KVM: Initialize kvm before registering the mmu notifier
[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         spin_lock_init(&kvm->mmu_lock);
472         kvm->mm = current->mm;
473         atomic_inc(&kvm->mm->mm_count);
474         kvm_eventfd_init(kvm);
475         mutex_init(&kvm->lock);
476         mutex_init(&kvm->irq_lock);
477         mutex_init(&kvm->slots_lock);
478         atomic_set(&kvm->users_count, 1);
479
480         r = kvm_init_mmu_notifier(kvm);
481         if (r)
482                 goto out_err;
483
484         raw_spin_lock(&kvm_lock);
485         list_add(&kvm->vm_list, &vm_list);
486         raw_spin_unlock(&kvm_lock);
487
488         return kvm;
489
490 out_err:
491         cleanup_srcu_struct(&kvm->srcu);
492 out_err_nosrcu:
493         hardware_disable_all();
494 out_err_nodisable:
495         for (i = 0; i < KVM_NR_BUSES; i++)
496                 kfree(kvm->buses[i]);
497         kfree(kvm->memslots);
498         kvm_arch_free_vm(kvm);
499         return ERR_PTR(r);
500 }
501
502 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
503 {
504         if (!memslot->dirty_bitmap)
505                 return;
506
507         if (2 * kvm_dirty_bitmap_bytes(memslot) > PAGE_SIZE)
508                 vfree(memslot->dirty_bitmap_head);
509         else
510                 kfree(memslot->dirty_bitmap_head);
511
512         memslot->dirty_bitmap = NULL;
513         memslot->dirty_bitmap_head = NULL;
514 }
515
516 /*
517  * Free any memory in @free but not in @dont.
518  */
519 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
520                                   struct kvm_memory_slot *dont)
521 {
522         int i;
523
524         if (!dont || free->rmap != dont->rmap)
525                 vfree(free->rmap);
526
527         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
528                 kvm_destroy_dirty_bitmap(free);
529
530
531         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
532                 if (!dont || free->lpage_info[i] != dont->lpage_info[i]) {
533                         vfree(free->lpage_info[i]);
534                         free->lpage_info[i] = NULL;
535                 }
536         }
537
538         free->npages = 0;
539         free->rmap = NULL;
540 }
541
542 void kvm_free_physmem(struct kvm *kvm)
543 {
544         int i;
545         struct kvm_memslots *slots = kvm->memslots;
546
547         for (i = 0; i < slots->nmemslots; ++i)
548                 kvm_free_physmem_slot(&slots->memslots[i], NULL);
549
550         kfree(kvm->memslots);
551 }
552
553 static void kvm_destroy_vm(struct kvm *kvm)
554 {
555         int i;
556         struct mm_struct *mm = kvm->mm;
557
558         kvm_arch_sync_events(kvm);
559         raw_spin_lock(&kvm_lock);
560         list_del(&kvm->vm_list);
561         raw_spin_unlock(&kvm_lock);
562         kvm_free_irq_routing(kvm);
563         for (i = 0; i < KVM_NR_BUSES; i++)
564                 kvm_io_bus_destroy(kvm->buses[i]);
565         kvm_coalesced_mmio_free(kvm);
566 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
567         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
568 #else
569         kvm_arch_flush_shadow(kvm);
570 #endif
571         kvm_arch_destroy_vm(kvm);
572         kvm_free_physmem(kvm);
573         cleanup_srcu_struct(&kvm->srcu);
574         kvm_arch_free_vm(kvm);
575         hardware_disable_all();
576         mmdrop(mm);
577 }
578
579 void kvm_get_kvm(struct kvm *kvm)
580 {
581         atomic_inc(&kvm->users_count);
582 }
583 EXPORT_SYMBOL_GPL(kvm_get_kvm);
584
585 void kvm_put_kvm(struct kvm *kvm)
586 {
587         if (atomic_dec_and_test(&kvm->users_count))
588                 kvm_destroy_vm(kvm);
589 }
590 EXPORT_SYMBOL_GPL(kvm_put_kvm);
591
592
593 static int kvm_vm_release(struct inode *inode, struct file *filp)
594 {
595         struct kvm *kvm = filp->private_data;
596
597         kvm_irqfd_release(kvm);
598
599         kvm_put_kvm(kvm);
600         return 0;
601 }
602
603 #ifndef CONFIG_S390
604 /*
605  * Allocation size is twice as large as the actual dirty bitmap size.
606  * This makes it possible to do double buffering: see x86's
607  * kvm_vm_ioctl_get_dirty_log().
608  */
609 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
610 {
611         unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
612
613         if (dirty_bytes > PAGE_SIZE)
614                 memslot->dirty_bitmap = vzalloc(dirty_bytes);
615         else
616                 memslot->dirty_bitmap = kzalloc(dirty_bytes, GFP_KERNEL);
617
618         if (!memslot->dirty_bitmap)
619                 return -ENOMEM;
620
621         memslot->dirty_bitmap_head = memslot->dirty_bitmap;
622         return 0;
623 }
624 #endif /* !CONFIG_S390 */
625
626 /*
627  * Allocate some memory and give it an address in the guest physical address
628  * space.
629  *
630  * Discontiguous memory is allowed, mostly for framebuffers.
631  *
632  * Must be called holding mmap_sem for write.
633  */
634 int __kvm_set_memory_region(struct kvm *kvm,
635                             struct kvm_userspace_memory_region *mem,
636                             int user_alloc)
637 {
638         int r;
639         gfn_t base_gfn;
640         unsigned long npages;
641         unsigned long i;
642         struct kvm_memory_slot *memslot;
643         struct kvm_memory_slot old, new;
644         struct kvm_memslots *slots, *old_memslots;
645
646         r = -EINVAL;
647         /* General sanity checks */
648         if (mem->memory_size & (PAGE_SIZE - 1))
649                 goto out;
650         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
651                 goto out;
652         /* We can read the guest memory with __xxx_user() later on. */
653         if (user_alloc &&
654             ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
655              !access_ok(VERIFY_WRITE,
656                         (void __user *)(unsigned long)mem->userspace_addr,
657                         mem->memory_size)))
658                 goto out;
659         if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
660                 goto out;
661         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
662                 goto out;
663
664         memslot = &kvm->memslots->memslots[mem->slot];
665         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
666         npages = mem->memory_size >> PAGE_SHIFT;
667
668         r = -EINVAL;
669         if (npages > KVM_MEM_MAX_NR_PAGES)
670                 goto out;
671
672         if (!npages)
673                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
674
675         new = old = *memslot;
676
677         new.id = mem->slot;
678         new.base_gfn = base_gfn;
679         new.npages = npages;
680         new.flags = mem->flags;
681
682         /* Disallow changing a memory slot's size. */
683         r = -EINVAL;
684         if (npages && old.npages && npages != old.npages)
685                 goto out_free;
686
687         /* Check for overlaps */
688         r = -EEXIST;
689         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
690                 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
691
692                 if (s == memslot || !s->npages)
693                         continue;
694                 if (!((base_gfn + npages <= s->base_gfn) ||
695                       (base_gfn >= s->base_gfn + s->npages)))
696                         goto out_free;
697         }
698
699         /* Free page dirty bitmap if unneeded */
700         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
701                 new.dirty_bitmap = NULL;
702
703         r = -ENOMEM;
704
705         /* Allocate if a slot is being created */
706 #ifndef CONFIG_S390
707         if (npages && !new.rmap) {
708                 new.rmap = vzalloc(npages * sizeof(*new.rmap));
709
710                 if (!new.rmap)
711                         goto out_free;
712
713                 new.user_alloc = user_alloc;
714                 new.userspace_addr = mem->userspace_addr;
715         }
716         if (!npages)
717                 goto skip_lpage;
718
719         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
720                 unsigned long ugfn;
721                 unsigned long j;
722                 int lpages;
723                 int level = i + 2;
724
725                 /* Avoid unused variable warning if no large pages */
726                 (void)level;
727
728                 if (new.lpage_info[i])
729                         continue;
730
731                 lpages = 1 + ((base_gfn + npages - 1)
732                              >> KVM_HPAGE_GFN_SHIFT(level));
733                 lpages -= base_gfn >> KVM_HPAGE_GFN_SHIFT(level);
734
735                 new.lpage_info[i] = vzalloc(lpages * sizeof(*new.lpage_info[i]));
736
737                 if (!new.lpage_info[i])
738                         goto out_free;
739
740                 if (base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
741                         new.lpage_info[i][0].write_count = 1;
742                 if ((base_gfn+npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
743                         new.lpage_info[i][lpages - 1].write_count = 1;
744                 ugfn = new.userspace_addr >> PAGE_SHIFT;
745                 /*
746                  * If the gfn and userspace address are not aligned wrt each
747                  * other, or if explicitly asked to, disable large page
748                  * support for this slot
749                  */
750                 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
751                     !largepages_enabled)
752                         for (j = 0; j < lpages; ++j)
753                                 new.lpage_info[i][j].write_count = 1;
754         }
755
756 skip_lpage:
757
758         /* Allocate page dirty bitmap if needed */
759         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
760                 if (kvm_create_dirty_bitmap(&new) < 0)
761                         goto out_free;
762                 /* destroy any largepage mappings for dirty tracking */
763         }
764 #else  /* not defined CONFIG_S390 */
765         new.user_alloc = user_alloc;
766         if (user_alloc)
767                 new.userspace_addr = mem->userspace_addr;
768 #endif /* not defined CONFIG_S390 */
769
770         if (!npages) {
771                 r = -ENOMEM;
772                 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
773                 if (!slots)
774                         goto out_free;
775                 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
776                 if (mem->slot >= slots->nmemslots)
777                         slots->nmemslots = mem->slot + 1;
778                 slots->generation++;
779                 slots->memslots[mem->slot].flags |= KVM_MEMSLOT_INVALID;
780
781                 old_memslots = kvm->memslots;
782                 rcu_assign_pointer(kvm->memslots, slots);
783                 synchronize_srcu_expedited(&kvm->srcu);
784                 /* From this point no new shadow pages pointing to a deleted
785                  * memslot will be created.
