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(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(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 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
211 kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
214 void kvm_make_scan_ioapic_request(struct kvm *kvm)
216 kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
219 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
224 mutex_init(&vcpu->mutex);
229 vcpu->halt_poll_ns = 0;
230 init_waitqueue_head(&vcpu->wq);
231 kvm_async_pf_vcpu_init(vcpu);
234 INIT_LIST_HEAD(&vcpu->blocked_vcpu_list);
236 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
241 vcpu->run = page_address(page);
243 kvm_vcpu_set_in_spin_loop(vcpu, false);
244 kvm_vcpu_set_dy_eligible(vcpu, false);
245 vcpu->preempted = false;
247 r = kvm_arch_vcpu_init(vcpu);
253 free_page((unsigned long)vcpu->run);
257 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
259 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
262 kvm_arch_vcpu_uninit(vcpu);
263 free_page((unsigned long)vcpu->run);
265 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
267 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
268 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
270 return container_of(mn, struct kvm, mmu_notifier);
273 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
274 struct mm_struct *mm,
275 unsigned long address)
277 struct kvm *kvm = mmu_notifier_to_kvm(mn);
278 int need_tlb_flush, idx;
281 * When ->invalidate_page runs, the linux pte has been zapped
282 * already but the page is still allocated until
283 * ->invalidate_page returns. So if we increase the sequence
284 * here the kvm page fault will notice if the spte can't be
285 * established because the page is going to be freed. If
286 * instead the kvm page fault establishes the spte before
287 * ->invalidate_page runs, kvm_unmap_hva will release it
290 * The sequence increase only need to be seen at spin_unlock
291 * time, and not at spin_lock time.
293 * Increasing the sequence after the spin_unlock would be
294 * unsafe because the kvm page fault could then establish the
295 * pte after kvm_unmap_hva returned, without noticing the page
296 * is going to be freed.
298 idx = srcu_read_lock(&kvm->srcu);
299 spin_lock(&kvm->mmu_lock);
301 kvm->mmu_notifier_seq++;
302 need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
303 /* we've to flush the tlb before the pages can be freed */
305 kvm_flush_remote_tlbs(kvm);
307 spin_unlock(&kvm->mmu_lock);
309 kvm_arch_mmu_notifier_invalidate_page(kvm, address);
311 srcu_read_unlock(&kvm->srcu, idx);
314 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
315 struct mm_struct *mm,
316 unsigned long address,
319 struct kvm *kvm = mmu_notifier_to_kvm(mn);
322 idx = srcu_read_lock(&kvm->srcu);
323 spin_lock(&kvm->mmu_lock);
324 kvm->mmu_notifier_seq++;
325 kvm_set_spte_hva(kvm, address, pte);
326 spin_unlock(&kvm->mmu_lock);
327 srcu_read_unlock(&kvm->srcu, idx);
330 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
331 struct mm_struct *mm,
335 struct kvm *kvm = mmu_notifier_to_kvm(mn);
336 int need_tlb_flush = 0, idx;
338 idx = srcu_read_lock(&kvm->srcu);
339 spin_lock(&kvm->mmu_lock);
341 * The count increase must become visible at unlock time as no
342 * spte can be established without taking the mmu_lock and
343 * count is also read inside the mmu_lock critical section.
345 kvm->mmu_notifier_count++;
346 need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
347 need_tlb_flush |= kvm->tlbs_dirty;
348 /* we've to flush the tlb before the pages can be freed */
350 kvm_flush_remote_tlbs(kvm);
352 spin_unlock(&kvm->mmu_lock);
353 srcu_read_unlock(&kvm->srcu, idx);
356 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
357 struct mm_struct *mm,
361 struct kvm *kvm = mmu_notifier_to_kvm(mn);
363 spin_lock(&kvm->mmu_lock);
365 * This sequence increase will notify the kvm page fault that
366 * the page that is going to be mapped in the spte could have
369 kvm->mmu_notifier_seq++;
372 * The above sequence increase must be visible before the
373 * below count decrease, which is ensured by the smp_wmb above
374 * in conjunction with the smp_rmb in mmu_notifier_retry().
376 kvm->mmu_notifier_count--;
377 spin_unlock(&kvm->mmu_lock);
379 BUG_ON(kvm->mmu_notifier_count < 0);
382 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
383 struct mm_struct *mm,
387 struct kvm *kvm = mmu_notifier_to_kvm(mn);
390 idx = srcu_read_lock(&kvm->srcu);
391 spin_lock(&kvm->mmu_lock);
393 young = kvm_age_hva(kvm, start, end);
395 kvm_flush_remote_tlbs(kvm);
397 spin_unlock(&kvm->mmu_lock);
398 srcu_read_unlock(&kvm->srcu, idx);
403 static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
404 struct mm_struct *mm,
408 struct kvm *kvm = mmu_notifier_to_kvm(mn);
411 idx = srcu_read_lock(&kvm->srcu);
412 spin_lock(&kvm->mmu_lock);
414 * Even though we do not flush TLB, this will still adversely
415 * affect performance on pre-Haswell Intel EPT, where there is
416 * no EPT Access Bit to clear so that we have to tear down EPT
417 * tables instead. If we find this unacceptable, we can always
418 * add a parameter to kvm_age_hva so that it effectively doesn't
419 * do anything on clear_young.
421 * Also note that currently we never issue secondary TLB flushes
422 * from clear_young, leaving this job up to the regular system
423 * cadence. If we find this inaccurate, we might come up with a
424 * more sophisticated heuristic later.
426 young = kvm_age_hva(kvm, start, end);
427 spin_unlock(&kvm->mmu_lock);
428 srcu_read_unlock(&kvm->srcu, idx);
433 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
434 struct mm_struct *mm,
435 unsigned long address)
437 struct kvm *kvm = mmu_notifier_to_kvm(mn);
440 idx = srcu_read_lock(&kvm->srcu);
441 spin_lock(&kvm->mmu_lock);
442 young = kvm_test_age_hva(kvm, address);
443 spin_unlock(&kvm->mmu_lock);
444 srcu_read_unlock(&kvm->srcu, idx);
449 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
450 struct mm_struct *mm)
452 struct kvm *kvm = mmu_notifier_to_kvm(mn);
455 idx = srcu_read_lock(&kvm->srcu);
456 kvm_arch_flush_shadow_all(kvm);
457 srcu_read_unlock(&kvm->srcu, idx);
460 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
461 .invalidate_page = kvm_mmu_notifier_invalidate_page,
462 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
463 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
464 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
465 .clear_young = kvm_mmu_notifier_clear_young,
466 .test_young = kvm_mmu_notifier_test_young,
467 .change_pte = kvm_mmu_notifier_change_pte,
468 .release = kvm_mmu_notifier_release,
471 static int kvm_init_mmu_notifier(struct kvm *kvm)
473 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
474 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
477 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
479 static int kvm_init_mmu_notifier(struct kvm *kvm)
484 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
486 static struct kvm_memslots *kvm_alloc_memslots(void)
489 struct kvm_memslots *slots;
491 slots = kvm_kvzalloc(sizeof(struct kvm_memslots));
496 * Init kvm generation close to the maximum to easily test the
497 * code of handling generation number wrap-around.
499 slots->generation = -150;
500 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
501 slots->id_to_index[i] = slots->memslots[i].id = i;
506 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
508 if (!memslot->dirty_bitmap)
511 kvfree(memslot->dirty_bitmap);
512 memslot->dirty_bitmap = NULL;
516 * Free any memory in @free but not in @dont.
