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 void kvm_io_bus_destroy(struct kvm_io_bus *bus);
90 static bool kvm_rebooting;
92 static bool largepages_enabled = true;
94 inline int kvm_is_mmio_pfn(pfn_t pfn)
97 struct page *page = compound_head(pfn_to_page(pfn));
98 return PageReserved(page);
105 * Switches to specified vcpu, until a matching vcpu_put()
107 void vcpu_load(struct kvm_vcpu *vcpu)
111 mutex_lock(&vcpu->mutex);
113 preempt_notifier_register(&vcpu->preempt_notifier);
114 kvm_arch_vcpu_load(vcpu, cpu);
118 void vcpu_put(struct kvm_vcpu *vcpu)
121 kvm_arch_vcpu_put(vcpu);
122 preempt_notifier_unregister(&vcpu->preempt_notifier);
124 mutex_unlock(&vcpu->mutex);
127 static void ack_flush(void *_completed)
131 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
136 struct kvm_vcpu *vcpu;
138 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
140 spin_lock(&kvm->requests_lock);
141 me = smp_processor_id();
142 kvm_for_each_vcpu(i, vcpu, kvm) {
143 if (test_and_set_bit(req, &vcpu->requests))
146 if (cpus != NULL && cpu != -1 && cpu != me)
147 cpumask_set_cpu(cpu, cpus);
149 if (unlikely(cpus == NULL))
150 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
151 else if (!cpumask_empty(cpus))
152 smp_call_function_many(cpus, ack_flush, NULL, 1);
155 spin_unlock(&kvm->requests_lock);
156 free_cpumask_var(cpus);
160 void kvm_flush_remote_tlbs(struct kvm *kvm)
162 if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
163 ++kvm->stat.remote_tlb_flush;
166 void kvm_reload_remote_mmus(struct kvm *kvm)
168 make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
171 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
176 mutex_init(&vcpu->mutex);
180 init_waitqueue_head(&vcpu->wq);
182 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
187 vcpu->run = page_address(page);
189 r = kvm_arch_vcpu_init(vcpu);
195 free_page((unsigned long)vcpu->run);
199 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
201 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
203 kvm_arch_vcpu_uninit(vcpu);
204 free_page((unsigned long)vcpu->run);
206 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
208 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
209 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
211 return container_of(mn, struct kvm, mmu_notifier);
214 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
215 struct mm_struct *mm,
216 unsigned long address)
218 struct kvm *kvm = mmu_notifier_to_kvm(mn);
219 int need_tlb_flush, idx;
222 * When ->invalidate_page runs, the linux pte has been zapped
223 * already but the page is still allocated until
224 * ->invalidate_page returns. So if we increase the sequence
225 * here the kvm page fault will notice if the spte can't be
226 * established because the page is going to be freed. If
227 * instead the kvm page fault establishes the spte before
228 * ->invalidate_page runs, kvm_unmap_hva will release it
231 * The sequence increase only need to be seen at spin_unlock
232 * time, and not at spin_lock time.
234 * Increasing the sequence after the spin_unlock would be
235 * unsafe because the kvm page fault could then establish the
236 * pte after kvm_unmap_hva returned, without noticing the page
237 * is going to be freed.
239 idx = srcu_read_lock(&kvm->srcu);
240 spin_lock(&kvm->mmu_lock);
241 kvm->mmu_notifier_seq++;
242 need_tlb_flush = kvm_unmap_hva(kvm, address);
243 spin_unlock(&kvm->mmu_lock);
244 srcu_read_unlock(&kvm->srcu, idx);
246 /* we've to flush the tlb before the pages can be freed */
248 kvm_flush_remote_tlbs(kvm);
252 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
253 struct mm_struct *mm,
254 unsigned long address,
257 struct kvm *kvm = mmu_notifier_to_kvm(mn);
260 idx = srcu_read_lock(&kvm->srcu);
261 spin_lock(&kvm->mmu_lock);
262 kvm->mmu_notifier_seq++;
263 kvm_set_spte_hva(kvm, address, pte);
264 spin_unlock(&kvm->mmu_lock);
265 srcu_read_unlock(&kvm->srcu, idx);
268 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
269 struct mm_struct *mm,
273 struct kvm *kvm = mmu_notifier_to_kvm(mn);
274 int need_tlb_flush = 0, idx;
276 idx = srcu_read_lock(&kvm->srcu);
277 spin_lock(&kvm->mmu_lock);
279 * The count increase must become visible at unlock time as no
280 * spte can be established without taking the mmu_lock and
281 * count is also read inside the mmu_lock critical section.
283 kvm->mmu_notifier_count++;
284 for (; start < end; start += PAGE_SIZE)
285 need_tlb_flush |= kvm_unmap_hva(kvm, start);
286 spin_unlock(&kvm->mmu_lock);
287 srcu_read_unlock(&kvm->srcu, idx);
289 /* we've to flush the tlb before the pages can be freed */
291 kvm_flush_remote_tlbs(kvm);
294 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
295 struct mm_struct *mm,
299 struct kvm *kvm = mmu_notifier_to_kvm(mn);
301 spin_lock(&kvm->mmu_lock);
303 * This sequence increase will notify the kvm page fault that
304 * the page that is going to be mapped in the spte could have
307 kvm->mmu_notifier_seq++;
309 * The above sequence increase must be visible before the
310 * below count decrease but both values are read by the kvm
311 * page fault under mmu_lock spinlock so we don't need to add
312 * a smb_wmb() here in between the two.
