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 /* Architectures should define their poll value according to the halt latency */
70 static unsigned int halt_poll_ns = KVM_HALT_POLL_NS_DEFAULT;
71 module_param(halt_poll_ns, uint, S_IRUGO | S_IWUSR);
73 /* Default doubles per-vcpu halt_poll_ns. */
74 static unsigned int halt_poll_ns_grow = 2;
75 module_param(halt_poll_ns_grow, int, S_IRUGO);
77 /* Default resets per-vcpu halt_poll_ns . */
78 static unsigned int halt_poll_ns_shrink;
79 module_param(halt_poll_ns_shrink, int, S_IRUGO);
84 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
87 DEFINE_SPINLOCK(kvm_lock);
88 static DEFINE_RAW_SPINLOCK(kvm_count_lock);
91 static cpumask_var_t cpus_hardware_enabled;
92 static int kvm_usage_count;
93 static atomic_t hardware_enable_failed;
95 struct kmem_cache *kvm_vcpu_cache;
96 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
98 static __read_mostly struct preempt_ops kvm_preempt_ops;
100 struct dentry *kvm_debugfs_dir;
101 EXPORT_SYMBOL_GPL(kvm_debugfs_dir);
103 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
105 #ifdef CONFIG_KVM_COMPAT
106 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
109 static int hardware_enable_all(void);
110 static void hardware_disable_all(void);
112 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
114 static void kvm_release_pfn_dirty(kvm_pfn_t pfn);
115 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot, gfn_t gfn);
117 __visible bool kvm_rebooting;
118 EXPORT_SYMBOL_GPL(kvm_rebooting);
120 static bool largepages_enabled = true;
122 bool kvm_is_reserved_pfn(kvm_pfn_t pfn)
125 return PageReserved(pfn_to_page(pfn));
131 * Switches to specified vcpu, until a matching vcpu_put()
133 int vcpu_load(struct kvm_vcpu *vcpu)
137 if (mutex_lock_killable(&vcpu->mutex))
140 preempt_notifier_register(&vcpu->preempt_notifier);
141 kvm_arch_vcpu_load(vcpu, cpu);
146 void vcpu_put(struct kvm_vcpu *vcpu)
149 kvm_arch_vcpu_put(vcpu);
150 preempt_notifier_unregister(&vcpu->preempt_notifier);
152 mutex_unlock(&vcpu->mutex);
155 static void ack_flush(void *_completed)
159 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
164 struct kvm_vcpu *vcpu;
166 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
169 kvm_for_each_vcpu(i, vcpu, kvm) {
170 kvm_make_request(req, vcpu);
173 /* Set ->requests bit before we read ->mode */
176 if (cpus != NULL && cpu != -1 && cpu != me &&
177 kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
178 cpumask_set_cpu(cpu, cpus);
180 if (unlikely(cpus == NULL))
181 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
182 else if (!cpumask_empty(cpus))
183 smp_call_function_many(cpus, ack_flush, NULL, 1);
187 free_cpumask_var(cpus);
191 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
192 void kvm_flush_remote_tlbs(struct kvm *kvm)
194 long dirty_count = kvm->tlbs_dirty;
197 if (kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
198 ++kvm->stat.remote_tlb_flush;
199 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
201 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
204 void kvm_reload_remote_mmus(struct kvm *kvm)
206 kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
209 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
214 mutex_init(&vcpu->mutex);
219 init_swait_queue_head(&vcpu->wq);
220 kvm_async_pf_vcpu_init(vcpu);
223 INIT_LIST_HEAD(&vcpu->blocked_vcpu_list);
225 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
230 vcpu->run = page_address(page);
232 kvm_vcpu_set_in_spin_loop(vcpu, false);
233 kvm_vcpu_set_dy_eligible(vcpu, false);
234 vcpu->preempted = false;
236 r = kvm_arch_vcpu_init(vcpu);
242 free_page((unsigned long)vcpu->run);
246 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
248 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
251 kvm_arch_vcpu_uninit(vcpu);
252 free_page((unsigned long)vcpu->run);
254 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
256 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
257 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
259 return container_of(mn, struct kvm, mmu_notifier);
262 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
263 struct mm_struct *mm,
264 unsigned long address)
266 struct kvm *kvm = mmu_notifier_to_kvm(mn);
267 int need_tlb_flush, idx;
270 * When ->invalidate_page runs, the linux pte has been zapped
271 * already but the page is still allocated until
272 * ->invalidate_page returns. So if we increase the sequence
273 * here the kvm page fault will notice if the spte can't be
274 * established because the page is going to be freed. If
275 * instead the kvm page fault establishes the spte before
276 * ->invalidate_page runs, kvm_unmap_hva will release it
279 * The sequence increase only need to be seen at spin_unlock
280 * time, and not at spin_lock time.
282 * Increasing the sequence after the spin_unlock would be
283 * unsafe because the kvm page fault could then establish the
284 * pte after kvm_unmap_hva returned, without noticing the page
285 * is going to be freed.
287 idx = srcu_read_lock(&kvm->srcu);
288 spin_lock(&kvm->mmu_lock);
290 kvm->mmu_notifier_seq++;
291 need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
292 /* we've to flush the tlb before the pages can be freed */
294 kvm_flush_remote_tlbs(kvm);
296 spin_unlock(&kvm->mmu_lock);
298 kvm_arch_mmu_notifier_invalidate_page(kvm, address);
300 srcu_read_unlock(&kvm->srcu, idx);
303 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
304 struct mm_struct *mm,
305 unsigned long address,
308 struct kvm *kvm = mmu_notifier_to_kvm(mn);
311 idx = srcu_read_lock(&kvm->srcu);
312 spin_lock(&kvm->mmu_lock);
313 kvm->mmu_notifier_seq++;
314 kvm_set_spte_hva(kvm, address, pte);
315 spin_unlock(&kvm->mmu_lock);
316 srcu_read_unlock(&kvm->srcu, idx);
319 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
320 struct mm_struct *mm,
324 struct kvm *kvm = mmu_notifier_to_kvm(mn);
325 int need_tlb_flush = 0, idx;
327 idx = srcu_read_lock(&kvm->srcu);
328 spin_lock(&kvm->mmu_lock);
330 * The count increase must become visible at unlock time as no
331 * spte can be established without taking the mmu_lock and
332 * count is also read inside the mmu_lock critical section.
334 kvm->mmu_notifier_count++;
335 need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
336 need_tlb_flush |= kvm->tlbs_dirty;
337 /* we've to flush the tlb before the pages can be freed */
339 kvm_flush_remote_tlbs(kvm);
341 spin_unlock(&kvm->mmu_lock);
342 srcu_read_unlock(&kvm->srcu, idx);
345 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
346 struct mm_struct *mm,
350 struct kvm *kvm = mmu_notifier_to_kvm(mn);
352 spin_lock(&kvm->mmu_lock);
354 * This sequence increase will notify the kvm page fault that
355 * the page that is going to be mapped in the spte could have
358 kvm->mmu_notifier_seq++;
361 * The above sequence increase must be visible before the
362 * below count decrease, which is ensured by the smp_wmb above
363 * in conjunction with the smp_rmb in mmu_notifier_retry().
365 kvm->mmu_notifier_count--;
366 spin_unlock(&kvm->mmu_lock);
368 BUG_ON(kvm->mmu_notifier_count < 0);
371 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
372 struct mm_struct *mm,
376 struct kvm *kvm = mmu_notifier_to_kvm(mn);
379 idx = srcu_read_lock(&kvm->srcu);
380 spin_lock(&kvm->mmu_lock);
382 young = kvm_age_hva(kvm, start, end);
384 kvm_flush_remote_tlbs(kvm);
386 spin_unlock(&kvm->mmu_lock);
387 srcu_read_unlock(&kvm->srcu, idx);
392 static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
393 struct mm_struct *mm,
397 struct kvm *kvm = mmu_notifier_to_kvm(mn);
400 idx = srcu_read_lock(&kvm->srcu);
401 spin_lock(&kvm->mmu_lock);
403 * Even though we do not flush TLB, this will still adversely
404 * affect performance on pre-Haswell Intel EPT, where there is
405 * no EPT Access Bit to clear so that we have to tear down EPT
406 * tables instead. If we find this unacceptable, we can always
407 * add a parameter to kvm_age_hva so that it effectively doesn't
408 * do anything on clear_young.
410 * Also note that currently we never issue secondary TLB flushes
411 * from clear_young, leaving this job up to the regular system
412 * cadence. If we find this inaccurate, we might come up with a
413 * more sophisticated heuristic later.
