virt/kvm/kvm_main.c

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * This module enables machines with Intel VT-x extensions to run virtual
   5  * machines without emulation or binary translation.
   6  *
   7  * Copyright (C) 2006 Qumranet, Inc.
   8  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
   9  *
  10  * Authors:
  11  *   Avi Kivity   <avi@qumranet.com>
  12  *   Yaniv Kamay  <yaniv@qumranet.com>
  13  *
  14  * This work is licensed under the terms of the GNU GPL, version 2.  See
  15  * the COPYING file in the top-level directory.
  16  *
  17  */
  18
  19 #include "iodev.h"
  20
  21 #include <linux/kvm_host.h>
  22 #include <linux/kvm.h>
  23 #include <linux/module.h>
  24 #include <linux/errno.h>
  25 #include <linux/percpu.h>
  26 #include <linux/mm.h>
  27 #include <linux/miscdevice.h>
  28 #include <linux/vmalloc.h>
  29 #include <linux/reboot.h>
  30 #include <linux/debugfs.h>
  31 #include <linux/highmem.h>
  32 #include <linux/file.h>
  33 #include <linux/syscore_ops.h>
  34 #include <linux/cpu.h>
  35 #include <linux/sched.h>
  36 #include <linux/cpumask.h>
  37 #include <linux/smp.h>
  38 #include <linux/anon_inodes.h>
  39 #include <linux/profile.h>
  40 #include <linux/kvm_para.h>
  41 #include <linux/pagemap.h>
  42 #include <linux/mman.h>
  43 #include <linux/swap.h>
  44 #include <linux/bitops.h>
  45 #include <linux/spinlock.h>
  46 #include <linux/compat.h>
  47 #include <linux/srcu.h>
  48 #include <linux/hugetlb.h>
  49 #include <linux/slab.h>
  50 #include <linux/sort.h>
  51 #include <linux/bsearch.h>
  52
  53 #include <asm/processor.h>
  54 #include <asm/io.h>
  55 #include <asm/uaccess.h>
  56 #include <asm/pgtable.h>
  57
  58 #include "coalesced_mmio.h"
  59 #include "async_pf.h"
  60
  61 #define CREATE_TRACE_POINTS
  62 #include <trace/events/kvm.h>
  63
  64 MODULE_AUTHOR("Qumranet");
  65 MODULE_LICENSE("GPL");
  66
  67 /*
  68  * Ordering of locks:
  69  *
  70  *              kvm->lock --> kvm->slots_lock --> kvm->irq_lock
  71  */
  72
  73 DEFINE_SPINLOCK(kvm_lock);
  74 static DEFINE_RAW_SPINLOCK(kvm_count_lock);
  75 LIST_HEAD(vm_list);
  76
  77 static cpumask_var_t cpus_hardware_enabled;
  78 static int kvm_usage_count = 0;
  79 static atomic_t hardware_enable_failed;
  80
  81 struct kmem_cache *kvm_vcpu_cache;
  82 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
  83
  84 static __read_mostly struct preempt_ops kvm_preempt_ops;
  85
  86 struct dentry *kvm_debugfs_dir;
  87
  88 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
  89                            unsigned long arg);
  90 #ifdef CONFIG_COMPAT
  91 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
  92                                   unsigned long arg);
  93 #endif
  94 static int hardware_enable_all(void);
  95 static void hardware_disable_all(void);
  96
  97 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
  98 static void update_memslots(struct kvm_memslots *slots,
  99                             struct kvm_memory_slot *new, u64 last_generation);
 100
 101 static void kvm_release_pfn_dirty(pfn_t pfn);
 102 static void mark_page_dirty_in_slot(struct kvm *kvm,
 103                                     struct kvm_memory_slot *memslot, gfn_t gfn);
 104
 105 __visible bool kvm_rebooting;
 106 EXPORT_SYMBOL_GPL(kvm_rebooting);
 107
 108 static bool largepages_enabled = true;
 109
 110 bool kvm_is_mmio_pfn(pfn_t pfn)
 111 {
 112         if (pfn_valid(pfn))
 113                 return PageReserved(pfn_to_page(pfn));
 114
 115         return true;
 116 }
 117
 118 /*
 119  * Switches to specified vcpu, until a matching vcpu_put()
 120  */
 121 int vcpu_load(struct kvm_vcpu *vcpu)
 122 {
 123         int cpu;
 124
 125         if (mutex_lock_killable(&vcpu->mutex))
 126                 return -EINTR;
 127         if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
 128                 /* The thread running this VCPU changed. */
 129                 struct pid *oldpid = vcpu->pid;
 130                 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
 131                 rcu_assign_pointer(vcpu->pid, newpid);
 132                 if (oldpid)
 133                         synchronize_rcu();
 134                 put_pid(oldpid);
 135         }
 136         cpu = get_cpu();
 137         preempt_notifier_register(&vcpu->preempt_notifier);
 138         kvm_arch_vcpu_load(vcpu, cpu);
 139         put_cpu();
 140         return 0;
 141 }
 142
 143 void vcpu_put(struct kvm_vcpu *vcpu)
 144 {
 145         preempt_disable();
 146         kvm_arch_vcpu_put(vcpu);
 147         preempt_notifier_unregister(&vcpu->preempt_notifier);
 148         preempt_enable();
 149         mutex_unlock(&vcpu->mutex);
 150 }
 151
 152 static void ack_flush(void *_completed)
 153 {
 154 }
 155
 156 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
 157 {
 158         int i, cpu, me;
 159         cpumask_var_t cpus;
 160         bool called = true;
 161         struct kvm_vcpu *vcpu;
 162
 163         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
 164
 165         me = get_cpu();
 166         kvm_for_each_vcpu(i, vcpu, kvm) {
 167                 kvm_make_request(req, vcpu);
 168                 cpu = vcpu->cpu;
 169
 170                 /* Set ->requests bit before we read ->mode */
 171                 smp_mb();
 172
 173                 if (cpus != NULL && cpu != -1 && cpu != me &&
 174                       kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
 175                         cpumask_set_cpu(cpu, cpus);
 176         }
 177         if (unlikely(cpus == NULL))
 178                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
 179         else if (!cpumask_empty(cpus))
 180                 smp_call_function_many(cpus, ack_flush, NULL, 1);
 181         else
 182                 called = false;
 183         put_cpu();
 184         free_cpumask_var(cpus);
 185         return called;
 186 }
 187
 188 void kvm_flush_remote_tlbs(struct kvm *kvm)
 189 {
 190         long dirty_count = kvm->tlbs_dirty;
 191
 192         smp_mb();
 193         if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
 194                 ++kvm->stat.remote_tlb_flush;
 195         cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
 196 }
 197 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
 198
 199 void kvm_reload_remote_mmus(struct kvm *kvm)
 200 {
 201         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
 202 }
 203
 204 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
 205 {
 206         make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
 207 }
 208
 209 void kvm_make_scan_ioapic_request(struct kvm *kvm)
 210 {
 211         make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
 212 }
 213
 214 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 215 {
 216         struct page *page;
 217         int r;
 218
 219         mutex_init(&vcpu->mutex);
 220         vcpu->cpu = -1;
 221         vcpu->kvm = kvm;
 222         vcpu->vcpu_id = id;
 223         vcpu->pid = NULL;
 224         init_waitqueue_head(&vcpu->wq);
 225         kvm_async_pf_vcpu_init(vcpu);
 226
 227         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 228         if (!page) {
 229                 r = -ENOMEM;
 230                 goto fail;
 231         }
 232         vcpu->run = page_address(page);
 233
 234         kvm_vcpu_set_in_spin_loop(vcpu, false);
 235         kvm_vcpu_set_dy_eligible(vcpu, false);
 236         vcpu->preempted = false;
 237
 238         r = kvm_arch_vcpu_init(vcpu);
 239         if (r < 0)
 240                 goto fail_free_run;
 241         return 0;
 242
 243 fail_free_run:
 244         free_page((unsigned long)vcpu->run);
 245 fail:
 246         return r;
 247 }
 248 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 249
 250 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 251 {
 252         put_pid(vcpu->pid);
 253         kvm_arch_vcpu_uninit(vcpu);
 254         free_page((unsigned long)vcpu->run);
 255 }
 256 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 257
 258 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 259 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
 260 {
 261         return container_of(mn, struct kvm, mmu_notifier);
 262 }
 263
 264 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
 265                                              struct mm_struct *mm,
 266                                              unsigned long address)
 267 {
 268         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 269         int need_tlb_flush, idx;
 270
 271         /*
 272          * When ->invalidate_page runs, the linux pte has been zapped
 273          * already but the page is still allocated until
 274          * ->invalidate_page returns. So if we increase the sequence
 275          * here the kvm page fault will notice if the spte can't be
 276          * established because the page is going to be freed. If
 277          * instead the kvm page fault establishes the spte before
 278          * ->invalidate_page runs, kvm_unmap_hva will release it
 279          * before returning.
 280          *
 281          * The sequence increase only need to be seen at spin_unlock
 282          * time, and not at spin_lock time.
 283          *
 284          * Increasing the sequence after the spin_unlock would be
 285          * unsafe because the kvm page fault could then establish the
 286          * pte after kvm_unmap_hva returned, without noticing the page
 287          * is going to be freed.
 288          */
 289         idx = srcu_read_lock(&kvm->srcu);
 290         spin_lock(&kvm->mmu_lock);
 291
 292         kvm->mmu_notifier_seq++;
 293         need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
 294         /* we've to flush the tlb before the pages can be freed */
 295         if (need_tlb_flush)
 296                 kvm_flush_remote_tlbs(kvm);
 297
 298         spin_unlock(&kvm->mmu_lock);
 299         srcu_read_unlock(&kvm->srcu, idx);
 300 }
 301
 302 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
 303                                         struct mm_struct *mm,
 304                                         unsigned long address,
 305                                         pte_t pte)
 306 {
 307         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 308         int idx;
 309
 310         idx = srcu_read_lock(&kvm->srcu);
 311         spin_lock(&kvm->mmu_lock);
 312         kvm->mmu_notifier_seq++;
 313         kvm_set_spte_hva(kvm, address, pte);
 314         spin_unlock(&kvm->mmu_lock);
 315         srcu_read_unlock(&kvm->srcu, idx);
 316 }
 317
 318 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 319                                                     struct mm_struct *mm,
 320                                                     unsigned long start,
 321                                                     unsigned long end)
 322 {
 323         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 324         int need_tlb_flush = 0, idx;
 325
 326         idx = srcu_read_lock(&kvm->srcu);
 327         spin_lock(&kvm->mmu_lock);
 328         /*
 329          * The count increase must become visible at unlock time as no
 330          * spte can be established without taking the mmu_lock and
 331          * count is also read inside the mmu_lock critical section.
 332          */
 333         kvm->mmu_notifier_count++;
 334         need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
 335         need_tlb_flush |= kvm->tlbs_dirty;
 336         /* we've to flush the tlb before the pages can be freed */
 337         if (need_tlb_flush)
 338                 kvm_flush_remote_tlbs(kvm);
 339
 340         spin_unlock(&kvm->mmu_lock);
 341         srcu_read_unlock(&kvm->srcu, idx);
 342 }
 343
 344 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
 345                                                   struct mm_struct *mm,
 346                                                   unsigned long start,
 347                                                   unsigned long end)
 348 {
 349         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 350
 351         spin_lock(&kvm->mmu_lock);
 352         /*
 353          * This sequence increase will notify the kvm page fault that
 354          * the page that is going to be mapped in the spte could have
 355          * been freed.
 356          */
 357         kvm->mmu_notifier_seq++;
 358         smp_wmb();
 359         /*
 360          * The above sequence increase must be visible before the
 361          * below count decrease, which is ensured by the smp_wmb above
 362          * in conjunction with the smp_rmb in mmu_notifier_retry().
