KVM: Add memory slot versioning and use it to provide fast guest write interface
[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 hva_to_pfn(struct kvm *kvm, unsigned long addr, bool atomic,
956                         bool *async)
957 {
958         struct page *page[1];
959         int npages = 0;
960         pfn_t pfn;
961
962         /* we can do it either atomically or asynchronously, not both */
963         BUG_ON(atomic && async);
964
965         if (atomic || async)
966                 npages = __get_user_pages_fast(addr, 1, 1, page);
967
968         if (unlikely(npages != 1) && !atomic) {
969                 might_sleep();
970                 npages = get_user_pages_fast(addr, 1, 1, page);
971         }
972
973         if (unlikely(npages != 1)) {
974                 struct vm_area_struct *vma;
975
976                 if (atomic)
977                         goto return_fault_page;
978
979                 down_read(&current->mm->mmap_sem);
980                 if (is_hwpoison_address(addr)) {
981                         up_read(&current->mm->mmap_sem);
982                         get_page(hwpoison_page);
983                         return page_to_pfn(hwpoison_page);
984                 }
985
986                 vma = find_vma(current->mm, addr);
987
988                 if (vma == NULL || addr < vma->vm_start ||
989                     !(vma->vm_flags & VM_PFNMAP)) {
990                         if (async && !(vma->vm_flags & VM_PFNMAP) &&
991                             (vma->vm_flags & VM_WRITE))
992                                 *async = true;
993                         up_read(&current->mm->mmap_sem);
994 return_fault_page:
995                         get_page(fault_page);
996                         return page_to_pfn(fault_page);
997                 }
998
999                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1000                 up_read(&current->mm->mmap_sem);
1001                 BUG_ON(!kvm_is_mmio_pfn(pfn));
1002         } else
1003                 pfn = page_to_pfn(page[0]);
1004
1005         return pfn;
1006 }
1007
1008 pfn_t hva_to_pfn_atomic(struct kvm *kvm, unsigned long addr)
1009 {
1010         return hva_to_pfn(kvm, addr, true, NULL);
1011 }
1012 EXPORT_SYMBOL_GPL(hva_to_pfn_atomic);
1013
1014 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async)
1015 {
1016         unsigned long addr;
1017
1018         if (async)
1019                 *async = false;
1020
1021         addr = gfn_to_hva(kvm, gfn);
1022         if (kvm_is_error_hva(addr)) {
1023                 get_page(bad_page);
1024                 return page_to_pfn(bad_page);
1025         }
1026
1027         return hva_to_pfn(kvm, addr, atomic, async);
1028 }
1029
1030 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1031 {
1032         return __gfn_to_pfn(kvm, gfn, true, NULL);
1033 }
1034 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1035
1036 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async)
1037 {
1038         return __gfn_to_pfn(kvm, gfn, false, async);
1039 }
1040 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1041
1042 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1043 {
1044         return __gfn_to_pfn(kvm, gfn, false, NULL);
1045 }
1046 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1047
1048 pfn_t gfn_to_pfn_memslot(struct kvm *kvm,
1049                          struct kvm_memory_slot *slot, gfn_t gfn)
1050 {
1051         unsigned long addr = gfn_to_hva_memslot(slot, gfn);
1052         return hva_to_pfn(kvm, addr, false, NULL);
1053 }
1054
1055 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1056                                                                   int nr_pages)
1057 {
1058         unsigned long addr;
1059         gfn_t entry;
1060
1061         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1062         if (kvm_is_error_hva(addr))
1063                 return -1;
1064
1065         if (entry < nr_pages)
1066                 return 0;
1067
1068         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1069 }
1070 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1071
1072 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1073 {
1074         pfn_t pfn;
1075
1076         pfn = gfn_to_pfn(kvm, gfn);
1077         if (!kvm_is_mmio_pfn(pfn))
1078                 return pfn_to_page(pfn);
1079
1080         WARN_ON(kvm_is_mmio_pfn(pfn));
1081
1082         get_page(bad_page);
1083         return bad_page;
1084 }
1085
1086 EXPORT_SYMBOL_GPL(gfn_to_page);
1087
1088 void kvm_release_page_clean(struct page *page)
1089 {
1090         kvm_release_pfn_clean(page_to_pfn(page));
1091 }
1092 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1093
1094 void kvm_release_pfn_clean(pfn_t pfn)
1095 {
1096         if (!kvm_is_mmio_pfn(pfn))
1097                 put_page(pfn_to_page(pfn));
1098 }
1099 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1100
1101 void kvm_release_page_dirty(struct page *page)
1102 {
1103         kvm_release_pfn_dirty(page_to_pfn(page));
1104 }
1105 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1106
1107 void kvm_release_pfn_dirty(pfn_t pfn)
1108 {
1109         kvm_set_pfn_dirty(pfn);
1110         kvm_release_pfn_clean(pfn);
1111 }
1112 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1113
1114 void kvm_set_page_dirty(struct page *page)
1115 {
1116         kvm_set_pfn_dirty(page_to_pfn(page));
1117 }
1118 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1119
1120 void kvm_set_pfn_dirty(pfn_t pfn)
1121 {
1122         if (!kvm_is_mmio_pfn(pfn)) {
1123                 struct page *page = pfn_to_page(pfn);
1124                 if (!PageReserved(page))
1125                         SetPageDirty(page);
1126         }
1127 }
1128 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1129
1130 void kvm_set_pfn_accessed(pfn_t pfn)
1131 {
1132         if (!kvm_is_mmio_pfn(pfn))
1133                 mark_page_accessed(pfn_to_page(pfn));
1134 }
1135 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1136
1137 void kvm_get_pfn(pfn_t pfn)
1138 {
