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.
19 #include <kvm/iodev.h>
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 static unsigned int halt_poll_ns;
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;
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;
92 EXPORT_SYMBOL_GPL(kvm_debugfs_dir);
94 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
96 #ifdef CONFIG_KVM_COMPAT
97 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
100 static int hardware_enable_all(void);
101 static void hardware_disable_all(void);
103 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
105 static void kvm_release_pfn_dirty(pfn_t pfn);
106 static void mark_page_dirty_in_slot(struct kvm *kvm,
107 struct kvm_memory_slot *memslot, gfn_t gfn);
109 __visible bool kvm_rebooting;
110 EXPORT_SYMBOL_GPL(kvm_rebooting);
112 static bool largepages_enabled = true;
114 bool kvm_is_reserved_pfn(pfn_t pfn)
117 return PageReserved(pfn_to_page(pfn));
123 * Switches to specified vcpu, until a matching vcpu_put()
125 int vcpu_load(struct kvm_vcpu *vcpu)
129 if (mutex_lock_killable(&vcpu->mutex))
132 preempt_notifier_register(&vcpu->preempt_notifier);
133 kvm_arch_vcpu_load(vcpu, cpu);
138 void vcpu_put(struct kvm_vcpu *vcpu)
141 kvm_arch_vcpu_put(vcpu);
142 preempt_notifier_unregister(&vcpu->preempt_notifier);
144 mutex_unlock(&vcpu->mutex);
147 static void ack_flush(void *_completed)
151 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
156 struct kvm_vcpu *vcpu;
158 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
161 kvm_for_each_vcpu(i, vcpu, kvm) {
162 kvm_make_request(req, vcpu);
165 /* Set ->requests bit before we read ->mode */
168 if (cpus != NULL && cpu != -1 && cpu != me &&
169 kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
170 cpumask_set_cpu(cpu, cpus);
172 if (unlikely(cpus == NULL))
173 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
174 else if (!cpumask_empty(cpus))
175 smp_call_function_many(cpus, ack_flush, NULL, 1);
179 free_cpumask_var(cpus);
183 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
184 void kvm_flush_remote_tlbs(struct kvm *kvm)
186 long dirty_count = kvm->tlbs_dirty;
189 if (kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
190 ++kvm->stat.remote_tlb_flush;
191 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
193 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
196 void kvm_reload_remote_mmus(struct kvm *kvm)
198 kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
201 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
203 kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
206 void kvm_make_scan_ioapic_request(struct kvm *kvm)
208 kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
211 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
216 mutex_init(&vcpu->mutex);
221 init_waitqueue_head(&vcpu->wq);
222 kvm_async_pf_vcpu_init(vcpu);
224 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
229 vcpu->run = page_address(page);
231 kvm_vcpu_set_in_spin_loop(vcpu, false);
232 kvm_vcpu_set_dy_eligible(vcpu, false);
233 vcpu->preempted = false;
235 r = kvm_arch_vcpu_init(vcpu);
241 free_page((unsigned long)vcpu->run);
245 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
247 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
250 kvm_arch_vcpu_uninit(vcpu);
251 free_page((unsigned long)vcpu->run);
253 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
255 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
256 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
258 return container_of(mn, struct kvm, mmu_notifier);
261 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
262 struct mm_struct *mm,
263 unsigned long address)
265 struct kvm *kvm = mmu_notifier_to_kvm(mn);
266 int need_tlb_flush, idx;
269 * When ->invalidate_page runs, the linux pte has been zapped
270 * already but the page is still allocated until
271 * ->invalidate_page returns. So if we increase the sequence
272 * here the kvm page fault will notice if the spte can't be
273 * established because the page is going to be freed. If
274 * instead the kvm page fault establishes the spte before
275 * ->invalidate_page runs, kvm_unmap_hva will release it
278 * The sequence increase only need to be seen at spin_unlock
279 * time, and not at spin_lock time.
281 * Increasing the sequence after the spin_unlock would be
282 * unsafe because the kvm page fault could then establish the
283 * pte after kvm_unmap_hva returned, without noticing the page
284 * is going to be freed.
286 idx = srcu_read_lock(&kvm->srcu);
287 spin_lock(&kvm->mmu_lock);
289 kvm->mmu_notifier_seq++;
290 need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
291 /* we've to flush the tlb before the pages can be freed */
293 kvm_flush_remote_tlbs(kvm);
295 spin_unlock(&kvm->mmu_lock);
297 kvm_arch_mmu_notifier_invalidate_page(kvm, address);
299 srcu_read_unlock(&kvm->srcu, idx);
302 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
303 struct mm_struct *mm,
304 unsigned long address,
307 struct kvm *kvm = mmu_notifier_to_kvm(mn);
310 idx = srcu_read_lock(&kvm->srcu);
311 spin_lock(&kvm->mmu_lock);
312 kvm->mmu_notifier_seq++;
313 kvm_set_spte_hva(kvm, address, pte);
314 spin_unlock(&kvm->mmu_lock);
315 srcu_read_unlock(&kvm->srcu, idx);
318 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
319 struct mm_struct *mm,
323 struct kvm *kvm = mmu_notifier_to_kvm(mn);
324 int need_tlb_flush = 0, idx;
326 idx = srcu_read_lock(&kvm->srcu);
327 spin_lock(&kvm->mmu_lock);
329 * The count increase must become visible at unlock time as no
330 * spte can be established without taking the mmu_lock and
331 * count is also read inside the mmu_lock critical section.
333 kvm->mmu_notifier_count++;
334 need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
335 need_tlb_flush |= kvm->tlbs_dirty;
336 /* we've to flush the tlb before the pages can be freed */
338 kvm_flush_remote_tlbs(kvm);
340 spin_unlock(&kvm->mmu_lock);
341 srcu_read_unlock(&kvm->srcu, idx);
344 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
345 struct mm_struct *mm,
349 struct kvm *kvm = mmu_notifier_to_kvm(mn);
351 spin_lock(&kvm->mmu_lock);
353 * This sequence increase will notify the kvm page fault that
354 * the page that is going to be mapped in the spte could have
357 kvm->mmu_notifier_seq++;
360 * The above sequence increase must be visible before the
361 * below count decrease, which is ensured by the smp_wmb above
362 * in conjunction with the smp_rmb in mmu_notifier_retry().
364 kvm->mmu_notifier_count--;
365 spin_unlock(&kvm->mmu_lock);
367 BUG_ON(kvm->mmu_notifier_count < 0);
370 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
371 struct mm_struct *mm,
375 struct kvm *kvm = mmu_notifier_to_kvm(mn);
378 idx = srcu_read_lock(&kvm->srcu);
379 spin_lock(&kvm->mmu_lock);
381 young = kvm_age_hva(kvm, start, end);
383 kvm_flush_remote_tlbs(kvm);
385 spin_unlock(&kvm->mmu_lock);
386 srcu_read_unlock(&kvm->srcu, idx);
391 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
392 struct mm_struct *mm,
393 unsigned long address)
395 struct kvm *kvm = mmu_notifier_to_kvm(mn);
398 idx = srcu_read_lock(&kvm->srcu);
399 spin_lock(&kvm->mmu_lock);
400 young = kvm_test_age_hva(kvm, address);
401 spin_unlock(&kvm->mmu_lock);
402 srcu_read_unlock(&kvm->srcu, idx);
407 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
408 struct mm_struct *mm)
410 struct kvm *kvm = mmu_notifier_to_kvm(mn);
413 idx = srcu_read_lock(&kvm->srcu);
414 kvm_arch_flush_shadow_all(kvm);
415 srcu_read_unlock(&kvm->srcu, idx);
418 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
419 .invalidate_page = kvm_mmu_notifier_invalidate_page,
420 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
421 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
422 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
423 .test_young = kvm_mmu_notifier_test_young,
424 .change_pte = kvm_mmu_notifier_change_pte,
425 .release = kvm_mmu_notifier_release,
428 static int kvm_init_mmu_notifier(struct kvm *kvm)
430 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
431 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
434 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
436 static int kvm_init_mmu_notifier(struct kvm *kvm)
441 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
443 static void kvm_init_memslots_id(struct kvm *kvm)
446 struct kvm_memslots *slots = kvm->memslots;
448 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
449 slots->id_to_index[i] = slots->memslots[i].id = i;
452 static struct kvm *kvm_create_vm(unsigned long type)
455 struct kvm *kvm = kvm_arch_alloc_vm();
458 return ERR_PTR(-ENOMEM);
460 r = kvm_arch_init_vm(kvm, type);
462 goto out_err_no_disable;
464 r = hardware_enable_all();
466 goto out_err_no_disable;
468 #ifdef CONFIG_HAVE_KVM_IRQFD
469 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
472 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
475 kvm->memslots = kvm_kvzalloc(sizeof(struct kvm_memslots));
477 goto out_err_no_srcu;
480 * Init kvm generation close to the maximum to easily test the
481 * code of handling generation number wrap-around.
