6330653480e467d0f79776d0e0a2c90ebe56f7e1
[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 int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1041         unsigned long start, int write, struct page **page)
1042 {
1043         int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1044
1045         if (write)
1046                 flags |= FOLL_WRITE;
1047
1048         return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1049 }
1050
1051 static inline int check_user_page_hwpoison(unsigned long addr)
1052 {
1053         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1054
1055         rc = __get_user_pages(current, current->mm, addr, 1,
1056                               flags, NULL, NULL, NULL);
1057         return rc == -EHWPOISON;
1058 }
1059
1060 static pfn_t hva_to_pfn(struct kvm *kvm, unsigned long addr, bool atomic,
1061                         bool *async, bool write_fault, bool *writable)
1062 {
1063         struct page *page[1];
1064         int npages = 0;
1065         pfn_t pfn;
1066
1067         /* we can do it either atomically or asynchronously, not both */
1068         BUG_ON(atomic && async);
1069
1070         BUG_ON(!write_fault && !writable);
1071
1072         if (writable)
1073                 *writable = true;
1074
1075         if (atomic || async)
1076                 npages = __get_user_pages_fast(addr, 1, 1, page);
1077
1078         if (unlikely(npages != 1) && !atomic) {
1079                 might_sleep();
1080
1081                 if (writable)
1082                         *writable = write_fault;
1083
1084                 if (async) {
1085                         down_read(&current->mm->mmap_sem);
1086                         npages = get_user_page_nowait(current, current->mm,
1087                                                      addr, write_fault, page);
1088                         up_read(&current->mm->mmap_sem);
1089                 } else
1090                         npages = get_user_pages_fast(addr, 1, write_fault,
1091                                                      page);
1092
1093                 /* map read fault as writable if possible */
1094                 if (unlikely(!write_fault) && npages == 1) {
1095                         struct page *wpage[1];
1096
1097                         npages = __get_user_pages_fast(addr, 1, 1, wpage);
1098                         if (npages == 1) {
1099                                 *writable = true;
1100                                 put_page(page[0]);
1101                                 page[0] = wpage[0];
1102                         }
1103                         npages = 1;
1104                 }
1105         }
1106
1107         if (unlikely(npages != 1)) {
1108                 struct vm_area_struct *vma;
1109
1110                 if (atomic)
1111                         return get_fault_pfn();
1112
1113                 down_read(&current->mm->mmap_sem);
1114                 if (npages == -EHWPOISON ||
1115                         (!async && check_user_page_hwpoison(addr))) {
1116                         up_read(&current->mm->mmap_sem);
1117                         get_page(hwpoison_page);
1118                         return page_to_pfn(hwpoison_page);
1119                 }
1120
1121                 vma = find_vma_intersection(current->mm, addr, addr+1);
1122
1123                 if (vma == NULL)
1124                         pfn = get_fault_pfn();
1125                 else if ((vma->vm_flags & VM_PFNMAP)) {
1126                         pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1127                                 vma->vm_pgoff;
1128                         BUG_ON(!kvm_is_mmio_pfn(pfn));
1129                 } else {
1130                         if (async && (vma->vm_flags & VM_WRITE))
1131                                 *async = true;
1132                         pfn = get_fault_pfn();
1133                 }
1134                 up_read(&current->mm->mmap_sem);
1135         } else
1136                 pfn = page_to_pfn(page[0]);
1137
1138         return pfn;
1139 }
1140
1141 pfn_t hva_to_pfn_atomic(struct kvm *kvm, unsigned long addr)
1142 {
1143         return hva_to_pfn(kvm, addr, true, NULL, true, NULL);
1144 }
1145 EXPORT_SYMBOL_GPL(hva_to_pfn_atomic);
1146
1147 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1148                           bool write_fault, bool *writable)
1149 {
1150         unsigned long addr;
1151
1152         if (async)
1153                 *async = false;
1154
1155         addr = gfn_to_hva(kvm, gfn);
1156         if (kvm_is_error_hva(addr)) {
1157                 get_page(bad_page);
1158                 return page_to_pfn(bad_page);
1159         }
1160
1161         return hva_to_pfn(kvm, addr, atomic, async, write_fault, writable);
1162 }
1163
1164 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1165 {
1166         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1167 }
1168 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1169
1170 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1171                        bool write_fault, bool *writable)
1172 {
1173         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1174 }
1175 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1176
1177 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1178 {
1179         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1180 }
1181 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1182
1183 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1184                       bool *writable)
1185 {
1186         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1187 }
1188 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1189
1190 pfn_t gfn_to_pfn_memslot(struct kvm *kvm,
1191                          struct kvm_memory_slot *slot, gfn_t gfn)
1192 {
1193         unsigned long addr = gfn_to_hva_memslot(slot, gfn);
1194         return hva_to_pfn(kvm, addr, false, NULL, true, NULL);
1195 }
1196
1197 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1198                                                                   int nr_pages)
1199 {
1200         unsigned long addr;
1201         gfn_t entry;
1202
1203         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1204         if (kvm_is_error_hva(addr))
1205                 return -1;
1206
1207         if (entry < nr_pages)
1208                 return 0;
1209
1210         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1211 }
1212 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1213
1214 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1215 {
1216         pfn_t pfn;
1217