1139         if (!kvm_is_mmio_pfn(pfn))
1140                 get_page(pfn_to_page(pfn));
1141 }
1142 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1143
1144 static int next_segment(unsigned long len, int offset)
1145 {
1146         if (len > PAGE_SIZE - offset)
1147                 return PAGE_SIZE - offset;
1148         else
1149                 return len;
1150 }
1151
1152 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1153                         int len)
1154 {
1155         int r;
1156         unsigned long addr;
1157
1158         addr = gfn_to_hva(kvm, gfn);
1159         if (kvm_is_error_hva(addr))
1160                 return -EFAULT;
1161         r = copy_from_user(data, (void __user *)addr + offset, len);
1162         if (r)
1163                 return -EFAULT;
1164         return 0;
1165 }
1166 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1167
1168 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1169 {
1170         gfn_t gfn = gpa >> PAGE_SHIFT;
1171         int seg;
1172         int offset = offset_in_page(gpa);
1173         int ret;
1174
1175         while ((seg = next_segment(len, offset)) != 0) {
1176                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1177                 if (ret < 0)
1178                         return ret;
1179                 offset = 0;
1180                 len -= seg;
1181                 data += seg;
1182                 ++gfn;
1183         }
1184         return 0;
1185 }
1186 EXPORT_SYMBOL_GPL(kvm_read_guest);
1187
1188 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1189                           unsigned long len)
1190 {
1191         int r;
1192         unsigned long addr;
1193         gfn_t gfn = gpa >> PAGE_SHIFT;
1194         int offset = offset_in_page(gpa);
1195
1196         addr = gfn_to_hva(kvm, gfn);
1197         if (kvm_is_error_hva(addr))
1198                 return -EFAULT;
1199         pagefault_disable();
1200         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1201         pagefault_enable();
1202         if (r)
1203                 return -EFAULT;
1204         return 0;
1205 }
1206 EXPORT_SYMBOL(kvm_read_guest_atomic);
1207
1208 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1209                          int offset, int len)
1210 {
1211         int r;
1212         unsigned long addr;
1213
1214         addr = gfn_to_hva(kvm, gfn);
1215         if (kvm_is_error_hva(addr))
1216                 return -EFAULT;
1217         r = copy_to_user((void __user *)addr + offset, data, len);
1218         if (r)
1219                 return -EFAULT;
1220         mark_page_dirty(kvm, gfn);
1221         return 0;
1222 }
1223 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1224
1225 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1226                     unsigned long len)
1227 {
1228         gfn_t gfn = gpa >> PAGE_SHIFT;
1229         int seg;
1230         int offset = offset_in_page(gpa);
1231         int ret;
1232
1233         while ((seg = next_segment(len, offset)) != 0) {
1234                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1235                 if (ret < 0)
1236                         return ret;
1237                 offset = 0;
1238                 len -= seg;
1239                 data += seg;
1240                 ++gfn;
1241         }
1242         return 0;
1243 }
1244
1245 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1246                               gpa_t gpa)
1247 {
1248         struct kvm_memslots *slots = kvm_memslots(kvm);
1249         int offset = offset_in_page(gpa);
1250         gfn_t gfn = gpa >> PAGE_SHIFT;
1251
1252         ghc->gpa = gpa;
1253         ghc->generation = slots->generation;
1254         ghc->memslot = __gfn_to_memslot(slots, gfn);
1255         ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1256         if (!kvm_is_error_hva(ghc->hva))
1257                 ghc->hva += offset;
1258         else
1259                 return -EFAULT;
1260
1261         return 0;
1262 }
1263 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1264
1265 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1266                            void *data, unsigned long len)
1267 {
1268         struct kvm_memslots *slots = kvm_memslots(kvm);
1269         int r;
1270
1271         if (slots->generation != ghc->generation)
1272                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1273
1274         if (kvm_is_error_hva(ghc->hva))
1275                 return -EFAULT;
1276
1277         r = copy_to_user((void __user *)ghc->hva, data, len);
1278         if (r)
1279                 return -EFAULT;
1280         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1281
1282         return 0;
1283 }
1284 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1285
1286 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1287 {
1288         return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
1289 }
1290 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1291
1292 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1293 {
1294         gfn_t gfn = gpa >> PAGE_SHIFT;
1295         int seg;
1296         int offset = offset_in_page(gpa);
1297         int ret;
1298
1299         while ((seg = next_segment(len, offset)) != 0) {
1300                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1301                 if (ret < 0)
1302                         return ret;
1303                 offset = 0;
1304                 len -= seg;
1305                 ++gfn;
1306         }
1307         return 0;
1308 }
1309 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1310
1311 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1312                              gfn_t gfn)
1313 {
1314         if (memslot && memslot->dirty_bitmap) {
1315                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1316
1317                 generic___set_le_bit(rel_gfn, memslot->dirty_bitmap);
1318         }
1319 }
1320
1321 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1322 {
1323         struct kvm_memory_slot *memslot;
