm_is_error_hva(ghc->hva))
1621 r = __copy_from_user(data, (void __user *)ghc->hva, len);
1627 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1629 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1631 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
1633 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
1635 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1637 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1639 gfn_t gfn = gpa >> PAGE_SHIFT;
1641 int offset = offset_in_page(gpa);
1644 while ((seg = next_segment(len, offset)) != 0) {
1645 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1654 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1656 static void mark_page_dirty_in_slot(struct kvm *kvm,
1657 struct kvm_memory_slot *memslot,
1660 if (memslot && memslot->dirty_bitmap) {
1661 unsigned long rel_gfn = gfn - memslot->base_gfn;
1663 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1667 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1669 struct kvm_memory_slot *memslot;
1671 memslot = gfn_to_memslot(kvm, gfn);
1672 mark_page_dirty_in_slot(kvm, memslot, gfn);
1674 EXPORT_SYMBOL_GPL(mark_page_dirty);
1677 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1679 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1684 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1686 if (kvm_arch_vcpu_runnable(vcpu)) {
1687 kvm_make_request(KVM_REQ_UNHALT, vcpu);
1690 if (kvm_cpu_has_pending_timer(vcpu))
1692 if (signal_pending(current))
1698 finish_wait(&vcpu->wq, &wait);
1700 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
1704 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1706 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1709 int cpu = vcpu->cpu;
1710 wait_queue_head_t *wqp;
1712 wqp = kvm_arch_vcpu_wq(vcpu);
1713 if (waitqueue_active(wqp)) {
1714 wake_up_interruptible(wqp);
1715 ++vcpu->stat.halt_wakeup;
1719 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1720 if (kvm_arch_vcpu_should_kick(vcpu))
1721 smp_send_reschedule(cpu);
1724 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
1725 #endif /* !CONFIG_S390 */
1727 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
1730 struct task_struct *task = NULL;
1734 pid = rcu_dereference(target->pid);
1736 task = get_pid_task(target->pid, PIDTYPE_PID);
1740 if (task->flags & PF_VCPU) {
1741 put_task_struct(task);
1744 ret = yield_to(task, 1);
1745 put_task_struct(task);
1749 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1752 * Helper that checks whether a VCPU is eligible for directed yield.
1753 * Most eligible candidate to yield is decided by following heuristics:
1755 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1756 * (preempted lock holder), indicated by @in_spin_loop.
1757 * Set at the beiginning and cleared at the end of interception/PLE handler.
1759 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1760 * chance last time (mostly it has become eligible now since we have probably
1761 * yielded to lockholder in last iteration. This is done by toggling
1762 * @dy_eligible each time a VCPU checked for eligibility.)
1764 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1765 * to preempted lock-holder could result in wrong VCPU selection and CPU
1766 * burning. Giving priority for a potential lock-holder increases lock
1769 * Since algorithm is based on heuristics, accessing another VCPU data without
1770 * locking does not harm. It may result in trying to yield to same VCPU, fail
1771 * and continue with next VCPU and so on.
1773 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1775 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1778 eligible = !vcpu->spin_loop.in_spin_loop ||
1779 (vcpu->spin_loop.in_spin_loop &&
1780 vcpu->spin_loop.dy_eligible);
1782 if (vcpu->spin_loop.in_spin_loop)
1783 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1791 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1793 struct kvm *kvm = me->kvm;
1794 struct kvm_vcpu *vcpu;
1795 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1801 kvm_vcpu_set_in_spin_loop(me, true);
1803 * We boost the priority of a VCPU that is runnable but not
1804 * currently running, because it got preempted by something
1805 * else and called schedule in __vcpu_run. Hopefully that
1806 * VCPU is holding the lock that we need and will release it.
1807 * We approximate round-robin by starting at the last boosted VCPU.
