Revert "mm, mmu_notifier: annotate mmu notifiers with blockable invalidate callbacks"
[muen/linux.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 #define KVM_COMPAT(c)   .compat_ioctl   = (c)
120 #else
121 static long kvm_no_compat_ioctl(struct file *file, unsigned int ioctl,
122                                 unsigned long arg) { return -EINVAL; }
123 #define KVM_COMPAT(c)   .compat_ioctl   = kvm_no_compat_ioctl
124 #endif
125 static int hardware_enable_all(void);
126 static void hardware_disable_all(void);
127
128 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
129
130 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot, gfn_t gfn);
131
132 __visible bool kvm_rebooting;
133 EXPORT_SYMBOL_GPL(kvm_rebooting);
134
135 static bool largepages_enabled = true;
136
137 #define KVM_EVENT_CREATE_VM 0
138 #define KVM_EVENT_DESTROY_VM 1
139 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm);
140 static unsigned long long kvm_createvm_count;
141 static unsigned long long kvm_active_vms;
142
143 __weak int kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
144                 unsigned long start, unsigned long end, bool blockable)
145 {
146         return 0;
147 }
148
149 bool kvm_is_reserved_pfn(kvm_pfn_t pfn)
150 {
151         if (pfn_valid(pfn))
152                 return PageReserved(pfn_to_page(pfn));
153
154         return true;
155 }
156
157 /*
158  * Switches to specified vcpu, until a matching vcpu_put()
159  */
160 void vcpu_load(struct kvm_vcpu *vcpu)
161 {
162         int cpu = get_cpu();
163         preempt_notifier_register(&vcpu->preempt_notifier);
164         kvm_arch_vcpu_load(vcpu, cpu);
165         put_cpu();
166 }
167 EXPORT_SYMBOL_GPL(vcpu_load);
168
169 void vcpu_put(struct kvm_vcpu *vcpu)
170 {
171         preempt_disable();
172         kvm_arch_vcpu_put(vcpu);
173         preempt_notifier_unregister(&vcpu->preempt_notifier);
174         preempt_enable();
175 }
176 EXPORT_SYMBOL_GPL(vcpu_put);
177
178 /* TODO: merge with kvm_arch_vcpu_should_kick */
179 static bool kvm_request_needs_ipi(struct kvm_vcpu *vcpu, unsigned req)
180 {
181         int mode = kvm_vcpu_exiting_guest_mode(vcpu);
182
183         /*
184          * We need to wait for the VCPU to reenable interrupts and get out of
185          * READING_SHADOW_PAGE_TABLES mode.
186          */
187         if (req & KVM_REQUEST_WAIT)
188                 return mode != OUTSIDE_GUEST_MODE;
189
190         /*
191          * Need to kick a running VCPU, but otherwise there is nothing to do.
192          */
193         return mode == IN_GUEST_MODE;
194 }
195
196 static void ack_flush(void *_completed)
197 {
198 }
199
200 static inline bool kvm_kick_many_cpus(const struct cpumask *cpus, bool wait)
201 {
202         if (unlikely(!cpus))
203                 cpus = cpu_online_mask;
204
205         if (cpumask_empty(cpus))
206                 return false;
207
208         smp_call_function_many(cpus, ack_flush, NULL, wait);
209         return true;
210 }
211
212 bool kvm_make_vcpus_request_mask(struct kvm *kvm, unsigned int req,
213                                  unsigned long *vcpu_bitmap, cpumask_var_t tmp)
214 {
215         int i, cpu, me;
216         struct kvm_vcpu *vcpu;
217         bool called;
218
219         me = get_cpu();
220
221         kvm_for_each_vcpu(i, vcpu, kvm) {
222                 if (vcpu_bitmap && !test_bit(i, vcpu_bitmap))
223                         continue;
224
225                 kvm_make_request(req, vcpu);
226                 cpu = vcpu->cpu;
227
228                 if (!(req & KVM_REQUEST_NO_WAKEUP) && kvm_vcpu_wake_up(vcpu))
229                         continue;
230
231                 if (tmp != NULL && cpu != -1 && cpu != me &&
232                     kvm_request_needs_ipi(vcpu, req))
233                         __cpumask_set_cpu(cpu, tmp);
234         }
235
236         called = kvm_kick_many_cpus(tmp, !!(req & KVM_REQUEST_WAIT));
237         put_cpu();
238
239         return called;
240 }
241
242 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
243 {
244         cpumask_var_t cpus;
245         bool called;
246
247         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
248
249         called = kvm_make_vcpus_request_mask(kvm, req, NULL, cpus);
250
251         free_cpumask_var(cpus);
252         return called;
253 }
254
255 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
256 void kvm_flush_remote_tlbs(struct kvm *kvm)
257 {
258         /*
259          * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in
260          * kvm_make_all_cpus_request.
261          */
262         long dirty_count = smp_load_acquire(&kvm->tlbs_dirty);
263
264         /*
265          * We want to publish modifications to the page tables before reading
266          * mode. Pairs with a memory barrier in arch-specific code.
267          * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
268          * and smp_mb in walk_shadow_page_lockless_begin/end.
269          * - powerpc: smp_mb in kvmppc_prepare_to_enter.
270          *
271          * There is already an smp_mb__after_atomic() before
272          * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
273          * barrier here.
274          */
275         if (!kvm_arch_flush_remote_tlb(kvm)
276             || kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
277                 ++kvm->stat.remote_tlb_flush;
278         cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
279 }
280 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
281 #endif
282
283 void kvm_reload_remote_mmus(struct kvm *kvm)
284 {
285         kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
286 }
287
288 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
289 {
290         struct page *page;
291         int r;
292
293         mutex_init(&vcpu->mutex);
294         vcpu->cpu = -1;
295         vcpu->kvm = kvm;
296         vcpu->vcpu_id = id;
297         vcpu->pid = NULL;
298         init_swait_queue_head(&vcpu->wq);
299         kvm_async_pf_vcpu_init(vcpu);
300
301         vcpu->pre_pcpu = -1;
302         INIT_LIST_HEAD(&vcpu->blocked_vcpu_list);
303
304         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
305         if (!page) {
306                 r = -ENOMEM;
307                 goto fail;
308         }
309         vcpu->run = page_address(page);
310
311         kvm_vcpu_set_in_spin_loop(vcpu, false);
312         kvm_vcpu_set_dy_eligible(vcpu, false);
313         vcpu->preempted = false;
314
315         r = kvm_arch_vcpu_init(vcpu);
316         if (r < 0)
317                 goto fail_free_run;
318         return 0;
319
320 fail_free_run:
321         free_page((unsigned long)vcpu->run);
322 fail:
323         return r;
324 }
325 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
326
327 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
328 {
329         /*
330          * no need for rcu_read_lock as VCPU_RUN is the only place that
331          * will change the vcpu->pid pointer and on uninit all file
332          * descriptors are already gone.
333          */
334         put_pid(rcu_dereference_protected(vcpu->pid, 1));
335         kvm_arch_vcpu_uninit(vcpu);
336         free_page((unsigned long)vcpu->run);
337 }
338 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
339
340 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
341 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
342 {
343         return container_of(mn, struct kvm, mmu_notifier);
344 }
345
346 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
347                                         struct mm_struct *mm,
348                                         unsigned long address,
349                                         pte_t pte)
350 {
351         struct kvm *kvm = mmu_notifier_to_kvm(mn);
352         int idx;
353
354         idx = srcu_read_lock(&kvm->srcu);
355         spin_lock(&kvm->mmu_lock);
356         kvm->mmu_notifier_seq++;
357         kvm_set_spte_hva(kvm, address, pte);
358         spin_unlock(&kvm->mmu_lock);
359         srcu_read_unlock(&kvm->srcu, idx);
360 }
361
362 static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
363                                                     struct mm_struct *mm,
364                                                     unsigned long start,
365                                                     unsigned long end,
366                                                     bool blockable)
367 {
368         struct kvm *kvm = mmu_notifier_to_kvm(mn);
369         int need_tlb_flush = 0, idx;
370         int ret;
371
372         idx = srcu_read_lock(&kvm->srcu);
373         spin_lock(&kvm->mmu_lock);
374         /*
375          * The count increase must become visible at unlock time as no
376          * spte can be established without taking the mmu_lock and
377          * count is also read inside the mmu_lock critical section.
378          */
379         kvm->mmu_notifier_count++;
380         need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
381         need_tlb_flush |= kvm->tlbs_dirty;
382         /* we've to flush the tlb before the pages can be freed */
383         if (need_tlb_flush)
384                 kvm_flush_remote_tlbs(kvm);
385
386         spin_unlock(&kvm->mmu_lock);
387
388         ret = kvm_arch_mmu_notifier_invalidate_range(kvm, start, end, blockable);
389
390         srcu_read_unlock(&kvm->srcu, idx);
391
392         return ret;
393 }
394
395 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
396                                                   struct mm_struct *mm,
397                                                   unsigned long start,
398                                                   unsigned long end)
399 {
400         struct kvm *kvm = mmu_notifier_to_kvm(mn);
401
402         spin_lock(&kvm->mmu_lock);
403         /*
404          * This sequence increase will notify the kvm page fault that
405          * the page that is going to be mapped in the spte could have
406          * been freed.
407          */
408         kvm->mmu_notifier_seq++;
409         smp_wmb();
410         /*
411          * The above sequence increase must be visible before the
412          * below count decrease, which is ensured by the smp_wmb above
413          * in conjunction with the smp_rmb in mmu_notifier_retry().
414          */
415         kvm->mmu_notifier_count--;
416         spin_unlock(&kvm->mmu_lock);
417
418         BUG_ON(kvm->mmu_notifier_count < 0);
419 }
420
421 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
422                                               struct mm_struct *mm,
423                                               unsigned long start,
424                                               unsigned long end)
425 {
426         struct kvm *kvm = mmu_notifier_to_kvm(mn);
427         int young, idx;
428
429         idx = srcu_read_lock(&kvm->srcu);
430         spin_lock(&kvm->mmu_lock);
431
432         young = kvm_age_hva(kvm, start, end);
433         if (young)
434                 kvm_flush_remote_tlbs(kvm);
435
436         spin_unlock(&kvm->mmu_lock);
437         srcu_read_unlock(&kvm->srcu, idx);
438
439         return young;
440 }
441
442 static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
443                                         struct mm_struct *mm,
444                                         unsigned long start,
445                                         unsigned long end)
446 {
447         struct kvm *kvm = mmu_notifier_to_kvm(mn);
448         int young, idx;
449
450         idx = srcu_read_lock(&kvm->srcu);
451         spin_lock(&kvm->mmu_lock);
452         /*
453          * Even though we do not flush TLB, this will still adversely
454          * affect performance on pre-Haswell Intel EPT, where there is
455          * no EPT Access Bit to clear so that we have to tear down EPT
456          * tables instead. If we find this unacceptable, we can always
457          * add a parameter to kvm_age_hva so that it effectively doesn't
458          * do anything on clear_young.
459          *
460          * Also note that currently we never issue secondary TLB flushes
461          * from clear_young, leaving this job up to the regular system
462          * cadence. If we find this inaccurate, we might come up with a
463          * more sophisticated heuristic later.
464          */
465         young = kvm_age_hva(kvm, start, end);
466         spin_unlock(&kvm->mmu_lock);
467         srcu_read_unlock(&kvm->srcu, idx);
468
469         return young;
470 }
471
472 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
473                                        struct mm_struct *mm,
474                                        unsigned long address)
475 {
476         struct kvm *kvm = mmu_notifier_to_kvm(mn);
477         int young, idx;
478
479         idx = srcu_read_lock(&kvm->srcu);
480         spin_lock(&kvm->mmu_lock);
481         young = kvm_test_age_hva(kvm, address);
482         spin_unlock(&kvm->mmu_lock);
483         srcu_read_unlock(&kvm->srcu, idx);
484
485         return young;
486 }
487
488 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
489                                      struct mm_struct *mm)
490 {
491         struct kvm *kvm = mmu_notifier_to_kvm(mn);
492         int idx;
493
494         idx = srcu_read_lock(&kvm->srcu);
495         kvm_arch_flush_shadow_all(kvm);
496         srcu_read_unlock(&kvm->srcu, idx);
497 }
498
499 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
500         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
501         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
502         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
503         .clear_young            = kvm_mmu_notifier_clear_young,
504         .test_young             = kvm_mmu_notifier_test_young,
505         .change_pte             = kvm_mmu_notifier_change_pte,
506         .release                = kvm_mmu_notifier_release,
507 };
508
509 static int kvm_init_mmu_notifier(struct kvm *kvm)
510 {
511         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
512         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
513 }
514
515 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
516
517 static int kvm_init_mmu_notifier(struct kvm *kvm)
518 {
519         return 0;
520 }
521
522 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
523
524 static struct kvm_memslots *kvm_alloc_memslots(void)
525 {
526         int i;
527         struct kvm_memslots *slots;
528
529         slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
530         if (!slots)
531                 return NULL;
532
533         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
534                 slots->id_to_index[i] = slots->memslots[i].id = i;
535
536         return slots;
537 }
538
539 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
540 {
541         if (!memslot->dirty_bitmap)
542                 return;
543
544         kvfree(memslot->dirty_bitmap);
545         memslot->dirty_bitmap = NULL;
546 }
547
548 /*
549  * Free any memory in @free but not in @dont.