1218         pfn = gfn_to_pfn(kvm, gfn);
1219         if (!kvm_is_mmio_pfn(pfn))
1220                 return pfn_to_page(pfn);
1221
1222         WARN_ON(kvm_is_mmio_pfn(pfn));
1223
1224         get_page(bad_page);
1225         return bad_page;
1226 }
1227
1228 EXPORT_SYMBOL_GPL(gfn_to_page);
1229
1230 void kvm_release_page_clean(struct page *page)
1231 {
1232         kvm_release_pfn_clean(page_to_pfn(page));
1233 }
1234 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1235
1236 void kvm_release_pfn_clean(pfn_t pfn)
1237 {
1238         if (!kvm_is_mmio_pfn(pfn))
1239                 put_page(pfn_to_page(pfn));
1240 }
1241 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1242
1243 void kvm_release_page_dirty(struct page *page)
1244 {
1245         kvm_release_pfn_dirty(page_to_pfn(page));
1246 }
1247 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1248
1249 void kvm_release_pfn_dirty(pfn_t pfn)
1250 {
1251         kvm_set_pfn_dirty(pfn);
1252         kvm_release_pfn_clean(pfn);
1253 }
1254 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1255
1256 void kvm_set_page_dirty(struct page *page)
1257 {
1258         kvm_set_pfn_dirty(page_to_pfn(page));
1259 }
1260 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1261
1262 void kvm_set_pfn_dirty(pfn_t pfn)
1263 {
1264         if (!kvm_is_mmio_pfn(pfn)) {
1265                 struct page *page = pfn_to_page(pfn);
1266                 if (!PageReserved(page))
1267                         SetPageDirty(page);
1268         }
1269 }
1270 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1271
1272 void kvm_set_pfn_accessed(pfn_t pfn)
1273 {
1274         if (!kvm_is_mmio_pfn(pfn))
1275                 mark_page_accessed(pfn_to_page(pfn));
1276 }
1277 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1278
1279 void kvm_get_pfn(pfn_t pfn)
1280 {
1281         if (!kvm_is_mmio_pfn(pfn))
1282                 get_page(pfn_to_page(pfn));
1283 }
1284 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1285
1286 static int next_segment(unsigned long len, int offset)
1287 {
1288         if (len > PAGE_SIZE - offset)
1289                 return PAGE_SIZE - offset;
1290         else
1291                 return len;
1292 }
1293
1294 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1295                         int len)
1296 {
1297         int r;
1298         unsigned long addr;
1299
1300         addr = gfn_to_hva(kvm, gfn);
1301         if (kvm_is_error_hva(addr))
1302                 return -EFAULT;
1303         r = copy_from_user(data, (void __user *)addr + offset, len);
1304         if (r)
1305                 return -EFAULT;
1306         return 0;
1307 }
1308 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1309
1310 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1311 {
1312         gfn_t gfn = gpa >> PAGE_SHIFT;
1313         int seg;
1314         int offset = offset_in_page(gpa);
1315         int ret;
1316
1317         while ((seg = next_segment(len, offset)) != 0) {
1318                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1319                 if (ret < 0)
1320                         return ret;
1321                 offset = 0;
1322                 len -= seg;
1323                 data += seg;
1324                 ++gfn;
1325         }
1326         return 0;
1327 }
1328 EXPORT_SYMBOL_GPL(kvm_read_guest);
1329
1330 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1331                           unsigned long len)
1332 {
1333         int r;
1334         unsigned long addr;
1335         gfn_t gfn = gpa >> PAGE_SHIFT;
1336         int offset = offset_in_page(gpa);
1337
1338         addr = gfn_to_hva(kvm, gfn);
1339         if (kvm_is_error_hva(addr))
1340                 return -EFAULT;
1341         pagefault_disable();
1342         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1343         pagefault_enable();
1344         if (r)
1345                 return -EFAULT;
1346         return 0;
1347 }
1348 EXPORT_SYMBOL(kvm_read_guest_atomic);
1349
1350 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1351                          int offset, int len)
1352 {
1353         int r;
1354         unsigned long addr;
1355
1356         addr = gfn_to_hva(kvm, gfn);
1357         if (kvm_is_error_hva(addr))
1358                 return -EFAULT;
1359         r = copy_to_user((void __user *)addr + offset, data, len);
1360         if (r)
1361                 return -EFAULT;
1362         mark_page_dirty(kvm, gfn);
1363         return 0;
1364 }
1365 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1366
1367 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1368                     unsigned long len)
1369 {
1370         gfn_t gfn = gpa >> PAGE_SHIFT;
1371         int seg;
1372         int offset = offset_in_page(gpa);
1373         int ret;
1374
1375         while ((seg = next_segment(len, offset)) != 0) {
1376                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1377                 if (ret < 0)
1378                         return ret;
1379                 offset = 0;
1380                 len -= seg;
1381                 data += seg;
1382                 ++gfn;
1383         }
1384         return 0;
1385 }
1386
1387 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1388                               gpa_t gpa)
1389 {
1390         struct kvm_memslots *slots = kvm_memslots(kvm);
1391         int offset = offset_in_page(gpa);
1392         gfn_t gfn = gpa >> PAGE_SHIFT;
1393
1394         ghc->gpa = gpa;
1395         ghc->generation = slots->generation;
1396         ghc->memslot = __gfn_to_memslot(slots, gfn);
1397         ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1398         if (!kvm_is_error_hva(ghc->hva))
1399                 ghc->hva += offset;
1400         else
1401                 return -EFAULT;
1402
1403         return 0;
1404 }
1405 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1406
1407 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1408                            void *data, unsigned long len)
1409 {
1410         struct kvm_memslots *slots = kvm_memslots(kvm);
1411         int r;
1412
1413         if (slots->generation != ghc->generation)
1414                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1415
1416         if (kvm_is_error_hva(ghc->hva))
1417                 return -EFAULT;
1418
1419         r = copy_to_user((void __user *)ghc->hva, data, len);
1420         if (r)
1421                 return -EFAULT;
1422         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1423
1424         return 0;
1425 }
1426 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1427
