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 !is_zero_pfn(pfn) && 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 bool kvm_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 (kvm_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 kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
204 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
206 kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
209 void kvm_make_scan_ioapic_request(struct kvm *kvm)
211 kvm_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);
300 kvm_arch_mmu_notifier_invalidate_page(kvm, address);
302 srcu_read_unlock(&kvm->srcu, idx);
305 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
306 struct mm_struct *mm,
307 unsigned long address,
310 struct kvm *kvm = mmu_notifier_to_kvm(mn);
313 idx = srcu_read_lock(&kvm->srcu);
314 spin_lock(&kvm->mmu_lock);
315 kvm->mmu_notifier_seq++;
316 kvm_set_spte_hva(kvm, address, pte);
317 spin_unlock(&kvm->mmu_lock);
318 srcu_read_unlock(&kvm->srcu, idx);
321 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
322 struct mm_struct *mm,
326 struct kvm *kvm = mmu_notifier_to_kvm(mn);
327 int need_tlb_flush = 0, idx;
329 idx = srcu_read_lock(&kvm->srcu);
330 spin_lock(&kvm->mmu_lock);
332 * The count increase must become visible at unlock time as no
333 * spte can be established without taking the mmu_lock and
334 * count is also read inside the mmu_lock critical section.
336 kvm->mmu_notifier_count++;
337 need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
338 need_tlb_flush |= kvm->tlbs_dirty;
339 /* we've to flush the tlb before the pages can be freed */
341 kvm_flush_remote_tlbs(kvm);
343 spin_unlock(&kvm->mmu_lock);
344 srcu_read_unlock(&kvm->srcu, idx);
347 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
348 struct mm_struct *mm,
352 struct kvm *kvm = mmu_notifier_to_kvm(mn);
354 spin_lock(&kvm->mmu_lock);
356 * This sequence increase will notify the kvm page fault that
357 * the page that is going to be mapped in the spte could have
360 kvm->mmu_notifier_seq++;
363 * The above sequence increase must be visible before the
364 * below count decrease, which is ensured by the smp_wmb above
365 * in conjunction with the smp_rmb in mmu_notifier_retry().
367 kvm->mmu_notifier_count--;
368 spin_unlock(&kvm->mmu_lock);
370 BUG_ON(kvm->mmu_notifier_count < 0);
373 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
374 struct mm_struct *mm,
378 struct kvm *kvm = mmu_notifier_to_kvm(mn);
381 idx = srcu_read_lock(&kvm->srcu);
382 spin_lock(&kvm->mmu_lock);
384 young = kvm_age_hva(kvm, start, end);
386 kvm_flush_remote_tlbs(kvm);
388 spin_unlock(&kvm->mmu_lock);
389 srcu_read_unlock(&kvm->srcu, idx);
394 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
395 struct mm_struct *mm,
396 unsigned long address)
398 struct kvm *kvm = mmu_notifier_to_kvm(mn);
401 idx = srcu_read_lock(&kvm->srcu);
402 spin_lock(&kvm->mmu_lock);
403 young = kvm_test_age_hva(kvm, address);
404 spin_unlock(&kvm->mmu_lock);
405 srcu_read_unlock(&kvm->srcu, idx);
410 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
411 struct mm_struct *mm)
413 struct kvm *kvm = mmu_notifier_to_kvm(mn);
416 idx = srcu_read_lock(&kvm->srcu);
417 kvm_arch_flush_shadow_all(kvm);
418 srcu_read_unlock(&kvm->srcu, idx);
421 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
422 .invalidate_page = kvm_mmu_notifier_invalidate_page,
423 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
424 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
425 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
426 .test_young = kvm_mmu_notifier_test_young,
427 .change_pte = kvm_mmu_notifier_change_pte,
428 .release = kvm_mmu_notifier_release,
431 static int kvm_init_mmu_notifier(struct kvm *kvm)
433 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
434 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
437 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
439 static int kvm_init_mmu_notifier(struct kvm *kvm)
444 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
446 static void kvm_init_memslots_id(struct kvm *kvm)
449 struct kvm_memslots *slots = kvm->memslots;
451 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
452 slots->id_to_index[i] = slots->memslots[i].id = i;
455 static struct kvm *kvm_create_vm(unsigned long type)
458 struct kvm *kvm = kvm_arch_alloc_vm();
461 return ERR_PTR(-ENOMEM);
463 r = kvm_arch_init_vm(kvm, type);
465 goto out_err_no_disable;
467 r = hardware_enable_all();
469 goto out_err_no_disable;
471 #ifdef CONFIG_HAVE_KVM_IRQFD
472 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
475 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
478 kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
480 goto out_err_no_srcu;
483 * Init kvm generation close to the maximum to easily test the
484 * code of handling generation number wrap-around.
486 kvm->memslots->generation = -150;
488 kvm_init_memslots_id(kvm);
489 if (init_srcu_struct(&kvm->srcu))
490 goto out_err_no_srcu;
491 if (init_srcu_struct(&kvm->irq_srcu))
492 goto out_err_no_irq_srcu;
493 for (i = 0; i < KVM_NR_BUSES; i++) {
494 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
500 spin_lock_init(&kvm->mmu_lock);
501 kvm->mm = current->mm;
502 atomic_inc(&kvm->mm->mm_count);
503 kvm_eventfd_init(kvm);
504 mutex_init(&kvm->lock);
505 mutex_init(&kvm->irq_lock);
506 mutex_init(&kvm->slots_lock);
507 atomic_set(&kvm->users_count, 1);
508 INIT_LIST_HEAD(&kvm->devices);
510 r = kvm_init_mmu_notifier(kvm);
514 spin_lock(&kvm_lock);
515 list_add(&kvm->vm_list, &vm_list);
516 spin_unlock(&kvm_lock);
521 cleanup_srcu_struct(&kvm->irq_srcu);
523 cleanup_srcu_struct(&kvm->srcu);
525 hardware_disable_all();
527 for (i = 0; i < KVM_NR_BUSES; i++)
528 kfree(kvm->buses[i]);
529 kfree(kvm->memslots);
530 kvm_arch_free_vm(kvm);
535 * Avoid using vmalloc for a small buffer.
536 * Should not be used when the size is statically known.
538 void *kvm_kvzalloc(unsigned long size)
540 if (size > PAGE_SIZE)
541 return vzalloc(size);
543 return kzalloc(size, GFP_KERNEL);
546 void kvm_kvfree(const void *addr)
548 if (is_vmalloc_addr(addr))
554 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
556 if (!memslot->dirty_bitmap)
559 kvm_kvfree(memslot->dirty_bitmap);
560 memslot->dirty_bitmap = NULL;
564 * Free any memory in @free but not in @dont.
