58146457bf97fafeef9306f4dcd59b91578b185b
[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 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1000                                      gfn_t *nr_pages)
1001 {
1002         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1003                 return bad_hva();
1004
1005         if (nr_pages)
1006                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1007
1008         return gfn_to_hva_memslot(slot, gfn);
1009 }
1010
1011 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1012 {
1013         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1014 }
1015 EXPORT_SYMBOL_GPL(gfn_to_hva);
1016
1017 static pfn_t get_fault_pfn(void)
1018 {
1019         get_page(fault_page);
1020         return fault_pfn;
1021 }
1022
1023 int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1024         unsigned long start, int write, struct page **page)
1025 {
1026         int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1027
1028         if (write)
1029                 flags |= FOLL_WRITE;
1030
1031         return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1032 }
1033
1034 static inline int check_user_page_hwpoison(unsigned long addr)
1035 {
1036         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1037
1038         rc = __get_user_pages(current, current->mm, addr, 1,
1039                               flags, NULL, NULL, NULL);
1040         return rc == -EHWPOISON;
1041 }
1042
1043 static pfn_t hva_to_pfn(struct kvm *kvm, unsigned long addr, bool atomic,
1044                         bool *async, bool write_fault, bool *writable)
1045 {
1046         struct page *page[1];
1047         int npages = 0;
1048         pfn_t pfn;
1049
1050         /* we can do it either atomically or asynchronously, not both */
1051         BUG_ON(atomic && async);
1052
1053         BUG_ON(!write_fault && !writable);
1054
1055         if (writable)
1056                 *writable = true;
1057
1058         if (atomic || async)
1059                 npages = __get_user_pages_fast(addr, 1, 1, page);
1060
1061         if (unlikely(npages != 1) && !atomic) {
1062                 might_sleep();
1063
1064                 if (writable)
1065                         *writable = write_fault;
1066
1067                 if (async) {
1068                         down_read(&current->mm->mmap_sem);
1069                         npages = get_user_page_nowait(current, current->mm,
1070                                                      addr, write_fault, page);
1071                         up_read(&current->mm->mmap_sem);
1072                 } else
1073                         npages = get_user_pages_fast(addr, 1, write_fault,
1074                                                      page);
1075
1076                 /* map read fault as writable if possible */
1077                 if (unlikely(!write_fault) && npages == 1) {
1078                         struct page *wpage[1];
1079
1080                         npages = __get_user_pages_fast(addr, 1, 1, wpage);
1081                         if (npages == 1) {
1082                                 *writable = true;
1083                                 put_page(page[0]);
1084                                 page[0] = wpage[0];
1085                         }
1086                         npages = 1;
1087                 }
1088         }
1089
1090         if (unlikely(npages != 1)) {
1091                 struct vm_area_struct *vma;
1092
1093                 if (atomic)
1094                         return get_fault_pfn();
1095
1096                 down_read(&current->mm->mmap_sem);
1097                 if (npages == -EHWPOISON ||
1098                         (!async && check_user_page_hwpoison(addr))) {
1099                         up_read(&current->mm->mmap_sem);
1100                         get_page(hwpoison_page);
1101                         return page_to_pfn(hwpoison_page);
1102                 }
1103
1104                 vma = find_vma_intersection(current->mm, addr, addr+1);
1105
1106                 if (vma == NULL)
1107                         pfn = get_fault_pfn();
1108                 else if ((vma->vm_flags & VM_PFNMAP)) {
1109                         pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1110                                 vma->vm_pgoff;
1111                         BUG_ON(!kvm_is_mmio_pfn(pfn));
1112                 } else {
1113                         if (async && (vma->vm_flags & VM_WRITE))
1114                                 *async = true;
1115                         pfn = get_fault_pfn();
1116                 }
1117                 up_read(&current->mm->mmap_sem);
1118         } else
1119                 pfn = page_to_pfn(page[0]);
1120
1121         return pfn;
1122 }
1123
1124 pfn_t hva_to_pfn_atomic(struct kvm *kvm, unsigned long addr)
1125 {
1126         return hva_to_pfn(kvm, addr, true, NULL, true, NULL);
1127 }
1128 EXPORT_SYMBOL_GPL(hva_to_pfn_atomic);
1129
1130 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1131                           bool write_fault, bool *writable)
1132 {
1133         unsigned long addr;
1134
1135         if (async)
1136                 *async = false;
1137
1138         addr = gfn_to_hva(kvm, gfn);
1139         if (kvm_is_error_hva(addr)) {
1140                 get_page(bad_page);
1141                 return page_to_pfn(bad_page);
1142         }
1143
1144         return hva_to_pfn(kvm, addr, atomic, async, write_fault, writable);
1145 }
1146
1147 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1148 {
1149         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1150 }
1151 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1152
1153 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1154                        bool write_fault, bool *writable)
1155 {
1156         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1157 }
1158 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1159
1160 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1161 {
1162         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1163 }
1164 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1165
1166 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1167                       bool *writable)
1168 {
1169         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1170 }
1171 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1172
1173 pfn_t gfn_to_pfn_memslot(struct kvm *kvm,
1174                          struct kvm_memory_slot *slot, gfn_t gfn)
1175 {
1176         unsigned long addr = gfn_to_hva_memslot(slot, gfn);
1177         return hva_to_pfn(kvm, addr, false, NULL, true, NULL);
1178 }
1179
1180 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1181                                                                   int nr_pages)
1182 {
1183         unsigned long addr;
1184         gfn_t entry;
