Merge branch 'syscore' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/suspen...
[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/syscore_ops.h>
34 #include <linux/cpu.h>
35 #include <linux/sched.h>
36 #include <linux/cpumask.h>
37 #include <linux/smp.h>
38 #include <linux/anon_inodes.h>
39 #include <linux/profile.h>
40 #include <linux/kvm_para.h>
41 #include <linux/pagemap.h>
42 #include <linux/mman.h>
43 #include <linux/swap.h>
44 #include <linux/bitops.h>
45 #include <linux/spinlock.h>
46 #include <linux/compat.h>
47 #include <linux/srcu.h>
48 #include <linux/hugetlb.h>
49 #include <linux/slab.h>
50
51 #include <asm/processor.h>
52 #include <asm/io.h>
53 #include <asm/uaccess.h>
54 #include <asm/pgtable.h>
55
56 #include "coalesced_mmio.h"
57 #include "async_pf.h"
58
59 #define CREATE_TRACE_POINTS
60 #include <trace/events/kvm.h>
61
62 MODULE_AUTHOR("Qumranet");
63 MODULE_LICENSE("GPL");
64
65 /*
66  * Ordering of locks:
67  *
68  *              kvm->lock --> kvm->slots_lock --> kvm->irq_lock
69  */
70
71 DEFINE_RAW_SPINLOCK(kvm_lock);
72 LIST_HEAD(vm_list);
73
74 static cpumask_var_t cpus_hardware_enabled;
75 static int kvm_usage_count = 0;
76 static atomic_t hardware_enable_failed;
77
78 struct kmem_cache *kvm_vcpu_cache;
79 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
80
81 static __read_mostly struct preempt_ops kvm_preempt_ops;
82
83 struct dentry *kvm_debugfs_dir;
84
85 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
86                            unsigned long arg);
87 static int hardware_enable_all(void);
88 static void hardware_disable_all(void);
89
90 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
91
92 bool kvm_rebooting;
93 EXPORT_SYMBOL_GPL(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                 int reserved;
107                 struct page *tail = pfn_to_page(pfn);
108                 struct page *head = compound_trans_head(tail);
109                 reserved = PageReserved(head);
110                 if (head != tail) {
111                         /*
112                          * "head" is not a dangling pointer
113                          * (compound_trans_head takes care of that)
114                          * but the hugepage may have been splitted
115                          * from under us (and we may not hold a
116                          * reference count on the head page so it can
117                          * be reused before we run PageReferenced), so
118                          * we've to check PageTail before returning
119                          * what we just read.
120                          */
121                         smp_rmb();
122                         if (PageTail(tail))
123                                 return reserved;
124                 }
125                 return PageReserved(tail);
126         }
127
128         return true;
129 }
130
131 /*
132  * Switches to specified vcpu, until a matching vcpu_put()
133  */
134 void vcpu_load(struct kvm_vcpu *vcpu)
135 {
136         int cpu;
137
138         mutex_lock(&vcpu->mutex);
139         if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
140                 /* The thread running this VCPU changed. */
141                 struct pid *oldpid = vcpu->pid;
142                 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
143                 rcu_assign_pointer(vcpu->pid, newpid);
144                 synchronize_rcu();
145                 put_pid(oldpid);
146         }
147         cpu = get_cpu();
148         preempt_notifier_register(&vcpu->preempt_notifier);
149         kvm_arch_vcpu_load(vcpu, cpu);
150         put_cpu();
151 }
152
153 void vcpu_put(struct kvm_vcpu *vcpu)
154 {
155         preempt_disable();
156         kvm_arch_vcpu_put(vcpu);
157         preempt_notifier_unregister(&vcpu->preempt_notifier);
158         preempt_enable();
159         mutex_unlock(&vcpu->mutex);
160 }
161
162 static void ack_flush(void *_completed)
163 {
164 }
165
166 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
167 {
168         int i, cpu, me;
169         cpumask_var_t cpus;
170         bool called = true;
171         struct kvm_vcpu *vcpu;
172
173         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
174
175         me = get_cpu();
176         kvm_for_each_vcpu(i, vcpu, kvm) {
177                 kvm_make_request(req, vcpu);
178                 cpu = vcpu->cpu;
179
180                 /* Set ->requests bit before we read ->mode */
181                 smp_mb();
182
183                 if (cpus != NULL && cpu != -1 && cpu != me &&
184                       kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
185                         cpumask_set_cpu(cpu, cpus);
186         }
187         if (unlikely(cpus == NULL))
188                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
189         else if (!cpumask_empty(cpus))
190                 smp_call_function_many(cpus, ack_flush, NULL, 1);
191         else
192                 called = false;
193         put_cpu();
194         free_cpumask_var(cpus);
195         return called;
196 }
197
198 void kvm_flush_remote_tlbs(struct kvm *kvm)
199 {
200         int dirty_count = kvm->tlbs_dirty;
201
202         smp_mb();
203         if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
204                 ++kvm->stat.remote_tlb_flush;
205         cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
206 }
207
208 void kvm_reload_remote_mmus(struct kvm *kvm)
209 {
210         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
211 }
212
213 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
214 {
215         struct page *page;
216         int r;
217
218         mutex_init(&vcpu->mutex);
219         vcpu->cpu = -1;
220         vcpu->kvm = kvm;
221         vcpu->vcpu_id = id;
222         vcpu->pid = NULL;
223         init_waitqueue_head(&vcpu->wq);
224         kvm_async_pf_vcpu_init(vcpu);
225
226         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
227         if (!page) {
228                 r = -ENOMEM;
229                 goto fail;
230         }
231         vcpu->run = page_address(page);
232
233         r = kvm_arch_vcpu_init(vcpu);
234         if (r < 0)
235                 goto fail_free_run;
236         return 0;
237
238 fail_free_run:
239         free_page((unsigned long)vcpu->run);
240 fail:
241         return r;
242 }
243 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
244
245 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
246 {
247         put_pid(vcpu->pid);
248         kvm_arch_vcpu_uninit(vcpu);
249         free_page((unsigned long)vcpu->run);
250 }
251 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
252
253 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
254 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
255 {
256         return container_of(mn, struct kvm, mmu_notifier);
257 }
258
259 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
260                                              struct mm_struct *mm,
261                                              unsigned long address)
262 {
263         struct kvm *kvm = mmu_notifier_to_kvm(mn);
264         int need_tlb_flush, idx;
265
266         /*
267          * When ->invalidate_page runs, the linux pte has been zapped
268          * already but the page is still allocated until
269          * ->invalidate_page returns. So if we increase the sequence
270          * here the kvm page fault will notice if the spte can't be
271          * established because the page is going to be freed. If
272          * instead the kvm page fault establishes the spte before
273          * ->invalidate_page runs, kvm_unmap_hva will release it
274          * before returning.
275          *
276          * The sequence increase only need to be seen at spin_unlock
277          * time, and not at spin_lock time.
278          *
279          * Increasing the sequence after the spin_unlock would be
280          * unsafe because the kvm page fault could then establish the
281          * pte after kvm_unmap_hva returned, without noticing the page
282          * is going to be freed.
283          */
284         idx = srcu_read_lock(&kvm->srcu);
285         spin_lock(&kvm->mmu_lock);
286         kvm->mmu_notifier_seq++;
287         need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
288         spin_unlock(&kvm->mmu_lock);
289         srcu_read_unlock(&kvm->srcu, idx);
290
291         /* we've to flush the tlb before the pages can be freed */
292         if (need_tlb_flush)
293                 kvm_flush_remote_tlbs(kvm);
294
295 }
296
297 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
298                                         struct mm_struct *mm,
299                                         unsigned long address,
300                                         pte_t pte)
301 {
302         struct kvm *kvm = mmu_notifier_to_kvm(mn);
303         int idx;
304
305         idx = srcu_read_lock(&kvm->srcu);
306         spin_lock(&kvm->mmu_lock);
307         kvm->mmu_notifier_seq++;
308         kvm_set_spte_hva(kvm, address, pte);
309         spin_unlock(&kvm->mmu_lock);
310         srcu_read_unlock(&kvm->srcu, idx);
311 }
312
313 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
314                                                     struct mm_struct *mm,
315                                                     unsigned long start,
316                                                     unsigned long end)
317 {
318         struct kvm *kvm = mmu_notifier_to_kvm(mn);
319         int need_tlb_flush = 0, idx;
320
321         idx = srcu_read_lock(&kvm->srcu);
322         spin_lock(&kvm->mmu_lock);
323         /*
324          * The count increase must become visible at unlock time as no
325          * spte can be established without taking the mmu_lock and
326          * count is also read inside the mmu_lock critical section.
327          */
328         kvm->mmu_notifier_count++;
329         for (; start < end; start += PAGE_SIZE)
330                 need_tlb_flush |= kvm_unmap_hva(kvm, start);
331         need_tlb_flush |= kvm->tlbs_dirty;
332         spin_unlock(&kvm->mmu_lock);
333         srcu_read_unlock(&kvm->srcu, idx);
334
335         /* we've to flush the tlb before the pages can be freed */
336         if (need_tlb_flush)
337                 kvm_flush_remote_tlbs(kvm);
338 }
339
340 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
341                                                   struct mm_struct *mm,
342                                                   unsigned long start,
343                                                   unsigned long end)
344 {
345         struct kvm *kvm = mmu_notifier_to_kvm(mn);
346
347         spin_lock(&kvm->mmu_lock);
348         /*
349          * This sequence increase will notify the kvm page fault that
350          * the page that is going to be mapped in the spte could have
351          * been freed.