1428 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1429 {
1430         return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1431                                     offset, len);
1432 }
1433 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1434
1435 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1436 {
1437         gfn_t gfn = gpa >> PAGE_SHIFT;
1438         int seg;
1439         int offset = offset_in_page(gpa);
1440         int ret;
1441
1442         while ((seg = next_segment(len, offset)) != 0) {
1443                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1444                 if (ret < 0)
1445                         return ret;
1446                 offset = 0;
1447                 len -= seg;
1448                 ++gfn;
1449         }
1450         return 0;
1451 }
1452 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1453
1454 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1455                              gfn_t gfn)
1456 {
1457         if (memslot && memslot->dirty_bitmap) {
1458                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1459
1460                 __set_bit_le(rel_gfn, memslot->dirty_bitmap);
1461         }
1462 }
1463
1464 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1465 {
1466         struct kvm_memory_slot *memslot;
1467
1468         memslot = gfn_to_memslot(kvm, gfn);
1469         mark_page_dirty_in_slot(kvm, memslot, gfn);
1470 }
1471
1472 /*
1473  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1474  */
1475 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1476 {
1477         DEFINE_WAIT(wait);
1478
1479         for (;;) {
1480                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1481
1482                 if (kvm_arch_vcpu_runnable(vcpu)) {
1483                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1484                         break;
1485                 }
1486                 if (kvm_cpu_has_pending_timer(vcpu))
1487                         break;
1488                 if (signal_pending(current))
1489                         break;
1490
1491                 schedule();
1492         }
1493
1494         finish_wait(&vcpu->wq, &wait);
1495 }
1496
1497 void kvm_resched(struct kvm_vcpu *vcpu)
1498 {
1499         if (!need_resched())
1500                 return;
1501         cond_resched();
1502 }
1503 EXPORT_SYMBOL_GPL(kvm_resched);
1504
1505 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1506 {
1507         struct kvm *kvm = me->kvm;
1508         struct kvm_vcpu *vcpu;
1509         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1510         int yielded = 0;
1511         int pass;
1512         int i;
1513
1514         /*
1515          * We boost the priority of a VCPU that is runnable but not
1516          * currently running, because it got preempted by something
1517          * else and called schedule in __vcpu_run.  Hopefully that
1518          * VCPU is holding the lock that we need and will release it.
1519          * We approximate round-robin by starting at the last boosted VCPU.
1520          */
1521         for (pass = 0; pass < 2 && !yielded; pass++) {
1522                 kvm_for_each_vcpu(i, vcpu, kvm) {
1523                         struct task_struct *task = NULL;
1524                         struct pid *pid;
1525                         if (!pass && i < last_boosted_vcpu) {
1526                                 i = last_boosted_vcpu;
1527                                 continue;
1528                         } else if (pass && i > last_boosted_vcpu)
1529                                 break;
1530                         if (vcpu == me)
1531                                 continue;
1532                         if (waitqueue_active(&vcpu->wq))
1533                                 continue;
1534                         rcu_read_lock();
1535                         pid = rcu_dereference(vcpu->pid);
1536                         if (pid)
1537                                 task = get_pid_task(vcpu->pid, PIDTYPE_PID);
1538                         rcu_read_unlock();
1539                         if (!task)
1540                                 continue;
1541                         if (task->flags & PF_VCPU) {
1542                                 put_task_struct(task);
1543                                 continue;
1544                         }
1545                         if (yield_to(task, 1)) {
1546                                 put_task_struct(task);
1547                                 kvm->last_boosted_vcpu = i;
1548                                 yielded = 1;
1549                                 break;
1550                         }
1551                         put_task_struct(task);
1552                 }
1553         }
1554 }
1555 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1556
1557 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1558 {
1559         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1560         struct page *page;
1561
1562         if (vmf->pgoff == 0)
1563                 page = virt_to_page(vcpu->run);
1564 #ifdef CONFIG_X86
1565         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1566                 page = virt_to_page(vcpu->arch.pio_data);
1567 #endif
1568 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1569         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1570                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1571 #endif
1572         else
1573                 return VM_FAULT_SIGBUS;
1574         get_page(page);
1575         vmf->page = page;
1576         return 0;
1577 }
1578
1579 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1580         .fault = kvm_vcpu_fault,
1581 };
1582
1583 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1584 {
1585         vma->vm_ops = &kvm_vcpu_vm_ops;
1586         return 0;
1587 }
1588
1589 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1590 {
1591         struct kvm_vcpu *vcpu = filp->private_data;
1592
1593         kvm_put_kvm(vcpu->kvm);
1594         return 0;
1595 }
1596
1597 static struct file_operations kvm_vcpu_fops = {
1598         .release        = kvm_vcpu_release,
1599         .unlocked_ioctl = kvm_vcpu_ioctl,
1600         .compat_ioctl   = kvm_vcpu_ioctl,
1601         .mmap           = kvm_vcpu_mmap,
1602         .llseek         = noop_llseek,
1603 };
1604
1605 /*
1606  * Allocates an inode for the vcpu.
1607  */
1608 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1609 {
1610         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1611 }
1612
1613 /*
1614  * Creates some virtual cpus.  Good luck creating more than one.