1185
1186         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1187         if (kvm_is_error_hva(addr))
1188                 return -1;
1189
1190         if (entry < nr_pages)
1191                 return 0;
1192
1193         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1194 }
1195 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1196
1197 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1198 {
1199         pfn_t pfn;
1200
1201         pfn = gfn_to_pfn(kvm, gfn);
1202         if (!kvm_is_mmio_pfn(pfn))
1203                 return pfn_to_page(pfn);
1204
1205         WARN_ON(kvm_is_mmio_pfn(pfn));
1206
1207         get_page(bad_page);
1208         return bad_page;
1209 }
1210
1211 EXPORT_SYMBOL_GPL(gfn_to_page);
1212
1213 void kvm_release_page_clean(struct page *page)
1214 {
1215         kvm_release_pfn_clean(page_to_pfn(page));
1216 }
1217 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1218
1219 void kvm_release_pfn_clean(pfn_t pfn)
1220 {
1221         if (!kvm_is_mmio_pfn(pfn))
1222                 put_page(pfn_to_page(pfn));
1223 }
1224 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1225
1226 void kvm_release_page_dirty(struct page *page)
1227 {
1228         kvm_release_pfn_dirty(page_to_pfn(page));
1229 }
1230 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1231
1232 void kvm_release_pfn_dirty(pfn_t pfn)
1233 {
1234         kvm_set_pfn_dirty(pfn);
1235         kvm_release_pfn_clean(pfn);
1236 }
1237 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1238
1239 void kvm_set_page_dirty(struct page *page)
1240 {
1241         kvm_set_pfn_dirty(page_to_pfn(page));
1242 }
1243 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1244
1245 void kvm_set_pfn_dirty(pfn_t pfn)
1246 {
1247         if (!kvm_is_mmio_pfn(pfn)) {
1248                 struct page *page = pfn_to_page(pfn);
1249                 if (!PageReserved(page))
1250                         SetPageDirty(page);
1251         }
1252 }
1253 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1254
1255 void kvm_set_pfn_accessed(pfn_t pfn)
1256 {
1257         if (!kvm_is_mmio_pfn(pfn))
1258                 mark_page_accessed(pfn_to_page(pfn));
1259 }
1260 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1261
1262 void kvm_get_pfn(pfn_t pfn)
1263 {
1264         if (!kvm_is_mmio_pfn(pfn))
1265                 get_page(pfn_to_page(pfn));
1266 }
1267 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1268
1269 static int next_segment(unsigned long len, int offset)
1270 {
1271         if (len > PAGE_SIZE - offset)
1272                 return PAGE_SIZE - offset;
1273         else
1274                 return len;
1275 }
1276
1277 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1278                         int len)
1279 {
1280         int r;
1281         unsigned long addr;
1282
1283         addr = gfn_to_hva(kvm, gfn);
1284         if (kvm_is_error_hva(addr))
1285                 return -EFAULT;
1286         r = copy_from_user(data, (void __user *)addr + offset, len);
1287         if (r)
1288                 return -EFAULT;
1289         return 0;
1290 }
1291 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1292
1293 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1294 {
1295         gfn_t gfn = gpa >> PAGE_SHIFT;
1296         int seg;
1297         int offset = offset_in_page(gpa);
1298         int ret;
1299
1300         while ((seg = next_segment(len, offset)) != 0) {
1301                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1302                 if (ret < 0)
1303                         return ret;
1304                 offset = 0;
1305                 len -= seg;
1306                 data += seg;
1307                 ++gfn;
1308         }
1309         return 0;
1310 }
1311 EXPORT_SYMBOL_GPL(kvm_read_guest);
1312
1313 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1314                           unsigned long len)
1315 {
1316         int r;
1317         unsigned long addr;
1318         gfn_t gfn = gpa >> PAGE_SHIFT;
1319         int offset = offset_in_page(gpa);
1320
1321         addr = gfn_to_hva(kvm, gfn);
1322         if (kvm_is_error_hva(addr))
1323                 return -EFAULT;
1324         pagefault_disable();
1325         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1326         pagefault_enable();
1327         if (r)
1328                 return -EFAULT;
1329         return 0;
1330 }
1331 EXPORT_SYMBOL(kvm_read_guest_atomic);
1332
1333 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1334                          int offset, int len)
1335 {
1336         int r;
1337         unsigned long addr;
1338
1339         addr = gfn_to_hva(kvm, gfn);
1340         if (kvm_is_error_hva(addr))
1341                 return -EFAULT;
1342         r = copy_to_user((void __user *)addr + offset, data, len);
1343         if (r)
1344                 return -EFAULT;
1345         mark_page_dirty(kvm, gfn);
1346         return 0;
1347 }
1348 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1349
1350 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1351                     unsigned long len)
1352 {
1353         gfn_t gfn = gpa >> PAGE_SHIFT;
1354         int seg;
1355         int offset = offset_in_page(gpa);
1356         int ret;
1357
1358         while ((seg = next_segment(len, offset)) != 0) {
1359                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1360                 if (ret < 0)
1361                         return ret;
1362                 offset = 0;
1363                 len -= seg;
1364                 data += seg;
1365                 ++gfn;
1366         }
1367         return 0;
1368 }
1369
1370 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1371                               gpa_t gpa)
1372 {
1373         struct kvm_memslots *slots = kvm_memslots(kvm);
1374         int offset = offset_in_page(gpa);
1375         gfn_t gfn = gpa >> PAGE_SHIFT;
1376
1377         ghc->gpa = gpa;
1378         ghc->generation = slots->generation;
1379         ghc->memslot = __gfn_to_memslot(slots, gfn);
1380         ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1381         if (!kvm_is_error_hva(ghc->hva))
1382                 ghc->hva += offset;
1383         else
1384                 return -EFAULT;
1385
1386         return 0;
1387 }
1388 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1389
1390 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1391                            void *data, unsigned long len)
1392 {
1393         struct kvm_memslots *slots = kvm_memslots(kvm);
1394         int r;
1395
1396         if (slots->generation != ghc->generation)
1397                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1398
1399         if (kvm_is_error_hva(ghc->hva))
1400                 return -EFAULT;
1401
1402         r = copy_to_user((void __user *)ghc->hva, data, len);
