KVM: reorganize hva_to_pfn
[linux-3.10.git] / virt / kvm / kvm_main.c
1 /*
2  * Kernel-based Virtual Machine driver for Linux
3  *
4  * This module enables machines with Intel VT-x extensions to run virtual
5  * machines without emulation or binary translation.
6  *
7  * Copyright (C) 2006 Qumranet, Inc.
8  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
9  *
10  * Authors:
11  *   Avi Kivity   <avi@qumranet.com>
12  *   Yaniv Kamay  <yaniv@qumranet.com>
13  *
14  * This work is licensed under the terms of the GNU GPL, version 2.  See
15  * the COPYING file in the top-level directory.
16  *
17  */
18
19 #include "iodev.h"
20
21 #include <linux/kvm_host.h>
22 #include <linux/kvm.h>
23 #include <linux/module.h>
24 #include <linux/errno.h>
25 #include <linux/percpu.h>
26 #include <linux/mm.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/syscore_ops.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>
48 #include <linux/hugetlb.h>
49 #include <linux/slab.h>
50 #include <linux/sort.h>
51 #include <linux/bsearch.h>
52
53 #include <asm/processor.h>
54 #include <asm/io.h>
55 #include <asm/uaccess.h>
56 #include <asm/pgtable.h>
57
58 #include "coalesced_mmio.h"
59 #include "async_pf.h"
60
61 #define CREATE_TRACE_POINTS
62 #include <trace/events/kvm.h>
63
64 MODULE_AUTHOR("Qumranet");
65 MODULE_LICENSE("GPL");
66
67 /*
68  * Ordering of locks:
69  *
70  *              kvm->lock --> kvm->slots_lock --> kvm->irq_lock
71  */
72
73 DEFINE_RAW_SPINLOCK(kvm_lock);
74 LIST_HEAD(vm_list);
75
76 static cpumask_var_t cpus_hardware_enabled;
77 static int kvm_usage_count = 0;
78 static atomic_t hardware_enable_failed;
79
80 struct kmem_cache *kvm_vcpu_cache;
81 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
82
83 static __read_mostly struct preempt_ops kvm_preempt_ops;
84
85 struct dentry *kvm_debugfs_dir;
86
87 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
88                            unsigned long arg);
89 #ifdef CONFIG_COMPAT
90 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
91                                   unsigned long arg);
92 #endif
93 static int hardware_enable_all(void);
94 static void hardware_disable_all(void);
95
96 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
97
98 bool kvm_rebooting;
99 EXPORT_SYMBOL_GPL(kvm_rebooting);
100
101 static bool largepages_enabled = true;
102
103 bool kvm_is_mmio_pfn(pfn_t pfn)
104 {
105         if (pfn_valid(pfn)) {
106                 int reserved;
107                 struct page *tail = pfn_to_page(pfn);
108                 struct page *head = compound_trans_head(tail);
109                 reserved = PageReserved(head);
110                 if (head != tail) {
111                         /*
112                          * "head" is not a dangling pointer
113                          * (compound_trans_head takes care of that)
114                          * but the hugepage may have been splitted
115                          * from under us (and we may not hold a
116                          * reference count on the head page so it can
117                          * be reused before we run PageReferenced), so
118                          * we've to check PageTail before returning
119                          * what we just read.
120                          */
121                         smp_rmb();
122                         if (PageTail(tail))
123                                 return reserved;
124                 }
125                 return PageReserved(tail);
126         }
127
128         return true;
129 }
130
131 /*
132  * Switches to specified vcpu, until a matching vcpu_put()
133  */
134 void vcpu_load(struct kvm_vcpu *vcpu)
135 {
136         int cpu;
137
138         mutex_lock(&vcpu->mutex);
139         if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
140                 /* The thread running this VCPU changed. */
141                 struct pid *oldpid = vcpu->pid;
142                 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
143                 rcu_assign_pointer(vcpu->pid, newpid);
144                 synchronize_rcu();
145                 put_pid(oldpid);
146         }
147         cpu = get_cpu();
148         preempt_notifier_register(&vcpu->preempt_notifier);
149         kvm_arch_vcpu_load(vcpu, cpu);
150         put_cpu();
151 }
152
153 void vcpu_put(struct kvm_vcpu *vcpu)
154 {
155         preempt_disable();
156         kvm_arch_vcpu_put(vcpu);
157         preempt_notifier_unregister(&vcpu->preempt_notifier);
158         preempt_enable();
159         mutex_unlock(&vcpu->mutex);
160 }
161
162 static void ack_flush(void *_completed)
163 {
164 }
165
166 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
167 {
168         int i, cpu, me;
169         cpumask_var_t cpus;
170         bool called = true;
171         struct kvm_vcpu *vcpu;
172
173         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
174
175         me = get_cpu();
176         kvm_for_each_vcpu(i, vcpu, kvm) {
177                 kvm_make_request(req, vcpu);
178                 cpu = vcpu->cpu;
179
180                 /* Set ->requests bit before we read ->mode */
181                 smp_mb();
182
183                 if (cpus != NULL && cpu != -1 && cpu != me &&
184                       kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
185                         cpumask_set_cpu(cpu, cpus);
186         }
187         if (unlikely(cpus == NULL))
188                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
189         else if (!cpumask_empty(cpus))
190                 smp_call_function_many(cpus, ack_flush, NULL, 1);
191         else
192                 called = false;
193         put_cpu();
194         free_cpumask_var(cpus);
195         return called;
196 }
197
198 void kvm_flush_remote_tlbs(struct kvm *kvm)
199 {
200         long dirty_count = kvm->tlbs_dirty;
201
202         smp_mb();
203         if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
204                 ++kvm->stat.remote_tlb_flush;
205         cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
206 }
207
208 void kvm_reload_remote_mmus(struct kvm *kvm)
209 {
210         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
211 }
212
213 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
214 {
215         struct page *page;
216         int r;
217
218         mutex_init(&vcpu->mutex);
219         vcpu->cpu = -1;
220         vcpu->kvm = kvm;
221         vcpu->vcpu_id = id;
222         vcpu->pid = NULL;
223         init_waitqueue_head(&vcpu->wq);
224         kvm_async_pf_vcpu_init(vcpu);
225
226         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
227         if (!page) {
228                 r = -ENOMEM;
229                 goto fail;
230         }
231         vcpu->run = page_address(page);
232
233         kvm_vcpu_set_in_spin_loop(vcpu, false);
234         kvm_vcpu_set_dy_eligible(vcpu, false);
235
236         r = kvm_arch_vcpu_init(vcpu);
237         if (r < 0)
238                 goto fail_free_run;
239         return 0;
240
241 fail_free_run:
242         free_page((unsigned long)vcpu->run);
243 fail:
244         return r;
245 }
246 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
247
248 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
249 {
250         put_pid(vcpu->pid);
251         kvm_arch_vcpu_uninit(vcpu);
252         free_page((unsigned long)vcpu->run);
253 }
254 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
255
256 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
257 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
258 {
259         return container_of(mn, struct kvm, mmu_notifier);
260 }
261
262 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
263                                              struct mm_struct *mm,
264                                              unsigned long address)
265 {
266         struct kvm *kvm = mmu_notifier_to_kvm(mn);
267         int need_tlb_flush, idx;
268
269         /*
270          * When ->invalidate_page runs, the linux pte has been zapped
271          * already but the page is still allocated until
272          * ->invalidate_page returns. So if we increase the sequence
273          * here the kvm page fault will notice if the spte can't be
274          * established because the page is going to be freed. If
275          * instead the kvm page fault establishes the spte before
276          * ->invalidate_page runs, kvm_unmap_hva will release it
277          * before returning.
278          *
279          * The sequence increase only need to be seen at spin_unlock
280          * time, and not at spin_lock time.
281          *
282          * Increasing the sequence after the spin_unlock would be
283          * unsafe because the kvm page fault could then establish the
284          * pte after kvm_unmap_hva returned, without noticing the page
285          * is going to be freed.
286          */
287         idx = srcu_read_lock(&kvm->srcu);
288         spin_lock(&kvm->mmu_lock);
289
290         kvm->mmu_notifier_seq++;
291         need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
292         /* we've to flush the tlb before the pages can be freed */
293         if (need_tlb_flush)
294                 kvm_flush_remote_tlbs(kvm);
295
296         spin_unlock(&kvm->mmu_lock);
297         srcu_read_unlock(&kvm->srcu, idx);
298 }
299
300 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
301                                         struct mm_struct *mm,
302                                         unsigned long address,
303                                         pte_t pte)
304 {
305         struct kvm *kvm = mmu_notifier_to_kvm(mn);
306         int idx;
307
308         idx = srcu_read_lock(&kvm->srcu);
309         spin_lock(&kvm->mmu_lock);
310         kvm->mmu_notifier_seq++;
311         kvm_set_spte_hva(kvm, address, pte);
312         spin_unlock(&kvm->mmu_lock);
313         srcu_read_unlock(&kvm->srcu, idx);
314 }
315
316 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
317                                                     struct mm_struct *mm,
318                                                     unsigned long start,
319                                                     unsigned long end)
320 {
321         struct kvm *kvm = mmu_notifier_to_kvm(mn);
322         int need_tlb_flush = 0, idx;
323
324         idx = srcu_read_lock(&kvm->srcu);
325         spin_lock(&kvm->mmu_lock);
326         /*
327          * The count increase must become visible at unlock time as no
328          * spte can be established without taking the mmu_lock and
329          * count is also read inside the mmu_lock critical section.
330          */
331         kvm->mmu_notifier_count++;
332         need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
333         need_tlb_flush |= kvm->tlbs_dirty;
334         /* we've to flush the tlb before the pages can be freed */
335         if (need_tlb_flush)
336                 kvm_flush_remote_tlbs(kvm);
337
338         spin_unlock(&kvm->mmu_lock);
339         srcu_read_unlock(&kvm->srcu, idx);
340 }
341
342 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
343                                                   struct mm_struct *mm,
344                                                   unsigned long start,
345                                                   unsigned long end)
346 {
347         struct kvm *kvm = mmu_notifier_to_kvm(mn);
348
349         spin_lock(&kvm->mmu_lock);
350         /*
351          * This sequence increase will notify the kvm page fault that
352          * the page that is going to be mapped in the spte could have
353          * been freed.
354          */
355         kvm->mmu_notifier_seq++;
356         smp_wmb();
357         /*
358          * The above sequence increase must be visible before the
359          * below count decrease, which is ensured by the smp_wmb above
360          * in conjunction with the smp_rmb in mmu_notifier_retry().
