~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/virt/kvm/kvm_main.c

Version: ~ [ linux-5.4-rc3 ] ~ [ linux-5.3.6 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.79 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.149 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.196 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.196 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.75 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-3.9.11 ] ~ [ linux-3.8.13 ] ~ [ linux-3.7.10 ] ~ [ linux-3.6.11 ] ~ [ linux-3.5.7 ] ~ [ linux-3.4.113 ] ~ [ linux-3.3.8 ] ~ [ linux-3.2.102 ] ~ [ linux-3.1.10 ] ~ [ linux-3.0.101 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  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 <kvm/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 #include <linux/sort.h>
 51 #include <linux/bsearch.h>
 52 
 53 #include <asm/processor.h>
 54 #include <asm/io.h>
 55 #include <asm/ioctl.h>
 56 #include <asm/uaccess.h>
 57 #include <asm/pgtable.h>
 58 
 59 #include "coalesced_mmio.h"
 60 #include "async_pf.h"
 61 #include "vfio.h"
 62 
 63 #define CREATE_TRACE_POINTS
 64 #include <trace/events/kvm.h>
 65 
 66 MODULE_AUTHOR("Qumranet");
 67 MODULE_LICENSE("GPL");
 68 
 69 static unsigned int halt_poll_ns;
 70 module_param(halt_poll_ns, uint, S_IRUGO | S_IWUSR);
 71 
 72 /*
 73  * Ordering of locks:
 74  *
 75  *      kvm->lock --> kvm->slots_lock --> kvm->irq_lock
 76  */
 77 
 78 DEFINE_SPINLOCK(kvm_lock);
 79 static DEFINE_RAW_SPINLOCK(kvm_count_lock);
 80 LIST_HEAD(vm_list);
 81 
 82 static cpumask_var_t cpus_hardware_enabled;
 83 static int kvm_usage_count;
 84 static atomic_t hardware_enable_failed;
 85 
 86 struct kmem_cache *kvm_vcpu_cache;
 87 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
 88 
 89 static __read_mostly struct preempt_ops kvm_preempt_ops;
 90 
 91 struct dentry *kvm_debugfs_dir;
 92 EXPORT_SYMBOL_GPL(kvm_debugfs_dir);
 93 
 94 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
 95                            unsigned long arg);
 96 #ifdef CONFIG_KVM_COMPAT
 97 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
 98                                   unsigned long arg);
 99 #endif
100 static int hardware_enable_all(void);
101 static void hardware_disable_all(void);
102 
103 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
104 
105 static void kvm_release_pfn_dirty(pfn_t pfn);
106 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot, gfn_t gfn);
107 
108 __visible bool kvm_rebooting;
109 EXPORT_SYMBOL_GPL(kvm_rebooting);
110 
111 static bool largepages_enabled = true;
112 
113 bool kvm_is_reserved_pfn(pfn_t pfn)
114 {
115         if (pfn_valid(pfn))
116                 return PageReserved(pfn_to_page(pfn));
117 
118         return true;
119 }
120 
121 /*
122  * Switches to specified vcpu, until a matching vcpu_put()
123  */
124 int vcpu_load(struct kvm_vcpu *vcpu)
125 {
126         int cpu;
127 
128         if (mutex_lock_killable(&vcpu->mutex))
129                 return -EINTR;
130         cpu = get_cpu();
131         preempt_notifier_register(&vcpu->preempt_notifier);
132         kvm_arch_vcpu_load(vcpu, cpu);
133         put_cpu();
134         return 0;
135 }
136 
137 void vcpu_put(struct kvm_vcpu *vcpu)
138 {
139         preempt_disable();
140         kvm_arch_vcpu_put(vcpu);
141         preempt_notifier_unregister(&vcpu->preempt_notifier);
142         preempt_enable();
143         mutex_unlock(&vcpu->mutex);
144 }
145 
146 static void ack_flush(void *_completed)
147 {
148 }
149 
150 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
151 {
152         int i, cpu, me;
153         cpumask_var_t cpus;
154         bool called = true;
155         struct kvm_vcpu *vcpu;
156 
157         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
158 
159         me = get_cpu();
160         kvm_for_each_vcpu(i, vcpu, kvm) {
161                 kvm_make_request(req, vcpu);
162                 cpu = vcpu->cpu;
163 
164                 /* Set ->requests bit before we read ->mode */
165                 smp_mb();
166 
167                 if (cpus != NULL && cpu != -1 && cpu != me &&
168                       kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
169                         cpumask_set_cpu(cpu, cpus);
170         }
171         if (unlikely(cpus == NULL))
172                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
173         else if (!cpumask_empty(cpus))
174                 smp_call_function_many(cpus, ack_flush, NULL, 1);
175         else
176                 called = false;
177         put_cpu();
178         free_cpumask_var(cpus);
179         return called;
180 }
181 
182 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
183 void kvm_flush_remote_tlbs(struct kvm *kvm)
184 {
185         long dirty_count = kvm->tlbs_dirty;
186 
187         smp_mb();
188         if (kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
189                 ++kvm->stat.remote_tlb_flush;
190         cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
191 }
192 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
193 #endif
194 
195 void kvm_reload_remote_mmus(struct kvm *kvm)
196 {
197         kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
198 }
199 
200 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
201 {
202         kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
203 }
204 
205 void kvm_make_scan_ioapic_request(struct kvm *kvm)
206 {
207         kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
208 }
209 
210 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
211 {
212         struct page *page;
213         int r;
214 
215         mutex_init(&vcpu->mutex);
216         vcpu->cpu = -1;
217         vcpu->kvm = kvm;
218         vcpu->vcpu_id = id;
219         vcpu->pid = NULL;
220         init_waitqueue_head(&vcpu->wq);
221         kvm_async_pf_vcpu_init(vcpu);
222 
223         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
224         if (!page) {
225                 r = -ENOMEM;
226                 goto fail;
227         }
228         vcpu->run = page_address(page);
229 
230         kvm_vcpu_set_in_spin_loop(vcpu, false);
231         kvm_vcpu_set_dy_eligible(vcpu, false);
232         vcpu->preempted = false;
233 
234         r = kvm_arch_vcpu_init(vcpu);
235         if (r < 0)
236                 goto fail_free_run;
237         return 0;
238 
239 fail_free_run:
240         free_page((unsigned long)vcpu->run);
241 fail:
242         return r;
243 }
244 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
245 
246 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
247 {
248         put_pid(vcpu->pid);
249         kvm_arch_vcpu_uninit(vcpu);
250         free_page((unsigned long)vcpu->run);
251 }
252 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
253 
254 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
255 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
256 {
257         return container_of(mn, struct kvm, mmu_notifier);
258 }
259 
260 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
261                                              struct mm_struct *mm,
262                                              unsigned long address)
263 {
264         struct kvm *kvm = mmu_notifier_to_kvm(mn);
265         int need_tlb_flush, idx;
266 
267         /*
268          * When ->invalidate_page runs, the linux pte has been zapped
269          * already but the page is still allocated until
270          * ->invalidate_page returns. So if we increase the sequence
271          * here the kvm page fault will notice if the spte can't be
272          * established because the page is going to be freed. If
273          * instead the kvm page fault establishes the spte before
274          * ->invalidate_page runs, kvm_unmap_hva will release it
275          * before returning.
276          *
277          * The sequence increase only need to be seen at spin_unlock
278          * time, and not at spin_lock time.
279          *
280          * Increasing the sequence after the spin_unlock would be
281          * unsafe because the kvm page fault could then establish the
282          * pte after kvm_unmap_hva returned, without noticing the page
283          * is going to be freed.
284          */
285         idx = srcu_read_lock(&kvm->srcu);
286         spin_lock(&kvm->mmu_lock);
287 
288         kvm->mmu_notifier_seq++;
289         need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
290         /* we've to flush the tlb before the pages can be freed */
291         if (need_tlb_flush)
292                 kvm_flush_remote_tlbs(kvm);
293 
294         spin_unlock(&kvm->mmu_lock);
295 
296         kvm_arch_mmu_notifier_invalidate_page(kvm, address);
297 
298         srcu_read_unlock(&kvm->srcu, idx);
299 }
300 
301 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
302                                         struct mm_struct *mm,
303                                         unsigned long address,
304                                         pte_t pte)
305 {
306         struct kvm *kvm = mmu_notifier_to_kvm(mn);
307         int idx;
308 
309         idx = srcu_read_lock(&kvm->srcu);
310         spin_lock(&kvm->mmu_lock);
311         kvm->mmu_notifier_seq++;
312         kvm_set_spte_hva(kvm, address, pte);
313         spin_unlock(&kvm->mmu_lock);
314         srcu_read_unlock(&kvm->srcu, idx);
315 }
316 
317 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
318                                                     struct mm_struct *mm,
319                                                     unsigned long start,
320                                                     unsigned long end)
321 {
322         struct kvm *kvm = mmu_notifier_to_kvm(mn);
323         int need_tlb_flush = 0, idx;
324 
325         idx = srcu_read_lock(&kvm->srcu);
326         spin_lock(&kvm->mmu_lock);
327         /*
328          * The count increase must become visible at unlock time as no
329          * spte can be established without taking the mmu_lock and
330          * count is also read inside the mmu_lock critical section.
331          */
332         kvm->mmu_notifier_count++;
333         need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
334         need_tlb_flush |= kvm->tlbs_dirty;
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         spin_unlock(&kvm->mmu_lock);
340         srcu_read_unlock(&kvm->srcu, idx);
341 }
342 
343 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
344                                                   struct mm_struct *mm,
345                                                   unsigned long start,
346                                                   unsigned long end)
347 {
348         struct kvm *kvm = mmu_notifier_to_kvm(mn);
349 
350         spin_lock(&kvm->mmu_lock);
351         /*
352          * This sequence increase will notify the kvm page fault that
353          * the page that is going to be mapped in the spte could have
354          * been freed.
355          */
356         kvm->mmu_notifier_seq++;
357         smp_wmb();
358         /*
359          * The above sequence increase must be visible before the
360          * below count decrease, which is ensured by the smp_wmb above
361          * in conjunction with the smp_rmb in mmu_notifier_retry().
362          */
363         kvm->mmu_notifier_count--;
364         spin_unlock(&kvm->mmu_lock);
365 
366         BUG_ON(kvm->mmu_notifier_count < 0);
367 }
368 
369 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
370                                               struct mm_struct *mm,
371                                               unsigned long start,
372                                               unsigned long end)
373 {
374         struct kvm *kvm = mmu_notifier_to_kvm(mn);
375         int young, idx;
376 
377         idx = srcu_read_lock(&kvm->srcu);
378         spin_lock(&kvm->mmu_lock);
379 
380         young = kvm_age_hva(kvm, start, end);
381         if (young)
382                 kvm_flush_remote_tlbs(kvm);
383 
384         spin_unlock(&kvm->mmu_lock);
385         srcu_read_unlock(&kvm->srcu, idx);
386 
387         return young;
388 }
389 
390 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
391                                        struct mm_struct *mm,
392                                        unsigned long address)
393 {
394         struct kvm *kvm = mmu_notifier_to_kvm(mn);
395         int young, idx;
396 
397         idx = srcu_read_lock(&kvm->srcu);
398         spin_lock(&kvm->mmu_lock);
399         young = kvm_test_age_hva(kvm, address);
400         spin_unlock(&kvm->mmu_lock);
401         srcu_read_unlock(&kvm->srcu, idx);
402 
403         return young;
404 }
405 
406 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
407                                      struct mm_struct *mm)
408 {
409         struct kvm *kvm = mmu_notifier_to_kvm(mn);
410         int idx;
411 
412         idx = srcu_read_lock(&kvm->srcu);
413         kvm_arch_flush_shadow_all(kvm);
414         srcu_read_unlock(&kvm->srcu, idx);
415 }
416 
417 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
418         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
419         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
420         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
421         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
422         .test_young             = kvm_mmu_notifier_test_young,
423         .change_pte             = kvm_mmu_notifier_change_pte,
424         .release                = kvm_mmu_notifier_release,
425 };
426 
427 static int kvm_init_mmu_notifier(struct kvm *kvm)
428 {
429         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
430         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
431 }
432 
433 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
434 
435 static int kvm_init_mmu_notifier(struct kvm *kvm)
436 {
437         return 0;
438 }
439 
440 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
441 
442 static struct kvm_memslots *kvm_alloc_memslots(void)
443 {
444         int i;
445         struct kvm_memslots *slots;
446 
447         slots = kvm_kvzalloc(sizeof(struct kvm_memslots));
448         if (!slots)
449                 return NULL;
450 
451         /*
452          * Init kvm generation close to the maximum to easily test the
453          * code of handling generation number wrap-around.
454          */
455         slots->generation = -150;
456         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
457                 slots->id_to_index[i] = slots->memslots[i].id = i;
458 
459         return slots;
460 }
461 
462 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
463 {
464         if (!memslot->dirty_bitmap)
465                 return;
466 
467         kvfree(memslot->dirty_bitmap);
468         memslot->dirty_bitmap = NULL;
469 }
470 
471 /*
472  * Free any memory in @free but not in @dont.
