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