f6cd14d2f0d97903b9acf081044cc32ed682f6c2
[linux-3.10.git] / virt / kvm / kvm_main.c
1 /*
2  * Kernel-based Virtual Machine driver for Linux
3  *
4  * This module enables machines with Intel VT-x extensions to run virtual
5  * machines without emulation or binary translation.
6  *
7  * Copyright (C) 2006 Qumranet, Inc.
8  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
9  *
10  * Authors:
11  *   Avi Kivity   <avi@qumranet.com>
12  *   Yaniv Kamay  <yaniv@qumranet.com>
13  *
14  * This work is licensed under the terms of the GNU GPL, version 2.  See
15  * the COPYING file in the top-level directory.
16  *
17  */
18
19 #include "iodev.h"
20
21 #include <linux/kvm_host.h>
22 #include <linux/kvm.h>
23 #include <linux/module.h>
24 #include <linux/errno.h>
25 #include <linux/percpu.h>
26 #include <linux/mm.h>
27 #include <linux/miscdevice.h>
28 #include <linux/vmalloc.h>
29 #include <linux/reboot.h>
30 #include <linux/debugfs.h>
31 #include <linux/highmem.h>
32 #include <linux/file.h>
33 #include <linux/syscore_ops.h>
34 #include <linux/cpu.h>
35 #include <linux/sched.h>
36 #include <linux/cpumask.h>
37 #include <linux/smp.h>
38 #include <linux/anon_inodes.h>
39 #include <linux/profile.h>
40 #include <linux/kvm_para.h>
41 #include <linux/pagemap.h>
42 #include <linux/mman.h>
43 #include <linux/swap.h>
44 #include <linux/bitops.h>
45 #include <linux/spinlock.h>
46 #include <linux/compat.h>
47 #include <linux/srcu.h>
48 #include <linux/hugetlb.h>
49 #include <linux/slab.h>
50 #include <linux/sort.h>
51 #include <linux/bsearch.h>
52
53 #include <asm/processor.h>
54 #include <asm/io.h>
55 #include <asm/uaccess.h>
56 #include <asm/pgtable.h>
57
58 #include "coalesced_mmio.h"
59 #include "async_pf.h"
60
61 #define CREATE_TRACE_POINTS
62 #include <trace/events/kvm.h>
63
64 MODULE_AUTHOR("Qumranet");
65 MODULE_LICENSE("GPL");
66
67 /*
68  * Ordering of locks:
69  *
70  *              kvm->lock --> kvm->slots_lock --> kvm->irq_lock
71  */
72
73 DEFINE_RAW_SPINLOCK(kvm_lock);
74 LIST_HEAD(vm_list);
75
76 static cpumask_var_t cpus_hardware_enabled;
77 static int kvm_usage_count = 0;
78 static atomic_t hardware_enable_failed;
79
80 struct kmem_cache *kvm_vcpu_cache;
81 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
82
83 static __read_mostly struct preempt_ops kvm_preempt_ops;
84
85 struct dentry *kvm_debugfs_dir;
86
87 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
88                            unsigned long arg);
89 #ifdef CONFIG_COMPAT
90 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
91                                   unsigned long arg);
92 #endif
93 static int hardware_enable_all(void);
94 static void hardware_disable_all(void);
95
96 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
97
98 bool kvm_rebooting;
99 EXPORT_SYMBOL_GPL(kvm_rebooting);
100
101 static bool largepages_enabled = true;
102
103 bool kvm_is_mmio_pfn(pfn_t pfn)
104 {
105         if (pfn_valid(pfn)) {
106                 int reserved;
107                 struct page *tail = pfn_to_page(pfn);
108                 struct page *head = compound_trans_head(tail);
109                 reserved = PageReserved(head);
110                 if (head != tail) {
111                         /*
112                          * "head" is not a dangling pointer
113                          * (compound_trans_head takes care of that)
114                          * but the hugepage may have been splitted
115                          * from under us (and we may not hold a
116                          * reference count on the head page so it can
117                          * be reused before we run PageReferenced), so
118                          * we've to check PageTail before returning
119                          * what we just read.
120                          */
121                         smp_rmb();
122                         if (PageTail(tail))
123                                 return reserved;
124                 }
125                 return PageReserved(tail);
126         }
127
128         return true;
129 }
130
131 /*
132  * Switches to specified vcpu, until a matching vcpu_put()
133  */
134 int vcpu_load(struct kvm_vcpu *vcpu)
135 {
136         int cpu;
137
138         if (mutex_lock_killable(&vcpu->mutex))
139                 return -EINTR;
140         if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
141                 /* The thread running this VCPU changed. */
142                 struct pid *oldpid = vcpu->pid;
143                 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
144                 rcu_assign_pointer(vcpu->pid, newpid);
145                 synchronize_rcu();
146                 put_pid(oldpid);
147         }
148         cpu = get_cpu();
149         preempt_notifier_register(&vcpu->preempt_notifier);
150         kvm_arch_vcpu_load(vcpu, cpu);
151         put_cpu();
152         return 0;
153 }
154
155 void vcpu_put(struct kvm_vcpu *vcpu)
156 {
157         preempt_disable();
158         kvm_arch_vcpu_put(vcpu);
159         preempt_notifier_unregister(&vcpu->preempt_notifier);
160         preempt_enable();
161         mutex_unlock(&vcpu->mutex);
162 }
163
164 static void ack_flush(void *_completed)
165 {
166 }
167
168 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
169 {
170         int i, cpu, me;
171         cpumask_var_t cpus;
172         bool called = true;
173         struct kvm_vcpu *vcpu;
174
175         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
176
177         me = get_cpu();
178         kvm_for_each_vcpu(i, vcpu, kvm) {
179                 kvm_make_request(req, vcpu);
180                 cpu = vcpu->cpu;
181
182                 /* Set ->requests bit before we read ->mode */
183                 smp_mb();
184
185                 if (cpus != NULL && cpu != -1 && cpu != me &&
186                       kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
187                         cpumask_set_cpu(cpu, cpus);
188         }
189         if (unlikely(cpus == NULL))
190                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
191         else if (!cpumask_empty(cpus))
192                 smp_call_function_many(cpus, ack_flush, NULL, 1);
193         else
194                 called = false;
195         put_cpu();
196         free_cpumask_var(cpus);
197         return called;
198 }
199
200 void kvm_flush_remote_tlbs(struct kvm *kvm)
201 {
202         long dirty_count = kvm->tlbs_dirty;
203
204         smp_mb();
205         if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
206                 ++kvm->stat.remote_tlb_flush;
207         cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
208 }
209
210 void kvm_reload_remote_mmus(struct kvm *kvm)
211 {
212         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
213 }
214
215 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
216 {
217         make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
218 }
219
220 void kvm_make_scan_ioapic_request(struct kvm *kvm)
221 {
222         make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
223 }
224
225 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
226 {
227         struct page *page;
228         int r;
229
230         mutex_init(&vcpu->mutex);
231         vcpu->cpu = -1;
232         vcpu->kvm = kvm;
233         vcpu->vcpu_id = id;
234         vcpu->pid = NULL;
235         init_waitqueue_head(&vcpu->wq);
236         kvm_async_pf_vcpu_init(vcpu);
237
238         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
239         if (!page) {
240                 r = -ENOMEM;
241                 goto fail;
242         }
243         vcpu->run = page_address(page);
244
245         kvm_vcpu_set_in_spin_loop(vcpu, false);
246         kvm_vcpu_set_dy_eligible(vcpu, false);
247         vcpu->preempted = false;
248
249         r = kvm_arch_vcpu_init(vcpu);
250         if (r < 0)
251                 goto fail_free_run;
252         return 0;
253
254 fail_free_run:
255         free_page((unsigned long)vcpu->run);
256 fail:
257         return r;
258 }
259 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
260
261 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
262 {
263         put_pid(vcpu->pid);
264         kvm_arch_vcpu_uninit(vcpu);
265         free_page((unsigned long)vcpu->run);
266 }
267 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
268
269 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
270 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
271 {
272         return container_of(mn, struct kvm, mmu_notifier);
273 }
274
275 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
276                                              struct mm_struct *mm,
277                                              unsigned long address)
278 {
279         struct kvm *kvm = mmu_notifier_to_kvm(mn);
280         int need_tlb_flush, idx;
281
282         /*
283          * When ->invalidate_page runs, the linux pte has been zapped
284          * already but the page is still allocated until
285          * ->invalidate_page returns. So if we increase the sequence
286          * here the kvm page fault will notice if the spte can't be
287          * established because the page is going to be freed. If
288          * instead the kvm page fault establishes the spte before
289          * ->invalidate_page runs, kvm_unmap_hva will release it
290          * before returning.
291          *
292          * The sequence increase only need to be seen at spin_unlock
293          * time, and not at spin_lock time.
294          *
295          * Increasing the sequence after the spin_unlock would be
296          * unsafe because the kvm page fault could then establish the
297          * pte after kvm_unmap_hva returned, without noticing the page
298          * is going to be freed.
299          */
300         idx = srcu_read_lock(&kvm->srcu);
301         spin_lock(&kvm->mmu_lock);
302
303         kvm->mmu_notifier_seq++;
304         need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
305         /* we've to flush the tlb before the pages can be freed */
306         if (need_tlb_flush)
307                 kvm_flush_remote_tlbs(kvm);
308
309         spin_unlock(&kvm->mmu_lock);
310         srcu_read_unlock(&kvm->srcu, idx);
311 }
312
313 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
314                                         struct mm_struct *mm,
315                                         unsigned long address,
316                                         pte_t pte)
317 {
318         struct kvm *kvm = mmu_notifier_to_kvm(mn);
319         int idx;
320
321         idx = srcu_read_lock(&kvm->srcu);
322         spin_lock(&kvm->mmu_lock);
323         kvm->mmu_notifier_seq++;
324         kvm_set_spte_hva(kvm, address, pte);
325         spin_unlock(&kvm->mmu_lock);
326         srcu_read_unlock(&kvm->srcu, idx);
327 }
328
329 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
330                                                     struct mm_struct *mm,
331                                                     unsigned long start,
332                                                     unsigned long end)
333 {
334         struct kvm *kvm = mmu_notifier_to_kvm(mn);
335         int need_tlb_flush = 0, idx;
336
337         idx = srcu_read_lock(&kvm->srcu);
338         spin_lock(&kvm->mmu_lock);
339         /*
340          * The count increase must become visible at unlock time as no
341          * spte can be established without taking the mmu_lock and
342          * count is also read inside the mmu_lock critical section.
343          */
344         kvm->mmu_notifier_count++;
345         need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
346         need_tlb_flush |= kvm->tlbs_dirty;
347         /* we've to flush the tlb before the pages can be freed */
348         if (need_tlb_flush)
349                 kvm_flush_remote_tlbs(kvm);
350
351         spin_unlock(&kvm->mmu_lock);
352         srcu_read_unlock(&kvm->srcu, idx);
353 }
354
355 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
356                                                   struct mm_struct *mm,
357                                                   unsigned long start,
358                                                   unsigned long end)
359 {
360         struct kvm *kvm = mmu_notifier_to_kvm(mn);
361
362         spin_lock(&kvm->mmu_lock);
363         /*
364          * This sequence increase will notify the kvm page fault that
365          * the page that is going to be mapped in the spte could have
366          * been freed.
367          */
368         kvm->mmu_notifier_seq++;
369         smp_wmb();
370         /*
371          * The above sequence increase must be visible before the
372          * below count decrease, which is ensured by the smp_wmb above
373          * in conjunction with the smp_rmb in mmu_notifier_retry().
