Merge tag 'drm-fixes-for-v4.16-rc2' of git://people.freedesktop.org/~airlied/linux
[sfrench/cifs-2.6.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 <kvm/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/signal.h>
36 #include <linux/sched/mm.h>
37 #include <linux/sched/stat.h>
38 #include <linux/cpumask.h>
39 #include <linux/smp.h>
40 #include <linux/anon_inodes.h>
41 #include <linux/profile.h>
42 #include <linux/kvm_para.h>
43 #include <linux/pagemap.h>
44 #include <linux/mman.h>
45 #include <linux/swap.h>
46 #include <linux/bitops.h>
47 #include <linux/spinlock.h>
48 #include <linux/compat.h>
49 #include <linux/srcu.h>
50 #include <linux/hugetlb.h>
51 #include <linux/slab.h>
52 #include <linux/sort.h>
53 #include <linux/bsearch.h>
54
55 #include <asm/processor.h>
56 #include <asm/io.h>
57 #include <asm/ioctl.h>
58 #include <linux/uaccess.h>
59 #include <asm/pgtable.h>
60
61 #include "coalesced_mmio.h"
62 #include "async_pf.h"
63 #include "vfio.h"
64
65 #define CREATE_TRACE_POINTS
66 #include <trace/events/kvm.h>
67
68 /* Worst case buffer size needed for holding an integer. */
69 #define ITOA_MAX_LEN 12
70
71 MODULE_AUTHOR("Qumranet");
72 MODULE_LICENSE("GPL");
73
74 /* Architectures should define their poll value according to the halt latency */
75 unsigned int halt_poll_ns = KVM_HALT_POLL_NS_DEFAULT;
76 module_param(halt_poll_ns, uint, 0644);
77 EXPORT_SYMBOL_GPL(halt_poll_ns);
78
79 /* Default doubles per-vcpu halt_poll_ns. */
80 unsigned int halt_poll_ns_grow = 2;
81 module_param(halt_poll_ns_grow, uint, 0644);
82 EXPORT_SYMBOL_GPL(halt_poll_ns_grow);
83
84 /* Default resets per-vcpu halt_poll_ns . */
85 unsigned int halt_poll_ns_shrink;
86 module_param(halt_poll_ns_shrink, uint, 0644);
87 EXPORT_SYMBOL_GPL(halt_poll_ns_shrink);
88
89 /*
90  * Ordering of locks:
91  *
92  *      kvm->lock --> kvm->slots_lock --> kvm->irq_lock
93  */
94
95 DEFINE_SPINLOCK(kvm_lock);
96 static DEFINE_RAW_SPINLOCK(kvm_count_lock);
97 LIST_HEAD(vm_list);
98
99 static cpumask_var_t cpus_hardware_enabled;
100 static int kvm_usage_count;
101 static atomic_t hardware_enable_failed;
102
103 struct kmem_cache *kvm_vcpu_cache;
104 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
105
106 static __read_mostly struct preempt_ops kvm_preempt_ops;
107
108 struct dentry *kvm_debugfs_dir;
109 EXPORT_SYMBOL_GPL(kvm_debugfs_dir);
110
111 static int kvm_debugfs_num_entries;
112 static const struct file_operations *stat_fops_per_vm[];
113
114 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
115                            unsigned long arg);
116 #ifdef CONFIG_KVM_COMPAT
117 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
118                                   unsigned long arg);
119 #endif
120 static int hardware_enable_all(void);
121 static void hardware_disable_all(void);
122
123 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
124
125 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot, gfn_t gfn);
126
127 __visible bool kvm_rebooting;
128 EXPORT_SYMBOL_GPL(kvm_rebooting);
129
130 static bool largepages_enabled = true;
131
132 #define KVM_EVENT_CREATE_VM 0
133 #define KVM_EVENT_DESTROY_VM 1
134 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm);
135 static unsigned long long kvm_createvm_count;
136 static unsigned long long kvm_active_vms;
137
138 __weak void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
139                 unsigned long start, unsigned long end)
140 {
141 }
142
143 bool kvm_is_reserved_pfn(kvm_pfn_t pfn)
144 {
145         if (pfn_valid(pfn))
146                 return PageReserved(pfn_to_page(pfn));
147
148         return true;
149 }
150
151 /*
152  * Switches to specified vcpu, until a matching vcpu_put()
153  */
154 void vcpu_load(struct kvm_vcpu *vcpu)
155 {
156         int cpu = get_cpu();
157         preempt_notifier_register(&vcpu->preempt_notifier);
158         kvm_arch_vcpu_load(vcpu, cpu);
159         put_cpu();
160 }
161 EXPORT_SYMBOL_GPL(vcpu_load);
162
163 void vcpu_put(struct kvm_vcpu *vcpu)
164 {
165         preempt_disable();
166         kvm_arch_vcpu_put(vcpu);
167         preempt_notifier_unregister(&vcpu->preempt_notifier);
168         preempt_enable();
169 }
170 EXPORT_SYMBOL_GPL(vcpu_put);
171
172 /* TODO: merge with kvm_arch_vcpu_should_kick */
173 static bool kvm_request_needs_ipi(struct kvm_vcpu *vcpu, unsigned req)
174 {
175         int mode = kvm_vcpu_exiting_guest_mode(vcpu);
176
177         /*
178          * We need to wait for the VCPU to reenable interrupts and get out of
179          * READING_SHADOW_PAGE_TABLES mode.
180          */
181         if (req & KVM_REQUEST_WAIT)
182                 return mode != OUTSIDE_GUEST_MODE;
183
184         /*
185          * Need to kick a running VCPU, but otherwise there is nothing to do.
186          */
187         return mode == IN_GUEST_MODE;
188 }
189
190 static void ack_flush(void *_completed)
191 {
192 }
193
194 static inline bool kvm_kick_many_cpus(const struct cpumask *cpus, bool wait)
195 {
196         if (unlikely(!cpus))
197                 cpus = cpu_online_mask;
198
199         if (cpumask_empty(cpus))
200                 return false;
201
202         smp_call_function_many(cpus, ack_flush, NULL, wait);
203         return true;
204 }
205
206 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
207 {
208         int i, cpu, me;
209         cpumask_var_t cpus;
210         bool called;
211         struct kvm_vcpu *vcpu;
212
213         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
214
215         me = get_cpu();
216         kvm_for_each_vcpu(i, vcpu, kvm) {
217                 kvm_make_request(req, vcpu);
218                 cpu = vcpu->cpu;
219
220                 if (!(req & KVM_REQUEST_NO_WAKEUP) && kvm_vcpu_wake_up(vcpu))
221                         continue;
222
223                 if (cpus != NULL && cpu != -1 && cpu != me &&
224                     kvm_request_needs_ipi(vcpu, req))
225                         __cpumask_set_cpu(cpu, cpus);
226         }
227         called = kvm_kick_many_cpus(cpus, !!(req & KVM_REQUEST_WAIT));
228         put_cpu();
229         free_cpumask_var(cpus);
230         return called;
231 }
232
233 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
234 void kvm_flush_remote_tlbs(struct kvm *kvm)
235 {
236         /*
237          * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in
238          * kvm_make_all_cpus_request.
239          */
240         long dirty_count = smp_load_acquire(&kvm->tlbs_dirty);
241
242         /*
243          * We want to publish modifications to the page tables before reading
244          * mode. Pairs with a memory barrier in arch-specific code.
245          * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
246          * and smp_mb in walk_shadow_page_lockless_begin/end.
247          * - powerpc: smp_mb in kvmppc_prepare_to_enter.
248          *
249          * There is already an smp_mb__after_atomic() before
250          * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
251          * barrier here.
252          */
253         if (kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
254                 ++kvm->stat.remote_tlb_flush;
255         cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
256 }
257 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
258 #endif
259
260 void kvm_reload_remote_mmus(struct kvm *kvm)
261 {
262         kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
263 }
264
265 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
266 {
267         struct page *page;
268         int r;
269
270         mutex_init(&vcpu->mutex);
271         vcpu->cpu = -1;
272         vcpu->kvm = kvm;
273         vcpu->vcpu_id = id;
274         vcpu->pid = NULL;
275         init_swait_queue_head(&vcpu->wq);
276         kvm_async_pf_vcpu_init(vcpu);
277
278         vcpu->pre_pcpu = -1;
279         INIT_LIST_HEAD(&vcpu->blocked_vcpu_list);
280
281         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
282         if (!page) {
283                 r = -ENOMEM;
284                 goto fail;
285         }
286         vcpu->run = page_address(page);
287
288         kvm_vcpu_set_in_spin_loop(vcpu, false);
289         kvm_vcpu_set_dy_eligible(vcpu, false);
290         vcpu->preempted = false;
291
292         r = kvm_arch_vcpu_init(vcpu);
293         if (r < 0)
294                 goto fail_free_run;
295         return 0;
296
297 fail_free_run:
298         free_page((unsigned long)vcpu->run);
299 fail:
300         return r;
301 }
302 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
303
304 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
305 {
306         /*
307          * no need for rcu_read_lock as VCPU_RUN is the only place that
308          * will change the vcpu->pid pointer and on uninit all file
309          * descriptors are already gone.
310          */
311         put_pid(rcu_dereference_protected(vcpu->pid, 1));
312         kvm_arch_vcpu_uninit(vcpu);
313         free_page((unsigned long)vcpu->run);
314 }
315 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
316
317 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
318 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
319 {
320         return container_of(mn, struct kvm, mmu_notifier);
321 }
322
323 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
324                                         struct mm_struct *mm,
325                                         unsigned long address,
326                                         pte_t pte)
327 {
328         struct kvm *kvm = mmu_notifier_to_kvm(mn);
329         int idx;
330
331         idx = srcu_read_lock(&kvm->srcu);
332         spin_lock(&kvm->mmu_lock);
333         kvm->mmu_notifier_seq++;
334         kvm_set_spte_hva(kvm, address, pte);
335         spin_unlock(&kvm->mmu_lock);
336         srcu_read_unlock(&kvm->srcu, idx);
337 }
338
339 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
340                                                     struct mm_struct *mm,
341                                                     unsigned long start,
342                                                     unsigned long end)
343 {
344         struct kvm *kvm = mmu_notifier_to_kvm(mn);
345         int need_tlb_flush = 0, idx;
346
347         idx = srcu_read_lock(&kvm->srcu);
348         spin_lock(&kvm->mmu_lock);
349         /*
350          * The count increase must become visible at unlock time as no
351          * spte can be established without taking the mmu_lock and
352          * count is also read inside the mmu_lock critical section.
353          */
354         kvm->mmu_notifier_count++;
355         need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
356         need_tlb_flush |= kvm->tlbs_dirty;
357         /* we've to flush the tlb before the pages can be freed */
358         if (need_tlb_flush)
359                 kvm_flush_remote_tlbs(kvm);
360
361         spin_unlock(&kvm->mmu_lock);
362
363         kvm_arch_mmu_notifier_invalidate_range(kvm, start, end);
364
365         srcu_read_unlock(&kvm->srcu, idx);
366 }
367
368 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
369                                                   struct mm_struct *mm,
370                                                   unsigned long start,
371                                                   unsigned long end)
372 {
373         struct kvm *kvm = mmu_notifier_to_kvm(mn);
374
375         spin_lock(&kvm->mmu_lock);
376         /*
377          * This sequence increase will notify the kvm page fault that
378          * the page that is going to be mapped in the spte could have
379          * been freed.
380          */
381         kvm->mmu_notifier_seq++;
382         smp_wmb();
383         /*
384          * The above sequence increase must be visible before the
385          * below count decrease, which is ensured by the smp_wmb above
386          * in conjunction with the smp_rmb in mmu_notifier_retry().
387          */
388         kvm->mmu_notifier_count--;
389         spin_unlock(&kvm->mmu_lock);
390
391         BUG_ON(kvm->mmu_notifier_count < 0);
392 }
393
394 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
395                                               struct mm_struct *mm,
396                                               unsigned long start,
397                                               unsigned long end)
398 {
399         struct kvm *kvm = mmu_notifier_to_kvm(mn);
400         int young, idx;
401
402         idx = srcu_read_lock(&kvm->srcu);
403         spin_lock(&kvm->mmu_lock);
404
405         young = kvm_age_hva(kvm, start, end);
406         if (young)
407                 kvm_flush_remote_tlbs(kvm);
408
409         spin_unlock(&kvm->mmu_lock);
410         srcu_read_unlock(&kvm->srcu, idx);
411
412         return young;
413 }
414
415 static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
416                                         struct mm_struct *mm,
417                                         unsigned long start,
418                                         unsigned long end)
419 {
420         struct kvm *kvm = mmu_notifier_to_kvm(mn);
421         int young, idx;
422
423         idx = srcu_read_lock(&kvm->srcu);
424         spin_lock(&kvm->mmu_lock);
425         /*
426          * Even though we do not flush TLB, this will still adversely
427          * affect performance on pre-Haswell Intel EPT, where there is
428          * no EPT Access Bit to clear so that we have to tear down EPT
429          * tables instead. If we find this unacceptable, we can always
430          * add a parameter to kvm_age_hva so that it effectively doesn't
431          * do anything on clear_young.