786                  *
787                  * validation of sp->gfn happens in:
788                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
789                  *      - kvm_is_visible_gfn (mmu_check_roots)
790                  */
791                 kvm_arch_flush_shadow(kvm);
792                 kfree(old_memslots);
793         }
794
795         r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
796         if (r)
797                 goto out_free;
798
799         /* map the pages in iommu page table */
800         if (npages) {
801                 r = kvm_iommu_map_pages(kvm, &new);
802                 if (r)
803                         goto out_free;
804         }
805
806         r = -ENOMEM;
807         slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
808         if (!slots)
809                 goto out_free;
810         memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
811         if (mem->slot >= slots->nmemslots)
812                 slots->nmemslots = mem->slot + 1;
813         slots->generation++;
814
815         /* actual memory is freed via old in kvm_free_physmem_slot below */
816         if (!npages) {
817                 new.rmap = NULL;
818                 new.dirty_bitmap = NULL;
819                 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i)
820                         new.lpage_info[i] = NULL;
821         }
822
823         slots->memslots[mem->slot] = new;
824         old_memslots = kvm->memslots;
825         rcu_assign_pointer(kvm->memslots, slots);
826         synchronize_srcu_expedited(&kvm->srcu);
827
828         kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
829
830         kvm_free_physmem_slot(&old, &new);
831         kfree(old_memslots);
832
833         return 0;
834
835 out_free:
836         kvm_free_physmem_slot(&new, &old);
837 out:
838         return r;
839
840 }
841 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
842
843 int kvm_set_memory_region(struct kvm *kvm,
844                           struct kvm_userspace_memory_region *mem,
845                           int user_alloc)
846 {
847         int r;
848
849         mutex_lock(&kvm->slots_lock);
850         r = __kvm_set_memory_region(kvm, mem, user_alloc);
851         mutex_unlock(&kvm->slots_lock);
852         return r;
853 }
854 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
855
856 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
857                                    struct
858                                    kvm_userspace_memory_region *mem,
859                                    int user_alloc)
860 {
861         if (mem->slot >= KVM_MEMORY_SLOTS)
862                 return -EINVAL;
863         return kvm_set_memory_region(kvm, mem, user_alloc);
864 }
865
866 int kvm_get_dirty_log(struct kvm *kvm,
867                         struct kvm_dirty_log *log, int *is_dirty)
868 {
869         struct kvm_memory_slot *memslot;
870         int r, i;
871         unsigned long n;
872         unsigned long any = 0;
873
874         r = -EINVAL;
875         if (log->slot >= KVM_MEMORY_SLOTS)
876                 goto out;
877
878         memslot = &kvm->memslots->memslots[log->slot];
879         r = -ENOENT;
880         if (!memslot->dirty_bitmap)
881                 goto out;
882
883         n = kvm_dirty_bitmap_bytes(memslot);
884
885         for (i = 0; !any && i < n/sizeof(long); ++i)
886                 any = memslot->dirty_bitmap[i];
887
888         r = -EFAULT;
889         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
890                 goto out;
891
892         if (any)
893                 *is_dirty = 1;
894
895         r = 0;
896 out:
897         return r;
898 }
899
900 void kvm_disable_largepages(void)
901 {
902         largepages_enabled = false;
903 }
904 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
905
906 int is_error_page(struct page *page)
907 {
908         return page == bad_page || page == hwpoison_page || page == fault_page;
909 }
910 EXPORT_SYMBOL_GPL(is_error_page);
911
912 int is_error_pfn(pfn_t pfn)
913 {
914         return pfn == bad_pfn || pfn == hwpoison_pfn || pfn == fault_pfn;
915 }
916 EXPORT_SYMBOL_GPL(is_error_pfn);
917
918 int is_hwpoison_pfn(pfn_t pfn)
919 {
920         return pfn == hwpoison_pfn;
921 }
922 EXPORT_SYMBOL_GPL(is_hwpoison_pfn);
923
924 int is_fault_pfn(pfn_t pfn)
925 {
926         return pfn == fault_pfn;
927 }
928 EXPORT_SYMBOL_GPL(is_fault_pfn);
929
930 static inline unsigned long bad_hva(void)
931 {
932         return PAGE_OFFSET;
933 }
934
935 int kvm_is_error_hva(unsigned long addr)
936 {
937         return addr == bad_hva();
938 }
939 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
940
941 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm_memslots *slots,
942                                                 gfn_t gfn)
943 {
944         int i;
945
946         for (i = 0; i < slots->nmemslots; ++i) {
947                 struct kvm_memory_slot *memslot = &slots->memslots[i];
948
949                 if (gfn >= memslot->base_gfn
950                     && gfn < memslot->base_gfn + memslot->npages)
951                         return memslot;
952         }
953         return NULL;
954 }
955
956 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
957 {
958         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
959 }
960 EXPORT_SYMBOL_GPL(gfn_to_memslot);
961
962 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
963 {
964         int i;
965         struct kvm_memslots *slots = kvm_memslots(kvm);
966
967         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
968                 struct kvm_memory_slot *memslot = &slots->memslots[i];
969
970                 if (memslot->flags & KVM_MEMSLOT_INVALID)
971                         continue;
972
973                 if (gfn >= memslot->base_gfn
974                     && gfn < memslot->base_gfn + memslot->npages)
975                         return 1;
976         }
977         return 0;
978 }
979 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
980
981 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
982 {
983         struct vm_area_struct *vma;
984         unsigned long addr, size;
985
986         size = PAGE_SIZE;
987
988         addr = gfn_to_hva(kvm, gfn);
989         if (kvm_is_error_hva(addr))
990                 return PAGE_SIZE;
991
992         down_read(&current->mm->mmap_sem);
993         vma = find_vma(current->mm, addr);
994         if (!vma)
995                 goto out;
996
997         size = vma_kernel_pagesize(vma);
998
999 out:
1000         up_read(&current->mm->mmap_sem);
1001
1002         return size;
1003 }
1004
1005 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1006                                      gfn_t *nr_pages)
1007 {
1008         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1009                 return bad_hva();
1010
1011         if (nr_pages)
1012                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1013
1014         return gfn_to_hva_memslot(slot, gfn);
1015 }
1016
1017 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1018 {
1019         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1020 }
1021 EXPORT_SYMBOL_GPL(gfn_to_hva);
1022
1023 static pfn_t get_fault_pfn(void)
1024 {
1025         get_page(fault_page);
1026         return fault_pfn;
1027 }
1028
1029 int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1030         unsigned long start, int write, struct page **page)
1031 {
1032         int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1033
1034         if (write)
1035                 flags |= FOLL_WRITE;
1036
1037         return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1038 }
1039
1040 static inline int check_user_page_hwpoison(unsigned long addr)
1041 {
1042         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1043
1044         rc = __get_user_pages(current, current->mm, addr, 1,
1045                               flags, NULL, NULL, NULL);
1046         return rc == -EHWPOISON;
1047 }
1048
1049 static pfn_t hva_to_pfn(struct kvm *kvm, unsigned long addr, bool atomic,
1050                         bool *async, bool write_fault, bool *writable)
1051 {
1052         struct page *page[1];
1053         int npages = 0;
1054         pfn_t pfn;
1055
1056         /* we can do it either atomically or asynchronously, not both */
1057         BUG_ON(atomic && async);
1058
1059         BUG_ON(!write_fault && !writable);
1060
1061         if (writable)
1062                 *writable = true;
1063
1064         if (atomic || async)
1065                 npages = __get_user_pages_fast(addr, 1, 1, page);
1066
1067         if (unlikely(npages != 1) && !atomic) {
1068                 might_sleep();
1069
1070                 if (writable)
1071                         *writable = write_fault;
1072
1073                 if (async) {
1074                         down_read(&current->mm->mmap_sem);
1075                         npages = get_user_page_nowait(current, current->mm,
