Merge tag 'kvm-3.7-2' of git://git.kernel.org/pub/scm/virt/kvm/kvm
[~shefty/rdma-dev.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 int vcpu_load(struct kvm_vcpu *vcpu)
135 {
136         int cpu;
137
138         if (mutex_lock_killable(&vcpu->mutex))
139                 return -EINTR;
140         if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
141                 /* The thread running this VCPU changed. */
142                 struct pid *oldpid = vcpu->pid;
143                 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
144                 rcu_assign_pointer(vcpu->pid, newpid);
145                 synchronize_rcu();
146                 put_pid(oldpid);
147         }
148         cpu = get_cpu();
149         preempt_notifier_register(&vcpu->preempt_notifier);
150         kvm_arch_vcpu_load(vcpu, cpu);
151         put_cpu();
152         return 0;
153 }
154
155 void vcpu_put(struct kvm_vcpu *vcpu)
156 {
157         preempt_disable();
158         kvm_arch_vcpu_put(vcpu);
159         preempt_notifier_unregister(&vcpu->preempt_notifier);
160         preempt_enable();
161         mutex_unlock(&vcpu->mutex);
162 }
163
164 static void ack_flush(void *_completed)
165 {
166 }
167
168 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
169 {
170         int i, cpu, me;
171         cpumask_var_t cpus;
172         bool called = true;
173         struct kvm_vcpu *vcpu;
174
175         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
176
177         me = get_cpu();
178         kvm_for_each_vcpu(i, vcpu, kvm) {
179                 kvm_make_request(req, vcpu);
180                 cpu = vcpu->cpu;
181
182                 /* Set ->requests bit before we read ->mode */
183                 smp_mb();
184
185                 if (cpus != NULL && cpu != -1 && cpu != me &&
186                       kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
187                         cpumask_set_cpu(cpu, cpus);
188         }
189         if (unlikely(cpus == NULL))
190                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
191         else if (!cpumask_empty(cpus))
192                 smp_call_function_many(cpus, ack_flush, NULL, 1);
193         else
194                 called = false;
195         put_cpu();
196         free_cpumask_var(cpus);
197         return called;
198 }
199
200 void kvm_flush_remote_tlbs(struct kvm *kvm)
201 {
202         long dirty_count = kvm->tlbs_dirty;
203
204         smp_mb();
205         if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
206                 ++kvm->stat.remote_tlb_flush;
207         cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
208 }
209
210 void kvm_reload_remote_mmus(struct kvm *kvm)
211 {
212         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
213 }
214
215 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
216 {
217         struct page *page;
218         int r;
219
220         mutex_init(&vcpu->mutex);
221         vcpu->cpu = -1;
222         vcpu->kvm = kvm;
223         vcpu->vcpu_id = id;
224         vcpu->pid = NULL;
225         init_waitqueue_head(&vcpu->wq);
226         kvm_async_pf_vcpu_init(vcpu);
227
228         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
229         if (!page) {
230                 r = -ENOMEM;
231                 goto fail;
232         }
233         vcpu->run = page_address(page);
234
235         kvm_vcpu_set_in_spin_loop(vcpu, false);
236         kvm_vcpu_set_dy_eligible(vcpu, false);
237
238         r = kvm_arch_vcpu_init(vcpu);
239         if (r < 0)
240                 goto fail_free_run;
241         return 0;
242
243 fail_free_run:
244         free_page((unsigned long)vcpu->run);
245 fail:
246         return r;
247 }
248 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
249
250 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
251 {
252         put_pid(vcpu->pid);
253         kvm_arch_vcpu_uninit(vcpu);
254         free_page((unsigned long)vcpu->run);
255 }
256 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
257
258 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
259 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
260 {
261         return container_of(mn, struct kvm, mmu_notifier);
262 }
263
264 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
265                                              struct mm_struct *mm,
266                                              unsigned long address)
267 {
268         struct kvm *kvm = mmu_notifier_to_kvm(mn);
269         int need_tlb_flush, idx;
270
271         /*
272          * When ->invalidate_page runs, the linux pte has been zapped
273          * already but the page is still allocated until
274          * ->invalidate_page returns. So if we increase the sequence
275          * here the kvm page fault will notice if the spte can't be
276          * established because the page is going to be freed. If
277          * instead the kvm page fault establishes the spte before
278          * ->invalidate_page runs, kvm_unmap_hva will release it
279          * before returning.
280          *
281          * The sequence increase only need to be seen at spin_unlock
282          * time, and not at spin_lock time.
283          *
284          * Increasing the sequence after the spin_unlock would be
285          * unsafe because the kvm page fault could then establish the
286          * pte after kvm_unmap_hva returned, without noticing the page
287          * is going to be freed.
288          */
289         idx = srcu_read_lock(&kvm->srcu);
290         spin_lock(&kvm->mmu_lock);
291
292         kvm->mmu_notifier_seq++;
293         need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
294         /* we've to flush the tlb before the pages can be freed */
295         if (need_tlb_flush)
296                 kvm_flush_remote_tlbs(kvm);
297
298         spin_unlock(&kvm->mmu_lock);
299         srcu_read_unlock(&kvm->srcu, idx);
300 }
301
302 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
303                                         struct mm_struct *mm,
304                                         unsigned long address,
305                                         pte_t pte)
306 {
307         struct kvm *kvm = mmu_notifier_to_kvm(mn);
308         int idx;
309
310         idx = srcu_read_lock(&kvm->srcu);
311         spin_lock(&kvm->mmu_lock);
312         kvm->mmu_notifier_seq++;
313         kvm_set_spte_hva(kvm, address, pte);
314         spin_unlock(&kvm->mmu_lock);
315         srcu_read_unlock(&kvm->srcu, idx);
316 }
317
318 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
319                                                     struct mm_struct *mm,
320                                                     unsigned long start,
321                                                     unsigned long end)
322 {
323         struct kvm *kvm = mmu_notifier_to_kvm(mn);
324         int need_tlb_flush = 0, idx;
325
326         idx = srcu_read_lock(&kvm->srcu);
327         spin_lock(&kvm->mmu_lock);
328         /*
329          * The count increase must become visible at unlock time as no
330          * spte can be established without taking the mmu_lock and
331          * count is also read inside the mmu_lock critical section.
332          */
333         kvm->mmu_notifier_count++;
334         need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
335         need_tlb_flush |= kvm->tlbs_dirty;
336         /* we've to flush the tlb before the pages can be freed */
337         if (need_tlb_flush)
338                 kvm_flush_remote_tlbs(kvm);
339
340         spin_unlock(&kvm->mmu_lock);
341         srcu_read_unlock(&kvm->srcu, idx);
342 }
343
344 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
345                                                   struct mm_struct *mm,
346                                                   unsigned long start,
347                                                   unsigned long end)
348 {
349         struct kvm *kvm = mmu_notifier_to_kvm(mn);
350
351         spin_lock(&kvm->mmu_lock);
352         /*
353          * This sequence increase will notify the kvm page fault that
354          * the page that is going to be mapped in the spte could have
355          * been freed.
356          */
357         kvm->mmu_notifier_seq++;
358         smp_wmb();
359         /*
360          * The above sequence increase must be visible before the
361          * below count decrease, which is ensured by the smp_wmb above
362          * in conjunction with the smp_rmb in mmu_notifier_retry().
363          */
364         kvm->mmu_notifier_count--;
365         spin_unlock(&kvm->mmu_lock);
366
367         BUG_ON(kvm->mmu_notifier_count < 0);
368 }
369
370 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
371                                               struct mm_struct *mm,
372                                               unsigned long address)
373 {
374         struct kvm *kvm = mmu_notifier_to_kvm(mn);
375         int young, idx;
376
377         idx = srcu_read_lock(&kvm->srcu);
378         spin_lock(&kvm->mmu_lock);
379
380         young = kvm_age_hva(kvm, address);
381         if (young)
382                 kvm_flush_remote_tlbs(kvm);
383
384         spin_unlock(&kvm->mmu_lock);
385         srcu_read_unlock(&kvm->srcu, idx);
386
387         return young;
388 }
389
390 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
391                                        struct mm_struct *mm,
392                                        unsigned long address)
393 {
394         struct kvm *kvm = mmu_notifier_to_kvm(mn);
395         int young, idx;
396
397         idx = srcu_read_lock(&kvm->srcu);
398         spin_lock(&kvm->mmu_lock);
399         young = kvm_test_age_hva(kvm, address);
400         spin_unlock(&kvm->mmu_lock);
401         srcu_read_unlock(&kvm->srcu, idx);
402
403         return young;
404 }
405
406 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
407                                      struct mm_struct *mm)
408 {
409         struct kvm *kvm = mmu_notifier_to_kvm(mn);
410         int idx;
411
412         idx = srcu_read_lock(&kvm->srcu);
413         kvm_arch_flush_shadow_all(kvm);
414         srcu_read_unlock(&kvm->srcu, idx);
415 }
416
417 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
418         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
419         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
420         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
421         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
422         .test_young             = kvm_mmu_notifier_test_young,
423         .change_pte             = kvm_mmu_notifier_change_pte,
424         .release                = kvm_mmu_notifier_release,
425 };
426
427 static int kvm_init_mmu_notifier(struct kvm *kvm)
428 {
429         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
430         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
431 }
432
433 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
434
435 static int kvm_init_mmu_notifier(struct kvm *kvm)
436 {
437         return 0;
438 }
439
440 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
441
442 static void kvm_init_memslots_id(struct kvm *kvm)
443 {
444         int i;
445         struct kvm_memslots *slots = kvm->memslots;
446
447         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
448                 slots->id_to_index[i] = slots->memslots[i].id = i;
449 }
450
451 static struct kvm *kvm_create_vm(unsigned long type)
452 {
453         int r, i;
454         struct kvm *kvm = kvm_arch_alloc_vm();
455
456         if (!kvm)
457                 return ERR_PTR(-ENOMEM);
458
459         r = kvm_arch_init_vm(kvm, type);
460         if (r)
461                 goto out_err_nodisable;
462
463         r = hardware_enable_all();
464         if (r)
465                 goto out_err_nodisable;
466
467 #ifdef CONFIG_HAVE_KVM_IRQCHIP
468         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
469         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
470 #endif
471
472         r = -ENOMEM;
473         kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
474         if (!kvm->memslots)
475                 goto out_err_nosrcu;
476         kvm_init_memslots_id(kvm);
477         if (init_srcu_struct(&kvm->srcu))
478                 goto out_err_nosrcu;
479         for (i = 0; i < KVM_NR_BUSES; i++) {
480                 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
481                                         GFP_KERNEL);
482                 if (!kvm->buses[i])
483                         goto out_err;
484         }
485
486         spin_lock_init(&kvm->mmu_lock);
487         kvm->mm = current->mm;
488         atomic_inc(&kvm->mm->mm_count);
489         kvm_eventfd_init(kvm);
490         mutex_init(&kvm->lock);
491         mutex_init(&kvm->irq_lock);
492         mutex_init(&kvm->slots_lock);
493         atomic_set(&kvm->users_count, 1);
494
495         r = kvm_init_mmu_notifier(kvm);
496         if (r)
497                 goto out_err;
498
499         raw_spin_lock(&kvm_lock);
500         list_add(&kvm->vm_list, &vm_list);
501         raw_spin_unlock(&kvm_lock);
502
503         return kvm;
504
505 out_err:
506         cleanup_srcu_struct(&kvm->srcu);
507 out_err_nosrcu:
508         hardware_disable_all();
509 out_err_nodisable:
510         for (i = 0; i < KVM_NR_BUSES; i++)
511                 kfree(kvm->buses[i]);
512         kfree(kvm->memslots);
513         kvm_arch_free_vm(kvm);
514         return ERR_PTR(r);
515 }
516
517 /*
518  * Avoid using vmalloc for a small buffer.