352          */
353         kvm->mmu_notifier_seq++;
354         /*
355          * The above sequence increase must be visible before the
356          * below count decrease but both values are read by the kvm
357          * page fault under mmu_lock spinlock so we don't need to add
358          * a smb_wmb() here in between the two.
359          */
360         kvm->mmu_notifier_count--;
361         spin_unlock(&kvm->mmu_lock);
362
363         BUG_ON(kvm->mmu_notifier_count < 0);
364 }
365
366 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
367                                               struct mm_struct *mm,
368                                               unsigned long address)
369 {
370         struct kvm *kvm = mmu_notifier_to_kvm(mn);
371         int young, idx;
372
373         idx = srcu_read_lock(&kvm->srcu);
374         spin_lock(&kvm->mmu_lock);
375         young = kvm_age_hva(kvm, address);
376         spin_unlock(&kvm->mmu_lock);
377         srcu_read_unlock(&kvm->srcu, idx);
378
379         if (young)
380                 kvm_flush_remote_tlbs(kvm);
381
382         return young;
383 }
384
385 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
386                                        struct mm_struct *mm,
387                                        unsigned long address)
388 {
389         struct kvm *kvm = mmu_notifier_to_kvm(mn);
390         int young, idx;
391
392         idx = srcu_read_lock(&kvm->srcu);
393         spin_lock(&kvm->mmu_lock);
394         young = kvm_test_age_hva(kvm, address);
395         spin_unlock(&kvm->mmu_lock);
396         srcu_read_unlock(&kvm->srcu, idx);
397
398         return young;
399 }
400
401 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
402                                      struct mm_struct *mm)
403 {
404         struct kvm *kvm = mmu_notifier_to_kvm(mn);
405         int idx;
406
407         idx = srcu_read_lock(&kvm->srcu);
408         kvm_arch_flush_shadow(kvm);
409         srcu_read_unlock(&kvm->srcu, idx);
410 }
411
412 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
413         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
414         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
415         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
416         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
417         .test_young             = kvm_mmu_notifier_test_young,
418         .change_pte             = kvm_mmu_notifier_change_pte,
419         .release                = kvm_mmu_notifier_release,
420 };
421
422 static int kvm_init_mmu_notifier(struct kvm *kvm)
423 {
424         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
425         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
426 }
427
428 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
429
430 static int kvm_init_mmu_notifier(struct kvm *kvm)
431 {
432         return 0;
433 }
434
435 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
436
437 static struct kvm *kvm_create_vm(void)
438 {
439         int r, i;
440         struct kvm *kvm = kvm_arch_alloc_vm();
441
442         if (!kvm)
443                 return ERR_PTR(-ENOMEM);
444
445         r = kvm_arch_init_vm(kvm);
446         if (r)
447                 goto out_err_nodisable;
448
449         r = hardware_enable_all();
450         if (r)
451                 goto out_err_nodisable;
452
453 #ifdef CONFIG_HAVE_KVM_IRQCHIP
454         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
455         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
456 #endif
457
458         r = -ENOMEM;
459         kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
460         if (!kvm->memslots)
461                 goto out_err_nosrcu;
462         if (init_srcu_struct(&kvm->srcu))
463                 goto out_err_nosrcu;
464         for (i = 0; i < KVM_NR_BUSES; i++) {
465                 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
466                                         GFP_KERNEL);
467                 if (!kvm->buses[i])
468                         goto out_err;
469         }
470
471         r = kvm_init_mmu_notifier(kvm);
472         if (r)
473                 goto out_err;
474
475         kvm->mm = current->mm;
476         atomic_inc(&kvm->mm->mm_count);
477         spin_lock_init(&kvm->mmu_lock);
478         kvm_eventfd_init(kvm);
479         mutex_init(&kvm->lock);
480         mutex_init(&kvm->irq_lock);
481         mutex_init(&kvm->slots_lock);
482         atomic_set(&kvm->users_count, 1);
483         raw_spin_lock(&kvm_lock);
484         list_add(&kvm->vm_list, &vm_list);
485         raw_spin_unlock(&kvm_lock);
486
487         return kvm;
488
489 out_err:
490         cleanup_srcu_struct(&kvm->srcu);
491 out_err_nosrcu:
492         hardware_disable_all();
493 out_err_nodisable:
494         for (i = 0; i < KVM_NR_BUSES; i++)
495                 kfree(kvm->buses[i]);
496         kfree(kvm->memslots);
497         kvm_arch_free_vm(kvm);
498         return ERR_PTR(r);
499 }
500
501 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
502 {
503         if (!memslot->dirty_bitmap)
504                 return;
505
506         if (2 * kvm_dirty_bitmap_bytes(memslot) > PAGE_SIZE)
507                 vfree(memslot->dirty_bitmap_head);
508         else
509                 kfree(memslot->dirty_bitmap_head);
510
511         memslot->dirty_bitmap = NULL;
512         memslot->dirty_bitmap_head = NULL;
513 }
514
515 /*
516  * Free any memory in @free but not in @dont.
517  */
518 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
519                                   struct kvm_memory_slot *dont)
520 {
521         int i;
522
523         if (!dont || free->rmap != dont->rmap)
524                 vfree(free->rmap);
525
526         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
527                 kvm_destroy_dirty_bitmap(free);
528
529
530         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
531                 if (!dont || free->lpage_info[i] != dont->lpage_info[i]) {
532                         vfree(free->lpage_info[i]);
533                         free->lpage_info[i] = NULL;
534                 }
535         }
536
537         free->npages = 0;
538         free->rmap = NULL;
539 }
540
541 void kvm_free_physmem(struct kvm *kvm)
542 {
543         int i;
544         struct kvm_memslots *slots = kvm->memslots;
545
546         for (i = 0; i < slots->nmemslots; ++i)
547                 kvm_free_physmem_slot(&slots->memslots[i], NULL);
548
549         kfree(kvm->memslots);
550 }
551
552 static void kvm_destroy_vm(struct kvm *kvm)
553 {
554         int i;
555         struct mm_struct *mm = kvm->mm;
556
557         kvm_arch_sync_events(kvm);
558         raw_spin_lock(&kvm_lock);
559         list_del(&kvm->vm_list);
560         raw_spin_unlock(&kvm_lock);
561         kvm_free_irq_routing(kvm);
562         for (i = 0; i < KVM_NR_BUSES; i++)
563                 kvm_io_bus_destroy(kvm->buses[i]);
564         kvm_coalesced_mmio_free(kvm);
565 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
566         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
567 #else
568         kvm_arch_flush_shadow(kvm);
569 #endif
570         kvm_arch_destroy_vm(kvm);
571         kvm_free_physmem(kvm);
572         cleanup_srcu_struct(&kvm->srcu);
573         kvm_arch_free_vm(kvm);
574         hardware_disable_all();
575         mmdrop(mm);
576 }
577
578 void kvm_get_kvm(struct kvm *kvm)
579 {
580         atomic_inc(&kvm->users_count);
581 }
582 EXPORT_SYMBOL_GPL(kvm_get_kvm);
583
584 void kvm_put_kvm(struct kvm *kvm)
585 {
586         if (atomic_dec_and_test(&kvm->users_count))
587                 kvm_destroy_vm(kvm);
588 }
589 EXPORT_SYMBOL_GPL(kvm_put_kvm);
590
591
592 static int kvm_vm_release(struct inode *inode, struct file *filp)
593 {
594         struct kvm *kvm = filp->private_data;
595
596         kvm_irqfd_release(kvm);
597
598         kvm_put_kvm(kvm);
599         return 0;
600 }
601
602 #ifndef CONFIG_S390
603 /*
604  * Allocation size is twice as large as the actual dirty bitmap size.
605  * This makes it possible to do double buffering: see x86's
606  * kvm_vm_ioctl_get_dirty_log().
607  */
608 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
609 {
610         unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
611
612         if (dirty_bytes > PAGE_SIZE)
613                 memslot->dirty_bitmap = vzalloc(dirty_bytes);
614         else
615                 memslot->dirty_bitmap = kzalloc(dirty_bytes, GFP_KERNEL);
616
617         if (!memslot->dirty_bitmap)
618                 return -ENOMEM;
619
620         memslot->dirty_bitmap_head = memslot->dirty_bitmap;
621         return 0;
622 }
623 #endif /* !CONFIG_S390 */
624
625 /*
626  * Allocate some memory and give it an address in the guest physical address
627  * space.
628  *
629  * Discontiguous memory is allowed, mostly for framebuffers.
630  *
631  * Must be called holding mmap_sem for write.