432          *
433          * Also note that currently we never issue secondary TLB flushes
434          * from clear_young, leaving this job up to the regular system
435          * cadence. If we find this inaccurate, we might come up with a
436          * more sophisticated heuristic later.
437          */
438         young = kvm_age_hva(kvm, start, end);
439         spin_unlock(&kvm->mmu_lock);
440         srcu_read_unlock(&kvm->srcu, idx);
441
442         return young;
443 }
444
445 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
446                                        struct mm_struct *mm,
447                                        unsigned long address)
448 {
449         struct kvm *kvm = mmu_notifier_to_kvm(mn);
450         int young, idx;
451
452         idx = srcu_read_lock(&kvm->srcu);
453         spin_lock(&kvm->mmu_lock);
454         young = kvm_test_age_hva(kvm, address);
455         spin_unlock(&kvm->mmu_lock);
456         srcu_read_unlock(&kvm->srcu, idx);
457
458         return young;
459 }
460
461 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
462                                      struct mm_struct *mm)
463 {
464         struct kvm *kvm = mmu_notifier_to_kvm(mn);
465         int idx;
466
467         idx = srcu_read_lock(&kvm->srcu);
468         kvm_arch_flush_shadow_all(kvm);
469         srcu_read_unlock(&kvm->srcu, idx);
470 }
471
472 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
473         .flags                  = MMU_INVALIDATE_DOES_NOT_BLOCK,
474         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
475         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
476         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
477         .clear_young            = kvm_mmu_notifier_clear_young,
478         .test_young             = kvm_mmu_notifier_test_young,
479         .change_pte             = kvm_mmu_notifier_change_pte,
480         .release                = kvm_mmu_notifier_release,
481 };
482
483 static int kvm_init_mmu_notifier(struct kvm *kvm)
484 {
485         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
486         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
487 }
488
489 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
490
491 static int kvm_init_mmu_notifier(struct kvm *kvm)
492 {
493         return 0;
494 }
495
496 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
497
498 static struct kvm_memslots *kvm_alloc_memslots(void)
499 {
500         int i;
501         struct kvm_memslots *slots;
502
503         slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
504         if (!slots)
505                 return NULL;
506
507         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
508                 slots->id_to_index[i] = slots->memslots[i].id = i;
509
510         return slots;
511 }
512
513 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
514 {
515         if (!memslot->dirty_bitmap)
516                 return;
517
518         kvfree(memslot->dirty_bitmap);
519         memslot->dirty_bitmap = NULL;
520 }
521
522 /*
523  * Free any memory in @free but not in @dont.
524  */
525 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
526                               struct kvm_memory_slot *dont)
527 {
528         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
529                 kvm_destroy_dirty_bitmap(free);
530
531         kvm_arch_free_memslot(kvm, free, dont);
532
533         free->npages = 0;
534 }
535
536 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
537 {
538         struct kvm_memory_slot *memslot;
539
540         if (!slots)
541                 return;
542
543         kvm_for_each_memslot(memslot, slots)
544                 kvm_free_memslot(kvm, memslot, NULL);
545
546         kvfree(slots);
547 }
548
549 static void kvm_destroy_vm_debugfs(struct kvm *kvm)
550 {
551         int i;
552
553         if (!kvm->debugfs_dentry)
554                 return;
555
556         debugfs_remove_recursive(kvm->debugfs_dentry);
557
558         if (kvm->debugfs_stat_data) {
559                 for (i = 0; i < kvm_debugfs_num_entries; i++)
560                         kfree(kvm->debugfs_stat_data[i]);
561                 kfree(kvm->debugfs_stat_data);
562         }
563 }
564
565 static int kvm_create_vm_debugfs(struct kvm *kvm, int fd)
566 {
567         char dir_name[ITOA_MAX_LEN * 2];
568         struct kvm_stat_data *stat_data;
569         struct kvm_stats_debugfs_item *p;
570
571         if (!debugfs_initialized())
572                 return 0;
573
574         snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd);
575         kvm->debugfs_dentry = debugfs_create_dir(dir_name,
576                                                  kvm_debugfs_dir);
577         if (!kvm->debugfs_dentry)
578                 return -ENOMEM;
579
580         kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries,
581                                          sizeof(*kvm->debugfs_stat_data),
582                                          GFP_KERNEL);
583         if (!kvm->debugfs_stat_data)
584                 return -ENOMEM;
585
586         for (p = debugfs_entries; p->name; p++) {
587                 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL);
588                 if (!stat_data)
589                         return -ENOMEM;
590
591                 stat_data->kvm = kvm;
592                 stat_data->offset = p->offset;
593                 kvm->debugfs_stat_data[p - debugfs_entries] = stat_data;
594                 if (!debugfs_create_file(p->name, 0644,
595                                          kvm->debugfs_dentry,
596                                          stat_data,
597                                          stat_fops_per_vm[p->kind]))
598                         return -ENOMEM;
599         }
600         return 0;
601 }
602
603 static struct kvm *kvm_create_vm(unsigned long type)
604 {
605         int r, i;
606         struct kvm *kvm = kvm_arch_alloc_vm();
607
608         if (!kvm)
609                 return ERR_PTR(-ENOMEM);
610
611         spin_lock_init(&kvm->mmu_lock);
612         mmgrab(current->mm);
613         kvm->mm = current->mm;
614         kvm_eventfd_init(kvm);
615         mutex_init(&kvm->lock);
616         mutex_init(&kvm->irq_lock);
617         mutex_init(&kvm->slots_lock);
618         refcount_set(&kvm->users_count, 1);
619         INIT_LIST_HEAD(&kvm->devices);
620
621         r = kvm_arch_init_vm(kvm, type);
622         if (r)
623                 goto out_err_no_disable;
624
625         r = hardware_enable_all();
626         if (r)
627                 goto out_err_no_disable;
628
629 #ifdef CONFIG_HAVE_KVM_IRQFD
630         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
631 #endif
632
633         BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
634
635         r = -ENOMEM;
636         for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
637                 struct kvm_memslots *slots = kvm_alloc_memslots();
638                 if (!slots)
639                         goto out_err_no_srcu;
640                 /*
641                  * Generations must be different for each address space.
642                  * Init kvm generation close to the maximum to easily test the
643                  * code of handling generation number wrap-around.
644                  */
645                 slots->generation = i * 2 - 150;
646                 rcu_assign_pointer(kvm->memslots[i], slots);
647         }
648
649         if (init_srcu_struct(&kvm->srcu))
650                 goto out_err_no_srcu;
651         if (init_srcu_struct(&kvm->irq_srcu))
652                 goto out_err_no_irq_srcu;
653         for (i = 0; i < KVM_NR_BUSES; i++) {
654                 rcu_assign_pointer(kvm->buses[i],
655                         kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL));
656                 if (!kvm->buses[i])
657                         goto out_err;
658         }
659
660         r = kvm_init_mmu_notifier(kvm);
661         if (r)
662                 goto out_err;
663
664         spin_lock(&kvm_lock);
665         list_add(&kvm->vm_list, &vm_list);
666         spin_unlock(&kvm_lock);
667
668         preempt_notifier_inc();
669
670         return kvm;
671
672 out_err:
673         cleanup_srcu_struct(&kvm->irq_srcu);
674 out_err_no_irq_srcu:
675         cleanup_srcu_struct(&kvm->srcu);
676 out_err_no_srcu:
677         hardware_disable_all();
678 out_err_no_disable:
679         refcount_set(&kvm->users_count, 0);
680         for (i = 0; i < KVM_NR_BUSES; i++)
681                 kfree(kvm_get_bus(kvm, i));
682         for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
683                 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
684         kvm_arch_free_vm(kvm);
685         mmdrop(current->mm);
686         return ERR_PTR(r);
687 }
688
689 static void kvm_destroy_devices(struct kvm *kvm)
690 {
691         struct kvm_device *dev, *tmp;
692
693         /*
694          * We do not need to take the kvm->lock here, because nobody else
695          * has a reference to the struct kvm at this point and therefore
696          * cannot access the devices list anyhow.
697          */
698         list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
699                 list_del(&dev->vm_node);
700                 dev->ops->destroy(dev);
701         }
702 }
703
704 static void kvm_destroy_vm(struct kvm *kvm)
705 {
706         int i;
707         struct mm_struct *mm = kvm->mm;
708
709         kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm);
710         kvm_destroy_vm_debugfs(kvm);
711         kvm_arch_sync_events(kvm);
712         spin_lock(&kvm_lock);
713         list_del(&kvm->vm_list);
714         spin_unlock(&kvm_lock);
715         kvm_free_irq_routing(kvm);
716         for (i = 0; i < KVM_NR_BUSES; i++) {
717                 struct kvm_io_bus *bus = kvm_get_bus(kvm, i);
718
719                 if (bus)
720                         kvm_io_bus_destroy(bus);
721                 kvm->buses[i] = NULL;
722         }
723         kvm_coalesced_mmio_free(kvm);
724 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
725         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
726 #else
727         kvm_arch_flush_shadow_all(kvm);
728 #endif
729         kvm_arch_destroy_vm(kvm);
730         kvm_destroy_devices(kvm);
731         for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
732                 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
733         cleanup_srcu_struct(&kvm->irq_srcu);
734         cleanup_srcu_struct(&kvm->srcu);
735         kvm_arch_free_vm(kvm);
736         preempt_notifier_dec();
737         hardware_disable_all();
738         mmdrop(mm);
739 }
740
741 void kvm_get_kvm(struct kvm *kvm)
742 {
743         refcount_inc(&kvm->users_count);
744 }
745 EXPORT_SYMBOL_GPL(kvm_get_kvm);
746
747 void kvm_put_kvm(struct kvm *kvm)
748 {
749         if (refcount_dec_and_test(&kvm->users_count))
750                 kvm_destroy_vm(kvm);
751 }
752 EXPORT_SYMBOL_GPL(kvm_put_kvm);
753
754
755 static int kvm_vm_release(struct inode *inode, struct file *filp)
756 {
757         struct kvm *kvm = filp->private_data;
758
759         kvm_irqfd_release(kvm);
760
761         kvm_put_kvm(kvm);
762         return 0;
763 }
764
765 /*
766  * Allocation size is twice as large as the actual dirty bitmap size.
767  * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
768  */
769 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
770 {
771         unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
772
773         memslot->dirty_bitmap = kvzalloc(dirty_bytes, GFP_KERNEL);
774         if (!memslot->dirty_bitmap)
775                 return -ENOMEM;
776
777         return 0;
778 }
779
780 /*
781  * Insert memslot and re-sort memslots based on their GFN,
782  * so binary search could be used to lookup GFN.
783  * Sorting algorithm takes advantage of having initially
784  * sorted array and known changed memslot position.
785  */
786 static void update_memslots(struct kvm_memslots *slots,
787                             struct kvm_memory_slot *new)
788 {
789         int id = new->id;
790         int i = slots->id_to_index[id];
791         struct kvm_memory_slot *mslots = slots->memslots;
792
793         WARN_ON(mslots[i].id != id);
794         if (!new->npages) {
795                 WARN_ON(!mslots[i].npages);
796                 if (mslots[i].npages)
797                         slots->used_slots--;
798         } else {
799                 if (!mslots[i].npages)
800                         slots->used_slots++;
801         }
802
803         while (i < KVM_MEM_SLOTS_NUM - 1 &&
804                new->base_gfn <= mslots[i + 1].base_gfn) {
805                 if (!mslots[i + 1].npages)
806                         break;
807                 mslots[i] = mslots[i + 1];
808                 slots->id_to_index[mslots[i].id] = i;
809                 i++;
810         }
811
812         /*
813          * The ">=" is needed when creating a slot with base_gfn == 0,
814          * so that it moves before all those with base_gfn == npages == 0.
815          *
816          * On the other hand, if new->npages is zero, the above loop has
817          * already left i pointing to the beginning of the empty part of
818          * mslots, and the ">=" would move the hole backwards in this
819          * case---which is wrong.  So skip the loop when deleting a slot.
