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>
53 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
54 #include "coalesced_mmio.h"
57 #define CREATE_TRACE_POINTS
58 #include <trace/events/kvm.h>
60 MODULE_AUTHOR("Qumranet");
61 MODULE_LICENSE("GPL");
66 * kvm->slots_lock --> kvm->lock --> kvm->irq_lock
69 DEFINE_SPINLOCK(kvm_lock);
72 static cpumask_var_t cpus_hardware_enabled;
73 static int kvm_usage_count = 0;
74 static atomic_t hardware_enable_failed;
76 struct kmem_cache *kvm_vcpu_cache;
77 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
79 static __read_mostly struct preempt_ops kvm_preempt_ops;
81 struct dentry *kvm_debugfs_dir;
83 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
85 static int hardware_enable_all(void);
86 static void hardware_disable_all(void);
88 static bool kvm_rebooting;
90 static bool largepages_enabled = true;
92 inline int kvm_is_mmio_pfn(pfn_t pfn)
95 struct page *page = compound_head(pfn_to_page(pfn));
96 return PageReserved(page);
103 * Switches to specified vcpu, until a matching vcpu_put()
105 void vcpu_load(struct kvm_vcpu *vcpu)
109 mutex_lock(&vcpu->mutex);
111 preempt_notifier_register(&vcpu->preempt_notifier);
112 kvm_arch_vcpu_load(vcpu, cpu);
116 void vcpu_put(struct kvm_vcpu *vcpu)
119 kvm_arch_vcpu_put(vcpu);
120 preempt_notifier_unregister(&vcpu->preempt_notifier);
122 mutex_unlock(&vcpu->mutex);
125 static void ack_flush(void *_completed)
129 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
134 struct kvm_vcpu *vcpu;
136 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
138 spin_lock(&kvm->requests_lock);
139 me = smp_processor_id();
140 kvm_for_each_vcpu(i, vcpu, kvm) {
141 if (test_and_set_bit(req, &vcpu->requests))
144 if (cpus != NULL && cpu != -1 && cpu != me)
145 cpumask_set_cpu(cpu, cpus);
147 if (unlikely(cpus == NULL))
148 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
149 else if (!cpumask_empty(cpus))
150 smp_call_function_many(cpus, ack_flush, NULL, 1);
153 spin_unlock(&kvm->requests_lock);
154 free_cpumask_var(cpus);
158 void kvm_flush_remote_tlbs(struct kvm *kvm)
160 if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
161 ++kvm->stat.remote_tlb_flush;
164 void kvm_reload_remote_mmus(struct kvm *kvm)
166 make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
169 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
174 mutex_init(&vcpu->mutex);
178 init_waitqueue_head(&vcpu->wq);
180 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
185 vcpu->run = page_address(page);
187 r = kvm_arch_vcpu_init(vcpu);
193 free_page((unsigned long)vcpu->run);
197 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
199 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
201 kvm_arch_vcpu_uninit(vcpu);
202 free_page((unsigned long)vcpu->run);
204 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
206 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
207 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
209 return container_of(mn, struct kvm, mmu_notifier);
212 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
213 struct mm_struct *mm,
214 unsigned long address)
216 struct kvm *kvm = mmu_notifier_to_kvm(mn);
220 * When ->invalidate_page runs, the linux pte has been zapped
221 * already but the page is still allocated until
222 * ->invalidate_page returns. So if we increase the sequence
223 * here the kvm page fault will notice if the spte can't be
224 * established because the page is going to be freed. If
225 * instead the kvm page fault establishes the spte before
226 * ->invalidate_page runs, kvm_unmap_hva will release it
229 * The sequence increase only need to be seen at spin_unlock
230 * time, and not at spin_lock time.
232 * Increasing the sequence after the spin_unlock would be
233 * unsafe because the kvm page fault could then establish the
234 * pte after kvm_unmap_hva returned, without noticing the page
235 * is going to be freed.
237 spin_lock(&kvm->mmu_lock);
238 kvm->mmu_notifier_seq++;
239 need_tlb_flush = kvm_unmap_hva(kvm, address);
240 spin_unlock(&kvm->mmu_lock);
242 /* we've to flush the tlb before the pages can be freed */
244 kvm_flush_remote_tlbs(kvm);
248 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
249 struct mm_struct *mm,
250 unsigned long address,
253 struct kvm *kvm = mmu_notifier_to_kvm(mn);
255 spin_lock(&kvm->mmu_lock);
256 kvm->mmu_notifier_seq++;
257 kvm_set_spte_hva(kvm, address, pte);
258 spin_unlock(&kvm->mmu_lock);
261 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
262 struct mm_struct *mm,
266 struct kvm *kvm = mmu_notifier_to_kvm(mn);
267 int need_tlb_flush = 0;
269 spin_lock(&kvm->mmu_lock);
271 * The count increase must become visible at unlock time as no
272 * spte can be established without taking the mmu_lock and
273 * count is also read inside the mmu_lock critical section.