473  */
474 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
475                               struct kvm_memory_slot *dont)
476 {
477         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
478                 kvm_destroy_dirty_bitmap(free);
479 
480         kvm_arch_free_memslot(kvm, free, dont);
481 
482         free->npages = 0;
483 }
484 
485 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
486 {
487         struct kvm_memory_slot *memslot;
488 
489         if (!slots)
490                 return;
491 
492         kvm_for_each_memslot(memslot, slots)
493                 kvm_free_memslot(kvm, memslot, NULL);
494 
495         kvfree(slots);
496 }
497 
498 static struct kvm *kvm_create_vm(unsigned long type)
499 {
500         int r, i;
501         struct kvm *kvm = kvm_arch_alloc_vm();
502 
503         if (!kvm)
504                 return ERR_PTR(-ENOMEM);
505 
506         r = kvm_arch_init_vm(kvm, type);
507         if (r)
508                 goto out_err_no_disable;
509 
510         r = hardware_enable_all();
511         if (r)
512                 goto out_err_no_disable;
513 
514 #ifdef CONFIG_HAVE_KVM_IRQFD
515         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
516 #endif
517 
518         BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
519 
520         r = -ENOMEM;
521         for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
522                 kvm->memslots[i] = kvm_alloc_memslots();
523                 if (!kvm->memslots[i])
524                         goto out_err_no_srcu;
525         }
526 
527         if (init_srcu_struct(&kvm->srcu))
528                 goto out_err_no_srcu;
529         if (init_srcu_struct(&kvm->irq_srcu))
530                 goto out_err_no_irq_srcu;
531         for (i = 0; i < KVM_NR_BUSES; i++) {
532                 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
533                                         GFP_KERNEL);
534                 if (!kvm->buses[i])
535                         goto out_err;
536         }
537 
538         spin_lock_init(&kvm->mmu_lock);
539         kvm->mm = current->mm;
540         atomic_inc(&kvm->mm->mm_count);
541         kvm_eventfd_init(kvm);
542         mutex_init(&kvm->lock);
543         mutex_init(&kvm->irq_lock);
544         mutex_init(&kvm->slots_lock);
545         atomic_set(&kvm->users_count, 1);
546         INIT_LIST_HEAD(&kvm->devices);
547 
548         r = kvm_init_mmu_notifier(kvm);
549         if (r)
550                 goto out_err;
551 
552         spin_lock(&kvm_lock);
553         list_add(&kvm->vm_list, &vm_list);
554         spin_unlock(&kvm_lock);
555 
556         preempt_notifier_inc();
557 
558         return kvm;
559 
560 out_err:
561         cleanup_srcu_struct(&kvm->irq_srcu);
562 out_err_no_irq_srcu:
563         cleanup_srcu_struct(&kvm->srcu);
564 out_err_no_srcu:
565         hardware_disable_all();
566 out_err_no_disable:
567         for (i = 0; i < KVM_NR_BUSES; i++)
568                 kfree(kvm->buses[i]);
569         for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
570                 kvm_free_memslots(kvm, kvm->memslots[i]);
571         kvm_arch_free_vm(kvm);
572         return ERR_PTR(r);
573 }
574 
575 /*
576  * Avoid using vmalloc for a small buffer.
577  * Should not be used when the size is statically known.
578  */
579 void *kvm_kvzalloc(unsigned long size)
580 {
581         if (size > PAGE_SIZE)
582                 return vzalloc(size);
583         else
584                 return kzalloc(size, GFP_KERNEL);
585 }
586 
587 static void kvm_destroy_devices(struct kvm *kvm)
588 {
589         struct list_head *node, *tmp;
590 
591         list_for_each_safe(node, tmp, &kvm->devices) {
592                 struct kvm_device *dev =
593                         list_entry(node, struct kvm_device, vm_node);
594 
595                 list_del(node);
596                 dev->ops->destroy(dev);
597         }
598 }
599 
600 static void kvm_destroy_vm(struct kvm *kvm)
601 {
602         int i;
603         struct mm_struct *mm = kvm->mm;
604 
605         kvm_arch_sync_events(kvm);
606         spin_lock(&kvm_lock);
607         list_del(&kvm->vm_list);
608         spin_unlock(&kvm_lock);
609         kvm_free_irq_routing(kvm);
610         for (i = 0; i < KVM_NR_BUSES; i++)
611                 kvm_io_bus_destroy(kvm->buses[i]);
612         kvm_coalesced_mmio_free(kvm);
613 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
614         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
615 #else
616         kvm_arch_flush_shadow_all(kvm);
617 #endif
618         kvm_arch_destroy_vm(kvm);
619         kvm_destroy_devices(kvm);
620         for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
621                 kvm_free_memslots(kvm, kvm->memslots[i]);
622         cleanup_srcu_struct(&kvm->irq_srcu);
623         cleanup_srcu_struct(&kvm->srcu);
624         kvm_arch_free_vm(kvm);
625         preempt_notifier_dec();
626         hardware_disable_all();
627         mmdrop(mm);
628 }
629 
630 void kvm_get_kvm(struct kvm *kvm)
631 {
632         atomic_inc(&kvm->users_count);
633 }
634 EXPORT_SYMBOL_GPL(kvm_get_kvm);
635 
636 void kvm_put_kvm(struct kvm *kvm)
637 {
638         if (atomic_dec_and_test(&kvm->users_count))
639                 kvm_destroy_vm(kvm);
640 }
641 EXPORT_SYMBOL_GPL(kvm_put_kvm);
642 
643 
644 static int kvm_vm_release(struct inode *inode, struct file *filp)
645 {
646         struct kvm *kvm = filp->private_data;
647 
648         kvm_irqfd_release(kvm);
649 
650         kvm_put_kvm(kvm);
651         return 0;
652 }
653 
654 /*
655  * Allocation size is twice as large as the actual dirty bitmap size.
656  * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
657  */
658 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
659 {
660         unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
661 
662         memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
663         if (!memslot->dirty_bitmap)
664                 return -ENOMEM;
665 
666         return 0;
667 }
668 
669 /*
670  * Insert memslot and re-sort memslots based on their GFN,
671  * so binary search could be used to lookup GFN.
672  * Sorting algorithm takes advantage of having initially
673  * sorted array and known changed memslot position.
674  */
675 static void update_memslots(struct kvm_memslots *slots,
676                             struct kvm_memory_slot *new)
677 {
678         int id = new->id;
679         int i = slots->id_to_index[id];
680         struct kvm_memory_slot *mslots = slots->memslots;
681 
682         WARN_ON(mslots[i].id != id);
683         if (!new->npages) {
684                 WARN_ON(!mslots[i].npages);
685                 if (mslots[i].npages)
686                         slots->used_slots--;
687         } else {
688                 if (!mslots[i].npages)
689                         slots->used_slots++;
690         }
691 
692         while (i < KVM_MEM_SLOTS_NUM - 1 &&
693                new->base_gfn <= mslots[i + 1].base_gfn) {
694                 if (!mslots[i + 1].npages)
695                         break;
696                 mslots[i] = mslots[i + 1];
697                 slots->id_to_index[mslots[i].id] = i;
698                 i++;
699         }
700 
701         /*
702          * The ">=" is needed when creating a slot with base_gfn == 0,
703          * so that it moves before all those with base_gfn == npages == 0.
704          *
705          * On the other hand, if new->npages is zero, the above loop has
706          * already left i pointing to the beginning of the empty part of
707          * mslots, and the ">=" would move the hole backwards in this
708          * case---which is wrong.  So skip the loop when deleting a slot.
709          */
710         if (new->npages) {
711                 while (i > 0 &&
712                        new->base_gfn >= mslots[i - 1].base_gfn) {
713                         mslots[i] = mslots[i - 1];
714                         slots->id_to_index[mslots[i].id] = i;
715                         i--;
716                 }
717         } else
718                 WARN_ON_ONCE(i != slots->used_slots);
719 
720         mslots[i] = *new;
721         slots->id_to_index[mslots[i].id] = i;
722 }
723 
724 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
725 {
726         u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
727 
728 #ifdef __KVM_HAVE_READONLY_MEM
729         valid_flags |= KVM_MEM_READONLY;
730 #endif
731 
732         if (mem->flags & ~valid_flags)
733                 return -EINVAL;
734 
735         return 0;
736 }
737 
738 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
739                 int as_id, struct kvm_memslots *slots)
740 {
741         struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id);
742 
743         /*
744          * Set the low bit in the generation, which disables SPTE caching
745          * until the end of synchronize_srcu_expedited.
746          */
747         WARN_ON(old_memslots->generation & 1);
748         slots->generation = old_memslots->generation + 1;
749 
750         rcu_assign_pointer(kvm->memslots[as_id], slots);
751         synchronize_srcu_expedited(&kvm->srcu);
752 
753         /*
754          * Increment the new memslot generation a second time. This prevents
755          * vm exits that race with memslot updates from caching a memslot
756          * generation that will (potentially) be valid forever.
757          */
758         slots->generation++;
759 
760         kvm_arch_memslots_updated(kvm, slots);
761 
762         return old_memslots;
763 }
764 
765 /*
766  * Allocate some memory and give it an address in the guest physical address
767  * space.
768  *
769  * Discontiguous memory is allowed, mostly for framebuffers.
770  *
771  * Must be called holding kvm->slots_lock for write.
772  */
773 int __kvm_set_memory_region(struct kvm *kvm,
774                             const struct kvm_userspace_memory_region *mem)
775 {
776         int r;
777         gfn_t base_gfn;
778         unsigned long npages;
779         struct kvm_memory_slot *slot;
780         struct kvm_memory_slot old, new;
781         struct kvm_memslots *slots = NULL, *old_memslots;
782         int as_id, id;
783         enum kvm_mr_change change;
784 
785         r = check_memory_region_flags(mem);
786         if (r)
787                 goto out;
788 
789         r = -EINVAL;
790         as_id = mem->slot >> 16;
791         id = (u16)mem->slot;
792 
793         /* General sanity checks */
794         if (mem->memory_size & (PAGE_SIZE - 1))
795                 goto out;
796         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
797                 goto out;
798         /* We can read the guest memory with __xxx_user() later on. */
799         if ((id < KVM_USER_MEM_SLOTS) &&
800             ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
801              !access_ok(VERIFY_WRITE,
802                         (void __user *)(unsigned long)mem->userspace_addr,
803                         mem->memory_size)))
804                 goto out;
805         if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
806                 goto out;
807         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
808                 goto out;
809 
810         slot = id_to_memslot(__kvm_memslots(kvm, as_id), id);
811         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
812         npages = mem->memory_size >> PAGE_SHIFT;
813 
814         if (npages > KVM_MEM_MAX_NR_PAGES)
815                 goto out;
816 
817         new = old = *slot;
818 
819         new.id = id;
820         new.base_gfn = base_gfn;
821         new.npages = npages;
822         new.flags = mem->flags;
823 
824         if (npages) {
825                 if (!old.npages)
826                         change = KVM_MR_CREATE;
827                 else { /* Modify an existing slot. */
828                         if ((mem->userspace_addr != old.userspace_addr) ||
829                             (npages != old.npages) ||
830                             ((new.flags ^ old.flags) & KVM_MEM_READONLY))
831                                 goto out;
832 
833                         if (base_gfn != old.base_gfn)
834                                 change = KVM_MR_MOVE;
835                         else if (new.flags != old.flags)
836                                 change = KVM_MR_FLAGS_ONLY;
837                         else { /* Nothing to change. */
838                                 r = 0;
839                                 goto out;
840                         }
841                 }
842         } else {
843                 if (!old.npages)
844                         goto out;
845 
846                 change = KVM_MR_DELETE;
847                 new.base_gfn = 0;
848                 new.flags = 0;
849         }
850 
851         if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
852                 /* Check for overlaps */
853                 r = -EEXIST;
854                 kvm_for_each_memslot(slot, __kvm_memslots(kvm, as_id)) {
855                         if ((slot->id >= KVM_USER_MEM_SLOTS) ||
856                             (slot->id == id))
857                                 continue;
858                         if (!((base_gfn + npages <= slot->base_gfn) ||
859                               (base_gfn >= slot->base_gfn + slot->npages)))
860                                 goto out;
861                 }
862         }
863 
864         /* Free page dirty bitmap if unneeded */
865         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
866                 new.dirty_bitmap = NULL;
867 
868         r = -ENOMEM;
869         if (change == KVM_MR_CREATE) {
870                 new.userspace_addr = mem->userspace_addr;
871 
872                 if (kvm_arch_create_memslot(kvm, &new, npages))
873                         goto out_free;
874         }
875 
876         /* Allocate page dirty bitmap if needed */
877         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
878                 if (kvm_create_dirty_bitmap(&new) < 0)
879                         goto out_free;
880         }
881 
882         slots = kvm_kvzalloc(sizeof(struct kvm_memslots));
883         if (!slots)
884                 goto out_free;
885         memcpy(slots, __kvm_memslots(kvm, as_id), sizeof(struct kvm_memslots));
886 
887         if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
888                 slot = id_to_memslot(slots, id);
889                 slot->flags |= KVM_MEMSLOT_INVALID;
890 
891                 old_memslots = install_new_memslots(kvm, as_id, slots);
892 
893                 /* slot was deleted or moved, clear iommu mapping */
894                 kvm_iommu_unmap_pages(kvm, &old);
895                 /* From this point no new shadow pages pointing to a deleted,
896                  * or moved, memslot will be created.
