KVM: add missing void __user * cast to access_ok() call
[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         spin_lock_init(&kvm->mmu_lock);
471
472         r = kvm_init_mmu_notifier(kvm);
473         if (r)
474                 goto out_err;
475
476         kvm->mm = current->mm;
477         atomic_inc(&kvm->mm->mm_count);
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         /* We can read the guest memory with __xxx_user() later on. */
652         if (user_alloc &&
653             ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
654              !access_ok(VERIFY_WRITE,
655                         (void __user *)(unsigned long)mem->userspace_addr,
656                         mem->memory_size)))
657                 goto out;
658         if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
659                 goto out;
660         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
661                 goto out;
662
663         memslot = &kvm->memslots->memslots[mem->slot];
664         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
665         npages = mem->memory_size >> PAGE_SHIFT;
666
667         r = -EINVAL;
668         if (npages > KVM_MEM_MAX_NR_PAGES)
669                 goto out;
670
671         if (!npages)
672                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
673
674         new = old = *memslot;
675
676         new.id = mem->slot;
677         new.base_gfn = base_gfn;
678         new.npages = npages;
679         new.flags = mem->flags;
680
681         /* Disallow changing a memory slot's size. */
682         r = -EINVAL;
683         if (npages && old.npages && npages != old.npages)
684                 goto out_free;
685
686         /* Check for overlaps */
687         r = -EEXIST;
688         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
689                 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
690
691                 if (s == memslot || !s->npages)
692                         continue;
693                 if (!((base_gfn + npages <= s->base_gfn) ||
694                       (base_gfn >= s->base_gfn + s->npages)))
695                         goto out_free;
696         }
697
698         /* Free page dirty bitmap if unneeded */
699         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
700                 new.dirty_bitmap = NULL;
701
702         r = -ENOMEM;
703
704         /* Allocate if a slot is being created */
705 #ifndef CONFIG_S390
706         if (npages && !new.rmap) {
707                 new.rmap = vzalloc(npages * sizeof(*new.rmap));
708
709                 if (!new.rmap)
710                         goto out_free;
711
712                 new.user_alloc = user_alloc;
713                 new.userspace_addr = mem->userspace_addr;
714         }
715         if (!npages)
716                 goto skip_lpage;
717
718         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
719                 unsigned long ugfn;
720                 unsigned long j;
721                 int lpages;
722                 int level = i + 2;
723
724                 /* Avoid unused variable warning if no large pages */
725                 (void)level;
726
727                 if (new.lpage_info[i])
728                         continue;
729
730                 lpages = 1 + ((base_gfn + npages - 1)
731                              >> KVM_HPAGE_GFN_SHIFT(level));
732                 lpages -= base_gfn >> KVM_HPAGE_GFN_SHIFT(level);
733
734                 new.lpage_info[i] = vzalloc(lpages * sizeof(*new.lpage_info[i]));
735
736                 if (!new.lpage_info[i])
737                         goto out_free;
738
739                 if (base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
740                         new.lpage_info[i][0].write_count = 1;
741                 if ((base_gfn+npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
742                         new.lpage_info[i][lpages - 1].write_count = 1;
743                 ugfn = new.userspace_addr >> PAGE_SHIFT;
744                 /*
745                  * If the gfn and userspace address are not aligned wrt each
746                  * other, or if explicitly asked to, disable large page
747                  * support for this slot
748                  */
749                 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
750                     !largepages_enabled)
751                         for (j = 0; j < lpages; ++j)
752                                 new.lpage_info[i][j].write_count = 1;
753         }
754
755 skip_lpage:
756
757         /* Allocate page dirty bitmap if needed */
758         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
759                 if (kvm_create_dirty_bitmap(&new) < 0)
760                         goto out_free;
761                 /* destroy any largepage mappings for dirty tracking */
762         }
763 #else  /* not defined CONFIG_S390 */
764         new.user_alloc = user_alloc;
765         if (user_alloc)
766                 new.userspace_addr = mem->userspace_addr;
767 #endif /* not defined CONFIG_S390 */
768
769         if (!npages) {
770                 r = -ENOMEM;
771                 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
772                 if (!slots)
773                         goto out_free;
774                 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
775                 if (mem->slot >= slots->nmemslots)
776                         slots->nmemslots = mem->slot + 1;
777                 slots->generation++;
778                 slots->memslots[mem->slot].flags |= KVM_MEMSLOT_INVALID;
779
780                 old_memslots = kvm->memslots;
781                 rcu_assign_pointer(kvm->memslots, slots);
782                 synchronize_srcu_expedited(&kvm->srcu);
783                 /* From this point no new shadow pages pointing to a deleted
784                  * memslot will be created.
785                  *
786                  * validation of sp->gfn happens in:
787                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
788                  *      - kvm_is_visible_gfn (mmu_check_roots)
789                  */
790                 kvm_arch_flush_shadow(kvm);
791                 kfree(old_memslots);
792         }
793
794         r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
795         if (r)
796                 goto out_free;
797
798         /* map the pages in iommu page table */
799         if (npages) {
800                 r = kvm_iommu_map_pages(kvm, &new);
801                 if (r)
802                         goto out_free;
803         }
804
805         r = -ENOMEM;
806         slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
807         if (!slots)
808                 goto out_free;
809         memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
810         if (mem->slot >= slots->nmemslots)
811                 slots->nmemslots = mem->slot + 1;
812         slots->generation++;
813
814         /* actual memory is freed via old in kvm_free_physmem_slot below */
815         if (!npages) {
816                 new.rmap = NULL;
817                 new.dirty_bitmap = NULL;
818                 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i)
819                         new.lpage_info[i] = NULL;
820         }
821
822         slots->memslots[mem->slot] = new;
823         old_memslots = kvm->memslots;
824         rcu_assign_pointer(kvm->memslots, slots);
825         synchronize_srcu_expedited(&kvm->srcu);
826
827         kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
828
829         kvm_free_physmem_slot(&old, &new);
830         kfree(old_memslots);
831
832         return 0;
833
834 out_free:
835         kvm_free_physmem_slot(&new, &old);
836 out:
837         return r;
838
839 }
840 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
841
842 int kvm_set_memory_region(struct kvm *kvm,
843                           struct kvm_userspace_memory_region *mem,
844                           int user_alloc)
845 {
846         int r;
847
848         mutex_lock(&kvm->slots_lock);
849         r = __kvm_set_memory_region(kvm, mem, user_alloc);
850         mutex_unlock(&kvm->slots_lock);
851         return r;
852 }
853 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
854
855 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
856                                    struct
857                                    kvm_userspace_memory_region *mem,
858                                    int user_alloc)
859 {
860         if (mem->slot >= KVM_MEMORY_SLOTS)
861                 return -EINVAL;
862         return kvm_set_memory_region(kvm, mem, user_alloc);
863 }
864
865 int kvm_get_dirty_log(struct kvm *kvm,
866                         struct kvm_dirty_log *log, int *is_dirty)
867 {
868         struct kvm_memory_slot *memslot;
869         int r, i;
870         unsigned long n;
871         unsigned long any = 0;
872