361          */
362         kvm->mmu_notifier_count--;
363         spin_unlock(&kvm->mmu_lock);
364
365         BUG_ON(kvm->mmu_notifier_count < 0);
366 }
367
368 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
369                                               struct mm_struct *mm,
370                                               unsigned long address)
371 {
372         struct kvm *kvm = mmu_notifier_to_kvm(mn);
373         int young, idx;
374
375         idx = srcu_read_lock(&kvm->srcu);
376         spin_lock(&kvm->mmu_lock);
377
378         young = kvm_age_hva(kvm, address);
379         if (young)
380                 kvm_flush_remote_tlbs(kvm);
381
382         spin_unlock(&kvm->mmu_lock);
383         srcu_read_unlock(&kvm->srcu, idx);
384
385         return young;
386 }
387
388 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
389                                        struct mm_struct *mm,
390                                        unsigned long address)
391 {
392         struct kvm *kvm = mmu_notifier_to_kvm(mn);
393         int young, idx;
394
395         idx = srcu_read_lock(&kvm->srcu);
396         spin_lock(&kvm->mmu_lock);
397         young = kvm_test_age_hva(kvm, address);
398         spin_unlock(&kvm->mmu_lock);
399         srcu_read_unlock(&kvm->srcu, idx);
400
401         return young;
402 }
403
404 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
405                                      struct mm_struct *mm)
406 {
407         struct kvm *kvm = mmu_notifier_to_kvm(mn);
408         int idx;
409
410         idx = srcu_read_lock(&kvm->srcu);
411         kvm_arch_flush_shadow(kvm);
412         srcu_read_unlock(&kvm->srcu, idx);
413 }
414
415 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
416         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
417         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
418         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
419         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
420         .test_young             = kvm_mmu_notifier_test_young,
421         .change_pte             = kvm_mmu_notifier_change_pte,
422         .release                = kvm_mmu_notifier_release,
423 };
424
425 static int kvm_init_mmu_notifier(struct kvm *kvm)
426 {
427         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
428         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
429 }
430
431 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
432
433 static int kvm_init_mmu_notifier(struct kvm *kvm)
434 {
435         return 0;
436 }
437
438 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
439
440 static void kvm_init_memslots_id(struct kvm *kvm)
441 {
442         int i;
443         struct kvm_memslots *slots = kvm->memslots;
444
445         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
446                 slots->id_to_index[i] = slots->memslots[i].id = i;
447 }
448
449 static struct kvm *kvm_create_vm(unsigned long type)
450 {
451         int r, i;
452         struct kvm *kvm = kvm_arch_alloc_vm();
453
454         if (!kvm)
455                 return ERR_PTR(-ENOMEM);
456
457         r = kvm_arch_init_vm(kvm, type);
458         if (r)
459                 goto out_err_nodisable;
460
461         r = hardware_enable_all();
462         if (r)
463                 goto out_err_nodisable;
464
465 #ifdef CONFIG_HAVE_KVM_IRQCHIP
466         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
467         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
468 #endif
469
470         r = -ENOMEM;
471         kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
472         if (!kvm->memslots)
473                 goto out_err_nosrcu;
474         kvm_init_memslots_id(kvm);
475         if (init_srcu_struct(&kvm->srcu))
476                 goto out_err_nosrcu;
477         for (i = 0; i < KVM_NR_BUSES; i++) {
478                 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
479                                         GFP_KERNEL);
480                 if (!kvm->buses[i])
481                         goto out_err;
482         }
483
484         spin_lock_init(&kvm->mmu_lock);
485         kvm->mm = current->mm;
486         atomic_inc(&kvm->mm->mm_count);
487         kvm_eventfd_init(kvm);
488         mutex_init(&kvm->lock);
489         mutex_init(&kvm->irq_lock);
490         mutex_init(&kvm->slots_lock);
491         atomic_set(&kvm->users_count, 1);
492
493         r = kvm_init_mmu_notifier(kvm);
494         if (r)
495                 goto out_err;
496
497         raw_spin_lock(&kvm_lock);
498         list_add(&kvm->vm_list, &vm_list);
499         raw_spin_unlock(&kvm_lock);
500
501         return kvm;
502
503 out_err:
504         cleanup_srcu_struct(&kvm->srcu);
505 out_err_nosrcu:
506         hardware_disable_all();
507 out_err_nodisable:
508         for (i = 0; i < KVM_NR_BUSES; i++)
509                 kfree(kvm->buses[i]);
510         kfree(kvm->memslots);
511         kvm_arch_free_vm(kvm);
512         return ERR_PTR(r);
513 }
514
515 /*
516  * Avoid using vmalloc for a small buffer.
517  * Should not be used when the size is statically known.
518  */
519 void *kvm_kvzalloc(unsigned long size)
520 {
521         if (size > PAGE_SIZE)
522                 return vzalloc(size);
523         else
524                 return kzalloc(size, GFP_KERNEL);
525 }
526
527 void kvm_kvfree(const void *addr)
528 {
529         if (is_vmalloc_addr(addr))
530                 vfree(addr);
531         else
532                 kfree(addr);
533 }
534
535 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
536 {
537         if (!memslot->dirty_bitmap)
538                 return;
539
540         kvm_kvfree(memslot->dirty_bitmap);
541         memslot->dirty_bitmap = NULL;
542 }
543
544 /*
545  * Free any memory in @free but not in @dont.
546  */
547 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
548                                   struct kvm_memory_slot *dont)
549 {
550         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
551                 kvm_destroy_dirty_bitmap(free);
552
553         kvm_arch_free_memslot(free, dont);
554
555         free->npages = 0;
556 }
557
558 void kvm_free_physmem(struct kvm *kvm)
559 {
560         struct kvm_memslots *slots = kvm->memslots;
561         struct kvm_memory_slot *memslot;
562
563         kvm_for_each_memslot(memslot, slots)
564                 kvm_free_physmem_slot(memslot, NULL);
565
566         kfree(kvm->memslots);
567 }
568
569 static void kvm_destroy_vm(struct kvm *kvm)
570 {
571         int i;
572         struct mm_struct *mm = kvm->mm;
573
574         kvm_arch_sync_events(kvm);
575         raw_spin_lock(&kvm_lock);
576         list_del(&kvm->vm_list);
577         raw_spin_unlock(&kvm_lock);
578         kvm_free_irq_routing(kvm);
579         for (i = 0; i < KVM_NR_BUSES; i++)
580                 kvm_io_bus_destroy(kvm->buses[i]);
581         kvm_coalesced_mmio_free(kvm);
582 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
583         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
584 #else
585         kvm_arch_flush_shadow(kvm);
586 #endif
587         kvm_arch_destroy_vm(kvm);
588         kvm_free_physmem(kvm);
589         cleanup_srcu_struct(&kvm->srcu);
590         kvm_arch_free_vm(kvm);
591         hardware_disable_all();
592         mmdrop(mm);
593 }
594
595 void kvm_get_kvm(struct kvm *kvm)
596 {
597         atomic_inc(&kvm->users_count);
598 }
599 EXPORT_SYMBOL_GPL(kvm_get_kvm);
600
601 void kvm_put_kvm(struct kvm *kvm)
602 {
603         if (atomic_dec_and_test(&kvm->users_count))
604                 kvm_destroy_vm(kvm);
605 }
606 EXPORT_SYMBOL_GPL(kvm_put_kvm);
607
608
609 static int kvm_vm_release(struct inode *inode, struct file *filp)
610 {
611         struct kvm *kvm = filp->private_data;
612
613         kvm_irqfd_release(kvm);
614
615         kvm_put_kvm(kvm);
616         return 0;
617 }
618
619 /*
620  * Allocation size is twice as large as the actual dirty bitmap size.
621  * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
622  */
623 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
624 {
625 #ifndef CONFIG_S390
626         unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
627
628         memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
629         if (!memslot->dirty_bitmap)
630                 return -ENOMEM;
631
632 #endif /* !CONFIG_S390 */
633         return 0;
634 }
635
636 static int cmp_memslot(const void *slot1, const void *slot2)
637 {
638         struct kvm_memory_slot *s1, *s2;
639
640         s1 = (struct kvm_memory_slot *)slot1;
641         s2 = (struct kvm_memory_slot *)slot2;
642
643         if (s1->npages < s2->npages)
644                 return 1;
645         if (s1->npages > s2->npages)
646                 return -1;
647
648         return 0;
649 }
650
651 /*
652  * Sort the memslots base on its size, so the larger slots
653  * will get better fit.
654  */
655 static void sort_memslots(struct kvm_memslots *slots)
656 {
657         int i;
658
659         sort(slots->memslots, KVM_MEM_SLOTS_NUM,
660               sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
661
662         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
663                 slots->id_to_index[slots->memslots[i].id] = i;
664 }
665
666 void update_memslots(struct kvm_memslots *slots, struct kvm_memory_slot *new)
667 {
668         if (new) {
669                 int id = new->id;
670                 struct kvm_memory_slot *old = id_to_memslot(slots, id);
671                 unsigned long npages = old->npages;
672
673                 *old = *new;
674                 if (new->npages != npages)
675                         sort_memslots(slots);
676         }
677
678         slots->generation++;
679 }
680
681 static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
682 {
683         if (mem->flags & ~KVM_MEM_LOG_DIRTY_PAGES)
684                 return -EINVAL;
685
686         return 0;
687 }
688
689 /*
690  * Allocate some memory and give it an address in the guest physical address
691  * space.
692  *
693  * Discontiguous memory is allowed, mostly for framebuffers.
694  *
695  * Must be called holding mmap_sem for write.
