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>
47 #include <linux/srcu.h>
49 #include <asm/processor.h>
51 #include <asm/uaccess.h>
52 #include <asm/pgtable.h>
53 #include <asm-generic/bitops/le.h>
55 #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->lock --> kvm->slots_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);
217 int need_tlb_flush, idx;
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 idx = srcu_read_lock(&kvm->srcu);
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);
242 srcu_read_unlock(&kvm->srcu, idx);
244 /* we've to flush the tlb before the pages can be freed */
246 kvm_flush_remote_tlbs(kvm);
250 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
251 struct mm_struct *mm,
252 unsigned long address,
255 struct kvm *kvm = mmu_notifier_to_kvm(mn);
258 idx = srcu_read_lock(&kvm->srcu);
259 spin_lock(&kvm->mmu_lock);
260 kvm->mmu_notifier_seq++;
261 kvm_set_spte_hva(kvm, address, pte);
262 spin_unlock(&kvm->mmu_lock);
263 srcu_read_unlock(&kvm->srcu, idx);
266 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
267 struct mm_struct *mm,
271 struct kvm *kvm = mmu_notifier_to_kvm(mn);
272 int need_tlb_flush = 0, idx;
274 idx = srcu_read_lock(&kvm->srcu);
275 spin_lock(&kvm->mmu_lock);
277 * The count increase must become visible at unlock time as no
278 * spte can be established without taking the mmu_lock and
279 * count is also read inside the mmu_lock critical section.
281 kvm->mmu_notifier_count++;
282 for (; start < end; start += PAGE_SIZE)
283 need_tlb_flush |= kvm_unmap_hva(kvm, start);
284 spin_unlock(&kvm->mmu_lock);
285 srcu_read_unlock(&kvm->srcu, idx);
287 /* we've to flush the tlb before the pages can be freed */
289 kvm_flush_remote_tlbs(kvm);
292 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
293 struct mm_struct *mm,
297 struct kvm *kvm = mmu_notifier_to_kvm(mn);
299 spin_lock(&kvm->mmu_lock);
301 * This sequence increase will notify the kvm page fault that
302 * the page that is going to be mapped in the spte could have
305 kvm->mmu_notifier_seq++;
307 * The above sequence increase must be visible before the
308 * below count decrease but both values are read by the kvm
309 * page fault under mmu_lock spinlock so we don't need to add
310 * a smb_wmb() here in between the two.
312 kvm->mmu_notifier_count--;
313 spin_unlock(&kvm->mmu_lock);
315 BUG_ON(kvm->mmu_notifier_count < 0);
318 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
319 struct mm_struct *mm,
320 unsigned long address)
322 struct kvm *kvm = mmu_notifier_to_kvm(mn);
325 idx = srcu_read_lock(&kvm->srcu);
326 spin_lock(&kvm->mmu_lock);
327 young = kvm_age_hva(kvm, address);
328 spin_unlock(&kvm->mmu_lock);
329 srcu_read_unlock(&kvm->srcu, idx);
332 kvm_flush_remote_tlbs(kvm);
337 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
338 struct mm_struct *mm)
340 struct kvm *kvm = mmu_notifier_to_kvm(mn);
341 kvm_arch_flush_shadow(kvm);
344 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
345 .invalidate_page = kvm_mmu_notifier_invalidate_page,
346 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
347 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
348 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
349 .change_pte = kvm_mmu_notifier_change_pte,
350 .release = kvm_mmu_notifier_release,
353 static int kvm_init_mmu_notifier(struct kvm *kvm)
355 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
356 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
359 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
361 static int kvm_init_mmu_notifier(struct kvm *kvm)
366 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
368 static struct kvm *kvm_create_vm(void)
371 struct kvm *kvm = kvm_arch_create_vm();
372 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
379 r = hardware_enable_all();
381 goto out_err_nodisable;
383 #ifdef CONFIG_HAVE_KVM_IRQCHIP
384 INIT_HLIST_HEAD(&kvm->mask_notifier_list);
385 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
389 kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
392 if (init_srcu_struct(&kvm->srcu))
395 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
396 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
398 cleanup_srcu_struct(&kvm->srcu);
402 kvm->coalesced_mmio_ring =
403 (struct kvm_coalesced_mmio_ring *)page_address(page);
406 r = kvm_init_mmu_notifier(kvm);
408 cleanup_srcu_struct(&kvm->srcu);
409 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
415 kvm->mm = current->mm;
416 atomic_inc(&kvm->mm->mm_count);
417 spin_lock_init(&kvm->mmu_lock);
418 spin_lock_init(&kvm->requests_lock);
419 kvm_io_bus_init(&kvm->pio_bus);
420 kvm_eventfd_init(kvm);
421 mutex_init(&kvm->lock);
422 mutex_init(&kvm->irq_lock);
423 kvm_io_bus_init(&kvm->mmio_bus);
424 init_rwsem(&kvm->slots_lock);
425 atomic_set(&kvm->users_count, 1);
426 spin_lock(&kvm_lock);
427 list_add(&kvm->vm_list, &vm_list);
428 spin_unlock(&kvm_lock);
429 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
430 kvm_coalesced_mmio_init(kvm);
436 hardware_disable_all();
438 kfree(kvm->memslots);
444 * Free any memory in @free but not in @dont.
