ed3c4e7c10089aee765d1cc291d9ee5436e018a3
[linux-2.6.git] / virt / kvm / kvm_main.c
1 /*
2  * Kernel-based Virtual Machine driver for Linux
3  *
4  * This module enables machines with Intel VT-x extensions to run virtual
5  * machines without emulation or binary translation.
6  *
7  * Copyright (C) 2006 Qumranet, Inc.
8  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
9  *
10  * Authors:
11  *   Avi Kivity   <avi@qumranet.com>
12  *   Yaniv Kamay  <yaniv@qumranet.com>
13  *
14  * This work is licensed under the terms of the GNU GPL, version 2.  See
15  * the COPYING file in the top-level directory.
16  *
17  */
18
19 #include "iodev.h"
20
21 #include <linux/kvm_host.h>
22 #include <linux/kvm.h>
23 #include <linux/module.h>
24 #include <linux/errno.h>
25 #include <linux/percpu.h>
26 #include <linux/mm.h>
27 #include <linux/miscdevice.h>
28 #include <linux/vmalloc.h>
29 #include <linux/reboot.h>
30 #include <linux/debugfs.h>
31 #include <linux/highmem.h>
32 #include <linux/file.h>
33 #include <linux/syscore_ops.h>
34 #include <linux/cpu.h>
35 #include <linux/sched.h>
36 #include <linux/cpumask.h>
37 #include <linux/smp.h>
38 #include <linux/anon_inodes.h>
39 #include <linux/profile.h>
40 #include <linux/kvm_para.h>
41 #include <linux/pagemap.h>
42 #include <linux/mman.h>
43 #include <linux/swap.h>
44 #include <linux/bitops.h>
45 #include <linux/spinlock.h>
46 #include <linux/compat.h>
47 #include <linux/srcu.h>
48 #include <linux/hugetlb.h>
49 #include <linux/slab.h>
50
51 #include <asm/processor.h>
52 #include <asm/io.h>
53 #include <asm/uaccess.h>
54 #include <asm/pgtable.h>
55
56 #include "coalesced_mmio.h"
57 #include "async_pf.h"
58
59 #define CREATE_TRACE_POINTS
60 #include <trace/events/kvm.h>
61
62 MODULE_AUTHOR("Qumranet");
63 MODULE_LICENSE("GPL");
64
65 /*
66  * Ordering of locks:
67  *
68  *              kvm->lock --> kvm->slots_lock --> kvm->irq_lock
69  */
70
71 DEFINE_RAW_SPINLOCK(kvm_lock);
72 LIST_HEAD(vm_list);
73
74 static cpumask_var_t cpus_hardware_enabled;
75 static int kvm_usage_count = 0;
76 static atomic_t hardware_enable_failed;
77
78 struct kmem_cache *kvm_vcpu_cache;
79 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
80
81 static __read_mostly struct preempt_ops kvm_preempt_ops;
82
83 struct dentry *kvm_debugfs_dir;
84
85 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
86                            unsigned long arg);
87 static int hardware_enable_all(void);
88 static void hardware_disable_all(void);
89
90 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
91
92 bool kvm_rebooting;
93 EXPORT_SYMBOL_GPL(kvm_rebooting);
94
95 static bool largepages_enabled = true;
96
97 static struct page *hwpoison_page;
98 static pfn_t hwpoison_pfn;
99
100 static struct page *fault_page;
101 static pfn_t fault_pfn;
102
103 inline int kvm_is_mmio_pfn(pfn_t pfn)
104 {
105         if (pfn_valid(pfn)) {
106                 int reserved;
107                 struct page *tail = pfn_to_page(pfn);
108                 struct page *head = compound_trans_head(tail);
109                 reserved = PageReserved(head);
110                 if (head != tail) {
111                         /*
112                          * "head" is not a dangling pointer
113                          * (compound_trans_head takes care of that)
114                          * but the hugepage may have been splitted
115                          * from under us (and we may not hold a
116                          * reference count on the head page so it can
117                          * be reused before we run PageReferenced), so
118                          * we've to check PageTail before returning
119                          * what we just read.
120                          */
121                         smp_rmb();
122                         if (PageTail(tail))
123                                 return reserved;
124                 }
125                 return PageReserved(tail);
126         }
127
128         return true;
129 }
130
131 /*
132  * Switches to specified vcpu, until a matching vcpu_put()
133  */
134 void vcpu_load(struct kvm_vcpu *vcpu)
135 {
136         int cpu;
137
138         mutex_lock(&vcpu->mutex);
139         if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
140                 /* The thread running this VCPU changed. */
141                 struct pid *oldpid = vcpu->pid;
142                 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
143                 rcu_assign_pointer(vcpu->pid, newpid);
144                 synchronize_rcu();
145                 put_pid(oldpid);
146         }
147         cpu = get_cpu();
148         preempt_notifier_register(&vcpu->preempt_notifier);
149         kvm_arch_vcpu_load(vcpu, cpu);
150         put_cpu();
151 }
152
153 void vcpu_put(struct kvm_vcpu *vcpu)
154 {
155         preempt_disable();
156         kvm_arch_vcpu_put(vcpu);
157         preempt_notifier_unregister(&vcpu->preempt_notifier);
158         preempt_enable();
159         mutex_unlock(&vcpu->mutex);
160 }
161
162 static void ack_flush(void *_completed)
163 {
164 }
165
166 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
167 {
168         int i, cpu, me;
169         cpumask_var_t cpus;
170         bool called = true;
171         struct kvm_vcpu *vcpu;
172
173         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
174
175         me = get_cpu();
176         kvm_for_each_vcpu(i, vcpu, kvm) {
177                 kvm_make_request(req, vcpu);
178                 cpu = vcpu->cpu;
179
180                 /* Set ->requests bit before we read ->mode */
181                 smp_mb();
182
183                 if (cpus != NULL && cpu != -1 && cpu != me &&
184                       kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
185                         cpumask_set_cpu(cpu, cpus);
186         }
187         if (unlikely(cpus == NULL))
188                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
189         else if (!cpumask_empty(cpus))
190                 smp_call_function_many(cpus, ack_flush, NULL, 1);
191         else
192                 called = false;
193         put_cpu();
194         free_cpumask_var(cpus);
195         return called;
196 }
197
198 void kvm_flush_remote_tlbs(struct kvm *kvm)
199 {
200         int dirty_count = kvm->tlbs_dirty;
201
202         smp_mb();
203         if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
204                 ++kvm->stat.remote_tlb_flush;
205         cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
206 }
207
208 void kvm_reload_remote_mmus(struct kvm *kvm)
209 {
210         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
211 }
212
213 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
214 {
215         struct page *page;
216         int r;
217
218         mutex_init(&vcpu->mutex);
219         vcpu->cpu = -1;
220         vcpu->kvm = kvm;
221         vcpu->vcpu_id = id;
222         vcpu->pid = NULL;
223         init_waitqueue_head(&vcpu->wq);
224         kvm_async_pf_vcpu_init(vcpu);
225
226         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
227         if (!page) {
228                 r = -ENOMEM;
229                 goto fail;
230         }
231         vcpu->run = page_address(page);
232
233         r = kvm_arch_vcpu_init(vcpu);
234         if (r < 0)
235                 goto fail_free_run;
236         return 0;
237
238 fail_free_run:
239         free_page((unsigned long)vcpu->run);
240 fail:
241         return r;
242 }
243 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
244
245 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
246 {
247         put_pid(vcpu->pid);
248         kvm_arch_vcpu_uninit(vcpu);
249         free_page((unsigned long)vcpu->run);
250 }
251 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
252
253 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
254 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
255 {
256         return container_of(mn, struct kvm, mmu_notifier);
257 }
258
259 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
260                                              struct mm_struct *mm,
261                                              unsigned long address)
262 {
263         struct kvm *kvm = mmu_notifier_to_kvm(mn);
264         int need_tlb_flush, idx;
265
266         /*
267          * When ->invalidate_page runs, the linux pte has been zapped
268          * already but the page is still allocated until
269          * ->invalidate_page returns. So if we increase the sequence
270          * here the kvm page fault will notice if the spte can't be
271          * established because the page is going to be freed. If
272          * instead the kvm page fault establishes the spte before
273          * ->invalidate_page runs, kvm_unmap_hva will release it
274          * before returning.
275          *
276          * The sequence increase only need to be seen at spin_unlock
277          * time, and not at spin_lock time.
278          *
279          * Increasing the sequence after the spin_unlock would be
280          * unsafe because the kvm page fault could then establish the
281          * pte after kvm_unmap_hva returned, without noticing the page
282          * is going to be freed.
283          */
284         idx = srcu_read_lock(&kvm->srcu);
285         spin_lock(&kvm->mmu_lock);
286         kvm->mmu_notifier_seq++;
287         need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
288         spin_unlock(&kvm->mmu_lock);
289         srcu_read_unlock(&kvm->srcu, idx);
290
291         /* we've to flush the tlb before the pages can be freed */
292         if (need_tlb_flush)
293                 kvm_flush_remote_tlbs(kvm);
294
295 }
296
297 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
298                                         struct mm_struct *mm,
299                                         unsigned long address,
300                                         pte_t pte)
301 {
302         struct kvm *kvm = mmu_notifier_to_kvm(mn);
303         int idx;
304
305         idx = srcu_read_lock(&kvm->srcu);
306         spin_lock(&kvm->mmu_lock);
307         kvm->mmu_notifier_seq++;
308         kvm_set_spte_hva(kvm, address, pte);
309         spin_unlock(&kvm->mmu_lock);
310         srcu_read_unlock(&kvm->srcu, idx);
311 }
312
313 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
314                                                     struct mm_struct *mm,
315                                                     unsigned long start,
316                                                     unsigned long end)
317 {
318         struct kvm *kvm = mmu_notifier_to_kvm(mn);
319         int need_tlb_flush = 0, idx;
320
321         idx = srcu_read_lock(&kvm->srcu);
322         spin_lock(&kvm->mmu_lock);
323         /*
324          * The count increase must become visible at unlock time as no
325          * spte can be established without taking the mmu_lock and
326          * count is also read inside the mmu_lock critical section.