1809 for (pass = 0; pass < 2 && !yielded && try; pass++) {
1810 kvm_for_each_vcpu(i, vcpu, kvm) {
1811 if (!pass && i <= last_boosted_vcpu) {
1812 i = last_boosted_vcpu;
1814 } else if (pass && i > last_boosted_vcpu)
1816 if (!ACCESS_ONCE(vcpu->preempted))
1820 if (waitqueue_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
1822 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1825 yielded = kvm_vcpu_yield_to(vcpu);
1827 kvm->last_boosted_vcpu = i;
1829 } else if (yielded < 0) {
1836 kvm_vcpu_set_in_spin_loop(me, false);
1838 /* Ensure vcpu is not eligible during next spinloop */
1839 kvm_vcpu_set_dy_eligible(me, false);
1841 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1843 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1845 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1848 if (vmf->pgoff == 0)
1849 page = virt_to_page(vcpu->run);
1851 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1852 page = virt_to_page(vcpu->arch.pio_data);
1854 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1855 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1856 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1859 return kvm_arch_vcpu_fault(vcpu, vmf);
1865 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1866 .fault = kvm_vcpu_fault,
1869 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1871 vma->vm_ops = &kvm_vcpu_vm_ops;
1875 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1877 struct kvm_vcpu *vcpu = filp->private_data;
1879 kvm_put_kvm(vcpu->kvm);
1883 static struct file_operations kvm_vcpu_fops = {
1884 .release = kvm_vcpu_release,
1885 .unlocked_ioctl = kvm_vcpu_ioctl,
1886 #ifdef CONFIG_COMPAT
1887 .compat_ioctl = kvm_vcpu_compat_ioctl,
1889 .mmap = kvm_vcpu_mmap,
1890 .llseek = noop_llseek,
1894 * Allocates an inode for the vcpu.
1896 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1898 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
1902 * Creates some virtual cpus. Good luck creating more than one.
1904 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1907 struct kvm_vcpu *vcpu, *v;
1909 if (id >= KVM_MAX_VCPUS)
1912 vcpu = kvm_arch_vcpu_create(kvm, id);
1914 return PTR_ERR(vcpu);
1916 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1918 r = kvm_arch_vcpu_setup(vcpu);
1922 mutex_lock(&kvm->lock);
1923 if (!kvm_vcpu_compatible(vcpu)) {
1925 goto unlock_vcpu_destroy;
1927 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1929 goto unlock_vcpu_destroy;
1932 kvm_for_each_vcpu(r, v, kvm)
1933 if (v->vcpu_id == id) {
1935 goto unlock_vcpu_destroy;
1938 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1940 /* Now it's all set up, let userspace reach it */
1942 r = create_vcpu_fd(vcpu);
1945 goto unlock_vcpu_destroy;
1948 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1950 atomic_inc(&kvm->online_vcpus);
1952 mutex_unlock(&kvm->lock);
1953 kvm_arch_vcpu_postcreate(vcpu);
1956 unlock_vcpu_destroy:
1957 mutex_unlock(&kvm->lock);
1959 kvm_arch_vcpu_destroy(vcpu);
1963 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1966 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1967 vcpu->sigset_active = 1;
1968 vcpu->sigset = *sigset;
1970 vcpu->sigset_active = 0;
1974 static long kvm_vcpu_ioctl(struct file *filp,
1975 unsigned int ioctl, unsigned long arg)
1977 struct kvm_vcpu *vcpu = filp->private_data;
1978 void __user *argp = (void __user *)arg;
1980 struct kvm_fpu *fpu = NULL;
1981 struct kvm_sregs *kvm_sregs = NULL;
1983 if (vcpu->kvm->mm != current->mm)
1986 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
1988 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1989 * so vcpu_load() would break it.