519  * Should not be used when the size is statically known.
520  */
521 void *kvm_kvzalloc(unsigned long size)
522 {
523         if (size > PAGE_SIZE)
524                 return vzalloc(size);
525         else
526                 return kzalloc(size, GFP_KERNEL);
527 }
528
529 void kvm_kvfree(const void *addr)
530 {
531         if (is_vmalloc_addr(addr))
532                 vfree(addr);
533         else
534                 kfree(addr);
535 }
536
537 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
538 {
539         if (!memslot->dirty_bitmap)
540                 return;
541
542         kvm_kvfree(memslot->dirty_bitmap);
543         memslot->dirty_bitmap = NULL;
544 }
545
546 /*
547  * Free any memory in @free but not in @dont.
548  */
549 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
550                                   struct kvm_memory_slot *dont)
551 {
552         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
553                 kvm_destroy_dirty_bitmap(free);
554
555         kvm_arch_free_memslot(free, dont);
556
557         free->npages = 0;
558 }
559
560 void kvm_free_physmem(struct kvm *kvm)
561 {
562         struct kvm_memslots *slots = kvm->memslots;
563         struct kvm_memory_slot *memslot;
564
565         kvm_for_each_memslot(memslot, slots)
566                 kvm_free_physmem_slot(memslot, NULL);
567
568         kfree(kvm->memslots);
569 }
570
571 static void kvm_destroy_vm(struct kvm *kvm)
572 {
573         int i;
574         struct mm_struct *mm = kvm->mm;
575
576         kvm_arch_sync_events(kvm);
577         raw_spin_lock(&kvm_lock);
578         list_del(&kvm->vm_list);
579         raw_spin_unlock(&kvm_lock);
580         kvm_free_irq_routing(kvm);
581         for (i = 0; i < KVM_NR_BUSES; i++)
582                 kvm_io_bus_destroy(kvm->buses[i]);
583         kvm_coalesced_mmio_free(kvm);
584 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
585         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
586 #else
587         kvm_arch_flush_shadow_all(kvm);
588 #endif
589         kvm_arch_destroy_vm(kvm);
590         kvm_free_physmem(kvm);
591         cleanup_srcu_struct(&kvm->srcu);
592         kvm_arch_free_vm(kvm);
593         hardware_disable_all();
594         mmdrop(mm);
595 }
596
597 void kvm_get_kvm(struct kvm *kvm)
598 {
599         atomic_inc(&kvm->users_count);
600 }
601 EXPORT_SYMBOL_GPL(kvm_get_kvm);
602
603 void kvm_put_kvm(struct kvm *kvm)
604 {
605         if (atomic_dec_and_test(&kvm->users_count))
606                 kvm_destroy_vm(kvm);
607 }
608 EXPORT_SYMBOL_GPL(kvm_put_kvm);
609
610
611 static int kvm_vm_release(struct inode *inode, struct file *filp)
612 {
613         struct kvm *kvm = filp->private_data;
614
615         kvm_irqfd_release(kvm);
616
617         kvm_put_kvm(kvm);
618         return 0;
619 }
620
621 /*
622  * Allocation size is twice as large as the actual dirty bitmap size.
623  * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
624  */
625 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
626 {
627 #ifndef CONFIG_S390
628         unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
629
630         memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
631         if (!memslot->dirty_bitmap)
632                 return -ENOMEM;
633
634 #endif /* !CONFIG_S390 */
635         return 0;
636 }
637
638 static int cmp_memslot(const void *slot1, const void *slot2)
639 {
640         struct kvm_memory_slot *s1, *s2;
641
642         s1 = (struct kvm_memory_slot *)slot1;
643         s2 = (struct kvm_memory_slot *)slot2;
644
645         if (s1->npages < s2->npages)
646                 return 1;
647         if (s1->npages > s2->npages)
648                 return -1;
649
650         return 0;
651 }
652
653 /*
654  * Sort the memslots base on its size, so the larger slots
655  * will get better fit.
656  */
657 static void sort_memslots(struct kvm_memslots *slots)
658 {
659         int i;
660
661         sort(slots->memslots, KVM_MEM_SLOTS_NUM,
662               sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
663
664         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
665                 slots->id_to_index[slots->memslots[i].id] = i;
666 }
667
668 void update_memslots(struct kvm_memslots *slots, struct kvm_memory_slot *new)
669 {
670         if (new) {
671                 int id = new->id;
672                 struct kvm_memory_slot *old = id_to_memslot(slots, id);
673                 unsigned long npages = old->npages;
674
675                 *old = *new;
676                 if (new->npages != npages)
677                         sort_memslots(slots);
678         }
679
680         slots->generation++;
681 }
682
683 static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
684 {
685         u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
686
687 #ifdef KVM_CAP_READONLY_MEM
688         valid_flags |= KVM_MEM_READONLY;
689 #endif
690
691         if (mem->flags & ~valid_flags)
692                 return -EINVAL;
693
694         return 0;
695 }
696
697 /*
698  * Allocate some memory and give it an address in the guest physical address
699  * space.
700  *
701  * Discontiguous memory is allowed, mostly for framebuffers.
702  *
703  * Must be called holding mmap_sem for write.
704  */
705 int __kvm_set_memory_region(struct kvm *kvm,
706                             struct kvm_userspace_memory_region *mem,
707                             int user_alloc)
708 {
709         int r;
710         gfn_t base_gfn;
711         unsigned long npages;
712         unsigned long i;
713         struct kvm_memory_slot *memslot;
714         struct kvm_memory_slot old, new;
715         struct kvm_memslots *slots, *old_memslots;
716
717         r = check_memory_region_flags(mem);
718         if (r)
719                 goto out;
720
721         r = -EINVAL;
722         /* General sanity checks */
723         if (mem->memory_size & (PAGE_SIZE - 1))
724                 goto out;
725         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
726                 goto out;
727         /* We can read the guest memory with __xxx_user() later on. */
728         if (user_alloc &&
729             ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
730              !access_ok(VERIFY_WRITE,
731                         (void __user *)(unsigned long)mem->userspace_addr,
732                         mem->memory_size)))
733                 goto out;
734         if (mem->slot >= KVM_MEM_SLOTS_NUM)
735                 goto out;
736         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
737                 goto out;
738
739         memslot = id_to_memslot(kvm->memslots, mem->slot);
740         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
741         npages = mem->memory_size >> PAGE_SHIFT;
742
743         r = -EINVAL;
744         if (npages > KVM_MEM_MAX_NR_PAGES)
745                 goto out;
746
747         if (!npages)
748                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
749
750         new = old = *memslot;
751
752         new.id = mem->slot;
753         new.base_gfn = base_gfn;
754         new.npages = npages;
755         new.flags = mem->flags;
756
757         /* Disallow changing a memory slot's size. */
758         r = -EINVAL;
759         if (npages && old.npages && npages != old.npages)
760                 goto out_free;
761
762         /* Check for overlaps */
763         r = -EEXIST;
764         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
765                 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
766
767                 if (s == memslot || !s->npages)
768                         continue;
769                 if (!((base_gfn + npages <= s->base_gfn) ||
770                       (base_gfn >= s->base_gfn + s->npages)))
771                         goto out_free;
772         }
773
774         /* Free page dirty bitmap if unneeded */
775         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
776                 new.dirty_bitmap = NULL;
777
778         r = -ENOMEM;
779
780         /* Allocate if a slot is being created */
781         if (npages && !old.npages) {
782                 new.user_alloc = user_alloc;
783                 new.userspace_addr = mem->userspace_addr;
784
785                 if (kvm_arch_create_memslot(&new, npages))
786                         goto out_free;
787         }
788
789         /* Allocate page dirty bitmap if needed */
790         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
791                 if (kvm_create_dirty_bitmap(&new) < 0)
792                         goto out_free;
793                 /* destroy any largepage mappings for dirty tracking */
794         }
795
796         if (!npages || base_gfn != old.base_gfn) {
797                 struct kvm_memory_slot *slot;
798
799                 r = -ENOMEM;
800                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
801                                 GFP_KERNEL);
802                 if (!slots)
803                         goto out_free;
804                 slot = id_to_memslot(slots, mem->slot);
805                 slot->flags |= KVM_MEMSLOT_INVALID;
806
807                 update_memslots(slots, NULL);
808
809                 old_memslots = kvm->memslots;
810                 rcu_assign_pointer(kvm->memslots, slots);
811                 synchronize_srcu_expedited(&kvm->srcu);
812                 /* From this point no new shadow pages pointing to a deleted,
813                  * or moved, memslot will be created.