327          */
328         kvm->mmu_notifier_count++;
329         for (; start < end; start += PAGE_SIZE)
330                 need_tlb_flush |= kvm_unmap_hva(kvm, start);
331         need_tlb_flush |= kvm->tlbs_dirty;
332         spin_unlock(&kvm->mmu_lock);
333         srcu_read_unlock(&kvm->srcu, idx);
334
335         /* we've to flush the tlb before the pages can be freed */
336         if (need_tlb_flush)
337                 kvm_flush_remote_tlbs(kvm);
338 }
339
340 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
341                                                   struct mm_struct *mm,
342                                                   unsigned long start,
343                                                   unsigned long end)
344 {
345         struct kvm *kvm = mmu_notifier_to_kvm(mn);
346
347         spin_lock(&kvm->mmu_lock);
348         /*
349          * This sequence increase will notify the kvm page fault that
350          * the page that is going to be mapped in the spte could have
351          * been freed.
352          */
353         kvm->mmu_notifier_seq++;
354         /*
355          * The above sequence increase must be visible before the
356          * below count decrease but both values are read by the kvm
357          * page fault under mmu_lock spinlock so we don't need to add
358          * a smb_wmb() here in between the two.
359          */
360         kvm->mmu_notifier_count--;
361         spin_unlock(&kvm->mmu_lock);
362
363         BUG_ON(kvm->mmu_notifier_count < 0);
364 }
365
366 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
367                                               struct mm_struct *mm,
368                                               unsigned long address)
369 {
370         struct kvm *kvm = mmu_notifier_to_kvm(mn);
371         int young, idx;
372
373         idx = srcu_read_lock(&kvm->srcu);
374         spin_lock(&kvm->mmu_lock);
375         young = kvm_age_hva(kvm, address);
376         spin_unlock(&kvm->mmu_lock);
377         srcu_read_unlock(&kvm->srcu, idx);
378
379         if (young)
380                 kvm_flush_remote_tlbs(kvm);
381
382         return young;
383 }
384
385 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
386                                        struct mm_struct *mm,
387                                        unsigned long address)
388 {
389         struct kvm *kvm = mmu_notifier_to_kvm(mn);
390         int young, idx;
391
392         idx = srcu_read_lock(&kvm->srcu);
393         spin_lock(&kvm->mmu_lock);
394         young = kvm_test_age_hva(kvm, address);
395         spin_unlock(&kvm->mmu_lock);
396         srcu_read_unlock(&kvm->srcu, idx);
397
398         return young;
399 }
400
401 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
402                                      struct mm_struct *mm)
403 {
404         struct kvm *kvm = mmu_notifier_to_kvm(mn);
405         int idx;
406
407         idx = srcu_read_lock(&kvm->srcu);
408         kvm_arch_flush_shadow(kvm);
409         srcu_read_unlock(&kvm->srcu, idx);
410 }
411
412 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
413         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
414         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
415         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
416         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
417         .test_young             = kvm_mmu_notifier_test_young,
418         .change_pte             = kvm_mmu_notifier_change_pte,
419         .release                = kvm_mmu_notifier_release,
420 };
421
422 static int kvm_init_mmu_notifier(struct kvm *kvm)
423 {
424         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
425         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
426 }
427
428 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
429
430 static int kvm_init_mmu_notifier(struct kvm *kvm)
431 {
432         return 0;
433 }
434
435 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
436
437 static struct kvm *kvm_create_vm(void)
438 {
439         int r, i;
440         struct kvm *kvm = kvm_arch_alloc_vm();
441
442         if (!kvm)
443                 return ERR_PTR(-ENOMEM);
444
445         r = kvm_arch_init_vm(kvm);
446         if (r)
447                 goto out_err_nodisable;
448
449         r = hardware_enable_all();
450         if (r)
451                 goto out_err_nodisable;
452
453 #ifdef CONFIG_HAVE_KVM_IRQCHIP
454         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
455         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
456 #endif
457
458         r = -ENOMEM;
459         kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
460         if (!kvm->memslots)
461                 goto out_err_nosrcu;
462         if (init_srcu_struct(&kvm->srcu))
463                 goto out_err_nosrcu;
464         for (i = 0; i < KVM_NR_BUSES; i++) {
465                 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
466                                         GFP_KERNEL);
467                 if (!kvm->buses[i])
468                         goto out_err;
469         }
470
471         r = kvm_init_mmu_notifier(kvm);
472         if (r)
473                 goto out_err;
474
475         kvm->mm = current->mm;
476         atomic_inc(&kvm->mm->mm_count);
477         spin_lock_init(&kvm->mmu_lock);
478         kvm_eventfd_init(kvm);
479         mutex_init(&kvm->lock);
480         mutex_init(&kvm->irq_lock);
481         mutex_init(&kvm->slots_lock);
482         atomic_set(&kvm->users_count, 1);
483         raw_spin_lock(&kvm_lock);
484         list_add(&kvm->vm_list, &vm_list);
485         raw_spin_unlock(&kvm_lock);
486
487         return kvm;
488
489 out_err:
490         cleanup_srcu_struct(&kvm->srcu);
491 out_err_nosrcu:
492         hardware_disable_all();
493 out_err_nodisable:
494         for (i = 0; i < KVM_NR_BUSES; i++)
495                 kfree(kvm->buses[i]);
496         kfree(kvm->memslots);
497         kvm_arch_free_vm(kvm);
498         return ERR_PTR(r);
499 }
500
501 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
502 {
503         if (!memslot->dirty_bitmap)
504                 return;
505
506         if (2 * kvm_dirty_bitmap_bytes(memslot) > PAGE_SIZE)
507                 vfree(memslot->dirty_bitmap_head);
508         else
509                 kfree(memslot->dirty_bitmap_head);
510
511         memslot->dirty_bitmap = NULL;
512         memslot->dirty_bitmap_head = NULL;
513 }
514
515 /*
516  * Free any memory in @free but not in @dont.
517  */
518 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
519                                   struct kvm_memory_slot *dont)
520 {
521         int i;
522
523         if (!dont || free->rmap != dont->rmap)
524                 vfree(free->rmap);
525
526         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
527                 kvm_destroy_dirty_bitmap(free);
528
529
530         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
531                 if (!dont || free->lpage_info[i] != dont->lpage_info[i]) {
532                         vfree(free->lpage_info[i]);
533                         free->lpage_info[i] = NULL;
534                 }
535         }
536
537         free->npages = 0;
538         free->rmap = NULL;
539 }
540
541 void kvm_free_physmem(struct kvm *kvm)
542 {
543         int i;
544         struct kvm_memslots *slots = kvm->memslots;
545
546         for (i = 0; i < slots->nmemslots; ++i)
547                 kvm_free_physmem_slot(&slots->memslots[i], NULL);
548
549         kfree(kvm->memslots);
550 }
551
552 static void kvm_destroy_vm(struct kvm *kvm)
553 {
554         int i;
555         struct mm_struct *mm = kvm->mm;
556
557         kvm_arch_sync_events(kvm);
558         raw_spin_lock(&kvm_lock);
559         list_del(&kvm->vm_list);
560         raw_spin_unlock(&kvm_lock);
561         kvm_free_irq_routing(kvm);
562         for (i = 0; i < KVM_NR_BUSES; i++)
563                 kvm_io_bus_destroy(kvm->buses[i]);
564         kvm_coalesced_mmio_free(kvm);
565 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
566         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
567 #else
568         kvm_arch_flush_shadow(kvm);
569 #endif
570         kvm_arch_destroy_vm(kvm);
571         kvm_free_physmem(kvm);
572         cleanup_srcu_struct(&kvm->srcu);
573         kvm_arch_free_vm(kvm);
574         hardware_disable_all();
575         mmdrop(mm);
576 }
577
578 void kvm_get_kvm(struct kvm *kvm)
579 {
580         atomic_inc(&kvm->users_count);
581 }
582 EXPORT_SYMBOL_GPL(kvm_get_kvm);
583
584 void kvm_put_kvm(struct kvm *kvm)
585 {
586         if (atomic_dec_and_test(&kvm->users_count))
587                 kvm_destroy_vm(kvm);
588 }
589 EXPORT_SYMBOL_GPL(kvm_put_kvm);
590
591
592 static int kvm_vm_release(struct inode *inode, struct file *filp)
593 {
594         struct kvm *kvm = filp->private_data;
595
596         kvm_irqfd_release(kvm);
597
598         kvm_put_kvm(kvm);
599         return 0;
600 }
601
602 #ifndef CONFIG_S390
603 /*
604  * Allocation size is twice as large as the actual dirty bitmap size.
605  * This makes it possible to do double buffering: see x86's
606  * kvm_vm_ioctl_get_dirty_log().
607  */
608 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
609 {
610         unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
611
612         if (dirty_bytes > PAGE_SIZE)
613                 memslot->dirty_bitmap = vzalloc(dirty_bytes);
614         else
615                 memslot->dirty_bitmap = kzalloc(dirty_bytes, GFP_KERNEL);
616
617         if (!memslot->dirty_bitmap)
618                 return -ENOMEM;
619
620         memslot->dirty_bitmap_head = memslot->dirty_bitmap;
621         return 0;
622 }
623 #endif /* !CONFIG_S390 */
624
625 /*
626  * Allocate some memory and give it an address in the guest physical address
627  * space.