550  */
551 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
552                               struct kvm_memory_slot *dont)
553 {
554         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
555                 kvm_destroy_dirty_bitmap(free);
556
557         kvm_arch_free_memslot(kvm, free, dont);
558
559         free->npages = 0;
560 }
561
562 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
563 {
564         struct kvm_memory_slot *memslot;
565
566         if (!slots)
567                 return;
568
569         kvm_for_each_memslot(memslot, slots)
570                 kvm_free_memslot(kvm, memslot, NULL);
571
572         kvfree(slots);
573 }
574
575 static void kvm_destroy_vm_debugfs(struct kvm *kvm)
576 {
577         int i;
578
579         if (!kvm->debugfs_dentry)
580                 return;
581
582         debugfs_remove_recursive(kvm->debugfs_dentry);
583
584         if (kvm->debugfs_stat_data) {
585                 for (i = 0; i < kvm_debugfs_num_entries; i++)
586                         kfree(kvm->debugfs_stat_data[i]);
587                 kfree(kvm->debugfs_stat_data);
588         }
589 }
590
591 static int kvm_create_vm_debugfs(struct kvm *kvm, int fd)
592 {
593         char dir_name[ITOA_MAX_LEN * 2];
594         struct kvm_stat_data *stat_data;
595         struct kvm_stats_debugfs_item *p;
596
597         if (!debugfs_initialized())
598                 return 0;
599
600         snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd);
601         kvm->debugfs_dentry = debugfs_create_dir(dir_name, kvm_debugfs_dir);
602
603         kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries,
604                                          sizeof(*kvm->debugfs_stat_data),
605                                          GFP_KERNEL);
606         if (!kvm->debugfs_stat_data)
607                 return -ENOMEM;
608
609         for (p = debugfs_entries; p->name; p++) {
610                 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL);
611                 if (!stat_data)
612                         return -ENOMEM;
613
614                 stat_data->kvm = kvm;
615                 stat_data->offset = p->offset;
616                 kvm->debugfs_stat_data[p - debugfs_entries] = stat_data;
617                 debugfs_create_file(p->name, 0644, kvm->debugfs_dentry,
618                                     stat_data, stat_fops_per_vm[p->kind]);
619         }
620         return 0;
621 }
622
623 static struct kvm *kvm_create_vm(unsigned long type)
624 {
625         int r, i;
626         struct kvm *kvm = kvm_arch_alloc_vm();
627
628         if (!kvm)
629                 return ERR_PTR(-ENOMEM);
630
631         spin_lock_init(&kvm->mmu_lock);
632         mmgrab(current->mm);
633         kvm->mm = current->mm;
634         kvm_eventfd_init(kvm);
635         mutex_init(&kvm->lock);
636         mutex_init(&kvm->irq_lock);
637         mutex_init(&kvm->slots_lock);
638         refcount_set(&kvm->users_count, 1);
639         INIT_LIST_HEAD(&kvm->devices);
640
641         r = kvm_arch_init_vm(kvm, type);
642         if (r)
643                 goto out_err_no_disable;
644
645         r = hardware_enable_all();
646         if (r)
647                 goto out_err_no_disable;
648
649 #ifdef CONFIG_HAVE_KVM_IRQFD
650         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
651 #endif
652
653         BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
654
655         r = -ENOMEM;
656         for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
657                 struct kvm_memslots *slots = kvm_alloc_memslots();
658                 if (!slots)
659                         goto out_err_no_srcu;
660                 /*
661                  * Generations must be different for each address space.
662                  * Init kvm generation close to the maximum to easily test the
663                  * code of handling generation number wrap-around.
664                  */
665                 slots->generation = i * 2 - 150;
666                 rcu_assign_pointer(kvm->memslots[i], slots);
667         }
668
669         if (init_srcu_struct(&kvm->srcu))
670                 goto out_err_no_srcu;
671         if (init_srcu_struct(&kvm->irq_srcu))
672                 goto out_err_no_irq_srcu;
673         for (i = 0; i < KVM_NR_BUSES; i++) {
674                 rcu_assign_pointer(kvm->buses[i],
675                         kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL));
676                 if (!kvm->buses[i])
677                         goto out_err;
678         }
679
680         r = kvm_init_mmu_notifier(kvm);
681         if (r)
682                 goto out_err;
683
684         spin_lock(&kvm_lock);
685         list_add(&kvm->vm_list, &vm_list);
686         spin_unlock(&kvm_lock);
687
688         preempt_notifier_inc();
689
690         return kvm;
691
692 out_err:
693         cleanup_srcu_struct(&kvm->irq_srcu);
694 out_err_no_irq_srcu:
695         cleanup_srcu_struct(&kvm->srcu);
696 out_err_no_srcu:
697         hardware_disable_all();
698 out_err_no_disable:
699         refcount_set(&kvm->users_count, 0);
700         for (i = 0; i < KVM_NR_BUSES; i++)
701                 kfree(kvm_get_bus(kvm, i));
702         for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
703                 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
704         kvm_arch_free_vm(kvm);
705         mmdrop(current->mm);
706         return ERR_PTR(r);
707 }
708
709 static void kvm_destroy_devices(struct kvm *kvm)
710 {
711         struct kvm_device *dev, *tmp;
712
713         /*
714          * We do not need to take the kvm->lock here, because nobody else
715          * has a reference to the struct kvm at this point and therefore
716          * cannot access the devices list anyhow.
717          */
718         list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
719                 list_del(&dev->vm_node);
720                 dev->ops->destroy(dev);
721         }
722 }
723
724 static void kvm_destroy_vm(struct kvm *kvm)
725 {
726         int i;
727         struct mm_struct *mm = kvm->mm;
728
729         kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm);
730         kvm_destroy_vm_debugfs(kvm);
731         kvm_arch_sync_events(kvm);
732         spin_lock(&kvm_lock);
733         list_del(&kvm->vm_list);
734         spin_unlock(&kvm_lock);
735         kvm_free_irq_routing(kvm);
736         for (i = 0; i < KVM_NR_BUSES; i++) {
737                 struct kvm_io_bus *bus = kvm_get_bus(kvm, i);
738
739                 if (bus)
740                         kvm_io_bus_destroy(bus);
741                 kvm->buses[i] = NULL;
742         }
743         kvm_coalesced_mmio_free(kvm);
744 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
745         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
746 #else
747         kvm_arch_flush_shadow_all(kvm);
748 #endif
749         kvm_arch_destroy_vm(kvm);
750         kvm_destroy_devices(kvm);
751         for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
752                 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
753         cleanup_srcu_struct(&kvm->irq_srcu);
754         cleanup_srcu_struct(&kvm->srcu);
755         kvm_arch_free_vm(kvm);
756         preempt_notifier_dec();
757         hardware_disable_all();
758         mmdrop(mm);
759 }
760
761 void kvm_get_kvm(struct kvm *kvm)
762 {
763         refcount_inc(&kvm->users_count);
764 }
765 EXPORT_SYMBOL_GPL(kvm_get_kvm);
766
767 void kvm_put_kvm(struct kvm *kvm)
768 {
769         if (refcount_dec_and_test(&kvm->users_count))
770                 kvm_destroy_vm(kvm);
771 }
772 EXPORT_SYMBOL_GPL(kvm_put_kvm);
773
774
775 static int kvm_vm_release(struct inode *inode, struct file *filp)
776 {
777         struct kvm *kvm = filp->private_data;
778
779         kvm_irqfd_release(kvm);
780
781         kvm_put_kvm(kvm);
782         return 0;
783 }
784
785 /*
786  * Allocation size is twice as large as the actual dirty bitmap size.
787  * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
788  */
789 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
790 {
791         unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
792
793         memslot->dirty_bitmap = kvzalloc(dirty_bytes, GFP_KERNEL);
794         if (!memslot->dirty_bitmap)
795                 return -ENOMEM;
796
797         return 0;
798 }
799
800 /*
801  * Insert memslot and re-sort memslots based on their GFN,
802  * so binary search could be used to lookup GFN.
803  * Sorting algorithm takes advantage of having initially
804  * sorted array and known changed memslot position.
805  */
806 static void update_memslots(struct kvm_memslots *slots,
807                             struct kvm_memory_slot *new,
808                             enum kvm_mr_change change)
809 {
810         int id = new->id;
811         int i = slots->id_to_index[id];
812         struct kvm_memory_slot *mslots = slots->memslots;
813
814         WARN_ON(mslots[i].id != id);
815         switch (change) {
816         case KVM_MR_CREATE:
817                 slots->used_slots++;
818                 WARN_ON(mslots[i].npages || !new->npages);
819                 break;
820         case KVM_MR_DELETE:
821                 slots->used_slots--;
822                 WARN_ON(new->npages || !mslots[i].npages);
823                 break;
824         default:
825                 break;
826         }
827
828         while (i < KVM_MEM_SLOTS_NUM - 1 &&
829                new->base_gfn <= mslots[i + 1].base_gfn) {
830                 if (!mslots[i + 1].npages)
831                         break;
832                 mslots[i] = mslots[i + 1];
833                 slots->id_to_index[mslots[i].id] = i;
834                 i++;
835         }
836
837         /*
838          * The ">=" is needed when creating a slot with base_gfn == 0,
839          * so that it moves before all those with base_gfn == npages == 0.
840          *
841          * On the other hand, if new->npages is zero, the above loop has
842          * already left i pointing to the beginning of the empty part of
843          * mslots, and the ">=" would move the hole backwards in this
844          * case---which is wrong.  So skip the loop when deleting a slot.
845          */
846         if (new->npages) {
847                 while (i > 0 &&
848                        new->base_gfn >= mslots[i - 1].base_gfn) {
849                         mslots[i] = mslots[i - 1];
850                         slots->id_to_index[mslots[i].id] = i;
851                         i--;
852                 }
853         } else
854                 WARN_ON_ONCE(i != slots->used_slots);
855
856         mslots[i] = *new;
857         slots->id_to_index[mslots[i].id] = i;
858 }
859
860 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
861 {
862         u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
863
864 #ifdef __KVM_HAVE_READONLY_MEM
865         valid_flags |= KVM_MEM_READONLY;
866 #endif
867
868         if (mem->flags & ~valid_flags)
869                 return -EINVAL;
870
871         return 0;
872 }
873
874 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
875                 int as_id, struct kvm_memslots *slots)
876 {
877         struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id);
878
879         /*
880          * Set the low bit in the generation, which disables SPTE caching
881          * until the end of synchronize_srcu_expedited.
882          */
883         WARN_ON(old_memslots->generation & 1);
884         slots->generation = old_memslots->generation + 1;
885
886         rcu_assign_pointer(kvm->memslots[as_id], slots);
887         synchronize_srcu_expedited(&kvm->srcu);
888
889         /*
890          * Increment the new memslot generation a second time. This prevents
891          * vm exits that race with memslot updates from caching a memslot
892          * generation that will (potentially) be valid forever.
893          *
894          * Generations must be unique even across address spaces.  We do not need
895          * a global counter for that, instead the generation space is evenly split
896          * across address spaces.  For example, with two address spaces, address
897          * space 0 will use generations 0, 4, 8, ... while * address space 1 will
898          * use generations 2, 6, 10, 14, ...
899          */
900         slots->generation += KVM_ADDRESS_SPACE_NUM * 2 - 1;
901
902         kvm_arch_memslots_updated(kvm, slots);
903
904         return old_memslots;
905 }
906
907 /*
908  * Allocate some memory and give it an address in the guest physical address
909  * space.
910  *
911  * Discontiguous memory is allowed, mostly for framebuffers.
912  *
913  * Must be called holding kvm->slots_lock for write.