566 static void kvm_free_physmem_slot(struct kvm *kvm, struct kvm_memory_slot *free,
567 struct kvm_memory_slot *dont)
569 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
570 kvm_destroy_dirty_bitmap(free);
572 kvm_arch_free_memslot(kvm, free, dont);
577 static void kvm_free_physmem(struct kvm *kvm)
579 struct kvm_memslots *slots = kvm->memslots;
580 struct kvm_memory_slot *memslot;
582 kvm_for_each_memslot(memslot, slots)
583 kvm_free_physmem_slot(kvm, memslot, NULL);
585 kfree(kvm->memslots);
588 static void kvm_destroy_devices(struct kvm *kvm)
590 struct list_head *node, *tmp;
592 list_for_each_safe(node, tmp, &kvm->devices) {
593 struct kvm_device *dev =
594 list_entry(node, struct kvm_device, vm_node);
597 dev->ops->destroy(dev);
601 static void kvm_destroy_vm(struct kvm *kvm)
604 struct mm_struct *mm = kvm->mm;
606 kvm_arch_sync_events(kvm);
607 spin_lock(&kvm_lock);
608 list_del(&kvm->vm_list);
609 spin_unlock(&kvm_lock);
610 kvm_free_irq_routing(kvm);
611 for (i = 0; i < KVM_NR_BUSES; i++)
612 kvm_io_bus_destroy(kvm->buses[i]);
613 kvm_coalesced_mmio_free(kvm);
614 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
615 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
617 kvm_arch_flush_shadow_all(kvm);
619 kvm_arch_destroy_vm(kvm);
620 kvm_destroy_devices(kvm);
621 kvm_free_physmem(kvm);
622 cleanup_srcu_struct(&kvm->irq_srcu);
623 cleanup_srcu_struct(&kvm->srcu);
624 kvm_arch_free_vm(kvm);
625 hardware_disable_all();
629 void kvm_get_kvm(struct kvm *kvm)
631 atomic_inc(&kvm->users_count);
633 EXPORT_SYMBOL_GPL(kvm_get_kvm);
635 void kvm_put_kvm(struct kvm *kvm)
637 if (atomic_dec_and_test(&kvm->users_count))
640 EXPORT_SYMBOL_GPL(kvm_put_kvm);
643 static int kvm_vm_release(struct inode *inode, struct file *filp)
645 struct kvm *kvm = filp->private_data;
647 kvm_irqfd_release(kvm);
654 * Allocation size is twice as large as the actual dirty bitmap size.
655 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
657 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
659 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
661 memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
662 if (!memslot->dirty_bitmap)
669 * Insert memslot and re-sort memslots based on their size,
670 * so the larger slots will get better fit. Sorting algorithm
671 * takes advantage of having initially sorted array and
672 * known changed memslot position.
674 static void update_memslots(struct kvm_memslots *slots,
675 struct kvm_memory_slot *new)
678 int i = slots->id_to_index[id];
679 struct kvm_memory_slot *mslots = slots->memslots;
681 WARN_ON(mslots[i].id != id);
682 if (new->npages != mslots[i].npages) {
683 if (new->npages < mslots[i].npages) {
684 while (i < KVM_MEM_SLOTS_NUM - 1 &&
685 new->npages < mslots[i + 1].npages) {
686 mslots[i] = mslots[i + 1];
687 slots->id_to_index[mslots[i].id] = i;
692 new->npages > mslots[i - 1].npages) {
693 mslots[i] = mslots[i - 1];
694 slots->id_to_index[mslots[i].id] = i;
701 slots->id_to_index[mslots[i].id] = i;
704 static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
706 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
708 #ifdef __KVM_HAVE_READONLY_MEM
709 valid_flags |= KVM_MEM_READONLY;
712 if (mem->flags & ~valid_flags)
718 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
719 struct kvm_memslots *slots)
721 struct kvm_memslots *old_memslots = kvm->memslots;
724 * Set the low bit in the generation, which disables SPTE caching
725 * until the end of synchronize_srcu_expedited.
727 WARN_ON(old_memslots->generation & 1);
728 slots->generation = old_memslots->generation + 1;
730 rcu_assign_pointer(kvm->memslots, slots);
731 synchronize_srcu_expedited(&kvm->srcu);
734 * Increment the new memslot generation a second time. This prevents
735 * vm exits that race with memslot updates from caching a memslot
736 * generation that will (potentially) be valid forever.
740 kvm_arch_memslots_updated(kvm);
746 * Allocate some memory and give it an address in the guest physical address
749 * Discontiguous memory is allowed, mostly for framebuffers.
751 * Must be called holding kvm->slots_lock for write.
753 int __kvm_set_memory_region(struct kvm *kvm,
754 struct kvm_userspace_memory_region *mem)
758 unsigned long npages;
759 struct kvm_memory_slot *slot;
760 struct kvm_memory_slot old, new;
761 struct kvm_memslots *slots = NULL, *old_memslots;
762 enum kvm_mr_change change;
764 r = check_memory_region_flags(mem);
769 /* General sanity checks */
770 if (mem->memory_size & (PAGE_SIZE - 1))
772 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
774 /* We can read the guest memory with __xxx_user() later on. */
775 if ((mem->slot < KVM_USER_MEM_SLOTS) &&
776 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
777 !access_ok(VERIFY_WRITE,
778 (void __user *)(unsigned long)mem->userspace_addr,
781 if (mem->slot >= KVM_MEM_SLOTS_NUM)
783 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
786 slot = id_to_memslot(kvm->memslots, mem->slot);
787 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
788 npages = mem->memory_size >> PAGE_SHIFT;
790 if (npages > KVM_MEM_MAX_NR_PAGES)
794 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
799 new.base_gfn = base_gfn;
801 new.flags = mem->flags;
805 change = KVM_MR_CREATE;
806 else { /* Modify an existing slot. */
807 if ((mem->userspace_addr != old.userspace_addr) ||
808 (npages != old.npages) ||
809 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
812 if (base_gfn != old.base_gfn)
813 change = KVM_MR_MOVE;
814 else if (new.flags != old.flags)
815 change = KVM_MR_FLAGS_ONLY;
816 else { /* Nothing to change. */
821 } else if (old.npages) {
822 change = KVM_MR_DELETE;
823 } else /* Modify a non-existent slot: disallowed. */
826 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
827 /* Check for overlaps */
829 kvm_for_each_memslot(slot, kvm->memslots) {
830 if ((slot->id >= KVM_USER_MEM_SLOTS) ||
831 (slot->id == mem->slot))
833 if (!((base_gfn + npages <= slot->base_gfn) ||
834 (base_gfn >= slot->base_gfn + slot->npages)))
839 /* Free page dirty bitmap if unneeded */
840 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
841 new.dirty_bitmap = NULL;
844 if (change == KVM_MR_CREATE) {
845 new.userspace_addr = mem->userspace_addr;
847 if (kvm_arch_create_memslot(kvm, &new, npages))
851 /* Allocate page dirty bitmap if needed */
852 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
853 if (kvm_create_dirty_bitmap(&new) < 0)
857 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
862 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
863 slot = id_to_memslot(slots, mem->slot);
864 slot->flags |= KVM_MEMSLOT_INVALID;
866 old_memslots = install_new_memslots(kvm, slots);
868 /* slot was deleted or moved, clear iommu mapping */
869 kvm_iommu_unmap_pages(kvm, &old);
870 /* From this point no new shadow pages pointing to a deleted,
871 * or moved, memslot will be created.