483 kvm->memslots->generation = -150;
485 kvm_init_memslots_id(kvm);
486 if (init_srcu_struct(&kvm->srcu))
487 goto out_err_no_srcu;
488 if (init_srcu_struct(&kvm->irq_srcu))
489 goto out_err_no_irq_srcu;
490 for (i = 0; i < KVM_NR_BUSES; i++) {
491 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
497 spin_lock_init(&kvm->mmu_lock);
498 kvm->mm = current->mm;
499 atomic_inc(&kvm->mm->mm_count);
500 kvm_eventfd_init(kvm);
501 mutex_init(&kvm->lock);
502 mutex_init(&kvm->irq_lock);
503 mutex_init(&kvm->slots_lock);
504 atomic_set(&kvm->users_count, 1);
505 INIT_LIST_HEAD(&kvm->devices);
507 r = kvm_init_mmu_notifier(kvm);
511 spin_lock(&kvm_lock);
512 list_add(&kvm->vm_list, &vm_list);
513 spin_unlock(&kvm_lock);
518 cleanup_srcu_struct(&kvm->irq_srcu);
520 cleanup_srcu_struct(&kvm->srcu);
522 hardware_disable_all();
524 for (i = 0; i < KVM_NR_BUSES; i++)
525 kfree(kvm->buses[i]);
526 kvfree(kvm->memslots);
527 kvm_arch_free_vm(kvm);
532 * Avoid using vmalloc for a small buffer.
533 * Should not be used when the size is statically known.
535 void *kvm_kvzalloc(unsigned long size)
537 if (size > PAGE_SIZE)
538 return vzalloc(size);
540 return kzalloc(size, GFP_KERNEL);
543 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
545 if (!memslot->dirty_bitmap)
548 kvfree(memslot->dirty_bitmap);
549 memslot->dirty_bitmap = NULL;
553 * Free any memory in @free but not in @dont.
555 static void kvm_free_physmem_slot(struct kvm *kvm, struct kvm_memory_slot *free,
556 struct kvm_memory_slot *dont)
558 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
559 kvm_destroy_dirty_bitmap(free);
561 kvm_arch_free_memslot(kvm, free, dont);
566 static void kvm_free_physmem(struct kvm *kvm)
568 struct kvm_memslots *slots = kvm->memslots;
569 struct kvm_memory_slot *memslot;
571 kvm_for_each_memslot(memslot, slots)
572 kvm_free_physmem_slot(kvm, memslot, NULL);
574 kvfree(kvm->memslots);
577 static void kvm_destroy_devices(struct kvm *kvm)
579 struct list_head *node, *tmp;
581 list_for_each_safe(node, tmp, &kvm->devices) {
582 struct kvm_device *dev =
583 list_entry(node, struct kvm_device, vm_node);
586 dev->ops->destroy(dev);
590 static void kvm_destroy_vm(struct kvm *kvm)
593 struct mm_struct *mm = kvm->mm;
595 kvm_arch_sync_events(kvm);
596 spin_lock(&kvm_lock);
597 list_del(&kvm->vm_list);
598 spin_unlock(&kvm_lock);
599 kvm_free_irq_routing(kvm);
600 for (i = 0; i < KVM_NR_BUSES; i++)
601 kvm_io_bus_destroy(kvm->buses[i]);
602 kvm_coalesced_mmio_free(kvm);
603 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
604 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
606 kvm_arch_flush_shadow_all(kvm);
608 kvm_arch_destroy_vm(kvm);
609 kvm_destroy_devices(kvm);
610 kvm_free_physmem(kvm);
611 cleanup_srcu_struct(&kvm->irq_srcu);
612 cleanup_srcu_struct(&kvm->srcu);
613 kvm_arch_free_vm(kvm);
614 hardware_disable_all();
618 void kvm_get_kvm(struct kvm *kvm)
620 atomic_inc(&kvm->users_count);
622 EXPORT_SYMBOL_GPL(kvm_get_kvm);
624 void kvm_put_kvm(struct kvm *kvm)
626 if (atomic_dec_and_test(&kvm->users_count))
629 EXPORT_SYMBOL_GPL(kvm_put_kvm);
632 static int kvm_vm_release(struct inode *inode, struct file *filp)
634 struct kvm *kvm = filp->private_data;
636 kvm_irqfd_release(kvm);
643 * Allocation size is twice as large as the actual dirty bitmap size.
644 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
646 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
648 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
650 memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
651 if (!memslot->dirty_bitmap)
658 * Insert memslot and re-sort memslots based on their GFN,
659 * so binary search could be used to lookup GFN.
660 * Sorting algorithm takes advantage of having initially
661 * sorted array and known changed memslot position.
663 static void update_memslots(struct kvm_memslots *slots,
664 struct kvm_memory_slot *new)
667 int i = slots->id_to_index[id];
668 struct kvm_memory_slot *mslots = slots->memslots;
670 WARN_ON(mslots[i].id != id);
672 WARN_ON(!mslots[i].npages);
675 if (mslots[i].npages)
678 if (!mslots[i].npages)
682 while (i < KVM_MEM_SLOTS_NUM - 1 &&
683 new->base_gfn <= mslots[i + 1].base_gfn) {
684 if (!mslots[i + 1].npages)
686 mslots[i] = mslots[i + 1];
687 slots->id_to_index[mslots[i].id] = i;
692 * The ">=" is needed when creating a slot with base_gfn == 0,
693 * so that it moves before all those with base_gfn == npages == 0.
695 * On the other hand, if new->npages is zero, the above loop has
696 * already left i pointing to the beginning of the empty part of
697 * mslots, and the ">=" would move the hole backwards in this
698 * case---which is wrong. So skip the loop when deleting a slot.
702 new->base_gfn >= mslots[i - 1].base_gfn) {
703 mslots[i] = mslots[i - 1];
704 slots->id_to_index[mslots[i].id] = i;
708 WARN_ON_ONCE(i != slots->used_slots);
711 slots->id_to_index[mslots[i].id] = i;
714 static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
716 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
718 #ifdef __KVM_HAVE_READONLY_MEM
719 valid_flags |= KVM_MEM_READONLY;
722 if (mem->flags & ~valid_flags)
728 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
729 struct kvm_memslots *slots)
731 struct kvm_memslots *old_memslots = kvm->memslots;
734 * Set the low bit in the generation, which disables SPTE caching
735 * until the end of synchronize_srcu_expedited.
737 WARN_ON(old_memslots->generation & 1);
738 slots->generation = old_memslots->generation + 1;
740 rcu_assign_pointer(kvm->memslots, slots);
741 synchronize_srcu_expedited(&kvm->srcu);
744 * Increment the new memslot generation a second time. This prevents
745 * vm exits that race with memslot updates from caching a memslot
746 * generation that will (potentially) be valid forever.
750 kvm_arch_memslots_updated(kvm);
756 * Allocate some memory and give it an address in the guest physical address
759 * Discontiguous memory is allowed, mostly for framebuffers.
761 * Must be called holding kvm->slots_lock for write.