873         r = -EINVAL;
874         if (log->slot >= KVM_MEMORY_SLOTS)
875                 goto out;
876
877         memslot = &kvm->memslots->memslots[log->slot];
878         r = -ENOENT;
879         if (!memslot->dirty_bitmap)
880                 goto out;
881
882         n = kvm_dirty_bitmap_bytes(memslot);
883
884         for (i = 0; !any && i < n/sizeof(long); ++i)
885                 any = memslot->dirty_bitmap[i];
886
887         r = -EFAULT;
888         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
889                 goto out;
890
891         if (any)
892                 *is_dirty = 1;
893
894         r = 0;
895 out:
896         return r;
897 }
898
899 void kvm_disable_largepages(void)
900 {
901         largepages_enabled = false;
902 }
903 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
904
905 int is_error_page(struct page *page)
906 {
907         return page == bad_page || page == hwpoison_page || page == fault_page;
908 }
909 EXPORT_SYMBOL_GPL(is_error_page);
910
911 int is_error_pfn(pfn_t pfn)
912 {
913         return pfn == bad_pfn || pfn == hwpoison_pfn || pfn == fault_pfn;
914 }
915 EXPORT_SYMBOL_GPL(is_error_pfn);
916
917 int is_hwpoison_pfn(pfn_t pfn)
918 {
919         return pfn == hwpoison_pfn;
920 }
921 EXPORT_SYMBOL_GPL(is_hwpoison_pfn);
922
923 int is_fault_pfn(pfn_t pfn)
924 {
925         return pfn == fault_pfn;
926 }
927 EXPORT_SYMBOL_GPL(is_fault_pfn);
928
929 static inline unsigned long bad_hva(void)
930 {
931         return PAGE_OFFSET;
932 }
933
934 int kvm_is_error_hva(unsigned long addr)
935 {
936         return addr == bad_hva();
937 }
938 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
939
940 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm_memslots *slots,
941                                                 gfn_t gfn)
942 {
943         int i;
944
945         for (i = 0; i < slots->nmemslots; ++i) {
946                 struct kvm_memory_slot *memslot = &slots->memslots[i];
947
948                 if (gfn >= memslot->base_gfn
949                     && gfn < memslot->base_gfn + memslot->npages)
950                         return memslot;
951         }
952         return NULL;
953 }
954
955 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
956 {
957         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
958 }
959 EXPORT_SYMBOL_GPL(gfn_to_memslot);
960
961 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
962 {
963         int i;
964         struct kvm_memslots *slots = kvm_memslots(kvm);
965
966         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
967                 struct kvm_memory_slot *memslot = &slots->memslots[i];
968
969                 if (memslot->flags & KVM_MEMSLOT_INVALID)
970                         continue;
971
972                 if (gfn >= memslot->base_gfn
973                     && gfn < memslot->base_gfn + memslot->npages)
974                         return 1;
975         }
976         return 0;
977 }
978 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
979
980 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
981 {
982         struct vm_area_struct *vma;
983         unsigned long addr, size;
984
985         size = PAGE_SIZE;
986
987         addr = gfn_to_hva(kvm, gfn);
988         if (kvm_is_error_hva(addr))
989                 return PAGE_SIZE;
990
991         down_read(&current->mm->mmap_sem);
992         vma = find_vma(current->mm, addr);
993         if (!vma)
994                 goto out;
995
996         size = vma_kernel_pagesize(vma);
997
998 out:
999         up_read(&current->mm->mmap_sem);
1000
1001         return size;
1002 }
1003
1004 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1005                                      gfn_t *nr_pages)
1006 {
1007         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1008                 return bad_hva();
1009
1010         if (nr_pages)
1011                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1012
1013         return gfn_to_hva_memslot(slot, gfn);
1014 }
1015
1016 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1017 {
1018         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1019 }
1020 EXPORT_SYMBOL_GPL(gfn_to_hva);
1021
1022 static pfn_t get_fault_pfn(void)
1023 {
1024         get_page(fault_page);
1025         return fault_pfn;
1026 }
1027
1028 int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1029         unsigned long start, int write, struct page **page)
1030 {
1031         int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1032
1033         if (write)
1034                 flags |= FOLL_WRITE;
1035
1036         return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1037 }
1038
1039 static inline int check_user_page_hwpoison(unsigned long addr)
1040 {
1041         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1042
1043         rc = __get_user_pages(current, current->mm, addr, 1,
1044                               flags, NULL, NULL, NULL);
1045         return rc == -EHWPOISON;
1046 }
1047
1048 static pfn_t hva_to_pfn(struct kvm *kvm, unsigned long addr, bool atomic,
1049                         bool *async, bool write_fault, bool *writable)
1050 {
1051         struct page *page[1];
1052         int npages = 0;
1053         pfn_t pfn;
1054
1055         /* we can do it either atomically or asynchronously, not both */
1056         BUG_ON(atomic && async);
1057
1058         BUG_ON(!write_fault && !writable);
1059
1060         if (writable)
1061                 *writable = true;
1062
1063         if (atomic || async)
1064                 npages = __get_user_pages_fast(addr, 1, 1, page);
1065
1066         if (unlikely(npages != 1) && !atomic) {
1067                 might_sleep();
1068
1069                 if (writable)
1070                         *writable = write_fault;
1071
1072                 if (async) {
1073                         down_read(&current->mm->mmap_sem);
1074                         npages = get_user_page_nowait(current, current->mm,
1075                                                      addr, write_fault, page);
1076                         up_read(&current->mm->mmap_sem);
1077                 } else
1078                         npages = get_user_pages_fast(addr, 1, write_fault,
1079                                                      page);
1080
1081                 /* map read fault as writable if possible */
1082                 if (unlikely(!write_fault) && npages == 1) {
1083                         struct page *wpage[1];
1084
1085                         npages = __get_user_pages_fast(addr, 1, 1, wpage);
1086                         if (npages == 1) {
1087                                 *writable = true;
1088                                 put_page(page[0]);
1089                                 page[0] = wpage[0];
1090                         }
1091                         npages = 1;
1092                 }
1093         }
1094
1095         if (unlikely(npages != 1)) {
1096                 struct vm_area_struct *vma;
1097
1098                 if (atomic)
1099                         return get_fault_pfn();
1100
1101                 down_read(&current->mm->mmap_sem);
1102                 if (npages == -EHWPOISON ||
1103                         (!async && check_user_page_hwpoison(addr))) {
1104                         up_read(&current->mm->mmap_sem);
1105                         get_page(hwpoison_page);
1106                         return page_to_pfn(hwpoison_page);
1107                 }
1108
1109                 vma = find_vma_intersection(current->mm, addr, addr+1);
1110
1111                 if (vma == NULL)
1112                         pfn = get_fault_pfn();
1113                 else if ((vma->vm_flags & VM_PFNMAP)) {
1114                         pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1115                                 vma->vm_pgoff;
1116                         BUG_ON(!kvm_is_mmio_pfn(pfn));
1117                 } else {
1118                         if (async && (vma->vm_flags & VM_WRITE))
1119                                 *async = true;
1120                         pfn = get_fault_pfn();
1121                 }
1122                 up_read(&current->mm->mmap_sem);
1123         } else
1124                 pfn = page_to_pfn(page[0]);
1125
1126         return pfn;
1127 }
1128
1129 pfn_t hva_to_pfn_atomic(struct kvm *kvm, unsigned long addr)
1130 {
1131         return hva_to_pfn(kvm, addr, true, NULL, true, NULL);
1132 }
1133 EXPORT_SYMBOL_GPL(hva_to_pfn_atomic);
1134
1135 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1136                           bool write_fault, bool *writable)
1137 {
1138         unsigned long addr;
1139
1140         if (async)
1141                 *async = false;
1142
1143         addr = gfn_to_hva(kvm, gfn);
1144         if (kvm_is_error_hva(addr)) {
1145                 get_page(bad_page);
1146                 return page_to_pfn(bad_page);
1147         }
1148