632  */
633 int __kvm_set_memory_region(struct kvm *kvm,
634                             struct kvm_userspace_memory_region *mem,
635                             int user_alloc)
636 {
637         int r;
638         gfn_t base_gfn;
639         unsigned long npages;
640         unsigned long i;
641         struct kvm_memory_slot *memslot;
642         struct kvm_memory_slot old, new;
643         struct kvm_memslots *slots, *old_memslots;
644
645         r = -EINVAL;
646         /* General sanity checks */
647         if (mem->memory_size & (PAGE_SIZE - 1))
648                 goto out;
649         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
650                 goto out;
651         if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
652                 goto out;
653         if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
654                 goto out;
655         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
656                 goto out;
657
658         memslot = &kvm->memslots->memslots[mem->slot];
659         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
660         npages = mem->memory_size >> PAGE_SHIFT;
661
662         r = -EINVAL;
663         if (npages > KVM_MEM_MAX_NR_PAGES)
664                 goto out;
665
666         if (!npages)
667                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
668
669         new = old = *memslot;
670
671         new.id = mem->slot;
672         new.base_gfn = base_gfn;
673         new.npages = npages;
674         new.flags = mem->flags;
675
676         /* Disallow changing a memory slot's size. */
677         r = -EINVAL;
678         if (npages && old.npages && npages != old.npages)
679                 goto out_free;
680
681         /* Check for overlaps */
682         r = -EEXIST;
683         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
684                 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
685
686                 if (s == memslot || !s->npages)
687                         continue;
688                 if (!((base_gfn + npages <= s->base_gfn) ||
689                       (base_gfn >= s->base_gfn + s->npages)))
690                         goto out_free;
691         }
692
693         /* Free page dirty bitmap if unneeded */
694         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
695                 new.dirty_bitmap = NULL;
696
697         r = -ENOMEM;
698
699         /* Allocate if a slot is being created */
700 #ifndef CONFIG_S390
701         if (npages && !new.rmap) {
702                 new.rmap = vzalloc(npages * sizeof(*new.rmap));
703
704                 if (!new.rmap)
705                         goto out_free;
706
707                 new.user_alloc = user_alloc;
708                 new.userspace_addr = mem->userspace_addr;
709         }
710         if (!npages)
711                 goto skip_lpage;
712
713         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
714                 unsigned long ugfn;
715                 unsigned long j;
716                 int lpages;
717                 int level = i + 2;
718
719                 /* Avoid unused variable warning if no large pages */
720                 (void)level;
721
722                 if (new.lpage_info[i])
723                         continue;
724
725                 lpages = 1 + ((base_gfn + npages - 1)
726                              >> KVM_HPAGE_GFN_SHIFT(level));
727                 lpages -= base_gfn >> KVM_HPAGE_GFN_SHIFT(level);
728
729                 new.lpage_info[i] = vzalloc(lpages * sizeof(*new.lpage_info[i]));
730
731                 if (!new.lpage_info[i])
732                         goto out_free;
733
734                 if (base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
735                         new.lpage_info[i][0].write_count = 1;
736                 if ((base_gfn+npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
737                         new.lpage_info[i][lpages - 1].write_count = 1;
738                 ugfn = new.userspace_addr >> PAGE_SHIFT;
739                 /*
740                  * If the gfn and userspace address are not aligned wrt each
741                  * other, or if explicitly asked to, disable large page
742                  * support for this slot
743                  */
744                 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
745                     !largepages_enabled)
746                         for (j = 0; j < lpages; ++j)
747                                 new.lpage_info[i][j].write_count = 1;
748         }
749
750 skip_lpage:
751
752         /* Allocate page dirty bitmap if needed */
753         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
754                 if (kvm_create_dirty_bitmap(&new) < 0)
755                         goto out_free;
756                 /* destroy any largepage mappings for dirty tracking */
757         }
758 #else  /* not defined CONFIG_S390 */
759         new.user_alloc = user_alloc;
760         if (user_alloc)
761                 new.userspace_addr = mem->userspace_addr;
762 #endif /* not defined CONFIG_S390 */
763
764         if (!npages) {
765                 r = -ENOMEM;
766                 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
767                 if (!slots)
768                         goto out_free;
769                 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
770                 if (mem->slot >= slots->nmemslots)
771                         slots->nmemslots = mem->slot + 1;
772                 slots->generation++;
773                 slots->memslots[mem->slot].flags |= KVM_MEMSLOT_INVALID;
774
775                 old_memslots = kvm->memslots;
776                 rcu_assign_pointer(kvm->memslots, slots);
777                 synchronize_srcu_expedited(&kvm->srcu);
778                 /* From this point no new shadow pages pointing to a deleted
779                  * memslot will be created.
780                  *
781                  * validation of sp->gfn happens in:
782                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
783                  *      - kvm_is_visible_gfn (mmu_check_roots)
784                  */
785                 kvm_arch_flush_shadow(kvm);
786                 kfree(old_memslots);
787         }
788
789         r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
790         if (r)
791                 goto out_free;
792
793         /* map the pages in iommu page table */
794         if (npages) {
795                 r = kvm_iommu_map_pages(kvm, &new);
796                 if (r)
797                         goto out_free;
798         }
799
800         r = -ENOMEM;
801         slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
802         if (!slots)
803                 goto out_free;
804         memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
805         if (mem->slot >= slots->nmemslots)
806                 slots->nmemslots = mem->slot + 1;
807         slots->generation++;
808
809         /* actual memory is freed via old in kvm_free_physmem_slot below */
810         if (!npages) {
811                 new.rmap = NULL;
812                 new.dirty_bitmap = NULL;
813                 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i)
814                         new.lpage_info[i] = NULL;
815         }
816
817         slots->memslots[mem->slot] = new;
818         old_memslots = kvm->memslots;
819         rcu_assign_pointer(kvm->memslots, slots);
820         synchronize_srcu_expedited(&kvm->srcu);
821
822         kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
823
824         kvm_free_physmem_slot(&old, &new);
825         kfree(old_memslots);
826
827         return 0;
828
829 out_free:
830         kvm_free_physmem_slot(&new, &old);
831 out:
832         return r;
833
834 }
835 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
836
837 int kvm_set_memory_region(struct kvm *kvm,
838                           struct kvm_userspace_memory_region *mem,
839                           int user_alloc)
840 {
841         int r;
842
843         mutex_lock(&kvm->slots_lock);
844         r = __kvm_set_memory_region(kvm, mem, user_alloc);
845         mutex_unlock(&kvm->slots_lock);
846         return r;
847 }
848 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
849
850 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
851                                    struct
852                                    kvm_userspace_memory_region *mem,
853                                    int user_alloc)
854 {
855         if (mem->slot >= KVM_MEMORY_SLOTS)
856                 return -EINVAL;
857         return kvm_set_memory_region(kvm, mem, user_alloc);
858 }
859
860 int kvm_get_dirty_log(struct kvm *kvm,
861                         struct kvm_dirty_log *log, int *is_dirty)
862 {
863         struct kvm_memory_slot *memslot;
864         int r, i;
865         unsigned long n;
866         unsigned long any = 0;
867
868         r = -EINVAL;
869         if (log->slot >= KVM_MEMORY_SLOTS)
870                 goto out;
871
872         memslot = &kvm->memslots->memslots[log->slot];
873         r = -ENOENT;
874         if (!memslot->dirty_bitmap)
875                 goto out;
876
877         n = kvm_dirty_bitmap_bytes(memslot);
878
879         for (i = 0; !any && i < n/sizeof(long); ++i)
880                 any = memslot->dirty_bitmap[i];
881
882         r = -EFAULT;
883         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
884                 goto out;
885
886         if (any)
887                 *is_dirty = 1;
888
889         r = 0;
890 out:
891         return r;
892 }
893
894 void kvm_disable_largepages(void)
895 {
896         largepages_enabled = false;
897 }
898 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
899
900 int is_error_page(struct page *page)
901 {
902         return page == bad_page || page == hwpoison_page || page == fault_page;
903 }
904 EXPORT_SYMBOL_GPL(is_error_page);
905
906 int is_error_pfn(pfn_t pfn)
907 {
908         return pfn == bad_pfn || pfn == hwpoison_pfn || pfn == fault_pfn;
909 }
910 EXPORT_SYMBOL_GPL(is_error_pfn);
911
912 int is_hwpoison_pfn(pfn_t pfn)
913 {
914         return pfn == hwpoison_pfn;
915 }
916 EXPORT_SYMBOL_GPL(is_hwpoison_pfn);
917
918 int is_fault_pfn(pfn_t pfn)
919 {
920         return pfn == fault_pfn;
921 }
922 EXPORT_SYMBOL_GPL(is_fault_pfn);
923
924 static inline unsigned long bad_hva(void)
925 {
926         return PAGE_OFFSET;
927 }
928
929 int kvm_is_error_hva(unsigned long addr)
930 {
931         return addr == bad_hva();
932 }
933 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
934
935 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm_memslots *slots,
936                                                 gfn_t gfn)
937 {
938         int i;
939
940         for (i = 0; i < slots->nmemslots; ++i) {
941                 struct kvm_memory_slot *memslot = &slots->memslots[i];
942
943                 if (gfn >= memslot->base_gfn
944                     && gfn < memslot->base_gfn + memslot->npages)
945                         return memslot;
946         }
947         return NULL;
948 }
949
950 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
951 {
952         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
953 }
954 EXPORT_SYMBOL_GPL(gfn_to_memslot);
955
956 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
957 {
958         int i;
959         struct kvm_memslots *slots = kvm_memslots(kvm);
960
961         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
962                 struct kvm_memory_slot *memslot = &slots->memslots[i];
963
964                 if (memslot->flags & KVM_MEMSLOT_INVALID)
965                         continue;
966
967                 if (gfn >= memslot->base_gfn
968                     && gfn < memslot->base_gfn + memslot->npages)