1403         if (r)
1404                 return -EFAULT;
1405         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1406
1407         return 0;
1408 }
1409 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1410
1411 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1412 {
1413         return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1414                                     offset, len);
1415 }
1416 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1417
1418 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1419 {
1420         gfn_t gfn = gpa >> PAGE_SHIFT;
1421         int seg;
1422         int offset = offset_in_page(gpa);
1423         int ret;
1424
1425         while ((seg = next_segment(len, offset)) != 0) {
1426                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1427                 if (ret < 0)
1428                         return ret;
1429                 offset = 0;
1430                 len -= seg;
1431                 ++gfn;
1432         }
1433         return 0;
1434 }
1435 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1436
1437 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1438                              gfn_t gfn)
1439 {
1440         if (memslot && memslot->dirty_bitmap) {
1441                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1442
1443                 __set_bit_le(rel_gfn, memslot->dirty_bitmap);
1444         }
1445 }
1446
1447 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1448 {
1449         struct kvm_memory_slot *memslot;
1450
1451         memslot = gfn_to_memslot(kvm, gfn);
1452         mark_page_dirty_in_slot(kvm, memslot, gfn);
1453 }
1454
1455 /*
1456  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1457  */
1458 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1459 {
1460         DEFINE_WAIT(wait);
1461
1462         for (;;) {
1463                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1464
1465                 if (kvm_arch_vcpu_runnable(vcpu)) {
1466                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1467                         break;
1468                 }
1469                 if (kvm_cpu_has_pending_timer(vcpu))
1470                         break;
1471                 if (signal_pending(current))
1472                         break;
1473
1474                 schedule();
1475         }
1476
1477         finish_wait(&vcpu->wq, &wait);
1478 }
1479
1480 void kvm_resched(struct kvm_vcpu *vcpu)
1481 {
1482         if (!need_resched())
1483                 return;
1484         cond_resched();
1485 }
1486 EXPORT_SYMBOL_GPL(kvm_resched);
1487
1488 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1489 {
1490         struct kvm *kvm = me->kvm;
1491         struct kvm_vcpu *vcpu;
1492         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1493         int yielded = 0;
1494         int pass;
1495         int i;
1496
1497         /*
1498          * We boost the priority of a VCPU that is runnable but not
1499          * currently running, because it got preempted by something
1500          * else and called schedule in __vcpu_run.  Hopefully that
1501          * VCPU is holding the lock that we need and will release it.
1502          * We approximate round-robin by starting at the last boosted VCPU.
1503          */
1504         for (pass = 0; pass < 2 && !yielded; pass++) {
1505                 kvm_for_each_vcpu(i, vcpu, kvm) {
1506                         struct task_struct *task = NULL;
1507                         struct pid *pid;
1508                         if (!pass && i < last_boosted_vcpu) {
1509                                 i = last_boosted_vcpu;
1510                                 continue;
1511                         } else if (pass && i > last_boosted_vcpu)
1512                                 break;
1513                         if (vcpu == me)
1514                                 continue;
1515                         if (waitqueue_active(&vcpu->wq))
1516                                 continue;
1517                         rcu_read_lock();
1518                         pid = rcu_dereference(vcpu->pid);
1519                         if (pid)
1520                                 task = get_pid_task(vcpu->pid, PIDTYPE_PID);
1521                         rcu_read_unlock();
1522                         if (!task)
1523                                 continue;
1524                         if (task->flags & PF_VCPU) {
1525                                 put_task_struct(task);
1526                                 continue;
1527                         }
1528                         if (yield_to(task, 1)) {
1529                                 put_task_struct(task);
1530                                 kvm->last_boosted_vcpu = i;
1531                                 yielded = 1;
1532                                 break;
1533                         }
1534                         put_task_struct(task);
1535                 }
1536         }
1537 }
1538 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1539
1540 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1541 {
1542         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1543         struct page *page;
1544
1545         if (vmf->pgoff == 0)
1546                 page = virt_to_page(vcpu->run);
1547 #ifdef CONFIG_X86
1548         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1549                 page = virt_to_page(vcpu->arch.pio_data);
1550 #endif
1551 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1552         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1553                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1554 #endif
1555         else
1556                 return VM_FAULT_SIGBUS;
1557         get_page(page);
1558         vmf->page = page;
1559         return 0;
1560 }
1561
1562 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1563         .fault = kvm_vcpu_fault,
1564 };
1565
1566 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1567 {
1568         vma->vm_ops = &kvm_vcpu_vm_ops;
1569         return 0;
1570 }
1571
1572 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1573 {
1574         struct kvm_vcpu *vcpu = filp->private_data;
1575
1576         kvm_put_kvm(vcpu->kvm);
1577         return 0;
1578 }
1579
1580 static struct file_operations kvm_vcpu_fops = {
1581         .release        = kvm_vcpu_release,
1582         .unlocked_ioctl = kvm_vcpu_ioctl,
1583         .compat_ioctl   = kvm_vcpu_ioctl,
1584         .mmap           = kvm_vcpu_mmap,
1585         .llseek         = noop_llseek,
1586 };
1587
1588 /*
1589  * Allocates an inode for the vcpu.
1590  */
1591 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1592 {
1593         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1594 }
1595
1596 /*
1597  * Creates some virtual cpus.  Good luck creating more than one.