914  */
915 int __kvm_set_memory_region(struct kvm *kvm,
916                             const struct kvm_userspace_memory_region *mem)
917 {
918         int r;
919         gfn_t base_gfn;
920         unsigned long npages;
921         struct kvm_memory_slot *slot;
922         struct kvm_memory_slot old, new;
923         struct kvm_memslots *slots = NULL, *old_memslots;
924         int as_id, id;
925         enum kvm_mr_change change;
926
927         r = check_memory_region_flags(mem);
928         if (r)
929                 goto out;
930
931         r = -EINVAL;
932         as_id = mem->slot >> 16;
933         id = (u16)mem->slot;
934
935         /* General sanity checks */
936         if (mem->memory_size & (PAGE_SIZE - 1))
937                 goto out;
938         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
939                 goto out;
940         /* We can read the guest memory with __xxx_user() later on. */
941         if ((id < KVM_USER_MEM_SLOTS) &&
942             ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
943              !access_ok(VERIFY_WRITE,
944                         (void __user *)(unsigned long)mem->userspace_addr,
945                         mem->memory_size)))
946                 goto out;
947         if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
948                 goto out;
949         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
950                 goto out;
951
952         slot = id_to_memslot(__kvm_memslots(kvm, as_id), id);
953         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
954         npages = mem->memory_size >> PAGE_SHIFT;
955
956         if (npages > KVM_MEM_MAX_NR_PAGES)
957                 goto out;
958
959         new = old = *slot;
960
961         new.id = id;
962         new.base_gfn = base_gfn;
963         new.npages = npages;
964         new.flags = mem->flags;
965
966         if (npages) {
967                 if (!old.npages)
968                         change = KVM_MR_CREATE;
969                 else { /* Modify an existing slot. */
970                         if ((mem->userspace_addr != old.userspace_addr) ||
971                             (npages != old.npages) ||
972                             ((new.flags ^ old.flags) & KVM_MEM_READONLY))
973                                 goto out;
974
975                         if (base_gfn != old.base_gfn)
976                                 change = KVM_MR_MOVE;
977                         else if (new.flags != old.flags)
978                                 change = KVM_MR_FLAGS_ONLY;
979                         else { /* Nothing to change. */
980                                 r = 0;
981                                 goto out;
982                         }
983                 }
984         } else {
985                 if (!old.npages)
986                         goto out;
987
988                 change = KVM_MR_DELETE;
989                 new.base_gfn = 0;
990                 new.flags = 0;
991         }
992
993         if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
994                 /* Check for overlaps */
995                 r = -EEXIST;
996                 kvm_for_each_memslot(slot, __kvm_memslots(kvm, as_id)) {
997                         if (slot->id == id)
998                                 continue;
999                         if (!((base_gfn + npages <= slot->base_gfn) ||
1000                               (base_gfn >= slot->base_gfn + slot->npages)))
1001                                 goto out;
1002                 }
1003         }
1004
1005         /* Free page dirty bitmap if unneeded */
1006         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
1007                 new.dirty_bitmap = NULL;
1008
1009         r = -ENOMEM;
1010         if (change == KVM_MR_CREATE) {
1011                 new.userspace_addr = mem->userspace_addr;
1012
1013                 if (kvm_arch_create_memslot(kvm, &new, npages))
1014                         goto out_free;
1015         }
1016
1017         /* Allocate page dirty bitmap if needed */
1018         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
1019                 if (kvm_create_dirty_bitmap(&new) < 0)
1020                         goto out_free;
1021         }
1022
1023         slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
1024         if (!slots)
1025                 goto out_free;
1026         memcpy(slots, __kvm_memslots(kvm, as_id), sizeof(struct kvm_memslots));
1027
1028         if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
1029                 slot = id_to_memslot(slots, id);
1030                 slot->flags |= KVM_MEMSLOT_INVALID;
1031
1032                 old_memslots = install_new_memslots(kvm, as_id, slots);
1033
1034                 /* From this point no new shadow pages pointing to a deleted,
1035                  * or moved, memslot will be created.
1036                  *
1037                  * validation of sp->gfn happens in:
1038                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1039                  *      - kvm_is_visible_gfn (mmu_check_roots)
1040                  */
1041                 kvm_arch_flush_shadow_memslot(kvm, slot);
1042
1043                 /*
1044                  * We can re-use the old_memslots from above, the only difference
1045                  * from the currently installed memslots is the invalid flag.  This
1046                  * will get overwritten by update_memslots anyway.
1047                  */
1048                 slots = old_memslots;
1049         }
1050
1051         r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
1052         if (r)
1053                 goto out_slots;
1054
1055         /* actual memory is freed via old in kvm_free_memslot below */
1056         if (change == KVM_MR_DELETE) {
1057                 new.dirty_bitmap = NULL;
1058                 memset(&new.arch, 0, sizeof(new.arch));
1059         }
1060
1061         update_memslots(slots, &new, change);
1062         old_memslots = install_new_memslots(kvm, as_id, slots);
1063
1064         kvm_arch_commit_memory_region(kvm, mem, &old, &new, change);
1065
1066         kvm_free_memslot(kvm, &old, &new);
1067         kvfree(old_memslots);
1068         return 0;
1069
1070 out_slots:
1071         kvfree(slots);
1072 out_free:
1073         kvm_free_memslot(kvm, &new, &old);
1074 out:
1075         return r;
1076 }
1077 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1078
1079 int kvm_set_memory_region(struct kvm *kvm,
1080                           const struct kvm_userspace_memory_region *mem)
1081 {
1082         int r;
1083
1084         mutex_lock(&kvm->slots_lock);
1085         r = __kvm_set_memory_region(kvm, mem);
1086         mutex_unlock(&kvm->slots_lock);
1087         return r;
1088 }
1089 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1090
1091 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1092                                           struct kvm_userspace_memory_region *mem)
1093 {
1094         if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1095                 return -EINVAL;
1096
1097         return kvm_set_memory_region(kvm, mem);
1098 }
1099
1100 int kvm_get_dirty_log(struct kvm *kvm,
1101                         struct kvm_dirty_log *log, int *is_dirty)
1102 {
1103         struct kvm_memslots *slots;
1104         struct kvm_memory_slot *memslot;
1105         int i, as_id, id;
1106         unsigned long n;
1107         unsigned long any = 0;
1108
1109         as_id = log->slot >> 16;
1110         id = (u16)log->slot;
1111         if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1112                 return -EINVAL;
1113
1114         slots = __kvm_memslots(kvm, as_id);
1115         memslot = id_to_memslot(slots, id);
1116         if (!memslot->dirty_bitmap)
1117                 return -ENOENT;
1118
1119         n = kvm_dirty_bitmap_bytes(memslot);
1120
1121         for (i = 0; !any && i < n/sizeof(long); ++i)
1122                 any = memslot->dirty_bitmap[i];
1123
1124         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1125                 return -EFAULT;
1126
1127         if (any)
1128                 *is_dirty = 1;
1129         return 0;
1130 }
1131 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1132
1133 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1134 /**
1135  * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages
1136  *      are dirty write protect them for next write.
1137  * @kvm:        pointer to kvm instance
1138  * @log:        slot id and address to which we copy the log
1139  * @is_dirty:   flag set if any page is dirty
1140  *
1141  * We need to keep it in mind that VCPU threads can write to the bitmap
1142  * concurrently. So, to avoid losing track of dirty pages we keep the
1143  * following order:
1144  *
1145  *    1. Take a snapshot of the bit and clear it if needed.
1146  *    2. Write protect the corresponding page.
1147  *    3. Copy the snapshot to the userspace.
1148  *    4. Upon return caller flushes TLB's if needed.
1149  *
1150  * Between 2 and 4, the guest may write to the page using the remaining TLB
1151  * entry.  This is not a problem because the page is reported dirty using
1152  * the snapshot taken before and step 4 ensures that writes done after
1153  * exiting to userspace will be logged for the next call.
1154  *
1155  */
1156 int kvm_get_dirty_log_protect(struct kvm *kvm,
1157                         struct kvm_dirty_log *log, bool *is_dirty)
1158 {
1159         struct kvm_memslots *slots;
1160         struct kvm_memory_slot *memslot;
1161         int i, as_id, id;
1162         unsigned long n;
1163         unsigned long *dirty_bitmap;
1164         unsigned long *dirty_bitmap_buffer;
1165
1166         as_id = log->slot >> 16;
1167         id = (u16)log->slot;
1168         if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1169                 return -EINVAL;
1170
1171         slots = __kvm_memslots(kvm, as_id);
1172         memslot = id_to_memslot(slots, id);
1173
1174         dirty_bitmap = memslot->dirty_bitmap;
1175         if (!dirty_bitmap)
1176                 return -ENOENT;
1177
1178         n = kvm_dirty_bitmap_bytes(memslot);
1179
1180         dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
1181         memset(dirty_bitmap_buffer, 0, n);
1182
1183         spin_lock(&kvm->mmu_lock);
1184         *is_dirty = false;
1185         for (i = 0; i < n / sizeof(long); i++) {
1186                 unsigned long mask;
1187                 gfn_t offset;
1188
1189                 if (!dirty_bitmap[i])
1190                         continue;
1191
1192                 *is_dirty = true;
1193
1194                 mask = xchg(&dirty_bitmap[i], 0);
1195                 dirty_bitmap_buffer[i] = mask;
1196
1197                 if (mask) {
1198                         offset = i * BITS_PER_LONG;
1199                         kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1200                                                                 offset, mask);
1201                 }
1202         }
1203
1204         spin_unlock(&kvm->mmu_lock);
1205         if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1206                 return -EFAULT;
1207         return 0;
1208 }
1209 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect);
1210 #endif
1211
1212 bool kvm_largepages_enabled(void)
1213 {
1214         return largepages_enabled;
1215 }
1216
1217 void kvm_disable_largepages(void)
1218 {
1219         largepages_enabled = false;
1220 }
1221 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1222
1223 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1224 {
1225         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1226 }
1227 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1228
1229 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
1230 {
1231         return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn);
1232 }
1233
1234 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1235 {
1236         struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1237
1238         if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1239               memslot->flags & KVM_MEMSLOT_INVALID)
1240                 return false;
1241
1242         return true;
1243 }
1244 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1245
1246 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1247 {
1248         struct vm_area_struct *vma;
1249         unsigned long addr, size;
1250
1251         size = PAGE_SIZE;
1252
1253         addr = gfn_to_hva(kvm, gfn);
1254         if (kvm_is_error_hva(addr))
1255                 return PAGE_SIZE;
1256
1257         down_read(&current->mm->mmap_sem);
1258         vma = find_vma(current->mm, addr);
1259         if (!vma)
1260                 goto out;
1261
1262         size = vma_kernel_pagesize(vma);
1263
1264 out:
1265         up_read(&current->mm->mmap_sem);
1266
1267         return size;
1268 }
1269
1270 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1271 {
1272         return slot->flags & KVM_MEM_READONLY;
1273 }
1274
1275 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1276                                        gfn_t *nr_pages, bool write)
1277 {
1278         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1279                 return KVM_HVA_ERR_BAD;
1280
1281         if (memslot_is_readonly(slot) && write)
1282                 return KVM_HVA_ERR_RO_BAD;
1283
1284         if (nr_pages)
1285                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1286
1287         return __gfn_to_hva_memslot(slot, gfn);
1288 }
1289
1290 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1291                                      gfn_t *nr_pages)
1292 {
1293         return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1294 }
1295
1296 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1297                                         gfn_t gfn)
1298 {
1299         return gfn_to_hva_many(slot, gfn, NULL);
1300 }
1301 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1302
1303 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1304 {
1305         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1306 }
1307 EXPORT_SYMBOL_GPL(gfn_to_hva);
1308
1309 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
1310 {
1311         return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
1312 }
1313 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
1314
1315 /*
1316  * Return the hva of a @gfn and the R/W attribute if possible.
1317  *
1318  * @slot: the kvm_memory_slot which contains @gfn
1319  * @gfn: the gfn to be translated
1320  * @writable: used to return the read/write attribute of the @slot if the hva
1321  * is valid and @writable is not NULL
1322  */
1323 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1324                                       gfn_t gfn, bool *writable)
1325 {
1326         unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1327
1328         if (!kvm_is_error_hva(hva) && writable)
1329                 *writable = !memslot_is_readonly(slot);
1330
1331         return hva;
1332 }
1333
1334 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1335 {
1336         struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1337
1338         return gfn_to_hva_memslot_prot(slot, gfn, writable);
1339 }
1340
1341 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
1342 {
1343         struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1344
1345         return gfn_to_hva_memslot_prot(slot, gfn, writable);
1346 }
1347
1348 static inline int check_user_page_hwpoison(unsigned long addr)
1349 {
1350         int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
1351
1352         rc = get_user_pages(addr, 1, flags, NULL, NULL);
1353         return rc == -EHWPOISON;
1354 }
1355
1356 /*
1357  * The fast path to get the writable pfn which will be stored in @pfn,
1358  * true indicates success, otherwise false is returned.  It's also the
1359  * only part that runs if we can are in atomic context.
1360  */
1361 static bool hva_to_pfn_fast(unsigned long addr, bool write_fault,
1362                             bool *writable, kvm_pfn_t *pfn)
1363 {
1364         struct page *page[1];
1365         int npages;
1366
1367         /*
1368          * Fast pin a writable pfn only if it is a write fault request
1369          * or the caller allows to map a writable pfn for a read fault
1370          * request.
1371          */
1372         if (!(write_fault || writable))
1373                 return false;
1374
1375         npages = __get_user_pages_fast(addr, 1, 1, page);
1376         if (npages == 1) {
1377                 *pfn = page_to_pfn(page[0]);
1378
1379                 if (writable)
1380                         *writable = true;
1381                 return true;
1382         }
1383
1384         return false;
1385 }
1386
1387 /*
1388  * The slow path to get the pfn of the specified host virtual address,
1389  * 1 indicates success, -errno is returned if error is detected.