628  *
629  * Discontiguous memory is allowed, mostly for framebuffers.
630  *
631  * Must be called holding mmap_sem for write.
632  */
633 int __kvm_set_memory_region(struct kvm *kvm,
634                             struct kvm_userspace_memory_region *mem,
635                             int user_alloc)
636 {
637         int r;
638         gfn_t base_gfn;
639         unsigned long npages;
640         unsigned long i;
641         struct kvm_memory_slot *memslot;
642         struct kvm_memory_slot old, new;
643         struct kvm_memslots *slots, *old_memslots;
644
645         r = -EINVAL;
646         /* General sanity checks */
647         if (mem->memory_size & (PAGE_SIZE - 1))
648                 goto out;
649         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
650                 goto out;
651         /* 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, mem->userspace_addr, mem->memory_size)))
655                 goto out;
656         if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
657                 goto out;
658         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
659                 goto out;
660
661         memslot = &kvm->memslots->memslots[mem->slot];
662         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
663         npages = mem->memory_size >> PAGE_SHIFT;
664
665         r = -EINVAL;
666         if (npages > KVM_MEM_MAX_NR_PAGES)
667                 goto out;
668
669         if (!npages)
670                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
671
672         new = old = *memslot;
673
674         new.id = mem->slot;
675         new.base_gfn = base_gfn;
676         new.npages = npages;
677         new.flags = mem->flags;
678
679         /* Disallow changing a memory slot's size. */
680         r = -EINVAL;
681         if (npages && old.npages && npages != old.npages)
682                 goto out_free;
683
684         /* Check for overlaps */
685         r = -EEXIST;
686         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
687                 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
688
689                 if (s == memslot || !s->npages)
690                         continue;
691                 if (!((base_gfn + npages <= s->base_gfn) ||
692                       (base_gfn >= s->base_gfn + s->npages)))
693                         goto out_free;
694         }
695
696         /* Free page dirty bitmap if unneeded */
697         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
698                 new.dirty_bitmap = NULL;
699
700         r = -ENOMEM;
701
702         /* Allocate if a slot is being created */
703 #ifndef CONFIG_S390
704         if (npages && !new.rmap) {
705                 new.rmap = vzalloc(npages * sizeof(*new.rmap));
706
707                 if (!new.rmap)
708                         goto out_free;
709
710                 new.user_alloc = user_alloc;
711                 new.userspace_addr = mem->userspace_addr;
712         }
713         if (!npages)
714                 goto skip_lpage;
715
716         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
717                 unsigned long ugfn;
718                 unsigned long j;
719                 int lpages;
720                 int level = i + 2;
721
722                 /* Avoid unused variable warning if no large pages */
723                 (void)level;
724
725                 if (new.lpage_info[i])
726                         continue;
727
728                 lpages = 1 + ((base_gfn + npages - 1)
729                              >> KVM_HPAGE_GFN_SHIFT(level));
730                 lpages -= base_gfn >> KVM_HPAGE_GFN_SHIFT(level);
731
732                 new.lpage_info[i] = vzalloc(lpages * sizeof(*new.lpage_info[i]));
733
734                 if (!new.lpage_info[i])
735                         goto out_free;
736
737                 if (base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
738                         new.lpage_info[i][0].write_count = 1;
739                 if ((base_gfn+npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
740                         new.lpage_info[i][lpages - 1].write_count = 1;
741                 ugfn = new.userspace_addr >> PAGE_SHIFT;
742                 /*
743                  * If the gfn and userspace address are not aligned wrt each
744                  * other, or if explicitly asked to, disable large page
745                  * support for this slot
746                  */
747                 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
748                     !largepages_enabled)
749                         for (j = 0; j < lpages; ++j)
750                                 new.lpage_info[i][j].write_count = 1;
751         }
752
753 skip_lpage:
754
755         /* Allocate page dirty bitmap if needed */
756         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
757                 if (kvm_create_dirty_bitmap(&new) < 0)
758                         goto out_free;
759                 /* destroy any largepage mappings for dirty tracking */
760         }
761 #else  /* not defined CONFIG_S390 */
762         new.user_alloc = user_alloc;
763         if (user_alloc)
764                 new.userspace_addr = mem->userspace_addr;
765 #endif /* not defined CONFIG_S390 */
766
767         if (!npages) {
768                 r = -ENOMEM;
769                 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
770                 if (!slots)
771                         goto out_free;
772                 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
773                 if (mem->slot >= slots->nmemslots)
774                         slots->nmemslots = mem->slot + 1;
775                 slots->generation++;
776                 slots->memslots[mem->slot].flags |= KVM_MEMSLOT_INVALID;
777
778                 old_memslots = kvm->memslots;
779                 rcu_assign_pointer(kvm->memslots, slots);
780                 synchronize_srcu_expedited(&kvm->srcu);
781                 /* From this point no new shadow pages pointing to a deleted
782                  * memslot will be created.
783                  *
784                  * validation of sp->gfn happens in:
785                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
786                  *      - kvm_is_visible_gfn (mmu_check_roots)
787                  */
788                 kvm_arch_flush_shadow(kvm);
789                 kfree(old_memslots);
790         }
791
792         r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
793         if (r)
794                 goto out_free;
795
796         /* map the pages in iommu page table */
797         if (npages) {
798                 r = kvm_iommu_map_pages(kvm, &new);
799                 if (r)
800                         goto out_free;
801         }
802
803         r = -ENOMEM;
804         slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
805         if (!slots)
806                 goto out_free;
807         memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
808         if (mem->slot >= slots->nmemslots)
809                 slots->nmemslots = mem->slot + 1;
810         slots->generation++;
811
812         /* actual memory is freed via old in kvm_free_physmem_slot below */
813         if (!npages) {
814                 new.rmap = NULL;
815                 new.dirty_bitmap = NULL;
816                 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i)
817                         new.lpage_info[i] = NULL;
818         }
819
820         slots->memslots[mem->slot] = new;
821         old_memslots = kvm->memslots;
822         rcu_assign_pointer(kvm->memslots, slots);
823         synchronize_srcu_expedited(&kvm->srcu);
824
825         kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
826
827         kvm_free_physmem_slot(&old, &new);
828         kfree(old_memslots);
829
830         return 0;
831
832 out_free:
833         kvm_free_physmem_slot(&new, &old);
834 out:
835         return r;
836
837 }
838 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
839
840 int kvm_set_memory_region(struct kvm *kvm,
841                           struct kvm_userspace_memory_region *mem,
842                           int user_alloc)
843 {
844         int r;
845
846         mutex_lock(&kvm->slots_lock);
847         r = __kvm_set_memory_region(kvm, mem, user_alloc);
848         mutex_unlock(&kvm->slots_lock);
849         return r;
850 }
851 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
852
853 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
854                                    struct
855                                    kvm_userspace_memory_region *mem,
856                                    int user_alloc)
857 {
858         if (mem->slot >= KVM_MEMORY_SLOTS)
859                 return -EINVAL;
860         return kvm_set_memory_region(kvm, mem, user_alloc);
861 }
862
863 int kvm_get_dirty_log(struct kvm *kvm,
864                         struct kvm_dirty_log *log, int *is_dirty)
865 {
866         struct kvm_memory_slot *memslot;
867         int r, i;
868         unsigned long n;
869         unsigned long any = 0;
870
871         r = -EINVAL;
872         if (log->slot >= KVM_MEMORY_SLOTS)
873                 goto out;
874
875         memslot = &kvm->memslots->memslots[log->slot];
876         r = -ENOENT;
877         if (!memslot->dirty_bitmap)
878                 goto out;
879
880         n = kvm_dirty_bitmap_bytes(memslot);
881
882         for (i = 0; !any && i < n/sizeof(long); ++i)
883                 any = memslot->dirty_bitmap[i];
884
885         r = -EFAULT;
886         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
887                 goto out;
888
889         if (any)
890                 *is_dirty = 1;
891
892         r = 0;
893 out:
894         return r;
895 }
896
897 void kvm_disable_largepages(void)
898 {
899         largepages_enabled = false;
900 }
901 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
902
903 int is_error_page(struct page *page)
904 {
905         return page == bad_page || page == hwpoison_page || page == fault_page;
906 }
907 EXPORT_SYMBOL_GPL(is_error_page);
908
909 int is_error_pfn(pfn_t pfn)
910 {
911         return pfn == bad_pfn || pfn == hwpoison_pfn || pfn == fault_pfn;
912 }
913 EXPORT_SYMBOL_GPL(is_error_pfn);
914
915 int is_hwpoison_pfn(pfn_t pfn)
916 {
917         return pfn == hwpoison_pfn;
918 }
919 EXPORT_SYMBOL_GPL(is_hwpoison_pfn);
920
921 int is_fault_pfn(pfn_t pfn)
922 {
923         return pfn == fault_pfn;
924 }
925 EXPORT_SYMBOL_GPL(is_fault_pfn);
926
927 static inline unsigned long bad_hva(void)
928 {
929         return PAGE_OFFSET;
930 }
931
932 int kvm_is_error_hva(unsigned long addr)
933 {
934         return addr == bad_hva();
935 }
936 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
937
938 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm_memslots *slots,
939                                                 gfn_t gfn)
940 {
941         int i;
942
943         for (i = 0; i < slots->nmemslots; ++i) {
944                 struct kvm_memory_slot *memslot = &slots->memslots[i];