873 * validation of sp->gfn happens in:
874 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
875 * - kvm_is_visible_gfn (mmu_check_roots)
877 kvm_arch_flush_shadow_memslot(kvm, slot);
880 * We can re-use the old_memslots from above, the only difference
881 * from the currently installed memslots is the invalid flag. This
882 * will get overwritten by update_memslots anyway.
884 slots = old_memslots;
887 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
891 /* actual memory is freed via old in kvm_free_physmem_slot below */
892 if (change == KVM_MR_DELETE) {
893 new.dirty_bitmap = NULL;
894 memset(&new.arch, 0, sizeof(new.arch));
897 update_memslots(slots, &new);
898 old_memslots = install_new_memslots(kvm, slots);
900 kvm_arch_commit_memory_region(kvm, mem, &old, change);
902 kvm_free_physmem_slot(kvm, &old, &new);
906 * IOMMU mapping: New slots need to be mapped. Old slots need to be
907 * un-mapped and re-mapped if their base changes. Since base change
908 * unmapping is handled above with slot deletion, mapping alone is
909 * needed here. Anything else the iommu might care about for existing
910 * slots (size changes, userspace addr changes and read-only flag
911 * changes) is disallowed above, so any other attribute changes getting
912 * here can be skipped.
914 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
915 r = kvm_iommu_map_pages(kvm, &new);
924 kvm_free_physmem_slot(kvm, &new, &old);
928 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
930 int kvm_set_memory_region(struct kvm *kvm,
931 struct kvm_userspace_memory_region *mem)
935 mutex_lock(&kvm->slots_lock);
936 r = __kvm_set_memory_region(kvm, mem);
937 mutex_unlock(&kvm->slots_lock);
940 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
942 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
943 struct kvm_userspace_memory_region *mem)
945 if (mem->slot >= KVM_USER_MEM_SLOTS)
947 return kvm_set_memory_region(kvm, mem);
950 int kvm_get_dirty_log(struct kvm *kvm,
951 struct kvm_dirty_log *log, int *is_dirty)
953 struct kvm_memory_slot *memslot;
956 unsigned long any = 0;
959 if (log->slot >= KVM_USER_MEM_SLOTS)
962 memslot = id_to_memslot(kvm->memslots, log->slot);
964 if (!memslot->dirty_bitmap)
967 n = kvm_dirty_bitmap_bytes(memslot);
969 for (i = 0; !any && i < n/sizeof(long); ++i)
970 any = memslot->dirty_bitmap[i];
973 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
983 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
985 bool kvm_largepages_enabled(void)
987 return largepages_enabled;
990 void kvm_disable_largepages(void)
992 largepages_enabled = false;
994 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
996 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
998 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1000 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1002 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1004 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1006 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1007 memslot->flags & KVM_MEMSLOT_INVALID)
1012 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1014 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1016 struct vm_area_struct *vma;
1017 unsigned long addr, size;
1021 addr = gfn_to_hva(kvm, gfn);
1022 if (kvm_is_error_hva(addr))
1025 down_read(¤t->mm->mmap_sem);
1026 vma = find_vma(current->mm, addr);
1030 size = vma_kernel_pagesize(vma);
1033 up_read(¤t->mm->mmap_sem);
1038 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1040 return slot->flags & KVM_MEM_READONLY;
1043 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1044 gfn_t *nr_pages, bool write)
1046 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1047 return KVM_HVA_ERR_BAD;
1049 if (memslot_is_readonly(slot) && write)
1050 return KVM_HVA_ERR_RO_BAD;
1053 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1055 return __gfn_to_hva_memslot(slot, gfn);
1058 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1061 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1064 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1067 return gfn_to_hva_many(slot, gfn, NULL);
1069 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1071 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1073 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1075 EXPORT_SYMBOL_GPL(gfn_to_hva);
1078 * If writable is set to false, the hva returned by this function is only
1079 * allowed to be read.
1081 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1082 gfn_t gfn, bool *writable)
1084 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1086 if (!kvm_is_error_hva(hva) && writable)
1087 *writable = !memslot_is_readonly(slot);
1092 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1094 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1096 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1099 static int kvm_read_hva(void *data, void __user *hva, int len)
1101 return __copy_from_user(data, hva, len);
1104 static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1106 return __copy_from_user_inatomic(data, hva, len);
1109 static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1110 unsigned long start, int write, struct page **page)
1112 int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1115 flags |= FOLL_WRITE;
1117 return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1120 int kvm_get_user_page_io(struct task_struct *tsk, struct mm_struct *mm,
1121 unsigned long addr, bool write_fault,
1122 struct page **pagep)
1126 int flags = FOLL_TOUCH | FOLL_HWPOISON |
1127 (pagep ? FOLL_GET : 0) |
1128 (write_fault ? FOLL_WRITE : 0);
1131 * If retrying the fault, we get here *not* having allowed the filemap
1132 * to wait on the page lock. We should now allow waiting on the IO with
1133 * the mmap semaphore released.
1135 down_read(&mm->mmap_sem);
1136 npages = __get_user_pages(tsk, mm, addr, 1, flags, pagep, NULL,
1145 * The previous call has now waited on the IO. Now we can
1146 * retry and complete. Pass TRIED to ensure we do not re
1147 * schedule async IO (see e.g. filemap_fault).
1149 down_read(&mm->mmap_sem);
1150 npages = __get_user_pages(tsk, mm, addr, 1, flags | FOLL_TRIED,
1153 up_read(&mm->mmap_sem);
1157 static inline int check_user_page_hwpoison(unsigned long addr)
1159 int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1161 rc = __get_user_pages(current, current->mm, addr, 1,
1162 flags, NULL, NULL, NULL);
1163 return rc == -EHWPOISON;
1167 * The atomic path to get the writable pfn which will be stored in @pfn,
1168 * true indicates success, otherwise false is returned.
1170 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1171 bool write_fault, bool *writable, pfn_t *pfn)
1173 struct page *page[1];
1176 if (!(async || atomic))
1180 * Fast pin a writable pfn only if it is a write fault request
1181 * or the caller allows to map a writable pfn for a read fault
1184 if (!(write_fault || writable))
1187 npages = __get_user_pages_fast(addr, 1, 1, page);
1189 *pfn = page_to_pfn(page[0]);
1200 * The slow path to get the pfn of the specified host virtual address,
1201 * 1 indicates success, -errno is returned if error is detected.
1203 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1204 bool *writable, pfn_t *pfn)
1206 struct page *page[1];
1212 *writable = write_fault;
1215 down_read(¤t->mm->mmap_sem);
1216 npages = get_user_page_nowait(current, current->mm,
1217 addr, write_fault, page);
1218 up_read(¤t->mm->mmap_sem);
1221 * By now we have tried gup_fast, and possibly async_pf, and we
1222 * are certainly not atomic. Time to retry the gup, allowing
1223 * mmap semaphore to be relinquished in the case of IO.