314 kvm->mmu_notifier_count--;
315 spin_unlock(&kvm->mmu_lock);
317 BUG_ON(kvm->mmu_notifier_count < 0);
320 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
321 struct mm_struct *mm,
322 unsigned long address)
324 struct kvm *kvm = mmu_notifier_to_kvm(mn);
327 idx = srcu_read_lock(&kvm->srcu);
328 spin_lock(&kvm->mmu_lock);
329 young = kvm_age_hva(kvm, address);
330 spin_unlock(&kvm->mmu_lock);
331 srcu_read_unlock(&kvm->srcu, idx);
334 kvm_flush_remote_tlbs(kvm);
339 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
340 struct mm_struct *mm)
342 struct kvm *kvm = mmu_notifier_to_kvm(mn);
343 kvm_arch_flush_shadow(kvm);
346 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
347 .invalidate_page = kvm_mmu_notifier_invalidate_page,
348 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
349 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
350 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
351 .change_pte = kvm_mmu_notifier_change_pte,
352 .release = kvm_mmu_notifier_release,
355 static int kvm_init_mmu_notifier(struct kvm *kvm)
357 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
358 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
361 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
363 static int kvm_init_mmu_notifier(struct kvm *kvm)
368 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
370 static struct kvm *kvm_create_vm(void)
373 struct kvm *kvm = kvm_arch_create_vm();
378 r = hardware_enable_all();
380 goto out_err_nodisable;
382 #ifdef CONFIG_HAVE_KVM_IRQCHIP
383 INIT_HLIST_HEAD(&kvm->mask_notifier_list);
384 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
388 kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
391 if (init_srcu_struct(&kvm->srcu))
393 for (i = 0; i < KVM_NR_BUSES; i++) {
394 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
396 if (!kvm->buses[i]) {
397 cleanup_srcu_struct(&kvm->srcu);
402 r = kvm_init_mmu_notifier(kvm);
404 cleanup_srcu_struct(&kvm->srcu);
408 kvm->mm = current->mm;
409 atomic_inc(&kvm->mm->mm_count);
410 spin_lock_init(&kvm->mmu_lock);
411 spin_lock_init(&kvm->requests_lock);
412 kvm_eventfd_init(kvm);
413 mutex_init(&kvm->lock);
414 mutex_init(&kvm->irq_lock);
415 mutex_init(&kvm->slots_lock);
416 atomic_set(&kvm->users_count, 1);
417 spin_lock(&kvm_lock);
418 list_add(&kvm->vm_list, &vm_list);
419 spin_unlock(&kvm_lock);
420 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
421 kvm_coalesced_mmio_init(kvm);
427 hardware_disable_all();
429 for (i = 0; i < KVM_NR_BUSES; i++)
430 kfree(kvm->buses[i]);
431 kfree(kvm->memslots);
437 * Free any memory in @free but not in @dont.
439 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
440 struct kvm_memory_slot *dont)
444 if (!dont || free->rmap != dont->rmap)
447 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
448 vfree(free->dirty_bitmap);
451 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
452 if (!dont || free->lpage_info[i] != dont->lpage_info[i]) {
453 vfree(free->lpage_info[i]);
454 free->lpage_info[i] = NULL;
459 free->dirty_bitmap = NULL;
463 void kvm_free_physmem(struct kvm *kvm)
466 struct kvm_memslots *slots = kvm->memslots;
468 for (i = 0; i < slots->nmemslots; ++i)
469 kvm_free_physmem_slot(&slots->memslots[i], NULL);
471 kfree(kvm->memslots);
474 static void kvm_destroy_vm(struct kvm *kvm)
477 struct mm_struct *mm = kvm->mm;
479 kvm_arch_sync_events(kvm);
480 spin_lock(&kvm_lock);
481 list_del(&kvm->vm_list);
482 spin_unlock(&kvm_lock);
483 kvm_free_irq_routing(kvm);
484 for (i = 0; i < KVM_NR_BUSES; i++)
485 kvm_io_bus_destroy(kvm->buses[i]);
486 kvm_coalesced_mmio_free(kvm);
487 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
488 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
490 kvm_arch_flush_shadow(kvm);
492 kvm_arch_destroy_vm(kvm);
493 hardware_disable_all();
497 void kvm_get_kvm(struct kvm *kvm)
499 atomic_inc(&kvm->users_count);
501 EXPORT_SYMBOL_GPL(kvm_get_kvm);
503 void kvm_put_kvm(struct kvm *kvm)
505 if (atomic_dec_and_test(&kvm->users_count))
508 EXPORT_SYMBOL_GPL(kvm_put_kvm);
511 static int kvm_vm_release(struct inode *inode, struct file *filp)
513 struct kvm *kvm = filp->private_data;
515 kvm_irqfd_release(kvm);
522 * Allocate some memory and give it an address in the guest physical address
525 * Discontiguous memory is allowed, mostly for framebuffers.
527 * Must be called holding mmap_sem for write.