814                  *
815                  * validation of sp->gfn happens in:
816                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
817                  *      - kvm_is_visible_gfn (mmu_check_roots)
818                  */
819                 kvm_arch_flush_shadow_memslot(kvm, slot);
820                 kfree(old_memslots);
821         }
822
823         r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
824         if (r)
825                 goto out_free;
826
827         /* map/unmap the pages in iommu page table */
828         if (npages) {
829                 r = kvm_iommu_map_pages(kvm, &new);
830                 if (r)
831                         goto out_free;
832         } else
833                 kvm_iommu_unmap_pages(kvm, &old);
834
835         r = -ENOMEM;
836         slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
837                         GFP_KERNEL);
838         if (!slots)
839                 goto out_free;
840
841         /* actual memory is freed via old in kvm_free_physmem_slot below */
842         if (!npages) {
843                 new.dirty_bitmap = NULL;
844                 memset(&new.arch, 0, sizeof(new.arch));
845         }
846
847         update_memslots(slots, &new);
848         old_memslots = kvm->memslots;
849         rcu_assign_pointer(kvm->memslots, slots);
850         synchronize_srcu_expedited(&kvm->srcu);
851
852         kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
853
854         kvm_free_physmem_slot(&old, &new);
855         kfree(old_memslots);
856
857         return 0;
858
859 out_free:
860         kvm_free_physmem_slot(&new, &old);
861 out:
862         return r;
863
864 }
865 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
866
867 int kvm_set_memory_region(struct kvm *kvm,
868                           struct kvm_userspace_memory_region *mem,
869                           int user_alloc)
870 {
871         int r;
872
873         mutex_lock(&kvm->slots_lock);
874         r = __kvm_set_memory_region(kvm, mem, user_alloc);
875         mutex_unlock(&kvm->slots_lock);
876         return r;
877 }
878 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
879
880 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
881                                    struct
882                                    kvm_userspace_memory_region *mem,
883                                    int user_alloc)
884 {
885         if (mem->slot >= KVM_MEMORY_SLOTS)
886                 return -EINVAL;
887         return kvm_set_memory_region(kvm, mem, user_alloc);
888 }
889
890 int kvm_get_dirty_log(struct kvm *kvm,
891                         struct kvm_dirty_log *log, int *is_dirty)
892 {
893         struct kvm_memory_slot *memslot;
894         int r, i;
895         unsigned long n;
896         unsigned long any = 0;
897
898         r = -EINVAL;
899         if (log->slot >= KVM_MEMORY_SLOTS)
900                 goto out;
901
902         memslot = id_to_memslot(kvm->memslots, log->slot);
903         r = -ENOENT;
904         if (!memslot->dirty_bitmap)
905                 goto out;
906
907         n = kvm_dirty_bitmap_bytes(memslot);
908
909         for (i = 0; !any && i < n/sizeof(long); ++i)
910                 any = memslot->dirty_bitmap[i];
911
912         r = -EFAULT;
913         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
914                 goto out;
915
916         if (any)
917                 *is_dirty = 1;
918
919         r = 0;
920 out:
921         return r;
922 }
923
924 bool kvm_largepages_enabled(void)
925 {
926         return largepages_enabled;
927 }
928
929 void kvm_disable_largepages(void)
930 {
931         largepages_enabled = false;
932 }
933 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
934
935 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
936 {
937         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
938 }
939 EXPORT_SYMBOL_GPL(gfn_to_memslot);
940
941 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
942 {
943         struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
944
945         if (!memslot || memslot->id >= KVM_MEMORY_SLOTS ||
946               memslot->flags & KVM_MEMSLOT_INVALID)
947                 return 0;
948
949         return 1;
950 }
951 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
952
953 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
954 {
955         struct vm_area_struct *vma;
956         unsigned long addr, size;
957
958         size = PAGE_SIZE;
959
960         addr = gfn_to_hva(kvm, gfn);
961         if (kvm_is_error_hva(addr))
962                 return PAGE_SIZE;
963
964         down_read(&current->mm->mmap_sem);
965         vma = find_vma(current->mm, addr);
966         if (!vma)
967                 goto out;
968
969         size = vma_kernel_pagesize(vma);
970
971 out:
972         up_read(&current->mm->mmap_sem);
973
974         return size;
975 }
976
977 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
978 {
979         return slot->flags & KVM_MEM_READONLY;
980 }
981
982 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
983                                        gfn_t *nr_pages, bool write)
984 {
985         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
986                 return KVM_HVA_ERR_BAD;
987
988         if (memslot_is_readonly(slot) && write)
989                 return KVM_HVA_ERR_RO_BAD;
990
991         if (nr_pages)
992                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
993
994         return __gfn_to_hva_memslot(slot, gfn);
995 }
996
997 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
998                                      gfn_t *nr_pages)
999 {
1000         return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1001 }
1002
1003 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1004                                  gfn_t gfn)
1005 {
1006         return gfn_to_hva_many(slot, gfn, NULL);
1007 }
1008 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1009
1010 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1011 {
1012         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1013 }
1014 EXPORT_SYMBOL_GPL(gfn_to_hva);
1015
1016 /*
1017  * The hva returned by this function is only allowed to be read.
1018  * It should pair with kvm_read_hva() or kvm_read_hva_atomic().
1019  */
1020 static unsigned long gfn_to_hva_read(struct kvm *kvm, gfn_t gfn)
1021 {
1022         return __gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL, false);
1023 }
1024
1025 static int kvm_read_hva(void *data, void __user *hva, int len)
1026 {
1027         return __copy_from_user(data, hva, len);
1028 }
1029
1030 static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1031 {
1032         return __copy_from_user_inatomic(data, hva, len);
1033 }
1034
1035 int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1036         unsigned long start, int write, struct page **page)
1037 {
1038         int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1039
1040         if (write)
1041                 flags |= FOLL_WRITE;
1042
1043         return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1044 }
1045
1046 static inline int check_user_page_hwpoison(unsigned long addr)
1047 {
1048         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1049
1050         rc = __get_user_pages(current, current->mm, addr, 1,
1051                               flags, NULL, NULL, NULL);
1052         return rc == -EHWPOISON;
1053 }
1054
1055 /*
1056  * The atomic path to get the writable pfn which will be stored in @pfn,
1057  * true indicates success, otherwise false is returned.
1058  */
1059 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1060                             bool write_fault, bool *writable, pfn_t *pfn)
1061 {
1062         struct page *page[1];
1063         int npages;
1064
1065         if (!(async || atomic))
1066                 return false;
1067
1068         /*
1069          * Fast pin a writable pfn only if it is a write fault request
1070          * or the caller allows to map a writable pfn for a read fault
1071          * request.
1072          */
1073         if (!(write_fault || writable))
1074                 return false;
1075
1076         npages = __get_user_pages_fast(addr, 1, 1, page);
1077         if (npages == 1) {
1078                 *pfn = page_to_pfn(page[0]);
1079
1080                 if (writable)
1081                         *writable = true;
1082                 return true;
1083         }
1084
1085         return false;
1086 }
1087
1088 /*
1089  * The slow path to get the pfn of the specified host virtual address,
1090  * 1 indicates success, -errno is returned if error is detected.
1091  */
1092 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1093                            bool *writable, pfn_t *pfn)
1094 {
1095         struct page *page[1];
1096         int npages = 0;
1097
1098         might_sleep();
1099
1100         if (writable)
1101                 *writable = write_fault;
1102
1103         if (async) {
1104                 down_read(&current->mm->mmap_sem);
1105                 npages = get_user_page_nowait(current, current->mm,
1106                                               addr, write_fault, page);
1107                 up_read(&current->mm->mmap_sem);
1108         } else
1109                 npages = get_user_pages_fast(addr, 1, write_fault,
1110                                              page);
1111         if (npages != 1)
1112                 return npages;
1113
1114         /* map read fault as writable if possible */
1115         if (unlikely(!write_fault) && writable) {
1116                 struct page *wpage[1];
1117
1118                 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1119                 if (npages == 1) {
1120                         *writable = true;
1121                         put_page(page[0]);
1122                         page[0] = wpage[0];
1123                 }
1124
1125                 npages = 1;
1126         }
1127         *pfn = page_to_pfn(page[0]);
1128         return npages;
1129 }
1130
1131 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1132 {
1133         if (unlikely(!(vma->vm_flags & VM_READ)))
1134                 return false;
1135
1136         if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1137                 return false;
1138
1139         return true;
1140 }
1141
1142 /*
1143  * Pin guest page in memory and return its pfn.
1144  * @addr: host virtual address which maps memory to the guest
1145  * @atomic: whether this function can sleep
1146  * @async: whether this function need to wait IO complete if the
1147  *         host page is not in the memory
1148  * @write_fault: whether we should get a writable host page
1149  * @writable: whether it allows to map a writable host page for !@write_fault
1150  *
1151  * The function will map a writable host page for these two cases:
1152  * 1): @write_fault = true
1153  * 2): @write_fault = false && @writable, @writable will tell the caller
1154  *     whether the mapping is writable.