374          */
375         kvm->mmu_notifier_count--;
376         spin_unlock(&kvm->mmu_lock);
377
378         BUG_ON(kvm->mmu_notifier_count < 0);
379 }
380
381 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
382                                               struct mm_struct *mm,
383                                               unsigned long address)
384 {
385         struct kvm *kvm = mmu_notifier_to_kvm(mn);
386         int young, idx;
387
388         idx = srcu_read_lock(&kvm->srcu);
389         spin_lock(&kvm->mmu_lock);
390
391         young = kvm_age_hva(kvm, address);
392         if (young)
393                 kvm_flush_remote_tlbs(kvm);
394
395         spin_unlock(&kvm->mmu_lock);
396         srcu_read_unlock(&kvm->srcu, idx);
397
398         return young;
399 }
400
401 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
402                                        struct mm_struct *mm,
403                                        unsigned long address)
404 {
405         struct kvm *kvm = mmu_notifier_to_kvm(mn);
406         int young, idx;
407
408         idx = srcu_read_lock(&kvm->srcu);
409         spin_lock(&kvm->mmu_lock);
410         young = kvm_test_age_hva(kvm, address);
411         spin_unlock(&kvm->mmu_lock);
412         srcu_read_unlock(&kvm->srcu, idx);
413
414         return young;
415 }
416
417 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
418                                      struct mm_struct *mm)
419 {
420         struct kvm *kvm = mmu_notifier_to_kvm(mn);
421         int idx;
422
423         idx = srcu_read_lock(&kvm->srcu);
424         kvm_arch_flush_shadow_all(kvm);
425         srcu_read_unlock(&kvm->srcu, idx);
426 }
427
428 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
429         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
430         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
431         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
432         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
433         .test_young             = kvm_mmu_notifier_test_young,
434         .change_pte             = kvm_mmu_notifier_change_pte,
435         .release                = kvm_mmu_notifier_release,
436 };
437
438 static int kvm_init_mmu_notifier(struct kvm *kvm)
439 {
440         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
441         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
442 }
443
444 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
445
446 static int kvm_init_mmu_notifier(struct kvm *kvm)
447 {
448         return 0;
449 }
450
451 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
452
453 static void kvm_init_memslots_id(struct kvm *kvm)
454 {
455         int i;
456         struct kvm_memslots *slots = kvm->memslots;
457
458         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
459                 slots->id_to_index[i] = slots->memslots[i].id = i;
460 }
461
462 static struct kvm *kvm_create_vm(unsigned long type)
463 {
464         int r, i;
465         struct kvm *kvm = kvm_arch_alloc_vm();
466
467         if (!kvm)
468                 return ERR_PTR(-ENOMEM);
469
470         r = kvm_arch_init_vm(kvm, type);
471         if (r)
472                 goto out_err_nodisable;
473
474         r = hardware_enable_all();
475         if (r)
476                 goto out_err_nodisable;
477
478 #ifdef CONFIG_HAVE_KVM_IRQCHIP
479         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
480         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
481 #endif
482
483         BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
484
485         r = -ENOMEM;
486         kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
487         if (!kvm->memslots)
488                 goto out_err_nosrcu;
489         kvm_init_memslots_id(kvm);
490         if (init_srcu_struct(&kvm->srcu))
491                 goto out_err_nosrcu;
492         for (i = 0; i < KVM_NR_BUSES; i++) {
493                 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
494                                         GFP_KERNEL);
495                 if (!kvm->buses[i])
496                         goto out_err;
497         }
498
499         spin_lock_init(&kvm->mmu_lock);
500         kvm->mm = current->mm;
501         atomic_inc(&kvm->mm->mm_count);
502         kvm_eventfd_init(kvm);
503         mutex_init(&kvm->lock);
504         mutex_init(&kvm->irq_lock);
505         mutex_init(&kvm->slots_lock);
506         atomic_set(&kvm->users_count, 1);
507
508         r = kvm_init_mmu_notifier(kvm);
509         if (r)
510                 goto out_err;
511
512         raw_spin_lock(&kvm_lock);
513         list_add(&kvm->vm_list, &vm_list);
514         raw_spin_unlock(&kvm_lock);
515
516         return kvm;
517
518 out_err:
519         cleanup_srcu_struct(&kvm->srcu);
520 out_err_nosrcu:
521         hardware_disable_all();
522 out_err_nodisable:
523         for (i = 0; i < KVM_NR_BUSES; i++)
524                 kfree(kvm->buses[i]);
525         kfree(kvm->memslots);
526         kvm_arch_free_vm(kvm);
527         return ERR_PTR(r);
528 }
529
530 /*
531  * Avoid using vmalloc for a small buffer.
532  * Should not be used when the size is statically known.
533  */
534 void *kvm_kvzalloc(unsigned long size)
535 {
536         if (size > PAGE_SIZE)
537                 return vzalloc(size);
538         else
539                 return kzalloc(size, GFP_KERNEL);
540 }
541
542 void kvm_kvfree(const void *addr)
543 {
544         if (is_vmalloc_addr(addr))
545                 vfree(addr);
546         else
547                 kfree(addr);
548 }
549
550 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
551 {
552         if (!memslot->dirty_bitmap)
553                 return;
554
555         kvm_kvfree(memslot->dirty_bitmap);
556         memslot->dirty_bitmap = NULL;
557 }
558
559 /*
560  * Free any memory in @free but not in @dont.
561  */
562 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
563                                   struct kvm_memory_slot *dont)
564 {
565         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
566                 kvm_destroy_dirty_bitmap(free);
567
568         kvm_arch_free_memslot(free, dont);
569
570         free->npages = 0;
571 }
572
573 void kvm_free_physmem(struct kvm *kvm)
574 {
575         struct kvm_memslots *slots = kvm->memslots;
576         struct kvm_memory_slot *memslot;
577
578         kvm_for_each_memslot(memslot, slots)
579                 kvm_free_physmem_slot(memslot, NULL);
580
581         kfree(kvm->memslots);
582 }
583
584 static void kvm_destroy_vm(struct kvm *kvm)
585 {
586         int i;
587         struct mm_struct *mm = kvm->mm;
588
589         kvm_arch_sync_events(kvm);
590         raw_spin_lock(&kvm_lock);
591         list_del(&kvm->vm_list);
592         raw_spin_unlock(&kvm_lock);
593         kvm_free_irq_routing(kvm);
594         for (i = 0; i < KVM_NR_BUSES; i++)
595                 kvm_io_bus_destroy(kvm->buses[i]);
596         kvm_coalesced_mmio_free(kvm);
597 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
598         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
599 #else
600         kvm_arch_flush_shadow_all(kvm);
601 #endif
602         kvm_arch_destroy_vm(kvm);
603         kvm_free_physmem(kvm);
604         cleanup_srcu_struct(&kvm->srcu);
605         kvm_arch_free_vm(kvm);
606         hardware_disable_all();
607         mmdrop(mm);
608 }
609
610 void kvm_get_kvm(struct kvm *kvm)
611 {
612         atomic_inc(&kvm->users_count);
613 }
614 EXPORT_SYMBOL_GPL(kvm_get_kvm);
615
616 void kvm_put_kvm(struct kvm *kvm)
617 {
618         if (atomic_dec_and_test(&kvm->users_count))
619                 kvm_destroy_vm(kvm);
620 }
621 EXPORT_SYMBOL_GPL(kvm_put_kvm);
622
623
624 static int kvm_vm_release(struct inode *inode, struct file *filp)
625 {
626         struct kvm *kvm = filp->private_data;
627
628         kvm_irqfd_release(kvm);
629
630         kvm_put_kvm(kvm);
631         return 0;
632 }
633
634 /*
635  * Allocation size is twice as large as the actual dirty bitmap size.
636  * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
637  */
638 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
639 {
640 #ifndef CONFIG_S390
641         unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
642
643         memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
644         if (!memslot->dirty_bitmap)
645                 return -ENOMEM;
646
647 #endif /* !CONFIG_S390 */
648         return 0;
649 }
650
651 static int cmp_memslot(const void *slot1, const void *slot2)
652 {
653         struct kvm_memory_slot *s1, *s2;
654
655         s1 = (struct kvm_memory_slot *)slot1;
656         s2 = (struct kvm_memory_slot *)slot2;
657
658         if (s1->npages < s2->npages)
659                 return 1;
660         if (s1->npages > s2->npages)
661                 return -1;
662
663         return 0;
664 }
665
666 /*
667  * Sort the memslots base on its size, so the larger slots
668  * will get better fit.
669  */
670 static void sort_memslots(struct kvm_memslots *slots)
671 {
672         int i;
673
674         sort(slots->memslots, KVM_MEM_SLOTS_NUM,
675               sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
676
677         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
678                 slots->id_to_index[slots->memslots[i].id] = i;
679 }
680
681 void update_memslots(struct kvm_memslots *slots, struct kvm_memory_slot *new,
682                      u64 last_generation)
683 {
684         if (new) {
685                 int id = new->id;
686                 struct kvm_memory_slot *old = id_to_memslot(slots, id);
687                 unsigned long npages = old->npages;
688
689                 *old = *new;
690                 if (new->npages != npages)
691                         sort_memslots(slots);
692         }
693
694         slots->generation = last_generation + 1;
695 }
696
697 static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
698 {
699         u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
700
701 #ifdef KVM_CAP_READONLY_MEM
702         valid_flags |= KVM_MEM_READONLY;
703 #endif
704
705         if (mem->flags & ~valid_flags)
706                 return -EINVAL;
707
708         return 0;
709 }
710
711 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
712                 struct kvm_memslots *slots, struct kvm_memory_slot *new)
713 {
714         struct kvm_memslots *old_memslots = kvm->memslots;
715
716         update_memslots(slots, new, kvm->memslots->generation);
717         rcu_assign_pointer(kvm->memslots, slots);
718         synchronize_srcu_expedited(&kvm->srcu);
719         return old_memslots; 
720 }
721
722 /*
723  * Allocate some memory and give it an address in the guest physical address
724  * space.
725  *
726  * Discontiguous memory is allowed, mostly for framebuffers.
727  *
728  * Must be called holding mmap_sem for write.