1076                                                      addr, write_fault, page);
1077                         up_read(&current->mm->mmap_sem);
1078                 } else
1079                         npages = get_user_pages_fast(addr, 1, write_fault,
1080                                                      page);
1081
1082                 /* map read fault as writable if possible */
1083                 if (unlikely(!write_fault) && npages == 1) {
1084                         struct page *wpage[1];
1085
1086                         npages = __get_user_pages_fast(addr, 1, 1, wpage);
1087                         if (npages == 1) {
1088                                 *writable = true;
1089                                 put_page(page[0]);
1090                                 page[0] = wpage[0];
1091                         }
1092                         npages = 1;
1093                 }
1094         }
1095
1096         if (unlikely(npages != 1)) {
1097                 struct vm_area_struct *vma;
1098
1099                 if (atomic)
1100                         return get_fault_pfn();
1101
1102                 down_read(&current->mm->mmap_sem);
1103                 if (npages == -EHWPOISON ||
1104                         (!async && check_user_page_hwpoison(addr))) {
1105                         up_read(&current->mm->mmap_sem);
1106                         get_page(hwpoison_page);
1107                         return page_to_pfn(hwpoison_page);
1108                 }
1109
1110                 vma = find_vma_intersection(current->mm, addr, addr+1);
1111
1112                 if (vma == NULL)
1113                         pfn = get_fault_pfn();
1114                 else if ((vma->vm_flags & VM_PFNMAP)) {
1115                         pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1116                                 vma->vm_pgoff;
1117                         BUG_ON(!kvm_is_mmio_pfn(pfn));
1118                 } else {
1119                         if (async && (vma->vm_flags & VM_WRITE))
1120                                 *async = true;
1121                         pfn = get_fault_pfn();
1122                 }
1123                 up_read(&current->mm->mmap_sem);
1124         } else
1125                 pfn = page_to_pfn(page[0]);
1126
1127         return pfn;
1128 }
1129
1130 pfn_t hva_to_pfn_atomic(struct kvm *kvm, unsigned long addr)
1131 {
1132         return hva_to_pfn(kvm, addr, true, NULL, true, NULL);
1133 }
1134 EXPORT_SYMBOL_GPL(hva_to_pfn_atomic);
1135
1136 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1137                           bool write_fault, bool *writable)
1138 {
1139         unsigned long addr;
1140
1141         if (async)
1142                 *async = false;
1143
1144         addr = gfn_to_hva(kvm, gfn);
1145         if (kvm_is_error_hva(addr)) {
1146                 get_page(bad_page);
1147                 return page_to_pfn(bad_page);
1148         }
1149
1150         return hva_to_pfn(kvm, addr, atomic, async, write_fault, writable);
1151 }
1152
1153 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1154 {
1155         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1156 }
1157 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1158
1159 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1160                        bool write_fault, bool *writable)
1161 {
1162         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1163 }
1164 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1165
1166 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1167 {
1168         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1169 }
1170 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1171
1172 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1173                       bool *writable)
1174 {
1175         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1176 }
1177 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1178
1179 pfn_t gfn_to_pfn_memslot(struct kvm *kvm,
1180                          struct kvm_memory_slot *slot, gfn_t gfn)
1181 {
1182         unsigned long addr = gfn_to_hva_memslot(slot, gfn);
1183         return hva_to_pfn(kvm, addr, false, NULL, true, NULL);
1184 }
1185
1186 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1187                                                                   int nr_pages)
1188 {
1189         unsigned long addr;
1190         gfn_t entry;
1191
1192         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1193         if (kvm_is_error_hva(addr))
1194                 return -1;
1195
1196         if (entry < nr_pages)
1197                 return 0;
1198
1199         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1200 }
1201 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1202
1203 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1204 {
1205         pfn_t pfn;
1206
1207         pfn = gfn_to_pfn(kvm, gfn);
1208         if (!kvm_is_mmio_pfn(pfn))
1209                 return pfn_to_page(pfn);
1210
1211         WARN_ON(kvm_is_mmio_pfn(pfn));
1212
1213         get_page(bad_page);
1214         return bad_page;
1215 }
1216
1217 EXPORT_SYMBOL_GPL(gfn_to_page);
1218
1219 void kvm_release_page_clean(struct page *page)
1220 {
1221         kvm_release_pfn_clean(page_to_pfn(page));
1222 }
1223 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1224
1225 void kvm_release_pfn_clean(pfn_t pfn)
1226 {
1227         if (!kvm_is_mmio_pfn(pfn))
1228                 put_page(pfn_to_page(pfn));
1229 }
1230 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1231
1232 void kvm_release_page_dirty(struct page *page)
1233 {
1234         kvm_release_pfn_dirty(page_to_pfn(page));
1235 }
1236 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1237
1238 void kvm_release_pfn_dirty(pfn_t pfn)
1239 {
1240         kvm_set_pfn_dirty(pfn);
1241         kvm_release_pfn_clean(pfn);
1242 }
1243 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1244
1245 void kvm_set_page_dirty(struct page *page)
1246 {
1247         kvm_set_pfn_dirty(page_to_pfn(page));
1248 }
1249 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1250
1251 void kvm_set_pfn_dirty(pfn_t pfn)
1252 {
1253         if (!kvm_is_mmio_pfn(pfn)) {
1254                 struct page *page = pfn_to_page(pfn);
1255                 if (!PageReserved(page))
1256                         SetPageDirty(page);
1257         }
1258 }
1259 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1260
1261 void kvm_set_pfn_accessed(pfn_t pfn)
1262 {
1263         if (!kvm_is_mmio_pfn(pfn))
1264                 mark_page_accessed(pfn_to_page(pfn));
1265 }
1266 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1267
1268 void kvm_get_pfn(pfn_t pfn)
1269 {
1270         if (!kvm_is_mmio_pfn(pfn))
1271                 get_page(pfn_to_page(pfn));
1272 }
1273 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1274
1275 static int next_segment(unsigned long len, int offset)
1276 {
1277         if (len > PAGE_SIZE - offset)
1278                 return PAGE_SIZE - offset;
1279         else
1280                 return len;
1281 }
1282
1283 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1284                         int len)
1285 {
1286         int r;
1287         unsigned long addr;
1288
1289         addr = gfn_to_hva(kvm, gfn);
1290         if (kvm_is_error_hva(addr))
1291                 return -EFAULT;
1292         r = __copy_from_user(data, (void __user *)addr + offset, len);
1293         if (r)
1294                 return -EFAULT;
1295         return 0;
1296 }
1297 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1298
1299 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1300 {
1301         gfn_t gfn = gpa >> PAGE_SHIFT;
1302         int seg;
1303         int offset = offset_in_page(gpa);
1304         int ret;
1305
1306         while ((seg = next_segment(len, offset)) != 0) {
1307                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1308                 if (ret < 0)
1309                         return ret;
1310                 offset = 0;
1311                 len -= seg;
1312                 data += seg;
1313                 ++gfn;
1314         }
1315         return 0;
1316 }
1317 EXPORT_SYMBOL_GPL(kvm_read_guest);
1318
1319 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1320                           unsigned long len)
1321 {
1322         int r;
1323         unsigned long addr;
1324         gfn_t gfn = gpa >> PAGE_SHIFT;
1325         int offset = offset_in_page(gpa);
1326
1327         addr = gfn_to_hva(kvm, gfn);
1328         if (kvm_is_error_hva(addr))
1329                 return -EFAULT;
1330         pagefault_disable();
1331         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1332         pagefault_enable();
1333         if (r)
1334                 return -EFAULT;
1335         return 0;
1336 }
1337 EXPORT_SYMBOL(kvm_read_guest_atomic);
1338
1339 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1340                          int offset, int len)