969                         return 1;
970         }
971         return 0;
972 }
973 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
974
975 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
976 {
977         struct vm_area_struct *vma;
978         unsigned long addr, size;
979
980         size = PAGE_SIZE;
981
982         addr = gfn_to_hva(kvm, gfn);
983         if (kvm_is_error_hva(addr))
984                 return PAGE_SIZE;
985
986         down_read(&current->mm->mmap_sem);
987         vma = find_vma(current->mm, addr);
988         if (!vma)
989                 goto out;
990
991         size = vma_kernel_pagesize(vma);
992
993 out:
994         up_read(&current->mm->mmap_sem);
995
996         return size;
997 }
998
999 int memslot_id(struct kvm *kvm, gfn_t gfn)
1000 {
1001         int i;
1002         struct kvm_memslots *slots = kvm_memslots(kvm);
1003         struct kvm_memory_slot *memslot = NULL;
1004
1005         for (i = 0; i < slots->nmemslots; ++i) {
1006                 memslot = &slots->memslots[i];
1007
1008                 if (gfn >= memslot->base_gfn
1009                     && gfn < memslot->base_gfn + memslot->npages)
1010                         break;
1011         }
1012
1013         return memslot - slots->memslots;
1014 }
1015
1016 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1017                                      gfn_t *nr_pages)
1018 {
1019         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1020                 return bad_hva();
1021
1022         if (nr_pages)
1023                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1024
1025         return gfn_to_hva_memslot(slot, gfn);
1026 }
1027
1028 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1029 {
1030         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1031 }
1032 EXPORT_SYMBOL_GPL(gfn_to_hva);
1033
1034 static pfn_t get_fault_pfn(void)
1035 {
1036         get_page(fault_page);
1037         return fault_pfn;
1038 }
1039
1040 static inline int check_user_page_hwpoison(unsigned long addr)
1041 {
1042         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1043
1044         rc = __get_user_pages(current, current->mm, addr, 1,
1045                               flags, NULL, NULL, NULL);
1046         return rc == -EHWPOISON;
1047 }
1048
1049 static pfn_t hva_to_pfn(struct kvm *kvm, unsigned long addr, bool atomic,
1050                         bool *async, bool write_fault, bool *writable)
1051 {
1052         struct page *page[1];
1053         int npages = 0;
1054         pfn_t pfn;
1055
1056         /* we can do it either atomically or asynchronously, not both */
1057         BUG_ON(atomic && async);
1058
1059         BUG_ON(!write_fault && !writable);
1060
1061         if (writable)
1062                 *writable = true;
1063
1064         if (atomic || async)
1065                 npages = __get_user_pages_fast(addr, 1, 1, page);
1066
1067         if (unlikely(npages != 1) && !atomic) {
1068                 might_sleep();
1069
1070                 if (writable)
1071                         *writable = write_fault;
1072
1073                 npages = get_user_pages_fast(addr, 1, write_fault, page);
1074
1075                 /* map read fault as writable if possible */
1076                 if (unlikely(!write_fault) && npages == 1) {
1077                         struct page *wpage[1];
1078
1079                         npages = __get_user_pages_fast(addr, 1, 1, wpage);
1080                         if (npages == 1) {
1081                                 *writable = true;
1082                                 put_page(page[0]);
1083                                 page[0] = wpage[0];
1084                         }
1085                         npages = 1;
1086                 }
1087         }
1088
1089         if (unlikely(npages != 1)) {
1090                 struct vm_area_struct *vma;
1091
1092                 if (atomic)
1093                         return get_fault_pfn();
1094
1095                 down_read(&current->mm->mmap_sem);
1096                 if (check_user_page_hwpoison(addr)) {
1097                         up_read(&current->mm->mmap_sem);
1098                         get_page(hwpoison_page);
1099                         return page_to_pfn(hwpoison_page);
1100                 }
1101
1102                 vma = find_vma_intersection(current->mm, addr, addr+1);
1103
1104                 if (vma == NULL)
1105                         pfn = get_fault_pfn();
1106                 else if ((vma->vm_flags & VM_PFNMAP)) {
1107                         pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1108                                 vma->vm_pgoff;
1109                         BUG_ON(!kvm_is_mmio_pfn(pfn));
1110                 } else {
1111                         if (async && (vma->vm_flags & VM_WRITE))
1112                                 *async = true;
1113                         pfn = get_fault_pfn();
1114                 }
1115                 up_read(&current->mm->mmap_sem);
1116         } else
1117                 pfn = page_to_pfn(page[0]);
1118
1119         return pfn;
1120 }
1121
1122 pfn_t hva_to_pfn_atomic(struct kvm *kvm, unsigned long addr)
1123 {
1124         return hva_to_pfn(kvm, addr, true, NULL, true, NULL);
1125 }
1126 EXPORT_SYMBOL_GPL(hva_to_pfn_atomic);
1127
1128 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1129                           bool write_fault, bool *writable)
1130 {
1131         unsigned long addr;
1132
1133         if (async)
1134                 *async = false;
1135
1136         addr = gfn_to_hva(kvm, gfn);
1137         if (kvm_is_error_hva(addr)) {
1138                 get_page(bad_page);
1139                 return page_to_pfn(bad_page);
1140         }
1141
1142         return hva_to_pfn(kvm, addr, atomic, async, write_fault, writable);
1143 }
1144
1145 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1146 {
1147         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1148 }
1149 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1150
1151 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1152                        bool write_fault, bool *writable)
1153 {
1154         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1155 }
1156 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1157
1158 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1159 {
1160         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1161 }
1162 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1163
1164 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1165                       bool *writable)
1166 {
1167         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1168 }
1169 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1170
1171 pfn_t gfn_to_pfn_memslot(struct kvm *kvm,
1172                          struct kvm_memory_slot *slot, gfn_t gfn)
1173 {
1174         unsigned long addr = gfn_to_hva_memslot(slot, gfn);
1175         return hva_to_pfn(kvm, addr, false, NULL, true, NULL);
1176 }
1177
1178 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1179                                                                   int nr_pages)
1180 {
1181         unsigned long addr;
1182         gfn_t entry;
1183
1184         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1185         if (kvm_is_error_hva(addr))
1186                 return -1;
1187
1188         if (entry < nr_pages)
1189                 return 0;
1190
1191         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1192 }
1193 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1194
1195 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1196 {
1197         pfn_t pfn;
1198
1199         pfn = gfn_to_pfn(kvm, gfn);
1200         if (!kvm_is_mmio_pfn(pfn))
1201                 return pfn_to_page(pfn);
1202
1203         WARN_ON(kvm_is_mmio_pfn(pfn));
1204
1205         get_page(bad_page);
1206         return bad_page;
1207 }
1208
1209 EXPORT_SYMBOL_GPL(gfn_to_page);
1210
1211 void kvm_release_page_clean(struct page *page)
1212 {
1213         kvm_release_pfn_clean(page_to_pfn(page));
1214 }
1215 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1216
1217 void kvm_release_pfn_clean(pfn_t pfn)
1218 {
1219         if (!kvm_is_mmio_pfn(pfn))
1220                 put_page(pfn_to_page(pfn));
1221 }
1222 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1223
1224 void kvm_release_page_dirty(struct page *page)
1225 {
1226         kvm_release_pfn_dirty(page_to_pfn(page));
1227 }
1228 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1229
1230 void kvm_release_pfn_dirty(pfn_t pfn)
1231 {
1232         kvm_set_pfn_dirty(pfn);
1233         kvm_release_pfn_clean(pfn);
1234 }
1235 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1236
1237 void kvm_set_page_dirty(struct page *page)
1238 {
1239         kvm_set_pfn_dirty(page_to_pfn(page));
1240 }
1241 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1242
1243 void kvm_set_pfn_dirty(pfn_t pfn)
1244 {
1245         if (!kvm_is_mmio_pfn(pfn)) {
1246                 struct page *page = pfn_to_page(pfn);
1247                 if (!PageReserved(page))
1248                         SetPageDirty(page);
1249         }
1250 }
1251 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1252
1253 void kvm_set_pfn_accessed(pfn_t pfn)
1254 {
1255         if (!kvm_is_mmio_pfn(pfn))
1256                 mark_page_accessed(pfn_to_page(pfn));
1257 }
1258 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1259
1260 void kvm_get_pfn(pfn_t pfn)
1261 {
1262         if (!kvm_is_mmio_pfn(pfn))
1263                 get_page(pfn_to_page(pfn));
1264 }
1265 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1266
1267 static int next_segment(unsigned long len, int offset)
1268 {
1269         if (len > PAGE_SIZE - offset)
1270                 return PAGE_SIZE - offset;
1271         else
1272                 return len;
1273 }
1274
1275 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1276                         int len)
1277 {
1278         int r;
1279         unsigned long addr;
1280
1281         addr = gfn_to_hva(kvm, gfn);
1282         if (kvm_is_error_hva(addr))
1283                 return -EFAULT;
1284         r = copy_from_user(data, (void __user *)addr + offset, len);
1285         if (r)
1286                 return -EFAULT;
1287         return 0;
1288 }
1289 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1290
1291 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1292 {
1293         gfn_t gfn = gpa >> PAGE_SHIFT;
1294         int seg;
1295         int offset = offset_in_page(gpa);
1296         int ret;
1297
1298         while ((seg = next_segment(len, offset)) != 0) {
1299                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1300                 if (ret < 0)
1301                         return ret;
1302                 offset = 0;
1303                 len -= seg;
1304                 data += seg;
1305                 ++gfn;
1306         }
1307         return 0;
1308 }
1309 EXPORT_SYMBOL_GPL(kvm_read_guest);
1310
1311 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1312                           unsigned long len)
1313 {
1314         int r;
1315         unsigned long addr;
1316         gfn_t gfn = gpa >> PAGE_SHIFT;
1317         int offset = offset_in_page(gpa);
1318
1319         addr = gfn_to_hva(kvm, gfn);
1320         if (kvm_is_error_hva(addr))
1321                 return -EFAULT;
1322         pagefault_disable();
1323         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1324         pagefault_enable();
1325         if (r)
1326                 return -EFAULT;
1327         return 0;
1328 }
1329 EXPORT_SYMBOL(kvm_read_guest_atomic);
1330
1331 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1332                          int offset, int len)
1333 {
