KVM: improve hva_to_pfn() readability
[linux-2.6.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/sysdev.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
51 #include <asm/processor.h>
52 #include <asm/io.h>
53 #include <asm/uaccess.h>
54 #include <asm/pgtable.h>
55 #include <asm-generic/bitops/le.h>
56
57 #include "coalesced_mmio.h"
58 #include "async_pf.h"
59
60 #define CREATE_TRACE_POINTS
61 #include <trace/events/kvm.h>
62
63 MODULE_AUTHOR("Qumranet");
64 MODULE_LICENSE("GPL");
65
66 /*
67  * Ordering of locks:
68  *
69  *              kvm->lock --> kvm->slots_lock --> kvm->irq_lock
70  */
71
72 DEFINE_SPINLOCK(kvm_lock);
73 LIST_HEAD(vm_list);
74
75 static cpumask_var_t cpus_hardware_enabled;
76 static int kvm_usage_count = 0;
77 static atomic_t hardware_enable_failed;
78
79 struct kmem_cache *kvm_vcpu_cache;
80 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
81
82 static __read_mostly struct preempt_ops kvm_preempt_ops;
83
84 struct dentry *kvm_debugfs_dir;
85
86 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
87                            unsigned long arg);
88 static int hardware_enable_all(void);
89 static void hardware_disable_all(void);
90
91 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
92
93 static bool kvm_rebooting;
94
95 static bool largepages_enabled = true;
96
97 static struct page *hwpoison_page;
98 static pfn_t hwpoison_pfn;
99
100 static struct page *fault_page;
101 static pfn_t fault_pfn;
102
103 inline int kvm_is_mmio_pfn(pfn_t pfn)
104 {
105         if (pfn_valid(pfn)) {
106                 struct page *page = compound_head(pfn_to_page(pfn));
107                 return PageReserved(page);
108         }
109
110         return true;
111 }
112
113 /*
114  * Switches to specified vcpu, until a matching vcpu_put()
115  */
116 void vcpu_load(struct kvm_vcpu *vcpu)
117 {
118         int cpu;
119
120         mutex_lock(&vcpu->mutex);
121         cpu = get_cpu();
122         preempt_notifier_register(&vcpu->preempt_notifier);
123         kvm_arch_vcpu_load(vcpu, cpu);
124         put_cpu();
125 }
126
127 void vcpu_put(struct kvm_vcpu *vcpu)
128 {
129         preempt_disable();
130         kvm_arch_vcpu_put(vcpu);
131         preempt_notifier_unregister(&vcpu->preempt_notifier);
132         preempt_enable();
133         mutex_unlock(&vcpu->mutex);
134 }
135
136 static void ack_flush(void *_completed)
137 {
138 }
139
140 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
141 {
142         int i, cpu, me;
143         cpumask_var_t cpus;
144         bool called = true;
145         struct kvm_vcpu *vcpu;
146
147         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
148
149         raw_spin_lock(&kvm->requests_lock);
150         me = smp_processor_id();
151         kvm_for_each_vcpu(i, vcpu, kvm) {
152                 if (kvm_make_check_request(req, vcpu))
153                         continue;
154                 cpu = vcpu->cpu;
155                 if (cpus != NULL && cpu != -1 && cpu != me)
156                         cpumask_set_cpu(cpu, cpus);
157         }
158         if (unlikely(cpus == NULL))
159                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
160         else if (!cpumask_empty(cpus))
161                 smp_call_function_many(cpus, ack_flush, NULL, 1);
162         else
163                 called = false;
164         raw_spin_unlock(&kvm->requests_lock);
165         free_cpumask_var(cpus);
166         return called;
167 }
168
169 void kvm_flush_remote_tlbs(struct kvm *kvm)
170 {
171         if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
172                 ++kvm->stat.remote_tlb_flush;
173 }
174
175 void kvm_reload_remote_mmus(struct kvm *kvm)
176 {
177         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
178 }
179
180 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
181 {
182         struct page *page;
183         int r;
184
185         mutex_init(&vcpu->mutex);
186         vcpu->cpu = -1;
187         vcpu->kvm = kvm;
188         vcpu->vcpu_id = id;
189         init_waitqueue_head(&vcpu->wq);
190         kvm_async_pf_vcpu_init(vcpu);
191
192         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
193         if (!page) {
194                 r = -ENOMEM;
195                 goto fail;
196         }
197         vcpu->run = page_address(page);
198
199         r = kvm_arch_vcpu_init(vcpu);
200         if (r < 0)
201                 goto fail_free_run;
202         return 0;
203
204 fail_free_run:
205         free_page((unsigned long)vcpu->run);
206 fail:
207         return r;
208 }
209 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
210
211 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
212 {
213         kvm_arch_vcpu_uninit(vcpu);
214         free_page((unsigned long)vcpu->run);
215 }
216 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
217
218 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
219 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
220 {
221         return container_of(mn, struct kvm, mmu_notifier);
222 }
223
224 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
225                                              struct mm_struct *mm,
226                                              unsigned long address)
227 {
228         struct kvm *kvm = mmu_notifier_to_kvm(mn);
229         int need_tlb_flush, idx;
230
231         /*
232          * When ->invalidate_page runs, the linux pte has been zapped
233          * already but the page is still allocated until
234          * ->invalidate_page returns. So if we increase the sequence
235          * here the kvm page fault will notice if the spte can't be
236          * established because the page is going to be freed. If
237          * instead the kvm page fault establishes the spte before
238          * ->invalidate_page runs, kvm_unmap_hva will release it
239          * before returning.
240          *
241          * The sequence increase only need to be seen at spin_unlock
242          * time, and not at spin_lock time.
243          *
244          * Increasing the sequence after the spin_unlock would be
245          * unsafe because the kvm page fault could then establish the
246          * pte after kvm_unmap_hva returned, without noticing the page
247          * is going to be freed.
248          */
249         idx = srcu_read_lock(&kvm->srcu);
250         spin_lock(&kvm->mmu_lock);
251         kvm->mmu_notifier_seq++;
252         need_tlb_flush = kvm_unmap_hva(kvm, address);
253         spin_unlock(&kvm->mmu_lock);
254         srcu_read_unlock(&kvm->srcu, idx);
255
256         /* we've to flush the tlb before the pages can be freed */
257         if (need_tlb_flush)
258                 kvm_flush_remote_tlbs(kvm);
259
260 }
261
262 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
263                                         struct mm_struct *mm,
264                                         unsigned long address,
265                                         pte_t pte)
266 {
267         struct kvm *kvm = mmu_notifier_to_kvm(mn);
268         int idx;
269
270         idx = srcu_read_lock(&kvm->srcu);
271         spin_lock(&kvm->mmu_lock);
272         kvm->mmu_notifier_seq++;
273         kvm_set_spte_hva(kvm, address, pte);
274         spin_unlock(&kvm->mmu_lock);
275         srcu_read_unlock(&kvm->srcu, idx);
276 }
277
278 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
279                                                     struct mm_struct *mm,
280                                                     unsigned long start,
281                                                     unsigned long end)
282 {
283         struct kvm *kvm = mmu_notifier_to_kvm(mn);
284         int need_tlb_flush = 0, idx;
285
286         idx = srcu_read_lock(&kvm->srcu);
287         spin_lock(&kvm->mmu_lock);
288         /*
289          * The count increase must become visible at unlock time as no
290          * spte can be established without taking the mmu_lock and
291          * count is also read inside the mmu_lock critical section.
292          */
293         kvm->mmu_notifier_count++;
294         for (; start < end; start += PAGE_SIZE)
295                 need_tlb_flush |= kvm_unmap_hva(kvm, start);
296         spin_unlock(&kvm->mmu_lock);
297         srcu_read_unlock(&kvm->srcu, idx);
298
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
304 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
305                                                   struct mm_struct *mm,
306                                                   unsigned long start,
307                                                   unsigned long end)
308 {
309         struct kvm *kvm = mmu_notifier_to_kvm(mn);
310
311         spin_lock(&kvm->mmu_lock);
312         /*
313          * This sequence increase will notify the kvm page fault that
314          * the page that is going to be mapped in the spte could have
315          * been freed.
316          */
317         kvm->mmu_notifier_seq++;
318         /*
319          * The above sequence increase must be visible before the
320          * below count decrease but both values are read by the kvm
321          * page fault under mmu_lock spinlock so we don't need to add
322          * a smb_wmb() here in between the two.