945
946                 if (gfn >= memslot->base_gfn
947                     && gfn < memslot->base_gfn + memslot->npages)
948                         return memslot;
949         }
950         return NULL;
951 }
952
953 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
954 {
955         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
956 }
957 EXPORT_SYMBOL_GPL(gfn_to_memslot);
958
959 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
960 {
961         int i;
962         struct kvm_memslots *slots = kvm_memslots(kvm);
963
964         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
965                 struct kvm_memory_slot *memslot = &slots->memslots[i];
966
967                 if (memslot->flags & KVM_MEMSLOT_INVALID)
968                         continue;
969
970                 if (gfn >= memslot->base_gfn
971                     && gfn < memslot->base_gfn + memslot->npages)
972                         return 1;
973         }
974         return 0;
975 }
976 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
977
978 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
979 {
980         struct vm_area_struct *vma;
981         unsigned long addr, size;
982
983         size = PAGE_SIZE;
984
985         addr = gfn_to_hva(kvm, gfn);
986         if (kvm_is_error_hva(addr))
987                 return PAGE_SIZE;
988
989         down_read(&current->mm->mmap_sem);
990         vma = find_vma(current->mm, addr);
991         if (!vma)
992                 goto out;
993
994         size = vma_kernel_pagesize(vma);
995
996 out:
997         up_read(&current->mm->mmap_sem);
998
999         return size;
1000 }
1001
1002 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1003                                      gfn_t *nr_pages)
1004 {
1005         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1006                 return bad_hva();
1007
1008         if (nr_pages)
1009                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1010
1011         return gfn_to_hva_memslot(slot, gfn);
1012 }
1013
1014 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1015 {
1016         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1017 }
1018 EXPORT_SYMBOL_GPL(gfn_to_hva);
1019
1020 static pfn_t get_fault_pfn(void)
1021 {
1022         get_page(fault_page);
1023         return fault_pfn;
1024 }
1025
1026 int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1027         unsigned long start, int write, struct page **page)
1028 {
1029         int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1030
1031         if (write)
1032                 flags |= FOLL_WRITE;
1033
1034         return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1035 }
1036
1037 static inline int check_user_page_hwpoison(unsigned long addr)
1038 {
1039         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1040
1041         rc = __get_user_pages(current, current->mm, addr, 1,
1042                               flags, NULL, NULL, NULL);
1043         return rc == -EHWPOISON;
1044 }
1045
1046 static pfn_t hva_to_pfn(struct kvm *kvm, unsigned long addr, bool atomic,
1047                         bool *async, bool write_fault, bool *writable)
1048 {
1049         struct page *page[1];
1050         int npages = 0;
1051         pfn_t pfn;
1052
1053         /* we can do it either atomically or asynchronously, not both */
1054         BUG_ON(atomic && async);
1055
1056         BUG_ON(!write_fault && !writable);
1057
1058         if (writable)
1059                 *writable = true;
1060
1061         if (atomic || async)
1062                 npages = __get_user_pages_fast(addr, 1, 1, page);
1063
1064         if (unlikely(npages != 1) && !atomic) {
1065                 might_sleep();
1066
1067                 if (writable)
1068                         *writable = write_fault;
1069
1070                 if (async) {
1071                         down_read(&current->mm->mmap_sem);
1072                         npages = get_user_page_nowait(current, current->mm,
1073                                                      addr, write_fault, page);
1074                         up_read(&current->mm->mmap_sem);
1075                 } else
1076                         npages = get_user_pages_fast(addr, 1, write_fault,
1077                                                      page);
1078
1079                 /* map read fault as writable if possible */
1080                 if (unlikely(!write_fault) && npages == 1) {
1081                         struct page *wpage[1];
1082
1083                         npages = __get_user_pages_fast(addr, 1, 1, wpage);
1084                         if (npages == 1) {
1085                                 *writable = true;
1086                                 put_page(page[0]);
1087                                 page[0] = wpage[0];
1088                         }
1089                         npages = 1;
1090                 }
1091         }
1092
1093         if (unlikely(npages != 1)) {
1094                 struct vm_area_struct *vma;
1095
1096                 if (atomic)
1097                         return get_fault_pfn();
1098
1099                 down_read(&current->mm->mmap_sem);
1100                 if (npages == -EHWPOISON ||
1101                         (!async && check_user_page_hwpoison(addr))) {
1102                         up_read(&current->mm->mmap_sem);
1103                         get_page(hwpoison_page);
1104                         return page_to_pfn(hwpoison_page);
1105                 }
1106
1107                 vma = find_vma_intersection(current->mm, addr, addr+1);
1108
1109                 if (vma == NULL)
1110                         pfn = get_fault_pfn();
1111                 else if ((vma->vm_flags & VM_PFNMAP)) {
1112                         pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1113                                 vma->vm_pgoff;
1114                         BUG_ON(!kvm_is_mmio_pfn(pfn));
1115                 } else {
1116                         if (async && (vma->vm_flags & VM_WRITE))
1117                                 *async = true;
1118                         pfn = get_fault_pfn();
1119                 }
1120                 up_read(&current->mm->mmap_sem);
1121         } else
1122                 pfn = page_to_pfn(page[0]);
1123
1124         return pfn;
1125 }
1126
1127 pfn_t hva_to_pfn_atomic(struct kvm *kvm, unsigned long addr)
1128 {
1129         return hva_to_pfn(kvm, addr, true, NULL, true, NULL);
1130 }
1131 EXPORT_SYMBOL_GPL(hva_to_pfn_atomic);
1132
1133 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1134                           bool write_fault, bool *writable)
1135 {
1136         unsigned long addr;
1137
1138         if (async)
1139                 *async = false;
1140
1141         addr = gfn_to_hva(kvm, gfn);
1142         if (kvm_is_error_hva(addr)) {
1143                 get_page(bad_page);
1144                 return page_to_pfn(bad_page);
1145         }
1146
1147         return hva_to_pfn(kvm, addr, atomic, async, write_fault, writable);
1148 }
1149
1150 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1151 {
1152         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1153 }
1154 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1155
1156 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1157                        bool write_fault, bool *writable)
1158 {
1159         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1160 }
1161 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1162
1163 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1164 {
1165         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1166 }
1167 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1168
1169 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1170                       bool *writable)
1171 {
1172         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1173 }
1174 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1175
1176 pfn_t gfn_to_pfn_memslot(struct kvm *kvm,
1177                          struct kvm_memory_slot *slot, gfn_t gfn)
1178 {
1179         unsigned long addr = gfn_to_hva_memslot(slot, gfn);
1180         return hva_to_pfn(kvm, addr, false, NULL, true, NULL);
1181 }
1182
1183 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1184                                                                   int nr_pages)
1185 {
1186         unsigned long addr;
1187         gfn_t entry;
1188
1189         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1190         if (kvm_is_error_hva(addr))
1191                 return -1;
1192
1193         if (entry < nr_pages)
1194                 return 0;
1195
1196         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1197 }
1198 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1199
1200 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1201 {
1202         pfn_t pfn;
1203
1204         pfn = gfn_to_pfn(kvm, gfn);
1205         if (!kvm_is_mmio_pfn(pfn))
1206                 return pfn_to_page(pfn);
1207
1208         WARN_ON(kvm_is_mmio_pfn(pfn));
1209
1210         get_page(bad_page);
1211         return bad_page;
1212 }
1213
1214 EXPORT_SYMBOL_GPL(gfn_to_page);
1215
1216 void kvm_release_page_clean(struct page *page)
1217 {
1218         kvm_release_pfn_clean(page_to_pfn(page));
1219 }
1220 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1221
1222 void kvm_release_pfn_clean(pfn_t pfn)
1223 {
1224         if (!kvm_is_mmio_pfn(pfn))
1225                 put_page(pfn_to_page(pfn));
1226 }
1227 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1228
1229 void kvm_release_page_dirty(struct page *page)
1230 {
1231         kvm_release_pfn_dirty(page_to_pfn(page));
1232 }
1233 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1234
1235 void kvm_release_pfn_dirty(pfn_t pfn)
1236 {
1237         kvm_set_pfn_dirty(pfn);
1238         kvm_release_pfn_clean(pfn);
1239 }
1240 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1241
1242 void kvm_set_page_dirty(struct page *page)
1243 {
1244         kvm_set_pfn_dirty(page_to_pfn(page));
1245 }
1246 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1247
1248 void kvm_set_pfn_dirty(pfn_t pfn)
1249 {
1250         if (!kvm_is_mmio_pfn(pfn)) {
1251                 struct page *page = pfn_to_page(pfn);
1252                 if (!PageReserved(page))
1253                         SetPageDirty(page);
1254         }
1255 }
1256 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1257
1258 void kvm_set_pfn_accessed(pfn_t pfn)
1259 {
1260         if (!kvm_is_mmio_pfn(pfn))
1261                 mark_page_accessed(pfn_to_page(pfn));
1262 }
1263 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1264
1265 void kvm_get_pfn(pfn_t pfn)
1266 {
1267         if (!kvm_is_mmio_pfn(pfn))
1268                 get_page(pfn_to_page(pfn));
1269 }
1270 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1271
1272 static int next_segment(unsigned long len, int offset)
1273 {
1274         if (len > PAGE_SIZE - offset)
1275                 return PAGE_SIZE - offset;
1276         else
1277                 return len;