415 young = kvm_age_hva(kvm, start, end);
416 spin_unlock(&kvm->mmu_lock);
417 srcu_read_unlock(&kvm->srcu, idx);
422 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
423 struct mm_struct *mm,
424 unsigned long address)
426 struct kvm *kvm = mmu_notifier_to_kvm(mn);
429 idx = srcu_read_lock(&kvm->srcu);
430 spin_lock(&kvm->mmu_lock);
431 young = kvm_test_age_hva(kvm, address);
432 spin_unlock(&kvm->mmu_lock);
433 srcu_read_unlock(&kvm->srcu, idx);
438 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
439 struct mm_struct *mm)
441 struct kvm *kvm = mmu_notifier_to_kvm(mn);
444 idx = srcu_read_lock(&kvm->srcu);
445 kvm_arch_flush_shadow_all(kvm);
446 srcu_read_unlock(&kvm->srcu, idx);
449 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
450 .invalidate_page = kvm_mmu_notifier_invalidate_page,
451 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
452 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
453 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
454 .clear_young = kvm_mmu_notifier_clear_young,
455 .test_young = kvm_mmu_notifier_test_young,
456 .change_pte = kvm_mmu_notifier_change_pte,
457 .release = kvm_mmu_notifier_release,
460 static int kvm_init_mmu_notifier(struct kvm *kvm)
462 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
463 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
466 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
468 static int kvm_init_mmu_notifier(struct kvm *kvm)
473 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
475 static struct kvm_memslots *kvm_alloc_memslots(void)
478 struct kvm_memslots *slots;
480 slots = kvm_kvzalloc(sizeof(struct kvm_memslots));
485 * Init kvm generation close to the maximum to easily test the
486 * code of handling generation number wrap-around.
488 slots->generation = -150;
489 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
490 slots->id_to_index[i] = slots->memslots[i].id = i;
495 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
497 if (!memslot->dirty_bitmap)
500 kvfree(memslot->dirty_bitmap);
501 memslot->dirty_bitmap = NULL;
505 * Free any memory in @free but not in @dont.
507 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
508 struct kvm_memory_slot *dont)
510 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
511 kvm_destroy_dirty_bitmap(free);
513 kvm_arch_free_memslot(kvm, free, dont);
518 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
520 struct kvm_memory_slot *memslot;
525 kvm_for_each_memslot(memslot, slots)
526 kvm_free_memslot(kvm, memslot, NULL);
531 static struct kvm *kvm_create_vm(unsigned long type)
534 struct kvm *kvm = kvm_arch_alloc_vm();
537 return ERR_PTR(-ENOMEM);
539 r = kvm_arch_init_vm(kvm, type);
541 goto out_err_no_disable;
543 r = hardware_enable_all();
545 goto out_err_no_disable;
547 #ifdef CONFIG_HAVE_KVM_IRQFD
548 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
551 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
554 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
555 kvm->memslots[i] = kvm_alloc_memslots();
556 if (!kvm->memslots[i])
557 goto out_err_no_srcu;
560 if (init_srcu_struct(&kvm->srcu))
561 goto out_err_no_srcu;
562 if (init_srcu_struct(&kvm->irq_srcu))
563 goto out_err_no_irq_srcu;
564 for (i = 0; i < KVM_NR_BUSES; i++) {
565 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
571 spin_lock_init(&kvm->mmu_lock);
572 kvm->mm = current->mm;
573 atomic_inc(&kvm->mm->mm_count);
574 kvm_eventfd_init(kvm);
575 mutex_init(&kvm->lock);
576 mutex_init(&kvm->irq_lock);
577 mutex_init(&kvm->slots_lock);
578 atomic_set(&kvm->users_count, 1);
579 INIT_LIST_HEAD(&kvm->devices);
581 r = kvm_init_mmu_notifier(kvm);
585 spin_lock(&kvm_lock);
586 list_add(&kvm->vm_list, &vm_list);
587 spin_unlock(&kvm_lock);
589 preempt_notifier_inc();
594 cleanup_srcu_struct(&kvm->irq_srcu);
596 cleanup_srcu_struct(&kvm->srcu);
598 hardware_disable_all();
600 for (i = 0; i < KVM_NR_BUSES; i++)
601 kfree(kvm->buses[i]);
602 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
603 kvm_free_memslots(kvm, kvm->memslots[i]);
604 kvm_arch_free_vm(kvm);
609 * Avoid using vmalloc for a small buffer.
610 * Should not be used when the size is statically known.
612 void *kvm_kvzalloc(unsigned long size)
614 if (size > PAGE_SIZE)
615 return vzalloc(size);
617 return kzalloc(size, GFP_KERNEL);
620 static void kvm_destroy_devices(struct kvm *kvm)
622 struct list_head *node, *tmp;
624 list_for_each_safe(node, tmp, &kvm->devices) {
625 struct kvm_device *dev =
626 list_entry(node, struct kvm_device, vm_node);
629 dev->ops->destroy(dev);
633 static void kvm_destroy_vm(struct kvm *kvm)
636 struct mm_struct *mm = kvm->mm;
638 kvm_arch_sync_events(kvm);
639 spin_lock(&kvm_lock);
640 list_del(&kvm->vm_list);
641 spin_unlock(&kvm_lock);
642 kvm_free_irq_routing(kvm);
643 for (i = 0; i < KVM_NR_BUSES; i++)
644 kvm_io_bus_destroy(kvm->buses[i]);
645 kvm_coalesced_mmio_free(kvm);
646 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
647 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
649 kvm_arch_flush_shadow_all(kvm);
651 kvm_arch_destroy_vm(kvm);
652 kvm_destroy_devices(kvm);
653 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
654 kvm_free_memslots(kvm, kvm->memslots[i]);
655 cleanup_srcu_struct(&kvm->irq_srcu);
656 cleanup_srcu_struct(&kvm->srcu);
657 kvm_arch_free_vm(kvm);
658 preempt_notifier_dec();
659 hardware_disable_all();
663 void kvm_get_kvm(struct kvm *kvm)
665 atomic_inc(&kvm->users_count);
667 EXPORT_SYMBOL_GPL(kvm_get_kvm);
669 void kvm_put_kvm(struct kvm *kvm)
671 if (atomic_dec_and_test(&kvm->users_count))
674 EXPORT_SYMBOL_GPL(kvm_put_kvm);
677 static int kvm_vm_release(struct inode *inode, struct file *filp)
679 struct kvm *kvm = filp->private_data;
681 kvm_irqfd_release(kvm);
688 * Allocation size is twice as large as the actual dirty bitmap size.
689 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
691 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
693 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
695 memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
696 if (!memslot->dirty_bitmap)
703 * Insert memslot and re-sort memslots based on their GFN,
704 * so binary search could be used to lookup GFN.
705 * Sorting algorithm takes advantage of having initially
706 * sorted array and known changed memslot position.
708 static void update_memslots(struct kvm_memslots *slots,
709 struct kvm_memory_slot *new)
712 int i = slots->id_to_index[id];
713 struct kvm_memory_slot *mslots = slots->memslots;
715 WARN_ON(mslots[i].id != id);
717 WARN_ON(!mslots[i].npages);
718 if (mslots[i].npages)
721 if (!mslots[i].npages)
725 while (i < KVM_MEM_SLOTS_NUM - 1 &&
726 new->base_gfn <= mslots[i + 1].base_gfn) {
727 if (!mslots[i + 1].npages)
729 mslots[i] = mslots[i + 1];
730 slots->id_to_index[mslots[i].id] = i;
735 * The ">=" is needed when creating a slot with base_gfn == 0,
736 * so that it moves before all those with base_gfn == npages == 0.
738 * On the other hand, if new->npages is zero, the above loop has
739 * already left i pointing to the beginning of the empty part of
740 * mslots, and the ">=" would move the hole backwards in this
741 * case---which is wrong. So skip the loop when deleting a slot.
745 new->base_gfn >= mslots[i - 1].base_gfn) {
746 mslots[i] = mslots[i - 1];
747 slots->id_to_index[mslots[i].id] = i;
751 WARN_ON_ONCE(i != slots->used_slots);
754 slots->id_to_index[mslots[i].id] = i;
757 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
759 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
761 #ifdef __KVM_HAVE_READONLY_MEM
762 valid_flags |= KVM_MEM_READONLY;
765 if (mem->flags & ~valid_flags)
771 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
772 int as_id, struct kvm_memslots *slots)
774 struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id);
777 * Set the low bit in the generation, which disables SPTE caching
778 * until the end of synchronize_srcu_expedited.
780 WARN_ON(old_memslots->generation & 1);
781 slots->generation = old_memslots->generation + 1;
783 rcu_assign_pointer(kvm->memslots[as_id], slots);
784 synchronize_srcu_expedited(&kvm->srcu);
787 * Increment the new memslot generation a second time. This prevents
788 * vm exits that race with memslot updates from caching a memslot
789 * generation that will (potentially) be valid forever.
793 kvm_arch_memslots_updated(kvm, slots);
799 * Allocate some memory and give it an address in the guest physical address
802 * Discontiguous memory is allowed, mostly for framebuffers.