1991 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1992 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1996 r = vcpu_load(vcpu);
2004 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2005 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2007 case KVM_GET_REGS: {
2008 struct kvm_regs *kvm_regs;
2011 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2014 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2018 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2025 case KVM_SET_REGS: {
2026 struct kvm_regs *kvm_regs;
2029 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2030 if (IS_ERR(kvm_regs)) {
2031 r = PTR_ERR(kvm_regs);
2034 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2038 case KVM_GET_SREGS: {
2039 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2043 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2047 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2052 case KVM_SET_SREGS: {
2053 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2054 if (IS_ERR(kvm_sregs)) {
2055 r = PTR_ERR(kvm_sregs);
2059 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2062 case KVM_GET_MP_STATE: {
2063 struct kvm_mp_state mp_state;
2065 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2069 if (copy_to_user(argp, &mp_state, sizeof mp_state))
2074 case KVM_SET_MP_STATE: {
2075 struct kvm_mp_state mp_state;
2078 if (copy_from_user(&mp_state, argp, sizeof mp_state))
2080 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2083 case KVM_TRANSLATE: {
2084 struct kvm_translation tr;
2087 if (copy_from_user(&tr, argp, sizeof tr))
2089 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2093 if (copy_to_user(argp, &tr, sizeof tr))
2098 case KVM_SET_GUEST_DEBUG: {
2099 struct kvm_guest_debug dbg;
2102 if (copy_from_user(&dbg, argp, sizeof dbg))
2104 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2107 case KVM_SET_SIGNAL_MASK: {
2108 struct kvm_signal_mask __user *sigmask_arg = argp;
2109 struct kvm_signal_mask kvm_sigmask;
2110 sigset_t sigset, *p;
2115 if (copy_from_user(&kvm_sigmask, argp,
2116 sizeof kvm_sigmask))
2119 if (kvm_sigmask.len != sizeof sigset)
2122 if (copy_from_user(&sigset, sigmask_arg->sigset,
2127 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2131 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2135 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2139 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2145 fpu = memdup_user(argp, sizeof(*fpu));
2151 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2155 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2164 #ifdef CONFIG_COMPAT
2165 static long kvm_vcpu_compat_ioctl(struct file *filp,
2166 unsigned int ioctl, unsigned long arg)
2168 struct kvm_vcpu *vcpu = filp->private_data;
2169 void __user *argp = compat_ptr(arg);
2172 if (vcpu->kvm->mm != current->mm)
2176 case KVM_SET_SIGNAL_MASK: {
2177 struct kvm_signal_mask __user *sigmask_arg = argp;
2178 struct kvm_signal_mask kvm_sigmask;
2179 compat_sigset_t csigset;
2184 if (copy_from_user(&kvm_sigmask, argp,
2185 sizeof kvm_sigmask))
2188 if (kvm_sigmask.len != sizeof csigset)
2191 if (copy_from_user(&csigset, sigmask_arg->sigset,
2194 sigset_from_compat(&sigset, &csigset);
2195 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2197 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2201 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2209 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2210 int (*accessor)(struct kvm_device *dev,
2211 struct kvm_device_attr *attr),
2214 struct kvm_device_attr attr;
2219 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2222 return accessor(dev, &attr);
2225 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2228 struct kvm_device *dev = filp->private_data;
2231 case KVM_SET_DEVICE_ATTR:
2232 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2233 case KVM_GET_DEVICE_ATTR:
2234 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2235 case KVM_HAS_DEVICE_ATTR:
2236 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2238 if (dev->ops->ioctl)