 363          */
 364         kvm->mmu_notifier_count--;
 365         spin_unlock(&kvm->mmu_lock);
 366
 367         BUG_ON(kvm->mmu_notifier_count < 0);
 368 }
 369
 370 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
 371                                               struct mm_struct *mm,
 372                                               unsigned long address)
 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, address);
 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 void kvm_init_memslots_id(struct kvm *kvm)
 443 {
 444         int i;
 445         struct kvm_memslots *slots = kvm->memslots;
 446
 447         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
 448                 slots->id_to_index[i] = slots->memslots[i].id = i;
 449 }
 450
 451 static struct kvm *kvm_create_vm(unsigned long type)
 452 {
 453         int r, i;
 454         struct kvm *kvm = kvm_arch_alloc_vm();
 455
 456         if (!kvm)
 457                 return ERR_PTR(-ENOMEM);
 458
 459         r = kvm_arch_init_vm(kvm, type);
 460         if (r)
 461                 goto out_err_no_disable;
 462
 463         r = hardware_enable_all();
 464         if (r)
 465                 goto out_err_no_disable;
 466
 467 #ifdef CONFIG_HAVE_KVM_IRQCHIP
 468         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
 469 #endif
 470 #ifdef CONFIG_HAVE_KVM_IRQFD
 471         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
 472 #endif
 473
 474         BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
 475
 476         r = -ENOMEM;
 477         kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 478         if (!kvm->memslots)
 479                 goto out_err_no_srcu;
 480         kvm_init_memslots_id(kvm);
 481         if (init_srcu_struct(&kvm->srcu))
 482                 goto out_err_no_srcu;
 483         if (init_srcu_struct(&kvm->irq_srcu))
 484                 goto out_err_no_irq_srcu;
 485         for (i = 0; i < KVM_NR_BUSES; i++) {
 486                 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
 487                                         GFP_KERNEL);
 488                 if (!kvm->buses[i])
 489                         goto out_err;
 490         }
 491
 492         spin_lock_init(&kvm->mmu_lock);
 493         kvm->mm = current->mm;
 494         atomic_inc(&kvm->mm->mm_count);
 495         kvm_eventfd_init(kvm);
 496         mutex_init(&kvm->lock);
 497         mutex_init(&kvm->irq_lock);
 498         mutex_init(&kvm->slots_lock);
 499         atomic_set(&kvm->users_count, 1);
 500         INIT_LIST_HEAD(&kvm->devices);
 501
 502         r = kvm_init_mmu_notifier(kvm);
 503         if (r)
 504                 goto out_err;
 505
 506         spin_lock(&kvm_lock);
 507         list_add(&kvm->vm_list, &vm_list);
 508         spin_unlock(&kvm_lock);
 509
 510         return kvm;
 511
 512 out_err:
 513         cleanup_srcu_struct(&kvm->irq_srcu);
 514 out_err_no_irq_srcu:
 515         cleanup_srcu_struct(&kvm->srcu);
 516 out_err_no_srcu:
 517         hardware_disable_all();
 518 out_err_no_disable:
 519         for (i = 0; i < KVM_NR_BUSES; i++)
 520                 kfree(kvm->buses[i]);
 521         kfree(kvm->memslots);
 522         kvm_arch_free_vm(kvm);
 523         return ERR_PTR(r);
 524 }
 525
 526 /*
 527  * Avoid using vmalloc for a small buffer.
 528  * Should not be used when the size is statically known.
 529  */
 530 void *kvm_kvzalloc(unsigned long size)
 531 {
 532         if (size > PAGE_SIZE)
 533                 return vzalloc(size);
 534         else
 535                 return kzalloc(size, GFP_KERNEL);
 536 }
 537
 538 void kvm_kvfree(const void *addr)
 539 {
 540         if (is_vmalloc_addr(addr))
 541                 vfree(addr);
 542         else
 543                 kfree(addr);
 544 }
 545
 546 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
 547 {
 548         if (!memslot->dirty_bitmap)
 549                 return;
 550
 551         kvm_kvfree(memslot->dirty_bitmap);
 552         memslot->dirty_bitmap = NULL;
 553 }
 554
 555 /*
 556  * Free any memory in @free but not in @dont.
 557  */
 558 static void kvm_free_physmem_slot(struct kvm *kvm, struct kvm_memory_slot *free,
 559                                   struct kvm_memory_slot *dont)
 560 {
 561         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 562                 kvm_destroy_dirty_bitmap(free);
 563
 564         kvm_arch_free_memslot(kvm, free, dont);
 565
 566         free->npages = 0;
 567 }
 568
 569 static void kvm_free_physmem(struct kvm *kvm)
 570 {
 571         struct kvm_memslots *slots = kvm->memslots;
 572         struct kvm_memory_slot *memslot;
 573
 574         kvm_for_each_memslot(memslot, slots)
 575                 kvm_free_physmem_slot(kvm, memslot, NULL);
 576
 577         kfree(kvm->memslots);
 578 }
 579
 580 static void kvm_destroy_devices(struct kvm *kvm)
 581 {
 582         struct list_head *node, *tmp;
 583
 584         list_for_each_safe(node, tmp, &kvm->devices) {
 585                 struct kvm_device *dev =
 586                         list_entry(node, struct kvm_device, vm_node);
 587
 588                 list_del(node);
 589                 dev->ops->destroy(dev);
 590         }
 591 }
 592
 593 static void kvm_destroy_vm(struct kvm *kvm)
 594 {
 595         int i;
 596         struct mm_struct *mm = kvm->mm;
 597
 598         kvm_arch_sync_events(kvm);
 599         spin_lock(&kvm_lock);
 600         list_del(&kvm->vm_list);
 601         spin_unlock(&kvm_lock);
 602         kvm_free_irq_routing(kvm);
 603         for (i = 0; i < KVM_NR_BUSES; i++)
 604                 kvm_io_bus_destroy(kvm->buses[i]);
 605         kvm_coalesced_mmio_free(kvm);
 606 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 607         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
 608 #else
 609         kvm_arch_flush_shadow_all(kvm);
 610 #endif
 611         kvm_arch_destroy_vm(kvm);
 612         kvm_destroy_devices(kvm);
 613         kvm_free_physmem(kvm);
 614         cleanup_srcu_struct(&kvm->irq_srcu);
 615         cleanup_srcu_struct(&kvm->srcu);
 616         kvm_arch_free_vm(kvm);
 617         hardware_disable_all();
 618         mmdrop(mm);
 619 }
 620
 621 void kvm_get_kvm(struct kvm *kvm)
 622 {
 623         atomic_inc(&kvm->users_count);
 624 }
 625 EXPORT_SYMBOL_GPL(kvm_get_kvm);
 626
 627 void kvm_put_kvm(struct kvm *kvm)
 628 {
 629         if (atomic_dec_and_test(&kvm->users_count))
 630                 kvm_destroy_vm(kvm);
 631 }
 632 EXPORT_SYMBOL_GPL(kvm_put_kvm);
 633
 634
 635 static int kvm_vm_release(struct inode *inode, struct file *filp)
 636 {
 637         struct kvm *kvm = filp->private_data;
 638
 639         kvm_irqfd_release(kvm);
 640
 641         kvm_put_kvm(kvm);
 642         return 0;
 643 }
 644
 645 /*
 646  * Allocation size is twice as large as the actual dirty bitmap size.
 647  * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
 648  */
 649 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
 650 {
 651         unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
 652
 653         memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
 654         if (!memslot->dirty_bitmap)
 655                 return -ENOMEM;
 656
 657         return 0;
 658 }
 659
 660 static int cmp_memslot(const void *slot1, const void *slot2)
 661 {
 662         struct kvm_memory_slot *s1, *s2;
 663
 664         s1 = (struct kvm_memory_slot *)slot1;
 665         s2 = (struct kvm_memory_slot *)slot2;
 666
 667         if (s1->npages < s2->npages)
 668                 return 1;
 669         if (s1->npages > s2->npages)
 670                 return -1;
 671
 672         return 0;
 673 }
 674
 675 /*
 676  * Sort the memslots base on its size, so the larger slots
 677  * will get better fit.
 678  */
 679 static void sort_memslots(struct kvm_memslots *slots)
 680 {
 681         int i;
 682
 683         sort(slots->memslots, KVM_MEM_SLOTS_NUM,
 684               sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
 685
 686         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
 687                 slots->id_to_index[slots->memslots[i].id] = i;
 688 }
 689
 690 static void update_memslots(struct kvm_memslots *slots,
 691                             struct kvm_memory_slot *new,
 692                             u64 last_generation)
 693 {
 694         if (new) {
 695                 int id = new->id;
 696                 struct kvm_memory_slot *old = id_to_memslot(slots, id);
 697                 unsigned long npages = old->npages;
 698
 699                 *old = *new;
 700                 if (new->npages != npages)
 701                         sort_memslots(slots);
 702         }
 703
 704         slots->generation = last_generation + 1;
 705 }
 706
 707 static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
 708 {
 709         u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
 710
 711 #ifdef KVM_CAP_READONLY_MEM
 712         valid_flags |= KVM_MEM_READONLY;
 713 #endif
 714
 715         if (mem->flags & ~valid_flags)
 716                 return -EINVAL;
 717
 718         return 0;
 719 }
 720
 721 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
 722                 struct kvm_memslots *slots, struct kvm_memory_slot *new)
 723 {
 724         struct kvm_memslots *old_memslots = kvm->memslots;
 725
 726         update_memslots(slots, new, kvm->memslots->generation);
 727         rcu_assign_pointer(kvm->memslots, slots);
 728         synchronize_srcu_expedited(&kvm->srcu);
 729
 730         kvm_arch_memslots_updated(kvm);
 731
 732         return old_memslots;
 733 }
 734
 735 /*
 736  * Allocate some memory and give it an address in the guest physical address
 737  * space.
 738  *
 739  * Discontiguous memory is allowed, mostly for framebuffers.
 740  *
 741  * Must be called holding mmap_sem for write.
