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");
69 unsigned int halt_poll_ns = 0;
70 module_param(halt_poll_ns, uint, S_IRUGO | S_IWUSR);
75 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
78 DEFINE_SPINLOCK(kvm_lock);
79 static DEFINE_RAW_SPINLOCK(kvm_count_lock);
82 static cpumask_var_t cpus_hardware_enabled;
83 static int kvm_usage_count = 0;
84 static atomic_t hardware_enable_failed;
86 struct kmem_cache *kvm_vcpu_cache;
87 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
89 static __read_mostly struct preempt_ops kvm_preempt_ops;
91 struct dentry *kvm_debugfs_dir;
93 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
95 #ifdef CONFIG_KVM_COMPAT
96 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
99 static int hardware_enable_all(void);
100 static void hardware_disable_all(void);
102 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
104 static void kvm_release_pfn_dirty(pfn_t pfn);
105 static void mark_page_dirty_in_slot(struct kvm *kvm,
106 struct kvm_memory_slot *memslot, gfn_t gfn);
108 __visible bool kvm_rebooting;
109 EXPORT_SYMBOL_GPL(kvm_rebooting);
111 static bool largepages_enabled = true;
113 bool kvm_is_reserved_pfn(pfn_t pfn)
116 return PageReserved(pfn_to_page(pfn));
122 * Switches to specified vcpu, until a matching vcpu_put()
124 int vcpu_load(struct kvm_vcpu *vcpu)
128 if (mutex_lock_killable(&vcpu->mutex))
131 preempt_notifier_register(&vcpu->preempt_notifier);
132 kvm_arch_vcpu_load(vcpu, cpu);
137 void vcpu_put(struct kvm_vcpu *vcpu)
140 kvm_arch_vcpu_put(vcpu);
141 preempt_notifier_unregister(&vcpu->preempt_notifier);
143 mutex_unlock(&vcpu->mutex);
146 static void ack_flush(void *_completed)
150 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
155 struct kvm_vcpu *vcpu;
157 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
160 kvm_for_each_vcpu(i, vcpu, kvm) {
161 kvm_make_request(req, vcpu);
164 /* Set ->requests bit before we read ->mode */
167 if (cpus != NULL && cpu != -1 && cpu != me &&
168 kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
169 cpumask_set_cpu(cpu, cpus);
171 if (unlikely(cpus == NULL))
172 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
173 else if (!cpumask_empty(cpus))
174 smp_call_function_many(cpus, ack_flush, NULL, 1);
178 free_cpumask_var(cpus);
182 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
183 void kvm_flush_remote_tlbs(struct kvm *kvm)
185 long dirty_count = kvm->tlbs_dirty;
188 if (kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
189 ++kvm->stat.remote_tlb_flush;
190 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
192 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
195 void kvm_reload_remote_mmus(struct kvm *kvm)
197 kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
200 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
202 kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
205 void kvm_make_scan_ioapic_request(struct kvm *kvm)
207 kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
210 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
215 mutex_init(&vcpu->mutex);
220 init_waitqueue_head(&vcpu->wq);
221 kvm_async_pf_vcpu_init(vcpu);
223 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
228 vcpu->run = page_address(page);
230 kvm_vcpu_set_in_spin_loop(vcpu, false);
231 kvm_vcpu_set_dy_eligible(vcpu, false);
232 vcpu->preempted = false;
234 r = kvm_arch_vcpu_init(vcpu);
240 free_page((unsigned long)vcpu->run);
244 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
246 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
249 kvm_arch_vcpu_uninit(vcpu);
250 free_page((unsigned long)vcpu->run);
252 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
254 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
255 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
257 return container_of(mn, struct kvm, mmu_notifier);
260 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
261 struct mm_struct *mm,
262 unsigned long address)
264 struct kvm *kvm = mmu_notifier_to_kvm(mn);
265 int need_tlb_flush, idx;
268 * When ->invalidate_page runs, the linux pte has been zapped
269 * already but the page is still allocated until
270 * ->invalidate_page returns. So if we increase the sequence
271 * here the kvm page fault will notice if the spte can't be
272 * established because the page is going to be freed. If
273 * instead the kvm page fault establishes the spte before
274 * ->invalidate_page runs, kvm_unmap_hva will release it
277 * The sequence increase only need to be seen at spin_unlock
278 * time, and not at spin_lock time.
280 * Increasing the sequence after the spin_unlock would be
281 * unsafe because the kvm page fault could then establish the
282 * pte after kvm_unmap_hva returned, without noticing the page
283 * is going to be freed.
285 idx = srcu_read_lock(&kvm->srcu);
286 spin_lock(&kvm->mmu_lock);
288 kvm->mmu_notifier_seq++;
289 need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
290 /* we've to flush the tlb before the pages can be freed */
292 kvm_flush_remote_tlbs(kvm);
294 spin_unlock(&kvm->mmu_lock);
296 kvm_arch_mmu_notifier_invalidate_page(kvm, address);
298 srcu_read_unlock(&kvm->srcu, idx);
301 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
302 struct mm_struct *mm,
303 unsigned long address,
306 struct kvm *kvm = mmu_notifier_to_kvm(mn);
309 idx = srcu_read_lock(&kvm->srcu);
310 spin_lock(&kvm->mmu_lock);
311 kvm->mmu_notifier_seq++;
312 kvm_set_spte_hva(kvm, address, pte);
313 spin_unlock(&kvm->mmu_lock);
314 srcu_read_unlock(&kvm->srcu, idx);
317 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
318 struct mm_struct *mm,
322 struct kvm *kvm = mmu_notifier_to_kvm(mn);
323 int need_tlb_flush = 0, idx;
325 idx = srcu_read_lock(&kvm->srcu);
326 spin_lock(&kvm->mmu_lock);
328 * The count increase must become visible at unlock time as no
329 * spte can be established without taking the mmu_lock and
330 * count is also read inside the mmu_lock critical section.
332 kvm->mmu_notifier_count++;
333 need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
334 need_tlb_flush |= kvm->tlbs_dirty;
335 /* we've to flush the tlb before the pages can be freed */
337 kvm_flush_remote_tlbs(kvm);
339 spin_unlock(&kvm->mmu_lock);
340 srcu_read_unlock(&kvm->srcu, idx);
343 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
344 struct mm_struct *mm,
348 struct kvm *kvm = mmu_notifier_to_kvm(mn);
350 spin_lock(&kvm->mmu_lock);
352 * This sequence increase will notify the kvm page fault that
353 * the page that is going to be mapped in the spte could have
356 kvm->mmu_notifier_seq++;
359 * The above sequence increase must be visible before the
360 * below count decrease, which is ensured by the smp_wmb above
361 * in conjunction with the smp_rmb in mmu_notifier_retry().
363 kvm->mmu_notifier_count--;
364 spin_unlock(&kvm->mmu_lock);
366 BUG_ON(kvm->mmu_notifier_count < 0);
369 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
370 struct mm_struct *mm,
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, start, end);
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_IRQFD
468 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
471 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
474 kvm->memslots = kvm_kvzalloc(sizeof(struct kvm_memslots));
476 goto out_err_no_srcu;
479 * Init kvm generation close to the maximum to easily test the
480 * code of handling generation number wrap-around.
482 kvm->memslots->generation = -150;
484 kvm_init_memslots_id(kvm);
485 if (init_srcu_struct(&kvm->srcu))
486 goto out_err_no_srcu;
487 if (init_srcu_struct(&kvm->irq_srcu))
488 goto out_err_no_irq_srcu;
489 for (i = 0; i < KVM_NR_BUSES; i++) {
490 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
496 spin_lock_init(&kvm->mmu_lock);
497 kvm->mm = current->mm;
498 atomic_inc(&kvm->mm->mm_count);
499 kvm_eventfd_init(kvm);
500 mutex_init(&kvm->lock);
501 mutex_init(&kvm->irq_lock);
502 mutex_init(&kvm->slots_lock);
503 atomic_set(&kvm->users_count, 1);
504 INIT_LIST_HEAD(&kvm->devices);
506 r = kvm_init_mmu_notifier(kvm);
510 spin_lock(&kvm_lock);
511 list_add(&kvm->vm_list, &vm_list);
512 spin_unlock(&kvm_lock);
517 cleanup_srcu_struct(&kvm->irq_srcu);
519 cleanup_srcu_struct(&kvm->srcu);
521 hardware_disable_all();
523 for (i = 0; i < KVM_NR_BUSES; i++)
524 kfree(kvm->buses[i]);
525 kvfree(kvm->memslots);
526 kvm_arch_free_vm(kvm);
531 * Avoid using vmalloc for a small buffer.