2239 return dev->ops->ioctl(dev, ioctl, arg);
2245 static int kvm_device_release(struct inode *inode, struct file *filp)
2247 struct kvm_device *dev = filp->private_data;
2248 struct kvm *kvm = dev->kvm;
2254 static const struct file_operations kvm_device_fops = {
2255 .unlocked_ioctl = kvm_device_ioctl,
2256 #ifdef CONFIG_COMPAT
2257 .compat_ioctl = kvm_device_ioctl,
2259 .release = kvm_device_release,
2262 struct kvm_device *kvm_device_from_filp(struct file *filp)
2264 if (filp->f_op != &kvm_device_fops)
2267 return filp->private_data;
2270 static int kvm_ioctl_create_device(struct kvm *kvm,
2271 struct kvm_create_device *cd)
2273 struct kvm_device_ops *ops = NULL;
2274 struct kvm_device *dev;
2275 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2279 #ifdef CONFIG_KVM_MPIC
2280 case KVM_DEV_TYPE_FSL_MPIC_20:
2281 case KVM_DEV_TYPE_FSL_MPIC_42:
2282 ops = &kvm_mpic_ops;
2285 #ifdef CONFIG_KVM_XICS
2286 case KVM_DEV_TYPE_XICS:
2287 ops = &kvm_xics_ops;
2290 #ifdef CONFIG_KVM_VFIO
2291 case KVM_DEV_TYPE_VFIO:
2292 ops = &kvm_vfio_ops;
2295 #ifdef CONFIG_KVM_ARM_VGIC
2296 case KVM_DEV_TYPE_ARM_VGIC_V2:
2297 ops = &kvm_arm_vgic_v2_ops;
2301 case KVM_DEV_TYPE_FLIC:
2302 ops = &kvm_flic_ops;
2312 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2319 ret = ops->create(dev, cd->type);
2325 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2331 list_add(&dev->vm_node, &kvm->devices);
2337 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2340 case KVM_CAP_USER_MEMORY:
2341 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2342 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2343 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2344 case KVM_CAP_SET_BOOT_CPU_ID:
2346 case KVM_CAP_INTERNAL_ERROR_DATA:
2347 #ifdef CONFIG_HAVE_KVM_MSI
2348 case KVM_CAP_SIGNAL_MSI:
2350 #ifdef CONFIG_HAVE_KVM_IRQFD
2351 case KVM_CAP_IRQFD_RESAMPLE:
2353 case KVM_CAP_CHECK_EXTENSION_VM:
2355 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2356 case KVM_CAP_IRQ_ROUTING:
2357 return KVM_MAX_IRQ_ROUTES;
2362 return kvm_vm_ioctl_check_extension(kvm, arg);
2365 static long kvm_vm_ioctl(struct file *filp,
2366 unsigned int ioctl, unsigned long arg)
2368 struct kvm *kvm = filp->private_data;
2369 void __user *argp = (void __user *)arg;
2372 if (kvm->mm != current->mm)
2375 case KVM_CREATE_VCPU:
2376 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2378 case KVM_SET_USER_MEMORY_REGION: {
2379 struct kvm_userspace_memory_region kvm_userspace_mem;
2382 if (copy_from_user(&kvm_userspace_mem, argp,
2383 sizeof kvm_userspace_mem))
2386 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2389 case KVM_GET_DIRTY_LOG: {
2390 struct kvm_dirty_log log;
2393 if (copy_from_user(&log, argp, sizeof log))
2395 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2398 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2399 case KVM_REGISTER_COALESCED_MMIO: {
2400 struct kvm_coalesced_mmio_zone zone;
2402 if (copy_from_user(&zone, argp, sizeof zone))
2404 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2407 case KVM_UNREGISTER_COALESCED_MMIO: {
2408 struct kvm_coalesced_mmio_zone zone;
2410 if (copy_from_user(&zone, argp, sizeof zone))
2412 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2417 struct kvm_irqfd data;
2420 if (copy_from_user(&data, argp, sizeof data))
2422 r = kvm_irqfd(kvm, &data);
2425 case KVM_IOEVENTFD: {
2426 struct kvm_ioeventfd data;
2429 if (copy_from_user(&data, argp, sizeof data))
2431 r = kvm_ioeventfd(kvm, &data);
2434 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2435 case KVM_SET_BOOT_CPU_ID:
2437 mutex_lock(&kvm->lock);
2438 if (atomic_read(&kvm->online_vcpus) != 0)
2441 kvm->bsp_vcpu_id = arg;
2442 mutex_unlock(&kvm->lock);
2445 #ifdef CONFIG_HAVE_KVM_MSI
2446 case KVM_SIGNAL_MSI: {
2450 if (copy_from_user(&msi, argp, sizeof msi))
2452 r = kvm_send_userspace_msi(kvm, &msi);
2456 #ifdef __KVM_HAVE_IRQ_LINE
2457 case KVM_IRQ_LINE_STATUS:
2458 case KVM_IRQ_LINE: {
2459 struct kvm_irq_level irq_event;