1324
1325         memslot = gfn_to_memslot(kvm, gfn);
1326         mark_page_dirty_in_slot(kvm, memslot, gfn);
1327 }
1328
1329 /*
1330  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1331  */
1332 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1333 {
1334         DEFINE_WAIT(wait);
1335
1336         for (;;) {
1337                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1338
1339                 if (kvm_arch_vcpu_runnable(vcpu)) {
1340                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1341                         break;
1342                 }
1343                 if (kvm_cpu_has_pending_timer(vcpu))
1344                         break;
1345                 if (signal_pending(current))
1346                         break;
1347
1348                 schedule();
1349         }
1350
1351         finish_wait(&vcpu->wq, &wait);
1352 }
1353
1354 void kvm_resched(struct kvm_vcpu *vcpu)
1355 {
1356         if (!need_resched())
1357                 return;
1358         cond_resched();
1359 }
1360 EXPORT_SYMBOL_GPL(kvm_resched);
1361
1362 void kvm_vcpu_on_spin(struct kvm_vcpu *vcpu)
1363 {
1364         ktime_t expires;
1365         DEFINE_WAIT(wait);
1366
1367         prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1368
1369         /* Sleep for 100 us, and hope lock-holder got scheduled */
1370         expires = ktime_add_ns(ktime_get(), 100000UL);
1371         schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
1372
1373         finish_wait(&vcpu->wq, &wait);
1374 }
1375 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1376
1377 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1378 {
1379         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1380         struct page *page;
1381
1382         if (vmf->pgoff == 0)
1383                 page = virt_to_page(vcpu->run);
1384 #ifdef CONFIG_X86
1385         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1386                 page = virt_to_page(vcpu->arch.pio_data);
1387 #endif
1388 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1389         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1390                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1391 #endif
1392         else
1393                 return VM_FAULT_SIGBUS;
1394         get_page(page);
1395         vmf->page = page;
1396         return 0;
1397 }
1398
1399 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1400         .fault = kvm_vcpu_fault,
1401 };
1402
1403 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1404 {
1405         vma->vm_ops = &kvm_vcpu_vm_ops;
1406         return 0;
1407 }
1408
1409 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1410 {
1411         struct kvm_vcpu *vcpu = filp->private_data;
1412
1413         kvm_put_kvm(vcpu->kvm);
1414         return 0;
1415 }
1416
1417 static struct file_operations kvm_vcpu_fops = {
1418         .release        = kvm_vcpu_release,
1419         .unlocked_ioctl = kvm_vcpu_ioctl,
1420         .compat_ioctl   = kvm_vcpu_ioctl,
1421         .mmap           = kvm_vcpu_mmap,
1422         .llseek         = noop_llseek,
1423 };
1424
1425 /*
1426  * Allocates an inode for the vcpu.
1427  */
1428 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1429 {
1430         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1431 }
1432
1433 /*
1434  * Creates some virtual cpus.  Good luck creating more than one.
1435  */
1436 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1437 {
1438         int r;
1439         struct kvm_vcpu *vcpu, *v;
1440
1441         vcpu = kvm_arch_vcpu_create(kvm, id);
1442         if (IS_ERR(vcpu))
1443                 return PTR_ERR(vcpu);
1444
1445         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1446
1447         r = kvm_arch_vcpu_setup(vcpu);
1448         if (r)
1449                 return r;
1450
1451         mutex_lock(&kvm->lock);
1452         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1453                 r = -EINVAL;
1454                 goto vcpu_destroy;
1455         }
1456
1457         kvm_for_each_vcpu(r, v, kvm)
1458                 if (v->vcpu_id == id) {
1459                         r = -EEXIST;
1460                         goto vcpu_destroy;
1461                 }
1462
1463         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1464
1465         /* Now it's all set up, let userspace reach it */
1466         kvm_get_kvm(kvm);
1467         r = create_vcpu_fd(vcpu);
1468         if (r < 0) {
1469                 kvm_put_kvm(kvm);
1470                 goto vcpu_destroy;
1471         }
1472
1473         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1474         smp_wmb();
1475         atomic_inc(&kvm->online_vcpus);
1476
1477 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1478         if (kvm->bsp_vcpu_id == id)
1479                 kvm->bsp_vcpu = vcpu;
1480 #endif
1481         mutex_unlock(&kvm->lock);
1482         return r;
1483
1484 vcpu_destroy:
1485         mutex_unlock(&kvm->lock);
1486         kvm_arch_vcpu_destroy(vcpu);
1487         return r;
1488 }
1489
1490 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1491 {
1492         if (sigset) {
1493                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1494                 vcpu->sigset_active = 1;
1495                 vcpu->sigset = *sigset;
1496         } else
1497                 vcpu->sigset_active = 0;
1498         return 0;
1499 }
1500
1501 static long kvm_vcpu_ioctl(struct file *filp,
1502                            unsigned int ioctl, unsigned long arg)
1503 {
1504         struct kvm_vcpu *vcpu = filp->private_data;
1505         void __user *argp = (void __user *)arg;
1506         int r;
1507         struct kvm_fpu *fpu = NULL;
1508         struct kvm_sregs *kvm_sregs = NULL;
1509
1510         if (vcpu->kvm->mm != current->mm)
1511                 return -EIO;
1512
1513 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1514         /*
1515          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1516          * so vcpu_load() would break it.