1334         int r;
1335         unsigned long addr;
1336
1337         addr = gfn_to_hva(kvm, gfn);
1338         if (kvm_is_error_hva(addr))
1339                 return -EFAULT;
1340         r = copy_to_user((void __user *)addr + offset, data, len);
1341         if (r)
1342                 return -EFAULT;
1343         mark_page_dirty(kvm, gfn);
1344         return 0;
1345 }
1346 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1347
1348 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1349                     unsigned long len)
1350 {
1351         gfn_t gfn = gpa >> PAGE_SHIFT;
1352         int seg;
1353         int offset = offset_in_page(gpa);
1354         int ret;
1355
1356         while ((seg = next_segment(len, offset)) != 0) {
1357                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1358                 if (ret < 0)
1359                         return ret;
1360                 offset = 0;
1361                 len -= seg;
1362                 data += seg;
1363                 ++gfn;
1364         }
1365         return 0;
1366 }
1367
1368 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1369                               gpa_t gpa)
1370 {
1371         struct kvm_memslots *slots = kvm_memslots(kvm);
1372         int offset = offset_in_page(gpa);
1373         gfn_t gfn = gpa >> PAGE_SHIFT;
1374
1375         ghc->gpa = gpa;
1376         ghc->generation = slots->generation;
1377         ghc->memslot = __gfn_to_memslot(slots, gfn);
1378         ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1379         if (!kvm_is_error_hva(ghc->hva))
1380                 ghc->hva += offset;
1381         else
1382                 return -EFAULT;
1383
1384         return 0;
1385 }
1386 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1387
1388 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1389                            void *data, unsigned long len)
1390 {
1391         struct kvm_memslots *slots = kvm_memslots(kvm);
1392         int r;
1393
1394         if (slots->generation != ghc->generation)
1395                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1396
1397         if (kvm_is_error_hva(ghc->hva))
1398                 return -EFAULT;
1399
1400         r = copy_to_user((void __user *)ghc->hva, data, len);
1401         if (r)
1402                 return -EFAULT;
1403         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1404
1405         return 0;
1406 }
1407 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1408
1409 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1410 {
1411         return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1412                                     offset, len);
1413 }
1414 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1415
1416 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1417 {
1418         gfn_t gfn = gpa >> PAGE_SHIFT;
1419         int seg;
1420         int offset = offset_in_page(gpa);
1421         int ret;
1422
1423         while ((seg = next_segment(len, offset)) != 0) {
1424                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1425                 if (ret < 0)
1426                         return ret;
1427                 offset = 0;
1428                 len -= seg;
1429                 ++gfn;
1430         }
1431         return 0;
1432 }
1433 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1434
1435 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1436                              gfn_t gfn)
1437 {
1438         if (memslot && memslot->dirty_bitmap) {
1439                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1440
1441                 __set_bit_le(rel_gfn, memslot->dirty_bitmap);
1442         }
1443 }
1444
1445 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1446 {
1447         struct kvm_memory_slot *memslot;
1448
1449         memslot = gfn_to_memslot(kvm, gfn);
1450         mark_page_dirty_in_slot(kvm, memslot, gfn);
1451 }
1452
1453 /*
1454  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1455  */
1456 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1457 {
1458         DEFINE_WAIT(wait);
1459
1460         for (;;) {
1461                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1462
1463                 if (kvm_arch_vcpu_runnable(vcpu)) {
1464                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1465                         break;
1466                 }
1467                 if (kvm_cpu_has_pending_timer(vcpu))
1468                         break;
1469                 if (signal_pending(current))
1470                         break;
1471
1472                 schedule();
1473         }
1474
1475         finish_wait(&vcpu->wq, &wait);
1476 }
1477
1478 void kvm_resched(struct kvm_vcpu *vcpu)
1479 {
1480         if (!need_resched())
1481                 return;
1482         cond_resched();
1483 }
1484 EXPORT_SYMBOL_GPL(kvm_resched);
1485
1486 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1487 {
1488         struct kvm *kvm = me->kvm;
1489         struct kvm_vcpu *vcpu;
1490         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1491         int yielded = 0;
1492         int pass;
1493         int i;
1494
1495         /*
1496          * We boost the priority of a VCPU that is runnable but not
1497          * currently running, because it got preempted by something
1498          * else and called schedule in __vcpu_run.  Hopefully that
1499          * VCPU is holding the lock that we need and will release it.
1500          * We approximate round-robin by starting at the last boosted VCPU.
1501          */
1502         for (pass = 0; pass < 2 && !yielded; pass++) {
1503                 kvm_for_each_vcpu(i, vcpu, kvm) {
1504                         struct task_struct *task = NULL;
1505                         struct pid *pid;
1506                         if (!pass && i < last_boosted_vcpu) {
1507                                 i = last_boosted_vcpu;
1508                                 continue;
1509                         } else if (pass && i > last_boosted_vcpu)
1510                                 break;
1511                         if (vcpu == me)
1512                                 continue;
1513                         if (waitqueue_active(&vcpu->wq))
1514                                 continue;
1515                         rcu_read_lock();
1516                         pid = rcu_dereference(vcpu->pid);
1517                         if (pid)
1518                                 task = get_pid_task(vcpu->pid, PIDTYPE_PID);
1519                         rcu_read_unlock();
1520                         if (!task)
1521                                 continue;
1522                         if (task->flags & PF_VCPU) {
1523                                 put_task_struct(task);
1524                                 continue;
1525                         }
1526                         if (yield_to(task, 1)) {
1527                                 put_task_struct(task);
1528                                 kvm->last_boosted_vcpu = i;
1529                                 yielded = 1;
1530                                 break;
1531                         }
1532                         put_task_struct(task);
1533                 }
1534         }
1535 }
1536 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1537
1538 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1539 {
1540         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1541         struct page *page;
1542
1543         if (vmf->pgoff == 0)
1544                 page = virt_to_page(vcpu->run);
1545 #ifdef CONFIG_X86
1546         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1547                 page = virt_to_page(vcpu->arch.pio_data);
1548 #endif
1549 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1550         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1551                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1552 #endif
1553         else
1554                 return VM_FAULT_SIGBUS;
1555         get_page(page);
1556         vmf->page = page;
1557         return 0;
1558 }
1559
1560 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1561         .fault = kvm_vcpu_fault,
1562 };
1563
1564 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1565 {
1566         vma->vm_ops = &kvm_vcpu_vm_ops;
1567         return 0;
1568 }
1569
1570 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1571 {
1572         struct kvm_vcpu *vcpu = filp->private_data;
1573
1574         kvm_put_kvm(vcpu->kvm);
1575         return 0;
1576 }
1577
1578 static struct file_operations kvm_vcpu_fops = {
1579         .release        = kvm_vcpu_release,
1580         .unlocked_ioctl = kvm_vcpu_ioctl,
1581         .compat_ioctl   = kvm_vcpu_ioctl,
1582         .mmap           = kvm_vcpu_mmap,
1583         .llseek         = noop_llseek,
1584 };
1585
1586 /*
1587  * Allocates an inode for the vcpu.
1588  */
1589 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1590 {
1591         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1592 }
1593
1594 /*
1595  * Creates some virtual cpus.  Good luck creating more than one.
1596  */
1597 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1598 {
1599         int r;
1600         struct kvm_vcpu *vcpu, *v;
1601
1602         vcpu = kvm_arch_vcpu_create(kvm, id);
1603         if (IS_ERR(vcpu))
1604                 return PTR_ERR(vcpu);
1605
1606         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1607
1608         r = kvm_arch_vcpu_setup(vcpu);
1609         if (r)
1610                 return r;
1611
1612         mutex_lock(&kvm->lock);
1613         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1614                 r = -EINVAL;
1615                 goto vcpu_destroy;
1616         }
1617
1618         kvm_for_each_vcpu(r, v, kvm)
1619                 if (v->vcpu_id == id) {
1620                         r = -EEXIST;
1621                         goto vcpu_destroy;
1622                 }
1623
1624         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1625
1626         /* Now it's all set up, let userspace reach it */
1627         kvm_get_kvm(kvm);
1628         r = create_vcpu_fd(vcpu);
1629         if (r < 0) {
1630                 kvm_put_kvm(kvm);
1631                 goto vcpu_destroy;
1632         }
1633
1634         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1635         smp_wmb();
1636         atomic_inc(&kvm->online_vcpus);
1637
1638 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1639         if (kvm->bsp_vcpu_id == id)
1640                 kvm->bsp_vcpu = vcpu;
1641 #endif
1642         mutex_unlock(&kvm->lock);
1643         return r;
1644
1645 vcpu_destroy:
1646         mutex_unlock(&kvm->lock);
1647         kvm_arch_vcpu_destroy(vcpu);
1648         return r;
1649 }
1650
1651 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1652 {
1653         if (sigset) {
1654                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1655                 vcpu->sigset_active = 1;
1656                 vcpu->sigset = *sigset;
1657         } else
1658                 vcpu->sigset_active = 0;
1659         return 0;
1660 }
1661
1662 static long kvm_vcpu_ioctl(struct file *filp,
1663                            unsigned int ioctl, unsigned long arg)
1664 {
1665         struct kvm_vcpu *vcpu = filp->private_data;
1666         void __user *argp = (void __user *)arg;
1667         int r;
1668         struct kvm_fpu *fpu = NULL;
1669         struct kvm_sregs *kvm_sregs = NULL;
1670
1671         if (vcpu->kvm->mm != current->mm)
1672                 return -EIO;
1673
1674 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1675         /*
1676          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1677          * so vcpu_load() would break it.