729  */
730 int __kvm_set_memory_region(struct kvm *kvm,
731                             struct kvm_userspace_memory_region *mem)
732 {
733         int r;
734         gfn_t base_gfn;
735         unsigned long npages;
736         struct kvm_memory_slot *slot;
737         struct kvm_memory_slot old, new;
738         struct kvm_memslots *slots = NULL, *old_memslots;
739         enum kvm_mr_change change;
740
741         r = check_memory_region_flags(mem);
742         if (r)
743                 goto out;
744
745         r = -EINVAL;
746         /* General sanity checks */
747         if (mem->memory_size & (PAGE_SIZE - 1))
748                 goto out;
749         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
750                 goto out;
751         /* We can read the guest memory with __xxx_user() later on. */
752         if ((mem->slot < KVM_USER_MEM_SLOTS) &&
753             ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
754              !access_ok(VERIFY_WRITE,
755                         (void __user *)(unsigned long)mem->userspace_addr,
756                         mem->memory_size)))
757                 goto out;
758         if (mem->slot >= KVM_MEM_SLOTS_NUM)
759                 goto out;
760         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
761                 goto out;
762
763         slot = id_to_memslot(kvm->memslots, mem->slot);
764         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
765         npages = mem->memory_size >> PAGE_SHIFT;
766
767         r = -EINVAL;
768         if (npages > KVM_MEM_MAX_NR_PAGES)
769                 goto out;
770
771         if (!npages)
772                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
773
774         new = old = *slot;
775
776         new.id = mem->slot;
777         new.base_gfn = base_gfn;
778         new.npages = npages;
779         new.flags = mem->flags;
780
781         r = -EINVAL;
782         if (npages) {
783                 if (!old.npages)
784                         change = KVM_MR_CREATE;
785                 else { /* Modify an existing slot. */
786                         if ((mem->userspace_addr != old.userspace_addr) ||
787                             (npages != old.npages) ||
788                             ((new.flags ^ old.flags) & KVM_MEM_READONLY))
789                                 goto out;
790
791                         if (base_gfn != old.base_gfn)
792                                 change = KVM_MR_MOVE;
793                         else if (new.flags != old.flags)
794                                 change = KVM_MR_FLAGS_ONLY;
795                         else { /* Nothing to change. */
796                                 r = 0;
797                                 goto out;
798                         }
799                 }
800         } else if (old.npages) {
801                 change = KVM_MR_DELETE;
802         } else /* Modify a non-existent slot: disallowed. */
803                 goto out;
804
805         if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
806                 /* Check for overlaps */
807                 r = -EEXIST;
808                 kvm_for_each_memslot(slot, kvm->memslots) {
809                         if ((slot->id >= KVM_USER_MEM_SLOTS) ||
810                             (slot->id == mem->slot))
811                                 continue;
812                         if (!((base_gfn + npages <= slot->base_gfn) ||
813                               (base_gfn >= slot->base_gfn + slot->npages)))
814                                 goto out;
815                 }
816         }
817
818         /* Free page dirty bitmap if unneeded */
819         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
820                 new.dirty_bitmap = NULL;
821
822         r = -ENOMEM;
823         if (change == KVM_MR_CREATE) {
824                 new.userspace_addr = mem->userspace_addr;
825
826                 if (kvm_arch_create_memslot(&new, npages))
827                         goto out_free;
828         }
829
830         /* Allocate page dirty bitmap if needed */
831         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
832                 if (kvm_create_dirty_bitmap(&new) < 0)
833                         goto out_free;
834         }
835
836         if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
837                 r = -ENOMEM;
838                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
839                                 GFP_KERNEL);
840                 if (!slots)
841                         goto out_free;
842                 slot = id_to_memslot(slots, mem->slot);
843                 slot->flags |= KVM_MEMSLOT_INVALID;
844
845                 old_memslots = install_new_memslots(kvm, slots, NULL);
846
847                 /* slot was deleted or moved, clear iommu mapping */
848                 kvm_iommu_unmap_pages(kvm, &old);
849                 /* From this point no new shadow pages pointing to a deleted,
850                  * or moved, memslot will be created.
851                  *
852                  * validation of sp->gfn happens in:
853                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
854                  *      - kvm_is_visible_gfn (mmu_check_roots)
855                  */
856                 kvm_arch_flush_shadow_memslot(kvm, slot);
857                 slots = old_memslots;
858         }
859
860         r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
861         if (r)
862                 goto out_slots;
863
864         r = -ENOMEM;
865         /*
866          * We can re-use the old_memslots from above, the only difference
867          * from the currently installed memslots is the invalid flag.  This
868          * will get overwritten by update_memslots anyway.
869          */
870         if (!slots) {
871                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
872                                 GFP_KERNEL);
873                 if (!slots)
874                         goto out_free;
875         }
876
877         /*
878          * IOMMU mapping:  New slots need to be mapped.  Old slots need to be
879          * un-mapped and re-mapped if their base changes.  Since base change
880          * unmapping is handled above with slot deletion, mapping alone is
881          * needed here.  Anything else the iommu might care about for existing
882          * slots (size changes, userspace addr changes and read-only flag
883          * changes) is disallowed above, so any other attribute changes getting
884          * here can be skipped.
885          */
886         if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
887                 r = kvm_iommu_map_pages(kvm, &new);
888                 if (r)
889                         goto out_slots;
890         }
891
892         /* actual memory is freed via old in kvm_free_physmem_slot below */
893         if (change == KVM_MR_DELETE) {
894                 new.dirty_bitmap = NULL;
895                 memset(&new.arch, 0, sizeof(new.arch));
896         }
897
898         old_memslots = install_new_memslots(kvm, slots, &new);
899
900         kvm_arch_commit_memory_region(kvm, mem, &old, change);
901
902         kvm_free_physmem_slot(&old, &new);
903         kfree(old_memslots);
904
905         return 0;
906
907 out_slots:
908         kfree(slots);
909 out_free:
910         kvm_free_physmem_slot(&new, &old);
911 out:
912         return r;
913 }
914 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
915
916 int kvm_set_memory_region(struct kvm *kvm,
917                           struct kvm_userspace_memory_region *mem)
918 {
919         int r;
920
921         mutex_lock(&kvm->slots_lock);
922         r = __kvm_set_memory_region(kvm, mem);
923         mutex_unlock(&kvm->slots_lock);
924         return r;
925 }
926 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
927
928 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
929                                    struct kvm_userspace_memory_region *mem)
930 {
931         if (mem->slot >= KVM_USER_MEM_SLOTS)
932                 return -EINVAL;
933         return kvm_set_memory_region(kvm, mem);
934 }
935
936 int kvm_get_dirty_log(struct kvm *kvm,
937                         struct kvm_dirty_log *log, int *is_dirty)
938 {
939         struct kvm_memory_slot *memslot;
940         int r, i;
941         unsigned long n;
942         unsigned long any = 0;
943
944         r = -EINVAL;
945         if (log->slot >= KVM_USER_MEM_SLOTS)
946                 goto out;
947
948         memslot = id_to_memslot(kvm->memslots, log->slot);
949         r = -ENOENT;
950         if (!memslot->dirty_bitmap)
951                 goto out;
952
953         n = kvm_dirty_bitmap_bytes(memslot);
954
955         for (i = 0; !any && i < n/sizeof(long); ++i)
956                 any = memslot->dirty_bitmap[i];
957
958         r = -EFAULT;
959         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
960                 goto out;
961
962         if (any)
963                 *is_dirty = 1;
964
965         r = 0;
966 out:
967         return r;
968 }
969
970 bool kvm_largepages_enabled(void)
971 {
972         return largepages_enabled;
973 }
974
975 void kvm_disable_largepages(void)
976 {
977         largepages_enabled = false;
978 }
979 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
980
981 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
982 {
983         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
984 }
985 EXPORT_SYMBOL_GPL(gfn_to_memslot);
986
987 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
988 {
989         struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
990
991         if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
992               memslot->flags & KVM_MEMSLOT_INVALID)
993                 return 0;
994
995         return 1;
996 }
997 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
998
999 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1000 {
1001         struct vm_area_struct *vma;
1002         unsigned long addr, size;
1003
1004         size = PAGE_SIZE;
1005
1006         addr = gfn_to_hva(kvm, gfn);
1007         if (kvm_is_error_hva(addr))
1008                 return PAGE_SIZE;
1009
1010         down_read(&current->mm->mmap_sem);
1011         vma = find_vma(current->mm, addr);
1012         if (!vma)
1013                 goto out;
1014
1015         size = vma_kernel_pagesize(vma);
1016
1017 out:
1018         up_read(&current->mm->mmap_sem);
1019
1020         return size;
1021 }
1022
1023 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1024 {
1025         return slot->flags & KVM_MEM_READONLY;
1026 }
1027
1028 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1029                                        gfn_t *nr_pages, bool write)
1030 {
1031         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1032                 return KVM_HVA_ERR_BAD;
1033
1034         if (memslot_is_readonly(slot) && write)
1035                 return KVM_HVA_ERR_RO_BAD;
1036
1037         if (nr_pages)
1038                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1039
1040         return __gfn_to_hva_memslot(slot, gfn);
1041 }
1042
1043 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1044                                      gfn_t *nr_pages)
1045 {
1046         return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1047 }
1048
1049 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1050                                  gfn_t gfn)
1051 {
1052         return gfn_to_hva_many(slot, gfn, NULL);
1053 }
1054 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1055
1056 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1057 {
1058         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1059 }
1060 EXPORT_SYMBOL_GPL(gfn_to_hva);
1061
1062 /*
1063  * The hva returned by this function is only allowed to be read.
1064  * It should pair with kvm_read_hva() or kvm_read_hva_atomic().
1065  */
1066 static unsigned long gfn_to_hva_read(struct kvm *kvm, gfn_t gfn)
1067 {
1068         return __gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL, false);
1069 }
1070
1071 static int kvm_read_hva(void *data, void __user *hva, int len)
1072 {
1073         return __copy_from_user(data, hva, len);
1074 }
1075
1076 static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1077 {
1078         return __copy_from_user_inatomic(data, hva, len);
1079 }
1080
1081 static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1082         unsigned long start, int write, struct page **page)
1083 {
1084         int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1085
1086         if (write)
1087                 flags |= FOLL_WRITE;
1088
1089         return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1090 }
1091
1092 static inline int check_user_page_hwpoison(unsigned long addr)
1093 {
1094         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1095
1096         rc = __get_user_pages(current, current->mm, addr, 1,
1097                               flags, NULL, NULL, NULL);
1098         return rc == -EHWPOISON;
1099 }
1100
1101 /*
1102  * The atomic path to get the writable pfn which will be stored in @pfn,
1103  * true indicates success, otherwise false is returned.
1104  */
1105 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1106                             bool write_fault, bool *writable, pfn_t *pfn)
1107 {
1108         struct page *page[1];
1109         int npages;
1110
1111         if (!(async || atomic))
1112                 return false;
1113
1114         /*
1115          * Fast pin a writable pfn only if it is a write fault request
1116          * or the caller allows to map a writable pfn for a read fault
1117          * request.
1118          */
1119         if (!(write_fault || writable))
1120                 return false;
1121
1122         npages = __get_user_pages_fast(addr, 1, 1, page);
1123         if (npages == 1) {
1124                 *pfn = page_to_pfn(page[0]);
1125
1126                 if (writable)
1127                         *writable = true;
1128                 return true;
1129         }
1130
1131         return false;
1132 }
1133
1134 /*
1135  * The slow path to get the pfn of the specified host virtual address,
1136  * 1 indicates success, -errno is returned if error is detected.
1137  */
1138 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1139                            bool *writable, pfn_t *pfn)
1140 {
1141         struct page *page[1];
1142         int npages = 0;
1143
1144         might_sleep();
1145
1146         if (writable)
1147                 *writable = write_fault;
1148
1149         if (async) {
1150                 down_read(&current->mm->mmap_sem);
1151                 npages = get_user_page_nowait(current, current->mm,
1152                                               addr, write_fault, page);
1153                 up_read(&current->mm->mmap_sem);
1154         } else
1155                 npages = get_user_pages_fast(addr, 1, write_fault,
1156                                              page);
1157         if (npages != 1)
1158                 return npages;
1159
1160         /* map read fault as writable if possible */
1161         if (unlikely(!write_fault) && writable) {
1162                 struct page *wpage[1];
1163
1164                 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1165                 if (npages == 1) {
1166                         *writable = true;
1167                         put_page(page[0]);
1168                         page[0] = wpage[0];
1169                 }
1170
1171                 npages = 1;
1172         }
1173         *pfn = page_to_pfn(page[0]);
1174         return npages;
1175 }
1176
1177 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1178 {
1179         if (unlikely(!(vma->vm_flags & VM_READ)))
1180                 return false;
1181
1182         if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1183                 return false;
1184
1185         return true;
1186 }
1187
1188 /*
1189  * Pin guest page in memory and return its pfn.