1149         return hva_to_pfn(kvm, addr, atomic, async, write_fault, writable);
1150 }
1151
1152 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1153 {
1154         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1155 }
1156 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1157
1158 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1159                        bool write_fault, bool *writable)
1160 {
1161         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1162 }
1163 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1164
1165 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1166 {
1167         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1168 }
1169 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1170
1171 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1172                       bool *writable)
1173 {
1174         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1175 }
1176 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1177
1178 pfn_t gfn_to_pfn_memslot(struct kvm *kvm,
1179                          struct kvm_memory_slot *slot, gfn_t gfn)
1180 {
1181         unsigned long addr = gfn_to_hva_memslot(slot, gfn);
1182         return hva_to_pfn(kvm, addr, false, NULL, true, NULL);
1183 }
1184
1185 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1186                                                                   int nr_pages)
1187 {
1188         unsigned long addr;
1189         gfn_t entry;
1190
1191         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1192         if (kvm_is_error_hva(addr))
1193                 return -1;
1194
1195         if (entry < nr_pages)
1196                 return 0;
1197
1198         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1199 }
1200 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1201
1202 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1203 {
1204         pfn_t pfn;
1205
1206         pfn = gfn_to_pfn(kvm, gfn);
1207         if (!kvm_is_mmio_pfn(pfn))
1208                 return pfn_to_page(pfn);
1209
1210         WARN_ON(kvm_is_mmio_pfn(pfn));
1211
1212         get_page(bad_page);
1213         return bad_page;
1214 }
1215
1216 EXPORT_SYMBOL_GPL(gfn_to_page);
1217
1218 void kvm_release_page_clean(struct page *page)
1219 {
1220         kvm_release_pfn_clean(page_to_pfn(page));
1221 }
1222 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1223
1224 void kvm_release_pfn_clean(pfn_t pfn)
1225 {
1226         if (!kvm_is_mmio_pfn(pfn))
1227                 put_page(pfn_to_page(pfn));
1228 }
1229 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1230
1231 void kvm_release_page_dirty(struct page *page)
1232 {
1233         kvm_release_pfn_dirty(page_to_pfn(page));
1234 }
1235 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1236
1237 void kvm_release_pfn_dirty(pfn_t pfn)
1238 {
1239         kvm_set_pfn_dirty(pfn);
1240         kvm_release_pfn_clean(pfn);
1241 }
1242 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1243
1244 void kvm_set_page_dirty(struct page *page)
1245 {
1246         kvm_set_pfn_dirty(page_to_pfn(page));
1247 }
1248 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1249
1250 void kvm_set_pfn_dirty(pfn_t pfn)
1251 {
1252         if (!kvm_is_mmio_pfn(pfn)) {
1253                 struct page *page = pfn_to_page(pfn);
1254                 if (!PageReserved(page))
1255                         SetPageDirty(page);
1256         }
1257 }
1258 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1259
1260 void kvm_set_pfn_accessed(pfn_t pfn)
1261 {
1262         if (!kvm_is_mmio_pfn(pfn))
1263                 mark_page_accessed(pfn_to_page(pfn));
1264 }
1265 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1266
1267 void kvm_get_pfn(pfn_t pfn)
1268 {
1269         if (!kvm_is_mmio_pfn(pfn))
1270                 get_page(pfn_to_page(pfn));
1271 }
1272 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1273
1274 static int next_segment(unsigned long len, int offset)
1275 {
1276         if (len > PAGE_SIZE - offset)
1277                 return PAGE_SIZE - offset;
1278         else
1279                 return len;
1280 }
1281
1282 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1283                         int len)
1284 {
1285         int r;
1286         unsigned long addr;
1287
1288         addr = gfn_to_hva(kvm, gfn);
1289         if (kvm_is_error_hva(addr))
1290                 return -EFAULT;
1291         r = __copy_from_user(data, (void __user *)addr + offset, len);
1292         if (r)
1293                 return -EFAULT;
1294         return 0;
1295 }
1296 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1297
1298 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1299 {
1300         gfn_t gfn = gpa >> PAGE_SHIFT;
1301         int seg;
1302         int offset = offset_in_page(gpa);
1303         int ret;
1304
1305         while ((seg = next_segment(len, offset)) != 0) {
1306                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1307                 if (ret < 0)
1308                         return ret;
1309                 offset = 0;
1310                 len -= seg;
1311                 data += seg;
1312                 ++gfn;
1313         }
1314         return 0;
1315 }
1316 EXPORT_SYMBOL_GPL(kvm_read_guest);
1317
1318 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1319                           unsigned long len)
1320 {
1321         int r;
1322         unsigned long addr;
1323         gfn_t gfn = gpa >> PAGE_SHIFT;
1324         int offset = offset_in_page(gpa);
1325
1326         addr = gfn_to_hva(kvm, gfn);
1327         if (kvm_is_error_hva(addr))
1328                 return -EFAULT;
1329         pagefault_disable();
1330         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1331         pagefault_enable();
1332         if (r)
1333                 return -EFAULT;
1334         return 0;
1335 }
1336 EXPORT_SYMBOL(kvm_read_guest_atomic);
1337
1338 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1339                          int offset, int len)
1340 {
1341         int r;
1342         unsigned long addr;
1343
1344         addr = gfn_to_hva(kvm, gfn);
1345         if (kvm_is_error_hva(addr))
1346                 return -EFAULT;
1347         r = copy_to_user((void __user *)addr + offset, data, len);
1348         if (r)
1349                 return -EFAULT;
1350         mark_page_dirty(kvm, gfn);
1351         return 0;
1352 }
1353 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1354
1355 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1356                     unsigned long len)
1357 {
1358         gfn_t gfn = gpa >> PAGE_SHIFT;
1359         int seg;
1360         int offset = offset_in_page(gpa);
1361         int ret;
1362
1363         while ((seg = next_segment(len, offset)) != 0) {
1364                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1365                 if (ret < 0)
1366                         return ret;
1367                 offset = 0;
1368                 len -= seg;
1369                 data += seg;
1370                 ++gfn;
1371         }
1372         return 0;
1373 }
1374
1375 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1376                               gpa_t gpa)
1377 {
1378         struct kvm_memslots *slots = kvm_memslots(kvm);
1379         int offset = offset_in_page(gpa);
1380         gfn_t gfn = gpa >> PAGE_SHIFT;
1381
1382         ghc->gpa = gpa;
1383         ghc->generation = slots->generation;
1384         ghc->memslot = __gfn_to_memslot(slots, gfn);
1385         ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1386         if (!kvm_is_error_hva(ghc->hva))
1387                 ghc->hva += offset;
1388         else
1389                 return -EFAULT;
1390
1391         return 0;
1392 }
1393 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1394
1395 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1396                            void *data, unsigned long len)
1397 {
1398         struct kvm_memslots *slots = kvm_memslots(kvm);
1399         int r;
1400
1401         if (slots->generation != ghc->generation)
1402                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1403
1404         if (kvm_is_error_hva(ghc->hva))
1405                 return -EFAULT;
1406
1407         r = copy_to_user((void __user *)ghc->hva, data, len);
1408         if (r)
1409                 return -EFAULT;
1410         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1411
1412         return 0;
1413 }
1414 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1415
1416 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1417 {
1418         return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1419                                     offset, len);
1420 }
1421 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1422
1423 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1424 {
1425         gfn_t gfn = gpa >> PAGE_SHIFT;