763 int __kvm_set_memory_region(struct kvm *kvm,
764 struct kvm_userspace_memory_region *mem)
768 unsigned long npages;
769 struct kvm_memory_slot *slot;
770 struct kvm_memory_slot old, new;
771 struct kvm_memslots *slots = NULL, *old_memslots;
772 enum kvm_mr_change change;
774 r = check_memory_region_flags(mem);
779 /* General sanity checks */
780 if (mem->memory_size & (PAGE_SIZE - 1))
782 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
784 /* We can read the guest memory with __xxx_user() later on. */
785 if ((mem->slot < KVM_USER_MEM_SLOTS) &&
786 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
787 !access_ok(VERIFY_WRITE,
788 (void __user *)(unsigned long)mem->userspace_addr,
791 if (mem->slot >= KVM_MEM_SLOTS_NUM)
793 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
796 slot = id_to_memslot(kvm->memslots, mem->slot);
797 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
798 npages = mem->memory_size >> PAGE_SHIFT;
800 if (npages > KVM_MEM_MAX_NR_PAGES)
804 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
809 new.base_gfn = base_gfn;
811 new.flags = mem->flags;
815 change = KVM_MR_CREATE;
816 else { /* Modify an existing slot. */
817 if ((mem->userspace_addr != old.userspace_addr) ||
818 (npages != old.npages) ||
819 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
822 if (base_gfn != old.base_gfn)
823 change = KVM_MR_MOVE;
824 else if (new.flags != old.flags)
825 change = KVM_MR_FLAGS_ONLY;
826 else { /* Nothing to change. */
831 } else if (old.npages) {
832 change = KVM_MR_DELETE;
833 } else /* Modify a non-existent slot: disallowed. */
836 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
837 /* Check for overlaps */
839 kvm_for_each_memslot(slot, kvm->memslots) {
840 if ((slot->id >= KVM_USER_MEM_SLOTS) ||
841 (slot->id == mem->slot))
843 if (!((base_gfn + npages <= slot->base_gfn) ||
844 (base_gfn >= slot->base_gfn + slot->npages)))
849 /* Free page dirty bitmap if unneeded */
850 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
851 new.dirty_bitmap = NULL;
854 if (change == KVM_MR_CREATE) {
855 new.userspace_addr = mem->userspace_addr;
857 if (kvm_arch_create_memslot(kvm, &new, npages))
861 /* Allocate page dirty bitmap if needed */
862 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
863 if (kvm_create_dirty_bitmap(&new) < 0)
867 slots = kvm_kvzalloc(sizeof(struct kvm_memslots));
870 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
872 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
873 slot = id_to_memslot(slots, mem->slot);
874 slot->flags |= KVM_MEMSLOT_INVALID;
876 old_memslots = install_new_memslots(kvm, slots);
878 /* slot was deleted or moved, clear iommu mapping */
879 kvm_iommu_unmap_pages(kvm, &old);
880 /* From this point no new shadow pages pointing to a deleted,
881 * or moved, memslot will be created.
883 * validation of sp->gfn happens in:
884 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
885 * - kvm_is_visible_gfn (mmu_check_roots)
887 kvm_arch_flush_shadow_memslot(kvm, slot);
890 * We can re-use the old_memslots from above, the only difference
891 * from the currently installed memslots is the invalid flag. This
892 * will get overwritten by update_memslots anyway.
894 slots = old_memslots;
897 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
901 /* actual memory is freed via old in kvm_free_physmem_slot below */
902 if (change == KVM_MR_DELETE) {
903 new.dirty_bitmap = NULL;
904 memset(&new.arch, 0, sizeof(new.arch));
907 update_memslots(slots, &new);
908 old_memslots = install_new_memslots(kvm, slots);
910 kvm_arch_commit_memory_region(kvm, mem, &old, change);
912 kvm_free_physmem_slot(kvm, &old, &new);
913 kvfree(old_memslots);
916 * IOMMU mapping: New slots need to be mapped. Old slots need to be
917 * un-mapped and re-mapped if their base changes. Since base change
918 * unmapping is handled above with slot deletion, mapping alone is
919 * needed here. Anything else the iommu might care about for existing
920 * slots (size changes, userspace addr changes and read-only flag
921 * changes) is disallowed above, so any other attribute changes getting
922 * here can be skipped.
924 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
925 r = kvm_iommu_map_pages(kvm, &new);
934 kvm_free_physmem_slot(kvm, &new, &old);
938 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
940 int kvm_set_memory_region(struct kvm *kvm,
941 struct kvm_userspace_memory_region *mem)
945 mutex_lock(&kvm->slots_lock);
946 r = __kvm_set_memory_region(kvm, mem);
947 mutex_unlock(&kvm->slots_lock);
950 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
952 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
953 struct kvm_userspace_memory_region *mem)
955 if (mem->slot >= KVM_USER_MEM_SLOTS)
957 return kvm_set_memory_region(kvm, mem);
960 int kvm_get_dirty_log(struct kvm *kvm,
961 struct kvm_dirty_log *log, int *is_dirty)
963 struct kvm_memory_slot *memslot;
966 unsigned long any = 0;
969 if (log->slot >= KVM_USER_MEM_SLOTS)
972 memslot = id_to_memslot(kvm->memslots, log->slot);
974 if (!memslot->dirty_bitmap)
977 n = kvm_dirty_bitmap_bytes(memslot);
979 for (i = 0; !any && i < n/sizeof(long); ++i)
980 any = memslot->dirty_bitmap[i];
983 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
993 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
995 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
997 * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages
998 * are dirty write protect them for next write.
999 * @kvm: pointer to kvm instance
1000 * @log: slot id and address to which we copy the log
1001 * @is_dirty: flag set if any page is dirty
1003 * We need to keep it in mind that VCPU threads can write to the bitmap
1004 * concurrently. So, to avoid losing track of dirty pages we keep the
1007 * 1. Take a snapshot of the bit and clear it if needed.
1008 * 2. Write protect the corresponding page.
1009 * 3. Copy the snapshot to the userspace.
1010 * 4. Upon return caller flushes TLB's if needed.
1012 * Between 2 and 4, the guest may write to the page using the remaining TLB
1013 * entry. This is not a problem because the page is reported dirty using
1014 * the snapshot taken before and step 4 ensures that writes done after
1015 * exiting to userspace will be logged for the next call.
1018 int kvm_get_dirty_log_protect(struct kvm *kvm,
1019 struct kvm_dirty_log *log, bool *is_dirty)
1021 struct kvm_memory_slot *memslot;
1024 unsigned long *dirty_bitmap;
1025 unsigned long *dirty_bitmap_buffer;
1028 if (log->slot >= KVM_USER_MEM_SLOTS)
1031 memslot = id_to_memslot(kvm->memslots, log->slot);
1033 dirty_bitmap = memslot->dirty_bitmap;
1038 n = kvm_dirty_bitmap_bytes(memslot);
1040 dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long);
1041 memset(dirty_bitmap_buffer, 0, n);
1043 spin_lock(&kvm->mmu_lock);
1045 for (i = 0; i < n / sizeof(long); i++) {
1049 if (!dirty_bitmap[i])
1054 mask = xchg(&dirty_bitmap[i], 0);
1055 dirty_bitmap_buffer[i] = mask;
1058 offset = i * BITS_PER_LONG;
1059 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1064 spin_unlock(&kvm->mmu_lock);
1067 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1074 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect);
1077 bool kvm_largepages_enabled(void)
1079 return largepages_enabled;
1082 void kvm_disable_largepages(void)
1084 largepages_enabled = false;
1086 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1088 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1090 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1092 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1094 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1096 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1098 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1099 memslot->flags & KVM_MEMSLOT_INVALID)
1104 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1106 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1108 struct vm_area_struct *vma;
1109 unsigned long addr, size;
1113 addr = gfn_to_hva(kvm, gfn);
1114 if (kvm_is_error_hva(addr))
1117 down_read(¤t->mm->mmap_sem);
1118 vma = find_vma(current->mm, addr);
1122 size = vma_kernel_pagesize(vma);
1125 up_read(¤t->mm->mmap_sem);
1130 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1132 return slot->flags & KVM_MEM_READONLY;
1135 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1136 gfn_t *nr_pages, bool write)
1138 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1139 return KVM_HVA_ERR_BAD;
1141 if (memslot_is_readonly(slot) && write)
1142 return KVM_HVA_ERR_RO_BAD;
1145 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1147 return __gfn_to_hva_memslot(slot, gfn);
1150 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1153 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1156 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1159 return gfn_to_hva_many(slot, gfn, NULL);
1161 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1163 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1165 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1167 EXPORT_SYMBOL_GPL(gfn_to_hva);
1170 * If writable is set to false, the hva returned by this function is only
1171 * allowed to be read.
