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 /* halt polling only reduces halt latency by 5-7 us, 500us is enough */
70 static unsigned int halt_poll_ns = 500000;
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);
233 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
238 vcpu->run = page_address(page);
240 kvm_vcpu_set_in_spin_loop(vcpu, false);
241 kvm_vcpu_set_dy_eligible(vcpu, false);
242 vcpu->preempted = false;
244 r = kvm_arch_vcpu_init(vcpu);
250 free_page((unsigned long)vcpu->run);
254 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
256 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
259 kvm_arch_vcpu_uninit(vcpu);
260 free_page((unsigned long)vcpu->run);
262 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
264 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
265 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
267 return container_of(mn, struct kvm, mmu_notifier);
270 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
271 struct mm_struct *mm,
272 unsigned long address)
274 struct kvm *kvm = mmu_notifier_to_kvm(mn);
275 int need_tlb_flush, idx;
278 * When ->invalidate_page runs, the linux pte has been zapped
279 * already but the page is still allocated until
280 * ->invalidate_page returns. So if we increase the sequence
281 * here the kvm page fault will notice if the spte can't be
282 * established because the page is going to be freed. If
283 * instead the kvm page fault establishes the spte before
284 * ->invalidate_page runs, kvm_unmap_hva will release it
287 * The sequence increase only need to be seen at spin_unlock
288 * time, and not at spin_lock time.
290 * Increasing the sequence after the spin_unlock would be
291 * unsafe because the kvm page fault could then establish the
292 * pte after kvm_unmap_hva returned, without noticing the page
293 * is going to be freed.
295 idx = srcu_read_lock(&kvm->srcu);
296 spin_lock(&kvm->mmu_lock);
298 kvm->mmu_notifier_seq++;
299 need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
300 /* we've to flush the tlb before the pages can be freed */
302 kvm_flush_remote_tlbs(kvm);
304 spin_unlock(&kvm->mmu_lock);
306 kvm_arch_mmu_notifier_invalidate_page(kvm, address);
308 srcu_read_unlock(&kvm->srcu, idx);
311 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
312 struct mm_struct *mm,
313 unsigned long address,
316 struct kvm *kvm = mmu_notifier_to_kvm(mn);
319 idx = srcu_read_lock(&kvm->srcu);
320 spin_lock(&kvm->mmu_lock);
321 kvm->mmu_notifier_seq++;
322 kvm_set_spte_hva(kvm, address, pte);
323 spin_unlock(&kvm->mmu_lock);
324 srcu_read_unlock(&kvm->srcu, idx);
327 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
328 struct mm_struct *mm,
332 struct kvm *kvm = mmu_notifier_to_kvm(mn);
333 int need_tlb_flush = 0, idx;
335 idx = srcu_read_lock(&kvm->srcu);
336 spin_lock(&kvm->mmu_lock);
338 * The count increase must become visible at unlock time as no
339 * spte can be established without taking the mmu_lock and
340 * count is also read inside the mmu_lock critical section.
342 kvm->mmu_notifier_count++;
343 need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
344 need_tlb_flush |= kvm->tlbs_dirty;
345 /* we've to flush the tlb before the pages can be freed */
347 kvm_flush_remote_tlbs(kvm);
349 spin_unlock(&kvm->mmu_lock);
350 srcu_read_unlock(&kvm->srcu, idx);
353 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
354 struct mm_struct *mm,
358 struct kvm *kvm = mmu_notifier_to_kvm(mn);
360 spin_lock(&kvm->mmu_lock);
362 * This sequence increase will notify the kvm page fault that
363 * the page that is going to be mapped in the spte could have
366 kvm->mmu_notifier_seq++;
369 * The above sequence increase must be visible before the
370 * below count decrease, which is ensured by the smp_wmb above
371 * in conjunction with the smp_rmb in mmu_notifier_retry().
373 kvm->mmu_notifier_count--;
374 spin_unlock(&kvm->mmu_lock);
376 BUG_ON(kvm->mmu_notifier_count < 0);
379 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
380 struct mm_struct *mm,
384 struct kvm *kvm = mmu_notifier_to_kvm(mn);
387 idx = srcu_read_lock(&kvm->srcu);
388 spin_lock(&kvm->mmu_lock);
390 young = kvm_age_hva(kvm, start, end);
392 kvm_flush_remote_tlbs(kvm);
394 spin_unlock(&kvm->mmu_lock);
395 srcu_read_unlock(&kvm->srcu, idx);
400 static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
401 struct mm_struct *mm,
405 struct kvm *kvm = mmu_notifier_to_kvm(mn);
408 idx = srcu_read_lock(&kvm->srcu);
409 spin_lock(&kvm->mmu_lock);
411 * Even though we do not flush TLB, this will still adversely
412 * affect performance on pre-Haswell Intel EPT, where there is
413 * no EPT Access Bit to clear so that we have to tear down EPT
414 * tables instead. If we find this unacceptable, we can always
415 * add a parameter to kvm_age_hva so that it effectively doesn't
416 * do anything on clear_young.
418 * Also note that currently we never issue secondary TLB flushes
419 * from clear_young, leaving this job up to the regular system
420 * cadence. If we find this inaccurate, we might come up with a
421 * more sophisticated heuristic later.
423 young = kvm_age_hva(kvm, start, end);
424 spin_unlock(&kvm->mmu_lock);
425 srcu_read_unlock(&kvm->srcu, idx);
430 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
431 struct mm_struct *mm,
432 unsigned long address)
434 struct kvm *kvm = mmu_notifier_to_kvm(mn);
437 idx = srcu_read_lock(&kvm->srcu);
438 spin_lock(&kvm->mmu_lock);
439 young = kvm_test_age_hva(kvm, address);
440 spin_unlock(&kvm->mmu_lock);
441 srcu_read_unlock(&kvm->srcu, idx);
446 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
447 struct mm_struct *mm)
449 struct kvm *kvm = mmu_notifier_to_kvm(mn);
452 idx = srcu_read_lock(&kvm->srcu);
453 kvm_arch_flush_shadow_all(kvm);
454 srcu_read_unlock(&kvm->srcu, idx);
457 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
458 .invalidate_page = kvm_mmu_notifier_invalidate_page,
459 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
460 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
461 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
462 .clear_young = kvm_mmu_notifier_clear_young,
463 .test_young = kvm_mmu_notifier_test_young,
464 .change_pte = kvm_mmu_notifier_change_pte,
465 .release = kvm_mmu_notifier_release,
468 static int kvm_init_mmu_notifier(struct kvm *kvm)
470 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
471 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
474 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
476 static int kvm_init_mmu_notifier(struct kvm *kvm)
481 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
483 static struct kvm_memslots *kvm_alloc_memslots(void)
486 struct kvm_memslots *slots;
488 slots = kvm_kvzalloc(sizeof(struct kvm_memslots));
493 * Init kvm generation close to the maximum to easily test the
494 * code of handling generation number wrap-around.
496 slots->generation = -150;
497 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
498 slots->id_to_index[i] = slots->memslots[i].id = i;
503 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
505 if (!memslot->dirty_bitmap)
508 kvfree(memslot->dirty_bitmap);
509 memslot->dirty_bitmap = NULL;
513 * Free any memory in @free but not in @dont.