1341 {
1342         int r;
1343         unsigned long addr;
1344
1345         addr = gfn_to_hva(kvm, gfn);
1346         if (kvm_is_error_hva(addr))
1347                 return -EFAULT;
1348         r = copy_to_user((void __user *)addr + offset, data, len);
1349         if (r)
1350                 return -EFAULT;
1351         mark_page_dirty(kvm, gfn);
1352         return 0;
1353 }
1354 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1355
1356 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1357                     unsigned long len)
1358 {
1359         gfn_t gfn = gpa >> PAGE_SHIFT;
1360         int seg;
1361         int offset = offset_in_page(gpa);
1362         int ret;
1363
1364         while ((seg = next_segment(len, offset)) != 0) {
1365                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1366                 if (ret < 0)
1367                         return ret;
1368                 offset = 0;
1369                 len -= seg;
1370                 data += seg;
1371                 ++gfn;
1372         }
1373         return 0;
1374 }
1375
1376 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1377                               gpa_t gpa)
1378 {
1379         struct kvm_memslots *slots = kvm_memslots(kvm);
1380         int offset = offset_in_page(gpa);
1381         gfn_t gfn = gpa >> PAGE_SHIFT;
1382
1383         ghc->gpa = gpa;
1384         ghc->generation = slots->generation;
1385         ghc->memslot = __gfn_to_memslot(slots, gfn);
1386         ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1387         if (!kvm_is_error_hva(ghc->hva))
1388                 ghc->hva += offset;
1389         else
1390                 return -EFAULT;
1391
1392         return 0;
1393 }
1394 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1395
1396 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1397                            void *data, unsigned long len)
1398 {
1399         struct kvm_memslots *slots = kvm_memslots(kvm);
1400         int r;
1401
1402         if (slots->generation != ghc->generation)
1403                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1404
1405         if (kvm_is_error_hva(ghc->hva))
1406                 return -EFAULT;
1407
1408         r = copy_to_user((void __user *)ghc->hva, data, len);
1409         if (r)
1410                 return -EFAULT;
1411         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1412
1413         return 0;
1414 }
1415 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1416
1417 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1418 {
1419         return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1420                                     offset, len);
1421 }
1422 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1423
1424 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1425 {
1426         gfn_t gfn = gpa >> PAGE_SHIFT;
1427         int seg;
1428         int offset = offset_in_page(gpa);
1429         int ret;
1430
1431         while ((seg = next_segment(len, offset)) != 0) {
1432                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1433                 if (ret < 0)
1434                         return ret;
1435                 offset = 0;
1436                 len -= seg;
1437                 ++gfn;
1438         }
1439         return 0;
1440 }
1441 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1442
1443 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1444                              gfn_t gfn)
1445 {
1446         if (memslot && memslot->dirty_bitmap) {
1447                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1448
1449                 __set_bit_le(rel_gfn, memslot->dirty_bitmap);
1450         }
1451 }
1452
1453 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1454 {
1455         struct kvm_memory_slot *memslot;
1456
1457         memslot = gfn_to_memslot(kvm, gfn);
1458         mark_page_dirty_in_slot(kvm, memslot, gfn);
1459 }
1460
1461 /*
1462  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1463  */
1464 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1465 {
1466         DEFINE_WAIT(wait);
1467
1468         for (;;) {
1469                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1470
1471                 if (kvm_arch_vcpu_runnable(vcpu)) {
1472                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1473                         break;
1474                 }
1475                 if (kvm_cpu_has_pending_timer(vcpu))
1476                         break;
1477                 if (signal_pending(current))
1478                         break;
1479
1480                 schedule();
1481         }
1482
1483         finish_wait(&vcpu->wq, &wait);
1484 }
1485
1486 void kvm_resched(struct kvm_vcpu *vcpu)
1487 {
1488         if (!need_resched())
1489                 return;
1490         cond_resched();
1491 }
1492 EXPORT_SYMBOL_GPL(kvm_resched);
1493
1494 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1495 {
1496         struct kvm *kvm = me->kvm;
1497         struct kvm_vcpu *vcpu;
1498         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1499         int yielded = 0;
1500         int pass;
1501         int i;
1502
1503         /*
1504          * We boost the priority of a VCPU that is runnable but not
1505          * currently running, because it got preempted by something
1506          * else and called schedule in __vcpu_run.  Hopefully that
1507          * VCPU is holding the lock that we need and will release it.
1508          * We approximate round-robin by starting at the last boosted VCPU.
1509          */
1510         for (pass = 0; pass < 2 && !yielded; pass++) {
1511                 kvm_for_each_vcpu(i, vcpu, kvm) {
1512                         struct task_struct *task = NULL;
1513                         struct pid *pid;
1514                         if (!pass && i < last_boosted_vcpu) {
1515                                 i = last_boosted_vcpu;
1516                                 continue;
1517                         } else if (pass && i > last_boosted_vcpu)
1518                                 break;
1519                         if (vcpu == me)
1520                                 continue;
1521                         if (waitqueue_active(&vcpu->wq))
1522                                 continue;
1523                         rcu_read_lock();
1524                         pid = rcu_dereference(vcpu->pid);
1525                         if (pid)
1526                                 task = get_pid_task(vcpu->pid, PIDTYPE_PID);
1527                         rcu_read_unlock();
1528                         if (!task)
1529                                 continue;
1530                         if (task->flags & PF_VCPU) {
1531                                 put_task_struct(task);
1532                                 continue;
1533                         }
1534                         if (yield_to(task, 1)) {
1535                                 put_task_struct(task);
1536                                 kvm->last_boosted_vcpu = i;
1537                                 yielded = 1;
1538                                 break;
1539                         }
1540                         put_task_struct(task);
1541                 }
1542         }
1543 }
1544 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1545
1546 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1547 {
1548         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1549         struct page *page;
1550
1551         if (vmf->pgoff == 0)
1552                 page = virt_to_page(vcpu->run);
1553 #ifdef CONFIG_X86
1554         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1555                 page = virt_to_page(vcpu->arch.pio_data);
1556 #endif
1557 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1558         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1559                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1560 #endif
1561         else
1562                 return VM_FAULT_SIGBUS;
1563         get_page(page);
1564         vmf->page = page;
1565         return 0;
1566 }
1567
1568 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1569         .fault = kvm_vcpu_fault,
1570 };
1571
1572 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1573 {
1574         vma->vm_ops = &kvm_vcpu_vm_ops;
1575         return 0;
1576 }
1577
1578 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1579 {
1580         struct kvm_vcpu *vcpu = filp->private_data;
1581
1582         kvm_put_kvm(vcpu->kvm);
1583         return 0;
1584 }
1585
1586 static struct file_operations kvm_vcpu_fops = {
1587         .release        = kvm_vcpu_release,
1588         .unlocked_ioctl = kvm_vcpu_ioctl,
1589         .compat_ioctl   = kvm_vcpu_ioctl,
1590         .mmap           = kvm_vcpu_mmap,
1591         .llseek         = noop_llseek,
1592 };
1593
1594 /*
1595  * Allocates an inode for the vcpu.