1390  */
1391 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1392                            bool *writable, kvm_pfn_t *pfn)
1393 {
1394         unsigned int flags = FOLL_HWPOISON;
1395         struct page *page;
1396         int npages = 0;
1397
1398         might_sleep();
1399
1400         if (writable)
1401                 *writable = write_fault;
1402
1403         if (write_fault)
1404                 flags |= FOLL_WRITE;
1405         if (async)
1406                 flags |= FOLL_NOWAIT;
1407
1408         npages = get_user_pages_unlocked(addr, 1, &page, flags);
1409         if (npages != 1)
1410                 return npages;
1411
1412         /* map read fault as writable if possible */
1413         if (unlikely(!write_fault) && writable) {
1414                 struct page *wpage;
1415
1416                 if (__get_user_pages_fast(addr, 1, 1, &wpage) == 1) {
1417                         *writable = true;
1418                         put_page(page);
1419                         page = wpage;
1420                 }
1421         }
1422         *pfn = page_to_pfn(page);
1423         return npages;
1424 }
1425
1426 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1427 {
1428         if (unlikely(!(vma->vm_flags & VM_READ)))
1429                 return false;
1430
1431         if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1432                 return false;
1433
1434         return true;
1435 }
1436
1437 static int hva_to_pfn_remapped(struct vm_area_struct *vma,
1438                                unsigned long addr, bool *async,
1439                                bool write_fault, bool *writable,
1440                                kvm_pfn_t *p_pfn)
1441 {
1442         unsigned long pfn;
1443         int r;
1444
1445         r = follow_pfn(vma, addr, &pfn);
1446         if (r) {
1447                 /*
1448                  * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
1449                  * not call the fault handler, so do it here.
1450                  */
1451                 bool unlocked = false;
1452                 r = fixup_user_fault(current, current->mm, addr,
1453                                      (write_fault ? FAULT_FLAG_WRITE : 0),
1454                                      &unlocked);
1455                 if (unlocked)
1456                         return -EAGAIN;
1457                 if (r)
1458                         return r;
1459
1460                 r = follow_pfn(vma, addr, &pfn);
1461                 if (r)
1462                         return r;
1463
1464         }
1465
1466         if (writable)
1467                 *writable = true;
1468
1469         /*
1470          * Get a reference here because callers of *hva_to_pfn* and
1471          * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
1472          * returned pfn.  This is only needed if the VMA has VM_MIXEDMAP
1473          * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will
1474          * simply do nothing for reserved pfns.
1475          *
1476          * Whoever called remap_pfn_range is also going to call e.g.
1477          * unmap_mapping_range before the underlying pages are freed,
1478          * causing a call to our MMU notifier.
1479          */ 
1480         kvm_get_pfn(pfn);
1481
1482         *p_pfn = pfn;
1483         return 0;
1484 }
1485
1486 /*
1487  * Pin guest page in memory and return its pfn.
1488  * @addr: host virtual address which maps memory to the guest
1489  * @atomic: whether this function can sleep
1490  * @async: whether this function need to wait IO complete if the
1491  *         host page is not in the memory
1492  * @write_fault: whether we should get a writable host page
1493  * @writable: whether it allows to map a writable host page for !@write_fault
1494  *
1495  * The function will map a writable host page for these two cases:
1496  * 1): @write_fault = true
1497  * 2): @write_fault = false && @writable, @writable will tell the caller
1498  *     whether the mapping is writable.
1499  */
1500 static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1501                         bool write_fault, bool *writable)
1502 {
1503         struct vm_area_struct *vma;
1504         kvm_pfn_t pfn = 0;
1505         int npages, r;
1506
1507         /* we can do it either atomically or asynchronously, not both */
1508         BUG_ON(atomic && async);
1509
1510         if (hva_to_pfn_fast(addr, write_fault, writable, &pfn))
1511                 return pfn;
1512
1513         if (atomic)
1514                 return KVM_PFN_ERR_FAULT;
1515
1516         npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1517         if (npages == 1)
1518                 return pfn;
1519
1520         down_read(&current->mm->mmap_sem);
1521         if (npages == -EHWPOISON ||
1522               (!async && check_user_page_hwpoison(addr))) {
1523                 pfn = KVM_PFN_ERR_HWPOISON;
1524                 goto exit;
1525         }
1526
1527 retry:
1528         vma = find_vma_intersection(current->mm, addr, addr + 1);
1529
1530         if (vma == NULL)
1531                 pfn = KVM_PFN_ERR_FAULT;
1532         else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
1533                 r = hva_to_pfn_remapped(vma, addr, async, write_fault, writable, &pfn);
1534                 if (r == -EAGAIN)
1535                         goto retry;
1536                 if (r < 0)
1537                         pfn = KVM_PFN_ERR_FAULT;
1538         } else {
1539                 if (async && vma_is_valid(vma, write_fault))
1540                         *async = true;
1541                 pfn = KVM_PFN_ERR_FAULT;
1542         }
1543 exit:
1544         up_read(&current->mm->mmap_sem);
1545         return pfn;
1546 }
1547
1548 kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn,
1549                                bool atomic, bool *async, bool write_fault,
1550                                bool *writable)
1551 {
1552         unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1553
1554         if (addr == KVM_HVA_ERR_RO_BAD) {
1555                 if (writable)
1556                         *writable = false;
1557                 return KVM_PFN_ERR_RO_FAULT;
1558         }
1559
1560         if (kvm_is_error_hva(addr)) {
1561                 if (writable)
1562                         *writable = false;
1563                 return KVM_PFN_NOSLOT;
1564         }
1565
1566         /* Do not map writable pfn in the readonly memslot. */
1567         if (writable && memslot_is_readonly(slot)) {
1568                 *writable = false;
1569                 writable = NULL;
1570         }
1571
1572         return hva_to_pfn(addr, atomic, async, write_fault,
1573                           writable);
1574 }
1575 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
1576
1577 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1578                       bool *writable)
1579 {
1580         return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
1581                                     write_fault, writable);
1582 }
1583 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1584
1585 kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1586 {
1587         return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1588 }
1589 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
1590
1591 kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1592 {
1593         return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1594 }
1595 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1596
1597 kvm_pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1598 {
1599         return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm, gfn), gfn);
1600 }
1601 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1602
1603 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
1604 {
1605         return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1606 }
1607 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
1608
1609 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1610 {
1611         return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
1612 }
1613 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1614
1615 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
1616 {
1617         return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1618 }
1619 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
1620
1621 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1622                             struct page **pages, int nr_pages)
1623 {
1624         unsigned long addr;
1625         gfn_t entry = 0;
1626
1627         addr = gfn_to_hva_many(slot, gfn, &entry);
1628         if (kvm_is_error_hva(addr))
1629                 return -1;
1630
1631         if (entry < nr_pages)
1632                 return 0;
1633
1634         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1635 }
1636 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1637
1638 static struct page *kvm_pfn_to_page(kvm_pfn_t pfn)
1639 {
1640         if (is_error_noslot_pfn(pfn))
1641                 return KVM_ERR_PTR_BAD_PAGE;
1642
1643         if (kvm_is_reserved_pfn(pfn)) {
1644                 WARN_ON(1);
1645                 return KVM_ERR_PTR_BAD_PAGE;
1646         }
1647
1648         return pfn_to_page(pfn);
1649 }
1650
1651 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1652 {
1653         kvm_pfn_t pfn;
1654
1655         pfn = gfn_to_pfn(kvm, gfn);
1656
1657         return kvm_pfn_to_page(pfn);
1658 }
1659 EXPORT_SYMBOL_GPL(gfn_to_page);
1660
1661 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
1662 {
1663         kvm_pfn_t pfn;
1664
1665         pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
1666
1667         return kvm_pfn_to_page(pfn);
1668 }
1669 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
1670
1671 void kvm_release_page_clean(struct page *page)
1672 {
1673         WARN_ON(is_error_page(page));
1674
1675         kvm_release_pfn_clean(page_to_pfn(page));
1676 }
1677 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1678
1679 void kvm_release_pfn_clean(kvm_pfn_t pfn)
1680 {
1681         if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
1682                 put_page(pfn_to_page(pfn));
1683 }
1684 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1685
1686 void kvm_release_page_dirty(struct page *page)
1687 {
1688         WARN_ON(is_error_page(page));
1689
1690         kvm_release_pfn_dirty(page_to_pfn(page));
1691 }
1692 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1693
1694 void kvm_release_pfn_dirty(kvm_pfn_t pfn)
1695 {
1696         kvm_set_pfn_dirty(pfn);
1697         kvm_release_pfn_clean(pfn);
1698 }
1699 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1700
1701 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
1702 {
1703         if (!kvm_is_reserved_pfn(pfn)) {
1704                 struct page *page = pfn_to_page(pfn);
1705
1706                 if (!PageReserved(page))
1707                         SetPageDirty(page);
1708         }
1709 }
1710 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1711
1712 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
1713 {
1714         if (!kvm_is_reserved_pfn(pfn))
1715                 mark_page_accessed(pfn_to_page(pfn));
1716 }
1717 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1718
1719 void kvm_get_pfn(kvm_pfn_t pfn)
1720 {
1721         if (!kvm_is_reserved_pfn(pfn))
1722                 get_page(pfn_to_page(pfn));
1723 }
1724 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1725
1726 static int next_segment(unsigned long len, int offset)
1727 {
1728         if (len > PAGE_SIZE - offset)
1729                 return PAGE_SIZE - offset;
1730         else
1731                 return len;
1732 }
1733
1734 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
1735                                  void *data, int offset, int len)
1736 {
1737         int r;
1738         unsigned long addr;
1739
1740         addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1741         if (kvm_is_error_hva(addr))
1742                 return -EFAULT;
1743         r = __copy_from_user(data, (void __user *)addr + offset, len);
1744         if (r)
1745                 return -EFAULT;
1746         return 0;
1747 }
1748
1749 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1750                         int len)
1751 {
1752         struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1753
1754         return __kvm_read_guest_page(slot, gfn, data, offset, len);
1755 }
1756 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1757
1758 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
1759                              int offset, int len)
1760 {
1761         struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1762
1763         return __kvm_read_guest_page(slot, gfn, data, offset, len);
1764 }
1765 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
1766
1767 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1768 {
1769         gfn_t gfn = gpa >> PAGE_SHIFT;
1770         int seg;
1771         int offset = offset_in_page(gpa);
1772         int ret;
1773
1774         while ((seg = next_segment(len, offset)) != 0) {
1775                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1776                 if (ret < 0)
1777                         return ret;
1778                 offset = 0;
1779                 len -= seg;
1780                 data += seg;
1781                 ++gfn;
1782         }
1783         return 0;
1784 }
1785 EXPORT_SYMBOL_GPL(kvm_read_guest);
1786
1787 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
1788 {
1789         gfn_t gfn = gpa >> PAGE_SHIFT;
1790         int seg;
1791         int offset = offset_in_page(gpa);
1792         int ret;
1793
1794         while ((seg = next_segment(len, offset)) != 0) {
1795                 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
1796                 if (ret < 0)
1797                         return ret;
1798                 offset = 0;
1799                 len -= seg;
1800                 data += seg;
1801                 ++gfn;
1802         }
1803         return 0;
1804 }
1805 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
1806
1807 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1808                                    void *data, int offset, unsigned long len)
1809 {
1810         int r;
1811         unsigned long addr;
1812
1813         addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1814         if (kvm_is_error_hva(addr))
1815                 return -EFAULT;
1816         pagefault_disable();
1817         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1818         pagefault_enable();
1819         if (r)
1820                 return -EFAULT;
1821         return 0;
1822 }
1823
1824 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1825                           unsigned long len)
1826 {
1827         gfn_t gfn = gpa >> PAGE_SHIFT;
1828         struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1829         int offset = offset_in_page(gpa);
1830
1831         return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1832 }
1833 EXPORT_SYMBOL_GPL(kvm_read_guest_atomic);
1834
1835 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
1836                                void *data, unsigned long len)
1837 {
1838         gfn_t gfn = gpa >> PAGE_SHIFT;
1839         struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1840         int offset = offset_in_page(gpa);
1841
1842         return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1843 }
1844 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
1845
1846 static int __kvm_write_guest_page(struct kvm_memory_slot *memslot, gfn_t gfn,
1847                                   const void *data, int offset, int len)
1848 {
1849         int r;
1850         unsigned long addr;
1851
1852         addr = gfn_to_hva_memslot(memslot, gfn);
1853         if (kvm_is_error_hva(addr))
1854                 return -EFAULT;
1855         r = __copy_to_user((void __user *)addr + offset, data, len);
1856         if (r)
1857                 return -EFAULT;
1858         mark_page_dirty_in_slot(memslot, gfn);
1859         return 0;
1860 }
1861
1862 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
1863                          const void *data, int offset, int len)
1864 {
1865         struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1866
1867         return __kvm_write_guest_page(slot, gfn, data, offset, len);
1868 }
1869 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1870
1871 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
1872                               const void *data, int offset, int len)
1873 {
1874         struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1875
1876         return __kvm_write_guest_page(slot, gfn, data, offset, len);
1877 }
1878 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
1879
1880 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1881                     unsigned long len)
1882 {
1883         gfn_t gfn = gpa >> PAGE_SHIFT;
1884         int seg;
1885         int offset = offset_in_page(gpa);
1886         int ret;
1887
1888         while ((seg = next_segment(len, offset)) != 0) {
1889                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1890                 if (ret < 0)
1891                         return ret;
1892                 offset = 0;
1893                 len -= seg;
1894                 data += seg;
1895                 ++gfn;
1896         }
1897         return 0;
1898 }
1899 EXPORT_SYMBOL_GPL(kvm_write_guest);
1900
1901 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
1902                          unsigned long len)
1903 {
1904         gfn_t gfn = gpa >> PAGE_SHIFT;
1905         int seg;
1906         int offset = offset_in_page(gpa);
1907         int ret;
1908
1909         while ((seg = next_segment(len, offset)) != 0) {
1910                 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
1911                 if (ret < 0)
1912                         return ret;
1913                 offset = 0;
1914                 len -= seg;
1915                 data += seg;
1916                 ++gfn;
1917         }
1918         return 0;
1919 }
1920 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
1921
1922 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
1923                                        struct gfn_to_hva_cache *ghc,
1924                                        gpa_t gpa, unsigned long len)
1925 {
1926         int offset = offset_in_page(gpa);
1927         gfn_t start_gfn = gpa >> PAGE_SHIFT;
1928         gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1929         gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1930         gfn_t nr_pages_avail;
1931
1932         ghc->gpa = gpa;
1933         ghc->generation = slots->generation;
1934         ghc->len = len;
1935         ghc->memslot = __gfn_to_memslot(slots, start_gfn);
1936         ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, NULL);
1937         if (!kvm_is_error_hva(ghc->hva) && nr_pages_needed <= 1) {
1938                 ghc->hva += offset;
1939         } else {
1940                 /*
1941                  * If the requested region crosses two memslots, we still
1942                  * verify that the entire region is valid here.