446 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
447 struct kvm_memory_slot *dont)
451 if (!dont || free->rmap != dont->rmap)
454 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
455 vfree(free->dirty_bitmap);
458 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
459 if (!dont || free->lpage_info[i] != dont->lpage_info[i]) {
460 vfree(free->lpage_info[i]);
461 free->lpage_info[i] = NULL;
466 free->dirty_bitmap = NULL;
470 void kvm_free_physmem(struct kvm *kvm)
473 struct kvm_memslots *slots = kvm->memslots;
475 for (i = 0; i < slots->nmemslots; ++i)
476 kvm_free_physmem_slot(&slots->memslots[i], NULL);
478 kfree(kvm->memslots);
481 static void kvm_destroy_vm(struct kvm *kvm)
483 struct mm_struct *mm = kvm->mm;
485 kvm_arch_sync_events(kvm);
486 spin_lock(&kvm_lock);
487 list_del(&kvm->vm_list);
488 spin_unlock(&kvm_lock);
489 kvm_free_irq_routing(kvm);
490 kvm_io_bus_destroy(&kvm->pio_bus);
491 kvm_io_bus_destroy(&kvm->mmio_bus);
492 kvm_coalesced_mmio_free(kvm);
493 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
494 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
496 kvm_arch_flush_shadow(kvm);
498 cleanup_srcu_struct(&kvm->srcu);
499 kvm_arch_destroy_vm(kvm);
500 hardware_disable_all();
504 void kvm_get_kvm(struct kvm *kvm)
506 atomic_inc(&kvm->users_count);
508 EXPORT_SYMBOL_GPL(kvm_get_kvm);
510 void kvm_put_kvm(struct kvm *kvm)
512 if (atomic_dec_and_test(&kvm->users_count))
515 EXPORT_SYMBOL_GPL(kvm_put_kvm);
518 static int kvm_vm_release(struct inode *inode, struct file *filp)
520 struct kvm *kvm = filp->private_data;
522 kvm_irqfd_release(kvm);
529 * Allocate some memory and give it an address in the guest physical address
532 * Discontiguous memory is allowed, mostly for framebuffers.
534 * Must be called holding mmap_sem for write.
536 int __kvm_set_memory_region(struct kvm *kvm,
537 struct kvm_userspace_memory_region *mem,
540 int r, flush_shadow = 0;
542 unsigned long npages;
544 struct kvm_memory_slot *memslot;
545 struct kvm_memory_slot old, new;
546 struct kvm_memslots *slots, *old_memslots;
549 /* General sanity checks */
550 if (mem->memory_size & (PAGE_SIZE - 1))
552 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
554 if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
556 if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
558 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
561 memslot = &kvm->memslots->memslots[mem->slot];
562 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
563 npages = mem->memory_size >> PAGE_SHIFT;
566 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
568 new = old = *memslot;
570 new.base_gfn = base_gfn;
572 new.flags = mem->flags;
574 /* Disallow changing a memory slot's size. */
576 if (npages && old.npages && npages != old.npages)
579 /* Check for overlaps */
581 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
582 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
584 if (s == memslot || !s->npages)
586 if (!((base_gfn + npages <= s->base_gfn) ||
587 (base_gfn >= s->base_gfn + s->npages)))
591 /* Free page dirty bitmap if unneeded */
592 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
593 new.dirty_bitmap = NULL;
597 /* Allocate if a slot is being created */
599 if (npages && !new.rmap) {
600 new.rmap = vmalloc(npages * sizeof(struct page *));
605 memset(new.rmap, 0, npages * sizeof(*new.rmap));
607 new.user_alloc = user_alloc;
608 new.userspace_addr = mem->userspace_addr;
613 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
619 /* Avoid unused variable warning if no large pages */
622 if (new.lpage_info[i])
625 lpages = 1 + (base_gfn + npages - 1) /
626 KVM_PAGES_PER_HPAGE(level);
627 lpages -= base_gfn / KVM_PAGES_PER_HPAGE(level);
629 new.lpage_info[i] = vmalloc(lpages * sizeof(*new.lpage_info[i]));
631 if (!new.lpage_info[i])
634 memset(new.lpage_info[i], 0,
635 lpages * sizeof(*new.lpage_info[i]));
637 if (base_gfn % KVM_PAGES_PER_HPAGE(level))
638 new.lpage_info[i][0].write_count = 1;
639 if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE(level))
640 new.lpage_info[i][lpages - 1].write_count = 1;
641 ugfn = new.userspace_addr >> PAGE_SHIFT;
643 * If the gfn and userspace address are not aligned wrt each
644 * other, or if explicitly asked to, disable large page
645 * support for this slot
647 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
649 for (j = 0; j < lpages; ++j)
650 new.lpage_info[i][j].write_count = 1;
655 /* Allocate page dirty bitmap if needed */
656 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
657 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
659 new.dirty_bitmap = vmalloc(dirty_bytes);
660 if (!new.dirty_bitmap)
662 memset(new.dirty_bitmap, 0, dirty_bytes);
663 /* destroy any largepage mappings for dirty tracking */
667 #else /* not defined CONFIG_S390 */
668 new.user_alloc = user_alloc;
670 new.userspace_addr = mem->userspace_addr;
671 #endif /* not defined CONFIG_S390 */
675 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
678 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
679 if (mem->slot >= slots->nmemslots)
680 slots->nmemslots = mem->slot + 1;
681 slots->memslots[mem->slot].flags |= KVM_MEMSLOT_INVALID;
683 old_memslots = kvm->memslots;
684 rcu_assign_pointer(kvm->memslots, slots);
685 synchronize_srcu_expedited(&kvm->srcu);
686 /* From this point no new shadow pages pointing to a deleted
687 * memslot will be created.