1190  * @addr: host virtual address which maps memory to the guest
1191  * @atomic: whether this function can sleep
1192  * @async: whether this function need to wait IO complete if the
1193  *         host page is not in the memory
1194  * @write_fault: whether we should get a writable host page
1195  * @writable: whether it allows to map a writable host page for !@write_fault
1196  *
1197  * The function will map a writable host page for these two cases:
1198  * 1): @write_fault = true
1199  * 2): @write_fault = false && @writable, @writable will tell the caller
1200  *     whether the mapping is writable.
1201  */
1202 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1203                         bool write_fault, bool *writable)
1204 {
1205         struct vm_area_struct *vma;
1206         pfn_t pfn = 0;
1207         int npages;
1208
1209         /* we can do it either atomically or asynchronously, not both */
1210         BUG_ON(atomic && async);
1211
1212         if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1213                 return pfn;
1214
1215         if (atomic)
1216                 return KVM_PFN_ERR_FAULT;
1217
1218         npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1219         if (npages == 1)
1220                 return pfn;
1221
1222         down_read(&current->mm->mmap_sem);
1223         if (npages == -EHWPOISON ||
1224               (!async && check_user_page_hwpoison(addr))) {
1225                 pfn = KVM_PFN_ERR_HWPOISON;
1226                 goto exit;
1227         }
1228
1229         vma = find_vma_intersection(current->mm, addr, addr + 1);
1230
1231         if (vma == NULL)
1232                 pfn = KVM_PFN_ERR_FAULT;
1233         else if ((vma->vm_flags & VM_PFNMAP)) {
1234                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1235                         vma->vm_pgoff;
1236                 BUG_ON(!kvm_is_mmio_pfn(pfn));
1237         } else {
1238                 if (async && vma_is_valid(vma, write_fault))
1239                         *async = true;
1240                 pfn = KVM_PFN_ERR_FAULT;
1241         }
1242 exit:
1243         up_read(&current->mm->mmap_sem);
1244         return pfn;
1245 }
1246
1247 static pfn_t
1248 __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1249                      bool *async, bool write_fault, bool *writable)
1250 {
1251         unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1252
1253         if (addr == KVM_HVA_ERR_RO_BAD)
1254                 return KVM_PFN_ERR_RO_FAULT;
1255
1256         if (kvm_is_error_hva(addr))
1257                 return KVM_PFN_NOSLOT;
1258
1259         /* Do not map writable pfn in the readonly memslot. */
1260         if (writable && memslot_is_readonly(slot)) {
1261                 *writable = false;
1262                 writable = NULL;
1263         }
1264
1265         return hva_to_pfn(addr, atomic, async, write_fault,
1266                           writable);
1267 }
1268
1269 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1270                           bool write_fault, bool *writable)
1271 {
1272         struct kvm_memory_slot *slot;
1273
1274         if (async)
1275                 *async = false;
1276
1277         slot = gfn_to_memslot(kvm, gfn);
1278
1279         return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1280                                     writable);
1281 }
1282
1283 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1284 {
1285         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1286 }
1287 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1288
1289 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1290                        bool write_fault, bool *writable)
1291 {
1292         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1293 }
1294 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1295
1296 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1297 {
1298         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1299 }
1300 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1301
1302 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1303                       bool *writable)
1304 {
1305         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1306 }
1307 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1308
1309 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1310 {
1311         return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1312 }
1313
1314 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1315 {
1316         return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1317 }
1318 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1319
1320 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1321                                                                   int nr_pages)
1322 {
1323         unsigned long addr;
1324         gfn_t entry;
1325
1326         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1327         if (kvm_is_error_hva(addr))
1328                 return -1;
1329
1330         if (entry < nr_pages)
1331                 return 0;
1332
1333         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1334 }
1335 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1336
1337 static struct page *kvm_pfn_to_page(pfn_t pfn)
1338 {
1339         if (is_error_noslot_pfn(pfn))
1340                 return KVM_ERR_PTR_BAD_PAGE;
1341
1342         if (kvm_is_mmio_pfn(pfn)) {
1343                 WARN_ON(1);
1344                 return KVM_ERR_PTR_BAD_PAGE;
1345         }
1346
1347         return pfn_to_page(pfn);
1348 }
1349
1350 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1351 {
1352         pfn_t pfn;
1353
1354         pfn = gfn_to_pfn(kvm, gfn);
1355
1356         return kvm_pfn_to_page(pfn);
1357 }
1358
1359 EXPORT_SYMBOL_GPL(gfn_to_page);
1360
1361 void kvm_release_page_clean(struct page *page)
1362 {
1363         WARN_ON(is_error_page(page));
1364
1365         kvm_release_pfn_clean(page_to_pfn(page));
1366 }
1367 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1368
1369 void kvm_release_pfn_clean(pfn_t pfn)
1370 {
1371         if (!is_error_noslot_pfn(pfn) && !kvm_is_mmio_pfn(pfn))
1372                 put_page(pfn_to_page(pfn));
1373 }
1374 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1375
1376 void kvm_release_page_dirty(struct page *page)
1377 {
1378         WARN_ON(is_error_page(page));
1379
1380         kvm_release_pfn_dirty(page_to_pfn(page));
1381 }
1382 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1383
1384 void kvm_release_pfn_dirty(pfn_t pfn)
1385 {
1386         kvm_set_pfn_dirty(pfn);
1387         kvm_release_pfn_clean(pfn);
1388 }
1389 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1390
1391 void kvm_set_page_dirty(struct page *page)
1392 {
1393         kvm_set_pfn_dirty(page_to_pfn(page));
1394 }
1395 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1396
1397 void kvm_set_pfn_dirty(pfn_t pfn)
1398 {
1399         if (!kvm_is_mmio_pfn(pfn)) {
1400                 struct page *page = pfn_to_page(pfn);
1401                 if (!PageReserved(page))
1402                         SetPageDirty(page);
1403         }
1404 }
1405 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1406
1407 void kvm_set_pfn_accessed(pfn_t pfn)
1408 {
1409         if (!kvm_is_mmio_pfn(pfn))
1410                 mark_page_accessed(pfn_to_page(pfn));
1411 }
1412 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1413
1414 void kvm_get_pfn(pfn_t pfn)
1415 {
1416         if (!kvm_is_mmio_pfn(pfn))
1417                 get_page(pfn_to_page(pfn));
1418 }
1419 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1420
1421 static int next_segment(unsigned long len, int offset)
1422 {
1423         if (len > PAGE_SIZE - offset)
1424                 return PAGE_SIZE - offset;
1425         else
1426                 return len;
1427 }
1428
1429 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1430                         int len)
1431 {
1432         int r;
1433         unsigned long addr;
1434
1435         addr = gfn_to_hva_read(kvm, gfn);
1436         if (kvm_is_error_hva(addr))
1437                 return -EFAULT;
1438         r = kvm_read_hva(data, (void __user *)addr + offset, len);
1439         if (r)
1440                 return -EFAULT;
1441         return 0;
1442 }
1443 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1444
1445 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1446 {
1447         gfn_t gfn = gpa >> PAGE_SHIFT;
1448         int seg;
1449         int offset = offset_in_page(gpa);
1450         int ret;
1451
1452         while ((seg = next_segment(len, offset)) != 0) {
1453                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1454                 if (ret < 0)
1455                         return ret;
1456                 offset = 0;
1457                 len -= seg;
1458                 data += seg;
1459                 ++gfn;
1460         }
1461         return 0;
1462 }
1463 EXPORT_SYMBOL_GPL(kvm_read_guest);
1464
1465 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1466                           unsigned long len)
1467 {
1468         int r;
1469         unsigned long addr;
1470         gfn_t gfn = gpa >> PAGE_SHIFT;
1471         int offset = offset_in_page(gpa);
1472
1473         addr = gfn_to_hva_read(kvm, gfn);
1474         if (kvm_is_error_hva(addr))
1475                 return -EFAULT;
1476         pagefault_disable();
1477         r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
1478         pagefault_enable();
1479         if (r)
1480                 return -EFAULT;
1481         return 0;
1482 }
1483 EXPORT_SYMBOL(kvm_read_guest_atomic);
1484
1485 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1486                          int offset, int len)
1487 {
1488         int r;
1489         unsigned long addr;
1490
1491         addr = gfn_to_hva(kvm, gfn);
1492         if (kvm_is_error_hva(addr))
1493                 return -EFAULT;
1494         r = __copy_to_user((void __user *)addr + offset, data, len);
1495         if (r)
1496                 return -EFAULT;
1497         mark_page_dirty(kvm, gfn);
1498         return 0;
1499 }
1500 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1501
1502 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1503                     unsigned long len)
1504 {
1505         gfn_t gfn = gpa >> PAGE_SHIFT;
1506         int seg;
1507         int offset = offset_in_page(gpa);
1508         int ret;
1509
1510         while ((seg = next_segment(len, offset)) != 0) {
1511                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1512                 if (ret < 0)
1513                         return ret;
1514                 offset = 0;
1515                 len -= seg;
1516                 data += seg;
1517                 ++gfn;
1518         }
1519         return 0;
1520 }
1521
1522 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1523                               gpa_t gpa)
1524 {
1525         struct kvm_memslots *slots = kvm_memslots(kvm);
1526         int offset = offset_in_page(gpa);
1527         gfn_t gfn = gpa >> PAGE_SHIFT;
1528
1529         ghc->gpa = gpa;
1530         ghc->generation = slots->generation;
1531         ghc->memslot = gfn_to_memslot(kvm, gfn);
1532         ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1533         if (!kvm_is_error_hva(ghc->hva))
1534                 ghc->hva += offset;
1535         else
1536                 return -EFAULT;
1537
1538         return 0;
1539 }
1540 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1541
1542 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1543                            void *data, unsigned long len)
1544 {
1545         struct kvm_memslots *slots = kvm_memslots(kvm);
1546         int r;
1547
1548         if (slots->generation != ghc->generation)
1549                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1550
1551         if (kvm_is_error_hva(ghc->hva))
1552                 return -EFAULT;
1553
1554         r = __copy_to_user((void __user *)ghc->hva, data, len);
1555         if (r)
1556                 return -EFAULT;
1557         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1558
1559         return 0;
1560 }
1561 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1562
1563 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1564                            void *data, unsigned long len)
1565 {
1566         struct kvm_memslots *slots = kvm_memslots(kvm);
1567         int r;
1568
1569         if (slots->generation != ghc->generation)
1570                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1571
1572         if (kvm_is_error_hva(ghc->hva))
1573                 return -EFAULT;
1574
1575         r = __copy_from_user(data, (void __user *)ghc->hva, len);
1576         if (r)
1577                 return -EFAULT;
1578
1579         return 0;
1580 }
1581 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1582
1583 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1584 {
1585         return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1586                                     offset, len);
1587 }
1588 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1589
1590 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1591 {
1592         gfn_t gfn = gpa >> PAGE_SHIFT;
1593         int seg;
1594         int offset = offset_in_page(gpa);
1595         int ret;
1596
1597         while ((seg = next_segment(len, offset)) != 0) {
1598                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1599                 if (ret < 0)
1600                         return ret;
1601                 offset = 0;
1602                 len -= seg;
1603                 ++gfn;
1604         }
1605         return 0;
1606 }
1607 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1608
1609 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1610                              gfn_t gfn)
1611 {
1612         if (memslot && memslot->dirty_bitmap) {
1613                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1614
1615                 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1616         }
1617 }
1618
1619 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1620 {
1621         struct kvm_memory_slot *memslot;
1622
1623         memslot = gfn_to_memslot(kvm, gfn);
1624         mark_page_dirty_in_slot(kvm, memslot, gfn);
1625 }
1626
1627 /*
1628  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1629  */
1630 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1631 {
1632         DEFINE_WAIT(wait);
1633
1634         for (;;) {
1635                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1636
1637                 if (kvm_arch_vcpu_runnable(vcpu)) {
1638                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1639                         break;
1640                 }
1641                 if (kvm_cpu_has_pending_timer(vcpu))
1642                         break;
1643                 if (signal_pending(current))
1644                         break;
1645
1646                 schedule();
1647         }
1648
1649         finish_wait(&vcpu->wq, &wait);
1650 }
1651
1652 #ifndef CONFIG_S390
1653 /*
1654  * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1655  */
1656 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1657 {
1658         int me;
1659         int cpu = vcpu->cpu;
1660         wait_queue_head_t *wqp;
1661
1662         wqp = kvm_arch_vcpu_wq(vcpu);
1663         if (waitqueue_active(wqp)) {
1664                 wake_up_interruptible(wqp);
1665                 ++vcpu->stat.halt_wakeup;
1666         }
1667
1668         me = get_cpu();
1669         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1670                 if (kvm_arch_vcpu_should_kick(vcpu))
1671                         smp_send_reschedule(cpu);
1672         put_cpu();
1673 }
1674 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
1675 #endif /* !CONFIG_S390 */
1676
1677 void kvm_resched(struct kvm_vcpu *vcpu)
1678 {
1679         if (!need_resched())
1680                 return;
1681         cond_resched();
1682 }
1683 EXPORT_SYMBOL_GPL(kvm_resched);
1684
1685 bool kvm_vcpu_yield_to(struct kvm_vcpu *target)
1686 {
1687         struct pid *pid;
1688         struct task_struct *task = NULL;
1689         bool ret = false;
1690
1691         rcu_read_lock();
1692         pid = rcu_dereference(target->pid);
1693         if (pid)
1694                 task = get_pid_task(target->pid, PIDTYPE_PID);
1695         rcu_read_unlock();
1696         if (!task)
1697                 return ret;
1698         if (task->flags & PF_VCPU) {
1699                 put_task_struct(task);
1700                 return ret;
1701         }
1702         ret = yield_to(task, 1);
1703         put_task_struct(task);
1704
1705         return ret;
1706 }
1707 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1708
1709 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1710 /*
1711  * Helper that checks whether a VCPU is eligible for directed yield.
