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>
25 #include <linux/gfp.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/sysdev.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>
48 #include <asm/processor.h>
50 #include <asm/uaccess.h>
51 #include <asm/pgtable.h>
52 #include <asm-generic/bitops/le.h>
54 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
55 #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->slots_lock --> kvm->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 bool kvm_rebooting;
91 static bool largepages_enabled = true;
93 inline int kvm_is_mmio_pfn(pfn_t pfn)
96 struct page *page = compound_head(pfn_to_page(pfn));
97 return PageReserved(page);
104 * Switches to specified vcpu, until a matching vcpu_put()
106 void vcpu_load(struct kvm_vcpu *vcpu)
110 mutex_lock(&vcpu->mutex);
112 preempt_notifier_register(&vcpu->preempt_notifier);
113 kvm_arch_vcpu_load(vcpu, cpu);
117 void vcpu_put(struct kvm_vcpu *vcpu)
120 kvm_arch_vcpu_put(vcpu);
121 preempt_notifier_unregister(&vcpu->preempt_notifier);
123 mutex_unlock(&vcpu->mutex);
126 static void ack_flush(void *_completed)
130 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
135 struct kvm_vcpu *vcpu;
137 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
139 spin_lock(&kvm->requests_lock);
140 me = smp_processor_id();
141 kvm_for_each_vcpu(i, vcpu, kvm) {
142 if (test_and_set_bit(req, &vcpu->requests))
145 if (cpus != NULL && cpu != -1 && cpu != me)
146 cpumask_set_cpu(cpu, cpus);
148 if (unlikely(cpus == NULL))
149 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
150 else if (!cpumask_empty(cpus))
151 smp_call_function_many(cpus, ack_flush, NULL, 1);
154 spin_unlock(&kvm->requests_lock);
155 free_cpumask_var(cpus);
159 void kvm_flush_remote_tlbs(struct kvm *kvm)
161 if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
162 ++kvm->stat.remote_tlb_flush;
165 void kvm_reload_remote_mmus(struct kvm *kvm)
167 make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
170 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
175 mutex_init(&vcpu->mutex);
179 init_waitqueue_head(&vcpu->wq);
181 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
186 vcpu->run = page_address(page);
188 r = kvm_arch_vcpu_init(vcpu);
194 free_page((unsigned long)vcpu->run);
198 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
200 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
202 kvm_arch_vcpu_uninit(vcpu);
203 free_page((unsigned long)vcpu->run);
205 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
207 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
208 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
210 return container_of(mn, struct kvm, mmu_notifier);
213 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
214 struct mm_struct *mm,
215 unsigned long address)
217 struct kvm *kvm = mmu_notifier_to_kvm(mn);
221 * When ->invalidate_page runs, the linux pte has been zapped
222 * already but the page is still allocated until
223 * ->invalidate_page returns. So if we increase the sequence
224 * here the kvm page fault will notice if the spte can't be
225 * established because the page is going to be freed. If
226 * instead the kvm page fault establishes the spte before
227 * ->invalidate_page runs, kvm_unmap_hva will release it
230 * The sequence increase only need to be seen at spin_unlock
231 * time, and not at spin_lock time.
233 * Increasing the sequence after the spin_unlock would be
234 * unsafe because the kvm page fault could then establish the
235 * pte after kvm_unmap_hva returned, without noticing the page
236 * is going to be freed.
238 spin_lock(&kvm->mmu_lock);
239 kvm->mmu_notifier_seq++;
240 need_tlb_flush = kvm_unmap_hva(kvm, address);
241 spin_unlock(&kvm->mmu_lock);
243 /* we've to flush the tlb before the pages can be freed */
245 kvm_flush_remote_tlbs(kvm);
249 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
250 struct mm_struct *mm,
251 unsigned long address,
254 struct kvm *kvm = mmu_notifier_to_kvm(mn);
256 spin_lock(&kvm->mmu_lock);
257 kvm->mmu_notifier_seq++;
258 kvm_set_spte_hva(kvm, address, pte);
259 spin_unlock(&kvm->mmu_lock);
262 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
263 struct mm_struct *mm,
267 struct kvm *kvm = mmu_notifier_to_kvm(mn);
268 int need_tlb_flush = 0;
270 spin_lock(&kvm->mmu_lock);
272 * The count increase must become visible at unlock time as no
273 * spte can be established without taking the mmu_lock and
274 * count is also read inside the mmu_lock critical section.
276 kvm->mmu_notifier_count++;
277 for (; start < end; start += PAGE_SIZE)
278 need_tlb_flush |= kvm_unmap_hva(kvm, start);
279 spin_unlock(&kvm->mmu_lock);
281 /* we've to flush the tlb before the pages can be freed */
283 kvm_flush_remote_tlbs(kvm);
286 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
287 struct mm_struct *mm,
291 struct kvm *kvm = mmu_notifier_to_kvm(mn);
293 spin_lock(&kvm->mmu_lock);
295 * This sequence increase will notify the kvm page fault that
296 * the page that is going to be mapped in the spte could have
299 kvm->mmu_notifier_seq++;
301 * The above sequence increase must be visible before the
302 * below count decrease but both values are read by the kvm
303 * page fault under mmu_lock spinlock so we don't need to add
304 * a smb_wmb() here in between the two.