 742  */
 743 int __kvm_set_memory_region(struct kvm *kvm,
 744                             struct kvm_userspace_memory_region *mem)
 745 {
 746         int r;
 747         gfn_t base_gfn;
 748         unsigned long npages;
 749         struct kvm_memory_slot *slot;
 750         struct kvm_memory_slot old, new;
 751         struct kvm_memslots *slots = NULL, *old_memslots;
 752         enum kvm_mr_change change;
 753
 754         r = check_memory_region_flags(mem);
 755         if (r)
 756                 goto out;
 757
 758         r = -EINVAL;
 759         /* General sanity checks */
 760         if (mem->memory_size & (PAGE_SIZE - 1))
 761                 goto out;
 762         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 763                 goto out;
 764         /* We can read the guest memory with __xxx_user() later on. */
 765         if ((mem->slot < KVM_USER_MEM_SLOTS) &&
 766             ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
 767              !access_ok(VERIFY_WRITE,
 768                         (void __user *)(unsigned long)mem->userspace_addr,
 769                         mem->memory_size)))
 770                 goto out;
 771         if (mem->slot >= KVM_MEM_SLOTS_NUM)
 772                 goto out;
 773         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 774                 goto out;
 775
 776         slot = id_to_memslot(kvm->memslots, mem->slot);
 777         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 778         npages = mem->memory_size >> PAGE_SHIFT;
 779
 780         r = -EINVAL;
 781         if (npages > KVM_MEM_MAX_NR_PAGES)
 782                 goto out;
 783
 784         if (!npages)
 785                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 786
 787         new = old = *slot;
 788
 789         new.id = mem->slot;
 790         new.base_gfn = base_gfn;
 791         new.npages = npages;
 792         new.flags = mem->flags;
 793
 794         r = -EINVAL;
 795         if (npages) {
 796                 if (!old.npages)
 797                         change = KVM_MR_CREATE;
 798                 else { /* Modify an existing slot. */
 799                         if ((mem->userspace_addr != old.userspace_addr) ||
 800                             (npages != old.npages) ||
 801                             ((new.flags ^ old.flags) & KVM_MEM_READONLY))
 802                                 goto out;
 803
 804                         if (base_gfn != old.base_gfn)
 805                                 change = KVM_MR_MOVE;
 806                         else if (new.flags != old.flags)
 807                                 change = KVM_MR_FLAGS_ONLY;
 808                         else { /* Nothing to change. */
 809                                 r = 0;
 810                                 goto out;
 811                         }
 812                 }
 813         } else if (old.npages) {
 814                 change = KVM_MR_DELETE;
 815         } else /* Modify a non-existent slot: disallowed. */
 816                 goto out;
 817
 818         if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
 819                 /* Check for overlaps */
 820                 r = -EEXIST;
 821                 kvm_for_each_memslot(slot, kvm->memslots) {
 822                         if ((slot->id >= KVM_USER_MEM_SLOTS) ||
 823                             (slot->id == mem->slot))
 824                                 continue;
 825                         if (!((base_gfn + npages <= slot->base_gfn) ||
 826                               (base_gfn >= slot->base_gfn + slot->npages)))
 827                                 goto out;
 828                 }
 829         }
 830
 831         /* Free page dirty bitmap if unneeded */
 832         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 833                 new.dirty_bitmap = NULL;
 834
 835         r = -ENOMEM;
 836         if (change == KVM_MR_CREATE) {
 837                 new.userspace_addr = mem->userspace_addr;
 838
 839                 if (kvm_arch_create_memslot(kvm, &new, npages))
 840                         goto out_free;
 841         }
 842
 843         /* Allocate page dirty bitmap if needed */
 844         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 845                 if (kvm_create_dirty_bitmap(&new) < 0)
 846                         goto out_free;
 847         }
 848
 849         if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
 850                 r = -ENOMEM;
 851                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 852                                 GFP_KERNEL);
 853                 if (!slots)
 854                         goto out_free;
 855                 slot = id_to_memslot(slots, mem->slot);
 856                 slot->flags |= KVM_MEMSLOT_INVALID;
 857
 858                 old_memslots = install_new_memslots(kvm, slots, NULL);
 859
 860                 /* slot was deleted or moved, clear iommu mapping */
 861                 kvm_iommu_unmap_pages(kvm, &old);
 862                 /* From this point no new shadow pages pointing to a deleted,
 863                  * or moved, memslot will be created.
 864                  *
 865                  * validation of sp->gfn happens in:
 866                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
 867                  *      - kvm_is_visible_gfn (mmu_check_roots)
 868                  */
 869                 kvm_arch_flush_shadow_memslot(kvm, slot);
 870                 slots = old_memslots;
 871         }
 872
 873         r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
 874         if (r)
 875                 goto out_slots;
 876
 877         r = -ENOMEM;
 878         /*
 879          * We can re-use the old_memslots from above, the only difference
 880          * from the currently installed memslots is the invalid flag.  This
 881          * will get overwritten by update_memslots anyway.
 882          */
 883         if (!slots) {
 884                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 885                                 GFP_KERNEL);
 886                 if (!slots)
 887                         goto out_free;
 888         }
 889
 890         /* actual memory is freed via old in kvm_free_physmem_slot below */
 891         if (change == KVM_MR_DELETE) {
 892                 new.dirty_bitmap = NULL;
 893                 memset(&new.arch, 0, sizeof(new.arch));
 894         }
 895
 896         old_memslots = install_new_memslots(kvm, slots, &new);
 897
 898         kvm_arch_commit_memory_region(kvm, mem, &old, change);
 899
 900         kvm_free_physmem_slot(kvm, &old, &new);
 901         kfree(old_memslots);
 902
 903         /*
 904          * IOMMU mapping:  New slots need to be mapped.  Old slots need to be
 905          * un-mapped and re-mapped if their base changes.  Since base change
 906          * unmapping is handled above with slot deletion, mapping alone is
 907          * needed here.  Anything else the iommu might care about for existing
 908          * slots (size changes, userspace addr changes and read-only flag
 909          * changes) is disallowed above, so any other attribute changes getting
 910          * here can be skipped.
 911          */
 912         if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
 913                 r = kvm_iommu_map_pages(kvm, &new);
 914                 return r;
 915         }
 916
 917         return 0;
 918
 919 out_slots:
 920         kfree(slots);
 921 out_free:
 922         kvm_free_physmem_slot(kvm, &new, &old);
 923 out:
 924         return r;
 925 }
 926 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 927
 928 int kvm_set_memory_region(struct kvm *kvm,
 929                           struct kvm_userspace_memory_region *mem)
 930 {
 931         int r;
 932
 933         mutex_lock(&kvm->slots_lock);
 934         r = __kvm_set_memory_region(kvm, mem);
 935         mutex_unlock(&kvm->slots_lock);
 936         return r;
 937 }
 938 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 939
 940 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 941                                           struct kvm_userspace_memory_region *mem)
 942 {
 943         if (mem->slot >= KVM_USER_MEM_SLOTS)
 944                 return -EINVAL;
 945         return kvm_set_memory_region(kvm, mem);
 946 }
 947
 948 int kvm_get_dirty_log(struct kvm *kvm,
 949                         struct kvm_dirty_log *log, int *is_dirty)
 950 {
 951         struct kvm_memory_slot *memslot;
 952         int r, i;
 953         unsigned long n;
 954         unsigned long any = 0;
 955
 956         r = -EINVAL;
 957         if (log->slot >= KVM_USER_MEM_SLOTS)
 958                 goto out;
 959
 960         memslot = id_to_memslot(kvm->memslots, log->slot);
 961         r = -ENOENT;
 962         if (!memslot->dirty_bitmap)
 963                 goto out;
 964
 965         n = kvm_dirty_bitmap_bytes(memslot);
 966
 967         for (i = 0; !any && i < n/sizeof(long); ++i)
 968                 any = memslot->dirty_bitmap[i];
 969
 970         r = -EFAULT;
 971         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 972                 goto out;
 973
 974         if (any)
 975                 *is_dirty = 1;
 976
 977         r = 0;
 978 out:
 979         return r;
 980 }
 981 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
 982
 983 bool kvm_largepages_enabled(void)
 984 {
 985         return largepages_enabled;
 986 }
 987
 988 void kvm_disable_largepages(void)
 989 {
 990         largepages_enabled = false;
 991 }
 992 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
 993
 994 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 995 {
 996         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
 997 }
 998 EXPORT_SYMBOL_GPL(gfn_to_memslot);
 999
1000 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1001 {
1002         struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1003
1004         if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1005               memslot->flags & KVM_MEMSLOT_INVALID)
1006                 return 0;
1007
1008         return 1;
1009 }
1010 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1011
1012 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1013 {
1014         struct vm_area_struct *vma;
1015         unsigned long addr, size;
1016
1017         size = PAGE_SIZE;
1018
1019         addr = gfn_to_hva(kvm, gfn);
1020         if (kvm_is_error_hva(addr))
1021                 return PAGE_SIZE;
1022
1023         down_read(&current->mm->mmap_sem);
1024         vma = find_vma(current->mm, addr);
1025         if (!vma)
1026                 goto out;
1027
1028         size = vma_kernel_pagesize(vma);
1029
1030 out:
1031         up_read(&current->mm->mmap_sem);
1032
1033         return size;
1034 }
1035
1036 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1037 {
1038         return slot->flags & KVM_MEM_READONLY;
1039 }
1040
1041 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1042                                        gfn_t *nr_pages, bool write)
1043 {
1044         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1045                 return KVM_HVA_ERR_BAD;
1046
1047         if (memslot_is_readonly(slot) && write)
1048                 return KVM_HVA_ERR_RO_BAD;
1049
1050         if (nr_pages)
1051                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1052
1053         return __gfn_to_hva_memslot(slot, gfn);
1054 }
1055
1056 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1057                                      gfn_t *nr_pages)
1058 {
1059         return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1060 }
1061
1062 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1063                                         gfn_t gfn)
1064 {
1065         return gfn_to_hva_many(slot, gfn, NULL);
1066 }
1067 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1068
1069 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1070 {
1071         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1072 }
1073 EXPORT_SYMBOL_GPL(gfn_to_hva);
1074
1075 /*
1076  * If writable is set to false, the hva returned by this function is only
1077  * allowed to be read.
1078  */
1079 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1080 {
1081         struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1082         unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1083
1084         if (!kvm_is_error_hva(hva) && writable)
1085                 *writable = !memslot_is_readonly(slot);
1086
1087         return hva;
1088 }
1089
1090 static int kvm_read_hva(void *data, void __user *hva, int len)
1091 {
1092         return __copy_from_user(data, hva, len);
1093 }
1094
1095 static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1096 {
1097         return __copy_from_user_inatomic(data, hva, len);
1098 }
1099
1100 static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1101         unsigned long start, int write, struct page **page)
1102 {
1103         int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1104
1105         if (write)
1106                 flags |= FOLL_WRITE;
1107
1108         return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1109 }
1110
1111 static inline int check_user_page_hwpoison(unsigned long addr)
1112 {
1113         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1114
1115         rc = __get_user_pages(current, current->mm, addr, 1,
1116                               flags, NULL, NULL, NULL);
1117         return rc == -EHWPOISON;
1118 }
1119
1120 /*
1121  * The atomic path to get the writable pfn which will be stored in @pfn,
1122  * true indicates success, otherwise false is returned.
1123  */
1124 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1125                             bool write_fault, bool *writable, pfn_t *pfn)
1126 {
1127         struct page *page[1];
1128         int npages;
1129
1130         if (!(async || atomic))
1131                 return false;
1132
1133         /*
1134          * Fast pin a writable pfn only if it is a write fault request
1135          * or the caller allows to map a writable pfn for a read fault
1136          * request.
1137          */
1138         if (!(write_fault || writable))
1139                 return false;
1140
1141         npages = __get_user_pages_fast(addr, 1, 1, page);
1142         if (npages == 1) {
1143                 *pfn = page_to_pfn(page[0]);
1144
1145                 if (writable)
1146                         *writable = true;
1147                 return true;
1148         }
1149
1150         return false;
1151 }
1152
1153 /*
1154  * The slow path to get the pfn of the specified host virtual address,
1155  * 1 indicates success, -errno is returned if error is detected.
1156  */
1157 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1158                            bool *writable, pfn_t *pfn)
1159 {
1160         struct page *page[1];
1161         int npages = 0;
1162
1163         might_sleep();
1164
1165         if (writable)
1166                 *writable = write_fault;
1167
1168         if (async) {
1169                 down_read(&current->mm->mmap_sem);
1170                 npages = get_user_page_nowait(current, current->mm,
1171                                               addr, write_fault, page);
1172                 up_read(&current->mm->mmap_sem);
1173         } else
1174                 npages = get_user_pages_fast(addr, 1, write_fault,
1175                                              page);
1176         if (npages != 1)
1177                 return npages;
1178
1179         /* map read fault as writable if possible */
1180         if (unlikely(!write_fault) && writable) {
1181                 struct page *wpage[1];
1182
1183                 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1184                 if (npages == 1) {
1185                         *writable = true;
1186                         put_page(page[0]);
1187                         page[0] = wpage[0];
1188                 }
1189
1190                 npages = 1;
1191         }
1192         *pfn = page_to_pfn(page[0]);
1193         return npages;
1194 }
1195
1196 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1197 {
1198         if (unlikely(!(vma->vm_flags & VM_READ)))
1199                 return false;
1200
1201         if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1202                 return false;
1203
1204         return true;
1205 }
1206
1207 /*
1208  * Pin guest page in memory and return its pfn.
1209  * @addr: host virtual address which maps memory to the guest
1210  * @atomic: whether this function can sleep
1211  * @async: whether this function need to wait IO complete if the
1212  *         host page is not in the memory
1213  * @write_fault: whether we should get a writable host page
1214  * @writable: whether it allows to map a writable host page for !@write_fault
1215  *
1216  * The function will map a writable host page for these two cases:
1217  * 1): @write_fault = true
1218  * 2): @write_fault = false && @writable, @writable will tell the caller
1219  *     whether the mapping is writable.