1712  * Most eligible candidate to yield is decided by following heuristics:
1713  *
1714  *  (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1715  *  (preempted lock holder), indicated by @in_spin_loop.
1716  *  Set at the beiginning and cleared at the end of interception/PLE handler.
1717  *
1718  *  (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1719  *  chance last time (mostly it has become eligible now since we have probably
1720  *  yielded to lockholder in last iteration. This is done by toggling
1721  *  @dy_eligible each time a VCPU checked for eligibility.)
1722  *
1723  *  Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1724  *  to preempted lock-holder could result in wrong VCPU selection and CPU
1725  *  burning. Giving priority for a potential lock-holder increases lock
1726  *  progress.
1727  *
1728  *  Since algorithm is based on heuristics, accessing another VCPU data without
1729  *  locking does not harm. It may result in trying to yield to  same VCPU, fail
1730  *  and continue with next VCPU and so on.
1731  */
1732 bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1733 {
1734         bool eligible;
1735
1736         eligible = !vcpu->spin_loop.in_spin_loop ||
1737                         (vcpu->spin_loop.in_spin_loop &&
1738                          vcpu->spin_loop.dy_eligible);
1739
1740         if (vcpu->spin_loop.in_spin_loop)
1741                 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1742
1743         return eligible;
1744 }
1745 #endif
1746
1747 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1748 {
1749         struct kvm *kvm = me->kvm;
1750         struct kvm_vcpu *vcpu;
1751         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1752         int yielded = 0;
1753         int try = 3;
1754         int pass;
1755         int i;
1756
1757         kvm_vcpu_set_in_spin_loop(me, true);
1758         /*
1759          * We boost the priority of a VCPU that is runnable but not
1760          * currently running, because it got preempted by something
1761          * else and called schedule in __vcpu_run.  Hopefully that
1762          * VCPU is holding the lock that we need and will release it.
1763          * We approximate round-robin by starting at the last boosted VCPU.
1764          */
1765         for (pass = 0; pass < 2 && !yielded && try; pass++) {
1766                 kvm_for_each_vcpu(i, vcpu, kvm) {
1767                         if (!pass && i <= last_boosted_vcpu) {
1768                                 i = last_boosted_vcpu;
1769                                 continue;
1770                         } else if (pass && i > last_boosted_vcpu)
1771                                 break;
1772                         if (!ACCESS_ONCE(vcpu->preempted))
1773                                 continue;
1774                         if (vcpu == me)
1775                                 continue;
1776                         if (waitqueue_active(&vcpu->wq))
1777                                 continue;
1778                         if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1779                                 continue;
1780
1781                         yielded = kvm_vcpu_yield_to(vcpu);
1782                         if (yielded > 0) {
1783                                 kvm->last_boosted_vcpu = i;
1784                                 break;
1785                         } else if (yielded < 0) {
1786                                 try--;
1787                                 if (!try)
1788                                         break;
1789                         }
1790                 }
1791         }
1792         kvm_vcpu_set_in_spin_loop(me, false);
1793
1794         /* Ensure vcpu is not eligible during next spinloop */
1795         kvm_vcpu_set_dy_eligible(me, false);
1796 }
1797 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1798
1799 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1800 {
1801         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1802         struct page *page;
1803
1804         if (vmf->pgoff == 0)
1805                 page = virt_to_page(vcpu->run);
1806 #ifdef CONFIG_X86
1807         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1808                 page = virt_to_page(vcpu->arch.pio_data);
1809 #endif
1810 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1811         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1812                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1813 #endif
1814         else
1815                 return kvm_arch_vcpu_fault(vcpu, vmf);
1816         get_page(page);
1817         vmf->page = page;
1818         return 0;
1819 }
1820
1821 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1822         .fault = kvm_vcpu_fault,
1823 };
1824
1825 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1826 {
1827         vma->vm_ops = &kvm_vcpu_vm_ops;
1828         return 0;
1829 }
1830
1831 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1832 {
1833         struct kvm_vcpu *vcpu = filp->private_data;
1834
1835         kvm_put_kvm(vcpu->kvm);
1836         return 0;
1837 }
1838
1839 static struct file_operations kvm_vcpu_fops = {
1840         .release        = kvm_vcpu_release,
1841         .unlocked_ioctl = kvm_vcpu_ioctl,
1842 #ifdef CONFIG_COMPAT
1843         .compat_ioctl   = kvm_vcpu_compat_ioctl,
1844 #endif
1845         .mmap           = kvm_vcpu_mmap,
1846         .llseek         = noop_llseek,
1847 };
1848
1849 /*
1850  * Allocates an inode for the vcpu.
1851  */
1852 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1853 {
1854         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1855 }
1856
1857 /*
1858  * Creates some virtual cpus.  Good luck creating more than one.
1859  */
1860 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1861 {
1862         int r;
1863         struct kvm_vcpu *vcpu, *v;
1864
1865         vcpu = kvm_arch_vcpu_create(kvm, id);
1866         if (IS_ERR(vcpu))
1867                 return PTR_ERR(vcpu);
1868
1869         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1870
1871         r = kvm_arch_vcpu_setup(vcpu);
1872         if (r)
1873                 goto vcpu_destroy;
1874
1875         mutex_lock(&kvm->lock);
1876         if (!kvm_vcpu_compatible(vcpu)) {
1877                 r = -EINVAL;
1878                 goto unlock_vcpu_destroy;
1879         }
1880         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1881                 r = -EINVAL;
1882                 goto unlock_vcpu_destroy;
1883         }
1884
1885         kvm_for_each_vcpu(r, v, kvm)
1886                 if (v->vcpu_id == id) {
1887                         r = -EEXIST;
1888                         goto unlock_vcpu_destroy;
1889                 }
1890
1891         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1892
1893         /* Now it's all set up, let userspace reach it */
1894         kvm_get_kvm(kvm);
1895         r = create_vcpu_fd(vcpu);
1896         if (r < 0) {
1897                 kvm_put_kvm(kvm);
1898                 goto unlock_vcpu_destroy;
1899         }
1900
1901         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1902         smp_wmb();
1903         atomic_inc(&kvm->online_vcpus);
1904
1905         mutex_unlock(&kvm->lock);
1906         kvm_arch_vcpu_postcreate(vcpu);
1907         return r;
1908
1909 unlock_vcpu_destroy:
1910         mutex_unlock(&kvm->lock);
1911 vcpu_destroy:
1912         kvm_arch_vcpu_destroy(vcpu);
1913         return r;
1914 }
1915
1916 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1917 {
1918         if (sigset) {
1919                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1920                 vcpu->sigset_active = 1;
1921                 vcpu->sigset = *sigset;
1922         } else
1923                 vcpu->sigset_active = 0;
1924         return 0;
1925 }
1926
1927 static long kvm_vcpu_ioctl(struct file *filp,
1928                            unsigned int ioctl, unsigned long arg)
1929 {
1930         struct kvm_vcpu *vcpu = filp->private_data;
1931         void __user *argp = (void __user *)arg;
1932         int r;
1933         struct kvm_fpu *fpu = NULL;
1934         struct kvm_sregs *kvm_sregs = NULL;
1935
1936         if (vcpu->kvm->mm != current->mm)
1937                 return -EIO;
1938
1939 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1940         /*
1941          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1942          * so vcpu_load() would break it.