696  */
697 int __kvm_set_memory_region(struct kvm *kvm,
698                             struct kvm_userspace_memory_region *mem,
699                             int user_alloc)
700 {
701         int r;
702         gfn_t base_gfn;
703         unsigned long npages;
704         unsigned long i;
705         struct kvm_memory_slot *memslot;
706         struct kvm_memory_slot old, new;
707         struct kvm_memslots *slots, *old_memslots;
708
709         r = check_memory_region_flags(mem);
710         if (r)
711                 goto out;
712
713         r = -EINVAL;
714         /* General sanity checks */
715         if (mem->memory_size & (PAGE_SIZE - 1))
716                 goto out;
717         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
718                 goto out;
719         /* We can read the guest memory with __xxx_user() later on. */
720         if (user_alloc &&
721             ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
722              !access_ok(VERIFY_WRITE,
723                         (void __user *)(unsigned long)mem->userspace_addr,
724                         mem->memory_size)))
725                 goto out;
726         if (mem->slot >= KVM_MEM_SLOTS_NUM)
727                 goto out;
728         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
729                 goto out;
730
731         memslot = id_to_memslot(kvm->memslots, mem->slot);
732         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
733         npages = mem->memory_size >> PAGE_SHIFT;
734
735         r = -EINVAL;
736         if (npages > KVM_MEM_MAX_NR_PAGES)
737                 goto out;
738
739         if (!npages)
740                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
741
742         new = old = *memslot;
743
744         new.id = mem->slot;
745         new.base_gfn = base_gfn;
746         new.npages = npages;
747         new.flags = mem->flags;
748
749         /* Disallow changing a memory slot's size. */
750         r = -EINVAL;
751         if (npages && old.npages && npages != old.npages)
752                 goto out_free;
753
754         /* Check for overlaps */
755         r = -EEXIST;
756         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
757                 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
758
759                 if (s == memslot || !s->npages)
760                         continue;
761                 if (!((base_gfn + npages <= s->base_gfn) ||
762                       (base_gfn >= s->base_gfn + s->npages)))
763                         goto out_free;
764         }
765
766         /* Free page dirty bitmap if unneeded */
767         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
768                 new.dirty_bitmap = NULL;
769
770         r = -ENOMEM;
771
772         /* Allocate if a slot is being created */
773         if (npages && !old.npages) {
774                 new.user_alloc = user_alloc;
775                 new.userspace_addr = mem->userspace_addr;
776
777                 if (kvm_arch_create_memslot(&new, npages))
778                         goto out_free;
779         }
780
781         /* Allocate page dirty bitmap if needed */
782         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
783                 if (kvm_create_dirty_bitmap(&new) < 0)
784                         goto out_free;
785                 /* destroy any largepage mappings for dirty tracking */
786         }
787
788         if (!npages) {
789                 struct kvm_memory_slot *slot;
790
791                 r = -ENOMEM;
792                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
793                                 GFP_KERNEL);
794                 if (!slots)
795                         goto out_free;
796                 slot = id_to_memslot(slots, mem->slot);
797                 slot->flags |= KVM_MEMSLOT_INVALID;
798
799                 update_memslots(slots, NULL);
800
801                 old_memslots = kvm->memslots;
802                 rcu_assign_pointer(kvm->memslots, slots);
803                 synchronize_srcu_expedited(&kvm->srcu);
804                 /* From this point no new shadow pages pointing to a deleted
805                  * memslot will be created.
806                  *
807                  * validation of sp->gfn happens in:
808                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
809                  *      - kvm_is_visible_gfn (mmu_check_roots)
810                  */
811                 kvm_arch_flush_shadow(kvm);
812                 kfree(old_memslots);
813         }
814
815         r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
816         if (r)
817                 goto out_free;
818
819         /* map/unmap the pages in iommu page table */
820         if (npages) {
821                 r = kvm_iommu_map_pages(kvm, &new);
822                 if (r)
823                         goto out_free;
824         } else
825                 kvm_iommu_unmap_pages(kvm, &old);
826
827         r = -ENOMEM;
828         slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
829                         GFP_KERNEL);
830         if (!slots)
831                 goto out_free;
832
833         /* actual memory is freed via old in kvm_free_physmem_slot below */
834         if (!npages) {
835                 new.dirty_bitmap = NULL;
836                 memset(&new.arch, 0, sizeof(new.arch));
837         }
838
839         update_memslots(slots, &new);
840         old_memslots = kvm->memslots;
841         rcu_assign_pointer(kvm->memslots, slots);
842         synchronize_srcu_expedited(&kvm->srcu);
843
844         kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
845
846         /*
847          * If the new memory slot is created, we need to clear all
848          * mmio sptes.
849          */
850         if (npages && old.base_gfn != mem->guest_phys_addr >> PAGE_SHIFT)
851                 kvm_arch_flush_shadow(kvm);
852
853         kvm_free_physmem_slot(&old, &new);
854         kfree(old_memslots);
855
856         return 0;
857
858 out_free:
859         kvm_free_physmem_slot(&new, &old);
860 out:
861         return r;
862
863 }
864 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
865
866 int kvm_set_memory_region(struct kvm *kvm,
867                           struct kvm_userspace_memory_region *mem,
868                           int user_alloc)
869 {
870         int r;
871
872         mutex_lock(&kvm->slots_lock);
873         r = __kvm_set_memory_region(kvm, mem, user_alloc);
874         mutex_unlock(&kvm->slots_lock);
875         return r;
876 }
877 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
878
879 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
880                                    struct
881                                    kvm_userspace_memory_region *mem,
882                                    int user_alloc)
883 {
884         if (mem->slot >= KVM_MEMORY_SLOTS)
885                 return -EINVAL;
886         return kvm_set_memory_region(kvm, mem, user_alloc);
887 }
888
889 int kvm_get_dirty_log(struct kvm *kvm,
890                         struct kvm_dirty_log *log, int *is_dirty)
891 {
892         struct kvm_memory_slot *memslot;
893         int r, i;
894         unsigned long n;
895         unsigned long any = 0;
896
897         r = -EINVAL;
898         if (log->slot >= KVM_MEMORY_SLOTS)
899                 goto out;
900
901         memslot = id_to_memslot(kvm->memslots, log->slot);
902         r = -ENOENT;
903         if (!memslot->dirty_bitmap)
904                 goto out;
905
906         n = kvm_dirty_bitmap_bytes(memslot);
907
908         for (i = 0; !any && i < n/sizeof(long); ++i)
909                 any = memslot->dirty_bitmap[i];
910
911         r = -EFAULT;
912         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
913                 goto out;
914
915         if (any)
916                 *is_dirty = 1;
917
918         r = 0;
919 out:
920         return r;
921 }
922
923 bool kvm_largepages_enabled(void)
924 {
925         return largepages_enabled;
926 }
927
928 void kvm_disable_largepages(void)
929 {
930         largepages_enabled = false;
931 }
932 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
933
934 static inline unsigned long bad_hva(void)
935 {
936         return PAGE_OFFSET;
937 }
938
939 int kvm_is_error_hva(unsigned long addr)
940 {
941         return addr == bad_hva();
942 }
943 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
944
945 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
946 {
947         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
948 }
949 EXPORT_SYMBOL_GPL(gfn_to_memslot);
950
951 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
952 {
953         struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
954
955         if (!memslot || memslot->id >= KVM_MEMORY_SLOTS ||
956               memslot->flags & KVM_MEMSLOT_INVALID)
957                 return 0;
958
959         return 1;
960 }
961 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
962
963 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
964 {
965         struct vm_area_struct *vma;
966         unsigned long addr, size;
967
968         size = PAGE_SIZE;
969
970         addr = gfn_to_hva(kvm, gfn);
971         if (kvm_is_error_hva(addr))
972                 return PAGE_SIZE;
973
974         down_read(&current->mm->mmap_sem);
975         vma = find_vma(current->mm, addr);
976         if (!vma)
977                 goto out;
978
979         size = vma_kernel_pagesize(vma);
980
981 out:
982         up_read(&current->mm->mmap_sem);
983
984         return size;
985 }
986
987 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
988                                      gfn_t *nr_pages)
989 {
990         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
991                 return bad_hva();
992
993         if (nr_pages)
994                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
995
996         return gfn_to_hva_memslot(slot, gfn);
997 }
998
999 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1000 {
1001         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1002 }
1003 EXPORT_SYMBOL_GPL(gfn_to_hva);
1004
1005 /*
1006  * The hva returned by this function is only allowed to be read.
1007  * It should pair with kvm_read_hva() or kvm_read_hva_atomic().
1008  */
1009 static unsigned long gfn_to_hva_read(struct kvm *kvm, gfn_t gfn)
1010 {
1011         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1012 }
1013
1014 static int kvm_read_hva(void *data, void __user *hva, int len)
1015 {
1016         return __copy_from_user(data, hva, len);
1017 }
1018
1019 static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1020 {
1021         return __copy_from_user_inatomic(data, hva, len);
1022 }
1023
1024 int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1025         unsigned long start, int write, struct page **page)
1026 {
1027         int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1028
1029         if (write)
1030                 flags |= FOLL_WRITE;
1031
1032         return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1033 }
1034
1035 static inline int check_user_page_hwpoison(unsigned long addr)
1036 {
1037         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1038
1039         rc = __get_user_pages(current, current->mm, addr, 1,
1040                               flags, NULL, NULL, NULL);
1041         return rc == -EHWPOISON;
1042 }
1043
1044 /*
1045  * The atomic path to get the writable pfn which will be stored in @pfn,
1046  * true indicates success, otherwise false is returned.
1047  */
1048 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1049                             bool write_fault, bool *writable, pfn_t *pfn)
1050 {
1051         struct page *page[1];
1052         int npages;
1053
1054         if (!(async || atomic))
1055                 return false;
1056
1057         npages = __get_user_pages_fast(addr, 1, 1, page);
1058         if (npages == 1) {
1059                 *pfn = page_to_pfn(page[0]);
1060
1061                 if (writable)
1062                         *writable = true;
1063                 return true;
1064         }
1065
1066         return false;
1067 }
1068
1069 /*
1070  * The slow path to get the pfn of the specified host virtual address,
1071  * 1 indicates success, -errno is returned if error is detected.