529 int __kvm_set_memory_region(struct kvm *kvm,
530 struct kvm_userspace_memory_region *mem,
533 int r, flush_shadow = 0;
535 unsigned long npages;
537 struct kvm_memory_slot *memslot;
538 struct kvm_memory_slot old, new;
539 struct kvm_memslots *slots, *old_memslots;
542 /* General sanity checks */
543 if (mem->memory_size & (PAGE_SIZE - 1))
545 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
547 if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
549 if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
551 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
554 memslot = &kvm->memslots->memslots[mem->slot];
555 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
556 npages = mem->memory_size >> PAGE_SHIFT;
559 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
561 new = old = *memslot;
563 new.base_gfn = base_gfn;
565 new.flags = mem->flags;
567 /* Disallow changing a memory slot's size. */
569 if (npages && old.npages && npages != old.npages)
572 /* Check for overlaps */
574 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
575 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
577 if (s == memslot || !s->npages)
579 if (!((base_gfn + npages <= s->base_gfn) ||
580 (base_gfn >= s->base_gfn + s->npages)))
584 /* Free page dirty bitmap if unneeded */
585 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
586 new.dirty_bitmap = NULL;
590 /* Allocate if a slot is being created */
592 if (npages && !new.rmap) {
593 new.rmap = vmalloc(npages * sizeof(struct page *));
598 memset(new.rmap, 0, npages * sizeof(*new.rmap));
600 new.user_alloc = user_alloc;
601 new.userspace_addr = mem->userspace_addr;
606 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
612 /* Avoid unused variable warning if no large pages */
615 if (new.lpage_info[i])
618 lpages = 1 + (base_gfn + npages - 1) /
619 KVM_PAGES_PER_HPAGE(level);
620 lpages -= base_gfn / KVM_PAGES_PER_HPAGE(level);
622 new.lpage_info[i] = vmalloc(lpages * sizeof(*new.lpage_info[i]));
624 if (!new.lpage_info[i])
627 memset(new.lpage_info[i], 0,
628 lpages * sizeof(*new.lpage_info[i]));
630 if (base_gfn % KVM_PAGES_PER_HPAGE(level))
631 new.lpage_info[i][0].write_count = 1;
632 if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE(level))
633 new.lpage_info[i][lpages - 1].write_count = 1;
634 ugfn = new.userspace_addr >> PAGE_SHIFT;
636 * If the gfn and userspace address are not aligned wrt each
637 * other, or if explicitly asked to, disable large page
638 * support for this slot
640 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
642 for (j = 0; j < lpages; ++j)
643 new.lpage_info[i][j].write_count = 1;
648 /* Allocate page dirty bitmap if needed */
649 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
650 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
652 new.dirty_bitmap = vmalloc(dirty_bytes);
653 if (!new.dirty_bitmap)
655 memset(new.dirty_bitmap, 0, dirty_bytes);
656 /* destroy any largepage mappings for dirty tracking */
660 #else /* not defined CONFIG_S390 */
661 new.user_alloc = user_alloc;
663 new.userspace_addr = mem->userspace_addr;
664 #endif /* not defined CONFIG_S390 */
668 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
671 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
672 if (mem->slot >= slots->nmemslots)
673 slots->nmemslots = mem->slot + 1;
674 slots->memslots[mem->slot].flags |= KVM_MEMSLOT_INVALID;
676 old_memslots = kvm->memslots;
677 rcu_assign_pointer(kvm->memslots, slots);
678 synchronize_srcu_expedited(&kvm->srcu);
679 /* From this point no new shadow pages pointing to a deleted
680 * memslot will be created.
682 * validation of sp->gfn happens in:
683 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
684 * - kvm_is_visible_gfn (mmu_check_roots)
686 kvm_arch_flush_shadow(kvm);
690 r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
695 /* map the pages in iommu page table */
697 r = kvm_iommu_map_pages(kvm, &new);
704 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
707 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
708 if (mem->slot >= slots->nmemslots)
709 slots->nmemslots = mem->slot + 1;
711 /* actual memory is freed via old in kvm_free_physmem_slot below */
714 new.dirty_bitmap = NULL;
715 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i)
716 new.lpage_info[i] = NULL;
719 slots->memslots[mem->slot] = new;
720 old_memslots = kvm->memslots;
721 rcu_assign_pointer(kvm->memslots, slots);
722 synchronize_srcu_expedited(&kvm->srcu);
724 kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
726 kvm_free_physmem_slot(&old, &new);
730 kvm_arch_flush_shadow(kvm);
735 kvm_free_physmem_slot(&new, &old);
740 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
742 int kvm_set_memory_region(struct kvm *kvm,
743 struct kvm_userspace_memory_region *mem,
748 mutex_lock(&kvm->slots_lock);
749 r = __kvm_set_memory_region(kvm, mem, user_alloc);
750 mutex_unlock(&kvm->slots_lock);
753 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
755 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
757 kvm_userspace_memory_region *mem,
760 if (mem->slot >= KVM_MEMORY_SLOTS)
762 return kvm_set_memory_region(kvm, mem, user_alloc);
765 int kvm_get_dirty_log(struct kvm *kvm,
766 struct kvm_dirty_log *log, int *is_dirty)
768 struct kvm_memory_slot *memslot;
771 unsigned long any = 0;
774 if (log->slot >= KVM_MEMORY_SLOTS)