1426         int seg;
1427         int offset = offset_in_page(gpa);
1428         int ret;
1429
1430         while ((seg = next_segment(len, offset)) != 0) {
1431                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1432                 if (ret < 0)
1433                         return ret;
1434                 offset = 0;
1435                 len -= seg;
1436                 ++gfn;
1437         }
1438         return 0;
1439 }
1440 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1441
1442 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1443                              gfn_t gfn)
1444 {
1445         if (memslot && memslot->dirty_bitmap) {
1446                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1447
1448                 __set_bit_le(rel_gfn, memslot->dirty_bitmap);
1449         }
1450 }
1451
1452 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1453 {
1454         struct kvm_memory_slot *memslot;
1455
1456         memslot = gfn_to_memslot(kvm, gfn);
1457         mark_page_dirty_in_slot(kvm, memslot, gfn);
1458 }
1459
1460 /*
1461  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1462  */
1463 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1464 {
1465         DEFINE_WAIT(wait);
1466
1467         for (;;) {
1468                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1469
1470                 if (kvm_arch_vcpu_runnable(vcpu)) {
1471                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1472                         break;
1473                 }
1474                 if (kvm_cpu_has_pending_timer(vcpu))
1475                         break;
1476                 if (signal_pending(current))
1477                         break;
1478
1479                 schedule();
1480         }
1481
1482         finish_wait(&vcpu->wq, &wait);
1483 }
1484
1485 void kvm_resched(struct kvm_vcpu *vcpu)
1486 {
1487         if (!need_resched())
1488                 return;
1489         cond_resched();
1490 }
1491 EXPORT_SYMBOL_GPL(kvm_resched);
1492
1493 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1494 {
1495         struct kvm *kvm = me->kvm;
1496         struct kvm_vcpu *vcpu;
1497         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1498         int yielded = 0;
1499         int pass;
1500         int i;
1501
1502         /*
1503          * We boost the priority of a VCPU that is runnable but not
1504          * currently running, because it got preempted by something
1505          * else and called schedule in __vcpu_run.  Hopefully that
1506          * VCPU is holding the lock that we need and will release it.
1507          * We approximate round-robin by starting at the last boosted VCPU.
1508          */
1509         for (pass = 0; pass < 2 && !yielded; pass++) {
1510                 kvm_for_each_vcpu(i, vcpu, kvm) {
1511                         struct task_struct *task = NULL;
1512                         struct pid *pid;
1513                         if (!pass && i < last_boosted_vcpu) {
1514                                 i = last_boosted_vcpu;
1515                                 continue;
1516                         } else if (pass && i > last_boosted_vcpu)
1517                                 break;
1518                         if (vcpu == me)
1519                                 continue;
1520                         if (waitqueue_active(&vcpu->wq))
1521                                 continue;
1522                         rcu_read_lock();
1523                         pid = rcu_dereference(vcpu->pid);
1524                         if (pid)
1525                                 task = get_pid_task(vcpu->pid, PIDTYPE_PID);
1526                         rcu_read_unlock();
1527                         if (!task)
1528                                 continue;
1529                         if (task->flags & PF_VCPU) {
1530                                 put_task_struct(task);
1531                                 continue;
1532                         }
1533                         if (yield_to(task, 1)) {
1534                                 put_task_struct(task);
1535                                 kvm->last_boosted_vcpu = i;
1536                                 yielded = 1;
1537                                 break;
1538                         }
1539                         put_task_struct(task);
1540                 }
1541         }
1542 }
1543 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1544
1545 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1546 {
1547         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1548         struct page *page;
1549
1550         if (vmf->pgoff == 0)
1551                 page = virt_to_page(vcpu->run);
1552 #ifdef CONFIG_X86
1553         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1554                 page = virt_to_page(vcpu->arch.pio_data);
1555 #endif
1556 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1557         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1558                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1559 #endif
1560         else
1561                 return VM_FAULT_SIGBUS;
1562         get_page(page);
1563         vmf->page = page;
1564         return 0;
1565 }
1566
1567 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1568         .fault = kvm_vcpu_fault,
1569 };
1570
1571 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1572 {
1573         vma->vm_ops = &kvm_vcpu_vm_ops;
1574         return 0;
1575 }
1576
1577 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1578 {
1579         struct kvm_vcpu *vcpu = filp->private_data;
1580
1581         kvm_put_kvm(vcpu->kvm);
1582         return 0;
1583 }
1584
1585 static struct file_operations kvm_vcpu_fops = {
1586         .release        = kvm_vcpu_release,
1587         .unlocked_ioctl = kvm_vcpu_ioctl,
1588         .compat_ioctl   = kvm_vcpu_ioctl,
1589         .mmap           = kvm_vcpu_mmap,
1590         .llseek         = noop_llseek,
1591 };
1592
1593 /*
1594  * Allocates an inode for the vcpu.
1595  */
1596 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1597 {
1598         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1599 }
1600
1601 /*
1602  * Creates some virtual cpus.  Good luck creating more than one.
1603  */
1604 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1605 {
1606         int r;
1607         struct kvm_vcpu *vcpu, *v;
1608
1609         vcpu = kvm_arch_vcpu_create(kvm, id);
1610         if (IS_ERR(vcpu))
1611                 return PTR_ERR(vcpu);
1612
1613         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1614
1615         r = kvm_arch_vcpu_setup(vcpu);
1616         if (r)
1617                 return r;
1618
1619         mutex_lock(&kvm->lock);
1620         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1621                 r = -EINVAL;
1622                 goto vcpu_destroy;
1623         }
1624
1625         kvm_for_each_vcpu(r, v, kvm)
1626                 if (v->vcpu_id == id) {
1627                         r = -EEXIST;
1628                         goto vcpu_destroy;
1629                 }
1630
1631         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1632
1633         /* Now it's all set up, let userspace reach it */
1634         kvm_get_kvm(kvm);
1635         r = create_vcpu_fd(vcpu);
1636         if (r < 0) {
1637                 kvm_put_kvm(kvm);
1638                 goto vcpu_destroy;
1639         }
1640
1641         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1642         smp_wmb();
1643         atomic_inc(&kvm->online_vcpus);
1644
1645 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1646         if (kvm->bsp_vcpu_id == id)
1647                 kvm->bsp_vcpu = vcpu;
1648 #endif
1649         mutex_unlock(&kvm->lock);
1650         return r;
1651
1652 vcpu_destroy:
1653         mutex_unlock(&kvm->lock);
1654         kvm_arch_vcpu_destroy(vcpu);
1655         return r;
1656 }
1657
1658 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1659 {
1660         if (sigset) {
1661                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1662                 vcpu->sigset_active = 1;
1663                 vcpu->sigset = *sigset;
1664         } else
1665                 vcpu->sigset_active = 0;
1666         return 0;
1667 }
1668
1669 static long kvm_vcpu_ioctl(struct file *filp,
1670                            unsigned int ioctl, unsigned long arg)
1671 {
1672         struct kvm_vcpu *vcpu = filp->private_data;
1673         void __user *argp = (void __user *)arg;
1674         int r;
1675         struct kvm_fpu *fpu = NULL;
1676         struct kvm_sregs *kvm_sregs = NULL;
1677
1678         if (vcpu->kvm->mm != current->mm)
1679                 return -EIO;
1680
1681 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1682         /*
1683          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1684          * so vcpu_load() would break it.