306 kvm->mmu_notifier_count--;
307 spin_unlock(&kvm->mmu_lock);
309 BUG_ON(kvm->mmu_notifier_count < 0);
312 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
313 struct mm_struct *mm,
314 unsigned long address)
316 struct kvm *kvm = mmu_notifier_to_kvm(mn);
319 spin_lock(&kvm->mmu_lock);
320 young = kvm_age_hva(kvm, address);
321 spin_unlock(&kvm->mmu_lock);
324 kvm_flush_remote_tlbs(kvm);
329 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
330 struct mm_struct *mm)
332 struct kvm *kvm = mmu_notifier_to_kvm(mn);
333 kvm_arch_flush_shadow(kvm);
336 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
337 .invalidate_page = kvm_mmu_notifier_invalidate_page,
338 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
339 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
340 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
341 .change_pte = kvm_mmu_notifier_change_pte,
342 .release = kvm_mmu_notifier_release,
344 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
346 static struct kvm *kvm_create_vm(void)
349 struct kvm *kvm = kvm_arch_create_vm();
350 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
357 r = hardware_enable_all();
359 goto out_err_nodisable;
361 #ifdef CONFIG_HAVE_KVM_IRQCHIP
362 INIT_HLIST_HEAD(&kvm->mask_notifier_list);
363 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
366 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
367 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
372 kvm->coalesced_mmio_ring =
373 (struct kvm_coalesced_mmio_ring *)page_address(page);
376 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
378 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
379 r = mmu_notifier_register(&kvm->mmu_notifier, current->mm);
381 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
389 kvm->mm = current->mm;
390 atomic_inc(&kvm->mm->mm_count);
391 spin_lock_init(&kvm->mmu_lock);
392 spin_lock_init(&kvm->requests_lock);
393 kvm_io_bus_init(&kvm->pio_bus);
394 kvm_eventfd_init(kvm);
395 mutex_init(&kvm->lock);
396 mutex_init(&kvm->irq_lock);
397 kvm_io_bus_init(&kvm->mmio_bus);
398 init_rwsem(&kvm->slots_lock);
399 atomic_set(&kvm->users_count, 1);
400 spin_lock(&kvm_lock);
401 list_add(&kvm->vm_list, &vm_list);
402 spin_unlock(&kvm_lock);
403 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
404 kvm_coalesced_mmio_init(kvm);
410 hardware_disable_all();
417 * Free any memory in @free but not in @dont.
419 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
420 struct kvm_memory_slot *dont)
424 if (!dont || free->rmap != dont->rmap)
427 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
428 vfree(free->dirty_bitmap);
431 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
432 if (!dont || free->lpage_info[i] != dont->lpage_info[i]) {
433 vfree(free->lpage_info[i]);
434 free->lpage_info[i] = NULL;
439 free->dirty_bitmap = NULL;
443 void kvm_free_physmem(struct kvm *kvm)
447 for (i = 0; i < kvm->nmemslots; ++i)
448 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
451 static void kvm_destroy_vm(struct kvm *kvm)
453 struct mm_struct *mm = kvm->mm;
455 kvm_arch_sync_events(kvm);
456 spin_lock(&kvm_lock);
457 list_del(&kvm->vm_list);
458 spin_unlock(&kvm_lock);
459 kvm_free_irq_routing(kvm);
460 kvm_io_bus_destroy(&kvm->pio_bus);
461 kvm_io_bus_destroy(&kvm->mmio_bus);
462 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
463 if (kvm->coalesced_mmio_ring != NULL)
464 free_page((unsigned long)kvm->coalesced_mmio_ring);
466 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
467 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
469 kvm_arch_flush_shadow(kvm);
471 kvm_arch_destroy_vm(kvm);
472 hardware_disable_all();
476 void kvm_get_kvm(struct kvm *kvm)
478 atomic_inc(&kvm->users_count);
480 EXPORT_SYMBOL_GPL(kvm_get_kvm);
482 void kvm_put_kvm(struct kvm *kvm)
484 if (atomic_dec_and_test(&kvm->users_count))
487 EXPORT_SYMBOL_GPL(kvm_put_kvm);
490 static int kvm_vm_release(struct inode *inode, struct file *filp)
492 struct kvm *kvm = filp->private_data;
494 kvm_irqfd_release(kvm);
501 * Allocate some memory and give it an address in the guest physical address
504 * Discontiguous memory is allowed, mostly for framebuffers.
506 * Must be called holding mmap_sem for write.