777 memslot = &kvm->memslots->memslots[log->slot];
779 if (!memslot->dirty_bitmap)
782 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
784 for (i = 0; !any && i < n/sizeof(long); ++i)
785 any = memslot->dirty_bitmap[i];
788 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
799 void kvm_disable_largepages(void)
801 largepages_enabled = false;
803 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
805 int is_error_page(struct page *page)
807 return page == bad_page;
809 EXPORT_SYMBOL_GPL(is_error_page);
811 int is_error_pfn(pfn_t pfn)
813 return pfn == bad_pfn;
815 EXPORT_SYMBOL_GPL(is_error_pfn);
817 static inline unsigned long bad_hva(void)
822 int kvm_is_error_hva(unsigned long addr)
824 return addr == bad_hva();
826 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
828 struct kvm_memory_slot *gfn_to_memslot_unaliased(struct kvm *kvm, gfn_t gfn)
831 struct kvm_memslots *slots = rcu_dereference(kvm->memslots);
833 for (i = 0; i < slots->nmemslots; ++i) {
834 struct kvm_memory_slot *memslot = &slots->memslots[i];
836 if (gfn >= memslot->base_gfn
837 && gfn < memslot->base_gfn + memslot->npages)
842 EXPORT_SYMBOL_GPL(gfn_to_memslot_unaliased);
844 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
846 gfn = unalias_gfn(kvm, gfn);
847 return gfn_to_memslot_unaliased(kvm, gfn);
850 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
853 struct kvm_memslots *slots = rcu_dereference(kvm->memslots);
855 gfn = unalias_gfn_instantiation(kvm, gfn);
856 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
857 struct kvm_memory_slot *memslot = &slots->memslots[i];
859 if (memslot->flags & KVM_MEMSLOT_INVALID)
862 if (gfn >= memslot->base_gfn
863 && gfn < memslot->base_gfn + memslot->npages)
868 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
870 int memslot_id(struct kvm *kvm, gfn_t gfn)
873 struct kvm_memslots *slots = rcu_dereference(kvm->memslots);
874 struct kvm_memory_slot *memslot = NULL;
876 gfn = unalias_gfn(kvm, gfn);
877 for (i = 0; i < slots->nmemslots; ++i) {
878 memslot = &slots->memslots[i];
880 if (gfn >= memslot->base_gfn
881 && gfn < memslot->base_gfn + memslot->npages)
885 return memslot - slots->memslots;
888 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
890 struct kvm_memory_slot *slot;
892 gfn = unalias_gfn_instantiation(kvm, gfn);
893 slot = gfn_to_memslot_unaliased(kvm, gfn);
894 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
896 return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
898 EXPORT_SYMBOL_GPL(gfn_to_hva);
900 static pfn_t hva_to_pfn(struct kvm *kvm, unsigned long addr)
902 struct page *page[1];
908 npages = get_user_pages_fast(addr, 1, 1, page);
910 if (unlikely(npages != 1)) {
911 struct vm_area_struct *vma;
913 down_read(¤t->mm->mmap_sem);
914 vma = find_vma(current->mm, addr);
916 if (vma == NULL || addr < vma->vm_start ||
917 !(vma->vm_flags & VM_PFNMAP)) {
918 up_read(¤t->mm->mmap_sem);
920 return page_to_pfn(bad_page);
923 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
924 up_read(¤t->mm->mmap_sem);
925 BUG_ON(!kvm_is_mmio_pfn(pfn));
927 pfn = page_to_pfn(page[0]);
932 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
936 addr = gfn_to_hva(kvm, gfn);
937 if (kvm_is_error_hva(addr)) {
939 return page_to_pfn(bad_page);
942 return hva_to_pfn(kvm, addr);
944 EXPORT_SYMBOL_GPL(gfn_to_pfn);
946 static unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
948 return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
951 pfn_t gfn_to_pfn_memslot(struct kvm *kvm,
952 struct kvm_memory_slot *slot, gfn_t gfn)
954 unsigned long addr = gfn_to_hva_memslot(slot, gfn);
955 return hva_to_pfn(kvm, addr);
958 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
962 pfn = gfn_to_pfn(kvm, gfn);
963 if (!kvm_is_mmio_pfn(pfn))
964 return pfn_to_page(pfn);
966 WARN_ON(kvm_is_mmio_pfn(pfn));
972 EXPORT_SYMBOL_GPL(gfn_to_page);
974 void kvm_release_page_clean(struct page *page)
976 kvm_release_pfn_clean(page_to_pfn(page));
978 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
980 void kvm_release_pfn_clean(pfn_t pfn)
982 if (!kvm_is_mmio_pfn(pfn))
983 put_page(pfn_to_page(pfn));
985 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
987 void kvm_release_page_dirty(struct page *page)
989 kvm_release_pfn_dirty(page_to_pfn(page));
991 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
993 void kvm_release_pfn_dirty(pfn_t pfn)
995 kvm_set_pfn_dirty(pfn);
996 kvm_release_pfn_clean(pfn);
998 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1000 void kvm_set_page_dirty(struct page *page)
1002 kvm_set_pfn_dirty(page_to_pfn(page));
1004 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1006 void kvm_set_pfn_dirty(pfn_t pfn)
1008 if (!kvm_is_mmio_pfn(pfn)) {
1009 struct page *page = pfn_to_page(pfn);
1010 if (!PageReserved(page))
1014 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1016 void kvm_set_pfn_accessed(pfn_t pfn)
1018 if (!kvm_is_mmio_pfn(pfn))
1019 mark_page_accessed(pfn_to_page(pfn));
1021 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1023 void kvm_get_pfn(pfn_t pfn)
1025 if (!kvm_is_mmio_pfn(pfn))
1026 get_page(pfn_to_page(pfn));
1028 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1030 static int next_segment(unsigned long len, int offset)
1032 if (len > PAGE_SIZE - offset)
1033 return PAGE_SIZE - offset;
1038 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1044 addr = gfn_to_hva(kvm, gfn);