1596  */
1597 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1598 {
1599         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1600 }
1601
1602 /*
1603  * Creates some virtual cpus.  Good luck creating more than one.
1604  */
1605 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1606 {
1607         int r;
1608         struct kvm_vcpu *vcpu, *v;
1609
1610         vcpu = kvm_arch_vcpu_create(kvm, id);
1611         if (IS_ERR(vcpu))
1612                 return PTR_ERR(vcpu);
1613
1614         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1615
1616         r = kvm_arch_vcpu_setup(vcpu);
1617         if (r)
1618                 return r;
1619
1620         mutex_lock(&kvm->lock);
1621         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1622                 r = -EINVAL;
1623                 goto vcpu_destroy;
1624         }
1625
1626         kvm_for_each_vcpu(r, v, kvm)
1627                 if (v->vcpu_id == id) {
1628                         r = -EEXIST;
1629                         goto vcpu_destroy;
1630                 }
1631
1632         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1633
1634         /* Now it's all set up, let userspace reach it */
1635         kvm_get_kvm(kvm);
1636         r = create_vcpu_fd(vcpu);
1637         if (r < 0) {
1638                 kvm_put_kvm(kvm);
1639                 goto vcpu_destroy;
1640         }
1641
1642         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1643         smp_wmb();
1644         atomic_inc(&kvm->online_vcpus);
1645
1646 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1647         if (kvm->bsp_vcpu_id == id)
1648                 kvm->bsp_vcpu = vcpu;
1649 #endif
1650         mutex_unlock(&kvm->lock);
1651         return r;
1652
1653 vcpu_destroy:
1654         mutex_unlock(&kvm->lock);
1655         kvm_arch_vcpu_destroy(vcpu);
1656         return r;
1657 }
1658
1659 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1660 {
1661         if (sigset) {
1662                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1663                 vcpu->sigset_active = 1;
1664                 vcpu->sigset = *sigset;
1665         } else
1666                 vcpu->sigset_active = 0;
1667         return 0;
1668 }
1669
1670 static long kvm_vcpu_ioctl(struct file *filp,
1671                            unsigned int ioctl, unsigned long arg)
1672 {
1673         struct kvm_vcpu *vcpu = filp->private_data;
1674         void __user *argp = (void __user *)arg;
1675         int r;
1676         struct kvm_fpu *fpu = NULL;
1677         struct kvm_sregs *kvm_sregs = NULL;
1678
1679         if (vcpu->kvm->mm != current->mm)
1680                 return -EIO;
1681
1682 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1683         /*
1684          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1685          * so vcpu_load() would break it.
1686          */
1687         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1688                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1689 #endif
1690
1691
1692         vcpu_load(vcpu);
1693         switch (ioctl) {
1694         case KVM_RUN:
1695                 r = -EINVAL;
1696                 if (arg)
1697                         goto out;
1698                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1699                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1700                 break;
1701         case KVM_GET_REGS: {
1702                 struct kvm_regs *kvm_regs;
1703
1704                 r = -ENOMEM;
1705                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1706                 if (!kvm_regs)
1707                         goto out;
1708                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1709                 if (r)
1710                         goto out_free1;
1711                 r = -EFAULT;
1712                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1713                         goto out_free1;
1714                 r = 0;
1715 out_free1:
1716                 kfree(kvm_regs);
1717                 break;
1718         }
1719         case KVM_SET_REGS: {
1720                 struct kvm_regs *kvm_regs;
1721
1722                 r = -ENOMEM;
1723                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1724                 if (!kvm_regs)
1725                         goto out;
1726                 r = -EFAULT;
1727                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1728                         goto out_free2;
1729                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1730                 if (r)
1731                         goto out_free2;
1732                 r = 0;
1733 out_free2:
1734                 kfree(kvm_regs);
1735                 break;
1736         }
1737         case KVM_GET_SREGS: {
1738                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1739                 r = -ENOMEM;
1740                 if (!kvm_sregs)
1741                         goto out;
1742                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1743                 if (r)
1744                         goto out;
1745                 r = -EFAULT;
1746                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1747                         goto out;
1748                 r = 0;
1749                 break;
1750         }
1751         case KVM_SET_SREGS: {
1752                 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1753                 r = -ENOMEM;
1754                 if (!kvm_sregs)
1755                         goto out;
1756                 r = -EFAULT;
1757                 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1758                         goto out;
1759                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1760                 if (r)
1761                         goto out;
1762                 r = 0;
1763                 break;
1764         }
1765         case KVM_GET_MP_STATE: {
1766                 struct kvm_mp_state mp_state;
1767
1768                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1769                 if (r)
1770                         goto out;
1771                 r = -EFAULT;
1772                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1773                         goto out;
1774                 r = 0;
1775                 break;
1776         }
1777         case KVM_SET_MP_STATE: {
1778                 struct kvm_mp_state mp_state;
1779
1780                 r = -EFAULT;
1781                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1782                         goto out;
1783                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1784                 if (r)
1785                         goto out;
1786                 r = 0;
1787                 break;
1788         }
1789         case KVM_TRANSLATE: {
1790                 struct kvm_translation tr;
1791
1792                 r = -EFAULT;
1793                 if (copy_from_user(&tr, argp, sizeof tr))
1794                         goto out;
1795                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1796                 if (r)
1797                         goto out;
1798                 r = -EFAULT;
1799                 if (copy_to_user(argp, &tr, sizeof tr))
1800                         goto out;
1801                 r = 0;
1802                 break;
1803         }
1804         case KVM_SET_GUEST_DEBUG: {
1805                 struct kvm_guest_debug dbg;
1806
1807                 r = -EFAULT;
1808                 if (copy_from_user(&dbg, argp, sizeof dbg))
1809                         goto out;
1810                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1811                 if (r)
1812                         goto out;
1813                 r = 0;
1814                 break;
1815         }
1816         case KVM_SET_SIGNAL_MASK: {
1817                 struct kvm_signal_mask __user *sigmask_arg = argp;
1818                 struct kvm_signal_mask kvm_sigmask;
1819                 sigset_t sigset, *p;
1820
1821                 p = NULL;
1822                 if (argp) {
1823                         r = -EFAULT;
1824                         if (copy_from_user(&kvm_sigmask, argp,
1825                                            sizeof kvm_sigmask))
1826                                 goto out;
1827                         r = -EINVAL;
1828                         if (kvm_sigmask.len != sizeof sigset)
1829                                 goto out;
1830                         r = -EFAULT;
1831                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1832                                            sizeof sigset))
1833                                 goto out;