1943          */
1944         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1945                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1946 #endif
1947
1948
1949         r = vcpu_load(vcpu);
1950         if (r)
1951                 return r;
1952         switch (ioctl) {
1953         case KVM_RUN:
1954                 r = -EINVAL;
1955                 if (arg)
1956                         goto out;
1957                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1958                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1959                 break;
1960         case KVM_GET_REGS: {
1961                 struct kvm_regs *kvm_regs;
1962
1963                 r = -ENOMEM;
1964                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1965                 if (!kvm_regs)
1966                         goto out;
1967                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1968                 if (r)
1969                         goto out_free1;
1970                 r = -EFAULT;
1971                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1972                         goto out_free1;
1973                 r = 0;
1974 out_free1:
1975                 kfree(kvm_regs);
1976                 break;
1977         }
1978         case KVM_SET_REGS: {
1979                 struct kvm_regs *kvm_regs;
1980
1981                 r = -ENOMEM;
1982                 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
1983                 if (IS_ERR(kvm_regs)) {
1984                         r = PTR_ERR(kvm_regs);
1985                         goto out;
1986                 }
1987                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1988                 kfree(kvm_regs);
1989                 break;
1990         }
1991         case KVM_GET_SREGS: {
1992                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1993                 r = -ENOMEM;
1994                 if (!kvm_sregs)
1995                         goto out;
1996                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1997                 if (r)
1998                         goto out;
1999                 r = -EFAULT;
2000                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2001                         goto out;
2002                 r = 0;
2003                 break;
2004         }
2005         case KVM_SET_SREGS: {
2006                 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2007                 if (IS_ERR(kvm_sregs)) {
2008                         r = PTR_ERR(kvm_sregs);
2009                         kvm_sregs = NULL;
2010                         goto out;
2011                 }
2012                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2013                 break;
2014         }
2015         case KVM_GET_MP_STATE: {
2016                 struct kvm_mp_state mp_state;
2017
2018                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2019                 if (r)
2020                         goto out;
2021                 r = -EFAULT;
2022                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
2023                         goto out;
2024                 r = 0;
2025                 break;
2026         }
2027         case KVM_SET_MP_STATE: {
2028                 struct kvm_mp_state mp_state;
2029
2030                 r = -EFAULT;
2031                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
2032                         goto out;
2033                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2034                 break;
2035         }
2036         case KVM_TRANSLATE: {
2037                 struct kvm_translation tr;
2038
2039                 r = -EFAULT;
2040                 if (copy_from_user(&tr, argp, sizeof tr))
2041                         goto out;
2042                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2043                 if (r)
2044                         goto out;
2045                 r = -EFAULT;
2046                 if (copy_to_user(argp, &tr, sizeof tr))
2047                         goto out;
2048                 r = 0;
2049                 break;
2050         }
2051         case KVM_SET_GUEST_DEBUG: {
2052                 struct kvm_guest_debug dbg;
2053
2054                 r = -EFAULT;
2055                 if (copy_from_user(&dbg, argp, sizeof dbg))
2056                         goto out;
2057                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2058                 break;
2059         }
2060         case KVM_SET_SIGNAL_MASK: {
2061                 struct kvm_signal_mask __user *sigmask_arg = argp;
2062                 struct kvm_signal_mask kvm_sigmask;
2063                 sigset_t sigset, *p;
2064
2065                 p = NULL;
2066                 if (argp) {
2067                         r = -EFAULT;
2068                         if (copy_from_user(&kvm_sigmask, argp,
2069                                            sizeof kvm_sigmask))
2070                                 goto out;
2071                         r = -EINVAL;
2072                         if (kvm_sigmask.len != sizeof sigset)
2073                                 goto out;
2074                         r = -EFAULT;
2075                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2076                                            sizeof sigset))
2077                                 goto out;
2078                         p = &sigset;
2079                 }
2080                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2081                 break;
2082         }
2083         case KVM_GET_FPU: {
2084                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2085                 r = -ENOMEM;
2086                 if (!fpu)
2087                         goto out;
2088                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2089                 if (r)
2090                         goto out;
2091                 r = -EFAULT;
2092                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2093                         goto out;
2094                 r = 0;
2095                 break;
2096         }
2097         case KVM_SET_FPU: {
2098                 fpu = memdup_user(argp, sizeof(*fpu));
2099                 if (IS_ERR(fpu)) {
2100                         r = PTR_ERR(fpu);
2101                         fpu = NULL;
2102                         goto out;
2103                 }
2104                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2105                 break;
2106         }
2107         default:
2108                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2109         }
2110 out:
2111         vcpu_put(vcpu);
2112         kfree(fpu);
2113         kfree(kvm_sregs);
2114         return r;
2115 }
2116
2117 #ifdef CONFIG_COMPAT
2118 static long kvm_vcpu_compat_ioctl(struct file *filp,
2119                                   unsigned int ioctl, unsigned long arg)
2120 {
2121         struct kvm_vcpu *vcpu = filp->private_data;
2122         void __user *argp = compat_ptr(arg);
2123         int r;
2124
2125         if (vcpu->kvm->mm != current->mm)
2126                 return -EIO;
2127
2128         switch (ioctl) {
2129         case KVM_SET_SIGNAL_MASK: {
2130                 struct kvm_signal_mask __user *sigmask_arg = argp;
2131                 struct kvm_signal_mask kvm_sigmask;
2132                 compat_sigset_t csigset;
2133                 sigset_t sigset;
2134
2135                 if (argp) {
2136                         r = -EFAULT;
2137                         if (copy_from_user(&kvm_sigmask, argp,
2138                                            sizeof kvm_sigmask))
2139                                 goto out;
2140                         r = -EINVAL;
2141                         if (kvm_sigmask.len != sizeof csigset)
2142                                 goto out;
2143                         r = -EFAULT;
2144                         if (copy_from_user(&csigset, sigmask_arg->sigset,
2145                                            sizeof csigset))
2146                                 goto out;
2147                         sigset_from_compat(&sigset, &csigset);
2148                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2149                 } else
2150                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2151                 break;
2152         }
2153         default:
2154                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2155         }
2156
2157 out:
2158         return r;
2159 }
2160 #endif
2161
2162 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2163                                  int (*accessor)(struct kvm_device *dev,
2164                                                  struct kvm_device_attr *attr),
2165                                  unsigned long arg)
2166 {
2167         struct kvm_device_attr attr;
2168
2169         if (!accessor)
2170                 return -EPERM;
2171
2172         if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2173                 return -EFAULT;
2174
2175         return accessor(dev, &attr);
2176 }
2177
2178 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2179                              unsigned long arg)
2180 {
2181         struct kvm_device *dev = filp->private_data;
2182
2183         switch (ioctl) {
2184         case KVM_SET_DEVICE_ATTR:
2185                 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2186         case KVM_GET_DEVICE_ATTR:
2187                 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2188         case KVM_HAS_DEVICE_ATTR:
2189                 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2190         default:
2191                 if (dev->ops->ioctl)
2192                         return dev->ops->ioctl(dev, ioctl, arg);
2193
2194                 return -ENOTTY;
2195         }
2196 }
2197
2198 void kvm_device_get(struct kvm_device *dev)
2199 {
2200         atomic_inc(&dev->users);
2201 }
2202
2203 void kvm_device_put(struct kvm_device *dev)
2204 {
2205         if (atomic_dec_and_test(&dev->users))
2206                 dev->ops->destroy(dev);
2207 }
2208
2209 static int kvm_device_release(struct inode *inode, struct file *filp)
2210 {
2211         struct kvm_device *dev = filp->private_data;
2212         struct kvm *kvm = dev->kvm;
2213
2214         kvm_device_put(dev);
2215         kvm_put_kvm(kvm);
2216         return 0;
2217 }
2218
2219 static const struct file_operations kvm_device_fops = {
2220         .unlocked_ioctl = kvm_device_ioctl,
2221         .release = kvm_device_release,
2222 };
2223
2224 struct kvm_device *kvm_device_from_filp(struct file *filp)
2225 {
2226         if (filp->f_op != &kvm_device_fops)
2227                 return NULL;
2228
2229         return filp->private_data;
2230 }
2231
2232 static int kvm_ioctl_create_device(struct kvm *kvm,
2233                                    struct kvm_create_device *cd)
2234 {
2235         struct kvm_device_ops *ops = NULL;
2236         struct kvm_device *dev;
2237         bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2238         int ret;
2239
2240         switch (cd->type) {
2241 #ifdef CONFIG_KVM_MPIC
2242         case KVM_DEV_TYPE_FSL_MPIC_20:
2243         case KVM_DEV_TYPE_FSL_MPIC_42:
2244                 ops = &kvm_mpic_ops;
2245                 break;
2246 #endif
2247         default:
2248                 return -ENODEV;
2249         }
2250
2251         if (test)
2252                 return 0;
2253
2254         dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2255         if (!dev)
2256                 return -ENOMEM;
2257
2258         dev->ops = ops;
2259         dev->kvm = kvm;
2260         atomic_set(&dev->users, 1);
2261
2262         ret = ops->create(dev, cd->type);
2263         if (ret < 0) {
2264                 kfree(dev);
2265                 return ret;
2266         }
2267
2268         ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR);
2269         if (ret < 0) {
2270                 ops->destroy(dev);
2271                 return ret;
2272         }
2273
2274         kvm_get_kvm(kvm);
2275         cd->fd = ret;
2276         return 0;
2277 }
2278
2279 static long kvm_vm_ioctl(struct file *filp,
2280                            unsigned int ioctl, unsigned long arg)
2281 {
2282         struct kvm *kvm = filp->private_data;
2283         void __user *argp = (void __user *)arg;
2284         int r;
2285
2286         if (kvm->mm != current->mm)
2287                 return -EIO;
2288         switch (ioctl) {
2289         case KVM_CREATE_VCPU:
2290                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2291                 break;
2292         case KVM_SET_USER_MEMORY_REGION: {
2293                 struct kvm_userspace_memory_region kvm_userspace_mem;
2294
2295                 r = -EFAULT;
2296                 if (copy_from_user(&kvm_userspace_mem, argp,
2297                                                 sizeof kvm_userspace_mem))
2298                         goto out;
2299
2300                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2301                 break;
2302         }
2303         case KVM_GET_DIRTY_LOG: {
2304                 struct kvm_dirty_log log;
2305
2306                 r = -EFAULT;
2307                 if (copy_from_user(&log, argp, sizeof log))
2308                         goto out;
2309                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2310                 break;
2311         }
2312 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2313         case KVM_REGISTER_COALESCED_MMIO: {
2314                 struct kvm_coalesced_mmio_zone zone;
2315                 r = -EFAULT;
2316                 if (copy_from_user(&zone, argp, sizeof zone))
2317                         goto out;
2318                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2319                 break;
2320         }
2321         case KVM_UNREGISTER_COALESCED_MMIO: {
2322                 struct kvm_coalesced_mmio_zone zone;
2323                 r = -EFAULT;
2324                 if (copy_from_user(&zone, argp, sizeof zone))
2325                         goto out;
2326                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2327                 break;
2328         }
2329 #endif
2330         case KVM_IRQFD: {
2331                 struct kvm_irqfd data;
2332
2333                 r = -EFAULT;
2334                 if (copy_from_user(&data, argp, sizeof data))
2335                         goto out;
2336                 r = kvm_irqfd(kvm, &data);
2337                 break;
2338         }
2339         case KVM_IOEVENTFD: {
2340                 struct kvm_ioeventfd data;
2341
2342                 r = -EFAULT;
2343                 if (copy_from_user(&data, argp, sizeof data))
2344                         goto out;
2345                 r = kvm_ioeventfd(kvm, &data);
2346                 break;
2347         }
2348 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2349         case KVM_SET_BOOT_CPU_ID:
2350                 r = 0;
2351                 mutex_lock(&kvm->lock);
2352                 if (atomic_read(&kvm->online_vcpus) != 0)
2353                         r = -EBUSY;
2354                 else
2355                         kvm->bsp_vcpu_id = arg;
2356                 mutex_unlock(&kvm->lock);
2357                 break;
2358 #endif
2359 #ifdef CONFIG_HAVE_KVM_MSI
2360         case KVM_SIGNAL_MSI: {
2361                 struct kvm_msi msi;
2362
2363                 r = -EFAULT;
2364                 if (copy_from_user(&msi, argp, sizeof msi))
2365                         goto out;
2366                 r = kvm_send_userspace_msi(kvm, &msi);
2367                 break;
2368         }
2369 #endif
2370 #ifdef __KVM_HAVE_IRQ_LINE
2371         case KVM_IRQ_LINE_STATUS:
2372         case KVM_IRQ_LINE: {
2373                 struct kvm_irq_level irq_event;
2374
2375                 r = -EFAULT;
2376                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2377                         goto out;
2378
2379                 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
2380                                         ioctl == KVM_IRQ_LINE_STATUS);
2381                 if (r)
2382                         goto out;
2383
2384                 r = -EFAULT;
2385                 if (ioctl == KVM_IRQ_LINE_STATUS) {
2386                         if (copy_to_user(argp, &irq_event, sizeof irq_event))
2387                                 goto out;
2388                 }
2389
2390                 r = 0;
2391                 break;
2392         }
2393 #endif
2394 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2395         case KVM_SET_GSI_ROUTING: {
2396                 struct kvm_irq_routing routing;
2397                 struct kvm_irq_routing __user *urouting;
2398                 struct kvm_irq_routing_entry *entries;
2399
2400                 r = -EFAULT;
2401                 if (copy_from_user(&routing, argp, sizeof(routing)))
2402                         goto out;
2403                 r = -EINVAL;
2404                 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2405                         goto out;
2406                 if (routing.flags)
2407                         goto out;
2408                 r = -ENOMEM;
2409                 entries = vmalloc(routing.nr * sizeof(*entries));
2410                 if (!entries)
2411                         goto out;
2412                 r = -EFAULT;
2413                 urouting = argp;
2414                 if (copy_from_user(entries, urouting->entries,
2415                                    routing.nr * sizeof(*entries)))
2416                         goto out_free_irq_routing;
2417                 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2418                                         routing.flags);
2419         out_free_irq_routing:
2420                 vfree(entries);
2421                 break;
2422         }
2423 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
2424         case KVM_CREATE_DEVICE: {
2425                 struct kvm_create_device cd;
2426
2427                 r = -EFAULT;
2428                 if (copy_from_user(&cd, argp, sizeof(cd)))
2429                         goto out;
2430
2431                 r = kvm_ioctl_create_device(kvm, &cd);
2432                 if (r)
2433                         goto out;
2434
2435                 r = -EFAULT;
2436                 if (copy_to_user(argp, &cd, sizeof(cd)))
2437                         goto out;
2438
2439                 r = 0;
2440                 break;
2441         }
2442         default:
2443                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2444                 if (r == -ENOTTY)
2445                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2446         }
2447 out:
2448         return r;
2449 }
2450
2451 #ifdef CONFIG_COMPAT
2452 struct compat_kvm_dirty_log {
2453         __u32 slot;
2454         __u32 padding1;
2455         union {
2456                 compat_uptr_t dirty_bitmap; /* one bit per page */
2457                 __u64 padding2;
2458         };
2459 };
2460
2461 static long kvm_vm_compat_ioctl(struct file *filp,
2462                            unsigned int ioctl, unsigned long arg)
2463 {
2464         struct kvm *kvm = filp->private_data;
2465         int r;
2466
2467         if (kvm->mm != current->mm)
2468                 return -EIO;
2469         switch (ioctl) {
2470         case KVM_GET_DIRTY_LOG: {
2471                 struct compat_kvm_dirty_log compat_log;
2472                 struct kvm_dirty_log log;
2473
2474                 r = -EFAULT;
2475                 if (copy_from_user(&compat_log, (void __user *)arg,
2476                                    sizeof(compat_log)))
2477                         goto out;
2478                 log.slot         = compat_log.slot;
2479                 log.padding1     = compat_log.padding1;
2480                 log.padding2     = compat_log.padding2;
2481                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2482
2483                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2484                 break;
2485         }
2486         default:
2487                 r = kvm_vm_ioctl(filp, ioctl, arg);
2488         }
2489
2490 out:
2491         return r;
2492 }
2493 #endif
2494
2495 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2496 {
2497         struct page *page[1];
2498         unsigned long addr;
2499         int npages;
2500         gfn_t gfn = vmf->pgoff;
2501         struct kvm *kvm = vma->vm_file->private_data;
2502
2503         addr = gfn_to_hva(kvm, gfn);
2504         if (kvm_is_error_hva(addr))
2505                 return VM_FAULT_SIGBUS;
2506
2507         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2508                                 NULL);
2509         if (unlikely(npages != 1))
2510                 return VM_FAULT_SIGBUS;
2511
2512         vmf->page = page[0];
2513         return 0;
2514 }
2515
2516 static const struct vm_operations_struct kvm_vm_vm_ops = {
2517         .fault = kvm_vm_fault,
2518 };
2519
2520 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2521 {
2522         vma->vm_ops = &kvm_vm_vm_ops;
2523         return 0;
2524 }
2525
2526 static struct file_operations kvm_vm_fops = {
2527         .release        = kvm_vm_release,
2528         .unlocked_ioctl = kvm_vm_ioctl,
2529 #ifdef CONFIG_COMPAT
2530         .compat_ioctl   = kvm_vm_compat_ioctl,
2531 #endif
2532         .mmap           = kvm_vm_mmap,
2533         .llseek         = noop_llseek,
2534 };
2535
2536 static int kvm_dev_ioctl_create_vm(unsigned long type)
2537 {
2538         int r;
2539         struct kvm *kvm;
2540
2541         kvm = kvm_create_vm(type);
2542         if (IS_ERR(kvm))
2543                 return PTR_ERR(kvm);
2544 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2545         r = kvm_coalesced_mmio_init(kvm);
2546         if (r < 0) {
2547                 kvm_put_kvm(kvm);
2548                 return r;
2549         }
2550 #endif
2551         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2552         if (r < 0)
2553                 kvm_put_kvm(kvm);
2554
2555         return r;
2556 }
2557
2558 static long kvm_dev_ioctl_check_extension_generic(long arg)
2559 {
2560         switch (arg) {
2561         case KVM_CAP_USER_MEMORY:
2562         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2563         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2564 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2565         case KVM_CAP_SET_BOOT_CPU_ID:
2566 #endif
2567         case KVM_CAP_INTERNAL_ERROR_DATA:
2568 #ifdef CONFIG_HAVE_KVM_MSI
2569         case KVM_CAP_SIGNAL_MSI:
2570 #endif
2571 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2572         case KVM_CAP_IRQFD_RESAMPLE:
2573 #endif
2574                 return 1;
2575 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2576         case KVM_CAP_IRQ_ROUTING:
2577                 return KVM_MAX_IRQ_ROUTES;
2578 #endif
2579         default:
2580                 break;
2581         }
2582         return kvm_dev_ioctl_check_extension(arg);
2583 }
2584
2585 static long kvm_dev_ioctl(struct file *filp,
2586                           unsigned int ioctl, unsigned long arg)
2587 {
2588         long r = -EINVAL;
2589
2590         switch (ioctl) {
2591         case KVM_GET_API_VERSION:
2592                 r = -EINVAL;
2593                 if (arg)
2594                         goto out;
2595                 r = KVM_API_VERSION;
2596                 break;
2597         case KVM_CREATE_VM:
2598                 r = kvm_dev_ioctl_create_vm(arg);
2599                 break;
2600         case KVM_CHECK_EXTENSION:
2601                 r = kvm_dev_ioctl_check_extension_generic(arg);
2602                 break;
2603         case KVM_GET_VCPU_MMAP_SIZE:
2604                 r = -EINVAL;
2605                 if (arg)
2606                         goto out;
2607                 r = PAGE_SIZE;     /* struct kvm_run */
2608 #ifdef CONFIG_X86
2609                 r += PAGE_SIZE;    /* pio data page */
2610 #endif
2611 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2612                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2613 #endif
2614                 break;
2615         case KVM_TRACE_ENABLE:
2616         case KVM_TRACE_PAUSE:
2617         case KVM_TRACE_DISABLE:
2618                 r = -EOPNOTSUPP;
2619                 break;
2620         default:
2621                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2622         }
2623 out:
2624         return r;
2625 }
2626
2627 static struct file_operations kvm_chardev_ops = {
2628         .unlocked_ioctl = kvm_dev_ioctl,
2629         .compat_ioctl   = kvm_dev_ioctl,
2630         .llseek         = noop_llseek,
2631 };
2632
2633 static struct miscdevice kvm_dev = {
2634         KVM_MINOR,
2635         "kvm",
2636         &kvm_chardev_ops,
2637 };
2638
2639 static void hardware_enable_nolock(void *junk)
2640 {
2641         int cpu = raw_smp_processor_id();
2642         int r;
2643
2644         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2645                 return;
2646
2647         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2648
2649         r = kvm_arch_hardware_enable(NULL);
2650
2651         if (r) {
2652                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2653                 atomic_inc(&hardware_enable_failed);
2654                 printk(KERN_INFO "kvm: enabling virtualization on "
2655                                  "CPU%d failed\n", cpu);
2656         }
2657 }
2658
2659 static void hardware_enable(void *junk)
2660 {
2661         raw_spin_lock(&kvm_lock);
2662         hardware_enable_nolock(junk);
2663         raw_spin_unlock(&kvm_lock);
2664 }
2665
2666 static void hardware_disable_nolock(void *junk)
2667 {
2668         int cpu = raw_smp_processor_id();
2669
2670         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2671                 return;
2672         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2673         kvm_arch_hardware_disable(NULL);
2674 }
2675
2676 static void hardware_disable(void *junk)
2677 {
2678         raw_spin_lock(&kvm_lock);
2679         hardware_disable_nolock(junk);
2680         raw_spin_unlock(&kvm_lock);
2681 }
2682
2683 static void hardware_disable_all_nolock(void)
2684 {
2685         BUG_ON(!kvm_usage_count);
2686
2687         kvm_usage_count--;
2688         if (!kvm_usage_count)
2689                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2690 }
2691
2692 static void hardware_disable_all(void)
2693 {
2694         raw_spin_lock(&kvm_lock);
2695         hardware_disable_all_nolock();
2696         raw_spin_unlock(&kvm_lock);
2697 }
2698
2699 static int hardware_enable_all(void)
2700 {
2701         int r = 0;
2702
2703         raw_spin_lock(&kvm_lock);
2704
2705         kvm_usage_count++;
2706         if (kvm_usage_count == 1) {
2707                 atomic_set(&hardware_enable_failed, 0);
2708                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2709
2710                 if (atomic_read(&hardware_enable_failed)) {
2711                         hardware_disable_all_nolock();
2712                         r = -EBUSY;
2713                 }
2714         }
2715
2716         raw_spin_unlock(&kvm_lock);
2717
2718         return r;
2719 }
2720
2721 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2722                            void *v)
2723 {
2724         int cpu = (long)v;
2725
2726         if (!kvm_usage_count)
2727                 return NOTIFY_OK;
2728
2729         val &= ~CPU_TASKS_FROZEN;
2730         switch (val) {
2731         case CPU_DYING:
2732                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2733                        cpu);
2734                 hardware_disable(NULL);
2735                 break;
2736         case CPU_STARTING:
2737                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2738                        cpu);
2739                 hardware_enable(NULL);
2740                 break;
2741         }
2742         return NOTIFY_OK;
2743 }
2744
2745 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2746                       void *v)
2747 {
2748         /*
2749          * Some (well, at least mine) BIOSes hang on reboot if
2750          * in vmx root mode.
2751          *
2752          * And Intel TXT required VMX off for all cpu when system shutdown.
