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_test_young(struct mmu_notifier *mn,
401 struct mm_struct *mm,
402 unsigned long address)
404 struct kvm *kvm = mmu_notifier_to_kvm(mn);
407 idx = srcu_read_lock(&kvm->srcu);
408 spin_lock(&kvm->mmu_lock);
409 young = kvm_test_age_hva(kvm, address);
410 spin_unlock(&kvm->mmu_lock);
411 srcu_read_unlock(&kvm->srcu, idx);
416 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
417 struct mm_struct *mm)
419 struct kvm *kvm = mmu_notifier_to_kvm(mn);
422 idx = srcu_read_lock(&kvm->srcu);
423 kvm_arch_flush_shadow_all(kvm);
424 srcu_read_unlock(&kvm->srcu, idx);
427 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
428 .invalidate_page = kvm_mmu_notifier_invalidate_page,
429 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
430 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
431 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
432 .test_young = kvm_mmu_notifier_test_young,
433 .change_pte = kvm_mmu_notifier_change_pte,
434 .release = kvm_mmu_notifier_release,
437 static int kvm_init_mmu_notifier(struct kvm *kvm)
439 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
440 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
443 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
445 static int kvm_init_mmu_notifier(struct kvm *kvm)
450 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
452 static struct kvm_memslots *kvm_alloc_memslots(void)
455 struct kvm_memslots *slots;
457 slots = kvm_kvzalloc(sizeof(struct kvm_memslots));
462 * Init kvm generation close to the maximum to easily test the
463 * code of handling generation number wrap-around.
465 slots->generation = -150;
466 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
467 slots->id_to_index[i] = slots->memslots[i].id = i;
472 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
474 if (!memslot->dirty_bitmap)
477 kvfree(memslot->dirty_bitmap);
478 memslot->dirty_bitmap = NULL;
482 * Free any memory in @free but not in @dont.
484 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
485 struct kvm_memory_slot *dont)
487 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
488 kvm_destroy_dirty_bitmap(free);
490 kvm_arch_free_memslot(kvm, free, dont);
495 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
497 struct kvm_memory_slot *memslot;
502 kvm_for_each_memslot(memslot, slots)
503 kvm_free_memslot(kvm, memslot, NULL);
508 static struct kvm *kvm_create_vm(unsigned long type)
511 struct kvm *kvm = kvm_arch_alloc_vm();
514 return ERR_PTR(-ENOMEM);
516 r = kvm_arch_init_vm(kvm, type);
518 goto out_err_no_disable;
520 r = hardware_enable_all();
522 goto out_err_no_disable;
524 #ifdef CONFIG_HAVE_KVM_IRQFD
525 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
528 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
531 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
532 kvm->memslots[i] = kvm_alloc_memslots();
533 if (!kvm->memslots[i])
534 goto out_err_no_srcu;
537 if (init_srcu_struct(&kvm->srcu))
538 goto out_err_no_srcu;
539 if (init_srcu_struct(&kvm->irq_srcu))
540 goto out_err_no_irq_srcu;
541 for (i = 0; i < KVM_NR_BUSES; i++) {
542 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
548 spin_lock_init(&kvm->mmu_lock);
549 kvm->mm = current->mm;
550 atomic_inc(&kvm->mm->mm_count);
551 kvm_eventfd_init(kvm);
552 mutex_init(&kvm->lock);
553 mutex_init(&kvm->irq_lock);
554 mutex_init(&kvm->slots_lock);
555 atomic_set(&kvm->users_count, 1);
556 INIT_LIST_HEAD(&kvm->devices);
558 r = kvm_init_mmu_notifier(kvm);
562 spin_lock(&kvm_lock);
563 list_add(&kvm->vm_list, &vm_list);
564 spin_unlock(&kvm_lock);
566 preempt_notifier_inc();
571 cleanup_srcu_struct(&kvm->irq_srcu);
573 cleanup_srcu_struct(&kvm->srcu);
575 hardware_disable_all();
577 for (i = 0; i < KVM_NR_BUSES; i++)
578 kfree(kvm->buses[i]);
579 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
580 kvm_free_memslots(kvm, kvm->memslots[i]);
581 kvm_arch_free_vm(kvm);
586 * Avoid using vmalloc for a small buffer.
587 * Should not be used when the size is statically known.
589 void *kvm_kvzalloc(unsigned long size)
591 if (size > PAGE_SIZE)
592 return vzalloc(size);
594 return kzalloc(size, GFP_KERNEL);
597 static void kvm_destroy_devices(struct kvm *kvm)
599 struct list_head *node, *tmp;
601 list_for_each_safe(node, tmp, &kvm->devices) {
602 struct kvm_device *dev =
603 list_entry(node, struct kvm_device, vm_node);
606 dev->ops->destroy(dev);
610 static void kvm_destroy_vm(struct kvm *kvm)
613 struct mm_struct *mm = kvm->mm;
615 kvm_arch_sync_events(kvm);
616 spin_lock(&kvm_lock);
617 list_del(&kvm->vm_list);
618 spin_unlock(&kvm_lock);
619 kvm_free_irq_routing(kvm);
620 for (i = 0; i < KVM_NR_BUSES; i++)
621 kvm_io_bus_destroy(kvm->buses[i]);
622 kvm_coalesced_mmio_free(kvm);
623 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
624 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
626 kvm_arch_flush_shadow_all(kvm);
628 kvm_arch_destroy_vm(kvm);
629 kvm_destroy_devices(kvm);
630 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
631 kvm_free_memslots(kvm, kvm->memslots[i]);
632 cleanup_srcu_struct(&kvm->irq_srcu);
633 cleanup_srcu_struct(&kvm->srcu);
634 kvm_arch_free_vm(kvm);
635 preempt_notifier_dec();
636 hardware_disable_all();
640 void kvm_get_kvm(struct kvm *kvm)
642 atomic_inc(&kvm->users_count);
644 EXPORT_SYMBOL_GPL(kvm_get_kvm);
646 void kvm_put_kvm(struct kvm *kvm)
648 if (atomic_dec_and_test(&kvm->users_count))
651 EXPORT_SYMBOL_GPL(kvm_put_kvm);
654 static int kvm_vm_release(struct inode *inode, struct file *filp)
656 struct kvm *kvm = filp->private_data;
658 kvm_irqfd_release(kvm);
665 * Allocation size is twice as large as the actual dirty bitmap size.
666 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
668 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
670 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
672 memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
673 if (!memslot->dirty_bitmap)
680 * Insert memslot and re-sort memslots based on their GFN,
681 * so binary search could be used to lookup GFN.
682 * Sorting algorithm takes advantage of having initially
683 * sorted array and known changed memslot position.
685 static void update_memslots(struct kvm_memslots *slots,
686 struct kvm_memory_slot *new)
689 int i = slots->id_to_index[id];
690 struct kvm_memory_slot *mslots = slots->memslots;
692 WARN_ON(mslots[i].id != id);
694 WARN_ON(!mslots[i].npages);
695 if (mslots[i].npages)
698 if (!mslots[i].npages)
702 while (i < KVM_MEM_SLOTS_NUM - 1 &&
703 new->base_gfn <= mslots[i + 1].base_gfn) {
704 if (!mslots[i + 1].npages)
706 mslots[i] = mslots[i + 1];
707 slots->id_to_index[mslots[i].id] = i;
712 * The ">=" is needed when creating a slot with base_gfn == 0,
713 * so that it moves before all those with base_gfn == npages == 0.
715 * On the other hand, if new->npages is zero, the above loop has
716 * already left i pointing to the beginning of the empty part of
717 * mslots, and the ">=" would move the hole backwards in this
718 * case---which is wrong. So skip the loop when deleting a slot.
722 new->base_gfn >= mslots[i - 1].base_gfn) {
723 mslots[i] = mslots[i - 1];
724 slots->id_to_index[mslots[i].id] = i;
728 WARN_ON_ONCE(i != slots->used_slots);
731 slots->id_to_index[mslots[i].id] = i;
734 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
736 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
738 #ifdef __KVM_HAVE_READONLY_MEM
739 valid_flags |= KVM_MEM_READONLY;
742 if (mem->flags & ~valid_flags)
748 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
749 int as_id, struct kvm_memslots *slots)
751 struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id);
754 * Set the low bit in the generation, which disables SPTE caching
755 * until the end of synchronize_srcu_expedited.
757 WARN_ON(old_memslots->generation & 1);
758 slots->generation = old_memslots->generation + 1;
760 rcu_assign_pointer(kvm->memslots[as_id], slots);
761 synchronize_srcu_expedited(&kvm->srcu);
764 * Increment the new memslot generation a second time. This prevents
765 * vm exits that race with memslot updates from caching a memslot
766 * generation that will (potentially) be valid forever.
770 kvm_arch_memslots_updated(kvm, slots);
776 * Allocate some memory and give it an address in the guest physical address
779 * Discontiguous memory is allowed, mostly for framebuffers.
781 * Must be called holding kvm->slots_lock for write.
783 int __kvm_set_memory_region(struct kvm *kvm,
784 const struct kvm_userspace_memory_region *mem)
788 unsigned long npages;
789 struct kvm_memory_slot *slot;
790 struct kvm_memory_slot old, new;
791 struct kvm_memslots *slots = NULL, *old_memslots;
793 enum kvm_mr_change change;
795 r = check_memory_region_flags(mem);
800 as_id = mem->slot >> 16;
803 /* General sanity checks */
804 if (mem->memory_size & (PAGE_SIZE - 1))
806 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
808 /* We can read the guest memory with __xxx_user() later on. */
809 if ((id < KVM_USER_MEM_SLOTS) &&
810 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
811 !access_ok(VERIFY_WRITE,
812 (void __user *)(unsigned long)mem->userspace_addr,
815 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
817 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
820 slot = id_to_memslot(__kvm_memslots(kvm, as_id), id);
821 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
822 npages = mem->memory_size >> PAGE_SHIFT;
824 if (npages > KVM_MEM_MAX_NR_PAGES)
830 new.base_gfn = base_gfn;
832 new.flags = mem->flags;
836 change = KVM_MR_CREATE;
837 else { /* Modify an existing slot. */
838 if ((mem->userspace_addr != old.userspace_addr) ||
839 (npages != old.npages) ||
840 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
843 if (base_gfn != old.base_gfn)
844 change = KVM_MR_MOVE;
845 else if (new.flags != old.flags)
846 change = KVM_MR_FLAGS_ONLY;
847 else { /* Nothing to change. */
856 change = KVM_MR_DELETE;
861 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
862 /* Check for overlaps */
864 kvm_for_each_memslot(slot, __kvm_memslots(kvm, as_id)) {
865 if ((slot->id >= KVM_USER_MEM_SLOTS) ||
868 if (!((base_gfn + npages <= slot->base_gfn) ||
869 (base_gfn >= slot->base_gfn + slot->npages)))
874 /* Free page dirty bitmap if unneeded */
875 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
876 new.dirty_bitmap = NULL;
879 if (change == KVM_MR_CREATE) {
880 new.userspace_addr = mem->userspace_addr;
882 if (kvm_arch_create_memslot(kvm, &new, npages))
886 /* Allocate page dirty bitmap if needed */
887 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
888 if (kvm_create_dirty_bitmap(&new) < 0)
892 slots = kvm_kvzalloc(sizeof(struct kvm_memslots));
895 memcpy(slots, __kvm_memslots(kvm, as_id), sizeof(struct kvm_memslots));
897 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
898 slot = id_to_memslot(slots, id);
899 slot->flags |= KVM_MEMSLOT_INVALID;
901 old_memslots = install_new_memslots(kvm, as_id, slots);
903 /* slot was deleted or moved, clear iommu mapping */
904 kvm_iommu_unmap_pages(kvm, &old);
905 /* From this point no new shadow pages pointing to a deleted,
906 * or moved, memslot will be created.
908 * validation of sp->gfn happens in:
909 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
910 * - kvm_is_visible_gfn (mmu_check_roots)
912 kvm_arch_flush_shadow_memslot(kvm, slot);
915 * We can re-use the old_memslots from above, the only difference
916 * from the currently installed memslots is the invalid flag. This
917 * will get overwritten by update_memslots anyway.
919 slots = old_memslots;
922 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
926 /* actual memory is freed via old in kvm_free_memslot below */
927 if (change == KVM_MR_DELETE) {
928 new.dirty_bitmap = NULL;
929 memset(&new.arch, 0, sizeof(new.arch));
932 update_memslots(slots, &new);
933 old_memslots = install_new_memslots(kvm, as_id, slots);
935 kvm_arch_commit_memory_region(kvm, mem, &old, &new, change);
937 kvm_free_memslot(kvm, &old, &new);
938 kvfree(old_memslots);
941 * IOMMU mapping: New slots need to be mapped. Old slots need to be
942 * un-mapped and re-mapped if their base changes. Since base change
943 * unmapping is handled above with slot deletion, mapping alone is
944 * needed here. Anything else the iommu might care about for existing
945 * slots (size changes, userspace addr changes and read-only flag
946 * changes) is disallowed above, so any other attribute changes getting
947 * here can be skipped.
949 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
950 r = kvm_iommu_map_pages(kvm, &new);
959 kvm_free_memslot(kvm, &new, &old);
963 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
965 int kvm_set_memory_region(struct kvm *kvm,
966 const struct kvm_userspace_memory_region *mem)
970 mutex_lock(&kvm->slots_lock);
971 r = __kvm_set_memory_region(kvm, mem);
972 mutex_unlock(&kvm->slots_lock);
975 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
977 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
978 struct kvm_userspace_memory_region *mem)
980 if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
983 return kvm_set_memory_region(kvm, mem);
986 int kvm_get_dirty_log(struct kvm *kvm,
987 struct kvm_dirty_log *log, int *is_dirty)
989 struct kvm_memslots *slots;
990 struct kvm_memory_slot *memslot;
993 unsigned long any = 0;
996 as_id = log->slot >> 16;
998 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1001 slots = __kvm_memslots(kvm, as_id);
1002 memslot = id_to_memslot(slots, id);
1004 if (!memslot->dirty_bitmap)
1007 n = kvm_dirty_bitmap_bytes(memslot);
1009 for (i = 0; !any && i < n/sizeof(long); ++i)
1010 any = memslot->dirty_bitmap[i];
1013 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1023 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1025 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1027 * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages
1028 * are dirty write protect them for next write.
1029 * @kvm: pointer to kvm instance
1030 * @log: slot id and address to which we copy the log
1031 * @is_dirty: flag set if any page is dirty
1033 * We need to keep it in mind that VCPU threads can write to the bitmap
1034 * concurrently. So, to avoid losing track of dirty pages we keep the
1037 * 1. Take a snapshot of the bit and clear it if needed.
1038 * 2. Write protect the corresponding page.
1039 * 3. Copy the snapshot to the userspace.
1040 * 4. Upon return caller flushes TLB's if needed.
1042 * Between 2 and 4, the guest may write to the page using the remaining TLB
1043 * entry. This is not a problem because the page is reported dirty using
1044 * the snapshot taken before and step 4 ensures that writes done after
1045 * exiting to userspace will be logged for the next call.
1048 int kvm_get_dirty_log_protect(struct kvm *kvm,
1049 struct kvm_dirty_log *log, bool *is_dirty)
1051 struct kvm_memslots *slots;
1052 struct kvm_memory_slot *memslot;
1053 int r, i, as_id, id;
1055 unsigned long *dirty_bitmap;
1056 unsigned long *dirty_bitmap_buffer;
1059 as_id = log->slot >> 16;
1060 id = (u16)log->slot;
1061 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1064 slots = __kvm_memslots(kvm, as_id);
1065 memslot = id_to_memslot(slots, id);
1067 dirty_bitmap = memslot->dirty_bitmap;
1072 n = kvm_dirty_bitmap_bytes(memslot);
1074 dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long);
1075 memset(dirty_bitmap_buffer, 0, n);
1077 spin_lock(&kvm->mmu_lock);
1079 for (i = 0; i < n / sizeof(long); i++) {
1083 if (!dirty_bitmap[i])
1088 mask = xchg(&dirty_bitmap[i], 0);
1089 dirty_bitmap_buffer[i] = mask;
1092 offset = i * BITS_PER_LONG;
1093 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1098 spin_unlock(&kvm->mmu_lock);
1101 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1108 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect);
1111 bool kvm_largepages_enabled(void)
1113 return largepages_enabled;
1116 void kvm_disable_largepages(void)
1118 largepages_enabled = false;
1120 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1122 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1124 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1126 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1128 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
1130 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn);
1133 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1135 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1137 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1138 memslot->flags & KVM_MEMSLOT_INVALID)
1143 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1145 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1147 struct vm_area_struct *vma;
1148 unsigned long addr, size;
1152 addr = gfn_to_hva(kvm, gfn);
1153 if (kvm_is_error_hva(addr))
1156 down_read(¤t->mm->mmap_sem);
1157 vma = find_vma(current->mm, addr);
1161 size = vma_kernel_pagesize(vma);
1164 up_read(¤t->mm->mmap_sem);
1169 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1171 return slot->flags & KVM_MEM_READONLY;
1174 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1175 gfn_t *nr_pages, bool write)
1177 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1178 return KVM_HVA_ERR_BAD;
1180 if (memslot_is_readonly(slot) && write)
1181 return KVM_HVA_ERR_RO_BAD;
1184 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1186 return __gfn_to_hva_memslot(slot, gfn);
1189 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1192 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1195 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1198 return gfn_to_hva_many(slot, gfn, NULL);
1200 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1202 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1204 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1206 EXPORT_SYMBOL_GPL(gfn_to_hva);
1208 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
1210 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
1212 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
1215 * If writable is set to false, the hva returned by this function is only
1216 * allowed to be read.
1218 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1219 gfn_t gfn, bool *writable)
1221 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1223 if (!kvm_is_error_hva(hva) && writable)
1224 *writable = !memslot_is_readonly(slot);
1229 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1231 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1233 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1236 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
1238 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1240 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1243 static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1244 unsigned long start, int write, struct page **page)
1246 int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1249 flags |= FOLL_WRITE;
1251 return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1254 static inline int check_user_page_hwpoison(unsigned long addr)
1256 int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1258 rc = __get_user_pages(current, current->mm, addr, 1,
1259 flags, NULL, NULL, NULL);
1260 return rc == -EHWPOISON;
1264 * The atomic path to get the writable pfn which will be stored in @pfn,
1265 * true indicates success, otherwise false is returned.
1267 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1268 bool write_fault, bool *writable, pfn_t *pfn)
1270 struct page *page[1];
1273 if (!(async || atomic))
1277 * Fast pin a writable pfn only if it is a write fault request
1278 * or the caller allows to map a writable pfn for a read fault
1281 if (!(write_fault || writable))
1284 npages = __get_user_pages_fast(addr, 1, 1, page);
1286 *pfn = page_to_pfn(page[0]);
1297 * The slow path to get the pfn of the specified host virtual address,
1298 * 1 indicates success, -errno is returned if error is detected.
1300 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1301 bool *writable, pfn_t *pfn)
1303 struct page *page[1];
1309 *writable = write_fault;
1312 down_read(¤t->mm->mmap_sem);
1313 npages = get_user_page_nowait(current, current->mm,
1314 addr, write_fault, page);
1315 up_read(¤t->mm->mmap_sem);
1317 npages = __get_user_pages_unlocked(current, current->mm, addr, 1,
1318 write_fault, 0, page,
1319 FOLL_TOUCH|FOLL_HWPOISON);
1323 /* map read fault as writable if possible */
1324 if (unlikely(!write_fault) && writable) {
1325 struct page *wpage[1];
1327 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1336 *pfn = page_to_pfn(page[0]);
1340 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1342 if (unlikely(!(vma->vm_flags & VM_READ)))
1345 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1352 * Pin guest page in memory and return its pfn.
1353 * @addr: host virtual address which maps memory to the guest
1354 * @atomic: whether this function can sleep
1355 * @async: whether this function need to wait IO complete if the
1356 * host page is not in the memory
1357 * @write_fault: whether we should get a writable host page
1358 * @writable: whether it allows to map a writable host page for !@write_fault
1360 * The function will map a writable host page for these two cases:
1361 * 1): @write_fault = true
1362 * 2): @write_fault = false && @writable, @writable will tell the caller
1363 * whether the mapping is writable.
1365 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1366 bool write_fault, bool *writable)
1368 struct vm_area_struct *vma;
1372 /* we can do it either atomically or asynchronously, not both */
1373 BUG_ON(atomic && async);
1375 if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1379 return KVM_PFN_ERR_FAULT;
1381 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1385 down_read(¤t->mm->mmap_sem);
1386 if (npages == -EHWPOISON ||
1387 (!async && check_user_page_hwpoison(addr))) {
1388 pfn = KVM_PFN_ERR_HWPOISON;
1392 vma = find_vma_intersection(current->mm, addr, addr + 1);
1395 pfn = KVM_PFN_ERR_FAULT;
1396 else if ((vma->vm_flags & VM_PFNMAP)) {
1397 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1399 BUG_ON(!kvm_is_reserved_pfn(pfn));
1401 if (async && vma_is_valid(vma, write_fault))
1403 pfn = KVM_PFN_ERR_FAULT;
1406 up_read(¤t->mm->mmap_sem);
1410 pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1411 bool *async, bool write_fault, bool *writable)
1413 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1415 if (addr == KVM_HVA_ERR_RO_BAD)
1416 return KVM_PFN_ERR_RO_FAULT;
1418 if (kvm_is_error_hva(addr))
1419 return KVM_PFN_NOSLOT;
1421 /* Do not map writable pfn in the readonly memslot. */
1422 if (writable && memslot_is_readonly(slot)) {
1427 return hva_to_pfn(addr, atomic, async, write_fault,
1430 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
1432 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1435 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
1436 write_fault, writable);
1438 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1440 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1442 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1444 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
1446 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1448 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1450 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1452 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1454 return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm, gfn), gfn);
1456 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1458 pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
1460 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1462 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
1464 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1466 return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
1468 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1470 pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
1472 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1474 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
1476 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1477 struct page **pages, int nr_pages)
1482 addr = gfn_to_hva_many(slot, gfn, &entry);
1483 if (kvm_is_error_hva(addr))
1486 if (entry < nr_pages)
1489 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1491 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1493 static struct page *kvm_pfn_to_page(pfn_t pfn)
1495 if (is_error_noslot_pfn(pfn))
1496 return KVM_ERR_PTR_BAD_PAGE;
1498 if (kvm_is_reserved_pfn(pfn)) {
1500 return KVM_ERR_PTR_BAD_PAGE;
1503 return pfn_to_page(pfn);
1506 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1510 pfn = gfn_to_pfn(kvm, gfn);
1512 return kvm_pfn_to_page(pfn);
1514 EXPORT_SYMBOL_GPL(gfn_to_page);
1516 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
1520 pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
1522 return kvm_pfn_to_page(pfn);
1524 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
1526 void kvm_release_page_clean(struct page *page)
1528 WARN_ON(is_error_page(page));
1530 kvm_release_pfn_clean(page_to_pfn(page));
1532 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1534 void kvm_release_pfn_clean(pfn_t pfn)
1536 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
1537 put_page(pfn_to_page(pfn));
1539 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1541 void kvm_release_page_dirty(struct page *page)
1543 WARN_ON(is_error_page(page));
1545 kvm_release_pfn_dirty(page_to_pfn(page));
1547 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1549 static void kvm_release_pfn_dirty(pfn_t pfn)
1551 kvm_set_pfn_dirty(pfn);
1552 kvm_release_pfn_clean(pfn);
1555 void kvm_set_pfn_dirty(pfn_t pfn)
1557 if (!kvm_is_reserved_pfn(pfn)) {
1558 struct page *page = pfn_to_page(pfn);
1560 if (!PageReserved(page))
1564 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1566 void kvm_set_pfn_accessed(pfn_t pfn)
1568 if (!kvm_is_reserved_pfn(pfn))
1569 mark_page_accessed(pfn_to_page(pfn));
1571 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1573 void kvm_get_pfn(pfn_t pfn)
1575 if (!kvm_is_reserved_pfn(pfn))
1576 get_page(pfn_to_page(pfn));
1578 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1580 static int next_segment(unsigned long len, int offset)
1582 if (len > PAGE_SIZE - offset)
1583 return PAGE_SIZE - offset;
1588 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
1589 void *data, int offset, int len)
1594 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1595 if (kvm_is_error_hva(addr))
1597 r = __copy_from_user(data, (void __user *)addr + offset, len);
1603 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1606 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1608 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1610 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1612 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
1613 int offset, int len)
1615 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1617 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1619 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
1621 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1623 gfn_t gfn = gpa >> PAGE_SHIFT;
1625 int offset = offset_in_page(gpa);
1628 while ((seg = next_segment(len, offset)) != 0) {
1629 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1639 EXPORT_SYMBOL_GPL(kvm_read_guest);
1641 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
1643 gfn_t gfn = gpa >> PAGE_SHIFT;
1645 int offset = offset_in_page(gpa);
1648 while ((seg = next_segment(len, offset)) != 0) {
1649 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
1659 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
1661 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1662 void *data, int offset, unsigned long len)
1667 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1668 if (kvm_is_error_hva(addr))
1670 pagefault_disable();
1671 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1678 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1681 gfn_t gfn = gpa >> PAGE_SHIFT;
1682 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1683 int offset = offset_in_page(gpa);
1685 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1687 EXPORT_SYMBOL_GPL(kvm_read_guest_atomic);
1689 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
1690 void *data, unsigned long len)
1692 gfn_t gfn = gpa >> PAGE_SHIFT;
1693 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1694 int offset = offset_in_page(gpa);
1696 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1698 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
1700 static int __kvm_write_guest_page(struct kvm_memory_slot *memslot, gfn_t gfn,
1701 const void *data, int offset, int len)
1706 addr = gfn_to_hva_memslot(memslot, gfn);
1707 if (kvm_is_error_hva(addr))
1709 r = __copy_to_user((void __user *)addr + offset, data, len);
1712 mark_page_dirty_in_slot(memslot, gfn);
1716 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
1717 const void *data, int offset, int len)
1719 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1721 return __kvm_write_guest_page(slot, gfn, data, offset, len);
1723 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1725 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
1726 const void *data, int offset, int len)
1728 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1730 return __kvm_write_guest_page(slot, gfn, data, offset, len);
1732 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
1734 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1737 gfn_t gfn = gpa >> PAGE_SHIFT;
1739 int offset = offset_in_page(gpa);
1742 while ((seg = next_segment(len, offset)) != 0) {
1743 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1753 EXPORT_SYMBOL_GPL(kvm_write_guest);
1755 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
1758 gfn_t gfn = gpa >> PAGE_SHIFT;
1760 int offset = offset_in_page(gpa);
1763 while ((seg = next_segment(len, offset)) != 0) {
1764 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
1774 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
1776 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1777 gpa_t gpa, unsigned long len)
1779 struct kvm_memslots *slots = kvm_memslots(kvm);
1780 int offset = offset_in_page(gpa);
1781 gfn_t start_gfn = gpa >> PAGE_SHIFT;
1782 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1783 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1784 gfn_t nr_pages_avail;
1787 ghc->generation = slots->generation;
1789 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1790 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, NULL);
1791 if (!kvm_is_error_hva(ghc->hva) && nr_pages_needed <= 1) {
1795 * If the requested region crosses two memslots, we still
1796 * verify that the entire region is valid here.
1798 while (start_gfn <= end_gfn) {
1799 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1800 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1802 if (kvm_is_error_hva(ghc->hva))
1804 start_gfn += nr_pages_avail;
1806 /* Use the slow path for cross page reads and writes. */
1807 ghc->memslot = NULL;
1811 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1813 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1814 void *data, unsigned long len)
1816 struct kvm_memslots *slots = kvm_memslots(kvm);
1819 BUG_ON(len > ghc->len);
1821 if (slots->generation != ghc->generation)
1822 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1824 if (unlikely(!ghc->memslot))
1825 return kvm_write_guest(kvm, ghc->gpa, data, len);
1827 if (kvm_is_error_hva(ghc->hva))
1830 r = __copy_to_user((void __user *)ghc->hva, data, len);
1833 mark_page_dirty_in_slot(ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1837 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1839 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1840 void *data, unsigned long len)
1842 struct kvm_memslots *slots = kvm_memslots(kvm);
1845 BUG_ON(len > ghc->len);
1847 if (slots->generation != ghc->generation)
1848 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1850 if (unlikely(!ghc->memslot))
1851 return kvm_read_guest(kvm, ghc->gpa, data, len);
1853 if (kvm_is_error_hva(ghc->hva))
1856 r = __copy_from_user(data, (void __user *)ghc->hva, len);
1862 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1864 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1866 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
1868 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
1870 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1872 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1874 gfn_t gfn = gpa >> PAGE_SHIFT;
1876 int offset = offset_in_page(gpa);
1879 while ((seg = next_segment(len, offset)) != 0) {
1880 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1889 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1891 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot,
1894 if (memslot && memslot->dirty_bitmap) {
1895 unsigned long rel_gfn = gfn - memslot->base_gfn;
1897 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1901 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1903 struct kvm_memory_slot *memslot;
1905 memslot = gfn_to_memslot(kvm, gfn);
1906 mark_page_dirty_in_slot(memslot, gfn);
1908 EXPORT_SYMBOL_GPL(mark_page_dirty);
1910 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
1912 struct kvm_memory_slot *memslot;
1914 memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1915 mark_page_dirty_in_slot(memslot, gfn);
1917 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
1919 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
1923 old = val = vcpu->halt_poll_ns;
1925 if (val == 0 && halt_poll_ns_grow)
1928 val *= halt_poll_ns_grow;
1930 vcpu->halt_poll_ns = val;
1931 trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
1934 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
1938 old = val = vcpu->halt_poll_ns;
1939 if (halt_poll_ns_shrink == 0)
1942 val /= halt_poll_ns_shrink;
1944 vcpu->halt_poll_ns = val;
1945 trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
1948 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
1950 if (kvm_arch_vcpu_runnable(vcpu)) {
1951 kvm_make_request(KVM_REQ_UNHALT, vcpu);
1954 if (kvm_cpu_has_pending_timer(vcpu))
1956 if (signal_pending(current))
1963 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1965 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1969 bool waited = false;
1972 start = cur = ktime_get();
1973 if (vcpu->halt_poll_ns) {
1974 ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns);
1978 * This sets KVM_REQ_UNHALT if an interrupt
1981 if (kvm_vcpu_check_block(vcpu) < 0) {
1982 ++vcpu->stat.halt_successful_poll;
1986 } while (single_task_running() && ktime_before(cur, stop));
1990 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1992 if (kvm_vcpu_check_block(vcpu) < 0)
1999 finish_wait(&vcpu->wq, &wait);
2003 block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
2006 if (block_ns <= vcpu->halt_poll_ns)
2008 /* we had a long block, shrink polling */
2009 else if (vcpu->halt_poll_ns && block_ns > halt_poll_ns)
2010 shrink_halt_poll_ns(vcpu);
2011 /* we had a short halt and our poll time is too small */
2012 else if (vcpu->halt_poll_ns < halt_poll_ns &&
2013 block_ns < halt_poll_ns)
2014 grow_halt_poll_ns(vcpu);
2017 trace_kvm_vcpu_wakeup(block_ns, waited);
2019 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
2023 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2025 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
2028 int cpu = vcpu->cpu;
2029 wait_queue_head_t *wqp;
2031 wqp = kvm_arch_vcpu_wq(vcpu);
2032 if (waitqueue_active(wqp)) {
2033 wake_up_interruptible(wqp);
2034 ++vcpu->stat.halt_wakeup;
2038 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
2039 if (kvm_arch_vcpu_should_kick(vcpu))
2040 smp_send_reschedule(cpu);
2043 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
2044 #endif /* !CONFIG_S390 */
2046 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
2049 struct task_struct *task = NULL;
2053 pid = rcu_dereference(target->pid);
2055 task = get_pid_task(pid, PIDTYPE_PID);
2059 ret = yield_to(task, 1);
2060 put_task_struct(task);
2064 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
2067 * Helper that checks whether a VCPU is eligible for directed yield.
2068 * Most eligible candidate to yield is decided by following heuristics:
2070 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2071 * (preempted lock holder), indicated by @in_spin_loop.
2072 * Set at the beiginning and cleared at the end of interception/PLE handler.
2074 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2075 * chance last time (mostly it has become eligible now since we have probably
2076 * yielded to lockholder in last iteration. This is done by toggling
2077 * @dy_eligible each time a VCPU checked for eligibility.)
2079 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2080 * to preempted lock-holder could result in wrong VCPU selection and CPU
2081 * burning. Giving priority for a potential lock-holder increases lock
2084 * Since algorithm is based on heuristics, accessing another VCPU data without
2085 * locking does not harm. It may result in trying to yield to same VCPU, fail
2086 * and continue with next VCPU and so on.
2088 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
2090 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2093 eligible = !vcpu->spin_loop.in_spin_loop ||
2094 vcpu->spin_loop.dy_eligible;
2096 if (vcpu->spin_loop.in_spin_loop)
2097 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
2105 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
2107 struct kvm *kvm = me->kvm;
2108 struct kvm_vcpu *vcpu;
2109 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
2115 kvm_vcpu_set_in_spin_loop(me, true);
2117 * We boost the priority of a VCPU that is runnable but not
2118 * currently running, because it got preempted by something
2119 * else and called schedule in __vcpu_run. Hopefully that
2120 * VCPU is holding the lock that we need and will release it.
2121 * We approximate round-robin by starting at the last boosted VCPU.
2123 for (pass = 0; pass < 2 && !yielded && try; pass++) {
2124 kvm_for_each_vcpu(i, vcpu, kvm) {
2125 if (!pass && i <= last_boosted_vcpu) {
2126 i = last_boosted_vcpu;
2128 } else if (pass && i > last_boosted_vcpu)
2130 if (!ACCESS_ONCE(vcpu->preempted))
2134 if (waitqueue_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
2136 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
2139 yielded = kvm_vcpu_yield_to(vcpu);
2141 kvm->last_boosted_vcpu = i;
2143 } else if (yielded < 0) {
2150 kvm_vcpu_set_in_spin_loop(me, false);
2152 /* Ensure vcpu is not eligible during next spinloop */
2153 kvm_vcpu_set_dy_eligible(me, false);
2155 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
2157 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2159 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
2162 if (vmf->pgoff == 0)
2163 page = virt_to_page(vcpu->run);
2165 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
2166 page = virt_to_page(vcpu->arch.pio_data);
2168 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2169 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
2170 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
2173 return kvm_arch_vcpu_fault(vcpu, vmf);
2179 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
2180 .fault = kvm_vcpu_fault,
2183 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2185 vma->vm_ops = &kvm_vcpu_vm_ops;
2189 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2191 struct kvm_vcpu *vcpu = filp->private_data;
2193 kvm_put_kvm(vcpu->kvm);
2197 static struct file_operations kvm_vcpu_fops = {
2198 .release = kvm_vcpu_release,
2199 .unlocked_ioctl = kvm_vcpu_ioctl,
2200 #ifdef CONFIG_KVM_COMPAT
2201 .compat_ioctl = kvm_vcpu_compat_ioctl,
2203 .mmap = kvm_vcpu_mmap,
2204 .llseek = noop_llseek,
2208 * Allocates an inode for the vcpu.
2210 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2212 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
2216 * Creates some virtual cpus. Good luck creating more than one.
2218 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
2221 struct kvm_vcpu *vcpu, *v;
2223 if (id >= KVM_MAX_VCPUS)
2226 vcpu = kvm_arch_vcpu_create(kvm, id);
2228 return PTR_ERR(vcpu);
2230 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2232 r = kvm_arch_vcpu_setup(vcpu);
2236 mutex_lock(&kvm->lock);
2237 if (!kvm_vcpu_compatible(vcpu)) {
2239 goto unlock_vcpu_destroy;
2241 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
2243 goto unlock_vcpu_destroy;
2246 kvm_for_each_vcpu(r, v, kvm)
2247 if (v->vcpu_id == id) {
2249 goto unlock_vcpu_destroy;
2252 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
2254 /* Now it's all set up, let userspace reach it */
2256 r = create_vcpu_fd(vcpu);
2259 goto unlock_vcpu_destroy;
2262 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
2265 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
2266 * before kvm->online_vcpu's incremented value.
2269 atomic_inc(&kvm->online_vcpus);
2271 mutex_unlock(&kvm->lock);
2272 kvm_arch_vcpu_postcreate(vcpu);
2275 unlock_vcpu_destroy:
2276 mutex_unlock(&kvm->lock);
2278 kvm_arch_vcpu_destroy(vcpu);
2282 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2285 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2286 vcpu->sigset_active = 1;
2287 vcpu->sigset = *sigset;
2289 vcpu->sigset_active = 0;
2293 static long kvm_vcpu_ioctl(struct file *filp,
2294 unsigned int ioctl, unsigned long arg)
2296 struct kvm_vcpu *vcpu = filp->private_data;
2297 void __user *argp = (void __user *)arg;
2299 struct kvm_fpu *fpu = NULL;
2300 struct kvm_sregs *kvm_sregs = NULL;
2302 if (vcpu->kvm->mm != current->mm)
2305 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2308 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
2310 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
2311 * so vcpu_load() would break it.
2313 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_S390_IRQ || ioctl == KVM_INTERRUPT)
2314 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2318 r = vcpu_load(vcpu);
2326 if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
2327 /* The thread running this VCPU changed. */
2328 struct pid *oldpid = vcpu->pid;
2329 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
2331 rcu_assign_pointer(vcpu->pid, newpid);
2336 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2337 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2339 case KVM_GET_REGS: {
2340 struct kvm_regs *kvm_regs;
2343 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2346 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2350 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2357 case KVM_SET_REGS: {
2358 struct kvm_regs *kvm_regs;
2361 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2362 if (IS_ERR(kvm_regs)) {
2363 r = PTR_ERR(kvm_regs);
2366 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2370 case KVM_GET_SREGS: {
2371 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2375 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2379 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2384 case KVM_SET_SREGS: {
2385 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2386 if (IS_ERR(kvm_sregs)) {
2387 r = PTR_ERR(kvm_sregs);
2391 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2394 case KVM_GET_MP_STATE: {
2395 struct kvm_mp_state mp_state;
2397 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2401 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
2406 case KVM_SET_MP_STATE: {
2407 struct kvm_mp_state mp_state;
2410 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
2412 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2415 case KVM_TRANSLATE: {
2416 struct kvm_translation tr;
2419 if (copy_from_user(&tr, argp, sizeof(tr)))
2421 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2425 if (copy_to_user(argp, &tr, sizeof(tr)))
2430 case KVM_SET_GUEST_DEBUG: {
2431 struct kvm_guest_debug dbg;
2434 if (copy_from_user(&dbg, argp, sizeof(dbg)))
2436 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2439 case KVM_SET_SIGNAL_MASK: {
2440 struct kvm_signal_mask __user *sigmask_arg = argp;
2441 struct kvm_signal_mask kvm_sigmask;
2442 sigset_t sigset, *p;
2447 if (copy_from_user(&kvm_sigmask, argp,
2448 sizeof(kvm_sigmask)))
2451 if (kvm_sigmask.len != sizeof(sigset))
2454 if (copy_from_user(&sigset, sigmask_arg->sigset,
2459 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2463 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2467 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2471 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2477 fpu = memdup_user(argp, sizeof(*fpu));
2483 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2487 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2496 #ifdef CONFIG_KVM_COMPAT
2497 static long kvm_vcpu_compat_ioctl(struct file *filp,
2498 unsigned int ioctl, unsigned long arg)
2500 struct kvm_vcpu *vcpu = filp->private_data;
2501 void __user *argp = compat_ptr(arg);
2504 if (vcpu->kvm->mm != current->mm)
2508 case KVM_SET_SIGNAL_MASK: {
2509 struct kvm_signal_mask __user *sigmask_arg = argp;
2510 struct kvm_signal_mask kvm_sigmask;
2511 compat_sigset_t csigset;
2516 if (copy_from_user(&kvm_sigmask, argp,
2517 sizeof(kvm_sigmask)))
2520 if (kvm_sigmask.len != sizeof(csigset))
2523 if (copy_from_user(&csigset, sigmask_arg->sigset,
2526 sigset_from_compat(&sigset, &csigset);
2527 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2529 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2533 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2541 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2542 int (*accessor)(struct kvm_device *dev,
2543 struct kvm_device_attr *attr),
2546 struct kvm_device_attr attr;
2551 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2554 return accessor(dev, &attr);
2557 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2560 struct kvm_device *dev = filp->private_data;
2563 case KVM_SET_DEVICE_ATTR:
2564 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2565 case KVM_GET_DEVICE_ATTR:
2566 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2567 case KVM_HAS_DEVICE_ATTR:
2568 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2570 if (dev->ops->ioctl)
2571 return dev->ops->ioctl(dev, ioctl, arg);
2577 static int kvm_device_release(struct inode *inode, struct file *filp)
2579 struct kvm_device *dev = filp->private_data;
2580 struct kvm *kvm = dev->kvm;
2586 static const struct file_operations kvm_device_fops = {
2587 .unlocked_ioctl = kvm_device_ioctl,
2588 #ifdef CONFIG_KVM_COMPAT
2589 .compat_ioctl = kvm_device_ioctl,
2591 .release = kvm_device_release,
2594 struct kvm_device *kvm_device_from_filp(struct file *filp)
2596 if (filp->f_op != &kvm_device_fops)
2599 return filp->private_data;
2602 static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
2603 #ifdef CONFIG_KVM_MPIC
2604 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
2605 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
2608 #ifdef CONFIG_KVM_XICS
2609 [KVM_DEV_TYPE_XICS] = &kvm_xics_ops,
2613 int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
2615 if (type >= ARRAY_SIZE(kvm_device_ops_table))
2618 if (kvm_device_ops_table[type] != NULL)
2621 kvm_device_ops_table[type] = ops;
2625 void kvm_unregister_device_ops(u32 type)
2627 if (kvm_device_ops_table[type] != NULL)
2628 kvm_device_ops_table[type] = NULL;
2631 static int kvm_ioctl_create_device(struct kvm *kvm,
2632 struct kvm_create_device *cd)
2634 struct kvm_device_ops *ops = NULL;
2635 struct kvm_device *dev;
2636 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2639 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
2642 ops = kvm_device_ops_table[cd->type];
2649 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2656 ret = ops->create(dev, cd->type);
2662 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2668 list_add(&dev->vm_node, &kvm->devices);
2674 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2677 case KVM_CAP_USER_MEMORY:
2678 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2679 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2680 case KVM_CAP_INTERNAL_ERROR_DATA:
2681 #ifdef CONFIG_HAVE_KVM_MSI
2682 case KVM_CAP_SIGNAL_MSI:
2684 #ifdef CONFIG_HAVE_KVM_IRQFD
2686 case KVM_CAP_IRQFD_RESAMPLE:
2688 case KVM_CAP_CHECK_EXTENSION_VM:
2690 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2691 case KVM_CAP_IRQ_ROUTING:
2692 return KVM_MAX_IRQ_ROUTES;
2694 #if KVM_ADDRESS_SPACE_NUM > 1
2695 case KVM_CAP_MULTI_ADDRESS_SPACE:
2696 return KVM_ADDRESS_SPACE_NUM;
2701 return kvm_vm_ioctl_check_extension(kvm, arg);
2704 static long kvm_vm_ioctl(struct file *filp,
2705 unsigned int ioctl, unsigned long arg)
2707 struct kvm *kvm = filp->private_data;
2708 void __user *argp = (void __user *)arg;
2711 if (kvm->mm != current->mm)
2714 case KVM_CREATE_VCPU:
2715 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2717 case KVM_SET_USER_MEMORY_REGION: {
2718 struct kvm_userspace_memory_region kvm_userspace_mem;
2721 if (copy_from_user(&kvm_userspace_mem, argp,
2722 sizeof(kvm_userspace_mem)))
2725 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2728 case KVM_GET_DIRTY_LOG: {
2729 struct kvm_dirty_log log;
2732 if (copy_from_user(&log, argp, sizeof(log)))
2734 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2737 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2738 case KVM_REGISTER_COALESCED_MMIO: {
2739 struct kvm_coalesced_mmio_zone zone;
2742 if (copy_from_user(&zone, argp, sizeof(zone)))
2744 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2747 case KVM_UNREGISTER_COALESCED_MMIO: {
2748 struct kvm_coalesced_mmio_zone zone;
2751 if (copy_from_user(&zone, argp, sizeof(zone)))
2753 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2758 struct kvm_irqfd data;
2761 if (copy_from_user(&data, argp, sizeof(data)))
2763 r = kvm_irqfd(kvm, &data);
2766 case KVM_IOEVENTFD: {
2767 struct kvm_ioeventfd data;
2770 if (copy_from_user(&data, argp, sizeof(data)))
2772 r = kvm_ioeventfd(kvm, &data);
2775 #ifdef CONFIG_HAVE_KVM_MSI
2776 case KVM_SIGNAL_MSI: {
2780 if (copy_from_user(&msi, argp, sizeof(msi)))
2782 r = kvm_send_userspace_msi(kvm, &msi);
2786 #ifdef __KVM_HAVE_IRQ_LINE
2787 case KVM_IRQ_LINE_STATUS:
2788 case KVM_IRQ_LINE: {
2789 struct kvm_irq_level irq_event;
2792 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
2795 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
2796 ioctl == KVM_IRQ_LINE_STATUS);
2801 if (ioctl == KVM_IRQ_LINE_STATUS) {
2802 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
2810 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2811 case KVM_SET_GSI_ROUTING: {
2812 struct kvm_irq_routing routing;
2813 struct kvm_irq_routing __user *urouting;
2814 struct kvm_irq_routing_entry *entries;
2817 if (copy_from_user(&routing, argp, sizeof(routing)))
2820 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2825 entries = vmalloc(routing.nr * sizeof(*entries));
2830 if (copy_from_user(entries, urouting->entries,
2831 routing.nr * sizeof(*entries)))
2832 goto out_free_irq_routing;
2833 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2835 out_free_irq_routing:
2839 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
2840 case KVM_CREATE_DEVICE: {
2841 struct kvm_create_device cd;
2844 if (copy_from_user(&cd, argp, sizeof(cd)))
2847 r = kvm_ioctl_create_device(kvm, &cd);
2852 if (copy_to_user(argp, &cd, sizeof(cd)))
2858 case KVM_CHECK_EXTENSION:
2859 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
2862 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2868 #ifdef CONFIG_KVM_COMPAT
2869 struct compat_kvm_dirty_log {
2873 compat_uptr_t dirty_bitmap; /* one bit per page */
2878 static long kvm_vm_compat_ioctl(struct file *filp,
2879 unsigned int ioctl, unsigned long arg)
2881 struct kvm *kvm = filp->private_data;
2884 if (kvm->mm != current->mm)
2887 case KVM_GET_DIRTY_LOG: {
2888 struct compat_kvm_dirty_log compat_log;
2889 struct kvm_dirty_log log;
2892 if (copy_from_user(&compat_log, (void __user *)arg,
2893 sizeof(compat_log)))
2895 log.slot = compat_log.slot;
2896 log.padding1 = compat_log.padding1;
2897 log.padding2 = compat_log.padding2;
2898 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2900 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2904 r = kvm_vm_ioctl(filp, ioctl, arg);
2912 static struct file_operations kvm_vm_fops = {
2913 .release = kvm_vm_release,
2914 .unlocked_ioctl = kvm_vm_ioctl,
2915 #ifdef CONFIG_KVM_COMPAT
2916 .compat_ioctl = kvm_vm_compat_ioctl,
2918 .llseek = noop_llseek,
2921 static int kvm_dev_ioctl_create_vm(unsigned long type)
2926 kvm = kvm_create_vm(type);
2928 return PTR_ERR(kvm);
2929 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2930 r = kvm_coalesced_mmio_init(kvm);
2936 r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC);
2943 static long kvm_dev_ioctl(struct file *filp,
2944 unsigned int ioctl, unsigned long arg)
2949 case KVM_GET_API_VERSION:
2952 r = KVM_API_VERSION;
2955 r = kvm_dev_ioctl_create_vm(arg);
2957 case KVM_CHECK_EXTENSION:
2958 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
2960 case KVM_GET_VCPU_MMAP_SIZE:
2963 r = PAGE_SIZE; /* struct kvm_run */
2965 r += PAGE_SIZE; /* pio data page */
2967 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2968 r += PAGE_SIZE; /* coalesced mmio ring page */
2971 case KVM_TRACE_ENABLE:
2972 case KVM_TRACE_PAUSE:
2973 case KVM_TRACE_DISABLE:
2977 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2983 static struct file_operations kvm_chardev_ops = {
2984 .unlocked_ioctl = kvm_dev_ioctl,
2985 .compat_ioctl = kvm_dev_ioctl,
2986 .llseek = noop_llseek,
2989 static struct miscdevice kvm_dev = {
2995 static void hardware_enable_nolock(void *junk)
2997 int cpu = raw_smp_processor_id();
3000 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
3003 cpumask_set_cpu(cpu, cpus_hardware_enabled);
3005 r = kvm_arch_hardware_enable();
3008 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3009 atomic_inc(&hardware_enable_failed);
3010 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
3014 static void hardware_enable(void)
3016 raw_spin_lock(&kvm_count_lock);
3017 if (kvm_usage_count)
3018 hardware_enable_nolock(NULL);
3019 raw_spin_unlock(&kvm_count_lock);
3022 static void hardware_disable_nolock(void *junk)
3024 int cpu = raw_smp_processor_id();
3026 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
3028 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3029 kvm_arch_hardware_disable();
3032 static void hardware_disable(void)
3034 raw_spin_lock(&kvm_count_lock);
3035 if (kvm_usage_count)
3036 hardware_disable_nolock(NULL);
3037 raw_spin_unlock(&kvm_count_lock);
3040 static void hardware_disable_all_nolock(void)
3042 BUG_ON(!kvm_usage_count);
3045 if (!kvm_usage_count)
3046 on_each_cpu(hardware_disable_nolock, NULL, 1);
3049 static void hardware_disable_all(void)
3051 raw_spin_lock(&kvm_count_lock);
3052 hardware_disable_all_nolock();
3053 raw_spin_unlock(&kvm_count_lock);
3056 static int hardware_enable_all(void)
3060 raw_spin_lock(&kvm_count_lock);
3063 if (kvm_usage_count == 1) {
3064 atomic_set(&hardware_enable_failed, 0);
3065 on_each_cpu(hardware_enable_nolock, NULL, 1);
3067 if (atomic_read(&hardware_enable_failed)) {
3068 hardware_disable_all_nolock();
3073 raw_spin_unlock(&kvm_count_lock);
3078 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
3081 val &= ~CPU_TASKS_FROZEN;
3093 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3097 * Some (well, at least mine) BIOSes hang on reboot if
3100 * And Intel TXT required VMX off for all cpu when system shutdown.
3102 pr_info("kvm: exiting hardware virtualization\n");
3103 kvm_rebooting = true;
3104 on_each_cpu(hardware_disable_nolock, NULL, 1);
3108 static struct notifier_block kvm_reboot_notifier = {
3109 .notifier_call = kvm_reboot,
3113 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3117 for (i = 0; i < bus->dev_count; i++) {
3118 struct kvm_io_device *pos = bus->range[i].dev;
3120 kvm_iodevice_destructor(pos);
3125 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
3126 const struct kvm_io_range *r2)
3128 if (r1->addr < r2->addr)
3130 if (r1->addr + r1->len > r2->addr + r2->len)
3135 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
3137 return kvm_io_bus_cmp(p1, p2);
3140 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
3141 gpa_t addr, int len)
3143 bus->range[bus->dev_count++] = (struct kvm_io_range) {
3149 sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
3150 kvm_io_bus_sort_cmp, NULL);
3155 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
3156 gpa_t addr, int len)
3158 struct kvm_io_range *range, key;
3161 key = (struct kvm_io_range) {
3166 range = bsearch(&key, bus->range, bus->dev_count,
3167 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
3171 off = range - bus->range;
3173 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
3179 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3180 struct kvm_io_range *range, const void *val)
3184 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3188 while (idx < bus->dev_count &&
3189 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3190 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
3199 /* kvm_io_bus_write - called under kvm->slots_lock */
3200 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3201 int len, const void *val)
3203 struct kvm_io_bus *bus;
3204 struct kvm_io_range range;
3207 range = (struct kvm_io_range) {
3212 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3213 r = __kvm_io_bus_write(vcpu, bus, &range, val);
3214 return r < 0 ? r : 0;
3217 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3218 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
3219 gpa_t addr, int len, const void *val, long cookie)
3221 struct kvm_io_bus *bus;
3222 struct kvm_io_range range;
3224 range = (struct kvm_io_range) {
3229 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3231 /* First try the device referenced by cookie. */
3232 if ((cookie >= 0) && (cookie < bus->dev_count) &&
3233 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
3234 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
3239 * cookie contained garbage; fall back to search and return the
3240 * correct cookie value.
3242 return __kvm_io_bus_write(vcpu, bus, &range, val);
3245 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3246 struct kvm_io_range *range, void *val)
3250 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3254 while (idx < bus->dev_count &&
3255 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3256 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
3264 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
3266 /* kvm_io_bus_read - called under kvm->slots_lock */
3267 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3270 struct kvm_io_bus *bus;
3271 struct kvm_io_range range;
3274 range = (struct kvm_io_range) {
3279 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3280 r = __kvm_io_bus_read(vcpu, bus, &range, val);
3281 return r < 0 ? r : 0;
3285 /* Caller must hold slots_lock. */
3286 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3287 int len, struct kvm_io_device *dev)
3289 struct kvm_io_bus *new_bus, *bus;
3291 bus = kvm->buses[bus_idx];
3292 /* exclude ioeventfd which is limited by maximum fd */
3293 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3296 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
3297 sizeof(struct kvm_io_range)), GFP_KERNEL);
3300 memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
3301 sizeof(struct kvm_io_range)));
3302 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
3303 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3304 synchronize_srcu_expedited(&kvm->srcu);
3310 /* Caller must hold slots_lock. */
3311 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3312 struct kvm_io_device *dev)
3315 struct kvm_io_bus *new_bus, *bus;
3317 bus = kvm->buses[bus_idx];
3319 for (i = 0; i < bus->dev_count; i++)
3320 if (bus->range[i].dev == dev) {
3328 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3329 sizeof(struct kvm_io_range)), GFP_KERNEL);
3333 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3334 new_bus->dev_count--;
3335 memcpy(new_bus->range + i, bus->range + i + 1,
3336 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3338 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3339 synchronize_srcu_expedited(&kvm->srcu);
3344 static struct notifier_block kvm_cpu_notifier = {
3345 .notifier_call = kvm_cpu_hotplug,
3348 static int vm_stat_get(void *_offset, u64 *val)
3350 unsigned offset = (long)_offset;
3354 spin_lock(&kvm_lock);
3355 list_for_each_entry(kvm, &vm_list, vm_list)
3356 *val += *(u32 *)((void *)kvm + offset);
3357 spin_unlock(&kvm_lock);
3361 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
3363 static int vcpu_stat_get(void *_offset, u64 *val)
3365 unsigned offset = (long)_offset;
3367 struct kvm_vcpu *vcpu;
3371 spin_lock(&kvm_lock);
3372 list_for_each_entry(kvm, &vm_list, vm_list)
3373 kvm_for_each_vcpu(i, vcpu, kvm)
3374 *val += *(u32 *)((void *)vcpu + offset);
3376 spin_unlock(&kvm_lock);
3380 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
3382 static const struct file_operations *stat_fops[] = {
3383 [KVM_STAT_VCPU] = &vcpu_stat_fops,
3384 [KVM_STAT_VM] = &vm_stat_fops,
3387 static int kvm_init_debug(void)
3390 struct kvm_stats_debugfs_item *p;
3392 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3393 if (kvm_debugfs_dir == NULL)
3396 for (p = debugfs_entries; p->name; ++p) {
3397 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
3398 (void *)(long)p->offset,
3399 stat_fops[p->kind]);
3400 if (p->dentry == NULL)
3407 debugfs_remove_recursive(kvm_debugfs_dir);
3412 static void kvm_exit_debug(void)
3414 struct kvm_stats_debugfs_item *p;
3416 for (p = debugfs_entries; p->name; ++p)
3417 debugfs_remove(p->dentry);
3418 debugfs_remove(kvm_debugfs_dir);
3421 static int kvm_suspend(void)
3423 if (kvm_usage_count)
3424 hardware_disable_nolock(NULL);
3428 static void kvm_resume(void)
3430 if (kvm_usage_count) {
3431 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3432 hardware_enable_nolock(NULL);
3436 static struct syscore_ops kvm_syscore_ops = {
3437 .suspend = kvm_suspend,
3438 .resume = kvm_resume,
3442 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3444 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3447 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3449 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3451 if (vcpu->preempted)
3452 vcpu->preempted = false;
3454 kvm_arch_sched_in(vcpu, cpu);
3456 kvm_arch_vcpu_load(vcpu, cpu);
3459 static void kvm_sched_out(struct preempt_notifier *pn,
3460 struct task_struct *next)
3462 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3464 if (current->state == TASK_RUNNING)
3465 vcpu->preempted = true;
3466 kvm_arch_vcpu_put(vcpu);
3469 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3470 struct module *module)
3475 r = kvm_arch_init(opaque);
3480 * kvm_arch_init makes sure there's at most one caller
3481 * for architectures that support multiple implementations,
3482 * like intel and amd on x86.
3483 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3484 * conflicts in case kvm is already setup for another implementation.
3486 r = kvm_irqfd_init();
3490 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
3495 r = kvm_arch_hardware_setup();
3499 for_each_online_cpu(cpu) {
3500 smp_call_function_single(cpu,
3501 kvm_arch_check_processor_compat,
3507 r = register_cpu_notifier(&kvm_cpu_notifier);
3510 register_reboot_notifier(&kvm_reboot_notifier);
3512 /* A kmem cache lets us meet the alignment requirements of fx_save. */
3514 vcpu_align = __alignof__(struct kvm_vcpu);
3515 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
3517 if (!kvm_vcpu_cache) {
3522 r = kvm_async_pf_init();
3526 kvm_chardev_ops.owner = module;
3527 kvm_vm_fops.owner = module;
3528 kvm_vcpu_fops.owner = module;
3530 r = misc_register(&kvm_dev);
3532 pr_err("kvm: misc device register failed\n");
3536 register_syscore_ops(&kvm_syscore_ops);
3538 kvm_preempt_ops.sched_in = kvm_sched_in;
3539 kvm_preempt_ops.sched_out = kvm_sched_out;
3541 r = kvm_init_debug();
3543 pr_err("kvm: create debugfs files failed\n");
3547 r = kvm_vfio_ops_init();
3553 unregister_syscore_ops(&kvm_syscore_ops);
3554 misc_deregister(&kvm_dev);
3556 kvm_async_pf_deinit();
3558 kmem_cache_destroy(kvm_vcpu_cache);
3560 unregister_reboot_notifier(&kvm_reboot_notifier);
3561 unregister_cpu_notifier(&kvm_cpu_notifier);
3564 kvm_arch_hardware_unsetup();
3566 free_cpumask_var(cpus_hardware_enabled);
3574 EXPORT_SYMBOL_GPL(kvm_init);
3579 misc_deregister(&kvm_dev);
3580 kmem_cache_destroy(kvm_vcpu_cache);
3581 kvm_async_pf_deinit();
3582 unregister_syscore_ops(&kvm_syscore_ops);
3583 unregister_reboot_notifier(&kvm_reboot_notifier);
3584 unregister_cpu_notifier(&kvm_cpu_notifier);
3585 on_each_cpu(hardware_disable_nolock, NULL, 1);
3586 kvm_arch_hardware_unsetup();
3589 free_cpumask_var(cpus_hardware_enabled);
3590 kvm_vfio_ops_exit();
3592 EXPORT_SYMBOL_GPL(kvm_exit);