1072  */
1073 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1074                            bool *writable, pfn_t *pfn)
1075 {
1076         struct page *page[1];
1077         int npages = 0;
1078
1079         might_sleep();
1080
1081         if (writable)
1082                 *writable = write_fault;
1083
1084         if (async) {
1085                 down_read(&current->mm->mmap_sem);
1086                 npages = get_user_page_nowait(current, current->mm,
1087                                               addr, write_fault, page);
1088                 up_read(&current->mm->mmap_sem);
1089         } else
1090                 npages = get_user_pages_fast(addr, 1, write_fault,
1091                                              page);
1092         if (npages != 1)
1093                 return npages;
1094
1095         /* map read fault as writable if possible */
1096         if (unlikely(!write_fault)) {
1097                 struct page *wpage[1];
1098
1099                 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1100                 if (npages == 1) {
1101                         *writable = true;
1102                         put_page(page[0]);
1103                         page[0] = wpage[0];
1104                 }
1105
1106                 npages = 1;
1107         }
1108         *pfn = page_to_pfn(page[0]);
1109         return npages;
1110 }
1111
1112 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1113                         bool write_fault, bool *writable)
1114 {
1115         struct vm_area_struct *vma;
1116         pfn_t pfn = 0;
1117         int npages;
1118
1119         /* we can do it either atomically or asynchronously, not both */
1120         BUG_ON(atomic && async);
1121
1122         BUG_ON(!write_fault && !writable);
1123
1124         if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1125                 return pfn;
1126
1127         if (atomic)
1128                 return KVM_PFN_ERR_FAULT;
1129
1130         npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1131         if (npages == 1)
1132                 return pfn;
1133
1134         down_read(&current->mm->mmap_sem);
1135         if (npages == -EHWPOISON ||
1136               (!async && check_user_page_hwpoison(addr))) {
1137                 pfn = KVM_PFN_ERR_HWPOISON;
1138                 goto exit;
1139         }
1140
1141         vma = find_vma_intersection(current->mm, addr, addr + 1);
1142
1143         if (vma == NULL)
1144                 pfn = KVM_PFN_ERR_FAULT;
1145         else if ((vma->vm_flags & VM_PFNMAP)) {
1146                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1147                         vma->vm_pgoff;
1148                 BUG_ON(!kvm_is_mmio_pfn(pfn));
1149         } else {
1150                 if (async && (vma->vm_flags & VM_WRITE))
1151                         *async = true;
1152                 pfn = KVM_PFN_ERR_FAULT;
1153         }
1154 exit:
1155         up_read(&current->mm->mmap_sem);
1156         return pfn;
1157 }
1158
1159 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1160                           bool write_fault, bool *writable)
1161 {
1162         unsigned long addr;
1163
1164         if (async)
1165                 *async = false;
1166
1167         addr = gfn_to_hva(kvm, gfn);
1168         if (kvm_is_error_hva(addr))
1169                 return KVM_PFN_ERR_BAD;
1170
1171         return hva_to_pfn(addr, atomic, async, write_fault, writable);
1172 }
1173
1174 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1175 {
1176         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1177 }
1178 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1179
1180 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1181                        bool write_fault, bool *writable)
1182 {
1183         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1184 }
1185 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1186
1187 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1188 {
1189         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1190 }
1191 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1192
1193 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1194                       bool *writable)
1195 {
1196         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1197 }
1198 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1199
1200 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1201 {
1202         unsigned long addr = gfn_to_hva_memslot(slot, gfn);
1203         return hva_to_pfn(addr, false, NULL, true, NULL);
1204 }
1205
1206 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1207 {
1208         unsigned long addr = gfn_to_hva_memslot(slot, gfn);
1209
1210         return hva_to_pfn(addr, true, NULL, true, NULL);
1211 }
1212 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1213
1214 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1215                                                                   int nr_pages)
1216 {
1217         unsigned long addr;
1218         gfn_t entry;
1219
1220         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1221         if (kvm_is_error_hva(addr))
1222                 return -1;
1223
1224         if (entry < nr_pages)
1225                 return 0;
1226
1227         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1228 }
1229 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1230
1231 static struct page *kvm_pfn_to_page(pfn_t pfn)
1232 {
1233         if (is_error_pfn(pfn))
1234                 return KVM_ERR_PTR_BAD_PAGE;
1235
1236         if (kvm_is_mmio_pfn(pfn)) {
1237                 WARN_ON(1);
1238                 return KVM_ERR_PTR_BAD_PAGE;
1239         }
1240
1241         return pfn_to_page(pfn);
1242 }
1243
1244 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1245 {
1246         pfn_t pfn;
1247
1248         pfn = gfn_to_pfn(kvm, gfn);
1249
1250         return kvm_pfn_to_page(pfn);
1251 }
1252
1253 EXPORT_SYMBOL_GPL(gfn_to_page);
1254
1255 void kvm_release_page_clean(struct page *page)
1256 {
1257         WARN_ON(is_error_page(page));
1258
1259         kvm_release_pfn_clean(page_to_pfn(page));
1260 }
1261 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1262
1263 void kvm_release_pfn_clean(pfn_t pfn)
1264 {
1265         WARN_ON(is_error_pfn(pfn));
1266
1267         if (!kvm_is_mmio_pfn(pfn))
1268                 put_page(pfn_to_page(pfn));
1269 }
1270 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1271
1272 void kvm_release_page_dirty(struct page *page)
1273 {
1274         WARN_ON(is_error_page(page));
1275
1276         kvm_release_pfn_dirty(page_to_pfn(page));
1277 }
1278 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1279
1280 void kvm_release_pfn_dirty(pfn_t pfn)
1281 {
1282         kvm_set_pfn_dirty(pfn);
1283         kvm_release_pfn_clean(pfn);
1284 }
1285 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1286
1287 void kvm_set_page_dirty(struct page *page)
1288 {
1289         kvm_set_pfn_dirty(page_to_pfn(page));
1290 }
1291 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1292
1293 void kvm_set_pfn_dirty(pfn_t pfn)
1294 {
1295         if (!kvm_is_mmio_pfn(pfn)) {
1296                 struct page *page = pfn_to_page(pfn);
1297                 if (!PageReserved(page))
1298                         SetPageDirty(page);
1299         }
1300 }
1301 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1302
1303 void kvm_set_pfn_accessed(pfn_t pfn)
1304 {
1305         if (!kvm_is_mmio_pfn(pfn))
1306                 mark_page_accessed(pfn_to_page(pfn));
1307 }
1308 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1309
1310 void kvm_get_pfn(pfn_t pfn)
1311 {
1312         if (!kvm_is_mmio_pfn(pfn))
1313                 get_page(pfn_to_page(pfn));
1314 }
1315 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1316
1317 static int next_segment(unsigned long len, int offset)
1318 {
1319         if (len > PAGE_SIZE - offset)
1320                 return PAGE_SIZE - offset;
1321         else
1322                 return len;
1323 }
1324
1325 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1326                         int len)
1327 {
1328         int r;
1329         unsigned long addr;
1330
1331         addr = gfn_to_hva_read(kvm, gfn);
1332         if (kvm_is_error_hva(addr))
1333                 return -EFAULT;
1334         r = kvm_read_hva(data, (void __user *)addr + offset, len);
1335         if (r)
1336                 return -EFAULT;
1337         return 0;
1338 }
1339 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1340
1341 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1342 {
1343         gfn_t gfn = gpa >> PAGE_SHIFT;
1344         int seg;
1345         int offset = offset_in_page(gpa);
1346         int ret;
1347
1348         while ((seg = next_segment(len, offset)) != 0) {
1349                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1350                 if (ret < 0)
1351                         return ret;
1352                 offset = 0;
1353                 len -= seg;
1354                 data += seg;
1355                 ++gfn;
1356         }
1357         return 0;
1358 }
1359 EXPORT_SYMBOL_GPL(kvm_read_guest);
1360
1361 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1362                           unsigned long len)
1363 {
1364         int r;
1365         unsigned long addr;
1366         gfn_t gfn = gpa >> PAGE_SHIFT;
1367         int offset = offset_in_page(gpa);
1368
1369         addr = gfn_to_hva_read(kvm, gfn);
1370         if (kvm_is_error_hva(addr))
1371                 return -EFAULT;
1372         pagefault_disable();
1373         r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
1374         pagefault_enable();
1375         if (r)
1376                 return -EFAULT;
1377         return 0;
1378 }
1379 EXPORT_SYMBOL(kvm_read_guest_atomic);
1380
1381 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1382                          int offset, int len)
1383 {
1384         int r;
1385         unsigned long addr;
1386
1387         addr = gfn_to_hva(kvm, gfn);
1388         if (kvm_is_error_hva(addr))
1389                 return -EFAULT;
1390         r = __copy_to_user((void __user *)addr + offset, data, len);
1391         if (r)
1392                 return -EFAULT;
1393         mark_page_dirty(kvm, gfn);
1394         return 0;
1395 }
1396 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1397
1398 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1399                     unsigned long len)
1400 {
1401         gfn_t gfn = gpa >> PAGE_SHIFT;
1402         int seg;
1403         int offset = offset_in_page(gpa);
1404         int ret;
1405
1406         while ((seg = next_segment(len, offset)) != 0) {
1407                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1408                 if (ret < 0)
1409                         return ret;
1410                 offset = 0;
1411                 len -= seg;
1412                 data += seg;
1413                 ++gfn;
1414         }
1415         return 0;
1416 }
1417
1418 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1419                               gpa_t gpa)
1420 {
1421         struct kvm_memslots *slots = kvm_memslots(kvm);
1422         int offset = offset_in_page(gpa);
1423         gfn_t gfn = gpa >> PAGE_SHIFT;
1424
1425         ghc->gpa = gpa;
1426         ghc->generation = slots->generation;
1427         ghc->memslot = gfn_to_memslot(kvm, gfn);
1428         ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1429         if (!kvm_is_error_hva(ghc->hva))
1430                 ghc->hva += offset;
1431         else
1432                 return -EFAULT;
1433
1434         return 0;
1435 }
1436 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1437
1438 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1439                            void *data, unsigned long len)
1440 {
1441         struct kvm_memslots *slots = kvm_memslots(kvm);
1442         int r;
1443
1444         if (slots->generation != ghc->generation)
1445                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1446
1447         if (kvm_is_error_hva(ghc->hva))
1448                 return -EFAULT;
1449
1450         r = __copy_to_user((void __user *)ghc->hva, data, len);
1451         if (r)
1452                 return -EFAULT;
1453         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1454
1455         return 0;
1456 }
1457 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1458
1459 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1460                            void *data, unsigned long len)
1461 {
1462         struct kvm_memslots *slots = kvm_memslots(kvm);
1463         int r;
1464
1465         if (slots->generation != ghc->generation)
1466                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1467
1468         if (kvm_is_error_hva(ghc->hva))
1469                 return -EFAULT;
1470
1471         r = __copy_from_user(data, (void __user *)ghc->hva, len);
1472         if (r)
1473                 return -EFAULT;
1474
1475         return 0;
1476 }
1477 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1478
1479 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1480 {
1481         return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1482                                     offset, len);
1483 }
1484 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1485
1486 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1487 {
1488         gfn_t gfn = gpa >> PAGE_SHIFT;
1489         int seg;
1490         int offset = offset_in_page(gpa);
1491         int ret;
1492
1493         while ((seg = next_segment(len, offset)) != 0) {
1494                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1495                 if (ret < 0)
1496                         return ret;
1497                 offset = 0;
1498                 len -= seg;
1499                 ++gfn;
1500         }
1501         return 0;
1502 }
1503 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1504
1505 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1506                              gfn_t gfn)
1507 {
1508         if (memslot && memslot->dirty_bitmap) {
1509                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1510
1511                 /* TODO: introduce set_bit_le() and use it */
1512                 test_and_set_bit_le(rel_gfn, memslot->dirty_bitmap);
1513         }
1514 }
1515
1516 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1517 {
1518         struct kvm_memory_slot *memslot;
1519
1520         memslot = gfn_to_memslot(kvm, gfn);
1521         mark_page_dirty_in_slot(kvm, memslot, gfn);
1522 }
1523
1524 /*
1525  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1526  */
1527 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1528 {
1529         DEFINE_WAIT(wait);
1530
1531         for (;;) {
1532                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1533
1534                 if (kvm_arch_vcpu_runnable(vcpu)) {
1535                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1536                         break;
1537                 }
1538                 if (kvm_cpu_has_pending_timer(vcpu))
1539                         break;
1540                 if (signal_pending(current))
1541                         break;
1542
1543                 schedule();
1544         }
1545
1546         finish_wait(&vcpu->wq, &wait);
1547 }
1548
1549 #ifndef CONFIG_S390
1550 /*
1551  * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1552  */
1553 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1554 {
1555         int me;
1556         int cpu = vcpu->cpu;
1557         wait_queue_head_t *wqp;
1558
1559         wqp = kvm_arch_vcpu_wq(vcpu);
1560         if (waitqueue_active(wqp)) {
1561                 wake_up_interruptible(wqp);
1562                 ++vcpu->stat.halt_wakeup;
1563         }
1564
1565         me = get_cpu();
1566         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1567                 if (kvm_arch_vcpu_should_kick(vcpu))
1568                         smp_send_reschedule(cpu);
1569         put_cpu();
1570 }
1571 #endif /* !CONFIG_S390 */
1572
1573 void kvm_resched(struct kvm_vcpu *vcpu)
1574 {
1575         if (!need_resched())
1576                 return;
1577         cond_resched();
1578 }
1579 EXPORT_SYMBOL_GPL(kvm_resched);
1580
1581 bool kvm_vcpu_yield_to(struct kvm_vcpu *target)
1582 {
1583         struct pid *pid;
1584         struct task_struct *task = NULL;
1585
1586         rcu_read_lock();
1587         pid = rcu_dereference(target->pid);
1588         if (pid)
1589                 task = get_pid_task(target->pid, PIDTYPE_PID);
1590         rcu_read_unlock();
1591         if (!task)
1592                 return false;
1593         if (task->flags & PF_VCPU) {
1594                 put_task_struct(task);
1595                 return false;
1596         }
1597         if (yield_to(task, 1)) {
1598                 put_task_struct(task);
1599                 return true;
1600         }
1601         put_task_struct(task);
1602         return false;
1603 }
1604 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1605
1606 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1607 /*
1608  * Helper that checks whether a VCPU is eligible for directed yield.