689 * validation of sp->gfn happens in:
690 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
691 * - kvm_is_visible_gfn (mmu_check_roots)
693 kvm_arch_flush_shadow(kvm);
697 r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
702 /* map the pages in iommu page table */
704 r = kvm_iommu_map_pages(kvm, &new);
711 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
714 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
715 if (mem->slot >= slots->nmemslots)
716 slots->nmemslots = mem->slot + 1;
718 /* actual memory is freed via old in kvm_free_physmem_slot below */
721 new.dirty_bitmap = NULL;
722 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i)
723 new.lpage_info[i] = NULL;
726 slots->memslots[mem->slot] = new;
727 old_memslots = kvm->memslots;
728 rcu_assign_pointer(kvm->memslots, slots);
729 synchronize_srcu_expedited(&kvm->srcu);
731 kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
733 kvm_free_physmem_slot(&old, &new);
737 kvm_arch_flush_shadow(kvm);
742 kvm_free_physmem_slot(&new, &old);
747 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
749 int kvm_set_memory_region(struct kvm *kvm,
750 struct kvm_userspace_memory_region *mem,
755 down_write(&kvm->slots_lock);
756 r = __kvm_set_memory_region(kvm, mem, user_alloc);
757 up_write(&kvm->slots_lock);
760 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
762 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
764 kvm_userspace_memory_region *mem,
767 if (mem->slot >= KVM_MEMORY_SLOTS)
769 return kvm_set_memory_region(kvm, mem, user_alloc);
772 int kvm_get_dirty_log(struct kvm *kvm,
773 struct kvm_dirty_log *log, int *is_dirty)
775 struct kvm_memory_slot *memslot;
778 unsigned long any = 0;
781 if (log->slot >= KVM_MEMORY_SLOTS)
784 memslot = &kvm->memslots->memslots[log->slot];
786 if (!memslot->dirty_bitmap)
789 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
791 for (i = 0; !any && i < n/sizeof(long); ++i)
792 any = memslot->dirty_bitmap[i];
795 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
806 void kvm_disable_largepages(void)
808 largepages_enabled = false;
810 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
812 int is_error_page(struct page *page)
814 return page == bad_page;
816 EXPORT_SYMBOL_GPL(is_error_page);
818 int is_error_pfn(pfn_t pfn)
820 return pfn == bad_pfn;
822 EXPORT_SYMBOL_GPL(is_error_pfn);
824 static inline unsigned long bad_hva(void)
829 int kvm_is_error_hva(unsigned long addr)
831 return addr == bad_hva();
833 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
835 struct kvm_memory_slot *gfn_to_memslot_unaliased(struct kvm *kvm, gfn_t gfn)
838 struct kvm_memslots *slots = rcu_dereference(kvm->memslots);
840 for (i = 0; i < slots->nmemslots; ++i) {
841 struct kvm_memory_slot *memslot = &slots->memslots[i];
843 if (gfn >= memslot->base_gfn
844 && gfn < memslot->base_gfn + memslot->npages)
849 EXPORT_SYMBOL_GPL(gfn_to_memslot_unaliased);
851 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
853 gfn = unalias_gfn(kvm, gfn);
854 return gfn_to_memslot_unaliased(kvm, gfn);
857 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
860 struct kvm_memslots *slots = rcu_dereference(kvm->memslots);
862 gfn = unalias_gfn(kvm, gfn);
863 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
864 struct kvm_memory_slot *memslot = &slots->memslots[i];
866 if (memslot->flags & KVM_MEMSLOT_INVALID)
869 if (gfn >= memslot->base_gfn
870 && gfn < memslot->base_gfn + memslot->npages)
875 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
877 int memslot_id(struct kvm *kvm, gfn_t gfn)
880 struct kvm_memslots *slots = rcu_dereference(kvm->memslots);
881 struct kvm_memory_slot *memslot = NULL;
883 gfn = unalias_gfn(kvm, gfn);
884 for (i = 0; i < slots->nmemslots; ++i) {
885 memslot = &slots->memslots[i];
887 if (gfn >= memslot->base_gfn
888 && gfn < memslot->base_gfn + memslot->npages)
892 return memslot - slots->memslots;
895 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
897 struct kvm_memory_slot *slot;
899 gfn = unalias_gfn(kvm, gfn);
900 slot = gfn_to_memslot_unaliased(kvm, gfn);
901 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
903 return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
905 EXPORT_SYMBOL_GPL(gfn_to_hva);
907 static pfn_t hva_to_pfn(struct kvm *kvm, unsigned long addr)
909 struct page *page[1];
915 npages = get_user_pages_fast(addr, 1, 1, page);
917 if (unlikely(npages != 1)) {
918 struct vm_area_struct *vma;
920 down_read(¤t->mm->mmap_sem);
921 vma = find_vma(current->mm, addr);
923 if (vma == NULL || addr < vma->vm_start ||
924 !(vma->vm_flags & VM_PFNMAP)) {
925 up_read(¤t->mm->mmap_sem);
927 return page_to_pfn(bad_page);
930 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