532 * Should not be used when the size is statically known.
534 void *kvm_kvzalloc(unsigned long size)
536 if (size > PAGE_SIZE)
537 return vzalloc(size);
539 return kzalloc(size, GFP_KERNEL);
542 void kvm_kvfree(const void *addr)
544 if (is_vmalloc_addr(addr))
550 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
552 if (!memslot->dirty_bitmap)
555 kvm_kvfree(memslot->dirty_bitmap);
556 memslot->dirty_bitmap = NULL;
560 * Free any memory in @free but not in @dont.
562 static void kvm_free_physmem_slot(struct kvm *kvm, struct kvm_memory_slot *free,
563 struct kvm_memory_slot *dont)
565 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
566 kvm_destroy_dirty_bitmap(free);
568 kvm_arch_free_memslot(kvm, free, dont);
573 static void kvm_free_physmem(struct kvm *kvm)
575 struct kvm_memslots *slots = kvm->memslots;
576 struct kvm_memory_slot *memslot;
578 kvm_for_each_memslot(memslot, slots)
579 kvm_free_physmem_slot(kvm, memslot, NULL);
581 kvfree(kvm->memslots);
584 static void kvm_destroy_devices(struct kvm *kvm)
586 struct list_head *node, *tmp;
588 list_for_each_safe(node, tmp, &kvm->devices) {
589 struct kvm_device *dev =
590 list_entry(node, struct kvm_device, vm_node);
593 dev->ops->destroy(dev);
597 static void kvm_destroy_vm(struct kvm *kvm)
600 struct mm_struct *mm = kvm->mm;
602 kvm_arch_sync_events(kvm);
603 spin_lock(&kvm_lock);
604 list_del(&kvm->vm_list);
605 spin_unlock(&kvm_lock);
606 kvm_free_irq_routing(kvm);
607 for (i = 0; i < KVM_NR_BUSES; i++)
608 kvm_io_bus_destroy(kvm->buses[i]);
609 kvm_coalesced_mmio_free(kvm);
610 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
611 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
613 kvm_arch_flush_shadow_all(kvm);
615 kvm_arch_destroy_vm(kvm);
616 kvm_destroy_devices(kvm);
617 kvm_free_physmem(kvm);
618 cleanup_srcu_struct(&kvm->irq_srcu);
619 cleanup_srcu_struct(&kvm->srcu);
620 kvm_arch_free_vm(kvm);
621 hardware_disable_all();
625 void kvm_get_kvm(struct kvm *kvm)
627 atomic_inc(&kvm->users_count);
629 EXPORT_SYMBOL_GPL(kvm_get_kvm);
631 void kvm_put_kvm(struct kvm *kvm)
633 if (atomic_dec_and_test(&kvm->users_count))
636 EXPORT_SYMBOL_GPL(kvm_put_kvm);
639 static int kvm_vm_release(struct inode *inode, struct file *filp)
641 struct kvm *kvm = filp->private_data;
643 kvm_irqfd_release(kvm);
650 * Allocation size is twice as large as the actual dirty bitmap size.
651 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
653 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
655 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
657 memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
658 if (!memslot->dirty_bitmap)
665 * Insert memslot and re-sort memslots based on their GFN,
666 * so binary search could be used to lookup GFN.
667 * Sorting algorithm takes advantage of having initially
668 * sorted array and known changed memslot position.
670 static void update_memslots(struct kvm_memslots *slots,
671 struct kvm_memory_slot *new)
674 int i = slots->id_to_index[id];
675 struct kvm_memory_slot *mslots = slots->memslots;
677 WARN_ON(mslots[i].id != id);
679 WARN_ON(!mslots[i].npages);
682 if (mslots[i].npages)
685 if (!mslots[i].npages)
689 while (i < KVM_MEM_SLOTS_NUM - 1 &&
690 new->base_gfn <= mslots[i + 1].base_gfn) {
691 if (!mslots[i + 1].npages)
693 mslots[i] = mslots[i + 1];
694 slots->id_to_index[mslots[i].id] = i;
699 * The ">=" is needed when creating a slot with base_gfn == 0,
700 * so that it moves before all those with base_gfn == npages == 0.
702 * On the other hand, if new->npages is zero, the above loop has
703 * already left i pointing to the beginning of the empty part of
704 * mslots, and the ">=" would move the hole backwards in this
705 * case---which is wrong. So skip the loop when deleting a slot.
709 new->base_gfn >= mslots[i - 1].base_gfn) {
710 mslots[i] = mslots[i - 1];
711 slots->id_to_index[mslots[i].id] = i;
715 WARN_ON_ONCE(i != slots->used_slots);
718 slots->id_to_index[mslots[i].id] = i;
721 static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
723 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
725 #ifdef __KVM_HAVE_READONLY_MEM
726 valid_flags |= KVM_MEM_READONLY;
729 if (mem->flags & ~valid_flags)
735 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
736 struct kvm_memslots *slots)
738 struct kvm_memslots *old_memslots = kvm->memslots;
741 * Set the low bit in the generation, which disables SPTE caching
742 * until the end of synchronize_srcu_expedited.
744 WARN_ON(old_memslots->generation & 1);
745 slots->generation = old_memslots->generation + 1;
747 rcu_assign_pointer(kvm->memslots, slots);
748 synchronize_srcu_expedited(&kvm->srcu);
751 * Increment the new memslot generation a second time. This prevents
752 * vm exits that race with memslot updates from caching a memslot
753 * generation that will (potentially) be valid forever.
757 kvm_arch_memslots_updated(kvm);
763 * Allocate some memory and give it an address in the guest physical address
766 * Discontiguous memory is allowed, mostly for framebuffers.
768 * Must be called holding kvm->slots_lock for write.
770 int __kvm_set_memory_region(struct kvm *kvm,
771 struct kvm_userspace_memory_region *mem)
775 unsigned long npages;
776 struct kvm_memory_slot *slot;
777 struct kvm_memory_slot old, new;
778 struct kvm_memslots *slots = NULL, *old_memslots;
779 enum kvm_mr_change change;
781 r = check_memory_region_flags(mem);
786 /* General sanity checks */
787 if (mem->memory_size & (PAGE_SIZE - 1))
789 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
791 /* We can read the guest memory with __xxx_user() later on. */
792 if ((mem->slot < KVM_USER_MEM_SLOTS) &&
793 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
794 !access_ok(VERIFY_WRITE,
795 (void __user *)(unsigned long)mem->userspace_addr,
798 if (mem->slot >= KVM_MEM_SLOTS_NUM)
800 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
803 slot = id_to_memslot(kvm->memslots, mem->slot);
804 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
805 npages = mem->memory_size >> PAGE_SHIFT;
807 if (npages > KVM_MEM_MAX_NR_PAGES)
811 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
816 new.base_gfn = base_gfn;
818 new.flags = mem->flags;
822 change = KVM_MR_CREATE;
823 else { /* Modify an existing slot. */
824 if ((mem->userspace_addr != old.userspace_addr) ||
825 (npages != old.npages) ||
826 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
829 if (base_gfn != old.base_gfn)
830 change = KVM_MR_MOVE;
831 else if (new.flags != old.flags)
832 change = KVM_MR_FLAGS_ONLY;
833 else { /* Nothing to change. */
838 } else if (old.npages) {
839 change = KVM_MR_DELETE;
840 } else /* Modify a non-existent slot: disallowed. */
843 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
844 /* Check for overlaps */
846 kvm_for_each_memslot(slot, kvm->memslots) {
847 if ((slot->id >= KVM_USER_MEM_SLOTS) ||
848 (slot->id == mem->slot))
850 if (!((base_gfn + npages <= slot->base_gfn) ||
851 (base_gfn >= slot->base_gfn + slot->npages)))
856 /* Free page dirty bitmap if unneeded */
857 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
858 new.dirty_bitmap = NULL;
861 if (change == KVM_MR_CREATE) {
862 new.userspace_addr = mem->userspace_addr;
864 if (kvm_arch_create_memslot(kvm, &new, npages))
868 /* Allocate page dirty bitmap if needed */
869 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
870 if (kvm_create_dirty_bitmap(&new) < 0)
874 slots = kvm_kvzalloc(sizeof(struct kvm_memslots));
877 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
879 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
880 slot = id_to_memslot(slots, mem->slot);
881 slot->flags |= KVM_MEMSLOT_INVALID;
883 old_memslots = install_new_memslots(kvm, slots);
885 /* slot was deleted or moved, clear iommu mapping */
886 kvm_iommu_unmap_pages(kvm, &old);
887 /* From this point no new shadow pages pointing to a deleted,
888 * or moved, memslot will be created.