518 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
519 struct kvm_memory_slot *dont)
521 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
522 kvm_destroy_dirty_bitmap(free);
524 kvm_arch_free_memslot(kvm, free, dont);
529 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
531 struct kvm_memory_slot *memslot;
536 kvm_for_each_memslot(memslot, slots)
537 kvm_free_memslot(kvm, memslot, NULL);
542 static struct kvm *kvm_create_vm(unsigned long type)
545 struct kvm *kvm = kvm_arch_alloc_vm();
548 return ERR_PTR(-ENOMEM);
550 spin_lock_init(&kvm->mmu_lock);
551 atomic_inc(¤t->mm->mm_count);
552 kvm->mm = current->mm;
553 kvm_eventfd_init(kvm);
554 mutex_init(&kvm->lock);
555 mutex_init(&kvm->irq_lock);
556 mutex_init(&kvm->slots_lock);
557 atomic_set(&kvm->users_count, 1);
558 INIT_LIST_HEAD(&kvm->devices);
560 r = kvm_arch_init_vm(kvm, type);
562 goto out_err_no_disable;
564 r = hardware_enable_all();
566 goto out_err_no_disable;
568 #ifdef CONFIG_HAVE_KVM_IRQFD
569 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
572 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
575 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
576 kvm->memslots[i] = kvm_alloc_memslots();
577 if (!kvm->memslots[i])
578 goto out_err_no_srcu;
581 if (init_srcu_struct(&kvm->srcu))
582 goto out_err_no_srcu;
583 if (init_srcu_struct(&kvm->irq_srcu))
584 goto out_err_no_irq_srcu;
585 for (i = 0; i < KVM_NR_BUSES; i++) {
586 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
592 r = kvm_init_mmu_notifier(kvm);
596 spin_lock(&kvm_lock);
597 list_add(&kvm->vm_list, &vm_list);
598 spin_unlock(&kvm_lock);
600 preempt_notifier_inc();
605 cleanup_srcu_struct(&kvm->irq_srcu);
607 cleanup_srcu_struct(&kvm->srcu);
609 hardware_disable_all();
611 for (i = 0; i < KVM_NR_BUSES; i++)
612 kfree(kvm->buses[i]);
613 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
614 kvm_free_memslots(kvm, kvm->memslots[i]);
615 kvm_arch_free_vm(kvm);
621 * Avoid using vmalloc for a small buffer.
622 * Should not be used when the size is statically known.
624 void *kvm_kvzalloc(unsigned long size)
626 if (size > PAGE_SIZE)
627 return vzalloc(size);
629 return kzalloc(size, GFP_KERNEL);
632 static void kvm_destroy_devices(struct kvm *kvm)
634 struct list_head *node, *tmp;
636 list_for_each_safe(node, tmp, &kvm->devices) {
637 struct kvm_device *dev =
638 list_entry(node, struct kvm_device, vm_node);
641 dev->ops->destroy(dev);
645 static void kvm_destroy_vm(struct kvm *kvm)
648 struct mm_struct *mm = kvm->mm;
650 kvm_arch_sync_events(kvm);
651 spin_lock(&kvm_lock);
652 list_del(&kvm->vm_list);
653 spin_unlock(&kvm_lock);
654 kvm_free_irq_routing(kvm);
655 for (i = 0; i < KVM_NR_BUSES; i++)
656 kvm_io_bus_destroy(kvm->buses[i]);
657 kvm_coalesced_mmio_free(kvm);
658 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
659 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
661 kvm_arch_flush_shadow_all(kvm);
663 kvm_arch_destroy_vm(kvm);
664 kvm_destroy_devices(kvm);
665 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
666 kvm_free_memslots(kvm, kvm->memslots[i]);
667 cleanup_srcu_struct(&kvm->irq_srcu);
668 cleanup_srcu_struct(&kvm->srcu);
669 kvm_arch_free_vm(kvm);
670 preempt_notifier_dec();
671 hardware_disable_all();
675 void kvm_get_kvm(struct kvm *kvm)
677 atomic_inc(&kvm->users_count);
679 EXPORT_SYMBOL_GPL(kvm_get_kvm);
681 void kvm_put_kvm(struct kvm *kvm)
683 if (atomic_dec_and_test(&kvm->users_count))
686 EXPORT_SYMBOL_GPL(kvm_put_kvm);
689 static int kvm_vm_release(struct inode *inode, struct file *filp)
691 struct kvm *kvm = filp->private_data;
693 kvm_irqfd_release(kvm);
700 * Allocation size is twice as large as the actual dirty bitmap size.
701 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
703 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
705 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
707 memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
708 if (!memslot->dirty_bitmap)
715 * Insert memslot and re-sort memslots based on their GFN,
716 * so binary search could be used to lookup GFN.
717 * Sorting algorithm takes advantage of having initially
718 * sorted array and known changed memslot position.
720 static void update_memslots(struct kvm_memslots *slots,
721 struct kvm_memory_slot *new)
724 int i = slots->id_to_index[id];
725 struct kvm_memory_slot *mslots = slots->memslots;
727 WARN_ON(mslots[i].id != id);
729 WARN_ON(!mslots[i].npages);
730 if (mslots[i].npages)
733 if (!mslots[i].npages)
737 while (i < KVM_MEM_SLOTS_NUM - 1 &&
738 new->base_gfn <= mslots[i + 1].base_gfn) {
739 if (!mslots[i + 1].npages)
741 mslots[i] = mslots[i + 1];
742 slots->id_to_index[mslots[i].id] = i;
747 * The ">=" is needed when creating a slot with base_gfn == 0,
748 * so that it moves before all those with base_gfn == npages == 0.
750 * On the other hand, if new->npages is zero, the above loop has
751 * already left i pointing to the beginning of the empty part of
752 * mslots, and the ">=" would move the hole backwards in this
753 * case---which is wrong. So skip the loop when deleting a slot.
757 new->base_gfn >= mslots[i - 1].base_gfn) {
758 mslots[i] = mslots[i - 1];
759 slots->id_to_index[mslots[i].id] = i;
763 WARN_ON_ONCE(i != slots->used_slots);
766 slots->id_to_index[mslots[i].id] = i;
769 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
771 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
773 #ifdef __KVM_HAVE_READONLY_MEM
774 valid_flags |= KVM_MEM_READONLY;
777 if (mem->flags & ~valid_flags)
783 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
784 int as_id, struct kvm_memslots *slots)
786 struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id);
789 * Set the low bit in the generation, which disables SPTE caching
790 * until the end of synchronize_srcu_expedited.
792 WARN_ON(old_memslots->generation & 1);
793 slots->generation = old_memslots->generation + 1;
795 rcu_assign_pointer(kvm->memslots[as_id], slots);
796 synchronize_srcu_expedited(&kvm->srcu);
799 * Increment the new memslot generation a second time. This prevents
800 * vm exits that race with memslot updates from caching a memslot
801 * generation that will (potentially) be valid forever.
805 kvm_arch_memslots_updated(kvm, slots);
811 * Allocate some memory and give it an address in the guest physical address
814 * Discontiguous memory is allowed, mostly for framebuffers.
816 * Must be called holding kvm->slots_lock for write.
818 int __kvm_set_memory_region(struct kvm *kvm,
819 const struct kvm_userspace_memory_region *mem)
823 unsigned long npages;
824 struct kvm_memory_slot *slot;
825 struct kvm_memory_slot old, new;
826 struct kvm_memslots *slots = NULL, *old_memslots;
828 enum kvm_mr_change change;
830 r = check_memory_region_flags(mem);
835 as_id = mem->slot >> 16;
838 /* General sanity checks */
839 if (mem->memory_size & (PAGE_SIZE - 1))
841 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
843 /* We can read the guest memory with __xxx_user() later on. */
844 if ((id < KVM_USER_MEM_SLOTS) &&
845 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
846 !access_ok(VERIFY_WRITE,
847 (void __user *)(unsigned long)mem->userspace_addr,
850 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
852 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
855 slot = id_to_memslot(__kvm_memslots(kvm, as_id), id);
856 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
857 npages = mem->memory_size >> PAGE_SHIFT;
859 if (npages > KVM_MEM_MAX_NR_PAGES)
865 new.base_gfn = base_gfn;
867 new.flags = mem->flags;
871 change = KVM_MR_CREATE;
872 else { /* Modify an existing slot. */
873 if ((mem->userspace_addr != old.userspace_addr) ||
874 (npages != old.npages) ||
875 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
878 if (base_gfn != old.base_gfn)
879 change = KVM_MR_MOVE;
880 else if (new.flags != old.flags)
881 change = KVM_MR_FLAGS_ONLY;
882 else { /* Nothing to change. */
891 change = KVM_MR_DELETE;
896 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
897 /* Check for overlaps */
899 kvm_for_each_memslot(slot, __kvm_memslots(kvm, as_id)) {
900 if ((slot->id >= KVM_USER_MEM_SLOTS) ||
903 if (!((base_gfn + npages <= slot->base_gfn) ||
904 (base_gfn >= slot->base_gfn + slot->npages)))
909 /* Free page dirty bitmap if unneeded */
910 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
911 new.dirty_bitmap = NULL;
914 if (change == KVM_MR_CREATE) {
915 new.userspace_addr = mem->userspace_addr;
917 if (kvm_arch_create_memslot(kvm, &new, npages))
921 /* Allocate page dirty bitmap if needed */
922 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
923 if (kvm_create_dirty_bitmap(&new) < 0)
927 slots = kvm_kvzalloc(sizeof(struct kvm_memslots));
930 memcpy(slots, __kvm_memslots(kvm, as_id), sizeof(struct kvm_memslots));
932 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
933 slot = id_to_memslot(slots, id);
934 slot->flags |= KVM_MEMSLOT_INVALID;
936 old_memslots = install_new_memslots(kvm, as_id, slots);
938 /* slot was deleted or moved, clear iommu mapping */
939 kvm_iommu_unmap_pages(kvm, &old);
940 /* From this point no new shadow pages pointing to a deleted,
941 * or moved, memslot will be created.