1278 }
1279
1280 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1281                         int len)
1282 {
1283         int r;
1284         unsigned long addr;
1285
1286         addr = gfn_to_hva(kvm, gfn);
1287         if (kvm_is_error_hva(addr))
1288                 return -EFAULT;
1289         r = __copy_from_user(data, (void __user *)addr + offset, len);
1290         if (r)
1291                 return -EFAULT;
1292         return 0;
1293 }
1294 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1295
1296 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1297 {
1298         gfn_t gfn = gpa >> PAGE_SHIFT;
1299         int seg;
1300         int offset = offset_in_page(gpa);
1301         int ret;
1302
1303         while ((seg = next_segment(len, offset)) != 0) {
1304                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1305                 if (ret < 0)
1306                         return ret;
1307                 offset = 0;
1308                 len -= seg;
1309                 data += seg;
1310                 ++gfn;
1311         }
1312         return 0;
1313 }
1314 EXPORT_SYMBOL_GPL(kvm_read_guest);
1315
1316 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1317                           unsigned long len)
1318 {
1319         int r;
1320         unsigned long addr;
1321         gfn_t gfn = gpa >> PAGE_SHIFT;
1322         int offset = offset_in_page(gpa);
1323
1324         addr = gfn_to_hva(kvm, gfn);
1325         if (kvm_is_error_hva(addr))
1326                 return -EFAULT;
1327         pagefault_disable();
1328         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1329         pagefault_enable();
1330         if (r)
1331                 return -EFAULT;
1332         return 0;
1333 }
1334 EXPORT_SYMBOL(kvm_read_guest_atomic);
1335
1336 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1337                          int offset, int len)
1338 {
1339         int r;
1340         unsigned long addr;
1341
1342         addr = gfn_to_hva(kvm, gfn);
1343         if (kvm_is_error_hva(addr))
1344                 return -EFAULT;
1345         r = copy_to_user((void __user *)addr + offset, data, len);
1346         if (r)
1347                 return -EFAULT;
1348         mark_page_dirty(kvm, gfn);
1349         return 0;
1350 }
1351 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1352
1353 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1354                     unsigned long len)
1355 {
1356         gfn_t gfn = gpa >> PAGE_SHIFT;
1357         int seg;
1358         int offset = offset_in_page(gpa);
1359         int ret;
1360
1361         while ((seg = next_segment(len, offset)) != 0) {
1362                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1363                 if (ret < 0)
1364                         return ret;
1365                 offset = 0;
1366                 len -= seg;
1367                 data += seg;
1368                 ++gfn;
1369         }
1370         return 0;
1371 }
1372
1373 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1374                               gpa_t gpa)
1375 {
1376         struct kvm_memslots *slots = kvm_memslots(kvm);
1377         int offset = offset_in_page(gpa);
1378         gfn_t gfn = gpa >> PAGE_SHIFT;
1379
1380         ghc->gpa = gpa;
1381         ghc->generation = slots->generation;
1382         ghc->memslot = __gfn_to_memslot(slots, gfn);
1383         ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1384         if (!kvm_is_error_hva(ghc->hva))
1385                 ghc->hva += offset;
1386         else
1387                 return -EFAULT;
1388
1389         return 0;
1390 }
1391 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1392
1393 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1394                            void *data, unsigned long len)
1395 {
1396         struct kvm_memslots *slots = kvm_memslots(kvm);
1397         int r;
1398
1399         if (slots->generation != ghc->generation)
1400                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1401
1402         if (kvm_is_error_hva(ghc->hva))
1403                 return -EFAULT;
1404
1405         r = copy_to_user((void __user *)ghc->hva, data, len);
1406         if (r)
1407                 return -EFAULT;
1408         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1409
1410         return 0;
1411 }
1412 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1413
1414 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1415 {
1416         return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1417                                     offset, len);
1418 }
1419 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1420
1421 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1422 {
1423         gfn_t gfn = gpa >> PAGE_SHIFT;
1424         int seg;
1425         int offset = offset_in_page(gpa);
1426         int ret;
1427
1428         while ((seg = next_segment(len, offset)) != 0) {
1429                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1430                 if (ret < 0)
1431                         return ret;
1432                 offset = 0;
1433                 len -= seg;
1434                 ++gfn;
1435         }
1436         return 0;
1437 }
1438 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1439
1440 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1441                              gfn_t gfn)
1442 {
1443         if (memslot && memslot->dirty_bitmap) {
1444                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1445
1446                 __set_bit_le(rel_gfn, memslot->dirty_bitmap);
1447         }
1448 }
1449
1450 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1451 {
1452         struct kvm_memory_slot *memslot;
1453
1454         memslot = gfn_to_memslot(kvm, gfn);
1455         mark_page_dirty_in_slot(kvm, memslot, gfn);
1456 }
1457
1458 /*
1459  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1460  */
1461 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1462 {
1463         DEFINE_WAIT(wait);
1464
1465         for (;;) {
1466                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1467
1468                 if (kvm_arch_vcpu_runnable(vcpu)) {
1469                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1470                         break;
1471                 }
1472                 if (kvm_cpu_has_pending_timer(vcpu))
1473                         break;
1474                 if (signal_pending(current))
1475                         break;
1476
1477                 schedule();
1478         }
1479
1480         finish_wait(&vcpu->wq, &wait);
1481 }
1482
1483 void kvm_resched(struct kvm_vcpu *vcpu)
1484 {
1485         if (!need_resched())
1486                 return;
1487         cond_resched();
1488 }
1489 EXPORT_SYMBOL_GPL(kvm_resched);
1490
1491 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1492 {
1493         struct kvm *kvm = me->kvm;
1494         struct kvm_vcpu *vcpu;
1495         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1496         int yielded = 0;
1497         int pass;
1498         int i;
1499
1500         /*
1501          * We boost the priority of a VCPU that is runnable but not
1502          * currently running, because it got preempted by something
1503          * else and called schedule in __vcpu_run.  Hopefully that
1504          * VCPU is holding the lock that we need and will release it.
1505          * We approximate round-robin by starting at the last boosted VCPU.
1506          */
1507         for (pass = 0; pass < 2 && !yielded; pass++) {
1508                 kvm_for_each_vcpu(i, vcpu, kvm) {
1509                         struct task_struct *task = NULL;
1510                         struct pid *pid;
1511                         if (!pass && i < last_boosted_vcpu) {
1512                                 i = last_boosted_vcpu;
1513                                 continue;
1514                         } else if (pass && i > last_boosted_vcpu)
1515                                 break;
1516                         if (vcpu == me)
1517                                 continue;
1518                         if (waitqueue_active(&vcpu->wq))
1519                                 continue;
1520                         rcu_read_lock();
1521                         pid = rcu_dereference(vcpu->pid);
1522                         if (pid)
1523                                 task = get_pid_task(vcpu->pid, PIDTYPE_PID);
1524                         rcu_read_unlock();
1525                         if (!task)
1526                                 continue;
1527                         if (task->flags & PF_VCPU) {
1528                                 put_task_struct(task);
1529                                 continue;
1530                         }
1531                         if (yield_to(task, 1)) {
1532                                 put_task_struct(task);
1533                                 kvm->last_boosted_vcpu = i;
1534                                 yielded = 1;
1535                                 break;
1536                         }
1537                         put_task_struct(task);
1538                 }
1539         }
1540 }
1541 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1542
1543 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1544 {
1545         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1546         struct page *page;
1547
1548         if (vmf->pgoff == 0)
1549                 page = virt_to_page(vcpu->run);
1550 #ifdef CONFIG_X86
1551         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1552                 page = virt_to_page(vcpu->arch.pio_data);
1553 #endif
1554 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1555         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1556                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1557 #endif
1558         else
1559                 return VM_FAULT_SIGBUS;
1560         get_page(page);
1561         vmf->page = page;
1562         return 0;
1563 }
1564
1565 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1566         .fault = kvm_vcpu_fault,
1567 };
1568
1569 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1570 {
1571         vma->vm_ops = &kvm_vcpu_vm_ops;
1572         return 0;
1573 }
1574
1575 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1576 {
1577         struct kvm_vcpu *vcpu = filp->private_data;
1578
1579         kvm_put_kvm(vcpu->kvm);
1580         return 0;
1581 }
1582
1583 static struct file_operations kvm_vcpu_fops = {
1584         .release        = kvm_vcpu_release,
1585         .unlocked_ioctl = kvm_vcpu_ioctl,
1586         .compat_ioctl   = kvm_vcpu_ioctl,
1587         .mmap           = kvm_vcpu_mmap,
1588         .llseek         = noop_llseek,
1589 };
1590
1591 /*
1592  * Allocates an inode for the vcpu.
1593  */
1594 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1595 {
1596         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1597 }
1598
1599 /*
1600  * Creates some virtual cpus.  Good luck creating more than one.