890 * validation of sp->gfn happens in:
891 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
892 * - kvm_is_visible_gfn (mmu_check_roots)
894 kvm_arch_flush_shadow_memslot(kvm, slot);
897 * We can re-use the old_memslots from above, the only difference
898 * from the currently installed memslots is the invalid flag. This
899 * will get overwritten by update_memslots anyway.
901 slots = old_memslots;
904 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
908 /* actual memory is freed via old in kvm_free_physmem_slot below */
909 if (change == KVM_MR_DELETE) {
910 new.dirty_bitmap = NULL;
911 memset(&new.arch, 0, sizeof(new.arch));
914 update_memslots(slots, &new);
915 old_memslots = install_new_memslots(kvm, slots);
917 kvm_arch_commit_memory_region(kvm, mem, &old, change);
919 kvm_free_physmem_slot(kvm, &old, &new);
920 kvfree(old_memslots);
923 * IOMMU mapping: New slots need to be mapped. Old slots need to be
924 * un-mapped and re-mapped if their base changes. Since base change
925 * unmapping is handled above with slot deletion, mapping alone is
926 * needed here. Anything else the iommu might care about for existing
927 * slots (size changes, userspace addr changes and read-only flag
928 * changes) is disallowed above, so any other attribute changes getting
929 * here can be skipped.
931 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
932 r = kvm_iommu_map_pages(kvm, &new);
941 kvm_free_physmem_slot(kvm, &new, &old);
945 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
947 int kvm_set_memory_region(struct kvm *kvm,
948 struct kvm_userspace_memory_region *mem)
952 mutex_lock(&kvm->slots_lock);
953 r = __kvm_set_memory_region(kvm, mem);
954 mutex_unlock(&kvm->slots_lock);
957 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
959 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
960 struct kvm_userspace_memory_region *mem)
962 if (mem->slot >= KVM_USER_MEM_SLOTS)
964 return kvm_set_memory_region(kvm, mem);
967 int kvm_get_dirty_log(struct kvm *kvm,
968 struct kvm_dirty_log *log, int *is_dirty)
970 struct kvm_memory_slot *memslot;
973 unsigned long any = 0;
976 if (log->slot >= KVM_USER_MEM_SLOTS)
979 memslot = id_to_memslot(kvm->memslots, log->slot);
981 if (!memslot->dirty_bitmap)
984 n = kvm_dirty_bitmap_bytes(memslot);
986 for (i = 0; !any && i < n/sizeof(long); ++i)
987 any = memslot->dirty_bitmap[i];
990 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1000 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1002 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1004 * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages
1005 * are dirty write protect them for next write.
1006 * @kvm: pointer to kvm instance
1007 * @log: slot id and address to which we copy the log
1008 * @is_dirty: flag set if any page is dirty
1010 * We need to keep it in mind that VCPU threads can write to the bitmap
1011 * concurrently. So, to avoid losing track of dirty pages we keep the
1014 * 1. Take a snapshot of the bit and clear it if needed.
1015 * 2. Write protect the corresponding page.
1016 * 3. Copy the snapshot to the userspace.
1017 * 4. Upon return caller flushes TLB's if needed.
1019 * Between 2 and 4, the guest may write to the page using the remaining TLB
1020 * entry. This is not a problem because the page is reported dirty using
1021 * the snapshot taken before and step 4 ensures that writes done after
1022 * exiting to userspace will be logged for the next call.
1025 int kvm_get_dirty_log_protect(struct kvm *kvm,
1026 struct kvm_dirty_log *log, bool *is_dirty)
1028 struct kvm_memory_slot *memslot;
1031 unsigned long *dirty_bitmap;
1032 unsigned long *dirty_bitmap_buffer;
1035 if (log->slot >= KVM_USER_MEM_SLOTS)
1038 memslot = id_to_memslot(kvm->memslots, log->slot);
1040 dirty_bitmap = memslot->dirty_bitmap;
1045 n = kvm_dirty_bitmap_bytes(memslot);
1047 dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long);
1048 memset(dirty_bitmap_buffer, 0, n);
1050 spin_lock(&kvm->mmu_lock);
1052 for (i = 0; i < n / sizeof(long); i++) {
1056 if (!dirty_bitmap[i])
1061 mask = xchg(&dirty_bitmap[i], 0);
1062 dirty_bitmap_buffer[i] = mask;
1064 offset = i * BITS_PER_LONG;
1065 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot, offset,
1069 spin_unlock(&kvm->mmu_lock);
1072 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1079 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect);
1082 bool kvm_largepages_enabled(void)
1084 return largepages_enabled;
1087 void kvm_disable_largepages(void)
1089 largepages_enabled = false;
1091 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1093 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1095 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1097 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1099 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1101 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1103 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1104 memslot->flags & KVM_MEMSLOT_INVALID)
1109 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1111 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1113 struct vm_area_struct *vma;
1114 unsigned long addr, size;
1118 addr = gfn_to_hva(kvm, gfn);
1119 if (kvm_is_error_hva(addr))
1122 down_read(¤t->mm->mmap_sem);
1123 vma = find_vma(current->mm, addr);
1127 size = vma_kernel_pagesize(vma);
1130 up_read(¤t->mm->mmap_sem);
1135 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1137 return slot->flags & KVM_MEM_READONLY;
1140 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1141 gfn_t *nr_pages, bool write)
1143 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1144 return KVM_HVA_ERR_BAD;
1146 if (memslot_is_readonly(slot) && write)
1147 return KVM_HVA_ERR_RO_BAD;
1150 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1152 return __gfn_to_hva_memslot(slot, gfn);
1155 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1158 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1161 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1164 return gfn_to_hva_many(slot, gfn, NULL);
1166 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1168 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1170 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1172 EXPORT_SYMBOL_GPL(gfn_to_hva);
1175 * If writable is set to false, the hva returned by this function is only
1176 * allowed to be read.
1178 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1179 gfn_t gfn, bool *writable)
1181 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1183 if (!kvm_is_error_hva(hva) && writable)
1184 *writable = !memslot_is_readonly(slot);
1189 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1191 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1193 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1196 static int kvm_read_hva(void *data, void __user *hva, int len)
1198 return __copy_from_user(data, hva, len);
1201 static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1203 return __copy_from_user_inatomic(data, hva, len);
1206 static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1207 unsigned long start, int write, struct page **page)
1209 int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1212 flags |= FOLL_WRITE;
1214 return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1217 static inline int check_user_page_hwpoison(unsigned long addr)
1219 int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1221 rc = __get_user_pages(current, current->mm, addr, 1,
1222 flags, NULL, NULL, NULL);
1223 return rc == -EHWPOISON;
1227 * The atomic path to get the writable pfn which will be stored in @pfn,
1228 * true indicates success, otherwise false is returned.
1230 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1231 bool write_fault, bool *writable, pfn_t *pfn)
1233 struct page *page[1];
1236 if (!(async || atomic))
1240 * Fast pin a writable pfn only if it is a write fault request
1241 * or the caller allows to map a writable pfn for a read fault
1244 if (!(write_fault || writable))
1247 npages = __get_user_pages_fast(addr, 1, 1, page);
1249 *pfn = page_to_pfn(page[0]);
1260 * The slow path to get the pfn of the specified host virtual address,
1261 * 1 indicates success, -errno is returned if error is detected.