943 * validation of sp->gfn happens in:
944 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
945 * - kvm_is_visible_gfn (mmu_check_roots)
947 kvm_arch_flush_shadow_memslot(kvm, slot);
950 * We can re-use the old_memslots from above, the only difference
951 * from the currently installed memslots is the invalid flag. This
952 * will get overwritten by update_memslots anyway.
954 slots = old_memslots;
957 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
961 /* actual memory is freed via old in kvm_free_memslot below */
962 if (change == KVM_MR_DELETE) {
963 new.dirty_bitmap = NULL;
964 memset(&new.arch, 0, sizeof(new.arch));
967 update_memslots(slots, &new);
968 old_memslots = install_new_memslots(kvm, as_id, slots);
970 kvm_arch_commit_memory_region(kvm, mem, &old, &new, change);
972 kvm_free_memslot(kvm, &old, &new);
973 kvfree(old_memslots);
976 * IOMMU mapping: New slots need to be mapped. Old slots need to be
977 * un-mapped and re-mapped if their base changes. Since base change
978 * unmapping is handled above with slot deletion, mapping alone is
979 * needed here. Anything else the iommu might care about for existing
980 * slots (size changes, userspace addr changes and read-only flag
981 * changes) is disallowed above, so any other attribute changes getting
982 * here can be skipped.
984 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
985 r = kvm_iommu_map_pages(kvm, &new);
994 kvm_free_memslot(kvm, &new, &old);
998 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1000 int kvm_set_memory_region(struct kvm *kvm,
1001 const struct kvm_userspace_memory_region *mem)
1005 mutex_lock(&kvm->slots_lock);
1006 r = __kvm_set_memory_region(kvm, mem);
1007 mutex_unlock(&kvm->slots_lock);
1010 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1012 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1013 struct kvm_userspace_memory_region *mem)
1015 if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1018 return kvm_set_memory_region(kvm, mem);
1021 int kvm_get_dirty_log(struct kvm *kvm,
1022 struct kvm_dirty_log *log, int *is_dirty)
1024 struct kvm_memslots *slots;
1025 struct kvm_memory_slot *memslot;
1026 int r, i, as_id, id;
1028 unsigned long any = 0;
1031 as_id = log->slot >> 16;
1032 id = (u16)log->slot;
1033 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1036 slots = __kvm_memslots(kvm, as_id);
1037 memslot = id_to_memslot(slots, id);
1039 if (!memslot->dirty_bitmap)
1042 n = kvm_dirty_bitmap_bytes(memslot);
1044 for (i = 0; !any && i < n/sizeof(long); ++i)
1045 any = memslot->dirty_bitmap[i];
1048 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1058 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1060 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1062 * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages
1063 * are dirty write protect them for next write.
1064 * @kvm: pointer to kvm instance
1065 * @log: slot id and address to which we copy the log
1066 * @is_dirty: flag set if any page is dirty
1068 * We need to keep it in mind that VCPU threads can write to the bitmap
1069 * concurrently. So, to avoid losing track of dirty pages we keep the
1072 * 1. Take a snapshot of the bit and clear it if needed.
1073 * 2. Write protect the corresponding page.
1074 * 3. Copy the snapshot to the userspace.
1075 * 4. Upon return caller flushes TLB's if needed.
1077 * Between 2 and 4, the guest may write to the page using the remaining TLB
1078 * entry. This is not a problem because the page is reported dirty using
1079 * the snapshot taken before and step 4 ensures that writes done after
1080 * exiting to userspace will be logged for the next call.
1083 int kvm_get_dirty_log_protect(struct kvm *kvm,
1084 struct kvm_dirty_log *log, bool *is_dirty)
1086 struct kvm_memslots *slots;
1087 struct kvm_memory_slot *memslot;
1088 int r, i, as_id, id;
1090 unsigned long *dirty_bitmap;
1091 unsigned long *dirty_bitmap_buffer;
1094 as_id = log->slot >> 16;
1095 id = (u16)log->slot;
1096 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1099 slots = __kvm_memslots(kvm, as_id);
1100 memslot = id_to_memslot(slots, id);
1102 dirty_bitmap = memslot->dirty_bitmap;
1107 n = kvm_dirty_bitmap_bytes(memslot);
1109 dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long);
1110 memset(dirty_bitmap_buffer, 0, n);
1112 spin_lock(&kvm->mmu_lock);
1114 for (i = 0; i < n / sizeof(long); i++) {
1118 if (!dirty_bitmap[i])
1123 mask = xchg(&dirty_bitmap[i], 0);
1124 dirty_bitmap_buffer[i] = mask;
1127 offset = i * BITS_PER_LONG;
1128 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1133 spin_unlock(&kvm->mmu_lock);
1136 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1143 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect);
1146 bool kvm_largepages_enabled(void)
1148 return largepages_enabled;
1151 void kvm_disable_largepages(void)
1153 largepages_enabled = false;
1155 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1157 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1159 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1161 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1163 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
1165 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn);
1168 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1170 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1172 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1173 memslot->flags & KVM_MEMSLOT_INVALID)
1178 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1180 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1182 struct vm_area_struct *vma;
1183 unsigned long addr, size;
1187 addr = gfn_to_hva(kvm, gfn);
1188 if (kvm_is_error_hva(addr))
1191 down_read(¤t->mm->mmap_sem);
1192 vma = find_vma(current->mm, addr);
1196 size = vma_kernel_pagesize(vma);
1199 up_read(¤t->mm->mmap_sem);
1204 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1206 return slot->flags & KVM_MEM_READONLY;
1209 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1210 gfn_t *nr_pages, bool write)
1212 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1213 return KVM_HVA_ERR_BAD;
1215 if (memslot_is_readonly(slot) && write)
1216 return KVM_HVA_ERR_RO_BAD;
1219 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1221 return __gfn_to_hva_memslot(slot, gfn);
1224 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1227 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1230 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1233 return gfn_to_hva_many(slot, gfn, NULL);
1235 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1237 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1239 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1241 EXPORT_SYMBOL_GPL(gfn_to_hva);
1243 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
1245 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
1247 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
1250 * If writable is set to false, the hva returned by this function is only
1251 * allowed to be read.
1253 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1254 gfn_t gfn, bool *writable)
1256 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1258 if (!kvm_is_error_hva(hva) && writable)
1259 *writable = !memslot_is_readonly(slot);
1264 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1266 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1268 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1271 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
1273 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1275 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1278 static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1279 unsigned long start, int write, struct page **page)
1281 int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1284 flags |= FOLL_WRITE;
1286 return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1289 static inline int check_user_page_hwpoison(unsigned long addr)
1291 int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1293 rc = __get_user_pages(current, current->mm, addr, 1,
1294 flags, NULL, NULL, NULL);
1295 return rc == -EHWPOISON;
1299 * The atomic path to get the writable pfn which will be stored in @pfn,
1300 * true indicates success, otherwise false is returned.
1302 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1303 bool write_fault, bool *writable, pfn_t *pfn)
1305 struct page *page[1];
1308 if (!(async || atomic))
1312 * Fast pin a writable pfn only if it is a write fault request
1313 * or the caller allows to map a writable pfn for a read fault
1316 if (!(write_fault || writable))
1319 npages = __get_user_pages_fast(addr, 1, 1, page);
1321 *pfn = page_to_pfn(page[0]);
1332 * The slow path to get the pfn of the specified host virtual address,
1333 * 1 indicates success, -errno is returned if error is detected.