275 kvm->mmu_notifier_count++;
276 for (; start < end; start += PAGE_SIZE)
277 need_tlb_flush |= kvm_unmap_hva(kvm, start);
278 spin_unlock(&kvm->mmu_lock);
280 /* we've to flush the tlb before the pages can be freed */
282 kvm_flush_remote_tlbs(kvm);
285 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
286 struct mm_struct *mm,
290 struct kvm *kvm = mmu_notifier_to_kvm(mn);
292 spin_lock(&kvm->mmu_lock);
294 * This sequence increase will notify the kvm page fault that
295 * the page that is going to be mapped in the spte could have
298 kvm->mmu_notifier_seq++;
300 * The above sequence increase must be visible before the
301 * below count decrease but both values are read by the kvm
302 * page fault under mmu_lock spinlock so we don't need to add
303 * a smb_wmb() here in between the two.
305 kvm->mmu_notifier_count--;
306 spin_unlock(&kvm->mmu_lock);
308 BUG_ON(kvm->mmu_notifier_count < 0);
311 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
312 struct mm_struct *mm,
313 unsigned long address)
315 struct kvm *kvm = mmu_notifier_to_kvm(mn);
318 spin_lock(&kvm->mmu_lock);
319 young = kvm_age_hva(kvm, address);
320 spin_unlock(&kvm->mmu_lock);
323 kvm_flush_remote_tlbs(kvm);
328 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
329 struct mm_struct *mm)
331 struct kvm *kvm = mmu_notifier_to_kvm(mn);
332 kvm_arch_flush_shadow(kvm);
335 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
336 .invalidate_page = kvm_mmu_notifier_invalidate_page,
337 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
338 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
339 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
340 .change_pte = kvm_mmu_notifier_change_pte,
341 .release = kvm_mmu_notifier_release,
343 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
345 static struct kvm *kvm_create_vm(void)
348 struct kvm *kvm = kvm_arch_create_vm();
349 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
356 r = hardware_enable_all();
358 goto out_err_nodisable;
360 #ifdef CONFIG_HAVE_KVM_IRQCHIP
361 INIT_HLIST_HEAD(&kvm->mask_notifier_list);
362 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
365 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
366 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
371 kvm->coalesced_mmio_ring =
372 (struct kvm_coalesced_mmio_ring *)page_address(page);
375 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
377 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
378 r = mmu_notifier_register(&kvm->mmu_notifier, current->mm);
380 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
388 kvm->mm = current->mm;
389 atomic_inc(&kvm->mm->mm_count);
390 spin_lock_init(&kvm->mmu_lock);
391 spin_lock_init(&kvm->requests_lock);
392 kvm_io_bus_init(&kvm->pio_bus);
393 kvm_eventfd_init(kvm);
394 mutex_init(&kvm->lock);
395 mutex_init(&kvm->irq_lock);
396 kvm_io_bus_init(&kvm->mmio_bus);
397 init_rwsem(&kvm->slots_lock);
398 atomic_set(&kvm->users_count, 1);
399 spin_lock(&kvm_lock);
400 list_add(&kvm->vm_list, &vm_list);
401 spin_unlock(&kvm_lock);
402 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
403 kvm_coalesced_mmio_init(kvm);
409 hardware_disable_all();
416 * Free any memory in @free but not in @dont.
418 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
419 struct kvm_memory_slot *dont)
423 if (!dont || free->rmap != dont->rmap)
426 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
427 vfree(free->dirty_bitmap);
430 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
431 if (!dont || free->lpage_info[i] != dont->lpage_info[i]) {
432 vfree(free->lpage_info[i]);
433 free->lpage_info[i] = NULL;
438 free->dirty_bitmap = NULL;
442 void kvm_free_physmem(struct kvm *kvm)
446 for (i = 0; i < kvm->nmemslots; ++i)
447 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
450 static void kvm_destroy_vm(struct kvm *kvm)
452 struct mm_struct *mm = kvm->mm;
454 kvm_arch_sync_events(kvm);
455 spin_lock(&kvm_lock);
456 list_del(&kvm->vm_list);
457 spin_unlock(&kvm_lock);
458 kvm_free_irq_routing(kvm);
459 kvm_io_bus_destroy(&kvm->pio_bus);
460 kvm_io_bus_destroy(&kvm->mmio_bus);
461 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
462 if (kvm->coalesced_mmio_ring != NULL)
463 free_page((unsigned long)kvm->coalesced_mmio_ring);
465 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
466 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
468 kvm_arch_flush_shadow(kvm);
470 kvm_arch_destroy_vm(kvm);
471 hardware_disable_all();
475 void kvm_get_kvm(struct kvm *kvm)
477 atomic_inc(&kvm->users_count);
479 EXPORT_SYMBOL_GPL(kvm_get_kvm);
481 void kvm_put_kvm(struct kvm *kvm)
483 if (atomic_dec_and_test(&kvm->users_count))
486 EXPORT_SYMBOL_GPL(kvm_put_kvm);
489 static int kvm_vm_release(struct inode *inode, struct file *filp)
491 struct kvm *kvm = filp->private_data;
493 kvm_irqfd_release(kvm);
500 * Allocate some memory and give it an address in the guest physical address
503 * Discontiguous memory is allowed, mostly for framebuffers.
505 * Must be called holding mmap_sem for write.