1615  */
1616 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1617 {
1618         int r;
1619         struct kvm_vcpu *vcpu, *v;
1620
1621         vcpu = kvm_arch_vcpu_create(kvm, id);
1622         if (IS_ERR(vcpu))
1623                 return PTR_ERR(vcpu);
1624
1625         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1626
1627         r = kvm_arch_vcpu_setup(vcpu);
1628         if (r)
1629                 return r;
1630
1631         mutex_lock(&kvm->lock);
1632         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1633                 r = -EINVAL;
1634                 goto vcpu_destroy;
1635         }
1636
1637         kvm_for_each_vcpu(r, v, kvm)
1638                 if (v->vcpu_id == id) {
1639                         r = -EEXIST;
1640                         goto vcpu_destroy;
1641                 }
1642
1643         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1644
1645         /* Now it's all set up, let userspace reach it */
1646         kvm_get_kvm(kvm);
1647         r = create_vcpu_fd(vcpu);
1648         if (r < 0) {
1649                 kvm_put_kvm(kvm);
1650                 goto vcpu_destroy;
1651         }
1652
1653         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1654         smp_wmb();
1655         atomic_inc(&kvm->online_vcpus);
1656
1657 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1658         if (kvm->bsp_vcpu_id == id)
1659                 kvm->bsp_vcpu = vcpu;
1660 #endif
1661         mutex_unlock(&kvm->lock);
1662         return r;
1663
1664 vcpu_destroy:
1665         mutex_unlock(&kvm->lock);
1666         kvm_arch_vcpu_destroy(vcpu);
1667         return r;
1668 }
1669
1670 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1671 {
1672         if (sigset) {
1673                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1674                 vcpu->sigset_active = 1;
1675                 vcpu->sigset = *sigset;
1676         } else
1677                 vcpu->sigset_active = 0;
1678         return 0;
1679 }
1680
1681 static long kvm_vcpu_ioctl(struct file *filp,
1682                            unsigned int ioctl, unsigned long arg)
1683 {
1684         struct kvm_vcpu *vcpu = filp->private_data;
1685         void __user *argp = (void __user *)arg;
1686         int r;
1687         struct kvm_fpu *fpu = NULL;
1688         struct kvm_sregs *kvm_sregs = NULL;
1689
1690         if (vcpu->kvm->mm != current->mm)
1691                 return -EIO;
1692
1693 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1694         /*
1695          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1696          * so vcpu_load() would break it.
1697          */
1698         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1699                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1700 #endif
1701
1702
1703         vcpu_load(vcpu);
1704         switch (ioctl) {
1705         case KVM_RUN:
1706                 r = -EINVAL;
1707                 if (arg)
1708                         goto out;
1709                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1710                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1711                 break;
1712         case KVM_GET_REGS: {
1713                 struct kvm_regs *kvm_regs;
1714
1715                 r = -ENOMEM;
1716                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1717                 if (!kvm_regs)
1718                         goto out;
1719                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1720                 if (r)
1721                         goto out_free1;
1722                 r = -EFAULT;
1723                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1724                         goto out_free1;
1725                 r = 0;
1726 out_free1:
1727                 kfree(kvm_regs);
1728                 break;
1729         }
1730         case KVM_SET_REGS: {
1731                 struct kvm_regs *kvm_regs;
1732
1733                 r = -ENOMEM;
1734                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1735                 if (!kvm_regs)
1736                         goto out;
1737                 r = -EFAULT;
1738                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1739                         goto out_free2;
1740                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1741                 if (r)
1742                         goto out_free2;
1743                 r = 0;
1744 out_free2:
1745                 kfree(kvm_regs);
1746                 break;
1747         }
1748         case KVM_GET_SREGS: {
1749                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1750                 r = -ENOMEM;
1751                 if (!kvm_sregs)
1752                         goto out;
1753                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1754                 if (r)
1755                         goto out;
1756                 r = -EFAULT;
1757                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1758                         goto out;
1759                 r = 0;
1760                 break;
1761         }
1762         case KVM_SET_SREGS: {
1763                 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1764                 r = -ENOMEM;
1765                 if (!kvm_sregs)
1766                         goto out;
1767                 r = -EFAULT;
1768                 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1769                         goto out;
1770                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1771                 if (r)
1772                         goto out;
1773                 r = 0;
1774                 break;
1775         }
1776         case KVM_GET_MP_STATE: {
1777                 struct kvm_mp_state mp_state;
1778
1779                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1780                 if (r)
1781                         goto out;
1782                 r = -EFAULT;
1783                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1784                         goto out;
1785                 r = 0;
1786                 break;
1787         }
1788         case KVM_SET_MP_STATE: {
1789                 struct kvm_mp_state mp_state;
1790
1791                 r = -EFAULT;
1792                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1793                         goto out;
1794                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1795                 if (r)
1796                         goto out;
1797                 r = 0;
1798                 break;
1799         }
1800         case KVM_TRANSLATE: {
1801                 struct kvm_translation tr;
1802
1803                 r = -EFAULT;
1804                 if (copy_from_user(&tr, argp, sizeof tr))
1805                         goto out;
1806                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1807                 if (r)
1808                         goto out;
1809                 r = -EFAULT;
1810                 if (copy_to_user(argp, &tr, sizeof tr))
1811                         goto out;
1812                 r = 0;
1813                 break;
1814         }
1815         case KVM_SET_GUEST_DEBUG: {
1816                 struct kvm_guest_debug dbg;
1817
1818                 r = -EFAULT;
1819                 if (copy_from_user(&dbg, argp, sizeof dbg))
1820                         goto out;
1821                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1822                 if (r)
1823                         goto out;
1824                 r = 0;
1825                 break;
1826         }
1827         case KVM_SET_SIGNAL_MASK: {
1828                 struct kvm_signal_mask __user *sigmask_arg = argp;
1829                 struct kvm_signal_mask kvm_sigmask;
1830                 sigset_t sigset, *p;
1831
1832                 p = NULL;