1943                  */
1944                 while (start_gfn <= end_gfn) {
1945                         nr_pages_avail = 0;
1946                         ghc->memslot = __gfn_to_memslot(slots, start_gfn);
1947                         ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1948                                                    &nr_pages_avail);
1949                         if (kvm_is_error_hva(ghc->hva))
1950                                 return -EFAULT;
1951                         start_gfn += nr_pages_avail;
1952                 }
1953                 /* Use the slow path for cross page reads and writes. */
1954                 ghc->memslot = NULL;
1955         }
1956         return 0;
1957 }
1958
1959 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1960                               gpa_t gpa, unsigned long len)
1961 {
1962         struct kvm_memslots *slots = kvm_memslots(kvm);
1963         return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len);
1964 }
1965 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1966
1967 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1968                            void *data, int offset, unsigned long len)
1969 {
1970         struct kvm_memslots *slots = kvm_memslots(kvm);
1971         int r;
1972         gpa_t gpa = ghc->gpa + offset;
1973
1974         BUG_ON(len + offset > ghc->len);
1975
1976         if (slots->generation != ghc->generation)
1977                 __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len);
1978
1979         if (unlikely(!ghc->memslot))
1980                 return kvm_write_guest(kvm, gpa, data, len);
1981
1982         if (kvm_is_error_hva(ghc->hva))
1983                 return -EFAULT;
1984
1985         r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
1986         if (r)
1987                 return -EFAULT;
1988         mark_page_dirty_in_slot(ghc->memslot, gpa >> PAGE_SHIFT);
1989
1990         return 0;
1991 }
1992 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached);
1993
1994 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1995                            void *data, unsigned long len)
1996 {
1997         return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len);
1998 }
1999 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
2000
2001 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2002                            void *data, unsigned long len)
2003 {
2004         struct kvm_memslots *slots = kvm_memslots(kvm);
2005         int r;
2006
2007         BUG_ON(len > ghc->len);
2008
2009         if (slots->generation != ghc->generation)
2010                 __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len);
2011
2012         if (unlikely(!ghc->memslot))
2013                 return kvm_read_guest(kvm, ghc->gpa, data, len);
2014
2015         if (kvm_is_error_hva(ghc->hva))
2016                 return -EFAULT;
2017
2018         r = __copy_from_user(data, (void __user *)ghc->hva, len);
2019         if (r)
2020                 return -EFAULT;
2021
2022         return 0;
2023 }
2024 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
2025
2026 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
2027 {
2028         const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
2029
2030         return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
2031 }
2032 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
2033
2034 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
2035 {
2036         gfn_t gfn = gpa >> PAGE_SHIFT;
2037         int seg;
2038         int offset = offset_in_page(gpa);
2039         int ret;
2040
2041         while ((seg = next_segment(len, offset)) != 0) {
2042                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
2043                 if (ret < 0)
2044                         return ret;
2045                 offset = 0;
2046                 len -= seg;
2047                 ++gfn;
2048         }
2049         return 0;
2050 }
2051 EXPORT_SYMBOL_GPL(kvm_clear_guest);
2052
2053 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot,
2054                                     gfn_t gfn)
2055 {
2056         if (memslot && memslot->dirty_bitmap) {
2057                 unsigned long rel_gfn = gfn - memslot->base_gfn;
2058
2059                 set_bit_le(rel_gfn, memslot->dirty_bitmap);
2060         }
2061 }
2062
2063 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
2064 {
2065         struct kvm_memory_slot *memslot;
2066
2067         memslot = gfn_to_memslot(kvm, gfn);
2068         mark_page_dirty_in_slot(memslot, gfn);
2069 }
2070 EXPORT_SYMBOL_GPL(mark_page_dirty);
2071
2072 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
2073 {
2074         struct kvm_memory_slot *memslot;
2075
2076         memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2077         mark_page_dirty_in_slot(memslot, gfn);
2078 }
2079 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
2080
2081 void kvm_sigset_activate(struct kvm_vcpu *vcpu)
2082 {
2083         if (!vcpu->sigset_active)
2084                 return;
2085
2086         /*
2087          * This does a lockless modification of ->real_blocked, which is fine
2088          * because, only current can change ->real_blocked and all readers of
2089          * ->real_blocked don't care as long ->real_blocked is always a subset
2090          * of ->blocked.
2091          */
2092         sigprocmask(SIG_SETMASK, &vcpu->sigset, &current->real_blocked);
2093 }
2094
2095 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu)
2096 {
2097         if (!vcpu->sigset_active)
2098                 return;
2099
2100         sigprocmask(SIG_SETMASK, &current->real_blocked, NULL);
2101         sigemptyset(&current->real_blocked);
2102 }
2103
2104 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
2105 {
2106         unsigned int old, val, grow;
2107
2108         old = val = vcpu->halt_poll_ns;
2109         grow = READ_ONCE(halt_poll_ns_grow);
2110         /* 10us base */
2111         if (val == 0 && grow)
2112                 val = 10000;
2113         else
2114                 val *= grow;
2115
2116         if (val > halt_poll_ns)
2117                 val = halt_poll_ns;
2118
2119         vcpu->halt_poll_ns = val;
2120         trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
2121 }
2122
2123 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
2124 {
2125         unsigned int old, val, shrink;
2126
2127         old = val = vcpu->halt_poll_ns;
2128         shrink = READ_ONCE(halt_poll_ns_shrink);
2129         if (shrink == 0)
2130                 val = 0;
2131         else
2132                 val /= shrink;
2133
2134         vcpu->halt_poll_ns = val;
2135         trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
2136 }
2137
2138 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
2139 {
2140         int ret = -EINTR;
2141         int idx = srcu_read_lock(&vcpu->kvm->srcu);
2142
2143         if (kvm_arch_vcpu_runnable(vcpu)) {
2144                 kvm_make_request(KVM_REQ_UNHALT, vcpu);
2145                 goto out;
2146         }
2147         if (kvm_cpu_has_pending_timer(vcpu))
2148                 goto out;
2149         if (signal_pending(current))
2150                 goto out;
2151
2152         ret = 0;
2153 out:
2154         srcu_read_unlock(&vcpu->kvm->srcu, idx);
2155         return ret;
2156 }
2157
2158 /*
2159  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2160  */
2161 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
2162 {
2163         ktime_t start, cur;
2164         DECLARE_SWAITQUEUE(wait);
2165         bool waited = false;
2166         u64 block_ns;
2167
2168         start = cur = ktime_get();
2169         if (vcpu->halt_poll_ns) {
2170                 ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns);
2171
2172                 ++vcpu->stat.halt_attempted_poll;
2173                 do {
2174                         /*
2175                          * This sets KVM_REQ_UNHALT if an interrupt
2176                          * arrives.
2177                          */
2178                         if (kvm_vcpu_check_block(vcpu) < 0) {
2179                                 ++vcpu->stat.halt_successful_poll;
2180                                 if (!vcpu_valid_wakeup(vcpu))
2181                                         ++vcpu->stat.halt_poll_invalid;
2182                                 goto out;
2183                         }
2184                         cur = ktime_get();
2185                 } while (single_task_running() && ktime_before(cur, stop));
2186         }
2187
2188         kvm_arch_vcpu_blocking(vcpu);
2189
2190         for (;;) {
2191                 prepare_to_swait_exclusive(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
2192
2193                 if (kvm_vcpu_check_block(vcpu) < 0)
2194                         break;
2195
2196                 waited = true;
2197                 schedule();
2198         }
2199
2200         finish_swait(&vcpu->wq, &wait);
2201         cur = ktime_get();
2202
2203         kvm_arch_vcpu_unblocking(vcpu);
2204 out:
2205         block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
2206
2207         if (!vcpu_valid_wakeup(vcpu))
2208                 shrink_halt_poll_ns(vcpu);
2209         else if (halt_poll_ns) {
2210                 if (block_ns <= vcpu->halt_poll_ns)
2211                         ;
2212                 /* we had a long block, shrink polling */
2213                 else if (vcpu->halt_poll_ns && block_ns > halt_poll_ns)
2214                         shrink_halt_poll_ns(vcpu);
2215                 /* we had a short halt and our poll time is too small */
2216                 else if (vcpu->halt_poll_ns < halt_poll_ns &&
2217                         block_ns < halt_poll_ns)
2218                         grow_halt_poll_ns(vcpu);
2219         } else
2220                 vcpu->halt_poll_ns = 0;
2221
2222         trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu));
2223         kvm_arch_vcpu_block_finish(vcpu);
2224 }
2225 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
2226
2227 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
2228 {
2229         struct swait_queue_head *wqp;
2230
2231         wqp = kvm_arch_vcpu_wq(vcpu);
2232         if (swq_has_sleeper(wqp)) {
2233                 swake_up_one(wqp);
2234                 ++vcpu->stat.halt_wakeup;
2235                 return true;
2236         }
2237
2238         return false;
2239 }
2240 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
2241
2242 #ifndef CONFIG_S390
2243 /*
2244  * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2245  */
2246 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
2247 {
2248         int me;
2249         int cpu = vcpu->cpu;
2250
2251         if (kvm_vcpu_wake_up(vcpu))
2252                 return;
2253
2254         me = get_cpu();
2255         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
2256                 if (kvm_arch_vcpu_should_kick(vcpu))
2257                         smp_send_reschedule(cpu);
2258         put_cpu();
2259 }
2260 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
2261 #endif /* !CONFIG_S390 */
2262
2263 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
2264 {
2265         struct pid *pid;
2266         struct task_struct *task = NULL;
2267         int ret = 0;
2268
2269         rcu_read_lock();
2270         pid = rcu_dereference(target->pid);
2271         if (pid)
2272                 task = get_pid_task(pid, PIDTYPE_PID);
2273         rcu_read_unlock();
2274         if (!task)
2275                 return ret;
2276         ret = yield_to(task, 1);
2277         put_task_struct(task);
2278
2279         return ret;
2280 }
2281 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
2282
2283 /*
2284  * Helper that checks whether a VCPU is eligible for directed yield.
2285  * Most eligible candidate to yield is decided by following heuristics:
2286  *
2287  *  (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2288  *  (preempted lock holder), indicated by @in_spin_loop.
2289  *  Set at the beiginning and cleared at the end of interception/PLE handler.
2290  *
2291  *  (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2292  *  chance last time (mostly it has become eligible now since we have probably
2293  *  yielded to lockholder in last iteration. This is done by toggling
2294  *  @dy_eligible each time a VCPU checked for eligibility.)
2295  *
2296  *  Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2297  *  to preempted lock-holder could result in wrong VCPU selection and CPU
2298  *  burning. Giving priority for a potential lock-holder increases lock
2299  *  progress.
2300  *
2301  *  Since algorithm is based on heuristics, accessing another VCPU data without
2302  *  locking does not harm. It may result in trying to yield to  same VCPU, fail
2303  *  and continue with next VCPU and so on.
2304  */
2305 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
2306 {
2307 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2308         bool eligible;
2309
2310         eligible = !vcpu->spin_loop.in_spin_loop ||
2311                     vcpu->spin_loop.dy_eligible;
2312
2313         if (vcpu->spin_loop.in_spin_loop)
2314                 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
2315
2316         return eligible;
2317 #else
2318         return true;
2319 #endif
2320 }
2321
2322 void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode)
2323 {
2324         struct kvm *kvm = me->kvm;
2325         struct kvm_vcpu *vcpu;
2326         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
2327         int yielded = 0;
2328         int try = 3;
2329         int pass;
2330         int i;
2331
2332         kvm_vcpu_set_in_spin_loop(me, true);
2333         /*
2334          * We boost the priority of a VCPU that is runnable but not
2335          * currently running, because it got preempted by something
2336          * else and called schedule in __vcpu_run.  Hopefully that
2337          * VCPU is holding the lock that we need and will release it.