1173 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1174 gfn_t gfn, bool *writable)
1176 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1178 if (!kvm_is_error_hva(hva) && writable)
1179 *writable = !memslot_is_readonly(slot);
1184 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1186 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1188 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1191 static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1192 unsigned long start, int write, struct page **page)
1194 int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1197 flags |= FOLL_WRITE;
1199 return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1202 static inline int check_user_page_hwpoison(unsigned long addr)
1204 int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1206 rc = __get_user_pages(current, current->mm, addr, 1,
1207 flags, NULL, NULL, NULL);
1208 return rc == -EHWPOISON;
1212 * The atomic path to get the writable pfn which will be stored in @pfn,
1213 * true indicates success, otherwise false is returned.
1215 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1216 bool write_fault, bool *writable, pfn_t *pfn)
1218 struct page *page[1];
1221 if (!(async || atomic))
1225 * Fast pin a writable pfn only if it is a write fault request
1226 * or the caller allows to map a writable pfn for a read fault
1229 if (!(write_fault || writable))
1232 npages = __get_user_pages_fast(addr, 1, 1, page);
1234 *pfn = page_to_pfn(page[0]);
1245 * The slow path to get the pfn of the specified host virtual address,
1246 * 1 indicates success, -errno is returned if error is detected.
1248 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1249 bool *writable, pfn_t *pfn)
1251 struct page *page[1];
1257 *writable = write_fault;
1260 down_read(¤t->mm->mmap_sem);
1261 npages = get_user_page_nowait(current, current->mm,
1262 addr, write_fault, page);
1263 up_read(¤t->mm->mmap_sem);
1265 npages = __get_user_pages_unlocked(current, current->mm, addr, 1,
1266 write_fault, 0, page,
1267 FOLL_TOUCH|FOLL_HWPOISON);
1271 /* map read fault as writable if possible */
1272 if (unlikely(!write_fault) && writable) {
1273 struct page *wpage[1];
1275 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1284 *pfn = page_to_pfn(page[0]);
1288 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1290 if (unlikely(!(vma->vm_flags & VM_READ)))
1293 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1300 * Pin guest page in memory and return its pfn.
1301 * @addr: host virtual address which maps memory to the guest
1302 * @atomic: whether this function can sleep
1303 * @async: whether this function need to wait IO complete if the
1304 * host page is not in the memory
1305 * @write_fault: whether we should get a writable host page
1306 * @writable: whether it allows to map a writable host page for !@write_fault
1308 * The function will map a writable host page for these two cases:
1309 * 1): @write_fault = true
1310 * 2): @write_fault = false && @writable, @writable will tell the caller
1311 * whether the mapping is writable.
1313 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1314 bool write_fault, bool *writable)
1316 struct vm_area_struct *vma;
1320 /* we can do it either atomically or asynchronously, not both */
1321 BUG_ON(atomic && async);
1323 if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1327 return KVM_PFN_ERR_FAULT;
1329 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1333 down_read(¤t->mm->mmap_sem);
1334 if (npages == -EHWPOISON ||
1335 (!async && check_user_page_hwpoison(addr))) {
1336 pfn = KVM_PFN_ERR_HWPOISON;
1340 vma = find_vma_intersection(current->mm, addr, addr + 1);
1343 pfn = KVM_PFN_ERR_FAULT;
1344 else if ((vma->vm_flags & VM_PFNMAP)) {
1345 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1347 BUG_ON(!kvm_is_reserved_pfn(pfn));
1349 if (async && vma_is_valid(vma, write_fault))
1351 pfn = KVM_PFN_ERR_FAULT;
1354 up_read(¤t->mm->mmap_sem);
1359 __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1360 bool *async, bool write_fault, bool *writable)
1362 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1364 if (addr == KVM_HVA_ERR_RO_BAD)
1365 return KVM_PFN_ERR_RO_FAULT;
1367 if (kvm_is_error_hva(addr))
1368 return KVM_PFN_NOSLOT;
1370 /* Do not map writable pfn in the readonly memslot. */
1371 if (writable && memslot_is_readonly(slot)) {
1376 return hva_to_pfn(addr, atomic, async, write_fault,
1380 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1381 bool write_fault, bool *writable)
1383 struct kvm_memory_slot *slot;
1388 slot = gfn_to_memslot(kvm, gfn);
1390 return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1394 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1396 return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1398 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1400 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1401 bool write_fault, bool *writable)
1403 return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1405 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1407 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1409 return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1411 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1413 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1416 return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1418 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1420 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1422 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1425 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1427 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1429 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1431 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1437 addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1438 if (kvm_is_error_hva(addr))
1441 if (entry < nr_pages)
1444 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1446 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1448 static struct page *kvm_pfn_to_page(pfn_t pfn)
1450 if (is_error_noslot_pfn(pfn))
1451 return KVM_ERR_PTR_BAD_PAGE;
1453 if (kvm_is_reserved_pfn(pfn)) {
1455 return KVM_ERR_PTR_BAD_PAGE;
1458 return pfn_to_page(pfn);
1461 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1465 pfn = gfn_to_pfn(kvm, gfn);
1467 return kvm_pfn_to_page(pfn);
1469 EXPORT_SYMBOL_GPL(gfn_to_page);
1471 void kvm_release_page_clean(struct page *page)
1473 WARN_ON(is_error_page(page));
1475 kvm_release_pfn_clean(page_to_pfn(page));
1477 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1479 void kvm_release_pfn_clean(pfn_t pfn)
1481 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
1482 put_page(pfn_to_page(pfn));
1484 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1486 void kvm_release_page_dirty(struct page *page)
1488 WARN_ON(is_error_page(page));
1490 kvm_release_pfn_dirty(page_to_pfn(page));
1492 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1494 static void kvm_release_pfn_dirty(pfn_t pfn)
1496 kvm_set_pfn_dirty(pfn);
1497 kvm_release_pfn_clean(pfn);
1500 void kvm_set_pfn_dirty(pfn_t pfn)
1502 if (!kvm_is_reserved_pfn(pfn)) {
1503 struct page *page = pfn_to_page(pfn);
1505 if (!PageReserved(page))
1509 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1511 void kvm_set_pfn_accessed(pfn_t pfn)
1513 if (!kvm_is_reserved_pfn(pfn))
1514 mark_page_accessed(pfn_to_page(pfn));
1516 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1518 void kvm_get_pfn(pfn_t pfn)
1520 if (!kvm_is_reserved_pfn(pfn))
1521 get_page(pfn_to_page(pfn));
1523 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1525 static int next_segment(unsigned long len, int offset)
1527 if (len > PAGE_SIZE - offset)
1528 return PAGE_SIZE - offset;
1533 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1539 addr = gfn_to_hva_prot(kvm, gfn, NULL);
1540 if (kvm_is_error_hva(addr))
1542 r = __copy_from_user(data, (void __user *)addr + offset, len);
1547 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1549 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1551 gfn_t gfn = gpa >> PAGE_SHIFT;
1553 int offset = offset_in_page(gpa);
1556 while ((seg = next_segment(len, offset)) != 0) {
1557 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1567 EXPORT_SYMBOL_GPL(kvm_read_guest);
1569 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1574 gfn_t gfn = gpa >> PAGE_SHIFT;
1575 int offset = offset_in_page(gpa);
1577 addr = gfn_to_hva_prot(kvm, gfn, NULL);
1578 if (kvm_is_error_hva(addr))
1580 pagefault_disable();
1581 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1587 EXPORT_SYMBOL(kvm_read_guest_atomic);
1589 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1590 int offset, int len)
1595 addr = gfn_to_hva(kvm, gfn);
1596 if (kvm_is_error_hva(addr))
1598 r = __copy_to_user((void __user *)addr + offset, data, len);
1601 mark_page_dirty(kvm, gfn);
1604 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1606 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1609 gfn_t gfn = gpa >> PAGE_SHIFT;
1611 int offset = offset_in_page(gpa);
1614 while ((seg = next_segment(len, offset)) != 0) {
1615 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1625 EXPORT_SYMBOL_GPL(kvm_write_guest);
1627 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1628 gpa_t gpa, unsigned long len)
1630 struct kvm_memslots *slots = kvm_memslots(kvm);
1631 int offset = offset_in_page(gpa);
1632 gfn_t start_gfn = gpa >> PAGE_SHIFT;
1633 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1634 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1635 gfn_t nr_pages_avail;
1638 ghc->generation = slots->generation;
1640 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1641 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, NULL);
1642 if (!kvm_is_error_hva(ghc->hva) && nr_pages_needed <= 1) {
1646 * If the requested region crosses two memslots, we still
1647 * verify that the entire region is valid here.