1263 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1264 bool *writable, pfn_t *pfn)
1266 struct page *page[1];
1272 *writable = write_fault;
1275 down_read(¤t->mm->mmap_sem);
1276 npages = get_user_page_nowait(current, current->mm,
1277 addr, write_fault, page);
1278 up_read(¤t->mm->mmap_sem);
1280 npages = __get_user_pages_unlocked(current, current->mm, addr, 1,
1281 write_fault, 0, page,
1282 FOLL_TOUCH|FOLL_HWPOISON);
1286 /* map read fault as writable if possible */
1287 if (unlikely(!write_fault) && writable) {
1288 struct page *wpage[1];
1290 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1299 *pfn = page_to_pfn(page[0]);
1303 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1305 if (unlikely(!(vma->vm_flags & VM_READ)))
1308 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1315 * Pin guest page in memory and return its pfn.
1316 * @addr: host virtual address which maps memory to the guest
1317 * @atomic: whether this function can sleep
1318 * @async: whether this function need to wait IO complete if the
1319 * host page is not in the memory
1320 * @write_fault: whether we should get a writable host page
1321 * @writable: whether it allows to map a writable host page for !@write_fault
1323 * The function will map a writable host page for these two cases:
1324 * 1): @write_fault = true
1325 * 2): @write_fault = false && @writable, @writable will tell the caller
1326 * whether the mapping is writable.
1328 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1329 bool write_fault, bool *writable)
1331 struct vm_area_struct *vma;
1335 /* we can do it either atomically or asynchronously, not both */
1336 BUG_ON(atomic && async);
1338 if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1342 return KVM_PFN_ERR_FAULT;
1344 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1348 down_read(¤t->mm->mmap_sem);
1349 if (npages == -EHWPOISON ||
1350 (!async && check_user_page_hwpoison(addr))) {
1351 pfn = KVM_PFN_ERR_HWPOISON;
1355 vma = find_vma_intersection(current->mm, addr, addr + 1);
1358 pfn = KVM_PFN_ERR_FAULT;
1359 else if ((vma->vm_flags & VM_PFNMAP)) {
1360 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1362 BUG_ON(!kvm_is_reserved_pfn(pfn));
1364 if (async && vma_is_valid(vma, write_fault))
1366 pfn = KVM_PFN_ERR_FAULT;
1369 up_read(¤t->mm->mmap_sem);
1374 __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1375 bool *async, bool write_fault, bool *writable)
1377 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1379 if (addr == KVM_HVA_ERR_RO_BAD)
1380 return KVM_PFN_ERR_RO_FAULT;
1382 if (kvm_is_error_hva(addr))
1383 return KVM_PFN_NOSLOT;
1385 /* Do not map writable pfn in the readonly memslot. */
1386 if (writable && memslot_is_readonly(slot)) {
1391 return hva_to_pfn(addr, atomic, async, write_fault,
1395 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1396 bool write_fault, bool *writable)
1398 struct kvm_memory_slot *slot;
1403 slot = gfn_to_memslot(kvm, gfn);
1405 return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1409 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1411 return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1413 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1415 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1416 bool write_fault, bool *writable)
1418 return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1420 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1422 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1424 return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1426 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1428 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1431 return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1433 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1435 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1437 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1440 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1442 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1444 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1446 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1452 addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1453 if (kvm_is_error_hva(addr))
1456 if (entry < nr_pages)
1459 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1461 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1463 static struct page *kvm_pfn_to_page(pfn_t pfn)
1465 if (is_error_noslot_pfn(pfn))
1466 return KVM_ERR_PTR_BAD_PAGE;
1468 if (kvm_is_reserved_pfn(pfn)) {
1470 return KVM_ERR_PTR_BAD_PAGE;
1473 return pfn_to_page(pfn);
1476 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1480 pfn = gfn_to_pfn(kvm, gfn);
1482 return kvm_pfn_to_page(pfn);
1485 EXPORT_SYMBOL_GPL(gfn_to_page);
1487 void kvm_release_page_clean(struct page *page)
1489 WARN_ON(is_error_page(page));
1491 kvm_release_pfn_clean(page_to_pfn(page));
1493 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1495 void kvm_release_pfn_clean(pfn_t pfn)
1497 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
1498 put_page(pfn_to_page(pfn));
1500 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1502 void kvm_release_page_dirty(struct page *page)
1504 WARN_ON(is_error_page(page));
1506 kvm_release_pfn_dirty(page_to_pfn(page));
1508 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1510 static void kvm_release_pfn_dirty(pfn_t pfn)
1512 kvm_set_pfn_dirty(pfn);
1513 kvm_release_pfn_clean(pfn);
1516 void kvm_set_pfn_dirty(pfn_t pfn)
1518 if (!kvm_is_reserved_pfn(pfn)) {
1519 struct page *page = pfn_to_page(pfn);
1520 if (!PageReserved(page))
1524 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1526 void kvm_set_pfn_accessed(pfn_t pfn)
1528 if (!kvm_is_reserved_pfn(pfn))
1529 mark_page_accessed(pfn_to_page(pfn));
1531 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1533 void kvm_get_pfn(pfn_t pfn)
1535 if (!kvm_is_reserved_pfn(pfn))
1536 get_page(pfn_to_page(pfn));
1538 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1540 static int next_segment(unsigned long len, int offset)
1542 if (len > PAGE_SIZE - offset)
1543 return PAGE_SIZE - offset;
1548 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1554 addr = gfn_to_hva_prot(kvm, gfn, NULL);
1555 if (kvm_is_error_hva(addr))
1557 r = kvm_read_hva(data, (void __user *)addr + offset, len);
1562 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1564 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1566 gfn_t gfn = gpa >> PAGE_SHIFT;
1568 int offset = offset_in_page(gpa);
1571 while ((seg = next_segment(len, offset)) != 0) {
1572 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1582 EXPORT_SYMBOL_GPL(kvm_read_guest);
1584 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1589 gfn_t gfn = gpa >> PAGE_SHIFT;
1590 int offset = offset_in_page(gpa);
1592 addr = gfn_to_hva_prot(kvm, gfn, NULL);
1593 if (kvm_is_error_hva(addr))
1595 pagefault_disable();
1596 r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
1602 EXPORT_SYMBOL(kvm_read_guest_atomic);
1604 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1605 int offset, int len)
1610 addr = gfn_to_hva(kvm, gfn);
1611 if (kvm_is_error_hva(addr))
1613 r = __copy_to_user((void __user *)addr + offset, data, len);
1616 mark_page_dirty(kvm, gfn);
1619 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1621 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1624 gfn_t gfn = gpa >> PAGE_SHIFT;
1626 int offset = offset_in_page(gpa);
1629 while ((seg = next_segment(len, offset)) != 0) {
1630 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1640 EXPORT_SYMBOL_GPL(kvm_write_guest);
1642 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1643 gpa_t gpa, unsigned long len)
1645 struct kvm_memslots *slots = kvm_memslots(kvm);
1646 int offset = offset_in_page(gpa);
1647 gfn_t start_gfn = gpa >> PAGE_SHIFT;
1648 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1649 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1650 gfn_t nr_pages_avail;
1653 ghc->generation = slots->generation;
1655 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1656 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, &nr_pages_avail);
1657 if (!kvm_is_error_hva(ghc->hva) && nr_pages_avail >= nr_pages_needed) {
1661 * If the requested region crosses two memslots, we still
1662 * verify that the entire region is valid here.