1609  * Most eligible candidate to yield is decided by following heuristics:
1610  *
1611  *  (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1612  *  (preempted lock holder), indicated by @in_spin_loop.
1613  *  Set at the beiginning and cleared at the end of interception/PLE handler.
1614  *
1615  *  (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1616  *  chance last time (mostly it has become eligible now since we have probably
1617  *  yielded to lockholder in last iteration. This is done by toggling
1618  *  @dy_eligible each time a VCPU checked for eligibility.)
1619  *
1620  *  Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1621  *  to preempted lock-holder could result in wrong VCPU selection and CPU
1622  *  burning. Giving priority for a potential lock-holder increases lock
1623  *  progress.
1624  *
1625  *  Since algorithm is based on heuristics, accessing another VCPU data without
1626  *  locking does not harm. It may result in trying to yield to  same VCPU, fail
1627  *  and continue with next VCPU and so on.
1628  */
1629 bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1630 {
1631         bool eligible;
1632
1633         eligible = !vcpu->spin_loop.in_spin_loop ||
1634                         (vcpu->spin_loop.in_spin_loop &&
1635                          vcpu->spin_loop.dy_eligible);
1636
1637         if (vcpu->spin_loop.in_spin_loop)
1638                 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1639
1640         return eligible;
1641 }
1642 #endif
1643 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1644 {
1645         struct kvm *kvm = me->kvm;
1646         struct kvm_vcpu *vcpu;
1647         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1648         int yielded = 0;
1649         int pass;
1650         int i;
1651
1652         kvm_vcpu_set_in_spin_loop(me, true);
1653         /*
1654          * We boost the priority of a VCPU that is runnable but not
1655          * currently running, because it got preempted by something
1656          * else and called schedule in __vcpu_run.  Hopefully that
1657          * VCPU is holding the lock that we need and will release it.
1658          * We approximate round-robin by starting at the last boosted VCPU.
1659          */
1660         for (pass = 0; pass < 2 && !yielded; pass++) {
1661                 kvm_for_each_vcpu(i, vcpu, kvm) {
1662                         if (!pass && i <= last_boosted_vcpu) {
1663                                 i = last_boosted_vcpu;
1664                                 continue;
1665                         } else if (pass && i > last_boosted_vcpu)
1666                                 break;
1667                         if (vcpu == me)
1668                                 continue;
1669                         if (waitqueue_active(&vcpu->wq))
1670                                 continue;
1671                         if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1672                                 continue;
1673                         if (kvm_vcpu_yield_to(vcpu)) {
1674                                 kvm->last_boosted_vcpu = i;
1675                                 yielded = 1;
1676                                 break;
1677                         }
1678                 }
1679         }
1680         kvm_vcpu_set_in_spin_loop(me, false);
1681
1682         /* Ensure vcpu is not eligible during next spinloop */
1683         kvm_vcpu_set_dy_eligible(me, false);
1684 }
1685 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1686
1687 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1688 {
1689         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1690         struct page *page;
1691
1692         if (vmf->pgoff == 0)
1693                 page = virt_to_page(vcpu->run);
1694 #ifdef CONFIG_X86
1695         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1696                 page = virt_to_page(vcpu->arch.pio_data);
1697 #endif
1698 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1699         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1700                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1701 #endif
1702         else
1703                 return kvm_arch_vcpu_fault(vcpu, vmf);
1704         get_page(page);
1705         vmf->page = page;
1706         return 0;
1707 }
1708
1709 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1710         .fault = kvm_vcpu_fault,
1711 };
1712
1713 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1714 {
1715         vma->vm_ops = &kvm_vcpu_vm_ops;
1716         return 0;
1717 }
1718
1719 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1720 {
1721         struct kvm_vcpu *vcpu = filp->private_data;
1722
1723         kvm_put_kvm(vcpu->kvm);
1724         return 0;
1725 }
1726
1727 static struct file_operations kvm_vcpu_fops = {
1728         .release        = kvm_vcpu_release,
1729         .unlocked_ioctl = kvm_vcpu_ioctl,
1730 #ifdef CONFIG_COMPAT
1731         .compat_ioctl   = kvm_vcpu_compat_ioctl,
1732 #endif
1733         .mmap           = kvm_vcpu_mmap,
1734         .llseek         = noop_llseek,
1735 };
1736
1737 /*
1738  * Allocates an inode for the vcpu.
1739  */
1740 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1741 {
1742         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1743 }
1744
1745 /*
1746  * Creates some virtual cpus.  Good luck creating more than one.
1747  */
1748 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1749 {
1750         int r;
1751         struct kvm_vcpu *vcpu, *v;
1752
1753         vcpu = kvm_arch_vcpu_create(kvm, id);
1754         if (IS_ERR(vcpu))
1755                 return PTR_ERR(vcpu);
1756
1757         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1758
1759         r = kvm_arch_vcpu_setup(vcpu);
1760         if (r)
1761                 goto vcpu_destroy;
1762
1763         mutex_lock(&kvm->lock);
1764         if (!kvm_vcpu_compatible(vcpu)) {
1765                 r = -EINVAL;
1766                 goto unlock_vcpu_destroy;
1767         }
1768         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1769                 r = -EINVAL;
1770                 goto unlock_vcpu_destroy;
1771         }
1772
1773         kvm_for_each_vcpu(r, v, kvm)
1774                 if (v->vcpu_id == id) {
1775                         r = -EEXIST;
1776                         goto unlock_vcpu_destroy;
1777                 }
1778
1779         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1780
1781         /* Now it's all set up, let userspace reach it */
1782         kvm_get_kvm(kvm);
1783         r = create_vcpu_fd(vcpu);
1784         if (r < 0) {
1785                 kvm_put_kvm(kvm);
1786                 goto unlock_vcpu_destroy;
1787         }
1788
1789         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1790         smp_wmb();
1791         atomic_inc(&kvm->online_vcpus);
1792
1793         mutex_unlock(&kvm->lock);
1794         return r;
1795
1796 unlock_vcpu_destroy:
1797         mutex_unlock(&kvm->lock);
1798 vcpu_destroy:
1799         kvm_arch_vcpu_destroy(vcpu);
1800         return r;
1801 }
1802
1803 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1804 {
1805         if (sigset) {
1806                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1807                 vcpu->sigset_active = 1;
1808                 vcpu->sigset = *sigset;
1809         } else
1810                 vcpu->sigset_active = 0;
1811         return 0;
1812 }
1813
1814 static long kvm_vcpu_ioctl(struct file *filp,
1815                            unsigned int ioctl, unsigned long arg)
1816 {
1817         struct kvm_vcpu *vcpu = filp->private_data;
1818         void __user *argp = (void __user *)arg;
1819         int r;
1820         struct kvm_fpu *fpu = NULL;
1821         struct kvm_sregs *kvm_sregs = NULL;
1822
1823         if (vcpu->kvm->mm != current->mm)
1824                 return -EIO;
1825
1826 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1827         /*
1828          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1829          * so vcpu_load() would break it.