1834                         p = &sigset;
1835                 }
1836                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
1837                 break;
1838         }
1839         case KVM_GET_FPU: {
1840                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1841                 r = -ENOMEM;
1842                 if (!fpu)
1843                         goto out;
1844                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1845                 if (r)
1846                         goto out;
1847                 r = -EFAULT;
1848                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1849                         goto out;
1850                 r = 0;
1851                 break;
1852         }
1853         case KVM_SET_FPU: {
1854                 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1855                 r = -ENOMEM;
1856                 if (!fpu)
1857                         goto out;
1858                 r = -EFAULT;
1859                 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1860                         goto out;
1861                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1862                 if (r)
1863                         goto out;
1864                 r = 0;
1865                 break;
1866         }
1867         default:
1868                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1869         }
1870 out:
1871         vcpu_put(vcpu);
1872         kfree(fpu);
1873         kfree(kvm_sregs);
1874         return r;
1875 }
1876
1877 static long kvm_vm_ioctl(struct file *filp,
1878                            unsigned int ioctl, unsigned long arg)
1879 {
1880         struct kvm *kvm = filp->private_data;
1881         void __user *argp = (void __user *)arg;
1882         int r;
1883
1884         if (kvm->mm != current->mm)
1885                 return -EIO;
1886         switch (ioctl) {
1887         case KVM_CREATE_VCPU:
1888                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1889                 if (r < 0)
1890                         goto out;
1891                 break;
1892         case KVM_SET_USER_MEMORY_REGION: {
1893                 struct kvm_userspace_memory_region kvm_userspace_mem;
1894
1895                 r = -EFAULT;
1896                 if (copy_from_user(&kvm_userspace_mem, argp,
1897                                                 sizeof kvm_userspace_mem))
1898                         goto out;
1899
1900                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1901                 if (r)
1902                         goto out;
1903                 break;
1904         }
1905         case KVM_GET_DIRTY_LOG: {
1906                 struct kvm_dirty_log log;
1907
1908                 r = -EFAULT;
1909                 if (copy_from_user(&log, argp, sizeof log))
1910                         goto out;
1911                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1912                 if (r)
1913                         goto out;
1914                 break;
1915         }
1916 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1917         case KVM_REGISTER_COALESCED_MMIO: {
1918                 struct kvm_coalesced_mmio_zone zone;
1919                 r = -EFAULT;
1920                 if (copy_from_user(&zone, argp, sizeof zone))
1921                         goto out;
1922                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1923                 if (r)
1924                         goto out;
1925                 r = 0;
1926                 break;
1927         }
1928         case KVM_UNREGISTER_COALESCED_MMIO: {
1929                 struct kvm_coalesced_mmio_zone zone;
1930                 r = -EFAULT;
1931                 if (copy_from_user(&zone, argp, sizeof zone))
1932                         goto out;
1933                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1934                 if (r)
1935                         goto out;
1936                 r = 0;
1937                 break;
1938         }
1939 #endif
1940         case KVM_IRQFD: {
1941                 struct kvm_irqfd data;
1942
1943                 r = -EFAULT;
1944                 if (copy_from_user(&data, argp, sizeof data))
1945                         goto out;
1946                 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
1947                 break;
1948         }
1949         case KVM_IOEVENTFD: {
1950                 struct kvm_ioeventfd data;
1951
1952                 r = -EFAULT;
1953                 if (copy_from_user(&data, argp, sizeof data))
1954                         goto out;
1955                 r = kvm_ioeventfd(kvm, &data);
1956                 break;
1957         }
1958 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1959         case KVM_SET_BOOT_CPU_ID:
1960                 r = 0;
1961                 mutex_lock(&kvm->lock);
1962                 if (atomic_read(&kvm->online_vcpus) != 0)
1963                         r = -EBUSY;
1964                 else
1965                         kvm->bsp_vcpu_id = arg;
1966                 mutex_unlock(&kvm->lock);
1967                 break;
1968 #endif
1969         default:
1970                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1971                 if (r == -ENOTTY)
1972                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
1973         }
1974 out:
1975         return r;
1976 }
1977
1978 #ifdef CONFIG_COMPAT
1979 struct compat_kvm_dirty_log {
1980         __u32 slot;
1981         __u32 padding1;
1982         union {
1983                 compat_uptr_t dirty_bitmap; /* one bit per page */
1984                 __u64 padding2;
1985         };
1986 };
1987
1988 static long kvm_vm_compat_ioctl(struct file *filp,
1989                            unsigned int ioctl, unsigned long arg)
1990 {
1991         struct kvm *kvm = filp->private_data;
1992         int r;
1993
1994         if (kvm->mm != current->mm)
1995                 return -EIO;
1996         switch (ioctl) {
1997         case KVM_GET_DIRTY_LOG: {
1998                 struct compat_kvm_dirty_log compat_log;
1999                 struct kvm_dirty_log log;
2000
2001                 r = -EFAULT;
2002                 if (copy_from_user(&compat_log, (void __user *)arg,
2003                                    sizeof(compat_log)))
2004                         goto out;
2005                 log.slot         = compat_log.slot;
2006                 log.padding1     = compat_log.padding1;
2007                 log.padding2     = compat_log.padding2;
2008                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2009
2010                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2011                 if (r)
2012                         goto out;
2013                 break;
2014         }
2015         default:
2016                 r = kvm_vm_ioctl(filp, ioctl, arg);
2017         }
2018
2019 out:
2020         return r;
2021 }
2022 #endif
2023
2024 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2025 {
2026         struct page *page[1];
2027         unsigned long addr;
2028         int npages;
2029         gfn_t gfn = vmf->pgoff;
2030         struct kvm *kvm = vma->vm_file->private_data;
2031
2032         addr = gfn_to_hva(kvm, gfn);
2033         if (kvm_is_error_hva(addr))
2034                 return VM_FAULT_SIGBUS;
2035
2036         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2037                                 NULL);
2038         if (unlikely(npages != 1))
2039                 return VM_FAULT_SIGBUS;
2040
2041         vmf->page = page[0];
2042         return 0;
2043 }
2044
2045 static const struct vm_operations_struct kvm_vm_vm_ops = {
2046         .fault = kvm_vm_fault,
2047 };
2048
2049 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2050 {
2051         vma->vm_ops = &kvm_vm_vm_ops;
2052         return 0;
2053 }
2054
2055 static struct file_operations kvm_vm_fops = {
2056         .release        = kvm_vm_release,
2057         .unlocked_ioctl = kvm_vm_ioctl,
2058 #ifdef CONFIG_COMPAT
2059         .compat_ioctl   = kvm_vm_compat_ioctl,
2060 #endif
2061         .mmap           = kvm_vm_mmap,
2062         .llseek         = noop_llseek,
2063 };
2064
2065 static int kvm_dev_ioctl_create_vm(void)
2066 {
2067         int r;
2068         struct kvm *kvm;
2069
2070         kvm = kvm_create_vm();
2071         if (IS_ERR(kvm))
2072                 return PTR_ERR(kvm);
2073 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2074         r = kvm_coalesced_mmio_init(kvm);
2075         if (r < 0) {
2076                 kvm_put_kvm(kvm);
2077                 return r;
2078         }
2079 #endif
2080         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2081         if (r < 0)