515 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
516 struct kvm_memory_slot *dont)
518 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
519 kvm_destroy_dirty_bitmap(free);
521 kvm_arch_free_memslot(kvm, free, dont);
526 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
528 struct kvm_memory_slot *memslot;
533 kvm_for_each_memslot(memslot, slots)
534 kvm_free_memslot(kvm, memslot, NULL);
539 static struct kvm *kvm_create_vm(unsigned long type)
542 struct kvm *kvm = kvm_arch_alloc_vm();
545 return ERR_PTR(-ENOMEM);
547 r = kvm_arch_init_vm(kvm, type);
549 goto out_err_no_disable;
551 r = hardware_enable_all();
553 goto out_err_no_disable;
555 #ifdef CONFIG_HAVE_KVM_IRQFD
556 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
559 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
562 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
563 kvm->memslots[i] = kvm_alloc_memslots();
564 if (!kvm->memslots[i])
565 goto out_err_no_srcu;
568 if (init_srcu_struct(&kvm->srcu))
569 goto out_err_no_srcu;
570 if (init_srcu_struct(&kvm->irq_srcu))
571 goto out_err_no_irq_srcu;
572 for (i = 0; i < KVM_NR_BUSES; i++) {
573 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
579 spin_lock_init(&kvm->mmu_lock);
580 kvm->mm = current->mm;
581 atomic_inc(&kvm->mm->mm_count);
582 kvm_eventfd_init(kvm);
583 mutex_init(&kvm->lock);
584 mutex_init(&kvm->irq_lock);
585 mutex_init(&kvm->slots_lock);
586 atomic_set(&kvm->users_count, 1);
587 INIT_LIST_HEAD(&kvm->devices);
589 r = kvm_init_mmu_notifier(kvm);
593 spin_lock(&kvm_lock);
594 list_add(&kvm->vm_list, &vm_list);
595 spin_unlock(&kvm_lock);
597 preempt_notifier_inc();
602 cleanup_srcu_struct(&kvm->irq_srcu);
604 cleanup_srcu_struct(&kvm->srcu);
606 hardware_disable_all();
608 for (i = 0; i < KVM_NR_BUSES; i++)
609 kfree(kvm->buses[i]);
610 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
611 kvm_free_memslots(kvm, kvm->memslots[i]);
612 kvm_arch_free_vm(kvm);
617 * Avoid using vmalloc for a small buffer.
618 * Should not be used when the size is statically known.
620 void *kvm_kvzalloc(unsigned long size)
622 if (size > PAGE_SIZE)
623 return vzalloc(size);
625 return kzalloc(size, GFP_KERNEL);
628 static void kvm_destroy_devices(struct kvm *kvm)
630 struct list_head *node, *tmp;
632 list_for_each_safe(node, tmp, &kvm->devices) {
633 struct kvm_device *dev =
634 list_entry(node, struct kvm_device, vm_node);
637 dev->ops->destroy(dev);
641 static void kvm_destroy_vm(struct kvm *kvm)
644 struct mm_struct *mm = kvm->mm;
646 kvm_arch_sync_events(kvm);
647 spin_lock(&kvm_lock);
648 list_del(&kvm->vm_list);
649 spin_unlock(&kvm_lock);
650 kvm_free_irq_routing(kvm);
651 for (i = 0; i < KVM_NR_BUSES; i++)
652 kvm_io_bus_destroy(kvm->buses[i]);
653 kvm_coalesced_mmio_free(kvm);
654 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
655 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
657 kvm_arch_flush_shadow_all(kvm);
659 kvm_arch_destroy_vm(kvm);
660 kvm_destroy_devices(kvm);
661 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
662 kvm_free_memslots(kvm, kvm->memslots[i]);
663 cleanup_srcu_struct(&kvm->irq_srcu);
664 cleanup_srcu_struct(&kvm->srcu);
665 kvm_arch_free_vm(kvm);
666 preempt_notifier_dec();
667 hardware_disable_all();
671 void kvm_get_kvm(struct kvm *kvm)
673 atomic_inc(&kvm->users_count);
675 EXPORT_SYMBOL_GPL(kvm_get_kvm);
677 void kvm_put_kvm(struct kvm *kvm)
679 if (atomic_dec_and_test(&kvm->users_count))
682 EXPORT_SYMBOL_GPL(kvm_put_kvm);
685 static int kvm_vm_release(struct inode *inode, struct file *filp)
687 struct kvm *kvm = filp->private_data;
689 kvm_irqfd_release(kvm);
696 * Allocation size is twice as large as the actual dirty bitmap size.
697 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
699 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
701 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
703 memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
704 if (!memslot->dirty_bitmap)
711 * Insert memslot and re-sort memslots based on their GFN,
712 * so binary search could be used to lookup GFN.
713 * Sorting algorithm takes advantage of having initially
714 * sorted array and known changed memslot position.
716 static void update_memslots(struct kvm_memslots *slots,
717 struct kvm_memory_slot *new)
720 int i = slots->id_to_index[id];
721 struct kvm_memory_slot *mslots = slots->memslots;
723 WARN_ON(mslots[i].id != id);
725 WARN_ON(!mslots[i].npages);
726 if (mslots[i].npages)
729 if (!mslots[i].npages)
733 while (i < KVM_MEM_SLOTS_NUM - 1 &&
734 new->base_gfn <= mslots[i + 1].base_gfn) {
735 if (!mslots[i + 1].npages)
737 mslots[i] = mslots[i + 1];
738 slots->id_to_index[mslots[i].id] = i;
743 * The ">=" is needed when creating a slot with base_gfn == 0,
744 * so that it moves before all those with base_gfn == npages == 0.
746 * On the other hand, if new->npages is zero, the above loop has
747 * already left i pointing to the beginning of the empty part of
748 * mslots, and the ">=" would move the hole backwards in this
749 * case---which is wrong. So skip the loop when deleting a slot.
753 new->base_gfn >= mslots[i - 1].base_gfn) {
754 mslots[i] = mslots[i - 1];
755 slots->id_to_index[mslots[i].id] = i;
759 WARN_ON_ONCE(i != slots->used_slots);
762 slots->id_to_index[mslots[i].id] = i;
765 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
767 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
769 #ifdef __KVM_HAVE_READONLY_MEM
770 valid_flags |= KVM_MEM_READONLY;
773 if (mem->flags & ~valid_flags)
779 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
780 int as_id, struct kvm_memslots *slots)
782 struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id);
785 * Set the low bit in the generation, which disables SPTE caching
786 * until the end of synchronize_srcu_expedited.
788 WARN_ON(old_memslots->generation & 1);
789 slots->generation = old_memslots->generation + 1;
791 rcu_assign_pointer(kvm->memslots[as_id], slots);
792 synchronize_srcu_expedited(&kvm->srcu);
795 * Increment the new memslot generation a second time. This prevents
796 * vm exits that race with memslot updates from caching a memslot
797 * generation that will (potentially) be valid forever.
801 kvm_arch_memslots_updated(kvm, slots);
807 * Allocate some memory and give it an address in the guest physical address
810 * Discontiguous memory is allowed, mostly for framebuffers.
812 * Must be called holding kvm->slots_lock for write.
814 int __kvm_set_memory_region(struct kvm *kvm,
815 const struct kvm_userspace_memory_region *mem)
819 unsigned long npages;
820 struct kvm_memory_slot *slot;
821 struct kvm_memory_slot old, new;
822 struct kvm_memslots *slots = NULL, *old_memslots;
824 enum kvm_mr_change change;
826 r = check_memory_region_flags(mem);
831 as_id = mem->slot >> 16;
834 /* General sanity checks */
835 if (mem->memory_size & (PAGE_SIZE - 1))
837 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
839 /* We can read the guest memory with __xxx_user() later on. */
840 if ((id < KVM_USER_MEM_SLOTS) &&
841 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
842 !access_ok(VERIFY_WRITE,
843 (void __user *)(unsigned long)mem->userspace_addr,
846 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
848 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
851 slot = id_to_memslot(__kvm_memslots(kvm, as_id), id);
852 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
853 npages = mem->memory_size >> PAGE_SHIFT;
855 if (npages > KVM_MEM_MAX_NR_PAGES)
861 new.base_gfn = base_gfn;
863 new.flags = mem->flags;
867 change = KVM_MR_CREATE;
868 else { /* Modify an existing slot. */
869 if ((mem->userspace_addr != old.userspace_addr) ||
870 (npages != old.npages) ||
871 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
874 if (base_gfn != old.base_gfn)
875 change = KVM_MR_MOVE;
876 else if (new.flags != old.flags)
877 change = KVM_MR_FLAGS_ONLY;
878 else { /* Nothing to change. */
887 change = KVM_MR_DELETE;
892 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
893 /* Check for overlaps */
895 kvm_for_each_memslot(slot, __kvm_memslots(kvm, as_id)) {
896 if ((slot->id >= KVM_USER_MEM_SLOTS) ||
899 if (!((base_gfn + npages <= slot->base_gfn) ||
900 (base_gfn >= slot->base_gfn + slot->npages)))
905 /* Free page dirty bitmap if unneeded */
906 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
907 new.dirty_bitmap = NULL;
910 if (change == KVM_MR_CREATE) {
911 new.userspace_addr = mem->userspace_addr;
913 if (kvm_arch_create_memslot(kvm, &new, npages))
917 /* Allocate page dirty bitmap if needed */
918 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
919 if (kvm_create_dirty_bitmap(&new) < 0)
923 slots = kvm_kvzalloc(sizeof(struct kvm_memslots));
926 memcpy(slots, __kvm_memslots(kvm, as_id), sizeof(struct kvm_memslots));
928 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
929 slot = id_to_memslot(slots, id);
930 slot->flags |= KVM_MEMSLOT_INVALID;
932 old_memslots = install_new_memslots(kvm, as_id, slots);
934 /* slot was deleted or moved, clear iommu mapping */
935 kvm_iommu_unmap_pages(kvm, &old);
936 /* From this point no new shadow pages pointing to a deleted,
937 * or moved, memslot will be created.
