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.
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");
72 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
75 DEFINE_SPINLOCK(kvm_lock);
76 static DEFINE_RAW_SPINLOCK(kvm_count_lock);
79 static cpumask_var_t cpus_hardware_enabled;
80 static int kvm_usage_count = 0;
81 static atomic_t hardware_enable_failed;
83 struct kmem_cache *kvm_vcpu_cache;
84 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
86 static __read_mostly struct preempt_ops kvm_preempt_ops;
88 struct dentry *kvm_debugfs_dir;
90 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
93 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
96 static int hardware_enable_all(void);
97 static void hardware_disable_all(void);
99 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
101 static void kvm_release_pfn_dirty(pfn_t pfn);
102 static void mark_page_dirty_in_slot(struct kvm *kvm,
103 struct kvm_memory_slot *memslot, gfn_t gfn);
105 __visible bool kvm_rebooting;
106 EXPORT_SYMBOL_GPL(kvm_rebooting);
108 static bool largepages_enabled = true;
110 bool kvm_is_mmio_pfn(pfn_t pfn)
113 return PageReserved(pfn_to_page(pfn));
119 * Switches to specified vcpu, until a matching vcpu_put()
121 int vcpu_load(struct kvm_vcpu *vcpu)
125 if (mutex_lock_killable(&vcpu->mutex))
127 if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
128 /* The thread running this VCPU changed. */
129 struct pid *oldpid = vcpu->pid;
130 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
131 rcu_assign_pointer(vcpu->pid, newpid);
137 preempt_notifier_register(&vcpu->preempt_notifier);
138 kvm_arch_vcpu_load(vcpu, cpu);
143 void vcpu_put(struct kvm_vcpu *vcpu)
146 kvm_arch_vcpu_put(vcpu);
147 preempt_notifier_unregister(&vcpu->preempt_notifier);
149 mutex_unlock(&vcpu->mutex);
152 static void ack_flush(void *_completed)
156 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
161 struct kvm_vcpu *vcpu;
163 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
166 kvm_for_each_vcpu(i, vcpu, kvm) {
167 kvm_make_request(req, vcpu);
170 /* Set ->requests bit before we read ->mode */
173 if (cpus != NULL && cpu != -1 && cpu != me &&
174 kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
175 cpumask_set_cpu(cpu, cpus);
177 if (unlikely(cpus == NULL))
178 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
179 else if (!cpumask_empty(cpus))
180 smp_call_function_many(cpus, ack_flush, NULL, 1);
184 free_cpumask_var(cpus);
188 void kvm_flush_remote_tlbs(struct kvm *kvm)
190 long dirty_count = kvm->tlbs_dirty;
193 if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
194 ++kvm->stat.remote_tlb_flush;
195 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
197 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
199 void kvm_reload_remote_mmus(struct kvm *kvm)
201 make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
204 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
206 make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
209 void kvm_make_scan_ioapic_request(struct kvm *kvm)
211 make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
214 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
219 mutex_init(&vcpu->mutex);
224 init_waitqueue_head(&vcpu->wq);
225 kvm_async_pf_vcpu_init(vcpu);
227 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
232 vcpu->run = page_address(page);
234 kvm_vcpu_set_in_spin_loop(vcpu, false);
235 kvm_vcpu_set_dy_eligible(vcpu, false);
236 vcpu->preempted = false;
238 r = kvm_arch_vcpu_init(vcpu);
244 free_page((unsigned long)vcpu->run);
248 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
250 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
253 kvm_arch_vcpu_uninit(vcpu);
254 free_page((unsigned long)vcpu->run);
256 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
258 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
259 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
261 return container_of(mn, struct kvm, mmu_notifier);
264 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
265 struct mm_struct *mm,
266 unsigned long address)
268 struct kvm *kvm = mmu_notifier_to_kvm(mn);
269 int need_tlb_flush, idx;
272 * When ->invalidate_page runs, the linux pte has been zapped
273 * already but the page is still allocated until
274 * ->invalidate_page returns. So if we increase the sequence
275 * here the kvm page fault will notice if the spte can't be
276 * established because the page is going to be freed. If
277 * instead the kvm page fault establishes the spte before
278 * ->invalidate_page runs, kvm_unmap_hva will release it
281 * The sequence increase only need to be seen at spin_unlock
282 * time, and not at spin_lock time.
284 * Increasing the sequence after the spin_unlock would be
285 * unsafe because the kvm page fault could then establish the
286 * pte after kvm_unmap_hva returned, without noticing the page
287 * is going to be freed.
289 idx = srcu_read_lock(&kvm->srcu);
290 spin_lock(&kvm->mmu_lock);
292 kvm->mmu_notifier_seq++;
293 need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
294 /* we've to flush the tlb before the pages can be freed */
296 kvm_flush_remote_tlbs(kvm);
298 spin_unlock(&kvm->mmu_lock);
299 srcu_read_unlock(&kvm->srcu, idx);
302 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
303 struct mm_struct *mm,
304 unsigned long address,
307 struct kvm *kvm = mmu_notifier_to_kvm(mn);
310 idx = srcu_read_lock(&kvm->srcu);
311 spin_lock(&kvm->mmu_lock);
312 kvm->mmu_notifier_seq++;
313 kvm_set_spte_hva(kvm, address, pte);
314 spin_unlock(&kvm->mmu_lock);
315 srcu_read_unlock(&kvm->srcu, idx);
318 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
319 struct mm_struct *mm,
323 struct kvm *kvm = mmu_notifier_to_kvm(mn);
324 int need_tlb_flush = 0, idx;
326 idx = srcu_read_lock(&kvm->srcu);
327 spin_lock(&kvm->mmu_lock);
329 * The count increase must become visible at unlock time as no
330 * spte can be established without taking the mmu_lock and
331 * count is also read inside the mmu_lock critical section.
333 kvm->mmu_notifier_count++;
334 need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
335 need_tlb_flush |= kvm->tlbs_dirty;
336 /* we've to flush the tlb before the pages can be freed */
338 kvm_flush_remote_tlbs(kvm);
340 spin_unlock(&kvm->mmu_lock);
341 srcu_read_unlock(&kvm->srcu, idx);
344 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
345 struct mm_struct *mm,
349 struct kvm *kvm = mmu_notifier_to_kvm(mn);
351 spin_lock(&kvm->mmu_lock);
353 * This sequence increase will notify the kvm page fault that
354 * the page that is going to be mapped in the spte could have
357 kvm->mmu_notifier_seq++;
360 * The above sequence increase must be visible before the
361 * below count decrease, which is ensured by the smp_wmb above
362 * in conjunction with the smp_rmb in mmu_notifier_retry().
364 kvm->mmu_notifier_count--;
365 spin_unlock(&kvm->mmu_lock);
367 BUG_ON(kvm->mmu_notifier_count < 0);
370 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
371 struct mm_struct *mm,
375 struct kvm *kvm = mmu_notifier_to_kvm(mn);
378 idx = srcu_read_lock(&kvm->srcu);
379 spin_lock(&kvm->mmu_lock);
381 young = kvm_age_hva(kvm, start, end);
383 kvm_flush_remote_tlbs(kvm);
385 spin_unlock(&kvm->mmu_lock);
386 srcu_read_unlock(&kvm->srcu, idx);
391 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
392 struct mm_struct *mm,
393 unsigned long address)
395 struct kvm *kvm = mmu_notifier_to_kvm(mn);
398 idx = srcu_read_lock(&kvm->srcu);
399 spin_lock(&kvm->mmu_lock);
400 young = kvm_test_age_hva(kvm, address);
401 spin_unlock(&kvm->mmu_lock);
402 srcu_read_unlock(&kvm->srcu, idx);
407 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
408 struct mm_struct *mm)
410 struct kvm *kvm = mmu_notifier_to_kvm(mn);
413 idx = srcu_read_lock(&kvm->srcu);
414 kvm_arch_flush_shadow_all(kvm);
415 srcu_read_unlock(&kvm->srcu, idx);
418 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
419 .invalidate_page = kvm_mmu_notifier_invalidate_page,
420 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
421 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
422 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
423 .test_young = kvm_mmu_notifier_test_young,
424 .change_pte = kvm_mmu_notifier_change_pte,
425 .release = kvm_mmu_notifier_release,
428 static int kvm_init_mmu_notifier(struct kvm *kvm)
430 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
431 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
434 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
436 static int kvm_init_mmu_notifier(struct kvm *kvm)
441 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
443 static void kvm_init_memslots_id(struct kvm *kvm)
446 struct kvm_memslots *slots = kvm->memslots;
448 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
449 slots->id_to_index[i] = slots->memslots[i].id = i;
452 static struct kvm *kvm_create_vm(unsigned long type)
455 struct kvm *kvm = kvm_arch_alloc_vm();
458 return ERR_PTR(-ENOMEM);
460 r = kvm_arch_init_vm(kvm, type);
462 goto out_err_no_disable;
464 r = hardware_enable_all();
466 goto out_err_no_disable;
468 #ifdef CONFIG_HAVE_KVM_IRQCHIP
469 INIT_HLIST_HEAD(&kvm->mask_notifier_list);
471 #ifdef CONFIG_HAVE_KVM_IRQFD
472 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
475 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
478 kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
480 goto out_err_no_srcu;
483 * Init kvm generation close to the maximum to easily test the
484 * code of handling generation number wrap-around.