1649 while (start_gfn <= end_gfn) {
1650 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1651 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1653 if (kvm_is_error_hva(ghc->hva))
1655 start_gfn += nr_pages_avail;
1657 /* Use the slow path for cross page reads and writes. */
1658 ghc->memslot = NULL;
1662 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1664 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1665 void *data, unsigned long len)
1667 struct kvm_memslots *slots = kvm_memslots(kvm);
1670 BUG_ON(len > ghc->len);
1672 if (slots->generation != ghc->generation)
1673 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1675 if (unlikely(!ghc->memslot))
1676 return kvm_write_guest(kvm, ghc->gpa, data, len);
1678 if (kvm_is_error_hva(ghc->hva))
1681 r = __copy_to_user((void __user *)ghc->hva, data, len);
1684 mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1688 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1690 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1691 void *data, unsigned long len)
1693 struct kvm_memslots *slots = kvm_memslots(kvm);
1696 BUG_ON(len > ghc->len);
1698 if (slots->generation != ghc->generation)
1699 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1701 if (unlikely(!ghc->memslot))
1702 return kvm_read_guest(kvm, ghc->gpa, data, len);
1704 if (kvm_is_error_hva(ghc->hva))
1707 r = __copy_from_user(data, (void __user *)ghc->hva, len);
1713 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1715 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1717 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
1719 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
1721 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1723 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1725 gfn_t gfn = gpa >> PAGE_SHIFT;
1727 int offset = offset_in_page(gpa);
1730 while ((seg = next_segment(len, offset)) != 0) {
1731 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1740 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1742 static void mark_page_dirty_in_slot(struct kvm *kvm,
1743 struct kvm_memory_slot *memslot,
1746 if (memslot && memslot->dirty_bitmap) {
1747 unsigned long rel_gfn = gfn - memslot->base_gfn;
1749 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1753 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1755 struct kvm_memory_slot *memslot;
1757 memslot = gfn_to_memslot(kvm, gfn);
1758 mark_page_dirty_in_slot(kvm, memslot, gfn);
1760 EXPORT_SYMBOL_GPL(mark_page_dirty);
1762 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
1764 if (kvm_arch_vcpu_runnable(vcpu)) {
1765 kvm_make_request(KVM_REQ_UNHALT, vcpu);
1768 if (kvm_cpu_has_pending_timer(vcpu))
1770 if (signal_pending(current))
1777 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1779 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1783 bool waited = false;
1785 start = cur = ktime_get();
1787 ktime_t stop = ktime_add_ns(ktime_get(), halt_poll_ns);
1791 * This sets KVM_REQ_UNHALT if an interrupt
1794 if (kvm_vcpu_check_block(vcpu) < 0) {
1795 ++vcpu->stat.halt_successful_poll;
1799 } while (single_task_running() && ktime_before(cur, stop));
1803 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1805 if (kvm_vcpu_check_block(vcpu) < 0)
1812 finish_wait(&vcpu->wq, &wait);
1816 trace_kvm_vcpu_wakeup(ktime_to_ns(cur) - ktime_to_ns(start), waited);
1818 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
1822 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1824 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1827 int cpu = vcpu->cpu;
1828 wait_queue_head_t *wqp;
1830 wqp = kvm_arch_vcpu_wq(vcpu);
1831 if (waitqueue_active(wqp)) {
1832 wake_up_interruptible(wqp);
1833 ++vcpu->stat.halt_wakeup;
1837 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1838 if (kvm_arch_vcpu_should_kick(vcpu))
1839 smp_send_reschedule(cpu);
1842 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
1843 #endif /* !CONFIG_S390 */
1845 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
1848 struct task_struct *task = NULL;
1852 pid = rcu_dereference(target->pid);
1854 task = get_pid_task(pid, PIDTYPE_PID);
1858 ret = yield_to(task, 1);
1859 put_task_struct(task);
1863 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1866 * Helper that checks whether a VCPU is eligible for directed yield.
1867 * Most eligible candidate to yield is decided by following heuristics:
1869 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1870 * (preempted lock holder), indicated by @in_spin_loop.
1871 * Set at the beiginning and cleared at the end of interception/PLE handler.
1873 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1874 * chance last time (mostly it has become eligible now since we have probably
1875 * yielded to lockholder in last iteration. This is done by toggling
1876 * @dy_eligible each time a VCPU checked for eligibility.)
1878 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1879 * to preempted lock-holder could result in wrong VCPU selection and CPU
1880 * burning. Giving priority for a potential lock-holder increases lock
1883 * Since algorithm is based on heuristics, accessing another VCPU data without
1884 * locking does not harm. It may result in trying to yield to same VCPU, fail
1885 * and continue with next VCPU and so on.
1887 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1889 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1892 eligible = !vcpu->spin_loop.in_spin_loop ||
1893 vcpu->spin_loop.dy_eligible;
1895 if (vcpu->spin_loop.in_spin_loop)
1896 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1904 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1906 struct kvm *kvm = me->kvm;
1907 struct kvm_vcpu *vcpu;
1908 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1914 kvm_vcpu_set_in_spin_loop(me, true);
1916 * We boost the priority of a VCPU that is runnable but not
1917 * currently running, because it got preempted by something
1918 * else and called schedule in __vcpu_run. Hopefully that
1919 * VCPU is holding the lock that we need and will release it.
1920 * We approximate round-robin by starting at the last boosted VCPU.
1922 for (pass = 0; pass < 2 && !yielded && try; pass++) {
1923 kvm_for_each_vcpu(i, vcpu, kvm) {
1924 if (!pass && i <= last_boosted_vcpu) {
1925 i = last_boosted_vcpu;
1927 } else if (pass && i > last_boosted_vcpu)
1929 if (!ACCESS_ONCE(vcpu->preempted))
1933 if (waitqueue_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
1935 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1938 yielded = kvm_vcpu_yield_to(vcpu);
1940 kvm->last_boosted_vcpu = i;
1942 } else if (yielded < 0) {
1949 kvm_vcpu_set_in_spin_loop(me, false);
1951 /* Ensure vcpu is not eligible during next spinloop */
1952 kvm_vcpu_set_dy_eligible(me, false);
1954 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1956 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1958 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1961 if (vmf->pgoff == 0)
1962 page = virt_to_page(vcpu->run);
1964 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1965 page = virt_to_page(vcpu->arch.pio_data);
1967 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1968 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1969 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1972 return kvm_arch_vcpu_fault(vcpu, vmf);
1978 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1979 .fault = kvm_vcpu_fault,
1982 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1984 vma->vm_ops = &kvm_vcpu_vm_ops;
1988 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1990 struct kvm_vcpu *vcpu = filp->private_data;
1992 kvm_put_kvm(vcpu->kvm);
1996 static struct file_operations kvm_vcpu_fops = {
1997 .release = kvm_vcpu_release,
1998 .unlocked_ioctl = kvm_vcpu_ioctl,
1999 #ifdef CONFIG_KVM_COMPAT
2000 .compat_ioctl = kvm_vcpu_compat_ioctl,
2002 .mmap = kvm_vcpu_mmap,
2003 .llseek = noop_llseek,
2007 * Allocates an inode for the vcpu.
2009 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2011 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
2015 * Creates some virtual cpus. Good luck creating more than one.