2462 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2465 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
2466 ioctl == KVM_IRQ_LINE_STATUS);
2471 if (ioctl == KVM_IRQ_LINE_STATUS) {
2472 if (copy_to_user(argp, &irq_event, sizeof irq_event))
2480 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2481 case KVM_SET_GSI_ROUTING: {
2482 struct kvm_irq_routing routing;
2483 struct kvm_irq_routing __user *urouting;
2484 struct kvm_irq_routing_entry *entries;
2487 if (copy_from_user(&routing, argp, sizeof(routing)))
2490 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2495 entries = vmalloc(routing.nr * sizeof(*entries));
2500 if (copy_from_user(entries, urouting->entries,
2501 routing.nr * sizeof(*entries)))
2502 goto out_free_irq_routing;
2503 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2505 out_free_irq_routing:
2509 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
2510 case KVM_CREATE_DEVICE: {
2511 struct kvm_create_device cd;
2514 if (copy_from_user(&cd, argp, sizeof(cd)))
2517 r = kvm_ioctl_create_device(kvm, &cd);
2522 if (copy_to_user(argp, &cd, sizeof(cd)))
2528 case KVM_CHECK_EXTENSION:
2529 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
2532 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2534 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2540 #ifdef CONFIG_COMPAT
2541 struct compat_kvm_dirty_log {
2545 compat_uptr_t dirty_bitmap; /* one bit per page */
2550 static long kvm_vm_compat_ioctl(struct file *filp,
2551 unsigned int ioctl, unsigned long arg)
2553 struct kvm *kvm = filp->private_data;
2556 if (kvm->mm != current->mm)
2559 case KVM_GET_DIRTY_LOG: {
2560 struct compat_kvm_dirty_log compat_log;
2561 struct kvm_dirty_log log;
2564 if (copy_from_user(&compat_log, (void __user *)arg,
2565 sizeof(compat_log)))
2567 log.slot = compat_log.slot;
2568 log.padding1 = compat_log.padding1;
2569 log.padding2 = compat_log.padding2;
2570 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2572 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2576 r = kvm_vm_ioctl(filp, ioctl, arg);
2584 static struct file_operations kvm_vm_fops = {
2585 .release = kvm_vm_release,
2586 .unlocked_ioctl = kvm_vm_ioctl,
2587 #ifdef CONFIG_COMPAT
2588 .compat_ioctl = kvm_vm_compat_ioctl,
2590 .llseek = noop_llseek,
2593 static int kvm_dev_ioctl_create_vm(unsigned long type)
2598 kvm = kvm_create_vm(type);
2600 return PTR_ERR(kvm);
2601 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2602 r = kvm_coalesced_mmio_init(kvm);
2608 r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC);
2615 static long kvm_dev_ioctl(struct file *filp,
2616 unsigned int ioctl, unsigned long arg)
2621 case KVM_GET_API_VERSION:
2625 r = KVM_API_VERSION;
2628 r = kvm_dev_ioctl_create_vm(arg);
2630 case KVM_CHECK_EXTENSION:
2631 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
2633 case KVM_GET_VCPU_MMAP_SIZE:
2637 r = PAGE_SIZE; /* struct kvm_run */
2639 r += PAGE_SIZE; /* pio data page */
2641 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2642 r += PAGE_SIZE; /* coalesced mmio ring page */
2645 case KVM_TRACE_ENABLE:
2646 case KVM_TRACE_PAUSE:
2647 case KVM_TRACE_DISABLE:
2651 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2657 static struct file_operations kvm_chardev_ops = {
2658 .unlocked_ioctl = kvm_dev_ioctl,
2659 .compat_ioctl = kvm_dev_ioctl,
2660 .llseek = noop_llseek,
2663 static struct miscdevice kvm_dev = {
2669 static void hardware_enable_nolock(void *junk)
2671 int cpu = raw_smp_processor_id();
2674 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2677 cpumask_set_cpu(cpu, cpus_hardware_enabled);
2679 r = kvm_arch_hardware_enable();
2682 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2683 atomic_inc(&hardware_enable_failed);
2684 printk(KERN_INFO "kvm: enabling virtualization on "
2685 "CPU%d failed\n", cpu);
2689 static void hardware_enable(void)
2691 raw_spin_lock(&kvm_count_lock);
2692 if (kvm_usage_count)
2693 hardware_enable_nolock(NULL);
2694 raw_spin_unlock(&kvm_count_lock);
2697 static void hardware_disable_nolock(void *junk)
2699 int cpu = raw_smp_processor_id();