1335 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1336 bool *writable, pfn_t *pfn)
1338 struct page *page[1];
1344 *writable = write_fault;
1347 down_read(¤t->mm->mmap_sem);
1348 npages = get_user_page_nowait(current, current->mm,
1349 addr, write_fault, page);
1350 up_read(¤t->mm->mmap_sem);
1352 npages = __get_user_pages_unlocked(current, current->mm, addr, 1,
1353 write_fault, 0, page,
1354 FOLL_TOUCH|FOLL_HWPOISON);
1358 /* map read fault as writable if possible */
1359 if (unlikely(!write_fault) && writable) {
1360 struct page *wpage[1];
1362 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1371 *pfn = page_to_pfn(page[0]);
1375 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1377 if (unlikely(!(vma->vm_flags & VM_READ)))
1380 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1387 * Pin guest page in memory and return its pfn.
1388 * @addr: host virtual address which maps memory to the guest
1389 * @atomic: whether this function can sleep
1390 * @async: whether this function need to wait IO complete if the
1391 * host page is not in the memory
1392 * @write_fault: whether we should get a writable host page
1393 * @writable: whether it allows to map a writable host page for !@write_fault
1395 * The function will map a writable host page for these two cases:
1396 * 1): @write_fault = true
1397 * 2): @write_fault = false && @writable, @writable will tell the caller
1398 * whether the mapping is writable.
1400 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1401 bool write_fault, bool *writable)
1403 struct vm_area_struct *vma;
1407 /* we can do it either atomically or asynchronously, not both */
1408 BUG_ON(atomic && async);
1410 if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1414 return KVM_PFN_ERR_FAULT;
1416 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1420 down_read(¤t->mm->mmap_sem);
1421 if (npages == -EHWPOISON ||
1422 (!async && check_user_page_hwpoison(addr))) {
1423 pfn = KVM_PFN_ERR_HWPOISON;
1427 vma = find_vma_intersection(current->mm, addr, addr + 1);
1430 pfn = KVM_PFN_ERR_FAULT;
1431 else if ((vma->vm_flags & VM_PFNMAP)) {
1432 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1434 BUG_ON(!kvm_is_reserved_pfn(pfn));
1436 if (async && vma_is_valid(vma, write_fault))
1438 pfn = KVM_PFN_ERR_FAULT;
1441 up_read(¤t->mm->mmap_sem);
1445 pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1446 bool *async, bool write_fault, bool *writable)
1448 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1450 if (addr == KVM_HVA_ERR_RO_BAD)
1451 return KVM_PFN_ERR_RO_FAULT;
1453 if (kvm_is_error_hva(addr))
1454 return KVM_PFN_NOSLOT;
1456 /* Do not map writable pfn in the readonly memslot. */
1457 if (writable && memslot_is_readonly(slot)) {
1462 return hva_to_pfn(addr, atomic, async, write_fault,
1465 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
1467 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1470 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
1471 write_fault, writable);
1473 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1475 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1477 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1479 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
1481 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1483 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1485 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1487 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1489 return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm, gfn), gfn);
1491 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1493 pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
1495 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1497 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
1499 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1501 return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
1503 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1505 pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
1507 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1509 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
1511 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1512 struct page **pages, int nr_pages)
1517 addr = gfn_to_hva_many(slot, gfn, &entry);
1518 if (kvm_is_error_hva(addr))
1521 if (entry < nr_pages)
1524 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1526 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1528 static struct page *kvm_pfn_to_page(pfn_t pfn)
1530 if (is_error_noslot_pfn(pfn))
1531 return KVM_ERR_PTR_BAD_PAGE;
1533 if (kvm_is_reserved_pfn(pfn)) {
1535 return KVM_ERR_PTR_BAD_PAGE;
1538 return pfn_to_page(pfn);
1541 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1545 pfn = gfn_to_pfn(kvm, gfn);
1547 return kvm_pfn_to_page(pfn);
1549 EXPORT_SYMBOL_GPL(gfn_to_page);
1551 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
1555 pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
1557 return kvm_pfn_to_page(pfn);
1559 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
1561 void kvm_release_page_clean(struct page *page)
1563 WARN_ON(is_error_page(page));
1565 kvm_release_pfn_clean(page_to_pfn(page));
1567 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1569 void kvm_release_pfn_clean(pfn_t pfn)
1571 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
1572 put_page(pfn_to_page(pfn));
1574 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1576 void kvm_release_page_dirty(struct page *page)
1578 WARN_ON(is_error_page(page));
1580 kvm_release_pfn_dirty(page_to_pfn(page));
1582 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1584 static void kvm_release_pfn_dirty(pfn_t pfn)
1586 kvm_set_pfn_dirty(pfn);
1587 kvm_release_pfn_clean(pfn);
1590 void kvm_set_pfn_dirty(pfn_t pfn)
1592 if (!kvm_is_reserved_pfn(pfn)) {
1593 struct page *page = pfn_to_page(pfn);
1595 if (!PageReserved(page))
1599 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1601 void kvm_set_pfn_accessed(pfn_t pfn)
1603 if (!kvm_is_reserved_pfn(pfn))
1604 mark_page_accessed(pfn_to_page(pfn));
1606 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1608 void kvm_get_pfn(pfn_t pfn)
1610 if (!kvm_is_reserved_pfn(pfn))
1611 get_page(pfn_to_page(pfn));
1613 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1615 static int next_segment(unsigned long len, int offset)
1617 if (len > PAGE_SIZE - offset)
1618 return PAGE_SIZE - offset;
1623 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
1624 void *data, int offset, int len)
1629 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1630 if (kvm_is_error_hva(addr))
1632 r = __copy_from_user(data, (void __user *)addr + offset, len);
1638 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1641 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1643 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1645 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1647 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
1648 int offset, int len)
1650 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1652 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1654 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
1656 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1658 gfn_t gfn = gpa >> PAGE_SHIFT;
1660 int offset = offset_in_page(gpa);
1663 while ((seg = next_segment(len, offset)) != 0) {
1664 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1674 EXPORT_SYMBOL_GPL(kvm_read_guest);
1676 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
1678 gfn_t gfn = gpa >> PAGE_SHIFT;
1680 int offset = offset_in_page(gpa);
1683 while ((seg = next_segment(len, offset)) != 0) {
1684 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
1694 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
1696 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1697 void *data, int offset, unsigned long len)
1702 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1703 if (kvm_is_error_hva(addr))
1705 pagefault_disable();
1706 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1713 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1716 gfn_t gfn = gpa >> PAGE_SHIFT;
1717 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1718 int offset = offset_in_page(gpa);
1720 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1722 EXPORT_SYMBOL_GPL(kvm_read_guest_atomic);
1724 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
1725 void *data, unsigned long len)
1727 gfn_t gfn = gpa >> PAGE_SHIFT;
1728 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1729 int offset = offset_in_page(gpa);
1731 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1733 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
1735 static int __kvm_write_guest_page(struct kvm_memory_slot *memslot, gfn_t gfn,
1736 const void *data, int offset, int len)
1741 addr = gfn_to_hva_memslot(memslot, gfn);
1742 if (kvm_is_error_hva(addr))
1744 r = __copy_to_user((void __user *)addr + offset, data, len);
1747 mark_page_dirty_in_slot(memslot, gfn);
1751 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
1752 const void *data, int offset, int len)
1754 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1756 return __kvm_write_guest_page(slot, gfn, data, offset, len);
1758 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1760 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
1761 const void *data, int offset, int len)
1763 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1765 return __kvm_write_guest_page(slot, gfn, data, offset, len);
1767 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
1769 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1772 gfn_t gfn = gpa >> PAGE_SHIFT;
1774 int offset = offset_in_page(gpa);
1777 while ((seg = next_segment(len, offset)) != 0) {
1778 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1788 EXPORT_SYMBOL_GPL(kvm_write_guest);
1790 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
1793 gfn_t gfn = gpa >> PAGE_SHIFT;
1795 int offset = offset_in_page(gpa);
1798 while ((seg = next_segment(len, offset)) != 0) {
1799 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
1809 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
1811 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1812 gpa_t gpa, unsigned long len)
1814 struct kvm_memslots *slots = kvm_memslots(kvm);
1815 int offset = offset_in_page(gpa);
1816 gfn_t start_gfn = gpa >> PAGE_SHIFT;
1817 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1818 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1819 gfn_t nr_pages_avail;
1822 ghc->generation = slots->generation;
1824 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1825 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, NULL);
1826 if (!kvm_is_error_hva(ghc->hva) && nr_pages_needed <= 1) {
1830 * If the requested region crosses two memslots, we still
1831 * verify that the entire region is valid here.