820          */
821         if (new->npages) {
822                 while (i > 0 &&
823                        new->base_gfn >= mslots[i - 1].base_gfn) {
824                         mslots[i] = mslots[i - 1];
825                         slots->id_to_index[mslots[i].id] = i;
826                         i--;
827                 }
828         } else
829                 WARN_ON_ONCE(i != slots->used_slots);
830
831         mslots[i] = *new;
832         slots->id_to_index[mslots[i].id] = i;
833 }
834
835 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
836 {
837         u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
838
839 #ifdef __KVM_HAVE_READONLY_MEM
840         valid_flags |= KVM_MEM_READONLY;
841 #endif
842
843         if (mem->flags & ~valid_flags)
844                 return -EINVAL;
845
846         return 0;
847 }
848
849 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
850                 int as_id, struct kvm_memslots *slots)
851 {
852         struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id);
853
854         /*
855          * Set the low bit in the generation, which disables SPTE caching
856          * until the end of synchronize_srcu_expedited.
857          */
858         WARN_ON(old_memslots->generation & 1);
859         slots->generation = old_memslots->generation + 1;
860
861         rcu_assign_pointer(kvm->memslots[as_id], slots);
862         synchronize_srcu_expedited(&kvm->srcu);
863
864         /*
865          * Increment the new memslot generation a second time. This prevents
866          * vm exits that race with memslot updates from caching a memslot
867          * generation that will (potentially) be valid forever.
868          *
869          * Generations must be unique even across address spaces.  We do not need
870          * a global counter for that, instead the generation space is evenly split
871          * across address spaces.  For example, with two address spaces, address
872          * space 0 will use generations 0, 4, 8, ... while * address space 1 will
873          * use generations 2, 6, 10, 14, ...
874          */
875         slots->generation += KVM_ADDRESS_SPACE_NUM * 2 - 1;
876
877         kvm_arch_memslots_updated(kvm, slots);
878
879         return old_memslots;
880 }
881
882 /*
883  * Allocate some memory and give it an address in the guest physical address
884  * space.
885  *
886  * Discontiguous memory is allowed, mostly for framebuffers.
887  *
888  * Must be called holding kvm->slots_lock for write.
889  */
890 int __kvm_set_memory_region(struct kvm *kvm,
891                             const struct kvm_userspace_memory_region *mem)
892 {
893         int r;
894         gfn_t base_gfn;
895         unsigned long npages;
896         struct kvm_memory_slot *slot;
897         struct kvm_memory_slot old, new;
898         struct kvm_memslots *slots = NULL, *old_memslots;
899         int as_id, id;
900         enum kvm_mr_change change;
901
902         r = check_memory_region_flags(mem);
903         if (r)
904                 goto out;
905
906         r = -EINVAL;
907         as_id = mem->slot >> 16;
908         id = (u16)mem->slot;
909
910         /* General sanity checks */
911         if (mem->memory_size & (PAGE_SIZE - 1))
912                 goto out;
913         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
914                 goto out;
915         /* We can read the guest memory with __xxx_user() later on. */
916         if ((id < KVM_USER_MEM_SLOTS) &&
917             ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
918              !access_ok(VERIFY_WRITE,
919                         (void __user *)(unsigned long)mem->userspace_addr,
920                         mem->memory_size)))
921                 goto out;
922         if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
923                 goto out;
924         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
925                 goto out;
926
927         slot = id_to_memslot(__kvm_memslots(kvm, as_id), id);
928         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
929         npages = mem->memory_size >> PAGE_SHIFT;
930
931         if (npages > KVM_MEM_MAX_NR_PAGES)
932                 goto out;
933
934         new = old = *slot;
935
936         new.id = id;
937         new.base_gfn = base_gfn;
938         new.npages = npages;
939         new.flags = mem->flags;
940
941         if (npages) {
942                 if (!old.npages)
943                         change = KVM_MR_CREATE;
944                 else { /* Modify an existing slot. */
945                         if ((mem->userspace_addr != old.userspace_addr) ||
946                             (npages != old.npages) ||
947                             ((new.flags ^ old.flags) & KVM_MEM_READONLY))
948                                 goto out;
949
950                         if (base_gfn != old.base_gfn)
951                                 change = KVM_MR_MOVE;
952                         else if (new.flags != old.flags)
953                                 change = KVM_MR_FLAGS_ONLY;
954                         else { /* Nothing to change. */
955                                 r = 0;
956                                 goto out;
957                         }
958                 }
959         } else {
960                 if (!old.npages)
961                         goto out;
962
963                 change = KVM_MR_DELETE;
964                 new.base_gfn = 0;
965                 new.flags = 0;
966         }
967
968         if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
969                 /* Check for overlaps */
970                 r = -EEXIST;
971                 kvm_for_each_memslot(slot, __kvm_memslots(kvm, as_id)) {
972                         if ((slot->id >= KVM_USER_MEM_SLOTS) ||
973                             (slot->id == id))
974                                 continue;
975                         if (!((base_gfn + npages <= slot->base_gfn) ||
976                               (base_gfn >= slot->base_gfn + slot->npages)))
977                                 goto out;
978                 }
979         }
980
981         /* Free page dirty bitmap if unneeded */
982         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
983                 new.dirty_bitmap = NULL;
984
985         r = -ENOMEM;
986         if (change == KVM_MR_CREATE) {
987                 new.userspace_addr = mem->userspace_addr;
988
989                 if (kvm_arch_create_memslot(kvm, &new, npages))
990                         goto out_free;
991         }
992
993         /* Allocate page dirty bitmap if needed */
994         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
995                 if (kvm_create_dirty_bitmap(&new) < 0)
996                         goto out_free;
997         }
998
999         slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
1000         if (!slots)
1001                 goto out_free;
1002         memcpy(slots, __kvm_memslots(kvm, as_id), sizeof(struct kvm_memslots));
1003
1004         if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
1005                 slot = id_to_memslot(slots, id);
1006                 slot->flags |= KVM_MEMSLOT_INVALID;
1007
1008                 old_memslots = install_new_memslots(kvm, as_id, slots);
1009
1010                 /* From this point no new shadow pages pointing to a deleted,
1011                  * or moved, memslot will be created.
1012                  *
1013                  * validation of sp->gfn happens in:
1014                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1015                  *      - kvm_is_visible_gfn (mmu_check_roots)
1016                  */
1017                 kvm_arch_flush_shadow_memslot(kvm, slot);
1018
1019                 /*
1020                  * We can re-use the old_memslots from above, the only difference
1021                  * from the currently installed memslots is the invalid flag.  This
1022                  * will get overwritten by update_memslots anyway.
1023                  */
1024                 slots = old_memslots;
1025         }
1026
1027         r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
1028         if (r)
1029                 goto out_slots;
1030
1031         /* actual memory is freed via old in kvm_free_memslot below */
1032         if (change == KVM_MR_DELETE) {
1033                 new.dirty_bitmap = NULL;
1034                 memset(&new.arch, 0, sizeof(new.arch));
1035         }
1036
1037         update_memslots(slots, &new);
1038         old_memslots = install_new_memslots(kvm, as_id, slots);
1039
1040         kvm_arch_commit_memory_region(kvm, mem, &old, &new, change);
1041
1042         kvm_free_memslot(kvm, &old, &new);
1043         kvfree(old_memslots);
1044         return 0;
1045
1046 out_slots:
1047         kvfree(slots);
1048 out_free:
1049         kvm_free_memslot(kvm, &new, &old);
1050 out:
1051         return r;
1052 }
1053 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1054
1055 int kvm_set_memory_region(struct kvm *kvm,
1056                           const struct kvm_userspace_memory_region *mem)
1057 {
1058         int r;
1059
1060         mutex_lock(&kvm->slots_lock);
1061         r = __kvm_set_memory_region(kvm, mem);
1062         mutex_unlock(&kvm->slots_lock);
1063         return r;
1064 }
1065 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1066
1067 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1068                                           struct kvm_userspace_memory_region *mem)
1069 {
1070         if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1071                 return -EINVAL;
1072
1073         return kvm_set_memory_region(kvm, mem);
1074 }
1075
1076 int kvm_get_dirty_log(struct kvm *kvm,
1077                         struct kvm_dirty_log *log, int *is_dirty)
1078 {
1079         struct kvm_memslots *slots;
1080         struct kvm_memory_slot *memslot;
1081         int i, as_id, id;
1082         unsigned long n;
1083         unsigned long any = 0;
1084
1085         as_id = log->slot >> 16;
1086         id = (u16)log->slot;
1087         if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1088                 return -EINVAL;
1089
1090         slots = __kvm_memslots(kvm, as_id);
1091         memslot = id_to_memslot(slots, id);
1092         if (!memslot->dirty_bitmap)
1093                 return -ENOENT;
1094
1095         n = kvm_dirty_bitmap_bytes(memslot);
1096
1097         for (i = 0; !any && i < n/sizeof(long); ++i)
1098                 any = memslot->dirty_bitmap[i];
1099
1100         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1101                 return -EFAULT;
1102
1103         if (any)
1104                 *is_dirty = 1;
1105         return 0;
1106 }
1107 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1108
1109 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1110 /**
1111  * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages
1112  *      are dirty write protect them for next write.
1113  * @kvm:        pointer to kvm instance
1114  * @log:        slot id and address to which we copy the log
1115  * @is_dirty:   flag set if any page is dirty
1116  *
1117  * We need to keep it in mind that VCPU threads can write to the bitmap
1118  * concurrently. So, to avoid losing track of dirty pages we keep the
1119  * following order:
1120  *
1121  *    1. Take a snapshot of the bit and clear it if needed.
1122  *    2. Write protect the corresponding page.
1123  *    3. Copy the snapshot to the userspace.
1124  *    4. Upon return caller flushes TLB's if needed.
1125  *
1126  * Between 2 and 4, the guest may write to the page using the remaining TLB
1127  * entry.  This is not a problem because the page is reported dirty using
1128  * the snapshot taken before and step 4 ensures that writes done after
1129  * exiting to userspace will be logged for the next call.
1130  *
1131  */
1132 int kvm_get_dirty_log_protect(struct kvm *kvm,
1133                         struct kvm_dirty_log *log, bool *is_dirty)
1134 {
1135         struct kvm_memslots *slots;
1136         struct kvm_memory_slot *memslot;
1137         int i, as_id, id;
1138         unsigned long n;
1139         unsigned long *dirty_bitmap;
1140         unsigned long *dirty_bitmap_buffer;
1141
1142         as_id = log->slot >> 16;
1143         id = (u16)log->slot;
1144         if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1145                 return -EINVAL;
1146
1147         slots = __kvm_memslots(kvm, as_id);
1148         memslot = id_to_memslot(slots, id);
1149
1150         dirty_bitmap = memslot->dirty_bitmap;
1151         if (!dirty_bitmap)
1152                 return -ENOENT;
1153
1154         n = kvm_dirty_bitmap_bytes(memslot);
1155
1156         dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long);
1157         memset(dirty_bitmap_buffer, 0, n);
1158
1159         spin_lock(&kvm->mmu_lock);
1160         *is_dirty = false;
1161         for (i = 0; i < n / sizeof(long); i++) {
1162                 unsigned long mask;
1163                 gfn_t offset;
1164
1165                 if (!dirty_bitmap[i])
1166                         continue;
1167
1168                 *is_dirty = true;
1169
1170                 mask = xchg(&dirty_bitmap[i], 0);
1171                 dirty_bitmap_buffer[i] = mask;
1172
1173                 if (mask) {
1174                         offset = i * BITS_PER_LONG;
1175                         kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1176                                                                 offset, mask);
1177                 }
1178         }
1179
1180         spin_unlock(&kvm->mmu_lock);
1181         if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1182                 return -EFAULT;
1183         return 0;
1184 }
1185 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect);
1186 #endif
1187
1188 bool kvm_largepages_enabled(void)
1189 {
1190         return largepages_enabled;
1191 }
1192
1193 void kvm_disable_largepages(void)
1194 {
1195         largepages_enabled = false;
1196 }
1197 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1198
1199 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1200 {
1201         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1202 }
1203 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1204
1205 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
1206 {
1207         return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn);
1208 }
1209
1210 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1211 {
1212         struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1213
1214         if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1215               memslot->flags & KVM_MEMSLOT_INVALID)
1216                 return false;
1217
1218         return true;
1219 }
1220 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1221
1222 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1223 {
1224         struct vm_area_struct *vma;
1225         unsigned long addr, size;
1226
1227         size = PAGE_SIZE;
1228
1229         addr = gfn_to_hva(kvm, gfn);
1230         if (kvm_is_error_hva(addr))
1231                 return PAGE_SIZE;
1232
1233         down_read(&current->mm->mmap_sem);
1234         vma = find_vma(current->mm, addr);
1235         if (!vma)
1236                 goto out;
1237
1238         size = vma_kernel_pagesize(vma);
1239
1240 out:
1241         up_read(&current->mm->mmap_sem);
1242
1243         return size;
1244 }
1245
1246 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1247 {
1248         return slot->flags & KVM_MEM_READONLY;
1249 }
1250
1251 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1252                                        gfn_t *nr_pages, bool write)
1253 {
1254         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1255                 return KVM_HVA_ERR_BAD;
1256
1257         if (memslot_is_readonly(slot) && write)
1258                 return KVM_HVA_ERR_RO_BAD;
1259
1260         if (nr_pages)
1261                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1262
1263         return __gfn_to_hva_memslot(slot, gfn);
1264 }
1265
1266 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1267                                      gfn_t *nr_pages)
1268 {
1269         return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1270 }
1271
1272 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1273                                         gfn_t gfn)
1274 {
1275         return gfn_to_hva_many(slot, gfn, NULL);
1276 }
1277 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1278
1279 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1280 {
1281         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1282 }
1283 EXPORT_SYMBOL_GPL(gfn_to_hva);
1284
1285 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
1286 {
1287         return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
1288 }
1289 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
1290
1291 /*
1292  * If writable is set to false, the hva returned by this function is only
1293  * allowed to be read.