931 up_read(¤t->mm->mmap_sem);
932 BUG_ON(!kvm_is_mmio_pfn(pfn));
934 pfn = page_to_pfn(page[0]);
939 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
943 addr = gfn_to_hva(kvm, gfn);
944 if (kvm_is_error_hva(addr)) {
946 return page_to_pfn(bad_page);
949 return hva_to_pfn(kvm, addr);
951 EXPORT_SYMBOL_GPL(gfn_to_pfn);
953 static unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
955 return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
958 pfn_t gfn_to_pfn_memslot(struct kvm *kvm,
959 struct kvm_memory_slot *slot, gfn_t gfn)
961 unsigned long addr = gfn_to_hva_memslot(slot, gfn);
962 return hva_to_pfn(kvm, addr);
965 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
969 pfn = gfn_to_pfn(kvm, gfn);
970 if (!kvm_is_mmio_pfn(pfn))
971 return pfn_to_page(pfn);
973 WARN_ON(kvm_is_mmio_pfn(pfn));
979 EXPORT_SYMBOL_GPL(gfn_to_page);
981 void kvm_release_page_clean(struct page *page)
983 kvm_release_pfn_clean(page_to_pfn(page));
985 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
987 void kvm_release_pfn_clean(pfn_t pfn)
989 if (!kvm_is_mmio_pfn(pfn))
990 put_page(pfn_to_page(pfn));
992 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
994 void kvm_release_page_dirty(struct page *page)
996 kvm_release_pfn_dirty(page_to_pfn(page));
998 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1000 void kvm_release_pfn_dirty(pfn_t pfn)
1002 kvm_set_pfn_dirty(pfn);
1003 kvm_release_pfn_clean(pfn);
1005 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1007 void kvm_set_page_dirty(struct page *page)
1009 kvm_set_pfn_dirty(page_to_pfn(page));
1011 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1013 void kvm_set_pfn_dirty(pfn_t pfn)
1015 if (!kvm_is_mmio_pfn(pfn)) {
1016 struct page *page = pfn_to_page(pfn);
1017 if (!PageReserved(page))
1021 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1023 void kvm_set_pfn_accessed(pfn_t pfn)
1025 if (!kvm_is_mmio_pfn(pfn))
1026 mark_page_accessed(pfn_to_page(pfn));
1028 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1030 void kvm_get_pfn(pfn_t pfn)
1032 if (!kvm_is_mmio_pfn(pfn))
1033 get_page(pfn_to_page(pfn));
1035 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1037 static int next_segment(unsigned long len, int offset)
1039 if (len > PAGE_SIZE - offset)
1040 return PAGE_SIZE - offset;
1045 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1051 addr = gfn_to_hva(kvm, gfn);
1052 if (kvm_is_error_hva(addr))
1054 r = copy_from_user(data, (void __user *)addr + offset, len);
1059 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1061 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1063 gfn_t gfn = gpa >> PAGE_SHIFT;
1065 int offset = offset_in_page(gpa);
1068 while ((seg = next_segment(len, offset)) != 0) {
1069 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1079 EXPORT_SYMBOL_GPL(kvm_read_guest);
1081 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1086 gfn_t gfn = gpa >> PAGE_SHIFT;
1087 int offset = offset_in_page(gpa);
1089 addr = gfn_to_hva(kvm, gfn);
1090 if (kvm_is_error_hva(addr))
1092 pagefault_disable();
1093 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1099 EXPORT_SYMBOL(kvm_read_guest_atomic);
1101 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1102 int offset, int len)
1107 addr = gfn_to_hva(kvm, gfn);
1108 if (kvm_is_error_hva(addr))
1110 r = copy_to_user((void __user *)addr + offset, data, len);
1113 mark_page_dirty(kvm, gfn);
1116 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1118 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1121 gfn_t gfn = gpa >> PAGE_SHIFT;
1123 int offset = offset_in_page(gpa);
1126 while ((seg = next_segment(len, offset)) != 0) {
1127 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1138 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1140 return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
1142 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1144 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1146 gfn_t gfn = gpa >> PAGE_SHIFT;
1148 int offset = offset_in_page(gpa);
1151 while ((seg = next_segment(len, offset)) != 0) {
1152 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1161 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1163 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1165 struct kvm_memory_slot *memslot;
1167 gfn = unalias_gfn(kvm, gfn);
1168 memslot = gfn_to_memslot_unaliased(kvm, gfn);
1169 if (memslot && memslot->dirty_bitmap) {
1170 unsigned long rel_gfn = gfn - memslot->base_gfn;
1173 if (!generic_test_le_bit(rel_gfn, memslot->dirty_bitmap))
1174 generic___set_le_bit(rel_gfn, memslot->dirty_bitmap);
1179 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1181 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1186 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1188 if (kvm_arch_vcpu_runnable(vcpu)) {