486 kvm->memslots->generation = -150;
488 kvm_init_memslots_id(kvm);
489 if (init_srcu_struct(&kvm->srcu))
490 goto out_err_no_srcu;
491 if (init_srcu_struct(&kvm->irq_srcu))
492 goto out_err_no_irq_srcu;
493 for (i = 0; i < KVM_NR_BUSES; i++) {
494 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
500 spin_lock_init(&kvm->mmu_lock);
501 kvm->mm = current->mm;
502 atomic_inc(&kvm->mm->mm_count);
503 kvm_eventfd_init(kvm);
504 mutex_init(&kvm->lock);
505 mutex_init(&kvm->irq_lock);
506 mutex_init(&kvm->slots_lock);
507 atomic_set(&kvm->users_count, 1);
508 INIT_LIST_HEAD(&kvm->devices);
510 r = kvm_init_mmu_notifier(kvm);
514 spin_lock(&kvm_lock);
515 list_add(&kvm->vm_list, &vm_list);
516 spin_unlock(&kvm_lock);
521 cleanup_srcu_struct(&kvm->irq_srcu);
523 cleanup_srcu_struct(&kvm->srcu);
525 hardware_disable_all();
527 for (i = 0; i < KVM_NR_BUSES; i++)
528 kfree(kvm->buses[i]);
529 kfree(kvm->memslots);
530 kvm_arch_free_vm(kvm);
535 * Avoid using vmalloc for a small buffer.
536 * Should not be used when the size is statically known.
538 void *kvm_kvzalloc(unsigned long size)
540 if (size > PAGE_SIZE)
541 return vzalloc(size);
543 return kzalloc(size, GFP_KERNEL);
546 void kvm_kvfree(const void *addr)
548 if (is_vmalloc_addr(addr))
554 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
556 if (!memslot->dirty_bitmap)
559 kvm_kvfree(memslot->dirty_bitmap);
560 memslot->dirty_bitmap = NULL;
564 * Free any memory in @free but not in @dont.
566 static void kvm_free_physmem_slot(struct kvm *kvm, struct kvm_memory_slot *free,
567 struct kvm_memory_slot *dont)
569 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
570 kvm_destroy_dirty_bitmap(free);
572 kvm_arch_free_memslot(kvm, free, dont);
577 static void kvm_free_physmem(struct kvm *kvm)
579 struct kvm_memslots *slots = kvm->memslots;
580 struct kvm_memory_slot *memslot;
582 kvm_for_each_memslot(memslot, slots)
583 kvm_free_physmem_slot(kvm, memslot, NULL);
585 kfree(kvm->memslots);
588 static void kvm_destroy_devices(struct kvm *kvm)
590 struct list_head *node, *tmp;
592 list_for_each_safe(node, tmp, &kvm->devices) {
593 struct kvm_device *dev =
594 list_entry(node, struct kvm_device, vm_node);
597 dev->ops->destroy(dev);
601 static void kvm_destroy_vm(struct kvm *kvm)
604 struct mm_struct *mm = kvm->mm;
606 kvm_arch_sync_events(kvm);
607 spin_lock(&kvm_lock);
608 list_del(&kvm->vm_list);
609 spin_unlock(&kvm_lock);
610 kvm_free_irq_routing(kvm);
611 for (i = 0; i < KVM_NR_BUSES; i++)
612 kvm_io_bus_destroy(kvm->buses[i]);
613 kvm_coalesced_mmio_free(kvm);
614 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
615 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
617 kvm_arch_flush_shadow_all(kvm);
619 kvm_arch_destroy_vm(kvm);
620 kvm_destroy_devices(kvm);
621 kvm_free_physmem(kvm);
622 cleanup_srcu_struct(&kvm->irq_srcu);
623 cleanup_srcu_struct(&kvm->srcu);
624 kvm_arch_free_vm(kvm);
625 hardware_disable_all();
629 void kvm_get_kvm(struct kvm *kvm)
631 atomic_inc(&kvm->users_count);
633 EXPORT_SYMBOL_GPL(kvm_get_kvm);
635 void kvm_put_kvm(struct kvm *kvm)
637 if (atomic_dec_and_test(&kvm->users_count))
640 EXPORT_SYMBOL_GPL(kvm_put_kvm);
643 static int kvm_vm_release(struct inode *inode, struct file *filp)
645 struct kvm *kvm = filp->private_data;
647 kvm_irqfd_release(kvm);
654 * Allocation size is twice as large as the actual dirty bitmap size.
655 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
657 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
659 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
661 memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
662 if (!memslot->dirty_bitmap)
668 static int cmp_memslot(const void *slot1, const void *slot2)
670 struct kvm_memory_slot *s1, *s2;
672 s1 = (struct kvm_memory_slot *)slot1;
673 s2 = (struct kvm_memory_slot *)slot2;
675 if (s1->npages < s2->npages)
677 if (s1->npages > s2->npages)
684 * Sort the memslots base on its size, so the larger slots
685 * will get better fit.
687 static void sort_memslots(struct kvm_memslots *slots)
691 sort(slots->memslots, KVM_MEM_SLOTS_NUM,
692 sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
694 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
695 slots->id_to_index[slots->memslots[i].id] = i;
698 static void update_memslots(struct kvm_memslots *slots,
699 struct kvm_memory_slot *new)
703 struct kvm_memory_slot *old = id_to_memslot(slots, id);
704 unsigned long npages = old->npages;
707 if (new->npages != npages)
708 sort_memslots(slots);
712 static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
714 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
716 #ifdef __KVM_HAVE_READONLY_MEM
717 valid_flags |= KVM_MEM_READONLY;
720 if (mem->flags & ~valid_flags)
726 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
727 struct kvm_memslots *slots, struct kvm_memory_slot *new)
729 struct kvm_memslots *old_memslots = kvm->memslots;
732 * Set the low bit in the generation, which disables SPTE caching
733 * until the end of synchronize_srcu_expedited.
735 WARN_ON(old_memslots->generation & 1);
736 slots->generation = old_memslots->generation + 1;
738 update_memslots(slots, new);
739 rcu_assign_pointer(kvm->memslots, slots);
740 synchronize_srcu_expedited(&kvm->srcu);
743 * Increment the new memslot generation a second time. This prevents
744 * vm exits that race with memslot updates from caching a memslot
745 * generation that will (potentially) be valid forever.
749 kvm_arch_memslots_updated(kvm);
755 * Allocate some memory and give it an address in the guest physical address
758 * Discontiguous memory is allowed, mostly for framebuffers.
760 * Must be called holding mmap_sem for write.