1598  */
1599 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1600 {
1601         int r;
1602         struct kvm_vcpu *vcpu, *v;
1603
1604         vcpu = kvm_arch_vcpu_create(kvm, id);
1605         if (IS_ERR(vcpu))
1606                 return PTR_ERR(vcpu);
1607
1608         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1609
1610         r = kvm_arch_vcpu_setup(vcpu);
1611         if (r)
1612                 return r;
1613
1614         mutex_lock(&kvm->lock);
1615         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1616                 r = -EINVAL;
1617                 goto vcpu_destroy;
1618         }
1619
1620         kvm_for_each_vcpu(r, v, kvm)
1621                 if (v->vcpu_id == id) {
1622                         r = -EEXIST;
1623                         goto vcpu_destroy;
1624                 }
1625
1626         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1627
1628         /* Now it's all set up, let userspace reach it */
1629         kvm_get_kvm(kvm);
1630         r = create_vcpu_fd(vcpu);
1631         if (r < 0) {
1632                 kvm_put_kvm(kvm);
1633                 goto vcpu_destroy;
1634         }
1635
1636         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1637         smp_wmb();
1638         atomic_inc(&kvm->online_vcpus);
1639
1640 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1641         if (kvm->bsp_vcpu_id == id)
1642                 kvm->bsp_vcpu = vcpu;
1643 #endif
1644         mutex_unlock(&kvm->lock);
1645         return r;
1646
1647 vcpu_destroy:
1648         mutex_unlock(&kvm->lock);
1649         kvm_arch_vcpu_destroy(vcpu);
1650         return r;
1651 }
1652
1653 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1654 {
1655         if (sigset) {
1656                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1657                 vcpu->sigset_active = 1;
1658                 vcpu->sigset = *sigset;
1659         } else
1660                 vcpu->sigset_active = 0;
1661         return 0;
1662 }
1663
1664 static long kvm_vcpu_ioctl(struct file *filp,
1665                            unsigned int ioctl, unsigned long arg)
1666 {
1667         struct kvm_vcpu *vcpu = filp->private_data;
1668         void __user *argp = (void __user *)arg;
1669         int r;
1670         struct kvm_fpu *fpu = NULL;
1671         struct kvm_sregs *kvm_sregs = NULL;
1672
1673         if (vcpu->kvm->mm != current->mm)
1674                 return -EIO;
1675
1676 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1677         /*
1678          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1679          * so vcpu_load() would break it.
1680          */
1681         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1682                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1683 #endif
1684
1685
1686         vcpu_load(vcpu);
1687         switch (ioctl) {
1688         case KVM_RUN:
1689                 r = -EINVAL;
1690                 if (arg)
1691                         goto out;
1692                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1693                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1694                 break;
1695         case KVM_GET_REGS: {
1696                 struct kvm_regs *kvm_regs;
1697
1698                 r = -ENOMEM;
1699                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1700                 if (!kvm_regs)
1701                         goto out;
1702                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1703                 if (r)
1704                         goto out_free1;
1705                 r = -EFAULT;
1706                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1707                         goto out_free1;
1708                 r = 0;
1709 out_free1:
1710                 kfree(kvm_regs);
1711                 break;
1712         }
1713         case KVM_SET_REGS: {
1714                 struct kvm_regs *kvm_regs;
1715
1716                 r = -ENOMEM;
1717                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1718                 if (!kvm_regs)
1719                         goto out;
1720                 r = -EFAULT;
1721                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1722                         goto out_free2;
1723                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1724                 if (r)
1725                         goto out_free2;
1726                 r = 0;
1727 out_free2:
1728                 kfree(kvm_regs);
1729                 break;
1730         }
1731         case KVM_GET_SREGS: {
1732                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1733                 r = -ENOMEM;
1734                 if (!kvm_sregs)
1735                         goto out;
1736                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1737                 if (r)
1738                         goto out;
1739                 r = -EFAULT;
1740                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1741                         goto out;
1742                 r = 0;
1743                 break;
1744         }
1745         case KVM_SET_SREGS: {
1746                 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1747                 r = -ENOMEM;
1748                 if (!kvm_sregs)
1749                         goto out;
1750                 r = -EFAULT;
1751                 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1752                         goto out;
1753                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1754                 if (r)
1755                         goto out;
1756                 r = 0;
1757                 break;
1758         }
1759         case KVM_GET_MP_STATE: {
1760                 struct kvm_mp_state mp_state;
1761
1762                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1763                 if (r)
1764                         goto out;
1765                 r = -EFAULT;
1766                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1767                         goto out;
1768                 r = 0;
1769                 break;
1770         }
1771         case KVM_SET_MP_STATE: {
1772                 struct kvm_mp_state mp_state;
1773
1774                 r = -EFAULT;
1775                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1776                         goto out;
1777                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1778                 if (r)
1779                         goto out;
1780                 r = 0;
1781                 break;
1782         }
1783         case KVM_TRANSLATE: {
1784                 struct kvm_translation tr;
1785
1786                 r = -EFAULT;
1787                 if (copy_from_user(&tr, argp, sizeof tr))
1788                         goto out;
1789                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1790                 if (r)
1791                         goto out;
1792                 r = -EFAULT;
1793                 if (copy_to_user(argp, &tr, sizeof tr))
1794                         goto out;
1795                 r = 0;
1796                 break;
1797         }
1798         case KVM_SET_GUEST_DEBUG: {
1799                 struct kvm_guest_debug dbg;
1800
1801                 r = -EFAULT;
1802                 if (copy_from_user(&dbg, argp, sizeof dbg))
1803                         goto out;
1804                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1805                 if (r)
1806                         goto out;
1807                 r = 0;
1808                 break;
1809         }
1810         case KVM_SET_SIGNAL_MASK: {
1811                 struct kvm_signal_mask __user *sigmask_arg = argp;
1812                 struct kvm_signal_mask kvm_sigmask;
1813                 sigset_t sigset, *p;
1814
1815                 p = NULL;