2082                 kvm_put_kvm(kvm);
2083
2084         return r;
2085 }
2086
2087 static long kvm_dev_ioctl_check_extension_generic(long arg)
2088 {
2089         switch (arg) {
2090         case KVM_CAP_USER_MEMORY:
2091         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2092         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2093 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2094         case KVM_CAP_SET_BOOT_CPU_ID:
2095 #endif
2096         case KVM_CAP_INTERNAL_ERROR_DATA:
2097                 return 1;
2098 #ifdef CONFIG_HAVE_KVM_IRQCHIP
2099         case KVM_CAP_IRQ_ROUTING:
2100                 return KVM_MAX_IRQ_ROUTES;
2101 #endif
2102         default:
2103                 break;
2104         }
2105         return kvm_dev_ioctl_check_extension(arg);
2106 }
2107
2108 static long kvm_dev_ioctl(struct file *filp,
2109                           unsigned int ioctl, unsigned long arg)
2110 {
2111         long r = -EINVAL;
2112
2113         switch (ioctl) {
2114         case KVM_GET_API_VERSION:
2115                 r = -EINVAL;
2116                 if (arg)
2117                         goto out;
2118                 r = KVM_API_VERSION;
2119                 break;
2120         case KVM_CREATE_VM:
2121                 r = -EINVAL;
2122                 if (arg)
2123                         goto out;
2124                 r = kvm_dev_ioctl_create_vm();
2125                 break;
2126         case KVM_CHECK_EXTENSION:
2127                 r = kvm_dev_ioctl_check_extension_generic(arg);
2128                 break;
2129         case KVM_GET_VCPU_MMAP_SIZE:
2130                 r = -EINVAL;
2131                 if (arg)
2132                         goto out;
2133                 r = PAGE_SIZE;     /* struct kvm_run */
2134 #ifdef CONFIG_X86
2135                 r += PAGE_SIZE;    /* pio data page */
2136 #endif
2137 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2138                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2139 #endif
2140                 break;
2141         case KVM_TRACE_ENABLE:
2142         case KVM_TRACE_PAUSE:
2143         case KVM_TRACE_DISABLE:
2144                 r = -EOPNOTSUPP;
2145                 break;
2146         default:
2147                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2148         }
2149 out:
2150         return r;
2151 }
2152
2153 static struct file_operations kvm_chardev_ops = {
2154         .unlocked_ioctl = kvm_dev_ioctl,
2155         .compat_ioctl   = kvm_dev_ioctl,
2156         .llseek         = noop_llseek,
2157 };
2158
2159 static struct miscdevice kvm_dev = {
2160         KVM_MINOR,
2161         "kvm",
2162         &kvm_chardev_ops,
2163 };
2164
2165 static void hardware_enable_nolock(void *junk)
2166 {
2167         int cpu = raw_smp_processor_id();
2168         int r;
2169
2170         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2171                 return;
2172
2173         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2174
2175         r = kvm_arch_hardware_enable(NULL);
2176
2177         if (r) {
2178                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2179                 atomic_inc(&hardware_enable_failed);
2180                 printk(KERN_INFO "kvm: enabling virtualization on "
2181                                  "CPU%d failed\n", cpu);
2182         }
2183 }
2184
2185 static void hardware_enable(void *junk)
2186 {
2187         raw_spin_lock(&kvm_lock);
2188         hardware_enable_nolock(junk);
2189         raw_spin_unlock(&kvm_lock);
2190 }
2191
2192 static void hardware_disable_nolock(void *junk)
2193 {
2194         int cpu = raw_smp_processor_id();
2195
2196         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2197                 return;
2198         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2199         kvm_arch_hardware_disable(NULL);
2200 }
2201
2202 static void hardware_disable(void *junk)
2203 {
2204         raw_spin_lock(&kvm_lock);
2205         hardware_disable_nolock(junk);
2206         raw_spin_unlock(&kvm_lock);
2207 }
2208
2209 static void hardware_disable_all_nolock(void)
2210 {
2211         BUG_ON(!kvm_usage_count);
2212
2213         kvm_usage_count--;
2214         if (!kvm_usage_count)
2215                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2216 }
2217
2218 static void hardware_disable_all(void)
2219 {
2220         raw_spin_lock(&kvm_lock);
2221         hardware_disable_all_nolock();
2222         raw_spin_unlock(&kvm_lock);
2223 }
2224
2225 static int hardware_enable_all(void)
2226 {
2227         int r = 0;
2228
2229         raw_spin_lock(&kvm_lock);
2230
2231         kvm_usage_count++;
2232         if (kvm_usage_count == 1) {
2233                 atomic_set(&hardware_enable_failed, 0);
2234                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2235
2236                 if (atomic_read(&hardware_enable_failed)) {
2237                         hardware_disable_all_nolock();
2238                         r = -EBUSY;
2239                 }
2240         }
2241
2242         raw_spin_unlock(&kvm_lock);
2243
2244         return r;
2245 }
2246
2247 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2248                            void *v)
2249 {
2250         int cpu = (long)v;
2251
2252         if (!kvm_usage_count)
2253                 return NOTIFY_OK;
2254
2255         val &= ~CPU_TASKS_FROZEN;
2256         switch (val) {
2257         case CPU_DYING:
2258                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2259                        cpu);
2260                 hardware_disable(NULL);
2261                 break;
2262         case CPU_STARTING:
2263                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2264                        cpu);
2265                 hardware_enable(NULL);
2266                 break;
2267         }
2268         return NOTIFY_OK;
2269 }
2270
2271
2272 asmlinkage void kvm_spurious_fault(void)
2273 {
2274         /* Fault while not rebooting.  We want the trace. */
2275         BUG();
2276 }
2277 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
2278
2279 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2280                       void *v)
2281 {
2282         /*
2283          * Some (well, at least mine) BIOSes hang on reboot if
2284          * in vmx root mode.
2285          *
2286          * And Intel TXT required VMX off for all cpu when system shutdown.
2287          */
2288         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2289         kvm_rebooting = true;
2290         on_each_cpu(hardware_disable_nolock, NULL, 1);
2291         return NOTIFY_OK;
2292 }
2293
2294 static struct notifier_block kvm_reboot_notifier = {
2295         .notifier_call = kvm_reboot,
2296         .priority = 0,
2297 };
2298
2299 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2300 {
2301         int i;
2302
2303         for (i = 0; i < bus->dev_count; i++) {
2304                 struct kvm_io_device *pos = bus->devs[i];
2305
2306                 kvm_iodevice_destructor(pos);
2307         }
2308         kfree(bus);
2309 }
2310
2311 /* kvm_io_bus_write - called under kvm->slots_lock */
2312 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2313                      int len, const void *val)
2314 {
2315         int i;
2316         struct kvm_io_bus *bus;
2317
2318         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2319         for (i = 0; i < bus->dev_count; i++)
2320                 if (!kvm_iodevice_write(bus->devs[i], addr, len, val))
2321                         return 0;
2322         return -EOPNOTSUPP;
2323 }
2324
2325 /* kvm_io_bus_read - called under kvm->slots_lock */
2326 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2327                     int len, void *val)
2328 {
2329         int i;
2330         struct kvm_io_bus *bus;
2331
2332         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2333         for (i = 0; i < bus->dev_count; i++)
2334                 if (!kvm_iodevice_read(bus->devs[i], addr, len, val))
2335                         return 0;
2336         return -EOPNOTSUPP;
2337 }
2338
2339 /* Caller must hold slots_lock. */
2340 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2341                             struct kvm_io_device *dev)
2342 {
2343         struct kvm_io_bus *new_bus, *bus;
2344
2345         bus = kvm->buses[bus_idx];
2346         if (bus->dev_count > NR_IOBUS_DEVS-1)
2347                 return -ENOSPC;
2348
2349         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2350         if (!new_bus)
2351                 return -ENOMEM;
2352         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2353         new_bus->devs[new_bus->dev_count++] = dev;