762 int __kvm_set_memory_region(struct kvm *kvm,
763 struct kvm_userspace_memory_region *mem)
767 unsigned long npages;
768 struct kvm_memory_slot *slot;
769 struct kvm_memory_slot old, new;
770 struct kvm_memslots *slots = NULL, *old_memslots;
771 enum kvm_mr_change change;
773 r = check_memory_region_flags(mem);
778 /* General sanity checks */
779 if (mem->memory_size & (PAGE_SIZE - 1))
781 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
783 /* We can read the guest memory with __xxx_user() later on. */
784 if ((mem->slot < KVM_USER_MEM_SLOTS) &&
785 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
786 !access_ok(VERIFY_WRITE,
787 (void __user *)(unsigned long)mem->userspace_addr,
790 if (mem->slot >= KVM_MEM_SLOTS_NUM)
792 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
795 slot = id_to_memslot(kvm->memslots, mem->slot);
796 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
797 npages = mem->memory_size >> PAGE_SHIFT;
799 if (npages > KVM_MEM_MAX_NR_PAGES)
803 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
808 new.base_gfn = base_gfn;
810 new.flags = mem->flags;
814 change = KVM_MR_CREATE;
815 else { /* Modify an existing slot. */
816 if ((mem->userspace_addr != old.userspace_addr) ||
817 (npages != old.npages) ||
818 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
821 if (base_gfn != old.base_gfn)
822 change = KVM_MR_MOVE;
823 else if (new.flags != old.flags)
824 change = KVM_MR_FLAGS_ONLY;
825 else { /* Nothing to change. */
830 } else if (old.npages) {
831 change = KVM_MR_DELETE;
832 } else /* Modify a non-existent slot: disallowed. */
835 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
836 /* Check for overlaps */
838 kvm_for_each_memslot(slot, kvm->memslots) {
839 if ((slot->id >= KVM_USER_MEM_SLOTS) ||
840 (slot->id == mem->slot))
842 if (!((base_gfn + npages <= slot->base_gfn) ||
843 (base_gfn >= slot->base_gfn + slot->npages)))
848 /* Free page dirty bitmap if unneeded */
849 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
850 new.dirty_bitmap = NULL;
853 if (change == KVM_MR_CREATE) {
854 new.userspace_addr = mem->userspace_addr;
856 if (kvm_arch_create_memslot(kvm, &new, npages))
860 /* Allocate page dirty bitmap if needed */
861 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
862 if (kvm_create_dirty_bitmap(&new) < 0)
866 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
867 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
871 slot = id_to_memslot(slots, mem->slot);
872 slot->flags |= KVM_MEMSLOT_INVALID;
874 old_memslots = install_new_memslots(kvm, slots, NULL);
876 /* slot was deleted or moved, clear iommu mapping */
877 kvm_iommu_unmap_pages(kvm, &old);
878 /* From this point no new shadow pages pointing to a deleted,
879 * or moved, memslot will be created.
881 * validation of sp->gfn happens in:
882 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
883 * - kvm_is_visible_gfn (mmu_check_roots)
885 kvm_arch_flush_shadow_memslot(kvm, slot);
886 slots = old_memslots;
889 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
895 * We can re-use the old_memslots from above, the only difference
896 * from the currently installed memslots is the invalid flag. This
897 * will get overwritten by update_memslots anyway.
900 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
906 /* actual memory is freed via old in kvm_free_physmem_slot below */
907 if (change == KVM_MR_DELETE) {
908 new.dirty_bitmap = NULL;
909 memset(&new.arch, 0, sizeof(new.arch));
912 old_memslots = install_new_memslots(kvm, slots, &new);
914 kvm_arch_commit_memory_region(kvm, mem, &old, change);
916 kvm_free_physmem_slot(kvm, &old, &new);
920 * IOMMU mapping: New slots need to be mapped. Old slots need to be
921 * un-mapped and re-mapped if their base changes. Since base change
922 * unmapping is handled above with slot deletion, mapping alone is
923 * needed here. Anything else the iommu might care about for existing
924 * slots (size changes, userspace addr changes and read-only flag
925 * changes) is disallowed above, so any other attribute changes getting
926 * here can be skipped.
928 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
929 r = kvm_iommu_map_pages(kvm, &new);
938 kvm_free_physmem_slot(kvm, &new, &old);
942 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
944 int kvm_set_memory_region(struct kvm *kvm,
945 struct kvm_userspace_memory_region *mem)
949 mutex_lock(&kvm->slots_lock);
950 r = __kvm_set_memory_region(kvm, mem);
951 mutex_unlock(&kvm->slots_lock);
954 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
956 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
957 struct kvm_userspace_memory_region *mem)
959 if (mem->slot >= KVM_USER_MEM_SLOTS)
961 return kvm_set_memory_region(kvm, mem);
964 int kvm_get_dirty_log(struct kvm *kvm,
965 struct kvm_dirty_log *log, int *is_dirty)
967 struct kvm_memory_slot *memslot;
970 unsigned long any = 0;
973 if (log->slot >= KVM_USER_MEM_SLOTS)
976 memslot = id_to_memslot(kvm->memslots, log->slot);
978 if (!memslot->dirty_bitmap)
981 n = kvm_dirty_bitmap_bytes(memslot);
983 for (i = 0; !any && i < n/sizeof(long); ++i)
984 any = memslot->dirty_bitmap[i];
987 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
997 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
999 bool kvm_largepages_enabled(void)
1001 return largepages_enabled;
1004 void kvm_disable_largepages(void)
1006 largepages_enabled = false;
1008 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1010 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1012 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1014 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1016 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1018 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1020 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1021 memslot->flags & KVM_MEMSLOT_INVALID)
1026 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1028 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1030 struct vm_area_struct *vma;
1031 unsigned long addr, size;
1035 addr = gfn_to_hva(kvm, gfn);
1036 if (kvm_is_error_hva(addr))
1039 down_read(¤t->mm->mmap_sem);
1040 vma = find_vma(current->mm, addr);
1044 size = vma_kernel_pagesize(vma);
1047 up_read(¤t->mm->mmap_sem);
1052 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1054 return slot->flags & KVM_MEM_READONLY;
1057 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1058 gfn_t *nr_pages, bool write)
1060 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1061 return KVM_HVA_ERR_BAD;
1063 if (memslot_is_readonly(slot) && write)
1064 return KVM_HVA_ERR_RO_BAD;
1067 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1069 return __gfn_to_hva_memslot(slot, gfn);
1072 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1075 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1078 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1081 return gfn_to_hva_many(slot, gfn, NULL);
1083 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1085 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1087 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1089 EXPORT_SYMBOL_GPL(gfn_to_hva);
1092 * If writable is set to false, the hva returned by this function is only
1093 * allowed to be read.
1095 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1097 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1098 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1100 if (!kvm_is_error_hva(hva) && writable)
1101 *writable = !memslot_is_readonly(slot);
1106 static int kvm_read_hva(void *data, void __user *hva, int len)
1108 return __copy_from_user(data, hva, len);
1111 static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1113 return __copy_from_user_inatomic(data, hva, len);
1116 static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1117 unsigned long start, int write, struct page **page)
1119 int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1122 flags |= FOLL_WRITE;
1124 return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1127 int kvm_get_user_page_io(struct task_struct *tsk, struct mm_struct *mm,
1128 unsigned long addr, bool write_fault,
1129 struct page **pagep)
1133 int flags = FOLL_TOUCH | FOLL_HWPOISON |
1134 (pagep ? FOLL_GET : 0) |
1135 (write_fault ? FOLL_WRITE : 0);
1138 * If retrying the fault, we get here *not* having allowed the filemap
1139 * to wait on the page lock. We should now allow waiting on the IO with
1140 * the mmap semaphore released.
1142 down_read(&mm->mmap_sem);
1143 npages = __get_user_pages(tsk, mm, addr, 1, flags, pagep, NULL,
1146 VM_BUG_ON(npages != -EBUSY);
1152 * The previous call has now waited on the IO. Now we can
1153 * retry and complete. Pass TRIED to ensure we do not re
1154 * schedule async IO (see e.g. filemap_fault).