1189 set_bit(KVM_REQ_UNHALT, &vcpu->requests);
1192 if (kvm_cpu_has_pending_timer(vcpu))
1194 if (signal_pending(current))
1200 finish_wait(&vcpu->wq, &wait);
1203 void kvm_resched(struct kvm_vcpu *vcpu)
1205 if (!need_resched())
1209 EXPORT_SYMBOL_GPL(kvm_resched);
1211 void kvm_vcpu_on_spin(struct kvm_vcpu *vcpu)
1216 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1218 /* Sleep for 100 us, and hope lock-holder got scheduled */
1219 expires = ktime_add_ns(ktime_get(), 100000UL);
1220 schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
1222 finish_wait(&vcpu->wq, &wait);
1224 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1226 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1228 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1231 if (vmf->pgoff == 0)
1232 page = virt_to_page(vcpu->run);
1234 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1235 page = virt_to_page(vcpu->arch.pio_data);
1237 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1238 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1239 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1242 return VM_FAULT_SIGBUS;
1248 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1249 .fault = kvm_vcpu_fault,
1252 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1254 vma->vm_ops = &kvm_vcpu_vm_ops;
1258 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1260 struct kvm_vcpu *vcpu = filp->private_data;
1262 kvm_put_kvm(vcpu->kvm);
1266 static struct file_operations kvm_vcpu_fops = {
1267 .release = kvm_vcpu_release,
1268 .unlocked_ioctl = kvm_vcpu_ioctl,
1269 .compat_ioctl = kvm_vcpu_ioctl,
1270 .mmap = kvm_vcpu_mmap,
1274 * Allocates an inode for the vcpu.
1276 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1278 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1282 * Creates some virtual cpus. Good luck creating more than one.
1284 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1287 struct kvm_vcpu *vcpu, *v;
1289 vcpu = kvm_arch_vcpu_create(kvm, id);
1291 return PTR_ERR(vcpu);
1293 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1295 r = kvm_arch_vcpu_setup(vcpu);
1299 mutex_lock(&kvm->lock);
1300 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1305 kvm_for_each_vcpu(r, v, kvm)
1306 if (v->vcpu_id == id) {
1311 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1313 /* Now it's all set up, let userspace reach it */
1315 r = create_vcpu_fd(vcpu);
1321 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1323 atomic_inc(&kvm->online_vcpus);
1325 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1326 if (kvm->bsp_vcpu_id == id)
1327 kvm->bsp_vcpu = vcpu;
1329 mutex_unlock(&kvm->lock);
1333 mutex_unlock(&kvm->lock);
1334 kvm_arch_vcpu_destroy(vcpu);
1338 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1341 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1342 vcpu->sigset_active = 1;
1343 vcpu->sigset = *sigset;
1345 vcpu->sigset_active = 0;
1349 static long kvm_vcpu_ioctl(struct file *filp,
1350 unsigned int ioctl, unsigned long arg)
1352 struct kvm_vcpu *vcpu = filp->private_data;
1353 void __user *argp = (void __user *)arg;
1355 struct kvm_fpu *fpu = NULL;
1356 struct kvm_sregs *kvm_sregs = NULL;
1358 if (vcpu->kvm->mm != current->mm)
1365 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1367 case KVM_GET_REGS: {
1368 struct kvm_regs *kvm_regs;
1371 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1374 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1378 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1385 case KVM_SET_REGS: {
1386 struct kvm_regs *kvm_regs;
1389 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1393 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1395 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1403 case KVM_GET_SREGS: {
1404 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1408 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1412 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1417 case KVM_SET_SREGS: {
1418 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1423 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1425 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1431 case KVM_GET_MP_STATE: {
1432 struct kvm_mp_state mp_state;
1434 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1438 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1443 case KVM_SET_MP_STATE: {
1444 struct kvm_mp_state mp_state;
1447 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1449 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1455 case KVM_TRANSLATE: {
1456 struct kvm_translation tr;
1459 if (copy_from_user(&tr, argp, sizeof tr))
1461 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1465 if (copy_to_user(argp, &tr, sizeof tr))
