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