939 * validation of sp->gfn happens in:
940 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
941 * - kvm_is_visible_gfn (mmu_check_roots)
943 kvm_arch_flush_shadow_memslot(kvm, slot);
946 * We can re-use the old_memslots from above, the only difference
947 * from the currently installed memslots is the invalid flag. This
948 * will get overwritten by update_memslots anyway.
950 slots = old_memslots;
953 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
957 /* actual memory is freed via old in kvm_free_memslot below */
958 if (change == KVM_MR_DELETE) {
959 new.dirty_bitmap = NULL;
960 memset(&new.arch, 0, sizeof(new.arch));
963 update_memslots(slots, &new);
964 old_memslots = install_new_memslots(kvm, as_id, slots);
966 kvm_arch_commit_memory_region(kvm, mem, &old, &new, change);
968 kvm_free_memslot(kvm, &old, &new);
969 kvfree(old_memslots);
972 * IOMMU mapping: New slots need to be mapped. Old slots need to be
973 * un-mapped and re-mapped if their base changes. Since base change
974 * unmapping is handled above with slot deletion, mapping alone is
975 * needed here. Anything else the iommu might care about for existing
976 * slots (size changes, userspace addr changes and read-only flag
977 * changes) is disallowed above, so any other attribute changes getting
978 * here can be skipped.
980 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
981 r = kvm_iommu_map_pages(kvm, &new);
990 kvm_free_memslot(kvm, &new, &old);
994 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
996 int kvm_set_memory_region(struct kvm *kvm,
997 const struct kvm_userspace_memory_region *mem)
1001 mutex_lock(&kvm->slots_lock);
1002 r = __kvm_set_memory_region(kvm, mem);
1003 mutex_unlock(&kvm->slots_lock);
1006 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1008 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1009 struct kvm_userspace_memory_region *mem)
1011 if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1014 return kvm_set_memory_region(kvm, mem);
1017 int kvm_get_dirty_log(struct kvm *kvm,
1018 struct kvm_dirty_log *log, int *is_dirty)
1020 struct kvm_memslots *slots;
1021 struct kvm_memory_slot *memslot;
1022 int r, i, as_id, id;
1024 unsigned long any = 0;
1027 as_id = log->slot >> 16;
1028 id = (u16)log->slot;
1029 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1032 slots = __kvm_memslots(kvm, as_id);
1033 memslot = id_to_memslot(slots, id);
1035 if (!memslot->dirty_bitmap)
1038 n = kvm_dirty_bitmap_bytes(memslot);
1040 for (i = 0; !any && i < n/sizeof(long); ++i)
1041 any = memslot->dirty_bitmap[i];
1044 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1054 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1056 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1058 * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages
1059 * are dirty write protect them for next write.
1060 * @kvm: pointer to kvm instance
1061 * @log: slot id and address to which we copy the log
1062 * @is_dirty: flag set if any page is dirty
1064 * We need to keep it in mind that VCPU threads can write to the bitmap
1065 * concurrently. So, to avoid losing track of dirty pages we keep the
1068 * 1. Take a snapshot of the bit and clear it if needed.
1069 * 2. Write protect the corresponding page.
1070 * 3. Copy the snapshot to the userspace.
1071 * 4. Upon return caller flushes TLB's if needed.
1073 * Between 2 and 4, the guest may write to the page using the remaining TLB
1074 * entry. This is not a problem because the page is reported dirty using
1075 * the snapshot taken before and step 4 ensures that writes done after
1076 * exiting to userspace will be logged for the next call.
1079 int kvm_get_dirty_log_protect(struct kvm *kvm,
1080 struct kvm_dirty_log *log, bool *is_dirty)
1082 struct kvm_memslots *slots;
1083 struct kvm_memory_slot *memslot;
1084 int r, i, as_id, id;
1086 unsigned long *dirty_bitmap;
1087 unsigned long *dirty_bitmap_buffer;
1090 as_id = log->slot >> 16;
1091 id = (u16)log->slot;
1092 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1095 slots = __kvm_memslots(kvm, as_id);
1096 memslot = id_to_memslot(slots, id);
1098 dirty_bitmap = memslot->dirty_bitmap;
1103 n = kvm_dirty_bitmap_bytes(memslot);
1105 dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long);
1106 memset(dirty_bitmap_buffer, 0, n);
1108 spin_lock(&kvm->mmu_lock);
1110 for (i = 0; i < n / sizeof(long); i++) {
1114 if (!dirty_bitmap[i])
1119 mask = xchg(&dirty_bitmap[i], 0);
1120 dirty_bitmap_buffer[i] = mask;
1123 offset = i * BITS_PER_LONG;
1124 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1129 spin_unlock(&kvm->mmu_lock);
1132 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1139 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect);
1142 bool kvm_largepages_enabled(void)
1144 return largepages_enabled;
1147 void kvm_disable_largepages(void)
1149 largepages_enabled = false;
1151 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1153 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1155 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1157 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1159 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
1161 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn);
1164 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1166 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1168 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1169 memslot->flags & KVM_MEMSLOT_INVALID)
1174 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1176 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1178 struct vm_area_struct *vma;
1179 unsigned long addr, size;
1183 addr = gfn_to_hva(kvm, gfn);
1184 if (kvm_is_error_hva(addr))
1187 down_read(¤t->mm->mmap_sem);
1188 vma = find_vma(current->mm, addr);
1192 size = vma_kernel_pagesize(vma);
1195 up_read(¤t->mm->mmap_sem);
1200 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1202 return slot->flags & KVM_MEM_READONLY;
1205 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1206 gfn_t *nr_pages, bool write)
1208 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1209 return KVM_HVA_ERR_BAD;
1211 if (memslot_is_readonly(slot) && write)
1212 return KVM_HVA_ERR_RO_BAD;
1215 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1217 return __gfn_to_hva_memslot(slot, gfn);
1220 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1223 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1226 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1229 return gfn_to_hva_many(slot, gfn, NULL);
1231 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1233 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1235 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1237 EXPORT_SYMBOL_GPL(gfn_to_hva);
1239 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
1241 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
1243 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
1246 * If writable is set to false, the hva returned by this function is only
1247 * allowed to be read.
1249 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1250 gfn_t gfn, bool *writable)
1252 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1254 if (!kvm_is_error_hva(hva) && writable)
1255 *writable = !memslot_is_readonly(slot);
1260 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1262 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1264 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1267 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
1269 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1271 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1274 static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1275 unsigned long start, int write, struct page **page)
1277 int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1280 flags |= FOLL_WRITE;
1282 return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1285 static inline int check_user_page_hwpoison(unsigned long addr)
1287 int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1289 rc = __get_user_pages(current, current->mm, addr, 1,
1290 flags, NULL, NULL, NULL);
1291 return rc == -EHWPOISON;
1295 * The atomic path to get the writable pfn which will be stored in @pfn,
1296 * true indicates success, otherwise false is returned.
1298 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1299 bool write_fault, bool *writable, pfn_t *pfn)
1301 struct page *page[1];
1304 if (!(async || atomic))
1308 * Fast pin a writable pfn only if it is a write fault request
1309 * or the caller allows to map a writable pfn for a read fault
1312 if (!(write_fault || writable))
1315 npages = __get_user_pages_fast(addr, 1, 1, page);
1317 *pfn = page_to_pfn(page[0]);
1328 * The slow path to get the pfn of the specified host virtual address,
1329 * 1 indicates success, -errno is returned if error is detected.