1816                 if (argp) {
1817                         r = -EFAULT;
1818                         if (copy_from_user(&kvm_sigmask, argp,
1819                                            sizeof kvm_sigmask))
1820                                 goto out;
1821                         r = -EINVAL;
1822                         if (kvm_sigmask.len != sizeof sigset)
1823                                 goto out;
1824                         r = -EFAULT;
1825                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1826                                            sizeof sigset))
1827                                 goto out;
1828                         p = &sigset;
1829                 }
1830                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
1831                 break;
1832         }
1833         case KVM_GET_FPU: {
1834                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1835                 r = -ENOMEM;
1836                 if (!fpu)
1837                         goto out;
1838                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1839                 if (r)
1840                         goto out;
1841                 r = -EFAULT;
1842                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1843                         goto out;
1844                 r = 0;
1845                 break;
1846         }
1847         case KVM_SET_FPU: {
1848                 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1849                 r = -ENOMEM;
1850                 if (!fpu)
1851                         goto out;
1852                 r = -EFAULT;
1853                 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1854                         goto out;
1855                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1856                 if (r)
1857                         goto out;
1858                 r = 0;
1859                 break;
1860         }
1861         default:
1862                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1863         }
1864 out:
1865         vcpu_put(vcpu);
1866         kfree(fpu);
1867         kfree(kvm_sregs);
1868         return r;
1869 }
1870
1871 static long kvm_vm_ioctl(struct file *filp,
1872                            unsigned int ioctl, unsigned long arg)
1873 {
1874         struct kvm *kvm = filp->private_data;
1875         void __user *argp = (void __user *)arg;
1876         int r;
1877
1878         if (kvm->mm != current->mm)
1879                 return -EIO;
1880         switch (ioctl) {
1881         case KVM_CREATE_VCPU:
1882                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1883                 if (r < 0)
1884                         goto out;
1885                 break;
1886         case KVM_SET_USER_MEMORY_REGION: {
1887                 struct kvm_userspace_memory_region kvm_userspace_mem;
1888
1889                 r = -EFAULT;
1890                 if (copy_from_user(&kvm_userspace_mem, argp,
1891                                                 sizeof kvm_userspace_mem))
1892                         goto out;
1893
1894                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1895                 if (r)
1896                         goto out;
1897                 break;
1898         }
1899         case KVM_GET_DIRTY_LOG: {
1900                 struct kvm_dirty_log log;
1901
1902                 r = -EFAULT;
1903                 if (copy_from_user(&log, argp, sizeof log))
1904                         goto out;
1905                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1906                 if (r)
1907                         goto out;
1908                 break;
1909         }
1910 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1911         case KVM_REGISTER_COALESCED_MMIO: {
1912                 struct kvm_coalesced_mmio_zone zone;
1913                 r = -EFAULT;
1914                 if (copy_from_user(&zone, argp, sizeof zone))
1915                         goto out;
1916                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1917                 if (r)
1918                         goto out;
1919                 r = 0;
1920                 break;
1921         }
1922         case KVM_UNREGISTER_COALESCED_MMIO: {
1923                 struct kvm_coalesced_mmio_zone zone;
1924                 r = -EFAULT;
1925                 if (copy_from_user(&zone, argp, sizeof zone))
1926                         goto out;
1927                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1928                 if (r)
1929                         goto out;
1930                 r = 0;
1931                 break;
1932         }
1933 #endif
1934         case KVM_IRQFD: {
1935                 struct kvm_irqfd data;
1936
1937                 r = -EFAULT;
1938                 if (copy_from_user(&data, argp, sizeof data))
1939                         goto out;
1940                 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
1941                 break;
1942         }
1943         case KVM_IOEVENTFD: {
1944                 struct kvm_ioeventfd data;
1945
1946                 r = -EFAULT;
1947                 if (copy_from_user(&data, argp, sizeof data))
1948                         goto out;
1949                 r = kvm_ioeventfd(kvm, &data);
1950                 break;
1951         }
1952 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1953         case KVM_SET_BOOT_CPU_ID:
1954                 r = 0;
1955                 mutex_lock(&kvm->lock);
1956                 if (atomic_read(&kvm->online_vcpus) != 0)
1957                         r = -EBUSY;
1958                 else
1959                         kvm->bsp_vcpu_id = arg;
1960                 mutex_unlock(&kvm->lock);
1961                 break;
1962 #endif
1963         default:
1964                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1965                 if (r == -ENOTTY)
1966                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
1967         }
1968 out:
1969         return r;
1970 }
1971
1972 #ifdef CONFIG_COMPAT
1973 struct compat_kvm_dirty_log {
1974         __u32 slot;
1975         __u32 padding1;
1976         union {
1977                 compat_uptr_t dirty_bitmap; /* one bit per page */
1978                 __u64 padding2;
1979         };
1980 };
1981
1982 static long kvm_vm_compat_ioctl(struct file *filp,
1983                            unsigned int ioctl, unsigned long arg)
1984 {
1985         struct kvm *kvm = filp->private_data;
1986         int r;
1987
1988         if (kvm->mm != current->mm)
1989                 return -EIO;
1990         switch (ioctl) {
1991         case KVM_GET_DIRTY_LOG: {
1992                 struct compat_kvm_dirty_log compat_log;
1993                 struct kvm_dirty_log log;
1994
1995                 r = -EFAULT;
1996                 if (copy_from_user(&compat_log, (void __user *)arg,
1997                                    sizeof(compat_log)))
1998                         goto out;
1999                 log.slot         = compat_log.slot;
2000                 log.padding1     = compat_log.padding1;
2001                 log.padding2     = compat_log.padding2;
2002                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2003
2004                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2005                 if (r)
2006                         goto out;
2007                 break;
2008         }
2009         default:
2010                 r = kvm_vm_ioctl(filp, ioctl, arg);
2011         }
2012
2013 out:
2014         return r;
2015 }
2016 #endif
2017
2018 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2019 {
2020         struct page *page[1];
2021         unsigned long addr;
2022         int npages;
2023         gfn_t gfn = vmf->pgoff;
2024         struct kvm *kvm = vma->vm_file->private_data;