1685          */
1686         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1687                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1688 #endif
1689
1690
1691         vcpu_load(vcpu);
1692         switch (ioctl) {
1693         case KVM_RUN:
1694                 r = -EINVAL;
1695                 if (arg)
1696                         goto out;
1697                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1698                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1699                 break;
1700         case KVM_GET_REGS: {
1701                 struct kvm_regs *kvm_regs;
1702
1703                 r = -ENOMEM;
1704                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1705                 if (!kvm_regs)
1706                         goto out;
1707                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1708                 if (r)
1709                         goto out_free1;
1710                 r = -EFAULT;
1711                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1712                         goto out_free1;
1713                 r = 0;
1714 out_free1:
1715                 kfree(kvm_regs);
1716                 break;
1717         }
1718         case KVM_SET_REGS: {
1719                 struct kvm_regs *kvm_regs;
1720
1721                 r = -ENOMEM;
1722                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1723                 if (!kvm_regs)
1724                         goto out;
1725                 r = -EFAULT;
1726                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1727                         goto out_free2;
1728                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1729                 if (r)
1730                         goto out_free2;
1731                 r = 0;
1732 out_free2:
1733                 kfree(kvm_regs);
1734                 break;
1735         }
1736         case KVM_GET_SREGS: {
1737                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1738                 r = -ENOMEM;
1739                 if (!kvm_sregs)
1740                         goto out;
1741                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1742                 if (r)
1743                         goto out;
1744                 r = -EFAULT;
1745                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1746                         goto out;
1747                 r = 0;
1748                 break;
1749         }
1750         case KVM_SET_SREGS: {
1751                 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1752                 r = -ENOMEM;
1753                 if (!kvm_sregs)
1754                         goto out;
1755                 r = -EFAULT;
1756                 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1757                         goto out;
1758                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1759                 if (r)
1760                         goto out;
1761                 r = 0;
1762                 break;
1763         }
1764         case KVM_GET_MP_STATE: {
1765                 struct kvm_mp_state mp_state;
1766
1767                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1768                 if (r)
1769                         goto out;
1770                 r = -EFAULT;
1771                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1772                         goto out;
1773                 r = 0;
1774                 break;
1775         }
1776         case KVM_SET_MP_STATE: {
1777                 struct kvm_mp_state mp_state;
1778
1779                 r = -EFAULT;
1780                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1781                         goto out;
1782                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1783                 if (r)
1784                         goto out;
1785                 r = 0;
1786                 break;
1787         }
1788         case KVM_TRANSLATE: {
1789                 struct kvm_translation tr;
1790
1791                 r = -EFAULT;
1792                 if (copy_from_user(&tr, argp, sizeof tr))
1793                         goto out;
1794                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1795                 if (r)
1796                         goto out;
1797                 r = -EFAULT;
1798                 if (copy_to_user(argp, &tr, sizeof tr))
1799                         goto out;
1800                 r = 0;
1801                 break;
1802         }
1803         case KVM_SET_GUEST_DEBUG: {
1804                 struct kvm_guest_debug dbg;
1805
1806                 r = -EFAULT;
1807                 if (copy_from_user(&dbg, argp, sizeof dbg))
1808                         goto out;
1809                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1810                 if (r)
1811                         goto out;
1812                 r = 0;
1813                 break;
1814         }
1815         case KVM_SET_SIGNAL_MASK: {
1816                 struct kvm_signal_mask __user *sigmask_arg = argp;
1817                 struct kvm_signal_mask kvm_sigmask;
1818                 sigset_t sigset, *p;
1819
1820                 p = NULL;
1821                 if (argp) {
1822                         r = -EFAULT;
1823                         if (copy_from_user(&kvm_sigmask, argp,
1824                                            sizeof kvm_sigmask))
1825                                 goto out;
1826                         r = -EINVAL;
1827                         if (kvm_sigmask.len != sizeof sigset)
1828                                 goto out;
1829                         r = -EFAULT;
1830                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1831                                            sizeof sigset))
1832                                 goto out;
1833                         p = &sigset;
1834                 }
1835                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
1836                 break;
1837         }
1838         case KVM_GET_FPU: {
1839                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1840                 r = -ENOMEM;
1841                 if (!fpu)
1842                         goto out;
1843                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1844                 if (r)
1845                         goto out;
1846                 r = -EFAULT;
1847                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1848                         goto out;
1849                 r = 0;
1850                 break;
1851         }
1852         case KVM_SET_FPU: {
1853                 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1854                 r = -ENOMEM;
1855                 if (!fpu)
1856                         goto out;
1857                 r = -EFAULT;
1858                 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1859                         goto out;
1860                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1861                 if (r)
1862                         goto out;
1863                 r = 0;
1864                 break;
1865         }
1866         default:
1867                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1868         }
1869 out:
1870         vcpu_put(vcpu);
1871         kfree(fpu);
1872         kfree(kvm_sregs);
1873         return r;
1874 }
1875
1876 static long kvm_vm_ioctl(struct file *filp,
1877                            unsigned int ioctl, unsigned long arg)
1878 {
1879         struct kvm *kvm = filp->private_data;
1880         void __user *argp = (void __user *)arg;
1881         int r;
1882
1883         if (kvm->mm != current->mm)
1884                 return -EIO;
1885         switch (ioctl) {
1886         case KVM_CREATE_VCPU:
1887                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1888                 if (r < 0)
1889                         goto out;
1890                 break;
1891         case KVM_SET_USER_MEMORY_REGION: {