1156 down_read(&mm->mmap_sem);
1157 npages = __get_user_pages(tsk, mm, addr, 1, flags | FOLL_TRIED,
1160 up_read(&mm->mmap_sem);
1164 static inline int check_user_page_hwpoison(unsigned long addr)
1166 int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1168 rc = __get_user_pages(current, current->mm, addr, 1,
1169 flags, NULL, NULL, NULL);
1170 return rc == -EHWPOISON;
1174 * The atomic path to get the writable pfn which will be stored in @pfn,
1175 * true indicates success, otherwise false is returned.
1177 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1178 bool write_fault, bool *writable, pfn_t *pfn)
1180 struct page *page[1];
1183 if (!(async || atomic))
1187 * Fast pin a writable pfn only if it is a write fault request
1188 * or the caller allows to map a writable pfn for a read fault
1191 if (!(write_fault || writable))
1194 npages = __get_user_pages_fast(addr, 1, 1, page);
1196 *pfn = page_to_pfn(page[0]);
1207 * The slow path to get the pfn of the specified host virtual address,
1208 * 1 indicates success, -errno is returned if error is detected.
1210 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1211 bool *writable, pfn_t *pfn)
1213 struct page *page[1];
1219 *writable = write_fault;
1222 down_read(¤t->mm->mmap_sem);
1223 npages = get_user_page_nowait(current, current->mm,
1224 addr, write_fault, page);
1225 up_read(¤t->mm->mmap_sem);
1228 * By now we have tried gup_fast, and possibly async_pf, and we
1229 * are certainly not atomic. Time to retry the gup, allowing
1230 * mmap semaphore to be relinquished in the case of IO.
1232 npages = kvm_get_user_page_io(current, current->mm, addr,
1238 /* map read fault as writable if possible */
1239 if (unlikely(!write_fault) && writable) {
1240 struct page *wpage[1];
1242 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1251 *pfn = page_to_pfn(page[0]);
1255 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1257 if (unlikely(!(vma->vm_flags & VM_READ)))
1260 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1267 * Pin guest page in memory and return its pfn.
1268 * @addr: host virtual address which maps memory to the guest
1269 * @atomic: whether this function can sleep
1270 * @async: whether this function need to wait IO complete if the
1271 * host page is not in the memory
1272 * @write_fault: whether we should get a writable host page
1273 * @writable: whether it allows to map a writable host page for !@write_fault
1275 * The function will map a writable host page for these two cases:
1276 * 1): @write_fault = true
1277 * 2): @write_fault = false && @writable, @writable will tell the caller
1278 * whether the mapping is writable.
1280 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1281 bool write_fault, bool *writable)
1283 struct vm_area_struct *vma;
1287 /* we can do it either atomically or asynchronously, not both */
1288 BUG_ON(atomic && async);
1290 if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1294 return KVM_PFN_ERR_FAULT;
1296 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1300 down_read(¤t->mm->mmap_sem);
1301 if (npages == -EHWPOISON ||
1302 (!async && check_user_page_hwpoison(addr))) {
1303 pfn = KVM_PFN_ERR_HWPOISON;
1307 vma = find_vma_intersection(current->mm, addr, addr + 1);
1310 pfn = KVM_PFN_ERR_FAULT;
1311 else if ((vma->vm_flags & VM_PFNMAP)) {
1312 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1314 BUG_ON(!kvm_is_mmio_pfn(pfn));
1316 if (async && vma_is_valid(vma, write_fault))
1318 pfn = KVM_PFN_ERR_FAULT;
1321 up_read(¤t->mm->mmap_sem);
1326 __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1327 bool *async, bool write_fault, bool *writable)
1329 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1331 if (addr == KVM_HVA_ERR_RO_BAD)
1332 return KVM_PFN_ERR_RO_FAULT;
1334 if (kvm_is_error_hva(addr))
1335 return KVM_PFN_NOSLOT;
1337 /* Do not map writable pfn in the readonly memslot. */
1338 if (writable && memslot_is_readonly(slot)) {
1343 return hva_to_pfn(addr, atomic, async, write_fault,
1347 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1348 bool write_fault, bool *writable)
1350 struct kvm_memory_slot *slot;
1355 slot = gfn_to_memslot(kvm, gfn);
1357 return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1361 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1363 return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1365 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1367 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1368 bool write_fault, bool *writable)
1370 return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1372 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1374 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1376 return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1378 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1380 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1383 return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1385 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1387 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1389 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1392 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1394 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1396 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1398 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1404 addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1405 if (kvm_is_error_hva(addr))
1408 if (entry < nr_pages)
1411 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1413 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1415 static struct page *kvm_pfn_to_page(pfn_t pfn)
1417 if (is_error_noslot_pfn(pfn))
1418 return KVM_ERR_PTR_BAD_PAGE;
1420 if (kvm_is_mmio_pfn(pfn)) {
1422 return KVM_ERR_PTR_BAD_PAGE;
1425 return pfn_to_page(pfn);
1428 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1432 pfn = gfn_to_pfn(kvm, gfn);
1434 return kvm_pfn_to_page(pfn);
1437 EXPORT_SYMBOL_GPL(gfn_to_page);
1439 void kvm_release_page_clean(struct page *page)
1441 WARN_ON(is_error_page(page));
1443 kvm_release_pfn_clean(page_to_pfn(page));
1445 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1447 void kvm_release_pfn_clean(pfn_t pfn)
1449 if (!is_error_noslot_pfn(pfn) && !kvm_is_mmio_pfn(pfn))
1450 put_page(pfn_to_page(pfn));
1452 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1454 void kvm_release_page_dirty(struct page *page)
1456 WARN_ON(is_error_page(page));
1458 kvm_release_pfn_dirty(page_to_pfn(page));
1460 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1462 static void kvm_release_pfn_dirty(pfn_t pfn)
1464 kvm_set_pfn_dirty(pfn);
1465 kvm_release_pfn_clean(pfn);
1468 void kvm_set_pfn_dirty(pfn_t pfn)
1470 if (!kvm_is_mmio_pfn(pfn)) {
1471 struct page *page = pfn_to_page(pfn);
1472 if (!PageReserved(page))
1476 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1478 void kvm_set_pfn_accessed(pfn_t pfn)
1480 if (!kvm_is_mmio_pfn(pfn))
1481 mark_page_accessed(pfn_to_page(pfn));
1483 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1485 void kvm_get_pfn(pfn_t pfn)
1487 if (!kvm_is_mmio_pfn(pfn))
1488 get_page(pfn_to_page(pfn));
1490 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1492 static int next_segment(unsigned long len, int offset)
1494 if (len > PAGE_SIZE - offset)
1495 return PAGE_SIZE - offset;
1500 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1506 addr = gfn_to_hva_prot(kvm, gfn, NULL);
1507 if (kvm_is_error_hva(addr))
1509 r = kvm_read_hva(data, (void __user *)addr + offset, len);
1514 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1516 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1518 gfn_t gfn = gpa >> PAGE_SHIFT;
1520 int offset = offset_in_page(gpa);
1523 while ((seg = next_segment(len, offset)) != 0) {
1524 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1534 EXPORT_SYMBOL_GPL(kvm_read_guest);
1536 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1541 gfn_t gfn = gpa >> PAGE_SHIFT;
1542 int offset = offset_in_page(gpa);
1544 addr = gfn_to_hva_prot(kvm, gfn, NULL);
1545 if (kvm_is_error_hva(addr))
1547 pagefault_disable();
1548 r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
1554 EXPORT_SYMBOL(kvm_read_guest_atomic);
1556 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1557 int offset, int len)
1562 addr = gfn_to_hva(kvm, gfn);
1563 if (kvm_is_error_hva(addr))
1565 r = __copy_to_user((void __user *)addr + offset, data, len);
1568 mark_page_dirty(kvm, gfn);
1571 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1573 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1576 gfn_t gfn = gpa >> PAGE_SHIFT;
1578 int offset = offset_in_page(gpa);
1581 while ((seg = next_segment(len, offset)) != 0) {
1582 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1593 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1594 gpa_t gpa, unsigned long len)
1596 struct kvm_memslots *slots = kvm_memslots(kvm);
1597 int offset = offset_in_page(gpa);
1598 gfn_t start_gfn = gpa >> PAGE_SHIFT;
1599 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1600 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1601 gfn_t nr_pages_avail;
1604 ghc->generation = slots->generation;
1606 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1607 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, &nr_pages_avail);
1608 if (!kvm_is_error_hva(ghc->hva) && nr_pages_avail >= nr_pages_needed) {
1612 * If the requested region crosses two memslots, we still
1613 * verify that the entire region is valid here.