1155  */
1156 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1157                         bool write_fault, bool *writable)
1158 {
1159         struct vm_area_struct *vma;
1160         pfn_t pfn = 0;
1161         int npages;
1162
1163         /* we can do it either atomically or asynchronously, not both */
1164         BUG_ON(atomic && async);
1165
1166         if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1167                 return pfn;
1168
1169         if (atomic)
1170                 return KVM_PFN_ERR_FAULT;
1171
1172         npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1173         if (npages == 1)
1174                 return pfn;
1175
1176         down_read(&current->mm->mmap_sem);
1177         if (npages == -EHWPOISON ||
1178               (!async && check_user_page_hwpoison(addr))) {
1179                 pfn = KVM_PFN_ERR_HWPOISON;
1180                 goto exit;
1181         }
1182
1183         vma = find_vma_intersection(current->mm, addr, addr + 1);
1184
1185         if (vma == NULL)
1186                 pfn = KVM_PFN_ERR_FAULT;
1187         else if ((vma->vm_flags & VM_PFNMAP)) {
1188                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1189                         vma->vm_pgoff;
1190                 BUG_ON(!kvm_is_mmio_pfn(pfn));
1191         } else {
1192                 if (async && vma_is_valid(vma, write_fault))
1193                         *async = true;
1194                 pfn = KVM_PFN_ERR_FAULT;
1195         }
1196 exit:
1197         up_read(&current->mm->mmap_sem);
1198         return pfn;
1199 }
1200
1201 static pfn_t
1202 __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1203                      bool *async, bool write_fault, bool *writable)
1204 {
1205         unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1206
1207         if (addr == KVM_HVA_ERR_RO_BAD)
1208                 return KVM_PFN_ERR_RO_FAULT;
1209
1210         if (kvm_is_error_hva(addr))
1211                 return KVM_PFN_ERR_BAD;
1212
1213         /* Do not map writable pfn in the readonly memslot. */
1214         if (writable && memslot_is_readonly(slot)) {
1215                 *writable = false;
1216                 writable = NULL;
1217         }
1218
1219         return hva_to_pfn(addr, atomic, async, write_fault,
1220                           writable);
1221 }
1222
1223 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1224                           bool write_fault, bool *writable)
1225 {
1226         struct kvm_memory_slot *slot;
1227
1228         if (async)
1229                 *async = false;
1230
1231         slot = gfn_to_memslot(kvm, gfn);
1232
1233         return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1234                                     writable);
1235 }
1236
1237 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1238 {
1239         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1240 }
1241 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1242
1243 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1244                        bool write_fault, bool *writable)
1245 {
1246         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1247 }
1248 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1249
1250 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1251 {
1252         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1253 }
1254 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1255
1256 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1257                       bool *writable)
1258 {
1259         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1260 }
1261 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1262
1263 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1264 {
1265         return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1266 }
1267
1268 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1269 {
1270         return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1271 }
1272 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1273
1274 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1275                                                                   int nr_pages)
1276 {
1277         unsigned long addr;
1278         gfn_t entry;
1279
1280         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1281         if (kvm_is_error_hva(addr))
1282                 return -1;
1283
1284         if (entry < nr_pages)
1285                 return 0;
1286
1287         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1288 }
1289 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1290
1291 static struct page *kvm_pfn_to_page(pfn_t pfn)
1292 {
1293         if (is_error_pfn(pfn))
1294                 return KVM_ERR_PTR_BAD_PAGE;
1295
1296         if (kvm_is_mmio_pfn(pfn)) {
1297                 WARN_ON(1);
1298                 return KVM_ERR_PTR_BAD_PAGE;
1299         }
1300
1301         return pfn_to_page(pfn);
1302 }
1303
1304 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1305 {
1306         pfn_t pfn;
1307
1308         pfn = gfn_to_pfn(kvm, gfn);
1309
1310         return kvm_pfn_to_page(pfn);
1311 }
1312
1313 EXPORT_SYMBOL_GPL(gfn_to_page);
1314
1315 void kvm_release_page_clean(struct page *page)
1316 {
1317         WARN_ON(is_error_page(page));
1318
1319         kvm_release_pfn_clean(page_to_pfn(page));
1320 }
1321 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1322
1323 void kvm_release_pfn_clean(pfn_t pfn)
1324 {
1325         if (!is_error_pfn(pfn) && !kvm_is_mmio_pfn(pfn))
1326                 put_page(pfn_to_page(pfn));
1327 }
1328 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1329
1330 void kvm_release_page_dirty(struct page *page)
1331 {
1332         WARN_ON(is_error_page(page));
1333
1334         kvm_release_pfn_dirty(page_to_pfn(page));
1335 }
1336 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1337
1338 void kvm_release_pfn_dirty(pfn_t pfn)
1339 {
1340         kvm_set_pfn_dirty(pfn);
1341         kvm_release_pfn_clean(pfn);
1342 }
1343 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1344
1345 void kvm_set_page_dirty(struct page *page)
1346 {
1347         kvm_set_pfn_dirty(page_to_pfn(page));
1348 }
1349 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1350
1351 void kvm_set_pfn_dirty(pfn_t pfn)
1352 {
1353         if (!kvm_is_mmio_pfn(pfn)) {
1354                 struct page *page = pfn_to_page(pfn);
1355                 if (!PageReserved(page))
1356                         SetPageDirty(page);
1357         }
1358 }
1359 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1360
1361 void kvm_set_pfn_accessed(pfn_t pfn)
1362 {
1363         if (!kvm_is_mmio_pfn(pfn))
1364                 mark_page_accessed(pfn_to_page(pfn));
1365 }
1366 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1367
1368 void kvm_get_pfn(pfn_t pfn)
1369 {
1370         if (!kvm_is_mmio_pfn(pfn))
1371                 get_page(pfn_to_page(pfn));
1372 }
1373 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1374
1375 static int next_segment(unsigned long len, int offset)
1376 {
1377         if (len > PAGE_SIZE - offset)
1378                 return PAGE_SIZE - offset;
1379         else
1380                 return len;
1381 }
1382
1383 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1384                         int len)
1385 {
1386         int r;
1387         unsigned long addr;
1388
1389         addr = gfn_to_hva_read(kvm, gfn);
1390         if (kvm_is_error_hva(addr))
1391                 return -EFAULT;
1392         r = kvm_read_hva(data, (void __user *)addr + offset, len);
1393         if (r)
1394                 return -EFAULT;
1395         return 0;
1396 }
1397 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1398
1399 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, 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_read_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 EXPORT_SYMBOL_GPL(kvm_read_guest);
1418
1419 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1420                           unsigned long len)
1421 {
1422         int r;
1423         unsigned long addr;
1424         gfn_t gfn = gpa >> PAGE_SHIFT;
1425         int offset = offset_in_page(gpa);
1426
1427         addr = gfn_to_hva_read(kvm, gfn);
1428         if (kvm_is_error_hva(addr))
1429                 return -EFAULT;
1430         pagefault_disable();
1431         r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
1432         pagefault_enable();
1433         if (r)
1434                 return -EFAULT;
1435         return 0;
1436 }
1437 EXPORT_SYMBOL(kvm_read_guest_atomic);
1438
1439 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1440                          int offset, int len)
1441 {
1442         int r;
1443         unsigned long addr;
1444
1445         addr = gfn_to_hva(kvm, gfn);
1446         if (kvm_is_error_hva(addr))
1447                 return -EFAULT;
1448         r = __copy_to_user((void __user *)addr + offset, data, len);
1449         if (r)
1450                 return -EFAULT;
1451         mark_page_dirty(kvm, gfn);
1452         return 0;
1453 }
1454 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1455
1456 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1457                     unsigned long len)
1458 {
1459         gfn_t gfn = gpa >> PAGE_SHIFT;
1460         int seg;
1461         int offset = offset_in_page(gpa);
1462         int ret;
1463
1464         while ((seg = next_segment(len, offset)) != 0) {
1465                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1466                 if (ret < 0)
1467                         return ret;
1468                 offset = 0;
1469                 len -= seg;
1470                 data += seg;
1471                 ++gfn;
1472         }
1473         return 0;
1474 }
1475
1476 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1477                               gpa_t gpa)
1478 {
1479         struct kvm_memslots *slots = kvm_memslots(kvm);
1480         int offset = offset_in_page(gpa);
1481         gfn_t gfn = gpa >> PAGE_SHIFT;
1482
1483         ghc->gpa = gpa;
1484         ghc->generation = slots->generation;
1485         ghc->memslot = gfn_to_memslot(kvm, gfn);
1486         ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1487         if (!kvm_is_error_hva(ghc->hva))
1488                 ghc->hva += offset;
1489         else
1490                 return -EFAULT;
1491
1492         return 0;
1493 }
1494 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1495
1496 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1497                            void *data, unsigned long len)
1498 {
1499         struct kvm_memslots *slots = kvm_memslots(kvm);
1500         int r;
1501
1502         if (slots->generation != ghc->generation)
1503                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1504
1505         if (kvm_is_error_hva(ghc->hva))
1506                 return -EFAULT;
1507
1508         r = __copy_to_user((void __user *)ghc->hva, data, len);
1509         if (r)
1510                 return -EFAULT;
1511         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1512
1513         return 0;
1514 }
1515 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1516
1517 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1518                            void *data, unsigned long len)
1519 {
1520         struct kvm_memslots *slots = kvm_memslots(kvm);
1521         int r;
1522
1523         if (slots->generation != ghc->generation)
1524                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1525
1526         if (kvm_is_error_hva(ghc->hva))
1527                 return -EFAULT;
1528
1529         r = __copy_from_user(data, (void __user *)ghc->hva, len);
1530         if (r)
1531                 return -EFAULT;
1532
1533         return 0;
1534 }
1535 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1536
1537 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1538 {
1539         return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1540                                     offset, len);
1541 }
1542 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1543
1544 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1545 {
1546         gfn_t gfn = gpa >> PAGE_SHIFT;
1547         int seg;
1548         int offset = offset_in_page(gpa);
1549         int ret;
1550
1551         while ((seg = next_segment(len, offset)) != 0) {
1552                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1553                 if (ret < 0)
1554                         return ret;
1555                 offset = 0;
1556                 len -= seg;
1557                 ++gfn;
1558         }
1559         return 0;
1560 }
1561 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1562
1563 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1564                              gfn_t gfn)
1565 {
1566         if (memslot && memslot->dirty_bitmap) {
1567                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1568
1569                 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1570         }
1571 }
1572
1573 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1574 {
1575         struct kvm_memory_slot *memslot;
1576
1577         memslot = gfn_to_memslot(kvm, gfn);
1578         mark_page_dirty_in_slot(kvm, memslot, gfn);
1579 }
1580
1581 /*
1582  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1583  */
1584 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1585 {
1586         DEFINE_WAIT(wait);
1587
1588         for (;;) {
1589                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1590
1591                 if (kvm_arch_vcpu_runnable(vcpu)) {
1592                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1593                         break;
1594                 }
1595                 if (kvm_cpu_has_pending_timer(vcpu))
1596                         break;
1597                 if (signal_pending(current))
1598                         break;
1599
1600                 schedule();
1601         }
1602
1603         finish_wait(&vcpu->wq, &wait);
1604 }
1605
1606 #ifndef CONFIG_S390
1607 /*
1608  * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1609  */
1610 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1611 {
1612         int me;
1613         int cpu = vcpu->cpu;
1614         wait_queue_head_t *wqp;
1615
1616         wqp = kvm_arch_vcpu_wq(vcpu);
1617         if (waitqueue_active(wqp)) {
1618                 wake_up_interruptible(wqp);
1619                 ++vcpu->stat.halt_wakeup;
1620         }
1621
1622         me = get_cpu();
1623         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1624                 if (kvm_arch_vcpu_should_kick(vcpu))
1625                         smp_send_reschedule(cpu);
1626         put_cpu();
1627 }
1628 #endif /* !CONFIG_S390 */
1629
1630 void kvm_resched(struct kvm_vcpu *vcpu)
1631 {
1632         if (!need_resched())
1633                 return;
1634         cond_resched();
1635 }
1636 EXPORT_SYMBOL_GPL(kvm_resched);
1637
1638 bool kvm_vcpu_yield_to(struct kvm_vcpu *target)
1639 {
1640         struct pid *pid;
1641         struct task_struct *task = NULL;
1642
1643         rcu_read_lock();
1644         pid = rcu_dereference(target->pid);
1645         if (pid)
1646                 task = get_pid_task(target->pid, PIDTYPE_PID);
1647         rcu_read_unlock();
1648         if (!task)
1649                 return false;
1650         if (task->flags & PF_VCPU) {
1651                 put_task_struct(task);
1652                 return false;
1653         }
1654         if (yield_to(task, 1)) {
1655                 put_task_struct(task);
1656                 return true;
1657         }
1658         put_task_struct(task);
1659         return false;
1660 }
1661 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1662
1663 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1664 /*
1665  * Helper that checks whether a VCPU is eligible for directed yield.