1331 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1332 bool *writable, pfn_t *pfn)
1334 struct page *page[1];
1340 *writable = write_fault;
1343 down_read(¤t->mm->mmap_sem);
1344 npages = get_user_page_nowait(current, current->mm,
1345 addr, write_fault, page);
1346 up_read(¤t->mm->mmap_sem);
1348 npages = __get_user_pages_unlocked(current, current->mm, addr, 1,
1349 write_fault, 0, page,
1350 FOLL_TOUCH|FOLL_HWPOISON);
1354 /* map read fault as writable if possible */
1355 if (unlikely(!write_fault) && writable) {
1356 struct page *wpage[1];
1358 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1367 *pfn = page_to_pfn(page[0]);
1371 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1373 if (unlikely(!(vma->vm_flags & VM_READ)))
1376 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1383 * Pin guest page in memory and return its pfn.
1384 * @addr: host virtual address which maps memory to the guest
1385 * @atomic: whether this function can sleep
1386 * @async: whether this function need to wait IO complete if the
1387 * host page is not in the memory
1388 * @write_fault: whether we should get a writable host page
1389 * @writable: whether it allows to map a writable host page for !@write_fault
1391 * The function will map a writable host page for these two cases:
1392 * 1): @write_fault = true
1393 * 2): @write_fault = false && @writable, @writable will tell the caller
1394 * whether the mapping is writable.
1396 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1397 bool write_fault, bool *writable)
1399 struct vm_area_struct *vma;
1403 /* we can do it either atomically or asynchronously, not both */
1404 BUG_ON(atomic && async);
1406 if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1410 return KVM_PFN_ERR_FAULT;
1412 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1416 down_read(¤t->mm->mmap_sem);
1417 if (npages == -EHWPOISON ||
1418 (!async && check_user_page_hwpoison(addr))) {
1419 pfn = KVM_PFN_ERR_HWPOISON;
1423 vma = find_vma_intersection(current->mm, addr, addr + 1);
1426 pfn = KVM_PFN_ERR_FAULT;
1427 else if ((vma->vm_flags & VM_PFNMAP)) {
1428 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1430 BUG_ON(!kvm_is_reserved_pfn(pfn));
1432 if (async && vma_is_valid(vma, write_fault))
1434 pfn = KVM_PFN_ERR_FAULT;
1437 up_read(¤t->mm->mmap_sem);
1441 pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1442 bool *async, bool write_fault, bool *writable)
1444 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1446 if (addr == KVM_HVA_ERR_RO_BAD)
1447 return KVM_PFN_ERR_RO_FAULT;
1449 if (kvm_is_error_hva(addr))
1450 return KVM_PFN_NOSLOT;
1452 /* Do not map writable pfn in the readonly memslot. */
1453 if (writable && memslot_is_readonly(slot)) {
1458 return hva_to_pfn(addr, atomic, async, write_fault,
1461 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
1463 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1466 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
1467 write_fault, writable);
1469 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1471 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1473 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1475 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
1477 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1479 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1481 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1483 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1485 return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm, gfn), gfn);
1487 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1489 pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
1491 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1493 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
1495 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1497 return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
1499 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1501 pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
1503 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1505 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
1507 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1508 struct page **pages, int nr_pages)
1513 addr = gfn_to_hva_many(slot, gfn, &entry);
1514 if (kvm_is_error_hva(addr))
1517 if (entry < nr_pages)
1520 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1522 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1524 static struct page *kvm_pfn_to_page(pfn_t pfn)
1526 if (is_error_noslot_pfn(pfn))
1527 return KVM_ERR_PTR_BAD_PAGE;
1529 if (kvm_is_reserved_pfn(pfn)) {
1531 return KVM_ERR_PTR_BAD_PAGE;
1534 return pfn_to_page(pfn);
1537 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1541 pfn = gfn_to_pfn(kvm, gfn);
1543 return kvm_pfn_to_page(pfn);
1545 EXPORT_SYMBOL_GPL(gfn_to_page);
1547 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
1551 pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
1553 return kvm_pfn_to_page(pfn);
1555 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
1557 void kvm_release_page_clean(struct page *page)
1559 WARN_ON(is_error_page(page));
1561 kvm_release_pfn_clean(page_to_pfn(page));
1563 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1565 void kvm_release_pfn_clean(pfn_t pfn)
1567 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
1568 put_page(pfn_to_page(pfn));
1570 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1572 void kvm_release_page_dirty(struct page *page)
1574 WARN_ON(is_error_page(page));
1576 kvm_release_pfn_dirty(page_to_pfn(page));
1578 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1580 static void kvm_release_pfn_dirty(pfn_t pfn)
1582 kvm_set_pfn_dirty(pfn);
1583 kvm_release_pfn_clean(pfn);
1586 void kvm_set_pfn_dirty(pfn_t pfn)
1588 if (!kvm_is_reserved_pfn(pfn)) {
1589 struct page *page = pfn_to_page(pfn);
1591 if (!PageReserved(page))
1595 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1597 void kvm_set_pfn_accessed(pfn_t pfn)
1599 if (!kvm_is_reserved_pfn(pfn))
1600 mark_page_accessed(pfn_to_page(pfn));
1602 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1604 void kvm_get_pfn(pfn_t pfn)
1606 if (!kvm_is_reserved_pfn(pfn))
1607 get_page(pfn_to_page(pfn));
1609 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1611 static int next_segment(unsigned long len, int offset)
1613 if (len > PAGE_SIZE - offset)
1614 return PAGE_SIZE - offset;
1619 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
1620 void *data, int offset, int len)
1625 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1626 if (kvm_is_error_hva(addr))
1628 r = __copy_from_user(data, (void __user *)addr + offset, len);
1634 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1637 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1639 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1641 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1643 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
1644 int offset, int len)
1646 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1648 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1650 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
1652 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1654 gfn_t gfn = gpa >> PAGE_SHIFT;
1656 int offset = offset_in_page(gpa);
1659 while ((seg = next_segment(len, offset)) != 0) {
1660 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1670 EXPORT_SYMBOL_GPL(kvm_read_guest);
1672 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
1674 gfn_t gfn = gpa >> PAGE_SHIFT;
1676 int offset = offset_in_page(gpa);
1679 while ((seg = next_segment(len, offset)) != 0) {
1680 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
1690 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
1692 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1693 void *data, int offset, unsigned long len)
1698 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1699 if (kvm_is_error_hva(addr))
1701 pagefault_disable();
1702 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1709 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1712 gfn_t gfn = gpa >> PAGE_SHIFT;
1713 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1714 int offset = offset_in_page(gpa);
1716 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1718 EXPORT_SYMBOL_GPL(kvm_read_guest_atomic);
1720 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
1721 void *data, unsigned long len)
1723 gfn_t gfn = gpa >> PAGE_SHIFT;
1724 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1725 int offset = offset_in_page(gpa);
1727 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1729 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
1731 static int __kvm_write_guest_page(struct kvm_memory_slot *memslot, gfn_t gfn,
1732 const void *data, int offset, int len)
1737 addr = gfn_to_hva_memslot(memslot, gfn);
1738 if (kvm_is_error_hva(addr))
1740 r = __copy_to_user((void __user *)addr + offset, data, len);
1743 mark_page_dirty_in_slot(memslot, gfn);
1747 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
1748 const void *data, int offset, int len)
1750 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1752 return __kvm_write_guest_page(slot, gfn, data, offset, len);
1754 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1756 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
1757 const void *data, int offset, int len)
1759 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1761 return __kvm_write_guest_page(slot, gfn, data, offset, len);
1763 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
1765 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1768 gfn_t gfn = gpa >> PAGE_SHIFT;
1770 int offset = offset_in_page(gpa);
1773 while ((seg = next_segment(len, offset)) != 0) {
1774 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1784 EXPORT_SYMBOL_GPL(kvm_write_guest);
1786 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
1789 gfn_t gfn = gpa >> PAGE_SHIFT;
1791 int offset = offset_in_page(gpa);
1794 while ((seg = next_segment(len, offset)) != 0) {
1795 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
1805 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
1807 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1808 gpa_t gpa, unsigned long len)
1810 struct kvm_memslots *slots = kvm_memslots(kvm);
1811 int offset = offset_in_page(gpa);
1812 gfn_t start_gfn = gpa >> PAGE_SHIFT;
1813 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1814 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1815 gfn_t nr_pages_avail;
1818 ghc->generation = slots->generation;
1820 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1821 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, NULL);
1822 if (!kvm_is_error_hva(ghc->hva) && nr_pages_needed <= 1) {
1826 * If the requested region crosses two memslots, we still
1827 * verify that the entire region is valid here.