1470 case KVM_SET_GUEST_DEBUG: {
1471 struct kvm_guest_debug dbg;
1474 if (copy_from_user(&dbg, argp, sizeof dbg))
1476 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1482 case KVM_SET_SIGNAL_MASK: {
1483 struct kvm_signal_mask __user *sigmask_arg = argp;
1484 struct kvm_signal_mask kvm_sigmask;
1485 sigset_t sigset, *p;
1490 if (copy_from_user(&kvm_sigmask, argp,
1491 sizeof kvm_sigmask))
1494 if (kvm_sigmask.len != sizeof sigset)
1497 if (copy_from_user(&sigset, sigmask_arg->sigset,
1502 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1506 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1510 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1514 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1520 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1525 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1527 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1534 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1542 static long kvm_vm_ioctl(struct file *filp,
1543 unsigned int ioctl, unsigned long arg)
1545 struct kvm *kvm = filp->private_data;
1546 void __user *argp = (void __user *)arg;
1549 if (kvm->mm != current->mm)
1552 case KVM_CREATE_VCPU:
1553 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1557 case KVM_SET_USER_MEMORY_REGION: {
1558 struct kvm_userspace_memory_region kvm_userspace_mem;
1561 if (copy_from_user(&kvm_userspace_mem, argp,
1562 sizeof kvm_userspace_mem))
1565 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1570 case KVM_GET_DIRTY_LOG: {
1571 struct kvm_dirty_log log;
1574 if (copy_from_user(&log, argp, sizeof log))
1576 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1581 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1582 case KVM_REGISTER_COALESCED_MMIO: {
1583 struct kvm_coalesced_mmio_zone zone;
1585 if (copy_from_user(&zone, argp, sizeof zone))
1588 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1594 case KVM_UNREGISTER_COALESCED_MMIO: {
1595 struct kvm_coalesced_mmio_zone zone;
1597 if (copy_from_user(&zone, argp, sizeof zone))
1600 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1608 struct kvm_irqfd data;
1611 if (copy_from_user(&data, argp, sizeof data))
1613 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
1616 case KVM_IOEVENTFD: {
1617 struct kvm_ioeventfd data;
1620 if (copy_from_user(&data, argp, sizeof data))
1622 r = kvm_ioeventfd(kvm, &data);
1625 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1626 case KVM_SET_BOOT_CPU_ID:
1628 mutex_lock(&kvm->lock);
1629 if (atomic_read(&kvm->online_vcpus) != 0)
1632 kvm->bsp_vcpu_id = arg;
1633 mutex_unlock(&kvm->lock);
1637 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1639 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
1645 #ifdef CONFIG_COMPAT
1646 struct compat_kvm_dirty_log {
1650 compat_uptr_t dirty_bitmap; /* one bit per page */
1655 static long kvm_vm_compat_ioctl(struct file *filp,
1656 unsigned int ioctl, unsigned long arg)
1658 struct kvm *kvm = filp->private_data;
1661 if (kvm->mm != current->mm)
1664 case KVM_GET_DIRTY_LOG: {
1665 struct compat_kvm_dirty_log compat_log;
1666 struct kvm_dirty_log log;
1669 if (copy_from_user(&compat_log, (void __user *)arg,
1670 sizeof(compat_log)))
1672 log.slot = compat_log.slot;
1673 log.padding1 = compat_log.padding1;
1674 log.padding2 = compat_log.padding2;
1675 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
1677 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1683 r = kvm_vm_ioctl(filp, ioctl, arg);
1691 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1693 struct page *page[1];
1696 gfn_t gfn = vmf->pgoff;
1697 struct kvm *kvm = vma->vm_file->private_data;
1699 addr = gfn_to_hva(kvm, gfn);
1700 if (kvm_is_error_hva(addr))
1701 return VM_FAULT_SIGBUS;
1703 npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
1705 if (unlikely(npages != 1))
1706 return VM_FAULT_SIGBUS;
1708 vmf->page = page[0];
1712 static const struct vm_operations_struct kvm_vm_vm_ops = {
1713 .fault = kvm_vm_fault,
1716 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1718 vma->vm_ops = &kvm_vm_vm_ops;
1722 static struct file_operations kvm_vm_fops = {
1723 .release = kvm_vm_release,
1724 .unlocked_ioctl = kvm_vm_ioctl,
1725 #ifdef CONFIG_COMPAT
1726 .compat_ioctl = kvm_vm_compat_ioctl,
1728 .mmap = kvm_vm_mmap,
1731 static int kvm_dev_ioctl_create_vm(void)
1736 kvm = kvm_create_vm();
1738 return PTR_ERR(kvm);
1739 fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
1746 static long kvm_dev_ioctl_check_extension_generic(long arg)
1749 case KVM_CAP_USER_MEMORY:
1750 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
1751 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