1833                 if (argp) {
1834                         r = -EFAULT;
1835                         if (copy_from_user(&kvm_sigmask, argp,
1836                                            sizeof kvm_sigmask))
1837                                 goto out;
1838                         r = -EINVAL;
1839                         if (kvm_sigmask.len != sizeof sigset)
1840                                 goto out;
1841                         r = -EFAULT;
1842                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1843                                            sizeof sigset))
1844                                 goto out;
1845                         p = &sigset;
1846                 }
1847                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
1848                 break;
1849         }
1850         case KVM_GET_FPU: {
1851                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1852                 r = -ENOMEM;
1853                 if (!fpu)
1854                         goto out;
1855                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1856                 if (r)
1857                         goto out;
1858                 r = -EFAULT;
1859                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1860                         goto out;
1861                 r = 0;
1862                 break;
1863         }
1864         case KVM_SET_FPU: {
1865                 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1866                 r = -ENOMEM;
1867                 if (!fpu)
1868                         goto out;
1869                 r = -EFAULT;
1870                 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1871                         goto out;
1872                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1873                 if (r)
1874                         goto out;
1875                 r = 0;
1876                 break;
1877         }
1878         default:
1879                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1880         }
1881 out:
1882         vcpu_put(vcpu);
1883         kfree(fpu);
1884         kfree(kvm_sregs);
1885         return r;
1886 }
1887
1888 static long kvm_vm_ioctl(struct file *filp,
1889                            unsigned int ioctl, unsigned long arg)
1890 {
1891         struct kvm *kvm = filp->private_data;
1892         void __user *argp = (void __user *)arg;
1893         int r;
1894
1895         if (kvm->mm != current->mm)
1896                 return -EIO;
1897         switch (ioctl) {
1898         case KVM_CREATE_VCPU:
1899                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1900                 if (r < 0)
1901                         goto out;
1902                 break;
1903         case KVM_SET_USER_MEMORY_REGION: {
1904                 struct kvm_userspace_memory_region kvm_userspace_mem;
1905
1906                 r = -EFAULT;
1907                 if (copy_from_user(&kvm_userspace_mem, argp,
1908                                                 sizeof kvm_userspace_mem))
1909                         goto out;
1910
1911                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1912                 if (r)
1913                         goto out;
1914                 break;
1915         }
1916         case KVM_GET_DIRTY_LOG: {
1917                 struct kvm_dirty_log log;
1918
1919                 r = -EFAULT;
1920                 if (copy_from_user(&log, argp, sizeof log))
1921                         goto out;
1922                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1923                 if (r)
1924                         goto out;
1925                 break;
1926         }
1927 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1928         case KVM_REGISTER_COALESCED_MMIO: {
1929                 struct kvm_coalesced_mmio_zone zone;
1930                 r = -EFAULT;
1931                 if (copy_from_user(&zone, argp, sizeof zone))
1932                         goto out;
1933                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1934                 if (r)
1935                         goto out;
1936                 r = 0;
1937                 break;
1938         }
1939         case KVM_UNREGISTER_COALESCED_MMIO: {
1940                 struct kvm_coalesced_mmio_zone zone;
1941                 r = -EFAULT;
1942                 if (copy_from_user(&zone, argp, sizeof zone))
1943                         goto out;
1944                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1945                 if (r)
1946                         goto out;
1947                 r = 0;
1948                 break;
1949         }
1950 #endif
1951         case KVM_IRQFD: {
1952                 struct kvm_irqfd data;
1953
1954                 r = -EFAULT;
1955                 if (copy_from_user(&data, argp, sizeof data))
1956                         goto out;
1957                 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
1958                 break;
1959         }
1960         case KVM_IOEVENTFD: {
1961                 struct kvm_ioeventfd data;
1962
1963                 r = -EFAULT;
1964                 if (copy_from_user(&data, argp, sizeof data))
1965                         goto out;
1966                 r = kvm_ioeventfd(kvm, &data);
1967                 break;
1968         }
1969 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1970         case KVM_SET_BOOT_CPU_ID:
1971                 r = 0;
1972                 mutex_lock(&kvm->lock);
1973                 if (atomic_read(&kvm->online_vcpus) != 0)
1974                         r = -EBUSY;
1975                 else
1976                         kvm->bsp_vcpu_id = arg;
1977                 mutex_unlock(&kvm->lock);
1978                 break;
1979 #endif
1980         default:
1981                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1982                 if (r == -ENOTTY)
1983                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
1984         }
1985 out:
1986         return r;
1987 }
1988
1989 #ifdef CONFIG_COMPAT
1990 struct compat_kvm_dirty_log {
1991         __u32 slot;
1992         __u32 padding1;
1993         union {
1994                 compat_uptr_t dirty_bitmap; /* one bit per page */
1995                 __u64 padding2;
1996         };
1997 };
1998
1999 static long kvm_vm_compat_ioctl(struct file *filp,
2000                            unsigned int ioctl, unsigned long arg)
2001 {
2002         struct kvm *kvm = filp->private_data;
2003         int r;
2004
2005         if (kvm->mm != current->mm)
2006                 return -EIO;
2007         switch (ioctl) {
2008         case KVM_GET_DIRTY_LOG: {
2009                 struct compat_kvm_dirty_log compat_log;
2010                 struct kvm_dirty_log log;
2011
2012                 r = -EFAULT;
2013                 if (copy_from_user(&compat_log, (void __user *)arg,
2014                                    sizeof(compat_log)))
2015                         goto out;
2016                 log.slot         = compat_log.slot;
2017                 log.padding1     = compat_log.padding1;
2018                 log.padding2     = compat_log.padding2;
2019                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2020
2021                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2022                 if (r)
2023                         goto out;
2024                 break;
2025         }
2026         default:
2027                 r = kvm_vm_ioctl(filp, ioctl, arg);
2028         }
2029
2030 out:
2031         return r;
2032 }
2033 #endif
2034
2035 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2036 {
2037         struct page *page[1];
2038         unsigned long addr;
2039         int npages;
2040         gfn_t gfn = vmf->pgoff;
2041         struct kvm *kvm = vma->vm_file->private_data;