897                  *
898                  * validation of sp->gfn happens in:
899                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
900                  *      - kvm_is_visible_gfn (mmu_check_roots)
901                  */
902                 kvm_arch_flush_shadow_memslot(kvm, slot);
903 
904                 /*
905                  * We can re-use the old_memslots from above, the only difference
906                  * from the currently installed memslots is the invalid flag.  This
907                  * will get overwritten by update_memslots anyway.
908                  */
909                 slots = old_memslots;
910         }
911 
912         r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
913         if (r)
914                 goto out_slots;
915 
916         /* actual memory is freed via old in kvm_free_memslot below */
917         if (change == KVM_MR_DELETE) {
918                 new.dirty_bitmap = NULL;
919                 memset(&new.arch, 0, sizeof(new.arch));
920         }
921 
922         update_memslots(slots, &new);
923         old_memslots = install_new_memslots(kvm, as_id, slots);
924 
925         kvm_arch_commit_memory_region(kvm, mem, &old, &new, change);
926 
927         kvm_free_memslot(kvm, &old, &new);
928         kvfree(old_memslots);
929 
930         /*
931          * IOMMU mapping:  New slots need to be mapped.  Old slots need to be
932          * un-mapped and re-mapped if their base changes.  Since base change
933          * unmapping is handled above with slot deletion, mapping alone is
934          * needed here.  Anything else the iommu might care about for existing
935          * slots (size changes, userspace addr changes and read-only flag
936          * changes) is disallowed above, so any other attribute changes getting
937          * here can be skipped.
938          */
939         if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
940                 r = kvm_iommu_map_pages(kvm, &new);
941                 return r;
942         }
943 
944         return 0;
945 
946 out_slots:
947         kvfree(slots);
948 out_free:
949         kvm_free_memslot(kvm, &new, &old);
950 out:
951         return r;
952 }
953 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
954 
955 int kvm_set_memory_region(struct kvm *kvm,
956                           const struct kvm_userspace_memory_region *mem)
957 {
958         int r;
959 
960         mutex_lock(&kvm->slots_lock);
961         r = __kvm_set_memory_region(kvm, mem);
962         mutex_unlock(&kvm->slots_lock);
963         return r;
964 }
965 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
966 
967 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
968                                           struct kvm_userspace_memory_region *mem)
969 {
970         if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
971                 return -EINVAL;
972 
973         return kvm_set_memory_region(kvm, mem);
974 }
975 
976 int kvm_get_dirty_log(struct kvm *kvm,
977                         struct kvm_dirty_log *log, int *is_dirty)
978 {
979         struct kvm_memslots *slots;
980         struct kvm_memory_slot *memslot;
981         int r, i, as_id, id;
982         unsigned long n;
983         unsigned long any = 0;
984 
985         r = -EINVAL;
986         as_id = log->slot >> 16;
987         id = (u16)log->slot;
988         if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
989                 goto out;
990 
991         slots = __kvm_memslots(kvm, as_id);
992         memslot = id_to_memslot(slots, id);
993         r = -ENOENT;
994         if (!memslot->dirty_bitmap)
995                 goto out;
996 
997         n = kvm_dirty_bitmap_bytes(memslot);
998 
999         for (i = 0; !any && i < n/sizeof(long); ++i)
1000                 any = memslot->dirty_bitmap[i];
1001 
1002         r = -EFAULT;
1003         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1004                 goto out;
1005 
1006         if (any)
1007                 *is_dirty = 1;
1008 
1009         r = 0;
1010 out:
1011         return r;
1012 }
1013 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1014 
1015 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1016 /**
1017  * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages
1018  *      are dirty write protect them for next write.
1019  * @kvm:        pointer to kvm instance
1020  * @log:        slot id and address to which we copy the log
1021  * @is_dirty:   flag set if any page is dirty
1022  *
1023  * We need to keep it in mind that VCPU threads can write to the bitmap
1024  * concurrently. So, to avoid losing track of dirty pages we keep the
1025  * following order:
1026  *
1027  *    1. Take a snapshot of the bit and clear it if needed.
1028  *    2. Write protect the corresponding page.
1029  *    3. Copy the snapshot to the userspace.
1030  *    4. Upon return caller flushes TLB's if needed.
1031  *
1032  * Between 2 and 4, the guest may write to the page using the remaining TLB
1033  * entry.  This is not a problem because the page is reported dirty using
1034  * the snapshot taken before and step 4 ensures that writes done after
1035  * exiting to userspace will be logged for the next call.
1036  *
1037  */
1038 int kvm_get_dirty_log_protect(struct kvm *kvm,
1039                         struct kvm_dirty_log *log, bool *is_dirty)
1040 {
1041         struct kvm_memslots *slots;
1042         struct kvm_memory_slot *memslot;
1043         int r, i, as_id, id;
1044         unsigned long n;
1045         unsigned long *dirty_bitmap;
1046         unsigned long *dirty_bitmap_buffer;
1047 
1048         r = -EINVAL;
1049         as_id = log->slot >> 16;
1050         id = (u16)log->slot;
1051         if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1052                 goto out;
1053 
1054         slots = __kvm_memslots(kvm, as_id);
1055         memslot = id_to_memslot(slots, id);
1056 
1057         dirty_bitmap = memslot->dirty_bitmap;
1058         r = -ENOENT;
1059         if (!dirty_bitmap)
1060                 goto out;
1061 
1062         n = kvm_dirty_bitmap_bytes(memslot);
1063 
1064         dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long);
1065         memset(dirty_bitmap_buffer, 0, n);
1066 
1067         spin_lock(&kvm->mmu_lock);
1068         *is_dirty = false;
1069         for (i = 0; i < n / sizeof(long); i++) {
1070                 unsigned long mask;
1071                 gfn_t offset;
1072 
1073                 if (!dirty_bitmap[i])
1074                         continue;
1075 
1076                 *is_dirty = true;
1077 
1078                 mask = xchg(&dirty_bitmap[i], 0);
1079                 dirty_bitmap_buffer[i] = mask;
1080 
1081                 if (mask) {
1082                         offset = i * BITS_PER_LONG;
1083                         kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1084                                                                 offset, mask);
1085                 }
1086         }
1087 
1088         spin_unlock(&kvm->mmu_lock);
1089 
1090         r = -EFAULT;
1091         if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1092                 goto out;
1093 
1094         r = 0;
1095 out:
1096         return r;
1097 }
1098 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect);
1099 #endif
1100 
1101 bool kvm_largepages_enabled(void)
1102 {
1103         return largepages_enabled;
1104 }
1105 
1106 void kvm_disable_largepages(void)
1107 {
1108         largepages_enabled = false;
1109 }
1110 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1111 
1112 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1113 {
1114         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1115 }
1116 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1117 
1118 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
1119 {
1120         return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn);
1121 }
1122 
1123 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1124 {
1125         struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1126 
1127         if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1128               memslot->flags & KVM_MEMSLOT_INVALID)
1129                 return 0;
1130 
1131         return 1;
1132 }
1133 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1134 
1135 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1136 {
1137         struct vm_area_struct *vma;
1138         unsigned long addr, size;
1139 
1140         size = PAGE_SIZE;
1141 
1142         addr = gfn_to_hva(kvm, gfn);
1143         if (kvm_is_error_hva(addr))
1144                 return PAGE_SIZE;
1145 
1146         down_read(&current->mm->mmap_sem);
1147         vma = find_vma(current->mm, addr);
1148         if (!vma)
1149                 goto out;
1150 
1151         size = vma_kernel_pagesize(vma);
1152 
1153 out:
1154         up_read(&current->mm->mmap_sem);
1155 
1156         return size;
1157 }
1158 
1159 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1160 {
1161         return slot->flags & KVM_MEM_READONLY;
1162 }
1163 
1164 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1165                                        gfn_t *nr_pages, bool write)
1166 {
1167         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1168                 return KVM_HVA_ERR_BAD;
1169 
1170         if (memslot_is_readonly(slot) && write)
1171                 return KVM_HVA_ERR_RO_BAD;
1172 
1173         if (nr_pages)
1174                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1175 
1176         return __gfn_to_hva_memslot(slot, gfn);
1177 }
1178 
1179 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1180                                      gfn_t *nr_pages)
1181 {
1182         return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1183 }
1184 
1185 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1186                                         gfn_t gfn)
1187 {
1188         return gfn_to_hva_many(slot, gfn, NULL);
1189 }
1190 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1191 
1192 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1193 {
1194         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1195 }
1196 EXPORT_SYMBOL_GPL(gfn_to_hva);
1197 
1198 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
1199 {
1200         return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
1201 }
1202 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
1203 
1204 /*
1205  * If writable is set to false, the hva returned by this function is only
1206  * allowed to be read.
1207  */
1208 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1209                                       gfn_t gfn, bool *writable)
1210 {
1211         unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1212 
1213         if (!kvm_is_error_hva(hva) && writable)
1214                 *writable = !memslot_is_readonly(slot);
1215 
1216         return hva;
1217 }
1218 
1219 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1220 {
1221         struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1222 
1223         return gfn_to_hva_memslot_prot(slot, gfn, writable);
1224 }
1225 
1226 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
1227 {
1228         struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1229 
1230         return gfn_to_hva_memslot_prot(slot, gfn, writable);
1231 }
1232 
1233 static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1234         unsigned long start, int write, struct page **page)
1235 {
1236         int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1237 
1238         if (write)
1239                 flags |= FOLL_WRITE;
1240 
1241         return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1242 }
1243 
1244 static inline int check_user_page_hwpoison(unsigned long addr)
1245 {
1246         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1247 
1248         rc = __get_user_pages(current, current->mm, addr, 1,
1249                               flags, NULL, NULL, NULL);
1250         return rc == -EHWPOISON;
1251 }
1252 
1253 /*
1254  * The atomic path to get the writable pfn which will be stored in @pfn,
1255  * true indicates success, otherwise false is returned.
1256  */
1257 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1258                             bool write_fault, bool *writable, pfn_t *pfn)
1259 {
1260         struct page *page[1];
1261         int npages;
1262 
1263         if (!(async || atomic))
1264                 return false;
1265 
1266         /*
1267          * Fast pin a writable pfn only if it is a write fault request
1268          * or the caller allows to map a writable pfn for a read fault
1269          * request.
1270          */
1271         if (!(write_fault || writable))
1272                 return false;
1273 
1274         npages = __get_user_pages_fast(addr, 1, 1, page);
1275         if (npages == 1) {
1276                 *pfn = page_to_pfn(page[0]);
1277 
1278                 if (writable)
1279                         *writable = true;
1280                 return true;
1281         }
1282 
1283         return false;
1284 }
1285 
1286 /*
1287  * The slow path to get the pfn of the specified host virtual address,
1288  * 1 indicates success, -errno is returned if error is detected.
1289  */
1290 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1291                            bool *writable, pfn_t *pfn)
1292 {
1293         struct page *page[1];
1294         int npages = 0;
1295 
1296         might_sleep();
1297 
1298         if (writable)
1299                 *writable = write_fault;
1300 
1301         if (async) {
1302                 down_read(&current->mm->mmap_sem);
1303                 npages = get_user_page_nowait(current, current->mm,
1304                                               addr, write_fault, page);
1305                 up_read(&current->mm->mmap_sem);
1306         } else
1307                 npages = __get_user_pages_unlocked(current, current->mm, addr, 1,
1308                                                    write_fault, 0, page,
1309                                                    FOLL_TOUCH|FOLL_HWPOISON);
1310         if (npages != 1)
1311                 return npages;
1312 
1313         /* map read fault as writable if possible */
1314         if (unlikely(!write_fault) && writable) {
1315                 struct page *wpage[1];
1316 
1317                 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1318                 if (npages == 1) {
1319                         *writable = true;
1320                         put_page(page[0]);
1321                         page[0] = wpage[0];
1322                 }
1323 
1324                 npages = 1;
1325         }
1326         *pfn = page_to_pfn(page[0]);
1327         return npages;
1328 }
1329 
1330 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1331 {
1332         if (unlikely(!(vma->vm_flags & VM_READ)))
1333                 return false;
1334 
1335         if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1336                 return false;
1337 
1338         return true;
1339 }
1340 
1341 /*
1342  * Pin guest page in memory and return its pfn.