1664 while (start_gfn <= end_gfn) {
1665 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1666 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1668 if (kvm_is_error_hva(ghc->hva))
1670 start_gfn += nr_pages_avail;
1672 /* Use the slow path for cross page reads and writes. */
1673 ghc->memslot = NULL;
1677 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1679 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1680 void *data, unsigned long len)
1682 struct kvm_memslots *slots = kvm_memslots(kvm);
1685 BUG_ON(len > ghc->len);
1687 if (slots->generation != ghc->generation)
1688 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1690 if (unlikely(!ghc->memslot))
1691 return kvm_write_guest(kvm, ghc->gpa, data, len);
1693 if (kvm_is_error_hva(ghc->hva))
1696 r = __copy_to_user((void __user *)ghc->hva, data, len);
1699 mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1703 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1705 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1706 void *data, unsigned long len)
1708 struct kvm_memslots *slots = kvm_memslots(kvm);
1711 BUG_ON(len > ghc->len);
1713 if (slots->generation != ghc->generation)
1714 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1716 if (unlikely(!ghc->memslot))
1717 return kvm_read_guest(kvm, ghc->gpa, data, len);
1719 if (kvm_is_error_hva(ghc->hva))
1722 r = __copy_from_user(data, (void __user *)ghc->hva, len);
1728 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1730 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1732 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
1734 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
1736 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1738 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1740 gfn_t gfn = gpa >> PAGE_SHIFT;
1742 int offset = offset_in_page(gpa);
1745 while ((seg = next_segment(len, offset)) != 0) {
1746 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1755 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1757 static void mark_page_dirty_in_slot(struct kvm *kvm,
1758 struct kvm_memory_slot *memslot,
1761 if (memslot && memslot->dirty_bitmap) {
1762 unsigned long rel_gfn = gfn - memslot->base_gfn;
1764 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1768 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1770 struct kvm_memory_slot *memslot;
1772 memslot = gfn_to_memslot(kvm, gfn);
1773 mark_page_dirty_in_slot(kvm, memslot, gfn);
1775 EXPORT_SYMBOL_GPL(mark_page_dirty);
1777 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
1779 if (kvm_arch_vcpu_runnable(vcpu)) {
1780 kvm_make_request(KVM_REQ_UNHALT, vcpu);
1783 if (kvm_cpu_has_pending_timer(vcpu))
1785 if (signal_pending(current))
1792 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1794 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1798 bool waited = false;
1800 start = cur = ktime_get();
1802 ktime_t stop = ktime_add_ns(ktime_get(), halt_poll_ns);
1805 * This sets KVM_REQ_UNHALT if an interrupt
1808 if (kvm_vcpu_check_block(vcpu) < 0) {
1809 ++vcpu->stat.halt_successful_poll;
1813 } while (single_task_running() && ktime_before(cur, stop));
1817 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1819 if (kvm_vcpu_check_block(vcpu) < 0)
1826 finish_wait(&vcpu->wq, &wait);
1830 trace_kvm_vcpu_wakeup(ktime_to_ns(cur) - ktime_to_ns(start), waited);
1832 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
1836 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1838 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1841 int cpu = vcpu->cpu;
1842 wait_queue_head_t *wqp;
1844 wqp = kvm_arch_vcpu_wq(vcpu);
1845 if (waitqueue_active(wqp)) {
1846 wake_up_interruptible(wqp);
1847 ++vcpu->stat.halt_wakeup;
1851 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1852 if (kvm_arch_vcpu_should_kick(vcpu))
1853 smp_send_reschedule(cpu);
1856 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
1857 #endif /* !CONFIG_S390 */
1859 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
1862 struct task_struct *task = NULL;
1866 pid = rcu_dereference(target->pid);
1868 task = get_pid_task(pid, PIDTYPE_PID);
1872 ret = yield_to(task, 1);
1873 put_task_struct(task);
1877 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1880 * Helper that checks whether a VCPU is eligible for directed yield.
1881 * Most eligible candidate to yield is decided by following heuristics:
1883 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1884 * (preempted lock holder), indicated by @in_spin_loop.
1885 * Set at the beiginning and cleared at the end of interception/PLE handler.
1887 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1888 * chance last time (mostly it has become eligible now since we have probably
1889 * yielded to lockholder in last iteration. This is done by toggling
1890 * @dy_eligible each time a VCPU checked for eligibility.)
1892 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1893 * to preempted lock-holder could result in wrong VCPU selection and CPU
1894 * burning. Giving priority for a potential lock-holder increases lock
1897 * Since algorithm is based on heuristics, accessing another VCPU data without
1898 * locking does not harm. It may result in trying to yield to same VCPU, fail
1899 * and continue with next VCPU and so on.
1901 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1903 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1906 eligible = !vcpu->spin_loop.in_spin_loop ||
1907 vcpu->spin_loop.dy_eligible;
1909 if (vcpu->spin_loop.in_spin_loop)
1910 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1918 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1920 struct kvm *kvm = me->kvm;
1921 struct kvm_vcpu *vcpu;
1922 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1928 kvm_vcpu_set_in_spin_loop(me, true);
1930 * We boost the priority of a VCPU that is runnable but not
1931 * currently running, because it got preempted by something
1932 * else and called schedule in __vcpu_run. Hopefully that
1933 * VCPU is holding the lock that we need and will release it.
1934 * We approximate round-robin by starting at the last boosted VCPU.
1936 for (pass = 0; pass < 2 && !yielded && try; pass++) {
1937 kvm_for_each_vcpu(i, vcpu, kvm) {
1938 if (!pass && i <= last_boosted_vcpu) {
1939 i = last_boosted_vcpu;
1941 } else if (pass && i > last_boosted_vcpu)
1943 if (!ACCESS_ONCE(vcpu->preempted))
1947 if (waitqueue_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
1949 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1952 yielded = kvm_vcpu_yield_to(vcpu);
1954 kvm->last_boosted_vcpu = i;
1956 } else if (yielded < 0) {
1963 kvm_vcpu_set_in_spin_loop(me, false);
1965 /* Ensure vcpu is not eligible during next spinloop */
1966 kvm_vcpu_set_dy_eligible(me, false);
1968 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1970 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1972 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1975 if (vmf->pgoff == 0)
1976 page = virt_to_page(vcpu->run);
1978 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1979 page = virt_to_page(vcpu->arch.pio_data);
1981 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1982 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1983 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1986 return kvm_arch_vcpu_fault(vcpu, vmf);
1992 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1993 .fault = kvm_vcpu_fault,
1996 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1998 vma->vm_ops = &kvm_vcpu_vm_ops;
2002 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2004 struct kvm_vcpu *vcpu = filp->private_data;
2006 kvm_put_kvm(vcpu->kvm);
2010 static struct file_operations kvm_vcpu_fops = {
2011 .release = kvm_vcpu_release,
2012 .unlocked_ioctl = kvm_vcpu_ioctl,
2013 #ifdef CONFIG_KVM_COMPAT
2014 .compat_ioctl = kvm_vcpu_compat_ioctl,
2016 .mmap = kvm_vcpu_mmap,
2017 .llseek = noop_llseek,
2021 * Allocates an inode for the vcpu.
2023 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2025 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
2029 * Creates some virtual cpus. Good luck creating more than one.
2031 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
2034 struct kvm_vcpu *vcpu, *v;
2036 if (id >= KVM_MAX_VCPUS)
2039 vcpu = kvm_arch_vcpu_create(kvm, id);
2041 return PTR_ERR(vcpu);
2043 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2045 r = kvm_arch_vcpu_setup(vcpu);
2049 mutex_lock(&kvm->lock);
2050 if (!kvm_vcpu_compatible(vcpu)) {
2052 goto unlock_vcpu_destroy;
2054 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
2056 goto unlock_vcpu_destroy;
2059 kvm_for_each_vcpu(r, v, kvm)
2060 if (v->vcpu_id == id) {
2062 goto unlock_vcpu_destroy;
2065 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
2067 /* Now it's all set up, let userspace reach it */
2069 r = create_vcpu_fd(vcpu);
2072 goto unlock_vcpu_destroy;
2075 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
2077 atomic_inc(&kvm->online_vcpus);
2079 mutex_unlock(&kvm->lock);
2080 kvm_arch_vcpu_postcreate(vcpu);
2083 unlock_vcpu_destroy:
2084 mutex_unlock(&kvm->lock);
2086 kvm_arch_vcpu_destroy(vcpu);
2090 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2093 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2094 vcpu->sigset_active = 1;
2095 vcpu->sigset = *sigset;
2097 vcpu->sigset_active = 0;
2101 static long kvm_vcpu_ioctl(struct file *filp,
2102 unsigned int ioctl, unsigned long arg)
2104 struct kvm_vcpu *vcpu = filp->private_data;
2105 void __user *argp = (void __user *)arg;
2107 struct kvm_fpu *fpu = NULL;
2108 struct kvm_sregs *kvm_sregs = NULL;
2110 if (vcpu->kvm->mm != current->mm)
2113 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2116 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
2118 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
2119 * so vcpu_load() would break it.