323          */
324         kvm->mmu_notifier_count--;
325         spin_unlock(&kvm->mmu_lock);
326
327         BUG_ON(kvm->mmu_notifier_count < 0);
328 }
329
330 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
331                                               struct mm_struct *mm,
332                                               unsigned long address)
333 {
334         struct kvm *kvm = mmu_notifier_to_kvm(mn);
335         int young, idx;
336
337         idx = srcu_read_lock(&kvm->srcu);
338         spin_lock(&kvm->mmu_lock);
339         young = kvm_age_hva(kvm, address);
340         spin_unlock(&kvm->mmu_lock);
341         srcu_read_unlock(&kvm->srcu, idx);
342
343         if (young)
344                 kvm_flush_remote_tlbs(kvm);
345
346         return young;
347 }
348
349 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
350                                      struct mm_struct *mm)
351 {
352         struct kvm *kvm = mmu_notifier_to_kvm(mn);
353         int idx;
354
355         idx = srcu_read_lock(&kvm->srcu);
356         kvm_arch_flush_shadow(kvm);
357         srcu_read_unlock(&kvm->srcu, idx);
358 }
359
360 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
361         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
362         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
363         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
364         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
365         .change_pte             = kvm_mmu_notifier_change_pte,
366         .release                = kvm_mmu_notifier_release,
367 };
368
369 static int kvm_init_mmu_notifier(struct kvm *kvm)
370 {
371         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
372         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
373 }
374
375 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
376
377 static int kvm_init_mmu_notifier(struct kvm *kvm)
378 {
379         return 0;
380 }
381
382 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
383
384 static struct kvm *kvm_create_vm(void)
385 {
386         int r = 0, i;
387         struct kvm *kvm = kvm_arch_create_vm();
388
389         if (IS_ERR(kvm))
390                 goto out;
391
392         r = hardware_enable_all();
393         if (r)
394                 goto out_err_nodisable;
395
396 #ifdef CONFIG_HAVE_KVM_IRQCHIP
397         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
398         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
399 #endif
400
401         r = -ENOMEM;
402         kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
403         if (!kvm->memslots)
404                 goto out_err;
405         if (init_srcu_struct(&kvm->srcu))
406                 goto out_err;
407         for (i = 0; i < KVM_NR_BUSES; i++) {
408                 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
409                                         GFP_KERNEL);
410                 if (!kvm->buses[i]) {
411                         cleanup_srcu_struct(&kvm->srcu);
412                         goto out_err;
413                 }
414         }
415
416         r = kvm_init_mmu_notifier(kvm);
417         if (r) {
418                 cleanup_srcu_struct(&kvm->srcu);
419                 goto out_err;
420         }
421
422         kvm->mm = current->mm;
423         atomic_inc(&kvm->mm->mm_count);
424         spin_lock_init(&kvm->mmu_lock);
425         raw_spin_lock_init(&kvm->requests_lock);
426         kvm_eventfd_init(kvm);
427         mutex_init(&kvm->lock);
428         mutex_init(&kvm->irq_lock);
429         mutex_init(&kvm->slots_lock);
430         atomic_set(&kvm->users_count, 1);
431         spin_lock(&kvm_lock);
432         list_add(&kvm->vm_list, &vm_list);
433         spin_unlock(&kvm_lock);
434 out:
435         return kvm;
436
437 out_err:
438         hardware_disable_all();
439 out_err_nodisable:
440         for (i = 0; i < KVM_NR_BUSES; i++)
441                 kfree(kvm->buses[i]);
442         kfree(kvm->memslots);
443         kfree(kvm);
444         return ERR_PTR(r);
445 }
446
447 /*
448  * Free any memory in @free but not in @dont.
449  */
450 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
451                                   struct kvm_memory_slot *dont)
452 {
453         int i;
454
455         if (!dont || free->rmap != dont->rmap)
456                 vfree(free->rmap);
457
458         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
459                 vfree(free->dirty_bitmap);
460
461
462         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
463                 if (!dont || free->lpage_info[i] != dont->lpage_info[i]) {
464                         vfree(free->lpage_info[i]);
465                         free->lpage_info[i] = NULL;
466                 }
467         }
468
469         free->npages = 0;
470         free->dirty_bitmap = NULL;
471         free->rmap = NULL;
472 }
473
474 void kvm_free_physmem(struct kvm *kvm)
475 {
476         int i;
477         struct kvm_memslots *slots = kvm->memslots;
478
479         for (i = 0; i < slots->nmemslots; ++i)
480                 kvm_free_physmem_slot(&slots->memslots[i], NULL);
481
482         kfree(kvm->memslots);
483 }
484
485 static void kvm_destroy_vm(struct kvm *kvm)
486 {
487         int i;
488         struct mm_struct *mm = kvm->mm;
489
490         kvm_arch_sync_events(kvm);
491         spin_lock(&kvm_lock);
492         list_del(&kvm->vm_list);
493         spin_unlock(&kvm_lock);
494         kvm_free_irq_routing(kvm);
495         for (i = 0; i < KVM_NR_BUSES; i++)
496                 kvm_io_bus_destroy(kvm->buses[i]);
497         kvm_coalesced_mmio_free(kvm);
498 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
499         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
500 #else
501         kvm_arch_flush_shadow(kvm);
502 #endif
503         kvm_arch_destroy_vm(kvm);
504         hardware_disable_all();
505         mmdrop(mm);
506 }
507
508 void kvm_get_kvm(struct kvm *kvm)
509 {
510         atomic_inc(&kvm->users_count);
511 }
512 EXPORT_SYMBOL_GPL(kvm_get_kvm);
513
514 void kvm_put_kvm(struct kvm *kvm)
515 {
516         if (atomic_dec_and_test(&kvm->users_count))
517                 kvm_destroy_vm(kvm);
518 }
519 EXPORT_SYMBOL_GPL(kvm_put_kvm);
520
521
522 static int kvm_vm_release(struct inode *inode, struct file *filp)
523 {
524         struct kvm *kvm = filp->private_data;
525
526         kvm_irqfd_release(kvm);
527
528         kvm_put_kvm(kvm);
529         return 0;
530 }
531
532 /*
533  * Allocate some memory and give it an address in the guest physical address
534  * space.
535  *
536  * Discontiguous memory is allowed, mostly for framebuffers.
537  *
538  * Must be called holding mmap_sem for write.
539  */
540 int __kvm_set_memory_region(struct kvm *kvm,
541                             struct kvm_userspace_memory_region *mem,
542                             int user_alloc)
543 {
544         int r, flush_shadow = 0;
545         gfn_t base_gfn;
546         unsigned long npages;
547         unsigned long i;
548         struct kvm_memory_slot *memslot;
549         struct kvm_memory_slot old, new;
550         struct kvm_memslots *slots, *old_memslots;
551
552         r = -EINVAL;
553         /* General sanity checks */
554         if (mem->memory_size & (PAGE_SIZE - 1))
555                 goto out;
556         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
557                 goto out;
558         if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
559                 goto out;
560         if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
561                 goto out;
562         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
563                 goto out;
564
565         memslot = &kvm->memslots->memslots[mem->slot];
566         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
567         npages = mem->memory_size >> PAGE_SHIFT;
568
569         r = -EINVAL;
570         if (npages > KVM_MEM_MAX_NR_PAGES)
571                 goto out;
572
573         if (!npages)
574                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
575
576         new = old = *memslot;
577
578         new.id = mem->slot;
579         new.base_gfn = base_gfn;
580         new.npages = npages;
581         new.flags = mem->flags;
582
583         /* Disallow changing a memory slot's size. */
584         r = -EINVAL;
585         if (npages && old.npages && npages != old.npages)
586                 goto out_free;
587
588         /* Check for overlaps */
589         r = -EEXIST;
590         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
591                 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
592
593                 if (s == memslot || !s->npages)
594                         continue;
595                 if (!((base_gfn + npages <= s->base_gfn) ||
596                       (base_gfn >= s->base_gfn + s->npages)))
597                         goto out_free;
598         }
599
600         /* Free page dirty bitmap if unneeded */
601         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
602                 new.dirty_bitmap = NULL;
603
604         r = -ENOMEM;
605
606         /* Allocate if a slot is being created */
607 #ifndef CONFIG_S390
608         if (npages && !new.rmap) {
609                 new.rmap = vmalloc(npages * sizeof(*new.rmap));
610
611                 if (!new.rmap)
612                         goto out_free;
613
614                 memset(new.rmap, 0, npages * sizeof(*new.rmap));
615
616                 new.user_alloc = user_alloc;
617                 new.userspace_addr = mem->userspace_addr;
618         }
619         if (!npages)
620                 goto skip_lpage;
621
622         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
623                 unsigned long ugfn;
624                 unsigned long j;
625                 int lpages;
626                 int level = i + 2;
627
628                 /* Avoid unused variable warning if no large pages */
629                 (void)level;
630
631                 if (new.lpage_info[i])
632                         continue;
633
634                 lpages = 1 + ((base_gfn + npages - 1)
635                              >> KVM_HPAGE_GFN_SHIFT(level));
636                 lpages -= base_gfn >> KVM_HPAGE_GFN_SHIFT(level);
637
638                 new.lpage_info[i] = vmalloc(lpages * sizeof(*new.lpage_info[i]));
639
640                 if (!new.lpage_info[i])
641                         goto out_free;
642
643                 memset(new.lpage_info[i], 0,
644                        lpages * sizeof(*new.lpage_info[i]));
645
646                 if (base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
647                         new.lpage_info[i][0].write_count = 1;
648                 if ((base_gfn+npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
649                         new.lpage_info[i][lpages - 1].write_count = 1;
650                 ugfn = new.userspace_addr >> PAGE_SHIFT;
651                 /*
652                  * If the gfn and userspace address are not aligned wrt each
653                  * other, or if explicitly asked to, disable large page
654                  * support for this slot
655                  */
656                 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
657                     !largepages_enabled)
658                         for (j = 0; j < lpages; ++j)
659                                 new.lpage_info[i][j].write_count = 1;
660         }
661
662 skip_lpage:
663
664         /* Allocate page dirty bitmap if needed */
665         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
666                 unsigned long dirty_bytes = kvm_dirty_bitmap_bytes(&new);
667
668                 new.dirty_bitmap = vmalloc(dirty_bytes);
669                 if (!new.dirty_bitmap)
670                         goto out_free;
671                 memset(new.dirty_bitmap, 0, dirty_bytes);
672                 /* destroy any largepage mappings for dirty tracking */
673                 if (old.npages)
674                         flush_shadow = 1;
675         }
676 #else  /* not defined CONFIG_S390 */
677         new.user_alloc = user_alloc;
678         if (user_alloc)
679                 new.userspace_addr = mem->userspace_addr;
680 #endif /* not defined CONFIG_S390 */
681
682         if (!npages) {
683                 r = -ENOMEM;
684                 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
685                 if (!slots)
686                         goto out_free;
687                 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
688                 if (mem->slot >= slots->nmemslots)
689                         slots->nmemslots = mem->slot + 1;
690                 slots->generation++;
691                 slots->memslots[mem->slot].flags |= KVM_MEMSLOT_INVALID;
692
693                 old_memslots = kvm->memslots;
694                 rcu_assign_pointer(kvm->memslots, slots);
695                 synchronize_srcu_expedited(&kvm->srcu);
696                 /* From this point no new shadow pages pointing to a deleted
697                  * memslot will be created.