804 * Must be called holding kvm->slots_lock for write.
806 int __kvm_set_memory_region(struct kvm *kvm,
807 const struct kvm_userspace_memory_region *mem)
811 unsigned long npages;
812 struct kvm_memory_slot *slot;
813 struct kvm_memory_slot old, new;
814 struct kvm_memslots *slots = NULL, *old_memslots;
816 enum kvm_mr_change change;
818 r = check_memory_region_flags(mem);
823 as_id = mem->slot >> 16;
826 /* General sanity checks */
827 if (mem->memory_size & (PAGE_SIZE - 1))
829 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
831 /* We can read the guest memory with __xxx_user() later on. */
832 if ((id < KVM_USER_MEM_SLOTS) &&
833 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
834 !access_ok(VERIFY_WRITE,
835 (void __user *)(unsigned long)mem->userspace_addr,
838 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
840 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
843 slot = id_to_memslot(__kvm_memslots(kvm, as_id), id);
844 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
845 npages = mem->memory_size >> PAGE_SHIFT;
847 if (npages > KVM_MEM_MAX_NR_PAGES)
853 new.base_gfn = base_gfn;
855 new.flags = mem->flags;
859 change = KVM_MR_CREATE;
860 else { /* Modify an existing slot. */
861 if ((mem->userspace_addr != old.userspace_addr) ||
862 (npages != old.npages) ||
863 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
866 if (base_gfn != old.base_gfn)
867 change = KVM_MR_MOVE;
868 else if (new.flags != old.flags)
869 change = KVM_MR_FLAGS_ONLY;
870 else { /* Nothing to change. */
879 change = KVM_MR_DELETE;
884 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
885 /* Check for overlaps */
887 kvm_for_each_memslot(slot, __kvm_memslots(kvm, as_id)) {
888 if ((slot->id >= KVM_USER_MEM_SLOTS) ||
891 if (!((base_gfn + npages <= slot->base_gfn) ||
892 (base_gfn >= slot->base_gfn + slot->npages)))
897 /* Free page dirty bitmap if unneeded */
898 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
899 new.dirty_bitmap = NULL;
902 if (change == KVM_MR_CREATE) {
903 new.userspace_addr = mem->userspace_addr;
905 if (kvm_arch_create_memslot(kvm, &new, npages))
909 /* Allocate page dirty bitmap if needed */
910 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
911 if (kvm_create_dirty_bitmap(&new) < 0)
915 slots = kvm_kvzalloc(sizeof(struct kvm_memslots));
918 memcpy(slots, __kvm_memslots(kvm, as_id), sizeof(struct kvm_memslots));
920 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
921 slot = id_to_memslot(slots, id);
922 slot->flags |= KVM_MEMSLOT_INVALID;
924 old_memslots = install_new_memslots(kvm, as_id, slots);
926 /* slot was deleted or moved, clear iommu mapping */
927 kvm_iommu_unmap_pages(kvm, &old);
928 /* From this point no new shadow pages pointing to a deleted,
929 * or moved, memslot will be created.
931 * validation of sp->gfn happens in:
932 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
933 * - kvm_is_visible_gfn (mmu_check_roots)
935 kvm_arch_flush_shadow_memslot(kvm, slot);
938 * We can re-use the old_memslots from above, the only difference
939 * from the currently installed memslots is the invalid flag. This
940 * will get overwritten by update_memslots anyway.
942 slots = old_memslots;
945 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
949 /* actual memory is freed via old in kvm_free_memslot below */
950 if (change == KVM_MR_DELETE) {
951 new.dirty_bitmap = NULL;
952 memset(&new.arch, 0, sizeof(new.arch));
955 update_memslots(slots, &new);
956 old_memslots = install_new_memslots(kvm, as_id, slots);
958 kvm_arch_commit_memory_region(kvm, mem, &old, &new, change);
960 kvm_free_memslot(kvm, &old, &new);
961 kvfree(old_memslots);
964 * IOMMU mapping: New slots need to be mapped. Old slots need to be
965 * un-mapped and re-mapped if their base changes. Since base change
966 * unmapping is handled above with slot deletion, mapping alone is
967 * needed here. Anything else the iommu might care about for existing
968 * slots (size changes, userspace addr changes and read-only flag
969 * changes) is disallowed above, so any other attribute changes getting
970 * here can be skipped.
972 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
973 r = kvm_iommu_map_pages(kvm, &new);
982 kvm_free_memslot(kvm, &new, &old);
986 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
988 int kvm_set_memory_region(struct kvm *kvm,
989 const struct kvm_userspace_memory_region *mem)
993 mutex_lock(&kvm->slots_lock);
994 r = __kvm_set_memory_region(kvm, mem);
995 mutex_unlock(&kvm->slots_lock);
998 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1000 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1001 struct kvm_userspace_memory_region *mem)
1003 if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1006 return kvm_set_memory_region(kvm, mem);
1009 int kvm_get_dirty_log(struct kvm *kvm,
1010 struct kvm_dirty_log *log, int *is_dirty)
1012 struct kvm_memslots *slots;
1013 struct kvm_memory_slot *memslot;
1014 int r, i, as_id, id;
1016 unsigned long any = 0;
1019 as_id = log->slot >> 16;
1020 id = (u16)log->slot;
1021 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1024 slots = __kvm_memslots(kvm, as_id);
1025 memslot = id_to_memslot(slots, id);
1027 if (!memslot->dirty_bitmap)
1030 n = kvm_dirty_bitmap_bytes(memslot);
1032 for (i = 0; !any && i < n/sizeof(long); ++i)
1033 any = memslot->dirty_bitmap[i];
1036 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1046 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1048 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1050 * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages
1051 * are dirty write protect them for next write.
1052 * @kvm: pointer to kvm instance
1053 * @log: slot id and address to which we copy the log
1054 * @is_dirty: flag set if any page is dirty
1056 * We need to keep it in mind that VCPU threads can write to the bitmap
1057 * concurrently. So, to avoid losing track of dirty pages we keep the
1060 * 1. Take a snapshot of the bit and clear it if needed.
1061 * 2. Write protect the corresponding page.
1062 * 3. Copy the snapshot to the userspace.
1063 * 4. Upon return caller flushes TLB's if needed.
1065 * Between 2 and 4, the guest may write to the page using the remaining TLB
1066 * entry. This is not a problem because the page is reported dirty using
1067 * the snapshot taken before and step 4 ensures that writes done after
1068 * exiting to userspace will be logged for the next call.
1071 int kvm_get_dirty_log_protect(struct kvm *kvm,
1072 struct kvm_dirty_log *log, bool *is_dirty)
1074 struct kvm_memslots *slots;
1075 struct kvm_memory_slot *memslot;
1076 int r, i, as_id, id;
1078 unsigned long *dirty_bitmap;
1079 unsigned long *dirty_bitmap_buffer;
1082 as_id = log->slot >> 16;
1083 id = (u16)log->slot;
1084 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1087 slots = __kvm_memslots(kvm, as_id);
1088 memslot = id_to_memslot(slots, id);
1090 dirty_bitmap = memslot->dirty_bitmap;
1095 n = kvm_dirty_bitmap_bytes(memslot);
1097 dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long);
1098 memset(dirty_bitmap_buffer, 0, n);
1100 spin_lock(&kvm->mmu_lock);
1102 for (i = 0; i < n / sizeof(long); i++) {
1106 if (!dirty_bitmap[i])
1111 mask = xchg(&dirty_bitmap[i], 0);
1112 dirty_bitmap_buffer[i] = mask;
1115 offset = i * BITS_PER_LONG;
1116 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1121 spin_unlock(&kvm->mmu_lock);
1124 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1131 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect);
1134 bool kvm_largepages_enabled(void)
1136 return largepages_enabled;
1139 void kvm_disable_largepages(void)
1141 largepages_enabled = false;
1143 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1145 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1147 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1149 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1151 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
1153 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn);
1156 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1158 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1160 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1161 memslot->flags & KVM_MEMSLOT_INVALID)
1166 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1168 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1170 struct vm_area_struct *vma;
1171 unsigned long addr, size;
1175 addr = gfn_to_hva(kvm, gfn);
1176 if (kvm_is_error_hva(addr))
1179 down_read(¤t->mm->mmap_sem);
1180 vma = find_vma(current->mm, addr);
1184 size = vma_kernel_pagesize(vma);
1187 up_read(¤t->mm->mmap_sem);
1192 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1194 return slot->flags & KVM_MEM_READONLY;
1197 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1198 gfn_t *nr_pages, bool write)
1200 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1201 return KVM_HVA_ERR_BAD;
1203 if (memslot_is_readonly(slot) && write)
1204 return KVM_HVA_ERR_RO_BAD;
1207 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1209 return __gfn_to_hva_memslot(slot, gfn);
1212 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1215 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1218 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1221 return gfn_to_hva_many(slot, gfn, NULL);
1223 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1225 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1227 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1229 EXPORT_SYMBOL_GPL(gfn_to_hva);
1231 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
1233 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
1235 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
1238 * If writable is set to false, the hva returned by this function is only
1239 * allowed to be read.