1829 while (start_gfn <= end_gfn) {
1830 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1831 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1833 if (kvm_is_error_hva(ghc->hva))
1835 start_gfn += nr_pages_avail;
1837 /* Use the slow path for cross page reads and writes. */
1838 ghc->memslot = NULL;
1842 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1844 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1845 void *data, unsigned long len)
1847 struct kvm_memslots *slots = kvm_memslots(kvm);
1850 BUG_ON(len > ghc->len);
1852 if (slots->generation != ghc->generation)
1853 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1855 if (unlikely(!ghc->memslot))
1856 return kvm_write_guest(kvm, ghc->gpa, data, len);
1858 if (kvm_is_error_hva(ghc->hva))
1861 r = __copy_to_user((void __user *)ghc->hva, data, len);
1864 mark_page_dirty_in_slot(ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1868 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1870 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1871 void *data, unsigned long len)
1873 struct kvm_memslots *slots = kvm_memslots(kvm);
1876 BUG_ON(len > ghc->len);
1878 if (slots->generation != ghc->generation)
1879 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1881 if (unlikely(!ghc->memslot))
1882 return kvm_read_guest(kvm, ghc->gpa, data, len);
1884 if (kvm_is_error_hva(ghc->hva))
1887 r = __copy_from_user(data, (void __user *)ghc->hva, len);
1893 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1895 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1897 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
1899 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
1901 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1903 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1905 gfn_t gfn = gpa >> PAGE_SHIFT;
1907 int offset = offset_in_page(gpa);
1910 while ((seg = next_segment(len, offset)) != 0) {
1911 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1920 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1922 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot,
1925 if (memslot && memslot->dirty_bitmap) {
1926 unsigned long rel_gfn = gfn - memslot->base_gfn;
1928 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1932 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1934 struct kvm_memory_slot *memslot;
1936 memslot = gfn_to_memslot(kvm, gfn);
1937 mark_page_dirty_in_slot(memslot, gfn);
1939 EXPORT_SYMBOL_GPL(mark_page_dirty);
1941 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
1943 struct kvm_memory_slot *memslot;
1945 memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1946 mark_page_dirty_in_slot(memslot, gfn);
1948 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
1950 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
1954 old = val = vcpu->halt_poll_ns;
1956 if (val == 0 && halt_poll_ns_grow)
1959 val *= halt_poll_ns_grow;
1961 vcpu->halt_poll_ns = val;
1962 trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
1965 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
1969 old = val = vcpu->halt_poll_ns;
1970 if (halt_poll_ns_shrink == 0)
1973 val /= halt_poll_ns_shrink;
1975 vcpu->halt_poll_ns = val;
1976 trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
1979 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
1981 if (kvm_arch_vcpu_runnable(vcpu)) {
1982 kvm_make_request(KVM_REQ_UNHALT, vcpu);
1985 if (kvm_cpu_has_pending_timer(vcpu))
1987 if (signal_pending(current))
1994 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1996 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
2000 bool waited = false;
2003 start = cur = ktime_get();
2004 if (vcpu->halt_poll_ns) {
2005 ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns);
2009 * This sets KVM_REQ_UNHALT if an interrupt
2012 if (kvm_vcpu_check_block(vcpu) < 0) {
2013 ++vcpu->stat.halt_successful_poll;
2017 } while (single_task_running() && ktime_before(cur, stop));
2021 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
2023 if (kvm_vcpu_check_block(vcpu) < 0)
2030 finish_wait(&vcpu->wq, &wait);
2034 block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
2037 if (block_ns <= vcpu->halt_poll_ns)
2039 /* we had a long block, shrink polling */
2040 else if (vcpu->halt_poll_ns && block_ns > halt_poll_ns)
2041 shrink_halt_poll_ns(vcpu);
2042 /* we had a short halt and our poll time is too small */
2043 else if (vcpu->halt_poll_ns < halt_poll_ns &&
2044 block_ns < halt_poll_ns)
2045 grow_halt_poll_ns(vcpu);
2048 trace_kvm_vcpu_wakeup(block_ns, waited);
2050 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
2054 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2056 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
2059 int cpu = vcpu->cpu;
2060 wait_queue_head_t *wqp;
2062 wqp = kvm_arch_vcpu_wq(vcpu);
2063 if (waitqueue_active(wqp)) {
2064 wake_up_interruptible(wqp);
2065 ++vcpu->stat.halt_wakeup;
2069 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
2070 if (kvm_arch_vcpu_should_kick(vcpu))
2071 smp_send_reschedule(cpu);
2074 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
2075 #endif /* !CONFIG_S390 */
2077 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
2080 struct task_struct *task = NULL;
2084 pid = rcu_dereference(target->pid);
2086 task = get_pid_task(pid, PIDTYPE_PID);
2090 ret = yield_to(task, 1);
2091 put_task_struct(task);
2095 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
2098 * Helper that checks whether a VCPU is eligible for directed yield.
2099 * Most eligible candidate to yield is decided by following heuristics:
2101 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2102 * (preempted lock holder), indicated by @in_spin_loop.
2103 * Set at the beiginning and cleared at the end of interception/PLE handler.
2105 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2106 * chance last time (mostly it has become eligible now since we have probably
2107 * yielded to lockholder in last iteration. This is done by toggling
2108 * @dy_eligible each time a VCPU checked for eligibility.)
2110 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2111 * to preempted lock-holder could result in wrong VCPU selection and CPU
2112 * burning. Giving priority for a potential lock-holder increases lock
2115 * Since algorithm is based on heuristics, accessing another VCPU data without
2116 * locking does not harm. It may result in trying to yield to same VCPU, fail
2117 * and continue with next VCPU and so on.
2119 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
2121 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2124 eligible = !vcpu->spin_loop.in_spin_loop ||
2125 vcpu->spin_loop.dy_eligible;
2127 if (vcpu->spin_loop.in_spin_loop)
2128 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
2136 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
2138 struct kvm *kvm = me->kvm;
2139 struct kvm_vcpu *vcpu;
2140 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
2146 kvm_vcpu_set_in_spin_loop(me, true);
2148 * We boost the priority of a VCPU that is runnable but not
2149 * currently running, because it got preempted by something
2150 * else and called schedule in __vcpu_run. Hopefully that
2151 * VCPU is holding the lock that we need and will release it.
2152 * We approximate round-robin by starting at the last boosted VCPU.
2154 for (pass = 0; pass < 2 && !yielded && try; pass++) {
2155 kvm_for_each_vcpu(i, vcpu, kvm) {
2156 if (!pass && i <= last_boosted_vcpu) {
2157 i = last_boosted_vcpu;
2159 } else if (pass && i > last_boosted_vcpu)
2161 if (!ACCESS_ONCE(vcpu->preempted))
2165 if (waitqueue_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
2167 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
2170 yielded = kvm_vcpu_yield_to(vcpu);
2172 kvm->last_boosted_vcpu = i;
2174 } else if (yielded < 0) {
2181 kvm_vcpu_set_in_spin_loop(me, false);
2183 /* Ensure vcpu is not eligible during next spinloop */
2184 kvm_vcpu_set_dy_eligible(me, false);
2186 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
2188 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2190 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
2193 if (vmf->pgoff == 0)
2194 page = virt_to_page(vcpu->run);
2196 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
2197 page = virt_to_page(vcpu->arch.pio_data);
2199 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2200 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
2201 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
2204 return kvm_arch_vcpu_fault(vcpu, vmf);
2210 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
2211 .fault = kvm_vcpu_fault,
2214 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2216 vma->vm_ops = &kvm_vcpu_vm_ops;
2220 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2222 struct kvm_vcpu *vcpu = filp->private_data;
2224 kvm_put_kvm(vcpu->kvm);
2228 static struct file_operations kvm_vcpu_fops = {
2229 .release = kvm_vcpu_release,
2230 .unlocked_ioctl = kvm_vcpu_ioctl,
2231 #ifdef CONFIG_KVM_COMPAT
2232 .compat_ioctl = kvm_vcpu_compat_ioctl,
2234 .mmap = kvm_vcpu_mmap,
2235 .llseek = noop_llseek,
2239 * Allocates an inode for the vcpu.
2241 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2243 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
2247 * Creates some virtual cpus. Good luck creating more than one.