2025
2026         addr = gfn_to_hva(kvm, gfn);
2027         if (kvm_is_error_hva(addr))
2028                 return VM_FAULT_SIGBUS;
2029
2030         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2031                                 NULL);
2032         if (unlikely(npages != 1))
2033                 return VM_FAULT_SIGBUS;
2034
2035         vmf->page = page[0];
2036         return 0;
2037 }
2038
2039 static const struct vm_operations_struct kvm_vm_vm_ops = {
2040         .fault = kvm_vm_fault,
2041 };
2042
2043 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2044 {
2045         vma->vm_ops = &kvm_vm_vm_ops;
2046         return 0;
2047 }
2048
2049 static struct file_operations kvm_vm_fops = {
2050         .release        = kvm_vm_release,
2051         .unlocked_ioctl = kvm_vm_ioctl,
2052 #ifdef CONFIG_COMPAT
2053         .compat_ioctl   = kvm_vm_compat_ioctl,
2054 #endif
2055         .mmap           = kvm_vm_mmap,
2056         .llseek         = noop_llseek,
2057 };
2058
2059 static int kvm_dev_ioctl_create_vm(void)
2060 {
2061         int r;
2062         struct kvm *kvm;
2063
2064         kvm = kvm_create_vm();
2065         if (IS_ERR(kvm))
2066                 return PTR_ERR(kvm);
2067 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2068         r = kvm_coalesced_mmio_init(kvm);
2069         if (r < 0) {
2070                 kvm_put_kvm(kvm);
2071                 return r;
2072         }
2073 #endif
2074         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2075         if (r < 0)
2076                 kvm_put_kvm(kvm);
2077
2078         return r;
2079 }
2080
2081 static long kvm_dev_ioctl_check_extension_generic(long arg)
2082 {
2083         switch (arg) {
2084         case KVM_CAP_USER_MEMORY:
2085         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2086         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2087 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2088         case KVM_CAP_SET_BOOT_CPU_ID:
2089 #endif
2090         case KVM_CAP_INTERNAL_ERROR_DATA:
2091                 return 1;
2092 #ifdef CONFIG_HAVE_KVM_IRQCHIP
2093         case KVM_CAP_IRQ_ROUTING:
2094                 return KVM_MAX_IRQ_ROUTES;
2095 #endif
2096         default:
2097                 break;
2098         }
2099         return kvm_dev_ioctl_check_extension(arg);
2100 }
2101
2102 static long kvm_dev_ioctl(struct file *filp,
2103                           unsigned int ioctl, unsigned long arg)
2104 {
2105         long r = -EINVAL;
2106
2107         switch (ioctl) {
2108         case KVM_GET_API_VERSION:
2109                 r = -EINVAL;
2110                 if (arg)
2111                         goto out;
2112                 r = KVM_API_VERSION;
2113                 break;
2114         case KVM_CREATE_VM:
2115                 r = -EINVAL;
2116                 if (arg)
2117                         goto out;
2118                 r = kvm_dev_ioctl_create_vm();
2119                 break;
2120         case KVM_CHECK_EXTENSION:
2121                 r = kvm_dev_ioctl_check_extension_generic(arg);
2122                 break;
2123         case KVM_GET_VCPU_MMAP_SIZE:
2124                 r = -EINVAL;
2125                 if (arg)
2126                         goto out;
2127                 r = PAGE_SIZE;     /* struct kvm_run */
2128 #ifdef CONFIG_X86
2129                 r += PAGE_SIZE;    /* pio data page */
2130 #endif
2131 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2132                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2133 #endif
2134                 break;
2135         case KVM_TRACE_ENABLE:
2136         case KVM_TRACE_PAUSE:
2137         case KVM_TRACE_DISABLE:
2138                 r = -EOPNOTSUPP;
2139                 break;
2140         default:
2141                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2142         }
2143 out:
2144         return r;
2145 }
2146
2147 static struct file_operations kvm_chardev_ops = {
2148         .unlocked_ioctl = kvm_dev_ioctl,
2149         .compat_ioctl   = kvm_dev_ioctl,
2150         .llseek         = noop_llseek,
2151 };
2152
2153 static struct miscdevice kvm_dev = {
2154         KVM_MINOR,
2155         "kvm",
2156         &kvm_chardev_ops,
2157 };
2158
2159 static void hardware_enable_nolock(void *junk)
2160 {
2161         int cpu = raw_smp_processor_id();
2162         int r;
2163
2164         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2165                 return;
2166
2167         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2168
2169         r = kvm_arch_hardware_enable(NULL);
2170
2171         if (r) {
2172                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2173                 atomic_inc(&hardware_enable_failed);
2174                 printk(KERN_INFO "kvm: enabling virtualization on "
2175                                  "CPU%d failed\n", cpu);
2176         }
2177 }
2178
2179 static void hardware_enable(void *junk)
2180 {
2181         raw_spin_lock(&kvm_lock);
2182         hardware_enable_nolock(junk);
2183         raw_spin_unlock(&kvm_lock);
2184 }
2185
2186 static void hardware_disable_nolock(void *junk)
2187 {
2188         int cpu = raw_smp_processor_id();
2189
2190         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2191                 return;
2192         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2193         kvm_arch_hardware_disable(NULL);
2194 }
2195
2196 static void hardware_disable(void *junk)
2197 {
2198         raw_spin_lock(&kvm_lock);
2199         hardware_disable_nolock(junk);
2200         raw_spin_unlock(&kvm_lock);
2201 }
2202
2203 static void hardware_disable_all_nolock(void)
2204 {
2205         BUG_ON(!kvm_usage_count);
2206
2207         kvm_usage_count--;
2208         if (!kvm_usage_count)
2209                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2210 }
2211
2212 static void hardware_disable_all(void)
2213 {
2214         raw_spin_lock(&kvm_lock);
2215         hardware_disable_all_nolock();
2216         raw_spin_unlock(&kvm_lock);
2217 }
2218
2219 static int hardware_enable_all(void)
2220 {
2221         int r = 0;
2222
2223         raw_spin_lock(&kvm_lock);
2224
2225         kvm_usage_count++;
2226         if (kvm_usage_count == 1) {
2227                 atomic_set(&hardware_enable_failed, 0);
2228                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2229
2230                 if (atomic_read(&hardware_enable_failed)) {
2231                         hardware_disable_all_nolock();
2232                         r = -EBUSY;
2233                 }
2234         }
2235
2236         raw_spin_unlock(&kvm_lock);
2237
2238         return r;
2239 }
2240
2241 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2242                            void *v)
2243 {
2244         int cpu = (long)v;
2245
2246         if (!kvm_usage_count)
2247                 return NOTIFY_OK;
2248
2249         val &= ~CPU_TASKS_FROZEN;
2250         switch (val) {
2251         case CPU_DYING:
2252                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2253                        cpu);
2254                 hardware_disable(NULL);
2255                 break;
2256         case CPU_STARTING:
2257                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2258                        cpu);
2259                 hardware_enable(NULL);
2260                 break;
2261         }
2262         return NOTIFY_OK;
2263 }
2264
2265
2266 asmlinkage void kvm_spurious_fault(void)
2267 {
2268         /* Fault while not rebooting.  We want the trace. */
2269         BUG();
2270 }
2271 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
2272
2273 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2274                       void *v)
2275 {
2276         /*
2277          * Some (well, at least mine) BIOSes hang on reboot if
2278          * in vmx root mode.