1833 while (start_gfn <= end_gfn) {
1834 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1835 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1837 if (kvm_is_error_hva(ghc->hva))
1839 start_gfn += nr_pages_avail;
1841 /* Use the slow path for cross page reads and writes. */
1842 ghc->memslot = NULL;
1846 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1848 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1849 void *data, unsigned long len)
1851 struct kvm_memslots *slots = kvm_memslots(kvm);
1854 BUG_ON(len > ghc->len);
1856 if (slots->generation != ghc->generation)
1857 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1859 if (unlikely(!ghc->memslot))
1860 return kvm_write_guest(kvm, ghc->gpa, data, len);
1862 if (kvm_is_error_hva(ghc->hva))
1865 r = __copy_to_user((void __user *)ghc->hva, data, len);
1868 mark_page_dirty_in_slot(ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1872 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1874 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1875 void *data, unsigned long len)
1877 struct kvm_memslots *slots = kvm_memslots(kvm);
1880 BUG_ON(len > ghc->len);
1882 if (slots->generation != ghc->generation)
1883 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1885 if (unlikely(!ghc->memslot))
1886 return kvm_read_guest(kvm, ghc->gpa, data, len);
1888 if (kvm_is_error_hva(ghc->hva))
1891 r = __copy_from_user(data, (void __user *)ghc->hva, len);
1897 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1899 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1901 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
1903 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
1905 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1907 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1909 gfn_t gfn = gpa >> PAGE_SHIFT;
1911 int offset = offset_in_page(gpa);
1914 while ((seg = next_segment(len, offset)) != 0) {
1915 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1924 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1926 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot,
1929 if (memslot && memslot->dirty_bitmap) {
1930 unsigned long rel_gfn = gfn - memslot->base_gfn;
1932 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1936 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1938 struct kvm_memory_slot *memslot;
1940 memslot = gfn_to_memslot(kvm, gfn);
1941 mark_page_dirty_in_slot(memslot, gfn);
1943 EXPORT_SYMBOL_GPL(mark_page_dirty);
1945 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
1947 struct kvm_memory_slot *memslot;
1949 memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1950 mark_page_dirty_in_slot(memslot, gfn);
1952 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
1954 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
1958 old = val = vcpu->halt_poll_ns;
1960 if (val == 0 && halt_poll_ns_grow)
1963 val *= halt_poll_ns_grow;
1965 if (val > halt_poll_ns)
1968 vcpu->halt_poll_ns = val;
1969 trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
1972 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
1976 old = val = vcpu->halt_poll_ns;
1977 if (halt_poll_ns_shrink == 0)
1980 val /= halt_poll_ns_shrink;
1982 vcpu->halt_poll_ns = val;
1983 trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
1986 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
1988 if (kvm_arch_vcpu_runnable(vcpu)) {
1989 kvm_make_request(KVM_REQ_UNHALT, vcpu);
1992 if (kvm_cpu_has_pending_timer(vcpu))
1994 if (signal_pending(current))
2001 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2003 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
2007 bool waited = false;
2010 start = cur = ktime_get();
2011 if (vcpu->halt_poll_ns) {
2012 ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns);
2014 ++vcpu->stat.halt_attempted_poll;
2017 * This sets KVM_REQ_UNHALT if an interrupt
2020 if (kvm_vcpu_check_block(vcpu) < 0) {
2021 ++vcpu->stat.halt_successful_poll;
2025 } while (single_task_running() && ktime_before(cur, stop));
2028 kvm_arch_vcpu_blocking(vcpu);
2031 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
2033 if (kvm_vcpu_check_block(vcpu) < 0)
2040 finish_wait(&vcpu->wq, &wait);
2043 kvm_arch_vcpu_unblocking(vcpu);
2045 block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
2048 if (block_ns <= vcpu->halt_poll_ns)
2050 /* we had a long block, shrink polling */
2051 else if (vcpu->halt_poll_ns && block_ns > halt_poll_ns)
2052 shrink_halt_poll_ns(vcpu);
2053 /* we had a short halt and our poll time is too small */
2054 else if (vcpu->halt_poll_ns < halt_poll_ns &&
2055 block_ns < halt_poll_ns)
2056 grow_halt_poll_ns(vcpu);
2058 vcpu->halt_poll_ns = 0;
2060 trace_kvm_vcpu_wakeup(block_ns, waited);
2062 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
2066 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2068 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
2071 int cpu = vcpu->cpu;
2072 wait_queue_head_t *wqp;
2074 wqp = kvm_arch_vcpu_wq(vcpu);
2075 if (waitqueue_active(wqp)) {
2076 wake_up_interruptible(wqp);
2077 ++vcpu->stat.halt_wakeup;
2081 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
2082 if (kvm_arch_vcpu_should_kick(vcpu))
2083 smp_send_reschedule(cpu);
2086 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
2087 #endif /* !CONFIG_S390 */
2089 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
2092 struct task_struct *task = NULL;
2096 pid = rcu_dereference(target->pid);
2098 task = get_pid_task(pid, PIDTYPE_PID);
2102 ret = yield_to(task, 1);
2103 put_task_struct(task);
2107 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
2110 * Helper that checks whether a VCPU is eligible for directed yield.
2111 * Most eligible candidate to yield is decided by following heuristics:
2113 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2114 * (preempted lock holder), indicated by @in_spin_loop.
2115 * Set at the beiginning and cleared at the end of interception/PLE handler.
2117 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2118 * chance last time (mostly it has become eligible now since we have probably
2119 * yielded to lockholder in last iteration. This is done by toggling
2120 * @dy_eligible each time a VCPU checked for eligibility.)
2122 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2123 * to preempted lock-holder could result in wrong VCPU selection and CPU
2124 * burning. Giving priority for a potential lock-holder increases lock
2127 * Since algorithm is based on heuristics, accessing another VCPU data without
2128 * locking does not harm. It may result in trying to yield to same VCPU, fail
2129 * and continue with next VCPU and so on.
2131 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
2133 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2136 eligible = !vcpu->spin_loop.in_spin_loop ||
2137 vcpu->spin_loop.dy_eligible;
2139 if (vcpu->spin_loop.in_spin_loop)
2140 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
2148 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
2150 struct kvm *kvm = me->kvm;
2151 struct kvm_vcpu *vcpu;
2152 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
2158 kvm_vcpu_set_in_spin_loop(me, true);
2160 * We boost the priority of a VCPU that is runnable but not
2161 * currently running, because it got preempted by something
2162 * else and called schedule in __vcpu_run. Hopefully that
2163 * VCPU is holding the lock that we need and will release it.
2164 * We approximate round-robin by starting at the last boosted VCPU.
2166 for (pass = 0; pass < 2 && !yielded && try; pass++) {
2167 kvm_for_each_vcpu(i, vcpu, kvm) {
2168 if (!pass && i <= last_boosted_vcpu) {
2169 i = last_boosted_vcpu;
2171 } else if (pass && i > last_boosted_vcpu)
2173 if (!ACCESS_ONCE(vcpu->preempted))
2177 if (waitqueue_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
2179 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
2182 yielded = kvm_vcpu_yield_to(vcpu);
2184 kvm->last_boosted_vcpu = i;
2186 } else if (yielded < 0) {
2193 kvm_vcpu_set_in_spin_loop(me, false);
2195 /* Ensure vcpu is not eligible during next spinloop */
2196 kvm_vcpu_set_dy_eligible(me, false);
2198 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
2200 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2202 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
2205 if (vmf->pgoff == 0)
2206 page = virt_to_page(vcpu->run);
2208 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
2209 page = virt_to_page(vcpu->arch.pio_data);
2211 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2212 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
2213 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
2216 return kvm_arch_vcpu_fault(vcpu, vmf);
2222 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
2223 .fault = kvm_vcpu_fault,
2226 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2228 vma->vm_ops = &kvm_vcpu_vm_ops;
2232 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2234 struct kvm_vcpu *vcpu = filp->private_data;
2236 kvm_put_kvm(vcpu->kvm);
2240 static struct file_operations kvm_vcpu_fops = {
2241 .release = kvm_vcpu_release,
2242 .unlocked_ioctl = kvm_vcpu_ioctl,
2243 #ifdef CONFIG_KVM_COMPAT
2244 .compat_ioctl = kvm_vcpu_compat_ioctl,
2246 .mmap = kvm_vcpu_mmap,
2247 .llseek = noop_llseek,
2251 * Allocates an inode for the vcpu.
2253 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2255 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
2259 * Creates some virtual cpus. Good luck creating more than one.