2017 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
2020 struct kvm_vcpu *vcpu, *v;
2022 if (id >= KVM_MAX_VCPUS)
2025 vcpu = kvm_arch_vcpu_create(kvm, id);
2027 return PTR_ERR(vcpu);
2029 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2031 r = kvm_arch_vcpu_setup(vcpu);
2035 mutex_lock(&kvm->lock);
2036 if (!kvm_vcpu_compatible(vcpu)) {
2038 goto unlock_vcpu_destroy;
2040 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
2042 goto unlock_vcpu_destroy;
2045 kvm_for_each_vcpu(r, v, kvm)
2046 if (v->vcpu_id == id) {
2048 goto unlock_vcpu_destroy;
2051 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
2053 /* Now it's all set up, let userspace reach it */
2055 r = create_vcpu_fd(vcpu);
2058 goto unlock_vcpu_destroy;
2061 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
2063 atomic_inc(&kvm->online_vcpus);
2065 mutex_unlock(&kvm->lock);
2066 kvm_arch_vcpu_postcreate(vcpu);
2069 unlock_vcpu_destroy:
2070 mutex_unlock(&kvm->lock);
2072 kvm_arch_vcpu_destroy(vcpu);
2076 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2079 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2080 vcpu->sigset_active = 1;
2081 vcpu->sigset = *sigset;
2083 vcpu->sigset_active = 0;
2087 static long kvm_vcpu_ioctl(struct file *filp,
2088 unsigned int ioctl, unsigned long arg)
2090 struct kvm_vcpu *vcpu = filp->private_data;
2091 void __user *argp = (void __user *)arg;
2093 struct kvm_fpu *fpu = NULL;
2094 struct kvm_sregs *kvm_sregs = NULL;
2096 if (vcpu->kvm->mm != current->mm)
2099 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2102 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
2104 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
2105 * so vcpu_load() would break it.
2107 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_S390_IRQ || ioctl == KVM_INTERRUPT)
2108 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2112 r = vcpu_load(vcpu);
2120 if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
2121 /* The thread running this VCPU changed. */
2122 struct pid *oldpid = vcpu->pid;
2123 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
2125 rcu_assign_pointer(vcpu->pid, newpid);
2130 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2131 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2133 case KVM_GET_REGS: {
2134 struct kvm_regs *kvm_regs;
2137 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2140 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2144 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2151 case KVM_SET_REGS: {
2152 struct kvm_regs *kvm_regs;
2155 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2156 if (IS_ERR(kvm_regs)) {
2157 r = PTR_ERR(kvm_regs);
2160 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2164 case KVM_GET_SREGS: {
2165 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2169 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2173 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2178 case KVM_SET_SREGS: {
2179 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2180 if (IS_ERR(kvm_sregs)) {
2181 r = PTR_ERR(kvm_sregs);
2185 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2188 case KVM_GET_MP_STATE: {
2189 struct kvm_mp_state mp_state;
2191 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2195 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
2200 case KVM_SET_MP_STATE: {
2201 struct kvm_mp_state mp_state;
2204 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
2206 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2209 case KVM_TRANSLATE: {
2210 struct kvm_translation tr;
2213 if (copy_from_user(&tr, argp, sizeof(tr)))
2215 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2219 if (copy_to_user(argp, &tr, sizeof(tr)))
2224 case KVM_SET_GUEST_DEBUG: {
2225 struct kvm_guest_debug dbg;
2228 if (copy_from_user(&dbg, argp, sizeof(dbg)))
2230 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2233 case KVM_SET_SIGNAL_MASK: {
2234 struct kvm_signal_mask __user *sigmask_arg = argp;
2235 struct kvm_signal_mask kvm_sigmask;
2236 sigset_t sigset, *p;
2241 if (copy_from_user(&kvm_sigmask, argp,
2242 sizeof(kvm_sigmask)))
2245 if (kvm_sigmask.len != sizeof(sigset))
2248 if (copy_from_user(&sigset, sigmask_arg->sigset,
2253 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2257 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2261 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2265 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2271 fpu = memdup_user(argp, sizeof(*fpu));
2277 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2281 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2290 #ifdef CONFIG_KVM_COMPAT
2291 static long kvm_vcpu_compat_ioctl(struct file *filp,
2292 unsigned int ioctl, unsigned long arg)
2294 struct kvm_vcpu *vcpu = filp->private_data;
2295 void __user *argp = compat_ptr(arg);
2298 if (vcpu->kvm->mm != current->mm)
2302 case KVM_SET_SIGNAL_MASK: {
2303 struct kvm_signal_mask __user *sigmask_arg = argp;
2304 struct kvm_signal_mask kvm_sigmask;
2305 compat_sigset_t csigset;
2310 if (copy_from_user(&kvm_sigmask, argp,
2311 sizeof(kvm_sigmask)))
2314 if (kvm_sigmask.len != sizeof(csigset))
2317 if (copy_from_user(&csigset, sigmask_arg->sigset,
2320 sigset_from_compat(&sigset, &csigset);
2321 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2323 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2327 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2335 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2336 int (*accessor)(struct kvm_device *dev,
2337 struct kvm_device_attr *attr),
2340 struct kvm_device_attr attr;
2345 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2348 return accessor(dev, &attr);
2351 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2354 struct kvm_device *dev = filp->private_data;
2357 case KVM_SET_DEVICE_ATTR:
2358 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2359 case KVM_GET_DEVICE_ATTR:
2360 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2361 case KVM_HAS_DEVICE_ATTR:
2362 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2364 if (dev->ops->ioctl)
2365 return dev->ops->ioctl(dev, ioctl, arg);
2371 static int kvm_device_release(struct inode *inode, struct file *filp)
2373 struct kvm_device *dev = filp->private_data;
2374 struct kvm *kvm = dev->kvm;
2380 static const struct file_operations kvm_device_fops = {
2381 .unlocked_ioctl = kvm_device_ioctl,
2382 #ifdef CONFIG_KVM_COMPAT
2383 .compat_ioctl = kvm_device_ioctl,
2385 .release = kvm_device_release,
2388 struct kvm_device *kvm_device_from_filp(struct file *filp)
2390 if (filp->f_op != &kvm_device_fops)
2393 return filp->private_data;
2396 static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
2397 #ifdef CONFIG_KVM_MPIC
2398 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
2399 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
2402 #ifdef CONFIG_KVM_XICS
2403 [KVM_DEV_TYPE_XICS] = &kvm_xics_ops,
2407 int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
2409 if (type >= ARRAY_SIZE(kvm_device_ops_table))
2412 if (kvm_device_ops_table[type] != NULL)
2415 kvm_device_ops_table[type] = ops;
2419 void kvm_unregister_device_ops(u32 type)
2421 if (kvm_device_ops_table[type] != NULL)
2422 kvm_device_ops_table[type] = NULL;
2425 static int kvm_ioctl_create_device(struct kvm *kvm,
2426 struct kvm_create_device *cd)
2428 struct kvm_device_ops *ops = NULL;
2429 struct kvm_device *dev;
2430 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2433 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
2436 ops = kvm_device_ops_table[cd->type];
2443 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2450 ret = ops->create(dev, cd->type);
2456 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2462 list_add(&dev->vm_node, &kvm->devices);
2468 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2471 case KVM_CAP_USER_MEMORY:
2472 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2473 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2474 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2475 case KVM_CAP_SET_BOOT_CPU_ID:
2477 case KVM_CAP_INTERNAL_ERROR_DATA:
2478 #ifdef CONFIG_HAVE_KVM_MSI
2479 case KVM_CAP_SIGNAL_MSI:
2481 #ifdef CONFIG_HAVE_KVM_IRQFD
2483 case KVM_CAP_IRQFD_RESAMPLE:
2485 case KVM_CAP_CHECK_EXTENSION_VM:
2487 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2488 case KVM_CAP_IRQ_ROUTING:
2489 return KVM_MAX_IRQ_ROUTES;
2494 return kvm_vm_ioctl_check_extension(kvm, arg);
2497 static long kvm_vm_ioctl(struct file *filp,
2498 unsigned int ioctl, unsigned long arg)
2500 struct kvm *kvm = filp->private_data;
2501 void __user *argp = (void __user *)arg;
2504 if (kvm->mm != current->mm)
2507 case KVM_CREATE_VCPU:
2508 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2510 case KVM_SET_USER_MEMORY_REGION: {
2511 struct kvm_userspace_memory_region kvm_userspace_mem;
2514 if (copy_from_user(&kvm_userspace_mem, argp,
2515 sizeof(kvm_userspace_mem)))
2518 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2521 case KVM_GET_DIRTY_LOG: {
2522 struct kvm_dirty_log log;
2525 if (copy_from_user(&log, argp, sizeof(log)))
2527 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2530 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2531 case KVM_REGISTER_COALESCED_MMIO: {
2532 struct kvm_coalesced_mmio_zone zone;
2535 if (copy_from_user(&zone, argp, sizeof(zone)))
2537 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2540 case KVM_UNREGISTER_COALESCED_MMIO: {
2541 struct kvm_coalesced_mmio_zone zone;
2544 if (copy_from_user(&zone, argp, sizeof(zone)))
2546 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2551 struct kvm_irqfd data;
2554 if (copy_from_user(&data, argp, sizeof(data)))
2556 r = kvm_irqfd(kvm, &data);
2559 case KVM_IOEVENTFD: {
2560 struct kvm_ioeventfd data;
2563 if (copy_from_user(&data, argp, sizeof(data)))
2565 r = kvm_ioeventfd(kvm, &data);
2568 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2569 case KVM_SET_BOOT_CPU_ID:
2571 mutex_lock(&kvm->lock);
2572 if (atomic_read(&kvm->online_vcpus) != 0)
2575 kvm->bsp_vcpu_id = arg;
2576 mutex_unlock(&kvm->lock);
2579 #ifdef CONFIG_HAVE_KVM_MSI
2580 case KVM_SIGNAL_MSI: {
2584 if (copy_from_user(&msi, argp, sizeof(msi)))
2586 r = kvm_send_userspace_msi(kvm, &msi);
2590 #ifdef __KVM_HAVE_IRQ_LINE
2591 case KVM_IRQ_LINE_STATUS:
2592 case KVM_IRQ_LINE: {
2593 struct kvm_irq_level irq_event;
2596 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
2599 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
2600 ioctl == KVM_IRQ_LINE_STATUS);
2605 if (ioctl == KVM_IRQ_LINE_STATUS) {
2606 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
2614 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2615 case KVM_SET_GSI_ROUTING: {
2616 struct kvm_irq_routing routing;
2617 struct kvm_irq_routing __user *urouting;
2618 struct kvm_irq_routing_entry *entries;
2621 if (copy_from_user(&routing, argp, sizeof(routing)))
2624 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2629 entries = vmalloc(routing.nr * sizeof(*entries));
2634 if (copy_from_user(entries, urouting->entries,
2635 routing.nr * sizeof(*entries)))
2636 goto out_free_irq_routing;
2637 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2639 out_free_irq_routing:
2643 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
2644 case KVM_CREATE_DEVICE: {
2645 struct kvm_create_device cd;
2648 if (copy_from_user(&cd, argp, sizeof(cd)))
2651 r = kvm_ioctl_create_device(kvm, &cd);
2656 if (copy_to_user(argp, &cd, sizeof(cd)))
2662 case KVM_CHECK_EXTENSION:
2663 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
2666 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2672 #ifdef CONFIG_KVM_COMPAT
2673 struct compat_kvm_dirty_log {
2677 compat_uptr_t dirty_bitmap; /* one bit per page */
2682 static long kvm_vm_compat_ioctl(struct file *filp,
2683 unsigned int ioctl, unsigned long arg)
2685 struct kvm *kvm = filp->private_data;
2688 if (kvm->mm != current->mm)
2691 case KVM_GET_DIRTY_LOG: {
2692 struct compat_kvm_dirty_log compat_log;
2693 struct kvm_dirty_log log;
2696 if (copy_from_user(&compat_log, (void __user *)arg,
2697 sizeof(compat_log)))
2699 log.slot = compat_log.slot;
2700 log.padding1 = compat_log.padding1;
2701 log.padding2 = compat_log.padding2;
2702 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2704 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2708 r = kvm_vm_ioctl(filp, ioctl, arg);
2716 static struct file_operations kvm_vm_fops = {
2717 .release = kvm_vm_release,
2718 .unlocked_ioctl = kvm_vm_ioctl,
2719 #ifdef CONFIG_KVM_COMPAT
2720 .compat_ioctl = kvm_vm_compat_ioctl,
2722 .llseek = noop_llseek,
2725 static int kvm_dev_ioctl_create_vm(unsigned long type)
2730 kvm = kvm_create_vm(type);
2732 return PTR_ERR(kvm);
2733 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2734 r = kvm_coalesced_mmio_init(kvm);
2740 r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC);
2747 static long kvm_dev_ioctl(struct file *filp,
2748 unsigned int ioctl, unsigned long arg)
2753 case KVM_GET_API_VERSION:
2756 r = KVM_API_VERSION;
2759 r = kvm_dev_ioctl_create_vm(arg);
2761 case KVM_CHECK_EXTENSION:
2762 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
2764 case KVM_GET_VCPU_MMAP_SIZE:
2767 r = PAGE_SIZE; /* struct kvm_run */
2769 r += PAGE_SIZE; /* pio data page */
2771 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2772 r += PAGE_SIZE; /* coalesced mmio ring page */
2775 case KVM_TRACE_ENABLE:
2776 case KVM_TRACE_PAUSE:
2777 case KVM_TRACE_DISABLE:
2781 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2787 static struct file_operations kvm_chardev_ops = {
2788 .unlocked_ioctl = kvm_dev_ioctl,
2789 .compat_ioctl = kvm_dev_ioctl,
2790 .llseek = noop_llseek,
2793 static struct miscdevice kvm_dev = {
2799 static void hardware_enable_nolock(void *junk)
2801 int cpu = raw_smp_processor_id();
2804 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2807 cpumask_set_cpu(cpu, cpus_hardware_enabled);
2809 r = kvm_arch_hardware_enable();
2812 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2813 atomic_inc(&hardware_enable_failed);
2814 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
2818 static void hardware_enable(void)
2820 raw_spin_lock(&kvm_count_lock);
2821 if (kvm_usage_count)
2822 hardware_enable_nolock(NULL);
2823 raw_spin_unlock(&kvm_count_lock);
2826 static void hardware_disable_nolock(void *junk)
2828 int cpu = raw_smp_processor_id();
2830 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2832 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2833 kvm_arch_hardware_disable();
2836 static void hardware_disable(void)
2838 raw_spin_lock(&kvm_count_lock);
2839 if (kvm_usage_count)
2840 hardware_disable_nolock(NULL);
2841 raw_spin_unlock(&kvm_count_lock);
2844 static void hardware_disable_all_nolock(void)
2846 BUG_ON(!kvm_usage_count);
2849 if (!kvm_usage_count)
2850 on_each_cpu(hardware_disable_nolock, NULL, 1);
2853 static void hardware_disable_all(void)
2855 raw_spin_lock(&kvm_count_lock);
2856 hardware_disable_all_nolock();
2857 raw_spin_unlock(&kvm_count_lock);
2860 static int hardware_enable_all(void)
2864 raw_spin_lock(&kvm_count_lock);
2867 if (kvm_usage_count == 1) {
2868 atomic_set(&hardware_enable_failed, 0);
2869 on_each_cpu(hardware_enable_nolock, NULL, 1);
2871 if (atomic_read(&hardware_enable_failed)) {
2872 hardware_disable_all_nolock();
2877 raw_spin_unlock(&kvm_count_lock);
2882 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2887 val &= ~CPU_TASKS_FROZEN;
2890 pr_info("kvm: disabling virtualization on CPU%d\n",
2895 pr_info("kvm: enabling virtualization on CPU%d\n",
2903 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2907 * Some (well, at least mine) BIOSes hang on reboot if
2910 * And Intel TXT required VMX off for all cpu when system shutdown.