508 int __kvm_set_memory_region(struct kvm *kvm,
509 struct kvm_userspace_memory_region *mem,
514 unsigned long npages;
516 struct kvm_memory_slot *memslot;
517 struct kvm_memory_slot old, new;
520 /* General sanity checks */
521 if (mem->memory_size & (PAGE_SIZE - 1))
523 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
525 if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
527 if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
529 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
532 memslot = &kvm->memslots[mem->slot];
533 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
534 npages = mem->memory_size >> PAGE_SHIFT;
537 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
539 new = old = *memslot;
541 new.base_gfn = base_gfn;
543 new.flags = mem->flags;
545 /* Disallow changing a memory slot's size. */
547 if (npages && old.npages && npages != old.npages)
550 /* Check for overlaps */
552 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
553 struct kvm_memory_slot *s = &kvm->memslots[i];
555 if (s == memslot || !s->npages)
557 if (!((base_gfn + npages <= s->base_gfn) ||
558 (base_gfn >= s->base_gfn + s->npages)))
562 /* Free page dirty bitmap if unneeded */
563 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
564 new.dirty_bitmap = NULL;
568 /* Allocate if a slot is being created */
570 if (npages && !new.rmap) {
571 new.rmap = vmalloc(npages * sizeof(struct page *));
576 memset(new.rmap, 0, npages * sizeof(*new.rmap));
578 new.user_alloc = user_alloc;
580 * hva_to_rmmap() serialzies with the mmu_lock and to be
581 * safe it has to ignore memslots with !user_alloc &&
585 new.userspace_addr = mem->userspace_addr;
587 new.userspace_addr = 0;
592 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
598 /* Avoid unused variable warning if no large pages */
601 if (new.lpage_info[i])
604 lpages = 1 + (base_gfn + npages - 1) /
605 KVM_PAGES_PER_HPAGE(level);
606 lpages -= base_gfn / KVM_PAGES_PER_HPAGE(level);
608 new.lpage_info[i] = vmalloc(lpages * sizeof(*new.lpage_info[i]));
610 if (!new.lpage_info[i])
613 memset(new.lpage_info[i], 0,
614 lpages * sizeof(*new.lpage_info[i]));
616 if (base_gfn % KVM_PAGES_PER_HPAGE(level))
617 new.lpage_info[i][0].write_count = 1;
618 if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE(level))
619 new.lpage_info[i][lpages - 1].write_count = 1;
620 ugfn = new.userspace_addr >> PAGE_SHIFT;
622 * If the gfn and userspace address are not aligned wrt each
623 * other, or if explicitly asked to, disable large page
624 * support for this slot
626 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
628 for (j = 0; j < lpages; ++j)
629 new.lpage_info[i][j].write_count = 1;
634 /* Allocate page dirty bitmap if needed */
635 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
636 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
638 new.dirty_bitmap = vmalloc(dirty_bytes);
639 if (!new.dirty_bitmap)
641 memset(new.dirty_bitmap, 0, dirty_bytes);
643 kvm_arch_flush_shadow(kvm);
645 #else /* not defined CONFIG_S390 */
646 new.user_alloc = user_alloc;
648 new.userspace_addr = mem->userspace_addr;
649 #endif /* not defined CONFIG_S390 */
652 kvm_arch_flush_shadow(kvm);
654 spin_lock(&kvm->mmu_lock);
655 if (mem->slot >= kvm->nmemslots)
656 kvm->nmemslots = mem->slot + 1;
659 spin_unlock(&kvm->mmu_lock);
661 r = kvm_arch_set_memory_region(kvm, mem, old, user_alloc);
663 spin_lock(&kvm->mmu_lock);
665 spin_unlock(&kvm->mmu_lock);
669 kvm_free_physmem_slot(&old, npages ? &new : NULL);
670 /* Slot deletion case: we have to update the current slot */
671 spin_lock(&kvm->mmu_lock);
674 spin_unlock(&kvm->mmu_lock);
676 /* map the pages in iommu page table */
677 r = kvm_iommu_map_pages(kvm, base_gfn, npages);
684 kvm_free_physmem_slot(&new, &old);
689 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
691 int kvm_set_memory_region(struct kvm *kvm,
692 struct kvm_userspace_memory_region *mem,
697 down_write(&kvm->slots_lock);
698 r = __kvm_set_memory_region(kvm, mem, user_alloc);
699 up_write(&kvm->slots_lock);
702 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
704 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
706 kvm_userspace_memory_region *mem,
709 if (mem->slot >= KVM_MEMORY_SLOTS)
711 return kvm_set_memory_region(kvm, mem, user_alloc);
714 int kvm_get_dirty_log(struct kvm *kvm,
715 struct kvm_dirty_log *log, int *is_dirty)
717 struct kvm_memory_slot *memslot;
720 unsigned long any = 0;
723 if (log->slot >= KVM_MEMORY_SLOTS)
726 memslot = &kvm->memslots[log->slot];
728 if (!memslot->dirty_bitmap)
731 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
733 for (i = 0; !any && i < n/sizeof(long); ++i)
734 any = memslot->dirty_bitmap[i];
737 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
748 void kvm_disable_largepages(void)
750 largepages_enabled = false;
752 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
754 int is_error_page(struct page *page)
756 return page == bad_page;
758 EXPORT_SYMBOL_GPL(is_error_page);