2249 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
2252 struct kvm_vcpu *vcpu, *v;
2254 if (id >= KVM_MAX_VCPUS)
2257 vcpu = kvm_arch_vcpu_create(kvm, id);
2259 return PTR_ERR(vcpu);
2261 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2263 r = kvm_arch_vcpu_setup(vcpu);
2267 mutex_lock(&kvm->lock);
2268 if (!kvm_vcpu_compatible(vcpu)) {
2270 goto unlock_vcpu_destroy;
2272 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
2274 goto unlock_vcpu_destroy;
2277 kvm_for_each_vcpu(r, v, kvm)
2278 if (v->vcpu_id == id) {
2280 goto unlock_vcpu_destroy;
2283 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
2285 /* Now it's all set up, let userspace reach it */
2287 r = create_vcpu_fd(vcpu);
2290 goto unlock_vcpu_destroy;
2293 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
2296 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
2297 * before kvm->online_vcpu's incremented value.
2300 atomic_inc(&kvm->online_vcpus);
2302 mutex_unlock(&kvm->lock);
2303 kvm_arch_vcpu_postcreate(vcpu);
2306 unlock_vcpu_destroy:
2307 mutex_unlock(&kvm->lock);
2309 kvm_arch_vcpu_destroy(vcpu);
2313 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2316 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2317 vcpu->sigset_active = 1;
2318 vcpu->sigset = *sigset;
2320 vcpu->sigset_active = 0;
2324 static long kvm_vcpu_ioctl(struct file *filp,
2325 unsigned int ioctl, unsigned long arg)
2327 struct kvm_vcpu *vcpu = filp->private_data;
2328 void __user *argp = (void __user *)arg;
2330 struct kvm_fpu *fpu = NULL;
2331 struct kvm_sregs *kvm_sregs = NULL;
2333 if (vcpu->kvm->mm != current->mm)
2336 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2339 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
2341 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
2342 * so vcpu_load() would break it.
2344 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_S390_IRQ || ioctl == KVM_INTERRUPT)
2345 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2349 r = vcpu_load(vcpu);
2357 if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
2358 /* The thread running this VCPU changed. */
2359 struct pid *oldpid = vcpu->pid;
2360 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
2362 rcu_assign_pointer(vcpu->pid, newpid);
2367 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2368 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2370 case KVM_GET_REGS: {
2371 struct kvm_regs *kvm_regs;
2374 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2377 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2381 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2388 case KVM_SET_REGS: {
2389 struct kvm_regs *kvm_regs;
2392 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2393 if (IS_ERR(kvm_regs)) {
2394 r = PTR_ERR(kvm_regs);
2397 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2401 case KVM_GET_SREGS: {
2402 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2406 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2410 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2415 case KVM_SET_SREGS: {
2416 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2417 if (IS_ERR(kvm_sregs)) {
2418 r = PTR_ERR(kvm_sregs);
2422 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2425 case KVM_GET_MP_STATE: {
2426 struct kvm_mp_state mp_state;
2428 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2432 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
2437 case KVM_SET_MP_STATE: {
2438 struct kvm_mp_state mp_state;
2441 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
2443 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2446 case KVM_TRANSLATE: {
2447 struct kvm_translation tr;
2450 if (copy_from_user(&tr, argp, sizeof(tr)))
2452 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2456 if (copy_to_user(argp, &tr, sizeof(tr)))
2461 case KVM_SET_GUEST_DEBUG: {
2462 struct kvm_guest_debug dbg;
2465 if (copy_from_user(&dbg, argp, sizeof(dbg)))
2467 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2470 case KVM_SET_SIGNAL_MASK: {
2471 struct kvm_signal_mask __user *sigmask_arg = argp;
2472 struct kvm_signal_mask kvm_sigmask;
2473 sigset_t sigset, *p;
2478 if (copy_from_user(&kvm_sigmask, argp,
2479 sizeof(kvm_sigmask)))
2482 if (kvm_sigmask.len != sizeof(sigset))
2485 if (copy_from_user(&sigset, sigmask_arg->sigset,
2490 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2494 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2498 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2502 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2508 fpu = memdup_user(argp, sizeof(*fpu));
2514 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2518 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2527 #ifdef CONFIG_KVM_COMPAT
2528 static long kvm_vcpu_compat_ioctl(struct file *filp,
2529 unsigned int ioctl, unsigned long arg)
2531 struct kvm_vcpu *vcpu = filp->private_data;
2532 void __user *argp = compat_ptr(arg);
2535 if (vcpu->kvm->mm != current->mm)
2539 case KVM_SET_SIGNAL_MASK: {
2540 struct kvm_signal_mask __user *sigmask_arg = argp;
2541 struct kvm_signal_mask kvm_sigmask;
2542 compat_sigset_t csigset;
2547 if (copy_from_user(&kvm_sigmask, argp,
2548 sizeof(kvm_sigmask)))
2551 if (kvm_sigmask.len != sizeof(csigset))
2554 if (copy_from_user(&csigset, sigmask_arg->sigset,
2557 sigset_from_compat(&sigset, &csigset);
2558 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2560 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2564 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2572 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2573 int (*accessor)(struct kvm_device *dev,
2574 struct kvm_device_attr *attr),
2577 struct kvm_device_attr attr;
2582 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2585 return accessor(dev, &attr);
2588 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2591 struct kvm_device *dev = filp->private_data;
2594 case KVM_SET_DEVICE_ATTR:
2595 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2596 case KVM_GET_DEVICE_ATTR:
2597 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2598 case KVM_HAS_DEVICE_ATTR:
2599 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2601 if (dev->ops->ioctl)
2602 return dev->ops->ioctl(dev, ioctl, arg);
2608 static int kvm_device_release(struct inode *inode, struct file *filp)
2610 struct kvm_device *dev = filp->private_data;
2611 struct kvm *kvm = dev->kvm;
2617 static const struct file_operations kvm_device_fops = {
2618 .unlocked_ioctl = kvm_device_ioctl,
2619 #ifdef CONFIG_KVM_COMPAT
2620 .compat_ioctl = kvm_device_ioctl,
2622 .release = kvm_device_release,
2625 struct kvm_device *kvm_device_from_filp(struct file *filp)
2627 if (filp->f_op != &kvm_device_fops)
2630 return filp->private_data;
2633 static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
2634 #ifdef CONFIG_KVM_MPIC
2635 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
2636 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
2639 #ifdef CONFIG_KVM_XICS
2640 [KVM_DEV_TYPE_XICS] = &kvm_xics_ops,
2644 int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
2646 if (type >= ARRAY_SIZE(kvm_device_ops_table))
2649 if (kvm_device_ops_table[type] != NULL)
2652 kvm_device_ops_table[type] = ops;
2656 void kvm_unregister_device_ops(u32 type)
2658 if (kvm_device_ops_table[type] != NULL)
2659 kvm_device_ops_table[type] = NULL;
2662 static int kvm_ioctl_create_device(struct kvm *kvm,
2663 struct kvm_create_device *cd)
2665 struct kvm_device_ops *ops = NULL;
2666 struct kvm_device *dev;
2667 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2670 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
2673 ops = kvm_device_ops_table[cd->type];
2680 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2687 ret = ops->create(dev, cd->type);