1666  * Most eligible candidate to yield is decided by following heuristics:
1667  *
1668  *  (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1669  *  (preempted lock holder), indicated by @in_spin_loop.
1670  *  Set at the beiginning and cleared at the end of interception/PLE handler.
1671  *
1672  *  (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1673  *  chance last time (mostly it has become eligible now since we have probably
1674  *  yielded to lockholder in last iteration. This is done by toggling
1675  *  @dy_eligible each time a VCPU checked for eligibility.)
1676  *
1677  *  Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1678  *  to preempted lock-holder could result in wrong VCPU selection and CPU
1679  *  burning. Giving priority for a potential lock-holder increases lock
1680  *  progress.
1681  *
1682  *  Since algorithm is based on heuristics, accessing another VCPU data without
1683  *  locking does not harm. It may result in trying to yield to  same VCPU, fail
1684  *  and continue with next VCPU and so on.
1685  */
1686 bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1687 {
1688         bool eligible;
1689
1690         eligible = !vcpu->spin_loop.in_spin_loop ||
1691                         (vcpu->spin_loop.in_spin_loop &&
1692                          vcpu->spin_loop.dy_eligible);
1693
1694         if (vcpu->spin_loop.in_spin_loop)
1695                 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1696
1697         return eligible;
1698 }
1699 #endif
1700 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1701 {
1702         struct kvm *kvm = me->kvm;
1703         struct kvm_vcpu *vcpu;
1704         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1705         int yielded = 0;
1706         int pass;
1707         int i;
1708
1709         kvm_vcpu_set_in_spin_loop(me, true);
1710         /*
1711          * We boost the priority of a VCPU that is runnable but not
1712          * currently running, because it got preempted by something
1713          * else and called schedule in __vcpu_run.  Hopefully that
1714          * VCPU is holding the lock that we need and will release it.
1715          * We approximate round-robin by starting at the last boosted VCPU.
1716          */
1717         for (pass = 0; pass < 2 && !yielded; pass++) {
1718                 kvm_for_each_vcpu(i, vcpu, kvm) {
1719                         if (!pass && i <= last_boosted_vcpu) {
1720                                 i = last_boosted_vcpu;
1721                                 continue;
1722                         } else if (pass && i > last_boosted_vcpu)
1723                                 break;
1724                         if (vcpu == me)
1725                                 continue;
1726                         if (waitqueue_active(&vcpu->wq))
1727                                 continue;
1728                         if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1729                                 continue;
1730                         if (kvm_vcpu_yield_to(vcpu)) {
1731                                 kvm->last_boosted_vcpu = i;
1732                                 yielded = 1;
1733                                 break;
1734                         }
1735                 }
1736         }
1737         kvm_vcpu_set_in_spin_loop(me, false);
1738
1739         /* Ensure vcpu is not eligible during next spinloop */
1740         kvm_vcpu_set_dy_eligible(me, false);
1741 }
1742 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1743
1744 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1745 {
1746         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1747         struct page *page;
1748
1749         if (vmf->pgoff == 0)
1750                 page = virt_to_page(vcpu->run);
1751 #ifdef CONFIG_X86
1752         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1753                 page = virt_to_page(vcpu->arch.pio_data);
1754 #endif
1755 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1756         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1757                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1758 #endif
1759         else
1760                 return kvm_arch_vcpu_fault(vcpu, vmf);
1761         get_page(page);
1762         vmf->page = page;
1763         return 0;
1764 }
1765
1766 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1767         .fault = kvm_vcpu_fault,
1768 };
1769
1770 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1771 {
1772         vma->vm_ops = &kvm_vcpu_vm_ops;
1773         return 0;
1774 }
1775
1776 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1777 {
1778         struct kvm_vcpu *vcpu = filp->private_data;
1779
1780         kvm_put_kvm(vcpu->kvm);
1781         return 0;
1782 }
1783
1784 static struct file_operations kvm_vcpu_fops = {
1785         .release        = kvm_vcpu_release,
1786         .unlocked_ioctl = kvm_vcpu_ioctl,
1787 #ifdef CONFIG_COMPAT
1788         .compat_ioctl   = kvm_vcpu_compat_ioctl,
1789 #endif
1790         .mmap           = kvm_vcpu_mmap,
1791         .llseek         = noop_llseek,
1792 };
1793
1794 /*
1795  * Allocates an inode for the vcpu.
1796  */
1797 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1798 {
1799         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1800 }
1801
1802 /*
1803  * Creates some virtual cpus.  Good luck creating more than one.
1804  */
1805 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1806 {
1807         int r;
1808         struct kvm_vcpu *vcpu, *v;
1809
1810         vcpu = kvm_arch_vcpu_create(kvm, id);
1811         if (IS_ERR(vcpu))
1812                 return PTR_ERR(vcpu);
1813
1814         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1815
1816         r = kvm_arch_vcpu_setup(vcpu);
1817         if (r)
1818                 goto vcpu_destroy;
1819
1820         mutex_lock(&kvm->lock);
1821         if (!kvm_vcpu_compatible(vcpu)) {
1822                 r = -EINVAL;
1823                 goto unlock_vcpu_destroy;
1824         }
1825         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1826                 r = -EINVAL;
1827                 goto unlock_vcpu_destroy;
1828         }
1829
1830         kvm_for_each_vcpu(r, v, kvm)
1831                 if (v->vcpu_id == id) {
1832                         r = -EEXIST;
1833                         goto unlock_vcpu_destroy;
1834                 }
1835
1836         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1837
1838         /* Now it's all set up, let userspace reach it */
1839         kvm_get_kvm(kvm);
1840         r = create_vcpu_fd(vcpu);
1841         if (r < 0) {
1842                 kvm_put_kvm(kvm);
1843                 goto unlock_vcpu_destroy;
1844         }
1845
1846         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1847         smp_wmb();
1848         atomic_inc(&kvm->online_vcpus);
1849
1850         mutex_unlock(&kvm->lock);
1851         return r;
1852
1853 unlock_vcpu_destroy:
1854         mutex_unlock(&kvm->lock);
1855 vcpu_destroy:
1856         kvm_arch_vcpu_destroy(vcpu);
1857         return r;
1858 }
1859
1860 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1861 {
1862         if (sigset) {
1863                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1864                 vcpu->sigset_active = 1;
1865                 vcpu->sigset = *sigset;
1866         } else
1867                 vcpu->sigset_active = 0;
1868         return 0;
1869 }
1870
1871 static long kvm_vcpu_ioctl(struct file *filp,
1872                            unsigned int ioctl, unsigned long arg)
1873 {
1874         struct kvm_vcpu *vcpu = filp->private_data;
1875         void __user *argp = (void __user *)arg;
1876         int r;
1877         struct kvm_fpu *fpu = NULL;
1878         struct kvm_sregs *kvm_sregs = NULL;
1879
1880         if (vcpu->kvm->mm != current->mm)
1881                 return -EIO;
1882
1883 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1884         /*
1885          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1886          * so vcpu_load() would break it.