760 int is_error_pfn(pfn_t pfn)
762 return pfn == bad_pfn;
764 EXPORT_SYMBOL_GPL(is_error_pfn);
766 static inline unsigned long bad_hva(void)
771 int kvm_is_error_hva(unsigned long addr)
773 return addr == bad_hva();
775 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
777 struct kvm_memory_slot *gfn_to_memslot_unaliased(struct kvm *kvm, gfn_t gfn)
781 for (i = 0; i < kvm->nmemslots; ++i) {
782 struct kvm_memory_slot *memslot = &kvm->memslots[i];
784 if (gfn >= memslot->base_gfn
785 && gfn < memslot->base_gfn + memslot->npages)
790 EXPORT_SYMBOL_GPL(gfn_to_memslot_unaliased);
792 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
794 gfn = unalias_gfn(kvm, gfn);
795 return gfn_to_memslot_unaliased(kvm, gfn);
798 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
802 gfn = unalias_gfn(kvm, gfn);
803 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
804 struct kvm_memory_slot *memslot = &kvm->memslots[i];
806 if (gfn >= memslot->base_gfn
807 && gfn < memslot->base_gfn + memslot->npages)
812 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
814 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
816 struct kvm_memory_slot *slot;
818 gfn = unalias_gfn(kvm, gfn);
819 slot = gfn_to_memslot_unaliased(kvm, gfn);
822 return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
824 EXPORT_SYMBOL_GPL(gfn_to_hva);
826 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
828 struct page *page[1];
835 addr = gfn_to_hva(kvm, gfn);
836 if (kvm_is_error_hva(addr)) {
838 return page_to_pfn(bad_page);
841 npages = get_user_pages_fast(addr, 1, 1, page);
843 if (unlikely(npages != 1)) {
844 struct vm_area_struct *vma;
846 down_read(¤t->mm->mmap_sem);
847 vma = find_vma(current->mm, addr);
849 if (vma == NULL || addr < vma->vm_start ||
850 !(vma->vm_flags & VM_PFNMAP)) {
851 up_read(¤t->mm->mmap_sem);
853 return page_to_pfn(bad_page);
856 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
857 up_read(¤t->mm->mmap_sem);
858 BUG_ON(!kvm_is_mmio_pfn(pfn));
860 pfn = page_to_pfn(page[0]);
865 EXPORT_SYMBOL_GPL(gfn_to_pfn);
867 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
871 pfn = gfn_to_pfn(kvm, gfn);
872 if (!kvm_is_mmio_pfn(pfn))
873 return pfn_to_page(pfn);
875 WARN_ON(kvm_is_mmio_pfn(pfn));
881 EXPORT_SYMBOL_GPL(gfn_to_page);
883 void kvm_release_page_clean(struct page *page)
885 kvm_release_pfn_clean(page_to_pfn(page));
887 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
889 void kvm_release_pfn_clean(pfn_t pfn)
891 if (!kvm_is_mmio_pfn(pfn))
892 put_page(pfn_to_page(pfn));
894 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
896 void kvm_release_page_dirty(struct page *page)
898 kvm_release_pfn_dirty(page_to_pfn(page));
900 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
902 void kvm_release_pfn_dirty(pfn_t pfn)
904 kvm_set_pfn_dirty(pfn);
905 kvm_release_pfn_clean(pfn);
907 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
909 void kvm_set_page_dirty(struct page *page)
911 kvm_set_pfn_dirty(page_to_pfn(page));
913 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
915 void kvm_set_pfn_dirty(pfn_t pfn)
917 if (!kvm_is_mmio_pfn(pfn)) {
918 struct page *page = pfn_to_page(pfn);
919 if (!PageReserved(page))
923 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
925 void kvm_set_pfn_accessed(pfn_t pfn)
927 if (!kvm_is_mmio_pfn(pfn))
928 mark_page_accessed(pfn_to_page(pfn));
930 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
932 void kvm_get_pfn(pfn_t pfn)
934 if (!kvm_is_mmio_pfn(pfn))
935 get_page(pfn_to_page(pfn));
937 EXPORT_SYMBOL_GPL(kvm_get_pfn);
939 static int next_segment(unsigned long len, int offset)
941 if (len > PAGE_SIZE - offset)
942 return PAGE_SIZE - offset;
947 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
953 addr = gfn_to_hva(kvm, gfn);
954 if (kvm_is_error_hva(addr))
956 r = copy_from_user(data, (void __user *)addr + offset, len);
961 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
963 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
965 gfn_t gfn = gpa >> PAGE_SHIFT;
967 int offset = offset_in_page(gpa);
970 while ((seg = next_segment(len, offset)) != 0) {
971 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
981 EXPORT_SYMBOL_GPL(kvm_read_guest);
983 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
988 gfn_t gfn = gpa >> PAGE_SHIFT;
989 int offset = offset_in_page(gpa);
991 addr = gfn_to_hva(kvm, gfn);
992 if (kvm_is_error_hva(addr))
995 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1001 EXPORT_SYMBOL(kvm_read_guest_atomic);
1003 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1004 int offset, int len)
1009 addr = gfn_to_hva(kvm, gfn);
1010 if (kvm_is_error_hva(addr))
1012 r = copy_to_user((void __user *)addr + offset, data, len);
1015 mark_page_dirty(kvm, gfn);
1018 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1020 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1023 gfn_t gfn = gpa >> PAGE_SHIFT;
1025 int offset = offset_in_page(gpa);
1028 while ((seg = next_segment(len, offset)) != 0) {
1029 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1040 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1042 return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
1044 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1046 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1048 gfn_t gfn = gpa >> PAGE_SHIFT;
1050 int offset = offset_in_page(gpa);
1053 while ((seg = next_segment(len, offset)) != 0) {
1054 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1063 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1065 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1067 struct kvm_memory_slot *memslot;
1069 gfn = unalias_gfn(kvm, gfn);
1070 memslot = gfn_to_memslot_unaliased(kvm, gfn);
1071 if (memslot && memslot->dirty_bitmap) {
1072 unsigned long rel_gfn = gfn - memslot->base_gfn;
1075 if (!generic_test_le_bit(rel_gfn, memslot->dirty_bitmap))
1076 generic___set_le_bit(rel_gfn, memslot->dirty_bitmap);
1081 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1083 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1088 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1090 if (kvm_arch_vcpu_runnable(vcpu)) {
1091 set_bit(KVM_REQ_UNHALT, &vcpu->requests);
1094 if (kvm_cpu_has_pending_timer(vcpu))
1096 if (signal_pending(current))
1102 finish_wait(&vcpu->wq, &wait);
1105 void kvm_resched(struct kvm_vcpu *vcpu)
1107 if (!need_resched())
1111 EXPORT_SYMBOL_GPL(kvm_resched);
1113 void kvm_vcpu_on_spin(struct kvm_vcpu *vcpu)
1118 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1120 /* Sleep for 100 us, and hope lock-holder got scheduled */
1121 expires = ktime_add_ns(ktime_get(), 100000UL);
1122 schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
1124 finish_wait(&vcpu->wq, &wait);
1126 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1128 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1130 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1133 if (vmf->pgoff == 0)
1134 page = virt_to_page(vcpu->run);
1136 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1137 page = virt_to_page(vcpu->arch.pio_data);
1139 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1140 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1141 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1144 return VM_FAULT_SIGBUS;
1150 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1151 .fault = kvm_vcpu_fault,
1154 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1156 vma->vm_ops = &kvm_vcpu_vm_ops;
1160 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1162 struct kvm_vcpu *vcpu = filp->private_data;
1164 kvm_put_kvm(vcpu->kvm);
1168 static struct file_operations kvm_vcpu_fops = {
1169 .release = kvm_vcpu_release,
1170 .unlocked_ioctl = kvm_vcpu_ioctl,
1171 .compat_ioctl = kvm_vcpu_ioctl,
1172 .mmap = kvm_vcpu_mmap,
1176 * Allocates an inode for the vcpu.
1178 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1180 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, 0);
1184 * Creates some virtual cpus. Good luck creating more than one.
1186 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1189 struct kvm_vcpu *vcpu, *v;
1191 vcpu = kvm_arch_vcpu_create(kvm, id);
1193 return PTR_ERR(vcpu);
1195 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1197 r = kvm_arch_vcpu_setup(vcpu);
1201 mutex_lock(&kvm->lock);
1202 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1207 kvm_for_each_vcpu(r, v, kvm)
1208 if (v->vcpu_id == id) {
1213 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1215 /* Now it's all set up, let userspace reach it */
1217 r = create_vcpu_fd(vcpu);
1223 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1225 atomic_inc(&kvm->online_vcpus);
1227 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1228 if (kvm->bsp_vcpu_id == id)
1229 kvm->bsp_vcpu = vcpu;
1231 mutex_unlock(&kvm->lock);
1235 mutex_unlock(&kvm->lock);
1236 kvm_arch_vcpu_destroy(vcpu);
1240 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1243 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1244 vcpu->sigset_active = 1;
1245 vcpu->sigset = *sigset;
1247 vcpu->sigset_active = 0;
1251 static long kvm_vcpu_ioctl(struct file *filp,
1252 unsigned int ioctl, unsigned long arg)
1254 struct kvm_vcpu *vcpu = filp->private_data;
1255 void __user *argp = (void __user *)arg;
1257 struct kvm_fpu *fpu = NULL;
1258 struct kvm_sregs *kvm_sregs = NULL;
1260 if (vcpu->kvm->mm != current->mm)
1267 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1269 case KVM_GET_REGS: {
1270 struct kvm_regs *kvm_regs;
1273 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1276 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1280 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1287 case KVM_SET_REGS: {
1288 struct kvm_regs *kvm_regs;
1291 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1295 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1297 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1305 case KVM_GET_SREGS: {
1306 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1310 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1314 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1319 case KVM_SET_SREGS: {
1320 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1325 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1327 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1333 case KVM_GET_MP_STATE: {
1334 struct kvm_mp_state mp_state;
1336 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1340 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1345 case KVM_SET_MP_STATE: {
1346 struct kvm_mp_state mp_state;
1349 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1351 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1357 case KVM_TRANSLATE: {
1358 struct kvm_translation tr;
1361 if (copy_from_user(&tr, argp, sizeof tr))
1363 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1367 if (copy_to_user(argp, &tr, sizeof tr))
1372 case KVM_SET_GUEST_DEBUG: {
1373 struct kvm_guest_debug dbg;
1376 if (copy_from_user(&dbg, argp, sizeof dbg))
1378 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1384 case KVM_SET_SIGNAL_MASK: {
1385 struct kvm_signal_mask __user *sigmask_arg = argp;
1386 struct kvm_signal_mask kvm_sigmask;
1387 sigset_t sigset, *p;
1392 if (copy_from_user(&kvm_sigmask, argp,
1393 sizeof kvm_sigmask))
1396 if (kvm_sigmask.len != sizeof sigset)
1399 if (copy_from_user(&sigset, sigmask_arg->sigset,
1404 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1408 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1412 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1416 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1422 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1427 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1429 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1436 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1444 static long kvm_vm_ioctl(struct file *filp,
1445 unsigned int ioctl, unsigned long arg)
1447 struct kvm *kvm = filp->private_data;
1448 void __user *argp = (void __user *)arg;
1451 if (kvm->mm != current->mm)
1454 case KVM_CREATE_VCPU:
1455 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1459 case KVM_SET_USER_MEMORY_REGION: {
1460 struct kvm_userspace_memory_region kvm_userspace_mem;
1463 if (copy_from_user(&kvm_userspace_mem, argp,
1464 sizeof kvm_userspace_mem))
1467 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1472 case KVM_GET_DIRTY_LOG: {
1473 struct kvm_dirty_log log;
1476 if (copy_from_user(&log, argp, sizeof log))
1478 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1483 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1484 case KVM_REGISTER_COALESCED_MMIO: {
1485 struct kvm_coalesced_mmio_zone zone;
1487 if (copy_from_user(&zone, argp, sizeof zone))
1490 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1496 case KVM_UNREGISTER_COALESCED_MMIO: {
1497 struct kvm_coalesced_mmio_zone zone;
1499 if (copy_from_user(&zone, argp, sizeof zone))
1502 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1510 struct kvm_irqfd data;
1513 if (copy_from_user(&data, argp, sizeof data))
1515 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
1518 case KVM_IOEVENTFD: {
1519 struct kvm_ioeventfd data;
1522 if (copy_from_user(&data, argp, sizeof data))
1524 r = kvm_ioeventfd(kvm, &data);
1527 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1528 case KVM_SET_BOOT_CPU_ID:
1530 mutex_lock(&kvm->lock);
1531 if (atomic_read(&kvm->online_vcpus) != 0)
1534 kvm->bsp_vcpu_id = arg;
1535 mutex_unlock(&kvm->lock);
1539 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1541 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
1547 #ifdef CONFIG_COMPAT
1548 struct compat_kvm_dirty_log {
1552 compat_uptr_t dirty_bitmap; /* one bit per page */
1557 static long kvm_vm_compat_ioctl(struct file *filp,
1558 unsigned int ioctl, unsigned long arg)
1560 struct kvm *kvm = filp->private_data;
1563 if (kvm->mm != current->mm)
1566 case KVM_GET_DIRTY_LOG: {
1567 struct compat_kvm_dirty_log compat_log;
1568 struct kvm_dirty_log log;
1571 if (copy_from_user(&compat_log, (void __user *)arg,
1572 sizeof(compat_log)))
1574 log.slot = compat_log.slot;
1575 log.padding1 = compat_log.padding1;
1576 log.padding2 = compat_log.padding2;
1577 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
1579 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1585 r = kvm_vm_ioctl(filp, ioctl, arg);
1593 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1595 struct page *page[1];
1598 gfn_t gfn = vmf->pgoff;
1599 struct kvm *kvm = vma->vm_file->private_data;
1601 addr = gfn_to_hva(kvm, gfn);
1602 if (kvm_is_error_hva(addr))
1603 return VM_FAULT_SIGBUS;
1605 npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
1607 if (unlikely(npages != 1))
1608 return VM_FAULT_SIGBUS;
1610 vmf->page = page[0];
1614 static const struct vm_operations_struct kvm_vm_vm_ops = {
1615 .fault = kvm_vm_fault,
1618 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1620 vma->vm_ops = &kvm_vm_vm_ops;
1624 static struct file_operations kvm_vm_fops = {
1625 .release = kvm_vm_release,
1626 .unlocked_ioctl = kvm_vm_ioctl,
1627 #ifdef CONFIG_COMPAT
1628 .compat_ioctl = kvm_vm_compat_ioctl,
1630 .mmap = kvm_vm_mmap,
1633 static int kvm_dev_ioctl_create_vm(void)
1638 kvm = kvm_create_vm();
1640 return PTR_ERR(kvm);
1641 fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, 0);
1648 static long kvm_dev_ioctl_check_extension_generic(long arg)
1651 case KVM_CAP_USER_MEMORY:
1652 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
1653 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
1654 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1655 case KVM_CAP_SET_BOOT_CPU_ID:
1657 case KVM_CAP_INTERNAL_ERROR_DATA:
1659 #ifdef CONFIG_HAVE_KVM_IRQCHIP
1660 case KVM_CAP_IRQ_ROUTING:
1661 return KVM_MAX_IRQ_ROUTES;
1666 return kvm_dev_ioctl_check_extension(arg);
1669 static long kvm_dev_ioctl(struct file *filp,
1670 unsigned int ioctl, unsigned long arg)
1675 case KVM_GET_API_VERSION:
1679 r = KVM_API_VERSION;
1685 r = kvm_dev_ioctl_create_vm();
1687 case KVM_CHECK_EXTENSION:
1688 r = kvm_dev_ioctl_check_extension_generic(arg);
1690 case KVM_GET_VCPU_MMAP_SIZE:
1694 r = PAGE_SIZE; /* struct kvm_run */
1696 r += PAGE_SIZE; /* pio data page */
1698 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1699 r += PAGE_SIZE; /* coalesced mmio ring page */
1702 case KVM_TRACE_ENABLE:
1703 case KVM_TRACE_PAUSE:
1704 case KVM_TRACE_DISABLE:
1708 return kvm_arch_dev_ioctl(filp, ioctl, arg);
1714 static struct file_operations kvm_chardev_ops = {
1715 .unlocked_ioctl = kvm_dev_ioctl,
1716 .compat_ioctl = kvm_dev_ioctl,
1719 static struct miscdevice kvm_dev = {
1725 static void hardware_enable(void *junk)
1727 int cpu = raw_smp_processor_id();
1730 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
1733 cpumask_set_cpu(cpu, cpus_hardware_enabled);
1735 r = kvm_arch_hardware_enable(NULL);
1738 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
1739 atomic_inc(&hardware_enable_failed);
1740 printk(KERN_INFO "kvm: enabling virtualization on "
1741 "CPU%d failed\n", cpu);
1745 static void hardware_disable(void *junk)
1747 int cpu = raw_smp_processor_id();
1749 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
1751 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
1752 kvm_arch_hardware_disable(NULL);
1755 static void hardware_disable_all_nolock(void)
1757 BUG_ON(!kvm_usage_count);
1760 if (!kvm_usage_count)
1761 on_each_cpu(hardware_disable, NULL, 1);
1764 static void hardware_disable_all(void)
1766 spin_lock(&kvm_lock);
1767 hardware_disable_all_nolock();
1768 spin_unlock(&kvm_lock);
1771 static int hardware_enable_all(void)
1775 spin_lock(&kvm_lock);
1778 if (kvm_usage_count == 1) {
1779 atomic_set(&hardware_enable_failed, 0);
1780 on_each_cpu(hardware_enable, NULL, 1);
1782 if (atomic_read(&hardware_enable_failed)) {
1783 hardware_disable_all_nolock();
1788 spin_unlock(&kvm_lock);
1793 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
1798 if (!kvm_usage_count)
1801 val &= ~CPU_TASKS_FROZEN;
1804 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1806 hardware_disable(NULL);
1808 case CPU_UP_CANCELED:
1809 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1811 smp_call_function_single(cpu, hardware_disable, NULL, 1);
1814 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
1816 smp_call_function_single(cpu, hardware_enable, NULL, 1);
1823 asmlinkage void kvm_handle_fault_on_reboot(void)
1826 /* spin while reset goes on */
1829 /* Fault while not rebooting. We want the trace. */
1832 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
1834 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
1838 * Some (well, at least mine) BIOSes hang on reboot if
1841 * And Intel TXT required VMX off for all cpu when system shutdown.
1843 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
1844 kvm_rebooting = true;
1845 on_each_cpu(hardware_disable, NULL, 1);
1849 static struct notifier_block kvm_reboot_notifier = {
1850 .notifier_call = kvm_reboot,
1854 void kvm_io_bus_init(struct kvm_io_bus *bus)
1856 memset(bus, 0, sizeof(*bus));
1859 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
1863 for (i = 0; i < bus->dev_count; i++) {
1864 struct kvm_io_device *pos = bus->devs[i];
1866 kvm_iodevice_destructor(pos);
1870 /* kvm_io_bus_write - called under kvm->slots_lock */
1871 int kvm_io_bus_write(struct kvm_io_bus *bus, gpa_t addr,
1872 int len, const void *val)
1875 for (i = 0; i < bus->dev_count; i++)
1876 if (!kvm_iodevice_write(bus->devs[i], addr, len, val))
1881 /* kvm_io_bus_read - called under kvm->slots_lock */
1882 int kvm_io_bus_read(struct kvm_io_bus *bus, gpa_t addr, int len, void *val)
1885 for (i = 0; i < bus->dev_count; i++)
1886 if (!kvm_iodevice_read(bus->devs[i], addr, len, val))
1891 int kvm_io_bus_register_dev(struct kvm *kvm, struct kvm_io_bus *bus,
1892 struct kvm_io_device *dev)
1896 down_write(&kvm->slots_lock);
1897 ret = __kvm_io_bus_register_dev(bus, dev);
1898 up_write(&kvm->slots_lock);
1903 /* An unlocked version. Caller must have write lock on slots_lock. */
1904 int __kvm_io_bus_register_dev(struct kvm_io_bus *bus,
1905 struct kvm_io_device *dev)
1907 if (bus->dev_count > NR_IOBUS_DEVS-1)
1910 bus->devs[bus->dev_count++] = dev;
1915 void kvm_io_bus_unregister_dev(struct kvm *kvm,
1916 struct kvm_io_bus *bus,
1917 struct kvm_io_device *dev)
1919 down_write(&kvm->slots_lock);
1920 __kvm_io_bus_unregister_dev(bus, dev);
1921 up_write(&kvm->slots_lock);
1924 /* An unlocked version. Caller must have write lock on slots_lock. */
1925 void __kvm_io_bus_unregister_dev(struct kvm_io_bus *bus,
1926 struct kvm_io_device *dev)
1930 for (i = 0; i < bus->dev_count; i++)
1931 if (bus->devs[i] == dev) {
1932 bus->devs[i] = bus->devs[--bus->dev_count];
1937 static struct notifier_block kvm_cpu_notifier = {
1938 .notifier_call = kvm_cpu_hotplug,
1939 .priority = 20, /* must be > scheduler priority */
1942 static int vm_stat_get(void *_offset, u64 *val)
1944 unsigned offset = (long)_offset;
1948 spin_lock(&kvm_lock);
1949 list_for_each_entry(kvm, &vm_list, vm_list)
1950 *val += *(u32 *)((void *)kvm + offset);
1951 spin_unlock(&kvm_lock);
1955 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
1957 static int vcpu_stat_get(void *_offset, u64 *val)
1959 unsigned offset = (long)_offset;
1961 struct kvm_vcpu *vcpu;
1965 spin_lock(&kvm_lock);
1966 list_for_each_entry(kvm, &vm_list, vm_list)
1967 kvm_for_each_vcpu(i, vcpu, kvm)
1968 *val += *(u32 *)((void *)vcpu + offset);
1970 spin_unlock(&kvm_lock);
1974 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
1976 static const struct file_operations *stat_fops[] = {
1977 [KVM_STAT_VCPU] = &vcpu_stat_fops,
1978 [KVM_STAT_VM] = &vm_stat_fops,
1981 static void kvm_init_debug(void)
1983 struct kvm_stats_debugfs_item *p;
1985 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
1986 for (p = debugfs_entries; p->name; ++p)
1987 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
1988 (void *)(long)p->offset,
1989 stat_fops[p->kind]);
1992 static void kvm_exit_debug(void)
1994 struct kvm_stats_debugfs_item *p;
1996 for (p = debugfs_entries; p->name; ++p)
1997 debugfs_remove(p->dentry);
1998 debugfs_remove(kvm_debugfs_dir);
2001 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2003 if (kvm_usage_count)
2004 hardware_disable(NULL);
2008 static int kvm_resume(struct sys_device *dev)
2010 if (kvm_usage_count)
2011 hardware_enable(NULL);
2015 static struct sysdev_class kvm_sysdev_class = {
2017 .suspend = kvm_suspend,
2018 .resume = kvm_resume,
2021 static struct sys_device kvm_sysdev = {
2023 .cls = &kvm_sysdev_class,
2026 struct page *bad_page;
2030 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2032 return container_of(pn, struct kvm_vcpu, preempt_notifier);
2035 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2037 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2039 kvm_arch_vcpu_load(vcpu, cpu);
2042 static void kvm_sched_out(struct preempt_notifier *pn,
2043 struct task_struct *next)
2045 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2047 kvm_arch_vcpu_put(vcpu);
2050 int kvm_init(void *opaque, unsigned int vcpu_size,
2051 struct module *module)
2056 r = kvm_arch_init(opaque);
2060 bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2062 if (bad_page == NULL) {
2067 bad_pfn = page_to_pfn(bad_page);
2069 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2074 r = kvm_arch_hardware_setup();
2078 for_each_online_cpu(cpu) {
2079 smp_call_function_single(cpu,
2080 kvm_arch_check_processor_compat,
2086 r = register_cpu_notifier(&kvm_cpu_notifier);
2089 register_reboot_notifier(&kvm_reboot_notifier);
2091 r = sysdev_class_register(&kvm_sysdev_class);
2095 r = sysdev_register(&kvm_sysdev);
2099 /* A kmem cache lets us meet the alignment requirements of fx_save. */
2100 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
2101 __alignof__(struct kvm_vcpu),
2103 if (!kvm_vcpu_cache) {
2108 kvm_chardev_ops.owner = module;
2109 kvm_vm_fops.owner = module;
2110 kvm_vcpu_fops.owner = module;
2112 r = misc_register(&kvm_dev);
2114 printk(KERN_ERR "kvm: misc device register failed\n");
2118 kvm_preempt_ops.sched_in = kvm_sched_in;
2119 kvm_preempt_ops.sched_out = kvm_sched_out;
2126 kmem_cache_destroy(kvm_vcpu_cache);
2128 sysdev_unregister(&kvm_sysdev);
2130 sysdev_class_unregister(&kvm_sysdev_class);
2132 unregister_reboot_notifier(&kvm_reboot_notifier);
2133 unregister_cpu_notifier(&kvm_cpu_notifier);
2136 kvm_arch_hardware_unsetup();
2138 free_cpumask_var(cpus_hardware_enabled);
2140 __free_page(bad_page);
2146 EXPORT_SYMBOL_GPL(kvm_init);
2150 tracepoint_synchronize_unregister();
2152 misc_deregister(&kvm_dev);
2153 kmem_cache_destroy(kvm_vcpu_cache);
2154 sysdev_unregister(&kvm_sysdev);
2155 sysdev_class_unregister(&kvm_sysdev_class);
2156 unregister_reboot_notifier(&kvm_reboot_notifier);
2157 unregister_cpu_notifier(&kvm_cpu_notifier);
2158 on_each_cpu(hardware_disable, NULL, 1);
2159 kvm_arch_hardware_unsetup();
2161 free_cpumask_var(cpus_hardware_enabled);
2162 __free_page(bad_page);
2164 EXPORT_SYMBOL_GPL(kvm_exit);