1892                 struct kvm_userspace_memory_region kvm_userspace_mem;
1893
1894                 r = -EFAULT;
1895                 if (copy_from_user(&kvm_userspace_mem, argp,
1896                                                 sizeof kvm_userspace_mem))
1897                         goto out;
1898
1899                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1900                 if (r)
1901                         goto out;
1902                 break;
1903         }
1904         case KVM_GET_DIRTY_LOG: {
1905                 struct kvm_dirty_log log;
1906
1907                 r = -EFAULT;
1908                 if (copy_from_user(&log, argp, sizeof log))
1909                         goto out;
1910                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1911                 if (r)
1912                         goto out;
1913                 break;
1914         }
1915 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1916         case KVM_REGISTER_COALESCED_MMIO: {
1917                 struct kvm_coalesced_mmio_zone zone;
1918                 r = -EFAULT;
1919                 if (copy_from_user(&zone, argp, sizeof zone))
1920                         goto out;
1921                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1922                 if (r)
1923                         goto out;
1924                 r = 0;
1925                 break;
1926         }
1927         case KVM_UNREGISTER_COALESCED_MMIO: {
1928                 struct kvm_coalesced_mmio_zone zone;
1929                 r = -EFAULT;
1930                 if (copy_from_user(&zone, argp, sizeof zone))
1931                         goto out;
1932                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1933                 if (r)
1934                         goto out;
1935                 r = 0;
1936                 break;
1937         }
1938 #endif
1939         case KVM_IRQFD: {
1940                 struct kvm_irqfd data;
1941
1942                 r = -EFAULT;
1943                 if (copy_from_user(&data, argp, sizeof data))
1944                         goto out;
1945                 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
1946                 break;
1947         }
1948         case KVM_IOEVENTFD: {
1949                 struct kvm_ioeventfd data;
1950
1951                 r = -EFAULT;
1952                 if (copy_from_user(&data, argp, sizeof data))
1953                         goto out;
1954                 r = kvm_ioeventfd(kvm, &data);
1955                 break;
1956         }
1957 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1958         case KVM_SET_BOOT_CPU_ID:
1959                 r = 0;
1960                 mutex_lock(&kvm->lock);
1961                 if (atomic_read(&kvm->online_vcpus) != 0)
1962                         r = -EBUSY;
1963                 else
1964                         kvm->bsp_vcpu_id = arg;
1965                 mutex_unlock(&kvm->lock);
1966                 break;
1967 #endif
1968         default:
1969                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1970                 if (r == -ENOTTY)
1971                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
1972         }
1973 out:
1974         return r;
1975 }
1976
1977 #ifdef CONFIG_COMPAT
1978 struct compat_kvm_dirty_log {
1979         __u32 slot;
1980         __u32 padding1;
1981         union {
1982                 compat_uptr_t dirty_bitmap; /* one bit per page */
1983                 __u64 padding2;
1984         };
1985 };
1986
1987 static long kvm_vm_compat_ioctl(struct file *filp,
1988                            unsigned int ioctl, unsigned long arg)
1989 {
1990         struct kvm *kvm = filp->private_data;
1991         int r;
1992
1993         if (kvm->mm != current->mm)
1994                 return -EIO;
1995         switch (ioctl) {
1996         case KVM_GET_DIRTY_LOG: {
1997                 struct compat_kvm_dirty_log compat_log;
1998                 struct kvm_dirty_log log;
1999
2000                 r = -EFAULT;
2001                 if (copy_from_user(&compat_log, (void __user *)arg,
2002                                    sizeof(compat_log)))
2003                         goto out;
2004                 log.slot         = compat_log.slot;
2005                 log.padding1     = compat_log.padding1;
2006                 log.padding2     = compat_log.padding2;
2007                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2008
2009                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2010                 if (r)
2011                         goto out;
2012                 break;
2013         }
2014         default:
2015                 r = kvm_vm_ioctl(filp, ioctl, arg);
2016         }
2017
2018 out:
2019         return r;
2020 }
2021 #endif
2022
2023 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2024 {
2025         struct page *page[1];
2026         unsigned long addr;
2027         int npages;
2028         gfn_t gfn = vmf->pgoff;
2029         struct kvm *kvm = vma->vm_file->private_data;
2030
2031         addr = gfn_to_hva(kvm, gfn);
2032         if (kvm_is_error_hva(addr))
2033                 return VM_FAULT_SIGBUS;
2034
2035         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2036                                 NULL);
2037         if (unlikely(npages != 1))
2038                 return VM_FAULT_SIGBUS;
2039
2040         vmf->page = page[0];
2041         return 0;
2042 }
2043
2044 static const struct vm_operations_struct kvm_vm_vm_ops = {
2045         .fault = kvm_vm_fault,
2046 };
2047
2048 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2049 {
2050         vma->vm_ops = &kvm_vm_vm_ops;
2051         return 0;
2052 }
2053
2054 static struct file_operations kvm_vm_fops = {
2055         .release        = kvm_vm_release,
2056         .unlocked_ioctl = kvm_vm_ioctl,
2057 #ifdef CONFIG_COMPAT
2058         .compat_ioctl   = kvm_vm_compat_ioctl,
2059 #endif
2060         .mmap           = kvm_vm_mmap,
2061         .llseek         = noop_llseek,
2062 };
2063
2064 static int kvm_dev_ioctl_create_vm(void)
2065 {
2066         int r;
2067         struct kvm *kvm;
2068
2069         kvm = kvm_create_vm();
2070         if (IS_ERR(kvm))
2071                 return PTR_ERR(kvm);
2072 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2073         r = kvm_coalesced_mmio_init(kvm);
2074         if (r < 0) {
2075                 kvm_put_kvm(kvm);
2076                 return r;
2077         }
2078 #endif
2079         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2080         if (r < 0)
2081                 kvm_put_kvm(kvm);
2082
2083         return r;
2084 }
2085
2086 static long kvm_dev_ioctl_check_extension_generic(long arg)
2087 {
2088         switch (arg) {
2089         case KVM_CAP_USER_MEMORY:
2090         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2091         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2092 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2093         case KVM_CAP_SET_BOOT_CPU_ID:
2094 #endif
2095         case KVM_CAP_INTERNAL_ERROR_DATA:
2096                 return 1;
2097 #ifdef CONFIG_HAVE_KVM_IRQCHIP
2098         case KVM_CAP_IRQ_ROUTING:
2099                 return KVM_MAX_IRQ_ROUTES;
2100 #endif
2101         default:
2102                 break;
2103         }
2104         return kvm_dev_ioctl_check_extension(arg);
2105 }
2106
2107 static long kvm_dev_ioctl(struct file *filp,
2108                           unsigned int ioctl, unsigned long arg)
2109 {
2110         long r = -EINVAL;
2111
2112         switch (ioctl) {
2113         case KVM_GET_API_VERSION:
2114                 r = -EINVAL;
2115                 if (arg)
2116                         goto out;
2117                 r = KVM_API_VERSION;
2118                 break;
2119         case KVM_CREATE_VM:
2120                 r = -EINVAL;
2121                 if (arg)
2122                         goto out;
2123                 r = kvm_dev_ioctl_create_vm();
2124                 break;