507 int __kvm_set_memory_region(struct kvm *kvm,
508 struct kvm_userspace_memory_region *mem,
513 unsigned long npages;
515 struct kvm_memory_slot *memslot;
516 struct kvm_memory_slot old, new;
519 /* General sanity checks */
520 if (mem->memory_size & (PAGE_SIZE - 1))
522 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
524 if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
526 if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
528 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
531 memslot = &kvm->memslots[mem->slot];
532 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
533 npages = mem->memory_size >> PAGE_SHIFT;
536 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
538 new = old = *memslot;
540 new.base_gfn = base_gfn;
542 new.flags = mem->flags;
544 /* Disallow changing a memory slot's size. */
546 if (npages && old.npages && npages != old.npages)
549 /* Check for overlaps */
551 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
552 struct kvm_memory_slot *s = &kvm->memslots[i];
554 if (s == memslot || !s->npages)
556 if (!((base_gfn + npages <= s->base_gfn) ||
557 (base_gfn >= s->base_gfn + s->npages)))
561 /* Free page dirty bitmap if unneeded */
562 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
563 new.dirty_bitmap = NULL;
567 /* Allocate if a slot is being created */
569 if (npages && !new.rmap) {
570 new.rmap = vmalloc(npages * sizeof(struct page *));
575 memset(new.rmap, 0, npages * sizeof(*new.rmap));
577 new.user_alloc = user_alloc;
579 * hva_to_rmmap() serialzies with the mmu_lock and to be
580 * safe it has to ignore memslots with !user_alloc &&
584 new.userspace_addr = mem->userspace_addr;
586 new.userspace_addr = 0;
591 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
597 /* Avoid unused variable warning if no large pages */
600 if (new.lpage_info[i])
603 lpages = 1 + (base_gfn + npages - 1) /
604 KVM_PAGES_PER_HPAGE(level);
605 lpages -= base_gfn / KVM_PAGES_PER_HPAGE(level);
607 new.lpage_info[i] = vmalloc(lpages * sizeof(*new.lpage_info[i]));
609 if (!new.lpage_info[i])
612 memset(new.lpage_info[i], 0,
613 lpages * sizeof(*new.lpage_info[i]));
615 if (base_gfn % KVM_PAGES_PER_HPAGE(level))
616 new.lpage_info[i][0].write_count = 1;
617 if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE(level))
618 new.lpage_info[i][lpages - 1].write_count = 1;
619 ugfn = new.userspace_addr >> PAGE_SHIFT;
621 * If the gfn and userspace address are not aligned wrt each
622 * other, or if explicitly asked to, disable large page
623 * support for this slot
625 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
627 for (j = 0; j < lpages; ++j)
628 new.lpage_info[i][j].write_count = 1;
633 /* Allocate page dirty bitmap if needed */
634 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
635 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
637 new.dirty_bitmap = vmalloc(dirty_bytes);
638 if (!new.dirty_bitmap)
640 memset(new.dirty_bitmap, 0, dirty_bytes);
642 kvm_arch_flush_shadow(kvm);
644 #else /* not defined CONFIG_S390 */
645 new.user_alloc = user_alloc;
647 new.userspace_addr = mem->userspace_addr;
648 #endif /* not defined CONFIG_S390 */
651 kvm_arch_flush_shadow(kvm);
653 spin_lock(&kvm->mmu_lock);
654 if (mem->slot >= kvm->nmemslots)
655 kvm->nmemslots = mem->slot + 1;
658 spin_unlock(&kvm->mmu_lock);
660 r = kvm_arch_set_memory_region(kvm, mem, old, user_alloc);
662 spin_lock(&kvm->mmu_lock);
664 spin_unlock(&kvm->mmu_lock);
668 kvm_free_physmem_slot(&old, npages ? &new : NULL);
669 /* Slot deletion case: we have to update the current slot */
670 spin_lock(&kvm->mmu_lock);
673 spin_unlock(&kvm->mmu_lock);
675 /* map the pages in iommu page table */
676 r = kvm_iommu_map_pages(kvm, base_gfn, npages);
683 kvm_free_physmem_slot(&new, &old);
688 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
690 int kvm_set_memory_region(struct kvm *kvm,
691 struct kvm_userspace_memory_region *mem,
696 down_write(&kvm->slots_lock);
697 r = __kvm_set_memory_region(kvm, mem, user_alloc);
698 up_write(&kvm->slots_lock);
701 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
703 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
705 kvm_userspace_memory_region *mem,
708 if (mem->slot >= KVM_MEMORY_SLOTS)
710 return kvm_set_memory_region(kvm, mem, user_alloc);
713 int kvm_get_dirty_log(struct kvm *kvm,
714 struct kvm_dirty_log *log, int *is_dirty)
716 struct kvm_memory_slot *memslot;
719 unsigned long any = 0;
722 if (log->slot >= KVM_MEMORY_SLOTS)
725 memslot = &kvm->memslots[log->slot];
727 if (!memslot->dirty_bitmap)
730 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
732 for (i = 0; !any && i < n/sizeof(long); ++i)
733 any = memslot->dirty_bitmap[i];
736 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
747 void kvm_disable_largepages(void)
749 largepages_enabled = false;
751 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
753 int is_error_page(struct page *page)
755 return page == bad_page;
757 EXPORT_SYMBOL_GPL(is_error_page);