1887          */
1888         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1889                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1890 #endif
1891
1892
1893         r = vcpu_load(vcpu);
1894         if (r)
1895                 return r;
1896         switch (ioctl) {
1897         case KVM_RUN:
1898                 r = -EINVAL;
1899                 if (arg)
1900                         goto out;
1901                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1902                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1903                 break;
1904         case KVM_GET_REGS: {
1905                 struct kvm_regs *kvm_regs;
1906
1907                 r = -ENOMEM;
1908                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1909                 if (!kvm_regs)
1910                         goto out;
1911                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1912                 if (r)
1913                         goto out_free1;
1914                 r = -EFAULT;
1915                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1916                         goto out_free1;
1917                 r = 0;
1918 out_free1:
1919                 kfree(kvm_regs);
1920                 break;
1921         }
1922         case KVM_SET_REGS: {
1923                 struct kvm_regs *kvm_regs;
1924
1925                 r = -ENOMEM;
1926                 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
1927                 if (IS_ERR(kvm_regs)) {
1928                         r = PTR_ERR(kvm_regs);
1929                         goto out;
1930                 }
1931                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1932                 if (r)
1933                         goto out_free2;
1934                 r = 0;
1935 out_free2:
1936                 kfree(kvm_regs);
1937                 break;
1938         }
1939         case KVM_GET_SREGS: {
1940                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1941                 r = -ENOMEM;
1942                 if (!kvm_sregs)
1943                         goto out;
1944                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1945                 if (r)
1946                         goto out;
1947                 r = -EFAULT;
1948                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1949                         goto out;
1950                 r = 0;
1951                 break;
1952         }
1953         case KVM_SET_SREGS: {
1954                 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
1955                 if (IS_ERR(kvm_sregs)) {
1956                         r = PTR_ERR(kvm_sregs);
1957                         goto out;
1958                 }
1959                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1960                 if (r)
1961                         goto out;
1962                 r = 0;
1963                 break;
1964         }
1965         case KVM_GET_MP_STATE: {
1966                 struct kvm_mp_state mp_state;
1967
1968                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1969                 if (r)
1970                         goto out;
1971                 r = -EFAULT;
1972                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1973                         goto out;
1974                 r = 0;
1975                 break;
1976         }
1977         case KVM_SET_MP_STATE: {
1978                 struct kvm_mp_state mp_state;
1979
1980                 r = -EFAULT;
1981                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1982                         goto out;
1983                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1984                 if (r)
1985                         goto out;
1986                 r = 0;
1987                 break;
1988         }
1989         case KVM_TRANSLATE: {
1990                 struct kvm_translation tr;
1991
1992                 r = -EFAULT;
1993                 if (copy_from_user(&tr, argp, sizeof tr))
1994                         goto out;
1995                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1996                 if (r)
1997                         goto out;
1998                 r = -EFAULT;
1999                 if (copy_to_user(argp, &tr, sizeof tr))
2000                         goto out;
2001                 r = 0;
2002                 break;
2003         }
2004         case KVM_SET_GUEST_DEBUG: {
2005                 struct kvm_guest_debug dbg;
2006
2007                 r = -EFAULT;
2008                 if (copy_from_user(&dbg, argp, sizeof dbg))
2009                         goto out;
2010                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2011                 if (r)
2012                         goto out;
2013                 r = 0;
2014                 break;
2015         }
2016         case KVM_SET_SIGNAL_MASK: {
2017                 struct kvm_signal_mask __user *sigmask_arg = argp;
2018                 struct kvm_signal_mask kvm_sigmask;
2019                 sigset_t sigset, *p;
2020
2021                 p = NULL;
2022                 if (argp) {
2023                         r = -EFAULT;
2024                         if (copy_from_user(&kvm_sigmask, argp,
2025                                            sizeof kvm_sigmask))
2026                                 goto out;
2027                         r = -EINVAL;
2028                         if (kvm_sigmask.len != sizeof sigset)
2029                                 goto out;
2030                         r = -EFAULT;
2031                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2032                                            sizeof sigset))
2033                                 goto out;
2034                         p = &sigset;
2035                 }
2036                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2037                 break;
2038         }
2039         case KVM_GET_FPU: {
2040                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2041                 r = -ENOMEM;
2042                 if (!fpu)
2043                         goto out;
2044                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2045                 if (r)
2046                         goto out;
2047                 r = -EFAULT;
2048                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2049                         goto out;
2050                 r = 0;
2051                 break;
2052         }
2053         case KVM_SET_FPU: {
2054                 fpu = memdup_user(argp, sizeof(*fpu));
2055                 if (IS_ERR(fpu)) {
2056                         r = PTR_ERR(fpu);
2057                         goto out;
2058                 }
2059                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2060                 if (r)
2061                         goto out;
2062                 r = 0;
2063                 break;
2064         }
2065         default:
2066                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2067         }
2068 out:
2069         vcpu_put(vcpu);
2070         kfree(fpu);
2071         kfree(kvm_sregs);
2072         return r;
2073 }
2074
2075 #ifdef CONFIG_COMPAT
2076 static long kvm_vcpu_compat_ioctl(struct file *filp,
2077                                   unsigned int ioctl, unsigned long arg)
2078 {
2079         struct kvm_vcpu *vcpu = filp->private_data;
2080         void __user *argp = compat_ptr(arg);
2081         int r;
2082
2083         if (vcpu->kvm->mm != current->mm)
2084                 return -EIO;
2085
2086         switch (ioctl) {
2087         case KVM_SET_SIGNAL_MASK: {
2088                 struct kvm_signal_mask __user *sigmask_arg = argp;
2089                 struct kvm_signal_mask kvm_sigmask;
2090                 compat_sigset_t csigset;
2091                 sigset_t sigset;
2092
2093                 if (argp) {
2094                         r = -EFAULT;
2095                         if (copy_from_user(&kvm_sigmask, argp,
2096                                            sizeof kvm_sigmask))
2097                                 goto out;
2098                         r = -EINVAL;
2099                         if (kvm_sigmask.len != sizeof csigset)
2100                                 goto out;
2101                         r = -EFAULT;
2102                         if (copy_from_user(&csigset, sigmask_arg->sigset,
2103                                            sizeof csigset))
2104                                 goto out;
2105                         sigset_from_compat(&sigset, &csigset);
2106                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2107                 } else
2108                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2109                 break;
2110         }
2111         default:
2112                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2113         }
2114
2115 out:
2116         return r;
2117 }
2118 #endif
2119
2120 static long kvm_vm_ioctl(struct file *filp,
2121                            unsigned int ioctl, unsigned long arg)
2122 {
2123         struct kvm *kvm = filp->private_data;
2124         void __user *argp = (void __user *)arg;
2125         int r;
2126
2127         if (kvm->mm != current->mm)
2128                 return -EIO;
2129         switch (ioctl) {
2130         case KVM_CREATE_VCPU:
2131                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2132                 if (r < 0)
2133                         goto out;
2134                 break;
2135         case KVM_SET_USER_MEMORY_REGION: {
2136                 struct kvm_userspace_memory_region kvm_userspace_mem;
2137
2138                 r = -EFAULT;
2139                 if (copy_from_user(&kvm_userspace_mem, argp,
2140                                                 sizeof kvm_userspace_mem))
2141                         goto out;
2142
2143                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
2144                 if (r)
2145                         goto out;
2146                 break;
2147         }
2148         case KVM_GET_DIRTY_LOG: {
2149                 struct kvm_dirty_log log;
2150
2151                 r = -EFAULT;
2152                 if (copy_from_user(&log, argp, sizeof log))
2153                         goto out;
2154                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2155                 if (r)
2156                         goto out;
2157                 break;
2158         }
2159 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2160         case KVM_REGISTER_COALESCED_MMIO: {
2161                 struct kvm_coalesced_mmio_zone zone;
2162                 r = -EFAULT;
2163                 if (copy_from_user(&zone, argp, sizeof zone))
2164                         goto out;
2165                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2166                 if (r)
2167                         goto out;
2168                 r = 0;
2169                 break;
2170         }
2171         case KVM_UNREGISTER_COALESCED_MMIO: {
2172                 struct kvm_coalesced_mmio_zone zone;
2173                 r = -EFAULT;
2174                 if (copy_from_user(&zone, argp, sizeof zone))
2175                         goto out;
2176                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2177                 if (r)
2178                         goto out;
2179                 r = 0;
2180                 break;
2181         }
2182 #endif
2183         case KVM_IRQFD: {
2184                 struct kvm_irqfd data;
2185
2186                 r = -EFAULT;
2187                 if (copy_from_user(&data, argp, sizeof data))
2188                         goto out;
2189                 r = kvm_irqfd(kvm, &data);
2190                 break;
2191         }
2192         case KVM_IOEVENTFD: {
2193                 struct kvm_ioeventfd data;