1601  */
1602 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1603 {
1604         int r;
1605         struct kvm_vcpu *vcpu, *v;
1606
1607         vcpu = kvm_arch_vcpu_create(kvm, id);
1608         if (IS_ERR(vcpu))
1609                 return PTR_ERR(vcpu);
1610
1611         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1612
1613         r = kvm_arch_vcpu_setup(vcpu);
1614         if (r)
1615                 return r;
1616
1617         mutex_lock(&kvm->lock);
1618         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1619                 r = -EINVAL;
1620                 goto vcpu_destroy;
1621         }
1622
1623         kvm_for_each_vcpu(r, v, kvm)
1624                 if (v->vcpu_id == id) {
1625                         r = -EEXIST;
1626                         goto vcpu_destroy;
1627                 }
1628
1629         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1630
1631         /* Now it's all set up, let userspace reach it */
1632         kvm_get_kvm(kvm);
1633         r = create_vcpu_fd(vcpu);
1634         if (r < 0) {
1635                 kvm_put_kvm(kvm);
1636                 goto vcpu_destroy;
1637         }
1638
1639         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1640         smp_wmb();
1641         atomic_inc(&kvm->online_vcpus);
1642
1643 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1644         if (kvm->bsp_vcpu_id == id)
1645                 kvm->bsp_vcpu = vcpu;
1646 #endif
1647         mutex_unlock(&kvm->lock);
1648         return r;
1649
1650 vcpu_destroy:
1651         mutex_unlock(&kvm->lock);
1652         kvm_arch_vcpu_destroy(vcpu);
1653         return r;
1654 }
1655
1656 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1657 {
1658         if (sigset) {
1659                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1660                 vcpu->sigset_active = 1;
1661                 vcpu->sigset = *sigset;
1662         } else
1663                 vcpu->sigset_active = 0;
1664         return 0;
1665 }
1666
1667 static long kvm_vcpu_ioctl(struct file *filp,
1668                            unsigned int ioctl, unsigned long arg)
1669 {
1670         struct kvm_vcpu *vcpu = filp->private_data;
1671         void __user *argp = (void __user *)arg;
1672         int r;
1673         struct kvm_fpu *fpu = NULL;
1674         struct kvm_sregs *kvm_sregs = NULL;
1675
1676         if (vcpu->kvm->mm != current->mm)
1677                 return -EIO;
1678
1679 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1680         /*
1681          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1682          * so vcpu_load() would break it.
1683          */
1684         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1685                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1686 #endif
1687
1688
1689         vcpu_load(vcpu);
1690         switch (ioctl) {
1691         case KVM_RUN:
1692                 r = -EINVAL;
1693                 if (arg)
1694                         goto out;
1695                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1696                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1697                 break;
1698         case KVM_GET_REGS: {
1699                 struct kvm_regs *kvm_regs;
1700
1701                 r = -ENOMEM;
1702                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1703                 if (!kvm_regs)
1704                         goto out;
1705                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1706                 if (r)
1707                         goto out_free1;
1708                 r = -EFAULT;
1709                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1710                         goto out_free1;
1711                 r = 0;
1712 out_free1:
1713                 kfree(kvm_regs);
1714                 break;
1715         }
1716         case KVM_SET_REGS: {
1717                 struct kvm_regs *kvm_regs;
1718
1719                 r = -ENOMEM;
1720                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1721                 if (!kvm_regs)
1722                         goto out;
1723                 r = -EFAULT;
1724                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1725                         goto out_free2;
1726                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1727                 if (r)
1728                         goto out_free2;
1729                 r = 0;
1730 out_free2:
1731                 kfree(kvm_regs);
1732                 break;
1733         }
1734         case KVM_GET_SREGS: {
1735                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1736                 r = -ENOMEM;
1737                 if (!kvm_sregs)
1738                         goto out;
1739                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1740                 if (r)
1741                         goto out;
1742                 r = -EFAULT;
1743                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1744                         goto out;
1745                 r = 0;
1746                 break;
1747         }
1748         case KVM_SET_SREGS: {
1749                 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1750                 r = -ENOMEM;
1751                 if (!kvm_sregs)
1752                         goto out;
1753                 r = -EFAULT;
1754                 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1755                         goto out;
1756                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1757                 if (r)
1758                         goto out;
1759                 r = 0;
1760                 break;
1761         }
1762         case KVM_GET_MP_STATE: {
1763                 struct kvm_mp_state mp_state;
1764
1765                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1766                 if (r)
1767                         goto out;
1768                 r = -EFAULT;
1769                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1770                         goto out;
1771                 r = 0;
1772                 break;
1773         }
1774         case KVM_SET_MP_STATE: {
1775                 struct kvm_mp_state mp_state;
1776
1777                 r = -EFAULT;
1778                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1779                         goto out;
1780                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1781                 if (r)
1782                         goto out;
1783                 r = 0;
1784                 break;
1785         }
1786         case KVM_TRANSLATE: {
1787                 struct kvm_translation tr;
1788
1789                 r = -EFAULT;
1790                 if (copy_from_user(&tr, argp, sizeof tr))
1791                         goto out;
1792                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1793                 if (r)
1794                         goto out;
1795                 r = -EFAULT;
1796                 if (copy_to_user(argp, &tr, sizeof tr))
1797                         goto out;
1798                 r = 0;
1799                 break;
1800         }
1801         case KVM_SET_GUEST_DEBUG: {
1802                 struct kvm_guest_debug dbg;
1803
1804                 r = -EFAULT;
1805                 if (copy_from_user(&dbg, argp, sizeof dbg))
1806                         goto out;
1807                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1808                 if (r)
1809                         goto out;
1810                 r = 0;
1811                 break;
1812         }
1813         case KVM_SET_SIGNAL_MASK: {
1814                 struct kvm_signal_mask __user *sigmask_arg = argp;
1815                 struct kvm_signal_mask kvm_sigmask;
1816                 sigset_t sigset, *p;
1817
1818                 p = NULL;
1819                 if (argp) {
1820                         r = -EFAULT;
1821                         if (copy_from_user(&kvm_sigmask, argp,
1822                                            sizeof kvm_sigmask))
1823                                 goto out;
1824                         r = -EINVAL;
1825                         if (kvm_sigmask.len != sizeof sigset)
1826                                 goto out;
1827                         r = -EFAULT;
1828                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1829                                            sizeof sigset))
1830                                 goto out;
1831                         p = &sigset;
1832                 }
1833                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
1834                 break;
1835         }
1836         case KVM_GET_FPU: {
1837                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1838                 r = -ENOMEM;
1839                 if (!fpu)
1840                         goto out;
1841                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1842                 if (r)
1843                         goto out;
1844                 r = -EFAULT;
1845                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1846                         goto out;
1847                 r = 0;
1848                 break;
1849         }
1850         case KVM_SET_FPU: {
1851                 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1852                 r = -ENOMEM;
1853                 if (!fpu)
1854                         goto out;
1855                 r = -EFAULT;
1856                 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1857                         goto out;
1858                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1859                 if (r)
1860                         goto out;
1861                 r = 0;
1862                 break;
1863         }
1864         default:
1865                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1866         }
1867 out:
1868         vcpu_put(vcpu);
1869         kfree(fpu);
1870         kfree(kvm_sregs);
1871         return r;
1872 }
1873
1874 static long kvm_vm_ioctl(struct file *filp,
1875                            unsigned int ioctl, unsigned long arg)
1876 {
1877         struct kvm *kvm = filp->private_data;
1878         void __user *argp = (void __user *)arg;
1879         int r;
1880
1881         if (kvm->mm != current->mm)
1882                 return -EIO;
1883         switch (ioctl) {
1884         case KVM_CREATE_VCPU:
1885                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1886                 if (r < 0)
1887                         goto out;
1888                 break;
1889         case KVM_SET_USER_MEMORY_REGION: {
1890                 struct kvm_userspace_memory_region kvm_userspace_mem;
1891
1892                 r = -EFAULT;
1893                 if (copy_from_user(&kvm_userspace_mem, argp,
1894                                                 sizeof kvm_userspace_mem))
1895                         goto out;
1896
1897                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1898                 if (r)
1899                         goto out;
1900                 break;
1901         }
1902         case KVM_GET_DIRTY_LOG: {
1903                 struct kvm_dirty_log log;
1904
1905                 r = -EFAULT;
1906                 if (copy_from_user(&log, argp, sizeof log))
1907                         goto out;
1908                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1909                 if (r)
1910                         goto out;
1911                 break;
1912         }
1913 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1914         case KVM_REGISTER_COALESCED_MMIO: {
1915                 struct kvm_coalesced_mmio_zone zone;
1916                 r = -EFAULT;
1917                 if (copy_from_user(&zone, argp, sizeof zone))