2701 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2703 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2704 kvm_arch_hardware_disable();
2707 static void hardware_disable(void)
2709 raw_spin_lock(&kvm_count_lock);
2710 if (kvm_usage_count)
2711 hardware_disable_nolock(NULL);
2712 raw_spin_unlock(&kvm_count_lock);
2715 static void hardware_disable_all_nolock(void)
2717 BUG_ON(!kvm_usage_count);
2720 if (!kvm_usage_count)
2721 on_each_cpu(hardware_disable_nolock, NULL, 1);
2724 static void hardware_disable_all(void)
2726 raw_spin_lock(&kvm_count_lock);
2727 hardware_disable_all_nolock();
2728 raw_spin_unlock(&kvm_count_lock);
2731 static int hardware_enable_all(void)
2735 raw_spin_lock(&kvm_count_lock);
2738 if (kvm_usage_count == 1) {
2739 atomic_set(&hardware_enable_failed, 0);
2740 on_each_cpu(hardware_enable_nolock, NULL, 1);
2742 if (atomic_read(&hardware_enable_failed)) {
2743 hardware_disable_all_nolock();
2748 raw_spin_unlock(&kvm_count_lock);
2753 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2758 val &= ~CPU_TASKS_FROZEN;
2761 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2766 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2774 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2778 * Some (well, at least mine) BIOSes hang on reboot if
2781 * And Intel TXT required VMX off for all cpu when system shutdown.
2783 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2784 kvm_rebooting = true;
2785 on_each_cpu(hardware_disable_nolock, NULL, 1);
2789 static struct notifier_block kvm_reboot_notifier = {
2790 .notifier_call = kvm_reboot,
2794 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2798 for (i = 0; i < bus->dev_count; i++) {
2799 struct kvm_io_device *pos = bus->range[i].dev;
2801 kvm_iodevice_destructor(pos);
2806 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
2807 const struct kvm_io_range *r2)
2809 if (r1->addr < r2->addr)
2811 if (r1->addr + r1->len > r2->addr + r2->len)
2816 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2818 return kvm_io_bus_cmp(p1, p2);
2821 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2822 gpa_t addr, int len)
2824 bus->range[bus->dev_count++] = (struct kvm_io_range) {
2830 sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2831 kvm_io_bus_sort_cmp, NULL);
2836 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2837 gpa_t addr, int len)
2839 struct kvm_io_range *range, key;
2842 key = (struct kvm_io_range) {
2847 range = bsearch(&key, bus->range, bus->dev_count,
2848 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2852 off = range - bus->range;
2854 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
2860 static int __kvm_io_bus_write(struct kvm_io_bus *bus,
2861 struct kvm_io_range *range, const void *val)
2865 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2869 while (idx < bus->dev_count &&
2870 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
2871 if (!kvm_iodevice_write(bus->range[idx].dev, range->addr,
2880 /* kvm_io_bus_write - called under kvm->slots_lock */
2881 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2882 int len, const void *val)
2884 struct kvm_io_bus *bus;
2885 struct kvm_io_range range;
2888 range = (struct kvm_io_range) {
2893 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2894 r = __kvm_io_bus_write(bus, &range, val);
2895 return r < 0 ? r : 0;
2898 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
2899 int kvm_io_bus_write_cookie(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2900 int len, const void *val, long cookie)
2902 struct kvm_io_bus *bus;
2903 struct kvm_io_range range;
2905 range = (struct kvm_io_range) {
2910 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2912 /* First try the device referenced by cookie. */
2913 if ((cookie >= 0) && (cookie < bus->dev_count) &&
2914 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
2915 if (!kvm_iodevice_write(bus->range[cookie].dev, addr, len,
2920 * cookie contained garbage; fall back to search and return the
2921 * correct cookie value.