1678          */
1679         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1680                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1681 #endif
1682
1683
1684         vcpu_load(vcpu);
1685         switch (ioctl) {
1686         case KVM_RUN:
1687                 r = -EINVAL;
1688                 if (arg)
1689                         goto out;
1690                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1691                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1692                 break;
1693         case KVM_GET_REGS: {
1694                 struct kvm_regs *kvm_regs;
1695
1696                 r = -ENOMEM;
1697                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1698                 if (!kvm_regs)
1699                         goto out;
1700                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1701                 if (r)
1702                         goto out_free1;
1703                 r = -EFAULT;
1704                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1705                         goto out_free1;
1706                 r = 0;
1707 out_free1:
1708                 kfree(kvm_regs);
1709                 break;
1710         }
1711         case KVM_SET_REGS: {
1712                 struct kvm_regs *kvm_regs;
1713
1714                 r = -ENOMEM;
1715                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1716                 if (!kvm_regs)
1717                         goto out;
1718                 r = -EFAULT;
1719                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1720                         goto out_free2;
1721                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1722                 if (r)
1723                         goto out_free2;
1724                 r = 0;
1725 out_free2:
1726                 kfree(kvm_regs);
1727                 break;
1728         }
1729         case KVM_GET_SREGS: {
1730                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1731                 r = -ENOMEM;
1732                 if (!kvm_sregs)
1733                         goto out;
1734                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1735                 if (r)
1736                         goto out;
1737                 r = -EFAULT;
1738                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1739                         goto out;
1740                 r = 0;
1741                 break;
1742         }
1743         case KVM_SET_SREGS: {
1744                 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1745                 r = -ENOMEM;
1746                 if (!kvm_sregs)
1747                         goto out;
1748                 r = -EFAULT;
1749                 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1750                         goto out;
1751                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1752                 if (r)
1753                         goto out;
1754                 r = 0;
1755                 break;
1756         }
1757         case KVM_GET_MP_STATE: {
1758                 struct kvm_mp_state mp_state;
1759
1760                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1761                 if (r)
1762                         goto out;
1763                 r = -EFAULT;
1764                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1765                         goto out;
1766                 r = 0;
1767                 break;
1768         }
1769         case KVM_SET_MP_STATE: {
1770                 struct kvm_mp_state mp_state;
1771
1772                 r = -EFAULT;
1773                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1774                         goto out;
1775                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1776                 if (r)
1777                         goto out;
1778                 r = 0;
1779                 break;
1780         }
1781         case KVM_TRANSLATE: {
1782                 struct kvm_translation tr;
1783
1784                 r = -EFAULT;
1785                 if (copy_from_user(&tr, argp, sizeof tr))
1786                         goto out;
1787                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1788                 if (r)
1789                         goto out;
1790                 r = -EFAULT;
1791                 if (copy_to_user(argp, &tr, sizeof tr))
1792                         goto out;
1793                 r = 0;
1794                 break;
1795         }
1796         case KVM_SET_GUEST_DEBUG: {
1797                 struct kvm_guest_debug dbg;
1798
1799                 r = -EFAULT;
1800                 if (copy_from_user(&dbg, argp, sizeof dbg))
1801                         goto out;
1802                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1803                 if (r)
1804                         goto out;
1805                 r = 0;
1806                 break;
1807         }
1808         case KVM_SET_SIGNAL_MASK: {
1809                 struct kvm_signal_mask __user *sigmask_arg = argp;
1810                 struct kvm_signal_mask kvm_sigmask;
1811                 sigset_t sigset, *p;
1812
1813                 p = NULL;
1814                 if (argp) {
1815                         r = -EFAULT;
1816                         if (copy_from_user(&kvm_sigmask, argp,
1817                                            sizeof kvm_sigmask))
1818                                 goto out;
1819                         r = -EINVAL;
1820                         if (kvm_sigmask.len != sizeof sigset)
1821                                 goto out;
1822                         r = -EFAULT;
1823                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1824                                            sizeof sigset))
1825                                 goto out;
1826                         p = &sigset;
1827                 }
1828                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
1829                 break;
1830         }
1831         case KVM_GET_FPU: {
1832                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1833                 r = -ENOMEM;
1834                 if (!fpu)
1835                         goto out;
1836                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1837                 if (r)
1838                         goto out;
1839                 r = -EFAULT;
1840                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1841                         goto out;
1842                 r = 0;
1843                 break;
1844         }
1845         case KVM_SET_FPU: {
1846                 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1847                 r = -ENOMEM;
1848                 if (!fpu)
1849                         goto out;
1850                 r = -EFAULT;
1851                 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1852                         goto out;
1853                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1854                 if (r)
1855                         goto out;
1856                 r = 0;
1857                 break;
1858         }
1859         default:
1860                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1861         }
1862 out:
1863         vcpu_put(vcpu);
1864         kfree(fpu);
1865         kfree(kvm_sregs);
1866         return r;
1867 }
1868
1869 static long kvm_vm_ioctl(struct file *filp,
1870                            unsigned int ioctl, unsigned long arg)
1871 {
1872         struct kvm *kvm = filp->private_data;
1873         void __user *argp = (void __user *)arg;
1874         int r;
1875
1876         if (kvm->mm != current->mm)
1877                 return -EIO;
1878         switch (ioctl) {
1879         case KVM_CREATE_VCPU:
1880                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1881                 if (r < 0)
1882                         goto out;
1883                 break;
1884         case KVM_SET_USER_MEMORY_REGION: {
1885                 struct kvm_userspace_memory_region kvm_userspace_mem;
1886
1887                 r = -EFAULT;
1888                 if (copy_from_user(&kvm_userspace_mem, argp,
1889                                                 sizeof kvm_userspace_mem))
1890                         goto out;
1891
1892                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1893                 if (r)
1894                         goto out;
1895                 break;
1896         }
1897         case KVM_GET_DIRTY_LOG: {
1898                 struct kvm_dirty_log log;
1899
1900                 r = -EFAULT;
1901                 if (copy_from_user(&log, argp, sizeof log))
1902                         goto out;
1903                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1904                 if (r)
1905                         goto out;
1906                 break;
1907         }
1908 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1909         case KVM_REGISTER_COALESCED_MMIO: {
1910                 struct kvm_coalesced_mmio_zone zone;
1911                 r = -EFAULT;
1912                 if (copy_from_user(&zone, argp, sizeof zone))
1913                         goto out;
1914                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1915                 if (r)
1916                         goto out;
1917                 r = 0;
1918                 break;
1919         }
1920         case KVM_UNREGISTER_COALESCED_MMIO: {
1921                 struct kvm_coalesced_mmio_zone zone;
1922                 r = -EFAULT;
1923                 if (copy_from_user(&zone, argp, sizeof zone))
1924                         goto out;
1925                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1926                 if (r)
1927                         goto out;
1928                 r = 0;
1929                 break;
1930         }
1931 #endif
1932         case KVM_IRQFD: {
1933                 struct kvm_irqfd data;
1934
1935                 r = -EFAULT;
1936                 if (copy_from_user(&data, argp, sizeof data))
1937                         goto out;
1938                 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
1939                 break;
1940         }
1941         case KVM_IOEVENTFD: {
1942                 struct kvm_ioeventfd data;
1943
1944                 r = -EFAULT;
1945                 if (copy_from_user(&data, argp, sizeof data))
1946                         goto out;
1947                 r = kvm_ioeventfd(kvm, &data);