1517          */
1518         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1519                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1520 #endif
1521
1522
1523         vcpu_load(vcpu);
1524         switch (ioctl) {
1525         case KVM_RUN:
1526                 r = -EINVAL;
1527                 if (arg)
1528                         goto out;
1529                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1530                 break;
1531         case KVM_GET_REGS: {
1532                 struct kvm_regs *kvm_regs;
1533
1534                 r = -ENOMEM;
1535                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1536                 if (!kvm_regs)
1537                         goto out;
1538                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1539                 if (r)
1540                         goto out_free1;
1541                 r = -EFAULT;
1542                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1543                         goto out_free1;
1544                 r = 0;
1545 out_free1:
1546                 kfree(kvm_regs);
1547                 break;
1548         }
1549         case KVM_SET_REGS: {
1550                 struct kvm_regs *kvm_regs;
1551
1552                 r = -ENOMEM;
1553                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1554                 if (!kvm_regs)
1555                         goto out;
1556                 r = -EFAULT;
1557                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1558                         goto out_free2;
1559                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1560                 if (r)
1561                         goto out_free2;
1562                 r = 0;
1563 out_free2:
1564                 kfree(kvm_regs);
1565                 break;
1566         }
1567         case KVM_GET_SREGS: {
1568                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1569                 r = -ENOMEM;
1570                 if (!kvm_sregs)
1571                         goto out;
1572                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1573                 if (r)
1574                         goto out;
1575                 r = -EFAULT;
1576                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1577                         goto out;
1578                 r = 0;
1579                 break;
1580         }
1581         case KVM_SET_SREGS: {
1582                 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1583                 r = -ENOMEM;
1584                 if (!kvm_sregs)
1585                         goto out;
1586                 r = -EFAULT;
1587                 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1588                         goto out;
1589                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1590                 if (r)
1591                         goto out;
1592                 r = 0;
1593                 break;
1594         }
1595         case KVM_GET_MP_STATE: {
1596                 struct kvm_mp_state mp_state;
1597
1598                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1599                 if (r)
1600                         goto out;
1601                 r = -EFAULT;
1602                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1603                         goto out;
1604                 r = 0;
1605                 break;
1606         }
1607         case KVM_SET_MP_STATE: {
1608                 struct kvm_mp_state mp_state;
1609
1610                 r = -EFAULT;
1611                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1612                         goto out;
1613                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1614                 if (r)
1615                         goto out;
1616                 r = 0;
1617                 break;
1618         }
1619         case KVM_TRANSLATE: {
1620                 struct kvm_translation tr;
1621
1622                 r = -EFAULT;
1623                 if (copy_from_user(&tr, argp, sizeof tr))
1624                         goto out;
1625                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1626                 if (r)
1627                         goto out;
1628                 r = -EFAULT;
1629                 if (copy_to_user(argp, &tr, sizeof tr))
1630                         goto out;
1631                 r = 0;
1632                 break;
1633         }
1634         case KVM_SET_GUEST_DEBUG: {
1635                 struct kvm_guest_debug dbg;
1636
1637                 r = -EFAULT;
1638                 if (copy_from_user(&dbg, argp, sizeof dbg))
1639                         goto out;
1640                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1641                 if (r)
1642                         goto out;
1643                 r = 0;
1644                 break;
1645         }
1646         case KVM_SET_SIGNAL_MASK: {
1647                 struct kvm_signal_mask __user *sigmask_arg = argp;
1648                 struct kvm_signal_mask kvm_sigmask;
1649                 sigset_t sigset, *p;
1650
1651                 p = NULL;
1652                 if (argp) {
1653                         r = -EFAULT;
1654                         if (copy_from_user(&kvm_sigmask, argp,
1655                                            sizeof kvm_sigmask))
1656                                 goto out;
1657                         r = -EINVAL;
1658                         if (kvm_sigmask.len != sizeof sigset)
1659                                 goto out;
1660                         r = -EFAULT;
1661                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1662                                            sizeof sigset))
1663                                 goto out;
1664                         p = &sigset;
1665                 }
1666                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
1667                 break;
1668         }
1669         case KVM_GET_FPU: {
1670                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1671                 r = -ENOMEM;
1672                 if (!fpu)
1673                         goto out;
1674                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1675                 if (r)
1676                         goto out;
1677                 r = -EFAULT;
1678                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1679                         goto out;
1680                 r = 0;
1681                 break;
1682         }
1683         case KVM_SET_FPU: {
1684                 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1685                 r = -ENOMEM;
1686                 if (!fpu)
1687                         goto out;
1688                 r = -EFAULT;
1689                 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1690                         goto out;
1691                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1692                 if (r)