2279          *
2280          * And Intel TXT required VMX off for all cpu when system shutdown.
2281          */
2282         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2283         kvm_rebooting = true;
2284         on_each_cpu(hardware_disable_nolock, NULL, 1);
2285         return NOTIFY_OK;
2286 }
2287
2288 static struct notifier_block kvm_reboot_notifier = {
2289         .notifier_call = kvm_reboot,
2290         .priority = 0,
2291 };
2292
2293 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2294 {
2295         int i;
2296
2297         for (i = 0; i < bus->dev_count; i++) {
2298                 struct kvm_io_device *pos = bus->devs[i];
2299
2300                 kvm_iodevice_destructor(pos);
2301         }
2302         kfree(bus);
2303 }
2304
2305 /* kvm_io_bus_write - called under kvm->slots_lock */
2306 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2307                      int len, const void *val)
2308 {
2309         int i;
2310         struct kvm_io_bus *bus;
2311
2312         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2313         for (i = 0; i < bus->dev_count; i++)
2314                 if (!kvm_iodevice_write(bus->devs[i], addr, len, val))
2315                         return 0;
2316         return -EOPNOTSUPP;
2317 }
2318
2319 /* kvm_io_bus_read - called under kvm->slots_lock */
2320 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2321                     int len, void *val)
2322 {
2323         int i;
2324         struct kvm_io_bus *bus;
2325
2326         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2327         for (i = 0; i < bus->dev_count; i++)
2328                 if (!kvm_iodevice_read(bus->devs[i], addr, len, val))
2329                         return 0;
2330         return -EOPNOTSUPP;
2331 }
2332
2333 /* Caller must hold slots_lock. */
2334 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2335                             struct kvm_io_device *dev)
2336 {
2337         struct kvm_io_bus *new_bus, *bus;
2338
2339         bus = kvm->buses[bus_idx];
2340         if (bus->dev_count > NR_IOBUS_DEVS-1)
2341                 return -ENOSPC;
2342
2343         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2344         if (!new_bus)
2345                 return -ENOMEM;
2346         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2347         new_bus->devs[new_bus->dev_count++] = dev;
2348         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2349         synchronize_srcu_expedited(&kvm->srcu);
2350         kfree(bus);
2351
2352         return 0;
2353 }
2354
2355 /* Caller must hold slots_lock. */
2356 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2357                               struct kvm_io_device *dev)
2358 {
2359         int i, r;
2360         struct kvm_io_bus *new_bus, *bus;
2361
2362         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2363         if (!new_bus)
2364                 return -ENOMEM;
2365
2366         bus = kvm->buses[bus_idx];
2367         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2368
2369         r = -ENOENT;
2370         for (i = 0; i < new_bus->dev_count; i++)
2371                 if (new_bus->devs[i] == dev) {
2372                         r = 0;
2373                         new_bus->devs[i] = new_bus->devs[--new_bus->dev_count];
2374                         break;
2375                 }
2376
2377         if (r) {
2378                 kfree(new_bus);
2379                 return r;
2380         }
2381
2382         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2383         synchronize_srcu_expedited(&kvm->srcu);
2384         kfree(bus);
2385         return r;
2386 }
2387
2388 static struct notifier_block kvm_cpu_notifier = {
2389         .notifier_call = kvm_cpu_hotplug,
2390 };
2391
2392 static int vm_stat_get(void *_offset, u64 *val)
2393 {
2394         unsigned offset = (long)_offset;
2395         struct kvm *kvm;
2396
2397         *val = 0;
2398         raw_spin_lock(&kvm_lock);
2399         list_for_each_entry(kvm, &vm_list, vm_list)
2400                 *val += *(u32 *)((void *)kvm + offset);
2401         raw_spin_unlock(&kvm_lock);
2402         return 0;
2403 }
2404
2405 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2406
2407 static int vcpu_stat_get(void *_offset, u64 *val)
2408 {
2409         unsigned offset = (long)_offset;
2410         struct kvm *kvm;
2411         struct kvm_vcpu *vcpu;
2412         int i;
2413
2414         *val = 0;
2415         raw_spin_lock(&kvm_lock);
2416         list_for_each_entry(kvm, &vm_list, vm_list)
2417                 kvm_for_each_vcpu(i, vcpu, kvm)
2418                         *val += *(u32 *)((void *)vcpu + offset);
2419
2420         raw_spin_unlock(&kvm_lock);
2421         return 0;
2422 }
2423
2424 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2425
2426 static const struct file_operations *stat_fops[] = {
2427         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2428         [KVM_STAT_VM]   = &vm_stat_fops,
2429 };
2430
2431 static void kvm_init_debug(void)
2432 {
2433         struct kvm_stats_debugfs_item *p;
2434
2435         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2436         for (p = debugfs_entries; p->name; ++p)
2437                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2438                                                 (void *)(long)p->offset,
2439                                                 stat_fops[p->kind]);
2440 }
2441
2442 static void kvm_exit_debug(void)
2443 {
2444         struct kvm_stats_debugfs_item *p;
2445
