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.
10 * Avi Kivity <avi@qumranet.com>
11 * Yaniv Kamay <yaniv@qumranet.com>
13 * This work is licensed under the terms of the GNU GPL, version 2. See
14 * the COPYING file in the top-level directory.
20 #include <linux/kvm_host.h>
21 #include <linux/kvm.h>
22 #include <linux/module.h>
23 #include <linux/errno.h>
24 #include <linux/percpu.h>
26 #include <linux/miscdevice.h>
27 #include <linux/vmalloc.h>
28 #include <linux/reboot.h>
29 #include <linux/debugfs.h>
30 #include <linux/highmem.h>
31 #include <linux/file.h>
32 #include <linux/sysdev.h>
33 #include <linux/cpu.h>
34 #include <linux/sched.h>
35 #include <linux/cpumask.h>
36 #include <linux/smp.h>
37 #include <linux/anon_inodes.h>
38 #include <linux/profile.h>
39 #include <linux/kvm_para.h>
40 #include <linux/pagemap.h>
41 #include <linux/mman.h>
42 #include <linux/swap.h>
43 #include <linux/bitops.h>
44 #include <linux/spinlock.h>
45 #include <linux/compat.h>
46 #include <linux/srcu.h>
47 #include <linux/hugetlb.h>
48 #include <linux/slab.h>
50 #include <asm/processor.h>
52 #include <asm/uaccess.h>
53 #include <asm/pgtable.h>
54 #include <asm-generic/bitops/le.h>
56 #include "coalesced_mmio.h"
58 #define CREATE_TRACE_POINTS
59 #include <trace/events/kvm.h>
61 MODULE_AUTHOR("Qumranet");
62 MODULE_LICENSE("GPL");
67 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
70 DEFINE_SPINLOCK(kvm_lock);
73 static cpumask_var_t cpus_hardware_enabled;
74 static int kvm_usage_count = 0;
75 static atomic_t hardware_enable_failed;
77 struct kmem_cache *kvm_vcpu_cache;
78 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
80 static __read_mostly struct preempt_ops kvm_preempt_ops;
82 struct dentry *kvm_debugfs_dir;
84 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
86 static int hardware_enable_all(void);
87 static void hardware_disable_all(void);
89 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
91 static bool kvm_rebooting;
93 static bool largepages_enabled = true;
95 inline int kvm_is_mmio_pfn(pfn_t pfn)
98 struct page *page = compound_head(pfn_to_page(pfn));
99 return PageReserved(page);
106 * Switches to specified vcpu, until a matching vcpu_put()
108 void vcpu_load(struct kvm_vcpu *vcpu)
112 mutex_lock(&vcpu->mutex);
114 preempt_notifier_register(&vcpu->preempt_notifier);
115 kvm_arch_vcpu_load(vcpu, cpu);
119 void vcpu_put(struct kvm_vcpu *vcpu)
122 kvm_arch_vcpu_put(vcpu);
123 preempt_notifier_unregister(&vcpu->preempt_notifier);
125 mutex_unlock(&vcpu->mutex);
128 static void ack_flush(void *_completed)
132 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
137 struct kvm_vcpu *vcpu;
139 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
141 raw_spin_lock(&kvm->requests_lock);
142 me = smp_processor_id();
143 kvm_for_each_vcpu(i, vcpu, kvm) {
144 if (test_and_set_bit(req, &vcpu->requests))
147 if (cpus != NULL && cpu != -1 && cpu != me)
148 cpumask_set_cpu(cpu, cpus);
150 if (unlikely(cpus == NULL))
151 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
152 else if (!cpumask_empty(cpus))
153 smp_call_function_many(cpus, ack_flush, NULL, 1);
156 raw_spin_unlock(&kvm->requests_lock);
157 free_cpumask_var(cpus);
161 void kvm_flush_remote_tlbs(struct kvm *kvm)
163 if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
164 ++kvm->stat.remote_tlb_flush;
167 void kvm_reload_remote_mmus(struct kvm *kvm)
169 make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
172 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
177 mutex_init(&vcpu->mutex);
181 init_waitqueue_head(&vcpu->wq);
183 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
188 vcpu->run = page_address(page);
190 r = kvm_arch_vcpu_init(vcpu);
196 free_page((unsigned long)vcpu->run);
200 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
202 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
204 kvm_arch_vcpu_uninit(vcpu);
205 free_page((unsigned long)vcpu->run);
207 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
209 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
210 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
212 return container_of(mn, struct kvm, mmu_notifier);
215 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
216 struct mm_struct *mm,
217 unsigned long address)
219 struct kvm *kvm = mmu_notifier_to_kvm(mn);
220 int need_tlb_flush, idx;
223 * When ->invalidate_page runs, the linux pte has been zapped
224 * already but the page is still allocated until
225 * ->invalidate_page returns. So if we increase the sequence
226 * here the kvm page fault will notice if the spte can't be
227 * established because the page is going to be freed. If
228 * instead the kvm page fault establishes the spte before
229 * ->invalidate_page runs, kvm_unmap_hva will release it
232 * The sequence increase only need to be seen at spin_unlock
233 * time, and not at spin_lock time.
235 * Increasing the sequence after the spin_unlock would be
236 * unsafe because the kvm page fault could then establish the
237 * pte after kvm_unmap_hva returned, without noticing the page
238 * is going to be freed.