1220  */
1221 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1222                         bool write_fault, bool *writable)
1223 {
1224         struct vm_area_struct *vma;
1225         pfn_t pfn = 0;
1226         int npages;
1227
1228         /* we can do it either atomically or asynchronously, not both */
1229         BUG_ON(atomic && async);
1230
1231         if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1232                 return pfn;
1233
1234         if (atomic)
1235                 return KVM_PFN_ERR_FAULT;
1236
1237         npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1238         if (npages == 1)
1239                 return pfn;
1240
1241         down_read(&current->mm->mmap_sem);
1242         if (npages == -EHWPOISON ||
1243               (!async && check_user_page_hwpoison(addr))) {
1244                 pfn = KVM_PFN_ERR_HWPOISON;
1245                 goto exit;
1246         }
1247
1248         vma = find_vma_intersection(current->mm, addr, addr + 1);
1249
1250         if (vma == NULL)
1251                 pfn = KVM_PFN_ERR_FAULT;
1252         else if ((vma->vm_flags & VM_PFNMAP)) {
1253                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1254                         vma->vm_pgoff;
1255                 BUG_ON(!kvm_is_mmio_pfn(pfn));
1256         } else {
1257                 if (async && vma_is_valid(vma, write_fault))
1258                         *async = true;
1259                 pfn = KVM_PFN_ERR_FAULT;
1260         }
1261 exit:
1262         up_read(&current->mm->mmap_sem);
1263         return pfn;
1264 }
1265
1266 static pfn_t
1267 __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1268                      bool *async, bool write_fault, bool *writable)
1269 {
1270         unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1271
1272         if (addr == KVM_HVA_ERR_RO_BAD)
1273                 return KVM_PFN_ERR_RO_FAULT;
1274
1275         if (kvm_is_error_hva(addr))
1276                 return KVM_PFN_NOSLOT;
1277
1278         /* Do not map writable pfn in the readonly memslot. */
1279         if (writable && memslot_is_readonly(slot)) {
1280                 *writable = false;
1281                 writable = NULL;
1282         }
1283
1284         return hva_to_pfn(addr, atomic, async, write_fault,
1285                           writable);
1286 }
1287
1288 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1289                           bool write_fault, bool *writable)
1290 {
1291         struct kvm_memory_slot *slot;
1292
1293         if (async)
1294                 *async = false;
1295
1296         slot = gfn_to_memslot(kvm, gfn);
1297
1298         return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1299                                     writable);
1300 }
1301
1302 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1303 {
1304         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1305 }
1306 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1307
1308 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1309                        bool write_fault, bool *writable)
1310 {
1311         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1312 }
1313 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1314
1315 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1316 {
1317         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1318 }
1319 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1320
1321 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1322                       bool *writable)
1323 {
1324         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1325 }
1326 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1327
1328 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1329 {
1330         return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1331 }
1332
1333 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1334 {
1335         return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1336 }
1337 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1338
1339 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1340                                                                   int nr_pages)
1341 {
1342         unsigned long addr;
1343         gfn_t entry;
1344
1345         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1346         if (kvm_is_error_hva(addr))
1347                 return -1;
1348
1349         if (entry < nr_pages)
1350                 return 0;
1351
1352         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1353 }
1354 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1355
1356 static struct page *kvm_pfn_to_page(pfn_t pfn)
1357 {
1358         if (is_error_noslot_pfn(pfn))
1359                 return KVM_ERR_PTR_BAD_PAGE;
1360
1361         if (kvm_is_mmio_pfn(pfn)) {
1362                 WARN_ON(1);
1363                 return KVM_ERR_PTR_BAD_PAGE;
1364         }
1365
1366         return pfn_to_page(pfn);
1367 }
1368
1369 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1370 {
1371         pfn_t pfn;
1372
1373         pfn = gfn_to_pfn(kvm, gfn);
1374
1375         return kvm_pfn_to_page(pfn);
1376 }
1377
1378 EXPORT_SYMBOL_GPL(gfn_to_page);
1379
1380 void kvm_release_page_clean(struct page *page)
1381 {
1382         WARN_ON(is_error_page(page));
1383
1384         kvm_release_pfn_clean(page_to_pfn(page));
1385 }
1386 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1387
1388 void kvm_release_pfn_clean(pfn_t pfn)
1389 {
1390         if (!is_error_noslot_pfn(pfn) && !kvm_is_mmio_pfn(pfn))
1391                 put_page(pfn_to_page(pfn));
1392 }
1393 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1394
1395 void kvm_release_page_dirty(struct page *page)
1396 {
1397         WARN_ON(is_error_page(page));
1398
1399         kvm_release_pfn_dirty(page_to_pfn(page));
1400 }
1401 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1402
1403 static void kvm_release_pfn_dirty(pfn_t pfn)
1404 {
1405         kvm_set_pfn_dirty(pfn);
1406         kvm_release_pfn_clean(pfn);
1407 }
1408
1409 void kvm_set_pfn_dirty(pfn_t pfn)
1410 {
1411         if (!kvm_is_mmio_pfn(pfn)) {
1412                 struct page *page = pfn_to_page(pfn);
1413                 if (!PageReserved(page))
1414                         SetPageDirty(page);
1415         }
1416 }
1417 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1418
1419 void kvm_set_pfn_accessed(pfn_t pfn)
1420 {
1421         if (!kvm_is_mmio_pfn(pfn))
1422                 mark_page_accessed(pfn_to_page(pfn));
1423 }
1424 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1425
1426 void kvm_get_pfn(pfn_t pfn)
1427 {
1428         if (!kvm_is_mmio_pfn(pfn))
1429                 get_page(pfn_to_page(pfn));
1430 }
1431 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1432
1433 static int next_segment(unsigned long len, int offset)
1434 {
1435         if (len > PAGE_SIZE - offset)
1436                 return PAGE_SIZE - offset;
1437         else
1438                 return len;
1439 }
1440
1441 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1442                         int len)
1443 {
1444         int r;
1445         unsigned long addr;
1446
1447         addr = gfn_to_hva_prot(kvm, gfn, NULL);
1448         if (kvm_is_error_hva(addr))
1449                 return -EFAULT;
1450         r = kvm_read_hva(data, (void __user *)addr + offset, len);
1451         if (r)
1452                 return -EFAULT;
1453         return 0;
1454 }
1455 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1456
1457 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1458 {
1459         gfn_t gfn = gpa >> PAGE_SHIFT;
1460         int seg;
1461         int offset = offset_in_page(gpa);
1462         int ret;
1463
1464         while ((seg = next_segment(len, offset)) != 0) {
1465                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1466                 if (ret < 0)
1467                         return ret;
1468                 offset = 0;
1469                 len -= seg;
1470                 data += seg;
1471                 ++gfn;
1472         }
1473         return 0;
1474 }
1475 EXPORT_SYMBOL_GPL(kvm_read_guest);
1476
1477 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1478                           unsigned long len)
1479 {
1480         int r;
1481         unsigned long addr;
1482         gfn_t gfn = gpa >> PAGE_SHIFT;
1483         int offset = offset_in_page(gpa);
1484
1485         addr = gfn_to_hva_prot(kvm, gfn, NULL);
1486         if (kvm_is_error_hva(addr))
1487                 return -EFAULT;
1488         pagefault_disable();
1489         r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
1490         pagefault_enable();
1491         if (r)
1492                 return -EFAULT;
1493         return 0;
1494 }
1495 EXPORT_SYMBOL(kvm_read_guest_atomic);
1496
1497 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1498                          int offset, int len)
1499 {
1500         int r;
1501         unsigned long addr;
1502
1503         addr = gfn_to_hva(kvm, gfn);
1504         if (kvm_is_error_hva(addr))
1505                 return -EFAULT;
1506         r = __copy_to_user((void __user *)addr + offset, data, len);
1507         if (r)
1508                 return -EFAULT;
1509         mark_page_dirty(kvm, gfn);
1510         return 0;
1511 }
1512 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1513
1514 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1515                     unsigned long len)
1516 {
1517         gfn_t gfn = gpa >> PAGE_SHIFT;
1518         int seg;
1519         int offset = offset_in_page(gpa);
1520         int ret;
1521
1522         while ((seg = next_segment(len, offset)) != 0) {
1523                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1524                 if (ret < 0)
1525                         return ret;
1526                 offset = 0;
1527                 len -= seg;
1528                 data += seg;
1529                 ++gfn;
1530         }
1531         return 0;
1532 }
1533
1534 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1535                               gpa_t gpa, unsigned long len)
1536 {
1537         struct kvm_memslots *slots = kvm_memslots(kvm);
1538         int offset = offset_in_page(gpa);
1539         gfn_t start_gfn = gpa >> PAGE_SHIFT;
1540         gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1541         gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1542         gfn_t nr_pages_avail;
1543
1544         ghc->gpa = gpa;
1545         ghc->generation = slots->generation;
1546         ghc->len = len;
1547         ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1548         ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, &nr_pages_avail);
1549         if (!kvm_is_error_hva(ghc->hva) && nr_pages_avail >= nr_pages_needed) {
1550                 ghc->hva += offset;
1551         } else {
1552                 /*
1553                  * If the requested region crosses two memslots, we still
1554                  * verify that the entire region is valid here.
1555                  */
1556                 while (start_gfn <= end_gfn) {
1557                         ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1558                         ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1559                                                    &nr_pages_avail);
1560                         if (kvm_is_error_hva(ghc->hva))
1561                                 return -EFAULT;
1562                         start_gfn += nr_pages_avail;
1563                 }
1564                 /* Use the slow path for cross page reads and writes. */
1565                 ghc->memslot = NULL;
1566         }
1567         return 0;
1568 }
1569 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1570
1571 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1572                            void *data, unsigned long len)
1573 {
1574         struct kvm_memslots *slots = kvm_memslots(kvm);
1575         int r;
1576
1577         BUG_ON(len > ghc->len);
1578
1579         if (slots->generation != ghc->generation)
1580                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1581
1582         if (unlikely(!ghc->memslot))
1583                 return kvm_write_guest(kvm, ghc->gpa, data, len);
1584
1585         if (kvm_is_error_hva(ghc->hva))
1586                 return -EFAULT;
1587
1588         r = __copy_to_user((void __user *)ghc->hva, data, len);
1589         if (r)
1590                 return -EFAULT;
1591         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1592
1593         return 0;
1594 }
1595 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1596
1597 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1598                            void *data, unsigned long len)
1599 {
1600         struct kvm_memslots *slots = kvm_memslots(kvm);
1601         int r;
1602
1603         BUG_ON(len > ghc->len);
1604
1605         if (slots->generation != ghc->generation)
1606                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1607
1608         if (unlikely(!ghc->memslot))
1609                 return kvm_read_guest(kvm, ghc->gpa, data, len);
1610
1611         if (kvm_is_error_hva(ghc->hva))
1612                 return -EFAULT;
1613
1614         r = __copy_from_user(data, (void __user *)ghc->hva, len);
1615         if (r)
1616                 return -EFAULT;
1617
1618         return 0;
1619 }
1620 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1621
1622 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1623 {
1624         const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
1625
1626         return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
1627 }
1628 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1629
1630 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1631 {
1632         gfn_t gfn = gpa >> PAGE_SHIFT;
1633         int seg;
1634         int offset = offset_in_page(gpa);
1635         int ret;
1636
1637         while ((seg = next_segment(len, offset)) != 0) {
1638                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1639                 if (ret < 0)
1640                         return ret;
1641                 offset = 0;
1642                 len -= seg;
1643                 ++gfn;
1644         }
1645         return 0;
1646 }
1647 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1648
1649 static void mark_page_dirty_in_slot(struct kvm *kvm,
1650                                     struct kvm_memory_slot *memslot,
1651                                     gfn_t gfn)
1652 {
1653         if (memslot && memslot->dirty_bitmap) {
1654                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1655
1656                 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1657         }
1658 }
1659
1660 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1661 {
1662         struct kvm_memory_slot *memslot;
1663
1664         memslot = gfn_to_memslot(kvm, gfn);
1665         mark_page_dirty_in_slot(kvm, memslot, gfn);
1666 }
1667 EXPORT_SYMBOL_GPL(mark_page_dirty);
1668
1669 /*
1670  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1671  */
1672 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1673 {
1674         DEFINE_WAIT(wait);
1675
1676         for (;;) {
1677                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1678
1679                 if (kvm_arch_vcpu_runnable(vcpu)) {
1680                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1681                         break;
1682                 }
1683                 if (kvm_cpu_has_pending_timer(vcpu))
1684                         break;
1685                 if (signal_pending(current))
1686                         break;
1687
1688                 schedule();
1689         }
1690
1691         finish_wait(&vcpu->wq, &wait);
1692 }
1693 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
1694
1695 #ifndef CONFIG_S390
1696 /*
1697  * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1698  */
1699 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1700 {
1701         int me;
1702         int cpu = vcpu->cpu;
1703         wait_queue_head_t *wqp;
1704
1705         wqp = kvm_arch_vcpu_wq(vcpu);
1706         if (waitqueue_active(wqp)) {
1707                 wake_up_interruptible(wqp);
1708                 ++vcpu->stat.halt_wakeup;
1709         }
1710
1711         me = get_cpu();
1712         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1713                 if (kvm_arch_vcpu_should_kick(vcpu))
1714                         smp_send_reschedule(cpu);
1715         put_cpu();
1716 }
1717 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
1718 #endif /* !CONFIG_S390 */
1719
1720 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
1721 {
1722         struct pid *pid;
1723         struct task_struct *task = NULL;
1724         int ret = 0;
1725
1726         rcu_read_lock();
1727         pid = rcu_dereference(target->pid);
1728         if (pid)
1729                 task = get_pid_task(target->pid, PIDTYPE_PID);
1730         rcu_read_unlock();
1731         if (!task)
1732                 return ret;
1733         if (task->flags & PF_VCPU) {
1734                 put_task_struct(task);
1735                 return ret;
1736         }
1737         ret = yield_to(task, 1);
1738         put_task_struct(task);
1739
1740         return ret;
1741 }
1742 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1743
1744 /*
1745  * Helper that checks whether a VCPU is eligible for directed yield.