1752 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1753 case KVM_CAP_SET_BOOT_CPU_ID:
1755 case KVM_CAP_INTERNAL_ERROR_DATA:
1757 #ifdef CONFIG_HAVE_KVM_IRQCHIP
1758 case KVM_CAP_IRQ_ROUTING:
1759 return KVM_MAX_IRQ_ROUTES;
1764 return kvm_dev_ioctl_check_extension(arg);
1767 static long kvm_dev_ioctl(struct file *filp,
1768 unsigned int ioctl, unsigned long arg)
1773 case KVM_GET_API_VERSION:
1777 r = KVM_API_VERSION;
1783 r = kvm_dev_ioctl_create_vm();
1785 case KVM_CHECK_EXTENSION:
1786 r = kvm_dev_ioctl_check_extension_generic(arg);
1788 case KVM_GET_VCPU_MMAP_SIZE:
1792 r = PAGE_SIZE; /* struct kvm_run */
1794 r += PAGE_SIZE; /* pio data page */
1796 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1797 r += PAGE_SIZE; /* coalesced mmio ring page */
1800 case KVM_TRACE_ENABLE:
1801 case KVM_TRACE_PAUSE:
1802 case KVM_TRACE_DISABLE:
1806 return kvm_arch_dev_ioctl(filp, ioctl, arg);
1812 static struct file_operations kvm_chardev_ops = {
1813 .unlocked_ioctl = kvm_dev_ioctl,
1814 .compat_ioctl = kvm_dev_ioctl,
1817 static struct miscdevice kvm_dev = {
1823 static void hardware_enable(void *junk)
1825 int cpu = raw_smp_processor_id();
1828 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
1831 cpumask_set_cpu(cpu, cpus_hardware_enabled);
1833 r = kvm_arch_hardware_enable(NULL);
1836 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
1837 atomic_inc(&hardware_enable_failed);
1838 printk(KERN_INFO "kvm: enabling virtualization on "
1839 "CPU%d failed\n", cpu);
1843 static void hardware_disable(void *junk)
1845 int cpu = raw_smp_processor_id();
1847 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
1849 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
1850 kvm_arch_hardware_disable(NULL);
1853 static void hardware_disable_all_nolock(void)
1855 BUG_ON(!kvm_usage_count);
1858 if (!kvm_usage_count)
1859 on_each_cpu(hardware_disable, NULL, 1);
1862 static void hardware_disable_all(void)
1864 spin_lock(&kvm_lock);
1865 hardware_disable_all_nolock();
1866 spin_unlock(&kvm_lock);
1869 static int hardware_enable_all(void)
1873 spin_lock(&kvm_lock);
1876 if (kvm_usage_count == 1) {
1877 atomic_set(&hardware_enable_failed, 0);
1878 on_each_cpu(hardware_enable, NULL, 1);
1880 if (atomic_read(&hardware_enable_failed)) {
1881 hardware_disable_all_nolock();
1886 spin_unlock(&kvm_lock);
1891 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
1896 if (!kvm_usage_count)
1899 val &= ~CPU_TASKS_FROZEN;
1902 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1904 hardware_disable(NULL);
1906 case CPU_UP_CANCELED:
1907 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1909 smp_call_function_single(cpu, hardware_disable, NULL, 1);
1912 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
1914 smp_call_function_single(cpu, hardware_enable, NULL, 1);
1921 asmlinkage void kvm_handle_fault_on_reboot(void)
1924 /* spin while reset goes on */
1927 /* Fault while not rebooting. We want the trace. */
1930 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
1932 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
1936 * Some (well, at least mine) BIOSes hang on reboot if
1939 * And Intel TXT required VMX off for all cpu when system shutdown.
1941 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
1942 kvm_rebooting = true;
1943 on_each_cpu(hardware_disable, NULL, 1);
1947 static struct notifier_block kvm_reboot_notifier = {
1948 .notifier_call = kvm_reboot,
1952 void kvm_io_bus_init(struct kvm_io_bus *bus)
1954 memset(bus, 0, sizeof(*bus));
1957 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
1961 for (i = 0; i < bus->dev_count; i++) {
1962 struct kvm_io_device *pos = bus->devs[i];
1964 kvm_iodevice_destructor(pos);
1968 /* kvm_io_bus_write - called under kvm->slots_lock */
1969 int kvm_io_bus_write(struct kvm_io_bus *bus, gpa_t addr,
1970 int len, const void *val)
1973 for (i = 0; i < bus->dev_count; i++)
1974 if (!kvm_iodevice_write(bus->devs[i], addr, len, val))
1979 /* kvm_io_bus_read - called under kvm->slots_lock */
1980 int kvm_io_bus_read(struct kvm_io_bus *bus, gpa_t addr, int len, void *val)
1983 for (i = 0; i < bus->dev_count; i++)
1984 if (!kvm_iodevice_read(bus->devs[i], addr, len, val))
1989 int kvm_io_bus_register_dev(struct kvm *kvm, struct kvm_io_bus *bus,
1990 struct kvm_io_device *dev)
1994 down_write(&kvm->slots_lock);
1995 ret = __kvm_io_bus_register_dev(bus, dev);
1996 up_write(&kvm->slots_lock);
2001 /* An unlocked version. Caller must have write lock on slots_lock. */
2002 int __kvm_io_bus_register_dev(struct kvm_io_bus *bus,
2003 struct kvm_io_device *dev)
2005 if (bus->dev_count > NR_IOBUS_DEVS-1)
2008 bus->devs[bus->dev_count++] = dev;
2013 void kvm_io_bus_unregister_dev(struct kvm *kvm,
2014 struct kvm_io_bus *bus,