1343  * @addr: host virtual address which maps memory to the guest
1344  * @atomic: whether this function can sleep
1345  * @async: whether this function need to wait IO complete if the
1346  *         host page is not in the memory
1347  * @write_fault: whether we should get a writable host page
1348  * @writable: whether it allows to map a writable host page for !@write_fault
1349  *
1350  * The function will map a writable host page for these two cases:
1351  * 1): @write_fault = true
1352  * 2): @write_fault = false && @writable, @writable will tell the caller
1353  *     whether the mapping is writable.
1354  */
1355 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1356                         bool write_fault, bool *writable)
1357 {
1358         struct vm_area_struct *vma;
1359         pfn_t pfn = 0;
1360         int npages;
1361 
1362         /* we can do it either atomically or asynchronously, not both */
1363         BUG_ON(atomic && async);
1364 
1365         if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1366                 return pfn;
1367 
1368         if (atomic)
1369                 return KVM_PFN_ERR_FAULT;
1370 
1371         npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1372         if (npages == 1)
1373                 return pfn;
1374 
1375         down_read(&current->mm->mmap_sem);
1376         if (npages == -EHWPOISON ||
1377               (!async && check_user_page_hwpoison(addr))) {
1378                 pfn = KVM_PFN_ERR_HWPOISON;
1379                 goto exit;
1380         }
1381 
1382         vma = find_vma_intersection(current->mm, addr, addr + 1);
1383 
1384         if (vma == NULL)
1385                 pfn = KVM_PFN_ERR_FAULT;
1386         else if ((vma->vm_flags & VM_PFNMAP)) {
1387                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1388                         vma->vm_pgoff;
1389                 BUG_ON(!kvm_is_reserved_pfn(pfn));
1390         } else {
1391                 if (async && vma_is_valid(vma, write_fault))
1392                         *async = true;
1393                 pfn = KVM_PFN_ERR_FAULT;
1394         }
1395 exit:
1396         up_read(&current->mm->mmap_sem);
1397         return pfn;
1398 }
1399 
1400 pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1401                            bool *async, bool write_fault, bool *writable)
1402 {
1403         unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1404 
1405         if (addr == KVM_HVA_ERR_RO_BAD)
1406                 return KVM_PFN_ERR_RO_FAULT;
1407 
1408         if (kvm_is_error_hva(addr))
1409                 return KVM_PFN_NOSLOT;
1410 
1411         /* Do not map writable pfn in the readonly memslot. */
1412         if (writable && memslot_is_readonly(slot)) {
1413                 *writable = false;
1414                 writable = NULL;
1415         }
1416 
1417         return hva_to_pfn(addr, atomic, async, write_fault,
1418                           writable);
1419 }
1420 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
1421 
1422 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1423                       bool *writable)
1424 {
1425         return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
1426                                     write_fault, writable);
1427 }
1428 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1429 
1430 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1431 {
1432         return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1433 }
1434 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
1435 
1436 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1437 {
1438         return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1439 }
1440 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1441 
1442 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1443 {
1444         return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm, gfn), gfn);
1445 }
1446 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1447 
1448 pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
1449 {
1450         return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1451 }
1452 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
1453 
1454 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1455 {
1456         return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
1457 }
1458 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1459 
1460 pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
1461 {
1462         return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1463 }
1464 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
1465 
1466 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1467                             struct page **pages, int nr_pages)
1468 {
1469         unsigned long addr;
1470         gfn_t entry;
1471 
1472         addr = gfn_to_hva_many(slot, gfn, &entry);
1473         if (kvm_is_error_hva(addr))
1474                 return -1;
1475 
1476         if (entry < nr_pages)
1477                 return 0;
1478 
1479         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1480 }
1481 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1482 
1483 static struct page *kvm_pfn_to_page(pfn_t pfn)
1484 {
1485         if (is_error_noslot_pfn(pfn))
1486                 return KVM_ERR_PTR_BAD_PAGE;
1487 
1488         if (kvm_is_reserved_pfn(pfn)) {
1489                 WARN_ON(1);
1490                 return KVM_ERR_PTR_BAD_PAGE;
1491         }
1492 
1493         return pfn_to_page(pfn);
1494 }
1495 
1496 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1497 {
1498         pfn_t pfn;
1499 
1500         pfn = gfn_to_pfn(kvm, gfn);
1501 
1502         return kvm_pfn_to_page(pfn);
1503 }
1504 EXPORT_SYMBOL_GPL(gfn_to_page);
1505 
1506 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
1507 {
1508         pfn_t pfn;
1509 
1510         pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
1511 
1512         return kvm_pfn_to_page(pfn);
1513 }
1514 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
1515 
1516 void kvm_release_page_clean(struct page *page)
1517 {
1518         WARN_ON(is_error_page(page));
1519 
1520         kvm_release_pfn_clean(page_to_pfn(page));
1521 }
1522 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1523 
1524 void kvm_release_pfn_clean(pfn_t pfn)
1525 {
1526         if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
1527                 put_page(pfn_to_page(pfn));
1528 }
1529 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1530 
1531 void kvm_release_page_dirty(struct page *page)
1532 {
1533         WARN_ON(is_error_page(page));
1534 
1535         kvm_release_pfn_dirty(page_to_pfn(page));
1536 }
1537 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1538 
1539 static void kvm_release_pfn_dirty(pfn_t pfn)
1540 {
1541         kvm_set_pfn_dirty(pfn);
1542         kvm_release_pfn_clean(pfn);
1543 }
1544 
1545 void kvm_set_pfn_dirty(pfn_t pfn)
1546 {
1547         if (!kvm_is_reserved_pfn(pfn)) {
1548                 struct page *page = pfn_to_page(pfn);
1549 
1550                 if (!PageReserved(page))
1551                         SetPageDirty(page);
1552         }
1553 }
1554 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1555 
1556 void kvm_set_pfn_accessed(pfn_t pfn)
1557 {
1558         if (!kvm_is_reserved_pfn(pfn))
1559                 mark_page_accessed(pfn_to_page(pfn));
1560 }
1561 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1562 
1563 void kvm_get_pfn(pfn_t pfn)
1564 {
1565         if (!kvm_is_reserved_pfn(pfn))
1566                 get_page(pfn_to_page(pfn));
1567 }
1568 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1569 
1570 static int next_segment(unsigned long len, int offset)
1571 {
1572         if (len > PAGE_SIZE - offset)
1573                 return PAGE_SIZE - offset;
1574         else
1575                 return len;
1576 }
1577 
1578 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
1579                                  void *data, int offset, int len)
1580 {
1581         int r;
1582         unsigned long addr;
1583 
1584         addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1585         if (kvm_is_error_hva(addr))
1586                 return -EFAULT;
1587         r = __copy_from_user(data, (void __user *)addr + offset, len);
1588         if (r)
1589                 return -EFAULT;
1590         return 0;
1591 }
1592 
1593 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1594                         int len)
1595 {
1596         struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1597 
1598         return __kvm_read_guest_page(slot, gfn, data, offset, len);
1599 }
1600 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1601 
1602 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
1603                              int offset, int len)
1604 {
1605         struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1606 
1607         return __kvm_read_guest_page(slot, gfn, data, offset, len);
1608 }
1609 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
1610 
1611 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1612 {
1613         gfn_t gfn = gpa >> PAGE_SHIFT;
1614         int seg;
1615         int offset = offset_in_page(gpa);
1616         int ret;
1617 
1618         while ((seg = next_segment(len, offset)) != 0) {
1619                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1620                 if (ret < 0)
1621                         return ret;
1622                 offset = 0;
1623                 len -= seg;
1624                 data += seg;
1625                 ++gfn;
1626         }
1627         return 0;
1628 }
1629 EXPORT_SYMBOL_GPL(kvm_read_guest);
1630 
1631 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
1632 {
1633         gfn_t gfn = gpa >> PAGE_SHIFT;
1634         int seg;
1635         int offset = offset_in_page(gpa);
1636         int ret;
1637 
1638         while ((seg = next_segment(len, offset)) != 0) {
1639                 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
1640                 if (ret < 0)
1641                         return ret;
1642                 offset = 0;
1643                 len -= seg;
1644                 data += seg;
1645                 ++gfn;
1646         }
1647         return 0;
1648 }
1649 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
1650 
1651 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1652                                    void *data, int offset, unsigned long len)
1653 {
1654         int r;
1655         unsigned long addr;
1656 
1657         addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1658         if (kvm_is_error_hva(addr))
1659                 return -EFAULT;
1660         pagefault_disable();
1661         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1662         pagefault_enable();
1663         if (r)
1664                 return -EFAULT;
1665         return 0;
1666 }
1667 
1668 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1669                           unsigned long len)
1670 {
1671         gfn_t gfn = gpa >> PAGE_SHIFT;
1672         struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1673         int offset = offset_in_page(gpa);
1674 
1675         return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1676 }
1677 EXPORT_SYMBOL_GPL(kvm_read_guest_atomic);
1678 
1679 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
1680                                void *data, unsigned long len)
1681 {
1682         gfn_t gfn = gpa >> PAGE_SHIFT;
1683         struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1684         int offset = offset_in_page(gpa);
1685 
1686         return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1687 }
1688 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
1689 
1690 static int __kvm_write_guest_page(struct kvm_memory_slot *memslot, gfn_t gfn,
1691                                   const void *data, int offset, int len)
1692 {
1693         int r;
1694         unsigned long addr;
1695 
1696         addr = gfn_to_hva_memslot(memslot, gfn);
1697         if (kvm_is_error_hva(addr))
1698                 return -EFAULT;
1699         r = __copy_to_user((void __user *)addr + offset, data, len);
1700         if (r)
1701                 return -EFAULT;
1702         mark_page_dirty_in_slot(memslot, gfn);
1703         return 0;
1704 }
1705 
1706 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
1707                          const void *data, int offset, int len)
1708 {
1709         struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1710 
1711         return __kvm_write_guest_page(slot, gfn, data, offset, len);
1712 }
1713 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1714 
1715 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
1716                               const void *data, int offset, int len)
1717 {
1718         struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1719 
1720         return __kvm_write_guest_page(slot, gfn, data, offset, len);
1721 }
1722 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
1723 
1724 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1725                     unsigned long len)
1726 {
1727         gfn_t gfn = gpa >> PAGE_SHIFT;
1728         int seg;
1729         int offset = offset_in_page(gpa);
1730         int ret;
1731 
1732         while ((seg = next_segment(len, offset)) != 0) {
1733                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1734                 if (ret < 0)
1735                         return ret;
1736                 offset = 0;
1737                 len -= seg;
1738                 data += seg;
1739                 ++gfn;
1740         }
1741         return 0;
1742 }
1743 EXPORT_SYMBOL_GPL(kvm_write_guest);
1744 
1745 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
1746                          unsigned long len)
1747 {
1748         gfn_t gfn = gpa >> PAGE_SHIFT;
1749         int seg;
1750         int offset = offset_in_page(gpa);
1751         int ret;
1752 
1753         while ((seg = next_segment(len, offset)) != 0) {
1754                 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
1755                 if (ret < 0)
1756                         return ret;
1757                 offset = 0;
1758                 len -= seg;
1759                 data += seg;
1760                 ++gfn;
1761         }
1762         return 0;
1763 }
1764 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
1765 
1766 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1767                               gpa_t gpa, unsigned long len)
1768 {
1769         struct kvm_memslots *slots = kvm_memslots(kvm);
1770         int offset = offset_in_page(gpa);
1771         gfn_t start_gfn = gpa >> PAGE_SHIFT;
1772         gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1773         gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1774         gfn_t nr_pages_avail;
1775 
1776         ghc->gpa = gpa;
1777         ghc->generation = slots->generation;
1778         ghc->len = len;
1779         ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1780         ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, NULL);
1781         if (!kvm_is_error_hva(ghc->hva) && nr_pages_needed <= 1) {
1782                 ghc->hva += offset;
1783         } else {
1784                 /*
1785                  * If the requested region crosses two memslots, we still
1786                  * verify that the entire region is valid here.