1241 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1242 gfn_t gfn, bool *writable)
1244 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1246 if (!kvm_is_error_hva(hva) && writable)
1247 *writable = !memslot_is_readonly(slot);
1252 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1254 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1256 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1259 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
1261 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1263 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1266 static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1267 unsigned long start, int write, struct page **page)
1269 int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1272 flags |= FOLL_WRITE;
1274 return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1277 static inline int check_user_page_hwpoison(unsigned long addr)
1279 int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1281 rc = __get_user_pages(current, current->mm, addr, 1,
1282 flags, NULL, NULL, NULL);
1283 return rc == -EHWPOISON;
1287 * The atomic path to get the writable pfn which will be stored in @pfn,
1288 * true indicates success, otherwise false is returned.
1290 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1291 bool write_fault, bool *writable, kvm_pfn_t *pfn)
1293 struct page *page[1];
1296 if (!(async || atomic))
1300 * Fast pin a writable pfn only if it is a write fault request
1301 * or the caller allows to map a writable pfn for a read fault
1304 if (!(write_fault || writable))
1307 npages = __get_user_pages_fast(addr, 1, 1, page);
1309 *pfn = page_to_pfn(page[0]);
1320 * The slow path to get the pfn of the specified host virtual address,
1321 * 1 indicates success, -errno is returned if error is detected.
1323 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1324 bool *writable, kvm_pfn_t *pfn)
1326 struct page *page[1];
1332 *writable = write_fault;
1335 down_read(¤t->mm->mmap_sem);
1336 npages = get_user_page_nowait(current, current->mm,
1337 addr, write_fault, page);
1338 up_read(¤t->mm->mmap_sem);
1340 npages = __get_user_pages_unlocked(current, current->mm, addr, 1,
1341 write_fault, 0, page,
1342 FOLL_TOUCH|FOLL_HWPOISON);
1346 /* map read fault as writable if possible */
1347 if (unlikely(!write_fault) && writable) {
1348 struct page *wpage[1];
1350 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1359 *pfn = page_to_pfn(page[0]);
1363 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1365 if (unlikely(!(vma->vm_flags & VM_READ)))
1368 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1375 * Pin guest page in memory and return its pfn.
1376 * @addr: host virtual address which maps memory to the guest
1377 * @atomic: whether this function can sleep
1378 * @async: whether this function need to wait IO complete if the
1379 * host page is not in the memory
1380 * @write_fault: whether we should get a writable host page
1381 * @writable: whether it allows to map a writable host page for !@write_fault
1383 * The function will map a writable host page for these two cases:
1384 * 1): @write_fault = true
1385 * 2): @write_fault = false && @writable, @writable will tell the caller
1386 * whether the mapping is writable.
1388 static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1389 bool write_fault, bool *writable)
1391 struct vm_area_struct *vma;
1395 /* we can do it either atomically or asynchronously, not both */
1396 BUG_ON(atomic && async);
1398 if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1402 return KVM_PFN_ERR_FAULT;
1404 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1408 down_read(¤t->mm->mmap_sem);
1409 if (npages == -EHWPOISON ||
1410 (!async && check_user_page_hwpoison(addr))) {
1411 pfn = KVM_PFN_ERR_HWPOISON;
1415 vma = find_vma_intersection(current->mm, addr, addr + 1);
1418 pfn = KVM_PFN_ERR_FAULT;
1419 else if ((vma->vm_flags & VM_PFNMAP)) {
1420 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1422 BUG_ON(!kvm_is_reserved_pfn(pfn));
1424 if (async && vma_is_valid(vma, write_fault))
1426 pfn = KVM_PFN_ERR_FAULT;
1429 up_read(¤t->mm->mmap_sem);
1433 kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn,
1434 bool atomic, bool *async, bool write_fault,
1437 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1439 if (addr == KVM_HVA_ERR_RO_BAD)
1440 return KVM_PFN_ERR_RO_FAULT;
1442 if (kvm_is_error_hva(addr))
1443 return KVM_PFN_NOSLOT;
1445 /* Do not map writable pfn in the readonly memslot. */
1446 if (writable && memslot_is_readonly(slot)) {
1451 return hva_to_pfn(addr, atomic, async, write_fault,
1454 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
1456 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1459 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
1460 write_fault, writable);
1462 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1464 kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1466 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1468 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
1470 kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1472 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1474 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1476 kvm_pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1478 return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm, gfn), gfn);
1480 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1482 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
1484 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1486 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
1488 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1490 return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
1492 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1494 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
1496 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1498 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
1500 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1501 struct page **pages, int nr_pages)
1506 addr = gfn_to_hva_many(slot, gfn, &entry);
1507 if (kvm_is_error_hva(addr))
1510 if (entry < nr_pages)
1513 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1515 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1517 static struct page *kvm_pfn_to_page(kvm_pfn_t pfn)
1519 if (is_error_noslot_pfn(pfn))
1520 return KVM_ERR_PTR_BAD_PAGE;
1522 if (kvm_is_reserved_pfn(pfn)) {
1524 return KVM_ERR_PTR_BAD_PAGE;
1527 return pfn_to_page(pfn);
1530 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1534 pfn = gfn_to_pfn(kvm, gfn);
1536 return kvm_pfn_to_page(pfn);
1538 EXPORT_SYMBOL_GPL(gfn_to_page);
1540 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
1544 pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
1546 return kvm_pfn_to_page(pfn);
1548 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
1550 void kvm_release_page_clean(struct page *page)
1552 WARN_ON(is_error_page(page));
1554 kvm_release_pfn_clean(page_to_pfn(page));
1556 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1558 void kvm_release_pfn_clean(kvm_pfn_t pfn)
1560 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
1561 put_page(pfn_to_page(pfn));
1563 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1565 void kvm_release_page_dirty(struct page *page)
1567 WARN_ON(is_error_page(page));
1569 kvm_release_pfn_dirty(page_to_pfn(page));
1571 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1573 static void kvm_release_pfn_dirty(kvm_pfn_t pfn)
1575 kvm_set_pfn_dirty(pfn);
1576 kvm_release_pfn_clean(pfn);
1579 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
1581 if (!kvm_is_reserved_pfn(pfn)) {
1582 struct page *page = pfn_to_page(pfn);
1584 if (!PageReserved(page))
1588 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1590 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
1592 if (!kvm_is_reserved_pfn(pfn))
1593 mark_page_accessed(pfn_to_page(pfn));
1595 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1597 void kvm_get_pfn(kvm_pfn_t pfn)
1599 if (!kvm_is_reserved_pfn(pfn))
1600 get_page(pfn_to_page(pfn));
1602 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1604 static int next_segment(unsigned long len, int offset)
1606 if (len > PAGE_SIZE - offset)
1607 return PAGE_SIZE - offset;
1612 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
1613 void *data, int offset, int len)
1618 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1619 if (kvm_is_error_hva(addr))
1621 r = __copy_from_user(data, (void __user *)addr + offset, len);
1627 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1630 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1632 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1634 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1636 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
1637 int offset, int len)
1639 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1641 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1643 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
1645 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1647 gfn_t gfn = gpa >> PAGE_SHIFT;
1649 int offset = offset_in_page(gpa);
1652 while ((seg = next_segment(len, offset)) != 0) {
1653 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1663 EXPORT_SYMBOL_GPL(kvm_read_guest);
1665 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
1667 gfn_t gfn = gpa >> PAGE_SHIFT;
1669 int offset = offset_in_page(gpa);
1672 while ((seg = next_segment(len, offset)) != 0) {
1673 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
1683 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
1685 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1686 void *data, int offset, unsigned long len)
1691 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1692 if (kvm_is_error_hva(addr))
1694 pagefault_disable();
1695 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1702 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1705 gfn_t gfn = gpa >> PAGE_SHIFT;
1706 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1707 int offset = offset_in_page(gpa);
1709 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1711 EXPORT_SYMBOL_GPL(kvm_read_guest_atomic);
1713 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
1714 void *data, unsigned long len)
1716 gfn_t gfn = gpa >> PAGE_SHIFT;
1717 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1718 int offset = offset_in_page(gpa);
1720 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1722 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
1724 static int __kvm_write_guest_page(struct kvm_memory_slot *memslot, gfn_t gfn,
1725 const void *data, int offset, int len)
1730 addr = gfn_to_hva_memslot(memslot, gfn);
1731 if (kvm_is_error_hva(addr))
1733 r = __copy_to_user((void __user *)addr + offset, data, len);
1736 mark_page_dirty_in_slot(memslot, gfn);
1740 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
1741 const void *data, int offset, int len)
1743 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1745 return __kvm_write_guest_page(slot, gfn, data, offset, len);
1747 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1749 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
1750 const void *data, int offset, int len)
1752 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1754 return __kvm_write_guest_page(slot, gfn, data, offset, len);
1756 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
1758 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1761 gfn_t gfn = gpa >> PAGE_SHIFT;
1763 int offset = offset_in_page(gpa);
1766 while ((seg = next_segment(len, offset)) != 0) {
1767 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1777 EXPORT_SYMBOL_GPL(kvm_write_guest);
1779 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
1782 gfn_t gfn = gpa >> PAGE_SHIFT;
1784 int offset = offset_in_page(gpa);
1787 while ((seg = next_segment(len, offset)) != 0) {
1788 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
1798 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
1800 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1801 gpa_t gpa, unsigned long len)
1803 struct kvm_memslots *slots = kvm_memslots(kvm);
1804 int offset = offset_in_page(gpa);
1805 gfn_t start_gfn = gpa >> PAGE_SHIFT;
1806 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1807 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1808 gfn_t nr_pages_avail;
1811 ghc->generation = slots->generation;
1813 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1814 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, NULL);
1815 if (!kvm_is_error_hva(ghc->hva) && nr_pages_needed <= 1) {
1819 * If the requested region crosses two memslots, we still
1820 * verify that the entire region is valid here.