2194
2195                 r = -EFAULT;
2196                 if (copy_from_user(&data, argp, sizeof data))
2197                         goto out;
2198                 r = kvm_ioeventfd(kvm, &data);
2199                 break;
2200         }
2201 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2202         case KVM_SET_BOOT_CPU_ID:
2203                 r = 0;
2204                 mutex_lock(&kvm->lock);
2205                 if (atomic_read(&kvm->online_vcpus) != 0)
2206                         r = -EBUSY;
2207                 else
2208                         kvm->bsp_vcpu_id = arg;
2209                 mutex_unlock(&kvm->lock);
2210                 break;
2211 #endif
2212 #ifdef CONFIG_HAVE_KVM_MSI
2213         case KVM_SIGNAL_MSI: {
2214                 struct kvm_msi msi;
2215
2216                 r = -EFAULT;
2217                 if (copy_from_user(&msi, argp, sizeof msi))
2218                         goto out;
2219                 r = kvm_send_userspace_msi(kvm, &msi);
2220                 break;
2221         }
2222 #endif
2223 #ifdef __KVM_HAVE_IRQ_LINE
2224         case KVM_IRQ_LINE_STATUS:
2225         case KVM_IRQ_LINE: {
2226                 struct kvm_irq_level irq_event;
2227
2228                 r = -EFAULT;
2229                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2230                         goto out;
2231
2232                 r = kvm_vm_ioctl_irq_line(kvm, &irq_event);
2233                 if (r)
2234                         goto out;
2235
2236                 r = -EFAULT;
2237                 if (ioctl == KVM_IRQ_LINE_STATUS) {
2238                         if (copy_to_user(argp, &irq_event, sizeof irq_event))
2239                                 goto out;
2240                 }
2241
2242                 r = 0;
2243                 break;
2244         }
2245 #endif
2246         default:
2247                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2248                 if (r == -ENOTTY)
2249                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2250         }
2251 out:
2252         return r;
2253 }
2254
2255 #ifdef CONFIG_COMPAT
2256 struct compat_kvm_dirty_log {
2257         __u32 slot;
2258         __u32 padding1;
2259         union {
2260                 compat_uptr_t dirty_bitmap; /* one bit per page */
2261                 __u64 padding2;
2262         };
2263 };
2264
2265 static long kvm_vm_compat_ioctl(struct file *filp,
2266                            unsigned int ioctl, unsigned long arg)
2267 {
2268         struct kvm *kvm = filp->private_data;
2269         int r;
2270
2271         if (kvm->mm != current->mm)
2272                 return -EIO;
2273         switch (ioctl) {
2274         case KVM_GET_DIRTY_LOG: {
2275                 struct compat_kvm_dirty_log compat_log;
2276                 struct kvm_dirty_log log;
2277
2278                 r = -EFAULT;
2279                 if (copy_from_user(&compat_log, (void __user *)arg,
2280                                    sizeof(compat_log)))
2281                         goto out;
2282                 log.slot         = compat_log.slot;
2283                 log.padding1     = compat_log.padding1;
2284                 log.padding2     = compat_log.padding2;
2285                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2286
2287                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2288                 if (r)
2289                         goto out;
2290                 break;
2291         }
2292         default:
2293                 r = kvm_vm_ioctl(filp, ioctl, arg);
2294         }
2295
2296 out:
2297         return r;
2298 }
2299 #endif
2300
2301 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2302 {
2303         struct page *page[1];
2304         unsigned long addr;
2305         int npages;
2306         gfn_t gfn = vmf->pgoff;
2307         struct kvm *kvm = vma->vm_file->private_data;
2308
2309         addr = gfn_to_hva(kvm, gfn);
2310         if (kvm_is_error_hva(addr))
2311                 return VM_FAULT_SIGBUS;
2312
2313         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2314                                 NULL);
2315         if (unlikely(npages != 1))
2316                 return VM_FAULT_SIGBUS;
2317
2318         vmf->page = page[0];
2319         return 0;
2320 }
2321
2322 static const struct vm_operations_struct kvm_vm_vm_ops = {
2323         .fault = kvm_vm_fault,
2324 };
2325
2326 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2327 {
2328         vma->vm_ops = &kvm_vm_vm_ops;
2329         return 0;
2330 }
2331
2332 static struct file_operations kvm_vm_fops = {
2333         .release        = kvm_vm_release,
2334         .unlocked_ioctl = kvm_vm_ioctl,
2335 #ifdef CONFIG_COMPAT
2336         .compat_ioctl   = kvm_vm_compat_ioctl,
2337 #endif
2338         .mmap           = kvm_vm_mmap,
2339         .llseek         = noop_llseek,
2340 };
2341
2342 static int kvm_dev_ioctl_create_vm(unsigned long type)
2343 {
2344         int r;
2345         struct kvm *kvm;
2346
2347         kvm = kvm_create_vm(type);
2348         if (IS_ERR(kvm))
2349                 return PTR_ERR(kvm);
2350 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2351         r = kvm_coalesced_mmio_init(kvm);
2352         if (r < 0) {
2353                 kvm_put_kvm(kvm);
2354                 return r;
2355         }
2356 #endif
2357         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2358         if (r < 0)
2359                 kvm_put_kvm(kvm);
2360
2361         return r;
2362 }
2363
2364 static long kvm_dev_ioctl_check_extension_generic(long arg)
2365 {
2366         switch (arg) {
2367         case KVM_CAP_USER_MEMORY:
2368         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2369         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2370 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2371         case KVM_CAP_SET_BOOT_CPU_ID:
2372 #endif
2373         case KVM_CAP_INTERNAL_ERROR_DATA:
2374 #ifdef CONFIG_HAVE_KVM_MSI
2375         case KVM_CAP_SIGNAL_MSI:
2376 #endif
2377                 return 1;
2378 #ifdef KVM_CAP_IRQ_ROUTING
2379         case KVM_CAP_IRQ_ROUTING:
2380                 return KVM_MAX_IRQ_ROUTES;
2381 #endif
2382         default:
2383                 break;
2384         }
2385         return kvm_dev_ioctl_check_extension(arg);
2386 }
2387
2388 static long kvm_dev_ioctl(struct file *filp,
2389                           unsigned int ioctl, unsigned long arg)
2390 {
2391         long r = -EINVAL;
2392
2393         switch (ioctl) {
2394         case KVM_GET_API_VERSION:
2395                 r = -EINVAL;
2396                 if (arg)
2397                         goto out;
2398                 r = KVM_API_VERSION;
2399                 break;
2400         case KVM_CREATE_VM:
2401                 r = kvm_dev_ioctl_create_vm(arg);
2402                 break;
2403         case KVM_CHECK_EXTENSION:
2404                 r = kvm_dev_ioctl_check_extension_generic(arg);
2405                 break;
2406         case KVM_GET_VCPU_MMAP_SIZE:
2407                 r = -EINVAL;
2408                 if (arg)
2409                         goto out;
2410                 r = PAGE_SIZE;     /* struct kvm_run */
2411 #ifdef CONFIG_X86
2412                 r += PAGE_SIZE;    /* pio data page */
2413 #endif
2414 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2415                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2416 #endif
2417                 break;
2418         case KVM_TRACE_ENABLE:
2419         case KVM_TRACE_PAUSE:
2420         case KVM_TRACE_DISABLE:
2421                 r = -EOPNOTSUPP;
2422                 break;
2423         default:
2424                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2425         }
2426 out:
2427         return r;
2428 }
2429
2430 static struct file_operations kvm_chardev_ops = {
2431         .unlocked_ioctl = kvm_dev_ioctl,
2432         .compat_ioctl   = kvm_dev_ioctl,
2433         .llseek         = noop_llseek,
2434 };
2435
2436 static struct miscdevice kvm_dev = {
2437         KVM_MINOR,
2438         "kvm",
2439         &kvm_chardev_ops,
2440 };
2441
2442 static void hardware_enable_nolock(void *junk)
2443 {
2444         int cpu = raw_smp_processor_id();
2445         int r;
2446
2447         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2448                 return;
2449
2450         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2451
2452         r = kvm_arch_hardware_enable(NULL);
2453
2454         if (r) {
2455                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2456                 atomic_inc(&hardware_enable_failed);
2457                 printk(KERN_INFO "kvm: enabling virtualization on "
2458                                  "CPU%d failed\n", cpu);
2459         }
2460 }
2461
2462 static void hardware_enable(void *junk)
2463 {
2464         raw_spin_lock(&kvm_lock);
2465         hardware_enable_nolock(junk);
2466         raw_spin_unlock(&kvm_lock);
2467 }
2468
2469 static void hardware_disable_nolock(void *junk)
2470 {
2471         int cpu = raw_smp_processor_id();
2472
2473         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2474                 return;
2475         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2476         kvm_arch_hardware_disable(NULL);
2477 }
2478
2479 static void hardware_disable(void *junk)
2480 {
2481         raw_spin_lock(&kvm_lock);
2482         hardware_disable_nolock(junk);
2483         raw_spin_unlock(&kvm_lock);
2484 }
2485
2486 static void hardware_disable_all_nolock(void)
2487 {
2488         BUG_ON(!kvm_usage_count);
2489
2490         kvm_usage_count--;
2491         if (!kvm_usage_count)
2492                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2493 }
2494
2495 static void hardware_disable_all(void)
2496 {
2497         raw_spin_lock(&kvm_lock);
2498         hardware_disable_all_nolock();
2499         raw_spin_unlock(&kvm_lock);
2500 }
2501
2502 static int hardware_enable_all(void)
2503 {
2504         int r = 0;
2505
2506         raw_spin_lock(&kvm_lock);
2507
2508         kvm_usage_count++;
2509         if (kvm_usage_count == 1) {
2510                 atomic_set(&hardware_enable_failed, 0);
2511                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2512
2513                 if (atomic_read(&hardware_enable_failed)) {
2514                         hardware_disable_all_nolock();
2515                         r = -EBUSY;
2516                 }
2517         }
2518
2519         raw_spin_unlock(&kvm_lock);
2520
2521         return r;
2522 }
2523
2524 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2525                            void *v)
2526 {
2527         int cpu = (long)v;
2528
2529         if (!kvm_usage_count)
2530                 return NOTIFY_OK;
2531
2532         val &= ~CPU_TASKS_FROZEN;
2533         switch (val) {
2534         case CPU_DYING:
2535                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2536                        cpu);
2537                 hardware_disable(NULL);
2538                 break;
2539         case CPU_STARTING:
2540                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2541                        cpu);
2542                 hardware_enable(NULL);
2543                 break;
2544         }
2545         return NOTIFY_OK;
2546 }
2547
2548
2549 asmlinkage void kvm_spurious_fault(void)
2550 {
2551         /* Fault while not rebooting.  We want the trace. */
2552         BUG();
2553 }
2554 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
2555
2556 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2557                       void *v)
2558 {
2559         /*
2560          * Some (well, at least mine) BIOSes hang on reboot if
2561          * in vmx root mode.
2562          *
2563          * And Intel TXT required VMX off for all cpu when system shutdown.