1948                 break;
1949         }
1950 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1951         case KVM_SET_BOOT_CPU_ID:
1952                 r = 0;
1953                 mutex_lock(&kvm->lock);
1954                 if (atomic_read(&kvm->online_vcpus) != 0)
1955                         r = -EBUSY;
1956                 else
1957                         kvm->bsp_vcpu_id = arg;
1958                 mutex_unlock(&kvm->lock);
1959                 break;
1960 #endif
1961         default:
1962                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1963                 if (r == -ENOTTY)
1964                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
1965         }
1966 out:
1967         return r;
1968 }
1969
1970 #ifdef CONFIG_COMPAT
1971 struct compat_kvm_dirty_log {
1972         __u32 slot;
1973         __u32 padding1;
1974         union {
1975                 compat_uptr_t dirty_bitmap; /* one bit per page */
1976                 __u64 padding2;
1977         };
1978 };
1979
1980 static long kvm_vm_compat_ioctl(struct file *filp,
1981                            unsigned int ioctl, unsigned long arg)
1982 {
1983         struct kvm *kvm = filp->private_data;
1984         int r;
1985
1986         if (kvm->mm != current->mm)
1987                 return -EIO;
1988         switch (ioctl) {
1989         case KVM_GET_DIRTY_LOG: {
1990                 struct compat_kvm_dirty_log compat_log;
1991                 struct kvm_dirty_log log;
1992
1993                 r = -EFAULT;
1994                 if (copy_from_user(&compat_log, (void __user *)arg,
1995                                    sizeof(compat_log)))
1996                         goto out;
1997                 log.slot         = compat_log.slot;
1998                 log.padding1     = compat_log.padding1;
1999                 log.padding2     = compat_log.padding2;
2000                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2001
2002                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2003                 if (r)
2004                         goto out;
2005                 break;
2006         }
2007         default:
2008                 r = kvm_vm_ioctl(filp, ioctl, arg);
2009         }
2010
2011 out:
2012         return r;
2013 }
2014 #endif
2015
2016 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2017 {
2018         struct page *page[1];
2019         unsigned long addr;
2020         int npages;
2021         gfn_t gfn = vmf->pgoff;
2022         struct kvm *kvm = vma->vm_file->private_data;
2023
2024         addr = gfn_to_hva(kvm, gfn);
2025         if (kvm_is_error_hva(addr))
2026                 return VM_FAULT_SIGBUS;
2027
2028         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2029                                 NULL);
2030         if (unlikely(npages != 1))
2031                 return VM_FAULT_SIGBUS;
2032
2033         vmf->page = page[0];
2034         return 0;
2035 }
2036
2037 static const struct vm_operations_struct kvm_vm_vm_ops = {
2038         .fault = kvm_vm_fault,
2039 };
2040
2041 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2042 {
2043         vma->vm_ops = &kvm_vm_vm_ops;
2044         return 0;
2045 }
2046
2047 static struct file_operations kvm_vm_fops = {
2048         .release        = kvm_vm_release,
2049         .unlocked_ioctl = kvm_vm_ioctl,
2050 #ifdef CONFIG_COMPAT
2051         .compat_ioctl   = kvm_vm_compat_ioctl,
2052 #endif
2053         .mmap           = kvm_vm_mmap,
2054         .llseek         = noop_llseek,
2055 };
2056
2057 static int kvm_dev_ioctl_create_vm(void)
2058 {
2059         int r;
2060         struct kvm *kvm;
2061
2062         kvm = kvm_create_vm();
2063         if (IS_ERR(kvm))
2064                 return PTR_ERR(kvm);
2065 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2066         r = kvm_coalesced_mmio_init(kvm);
2067         if (r < 0) {
2068                 kvm_put_kvm(kvm);
2069                 return r;
2070         }
2071 #endif
2072         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2073         if (r < 0)
2074                 kvm_put_kvm(kvm);
2075
2076         return r;
2077 }
2078
2079 static long kvm_dev_ioctl_check_extension_generic(long arg)
2080 {
2081         switch (arg) {
2082         case KVM_CAP_USER_MEMORY:
2083         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2084         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2085 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2086         case KVM_CAP_SET_BOOT_CPU_ID:
2087 #endif
2088         case KVM_CAP_INTERNAL_ERROR_DATA:
2089                 return 1;
2090 #ifdef CONFIG_HAVE_KVM_IRQCHIP
2091         case KVM_CAP_IRQ_ROUTING:
2092                 return KVM_MAX_IRQ_ROUTES;
2093 #endif
2094         default:
2095                 break;
2096         }
2097         return kvm_dev_ioctl_check_extension(arg);
2098 }
2099
2100 static long kvm_dev_ioctl(struct file *filp,
2101                           unsigned int ioctl, unsigned long arg)
2102 {
2103         long r = -EINVAL;
2104
2105         switch (ioctl) {
2106         case KVM_GET_API_VERSION:
2107                 r = -EINVAL;
2108                 if (arg)
2109                         goto out;
2110                 r = KVM_API_VERSION;
2111                 break;
2112         case KVM_CREATE_VM:
2113                 r = -EINVAL;
2114                 if (arg)
2115                         goto out;
2116                 r = kvm_dev_ioctl_create_vm();
2117                 break;
2118         case KVM_CHECK_EXTENSION:
2119                 r = kvm_dev_ioctl_check_extension_generic(arg);
2120                 break;
2121         case KVM_GET_VCPU_MMAP_SIZE:
2122                 r = -EINVAL;
2123                 if (arg)
2124                         goto out;
2125                 r = PAGE_SIZE;     /* struct kvm_run */
2126 #ifdef CONFIG_X86
2127                 r += PAGE_SIZE;    /* pio data page */
2128 #endif
2129 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2130                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2131 #endif
2132                 break;
2133         case KVM_TRACE_ENABLE:
2134         case KVM_TRACE_PAUSE:
2135         case KVM_TRACE_DISABLE:
2136                 r = -EOPNOTSUPP;
2137                 break;
2138         default:
2139                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2140         }
2141 out:
2142         return r;
2143 }
2144
2145 static struct file_operations kvm_chardev_ops = {
2146         .unlocked_ioctl = kvm_dev_ioctl,
2147         .compat_ioctl   = kvm_dev_ioctl,
2148         .llseek         = noop_llseek,
2149 };
2150
2151 static struct miscdevice kvm_dev = {
2152         KVM_MINOR,
2153         "kvm",
2154         &kvm_chardev_ops,
2155 };
2156
2157 static void hardware_enable_nolock(void *junk)
2158 {
2159         int cpu = raw_smp_processor_id();
2160         int r;
2161
2162         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2163                 return;
2164
2165         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2166
2167         r = kvm_arch_hardware_enable(NULL);
2168
2169         if (r) {
2170                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2171                 atomic_inc(&hardware_enable_failed);
2172                 printk(KERN_INFO "kvm: enabling virtualization on "
2173                                  "CPU%d failed\n", cpu);
2174         }
2175 }
2176
2177 static void hardware_enable(void *junk)
2178 {
2179         raw_spin_lock(&kvm_lock);
2180         hardware_enable_nolock(junk);
2181         raw_spin_unlock(&kvm_lock);
2182 }
2183
2184 static void hardware_disable_nolock(void *junk)
2185 {
2186         int cpu = raw_smp_processor_id();
2187
2188         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2189                 return;
2190         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2191         kvm_arch_hardware_disable(NULL);
2192 }
2193
2194 static void hardware_disable(void *junk)
2195 {
2196         raw_spin_lock(&kvm_lock);
2197         hardware_disable_nolock(junk);
2198         raw_spin_unlock(&kvm_lock);
2199 }
2200
2201 static void hardware_disable_all_nolock(void)
2202 {
2203         BUG_ON(!kvm_usage_count);
2204
2205         kvm_usage_count--;
2206         if (!kvm_usage_count)
2207                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2208 }
2209
2210 static void hardware_disable_all(void)
2211 {
2212         raw_spin_lock(&kvm_lock);
2213         hardware_disable_all_nolock();
2214         raw_spin_unlock(&kvm_lock);
2215 }
2216
2217 static int hardware_enable_all(void)
2218 {
2219         int r = 0;
2220
2221         raw_spin_lock(&kvm_lock);
2222
2223         kvm_usage_count++;
2224         if (kvm_usage_count == 1) {
2225                 atomic_set(&hardware_enable_failed, 0);
2226                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2227
2228                 if (atomic_read(&hardware_enable_failed)) {
2229                         hardware_disable_all_nolock();
2230                         r = -EBUSY;
2231                 }
2232         }
2233
2234         raw_spin_unlock(&kvm_lock);
2235
2236         return r;
2237 }
2238
2239 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2240                            void *v)
2241 {
2242         int cpu = (long)v;
2243
2244         if (!kvm_usage_count)
2245                 return NOTIFY_OK;
2246
2247         val &= ~CPU_TASKS_FROZEN;
2248         switch (val) {
2249         case CPU_DYING:
2250                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2251                        cpu);
2252                 hardware_disable(NULL);
2253                 break;
2254         case CPU_STARTING:
2255                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2256                        cpu);
2257                 hardware_enable(NULL);
2258                 break;
2259         }
2260         return NOTIFY_OK;
2261 }
2262
2263
2264 asmlinkage void kvm_spurious_fault(void)
2265 {
2266         /* Fault while not rebooting.  We want the trace. */
2267         BUG();
2268 }
2269 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
2270
2271 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2272                       void *v)
2273 {
2274         /*
2275          * Some (well, at least mine) BIOSes hang on reboot if
2276          * in vmx root mode.
2277          *
2278          * And Intel TXT required VMX off for all cpu when system shutdown.