1787                  */
1788                 while (start_gfn <= end_gfn) {
1789                         ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1790                         ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1791                                                    &nr_pages_avail);
1792                         if (kvm_is_error_hva(ghc->hva))
1793                                 return -EFAULT;
1794                         start_gfn += nr_pages_avail;
1795                 }
1796                 /* Use the slow path for cross page reads and writes. */
1797                 ghc->memslot = NULL;
1798         }
1799         return 0;
1800 }
1801 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1802 
1803 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1804                            void *data, unsigned long len)
1805 {
1806         struct kvm_memslots *slots = kvm_memslots(kvm);
1807         int r;
1808 
1809         BUG_ON(len > ghc->len);
1810 
1811         if (slots->generation != ghc->generation)
1812                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1813 
1814         if (unlikely(!ghc->memslot))
1815                 return kvm_write_guest(kvm, ghc->gpa, data, len);
1816 
1817         if (kvm_is_error_hva(ghc->hva))
1818                 return -EFAULT;
1819 
1820         r = __copy_to_user((void __user *)ghc->hva, data, len);
1821         if (r)
1822                 return -EFAULT;
1823         mark_page_dirty_in_slot(ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1824 
1825         return 0;
1826 }
1827 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1828 
1829 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1830                            void *data, unsigned long len)
1831 {
1832         struct kvm_memslots *slots = kvm_memslots(kvm);
1833         int r;
1834 
1835         BUG_ON(len > ghc->len);
1836 
1837         if (slots->generation != ghc->generation)
1838                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1839 
1840         if (unlikely(!ghc->memslot))
1841                 return kvm_read_guest(kvm, ghc->gpa, data, len);
1842 
1843         if (kvm_is_error_hva(ghc->hva))
1844                 return -EFAULT;
1845 
1846         r = __copy_from_user(data, (void __user *)ghc->hva, len);
1847         if (r)
1848                 return -EFAULT;
1849 
1850         return 0;
1851 }
1852 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1853 
1854 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1855 {
1856         const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
1857 
1858         return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
1859 }
1860 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1861 
1862 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1863 {
1864         gfn_t gfn = gpa >> PAGE_SHIFT;
1865         int seg;
1866         int offset = offset_in_page(gpa);
1867         int ret;
1868 
1869         while ((seg = next_segment(len, offset)) != 0) {
1870                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1871                 if (ret < 0)
1872                         return ret;
1873                 offset = 0;
1874                 len -= seg;
1875                 ++gfn;
1876         }
1877         return 0;
1878 }
1879 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1880 
1881 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot,
1882                                     gfn_t gfn)
1883 {
1884         if (memslot && memslot->dirty_bitmap) {
1885                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1886 
1887                 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1888         }
1889 }
1890 
1891 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1892 {
1893         struct kvm_memory_slot *memslot;
1894 
1895         memslot = gfn_to_memslot(kvm, gfn);
1896         mark_page_dirty_in_slot(memslot, gfn);
1897 }
1898 EXPORT_SYMBOL_GPL(mark_page_dirty);
1899 
1900 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
1901 {
1902         struct kvm_memory_slot *memslot;
1903 
1904         memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1905         mark_page_dirty_in_slot(memslot, gfn);
1906 }
1907 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
1908 
1909 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
1910 {
1911         if (kvm_arch_vcpu_runnable(vcpu)) {
1912                 kvm_make_request(KVM_REQ_UNHALT, vcpu);
1913                 return -EINTR;
1914         }
1915         if (kvm_cpu_has_pending_timer(vcpu))
1916                 return -EINTR;
1917         if (signal_pending(current))
1918                 return -EINTR;
1919 
1920         return 0;
1921 }
1922 
1923 /*
1924  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1925  */
1926 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1927 {
1928         ktime_t start, cur;
1929         DEFINE_WAIT(wait);
1930         bool waited = false;
1931 
1932         start = cur = ktime_get();
1933         if (halt_poll_ns) {
1934                 ktime_t stop = ktime_add_ns(ktime_get(), halt_poll_ns);
1935 
1936                 do {
1937                         /*
1938                          * This sets KVM_REQ_UNHALT if an interrupt
1939                          * arrives.
1940                          */
1941                         if (kvm_vcpu_check_block(vcpu) < 0) {
1942                                 ++vcpu->stat.halt_successful_poll;
1943                                 goto out;
1944                         }
1945                         cur = ktime_get();
1946                 } while (single_task_running() && ktime_before(cur, stop));
1947         }
1948 
1949         for (;;) {
1950                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1951 
1952                 if (kvm_vcpu_check_block(vcpu) < 0)
1953                         break;
1954 
1955                 waited = true;
1956                 schedule();
1957         }
1958 
1959         finish_wait(&vcpu->wq, &wait);
1960         cur = ktime_get();
1961 
1962 out:
1963         trace_kvm_vcpu_wakeup(ktime_to_ns(cur) - ktime_to_ns(start), waited);
1964 }
1965 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
1966 
1967 #ifndef CONFIG_S390
1968 /*
1969  * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1970  */
1971 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1972 {
1973         int me;
1974         int cpu = vcpu->cpu;
1975         wait_queue_head_t *wqp;
1976 
1977         wqp = kvm_arch_vcpu_wq(vcpu);
1978         if (waitqueue_active(wqp)) {
1979                 wake_up_interruptible(wqp);
1980                 ++vcpu->stat.halt_wakeup;
1981         }
1982 
1983         me = get_cpu();
1984         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1985                 if (kvm_arch_vcpu_should_kick(vcpu))
1986                         smp_send_reschedule(cpu);
1987         put_cpu();
1988 }
1989 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
1990 #endif /* !CONFIG_S390 */
1991 
1992 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
1993 {
1994         struct pid *pid;
1995         struct task_struct *task = NULL;
1996         int ret = 0;
1997 
1998         rcu_read_lock();
1999         pid = rcu_dereference(target->pid);
2000         if (pid)
2001                 task = get_pid_task(pid, PIDTYPE_PID);
2002         rcu_read_unlock();
2003         if (!task)
2004                 return ret;
2005         ret = yield_to(task, 1);
2006         put_task_struct(task);
2007 
2008         return ret;
2009 }
2010 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
2011 
2012 /*
2013  * Helper that checks whether a VCPU is eligible for directed yield.
2014  * Most eligible candidate to yield is decided by following heuristics:
2015  *
2016  *  (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2017  *  (preempted lock holder), indicated by @in_spin_loop.
2018  *  Set at the beiginning and cleared at the end of interception/PLE handler.
2019  *
2020  *  (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2021  *  chance last time (mostly it has become eligible now since we have probably
2022  *  yielded to lockholder in last iteration. This is done by toggling
2023  *  @dy_eligible each time a VCPU checked for eligibility.)
2024  *
2025  *  Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2026  *  to preempted lock-holder could result in wrong VCPU selection and CPU
2027  *  burning. Giving priority for a potential lock-holder increases lock
2028  *  progress.
2029  *
2030  *  Since algorithm is based on heuristics, accessing another VCPU data without
2031  *  locking does not harm. It may result in trying to yield to  same VCPU, fail
2032  *  and continue with next VCPU and so on.
2033  */
2034 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
2035 {
2036 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2037         bool eligible;
2038 
2039         eligible = !vcpu->spin_loop.in_spin_loop ||
2040                     vcpu->spin_loop.dy_eligible;
2041 
2042         if (vcpu->spin_loop.in_spin_loop)
2043                 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
2044 
2045         return eligible;
2046 #else
2047         return true;
2048 #endif
2049 }
2050 
2051 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
2052 {
2053         struct kvm *kvm = me->kvm;
2054         struct kvm_vcpu *vcpu;
2055         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
2056         int yielded = 0;
2057         int try = 3;
2058         int pass;
2059         int i;
2060 
2061         kvm_vcpu_set_in_spin_loop(me, true);
2062         /*
2063          * We boost the priority of a VCPU that is runnable but not
2064          * currently running, because it got preempted by something
2065          * else and called schedule in __vcpu_run.  Hopefully that
2066          * VCPU is holding the lock that we need and will release it.
2067          * We approximate round-robin by starting at the last boosted VCPU.
2068          */
2069         for (pass = 0; pass < 2 && !yielded && try; pass++) {
2070                 kvm_for_each_vcpu(i, vcpu, kvm) {
2071                         if (!pass && i <= last_boosted_vcpu) {
2072                                 i = last_boosted_vcpu;
2073                                 continue;
2074                         } else if (pass && i > last_boosted_vcpu)
2075                                 break;
2076                         if (!ACCESS_ONCE(vcpu->preempted))
2077                                 continue;
2078                         if (vcpu == me)
2079                                 continue;
2080                         if (waitqueue_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
2081                                 continue;
2082                         if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
2083                                 continue;
2084 
2085                         yielded = kvm_vcpu_yield_to(vcpu);
2086                         if (yielded > 0) {
2087                                 kvm->last_boosted_vcpu = i;
2088                                 break;
2089                         } else if (yielded < 0) {
2090                                 try--;
2091                                 if (!try)
2092                                         break;
2093                         }
2094                 }
2095         }
2096         kvm_vcpu_set_in_spin_loop(me, false);
2097 
2098         /* Ensure vcpu is not eligible during next spinloop */
2099         kvm_vcpu_set_dy_eligible(me, false);
2100 }
2101 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
2102 
2103 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2104 {
2105         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
2106         struct page *page;
2107 
2108         if (vmf->pgoff == 0)
2109                 page = virt_to_page(vcpu->run);
2110 #ifdef CONFIG_X86
2111         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
2112                 page = virt_to_page(vcpu->arch.pio_data);
2113 #endif
2114 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2115         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
2116                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
2117 #endif
2118         else
2119                 return kvm_arch_vcpu_fault(vcpu, vmf);
2120         get_page(page);
2121         vmf->page = page;
2122         return 0;
2123 }
2124 
2125 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
2126         .fault = kvm_vcpu_fault,
2127 };
2128 
2129 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2130 {
2131         vma->vm_ops = &kvm_vcpu_vm_ops;
2132         return 0;
2133 }
2134 
2135 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2136 {
2137         struct kvm_vcpu *vcpu = filp->private_data;
2138 
2139         kvm_put_kvm(vcpu->kvm);
2140         return 0;
2141 }
2142 
2143 static struct file_operations kvm_vcpu_fops = {
2144         .release        = kvm_vcpu_release,
2145         .unlocked_ioctl = kvm_vcpu_ioctl,
2146 #ifdef CONFIG_KVM_COMPAT
2147         .compat_ioctl   = kvm_vcpu_compat_ioctl,
2148 #endif
2149         .mmap           = kvm_vcpu_mmap,
2150         .llseek         = noop_llseek,
2151 };
2152 
2153 /*
2154  * Allocates an inode for the vcpu.
2155  */
2156 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2157 {
2158         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
2159 }
2160 
2161 /*
2162  * Creates some virtual cpus.  Good luck creating more than one.
2163  */
2164 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
2165 {
2166         int r;
2167         struct kvm_vcpu *vcpu, *v;
2168 
2169         if (id >= KVM_MAX_VCPUS)
2170                 return -EINVAL;
2171 
2172         vcpu = kvm_arch_vcpu_create(kvm, id);
2173         if (IS_ERR(vcpu))
2174                 return PTR_ERR(vcpu);
2175 
2176         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2177 
2178         r = kvm_arch_vcpu_setup(vcpu);
2179         if (r)
2180                 goto vcpu_destroy;
2181 
2182         mutex_lock(&kvm->lock);
2183         if (!kvm_vcpu_compatible(vcpu)) {
2184                 r = -EINVAL;
2185                 goto unlock_vcpu_destroy;
2186         }
2187         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
2188                 r = -EINVAL;
2189                 goto unlock_vcpu_destroy;
2190         }
2191 
2192         kvm_for_each_vcpu(r, v, kvm)
2193                 if (v->vcpu_id == id) {
2194                         r = -EEXIST;
2195                         goto unlock_vcpu_destroy;
2196                 }
2197 
2198         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
2199 
2200         /* Now it's all set up, let userspace reach it */
2201         kvm_get_kvm(kvm);
2202         r = create_vcpu_fd(vcpu);
2203         if (r < 0) {
2204                 kvm_put_kvm(kvm);
2205                 goto unlock_vcpu_destroy;
2206         }
2207 
2208         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
2209         smp_wmb();
2210         atomic_inc(&kvm->online_vcpus);
2211 
2212         mutex_unlock(&kvm->lock);
2213         kvm_arch_vcpu_postcreate(vcpu);
2214         return r;
2215 
2216 unlock_vcpu_destroy:
2217         mutex_unlock(&kvm->lock);
2218 vcpu_destroy:
2219         kvm_arch_vcpu_destroy(vcpu);
2220         return r;
2221 }
2222 
2223 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2224 {
2225         if (sigset) {
2226                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2227                 vcpu->sigset_active = 1;
2228                 vcpu->sigset = *sigset;
2229         } else
2230                 vcpu->sigset_active = 0;
2231         return 0;
2232 }
2233 
2234 static long kvm_vcpu_ioctl(struct file *filp,
2235                            unsigned int ioctl, unsigned long arg)
2236 {
2237         struct kvm_vcpu *vcpu = filp->private_data;
2238         void __user *argp = (void __user *)arg;
2239         int r;
2240         struct kvm_fpu *fpu = NULL;
2241         struct kvm_sregs *kvm_sregs = NULL;
2242 
2243         if (vcpu->kvm->mm != current->mm)
2244                 return -EIO;
2245 
2246         if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2247                 return -EINVAL;
2248 
2249 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
2250         /*
2251          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
2252          * so vcpu_load() would break it.