1294  */
1295 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1296                                       gfn_t gfn, bool *writable)
1297 {
1298         unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1299
1300         if (!kvm_is_error_hva(hva) && writable)
1301                 *writable = !memslot_is_readonly(slot);
1302
1303         return hva;
1304 }
1305
1306 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1307 {
1308         struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1309
1310         return gfn_to_hva_memslot_prot(slot, gfn, writable);
1311 }
1312
1313 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
1314 {
1315         struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1316
1317         return gfn_to_hva_memslot_prot(slot, gfn, writable);
1318 }
1319
1320 static inline int check_user_page_hwpoison(unsigned long addr)
1321 {
1322         int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
1323
1324         rc = get_user_pages(addr, 1, flags, NULL, NULL);
1325         return rc == -EHWPOISON;
1326 }
1327
1328 /*
1329  * The atomic path to get the writable pfn which will be stored in @pfn,
1330  * true indicates success, otherwise false is returned.
1331  */
1332 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1333                             bool write_fault, bool *writable, kvm_pfn_t *pfn)
1334 {
1335         struct page *page[1];
1336         int npages;
1337
1338         if (!(async || atomic))
1339                 return false;
1340
1341         /*
1342          * Fast pin a writable pfn only if it is a write fault request
1343          * or the caller allows to map a writable pfn for a read fault
1344          * request.
1345          */
1346         if (!(write_fault || writable))
1347                 return false;
1348
1349         npages = __get_user_pages_fast(addr, 1, 1, page);
1350         if (npages == 1) {
1351                 *pfn = page_to_pfn(page[0]);
1352
1353                 if (writable)
1354                         *writable = true;
1355                 return true;
1356         }
1357
1358         return false;
1359 }
1360
1361 /*
1362  * The slow path to get the pfn of the specified host virtual address,
1363  * 1 indicates success, -errno is returned if error is detected.
1364  */
1365 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1366                            bool *writable, kvm_pfn_t *pfn)
1367 {
1368         unsigned int flags = FOLL_HWPOISON;
1369         struct page *page;
1370         int npages = 0;
1371
1372         might_sleep();
1373
1374         if (writable)
1375                 *writable = write_fault;
1376
1377         if (write_fault)
1378                 flags |= FOLL_WRITE;
1379         if (async)
1380                 flags |= FOLL_NOWAIT;
1381
1382         npages = get_user_pages_unlocked(addr, 1, &page, flags);
1383         if (npages != 1)
1384                 return npages;
1385
1386         /* map read fault as writable if possible */
1387         if (unlikely(!write_fault) && writable) {
1388                 struct page *wpage;
1389
1390                 if (__get_user_pages_fast(addr, 1, 1, &wpage) == 1) {
1391                         *writable = true;
1392                         put_page(page);
1393                         page = wpage;
1394                 }
1395         }
1396         *pfn = page_to_pfn(page);
1397         return npages;
1398 }
1399
1400 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1401 {
1402         if (unlikely(!(vma->vm_flags & VM_READ)))
1403                 return false;
1404
1405         if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1406                 return false;
1407
1408         return true;
1409 }
1410
1411 static int hva_to_pfn_remapped(struct vm_area_struct *vma,
1412                                unsigned long addr, bool *async,
1413                                bool write_fault, bool *writable,
1414                                kvm_pfn_t *p_pfn)
1415 {
1416         unsigned long pfn;
1417         int r;
1418
1419         r = follow_pfn(vma, addr, &pfn);
1420         if (r) {
1421                 /*
1422                  * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
1423                  * not call the fault handler, so do it here.
1424                  */
1425                 bool unlocked = false;
1426                 r = fixup_user_fault(current, current->mm, addr,
1427                                      (write_fault ? FAULT_FLAG_WRITE : 0),
1428                                      &unlocked);
1429                 if (unlocked)
1430                         return -EAGAIN;
1431                 if (r)
1432                         return r;
1433
1434                 r = follow_pfn(vma, addr, &pfn);
1435                 if (r)
1436                         return r;
1437
1438         }
1439
1440         if (writable)
1441                 *writable = true;
1442
1443         /*
1444          * Get a reference here because callers of *hva_to_pfn* and
1445          * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
1446          * returned pfn.  This is only needed if the VMA has VM_MIXEDMAP
1447          * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will
1448          * simply do nothing for reserved pfns.
1449          *
1450          * Whoever called remap_pfn_range is also going to call e.g.
1451          * unmap_mapping_range before the underlying pages are freed,
1452          * causing a call to our MMU notifier.
1453          */ 
1454         kvm_get_pfn(pfn);
1455
1456         *p_pfn = pfn;
1457         return 0;
1458 }
1459
1460 /*
1461  * Pin guest page in memory and return its pfn.
1462  * @addr: host virtual address which maps memory to the guest
1463  * @atomic: whether this function can sleep
1464  * @async: whether this function need to wait IO complete if the
1465  *         host page is not in the memory
1466  * @write_fault: whether we should get a writable host page
1467  * @writable: whether it allows to map a writable host page for !@write_fault
1468  *
1469  * The function will map a writable host page for these two cases:
1470  * 1): @write_fault = true
1471  * 2): @write_fault = false && @writable, @writable will tell the caller
1472  *     whether the mapping is writable.
1473  */
1474 static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1475                         bool write_fault, bool *writable)
1476 {
1477         struct vm_area_struct *vma;
1478         kvm_pfn_t pfn = 0;
1479         int npages, r;
1480
1481         /* we can do it either atomically or asynchronously, not both */
1482         BUG_ON(atomic && async);
1483
1484         if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1485                 return pfn;
1486
1487         if (atomic)
1488                 return KVM_PFN_ERR_FAULT;
1489
1490         npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1491         if (npages == 1)
1492                 return pfn;
1493
1494         down_read(&current->mm->mmap_sem);
1495         if (npages == -EHWPOISON ||
1496               (!async && check_user_page_hwpoison(addr))) {
1497                 pfn = KVM_PFN_ERR_HWPOISON;
1498                 goto exit;
1499         }
1500
1501 retry:
1502         vma = find_vma_intersection(current->mm, addr, addr + 1);
1503
1504         if (vma == NULL)
1505                 pfn = KVM_PFN_ERR_FAULT;
1506         else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
1507                 r = hva_to_pfn_remapped(vma, addr, async, write_fault, writable, &pfn);
1508                 if (r == -EAGAIN)
1509                         goto retry;
1510                 if (r < 0)
1511                         pfn = KVM_PFN_ERR_FAULT;
1512         } else {
1513                 if (async && vma_is_valid(vma, write_fault))
1514                         *async = true;
1515                 pfn = KVM_PFN_ERR_FAULT;
1516         }
1517 exit:
1518         up_read(&current->mm->mmap_sem);
1519         return pfn;
1520 }
1521
1522 kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn,
1523                                bool atomic, bool *async, bool write_fault,
1524                                bool *writable)
1525 {
1526         unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1527
1528         if (addr == KVM_HVA_ERR_RO_BAD) {
1529                 if (writable)
1530                         *writable = false;
1531                 return KVM_PFN_ERR_RO_FAULT;
1532         }
1533
1534         if (kvm_is_error_hva(addr)) {
1535                 if (writable)
1536                         *writable = false;
1537                 return KVM_PFN_NOSLOT;
1538         }
1539
1540         /* Do not map writable pfn in the readonly memslot. */
1541         if (writable && memslot_is_readonly(slot)) {
1542                 *writable = false;
1543                 writable = NULL;
1544         }
1545
1546         return hva_to_pfn(addr, atomic, async, write_fault,
1547                           writable);
1548 }
1549 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
1550
1551 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1552                       bool *writable)
1553 {
1554         return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
1555                                     write_fault, writable);
1556 }
1557 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1558
1559 kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1560 {
1561         return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1562 }
1563 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
1564
1565 kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1566 {
1567         return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1568 }
1569 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1570
1571 kvm_pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1572 {
1573         return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm, gfn), gfn);
1574 }
1575 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1576
1577 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
1578 {
1579         return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1580 }
1581 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
1582
1583 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1584 {
1585         return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
1586 }
1587 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1588
1589 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
1590 {
1591         return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1592 }
1593 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
1594
1595 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1596                             struct page **pages, int nr_pages)
1597 {
1598         unsigned long addr;
1599         gfn_t entry = 0;
1600
1601         addr = gfn_to_hva_many(slot, gfn, &entry);
1602         if (kvm_is_error_hva(addr))
1603                 return -1;
1604
1605         if (entry < nr_pages)
1606                 return 0;
1607
1608         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1609 }
1610 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1611
1612 static struct page *kvm_pfn_to_page(kvm_pfn_t pfn)
1613 {
1614         if (is_error_noslot_pfn(pfn))
1615                 return KVM_ERR_PTR_BAD_PAGE;
1616
1617         if (kvm_is_reserved_pfn(pfn)) {
1618                 WARN_ON(1);
1619                 return KVM_ERR_PTR_BAD_PAGE;
1620         }
1621
1622         return pfn_to_page(pfn);
1623 }
1624
1625 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1626 {
1627         kvm_pfn_t pfn;
1628
1629         pfn = gfn_to_pfn(kvm, gfn);
1630
1631         return kvm_pfn_to_page(pfn);
1632 }
1633 EXPORT_SYMBOL_GPL(gfn_to_page);
1634
1635 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
1636 {
1637         kvm_pfn_t pfn;
1638
1639         pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
1640
1641         return kvm_pfn_to_page(pfn);
1642 }
1643 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
1644
1645 void kvm_release_page_clean(struct page *page)
1646 {
1647         WARN_ON(is_error_page(page));
1648
1649         kvm_release_pfn_clean(page_to_pfn(page));
1650 }
1651 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1652
1653 void kvm_release_pfn_clean(kvm_pfn_t pfn)
1654 {
1655         if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
1656                 put_page(pfn_to_page(pfn));
1657 }
1658 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1659
1660 void kvm_release_page_dirty(struct page *page)
1661 {
1662         WARN_ON(is_error_page(page));
1663
1664         kvm_release_pfn_dirty(page_to_pfn(page));
1665 }
1666 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1667
1668 void kvm_release_pfn_dirty(kvm_pfn_t pfn)
1669 {
1670         kvm_set_pfn_dirty(pfn);
1671         kvm_release_pfn_clean(pfn);
1672 }
1673 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1674
1675 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
1676 {
1677         if (!kvm_is_reserved_pfn(pfn)) {
1678                 struct page *page = pfn_to_page(pfn);
1679
1680                 if (!PageReserved(page))
1681                         SetPageDirty(page);
1682         }
1683 }
1684 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1685
1686 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
1687 {
1688         if (!kvm_is_reserved_pfn(pfn))
1689                 mark_page_accessed(pfn_to_page(pfn));
1690 }
1691 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1692
1693 void kvm_get_pfn(kvm_pfn_t pfn)
1694 {
1695         if (!kvm_is_reserved_pfn(pfn))
1696                 get_page(pfn_to_page(pfn));
1697 }
1698 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1699
1700 static int next_segment(unsigned long len, int offset)
1701 {
1702         if (len > PAGE_SIZE - offset)
1703                 return PAGE_SIZE - offset;
1704         else
1705                 return len;
1706 }
1707
1708 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
1709                                  void *data, int offset, int len)
1710 {
1711         int r;
1712         unsigned long addr;
1713
1714         addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1715         if (kvm_is_error_hva(addr))
1716                 return -EFAULT;
1717         r = __copy_from_user(data, (void __user *)addr + offset, len);
1718         if (r)
1719                 return -EFAULT;
1720         return 0;
1721 }
1722
1723 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1724                         int len)
1725 {
1726         struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1727
