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 <kvm/iodev.h>
  20
  21 #include <linux/kvm_host.h>
  22 #include <linux/kvm.h>
  23 #include <linux/module.h>
  24 #include <linux/errno.h>
  25 #include <linux/percpu.h>
  26 #include <linux/mm.h>
  27 #include <linux/miscdevice.h>
  28 #include <linux/vmalloc.h>
  29 #include <linux/reboot.h>
  30 #include <linux/debugfs.h>
  31 #include <linux/highmem.h>
  32 #include <linux/file.h>
  33 #include <linux/syscore_ops.h>
  34 #include <linux/cpu.h>
  35 #include <linux/sched/signal.h>
  36 #include <linux/sched/mm.h>
  37 #include <linux/sched/stat.h>
  38 #include <linux/cpumask.h>
  39 #include <linux/smp.h>
  40 #include <linux/anon_inodes.h>
  41 #include <linux/profile.h>
  42 #include <linux/kvm_para.h>
  43 #include <linux/pagemap.h>
  44 #include <linux/mman.h>
  45 #include <linux/swap.h>
  46 #include <linux/bitops.h>
  47 #include <linux/spinlock.h>
  48 #include <linux/compat.h>
  49 #include <linux/srcu.h>
  50 #include <linux/hugetlb.h>
  51 #include <linux/slab.h>
  52 #include <linux/sort.h>
  53 #include <linux/bsearch.h>
  54
  55 #include <asm/processor.h>
  56 #include <asm/io.h>
  57 #include <asm/ioctl.h>
  58 #include <linux/uaccess.h>
  59 #include <asm/pgtable.h>
  60
  61 #include "coalesced_mmio.h"
  62 #include "async_pf.h"
  63 #include "vfio.h"
  64
  65 #define CREATE_TRACE_POINTS
  66 #include <trace/events/kvm.h>
  67
  68 /* Worst case buffer size needed for holding an integer. */
  69 #define ITOA_MAX_LEN 12
  70
  71 MODULE_AUTHOR("Qumranet");
  72 MODULE_LICENSE("GPL");
  73
  74 /* Architectures should define their poll value according to the halt latency */
  75 unsigned int halt_poll_ns = KVM_HALT_POLL_NS_DEFAULT;
  76 module_param(halt_poll_ns, uint, 0644);
  77 EXPORT_SYMBOL_GPL(halt_poll_ns);
  78
  79 /* Default doubles per-vcpu halt_poll_ns. */
  80 unsigned int halt_poll_ns_grow = 2;
  81 module_param(halt_poll_ns_grow, uint, 0644);
  82 EXPORT_SYMBOL_GPL(halt_poll_ns_grow);
  83
  84 /* Default resets per-vcpu halt_poll_ns . */
  85 unsigned int halt_poll_ns_shrink;
  86 module_param(halt_poll_ns_shrink, uint, 0644);
  87 EXPORT_SYMBOL_GPL(halt_poll_ns_shrink);
  88
  89 /*
  90  * Ordering of locks:
  91  *
  92  *      kvm->lock --> kvm->slots_lock --> kvm->irq_lock
  93  */
  94
  95 DEFINE_SPINLOCK(kvm_lock);
  96 static DEFINE_RAW_SPINLOCK(kvm_count_lock);
  97 LIST_HEAD(vm_list);
  98
  99 static cpumask_var_t cpus_hardware_enabled;
 100 static int kvm_usage_count;
 101 static atomic_t hardware_enable_failed;
 102
 103 struct kmem_cache *kvm_vcpu_cache;
 104 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
 105
 106 static __read_mostly struct preempt_ops kvm_preempt_ops;
 107
 108 struct dentry *kvm_debugfs_dir;
 109 EXPORT_SYMBOL_GPL(kvm_debugfs_dir);
 110
 111 static int kvm_debugfs_num_entries;
 112 static const struct file_operations *stat_fops_per_vm[];
 113
 114 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
 115                            unsigned long arg);
 116 #ifdef CONFIG_KVM_COMPAT
 117 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
 118                                   unsigned long arg);
 119 #endif
 120 static int hardware_enable_all(void);
 121 static void hardware_disable_all(void);
 122
 123 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
 124
 125 static void kvm_release_pfn_dirty(kvm_pfn_t pfn);
 126 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot, gfn_t gfn);
 127
 128 __visible bool kvm_rebooting;
 129 EXPORT_SYMBOL_GPL(kvm_rebooting);
 130
 131 static bool largepages_enabled = true;
 132
 133 #define KVM_EVENT_CREATE_VM 0
 134 #define KVM_EVENT_DESTROY_VM 1
 135 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm);
 136 static unsigned long long kvm_createvm_count;
 137 static unsigned long long kvm_active_vms;
 138
 139 bool kvm_is_reserved_pfn(kvm_pfn_t pfn)
 140 {
 141         if (pfn_valid(pfn))
 142                 return PageReserved(pfn_to_page(pfn));
 143
 144         return true;
 145 }
 146
 147 /*
 148  * Switches to specified vcpu, until a matching vcpu_put()
 149  */
 150 int vcpu_load(struct kvm_vcpu *vcpu)
 151 {
 152         int cpu;
 153
 154         if (mutex_lock_killable(&vcpu->mutex))
 155                 return -EINTR;
 156         cpu = get_cpu();
 157         preempt_notifier_register(&vcpu->preempt_notifier);
 158         kvm_arch_vcpu_load(vcpu, cpu);
 159         put_cpu();
 160         return 0;
 161 }
 162 EXPORT_SYMBOL_GPL(vcpu_load);
 163
 164 void vcpu_put(struct kvm_vcpu *vcpu)
 165 {
 166         preempt_disable();
 167         kvm_arch_vcpu_put(vcpu);
 168         preempt_notifier_unregister(&vcpu->preempt_notifier);
 169         preempt_enable();
 170         mutex_unlock(&vcpu->mutex);
 171 }
 172 EXPORT_SYMBOL_GPL(vcpu_put);
 173
 174 /* TODO: merge with kvm_arch_vcpu_should_kick */
 175 static bool kvm_request_needs_ipi(struct kvm_vcpu *vcpu, unsigned req)
 176 {
 177         int mode = kvm_vcpu_exiting_guest_mode(vcpu);
 178
 179         /*
 180          * We need to wait for the VCPU to reenable interrupts and get out of
 181          * READING_SHADOW_PAGE_TABLES mode.
 182          */
 183         if (req & KVM_REQUEST_WAIT)
 184                 return mode != OUTSIDE_GUEST_MODE;
 185
 186         /*
 187          * Need to kick a running VCPU, but otherwise there is nothing to do.
 188          */
 189         return mode == IN_GUEST_MODE;
 190 }
 191
 192 static void ack_flush(void *_completed)
 193 {
 194 }
 195
 196 static inline bool kvm_kick_many_cpus(const struct cpumask *cpus, bool wait)
 197 {
 198         if (unlikely(!cpus))
 199                 cpus = cpu_online_mask;
 200
 201         if (cpumask_empty(cpus))
 202                 return false;
 203
 204         smp_call_function_many(cpus, ack_flush, NULL, wait);
 205         return true;
 206 }
 207
 208 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
 209 {
 210         int i, cpu, me;
 211         cpumask_var_t cpus;
 212         bool called;
 213         struct kvm_vcpu *vcpu;
 214
 215         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
 216
 217         me = get_cpu();
 218         kvm_for_each_vcpu(i, vcpu, kvm) {
 219                 kvm_make_request(req, vcpu);
 220                 cpu = vcpu->cpu;
 221
 222                 if (!(req & KVM_REQUEST_NO_WAKEUP) && kvm_vcpu_wake_up(vcpu))
 223                         continue;
 224
 225                 if (cpus != NULL && cpu != -1 && cpu != me &&
 226                     kvm_request_needs_ipi(vcpu, req))
 227                         __cpumask_set_cpu(cpu, cpus);
 228         }
 229         called = kvm_kick_many_cpus(cpus, !!(req & KVM_REQUEST_WAIT));
 230         put_cpu();
 231         free_cpumask_var(cpus);
 232         return called;
 233 }
 234
 235 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
 236 void kvm_flush_remote_tlbs(struct kvm *kvm)
 237 {
 238         /*
 239          * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in
 240          * kvm_make_all_cpus_request.
 241          */
 242         long dirty_count = smp_load_acquire(&kvm->tlbs_dirty);
 243
 244         /*
 245          * We want to publish modifications to the page tables before reading
 246          * mode. Pairs with a memory barrier in arch-specific code.
 247          * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
 248          * and smp_mb in walk_shadow_page_lockless_begin/end.
 249          * - powerpc: smp_mb in kvmppc_prepare_to_enter.
 250          *
 251          * There is already an smp_mb__after_atomic() before
 252          * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
 253          * barrier here.
 254          */
 255         if (kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
 256                 ++kvm->stat.remote_tlb_flush;
 257         cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
 258 }
 259 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
 260 #endif
 261
 262 void kvm_reload_remote_mmus(struct kvm *kvm)
 263 {
 264         kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
 265 }
 266
 267 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 268 {
 269         struct page *page;
 270         int r;
 271
 272         mutex_init(&vcpu->mutex);
 273         vcpu->cpu = -1;
 274         vcpu->kvm = kvm;
 275         vcpu->vcpu_id = id;
 276         vcpu->pid = NULL;
 277         init_swait_queue_head(&vcpu->wq);
 278         kvm_async_pf_vcpu_init(vcpu);
 279
 280         vcpu->pre_pcpu = -1;
 281         INIT_LIST_HEAD(&vcpu->blocked_vcpu_list);
 282
 283         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 284         if (!page) {
 285                 r = -ENOMEM;
 286                 goto fail;
 287         }
 288         vcpu->run = page_address(page);
 289
 290         kvm_vcpu_set_in_spin_loop(vcpu, false);
 291         kvm_vcpu_set_dy_eligible(vcpu, false);
 292         vcpu->preempted = false;
 293
 294         r = kvm_arch_vcpu_init(vcpu);
 295         if (r < 0)
 296                 goto fail_free_run;
 297         return 0;
 298
 299 fail_free_run:
 300         free_page((unsigned long)vcpu->run);
 301 fail:
 302         return r;
 303 }
 304 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 305
 306 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 307 {
 308         /*
 309          * no need for rcu_read_lock as VCPU_RUN is the only place that
 310          * will change the vcpu->pid pointer and on uninit all file
 311          * descriptors are already gone.
 312          */
 313         put_pid(rcu_dereference_protected(vcpu->pid, 1));
 314         kvm_arch_vcpu_uninit(vcpu);
 315         free_page((unsigned long)vcpu->run);
 316 }
 317 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 318
 319 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 320 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
 321 {
 322         return container_of(mn, struct kvm, mmu_notifier);
 323 }
 324
 325 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
 326                                         struct mm_struct *mm,
 327                                         unsigned long address,
 328                                         pte_t pte)
 329 {
 330         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 331         int idx;
 332
 333         idx = srcu_read_lock(&kvm->srcu);
 334         spin_lock(&kvm->mmu_lock);
 335         kvm->mmu_notifier_seq++;
 336         kvm_set_spte_hva(kvm, address, pte);
 337         spin_unlock(&kvm->mmu_lock);
 338         srcu_read_unlock(&kvm->srcu, idx);
 339 }
 340
 341 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 342                                                     struct mm_struct *mm,
 343                                                     unsigned long start,
 344                                                     unsigned long end)
 345 {
 346         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 347         int need_tlb_flush = 0, idx;
 348
 349         idx = srcu_read_lock(&kvm->srcu);
 350         spin_lock(&kvm->mmu_lock);
 351         /*
 352          * The count increase must become visible at unlock time as no
 353          * spte can be established without taking the mmu_lock and
 354          * count is also read inside the mmu_lock critical section.
 355          */
 356         kvm->mmu_notifier_count++;
 357         need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
 358         need_tlb_flush |= kvm->tlbs_dirty;
 359         /* we've to flush the tlb before the pages can be freed */
 360         if (need_tlb_flush)
 361                 kvm_flush_remote_tlbs(kvm);
 362
 363         spin_unlock(&kvm->mmu_lock);
 364         srcu_read_unlock(&kvm->srcu, idx);
 365 }
 366
 367 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
 368                                                   struct mm_struct *mm,
 369                                                   unsigned long start,
 370                                                   unsigned long end)
 371 {
 372         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 373
 374         spin_lock(&kvm->mmu_lock);
 375         /*
 376          * This sequence increase will notify the kvm page fault that
 377          * the page that is going to be mapped in the spte could have
 378          * been freed.
 379          */
 380         kvm->mmu_notifier_seq++;
 381         smp_wmb();
 382         /*
 383          * The above sequence increase must be visible before the
 384          * below count decrease, which is ensured by the smp_wmb above
 385          * in conjunction with the smp_rmb in mmu_notifier_retry().
 386          */
 387         kvm->mmu_notifier_count--;
 388         spin_unlock(&kvm->mmu_lock);
 389
 390         BUG_ON(kvm->mmu_notifier_count < 0);
 391 }
 392
 393 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
 394                                               struct mm_struct *mm,
 395                                               unsigned long start,
 396                                               unsigned long end)
 397 {
 398         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 399         int young, idx;
 400
 401         idx = srcu_read_lock(&kvm->srcu);
 402         spin_lock(&kvm->mmu_lock);
 403
 404         young = kvm_age_hva(kvm, start, end);
 405         if (young)
 406                 kvm_flush_remote_tlbs(kvm);
 407
 408         spin_unlock(&kvm->mmu_lock);
 409         srcu_read_unlock(&kvm->srcu, idx);
 410
 411         return young;
 412 }
 413
 414 static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
 415                                         struct mm_struct *mm,
 416                                         unsigned long start,
 417                                         unsigned long end)
 418 {
 419         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 420         int young, idx;
 421
 422         idx = srcu_read_lock(&kvm->srcu);
 423         spin_lock(&kvm->mmu_lock);
 424         /*
 425          * Even though we do not flush TLB, this will still adversely
 426          * affect performance on pre-Haswell Intel EPT, where there is
 427          * no EPT Access Bit to clear so that we have to tear down EPT
 428          * tables instead. If we find this unacceptable, we can always
 429          * add a parameter to kvm_age_hva so that it effectively doesn't
 430          * do anything on clear_young.