2446         for (p = debugfs_entries; p->name; ++p)
2447                 debugfs_remove(p->dentry);
2448         debugfs_remove(kvm_debugfs_dir);
2449 }
2450
2451 static int kvm_suspend(void)
2452 {
2453         if (kvm_usage_count)
2454                 hardware_disable_nolock(NULL);
2455         return 0;
2456 }
2457
2458 static void kvm_resume(void)
2459 {
2460         if (kvm_usage_count) {
2461                 WARN_ON(raw_spin_is_locked(&kvm_lock));
2462                 hardware_enable_nolock(NULL);
2463         }
2464 }
2465
2466 static struct syscore_ops kvm_syscore_ops = {
2467         .suspend = kvm_suspend,
2468         .resume = kvm_resume,
2469 };
2470
2471 struct page *bad_page;
2472 pfn_t bad_pfn;
2473
2474 static inline
2475 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2476 {
2477         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2478 }
2479
2480 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2481 {
2482         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2483
2484         kvm_arch_vcpu_load(vcpu, cpu);
2485 }
2486
2487 static void kvm_sched_out(struct preempt_notifier *pn,
2488                           struct task_struct *next)
2489 {
2490         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2491
2492         kvm_arch_vcpu_put(vcpu);
2493 }
2494
2495 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2496                   struct module *module)
2497 {
2498         int r;
2499         int cpu;
2500
2501         r = kvm_arch_init(opaque);
2502         if (r)
2503                 goto out_fail;
2504
2505         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2506
2507         if (bad_page == NULL) {
2508                 r = -ENOMEM;
2509                 goto out;
2510         }
2511
2512         bad_pfn = page_to_pfn(bad_page);
2513
2514         hwpoison_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2515
2516         if (hwpoison_page == NULL) {
2517                 r = -ENOMEM;
2518                 goto out_free_0;
2519         }
2520
2521         hwpoison_pfn = page_to_pfn(hwpoison_page);
2522
2523         fault_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2524
2525         if (fault_page == NULL) {
2526                 r = -ENOMEM;
2527                 goto out_free_0;
2528         }
2529
2530         fault_pfn = page_to_pfn(fault_page);
2531
2532         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2533                 r = -ENOMEM;
2534                 goto out_free_0;
2535         }
2536
2537         r = kvm_arch_hardware_setup();
2538         if (r < 0)
2539                 goto out_free_0a;
2540
2541         for_each_online_cpu(cpu) {
2542                 smp_call_function_single(cpu,
2543                                 kvm_arch_check_processor_compat,
2544                                 &r, 1);
2545                 if (r < 0)
2546                         goto out_free_1;
2547         }
2548
2549         r = register_cpu_notifier(&kvm_cpu_notifier);
2550         if (r)
2551                 goto out_free_2;
2552         register_reboot_notifier(&kvm_reboot_notifier);
2553
2554         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2555         if (!vcpu_align)
2556                 vcpu_align = __alignof__(struct kvm_vcpu);
2557         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2558                                            0, NULL);
2559         if (!kvm_vcpu_cache) {
2560                 r = -ENOMEM;
2561                 goto out_free_3;
2562         }
2563
2564         r = kvm_async_pf_init();
2565         if (r)
2566                 goto out_free;
2567
2568         kvm_chardev_ops.owner = module;
2569         kvm_vm_fops.owner = module;
2570         kvm_vcpu_fops.owner = module;
2571
2572         r = misc_register(&kvm_dev);
2573         if (r) {
2574                 printk(KERN_ERR "kvm: misc device register failed\n");
2575                 goto out_unreg;
2576         }
2577
2578         register_syscore_ops(&kvm_syscore_ops);
2579
2580         kvm_preempt_ops.sched_in = kvm_sched_in;
2581         kvm_preempt_ops.sched_out = kvm_sched_out;
2582
2583         kvm_init_debug();
2584
2585         return 0;
2586
2587 out_unreg:
2588         kvm_async_pf_deinit();
2589 out_free:
2590         kmem_cache_destroy(kvm_vcpu_cache);
2591 out_free_3:
2592         unregister_reboot_notifier(&kvm_reboot_notifier);
2593         unregister_cpu_notifier(&kvm_cpu_notifier);
2594 out_free_2:
2595 out_free_1:
2596         kvm_arch_hardware_unsetup();
2597 out_free_0a:
2598         free_cpumask_var(cpus_hardware_enabled);
2599 out_free_0:
2600         if (fault_page)
2601                 __free_page(fault_page);
2602         if (hwpoison_page)
2603                 __free_page(hwpoison_page);
2604         __free_page(bad_page);
2605 out:
2606         kvm_arch_exit();
2607 out_fail:
2608         return r;
2609 }
2610 EXPORT_SYMBOL_GPL(kvm_init);
2611
2612 void kvm_exit(void)
2613 {
2614         kvm_exit_debug();
2615         misc_deregister(&kvm_dev);
2616         kmem_cache_destroy(kvm_vcpu_cache);
2617         kvm_async_pf_deinit();
2618         unregister_syscore_ops(&kvm_syscore_ops);
2619         unregister_reboot_notifier(&kvm_reboot_notifier);
2620         unregister_cpu_notifier(&kvm_cpu_notifier);
2621         on_each_cpu(hardware_disable_nolock, NULL, 1);
2622         kvm_arch_hardware_unsetup();
2623         kvm_arch_exit();
2624         free_cpumask_var(cpus_hardware_enabled);
2625         __free_page(hwpoison_page);
2626         __free_page(bad_page);
2627 }
2628 EXPORT_SYMBOL_GPL(kvm_exit);