2121 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
2122 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2126 r = vcpu_load(vcpu);
2134 if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
2135 /* The thread running this VCPU changed. */
2136 struct pid *oldpid = vcpu->pid;
2137 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
2138 rcu_assign_pointer(vcpu->pid, newpid);
2143 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2144 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2146 case KVM_GET_REGS: {
2147 struct kvm_regs *kvm_regs;
2150 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2153 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2157 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2164 case KVM_SET_REGS: {
2165 struct kvm_regs *kvm_regs;
2168 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2169 if (IS_ERR(kvm_regs)) {
2170 r = PTR_ERR(kvm_regs);
2173 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2177 case KVM_GET_SREGS: {
2178 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2182 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2186 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2191 case KVM_SET_SREGS: {
2192 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2193 if (IS_ERR(kvm_sregs)) {
2194 r = PTR_ERR(kvm_sregs);
2198 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2201 case KVM_GET_MP_STATE: {
2202 struct kvm_mp_state mp_state;
2204 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2208 if (copy_to_user(argp, &mp_state, sizeof mp_state))
2213 case KVM_SET_MP_STATE: {
2214 struct kvm_mp_state mp_state;
2217 if (copy_from_user(&mp_state, argp, sizeof mp_state))
2219 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2222 case KVM_TRANSLATE: {
2223 struct kvm_translation tr;
2226 if (copy_from_user(&tr, argp, sizeof tr))
2228 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2232 if (copy_to_user(argp, &tr, sizeof tr))
2237 case KVM_SET_GUEST_DEBUG: {
2238 struct kvm_guest_debug dbg;
2241 if (copy_from_user(&dbg, argp, sizeof dbg))
2243 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2246 case KVM_SET_SIGNAL_MASK: {
2247 struct kvm_signal_mask __user *sigmask_arg = argp;
2248 struct kvm_signal_mask kvm_sigmask;
2249 sigset_t sigset, *p;
2254 if (copy_from_user(&kvm_sigmask, argp,
2255 sizeof kvm_sigmask))
2258 if (kvm_sigmask.len != sizeof sigset)
2261 if (copy_from_user(&sigset, sigmask_arg->sigset,
2266 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2270 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2274 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2278 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2284 fpu = memdup_user(argp, sizeof(*fpu));
2290 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2294 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2303 #ifdef CONFIG_KVM_COMPAT
2304 static long kvm_vcpu_compat_ioctl(struct file *filp,
2305 unsigned int ioctl, unsigned long arg)
2307 struct kvm_vcpu *vcpu = filp->private_data;
2308 void __user *argp = compat_ptr(arg);
2311 if (vcpu->kvm->mm != current->mm)
2315 case KVM_SET_SIGNAL_MASK: {
2316 struct kvm_signal_mask __user *sigmask_arg = argp;
2317 struct kvm_signal_mask kvm_sigmask;
2318 compat_sigset_t csigset;
2323 if (copy_from_user(&kvm_sigmask, argp,
2324 sizeof kvm_sigmask))
2327 if (kvm_sigmask.len != sizeof csigset)
2330 if (copy_from_user(&csigset, sigmask_arg->sigset,
2333 sigset_from_compat(&sigset, &csigset);
2334 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2336 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2340 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2348 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2349 int (*accessor)(struct kvm_device *dev,
2350 struct kvm_device_attr *attr),
2353 struct kvm_device_attr attr;
2358 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2361 return accessor(dev, &attr);
2364 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2367 struct kvm_device *dev = filp->private_data;
2370 case KVM_SET_DEVICE_ATTR:
2371 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2372 case KVM_GET_DEVICE_ATTR:
2373 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2374 case KVM_HAS_DEVICE_ATTR:
2375 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2377 if (dev->ops->ioctl)
2378 return dev->ops->ioctl(dev, ioctl, arg);
2384 static int kvm_device_release(struct inode *inode, struct file *filp)
2386 struct kvm_device *dev = filp->private_data;
2387 struct kvm *kvm = dev->kvm;
2393 static const struct file_operations kvm_device_fops = {
2394 .unlocked_ioctl = kvm_device_ioctl,
2395 #ifdef CONFIG_KVM_COMPAT
2396 .compat_ioctl = kvm_device_ioctl,
2398 .release = kvm_device_release,
2401 struct kvm_device *kvm_device_from_filp(struct file *filp)
2403 if (filp->f_op != &kvm_device_fops)
2406 return filp->private_data;
2409 static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
2410 #ifdef CONFIG_KVM_MPIC
2411 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
2412 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
2415 #ifdef CONFIG_KVM_XICS
2416 [KVM_DEV_TYPE_XICS] = &kvm_xics_ops,
2420 int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
2422 if (type >= ARRAY_SIZE(kvm_device_ops_table))
2425 if (kvm_device_ops_table[type] != NULL)
2428 kvm_device_ops_table[type] = ops;
2432 void kvm_unregister_device_ops(u32 type)
2434 if (kvm_device_ops_table[type] != NULL)
2435 kvm_device_ops_table[type] = NULL;
2438 static int kvm_ioctl_create_device(struct kvm *kvm,
2439 struct kvm_create_device *cd)
2441 struct kvm_device_ops *ops = NULL;
2442 struct kvm_device *dev;
2443 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2446 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
2449 ops = kvm_device_ops_table[cd->type];
2456 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2463 ret = ops->create(dev, cd->type);
2469 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2475 list_add(&dev->vm_node, &kvm->devices);
2481 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2484 case KVM_CAP_USER_MEMORY:
2485 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2486 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2487 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2488 case KVM_CAP_SET_BOOT_CPU_ID:
2490 case KVM_CAP_INTERNAL_ERROR_DATA:
2491 #ifdef CONFIG_HAVE_KVM_MSI
2492 case KVM_CAP_SIGNAL_MSI:
2494 #ifdef CONFIG_HAVE_KVM_IRQFD
2496 case KVM_CAP_IRQFD_RESAMPLE:
2498 case KVM_CAP_CHECK_EXTENSION_VM:
2500 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2501 case KVM_CAP_IRQ_ROUTING:
2502 return KVM_MAX_IRQ_ROUTES;
2507 return kvm_vm_ioctl_check_extension(kvm, arg);
2510 static long kvm_vm_ioctl(struct file *filp,
2511 unsigned int ioctl, unsigned long arg)
2513 struct kvm *kvm = filp->private_data;
2514 void __user *argp = (void __user *)arg;
2517 if (kvm->mm != current->mm)
2520 case KVM_CREATE_VCPU:
2521 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2523 case KVM_SET_USER_MEMORY_REGION: {
2524 struct kvm_userspace_memory_region kvm_userspace_mem;
2527 if (copy_from_user(&kvm_userspace_mem, argp,
2528 sizeof kvm_userspace_mem))
2531 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2534 case KVM_GET_DIRTY_LOG: {
2535 struct kvm_dirty_log log;
2538 if (copy_from_user(&log, argp, sizeof log))
2540 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2543 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2544 case KVM_REGISTER_COALESCED_MMIO: {
2545 struct kvm_coalesced_mmio_zone zone;
2547 if (copy_from_user(&zone, argp, sizeof zone))
2549 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2552 case KVM_UNREGISTER_COALESCED_MMIO: {
2553 struct kvm_coalesced_mmio_zone zone;
2555 if (copy_from_user(&zone, argp, sizeof zone))
2557 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2562 struct kvm_irqfd data;
2565 if (copy_from_user(&data, argp, sizeof data))
2567 r = kvm_irqfd(kvm, &data);
2570 case KVM_IOEVENTFD: {
2571 struct kvm_ioeventfd data;
2574 if (copy_from_user(&data, argp, sizeof data))
2576 r = kvm_ioeventfd(kvm, &data);
2579 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2580 case KVM_SET_BOOT_CPU_ID:
2582 mutex_lock(&kvm->lock);
2583 if (atomic_read(&kvm->online_vcpus) != 0)
2586 kvm->bsp_vcpu_id = arg;
2587 mutex_unlock(&kvm->lock);
2590 #ifdef CONFIG_HAVE_KVM_MSI
