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 r = kvm_arch_init_vm(kvm, type);
552 goto out_err_no_disable;
554 r = hardware_enable_all();
556 goto out_err_no_disable;
558 #ifdef CONFIG_HAVE_KVM_IRQFD
559 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
562 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
565 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
566 kvm->memslots[i] = kvm_alloc_memslots();
567 if (!kvm->memslots[i])
568 goto out_err_no_srcu;
571 if (init_srcu_struct(&kvm->srcu))
572 goto out_err_no_srcu;
573 if (init_srcu_struct(&kvm->irq_srcu))
574 goto out_err_no_irq_srcu;
575 for (i = 0; i < KVM_NR_BUSES; i++) {
576 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
582 spin_lock_init(&kvm->mmu_lock);
583 kvm->mm = current->mm;
584 atomic_inc(&kvm->mm->mm_count);
585 kvm_eventfd_init(kvm);
586 mutex_init(&kvm->lock);
587 mutex_init(&kvm->irq_lock);
588 mutex_init(&kvm->slots_lock);
589 atomic_set(&kvm->users_count, 1);
590 INIT_LIST_HEAD(&kvm->devices);
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);
620 * Avoid using vmalloc for a small buffer.
621 * Should not be used when the size is statically known.
623 void *kvm_kvzalloc(unsigned long size)
625 if (size > PAGE_SIZE)
626 return vzalloc(size);
628 return kzalloc(size, GFP_KERNEL);
631 static void kvm_destroy_devices(struct kvm *kvm)
633 struct list_head *node, *tmp;
635 list_for_each_safe(node, tmp, &kvm->devices) {
636 struct kvm_device *dev =
637 list_entry(node, struct kvm_device, vm_node);
640 dev->ops->destroy(dev);
644 static void kvm_destroy_vm(struct kvm *kvm)
647 struct mm_struct *mm = kvm->mm;
649 kvm_arch_sync_events(kvm);
650 spin_lock(&kvm_lock);
651 list_del(&kvm->vm_list);
652 spin_unlock(&kvm_lock);
653 kvm_free_irq_routing(kvm);
654 for (i = 0; i < KVM_NR_BUSES; i++)
655 kvm_io_bus_destroy(kvm->buses[i]);
656 kvm_coalesced_mmio_free(kvm);
657 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
658 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
660 kvm_arch_flush_shadow_all(kvm);
662 kvm_arch_destroy_vm(kvm);
663 kvm_destroy_devices(kvm);
664 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
665 kvm_free_memslots(kvm, kvm->memslots[i]);
666 cleanup_srcu_struct(&kvm->irq_srcu);
667 cleanup_srcu_struct(&kvm->srcu);
668 kvm_arch_free_vm(kvm);
669 preempt_notifier_dec();
670 hardware_disable_all();
674 void kvm_get_kvm(struct kvm *kvm)
676 atomic_inc(&kvm->users_count);
678 EXPORT_SYMBOL_GPL(kvm_get_kvm);
680 void kvm_put_kvm(struct kvm *kvm)
682 if (atomic_dec_and_test(&kvm->users_count))
685 EXPORT_SYMBOL_GPL(kvm_put_kvm);
688 static int kvm_vm_release(struct inode *inode, struct file *filp)
690 struct kvm *kvm = filp->private_data;
692 kvm_irqfd_release(kvm);
699 * Allocation size is twice as large as the actual dirty bitmap size.
700 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
702 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
704 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
706 memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
707 if (!memslot->dirty_bitmap)
714 * Insert memslot and re-sort memslots based on their GFN,
715 * so binary search could be used to lookup GFN.
716 * Sorting algorithm takes advantage of having initially
717 * sorted array and known changed memslot position.
719 static void update_memslots(struct kvm_memslots *slots,
720 struct kvm_memory_slot *new)
723 int i = slots->id_to_index[id];
724 struct kvm_memory_slot *mslots = slots->memslots;
726 WARN_ON(mslots[i].id != id);
728 WARN_ON(!mslots[i].npages);
729 if (mslots[i].npages)
732 if (!mslots[i].npages)
736 while (i < KVM_MEM_SLOTS_NUM - 1 &&
737 new->base_gfn <= mslots[i + 1].base_gfn) {
738 if (!mslots[i + 1].npages)
740 mslots[i] = mslots[i + 1];
741 slots->id_to_index[mslots[i].id] = i;
746 * The ">=" is needed when creating a slot with base_gfn == 0,
747 * so that it moves before all those with base_gfn == npages == 0.
749 * On the other hand, if new->npages is zero, the above loop has
750 * already left i pointing to the beginning of the empty part of
751 * mslots, and the ">=" would move the hole backwards in this
752 * case---which is wrong. So skip the loop when deleting a slot.
756 new->base_gfn >= mslots[i - 1].base_gfn) {
757 mslots[i] = mslots[i - 1];
758 slots->id_to_index[mslots[i].id] = i;
762 WARN_ON_ONCE(i != slots->used_slots);
765 slots->id_to_index[mslots[i].id] = i;
768 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
770 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
772 #ifdef __KVM_HAVE_READONLY_MEM
773 valid_flags |= KVM_MEM_READONLY;
776 if (mem->flags & ~valid_flags)
782 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
783 int as_id, struct kvm_memslots *slots)
785 struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id);
788 * Set the low bit in the generation, which disables SPTE caching
789 * until the end of synchronize_srcu_expedited.
791 WARN_ON(old_memslots->generation & 1);
792 slots->generation = old_memslots->generation + 1;
794 rcu_assign_pointer(kvm->memslots[as_id], slots);
795 synchronize_srcu_expedited(&kvm->srcu);
798 * Increment the new memslot generation a second time. This prevents
799 * vm exits that race with memslot updates from caching a memslot
800 * generation that will (potentially) be valid forever.
804 kvm_arch_memslots_updated(kvm, slots);
810 * Allocate some memory and give it an address in the guest physical address
813 * Discontiguous memory is allowed, mostly for framebuffers.
815 * Must be called holding kvm->slots_lock for write.
817 int __kvm_set_memory_region(struct kvm *kvm,
818 const struct kvm_userspace_memory_region *mem)
822 unsigned long npages;
823 struct kvm_memory_slot *slot;
824 struct kvm_memory_slot old, new;
825 struct kvm_memslots *slots = NULL, *old_memslots;
827 enum kvm_mr_change change;
829 r = check_memory_region_flags(mem);
834 as_id = mem->slot >> 16;
837 /* General sanity checks */
838 if (mem->memory_size & (PAGE_SIZE - 1))
840 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
842 /* We can read the guest memory with __xxx_user() later on. */
843 if ((id < KVM_USER_MEM_SLOTS) &&
844 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
845 !access_ok(VERIFY_WRITE,
846 (void __user *)(unsigned long)mem->userspace_addr,
849 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
851 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
854 slot = id_to_memslot(__kvm_memslots(kvm, as_id), id);
855 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
856 npages = mem->memory_size >> PAGE_SHIFT;
858 if (npages > KVM_MEM_MAX_NR_PAGES)
864 new.base_gfn = base_gfn;
866 new.flags = mem->flags;
870 change = KVM_MR_CREATE;
871 else { /* Modify an existing slot. */
872 if ((mem->userspace_addr != old.userspace_addr) ||
873 (npages != old.npages) ||
874 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
877 if (base_gfn != old.base_gfn)
878 change = KVM_MR_MOVE;
879 else if (new.flags != old.flags)
880 change = KVM_MR_FLAGS_ONLY;
881 else { /* Nothing to change. */
890 change = KVM_MR_DELETE;
895 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
896 /* Check for overlaps */
898 kvm_for_each_memslot(slot, __kvm_memslots(kvm, as_id)) {
899 if ((slot->id >= KVM_USER_MEM_SLOTS) ||
902 if (!((base_gfn + npages <= slot->base_gfn) ||
903 (base_gfn >= slot->base_gfn + slot->npages)))
908 /* Free page dirty bitmap if unneeded */
909 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
910 new.dirty_bitmap = NULL;
913 if (change == KVM_MR_CREATE) {
914 new.userspace_addr = mem->userspace_addr;
916 if (kvm_arch_create_memslot(kvm, &new, npages))
920 /* Allocate page dirty bitmap if needed */
921 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
922 if (kvm_create_dirty_bitmap(&new) < 0)
926 slots = kvm_kvzalloc(sizeof(struct kvm_memslots));
929 memcpy(slots, __kvm_memslots(kvm, as_id), sizeof(struct kvm_memslots));
931 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
932 slot = id_to_memslot(slots, id);
933 slot->flags |= KVM_MEMSLOT_INVALID;
935 old_memslots = install_new_memslots(kvm, as_id, slots);
937 /* slot was deleted or moved, clear iommu mapping */
938 kvm_iommu_unmap_pages(kvm, &old);
939 /* From this point no new shadow pages pointing to a deleted,
940 * or moved, memslot will be created.
942 * validation of sp->gfn happens in:
943 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
944 * - kvm_is_visible_gfn (mmu_check_roots)
946 kvm_arch_flush_shadow_memslot(kvm, slot);
949 * We can re-use the old_memslots from above, the only difference
950 * from the currently installed memslots is the invalid flag. This
951 * will get overwritten by update_memslots anyway.
953 slots = old_memslots;
956 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
960 /* actual memory is freed via old in kvm_free_memslot below */
961 if (change == KVM_MR_DELETE) {
962 new.dirty_bitmap = NULL;
963 memset(&new.arch, 0, sizeof(new.arch));
966 update_memslots(slots, &new);
967 old_memslots = install_new_memslots(kvm, as_id, slots);
969 kvm_arch_commit_memory_region(kvm, mem, &old, &new, change);
971 kvm_free_memslot(kvm, &old, &new);
972 kvfree(old_memslots);
975 * IOMMU mapping: New slots need to be mapped. Old slots need to be
976 * un-mapped and re-mapped if their base changes. Since base change
977 * unmapping is handled above with slot deletion, mapping alone is
978 * needed here. Anything else the iommu might care about for existing
979 * slots (size changes, userspace addr changes and read-only flag
980 * changes) is disallowed above, so any other attribute changes getting
981 * here can be skipped.
983 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
984 r = kvm_iommu_map_pages(kvm, &new);
993 kvm_free_memslot(kvm, &new, &old);
997 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
999 int kvm_set_memory_region(struct kvm *kvm,
1000 const struct kvm_userspace_memory_region *mem)
1004 mutex_lock(&kvm->slots_lock);
1005 r = __kvm_set_memory_region(kvm, mem);
1006 mutex_unlock(&kvm->slots_lock);
1009 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1011 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1012 struct kvm_userspace_memory_region *mem)
1014 if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1017 return kvm_set_memory_region(kvm, mem);
1020 int kvm_get_dirty_log(struct kvm *kvm,
1021 struct kvm_dirty_log *log, int *is_dirty)
1023 struct kvm_memslots *slots;
1024 struct kvm_memory_slot *memslot;
1025 int r, i, as_id, id;
1027 unsigned long any = 0;
1030 as_id = log->slot >> 16;
1031 id = (u16)log->slot;
1032 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1035 slots = __kvm_memslots(kvm, as_id);
1036 memslot = id_to_memslot(slots, id);
1038 if (!memslot->dirty_bitmap)
1041 n = kvm_dirty_bitmap_bytes(memslot);
1043 for (i = 0; !any && i < n/sizeof(long); ++i)
1044 any = memslot->dirty_bitmap[i];
1047 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1057 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1059 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1061 * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages
1062 * are dirty write protect them for next write.
1063 * @kvm: pointer to kvm instance
1064 * @log: slot id and address to which we copy the log
1065 * @is_dirty: flag set if any page is dirty
1067 * We need to keep it in mind that VCPU threads can write to the bitmap
1068 * concurrently. So, to avoid losing track of dirty pages we keep the
1071 * 1. Take a snapshot of the bit and clear it if needed.
1072 * 2. Write protect the corresponding page.
1073 * 3. Copy the snapshot to the userspace.
1074 * 4. Upon return caller flushes TLB's if needed.
1076 * Between 2 and 4, the guest may write to the page using the remaining TLB
1077 * entry. This is not a problem because the page is reported dirty using
1078 * the snapshot taken before and step 4 ensures that writes done after
1079 * exiting to userspace will be logged for the next call.
1082 int kvm_get_dirty_log_protect(struct kvm *kvm,
1083 struct kvm_dirty_log *log, bool *is_dirty)
1085 struct kvm_memslots *slots;
1086 struct kvm_memory_slot *memslot;
1087 int r, i, as_id, id;
1089 unsigned long *dirty_bitmap;
1090 unsigned long *dirty_bitmap_buffer;
1093 as_id = log->slot >> 16;
1094 id = (u16)log->slot;
1095 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1098 slots = __kvm_memslots(kvm, as_id);
1099 memslot = id_to_memslot(slots, id);
1101 dirty_bitmap = memslot->dirty_bitmap;
1106 n = kvm_dirty_bitmap_bytes(memslot);
1108 dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long);
1109 memset(dirty_bitmap_buffer, 0, n);
1111 spin_lock(&kvm->mmu_lock);
1113 for (i = 0; i < n / sizeof(long); i++) {
1117 if (!dirty_bitmap[i])
1122 mask = xchg(&dirty_bitmap[i], 0);
1123 dirty_bitmap_buffer[i] = mask;
1126 offset = i * BITS_PER_LONG;
1127 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1132 spin_unlock(&kvm->mmu_lock);
1135 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1142 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect);
1145 bool kvm_largepages_enabled(void)
1147 return largepages_enabled;
1150 void kvm_disable_largepages(void)
1152 largepages_enabled = false;
1154 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1156 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1158 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1160 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1162 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
1164 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn);
1167 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1169 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1171 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1172 memslot->flags & KVM_MEMSLOT_INVALID)
1177 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1179 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1181 struct vm_area_struct *vma;
1182 unsigned long addr, size;
1186 addr = gfn_to_hva(kvm, gfn);
1187 if (kvm_is_error_hva(addr))
1190 down_read(¤t->mm->mmap_sem);
1191 vma = find_vma(current->mm, addr);
1195 size = vma_kernel_pagesize(vma);
1198 up_read(¤t->mm->mmap_sem);
1203 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1205 return slot->flags & KVM_MEM_READONLY;
1208 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1209 gfn_t *nr_pages, bool write)
1211 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1212 return KVM_HVA_ERR_BAD;
1214 if (memslot_is_readonly(slot) && write)
1215 return KVM_HVA_ERR_RO_BAD;
1218 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1220 return __gfn_to_hva_memslot(slot, gfn);
1223 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1226 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1229 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1232 return gfn_to_hva_many(slot, gfn, NULL);
1234 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1236 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1238 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1240 EXPORT_SYMBOL_GPL(gfn_to_hva);
1242 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
1244 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
1246 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
1249 * If writable is set to false, the hva returned by this function is only
1250 * allowed to be read.
1252 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1253 gfn_t gfn, bool *writable)
1255 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1257 if (!kvm_is_error_hva(hva) && writable)
1258 *writable = !memslot_is_readonly(slot);
1263 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1265 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1267 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1270 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
1272 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1274 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1277 static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1278 unsigned long start, int write, struct page **page)
1280 int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1283 flags |= FOLL_WRITE;
1285 return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1288 static inline int check_user_page_hwpoison(unsigned long addr)
1290 int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1292 rc = __get_user_pages(current, current->mm, addr, 1,
1293 flags, NULL, NULL, NULL);
1294 return rc == -EHWPOISON;
1298 * The atomic path to get the writable pfn which will be stored in @pfn,
1299 * true indicates success, otherwise false is returned.
1301 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1302 bool write_fault, bool *writable, pfn_t *pfn)
1304 struct page *page[1];
1307 if (!(async || atomic))
1311 * Fast pin a writable pfn only if it is a write fault request
1312 * or the caller allows to map a writable pfn for a read fault
1315 if (!(write_fault || writable))
1318 npages = __get_user_pages_fast(addr, 1, 1, page);
1320 *pfn = page_to_pfn(page[0]);
1331 * The slow path to get the pfn of the specified host virtual address,
1332 * 1 indicates success, -errno is returned if error is detected.
1334 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1335 bool *writable, pfn_t *pfn)
1337 struct page *page[1];
1343 *writable = write_fault;
1346 down_read(¤t->mm->mmap_sem);
1347 npages = get_user_page_nowait(current, current->mm,
1348 addr, write_fault, page);
1349 up_read(¤t->mm->mmap_sem);
1351 npages = __get_user_pages_unlocked(current, current->mm, addr, 1,
1352 write_fault, 0, page,
1353 FOLL_TOUCH|FOLL_HWPOISON);
1357 /* map read fault as writable if possible */
1358 if (unlikely(!write_fault) && writable) {
1359 struct page *wpage[1];
1361 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1370 *pfn = page_to_pfn(page[0]);
1374 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1376 if (unlikely(!(vma->vm_flags & VM_READ)))
1379 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1386 * Pin guest page in memory and return its pfn.
1387 * @addr: host virtual address which maps memory to the guest
1388 * @atomic: whether this function can sleep
1389 * @async: whether this function need to wait IO complete if the
1390 * host page is not in the memory
1391 * @write_fault: whether we should get a writable host page
1392 * @writable: whether it allows to map a writable host page for !@write_fault
1394 * The function will map a writable host page for these two cases:
1395 * 1): @write_fault = true
1396 * 2): @write_fault = false && @writable, @writable will tell the caller
1397 * whether the mapping is writable.
1399 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1400 bool write_fault, bool *writable)
1402 struct vm_area_struct *vma;
1406 /* we can do it either atomically or asynchronously, not both */
1407 BUG_ON(atomic && async);
1409 if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1413 return KVM_PFN_ERR_FAULT;
1415 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1419 down_read(¤t->mm->mmap_sem);
1420 if (npages == -EHWPOISON ||
1421 (!async && check_user_page_hwpoison(addr))) {
1422 pfn = KVM_PFN_ERR_HWPOISON;
1426 vma = find_vma_intersection(current->mm, addr, addr + 1);
1429 pfn = KVM_PFN_ERR_FAULT;
1430 else if ((vma->vm_flags & VM_PFNMAP)) {
1431 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1433 BUG_ON(!kvm_is_reserved_pfn(pfn));
1435 if (async && vma_is_valid(vma, write_fault))
1437 pfn = KVM_PFN_ERR_FAULT;
1440 up_read(¤t->mm->mmap_sem);
1444 pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1445 bool *async, bool write_fault, bool *writable)
1447 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1449 if (addr == KVM_HVA_ERR_RO_BAD)
1450 return KVM_PFN_ERR_RO_FAULT;
1452 if (kvm_is_error_hva(addr))
1453 return KVM_PFN_NOSLOT;
1455 /* Do not map writable pfn in the readonly memslot. */
1456 if (writable && memslot_is_readonly(slot)) {
1461 return hva_to_pfn(addr, atomic, async, write_fault,
1464 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
1466 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1469 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
1470 write_fault, writable);
1472 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1474 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1476 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1478 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
1480 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1482 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1484 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1486 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1488 return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm, gfn), gfn);
1490 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1492 pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
1494 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1496 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
1498 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1500 return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
1502 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1504 pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
1506 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1508 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
1510 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1511 struct page **pages, int nr_pages)
1516 addr = gfn_to_hva_many(slot, gfn, &entry);
1517 if (kvm_is_error_hva(addr))
1520 if (entry < nr_pages)
1523 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1525 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1527 static struct page *kvm_pfn_to_page(pfn_t pfn)
1529 if (is_error_noslot_pfn(pfn))
1530 return KVM_ERR_PTR_BAD_PAGE;
1532 if (kvm_is_reserved_pfn(pfn)) {
1534 return KVM_ERR_PTR_BAD_PAGE;
1537 return pfn_to_page(pfn);
1540 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1544 pfn = gfn_to_pfn(kvm, gfn);
1546 return kvm_pfn_to_page(pfn);
1548 EXPORT_SYMBOL_GPL(gfn_to_page);
1550 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
1554 pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
1556 return kvm_pfn_to_page(pfn);
1558 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
1560 void kvm_release_page_clean(struct page *page)
1562 WARN_ON(is_error_page(page));
1564 kvm_release_pfn_clean(page_to_pfn(page));
1566 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1568 void kvm_release_pfn_clean(pfn_t pfn)
1570 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
1571 put_page(pfn_to_page(pfn));
1573 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1575 void kvm_release_page_dirty(struct page *page)
1577 WARN_ON(is_error_page(page));
1579 kvm_release_pfn_dirty(page_to_pfn(page));
1581 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1583 static void kvm_release_pfn_dirty(pfn_t pfn)
1585 kvm_set_pfn_dirty(pfn);
1586 kvm_release_pfn_clean(pfn);
1589 void kvm_set_pfn_dirty(pfn_t pfn)
1591 if (!kvm_is_reserved_pfn(pfn)) {
1592 struct page *page = pfn_to_page(pfn);
1594 if (!PageReserved(page))
1598 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1600 void kvm_set_pfn_accessed(pfn_t pfn)
1602 if (!kvm_is_reserved_pfn(pfn))
1603 mark_page_accessed(pfn_to_page(pfn));
1605 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1607 void kvm_get_pfn(pfn_t pfn)
1609 if (!kvm_is_reserved_pfn(pfn))
1610 get_page(pfn_to_page(pfn));
1612 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1614 static int next_segment(unsigned long len, int offset)
1616 if (len > PAGE_SIZE - offset)
1617 return PAGE_SIZE - offset;
1622 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
1623 void *data, int offset, int len)
1628 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1629 if (kvm_is_error_hva(addr))
1631 r = __copy_from_user(data, (void __user *)addr + offset, len);
1637 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1640 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1642 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1644 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1646 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
1647 int offset, int len)
1649 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1651 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1653 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
1655 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1657 gfn_t gfn = gpa >> PAGE_SHIFT;
1659 int offset = offset_in_page(gpa);
1662 while ((seg = next_segment(len, offset)) != 0) {
1663 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1673 EXPORT_SYMBOL_GPL(kvm_read_guest);
1675 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
1677 gfn_t gfn = gpa >> PAGE_SHIFT;
1679 int offset = offset_in_page(gpa);
1682 while ((seg = next_segment(len, offset)) != 0) {
1683 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
1693 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
1695 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1696 void *data, int offset, unsigned long len)
1701 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1702 if (kvm_is_error_hva(addr))
1704 pagefault_disable();
1705 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1712 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1715 gfn_t gfn = gpa >> PAGE_SHIFT;
1716 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1717 int offset = offset_in_page(gpa);
1719 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1721 EXPORT_SYMBOL_GPL(kvm_read_guest_atomic);
1723 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
1724 void *data, unsigned long len)
1726 gfn_t gfn = gpa >> PAGE_SHIFT;
1727 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1728 int offset = offset_in_page(gpa);
1730 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1732 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
1734 static int __kvm_write_guest_page(struct kvm_memory_slot *memslot, gfn_t gfn,
1735 const void *data, int offset, int len)
1740 addr = gfn_to_hva_memslot(memslot, gfn);
1741 if (kvm_is_error_hva(addr))
1743 r = __copy_to_user((void __user *)addr + offset, data, len);
1746 mark_page_dirty_in_slot(memslot, gfn);
1750 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
1751 const void *data, int offset, int len)
1753 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1755 return __kvm_write_guest_page(slot, gfn, data, offset, len);
1757 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1759 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
1760 const void *data, int offset, int len)
1762 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1764 return __kvm_write_guest_page(slot, gfn, data, offset, len);
1766 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
1768 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1771 gfn_t gfn = gpa >> PAGE_SHIFT;
1773 int offset = offset_in_page(gpa);
1776 while ((seg = next_segment(len, offset)) != 0) {
1777 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1787 EXPORT_SYMBOL_GPL(kvm_write_guest);
1789 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
1792 gfn_t gfn = gpa >> PAGE_SHIFT;
1794 int offset = offset_in_page(gpa);
1797 while ((seg = next_segment(len, offset)) != 0) {
1798 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
1808 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
1810 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1811 gpa_t gpa, unsigned long len)
1813 struct kvm_memslots *slots = kvm_memslots(kvm);
1814 int offset = offset_in_page(gpa);
1815 gfn_t start_gfn = gpa >> PAGE_SHIFT;
1816 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1817 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1818 gfn_t nr_pages_avail;
1821 ghc->generation = slots->generation;
1823 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1824 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, NULL);
1825 if (!kvm_is_error_hva(ghc->hva) && nr_pages_needed <= 1) {
1829 * If the requested region crosses two memslots, we still
1830 * verify that the entire region is valid here.
1832 while (start_gfn <= end_gfn) {
1833 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1834 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1836 if (kvm_is_error_hva(ghc->hva))
1838 start_gfn += nr_pages_avail;
1840 /* Use the slow path for cross page reads and writes. */
1841 ghc->memslot = NULL;
1845 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1847 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1848 void *data, unsigned long len)
1850 struct kvm_memslots *slots = kvm_memslots(kvm);
1853 BUG_ON(len > ghc->len);
1855 if (slots->generation != ghc->generation)
1856 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1858 if (unlikely(!ghc->memslot))
1859 return kvm_write_guest(kvm, ghc->gpa, data, len);
1861 if (kvm_is_error_hva(ghc->hva))
1864 r = __copy_to_user((void __user *)ghc->hva, data, len);
1867 mark_page_dirty_in_slot(ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1871 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1873 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1874 void *data, unsigned long len)
1876 struct kvm_memslots *slots = kvm_memslots(kvm);
1879 BUG_ON(len > ghc->len);
1881 if (slots->generation != ghc->generation)
1882 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1884 if (unlikely(!ghc->memslot))
1885 return kvm_read_guest(kvm, ghc->gpa, data, len);
1887 if (kvm_is_error_hva(ghc->hva))
1890 r = __copy_from_user(data, (void __user *)ghc->hva, len);
1896 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1898 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1900 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
1902 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
1904 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1906 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1908 gfn_t gfn = gpa >> PAGE_SHIFT;
1910 int offset = offset_in_page(gpa);
1913 while ((seg = next_segment(len, offset)) != 0) {
1914 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1923 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1925 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot,
1928 if (memslot && memslot->dirty_bitmap) {
1929 unsigned long rel_gfn = gfn - memslot->base_gfn;
1931 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1935 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1937 struct kvm_memory_slot *memslot;
1939 memslot = gfn_to_memslot(kvm, gfn);
1940 mark_page_dirty_in_slot(memslot, gfn);
1942 EXPORT_SYMBOL_GPL(mark_page_dirty);
1944 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
1946 struct kvm_memory_slot *memslot;
1948 memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1949 mark_page_dirty_in_slot(memslot, gfn);
1951 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
1953 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
1957 old = val = vcpu->halt_poll_ns;
1959 if (val == 0 && halt_poll_ns_grow)
1962 val *= halt_poll_ns_grow;
1964 if (val > halt_poll_ns)
1967 vcpu->halt_poll_ns = val;
1968 trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
1971 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
1975 old = val = vcpu->halt_poll_ns;
1976 if (halt_poll_ns_shrink == 0)
1979 val /= halt_poll_ns_shrink;
1981 vcpu->halt_poll_ns = val;
1982 trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
1985 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
1987 if (kvm_arch_vcpu_runnable(vcpu)) {
1988 kvm_make_request(KVM_REQ_UNHALT, vcpu);
1991 if (kvm_cpu_has_pending_timer(vcpu))
1993 if (signal_pending(current))
2000 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2002 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
2006 bool waited = false;
2009 start = cur = ktime_get();
2010 if (vcpu->halt_poll_ns) {
2011 ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns);
2013 ++vcpu->stat.halt_attempted_poll;
2016 * This sets KVM_REQ_UNHALT if an interrupt
2019 if (kvm_vcpu_check_block(vcpu) < 0) {
2020 ++vcpu->stat.halt_successful_poll;
2024 } while (single_task_running() && ktime_before(cur, stop));
2027 kvm_arch_vcpu_blocking(vcpu);
2030 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
2032 if (kvm_vcpu_check_block(vcpu) < 0)
2039 finish_wait(&vcpu->wq, &wait);
2042 kvm_arch_vcpu_unblocking(vcpu);
2044 block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
2047 if (block_ns <= vcpu->halt_poll_ns)
2049 /* we had a long block, shrink polling */
2050 else if (vcpu->halt_poll_ns && block_ns > halt_poll_ns)
2051 shrink_halt_poll_ns(vcpu);
2052 /* we had a short halt and our poll time is too small */
2053 else if (vcpu->halt_poll_ns < halt_poll_ns &&
2054 block_ns < halt_poll_ns)
2055 grow_halt_poll_ns(vcpu);
2057 vcpu->halt_poll_ns = 0;
2059 trace_kvm_vcpu_wakeup(block_ns, waited);
2061 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
2065 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2067 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
2070 int cpu = vcpu->cpu;
2071 wait_queue_head_t *wqp;
2073 wqp = kvm_arch_vcpu_wq(vcpu);
2074 if (waitqueue_active(wqp)) {
2075 wake_up_interruptible(wqp);
2076 ++vcpu->stat.halt_wakeup;
2080 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
2081 if (kvm_arch_vcpu_should_kick(vcpu))
2082 smp_send_reschedule(cpu);
2085 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
2086 #endif /* !CONFIG_S390 */
2088 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
2091 struct task_struct *task = NULL;
2095 pid = rcu_dereference(target->pid);
2097 task = get_pid_task(pid, PIDTYPE_PID);
2101 ret = yield_to(task, 1);
2102 put_task_struct(task);
2106 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
2109 * Helper that checks whether a VCPU is eligible for directed yield.
2110 * Most eligible candidate to yield is decided by following heuristics:
2112 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2113 * (preempted lock holder), indicated by @in_spin_loop.
2114 * Set at the beiginning and cleared at the end of interception/PLE handler.
2116 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2117 * chance last time (mostly it has become eligible now since we have probably
2118 * yielded to lockholder in last iteration. This is done by toggling
2119 * @dy_eligible each time a VCPU checked for eligibility.)
2121 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2122 * to preempted lock-holder could result in wrong VCPU selection and CPU
2123 * burning. Giving priority for a potential lock-holder increases lock
2126 * Since algorithm is based on heuristics, accessing another VCPU data without
2127 * locking does not harm. It may result in trying to yield to same VCPU, fail
2128 * and continue with next VCPU and so on.
2130 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
2132 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2135 eligible = !vcpu->spin_loop.in_spin_loop ||
2136 vcpu->spin_loop.dy_eligible;
2138 if (vcpu->spin_loop.in_spin_loop)
2139 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
2147 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
2149 struct kvm *kvm = me->kvm;
2150 struct kvm_vcpu *vcpu;
2151 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
2157 kvm_vcpu_set_in_spin_loop(me, true);
2159 * We boost the priority of a VCPU that is runnable but not
2160 * currently running, because it got preempted by something
2161 * else and called schedule in __vcpu_run. Hopefully that
2162 * VCPU is holding the lock that we need and will release it.
2163 * We approximate round-robin by starting at the last boosted VCPU.
2165 for (pass = 0; pass < 2 && !yielded && try; pass++) {
2166 kvm_for_each_vcpu(i, vcpu, kvm) {
2167 if (!pass && i <= last_boosted_vcpu) {
2168 i = last_boosted_vcpu;
2170 } else if (pass && i > last_boosted_vcpu)
2172 if (!ACCESS_ONCE(vcpu->preempted))
2176 if (waitqueue_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
2178 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
2181 yielded = kvm_vcpu_yield_to(vcpu);
2183 kvm->last_boosted_vcpu = i;
2185 } else if (yielded < 0) {
2192 kvm_vcpu_set_in_spin_loop(me, false);
2194 /* Ensure vcpu is not eligible during next spinloop */
2195 kvm_vcpu_set_dy_eligible(me, false);
2197 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
2199 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2201 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
2204 if (vmf->pgoff == 0)
2205 page = virt_to_page(vcpu->run);
2207 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
2208 page = virt_to_page(vcpu->arch.pio_data);
2210 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2211 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
2212 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
2215 return kvm_arch_vcpu_fault(vcpu, vmf);
2221 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
2222 .fault = kvm_vcpu_fault,
2225 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2227 vma->vm_ops = &kvm_vcpu_vm_ops;
2231 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2233 struct kvm_vcpu *vcpu = filp->private_data;
2235 kvm_put_kvm(vcpu->kvm);
2239 static struct file_operations kvm_vcpu_fops = {
2240 .release = kvm_vcpu_release,
2241 .unlocked_ioctl = kvm_vcpu_ioctl,
2242 #ifdef CONFIG_KVM_COMPAT
2243 .compat_ioctl = kvm_vcpu_compat_ioctl,
2245 .mmap = kvm_vcpu_mmap,
2246 .llseek = noop_llseek,
2250 * Allocates an inode for the vcpu.
2252 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2254 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
2258 * Creates some virtual cpus. Good luck creating more than one.
2260 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
2263 struct kvm_vcpu *vcpu, *v;
2265 if (id >= KVM_MAX_VCPUS)
2268 vcpu = kvm_arch_vcpu_create(kvm, id);
2270 return PTR_ERR(vcpu);
2272 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2274 r = kvm_arch_vcpu_setup(vcpu);
2278 mutex_lock(&kvm->lock);
2279 if (!kvm_vcpu_compatible(vcpu)) {
2281 goto unlock_vcpu_destroy;
2283 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
2285 goto unlock_vcpu_destroy;
2288 kvm_for_each_vcpu(r, v, kvm)
2289 if (v->vcpu_id == id) {
2291 goto unlock_vcpu_destroy;
2294 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
2296 /* Now it's all set up, let userspace reach it */
2298 r = create_vcpu_fd(vcpu);
2301 goto unlock_vcpu_destroy;
2304 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
2307 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
2308 * before kvm->online_vcpu's incremented value.
2311 atomic_inc(&kvm->online_vcpus);
2313 mutex_unlock(&kvm->lock);
2314 kvm_arch_vcpu_postcreate(vcpu);
2317 unlock_vcpu_destroy:
2318 mutex_unlock(&kvm->lock);
2320 kvm_arch_vcpu_destroy(vcpu);
2324 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2327 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2328 vcpu->sigset_active = 1;
2329 vcpu->sigset = *sigset;
2331 vcpu->sigset_active = 0;
2335 static long kvm_vcpu_ioctl(struct file *filp,
2336 unsigned int ioctl, unsigned long arg)
2338 struct kvm_vcpu *vcpu = filp->private_data;
2339 void __user *argp = (void __user *)arg;
2341 struct kvm_fpu *fpu = NULL;
2342 struct kvm_sregs *kvm_sregs = NULL;
2344 if (vcpu->kvm->mm != current->mm)
2347 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2350 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
2352 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
2353 * so vcpu_load() would break it.
2355 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_S390_IRQ || ioctl == KVM_INTERRUPT)
2356 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2360 r = vcpu_load(vcpu);
2368 if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
2369 /* The thread running this VCPU changed. */
2370 struct pid *oldpid = vcpu->pid;
2371 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
2373 rcu_assign_pointer(vcpu->pid, newpid);
2378 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2379 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2381 case KVM_GET_REGS: {
2382 struct kvm_regs *kvm_regs;
2385 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2388 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2392 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2399 case KVM_SET_REGS: {
2400 struct kvm_regs *kvm_regs;
2403 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2404 if (IS_ERR(kvm_regs)) {
2405 r = PTR_ERR(kvm_regs);
2408 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2412 case KVM_GET_SREGS: {
2413 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2417 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2421 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2426 case KVM_SET_SREGS: {
2427 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2428 if (IS_ERR(kvm_sregs)) {
2429 r = PTR_ERR(kvm_sregs);
2433 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2436 case KVM_GET_MP_STATE: {
2437 struct kvm_mp_state mp_state;
2439 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2443 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
2448 case KVM_SET_MP_STATE: {
2449 struct kvm_mp_state mp_state;
2452 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
2454 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2457 case KVM_TRANSLATE: {
2458 struct kvm_translation tr;
2461 if (copy_from_user(&tr, argp, sizeof(tr)))
2463 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2467 if (copy_to_user(argp, &tr, sizeof(tr)))
2472 case KVM_SET_GUEST_DEBUG: {
2473 struct kvm_guest_debug dbg;
2476 if (copy_from_user(&dbg, argp, sizeof(dbg)))
2478 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2481 case KVM_SET_SIGNAL_MASK: {
2482 struct kvm_signal_mask __user *sigmask_arg = argp;
2483 struct kvm_signal_mask kvm_sigmask;
2484 sigset_t sigset, *p;
2489 if (copy_from_user(&kvm_sigmask, argp,
2490 sizeof(kvm_sigmask)))
2493 if (kvm_sigmask.len != sizeof(sigset))
2496 if (copy_from_user(&sigset, sigmask_arg->sigset,
2501 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2505 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2509 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2513 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2519 fpu = memdup_user(argp, sizeof(*fpu));
2525 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2529 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2538 #ifdef CONFIG_KVM_COMPAT
2539 static long kvm_vcpu_compat_ioctl(struct file *filp,
2540 unsigned int ioctl, unsigned long arg)
2542 struct kvm_vcpu *vcpu = filp->private_data;
2543 void __user *argp = compat_ptr(arg);
2546 if (vcpu->kvm->mm != current->mm)
2550 case KVM_SET_SIGNAL_MASK: {
2551 struct kvm_signal_mask __user *sigmask_arg = argp;
2552 struct kvm_signal_mask kvm_sigmask;
2553 compat_sigset_t csigset;
2558 if (copy_from_user(&kvm_sigmask, argp,
2559 sizeof(kvm_sigmask)))
2562 if (kvm_sigmask.len != sizeof(csigset))
2565 if (copy_from_user(&csigset, sigmask_arg->sigset,
2568 sigset_from_compat(&sigset, &csigset);
2569 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2571 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2575 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2583 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2584 int (*accessor)(struct kvm_device *dev,
2585 struct kvm_device_attr *attr),
2588 struct kvm_device_attr attr;
2593 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2596 return accessor(dev, &attr);
2599 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2602 struct kvm_device *dev = filp->private_data;
2605 case KVM_SET_DEVICE_ATTR:
2606 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2607 case KVM_GET_DEVICE_ATTR:
2608 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2609 case KVM_HAS_DEVICE_ATTR:
2610 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2612 if (dev->ops->ioctl)
2613 return dev->ops->ioctl(dev, ioctl, arg);
2619 static int kvm_device_release(struct inode *inode, struct file *filp)
2621 struct kvm_device *dev = filp->private_data;
2622 struct kvm *kvm = dev->kvm;
2628 static const struct file_operations kvm_device_fops = {
2629 .unlocked_ioctl = kvm_device_ioctl,
2630 #ifdef CONFIG_KVM_COMPAT
2631 .compat_ioctl = kvm_device_ioctl,
2633 .release = kvm_device_release,
2636 struct kvm_device *kvm_device_from_filp(struct file *filp)
2638 if (filp->f_op != &kvm_device_fops)
2641 return filp->private_data;
2644 static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
2645 #ifdef CONFIG_KVM_MPIC
2646 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
2647 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
2650 #ifdef CONFIG_KVM_XICS
2651 [KVM_DEV_TYPE_XICS] = &kvm_xics_ops,
2655 int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
2657 if (type >= ARRAY_SIZE(kvm_device_ops_table))
2660 if (kvm_device_ops_table[type] != NULL)
2663 kvm_device_ops_table[type] = ops;
2667 void kvm_unregister_device_ops(u32 type)
2669 if (kvm_device_ops_table[type] != NULL)
2670 kvm_device_ops_table[type] = NULL;
2673 static int kvm_ioctl_create_device(struct kvm *kvm,
2674 struct kvm_create_device *cd)
2676 struct kvm_device_ops *ops = NULL;
2677 struct kvm_device *dev;
2678 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2681 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
2684 ops = kvm_device_ops_table[cd->type];
2691 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2698 ret = ops->create(dev, cd->type);
2704 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2710 list_add(&dev->vm_node, &kvm->devices);
2716 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2719 case KVM_CAP_USER_MEMORY:
2720 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2721 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2722 case KVM_CAP_INTERNAL_ERROR_DATA:
2723 #ifdef CONFIG_HAVE_KVM_MSI
2724 case KVM_CAP_SIGNAL_MSI:
2726 #ifdef CONFIG_HAVE_KVM_IRQFD
2728 case KVM_CAP_IRQFD_RESAMPLE:
2730 case KVM_CAP_IOEVENTFD_ANY_LENGTH:
2731 case KVM_CAP_CHECK_EXTENSION_VM:
2733 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2734 case KVM_CAP_IRQ_ROUTING:
2735 return KVM_MAX_IRQ_ROUTES;
2737 #if KVM_ADDRESS_SPACE_NUM > 1
2738 case KVM_CAP_MULTI_ADDRESS_SPACE:
2739 return KVM_ADDRESS_SPACE_NUM;
2744 return kvm_vm_ioctl_check_extension(kvm, arg);
2747 static long kvm_vm_ioctl(struct file *filp,
2748 unsigned int ioctl, unsigned long arg)
2750 struct kvm *kvm = filp->private_data;
2751 void __user *argp = (void __user *)arg;
2754 if (kvm->mm != current->mm)
2757 case KVM_CREATE_VCPU:
2758 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2760 case KVM_SET_USER_MEMORY_REGION: {
2761 struct kvm_userspace_memory_region kvm_userspace_mem;
2764 if (copy_from_user(&kvm_userspace_mem, argp,
2765 sizeof(kvm_userspace_mem)))
2768 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2771 case KVM_GET_DIRTY_LOG: {
2772 struct kvm_dirty_log log;
2775 if (copy_from_user(&log, argp, sizeof(log)))
2777 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2780 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2781 case KVM_REGISTER_COALESCED_MMIO: {
2782 struct kvm_coalesced_mmio_zone zone;
2785 if (copy_from_user(&zone, argp, sizeof(zone)))
2787 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2790 case KVM_UNREGISTER_COALESCED_MMIO: {
2791 struct kvm_coalesced_mmio_zone zone;
2794 if (copy_from_user(&zone, argp, sizeof(zone)))
2796 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2801 struct kvm_irqfd data;
2804 if (copy_from_user(&data, argp, sizeof(data)))
2806 r = kvm_irqfd(kvm, &data);
2809 case KVM_IOEVENTFD: {
2810 struct kvm_ioeventfd data;
2813 if (copy_from_user(&data, argp, sizeof(data)))
2815 r = kvm_ioeventfd(kvm, &data);
2818 #ifdef CONFIG_HAVE_KVM_MSI
2819 case KVM_SIGNAL_MSI: {
2823 if (copy_from_user(&msi, argp, sizeof(msi)))
2825 r = kvm_send_userspace_msi(kvm, &msi);
2829 #ifdef __KVM_HAVE_IRQ_LINE
2830 case KVM_IRQ_LINE_STATUS:
2831 case KVM_IRQ_LINE: {
2832 struct kvm_irq_level irq_event;
2835 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
2838 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
2839 ioctl == KVM_IRQ_LINE_STATUS);
2844 if (ioctl == KVM_IRQ_LINE_STATUS) {
2845 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
2853 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2854 case KVM_SET_GSI_ROUTING: {
2855 struct kvm_irq_routing routing;
2856 struct kvm_irq_routing __user *urouting;
2857 struct kvm_irq_routing_entry *entries;
2860 if (copy_from_user(&routing, argp, sizeof(routing)))
2863 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2868 entries = vmalloc(routing.nr * sizeof(*entries));
2873 if (copy_from_user(entries, urouting->entries,
2874 routing.nr * sizeof(*entries)))
2875 goto out_free_irq_routing;
2876 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2878 out_free_irq_routing:
2882 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
2883 case KVM_CREATE_DEVICE: {
2884 struct kvm_create_device cd;
2887 if (copy_from_user(&cd, argp, sizeof(cd)))
2890 r = kvm_ioctl_create_device(kvm, &cd);
2895 if (copy_to_user(argp, &cd, sizeof(cd)))
2901 case KVM_CHECK_EXTENSION:
2902 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
2905 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2911 #ifdef CONFIG_KVM_COMPAT
2912 struct compat_kvm_dirty_log {
2916 compat_uptr_t dirty_bitmap; /* one bit per page */
2921 static long kvm_vm_compat_ioctl(struct file *filp,
2922 unsigned int ioctl, unsigned long arg)
2924 struct kvm *kvm = filp->private_data;
2927 if (kvm->mm != current->mm)
2930 case KVM_GET_DIRTY_LOG: {
2931 struct compat_kvm_dirty_log compat_log;
2932 struct kvm_dirty_log log;
2935 if (copy_from_user(&compat_log, (void __user *)arg,
2936 sizeof(compat_log)))
2938 log.slot = compat_log.slot;
2939 log.padding1 = compat_log.padding1;
2940 log.padding2 = compat_log.padding2;
2941 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2943 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2947 r = kvm_vm_ioctl(filp, ioctl, arg);
2955 static struct file_operations kvm_vm_fops = {
2956 .release = kvm_vm_release,
2957 .unlocked_ioctl = kvm_vm_ioctl,
2958 #ifdef CONFIG_KVM_COMPAT
2959 .compat_ioctl = kvm_vm_compat_ioctl,
2961 .llseek = noop_llseek,
2964 static int kvm_dev_ioctl_create_vm(unsigned long type)
2969 kvm = kvm_create_vm(type);
2971 return PTR_ERR(kvm);
2972 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2973 r = kvm_coalesced_mmio_init(kvm);
2979 r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC);
2986 static long kvm_dev_ioctl(struct file *filp,
2987 unsigned int ioctl, unsigned long arg)
2992 case KVM_GET_API_VERSION:
2995 r = KVM_API_VERSION;
2998 r = kvm_dev_ioctl_create_vm(arg);
3000 case KVM_CHECK_EXTENSION:
3001 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
3003 case KVM_GET_VCPU_MMAP_SIZE:
3006 r = PAGE_SIZE; /* struct kvm_run */
3008 r += PAGE_SIZE; /* pio data page */
3010 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
3011 r += PAGE_SIZE; /* coalesced mmio ring page */
3014 case KVM_TRACE_ENABLE:
3015 case KVM_TRACE_PAUSE:
3016 case KVM_TRACE_DISABLE:
3020 return kvm_arch_dev_ioctl(filp, ioctl, arg);
3026 static struct file_operations kvm_chardev_ops = {
3027 .unlocked_ioctl = kvm_dev_ioctl,
3028 .compat_ioctl = kvm_dev_ioctl,
3029 .llseek = noop_llseek,
3032 static struct miscdevice kvm_dev = {
3038 static void hardware_enable_nolock(void *junk)
3040 int cpu = raw_smp_processor_id();
3043 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
3046 cpumask_set_cpu(cpu, cpus_hardware_enabled);
3048 r = kvm_arch_hardware_enable();
3051 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3052 atomic_inc(&hardware_enable_failed);
3053 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
3057 static void hardware_enable(void)
3059 raw_spin_lock(&kvm_count_lock);
3060 if (kvm_usage_count)
3061 hardware_enable_nolock(NULL);
3062 raw_spin_unlock(&kvm_count_lock);
3065 static void hardware_disable_nolock(void *junk)
3067 int cpu = raw_smp_processor_id();
3069 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
3071 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3072 kvm_arch_hardware_disable();
3075 static void hardware_disable(void)
3077 raw_spin_lock(&kvm_count_lock);
3078 if (kvm_usage_count)
3079 hardware_disable_nolock(NULL);
3080 raw_spin_unlock(&kvm_count_lock);
3083 static void hardware_disable_all_nolock(void)
3085 BUG_ON(!kvm_usage_count);
3088 if (!kvm_usage_count)
3089 on_each_cpu(hardware_disable_nolock, NULL, 1);
3092 static void hardware_disable_all(void)
3094 raw_spin_lock(&kvm_count_lock);
3095 hardware_disable_all_nolock();
3096 raw_spin_unlock(&kvm_count_lock);
3099 static int hardware_enable_all(void)
3103 raw_spin_lock(&kvm_count_lock);
3106 if (kvm_usage_count == 1) {
3107 atomic_set(&hardware_enable_failed, 0);
3108 on_each_cpu(hardware_enable_nolock, NULL, 1);
3110 if (atomic_read(&hardware_enable_failed)) {
3111 hardware_disable_all_nolock();
3116 raw_spin_unlock(&kvm_count_lock);
3121 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
3124 val &= ~CPU_TASKS_FROZEN;
3136 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3140 * Some (well, at least mine) BIOSes hang on reboot if
3143 * And Intel TXT required VMX off for all cpu when system shutdown.
3145 pr_info("kvm: exiting hardware virtualization\n");
3146 kvm_rebooting = true;
3147 on_each_cpu(hardware_disable_nolock, NULL, 1);
3151 static struct notifier_block kvm_reboot_notifier = {
3152 .notifier_call = kvm_reboot,
3156 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3160 for (i = 0; i < bus->dev_count; i++) {
3161 struct kvm_io_device *pos = bus->range[i].dev;
3163 kvm_iodevice_destructor(pos);
3168 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
3169 const struct kvm_io_range *r2)
3171 gpa_t addr1 = r1->addr;
3172 gpa_t addr2 = r2->addr;
3177 /* If r2->len == 0, match the exact address. If r2->len != 0,
3178 * accept any overlapping write. Any order is acceptable for
3179 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
3180 * we process all of them.
3193 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
3195 return kvm_io_bus_cmp(p1, p2);
3198 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
3199 gpa_t addr, int len)
3201 bus->range[bus->dev_count++] = (struct kvm_io_range) {
3207 sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
3208 kvm_io_bus_sort_cmp, NULL);
3213 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
3214 gpa_t addr, int len)
3216 struct kvm_io_range *range, key;
3219 key = (struct kvm_io_range) {
3224 range = bsearch(&key, bus->range, bus->dev_count,
3225 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
3229 off = range - bus->range;
3231 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
3237 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3238 struct kvm_io_range *range, const void *val)
3242 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3246 while (idx < bus->dev_count &&
3247 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3248 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
3257 /* kvm_io_bus_write - called under kvm->slots_lock */
3258 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3259 int len, const void *val)
3261 struct kvm_io_bus *bus;
3262 struct kvm_io_range range;
3265 range = (struct kvm_io_range) {
3270 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3271 r = __kvm_io_bus_write(vcpu, bus, &range, val);
3272 return r < 0 ? r : 0;
3275 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3276 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
3277 gpa_t addr, int len, const void *val, long cookie)
3279 struct kvm_io_bus *bus;
3280 struct kvm_io_range range;
3282 range = (struct kvm_io_range) {
3287 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3289 /* First try the device referenced by cookie. */
3290 if ((cookie >= 0) && (cookie < bus->dev_count) &&
3291 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
3292 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
3297 * cookie contained garbage; fall back to search and return the
3298 * correct cookie value.
3300 return __kvm_io_bus_write(vcpu, bus, &range, val);
3303 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3304 struct kvm_io_range *range, void *val)
3308 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3312 while (idx < bus->dev_count &&
3313 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3314 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
3322 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
3324 /* kvm_io_bus_read - called under kvm->slots_lock */
3325 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3328 struct kvm_io_bus *bus;
3329 struct kvm_io_range range;
3332 range = (struct kvm_io_range) {
3337 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3338 r = __kvm_io_bus_read(vcpu, bus, &range, val);
3339 return r < 0 ? r : 0;
3343 /* Caller must hold slots_lock. */
3344 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3345 int len, struct kvm_io_device *dev)
3347 struct kvm_io_bus *new_bus, *bus;
3349 bus = kvm->buses[bus_idx];
3350 /* exclude ioeventfd which is limited by maximum fd */
3351 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3354 new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count + 1) *
3355 sizeof(struct kvm_io_range)), GFP_KERNEL);
3358 memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
3359 sizeof(struct kvm_io_range)));
3360 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
3361 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3362 synchronize_srcu_expedited(&kvm->srcu);
3368 /* Caller must hold slots_lock. */
3369 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3370 struct kvm_io_device *dev)
3373 struct kvm_io_bus *new_bus, *bus;
3375 bus = kvm->buses[bus_idx];
3377 for (i = 0; i < bus->dev_count; i++)
3378 if (bus->range[i].dev == dev) {
3386 new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3387 sizeof(struct kvm_io_range)), GFP_KERNEL);
3391 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3392 new_bus->dev_count--;
3393 memcpy(new_bus->range + i, bus->range + i + 1,
3394 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3396 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3397 synchronize_srcu_expedited(&kvm->srcu);
3402 static struct notifier_block kvm_cpu_notifier = {
3403 .notifier_call = kvm_cpu_hotplug,
3406 static int vm_stat_get(void *_offset, u64 *val)
3408 unsigned offset = (long)_offset;
3412 spin_lock(&kvm_lock);
3413 list_for_each_entry(kvm, &vm_list, vm_list)
3414 *val += *(u32 *)((void *)kvm + offset);
3415 spin_unlock(&kvm_lock);
3419 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
3421 static int vcpu_stat_get(void *_offset, u64 *val)
3423 unsigned offset = (long)_offset;
3425 struct kvm_vcpu *vcpu;
3429 spin_lock(&kvm_lock);
3430 list_for_each_entry(kvm, &vm_list, vm_list)
3431 kvm_for_each_vcpu(i, vcpu, kvm)
3432 *val += *(u32 *)((void *)vcpu + offset);
3434 spin_unlock(&kvm_lock);
3438 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
3440 static const struct file_operations *stat_fops[] = {
3441 [KVM_STAT_VCPU] = &vcpu_stat_fops,
3442 [KVM_STAT_VM] = &vm_stat_fops,
3445 static int kvm_init_debug(void)
3448 struct kvm_stats_debugfs_item *p;
3450 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3451 if (kvm_debugfs_dir == NULL)
3454 for (p = debugfs_entries; p->name; ++p) {
3455 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
3456 (void *)(long)p->offset,
3457 stat_fops[p->kind]);
3458 if (p->dentry == NULL)
3465 debugfs_remove_recursive(kvm_debugfs_dir);
3470 static void kvm_exit_debug(void)
3472 struct kvm_stats_debugfs_item *p;
3474 for (p = debugfs_entries; p->name; ++p)
3475 debugfs_remove(p->dentry);
3476 debugfs_remove(kvm_debugfs_dir);
3479 static int kvm_suspend(void)
3481 if (kvm_usage_count)
3482 hardware_disable_nolock(NULL);
3486 static void kvm_resume(void)
3488 if (kvm_usage_count) {
3489 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3490 hardware_enable_nolock(NULL);
3494 static struct syscore_ops kvm_syscore_ops = {
3495 .suspend = kvm_suspend,
3496 .resume = kvm_resume,
3500 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3502 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3505 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3507 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3509 if (vcpu->preempted)
3510 vcpu->preempted = false;
3512 kvm_arch_sched_in(vcpu, cpu);
3514 kvm_arch_vcpu_load(vcpu, cpu);
3517 static void kvm_sched_out(struct preempt_notifier *pn,
3518 struct task_struct *next)
3520 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3522 if (current->state == TASK_RUNNING)
3523 vcpu->preempted = true;
3524 kvm_arch_vcpu_put(vcpu);
3527 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3528 struct module *module)
3533 r = kvm_arch_init(opaque);
3538 * kvm_arch_init makes sure there's at most one caller
3539 * for architectures that support multiple implementations,
3540 * like intel and amd on x86.
3541 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3542 * conflicts in case kvm is already setup for another implementation.
3544 r = kvm_irqfd_init();
3548 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
3553 r = kvm_arch_hardware_setup();
3557 for_each_online_cpu(cpu) {
3558 smp_call_function_single(cpu,
3559 kvm_arch_check_processor_compat,
3565 r = register_cpu_notifier(&kvm_cpu_notifier);
3568 register_reboot_notifier(&kvm_reboot_notifier);
3570 /* A kmem cache lets us meet the alignment requirements of fx_save. */
3572 vcpu_align = __alignof__(struct kvm_vcpu);
3573 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
3575 if (!kvm_vcpu_cache) {
3580 r = kvm_async_pf_init();
3584 kvm_chardev_ops.owner = module;
3585 kvm_vm_fops.owner = module;
3586 kvm_vcpu_fops.owner = module;
3588 r = misc_register(&kvm_dev);
3590 pr_err("kvm: misc device register failed\n");
3594 register_syscore_ops(&kvm_syscore_ops);
3596 kvm_preempt_ops.sched_in = kvm_sched_in;
3597 kvm_preempt_ops.sched_out = kvm_sched_out;
3599 r = kvm_init_debug();
3601 pr_err("kvm: create debugfs files failed\n");
3605 r = kvm_vfio_ops_init();
3611 unregister_syscore_ops(&kvm_syscore_ops);
3612 misc_deregister(&kvm_dev);
3614 kvm_async_pf_deinit();
3616 kmem_cache_destroy(kvm_vcpu_cache);
3618 unregister_reboot_notifier(&kvm_reboot_notifier);
3619 unregister_cpu_notifier(&kvm_cpu_notifier);
3622 kvm_arch_hardware_unsetup();
3624 free_cpumask_var(cpus_hardware_enabled);
3632 EXPORT_SYMBOL_GPL(kvm_init);
3637 misc_deregister(&kvm_dev);
3638 kmem_cache_destroy(kvm_vcpu_cache);
3639 kvm_async_pf_deinit();
3640 unregister_syscore_ops(&kvm_syscore_ops);
3641 unregister_reboot_notifier(&kvm_reboot_notifier);
3642 unregister_cpu_notifier(&kvm_cpu_notifier);
3643 on_each_cpu(hardware_disable_nolock, NULL, 1);
3644 kvm_arch_hardware_unsetup();
3647 free_cpumask_var(cpus_hardware_enabled);
3648 kvm_vfio_ops_exit();
3650 EXPORT_SYMBOL_GPL(kvm_exit);