2042
2043         addr = gfn_to_hva(kvm, gfn);
2044         if (kvm_is_error_hva(addr))
2045                 return VM_FAULT_SIGBUS;
2046
2047         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2048                                 NULL);
2049         if (unlikely(npages != 1))
2050                 return VM_FAULT_SIGBUS;
2051
2052         vmf->page = page[0];
2053         return 0;
2054 }
2055
2056 static const struct vm_operations_struct kvm_vm_vm_ops = {
2057         .fault = kvm_vm_fault,
2058 };
2059
2060 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2061 {
2062         vma->vm_ops = &kvm_vm_vm_ops;
2063         return 0;
2064 }
2065
2066 static struct file_operations kvm_vm_fops = {
2067         .release        = kvm_vm_release,
2068         .unlocked_ioctl = kvm_vm_ioctl,
2069 #ifdef CONFIG_COMPAT
2070         .compat_ioctl   = kvm_vm_compat_ioctl,
2071 #endif
2072         .mmap           = kvm_vm_mmap,
2073         .llseek         = noop_llseek,
2074 };
2075
2076 static int kvm_dev_ioctl_create_vm(void)
2077 {
2078         int r;
2079         struct kvm *kvm;
2080
2081         kvm = kvm_create_vm();
2082         if (IS_ERR(kvm))
2083                 return PTR_ERR(kvm);
2084 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2085         r = kvm_coalesced_mmio_init(kvm);
2086         if (r < 0) {
2087                 kvm_put_kvm(kvm);
2088                 return r;
2089         }
2090 #endif
2091         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2092         if (r < 0)
2093                 kvm_put_kvm(kvm);
2094
2095         return r;
2096 }
2097
2098 static long kvm_dev_ioctl_check_extension_generic(long arg)
2099 {
2100         switch (arg) {
2101         case KVM_CAP_USER_MEMORY:
2102         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2103         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2104 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2105         case KVM_CAP_SET_BOOT_CPU_ID:
2106 #endif
2107         case KVM_CAP_INTERNAL_ERROR_DATA:
2108                 return 1;
2109 #ifdef CONFIG_HAVE_KVM_IRQCHIP
2110         case KVM_CAP_IRQ_ROUTING:
2111                 return KVM_MAX_IRQ_ROUTES;
2112 #endif
2113         default:
2114                 break;
2115         }
2116         return kvm_dev_ioctl_check_extension(arg);
2117 }
2118
2119 static long kvm_dev_ioctl(struct file *filp,
2120                           unsigned int ioctl, unsigned long arg)
2121 {
2122         long r = -EINVAL;
2123
2124         switch (ioctl) {
2125         case KVM_GET_API_VERSION:
2126                 r = -EINVAL;
2127                 if (arg)
2128                         goto out;
2129                 r = KVM_API_VERSION;
2130                 break;
2131         case KVM_CREATE_VM:
2132                 r = -EINVAL;
2133                 if (arg)
2134                         goto out;
2135                 r = kvm_dev_ioctl_create_vm();
2136                 break;
2137         case KVM_CHECK_EXTENSION:
2138                 r = kvm_dev_ioctl_check_extension_generic(arg);
2139                 break;
2140         case KVM_GET_VCPU_MMAP_SIZE:
2141                 r = -EINVAL;
2142                 if (arg)
2143                         goto out;
2144                 r = PAGE_SIZE;     /* struct kvm_run */
2145 #ifdef CONFIG_X86
2146                 r += PAGE_SIZE;    /* pio data page */
2147 #endif
2148 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2149                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2150 #endif
2151                 break;
2152         case KVM_TRACE_ENABLE:
2153         case KVM_TRACE_PAUSE:
2154         case KVM_TRACE_DISABLE:
2155                 r = -EOPNOTSUPP;
2156                 break;
2157         default:
2158                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2159         }
2160 out:
2161         return r;
2162 }
2163
2164 static struct file_operations kvm_chardev_ops = {
2165         .unlocked_ioctl = kvm_dev_ioctl,
2166         .compat_ioctl   = kvm_dev_ioctl,
2167         .llseek         = noop_llseek,
2168 };
2169
2170 static struct miscdevice kvm_dev = {
2171         KVM_MINOR,
2172         "kvm",
2173         &kvm_chardev_ops,
2174 };
2175
2176 static void hardware_enable_nolock(void *junk)
2177 {
2178         int cpu = raw_smp_processor_id();
2179         int r;
2180
2181         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2182                 return;
2183
2184         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2185
2186         r = kvm_arch_hardware_enable(NULL);
2187
2188         if (r) {
2189                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2190                 atomic_inc(&hardware_enable_failed);
2191                 printk(KERN_INFO "kvm: enabling virtualization on "
2192                                  "CPU%d failed\n", cpu);
2193         }
2194 }
2195
2196 static void hardware_enable(void *junk)
2197 {
2198         raw_spin_lock(&kvm_lock);
2199         hardware_enable_nolock(junk);
2200         raw_spin_unlock(&kvm_lock);
2201 }
2202
2203 static void hardware_disable_nolock(void *junk)
2204 {
2205         int cpu = raw_smp_processor_id();
2206
2207         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2208                 return;
2209         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2210         kvm_arch_hardware_disable(NULL);
2211 }
2212
2213 static void hardware_disable(void *junk)
2214 {
2215         raw_spin_lock(&kvm_lock);
2216         hardware_disable_nolock(junk);
2217         raw_spin_unlock(&kvm_lock);
2218 }
2219
2220 static void hardware_disable_all_nolock(void)
2221 {
2222         BUG_ON(!kvm_usage_count);
2223
2224         kvm_usage_count--;
2225         if (!kvm_usage_count)
2226                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2227 }
2228
2229 static void hardware_disable_all(void)
2230 {
2231         raw_spin_lock(&kvm_lock);
2232         hardware_disable_all_nolock();
2233         raw_spin_unlock(&kvm_lock);
2234 }
2235
2236 static int hardware_enable_all(void)
2237 {
2238         int r = 0;
2239
2240         raw_spin_lock(&kvm_lock);
2241
2242         kvm_usage_count++;
2243         if (kvm_usage_count == 1) {
2244                 atomic_set(&hardware_enable_failed, 0);
2245                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2246
2247                 if (atomic_read(&hardware_enable_failed)) {
2248                         hardware_disable_all_nolock();
2249                         r = -EBUSY;
2250                 }
2251         }
2252
2253         raw_spin_unlock(&kvm_lock);
2254
2255         return r;
2256 }
2257
2258 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2259                            void *v)
2260 {
2261         int cpu = (long)v;
2262
2263         if (!kvm_usage_count)
2264                 return NOTIFY_OK;
2265
2266         val &= ~CPU_TASKS_FROZEN;
2267         switch (val) {
2268         case CPU_DYING:
2269                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2270                        cpu);
2271                 hardware_disable(NULL);
2272                 break;
2273         case CPU_STARTING:
2274                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2275                        cpu);
2276                 hardware_enable(NULL);
2277                 break;
2278         }
2279         return NOTIFY_OK;
2280 }
2281
2282
2283 asmlinkage void kvm_spurious_fault(void)
2284 {
2285         /* Fault while not rebooting.  We want the trace. */
2286         BUG();
2287 }
2288 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
2289
2290 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2291                       void *v)
2292 {
2293         /*
2294          * Some (well, at least mine) BIOSes hang on reboot if
2295          * in vmx root mode.