1830          */
1831         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1832                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1833 #endif
1834
1835
1836         vcpu_load(vcpu);
1837         switch (ioctl) {
1838         case KVM_RUN:
1839                 r = -EINVAL;
1840                 if (arg)
1841                         goto out;
1842                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1843                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1844                 break;
1845         case KVM_GET_REGS: {
1846                 struct kvm_regs *kvm_regs;
1847
1848                 r = -ENOMEM;
1849                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1850                 if (!kvm_regs)
1851                         goto out;
1852                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1853                 if (r)
1854                         goto out_free1;
1855                 r = -EFAULT;
1856                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1857                         goto out_free1;
1858                 r = 0;
1859 out_free1:
1860                 kfree(kvm_regs);
1861                 break;
1862         }
1863         case KVM_SET_REGS: {
1864                 struct kvm_regs *kvm_regs;
1865
1866                 r = -ENOMEM;
1867                 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
1868                 if (IS_ERR(kvm_regs)) {
1869                         r = PTR_ERR(kvm_regs);
1870                         goto out;
1871                 }
1872                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1873                 if (r)
1874                         goto out_free2;
1875                 r = 0;
1876 out_free2:
1877                 kfree(kvm_regs);
1878                 break;
1879         }
1880         case KVM_GET_SREGS: {
1881                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1882                 r = -ENOMEM;
1883                 if (!kvm_sregs)
1884                         goto out;
1885                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1886                 if (r)
1887                         goto out;
1888                 r = -EFAULT;
1889                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1890                         goto out;
1891                 r = 0;
1892                 break;
1893         }
1894         case KVM_SET_SREGS: {
1895                 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
1896                 if (IS_ERR(kvm_sregs)) {
1897                         r = PTR_ERR(kvm_sregs);
1898                         goto out;
1899                 }
1900                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1901                 if (r)
1902                         goto out;
1903                 r = 0;
1904                 break;
1905         }
1906         case KVM_GET_MP_STATE: {
1907                 struct kvm_mp_state mp_state;
1908
1909                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1910                 if (r)
1911                         goto out;
1912                 r = -EFAULT;
1913                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1914                         goto out;
1915                 r = 0;
1916                 break;
1917         }
1918         case KVM_SET_MP_STATE: {
1919                 struct kvm_mp_state mp_state;
1920
1921                 r = -EFAULT;
1922                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1923                         goto out;
1924                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1925                 if (r)
1926                         goto out;
1927                 r = 0;
1928                 break;
1929         }
1930         case KVM_TRANSLATE: {
1931                 struct kvm_translation tr;
1932
1933                 r = -EFAULT;
1934                 if (copy_from_user(&tr, argp, sizeof tr))
1935                         goto out;
1936                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1937                 if (r)
1938                         goto out;
1939                 r = -EFAULT;
1940                 if (copy_to_user(argp, &tr, sizeof tr))
1941                         goto out;
1942                 r = 0;
1943                 break;
1944         }
1945         case KVM_SET_GUEST_DEBUG: {
1946                 struct kvm_guest_debug dbg;
1947
1948                 r = -EFAULT;
1949                 if (copy_from_user(&dbg, argp, sizeof dbg))
1950                         goto out;
1951                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1952                 if (r)
1953                         goto out;
1954                 r = 0;
1955                 break;
1956         }
1957         case KVM_SET_SIGNAL_MASK: {
1958                 struct kvm_signal_mask __user *sigmask_arg = argp;
1959                 struct kvm_signal_mask kvm_sigmask;
1960                 sigset_t sigset, *p;
1961
1962                 p = NULL;
1963                 if (argp) {
1964                         r = -EFAULT;
1965                         if (copy_from_user(&kvm_sigmask, argp,
1966                                            sizeof kvm_sigmask))
1967                                 goto out;
1968                         r = -EINVAL;
1969                         if (kvm_sigmask.len != sizeof sigset)
1970                                 goto out;
1971                         r = -EFAULT;
1972                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1973                                            sizeof sigset))
1974                                 goto out;
1975                         p = &sigset;
1976                 }
1977                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
1978                 break;
1979         }
1980         case KVM_GET_FPU: {
1981                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1982                 r = -ENOMEM;
1983                 if (!fpu)
1984                         goto out;
1985                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1986                 if (r)
1987                         goto out;
1988                 r = -EFAULT;
1989                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1990                         goto out;
1991                 r = 0;
1992                 break;
1993         }
1994         case KVM_SET_FPU: {
1995                 fpu = memdup_user(argp, sizeof(*fpu));
1996                 if (IS_ERR(fpu)) {
1997                         r = PTR_ERR(fpu);
1998                         goto out;
1999                 }
2000                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2001                 if (r)
2002                         goto out;
2003                 r = 0;
2004                 break;
2005         }
2006         default:
2007                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2008         }
2009 out:
2010         vcpu_put(vcpu);
2011         kfree(fpu);
2012         kfree(kvm_sregs);
2013         return r;
2014 }
2015
2016 #ifdef CONFIG_COMPAT
2017 static long kvm_vcpu_compat_ioctl(struct file *filp,
2018                                   unsigned int ioctl, unsigned long arg)
2019 {
2020         struct kvm_vcpu *vcpu = filp->private_data;
2021         void __user *argp = compat_ptr(arg);
2022         int r;
2023
2024         if (vcpu->kvm->mm != current->mm)
2025                 return -EIO;
2026
2027         switch (ioctl) {
2028         case KVM_SET_SIGNAL_MASK: {
2029                 struct kvm_signal_mask __user *sigmask_arg = argp;
2030                 struct kvm_signal_mask kvm_sigmask;
2031                 compat_sigset_t csigset;
2032                 sigset_t sigset;
2033
2034                 if (argp) {
2035                         r = -EFAULT;
2036                         if (copy_from_user(&kvm_sigmask, argp,
2037                                            sizeof kvm_sigmask))
2038                                 goto out;
2039                         r = -EINVAL;
2040                         if (kvm_sigmask.len != sizeof csigset)
2041                                 goto out;
2042                         r = -EFAULT;
2043                         if (copy_from_user(&csigset, sigmask_arg->sigset,
2044                                            sizeof csigset))
2045                                 goto out;
2046                 }
2047                 sigset_from_compat(&sigset, &csigset);
2048                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2049                 break;
2050         }
2051         default:
2052                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2053         }
2054
2055 out:
2056         return r;
2057 }
2058 #endif
2059
2060 static long kvm_vm_ioctl(struct file *filp,
2061                            unsigned int ioctl, unsigned long arg)
2062 {
2063         struct kvm *kvm = filp->private_data;
2064         void __user *argp = (void __user *)arg;
2065         int r;
2066
2067         if (kvm->mm != current->mm)
2068                 return -EIO;
2069         switch (ioctl) {
2070         case KVM_CREATE_VCPU:
2071                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2072                 if (r < 0)
2073                         goto out;
2074                 break;
2075         case KVM_SET_USER_MEMORY_REGION: {
2076                 struct kvm_userspace_memory_region kvm_userspace_mem;
2077
2078                 r = -EFAULT;
2079                 if (copy_from_user(&kvm_userspace_mem, argp,
2080                                                 sizeof kvm_userspace_mem))
2081                         goto out;
2082
2083                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
2084                 if (r)
2085                         goto out;
2086                 break;
2087         }
2088         case KVM_GET_DIRTY_LOG: {
2089                 struct kvm_dirty_log log;
2090
2091                 r = -EFAULT;
2092                 if (copy_from_user(&log, argp, sizeof log))
2093                         goto out;
2094                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2095                 if (r)
2096                         goto out;
2097                 break;
2098         }
2099 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2100         case KVM_REGISTER_COALESCED_MMIO: {
2101                 struct kvm_coalesced_mmio_zone zone;
2102                 r = -EFAULT;
2103                 if (copy_from_user(&zone, argp, sizeof zone))
2104                         goto out;
2105                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2106                 if (r)
2107                         goto out;
2108                 r = 0;
2109                 break;
2110         }
2111         case KVM_UNREGISTER_COALESCED_MMIO: {
2112                 struct kvm_coalesced_mmio_zone zone;
2113                 r = -EFAULT;
2114                 if (copy_from_user(&zone, argp, sizeof zone))
2115                         goto out;
2116                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2117                 if (r)
2118                         goto out;
2119                 r = 0;
2120                 break;
2121         }
2122 #endif
2123         case KVM_IRQFD: {
2124                 struct kvm_irqfd data;
2125
2126                 r = -EFAULT;
2127                 if (copy_from_user(&data, argp, sizeof data))
2128                         goto out;
2129                 r = kvm_irqfd(kvm, &data);
2130                 break;
2131         }
2132         case KVM_IOEVENTFD: {
2133                 struct kvm_ioeventfd data;
2134
2135                 r = -EFAULT;