 431          *
 432          * Also note that currently we never issue secondary TLB flushes
 433          * from clear_young, leaving this job up to the regular system
 434          * cadence. If we find this inaccurate, we might come up with a
 435          * more sophisticated heuristic later.
 436          */
 437         young = kvm_age_hva(kvm, start, end);
 438         spin_unlock(&kvm->mmu_lock);
 439         srcu_read_unlock(&kvm->srcu, idx);
 440
 441         return young;
 442 }
 443
 444 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
 445                                        struct mm_struct *mm,
 446                                        unsigned long address)
 447 {
 448         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 449         int young, idx;
 450
 451         idx = srcu_read_lock(&kvm->srcu);
 452         spin_lock(&kvm->mmu_lock);
 453         young = kvm_test_age_hva(kvm, address);
 454         spin_unlock(&kvm->mmu_lock);
 455         srcu_read_unlock(&kvm->srcu, idx);
 456
 457         return young;
 458 }
 459
 460 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
 461                                      struct mm_struct *mm)
 462 {
 463         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 464         int idx;
 465
 466         idx = srcu_read_lock(&kvm->srcu);
 467         kvm_arch_flush_shadow_all(kvm);
 468         srcu_read_unlock(&kvm->srcu, idx);
 469 }
 470
 471 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 472         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
 473         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
 474         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
 475         .clear_young            = kvm_mmu_notifier_clear_young,
 476         .test_young             = kvm_mmu_notifier_test_young,
 477         .change_pte             = kvm_mmu_notifier_change_pte,
 478         .release                = kvm_mmu_notifier_release,
 479 };
 480
 481 static int kvm_init_mmu_notifier(struct kvm *kvm)
 482 {
 483         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
 484         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
 485 }
 486
 487 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
 488
 489 static int kvm_init_mmu_notifier(struct kvm *kvm)
 490 {
 491         return 0;
 492 }
 493
 494 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
 495
 496 static struct kvm_memslots *kvm_alloc_memslots(void)
 497 {
 498         int i;
 499         struct kvm_memslots *slots;
 500
 501         slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 502         if (!slots)
 503                 return NULL;
 504
 505         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
 506                 slots->id_to_index[i] = slots->memslots[i].id = i;
 507
 508         return slots;
 509 }
 510
 511 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
 512 {
 513         if (!memslot->dirty_bitmap)
 514                 return;
 515
 516         kvfree(memslot->dirty_bitmap);
 517         memslot->dirty_bitmap = NULL;
 518 }
 519
 520 /*
 521  * Free any memory in @free but not in @dont.
 522  */
 523 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
 524                               struct kvm_memory_slot *dont)
 525 {
 526         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 527                 kvm_destroy_dirty_bitmap(free);
 528
 529         kvm_arch_free_memslot(kvm, free, dont);
 530
 531         free->npages = 0;
 532 }
 533
 534 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
 535 {
 536         struct kvm_memory_slot *memslot;
 537
 538         if (!slots)
 539                 return;
 540
 541         kvm_for_each_memslot(memslot, slots)
 542                 kvm_free_memslot(kvm, memslot, NULL);
 543
 544         kvfree(slots);
 545 }
 546
 547 static void kvm_destroy_vm_debugfs(struct kvm *kvm)
 548 {
 549         int i;
 550
 551         if (!kvm->debugfs_dentry)
 552                 return;
 553
 554         debugfs_remove_recursive(kvm->debugfs_dentry);
 555
 556         if (kvm->debugfs_stat_data) {
 557                 for (i = 0; i < kvm_debugfs_num_entries; i++)
 558                         kfree(kvm->debugfs_stat_data[i]);
 559                 kfree(kvm->debugfs_stat_data);
 560         }
 561 }
 562
 563 static int kvm_create_vm_debugfs(struct kvm *kvm, int fd)
 564 {
 565         char dir_name[ITOA_MAX_LEN * 2];
 566         struct kvm_stat_data *stat_data;
 567         struct kvm_stats_debugfs_item *p;
 568
 569         if (!debugfs_initialized())
 570                 return 0;
 571
 572         snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd);
 573         kvm->debugfs_dentry = debugfs_create_dir(dir_name,
 574                                                  kvm_debugfs_dir);
 575         if (!kvm->debugfs_dentry)
 576                 return -ENOMEM;
 577
 578         kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries,
 579                                          sizeof(*kvm->debugfs_stat_data),
 580                                          GFP_KERNEL);
 581         if (!kvm->debugfs_stat_data)
 582                 return -ENOMEM;
 583
 584         for (p = debugfs_entries; p->name; p++) {
 585                 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL);
 586                 if (!stat_data)
 587                         return -ENOMEM;
 588
 589                 stat_data->kvm = kvm;
 590                 stat_data->offset = p->offset;
 591                 kvm->debugfs_stat_data[p - debugfs_entries] = stat_data;
 592                 if (!debugfs_create_file(p->name, 0644,
 593                                          kvm->debugfs_dentry,
 594                                          stat_data,
 595                                          stat_fops_per_vm[p->kind]))
 596                         return -ENOMEM;
 597         }
 598         return 0;
 599 }
 600
 601 static struct kvm *kvm_create_vm(unsigned long type)
 602 {
 603         int r, i;
 604         struct kvm *kvm = kvm_arch_alloc_vm();
 605
 606         if (!kvm)
 607                 return ERR_PTR(-ENOMEM);
 608
 609         spin_lock_init(&kvm->mmu_lock);
 610         mmgrab(current->mm);
 611         kvm->mm = current->mm;
 612         kvm_eventfd_init(kvm);
 613         mutex_init(&kvm->lock);
 614         mutex_init(&kvm->irq_lock);
 615         mutex_init(&kvm->slots_lock);
 616         refcount_set(&kvm->users_count, 1);
 617         INIT_LIST_HEAD(&kvm->devices);
 618
 619         r = kvm_arch_init_vm(kvm, type);
 620         if (r)
 621                 goto out_err_no_disable;
 622
 623         r = hardware_enable_all();
 624         if (r)
 625                 goto out_err_no_disable;
 626
 627 #ifdef CONFIG_HAVE_KVM_IRQFD
 628         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
 629 #endif
 630
 631         BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
 632
 633         r = -ENOMEM;
 634         for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
 635                 struct kvm_memslots *slots = kvm_alloc_memslots();
 636                 if (!slots)
 637                         goto out_err_no_srcu;
 638                 /*
 639                  * Generations must be different for each address space.
 640                  * Init kvm generation close to the maximum to easily test the
 641                  * code of handling generation number wrap-around.
 642                  */
 643                 slots->generation = i * 2 - 150;
 644                 rcu_assign_pointer(kvm->memslots[i], slots);
 645         }
 646
 647         if (init_srcu_struct(&kvm->srcu))
 648                 goto out_err_no_srcu;
 649         if (init_srcu_struct(&kvm->irq_srcu))
 650                 goto out_err_no_irq_srcu;
 651         for (i = 0; i < KVM_NR_BUSES; i++) {
 652                 rcu_assign_pointer(kvm->buses[i],
 653                         kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL));
 654                 if (!kvm->buses[i])
 655                         goto out_err;
 656         }
 657
 658         r = kvm_init_mmu_notifier(kvm);
 659         if (r)
 660                 goto out_err;
 661
 662         spin_lock(&kvm_lock);
 663         list_add(&kvm->vm_list, &vm_list);
 664         spin_unlock(&kvm_lock);
 665
 666         preempt_notifier_inc();
 667
 668         return kvm;
 669
 670 out_err:
 671         cleanup_srcu_struct(&kvm->irq_srcu);
 672 out_err_no_irq_srcu:
 673         cleanup_srcu_struct(&kvm->srcu);
 674 out_err_no_srcu:
 675         hardware_disable_all();
 676 out_err_no_disable:
 677         for (i = 0; i < KVM_NR_BUSES; i++)
 678                 kfree(kvm_get_bus(kvm, i));
 679         for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
 680                 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
 681         kvm_arch_free_vm(kvm);
 682         mmdrop(current->mm);
 683         return ERR_PTR(r);
 684 }
 685
 686 static void kvm_destroy_devices(struct kvm *kvm)
 687 {
 688         struct kvm_device *dev, *tmp;
 689
 690         /*
 691          * We do not need to take the kvm->lock here, because nobody else
 692          * has a reference to the struct kvm at this point and therefore
 693          * cannot access the devices list anyhow.
 694          */
 695         list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
 696                 list_del(&dev->vm_node);
 697                 dev->ops->destroy(dev);
 698         }
 699 }
 700
 701 static void kvm_destroy_vm(struct kvm *kvm)
 702 {
 703         int i;
 704         struct mm_struct *mm = kvm->mm;
 705
 706         kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm);
 707         kvm_destroy_vm_debugfs(kvm);
 708         kvm_arch_sync_events(kvm);
 709         spin_lock(&kvm_lock);
 710         list_del(&kvm->vm_list);
 711         spin_unlock(&kvm_lock);
 712         kvm_free_irq_routing(kvm);
 713         for (i = 0; i < KVM_NR_BUSES; i++) {
 714                 struct kvm_io_bus *bus = kvm_get_bus(kvm, i);
 715
 716                 if (bus)
 717                         kvm_io_bus_destroy(bus);
 718                 kvm->buses[i] = NULL;
 719         }
 720         kvm_coalesced_mmio_free(kvm);
 721 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 722         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
 723 #else
 724         kvm_arch_flush_shadow_all(kvm);
 725 #endif
 726         kvm_arch_destroy_vm(kvm);
 727         kvm_destroy_devices(kvm);
 728         for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
 729                 kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
 730         cleanup_srcu_struct(&kvm->irq_srcu);
 731         cleanup_srcu_struct(&kvm->srcu);
 732         kvm_arch_free_vm(kvm);
 733         preempt_notifier_dec();
 734         hardware_disable_all();
 735         mmdrop(mm);
 736 }
 737
 738 void kvm_get_kvm(struct kvm *kvm)
 739 {
 740         refcount_inc(&kvm->users_count);
 741 }
 742 EXPORT_SYMBOL_GPL(kvm_get_kvm);
 743
 744 void kvm_put_kvm(struct kvm *kvm)
 745 {
 746         if (refcount_dec_and_test(&kvm->users_count))
 747                 kvm_destroy_vm(kvm);
 748 }
 749 EXPORT_SYMBOL_GPL(kvm_put_kvm);
 750
 751
 752 static int kvm_vm_release(struct inode *inode, struct file *filp)
 753 {
 754         struct kvm *kvm = filp->private_data;
 755
 756         kvm_irqfd_release(kvm);
 757
 758         kvm_put_kvm(kvm);
 759         return 0;
 760 }
 761
 762 /*
 763  * Allocation size is twice as large as the actual dirty bitmap size.
 764  * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
 765  */
 766 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
 767 {
 768         unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
 769
 770         memslot->dirty_bitmap = kvzalloc(dirty_bytes, GFP_KERNEL);
 771         if (!memslot->dirty_bitmap)
 772                 return -ENOMEM;
 773
 774         return 0;
 775 }
 776
 777 /*
 778  * Insert memslot and re-sort memslots based on their GFN,
 779  * so binary search could be used to lookup GFN.
 780  * Sorting algorithm takes advantage of having initially
 781  * sorted array and known changed memslot position.
 782  */
 783 static void update_memslots(struct kvm_memslots *slots,
 784                             struct kvm_memory_slot *new)
 785 {
 786         int id = new->id;
 787         int i = slots->id_to_index[id];
 788         struct kvm_memory_slot *mslots = slots->memslots;
 789
 790         WARN_ON(mslots[i].id != id);
 791         if (!new->npages) {
 792                 WARN_ON(!mslots[i].npages);
 793                 if (mslots[i].npages)
 794                         slots->used_slots--;
 795         } else {
 796                 if (!mslots[i].npages)
 797                         slots->used_slots++;
 798         }
 799
 800         while (i < KVM_MEM_SLOTS_NUM - 1 &&
 801                new->base_gfn <= mslots[i + 1].base_gfn) {
 802                 if (!mslots[i + 1].npages)
 803                         break;
 804                 mslots[i] = mslots[i + 1];
 805                 slots->id_to_index[mslots[i].id] = i;
 806                 i++;
 807         }
 808
 809         /*
 810          * The ">=" is needed when creating a slot with base_gfn == 0,
 811          * so that it moves before all those with base_gfn == npages == 0.
 812          *
 813          * On the other hand, if new->npages is zero, the above loop has
 814          * already left i pointing to the beginning of the empty part of
 815          * mslots, and the ">=" would move the hole backwards in this
 816          * case---which is wrong.  So skip the loop when deleting a slot.
 817          */
 818         if (new->npages) {
 819                 while (i > 0 &&
 820                        new->base_gfn >= mslots[i - 1].base_gfn) {
 821                         mslots[i] = mslots[i - 1];
 822                         slots->id_to_index[mslots[i].id] = i;
 823                         i--;
 824                 }
 825         } else
 826                 WARN_ON_ONCE(i != slots->used_slots);
 827
 828         mslots[i] = *new;
 829         slots->id_to_index[mslots[i].id] = i;
 830 }
 831
 832 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
 833 {
 834         u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
 835
 836 #ifdef __KVM_HAVE_READONLY_MEM
 837         valid_flags |= KVM_MEM_READONLY;
 838 #endif
 839
 840         if (mem->flags & ~valid_flags)
 841                 return -EINVAL;
 842
 843         return 0;
 844 }
 845
 846 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
 847                 int as_id, struct kvm_memslots *slots)
 848 {
 849         struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id);
 850
 851         /*
 852          * Set the low bit in the generation, which disables SPTE caching
 853          * until the end of synchronize_srcu_expedited.
 854          */
 855         WARN_ON(old_memslots->generation & 1);
 856         slots->generation = old_memslots->generation + 1;
 857
 858         rcu_assign_pointer(kvm->memslots[as_id], slots);
 859         synchronize_srcu_expedited(&kvm->srcu);
 860
 861         /*
 862          * Increment the new memslot generation a second time. This prevents
 863          * vm exits that race with memslot updates from caching a memslot
 864          * generation that will (potentially) be valid forever.
 865          *
 866          * Generations must be unique even across address spaces.  We do not need
 867          * a global counter for that, instead the generation space is evenly split
 868          * across address spaces.  For example, with two address spaces, address
 869          * space 0 will use generations 0, 4, 8, ... while * address space 1 will
 870          * use generations 2, 6, 10, 14, ...
 871          */
 872         slots->generation += KVM_ADDRESS_SPACE_NUM * 2 - 1;
 873
 874         kvm_arch_memslots_updated(kvm, slots);
 875
 876         return old_memslots;
 877 }
 878
 879 /*
 880  * Allocate some memory and give it an address in the guest physical address
 881  * space.
 882  *
 883  * Discontiguous memory is allowed, mostly for framebuffers.
 884  *
 885  * Must be called holding kvm->slots_lock for write.
 886  */
 887 int __kvm_set_memory_region(struct kvm *kvm,
 888                             const struct kvm_userspace_memory_region *mem)
 889 {
 890         int r;
 891         gfn_t base_gfn;
 892         unsigned long npages;
 893         struct kvm_memory_slot *slot;
 894         struct kvm_memory_slot old, new;
 895         struct kvm_memslots *slots = NULL, *old_memslots;
 896         int as_id, id;
 897         enum kvm_mr_change change;
 898
 899         r = check_memory_region_flags(mem);
 900         if (r)
 901                 goto out;
 902
 903         r = -EINVAL;
 904         as_id = mem->slot >> 16;
 905         id = (u16)mem->slot;
 906
 907         /* General sanity checks */
 908         if (mem->memory_size & (PAGE_SIZE - 1))
 909                 goto out;
 910         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 911                 goto out;
 912         /* We can read the guest memory with __xxx_user() later on. */
 913         if ((id < KVM_USER_MEM_SLOTS) &&
 914             ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
 915              !access_ok(VERIFY_WRITE,
 916                         (void __user *)(unsigned long)mem->userspace_addr,
 917                         mem->memory_size)))
 918                 goto out;
 919         if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
 920                 goto out;
 921         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 922                 goto out;
 923
 924         slot = id_to_memslot(__kvm_memslots(kvm, as_id), id);
 925         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 926         npages = mem->memory_size >> PAGE_SHIFT;
 927
 928         if (npages > KVM_MEM_MAX_NR_PAGES)
 929                 goto out;
 930
 931         new = old = *slot;
 932
 933         new.id = id;
 934         new.base_gfn = base_gfn;
 935         new.npages = npages;
 936         new.flags = mem->flags;
 937
 938         if (npages) {
 939                 if (!old.npages)
 940                         change = KVM_MR_CREATE;
 941                 else { /* Modify an existing slot. */
 942                         if ((mem->userspace_addr != old.userspace_addr) ||
 943                             (npages != old.npages) ||
 944                             ((new.flags ^ old.flags) & KVM_MEM_READONLY))
 945                                 goto out;
 946
 947                         if (base_gfn != old.base_gfn)
 948                                 change = KVM_MR_MOVE;
 949                         else if (new.flags != old.flags)
 950                                 change = KVM_MR_FLAGS_ONLY;
 951                         else { /* Nothing to change. */
 952                                 r = 0;
 953                                 goto out;
 954                         }
 955                 }
 956         } else {
 957                 if (!old.npages)
 958                         goto out;
 959
 960                 change = KVM_MR_DELETE;
 961                 new.base_gfn = 0;
 962                 new.flags = 0;
 963         }
 964
 965         if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
 966                 /* Check for overlaps */
 967                 r = -EEXIST;
 968                 kvm_for_each_memslot(slot, __kvm_memslots(kvm, as_id)) {
 969                         if ((slot->id >= KVM_USER_MEM_SLOTS) ||
 970                             (slot->id == id))
 971                                 continue;
 972                         if (!((base_gfn + npages <= slot->base_gfn) ||
 973                               (base_gfn >= slot->base_gfn + slot->npages)))
 974                                 goto out;
 975                 }
 976         }
 977
 978         /* Free page dirty bitmap if unneeded */
 979         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 980                 new.dirty_bitmap = NULL;
 981
 982         r = -ENOMEM;
 983         if (change == KVM_MR_CREATE) {
 984                 new.userspace_addr = mem->userspace_addr;
 985
 986                 if (kvm_arch_create_memslot(kvm, &new, npages))
 987                         goto out_free;
 988         }
 989
 990         /* Allocate page dirty bitmap if needed */
 991         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 992                 if (kvm_create_dirty_bitmap(&new) < 0)
 993                         goto out_free;
 994         }
 995
 996         slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 997         if (!slots)
 998                 goto out_free;
 999         memcpy(slots, __kvm_memslots(kvm, as_id), sizeof(struct kvm_memslots));
1000
1001         if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
1002                 slot = id_to_memslot(slots, id);
1003                 slot->flags |= KVM_MEMSLOT_INVALID;
1004
1005                 old_memslots = install_new_memslots(kvm, as_id, slots);
1006
1007                 /* From this point no new shadow pages pointing to a deleted,
1008                  * or moved, memslot will be created.
1009                  *
1010                  * validation of sp->gfn happens in:
1011                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1012                  *      - kvm_is_visible_gfn (mmu_check_roots)
1013                  */
1014                 kvm_arch_flush_shadow_memslot(kvm, slot);
1015
1016                 /*
1017                  * We can re-use the old_memslots from above, the only difference
1018                  * from the currently installed memslots is the invalid flag.  This
1019                  * will get overwritten by update_memslots anyway.
1020                  */
1021                 slots = old_memslots;
1022         }
1023
1024         r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
1025         if (r)
1026                 goto out_slots;
1027
1028         /* actual memory is freed via old in kvm_free_memslot below */
1029         if (change == KVM_MR_DELETE) {
1030                 new.dirty_bitmap = NULL;
1031                 memset(&new.arch, 0, sizeof(new.arch));
1032         }
1033
1034         update_memslots(slots, &new);
1035         old_memslots = install_new_memslots(kvm, as_id, slots);
1036
1037         kvm_arch_commit_memory_region(kvm, mem, &old, &new, change);
1038
1039         kvm_free_memslot(kvm, &old, &new);
1040         kvfree(old_memslots);
1041         return 0;
1042
1043 out_slots:
1044         kvfree(slots);
1045 out_free:
1046         kvm_free_memslot(kvm, &new, &old);
1047 out:
1048         return r;
1049 }
1050 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1051
1052 int kvm_set_memory_region(struct kvm *kvm,
1053                           const struct kvm_userspace_memory_region *mem)
1054 {
1055         int r;
1056
1057         mutex_lock(&kvm->slots_lock);
1058         r = __kvm_set_memory_region(kvm, mem);
1059         mutex_unlock(&kvm->slots_lock);
1060         return r;
1061 }
1062 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1063
1064 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1065                                           struct kvm_userspace_memory_region *mem)
1066 {
1067         if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1068                 return -EINVAL;
1069
1070         return kvm_set_memory_region(kvm, mem);
1071 }
1072
1073 int kvm_get_dirty_log(struct kvm *kvm,
1074                         struct kvm_dirty_log *log, int *is_dirty)
1075 {
1076         struct kvm_memslots *slots;
1077         struct kvm_memory_slot *memslot;
1078         int i, as_id, id;
1079         unsigned long n;
1080         unsigned long any = 0;
1081
1082         as_id = log->slot >> 16;
1083         id = (u16)log->slot;
1084         if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1085                 return -EINVAL;
1086
1087         slots = __kvm_memslots(kvm, as_id);
1088         memslot = id_to_memslot(slots, id);
1089         if (!memslot->dirty_bitmap)
1090                 return -ENOENT;
1091
1092         n = kvm_dirty_bitmap_bytes(memslot);
1093
1094         for (i = 0; !any && i < n/sizeof(long); ++i)
1095                 any = memslot->dirty_bitmap[i];
1096
1097         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1098                 return -EFAULT;
1099
1100         if (any)
1101                 *is_dirty = 1;
1102         return 0;
1103 }
1104 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1105
1106 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1107 /**
1108  * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages
1109  *      are dirty write protect them for next write.
1110  * @kvm:        pointer to kvm instance
1111  * @log:        slot id and address to which we copy the log
1112  * @is_dirty:   flag set if any page is dirty
1113  *
1114  * We need to keep it in mind that VCPU threads can write to the bitmap
1115  * concurrently. So, to avoid losing track of dirty pages we keep the
1116  * following order:
1117  *
1118  *    1. Take a snapshot of the bit and clear it if needed.
1119  *    2. Write protect the corresponding page.
1120  *    3. Copy the snapshot to the userspace.
1121  *    4. Upon return caller flushes TLB's if needed.
1122  *
1123  * Between 2 and 4, the guest may write to the page using the remaining TLB
1124  * entry.  This is not a problem because the page is reported dirty using
1125  * the snapshot taken before and step 4 ensures that writes done after
1126  * exiting to userspace will be logged for the next call.
1127  *
1128  */
1129 int kvm_get_dirty_log_protect(struct kvm *kvm,
1130                         struct kvm_dirty_log *log, bool *is_dirty)
1131 {
1132         struct kvm_memslots *slots;
1133         struct kvm_memory_slot *memslot;
1134         int i, as_id, id;
1135         unsigned long n;
1136         unsigned long *dirty_bitmap;
1137         unsigned long *dirty_bitmap_buffer;
1138
1139         as_id = log->slot >> 16;
1140         id = (u16)log->slot;
1141         if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1142                 return -EINVAL;
1143
1144         slots = __kvm_memslots(kvm, as_id);
1145         memslot = id_to_memslot(slots, id);
1146
1147         dirty_bitmap = memslot->dirty_bitmap;
1148         if (!dirty_bitmap)
1149                 return -ENOENT;
1150
1151         n = kvm_dirty_bitmap_bytes(memslot);
1152
1153         dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long);
1154         memset(dirty_bitmap_buffer, 0, n);
1155
1156         spin_lock(&kvm->mmu_lock);
1157         *is_dirty = false;
1158         for (i = 0; i < n / sizeof(long); i++) {
1159                 unsigned long mask;
1160                 gfn_t offset;
1161
1162                 if (!dirty_bitmap[i])
1163                         continue;
1164
1165                 *is_dirty = true;
1166
1167                 mask = xchg(&dirty_bitmap[i], 0);
1168                 dirty_bitmap_buffer[i] = mask;
1169
1170                 if (mask) {
1171                         offset = i * BITS_PER_LONG;
1172                         kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1173                                                                 offset, mask);
1174                 }
1175         }
1176
1177         spin_unlock(&kvm->mmu_lock);
1178         if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1179                 return -EFAULT;
1180         return 0;
1181 }
1182 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect);
1183 #endif
1184
1185 bool kvm_largepages_enabled(void)
1186 {
1187         return largepages_enabled;
1188 }
1189
1190 void kvm_disable_largepages(void)
1191 {
1192         largepages_enabled = false;
1193 }
1194 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1195
1196 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1197 {
1198         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1199 }
1200 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1201
1202 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
1203 {
1204         return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn);
1205 }
1206
1207 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1208 {
1209         struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1210
1211         if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1212               memslot->flags & KVM_MEMSLOT_INVALID)
1213                 return false;
1214
1215         return true;
1216 }
1217 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1218
1219 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1220 {
1221         struct vm_area_struct *vma;
1222         unsigned long addr, size;
1223
1224         size = PAGE_SIZE;
1225
1226         addr = gfn_to_hva(kvm, gfn);
1227         if (kvm_is_error_hva(addr))
1228                 return PAGE_SIZE;
1229
1230         down_read(&current->mm->mmap_sem);
1231         vma = find_vma(current->mm, addr);
1232         if (!vma)
1233                 goto out;
1234
1235         size = vma_kernel_pagesize(vma);
1236
1237 out:
1238         up_read(&current->mm->mmap_sem);
1239
1240         return size;
1241 }
1242
1243 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1244 {
1245         return slot->flags & KVM_MEM_READONLY;
1246 }
1247
1248 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1249                                        gfn_t *nr_pages, bool write)
1250 {
1251         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1252                 return KVM_HVA_ERR_BAD;
1253
1254         if (memslot_is_readonly(slot) && write)
1255                 return KVM_HVA_ERR_RO_BAD;
1256
1257         if (nr_pages)
1258                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1259
1260         return __gfn_to_hva_memslot(slot, gfn);
1261 }
1262
1263 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1264                                      gfn_t *nr_pages)
1265 {
1266         return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1267 }
1268
1269 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1270                                         gfn_t gfn)
1271 {
1272         return gfn_to_hva_many(slot, gfn, NULL);
1273 }
1274 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1275
1276 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1277 {
1278         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1279 }
1280 EXPORT_SYMBOL_GPL(gfn_to_hva);
1281
1282 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
1283 {
1284         return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
1285 }
1286 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
1287
1288 /*
1289  * If writable is set to false, the hva returned by this function is only
1290  * allowed to be read.
1291  */
1292 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1293                                       gfn_t gfn, bool *writable)
1294 {
1295         unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1296
1297         if (!kvm_is_error_hva(hva) && writable)
1298                 *writable = !memslot_is_readonly(slot);
1299
1300         return hva;
1301 }
1302
1303 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1304 {
1305         struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1306
1307         return gfn_to_hva_memslot_prot(slot, gfn, writable);
1308 }
1309
1310 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
1311 {
1312         struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1313
1314         return gfn_to_hva_memslot_prot(slot, gfn, writable);
1315 }
1316
1317 static int get_user_page_nowait(unsigned long start, int write,
1318                 struct page **page)
1319 {
1320         int flags = FOLL_NOWAIT | FOLL_HWPOISON;
1321
1322         if (write)
1323                 flags |= FOLL_WRITE;
1324
1325         return get_user_pages(start, 1, flags, page, NULL);
1326 }
1327
1328 static inline int check_user_page_hwpoison(unsigned long addr)
1329 {
1330         int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
1331
1332         rc = get_user_pages(addr, 1, flags, NULL, NULL);
1333         return rc == -EHWPOISON;
1334 }
1335
1336 /*
1337  * The atomic path to get the writable pfn which will be stored in @pfn,
1338  * true indicates success, otherwise false is returned.
1339  */
1340 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1341                             bool write_fault, bool *writable, kvm_pfn_t *pfn)
1342 {
1343         struct page *page[1];
1344         int npages;
1345
1346         if (!(async || atomic))
1347                 return false;
1348
1349         /*
1350          * Fast pin a writable pfn only if it is a write fault request
1351          * or the caller allows to map a writable pfn for a read fault
1352          * request.
1353          */
1354         if (!(write_fault || writable))
1355                 return false;
1356
1357         npages = __get_user_pages_fast(addr, 1, 1, page);
1358         if (npages == 1) {
1359                 *pfn = page_to_pfn(page[0]);
1360
1361                 if (writable)
1362                         *writable = true;
1363                 return true;
1364         }
1365
1366         return false;
1367 }
1368
1369 /*
1370  * The slow path to get the pfn of the specified host virtual address,
1371  * 1 indicates success, -errno is returned if error is detected.
1372  */
1373 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1374                            bool *writable, kvm_pfn_t *pfn)
1375 {
1376         struct page *page[1];
1377         int npages = 0;
1378
1379         might_sleep();
1380
1381         if (writable)
1382                 *writable = write_fault;
1383
1384         if (async) {
1385                 down_read(&current->mm->mmap_sem);
1386                 npages = get_user_page_nowait(addr, write_fault, page);
1387                 up_read(&current->mm->mmap_sem);
1388         } else {
1389                 unsigned int flags = FOLL_HWPOISON;
1390
1391                 if (write_fault)
1392                         flags |= FOLL_WRITE;
1393
1394                 npages = get_user_pages_unlocked(addr, 1, page, flags);
1395         }
1396         if (npages != 1)
1397                 return npages;
1398
1399         /* map read fault as writable if possible */
1400         if (unlikely(!write_fault) && writable) {
1401                 struct page *wpage[1];
1402
1403                 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1404                 if (npages == 1) {
1405                         *writable = true;
1406                         put_page(page[0]);
1407                         page[0] = wpage[0];
1408                 }
1409
1410                 npages = 1;
1411         }
1412         *pfn = page_to_pfn(page[0]);
1413         return npages;
1414 }
1415
1416 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1417 {
1418         if (unlikely(!(vma->vm_flags & VM_READ)))
1419                 return false;
1420
1421         if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1422                 return false;
1423
1424         return true;
1425 }
1426
1427 static int hva_to_pfn_remapped(struct vm_area_struct *vma,
1428                                unsigned long addr, bool *async,
1429                                bool write_fault, kvm_pfn_t *p_pfn)
1430 {
1431         unsigned long pfn;
1432         int r;
1433
1434         r = follow_pfn(vma, addr, &pfn);
1435         if (r) {
1436                 /*
1437                  * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
1438                  * not call the fault handler, so do it here.
1439                  */
1440                 bool unlocked = false;
1441                 r = fixup_user_fault(current, current->mm, addr,
1442                                      (write_fault ? FAULT_FLAG_WRITE : 0),
1443                                      &unlocked);
1444                 if (unlocked)
1445                         return -EAGAIN;
1446                 if (r)
1447                         return r;
1448
1449                 r = follow_pfn(vma, addr, &pfn);
1450                 if (r)
1451                         return r;
1452
1453         }
1454
1455
1456         /*
1457          * Get a reference here because callers of *hva_to_pfn* and
1458          * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
1459          * returned pfn.  This is only needed if the VMA has VM_MIXEDMAP
1460          * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will
1461          * simply do nothing for reserved pfns.
1462          *
1463          * Whoever called remap_pfn_range is also going to call e.g.
1464          * unmap_mapping_range before the underlying pages are freed,
1465          * causing a call to our MMU notifier.
1466          */
1467         kvm_get_pfn(pfn);
1468
1469         *p_pfn = pfn;
1470         return 0;
1471 }
1472
1473 /*
1474  * Pin guest page in memory and return its pfn.
1475  * @addr: host virtual address which maps memory to the guest
1476  * @atomic: whether this function can sleep
1477  * @async: whether this function need to wait IO complete if the
1478  *         host page is not in the memory
1479  * @write_fault: whether we should get a writable host page
1480  * @writable: whether it allows to map a writable host page for !@write_fault
1481  *
1482  * The function will map a writable host page for these two cases:
1483  * 1): @write_fault = true
1484  * 2): @write_fault = false && @writable, @writable will tell the caller
1485  *     whether the mapping is writable.
1486  */
1487 static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1488                         bool write_fault, bool *writable)
1489 {
1490         struct vm_area_struct *vma;
1491         kvm_pfn_t pfn = 0;
1492         int npages, r;
1493
1494         /* we can do it either atomically or asynchronously, not both */
1495         BUG_ON(atomic && async);
1496
1497         if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1498                 return pfn;
1499
1500         if (atomic)
1501                 return KVM_PFN_ERR_FAULT;
1502
1503         npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1504         if (npages == 1)
1505                 return pfn;
1506
1507         down_read(&current->mm->mmap_sem);
1508         if (npages == -EHWPOISON ||
1509               (!async && check_user_page_hwpoison(addr))) {
1510                 pfn = KVM_PFN_ERR_HWPOISON;
1511                 goto exit;
1512         }
1513
1514 retry:
1515         vma = find_vma_intersection(current->mm, addr, addr + 1);
1516
1517         if (vma == NULL)
1518                 pfn = KVM_PFN_ERR_FAULT;
1519         else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
1520                 r = hva_to_pfn_remapped(vma, addr, async, write_fault, &pfn);
1521                 if (r == -EAGAIN)
1522                         goto retry;
1523                 if (r < 0)
1524                         pfn = KVM_PFN_ERR_FAULT;
1525         } else {
1526                 if (async && vma_is_valid(vma, write_fault))
1527                         *async = true;
1528                 pfn = KVM_PFN_ERR_FAULT;
1529         }
1530 exit:
1531         up_read(&current->mm->mmap_sem);
1532         return pfn;
1533 }
1534
1535 kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn,
1536                                bool atomic, bool *async, bool write_fault,
1537                                bool *writable)
1538 {
1539         unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1540
1541         if (addr == KVM_HVA_ERR_RO_BAD) {
1542                 if (writable)
1543                         *writable = false;
1544                 return KVM_PFN_ERR_RO_FAULT;
1545         }
1546
1547         if (kvm_is_error_hva(addr)) {
1548                 if (writable)
1549                         *writable = false;
1550                 return KVM_PFN_NOSLOT;
1551         }
1552
1553         /* Do not map writable pfn in the readonly memslot. */
1554         if (writable && memslot_is_readonly(slot)) {
1555                 *writable = false;
1556                 writable = NULL;
1557         }
1558
1559         return hva_to_pfn(addr, atomic, async, write_fault,
1560                           writable);
1561 }
1562 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
1563
1564 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1565                       bool *writable)
1566 {
1567         return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
1568                                     write_fault, writable);
1569 }
1570 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1571
1572 kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1573 {
1574         return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1575 }
1576 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
1577
1578 kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1579 {
1580         return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1581 }
1582 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1583
1584 kvm_pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1585 {
1586         return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm, gfn), gfn);
1587 }
1588 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1589
1590 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
1591 {
1592         return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1593 }
1594 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
1595
1596 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1597 {
1598         return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
1599 }
1600 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1601
1602 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
1603 {
1604         return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1605 }
1606 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
1607
1608 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1609                             struct page **pages, int nr_pages)
1610 {
1611         unsigned long addr;
1612         gfn_t entry = 0;
1613
1614         addr = gfn_to_hva_many(slot, gfn, &entry);
1615         if (kvm_is_error_hva(addr))
1616                 return -1;
1617
1618         if (entry < nr_pages)
1619                 return 0;
1620
1621         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1622 }
1623 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1624
1625 static struct page *kvm_pfn_to_page(kvm_pfn_t pfn)
1626 {
1627         if (is_error_noslot_pfn(pfn))
1628                 return KVM_ERR_PTR_BAD_PAGE;
1629
1630         if (kvm_is_reserved_pfn(pfn)) {
1631                 WARN_ON(1);
1632                 return KVM_ERR_PTR_BAD_PAGE;
1633         }
1634
1635         return pfn_to_page(pfn);
1636 }
1637
1638 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1639 {
1640         kvm_pfn_t pfn;
1641
1642         pfn = gfn_to_pfn(kvm, gfn);
1643
1644         return kvm_pfn_to_page(pfn);
1645 }
1646 EXPORT_SYMBOL_GPL(gfn_to_page);
1647
1648 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
1649 {
1650         kvm_pfn_t pfn;
1651
1652         pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
1653
1654         return kvm_pfn_to_page(pfn);
1655 }
1656 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
1657
1658 void kvm_release_page_clean(struct page *page)
1659 {
1660         WARN_ON(is_error_page(page));
1661
1662         kvm_release_pfn_clean(page_to_pfn(page));
1663 }
1664 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1665
1666 void kvm_release_pfn_clean(kvm_pfn_t pfn)
1667 {
1668         if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
1669                 put_page(pfn_to_page(pfn));
1670 }
1671 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1672
1673 void kvm_release_page_dirty(struct page *page)
1674 {
1675         WARN_ON(is_error_page(page));
1676
1677         kvm_release_pfn_dirty(page_to_pfn(page));
1678 }
1679 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1680
1681 static void kvm_release_pfn_dirty(kvm_pfn_t pfn)
1682 {
1683         kvm_set_pfn_dirty(pfn);
1684         kvm_release_pfn_clean(pfn);
1685 }
1686
1687 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
1688 {
1689         if (!kvm_is_reserved_pfn(pfn)) {
1690                 struct page *page = pfn_to_page(pfn);
1691
1692                 if (!PageReserved(page))
1693                         SetPageDirty(page);
1694         }
1695 }
1696 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1697
1698 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
1699 {
1700         if (!kvm_is_reserved_pfn(pfn))
1701                 mark_page_accessed(pfn_to_page(pfn));
1702 }
1703 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1704
1705 void kvm_get_pfn(kvm_pfn_t pfn)
1706 {
1707         if (!kvm_is_reserved_pfn(pfn))
1708                 get_page(pfn_to_page(pfn));
1709 }
1710 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1711
1712 static int next_segment(unsigned long len, int offset)
1713 {
1714         if (len > PAGE_SIZE - offset)
1715                 return PAGE_SIZE - offset;
1716         else
1717                 return len;
1718 }
1719
1720 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
1721                                  void *data, int offset, int len)
1722 {
1723         int r;
1724         unsigned long addr;
1725
1726         addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1727         if (kvm_is_error_hva(addr))
1728                 return -EFAULT;
1729         r = __copy_from_user(data, (void __user *)addr + offset, len);
1730         if (r)
1731                 return -EFAULT;
1732         return 0;
1733 }
1734
1735 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1736                         int len)
1737 {
1738         struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1739
1740         return __kvm_read_guest_page(slot, gfn, data, offset, len);
1741 }
1742 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1743
1744 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
1745                              int offset, int len)
1746 {
1747         struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1748
1749         return __kvm_read_guest_page(slot, gfn, data, offset, len);
1750 }
1751 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
1752
1753 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1754 {
1755         gfn_t gfn = gpa >> PAGE_SHIFT;
1756         int seg;
1757         int offset = offset_in_page(gpa);
1758         int ret;
1759
1760         while ((seg = next_segment(len, offset)) != 0) {
1761                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1762                 if (ret < 0)
1763                         return ret;
1764                 offset = 0;
1765                 len -= seg;
1766                 data += seg;
1767                 ++gfn;
1768         }
1769         return 0;
1770 }
1771 EXPORT_SYMBOL_GPL(kvm_read_guest);
1772
1773 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
1774 {
1775         gfn_t gfn = gpa >> PAGE_SHIFT;
1776         int seg;
1777         int offset = offset_in_page(gpa);
1778         int ret;
1779
1780         while ((seg = next_segment(len, offset)) != 0) {
1781                 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
1782                 if (ret < 0)
1783                         return ret;
1784                 offset = 0;
1785                 len -= seg;
1786                 data += seg;
1787                 ++gfn;
1788         }
1789         return 0;
1790 }
1791 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
1792
1793 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1794                                    void *data, int offset, unsigned long len)
1795 {
1796         int r;
1797         unsigned long addr;
1798
1799         addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1800         if (kvm_is_error_hva(addr))
1801                 return -EFAULT;
1802         pagefault_disable();
1803         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1804         pagefault_enable();
1805         if (r)
1806                 return -EFAULT;
1807         return 0;
1808 }
1809
1810 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1811                           unsigned long len)
1812 {
1813         gfn_t gfn = gpa >> PAGE_SHIFT;
1814         struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1815         int offset = offset_in_page(gpa);
1816
1817         return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1818 }
1819 EXPORT_SYMBOL_GPL(kvm_read_guest_atomic);
1820
1821 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
1822                                void *data, unsigned long len)
1823 {
1824         gfn_t gfn = gpa >> PAGE_SHIFT;
1825         struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1826         int offset = offset_in_page(gpa);
1827
1828         return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1829 }
1830 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
1831
1832 static int __kvm_write_guest_page(struct kvm_memory_slot *memslot, gfn_t gfn,
1833                                   const void *data, int offset, int len)
1834 {
1835         int r;
1836         unsigned long addr;
1837
1838         addr = gfn_to_hva_memslot(memslot, gfn);
1839         if (kvm_is_error_hva(addr))
1840                 return -EFAULT;
1841         r = __copy_to_user((void __user *)addr + offset, data, len);
1842         if (r)
1843                 return -EFAULT;
1844         mark_page_dirty_in_slot(memslot, gfn);
1845         return 0;
1846 }
1847
1848 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
1849                          const void *data, int offset, int len)
1850 {
1851         struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1852
1853         return __kvm_write_guest_page(slot, gfn, data, offset, len);
1854 }
1855 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1856
1857 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
1858                               const void *data, int offset, int len)
1859 {
1860         struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1861
1862         return __kvm_write_guest_page(slot, gfn, data, offset, len);
1863 }
1864 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
1865
1866 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1867                     unsigned long len)
1868 {
1869         gfn_t gfn = gpa >> PAGE_SHIFT;
1870         int seg;
1871         int offset = offset_in_page(gpa);
1872         int ret;
1873
1874         while ((seg = next_segment(len, offset)) != 0) {
1875                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1876                 if (ret < 0)
1877                         return ret;
1878                 offset = 0;
1879                 len -= seg;
1880                 data += seg;
1881                 ++gfn;
1882         }
1883         return 0;
1884 }
1885 EXPORT_SYMBOL_GPL(kvm_write_guest);
1886
1887 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
1888                          unsigned long len)
1889 {
1890         gfn_t gfn = gpa >> PAGE_SHIFT;
1891         int seg;
1892         int offset = offset_in_page(gpa);
1893         int ret;
1894
1895         while ((seg = next_segment(len, offset)) != 0) {
1896                 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
1897                 if (ret < 0)
1898                         return ret;
1899                 offset = 0;
1900                 len -= seg;
1901                 data += seg;
1902                 ++gfn;
1903         }
1904         return 0;
1905 }
1906 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
1907
1908 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
1909                                        struct gfn_to_hva_cache *ghc,
1910                                        gpa_t gpa, unsigned long len)
1911 {
1912         int offset = offset_in_page(gpa);
1913         gfn_t start_gfn = gpa >> PAGE_SHIFT;
1914         gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1915         gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1916         gfn_t nr_pages_avail;
1917
1918         ghc->gpa = gpa;
1919         ghc->generation = slots->generation;
1920         ghc->len = len;
1921         ghc->memslot = __gfn_to_memslot(slots, start_gfn);
1922         ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, NULL);
1923         if (!kvm_is_error_hva(ghc->hva) && nr_pages_needed <= 1) {
1924                 ghc->hva += offset;
1925         } else {
1926                 /*
1927                  * If the requested region crosses two memslots, we still
1928                  * verify that the entire region is valid here.
1929                  */
1930                 while (start_gfn <= end_gfn) {
1931                         nr_pages_avail = 0;
1932                         ghc->memslot = __gfn_to_memslot(slots, start_gfn);
1933                         ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1934                                                    &nr_pages_avail);
1935                         if (kvm_is_error_hva(ghc->hva))
1936                                 return -EFAULT;
1937                         start_gfn += nr_pages_avail;
1938                 }
1939                 /* Use the slow path for cross page reads and writes. */
1940                 ghc->memslot = NULL;
1941         }
1942         return 0;
1943 }
1944
1945 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1946                               gpa_t gpa, unsigned long len)
1947 {
1948         struct kvm_memslots *slots = kvm_memslots(kvm);
1949         return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len);
1950 }
1951 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1952
1953 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1954                            void *data, int offset, unsigned long len)
1955 {
1956         struct kvm_memslots *slots = kvm_memslots(kvm);
1957         int r;
1958         gpa_t gpa = ghc->gpa + offset;
1959
1960         BUG_ON(len + offset > ghc->len);
1961
1962         if (slots->generation != ghc->generation)
1963                 __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len);
1964
1965         if (unlikely(!ghc->memslot))
1966                 return kvm_write_guest(kvm, gpa, data, len);
1967
1968         if (kvm_is_error_hva(ghc->hva))
1969                 return -EFAULT;
1970
1971         r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
1972         if (r)
1973                 return -EFAULT;
1974         mark_page_dirty_in_slot(ghc->memslot, gpa >> PAGE_SHIFT);
1975
1976         return 0;
1977 }
1978 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached);
1979
1980 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1981                            void *data, unsigned long len)
1982 {
1983         return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len);
1984 }
1985 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1986
1987 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1988                            void *data, unsigned long len)
1989 {
1990         struct kvm_memslots *slots = kvm_memslots(kvm);
1991         int r;
1992
1993         BUG_ON(len > ghc->len);
1994
1995         if (slots->generation != ghc->generation)
1996                 __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len);
1997
1998         if (unlikely(!ghc->memslot))
1999                 return kvm_read_guest(kvm, ghc->gpa, data, len);
2000
2001         if (kvm_is_error_hva(ghc->hva))
2002                 return -EFAULT;
2003
2004         r = __copy_from_user(data, (void __user *)ghc->hva, len);
2005         if (r)
2006                 return -EFAULT;
2007
2008         return 0;
2009 }
2010 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
2011
2012 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
2013 {
2014         const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
2015
2016         return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
2017 }
2018 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
2019
2020 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
2021 {
2022         gfn_t gfn = gpa >> PAGE_SHIFT;
2023         int seg;
2024         int offset = offset_in_page(gpa);
2025         int ret;
2026
2027         while ((seg = next_segment(len, offset)) != 0) {
2028                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
2029                 if (ret < 0)
2030                         return ret;
2031                 offset = 0;
2032                 len -= seg;
2033                 ++gfn;
2034         }
2035         return 0;
2036 }
2037 EXPORT_SYMBOL_GPL(kvm_clear_guest);
2038
2039 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot,
2040                                     gfn_t gfn)
2041 {
2042         if (memslot && memslot->dirty_bitmap) {
2043                 unsigned long rel_gfn = gfn - memslot->base_gfn;
2044
2045                 set_bit_le(rel_gfn, memslot->dirty_bitmap);
2046         }
2047 }
2048
2049 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
2050 {
2051         struct kvm_memory_slot *memslot;
2052
2053         memslot = gfn_to_memslot(kvm, gfn);
2054         mark_page_dirty_in_slot(memslot, gfn);
2055 }
2056 EXPORT_SYMBOL_GPL(mark_page_dirty);
2057
2058 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
2059 {
2060         struct kvm_memory_slot *memslot;
2061
2062         memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2063         mark_page_dirty_in_slot(memslot, gfn);
2064 }
2065 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
2066
2067 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
2068 {
2069         unsigned int old, val, grow;
2070
2071         old = val = vcpu->halt_poll_ns;
2072         grow = READ_ONCE(halt_poll_ns_grow);
2073         /* 10us base */
2074         if (val == 0 && grow)
2075                 val = 10000;
2076         else
2077                 val *= grow;
2078
2079         if (val > halt_poll_ns)
2080                 val = halt_poll_ns;
2081
2082         vcpu->halt_poll_ns = val;
2083         trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
2084 }
2085
2086 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
2087 {
2088         unsigned int old, val, shrink;
2089
2090         old = val = vcpu->halt_poll_ns;
2091         shrink = READ_ONCE(halt_poll_ns_shrink);
2092         if (shrink == 0)
2093                 val = 0;
2094         else
2095                 val /= shrink;
2096
2097         vcpu->halt_poll_ns = val;
2098         trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
2099 }
2100
2101 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
2102 {
2103         if (kvm_arch_vcpu_runnable(vcpu)) {
2104                 kvm_make_request(KVM_REQ_UNHALT, vcpu);
2105                 return -EINTR;
2106         }
2107         if (kvm_cpu_has_pending_timer(vcpu))
2108                 return -EINTR;
2109         if (signal_pending(current))
2110                 return -EINTR;
2111
2112         return 0;
2113 }
2114
2115 /*
2116  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2117  */
2118 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
2119 {
2120         ktime_t start, cur;
2121         DECLARE_SWAITQUEUE(wait);
2122         bool waited = false;
2123         u64 block_ns;
2124
2125         start = cur = ktime_get();
2126         if (vcpu->halt_poll_ns) {
2127                 ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns);
2128
2129                 ++vcpu->stat.halt_attempted_poll;
2130                 do {
2131                         /*
2132                          * This sets KVM_REQ_UNHALT if an interrupt
2133                          * arrives.
2134                          */
2135                         if (kvm_vcpu_check_block(vcpu) < 0) {
2136                                 ++vcpu->stat.halt_successful_poll;
2137                                 if (!vcpu_valid_wakeup(vcpu))
2138                                         ++vcpu->stat.halt_poll_invalid;
2139                                 goto out;
2140                         }
2141                         cur = ktime_get();
2142                 } while (single_task_running() && ktime_before(cur, stop));
2143         }
2144
2145         kvm_arch_vcpu_blocking(vcpu);
2146
2147         for (;;) {
2148                 prepare_to_swait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
2149
2150                 if (kvm_vcpu_check_block(vcpu) < 0)
2151                         break;
2152
2153                 waited = true;
2154                 schedule();
2155         }
2156
2157         finish_swait(&vcpu->wq, &wait);
2158         cur = ktime_get();
2159
2160         kvm_arch_vcpu_unblocking(vcpu);
2161 out:
2162         block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
2163
2164         if (!vcpu_valid_wakeup(vcpu))
2165                 shrink_halt_poll_ns(vcpu);
2166         else if (halt_poll_ns) {
2167                 if (block_ns <= vcpu->halt_poll_ns)
2168                         ;
2169                 /* we had a long block, shrink polling */
2170                 else if (vcpu->halt_poll_ns && block_ns > halt_poll_ns)
2171                         shrink_halt_poll_ns(vcpu);
2172                 /* we had a short halt and our poll time is too small */
2173                 else if (vcpu->halt_poll_ns < halt_poll_ns &&
2174                         block_ns < halt_poll_ns)
2175                         grow_halt_poll_ns(vcpu);
2176         } else
2177                 vcpu->halt_poll_ns = 0;
2178
2179         trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu));
2180         kvm_arch_vcpu_block_finish(vcpu);
2181 }
2182 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
2183
2184 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
2185 {
2186         struct swait_queue_head *wqp;
2187
2188         wqp = kvm_arch_vcpu_wq(vcpu);
2189         if (swait_active(wqp)) {
2190                 swake_up(wqp);
2191                 ++vcpu->stat.halt_wakeup;
2192                 return true;
2193         }
2194
2195         return false;
2196 }
2197 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
2198
2199 #ifndef CONFIG_S390
2200 /*
2201  * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2202  */
2203 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
2204 {
2205         int me;
2206         int cpu = vcpu->cpu;
2207
2208         if (kvm_vcpu_wake_up(vcpu))
2209                 return;
2210
2211         me = get_cpu();
2212         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
2213                 if (kvm_arch_vcpu_should_kick(vcpu))
2214                         smp_send_reschedule(cpu);
2215         put_cpu();
2216 }
2217 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
2218 #endif /* !CONFIG_S390 */
2219
2220 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
2221 {
2222         struct pid *pid;
2223         struct task_struct *task = NULL;
2224         int ret = 0;
2225
2226         rcu_read_lock();
2227         pid = rcu_dereference(target->pid);
2228         if (pid)
2229                 task = get_pid_task(pid, PIDTYPE_PID);
2230         rcu_read_unlock();
2231         if (!task)
2232                 return ret;
2233         ret = yield_to(task, 1);
2234         put_task_struct(task);
2235
2236         return ret;
2237 }
2238 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
2239
2240 /*
2241  * Helper that checks whether a VCPU is eligible for directed yield.
2242  * Most eligible candidate to yield is decided by following heuristics:
2243  *
2244  *  (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2245  *  (preempted lock holder), indicated by @in_spin_loop.
2246  *  Set at the beiginning and cleared at the end of interception/PLE handler.
2247  *
2248  *  (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2249  *  chance last time (mostly it has become eligible now since we have probably
2250  *  yielded to lockholder in last iteration. This is done by toggling
2251  *  @dy_eligible each time a VCPU checked for eligibility.)
2252  *
2253  *  Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2254  *  to preempted lock-holder could result in wrong VCPU selection and CPU
2255  *  burning. Giving priority for a potential lock-holder increases lock
2256  *  progress.
2257  *
2258  *  Since algorithm is based on heuristics, accessing another VCPU data without
2259  *  locking does not harm. It may result in trying to yield to  same VCPU, fail
2260  *  and continue with next VCPU and so on.
2261  */
2262 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
2263 {
2264 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2265         bool eligible;
2266
2267         eligible = !vcpu->spin_loop.in_spin_loop ||
2268                     vcpu->spin_loop.dy_eligible;
2269
2270         if (vcpu->spin_loop.in_spin_loop)
2271                 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
2272
2273         return eligible;
2274 #else
2275         return true;
2276 #endif
2277 }
2278
2279 void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode)
2280 {
2281         struct kvm *kvm = me->kvm;
2282         struct kvm_vcpu *vcpu;
2283         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
2284         int yielded = 0;
2285         int try = 3;
2286         int pass;
2287         int i;
2288
2289         kvm_vcpu_set_in_spin_loop(me, true);
2290         /*
2291          * We boost the priority of a VCPU that is runnable but not
2292          * currently running, because it got preempted by something
2293          * else and called schedule in __vcpu_run.  Hopefully that
2294          * VCPU is holding the lock that we need and will release it.
2295          * We approximate round-robin by starting at the last boosted VCPU.
2296          */
2297         for (pass = 0; pass < 2 && !yielded && try; pass++) {
2298                 kvm_for_each_vcpu(i, vcpu, kvm) {
2299                         if (!pass && i <= last_boosted_vcpu) {
2300                                 i = last_boosted_vcpu;
2301                                 continue;
2302                         } else if (pass && i > last_boosted_vcpu)
2303                                 break;
2304                         if (!ACCESS_ONCE(vcpu->preempted))
2305                                 continue;
2306                         if (vcpu == me)
2307                                 continue;
2308                         if (swait_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
2309                                 continue;
2310                         if (yield_to_kernel_mode && !kvm_arch_vcpu_in_kernel(vcpu))
2311                                 continue;
2312                         if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
2313                                 continue;
2314
2315                         yielded = kvm_vcpu_yield_to(vcpu);
2316                         if (yielded > 0) {
2317                                 kvm->last_boosted_vcpu = i;
2318                                 break;
2319                         } else if (yielded < 0) {
2320                                 try--;
2321                                 if (!try)
2322                                         break;
2323                         }
2324                 }
2325         }
2326         kvm_vcpu_set_in_spin_loop(me, false);
2327
2328         /* Ensure vcpu is not eligible during next spinloop */
2329         kvm_vcpu_set_dy_eligible(me, false);
2330 }
2331 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
2332
2333 static int kvm_vcpu_fault(struct vm_fault *vmf)
2334 {
2335         struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data;
2336         struct page *page;
2337
2338         if (vmf->pgoff == 0)
2339                 page = virt_to_page(vcpu->run);
2340 #ifdef CONFIG_X86
2341         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
2342                 page = virt_to_page(vcpu->arch.pio_data);
2343 #endif
2344 #ifdef CONFIG_KVM_MMIO
2345         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
2346                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
2347 #endif
2348         else
2349                 return kvm_arch_vcpu_fault(vcpu, vmf);
2350         get_page(page);
2351         vmf->page = page;
2352         return 0;
2353 }
2354
2355 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
2356         .fault = kvm_vcpu_fault,
2357 };
2358
2359 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2360 {
2361         vma->vm_ops = &kvm_vcpu_vm_ops;
2362         return 0;
2363 }
2364
2365 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2366 {
2367         struct kvm_vcpu *vcpu = filp->private_data;
2368
2369         debugfs_remove_recursive(vcpu->debugfs_dentry);
2370         kvm_put_kvm(vcpu->kvm);
2371         return 0;
2372 }
2373
2374 static struct file_operations kvm_vcpu_fops = {
2375         .release        = kvm_vcpu_release,
2376         .unlocked_ioctl = kvm_vcpu_ioctl,
2377 #ifdef CONFIG_KVM_COMPAT
2378         .compat_ioctl   = kvm_vcpu_compat_ioctl,
2379 #endif
2380         .mmap           = kvm_vcpu_mmap,
2381         .llseek         = noop_llseek,
2382 };
2383
2384 /*
2385  * Allocates an inode for the vcpu.
2386  */
2387 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2388 {
2389         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
2390 }
2391
2392 static int kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
2393 {
2394         char dir_name[ITOA_MAX_LEN * 2];
2395         int ret;
2396
2397         if (!kvm_arch_has_vcpu_debugfs())
2398                 return 0;
2399
2400         if (!debugfs_initialized())
2401                 return 0;
2402
2403         snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
2404         vcpu->debugfs_dentry = debugfs_create_dir(dir_name,
2405                                                                 vcpu->kvm->debugfs_dentry);
2406         if (!vcpu->debugfs_dentry)
2407                 return -ENOMEM;
2408
2409         ret = kvm_arch_create_vcpu_debugfs(vcpu);
2410         if (ret < 0) {
2411                 debugfs_remove_recursive(vcpu->debugfs_dentry);
2412                 return ret;
2413         }
2414
2415         return 0;
2416 }
2417
2418 /*
2419  * Creates some virtual cpus.  Good luck creating more than one.
2420  */
2421 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
2422 {
2423         int r;
2424         struct kvm_vcpu *vcpu;
2425
2426         if (id >= KVM_MAX_VCPU_ID)
2427                 return -EINVAL;
2428
2429         mutex_lock(&kvm->lock);
2430         if (kvm->created_vcpus == KVM_MAX_VCPUS) {
2431                 mutex_unlock(&kvm->lock);
2432                 return -EINVAL;
2433         }
2434
2435         kvm->created_vcpus++;
2436         mutex_unlock(&kvm->lock);
2437
2438         vcpu = kvm_arch_vcpu_create(kvm, id);
2439         if (IS_ERR(vcpu)) {
2440                 r = PTR_ERR(vcpu);
2441                 goto vcpu_decrement;
2442         }
2443
2444         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2445
2446         r = kvm_arch_vcpu_setup(vcpu);
2447         if (r)
2448                 goto vcpu_destroy;
2449
2450         r = kvm_create_vcpu_debugfs(vcpu);
2451         if (r)
2452                 goto vcpu_destroy;
2453
2454         mutex_lock(&kvm->lock);
2455         if (kvm_get_vcpu_by_id(kvm, id)) {
2456                 r = -EEXIST;
2457                 goto unlock_vcpu_destroy;
2458         }
2459
2460         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
2461
2462         /* Now it's all set up, let userspace reach it */
2463         kvm_get_kvm(kvm);
2464         r = create_vcpu_fd(vcpu);
2465         if (r < 0) {
2466                 kvm_put_kvm(kvm);
2467                 goto unlock_vcpu_destroy;
2468         }
2469
2470         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
2471
2472         /*
2473          * Pairs with smp_rmb() in kvm_get_vcpu.  Write kvm->vcpus
2474          * before kvm->online_vcpu's incremented value.
2475          */
2476         smp_wmb();
2477         atomic_inc(&kvm->online_vcpus);
2478
2479         mutex_unlock(&kvm->lock);
2480         kvm_arch_vcpu_postcreate(vcpu);
2481         return r;
2482
2483 unlock_vcpu_destroy:
2484         mutex_unlock(&kvm->lock);
2485         debugfs_remove_recursive(vcpu->debugfs_dentry);
2486 vcpu_destroy:
2487         kvm_arch_vcpu_destroy(vcpu);
2488 vcpu_decrement:
2489         mutex_lock(&kvm->lock);
2490         kvm->created_vcpus--;
2491         mutex_unlock(&kvm->lock);
2492         return r;
2493 }
2494
2495 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2496 {
2497         if (sigset) {
2498                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2499                 vcpu->sigset_active = 1;
2500                 vcpu->sigset = *sigset;
2501         } else
2502                 vcpu->sigset_active = 0;
2503         return 0;
2504 }
2505
2506 static long kvm_vcpu_ioctl(struct file *filp,
2507                            unsigned int ioctl, unsigned long arg)
2508 {
2509         struct kvm_vcpu *vcpu = filp->private_data;
2510         void __user *argp = (void __user *)arg;
2511         int r;
2512         struct kvm_fpu *fpu = NULL;
2513         struct kvm_sregs *kvm_sregs = NULL;
2514
2515         if (vcpu->kvm->mm != current->mm)
2516                 return -EIO;
2517
2518         if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2519                 return -EINVAL;
2520
2521 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
2522         /*
2523          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
2524          * so vcpu_load() would break it.
2525          */
2526         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_S390_IRQ || ioctl == KVM_INTERRUPT)
2527                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2528 #endif
2529
2530
2531         r = vcpu_load(vcpu);
2532         if (r)
2533                 return r;
2534         switch (ioctl) {
2535         case KVM_RUN: {
2536                 struct pid *oldpid;
2537                 r = -EINVAL;
2538                 if (arg)
2539                         goto out;
2540                 oldpid = rcu_access_pointer(vcpu->pid);
2541                 if (unlikely(oldpid != current->pids[PIDTYPE_PID].pid)) {
2542                         /* The thread running this VCPU changed. */
2543                         struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
2544
2545                         rcu_assign_pointer(vcpu->pid, newpid);
2546                         if (oldpid)
2547                                 synchronize_rcu();
2548                         put_pid(oldpid);
2549                 }
2550                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2551                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2552                 break;
2553         }
2554         case KVM_GET_REGS: {
2555                 struct kvm_regs *kvm_regs;
2556
2557                 r = -ENOMEM;
2558                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2559                 if (!kvm_regs)
2560                         goto out;
2561                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2562                 if (r)
2563                         goto out_free1;
2564                 r = -EFAULT;
2565                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2566                         goto out_free1;
2567                 r = 0;
2568 out_free1:
2569                 kfree(kvm_regs);
2570                 break;
2571         }
2572         case KVM_SET_REGS: {
2573                 struct kvm_regs *kvm_regs;
2574
2575                 r = -ENOMEM;
2576                 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2577                 if (IS_ERR(kvm_regs)) {
2578                         r = PTR_ERR(kvm_regs);
2579                         goto out;
2580                 }
2581                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2582                 kfree(kvm_regs);
2583                 break;
2584         }
2585         case KVM_GET_SREGS: {
2586                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2587                 r = -ENOMEM;
2588                 if (!kvm_sregs)
2589                         goto out;
2590                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2591                 if (r)
2592                         goto out;
2593                 r = -EFAULT;
2594                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2595                         goto out;
2596                 r = 0;
2597                 break;
2598         }
2599         case KVM_SET_SREGS: {
2600                 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2601                 if (IS_ERR(kvm_sregs)) {
2602                         r = PTR_ERR(kvm_sregs);
2603                         kvm_sregs = NULL;
2604                         goto out;
2605                 }
2606                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2607                 break;
2608         }
2609         case KVM_GET_MP_STATE: {
2610                 struct kvm_mp_state mp_state;
2611
2612                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2613                 if (r)
2614                         goto out;
2615                 r = -EFAULT;
2616                 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
2617                         goto out;
2618                 r = 0;
2619                 break;
2620         }
2621         case KVM_SET_MP_STATE: {
2622                 struct kvm_mp_state mp_state;
2623
2624                 r = -EFAULT;
2625                 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
2626                         goto out;
2627                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2628                 break;
2629         }
2630         case KVM_TRANSLATE: {
2631                 struct kvm_translation tr;
2632
2633                 r = -EFAULT;
2634                 if (copy_from_user(&tr, argp, sizeof(tr)))
2635                         goto out;
2636                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2637                 if (r)
2638                         goto out;
2639                 r = -EFAULT;
2640                 if (copy_to_user(argp, &tr, sizeof(tr)))
2641                         goto out;
2642                 r = 0;
2643                 break;
2644         }
2645         case KVM_SET_GUEST_DEBUG: {
2646                 struct kvm_guest_debug dbg;
2647
2648                 r = -EFAULT;
2649                 if (copy_from_user(&dbg, argp, sizeof(dbg)))
2650                         goto out;
2651                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2652                 break;
2653         }
2654         case KVM_SET_SIGNAL_MASK: {
2655                 struct kvm_signal_mask __user *sigmask_arg = argp;
2656                 struct kvm_signal_mask kvm_sigmask;
2657                 sigset_t sigset, *p;
2658
2659                 p = NULL;
2660                 if (argp) {
2661                         r = -EFAULT;
2662                         if (copy_from_user(&kvm_sigmask, argp,
2663                                            sizeof(kvm_sigmask)))
2664                                 goto out;
2665                         r = -EINVAL;
2666                         if (kvm_sigmask.len != sizeof(sigset))
2667                                 goto out;
2668                         r = -EFAULT;
2669                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2670                                            sizeof(sigset)))
2671                                 goto out;
2672                         p = &sigset;
2673                 }
2674                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2675                 break;
2676         }
2677         case KVM_GET_FPU: {
2678                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2679                 r = -ENOMEM;
2680                 if (!fpu)
2681                         goto out;
2682                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2683                 if (r)
2684                         goto out;
2685                 r = -EFAULT;
2686                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2687                         goto out;
2688                 r = 0;
2689                 break;
2690         }
2691         case KVM_SET_FPU: {
2692                 fpu = memdup_user(argp, sizeof(*fpu));
2693                 if (IS_ERR(fpu)) {
2694                         r = PTR_ERR(fpu);
2695                         fpu = NULL;
2696                         goto out;
2697                 }
2698                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2699                 break;
2700         }
2701         default:
2702                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2703         }
2704 out:
2705         vcpu_put(vcpu);
2706         kfree(fpu);
2707         kfree(kvm_sregs);
2708         return r;
2709 }
2710
2711 #ifdef CONFIG_KVM_COMPAT
2712 static long kvm_vcpu_compat_ioctl(struct file *filp,
2713                                   unsigned int ioctl, unsigned long arg)
2714 {
2715         struct kvm_vcpu *vcpu = filp->private_data;
2716         void __user *argp = compat_ptr(arg);
2717         int r;
2718
2719         if (vcpu->kvm->mm != current->mm)
2720                 return -EIO;
2721
2722         switch (ioctl) {
2723         case KVM_SET_SIGNAL_MASK: {
2724                 struct kvm_signal_mask __user *sigmask_arg = argp;
2725                 struct kvm_signal_mask kvm_sigmask;
2726                 compat_sigset_t csigset;
2727                 sigset_t sigset;
2728
2729                 if (argp) {
2730                         r = -EFAULT;
2731                         if (copy_from_user(&kvm_sigmask, argp,
2732                                            sizeof(kvm_sigmask)))
2733                                 goto out;
2734                         r = -EINVAL;
2735                         if (kvm_sigmask.len != sizeof(csigset))
2736                                 goto out;
2737                         r = -EFAULT;
2738                         if (copy_from_user(&csigset, sigmask_arg->sigset,
2739                                            sizeof(csigset)))
2740                                 goto out;
2741                         sigset_from_compat(&sigset, &csigset);
2742                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2743                 } else
2744                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2745                 break;
2746         }
2747         default:
2748                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2749         }
2750
2751 out:
2752         return r;
2753 }
2754 #endif
2755
2756 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2757                                  int (*accessor)(struct kvm_device *dev,
2758                                                  struct kvm_device_attr *attr),
2759                                  unsigned long arg)
2760 {
2761         struct kvm_device_attr attr;
2762
2763         if (!accessor)
2764                 return -EPERM;
2765
2766         if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2767                 return -EFAULT;
2768
2769         return accessor(dev, &attr);
2770 }
2771
2772 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2773                              unsigned long arg)
2774 {
2775         struct kvm_device *dev = filp->private_data;
2776
2777         switch (ioctl) {
2778         case KVM_SET_DEVICE_ATTR:
2779                 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2780         case KVM_GET_DEVICE_ATTR:
2781                 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2782         case KVM_HAS_DEVICE_ATTR:
2783                 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2784         default:
2785                 if (dev->ops->ioctl)
2786                         return dev->ops->ioctl(dev, ioctl, arg);
2787
2788                 return -ENOTTY;
2789         }
2790 }
2791
2792 static int kvm_device_release(struct inode *inode, struct file *filp)
2793 {
2794         struct kvm_device *dev = filp->private_data;
2795         struct kvm *kvm = dev->kvm;
2796
2797         kvm_put_kvm(kvm);
2798         return 0;
2799 }
2800
2801 static const struct file_operations kvm_device_fops = {
2802         .unlocked_ioctl = kvm_device_ioctl,
2803 #ifdef CONFIG_KVM_COMPAT
2804         .compat_ioctl = kvm_device_ioctl,
2805 #endif
2806         .release = kvm_device_release,
2807 };
2808
2809 struct kvm_device *kvm_device_from_filp(struct file *filp)
2810 {
2811         if (filp->f_op != &kvm_device_fops)
2812                 return NULL;
2813
2814         return filp->private_data;
2815 }
2816
2817 static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
2818 #ifdef CONFIG_KVM_MPIC
2819         [KVM_DEV_TYPE_FSL_MPIC_20]      = &kvm_mpic_ops,
2820         [KVM_DEV_TYPE_FSL_MPIC_42]      = &kvm_mpic_ops,
2821 #endif
2822 };
2823
2824 int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
2825 {
2826         if (type >= ARRAY_SIZE(kvm_device_ops_table))
2827                 return -ENOSPC;
2828
2829         if (kvm_device_ops_table[type] != NULL)
2830                 return -EEXIST;
2831
2832         kvm_device_ops_table[type] = ops;
2833         return 0;
2834 }
2835
2836 void kvm_unregister_device_ops(u32 type)
2837 {
2838         if (kvm_device_ops_table[type] != NULL)
2839                 kvm_device_ops_table[type] = NULL;
2840 }
2841
2842 static int kvm_ioctl_create_device(struct kvm *kvm,
2843                                    struct kvm_create_device *cd)
2844 {
2845         struct kvm_device_ops *ops = NULL;
2846         struct kvm_device *dev;
2847         bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2848         int ret;
2849
2850         if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
2851                 return -ENODEV;
2852
2853         ops = kvm_device_ops_table[cd->type];
2854         if (ops == NULL)
2855                 return -ENODEV;
2856
2857         if (test)
2858                 return 0;
2859
2860         dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2861         if (!dev)
2862                 return -ENOMEM;
2863
2864         dev->ops = ops;
2865         dev->kvm = kvm;
2866
2867         mutex_lock(&kvm->lock);
2868         ret = ops->create(dev, cd->type);
2869         if (ret < 0) {
2870                 mutex_unlock(&kvm->lock);
2871                 kfree(dev);
2872                 return ret;
2873         }
2874         list_add(&dev->vm_node, &kvm->devices);
2875         mutex_unlock(&kvm->lock);
2876
2877         if (ops->init)
2878                 ops->init(dev);
2879
2880         ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2881         if (ret < 0) {
2882                 mutex_lock(&kvm->lock);
2883                 list_del(&dev->vm_node);
2884                 mutex_unlock(&kvm->lock);
2885                 ops->destroy(dev);
2886                 return ret;
2887         }
2888
2889         kvm_get_kvm(kvm);
2890         cd->fd = ret;
2891         return 0;
2892 }
2893
2894 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2895 {
2896         switch (arg) {
2897         case KVM_CAP_USER_MEMORY:
2898         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2899         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2900         case KVM_CAP_INTERNAL_ERROR_DATA:
2901 #ifdef CONFIG_HAVE_KVM_MSI
2902         case KVM_CAP_SIGNAL_MSI:
2903 #endif
2904 #ifdef CONFIG_HAVE_KVM_IRQFD
2905         case KVM_CAP_IRQFD:
2906         case KVM_CAP_IRQFD_RESAMPLE:
2907 #endif
2908         case KVM_CAP_IOEVENTFD_ANY_LENGTH:
2909         case KVM_CAP_CHECK_EXTENSION_VM:
2910                 return 1;
2911 #ifdef CONFIG_KVM_MMIO
2912         case KVM_CAP_COALESCED_MMIO:
2913                 return KVM_COALESCED_MMIO_PAGE_OFFSET;
2914 #endif
2915 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2916         case KVM_CAP_IRQ_ROUTING:
2917                 return KVM_MAX_IRQ_ROUTES;
2918 #endif
2919 #if KVM_ADDRESS_SPACE_NUM > 1
2920         case KVM_CAP_MULTI_ADDRESS_SPACE:
2921                 return KVM_ADDRESS_SPACE_NUM;
2922 #endif
2923         case KVM_CAP_MAX_VCPU_ID:
2924                 return KVM_MAX_VCPU_ID;
2925         default:
2926                 break;
2927         }
2928         return kvm_vm_ioctl_check_extension(kvm, arg);
2929 }
2930
2931 static long kvm_vm_ioctl(struct file *filp,
2932                            unsigned int ioctl, unsigned long arg)
2933 {
2934         struct kvm *kvm = filp->private_data;
2935         void __user *argp = (void __user *)arg;
2936         int r;
2937
2938         if (kvm->mm != current->mm)
2939                 return -EIO;
2940         switch (ioctl) {
2941         case KVM_CREATE_VCPU:
2942                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2943                 break;
2944         case KVM_SET_USER_MEMORY_REGION: {
2945                 struct kvm_userspace_memory_region kvm_userspace_mem;
2946
2947                 r = -EFAULT;
2948                 if (copy_from_user(&kvm_userspace_mem, argp,
2949                                                 sizeof(kvm_userspace_mem)))
2950                         goto out;
2951
2952                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2953                 break;
2954         }
2955         case KVM_GET_DIRTY_LOG: {
2956                 struct kvm_dirty_log log;
2957
2958                 r = -EFAULT;
2959                 if (copy_from_user(&log, argp, sizeof(log)))
2960                         goto out;
2961                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2962                 break;
2963         }
2964 #ifdef CONFIG_KVM_MMIO
2965         case KVM_REGISTER_COALESCED_MMIO: {
2966                 struct kvm_coalesced_mmio_zone zone;
2967
2968                 r = -EFAULT;
2969                 if (copy_from_user(&zone, argp, sizeof(zone)))
2970                         goto out;
2971                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2972                 break;
2973         }
2974         case KVM_UNREGISTER_COALESCED_MMIO: {
2975                 struct kvm_coalesced_mmio_zone zone;
2976
2977                 r = -EFAULT;
2978                 if (copy_from_user(&zone, argp, sizeof(zone)))
2979                         goto out;
2980                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2981                 break;
2982         }
2983 #endif
2984         case KVM_IRQFD: {
2985                 struct kvm_irqfd data;
2986
2987                 r = -EFAULT;
2988                 if (copy_from_user(&data, argp, sizeof(data)))
2989                         goto out;
2990                 r = kvm_irqfd(kvm, &data);
2991                 break;
2992         }
2993         case KVM_IOEVENTFD: {
2994                 struct kvm_ioeventfd data;
2995
2996                 r = -EFAULT;
2997                 if (copy_from_user(&data, argp, sizeof(data)))
2998                         goto out;
2999                 r = kvm_ioeventfd(kvm, &data);
3000                 break;
3001         }
3002 #ifdef CONFIG_HAVE_KVM_MSI
3003         case KVM_SIGNAL_MSI: {
3004                 struct kvm_msi msi;
3005
3006                 r = -EFAULT;
3007                 if (copy_from_user(&msi, argp, sizeof(msi)))
3008                         goto out;
3009                 r = kvm_send_userspace_msi(kvm, &msi);
3010                 break;
3011         }
3012 #endif
3013 #ifdef __KVM_HAVE_IRQ_LINE
3014         case KVM_IRQ_LINE_STATUS:
3015         case KVM_IRQ_LINE: {
3016                 struct kvm_irq_level irq_event;
3017
3018                 r = -EFAULT;
3019                 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
3020                         goto out;
3021
3022                 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
3023                                         ioctl == KVM_IRQ_LINE_STATUS);
3024                 if (r)
3025                         goto out;
3026
3027                 r = -EFAULT;
3028                 if (ioctl == KVM_IRQ_LINE_STATUS) {
3029                         if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
3030                                 goto out;
3031                 }
3032
3033                 r = 0;
3034                 break;
3035         }
3036 #endif
3037 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3038         case KVM_SET_GSI_ROUTING: {
3039                 struct kvm_irq_routing routing;
3040                 struct kvm_irq_routing __user *urouting;
3041                 struct kvm_irq_routing_entry *entries = NULL;
3042
3043                 r = -EFAULT;
3044                 if (copy_from_user(&routing, argp, sizeof(routing)))
3045                         goto out;
3046                 r = -EINVAL;
3047                 if (!kvm_arch_can_set_irq_routing(kvm))
3048                         goto out;
3049                 if (routing.nr > KVM_MAX_IRQ_ROUTES)
3050                         goto out;
3051                 if (routing.flags)
3052                         goto out;
3053                 if (routing.nr) {
3054                         r = -ENOMEM;
3055                         entries = vmalloc(routing.nr * sizeof(*entries));
3056                         if (!entries)
3057                                 goto out;
3058                         r = -EFAULT;
3059                         urouting = argp;
3060                         if (copy_from_user(entries, urouting->entries,
3061                                            routing.nr * sizeof(*entries)))
3062                                 goto out_free_irq_routing;
3063                 }
3064                 r = kvm_set_irq_routing(kvm, entries, routing.nr,
3065                                         routing.flags);
3066 out_free_irq_routing:
3067                 vfree(entries);
3068                 break;
3069         }
3070 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
3071         case KVM_CREATE_DEVICE: {
3072                 struct kvm_create_device cd;
3073
3074                 r = -EFAULT;
3075                 if (copy_from_user(&cd, argp, sizeof(cd)))
3076                         goto out;
3077
3078                 r = kvm_ioctl_create_device(kvm, &cd);
3079                 if (r)
3080                         goto out;
3081
3082                 r = -EFAULT;
3083                 if (copy_to_user(argp, &cd, sizeof(cd)))
3084                         goto out;
3085
3086                 r = 0;
3087                 break;
3088         }
3089         case KVM_CHECK_EXTENSION:
3090                 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
3091                 break;
3092         default:
3093                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
3094         }
3095 out:
3096         return r;
3097 }
3098
3099 #ifdef CONFIG_KVM_COMPAT
3100 struct compat_kvm_dirty_log {
3101         __u32 slot;
3102         __u32 padding1;
3103         union {
3104                 compat_uptr_t dirty_bitmap; /* one bit per page */
3105                 __u64 padding2;
3106         };
3107 };
3108
3109 static long kvm_vm_compat_ioctl(struct file *filp,
3110                            unsigned int ioctl, unsigned long arg)
3111 {
3112         struct kvm *kvm = filp->private_data;
3113         int r;
3114
3115         if (kvm->mm != current->mm)
3116                 return -EIO;
3117         switch (ioctl) {
3118         case KVM_GET_DIRTY_LOG: {
3119                 struct compat_kvm_dirty_log compat_log;
3120                 struct kvm_dirty_log log;
3121
3122                 if (copy_from_user(&compat_log, (void __user *)arg,
3123                                    sizeof(compat_log)))
3124                         return -EFAULT;
3125                 log.slot         = compat_log.slot;
3126                 log.padding1     = compat_log.padding1;
3127                 log.padding2     = compat_log.padding2;
3128                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
3129
3130                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3131                 break;
3132         }
3133         default:
3134                 r = kvm_vm_ioctl(filp, ioctl, arg);
3135         }
3136         return r;
3137 }
3138 #endif
3139
3140 static struct file_operations kvm_vm_fops = {
3141         .release        = kvm_vm_release,
3142         .unlocked_ioctl = kvm_vm_ioctl,
3143 #ifdef CONFIG_KVM_COMPAT
3144         .compat_ioctl   = kvm_vm_compat_ioctl,
3145 #endif
3146         .llseek         = noop_llseek,
3147 };
3148
3149 static int kvm_dev_ioctl_create_vm(unsigned long type)
3150 {
3151         int r;
3152         struct kvm *kvm;
3153         struct file *file;
3154
3155         kvm = kvm_create_vm(type);
3156         if (IS_ERR(kvm))
3157                 return PTR_ERR(kvm);
3158 #ifdef CONFIG_KVM_MMIO
3159         r = kvm_coalesced_mmio_init(kvm);
3160         if (r < 0) {
3161                 kvm_put_kvm(kvm);
3162                 return r;
3163         }
3164 #endif
3165         r = get_unused_fd_flags(O_CLOEXEC);
3166         if (r < 0) {
3167                 kvm_put_kvm(kvm);
3168                 return r;
3169         }
3170         file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
3171         if (IS_ERR(file)) {
3172                 put_unused_fd(r);
3173                 kvm_put_kvm(kvm);
3174                 return PTR_ERR(file);
3175         }
3176
3177         /*
3178          * Don't call kvm_put_kvm anymore at this point; file->f_op is
3179          * already set, with ->release() being kvm_vm_release().  In error
3180          * cases it will be called by the final fput(file) and will take
3181          * care of doing kvm_put_kvm(kvm).