2693 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2699 list_add(&dev->vm_node, &kvm->devices);
2705 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2708 case KVM_CAP_USER_MEMORY:
2709 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2710 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2711 case KVM_CAP_INTERNAL_ERROR_DATA:
2712 #ifdef CONFIG_HAVE_KVM_MSI
2713 case KVM_CAP_SIGNAL_MSI:
2715 #ifdef CONFIG_HAVE_KVM_IRQFD
2717 case KVM_CAP_IRQFD_RESAMPLE:
2719 case KVM_CAP_CHECK_EXTENSION_VM:
2721 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2722 case KVM_CAP_IRQ_ROUTING:
2723 return KVM_MAX_IRQ_ROUTES;
2725 #if KVM_ADDRESS_SPACE_NUM > 1
2726 case KVM_CAP_MULTI_ADDRESS_SPACE:
2727 return KVM_ADDRESS_SPACE_NUM;
2732 return kvm_vm_ioctl_check_extension(kvm, arg);
2735 static long kvm_vm_ioctl(struct file *filp,
2736 unsigned int ioctl, unsigned long arg)
2738 struct kvm *kvm = filp->private_data;
2739 void __user *argp = (void __user *)arg;
2742 if (kvm->mm != current->mm)
2745 case KVM_CREATE_VCPU:
2746 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2748 case KVM_SET_USER_MEMORY_REGION: {
2749 struct kvm_userspace_memory_region kvm_userspace_mem;
2752 if (copy_from_user(&kvm_userspace_mem, argp,
2753 sizeof(kvm_userspace_mem)))
2756 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2759 case KVM_GET_DIRTY_LOG: {
2760 struct kvm_dirty_log log;
2763 if (copy_from_user(&log, argp, sizeof(log)))
2765 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2768 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2769 case KVM_REGISTER_COALESCED_MMIO: {
2770 struct kvm_coalesced_mmio_zone zone;
2773 if (copy_from_user(&zone, argp, sizeof(zone)))
2775 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2778 case KVM_UNREGISTER_COALESCED_MMIO: {
2779 struct kvm_coalesced_mmio_zone zone;
2782 if (copy_from_user(&zone, argp, sizeof(zone)))
2784 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2789 struct kvm_irqfd data;
2792 if (copy_from_user(&data, argp, sizeof(data)))
2794 r = kvm_irqfd(kvm, &data);
2797 case KVM_IOEVENTFD: {
2798 struct kvm_ioeventfd data;
2801 if (copy_from_user(&data, argp, sizeof(data)))
2803 r = kvm_ioeventfd(kvm, &data);
2806 #ifdef CONFIG_HAVE_KVM_MSI
2807 case KVM_SIGNAL_MSI: {
2811 if (copy_from_user(&msi, argp, sizeof(msi)))
2813 r = kvm_send_userspace_msi(kvm, &msi);
2817 #ifdef __KVM_HAVE_IRQ_LINE
2818 case KVM_IRQ_LINE_STATUS:
2819 case KVM_IRQ_LINE: {
2820 struct kvm_irq_level irq_event;
2823 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
2826 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
2827 ioctl == KVM_IRQ_LINE_STATUS);
2832 if (ioctl == KVM_IRQ_LINE_STATUS) {
2833 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
2841 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2842 case KVM_SET_GSI_ROUTING: {
2843 struct kvm_irq_routing routing;
2844 struct kvm_irq_routing __user *urouting;
2845 struct kvm_irq_routing_entry *entries;
2848 if (copy_from_user(&routing, argp, sizeof(routing)))
2851 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2856 entries = vmalloc(routing.nr * sizeof(*entries));
2861 if (copy_from_user(entries, urouting->entries,
2862 routing.nr * sizeof(*entries)))
2863 goto out_free_irq_routing;
2864 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2866 out_free_irq_routing:
2870 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
2871 case KVM_CREATE_DEVICE: {
2872 struct kvm_create_device cd;
2875 if (copy_from_user(&cd, argp, sizeof(cd)))
2878 r = kvm_ioctl_create_device(kvm, &cd);
2883 if (copy_to_user(argp, &cd, sizeof(cd)))
2889 case KVM_CHECK_EXTENSION:
2890 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
2893 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2899 #ifdef CONFIG_KVM_COMPAT
2900 struct compat_kvm_dirty_log {
2904 compat_uptr_t dirty_bitmap; /* one bit per page */
2909 static long kvm_vm_compat_ioctl(struct file *filp,
2910 unsigned int ioctl, unsigned long arg)
2912 struct kvm *kvm = filp->private_data;
2915 if (kvm->mm != current->mm)
2918 case KVM_GET_DIRTY_LOG: {
2919 struct compat_kvm_dirty_log compat_log;
2920 struct kvm_dirty_log log;
2923 if (copy_from_user(&compat_log, (void __user *)arg,
2924 sizeof(compat_log)))
2926 log.slot = compat_log.slot;
2927 log.padding1 = compat_log.padding1;
2928 log.padding2 = compat_log.padding2;
2929 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2931 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2935 r = kvm_vm_ioctl(filp, ioctl, arg);
2943 static struct file_operations kvm_vm_fops = {
2944 .release = kvm_vm_release,
2945 .unlocked_ioctl = kvm_vm_ioctl,
2946 #ifdef CONFIG_KVM_COMPAT
2947 .compat_ioctl = kvm_vm_compat_ioctl,
2949 .llseek = noop_llseek,
2952 static int kvm_dev_ioctl_create_vm(unsigned long type)
2957 kvm = kvm_create_vm(type);
2959 return PTR_ERR(kvm);
2960 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2961 r = kvm_coalesced_mmio_init(kvm);
2967 r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC);
2974 static long kvm_dev_ioctl(struct file *filp,
2975 unsigned int ioctl, unsigned long arg)
2980 case KVM_GET_API_VERSION:
2983 r = KVM_API_VERSION;
2986 r = kvm_dev_ioctl_create_vm(arg);
2988 case KVM_CHECK_EXTENSION:
2989 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
2991 case KVM_GET_VCPU_MMAP_SIZE:
2994 r = PAGE_SIZE; /* struct kvm_run */
2996 r += PAGE_SIZE; /* pio data page */
2998 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2999 r += PAGE_SIZE; /* coalesced mmio ring page */
3002 case KVM_TRACE_ENABLE:
3003 case KVM_TRACE_PAUSE:
3004 case KVM_TRACE_DISABLE:
3008 return kvm_arch_dev_ioctl(filp, ioctl, arg);
3014 static struct file_operations kvm_chardev_ops = {
3015 .unlocked_ioctl = kvm_dev_ioctl,
3016 .compat_ioctl = kvm_dev_ioctl,
3017 .llseek = noop_llseek,
3020 static struct miscdevice kvm_dev = {
3026 static void hardware_enable_nolock(void *junk)
3028 int cpu = raw_smp_processor_id();
3031 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
3034 cpumask_set_cpu(cpu, cpus_hardware_enabled);
3036 r = kvm_arch_hardware_enable();
3039 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3040 atomic_inc(&hardware_enable_failed);
3041 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
3045 static void hardware_enable(void)
3047 raw_spin_lock(&kvm_count_lock);
3048 if (kvm_usage_count)
3049 hardware_enable_nolock(NULL);
3050 raw_spin_unlock(&kvm_count_lock);
3053 static void hardware_disable_nolock(void *junk)
3055 int cpu = raw_smp_processor_id();
3057 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
3059 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3060 kvm_arch_hardware_disable();
3063 static void hardware_disable(void)
3065 raw_spin_lock(&kvm_count_lock);
3066 if (kvm_usage_count)
3067 hardware_disable_nolock(NULL);
3068 raw_spin_unlock(&kvm_count_lock);
3071 static void hardware_disable_all_nolock(void)
3073 BUG_ON(!kvm_usage_count);
3076 if (!kvm_usage_count)
3077 on_each_cpu(hardware_disable_nolock, NULL, 1);
3080 static void hardware_disable_all(void)
3082 raw_spin_lock(&kvm_count_lock);
3083 hardware_disable_all_nolock();
3084 raw_spin_unlock(&kvm_count_lock);
3087 static int hardware_enable_all(void)
3091 raw_spin_lock(&kvm_count_lock);
3094 if (kvm_usage_count == 1) {
3095 atomic_set(&hardware_enable_failed, 0);
3096 on_each_cpu(hardware_enable_nolock, NULL, 1);
3098 if (atomic_read(&hardware_enable_failed)) {
3099 hardware_disable_all_nolock();
3104 raw_spin_unlock(&kvm_count_lock);
3109 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
3112 val &= ~CPU_TASKS_FROZEN;
3124 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3128 * Some (well, at least mine) BIOSes hang on reboot if
3131 * And Intel TXT required VMX off for all cpu when system shutdown.