2564          */
2565         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2566         kvm_rebooting = true;
2567         on_each_cpu(hardware_disable_nolock, NULL, 1);
2568         return NOTIFY_OK;
2569 }
2570
2571 static struct notifier_block kvm_reboot_notifier = {
2572         .notifier_call = kvm_reboot,
2573         .priority = 0,
2574 };
2575
2576 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2577 {
2578         int i;
2579
2580         for (i = 0; i < bus->dev_count; i++) {
2581                 struct kvm_io_device *pos = bus->range[i].dev;
2582
2583                 kvm_iodevice_destructor(pos);
2584         }
2585         kfree(bus);
2586 }
2587
2588 int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2589 {
2590         const struct kvm_io_range *r1 = p1;
2591         const struct kvm_io_range *r2 = p2;
2592
2593         if (r1->addr < r2->addr)
2594                 return -1;
2595         if (r1->addr + r1->len > r2->addr + r2->len)
2596                 return 1;
2597         return 0;
2598 }
2599
2600 int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2601                           gpa_t addr, int len)
2602 {
2603         bus->range[bus->dev_count++] = (struct kvm_io_range) {
2604                 .addr = addr,
2605                 .len = len,
2606                 .dev = dev,
2607         };
2608
2609         sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2610                 kvm_io_bus_sort_cmp, NULL);
2611
2612         return 0;
2613 }
2614
2615 int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2616                              gpa_t addr, int len)
2617 {
2618         struct kvm_io_range *range, key;
2619         int off;
2620
2621         key = (struct kvm_io_range) {
2622                 .addr = addr,
2623                 .len = len,
2624         };
2625
2626         range = bsearch(&key, bus->range, bus->dev_count,
2627                         sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2628         if (range == NULL)
2629                 return -ENOENT;
2630
2631         off = range - bus->range;
2632
2633         while (off > 0 && kvm_io_bus_sort_cmp(&key, &bus->range[off-1]) == 0)
2634                 off--;
2635
2636         return off;
2637 }
2638
2639 /* kvm_io_bus_write - called under kvm->slots_lock */
2640 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2641                      int len, const void *val)
2642 {
2643         int idx;
2644         struct kvm_io_bus *bus;
2645         struct kvm_io_range range;
2646
2647         range = (struct kvm_io_range) {
2648                 .addr = addr,
2649                 .len = len,
2650         };
2651
2652         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2653         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2654         if (idx < 0)
2655                 return -EOPNOTSUPP;
2656
2657         while (idx < bus->dev_count &&
2658                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2659                 if (!kvm_iodevice_write(bus->range[idx].dev, addr, len, val))
2660                         return 0;
2661                 idx++;
2662         }
2663
2664         return -EOPNOTSUPP;
2665 }
2666
2667 /* kvm_io_bus_read - called under kvm->slots_lock */
2668 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2669                     int len, void *val)
2670 {
2671         int idx;
2672         struct kvm_io_bus *bus;
2673         struct kvm_io_range range;
2674
2675         range = (struct kvm_io_range) {
2676                 .addr = addr,
2677                 .len = len,
2678         };
2679
2680         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2681         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2682         if (idx < 0)
2683                 return -EOPNOTSUPP;
2684
2685         while (idx < bus->dev_count &&
2686                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2687                 if (!kvm_iodevice_read(bus->range[idx].dev, addr, len, val))
2688                         return 0;
2689                 idx++;
2690         }
2691
2692         return -EOPNOTSUPP;
2693 }
2694
2695 /* Caller must hold slots_lock. */
2696 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2697                             int len, struct kvm_io_device *dev)
2698 {
2699         struct kvm_io_bus *new_bus, *bus;
2700
2701         bus = kvm->buses[bus_idx];
2702         if (bus->dev_count > NR_IOBUS_DEVS - 1)
2703                 return -ENOSPC;
2704
2705         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
2706                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2707         if (!new_bus)
2708                 return -ENOMEM;
2709         memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
2710                sizeof(struct kvm_io_range)));
2711         kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2712         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2713         synchronize_srcu_expedited(&kvm->srcu);
2714         kfree(bus);
2715
2716         return 0;
2717 }
2718
2719 /* Caller must hold slots_lock. */
2720 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2721                               struct kvm_io_device *dev)
2722 {
2723         int i, r;
2724         struct kvm_io_bus *new_bus, *bus;
2725
2726         bus = kvm->buses[bus_idx];
2727         r = -ENOENT;
2728         for (i = 0; i < bus->dev_count; i++)
2729                 if (bus->range[i].dev == dev) {
2730                         r = 0;
2731                         break;
2732                 }
2733
2734         if (r)
2735                 return r;
2736
2737         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
2738                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2739         if (!new_bus)
2740                 return -ENOMEM;
2741
2742         memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
2743         new_bus->dev_count--;
2744         memcpy(new_bus->range + i, bus->range + i + 1,
2745                (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
2746
2747         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2748         synchronize_srcu_expedited(&kvm->srcu);
2749         kfree(bus);
2750         return r;
2751 }
2752
2753 static struct notifier_block kvm_cpu_notifier = {
2754         .notifier_call = kvm_cpu_hotplug,
2755 };
2756
2757 static int vm_stat_get(void *_offset, u64 *val)
2758 {
2759         unsigned offset = (long)_offset;
2760         struct kvm *kvm;
2761
2762         *val = 0;
2763         raw_spin_lock(&kvm_lock);
2764         list_for_each_entry(kvm, &vm_list, vm_list)
2765                 *val += *(u32 *)((void *)kvm + offset);
2766         raw_spin_unlock(&kvm_lock);
2767         return 0;
2768 }
2769
2770 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2771
2772 static int vcpu_stat_get(void *_offset, u64 *val)
2773 {
2774         unsigned offset = (long)_offset;
2775         struct kvm *kvm;
2776         struct kvm_vcpu *vcpu;
2777         int i;
2778
2779         *val = 0;
2780         raw_spin_lock(&kvm_lock);
2781         list_for_each_entry(kvm, &vm_list, vm_list)
2782                 kvm_for_each_vcpu(i, vcpu, kvm)
2783                         *val += *(u32 *)((void *)vcpu + offset);
2784
2785         raw_spin_unlock(&kvm_lock);
2786         return 0;
2787 }
2788
2789 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2790
2791 static const struct file_operations *stat_fops[] = {
2792         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2793         [KVM_STAT_VM]   = &vm_stat_fops,
2794 };
2795
2796 static int kvm_init_debug(void)
2797 {
2798         int r = -EFAULT;
2799         struct kvm_stats_debugfs_item *p;
2800
2801         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2802         if (kvm_debugfs_dir == NULL)
2803                 goto out;
2804
2805         for (p = debugfs_entries; p->name; ++p) {
2806                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2807                                                 (void *)(long)p->offset,
2808                                                 stat_fops[p->kind]);
2809                 if (p->dentry == NULL)
2810                         goto out_dir;
2811         }
2812
2813         return 0;
2814
2815 out_dir:
2816         debugfs_remove_recursive(kvm_debugfs_dir);
2817 out:
2818         return r;
2819 }
2820
2821 static void kvm_exit_debug(void)
2822 {
2823         struct kvm_stats_debugfs_item *p;
2824
2825         for (p = debugfs_entries; p->name; ++p)
2826                 debugfs_remove(p->dentry);
2827         debugfs_remove(kvm_debugfs_dir);
2828 }
2829
2830 static int kvm_suspend(void)
2831 {
2832         if (kvm_usage_count)
2833                 hardware_disable_nolock(NULL);
2834         return 0;
2835 }
2836
2837 static void kvm_resume(void)
2838 {
2839         if (kvm_usage_count) {
2840                 WARN_ON(raw_spin_is_locked(&kvm_lock));
2841                 hardware_enable_nolock(NULL);
2842         }
2843 }
2844
2845 static struct syscore_ops kvm_syscore_ops = {
2846         .suspend = kvm_suspend,
2847         .resume = kvm_resume,
2848 };
2849
2850 static inline
2851 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2852 {
2853         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2854 }
2855
2856 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2857 {
2858         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2859
2860         kvm_arch_vcpu_load(vcpu, cpu);
2861 }
2862
2863 static void kvm_sched_out(struct preempt_notifier *pn,
2864                           struct task_struct *next)
2865 {
2866         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2867
2868         kvm_arch_vcpu_put(vcpu);
2869 }
2870
2871 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2872                   struct module *module)
2873 {
2874         int r;
2875         int cpu;
2876
2877         r = kvm_arch_init(opaque);
2878         if (r)
2879                 goto out_fail;
2880
2881         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2882                 r = -ENOMEM;
2883                 goto out_free_0;
2884         }
2885
2886         r = kvm_arch_hardware_setup();
2887         if (r < 0)
2888                 goto out_free_0a;
2889
2890         for_each_online_cpu(cpu) {
2891                 smp_call_function_single(cpu,
2892                                 kvm_arch_check_processor_compat,
2893                                 &r, 1);
2894                 if (r < 0)
2895                         goto out_free_1;
2896         }
2897
2898         r = register_cpu_notifier(&kvm_cpu_notifier);
2899         if (r)
2900                 goto out_free_2;
2901         register_reboot_notifier(&kvm_reboot_notifier);
2902
2903         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2904         if (!vcpu_align)
2905                 vcpu_align = __alignof__(struct kvm_vcpu);
2906         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2907                                            0, NULL);
2908         if (!kvm_vcpu_cache) {
2909                 r = -ENOMEM;
2910                 goto out_free_3;
2911         }
2912
2913         r = kvm_async_pf_init();
2914         if (r)
2915                 goto out_free;
2916
2917         kvm_chardev_ops.owner = module;
2918         kvm_vm_fops.owner = module;
2919         kvm_vcpu_fops.owner = module;
2920
2921         r = misc_register(&kvm_dev);
2922         if (r) {
2923                 printk(KERN_ERR "kvm: misc device register failed\n");
2924                 goto out_unreg;
2925         }
2926
2927         register_syscore_ops(&kvm_syscore_ops);
2928
2929         kvm_preempt_ops.sched_in = kvm_sched_in;
2930         kvm_preempt_ops.sched_out = kvm_sched_out;
2931
2932         r = kvm_init_debug();
2933         if (r) {
2934                 printk(KERN_ERR "kvm: create debugfs files failed\n");
2935                 goto out_undebugfs;
2936         }
2937
2938         return 0;
2939
2940 out_undebugfs:
2941         unregister_syscore_ops(&kvm_syscore_ops);
2942 out_unreg:
2943         kvm_async_pf_deinit();
2944 out_free:
2945         kmem_cache_destroy(kvm_vcpu_cache);
2946 out_free_3:
2947         unregister_reboot_notifier(&kvm_reboot_notifier);
2948         unregister_cpu_notifier(&kvm_cpu_notifier);
2949 out_free_2:
2950 out_free_1:
2951         kvm_arch_hardware_unsetup();
2952 out_free_0a:
2953         free_cpumask_var(cpus_hardware_enabled);
2954 out_free_0:
2955         kvm_arch_exit();
2956 out_fail:
2957         return r;
2958 }
2959 EXPORT_SYMBOL_GPL(kvm_init);
2960
2961 void kvm_exit(void)
2962 {
2963         kvm_exit_debug();
2964         misc_deregister(&kvm_dev);
2965         kmem_cache_destroy(kvm_vcpu_cache);
2966         kvm_async_pf_deinit();
2967         unregister_syscore_ops(&kvm_syscore_ops);
2968         unregister_reboot_notifier(&kvm_reboot_notifier);
2969         unregister_cpu_notifier(&kvm_cpu_notifier);
2970         on_each_cpu(hardware_disable_nolock, NULL, 1);
2971         kvm_arch_hardware_unsetup();
2972         kvm_arch_exit();
2973         free_cpumask_var(cpus_hardware_enabled);
2974 }
2975 EXPORT_SYMBOL_GPL(kvm_exit);