698                  *
699                  * validation of sp->gfn happens in:
700                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
701                  *      - kvm_is_visible_gfn (mmu_check_roots)
702                  */
703                 kvm_arch_flush_shadow(kvm);
704                 kfree(old_memslots);
705         }
706
707         r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
708         if (r)
709                 goto out_free;
710
711         /* map the pages in iommu page table */
712         if (npages) {
713                 r = kvm_iommu_map_pages(kvm, &new);
714                 if (r)
715                         goto out_free;
716         }
717
718         r = -ENOMEM;
719         slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
720         if (!slots)
721                 goto out_free;
722         memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
723         if (mem->slot >= slots->nmemslots)
724                 slots->nmemslots = mem->slot + 1;
725         slots->generation++;
726
727         /* actual memory is freed via old in kvm_free_physmem_slot below */
728         if (!npages) {
729                 new.rmap = NULL;
730                 new.dirty_bitmap = NULL;
731                 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i)
732                         new.lpage_info[i] = NULL;
733         }
734
735         slots->memslots[mem->slot] = new;
736         old_memslots = kvm->memslots;
737         rcu_assign_pointer(kvm->memslots, slots);
738         synchronize_srcu_expedited(&kvm->srcu);
739
740         kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
741
742         kvm_free_physmem_slot(&old, &new);
743         kfree(old_memslots);
744
745         if (flush_shadow)
746                 kvm_arch_flush_shadow(kvm);
747
748         return 0;
749
750 out_free:
751         kvm_free_physmem_slot(&new, &old);
752 out:
753         return r;
754
755 }
756 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
757
758 int kvm_set_memory_region(struct kvm *kvm,
759                           struct kvm_userspace_memory_region *mem,
760                           int user_alloc)
761 {
762         int r;
763
764         mutex_lock(&kvm->slots_lock);
765         r = __kvm_set_memory_region(kvm, mem, user_alloc);
766         mutex_unlock(&kvm->slots_lock);
767         return r;
768 }
769 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
770
771 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
772                                    struct
773                                    kvm_userspace_memory_region *mem,
774                                    int user_alloc)
775 {
776         if (mem->slot >= KVM_MEMORY_SLOTS)
777                 return -EINVAL;
778         return kvm_set_memory_region(kvm, mem, user_alloc);
779 }
780
781 int kvm_get_dirty_log(struct kvm *kvm,
782                         struct kvm_dirty_log *log, int *is_dirty)
783 {
784         struct kvm_memory_slot *memslot;
785         int r, i;
786         unsigned long n;
787         unsigned long any = 0;
788
789         r = -EINVAL;
790         if (log->slot >= KVM_MEMORY_SLOTS)
791                 goto out;
792
793         memslot = &kvm->memslots->memslots[log->slot];
794         r = -ENOENT;
795         if (!memslot->dirty_bitmap)
796                 goto out;
797
798         n = kvm_dirty_bitmap_bytes(memslot);
799
800         for (i = 0; !any && i < n/sizeof(long); ++i)
801                 any = memslot->dirty_bitmap[i];
802
803         r = -EFAULT;
804         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
805                 goto out;
806
807         if (any)
808                 *is_dirty = 1;
809
810         r = 0;
811 out:
812         return r;
813 }
814
815 void kvm_disable_largepages(void)
816 {
817         largepages_enabled = false;
818 }
819 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
820
821 int is_error_page(struct page *page)
822 {
823         return page == bad_page || page == hwpoison_page || page == fault_page;
824 }
825 EXPORT_SYMBOL_GPL(is_error_page);
826
827 int is_error_pfn(pfn_t pfn)
828 {
829         return pfn == bad_pfn || pfn == hwpoison_pfn || pfn == fault_pfn;
830 }
831 EXPORT_SYMBOL_GPL(is_error_pfn);
832
833 int is_hwpoison_pfn(pfn_t pfn)
834 {
835         return pfn == hwpoison_pfn;
836 }
837 EXPORT_SYMBOL_GPL(is_hwpoison_pfn);
838
839 int is_fault_pfn(pfn_t pfn)
840 {
841         return pfn == fault_pfn;
842 }
843 EXPORT_SYMBOL_GPL(is_fault_pfn);
844
845 static inline unsigned long bad_hva(void)
846 {
847         return PAGE_OFFSET;
848 }
849
850 int kvm_is_error_hva(unsigned long addr)
851 {
852         return addr == bad_hva();
853 }
854 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
855
856 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm_memslots *slots,
857                                                 gfn_t gfn)
858 {
859         int i;
860
861         for (i = 0; i < slots->nmemslots; ++i) {
862                 struct kvm_memory_slot *memslot = &slots->memslots[i];
863
864                 if (gfn >= memslot->base_gfn
865                     && gfn < memslot->base_gfn + memslot->npages)
866                         return memslot;
867         }
868         return NULL;
869 }
870
871 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
872 {
873         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
874 }
875 EXPORT_SYMBOL_GPL(gfn_to_memslot);
876
877 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
878 {
879         int i;
880         struct kvm_memslots *slots = kvm_memslots(kvm);
881
882         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
883                 struct kvm_memory_slot *memslot = &slots->memslots[i];
884
885                 if (memslot->flags & KVM_MEMSLOT_INVALID)
886                         continue;
887
888                 if (gfn >= memslot->base_gfn
889                     && gfn < memslot->base_gfn + memslot->npages)
890                         return 1;
891         }
892         return 0;
893 }
894 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
895
896 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
897 {
898         struct vm_area_struct *vma;
899         unsigned long addr, size;
900
901         size = PAGE_SIZE;
902
903         addr = gfn_to_hva(kvm, gfn);
904         if (kvm_is_error_hva(addr))
905                 return PAGE_SIZE;
906
907         down_read(&current->mm->mmap_sem);
908         vma = find_vma(current->mm, addr);
909         if (!vma)
910                 goto out;
911
912         size = vma_kernel_pagesize(vma);
913
914 out:
915         up_read(&current->mm->mmap_sem);
916
917         return size;
918 }
919
920 int memslot_id(struct kvm *kvm, gfn_t gfn)
921 {
922         int i;
923         struct kvm_memslots *slots = kvm_memslots(kvm);
924         struct kvm_memory_slot *memslot = NULL;
925
926         for (i = 0; i < slots->nmemslots; ++i) {
927                 memslot = &slots->memslots[i];
928
929                 if (gfn >= memslot->base_gfn
930                     && gfn < memslot->base_gfn + memslot->npages)
931                         break;
932         }
933
934         return memslot - slots->memslots;
935 }
936
937 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
938                                      gfn_t *nr_pages)
939 {
940         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
941                 return bad_hva();
942
943         if (nr_pages)
944                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
945
946         return gfn_to_hva_memslot(slot, gfn);
947 }
948
949 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
950 {
951         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
952 }
953 EXPORT_SYMBOL_GPL(gfn_to_hva);
954
955 static pfn_t get_fault_pfn(void)
956 {
957         get_page(fault_page);
958         return fault_pfn;
959 }
960
961 static pfn_t hva_to_pfn(struct kvm *kvm, unsigned long addr, bool atomic,
962                         bool *async)
963 {
964         struct page *page[1];
965         int npages = 0;
966         pfn_t pfn;
967
968         /* we can do it either atomically or asynchronously, not both */
969         BUG_ON(atomic && async);
970
971         if (atomic || async)
972                 npages = __get_user_pages_fast(addr, 1, 1, page);
973
974         if (unlikely(npages != 1) && !atomic) {
975                 might_sleep();
976                 npages = get_user_pages_fast(addr, 1, 1, page);
977         }
978
979         if (unlikely(npages != 1)) {
980                 struct vm_area_struct *vma;
981
982                 if (atomic)