1822 while (start_gfn <= end_gfn) {
1823 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1824 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1826 if (kvm_is_error_hva(ghc->hva))
1828 start_gfn += nr_pages_avail;
1830 /* Use the slow path for cross page reads and writes. */
1831 ghc->memslot = NULL;
1835 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1837 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1838 void *data, unsigned long len)
1840 struct kvm_memslots *slots = kvm_memslots(kvm);
1843 BUG_ON(len > ghc->len);
1845 if (slots->generation != ghc->generation)
1846 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1848 if (unlikely(!ghc->memslot))
1849 return kvm_write_guest(kvm, ghc->gpa, data, len);
1851 if (kvm_is_error_hva(ghc->hva))
1854 r = __copy_to_user((void __user *)ghc->hva, data, len);
1857 mark_page_dirty_in_slot(ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1861 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1863 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1864 void *data, unsigned long len)
1866 struct kvm_memslots *slots = kvm_memslots(kvm);
1869 BUG_ON(len > ghc->len);
1871 if (slots->generation != ghc->generation)
1872 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1874 if (unlikely(!ghc->memslot))
1875 return kvm_read_guest(kvm, ghc->gpa, data, len);
1877 if (kvm_is_error_hva(ghc->hva))
1880 r = __copy_from_user(data, (void __user *)ghc->hva, len);
1886 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1888 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1890 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
1892 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
1894 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1896 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1898 gfn_t gfn = gpa >> PAGE_SHIFT;
1900 int offset = offset_in_page(gpa);
1903 while ((seg = next_segment(len, offset)) != 0) {
1904 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1913 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1915 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot,
1918 if (memslot && memslot->dirty_bitmap) {
1919 unsigned long rel_gfn = gfn - memslot->base_gfn;
1921 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1925 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1927 struct kvm_memory_slot *memslot;
1929 memslot = gfn_to_memslot(kvm, gfn);
1930 mark_page_dirty_in_slot(memslot, gfn);
1932 EXPORT_SYMBOL_GPL(mark_page_dirty);
1934 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
1936 struct kvm_memory_slot *memslot;
1938 memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1939 mark_page_dirty_in_slot(memslot, gfn);
1941 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
1943 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
1947 old = val = vcpu->halt_poll_ns;
1949 if (val == 0 && halt_poll_ns_grow)
1952 val *= halt_poll_ns_grow;
1954 if (val > halt_poll_ns)
1957 vcpu->halt_poll_ns = val;
1958 trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
1961 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
1965 old = val = vcpu->halt_poll_ns;
1966 if (halt_poll_ns_shrink == 0)
1969 val /= halt_poll_ns_shrink;
1971 vcpu->halt_poll_ns = val;
1972 trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
1975 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
1977 if (kvm_arch_vcpu_runnable(vcpu)) {
1978 kvm_make_request(KVM_REQ_UNHALT, vcpu);
1981 if (kvm_cpu_has_pending_timer(vcpu))
1983 if (signal_pending(current))
1990 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1992 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1995 DECLARE_SWAITQUEUE(wait);
1996 bool waited = false;
1999 start = cur = ktime_get();
2000 if (vcpu->halt_poll_ns) {
2001 ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns);
2003 ++vcpu->stat.halt_attempted_poll;
2006 * This sets KVM_REQ_UNHALT if an interrupt
2009 if (kvm_vcpu_check_block(vcpu) < 0) {
2010 ++vcpu->stat.halt_successful_poll;
2014 } while (single_task_running() && ktime_before(cur, stop));
2017 kvm_arch_vcpu_blocking(vcpu);
2020 prepare_to_swait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
2022 if (kvm_vcpu_check_block(vcpu) < 0)
2029 finish_swait(&vcpu->wq, &wait);
2032 kvm_arch_vcpu_unblocking(vcpu);
2034 block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
2037 if (block_ns <= vcpu->halt_poll_ns)
2039 /* we had a long block, shrink polling */
2040 else if (vcpu->halt_poll_ns && block_ns > halt_poll_ns)
2041 shrink_halt_poll_ns(vcpu);
2042 /* we had a short halt and our poll time is too small */
2043 else if (vcpu->halt_poll_ns < halt_poll_ns &&
2044 block_ns < halt_poll_ns)
2045 grow_halt_poll_ns(vcpu);
2047 vcpu->halt_poll_ns = 0;
2049 trace_kvm_vcpu_wakeup(block_ns, waited);
2051 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
2055 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2057 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
2060 int cpu = vcpu->cpu;
2061 struct swait_queue_head *wqp;
2063 wqp = kvm_arch_vcpu_wq(vcpu);
2064 if (swait_active(wqp)) {
2066 ++vcpu->stat.halt_wakeup;
2070 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
2071 if (kvm_arch_vcpu_should_kick(vcpu))
2072 smp_send_reschedule(cpu);
2075 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
2076 #endif /* !CONFIG_S390 */
2078 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
2081 struct task_struct *task = NULL;
2085 pid = rcu_dereference(target->pid);
2087 task = get_pid_task(pid, PIDTYPE_PID);
2091 ret = yield_to(task, 1);
2092 put_task_struct(task);
2096 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
2099 * Helper that checks whether a VCPU is eligible for directed yield.
2100 * Most eligible candidate to yield is decided by following heuristics:
2102 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2103 * (preempted lock holder), indicated by @in_spin_loop.
2104 * Set at the beiginning and cleared at the end of interception/PLE handler.
2106 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2107 * chance last time (mostly it has become eligible now since we have probably
2108 * yielded to lockholder in last iteration. This is done by toggling
2109 * @dy_eligible each time a VCPU checked for eligibility.)
2111 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2112 * to preempted lock-holder could result in wrong VCPU selection and CPU
2113 * burning. Giving priority for a potential lock-holder increases lock
2116 * Since algorithm is based on heuristics, accessing another VCPU data without
2117 * locking does not harm. It may result in trying to yield to same VCPU, fail
2118 * and continue with next VCPU and so on.
2120 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
2122 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2125 eligible = !vcpu->spin_loop.in_spin_loop ||
2126 vcpu->spin_loop.dy_eligible;
2128 if (vcpu->spin_loop.in_spin_loop)
2129 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
2137 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
2139 struct kvm *kvm = me->kvm;
2140 struct kvm_vcpu *vcpu;
2141 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
2147 kvm_vcpu_set_in_spin_loop(me, true);
2149 * We boost the priority of a VCPU that is runnable but not
2150 * currently running, because it got preempted by something
2151 * else and called schedule in __vcpu_run. Hopefully that
2152 * VCPU is holding the lock that we need and will release it.
2153 * We approximate round-robin by starting at the last boosted VCPU.