240 idx = srcu_read_lock(&kvm->srcu);
241 spin_lock(&kvm->mmu_lock);
242 kvm->mmu_notifier_seq++;
243 need_tlb_flush = kvm_unmap_hva(kvm, address);
244 spin_unlock(&kvm->mmu_lock);
245 srcu_read_unlock(&kvm->srcu, idx);
247 /* we've to flush the tlb before the pages can be freed */
249 kvm_flush_remote_tlbs(kvm);
253 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
254 struct mm_struct *mm,
255 unsigned long address,
258 struct kvm *kvm = mmu_notifier_to_kvm(mn);
261 idx = srcu_read_lock(&kvm->srcu);
262 spin_lock(&kvm->mmu_lock);
263 kvm->mmu_notifier_seq++;
264 kvm_set_spte_hva(kvm, address, pte);
265 spin_unlock(&kvm->mmu_lock);
266 srcu_read_unlock(&kvm->srcu, idx);
269 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
270 struct mm_struct *mm,
274 struct kvm *kvm = mmu_notifier_to_kvm(mn);
275 int need_tlb_flush = 0, idx;
277 idx = srcu_read_lock(&kvm->srcu);
278 spin_lock(&kvm->mmu_lock);
280 * The count increase must become visible at unlock time as no
281 * spte can be established without taking the mmu_lock and
282 * count is also read inside the mmu_lock critical section.
284 kvm->mmu_notifier_count++;
285 for (; start < end; start += PAGE_SIZE)
286 need_tlb_flush |= kvm_unmap_hva(kvm, start);
287 spin_unlock(&kvm->mmu_lock);
288 srcu_read_unlock(&kvm->srcu, idx);
290 /* we've to flush the tlb before the pages can be freed */
292 kvm_flush_remote_tlbs(kvm);
295 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
296 struct mm_struct *mm,
300 struct kvm *kvm = mmu_notifier_to_kvm(mn);
302 spin_lock(&kvm->mmu_lock);
304 * This sequence increase will notify the kvm page fault that
305 * the page that is going to be mapped in the spte could have
308 kvm->mmu_notifier_seq++;
310 * The above sequence increase must be visible before the
311 * below count decrease but both values are read by the kvm
312 * page fault under mmu_lock spinlock so we don't need to add
313 * a smb_wmb() here in between the two.
315 kvm->mmu_notifier_count--;
316 spin_unlock(&kvm->mmu_lock);
318 BUG_ON(kvm->mmu_notifier_count < 0);
321 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
322 struct mm_struct *mm,
323 unsigned long address)
325 struct kvm *kvm = mmu_notifier_to_kvm(mn);
328 idx = srcu_read_lock(&kvm->srcu);
329 spin_lock(&kvm->mmu_lock);
330 young = kvm_age_hva(kvm, address);
331 spin_unlock(&kvm->mmu_lock);
332 srcu_read_unlock(&kvm->srcu, idx);
335 kvm_flush_remote_tlbs(kvm);
340 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
341 struct mm_struct *mm)
343 struct kvm *kvm = mmu_notifier_to_kvm(mn);
346 idx = srcu_read_lock(&kvm->srcu);
347 kvm_arch_flush_shadow(kvm);
348 srcu_read_unlock(&kvm->srcu, idx);
351 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
352 .invalidate_page = kvm_mmu_notifier_invalidate_page,
353 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
354 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
355 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
356 .change_pte = kvm_mmu_notifier_change_pte,
357 .release = kvm_mmu_notifier_release,
360 static int kvm_init_mmu_notifier(struct kvm *kvm)
362 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
363 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
366 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
368 static int kvm_init_mmu_notifier(struct kvm *kvm)
373 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
375 static struct kvm *kvm_create_vm(void)
378 struct kvm *kvm = kvm_arch_create_vm();
383 r = hardware_enable_all();
385 goto out_err_nodisable;
387 #ifdef CONFIG_HAVE_KVM_IRQCHIP
388 INIT_HLIST_HEAD(&kvm->mask_notifier_list);
389 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
393 kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
396 if (init_srcu_struct(&kvm->srcu))
398 for (i = 0; i < KVM_NR_BUSES; i++) {
399 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
401 if (!kvm->buses[i]) {
402 cleanup_srcu_struct(&kvm->srcu);
407 r = kvm_init_mmu_notifier(kvm);
409 cleanup_srcu_struct(&kvm->srcu);
413 kvm->mm = current->mm;
414 atomic_inc(&kvm->mm->mm_count);
415 spin_lock_init(&kvm->mmu_lock);
416 raw_spin_lock_init(&kvm->requests_lock);
417 kvm_eventfd_init(kvm);
418 mutex_init(&kvm->lock);
419 mutex_init(&kvm->irq_lock);
420 mutex_init(&kvm->slots_lock);
421 atomic_set(&kvm->users_count, 1);
422 spin_lock(&kvm_lock);
423 list_add(&kvm->vm_list, &vm_list);
424 spin_unlock(&kvm_lock);
429 hardware_disable_all();
431 for (i = 0; i < KVM_NR_BUSES; i++)
432 kfree(kvm->buses[i]);
433 kfree(kvm->memslots);
439 * Free any memory in @free but not in @dont.
441 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
442 struct kvm_memory_slot *dont)
446 if (!dont || free->rmap != dont->rmap)
449 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
450 vfree(free->dirty_bitmap);
453 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
454 if (!dont || free->lpage_info[i] != dont->lpage_info[i]) {
455 vfree(free->lpage_info[i]);
456 free->lpage_info[i] = NULL;
461 free->dirty_bitmap = NULL;
465 void kvm_free_physmem(struct kvm *kvm)
468 struct kvm_memslots *slots = kvm->memslots;
470 for (i = 0; i < slots->nmemslots; ++i)
471 kvm_free_physmem_slot(&slots->memslots[i], NULL);
473 kfree(kvm->memslots);
476 static void kvm_destroy_vm(struct kvm *kvm)
479 struct mm_struct *mm = kvm->mm;
481 kvm_arch_sync_events(kvm);
482 spin_lock(&kvm_lock);
483 list_del(&kvm->vm_list);
484 spin_unlock(&kvm_lock);
485 kvm_free_irq_routing(kvm);
486 for (i = 0; i < KVM_NR_BUSES; i++)
487 kvm_io_bus_destroy(kvm->buses[i]);
488 kvm_coalesced_mmio_free(kvm);
489 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
490 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
492 kvm_arch_flush_shadow(kvm);
494 kvm_arch_destroy_vm(kvm);
495 hardware_disable_all();
499 void kvm_get_kvm(struct kvm *kvm)
501 atomic_inc(&kvm->users_count);
503 EXPORT_SYMBOL_GPL(kvm_get_kvm);
505 void kvm_put_kvm(struct kvm *kvm)
507 if (atomic_dec_and_test(&kvm->users_count))
510 EXPORT_SYMBOL_GPL(kvm_put_kvm);
513 static int kvm_vm_release(struct inode *inode, struct file *filp)
515 struct kvm *kvm = filp->private_data;
517 kvm_irqfd_release(kvm);
524 * Allocate some memory and give it an address in the guest physical address
527 * Discontiguous memory is allowed, mostly for framebuffers.
529 * Must be called holding mmap_sem for write.
531 int __kvm_set_memory_region(struct kvm *kvm,
532 struct kvm_userspace_memory_region *mem,
535 int r, flush_shadow = 0;
537 unsigned long npages;
539 struct kvm_memory_slot *memslot;
540 struct kvm_memory_slot old, new;
541 struct kvm_memslots *slots, *old_memslots;
544 /* General sanity checks */
545 if (mem->memory_size & (PAGE_SIZE - 1))
547 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
549 if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
551 if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
553 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
556 memslot = &kvm->memslots->memslots[mem->slot];
557 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
558 npages = mem->memory_size >> PAGE_SHIFT;
561 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
563 new = old = *memslot;
565 new.base_gfn = base_gfn;
567 new.flags = mem->flags;
569 /* Disallow changing a memory slot's size. */
571 if (npages && old.npages && npages != old.npages)
574 /* Check for overlaps */
576 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
577 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
579 if (s == memslot || !s->npages)
581 if (!((base_gfn + npages <= s->base_gfn) ||
582 (base_gfn >= s->base_gfn + s->npages)))
586 /* Free page dirty bitmap if unneeded */
587 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
588 new.dirty_bitmap = NULL;
592 /* Allocate if a slot is being created */
594 if (npages && !new.rmap) {
595 new.rmap = vmalloc(npages * sizeof(struct page *));
600 memset(new.rmap, 0, npages * sizeof(*new.rmap));
602 new.user_alloc = user_alloc;
603 new.userspace_addr = mem->userspace_addr;
608 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
614 /* Avoid unused variable warning if no large pages */
617 if (new.lpage_info[i])
620 lpages = 1 + (base_gfn + npages - 1) /
621 KVM_PAGES_PER_HPAGE(level);
622 lpages -= base_gfn / KVM_PAGES_PER_HPAGE(level);
624 new.lpage_info[i] = vmalloc(lpages * sizeof(*new.lpage_info[i]));
626 if (!new.lpage_info[i])
629 memset(new.lpage_info[i], 0,
630 lpages * sizeof(*new.lpage_info[i]));
632 if (base_gfn % KVM_PAGES_PER_HPAGE(level))
633 new.lpage_info[i][0].write_count = 1;
634 if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE(level))
635 new.lpage_info[i][lpages - 1].write_count = 1;
636 ugfn = new.userspace_addr >> PAGE_SHIFT;
638 * If the gfn and userspace address are not aligned wrt each
639 * other, or if explicitly asked to, disable large page
640 * support for this slot
642 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
644 for (j = 0; j < lpages; ++j)
645 new.lpage_info[i][j].write_count = 1;
650 /* Allocate page dirty bitmap if needed */
651 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
652 unsigned long dirty_bytes = kvm_dirty_bitmap_bytes(&new);
654 new.dirty_bitmap = vmalloc(dirty_bytes);
655 if (!new.dirty_bitmap)
657 memset(new.dirty_bitmap, 0, dirty_bytes);
658 /* destroy any largepage mappings for dirty tracking */
662 #else /* not defined CONFIG_S390 */
663 new.user_alloc = user_alloc;
665 new.userspace_addr = mem->userspace_addr;
666 #endif /* not defined CONFIG_S390 */
670 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
673 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
674 if (mem->slot >= slots->nmemslots)
675 slots->nmemslots = mem->slot + 1;
676 slots->memslots[mem->slot].flags |= KVM_MEMSLOT_INVALID;
678 old_memslots = kvm->memslots;
679 rcu_assign_pointer(kvm->memslots, slots);
680 synchronize_srcu_expedited(&kvm->srcu);
681 /* From this point no new shadow pages pointing to a deleted
682 * memslot will be created.
684 * validation of sp->gfn happens in:
685 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
686 * - kvm_is_visible_gfn (mmu_check_roots)
688 kvm_arch_flush_shadow(kvm);
692 r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
697 /* map the pages in iommu page table */
699 r = kvm_iommu_map_pages(kvm, &new);
706 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
709 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
710 if (mem->slot >= slots->nmemslots)
711 slots->nmemslots = mem->slot + 1;
713 /* actual memory is freed via old in kvm_free_physmem_slot below */
716 new.dirty_bitmap = NULL;
717 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i)
718 new.lpage_info[i] = NULL;
721 slots->memslots[mem->slot] = new;
722 old_memslots = kvm->memslots;
723 rcu_assign_pointer(kvm->memslots, slots);
724 synchronize_srcu_expedited(&kvm->srcu);
726 kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
728 kvm_free_physmem_slot(&old, &new);
732 kvm_arch_flush_shadow(kvm);
737 kvm_free_physmem_slot(&new, &old);
742 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
744 int kvm_set_memory_region(struct kvm *kvm,
745 struct kvm_userspace_memory_region *mem,
750 mutex_lock(&kvm->slots_lock);
751 r = __kvm_set_memory_region(kvm, mem, user_alloc);
752 mutex_unlock(&kvm->slots_lock);
755 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
757 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
759 kvm_userspace_memory_region *mem,
762 if (mem->slot >= KVM_MEMORY_SLOTS)
764 return kvm_set_memory_region(kvm, mem, user_alloc);
767 int kvm_get_dirty_log(struct kvm *kvm,
768 struct kvm_dirty_log *log, int *is_dirty)
770 struct kvm_memory_slot *memslot;
773 unsigned long any = 0;
776 if (log->slot >= KVM_MEMORY_SLOTS)
779 memslot = &kvm->memslots->memslots[log->slot];
781 if (!memslot->dirty_bitmap)
784 n = kvm_dirty_bitmap_bytes(memslot);
786 for (i = 0; !any && i < n/sizeof(long); ++i)
787 any = memslot->dirty_bitmap[i];
790 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
801 void kvm_disable_largepages(void)
803 largepages_enabled = false;
805 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
807 int is_error_page(struct page *page)
809 return page == bad_page;
811 EXPORT_SYMBOL_GPL(is_error_page);
813 int is_error_pfn(pfn_t pfn)
815 return pfn == bad_pfn;
817 EXPORT_SYMBOL_GPL(is_error_pfn);
819 static inline unsigned long bad_hva(void)
824 int kvm_is_error_hva(unsigned long addr)
826 return addr == bad_hva();
828 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
830 struct kvm_memory_slot *gfn_to_memslot_unaliased(struct kvm *kvm, gfn_t gfn)
833 struct kvm_memslots *slots = rcu_dereference(kvm->memslots);
835 for (i = 0; i < slots->nmemslots; ++i) {
836 struct kvm_memory_slot *memslot = &slots->memslots[i];
838 if (gfn >= memslot->base_gfn
839 && gfn < memslot->base_gfn + memslot->npages)
844 EXPORT_SYMBOL_GPL(gfn_to_memslot_unaliased);
846 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
848 gfn = unalias_gfn(kvm, gfn);
849 return gfn_to_memslot_unaliased(kvm, gfn);
852 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
855 struct kvm_memslots *slots = rcu_dereference(kvm->memslots);
857 gfn = unalias_gfn_instantiation(kvm, gfn);
858 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
859 struct kvm_memory_slot *memslot = &slots->memslots[i];
861 if (memslot->flags & KVM_MEMSLOT_INVALID)
864 if (gfn >= memslot->base_gfn
865 && gfn < memslot->base_gfn + memslot->npages)
870 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
872 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
874 struct vm_area_struct *vma;
875 unsigned long addr, size;
879 addr = gfn_to_hva(kvm, gfn);
880 if (kvm_is_error_hva(addr))
883 down_read(¤t->mm->mmap_sem);
884 vma = find_vma(current->mm, addr);
888 size = vma_kernel_pagesize(vma);
891 up_read(¤t->mm->mmap_sem);
896 int memslot_id(struct kvm *kvm, gfn_t gfn)
899 struct kvm_memslots *slots = rcu_dereference(kvm->memslots);
900 struct kvm_memory_slot *memslot = NULL;
902 gfn = unalias_gfn(kvm, gfn);
903 for (i = 0; i < slots->nmemslots; ++i) {
904 memslot = &slots->memslots[i];
906 if (gfn >= memslot->base_gfn
907 && gfn < memslot->base_gfn + memslot->npages)
911 return memslot - slots->memslots;
914 static unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
916 return slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE;
919 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
921 struct kvm_memory_slot *slot;
923 gfn = unalias_gfn_instantiation(kvm, gfn);
924 slot = gfn_to_memslot_unaliased(kvm, gfn);
925 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
927 return gfn_to_hva_memslot(slot, gfn);
929 EXPORT_SYMBOL_GPL(gfn_to_hva);
931 static pfn_t hva_to_pfn(struct kvm *kvm, unsigned long addr)
933 struct page *page[1];
939 npages = get_user_pages_fast(addr, 1, 1, page);
941 if (unlikely(npages != 1)) {
942 struct vm_area_struct *vma;
944 down_read(¤t->mm->mmap_sem);
945 vma = find_vma(current->mm, addr);
947 if (vma == NULL || addr < vma->vm_start ||
948 !(vma->vm_flags & VM_PFNMAP)) {
949 up_read(¤t->mm->mmap_sem);
951 return page_to_pfn(bad_page);
954 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
955 up_read(¤t->mm->mmap_sem);
956 BUG_ON(!kvm_is_mmio_pfn(pfn));
958 pfn = page_to_pfn(page[0]);
963 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
967 addr = gfn_to_hva(kvm, gfn);
968 if (kvm_is_error_hva(addr)) {
970 return page_to_pfn(bad_page);
973 return hva_to_pfn(kvm, addr);
975 EXPORT_SYMBOL_GPL(gfn_to_pfn);
977 pfn_t gfn_to_pfn_memslot(struct kvm *kvm,
978 struct kvm_memory_slot *slot, gfn_t gfn)
980 unsigned long addr = gfn_to_hva_memslot(slot, gfn);
981 return hva_to_pfn(kvm, addr);
984 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
988 pfn = gfn_to_pfn(kvm, gfn);
989 if (!kvm_is_mmio_pfn(pfn))
990 return pfn_to_page(pfn);
992 WARN_ON(kvm_is_mmio_pfn(pfn));
998 EXPORT_SYMBOL_GPL(gfn_to_page);
1000 void kvm_release_page_clean(struct page *page)
1002 kvm_release_pfn_clean(page_to_pfn(page));
1004 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1006 void kvm_release_pfn_clean(pfn_t pfn)
1008 if (!kvm_is_mmio_pfn(pfn))
1009 put_page(pfn_to_page(pfn));
1011 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1013 void kvm_release_page_dirty(struct page *page)
1015 kvm_release_pfn_dirty(page_to_pfn(page));
1017 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1019 void kvm_release_pfn_dirty(pfn_t pfn)
1021 kvm_set_pfn_dirty(pfn);
1022 kvm_release_pfn_clean(pfn);
1024 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1026 void kvm_set_page_dirty(struct page *page)
1028 kvm_set_pfn_dirty(page_to_pfn(page));
1030 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1032 void kvm_set_pfn_dirty(pfn_t pfn)
1034 if (!kvm_is_mmio_pfn(pfn)) {
1035 struct page *page = pfn_to_page(pfn);
1036 if (!PageReserved(page))
1040 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1042 void kvm_set_pfn_accessed(pfn_t pfn)
1044 if (!kvm_is_mmio_pfn(pfn))
1045 mark_page_accessed(pfn_to_page(pfn));
1047 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1049 void kvm_get_pfn(pfn_t pfn)
1051 if (!kvm_is_mmio_pfn(pfn))
1052 get_page(pfn_to_page(pfn));
1054 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1056 static int next_segment(unsigned long len, int offset)
1058 if (len > PAGE_SIZE - offset)
1059 return PAGE_SIZE - offset;
1064 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1070 addr = gfn_to_hva(kvm, gfn);
1071 if (kvm_is_error_hva(addr))
1073 r = copy_from_user(data, (void __user *)addr + offset, len);
1078 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1080 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1082 gfn_t gfn = gpa >> PAGE_SHIFT;
1084 int offset = offset_in_page(gpa);
1087 while ((seg = next_segment(len, offset)) != 0) {
1088 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1098 EXPORT_SYMBOL_GPL(kvm_read_guest);
1100 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1105 gfn_t gfn = gpa >> PAGE_SHIFT;
1106 int offset = offset_in_page(gpa);
1108 addr = gfn_to_hva(kvm, gfn);
1109 if (kvm_is_error_hva(addr))
1111 pagefault_disable();
1112 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1118 EXPORT_SYMBOL(kvm_read_guest_atomic);
1120 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1121 int offset, int len)
1126 addr = gfn_to_hva(kvm, gfn);
1127 if (kvm_is_error_hva(addr))
1129 r = copy_to_user((void __user *)addr + offset, data, len);
1132 mark_page_dirty(kvm, gfn);
1135 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1137 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1140 gfn_t gfn = gpa >> PAGE_SHIFT;
1142 int offset = offset_in_page(gpa);
1145 while ((seg = next_segment(len, offset)) != 0) {
1146 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1157 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1159 return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
1161 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1163 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1165 gfn_t gfn = gpa >> PAGE_SHIFT;
1167 int offset = offset_in_page(gpa);
1170 while ((seg = next_segment(len, offset)) != 0) {
1171 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1180 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1182 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1184 struct kvm_memory_slot *memslot;
1186 gfn = unalias_gfn(kvm, gfn);
1187 memslot = gfn_to_memslot_unaliased(kvm, gfn);
1188 if (memslot && memslot->dirty_bitmap) {
1189 unsigned long rel_gfn = gfn - memslot->base_gfn;
1190 unsigned long *p = memslot->dirty_bitmap +
1191 rel_gfn / BITS_PER_LONG;
1192 int offset = rel_gfn % BITS_PER_LONG;
1195 if (!generic_test_le_bit(offset, p))
1196 generic___set_le_bit(offset, p);
1201 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1203 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1208 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1210 if (kvm_arch_vcpu_runnable(vcpu)) {
1211 set_bit(KVM_REQ_UNHALT, &vcpu->requests);
1214 if (kvm_cpu_has_pending_timer(vcpu))
1216 if (signal_pending(current))
1222 finish_wait(&vcpu->wq, &wait);
1225 void kvm_resched(struct kvm_vcpu *vcpu)
1227 if (!need_resched())
1231 EXPORT_SYMBOL_GPL(kvm_resched);
1233 void kvm_vcpu_on_spin(struct kvm_vcpu *vcpu)
1238 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1240 /* Sleep for 100 us, and hope lock-holder got scheduled */
1241 expires = ktime_add_ns(ktime_get(), 100000UL);
1242 schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
1244 finish_wait(&vcpu->wq, &wait);
1246 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1248 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1250 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1253 if (vmf->pgoff == 0)
1254 page = virt_to_page(vcpu->run);
1256 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1257 page = virt_to_page(vcpu->arch.pio_data);
1259 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1260 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1261 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1264 return VM_FAULT_SIGBUS;
1270 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1271 .fault = kvm_vcpu_fault,
1274 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1276 vma->vm_ops = &kvm_vcpu_vm_ops;
1280 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1282 struct kvm_vcpu *vcpu = filp->private_data;
1284 kvm_put_kvm(vcpu->kvm);
1288 static struct file_operations kvm_vcpu_fops = {
1289 .release = kvm_vcpu_release,
1290 .unlocked_ioctl = kvm_vcpu_ioctl,
1291 .compat_ioctl = kvm_vcpu_ioctl,
1292 .mmap = kvm_vcpu_mmap,
1296 * Allocates an inode for the vcpu.
1298 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1300 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1304 * Creates some virtual cpus. Good luck creating more than one.