2136                 if (copy_from_user(&data, argp, sizeof data))
2137                         goto out;
2138                 r = kvm_ioeventfd(kvm, &data);
2139                 break;
2140         }
2141 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2142         case KVM_SET_BOOT_CPU_ID:
2143                 r = 0;
2144                 mutex_lock(&kvm->lock);
2145                 if (atomic_read(&kvm->online_vcpus) != 0)
2146                         r = -EBUSY;
2147                 else
2148                         kvm->bsp_vcpu_id = arg;
2149                 mutex_unlock(&kvm->lock);
2150                 break;
2151 #endif
2152 #ifdef CONFIG_HAVE_KVM_MSI
2153         case KVM_SIGNAL_MSI: {
2154                 struct kvm_msi msi;
2155
2156                 r = -EFAULT;
2157                 if (copy_from_user(&msi, argp, sizeof msi))
2158                         goto out;
2159                 r = kvm_send_userspace_msi(kvm, &msi);
2160                 break;
2161         }
2162 #endif
2163 #ifdef __KVM_HAVE_IRQ_LINE
2164         case KVM_IRQ_LINE_STATUS:
2165         case KVM_IRQ_LINE: {
2166                 struct kvm_irq_level irq_event;
2167
2168                 r = -EFAULT;
2169                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2170                         goto out;
2171
2172                 r = kvm_vm_ioctl_irq_line(kvm, &irq_event);
2173                 if (r)
2174                         goto out;
2175
2176                 r = -EFAULT;
2177                 if (ioctl == KVM_IRQ_LINE_STATUS) {
2178                         if (copy_to_user(argp, &irq_event, sizeof irq_event))
2179                                 goto out;
2180                 }
2181
2182                 r = 0;
2183                 break;
2184         }
2185 #endif
2186         default:
2187                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2188                 if (r == -ENOTTY)
2189                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2190         }
2191 out:
2192         return r;
2193 }
2194
2195 #ifdef CONFIG_COMPAT
2196 struct compat_kvm_dirty_log {
2197         __u32 slot;
2198         __u32 padding1;
2199         union {
2200                 compat_uptr_t dirty_bitmap; /* one bit per page */
2201                 __u64 padding2;
2202         };
2203 };
2204
2205 static long kvm_vm_compat_ioctl(struct file *filp,
2206                            unsigned int ioctl, unsigned long arg)
2207 {
2208         struct kvm *kvm = filp->private_data;
2209         int r;
2210
2211         if (kvm->mm != current->mm)
2212                 return -EIO;
2213         switch (ioctl) {
2214         case KVM_GET_DIRTY_LOG: {
2215                 struct compat_kvm_dirty_log compat_log;
2216                 struct kvm_dirty_log log;
2217
2218                 r = -EFAULT;
2219                 if (copy_from_user(&compat_log, (void __user *)arg,
2220                                    sizeof(compat_log)))
2221                         goto out;
2222                 log.slot         = compat_log.slot;
2223                 log.padding1     = compat_log.padding1;
2224                 log.padding2     = compat_log.padding2;
2225                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2226
2227                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2228                 if (r)
2229                         goto out;
2230                 break;
2231         }
2232         default:
2233                 r = kvm_vm_ioctl(filp, ioctl, arg);
2234         }
2235
2236 out:
2237         return r;
2238 }
2239 #endif
2240
2241 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2242 {
2243         struct page *page[1];
2244         unsigned long addr;
2245         int npages;
2246         gfn_t gfn = vmf->pgoff;
2247         struct kvm *kvm = vma->vm_file->private_data;
2248
2249         addr = gfn_to_hva(kvm, gfn);
2250         if (kvm_is_error_hva(addr))
2251                 return VM_FAULT_SIGBUS;
2252
2253         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2254                                 NULL);
2255         if (unlikely(npages != 1))
2256                 return VM_FAULT_SIGBUS;
2257
2258         vmf->page = page[0];
2259         return 0;
2260 }
2261
2262 static const struct vm_operations_struct kvm_vm_vm_ops = {
2263         .fault = kvm_vm_fault,
2264 };
2265
2266 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2267 {
2268         vma->vm_ops = &kvm_vm_vm_ops;
2269         return 0;
2270 }
2271
2272 static struct file_operations kvm_vm_fops = {
2273         .release        = kvm_vm_release,
2274         .unlocked_ioctl = kvm_vm_ioctl,
2275 #ifdef CONFIG_COMPAT
2276         .compat_ioctl   = kvm_vm_compat_ioctl,
2277 #endif
2278         .mmap           = kvm_vm_mmap,
2279         .llseek         = noop_llseek,
2280 };
2281
2282 static int kvm_dev_ioctl_create_vm(unsigned long type)
2283 {
2284         int r;
2285         struct kvm *kvm;
2286
2287         kvm = kvm_create_vm(type);
2288         if (IS_ERR(kvm))
2289                 return PTR_ERR(kvm);
2290 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2291         r = kvm_coalesced_mmio_init(kvm);
2292         if (r < 0) {
2293                 kvm_put_kvm(kvm);
2294                 return r;
2295         }
2296 #endif
2297         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2298         if (r < 0)
2299                 kvm_put_kvm(kvm);
2300
2301         return r;
2302 }
2303
2304 static long kvm_dev_ioctl_check_extension_generic(long arg)
2305 {
2306         switch (arg) {
2307         case KVM_CAP_USER_MEMORY:
2308         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2309         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2310 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2311         case KVM_CAP_SET_BOOT_CPU_ID:
2312 #endif
2313         case KVM_CAP_INTERNAL_ERROR_DATA:
2314 #ifdef CONFIG_HAVE_KVM_MSI
2315         case KVM_CAP_SIGNAL_MSI:
2316 #endif
2317                 return 1;
2318 #ifdef KVM_CAP_IRQ_ROUTING
2319         case KVM_CAP_IRQ_ROUTING:
2320                 return KVM_MAX_IRQ_ROUTES;
2321 #endif
2322         default:
2323                 break;
2324         }
2325         return kvm_dev_ioctl_check_extension(arg);
2326 }
2327
2328 static long kvm_dev_ioctl(struct file *filp,
2329                           unsigned int ioctl, unsigned long arg)
2330 {
2331         long r = -EINVAL;
2332
2333         switch (ioctl) {
2334         case KVM_GET_API_VERSION:
2335                 r = -EINVAL;
2336                 if (arg)
2337                         goto out;
2338                 r = KVM_API_VERSION;
2339                 break;
2340         case KVM_CREATE_VM:
2341                 r = kvm_dev_ioctl_create_vm(arg);
2342                 break;
2343         case KVM_CHECK_EXTENSION:
2344                 r = kvm_dev_ioctl_check_extension_generic(arg);
2345                 break;
2346         case KVM_GET_VCPU_MMAP_SIZE:
2347                 r = -EINVAL;
2348                 if (arg)
2349                         goto out;
2350                 r = PAGE_SIZE;     /* struct kvm_run */
2351 #ifdef CONFIG_X86
2352                 r += PAGE_SIZE;    /* pio data page */
2353 #endif
2354 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2355                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2356 #endif
2357                 break;
2358         case KVM_TRACE_ENABLE:
2359         case KVM_TRACE_PAUSE:
2360         case KVM_TRACE_DISABLE:
2361                 r = -EOPNOTSUPP;
2362                 break;
2363         default:
2364                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2365         }
2366 out:
2367         return r;
2368 }
2369
2370 static struct file_operations kvm_chardev_ops = {
2371         .unlocked_ioctl = kvm_dev_ioctl,
2372         .compat_ioctl   = kvm_dev_ioctl,
2373         .llseek         = noop_llseek,
2374 };
2375
2376 static struct miscdevice kvm_dev = {
2377         KVM_MINOR,
2378         "kvm",
2379         &kvm_chardev_ops,
2380 };
2381
2382 static void hardware_enable_nolock(void *junk)
2383 {
2384         int cpu = raw_smp_processor_id();
2385         int r;
2386
2387         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2388                 return;
2389
2390         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2391
2392         r = kvm_arch_hardware_enable(NULL);
2393
2394         if (r) {
2395                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2396                 atomic_inc(&hardware_enable_failed);
2397                 printk(KERN_INFO "kvm: enabling virtualization on "
2398                                  "CPU%d failed\n", cpu);
2399         }
2400 }
2401
2402 static void hardware_enable(void *junk)
2403 {
2404         raw_spin_lock(&kvm_lock);
2405         hardware_enable_nolock(junk);
2406         raw_spin_unlock(&kvm_lock);
2407 }
2408
2409 static void hardware_disable_nolock(void *junk)
2410 {
2411         int cpu = raw_smp_processor_id();
2412
2413         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2414                 return;
2415         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2416         kvm_arch_hardware_disable(NULL);
2417 }
2418
2419 static void hardware_disable(void *junk)
2420 {
2421         raw_spin_lock(&kvm_lock);
2422         hardware_disable_nolock(junk);
2423         raw_spin_unlock(&kvm_lock);
2424 }
2425
2426 static void hardware_disable_all_nolock(void)
2427 {
2428         BUG_ON(!kvm_usage_count);
2429
2430         kvm_usage_count--;
2431         if (!kvm_usage_count)
2432                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2433 }
2434
2435 static void hardware_disable_all(void)
2436 {
2437         raw_spin_lock(&kvm_lock);
2438         hardware_disable_all_nolock();
2439         raw_spin_unlock(&kvm_lock);
2440 }
2441
2442 static int hardware_enable_all(void)
2443 {
2444         int r = 0;
2445
2446         raw_spin_lock(&kvm_lock);
2447
2448         kvm_usage_count++;
2449         if (kvm_usage_count == 1) {
2450                 atomic_set(&hardware_enable_failed, 0);
2451                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2452
2453                 if (atomic_read(&hardware_enable_failed)) {
2454                         hardware_disable_all_nolock();
2455                         r = -EBUSY;
2456                 }
2457         }
2458
2459         raw_spin_unlock(&kvm_lock);
2460
2461         return r;
2462 }
2463
2464 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2465                            void *v)
2466 {
2467         int cpu = (long)v;
2468
2469         if (!kvm_usage_count)
2470                 return NOTIFY_OK;
2471
2472         val &= ~CPU_TASKS_FROZEN;
2473         switch (val) {
2474         case CPU_DYING:
2475                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2476                        cpu);
2477                 hardware_disable(NULL);
2478                 break;
2479         case CPU_STARTING:
2480                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2481                        cpu);
2482                 hardware_enable(NULL);
2483                 break;
2484         }
2485         return NOTIFY_OK;
2486 }
2487
2488
2489 asmlinkage void kvm_spurious_fault(void)
2490 {
2491         /* Fault while not rebooting.  We want the trace. */
2492         BUG();
2493 }
2494 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
2495
2496 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2497                       void *v)
2498 {
2499         /*
2500          * Some (well, at least mine) BIOSes hang on reboot if
2501          * in vmx root mode.
2502          *
2503          * And Intel TXT required VMX off for all cpu when system shutdown.
