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/ioctl.h>
56 #include <asm/uaccess.h>
57 #include <asm/pgtable.h>
59 #include "coalesced_mmio.h"
63 #define CREATE_TRACE_POINTS
64 #include <trace/events/kvm.h>
66 MODULE_AUTHOR("Qumranet");
67 MODULE_LICENSE("GPL");
72 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
75 DEFINE_SPINLOCK(kvm_lock);
76 static DEFINE_RAW_SPINLOCK(kvm_count_lock);
79 static cpumask_var_t cpus_hardware_enabled;
80 static int kvm_usage_count = 0;
81 static atomic_t hardware_enable_failed;
83 struct kmem_cache *kvm_vcpu_cache;
84 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
86 static __read_mostly struct preempt_ops kvm_preempt_ops;
88 struct dentry *kvm_debugfs_dir;
90 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
93 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
96 static int hardware_enable_all(void);
97 static void hardware_disable_all(void);
99 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
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)
702 struct kvm_memory_slot *old = id_to_memslot(slots, id);
703 unsigned long npages = old->npages;
706 if (new->npages != npages)
707 sort_memslots(slots);
711 static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
713 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
715 #ifdef __KVM_HAVE_READONLY_MEM
716 valid_flags |= KVM_MEM_READONLY;
719 if (mem->flags & ~valid_flags)
725 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
726 struct kvm_memslots *slots, struct kvm_memory_slot *new)
728 struct kvm_memslots *old_memslots = kvm->memslots;
731 * Set the low bit in the generation, which disables SPTE caching
732 * until the end of synchronize_srcu_expedited.
734 WARN_ON(old_memslots->generation & 1);
735 slots->generation = old_memslots->generation + 1;
737 update_memslots(slots, new);
738 rcu_assign_pointer(kvm->memslots, slots);
739 synchronize_srcu_expedited(&kvm->srcu);
742 * Increment the new memslot generation a second time. This prevents
743 * vm exits that race with memslot updates from caching a memslot
744 * generation that will (potentially) be valid forever.
748 kvm_arch_memslots_updated(kvm);
754 * Allocate some memory and give it an address in the guest physical address
757 * Discontiguous memory is allowed, mostly for framebuffers.
759 * Must be called holding mmap_sem for write.
761 int __kvm_set_memory_region(struct kvm *kvm,
762 struct kvm_userspace_memory_region *mem)
766 unsigned long npages;
767 struct kvm_memory_slot *slot;
768 struct kvm_memory_slot old, new;
769 struct kvm_memslots *slots = NULL, *old_memslots;
770 enum kvm_mr_change change;
772 r = check_memory_region_flags(mem);
777 /* General sanity checks */
778 if (mem->memory_size & (PAGE_SIZE - 1))
780 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
782 /* We can read the guest memory with __xxx_user() later on. */
783 if ((mem->slot < KVM_USER_MEM_SLOTS) &&
784 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
785 !access_ok(VERIFY_WRITE,
786 (void __user *)(unsigned long)mem->userspace_addr,
789 if (mem->slot >= KVM_MEM_SLOTS_NUM)
791 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
794 slot = id_to_memslot(kvm->memslots, mem->slot);
795 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
796 npages = mem->memory_size >> PAGE_SHIFT;
798 if (npages > KVM_MEM_MAX_NR_PAGES)
802 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
807 new.base_gfn = base_gfn;
809 new.flags = mem->flags;
813 change = KVM_MR_CREATE;
814 else { /* Modify an existing slot. */
815 if ((mem->userspace_addr != old.userspace_addr) ||
816 (npages != old.npages) ||
817 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
820 if (base_gfn != old.base_gfn)
821 change = KVM_MR_MOVE;
822 else if (new.flags != old.flags)
823 change = KVM_MR_FLAGS_ONLY;
824 else { /* Nothing to change. */
829 } else if (old.npages) {
830 change = KVM_MR_DELETE;
831 } else /* Modify a non-existent slot: disallowed. */
834 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
835 /* Check for overlaps */
837 kvm_for_each_memslot(slot, kvm->memslots) {
838 if ((slot->id >= KVM_USER_MEM_SLOTS) ||
839 (slot->id == mem->slot))
841 if (!((base_gfn + npages <= slot->base_gfn) ||
842 (base_gfn >= slot->base_gfn + slot->npages)))
847 /* Free page dirty bitmap if unneeded */
848 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
849 new.dirty_bitmap = NULL;
852 if (change == KVM_MR_CREATE) {
853 new.userspace_addr = mem->userspace_addr;
855 if (kvm_arch_create_memslot(kvm, &new, npages))
859 /* Allocate page dirty bitmap if needed */
860 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
861 if (kvm_create_dirty_bitmap(&new) < 0)
865 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
866 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
870 slot = id_to_memslot(slots, mem->slot);
871 slot->flags |= KVM_MEMSLOT_INVALID;
873 old_memslots = install_new_memslots(kvm, slots, NULL);
875 /* slot was deleted or moved, clear iommu mapping */
876 kvm_iommu_unmap_pages(kvm, &old);
877 /* From this point no new shadow pages pointing to a deleted,
878 * or moved, memslot will be created.
880 * validation of sp->gfn happens in:
881 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
882 * - kvm_is_visible_gfn (mmu_check_roots)
884 kvm_arch_flush_shadow_memslot(kvm, slot);
885 slots = old_memslots;
888 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
894 * We can re-use the old_memslots from above, the only difference
895 * from the currently installed memslots is the invalid flag. This
896 * will get overwritten by update_memslots anyway.
899 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
905 /* actual memory is freed via old in kvm_free_physmem_slot below */
906 if (change == KVM_MR_DELETE) {
907 new.dirty_bitmap = NULL;
908 memset(&new.arch, 0, sizeof(new.arch));
911 old_memslots = install_new_memslots(kvm, slots, &new);
913 kvm_arch_commit_memory_region(kvm, mem, &old, change);
915 kvm_free_physmem_slot(kvm, &old, &new);
919 * IOMMU mapping: New slots need to be mapped. Old slots need to be
920 * un-mapped and re-mapped if their base changes. Since base change
921 * unmapping is handled above with slot deletion, mapping alone is
922 * needed here. Anything else the iommu might care about for existing
923 * slots (size changes, userspace addr changes and read-only flag
924 * changes) is disallowed above, so any other attribute changes getting
925 * here can be skipped.
927 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
928 r = kvm_iommu_map_pages(kvm, &new);
937 kvm_free_physmem_slot(kvm, &new, &old);
941 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
943 int kvm_set_memory_region(struct kvm *kvm,
944 struct kvm_userspace_memory_region *mem)
948 mutex_lock(&kvm->slots_lock);
949 r = __kvm_set_memory_region(kvm, mem);
950 mutex_unlock(&kvm->slots_lock);
953 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
955 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
956 struct kvm_userspace_memory_region *mem)
958 if (mem->slot >= KVM_USER_MEM_SLOTS)
960 return kvm_set_memory_region(kvm, mem);
963 int kvm_get_dirty_log(struct kvm *kvm,
964 struct kvm_dirty_log *log, int *is_dirty)
966 struct kvm_memory_slot *memslot;
969 unsigned long any = 0;
972 if (log->slot >= KVM_USER_MEM_SLOTS)
975 memslot = id_to_memslot(kvm->memslots, log->slot);
977 if (!memslot->dirty_bitmap)
980 n = kvm_dirty_bitmap_bytes(memslot);
982 for (i = 0; !any && i < n/sizeof(long); ++i)
983 any = memslot->dirty_bitmap[i];
986 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
996 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
998 bool kvm_largepages_enabled(void)
1000 return largepages_enabled;
1003 void kvm_disable_largepages(void)
1005 largepages_enabled = false;
1007 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1009 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1011 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1013 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1015 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1017 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1019 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1020 memslot->flags & KVM_MEMSLOT_INVALID)
1025 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1027 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1029 struct vm_area_struct *vma;
1030 unsigned long addr, size;
1034 addr = gfn_to_hva(kvm, gfn);
1035 if (kvm_is_error_hva(addr))
1038 down_read(¤t->mm->mmap_sem);
1039 vma = find_vma(current->mm, addr);
1043 size = vma_kernel_pagesize(vma);
1046 up_read(¤t->mm->mmap_sem);
1051 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1053 return slot->flags & KVM_MEM_READONLY;
1056 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1057 gfn_t *nr_pages, bool write)
1059 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1060 return KVM_HVA_ERR_BAD;
1062 if (memslot_is_readonly(slot) && write)
1063 return KVM_HVA_ERR_RO_BAD;
1066 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1068 return __gfn_to_hva_memslot(slot, gfn);
1071 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1074 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1077 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1080 return gfn_to_hva_many(slot, gfn, NULL);
1082 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1084 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1086 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1088 EXPORT_SYMBOL_GPL(gfn_to_hva);
1091 * If writable is set to false, the hva returned by this function is only
1092 * allowed to be read.
1094 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1096 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1097 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1099 if (!kvm_is_error_hva(hva) && writable)
1100 *writable = !memslot_is_readonly(slot);
1105 static int kvm_read_hva(void *data, void __user *hva, int len)
1107 return __copy_from_user(data, hva, len);
1110 static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1112 return __copy_from_user_inatomic(data, hva, len);
1115 static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1116 unsigned long start, int write, struct page **page)
1118 int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1121 flags |= FOLL_WRITE;
1123 return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1126 int kvm_get_user_page_io(struct task_struct *tsk, struct mm_struct *mm,
1127 unsigned long addr, bool write_fault,
1128 struct page **pagep)
1132 int flags = FOLL_TOUCH | FOLL_HWPOISON |
1133 (pagep ? FOLL_GET : 0) |
1134 (write_fault ? FOLL_WRITE : 0);
1137 * If retrying the fault, we get here *not* having allowed the filemap
1138 * to wait on the page lock. We should now allow waiting on the IO with
1139 * the mmap semaphore released.
1141 down_read(&mm->mmap_sem);
1142 npages = __get_user_pages(tsk, mm, addr, 1, flags, pagep, NULL,
1145 VM_BUG_ON(npages != -EBUSY);
1151 * The previous call has now waited on the IO. Now we can
1152 * retry and complete. Pass TRIED to ensure we do not re
1153 * schedule async IO (see e.g. filemap_fault).
1155 down_read(&mm->mmap_sem);
1156 npages = __get_user_pages(tsk, mm, addr, 1, flags | FOLL_TRIED,
1159 up_read(&mm->mmap_sem);
1163 static inline int check_user_page_hwpoison(unsigned long addr)
1165 int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1167 rc = __get_user_pages(current, current->mm, addr, 1,
1168 flags, NULL, NULL, NULL);
1169 return rc == -EHWPOISON;
1173 * The atomic path to get the writable pfn which will be stored in @pfn,
1174 * true indicates success, otherwise false is returned.
1176 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1177 bool write_fault, bool *writable, pfn_t *pfn)
1179 struct page *page[1];
1182 if (!(async || atomic))
1186 * Fast pin a writable pfn only if it is a write fault request
1187 * or the caller allows to map a writable pfn for a read fault
1190 if (!(write_fault || writable))
1193 npages = __get_user_pages_fast(addr, 1, 1, page);
1195 *pfn = page_to_pfn(page[0]);
1206 * The slow path to get the pfn of the specified host virtual address,
1207 * 1 indicates success, -errno is returned if error is detected.
1209 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1210 bool *writable, pfn_t *pfn)
1212 struct page *page[1];
1218 *writable = write_fault;
1221 down_read(¤t->mm->mmap_sem);
1222 npages = get_user_page_nowait(current, current->mm,
1223 addr, write_fault, page);
1224 up_read(¤t->mm->mmap_sem);
1227 * By now we have tried gup_fast, and possibly async_pf, and we
1228 * are certainly not atomic. Time to retry the gup, allowing
1229 * mmap semaphore to be relinquished in the case of IO.
1231 npages = kvm_get_user_page_io(current, current->mm, addr,
1237 /* map read fault as writable if possible */
1238 if (unlikely(!write_fault) && writable) {
1239 struct page *wpage[1];
1241 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1250 *pfn = page_to_pfn(page[0]);
1254 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1256 if (unlikely(!(vma->vm_flags & VM_READ)))
1259 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1266 * Pin guest page in memory and return its pfn.
1267 * @addr: host virtual address which maps memory to the guest
1268 * @atomic: whether this function can sleep
1269 * @async: whether this function need to wait IO complete if the
1270 * host page is not in the memory
1271 * @write_fault: whether we should get a writable host page
1272 * @writable: whether it allows to map a writable host page for !@write_fault
1274 * The function will map a writable host page for these two cases:
1275 * 1): @write_fault = true
1276 * 2): @write_fault = false && @writable, @writable will tell the caller
1277 * whether the mapping is writable.
1279 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1280 bool write_fault, bool *writable)
1282 struct vm_area_struct *vma;
1286 /* we can do it either atomically or asynchronously, not both */
1287 BUG_ON(atomic && async);
1289 if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1293 return KVM_PFN_ERR_FAULT;
1295 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1299 down_read(¤t->mm->mmap_sem);
1300 if (npages == -EHWPOISON ||
1301 (!async && check_user_page_hwpoison(addr))) {
1302 pfn = KVM_PFN_ERR_HWPOISON;
1306 vma = find_vma_intersection(current->mm, addr, addr + 1);
1309 pfn = KVM_PFN_ERR_FAULT;
1310 else if ((vma->vm_flags & VM_PFNMAP)) {
1311 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1313 BUG_ON(!kvm_is_mmio_pfn(pfn));
1315 if (async && vma_is_valid(vma, write_fault))
1317 pfn = KVM_PFN_ERR_FAULT;
1320 up_read(¤t->mm->mmap_sem);
1325 __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1326 bool *async, bool write_fault, bool *writable)
1328 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1330 if (addr == KVM_HVA_ERR_RO_BAD)
1331 return KVM_PFN_ERR_RO_FAULT;
1333 if (kvm_is_error_hva(addr))
1334 return KVM_PFN_NOSLOT;
1336 /* Do not map writable pfn in the readonly memslot. */
1337 if (writable && memslot_is_readonly(slot)) {
1342 return hva_to_pfn(addr, atomic, async, write_fault,
1346 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1347 bool write_fault, bool *writable)
1349 struct kvm_memory_slot *slot;
1354 slot = gfn_to_memslot(kvm, gfn);
1356 return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1360 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1362 return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1364 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1366 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1367 bool write_fault, bool *writable)
1369 return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1371 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1373 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1375 return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1377 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1379 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1382 return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1384 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1386 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1388 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1391 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1393 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1395 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1397 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1403 addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1404 if (kvm_is_error_hva(addr))
1407 if (entry < nr_pages)
1410 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1412 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1414 static struct page *kvm_pfn_to_page(pfn_t pfn)
1416 if (is_error_noslot_pfn(pfn))
1417 return KVM_ERR_PTR_BAD_PAGE;
1419 if (kvm_is_mmio_pfn(pfn)) {
1421 return KVM_ERR_PTR_BAD_PAGE;
1424 return pfn_to_page(pfn);
1427 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1431 pfn = gfn_to_pfn(kvm, gfn);
1433 return kvm_pfn_to_page(pfn);
1436 EXPORT_SYMBOL_GPL(gfn_to_page);
1438 void kvm_release_page_clean(struct page *page)
1440 WARN_ON(is_error_page(page));
1442 kvm_release_pfn_clean(page_to_pfn(page));
1444 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1446 void kvm_release_pfn_clean(pfn_t pfn)
1448 if (!is_error_noslot_pfn(pfn) && !kvm_is_mmio_pfn(pfn))
1449 put_page(pfn_to_page(pfn));
1451 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1453 void kvm_release_page_dirty(struct page *page)
1455 WARN_ON(is_error_page(page));
1457 kvm_release_pfn_dirty(page_to_pfn(page));
1459 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1461 static void kvm_release_pfn_dirty(pfn_t pfn)
1463 kvm_set_pfn_dirty(pfn);
1464 kvm_release_pfn_clean(pfn);
1467 void kvm_set_pfn_dirty(pfn_t pfn)
1469 if (!kvm_is_mmio_pfn(pfn)) {
1470 struct page *page = pfn_to_page(pfn);
1471 if (!PageReserved(page))
1475 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1477 void kvm_set_pfn_accessed(pfn_t pfn)
1479 if (!kvm_is_mmio_pfn(pfn))
1480 mark_page_accessed(pfn_to_page(pfn));
1482 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1484 void kvm_get_pfn(pfn_t pfn)
1486 if (!kvm_is_mmio_pfn(pfn))
1487 get_page(pfn_to_page(pfn));
1489 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1491 static int next_segment(unsigned long len, int offset)
1493 if (len > PAGE_SIZE - offset)
1494 return PAGE_SIZE - offset;
1499 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1505 addr = gfn_to_hva_prot(kvm, gfn, NULL);
1506 if (kvm_is_error_hva(addr))
1508 r = kvm_read_hva(data, (void __user *)addr + offset, len);
1513 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1515 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1517 gfn_t gfn = gpa >> PAGE_SHIFT;
1519 int offset = offset_in_page(gpa);
1522 while ((seg = next_segment(len, offset)) != 0) {
1523 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1533 EXPORT_SYMBOL_GPL(kvm_read_guest);
1535 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1540 gfn_t gfn = gpa >> PAGE_SHIFT;
1541 int offset = offset_in_page(gpa);
1543 addr = gfn_to_hva_prot(kvm, gfn, NULL);
1544 if (kvm_is_error_hva(addr))
1546 pagefault_disable();
1547 r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
1553 EXPORT_SYMBOL(kvm_read_guest_atomic);
1555 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1556 int offset, int len)
1561 addr = gfn_to_hva(kvm, gfn);
1562 if (kvm_is_error_hva(addr))
1564 r = __copy_to_user((void __user *)addr + offset, data, len);
1567 mark_page_dirty(kvm, gfn);
1570 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1572 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1575 gfn_t gfn = gpa >> PAGE_SHIFT;
1577 int offset = offset_in_page(gpa);
1580 while ((seg = next_segment(len, offset)) != 0) {
1581 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1592 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1593 gpa_t gpa, unsigned long len)
1595 struct kvm_memslots *slots = kvm_memslots(kvm);
1596 int offset = offset_in_page(gpa);
1597 gfn_t start_gfn = gpa >> PAGE_SHIFT;
1598 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1599 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1600 gfn_t nr_pages_avail;
1603 ghc->generation = slots->generation;
1605 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1606 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, &nr_pages_avail);
1607 if (!kvm_is_error_hva(ghc->hva) && nr_pages_avail >= nr_pages_needed) {
1611 * If the requested region crosses two memslots, we still
1612 * verify that the entire region is valid here.
1614 while (start_gfn <= end_gfn) {
1615 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1616 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1618 if (kvm_is_error_hva(ghc->hva))
1620 start_gfn += nr_pages_avail;
1622 /* Use the slow path for cross page reads and writes. */
1623 ghc->memslot = NULL;
1627 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1629 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1630 void *data, unsigned long len)
1632 struct kvm_memslots *slots = kvm_memslots(kvm);
1635 BUG_ON(len > ghc->len);
1637 if (slots->generation != ghc->generation)
1638 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1640 if (unlikely(!ghc->memslot))
1641 return kvm_write_guest(kvm, ghc->gpa, data, len);
1643 if (kvm_is_error_hva(ghc->hva))
1646 r = __copy_to_user((void __user *)ghc->hva, data, len);
1649 mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1653 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1655 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1656 void *data, unsigned long len)
1658 struct kvm_memslots *slots = kvm_memslots(kvm);
1661 BUG_ON(len > ghc->len);
1663 if (slots->generation != ghc->generation)
1664 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1666 if (unlikely(!ghc->memslot))
1667 return kvm_read_guest(kvm, ghc->gpa, data, len);
1669 if (kvm_is_error_hva(ghc->hva))
1672 r = __copy_from_user(data, (void __user *)ghc->hva, len);
1678 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1680 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1682 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
1684 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
1686 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1688 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1690 gfn_t gfn = gpa >> PAGE_SHIFT;
1692 int offset = offset_in_page(gpa);
1695 while ((seg = next_segment(len, offset)) != 0) {
1696 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1705 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1707 static void mark_page_dirty_in_slot(struct kvm *kvm,
1708 struct kvm_memory_slot *memslot,
1711 if (memslot && memslot->dirty_bitmap) {
1712 unsigned long rel_gfn = gfn - memslot->base_gfn;
1714 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1718 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1720 struct kvm_memory_slot *memslot;
1722 memslot = gfn_to_memslot(kvm, gfn);
1723 mark_page_dirty_in_slot(kvm, memslot, gfn);
1725 EXPORT_SYMBOL_GPL(mark_page_dirty);
1728 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1730 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1735 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1737 if (kvm_arch_vcpu_runnable(vcpu)) {
1738 kvm_make_request(KVM_REQ_UNHALT, vcpu);
1741 if (kvm_cpu_has_pending_timer(vcpu))
1743 if (signal_pending(current))
1749 finish_wait(&vcpu->wq, &wait);
1751 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
1755 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1757 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1760 int cpu = vcpu->cpu;
1761 wait_queue_head_t *wqp;
1763 wqp = kvm_arch_vcpu_wq(vcpu);
1764 if (waitqueue_active(wqp)) {
1765 wake_up_interruptible(wqp);
1766 ++vcpu->stat.halt_wakeup;
1770 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1771 if (kvm_arch_vcpu_should_kick(vcpu))
1772 smp_send_reschedule(cpu);
1775 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
1776 #endif /* !CONFIG_S390 */
1778 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
1781 struct task_struct *task = NULL;
1785 pid = rcu_dereference(target->pid);
1787 task = get_pid_task(target->pid, PIDTYPE_PID);
1791 if (task->flags & PF_VCPU) {
1792 put_task_struct(task);
1795 ret = yield_to(task, 1);
1796 put_task_struct(task);
1800 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1803 * Helper that checks whether a VCPU is eligible for directed yield.
1804 * Most eligible candidate to yield is decided by following heuristics:
1806 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1807 * (preempted lock holder), indicated by @in_spin_loop.
1808 * Set at the beiginning and cleared at the end of interception/PLE handler.
1810 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1811 * chance last time (mostly it has become eligible now since we have probably
1812 * yielded to lockholder in last iteration. This is done by toggling
1813 * @dy_eligible each time a VCPU checked for eligibility.)
1815 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1816 * to preempted lock-holder could result in wrong VCPU selection and CPU
1817 * burning. Giving priority for a potential lock-holder increases lock
1820 * Since algorithm is based on heuristics, accessing another VCPU data without
1821 * locking does not harm. It may result in trying to yield to same VCPU, fail
1822 * and continue with next VCPU and so on.
1824 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1826 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1829 eligible = !vcpu->spin_loop.in_spin_loop ||
1830 vcpu->spin_loop.dy_eligible;
1832 if (vcpu->spin_loop.in_spin_loop)
1833 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1841 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1843 struct kvm *kvm = me->kvm;
1844 struct kvm_vcpu *vcpu;
1845 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1851 kvm_vcpu_set_in_spin_loop(me, true);
1853 * We boost the priority of a VCPU that is runnable but not
1854 * currently running, because it got preempted by something
1855 * else and called schedule in __vcpu_run. Hopefully that
1856 * VCPU is holding the lock that we need and will release it.
1857 * We approximate round-robin by starting at the last boosted VCPU.
1859 for (pass = 0; pass < 2 && !yielded && try; pass++) {
1860 kvm_for_each_vcpu(i, vcpu, kvm) {
1861 if (!pass && i <= last_boosted_vcpu) {
1862 i = last_boosted_vcpu;
1864 } else if (pass && i > last_boosted_vcpu)
1866 if (!ACCESS_ONCE(vcpu->preempted))
1870 if (waitqueue_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
1872 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1875 yielded = kvm_vcpu_yield_to(vcpu);
1877 kvm->last_boosted_vcpu = i;
1879 } else if (yielded < 0) {
1886 kvm_vcpu_set_in_spin_loop(me, false);
1888 /* Ensure vcpu is not eligible during next spinloop */
1889 kvm_vcpu_set_dy_eligible(me, false);
1891 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1893 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1895 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1898 if (vmf->pgoff == 0)
1899 page = virt_to_page(vcpu->run);
1901 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1902 page = virt_to_page(vcpu->arch.pio_data);
1904 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1905 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1906 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1909 return kvm_arch_vcpu_fault(vcpu, vmf);
1915 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1916 .fault = kvm_vcpu_fault,
1919 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1921 vma->vm_ops = &kvm_vcpu_vm_ops;
1925 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1927 struct kvm_vcpu *vcpu = filp->private_data;
1929 kvm_put_kvm(vcpu->kvm);
1933 static struct file_operations kvm_vcpu_fops = {
1934 .release = kvm_vcpu_release,
1935 .unlocked_ioctl = kvm_vcpu_ioctl,
1936 #ifdef CONFIG_COMPAT
1937 .compat_ioctl = kvm_vcpu_compat_ioctl,
1939 .mmap = kvm_vcpu_mmap,
1940 .llseek = noop_llseek,
1944 * Allocates an inode for the vcpu.
1946 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1948 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
1952 * Creates some virtual cpus. Good luck creating more than one.
1954 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1957 struct kvm_vcpu *vcpu, *v;
1959 if (id >= KVM_MAX_VCPUS)
1962 vcpu = kvm_arch_vcpu_create(kvm, id);
1964 return PTR_ERR(vcpu);
1966 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1968 r = kvm_arch_vcpu_setup(vcpu);
1972 mutex_lock(&kvm->lock);
1973 if (!kvm_vcpu_compatible(vcpu)) {
1975 goto unlock_vcpu_destroy;
1977 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1979 goto unlock_vcpu_destroy;
1982 kvm_for_each_vcpu(r, v, kvm)
1983 if (v->vcpu_id == id) {
1985 goto unlock_vcpu_destroy;
1988 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1990 /* Now it's all set up, let userspace reach it */
1992 r = create_vcpu_fd(vcpu);
1995 goto unlock_vcpu_destroy;
1998 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
2000 atomic_inc(&kvm->online_vcpus);
2002 mutex_unlock(&kvm->lock);
2003 kvm_arch_vcpu_postcreate(vcpu);
2006 unlock_vcpu_destroy:
2007 mutex_unlock(&kvm->lock);
2009 kvm_arch_vcpu_destroy(vcpu);
2013 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2016 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2017 vcpu->sigset_active = 1;
2018 vcpu->sigset = *sigset;
2020 vcpu->sigset_active = 0;
2024 static long kvm_vcpu_ioctl(struct file *filp,
2025 unsigned int ioctl, unsigned long arg)
2027 struct kvm_vcpu *vcpu = filp->private_data;
2028 void __user *argp = (void __user *)arg;
2030 struct kvm_fpu *fpu = NULL;
2031 struct kvm_sregs *kvm_sregs = NULL;
2033 if (vcpu->kvm->mm != current->mm)
2036 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2039 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
2041 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
2042 * so vcpu_load() would break it.
2044 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
2045 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2049 r = vcpu_load(vcpu);
2057 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2058 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2060 case KVM_GET_REGS: {
2061 struct kvm_regs *kvm_regs;
2064 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2067 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2071 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2078 case KVM_SET_REGS: {
2079 struct kvm_regs *kvm_regs;
2082 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2083 if (IS_ERR(kvm_regs)) {
2084 r = PTR_ERR(kvm_regs);
2087 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2091 case KVM_GET_SREGS: {
2092 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2096 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2100 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2105 case KVM_SET_SREGS: {
2106 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2107 if (IS_ERR(kvm_sregs)) {
2108 r = PTR_ERR(kvm_sregs);
2112 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2115 case KVM_GET_MP_STATE: {
2116 struct kvm_mp_state mp_state;
2118 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2122 if (copy_to_user(argp, &mp_state, sizeof mp_state))
2127 case KVM_SET_MP_STATE: {
2128 struct kvm_mp_state mp_state;
2131 if (copy_from_user(&mp_state, argp, sizeof mp_state))
2133 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2136 case KVM_TRANSLATE: {
2137 struct kvm_translation tr;
2140 if (copy_from_user(&tr, argp, sizeof tr))
2142 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2146 if (copy_to_user(argp, &tr, sizeof tr))
2151 case KVM_SET_GUEST_DEBUG: {
2152 struct kvm_guest_debug dbg;
2155 if (copy_from_user(&dbg, argp, sizeof dbg))
2157 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2160 case KVM_SET_SIGNAL_MASK: {
2161 struct kvm_signal_mask __user *sigmask_arg = argp;
2162 struct kvm_signal_mask kvm_sigmask;
2163 sigset_t sigset, *p;
2168 if (copy_from_user(&kvm_sigmask, argp,
2169 sizeof kvm_sigmask))
2172 if (kvm_sigmask.len != sizeof sigset)
2175 if (copy_from_user(&sigset, sigmask_arg->sigset,
2180 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2184 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2188 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2192 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2198 fpu = memdup_user(argp, sizeof(*fpu));
2204 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2208 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2217 #ifdef CONFIG_COMPAT
2218 static long kvm_vcpu_compat_ioctl(struct file *filp,
2219 unsigned int ioctl, unsigned long arg)
2221 struct kvm_vcpu *vcpu = filp->private_data;
2222 void __user *argp = compat_ptr(arg);
2225 if (vcpu->kvm->mm != current->mm)
2229 case KVM_SET_SIGNAL_MASK: {
2230 struct kvm_signal_mask __user *sigmask_arg = argp;
2231 struct kvm_signal_mask kvm_sigmask;
2232 compat_sigset_t csigset;
2237 if (copy_from_user(&kvm_sigmask, argp,
2238 sizeof kvm_sigmask))
2241 if (kvm_sigmask.len != sizeof csigset)
2244 if (copy_from_user(&csigset, sigmask_arg->sigset,
2247 sigset_from_compat(&sigset, &csigset);
2248 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2250 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2254 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2262 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2263 int (*accessor)(struct kvm_device *dev,
2264 struct kvm_device_attr *attr),
2267 struct kvm_device_attr attr;
2272 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2275 return accessor(dev, &attr);
2278 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2281 struct kvm_device *dev = filp->private_data;
2284 case KVM_SET_DEVICE_ATTR:
2285 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2286 case KVM_GET_DEVICE_ATTR:
2287 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2288 case KVM_HAS_DEVICE_ATTR:
2289 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2291 if (dev->ops->ioctl)
2292 return dev->ops->ioctl(dev, ioctl, arg);
2298 static int kvm_device_release(struct inode *inode, struct file *filp)
2300 struct kvm_device *dev = filp->private_data;
2301 struct kvm *kvm = dev->kvm;
2307 static const struct file_operations kvm_device_fops = {
2308 .unlocked_ioctl = kvm_device_ioctl,
2309 #ifdef CONFIG_COMPAT
2310 .compat_ioctl = kvm_device_ioctl,
2312 .release = kvm_device_release,
2315 struct kvm_device *kvm_device_from_filp(struct file *filp)
2317 if (filp->f_op != &kvm_device_fops)
2320 return filp->private_data;
2323 static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
2324 #ifdef CONFIG_KVM_MPIC
2325 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
2326 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
2329 #ifdef CONFIG_KVM_XICS
2330 [KVM_DEV_TYPE_XICS] = &kvm_xics_ops,
2334 int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
2336 if (type >= ARRAY_SIZE(kvm_device_ops_table))
2339 if (kvm_device_ops_table[type] != NULL)
2342 kvm_device_ops_table[type] = ops;
2346 static int kvm_ioctl_create_device(struct kvm *kvm,
2347 struct kvm_create_device *cd)
2349 struct kvm_device_ops *ops = NULL;
2350 struct kvm_device *dev;
2351 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2354 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
2357 ops = kvm_device_ops_table[cd->type];
2364 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2371 ret = ops->create(dev, cd->type);
2377 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2383 list_add(&dev->vm_node, &kvm->devices);
2389 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2392 case KVM_CAP_USER_MEMORY:
2393 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2394 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2395 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2396 case KVM_CAP_SET_BOOT_CPU_ID:
2398 case KVM_CAP_INTERNAL_ERROR_DATA:
2399 #ifdef CONFIG_HAVE_KVM_MSI
2400 case KVM_CAP_SIGNAL_MSI:
2402 #ifdef CONFIG_HAVE_KVM_IRQFD
2403 case KVM_CAP_IRQFD_RESAMPLE:
2405 case KVM_CAP_CHECK_EXTENSION_VM:
2407 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2408 case KVM_CAP_IRQ_ROUTING:
2409 return KVM_MAX_IRQ_ROUTES;
2414 return kvm_vm_ioctl_check_extension(kvm, arg);
2417 static long kvm_vm_ioctl(struct file *filp,
2418 unsigned int ioctl, unsigned long arg)
2420 struct kvm *kvm = filp->private_data;
2421 void __user *argp = (void __user *)arg;
2424 if (kvm->mm != current->mm)
2427 case KVM_CREATE_VCPU:
2428 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2430 case KVM_SET_USER_MEMORY_REGION: {
2431 struct kvm_userspace_memory_region kvm_userspace_mem;
2434 if (copy_from_user(&kvm_userspace_mem, argp,
2435 sizeof kvm_userspace_mem))
2438 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2441 case KVM_GET_DIRTY_LOG: {
2442 struct kvm_dirty_log log;
2445 if (copy_from_user(&log, argp, sizeof log))
2447 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2450 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2451 case KVM_REGISTER_COALESCED_MMIO: {
2452 struct kvm_coalesced_mmio_zone zone;
2454 if (copy_from_user(&zone, argp, sizeof zone))
2456 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2459 case KVM_UNREGISTER_COALESCED_MMIO: {
2460 struct kvm_coalesced_mmio_zone zone;
2462 if (copy_from_user(&zone, argp, sizeof zone))
2464 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2469 struct kvm_irqfd data;
2472 if (copy_from_user(&data, argp, sizeof data))
2474 r = kvm_irqfd(kvm, &data);
2477 case KVM_IOEVENTFD: {
2478 struct kvm_ioeventfd data;
2481 if (copy_from_user(&data, argp, sizeof data))
2483 r = kvm_ioeventfd(kvm, &data);
2486 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2487 case KVM_SET_BOOT_CPU_ID:
2489 mutex_lock(&kvm->lock);
2490 if (atomic_read(&kvm->online_vcpus) != 0)
2493 kvm->bsp_vcpu_id = arg;
2494 mutex_unlock(&kvm->lock);
2497 #ifdef CONFIG_HAVE_KVM_MSI
2498 case KVM_SIGNAL_MSI: {
2502 if (copy_from_user(&msi, argp, sizeof msi))
2504 r = kvm_send_userspace_msi(kvm, &msi);
2508 #ifdef __KVM_HAVE_IRQ_LINE
2509 case KVM_IRQ_LINE_STATUS:
2510 case KVM_IRQ_LINE: {
2511 struct kvm_irq_level irq_event;
2514 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2517 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
2518 ioctl == KVM_IRQ_LINE_STATUS);
2523 if (ioctl == KVM_IRQ_LINE_STATUS) {
2524 if (copy_to_user(argp, &irq_event, sizeof irq_event))
2532 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2533 case KVM_SET_GSI_ROUTING: {
2534 struct kvm_irq_routing routing;
2535 struct kvm_irq_routing __user *urouting;
2536 struct kvm_irq_routing_entry *entries;
2539 if (copy_from_user(&routing, argp, sizeof(routing)))
2542 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2547 entries = vmalloc(routing.nr * sizeof(*entries));
2552 if (copy_from_user(entries, urouting->entries,
2553 routing.nr * sizeof(*entries)))
2554 goto out_free_irq_routing;
2555 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2557 out_free_irq_routing:
2561 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
2562 case KVM_CREATE_DEVICE: {
2563 struct kvm_create_device cd;
2566 if (copy_from_user(&cd, argp, sizeof(cd)))
2569 r = kvm_ioctl_create_device(kvm, &cd);
2574 if (copy_to_user(argp, &cd, sizeof(cd)))
2580 case KVM_CHECK_EXTENSION:
2581 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
2584 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2586 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2592 #ifdef CONFIG_COMPAT
2593 struct compat_kvm_dirty_log {
2597 compat_uptr_t dirty_bitmap; /* one bit per page */
2602 static long kvm_vm_compat_ioctl(struct file *filp,
2603 unsigned int ioctl, unsigned long arg)
2605 struct kvm *kvm = filp->private_data;
2608 if (kvm->mm != current->mm)
2611 case KVM_GET_DIRTY_LOG: {
2612 struct compat_kvm_dirty_log compat_log;
2613 struct kvm_dirty_log log;
2616 if (copy_from_user(&compat_log, (void __user *)arg,
2617 sizeof(compat_log)))
2619 log.slot = compat_log.slot;
2620 log.padding1 = compat_log.padding1;
2621 log.padding2 = compat_log.padding2;
2622 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2624 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2628 r = kvm_vm_ioctl(filp, ioctl, arg);
2636 static struct file_operations kvm_vm_fops = {
2637 .release = kvm_vm_release,
2638 .unlocked_ioctl = kvm_vm_ioctl,
2639 #ifdef CONFIG_COMPAT
2640 .compat_ioctl = kvm_vm_compat_ioctl,
2642 .llseek = noop_llseek,
2645 static int kvm_dev_ioctl_create_vm(unsigned long type)
2650 kvm = kvm_create_vm(type);
2652 return PTR_ERR(kvm);
2653 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2654 r = kvm_coalesced_mmio_init(kvm);
2660 r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC);
2667 static long kvm_dev_ioctl(struct file *filp,
2668 unsigned int ioctl, unsigned long arg)
2673 case KVM_GET_API_VERSION:
2676 r = KVM_API_VERSION;
2679 r = kvm_dev_ioctl_create_vm(arg);
2681 case KVM_CHECK_EXTENSION:
2682 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
2684 case KVM_GET_VCPU_MMAP_SIZE:
2687 r = PAGE_SIZE; /* struct kvm_run */
2689 r += PAGE_SIZE; /* pio data page */
2691 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2692 r += PAGE_SIZE; /* coalesced mmio ring page */
2695 case KVM_TRACE_ENABLE:
2696 case KVM_TRACE_PAUSE:
2697 case KVM_TRACE_DISABLE:
2701 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2707 static struct file_operations kvm_chardev_ops = {
2708 .unlocked_ioctl = kvm_dev_ioctl,
2709 .compat_ioctl = kvm_dev_ioctl,
2710 .llseek = noop_llseek,
2713 static struct miscdevice kvm_dev = {
2719 static void hardware_enable_nolock(void *junk)
2721 int cpu = raw_smp_processor_id();
2724 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2727 cpumask_set_cpu(cpu, cpus_hardware_enabled);
2729 r = kvm_arch_hardware_enable();
2732 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2733 atomic_inc(&hardware_enable_failed);
2734 printk(KERN_INFO "kvm: enabling virtualization on "
2735 "CPU%d failed\n", cpu);
2739 static void hardware_enable(void)
2741 raw_spin_lock(&kvm_count_lock);
2742 if (kvm_usage_count)
2743 hardware_enable_nolock(NULL);
2744 raw_spin_unlock(&kvm_count_lock);
2747 static void hardware_disable_nolock(void *junk)
2749 int cpu = raw_smp_processor_id();
2751 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2753 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2754 kvm_arch_hardware_disable();
2757 static void hardware_disable(void)
2759 raw_spin_lock(&kvm_count_lock);
2760 if (kvm_usage_count)
2761 hardware_disable_nolock(NULL);
2762 raw_spin_unlock(&kvm_count_lock);
2765 static void hardware_disable_all_nolock(void)
2767 BUG_ON(!kvm_usage_count);
2770 if (!kvm_usage_count)
2771 on_each_cpu(hardware_disable_nolock, NULL, 1);
2774 static void hardware_disable_all(void)
2776 raw_spin_lock(&kvm_count_lock);
2777 hardware_disable_all_nolock();
2778 raw_spin_unlock(&kvm_count_lock);
2781 static int hardware_enable_all(void)
2785 raw_spin_lock(&kvm_count_lock);
2788 if (kvm_usage_count == 1) {
2789 atomic_set(&hardware_enable_failed, 0);
2790 on_each_cpu(hardware_enable_nolock, NULL, 1);
2792 if (atomic_read(&hardware_enable_failed)) {
2793 hardware_disable_all_nolock();
2798 raw_spin_unlock(&kvm_count_lock);
2803 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2808 val &= ~CPU_TASKS_FROZEN;
2811 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2816 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2824 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2828 * Some (well, at least mine) BIOSes hang on reboot if
2831 * And Intel TXT required VMX off for all cpu when system shutdown.
2833 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2834 kvm_rebooting = true;
2835 on_each_cpu(hardware_disable_nolock, NULL, 1);
2839 static struct notifier_block kvm_reboot_notifier = {
2840 .notifier_call = kvm_reboot,
2844 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2848 for (i = 0; i < bus->dev_count; i++) {
2849 struct kvm_io_device *pos = bus->range[i].dev;
2851 kvm_iodevice_destructor(pos);
2856 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
2857 const struct kvm_io_range *r2)
2859 if (r1->addr < r2->addr)
2861 if (r1->addr + r1->len > r2->addr + r2->len)
2866 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2868 return kvm_io_bus_cmp(p1, p2);
2871 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2872 gpa_t addr, int len)
2874 bus->range[bus->dev_count++] = (struct kvm_io_range) {
2880 sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2881 kvm_io_bus_sort_cmp, NULL);
2886 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2887 gpa_t addr, int len)
2889 struct kvm_io_range *range, key;
2892 key = (struct kvm_io_range) {
2897 range = bsearch(&key, bus->range, bus->dev_count,
2898 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2902 off = range - bus->range;
2904 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
2910 static int __kvm_io_bus_write(struct kvm_io_bus *bus,
2911 struct kvm_io_range *range, const void *val)
2915 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2919 while (idx < bus->dev_count &&
2920 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
2921 if (!kvm_iodevice_write(bus->range[idx].dev, range->addr,
2930 /* kvm_io_bus_write - called under kvm->slots_lock */
2931 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2932 int len, const void *val)
2934 struct kvm_io_bus *bus;
2935 struct kvm_io_range range;
2938 range = (struct kvm_io_range) {
2943 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2944 r = __kvm_io_bus_write(bus, &range, val);
2945 return r < 0 ? r : 0;
2948 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
2949 int kvm_io_bus_write_cookie(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2950 int len, const void *val, long cookie)
2952 struct kvm_io_bus *bus;
2953 struct kvm_io_range range;
2955 range = (struct kvm_io_range) {
2960 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2962 /* First try the device referenced by cookie. */
2963 if ((cookie >= 0) && (cookie < bus->dev_count) &&
2964 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
2965 if (!kvm_iodevice_write(bus->range[cookie].dev, addr, len,
2970 * cookie contained garbage; fall back to search and return the
2971 * correct cookie value.
2973 return __kvm_io_bus_write(bus, &range, val);
2976 static int __kvm_io_bus_read(struct kvm_io_bus *bus, struct kvm_io_range *range,
2981 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2985 while (idx < bus->dev_count &&
2986 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
2987 if (!kvm_iodevice_read(bus->range[idx].dev, range->addr,
2995 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
2997 /* kvm_io_bus_read - called under kvm->slots_lock */
2998 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3001 struct kvm_io_bus *bus;
3002 struct kvm_io_range range;
3005 range = (struct kvm_io_range) {
3010 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
3011 r = __kvm_io_bus_read(bus, &range, val);
3012 return r < 0 ? r : 0;
3016 /* Caller must hold slots_lock. */
3017 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3018 int len, struct kvm_io_device *dev)
3020 struct kvm_io_bus *new_bus, *bus;
3022 bus = kvm->buses[bus_idx];
3023 /* exclude ioeventfd which is limited by maximum fd */
3024 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3027 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
3028 sizeof(struct kvm_io_range)), GFP_KERNEL);
3031 memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
3032 sizeof(struct kvm_io_range)));
3033 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
3034 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3035 synchronize_srcu_expedited(&kvm->srcu);
3041 /* Caller must hold slots_lock. */
3042 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3043 struct kvm_io_device *dev)
3046 struct kvm_io_bus *new_bus, *bus;
3048 bus = kvm->buses[bus_idx];
3050 for (i = 0; i < bus->dev_count; i++)
3051 if (bus->range[i].dev == dev) {
3059 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3060 sizeof(struct kvm_io_range)), GFP_KERNEL);
3064 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3065 new_bus->dev_count--;
3066 memcpy(new_bus->range + i, bus->range + i + 1,
3067 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3069 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3070 synchronize_srcu_expedited(&kvm->srcu);
3075 static struct notifier_block kvm_cpu_notifier = {
3076 .notifier_call = kvm_cpu_hotplug,
3079 static int vm_stat_get(void *_offset, u64 *val)
3081 unsigned offset = (long)_offset;
3085 spin_lock(&kvm_lock);
3086 list_for_each_entry(kvm, &vm_list, vm_list)
3087 *val += *(u32 *)((void *)kvm + offset);
3088 spin_unlock(&kvm_lock);
3092 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
3094 static int vcpu_stat_get(void *_offset, u64 *val)
3096 unsigned offset = (long)_offset;
3098 struct kvm_vcpu *vcpu;
3102 spin_lock(&kvm_lock);
3103 list_for_each_entry(kvm, &vm_list, vm_list)
3104 kvm_for_each_vcpu(i, vcpu, kvm)
3105 *val += *(u32 *)((void *)vcpu + offset);
3107 spin_unlock(&kvm_lock);
3111 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
3113 static const struct file_operations *stat_fops[] = {
3114 [KVM_STAT_VCPU] = &vcpu_stat_fops,
3115 [KVM_STAT_VM] = &vm_stat_fops,
3118 static int kvm_init_debug(void)
3121 struct kvm_stats_debugfs_item *p;
3123 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3124 if (kvm_debugfs_dir == NULL)
3127 for (p = debugfs_entries; p->name; ++p) {
3128 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
3129 (void *)(long)p->offset,
3130 stat_fops[p->kind]);
3131 if (p->dentry == NULL)
3138 debugfs_remove_recursive(kvm_debugfs_dir);
3143 static void kvm_exit_debug(void)
3145 struct kvm_stats_debugfs_item *p;
3147 for (p = debugfs_entries; p->name; ++p)
3148 debugfs_remove(p->dentry);
3149 debugfs_remove(kvm_debugfs_dir);
3152 static int kvm_suspend(void)
3154 if (kvm_usage_count)
3155 hardware_disable_nolock(NULL);
3159 static void kvm_resume(void)
3161 if (kvm_usage_count) {
3162 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3163 hardware_enable_nolock(NULL);
3167 static struct syscore_ops kvm_syscore_ops = {
3168 .suspend = kvm_suspend,
3169 .resume = kvm_resume,
3173 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3175 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3178 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3180 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3181 if (vcpu->preempted)
3182 vcpu->preempted = false;
3184 kvm_arch_sched_in(vcpu, cpu);
3186 kvm_arch_vcpu_load(vcpu, cpu);
3189 static void kvm_sched_out(struct preempt_notifier *pn,
3190 struct task_struct *next)
3192 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3194 if (current->state == TASK_RUNNING)
3195 vcpu->preempted = true;
3196 kvm_arch_vcpu_put(vcpu);
3199 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3200 struct module *module)
3205 r = kvm_arch_init(opaque);
3210 * kvm_arch_init makes sure there's at most one caller
3211 * for architectures that support multiple implementations,
3212 * like intel and amd on x86.
3213 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3214 * conflicts in case kvm is already setup for another implementation.
3216 r = kvm_irqfd_init();
3220 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
3225 r = kvm_arch_hardware_setup();
3229 for_each_online_cpu(cpu) {
3230 smp_call_function_single(cpu,
3231 kvm_arch_check_processor_compat,
3237 r = register_cpu_notifier(&kvm_cpu_notifier);
3240 register_reboot_notifier(&kvm_reboot_notifier);
3242 /* A kmem cache lets us meet the alignment requirements of fx_save. */
3244 vcpu_align = __alignof__(struct kvm_vcpu);
3245 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
3247 if (!kvm_vcpu_cache) {
3252 r = kvm_async_pf_init();
3256 kvm_chardev_ops.owner = module;
3257 kvm_vm_fops.owner = module;
3258 kvm_vcpu_fops.owner = module;
3260 r = misc_register(&kvm_dev);
3262 printk(KERN_ERR "kvm: misc device register failed\n");
3266 register_syscore_ops(&kvm_syscore_ops);
3268 kvm_preempt_ops.sched_in = kvm_sched_in;
3269 kvm_preempt_ops.sched_out = kvm_sched_out;
3271 r = kvm_init_debug();
3273 printk(KERN_ERR "kvm: create debugfs files failed\n");
3277 r = kvm_vfio_ops_init();
3283 unregister_syscore_ops(&kvm_syscore_ops);
3284 misc_deregister(&kvm_dev);
3286 kvm_async_pf_deinit();
3288 kmem_cache_destroy(kvm_vcpu_cache);
3290 unregister_reboot_notifier(&kvm_reboot_notifier);
3291 unregister_cpu_notifier(&kvm_cpu_notifier);
3294 kvm_arch_hardware_unsetup();
3296 free_cpumask_var(cpus_hardware_enabled);
3304 EXPORT_SYMBOL_GPL(kvm_init);
3309 misc_deregister(&kvm_dev);
3310 kmem_cache_destroy(kvm_vcpu_cache);
3311 kvm_async_pf_deinit();
3312 unregister_syscore_ops(&kvm_syscore_ops);
3313 unregister_reboot_notifier(&kvm_reboot_notifier);
3314 unregister_cpu_notifier(&kvm_cpu_notifier);
3315 on_each_cpu(hardware_disable_nolock, NULL, 1);
3316 kvm_arch_hardware_unsetup();
3319 free_cpumask_var(cpus_hardware_enabled);
3321 EXPORT_SYMBOL_GPL(kvm_exit);