2253          */
2254         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_S390_IRQ || ioctl == KVM_INTERRUPT)
2255                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2256 #endif
2257 
2258 
2259         r = vcpu_load(vcpu);
2260         if (r)
2261                 return r;
2262         switch (ioctl) {
2263         case KVM_RUN:
2264                 r = -EINVAL;
2265                 if (arg)
2266                         goto out;
2267                 if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
2268                         /* The thread running this VCPU changed. */
2269                         struct pid *oldpid = vcpu->pid;
2270                         struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
2271 
2272                         rcu_assign_pointer(vcpu->pid, newpid);
2273                         if (oldpid)
2274                                 synchronize_rcu();
2275                         put_pid(oldpid);
2276                 }
2277                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2278                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2279                 break;
2280         case KVM_GET_REGS: {
2281                 struct kvm_regs *kvm_regs;
2282 
2283                 r = -ENOMEM;
2284                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2285                 if (!kvm_regs)
2286                         goto out;
2287                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2288                 if (r)
2289                         goto out_free1;
2290                 r = -EFAULT;
2291                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2292                         goto out_free1;
2293                 r = 0;
2294 out_free1:
2295                 kfree(kvm_regs);
2296                 break;
2297         }
2298         case KVM_SET_REGS: {
2299                 struct kvm_regs *kvm_regs;
2300 
2301                 r = -ENOMEM;
2302                 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2303                 if (IS_ERR(kvm_regs)) {
2304                         r = PTR_ERR(kvm_regs);
2305                         goto out;
2306                 }
2307                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2308                 kfree(kvm_regs);
2309                 break;
2310         }
2311         case KVM_GET_SREGS: {
2312                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2313                 r = -ENOMEM;
2314                 if (!kvm_sregs)
2315                         goto out;
2316                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2317                 if (r)
2318                         goto out;
2319                 r = -EFAULT;
2320                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2321                         goto out;
2322                 r = 0;
2323                 break;
2324         }
2325         case KVM_SET_SREGS: {
2326                 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2327                 if (IS_ERR(kvm_sregs)) {
2328                         r = PTR_ERR(kvm_sregs);
2329                         kvm_sregs = NULL;
2330                         goto out;
2331                 }
2332                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2333                 break;
2334         }
2335         case KVM_GET_MP_STATE: {
2336                 struct kvm_mp_state mp_state;
2337 
2338                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2339                 if (r)
2340                         goto out;
2341                 r = -EFAULT;
2342                 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
2343                         goto out;
2344                 r = 0;
2345                 break;
2346         }
2347         case KVM_SET_MP_STATE: {
2348                 struct kvm_mp_state mp_state;
2349 
2350                 r = -EFAULT;
2351                 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
2352                         goto out;
2353                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2354                 break;
2355         }
2356         case KVM_TRANSLATE: {
2357                 struct kvm_translation tr;
2358 
2359                 r = -EFAULT;
2360                 if (copy_from_user(&tr, argp, sizeof(tr)))
2361                         goto out;
2362                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2363                 if (r)
2364                         goto out;
2365                 r = -EFAULT;
2366                 if (copy_to_user(argp, &tr, sizeof(tr)))
2367                         goto out;
2368                 r = 0;
2369                 break;
2370         }
2371         case KVM_SET_GUEST_DEBUG: {
2372                 struct kvm_guest_debug dbg;
2373 
2374                 r = -EFAULT;
2375                 if (copy_from_user(&dbg, argp, sizeof(dbg)))
2376                         goto out;
2377                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2378                 break;
2379         }
2380         case KVM_SET_SIGNAL_MASK: {
2381                 struct kvm_signal_mask __user *sigmask_arg = argp;
2382                 struct kvm_signal_mask kvm_sigmask;
2383                 sigset_t sigset, *p;
2384 
2385                 p = NULL;
2386                 if (argp) {
2387                         r = -EFAULT;
2388                         if (copy_from_user(&kvm_sigmask, argp,
2389                                            sizeof(kvm_sigmask)))
2390                                 goto out;
2391                         r = -EINVAL;
2392                         if (kvm_sigmask.len != sizeof(sigset))
2393                                 goto out;
2394                         r = -EFAULT;
2395                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2396                                            sizeof(sigset)))
2397                                 goto out;
2398                         p = &sigset;
2399                 }
2400                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2401                 break;
2402         }
2403         case KVM_GET_FPU: {
2404                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2405                 r = -ENOMEM;
2406                 if (!fpu)
2407                         goto out;
2408                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2409                 if (r)
2410                         goto out;
2411                 r = -EFAULT;
2412                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2413                         goto out;
2414                 r = 0;
2415                 break;
2416         }
2417         case KVM_SET_FPU: {
2418                 fpu = memdup_user(argp, sizeof(*fpu));
2419                 if (IS_ERR(fpu)) {
2420                         r = PTR_ERR(fpu);
2421                         fpu = NULL;
2422                         goto out;
2423                 }
2424                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2425                 break;
2426         }
2427         default:
2428                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2429         }
2430 out:
2431         vcpu_put(vcpu);
2432         kfree(fpu);
2433         kfree(kvm_sregs);
2434         return r;
2435 }
2436 
2437 #ifdef CONFIG_KVM_COMPAT
2438 static long kvm_vcpu_compat_ioctl(struct file *filp,
2439                                   unsigned int ioctl, unsigned long arg)
2440 {
2441         struct kvm_vcpu *vcpu = filp->private_data;
2442         void __user *argp = compat_ptr(arg);
2443         int r;
2444 
2445         if (vcpu->kvm->mm != current->mm)
2446                 return -EIO;
2447 
2448         switch (ioctl) {
2449         case KVM_SET_SIGNAL_MASK: {
2450                 struct kvm_signal_mask __user *sigmask_arg = argp;
2451                 struct kvm_signal_mask kvm_sigmask;
2452                 compat_sigset_t csigset;
2453                 sigset_t sigset;
2454 
2455                 if (argp) {
2456                         r = -EFAULT;
2457                         if (copy_from_user(&kvm_sigmask, argp,
2458                                            sizeof(kvm_sigmask)))
2459                                 goto out;
2460                         r = -EINVAL;
2461                         if (kvm_sigmask.len != sizeof(csigset))
2462                                 goto out;
2463                         r = -EFAULT;
2464                         if (copy_from_user(&csigset, sigmask_arg->sigset,
2465                                            sizeof(csigset)))
2466                                 goto out;
2467                         sigset_from_compat(&sigset, &csigset);
2468                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2469                 } else
2470                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2471                 break;
2472         }
2473         default:
2474                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2475         }
2476 
2477 out:
2478         return r;
2479 }
2480 #endif
2481 
2482 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2483                                  int (*accessor)(struct kvm_device *dev,
2484                                                  struct kvm_device_attr *attr),
2485                                  unsigned long arg)
2486 {
2487         struct kvm_device_attr attr;
2488 
2489         if (!accessor)
2490                 return -EPERM;
2491 
2492         if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2493                 return -EFAULT;
2494 
2495         return accessor(dev, &attr);
2496 }
2497 
2498 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2499                              unsigned long arg)
2500 {
2501         struct kvm_device *dev = filp->private_data;
2502 
2503         switch (ioctl) {
2504         case KVM_SET_DEVICE_ATTR:
2505                 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2506         case KVM_GET_DEVICE_ATTR:
2507                 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2508         case KVM_HAS_DEVICE_ATTR:
2509                 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2510         default:
2511                 if (dev->ops->ioctl)
2512                         return dev->ops->ioctl(dev, ioctl, arg);
2513 
2514                 return -ENOTTY;
2515         }
2516 }
2517 
2518 static int kvm_device_release(struct inode *inode, struct file *filp)
2519 {
2520         struct kvm_device *dev = filp->private_data;
2521         struct kvm *kvm = dev->kvm;
2522 
2523         kvm_put_kvm(kvm);
2524         return 0;
2525 }
2526 
2527 static const struct file_operations kvm_device_fops = {
2528         .unlocked_ioctl = kvm_device_ioctl,
2529 #ifdef CONFIG_KVM_COMPAT
2530         .compat_ioctl = kvm_device_ioctl,
2531 #endif
2532         .release = kvm_device_release,
2533 };
2534 
2535 struct kvm_device *kvm_device_from_filp(struct file *filp)
2536 {
2537         if (filp->f_op != &kvm_device_fops)
2538                 return NULL;
2539 
2540         return filp->private_data;
2541 }
2542 
2543 static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
2544 #ifdef CONFIG_KVM_MPIC
2545         [KVM_DEV_TYPE_FSL_MPIC_20]      = &kvm_mpic_ops,
2546         [KVM_DEV_TYPE_FSL_MPIC_42]      = &kvm_mpic_ops,
2547 #endif
2548 
2549 #ifdef CONFIG_KVM_XICS
2550         [KVM_DEV_TYPE_XICS]             = &kvm_xics_ops,
2551 #endif
2552 };
2553 
2554 int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
2555 {
2556         if (type >= ARRAY_SIZE(kvm_device_ops_table))
2557                 return -ENOSPC;
2558 
2559         if (kvm_device_ops_table[type] != NULL)
2560                 return -EEXIST;
2561 
2562         kvm_device_ops_table[type] = ops;
2563         return 0;
2564 }
2565 
2566 void kvm_unregister_device_ops(u32 type)
2567 {
2568         if (kvm_device_ops_table[type] != NULL)
2569                 kvm_device_ops_table[type] = NULL;
2570 }
2571 
2572 static int kvm_ioctl_create_device(struct kvm *kvm,
2573                                    struct kvm_create_device *cd)
2574 {
2575         struct kvm_device_ops *ops = NULL;
2576         struct kvm_device *dev;
2577         bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2578         int ret;
2579 
2580         if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
2581                 return -ENODEV;
2582 
2583         ops = kvm_device_ops_table[cd->type];
2584         if (ops == NULL)
2585                 return -ENODEV;
2586 
2587         if (test)
2588                 return 0;
2589 
2590         dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2591         if (!dev)
2592                 return -ENOMEM;
2593 
2594         dev->ops = ops;
2595         dev->kvm = kvm;
2596 
2597         ret = ops->create(dev, cd->type);
2598         if (ret < 0) {
2599                 kfree(dev);
2600                 return ret;
2601         }
2602 
2603         ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2604         if (ret < 0) {
2605                 ops->destroy(dev);
2606                 return ret;
2607         }
2608 
2609         list_add(&dev->vm_node, &kvm->devices);
2610         kvm_get_kvm(kvm);
2611         cd->fd = ret;
2612         return 0;
2613 }
2614 
2615 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2616 {
2617         switch (arg) {
2618         case KVM_CAP_USER_MEMORY:
2619         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2620         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2621 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2622         case KVM_CAP_SET_BOOT_CPU_ID:
2623 #endif
2624         case KVM_CAP_INTERNAL_ERROR_DATA:
2625 #ifdef CONFIG_HAVE_KVM_MSI
2626         case KVM_CAP_SIGNAL_MSI:
2627 #endif
2628 #ifdef CONFIG_HAVE_KVM_IRQFD
2629         case KVM_CAP_IRQFD:
2630         case KVM_CAP_IRQFD_RESAMPLE:
2631 #endif
2632         case KVM_CAP_CHECK_EXTENSION_VM:
2633                 return 1;
2634 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2635         case KVM_CAP_IRQ_ROUTING:
2636                 return KVM_MAX_IRQ_ROUTES;
2637 #endif
2638 #if KVM_ADDRESS_SPACE_NUM > 1
2639         case KVM_CAP_MULTI_ADDRESS_SPACE:
2640                 return KVM_ADDRESS_SPACE_NUM;
2641 #endif
2642         default:
2643                 break;
2644         }
2645         return kvm_vm_ioctl_check_extension(kvm, arg);
2646 }
2647 
2648 static long kvm_vm_ioctl(struct file *filp,
2649                            unsigned int ioctl, unsigned long arg)
2650 {
2651         struct kvm *kvm = filp->private_data;
2652         void __user *argp = (void __user *)arg;
2653         int r;
2654 
2655         if (kvm->mm != current->mm)
2656                 return -EIO;
2657         switch (ioctl) {
2658         case KVM_CREATE_VCPU:
2659                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2660                 break;
2661         case KVM_SET_USER_MEMORY_REGION: {
2662                 struct kvm_userspace_memory_region kvm_userspace_mem;
2663 
2664                 r = -EFAULT;
2665                 if (copy_from_user(&kvm_userspace_mem, argp,
2666                                                 sizeof(kvm_userspace_mem)))
2667                         goto out;
2668 
2669                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2670                 break;
2671         }
2672         case KVM_GET_DIRTY_LOG: {
2673                 struct kvm_dirty_log log;
2674 
2675                 r = -EFAULT;
2676                 if (copy_from_user(&log, argp, sizeof(log)))
2677                         goto out;
2678                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2679                 break;
2680         }
2681 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2682         case KVM_REGISTER_COALESCED_MMIO: {
2683                 struct kvm_coalesced_mmio_zone zone;
2684 
2685                 r = -EFAULT;
2686                 if (copy_from_user(&zone, argp, sizeof(zone)))
2687                         goto out;
2688                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2689                 break;
2690         }
2691         case KVM_UNREGISTER_COALESCED_MMIO: {
2692                 struct kvm_coalesced_mmio_zone zone;
2693 
2694                 r = -EFAULT;
2695                 if (copy_from_user(&zone, argp, sizeof(zone)))
2696                         goto out;
2697                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2698                 break;
2699         }
2700 #endif
2701         case KVM_IRQFD: {
2702                 struct kvm_irqfd data;
2703 
2704                 r = -EFAULT;
2705                 if (copy_from_user(&data, argp, sizeof(data)))
2706                         goto out;
2707                 r = kvm_irqfd(kvm, &data);
2708                 break;
2709         }
2710         case KVM_IOEVENTFD: {
2711                 struct kvm_ioeventfd data;
2712 
2713                 r = -EFAULT;
2714                 if (copy_from_user(&data, argp, sizeof(data)))
2715                         goto out;
2716                 r = kvm_ioeventfd(kvm, &data);
2717                 break;
2718         }
2719 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2720         case KVM_SET_BOOT_CPU_ID:
2721                 r = 0;
2722                 mutex_lock(&kvm->lock);
2723                 if (atomic_read(&kvm->online_vcpus) != 0)
2724                         r = -EBUSY;
2725                 else
2726                         kvm->bsp_vcpu_id = arg;
2727                 mutex_unlock(&kvm->lock);
2728                 break;
2729 #endif
2730 #ifdef CONFIG_HAVE_KVM_MSI
2731         case KVM_SIGNAL_MSI: {
2732                 struct kvm_msi msi;
2733 
2734                 r = -EFAULT;
2735                 if (copy_from_user(&msi, argp, sizeof(msi)))
2736                         goto out;
2737                 r = kvm_send_userspace_msi(kvm, &msi);
2738                 break;
2739         }
2740 #endif
2741 #ifdef __KVM_HAVE_IRQ_LINE
2742         case KVM_IRQ_LINE_STATUS:
2743         case KVM_IRQ_LINE: {
2744                 struct kvm_irq_level irq_event;
2745 
2746                 r = -EFAULT;
2747                 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
2748                         goto out;
2749 
2750                 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
2751                                         ioctl == KVM_IRQ_LINE_STATUS);
2752                 if (r)
2753                         goto out;
2754 
2755                 r = -EFAULT;
2756                 if (ioctl == KVM_IRQ_LINE_STATUS) {
2757                         if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
2758                                 goto out;
2759                 }
2760 
2761                 r = 0;
2762                 break;
2763         }
2764 #endif
2765 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2766         case KVM_SET_GSI_ROUTING: {
2767                 struct kvm_irq_routing routing;
2768                 struct kvm_irq_routing __user *urouting;
2769                 struct kvm_irq_routing_entry *entries;
2770 
2771                 r = -EFAULT;
2772                 if (copy_from_user(&routing, argp, sizeof(routing)))
2773                         goto out;
2774                 r = -EINVAL;
2775                 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2776                         goto out;
2777                 if (routing.flags)
2778                         goto out;
2779                 r = -ENOMEM;
2780                 entries = vmalloc(routing.nr * sizeof(*entries));
2781                 if (!entries)
2782                         goto out;
2783                 r = -EFAULT;
2784                 urouting = argp;
2785                 if (copy_from_user(entries, urouting->entries,
2786                                    routing.nr * sizeof(*entries)))
2787                         goto out_free_irq_routing;
2788                 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2789                                         routing.flags);
2790 out_free_irq_routing:
2791                 vfree(entries);
2792                 break;
2793         }
2794 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
2795         case KVM_CREATE_DEVICE: {
2796                 struct kvm_create_device cd;
2797 
2798                 r = -EFAULT;
2799                 if (copy_from_user(&cd, argp, sizeof(cd)))
2800                         goto out;
2801 
2802                 r = kvm_ioctl_create_device(kvm, &cd);
2803                 if (r)
2804                         goto out;
2805 
2806                 r = -EFAULT;
2807                 if (copy_to_user(argp, &cd, sizeof(cd)))
2808                         goto out;
2809 
2810                 r = 0;
2811                 break;
2812         }
2813         case KVM_CHECK_EXTENSION:
2814                 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
2815                 break;
2816         default:
2817                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2818         }
2819 out:
2820         return r;
2821 }
2822 
2823 #ifdef CONFIG_KVM_COMPAT
2824 struct compat_kvm_dirty_log {
2825         __u32 slot;
2826         __u32 padding1;
2827         union {
2828                 compat_uptr_t dirty_bitmap; /* one bit per page */
2829                 __u64 padding2;
2830         };
2831 };
2832 
2833 static long kvm_vm_compat_ioctl(struct file *filp,
2834                            unsigned int ioctl, unsigned long arg)
2835 {
2836         struct kvm *kvm = filp->private_data;
2837         int r;
2838 
2839         if (kvm->mm != current->mm)
2840                 return -EIO;
2841         switch (ioctl) {
2842         case KVM_GET_DIRTY_LOG: {
2843                 struct compat_kvm_dirty_log compat_log;
2844                 struct kvm_dirty_log log;
2845 
2846                 r = -EFAULT;
2847                 if (copy_from_user(&compat_log, (void __user *)arg,
2848                                    sizeof(compat_log)))
2849                         goto out;
2850                 log.slot         = compat_log.slot;
2851                 log.padding1     = compat_log.padding1;
2852                 log.padding2     = compat_log.padding2;
2853                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2854 
2855                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2856                 break;
2857         }
2858         default:
2859                 r = kvm_vm_ioctl(filp, ioctl, arg);
2860         }
2861 
2862 out:
2863         return r;
2864 }
2865 #endif
2866 
2867 static struct file_operations kvm_vm_fops = {
2868         .release        = kvm_vm_release,
2869         .unlocked_ioctl = kvm_vm_ioctl,
2870 #ifdef CONFIG_KVM_COMPAT
2871         .compat_ioctl   = kvm_vm_compat_ioctl,
2872 #endif
2873         .llseek         = noop_llseek,
2874 };
2875 
2876 static int kvm_dev_ioctl_create_vm(unsigned long type)
2877 {
2878         int r;
2879         struct kvm *kvm;
2880 
2881         kvm = kvm_create_vm(type);
2882         if (IS_ERR(kvm))
2883                 return PTR_ERR(kvm);
2884 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2885         r = kvm_coalesced_mmio_init(kvm);
2886         if (r < 0) {
2887                 kvm_put_kvm(kvm);
2888                 return r;
2889         }
2890 #endif
2891         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC);
2892         if (r < 0)
2893                 kvm_put_kvm(kvm);
2894 
2895         return r;
2896 }
2897 
2898 static long kvm_dev_ioctl(struct file *filp,
2899                           unsigned int ioctl, unsigned long arg)
2900 {
2901         long r = -EINVAL;
2902 
2903         switch (ioctl) {
2904         case KVM_GET_API_VERSION:
2905                 if (arg)
2906                         goto out;
2907                 r = KVM_API_VERSION;
2908                 break;
2909         case KVM_CREATE_VM:
2910                 r = kvm_dev_ioctl_create_vm(arg);
2911                 break;
2912         case KVM_CHECK_EXTENSION:
2913                 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
2914                 break;
2915         case KVM_GET_VCPU_MMAP_SIZE:
2916                 if (arg)
2917                         goto out;
2918                 r = PAGE_SIZE;     /* struct kvm_run */
2919 #ifdef CONFIG_X86
2920                 r += PAGE_SIZE;    /* pio data page */
2921 #endif
2922 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2923                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2924 #endif
2925                 break;
2926         case KVM_TRACE_ENABLE:
2927         case KVM_TRACE_PAUSE:
2928         case KVM_TRACE_DISABLE:
2929                 r = -EOPNOTSUPP;
2930                 break;
2931         default:
2932                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2933         }
2934 out:
2935         return r;
2936 }
2937 
2938 static struct file_operations kvm_chardev_ops = {
2939         .unlocked_ioctl = kvm_dev_ioctl,
2940         .compat_ioctl   = kvm_dev_ioctl,
2941         .llseek         = noop_llseek,
2942 };
2943 
2944 static struct miscdevice kvm_dev = {
2945         KVM_MINOR,
2946         "kvm",
2947         &kvm_chardev_ops,
2948 };
2949 
2950 static void hardware_enable_nolock(void *junk)
2951 {
2952         int cpu = raw_smp_processor_id();
2953         int r;
2954 
2955         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2956                 return;
2957 
2958         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2959 
2960         r = kvm_arch_hardware_enable();
2961 
2962         if (r) {
2963                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2964                 atomic_inc(&hardware_enable_failed);
2965                 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
2966         }
2967 }
2968 
2969 static void hardware_enable(void)
2970 {
2971         raw_spin_lock(&kvm_count_lock);
2972         if (kvm_usage_count)
2973                 hardware_enable_nolock(NULL);
2974         raw_spin_unlock(&kvm_count_lock);
2975 }
2976 
2977 static void hardware_disable_nolock(void *junk)
2978 {
2979         int cpu = raw_smp_processor_id();
2980 
2981         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2982                 return;
2983         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2984         kvm_arch_hardware_disable();
2985 }
2986 
2987 static void hardware_disable(void)
2988 {
2989         raw_spin_lock(&kvm_count_lock);
2990         if (kvm_usage_count)
2991                 hardware_disable_nolock(NULL);
2992         raw_spin_unlock(&kvm_count_lock);
2993 }
2994 
2995 static void hardware_disable_all_nolock(void)
2996 {
2997         BUG_ON(!kvm_usage_count);
2998 
2999         kvm_usage_count--;
3000         if (!kvm_usage_count)
3001                 on_each_cpu(hardware_disable_nolock, NULL, 1);
3002 }
3003 
3004 static void hardware_disable_all(void)
3005 {
3006         raw_spin_lock(&kvm_count_lock);
3007         hardware_disable_all_nolock();
3008         raw_spin_unlock(&kvm_count_lock);
3009 }
3010 
3011 static int hardware_enable_all(void)
3012 {
3013         int r = 0;
3014 
3015         raw_spin_lock(&kvm_count_lock);
3016 
3017         kvm_usage_count++;
3018         if (kvm_usage_count == 1) {
3019                 atomic_set(&hardware_enable_failed, 0);
3020                 on_each_cpu(hardware_enable_nolock, NULL, 1);
3021 
3022                 if (atomic_read(&hardware_enable_failed)) {
3023                         hardware_disable_all_nolock();
3024                         r = -EBUSY;
3025                 }
3026         }
3027 
3028         raw_spin_unlock(&kvm_count_lock);
3029 
3030         return r;
3031 }
3032 
3033 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
3034                            void *v)
3035 {
3036         val &= ~CPU_TASKS_FROZEN;
3037         switch (val) {
3038         case CPU_DYING:
3039                 hardware_disable();
3040                 break;
3041         case CPU_STARTING:
3042                 hardware_enable();
3043                 break;
3044         }
3045         return NOTIFY_OK;
3046 }
3047 
3048 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3049                       void *v)
3050 {
3051         /*
3052          * Some (well, at least mine) BIOSes hang on reboot if
3053          * in vmx root mode.
3054          *
3055          * And Intel TXT required VMX off for all cpu when system shutdown.
3056          */
3057         pr_info("kvm: exiting hardware virtualization\n");
3058         kvm_rebooting = true;
3059         on_each_cpu(hardware_disable_nolock, NULL, 1);
3060         return NOTIFY_OK;
3061 }
3062 
3063 static struct notifier_block kvm_reboot_notifier = {
3064         .notifier_call = kvm_reboot,
3065         .priority = 0,
3066 };
3067 
3068 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3069 {
3070         int i;
3071 
3072         for (i = 0; i < bus->dev_count; i++) {
3073                 struct kvm_io_device *pos = bus->range[i].dev;
3074 
3075                 kvm_iodevice_destructor(pos);
3076         }
3077         kfree(bus);
3078 }
3079 
3080 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
3081                                  const struct kvm_io_range *r2)
3082 {
3083         gpa_t addr1 = r1->addr;
3084         gpa_t addr2 = r2->addr;
3085 
3086         if (addr1 < addr2)
3087                 return -1;
3088 
3089         /* If r2->len == 0, match the exact address.  If r2->len != 0,
3090          * accept any overlapping write.  Any order is acceptable for
3091          * overlapping ranges, because kvm_io_bus_get_first_dev ensures
3092          * we process all of them.