2912 pr_info("kvm: exiting hardware virtualization\n");
2913 kvm_rebooting = true;
2914 on_each_cpu(hardware_disable_nolock, NULL, 1);
2918 static struct notifier_block kvm_reboot_notifier = {
2919 .notifier_call = kvm_reboot,
2923 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2927 for (i = 0; i < bus->dev_count; i++) {
2928 struct kvm_io_device *pos = bus->range[i].dev;
2930 kvm_iodevice_destructor(pos);
2935 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
2936 const struct kvm_io_range *r2)
2938 if (r1->addr < r2->addr)
2940 if (r1->addr + r1->len > r2->addr + r2->len)
2945 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2947 return kvm_io_bus_cmp(p1, p2);
2950 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2951 gpa_t addr, int len)
2953 bus->range[bus->dev_count++] = (struct kvm_io_range) {
2959 sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2960 kvm_io_bus_sort_cmp, NULL);
2965 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2966 gpa_t addr, int len)
2968 struct kvm_io_range *range, key;
2971 key = (struct kvm_io_range) {
2976 range = bsearch(&key, bus->range, bus->dev_count,
2977 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2981 off = range - bus->range;
2983 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
2989 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
2990 struct kvm_io_range *range, const void *val)
2994 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2998 while (idx < bus->dev_count &&
2999 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3000 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
3009 /* kvm_io_bus_write - called under kvm->slots_lock */
3010 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3011 int len, const void *val)
3013 struct kvm_io_bus *bus;
3014 struct kvm_io_range range;
3017 range = (struct kvm_io_range) {
3022 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3023 r = __kvm_io_bus_write(vcpu, bus, &range, val);
3024 return r < 0 ? r : 0;
3027 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3028 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
3029 gpa_t addr, int len, const void *val, long cookie)
3031 struct kvm_io_bus *bus;
3032 struct kvm_io_range range;
3034 range = (struct kvm_io_range) {
3039 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3041 /* First try the device referenced by cookie. */
3042 if ((cookie >= 0) && (cookie < bus->dev_count) &&
3043 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
3044 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
3049 * cookie contained garbage; fall back to search and return the
3050 * correct cookie value.
3052 return __kvm_io_bus_write(vcpu, bus, &range, val);
3055 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3056 struct kvm_io_range *range, void *val)
3060 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3064 while (idx < bus->dev_count &&
3065 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3066 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
3074 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
3076 /* kvm_io_bus_read - called under kvm->slots_lock */
3077 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3080 struct kvm_io_bus *bus;
3081 struct kvm_io_range range;
3084 range = (struct kvm_io_range) {
3089 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3090 r = __kvm_io_bus_read(vcpu, bus, &range, val);
3091 return r < 0 ? r : 0;
3095 /* Caller must hold slots_lock. */
3096 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3097 int len, struct kvm_io_device *dev)
3099 struct kvm_io_bus *new_bus, *bus;
3101 bus = kvm->buses[bus_idx];
3102 /* exclude ioeventfd which is limited by maximum fd */
3103 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3106 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
3107 sizeof(struct kvm_io_range)), GFP_KERNEL);
3110 memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
3111 sizeof(struct kvm_io_range)));
3112 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
3113 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3114 synchronize_srcu_expedited(&kvm->srcu);
3120 /* Caller must hold slots_lock. */
3121 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3122 struct kvm_io_device *dev)
3125 struct kvm_io_bus *new_bus, *bus;
3127 bus = kvm->buses[bus_idx];
3129 for (i = 0; i < bus->dev_count; i++)
3130 if (bus->range[i].dev == dev) {
3138 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3139 sizeof(struct kvm_io_range)), GFP_KERNEL);
3143 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3144 new_bus->dev_count--;
3145 memcpy(new_bus->range + i, bus->range + i + 1,
3146 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3148 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3149 synchronize_srcu_expedited(&kvm->srcu);
3154 static struct notifier_block kvm_cpu_notifier = {
3155 .notifier_call = kvm_cpu_hotplug,
3158 static int vm_stat_get(void *_offset, u64 *val)
3160 unsigned offset = (long)_offset;
3164 spin_lock(&kvm_lock);
3165 list_for_each_entry(kvm, &vm_list, vm_list)
3166 *val += *(u32 *)((void *)kvm + offset);
3167 spin_unlock(&kvm_lock);
3171 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
3173 static int vcpu_stat_get(void *_offset, u64 *val)
3175 unsigned offset = (long)_offset;
3177 struct kvm_vcpu *vcpu;
3181 spin_lock(&kvm_lock);
3182 list_for_each_entry(kvm, &vm_list, vm_list)
3183 kvm_for_each_vcpu(i, vcpu, kvm)
3184 *val += *(u32 *)((void *)vcpu + offset);
3186 spin_unlock(&kvm_lock);
3190 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
3192 static const struct file_operations *stat_fops[] = {
3193 [KVM_STAT_VCPU] = &vcpu_stat_fops,
3194 [KVM_STAT_VM] = &vm_stat_fops,
3197 static int kvm_init_debug(void)
3200 struct kvm_stats_debugfs_item *p;
3202 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3203 if (kvm_debugfs_dir == NULL)
3206 for (p = debugfs_entries; p->name; ++p) {
3207 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
3208 (void *)(long)p->offset,
3209 stat_fops[p->kind]);
3210 if (p->dentry == NULL)
3217 debugfs_remove_recursive(kvm_debugfs_dir);
3222 static void kvm_exit_debug(void)
3224 struct kvm_stats_debugfs_item *p;
3226 for (p = debugfs_entries; p->name; ++p)
3227 debugfs_remove(p->dentry);
3228 debugfs_remove(kvm_debugfs_dir);
3231 static int kvm_suspend(void)
3233 if (kvm_usage_count)
3234 hardware_disable_nolock(NULL);
3238 static void kvm_resume(void)
3240 if (kvm_usage_count) {
3241 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3242 hardware_enable_nolock(NULL);
3246 static struct syscore_ops kvm_syscore_ops = {
3247 .suspend = kvm_suspend,
3248 .resume = kvm_resume,
3252 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3254 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3257 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3259 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3261 if (vcpu->preempted)
3262 vcpu->preempted = false;
3264 kvm_arch_sched_in(vcpu, cpu);
3266 kvm_arch_vcpu_load(vcpu, cpu);
3269 static void kvm_sched_out(struct preempt_notifier *pn,
3270 struct task_struct *next)
3272 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3274 if (current->state == TASK_RUNNING)
3275 vcpu->preempted = true;
3276 kvm_arch_vcpu_put(vcpu);
3279 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3280 struct module *module)
3285 r = kvm_arch_init(opaque);
3290 * kvm_arch_init makes sure there's at most one caller
3291 * for architectures that support multiple implementations,
3292 * like intel and amd on x86.
3293 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3294 * conflicts in case kvm is already setup for another implementation.
3296 r = kvm_irqfd_init();
3300 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
3305 r = kvm_arch_hardware_setup();
3309 for_each_online_cpu(cpu) {
3310 smp_call_function_single(cpu,
3311 kvm_arch_check_processor_compat,
3317 r = register_cpu_notifier(&kvm_cpu_notifier);
3320 register_reboot_notifier(&kvm_reboot_notifier);
3322 /* A kmem cache lets us meet the alignment requirements of fx_save. */
3324 vcpu_align = __alignof__(struct kvm_vcpu);
3325 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
3327 if (!kvm_vcpu_cache) {
3332 r = kvm_async_pf_init();
3336 kvm_chardev_ops.owner = module;
3337 kvm_vm_fops.owner = module;
3338 kvm_vcpu_fops.owner = module;
3340 r = misc_register(&kvm_dev);
3342 pr_err("kvm: misc device register failed\n");
3346 register_syscore_ops(&kvm_syscore_ops);
3348 kvm_preempt_ops.sched_in = kvm_sched_in;
3349 kvm_preempt_ops.sched_out = kvm_sched_out;
3351 r = kvm_init_debug();
3353 pr_err("kvm: create debugfs files failed\n");
3357 r = kvm_vfio_ops_init();
3363 unregister_syscore_ops(&kvm_syscore_ops);
3364 misc_deregister(&kvm_dev);
3366 kvm_async_pf_deinit();
3368 kmem_cache_destroy(kvm_vcpu_cache);
3370 unregister_reboot_notifier(&kvm_reboot_notifier);
3371 unregister_cpu_notifier(&kvm_cpu_notifier);
3374 kvm_arch_hardware_unsetup();
3376 free_cpumask_var(cpus_hardware_enabled);
3384 EXPORT_SYMBOL_GPL(kvm_init);
3389 misc_deregister(&kvm_dev);
3390 kmem_cache_destroy(kvm_vcpu_cache);
3391 kvm_async_pf_deinit();
3392 unregister_syscore_ops(&kvm_syscore_ops);
3393 unregister_reboot_notifier(&kvm_reboot_notifier);
3394 unregister_cpu_notifier(&kvm_cpu_notifier);
3395 on_each_cpu(hardware_disable_nolock, NULL, 1);
3396 kvm_arch_hardware_unsetup();
3399 free_cpumask_var(cpus_hardware_enabled);
3400 kvm_vfio_ops_exit();
3402 EXPORT_SYMBOL_GPL(kvm_exit);