2155 for (pass = 0; pass < 2 && !yielded && try; pass++) {
2156 kvm_for_each_vcpu(i, vcpu, kvm) {
2157 if (!pass && i <= last_boosted_vcpu) {
2158 i = last_boosted_vcpu;
2160 } else if (pass && i > last_boosted_vcpu)
2162 if (!ACCESS_ONCE(vcpu->preempted))
2166 if (swait_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
2168 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
2171 yielded = kvm_vcpu_yield_to(vcpu);
2173 kvm->last_boosted_vcpu = i;
2175 } else if (yielded < 0) {
2182 kvm_vcpu_set_in_spin_loop(me, false);
2184 /* Ensure vcpu is not eligible during next spinloop */
2185 kvm_vcpu_set_dy_eligible(me, false);
2187 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
2189 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2191 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
2194 if (vmf->pgoff == 0)
2195 page = virt_to_page(vcpu->run);
2197 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
2198 page = virt_to_page(vcpu->arch.pio_data);
2200 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2201 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
2202 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
2205 return kvm_arch_vcpu_fault(vcpu, vmf);
2211 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
2212 .fault = kvm_vcpu_fault,
2215 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2217 vma->vm_ops = &kvm_vcpu_vm_ops;
2221 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2223 struct kvm_vcpu *vcpu = filp->private_data;
2225 kvm_put_kvm(vcpu->kvm);
2229 static struct file_operations kvm_vcpu_fops = {
2230 .release = kvm_vcpu_release,
2231 .unlocked_ioctl = kvm_vcpu_ioctl,
2232 #ifdef CONFIG_KVM_COMPAT
2233 .compat_ioctl = kvm_vcpu_compat_ioctl,
2235 .mmap = kvm_vcpu_mmap,
2236 .llseek = noop_llseek,
2240 * Allocates an inode for the vcpu.
2242 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2244 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
2248 * Creates some virtual cpus. Good luck creating more than one.
2250 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
2253 struct kvm_vcpu *vcpu;
2255 if (id >= KVM_MAX_VCPUS)
2258 vcpu = kvm_arch_vcpu_create(kvm, id);
2260 return PTR_ERR(vcpu);
2262 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2264 r = kvm_arch_vcpu_setup(vcpu);
2268 mutex_lock(&kvm->lock);
2269 if (!kvm_vcpu_compatible(vcpu)) {
2271 goto unlock_vcpu_destroy;
2273 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
2275 goto unlock_vcpu_destroy;
2277 if (kvm_get_vcpu_by_id(kvm, id)) {
2279 goto unlock_vcpu_destroy;
2282 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
2284 /* Now it's all set up, let userspace reach it */
2286 r = create_vcpu_fd(vcpu);
2289 goto unlock_vcpu_destroy;
2292 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
2295 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
2296 * before kvm->online_vcpu's incremented value.
2299 atomic_inc(&kvm->online_vcpus);
2301 mutex_unlock(&kvm->lock);
2302 kvm_arch_vcpu_postcreate(vcpu);
2305 unlock_vcpu_destroy:
2306 mutex_unlock(&kvm->lock);
2308 kvm_arch_vcpu_destroy(vcpu);
2312 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2315 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2316 vcpu->sigset_active = 1;
2317 vcpu->sigset = *sigset;
2319 vcpu->sigset_active = 0;
2323 static long kvm_vcpu_ioctl(struct file *filp,
2324 unsigned int ioctl, unsigned long arg)
2326 struct kvm_vcpu *vcpu = filp->private_data;
2327 void __user *argp = (void __user *)arg;
2329 struct kvm_fpu *fpu = NULL;
2330 struct kvm_sregs *kvm_sregs = NULL;
2332 if (vcpu->kvm->mm != current->mm)
2335 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2338 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
2340 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
2341 * so vcpu_load() would break it.
2343 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_S390_IRQ || ioctl == KVM_INTERRUPT)
2344 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2348 r = vcpu_load(vcpu);
2356 if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
2357 /* The thread running this VCPU changed. */
2358 struct pid *oldpid = vcpu->pid;
2359 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
2361 rcu_assign_pointer(vcpu->pid, newpid);
2366 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2367 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2369 case KVM_GET_REGS: {
2370 struct kvm_regs *kvm_regs;
2373 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2376 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2380 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2387 case KVM_SET_REGS: {
2388 struct kvm_regs *kvm_regs;
2391 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2392 if (IS_ERR(kvm_regs)) {
2393 r = PTR_ERR(kvm_regs);
2396 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2400 case KVM_GET_SREGS: {
2401 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2405 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2409 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2414 case KVM_SET_SREGS: {
2415 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2416 if (IS_ERR(kvm_sregs)) {
2417 r = PTR_ERR(kvm_sregs);
2421 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2424 case KVM_GET_MP_STATE: {
2425 struct kvm_mp_state mp_state;
2427 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2431 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
2436 case KVM_SET_MP_STATE: {
2437 struct kvm_mp_state mp_state;
2440 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
2442 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2445 case KVM_TRANSLATE: {
2446 struct kvm_translation tr;
2449 if (copy_from_user(&tr, argp, sizeof(tr)))
2451 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2455 if (copy_to_user(argp, &tr, sizeof(tr)))
2460 case KVM_SET_GUEST_DEBUG: {
2461 struct kvm_guest_debug dbg;
2464 if (copy_from_user(&dbg, argp, sizeof(dbg)))
2466 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2469 case KVM_SET_SIGNAL_MASK: {
2470 struct kvm_signal_mask __user *sigmask_arg = argp;
2471 struct kvm_signal_mask kvm_sigmask;
2472 sigset_t sigset, *p;
2477 if (copy_from_user(&kvm_sigmask, argp,
2478 sizeof(kvm_sigmask)))
2481 if (kvm_sigmask.len != sizeof(sigset))
2484 if (copy_from_user(&sigset, sigmask_arg->sigset,
2489 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2493 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2497 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2501 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2507 fpu = memdup_user(argp, sizeof(*fpu));
2513 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2517 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2526 #ifdef CONFIG_KVM_COMPAT
2527 static long kvm_vcpu_compat_ioctl(struct file *filp,
2528 unsigned int ioctl, unsigned long arg)
2530 struct kvm_vcpu *vcpu = filp->private_data;
2531 void __user *argp = compat_ptr(arg);
2534 if (vcpu->kvm->mm != current->mm)
2538 case KVM_SET_SIGNAL_MASK: {
2539 struct kvm_signal_mask __user *sigmask_arg = argp;
2540 struct kvm_signal_mask kvm_sigmask;
2541 compat_sigset_t csigset;
2546 if (copy_from_user(&kvm_sigmask, argp,
2547 sizeof(kvm_sigmask)))
2550 if (kvm_sigmask.len != sizeof(csigset))
2553 if (copy_from_user(&csigset, sigmask_arg->sigset,
2556 sigset_from_compat(&sigset, &csigset);
2557 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2559 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2563 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2571 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2572 int (*accessor)(struct kvm_device *dev,
2573 struct kvm_device_attr *attr),
2576 struct kvm_device_attr attr;
2581 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2584 return accessor(dev, &attr);
2587 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2590 struct kvm_device *dev = filp->private_data;
2593 case KVM_SET_DEVICE_ATTR:
2594 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2595 case KVM_GET_DEVICE_ATTR:
2596 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2597 case KVM_HAS_DEVICE_ATTR:
2598 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2600 if (dev->ops->ioctl)
2601 return dev->ops->ioctl(dev, ioctl, arg);
2607 static int kvm_device_release(struct inode *inode, struct file *filp)
2609 struct kvm_device *dev = filp->private_data;
2610 struct kvm *kvm = dev->kvm;
2616 static const struct file_operations kvm_device_fops = {
2617 .unlocked_ioctl = kvm_device_ioctl,
2618 #ifdef CONFIG_KVM_COMPAT
2619 .compat_ioctl = kvm_device_ioctl,
2621 .release = kvm_device_release,
2624 struct kvm_device *kvm_device_from_filp(struct file *filp)
2626 if (filp->f_op != &kvm_device_fops)
2629 return filp->private_data;
2632 static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
2633 #ifdef CONFIG_KVM_MPIC
2634 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
2635 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
2638 #ifdef CONFIG_KVM_XICS
2639 [KVM_DEV_TYPE_XICS] = &kvm_xics_ops,
2643 int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
2645 if (type >= ARRAY_SIZE(kvm_device_ops_table))
2648 if (kvm_device_ops_table[type] != NULL)
2651 kvm_device_ops_table[type] = ops;