3093          */
3094         if (r2->len) {
3095                 addr1 += r1->len;
3096                 addr2 += r2->len;
3097         }
3098 
3099         if (addr1 > addr2)
3100                 return 1;
3101 
3102         return 0;
3103 }
3104 
3105 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
3106 {
3107         return kvm_io_bus_cmp(p1, p2);
3108 }
3109 
3110 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
3111                           gpa_t addr, int len)
3112 {
3113         bus->range[bus->dev_count++] = (struct kvm_io_range) {
3114                 .addr = addr,
3115                 .len = len,
3116                 .dev = dev,
3117         };
3118 
3119         sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
3120                 kvm_io_bus_sort_cmp, NULL);
3121 
3122         return 0;
3123 }
3124 
3125 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
3126                              gpa_t addr, int len)
3127 {
3128         struct kvm_io_range *range, key;
3129         int off;
3130 
3131         key = (struct kvm_io_range) {
3132                 .addr = addr,
3133                 .len = len,
3134         };
3135 
3136         range = bsearch(&key, bus->range, bus->dev_count,
3137                         sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
3138         if (range == NULL)
3139                 return -ENOENT;
3140 
3141         off = range - bus->range;
3142 
3143         while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
3144                 off--;
3145 
3146         return off;
3147 }
3148 
3149 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3150                               struct kvm_io_range *range, const void *val)
3151 {
3152         int idx;
3153 
3154         idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3155         if (idx < 0)
3156                 return -EOPNOTSUPP;
3157 
3158         while (idx < bus->dev_count &&
3159                 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3160                 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
3161                                         range->len, val))
3162                         return idx;
3163                 idx++;
3164         }
3165 
3166         return -EOPNOTSUPP;
3167 }
3168 
3169 /* kvm_io_bus_write - called under kvm->slots_lock */
3170 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3171                      int len, const void *val)
3172 {
3173         struct kvm_io_bus *bus;
3174         struct kvm_io_range range;
3175         int r;
3176 
3177         range = (struct kvm_io_range) {
3178                 .addr = addr,
3179                 .len = len,
3180         };
3181 
3182         bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3183         r = __kvm_io_bus_write(vcpu, bus, &range, val);
3184         return r < 0 ? r : 0;
3185 }
3186 
3187 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3188 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
3189                             gpa_t addr, int len, const void *val, long cookie)
3190 {
3191         struct kvm_io_bus *bus;
3192         struct kvm_io_range range;
3193 
3194         range = (struct kvm_io_range) {
3195                 .addr = addr,
3196                 .len = len,
3197         };
3198 
3199         bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3200 
3201         /* First try the device referenced by cookie. */
3202         if ((cookie >= 0) && (cookie < bus->dev_count) &&
3203             (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
3204                 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
3205                                         val))
3206                         return cookie;
3207 
3208         /*
3209          * cookie contained garbage; fall back to search and return the
3210          * correct cookie value.
3211          */
3212         return __kvm_io_bus_write(vcpu, bus, &range, val);
3213 }
3214 
3215 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3216                              struct kvm_io_range *range, void *val)
3217 {
3218         int idx;
3219 
3220         idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3221         if (idx < 0)
3222                 return -EOPNOTSUPP;
3223 
3224         while (idx < bus->dev_count &&
3225                 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3226                 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
3227                                        range->len, val))
3228                         return idx;
3229                 idx++;
3230         }
3231 
3232         return -EOPNOTSUPP;
3233 }
3234 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
3235 
3236 /* kvm_io_bus_read - called under kvm->slots_lock */
3237 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3238                     int len, void *val)
3239 {
3240         struct kvm_io_bus *bus;
3241         struct kvm_io_range range;
3242         int r;
3243 
3244         range = (struct kvm_io_range) {
3245                 .addr = addr,
3246                 .len = len,
3247         };
3248 
3249         bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3250         r = __kvm_io_bus_read(vcpu, bus, &range, val);
3251         return r < 0 ? r : 0;
3252 }
3253 
3254 
3255 /* Caller must hold slots_lock. */
3256 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3257                             int len, struct kvm_io_device *dev)
3258 {
3259         struct kvm_io_bus *new_bus, *bus;
3260 
3261         bus = kvm->buses[bus_idx];
3262         /* exclude ioeventfd which is limited by maximum fd */
3263         if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3264                 return -ENOSPC;
3265 
3266         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
3267                           sizeof(struct kvm_io_range)), GFP_KERNEL);
3268         if (!new_bus)
3269                 return -ENOMEM;
3270         memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
3271                sizeof(struct kvm_io_range)));
3272         kvm_io_bus_insert_dev(new_bus, dev, addr, len);
3273         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3274         synchronize_srcu_expedited(&kvm->srcu);
3275         kfree(bus);
3276 
3277         return 0;
3278 }
3279 
3280 /* Caller must hold slots_lock. */
3281 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3282                               struct kvm_io_device *dev)
3283 {
3284         int i, r;
3285         struct kvm_io_bus *new_bus, *bus;
3286 
3287         bus = kvm->buses[bus_idx];
3288         r = -ENOENT;
3289         for (i = 0; i < bus->dev_count; i++)
3290                 if (bus->range[i].dev == dev) {
3291                         r = 0;
3292                         break;
3293                 }
3294 
3295         if (r)
3296                 return r;
3297 
3298         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3299                           sizeof(struct kvm_io_range)), GFP_KERNEL);
3300         if (!new_bus)
3301                 return -ENOMEM;
3302 
3303         memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3304         new_bus->dev_count--;
3305         memcpy(new_bus->range + i, bus->range + i + 1,
3306                (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3307 
3308         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3309         synchronize_srcu_expedited(&kvm->srcu);
3310         kfree(bus);
3311         return r;
3312 }
3313 
3314 static struct notifier_block kvm_cpu_notifier = {
3315         .notifier_call = kvm_cpu_hotplug,
3316 };
3317 
3318 static int vm_stat_get(void *_offset, u64 *val)
3319 {
3320         unsigned offset = (long)_offset;
3321         struct kvm *kvm;
3322 
3323         *val = 0;
3324         spin_lock(&kvm_lock);
3325         list_for_each_entry(kvm, &vm_list, vm_list)
3326                 *val += *(u32 *)((void *)kvm + offset);
3327         spin_unlock(&kvm_lock);
3328         return 0;
3329 }
3330 
3331 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
3332 
3333 static int vcpu_stat_get(void *_offset, u64 *val)
3334 {
3335         unsigned offset = (long)_offset;
3336         struct kvm *kvm;
3337         struct kvm_vcpu *vcpu;
3338         int i;
3339 
3340         *val = 0;
3341         spin_lock(&kvm_lock);
3342         list_for_each_entry(kvm, &vm_list, vm_list)
3343                 kvm_for_each_vcpu(i, vcpu, kvm)
3344                         *val += *(u32 *)((void *)vcpu + offset);
3345 
3346         spin_unlock(&kvm_lock);
3347         return 0;
3348 }
3349 
3350 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
3351 
3352 static const struct file_operations *stat_fops[] = {
3353         [KVM_STAT_VCPU] = &vcpu_stat_fops,
3354         [KVM_STAT_VM]   = &vm_stat_fops,
3355 };
3356 
3357 static int kvm_init_debug(void)
3358 {
3359         int r = -EEXIST;
3360         struct kvm_stats_debugfs_item *p;
3361 
3362         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3363         if (kvm_debugfs_dir == NULL)
3364                 goto out;
3365 
3366         for (p = debugfs_entries; p->name; ++p) {
3367                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
3368                                                 (void *)(long)p->offset,
3369                                                 stat_fops[p->kind]);
3370                 if (p->dentry == NULL)
3371                         goto out_dir;
3372         }
3373 
3374         return 0;
3375 
3376 out_dir:
3377         debugfs_remove_recursive(kvm_debugfs_dir);
3378 out:
3379         return r;
3380 }
3381 
3382 static void kvm_exit_debug(void)
3383 {
3384         struct kvm_stats_debugfs_item *p;
3385 
3386         for (p = debugfs_entries; p->name; ++p)
3387                 debugfs_remove(p->dentry);
3388         debugfs_remove(kvm_debugfs_dir);
3389 }
3390 
3391 static int kvm_suspend(void)
3392 {
3393         if (kvm_usage_count)
3394                 hardware_disable_nolock(NULL);
3395         return 0;
3396 }
3397 
3398 static void kvm_resume(void)
3399 {
3400         if (kvm_usage_count) {
3401                 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3402                 hardware_enable_nolock(NULL);
3403         }
3404 }
3405 
3406 static struct syscore_ops kvm_syscore_ops = {
3407         .suspend = kvm_suspend,
3408         .resume = kvm_resume,
3409 };
3410 
3411 static inline
3412 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3413 {
3414         return container_of(pn, struct kvm_vcpu, preempt_notifier);
3415 }
3416 
3417 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3418 {
3419         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3420 
3421         if (vcpu->preempted)
3422                 vcpu->preempted = false;
3423 
3424         kvm_arch_sched_in(vcpu, cpu);
3425 
3426         kvm_arch_vcpu_load(vcpu, cpu);
3427 }
3428 
3429 static void kvm_sched_out(struct preempt_notifier *pn,
3430                           struct task_struct *next)
3431 {
3432         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3433 
3434         if (current->state == TASK_RUNNING)
3435                 vcpu->preempted = true;
3436         kvm_arch_vcpu_put(vcpu);
3437 }
3438 
3439 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3440                   struct module *module)
3441 {
3442         int r;
3443         int cpu;
3444 
3445         r = kvm_arch_init(opaque);
3446         if (r)
3447                 goto out_fail;
3448 
3449         /*
3450          * kvm_arch_init makes sure there's at most one caller
3451          * for architectures that support multiple implementations,
3452          * like intel and amd on x86.
3453          * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3454          * conflicts in case kvm is already setup for another implementation.
3455          */
3456         r = kvm_irqfd_init();
3457         if (r)
3458                 goto out_irqfd;
3459 
3460         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
3461                 r = -ENOMEM;
3462                 goto out_free_0;
3463         }
3464 
3465         r = kvm_arch_hardware_setup();
3466         if (r < 0)
3467                 goto out_free_0a;
3468 
3469         for_each_online_cpu(cpu) {
3470                 smp_call_function_single(cpu,
3471                                 kvm_arch_check_processor_compat,
3472                                 &r, 1);
3473                 if (r < 0)
3474                         goto out_free_1;
3475         }
3476 
3477         r = register_cpu_notifier(&kvm_cpu_notifier);
3478         if (r)
3479                 goto out_free_2;
3480         register_reboot_notifier(&kvm_reboot_notifier);
3481 
3482         /* A kmem cache lets us meet the alignment requirements of fx_save. */
3483         if (!vcpu_align)
3484                 vcpu_align = __alignof__(struct kvm_vcpu);
3485         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
3486                                            0, NULL);
3487         if (!kvm_vcpu_cache) {
3488                 r = -ENOMEM;
3489                 goto out_free_3;
3490         }
3491 
3492         r = kvm_async_pf_init();
3493         if (r)
3494                 goto out_free;
3495 
3496         kvm_chardev_ops.owner = module;
3497         kvm_vm_fops.owner = module;
3498         kvm_vcpu_fops.owner = module;
3499 
3500         r = misc_register(&kvm_dev);
3501         if (r) {
3502                 pr_err("kvm: misc device register failed\n");
3503                 goto out_unreg;
3504         }
3505 
3506         register_syscore_ops(&kvm_syscore_ops);
3507 
3508         kvm_preempt_ops.sched_in = kvm_sched_in;
3509         kvm_preempt_ops.sched_out = kvm_sched_out;
3510 
3511         r = kvm_init_debug();
3512         if (r) {
3513                 pr_err("kvm: create debugfs files failed\n");
3514                 goto out_undebugfs;
3515         }
3516 
3517         r = kvm_vfio_ops_init();
3518         WARN_ON(r);
3519 
3520         return 0;
3521 
3522 out_undebugfs:
3523         unregister_syscore_ops(&kvm_syscore_ops);
3524         misc_deregister(&kvm_dev);
3525 out_unreg:
3526         kvm_async_pf_deinit();
3527 out_free:
3528         kmem_cache_destroy(kvm_vcpu_cache);
3529 out_free_3:
3530         unregister_reboot_notifier(&kvm_reboot_notifier);
3531         unregister_cpu_notifier(&kvm_cpu_notifier);
3532 out_free_2:
3533 out_free_1:
3534         kvm_arch_hardware_unsetup();
3535 out_free_0a:
3536         free_cpumask_var(cpus_hardware_enabled);
3537 out_free_0:
3538         kvm_irqfd_exit();
3539 out_irqfd:
3540         kvm_arch_exit();
3541 out_fail:
3542         return r;
3543 }
3544 EXPORT_SYMBOL_GPL(kvm_init);
3545 
3546 void kvm_exit(void)
3547 {
3548         kvm_exit_debug();
3549         misc_deregister(&kvm_dev);
3550         kmem_cache_destroy(kvm_vcpu_cache);
3551         kvm_async_pf_deinit();
3552         unregister_syscore_ops(&kvm_syscore_ops);
3553         unregister_reboot_notifier(&kvm_reboot_notifier);
3554         unregister_cpu_notifier(&kvm_cpu_notifier);
3555         on_each_cpu(hardware_disable_nolock, NULL, 1);
3556         kvm_arch_hardware_unsetup();
3557         kvm_arch_exit();
3558         kvm_irqfd_exit();
3559         free_cpumask_var(cpus_hardware_enabled);
3560         kvm_vfio_ops_exit();
3561 }
3562 EXPORT_SYMBOL_GPL(kvm_exit);
3563 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp