5e709ebb7c408e5e90a7b9cb156062edad54f34f
[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 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
221 {
222         struct page *page;
223         int r;
224
225         mutex_init(&vcpu->mutex);
226         vcpu->cpu = -1;
227         vcpu->kvm = kvm;
228         vcpu->vcpu_id = id;
229         vcpu->pid = NULL;
230         init_waitqueue_head(&vcpu->wq);
231         kvm_async_pf_vcpu_init(vcpu);
232
233         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
234         if (!page) {
235                 r = -ENOMEM;
236                 goto fail;
237         }
238         vcpu->run = page_address(page);
239
240         kvm_vcpu_set_in_spin_loop(vcpu, false);
241         kvm_vcpu_set_dy_eligible(vcpu, false);
242
243         r = kvm_arch_vcpu_init(vcpu);
244         if (r < 0)
245                 goto fail_free_run;
246         return 0;
247
248 fail_free_run:
249         free_page((unsigned long)vcpu->run);
250 fail:
251         return r;
252 }
253 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
254
255 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
256 {
257         put_pid(vcpu->pid);
258         kvm_arch_vcpu_uninit(vcpu);
259         free_page((unsigned long)vcpu->run);
260 }
261 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
262
263 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
264 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
265 {
266         return container_of(mn, struct kvm, mmu_notifier);
267 }
268
269 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
270                                              struct mm_struct *mm,
271                                              unsigned long address)
272 {
273         struct kvm *kvm = mmu_notifier_to_kvm(mn);
274         int need_tlb_flush, idx;
275
276         /*
277          * When ->invalidate_page runs, the linux pte has been zapped
278          * already but the page is still allocated until
279          * ->invalidate_page returns. So if we increase the sequence
280          * here the kvm page fault will notice if the spte can't be
281          * established because the page is going to be freed. If
282          * instead the kvm page fault establishes the spte before
283          * ->invalidate_page runs, kvm_unmap_hva will release it
284          * before returning.
285          *
286          * The sequence increase only need to be seen at spin_unlock
287          * time, and not at spin_lock time.
288          *
289          * Increasing the sequence after the spin_unlock would be
290          * unsafe because the kvm page fault could then establish the
291          * pte after kvm_unmap_hva returned, without noticing the page
292          * is going to be freed.
293          */
294         idx = srcu_read_lock(&kvm->srcu);
295         spin_lock(&kvm->mmu_lock);
296
297         kvm->mmu_notifier_seq++;
298         need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
299         /* we've to flush the tlb before the pages can be freed */
300         if (need_tlb_flush)
301                 kvm_flush_remote_tlbs(kvm);
302
303         spin_unlock(&kvm->mmu_lock);
304         srcu_read_unlock(&kvm->srcu, idx);
305 }
306
307 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
308                                         struct mm_struct *mm,
309                                         unsigned long address,
310                                         pte_t pte)
311 {
312         struct kvm *kvm = mmu_notifier_to_kvm(mn);
313         int idx;
314
315         idx = srcu_read_lock(&kvm->srcu);
316         spin_lock(&kvm->mmu_lock);
317         kvm->mmu_notifier_seq++;
318         kvm_set_spte_hva(kvm, address, pte);
319         spin_unlock(&kvm->mmu_lock);
320         srcu_read_unlock(&kvm->srcu, idx);
321 }
322
323 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
324                                                     struct mm_struct *mm,
325                                                     unsigned long start,
326                                                     unsigned long end)
327 {
328         struct kvm *kvm = mmu_notifier_to_kvm(mn);
329         int need_tlb_flush = 0, idx;
330
331         idx = srcu_read_lock(&kvm->srcu);
332         spin_lock(&kvm->mmu_lock);
333         /*
334          * The count increase must become visible at unlock time as no
335          * spte can be established without taking the mmu_lock and
336          * count is also read inside the mmu_lock critical section.
337          */
338         kvm->mmu_notifier_count++;
339         need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
340         need_tlb_flush |= kvm->tlbs_dirty;
341         /* we've to flush the tlb before the pages can be freed */
342         if (need_tlb_flush)
343                 kvm_flush_remote_tlbs(kvm);
344
345         spin_unlock(&kvm->mmu_lock);
346         srcu_read_unlock(&kvm->srcu, idx);
347 }
348
349 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
350                                                   struct mm_struct *mm,
351                                                   unsigned long start,
352                                                   unsigned long end)
353 {
354         struct kvm *kvm = mmu_notifier_to_kvm(mn);
355
356         spin_lock(&kvm->mmu_lock);
357         /*
358          * This sequence increase will notify the kvm page fault that
359          * the page that is going to be mapped in the spte could have
360          * been freed.
361          */
362         kvm->mmu_notifier_seq++;
363         smp_wmb();
364         /*
365          * The above sequence increase must be visible before the
366          * below count decrease, which is ensured by the smp_wmb above
367          * in conjunction with the smp_rmb in mmu_notifier_retry().
368          */
369         kvm->mmu_notifier_count--;
370         spin_unlock(&kvm->mmu_lock);
371
372         BUG_ON(kvm->mmu_notifier_count < 0);
373 }
374
375 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
376                                               struct mm_struct *mm,
377                                               unsigned long address)
378 {
379         struct kvm *kvm = mmu_notifier_to_kvm(mn);
380         int young, idx;
381
382         idx = srcu_read_lock(&kvm->srcu);
383         spin_lock(&kvm->mmu_lock);
384
385         young = kvm_age_hva(kvm, address);
386         if (young)
387                 kvm_flush_remote_tlbs(kvm);
388
389         spin_unlock(&kvm->mmu_lock);
390         srcu_read_unlock(&kvm->srcu, idx);
391
392         return young;
393 }
394
395 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
396                                        struct mm_struct *mm,
397                                        unsigned long address)
398 {
399         struct kvm *kvm = mmu_notifier_to_kvm(mn);
400         int young, idx;
401
402         idx = srcu_read_lock(&kvm->srcu);
403         spin_lock(&kvm->mmu_lock);
404         young = kvm_test_age_hva(kvm, address);
405         spin_unlock(&kvm->mmu_lock);
406         srcu_read_unlock(&kvm->srcu, idx);
407
408         return young;
409 }
410
411 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
412                                      struct mm_struct *mm)
413 {
414         struct kvm *kvm = mmu_notifier_to_kvm(mn);
415         int idx;
416
417         idx = srcu_read_lock(&kvm->srcu);
418         kvm_arch_flush_shadow_all(kvm);
419         srcu_read_unlock(&kvm->srcu, idx);
420 }
421
422 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
423         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
424         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
425         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
426         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
427         .test_young             = kvm_mmu_notifier_test_young,
428         .change_pte             = kvm_mmu_notifier_change_pte,
429         .release                = kvm_mmu_notifier_release,
430 };
431
432 static int kvm_init_mmu_notifier(struct kvm *kvm)
433 {
434         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
435         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
436 }
437
438 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
439
440 static int kvm_init_mmu_notifier(struct kvm *kvm)
441 {
442         return 0;
443 }
444
445 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
446
447 static void kvm_init_memslots_id(struct kvm *kvm)
448 {
449         int i;
450         struct kvm_memslots *slots = kvm->memslots;
451
452         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
453                 slots->id_to_index[i] = slots->memslots[i].id = i;
454 }
455
456 static struct kvm *kvm_create_vm(unsigned long type)
457 {
458         int r, i;
459         struct kvm *kvm = kvm_arch_alloc_vm();
460
461         if (!kvm)
462                 return ERR_PTR(-ENOMEM);
463
464         r = kvm_arch_init_vm(kvm, type);
465         if (r)
466                 goto out_err_nodisable;
467
468         r = hardware_enable_all();
469         if (r)
470                 goto out_err_nodisable;
471
472 #ifdef CONFIG_HAVE_KVM_IRQCHIP
473         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
474         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
475 #endif
476
477         BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
478
479         r = -ENOMEM;
480         kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
481         if (!kvm->memslots)
482                 goto out_err_nosrcu;
483         kvm_init_memslots_id(kvm);
484         if (init_srcu_struct(&kvm->srcu))
485                 goto out_err_nosrcu;
486         for (i = 0; i < KVM_NR_BUSES; i++) {
487                 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
488                                         GFP_KERNEL);
489                 if (!kvm->buses[i])
490                         goto out_err;
491         }
492
493         spin_lock_init(&kvm->mmu_lock);
494         kvm->mm = current->mm;
495         atomic_inc(&kvm->mm->mm_count);
496         kvm_eventfd_init(kvm);
497         mutex_init(&kvm->lock);
498         mutex_init(&kvm->irq_lock);
499         mutex_init(&kvm->slots_lock);
500         atomic_set(&kvm->users_count, 1);
501
502         r = kvm_init_mmu_notifier(kvm);
503         if (r)
504                 goto out_err;
505
506         raw_spin_lock(&kvm_lock);
507         list_add(&kvm->vm_list, &vm_list);
508         raw_spin_unlock(&kvm_lock);
509
510         return kvm;
511
512 out_err:
513         cleanup_srcu_struct(&kvm->srcu);
514 out_err_nosrcu:
515         hardware_disable_all();
516 out_err_nodisable:
517         for (i = 0; i < KVM_NR_BUSES; i++)
518                 kfree(kvm->buses[i]);
519         kfree(kvm->memslots);
520         kvm_arch_free_vm(kvm);
521         return ERR_PTR(r);
522 }
523
524 /*
525  * Avoid using vmalloc for a small buffer.
526  * Should not be used when the size is statically known.
527  */
528 void *kvm_kvzalloc(unsigned long size)
529 {
530         if (size > PAGE_SIZE)
531                 return vzalloc(size);
532         else
533                 return kzalloc(size, GFP_KERNEL);
534 }
535
536 void kvm_kvfree(const void *addr)
537 {
538         if (is_vmalloc_addr(addr))
539                 vfree(addr);
540         else
541                 kfree(addr);
542 }
543
544 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
545 {
546         if (!memslot->dirty_bitmap)
547                 return;
548
549         kvm_kvfree(memslot->dirty_bitmap);
550         memslot->dirty_bitmap = NULL;
551 }
552
553 /*
554  * Free any memory in @free but not in @dont.
555  */
556 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
557                                   struct kvm_memory_slot *dont)
558 {
559         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
560                 kvm_destroy_dirty_bitmap(free);
561
562         kvm_arch_free_memslot(free, dont);
563
564         free->npages = 0;
565 }
566
567 void kvm_free_physmem(struct kvm *kvm)
568 {
569         struct kvm_memslots *slots = kvm->memslots;
570         struct kvm_memory_slot *memslot;
571
572         kvm_for_each_memslot(memslot, slots)
573                 kvm_free_physmem_slot(memslot, NULL);
574
575         kfree(kvm->memslots);
576 }
577
578 static void kvm_destroy_vm(struct kvm *kvm)
579 {
580         int i;
581         struct mm_struct *mm = kvm->mm;
582
583         kvm_arch_sync_events(kvm);
584         raw_spin_lock(&kvm_lock);
585         list_del(&kvm->vm_list);
586         raw_spin_unlock(&kvm_lock);
587         kvm_free_irq_routing(kvm);
588         for (i = 0; i < KVM_NR_BUSES; i++)
589                 kvm_io_bus_destroy(kvm->buses[i]);
590         kvm_coalesced_mmio_free(kvm);
591 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
592         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
593 #else
594         kvm_arch_flush_shadow_all(kvm);
595 #endif
596         kvm_arch_destroy_vm(kvm);
597         kvm_free_physmem(kvm);
598         cleanup_srcu_struct(&kvm->srcu);
599         kvm_arch_free_vm(kvm);
600         hardware_disable_all();
601         mmdrop(mm);
602 }
603
604 void kvm_get_kvm(struct kvm *kvm)
605 {
606         atomic_inc(&kvm->users_count);
607 }
608 EXPORT_SYMBOL_GPL(kvm_get_kvm);
609
610 void kvm_put_kvm(struct kvm *kvm)
611 {
612         if (atomic_dec_and_test(&kvm->users_count))
613                 kvm_destroy_vm(kvm);
614 }
615 EXPORT_SYMBOL_GPL(kvm_put_kvm);
616
617
618 static int kvm_vm_release(struct inode *inode, struct file *filp)
619 {
620         struct kvm *kvm = filp->private_data;
621
622         kvm_irqfd_release(kvm);
623
624         kvm_put_kvm(kvm);
625         return 0;
626 }
627
628 /*
629  * Allocation size is twice as large as the actual dirty bitmap size.
630  * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
631  */
632 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
633 {
634 #ifndef CONFIG_S390
635         unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
636
637         memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
638         if (!memslot->dirty_bitmap)
639                 return -ENOMEM;
640
641 #endif /* !CONFIG_S390 */
642         return 0;
643 }
644
645 static int cmp_memslot(const void *slot1, const void *slot2)
646 {
647         struct kvm_memory_slot *s1, *s2;
648
649         s1 = (struct kvm_memory_slot *)slot1;
650         s2 = (struct kvm_memory_slot *)slot2;
651
652         if (s1->npages < s2->npages)
653                 return 1;
654         if (s1->npages > s2->npages)
655                 return -1;
656
657         return 0;
658 }
659
660 /*
661  * Sort the memslots base on its size, so the larger slots
662  * will get better fit.
663  */
664 static void sort_memslots(struct kvm_memslots *slots)
665 {
666         int i;
667
668         sort(slots->memslots, KVM_MEM_SLOTS_NUM,
669               sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
670
671         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
672                 slots->id_to_index[slots->memslots[i].id] = i;
673 }
674
675 void update_memslots(struct kvm_memslots *slots, struct kvm_memory_slot *new,
676                      u64 last_generation)
677 {
678         if (new) {
679                 int id = new->id;
680                 struct kvm_memory_slot *old = id_to_memslot(slots, id);
681                 unsigned long npages = old->npages;
682
683                 *old = *new;
684                 if (new->npages != npages)
685                         sort_memslots(slots);
686         }
687
688         slots->generation = last_generation + 1;
689 }
690
691 static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
692 {
693         u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
694
695 #ifdef KVM_CAP_READONLY_MEM
696         valid_flags |= KVM_MEM_READONLY;
697 #endif
698
699         if (mem->flags & ~valid_flags)
700                 return -EINVAL;
701
702         return 0;
703 }
704
705 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
706                 struct kvm_memslots *slots, struct kvm_memory_slot *new)
707 {
708         struct kvm_memslots *old_memslots = kvm->memslots;
709
710         update_memslots(slots, new, kvm->memslots->generation);
711         rcu_assign_pointer(kvm->memslots, slots);
712         synchronize_srcu_expedited(&kvm->srcu);
713         return old_memslots; 
714 }
715
716 /*
717  * Allocate some memory and give it an address in the guest physical address
718  * space.
719  *
720  * Discontiguous memory is allowed, mostly for framebuffers.
721  *
722  * Must be called holding mmap_sem for write.
723  */
724 int __kvm_set_memory_region(struct kvm *kvm,
725                             struct kvm_userspace_memory_region *mem,
726                             bool user_alloc)
727 {
728         int r;
729         gfn_t base_gfn;
730         unsigned long npages;
731         struct kvm_memory_slot *slot;
732         struct kvm_memory_slot old, new;
733         struct kvm_memslots *slots = NULL, *old_memslots;
734
735         r = check_memory_region_flags(mem);
736         if (r)
737                 goto out;
738
739         r = -EINVAL;
740         /* General sanity checks */
741         if (mem->memory_size & (PAGE_SIZE - 1))
742                 goto out;
743         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
744                 goto out;
745         /* We can read the guest memory with __xxx_user() later on. */
746         if (user_alloc &&
747             ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
748              !access_ok(VERIFY_WRITE,
749                         (void __user *)(unsigned long)mem->userspace_addr,
750                         mem->memory_size)))
751                 goto out;
752         if (mem->slot >= KVM_MEM_SLOTS_NUM)
753                 goto out;
754         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
755                 goto out;
756
757         slot = id_to_memslot(kvm->memslots, mem->slot);
758         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
759         npages = mem->memory_size >> PAGE_SHIFT;
760
761         r = -EINVAL;
762         if (npages > KVM_MEM_MAX_NR_PAGES)
763                 goto out;
764
765         if (!npages)
766                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
767
768         new = old = *slot;
769
770         new.id = mem->slot;
771         new.base_gfn = base_gfn;
772         new.npages = npages;
773         new.flags = mem->flags;
774
775         /*
776          * Disallow changing a memory slot's size or changing anything about
777          * zero sized slots that doesn't involve making them non-zero.
778          */
779         r = -EINVAL;
780         if (npages && old.npages && npages != old.npages)
781                 goto out;
782         if (!npages && !old.npages)
783                 goto out;
784
785         if ((npages && !old.npages) || (base_gfn != old.base_gfn)) {
786                 /* Check for overlaps */
787                 r = -EEXIST;
788                 kvm_for_each_memslot(slot, kvm->memslots) {
789                         if ((slot->id >= KVM_USER_MEM_SLOTS) ||
790                             (slot->id == mem->slot))
791                                 continue;
792                         if (!((base_gfn + npages <= slot->base_gfn) ||
793                               (base_gfn >= slot->base_gfn + slot->npages)))
794                                 goto out;
795                 }
796         }
797
798         /* Free page dirty bitmap if unneeded */
799         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
800                 new.dirty_bitmap = NULL;
801
802         r = -ENOMEM;
803
804         /*
805          * Allocate if a slot is being created.  If modifying a slot,
806          * the userspace_addr cannot change.
807          */
808         if (!old.npages) {
809                 new.user_alloc = user_alloc;
810                 new.userspace_addr = mem->userspace_addr;
811
812                 if (kvm_arch_create_memslot(&new, npages))
813                         goto out_free;
814         } else if (npages && mem->userspace_addr != old.userspace_addr) {
815                 r = -EINVAL;
816                 goto out_free;
817         }
818
819         /* Allocate page dirty bitmap if needed */
820         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
821                 if (kvm_create_dirty_bitmap(&new) < 0)
822                         goto out_free;
823         }
824
825         if (!npages || base_gfn != old.base_gfn) {
826                 r = -ENOMEM;
827                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
828                                 GFP_KERNEL);
829                 if (!slots)
830                         goto out_free;
831                 slot = id_to_memslot(slots, mem->slot);
832                 slot->flags |= KVM_MEMSLOT_INVALID;
833
834                 old_memslots = install_new_memslots(kvm, slots, NULL);
835
836                 /* slot was deleted or moved, clear iommu mapping */
837                 kvm_iommu_unmap_pages(kvm, &old);
838                 /* From this point no new shadow pages pointing to a deleted,
839                  * or moved, memslot will be created.
840                  *
841                  * validation of sp->gfn happens in:
842                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
843                  *      - kvm_is_visible_gfn (mmu_check_roots)
844                  */
845                 kvm_arch_flush_shadow_memslot(kvm, slot);
846                 slots = old_memslots;
847         }
848
849         r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
850         if (r)
851                 goto out_slots;
852
853         r = -ENOMEM;
854         /*
855          * We can re-use the old_memslots from above, the only difference
856          * from the currently installed memslots is the invalid flag.  This
857          * will get overwritten by update_memslots anyway.
858          */
859         if (!slots) {
860                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
861                                 GFP_KERNEL);
862                 if (!slots)
863                         goto out_free;
864         }
865
866         /* map new memory slot into the iommu */
867         if (npages) {
868                 r = kvm_iommu_map_pages(kvm, &new);
869                 if (r)
870                         goto out_slots;
871         }
872
873         /* actual memory is freed via old in kvm_free_physmem_slot below */
874         if (!npages) {
875                 new.dirty_bitmap = NULL;
876                 memset(&new.arch, 0, sizeof(new.arch));
877         }
878
879         old_memslots = install_new_memslots(kvm, slots, &new);
880
881         kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
882
883         kvm_free_physmem_slot(&old, &new);
884         kfree(old_memslots);
885
886         return 0;
887
888 out_slots:
889         kfree(slots);
890 out_free:
891         kvm_free_physmem_slot(&new, &old);
892 out:
893         return r;
894 }
895 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
896
897 int kvm_set_memory_region(struct kvm *kvm,
898                           struct kvm_userspace_memory_region *mem,
899                           bool user_alloc)
900 {
901         int r;
902
903         mutex_lock(&kvm->slots_lock);
904         r = __kvm_set_memory_region(kvm, mem, user_alloc);
905         mutex_unlock(&kvm->slots_lock);
906         return r;
907 }
908 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
909
910 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
911                                    struct
912                                    kvm_userspace_memory_region *mem,
913                                    bool user_alloc)
914 {
915         if (mem->slot >= KVM_USER_MEM_SLOTS)
916                 return -EINVAL;
917         return kvm_set_memory_region(kvm, mem, user_alloc);
918 }
919
920 int kvm_get_dirty_log(struct kvm *kvm,
921                         struct kvm_dirty_log *log, int *is_dirty)
922 {
923         struct kvm_memory_slot *memslot;
924         int r, i;
925         unsigned long n;
926         unsigned long any = 0;
927
928         r = -EINVAL;
929         if (log->slot >= KVM_USER_MEM_SLOTS)
930                 goto out;
931
932         memslot = id_to_memslot(kvm->memslots, log->slot);
933         r = -ENOENT;
934         if (!memslot->dirty_bitmap)
935                 goto out;
936
937         n = kvm_dirty_bitmap_bytes(memslot);
938
939         for (i = 0; !any && i < n/sizeof(long); ++i)
940                 any = memslot->dirty_bitmap[i];
941
942         r = -EFAULT;
943         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
944                 goto out;
945
946         if (any)
947                 *is_dirty = 1;
948
949         r = 0;
950 out:
951         return r;
952 }
953
954 bool kvm_largepages_enabled(void)
955 {
956         return largepages_enabled;
957 }
958
959 void kvm_disable_largepages(void)
960 {
961         largepages_enabled = false;
962 }
963 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
964
965 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
966 {
967         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
968 }
969 EXPORT_SYMBOL_GPL(gfn_to_memslot);
970
971 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
972 {
973         struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
974
975         if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
976               memslot->flags & KVM_MEMSLOT_INVALID)
977                 return 0;
978
979         return 1;
980 }
981 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
982
983 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
984 {
985         struct vm_area_struct *vma;
986         unsigned long addr, size;
987
988         size = PAGE_SIZE;
989
990         addr = gfn_to_hva(kvm, gfn);
991         if (kvm_is_error_hva(addr))
992                 return PAGE_SIZE;
993
994         down_read(&current->mm->mmap_sem);
995         vma = find_vma(current->mm, addr);
996         if (!vma)
997                 goto out;
998
999         size = vma_kernel_pagesize(vma);
1000
1001 out:
1002         up_read(&current->mm->mmap_sem);
1003
1004         return size;
1005 }
1006
1007 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1008 {
1009         return slot->flags & KVM_MEM_READONLY;
1010 }
1011
1012 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1013                                        gfn_t *nr_pages, bool write)
1014 {
1015         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1016                 return KVM_HVA_ERR_BAD;
1017
1018         if (memslot_is_readonly(slot) && write)
1019                 return KVM_HVA_ERR_RO_BAD;
1020
1021         if (nr_pages)
1022                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1023
1024         return __gfn_to_hva_memslot(slot, gfn);
1025 }
1026
1027 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1028                                      gfn_t *nr_pages)
1029 {
1030         return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1031 }
1032
1033 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1034                                  gfn_t gfn)
1035 {
1036         return gfn_to_hva_many(slot, gfn, NULL);
1037 }
1038 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1039
1040 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1041 {
1042         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1043 }
1044 EXPORT_SYMBOL_GPL(gfn_to_hva);
1045
1046 /*
1047  * The hva returned by this function is only allowed to be read.
1048  * It should pair with kvm_read_hva() or kvm_read_hva_atomic().
1049  */
1050 static unsigned long gfn_to_hva_read(struct kvm *kvm, gfn_t gfn)
1051 {
1052         return __gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL, false);
1053 }
1054
1055 static int kvm_read_hva(void *data, void __user *hva, int len)
1056 {
1057         return __copy_from_user(data, hva, len);
1058 }
1059
1060 static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1061 {
1062         return __copy_from_user_inatomic(data, hva, len);
1063 }
1064
1065 int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1066         unsigned long start, int write, struct page **page)
1067 {
1068         int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1069
1070         if (write)
1071                 flags |= FOLL_WRITE;
1072
1073         return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1074 }
1075
1076 static inline int check_user_page_hwpoison(unsigned long addr)
1077 {
1078         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1079
1080         rc = __get_user_pages(current, current->mm, addr, 1,
1081                               flags, NULL, NULL, NULL);
1082         return rc == -EHWPOISON;
1083 }
1084
1085 /*
1086  * The atomic path to get the writable pfn which will be stored in @pfn,
1087  * true indicates success, otherwise false is returned.
1088  */
1089 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1090                             bool write_fault, bool *writable, pfn_t *pfn)
1091 {
1092         struct page *page[1];
1093         int npages;
1094
1095         if (!(async || atomic))
1096                 return false;
1097
1098         /*
1099          * Fast pin a writable pfn only if it is a write fault request
1100          * or the caller allows to map a writable pfn for a read fault
1101          * request.
1102          */
1103         if (!(write_fault || writable))
1104                 return false;
1105
1106         npages = __get_user_pages_fast(addr, 1, 1, page);
1107         if (npages == 1) {
1108                 *pfn = page_to_pfn(page[0]);
1109
1110                 if (writable)
1111                         *writable = true;
1112                 return true;
1113         }
1114
1115         return false;
1116 }
1117
1118 /*
1119  * The slow path to get the pfn of the specified host virtual address,
1120  * 1 indicates success, -errno is returned if error is detected.
1121  */
1122 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1123                            bool *writable, pfn_t *pfn)
1124 {
1125         struct page *page[1];
1126         int npages = 0;
1127
1128         might_sleep();
1129
1130         if (writable)
1131                 *writable = write_fault;
1132
1133         if (async) {
1134                 down_read(&current->mm->mmap_sem);
1135                 npages = get_user_page_nowait(current, current->mm,
1136                                               addr, write_fault, page);
1137                 up_read(&current->mm->mmap_sem);
1138         } else
1139                 npages = get_user_pages_fast(addr, 1, write_fault,
1140                                              page);
1141         if (npages != 1)
1142                 return npages;
1143
1144         /* map read fault as writable if possible */
1145         if (unlikely(!write_fault) && writable) {
1146                 struct page *wpage[1];
1147
1148                 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1149                 if (npages == 1) {
1150                         *writable = true;
1151                         put_page(page[0]);
1152                         page[0] = wpage[0];
1153                 }
1154
1155                 npages = 1;
1156         }
1157         *pfn = page_to_pfn(page[0]);
1158         return npages;
1159 }
1160
1161 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1162 {
1163         if (unlikely(!(vma->vm_flags & VM_READ)))
1164                 return false;
1165
1166         if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1167                 return false;
1168
1169         return true;
1170 }
1171
1172 /*
1173  * Pin guest page in memory and return its pfn.
1174  * @addr: host virtual address which maps memory to the guest
1175  * @atomic: whether this function can sleep
1176  * @async: whether this function need to wait IO complete if the
1177  *         host page is not in the memory
1178  * @write_fault: whether we should get a writable host page
1179  * @writable: whether it allows to map a writable host page for !@write_fault
1180  *
1181  * The function will map a writable host page for these two cases:
1182  * 1): @write_fault = true
1183  * 2): @write_fault = false && @writable, @writable will tell the caller
1184  *     whether the mapping is writable.
1185  */
1186 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1187                         bool write_fault, bool *writable)
1188 {
1189         struct vm_area_struct *vma;
1190         pfn_t pfn = 0;
1191         int npages;
1192
1193         /* we can do it either atomically or asynchronously, not both */
1194         BUG_ON(atomic && async);
1195
1196         if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1197                 return pfn;
1198
1199         if (atomic)
1200                 return KVM_PFN_ERR_FAULT;
1201
1202         npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1203         if (npages == 1)
1204                 return pfn;
1205
1206         down_read(&current->mm->mmap_sem);
1207         if (npages == -EHWPOISON ||
1208               (!async && check_user_page_hwpoison(addr))) {
1209                 pfn = KVM_PFN_ERR_HWPOISON;
1210                 goto exit;
1211         }
1212
1213         vma = find_vma_intersection(current->mm, addr, addr + 1);
1214
1215         if (vma == NULL)
1216                 pfn = KVM_PFN_ERR_FAULT;
1217         else if ((vma->vm_flags & VM_PFNMAP)) {
1218                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1219                         vma->vm_pgoff;
1220                 BUG_ON(!kvm_is_mmio_pfn(pfn));
1221         } else {
1222                 if (async && vma_is_valid(vma, write_fault))
1223                         *async = true;
1224                 pfn = KVM_PFN_ERR_FAULT;
1225         }
1226 exit:
1227         up_read(&current->mm->mmap_sem);
1228         return pfn;
1229 }
1230
1231 static pfn_t
1232 __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1233                      bool *async, bool write_fault, bool *writable)
1234 {
1235         unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1236
1237         if (addr == KVM_HVA_ERR_RO_BAD)
1238                 return KVM_PFN_ERR_RO_FAULT;
1239
1240         if (kvm_is_error_hva(addr))
1241                 return KVM_PFN_NOSLOT;
1242
1243         /* Do not map writable pfn in the readonly memslot. */
1244         if (writable && memslot_is_readonly(slot)) {
1245                 *writable = false;
1246                 writable = NULL;
1247         }
1248
1249         return hva_to_pfn(addr, atomic, async, write_fault,
1250                           writable);
1251 }
1252
1253 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1254                           bool write_fault, bool *writable)
1255 {
1256         struct kvm_memory_slot *slot;
1257
1258         if (async)
1259                 *async = false;
1260
1261         slot = gfn_to_memslot(kvm, gfn);
1262
1263         return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1264                                     writable);
1265 }
1266
1267 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1268 {
1269         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1270 }
1271 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1272
1273 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1274                        bool write_fault, bool *writable)
1275 {
1276         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1277 }
1278 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1279
1280 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1281 {
1282         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1283 }
1284 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1285
1286 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1287                       bool *writable)
1288 {
1289         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1290 }
1291 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1292
1293 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1294 {
1295         return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1296 }
1297
1298 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1299 {
1300         return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1301 }
1302 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1303
1304 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1305                                                                   int nr_pages)
1306 {
1307         unsigned long addr;
1308         gfn_t entry;
1309
1310         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1311         if (kvm_is_error_hva(addr))
1312                 return -1;
1313
1314         if (entry < nr_pages)
1315                 return 0;
1316
1317         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1318 }
1319 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1320
1321 static struct page *kvm_pfn_to_page(pfn_t pfn)
1322 {
1323         if (is_error_noslot_pfn(pfn))
1324                 return KVM_ERR_PTR_BAD_PAGE;
1325
1326         if (kvm_is_mmio_pfn(pfn)) {
1327                 WARN_ON(1);
1328                 return KVM_ERR_PTR_BAD_PAGE;
1329         }
1330
1331         return pfn_to_page(pfn);
1332 }
1333
1334 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1335 {
1336         pfn_t pfn;
1337
1338         pfn = gfn_to_pfn(kvm, gfn);
1339
1340         return kvm_pfn_to_page(pfn);
1341 }
1342
1343 EXPORT_SYMBOL_GPL(gfn_to_page);
1344
1345 void kvm_release_page_clean(struct page *page)
1346 {
1347         WARN_ON(is_error_page(page));
1348
1349         kvm_release_pfn_clean(page_to_pfn(page));
1350 }
1351 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1352
1353 void kvm_release_pfn_clean(pfn_t pfn)
1354 {
1355         if (!is_error_noslot_pfn(pfn) && !kvm_is_mmio_pfn(pfn))
1356                 put_page(pfn_to_page(pfn));
1357 }
1358 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1359
1360 void kvm_release_page_dirty(struct page *page)
1361 {
1362         WARN_ON(is_error_page(page));
1363
1364         kvm_release_pfn_dirty(page_to_pfn(page));
1365 }
1366 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1367
1368 void kvm_release_pfn_dirty(pfn_t pfn)
1369 {
1370         kvm_set_pfn_dirty(pfn);
1371         kvm_release_pfn_clean(pfn);
1372 }
1373 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1374
1375 void kvm_set_page_dirty(struct page *page)
1376 {
1377         kvm_set_pfn_dirty(page_to_pfn(page));
1378 }
1379 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1380
1381 void kvm_set_pfn_dirty(pfn_t pfn)
1382 {
1383         if (!kvm_is_mmio_pfn(pfn)) {
1384                 struct page *page = pfn_to_page(pfn);
1385                 if (!PageReserved(page))
1386                         SetPageDirty(page);
1387         }
1388 }
1389 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1390
1391 void kvm_set_pfn_accessed(pfn_t pfn)
1392 {
1393         if (!kvm_is_mmio_pfn(pfn))
1394                 mark_page_accessed(pfn_to_page(pfn));
1395 }
1396 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1397
1398 void kvm_get_pfn(pfn_t pfn)
1399 {
1400         if (!kvm_is_mmio_pfn(pfn))
1401                 get_page(pfn_to_page(pfn));
1402 }
1403 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1404
1405 static int next_segment(unsigned long len, int offset)
1406 {
1407         if (len > PAGE_SIZE - offset)
1408                 return PAGE_SIZE - offset;
1409         else
1410                 return len;
1411 }
1412
1413 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1414                         int len)
1415 {
1416         int r;
1417         unsigned long addr;
1418
1419         addr = gfn_to_hva_read(kvm, gfn);
1420         if (kvm_is_error_hva(addr))
1421                 return -EFAULT;
1422         r = kvm_read_hva(data, (void __user *)addr + offset, len);
1423         if (r)
1424                 return -EFAULT;
1425         return 0;
1426 }
1427 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1428
1429 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1430 {
1431         gfn_t gfn = gpa >> PAGE_SHIFT;
1432         int seg;
1433         int offset = offset_in_page(gpa);
1434         int ret;
1435
1436         while ((seg = next_segment(len, offset)) != 0) {
1437                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1438                 if (ret < 0)
1439                         return ret;
1440                 offset = 0;
1441                 len -= seg;
1442                 data += seg;
1443                 ++gfn;
1444         }
1445         return 0;
1446 }
1447 EXPORT_SYMBOL_GPL(kvm_read_guest);
1448
1449 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1450                           unsigned long len)
1451 {
1452         int r;
1453         unsigned long addr;
1454         gfn_t gfn = gpa >> PAGE_SHIFT;
1455         int offset = offset_in_page(gpa);
1456
1457         addr = gfn_to_hva_read(kvm, gfn);
1458         if (kvm_is_error_hva(addr))
1459                 return -EFAULT;
1460         pagefault_disable();
1461         r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
1462         pagefault_enable();
1463         if (r)
1464                 return -EFAULT;
1465         return 0;
1466 }
1467 EXPORT_SYMBOL(kvm_read_guest_atomic);
1468
1469 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1470                          int offset, int len)
1471 {
1472         int r;
1473         unsigned long addr;
1474
1475         addr = gfn_to_hva(kvm, gfn);
1476         if (kvm_is_error_hva(addr))
1477                 return -EFAULT;
1478         r = __copy_to_user((void __user *)addr + offset, data, len);
1479         if (r)
1480                 return -EFAULT;
1481         mark_page_dirty(kvm, gfn);
1482         return 0;
1483 }
1484 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1485
1486 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1487                     unsigned long len)
1488 {
1489         gfn_t gfn = gpa >> PAGE_SHIFT;
1490         int seg;
1491         int offset = offset_in_page(gpa);
1492         int ret;
1493
1494         while ((seg = next_segment(len, offset)) != 0) {
1495                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1496                 if (ret < 0)
1497                         return ret;
1498                 offset = 0;
1499                 len -= seg;
1500                 data += seg;
1501                 ++gfn;
1502         }
1503         return 0;
1504 }
1505
1506 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1507                               gpa_t gpa)
1508 {
1509         struct kvm_memslots *slots = kvm_memslots(kvm);
1510         int offset = offset_in_page(gpa);
1511         gfn_t gfn = gpa >> PAGE_SHIFT;
1512
1513         ghc->gpa = gpa;
1514         ghc->generation = slots->generation;
1515         ghc->memslot = gfn_to_memslot(kvm, gfn);
1516         ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1517         if (!kvm_is_error_hva(ghc->hva))
1518                 ghc->hva += offset;
1519         else
1520                 return -EFAULT;
1521
1522         return 0;
1523 }
1524 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1525
1526 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1527                            void *data, unsigned long len)
1528 {
1529         struct kvm_memslots *slots = kvm_memslots(kvm);
1530         int r;
1531
1532         if (slots->generation != ghc->generation)
1533                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1534
1535         if (kvm_is_error_hva(ghc->hva))
1536                 return -EFAULT;
1537
1538         r = __copy_to_user((void __user *)ghc->hva, data, len);
1539         if (r)
1540                 return -EFAULT;
1541         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1542
1543         return 0;
1544 }
1545 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1546
1547 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1548                            void *data, unsigned long len)
1549 {
1550         struct kvm_memslots *slots = kvm_memslots(kvm);
1551         int r;
1552
1553         if (slots->generation != ghc->generation)
1554                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1555
1556         if (kvm_is_error_hva(ghc->hva))
1557                 return -EFAULT;
1558
1559         r = __copy_from_user(data, (void __user *)ghc->hva, len);
1560         if (r)
1561                 return -EFAULT;
1562
1563         return 0;
1564 }
1565 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1566
1567 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1568 {
1569         return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1570                                     offset, len);
1571 }
1572 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1573
1574 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1575 {
1576         gfn_t gfn = gpa >> PAGE_SHIFT;
1577         int seg;
1578         int offset = offset_in_page(gpa);
1579         int ret;
1580
1581         while ((seg = next_segment(len, offset)) != 0) {
1582                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1583                 if (ret < 0)
1584                         return ret;
1585                 offset = 0;
1586                 len -= seg;
1587                 ++gfn;
1588         }
1589         return 0;
1590 }
1591 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1592
1593 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1594                              gfn_t gfn)
1595 {
1596         if (memslot && memslot->dirty_bitmap) {
1597                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1598
1599                 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1600         }
1601 }
1602
1603 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1604 {
1605         struct kvm_memory_slot *memslot;
1606
1607         memslot = gfn_to_memslot(kvm, gfn);
1608         mark_page_dirty_in_slot(kvm, memslot, gfn);
1609 }
1610
1611 /*
1612  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1613  */
1614 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1615 {
1616         DEFINE_WAIT(wait);
1617
1618         for (;;) {
1619                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1620
1621                 if (kvm_arch_vcpu_runnable(vcpu)) {
1622                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1623                         break;
1624                 }
1625                 if (kvm_cpu_has_pending_timer(vcpu))
1626                         break;
1627                 if (signal_pending(current))
1628                         break;
1629
1630                 schedule();
1631         }
1632
1633         finish_wait(&vcpu->wq, &wait);
1634 }
1635
1636 #ifndef CONFIG_S390
1637 /*
1638  * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1639  */
1640 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1641 {
1642         int me;
1643         int cpu = vcpu->cpu;
1644         wait_queue_head_t *wqp;
1645
1646         wqp = kvm_arch_vcpu_wq(vcpu);
1647         if (waitqueue_active(wqp)) {
1648                 wake_up_interruptible(wqp);
1649                 ++vcpu->stat.halt_wakeup;
1650         }
1651
1652         me = get_cpu();
1653         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1654                 if (kvm_arch_vcpu_should_kick(vcpu))
1655                         smp_send_reschedule(cpu);
1656         put_cpu();
1657 }
1658 #endif /* !CONFIG_S390 */
1659
1660 void kvm_resched(struct kvm_vcpu *vcpu)
1661 {
1662         if (!need_resched())
1663                 return;
1664         cond_resched();
1665 }
1666 EXPORT_SYMBOL_GPL(kvm_resched);
1667
1668 bool kvm_vcpu_yield_to(struct kvm_vcpu *target)
1669 {
1670         struct pid *pid;
1671         struct task_struct *task = NULL;
1672
1673         rcu_read_lock();
1674         pid = rcu_dereference(target->pid);
1675         if (pid)
1676                 task = get_pid_task(target->pid, PIDTYPE_PID);
1677         rcu_read_unlock();
1678         if (!task)
1679                 return false;
1680         if (task->flags & PF_VCPU) {
1681                 put_task_struct(task);
1682                 return false;
1683         }
1684         if (yield_to(task, 1)) {
1685                 put_task_struct(task);
1686                 return true;
1687         }
1688         put_task_struct(task);
1689         return false;
1690 }
1691 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1692
1693 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1694 /*
1695  * Helper that checks whether a VCPU is eligible for directed yield.
1696  * Most eligible candidate to yield is decided by following heuristics:
1697  *
1698  *  (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1699  *  (preempted lock holder), indicated by @in_spin_loop.
1700  *  Set at the beiginning and cleared at the end of interception/PLE handler.
1701  *
1702  *  (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1703  *  chance last time (mostly it has become eligible now since we have probably
1704  *  yielded to lockholder in last iteration. This is done by toggling
1705  *  @dy_eligible each time a VCPU checked for eligibility.)
1706  *
1707  *  Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1708  *  to preempted lock-holder could result in wrong VCPU selection and CPU
1709  *  burning. Giving priority for a potential lock-holder increases lock
1710  *  progress.
1711  *
1712  *  Since algorithm is based on heuristics, accessing another VCPU data without
1713  *  locking does not harm. It may result in trying to yield to  same VCPU, fail
1714  *  and continue with next VCPU and so on.
1715  */
1716 bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1717 {
1718         bool eligible;
1719
1720         eligible = !vcpu->spin_loop.in_spin_loop ||
1721                         (vcpu->spin_loop.in_spin_loop &&
1722                          vcpu->spin_loop.dy_eligible);
1723
1724         if (vcpu->spin_loop.in_spin_loop)
1725                 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1726
1727         return eligible;
1728 }
1729 #endif
1730 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1731 {
1732         struct kvm *kvm = me->kvm;
1733         struct kvm_vcpu *vcpu;
1734         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1735         int yielded = 0;
1736         int pass;
1737         int i;
1738
1739         kvm_vcpu_set_in_spin_loop(me, true);
1740         /*
1741          * We boost the priority of a VCPU that is runnable but not
1742          * currently running, because it got preempted by something
1743          * else and called schedule in __vcpu_run.  Hopefully that
1744          * VCPU is holding the lock that we need and will release it.
1745          * We approximate round-robin by starting at the last boosted VCPU.
1746          */
1747         for (pass = 0; pass < 2 && !yielded; pass++) {
1748                 kvm_for_each_vcpu(i, vcpu, kvm) {
1749                         if (!pass && i <= last_boosted_vcpu) {
1750                                 i = last_boosted_vcpu;
1751                                 continue;
1752                         } else if (pass && i > last_boosted_vcpu)
1753                                 break;
1754                         if (vcpu == me)
1755                                 continue;
1756                         if (waitqueue_active(&vcpu->wq))
1757                                 continue;
1758                         if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1759                                 continue;
1760                         if (kvm_vcpu_yield_to(vcpu)) {
1761                                 kvm->last_boosted_vcpu = i;
1762                                 yielded = 1;
1763                                 break;
1764                         }
1765                 }
1766         }
1767         kvm_vcpu_set_in_spin_loop(me, false);
1768
1769         /* Ensure vcpu is not eligible during next spinloop */
1770         kvm_vcpu_set_dy_eligible(me, false);
1771 }
1772 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1773
1774 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1775 {
1776         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1777         struct page *page;
1778
1779         if (vmf->pgoff == 0)
1780                 page = virt_to_page(vcpu->run);
1781 #ifdef CONFIG_X86
1782         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1783                 page = virt_to_page(vcpu->arch.pio_data);
1784 #endif
1785 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1786         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1787                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1788 #endif
1789         else
1790                 return kvm_arch_vcpu_fault(vcpu, vmf);
1791         get_page(page);
1792         vmf->page = page;
1793         return 0;
1794 }
1795
1796 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1797         .fault = kvm_vcpu_fault,
1798 };
1799
1800 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1801 {
1802         vma->vm_ops = &kvm_vcpu_vm_ops;
1803         return 0;
1804 }
1805
1806 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1807 {
1808         struct kvm_vcpu *vcpu = filp->private_data;
1809
1810         kvm_put_kvm(vcpu->kvm);
1811         return 0;
1812 }
1813
1814 static struct file_operations kvm_vcpu_fops = {
1815         .release        = kvm_vcpu_release,
1816         .unlocked_ioctl = kvm_vcpu_ioctl,
1817 #ifdef CONFIG_COMPAT
1818         .compat_ioctl   = kvm_vcpu_compat_ioctl,
1819 #endif
1820         .mmap           = kvm_vcpu_mmap,
1821         .llseek         = noop_llseek,
1822 };
1823
1824 /*
1825  * Allocates an inode for the vcpu.
1826  */
1827 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1828 {
1829         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1830 }
1831
1832 /*
1833  * Creates some virtual cpus.  Good luck creating more than one.
1834  */
1835 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1836 {
1837         int r;
1838         struct kvm_vcpu *vcpu, *v;
1839
1840         vcpu = kvm_arch_vcpu_create(kvm, id);
1841         if (IS_ERR(vcpu))
1842                 return PTR_ERR(vcpu);
1843
1844         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1845
1846         r = kvm_arch_vcpu_setup(vcpu);
1847         if (r)
1848                 goto vcpu_destroy;
1849
1850         mutex_lock(&kvm->lock);
1851         if (!kvm_vcpu_compatible(vcpu)) {
1852                 r = -EINVAL;
1853                 goto unlock_vcpu_destroy;
1854         }
1855         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1856                 r = -EINVAL;
1857                 goto unlock_vcpu_destroy;
1858         }
1859
1860         kvm_for_each_vcpu(r, v, kvm)
1861                 if (v->vcpu_id == id) {
1862                         r = -EEXIST;
1863                         goto unlock_vcpu_destroy;
1864                 }
1865
1866         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1867
1868         /* Now it's all set up, let userspace reach it */
1869         kvm_get_kvm(kvm);
1870         r = create_vcpu_fd(vcpu);
1871         if (r < 0) {
1872                 kvm_put_kvm(kvm);
1873                 goto unlock_vcpu_destroy;
1874         }
1875
1876         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1877         smp_wmb();
1878         atomic_inc(&kvm->online_vcpus);
1879
1880         mutex_unlock(&kvm->lock);
1881         kvm_arch_vcpu_postcreate(vcpu);
1882         return r;
1883
1884 unlock_vcpu_destroy:
1885         mutex_unlock(&kvm->lock);
1886 vcpu_destroy:
1887         kvm_arch_vcpu_destroy(vcpu);
1888         return r;
1889 }
1890
1891 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1892 {
1893         if (sigset) {
1894                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1895                 vcpu->sigset_active = 1;
1896                 vcpu->sigset = *sigset;
1897         } else
1898                 vcpu->sigset_active = 0;
1899         return 0;
1900 }
1901
1902 static long kvm_vcpu_ioctl(struct file *filp,
1903                            unsigned int ioctl, unsigned long arg)
1904 {
1905         struct kvm_vcpu *vcpu = filp->private_data;
1906         void __user *argp = (void __user *)arg;
1907         int r;
1908         struct kvm_fpu *fpu = NULL;
1909         struct kvm_sregs *kvm_sregs = NULL;
1910
1911         if (vcpu->kvm->mm != current->mm)
1912                 return -EIO;
1913
1914 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1915         /*
1916          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1917          * so vcpu_load() would break it.
1918          */
1919         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1920                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1921 #endif
1922
1923
1924         r = vcpu_load(vcpu);
1925         if (r)
1926                 return r;
1927         switch (ioctl) {
1928         case KVM_RUN:
1929                 r = -EINVAL;
1930                 if (arg)
1931                         goto out;
1932                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1933                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1934                 break;
1935         case KVM_GET_REGS: {
1936                 struct kvm_regs *kvm_regs;
1937
1938                 r = -ENOMEM;
1939                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1940                 if (!kvm_regs)
1941                         goto out;
1942                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1943                 if (r)
1944                         goto out_free1;
1945                 r = -EFAULT;
1946                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1947                         goto out_free1;
1948                 r = 0;
1949 out_free1:
1950                 kfree(kvm_regs);
1951                 break;
1952         }
1953         case KVM_SET_REGS: {
1954                 struct kvm_regs *kvm_regs;
1955
1956                 r = -ENOMEM;
1957                 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
1958                 if (IS_ERR(kvm_regs)) {
1959                         r = PTR_ERR(kvm_regs);
1960                         goto out;
1961                 }
1962                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1963                 kfree(kvm_regs);
1964                 break;
1965         }
1966         case KVM_GET_SREGS: {
1967                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1968                 r = -ENOMEM;
1969                 if (!kvm_sregs)
1970                         goto out;
1971                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1972                 if (r)
1973                         goto out;
1974                 r = -EFAULT;
1975                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1976                         goto out;
1977                 r = 0;
1978                 break;
1979         }
1980         case KVM_SET_SREGS: {
1981                 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
1982                 if (IS_ERR(kvm_sregs)) {
1983                         r = PTR_ERR(kvm_sregs);
1984                         kvm_sregs = NULL;
1985                         goto out;
1986                 }
1987                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1988                 break;
1989         }
1990         case KVM_GET_MP_STATE: {
1991                 struct kvm_mp_state mp_state;
1992
1993                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1994                 if (r)
1995                         goto out;
1996                 r = -EFAULT;
1997                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1998                         goto out;
1999                 r = 0;
2000                 break;
2001         }
2002         case KVM_SET_MP_STATE: {
2003                 struct kvm_mp_state mp_state;
2004
2005                 r = -EFAULT;
2006                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
2007                         goto out;
2008                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2009                 break;
2010         }
2011         case KVM_TRANSLATE: {
2012                 struct kvm_translation tr;
2013
2014                 r = -EFAULT;
2015                 if (copy_from_user(&tr, argp, sizeof tr))
2016                         goto out;
2017                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2018                 if (r)
2019                         goto out;
2020                 r = -EFAULT;
2021                 if (copy_to_user(argp, &tr, sizeof tr))
2022                         goto out;
2023                 r = 0;
2024                 break;
2025         }
2026         case KVM_SET_GUEST_DEBUG: {
2027                 struct kvm_guest_debug dbg;
2028
2029                 r = -EFAULT;
2030                 if (copy_from_user(&dbg, argp, sizeof dbg))
2031                         goto out;
2032                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2033                 break;
2034         }
2035         case KVM_SET_SIGNAL_MASK: {
2036                 struct kvm_signal_mask __user *sigmask_arg = argp;
2037                 struct kvm_signal_mask kvm_sigmask;
2038                 sigset_t sigset, *p;
2039
2040                 p = NULL;
2041                 if (argp) {
2042                         r = -EFAULT;
2043                         if (copy_from_user(&kvm_sigmask, argp,
2044                                            sizeof kvm_sigmask))
2045                                 goto out;
2046                         r = -EINVAL;
2047                         if (kvm_sigmask.len != sizeof sigset)
2048                                 goto out;
2049                         r = -EFAULT;
2050                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2051                                            sizeof sigset))
2052                                 goto out;
2053                         p = &sigset;
2054                 }
2055                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2056                 break;
2057         }
2058         case KVM_GET_FPU: {
2059                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2060                 r = -ENOMEM;
2061                 if (!fpu)
2062                         goto out;
2063                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2064                 if (r)
2065                         goto out;
2066                 r = -EFAULT;
2067                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2068                         goto out;
2069                 r = 0;
2070                 break;
2071         }
2072         case KVM_SET_FPU: {
2073                 fpu = memdup_user(argp, sizeof(*fpu));
2074                 if (IS_ERR(fpu)) {
2075                         r = PTR_ERR(fpu);
2076                         fpu = NULL;
2077                         goto out;
2078                 }
2079                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2080                 break;
2081         }
2082         default:
2083                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2084         }
2085 out:
2086         vcpu_put(vcpu);
2087         kfree(fpu);
2088         kfree(kvm_sregs);
2089         return r;
2090 }
2091
2092 #ifdef CONFIG_COMPAT
2093 static long kvm_vcpu_compat_ioctl(struct file *filp,
2094                                   unsigned int ioctl, unsigned long arg)
2095 {
2096         struct kvm_vcpu *vcpu = filp->private_data;
2097         void __user *argp = compat_ptr(arg);
2098         int r;
2099
2100         if (vcpu->kvm->mm != current->mm)
2101                 return -EIO;
2102
2103         switch (ioctl) {
2104         case KVM_SET_SIGNAL_MASK: {
2105                 struct kvm_signal_mask __user *sigmask_arg = argp;
2106                 struct kvm_signal_mask kvm_sigmask;
2107                 compat_sigset_t csigset;
2108                 sigset_t sigset;
2109
2110                 if (argp) {
2111                         r = -EFAULT;
2112                         if (copy_from_user(&kvm_sigmask, argp,
2113                                            sizeof kvm_sigmask))
2114                                 goto out;
2115                         r = -EINVAL;
2116                         if (kvm_sigmask.len != sizeof csigset)
2117                                 goto out;
2118                         r = -EFAULT;
2119                         if (copy_from_user(&csigset, sigmask_arg->sigset,
2120                                            sizeof csigset))
2121                                 goto out;
2122                         sigset_from_compat(&sigset, &csigset);
2123                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2124                 } else
2125                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2126                 break;
2127         }
2128         default:
2129                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2130         }
2131
2132 out:
2133         return r;
2134 }
2135 #endif
2136
2137 static long kvm_vm_ioctl(struct file *filp,
2138                            unsigned int ioctl, unsigned long arg)
2139 {
2140         struct kvm *kvm = filp->private_data;
2141         void __user *argp = (void __user *)arg;
2142         int r;
2143
2144         if (kvm->mm != current->mm)
2145                 return -EIO;
2146         switch (ioctl) {
2147         case KVM_CREATE_VCPU:
2148                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2149                 break;
2150         case KVM_SET_USER_MEMORY_REGION: {
2151                 struct kvm_userspace_memory_region kvm_userspace_mem;
2152
2153                 r = -EFAULT;
2154                 if (copy_from_user(&kvm_userspace_mem, argp,
2155                                                 sizeof kvm_userspace_mem))
2156                         goto out;
2157
2158                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, true);
2159                 break;
2160         }
2161         case KVM_GET_DIRTY_LOG: {
2162                 struct kvm_dirty_log log;
2163
2164                 r = -EFAULT;
2165                 if (copy_from_user(&log, argp, sizeof log))
2166                         goto out;
2167                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2168                 break;
2169         }
2170 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2171         case KVM_REGISTER_COALESCED_MMIO: {
2172                 struct kvm_coalesced_mmio_zone zone;
2173                 r = -EFAULT;
2174                 if (copy_from_user(&zone, argp, sizeof zone))
2175                         goto out;
2176                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2177                 break;
2178         }
2179         case KVM_UNREGISTER_COALESCED_MMIO: {
2180                 struct kvm_coalesced_mmio_zone zone;
2181                 r = -EFAULT;
2182                 if (copy_from_user(&zone, argp, sizeof zone))
2183                         goto out;
2184                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2185                 break;
2186         }
2187 #endif
2188         case KVM_IRQFD: {
2189                 struct kvm_irqfd data;
2190
2191                 r = -EFAULT;
2192                 if (copy_from_user(&data, argp, sizeof data))
2193                         goto out;
2194                 r = kvm_irqfd(kvm, &data);
2195                 break;
2196         }
2197         case KVM_IOEVENTFD: {
2198                 struct kvm_ioeventfd data;
2199
2200                 r = -EFAULT;
2201                 if (copy_from_user(&data, argp, sizeof data))
2202                         goto out;
2203                 r = kvm_ioeventfd(kvm, &data);
2204                 break;
2205         }
2206 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2207         case KVM_SET_BOOT_CPU_ID:
2208                 r = 0;
2209                 mutex_lock(&kvm->lock);
2210                 if (atomic_read(&kvm->online_vcpus) != 0)
2211                         r = -EBUSY;
2212                 else
2213                         kvm->bsp_vcpu_id = arg;
2214                 mutex_unlock(&kvm->lock);
2215                 break;
2216 #endif
2217 #ifdef CONFIG_HAVE_KVM_MSI
2218         case KVM_SIGNAL_MSI: {
2219                 struct kvm_msi msi;
2220
2221                 r = -EFAULT;
2222                 if (copy_from_user(&msi, argp, sizeof msi))
2223                         goto out;
2224                 r = kvm_send_userspace_msi(kvm, &msi);
2225                 break;
2226         }
2227 #endif
2228 #ifdef __KVM_HAVE_IRQ_LINE
2229         case KVM_IRQ_LINE_STATUS:
2230         case KVM_IRQ_LINE: {
2231                 struct kvm_irq_level irq_event;
2232
2233                 r = -EFAULT;
2234                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2235                         goto out;
2236
2237                 r = kvm_vm_ioctl_irq_line(kvm, &irq_event);
2238                 if (r)
2239                         goto out;
2240
2241                 r = -EFAULT;
2242                 if (ioctl == KVM_IRQ_LINE_STATUS) {
2243                         if (copy_to_user(argp, &irq_event, sizeof irq_event))
2244                                 goto out;
2245                 }
2246
2247                 r = 0;
2248                 break;
2249         }
2250 #endif
2251         default:
2252                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2253                 if (r == -ENOTTY)
2254                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2255         }
2256 out:
2257         return r;
2258 }
2259
2260 #ifdef CONFIG_COMPAT
2261 struct compat_kvm_dirty_log {
2262         __u32 slot;
2263         __u32 padding1;
2264         union {
2265                 compat_uptr_t dirty_bitmap; /* one bit per page */
2266                 __u64 padding2;
2267         };
2268 };
2269
2270 static long kvm_vm_compat_ioctl(struct file *filp,
2271                            unsigned int ioctl, unsigned long arg)
2272 {
2273         struct kvm *kvm = filp->private_data;
2274         int r;
2275
2276         if (kvm->mm != current->mm)
2277                 return -EIO;
2278         switch (ioctl) {
2279         case KVM_GET_DIRTY_LOG: {
2280                 struct compat_kvm_dirty_log compat_log;
2281                 struct kvm_dirty_log log;
2282
2283                 r = -EFAULT;
2284                 if (copy_from_user(&compat_log, (void __user *)arg,
2285                                    sizeof(compat_log)))
2286                         goto out;
2287                 log.slot         = compat_log.slot;
2288                 log.padding1     = compat_log.padding1;
2289                 log.padding2     = compat_log.padding2;
2290                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2291
2292                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2293                 break;
2294         }
2295         default:
2296                 r = kvm_vm_ioctl(filp, ioctl, arg);
2297         }
2298
2299 out:
2300         return r;
2301 }
2302 #endif
2303
2304 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2305 {
2306         struct page *page[1];
2307         unsigned long addr;
2308         int npages;
2309         gfn_t gfn = vmf->pgoff;
2310         struct kvm *kvm = vma->vm_file->private_data;
2311
2312         addr = gfn_to_hva(kvm, gfn);
2313         if (kvm_is_error_hva(addr))
2314                 return VM_FAULT_SIGBUS;
2315
2316         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2317                                 NULL);
2318         if (unlikely(npages != 1))
2319                 return VM_FAULT_SIGBUS;
2320
2321         vmf->page = page[0];
2322         return 0;
2323 }
2324
2325 static const struct vm_operations_struct kvm_vm_vm_ops = {
2326         .fault = kvm_vm_fault,
2327 };
2328
2329 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2330 {
2331         vma->vm_ops = &kvm_vm_vm_ops;
2332         return 0;
2333 }
2334
2335 static struct file_operations kvm_vm_fops = {
2336         .release        = kvm_vm_release,
2337         .unlocked_ioctl = kvm_vm_ioctl,
2338 #ifdef CONFIG_COMPAT
2339         .compat_ioctl   = kvm_vm_compat_ioctl,
2340 #endif
2341         .mmap           = kvm_vm_mmap,
2342         .llseek         = noop_llseek,
2343 };
2344
2345 static int kvm_dev_ioctl_create_vm(unsigned long type)
2346 {
2347         int r;
2348         struct kvm *kvm;
2349
2350         kvm = kvm_create_vm(type);
2351         if (IS_ERR(kvm))
2352                 return PTR_ERR(kvm);
2353 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2354         r = kvm_coalesced_mmio_init(kvm);
2355         if (r < 0) {
2356                 kvm_put_kvm(kvm);
2357                 return r;
2358         }
2359 #endif
2360         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2361         if (r < 0)
2362                 kvm_put_kvm(kvm);
2363
2364         return r;
2365 }
2366
2367 static long kvm_dev_ioctl_check_extension_generic(long arg)
2368 {
2369         switch (arg) {
2370         case KVM_CAP_USER_MEMORY:
2371         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2372         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2373 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2374         case KVM_CAP_SET_BOOT_CPU_ID:
2375 #endif
2376         case KVM_CAP_INTERNAL_ERROR_DATA:
2377 #ifdef CONFIG_HAVE_KVM_MSI
2378         case KVM_CAP_SIGNAL_MSI:
2379 #endif
2380                 return 1;
2381 #ifdef KVM_CAP_IRQ_ROUTING
2382         case KVM_CAP_IRQ_ROUTING:
2383                 return KVM_MAX_IRQ_ROUTES;
2384 #endif
2385         default:
2386                 break;
2387         }
2388         return kvm_dev_ioctl_check_extension(arg);
2389 }
2390
2391 static long kvm_dev_ioctl(struct file *filp,
2392                           unsigned int ioctl, unsigned long arg)
2393 {
2394         long r = -EINVAL;
2395
2396         switch (ioctl) {
2397         case KVM_GET_API_VERSION:
2398                 r = -EINVAL;
2399                 if (arg)
2400                         goto out;
2401                 r = KVM_API_VERSION;
2402                 break;
2403         case KVM_CREATE_VM:
2404                 r = kvm_dev_ioctl_create_vm(arg);
2405                 break;
2406         case KVM_CHECK_EXTENSION:
2407                 r = kvm_dev_ioctl_check_extension_generic(arg);
2408                 break;
2409         case KVM_GET_VCPU_MMAP_SIZE:
2410                 r = -EINVAL;
2411                 if (arg)
2412                         goto out;
2413                 r = PAGE_SIZE;     /* struct kvm_run */
2414 #ifdef CONFIG_X86
2415                 r += PAGE_SIZE;    /* pio data page */
2416 #endif
2417 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2418                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2419 #endif
2420                 break;
2421         case KVM_TRACE_ENABLE:
2422         case KVM_TRACE_PAUSE:
2423         case KVM_TRACE_DISABLE:
2424                 r = -EOPNOTSUPP;
2425                 break;
2426         default:
2427                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2428         }
2429 out:
2430         return r;
2431 }
2432
2433 static struct file_operations kvm_chardev_ops = {
2434         .unlocked_ioctl = kvm_dev_ioctl,
2435         .compat_ioctl   = kvm_dev_ioctl,
2436         .llseek         = noop_llseek,
2437 };
2438
2439 static struct miscdevice kvm_dev = {
2440         KVM_MINOR,
2441         "kvm",
2442         &kvm_chardev_ops,
2443 };
2444
2445 static void hardware_enable_nolock(void *junk)
2446 {
2447         int cpu = raw_smp_processor_id();
2448         int r;
2449
2450         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2451                 return;
2452
2453         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2454
2455         r = kvm_arch_hardware_enable(NULL);
2456
2457         if (r) {
2458                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2459                 atomic_inc(&hardware_enable_failed);
2460                 printk(KERN_INFO "kvm: enabling virtualization on "
2461                                  "CPU%d failed\n", cpu);
2462         }
2463 }
2464
2465 static void hardware_enable(void *junk)
2466 {
2467         raw_spin_lock(&kvm_lock);
2468         hardware_enable_nolock(junk);
2469         raw_spin_unlock(&kvm_lock);
2470 }
2471
2472 static void hardware_disable_nolock(void *junk)
2473 {
2474         int cpu = raw_smp_processor_id();
2475
2476         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2477                 return;
2478         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2479         kvm_arch_hardware_disable(NULL);
2480 }
2481
2482 static void hardware_disable(void *junk)
2483 {
2484         raw_spin_lock(&kvm_lock);
2485         hardware_disable_nolock(junk);
2486         raw_spin_unlock(&kvm_lock);
2487 }
2488
2489 static void hardware_disable_all_nolock(void)
2490 {
2491         BUG_ON(!kvm_usage_count);
2492
2493         kvm_usage_count--;
2494         if (!kvm_usage_count)
2495                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2496 }
2497
2498 static void hardware_disable_all(void)
2499 {
2500         raw_spin_lock(&kvm_lock);
2501         hardware_disable_all_nolock();
2502         raw_spin_unlock(&kvm_lock);
2503 }
2504
2505 static int hardware_enable_all(void)
2506 {
2507         int r = 0;
2508
2509         raw_spin_lock(&kvm_lock);
2510
2511         kvm_usage_count++;
2512         if (kvm_usage_count == 1) {
2513                 atomic_set(&hardware_enable_failed, 0);
2514                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2515
2516                 if (atomic_read(&hardware_enable_failed)) {
2517                         hardware_disable_all_nolock();
2518                         r = -EBUSY;
2519                 }
2520         }
2521
2522         raw_spin_unlock(&kvm_lock);
2523
2524         return r;
2525 }
2526
2527 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2528                            void *v)
2529 {
2530         int cpu = (long)v;
2531
2532         if (!kvm_usage_count)
2533                 return NOTIFY_OK;
2534
2535         val &= ~CPU_TASKS_FROZEN;
2536         switch (val) {
2537         case CPU_DYING:
2538                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2539                        cpu);
2540                 hardware_disable(NULL);
2541                 break;
2542         case CPU_STARTING:
2543                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2544                        cpu);
2545                 hardware_enable(NULL);
2546                 break;
2547         }
2548         return NOTIFY_OK;
2549 }
2550
2551
2552 asmlinkage void kvm_spurious_fault(void)
2553 {
2554         /* Fault while not rebooting.  We want the trace. */
2555         BUG();
2556 }
2557 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
2558
2559 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2560                       void *v)
2561 {
2562         /*
2563          * Some (well, at least mine) BIOSes hang on reboot if
2564          * in vmx root mode.
2565          *
2566          * And Intel TXT required VMX off for all cpu when system shutdown.
2567          */
2568         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2569         kvm_rebooting = true;
2570         on_each_cpu(hardware_disable_nolock, NULL, 1);
2571         return NOTIFY_OK;
2572 }
2573
2574 static struct notifier_block kvm_reboot_notifier = {
2575         .notifier_call = kvm_reboot,
2576         .priority = 0,
2577 };
2578
2579 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2580 {
2581         int i;
2582
2583         for (i = 0; i < bus->dev_count; i++) {
2584                 struct kvm_io_device *pos = bus->range[i].dev;
2585
2586                 kvm_iodevice_destructor(pos);
2587         }
2588         kfree(bus);
2589 }
2590
2591 int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2592 {
2593         const struct kvm_io_range *r1 = p1;
2594         const struct kvm_io_range *r2 = p2;
2595
2596         if (r1->addr < r2->addr)
2597                 return -1;
2598         if (r1->addr + r1->len > r2->addr + r2->len)
2599                 return 1;
2600         return 0;
2601 }
2602
2603 int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2604                           gpa_t addr, int len)
2605 {
2606         bus->range[bus->dev_count++] = (struct kvm_io_range) {
2607                 .addr = addr,
2608                 .len = len,
2609                 .dev = dev,
2610         };
2611
2612         sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2613                 kvm_io_bus_sort_cmp, NULL);
2614
2615         return 0;
2616 }
2617
2618 int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2619                              gpa_t addr, int len)
2620 {
2621         struct kvm_io_range *range, key;
2622         int off;
2623
2624         key = (struct kvm_io_range) {
2625                 .addr = addr,
2626                 .len = len,
2627         };
2628
2629         range = bsearch(&key, bus->range, bus->dev_count,
2630                         sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2631         if (range == NULL)
2632                 return -ENOENT;
2633
2634         off = range - bus->range;
2635
2636         while (off > 0 && kvm_io_bus_sort_cmp(&key, &bus->range[off-1]) == 0)
2637                 off--;
2638
2639         return off;
2640 }
2641
2642 /* kvm_io_bus_write - called under kvm->slots_lock */
2643 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2644                      int len, const void *val)
2645 {
2646         int idx;
2647         struct kvm_io_bus *bus;
2648         struct kvm_io_range range;
2649
2650         range = (struct kvm_io_range) {
2651                 .addr = addr,
2652                 .len = len,
2653         };
2654
2655         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2656         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2657         if (idx < 0)
2658                 return -EOPNOTSUPP;
2659
2660         while (idx < bus->dev_count &&
2661                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2662                 if (!kvm_iodevice_write(bus->range[idx].dev, addr, len, val))
2663                         return 0;
2664                 idx++;
2665         }
2666
2667         return -EOPNOTSUPP;
2668 }
2669
2670 /* kvm_io_bus_read - called under kvm->slots_lock */
2671 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2672                     int len, void *val)
2673 {
2674         int idx;
2675         struct kvm_io_bus *bus;
2676         struct kvm_io_range range;
2677
2678         range = (struct kvm_io_range) {
2679                 .addr = addr,
2680                 .len = len,
2681         };
2682
2683         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2684         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2685         if (idx < 0)
2686                 return -EOPNOTSUPP;
2687
2688         while (idx < bus->dev_count &&
2689                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2690                 if (!kvm_iodevice_read(bus->range[idx].dev, addr, len, val))
2691                         return 0;
2692                 idx++;
2693         }
2694
2695         return -EOPNOTSUPP;
2696 }
2697
2698 /* Caller must hold slots_lock. */
2699 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2700                             int len, struct kvm_io_device *dev)
2701 {
2702         struct kvm_io_bus *new_bus, *bus;
2703
2704         bus = kvm->buses[bus_idx];
2705         if (bus->dev_count > NR_IOBUS_DEVS - 1)
2706                 return -ENOSPC;
2707
2708         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
2709                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2710         if (!new_bus)
2711                 return -ENOMEM;
2712         memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
2713                sizeof(struct kvm_io_range)));
2714         kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2715         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2716         synchronize_srcu_expedited(&kvm->srcu);
2717         kfree(bus);
2718
2719         return 0;
2720 }
2721
2722 /* Caller must hold slots_lock. */
2723 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2724                               struct kvm_io_device *dev)
2725 {
2726         int i, r;
2727         struct kvm_io_bus *new_bus, *bus;
2728
2729         bus = kvm->buses[bus_idx];
2730         r = -ENOENT;
2731         for (i = 0; i < bus->dev_count; i++)
2732                 if (bus->range[i].dev == dev) {
2733                         r = 0;
2734                         break;
2735                 }
2736
2737         if (r)
2738                 return r;
2739
2740         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
2741                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2742         if (!new_bus)
2743                 return -ENOMEM;
2744
2745         memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
2746         new_bus->dev_count--;
2747         memcpy(new_bus->range + i, bus->range + i + 1,
2748                (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
2749
2750         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2751         synchronize_srcu_expedited(&kvm->srcu);
2752         kfree(bus);
2753         return r;
2754 }
2755
2756 static struct notifier_block kvm_cpu_notifier = {
2757         .notifier_call = kvm_cpu_hotplug,
2758 };
2759
2760 static int vm_stat_get(void *_offset, u64 *val)
2761 {
2762         unsigned offset = (long)_offset;
2763         struct kvm *kvm;
2764
2765         *val = 0;
2766         raw_spin_lock(&kvm_lock);
2767         list_for_each_entry(kvm, &vm_list, vm_list)
2768                 *val += *(u32 *)((void *)kvm + offset);
2769         raw_spin_unlock(&kvm_lock);
2770         return 0;
2771 }
2772
2773 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2774
2775 static int vcpu_stat_get(void *_offset, u64 *val)
2776 {
2777         unsigned offset = (long)_offset;
2778         struct kvm *kvm;
2779         struct kvm_vcpu *vcpu;
2780         int i;
2781
2782         *val = 0;
2783         raw_spin_lock(&kvm_lock);
2784         list_for_each_entry(kvm, &vm_list, vm_list)
2785                 kvm_for_each_vcpu(i, vcpu, kvm)
2786                         *val += *(u32 *)((void *)vcpu + offset);
2787
2788         raw_spin_unlock(&kvm_lock);
2789         return 0;
2790 }
2791
2792 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2793
2794 static const struct file_operations *stat_fops[] = {
2795         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2796         [KVM_STAT_VM]   = &vm_stat_fops,
2797 };
2798
2799 static int kvm_init_debug(void)
2800 {
2801         int r = -EFAULT;
2802         struct kvm_stats_debugfs_item *p;
2803
2804         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2805         if (kvm_debugfs_dir == NULL)
2806                 goto out;
2807
2808         for (p = debugfs_entries; p->name; ++p) {
2809                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2810                                                 (void *)(long)p->offset,
2811                                                 stat_fops[p->kind]);
2812                 if (p->dentry == NULL)
2813                         goto out_dir;
2814         }
2815
2816         return 0;
2817
2818 out_dir:
2819         debugfs_remove_recursive(kvm_debugfs_dir);
2820 out:
2821         return r;
2822 }
2823
2824 static void kvm_exit_debug(void)
2825 {
2826         struct kvm_stats_debugfs_item *p;
2827
2828         for (p = debugfs_entries; p->name; ++p)
2829                 debugfs_remove(p->dentry);
2830         debugfs_remove(kvm_debugfs_dir);
2831 }
2832
2833 static int kvm_suspend(void)
2834 {
2835         if (kvm_usage_count)
2836                 hardware_disable_nolock(NULL);
2837         return 0;
2838 }
2839
2840 static void kvm_resume(void)
2841 {
2842         if (kvm_usage_count) {
2843                 WARN_ON(raw_spin_is_locked(&kvm_lock));
2844                 hardware_enable_nolock(NULL);
2845         }
2846 }
2847
2848 static struct syscore_ops kvm_syscore_ops = {
2849         .suspend = kvm_suspend,
2850         .resume = kvm_resume,
2851 };
2852
2853 static inline
2854 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2855 {
2856         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2857 }
2858
2859 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2860 {
2861         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2862
2863         kvm_arch_vcpu_load(vcpu, cpu);
2864 }
2865
2866 static void kvm_sched_out(struct preempt_notifier *pn,
2867                           struct task_struct *next)
2868 {
2869         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2870
2871         kvm_arch_vcpu_put(vcpu);
2872 }
2873
2874 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2875                   struct module *module)
2876 {
2877         int r;
2878         int cpu;
2879
2880         r = kvm_arch_init(opaque);
2881         if (r)
2882                 goto out_fail;
2883
2884         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2885                 r = -ENOMEM;
2886                 goto out_free_0;
2887         }
2888
2889         r = kvm_arch_hardware_setup();
2890         if (r < 0)
2891                 goto out_free_0a;
2892
2893         for_each_online_cpu(cpu) {
2894                 smp_call_function_single(cpu,
2895                                 kvm_arch_check_processor_compat,
2896                                 &r, 1);
2897                 if (r < 0)
2898                         goto out_free_1;
2899         }
2900
2901         r = register_cpu_notifier(&kvm_cpu_notifier);
2902         if (r)
2903                 goto out_free_2;
2904         register_reboot_notifier(&kvm_reboot_notifier);
2905
2906         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2907         if (!vcpu_align)
2908                 vcpu_align = __alignof__(struct kvm_vcpu);
2909         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2910                                            0, NULL);
2911         if (!kvm_vcpu_cache) {
2912                 r = -ENOMEM;
2913                 goto out_free_3;
2914         }
2915
2916         r = kvm_async_pf_init();
2917         if (r)
2918                 goto out_free;
2919
2920         kvm_chardev_ops.owner = module;
2921         kvm_vm_fops.owner = module;
2922         kvm_vcpu_fops.owner = module;
2923
2924         r = misc_register(&kvm_dev);
2925         if (r) {
2926                 printk(KERN_ERR "kvm: misc device register failed\n");
2927                 goto out_unreg;
2928         }
2929
2930         register_syscore_ops(&kvm_syscore_ops);
2931
2932         kvm_preempt_ops.sched_in = kvm_sched_in;
2933         kvm_preempt_ops.sched_out = kvm_sched_out;
2934
2935         r = kvm_init_debug();
2936         if (r) {
2937                 printk(KERN_ERR "kvm: create debugfs files failed\n");
2938                 goto out_undebugfs;
2939         }
2940
2941         return 0;
2942
2943 out_undebugfs:
2944         unregister_syscore_ops(&kvm_syscore_ops);
2945 out_unreg:
2946         kvm_async_pf_deinit();
2947 out_free:
2948         kmem_cache_destroy(kvm_vcpu_cache);
2949 out_free_3:
2950         unregister_reboot_notifier(&kvm_reboot_notifier);
2951         unregister_cpu_notifier(&kvm_cpu_notifier);
2952 out_free_2:
2953 out_free_1:
2954         kvm_arch_hardware_unsetup();
2955 out_free_0a:
2956         free_cpumask_var(cpus_hardware_enabled);
2957 out_free_0:
2958         kvm_arch_exit();
2959 out_fail:
2960         return r;
2961 }
2962 EXPORT_SYMBOL_GPL(kvm_init);
2963
2964 void kvm_exit(void)
2965 {
2966         kvm_exit_debug();
2967         misc_deregister(&kvm_dev);
2968         kmem_cache_destroy(kvm_vcpu_cache);
2969         kvm_async_pf_deinit();
2970         unregister_syscore_ops(&kvm_syscore_ops);
2971         unregister_reboot_notifier(&kvm_reboot_notifier);
2972         unregister_cpu_notifier(&kvm_cpu_notifier);
2973         on_each_cpu(hardware_disable_nolock, NULL, 1);
2974         kvm_arch_hardware_unsetup();
2975         kvm_arch_exit();
2976         free_cpumask_var(cpus_hardware_enabled);
2977 }
2978 EXPORT_SYMBOL_GPL(kvm_exit);