1615 while (start_gfn <= end_gfn) {
1616 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1617 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1619 if (kvm_is_error_hva(ghc->hva))
1621 start_gfn += nr_pages_avail;
1623 /* Use the slow path for cross page reads and writes. */
1624 ghc->memslot = NULL;
1628 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1630 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1631 void *data, unsigned long len)
1633 struct kvm_memslots *slots = kvm_memslots(kvm);
1636 BUG_ON(len > ghc->len);
1638 if (slots->generation != ghc->generation)
1639 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1641 if (unlikely(!ghc->memslot))
1642 return kvm_write_guest(kvm, ghc->gpa, data, len);
1644 if (kvm_is_error_hva(ghc->hva))
1647 r = __copy_to_user((void __user *)ghc->hva, data, len);
1650 mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1654 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1656 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1657 void *data, unsigned long len)
1659 struct kvm_memslots *slots = kvm_memslots(kvm);
1662 BUG_ON(len > ghc->len);
1664 if (slots->generation != ghc->generation)
1665 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1667 if (unlikely(!ghc->memslot))
1668 return kvm_read_guest(kvm, ghc->gpa, data, len);
1670 if (kvm_is_error_hva(ghc->hva))
1673 r = __copy_from_user(data, (void __user *)ghc->hva, len);
1679 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1681 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1683 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
1685 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
1687 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1689 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1691 gfn_t gfn = gpa >> PAGE_SHIFT;
1693 int offset = offset_in_page(gpa);
1696 while ((seg = next_segment(len, offset)) != 0) {
1697 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1706 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1708 static void mark_page_dirty_in_slot(struct kvm *kvm,
1709 struct kvm_memory_slot *memslot,
1712 if (memslot && memslot->dirty_bitmap) {
1713 unsigned long rel_gfn = gfn - memslot->base_gfn;
1715 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1719 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1721 struct kvm_memory_slot *memslot;
1723 memslot = gfn_to_memslot(kvm, gfn);
1724 mark_page_dirty_in_slot(kvm, memslot, gfn);
1726 EXPORT_SYMBOL_GPL(mark_page_dirty);
1729 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1731 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1736 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1738 if (kvm_arch_vcpu_runnable(vcpu)) {
1739 kvm_make_request(KVM_REQ_UNHALT, vcpu);
1742 if (kvm_cpu_has_pending_timer(vcpu))
1744 if (signal_pending(current))
1750 finish_wait(&vcpu->wq, &wait);
1752 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
1756 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1758 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1761 int cpu = vcpu->cpu;
1762 wait_queue_head_t *wqp;
1764 wqp = kvm_arch_vcpu_wq(vcpu);
1765 if (waitqueue_active(wqp)) {
1766 wake_up_interruptible(wqp);
1767 ++vcpu->stat.halt_wakeup;
1771 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1772 if (kvm_arch_vcpu_should_kick(vcpu))
1773 smp_send_reschedule(cpu);
1776 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
1777 #endif /* !CONFIG_S390 */
1779 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
1782 struct task_struct *task = NULL;
1786 pid = rcu_dereference(target->pid);
1788 task = get_pid_task(target->pid, PIDTYPE_PID);
1792 if (task->flags & PF_VCPU) {
1793 put_task_struct(task);
1796 ret = yield_to(task, 1);
1797 put_task_struct(task);
1801 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1804 * Helper that checks whether a VCPU is eligible for directed yield.
1805 * Most eligible candidate to yield is decided by following heuristics:
1807 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1808 * (preempted lock holder), indicated by @in_spin_loop.
1809 * Set at the beiginning and cleared at the end of interception/PLE handler.
1811 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1812 * chance last time (mostly it has become eligible now since we have probably
1813 * yielded to lockholder in last iteration. This is done by toggling
1814 * @dy_eligible each time a VCPU checked for eligibility.)
1816 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1817 * to preempted lock-holder could result in wrong VCPU selection and CPU
1818 * burning. Giving priority for a potential lock-holder increases lock
1821 * Since algorithm is based on heuristics, accessing another VCPU data without
1822 * locking does not harm. It may result in trying to yield to same VCPU, fail
1823 * and continue with next VCPU and so on.
1825 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1827 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1830 eligible = !vcpu->spin_loop.in_spin_loop ||
1831 vcpu->spin_loop.dy_eligible;
1833 if (vcpu->spin_loop.in_spin_loop)
1834 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1842 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1844 struct kvm *kvm = me->kvm;
1845 struct kvm_vcpu *vcpu;
1846 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1852 kvm_vcpu_set_in_spin_loop(me, true);
1854 * We boost the priority of a VCPU that is runnable but not
1855 * currently running, because it got preempted by something
1856 * else and called schedule in __vcpu_run. Hopefully that
1857 * VCPU is holding the lock that we need and will release it.
1858 * We approximate round-robin by starting at the last boosted VCPU.
1860 for (pass = 0; pass < 2 && !yielded && try; pass++) {
1861 kvm_for_each_vcpu(i, vcpu, kvm) {
1862 if (!pass && i <= last_boosted_vcpu) {
1863 i = last_boosted_vcpu;
1865 } else if (pass && i > last_boosted_vcpu)
1867 if (!ACCESS_ONCE(vcpu->preempted))
1871 if (waitqueue_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
1873 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1876 yielded = kvm_vcpu_yield_to(vcpu);
1878 kvm->last_boosted_vcpu = i;
1880 } else if (yielded < 0) {
1887 kvm_vcpu_set_in_spin_loop(me, false);
1889 /* Ensure vcpu is not eligible during next spinloop */
1890 kvm_vcpu_set_dy_eligible(me, false);
1892 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1894 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1896 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1899 if (vmf->pgoff == 0)
1900 page = virt_to_page(vcpu->run);
1902 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1903 page = virt_to_page(vcpu->arch.pio_data);
1905 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1906 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1907 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1910 return kvm_arch_vcpu_fault(vcpu, vmf);
1916 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1917 .fault = kvm_vcpu_fault,
1920 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1922 vma->vm_ops = &kvm_vcpu_vm_ops;
1926 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1928 struct kvm_vcpu *vcpu = filp->private_data;
1930 kvm_put_kvm(vcpu->kvm);
1934 static struct file_operations kvm_vcpu_fops = {
1935 .release = kvm_vcpu_release,
1936 .unlocked_ioctl = kvm_vcpu_ioctl,
1937 #ifdef CONFIG_COMPAT
1938 .compat_ioctl = kvm_vcpu_compat_ioctl,
1940 .mmap = kvm_vcpu_mmap,
1941 .llseek = noop_llseek,
1945 * Allocates an inode for the vcpu.
1947 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1949 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
1953 * Creates some virtual cpus. Good luck creating more than one.
1955 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1958 struct kvm_vcpu *vcpu, *v;
1960 if (id >= KVM_MAX_VCPUS)
1963 vcpu = kvm_arch_vcpu_create(kvm, id);
1965 return PTR_ERR(vcpu);
1967 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1969 r = kvm_arch_vcpu_setup(vcpu);
1973 mutex_lock(&kvm->lock);
1974 if (!kvm_vcpu_compatible(vcpu)) {
1976 goto unlock_vcpu_destroy;
1978 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1980 goto unlock_vcpu_destroy;
1983 kvm_for_each_vcpu(r, v, kvm)
1984 if (v->vcpu_id == id) {
1986 goto unlock_vcpu_destroy;
1989 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1991 /* Now it's all set up, let userspace reach it */
1993 r = create_vcpu_fd(vcpu);
1996 goto unlock_vcpu_destroy;
1999 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
2001 atomic_inc(&kvm->online_vcpus);
2003 mutex_unlock(&kvm->lock);
2004 kvm_arch_vcpu_postcreate(vcpu);
2007 unlock_vcpu_destroy:
2008 mutex_unlock(&kvm->lock);
2010 kvm_arch_vcpu_destroy(vcpu);
2014 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2017 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2018 vcpu->sigset_active = 1;
2019 vcpu->sigset = *sigset;
2021 vcpu->sigset_active = 0;
2025 static long kvm_vcpu_ioctl(struct file *filp,
2026 unsigned int ioctl, unsigned long arg)
2028 struct kvm_vcpu *vcpu = filp->private_data;
2029 void __user *argp = (void __user *)arg;
2031 struct kvm_fpu *fpu = NULL;
2032 struct kvm_sregs *kvm_sregs = NULL;
2034 if (vcpu->kvm->mm != current->mm)
2037 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2040 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
2042 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
2043 * so vcpu_load() would break it.