1306 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1309 struct kvm_vcpu *vcpu, *v;
1311 vcpu = kvm_arch_vcpu_create(kvm, id);
1313 return PTR_ERR(vcpu);
1315 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1317 r = kvm_arch_vcpu_setup(vcpu);
1321 mutex_lock(&kvm->lock);
1322 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1327 kvm_for_each_vcpu(r, v, kvm)
1328 if (v->vcpu_id == id) {
1333 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1335 /* Now it's all set up, let userspace reach it */
1337 r = create_vcpu_fd(vcpu);
1343 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1345 atomic_inc(&kvm->online_vcpus);
1347 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1348 if (kvm->bsp_vcpu_id == id)
1349 kvm->bsp_vcpu = vcpu;
1351 mutex_unlock(&kvm->lock);
1355 mutex_unlock(&kvm->lock);
1356 kvm_arch_vcpu_destroy(vcpu);
1360 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1363 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1364 vcpu->sigset_active = 1;
1365 vcpu->sigset = *sigset;
1367 vcpu->sigset_active = 0;
1371 static long kvm_vcpu_ioctl(struct file *filp,
1372 unsigned int ioctl, unsigned long arg)
1374 struct kvm_vcpu *vcpu = filp->private_data;
1375 void __user *argp = (void __user *)arg;
1377 struct kvm_fpu *fpu = NULL;
1378 struct kvm_sregs *kvm_sregs = NULL;
1380 if (vcpu->kvm->mm != current->mm)
1387 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1389 case KVM_GET_REGS: {
1390 struct kvm_regs *kvm_regs;
1393 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1396 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1400 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1407 case KVM_SET_REGS: {
1408 struct kvm_regs *kvm_regs;
1411 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1415 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1417 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1425 case KVM_GET_SREGS: {
1426 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1430 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1434 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1439 case KVM_SET_SREGS: {
1440 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1445 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1447 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1453 case KVM_GET_MP_STATE: {
1454 struct kvm_mp_state mp_state;
1456 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1460 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1465 case KVM_SET_MP_STATE: {
1466 struct kvm_mp_state mp_state;
1469 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1471 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1477 case KVM_TRANSLATE: {
1478 struct kvm_translation tr;
1481 if (copy_from_user(&tr, argp, sizeof tr))
1483 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1487 if (copy_to_user(argp, &tr, sizeof tr))
1492 case KVM_SET_GUEST_DEBUG: {
1493 struct kvm_guest_debug dbg;
1496 if (copy_from_user(&dbg, argp, sizeof dbg))
1498 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1504 case KVM_SET_SIGNAL_MASK: {
1505 struct kvm_signal_mask __user *sigmask_arg = argp;
1506 struct kvm_signal_mask kvm_sigmask;
1507 sigset_t sigset, *p;
1512 if (copy_from_user(&kvm_sigmask, argp,
1513 sizeof kvm_sigmask))
1516 if (kvm_sigmask.len != sizeof sigset)
1519 if (copy_from_user(&sigset, sigmask_arg->sigset,
1524 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1528 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1532 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1536 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1542 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1547 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1549 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1556 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1564 static long kvm_vm_ioctl(struct file *filp,
1565 unsigned int ioctl, unsigned long arg)
1567 struct kvm *kvm = filp->private_data;
1568 void __user *argp = (void __user *)arg;
1571 if (kvm->mm != current->mm)
1574 case KVM_CREATE_VCPU:
1575 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1579 case KVM_SET_USER_MEMORY_REGION: {
1580 struct kvm_userspace_memory_region kvm_userspace_mem;
1583 if (copy_from_user(&kvm_userspace_mem, argp,
1584 sizeof kvm_userspace_mem))
1587 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1592 case KVM_GET_DIRTY_LOG: {
1593 struct kvm_dirty_log log;
1596 if (copy_from_user(&log, argp, sizeof log))
1598 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1603 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1604 case KVM_REGISTER_COALESCED_MMIO: {
1605 struct kvm_coalesced_mmio_zone zone;
1607 if (copy_from_user(&zone, argp, sizeof zone))
1609 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1615 case KVM_UNREGISTER_COALESCED_MMIO: {
1616 struct kvm_coalesced_mmio_zone zone;
1618 if (copy_from_user(&zone, argp, sizeof zone))
1620 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1628 struct kvm_irqfd data;
1631 if (copy_from_user(&data, argp, sizeof data))
1633 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
1636 case KVM_IOEVENTFD: {
1637 struct kvm_ioeventfd data;
1640 if (copy_from_user(&data, argp, sizeof data))
1642 r = kvm_ioeventfd(kvm, &data);
1645 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1646 case KVM_SET_BOOT_CPU_ID:
1648 mutex_lock(&kvm->lock);
1649 if (atomic_read(&kvm->online_vcpus) != 0)
1652 kvm->bsp_vcpu_id = arg;
1653 mutex_unlock(&kvm->lock);
1657 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1659 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
1665 #ifdef CONFIG_COMPAT
1666 struct compat_kvm_dirty_log {
1670 compat_uptr_t dirty_bitmap; /* one bit per page */
1675 static long kvm_vm_compat_ioctl(struct file *filp,
1676 unsigned int ioctl, unsigned long arg)
1678 struct kvm *kvm = filp->private_data;
1681 if (kvm->mm != current->mm)
1684 case KVM_GET_DIRTY_LOG: {
1685 struct compat_kvm_dirty_log compat_log;
1686 struct kvm_dirty_log log;
1689 if (copy_from_user(&compat_log, (void __user *)arg,
1690 sizeof(compat_log)))
1692 log.slot = compat_log.slot;
1693 log.padding1 = compat_log.padding1;
1694 log.padding2 = compat_log.padding2;
1695 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
1697 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1703 r = kvm_vm_ioctl(filp, ioctl, arg);
1711 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1713 struct page *page[1];
1716 gfn_t gfn = vmf->pgoff;
1717 struct kvm *kvm = vma->vm_file->private_data;
1719 addr = gfn_to_hva(kvm, gfn);
1720 if (kvm_is_error_hva(addr))
1721 return VM_FAULT_SIGBUS;
1723 npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
1725 if (unlikely(npages != 1))
1726 return VM_FAULT_SIGBUS;
1728 vmf->page = page[0];
1732 static const struct vm_operations_struct kvm_vm_vm_ops = {
1733 .fault = kvm_vm_fault,
1736 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1738 vma->vm_ops = &kvm_vm_vm_ops;
1742 static struct file_operations kvm_vm_fops = {
1743 .release = kvm_vm_release,
1744 .unlocked_ioctl = kvm_vm_ioctl,
1745 #ifdef CONFIG_COMPAT
1746 .compat_ioctl = kvm_vm_compat_ioctl,
1748 .mmap = kvm_vm_mmap,
1751 static int kvm_dev_ioctl_create_vm(void)
1756 kvm = kvm_create_vm();
1758 return PTR_ERR(kvm);
1759 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1760 r = kvm_coalesced_mmio_init(kvm);
1766 fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
1773 static long kvm_dev_ioctl_check_extension_generic(long arg)
1776 case KVM_CAP_USER_MEMORY:
1777 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
1778 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
1779 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1780 case KVM_CAP_SET_BOOT_CPU_ID:
1782 case KVM_CAP_INTERNAL_ERROR_DATA:
1784 #ifdef CONFIG_HAVE_KVM_IRQCHIP
1785 case KVM_CAP_IRQ_ROUTING:
1786 return KVM_MAX_IRQ_ROUTES;
1791 return kvm_dev_ioctl_check_extension(arg);
1794 static long kvm_dev_ioctl(struct file *filp,
1795 unsigned int ioctl, unsigned long arg)
1800 case KVM_GET_API_VERSION:
1804 r = KVM_API_VERSION;
1810 r = kvm_dev_ioctl_create_vm();
1812 case KVM_CHECK_EXTENSION:
1813 r = kvm_dev_ioctl_check_extension_generic(arg);
1815 case KVM_GET_VCPU_MMAP_SIZE:
1819 r = PAGE_SIZE; /* struct kvm_run */
1821 r += PAGE_SIZE; /* pio data page */
1823 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1824 r += PAGE_SIZE; /* coalesced mmio ring page */
1827 case KVM_TRACE_ENABLE:
1828 case KVM_TRACE_PAUSE:
1829 case KVM_TRACE_DISABLE:
1833 return kvm_arch_dev_ioctl(filp, ioctl, arg);
1839 static struct file_operations kvm_chardev_ops = {
1840 .unlocked_ioctl = kvm_dev_ioctl,
1841 .compat_ioctl = kvm_dev_ioctl,
1844 static struct miscdevice kvm_dev = {
1850 static void hardware_enable(void *junk)
1852 int cpu = raw_smp_processor_id();
1855 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
1858 cpumask_set_cpu(cpu, cpus_hardware_enabled);
1860 r = kvm_arch_hardware_enable(NULL);
1863 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
1864 atomic_inc(&hardware_enable_failed);
1865 printk(KERN_INFO "kvm: enabling virtualization on "
1866 "CPU%d failed\n", cpu);
1870 static void hardware_disable(void *junk)
1872 int cpu = raw_smp_processor_id();
1874 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
1876 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
1877 kvm_arch_hardware_disable(NULL);
1880 static void hardware_disable_all_nolock(void)
1882 BUG_ON(!kvm_usage_count);
1885 if (!kvm_usage_count)
1886 on_each_cpu(hardware_disable, NULL, 1);
1889 static void hardware_disable_all(void)
1891 spin_lock(&kvm_lock);
1892 hardware_disable_all_nolock();
1893 spin_unlock(&kvm_lock);
1896 static int hardware_enable_all(void)
1900 spin_lock(&kvm_lock);
1903 if (kvm_usage_count == 1) {
1904 atomic_set(&hardware_enable_failed, 0);
1905 on_each_cpu(hardware_enable, NULL, 1);
1907 if (atomic_read(&hardware_enable_failed)) {
1908 hardware_disable_all_nolock();
1913 spin_unlock(&kvm_lock);
1918 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
1923 if (!kvm_usage_count)
1926 val &= ~CPU_TASKS_FROZEN;
1929 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1931 hardware_disable(NULL);
1933 case CPU_UP_CANCELED:
1934 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1936 smp_call_function_single(cpu, hardware_disable, NULL, 1);
1939 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
1941 smp_call_function_single(cpu, hardware_enable, NULL, 1);
1948 asmlinkage void kvm_handle_fault_on_reboot(void)
1951 /* spin while reset goes on */
1954 /* Fault while not rebooting. We want the trace. */
1957 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
1959 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
1963 * Some (well, at least mine) BIOSes hang on reboot if
1966 * And Intel TXT required VMX off for all cpu when system shutdown.
1968 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
1969 kvm_rebooting = true;
1970 on_each_cpu(hardware_disable, NULL, 1);
1974 static struct notifier_block kvm_reboot_notifier = {
1975 .notifier_call = kvm_reboot,
1979 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
1983 for (i = 0; i < bus->dev_count; i++) {
1984 struct kvm_io_device *pos = bus->devs[i];
1986 kvm_iodevice_destructor(pos);
1991 /* kvm_io_bus_write - called under kvm->slots_lock */
1992 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
1993 int len, const void *val)
1996 struct kvm_io_bus *bus = rcu_dereference(kvm->buses[bus_idx]);
1997 for (i = 0; i < bus->dev_count; i++)
1998 if (!kvm_iodevice_write(bus->devs[i], addr, len, val))
2003 /* kvm_io_bus_read - called under kvm->slots_lock */
2004 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2008 struct kvm_io_bus *bus = rcu_dereference(kvm->buses[bus_idx]);
2010 for (i = 0; i < bus->dev_count; i++)
2011 if (!kvm_iodevice_read(bus->devs[i], addr, len, val))
2016 /* Caller must hold slots_lock. */
2017 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2018 struct kvm_io_device *dev)
2020 struct kvm_io_bus *new_bus, *bus;
2022 bus = kvm->buses[bus_idx];
2023 if (bus->dev_count > NR_IOBUS_DEVS-1)
2026 new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2029 memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2030 new_bus->devs[new_bus->dev_count++] = dev;
2031 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2032 synchronize_srcu_expedited(&kvm->srcu);
2038 /* Caller must hold slots_lock. */
2039 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2040 struct kvm_io_device *dev)
2043 struct kvm_io_bus *new_bus, *bus;
2045 new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2049 bus = kvm->buses[bus_idx];
2050 memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2053 for (i = 0; i < new_bus->dev_count; i++)
2054 if (new_bus->devs[i] == dev) {
2056 new_bus->devs[i] = new_bus->devs[--new_bus->dev_count];
2065 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2066 synchronize_srcu_expedited(&kvm->srcu);
2071 static struct notifier_block kvm_cpu_notifier = {
2072 .notifier_call = kvm_cpu_hotplug,
2073 .priority = 20, /* must be > scheduler priority */
2076 static int vm_stat_get(void *_offset, u64 *val)
2078 unsigned offset = (long)_offset;
2082 spin_lock(&kvm_lock);
2083 list_for_each_entry(kvm, &vm_list, vm_list)
2084 *val += *(u32 *)((void *)kvm + offset);
2085 spin_unlock(&kvm_lock);
2089 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2091 static int vcpu_stat_get(void *_offset, u64 *val)
2093 unsigned offset = (long)_offset;
2095 struct kvm_vcpu *vcpu;
2099 spin_lock(&kvm_lock);
2100 list_for_each_entry(kvm, &vm_list, vm_list)
2101 kvm_for_each_vcpu(i, vcpu, kvm)
2102 *val += *(u32 *)((void *)vcpu + offset);
2104 spin_unlock(&kvm_lock);
2108 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2110 static const struct file_operations *stat_fops[] = {
2111 [KVM_STAT_VCPU] = &vcpu_stat_fops,
2112 [KVM_STAT_VM] = &vm_stat_fops,
2115 static void kvm_init_debug(void)
2117 struct kvm_stats_debugfs_item *p;
2119 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2120 for (p = debugfs_entries; p->name; ++p)
2121 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2122 (void *)(long)p->offset,
2123 stat_fops[p->kind]);
2126 static void kvm_exit_debug(void)
2128 struct kvm_stats_debugfs_item *p;
2130 for (p = debugfs_entries; p->name; ++p)
2131 debugfs_remove(p->dentry);
2132 debugfs_remove(kvm_debugfs_dir);
2135 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2137 if (kvm_usage_count)
2138 hardware_disable(NULL);
2142 static int kvm_resume(struct sys_device *dev)
2144 if (kvm_usage_count)
2145 hardware_enable(NULL);
2149 static struct sysdev_class kvm_sysdev_class = {
2151 .suspend = kvm_suspend,
2152 .resume = kvm_resume,
2155 static struct sys_device kvm_sysdev = {
2157 .cls = &kvm_sysdev_class,
2160 struct page *bad_page;
2164 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2166 return container_of(pn, struct kvm_vcpu, preempt_notifier);
2169 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2171 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2173 kvm_arch_vcpu_load(vcpu, cpu);
2176 static void kvm_sched_out(struct preempt_notifier *pn,
2177 struct task_struct *next)
2179 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2181 kvm_arch_vcpu_put(vcpu);
2184 int kvm_init(void *opaque, unsigned int vcpu_size,
2185 struct module *module)
2190 r = kvm_arch_init(opaque);
2194 bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2196 if (bad_page == NULL) {
2201 bad_pfn = page_to_pfn(bad_page);
2203 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2208 r = kvm_arch_hardware_setup();
2212 for_each_online_cpu(cpu) {
2213 smp_call_function_single(cpu,
2214 kvm_arch_check_processor_compat,
2220 r = register_cpu_notifier(&kvm_cpu_notifier);
2223 register_reboot_notifier(&kvm_reboot_notifier);
2225 r = sysdev_class_register(&kvm_sysdev_class);
2229 r = sysdev_register(&kvm_sysdev);
2233 /* A kmem cache lets us meet the alignment requirements of fx_save. */
2234 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
2235 __alignof__(struct kvm_vcpu),
2237 if (!kvm_vcpu_cache) {
2242 kvm_chardev_ops.owner = module;
2243 kvm_vm_fops.owner = module;
2244 kvm_vcpu_fops.owner = module;
2246 r = misc_register(&kvm_dev);
2248 printk(KERN_ERR "kvm: misc device register failed\n");
2252 kvm_preempt_ops.sched_in = kvm_sched_in;
2253 kvm_preempt_ops.sched_out = kvm_sched_out;
2260 kmem_cache_destroy(kvm_vcpu_cache);
2262 sysdev_unregister(&kvm_sysdev);
2264 sysdev_class_unregister(&kvm_sysdev_class);
2266 unregister_reboot_notifier(&kvm_reboot_notifier);
2267 unregister_cpu_notifier(&kvm_cpu_notifier);
2270 kvm_arch_hardware_unsetup();
2272 free_cpumask_var(cpus_hardware_enabled);
2274 __free_page(bad_page);
2280 EXPORT_SYMBOL_GPL(kvm_init);
2285 misc_deregister(&kvm_dev);
2286 kmem_cache_destroy(kvm_vcpu_cache);
2287 sysdev_unregister(&kvm_sysdev);
2288 sysdev_class_unregister(&kvm_sysdev_class);
2289 unregister_reboot_notifier(&kvm_reboot_notifier);
2290 unregister_cpu_notifier(&kvm_cpu_notifier);
2291 on_each_cpu(hardware_disable, NULL, 1);
2292 kvm_arch_hardware_unsetup();
2294 free_cpumask_var(cpus_hardware_enabled);
2295 __free_page(bad_page);
2297 EXPORT_SYMBOL_GPL(kvm_exit);