2296          *
2297          * And Intel TXT required VMX off for all cpu when system shutdown.
2298          */
2299         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2300         kvm_rebooting = true;
2301         on_each_cpu(hardware_disable_nolock, NULL, 1);
2302         return NOTIFY_OK;
2303 }
2304
2305 static struct notifier_block kvm_reboot_notifier = {
2306         .notifier_call = kvm_reboot,
2307         .priority = 0,
2308 };
2309
2310 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2311 {
2312         int i;
2313
2314         for (i = 0; i < bus->dev_count; i++) {
2315                 struct kvm_io_device *pos = bus->devs[i];
2316
2317                 kvm_iodevice_destructor(pos);
2318         }
2319         kfree(bus);
2320 }
2321
2322 /* kvm_io_bus_write - called under kvm->slots_lock */
2323 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2324                      int len, const void *val)
2325 {
2326         int i;
2327         struct kvm_io_bus *bus;
2328
2329         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2330         for (i = 0; i < bus->dev_count; i++)
2331                 if (!kvm_iodevice_write(bus->devs[i], addr, len, val))
2332                         return 0;
2333         return -EOPNOTSUPP;
2334 }
2335
2336 /* kvm_io_bus_read - called under kvm->slots_lock */
2337 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2338                     int len, void *val)
2339 {
2340         int i;
2341         struct kvm_io_bus *bus;
2342
2343         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2344         for (i = 0; i < bus->dev_count; i++)
2345                 if (!kvm_iodevice_read(bus->devs[i], addr, len, val))
2346                         return 0;
2347         return -EOPNOTSUPP;
2348 }
2349
2350 /* Caller must hold slots_lock. */
2351 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2352                             struct kvm_io_device *dev)
2353 {
2354         struct kvm_io_bus *new_bus, *bus;
2355
2356         bus = kvm->buses[bus_idx];
2357         if (bus->dev_count > NR_IOBUS_DEVS-1)
2358                 return -ENOSPC;
2359
2360         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2361         if (!new_bus)
2362                 return -ENOMEM;
2363         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2364         new_bus->devs[new_bus->dev_count++] = dev;
2365         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2366         synchronize_srcu_expedited(&kvm->srcu);
2367         kfree(bus);
2368
2369         return 0;
2370 }
2371
2372 /* Caller must hold slots_lock. */
2373 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2374                               struct kvm_io_device *dev)
2375 {
2376         int i, r;
2377         struct kvm_io_bus *new_bus, *bus;
2378
2379         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2380         if (!new_bus)
2381                 return -ENOMEM;
2382
2383         bus = kvm->buses[bus_idx];
2384         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2385
2386         r = -ENOENT;
2387         for (i = 0; i < new_bus->dev_count; i++)
2388                 if (new_bus->devs[i] == dev) {
2389                         r = 0;
2390                         new_bus->devs[i] = new_bus->devs[--new_bus->dev_count];
2391                         break;
2392                 }
2393
2394         if (r) {
2395                 kfree(new_bus);
2396                 return r;
2397         }
2398
2399         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2400         synchronize_srcu_expedited(&kvm->srcu);
2401         kfree(bus);
2402         return r;
2403 }
2404
2405 static struct notifier_block kvm_cpu_notifier = {
2406         .notifier_call = kvm_cpu_hotplug,
2407 };
2408
2409 static int vm_stat_get(void *_offset, u64 *val)
2410 {
2411         unsigned offset = (long)_offset;
2412         struct kvm *kvm;
2413
2414         *val = 0;
2415         raw_spin_lock(&kvm_lock);
2416         list_for_each_entry(kvm, &vm_list, vm_list)
2417                 *val += *(u32 *)((void *)kvm + offset);
2418         raw_spin_unlock(&kvm_lock);
2419         return 0;
2420 }
2421
2422 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2423
2424 static int vcpu_stat_get(void *_offset, u64 *val)
2425 {
2426         unsigned offset = (long)_offset;
2427         struct kvm *kvm;
2428         struct kvm_vcpu *vcpu;
2429         int i;
2430
2431         *val = 0;
2432         raw_spin_lock(&kvm_lock);
2433         list_for_each_entry(kvm, &vm_list, vm_list)
2434                 kvm_for_each_vcpu(i, vcpu, kvm)
2435                         *val += *(u32 *)((void *)vcpu + offset);
2436
2437         raw_spin_unlock(&kvm_lock);
2438         return 0;
2439 }
2440
2441 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2442
2443 static const struct file_operations *stat_fops[] = {
2444         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2445         [KVM_STAT_VM]   = &vm_stat_fops,
2446 };
2447
2448 static void kvm_init_debug(void)
2449 {
2450         struct kvm_stats_debugfs_item *p;
2451
2452         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2453         for (p = debugfs_entries; p->name; ++p)
2454                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2455                                                 (void *)(long)p->offset,
2456                                                 stat_fops[p->kind]);
2457 }
2458
2459 static void kvm_exit_debug(void)
2460 {
2461         struct kvm_stats_debugfs_item *p;
2462