2045 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
2046 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2050 r = vcpu_load(vcpu);
2058 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2059 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2061 case KVM_GET_REGS: {
2062 struct kvm_regs *kvm_regs;
2065 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2068 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2072 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2079 case KVM_SET_REGS: {
2080 struct kvm_regs *kvm_regs;
2083 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2084 if (IS_ERR(kvm_regs)) {
2085 r = PTR_ERR(kvm_regs);
2088 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2092 case KVM_GET_SREGS: {
2093 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2097 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2101 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2106 case KVM_SET_SREGS: {
2107 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2108 if (IS_ERR(kvm_sregs)) {
2109 r = PTR_ERR(kvm_sregs);
2113 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2116 case KVM_GET_MP_STATE: {
2117 struct kvm_mp_state mp_state;
2119 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2123 if (copy_to_user(argp, &mp_state, sizeof mp_state))
2128 case KVM_SET_MP_STATE: {
2129 struct kvm_mp_state mp_state;
2132 if (copy_from_user(&mp_state, argp, sizeof mp_state))
2134 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2137 case KVM_TRANSLATE: {
2138 struct kvm_translation tr;
2141 if (copy_from_user(&tr, argp, sizeof tr))
2143 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2147 if (copy_to_user(argp, &tr, sizeof tr))
2152 case KVM_SET_GUEST_DEBUG: {
2153 struct kvm_guest_debug dbg;
2156 if (copy_from_user(&dbg, argp, sizeof dbg))
2158 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2161 case KVM_SET_SIGNAL_MASK: {
2162 struct kvm_signal_mask __user *sigmask_arg = argp;
2163 struct kvm_signal_mask kvm_sigmask;
2164 sigset_t sigset, *p;
2169 if (copy_from_user(&kvm_sigmask, argp,
2170 sizeof kvm_sigmask))
2173 if (kvm_sigmask.len != sizeof sigset)
2176 if (copy_from_user(&sigset, sigmask_arg->sigset,
2181 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2185 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2189 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2193 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2199 fpu = memdup_user(argp, sizeof(*fpu));
2205 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2209 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2218 #ifdef CONFIG_COMPAT
2219 static long kvm_vcpu_compat_ioctl(struct file *filp,
2220 unsigned int ioctl, unsigned long arg)
2222 struct kvm_vcpu *vcpu = filp->private_data;
2223 void __user *argp = compat_ptr(arg);
2226 if (vcpu->kvm->mm != current->mm)
2230 case KVM_SET_SIGNAL_MASK: {
2231 struct kvm_signal_mask __user *sigmask_arg = argp;
2232 struct kvm_signal_mask kvm_sigmask;
2233 compat_sigset_t csigset;
2238 if (copy_from_user(&kvm_sigmask, argp,
2239 sizeof kvm_sigmask))
2242 if (kvm_sigmask.len != sizeof csigset)
2245 if (copy_from_user(&csigset, sigmask_arg->sigset,
2248 sigset_from_compat(&sigset, &csigset);
2249 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2251 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2255 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2263 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2264 int (*accessor)(struct kvm_device *dev,
2265 struct kvm_device_attr *attr),
2268 struct kvm_device_attr attr;
2273 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2276 return accessor(dev, &attr);
2279 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2282 struct kvm_device *dev = filp->private_data;
2285 case KVM_SET_DEVICE_ATTR:
2286 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2287 case KVM_GET_DEVICE_ATTR:
2288 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2289 case KVM_HAS_DEVICE_ATTR:
2290 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2292 if (dev->ops->ioctl)
2293 return dev->ops->ioctl(dev, ioctl, arg);
2299 static int kvm_device_release(struct inode *inode, struct file *filp)
2301 struct kvm_device *dev = filp->private_data;
2302 struct kvm *kvm = dev->kvm;
2308 static const struct file_operations kvm_device_fops = {
2309 .unlocked_ioctl = kvm_device_ioctl,
2310 #ifdef CONFIG_COMPAT
2311 .compat_ioctl = kvm_device_ioctl,
2313 .release = kvm_device_release,
2316 struct kvm_device *kvm_device_from_filp(struct file *filp)
2318 if (filp->f_op != &kvm_device_fops)
2321 return filp->private_data;
2324 static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
2325 #ifdef CONFIG_KVM_MPIC
2326 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
2327 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
2330 #ifdef CONFIG_KVM_XICS
2331 [KVM_DEV_TYPE_XICS] = &kvm_xics_ops,
2335 int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
2337 if (type >= ARRAY_SIZE(kvm_device_ops_table))
2340 if (kvm_device_ops_table[type] != NULL)
2343 kvm_device_ops_table[type] = ops;
2347 static int kvm_ioctl_create_device(struct kvm *kvm,
2348 struct kvm_create_device *cd)
2350 struct kvm_device_ops *ops = NULL;
2351 struct kvm_device *dev;
2352 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2355 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
2358 ops = kvm_device_ops_table[cd->type];
2365 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2372 ret = ops->create(dev, cd->type);
2378 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2384 list_add(&dev->vm_node, &kvm->devices);
2390 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2393 case KVM_CAP_USER_MEMORY:
2394 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2395 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2396 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2397 case KVM_CAP_SET_BOOT_CPU_ID:
2399 case KVM_CAP_INTERNAL_ERROR_DATA:
2400 #ifdef CONFIG_HAVE_KVM_MSI
2401 case KVM_CAP_SIGNAL_MSI:
2403 #ifdef CONFIG_HAVE_KVM_IRQFD
2404 case KVM_CAP_IRQFD_RESAMPLE:
2406 case KVM_CAP_CHECK_EXTENSION_VM:
2408 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2409 case KVM_CAP_IRQ_ROUTING:
2410 return KVM_MAX_IRQ_ROUTES;
2415 return kvm_vm_ioctl_check_extension(kvm, arg);
2418 static long kvm_vm_ioctl(struct file *filp,
2419 unsigned int ioctl, unsigned long arg)
2421 struct kvm *kvm = filp->private_data;
2422 void __user *argp = (void __user *)arg;
2425 if (kvm->mm != current->mm)
2428 case KVM_CREATE_VCPU:
2429 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2431 case KVM_SET_USER_MEMORY_REGION: {
2432 struct kvm_userspace_memory_region kvm_userspace_mem;
2435 if (copy_from_user(&kvm_userspace_mem, argp,
2436 sizeof kvm_userspace_mem))
2439 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2442 case KVM_GET_DIRTY_LOG: {
2443 struct kvm_dirty_log log;
2446 if (copy_from_user(&log, argp, sizeof log))
2448 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2451 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2452 case KVM_REGISTER_COALESCED_MMIO: {
2453 struct kvm_coalesced_mmio_zone zone;
2455 if (copy_from_user(&zone, argp, sizeof zone))
2457 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2460 case KVM_UNREGISTER_COALESCED_MMIO: {
2461 struct kvm_coalesced_mmio_zone zone;
2463 if (copy_from_user(&zone, argp, sizeof zone))
2465 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2470 struct kvm_irqfd data;
2473 if (copy_from_user(&data, argp, sizeof data))
2475 r = kvm_irqfd(kvm, &data);
2478 case KVM_IOEVENTFD: {
2479 struct kvm_ioeventfd data;
2482 if (copy_from_user(&data, argp, sizeof data))
2484 r = kvm_ioeventfd(kvm, &data);
2487 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2488 case KVM_SET_BOOT_CPU_ID:
2490 mutex_lock(&kvm->lock);
2491 if (atomic_read(&kvm->online_vcpus) != 0)
2494 kvm->bsp_vcpu_id = arg;
2495 mutex_unlock(&kvm->lock);
2498 #ifdef CONFIG_HAVE_KVM_MSI
2499 case KVM_SIGNAL_MSI: {
2503 if (copy_from_user(&msi, argp, sizeof msi))
2505 r = kvm_send_userspace_msi(kvm, &msi);
2509 #ifdef __KVM_HAVE_IRQ_LINE
2510 case KVM_IRQ_LINE_STATUS:
2511 case KVM_IRQ_LINE: {