983                         return get_fault_pfn();
984
985                 down_read(&current->mm->mmap_sem);
986                 if (is_hwpoison_address(addr)) {
987                         up_read(&current->mm->mmap_sem);
988                         get_page(hwpoison_page);
989                         return page_to_pfn(hwpoison_page);
990                 }
991
992                 vma = find_vma_intersection(current->mm, addr, addr+1);
993
994                 if (vma == NULL)
995                         pfn = get_fault_pfn();
996                 else if ((vma->vm_flags & VM_PFNMAP)) {
997                         pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
998                                 vma->vm_pgoff;
999                         BUG_ON(!kvm_is_mmio_pfn(pfn));
1000                 } else {
1001                         if (async && (vma->vm_flags & VM_WRITE))
1002                                 *async = true;
1003                         pfn = get_fault_pfn();
1004                 }
1005                 up_read(&current->mm->mmap_sem);
1006         } else
1007                 pfn = page_to_pfn(page[0]);
1008
1009         return pfn;
1010 }
1011
1012 pfn_t hva_to_pfn_atomic(struct kvm *kvm, unsigned long addr)
1013 {
1014         return hva_to_pfn(kvm, addr, true, NULL);
1015 }
1016 EXPORT_SYMBOL_GPL(hva_to_pfn_atomic);
1017
1018 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async)
1019 {
1020         unsigned long addr;
1021
1022         if (async)
1023                 *async = false;
1024
1025         addr = gfn_to_hva(kvm, gfn);
1026         if (kvm_is_error_hva(addr)) {
1027                 get_page(bad_page);
1028                 return page_to_pfn(bad_page);
1029         }
1030
1031         return hva_to_pfn(kvm, addr, atomic, async);
1032 }
1033
1034 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1035 {
1036         return __gfn_to_pfn(kvm, gfn, true, NULL);
1037 }
1038 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1039
1040 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async)
1041 {
1042         return __gfn_to_pfn(kvm, gfn, false, async);
1043 }
1044 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1045
1046 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1047 {
1048         return __gfn_to_pfn(kvm, gfn, false, NULL);
1049 }
1050 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1051
1052 pfn_t gfn_to_pfn_memslot(struct kvm *kvm,
1053                          struct kvm_memory_slot *slot, gfn_t gfn)
1054 {
1055         unsigned long addr = gfn_to_hva_memslot(slot, gfn);
1056         return hva_to_pfn(kvm, addr, false, NULL);
1057 }
1058
1059 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1060                                                                   int nr_pages)
1061 {
1062         unsigned long addr;
1063         gfn_t entry;
1064
1065         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1066         if (kvm_is_error_hva(addr))
1067                 return -1;
1068
1069         if (entry < nr_pages)
1070                 return 0;
1071
1072         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1073 }
1074 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1075
1076 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1077 {
1078         pfn_t pfn;
1079
1080         pfn = gfn_to_pfn(kvm, gfn);
1081         if (!kvm_is_mmio_pfn(pfn))
1082                 return pfn_to_page(pfn);
1083
1084         WARN_ON(kvm_is_mmio_pfn(pfn));
1085
1086         get_page(bad_page);
1087         return bad_page;
1088 }
1089
1090 EXPORT_SYMBOL_GPL(gfn_to_page);
1091
1092 void kvm_release_page_clean(struct page *page)
1093 {
1094         kvm_release_pfn_clean(page_to_pfn(page));
1095 }
1096 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1097
1098 void kvm_release_pfn_clean(pfn_t pfn)
1099 {
1100         if (!kvm_is_mmio_pfn(pfn))
1101                 put_page(pfn_to_page(pfn));
1102 }
1103 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1104
1105 void kvm_release_page_dirty(struct page *page)
1106 {
1107         kvm_release_pfn_dirty(page_to_pfn(page));
1108 }
1109 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1110
1111 void kvm_release_pfn_dirty(pfn_t pfn)
1112 {
1113         kvm_set_pfn_dirty(pfn);
1114         kvm_release_pfn_clean(pfn);
1115 }
1116 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1117
1118 void kvm_set_page_dirty(struct page *page)
1119 {
1120         kvm_set_pfn_dirty(page_to_pfn(page));
1121 }
1122 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1123
1124 void kvm_set_pfn_dirty(pfn_t pfn)
1125 {
1126         if (!kvm_is_mmio_pfn(pfn)) {
1127                 struct page *page = pfn_to_page(pfn);
1128                 if (!PageReserved(page))
1129                         SetPageDirty(page);
1130         }
1131 }
1132 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1133
1134 void kvm_set_pfn_accessed(pfn_t pfn)
1135 {
1136         if (!kvm_is_mmio_pfn(pfn))
1137                 mark_page_accessed(pfn_to_page(pfn));
1138 }
1139 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1140
1141 void kvm_get_pfn(pfn_t pfn)
1142 {
1143         if (!kvm_is_mmio_pfn(pfn))
1144                 get_page(pfn_to_page(pfn));
1145 }
1146 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1147
1148 static int next_segment(unsigned long len, int offset)
1149 {
1150         if (len > PAGE_SIZE - offset)
1151                 return PAGE_SIZE - offset;
1152         else
1153                 return len;
1154 }
1155
1156 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1157                         int len)
1158 {
1159         int r;
1160         unsigned long addr;
1161
1162         addr = gfn_to_hva(kvm, gfn);
1163         if (kvm_is_error_hva(addr))
1164                 return -EFAULT;
1165         r = copy_from_user(data, (void __user *)addr + offset, len);
1166         if (r)
1167                 return -EFAULT;
1168         return 0;
1169 }
1170 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1171
1172 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1173 {
1174         gfn_t gfn = gpa >> PAGE_SHIFT;
1175         int seg;
1176         int offset = offset_in_page(gpa);
1177         int ret;
1178
1179         while ((seg = next_segment(len, offset)) != 0) {
1180                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1181                 if (ret < 0)
1182                         return ret;
1183                 offset = 0;
1184                 len -= seg;
1185                 data += seg;
1186                 ++gfn;
1187         }
1188         return 0;
1189 }
1190 EXPORT_SYMBOL_GPL(kvm_read_guest);
1191
1192 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1193                           unsigned long len)
1194 {
1195         int r;
1196         unsigned long addr;
1197         gfn_t gfn = gpa >> PAGE_SHIFT;
1198         int offset = offset_in_page(gpa);
1199
1200         addr = gfn_to_hva(kvm, gfn);
1201         if (kvm_is_error_hva(addr))
1202                 return -EFAULT;
1203         pagefault_disable();
1204         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1205         pagefault_enable();
1206         if (r)
1207                 return -EFAULT;
1208         return 0;
1209 }
1210 EXPORT_SYMBOL(kvm_read_guest_atomic);
1211
1212 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1213                          int offset, int len)
1214 {
1215         int r;
1216         unsigned long addr;
1217
1218         addr = gfn_to_hva(kvm, gfn);
1219         if (kvm_is_error_hva(addr))
1220                 return -EFAULT;
1221         r = copy_to_user((void __user *)addr + offset, data, len);
1222         if (r)
1223                 return -EFAULT;
1224         mark_page_dirty(kvm, gfn);
1225         return 0;
1226 }
1227 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1228
1229 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1230                     unsigned long len)
1231 {
1232         gfn_t gfn = gpa >> PAGE_SHIFT;
1233         int seg;
1234         int offset = offset_in_page(gpa);
1235         int ret;
1236
1237         while ((seg = next_segment(len, offset)) != 0) {
1238                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1239                 if (ret < 0)
1240                         return ret;
1241                 offset = 0;
1242                 len -= seg;
1243                 data += seg;
1244                 ++gfn;
1245         }
1246         return 0;
1247 }
1248
1249 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1250                               gpa_t gpa)
1251 {