2591 case KVM_SIGNAL_MSI: {
2595 if (copy_from_user(&msi, argp, sizeof msi))
2597 r = kvm_send_userspace_msi(kvm, &msi);
2601 #ifdef __KVM_HAVE_IRQ_LINE
2602 case KVM_IRQ_LINE_STATUS:
2603 case KVM_IRQ_LINE: {
2604 struct kvm_irq_level irq_event;
2607 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2610 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
2611 ioctl == KVM_IRQ_LINE_STATUS);
2616 if (ioctl == KVM_IRQ_LINE_STATUS) {
2617 if (copy_to_user(argp, &irq_event, sizeof irq_event))
2625 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2626 case KVM_SET_GSI_ROUTING: {
2627 struct kvm_irq_routing routing;
2628 struct kvm_irq_routing __user *urouting;
2629 struct kvm_irq_routing_entry *entries;
2632 if (copy_from_user(&routing, argp, sizeof(routing)))
2635 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2640 entries = vmalloc(routing.nr * sizeof(*entries));
2645 if (copy_from_user(entries, urouting->entries,
2646 routing.nr * sizeof(*entries)))
2647 goto out_free_irq_routing;
2648 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2650 out_free_irq_routing:
2654 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
2655 case KVM_CREATE_DEVICE: {
2656 struct kvm_create_device cd;
2659 if (copy_from_user(&cd, argp, sizeof(cd)))
2662 r = kvm_ioctl_create_device(kvm, &cd);
2667 if (copy_to_user(argp, &cd, sizeof(cd)))
2673 case KVM_CHECK_EXTENSION:
2674 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
2677 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2683 #ifdef CONFIG_KVM_COMPAT
2684 struct compat_kvm_dirty_log {
2688 compat_uptr_t dirty_bitmap; /* one bit per page */
2693 static long kvm_vm_compat_ioctl(struct file *filp,
2694 unsigned int ioctl, unsigned long arg)
2696 struct kvm *kvm = filp->private_data;
2699 if (kvm->mm != current->mm)
2702 case KVM_GET_DIRTY_LOG: {
2703 struct compat_kvm_dirty_log compat_log;
2704 struct kvm_dirty_log log;
2707 if (copy_from_user(&compat_log, (void __user *)arg,
2708 sizeof(compat_log)))
2710 log.slot = compat_log.slot;
2711 log.padding1 = compat_log.padding1;
2712 log.padding2 = compat_log.padding2;
2713 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2715 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2719 r = kvm_vm_ioctl(filp, ioctl, arg);
2727 static struct file_operations kvm_vm_fops = {
2728 .release = kvm_vm_release,
2729 .unlocked_ioctl = kvm_vm_ioctl,
2730 #ifdef CONFIG_KVM_COMPAT
2731 .compat_ioctl = kvm_vm_compat_ioctl,
2733 .llseek = noop_llseek,
2736 static int kvm_dev_ioctl_create_vm(unsigned long type)
2741 kvm = kvm_create_vm(type);
2743 return PTR_ERR(kvm);
2744 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2745 r = kvm_coalesced_mmio_init(kvm);
2751 r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC);
2758 static long kvm_dev_ioctl(struct file *filp,
2759 unsigned int ioctl, unsigned long arg)
2764 case KVM_GET_API_VERSION:
2767 r = KVM_API_VERSION;
2770 r = kvm_dev_ioctl_create_vm(arg);
2772 case KVM_CHECK_EXTENSION:
2773 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
2775 case KVM_GET_VCPU_MMAP_SIZE:
2778 r = PAGE_SIZE; /* struct kvm_run */
2780 r += PAGE_SIZE; /* pio data page */
2782 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2783 r += PAGE_SIZE; /* coalesced mmio ring page */
2786 case KVM_TRACE_ENABLE:
2787 case KVM_TRACE_PAUSE:
2788 case KVM_TRACE_DISABLE:
2792 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2798 static struct file_operations kvm_chardev_ops = {
2799 .unlocked_ioctl = kvm_dev_ioctl,
2800 .compat_ioctl = kvm_dev_ioctl,
2801 .llseek = noop_llseek,
2804 static struct miscdevice kvm_dev = {
2810 static void hardware_enable_nolock(void *junk)
2812 int cpu = raw_smp_processor_id();
2815 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2818 cpumask_set_cpu(cpu, cpus_hardware_enabled);
2820 r = kvm_arch_hardware_enable();
2823 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2824 atomic_inc(&hardware_enable_failed);
2825 printk(KERN_INFO "kvm: enabling virtualization on "
2826 "CPU%d failed\n", cpu);
2830 static void hardware_enable(void)
2832 raw_spin_lock(&kvm_count_lock);
2833 if (kvm_usage_count)
2834 hardware_enable_nolock(NULL);
2835 raw_spin_unlock(&kvm_count_lock);
2838 static void hardware_disable_nolock(void *junk)
2840 int cpu = raw_smp_processor_id();
2842 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2844 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2845 kvm_arch_hardware_disable();
2848 static void hardware_disable(void)
2850 raw_spin_lock(&kvm_count_lock);
2851 if (kvm_usage_count)
2852 hardware_disable_nolock(NULL);
2853 raw_spin_unlock(&kvm_count_lock);
2856 static void hardware_disable_all_nolock(void)
2858 BUG_ON(!kvm_usage_count);
2861 if (!kvm_usage_count)
2862 on_each_cpu(hardware_disable_nolock, NULL, 1);
2865 static void hardware_disable_all(void)
2867 raw_spin_lock(&kvm_count_lock);
2868 hardware_disable_all_nolock();
2869 raw_spin_unlock(&kvm_count_lock);
2872 static int hardware_enable_all(void)
2876 raw_spin_lock(&kvm_count_lock);
2879 if (kvm_usage_count == 1) {
2880 atomic_set(&hardware_enable_failed, 0);
2881 on_each_cpu(hardware_enable_nolock, NULL, 1);
2883 if (atomic_read(&hardware_enable_failed)) {
2884 hardware_disable_all_nolock();
2889 raw_spin_unlock(&kvm_count_lock);
2894 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2899 val &= ~CPU_TASKS_FROZEN;
2902 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2907 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2915 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2919 * Some (well, at least mine) BIOSes hang on reboot if
2922 * And Intel TXT required VMX off for all cpu when system shutdown.
2924 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2925 kvm_rebooting = true;
2926 on_each_cpu(hardware_disable_nolock, NULL, 1);
2930 static struct notifier_block kvm_reboot_notifier = {
2931 .notifier_call = kvm_reboot,
2935 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2939 for (i = 0; i < bus->dev_count; i++) {
2940 struct kvm_io_device *pos = bus->range[i].dev;
2942 kvm_iodevice_destructor(pos);
2947 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
2948 const struct kvm_io_range *r2)
2950 if (r1->addr < r2->addr)
2952 if (r1->addr + r1->len > r2->addr + r2->len)
2957 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2959 return kvm_io_bus_cmp(p1, p2);
2962 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2963 gpa_t addr, int len)
2965 bus->range[bus->dev_count++] = (struct kvm_io_range) {
2971 sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2972 kvm_io_bus_sort_cmp, NULL);
2977 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2978 gpa_t addr, int len)
2980 struct kvm_io_range *range, key;
2983 key = (struct kvm_io_range) {
2988 range = bsearch(&key, bus->range, bus->dev_count,
2989 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2993 off = range - bus->range;
2995 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
3001 static int __kvm_io_bus_write(struct kvm_io_bus *bus,
3002 struct kvm_io_range *range, const void *val)
3006 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3010 while (idx < bus->dev_count &&
3011 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3012 if (!kvm_iodevice_write(bus->range[idx].dev, range->addr,
3021 /* kvm_io_bus_write - called under kvm->slots_lock */
3022 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3023 int len, const void *val)
3025 struct kvm_io_bus *bus;
3026 struct kvm_io_range range;
3029 range = (struct kvm_io_range) {
3034 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
3035 r = __kvm_io_bus_write(bus, &range, val);
3036 return r < 0 ? r : 0;
3039 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3040 int kvm_io_bus_write_cookie(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3041 int len, const void *val, long cookie)
3043 struct kvm_io_bus *bus;
3044 struct kvm_io_range range;
3046 range = (struct kvm_io_range) {
3051 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
3053 /* First try the device referenced by cookie. */
3054 if ((cookie >= 0) && (cookie < bus->dev_count) &&
3055 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
3056 if (!kvm_iodevice_write(bus->range[cookie].dev, addr, len,
3061 * cookie contained garbage; fall back to search and return the
3062 * correct cookie value.