2512 struct kvm_irq_level irq_event;
2515 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2518 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
2519 ioctl == KVM_IRQ_LINE_STATUS);
2524 if (ioctl == KVM_IRQ_LINE_STATUS) {
2525 if (copy_to_user(argp, &irq_event, sizeof irq_event))
2533 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2534 case KVM_SET_GSI_ROUTING: {
2535 struct kvm_irq_routing routing;
2536 struct kvm_irq_routing __user *urouting;
2537 struct kvm_irq_routing_entry *entries;
2540 if (copy_from_user(&routing, argp, sizeof(routing)))
2543 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2548 entries = vmalloc(routing.nr * sizeof(*entries));
2553 if (copy_from_user(entries, urouting->entries,
2554 routing.nr * sizeof(*entries)))
2555 goto out_free_irq_routing;
2556 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2558 out_free_irq_routing:
2562 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
2563 case KVM_CREATE_DEVICE: {
2564 struct kvm_create_device cd;
2567 if (copy_from_user(&cd, argp, sizeof(cd)))
2570 r = kvm_ioctl_create_device(kvm, &cd);
2575 if (copy_to_user(argp, &cd, sizeof(cd)))
2581 case KVM_CHECK_EXTENSION:
2582 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
2585 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2587 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2593 #ifdef CONFIG_COMPAT
2594 struct compat_kvm_dirty_log {
2598 compat_uptr_t dirty_bitmap; /* one bit per page */
2603 static long kvm_vm_compat_ioctl(struct file *filp,
2604 unsigned int ioctl, unsigned long arg)
2606 struct kvm *kvm = filp->private_data;
2609 if (kvm->mm != current->mm)
2612 case KVM_GET_DIRTY_LOG: {
2613 struct compat_kvm_dirty_log compat_log;
2614 struct kvm_dirty_log log;
2617 if (copy_from_user(&compat_log, (void __user *)arg,
2618 sizeof(compat_log)))
2620 log.slot = compat_log.slot;
2621 log.padding1 = compat_log.padding1;
2622 log.padding2 = compat_log.padding2;
2623 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2625 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2629 r = kvm_vm_ioctl(filp, ioctl, arg);
2637 static struct file_operations kvm_vm_fops = {
2638 .release = kvm_vm_release,
2639 .unlocked_ioctl = kvm_vm_ioctl,
2640 #ifdef CONFIG_COMPAT
2641 .compat_ioctl = kvm_vm_compat_ioctl,
2643 .llseek = noop_llseek,
2646 static int kvm_dev_ioctl_create_vm(unsigned long type)
2651 kvm = kvm_create_vm(type);
2653 return PTR_ERR(kvm);
2654 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2655 r = kvm_coalesced_mmio_init(kvm);
2661 r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC);
2668 static long kvm_dev_ioctl(struct file *filp,
2669 unsigned int ioctl, unsigned long arg)
2674 case KVM_GET_API_VERSION:
2677 r = KVM_API_VERSION;
2680 r = kvm_dev_ioctl_create_vm(arg);
2682 case KVM_CHECK_EXTENSION:
2683 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
2685 case KVM_GET_VCPU_MMAP_SIZE:
2688 r = PAGE_SIZE; /* struct kvm_run */
2690 r += PAGE_SIZE; /* pio data page */
2692 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2693 r += PAGE_SIZE; /* coalesced mmio ring page */
2696 case KVM_TRACE_ENABLE:
2697 case KVM_TRACE_PAUSE:
2698 case KVM_TRACE_DISABLE:
2702 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2708 static struct file_operations kvm_chardev_ops = {
2709 .unlocked_ioctl = kvm_dev_ioctl,
2710 .compat_ioctl = kvm_dev_ioctl,
2711 .llseek = noop_llseek,
2714 static struct miscdevice kvm_dev = {
2720 static void hardware_enable_nolock(void *junk)
2722 int cpu = raw_smp_processor_id();
2725 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2728 cpumask_set_cpu(cpu, cpus_hardware_enabled);
2730 r = kvm_arch_hardware_enable();
2733 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2734 atomic_inc(&hardware_enable_failed);
2735 printk(KERN_INFO "kvm: enabling virtualization on "
2736 "CPU%d failed\n", cpu);
2740 static void hardware_enable(void)
2742 raw_spin_lock(&kvm_count_lock);
2743 if (kvm_usage_count)
2744 hardware_enable_nolock(NULL);
2745 raw_spin_unlock(&kvm_count_lock);
2748 static void hardware_disable_nolock(void *junk)
2750 int cpu = raw_smp_processor_id();
2752 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2754 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2755 kvm_arch_hardware_disable();
2758 static void hardware_disable(void)
2760 raw_spin_lock(&kvm_count_lock);
2761 if (kvm_usage_count)
2762 hardware_disable_nolock(NULL);
2763 raw_spin_unlock(&kvm_count_lock);
2766 static void hardware_disable_all_nolock(void)
2768 BUG_ON(!kvm_usage_count);
2771 if (!kvm_usage_count)
2772 on_each_cpu(hardware_disable_nolock, NULL, 1);
2775 static void hardware_disable_all(void)
2777 raw_spin_lock(&kvm_count_lock);
2778 hardware_disable_all_nolock();
2779 raw_spin_unlock(&kvm_count_lock);
2782 static int hardware_enable_all(void)
2786 raw_spin_lock(&kvm_count_lock);
2789 if (kvm_usage_count == 1) {
2790 atomic_set(&hardware_enable_failed, 0);
2791 on_each_cpu(hardware_enable_nolock, NULL, 1);
2793 if (atomic_read(&hardware_enable_failed)) {
2794 hardware_disable_all_nolock();
2799 raw_spin_unlock(&kvm_count_lock);
2804 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2809 val &= ~CPU_TASKS_FROZEN;
2812 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2817 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2825 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2829 * Some (well, at least mine) BIOSes hang on reboot if
2832 * And Intel TXT required VMX off for all cpu when system shutdown.
2834 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2835 kvm_rebooting = true;
2836 on_each_cpu(hardware_disable_nolock, NULL, 1);
2840 static struct notifier_block kvm_reboot_notifier = {
2841 .notifier_call = kvm_reboot,
2845 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2849 for (i = 0; i < bus->dev_count; i++) {
2850 struct kvm_io_device *pos = bus->range[i].dev;
2852 kvm_iodevice_destructor(pos);
2857 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
2858 const struct kvm_io_range *r2)
2860 if (r1->addr < r2->addr)
2862 if (r1->addr + r1->len > r2->addr + r2->len)
2867 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2869 return kvm_io_bus_cmp(p1, p2);
2872 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2873 gpa_t addr, int len)
2875 bus->range[bus->dev_count++] = (struct kvm_io_range) {
2881 sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2882 kvm_io_bus_sort_cmp, NULL);
2887 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2888 gpa_t addr, int len)
2890 struct kvm_io_range *range, key;
2893 key = (struct kvm_io_range) {
2898 range = bsearch(&key, bus->range, bus->dev_count,
2899 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2903 off = range - bus->range;
2905 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
2911 static int __kvm_io_bus_write(struct kvm_io_bus *bus,
2912 struct kvm_io_range *range, const void *val)
2916 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2920 while (idx < bus->dev_count &&
2921 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
2922 if (!kvm_iodevice_write(bus->range[idx].dev, range->addr,
2931 /* kvm_io_bus_write - called under kvm->slots_lock */
2932 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2933 int len, const void *val)
2935 struct kvm_io_bus *bus;
2936 struct kvm_io_range range;
2939 range = (struct kvm_io_range) {
2944 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2945 r = __kvm_io_bus_write(bus, &range, val);
2946 return r < 0 ? r : 0;
2949 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
2950 int kvm_io_bus_write_cookie(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2951 int len, const void *val, long cookie)
2953 struct kvm_io_bus *bus;
2954 struct kvm_io_range range;
2956 range = (struct kvm_io_range) {
2961 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2963 /* First try the device referenced by cookie. */
2964 if ((cookie >= 0) && (cookie < bus->dev_count) &&
2965 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
2966 if (!kvm_iodevice_write(bus->range[cookie].dev, addr, len,
2971 * cookie contained garbage; fall back to search and return the
2972 * correct cookie value.