2923 return __kvm_io_bus_write(bus, &range, val);
2926 static int __kvm_io_bus_read(struct kvm_io_bus *bus, struct kvm_io_range *range,
2931 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2935 while (idx < bus->dev_count &&
2936 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
2937 if (!kvm_iodevice_read(bus->range[idx].dev, range->addr,
2945 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
2947 /* kvm_io_bus_read - called under kvm->slots_lock */
2948 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2951 struct kvm_io_bus *bus;
2952 struct kvm_io_range range;
2955 range = (struct kvm_io_range) {
2960 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2961 r = __kvm_io_bus_read(bus, &range, val);
2962 return r < 0 ? r : 0;
2966 /* Caller must hold slots_lock. */
2967 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2968 int len, struct kvm_io_device *dev)
2970 struct kvm_io_bus *new_bus, *bus;
2972 bus = kvm->buses[bus_idx];
2973 /* exclude ioeventfd which is limited by maximum fd */
2974 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
2977 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
2978 sizeof(struct kvm_io_range)), GFP_KERNEL);
2981 memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
2982 sizeof(struct kvm_io_range)));
2983 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2984 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2985 synchronize_srcu_expedited(&kvm->srcu);
2991 /* Caller must hold slots_lock. */
2992 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2993 struct kvm_io_device *dev)
2996 struct kvm_io_bus *new_bus, *bus;
2998 bus = kvm->buses[bus_idx];
3000 for (i = 0; i < bus->dev_count; i++)
3001 if (bus->range[i].dev == dev) {
3009 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3010 sizeof(struct kvm_io_range)), GFP_KERNEL);
3014 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3015 new_bus->dev_count--;
3016 memcpy(new_bus->range + i, bus->range + i + 1,
3017 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3019 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3020 synchronize_srcu_expedited(&kvm->srcu);
3025 static struct notifier_block kvm_cpu_notifier = {
3026 .notifier_call = kvm_cpu_hotplug,
3029 static int vm_stat_get(void *_offset, u64 *val)
3031 unsigned offset = (long)_offset;
3035 spin_lock(&kvm_lock);
3036 list_for_each_entry(kvm, &vm_list, vm_list)
3037 *val += *(u32 *)((void *)kvm + offset);
3038 spin_unlock(&kvm_lock);
3042 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
3044 static int vcpu_stat_get(void *_offset, u64 *val)
3046 unsigned offset = (long)_offset;
3048 struct kvm_vcpu *vcpu;
3052 spin_lock(&kvm_lock);
3053 list_for_each_entry(kvm, &vm_list, vm_list)
3054 kvm_for_each_vcpu(i, vcpu, kvm)
3055 *val += *(u32 *)((void *)vcpu + offset);
3057 spin_unlock(&kvm_lock);
3061 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
3063 static const struct file_operations *stat_fops[] = {
3064 [KVM_STAT_VCPU] = &vcpu_stat_fops,
3065 [KVM_STAT_VM] = &vm_stat_fops,
3068 static int kvm_init_debug(void)
3071 struct kvm_stats_debugfs_item *p;
3073 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3074 if (kvm_debugfs_dir == NULL)
3077 for (p = debugfs_entries; p->name; ++p) {
3078 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
3079 (void *)(long)p->offset,
3080 stat_fops[p->kind]);
3081 if (p->dentry == NULL)