2261 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
2264 struct kvm_vcpu *vcpu, *v;
2266 if (id >= KVM_MAX_VCPUS)
2269 vcpu = kvm_arch_vcpu_create(kvm, id);
2271 return PTR_ERR(vcpu);
2273 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2275 r = kvm_arch_vcpu_setup(vcpu);
2279 mutex_lock(&kvm->lock);
2280 if (!kvm_vcpu_compatible(vcpu)) {
2282 goto unlock_vcpu_destroy;
2284 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
2286 goto unlock_vcpu_destroy;
2289 kvm_for_each_vcpu(r, v, kvm)
2290 if (v->vcpu_id == id) {
2292 goto unlock_vcpu_destroy;
2295 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
2297 /* Now it's all set up, let userspace reach it */
2299 r = create_vcpu_fd(vcpu);
2302 goto unlock_vcpu_destroy;
2305 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
2308 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
2309 * before kvm->online_vcpu's incremented value.
2312 atomic_inc(&kvm->online_vcpus);
2314 mutex_unlock(&kvm->lock);
2315 kvm_arch_vcpu_postcreate(vcpu);
2318 unlock_vcpu_destroy:
2319 mutex_unlock(&kvm->lock);
2321 kvm_arch_vcpu_destroy(vcpu);
2325 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2328 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2329 vcpu->sigset_active = 1;
2330 vcpu->sigset = *sigset;
2332 vcpu->sigset_active = 0;
2336 static long kvm_vcpu_ioctl(struct file *filp,
2337 unsigned int ioctl, unsigned long arg)
2339 struct kvm_vcpu *vcpu = filp->private_data;
2340 void __user *argp = (void __user *)arg;
2342 struct kvm_fpu *fpu = NULL;
2343 struct kvm_sregs *kvm_sregs = NULL;
2345 if (vcpu->kvm->mm != current->mm)
2348 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2351 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
2353 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
2354 * so vcpu_load() would break it.
2356 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_S390_IRQ || ioctl == KVM_INTERRUPT)
2357 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2361 r = vcpu_load(vcpu);
2369 if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
2370 /* The thread running this VCPU changed. */
2371 struct pid *oldpid = vcpu->pid;
2372 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
2374 rcu_assign_pointer(vcpu->pid, newpid);
2379 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2380 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2382 case KVM_GET_REGS: {
2383 struct kvm_regs *kvm_regs;
2386 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2389 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2393 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2400 case KVM_SET_REGS: {
2401 struct kvm_regs *kvm_regs;
2404 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2405 if (IS_ERR(kvm_regs)) {
2406 r = PTR_ERR(kvm_regs);
2409 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2413 case KVM_GET_SREGS: {
2414 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2418 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2422 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2427 case KVM_SET_SREGS: {
2428 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2429 if (IS_ERR(kvm_sregs)) {
2430 r = PTR_ERR(kvm_sregs);
2434 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2437 case KVM_GET_MP_STATE: {
2438 struct kvm_mp_state mp_state;
2440 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2444 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
2449 case KVM_SET_MP_STATE: {
2450 struct kvm_mp_state mp_state;
2453 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
2455 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2458 case KVM_TRANSLATE: {
2459 struct kvm_translation tr;
2462 if (copy_from_user(&tr, argp, sizeof(tr)))
2464 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2468 if (copy_to_user(argp, &tr, sizeof(tr)))
2473 case KVM_SET_GUEST_DEBUG: {
2474 struct kvm_guest_debug dbg;
2477 if (copy_from_user(&dbg, argp, sizeof(dbg)))
2479 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2482 case KVM_SET_SIGNAL_MASK: {
2483 struct kvm_signal_mask __user *sigmask_arg = argp;
2484 struct kvm_signal_mask kvm_sigmask;
2485 sigset_t sigset, *p;
2490 if (copy_from_user(&kvm_sigmask, argp,
2491 sizeof(kvm_sigmask)))
2494 if (kvm_sigmask.len != sizeof(sigset))
2497 if (copy_from_user(&sigset, sigmask_arg->sigset,
2502 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2506 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2510 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2514 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2520 fpu = memdup_user(argp, sizeof(*fpu));
2526 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2530 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2539 #ifdef CONFIG_KVM_COMPAT
2540 static long kvm_vcpu_compat_ioctl(struct file *filp,
2541 unsigned int ioctl, unsigned long arg)
2543 struct kvm_vcpu *vcpu = filp->private_data;
2544 void __user *argp = compat_ptr(arg);
2547 if (vcpu->kvm->mm != current->mm)
2551 case KVM_SET_SIGNAL_MASK: {
2552 struct kvm_signal_mask __user *sigmask_arg = argp;
2553 struct kvm_signal_mask kvm_sigmask;
2554 compat_sigset_t csigset;
2559 if (copy_from_user(&kvm_sigmask, argp,
2560 sizeof(kvm_sigmask)))
2563 if (kvm_sigmask.len != sizeof(csigset))
2566 if (copy_from_user(&csigset, sigmask_arg->sigset,
2569 sigset_from_compat(&sigset, &csigset);
2570 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2572 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2576 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2584 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2585 int (*accessor)(struct kvm_device *dev,
2586 struct kvm_device_attr *attr),
2589 struct kvm_device_attr attr;
2594 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2597 return accessor(dev, &attr);
2600 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2603 struct kvm_device *dev = filp->private_data;
2606 case KVM_SET_DEVICE_ATTR:
2607 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2608 case KVM_GET_DEVICE_ATTR:
2609 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2610 case KVM_HAS_DEVICE_ATTR:
2611 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2613 if (dev->ops->ioctl)
2614 return dev->ops->ioctl(dev, ioctl, arg);
2620 static int kvm_device_release(struct inode *inode, struct file *filp)
2622 struct kvm_device *dev = filp->private_data;
2623 struct kvm *kvm = dev->kvm;
2629 static const struct file_operations kvm_device_fops = {
2630 .unlocked_ioctl = kvm_device_ioctl,
2631 #ifdef CONFIG_KVM_COMPAT
2632 .compat_ioctl = kvm_device_ioctl,
2634 .release = kvm_device_release,
2637 struct kvm_device *kvm_device_from_filp(struct file *filp)
2639 if (filp->f_op != &kvm_device_fops)
2642 return filp->private_data;
2645 static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
2646 #ifdef CONFIG_KVM_MPIC
2647 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
2648 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
2651 #ifdef CONFIG_KVM_XICS
2652 [KVM_DEV_TYPE_XICS] = &kvm_xics_ops,
2656 int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
2658 if (type >= ARRAY_SIZE(kvm_device_ops_table))
2661 if (kvm_device_ops_table[type] != NULL)
2664 kvm_device_ops_table[type] = ops;
2668 void kvm_unregister_device_ops(u32 type)
2670 if (kvm_device_ops_table[type] != NULL)
2671 kvm_device_ops_table[type] = NULL;
2674 static int kvm_ioctl_create_device(struct kvm *kvm,
2675 struct kvm_create_device *cd)
2677 struct kvm_device_ops *ops = NULL;
2678 struct kvm_device *dev;
2679 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2682 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
2685 ops = kvm_device_ops_table[cd->type];
2692 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2699 ret = ops->create(dev, cd->type);
2705 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2711 list_add(&dev->vm_node, &kvm->devices);
2717 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2720 case KVM_CAP_USER_MEMORY:
2721 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2722 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2723 case KVM_CAP_INTERNAL_ERROR_DATA:
2724 #ifdef CONFIG_HAVE_KVM_MSI
2725 case KVM_CAP_SIGNAL_MSI:
2727 #ifdef CONFIG_HAVE_KVM_IRQFD
2729 case KVM_CAP_IRQFD_RESAMPLE:
2731 case KVM_CAP_IOEVENTFD_ANY_LENGTH:
2732 case KVM_CAP_CHECK_EXTENSION_VM:
2734 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2735 case KVM_CAP_IRQ_ROUTING:
2736 return KVM_MAX_IRQ_ROUTES;
2738 #if KVM_ADDRESS_SPACE_NUM > 1
2739 case KVM_CAP_MULTI_ADDRESS_SPACE:
2740 return KVM_ADDRESS_SPACE_NUM;
2745 return kvm_vm_ioctl_check_extension(kvm, arg);
2748 static long kvm_vm_ioctl(struct file *filp,
2749 unsigned int ioctl, unsigned long arg)
2751 struct kvm *kvm = filp->private_data;
2752 void __user *argp = (void __user *)arg;
2755 if (kvm->mm != current->mm)
2758 case KVM_CREATE_VCPU:
2759 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2761 case KVM_SET_USER_MEMORY_REGION: {
2762 struct kvm_userspace_memory_region kvm_userspace_mem;
2765 if (copy_from_user(&kvm_userspace_mem, argp,
2766 sizeof(kvm_userspace_mem)))
2769 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2772 case KVM_GET_DIRTY_LOG: {
2773 struct kvm_dirty_log log;
2776 if (copy_from_user(&log, argp, sizeof(log)))
2778 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2781 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2782 case KVM_REGISTER_COALESCED_MMIO: {
2783 struct kvm_coalesced_mmio_zone zone;
2786 if (copy_from_user(&zone, argp, sizeof(zone)))
2788 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2791 case KVM_UNREGISTER_COALESCED_MMIO: {
2792 struct kvm_coalesced_mmio_zone zone;
2795 if (copy_from_user(&zone, argp, sizeof(zone)))
2797 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2802 struct kvm_irqfd data;
2805 if (copy_from_user(&data, argp, sizeof(data)))
2807 r = kvm_irqfd(kvm, &data);
2810 case KVM_IOEVENTFD: {
2811 struct kvm_ioeventfd data;
2814 if (copy_from_user(&data, argp, sizeof(data)))
2816 r = kvm_ioeventfd(kvm, &data);
2819 #ifdef CONFIG_HAVE_KVM_MSI
2820 case KVM_SIGNAL_MSI: {
2824 if (copy_from_user(&msi, argp, sizeof(msi)))
2826 r = kvm_send_userspace_msi(kvm, &msi);
2830 #ifdef __KVM_HAVE_IRQ_LINE
2831 case KVM_IRQ_LINE_STATUS:
2832 case KVM_IRQ_LINE: {
2833 struct kvm_irq_level irq_event;
2836 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
2839 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
2840 ioctl == KVM_IRQ_LINE_STATUS);
2845 if (ioctl == KVM_IRQ_LINE_STATUS) {
2846 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
2854 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2855 case KVM_SET_GSI_ROUTING: {
2856 struct kvm_irq_routing routing;
2857 struct kvm_irq_routing __user *urouting;
2858 struct kvm_irq_routing_entry *entries;
2861 if (copy_from_user(&routing, argp, sizeof(routing)))
2864 if (routing.nr > KVM_MAX_IRQ_ROUTES)
2869 entries = vmalloc(routing.nr * sizeof(*entries));
2874 if (copy_from_user(entries, urouting->entries,
2875 routing.nr * sizeof(*entries)))
2876 goto out_free_irq_routing;
2877 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2879 out_free_irq_routing:
2883 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
2884 case KVM_CREATE_DEVICE: {
2885 struct kvm_create_device cd;
2888 if (copy_from_user(&cd, argp, sizeof(cd)))
2891 r = kvm_ioctl_create_device(kvm, &cd);
2896 if (copy_to_user(argp, &cd, sizeof(cd)))
2902 case KVM_CHECK_EXTENSION:
2903 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
2906 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2912 #ifdef CONFIG_KVM_COMPAT
2913 struct compat_kvm_dirty_log {
2917 compat_uptr_t dirty_bitmap; /* one bit per page */
2922 static long kvm_vm_compat_ioctl(struct file *filp,
2923 unsigned int ioctl, unsigned long arg)
2925 struct kvm *kvm = filp->private_data;
2928 if (kvm->mm != current->mm)
2931 case KVM_GET_DIRTY_LOG: {
2932 struct compat_kvm_dirty_log compat_log;
2933 struct kvm_dirty_log log;
2936 if (copy_from_user(&compat_log, (void __user *)arg,
2937 sizeof(compat_log)))
2939 log.slot = compat_log.slot;
2940 log.padding1 = compat_log.padding1;
2941 log.padding2 = compat_log.padding2;
2942 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2944 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2948 r = kvm_vm_ioctl(filp, ioctl, arg);
2956 static struct file_operations kvm_vm_fops = {
2957 .release = kvm_vm_release,
2958 .unlocked_ioctl = kvm_vm_ioctl,
2959 #ifdef CONFIG_KVM_COMPAT
2960 .compat_ioctl = kvm_vm_compat_ioctl,
2962 .llseek = noop_llseek,
2965 static int kvm_dev_ioctl_create_vm(unsigned long type)
2970 kvm = kvm_create_vm(type);
2972 return PTR_ERR(kvm);
2973 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2974 r = kvm_coalesced_mmio_init(kvm);
2980 r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC);
2987 static long kvm_dev_ioctl(struct file *filp,
2988 unsigned int ioctl, unsigned long arg)
2993 case KVM_GET_API_VERSION:
2996 r = KVM_API_VERSION;
2999 r = kvm_dev_ioctl_create_vm(arg);
3001 case KVM_CHECK_EXTENSION:
3002 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
3004 case KVM_GET_VCPU_MMAP_SIZE:
3007 r = PAGE_SIZE; /* struct kvm_run */
3009 r += PAGE_SIZE; /* pio data page */
3011 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
3012 r += PAGE_SIZE; /* coalesced mmio ring page */
3015 case KVM_TRACE_ENABLE:
3016 case KVM_TRACE_PAUSE:
3017 case KVM_TRACE_DISABLE:
3021 return kvm_arch_dev_ioctl(filp, ioctl, arg);
3027 static struct file_operations kvm_chardev_ops = {
3028 .unlocked_ioctl = kvm_dev_ioctl,
3029 .compat_ioctl = kvm_dev_ioctl,
3030 .llseek = noop_llseek,
3033 static struct miscdevice kvm_dev = {
3039 static void hardware_enable_nolock(void *junk)
3041 int cpu = raw_smp_processor_id();
3044 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
3047 cpumask_set_cpu(cpu, cpus_hardware_enabled);
3049 r = kvm_arch_hardware_enable();
3052 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3053 atomic_inc(&hardware_enable_failed);
3054 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
3058 static void hardware_enable(void)
3060 raw_spin_lock(&kvm_count_lock);
3061 if (kvm_usage_count)
3062 hardware_enable_nolock(NULL);
3063 raw_spin_unlock(&kvm_count_lock);
3066 static void hardware_disable_nolock(void *junk)
3068 int cpu = raw_smp_processor_id();
3070 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
3072 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3073 kvm_arch_hardware_disable();
3076 static void hardware_disable(void)
3078 raw_spin_lock(&kvm_count_lock);
3079 if (kvm_usage_count)
3080 hardware_disable_nolock(NULL);
3081 raw_spin_unlock(&kvm_count_lock);
3084 static void hardware_disable_all_nolock(void)
3086 BUG_ON(!kvm_usage_count);
3089 if (!kvm_usage_count)
3090 on_each_cpu(hardware_disable_nolock, NULL, 1);
3093 static void hardware_disable_all(void)
3095 raw_spin_lock(&kvm_count_lock);
3096 hardware_disable_all_nolock();
3097 raw_spin_unlock(&kvm_count_lock);
3100 static int hardware_enable_all(void)
3104 raw_spin_lock(&kvm_count_lock);
3107 if (kvm_usage_count == 1) {
3108 atomic_set(&hardware_enable_failed, 0);
3109 on_each_cpu(hardware_enable_nolock, NULL, 1);
3111 if (atomic_read(&hardware_enable_failed)) {
3112 hardware_disable_all_nolock();
3117 raw_spin_unlock(&kvm_count_lock);
3122 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
3125 val &= ~CPU_TASKS_FROZEN;
3137 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3141 * Some (well, at least mine) BIOSes hang on reboot if
3144 * And Intel TXT required VMX off for all cpu when system shutdown.
3146 pr_info("kvm: exiting hardware virtualization\n");
3147 kvm_rebooting = true;
3148 on_each_cpu(hardware_disable_nolock, NULL, 1);
3152 static struct notifier_block kvm_reboot_notifier = {
3153 .notifier_call = kvm_reboot,
3157 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3161 for (i = 0; i < bus->dev_count; i++) {
3162 struct kvm_io_device *pos = bus->range[i].dev;
3164 kvm_iodevice_destructor(pos);
3169 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
3170 const struct kvm_io_range *r2)
3172 gpa_t addr1 = r1->addr;
3173 gpa_t addr2 = r2->addr;
3178 /* If r2->len == 0, match the exact address. If r2->len != 0,
3179 * accept any overlapping write. Any order is acceptable for
3180 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
3181 * we process all of them.
3194 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
3196 return kvm_io_bus_cmp(p1, p2);
3199 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
3200 gpa_t addr, int len)
3202 bus->range[bus->dev_count++] = (struct kvm_io_range) {
3208 sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
3209 kvm_io_bus_sort_cmp, NULL);
3214 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
3215 gpa_t addr, int len)
3217 struct kvm_io_range *range, key;
3220 key = (struct kvm_io_range) {
3225 range = bsearch(&key, bus->range, bus->dev_count,
3226 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
3230 off = range - bus->range;
3232 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
3238 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3239 struct kvm_io_range *range, const void *val)
3243 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3247 while (idx < bus->dev_count &&
3248 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3249 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
3258 /* kvm_io_bus_write - called under kvm->slots_lock */
3259 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3260 int len, const void *val)
3262 struct kvm_io_bus *bus;
3263 struct kvm_io_range range;
3266 range = (struct kvm_io_range) {
3271 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3272 r = __kvm_io_bus_write(vcpu, bus, &range, val);
3273 return r < 0 ? r : 0;
3276 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3277 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
3278 gpa_t addr, int len, const void *val, long cookie)
3280 struct kvm_io_bus *bus;
3281 struct kvm_io_range range;
3283 range = (struct kvm_io_range) {
3288 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3290 /* First try the device referenced by cookie. */
3291 if ((cookie >= 0) && (cookie < bus->dev_count) &&
3292 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
3293 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
3298 * cookie contained garbage; fall back to search and return the
3299 * correct cookie value.
3301 return __kvm_io_bus_write(vcpu, bus, &range, val);
3304 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3305 struct kvm_io_range *range, void *val)
3309 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3313 while (idx < bus->dev_count &&
3314 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3315 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
3323 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
3325 /* kvm_io_bus_read - called under kvm->slots_lock */
3326 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3329 struct kvm_io_bus *bus;
3330 struct kvm_io_range range;
3333 range = (struct kvm_io_range) {
3338 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3339 r = __kvm_io_bus_read(vcpu, bus, &range, val);
3340 return r < 0 ? r : 0;
3344 /* Caller must hold slots_lock. */
3345 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3346 int len, struct kvm_io_device *dev)
3348 struct kvm_io_bus *new_bus, *bus;
3350 bus = kvm->buses[bus_idx];
3351 /* exclude ioeventfd which is limited by maximum fd */
3352 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3355 new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count + 1) *
3356 sizeof(struct kvm_io_range)), GFP_KERNEL);
3359 memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
3360 sizeof(struct kvm_io_range)));
3361 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
3362 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3363 synchronize_srcu_expedited(&kvm->srcu);
3369 /* Caller must hold slots_lock. */
3370 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3371 struct kvm_io_device *dev)
3374 struct kvm_io_bus *new_bus, *bus;
3376 bus = kvm->buses[bus_idx];
3378 for (i = 0; i < bus->dev_count; i++)
3379 if (bus->range[i].dev == dev) {
3387 new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3388 sizeof(struct kvm_io_range)), GFP_KERNEL);
3392 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3393 new_bus->dev_count--;
3394 memcpy(new_bus->range + i, bus->range + i + 1,
3395 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3397 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3398 synchronize_srcu_expedited(&kvm->srcu);
3403 static struct notifier_block kvm_cpu_notifier = {
3404 .notifier_call = kvm_cpu_hotplug,
3407 static int vm_stat_get(void *_offset, u64 *val)
3409 unsigned offset = (long)_offset;
3413 spin_lock(&kvm_lock);
3414 list_for_each_entry(kvm, &vm_list, vm_list)
3415 *val += *(u32 *)((void *)kvm + offset);
3416 spin_unlock(&kvm_lock);
3420 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
3422 static int vcpu_stat_get(void *_offset, u64 *val)
3424 unsigned offset = (long)_offset;
3426 struct kvm_vcpu *vcpu;
3430 spin_lock(&kvm_lock);
3431 list_for_each_entry(kvm, &vm_list, vm_list)
3432 kvm_for_each_vcpu(i, vcpu, kvm)
3433 *val += *(u32 *)((void *)vcpu + offset);
3435 spin_unlock(&kvm_lock);
3439 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
3441 static const struct file_operations *stat_fops[] = {
3442 [KVM_STAT_VCPU] = &vcpu_stat_fops,
3443 [KVM_STAT_VM] = &vm_stat_fops,
3446 static int kvm_init_debug(void)
3449 struct kvm_stats_debugfs_item *p;
3451 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3452 if (kvm_debugfs_dir == NULL)
3455 for (p = debugfs_entries; p->name; ++p) {
3456 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
3457 (void *)(long)p->offset,
3458 stat_fops[p->kind]);
3459 if (p->dentry == NULL)
3466 debugfs_remove_recursive(kvm_debugfs_dir);
3471 static void kvm_exit_debug(void)
3473 struct kvm_stats_debugfs_item *p;
3475 for (p = debugfs_entries; p->name; ++p)
3476 debugfs_remove(p->dentry);
3477 debugfs_remove(kvm_debugfs_dir);
3480 static int kvm_suspend(void)
3482 if (kvm_usage_count)
3483 hardware_disable_nolock(NULL);
3487 static void kvm_resume(void)
3489 if (kvm_usage_count) {
3490 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3491 hardware_enable_nolock(NULL);
3495 static struct syscore_ops kvm_syscore_ops = {
3496 .suspend = kvm_suspend,
3497 .resume = kvm_resume,
3501 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3503 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3506 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3508 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3510 if (vcpu->preempted)
3511 vcpu->preempted = false;
3513 kvm_arch_sched_in(vcpu, cpu);
3515 kvm_arch_vcpu_load(vcpu, cpu);
3518 static void kvm_sched_out(struct preempt_notifier *pn,
3519 struct task_struct *next)
3521 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3523 if (current->state == TASK_RUNNING)
3524 vcpu->preempted = true;
3525 kvm_arch_vcpu_put(vcpu);
3528 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3529 struct module *module)
3534 r = kvm_arch_init(opaque);
3539 * kvm_arch_init makes sure there's at most one caller
3540 * for architectures that support multiple implementations,
3541 * like intel and amd on x86.
3542 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3543 * conflicts in case kvm is already setup for another implementation.
3545 r = kvm_irqfd_init();
3549 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
3554 r = kvm_arch_hardware_setup();
3558 for_each_online_cpu(cpu) {
3559 smp_call_function_single(cpu,
3560 kvm_arch_check_processor_compat,
3566 r = register_cpu_notifier(&kvm_cpu_notifier);
3569 register_reboot_notifier(&kvm_reboot_notifier);
3571 /* A kmem cache lets us meet the alignment requirements of fx_save. */
3573 vcpu_align = __alignof__(struct kvm_vcpu);
3574 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
3576 if (!kvm_vcpu_cache) {
3581 r = kvm_async_pf_init();
3585 kvm_chardev_ops.owner = module;
3586 kvm_vm_fops.owner = module;
3587 kvm_vcpu_fops.owner = module;
3589 r = misc_register(&kvm_dev);
3591 pr_err("kvm: misc device register failed\n");
3595 register_syscore_ops(&kvm_syscore_ops);
3597 kvm_preempt_ops.sched_in = kvm_sched_in;
3598 kvm_preempt_ops.sched_out = kvm_sched_out;
3600 r = kvm_init_debug();
3602 pr_err("kvm: create debugfs files failed\n");
3606 r = kvm_vfio_ops_init();
3612 unregister_syscore_ops(&kvm_syscore_ops);
3613 misc_deregister(&kvm_dev);
3615 kvm_async_pf_deinit();
3617 kmem_cache_destroy(kvm_vcpu_cache);
3619 unregister_reboot_notifier(&kvm_reboot_notifier);
3620 unregister_cpu_notifier(&kvm_cpu_notifier);
3623 kvm_arch_hardware_unsetup();
3625 free_cpumask_var(cpus_hardware_enabled);
3633 EXPORT_SYMBOL_GPL(kvm_init);
3638 misc_deregister(&kvm_dev);
3639 kmem_cache_destroy(kvm_vcpu_cache);
3640 kvm_async_pf_deinit();
3641 unregister_syscore_ops(&kvm_syscore_ops);
3642 unregister_reboot_notifier(&kvm_reboot_notifier);
3643 unregister_cpu_notifier(&kvm_cpu_notifier);
3644 on_each_cpu(hardware_disable_nolock, NULL, 1);
3645 kvm_arch_hardware_unsetup();
3648 free_cpumask_var(cpus_hardware_enabled);
3649 kvm_vfio_ops_exit();
3651 EXPORT_SYMBOL_GPL(kvm_exit);