1728         return __kvm_read_guest_page(slot, gfn, data, offset, len);
1729 }
1730 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1731
1732 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
1733                              int offset, int len)
1734 {
1735         struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1736
1737         return __kvm_read_guest_page(slot, gfn, data, offset, len);
1738 }
1739 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
1740
1741 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1742 {
1743         gfn_t gfn = gpa >> PAGE_SHIFT;
1744         int seg;
1745         int offset = offset_in_page(gpa);
1746         int ret;
1747
1748         while ((seg = next_segment(len, offset)) != 0) {
1749                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1750                 if (ret < 0)
1751                         return ret;
1752                 offset = 0;
1753                 len -= seg;
1754                 data += seg;
1755                 ++gfn;
1756         }
1757         return 0;
1758 }
1759 EXPORT_SYMBOL_GPL(kvm_read_guest);
1760
1761 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
1762 {
1763         gfn_t gfn = gpa >> PAGE_SHIFT;
1764         int seg;
1765         int offset = offset_in_page(gpa);
1766         int ret;
1767
1768         while ((seg = next_segment(len, offset)) != 0) {
1769                 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
1770                 if (ret < 0)
1771                         return ret;
1772                 offset = 0;
1773                 len -= seg;
1774                 data += seg;
1775                 ++gfn;
1776         }
1777         return 0;
1778 }
1779 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
1780
1781 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1782                                    void *data, int offset, unsigned long len)
1783 {
1784         int r;
1785         unsigned long addr;
1786
1787         addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1788         if (kvm_is_error_hva(addr))
1789                 return -EFAULT;
1790         pagefault_disable();
1791         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1792         pagefault_enable();
1793         if (r)
1794                 return -EFAULT;
1795         return 0;
1796 }
1797
1798 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1799                           unsigned long len)
1800 {
1801         gfn_t gfn = gpa >> PAGE_SHIFT;
1802         struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1803         int offset = offset_in_page(gpa);
1804
1805         return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1806 }
1807 EXPORT_SYMBOL_GPL(kvm_read_guest_atomic);
1808
1809 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
1810                                void *data, unsigned long len)
1811 {
1812         gfn_t gfn = gpa >> PAGE_SHIFT;
1813         struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1814         int offset = offset_in_page(gpa);
1815
1816         return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1817 }
1818 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
1819
1820 static int __kvm_write_guest_page(struct kvm_memory_slot *memslot, gfn_t gfn,
1821                                   const void *data, int offset, int len)
1822 {
1823         int r;
1824         unsigned long addr;
1825
1826         addr = gfn_to_hva_memslot(memslot, gfn);
1827         if (kvm_is_error_hva(addr))
1828                 return -EFAULT;
1829         r = __copy_to_user((void __user *)addr + offset, data, len);
1830         if (r)
1831                 return -EFAULT;
1832         mark_page_dirty_in_slot(memslot, gfn);
1833         return 0;
1834 }
1835
1836 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
1837                          const void *data, int offset, int len)
1838 {
1839         struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1840
1841         return __kvm_write_guest_page(slot, gfn, data, offset, len);
1842 }
1843 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1844
1845 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
1846                               const void *data, int offset, int len)
1847 {
1848         struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1849
1850         return __kvm_write_guest_page(slot, gfn, data, offset, len);
1851 }
1852 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
1853
1854 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1855                     unsigned long len)
1856 {
1857         gfn_t gfn = gpa >> PAGE_SHIFT;
1858         int seg;
1859         int offset = offset_in_page(gpa);
1860         int ret;
1861
1862         while ((seg = next_segment(len, offset)) != 0) {
1863                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1864                 if (ret < 0)
1865                         return ret;
1866                 offset = 0;
1867                 len -= seg;
1868                 data += seg;
1869                 ++gfn;
1870         }
1871         return 0;
1872 }
1873 EXPORT_SYMBOL_GPL(kvm_write_guest);
1874
1875 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
1876                          unsigned long len)
1877 {
1878         gfn_t gfn = gpa >> PAGE_SHIFT;
1879         int seg;
1880         int offset = offset_in_page(gpa);
1881         int ret;
1882
1883         while ((seg = next_segment(len, offset)) != 0) {
1884                 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
1885                 if (ret < 0)
1886                         return ret;
1887                 offset = 0;
1888                 len -= seg;
1889                 data += seg;
1890                 ++gfn;
1891         }
1892         return 0;
1893 }
1894 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
1895
1896 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
1897                                        struct gfn_to_hva_cache *ghc,
1898                                        gpa_t gpa, unsigned long len)
1899 {
1900         int offset = offset_in_page(gpa);
1901         gfn_t start_gfn = gpa >> PAGE_SHIFT;
1902         gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1903         gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1904         gfn_t nr_pages_avail;
1905
1906         ghc->gpa = gpa;
1907         ghc->generation = slots->generation;
1908         ghc->len = len;
1909         ghc->memslot = __gfn_to_memslot(slots, start_gfn);
1910         ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, NULL);
1911         if (!kvm_is_error_hva(ghc->hva) && nr_pages_needed <= 1) {
1912                 ghc->hva += offset;
1913         } else {
1914                 /*
1915                  * If the requested region crosses two memslots, we still
1916                  * verify that the entire region is valid here.
1917                  */
1918                 while (start_gfn <= end_gfn) {
1919                         nr_pages_avail = 0;
1920                         ghc->memslot = __gfn_to_memslot(slots, start_gfn);
1921                         ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1922                                                    &nr_pages_avail);
1923                         if (kvm_is_error_hva(ghc->hva))
1924                                 return -EFAULT;
1925                         start_gfn += nr_pages_avail;
1926                 }
1927                 /* Use the slow path for cross page reads and writes. */
1928                 ghc->memslot = NULL;
1929         }
1930         return 0;
1931 }
1932
1933 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1934                               gpa_t gpa, unsigned long len)
1935 {
1936         struct kvm_memslots *slots = kvm_memslots(kvm);
1937         return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len);
1938 }
1939 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1940
1941 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1942                            void *data, int offset, unsigned long len)
1943 {
1944         struct kvm_memslots *slots = kvm_memslots(kvm);
1945         int r;
1946         gpa_t gpa = ghc->gpa + offset;
1947
1948         BUG_ON(len + offset > ghc->len);
1949
1950         if (slots->generation != ghc->generation)
1951                 __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len);
1952
1953         if (unlikely(!ghc->memslot))
1954                 return kvm_write_guest(kvm, gpa, data, len);
1955
1956         if (kvm_is_error_hva(ghc->hva))
1957                 return -EFAULT;
1958
1959         r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
1960         if (r)
1961                 return -EFAULT;
1962         mark_page_dirty_in_slot(ghc->memslot, gpa >> PAGE_SHIFT);
1963
1964         return 0;
1965 }
1966 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached);
1967
1968 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1969                            void *data, unsigned long len)
1970 {
1971         return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len);
1972 }
1973 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1974
1975 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1976                            void *data, unsigned long len)
1977 {
1978         struct kvm_memslots *slots = kvm_memslots(kvm);
1979         int r;
1980
1981         BUG_ON(len > ghc->len);
1982
1983         if (slots->generation != ghc->generation)
1984                 __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len);
1985
1986         if (unlikely(!ghc->memslot))
1987                 return kvm_read_guest(kvm, ghc->gpa, data, len);
1988
1989         if (kvm_is_error_hva(ghc->hva))
1990                 return -EFAULT;
1991
1992         r = __copy_from_user(data, (void __user *)ghc->hva, len);
1993         if (r)
1994                 return -EFAULT;
1995
1996         return 0;
1997 }
1998 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1999
2000 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
2001 {
2002         const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
2003
2004         return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
2005 }
2006 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
2007
2008 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
2009 {
2010         gfn_t gfn = gpa >> PAGE_SHIFT;
2011         int seg;
2012         int offset = offset_in_page(gpa);
2013         int ret;
2014
2015         while ((seg = next_segment(len, offset)) != 0) {
2016                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
2017                 if (ret < 0)
2018                         return ret;
2019                 offset = 0;
2020                 len -= seg;
2021                 ++gfn;
2022         }
2023         return 0;
2024 }
2025 EXPORT_SYMBOL_GPL(kvm_clear_guest);
2026
2027 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot,
2028                                     gfn_t gfn)
2029 {
2030         if (memslot && memslot->dirty_bitmap) {
2031                 unsigned long rel_gfn = gfn - memslot->base_gfn;
2032
2033                 set_bit_le(rel_gfn, memslot->dirty_bitmap);
2034         }
2035 }
2036
2037 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
2038 {
2039         struct kvm_memory_slot *memslot;
2040
2041         memslot = gfn_to_memslot(kvm, gfn);
2042         mark_page_dirty_in_slot(memslot, gfn);
2043 }
2044 EXPORT_SYMBOL_GPL(mark_page_dirty);
2045
2046 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
2047 {
2048         struct kvm_memory_slot *memslot;
2049
2050         memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2051         mark_page_dirty_in_slot(memslot, gfn);
2052 }
2053 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
2054
2055 void kvm_sigset_activate(struct kvm_vcpu *vcpu)
2056 {
2057         if (!vcpu->sigset_active)
2058                 return;
2059
2060         /*
2061          * This does a lockless modification of ->real_blocked, which is fine
2062          * because, only current can change ->real_blocked and all readers of
2063          * ->real_blocked don't care as long ->real_blocked is always a subset
2064          * of ->blocked.
2065          */
2066         sigprocmask(SIG_SETMASK, &vcpu->sigset, &current->real_blocked);
2067 }
2068
2069 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu)
2070 {
2071         if (!vcpu->sigset_active)
2072                 return;
2073
2074         sigprocmask(SIG_SETMASK, &current->real_blocked, NULL);
2075         sigemptyset(&current->real_blocked);
2076 }
2077
2078 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
2079 {
2080         unsigned int old, val, grow;
2081
2082         old = val = vcpu->halt_poll_ns;
2083         grow = READ_ONCE(halt_poll_ns_grow);
2084         /* 10us base */
2085         if (val == 0 && grow)
2086                 val = 10000;
2087         else
2088                 val *= grow;
2089
2090         if (val > halt_poll_ns)
2091                 val = halt_poll_ns;
2092
2093         vcpu->halt_poll_ns = val;
2094         trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
2095 }
2096
2097 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
2098 {
2099         unsigned int old, val, shrink;
2100
2101         old = val = vcpu->halt_poll_ns;
2102         shrink = READ_ONCE(halt_poll_ns_shrink);
2103         if (shrink == 0)
2104                 val = 0;
2105         else
2106                 val /= shrink;
2107
2108         vcpu->halt_poll_ns = val;
2109         trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
2110 }
2111
2112 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
2113 {
2114         if (kvm_arch_vcpu_runnable(vcpu)) {
2115                 kvm_make_request(KVM_REQ_UNHALT, vcpu);
2116                 return -EINTR;
2117         }
2118         if (kvm_cpu_has_pending_timer(vcpu))
2119                 return -EINTR;
2120         if (signal_pending(current))
2121                 return -EINTR;
2122
2123         return 0;
2124 }
2125
2126 /*
2127  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2128  */
2129 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
2130 {
2131         ktime_t start, cur;
2132         DECLARE_SWAITQUEUE(wait);
2133         bool waited = false;
2134         u64 block_ns;
2135
2136         start = cur = ktime_get();
2137         if (vcpu->halt_poll_ns) {
2138                 ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns);
2139
2140                 ++vcpu->stat.halt_attempted_poll;
2141                 do {
2142                         /*
2143                          * This sets KVM_REQ_UNHALT if an interrupt
2144                          * arrives.