759 int is_error_pfn(pfn_t pfn)
761 return pfn == bad_pfn;
763 EXPORT_SYMBOL_GPL(is_error_pfn);
765 static inline unsigned long bad_hva(void)
770 int kvm_is_error_hva(unsigned long addr)
772 return addr == bad_hva();
774 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
776 struct kvm_memory_slot *gfn_to_memslot_unaliased(struct kvm *kvm, gfn_t gfn)
780 for (i = 0; i < kvm->nmemslots; ++i) {
781 struct kvm_memory_slot *memslot = &kvm->memslots[i];
783 if (gfn >= memslot->base_gfn
784 && gfn < memslot->base_gfn + memslot->npages)
789 EXPORT_SYMBOL_GPL(gfn_to_memslot_unaliased);
791 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
793 gfn = unalias_gfn(kvm, gfn);
794 return gfn_to_memslot_unaliased(kvm, gfn);
797 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
801 gfn = unalias_gfn(kvm, gfn);
802 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
803 struct kvm_memory_slot *memslot = &kvm->memslots[i];
805 if (gfn >= memslot->base_gfn
806 && gfn < memslot->base_gfn + memslot->npages)
811 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
813 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
815 struct kvm_memory_slot *slot;
817 gfn = unalias_gfn(kvm, gfn);
818 slot = gfn_to_memslot_unaliased(kvm, gfn);
821 return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
823 EXPORT_SYMBOL_GPL(gfn_to_hva);
825 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
827 struct page *page[1];
834 addr = gfn_to_hva(kvm, gfn);
835 if (kvm_is_error_hva(addr)) {
837 return page_to_pfn(bad_page);
840 npages = get_user_pages_fast(addr, 1, 1, page);
842 if (unlikely(npages != 1)) {
843 struct vm_area_struct *vma;
845 down_read(¤t->mm->mmap_sem);
846 vma = find_vma(current->mm, addr);
848 if (vma == NULL || addr < vma->vm_start ||
849 !(vma->vm_flags & VM_PFNMAP)) {
850 up_read(¤t->mm->mmap_sem);
852 return page_to_pfn(bad_page);
855 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
856 up_read(¤t->mm->mmap_sem);
857 BUG_ON(!kvm_is_mmio_pfn(pfn));
859 pfn = page_to_pfn(page[0]);
864 EXPORT_SYMBOL_GPL(gfn_to_pfn);
866 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
870 pfn = gfn_to_pfn(kvm, gfn);
871 if (!kvm_is_mmio_pfn(pfn))
872 return pfn_to_page(pfn);
874 WARN_ON(kvm_is_mmio_pfn(pfn));
880 EXPORT_SYMBOL_GPL(gfn_to_page);
882 void kvm_release_page_clean(struct page *page)
884 kvm_release_pfn_clean(page_to_pfn(page));
886 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
888 void kvm_release_pfn_clean(pfn_t pfn)
890 if (!kvm_is_mmio_pfn(pfn))
891 put_page(pfn_to_page(pfn));
893 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
895 void kvm_release_page_dirty(struct page *page)
897 kvm_release_pfn_dirty(page_to_pfn(page));
899 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
901 void kvm_release_pfn_dirty(pfn_t pfn)
903 kvm_set_pfn_dirty(pfn);
904 kvm_release_pfn_clean(pfn);
906 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
908 void kvm_set_page_dirty(struct page *page)
910 kvm_set_pfn_dirty(page_to_pfn(page));
912 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
914 void kvm_set_pfn_dirty(pfn_t pfn)
916 if (!kvm_is_mmio_pfn(pfn)) {
917 struct page *page = pfn_to_page(pfn);
918 if (!PageReserved(page))
922 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
924 void kvm_set_pfn_accessed(pfn_t pfn)
926 if (!kvm_is_mmio_pfn(pfn))
927 mark_page_accessed(pfn_to_page(pfn));
929 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
931 void kvm_get_pfn(pfn_t pfn)
933 if (!kvm_is_mmio_pfn(pfn))
934 get_page(pfn_to_page(pfn));
936 EXPORT_SYMBOL_GPL(kvm_get_pfn);
938 static int next_segment(unsigned long len, int offset)
940 if (len > PAGE_SIZE - offset)
941 return PAGE_SIZE - offset;
946 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
952 addr = gfn_to_hva(kvm, gfn);
953 if (kvm_is_error_hva(addr))
955 r = copy_from_user(data, (void __user *)addr + offset, len);
960 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
962 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
964 gfn_t gfn = gpa >> PAGE_SHIFT;
966 int offset = offset_in_page(gpa);
969 while ((seg = next_segment(len, offset)) != 0) {
970 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
980 EXPORT_SYMBOL_GPL(kvm_read_guest);
982 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
987 gfn_t gfn = gpa >> PAGE_SHIFT;
988 int offset = offset_in_page(gpa);
990 addr = gfn_to_hva(kvm, gfn);
991 if (kvm_is_error_hva(addr))
994 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1000 EXPORT_SYMBOL(kvm_read_guest_atomic);
1002 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1003 int offset, int len)
1008 addr = gfn_to_hva(kvm, gfn);
1009 if (kvm_is_error_hva(addr))
1011 r = copy_to_user((void __user *)addr + offset, data, len);
1014 mark_page_dirty(kvm, gfn);
1017 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1019 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1022 gfn_t gfn = gpa >> PAGE_SHIFT;
1024 int offset = offset_in_page(gpa);
1027 while ((seg = next_segment(len, offset)) != 0) {