1746  * Most eligible candidate to yield is decided by following heuristics:
1747  *
1748  *  (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1749  *  (preempted lock holder), indicated by @in_spin_loop.
1750  *  Set at the beiginning and cleared at the end of interception/PLE handler.
1751  *
1752  *  (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1753  *  chance last time (mostly it has become eligible now since we have probably
1754  *  yielded to lockholder in last iteration. This is done by toggling
1755  *  @dy_eligible each time a VCPU checked for eligibility.)
1756  *
1757  *  Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1758  *  to preempted lock-holder could result in wrong VCPU selection and CPU
1759  *  burning. Giving priority for a potential lock-holder increases lock
1760  *  progress.
1761  *
1762  *  Since algorithm is based on heuristics, accessing another VCPU data without
1763  *  locking does not harm. It may result in trying to yield to  same VCPU, fail
1764  *  and continue with next VCPU and so on.
1765  */
1766 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1767 {
1768 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1769         bool eligible;
1770
1771         eligible = !vcpu->spin_loop.in_spin_loop ||
1772                         (vcpu->spin_loop.in_spin_loop &&
1773                          vcpu->spin_loop.dy_eligible);
1774
1775         if (vcpu->spin_loop.in_spin_loop)
1776                 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1777
1778         return eligible;
1779 #else
1780         return true;
1781 #endif
1782 }
1783
1784 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1785 {
1786         struct kvm *kvm = me->kvm;
1787         struct kvm_vcpu *vcpu;
1788         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1789         int yielded = 0;
1790         int try = 3;
1791         int pass;
1792         int i;
1793
1794         kvm_vcpu_set_in_spin_loop(me, true);
1795         /*
1796          * We boost the priority of a VCPU that is runnable but not
1797          * currently running, because it got preempted by something
1798          * else and called schedule in __vcpu_run.  Hopefully that
1799          * VCPU is holding the lock that we need and will release it.
1800          * We approximate round-robin by starting at the last boosted VCPU.
1801          */
1802         for (pass = 0; pass < 2 && !yielded && try; pass++) {
1803                 kvm_for_each_vcpu(i, vcpu, kvm) {
1804                         if (!pass && i <= last_boosted_vcpu) {
1805                                 i = last_boosted_vcpu;
1806                                 continue;
1807                         } else if (pass && i > last_boosted_vcpu)
1808                                 break;
1809                         if (!ACCESS_ONCE(vcpu->preempted))
1810                                 continue;
1811                         if (vcpu == me)
1812                                 continue;
1813                         if (waitqueue_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
1814                                 continue;
1815                         if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1816                                 continue;
1817
1818                         yielded = kvm_vcpu_yield_to(vcpu);
1819                         if (yielded > 0) {
1820                                 kvm->last_boosted_vcpu = i;
1821                                 break;
1822                         } else if (yielded < 0) {
1823                                 try--;
1824                                 if (!try)
1825                                         break;
1826                         }
1827                 }
1828         }
1829         kvm_vcpu_set_in_spin_loop(me, false);
1830
1831         /* Ensure vcpu is not eligible during next spinloop */
1832         kvm_vcpu_set_dy_eligible(me, false);
1833 }
1834 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1835
1836 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1837 {
1838         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1839         struct page *page;
1840
1841         if (vmf->pgoff == 0)
1842                 page = virt_to_page(vcpu->run);
1843 #ifdef CONFIG_X86
1844         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1845                 page = virt_to_page(vcpu->arch.pio_data);
1846 #endif
1847 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1848         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1849                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1850 #endif
1851         else
1852                 return kvm_arch_vcpu_fault(vcpu, vmf);
1853         get_page(page);
1854         vmf->page = page;
1855         return 0;
1856 }
1857
1858 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1859         .fault = kvm_vcpu_fault,
1860 };
1861
1862 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1863 {
1864         vma->vm_ops = &kvm_vcpu_vm_ops;
1865         return 0;
1866 }
1867
1868 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1869 {
1870         struct kvm_vcpu *vcpu = filp->private_data;
1871
1872         kvm_put_kvm(vcpu->kvm);
1873         return 0;
1874 }
1875
1876 static struct file_operations kvm_vcpu_fops = {
1877         .release        = kvm_vcpu_release,
1878         .unlocked_ioctl = kvm_vcpu_ioctl,
1879 #ifdef CONFIG_COMPAT
1880         .compat_ioctl   = kvm_vcpu_compat_ioctl,
1881 #endif
1882         .mmap           = kvm_vcpu_mmap,
1883         .llseek         = noop_llseek,
1884 };
1885
1886 /*
1887  * Allocates an inode for the vcpu.
1888  */
1889 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1890 {
1891         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
1892 }
1893
1894 /*
1895  * Creates some virtual cpus.  Good luck creating more than one.
1896  */
1897 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1898 {
1899         int r;
1900         struct kvm_vcpu *vcpu, *v;
1901
1902         if (id >= KVM_MAX_VCPUS)
1903                 return -EINVAL;
1904
1905         vcpu = kvm_arch_vcpu_create(kvm, id);
1906         if (IS_ERR(vcpu))
1907                 return PTR_ERR(vcpu);
1908
1909         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1910
1911         r = kvm_arch_vcpu_setup(vcpu);
1912         if (r)
1913                 goto vcpu_destroy;
1914
1915         mutex_lock(&kvm->lock);
1916         if (!kvm_vcpu_compatible(vcpu)) {
1917                 r = -EINVAL;
1918                 goto unlock_vcpu_destroy;
1919         }
1920         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1921                 r = -EINVAL;
1922                 goto unlock_vcpu_destroy;
1923         }
1924
1925         kvm_for_each_vcpu(r, v, kvm)
1926                 if (v->vcpu_id == id) {
1927                         r = -EEXIST;
1928                         goto unlock_vcpu_destroy;
1929                 }
1930
1931         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1932
1933         /* Now it's all set up, let userspace reach it */
1934         kvm_get_kvm(kvm);
1935         r = create_vcpu_fd(vcpu);
1936         if (r < 0) {
1937                 kvm_put_kvm(kvm);
1938                 goto unlock_vcpu_destroy;
1939         }
1940
1941         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1942         smp_wmb();
1943         atomic_inc(&kvm->online_vcpus);
1944
1945         mutex_unlock(&kvm->lock);
1946         kvm_arch_vcpu_postcreate(vcpu);
1947         return r;
1948
1949 unlock_vcpu_destroy:
1950         mutex_unlock(&kvm->lock);
1951 vcpu_destroy:
1952         kvm_arch_vcpu_destroy(vcpu);
1953         return r;
1954 }
1955
1956 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1957 {
1958         if (sigset) {
1959                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1960                 vcpu->sigset_active = 1;
1961                 vcpu->sigset = *sigset;
1962         } else
1963                 vcpu->sigset_active = 0;
1964         return 0;
1965 }
1966
1967 static long kvm_vcpu_ioctl(struct file *filp,
1968                            unsigned int ioctl, unsigned long arg)
1969 {
1970         struct kvm_vcpu *vcpu = filp->private_data;
1971         void __user *argp = (void __user *)arg;
1972         int r;
1973         struct kvm_fpu *fpu = NULL;
1974         struct kvm_sregs *kvm_sregs = NULL;
1975
1976         if (vcpu->kvm->mm != current->mm)
1977                 return -EIO;
1978
1979 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
1980         /*
1981          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1982          * so vcpu_load() would break it.