2125         case KVM_CHECK_EXTENSION:
2126                 r = kvm_dev_ioctl_check_extension_generic(arg);
2127                 break;
2128         case KVM_GET_VCPU_MMAP_SIZE:
2129                 r = -EINVAL;
2130                 if (arg)
2131                         goto out;
2132                 r = PAGE_SIZE;     /* struct kvm_run */
2133 #ifdef CONFIG_X86
2134                 r += PAGE_SIZE;    /* pio data page */
2135 #endif
2136 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2137                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2138 #endif
2139                 break;
2140         case KVM_TRACE_ENABLE:
2141         case KVM_TRACE_PAUSE:
2142         case KVM_TRACE_DISABLE:
2143                 r = -EOPNOTSUPP;
2144                 break;
2145         default:
2146                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2147         }
2148 out:
2149         return r;
2150 }
2151
2152 static struct file_operations kvm_chardev_ops = {
2153         .unlocked_ioctl = kvm_dev_ioctl,
2154         .compat_ioctl   = kvm_dev_ioctl,
2155         .llseek         = noop_llseek,
2156 };
2157
2158 static struct miscdevice kvm_dev = {
2159         KVM_MINOR,
2160         "kvm",
2161         &kvm_chardev_ops,
2162 };
2163
2164 static void hardware_enable_nolock(void *junk)
2165 {
2166         int cpu = raw_smp_processor_id();
2167         int r;
2168
2169         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2170                 return;
2171
2172         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2173
2174         r = kvm_arch_hardware_enable(NULL);
2175
2176         if (r) {
2177                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2178                 atomic_inc(&hardware_enable_failed);
2179                 printk(KERN_INFO "kvm: enabling virtualization on "
2180                                  "CPU%d failed\n", cpu);
2181         }
2182 }
2183
2184 static void hardware_enable(void *junk)
2185 {
2186         raw_spin_lock(&kvm_lock);
2187         hardware_enable_nolock(junk);
2188         raw_spin_unlock(&kvm_lock);
2189 }
2190
2191 static void hardware_disable_nolock(void *junk)
2192 {
2193         int cpu = raw_smp_processor_id();
2194
2195         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2196                 return;
2197         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2198         kvm_arch_hardware_disable(NULL);
2199 }
2200
2201 static void hardware_disable(void *junk)
2202 {
2203         raw_spin_lock(&kvm_lock);
2204         hardware_disable_nolock(junk);
2205         raw_spin_unlock(&kvm_lock);
2206 }
2207
2208 static void hardware_disable_all_nolock(void)
2209 {
2210         BUG_ON(!kvm_usage_count);
2211
2212         kvm_usage_count--;
2213         if (!kvm_usage_count)
2214                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2215 }
2216
2217 static void hardware_disable_all(void)
2218 {
2219         raw_spin_lock(&kvm_lock);
2220         hardware_disable_all_nolock();
2221         raw_spin_unlock(&kvm_lock);
2222 }
2223
2224 static int hardware_enable_all(void)
2225 {
2226         int r = 0;
2227
2228         raw_spin_lock(&kvm_lock);
2229
2230         kvm_usage_count++;
2231         if (kvm_usage_count == 1) {
2232                 atomic_set(&hardware_enable_failed, 0);
2233                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2234
2235                 if (atomic_read(&hardware_enable_failed)) {
2236                         hardware_disable_all_nolock();
2237                         r = -EBUSY;
2238                 }
2239         }
2240
2241         raw_spin_unlock(&kvm_lock);
2242
2243         return r;
2244 }
2245
2246 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2247                            void *v)
2248 {
2249         int cpu = (long)v;
2250
2251         if (!kvm_usage_count)
2252                 return NOTIFY_OK;
2253
2254         val &= ~CPU_TASKS_FROZEN;
2255         switch (val) {
2256         case CPU_DYING:
2257                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2258                        cpu);
2259                 hardware_disable(NULL);
2260                 break;
2261         case CPU_STARTING:
2262                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2263                        cpu);
2264                 hardware_enable(NULL);
2265                 break;
2266         }
2267         return NOTIFY_OK;
2268 }
2269
2270
2271 asmlinkage void kvm_spurious_fault(void)
2272 {
2273         /* Fault while not rebooting.  We want the trace. */
2274         BUG();
2275 }
2276 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
2277
2278 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2279                       void *v)
2280 {
2281         /*
2282          * Some (well, at least mine) BIOSes hang on reboot if
2283          * in vmx root mode.
2284          *
2285          * And Intel TXT required VMX off for all cpu when system shutdown.
2286          */
2287         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2288         kvm_rebooting = true;
2289         on_each_cpu(hardware_disable_nolock, NULL, 1);
2290         return NOTIFY_OK;
2291 }
2292
2293 static struct notifier_block kvm_reboot_notifier = {
2294         .notifier_call = kvm_reboot,
2295         .priority = 0,
2296 };
2297
2298 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2299 {
2300         int i;
2301
2302         for (i = 0; i < bus->dev_count; i++) {
2303                 struct kvm_io_device *pos = bus->devs[i];
2304
2305                 kvm_iodevice_destructor(pos);
2306         }
2307         kfree(bus);
2308 }
2309
2310 /* kvm_io_bus_write - called under kvm->slots_lock */
2311 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2312                      int len, const void *val)
2313 {
2314         int i;
2315         struct kvm_io_bus *bus;
2316
2317         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2318         for (i = 0; i < bus->dev_count; i++)
2319                 if (!kvm_iodevice_write(bus->devs[i], addr, len, val))
2320                         return 0;
2321         return -EOPNOTSUPP;
2322 }
2323
2324 /* kvm_io_bus_read - called under kvm->slots_lock */
2325 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2326                     int len, void *val)
2327 {
2328         int i;
2329         struct kvm_io_bus *bus;
2330
2331         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2332         for (i = 0; i < bus->dev_count; i++)
2333                 if (!kvm_iodevice_read(bus->devs[i], addr, len, val))
2334                         return 0;
2335         return -EOPNOTSUPP;
2336 }
2337
2338 /* Caller must hold slots_lock. */
2339 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2340                             struct kvm_io_device *dev)
2341 {
2342         struct kvm_io_bus *new_bus, *bus;
2343
2344         bus = kvm->buses[bus_idx];
2345         if (bus->dev_count > NR_IOBUS_DEVS-1)
2346                 return -ENOSPC;
2347
2348         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2349         if (!new_bus)
2350                 return -ENOMEM;
2351         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2352         new_bus->devs[new_bus->dev_count++] = dev;
2353         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2354         synchronize_srcu_expedited(&kvm->srcu);
2355         kfree(bus);
2356
2357         return 0;
2358 }
2359
2360 /* Caller must hold slots_lock. */
2361 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2362                               struct kvm_io_device *dev)
2363 {
2364         int i, r;
2365         struct kvm_io_bus *new_bus, *bus;
2366
2367         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2368         if (!new_bus)
2369                 return -ENOMEM;
2370
2371         bus = kvm->buses[bus_idx];
2372         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2373
2374         r = -ENOENT;
2375         for (i = 0; i < new_bus->dev_count; i++)