2354         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2355         synchronize_srcu_expedited(&kvm->srcu);
2356         kfree(bus);
2357
2358         return 0;
2359 }
2360
2361 /* Caller must hold slots_lock. */
2362 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2363                               struct kvm_io_device *dev)
2364 {
2365         int i, r;
2366         struct kvm_io_bus *new_bus, *bus;
2367
2368         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2369         if (!new_bus)
2370                 return -ENOMEM;
2371
2372         bus = kvm->buses[bus_idx];
2373         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2374
2375         r = -ENOENT;
2376         for (i = 0; i < new_bus->dev_count; i++)
2377                 if (new_bus->devs[i] == dev) {
2378                         r = 0;
2379                         new_bus->devs[i] = new_bus->devs[--new_bus->dev_count];
2380                         break;
2381                 }
2382
2383         if (r) {
2384                 kfree(new_bus);
2385                 return r;
2386         }
2387
2388         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2389         synchronize_srcu_expedited(&kvm->srcu);
2390         kfree(bus);
2391         return r;
2392 }
2393
2394 static struct notifier_block kvm_cpu_notifier = {
2395         .notifier_call = kvm_cpu_hotplug,
2396 };
2397
2398 static int vm_stat_get(void *_offset, u64 *val)
2399 {
2400         unsigned offset = (long)_offset;
2401         struct kvm *kvm;
2402
2403         *val = 0;
2404         raw_spin_lock(&kvm_lock);
2405         list_for_each_entry(kvm, &vm_list, vm_list)
2406                 *val += *(u32 *)((void *)kvm + offset);
2407         raw_spin_unlock(&kvm_lock);
2408         return 0;
2409 }
2410
2411 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2412
2413 static int vcpu_stat_get(void *_offset, u64 *val)
2414 {
2415         unsigned offset = (long)_offset;
2416         struct kvm *kvm;
2417         struct kvm_vcpu *vcpu;
2418         int i;
2419
2420         *val = 0;
2421         raw_spin_lock(&kvm_lock);
2422         list_for_each_entry(kvm, &vm_list, vm_list)
2423                 kvm_for_each_vcpu(i, vcpu, kvm)
2424                         *val += *(u32 *)((void *)vcpu + offset);
2425
2426         raw_spin_unlock(&kvm_lock);
2427         return 0;
2428 }
2429
2430 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2431
2432 static const struct file_operations *stat_fops[] = {
2433         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2434         [KVM_STAT_VM]   = &vm_stat_fops,
2435 };
2436
2437 static void kvm_init_debug(void)
2438 {
2439         struct kvm_stats_debugfs_item *p;
2440
2441         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2442         for (p = debugfs_entries; p->name; ++p)
2443                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2444                                                 (void *)(long)p->offset,
2445                                                 stat_fops[p->kind]);
2446 }
2447
2448 static void kvm_exit_debug(void)
2449 {
2450         struct kvm_stats_debugfs_item *p;
2451
2452         for (p = debugfs_entries; p->name; ++p)
2453                 debugfs_remove(p->dentry);
2454         debugfs_remove(kvm_debugfs_dir);
2455 }
2456
2457 static int kvm_suspend(void)
2458 {
2459         if (kvm_usage_count)
2460                 hardware_disable_nolock(NULL);
2461         return 0;
2462 }
2463
2464 static void kvm_resume(void)
2465 {
2466         if (kvm_usage_count) {
2467                 WARN_ON(raw_spin_is_locked(&kvm_lock));
2468                 hardware_enable_nolock(NULL);
2469         }
2470 }
2471
2472 static struct syscore_ops kvm_syscore_ops = {
2473         .suspend = kvm_suspend,
2474         .resume = kvm_resume,
2475 };
2476
2477 struct page *bad_page;
2478 pfn_t bad_pfn;
2479
2480 static inline
2481 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2482 {
2483         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2484 }
2485
2486 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2487 {
2488         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2489
2490         kvm_arch_vcpu_load(vcpu, cpu);
2491 }
2492
2493 static void kvm_sched_out(struct preempt_notifier *pn,
2494                           struct task_struct *next)
2495 {
2496         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2497
2498         kvm_arch_vcpu_put(vcpu);
2499 }
2500
2501 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2502                   struct module *module)
2503 {
2504         int r;
2505         int cpu;
2506
2507         r = kvm_arch_init(opaque);
2508         if (r)
2509                 goto out_fail;
2510
2511         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2512
2513         if (bad_page == NULL) {
2514                 r = -ENOMEM;
2515                 goto out;
2516         }
2517
2518         bad_pfn = page_to_pfn(bad_page);
2519
2520         hwpoison_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2521
2522         if (hwpoison_page == NULL) {
2523                 r = -ENOMEM;
2524                 goto out_free_0;
2525         }
2526
2527         hwpoison_pfn = page_to_pfn(hwpoison_page);
2528
2529         fault_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2530
2531         if (fault_page == NULL) {
2532                 r = -ENOMEM;
2533                 goto out_free_0;
2534         }
2535
2536         fault_pfn = page_to_pfn(fault_page);
2537
2538         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2539                 r = -ENOMEM;
2540                 goto out_free_0;
2541         }
2542
2543         r = kvm_arch_hardware_setup();
2544         if (r < 0)
2545                 goto out_free_0a;
2546
2547         for_each_online_cpu(cpu) {
2548                 smp_call_function_single(cpu,
2549                                 kvm_arch_check_processor_compat,
2550                                 &r, 1);
2551                 if (r < 0)
2552                         goto out_free_1;
2553         }
2554
2555         r = register_cpu_notifier(&kvm_cpu_notifier);
2556         if (r)
2557                 goto out_free_2;
2558         register_reboot_notifier(&kvm_reboot_notifier);
2559
2560         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2561         if (!vcpu_align)
2562                 vcpu_align = __alignof__(struct kvm_vcpu);
2563         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2564                                            0, NULL);
2565         if (!kvm_vcpu_cache) {
2566                 r = -ENOMEM;
2567                 goto out_free_3;
2568         }
2569
2570         r = kvm_async_pf_init();
2571         if (r)
2572                 goto out_free;
2573
2574         kvm_chardev_ops.owner = module;
2575         kvm_vm_fops.owner = module;
2576         kvm_vcpu_fops.owner = module;
2577
2578         r = misc_register(&kvm_dev);
2579         if (r) {
2580                 printk(KERN_ERR "kvm: misc device register failed\n");
2581                 goto out_unreg;
2582         }
2583
2584         register_syscore_ops(&kvm_syscore_ops);
2585
2586         kvm_preempt_ops.sched_in = kvm_sched_in;
2587         kvm_preempt_ops.sched_out = kvm_sched_out;
2588
2589         kvm_init_debug();
2590
2591         return 0;
2592
2593 out_unreg:
2594         kvm_async_pf_deinit();
2595 out_free:
2596         kmem_cache_destroy(kvm_vcpu_cache);
2597 out_free_3:
2598         unregister_reboot_notifier(&kvm_reboot_notifier);
2599         unregister_cpu_notifier(&kvm_cpu_notifier);
2600 out_free_2:
2601 out_free_1:
2602         kvm_arch_hardware_unsetup();
2603 out_free_0a:
2604         free_cpumask_var(cpus_hardware_enabled);
2605 out_free_0:
2606         if (fault_page)
2607                 __free_page(fault_page);
2608         if (hwpoison_page)
2609                 __free_page(hwpoison_page);
2610         __free_page(bad_page);
2611 out:
2612         kvm_arch_exit();
2613 out_fail:
2614         return r;
2615 }
2616 EXPORT_SYMBOL_GPL(kvm_init);
2617
2618 void kvm_exit(void)
2619 {
2620         kvm_exit_debug();
2621         misc_deregister(&kvm_dev);
2622         kmem_cache_destroy(kvm_vcpu_cache);
2623         kvm_async_pf_deinit();
2624         unregister_syscore_ops(&kvm_syscore_ops);
2625         unregister_reboot_notifier(&kvm_reboot_notifier);
2626         unregister_cpu_notifier(&kvm_cpu_notifier);
2627         on_each_cpu(hardware_disable_nolock, NULL, 1);
2628         kvm_arch_hardware_unsetup();
2629         kvm_arch_exit();
2630         free_cpumask_var(cpus_hardware_enabled);
2631         __free_page(hwpoison_page);
2632         __free_page(bad_page);
2633 }
2634 EXPORT_SYMBOL_GPL(kvm_exit);