1693                         goto out;
1694                 r = 0;
1695                 break;
1696         }
1697         default:
1698                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1699         }
1700 out:
1701         vcpu_put(vcpu);
1702         kfree(fpu);
1703         kfree(kvm_sregs);
1704         return r;
1705 }
1706
1707 static long kvm_vm_ioctl(struct file *filp,
1708                            unsigned int ioctl, unsigned long arg)
1709 {
1710         struct kvm *kvm = filp->private_data;
1711         void __user *argp = (void __user *)arg;
1712         int r;
1713
1714         if (kvm->mm != current->mm)
1715                 return -EIO;
1716         switch (ioctl) {
1717         case KVM_CREATE_VCPU:
1718                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1719                 if (r < 0)
1720                         goto out;
1721                 break;
1722         case KVM_SET_USER_MEMORY_REGION: {
1723                 struct kvm_userspace_memory_region kvm_userspace_mem;
1724
1725                 r = -EFAULT;
1726                 if (copy_from_user(&kvm_userspace_mem, argp,
1727                                                 sizeof kvm_userspace_mem))
1728                         goto out;
1729
1730                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1731                 if (r)
1732                         goto out;
1733                 break;
1734         }
1735         case KVM_GET_DIRTY_LOG: {
1736                 struct kvm_dirty_log log;
1737
1738                 r = -EFAULT;
1739                 if (copy_from_user(&log, argp, sizeof log))
1740                         goto out;
1741                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1742                 if (r)
1743                         goto out;
1744                 break;
1745         }
1746 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1747         case KVM_REGISTER_COALESCED_MMIO: {
1748                 struct kvm_coalesced_mmio_zone zone;
1749                 r = -EFAULT;
1750                 if (copy_from_user(&zone, argp, sizeof zone))
1751                         goto out;
1752                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1753                 if (r)
1754                         goto out;
1755                 r = 0;
1756                 break;
1757         }
1758         case KVM_UNREGISTER_COALESCED_MMIO: {
1759                 struct kvm_coalesced_mmio_zone zone;
1760                 r = -EFAULT;
1761                 if (copy_from_user(&zone, argp, sizeof zone))
1762                         goto out;
1763                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1764                 if (r)
1765                         goto out;
1766                 r = 0;
1767                 break;
1768         }
1769 #endif
1770         case KVM_IRQFD: {
1771                 struct kvm_irqfd data;
1772
1773                 r = -EFAULT;
1774                 if (copy_from_user(&data, argp, sizeof data))
1775                         goto out;
1776                 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
1777                 break;
1778         }
1779         case KVM_IOEVENTFD: {
1780                 struct kvm_ioeventfd data;
1781
1782                 r = -EFAULT;
1783                 if (copy_from_user(&data, argp, sizeof data))
1784                         goto out;
1785                 r = kvm_ioeventfd(kvm, &data);
1786                 break;
1787         }
1788 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1789         case KVM_SET_BOOT_CPU_ID:
1790                 r = 0;
1791                 mutex_lock(&kvm->lock);
1792                 if (atomic_read(&kvm->online_vcpus) != 0)
1793                         r = -EBUSY;
1794                 else
1795                         kvm->bsp_vcpu_id = arg;
1796                 mutex_unlock(&kvm->lock);
1797                 break;
1798 #endif
1799         default:
1800                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1801                 if (r == -ENOTTY)
1802                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
1803         }
1804 out:
1805         return r;
1806 }
1807
1808 #ifdef CONFIG_COMPAT
1809 struct compat_kvm_dirty_log {
1810         __u32 slot;
1811         __u32 padding1;
1812         union {
1813                 compat_uptr_t dirty_bitmap; /* one bit per page */
1814                 __u64 padding2;
1815         };
1816 };
1817
1818 static long kvm_vm_compat_ioctl(struct file *filp,
1819                            unsigned int ioctl, unsigned long arg)
1820 {
1821         struct kvm *kvm = filp->private_data;
1822         int r;
1823
1824         if (kvm->mm != current->mm)
1825                 return -EIO;
1826         switch (ioctl) {
1827         case KVM_GET_DIRTY_LOG: {
1828                 struct compat_kvm_dirty_log compat_log;
1829                 struct kvm_dirty_log log;
1830
1831                 r = -EFAULT;
1832                 if (copy_from_user(&compat_log, (void __user *)arg,
1833                                    sizeof(compat_log)))
1834                         goto out;
1835                 log.slot         = compat_log.slot;
1836                 log.padding1     = compat_log.padding1;
1837                 log.padding2     = compat_log.padding2;
1838                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
1839
1840                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1841                 if (r)
1842                         goto out;
1843                 break;
1844         }
1845         default:
1846                 r = kvm_vm_ioctl(filp, ioctl, arg);
1847         }
1848
1849 out:
1850         return r;
1851 }
1852 #endif
1853
1854 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1855 {
1856         struct page *page[1];
1857         unsigned long addr;
1858         int npages;
1859         gfn_t gfn = vmf->pgoff;
1860         struct kvm *kvm = vma->vm_file->private_data;
1861
1862         addr = gfn_to_hva(kvm, gfn);
1863         if (kvm_is_error_hva(addr))
1864                 return VM_FAULT_SIGBUS;
1865
1866         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
1867                                 NULL);
1868         if (unlikely(npages != 1))
1869                 return VM_FAULT_SIGBUS;
1870
1871         vmf->page = page[0];
1872         return 0;
1873 }
1874
1875 static const struct vm_operations_struct kvm_vm_vm_ops = {
1876         .fault = kvm_vm_fault,
1877 };
1878
1879 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1880 {
1881         vma->vm_ops = &kvm_vm_vm_ops;
1882         return 0;
1883 }
1884
1885 static struct file_operations kvm_vm_fops = {
1886         .release        = kvm_vm_release,
1887         .unlocked_ioctl = kvm_vm_ioctl,
1888 #ifdef CONFIG_COMPAT
1889         .compat_ioctl   = kvm_vm_compat_ioctl,
1890 #endif