1252         struct kvm_memslots *slots = kvm_memslots(kvm);
1253         int offset = offset_in_page(gpa);
1254         gfn_t gfn = gpa >> PAGE_SHIFT;
1255
1256         ghc->gpa = gpa;
1257         ghc->generation = slots->generation;
1258         ghc->memslot = __gfn_to_memslot(slots, gfn);
1259         ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1260         if (!kvm_is_error_hva(ghc->hva))
1261                 ghc->hva += offset;
1262         else
1263                 return -EFAULT;
1264
1265         return 0;
1266 }
1267 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1268
1269 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1270                            void *data, unsigned long len)
1271 {
1272         struct kvm_memslots *slots = kvm_memslots(kvm);
1273         int r;
1274
1275         if (slots->generation != ghc->generation)
1276                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1277
1278         if (kvm_is_error_hva(ghc->hva))
1279                 return -EFAULT;
1280
1281         r = copy_to_user((void __user *)ghc->hva, data, len);
1282         if (r)
1283                 return -EFAULT;
1284         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1285
1286         return 0;
1287 }
1288 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1289
1290 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1291 {
1292         return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
1293 }
1294 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1295
1296 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1297 {
1298         gfn_t gfn = gpa >> PAGE_SHIFT;
1299         int seg;
1300         int offset = offset_in_page(gpa);
1301         int ret;
1302
1303         while ((seg = next_segment(len, offset)) != 0) {
1304                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1305                 if (ret < 0)
1306                         return ret;
1307                 offset = 0;
1308                 len -= seg;
1309                 ++gfn;
1310         }
1311         return 0;
1312 }
1313 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1314
1315 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1316                              gfn_t gfn)
1317 {
1318         if (memslot && memslot->dirty_bitmap) {
1319                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1320
1321                 generic___set_le_bit(rel_gfn, memslot->dirty_bitmap);
1322         }
1323 }
1324
1325 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1326 {
1327         struct kvm_memory_slot *memslot;
1328
1329         memslot = gfn_to_memslot(kvm, gfn);
1330         mark_page_dirty_in_slot(kvm, memslot, gfn);
1331 }
1332
1333 /*
1334  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1335  */
1336 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1337 {
1338         DEFINE_WAIT(wait);
1339
1340         for (;;) {
1341                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1342
1343                 if (kvm_arch_vcpu_runnable(vcpu)) {
1344                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1345                         break;
1346                 }
1347                 if (kvm_cpu_has_pending_timer(vcpu))
1348                         break;
1349                 if (signal_pending(current))
1350                         break;
1351
1352                 schedule();
1353         }
1354
1355         finish_wait(&vcpu->wq, &wait);
1356 }
1357
1358 void kvm_resched(struct kvm_vcpu *vcpu)
1359 {
1360         if (!need_resched())
1361                 return;
1362         cond_resched();
1363 }
1364 EXPORT_SYMBOL_GPL(kvm_resched);
1365
1366 void kvm_vcpu_on_spin(struct kvm_vcpu *vcpu)
1367 {
1368         ktime_t expires;
1369         DEFINE_WAIT(wait);
1370
1371         prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1372
1373         /* Sleep for 100 us, and hope lock-holder got scheduled */
1374         expires = ktime_add_ns(ktime_get(), 100000UL);
1375         schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
1376
1377         finish_wait(&vcpu->wq, &wait);
1378 }
1379 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1380
1381 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1382 {
1383         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1384         struct page *page;
1385
1386         if (vmf->pgoff == 0)
1387                 page = virt_to_page(vcpu->run);
1388 #ifdef CONFIG_X86
1389         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1390                 page = virt_to_page(vcpu->arch.pio_data);
1391 #endif
1392 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1393         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1394                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1395 #endif
1396         else
1397                 return VM_FAULT_SIGBUS;
1398         get_page(page);
1399         vmf->page = page;
1400         return 0;
1401 }
1402
1403 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1404         .fault = kvm_vcpu_fault,
1405 };
1406
1407 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1408 {
1409         vma->vm_ops = &kvm_vcpu_vm_ops;
1410         return 0;
1411 }
1412
1413 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1414 {
1415         struct kvm_vcpu *vcpu = filp->private_data;
1416
1417         kvm_put_kvm(vcpu->kvm);
1418         return 0;
1419 }
1420
1421 static struct file_operations kvm_vcpu_fops = {
1422         .release        = kvm_vcpu_release,
1423         .unlocked_ioctl = kvm_vcpu_ioctl,
1424         .compat_ioctl   = kvm_vcpu_ioctl,
1425         .mmap           = kvm_vcpu_mmap,
1426         .llseek         = noop_llseek,
1427 };
1428
1429 /*
1430  * Allocates an inode for the vcpu.
1431  */
1432 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1433 {
1434         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1435 }
1436
1437 /*
1438  * Creates some virtual cpus.  Good luck creating more than one.
1439  */
1440 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1441 {
1442         int r;
1443         struct kvm_vcpu *vcpu, *v;
1444
1445         vcpu = kvm_arch_vcpu_create(kvm, id);
1446         if (IS_ERR(vcpu))
1447                 return PTR_ERR(vcpu);
1448
1449         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1450
1451         r = kvm_arch_vcpu_setup(vcpu);
1452         if (r)
1453                 return r;
1454
1455         mutex_lock(&kvm->lock);
1456         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1457                 r = -EINVAL;
1458                 goto vcpu_destroy;
1459         }
1460
1461         kvm_for_each_vcpu(r, v, kvm)
1462                 if (v->vcpu_id == id) {
1463                         r = -EEXIST;
1464                         goto vcpu_destroy;
1465                 }
1466
1467         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1468
1469         /* Now it's all set up, let userspace reach it */
1470         kvm_get_kvm(kvm);
1471         r = create_vcpu_fd(vcpu);
1472         if (r < 0) {
1473                 kvm_put_kvm(kvm);
1474                 goto vcpu_destroy;
1475         }
1476
1477         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1478         smp_wmb();
1479         atomic_inc(&kvm->online_vcpus);
1480
1481 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1482         if (kvm->bsp_vcpu_id == id)
1483                 kvm->bsp_vcpu = vcpu;
1484 #endif
1485         mutex_unlock(&kvm->lock);
1486         return r;
1487
1488 vcpu_destroy:
1489         mutex_unlock(&kvm->lock);
1490         kvm_arch_vcpu_destroy(vcpu);
1491         return r;
1492 }
1493
1494 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1495 {
1496         if (sigset) {
1497                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1498                 vcpu->sigset_active = 1;
1499                 vcpu->sigset = *sigset;
1500         } else
1501                 vcpu->sigset_active = 0;
1502         return 0;
1503 }
1504
1505 static long kvm_vcpu_ioctl(struct file *filp,
1506                            unsigned int ioctl, unsigned long arg)
1507 {
1508         struct kvm_vcpu *vcpu = filp->private_data;
1509         void __user *argp = (void __user *)arg;
1510         int r;
1511         struct kvm_fpu *fpu = NULL;
1512         struct kvm_sregs *kvm_sregs = NULL;
1513
1514         if (vcpu->kvm->mm != current->mm)
1515                 return -EIO;
1516
1517 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1518         /*
1519          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1520          * so vcpu_load() would break it.