3088 debugfs_remove_recursive(kvm_debugfs_dir);
3093 static void kvm_exit_debug(void)
3095 struct kvm_stats_debugfs_item *p;
3097 for (p = debugfs_entries; p->name; ++p)
3098 debugfs_remove(p->dentry);
3099 debugfs_remove(kvm_debugfs_dir);
3102 static int kvm_suspend(void)
3104 if (kvm_usage_count)
3105 hardware_disable_nolock(NULL);
3109 static void kvm_resume(void)
3111 if (kvm_usage_count) {
3112 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3113 hardware_enable_nolock(NULL);
3117 static struct syscore_ops kvm_syscore_ops = {
3118 .suspend = kvm_suspend,
3119 .resume = kvm_resume,
3123 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3125 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3128 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3130 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3131 if (vcpu->preempted)
3132 vcpu->preempted = false;
3134 kvm_arch_sched_in(vcpu, cpu);
3136 kvm_arch_vcpu_load(vcpu, cpu);
3139 static void kvm_sched_out(struct preempt_notifier *pn,
3140 struct task_struct *next)
3142 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3144 if (current->state == TASK_RUNNING)
3145 vcpu->preempted = true;
3146 kvm_arch_vcpu_put(vcpu);
3149 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3150 struct module *module)
3155 r = kvm_arch_init(opaque);
3160 * kvm_arch_init makes sure there's at most one caller
3161 * for architectures that support multiple implementations,
3162 * like intel and amd on x86.
3163 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3164 * conflicts in case kvm is already setup for another implementation.
3166 r = kvm_irqfd_init();
3170 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
3175 r = kvm_arch_hardware_setup();
3179 for_each_online_cpu(cpu) {
3180 smp_call_function_single(cpu,
3181 kvm_arch_check_processor_compat,
3187 r = register_cpu_notifier(&kvm_cpu_notifier);
3190 register_reboot_notifier(&kvm_reboot_notifier);
3192 /* A kmem cache lets us meet the alignment requirements of fx_save. */
3194 vcpu_align = __alignof__(struct kvm_vcpu);
3195 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
3197 if (!kvm_vcpu_cache) {
3202 r = kvm_async_pf_init();
3206 kvm_chardev_ops.owner = module;
3207 kvm_vm_fops.owner = module;
3208 kvm_vcpu_fops.owner = module;
3210 r = misc_register(&kvm_dev);
3212 printk(KERN_ERR "kvm: misc device register failed\n");
3216 register_syscore_ops(&kvm_syscore_ops);
3218 kvm_preempt_ops.sched_in = kvm_sched_in;
3219 kvm_preempt_ops.sched_out = kvm_sched_out;
3221 r = kvm_init_debug();
3223 printk(KERN_ERR "kvm: create debugfs files failed\n");
3230 unregister_syscore_ops(&kvm_syscore_ops);
3231 misc_deregister(&kvm_dev);
3233 kvm_async_pf_deinit();
3235 kmem_cache_destroy(kvm_vcpu_cache);
3237 unregister_reboot_notifier(&kvm_reboot_notifier);
3238 unregister_cpu_notifier(&kvm_cpu_notifier);
3241 kvm_arch_hardware_unsetup();
3243 free_cpumask_var(cpus_hardware_enabled);
3251 EXPORT_SYMBOL_GPL(kvm_init);
3256 misc_deregister(&kvm_dev);
3257 kmem_cache_destroy(kvm_vcpu_cache);
3258 kvm_async_pf_deinit();
3259 unregister_syscore_ops(&kvm_syscore_ops);
3260 unregister_reboot_notifier(&kvm_reboot_notifier);
3261 unregister_cpu_notifier(&kvm_cpu_notifier);
3262 on_each_cpu(hardware_disable_nolock, NULL, 1);
3263 kvm_arch_hardware_unsetup();
3266 free_cpumask_var(cpus_hardware_enabled);
3268 EXPORT_SYMBOL_GPL(kvm_exit);