2338          * We approximate round-robin by starting at the last boosted VCPU.
2339          */
2340         for (pass = 0; pass < 2 && !yielded && try; pass++) {
2341                 kvm_for_each_vcpu(i, vcpu, kvm) {
2342                         if (!pass && i <= last_boosted_vcpu) {
2343                                 i = last_boosted_vcpu;
2344                                 continue;
2345                         } else if (pass && i > last_boosted_vcpu)
2346                                 break;
2347                         if (!READ_ONCE(vcpu->preempted))
2348                                 continue;
2349                         if (vcpu == me)
2350                                 continue;
2351                         if (swait_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
2352                                 continue;
2353                         if (yield_to_kernel_mode && !kvm_arch_vcpu_in_kernel(vcpu))
2354                                 continue;
2355                         if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
2356                                 continue;
2357
2358                         yielded = kvm_vcpu_yield_to(vcpu);
2359                         if (yielded > 0) {
2360                                 kvm->last_boosted_vcpu = i;
2361                                 break;
2362                         } else if (yielded < 0) {
2363                                 try--;
2364                                 if (!try)
2365                                         break;
2366                         }
2367                 }
2368         }
2369         kvm_vcpu_set_in_spin_loop(me, false);
2370
2371         /* Ensure vcpu is not eligible during next spinloop */
2372         kvm_vcpu_set_dy_eligible(me, false);
2373 }
2374 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
2375
2376 static vm_fault_t kvm_vcpu_fault(struct vm_fault *vmf)
2377 {
2378         struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data;
2379         struct page *page;
2380
2381         if (vmf->pgoff == 0)
2382                 page = virt_to_page(vcpu->run);
2383 #ifdef CONFIG_X86
2384         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
2385                 page = virt_to_page(vcpu->arch.pio_data);
2386 #endif
2387 #ifdef CONFIG_KVM_MMIO
2388         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
2389                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
2390 #endif
2391         else
2392                 return kvm_arch_vcpu_fault(vcpu, vmf);
2393         get_page(page);
2394         vmf->page = page;
2395         return 0;
2396 }
2397
2398 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
2399         .fault = kvm_vcpu_fault,
2400 };
2401
2402 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2403 {
2404         vma->vm_ops = &kvm_vcpu_vm_ops;
2405         return 0;
2406 }
2407
2408 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2409 {
2410         struct kvm_vcpu *vcpu = filp->private_data;
2411
2412         debugfs_remove_recursive(vcpu->debugfs_dentry);
2413         kvm_put_kvm(vcpu->kvm);
2414         return 0;
2415 }
2416
2417 static struct file_operations kvm_vcpu_fops = {
2418         .release        = kvm_vcpu_release,
2419         .unlocked_ioctl = kvm_vcpu_ioctl,
2420         .mmap           = kvm_vcpu_mmap,
2421         .llseek         = noop_llseek,
2422         KVM_COMPAT(kvm_vcpu_compat_ioctl),
2423 };
2424
2425 /*
2426  * Allocates an inode for the vcpu.
2427  */
2428 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2429 {
2430         char name[8 + 1 + ITOA_MAX_LEN + 1];
2431
2432         snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id);
2433         return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
2434 }
2435
2436 static int kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
2437 {
2438         char dir_name[ITOA_MAX_LEN * 2];
2439         int ret;
2440
2441         if (!kvm_arch_has_vcpu_debugfs())
2442                 return 0;
2443
2444         if (!debugfs_initialized())
2445                 return 0;
2446
2447         snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
2448         vcpu->debugfs_dentry = debugfs_create_dir(dir_name,
2449                                                                 vcpu->kvm->debugfs_dentry);
2450         if (!vcpu->debugfs_dentry)
2451                 return -ENOMEM;
2452
2453         ret = kvm_arch_create_vcpu_debugfs(vcpu);
2454         if (ret < 0) {
2455                 debugfs_remove_recursive(vcpu->debugfs_dentry);
2456                 return ret;
2457         }
2458
2459         return 0;
2460 }
2461
2462 /*
2463  * Creates some virtual cpus.  Good luck creating more than one.
2464  */
2465 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
2466 {
2467         int r;
2468         struct kvm_vcpu *vcpu;
2469
2470         if (id >= KVM_MAX_VCPU_ID)
2471                 return -EINVAL;
2472
2473         mutex_lock(&kvm->lock);
2474         if (kvm->created_vcpus == KVM_MAX_VCPUS) {
2475                 mutex_unlock(&kvm->lock);
2476                 return -EINVAL;
2477         }
2478
2479         kvm->created_vcpus++;
2480         mutex_unlock(&kvm->lock);
2481
2482         vcpu = kvm_arch_vcpu_create(kvm, id);
2483         if (IS_ERR(vcpu)) {
2484                 r = PTR_ERR(vcpu);
2485                 goto vcpu_decrement;
2486         }
2487
2488         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2489
2490         r = kvm_arch_vcpu_setup(vcpu);
2491         if (r)
2492                 goto vcpu_destroy;
2493
2494         r = kvm_create_vcpu_debugfs(vcpu);
2495         if (r)
2496                 goto vcpu_destroy;
2497
2498         mutex_lock(&kvm->lock);
2499         if (kvm_get_vcpu_by_id(kvm, id)) {
2500                 r = -EEXIST;
2501                 goto unlock_vcpu_destroy;
2502         }
2503
2504         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
2505
2506         /* Now it's all set up, let userspace reach it */
2507         kvm_get_kvm(kvm);
2508         r = create_vcpu_fd(vcpu);
2509         if (r < 0) {
2510                 kvm_put_kvm(kvm);
2511                 goto unlock_vcpu_destroy;
2512         }
2513
2514         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
2515
2516         /*
2517          * Pairs with smp_rmb() in kvm_get_vcpu.  Write kvm->vcpus
2518          * before kvm->online_vcpu's incremented value.
2519          */
2520         smp_wmb();
2521         atomic_inc(&kvm->online_vcpus);
2522
2523         mutex_unlock(&kvm->lock);
2524         kvm_arch_vcpu_postcreate(vcpu);
2525         return r;
2526
2527 unlock_vcpu_destroy:
2528         mutex_unlock(&kvm->lock);
2529         debugfs_remove_recursive(vcpu->debugfs_dentry);
2530 vcpu_destroy:
2531         kvm_arch_vcpu_destroy(vcpu);
2532 vcpu_decrement:
2533         mutex_lock(&kvm->lock);
2534         kvm->created_vcpus--;
2535         mutex_unlock(&kvm->lock);
2536         return r;
2537 }
2538
2539 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2540 {
2541         if (sigset) {
2542                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2543                 vcpu->sigset_active = 1;
2544                 vcpu->sigset = *sigset;
2545         } else
2546                 vcpu->sigset_active = 0;
2547         return 0;
2548 }
2549
2550 static long kvm_vcpu_ioctl(struct file *filp,
2551                            unsigned int ioctl, unsigned long arg)
2552 {
2553         struct kvm_vcpu *vcpu = filp->private_data;
2554         void __user *argp = (void __user *)arg;
2555         int r;
2556         struct kvm_fpu *fpu = NULL;
2557         struct kvm_sregs *kvm_sregs = NULL;
2558
2559         if (vcpu->kvm->mm != current->mm)
2560                 return -EIO;
2561
2562         if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2563                 return -EINVAL;
2564
2565         /*
2566          * Some architectures have vcpu ioctls that are asynchronous to vcpu
2567          * execution; mutex_lock() would break them.
2568          */
2569         r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg);
2570         if (r != -ENOIOCTLCMD)
2571                 return r;
2572
2573         if (mutex_lock_killable(&vcpu->mutex))
2574                 return -EINTR;
2575         switch (ioctl) {
2576         case KVM_RUN: {
2577                 struct pid *oldpid;
2578                 r = -EINVAL;
2579                 if (arg)
2580                         goto out;
2581                 oldpid = rcu_access_pointer(vcpu->pid);
2582                 if (unlikely(oldpid != task_pid(current))) {
2583                         /* The thread running this VCPU changed. */
2584                         struct pid *newpid;
2585
2586                         r = kvm_arch_vcpu_run_pid_change(vcpu);
2587                         if (r)
2588                                 break;
2589
2590                         newpid = get_task_pid(current, PIDTYPE_PID);
2591                         rcu_assign_pointer(vcpu->pid, newpid);
2592                         if (oldpid)
2593                                 synchronize_rcu();
2594                         put_pid(oldpid);
2595                 }
2596                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2597                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2598                 break;
2599         }
2600         case KVM_GET_REGS: {
2601                 struct kvm_regs *kvm_regs;
2602
2603                 r = -ENOMEM;
2604                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2605                 if (!kvm_regs)
2606                         goto out;
2607                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2608                 if (r)
2609                         goto out_free1;
2610                 r = -EFAULT;
2611                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2612                         goto out_free1;
2613                 r = 0;
2614 out_free1:
2615                 kfree(kvm_regs);
2616                 break;
2617         }
2618         case KVM_SET_REGS: {
2619                 struct kvm_regs *kvm_regs;
2620
2621                 r = -ENOMEM;
2622                 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2623                 if (IS_ERR(kvm_regs)) {
2624                         r = PTR_ERR(kvm_regs);
2625                         goto out;
2626                 }
2627                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2628                 kfree(kvm_regs);
2629                 break;
2630         }
2631         case KVM_GET_SREGS: {
2632                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2633                 r = -ENOMEM;
2634                 if (!kvm_sregs)
2635                         goto out;
2636                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2637                 if (r)
2638                         goto out;
2639                 r = -EFAULT;
2640                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2641                         goto out;
2642                 r = 0;
2643                 break;
2644         }
2645         case KVM_SET_SREGS: {
2646                 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2647                 if (IS_ERR(kvm_sregs)) {
2648                         r = PTR_ERR(kvm_sregs);
2649                         kvm_sregs = NULL;
2650                         goto out;
2651                 }
2652                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2653                 break;
2654         }
2655         case KVM_GET_MP_STATE: {
2656                 struct kvm_mp_state mp_state;
2657
2658                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2659                 if (r)
2660                         goto out;
2661                 r = -EFAULT;
2662                 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
2663                         goto out;
2664                 r = 0;
2665                 break;
2666         }
2667         case KVM_SET_MP_STATE: {
2668                 struct kvm_mp_state mp_state;
2669
2670                 r = -EFAULT;
2671                 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
2672                         goto out;
2673                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2674                 break;
2675         }
2676         case KVM_TRANSLATE: {
2677                 struct kvm_translation tr;
2678
2679                 r = -EFAULT;
2680                 if (copy_from_user(&tr, argp, sizeof(tr)))
2681                         goto out;
2682                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2683                 if (r)
2684                         goto out;
2685                 r = -EFAULT;
2686                 if (copy_to_user(argp, &tr, sizeof(tr)))
2687                         goto out;
2688                 r = 0;
2689                 break;
2690         }
2691         case KVM_SET_GUEST_DEBUG: {
2692                 struct kvm_guest_debug dbg;
2693
2694                 r = -EFAULT;
2695                 if (copy_from_user(&dbg, argp, sizeof(dbg)))
2696                         goto out;
2697                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2698                 break;
2699         }
2700         case KVM_SET_SIGNAL_MASK: {
2701                 struct kvm_signal_mask __user *sigmask_arg = argp;
2702                 struct kvm_signal_mask kvm_sigmask;
2703                 sigset_t sigset, *p;
2704
2705                 p = NULL;
2706                 if (argp) {
2707                         r = -EFAULT;
2708                         if (copy_from_user(&kvm_sigmask, argp,
2709                                            sizeof(kvm_sigmask)))
2710                                 goto out;
2711                         r = -EINVAL;
2712                         if (kvm_sigmask.len != sizeof(sigset))
2713                                 goto out;
2714                         r = -EFAULT;
2715                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2716                                            sizeof(sigset)))
2717                                 goto out;
2718                         p = &sigset;
2719                 }
2720                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2721                 break;
2722         }
2723         case KVM_GET_FPU: {
2724                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2725                 r = -ENOMEM;
2726                 if (!fpu)
2727                         goto out;
2728                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2729                 if (r)
2730                         goto out;
2731                 r = -EFAULT;
2732                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2733                         goto out;
2734                 r = 0;
2735                 break;
2736         }
2737         case KVM_SET_FPU: {
2738                 fpu = memdup_user(argp, sizeof(*fpu));
2739                 if (IS_ERR(fpu)) {
2740                         r = PTR_ERR(fpu);
2741                         fpu = NULL;
2742                         goto out;
2743                 }
2744                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2745                 break;
2746         }
2747         default:
2748                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2749         }
2750 out:
2751         mutex_unlock(&vcpu->mutex);
2752         kfree(fpu);
2753         kfree(kvm_sregs);
2754         return r;
2755 }
2756
2757 #ifdef CONFIG_KVM_COMPAT
2758 static long kvm_vcpu_compat_ioctl(struct file *filp,
2759                                   unsigned int ioctl, unsigned long arg)
2760 {
2761         struct kvm_vcpu *vcpu = filp->private_data;
2762         void __user *argp = compat_ptr(arg);
2763         int r;
2764
2765         if (vcpu->kvm->mm != current->mm)
2766                 return -EIO;
2767
2768         switch (ioctl) {
2769         case KVM_SET_SIGNAL_MASK: {
2770                 struct kvm_signal_mask __user *sigmask_arg = argp;