1045 if (kvm_is_error_hva(addr))
1047 r = copy_from_user(data, (void __user *)addr + offset, len);
1052 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1054 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1056 gfn_t gfn = gpa >> PAGE_SHIFT;
1058 int offset = offset_in_page(gpa);
1061 while ((seg = next_segment(len, offset)) != 0) {
1062 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1072 EXPORT_SYMBOL_GPL(kvm_read_guest);
1074 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1079 gfn_t gfn = gpa >> PAGE_SHIFT;
1080 int offset = offset_in_page(gpa);
1082 addr = gfn_to_hva(kvm, gfn);
1083 if (kvm_is_error_hva(addr))
1085 pagefault_disable();
1086 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1092 EXPORT_SYMBOL(kvm_read_guest_atomic);
1094 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1095 int offset, int len)
1100 addr = gfn_to_hva(kvm, gfn);
1101 if (kvm_is_error_hva(addr))
1103 r = copy_to_user((void __user *)addr + offset, data, len);
1106 mark_page_dirty(kvm, gfn);
1109 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1111 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1114 gfn_t gfn = gpa >> PAGE_SHIFT;
1116 int offset = offset_in_page(gpa);
1119 while ((seg = next_segment(len, offset)) != 0) {
1120 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1131 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1133 return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
1135 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1137 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1139 gfn_t gfn = gpa >> PAGE_SHIFT;
1141 int offset = offset_in_page(gpa);
1144 while ((seg = next_segment(len, offset)) != 0) {
1145 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1154 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1156 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1158 struct kvm_memory_slot *memslot;
1160 gfn = unalias_gfn(kvm, gfn);
1161 memslot = gfn_to_memslot_unaliased(kvm, gfn);
1162 if (memslot && memslot->dirty_bitmap) {
1163 unsigned long rel_gfn = gfn - memslot->base_gfn;
1166 if (!generic_test_le_bit(rel_gfn, memslot->dirty_bitmap))
1167 generic___set_le_bit(rel_gfn, memslot->dirty_bitmap);
1172 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1174 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1179 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1181 if (kvm_arch_vcpu_runnable(vcpu)) {
1182 set_bit(KVM_REQ_UNHALT, &vcpu->requests);
1185 if (kvm_cpu_has_pending_timer(vcpu))
1187 if (signal_pending(current))
1193 finish_wait(&vcpu->wq, &wait);
1196 void kvm_resched(struct kvm_vcpu *vcpu)
1198 if (!need_resched())
1202 EXPORT_SYMBOL_GPL(kvm_resched);
1204 void kvm_vcpu_on_spin(struct kvm_vcpu *vcpu)
1209 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1211 /* Sleep for 100 us, and hope lock-holder got scheduled */
1212 expires = ktime_add_ns(ktime_get(), 100000UL);
1213 schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
1215 finish_wait(&vcpu->wq, &wait);
1217 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1219 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1221 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1224 if (vmf->pgoff == 0)
1225 page = virt_to_page(vcpu->run);
1227 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1228 page = virt_to_page(vcpu->arch.pio_data);
1230 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1231 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1232 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1235 return VM_FAULT_SIGBUS;
1241 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1242 .fault = kvm_vcpu_fault,
1245 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1247 vma->vm_ops = &kvm_vcpu_vm_ops;
1251 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1253 struct kvm_vcpu *vcpu = filp->private_data;
1255 kvm_put_kvm(vcpu->kvm);
1259 static struct file_operations kvm_vcpu_fops = {
1260 .release = kvm_vcpu_release,
1261 .unlocked_ioctl = kvm_vcpu_ioctl,
1262 .compat_ioctl = kvm_vcpu_ioctl,
1263 .mmap = kvm_vcpu_mmap,
1267 * Allocates an inode for the vcpu.
1269 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1271 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1275 * Creates some virtual cpus. Good luck creating more than one.
1277 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1280 struct kvm_vcpu *vcpu, *v;
1282 vcpu = kvm_arch_vcpu_create(kvm, id);
1284 return PTR_ERR(vcpu);
1286 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1288 r = kvm_arch_vcpu_setup(vcpu);
1292 mutex_lock(&kvm->lock);
1293 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1298 kvm_for_each_vcpu(r, v, kvm)
1299 if (v->vcpu_id == id) {
1304 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1306 /* Now it's all set up, let userspace reach it */
1308 r = create_vcpu_fd(vcpu);
1314 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1316 atomic_inc(&kvm->online_vcpus);
1318 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1319 if (kvm->bsp_vcpu_id == id)
1320 kvm->bsp_vcpu = vcpu;
1322 mutex_unlock(&kvm->lock);
1326 mutex_unlock(&kvm->lock);
1327 kvm_arch_vcpu_destroy(vcpu);
1331 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1334 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1335 vcpu->sigset_active = 1;
1336 vcpu->sigset = *sigset;
1338 vcpu->sigset_active = 0;