2279          */
2280         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2281         kvm_rebooting = true;
2282         on_each_cpu(hardware_disable_nolock, NULL, 1);
2283         return NOTIFY_OK;
2284 }
2285
2286 static struct notifier_block kvm_reboot_notifier = {
2287         .notifier_call = kvm_reboot,
2288         .priority = 0,
2289 };
2290
2291 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2292 {
2293         int i;
2294
2295         for (i = 0; i < bus->dev_count; i++) {
2296                 struct kvm_io_device *pos = bus->devs[i];
2297
2298                 kvm_iodevice_destructor(pos);
2299         }
2300         kfree(bus);
2301 }
2302
2303 /* kvm_io_bus_write - called under kvm->slots_lock */
2304 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2305                      int len, const void *val)
2306 {
2307         int i;
2308         struct kvm_io_bus *bus;
2309
2310         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2311         for (i = 0; i < bus->dev_count; i++)
2312                 if (!kvm_iodevice_write(bus->devs[i], addr, len, val))
2313                         return 0;
2314         return -EOPNOTSUPP;
2315 }
2316
2317 /* kvm_io_bus_read - called under kvm->slots_lock */
2318 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2319                     int len, void *val)
2320 {
2321         int i;
2322         struct kvm_io_bus *bus;
2323
2324         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2325         for (i = 0; i < bus->dev_count; i++)
2326                 if (!kvm_iodevice_read(bus->devs[i], addr, len, val))
2327                         return 0;
2328         return -EOPNOTSUPP;
2329 }
2330
2331 /* Caller must hold slots_lock. */
2332 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2333                             struct kvm_io_device *dev)
2334 {
2335         struct kvm_io_bus *new_bus, *bus;
2336
2337         bus = kvm->buses[bus_idx];
2338         if (bus->dev_count > NR_IOBUS_DEVS-1)
2339                 return -ENOSPC;
2340
2341         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2342         if (!new_bus)
2343                 return -ENOMEM;
2344         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2345         new_bus->devs[new_bus->dev_count++] = dev;
2346         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2347         synchronize_srcu_expedited(&kvm->srcu);
2348         kfree(bus);
2349
2350         return 0;
2351 }
2352
2353 /* Caller must hold slots_lock. */
2354 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2355                               struct kvm_io_device *dev)
2356 {
2357         int i, r;
2358         struct kvm_io_bus *new_bus, *bus;
2359
2360         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2361         if (!new_bus)
2362                 return -ENOMEM;
2363
2364         bus = kvm->buses[bus_idx];
2365         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2366
2367         r = -ENOENT;
2368         for (i = 0; i < new_bus->dev_count; i++)
2369                 if (new_bus->devs[i] == dev) {
2370                         r = 0;
2371                         new_bus->devs[i] = new_bus->devs[--new_bus->dev_count];
2372                         break;
2373                 }
2374
2375         if (r) {
2376                 kfree(new_bus);
2377                 return r;
2378         }
2379
2380         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2381         synchronize_srcu_expedited(&kvm->srcu);
2382         kfree(bus);
2383         return r;
2384 }
2385
2386 static struct notifier_block kvm_cpu_notifier = {
2387         .notifier_call = kvm_cpu_hotplug,
2388 };
2389
2390 static int vm_stat_get(void *_offset, u64 *val)
2391 {
2392         unsigned offset = (long)_offset;
2393         struct kvm *kvm;
2394
2395         *val = 0;
2396         raw_spin_lock(&kvm_lock);
2397         list_for_each_entry(kvm, &vm_list, vm_list)
2398                 *val += *(u32 *)((void *)kvm + offset);
2399         raw_spin_unlock(&kvm_lock);
2400         return 0;
2401 }
2402
2403 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2404
2405 static int vcpu_stat_get(void *_offset, u64 *val)
2406 {
2407         unsigned offset = (long)_offset;
2408         struct kvm *kvm;
2409         struct kvm_vcpu *vcpu;
2410         int i;
2411
2412         *val = 0;
2413         raw_spin_lock(&kvm_lock);
2414         list_for_each_entry(kvm, &vm_list, vm_list)
2415                 kvm_for_each_vcpu(i, vcpu, kvm)
2416                         *val += *(u32 *)((void *)vcpu + offset);
2417
2418         raw_spin_unlock(&kvm_lock);
2419         return 0;
2420 }
2421
2422 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2423
2424 static const struct file_operations *stat_fops[] = {
2425         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2426         [KVM_STAT_VM]   = &vm_stat_fops,
2427 };
2428
2429 static void kvm_init_debug(void)
2430 {
2431         struct kvm_stats_debugfs_item *p;
2432
2433         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2434         for (p = debugfs_entries; p->name; ++p)
2435                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2436                                                 (void *)(long)p->offset,
2437                                                 stat_fops[p->kind]);
2438 }
2439
2440 static void kvm_exit_debug(void)
2441 {
2442         struct kvm_stats_debugfs_item *p;
2443
2444         for (p = debugfs_entries; p->name; ++p)
2445                 debugfs_remove(p->dentry);
2446         debugfs_remove(kvm_debugfs_dir);
2447 }
2448
2449 static int kvm_suspend(void)
2450 {
2451         if (kvm_usage_count)
2452                 hardware_disable_nolock(NULL);
2453         return 0;
2454 }
2455
2456 static void kvm_resume(void)
2457 {
2458         if (kvm_usage_count) {
2459                 WARN_ON(raw_spin_is_locked(&kvm_lock));
2460                 hardware_enable_nolock(NULL);
2461         }
2462 }
2463
2464 static struct syscore_ops kvm_syscore_ops = {
2465         .suspend = kvm_suspend,
2466         .resume = kvm_resume,
2467 };
2468
2469 struct page *bad_page;
2470 pfn_t bad_pfn;
2471
2472 static inline
2473 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2474 {
2475         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2476 }
2477
2478 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2479 {
2480         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2481
2482         kvm_arch_vcpu_load(vcpu, cpu);
2483 }
2484
2485 static void kvm_sched_out(struct preempt_notifier *pn,
2486                           struct task_struct *next)
2487 {
2488         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2489
2490         kvm_arch_vcpu_put(vcpu);
2491 }
2492
2493 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2494                   struct module *module)
2495 {
2496         int r;
2497         int cpu;
2498
2499         r = kvm_arch_init(opaque);
2500         if (r)
2501                 goto out_fail;
2502
2503         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2504
2505         if (bad_page == NULL) {
2506                 r = -ENOMEM;
2507                 goto out;
2508         }
2509
2510         bad_pfn = page_to_pfn(bad_page);
2511
2512         hwpoison_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2513
2514         if (hwpoison_page == NULL) {
2515                 r = -ENOMEM;
2516                 goto out_free_0;
2517         }
2518
2519         hwpoison_pfn = page_to_pfn(hwpoison_page);
2520
2521         fault_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2522
2523         if (fault_page == NULL) {
2524                 r = -ENOMEM;
2525                 goto out_free_0;
2526         }
2527
2528         fault_pfn = page_to_pfn(fault_page);
2529
2530         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2531                 r = -ENOMEM;
2532                 goto out_free_0;
2533         }
2534
2535         r = kvm_arch_hardware_setup();
2536         if (r < 0)
2537                 goto out_free_0a;
2538
2539         for_each_online_cpu(cpu) {
2540                 smp_call_function_single(cpu,
2541                                 kvm_arch_check_processor_compat,
2542                                 &r, 1);
2543                 if (r < 0)
2544                         goto out_free_1;
2545         }
2546
2547         r = register_cpu_notifier(&kvm_cpu_notifier);
2548         if (r)
2549                 goto out_free_2;
2550         register_reboot_notifier(&kvm_reboot_notifier);
2551
2552         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2553         if (!vcpu_align)
2554                 vcpu_align = __alignof__(struct kvm_vcpu);
2555         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2556                                            0, NULL);
2557         if (!kvm_vcpu_cache) {
2558                 r = -ENOMEM;
2559                 goto out_free_3;
2560         }
2561
2562         r = kvm_async_pf_init();
2563         if (r)
2564                 goto out_free;
2565
2566         kvm_chardev_ops.owner = module;
2567         kvm_vm_fops.owner = module;
2568         kvm_vcpu_fops.owner = module;
2569
2570         r = misc_register(&kvm_dev);
2571         if (r) {
2572                 printk(KERN_ERR "kvm: misc device register failed\n");
2573                 goto out_unreg;
2574         }
2575
2576         register_syscore_ops(&kvm_syscore_ops);
2577
2578         kvm_preempt_ops.sched_in = kvm_sched_in;
2579         kvm_preempt_ops.sched_out = kvm_sched_out;
2580
2581         kvm_init_debug();
2582
2583         return 0;
2584
2585 out_unreg:
2586         kvm_async_pf_deinit();
2587 out_free:
2588         kmem_cache_destroy(kvm_vcpu_cache);
2589 out_free_3:
2590         unregister_reboot_notifier(&kvm_reboot_notifier);
2591         unregister_cpu_notifier(&kvm_cpu_notifier);
2592 out_free_2:
2593 out_free_1:
2594         kvm_arch_hardware_unsetup();
2595 out_free_0a:
2596         free_cpumask_var(cpus_hardware_enabled);
2597 out_free_0:
2598         if (fault_page)
2599                 __free_page(fault_page);
2600         if (hwpoison_page)
2601                 __free_page(hwpoison_page);
2602         __free_page(bad_page);
2603 out:
2604         kvm_arch_exit();
2605 out_fail:
2606         return r;
2607 }
2608 EXPORT_SYMBOL_GPL(kvm_init);
2609
2610 void kvm_exit(void)
2611 {
2612         kvm_exit_debug();
2613         misc_deregister(&kvm_dev);
2614         kmem_cache_destroy(kvm_vcpu_cache);
2615         kvm_async_pf_deinit();
2616         unregister_syscore_ops(&kvm_syscore_ops);
2617         unregister_reboot_notifier(&kvm_reboot_notifier);
2618         unregister_cpu_notifier(&kvm_cpu_notifier);
2619         on_each_cpu(hardware_disable_nolock, NULL, 1);
2620         kvm_arch_hardware_unsetup();
2621         kvm_arch_exit();
2622         free_cpumask_var(cpus_hardware_enabled);
2623         __free_page(hwpoison_page);
2624         __free_page(bad_page);
2625 }
2626 EXPORT_SYMBOL_GPL(kvm_exit);