1521          */
1522         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1523                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1524 #endif
1525
1526
1527         vcpu_load(vcpu);
1528         switch (ioctl) {
1529         case KVM_RUN:
1530                 r = -EINVAL;
1531                 if (arg)
1532                         goto out;
1533                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1534                 break;
1535         case KVM_GET_REGS: {
1536                 struct kvm_regs *kvm_regs;
1537
1538                 r = -ENOMEM;
1539                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1540                 if (!kvm_regs)
1541                         goto out;
1542                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1543                 if (r)
1544                         goto out_free1;
1545                 r = -EFAULT;
1546                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1547                         goto out_free1;
1548                 r = 0;
1549 out_free1:
1550                 kfree(kvm_regs);
1551                 break;
1552         }
1553         case KVM_SET_REGS: {
1554                 struct kvm_regs *kvm_regs;
1555
1556                 r = -ENOMEM;
1557                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1558                 if (!kvm_regs)
1559                         goto out;
1560                 r = -EFAULT;
1561                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1562                         goto out_free2;
1563                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1564                 if (r)
1565                         goto out_free2;
1566                 r = 0;
1567 out_free2:
1568                 kfree(kvm_regs);
1569                 break;
1570         }
1571         case KVM_GET_SREGS: {
1572                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1573                 r = -ENOMEM;
1574                 if (!kvm_sregs)
1575                         goto out;
1576                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1577                 if (r)
1578                         goto out;
1579                 r = -EFAULT;
1580                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1581                         goto out;
1582                 r = 0;
1583                 break;
1584         }
1585         case KVM_SET_SREGS: {
1586                 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1587                 r = -ENOMEM;
1588                 if (!kvm_sregs)
1589                         goto out;
1590                 r = -EFAULT;
1591                 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1592                         goto out;
1593                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1594                 if (r)
1595                         goto out;
1596                 r = 0;
1597                 break;
1598         }
1599         case KVM_GET_MP_STATE: {
1600                 struct kvm_mp_state mp_state;
1601
1602                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1603                 if (r)
1604                         goto out;
1605                 r = -EFAULT;
1606                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1607                         goto out;
1608                 r = 0;
1609                 break;
1610         }
1611         case KVM_SET_MP_STATE: {
1612                 struct kvm_mp_state mp_state;
1613
1614                 r = -EFAULT;
1615                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1616                         goto out;
1617                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1618                 if (r)
1619                         goto out;
1620                 r = 0;
1621                 break;
1622         }
1623         case KVM_TRANSLATE: {
1624                 struct kvm_translation tr;
1625
1626                 r = -EFAULT;
1627                 if (copy_from_user(&tr, argp, sizeof tr))
1628                         goto out;
1629                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1630                 if (r)
1631                         goto out;
1632                 r = -EFAULT;
1633                 if (copy_to_user(argp, &tr, sizeof tr))
1634                         goto out;
1635                 r = 0;
1636                 break;
1637         }
1638         case KVM_SET_GUEST_DEBUG: {
1639                 struct kvm_guest_debug dbg;
1640
1641                 r = -EFAULT;
1642                 if (copy_from_user(&dbg, argp, sizeof dbg))
1643                         goto out;
1644                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1645                 if (r)
1646                         goto out;
1647                 r = 0;
1648                 break;
1649         }
1650         case KVM_SET_SIGNAL_MASK: {
1651                 struct kvm_signal_mask __user *sigmask_arg = argp;
1652                 struct kvm_signal_mask kvm_sigmask;
1653                 sigset_t sigset, *p;
1654
1655                 p = NULL;
1656                 if (argp) {
1657                         r = -EFAULT;
1658                         if (copy_from_user(&kvm_sigmask, argp,
1659                                            sizeof kvm_sigmask))
1660                                 goto out;
1661                         r = -EINVAL;
1662                         if (kvm_sigmask.len != sizeof sigset)
1663                                 goto out;
1664                         r = -EFAULT;
1665                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1666                                            sizeof sigset))
1667                                 goto out;
1668                         p = &sigset;
1669                 }
1670                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
1671                 break;
1672         }
1673         case KVM_GET_FPU: {
1674                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1675                 r = -ENOMEM;
1676                 if (!fpu)
1677                         goto out;
1678                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1679                 if (r)
1680                         goto out;
1681                 r = -EFAULT;
1682                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1683                         goto out;
1684                 r = 0;
1685                 break;
1686         }
1687         case KVM_SET_FPU: {
1688                 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1689                 r = -ENOMEM;
1690                 if (!fpu)
1691                         goto out;
1692                 r = -EFAULT;
1693                 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1694                         goto out;
1695                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1696                 if (r)
1697                         goto out;
1698                 r = 0;
1699                 break;
1700         }
1701         default:
1702                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1703         }
1704 out:
1705         vcpu_put(vcpu);
1706         kfree(fpu);
1707         kfree(kvm_sregs);
1708         return r;
1709 }
1710
1711 static long kvm_vm_ioctl(struct file *filp,
1712                            unsigned int ioctl, unsigned long arg)
1713 {
1714         struct kvm *kvm = filp->private_data;
1715         void __user *argp = (void __user *)arg;
1716         int r;
1717
1718         if (kvm->mm != current->mm)
1719                 return -EIO;
1720         switch (ioctl) {
1721         case KVM_CREATE_VCPU:
1722                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1723                 if (r < 0)
1724                         goto out;
1725                 break;
1726         case KVM_SET_USER_MEMORY_REGION: {
1727                 struct kvm_userspace_memory_region kvm_userspace_mem;
1728
1729                 r = -EFAULT;
1730                 if (copy_from_user(&kvm_userspace_mem, argp,
1731                                                 sizeof kvm_userspace_mem))
1732                         goto out;
1733
1734                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1735                 if (r)
1736                         goto out;
1737                 break;
1738         }
1739         case KVM_GET_DIRTY_LOG: {
1740                 struct kvm_dirty_log log;
1741
1742                 r = -EFAULT;
1743                 if (copy_from_user(&log, argp, sizeof log))
1744                         goto out;
1745                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1746                 if (r)
1747                         goto out;
1748                 break;
1749         }
1750 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1751         case KVM_REGISTER_COALESCED_MMIO: {
1752                 struct kvm_coalesced_mmio_zone zone;
1753                 r = -EFAULT;
1754                 if (copy_from_user(&zone, argp, sizeof zone))
1755                         goto out;
1756                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1757                 if (r)
1758                         goto out;
1759                 r = 0;
1760                 break;
1761         }
1762         case KVM_UNREGISTER_COALESCED_MMIO: {
1763                 struct kvm_coalesced_mmio_zone zone;
1764                 r = -EFAULT;
1765                 if (copy_from_user(&zone, argp, sizeof zone))
1766                         goto out;
1767                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1768                 if (r)
1769                         goto out;
1770                 r = 0;
1771                 break;
1772         }
1773 #endif
1774         case KVM_IRQFD: {
1775                 struct kvm_irqfd data;
1776
1777                 r = -EFAULT;
1778                 if (copy_from_user(&data, argp, sizeof data))
1779                         goto out;
1780                 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
1781                 break;
1782         }
1783         case KVM_IOEVENTFD: {
1784                 struct kvm_ioeventfd data;
1785
1786                 r = -EFAULT;
1787                 if (copy_from_user(&data, argp, sizeof data))
1788                         goto out;
1789                 r = kvm_ioeventfd(kvm, &data);
1790                 break;
1791         }
1792 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1793         case KVM_SET_BOOT_CPU_ID:
1794                 r = 0;
1795                 mutex_lock(&kvm->lock);
1796                 if (atomic_read(&kvm->online_vcpus) != 0)
1797                         r = -EBUSY;
1798                 else
1799                         kvm->bsp_vcpu_id = arg;
1800                 mutex_unlock(&kvm->lock);
1801                 break;
1802 #endif
1803         default:
1804                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1805                 if (r == -ENOTTY)
1806                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
1807         }
1808 out:
1809         return r;
1810 }
1811
1812 #ifdef CONFIG_COMPAT
1813 struct compat_kvm_dirty_log {
1814         __u32 slot;
1815         __u32 padding1;
1816         union {
1817                 compat_uptr_t dirty_bitmap; /* one bit per page */
1818                 __u64 padding2;
1819         };
1820 };
1821
1822 static long kvm_vm_compat_ioctl(struct file *filp,
1823                            unsigned int ioctl, unsigned long arg)
1824 {
1825         struct kvm *kvm = filp->private_data;
1826         int r;
1827
1828         if (kvm->mm != current->mm)
1829                 return -EIO;
1830         switch (ioctl) {
1831         case KVM_GET_DIRTY_LOG: {
1832                 struct compat_kvm_dirty_log compat_log;
1833                 struct kvm_dirty_log log;
1834
1835                 r = -EFAULT;
1836                 if (copy_from_user(&compat_log, (void __user *)arg,
1837                                    sizeof(compat_log)))
1838                         goto out;
1839                 log.slot         = compat_log.slot;
1840                 log.padding1     = compat_log.padding1;
1841                 log.padding2     = compat_log.padding2;
1842                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
1843
1844                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1845                 if (r)
1846                         goto out;
1847                 break;
1848         }
1849         default:
1850                 r = kvm_vm_ioctl(filp, ioctl, arg);
1851         }
1852
1853 out:
1854         return r;
1855 }
1856 #endif
1857
1858 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1859 {
1860         struct page *page[1];
1861         unsigned long addr;
1862         int npages;
1863         gfn_t gfn = vmf->pgoff;
1864         struct kvm *kvm = vma->vm_file->private_data;
1865
1866         addr = gfn_to_hva(kvm, gfn);
1867         if (kvm_is_error_hva(addr))
1868                 return VM_FAULT_SIGBUS;
1869
1870         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
1871                                 NULL);
1872         if (unlikely(npages != 1))
1873                 return VM_FAULT_SIGBUS;
1874
1875         vmf->page = page[0];
1876         return 0;
1877 }
1878
1879 static const struct vm_operations_struct kvm_vm_vm_ops = {
1880         .fault = kvm_vm_fault,
1881 };
1882
1883 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1884 {
1885         vma->vm_ops = &kvm_vm_vm_ops;
1886         return 0;
1887 }
1888
1889 static struct file_operations kvm_vm_fops = {
1890         .release        = kvm_vm_release,
1891         .unlocked_ioctl = kvm_vm_ioctl,
1892 #ifdef CONFIG_COMPAT
1893         .compat_ioctl   = kvm_vm_compat_ioctl,
1894 #endif
1895         .mmap           = kvm_vm_mmap,
1896         .llseek         = noop_llseek,
1897 };
1898
1899 static int kvm_dev_ioctl_create_vm(void)
1900 {
1901         int fd, r;
1902         struct kvm *kvm;
1903
1904         kvm = kvm_create_vm();
1905         if (IS_ERR(kvm))
1906                 return PTR_ERR(kvm);
1907 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1908         r = kvm_coalesced_mmio_init(kvm);
1909         if (r < 0) {
1910                 kvm_put_kvm(kvm);
1911                 return r;
1912         }
1913 #endif
1914         fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
1915         if (fd < 0)
1916                 kvm_put_kvm(kvm);
1917
1918         return fd;
1919 }
1920
1921 static long kvm_dev_ioctl_check_extension_generic(long arg)
1922 {
1923         switch (arg) {
1924         case KVM_CAP_USER_MEMORY:
1925         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
1926         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
1927 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1928         case KVM_CAP_SET_BOOT_CPU_ID:
1929 #endif
1930         case KVM_CAP_INTERNAL_ERROR_DATA:
1931                 return 1;
1932 #ifdef CONFIG_HAVE_KVM_IRQCHIP
1933         case KVM_CAP_IRQ_ROUTING:
1934                 return KVM_MAX_IRQ_ROUTES;
1935 #endif
1936         default:
1937                 break;
1938         }
1939         return kvm_dev_ioctl_check_extension(arg);
1940 }
1941
1942 static long kvm_dev_ioctl(struct file *filp,
1943                           unsigned int ioctl, unsigned long arg)
1944 {
1945         long r = -EINVAL;
1946
1947         switch (ioctl) {
1948         case KVM_GET_API_VERSION:
1949                 r = -EINVAL;
1950                 if (arg)
1951                         goto out;
1952                 r = KVM_API_VERSION;
1953                 break;
1954         case KVM_CREATE_VM:
1955                 r = -EINVAL;
1956                 if (arg)
1957                         goto out;
1958                 r = kvm_dev_ioctl_create_vm();
1959                 break;
1960         case KVM_CHECK_EXTENSION:
1961                 r = kvm_dev_ioctl_check_extension_generic(arg);
1962                 break;
1963         case KVM_GET_VCPU_MMAP_SIZE:
1964                 r = -EINVAL;
1965                 if (arg)
1966                         goto out;
1967                 r = PAGE_SIZE;     /* struct kvm_run */
1968 #ifdef CONFIG_X86
1969                 r += PAGE_SIZE;    /* pio data page */
1970 #endif
1971 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1972                 r += PAGE_SIZE;    /* coalesced mmio ring page */
1973 #endif
1974                 break;
1975         case KVM_TRACE_ENABLE:
1976         case KVM_TRACE_PAUSE:
1977         case KVM_TRACE_DISABLE:
1978                 r = -EOPNOTSUPP;
1979                 break;
1980         default:
1981                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
1982         }
1983 out:
1984         return r;
1985 }
1986
1987 static struct file_operations kvm_chardev_ops = {
1988         .unlocked_ioctl = kvm_dev_ioctl,
1989         .compat_ioctl   = kvm_dev_ioctl,
1990         .llseek         = noop_llseek,
1991 };
1992
1993 static struct miscdevice kvm_dev = {
1994         KVM_MINOR,
1995         "kvm",
1996         &kvm_chardev_ops,
1997 };
1998
1999 static void hardware_enable(void *junk)
2000 {
2001         int cpu = raw_smp_processor_id();
2002         int r;
2003
2004         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2005                 return;
2006
2007         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2008
2009         r = kvm_arch_hardware_enable(NULL);
2010
2011         if (r) {
2012                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2013                 atomic_inc(&hardware_enable_failed);
2014                 printk(KERN_INFO "kvm: enabling virtualization on "
2015                                  "CPU%d failed\n", cpu);
2016         }
2017 }
2018
2019 static void hardware_disable(void *junk)
2020 {
2021         int cpu = raw_smp_processor_id();
2022
2023         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2024                 return;
2025         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2026         kvm_arch_hardware_disable(NULL);
2027 }
2028
2029 static void hardware_disable_all_nolock(void)
2030 {
2031         BUG_ON(!kvm_usage_count);
2032
2033         kvm_usage_count--;
2034         if (!kvm_usage_count)
2035                 on_each_cpu(hardware_disable, NULL, 1);
2036 }
2037
2038 static void hardware_disable_all(void)
2039 {
2040         spin_lock(&kvm_lock);
2041         hardware_disable_all_nolock();
2042         spin_unlock(&kvm_lock);
2043 }
2044
2045 static int hardware_enable_all(void)
2046 {
2047         int r = 0;
2048
2049         spin_lock(&kvm_lock);
2050
2051         kvm_usage_count++;
2052         if (kvm_usage_count == 1) {
2053                 atomic_set(&hardware_enable_failed, 0);
2054                 on_each_cpu(hardware_enable, NULL, 1);
2055
2056                 if (atomic_read(&hardware_enable_failed)) {
2057                         hardware_disable_all_nolock();
2058                         r = -EBUSY;
2059                 }
2060         }
2061
2062         spin_unlock(&kvm_lock);
2063
2064         return r;
2065 }
2066
2067 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2068                            void *v)
2069 {
2070         int cpu = (long)v;
2071
2072         if (!kvm_usage_count)
2073                 return NOTIFY_OK;
2074
2075         val &= ~CPU_TASKS_FROZEN;
2076         switch (val) {
2077         case CPU_DYING:
2078                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2079                        cpu);
2080                 hardware_disable(NULL);
2081                 break;
2082         case CPU_STARTING:
2083                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2084                        cpu);
2085                 spin_lock(&kvm_lock);
2086                 hardware_enable(NULL);
2087                 spin_unlock(&kvm_lock);
2088                 break;
2089         }
2090         return NOTIFY_OK;
2091 }
2092
2093
2094 asmlinkage void kvm_handle_fault_on_reboot(void)
2095 {
2096         if (kvm_rebooting) {
2097                 /* spin while reset goes on */
2098                 local_irq_enable();
2099                 while (true)
2100                         cpu_relax();
2101         }
2102         /* Fault while not rebooting.  We want the trace. */
2103         BUG();
2104 }
2105 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
2106
2107 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2108                       void *v)
2109 {
2110         /*
2111          * Some (well, at least mine) BIOSes hang on reboot if
2112          * in vmx root mode.
2113          *
2114          * And Intel TXT required VMX off for all cpu when system shutdown.
2115          */
2116         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2117         kvm_rebooting = true;
2118         on_each_cpu(hardware_disable, NULL, 1);
2119         return NOTIFY_OK;
2120 }
2121
2122 static struct notifier_block kvm_reboot_notifier = {
2123         .notifier_call = kvm_reboot,
2124         .priority = 0,
2125 };
2126
2127 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2128 {
2129         int i;
2130
2131         for (i = 0; i < bus->dev_count; i++) {
2132                 struct kvm_io_device *pos = bus->devs[i];
2133
2134                 kvm_iodevice_destructor(pos);
2135         }
2136         kfree(bus);
2137 }
2138
2139 /* kvm_io_bus_write - called under kvm->slots_lock */
2140 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2141                      int len, const void *val)
2142 {
2143         int i;
2144         struct kvm_io_bus *bus;
2145
2146         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2147         for (i = 0; i < bus->dev_count; i++)
2148                 if (!kvm_iodevice_write(bus->devs[i], addr, len, val))
2149                         return 0;
2150         return -EOPNOTSUPP;
2151 }
2152
2153 /* kvm_io_bus_read - called under kvm->slots_lock */
2154 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2155                     int len, void *val)
2156 {
2157         int i;
2158         struct kvm_io_bus *bus;
2159
2160         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2161         for (i = 0; i < bus->dev_count; i++)
2162                 if (!kvm_iodevice_read(bus->devs[i], addr, len, val))
2163                         return 0;
2164         return -EOPNOTSUPP;
2165 }
2166
2167 /* Caller must hold slots_lock. */
2168 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2169                             struct kvm_io_device *dev)
2170 {
2171         struct kvm_io_bus *new_bus, *bus;
2172
2173         bus = kvm->buses[bus_idx];
2174         if (bus->dev_count > NR_IOBUS_DEVS-1)
2175                 return -ENOSPC;
2176
2177         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2178         if (!new_bus)
2179                 return -ENOMEM;
2180         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2181         new_bus->devs[new_bus->dev_count++] = dev;
2182         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2183         synchronize_srcu_expedited(&kvm->srcu);
2184         kfree(bus);
2185
2186         return 0;
2187 }
2188
2189 /* Caller must hold slots_lock. */
2190 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2191                               struct kvm_io_device *dev)
2192 {
2193         int i, r;
2194         struct kvm_io_bus *new_bus, *bus;
2195
2196         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2197         if (!new_bus)
2198                 return -ENOMEM;
2199
2200         bus = kvm->buses[bus_idx];
2201         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2202
2203         r = -ENOENT;
2204         for (i = 0; i < new_bus->dev_count; i++)
2205                 if (new_bus->devs[i] == dev) {
2206                         r = 0;
2207                         new_bus->devs[i] = new_bus->devs[--new_bus->dev_count];
2208                         break;
2209                 }
2210
2211         if (r) {
2212                 kfree(new_bus);
2213                 return r;
2214         }
2215
2216         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2217         synchronize_srcu_expedited(&kvm->srcu);
2218         kfree(bus);
2219         return r;
2220 }
2221
2222 static struct notifier_block kvm_cpu_notifier = {
2223         .notifier_call = kvm_cpu_hotplug,
2224 };
2225
2226 static int vm_stat_get(void *_offset, u64 *val)
2227 {
2228         unsigned offset = (long)_offset;
2229         struct kvm *kvm;
2230
2231         *val = 0;
2232         spin_lock(&kvm_lock);
2233         list_for_each_entry(kvm, &vm_list, vm_list)
2234                 *val += *(u32 *)((void *)kvm + offset);
2235         spin_unlock(&kvm_lock);
2236         return 0;
2237 }
2238
2239 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2240
2241 static int vcpu_stat_get(void *_offset, u64 *val)
2242 {
2243         unsigned offset = (long)_offset;
2244         struct kvm *kvm;
2245         struct kvm_vcpu *vcpu;
2246         int i;
2247
2248         *val = 0;
2249         spin_lock(&kvm_lock);
2250         list_for_each_entry(kvm, &vm_list, vm_list)
2251                 kvm_for_each_vcpu(i, vcpu, kvm)
2252                         *val += *(u32 *)((void *)vcpu + offset);
2253
2254         spin_unlock(&kvm_lock);
2255         return 0;
2256 }
2257
2258 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2259
2260 static const struct file_operations *stat_fops[] = {
2261         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2262         [KVM_STAT_VM]   = &vm_stat_fops,
2263 };
2264
2265 static void kvm_init_debug(void)
2266 {
2267         struct kvm_stats_debugfs_item *p;
2268
2269         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2270         for (p = debugfs_entries; p->name; ++p)
2271                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2272                                                 (void *)(long)p->offset,
2273                                                 stat_fops[p->kind]);
2274 }
2275
2276 static void kvm_exit_debug(void)
2277 {
2278         struct kvm_stats_debugfs_item *p;
2279
2280         for (p = debugfs_entries; p->name; ++p)
2281                 debugfs_remove(p->dentry);
2282         debugfs_remove(kvm_debugfs_dir);
2283 }
2284
2285 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2286 {
2287         if (kvm_usage_count)
2288                 hardware_disable(NULL);
2289         return 0;
2290 }
2291
2292 static int kvm_resume(struct sys_device *dev)
2293 {
2294         if (kvm_usage_count) {
2295                 WARN_ON(spin_is_locked(&kvm_lock));
2296                 hardware_enable(NULL);
2297         }
2298         return 0;
2299 }
2300
2301 static struct sysdev_class kvm_sysdev_class = {
2302         .name = "kvm",
2303         .suspend = kvm_suspend,
2304         .resume = kvm_resume,
2305 };
2306
2307 static struct sys_device kvm_sysdev = {
2308         .id = 0,
2309         .cls = &kvm_sysdev_class,
2310 };
2311
2312 struct page *bad_page;
2313 pfn_t bad_pfn;
2314
2315 static inline
2316 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2317 {
2318         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2319 }
2320
2321 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2322 {
2323         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2324
2325         kvm_arch_vcpu_load(vcpu, cpu);
2326 }
2327
2328 static void kvm_sched_out(struct preempt_notifier *pn,
2329                           struct task_struct *next)
2330 {
2331         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2332
2333         kvm_arch_vcpu_put(vcpu);
2334 }
2335
2336 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2337                   struct module *module)
2338 {
2339         int r;
2340         int cpu;
2341
2342         r = kvm_arch_init(opaque);
2343         if (r)
2344                 goto out_fail;
2345
2346         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2347
2348         if (bad_page == NULL) {
2349                 r = -ENOMEM;
2350                 goto out;
2351         }
2352
2353         bad_pfn = page_to_pfn(bad_page);
2354
2355         hwpoison_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2356
2357         if (hwpoison_page == NULL) {
2358                 r = -ENOMEM;
2359                 goto out_free_0;
2360         }
2361
2362         hwpoison_pfn = page_to_pfn(hwpoison_page);
2363
2364         fault_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2365
2366         if (fault_page == NULL) {
2367                 r = -ENOMEM;
2368                 goto out_free_0;
2369         }
2370
2371         fault_pfn = page_to_pfn(fault_page);
2372
2373         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2374                 r = -ENOMEM;
2375                 goto out_free_0;
2376         }
2377
2378         r = kvm_arch_hardware_setup();
2379         if (r < 0)
2380                 goto out_free_0a;
2381
2382         for_each_online_cpu(cpu) {
2383                 smp_call_function_single(cpu,
2384                                 kvm_arch_check_processor_compat,
2385                                 &r, 1);
2386                 if (r < 0)
2387                         goto out_free_1;
2388         }
2389
2390         r = register_cpu_notifier(&kvm_cpu_notifier);
2391         if (r)
2392                 goto out_free_2;
2393         register_reboot_notifier(&kvm_reboot_notifier);
2394
2395         r = sysdev_class_register(&kvm_sysdev_class);
2396         if (r)
2397                 goto out_free_3;
2398
2399         r = sysdev_register(&kvm_sysdev);
2400         if (r)
2401                 goto out_free_4;
2402
2403         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2404         if (!vcpu_align)
2405                 vcpu_align = __alignof__(struct kvm_vcpu);
2406         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2407                                            0, NULL);
2408         if (!kvm_vcpu_cache) {
2409                 r = -ENOMEM;
2410                 goto out_free_5;
2411         }
2412
2413         r = kvm_async_pf_init();
2414         if (r)
2415                 goto out_free;
2416
2417         kvm_chardev_ops.owner = module;
2418         kvm_vm_fops.owner = module;
2419         kvm_vcpu_fops.owner = module;
2420
2421         r = misc_register(&kvm_dev);
2422         if (r) {
2423                 printk(KERN_ERR "kvm: misc device register failed\n");
2424                 goto out_unreg;
2425         }
2426
2427         kvm_preempt_ops.sched_in = kvm_sched_in;
2428         kvm_preempt_ops.sched_out = kvm_sched_out;
2429
2430         kvm_init_debug();
2431
2432         return 0;
2433
2434 out_unreg:
2435         kvm_async_pf_deinit();
2436 out_free:
2437         kmem_cache_destroy(kvm_vcpu_cache);
2438 out_free_5:
2439         sysdev_unregister(&kvm_sysdev);
2440 out_free_4:
2441         sysdev_class_unregister(&kvm_sysdev_class);
2442 out_free_3:
2443         unregister_reboot_notifier(&kvm_reboot_notifier);
2444         unregister_cpu_notifier(&kvm_cpu_notifier);
2445 out_free_2:
2446 out_free_1:
2447         kvm_arch_hardware_unsetup();
2448 out_free_0a:
2449         free_cpumask_var(cpus_hardware_enabled);
2450 out_free_0:
2451         if (fault_page)
2452                 __free_page(fault_page);
2453         if (hwpoison_page)
2454                 __free_page(hwpoison_page);
2455         __free_page(bad_page);
2456 out:
2457         kvm_arch_exit();
2458 out_fail:
2459         return r;
2460 }
2461 EXPORT_SYMBOL_GPL(kvm_init);
2462
2463 void kvm_exit(void)
2464 {
2465         kvm_exit_debug();
2466         misc_deregister(&kvm_dev);
2467         kmem_cache_destroy(kvm_vcpu_cache);
2468         kvm_async_pf_deinit();
2469         sysdev_unregister(&kvm_sysdev);
2470         sysdev_class_unregister(&kvm_sysdev_class);
2471         unregister_reboot_notifier(&kvm_reboot_notifier);
2472         unregister_cpu_notifier(&kvm_cpu_notifier);
2473         on_each_cpu(hardware_disable, NULL, 1);
2474         kvm_arch_hardware_unsetup();
2475         kvm_arch_exit();
2476         free_cpumask_var(cpus_hardware_enabled);
2477         __free_page(hwpoison_page);
2478         __free_page(bad_page);
2479 }
2480 EXPORT_SYMBOL_GPL(kvm_exit);