2463         for (p = debugfs_entries; p->name; ++p)
2464                 debugfs_remove(p->dentry);
2465         debugfs_remove(kvm_debugfs_dir);
2466 }
2467
2468 static int kvm_suspend(void)
2469 {
2470         if (kvm_usage_count)
2471                 hardware_disable_nolock(NULL);
2472         return 0;
2473 }
2474
2475 static void kvm_resume(void)
2476 {
2477         if (kvm_usage_count) {
2478                 WARN_ON(raw_spin_is_locked(&kvm_lock));
2479                 hardware_enable_nolock(NULL);
2480         }
2481 }
2482
2483 static struct syscore_ops kvm_syscore_ops = {
2484         .suspend = kvm_suspend,
2485         .resume = kvm_resume,
2486 };
2487
2488 struct page *bad_page;
2489 pfn_t bad_pfn;
2490
2491 static inline
2492 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2493 {
2494         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2495 }
2496
2497 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2498 {
2499         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2500
2501         kvm_arch_vcpu_load(vcpu, cpu);
2502 }
2503
2504 static void kvm_sched_out(struct preempt_notifier *pn,
2505                           struct task_struct *next)
2506 {
2507         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2508
2509         kvm_arch_vcpu_put(vcpu);
2510 }
2511
2512 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2513                   struct module *module)
2514 {
2515         int r;
2516         int cpu;
2517
2518         r = kvm_arch_init(opaque);
2519         if (r)
2520                 goto out_fail;
2521
2522         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2523
2524         if (bad_page == NULL) {
2525                 r = -ENOMEM;
2526                 goto out;
2527         }
2528
2529         bad_pfn = page_to_pfn(bad_page);
2530
2531         hwpoison_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2532
2533         if (hwpoison_page == NULL) {
2534                 r = -ENOMEM;
2535                 goto out_free_0;
2536         }
2537
2538         hwpoison_pfn = page_to_pfn(hwpoison_page);
2539
2540         fault_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2541
2542         if (fault_page == NULL) {
2543                 r = -ENOMEM;
2544                 goto out_free_0;
2545         }
2546
2547         fault_pfn = page_to_pfn(fault_page);
2548
2549         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2550                 r = -ENOMEM;
2551                 goto out_free_0;
2552         }
2553
2554         r = kvm_arch_hardware_setup();
2555         if (r < 0)
2556                 goto out_free_0a;
2557
2558         for_each_online_cpu(cpu) {
2559                 smp_call_function_single(cpu,
2560                                 kvm_arch_check_processor_compat,
2561                                 &r, 1);
2562                 if (r < 0)
2563                         goto out_free_1;
2564         }
2565
2566         r = register_cpu_notifier(&kvm_cpu_notifier);
2567         if (r)
2568                 goto out_free_2;
2569         register_reboot_notifier(&kvm_reboot_notifier);
2570
2571         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2572         if (!vcpu_align)
2573                 vcpu_align = __alignof__(struct kvm_vcpu);
2574         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2575                                            0, NULL);
2576         if (!kvm_vcpu_cache) {
2577                 r = -ENOMEM;
2578                 goto out_free_3;
2579         }
2580
2581         r = kvm_async_pf_init();
2582         if (r)
2583                 goto out_free;
2584
2585         kvm_chardev_ops.owner = module;
2586         kvm_vm_fops.owner = module;
2587         kvm_vcpu_fops.owner = module;
2588
2589         r = misc_register(&kvm_dev);
2590         if (r) {
2591                 printk(KERN_ERR "kvm: misc device register failed\n");
2592                 goto out_unreg;
2593         }
2594
2595         register_syscore_ops(&kvm_syscore_ops);
2596
2597         kvm_preempt_ops.sched_in = kvm_sched_in;
2598         kvm_preempt_ops.sched_out = kvm_sched_out;
2599
2600         kvm_init_debug();
2601
2602         return 0;
2603
2604 out_unreg:
2605         kvm_async_pf_deinit();
2606 out_free:
2607         kmem_cache_destroy(kvm_vcpu_cache);
2608 out_free_3:
2609         unregister_reboot_notifier(&kvm_reboot_notifier);
2610         unregister_cpu_notifier(&kvm_cpu_notifier);
2611 out_free_2:
2612 out_free_1:
2613         kvm_arch_hardware_unsetup();
2614 out_free_0a:
2615         free_cpumask_var(cpus_hardware_enabled);
2616 out_free_0:
2617         if (fault_page)
2618                 __free_page(fault_page);
2619         if (hwpoison_page)
2620                 __free_page(hwpoison_page);
2621         __free_page(bad_page);
2622 out:
2623         kvm_arch_exit();
2624 out_fail:
2625         return r;
2626 }
2627 EXPORT_SYMBOL_GPL(kvm_init);
2628
2629 void kvm_exit(void)
2630 {
2631         kvm_exit_debug();
2632         misc_deregister(&kvm_dev);
2633         kmem_cache_destroy(kvm_vcpu_cache);
2634         kvm_async_pf_deinit();
2635         unregister_syscore_ops(&kvm_syscore_ops);
2636         unregister_reboot_notifier(&kvm_reboot_notifier);
2637         unregister_cpu_notifier(&kvm_cpu_notifier);
2638         on_each_cpu(hardware_disable_nolock, NULL, 1);
2639         kvm_arch_hardware_unsetup();
2640         kvm_arch_exit();
2641         free_cpumask_var(cpus_hardware_enabled);
2642         __free_page(hwpoison_page);
2643         __free_page(bad_page);
2644 }
2645 EXPORT_SYMBOL_GPL(kvm_exit);