1028 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1039 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1041 return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
1043 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1045 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1047 gfn_t gfn = gpa >> PAGE_SHIFT;
1049 int offset = offset_in_page(gpa);
1052 while ((seg = next_segment(len, offset)) != 0) {
1053 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1062 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1064 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1066 struct kvm_memory_slot *memslot;
1068 gfn = unalias_gfn(kvm, gfn);
1069 memslot = gfn_to_memslot_unaliased(kvm, gfn);
1070 if (memslot && memslot->dirty_bitmap) {
1071 unsigned long rel_gfn = gfn - memslot->base_gfn;
1074 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
1075 set_bit(rel_gfn, memslot->dirty_bitmap);
1080 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1082 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1087 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1089 if (kvm_arch_vcpu_runnable(vcpu)) {
1090 set_bit(KVM_REQ_UNHALT, &vcpu->requests);
1093 if (kvm_cpu_has_pending_timer(vcpu))
1095 if (signal_pending(current))
1101 finish_wait(&vcpu->wq, &wait);
1104 void kvm_resched(struct kvm_vcpu *vcpu)
1106 if (!need_resched())
1110 EXPORT_SYMBOL_GPL(kvm_resched);
1112 void kvm_vcpu_on_spin(struct kvm_vcpu *vcpu)
1117 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1119 /* Sleep for 100 us, and hope lock-holder got scheduled */
1120 expires = ktime_add_ns(ktime_get(), 100000UL);
1121 schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
1123 finish_wait(&vcpu->wq, &wait);
1125 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1127 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1129 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1132 if (vmf->pgoff == 0)
1133 page = virt_to_page(vcpu->run);
1135 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1136 page = virt_to_page(vcpu->arch.pio_data);
1138 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1139 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1140 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1143 return VM_FAULT_SIGBUS;
1149 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1150 .fault = kvm_vcpu_fault,
1153 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1155 vma->vm_ops = &kvm_vcpu_vm_ops;
1159 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1161 struct kvm_vcpu *vcpu = filp->private_data;
1163 kvm_put_kvm(vcpu->kvm);
1167 static struct file_operations kvm_vcpu_fops = {
1168 .release = kvm_vcpu_release,
1169 .unlocked_ioctl = kvm_vcpu_ioctl,
1170 .compat_ioctl = kvm_vcpu_ioctl,
1171 .mmap = kvm_vcpu_mmap,
1175 * Allocates an inode for the vcpu.
1177 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1179 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, 0);
1183 * Creates some virtual cpus. Good luck creating more than one.
1185 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1188 struct kvm_vcpu *vcpu, *v;
1190 vcpu = kvm_arch_vcpu_create(kvm, id);
1192 return PTR_ERR(vcpu);
1194 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1196 r = kvm_arch_vcpu_setup(vcpu);
1200 mutex_lock(&kvm->lock);
1201 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1206 kvm_for_each_vcpu(r, v, kvm)
1207 if (v->vcpu_id == id) {
1212 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1214 /* Now it's all set up, let userspace reach it */
1216 r = create_vcpu_fd(vcpu);
1222 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1224 atomic_inc(&kvm->online_vcpus);
1226 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1227 if (kvm->bsp_vcpu_id == id)
1228 kvm->bsp_vcpu = vcpu;
1230 mutex_unlock(&kvm->lock);
1234 mutex_unlock(&kvm->lock);
1235 kvm_arch_vcpu_destroy(vcpu);
1239 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1242 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1243 vcpu->sigset_active = 1;
1244 vcpu->sigset = *sigset;
1246 vcpu->sigset_active = 0;
1250 static long kvm_vcpu_ioctl(struct file *filp,
1251 unsigned int ioctl, unsigned long arg)
1253 struct kvm_vcpu *vcpu = filp->private_data;
1254 void __user *argp = (void __user *)arg;
1256 struct kvm_fpu *fpu = NULL;
1257 struct kvm_sregs *kvm_sregs = NULL;
1259 if (vcpu->kvm->mm != current->mm)
1266 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1268 case KVM_GET_REGS: {
1269 struct kvm_regs *kvm_regs;
1272 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1275 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1279 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1286 case KVM_SET_REGS: {
1287 struct kvm_regs *kvm_regs;
1290 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1294 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1296 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1304 case KVM_GET_SREGS: {
1305 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1309 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1313 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1318 case KVM_SET_SREGS: {
1319 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1324 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1326 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1332 case KVM_GET_MP_STATE: {
1333 struct kvm_mp_state mp_state;
1335 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1339 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1344 case KVM_SET_MP_STATE: {
1345 struct kvm_mp_state mp_state;
1348 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1350 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1356 case KVM_TRANSLATE: {
1357 struct kvm_translation tr;
1360 if (copy_from_user(&tr, argp, sizeof tr))
1362 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1366 if (copy_to_user(argp, &tr, sizeof tr))
1371 case KVM_SET_GUEST_DEBUG: {
1372 struct kvm_guest_debug dbg;
1375 if (copy_from_user(&dbg, argp, sizeof dbg))
1377 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1383 case KVM_SET_SIGNAL_MASK: {
1384 struct kvm_signal_mask __user *sigmask_arg = argp;
1385 struct kvm_signal_mask kvm_sigmask;
1386 sigset_t sigset, *p;
1391 if (copy_from_user(&kvm_sigmask, argp,
1392 sizeof kvm_sigmask))
1395 if (kvm_sigmask.len != sizeof sigset)
1398 if (copy_from_user(&sigset, sigmask_arg->sigset,
1403 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1407 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1411 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1415 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1421 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1426 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1428 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1435 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1443 static long kvm_vm_ioctl(struct file *filp,
1444 unsigned int ioctl, unsigned long arg)
1446 struct kvm *kvm = filp->private_data;
1447 void __user *argp = (void __user *)arg;
1450 if (kvm->mm != current->mm)
1453 case KVM_CREATE_VCPU:
1454 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1458 case KVM_SET_USER_MEMORY_REGION: {
1459 struct kvm_userspace_memory_region kvm_userspace_mem;
1462 if (copy_from_user(&kvm_userspace_mem, argp,
1463 sizeof kvm_userspace_mem))
1466 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1471 case KVM_GET_DIRTY_LOG: {
1472 struct kvm_dirty_log log;
1475 if (copy_from_user(&log, argp, sizeof log))
1477 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1482 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1483 case KVM_REGISTER_COALESCED_MMIO: {
1484 struct kvm_coalesced_mmio_zone zone;
1486 if (copy_from_user(&zone, argp, sizeof zone))
1489 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1495 case KVM_UNREGISTER_COALESCED_MMIO: {
1496 struct kvm_coalesced_mmio_zone zone;
1498 if (copy_from_user(&zone, argp, sizeof zone))
1501 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1509 struct kvm_irqfd data;
1512 if (copy_from_user(&data, argp, sizeof data))
1514 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
1517 case KVM_IOEVENTFD: {
1518 struct kvm_ioeventfd data;
1521 if (copy_from_user(&data, argp, sizeof data))
1523 r = kvm_ioeventfd(kvm, &data);
1526 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1527 case KVM_SET_BOOT_CPU_ID:
1529 mutex_lock(&kvm->lock);
1530 if (atomic_read(&kvm->online_vcpus) != 0)
1533 kvm->bsp_vcpu_id = arg;
1534 mutex_unlock(&kvm->lock);
1538 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1540 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
1546 #ifdef CONFIG_COMPAT
1547 struct compat_kvm_dirty_log {
1551 compat_uptr_t dirty_bitmap; /* one bit per page */
1556 static long kvm_vm_compat_ioctl(struct file *filp,
1557 unsigned int ioctl, unsigned long arg)
1559 struct kvm *kvm = filp->private_data;
1562 if (kvm->mm != current->mm)
1565 case KVM_GET_DIRTY_LOG: {
1566 struct compat_kvm_dirty_log compat_log;
1567 struct kvm_dirty_log log;
1570 if (copy_from_user(&compat_log, (void __user *)arg,
1571 sizeof(compat_log)))
1573 log.slot = compat_log.slot;
1574 log.padding1 = compat_log.padding1;
1575 log.padding2 = compat_log.padding2;
1576 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
1578 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1584 r = kvm_vm_ioctl(filp, ioctl, arg);
1592 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1594 struct page *page[1];
1597 gfn_t gfn = vmf->pgoff;
1598 struct kvm *kvm = vma->vm_file->private_data;
1600 addr = gfn_to_hva(kvm, gfn);
1601 if (kvm_is_error_hva(addr))
1602 return VM_FAULT_SIGBUS;
1604 npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
1606 if (unlikely(npages != 1))
1607 return VM_FAULT_SIGBUS;
1609 vmf->page = page[0];
1613 static const struct vm_operations_struct kvm_vm_vm_ops = {
1614 .fault = kvm_vm_fault,
1617 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1619 vma->vm_ops = &kvm_vm_vm_ops;
1623 static struct file_operations kvm_vm_fops = {
1624 .release = kvm_vm_release,
1625 .unlocked_ioctl = kvm_vm_ioctl,
1626 #ifdef CONFIG_COMPAT
1627 .compat_ioctl = kvm_vm_compat_ioctl,
1629 .mmap = kvm_vm_mmap,
1632 static int kvm_dev_ioctl_create_vm(void)
1637 kvm = kvm_create_vm();
1639 return PTR_ERR(kvm);
1640 fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, 0);
1647 static long kvm_dev_ioctl_check_extension_generic(long arg)
1650 case KVM_CAP_USER_MEMORY:
1651 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
1652 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
1653 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1654 case KVM_CAP_SET_BOOT_CPU_ID:
1656 case KVM_CAP_INTERNAL_ERROR_DATA:
1658 #ifdef CONFIG_HAVE_KVM_IRQCHIP
1659 case KVM_CAP_IRQ_ROUTING:
1660 return KVM_MAX_IRQ_ROUTES;
1665 return kvm_dev_ioctl_check_extension(arg);
1668 static long kvm_dev_ioctl(struct file *filp,
1669 unsigned int ioctl, unsigned long arg)
1674 case KVM_GET_API_VERSION:
1678 r = KVM_API_VERSION;
1684 r = kvm_dev_ioctl_create_vm();
1686 case KVM_CHECK_EXTENSION:
1687 r = kvm_dev_ioctl_check_extension_generic(arg);
1689 case KVM_GET_VCPU_MMAP_SIZE:
1693 r = PAGE_SIZE; /* struct kvm_run */
1695 r += PAGE_SIZE; /* pio data page */
1697 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1698 r += PAGE_SIZE; /* coalesced mmio ring page */
1701 case KVM_TRACE_ENABLE:
1702 case KVM_TRACE_PAUSE:
1703 case KVM_TRACE_DISABLE:
1707 return kvm_arch_dev_ioctl(filp, ioctl, arg);
1713 static struct file_operations kvm_chardev_ops = {
1714 .unlocked_ioctl = kvm_dev_ioctl,
1715 .compat_ioctl = kvm_dev_ioctl,
1718 static struct miscdevice kvm_dev = {
1724 static void hardware_enable(void *junk)
1726 int cpu = raw_smp_processor_id();
1729 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
1732 cpumask_set_cpu(cpu, cpus_hardware_enabled);
1734 r = kvm_arch_hardware_enable(NULL);
1737 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
1738 atomic_inc(&hardware_enable_failed);
1739 printk(KERN_INFO "kvm: enabling virtualization on "
1740 "CPU%d failed\n", cpu);
1744 static void hardware_disable(void *junk)
1746 int cpu = raw_smp_processor_id();
1748 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
1750 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
1751 kvm_arch_hardware_disable(NULL);
1754 static void hardware_disable_all_nolock(void)
1756 BUG_ON(!kvm_usage_count);
1759 if (!kvm_usage_count)
1760 on_each_cpu(hardware_disable, NULL, 1);
1763 static void hardware_disable_all(void)
1765 spin_lock(&kvm_lock);
1766 hardware_disable_all_nolock();
1767 spin_unlock(&kvm_lock);
1770 static int hardware_enable_all(void)
1774 spin_lock(&kvm_lock);
1777 if (kvm_usage_count == 1) {
1778 atomic_set(&hardware_enable_failed, 0);
1779 on_each_cpu(hardware_enable, NULL, 1);
1781 if (atomic_read(&hardware_enable_failed)) {
1782 hardware_disable_all_nolock();
1787 spin_unlock(&kvm_lock);
1792 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
1797 if (!kvm_usage_count)
1800 val &= ~CPU_TASKS_FROZEN;
1803 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1805 hardware_disable(NULL);
1807 case CPU_UP_CANCELED:
1808 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1810 smp_call_function_single(cpu, hardware_disable, NULL, 1);
1813 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
1815 smp_call_function_single(cpu, hardware_enable, NULL, 1);
1822 asmlinkage void kvm_handle_fault_on_reboot(void)
1825 /* spin while reset goes on */
1828 /* Fault while not rebooting. We want the trace. */
1831 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
1833 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
1837 * Some (well, at least mine) BIOSes hang on reboot if
1840 * And Intel TXT required VMX off for all cpu when system shutdown.
1842 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
1843 kvm_rebooting = true;
1844 on_each_cpu(hardware_disable, NULL, 1);
1848 static struct notifier_block kvm_reboot_notifier = {
1849 .notifier_call = kvm_reboot,
1853 void kvm_io_bus_init(struct kvm_io_bus *bus)
1855 memset(bus, 0, sizeof(*bus));
1858 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
1862 for (i = 0; i < bus->dev_count; i++) {
1863 struct kvm_io_device *pos = bus->devs[i];
1865 kvm_iodevice_destructor(pos);
1869 /* kvm_io_bus_write - called under kvm->slots_lock */
1870 int kvm_io_bus_write(struct kvm_io_bus *bus, gpa_t addr,
1871 int len, const void *val)
1874 for (i = 0; i < bus->dev_count; i++)
1875 if (!kvm_iodevice_write(bus->devs[i], addr, len, val))
1880 /* kvm_io_bus_read - called under kvm->slots_lock */
1881 int kvm_io_bus_read(struct kvm_io_bus *bus, gpa_t addr, int len, void *val)
1884 for (i = 0; i < bus->dev_count; i++)
1885 if (!kvm_iodevice_read(bus->devs[i], addr, len, val))
1890 int kvm_io_bus_register_dev(struct kvm *kvm, struct kvm_io_bus *bus,
1891 struct kvm_io_device *dev)
1895 down_write(&kvm->slots_lock);
1896 ret = __kvm_io_bus_register_dev(bus, dev);
1897 up_write(&kvm->slots_lock);
1902 /* An unlocked version. Caller must have write lock on slots_lock. */
1903 int __kvm_io_bus_register_dev(struct kvm_io_bus *bus,
1904 struct kvm_io_device *dev)
1906 if (bus->dev_count > NR_IOBUS_DEVS-1)
1909 bus->devs[bus->dev_count++] = dev;
1914 void kvm_io_bus_unregister_dev(struct kvm *kvm,
1915 struct kvm_io_bus *bus,
1916 struct kvm_io_device *dev)
1918 down_write(&kvm->slots_lock);
1919 __kvm_io_bus_unregister_dev(bus, dev);
1920 up_write(&kvm->slots_lock);
1923 /* An unlocked version. Caller must have write lock on slots_lock. */
1924 void __kvm_io_bus_unregister_dev(struct kvm_io_bus *bus,
1925 struct kvm_io_device *dev)
1929 for (i = 0; i < bus->dev_count; i++)
1930 if (bus->devs[i] == dev) {
1931 bus->devs[i] = bus->devs[--bus->dev_count];
1936 static struct notifier_block kvm_cpu_notifier = {
1937 .notifier_call = kvm_cpu_hotplug,
1938 .priority = 20, /* must be > scheduler priority */
1941 static int vm_stat_get(void *_offset, u64 *val)
1943 unsigned offset = (long)_offset;
1947 spin_lock(&kvm_lock);
1948 list_for_each_entry(kvm, &vm_list, vm_list)
1949 *val += *(u32 *)((void *)kvm + offset);
1950 spin_unlock(&kvm_lock);
1954 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
1956 static int vcpu_stat_get(void *_offset, u64 *val)
1958 unsigned offset = (long)_offset;
1960 struct kvm_vcpu *vcpu;
1964 spin_lock(&kvm_lock);
1965 list_for_each_entry(kvm, &vm_list, vm_list)
1966 kvm_for_each_vcpu(i, vcpu, kvm)
1967 *val += *(u32 *)((void *)vcpu + offset);
1969 spin_unlock(&kvm_lock);
1973 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
1975 static const struct file_operations *stat_fops[] = {
1976 [KVM_STAT_VCPU] = &vcpu_stat_fops,
1977 [KVM_STAT_VM] = &vm_stat_fops,
1980 static void kvm_init_debug(void)
1982 struct kvm_stats_debugfs_item *p;
1984 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
1985 for (p = debugfs_entries; p->name; ++p)
1986 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
1987 (void *)(long)p->offset,
1988 stat_fops[p->kind]);
1991 static void kvm_exit_debug(void)
1993 struct kvm_stats_debugfs_item *p;
1995 for (p = debugfs_entries; p->name; ++p)
1996 debugfs_remove(p->dentry);
1997 debugfs_remove(kvm_debugfs_dir);
2000 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2002 if (kvm_usage_count)
2003 hardware_disable(NULL);
2007 static int kvm_resume(struct sys_device *dev)
2009 if (kvm_usage_count)
2010 hardware_enable(NULL);
2014 static struct sysdev_class kvm_sysdev_class = {
2016 .suspend = kvm_suspend,
2017 .resume = kvm_resume,
2020 static struct sys_device kvm_sysdev = {
2022 .cls = &kvm_sysdev_class,
2025 struct page *bad_page;
2029 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2031 return container_of(pn, struct kvm_vcpu, preempt_notifier);
2034 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2036 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2038 kvm_arch_vcpu_load(vcpu, cpu);
2041 static void kvm_sched_out(struct preempt_notifier *pn,
2042 struct task_struct *next)
2044 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2046 kvm_arch_vcpu_put(vcpu);
2049 int kvm_init(void *opaque, unsigned int vcpu_size,
2050 struct module *module)
2055 r = kvm_arch_init(opaque);
2059 bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2061 if (bad_page == NULL) {
2066 bad_pfn = page_to_pfn(bad_page);
2068 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2073 r = kvm_arch_hardware_setup();
2077 for_each_online_cpu(cpu) {
2078 smp_call_function_single(cpu,
2079 kvm_arch_check_processor_compat,
2085 r = register_cpu_notifier(&kvm_cpu_notifier);
2088 register_reboot_notifier(&kvm_reboot_notifier);
2090 r = sysdev_class_register(&kvm_sysdev_class);
2094 r = sysdev_register(&kvm_sysdev);
2098 /* A kmem cache lets us meet the alignment requirements of fx_save. */
2099 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
2100 __alignof__(struct kvm_vcpu),
2102 if (!kvm_vcpu_cache) {
2107 kvm_chardev_ops.owner = module;
2108 kvm_vm_fops.owner = module;
2109 kvm_vcpu_fops.owner = module;
2111 r = misc_register(&kvm_dev);
2113 printk(KERN_ERR "kvm: misc device register failed\n");
2117 kvm_preempt_ops.sched_in = kvm_sched_in;
2118 kvm_preempt_ops.sched_out = kvm_sched_out;
2125 kmem_cache_destroy(kvm_vcpu_cache);
2127 sysdev_unregister(&kvm_sysdev);
2129 sysdev_class_unregister(&kvm_sysdev_class);
2131 unregister_reboot_notifier(&kvm_reboot_notifier);
2132 unregister_cpu_notifier(&kvm_cpu_notifier);
2135 kvm_arch_hardware_unsetup();
2137 free_cpumask_var(cpus_hardware_enabled);
2139 __free_page(bad_page);
2145 EXPORT_SYMBOL_GPL(kvm_init);
2149 tracepoint_synchronize_unregister();
2151 misc_deregister(&kvm_dev);
2152 kmem_cache_destroy(kvm_vcpu_cache);
2153 sysdev_unregister(&kvm_sysdev);
2154 sysdev_class_unregister(&kvm_sysdev_class);
2155 unregister_reboot_notifier(&kvm_reboot_notifier);
2156 unregister_cpu_notifier(&kvm_cpu_notifier);
2157 on_each_cpu(hardware_disable, NULL, 1);
2158 kvm_arch_hardware_unsetup();
2160 free_cpumask_var(cpus_hardware_enabled);
2161 __free_page(bad_page);
2163 EXPORT_SYMBOL_GPL(kvm_exit);