2504          */
2505         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2506         kvm_rebooting = true;
2507         on_each_cpu(hardware_disable_nolock, NULL, 1);
2508         return NOTIFY_OK;
2509 }
2510
2511 static struct notifier_block kvm_reboot_notifier = {
2512         .notifier_call = kvm_reboot,
2513         .priority = 0,
2514 };
2515
2516 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2517 {
2518         int i;
2519
2520         for (i = 0; i < bus->dev_count; i++) {
2521                 struct kvm_io_device *pos = bus->range[i].dev;
2522
2523                 kvm_iodevice_destructor(pos);
2524         }
2525         kfree(bus);
2526 }
2527
2528 int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2529 {
2530         const struct kvm_io_range *r1 = p1;
2531         const struct kvm_io_range *r2 = p2;
2532
2533         if (r1->addr < r2->addr)
2534                 return -1;
2535         if (r1->addr + r1->len > r2->addr + r2->len)
2536                 return 1;
2537         return 0;
2538 }
2539
2540 int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2541                           gpa_t addr, int len)
2542 {
2543         bus->range[bus->dev_count++] = (struct kvm_io_range) {
2544                 .addr = addr,
2545                 .len = len,
2546                 .dev = dev,
2547         };
2548
2549         sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2550                 kvm_io_bus_sort_cmp, NULL);
2551
2552         return 0;
2553 }
2554
2555 int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2556                              gpa_t addr, int len)
2557 {
2558         struct kvm_io_range *range, key;
2559         int off;
2560
2561         key = (struct kvm_io_range) {
2562                 .addr = addr,
2563                 .len = len,
2564         };
2565
2566         range = bsearch(&key, bus->range, bus->dev_count,
2567                         sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2568         if (range == NULL)
2569                 return -ENOENT;
2570
2571         off = range - bus->range;
2572
2573         while (off > 0 && kvm_io_bus_sort_cmp(&key, &bus->range[off-1]) == 0)
2574                 off--;
2575
2576         return off;
2577 }
2578
2579 /* kvm_io_bus_write - called under kvm->slots_lock */
2580 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2581                      int len, const void *val)
2582 {
2583         int idx;
2584         struct kvm_io_bus *bus;
2585         struct kvm_io_range range;
2586
2587         range = (struct kvm_io_range) {
2588                 .addr = addr,
2589                 .len = len,
2590         };
2591
2592         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2593         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2594         if (idx < 0)
2595                 return -EOPNOTSUPP;
2596
2597         while (idx < bus->dev_count &&
2598                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2599                 if (!kvm_iodevice_write(bus->range[idx].dev, addr, len, val))
2600                         return 0;
2601                 idx++;
2602         }
2603
2604         return -EOPNOTSUPP;
2605 }
2606
2607 /* kvm_io_bus_read - called under kvm->slots_lock */
2608 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2609                     int len, void *val)
2610 {
2611         int idx;
2612         struct kvm_io_bus *bus;
2613         struct kvm_io_range range;
2614
2615         range = (struct kvm_io_range) {
2616                 .addr = addr,
2617                 .len = len,
2618         };
2619
2620         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2621         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2622         if (idx < 0)
2623                 return -EOPNOTSUPP;
2624
2625         while (idx < bus->dev_count &&
2626                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2627                 if (!kvm_iodevice_read(bus->range[idx].dev, addr, len, val))
2628                         return 0;
2629                 idx++;
2630         }
2631
2632         return -EOPNOTSUPP;
2633 }
2634
2635 /* Caller must hold slots_lock. */
2636 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2637                             int len, struct kvm_io_device *dev)
2638 {
2639         struct kvm_io_bus *new_bus, *bus;
2640
2641         bus = kvm->buses[bus_idx];
2642         if (bus->dev_count > NR_IOBUS_DEVS - 1)
2643                 return -ENOSPC;
2644
2645         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
2646                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2647         if (!new_bus)
2648                 return -ENOMEM;
2649         memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
2650                sizeof(struct kvm_io_range)));
2651         kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2652         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2653         synchronize_srcu_expedited(&kvm->srcu);
2654         kfree(bus);
2655
2656         return 0;
2657 }
2658
2659 /* Caller must hold slots_lock. */
2660 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2661                               struct kvm_io_device *dev)
2662 {
2663         int i, r;
2664         struct kvm_io_bus *new_bus, *bus;
2665
2666         bus = kvm->buses[bus_idx];
2667         r = -ENOENT;
2668         for (i = 0; i < bus->dev_count; i++)
2669                 if (bus->range[i].dev == dev) {
2670                         r = 0;
2671                         break;
2672                 }
2673
2674         if (r)
2675                 return r;
2676
2677         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
2678                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2679         if (!new_bus)
2680                 return -ENOMEM;
2681
2682         memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
2683         new_bus->dev_count--;
2684         memcpy(new_bus->range + i, bus->range + i + 1,
2685                (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
2686
2687         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2688         synchronize_srcu_expedited(&kvm->srcu);
2689         kfree(bus);
2690         return r;
2691 }
2692
2693 static struct notifier_block kvm_cpu_notifier = {
2694         .notifier_call = kvm_cpu_hotplug,
2695 };
2696
2697 static int vm_stat_get(void *_offset, u64 *val)
2698 {
2699         unsigned offset = (long)_offset;
2700         struct kvm *kvm;
2701
2702         *val = 0;
2703         raw_spin_lock(&kvm_lock);
2704         list_for_each_entry(kvm, &vm_list, vm_list)
2705                 *val += *(u32 *)((void *)kvm + offset);
2706         raw_spin_unlock(&kvm_lock);
2707         return 0;
2708 }
2709
2710 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2711
2712 static int vcpu_stat_get(void *_offset, u64 *val)
2713 {
2714         unsigned offset = (long)_offset;
2715         struct kvm *kvm;
2716         struct kvm_vcpu *vcpu;
2717         int i;
2718
2719         *val = 0;
2720         raw_spin_lock(&kvm_lock);
2721         list_for_each_entry(kvm, &vm_list, vm_list)
2722                 kvm_for_each_vcpu(i, vcpu, kvm)
2723                         *val += *(u32 *)((void *)vcpu + offset);
2724
2725         raw_spin_unlock(&kvm_lock);
2726         return 0;
2727 }
2728
2729 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2730
2731 static const struct file_operations *stat_fops[] = {
2732         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2733         [KVM_STAT_VM]   = &vm_stat_fops,
2734 };
2735
2736 static int kvm_init_debug(void)
2737 {
2738         int r = -EFAULT;
2739         struct kvm_stats_debugfs_item *p;
2740
2741         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2742         if (kvm_debugfs_dir == NULL)
2743                 goto out;
2744
2745         for (p = debugfs_entries; p->name; ++p) {
2746                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2747                                                 (void *)(long)p->offset,
2748                                                 stat_fops[p->kind]);
2749                 if (p->dentry == NULL)
2750                         goto out_dir;
2751         }
2752
2753         return 0;
2754
2755 out_dir:
2756         debugfs_remove_recursive(kvm_debugfs_dir);
2757 out:
2758         return r;
2759 }
2760
2761 static void kvm_exit_debug(void)
2762 {
2763         struct kvm_stats_debugfs_item *p;
2764
2765         for (p = debugfs_entries; p->name; ++p)
2766                 debugfs_remove(p->dentry);
2767         debugfs_remove(kvm_debugfs_dir);
2768 }
2769
2770 static int kvm_suspend(void)
2771 {
2772         if (kvm_usage_count)
2773                 hardware_disable_nolock(NULL);
2774         return 0;
2775 }
2776
2777 static void kvm_resume(void)
2778 {
2779         if (kvm_usage_count) {
2780                 WARN_ON(raw_spin_is_locked(&kvm_lock));
2781                 hardware_enable_nolock(NULL);
2782         }
2783 }
2784
2785 static struct syscore_ops kvm_syscore_ops = {
2786         .suspend = kvm_suspend,
2787         .resume = kvm_resume,
2788 };
2789
2790 static inline
2791 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2792 {
2793         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2794 }
2795
2796 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2797 {
2798         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2799
2800         kvm_arch_vcpu_load(vcpu, cpu);
2801 }
2802
2803 static void kvm_sched_out(struct preempt_notifier *pn,
2804                           struct task_struct *next)
2805 {
2806         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2807
2808         kvm_arch_vcpu_put(vcpu);
2809 }
2810
2811 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2812                   struct module *module)
2813 {
2814         int r;
2815         int cpu;
2816
2817         r = kvm_arch_init(opaque);
2818         if (r)
2819                 goto out_fail;
2820
2821         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2822                 r = -ENOMEM;
2823                 goto out_free_0;
2824         }
2825
2826         r = kvm_arch_hardware_setup();
2827         if (r < 0)
2828                 goto out_free_0a;
2829
2830         for_each_online_cpu(cpu) {
2831                 smp_call_function_single(cpu,
2832                                 kvm_arch_check_processor_compat,
2833                                 &r, 1);
2834                 if (r < 0)
2835                         goto out_free_1;
2836         }
2837
2838         r = register_cpu_notifier(&kvm_cpu_notifier);
2839         if (r)
2840                 goto out_free_2;
2841         register_reboot_notifier(&kvm_reboot_notifier);
2842
2843         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2844         if (!vcpu_align)
2845                 vcpu_align = __alignof__(struct kvm_vcpu);
2846         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2847                                            0, NULL);
2848         if (!kvm_vcpu_cache) {
2849                 r = -ENOMEM;
2850                 goto out_free_3;
2851         }
2852
2853         r = kvm_async_pf_init();
2854         if (r)
2855                 goto out_free;
2856
2857         kvm_chardev_ops.owner = module;
2858         kvm_vm_fops.owner = module;
2859         kvm_vcpu_fops.owner = module;
2860
2861         r = misc_register(&kvm_dev);
2862         if (r) {
2863                 printk(KERN_ERR "kvm: misc device register failed\n");
2864                 goto out_unreg;
2865         }
2866
2867         register_syscore_ops(&kvm_syscore_ops);
2868
2869         kvm_preempt_ops.sched_in = kvm_sched_in;
2870         kvm_preempt_ops.sched_out = kvm_sched_out;
2871
2872         r = kvm_init_debug();
2873         if (r) {
2874                 printk(KERN_ERR "kvm: create debugfs files failed\n");
2875                 goto out_undebugfs;
2876         }
2877
2878         return 0;
2879
2880 out_undebugfs:
2881         unregister_syscore_ops(&kvm_syscore_ops);
2882 out_unreg:
2883         kvm_async_pf_deinit();
2884 out_free:
2885         kmem_cache_destroy(kvm_vcpu_cache);
2886 out_free_3:
2887         unregister_reboot_notifier(&kvm_reboot_notifier);
2888         unregister_cpu_notifier(&kvm_cpu_notifier);
2889 out_free_2:
2890 out_free_1:
2891         kvm_arch_hardware_unsetup();
2892 out_free_0a:
2893         free_cpumask_var(cpus_hardware_enabled);
2894 out_free_0:
2895         kvm_arch_exit();
2896 out_fail:
2897         return r;
2898 }
2899 EXPORT_SYMBOL_GPL(kvm_init);
2900
2901 void kvm_exit(void)
2902 {
2903         kvm_exit_debug();
2904         misc_deregister(&kvm_dev);
2905         kmem_cache_destroy(kvm_vcpu_cache);
2906         kvm_async_pf_deinit();
2907         unregister_syscore_ops(&kvm_syscore_ops);
2908         unregister_reboot_notifier(&kvm_reboot_notifier);
2909         unregister_cpu_notifier(&kvm_cpu_notifier);
2910         on_each_cpu(hardware_disable_nolock, NULL, 1);
2911         kvm_arch_hardware_unsetup();
2912         kvm_arch_exit();
2913         free_cpumask_var(cpus_hardware_enabled);
2914 }
2915 EXPORT_SYMBOL_GPL(kvm_exit);