2145                          */
2146                         if (kvm_vcpu_check_block(vcpu) < 0) {
2147                                 ++vcpu->stat.halt_successful_poll;
2148                                 if (!vcpu_valid_wakeup(vcpu))
2149                                         ++vcpu->stat.halt_poll_invalid;
2150                                 goto out;
2151                         }
2152                         cur = ktime_get();
2153                 } while (single_task_running() && ktime_before(cur, stop));
2154         }
2155
2156         kvm_arch_vcpu_blocking(vcpu);
2157
2158         for (;;) {
2159                 prepare_to_swait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
2160
2161                 if (kvm_vcpu_check_block(vcpu) < 0)
2162                         break;
2163
2164                 waited = true;
2165                 schedule();
2166         }
2167
2168         finish_swait(&vcpu->wq, &wait);
2169         cur = ktime_get();
2170
2171         kvm_arch_vcpu_unblocking(vcpu);
2172 out:
2173         block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
2174
2175         if (!vcpu_valid_wakeup(vcpu))
2176                 shrink_halt_poll_ns(vcpu);
2177         else if (halt_poll_ns) {
2178                 if (block_ns <= vcpu->halt_poll_ns)
2179                         ;
2180                 /* we had a long block, shrink polling */
2181                 else if (vcpu->halt_poll_ns && block_ns > halt_poll_ns)
2182                         shrink_halt_poll_ns(vcpu);
2183                 /* we had a short halt and our poll time is too small */
2184                 else if (vcpu->halt_poll_ns < halt_poll_ns &&
2185                         block_ns < halt_poll_ns)
2186                         grow_halt_poll_ns(vcpu);
2187         } else
2188                 vcpu->halt_poll_ns = 0;
2189
2190         trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu));
2191         kvm_arch_vcpu_block_finish(vcpu);
2192 }
2193 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
2194
2195 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
2196 {
2197         struct swait_queue_head *wqp;
2198
2199         wqp = kvm_arch_vcpu_wq(vcpu);
2200         if (swq_has_sleeper(wqp)) {
2201                 swake_up(wqp);
2202                 ++vcpu->stat.halt_wakeup;
2203                 return true;
2204         }
2205
2206         return false;
2207 }
2208 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
2209
2210 #ifndef CONFIG_S390
2211 /*
2212  * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2213  */
2214 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
2215 {
2216         int me;
2217         int cpu = vcpu->cpu;
2218
2219         if (kvm_vcpu_wake_up(vcpu))
2220                 return;
2221
2222         me = get_cpu();
2223         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
2224                 if (kvm_arch_vcpu_should_kick(vcpu))
2225                         smp_send_reschedule(cpu);
2226         put_cpu();
2227 }
2228 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
2229 #endif /* !CONFIG_S390 */
2230
2231 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
2232 {
2233         struct pid *pid;
2234         struct task_struct *task = NULL;
2235         int ret = 0;
2236
2237         rcu_read_lock();
2238         pid = rcu_dereference(target->pid);
2239         if (pid)
2240                 task = get_pid_task(pid, PIDTYPE_PID);
2241         rcu_read_unlock();
2242         if (!task)
2243                 return ret;
2244         ret = yield_to(task, 1);
2245         put_task_struct(task);
2246
2247         return ret;
2248 }
2249 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
2250
2251 /*
2252  * Helper that checks whether a VCPU is eligible for directed yield.
2253  * Most eligible candidate to yield is decided by following heuristics:
2254  *
2255  *  (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2256  *  (preempted lock holder), indicated by @in_spin_loop.
2257  *  Set at the beiginning and cleared at the end of interception/PLE handler.
2258  *
2259  *  (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2260  *  chance last time (mostly it has become eligible now since we have probably
2261  *  yielded to lockholder in last iteration. This is done by toggling
2262  *  @dy_eligible each time a VCPU checked for eligibility.)
2263  *
2264  *  Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2265  *  to preempted lock-holder could result in wrong VCPU selection and CPU
2266  *  burning. Giving priority for a potential lock-holder increases lock
2267  *  progress.
2268  *
2269  *  Since algorithm is based on heuristics, accessing another VCPU data without
2270  *  locking does not harm. It may result in trying to yield to  same VCPU, fail
2271  *  and continue with next VCPU and so on.
2272  */
2273 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
2274 {
2275 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2276         bool eligible;
2277
2278         eligible = !vcpu->spin_loop.in_spin_loop ||
2279                     vcpu->spin_loop.dy_eligible;
2280
2281         if (vcpu->spin_loop.in_spin_loop)
2282                 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
2283
2284         return eligible;
2285 #else
2286         return true;
2287 #endif
2288 }
2289
2290 void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode)
2291 {
2292         struct kvm *kvm = me->kvm;
2293         struct kvm_vcpu *vcpu;
2294         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
2295         int yielded = 0;
2296         int try = 3;
2297         int pass;
2298         int i;
2299
2300         kvm_vcpu_set_in_spin_loop(me, true);
2301         /*
2302          * We boost the priority of a VCPU that is runnable but not
2303          * currently running, because it got preempted by something
2304          * else and called schedule in __vcpu_run.  Hopefully that
2305          * VCPU is holding the lock that we need and will release it.
2306          * We approximate round-robin by starting at the last boosted VCPU.
2307          */
2308         for (pass = 0; pass < 2 && !yielded && try; pass++) {
2309                 kvm_for_each_vcpu(i, vcpu, kvm) {
2310                         if (!pass && i <= last_boosted_vcpu) {
2311                                 i = last_boosted_vcpu;
2312                                 continue;
2313                         } else if (pass && i > last_boosted_vcpu)
2314                                 break;
2315                         if (!READ_ONCE(vcpu->preempted))
2316                                 continue;
2317                         if (vcpu == me)
2318                                 continue;
2319                         if (swait_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
2320                                 continue;
2321                         if (yield_to_kernel_mode && !kvm_arch_vcpu_in_kernel(vcpu))
2322                                 continue;
2323                         if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
2324                                 continue;
2325
2326                         yielded = kvm_vcpu_yield_to(vcpu);
2327                         if (yielded > 0) {
2328                                 kvm->last_boosted_vcpu = i;
2329                                 break;
2330                         } else if (yielded < 0) {
2331                                 try--;
2332                                 if (!try)
2333                                         break;
2334                         }
2335                 }
2336         }
2337         kvm_vcpu_set_in_spin_loop(me, false);
2338
2339         /* Ensure vcpu is not eligible during next spinloop */
2340         kvm_vcpu_set_dy_eligible(me, false);
2341 }
2342 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
2343
2344 static int kvm_vcpu_fault(struct vm_fault *vmf)
2345 {
2346         struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data;
2347         struct page *page;
2348
2349         if (vmf->pgoff == 0)
2350                 page = virt_to_page(vcpu->run);
2351 #ifdef CONFIG_X86
2352         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
2353                 page = virt_to_page(vcpu->arch.pio_data);
2354 #endif
2355 #ifdef CONFIG_KVM_MMIO
2356         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
2357                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
2358 #endif
2359         else
2360                 return kvm_arch_vcpu_fault(vcpu, vmf);
2361         get_page(page);
2362         vmf->page = page;
2363         return 0;
2364 }
2365
2366 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
2367         .fault = kvm_vcpu_fault,
2368 };
2369
2370 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2371 {
2372         vma->vm_ops = &kvm_vcpu_vm_ops;
2373         return 0;
2374 }
2375
2376 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2377 {
2378         struct kvm_vcpu *vcpu = filp->private_data;
2379
2380         debugfs_remove_recursive(vcpu->debugfs_dentry);
2381         kvm_put_kvm(vcpu->kvm);
2382         return 0;
2383 }
2384
2385 static struct file_operations kvm_vcpu_fops = {
2386         .release        = kvm_vcpu_release,
2387         .unlocked_ioctl = kvm_vcpu_ioctl,
2388 #ifdef CONFIG_KVM_COMPAT
2389         .compat_ioctl   = kvm_vcpu_compat_ioctl,
2390 #endif
2391         .mmap           = kvm_vcpu_mmap,
2392         .llseek         = noop_llseek,
2393 };
2394
2395 /*
2396  * Allocates an inode for the vcpu.
2397  */
2398 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2399 {
2400         char name[8 + 1 + ITOA_MAX_LEN + 1];
2401
2402         snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id);
2403         return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
2404 }
2405
2406 static int kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
2407 {
2408         char dir_name[ITOA_MAX_LEN * 2];
2409         int ret;
2410
2411         if (!kvm_arch_has_vcpu_debugfs())
2412                 return 0;
2413
2414         if (!debugfs_initialized())
2415                 return 0;
2416
2417         snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
2418         vcpu->debugfs_dentry = debugfs_create_dir(dir_name,
2419                                                                 vcpu->kvm->debugfs_dentry);
2420         if (!vcpu->debugfs_dentry)
2421                 return -ENOMEM;
2422
2423         ret = kvm_arch_create_vcpu_debugfs(vcpu);
2424         if (ret < 0) {
2425                 debugfs_remove_recursive(vcpu->debugfs_dentry);
2426                 return ret;
2427         }
2428
2429         return 0;
2430 }
2431
2432 /*
2433  * Creates some virtual cpus.  Good luck creating more than one.
2434  */
2435 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
2436 {
2437         int r;
2438         struct kvm_vcpu *vcpu;
2439
2440         if (id >= KVM_MAX_VCPU_ID)
2441                 return -EINVAL;
2442
2443         mutex_lock(&kvm->lock);
2444         if (kvm->created_vcpus == KVM_MAX_VCPUS) {
2445                 mutex_unlock(&kvm->lock);
2446                 return -EINVAL;
2447         }
2448
2449         kvm->created_vcpus++;
2450         mutex_unlock(&kvm->lock);
2451
2452         vcpu = kvm_arch_vcpu_create(kvm, id);
2453         if (IS_ERR(vcpu)) {
2454                 r = PTR_ERR(vcpu);
2455                 goto vcpu_decrement;
2456         }
2457
2458         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2459
2460         r = kvm_arch_vcpu_setup(vcpu);
2461         if (r)
2462                 goto vcpu_destroy;
2463
2464         r = kvm_create_vcpu_debugfs(vcpu);
2465         if (r)
2466                 goto vcpu_destroy;
2467
2468         mutex_lock(&kvm->lock);
2469         if (kvm_get_vcpu_by_id(kvm, id)) {
2470                 r = -EEXIST;
2471                 goto unlock_vcpu_destroy;
2472         }
2473
2474         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
2475
2476         /* Now it's all set up, let userspace reach it */
2477         kvm_get_kvm(kvm);
2478         r = create_vcpu_fd(vcpu);
2479         if (r < 0) {
2480                 kvm_put_kvm(kvm);
2481                 goto unlock_vcpu_destroy;
2482         }
2483
2484         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
2485
2486         /*
2487          * Pairs with smp_rmb() in kvm_get_vcpu.  Write kvm->vcpus
2488          * before kvm->online_vcpu's incremented value.
2489          */
2490         smp_wmb();
2491         atomic_inc(&kvm->online_vcpus);
2492
2493         mutex_unlock(&kvm->lock);
2494         kvm_arch_vcpu_postcreate(vcpu);
2495         return r;
2496
2497 unlock_vcpu_destroy:
2498         mutex_unlock(&kvm->lock);
2499         debugfs_remove_recursive(vcpu->debugfs_dentry);
2500 vcpu_destroy:
2501         kvm_arch_vcpu_destroy(vcpu);
2502 vcpu_decrement:
2503         mutex_lock(&kvm->lock);
2504         kvm->created_vcpus--;
2505         mutex_unlock(&kvm->lock);
2506         return r;
2507 }
2508
2509 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2510 {
2511         if (sigset) {
2512                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2513                 vcpu->sigset_active = 1;
2514                 vcpu->sigset = *sigset;
2515         } else
2516                 vcpu->sigset_active = 0;
2517         return 0;
2518 }
2519
2520 static long kvm_vcpu_ioctl(struct file *filp,
2521                            unsigned int ioctl, unsigned long arg)
2522 {
2523         struct kvm_vcpu *vcpu = filp->private_data;
2524         void __user *argp = (void __user *)arg;
2525         int r;
2526         struct kvm_fpu *fpu = NULL;
2527         struct kvm_sregs *kvm_sregs = NULL;
2528
2529         if (vcpu->kvm->mm != current->mm)
2530                 return -EIO;
2531
2532         if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2533                 return -EINVAL;
2534
2535         /*
2536          * Some architectures have vcpu ioctls that are asynchronous to vcpu
2537          * execution; mutex_lock() would break them.