2655 void kvm_unregister_device_ops(u32 type)
2657 if (kvm_device_ops_table[type] != NULL)
2658 kvm_device_ops_table[type] = NULL;
2661 static int kvm_ioctl_create_device(struct kvm *kvm,
2662 struct kvm_create_device *cd)
2664 struct kvm_device_ops *ops = NULL;
2665 struct kvm_device *dev;
2666 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2669 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
2672 ops = kvm_device_ops_table[cd->type];
2679 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2686 ret = ops->create(dev, cd->type);
2692 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2698 list_add(&dev->vm_node, &kvm->devices);
2704 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2707 case KVM_CAP_USER_MEMORY:
2708 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2709 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2710 case KVM_CAP_INTERNAL_ERROR_DATA:
2711 #ifdef CONFIG_HAVE_KVM_MSI
2712 case KVM_CAP_SIGNAL_MSI:
2714 #ifdef CONFIG_HAVE_KVM_IRQFD
2716 case KVM_CAP_IRQFD_RESAMPLE:
2718 case KVM_CAP_IOEVENTFD_ANY_LENGTH:
2719 case KVM_CAP_CHECK_EXTENSION_VM:
2721 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2722 case KVM_CAP_IRQ_ROUTING:
2723 return KVM_MAX_IRQ_ROUTES;
2725 #if KVM_ADDRESS_SPACE_NUM > 1
2726 case KVM_CAP_MULTI_ADDRESS_SPACE:
2727 return KVM_ADDRESS_SPACE_NUM;
2732 return kvm_vm_ioctl_check_extension(kvm, arg);
2735 static long kvm_vm_ioctl(struct file *filp,
2736 unsigned int ioctl, unsigned long arg)
2738 struct kvm *kvm = filp->private_data;
2739 void __user *argp = (void __user *)arg;
2742 if (kvm->mm != current->mm)
2745 case KVM_CREATE_VCPU:
2746 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2748 case KVM_SET_USER_MEMORY_REGION: {
2749 struct kvm_userspace_memory_region kvm_userspace_mem;
2752 if (copy_from_user(&kvm_userspace_mem, argp,
2753 sizeof(kvm_userspace_mem)))
2756 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2759 case KVM_GET_DIRTY_LOG: {
2760 struct kvm_dirty_log log;
2763 if (copy_from_user(&log, argp, sizeof(log)))
2765 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2768 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2769 case KVM_REGISTER_COALESCED_MMIO: {
2770 struct kvm_coalesced_mmio_zone zone;
2773 if (copy_from_user(&zone, argp, sizeof(zone)))
2775 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2778 case KVM_UNREGISTER_COALESCED_MMIO: {
2779 struct kvm_coalesced_mmio_zone zone;
2782 if (copy_from_user(&zone, argp, sizeof(zone)))
2784 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2789 struct kvm_irqfd data;
2792 if (copy_from_user(&data, argp, sizeof(data)))
2794 r = kvm_irqfd(kvm, &data);
2797 case KVM_IOEVENTFD: {
2798 struct kvm_ioeventfd data;
2801 if (copy_from_user(&data, argp, sizeof(data)))
2803 r = kvm_ioeventfd(kvm, &data);
2806 #ifdef CONFIG_HAVE_KVM_MSI
2807 case KVM_SIGNAL_MSI: {
2811 if (copy_from_user(&msi, argp, sizeof(msi)))
2813 r = kvm_send_userspace_msi(kvm, &msi);
2817 #ifdef __KVM_HAVE_IRQ_LINE
2818 case KVM_IRQ_LINE_STATUS:
2819 case KVM_IRQ_LINE: {
2820 struct kvm_irq_level irq_event;
2823 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
2826 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
2827 ioctl == KVM_IRQ_LINE_STATUS);
2832 if (ioctl == KVM_IRQ_LINE_STATUS) {
2833 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
2841 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2842 case KVM_SET_GSI_ROUTING: {
2843 struct kvm_irq_routing routing;
2844 struct kvm_irq_routing __user *urouting;
2845 struct kvm_irq_routing_entry *entries;
2848 if (copy_from_user(&routing, argp, sizeof(routing)))
2851 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2856 entries = vmalloc(routing.nr * sizeof(*entries));
2861 if (copy_from_user(entries, urouting->entries,
2862 routing.nr * sizeof(*entries)))
2863 goto out_free_irq_routing;
2864 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2866 out_free_irq_routing:
2870 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
2871 case KVM_CREATE_DEVICE: {
2872 struct kvm_create_device cd;
2875 if (copy_from_user(&cd, argp, sizeof(cd)))
2878 r = kvm_ioctl_create_device(kvm, &cd);
2883 if (copy_to_user(argp, &cd, sizeof(cd)))
2889 case KVM_CHECK_EXTENSION:
2890 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
2893 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2899 #ifdef CONFIG_KVM_COMPAT
2900 struct compat_kvm_dirty_log {
2904 compat_uptr_t dirty_bitmap; /* one bit per page */
2909 static long kvm_vm_compat_ioctl(struct file *filp,
2910 unsigned int ioctl, unsigned long arg)
2912 struct kvm *kvm = filp->private_data;
2915 if (kvm->mm != current->mm)
2918 case KVM_GET_DIRTY_LOG: {
2919 struct compat_kvm_dirty_log compat_log;
2920 struct kvm_dirty_log log;
2923 if (copy_from_user(&compat_log, (void __user *)arg,
2924 sizeof(compat_log)))
2926 log.slot = compat_log.slot;
2927 log.padding1 = compat_log.padding1;
2928 log.padding2 = compat_log.padding2;
2929 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2931 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2935 r = kvm_vm_ioctl(filp, ioctl, arg);
2943 static struct file_operations kvm_vm_fops = {
2944 .release = kvm_vm_release,
2945 .unlocked_ioctl = kvm_vm_ioctl,
2946 #ifdef CONFIG_KVM_COMPAT
2947 .compat_ioctl = kvm_vm_compat_ioctl,
2949 .llseek = noop_llseek,
2952 static int kvm_dev_ioctl_create_vm(unsigned long type)
2957 kvm = kvm_create_vm(type);
2959 return PTR_ERR(kvm);
2960 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2961 r = kvm_coalesced_mmio_init(kvm);
2967 r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC);
2974 static long kvm_dev_ioctl(struct file *filp,
2975 unsigned int ioctl, unsigned long arg)
2980 case KVM_GET_API_VERSION:
2983 r = KVM_API_VERSION;
2986 r = kvm_dev_ioctl_create_vm(arg);
2988 case KVM_CHECK_EXTENSION:
2989 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
2991 case KVM_GET_VCPU_MMAP_SIZE:
2994 r = PAGE_SIZE; /* struct kvm_run */
2996 r += PAGE_SIZE; /* pio data page */
2998 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2999 r += PAGE_SIZE; /* coalesced mmio ring page */
3002 case KVM_TRACE_ENABLE:
3003 case KVM_TRACE_PAUSE:
3004 case KVM_TRACE_DISABLE:
3008 return kvm_arch_dev_ioctl(filp, ioctl, arg);
3014 static struct file_operations kvm_chardev_ops = {
3015 .unlocked_ioctl = kvm_dev_ioctl,
3016 .compat_ioctl = kvm_dev_ioctl,
3017 .llseek = noop_llseek,
3020 static struct miscdevice kvm_dev = {
3026 static void hardware_enable_nolock(void *junk)
3028 int cpu = raw_smp_processor_id();
3031 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
3034 cpumask_set_cpu(cpu, cpus_hardware_enabled);
3036 r = kvm_arch_hardware_enable();
3039 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3040 atomic_inc(&hardware_enable_failed);
3041 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
3045 static void hardware_enable(void)
3047 raw_spin_lock(&kvm_count_lock);
3048 if (kvm_usage_count)
3049 hardware_enable_nolock(NULL);
3050 raw_spin_unlock(&kvm_count_lock);
3053 static void hardware_disable_nolock(void *junk)
3055 int cpu = raw_smp_processor_id();
3057 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
3059 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3060 kvm_arch_hardware_disable();
3063 static void hardware_disable(void)
3065 raw_spin_lock(&kvm_count_lock);
3066 if (kvm_usage_count)
3067 hardware_disable_nolock(NULL);
3068 raw_spin_unlock(&kvm_count_lock);
3071 static void hardware_disable_all_nolock(void)
3073 BUG_ON(!kvm_usage_count);
3076 if (!kvm_usage_count)
3077 on_each_cpu(hardware_disable_nolock, NULL, 1);
3080 static void hardware_disable_all(void)
3082 raw_spin_lock(&kvm_count_lock);
3083 hardware_disable_all_nolock();
3084 raw_spin_unlock(&kvm_count_lock);
3087 static int hardware_enable_all(void)
3091 raw_spin_lock(&kvm_count_lock);
3094 if (kvm_usage_count == 1) {
3095 atomic_set(&hardware_enable_failed, 0);
3096 on_each_cpu(hardware_enable_nolock, NULL, 1);
3098 if (atomic_read(&hardware_enable_failed)) {
3099 hardware_disable_all_nolock();
3104 raw_spin_unlock(&kvm_count_lock);
3109 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
3112 val &= ~CPU_TASKS_FROZEN;
3124 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3128 * Some (well, at least mine) BIOSes hang on reboot if
3131 * And Intel TXT required VMX off for all cpu when system shutdown.
3133 pr_info("kvm: exiting hardware virtualization\n");
3134 kvm_rebooting = true;
3135 on_each_cpu(hardware_disable_nolock, NULL, 1);
3139 static struct notifier_block kvm_reboot_notifier = {
3140 .notifier_call = kvm_reboot,
3144 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3148 for (i = 0; i < bus->dev_count; i++) {
3149 struct kvm_io_device *pos = bus->range[i].dev;
3151 kvm_iodevice_destructor(pos);
3156 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
3157 const struct kvm_io_range *r2)
3159 gpa_t addr1 = r1->addr;
3160 gpa_t addr2 = r2->addr;
3165 /* If r2->len == 0, match the exact address. If r2->len != 0,
3166 * accept any overlapping write. Any order is acceptable for
3167 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
3168 * we process all of them.