1983          */
1984         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1985                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1986 #endif
1987
1988
1989         r = vcpu_load(vcpu);
1990         if (r)
1991                 return r;
1992         switch (ioctl) {
1993         case KVM_RUN:
1994                 r = -EINVAL;
1995                 if (arg)
1996                         goto out;
1997                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1998                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1999                 break;
2000         case KVM_GET_REGS: {
2001                 struct kvm_regs *kvm_regs;
2002
2003                 r = -ENOMEM;
2004                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2005                 if (!kvm_regs)
2006                         goto out;
2007                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2008                 if (r)
2009                         goto out_free1;
2010                 r = -EFAULT;
2011                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2012                         goto out_free1;
2013                 r = 0;
2014 out_free1:
2015                 kfree(kvm_regs);
2016                 break;
2017         }
2018         case KVM_SET_REGS: {
2019                 struct kvm_regs *kvm_regs;
2020
2021                 r = -ENOMEM;
2022                 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2023                 if (IS_ERR(kvm_regs)) {
2024                         r = PTR_ERR(kvm_regs);
2025                         goto out;
2026                 }
2027                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2028                 kfree(kvm_regs);
2029                 break;
2030         }
2031         case KVM_GET_SREGS: {
2032                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2033                 r = -ENOMEM;
2034                 if (!kvm_sregs)
2035                         goto out;
2036                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2037                 if (r)
2038                         goto out;
2039                 r = -EFAULT;
2040                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2041                         goto out;
2042                 r = 0;
2043                 break;
2044         }
2045         case KVM_SET_SREGS: {
2046                 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2047                 if (IS_ERR(kvm_sregs)) {
2048                         r = PTR_ERR(kvm_sregs);
2049                         kvm_sregs = NULL;
2050                         goto out;
2051                 }
2052                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2053                 break;
2054         }
2055         case KVM_GET_MP_STATE: {
2056                 struct kvm_mp_state mp_state;
2057
2058                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2059                 if (r)
2060                         goto out;
2061                 r = -EFAULT;
2062                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
2063                         goto out;
2064                 r = 0;
2065                 break;
2066         }
2067         case KVM_SET_MP_STATE: {
2068                 struct kvm_mp_state mp_state;
2069
2070                 r = -EFAULT;
2071                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
2072                         goto out;
2073                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2074                 break;
2075         }
2076         case KVM_TRANSLATE: {
2077                 struct kvm_translation tr;
2078
2079                 r = -EFAULT;
2080                 if (copy_from_user(&tr, argp, sizeof tr))
2081                         goto out;
2082                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2083                 if (r)
2084                         goto out;
2085                 r = -EFAULT;
2086                 if (copy_to_user(argp, &tr, sizeof tr))
2087                         goto out;
2088                 r = 0;
2089                 break;
2090         }
2091         case KVM_SET_GUEST_DEBUG: {
2092                 struct kvm_guest_debug dbg;
2093
2094                 r = -EFAULT;
2095                 if (copy_from_user(&dbg, argp, sizeof dbg))
2096                         goto out;
2097                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2098                 break;
2099         }
2100         case KVM_SET_SIGNAL_MASK: {
2101                 struct kvm_signal_mask __user *sigmask_arg = argp;
2102                 struct kvm_signal_mask kvm_sigmask;
2103                 sigset_t sigset, *p;
2104
2105                 p = NULL;
2106                 if (argp) {
2107                         r = -EFAULT;
2108                         if (copy_from_user(&kvm_sigmask, argp,
2109                                            sizeof kvm_sigmask))
2110                                 goto out;
2111                         r = -EINVAL;
2112                         if (kvm_sigmask.len != sizeof sigset)
2113                                 goto out;
2114                         r = -EFAULT;
2115                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2116                                            sizeof sigset))
2117                                 goto out;
2118                         p = &sigset;
2119                 }
2120                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2121                 break;
2122         }
2123         case KVM_GET_FPU: {
2124                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2125                 r = -ENOMEM;
2126                 if (!fpu)
2127                         goto out;
2128                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2129                 if (r)
2130                         goto out;
2131                 r = -EFAULT;
2132                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2133                         goto out;
2134                 r = 0;
2135                 break;
2136         }
2137         case KVM_SET_FPU: {
2138                 fpu = memdup_user(argp, sizeof(*fpu));
2139                 if (IS_ERR(fpu)) {
2140                         r = PTR_ERR(fpu);
2141                         fpu = NULL;
2142                         goto out;
2143                 }
2144                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2145                 break;
2146         }
2147         default:
2148                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2149         }
2150 out:
2151         vcpu_put(vcpu);
2152         kfree(fpu);
2153         kfree(kvm_sregs);
2154         return r;
2155 }
2156
2157 #ifdef CONFIG_COMPAT
2158 static long kvm_vcpu_compat_ioctl(struct file *filp,
2159                                   unsigned int ioctl, unsigned long arg)
2160 {
2161         struct kvm_vcpu *vcpu = filp->private_data;
2162         void __user *argp = compat_ptr(arg);
2163         int r;
2164
2165         if (vcpu->kvm->mm != current->mm)
2166                 return -EIO;
2167
2168         switch (ioctl) {
2169         case KVM_SET_SIGNAL_MASK: {
2170                 struct kvm_signal_mask __user *sigmask_arg = argp;
2171                 struct kvm_signal_mask kvm_sigmask;
2172                 compat_sigset_t csigset;
2173                 sigset_t sigset;
2174
2175                 if (argp) {
2176                         r = -EFAULT;
2177                         if (copy_from_user(&kvm_sigmask, argp,
2178                                            sizeof kvm_sigmask))
2179                                 goto out;
2180                         r = -EINVAL;
2181                         if (kvm_sigmask.len != sizeof csigset)
2182                                 goto out;
2183                         r = -EFAULT;
2184                         if (copy_from_user(&csigset, sigmask_arg->sigset,
2185                                            sizeof csigset))
2186                                 goto out;
2187                         sigset_from_compat(&sigset, &csigset);
2188                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2189                 } else
2190                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2191                 break;
2192         }
2193         default:
2194                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2195         }
2196
2197 out:
2198         return r;
2199 }
2200 #endif
2201
2202 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2203                                  int (*accessor)(struct kvm_device *dev,
2204                                                  struct kvm_device_attr *attr),
2205                                  unsigned long arg)
2206 {
2207         struct kvm_device_attr attr;
2208
2209         if (!accessor)
2210                 return -EPERM;
2211
2212         if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2213                 return -EFAULT;
2214
2215         return accessor(dev, &attr);
2216 }
2217
2218 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2219                              unsigned long arg)
2220 {
2221         struct kvm_device *dev = filp->private_data;
2222
2223         switch (ioctl) {
2224         case KVM_SET_DEVICE_ATTR:
2225                 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2226         case KVM_GET_DEVICE_ATTR:
2227                 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2228         case KVM_HAS_DEVICE_ATTR:
2229                 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2230         default:
2231                 if (dev->ops->ioctl)
2232                         return dev->ops->ioctl(dev, ioctl, arg);
2233
2234                 return -ENOTTY;
2235         }
2236 }
2237
2238 static int kvm_device_release(struct inode *inode, struct file *filp)
2239 {
2240         struct kvm_device *dev = filp->private_data;
2241         struct kvm *kvm = dev->kvm;
2242
2243         kvm_put_kvm(kvm);
2244         return 0;
2245 }
2246
2247 static const struct file_operations kvm_device_fops = {
2248         .unlocked_ioctl = kvm_device_ioctl,
2249 #ifdef CONFIG_COMPAT
2250         .compat_ioctl = kvm_device_ioctl,
2251 #endif
2252         .release = kvm_device_release,
2253 };
2254
2255 struct kvm_device *kvm_device_from_filp(struct file *filp)
2256 {
2257         if (filp->f_op != &kvm_device_fops)
2258                 return NULL;
2259
2260         return filp->private_data;
2261 }
2262
2263 static int kvm_ioctl_create_device(struct kvm *kvm,
2264                                    struct kvm_create_device *cd)
2265 {
2266         struct kvm_device_ops *ops = NULL;
2267         struct kvm_device *dev;
2268         bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2269         int ret;
2270
2271         switch (cd->type) {
2272 #ifdef CONFIG_KVM_MPIC
2273         case KVM_DEV_TYPE_FSL_MPIC_20:
2274         case KVM_DEV_TYPE_FSL_MPIC_42:
2275                 ops = &kvm_mpic_ops;
2276                 break;
2277 #endif
2278 #ifdef CONFIG_KVM_XICS
2279         case KVM_DEV_TYPE_XICS:
2280                 ops = &kvm_xics_ops;
2281                 break;
2282 #endif
2283 #ifdef CONFIG_KVM_VFIO
2284         case KVM_DEV_TYPE_VFIO:
2285                 ops = &kvm_vfio_ops;
2286                 break;
2287 #endif
2288 #ifdef CONFIG_KVM_ARM_VGIC
2289         case KVM_DEV_TYPE_ARM_VGIC_V2:
2290                 ops = &kvm_arm_vgic_v2_ops;
2291                 break;
2292 #endif
2293 #ifdef CONFIG_S390
2294         case KVM_DEV_TYPE_FLIC:
2295                 ops = &kvm_flic_ops;
2296                 break;
2297 #endif
2298         default:
2299                 return -ENODEV;
2300         }
2301
2302         if (test)
2303                 return 0;
2304
2305         dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2306         if (!dev)
2307                 return -ENOMEM;
2308
2309         dev->ops = ops;
2310         dev->kvm = kvm;
2311
2312         ret = ops->create(dev, cd->type);
2313         if (ret < 0) {
2314                 kfree(dev);
2315                 return ret;
2316         }
2317
2318         ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2319         if (ret < 0) {
2320                 ops->destroy(dev);
2321                 return ret;
2322         }
2323
2324         list_add(&dev->vm_node, &kvm->devices);
2325         kvm_get_kvm(kvm);
2326         cd->fd = ret;
2327         return 0;
2328 }
2329
2330 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2331 {
2332         switch (arg) {
2333         case KVM_CAP_USER_MEMORY:
2334         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2335         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2336 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2337         case KVM_CAP_SET_BOOT_CPU_ID:
2338 #endif
2339         case KVM_CAP_INTERNAL_ERROR_DATA:
2340 #ifdef CONFIG_HAVE_KVM_MSI
2341         case KVM_CAP_SIGNAL_MSI:
2342 #endif
2343 #ifdef CONFIG_HAVE_KVM_IRQFD
2344         case KVM_CAP_IRQFD_RESAMPLE:
2345 #endif
2346         case KVM_CAP_CHECK_EXTENSION_VM:
2347                 return 1;
2348 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2349         case KVM_CAP_IRQ_ROUTING:
2350                 return KVM_MAX_IRQ_ROUTES;
2351 #endif
2352         default:
2353                 break;
2354         }
2355         return kvm_vm_ioctl_check_extension(kvm, arg);
2356 }
2357
2358 static long kvm_vm_ioctl(struct file *filp,
2359                            unsigned int ioctl, unsigned long arg)
2360 {
2361         struct kvm *kvm = filp->private_data;
2362         void __user *argp = (void __user *)arg;
2363         int r;
2364
2365         if (kvm->mm != current->mm)
2366                 return -EIO;
2367         switch (ioctl) {
2368         case KVM_CREATE_VCPU:
2369                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2370                 break;
2371         case KVM_SET_USER_MEMORY_REGION: {
2372                 struct kvm_userspace_memory_region kvm_userspace_mem;
2373
2374                 r = -EFAULT;
2375                 if (copy_from_user(&kvm_userspace_mem, argp,
2376                                                 sizeof kvm_userspace_mem))
2377                         goto out;
2378
2379                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2380                 break;
2381         }
2382         case KVM_GET_DIRTY_LOG: {
2383                 struct kvm_dirty_log log;
2384
2385                 r = -EFAULT;
2386                 if (copy_from_user(&log, argp, sizeof log))
2387                         goto out;
2388                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2389                 break;
2390         }
2391 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2392         case KVM_REGISTER_COALESCED_MMIO: {
2393                 struct kvm_coalesced_mmio_zone zone;
2394                 r = -EFAULT;
2395                 if (copy_from_user(&zone, argp, sizeof zone))
2396                         goto out;
2397                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2398                 break;
2399         }
2400         case KVM_UNREGISTER_COALESCED_MMIO: {
2401                 struct kvm_coalesced_mmio_zone zone;
2402                 r = -EFAULT;
2403                 if (copy_from_user(&zone, argp, sizeof zone))
2404                         goto out;
2405                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2406                 break;
2407         }
2408 #endif
2409         case KVM_IRQFD: {
2410                 struct kvm_irqfd data;
2411
2412                 r = -EFAULT;
2413                 if (copy_from_user(&data, argp, sizeof data))
2414                         goto out;
2415                 r = kvm_irqfd(kvm, &data);
2416                 break;
2417         }
2418         case KVM_IOEVENTFD: {
2419                 struct kvm_ioeventfd data;
2420
2421                 r = -EFAULT;
2422                 if (copy_from_user(&data, argp, sizeof data))
2423                         goto out;
2424                 r = kvm_ioeventfd(kvm, &data);
2425                 break;
2426         }
2427 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2428         case KVM_SET_BOOT_CPU_ID:
2429                 r = 0;
2430                 mutex_lock(&kvm->lock);
2431                 if (atomic_read(&kvm->online_vcpus) != 0)
2432                         r = -EBUSY;
2433                 else
2434                         kvm->bsp_vcpu_id = arg;
2435                 mutex_unlock(&kvm->lock);
2436                 break;
2437 #endif
2438 #ifdef CONFIG_HAVE_KVM_MSI
2439         case KVM_SIGNAL_MSI: {
2440                 struct kvm_msi msi;
2441
2442                 r = -EFAULT;
2443                 if (copy_from_user(&msi, argp, sizeof msi))
2444                         goto out;
2445                 r = kvm_send_userspace_msi(kvm, &msi);
2446                 break;
2447         }
2448 #endif
2449 #ifdef __KVM_HAVE_IRQ_LINE
2450         case KVM_IRQ_LINE_STATUS:
2451         case KVM_IRQ_LINE: {
2452                 struct kvm_irq_level irq_event;
2453
2454                 r = -EFAULT;
2455                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2456                         goto out;
2457
2458                 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
2459                                         ioctl == KVM_IRQ_LINE_STATUS);
2460                 if (r)
2461                         goto out;
2462
2463                 r = -EFAULT;
2464                 if (ioctl == KVM_IRQ_LINE_STATUS) {
2465                         if (copy_to_user(argp, &irq_event, sizeof irq_event))
2466                                 goto out;
2467                 }
2468
2469                 r = 0;
2470                 break;
2471         }
2472 #endif
2473 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2474         case KVM_SET_GSI_ROUTING: {
2475                 struct kvm_irq_routing routing;
2476                 struct kvm_irq_routing __user *urouting;
2477                 struct kvm_irq_routing_entry *entries;
2478
2479                 r = -EFAULT;
2480                 if (copy_from_user(&routing, argp, sizeof(routing)))
2481                         goto out;
2482                 r = -EINVAL;
2483                 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2484                         goto out;
2485                 if (routing.flags)
2486                         goto out;
2487                 r = -ENOMEM;
2488                 entries = vmalloc(routing.nr * sizeof(*entries));
2489                 if (!entries)
2490                         goto out;
2491                 r = -EFAULT;
2492                 urouting = argp;
2493                 if (copy_from_user(entries, urouting->entries,
2494                                    routing.nr * sizeof(*entries)))
2495                         goto out_free_irq_routing;
2496                 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2497                                         routing.flags);
2498         out_free_irq_routing:
2499                 vfree(entries);
2500                 break;
2501         }
2502 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
2503         case KVM_CREATE_DEVICE: {
2504                 struct kvm_create_device cd;
2505
2506                 r = -EFAULT;
2507                 if (copy_from_user(&cd, argp, sizeof(cd)))
2508                         goto out;
2509
2510                 r = kvm_ioctl_create_device(kvm, &cd);
2511                 if (r)
2512                         goto out;
2513
2514                 r = -EFAULT;
2515                 if (copy_to_user(argp, &cd, sizeof(cd)))
2516                         goto out;
2517
2518                 r = 0;
2519                 break;
2520         }
2521         case KVM_CHECK_EXTENSION:
2522                 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
2523                 break;
2524         default:
2525                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2526                 if (r == -ENOTTY)
2527                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2528         }
2529 out:
2530         return r;
2531 }
2532
2533 #ifdef CONFIG_COMPAT
2534 struct compat_kvm_dirty_log {
2535         __u32 slot;
2536         __u32 padding1;
2537         union {
2538                 compat_uptr_t dirty_bitmap; /* one bit per page */
2539                 __u64 padding2;
2540         };
2541 };
2542
2543 static long kvm_vm_compat_ioctl(struct file *filp,
2544                            unsigned int ioctl, unsigned long arg)
2545 {
2546         struct kvm *kvm = filp->private_data;
2547         int r;
2548
2549         if (kvm->mm != current->mm)
2550                 return -EIO;
2551         switch (ioctl) {
2552         case KVM_GET_DIRTY_LOG: {
2553                 struct compat_kvm_dirty_log compat_log;
2554                 struct kvm_dirty_log log;
2555
2556                 r = -EFAULT;
2557                 if (copy_from_user(&compat_log, (void __user *)arg,
2558                                    sizeof(compat_log)))
2559                         goto out;
2560                 log.slot         = compat_log.slot;
2561                 log.padding1     = compat_log.padding1;
2562                 log.padding2     = compat_log.padding2;
2563                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2564
2565                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2566                 break;
2567         }
2568         default:
2569                 r = kvm_vm_ioctl(filp, ioctl, arg);
2570         }
2571
2572 out:
2573         return r;
2574 }
2575 #endif
2576
2577 static struct file_operations kvm_vm_fops = {
2578         .release        = kvm_vm_release,
2579         .unlocked_ioctl = kvm_vm_ioctl,
2580 #ifdef CONFIG_COMPAT
2581         .compat_ioctl   = kvm_vm_compat_ioctl,
2582 #endif
2583         .llseek         = noop_llseek,
2584 };
2585
2586 static int kvm_dev_ioctl_create_vm(unsigned long type)
2587 {
2588         int r;
2589         struct kvm *kvm;
2590
2591         kvm = kvm_create_vm(type);
2592         if (IS_ERR(kvm))
2593                 return PTR_ERR(kvm);
2594 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2595         r = kvm_coalesced_mmio_init(kvm);
2596         if (r < 0) {
2597                 kvm_put_kvm(kvm);
2598                 return r;
2599         }
2600 #endif
2601         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC);
2602         if (r < 0)
2603                 kvm_put_kvm(kvm);
2604
2605         return r;
2606 }
2607
2608 static long kvm_dev_ioctl(struct file *filp,
2609                           unsigned int ioctl, unsigned long arg)
2610 {
2611         long r = -EINVAL;
2612
2613         switch (ioctl) {
2614         case KVM_GET_API_VERSION:
2615                 r = -EINVAL;
2616                 if (arg)
2617                         goto out;
2618                 r = KVM_API_VERSION;
2619                 break;
2620         case KVM_CREATE_VM:
2621                 r = kvm_dev_ioctl_create_vm(arg);
2622                 break;
2623         case KVM_CHECK_EXTENSION:
2624                 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
2625                 break;
2626         case KVM_GET_VCPU_MMAP_SIZE:
2627                 r = -EINVAL;
2628                 if (arg)
2629                         goto out;
2630                 r = PAGE_SIZE;     /* struct kvm_run */
2631 #ifdef CONFIG_X86
2632                 r += PAGE_SIZE;    /* pio data page */
2633 #endif
2634 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2635                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2636 #endif
2637                 break;
2638         case KVM_TRACE_ENABLE:
2639         case KVM_TRACE_PAUSE:
2640         case KVM_TRACE_DISABLE:
2641                 r = -EOPNOTSUPP;
2642                 break;
2643         default:
2644                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2645         }
2646 out:
2647         return r;
2648 }
2649
2650 static struct file_operations kvm_chardev_ops = {
2651         .unlocked_ioctl = kvm_dev_ioctl,
2652         .compat_ioctl   = kvm_dev_ioctl,
2653         .llseek         = noop_llseek,
2654 };
2655
2656 static struct miscdevice kvm_dev = {
2657         KVM_MINOR,
2658         "kvm",
2659         &kvm_chardev_ops,
2660 };
2661
2662 static void hardware_enable_nolock(void *junk)
2663 {
2664         int cpu = raw_smp_processor_id();
2665         int r;
2666
2667         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2668                 return;
2669
2670         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2671
2672         r = kvm_arch_hardware_enable(NULL);
2673
2674         if (r) {
2675                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2676                 atomic_inc(&hardware_enable_failed);
2677                 printk(KERN_INFO "kvm: enabling virtualization on "
2678                                  "CPU%d failed\n", cpu);
2679         }
2680 }
2681
2682 static void hardware_enable(void)
2683 {
2684         raw_spin_lock(&kvm_count_lock);
2685         if (kvm_usage_count)
2686                 hardware_enable_nolock(NULL);
2687         raw_spin_unlock(&kvm_count_lock);
2688 }
2689
2690 static void hardware_disable_nolock(void *junk)
2691 {
2692         int cpu = raw_smp_processor_id();
2693
2694         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2695                 return;
2696         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2697         kvm_arch_hardware_disable(NULL);
2698 }
2699
2700 static void hardware_disable(void)
2701 {
2702         raw_spin_lock(&kvm_count_lock);
2703         if (kvm_usage_count)
2704                 hardware_disable_nolock(NULL);
2705         raw_spin_unlock(&kvm_count_lock);
2706 }
2707
2708 static void hardware_disable_all_nolock(void)
2709 {
2710         BUG_ON(!kvm_usage_count);
2711
2712         kvm_usage_count--;
2713         if (!kvm_usage_count)
2714                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2715 }
2716
2717 static void hardware_disable_all(void)
2718 {
2719         raw_spin_lock(&kvm_count_lock);
2720         hardware_disable_all_nolock();
2721         raw_spin_unlock(&kvm_count_lock);
2722 }
2723
2724 static int hardware_enable_all(void)
2725 {
2726         int r = 0;
2727
2728         raw_spin_lock(&kvm_count_lock);
2729
2730         kvm_usage_count++;
2731         if (kvm_usage_count == 1) {
2732                 atomic_set(&hardware_enable_failed, 0);
2733                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2734
2735                 if (atomic_read(&hardware_enable_failed)) {
2736                         hardware_disable_all_nolock();
2737                         r = -EBUSY;
2738                 }
2739         }
2740
2741         raw_spin_unlock(&kvm_count_lock);
2742
2743         return r;
2744 }
2745
2746 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2747                            void *v)
2748 {
2749         int cpu = (long)v;
2750
2751         val &= ~CPU_TASKS_FROZEN;
2752         switch (val) {
2753         case CPU_DYING:
2754                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2755                        cpu);
2756                 hardware_disable();
2757                 break;
2758         case CPU_STARTING:
2759                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2760                        cpu);
2761                 hardware_enable();
2762                 break;
2763         }
2764         return NOTIFY_OK;
2765 }
2766
2767 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2768                       void *v)
2769 {
2770         /*
2771          * Some (well, at least mine) BIOSes hang on reboot if
2772          * in vmx root mode.
2773          *
2774          * And Intel TXT required VMX off for all cpu when system shutdown.
2775          */
2776         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2777         kvm_rebooting = true;
2778         on_each_cpu(hardware_disable_nolock, NULL, 1);
2779         return NOTIFY_OK;
2780 }
2781
2782 static struct notifier_block kvm_reboot_notifier = {
2783         .notifier_call = kvm_reboot,
2784         .priority = 0,
2785 };
2786
2787 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2788 {
2789         int i;
2790
2791         for (i = 0; i < bus->dev_count; i++) {
2792                 struct kvm_io_device *pos = bus->range[i].dev;
2793
2794                 kvm_iodevice_destructor(pos);
2795         }
2796         kfree(bus);
2797 }
2798
2799 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
2800                                  const struct kvm_io_range *r2)
2801 {
2802         if (r1->addr < r2->addr)
2803                 return -1;
2804         if (r1->addr + r1->len > r2->addr + r2->len)
2805                 return 1;
2806         return 0;
2807 }
2808
2809 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2810 {
2811         return kvm_io_bus_cmp(p1, p2);
2812 }
2813
2814 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2815                           gpa_t addr, int len)
2816 {
2817         bus->range[bus->dev_count++] = (struct kvm_io_range) {
2818                 .addr = addr,
2819                 .len = len,
2820                 .dev = dev,
2821         };
2822
2823         sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2824                 kvm_io_bus_sort_cmp, NULL);
2825
2826         return 0;
2827 }
2828
2829 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2830                              gpa_t addr, int len)
2831 {
2832         struct kvm_io_range *range, key;
2833         int off;
2834
2835         key = (struct kvm_io_range) {
2836                 .addr = addr,
2837                 .len = len,
2838         };
2839
2840         range = bsearch(&key, bus->range, bus->dev_count,
2841                         sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2842         if (range == NULL)
2843                 return -ENOENT;
2844
2845         off = range - bus->range;
2846
2847         while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
2848                 off--;
2849
2850         return off;
2851 }
2852
2853 static int __kvm_io_bus_write(struct kvm_io_bus *bus,
2854                               struct kvm_io_range *range, const void *val)
2855 {
2856         int idx;
2857
2858         idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2859         if (idx < 0)
2860                 return -EOPNOTSUPP;
2861
2862         while (idx < bus->dev_count &&
2863                 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
2864                 if (!kvm_iodevice_write(bus->range[idx].dev, range->addr,
2865                                         range->len, val))
2866                         return idx;
2867                 idx++;
2868         }
2869
2870         return -EOPNOTSUPP;
2871 }
2872
2873 /* kvm_io_bus_write - called under kvm->slots_lock */
2874 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2875                      int len, const void *val)
2876 {
2877         struct kvm_io_bus *bus;
2878         struct kvm_io_range range;
2879         int r;
2880
2881         range = (struct kvm_io_range) {
2882                 .addr = addr,
2883                 .len = len,
2884         };
2885
2886         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2887         r = __kvm_io_bus_write(bus, &range, val);
2888         return r < 0 ? r : 0;
2889 }
2890
2891 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
2892 int kvm_io_bus_write_cookie(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2893                             int len, const void *val, long cookie)
2894 {
2895         struct kvm_io_bus *bus;
2896         struct kvm_io_range range;
2897
2898         range = (struct kvm_io_range) {
2899                 .addr = addr,
2900                 .len = len,
2901         };
2902
2903         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2904
2905         /* First try the device referenced by cookie. */
2906         if ((cookie >= 0) && (cookie < bus->dev_count) &&
2907             (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
2908                 if (!kvm_iodevice_write(bus->range[cookie].dev, addr, len,
2909                                         val))
2910                         return cookie;
2911
2912         /*
2913          * cookie contained garbage; fall back to search and return the
2914          * correct cookie value.
2915          */
2916         return __kvm_io_bus_write(bus, &range, val);
2917 }
2918
2919 static int __kvm_io_bus_read(struct kvm_io_bus *bus, struct kvm_io_range *range,
2920                              void *val)
2921 {
2922         int idx;
2923
2924         idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2925         if (idx < 0)
2926                 return -EOPNOTSUPP;
2927
2928         while (idx < bus->dev_count &&
2929                 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
2930                 if (!kvm_iodevice_read(bus->range[idx].dev, range->addr,
2931                                        range->len, val))
2932                         return idx;
2933                 idx++;
2934         }
2935
2936         return -EOPNOTSUPP;
2937 }
2938 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
2939
2940 /* kvm_io_bus_read - called under kvm->slots_lock */
2941 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2942                     int len, void *val)
2943 {
2944         struct kvm_io_bus *bus;
2945         struct kvm_io_range range;
2946         int r;
2947
2948         range = (struct kvm_io_range) {
2949                 .addr = addr,
2950                 .len = len,
2951         };
2952
2953         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2954         r = __kvm_io_bus_read(bus, &range, val);
2955         return r < 0 ? r : 0;
2956 }
2957
2958
2959 /* Caller must hold slots_lock. */
2960 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2961                             int len, struct kvm_io_device *dev)
2962 {
2963         struct kvm_io_bus *new_bus, *bus;
2964
2965         bus = kvm->buses[bus_idx];
2966         /* exclude ioeventfd which is limited by maximum fd */
2967         if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
2968                 return -ENOSPC;
2969
2970         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
2971                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2972         if (!new_bus)
2973                 return -ENOMEM;
2974         memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
2975                sizeof(struct kvm_io_range)));
2976         kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2977         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2978         synchronize_srcu_expedited(&kvm->srcu);
2979         kfree(bus);
2980
2981         return 0;
2982 }
2983
2984 /* Caller must hold slots_lock. */
2985 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2986                               struct kvm_io_device *dev)
2987 {
2988         int i, r;
2989         struct kvm_io_bus *new_bus, *bus;
2990
2991         bus = kvm->buses[bus_idx];
2992         r = -ENOENT;
2993         for (i = 0; i < bus->dev_count; i++)
2994                 if (bus->range[i].dev == dev) {
2995                         r = 0;
2996                         break;
2997                 }
2998
2999         if (r)
3000                 return r;
3001
3002         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3003                           sizeof(struct kvm_io_range)), GFP_KERNEL);
3004         if (!new_bus)
3005                 return -ENOMEM;
3006
3007         memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3008         new_bus->dev_count--;
3009         memcpy(new_bus->range + i, bus->range + i + 1,
3010                (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3011
3012         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3013         synchronize_srcu_expedited(&kvm->srcu);
3014         kfree(bus);
3015         return r;
3016 }
3017
3018 static struct notifier_block kvm_cpu_notifier = {
3019         .notifier_call = kvm_cpu_hotplug,
3020 };
3021
3022 static int vm_stat_get(void *_offset, u64 *val)
3023 {
3024         unsigned offset = (long)_offset;
3025         struct kvm *kvm;
3026
3027         *val = 0;
3028         spin_lock(&kvm_lock);
3029         list_for_each_entry(kvm, &vm_list, vm_list)
3030                 *val += *(u32 *)((void *)kvm + offset);
3031         spin_unlock(&kvm_lock);
3032         return 0;
3033 }
3034
3035 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
3036
3037 static int vcpu_stat_get(void *_offset, u64 *val)
3038 {
3039         unsigned offset = (long)_offset;
3040         struct kvm *kvm;
3041         struct kvm_vcpu *vcpu;
3042         int i;
3043
3044         *val = 0;
3045         spin_lock(&kvm_lock);
3046         list_for_each_entry(kvm, &vm_list, vm_list)
3047                 kvm_for_each_vcpu(i, vcpu, kvm)
3048                         *val += *(u32 *)((void *)vcpu + offset);
3049
3050         spin_unlock(&kvm_lock);
3051         return 0;
3052 }
3053
3054 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
3055
3056 static const struct file_operations *stat_fops[] = {
3057         [KVM_STAT_VCPU] = &vcpu_stat_fops,
3058         [KVM_STAT_VM]   = &vm_stat_fops,
3059 };
3060
3061 static int kvm_init_debug(void)
3062 {
3063         int r = -EEXIST;
3064         struct kvm_stats_debugfs_item *p;
3065
3066         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3067         if (kvm_debugfs_dir == NULL)
3068                 goto out;
3069
3070         for (p = debugfs_entries; p->name; ++p) {
3071                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
3072                                                 (void *)(long)p->offset,
3073                                                 stat_fops[p->kind]);
3074                 if (p->dentry == NULL)
3075                         goto out_dir;
3076         }
3077
3078         return 0;
3079
3080 out_dir:
3081         debugfs_remove_recursive(kvm_debugfs_dir);
3082 out:
3083         return r;
3084 }
3085
3086 static void kvm_exit_debug(void)
3087 {
3088         struct kvm_stats_debugfs_item *p;
3089
3090         for (p = debugfs_entries; p->name; ++p)
3091                 debugfs_remove(p->dentry);
3092         debugfs_remove(kvm_debugfs_dir);
3093 }
3094
3095 static int kvm_suspend(void)
3096 {
3097         if (kvm_usage_count)
3098                 hardware_disable_nolock(NULL);
3099         return 0;
3100 }
3101
3102 static void kvm_resume(void)
3103 {
3104         if (kvm_usage_count) {
3105                 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3106                 hardware_enable_nolock(NULL);
3107         }
3108 }
3109
3110 static struct syscore_ops kvm_syscore_ops = {
3111         .suspend = kvm_suspend,
3112         .resume = kvm_resume,
3113 };
3114
3115 static inline
3116 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3117 {
3118         return container_of(pn, struct kvm_vcpu, preempt_notifier);
3119 }
3120
3121 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3122 {
3123         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3124         if (vcpu->preempted)
3125                 vcpu->preempted = false;
3126
3127         kvm_arch_sched_in(vcpu, cpu);
3128
3129         kvm_arch_vcpu_load(vcpu, cpu);
3130 }
3131
3132 static void kvm_sched_out(struct preempt_notifier *pn,
3133                           struct task_struct *next)
3134 {
3135         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3136
3137         if (current->state == TASK_RUNNING)
3138                 vcpu->preempted = true;
3139         kvm_arch_vcpu_put(vcpu);
3140 }
3141
3142 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3143                   struct module *module)
3144 {
3145         int r;
3146         int cpu;
3147
3148         r = kvm_arch_init(opaque);
3149         if (r)
3150                 goto out_fail;
3151
3152         /*
3153          * kvm_arch_init makes sure there's at most one caller
3154          * for architectures that support multiple implementations,
3155          * like intel and amd on x86.
3156          * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3157          * conflicts in case kvm is already setup for another implementation.
3158          */
3159         r = kvm_irqfd_init();
3160         if (r)
3161                 goto out_irqfd;
3162
3163         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
3164                 r = -ENOMEM;
3165                 goto out_free_0;
3166         }
3167
3168         r = kvm_arch_hardware_setup();
3169         if (r < 0)
3170                 goto out_free_0a;
3171
3172         for_each_online_cpu(cpu) {
3173                 smp_call_function_single(cpu,
3174                                 kvm_arch_check_processor_compat,
3175                                 &r, 1);
3176                 if (r < 0)
3177                         goto out_free_1;
3178         }
3179
3180         r = register_cpu_notifier(&kvm_cpu_notifier);
3181         if (r)
3182                 goto out_free_2;
3183         register_reboot_notifier(&kvm_reboot_notifier);
3184
3185         /* A kmem cache lets us meet the alignment requirements of fx_save. */
3186         if (!vcpu_align)
3187                 vcpu_align = __alignof__(struct kvm_vcpu);
3188         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
3189                                            0, NULL);
3190         if (!kvm_vcpu_cache) {
3191                 r = -ENOMEM;
3192                 goto out_free_3;
3193         }
3194
3195         r = kvm_async_pf_init();
3196         if (r)
3197                 goto out_free;
3198
3199         kvm_chardev_ops.owner = module;
3200         kvm_vm_fops.owner = module;
3201         kvm_vcpu_fops.owner = module;
3202
3203         r = misc_register(&kvm_dev);
3204         if (r) {
3205                 printk(KERN_ERR "kvm: misc device register failed\n");
3206                 goto out_unreg;
3207         }
3208
3209         register_syscore_ops(&kvm_syscore_ops);
3210
3211         kvm_preempt_ops.sched_in = kvm_sched_in;
3212         kvm_preempt_ops.sched_out = kvm_sched_out;
3213
3214         r = kvm_init_debug();
3215         if (r) {
3216                 printk(KERN_ERR "kvm: create debugfs files failed\n");
3217                 goto out_undebugfs;
3218         }
3219
3220         return 0;
3221
3222 out_undebugfs:
3223         unregister_syscore_ops(&kvm_syscore_ops);
3224         misc_deregister(&kvm_dev);
3225 out_unreg:
3226         kvm_async_pf_deinit();
3227 out_free:
3228         kmem_cache_destroy(kvm_vcpu_cache);
3229 out_free_3:
3230         unregister_reboot_notifier(&kvm_reboot_notifier);
3231         unregister_cpu_notifier(&kvm_cpu_notifier);
3232 out_free_2:
3233 out_free_1:
3234         kvm_arch_hardware_unsetup();
3235 out_free_0a:
3236         free_cpumask_var(cpus_hardware_enabled);
3237 out_free_0:
3238         kvm_irqfd_exit();
3239 out_irqfd:
3240         kvm_arch_exit();
3241 out_fail:
3242         return r;
3243 }
3244 EXPORT_SYMBOL_GPL(kvm_init);
3245
3246 void kvm_exit(void)
3247 {
3248         kvm_exit_debug();
3249         misc_deregister(&kvm_dev);
3250         kmem_cache_destroy(kvm_vcpu_cache);
3251         kvm_async_pf_deinit();
3252         unregister_syscore_ops(&kvm_syscore_ops);
3253         unregister_reboot_notifier(&kvm_reboot_notifier);
3254         unregister_cpu_notifier(&kvm_cpu_notifier);
3255         on_each_cpu(hardware_disable_nolock, NULL, 1);
3256         kvm_arch_hardware_unsetup();
3257         kvm_arch_exit();
3258         kvm_irqfd_exit();
3259         free_cpumask_var(cpus_hardware_enabled);
3260 }
3261 EXPORT_SYMBOL_GPL(kvm_exit);