1225 npages = kvm_get_user_page_io(current, current->mm, addr,
1231 /* map read fault as writable if possible */
1232 if (unlikely(!write_fault) && writable) {
1233 struct page *wpage[1];
1235 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1244 *pfn = page_to_pfn(page[0]);
1248 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1250 if (unlikely(!(vma->vm_flags & VM_READ)))
1253 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1260 * Pin guest page in memory and return its pfn.
1261 * @addr: host virtual address which maps memory to the guest
1262 * @atomic: whether this function can sleep
1263 * @async: whether this function need to wait IO complete if the
1264 * host page is not in the memory
1265 * @write_fault: whether we should get a writable host page
1266 * @writable: whether it allows to map a writable host page for !@write_fault
1268 * The function will map a writable host page for these two cases:
1269 * 1): @write_fault = true
1270 * 2): @write_fault = false && @writable, @writable will tell the caller
1271 * whether the mapping is writable.
1273 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1274 bool write_fault, bool *writable)
1276 struct vm_area_struct *vma;
1280 /* we can do it either atomically or asynchronously, not both */
1281 BUG_ON(atomic && async);
1283 if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1287 return KVM_PFN_ERR_FAULT;
1289 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1293 down_read(¤t->mm->mmap_sem);
1294 if (npages == -EHWPOISON ||
1295 (!async && check_user_page_hwpoison(addr))) {
1296 pfn = KVM_PFN_ERR_HWPOISON;
1300 vma = find_vma_intersection(current->mm, addr, addr + 1);
1303 pfn = KVM_PFN_ERR_FAULT;
1304 else if ((vma->vm_flags & VM_PFNMAP)) {
1305 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1307 BUG_ON(!kvm_is_mmio_pfn(pfn));
1309 if (async && vma_is_valid(vma, write_fault))
1311 pfn = KVM_PFN_ERR_FAULT;
1314 up_read(¤t->mm->mmap_sem);
1319 __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1320 bool *async, bool write_fault, bool *writable)
1322 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1324 if (addr == KVM_HVA_ERR_RO_BAD)
1325 return KVM_PFN_ERR_RO_FAULT;
1327 if (kvm_is_error_hva(addr))
1328 return KVM_PFN_NOSLOT;
1330 /* Do not map writable pfn in the readonly memslot. */
1331 if (writable && memslot_is_readonly(slot)) {
1336 return hva_to_pfn(addr, atomic, async, write_fault,
1340 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1341 bool write_fault, bool *writable)
1343 struct kvm_memory_slot *slot;
1348 slot = gfn_to_memslot(kvm, gfn);
1350 return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1354 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1356 return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1358 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1360 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1361 bool write_fault, bool *writable)
1363 return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1365 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1367 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1369 return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1371 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1373 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1376 return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1378 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1380 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1382 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1385 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1387 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1389 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1391 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1397 addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1398 if (kvm_is_error_hva(addr))
1401 if (entry < nr_pages)
1404 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1406 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1408 static struct page *kvm_pfn_to_page(pfn_t pfn)
1410 if (is_error_noslot_pfn(pfn))
1411 return KVM_ERR_PTR_BAD_PAGE;
1413 if (kvm_is_mmio_pfn(pfn)) {
1415 return KVM_ERR_PTR_BAD_PAGE;
1418 return pfn_to_page(pfn);
1421 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1425 pfn = gfn_to_pfn(kvm, gfn);
1427 return kvm_pfn_to_page(pfn);
1430 EXPORT_SYMBOL_GPL(gfn_to_page);
1432 void kvm_release_page_clean(struct page *page)
1434 WARN_ON(is_error_page(page));
1436 kvm_release_pfn_clean(page_to_pfn(page));
1438 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1440 void kvm_release_pfn_clean(pfn_t pfn)
1442 if (!is_error_noslot_pfn(pfn) && !kvm_is_mmio_pfn(pfn))
1443 put_page(pfn_to_page(pfn));
1445 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1447 void kvm_release_page_dirty(struct page *page)
1449 WARN_ON(is_error_page(page));
1451 kvm_release_pfn_dirty(page_to_pfn(page));
1453 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1455 static void kvm_release_pfn_dirty(pfn_t pfn)
1457 kvm_set_pfn_dirty(pfn);
1458 kvm_release_pfn_clean(pfn);
1461 void kvm_set_pfn_dirty(pfn_t pfn)
1463 if (!kvm_is_mmio_pfn(pfn)) {
1464 struct page *page = pfn_to_page(pfn);
1465 if (!PageReserved(page))
1469 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1471 void kvm_set_pfn_accessed(pfn_t pfn)
1473 if (!kvm_is_mmio_pfn(pfn))
1474 mark_page_accessed(pfn_to_page(pfn));
1476 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1478 void kvm_get_pfn(pfn_t pfn)
1480 if (!kvm_is_mmio_pfn(pfn))
1481 get_page(pfn_to_page(pfn));
1483 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1485 static int next_segment(unsigned long len, int offset)
1487 if (len > PAGE_SIZE - offset)
1488 return PAGE_SIZE - offset;
1493 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1499 addr = gfn_to_hva_prot(kvm, gfn, NULL);
1500 if (kvm_is_error_hva(addr))
1502 r = kvm_read_hva(data, (void __user *)addr + offset, len);
1507 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1509 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1511 gfn_t gfn = gpa >> PAGE_SHIFT;
1513 int offset = offset_in_page(gpa);
1516 while ((seg = next_segment(len, offset)) != 0) {
1517 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1527 EXPORT_SYMBOL_GPL(kvm_read_guest);
1529 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1534 gfn_t gfn = gpa >> PAGE_SHIFT;
1535 int offset = offset_in_page(gpa);
1537 addr = gfn_to_hva_prot(kvm, gfn, NULL);
1538 if (kvm_is_error_hva(addr))
1540 pagefault_disable();
1541 r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
1547 EXPORT_SYMBOL(kvm_read_guest_atomic);
1549 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1550 int offset, int len)
1555 addr = gfn_to_hva(kvm, gfn);
1556 if (kvm_is_error_hva(addr))
1558 r = __copy_to_user((void __user *)addr + offset, data, len);
1561 mark_page_dirty(kvm, gfn);
1564 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1566 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1569 gfn_t gfn = gpa >> PAGE_SHIFT;
1571 int offset = offset_in_page(gpa);
1574 while ((seg = next_segment(len, offset)) != 0) {
1575 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1586 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1587 gpa_t gpa, unsigned long len)
1589 struct kvm_memslots *slots = kvm_memslots(kvm);
1590 int offset = offset_in_page(gpa);
1591 gfn_t start_gfn = gpa >> PAGE_SHIFT;
1592 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1593 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1594 gfn_t nr_pages_avail;
1597 ghc->generation = slots->generation;
1599 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1600 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, &nr_pages_avail);
1601 if (!kvm_is_error_hva(ghc->hva) && nr_pages_avail >= nr_pages_needed) {
1605 * If the requested region crosses two memslots, we still
1606 * verify that the entire region is valid here.