1918                         goto out;
1919                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1920                 if (r)
1921                         goto out;
1922                 r = 0;
1923                 break;
1924         }
1925         case KVM_UNREGISTER_COALESCED_MMIO: {
1926                 struct kvm_coalesced_mmio_zone zone;
1927                 r = -EFAULT;
1928                 if (copy_from_user(&zone, argp, sizeof zone))
1929                         goto out;
1930                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1931                 if (r)
1932                         goto out;
1933                 r = 0;
1934                 break;
1935         }
1936 #endif
1937         case KVM_IRQFD: {
1938                 struct kvm_irqfd data;
1939
1940                 r = -EFAULT;
1941                 if (copy_from_user(&data, argp, sizeof data))
1942                         goto out;
1943                 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
1944                 break;
1945         }
1946         case KVM_IOEVENTFD: {
1947                 struct kvm_ioeventfd data;
1948
1949                 r = -EFAULT;
1950                 if (copy_from_user(&data, argp, sizeof data))
1951                         goto out;
1952                 r = kvm_ioeventfd(kvm, &data);
1953                 break;
1954         }
1955 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1956         case KVM_SET_BOOT_CPU_ID:
1957                 r = 0;
1958                 mutex_lock(&kvm->lock);
1959                 if (atomic_read(&kvm->online_vcpus) != 0)
1960                         r = -EBUSY;
1961                 else
1962                         kvm->bsp_vcpu_id = arg;
1963                 mutex_unlock(&kvm->lock);
1964                 break;
1965 #endif
1966         default:
1967                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1968                 if (r == -ENOTTY)
1969                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
1970         }
1971 out:
1972         return r;
1973 }
1974
1975 #ifdef CONFIG_COMPAT
1976 struct compat_kvm_dirty_log {
1977         __u32 slot;
1978         __u32 padding1;
1979         union {
1980                 compat_uptr_t dirty_bitmap; /* one bit per page */
1981                 __u64 padding2;
1982         };
1983 };
1984
1985 static long kvm_vm_compat_ioctl(struct file *filp,
1986                            unsigned int ioctl, unsigned long arg)
1987 {
1988         struct kvm *kvm = filp->private_data;
1989         int r;
1990
1991         if (kvm->mm != current->mm)
1992                 return -EIO;
1993         switch (ioctl) {
1994         case KVM_GET_DIRTY_LOG: {
1995                 struct compat_kvm_dirty_log compat_log;
1996                 struct kvm_dirty_log log;
1997
1998                 r = -EFAULT;
1999                 if (copy_from_user(&compat_log, (void __user *)arg,
2000                                    sizeof(compat_log)))
2001                         goto out;
2002                 log.slot         = compat_log.slot;
2003                 log.padding1     = compat_log.padding1;
2004                 log.padding2     = compat_log.padding2;
2005                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2006
2007                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2008                 if (r)
2009                         goto out;
2010                 break;
2011         }
2012         default:
2013                 r = kvm_vm_ioctl(filp, ioctl, arg);
2014         }
2015
2016 out:
2017         return r;
2018 }
2019 #endif
2020
2021 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2022 {
2023         struct page *page[1];
2024         unsigned long addr;
2025         int npages;
2026         gfn_t gfn = vmf->pgoff;
2027         struct kvm *kvm = vma->vm_file->private_data;
2028
2029         addr = gfn_to_hva(kvm, gfn);
2030         if (kvm_is_error_hva(addr))
2031                 return VM_FAULT_SIGBUS;
2032
2033         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2034                                 NULL);
2035         if (unlikely(npages != 1))
2036                 return VM_FAULT_SIGBUS;
2037
2038         vmf->page = page[0];
2039         return 0;
2040 }
2041
2042 static const struct vm_operations_struct kvm_vm_vm_ops = {
2043         .fault = kvm_vm_fault,
2044 };
2045
2046 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2047 {
2048         vma->vm_ops = &kvm_vm_vm_ops;
2049         return 0;
2050 }
2051
2052 static struct file_operations kvm_vm_fops = {
2053         .release        = kvm_vm_release,
2054         .unlocked_ioctl = kvm_vm_ioctl,
2055 #ifdef CONFIG_COMPAT
2056         .compat_ioctl   = kvm_vm_compat_ioctl,
2057 #endif
2058         .mmap           = kvm_vm_mmap,
2059         .llseek         = noop_llseek,
2060 };
2061
2062 static int kvm_dev_ioctl_create_vm(void)
2063 {
2064         int r;
2065         struct kvm *kvm;
2066
2067         kvm = kvm_create_vm();
2068         if (IS_ERR(kvm))
2069                 return PTR_ERR(kvm);
2070 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2071         r = kvm_coalesced_mmio_init(kvm);
2072         if (r < 0) {
2073                 kvm_put_kvm(kvm);
2074                 return r;
2075         }
2076 #endif
2077         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2078         if (r < 0)
2079                 kvm_put_kvm(kvm);
2080
2081         return r;
2082 }
2083
2084 static long kvm_dev_ioctl_check_extension_generic(long arg)
2085 {
2086         switch (arg) {
2087         case KVM_CAP_USER_MEMORY:
2088         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2089         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2090 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2091         case KVM_CAP_SET_BOOT_CPU_ID:
2092 #endif
2093         case KVM_CAP_INTERNAL_ERROR_DATA:
2094                 return 1;
2095 #ifdef CONFIG_HAVE_KVM_IRQCHIP
2096         case KVM_CAP_IRQ_ROUTING:
2097                 return KVM_MAX_IRQ_ROUTES;
2098 #endif
2099         default:
2100                 break;
2101         }
2102         return kvm_dev_ioctl_check_extension(arg);
2103 }
2104
2105 static long kvm_dev_ioctl(struct file *filp,
2106                           unsigned int ioctl, unsigned long arg)
2107 {
2108         long r = -EINVAL;
2109
2110         switch (ioctl) {
2111         case KVM_GET_API_VERSION:
2112                 r = -EINVAL;
2113                 if (arg)
2114                         goto out;
2115                 r = KVM_API_VERSION;
2116                 break;
2117         case KVM_CREATE_VM:
2118                 r = -EINVAL;
2119                 if (arg)
2120                         goto out;
2121                 r = kvm_dev_ioctl_create_vm();
2122                 break;
2123         case KVM_CHECK_EXTENSION:
2124                 r = kvm_dev_ioctl_check_extension_generic(arg);
2125                 break;
2126         case KVM_GET_VCPU_MMAP_SIZE:
2127                 r = -EINVAL;
2128                 if (arg)
2129                         goto out;
2130                 r = PAGE_SIZE;     /* struct kvm_run */
2131 #ifdef CONFIG_X86
2132                 r += PAGE_SIZE;    /* pio data page */
2133 #endif
2134 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2135                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2136 #endif
2137                 break;
2138         case KVM_TRACE_ENABLE:
2139         case KVM_TRACE_PAUSE:
2140         case KVM_TRACE_DISABLE:
2141                 r = -EOPNOTSUPP;
2142                 break;
2143         default:
2144                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2145         }
2146 out:
2147         return r;
2148 }
2149
2150 static struct file_operations kvm_chardev_ops = {
2151         .unlocked_ioctl = kvm_dev_ioctl,
2152         .compat_ioctl   = kvm_dev_ioctl,
2153         .llseek         = noop_llseek,
2154 };
2155
2156 static struct miscdevice kvm_dev = {
2157         KVM_MINOR,
2158         "kvm",
2159         &kvm_chardev_ops,
2160 };
2161
2162 static void hardware_enable_nolock(void *junk)
2163 {
2164         int cpu = raw_smp_processor_id();
2165         int r;
2166
2167         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2168                 return;
2169
2170         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2171
2172         r = kvm_arch_hardware_enable(NULL);
2173
2174         if (r) {
2175                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2176                 atomic_inc(&hardware_enable_failed);
2177                 printk(KERN_INFO "kvm: enabling virtualization on "
2178                                  "CPU%d failed\n", cpu);
2179         }
2180 }
2181
2182 static void hardware_enable(void *junk)
2183 {
2184         raw_spin_lock(&kvm_lock);
2185         hardware_enable_nolock(junk);
2186         raw_spin_unlock(&kvm_lock);
2187 }
2188
2189 static void hardware_disable_nolock(void *junk)
2190 {
2191         int cpu = raw_smp_processor_id();
2192
2193         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2194                 return;
2195         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2196         kvm_arch_hardware_disable(NULL);
2197 }
2198
2199 static void hardware_disable(void *junk)
2200 {
2201         raw_spin_lock(&kvm_lock);
2202         hardware_disable_nolock(junk);
2203         raw_spin_unlock(&kvm_lock);
2204 }
2205
2206 static void hardware_disable_all_nolock(void)
2207 {
2208         BUG_ON(!kvm_usage_count);
2209
2210         kvm_usage_count--;
2211         if (!kvm_usage_count)
2212                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2213 }
2214
2215 static void hardware_disable_all(void)
2216 {
2217         raw_spin_lock(&kvm_lock);
2218         hardware_disable_all_nolock();
2219         raw_spin_unlock(&kvm_lock);
2220 }
2221
2222 static int hardware_enable_all(void)
2223 {
2224         int r = 0;
2225
2226         raw_spin_lock(&kvm_lock);
2227
2228         kvm_usage_count++;
2229         if (kvm_usage_count == 1) {
2230                 atomic_set(&hardware_enable_failed, 0);
2231                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2232
2233                 if (atomic_read(&hardware_enable_failed)) {
2234                         hardware_disable_all_nolock();
2235                         r = -EBUSY;
2236                 }
2237         }
2238
2239         raw_spin_unlock(&kvm_lock);
2240
2241         return r;
2242 }
2243
2244 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2245                            void *v)
2246 {
2247         int cpu = (long)v;
2248
2249         if (!kvm_usage_count)
2250                 return NOTIFY_OK;
2251
2252         val &= ~CPU_TASKS_FROZEN;
2253         switch (val) {
2254         case CPU_DYING:
2255                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2256                        cpu);
2257                 hardware_disable(NULL);
2258                 break;
2259         case CPU_STARTING:
2260                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2261                        cpu);
2262                 hardware_enable(NULL);
2263                 break;
2264         }
2265         return NOTIFY_OK;
2266 }
2267
2268
2269 asmlinkage void kvm_spurious_fault(void)
2270 {
2271         /* Fault while not rebooting.  We want the trace. */
2272         BUG();
2273 }
2274 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
2275
2276 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2277                       void *v)
2278 {
2279         /*
2280          * Some (well, at least mine) BIOSes hang on reboot if
2281          * in vmx root mode.
2282          *
2283          * And Intel TXT required VMX off for all cpu when system shutdown.