2974 return __kvm_io_bus_write(bus, &range, val);
2977 static int __kvm_io_bus_read(struct kvm_io_bus *bus, struct kvm_io_range *range,
2982 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2986 while (idx < bus->dev_count &&
2987 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
2988 if (!kvm_iodevice_read(bus->range[idx].dev, range->addr,
2996 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
2998 /* kvm_io_bus_read - called under kvm->slots_lock */
2999 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3002 struct kvm_io_bus *bus;
3003 struct kvm_io_range range;
3006 range = (struct kvm_io_range) {
3011 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
3012 r = __kvm_io_bus_read(bus, &range, val);
3013 return r < 0 ? r : 0;
3017 /* Caller must hold slots_lock. */
3018 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3019 int len, struct kvm_io_device *dev)
3021 struct kvm_io_bus *new_bus, *bus;
3023 bus = kvm->buses[bus_idx];
3024 /* exclude ioeventfd which is limited by maximum fd */
3025 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3028 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
3029 sizeof(struct kvm_io_range)), GFP_KERNEL);
3032 memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
3033 sizeof(struct kvm_io_range)));
3034 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
3035 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3036 synchronize_srcu_expedited(&kvm->srcu);
3042 /* Caller must hold slots_lock. */
3043 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3044 struct kvm_io_device *dev)
3047 struct kvm_io_bus *new_bus, *bus;
3049 bus = kvm->buses[bus_idx];
3051 for (i = 0; i < bus->dev_count; i++)
3052 if (bus->range[i].dev == dev) {
3060 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3061 sizeof(struct kvm_io_range)), GFP_KERNEL);
3065 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3066 new_bus->dev_count--;
3067 memcpy(new_bus->range + i, bus->range + i + 1,
3068 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3070 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3071 synchronize_srcu_expedited(&kvm->srcu);
3076 static struct notifier_block kvm_cpu_notifier = {
3077 .notifier_call = kvm_cpu_hotplug,
3080 static int vm_stat_get(void *_offset, u64 *val)
3082 unsigned offset = (long)_offset;
3086 spin_lock(&kvm_lock);
3087 list_for_each_entry(kvm, &vm_list, vm_list)
3088 *val += *(u32 *)((void *)kvm + offset);
3089 spin_unlock(&kvm_lock);
3093 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
3095 static int vcpu_stat_get(void *_offset, u64 *val)
3097 unsigned offset = (long)_offset;
3099 struct kvm_vcpu *vcpu;
3103 spin_lock(&kvm_lock);
3104 list_for_each_entry(kvm, &vm_list, vm_list)
3105 kvm_for_each_vcpu(i, vcpu, kvm)
3106 *val += *(u32 *)((void *)vcpu + offset);
3108 spin_unlock(&kvm_lock);
3112 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
3114 static const struct file_operations *stat_fops[] = {
3115 [KVM_STAT_VCPU] = &vcpu_stat_fops,
3116 [KVM_STAT_VM] = &vm_stat_fops,
3119 static int kvm_init_debug(void)
3122 struct kvm_stats_debugfs_item *p;
3124 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3125 if (kvm_debugfs_dir == NULL)
3128 for (p = debugfs_entries; p->name; ++p) {
3129 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
3130 (void *)(long)p->offset,
3131 stat_fops[p->kind]);
3132 if (p->dentry == NULL)
3139 debugfs_remove_recursive(kvm_debugfs_dir);
3144 static void kvm_exit_debug(void)
3146 struct kvm_stats_debugfs_item *p;
3148 for (p = debugfs_entries; p->name; ++p)
3149 debugfs_remove(p->dentry);
3150 debugfs_remove(kvm_debugfs_dir);
3153 static int kvm_suspend(void)
3155 if (kvm_usage_count)
3156 hardware_disable_nolock(NULL);
3160 static void kvm_resume(void)
3162 if (kvm_usage_count) {
3163 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3164 hardware_enable_nolock(NULL);
3168 static struct syscore_ops kvm_syscore_ops = {
3169 .suspend = kvm_suspend,
3170 .resume = kvm_resume,
3174 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3176 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3179 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3181 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3182 if (vcpu->preempted)
3183 vcpu->preempted = false;
3185 kvm_arch_sched_in(vcpu, cpu);
3187 kvm_arch_vcpu_load(vcpu, cpu);
3190 static void kvm_sched_out(struct preempt_notifier *pn,
3191 struct task_struct *next)
3193 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3195 if (current->state == TASK_RUNNING)
3196 vcpu->preempted = true;
3197 kvm_arch_vcpu_put(vcpu);
3200 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3201 struct module *module)
3206 r = kvm_arch_init(opaque);
3211 * kvm_arch_init makes sure there's at most one caller
3212 * for architectures that support multiple implementations,
3213 * like intel and amd on x86.
3214 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3215 * conflicts in case kvm is already setup for another implementation.
3217 r = kvm_irqfd_init();
3221 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
3226 r = kvm_arch_hardware_setup();
3230 for_each_online_cpu(cpu) {
3231 smp_call_function_single(cpu,
3232 kvm_arch_check_processor_compat,
3238 r = register_cpu_notifier(&kvm_cpu_notifier);
3241 register_reboot_notifier(&kvm_reboot_notifier);
3243 /* A kmem cache lets us meet the alignment requirements of fx_save. */
3245 vcpu_align = __alignof__(struct kvm_vcpu);
3246 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
3248 if (!kvm_vcpu_cache) {
3253 r = kvm_async_pf_init();
3257 kvm_chardev_ops.owner = module;
3258 kvm_vm_fops.owner = module;
3259 kvm_vcpu_fops.owner = module;
3261 r = misc_register(&kvm_dev);
3263 printk(KERN_ERR "kvm: misc device register failed\n");
3267 register_syscore_ops(&kvm_syscore_ops);
3269 kvm_preempt_ops.sched_in = kvm_sched_in;
3270 kvm_preempt_ops.sched_out = kvm_sched_out;
3272 r = kvm_init_debug();
3274 printk(KERN_ERR "kvm: create debugfs files failed\n");
3278 r = kvm_vfio_ops_init();
3284 unregister_syscore_ops(&kvm_syscore_ops);
3285 misc_deregister(&kvm_dev);
3287 kvm_async_pf_deinit();
3289 kmem_cache_destroy(kvm_vcpu_cache);
3291 unregister_reboot_notifier(&kvm_reboot_notifier);
3292 unregister_cpu_notifier(&kvm_cpu_notifier);
3295 kvm_arch_hardware_unsetup();
3297 free_cpumask_var(cpus_hardware_enabled);
3305 EXPORT_SYMBOL_GPL(kvm_init);
3310 misc_deregister(&kvm_dev);
3311 kmem_cache_destroy(kvm_vcpu_cache);
3312 kvm_async_pf_deinit();
3313 unregister_syscore_ops(&kvm_syscore_ops);
3314 unregister_reboot_notifier(&kvm_reboot_notifier);
3315 unregister_cpu_notifier(&kvm_cpu_notifier);
3316 on_each_cpu(hardware_disable_nolock, NULL, 1);
3317 kvm_arch_hardware_unsetup();
3320 free_cpumask_var(cpus_hardware_enabled);
3322 EXPORT_SYMBOL_GPL(kvm_exit);