1608 while (start_gfn <= end_gfn) {
1609 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1610 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1612 if (kvm_is_error_hva(ghc->hva))
1614 start_gfn += nr_pages_avail;
1616 /* Use the slow path for cross page reads and writes. */
1617 ghc->memslot = NULL;
1621 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1623 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1624 void *data, unsigned long len)
1626 struct kvm_memslots *slots = kvm_memslots(kvm);
1629 BUG_ON(len > ghc->len);
1631 if (slots->generation != ghc->generation)
1632 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1634 if (unlikely(!ghc->memslot))
1635 return kvm_write_guest(kvm, ghc->gpa, data, len);
1637 if (kvm_is_error_hva(ghc->hva))
1640 r = __copy_to_user((void __user *)ghc->hva, data, len);
1643 mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1647 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1649 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1650 void *data, unsigned long len)
1652 struct kvm_memslots *slots = kvm_memslots(kvm);
1655 BUG_ON(len > ghc->len);
1657 if (slots->generation != ghc->generation)
1658 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1660 if (unlikely(!ghc->memslot))
1661 return kvm_read_guest(kvm, ghc->gpa, data, len);
1663 if (kvm_is_error_hva(ghc->hva))
1666 r = __copy_from_user(data, (void __user *)ghc->hva, len);
1672 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1674 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1676 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
1678 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
1680 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1682 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1684 gfn_t gfn = gpa >> PAGE_SHIFT;
1686 int offset = offset_in_page(gpa);
1689 while ((seg = next_segment(len, offset)) != 0) {
1690 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1699 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1701 static void mark_page_dirty_in_slot(struct kvm *kvm,
1702 struct kvm_memory_slot *memslot,
1705 if (memslot && memslot->dirty_bitmap) {
1706 unsigned long rel_gfn = gfn - memslot->base_gfn;
1708 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1712 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1714 struct kvm_memory_slot *memslot;
1716 memslot = gfn_to_memslot(kvm, gfn);
1717 mark_page_dirty_in_slot(kvm, memslot, gfn);
1719 EXPORT_SYMBOL_GPL(mark_page_dirty);
1722 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1724 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1729 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1731 if (kvm_arch_vcpu_runnable(vcpu)) {
1732 kvm_make_request(KVM_REQ_UNHALT, vcpu);
1735 if (kvm_cpu_has_pending_timer(vcpu))
1737 if (signal_pending(current))
1743 finish_wait(&vcpu->wq, &wait);
1745 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
1749 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1751 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1754 int cpu = vcpu->cpu;
1755 wait_queue_head_t *wqp;
1757 wqp = kvm_arch_vcpu_wq(vcpu);
1758 if (waitqueue_active(wqp)) {
1759 wake_up_interruptible(wqp);
1760 ++vcpu->stat.halt_wakeup;
1764 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1765 if (kvm_arch_vcpu_should_kick(vcpu))
1766 smp_send_reschedule(cpu);
1769 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
1770 #endif /* !CONFIG_S390 */
1772 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
1775 struct task_struct *task = NULL;
1779 pid = rcu_dereference(target->pid);
1781 task = get_pid_task(pid, PIDTYPE_PID);
1785 if (task->flags & PF_VCPU) {
1786 put_task_struct(task);
1789 ret = yield_to(task, 1);
1790 put_task_struct(task);
1794 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1797 * Helper that checks whether a VCPU is eligible for directed yield.
1798 * Most eligible candidate to yield is decided by following heuristics:
1800 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1801 * (preempted lock holder), indicated by @in_spin_loop.
1802 * Set at the beiginning and cleared at the end of interception/PLE handler.
1804 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1805 * chance last time (mostly it has become eligible now since we have probably
1806 * yielded to lockholder in last iteration. This is done by toggling
1807 * @dy_eligible each time a VCPU checked for eligibility.)
1809 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1810 * to preempted lock-holder could result in wrong VCPU selection and CPU
1811 * burning. Giving priority for a potential lock-holder increases lock
1814 * Since algorithm is based on heuristics, accessing another VCPU data without
1815 * locking does not harm. It may result in trying to yield to same VCPU, fail
1816 * and continue with next VCPU and so on.
1818 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1820 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1823 eligible = !vcpu->spin_loop.in_spin_loop ||
1824 vcpu->spin_loop.dy_eligible;
1826 if (vcpu->spin_loop.in_spin_loop)
1827 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1835 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1837 struct kvm *kvm = me->kvm;
1838 struct kvm_vcpu *vcpu;
1839 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1845 kvm_vcpu_set_in_spin_loop(me, true);
1847 * We boost the priority of a VCPU that is runnable but not
1848 * currently running, because it got preempted by something
1849 * else and called schedule in __vcpu_run. Hopefully that
1850 * VCPU is holding the lock that we need and will release it.
1851 * We approximate round-robin by starting at the last boosted VCPU.
1853 for (pass = 0; pass < 2 && !yielded && try; pass++) {
1854 kvm_for_each_vcpu(i, vcpu, kvm) {
1855 if (!pass && i <= last_boosted_vcpu) {
1856 i = last_boosted_vcpu;
1858 } else if (pass && i > last_boosted_vcpu)
1860 if (!ACCESS_ONCE(vcpu->preempted))
1864 if (waitqueue_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
1866 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1869 yielded = kvm_vcpu_yield_to(vcpu);
1871 kvm->last_boosted_vcpu = i;
1873 } else if (yielded < 0) {
1880 kvm_vcpu_set_in_spin_loop(me, false);
1882 /* Ensure vcpu is not eligible during next spinloop */
1883 kvm_vcpu_set_dy_eligible(me, false);
1885 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1887 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1889 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1892 if (vmf->pgoff == 0)
1893 page = virt_to_page(vcpu->run);
1895 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1896 page = virt_to_page(vcpu->arch.pio_data);
1898 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1899 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1900 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1903 return kvm_arch_vcpu_fault(vcpu, vmf);
1909 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1910 .fault = kvm_vcpu_fault,
1913 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1915 vma->vm_ops = &kvm_vcpu_vm_ops;
1919 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1921 struct kvm_vcpu *vcpu = filp->private_data;
1923 kvm_put_kvm(vcpu->kvm);
1927 static struct file_operations kvm_vcpu_fops = {
1928 .release = kvm_vcpu_release,
1929 .unlocked_ioctl = kvm_vcpu_ioctl,
1930 #ifdef CONFIG_COMPAT
1931 .compat_ioctl = kvm_vcpu_compat_ioctl,
1933 .mmap = kvm_vcpu_mmap,
1934 .llseek = noop_llseek,
1938 * Allocates an inode for the vcpu.
1940 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1942 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
1946 * Creates some virtual cpus. Good luck creating more than one.
1948 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1951 struct kvm_vcpu *vcpu, *v;
1953 if (id >= KVM_MAX_VCPUS)
1956 vcpu = kvm_arch_vcpu_create(kvm, id);
1958 return PTR_ERR(vcpu);
1960 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1962 r = kvm_arch_vcpu_setup(vcpu);
1966 mutex_lock(&kvm->lock);
1967 if (!kvm_vcpu_compatible(vcpu)) {
1969 goto unlock_vcpu_destroy;
1971 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1973 goto unlock_vcpu_destroy;
1976 kvm_for_each_vcpu(r, v, kvm)
1977 if (v->vcpu_id == id) {
1979 goto unlock_vcpu_destroy;
1982 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1984 /* Now it's all set up, let userspace reach it */
1986 r = create_vcpu_fd(vcpu);
1989 goto unlock_vcpu_destroy;
1992 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1994 atomic_inc(&kvm->online_vcpus);
1996 mutex_unlock(&kvm->lock);
1997 kvm_arch_vcpu_postcreate(vcpu);
2000 unlock_vcpu_destroy:
2001 mutex_unlock(&kvm->lock);
2003 kvm_arch_vcpu_destroy(vcpu);
2007 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2010 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2011 vcpu->sigset_active = 1;
2012 vcpu->sigset = *sigset;
2014 vcpu->sigset_active = 0;
2018 static long kvm_vcpu_ioctl(struct file *filp,
2019 unsigned int ioctl, unsigned long arg)
2021 struct kvm_vcpu *vcpu = filp->private_data;
2022 void __user *argp = (void __user *)arg;
2024 struct kvm_fpu *fpu = NULL;
2025 struct kvm_sregs *kvm_sregs = NULL;
2027 if (vcpu->kvm->mm != current->mm)
2030 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2033 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
2035 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
2036 * so vcpu_load() would break it.
2038 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
2039 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2043 r = vcpu_load(vcpu);
2051 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2052 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2054 case KVM_GET_REGS: {
2055 struct kvm_regs *kvm_regs;
2058 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2061 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2065 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2072 case KVM_SET_REGS: {
2073 struct kvm_regs *kvm_regs;
2076 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2077 if (IS_ERR(kvm_regs)) {
2078 r = PTR_ERR(kvm_regs);
2081 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2085 case KVM_GET_SREGS: {
2086 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2090 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2094 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2099 case KVM_SET_SREGS: {
2100 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2101 if (IS_ERR(kvm_sregs)) {
2102 r = PTR_ERR(kvm_sregs);
2106 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2109 case KVM_GET_MP_STATE: {
2110 struct kvm_mp_state mp_state;
2112 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2116 if (copy_to_user(argp, &mp_state, sizeof mp_state))
2121 case KVM_SET_MP_STATE: {
2122 struct kvm_mp_state mp_state;
2125 if (copy_from_user(&mp_state, argp, sizeof mp_state))
2127 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2130 case KVM_TRANSLATE: {
2131 struct kvm_translation tr;
2134 if (copy_from_user(&tr, argp, sizeof tr))
2136 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2140 if (copy_to_user(argp, &tr, sizeof tr))
2145 case KVM_SET_GUEST_DEBUG: {
2146 struct kvm_guest_debug dbg;
2149 if (copy_from_user(&dbg, argp, sizeof dbg))
2151 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2154 case KVM_SET_SIGNAL_MASK: {
2155 struct kvm_signal_mask __user *sigmask_arg = argp;
2156 struct kvm_signal_mask kvm_sigmask;
2157 sigset_t sigset, *p;
2162 if (copy_from_user(&kvm_sigmask, argp,
2163 sizeof kvm_sigmask))
2166 if (kvm_sigmask.len != sizeof sigset)
2169 if (copy_from_user(&sigset, sigmask_arg->sigset,
2174 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2178 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2182 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2186 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2192 fpu = memdup_user(argp, sizeof(*fpu));
2198 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2202 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2211 #ifdef CONFIG_COMPAT
2212 static long kvm_vcpu_compat_ioctl(struct file *filp,
2213 unsigned int ioctl, unsigned long arg)
2215 struct kvm_vcpu *vcpu = filp->private_data;
2216 void __user *argp = compat_ptr(arg);
2219 if (vcpu->kvm->mm != current->mm)
2223 case KVM_SET_SIGNAL_MASK: {
2224 struct kvm_signal_mask __user *sigmask_arg = argp;
2225 struct kvm_signal_mask kvm_sigmask;
2226 compat_sigset_t csigset;
2231 if (copy_from_user(&kvm_sigmask, argp,
2232 sizeof kvm_sigmask))
2235 if (kvm_sigmask.len != sizeof csigset)
2238 if (copy_from_user(&csigset, sigmask_arg->sigset,
2241 sigset_from_compat(&sigset, &csigset);
2242 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2244 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2248 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2256 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2257 int (*accessor)(struct kvm_device *dev,
2258 struct kvm_device_attr *attr),
2261 struct kvm_device_attr attr;
2266 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2269 return accessor(dev, &attr);
2272 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2275 struct kvm_device *dev = filp->private_data;
2278 case KVM_SET_DEVICE_ATTR:
2279 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2280 case KVM_GET_DEVICE_ATTR:
2281 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2282 case KVM_HAS_DEVICE_ATTR:
2283 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2285 if (dev->ops->ioctl)
2286 return dev->ops->ioctl(dev, ioctl, arg);
2292 static int kvm_device_release(struct inode *inode, struct file *filp)
2294 struct kvm_device *dev = filp->private_data;
2295 struct kvm *kvm = dev->kvm;
2301 static const struct file_operations kvm_device_fops = {
2302 .unlocked_ioctl = kvm_device_ioctl,
2303 #ifdef CONFIG_COMPAT
2304 .compat_ioctl = kvm_device_ioctl,
2306 .release = kvm_device_release,
2309 struct kvm_device *kvm_device_from_filp(struct file *filp)
2311 if (filp->f_op != &kvm_device_fops)
2314 return filp->private_data;
2317 static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
2318 #ifdef CONFIG_KVM_MPIC
2319 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
2320 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
2323 #ifdef CONFIG_KVM_XICS
2324 [KVM_DEV_TYPE_XICS] = &kvm_xics_ops,
2328 int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
2330 if (type >= ARRAY_SIZE(kvm_device_ops_table))
2333 if (kvm_device_ops_table[type] != NULL)
2336 kvm_device_ops_table[type] = ops;
2340 void kvm_unregister_device_ops(u32 type)
2342 if (kvm_device_ops_table[type] != NULL)
2343 kvm_device_ops_table[type] = NULL;
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);
2590 #ifdef CONFIG_COMPAT
2591 struct compat_kvm_dirty_log {
2595 compat_uptr_t dirty_bitmap; /* one bit per page */
2600 static long kvm_vm_compat_ioctl(struct file *filp,
2601 unsigned int ioctl, unsigned long arg)
2603 struct kvm *kvm = filp->private_data;
2606 if (kvm->mm != current->mm)
2609 case KVM_GET_DIRTY_LOG: {
2610 struct compat_kvm_dirty_log compat_log;
2611 struct kvm_dirty_log log;
2614 if (copy_from_user(&compat_log, (void __user *)arg,
2615 sizeof(compat_log)))
2617 log.slot = compat_log.slot;
2618 log.padding1 = compat_log.padding1;
2619 log.padding2 = compat_log.padding2;
2620 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2622 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2626 r = kvm_vm_ioctl(filp, ioctl, arg);
2634 static struct file_operations kvm_vm_fops = {
2635 .release = kvm_vm_release,
2636 .unlocked_ioctl = kvm_vm_ioctl,
2637 #ifdef CONFIG_COMPAT
2638 .compat_ioctl = kvm_vm_compat_ioctl,
2640 .llseek = noop_llseek,
2643 static int kvm_dev_ioctl_create_vm(unsigned long type)
2648 kvm = kvm_create_vm(type);
2650 return PTR_ERR(kvm);
2651 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2652 r = kvm_coalesced_mmio_init(kvm);
2658 r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC);
2665 static long kvm_dev_ioctl(struct file *filp,
2666 unsigned int ioctl, unsigned long arg)
2671 case KVM_GET_API_VERSION:
2674 r = KVM_API_VERSION;
2677 r = kvm_dev_ioctl_create_vm(arg);
2679 case KVM_CHECK_EXTENSION:
2680 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
2682 case KVM_GET_VCPU_MMAP_SIZE:
2685 r = PAGE_SIZE; /* struct kvm_run */
2687 r += PAGE_SIZE; /* pio data page */
2689 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2690 r += PAGE_SIZE; /* coalesced mmio ring page */
2693 case KVM_TRACE_ENABLE:
2694 case KVM_TRACE_PAUSE:
2695 case KVM_TRACE_DISABLE:
2699 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2705 static struct file_operations kvm_chardev_ops = {
2706 .unlocked_ioctl = kvm_dev_ioctl,
2707 .compat_ioctl = kvm_dev_ioctl,
2708 .llseek = noop_llseek,
2711 static struct miscdevice kvm_dev = {
2717 static void hardware_enable_nolock(void *junk)
2719 int cpu = raw_smp_processor_id();
2722 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2725 cpumask_set_cpu(cpu, cpus_hardware_enabled);
2727 r = kvm_arch_hardware_enable();
2730 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2731 atomic_inc(&hardware_enable_failed);
2732 printk(KERN_INFO "kvm: enabling virtualization on "
2733 "CPU%d failed\n", cpu);
2737 static void hardware_enable(void)
2739 raw_spin_lock(&kvm_count_lock);
2740 if (kvm_usage_count)
2741 hardware_enable_nolock(NULL);
2742 raw_spin_unlock(&kvm_count_lock);
2745 static void hardware_disable_nolock(void *junk)
2747 int cpu = raw_smp_processor_id();
2749 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2751 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2752 kvm_arch_hardware_disable();
2755 static void hardware_disable(void)
2757 raw_spin_lock(&kvm_count_lock);
2758 if (kvm_usage_count)
2759 hardware_disable_nolock(NULL);
2760 raw_spin_unlock(&kvm_count_lock);
2763 static void hardware_disable_all_nolock(void)
2765 BUG_ON(!kvm_usage_count);
2768 if (!kvm_usage_count)
2769 on_each_cpu(hardware_disable_nolock, NULL, 1);
2772 static void hardware_disable_all(void)
2774 raw_spin_lock(&kvm_count_lock);
2775 hardware_disable_all_nolock();
2776 raw_spin_unlock(&kvm_count_lock);
2779 static int hardware_enable_all(void)
2783 raw_spin_lock(&kvm_count_lock);
2786 if (kvm_usage_count == 1) {
2787 atomic_set(&hardware_enable_failed, 0);
2788 on_each_cpu(hardware_enable_nolock, NULL, 1);
2790 if (atomic_read(&hardware_enable_failed)) {
2791 hardware_disable_all_nolock();
2796 raw_spin_unlock(&kvm_count_lock);
2801 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2806 val &= ~CPU_TASKS_FROZEN;
2809 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2814 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2822 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2826 * Some (well, at least mine) BIOSes hang on reboot if
2829 * And Intel TXT required VMX off for all cpu when system shutdown.
2831 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2832 kvm_rebooting = true;
2833 on_each_cpu(hardware_disable_nolock, NULL, 1);
2837 static struct notifier_block kvm_reboot_notifier = {
2838 .notifier_call = kvm_reboot,
2842 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2846 for (i = 0; i < bus->dev_count; i++) {
2847 struct kvm_io_device *pos = bus->range[i].dev;
2849 kvm_iodevice_destructor(pos);
2854 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
2855 const struct kvm_io_range *r2)
2857 if (r1->addr < r2->addr)
2859 if (r1->addr + r1->len > r2->addr + r2->len)
2864 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2866 return kvm_io_bus_cmp(p1, p2);
2869 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2870 gpa_t addr, int len)
2872 bus->range[bus->dev_count++] = (struct kvm_io_range) {
2878 sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2879 kvm_io_bus_sort_cmp, NULL);
2884 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2885 gpa_t addr, int len)
2887 struct kvm_io_range *range, key;
2890 key = (struct kvm_io_range) {
2895 range = bsearch(&key, bus->range, bus->dev_count,
2896 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2900 off = range - bus->range;
2902 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
2908 static int __kvm_io_bus_write(struct kvm_io_bus *bus,
2909 struct kvm_io_range *range, const void *val)
2913 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2917 while (idx < bus->dev_count &&
2918 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
2919 if (!kvm_iodevice_write(bus->range[idx].dev, range->addr,
2928 /* kvm_io_bus_write - called under kvm->slots_lock */
2929 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2930 int len, const void *val)
2932 struct kvm_io_bus *bus;
2933 struct kvm_io_range range;
2936 range = (struct kvm_io_range) {
2941 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2942 r = __kvm_io_bus_write(bus, &range, val);
2943 return r < 0 ? r : 0;
2946 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
2947 int kvm_io_bus_write_cookie(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2948 int len, const void *val, long cookie)
2950 struct kvm_io_bus *bus;
2951 struct kvm_io_range range;
2953 range = (struct kvm_io_range) {
2958 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2960 /* First try the device referenced by cookie. */
2961 if ((cookie >= 0) && (cookie < bus->dev_count) &&
2962 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
2963 if (!kvm_iodevice_write(bus->range[cookie].dev, addr, len,
2968 * cookie contained garbage; fall back to search and return the
2969 * correct cookie value.
2971 return __kvm_io_bus_write(bus, &range, val);
2974 static int __kvm_io_bus_read(struct kvm_io_bus *bus, struct kvm_io_range *range,
2979 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2983 while (idx < bus->dev_count &&
2984 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
2985 if (!kvm_iodevice_read(bus->range[idx].dev, range->addr,
2993 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
2995 /* kvm_io_bus_read - called under kvm->slots_lock */
2996 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2999 struct kvm_io_bus *bus;
3000 struct kvm_io_range range;
3003 range = (struct kvm_io_range) {
3008 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
3009 r = __kvm_io_bus_read(bus, &range, val);
3010 return r < 0 ? r : 0;
3014 /* Caller must hold slots_lock. */
3015 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3016 int len, struct kvm_io_device *dev)
3018 struct kvm_io_bus *new_bus, *bus;
3020 bus = kvm->buses[bus_idx];
3021 /* exclude ioeventfd which is limited by maximum fd */
3022 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3025 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
3026 sizeof(struct kvm_io_range)), GFP_KERNEL);
3029 memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
3030 sizeof(struct kvm_io_range)));
3031 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
3032 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3033 synchronize_srcu_expedited(&kvm->srcu);
3039 /* Caller must hold slots_lock. */
3040 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3041 struct kvm_io_device *dev)
3044 struct kvm_io_bus *new_bus, *bus;
3046 bus = kvm->buses[bus_idx];
3048 for (i = 0; i < bus->dev_count; i++)
3049 if (bus->range[i].dev == dev) {
3057 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3058 sizeof(struct kvm_io_range)), GFP_KERNEL);
3062 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3063 new_bus->dev_count--;
3064 memcpy(new_bus->range + i, bus->range + i + 1,
3065 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3067 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3068 synchronize_srcu_expedited(&kvm->srcu);
3073 static struct notifier_block kvm_cpu_notifier = {
3074 .notifier_call = kvm_cpu_hotplug,
3077 static int vm_stat_get(void *_offset, u64 *val)
3079 unsigned offset = (long)_offset;
3083 spin_lock(&kvm_lock);
3084 list_for_each_entry(kvm, &vm_list, vm_list)
3085 *val += *(u32 *)((void *)kvm + offset);
3086 spin_unlock(&kvm_lock);
3090 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
3092 static int vcpu_stat_get(void *_offset, u64 *val)
3094 unsigned offset = (long)_offset;
3096 struct kvm_vcpu *vcpu;
3100 spin_lock(&kvm_lock);
3101 list_for_each_entry(kvm, &vm_list, vm_list)
3102 kvm_for_each_vcpu(i, vcpu, kvm)
3103 *val += *(u32 *)((void *)vcpu + offset);
3105 spin_unlock(&kvm_lock);
3109 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
3111 static const struct file_operations *stat_fops[] = {
3112 [KVM_STAT_VCPU] = &vcpu_stat_fops,
3113 [KVM_STAT_VM] = &vm_stat_fops,
3116 static int kvm_init_debug(void)
3119 struct kvm_stats_debugfs_item *p;
3121 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3122 if (kvm_debugfs_dir == NULL)
3125 for (p = debugfs_entries; p->name; ++p) {
3126 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
3127 (void *)(long)p->offset,
3128 stat_fops[p->kind]);
3129 if (p->dentry == NULL)
3136 debugfs_remove_recursive(kvm_debugfs_dir);
3141 static void kvm_exit_debug(void)
3143 struct kvm_stats_debugfs_item *p;
3145 for (p = debugfs_entries; p->name; ++p)
3146 debugfs_remove(p->dentry);
3147 debugfs_remove(kvm_debugfs_dir);
3150 static int kvm_suspend(void)
3152 if (kvm_usage_count)
3153 hardware_disable_nolock(NULL);
3157 static void kvm_resume(void)
3159 if (kvm_usage_count) {
3160 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3161 hardware_enable_nolock(NULL);
3165 static struct syscore_ops kvm_syscore_ops = {
3166 .suspend = kvm_suspend,
3167 .resume = kvm_resume,
3171 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3173 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3176 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3178 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3179 if (vcpu->preempted)
3180 vcpu->preempted = false;
3182 kvm_arch_sched_in(vcpu, cpu);
3184 kvm_arch_vcpu_load(vcpu, cpu);
3187 static void kvm_sched_out(struct preempt_notifier *pn,
3188 struct task_struct *next)
3190 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3192 if (current->state == TASK_RUNNING)
3193 vcpu->preempted = true;
3194 kvm_arch_vcpu_put(vcpu);
3197 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3198 struct module *module)
3203 r = kvm_arch_init(opaque);
3208 * kvm_arch_init makes sure there's at most one caller
3209 * for architectures that support multiple implementations,
3210 * like intel and amd on x86.
3211 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3212 * conflicts in case kvm is already setup for another implementation.
3214 r = kvm_irqfd_init();
3218 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
3223 r = kvm_arch_hardware_setup();
3227 for_each_online_cpu(cpu) {
3228 smp_call_function_single(cpu,
3229 kvm_arch_check_processor_compat,
3235 r = register_cpu_notifier(&kvm_cpu_notifier);
3238 register_reboot_notifier(&kvm_reboot_notifier);
3240 /* A kmem cache lets us meet the alignment requirements of fx_save. */
3242 vcpu_align = __alignof__(struct kvm_vcpu);
3243 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
3245 if (!kvm_vcpu_cache) {
3250 r = kvm_async_pf_init();
3254 kvm_chardev_ops.owner = module;
3255 kvm_vm_fops.owner = module;
3256 kvm_vcpu_fops.owner = module;
3258 r = misc_register(&kvm_dev);
3260 printk(KERN_ERR "kvm: misc device register failed\n");
3264 register_syscore_ops(&kvm_syscore_ops);
3266 kvm_preempt_ops.sched_in = kvm_sched_in;
3267 kvm_preempt_ops.sched_out = kvm_sched_out;
3269 r = kvm_init_debug();
3271 printk(KERN_ERR "kvm: create debugfs files failed\n");
3275 r = kvm_vfio_ops_init();
3281 unregister_syscore_ops(&kvm_syscore_ops);
3282 misc_deregister(&kvm_dev);
3284 kvm_async_pf_deinit();
3286 kmem_cache_destroy(kvm_vcpu_cache);
3288 unregister_reboot_notifier(&kvm_reboot_notifier);
3289 unregister_cpu_notifier(&kvm_cpu_notifier);
3292 kvm_arch_hardware_unsetup();
3294 free_cpumask_var(cpus_hardware_enabled);
3302 EXPORT_SYMBOL_GPL(kvm_init);
3307 misc_deregister(&kvm_dev);
3308 kmem_cache_destroy(kvm_vcpu_cache);
3309 kvm_async_pf_deinit();
3310 unregister_syscore_ops(&kvm_syscore_ops);
3311 unregister_reboot_notifier(&kvm_reboot_notifier);
3312 unregister_cpu_notifier(&kvm_cpu_notifier);
3313 on_each_cpu(hardware_disable_nolock, NULL, 1);
3314 kvm_arch_hardware_unsetup();
3317 free_cpumask_var(cpus_hardware_enabled);
3318 kvm_vfio_ops_exit();
3320 EXPORT_SYMBOL_GPL(kvm_exit);