2015 struct kvm_io_device *dev)
2017 down_write(&kvm->slots_lock);
2018 __kvm_io_bus_unregister_dev(bus, dev);
2019 up_write(&kvm->slots_lock);
2022 /* An unlocked version. Caller must have write lock on slots_lock. */
2023 void __kvm_io_bus_unregister_dev(struct kvm_io_bus *bus,
2024 struct kvm_io_device *dev)
2028 for (i = 0; i < bus->dev_count; i++)
2029 if (bus->devs[i] == dev) {
2030 bus->devs[i] = bus->devs[--bus->dev_count];
2035 static struct notifier_block kvm_cpu_notifier = {
2036 .notifier_call = kvm_cpu_hotplug,
2037 .priority = 20, /* must be > scheduler priority */
2040 static int vm_stat_get(void *_offset, u64 *val)
2042 unsigned offset = (long)_offset;
2046 spin_lock(&kvm_lock);
2047 list_for_each_entry(kvm, &vm_list, vm_list)
2048 *val += *(u32 *)((void *)kvm + offset);
2049 spin_unlock(&kvm_lock);
2053 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2055 static int vcpu_stat_get(void *_offset, u64 *val)
2057 unsigned offset = (long)_offset;
2059 struct kvm_vcpu *vcpu;
2063 spin_lock(&kvm_lock);
2064 list_for_each_entry(kvm, &vm_list, vm_list)
2065 kvm_for_each_vcpu(i, vcpu, kvm)
2066 *val += *(u32 *)((void *)vcpu + offset);
2068 spin_unlock(&kvm_lock);
2072 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2074 static const struct file_operations *stat_fops[] = {
2075 [KVM_STAT_VCPU] = &vcpu_stat_fops,
2076 [KVM_STAT_VM] = &vm_stat_fops,
2079 static void kvm_init_debug(void)
2081 struct kvm_stats_debugfs_item *p;
2083 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2084 for (p = debugfs_entries; p->name; ++p)
2085 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2086 (void *)(long)p->offset,
2087 stat_fops[p->kind]);
2090 static void kvm_exit_debug(void)
2092 struct kvm_stats_debugfs_item *p;
2094 for (p = debugfs_entries; p->name; ++p)
2095 debugfs_remove(p->dentry);
2096 debugfs_remove(kvm_debugfs_dir);
2099 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2101 if (kvm_usage_count)
2102 hardware_disable(NULL);
2106 static int kvm_resume(struct sys_device *dev)
2108 if (kvm_usage_count)
2109 hardware_enable(NULL);
2113 static struct sysdev_class kvm_sysdev_class = {
2115 .suspend = kvm_suspend,
2116 .resume = kvm_resume,
2119 static struct sys_device kvm_sysdev = {
2121 .cls = &kvm_sysdev_class,
2124 struct page *bad_page;
2128 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2130 return container_of(pn, struct kvm_vcpu, preempt_notifier);
2133 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2135 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2137 kvm_arch_vcpu_load(vcpu, cpu);
2140 static void kvm_sched_out(struct preempt_notifier *pn,
2141 struct task_struct *next)
2143 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2145 kvm_arch_vcpu_put(vcpu);
2148 int kvm_init(void *opaque, unsigned int vcpu_size,
2149 struct module *module)
2154 r = kvm_arch_init(opaque);
2158 bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2160 if (bad_page == NULL) {
2165 bad_pfn = page_to_pfn(bad_page);
2167 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2172 r = kvm_arch_hardware_setup();
2176 for_each_online_cpu(cpu) {
2177 smp_call_function_single(cpu,
2178 kvm_arch_check_processor_compat,
2184 r = register_cpu_notifier(&kvm_cpu_notifier);
2187 register_reboot_notifier(&kvm_reboot_notifier);
2189 r = sysdev_class_register(&kvm_sysdev_class);
2193 r = sysdev_register(&kvm_sysdev);
2197 /* A kmem cache lets us meet the alignment requirements of fx_save. */
2198 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
2199 __alignof__(struct kvm_vcpu),
2201 if (!kvm_vcpu_cache) {
2206 kvm_chardev_ops.owner = module;
2207 kvm_vm_fops.owner = module;
2208 kvm_vcpu_fops.owner = module;
2210 r = misc_register(&kvm_dev);
2212 printk(KERN_ERR "kvm: misc device register failed\n");
2216 kvm_preempt_ops.sched_in = kvm_sched_in;
2217 kvm_preempt_ops.sched_out = kvm_sched_out;
2224 kmem_cache_destroy(kvm_vcpu_cache);
2226 sysdev_unregister(&kvm_sysdev);
2228 sysdev_class_unregister(&kvm_sysdev_class);
2230 unregister_reboot_notifier(&kvm_reboot_notifier);
2231 unregister_cpu_notifier(&kvm_cpu_notifier);
2234 kvm_arch_hardware_unsetup();
2236 free_cpumask_var(cpus_hardware_enabled);
2238 __free_page(bad_page);
2244 EXPORT_SYMBOL_GPL(kvm_init);
2248 tracepoint_synchronize_unregister();
2250 misc_deregister(&kvm_dev);
2251 kmem_cache_destroy(kvm_vcpu_cache);
2252 sysdev_unregister(&kvm_sysdev);
2253 sysdev_class_unregister(&kvm_sysdev_class);
2254 unregister_reboot_notifier(&kvm_reboot_notifier);
2255 unregister_cpu_notifier(&kvm_cpu_notifier);
2256 on_each_cpu(hardware_disable, NULL, 1);
2257 kvm_arch_hardware_unsetup();
2259 free_cpumask_var(cpus_hardware_enabled);
2260 __free_page(bad_page);
2262 EXPORT_SYMBOL_GPL(kvm_exit);