1342 static long kvm_vcpu_ioctl(struct file *filp,
1343 unsigned int ioctl, unsigned long arg)
1345 struct kvm_vcpu *vcpu = filp->private_data;
1346 void __user *argp = (void __user *)arg;
1348 struct kvm_fpu *fpu = NULL;
1349 struct kvm_sregs *kvm_sregs = NULL;
1351 if (vcpu->kvm->mm != current->mm)
1358 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1360 case KVM_GET_REGS: {
1361 struct kvm_regs *kvm_regs;
1364 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1367 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1371 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1378 case KVM_SET_REGS: {
1379 struct kvm_regs *kvm_regs;
1382 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1386 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1388 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1396 case KVM_GET_SREGS: {
1397 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1401 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1405 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1410 case KVM_SET_SREGS: {
1411 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1416 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1418 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1424 case KVM_GET_MP_STATE: {
1425 struct kvm_mp_state mp_state;
1427 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1431 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1436 case KVM_SET_MP_STATE: {
1437 struct kvm_mp_state mp_state;
1440 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1442 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1448 case KVM_TRANSLATE: {
1449 struct kvm_translation tr;
1452 if (copy_from_user(&tr, argp, sizeof tr))
1454 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1458 if (copy_to_user(argp, &tr, sizeof tr))
1463 case KVM_SET_GUEST_DEBUG: {
1464 struct kvm_guest_debug dbg;
1467 if (copy_from_user(&dbg, argp, sizeof dbg))
1469 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1475 case KVM_SET_SIGNAL_MASK: {
1476 struct kvm_signal_mask __user *sigmask_arg = argp;
1477 struct kvm_signal_mask kvm_sigmask;
1478 sigset_t sigset, *p;
1483 if (copy_from_user(&kvm_sigmask, argp,
1484 sizeof kvm_sigmask))
1487 if (kvm_sigmask.len != sizeof sigset)
1490 if (copy_from_user(&sigset, sigmask_arg->sigset,
1495 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1499 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1503 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1507 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1513 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1518 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1520 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1527 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1535 static long kvm_vm_ioctl(struct file *filp,
1536 unsigned int ioctl, unsigned long arg)
1538 struct kvm *kvm = filp->private_data;
1539 void __user *argp = (void __user *)arg;
1542 if (kvm->mm != current->mm)
1545 case KVM_CREATE_VCPU:
1546 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1550 case KVM_SET_USER_MEMORY_REGION: {
1551 struct kvm_userspace_memory_region kvm_userspace_mem;
1554 if (copy_from_user(&kvm_userspace_mem, argp,
1555 sizeof kvm_userspace_mem))
1558 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1563 case KVM_GET_DIRTY_LOG: {
1564 struct kvm_dirty_log log;
1567 if (copy_from_user(&log, argp, sizeof log))
1569 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1574 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1575 case KVM_REGISTER_COALESCED_MMIO: {
1576 struct kvm_coalesced_mmio_zone zone;
1578 if (copy_from_user(&zone, argp, sizeof zone))
1581 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1587 case KVM_UNREGISTER_COALESCED_MMIO: {
1588 struct kvm_coalesced_mmio_zone zone;
1590 if (copy_from_user(&zone, argp, sizeof zone))
1593 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1601 struct kvm_irqfd data;
1604 if (copy_from_user(&data, argp, sizeof data))
1606 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
1609 case KVM_IOEVENTFD: {
1610 struct kvm_ioeventfd data;
1613 if (copy_from_user(&data, argp, sizeof data))
1615 r = kvm_ioeventfd(kvm, &data);
1618 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1619 case KVM_SET_BOOT_CPU_ID:
1621 mutex_lock(&kvm->lock);
1622 if (atomic_read(&kvm->online_vcpus) != 0)
1625 kvm->bsp_vcpu_id = arg;
1626 mutex_unlock(&kvm->lock);
1630 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1632 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
1638 #ifdef CONFIG_COMPAT
1639 struct compat_kvm_dirty_log {
1643 compat_uptr_t dirty_bitmap; /* one bit per page */
1648 static long kvm_vm_compat_ioctl(struct file *filp,
1649 unsigned int ioctl, unsigned long arg)
1651 struct kvm *kvm = filp->private_data;
1654 if (kvm->mm != current->mm)
1657 case KVM_GET_DIRTY_LOG: {
1658 struct compat_kvm_dirty_log compat_log;
1659 struct kvm_dirty_log log;
1662 if (copy_from_user(&compat_log, (void __user *)arg,
1663 sizeof(compat_log)))
1665 log.slot = compat_log.slot;
1666 log.padding1 = compat_log.padding1;
1667 log.padding2 = compat_log.padding2;
1668 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
1670 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1676 r = kvm_vm_ioctl(filp, ioctl, arg);
1684 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1686 struct page *page[1];