2376                 if (new_bus->devs[i] == dev) {
2377                         r = 0;
2378                         new_bus->devs[i] = new_bus->devs[--new_bus->dev_count];
2379                         break;
2380                 }
2381
2382         if (r) {
2383                 kfree(new_bus);
2384                 return r;
2385         }
2386
2387         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2388         synchronize_srcu_expedited(&kvm->srcu);
2389         kfree(bus);
2390         return r;
2391 }
2392
2393 static struct notifier_block kvm_cpu_notifier = {
2394         .notifier_call = kvm_cpu_hotplug,
2395 };
2396
2397 static int vm_stat_get(void *_offset, u64 *val)
2398 {
2399         unsigned offset = (long)_offset;
2400         struct kvm *kvm;
2401
2402         *val = 0;
2403         raw_spin_lock(&kvm_lock);
2404         list_for_each_entry(kvm, &vm_list, vm_list)
2405                 *val += *(u32 *)((void *)kvm + offset);
2406         raw_spin_unlock(&kvm_lock);
2407         return 0;
2408 }
2409
2410 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2411
2412 static int vcpu_stat_get(void *_offset, u64 *val)
2413 {
2414         unsigned offset = (long)_offset;
2415         struct kvm *kvm;
2416         struct kvm_vcpu *vcpu;
2417         int i;
2418
2419         *val = 0;
2420         raw_spin_lock(&kvm_lock);
2421         list_for_each_entry(kvm, &vm_list, vm_list)
2422                 kvm_for_each_vcpu(i, vcpu, kvm)
2423                         *val += *(u32 *)((void *)vcpu + offset);
2424
2425         raw_spin_unlock(&kvm_lock);
2426         return 0;
2427 }
2428
2429 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2430
2431 static const struct file_operations *stat_fops[] = {
2432         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2433         [KVM_STAT_VM]   = &vm_stat_fops,
2434 };
2435
2436 static void kvm_init_debug(void)
2437 {
2438         struct kvm_stats_debugfs_item *p;
2439
2440         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2441         for (p = debugfs_entries; p->name; ++p)
2442                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2443                                                 (void *)(long)p->offset,
2444                                                 stat_fops[p->kind]);
2445 }
2446
2447 static void kvm_exit_debug(void)
2448 {
2449         struct kvm_stats_debugfs_item *p;
2450
2451         for (p = debugfs_entries; p->name; ++p)
2452                 debugfs_remove(p->dentry);
2453         debugfs_remove(kvm_debugfs_dir);
2454 }
2455
2456 static int kvm_suspend(void)
2457 {
2458         if (kvm_usage_count)
2459                 hardware_disable_nolock(NULL);
2460         return 0;
2461 }
2462
2463 static void kvm_resume(void)
2464 {
2465         if (kvm_usage_count) {
2466                 WARN_ON(raw_spin_is_locked(&kvm_lock));
2467                 hardware_enable_nolock(NULL);
2468         }
2469 }
2470
2471 static struct syscore_ops kvm_syscore_ops = {
2472         .suspend = kvm_suspend,
2473         .resume = kvm_resume,
2474 };
2475
2476 struct page *bad_page;
2477 pfn_t bad_pfn;
2478
2479 static inline
2480 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2481 {
2482         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2483 }
2484
2485 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2486 {
2487         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2488
2489         kvm_arch_vcpu_load(vcpu, cpu);
2490 }
2491
2492 static void kvm_sched_out(struct preempt_notifier *pn,
2493                           struct task_struct *next)
2494 {
2495         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2496
2497         kvm_arch_vcpu_put(vcpu);
2498 }
2499
2500 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2501                   struct module *module)
2502 {
2503         int r;
2504         int cpu;
2505
2506         r = kvm_arch_init(opaque);
2507         if (r)
2508                 goto out_fail;
2509
2510         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2511
2512         if (bad_page == NULL) {
2513                 r = -ENOMEM;
2514                 goto out;
2515         }
2516
2517         bad_pfn = page_to_pfn(bad_page);
2518
2519         hwpoison_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2520
2521         if (hwpoison_page == NULL) {
2522                 r = -ENOMEM;
2523                 goto out_free_0;
2524         }
2525
2526         hwpoison_pfn = page_to_pfn(hwpoison_page);
2527
2528         fault_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2529
2530         if (fault_page == NULL) {
2531                 r = -ENOMEM;
2532                 goto out_free_0;
2533         }
2534
2535         fault_pfn = page_to_pfn(fault_page);
2536
2537         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2538                 r = -ENOMEM;
2539                 goto out_free_0;
2540         }
2541
2542         r = kvm_arch_hardware_setup();
2543         if (r < 0)
2544                 goto out_free_0a;
2545
2546         for_each_online_cpu(cpu) {
2547                 smp_call_function_single(cpu,
2548                                 kvm_arch_check_processor_compat,
2549                                 &r, 1);
2550                 if (r < 0)
2551                         goto out_free_1;
2552         }
2553
2554         r = register_cpu_notifier(&kvm_cpu_notifier);
2555         if (r)
2556                 goto out_free_2;
2557         register_reboot_notifier(&kvm_reboot_notifier);
2558
2559         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2560         if (!vcpu_align)
2561                 vcpu_align = __alignof__(struct kvm_vcpu);
2562         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2563                                            0, NULL);
2564         if (!kvm_vcpu_cache) {
2565                 r = -ENOMEM;
2566                 goto out_free_3;
2567         }
2568
2569         r = kvm_async_pf_init();
2570         if (r)
2571                 goto out_free;
2572
2573         kvm_chardev_ops.owner = module;
2574         kvm_vm_fops.owner = module;
2575         kvm_vcpu_fops.owner = module;
2576
2577         r = misc_register(&kvm_dev);
2578         if (r) {
2579                 printk(KERN_ERR "kvm: misc device register failed\n");
2580                 goto out_unreg;
2581         }
2582
2583         register_syscore_ops(&kvm_syscore_ops);
2584
2585         kvm_preempt_ops.sched_in = kvm_sched_in;
2586         kvm_preempt_ops.sched_out = kvm_sched_out;
2587
2588         kvm_init_debug();
2589
2590         return 0;
2591
2592 out_unreg:
2593         kvm_async_pf_deinit();
2594 out_free:
2595         kmem_cache_destroy(kvm_vcpu_cache);
2596 out_free_3:
2597         unregister_reboot_notifier(&kvm_reboot_notifier);
2598         unregister_cpu_notifier(&kvm_cpu_notifier);
2599 out_free_2:
2600 out_free_1:
2601         kvm_arch_hardware_unsetup();
2602 out_free_0a:
2603         free_cpumask_var(cpus_hardware_enabled);
2604 out_free_0:
2605         if (fault_page)
2606                 __free_page(fault_page);
2607         if (hwpoison_page)
2608                 __free_page(hwpoison_page);
2609         __free_page(bad_page);
2610 out:
2611         kvm_arch_exit();
2612 out_fail:
2613         return r;
2614 }
2615 EXPORT_SYMBOL_GPL(kvm_init);
2616
2617 void kvm_exit(void)
2618 {
2619         kvm_exit_debug();
2620         misc_deregister(&kvm_dev);
2621         kmem_cache_destroy(kvm_vcpu_cache);
2622         kvm_async_pf_deinit();
2623         unregister_syscore_ops(&kvm_syscore_ops);
2624         unregister_reboot_notifier(&kvm_reboot_notifier);
2625         unregister_cpu_notifier(&kvm_cpu_notifier);
2626         on_each_cpu(hardware_disable_nolock, NULL, 1);
2627         kvm_arch_hardware_unsetup();
2628         kvm_arch_exit();
2629         free_cpumask_var(cpus_hardware_enabled);
2630         __free_page(hwpoison_page);
2631         __free_page(bad_page);
2632 }
2633 EXPORT_SYMBOL_GPL(kvm_exit);