2771                 struct kvm_signal_mask kvm_sigmask;
2772                 sigset_t sigset;
2773
2774                 if (argp) {
2775                         r = -EFAULT;
2776                         if (copy_from_user(&kvm_sigmask, argp,
2777                                            sizeof(kvm_sigmask)))
2778                                 goto out;
2779                         r = -EINVAL;
2780                         if (kvm_sigmask.len != sizeof(compat_sigset_t))
2781                                 goto out;
2782                         r = -EFAULT;
2783                         if (get_compat_sigset(&sigset, (void *)sigmask_arg->sigset))
2784                                 goto out;
2785                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2786                 } else
2787                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2788                 break;
2789         }
2790         default:
2791                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2792         }
2793
2794 out:
2795         return r;
2796 }
2797 #endif
2798
2799 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2800                                  int (*accessor)(struct kvm_device *dev,
2801                                                  struct kvm_device_attr *attr),
2802                                  unsigned long arg)
2803 {
2804         struct kvm_device_attr attr;
2805
2806         if (!accessor)
2807                 return -EPERM;
2808
2809         if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2810                 return -EFAULT;
2811
2812         return accessor(dev, &attr);
2813 }
2814
2815 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2816                              unsigned long arg)
2817 {
2818         struct kvm_device *dev = filp->private_data;
2819
2820         switch (ioctl) {
2821         case KVM_SET_DEVICE_ATTR:
2822                 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2823         case KVM_GET_DEVICE_ATTR:
2824                 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2825         case KVM_HAS_DEVICE_ATTR:
2826                 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2827         default:
2828                 if (dev->ops->ioctl)
2829                         return dev->ops->ioctl(dev, ioctl, arg);
2830
2831                 return -ENOTTY;
2832         }
2833 }
2834
2835 static int kvm_device_release(struct inode *inode, struct file *filp)
2836 {
2837         struct kvm_device *dev = filp->private_data;
2838         struct kvm *kvm = dev->kvm;
2839
2840         kvm_put_kvm(kvm);
2841         return 0;
2842 }
2843
2844 static const struct file_operations kvm_device_fops = {
2845         .unlocked_ioctl = kvm_device_ioctl,
2846         .release = kvm_device_release,
2847         KVM_COMPAT(kvm_device_ioctl),
2848 };
2849
2850 struct kvm_device *kvm_device_from_filp(struct file *filp)
2851 {
2852         if (filp->f_op != &kvm_device_fops)
2853                 return NULL;
2854
2855         return filp->private_data;
2856 }
2857
2858 static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
2859 #ifdef CONFIG_KVM_MPIC
2860         [KVM_DEV_TYPE_FSL_MPIC_20]      = &kvm_mpic_ops,
2861         [KVM_DEV_TYPE_FSL_MPIC_42]      = &kvm_mpic_ops,
2862 #endif
2863 };
2864
2865 int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
2866 {
2867         if (type >= ARRAY_SIZE(kvm_device_ops_table))
2868                 return -ENOSPC;
2869
2870         if (kvm_device_ops_table[type] != NULL)
2871                 return -EEXIST;
2872
2873         kvm_device_ops_table[type] = ops;
2874         return 0;
2875 }
2876
2877 void kvm_unregister_device_ops(u32 type)
2878 {
2879         if (kvm_device_ops_table[type] != NULL)
2880                 kvm_device_ops_table[type] = NULL;
2881 }
2882
2883 static int kvm_ioctl_create_device(struct kvm *kvm,
2884                                    struct kvm_create_device *cd)
2885 {
2886         struct kvm_device_ops *ops = NULL;
2887         struct kvm_device *dev;
2888         bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2889         int ret;
2890
2891         if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
2892                 return -ENODEV;
2893
2894         ops = kvm_device_ops_table[cd->type];
2895         if (ops == NULL)
2896                 return -ENODEV;
2897
2898         if (test)
2899                 return 0;
2900
2901         dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2902         if (!dev)
2903                 return -ENOMEM;
2904
2905         dev->ops = ops;
2906         dev->kvm = kvm;
2907
2908         mutex_lock(&kvm->lock);
2909         ret = ops->create(dev, cd->type);
2910         if (ret < 0) {
2911                 mutex_unlock(&kvm->lock);
2912                 kfree(dev);
2913                 return ret;
2914         }
2915         list_add(&dev->vm_node, &kvm->devices);
2916         mutex_unlock(&kvm->lock);
2917
2918         if (ops->init)
2919                 ops->init(dev);
2920
2921         ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2922         if (ret < 0) {
2923                 mutex_lock(&kvm->lock);
2924                 list_del(&dev->vm_node);
2925                 mutex_unlock(&kvm->lock);
2926                 ops->destroy(dev);
2927                 return ret;
2928         }
2929
2930         kvm_get_kvm(kvm);
2931         cd->fd = ret;
2932         return 0;
2933 }
2934
2935 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2936 {
2937         switch (arg) {
2938         case KVM_CAP_USER_MEMORY:
2939         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2940         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2941         case KVM_CAP_INTERNAL_ERROR_DATA:
2942 #ifdef CONFIG_HAVE_KVM_MSI
2943         case KVM_CAP_SIGNAL_MSI:
2944 #endif
2945 #ifdef CONFIG_HAVE_KVM_IRQFD
2946         case KVM_CAP_IRQFD:
2947         case KVM_CAP_IRQFD_RESAMPLE:
2948 #endif
2949         case KVM_CAP_IOEVENTFD_ANY_LENGTH:
2950         case KVM_CAP_CHECK_EXTENSION_VM:
2951                 return 1;
2952 #ifdef CONFIG_KVM_MMIO
2953         case KVM_CAP_COALESCED_MMIO:
2954                 return KVM_COALESCED_MMIO_PAGE_OFFSET;
2955         case KVM_CAP_COALESCED_PIO:
2956                 return 1;
2957 #endif
2958 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2959         case KVM_CAP_IRQ_ROUTING:
2960                 return KVM_MAX_IRQ_ROUTES;
2961 #endif
2962 #if KVM_ADDRESS_SPACE_NUM > 1
2963         case KVM_CAP_MULTI_ADDRESS_SPACE:
2964                 return KVM_ADDRESS_SPACE_NUM;
2965 #endif
2966         case KVM_CAP_MAX_VCPU_ID:
2967                 return KVM_MAX_VCPU_ID;
2968         default:
2969                 break;
2970         }
2971         return kvm_vm_ioctl_check_extension(kvm, arg);
2972 }
2973
2974 static long kvm_vm_ioctl(struct file *filp,
2975                            unsigned int ioctl, unsigned long arg)
2976 {
2977         struct kvm *kvm = filp->private_data;
2978         void __user *argp = (void __user *)arg;
2979         int r;
2980
2981         if (kvm->mm != current->mm)
2982                 return -EIO;
2983         switch (ioctl) {
2984         case KVM_CREATE_VCPU:
2985                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2986                 break;
2987         case KVM_SET_USER_MEMORY_REGION: {
2988                 struct kvm_userspace_memory_region kvm_userspace_mem;
2989
2990                 r = -EFAULT;
2991                 if (copy_from_user(&kvm_userspace_mem, argp,
2992                                                 sizeof(kvm_userspace_mem)))
2993                         goto out;
2994
2995                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2996                 break;
2997         }
2998         case KVM_GET_DIRTY_LOG: {
2999                 struct kvm_dirty_log log;
3000
3001                 r = -EFAULT;
3002                 if (copy_from_user(&log, argp, sizeof(log)))
3003                         goto out;
3004                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3005                 break;
3006         }
3007 #ifdef CONFIG_KVM_MMIO
3008         case KVM_REGISTER_COALESCED_MMIO: {
3009                 struct kvm_coalesced_mmio_zone zone;
3010
3011                 r = -EFAULT;
3012                 if (copy_from_user(&zone, argp, sizeof(zone)))
3013                         goto out;
3014                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
3015                 break;
3016         }
3017         case KVM_UNREGISTER_COALESCED_MMIO: {
3018                 struct kvm_coalesced_mmio_zone zone;
3019
3020                 r = -EFAULT;
3021                 if (copy_from_user(&zone, argp, sizeof(zone)))
3022                         goto out;
3023                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
3024                 break;
3025         }
3026 #endif
3027         case KVM_IRQFD: {
3028                 struct kvm_irqfd data;
3029
3030                 r = -EFAULT;
3031                 if (copy_from_user(&data, argp, sizeof(data)))
3032                         goto out;
3033                 r = kvm_irqfd(kvm, &data);
3034                 break;
3035         }
3036         case KVM_IOEVENTFD: {
3037                 struct kvm_ioeventfd data;
3038
3039                 r = -EFAULT;
3040                 if (copy_from_user(&data, argp, sizeof(data)))
3041                         goto out;
3042                 r = kvm_ioeventfd(kvm, &data);
3043                 break;
3044         }
3045 #ifdef CONFIG_HAVE_KVM_MSI
3046         case KVM_SIGNAL_MSI: {
3047                 struct kvm_msi msi;
3048
3049                 r = -EFAULT;
3050                 if (copy_from_user(&msi, argp, sizeof(msi)))
3051                         goto out;
3052                 r = kvm_send_userspace_msi(kvm, &msi);
3053                 break;
3054         }
3055 #endif
3056 #ifdef __KVM_HAVE_IRQ_LINE
3057         case KVM_IRQ_LINE_STATUS:
3058         case KVM_IRQ_LINE: {
3059                 struct kvm_irq_level irq_event;
3060
3061                 r = -EFAULT;
3062                 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
3063                         goto out;
3064
3065                 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
3066                                         ioctl == KVM_IRQ_LINE_STATUS);
3067                 if (r)
3068                         goto out;
3069
3070                 r = -EFAULT;
3071                 if (ioctl == KVM_IRQ_LINE_STATUS) {
3072                         if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
3073                                 goto out;
3074                 }
3075
3076                 r = 0;
3077                 break;
3078         }
3079 #endif
3080 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3081         case KVM_SET_GSI_ROUTING: {
3082                 struct kvm_irq_routing routing;
3083                 struct kvm_irq_routing __user *urouting;
3084                 struct kvm_irq_routing_entry *entries = NULL;
3085
3086                 r = -EFAULT;
3087                 if (copy_from_user(&routing, argp, sizeof(routing)))
3088                         goto out;
3089                 r = -EINVAL;
3090                 if (!kvm_arch_can_set_irq_routing(kvm))
3091                         goto out;
3092                 if (routing.nr > KVM_MAX_IRQ_ROUTES)
3093                         goto out;
3094                 if (routing.flags)
3095                         goto out;
3096                 if (routing.nr) {
3097                         r = -ENOMEM;
3098                         entries = vmalloc(array_size(sizeof(*entries),
3099                                                      routing.nr));
3100                         if (!entries)
3101                                 goto out;
3102                         r = -EFAULT;
3103                         urouting = argp;
3104                         if (copy_from_user(entries, urouting->entries,
3105                                            routing.nr * sizeof(*entries)))
3106                                 goto out_free_irq_routing;
3107                 }
3108                 r = kvm_set_irq_routing(kvm, entries, routing.nr,
3109                                         routing.flags);
3110 out_free_irq_routing:
3111                 vfree(entries);
3112                 break;
3113         }
3114 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
3115         case KVM_CREATE_DEVICE: {
3116                 struct kvm_create_device cd;
3117
3118                 r = -EFAULT;
3119                 if (copy_from_user(&cd, argp, sizeof(cd)))
3120                         goto out;
3121
3122                 r = kvm_ioctl_create_device(kvm, &cd);
3123                 if (r)
3124                         goto out;
3125
3126                 r = -EFAULT;
3127                 if (copy_to_user(argp, &cd, sizeof(cd)))
3128                         goto out;
3129
3130                 r = 0;
3131                 break;
3132         }
3133         case KVM_CHECK_EXTENSION:
3134                 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
3135                 break;
3136         default:
3137                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
3138         }
3139 out:
3140         return r;
3141 }
3142
3143 #ifdef CONFIG_KVM_COMPAT
3144 struct compat_kvm_dirty_log {
3145         __u32 slot;
3146         __u32 padding1;
3147         union {
3148                 compat_uptr_t dirty_bitmap; /* one bit per page */
3149                 __u64 padding2;
3150         };
3151 };
3152
3153 static long kvm_vm_compat_ioctl(struct file *filp,
3154                            unsigned int ioctl, unsigned long arg)
3155 {
3156         struct kvm *kvm = filp->private_data;
3157         int r;
3158
3159         if (kvm->mm != current->mm)
3160                 return -EIO;
3161         switch (ioctl) {
3162         case KVM_GET_DIRTY_LOG: {
3163                 struct compat_kvm_dirty_log compat_log;
3164                 struct kvm_dirty_log log;
3165
3166                 if (copy_from_user(&compat_log, (void __user *)arg,
3167                                    sizeof(compat_log)))
3168                         return -EFAULT;
3169                 log.slot         = compat_log.slot;
3170                 log.padding1     = compat_log.padding1;
3171                 log.padding2     = compat_log.padding2;
3172                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
3173
3174                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3175                 break;
3176         }
3177         default:
3178                 r = kvm_vm_ioctl(filp, ioctl, arg);
3179         }
3180         return r;
3181 }
3182 #endif
3183
3184 static struct file_operations kvm_vm_fops = {
3185         .release        = kvm_vm_release,
3186         .unlocked_ioctl = kvm_vm_ioctl,
3187         .llseek         = noop_llseek,
3188         KVM_COMPAT(kvm_vm_compat_ioctl),
3189 };
3190
3191 static int kvm_dev_ioctl_create_vm(unsigned long type)
3192 {
3193         int r;
3194         struct kvm *kvm;
3195         struct file *file;
3196
3197         kvm = kvm_create_vm(type);
3198         if (IS_ERR(kvm))
3199                 return PTR_ERR(kvm);
3200 #ifdef CONFIG_KVM_MMIO
3201         r = kvm_coalesced_mmio_init(kvm);
3202         if (r < 0)
3203                 goto put_kvm;
3204 #endif
3205         r = get_unused_fd_flags(O_CLOEXEC);
3206         if (r < 0)
3207                 goto put_kvm;
3208
3209         file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
3210         if (IS_ERR(file)) {
3211                 put_unused_fd(r);
3212                 r = PTR_ERR(file);
3213                 goto put_kvm;
3214         }
3215
3216         /*
3217          * Don't call kvm_put_kvm anymore at this point; file->f_op is
3218          * already set, with ->release() being kvm_vm_release().  In error
3219          * cases it will be called by the final fput(file) and will take
3220          * care of doing kvm_put_kvm(kvm).