2538          */
2539         r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg);
2540         if (r != -ENOIOCTLCMD)
2541                 return r;
2542
2543         if (mutex_lock_killable(&vcpu->mutex))
2544                 return -EINTR;
2545         switch (ioctl) {
2546         case KVM_RUN: {
2547                 struct pid *oldpid;
2548                 r = -EINVAL;
2549                 if (arg)
2550                         goto out;
2551                 oldpid = rcu_access_pointer(vcpu->pid);
2552                 if (unlikely(oldpid != current->pids[PIDTYPE_PID].pid)) {
2553                         /* The thread running this VCPU changed. */
2554                         struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
2555
2556                         rcu_assign_pointer(vcpu->pid, newpid);
2557                         if (oldpid)
2558                                 synchronize_rcu();
2559                         put_pid(oldpid);
2560                 }
2561                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2562                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2563                 break;
2564         }
2565         case KVM_GET_REGS: {
2566                 struct kvm_regs *kvm_regs;
2567
2568                 r = -ENOMEM;
2569                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2570                 if (!kvm_regs)
2571                         goto out;
2572                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2573                 if (r)
2574                         goto out_free1;
2575                 r = -EFAULT;
2576                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2577                         goto out_free1;
2578                 r = 0;
2579 out_free1:
2580                 kfree(kvm_regs);
2581                 break;
2582         }
2583         case KVM_SET_REGS: {
2584                 struct kvm_regs *kvm_regs;
2585
2586                 r = -ENOMEM;
2587                 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2588                 if (IS_ERR(kvm_regs)) {
2589                         r = PTR_ERR(kvm_regs);
2590                         goto out;
2591                 }
2592                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2593                 kfree(kvm_regs);
2594                 break;
2595         }
2596         case KVM_GET_SREGS: {
2597                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2598                 r = -ENOMEM;
2599                 if (!kvm_sregs)
2600                         goto out;
2601                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2602                 if (r)
2603                         goto out;
2604                 r = -EFAULT;
2605                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2606                         goto out;
2607                 r = 0;
2608                 break;
2609         }
2610         case KVM_SET_SREGS: {
2611                 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2612                 if (IS_ERR(kvm_sregs)) {
2613                         r = PTR_ERR(kvm_sregs);
2614                         kvm_sregs = NULL;
2615                         goto out;
2616                 }
2617                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2618                 break;
2619         }
2620         case KVM_GET_MP_STATE: {
2621                 struct kvm_mp_state mp_state;
2622
2623                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2624                 if (r)
2625                         goto out;
2626                 r = -EFAULT;
2627                 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
2628                         goto out;
2629                 r = 0;
2630                 break;
2631         }
2632         case KVM_SET_MP_STATE: {
2633                 struct kvm_mp_state mp_state;
2634
2635                 r = -EFAULT;
2636                 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
2637                         goto out;
2638                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2639                 break;
2640         }
2641         case KVM_TRANSLATE: {
2642                 struct kvm_translation tr;
2643
2644                 r = -EFAULT;
2645                 if (copy_from_user(&tr, argp, sizeof(tr)))
2646                         goto out;
2647                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2648                 if (r)
2649                         goto out;
2650                 r = -EFAULT;
2651                 if (copy_to_user(argp, &tr, sizeof(tr)))
2652                         goto out;
2653                 r = 0;
2654                 break;
2655         }
2656         case KVM_SET_GUEST_DEBUG: {
2657                 struct kvm_guest_debug dbg;
2658
2659                 r = -EFAULT;
2660                 if (copy_from_user(&dbg, argp, sizeof(dbg)))
2661                         goto out;
2662                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2663                 break;
2664         }
2665         case KVM_SET_SIGNAL_MASK: {
2666                 struct kvm_signal_mask __user *sigmask_arg = argp;
2667                 struct kvm_signal_mask kvm_sigmask;
2668                 sigset_t sigset, *p;
2669
2670                 p = NULL;
2671                 if (argp) {
2672                         r = -EFAULT;
2673                         if (copy_from_user(&kvm_sigmask, argp,
2674                                            sizeof(kvm_sigmask)))
2675                                 goto out;
2676                         r = -EINVAL;
2677                         if (kvm_sigmask.len != sizeof(sigset))
2678                                 goto out;
2679                         r = -EFAULT;
2680                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2681                                            sizeof(sigset)))
2682                                 goto out;
2683                         p = &sigset;
2684                 }
2685                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2686                 break;
2687         }
2688         case KVM_GET_FPU: {
2689                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2690                 r = -ENOMEM;
2691                 if (!fpu)
2692                         goto out;
2693                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2694                 if (r)
2695                         goto out;
2696                 r = -EFAULT;
2697                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2698                         goto out;
2699                 r = 0;
2700                 break;
2701         }
2702         case KVM_SET_FPU: {
2703                 fpu = memdup_user(argp, sizeof(*fpu));
2704                 if (IS_ERR(fpu)) {
2705                         r = PTR_ERR(fpu);
2706                         fpu = NULL;
2707                         goto out;
2708                 }
2709                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2710                 break;
2711         }
2712         default:
2713                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2714         }
2715 out:
2716         mutex_unlock(&vcpu->mutex);
2717         kfree(fpu);
2718         kfree(kvm_sregs);
2719         return r;
2720 }
2721
2722 #ifdef CONFIG_KVM_COMPAT
2723 static long kvm_vcpu_compat_ioctl(struct file *filp,
2724                                   unsigned int ioctl, unsigned long arg)
2725 {
2726         struct kvm_vcpu *vcpu = filp->private_data;
2727         void __user *argp = compat_ptr(arg);
2728         int r;
2729
2730         if (vcpu->kvm->mm != current->mm)
2731                 return -EIO;
2732
2733         switch (ioctl) {
2734         case KVM_SET_SIGNAL_MASK: {
2735                 struct kvm_signal_mask __user *sigmask_arg = argp;
2736                 struct kvm_signal_mask kvm_sigmask;
2737                 sigset_t sigset;
2738
2739                 if (argp) {
2740                         r = -EFAULT;
2741                         if (copy_from_user(&kvm_sigmask, argp,
2742                                            sizeof(kvm_sigmask)))
2743                                 goto out;
2744                         r = -EINVAL;
2745                         if (kvm_sigmask.len != sizeof(compat_sigset_t))
2746                                 goto out;
2747                         r = -EFAULT;
2748                         if (get_compat_sigset(&sigset, (void *)sigmask_arg->sigset))
2749                                 goto out;
2750                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2751                 } else
2752                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2753                 break;
2754         }
2755         default:
2756                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2757         }
2758
2759 out:
2760         return r;
2761 }
2762 #endif
2763
2764 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2765                                  int (*accessor)(struct kvm_device *dev,
2766                                                  struct kvm_device_attr *attr),
2767                                  unsigned long arg)
2768 {
2769         struct kvm_device_attr attr;
2770
2771         if (!accessor)
2772                 return -EPERM;
2773
2774         if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2775                 return -EFAULT;
2776
2777         return accessor(dev, &attr);
2778 }
2779
2780 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2781                              unsigned long arg)
2782 {
2783         struct kvm_device *dev = filp->private_data;
2784
2785         switch (ioctl) {
2786         case KVM_SET_DEVICE_ATTR:
2787                 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2788         case KVM_GET_DEVICE_ATTR:
2789                 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2790         case KVM_HAS_DEVICE_ATTR:
2791                 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2792         default:
2793                 if (dev->ops->ioctl)
2794                         return dev->ops->ioctl(dev, ioctl, arg);
2795
2796                 return -ENOTTY;
2797         }
2798 }
2799
2800 static int kvm_device_release(struct inode *inode, struct file *filp)
2801 {
2802         struct kvm_device *dev = filp->private_data;
2803         struct kvm *kvm = dev->kvm;
2804
2805         kvm_put_kvm(kvm);
2806         return 0;
2807 }
2808
2809 static const struct file_operations kvm_device_fops = {
2810         .unlocked_ioctl = kvm_device_ioctl,
2811 #ifdef CONFIG_KVM_COMPAT
2812         .compat_ioctl = kvm_device_ioctl,
2813 #endif
2814         .release = kvm_device_release,
2815 };
2816
2817 struct kvm_device *kvm_device_from_filp(struct file *filp)
2818 {
2819         if (filp->f_op != &kvm_device_fops)
2820                 return NULL;
2821
2822         return filp->private_data;
2823 }
2824
2825 static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
2826 #ifdef CONFIG_KVM_MPIC
2827         [KVM_DEV_TYPE_FSL_MPIC_20]      = &kvm_mpic_ops,
2828         [KVM_DEV_TYPE_FSL_MPIC_42]      = &kvm_mpic_ops,
2829 #endif
2830 };
2831
2832 int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
2833 {
2834         if (type >= ARRAY_SIZE(kvm_device_ops_table))
2835                 return -ENOSPC;
2836
2837         if (kvm_device_ops_table[type] != NULL)
2838                 return -EEXIST;
2839
2840         kvm_device_ops_table[type] = ops;
2841         return 0;
2842 }
2843
2844 void kvm_unregister_device_ops(u32 type)
2845 {
2846         if (kvm_device_ops_table[type] != NULL)
2847                 kvm_device_ops_table[type] = NULL;
2848 }
2849
2850 static int kvm_ioctl_create_device(struct kvm *kvm,
2851                                    struct kvm_create_device *cd)
2852 {
2853         struct kvm_device_ops *ops = NULL;
2854         struct kvm_device *dev;
2855         bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2856         int ret;
2857
2858         if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
2859                 return -ENODEV;
2860
2861         ops = kvm_device_ops_table[cd->type];
2862         if (ops == NULL)
2863                 return -ENODEV;
2864
2865         if (test)
2866                 return 0;
2867
2868         dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2869         if (!dev)
2870                 return -ENOMEM;
2871
2872         dev->ops = ops;
2873         dev->kvm = kvm;
2874
2875         mutex_lock(&kvm->lock);
2876         ret = ops->create(dev, cd->type);
2877         if (ret < 0) {
2878                 mutex_unlock(&kvm->lock);
2879                 kfree(dev);
2880                 return ret;
2881         }
2882         list_add(&dev->vm_node, &kvm->devices);
2883         mutex_unlock(&kvm->lock);
2884
2885         if (ops->init)
2886                 ops->init(dev);
2887
2888         ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2889         if (ret < 0) {
2890                 mutex_lock(&kvm->lock);
2891                 list_del(&dev->vm_node);
2892                 mutex_unlock(&kvm->lock);
2893                 ops->destroy(dev);
2894                 return ret;
2895         }
2896
2897         kvm_get_kvm(kvm);
2898         cd->fd = ret;
2899         return 0;
2900 }
2901
2902 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2903 {
2904         switch (arg) {
2905         case KVM_CAP_USER_MEMORY:
2906         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2907         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2908         case KVM_CAP_INTERNAL_ERROR_DATA:
2909 #ifdef CONFIG_HAVE_KVM_MSI
2910         case KVM_CAP_SIGNAL_MSI:
2911 #endif
2912 #ifdef CONFIG_HAVE_KVM_IRQFD
2913         case KVM_CAP_IRQFD:
2914         case KVM_CAP_IRQFD_RESAMPLE:
2915 #endif
2916         case KVM_CAP_IOEVENTFD_ANY_LENGTH:
2917         case KVM_CAP_CHECK_EXTENSION_VM:
2918                 return 1;
2919 #ifdef CONFIG_KVM_MMIO
2920         case KVM_CAP_COALESCED_MMIO:
2921                 return KVM_COALESCED_MMIO_PAGE_OFFSET;
2922 #endif
2923 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2924         case KVM_CAP_IRQ_ROUTING:
2925                 return KVM_MAX_IRQ_ROUTES;
2926 #endif
2927 #if KVM_ADDRESS_SPACE_NUM > 1
2928         case KVM_CAP_MULTI_ADDRESS_SPACE:
2929                 return KVM_ADDRESS_SPACE_NUM;
2930 #endif
2931         case KVM_CAP_MAX_VCPU_ID:
2932                 return KVM_MAX_VCPU_ID;
2933         default:
2934                 break;
2935         }
2936         return kvm_vm_ioctl_check_extension(kvm, arg);
2937 }
2938
2939 static long kvm_vm_ioctl(struct file *filp,
2940                            unsigned int ioctl, unsigned long arg)
2941 {
2942         struct kvm *kvm = filp->private_data;
2943         void __user *argp = (void __user *)arg;
2944         int r;
2945
2946         if (kvm->mm != current->mm)
2947                 return -EIO;
2948         switch (ioctl) {
2949         case KVM_CREATE_VCPU:
2950                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2951                 break;
2952         case KVM_SET_USER_MEMORY_REGION: {
2953                 struct kvm_userspace_memory_region kvm_userspace_mem;
2954
2955                 r = -EFAULT;
2956                 if (copy_from_user(&kvm_userspace_mem, argp,
2957                                                 sizeof(kvm_userspace_mem)))