3064 return __kvm_io_bus_write(bus, &range, val);
3067 static int __kvm_io_bus_read(struct kvm_io_bus *bus, struct kvm_io_range *range,
3072 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3076 while (idx < bus->dev_count &&
3077 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3078 if (!kvm_iodevice_read(bus->range[idx].dev, range->addr,
3086 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
3088 /* kvm_io_bus_read - called under kvm->slots_lock */
3089 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3092 struct kvm_io_bus *bus;
3093 struct kvm_io_range range;
3096 range = (struct kvm_io_range) {
3101 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
3102 r = __kvm_io_bus_read(bus, &range, val);
3103 return r < 0 ? r : 0;
3107 /* Caller must hold slots_lock. */
3108 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3109 int len, struct kvm_io_device *dev)
3111 struct kvm_io_bus *new_bus, *bus;
3113 bus = kvm->buses[bus_idx];
3114 /* exclude ioeventfd which is limited by maximum fd */
3115 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3118 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
3119 sizeof(struct kvm_io_range)), GFP_KERNEL);
3122 memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
3123 sizeof(struct kvm_io_range)));
3124 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
3125 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3126 synchronize_srcu_expedited(&kvm->srcu);
3132 /* Caller must hold slots_lock. */
3133 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3134 struct kvm_io_device *dev)
3137 struct kvm_io_bus *new_bus, *bus;
3139 bus = kvm->buses[bus_idx];
3141 for (i = 0; i < bus->dev_count; i++)
3142 if (bus->range[i].dev == dev) {
3150 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3151 sizeof(struct kvm_io_range)), GFP_KERNEL);
3155 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3156 new_bus->dev_count--;
3157 memcpy(new_bus->range + i, bus->range + i + 1,
3158 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3160 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3161 synchronize_srcu_expedited(&kvm->srcu);
3166 static struct notifier_block kvm_cpu_notifier = {
3167 .notifier_call = kvm_cpu_hotplug,
3170 static int vm_stat_get(void *_offset, u64 *val)
3172 unsigned offset = (long)_offset;
3176 spin_lock(&kvm_lock);
3177 list_for_each_entry(kvm, &vm_list, vm_list)
3178 *val += *(u32 *)((void *)kvm + offset);
3179 spin_unlock(&kvm_lock);
3183 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
3185 static int vcpu_stat_get(void *_offset, u64 *val)
3187 unsigned offset = (long)_offset;
3189 struct kvm_vcpu *vcpu;
3193 spin_lock(&kvm_lock);
3194 list_for_each_entry(kvm, &vm_list, vm_list)
3195 kvm_for_each_vcpu(i, vcpu, kvm)
3196 *val += *(u32 *)((void *)vcpu + offset);
3198 spin_unlock(&kvm_lock);
3202 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
3204 static const struct file_operations *stat_fops[] = {
3205 [KVM_STAT_VCPU] = &vcpu_stat_fops,
3206 [KVM_STAT_VM] = &vm_stat_fops,
3209 static int kvm_init_debug(void)
3212 struct kvm_stats_debugfs_item *p;
3214 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3215 if (kvm_debugfs_dir == NULL)
3218 for (p = debugfs_entries; p->name; ++p) {
3219 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
3220 (void *)(long)p->offset,
3221 stat_fops[p->kind]);
3222 if (p->dentry == NULL)
3229 debugfs_remove_recursive(kvm_debugfs_dir);
3234 static void kvm_exit_debug(void)
3236 struct kvm_stats_debugfs_item *p;
3238 for (p = debugfs_entries; p->name; ++p)
3239 debugfs_remove(p->dentry);
3240 debugfs_remove(kvm_debugfs_dir);
3243 static int kvm_suspend(void)
3245 if (kvm_usage_count)
3246 hardware_disable_nolock(NULL);
3250 static void kvm_resume(void)
3252 if (kvm_usage_count) {
3253 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3254 hardware_enable_nolock(NULL);
3258 static struct syscore_ops kvm_syscore_ops = {
3259 .suspend = kvm_suspend,
3260 .resume = kvm_resume,
3264 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3266 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3269 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3271 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3272 if (vcpu->preempted)
3273 vcpu->preempted = false;
3275 kvm_arch_sched_in(vcpu, cpu);
3277 kvm_arch_vcpu_load(vcpu, cpu);
3280 static void kvm_sched_out(struct preempt_notifier *pn,
3281 struct task_struct *next)
3283 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3285 if (current->state == TASK_RUNNING)
3286 vcpu->preempted = true;
3287 kvm_arch_vcpu_put(vcpu);
3290 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3291 struct module *module)
3296 r = kvm_arch_init(opaque);
3301 * kvm_arch_init makes sure there's at most one caller
3302 * for architectures that support multiple implementations,
3303 * like intel and amd on x86.
3304 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3305 * conflicts in case kvm is already setup for another implementation.
3307 r = kvm_irqfd_init();
3311 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
3316 r = kvm_arch_hardware_setup();
3320 for_each_online_cpu(cpu) {
3321 smp_call_function_single(cpu,
3322 kvm_arch_check_processor_compat,
3328 r = register_cpu_notifier(&kvm_cpu_notifier);
3331 register_reboot_notifier(&kvm_reboot_notifier);
3333 /* A kmem cache lets us meet the alignment requirements of fx_save. */
3335 vcpu_align = __alignof__(struct kvm_vcpu);
3336 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
3338 if (!kvm_vcpu_cache) {
3343 r = kvm_async_pf_init();
3347 kvm_chardev_ops.owner = module;
3348 kvm_vm_fops.owner = module;
3349 kvm_vcpu_fops.owner = module;
3351 r = misc_register(&kvm_dev);
3353 printk(KERN_ERR "kvm: misc device register failed\n");
3357 register_syscore_ops(&kvm_syscore_ops);
3359 kvm_preempt_ops.sched_in = kvm_sched_in;
3360 kvm_preempt_ops.sched_out = kvm_sched_out;
3362 r = kvm_init_debug();
3364 printk(KERN_ERR "kvm: create debugfs files failed\n");
3368 r = kvm_vfio_ops_init();
3374 unregister_syscore_ops(&kvm_syscore_ops);
3375 misc_deregister(&kvm_dev);
3377 kvm_async_pf_deinit();
3379 kmem_cache_destroy(kvm_vcpu_cache);
3381 unregister_reboot_notifier(&kvm_reboot_notifier);
3382 unregister_cpu_notifier(&kvm_cpu_notifier);
3385 kvm_arch_hardware_unsetup();
3387 free_cpumask_var(cpus_hardware_enabled);
3395 EXPORT_SYMBOL_GPL(kvm_init);
3400 misc_deregister(&kvm_dev);
3401 kmem_cache_destroy(kvm_vcpu_cache);
3402 kvm_async_pf_deinit();
3403 unregister_syscore_ops(&kvm_syscore_ops);
3404 unregister_reboot_notifier(&kvm_reboot_notifier);
3405 unregister_cpu_notifier(&kvm_cpu_notifier);
3406 on_each_cpu(hardware_disable_nolock, NULL, 1);
3407 kvm_arch_hardware_unsetup();
3410 free_cpumask_var(cpus_hardware_enabled);
3411 kvm_vfio_ops_exit();
3413 EXPORT_SYMBOL_GPL(kvm_exit);