3181 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
3183 return kvm_io_bus_cmp(p1, p2);
3186 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
3187 gpa_t addr, int len)
3189 bus->range[bus->dev_count++] = (struct kvm_io_range) {
3195 sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
3196 kvm_io_bus_sort_cmp, NULL);
3201 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
3202 gpa_t addr, int len)
3204 struct kvm_io_range *range, key;
3207 key = (struct kvm_io_range) {
3212 range = bsearch(&key, bus->range, bus->dev_count,
3213 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
3217 off = range - bus->range;
3219 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
3225 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3226 struct kvm_io_range *range, const void *val)
3230 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3234 while (idx < bus->dev_count &&
3235 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3236 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
3245 /* kvm_io_bus_write - called under kvm->slots_lock */
3246 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3247 int len, const void *val)
3249 struct kvm_io_bus *bus;
3250 struct kvm_io_range range;
3253 range = (struct kvm_io_range) {
3258 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3259 r = __kvm_io_bus_write(vcpu, bus, &range, val);
3260 return r < 0 ? r : 0;
3263 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3264 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
3265 gpa_t addr, int len, const void *val, long cookie)
3267 struct kvm_io_bus *bus;
3268 struct kvm_io_range range;
3270 range = (struct kvm_io_range) {
3275 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3277 /* First try the device referenced by cookie. */
3278 if ((cookie >= 0) && (cookie < bus->dev_count) &&
3279 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
3280 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
3285 * cookie contained garbage; fall back to search and return the
3286 * correct cookie value.
3288 return __kvm_io_bus_write(vcpu, bus, &range, val);
3291 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3292 struct kvm_io_range *range, void *val)
3296 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3300 while (idx < bus->dev_count &&
3301 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3302 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
3310 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
3312 /* kvm_io_bus_read - called under kvm->slots_lock */
3313 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3316 struct kvm_io_bus *bus;
3317 struct kvm_io_range range;
3320 range = (struct kvm_io_range) {
3325 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3326 r = __kvm_io_bus_read(vcpu, bus, &range, val);
3327 return r < 0 ? r : 0;
3331 /* Caller must hold slots_lock. */
3332 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3333 int len, struct kvm_io_device *dev)
3335 struct kvm_io_bus *new_bus, *bus;
3337 bus = kvm->buses[bus_idx];
3338 /* exclude ioeventfd which is limited by maximum fd */
3339 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3342 new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count + 1) *
3343 sizeof(struct kvm_io_range)), GFP_KERNEL);
3346 memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
3347 sizeof(struct kvm_io_range)));
3348 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
3349 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3350 synchronize_srcu_expedited(&kvm->srcu);
3356 /* Caller must hold slots_lock. */
3357 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3358 struct kvm_io_device *dev)
3361 struct kvm_io_bus *new_bus, *bus;
3363 bus = kvm->buses[bus_idx];
3365 for (i = 0; i < bus->dev_count; i++)
3366 if (bus->range[i].dev == dev) {
3374 new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3375 sizeof(struct kvm_io_range)), GFP_KERNEL);
3379 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3380 new_bus->dev_count--;
3381 memcpy(new_bus->range + i, bus->range + i + 1,
3382 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3384 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3385 synchronize_srcu_expedited(&kvm->srcu);
3390 static struct notifier_block kvm_cpu_notifier = {
3391 .notifier_call = kvm_cpu_hotplug,
3394 static int vm_stat_get(void *_offset, u64 *val)
3396 unsigned offset = (long)_offset;
3400 spin_lock(&kvm_lock);
3401 list_for_each_entry(kvm, &vm_list, vm_list)
3402 *val += *(u32 *)((void *)kvm + offset);
3403 spin_unlock(&kvm_lock);
3407 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
3409 static int vcpu_stat_get(void *_offset, u64 *val)
3411 unsigned offset = (long)_offset;
3413 struct kvm_vcpu *vcpu;
3417 spin_lock(&kvm_lock);
3418 list_for_each_entry(kvm, &vm_list, vm_list)
3419 kvm_for_each_vcpu(i, vcpu, kvm)
3420 *val += *(u32 *)((void *)vcpu + offset);
3422 spin_unlock(&kvm_lock);
3426 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
3428 static const struct file_operations *stat_fops[] = {
3429 [KVM_STAT_VCPU] = &vcpu_stat_fops,
3430 [KVM_STAT_VM] = &vm_stat_fops,
3433 static int kvm_init_debug(void)
3436 struct kvm_stats_debugfs_item *p;
3438 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3439 if (kvm_debugfs_dir == NULL)
3442 for (p = debugfs_entries; p->name; ++p) {
3443 if (!debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
3444 (void *)(long)p->offset,
3445 stat_fops[p->kind]))
3452 debugfs_remove_recursive(kvm_debugfs_dir);
3457 static int kvm_suspend(void)
3459 if (kvm_usage_count)
3460 hardware_disable_nolock(NULL);
3464 static void kvm_resume(void)
3466 if (kvm_usage_count) {
3467 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3468 hardware_enable_nolock(NULL);
3472 static struct syscore_ops kvm_syscore_ops = {
3473 .suspend = kvm_suspend,
3474 .resume = kvm_resume,
3478 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3480 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3483 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3485 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3487 if (vcpu->preempted)
3488 vcpu->preempted = false;
3490 kvm_arch_sched_in(vcpu, cpu);
3492 kvm_arch_vcpu_load(vcpu, cpu);
3495 static void kvm_sched_out(struct preempt_notifier *pn,
3496 struct task_struct *next)
3498 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3500 if (current->state == TASK_RUNNING)
3501 vcpu->preempted = true;
3502 kvm_arch_vcpu_put(vcpu);
3505 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3506 struct module *module)
3511 r = kvm_arch_init(opaque);
3516 * kvm_arch_init makes sure there's at most one caller
3517 * for architectures that support multiple implementations,
3518 * like intel and amd on x86.
3519 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3520 * conflicts in case kvm is already setup for another implementation.
3522 r = kvm_irqfd_init();
3526 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
3531 r = kvm_arch_hardware_setup();
3535 for_each_online_cpu(cpu) {
3536 smp_call_function_single(cpu,
3537 kvm_arch_check_processor_compat,
3543 r = register_cpu_notifier(&kvm_cpu_notifier);
3546 register_reboot_notifier(&kvm_reboot_notifier);
3548 /* A kmem cache lets us meet the alignment requirements of fx_save. */
3550 vcpu_align = __alignof__(struct kvm_vcpu);
3551 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
3553 if (!kvm_vcpu_cache) {
3558 r = kvm_async_pf_init();
3562 kvm_chardev_ops.owner = module;
3563 kvm_vm_fops.owner = module;
3564 kvm_vcpu_fops.owner = module;
3566 r = misc_register(&kvm_dev);
3568 pr_err("kvm: misc device register failed\n");
3572 register_syscore_ops(&kvm_syscore_ops);
3574 kvm_preempt_ops.sched_in = kvm_sched_in;
3575 kvm_preempt_ops.sched_out = kvm_sched_out;
3577 r = kvm_init_debug();
3579 pr_err("kvm: create debugfs files failed\n");
3583 r = kvm_vfio_ops_init();
3589 unregister_syscore_ops(&kvm_syscore_ops);
3590 misc_deregister(&kvm_dev);
3592 kvm_async_pf_deinit();
3594 kmem_cache_destroy(kvm_vcpu_cache);
3596 unregister_reboot_notifier(&kvm_reboot_notifier);
3597 unregister_cpu_notifier(&kvm_cpu_notifier);
3600 kvm_arch_hardware_unsetup();
3602 free_cpumask_var(cpus_hardware_enabled);
3610 EXPORT_SYMBOL_GPL(kvm_init);
3614 debugfs_remove_recursive(kvm_debugfs_dir);
3615 misc_deregister(&kvm_dev);
3616 kmem_cache_destroy(kvm_vcpu_cache);
3617 kvm_async_pf_deinit();
3618 unregister_syscore_ops(&kvm_syscore_ops);
3619 unregister_reboot_notifier(&kvm_reboot_notifier);
3620 unregister_cpu_notifier(&kvm_cpu_notifier);
3621 on_each_cpu(hardware_disable_nolock, NULL, 1);
3622 kvm_arch_hardware_unsetup();
3625 free_cpumask_var(cpus_hardware_enabled);
3626 kvm_vfio_ops_exit();
3628 EXPORT_SYMBOL_GPL(kvm_exit);