2958                         goto out;
2959
2960                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2961                 break;
2962         }
2963         case KVM_GET_DIRTY_LOG: {
2964                 struct kvm_dirty_log log;
2965
2966                 r = -EFAULT;
2967                 if (copy_from_user(&log, argp, sizeof(log)))
2968                         goto out;
2969                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2970                 break;
2971         }
2972 #ifdef CONFIG_KVM_MMIO
2973         case KVM_REGISTER_COALESCED_MMIO: {
2974                 struct kvm_coalesced_mmio_zone zone;
2975
2976                 r = -EFAULT;
2977                 if (copy_from_user(&zone, argp, sizeof(zone)))
2978                         goto out;
2979                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2980                 break;
2981         }
2982         case KVM_UNREGISTER_COALESCED_MMIO: {
2983                 struct kvm_coalesced_mmio_zone zone;
2984
2985                 r = -EFAULT;
2986                 if (copy_from_user(&zone, argp, sizeof(zone)))
2987                         goto out;
2988                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2989                 break;
2990         }
2991 #endif
2992         case KVM_IRQFD: {
2993                 struct kvm_irqfd data;
2994
2995                 r = -EFAULT;
2996                 if (copy_from_user(&data, argp, sizeof(data)))
2997                         goto out;
2998                 r = kvm_irqfd(kvm, &data);
2999                 break;
3000         }
3001         case KVM_IOEVENTFD: {
3002                 struct kvm_ioeventfd data;
3003
3004                 r = -EFAULT;
3005                 if (copy_from_user(&data, argp, sizeof(data)))
3006                         goto out;
3007                 r = kvm_ioeventfd(kvm, &data);
3008                 break;
3009         }
3010 #ifdef CONFIG_HAVE_KVM_MSI
3011         case KVM_SIGNAL_MSI: {
3012                 struct kvm_msi msi;
3013
3014                 r = -EFAULT;
3015                 if (copy_from_user(&msi, argp, sizeof(msi)))
3016                         goto out;
3017                 r = kvm_send_userspace_msi(kvm, &msi);
3018                 break;
3019         }
3020 #endif
3021 #ifdef __KVM_HAVE_IRQ_LINE
3022         case KVM_IRQ_LINE_STATUS:
3023         case KVM_IRQ_LINE: {
3024                 struct kvm_irq_level irq_event;
3025
3026                 r = -EFAULT;
3027                 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
3028                         goto out;
3029
3030                 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
3031                                         ioctl == KVM_IRQ_LINE_STATUS);
3032                 if (r)
3033                         goto out;
3034
3035                 r = -EFAULT;
3036                 if (ioctl == KVM_IRQ_LINE_STATUS) {
3037                         if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
3038                                 goto out;
3039                 }
3040
3041                 r = 0;
3042                 break;
3043         }
3044 #endif
3045 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3046         case KVM_SET_GSI_ROUTING: {
3047                 struct kvm_irq_routing routing;
3048                 struct kvm_irq_routing __user *urouting;
3049                 struct kvm_irq_routing_entry *entries = NULL;
3050
3051                 r = -EFAULT;
3052                 if (copy_from_user(&routing, argp, sizeof(routing)))
3053                         goto out;
3054                 r = -EINVAL;
3055                 if (!kvm_arch_can_set_irq_routing(kvm))
3056                         goto out;
3057                 if (routing.nr > KVM_MAX_IRQ_ROUTES)
3058                         goto out;
3059                 if (routing.flags)
3060                         goto out;
3061                 if (routing.nr) {
3062                         r = -ENOMEM;
3063                         entries = vmalloc(routing.nr * sizeof(*entries));
3064                         if (!entries)
3065                                 goto out;
3066                         r = -EFAULT;
3067                         urouting = argp;
3068                         if (copy_from_user(entries, urouting->entries,
3069                                            routing.nr * sizeof(*entries)))
3070                                 goto out_free_irq_routing;
3071                 }
3072                 r = kvm_set_irq_routing(kvm, entries, routing.nr,
3073                                         routing.flags);
3074 out_free_irq_routing:
3075                 vfree(entries);
3076                 break;
3077         }
3078 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
3079         case KVM_CREATE_DEVICE: {
3080                 struct kvm_create_device cd;
3081
3082                 r = -EFAULT;
3083                 if (copy_from_user(&cd, argp, sizeof(cd)))
3084                         goto out;
3085
3086                 r = kvm_ioctl_create_device(kvm, &cd);
3087                 if (r)
3088                         goto out;
3089
3090                 r = -EFAULT;
3091                 if (copy_to_user(argp, &cd, sizeof(cd)))
3092                         goto out;
3093
3094                 r = 0;
3095                 break;
3096         }
3097         case KVM_CHECK_EXTENSION:
3098                 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
3099                 break;
3100         default:
3101                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
3102         }
3103 out:
3104         return r;
3105 }
3106
3107 #ifdef CONFIG_KVM_COMPAT
3108 struct compat_kvm_dirty_log {
3109         __u32 slot;
3110         __u32 padding1;
3111         union {
3112                 compat_uptr_t dirty_bitmap; /* one bit per page */
3113                 __u64 padding2;
3114         };
3115 };
3116
3117 static long kvm_vm_compat_ioctl(struct file *filp,
3118                            unsigned int ioctl, unsigned long arg)
3119 {
3120         struct kvm *kvm = filp->private_data;
3121         int r;
3122
3123         if (kvm->mm != current->mm)
3124                 return -EIO;
3125         switch (ioctl) {
3126         case KVM_GET_DIRTY_LOG: {
3127                 struct compat_kvm_dirty_log compat_log;
3128                 struct kvm_dirty_log log;
3129
3130                 if (copy_from_user(&compat_log, (void __user *)arg,
3131                                    sizeof(compat_log)))
3132                         return -EFAULT;
3133                 log.slot         = compat_log.slot;
3134                 log.padding1     = compat_log.padding1;
3135                 log.padding2     = compat_log.padding2;
3136                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
3137
3138                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3139                 break;
3140         }
3141         default:
3142                 r = kvm_vm_ioctl(filp, ioctl, arg);
3143         }
3144         return r;
3145 }
3146 #endif
3147
3148 static struct file_operations kvm_vm_fops = {
3149         .release        = kvm_vm_release,
3150         .unlocked_ioctl = kvm_vm_ioctl,
3151 #ifdef CONFIG_KVM_COMPAT
3152         .compat_ioctl   = kvm_vm_compat_ioctl,
3153 #endif
3154         .llseek         = noop_llseek,
3155 };
3156
3157 static int kvm_dev_ioctl_create_vm(unsigned long type)
3158 {
3159         int r;
3160         struct kvm *kvm;
3161         struct file *file;
3162
3163         kvm = kvm_create_vm(type);
3164         if (IS_ERR(kvm))
3165                 return PTR_ERR(kvm);
3166 #ifdef CONFIG_KVM_MMIO
3167         r = kvm_coalesced_mmio_init(kvm);
3168         if (r < 0)
3169                 goto put_kvm;
3170 #endif
3171         r = get_unused_fd_flags(O_CLOEXEC);
3172         if (r < 0)
3173                 goto put_kvm;
3174
3175         file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
3176         if (IS_ERR(file)) {
3177                 put_unused_fd(r);
3178                 r = PTR_ERR(file);
3179                 goto put_kvm;
3180         }
3181
3182         /*
3183          * Don't call kvm_put_kvm anymore at this point; file->f_op is
3184          * already set, with ->release() being kvm_vm_release().  In error
3185          * cases it will be called by the final fput(file) and will take
3186          * care of doing kvm_put_kvm(kvm).
3187          */
3188         if (kvm_create_vm_debugfs(kvm, r) < 0) {
3189                 put_unused_fd(r);
3190                 fput(file);
3191                 return -ENOMEM;
3192         }
3193         kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm);
3194
3195         fd_install(r, file);
3196         return r;
3197
3198 put_kvm:
3199         kvm_put_kvm(kvm);
3200         return r;
3201 }
3202
3203 static long kvm_dev_ioctl(struct file *filp,
3204                           unsigned int ioctl, unsigned long arg)
3205 {
3206         long r = -EINVAL;
3207
3208         switch (ioctl) {
3209         case KVM_GET_API_VERSION:
3210                 if (arg)
3211                         goto out;
3212                 r = KVM_API_VERSION;
3213                 break;
3214         case KVM_CREATE_VM:
3215                 r = kvm_dev_ioctl_create_vm(arg);
3216                 break;
3217         case KVM_CHECK_EXTENSION:
3218                 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
3219                 break;
3220         case KVM_GET_VCPU_MMAP_SIZE:
3221                 if (arg)
3222                         goto out;
3223                 r = PAGE_SIZE;     /* struct kvm_run */
3224 #ifdef CONFIG_X86
3225                 r += PAGE_SIZE;    /* pio data page */
3226 #endif
3227 #ifdef CONFIG_KVM_MMIO
3228                 r += PAGE_SIZE;    /* coalesced mmio ring page */
3229 #endif
3230                 break;
3231         case KVM_TRACE_ENABLE:
3232         case KVM_TRACE_PAUSE:
3233         case KVM_TRACE_DISABLE:
3234                 r = -EOPNOTSUPP;
3235                 break;
3236         default:
3237                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
3238         }
3239 out:
3240         return r;
3241 }
3242
3243 static struct file_operations kvm_chardev_ops = {
3244         .unlocked_ioctl = kvm_dev_ioctl,
3245         .compat_ioctl   = kvm_dev_ioctl,
3246         .llseek         = noop_llseek,
3247 };
3248
3249 static struct miscdevice kvm_dev = {
3250         KVM_MINOR,
3251         "kvm",
3252         &kvm_chardev_ops,
3253 };
3254
3255 static void hardware_enable_nolock(void *junk)
3256 {
3257         int cpu = raw_smp_processor_id();
3258         int r;
3259
3260         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
3261                 return;
3262
3263         cpumask_set_cpu(cpu, cpus_hardware_enabled);
3264
3265         r = kvm_arch_hardware_enable();
3266
3267         if (r) {
3268                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3269                 atomic_inc(&hardware_enable_failed);
3270                 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
3271         }
3272 }
3273
3274 static int kvm_starting_cpu(unsigned int cpu)
3275 {
3276         raw_spin_lock(&kvm_count_lock);
3277         if (kvm_usage_count)
3278                 hardware_enable_nolock(NULL);
3279         raw_spin_unlock(&kvm_count_lock);
3280         return 0;
3281 }
3282
3283 static void hardware_disable_nolock(void *junk)
3284 {
3285         int cpu = raw_smp_processor_id();
3286
3287         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
3288                 return;
3289         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3290         kvm_arch_hardware_disable();
3291 }
3292
3293 static int kvm_dying_cpu(unsigned int cpu)
3294 {
3295         raw_spin_lock(&kvm_count_lock);
3296         if (kvm_usage_count)
3297                 hardware_disable_nolock(NULL);
3298         raw_spin_unlock(&kvm_count_lock);
3299         return 0;
3300 }
3301
3302 static void hardware_disable_all_nolock(void)
3303 {
3304         BUG_ON(!kvm_usage_count);
3305
3306         kvm_usage_count--;
3307         if (!kvm_usage_count)
3308                 on_each_cpu(hardware_disable_nolock, NULL, 1);
3309 }
3310
3311 static void hardware_disable_all(void)
3312 {
3313         raw_spin_lock(&kvm_count_lock);
3314         hardware_disable_all_nolock();
3315         raw_spin_unlock(&kvm_count_lock);
3316 }
3317
3318 static int hardware_enable_all(void)
3319 {
3320         int r = 0;
3321
3322         raw_spin_lock(&kvm_count_lock);
3323
3324         kvm_usage_count++;
3325         if (kvm_usage_count == 1) {
3326                 atomic_set(&hardware_enable_failed, 0);
3327                 on_each_cpu(hardware_enable_nolock, NULL, 1);
3328
3329                 if (atomic_read(&hardware_enable_failed)) {
3330                         hardware_disable_all_nolock();
3331                         r = -EBUSY;
3332                 }
3333         }
3334
3335         raw_spin_unlock(&kvm_count_lock);
3336
3337         return r;
3338 }
3339
3340 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3341                       void *v)
3342 {
3343         /*
3344          * Some (well, at least mine) BIOSes hang on reboot if
3345          * in vmx root mode.
3346          *
3347          * And Intel TXT required VMX off for all cpu when system shutdown.
3348          */
3349         pr_info("kvm: exiting hardware virtualization\n");
3350         kvm_rebooting = true;
3351         on_each_cpu(hardware_disable_nolock, NULL, 1);
3352         return NOTIFY_OK;
3353 }
3354
3355 static struct notifier_block kvm_reboot_notifier = {
3356         .notifier_call = kvm_reboot,
3357         .priority = 0,
3358 };
3359
3360 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3361 {
3362         int i;
3363
3364         for (i = 0; i < bus->dev_count; i++) {
3365                 struct kvm_io_device *pos = bus->range[i].dev;
3366
3367                 kvm_iodevice_destructor(pos);
3368         }
3369         kfree(bus);
3370 }
3371
3372 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
3373                                  const struct kvm_io_range *r2)
3374 {
3375         gpa_t addr1 = r1->addr;
3376         gpa_t addr2 = r2->addr;
3377
3378         if (addr1 < addr2)
3379                 return -1;
3380
3381         /* If r2->len == 0, match the exact address.  If r2->len != 0,
3382          * accept any overlapping write.  Any order is acceptable for
3383          * overlapping ranges, because kvm_io_bus_get_first_dev ensures
3384          * we process all of them.
3385          */
3386         if (r2->len) {
3387                 addr1 += r1->len;
3388                 addr2 += r2->len;
3389         }
3390
3391         if (addr1 > addr2)
3392                 return 1;
3393
3394         return 0;
3395 }
3396
3397 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
3398 {
3399         return kvm_io_bus_cmp(p1, p2);
3400 }
3401
3402 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
3403                           gpa_t addr, int len)
3404 {
3405         bus->range[bus->dev_count++] = (struct kvm_io_range) {
3406                 .addr = addr,
3407                 .len = len,
3408                 .dev = dev,
3409         };
3410
3411         sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
3412                 kvm_io_bus_sort_cmp, NULL);
3413
3414         return 0;
3415 }
3416
3417 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,