1689 gfn_t gfn = vmf->pgoff;
1690 struct kvm *kvm = vma->vm_file->private_data;
1692 addr = gfn_to_hva(kvm, gfn);
1693 if (kvm_is_error_hva(addr))
1694 return VM_FAULT_SIGBUS;
1696 npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
1698 if (unlikely(npages != 1))
1699 return VM_FAULT_SIGBUS;
1701 vmf->page = page[0];
1705 static const struct vm_operations_struct kvm_vm_vm_ops = {
1706 .fault = kvm_vm_fault,
1709 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1711 vma->vm_ops = &kvm_vm_vm_ops;
1715 static struct file_operations kvm_vm_fops = {
1716 .release = kvm_vm_release,
1717 .unlocked_ioctl = kvm_vm_ioctl,
1718 #ifdef CONFIG_COMPAT
1719 .compat_ioctl = kvm_vm_compat_ioctl,
1721 .mmap = kvm_vm_mmap,
1724 static int kvm_dev_ioctl_create_vm(void)
1729 kvm = kvm_create_vm();
1731 return PTR_ERR(kvm);
1732 fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
1739 static long kvm_dev_ioctl_check_extension_generic(long arg)
1742 case KVM_CAP_USER_MEMORY:
1743 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
1744 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
1745 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1746 case KVM_CAP_SET_BOOT_CPU_ID:
1748 case KVM_CAP_INTERNAL_ERROR_DATA:
1750 #ifdef CONFIG_HAVE_KVM_IRQCHIP
1751 case KVM_CAP_IRQ_ROUTING:
1752 return KVM_MAX_IRQ_ROUTES;
1757 return kvm_dev_ioctl_check_extension(arg);
1760 static long kvm_dev_ioctl(struct file *filp,
1761 unsigned int ioctl, unsigned long arg)
1766 case KVM_GET_API_VERSION:
1770 r = KVM_API_VERSION;
1776 r = kvm_dev_ioctl_create_vm();
1778 case KVM_CHECK_EXTENSION:
1779 r = kvm_dev_ioctl_check_extension_generic(arg);
1781 case KVM_GET_VCPU_MMAP_SIZE:
1785 r = PAGE_SIZE; /* struct kvm_run */
1787 r += PAGE_SIZE; /* pio data page */
1789 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1790 r += PAGE_SIZE; /* coalesced mmio ring page */
1793 case KVM_TRACE_ENABLE:
1794 case KVM_TRACE_PAUSE:
1795 case KVM_TRACE_DISABLE:
1799 return kvm_arch_dev_ioctl(filp, ioctl, arg);
1805 static struct file_operations kvm_chardev_ops = {
1806 .unlocked_ioctl = kvm_dev_ioctl,
1807 .compat_ioctl = kvm_dev_ioctl,
1810 static struct miscdevice kvm_dev = {
1816 static void hardware_enable(void *junk)
1818 int cpu = raw_smp_processor_id();
1821 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
1824 cpumask_set_cpu(cpu, cpus_hardware_enabled);
1826 r = kvm_arch_hardware_enable(NULL);
1829 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
1830 atomic_inc(&hardware_enable_failed);
1831 printk(KERN_INFO "kvm: enabling virtualization on "
1832 "CPU%d failed\n", cpu);
1836 static void hardware_disable(void *junk)
1838 int cpu = raw_smp_processor_id();
1840 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
1842 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
1843 kvm_arch_hardware_disable(NULL);
1846 static void hardware_disable_all_nolock(void)
1848 BUG_ON(!kvm_usage_count);
1851 if (!kvm_usage_count)
1852 on_each_cpu(hardware_disable, NULL, 1);
1855 static void hardware_disable_all(void)
1857 spin_lock(&kvm_lock);
1858 hardware_disable_all_nolock();
1859 spin_unlock(&kvm_lock);
1862 static int hardware_enable_all(void)
1866 spin_lock(&kvm_lock);
1869 if (kvm_usage_count == 1) {
1870 atomic_set(&hardware_enable_failed, 0);
1871 on_each_cpu(hardware_enable, NULL, 1);
1873 if (atomic_read(&hardware_enable_failed)) {
1874 hardware_disable_all_nolock();
1879 spin_unlock(&kvm_lock);
1884 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
1889 if (!kvm_usage_count)
1892 val &= ~CPU_TASKS_FROZEN;
1895 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1897 hardware_disable(NULL);
1899 case CPU_UP_CANCELED:
1900 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1902 smp_call_function_single(cpu, hardware_disable, NULL, 1);
1905 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
1907 smp_call_function_single(cpu, hardware_enable, NULL, 1);
1914 asmlinkage void kvm_handle_fault_on_reboot(void)
1917 /* spin while reset goes on */
1920 /* Fault while not rebooting. We want the trace. */
1923 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
1925 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
1929 * Some (well, at least mine) BIOSes hang on reboot if
1932 * And Intel TXT required VMX off for all cpu when system shutdown.
1934 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
1935 kvm_rebooting = true;
1936 on_each_cpu(hardware_disable, NULL, 1);
1940 static struct notifier_block kvm_reboot_notifier = {
1941 .notifier_call = kvm_reboot,
1945 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
1949 for (i = 0; i < bus->dev_count; i++) {
1950 struct kvm_io_device *pos = bus->devs[i];
1952 kvm_iodevice_destructor(pos);
1957 /* kvm_io_bus_write - called under kvm->slots_lock */
1958 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
1959 int len, const void *val)
1962 struct kvm_io_bus *bus = rcu_dereference(kvm->buses[bus_idx]);
1963 for (i = 0; i < bus->dev_count; i++)
1964 if (!kvm_iodevice_write(bus->devs[i], addr, len, val))
1969 /* kvm_io_bus_read - called under kvm->slots_lock */
1970 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
1974 struct kvm_io_bus *bus = rcu_dereference(kvm->buses[bus_idx]);
1976 for (i = 0; i < bus->dev_count; i++)
1977 if (!kvm_iodevice_read(bus->devs[i], addr, len, val))
1982 /* Caller must hold slots_lock. */