2284          */
2285         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2286         kvm_rebooting = true;
2287         on_each_cpu(hardware_disable_nolock, NULL, 1);
2288         return NOTIFY_OK;
2289 }
2290
2291 static struct notifier_block kvm_reboot_notifier = {
2292         .notifier_call = kvm_reboot,
2293         .priority = 0,
2294 };
2295
2296 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2297 {
2298         int i;
2299
2300         for (i = 0; i < bus->dev_count; i++) {
2301                 struct kvm_io_device *pos = bus->devs[i];
2302
2303                 kvm_iodevice_destructor(pos);
2304         }
2305         kfree(bus);
2306 }
2307
2308 /* kvm_io_bus_write - called under kvm->slots_lock */
2309 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2310                      int len, const void *val)
2311 {
2312         int i;
2313         struct kvm_io_bus *bus;
2314
2315         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2316         for (i = 0; i < bus->dev_count; i++)
2317                 if (!kvm_iodevice_write(bus->devs[i], addr, len, val))
2318                         return 0;
2319         return -EOPNOTSUPP;
2320 }
2321
2322 /* kvm_io_bus_read - called under kvm->slots_lock */
2323 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2324                     int len, void *val)
2325 {
2326         int i;
2327         struct kvm_io_bus *bus;
2328
2329         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2330         for (i = 0; i < bus->dev_count; i++)
2331                 if (!kvm_iodevice_read(bus->devs[i], addr, len, val))
2332                         return 0;
2333         return -EOPNOTSUPP;
2334 }
2335
2336 /* Caller must hold slots_lock. */
2337 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2338                             struct kvm_io_device *dev)
2339 {
2340         struct kvm_io_bus *new_bus, *bus;
2341
2342         bus = kvm->buses[bus_idx];
2343         if (bus->dev_count > NR_IOBUS_DEVS-1)
2344                 return -ENOSPC;
2345
2346         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2347         if (!new_bus)
2348                 return -ENOMEM;
2349         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2350         new_bus->devs[new_bus->dev_count++] = dev;
2351         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2352         synchronize_srcu_expedited(&kvm->srcu);
2353         kfree(bus);
2354
2355         return 0;
2356 }
2357
2358 /* Caller must hold slots_lock. */
2359 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2360                               struct kvm_io_device *dev)
2361 {
2362         int i, r;
2363         struct kvm_io_bus *new_bus, *bus;
2364
2365         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2366         if (!new_bus)
2367                 return -ENOMEM;
2368
2369         bus = kvm->buses[bus_idx];
2370         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2371
2372         r = -ENOENT;
2373         for (i = 0; i < new_bus->dev_count; i++)
2374                 if (new_bus->devs[i] == dev) {
2375                         r = 0;
2376                         new_bus->devs[i] = new_bus->devs[--new_bus->dev_count];
2377                         break;
2378                 }
2379
2380         if (r) {
2381                 kfree(new_bus);
2382                 return r;
2383         }
2384
2385         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2386         synchronize_srcu_expedited(&kvm->srcu);
2387         kfree(bus);
2388         return r;
2389 }
2390
2391 static struct notifier_block kvm_cpu_notifier = {
2392         .notifier_call = kvm_cpu_hotplug,
2393 };
2394
2395 static int vm_stat_get(void *_offset, u64 *val)
2396 {
2397         unsigned offset = (long)_offset;
2398         struct kvm *kvm;
2399
2400         *val = 0;
2401         raw_spin_lock(&kvm_lock);
2402         list_for_each_entry(kvm, &vm_list, vm_list)
2403                 *val += *(u32 *)((void *)kvm + offset);
2404         raw_spin_unlock(&kvm_lock);
2405         return 0;
2406 }
2407
2408 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2409
2410 static int vcpu_stat_get(void *_offset, u64 *val)
2411 {
2412         unsigned offset = (long)_offset;
2413         struct kvm *kvm;
2414         struct kvm_vcpu *vcpu;
2415         int i;
2416
2417         *val = 0;
2418         raw_spin_lock(&kvm_lock);
2419         list_for_each_entry(kvm, &vm_list, vm_list)
2420                 kvm_for_each_vcpu(i, vcpu, kvm)
2421                         *val += *(u32 *)((void *)vcpu + offset);
2422
2423         raw_spin_unlock(&kvm_lock);
2424         return 0;
2425 }
2426
2427 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2428
2429 static const struct file_operations *stat_fops[] = {
2430         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2431         [KVM_STAT_VM]   = &vm_stat_fops,
2432 };
2433
2434 static void kvm_init_debug(void)
2435 {
2436         struct kvm_stats_debugfs_item *p;
2437
2438         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2439         for (p = debugfs_entries; p->name; ++p)
2440                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2441                                                 (void *)(long)p->offset,
2442                                                 stat_fops[p->kind]);
2443 }
2444
2445 static void kvm_exit_debug(void)
2446 {
2447         struct kvm_stats_debugfs_item *p;
2448
2449         for (p = debugfs_entries; p->name; ++p)
2450                 debugfs_remove(p->dentry);
2451         debugfs_remove(kvm_debugfs_dir);
2452 }
2453
2454 static int kvm_suspend(void)
2455 {
2456         if (kvm_usage_count)
2457                 hardware_disable_nolock(NULL);
2458         return 0;
2459 }
2460
2461 static void kvm_resume(void)
2462 {
2463         if (kvm_usage_count) {
2464                 WARN_ON(raw_spin_is_locked(&kvm_lock));
2465                 hardware_enable_nolock(NULL);
2466         }
2467 }
2468
2469 static struct syscore_ops kvm_syscore_ops = {
2470         .suspend = kvm_suspend,
2471         .resume = kvm_resume,
2472 };
2473
2474 struct page *bad_page;
2475 pfn_t bad_pfn;
2476
2477 static inline
2478 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2479 {
2480         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2481 }
2482
2483 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2484 {
2485         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2486
2487         kvm_arch_vcpu_load(vcpu, cpu);
2488 }
2489
2490 static void kvm_sched_out(struct preempt_notifier *pn,
2491                           struct task_struct *next)
2492 {
2493         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2494
2495         kvm_arch_vcpu_put(vcpu);
2496 }
2497
2498 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2499                   struct module *module)
2500 {
2501         int r;
2502         int cpu;
2503
2504         r = kvm_arch_init(opaque);
2505         if (r)
2506                 goto out_fail;
2507
2508         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2509
2510         if (bad_page == NULL) {
2511                 r = -ENOMEM;
2512                 goto out;
2513         }
2514
2515         bad_pfn = page_to_pfn(bad_page);
2516
2517         hwpoison_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2518
2519         if (hwpoison_page == NULL) {
2520                 r = -ENOMEM;
2521                 goto out_free_0;
2522         }
2523
2524         hwpoison_pfn = page_to_pfn(hwpoison_page);
2525
2526         fault_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2527
2528         if (fault_page == NULL) {
2529                 r = -ENOMEM;
2530                 goto out_free_0;
2531         }
2532
2533         fault_pfn = page_to_pfn(fault_page);
2534
2535         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2536                 r = -ENOMEM;
2537                 goto out_free_0;
2538         }
2539
2540         r = kvm_arch_hardware_setup();
2541         if (r < 0)
2542                 goto out_free_0a;
2543
2544         for_each_online_cpu(cpu) {
2545                 smp_call_function_single(cpu,
2546                                 kvm_arch_check_processor_compat,
2547                                 &r, 1);
2548                 if (r < 0)
2549                         goto out_free_1;
2550         }
2551
2552         r = register_cpu_notifier(&kvm_cpu_notifier);
2553         if (r)
2554                 goto out_free_2;
2555         register_reboot_notifier(&kvm_reboot_notifier);
2556
2557         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2558         if (!vcpu_align)
2559                 vcpu_align = __alignof__(struct kvm_vcpu);
2560         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2561                                            0, NULL);
2562         if (!kvm_vcpu_cache) {
2563                 r = -ENOMEM;
2564                 goto out_free_3;
2565         }
2566
2567         r = kvm_async_pf_init();
2568         if (r)
2569                 goto out_free;
2570
2571         kvm_chardev_ops.owner = module;
2572         kvm_vm_fops.owner = module;
2573         kvm_vcpu_fops.owner = module;
2574
2575         r = misc_register(&kvm_dev);
2576         if (r) {
2577                 printk(KERN_ERR "kvm: misc device register failed\n");
2578                 goto out_unreg;
2579         }
2580
2581         register_syscore_ops(&kvm_syscore_ops);
2582
2583         kvm_preempt_ops.sched_in = kvm_sched_in;
2584         kvm_preempt_ops.sched_out = kvm_sched_out;
2585
2586         kvm_init_debug();
2587
2588         return 0;
2589
2590 out_unreg:
2591         kvm_async_pf_deinit();
2592 out_free:
2593         kmem_cache_destroy(kvm_vcpu_cache);
2594 out_free_3:
2595         unregister_reboot_notifier(&kvm_reboot_notifier);
2596         unregister_cpu_notifier(&kvm_cpu_notifier);
2597 out_free_2:
2598 out_free_1:
2599         kvm_arch_hardware_unsetup();
2600 out_free_0a:
2601         free_cpumask_var(cpus_hardware_enabled);
2602 out_free_0:
2603         if (fault_page)
2604                 __free_page(fault_page);
2605         if (hwpoison_page)
2606                 __free_page(hwpoison_page);
2607         __free_page(bad_page);
2608 out:
2609         kvm_arch_exit();
2610 out_fail:
2611         return r;
2612 }
2613 EXPORT_SYMBOL_GPL(kvm_init);
2614
2615 void kvm_exit(void)
2616 {
2617         kvm_exit_debug();
2618         misc_deregister(&kvm_dev);
2619         kmem_cache_destroy(kvm_vcpu_cache);
2620         kvm_async_pf_deinit();
2621         unregister_syscore_ops(&kvm_syscore_ops);
2622         unregister_reboot_notifier(&kvm_reboot_notifier);
2623         unregister_cpu_notifier(&kvm_cpu_notifier);
2624         on_each_cpu(hardware_disable_nolock, NULL, 1);
2625         kvm_arch_hardware_unsetup();
2626         kvm_arch_exit();
2627         free_cpumask_var(cpus_hardware_enabled);
2628         __free_page(hwpoison_page);
2629         __free_page(bad_page);
2630 }
2631 EXPORT_SYMBOL_GPL(kvm_exit);