3221          */
3222         if (kvm_create_vm_debugfs(kvm, r) < 0) {
3223                 put_unused_fd(r);
3224                 fput(file);
3225                 return -ENOMEM;
3226         }
3227         kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm);
3228
3229         fd_install(r, file);
3230         return r;
3231
3232 put_kvm:
3233         kvm_put_kvm(kvm);
3234         return r;
3235 }
3236
3237 static long kvm_dev_ioctl(struct file *filp,
3238                           unsigned int ioctl, unsigned long arg)
3239 {
3240         long r = -EINVAL;
3241
3242         switch (ioctl) {
3243         case KVM_GET_API_VERSION:
3244                 if (arg)
3245                         goto out;
3246                 r = KVM_API_VERSION;
3247                 break;
3248         case KVM_CREATE_VM:
3249                 r = kvm_dev_ioctl_create_vm(arg);
3250                 break;
3251         case KVM_CHECK_EXTENSION:
3252                 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
3253                 break;
3254         case KVM_GET_VCPU_MMAP_SIZE:
3255                 if (arg)
3256                         goto out;
3257                 r = PAGE_SIZE;     /* struct kvm_run */
3258 #ifdef CONFIG_X86
3259                 r += PAGE_SIZE;    /* pio data page */
3260 #endif
3261 #ifdef CONFIG_KVM_MMIO
3262                 r += PAGE_SIZE;    /* coalesced mmio ring page */
3263 #endif
3264                 break;
3265         case KVM_TRACE_ENABLE:
3266         case KVM_TRACE_PAUSE:
3267         case KVM_TRACE_DISABLE:
3268                 r = -EOPNOTSUPP;
3269                 break;
3270         default:
3271                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
3272         }
3273 out:
3274         return r;
3275 }
3276
3277 static struct file_operations kvm_chardev_ops = {
3278         .unlocked_ioctl = kvm_dev_ioctl,
3279         .llseek         = noop_llseek,
3280         KVM_COMPAT(kvm_dev_ioctl),
3281 };
3282
3283 static struct miscdevice kvm_dev = {
3284         KVM_MINOR,
3285         "kvm",
3286         &kvm_chardev_ops,
3287 };
3288
3289 static void hardware_enable_nolock(void *junk)
3290 {
3291         int cpu = raw_smp_processor_id();
3292         int r;
3293
3294         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
3295                 return;
3296
3297         cpumask_set_cpu(cpu, cpus_hardware_enabled);
3298
3299         r = kvm_arch_hardware_enable();
3300
3301         if (r) {
3302                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3303                 atomic_inc(&hardware_enable_failed);
3304                 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
3305         }
3306 }
3307
3308 static int kvm_starting_cpu(unsigned int cpu)
3309 {
3310         raw_spin_lock(&kvm_count_lock);
3311         if (kvm_usage_count)
3312                 hardware_enable_nolock(NULL);
3313         raw_spin_unlock(&kvm_count_lock);
3314         return 0;
3315 }
3316
3317 static void hardware_disable_nolock(void *junk)
3318 {
3319         int cpu = raw_smp_processor_id();
3320
3321         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
3322                 return;
3323         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3324         kvm_arch_hardware_disable();
3325 }
3326
3327 static int kvm_dying_cpu(unsigned int cpu)
3328 {
3329         raw_spin_lock(&kvm_count_lock);
3330         if (kvm_usage_count)
3331                 hardware_disable_nolock(NULL);
3332         raw_spin_unlock(&kvm_count_lock);
3333         return 0;
3334 }
3335
3336 static void hardware_disable_all_nolock(void)
3337 {
3338         BUG_ON(!kvm_usage_count);
3339
3340         kvm_usage_count--;
3341         if (!kvm_usage_count)
3342                 on_each_cpu(hardware_disable_nolock, NULL, 1);
3343 }
3344
3345 static void hardware_disable_all(void)
3346 {
3347         raw_spin_lock(&kvm_count_lock);
3348         hardware_disable_all_nolock();
3349         raw_spin_unlock(&kvm_count_lock);
3350 }
3351
3352 static int hardware_enable_all(void)
3353 {
3354         int r = 0;
3355
3356         raw_spin_lock(&kvm_count_lock);
3357
3358         kvm_usage_count++;
3359         if (kvm_usage_count == 1) {
3360                 atomic_set(&hardware_enable_failed, 0);
3361                 on_each_cpu(hardware_enable_nolock, NULL, 1);
3362
3363                 if (atomic_read(&hardware_enable_failed)) {
3364                         hardware_disable_all_nolock();
3365                         r = -EBUSY;
3366                 }
3367         }
3368
3369         raw_spin_unlock(&kvm_count_lock);
3370
3371         return r;
3372 }
3373
3374 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3375                       void *v)
3376 {
3377         /*
3378          * Some (well, at least mine) BIOSes hang on reboot if
3379          * in vmx root mode.
3380          *
3381          * And Intel TXT required VMX off for all cpu when system shutdown.
3382          */
3383         pr_info("kvm: exiting hardware virtualization\n");
3384         kvm_rebooting = true;
3385         on_each_cpu(hardware_disable_nolock, NULL, 1);
3386         return NOTIFY_OK;
3387 }
3388
3389 static struct notifier_block kvm_reboot_notifier = {
3390         .notifier_call = kvm_reboot,
3391         .priority = 0,
3392 };
3393
3394 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3395 {
3396         int i;
3397
3398         for (i = 0; i < bus->dev_count; i++) {
3399                 struct kvm_io_device *pos = bus->range[i].dev;
3400
3401                 kvm_iodevice_destructor(pos);
3402         }
3403         kfree(bus);
3404 }
3405
3406 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
3407                                  const struct kvm_io_range *r2)
3408 {
3409         gpa_t addr1 = r1->addr;
3410         gpa_t addr2 = r2->addr;
3411
3412         if (addr1 < addr2)
3413                 return -1;
3414
3415         /* If r2->len == 0, match the exact address.  If r2->len != 0,
3416          * accept any overlapping write.  Any order is acceptable for
3417          * overlapping ranges, because kvm_io_bus_get_first_dev ensures
3418          * we process all of them.
3419          */
3420         if (r2->len) {
3421                 addr1 += r1->len;
3422                 addr2 += r2->len;
3423         }
3424
3425         if (addr1 > addr2)
3426                 return 1;
3427
3428         return 0;
3429 }
3430
3431 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
3432 {
3433         return kvm_io_bus_cmp(p1, p2);
3434 }
3435
3436 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
3437                              gpa_t addr, int len)
3438 {
3439         struct kvm_io_range *range, key;
3440         int off;
3441
3442         key = (struct kvm_io_range) {
3443                 .addr = addr,
3444                 .len = len,
3445         };
3446
3447         range = bsearch(&key, bus->range, bus->dev_count,
3448                         sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
3449         if (range == NULL)
3450                 return -ENOENT;
3451
3452         off = range - bus->range;
3453
3454         while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
3455                 off--;
3456
3457         return off;
3458 }
3459
3460 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3461                               struct kvm_io_range *range, const void *val)
3462 {
3463         int idx;
3464
3465         idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3466         if (idx < 0)
3467                 return -EOPNOTSUPP;
3468
3469         while (idx < bus->dev_count &&
3470                 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3471                 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
3472                                         range->len, val))
3473                         return idx;
3474                 idx++;
3475         }
3476
3477         return -EOPNOTSUPP;
3478 }
3479
3480 /* kvm_io_bus_write - called under kvm->slots_lock */
3481 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3482                      int len, const void *val)
3483 {
3484         struct kvm_io_bus *bus;
3485         struct kvm_io_range range;
3486         int r;
3487
3488         range = (struct kvm_io_range) {
3489                 .addr = addr,
3490                 .len = len,
3491         };
3492
3493         bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3494         if (!bus)
3495                 return -ENOMEM;
3496         r = __kvm_io_bus_write(vcpu, bus, &range, val);
3497         return r < 0 ? r : 0;
3498 }
3499
3500 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3501 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
3502                             gpa_t addr, int len, const void *val, long cookie)
3503 {
3504         struct kvm_io_bus *bus;
3505         struct kvm_io_range range;
3506
3507         range = (struct kvm_io_range) {
3508                 .addr = addr,
3509                 .len = len,
3510         };
3511
3512         bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3513         if (!bus)
3514                 return -ENOMEM;
3515
3516         /* First try the device referenced by cookie. */
3517         if ((cookie >= 0) && (cookie < bus->dev_count) &&
3518             (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
3519                 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
3520                                         val))
3521                         return cookie;
3522
3523         /*
3524          * cookie contained garbage; fall back to search and return the
3525          * correct cookie value.
3526          */
3527         return __kvm_io_bus_write(vcpu, bus, &range, val);
3528 }
3529
3530 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3531                              struct kvm_io_range *range, void *val)
3532 {
3533         int idx;
3534
3535         idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3536         if (idx < 0)
3537                 return -EOPNOTSUPP;
3538
3539         while (idx < bus->dev_count &&
3540                 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3541                 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
3542                                        range->len, val))
3543                         return idx;
3544                 idx++;
3545         }
3546
3547         return -EOPNOTSUPP;
3548 }
3549 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
3550
3551 /* kvm_io_bus_read - called under kvm->slots_lock */
3552 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3553                     int len, void *val)
3554 {
3555         struct kvm_io_bus *bus;
3556         struct kvm_io_range range;
3557         int r;
3558
3559         range = (struct kvm_io_range) {
3560                 .addr = addr,
3561                 .len = len,
3562         };
3563
3564         bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3565         if (!bus)
3566                 return -ENOMEM;
3567         r = __kvm_io_bus_read(vcpu, bus, &range, val);
3568         return r < 0 ? r : 0;
3569 }
3570
3571
3572 /* Caller must hold slots_lock. */
3573 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3574                             int len, struct kvm_io_device *dev)
3575 {
3576         int i;
3577         struct kvm_io_bus *new_bus, *bus;
3578         struct kvm_io_range range;
3579
3580         bus = kvm_get_bus(kvm, bus_idx);
3581         if (!bus)
3582                 return -ENOMEM;
3583
3584         /* exclude ioeventfd which is limited by maximum fd */
3585         if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3586                 return -ENOSPC;
3587
3588         new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count + 1) *
3589                           sizeof(struct kvm_io_range)), GFP_KERNEL);
3590         if (!new_bus)
3591                 return -ENOMEM;
3592
3593         range = (struct kvm_io_range) {
3594                 .addr = addr,
3595                 .len = len,
3596                 .dev = dev,
3597         };
3598
3599         for (i = 0; i < bus->dev_count; i++)
3600                 if (kvm_io_bus_cmp(&bus->range[i], &range) > 0)
3601                         break;
3602
3603         memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3604         new_bus->dev_count++;