2753          */
2754         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2755         kvm_rebooting = true;
2756         on_each_cpu(hardware_disable_nolock, NULL, 1);
2757         return NOTIFY_OK;
2758 }
2759
2760 static struct notifier_block kvm_reboot_notifier = {
2761         .notifier_call = kvm_reboot,
2762         .priority = 0,
2763 };
2764
2765 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2766 {
2767         int i;
2768
2769         for (i = 0; i < bus->dev_count; i++) {
2770                 struct kvm_io_device *pos = bus->range[i].dev;
2771
2772                 kvm_iodevice_destructor(pos);
2773         }
2774         kfree(bus);
2775 }
2776
2777 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2778 {
2779         const struct kvm_io_range *r1 = p1;
2780         const struct kvm_io_range *r2 = p2;
2781
2782         if (r1->addr < r2->addr)
2783                 return -1;
2784         if (r1->addr + r1->len > r2->addr + r2->len)
2785                 return 1;
2786         return 0;
2787 }
2788
2789 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2790                           gpa_t addr, int len)
2791 {
2792         bus->range[bus->dev_count++] = (struct kvm_io_range) {
2793                 .addr = addr,
2794                 .len = len,
2795                 .dev = dev,
2796         };
2797
2798         sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2799                 kvm_io_bus_sort_cmp, NULL);
2800
2801         return 0;
2802 }
2803
2804 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2805                              gpa_t addr, int len)
2806 {
2807         struct kvm_io_range *range, key;
2808         int off;
2809
2810         key = (struct kvm_io_range) {
2811                 .addr = addr,
2812                 .len = len,
2813         };
2814
2815         range = bsearch(&key, bus->range, bus->dev_count,
2816                         sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2817         if (range == NULL)
2818                 return -ENOENT;
2819
2820         off = range - bus->range;
2821
2822         while (off > 0 && kvm_io_bus_sort_cmp(&key, &bus->range[off-1]) == 0)
2823                 off--;
2824
2825         return off;
2826 }
2827
2828 /* kvm_io_bus_write - called under kvm->slots_lock */
2829 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2830                      int len, const void *val)
2831 {
2832         int idx;
2833         struct kvm_io_bus *bus;
2834         struct kvm_io_range range;
2835
2836         range = (struct kvm_io_range) {
2837                 .addr = addr,
2838                 .len = len,
2839         };
2840
2841         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2842         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2843         if (idx < 0)
2844                 return -EOPNOTSUPP;
2845
2846         while (idx < bus->dev_count &&
2847                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2848                 if (!kvm_iodevice_write(bus->range[idx].dev, addr, len, val))
2849                         return 0;
2850                 idx++;
2851         }
2852
2853         return -EOPNOTSUPP;
2854 }
2855
2856 /* kvm_io_bus_read - called under kvm->slots_lock */
2857 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2858                     int len, void *val)
2859 {
2860         int idx;
2861         struct kvm_io_bus *bus;
2862         struct kvm_io_range range;
2863
2864         range = (struct kvm_io_range) {
2865                 .addr = addr,
2866                 .len = len,
2867         };
2868
2869         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2870         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2871         if (idx < 0)
2872                 return -EOPNOTSUPP;
2873
2874         while (idx < bus->dev_count &&
2875                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2876                 if (!kvm_iodevice_read(bus->range[idx].dev, addr, len, val))
2877                         return 0;
2878                 idx++;
2879         }
2880
2881         return -EOPNOTSUPP;
2882 }
2883
2884 /* Caller must hold slots_lock. */
2885 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2886                             int len, struct kvm_io_device *dev)
2887 {
2888         struct kvm_io_bus *new_bus, *bus;
2889
2890         bus = kvm->buses[bus_idx];
2891         if (bus->dev_count > NR_IOBUS_DEVS - 1)
2892                 return -ENOSPC;
2893
2894         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
2895                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2896         if (!new_bus)
2897                 return -ENOMEM;
2898         memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
2899                sizeof(struct kvm_io_range)));
2900         kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2901         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2902         synchronize_srcu_expedited(&kvm->srcu);
2903         kfree(bus);
2904
2905         return 0;
2906 }
2907
2908 /* Caller must hold slots_lock. */
2909 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2910                               struct kvm_io_device *dev)
2911 {
2912         int i, r;
2913         struct kvm_io_bus *new_bus, *bus;
2914
2915         bus = kvm->buses[bus_idx];
2916         r = -ENOENT;
2917         for (i = 0; i < bus->dev_count; i++)
2918                 if (bus->range[i].dev == dev) {
2919                         r = 0;
2920                         break;
2921                 }
2922
2923         if (r)
2924                 return r;
2925
2926         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
2927                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2928         if (!new_bus)
2929                 return -ENOMEM;
2930
2931         memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
2932         new_bus->dev_count--;
2933         memcpy(new_bus->range + i, bus->range + i + 1,
2934                (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
2935
2936         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2937         synchronize_srcu_expedited(&kvm->srcu);
2938         kfree(bus);
2939         return r;
2940 }
2941
2942 static struct notifier_block kvm_cpu_notifier = {
2943         .notifier_call = kvm_cpu_hotplug,
2944 };
2945
2946 static int vm_stat_get(void *_offset, u64 *val)
2947 {
2948         unsigned offset = (long)_offset;
2949         struct kvm *kvm;
2950
2951         *val = 0;
2952         raw_spin_lock(&kvm_lock);
2953         list_for_each_entry(kvm, &vm_list, vm_list)
2954                 *val += *(u32 *)((void *)kvm + offset);
2955         raw_spin_unlock(&kvm_lock);
2956         return 0;
2957 }
2958
2959 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2960
2961 static int vcpu_stat_get(void *_offset, u64 *val)
2962 {
2963         unsigned offset = (long)_offset;
2964         struct kvm *kvm;
2965         struct kvm_vcpu *vcpu;
2966         int i;
2967
2968         *val = 0;
2969         raw_spin_lock(&kvm_lock);
2970         list_for_each_entry(kvm, &vm_list, vm_list)
2971                 kvm_for_each_vcpu(i, vcpu, kvm)
2972                         *val += *(u32 *)((void *)vcpu + offset);
2973
2974         raw_spin_unlock(&kvm_lock);
2975         return 0;
2976 }
2977
2978 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2979
2980 static const struct file_operations *stat_fops[] = {
2981         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2982         [KVM_STAT_VM]   = &vm_stat_fops,
2983 };
2984
2985 static int kvm_init_debug(void)
2986 {
2987         int r = -EFAULT;
2988         struct kvm_stats_debugfs_item *p;
2989
2990         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2991         if (kvm_debugfs_dir == NULL)
2992                 goto out;
2993
2994         for (p = debugfs_entries; p->name; ++p) {
2995                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2996                                                 (void *)(long)p->offset,
2997                                                 stat_fops[p->kind]);
2998                 if (p->dentry == NULL)
2999                         goto out_dir;
3000         }
3001
3002         return 0;
3003
3004 out_dir:
3005         debugfs_remove_recursive(kvm_debugfs_dir);
3006 out:
3007         return r;
3008 }
3009
3010 static void kvm_exit_debug(void)
3011 {
3012         struct kvm_stats_debugfs_item *p;
3013
3014         for (p = debugfs_entries; p->name; ++p)
3015                 debugfs_remove(p->dentry);
3016         debugfs_remove(kvm_debugfs_dir);
3017 }
3018
3019 static int kvm_suspend(void)
3020 {
3021         if (kvm_usage_count)
3022                 hardware_disable_nolock(NULL);
3023         return 0;
3024 }
3025
3026 static void kvm_resume(void)
3027 {
3028         if (kvm_usage_count) {
3029                 WARN_ON(raw_spin_is_locked(&kvm_lock));
3030                 hardware_enable_nolock(NULL);
3031         }
3032 }
3033
3034 static struct syscore_ops kvm_syscore_ops = {
3035         .suspend = kvm_suspend,
3036         .resume = kvm_resume,
3037 };
3038
3039 static inline
3040 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3041 {
3042         return container_of(pn, struct kvm_vcpu, preempt_notifier);
3043 }
3044
3045 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3046 {
3047         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3048         if (vcpu->preempted)
3049                 vcpu->preempted = false;
3050
3051         kvm_arch_vcpu_load(vcpu, cpu);
3052 }
3053
3054 static void kvm_sched_out(struct preempt_notifier *pn,
3055                           struct task_struct *next)
3056 {
3057         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3058
3059         if (current->state == TASK_RUNNING)
3060                 vcpu->preempted = true;
3061         kvm_arch_vcpu_put(vcpu);
3062 }
3063
3064 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3065                   struct module *module)
3066 {
3067         int r;
3068         int cpu;
3069
3070         r = kvm_irqfd_init();
3071         if (r)
3072                 goto out_irqfd;
3073         r = kvm_arch_init(opaque);
3074         if (r)
3075                 goto out_fail;
3076
3077         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
3078                 r = -ENOMEM;
3079                 goto out_free_0;
3080         }
3081
3082         r = kvm_arch_hardware_setup();
3083         if (r < 0)
3084                 goto out_free_0a;
3085
3086         for_each_online_cpu(cpu) {
3087                 smp_call_function_single(cpu,
3088                                 kvm_arch_check_processor_compat,
3089                                 &r, 1);
3090                 if (r < 0)
3091                         goto out_free_1;
3092         }
3093
3094         r = register_cpu_notifier(&kvm_cpu_notifier);
3095         if (r)
3096                 goto out_free_2;
3097         register_reboot_notifier(&kvm_reboot_notifier);
3098
3099         /* A kmem cache lets us meet the alignment requirements of fx_save. */
3100         if (!vcpu_align)
3101                 vcpu_align = __alignof__(struct kvm_vcpu);
3102         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
3103                                            0, NULL);
3104         if (!kvm_vcpu_cache) {
3105                 r = -ENOMEM;
3106                 goto out_free_3;
3107         }
3108
3109         r = kvm_async_pf_init();
3110         if (r)
3111                 goto out_free;
3112
3113         kvm_chardev_ops.owner = module;
3114         kvm_vm_fops.owner = module;
3115         kvm_vcpu_fops.owner = module;
3116
3117         r = misc_register(&kvm_dev);
3118         if (r) {
3119                 printk(KERN_ERR "kvm: misc device register failed\n");
3120                 goto out_unreg;
3121         }
3122
3123         register_syscore_ops(&kvm_syscore_ops);
3124
3125         kvm_preempt_ops.sched_in = kvm_sched_in;
3126         kvm_preempt_ops.sched_out = kvm_sched_out;
3127
3128         r = kvm_init_debug();
3129         if (r) {
3130                 printk(KERN_ERR "kvm: create debugfs files failed\n");
3131                 goto out_undebugfs;
3132         }
3133
3134         return 0;
3135
3136 out_undebugfs:
3137         unregister_syscore_ops(&kvm_syscore_ops);
3138 out_unreg:
3139         kvm_async_pf_deinit();
3140 out_free:
3141         kmem_cache_destroy(kvm_vcpu_cache);
3142 out_free_3:
3143         unregister_reboot_notifier(&kvm_reboot_notifier);
3144         unregister_cpu_notifier(&kvm_cpu_notifier);
3145 out_free_2:
3146 out_free_1:
3147         kvm_arch_hardware_unsetup();
3148 out_free_0a:
3149         free_cpumask_var(cpus_hardware_enabled);
3150 out_free_0:
3151         kvm_arch_exit();
3152 out_fail:
3153         kvm_irqfd_exit();
3154 out_irqfd:
3155         return r;
3156 }
3157 EXPORT_SYMBOL_GPL(kvm_init);
3158
3159 void kvm_exit(void)
3160 {
3161         kvm_exit_debug();
3162         misc_deregister(&kvm_dev);
3163         kmem_cache_destroy(kvm_vcpu_cache);
3164         kvm_async_pf_deinit();
3165         unregister_syscore_ops(&kvm_syscore_ops);
3166         unregister_reboot_notifier(&kvm_reboot_notifier);
3167         unregister_cpu_notifier(&kvm_cpu_notifier);
3168         on_each_cpu(hardware_disable_nolock, NULL, 1);
3169         kvm_arch_hardware_unsetup();
3170         kvm_arch_exit();
3171         kvm_irqfd_exit();
3172         free_cpumask_var(cpus_hardware_enabled);
3173 }
3174 EXPORT_SYMBOL_GPL(kvm_exit);