1983 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx,
1984 struct kvm_io_device *dev)
1986 struct kvm_io_bus *new_bus, *bus;
1988 bus = kvm->buses[bus_idx];
1989 if (bus->dev_count > NR_IOBUS_DEVS-1)
1992 new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
1995 memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
1996 new_bus->devs[new_bus->dev_count++] = dev;
1997 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
1998 synchronize_srcu_expedited(&kvm->srcu);
2004 /* Caller must hold slots_lock. */
2005 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2006 struct kvm_io_device *dev)
2009 struct kvm_io_bus *new_bus, *bus;
2011 new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2015 bus = kvm->buses[bus_idx];
2016 memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2019 for (i = 0; i < new_bus->dev_count; i++)
2020 if (new_bus->devs[i] == dev) {
2022 new_bus->devs[i] = new_bus->devs[--new_bus->dev_count];
2031 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2032 synchronize_srcu_expedited(&kvm->srcu);
2037 static struct notifier_block kvm_cpu_notifier = {
2038 .notifier_call = kvm_cpu_hotplug,
2039 .priority = 20, /* must be > scheduler priority */
2042 static int vm_stat_get(void *_offset, u64 *val)
2044 unsigned offset = (long)_offset;
2048 spin_lock(&kvm_lock);
2049 list_for_each_entry(kvm, &vm_list, vm_list)
2050 *val += *(u32 *)((void *)kvm + offset);
2051 spin_unlock(&kvm_lock);
2055 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2057 static int vcpu_stat_get(void *_offset, u64 *val)
2059 unsigned offset = (long)_offset;
2061 struct kvm_vcpu *vcpu;
2065 spin_lock(&kvm_lock);
2066 list_for_each_entry(kvm, &vm_list, vm_list)
2067 kvm_for_each_vcpu(i, vcpu, kvm)
2068 *val += *(u32 *)((void *)vcpu + offset);
2070 spin_unlock(&kvm_lock);
2074 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2076 static const struct file_operations *stat_fops[] = {
2077 [KVM_STAT_VCPU] = &vcpu_stat_fops,
2078 [KVM_STAT_VM] = &vm_stat_fops,
2081 static void kvm_init_debug(void)
2083 struct kvm_stats_debugfs_item *p;
2085 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2086 for (p = debugfs_entries; p->name; ++p)
2087 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2088 (void *)(long)p->offset,
2089 stat_fops[p->kind]);
2092 static void kvm_exit_debug(void)
2094 struct kvm_stats_debugfs_item *p;
2096 for (p = debugfs_entries; p->name; ++p)
2097 debugfs_remove(p->dentry);
2098 debugfs_remove(kvm_debugfs_dir);
2101 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2103 if (kvm_usage_count)
2104 hardware_disable(NULL);
2108 static int kvm_resume(struct sys_device *dev)
2110 if (kvm_usage_count)
2111 hardware_enable(NULL);
2115 static struct sysdev_class kvm_sysdev_class = {
2117 .suspend = kvm_suspend,
2118 .resume = kvm_resume,
2121 static struct sys_device kvm_sysdev = {
2123 .cls = &kvm_sysdev_class,
2126 struct page *bad_page;
2130 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2132 return container_of(pn, struct kvm_vcpu, preempt_notifier);
2135 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2137 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2139 kvm_arch_vcpu_load(vcpu, cpu);
2142 static void kvm_sched_out(struct preempt_notifier *pn,
2143 struct task_struct *next)
2145 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2147 kvm_arch_vcpu_put(vcpu);
2150 int kvm_init(void *opaque, unsigned int vcpu_size,
2151 struct module *module)
2156 r = kvm_arch_init(opaque);
2160 bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2162 if (bad_page == NULL) {
2167 bad_pfn = page_to_pfn(bad_page);
2169 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2174 r = kvm_arch_hardware_setup();
2178 for_each_online_cpu(cpu) {
2179 smp_call_function_single(cpu,
2180 kvm_arch_check_processor_compat,
2186 r = register_cpu_notifier(&kvm_cpu_notifier);
2189 register_reboot_notifier(&kvm_reboot_notifier);
2191 r = sysdev_class_register(&kvm_sysdev_class);
2195 r = sysdev_register(&kvm_sysdev);
2199 /* A kmem cache lets us meet the alignment requirements of fx_save. */
2200 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
2201 __alignof__(struct kvm_vcpu),
2203 if (!kvm_vcpu_cache) {
2208 kvm_chardev_ops.owner = module;
2209 kvm_vm_fops.owner = module;
2210 kvm_vcpu_fops.owner = module;
2212 r = misc_register(&kvm_dev);
2214 printk(KERN_ERR "kvm: misc device register failed\n");
2218 kvm_preempt_ops.sched_in = kvm_sched_in;
2219 kvm_preempt_ops.sched_out = kvm_sched_out;
2226 kmem_cache_destroy(kvm_vcpu_cache);
2228 sysdev_unregister(&kvm_sysdev);
2230 sysdev_class_unregister(&kvm_sysdev_class);
2232 unregister_reboot_notifier(&kvm_reboot_notifier);
2233 unregister_cpu_notifier(&kvm_cpu_notifier);
2236 kvm_arch_hardware_unsetup();
2238 free_cpumask_var(cpus_hardware_enabled);
2240 __free_page(bad_page);
2246 EXPORT_SYMBOL_GPL(kvm_init);
2250 tracepoint_synchronize_unregister();
2252 misc_deregister(&kvm_dev);
2253 kmem_cache_destroy(kvm_vcpu_cache);
2254 sysdev_unregister(&kvm_sysdev);
2255 sysdev_class_unregister(&kvm_sysdev_class);
2256 unregister_reboot_notifier(&kvm_reboot_notifier);
2257 unregister_cpu_notifier(&kvm_cpu_notifier);
2258 on_each_cpu(hardware_disable, NULL, 1);
2259 kvm_arch_hardware_unsetup();
2261 free_cpumask_var(cpus_hardware_enabled);
2262 __free_page(bad_page);
2264 EXPORT_SYMBOL_GPL(kvm_exit);