3133 pr_info("kvm: exiting hardware virtualization\n");
3134 kvm_rebooting = true;
3135 on_each_cpu(hardware_disable_nolock, NULL, 1);
3139 static struct notifier_block kvm_reboot_notifier = {
3140 .notifier_call = kvm_reboot,
3144 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3148 for (i = 0; i < bus->dev_count; i++) {
3149 struct kvm_io_device *pos = bus->range[i].dev;
3151 kvm_iodevice_destructor(pos);
3156 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
3157 const struct kvm_io_range *r2)
3159 if (r1->addr < r2->addr)
3161 if (r1->addr + r1->len > r2->addr + r2->len)
3166 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
3168 return kvm_io_bus_cmp(p1, p2);
3171 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
3172 gpa_t addr, int len)
3174 bus->range[bus->dev_count++] = (struct kvm_io_range) {
3180 sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
3181 kvm_io_bus_sort_cmp, NULL);
3186 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
3187 gpa_t addr, int len)
3189 struct kvm_io_range *range, key;
3192 key = (struct kvm_io_range) {
3197 range = bsearch(&key, bus->range, bus->dev_count,
3198 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
3202 off = range - bus->range;
3204 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
3210 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3211 struct kvm_io_range *range, const void *val)
3215 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3219 while (idx < bus->dev_count &&
3220 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3221 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
3230 /* kvm_io_bus_write - called under kvm->slots_lock */
3231 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3232 int len, const void *val)
3234 struct kvm_io_bus *bus;
3235 struct kvm_io_range range;
3238 range = (struct kvm_io_range) {
3243 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3244 r = __kvm_io_bus_write(vcpu, bus, &range, val);
3245 return r < 0 ? r : 0;
3248 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3249 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
3250 gpa_t addr, int len, const void *val, long cookie)
3252 struct kvm_io_bus *bus;
3253 struct kvm_io_range range;
3255 range = (struct kvm_io_range) {
3260 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3262 /* First try the device referenced by cookie. */
3263 if ((cookie >= 0) && (cookie < bus->dev_count) &&
3264 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
3265 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
3270 * cookie contained garbage; fall back to search and return the
3271 * correct cookie value.
3273 return __kvm_io_bus_write(vcpu, bus, &range, val);
3276 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3277 struct kvm_io_range *range, void *val)
3281 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3285 while (idx < bus->dev_count &&
3286 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3287 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
3295 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
3297 /* kvm_io_bus_read - called under kvm->slots_lock */
3298 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3301 struct kvm_io_bus *bus;
3302 struct kvm_io_range range;
3305 range = (struct kvm_io_range) {
3310 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3311 r = __kvm_io_bus_read(vcpu, bus, &range, val);
3312 return r < 0 ? r : 0;
3316 /* Caller must hold slots_lock. */
3317 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3318 int len, struct kvm_io_device *dev)
3320 struct kvm_io_bus *new_bus, *bus;
3322 bus = kvm->buses[bus_idx];
3323 /* exclude ioeventfd which is limited by maximum fd */
3324 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3327 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
3328 sizeof(struct kvm_io_range)), GFP_KERNEL);
3331 memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
3332 sizeof(struct kvm_io_range)));
3333 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
3334 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3335 synchronize_srcu_expedited(&kvm->srcu);
3341 /* Caller must hold slots_lock. */
3342 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3343 struct kvm_io_device *dev)
3346 struct kvm_io_bus *new_bus, *bus;
3348 bus = kvm->buses[bus_idx];
3350 for (i = 0; i < bus->dev_count; i++)
3351 if (bus->range[i].dev == dev) {
3359 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3360 sizeof(struct kvm_io_range)), GFP_KERNEL);
3364 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3365 new_bus->dev_count--;
3366 memcpy(new_bus->range + i, bus->range + i + 1,
3367 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3369 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3370 synchronize_srcu_expedited(&kvm->srcu);
3375 static struct notifier_block kvm_cpu_notifier = {
3376 .notifier_call = kvm_cpu_hotplug,
3379 static int vm_stat_get(void *_offset, u64 *val)
3381 unsigned offset = (long)_offset;
3385 spin_lock(&kvm_lock);
3386 list_for_each_entry(kvm, &vm_list, vm_list)
3387 *val += *(u32 *)((void *)kvm + offset);
3388 spin_unlock(&kvm_lock);
3392 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
3394 static int vcpu_stat_get(void *_offset, u64 *val)
3396 unsigned offset = (long)_offset;
3398 struct kvm_vcpu *vcpu;
3402 spin_lock(&kvm_lock);
3403 list_for_each_entry(kvm, &vm_list, vm_list)
3404 kvm_for_each_vcpu(i, vcpu, kvm)
3405 *val += *(u32 *)((void *)vcpu + offset);
3407 spin_unlock(&kvm_lock);
3411 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
3413 static const struct file_operations *stat_fops[] = {
3414 [KVM_STAT_VCPU] = &vcpu_stat_fops,
3415 [KVM_STAT_VM] = &vm_stat_fops,
3418 static int kvm_init_debug(void)
3421 struct kvm_stats_debugfs_item *p;
3423 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3424 if (kvm_debugfs_dir == NULL)
3427 for (p = debugfs_entries; p->name; ++p) {
3428 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
3429 (void *)(long)p->offset,
3430 stat_fops[p->kind]);
3431 if (p->dentry == NULL)
3438 debugfs_remove_recursive(kvm_debugfs_dir);
3443 static void kvm_exit_debug(void)
3445 struct kvm_stats_debugfs_item *p;
3447 for (p = debugfs_entries; p->name; ++p)
3448 debugfs_remove(p->dentry);
3449 debugfs_remove(kvm_debugfs_dir);
3452 static int kvm_suspend(void)
3454 if (kvm_usage_count)
3455 hardware_disable_nolock(NULL);
3459 static void kvm_resume(void)
3461 if (kvm_usage_count) {
3462 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3463 hardware_enable_nolock(NULL);
3467 static struct syscore_ops kvm_syscore_ops = {
3468 .suspend = kvm_suspend,
3469 .resume = kvm_resume,
3473 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3475 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3478 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3480 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3482 if (vcpu->preempted)
3483 vcpu->preempted = false;
3485 kvm_arch_sched_in(vcpu, cpu);
3487 kvm_arch_vcpu_load(vcpu, cpu);
3490 static void kvm_sched_out(struct preempt_notifier *pn,
3491 struct task_struct *next)
3493 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3495 if (current->state == TASK_RUNNING)
3496 vcpu->preempted = true;
3497 kvm_arch_vcpu_put(vcpu);
3500 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3501 struct module *module)
3506 r = kvm_arch_init(opaque);
3511 * kvm_arch_init makes sure there's at most one caller
3512 * for architectures that support multiple implementations,
3513 * like intel and amd on x86.
3514 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3515 * conflicts in case kvm is already setup for another implementation.
3517 r = kvm_irqfd_init();
3521 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
3526 r = kvm_arch_hardware_setup();
3530 for_each_online_cpu(cpu) {
3531 smp_call_function_single(cpu,
3532 kvm_arch_check_processor_compat,
3538 r = register_cpu_notifier(&kvm_cpu_notifier);
3541 register_reboot_notifier(&kvm_reboot_notifier);
3543 /* A kmem cache lets us meet the alignment requirements of fx_save. */
3545 vcpu_align = __alignof__(struct kvm_vcpu);
3546 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
3548 if (!kvm_vcpu_cache) {
3553 r = kvm_async_pf_init();
3557 kvm_chardev_ops.owner = module;
3558 kvm_vm_fops.owner = module;
3559 kvm_vcpu_fops.owner = module;
3561 r = misc_register(&kvm_dev);
3563 pr_err("kvm: misc device register failed\n");
3567 register_syscore_ops(&kvm_syscore_ops);
3569 kvm_preempt_ops.sched_in = kvm_sched_in;
3570 kvm_preempt_ops.sched_out = kvm_sched_out;
3572 r = kvm_init_debug();
3574 pr_err("kvm: create debugfs files failed\n");
3578 r = kvm_vfio_ops_init();
3584 unregister_syscore_ops(&kvm_syscore_ops);
3585 misc_deregister(&kvm_dev);
3587 kvm_async_pf_deinit();
3589 kmem_cache_destroy(kvm_vcpu_cache);
3591 unregister_reboot_notifier(&kvm_reboot_notifier);
3592 unregister_cpu_notifier(&kvm_cpu_notifier);
3595 kvm_arch_hardware_unsetup();
3597 free_cpumask_var(cpus_hardware_enabled);
3605 EXPORT_SYMBOL_GPL(kvm_init);
3610 misc_deregister(&kvm_dev);
3611 kmem_cache_destroy(kvm_vcpu_cache);
3612 kvm_async_pf_deinit();
3613 unregister_syscore_ops(&kvm_syscore_ops);
3614 unregister_reboot_notifier(&kvm_reboot_notifier);
3615 unregister_cpu_notifier(&kvm_cpu_notifier);
3616 on_each_cpu(hardware_disable_nolock, NULL, 1);
3617 kvm_arch_hardware_unsetup();
3620 free_cpumask_var(cpus_hardware_enabled);
3621 kvm_vfio_ops_exit();
3623 EXPORT_SYMBOL_GPL(kvm_exit);