1891         .mmap           = kvm_vm_mmap,
1892         .llseek         = noop_llseek,
1893 };
1894
1895 static int kvm_dev_ioctl_create_vm(void)
1896 {
1897         int fd, r;
1898         struct kvm *kvm;
1899
1900         kvm = kvm_create_vm();
1901         if (IS_ERR(kvm))
1902                 return PTR_ERR(kvm);
1903 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1904         r = kvm_coalesced_mmio_init(kvm);
1905         if (r < 0) {
1906                 kvm_put_kvm(kvm);
1907                 return r;
1908         }
1909 #endif
1910         fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
1911         if (fd < 0)
1912                 kvm_put_kvm(kvm);
1913
1914         return fd;
1915 }
1916
1917 static long kvm_dev_ioctl_check_extension_generic(long arg)
1918 {
1919         switch (arg) {
1920         case KVM_CAP_USER_MEMORY:
1921         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
1922         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
1923 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1924         case KVM_CAP_SET_BOOT_CPU_ID:
1925 #endif
1926         case KVM_CAP_INTERNAL_ERROR_DATA:
1927                 return 1;
1928 #ifdef CONFIG_HAVE_KVM_IRQCHIP
1929         case KVM_CAP_IRQ_ROUTING:
1930                 return KVM_MAX_IRQ_ROUTES;
1931 #endif
1932         default:
1933                 break;
1934         }
1935         return kvm_dev_ioctl_check_extension(arg);
1936 }
1937
1938 static long kvm_dev_ioctl(struct file *filp,
1939                           unsigned int ioctl, unsigned long arg)
1940 {
1941         long r = -EINVAL;
1942
1943         switch (ioctl) {
1944         case KVM_GET_API_VERSION:
1945                 r = -EINVAL;
1946                 if (arg)
1947                         goto out;
1948                 r = KVM_API_VERSION;
1949                 break;
1950         case KVM_CREATE_VM:
1951                 r = -EINVAL;
1952                 if (arg)
1953                         goto out;
1954                 r = kvm_dev_ioctl_create_vm();
1955                 break;
1956         case KVM_CHECK_EXTENSION:
1957                 r = kvm_dev_ioctl_check_extension_generic(arg);
1958                 break;
1959         case KVM_GET_VCPU_MMAP_SIZE:
1960                 r = -EINVAL;
1961                 if (arg)
1962                         goto out;
1963                 r = PAGE_SIZE;     /* struct kvm_run */
1964 #ifdef CONFIG_X86
1965                 r += PAGE_SIZE;    /* pio data page */
1966 #endif
1967 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1968                 r += PAGE_SIZE;    /* coalesced mmio ring page */
1969 #endif
1970                 break;
1971         case KVM_TRACE_ENABLE:
1972         case KVM_TRACE_PAUSE:
1973         case KVM_TRACE_DISABLE:
1974                 r = -EOPNOTSUPP;
1975                 break;
1976         default:
1977                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
1978         }
1979 out:
1980         return r;
1981 }
1982
1983 static struct file_operations kvm_chardev_ops = {
1984         .unlocked_ioctl = kvm_dev_ioctl,
1985         .compat_ioctl   = kvm_dev_ioctl,
1986         .llseek         = noop_llseek,
1987 };
1988
1989 static struct miscdevice kvm_dev = {
1990         KVM_MINOR,
1991         "kvm",
1992         &kvm_chardev_ops,
1993 };
1994
1995 static void hardware_enable(void *junk)
1996 {
1997         int cpu = raw_smp_processor_id();
1998         int r;
1999
2000         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2001                 return;
2002
2003         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2004
2005         r = kvm_arch_hardware_enable(NULL);
2006
2007         if (r) {
2008                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2009                 atomic_inc(&hardware_enable_failed);
2010                 printk(KERN_INFO "kvm: enabling virtualization on "
2011                                  "CPU%d failed\n", cpu);
2012         }
2013 }
2014
2015 static void hardware_disable(void *junk)
2016 {
2017         int cpu = raw_smp_processor_id();
2018
2019         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2020                 return;
2021         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2022         kvm_arch_hardware_disable(NULL);
2023 }
2024
2025 static void hardware_disable_all_nolock(void)
2026 {
2027         BUG_ON(!kvm_usage_count);
2028
2029         kvm_usage_count--;
2030         if (!kvm_usage_count)
2031                 on_each_cpu(hardware_disable, NULL, 1);
2032 }
2033
2034 static void hardware_disable_all(void)
2035 {
2036         spin_lock(&kvm_lock);
2037         hardware_disable_all_nolock();
2038         spin_unlock(&kvm_lock);
2039 }
2040
2041 static int hardware_enable_all(void)
2042 {
2043         int r = 0;
2044
2045         spin_lock(&kvm_lock);
2046
2047         kvm_usage_count++;
2048         if (kvm_usage_count == 1) {
2049                 atomic_set(&hardware_enable_failed, 0);
2050                 on_each_cpu(hardware_enable, NULL, 1);
2051
2052                 if (atomic_read(&hardware_enable_failed)) {
2053                         hardware_disable_all_nolock();
2054                         r = -EBUSY;
2055                 }
2056         }
2057
2058         spin_unlock(&kvm_lock);
2059
2060         return r;
2061 }
2062
2063 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2064                            void *v)
2065 {
2066         int cpu = (long)v;
2067
2068         if (!kvm_usage_count)
2069                 return NOTIFY_OK;
2070
2071         val &= ~CPU_TASKS_FROZEN;
2072         switch (val) {
2073         case CPU_DYING:
2074                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2075                        cpu);
2076                 hardware_disable(NULL);
2077                 break;
2078         case CPU_STARTING:
2079                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2080                        cpu);
2081                 spin_lock(&kvm_lock);
2082                 hardware_enable(NULL);
2083                 spin_unlock(&kvm_lock);
2084                 break;
2085         }
2086         return NOTIFY_OK;
2087 }
2088
2089
2090 asmlinkage void kvm_handle_fault_on_reboot(void)
2091 {
2092         if (kvm_rebooting) {
2093                 /* spin while reset goes on */
2094                 local_irq_enable();
2095                 while (true)
2096                         cpu_relax();
2097         }
2098         /* Fault while not rebooting.  We want the trace. */
2099         BUG();
2100 }
2101 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
2102
2103 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2104                       void *v)
2105 {
2106         /*
2107          * Some (well, at least mine) BIOSes hang on reboot if
2108          * in vmx root mode.
2109          *
2110          * And Intel TXT required VMX off for all cpu when system shutdown.
2111          */
2112         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2113         kvm_rebooting = true;
2114         on_each_cpu(hardware_disable, NULL, 1);
2115         return NOTIFY_OK;
2116 }
2117
2118 static struct notifier_block kvm_reboot_notifier = {
2119         .notifier_call = kvm_reboot,
2120         .priority = 0,
2121 };
2122
2123 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2124 {
2125         int i;
2126
2127         for (i = 0; i < bus->dev_count; i++) {
2128                 struct kvm_io_device *pos = bus->devs[i];
2129
2130                 kvm_iodevice_destructor(pos);
2131         }
2132         kfree(bus);
2133 }
2134
2135 /* kvm_io_bus_write - called under kvm->slots_lock */
2136 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2137                      int len, const void *val)
2138 {
2139         int i;
2140         struct kvm_io_bus *bus;
2141
2142         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2143         for (i = 0; i < bus->dev_count; i++)
2144                 if (!kvm_iodevice_write(bus->devs[i], addr, len, val))
2145                         return 0;
2146         return -EOPNOTSUPP;
2147 }
2148
2149 /* kvm_io_bus_read - called under kvm->slots_lock */
2150 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2151                     int len, void *val)
2152 {
2153         int i;
2154         struct kvm_io_bus *bus;
2155
2156         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2157         for (i = 0; i < bus->dev_count; i++)
2158                 if (!kvm_iodevice_read(bus->devs[i], addr, len, val))
2159                         return 0;
2160         return -EOPNOTSUPP;
2161 }
2162
2163 /* Caller must hold slots_lock. */
2164 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2165                             struct kvm_io_device *dev)
2166 {
2167         struct kvm_io_bus *new_bus, *bus;
2168
2169         bus = kvm->buses[bus_idx];
2170         if (bus->dev_count > NR_IOBUS_DEVS-1)
2171                 return -ENOSPC;
2172
2173         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2174         if (!new_bus)
2175                 return -ENOMEM;
2176         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2177         new_bus->devs[new_bus->dev_count++] = dev;
2178         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2179         synchronize_srcu_expedited(&kvm->srcu);
2180         kfree(bus);
2181
2182         return 0;
2183 }
2184
2185 /* Caller must hold slots_lock. */
2186 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2187                               struct kvm_io_device *dev)
2188 {
2189         int i, r;
2190         struct kvm_io_bus *new_bus, *bus;
2191
2192         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2193         if (!new_bus)
2194                 return -ENOMEM;
2195
2196         bus = kvm->buses[bus_idx];
2197         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2198
2199         r = -ENOENT;
2200         for (i = 0; i < new_bus->dev_count; i++)
2201                 if (new_bus->devs[i] == dev) {
2202                         r = 0;
2203                         new_bus->devs[i] = new_bus->devs[--new_bus->dev_count];
2204                         break;
2205                 }
2206
2207         if (r) {
2208                 kfree(new_bus);
2209                 return r;
2210         }
2211
2212         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2213         synchronize_srcu_expedited(&kvm->srcu);
2214         kfree(bus);
2215         return r;
2216 }
2217
2218 static struct notifier_block kvm_cpu_notifier = {
2219         .notifier_call = kvm_cpu_hotplug,
2220 };
2221
2222 static int vm_stat_get(void *_offset, u64 *val)
2223 {
2224         unsigned offset = (long)_offset;
2225         struct kvm *kvm;
2226
2227         *val = 0;
2228         spin_lock(&kvm_lock);
2229         list_for_each_entry(kvm, &vm_list, vm_list)
2230                 *val += *(u32 *)((void *)kvm + offset);
2231         spin_unlock(&kvm_lock);
2232         return 0;
2233 }
2234
2235 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2236
2237 static int vcpu_stat_get(void *_offset, u64 *val)
2238 {
2239         unsigned offset = (long)_offset;
2240         struct kvm *kvm;
2241         struct kvm_vcpu *vcpu;
2242         int i;
2243
2244         *val = 0;
2245         spin_lock(&kvm_lock);
2246         list_for_each_entry(kvm, &vm_list, vm_list)
2247                 kvm_for_each_vcpu(i, vcpu, kvm)
2248                         *val += *(u32 *)((void *)vcpu + offset);
2249
2250         spin_unlock(&kvm_lock);
2251         return 0;
2252 }
2253
2254 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2255
2256 static const struct file_operations *stat_fops[] = {
2257         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2258         [KVM_STAT_VM]   = &vm_stat_fops,
2259 };
2260
2261 static void kvm_init_debug(void)
2262 {
2263         struct kvm_stats_debugfs_item *p;
2264
2265         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2266         for (p = debugfs_entries; p->name; ++p)
2267                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2268                                                 (void *)(long)p->offset,
2269                                                 stat_fops[p->kind]);
2270 }
2271
2272 static void kvm_exit_debug(void)
2273 {
2274         struct kvm_stats_debugfs_item *p;
2275
2276         for (p = debugfs_entries; p->name; ++p)
2277                 debugfs_remove(p->dentry);
2278         debugfs_remove(kvm_debugfs_dir);
2279 }
2280
2281 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2282 {
2283         if (kvm_usage_count)
2284                 hardware_disable(NULL);
2285         return 0;
2286 }
2287
2288 static int kvm_resume(struct sys_device *dev)
2289 {
2290         if (kvm_usage_count) {
2291                 WARN_ON(spin_is_locked(&kvm_lock));
2292                 hardware_enable(NULL);
2293         }
2294         return 0;
2295 }
2296
2297 static struct sysdev_class kvm_sysdev_class = {
2298         .name = "kvm",
2299         .suspend = kvm_suspend,
2300         .resume = kvm_resume,
2301 };
2302
2303 static struct sys_device kvm_sysdev = {
2304         .id = 0,
2305         .cls = &kvm_sysdev_class,
2306 };
2307
2308 struct page *bad_page;
2309 pfn_t bad_pfn;
2310
2311 static inline
2312 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2313 {
2314         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2315 }
2316
2317 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2318 {
2319         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2320
2321         kvm_arch_vcpu_load(vcpu, cpu);
2322 }
2323
2324 static void kvm_sched_out(struct preempt_notifier *pn,
2325                           struct task_struct *next)
2326 {
2327         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2328
2329         kvm_arch_vcpu_put(vcpu);
2330 }
2331
2332 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2333                   struct module *module)
2334 {
2335         int r;
2336         int cpu;
2337
2338         r = kvm_arch_init(opaque);
2339         if (r)
2340                 goto out_fail;
2341
2342         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2343
2344         if (bad_page == NULL) {
2345                 r = -ENOMEM;
2346                 goto out;
2347         }
2348
2349         bad_pfn = page_to_pfn(bad_page);
2350
2351         hwpoison_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2352
2353         if (hwpoison_page == NULL) {
2354                 r = -ENOMEM;
2355                 goto out_free_0;
2356         }
2357
2358         hwpoison_pfn = page_to_pfn(hwpoison_page);
2359
2360         fault_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2361
2362         if (fault_page == NULL) {
2363                 r = -ENOMEM;
2364                 goto out_free_0;
2365         }
2366
2367         fault_pfn = page_to_pfn(fault_page);
2368
2369         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2370                 r = -ENOMEM;
2371                 goto out_free_0;
2372         }
2373
2374         r = kvm_arch_hardware_setup();
2375         if (r < 0)
2376                 goto out_free_0a;
2377
2378         for_each_online_cpu(cpu) {
2379                 smp_call_function_single(cpu,
2380                                 kvm_arch_check_processor_compat,
2381                                 &r, 1);
2382                 if (r < 0)
2383                         goto out_free_1;
2384         }
2385
2386         r = register_cpu_notifier(&kvm_cpu_notifier);
2387         if (r)
2388                 goto out_free_2;
2389         register_reboot_notifier(&kvm_reboot_notifier);
2390
2391         r = sysdev_class_register(&kvm_sysdev_class);
2392         if (r)
2393                 goto out_free_3;
2394
2395         r = sysdev_register(&kvm_sysdev);
2396         if (r)
2397                 goto out_free_4;
2398
2399         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2400         if (!vcpu_align)
2401                 vcpu_align = __alignof__(struct kvm_vcpu);
2402         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2403                                            0, NULL);
2404         if (!kvm_vcpu_cache) {
2405                 r = -ENOMEM;
2406                 goto out_free_5;
2407         }
2408
2409         r = kvm_async_pf_init();
2410         if (r)
2411                 goto out_free;
2412
2413         kvm_chardev_ops.owner = module;
2414         kvm_vm_fops.owner = module;
2415         kvm_vcpu_fops.owner = module;
2416
2417         r = misc_register(&kvm_dev);
2418         if (r) {
2419                 printk(KERN_ERR "kvm: misc device register failed\n");
2420                 goto out_unreg;
2421         }
2422
2423         kvm_preempt_ops.sched_in = kvm_sched_in;
2424         kvm_preempt_ops.sched_out = kvm_sched_out;
2425
2426         kvm_init_debug();
2427
2428         return 0;
2429
2430 out_unreg:
2431         kvm_async_pf_deinit();
2432 out_free:
2433         kmem_cache_destroy(kvm_vcpu_cache);
2434 out_free_5:
2435         sysdev_unregister(&kvm_sysdev);
2436 out_free_4:
2437         sysdev_class_unregister(&kvm_sysdev_class);
2438 out_free_3:
2439         unregister_reboot_notifier(&kvm_reboot_notifier);
2440         unregister_cpu_notifier(&kvm_cpu_notifier);
2441 out_free_2:
2442 out_free_1:
2443         kvm_arch_hardware_unsetup();
2444 out_free_0a:
2445         free_cpumask_var(cpus_hardware_enabled);
2446 out_free_0:
2447         if (fault_page)
2448                 __free_page(fault_page);
2449         if (hwpoison_page)
2450                 __free_page(hwpoison_page);
2451         __free_page(bad_page);
2452 out:
2453         kvm_arch_exit();
2454 out_fail:
2455         return r;
2456 }
2457 EXPORT_SYMBOL_GPL(kvm_init);
2458
2459 void kvm_exit(void)
2460 {
2461         kvm_exit_debug();
2462         misc_deregister(&kvm_dev);
2463         kmem_cache_destroy(kvm_vcpu_cache);
2464         kvm_async_pf_deinit();
2465         sysdev_unregister(&kvm_sysdev);
2466         sysdev_class_unregister(&kvm_sysdev_class);
2467         unregister_reboot_notifier(&kvm_reboot_notifier);
2468         unregister_cpu_notifier(&kvm_cpu_notifier);
2469         on_each_cpu(hardware_disable, NULL, 1);
2470         kvm_arch_hardware_unsetup();
2471         kvm_arch_exit();
2472         free_cpumask_var(cpus_hardware_enabled);
2473         __free_page(hwpoison_page);
2474         __free_page(bad_page);
2475 }
2476 EXPORT_SYMBOL_GPL(kvm_exit);