3182          */
3183         if (kvm_create_vm_debugfs(kvm, r) < 0) {
3184                 put_unused_fd(r);
3185                 fput(file);
3186                 return -ENOMEM;
3187         }
3188         kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm);
3189
3190         fd_install(r, file);
3191         return r;
3192 }
3193
3194 static long kvm_dev_ioctl(struct file *filp,
3195                           unsigned int ioctl, unsigned long arg)
3196 {
3197         long r = -EINVAL;
3198
3199         switch (ioctl) {
3200         case KVM_GET_API_VERSION:
3201                 if (arg)
3202                         goto out;
3203                 r = KVM_API_VERSION;
3204                 break;
3205         case KVM_CREATE_VM:
3206                 r = kvm_dev_ioctl_create_vm(arg);
3207                 break;
3208         case KVM_CHECK_EXTENSION:
3209                 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
3210                 break;
3211         case KVM_GET_VCPU_MMAP_SIZE:
3212                 if (arg)
3213                         goto out;
3214                 r = PAGE_SIZE;     /* struct kvm_run */
3215 #ifdef CONFIG_X86
3216                 r += PAGE_SIZE;    /* pio data page */
3217 #endif
3218 #ifdef CONFIG_KVM_MMIO
3219                 r += PAGE_SIZE;    /* coalesced mmio ring page */
3220 #endif
3221                 break;
3222         case KVM_TRACE_ENABLE:
3223         case KVM_TRACE_PAUSE:
3224         case KVM_TRACE_DISABLE:
3225                 r = -EOPNOTSUPP;
3226                 break;
3227         default:
3228                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
3229         }
3230 out:
3231         return r;
3232 }
3233
3234 static struct file_operations kvm_chardev_ops = {
3235         .unlocked_ioctl = kvm_dev_ioctl,
3236         .compat_ioctl   = kvm_dev_ioctl,
3237         .llseek         = noop_llseek,
3238 };
3239
3240 static struct miscdevice kvm_dev = {
3241         KVM_MINOR,
3242         "kvm",
3243         &kvm_chardev_ops,
3244 };
3245
3246 static void hardware_enable_nolock(void *junk)
3247 {
3248         int cpu = raw_smp_processor_id();
3249         int r;
3250
3251         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
3252                 return;
3253
3254         cpumask_set_cpu(cpu, cpus_hardware_enabled);
3255
3256         r = kvm_arch_hardware_enable();
3257
3258         if (r) {
3259                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3260                 atomic_inc(&hardware_enable_failed);
3261                 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
3262         }
3263 }
3264
3265 static int kvm_starting_cpu(unsigned int cpu)
3266 {
3267         raw_spin_lock(&kvm_count_lock);
3268         if (kvm_usage_count)
3269                 hardware_enable_nolock(NULL);
3270         raw_spin_unlock(&kvm_count_lock);
3271         return 0;
3272 }
3273
3274 static void hardware_disable_nolock(void *junk)
3275 {
3276         int cpu = raw_smp_processor_id();
3277
3278         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
3279                 return;
3280         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3281         kvm_arch_hardware_disable();
3282 }
3283
3284 static int kvm_dying_cpu(unsigned int cpu)
3285 {
3286         raw_spin_lock(&kvm_count_lock);
3287         if (kvm_usage_count)
3288                 hardware_disable_nolock(NULL);
3289         raw_spin_unlock(&kvm_count_lock);
3290         return 0;
3291 }
3292
3293 static void hardware_disable_all_nolock(void)
3294 {
3295         BUG_ON(!kvm_usage_count);
3296
3297         kvm_usage_count--;
3298         if (!kvm_usage_count)
3299                 on_each_cpu(hardware_disable_nolock, NULL, 1);
3300 }
3301
3302 static void hardware_disable_all(void)
3303 {
3304         raw_spin_lock(&kvm_count_lock);
3305         hardware_disable_all_nolock();
3306         raw_spin_unlock(&kvm_count_lock);
3307 }
3308
3309 static int hardware_enable_all(void)
3310 {
3311         int r = 0;
3312
3313         raw_spin_lock(&kvm_count_lock);
3314
3315         kvm_usage_count++;
3316         if (kvm_usage_count == 1) {
3317                 atomic_set(&hardware_enable_failed, 0);
3318                 on_each_cpu(hardware_enable_nolock, NULL, 1);
3319
3320                 if (atomic_read(&hardware_enable_failed)) {
3321                         hardware_disable_all_nolock();
3322                         r = -EBUSY;
3323                 }
3324         }
3325
3326         raw_spin_unlock(&kvm_count_lock);
3327
3328         return r;
3329 }
3330
3331 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3332                       void *v)
3333 {
3334         /*
3335          * Some (well, at least mine) BIOSes hang on reboot if
3336          * in vmx root mode.
3337          *
3338          * And Intel TXT required VMX off for all cpu when system shutdown.
3339          */
3340         pr_info("kvm: exiting hardware virtualization\n");
3341         kvm_rebooting = true;
3342         on_each_cpu(hardware_disable_nolock, NULL, 1);
3343         return NOTIFY_OK;
3344 }
3345
3346 static struct notifier_block kvm_reboot_notifier = {
3347         .notifier_call = kvm_reboot,
3348         .priority = 0,
3349 };
3350
3351 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3352 {
3353         int i;
3354
3355         for (i = 0; i < bus->dev_count; i++) {
3356                 struct kvm_io_device *pos = bus->range[i].dev;
3357
3358                 kvm_iodevice_destructor(pos);
3359         }
3360         kfree(bus);
3361 }
3362
3363 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
3364                                  const struct kvm_io_range *r2)
3365 {
3366         gpa_t addr1 = r1->addr;
3367         gpa_t addr2 = r2->addr;
3368
3369         if (addr1 < addr2)
3370                 return -1;
3371
3372         /* If r2->len == 0, match the exact address.  If r2->len != 0,
3373          * accept any overlapping write.  Any order is acceptable for
3374          * overlapping ranges, because kvm_io_bus_get_first_dev ensures
3375          * we process all of them.
3376          */
3377         if (r2->len) {
3378                 addr1 += r1->len;
3379                 addr2 += r2->len;
3380         }
3381
3382         if (addr1 > addr2)
3383                 return 1;
3384
3385         return 0;
3386 }
3387
3388 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
3389 {
3390         return kvm_io_bus_cmp(p1, p2);
3391 }
3392
3393 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
3394                           gpa_t addr, int len)
3395 {
3396         bus->range[bus->dev_count++] = (struct kvm_io_range) {
3397                 .addr = addr,
3398                 .len = len,
3399                 .dev = dev,
3400         };
3401
3402         sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
3403                 kvm_io_bus_sort_cmp, NULL);
3404
3405         return 0;
3406 }
3407
3408 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
3409                              gpa_t addr, int len)
3410 {
3411         struct kvm_io_range *range, key;
3412         int off;
3413
3414         key = (struct kvm_io_range) {
3415                 .addr = addr,
3416                 .len = len,
3417         };
3418
3419         range = bsearch(&key, bus->range, bus->dev_count,
3420                         sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
3421         if (range == NULL)
3422                 return -ENOENT;
3423
3424         off = range - bus->range;
3425
3426         while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
3427                 off--;
3428
3429         return off;
3430 }
3431
3432 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3433                               struct kvm_io_range *range, const void *val)
3434 {
3435         int idx;
3436
3437         idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3438         if (idx < 0)
3439                 return -EOPNOTSUPP;
3440
3441         while (idx < bus->dev_count &&
3442                 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3443                 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
3444                                         range->len, val))
3445                         return idx;
3446                 idx++;
3447         }
3448
3449         return -EOPNOTSUPP;
3450 }
3451
3452 /* kvm_io_bus_write - called under kvm->slots_lock */
3453 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3454                      int len, const void *val)
3455 {
3456         struct kvm_io_bus *bus;
3457         struct kvm_io_range range;
3458         int r;
3459
3460         range = (struct kvm_io_range) {
3461                 .addr = addr,
3462                 .len = len,
3463         };
3464
3465         bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3466         if (!bus)
3467                 return -ENOMEM;
3468         r = __kvm_io_bus_write(vcpu, bus, &range, val);
3469         return r < 0 ? r : 0;
3470 }
3471
3472 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3473 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
3474                             gpa_t addr, int len, const void *val, long cookie)
3475 {
3476         struct kvm_io_bus *bus;
3477         struct kvm_io_range range;
3478
3479         range = (struct kvm_io_range) {
3480                 .addr = addr,
3481                 .len = len,
3482         };
3483
3484         bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3485         if (!bus)
3486                 return -ENOMEM;
3487
3488         /* First try the device referenced by cookie. */
3489         if ((cookie >= 0) && (cookie < bus->dev_count) &&
3490             (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
3491                 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
3492                                         val))
3493                         return cookie;
3494
3495         /*
3496          * cookie contained garbage; fall back to search and return the
3497          * correct cookie value.
3498          */
3499         return __kvm_io_bus_write(vcpu, bus, &range, val);
3500 }
3501
3502 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3503                              struct kvm_io_range *range, void *val)
3504 {
3505         int idx;
3506
3507         idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3508         if (idx < 0)
3509                 return -EOPNOTSUPP;
3510
3511         while (idx < bus->dev_count &&
3512                 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3513                 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
3514                                        range->len, val))
3515                         return idx;
3516                 idx++;
3517         }
3518
3519         return -EOPNOTSUPP;
3520 }
3521 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
3522
3523 /* kvm_io_bus_read - called under kvm->slots_lock */
3524 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3525                     int len, void *val)
3526 {
3527         struct kvm_io_bus *bus;
3528         struct kvm_io_range range;
3529         int r;
3530
3531         range = (struct kvm_io_range) {
3532                 .addr = addr,
3533                 .len = len,
3534         };
3535
3536         bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3537         if (!bus)
3538                 return -ENOMEM;
3539         r = __kvm_io_bus_read(vcpu, bus, &range, val);
3540         return r < 0 ? r : 0;
3541 }
3542
3543
3544 /* Caller must hold slots_lock. */
3545 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3546                             int len, struct kvm_io_device *dev)
3547 {
3548         struct kvm_io_bus *new_bus, *bus;
3549
3550         bus = kvm_get_bus(kvm, bus_idx);
3551         if (!bus)
3552                 return -ENOMEM;
3553
3554         /* exclude ioeventfd which is limited by maximum fd */
3555         if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3556                 return -ENOSPC;
3557
3558         new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count + 1) *
3559                           sizeof(struct kvm_io_range)), GFP_KERNEL);
3560         if (!new_bus)
3561                 return -ENOMEM;
3562         memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
3563                sizeof(struct kvm_io_range)));
3564         kvm_io_bus_insert_dev(new_bus, dev, addr, len);
3565         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3566         synchronize_srcu_expedited(&kvm->srcu);
3567         kfree(bus);
3568
3569         return 0;
3570 }
3571
3572 /* Caller must hold slots_lock. */
3573 void kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3574                                struct kvm_io_device *dev)
3575 {
3576         int i;
3577         struct kvm_io_bus *new_bus, *bus;
3578
3579         bus = kvm_get_bus(kvm, bus_idx);
3580         if (!bus)
3581                 return;
3582
3583         for (i = 0; i < bus->dev_count; i++)
3584                 if (bus->range[i].dev == dev) {
3585                         break;
3586                 }
3587
3588         if (i == bus->dev_count)
3589                 return;
3590
3591         new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3592                           sizeof(struct kvm_io_range)), GFP_KERNEL);
3593         if (!new_bus)  {
3594                 pr_err("kvm: failed to shrink bus, removing it completely\n");
3595                 goto broken;
3596         }
3597
3598         memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3599         new_bus->dev_count--;
3600         memcpy(new_bus->range + i, bus->range + i + 1,
3601                (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3602
3603 broken:
3604         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3605         synchronize_srcu_expedited(&kvm->srcu);
3606         kfree(bus);
3607         return;
3608 }
3609
3610 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3611                                          gpa_t addr)
3612 {
3613         struct kvm_io_bus *bus;
3614         int dev_idx, srcu_idx;
3615         struct kvm_io_device *iodev = NULL;
3616
3617         srcu_idx = srcu_read_lock(&kvm->srcu);
3618
3619         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
3620         if (!bus)
3621                 goto out_unlock;
3622
3623         dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
3624         if (dev_idx < 0)
3625                 goto out_unlock;
3626
3627         iodev = bus->range[dev_idx].dev;
3628
3629 out_unlock:
3630         srcu_read_unlock(&kvm->srcu, srcu_idx);
3631
3632         return iodev;
3633 }
3634 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
3635
3636 static int kvm_debugfs_open(struct inode *inode, struct file *file,
3637                            int (*get)(void *, u64 *), int (*set)(void *, u64),
3638                            const char *fmt)
3639 {
3640         struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
3641                                           inode->i_private;
3642
3643         /* The debugfs files are a reference to the kvm struct which
3644          * is still valid when kvm_destroy_vm is called.
3645          * To avoid the race between open and the removal of the debugfs
3646          * directory we test against the users count.
3647          */
3648         if (!refcount_inc_not_zero(&stat_data->kvm->users_count))
3649                 return -ENOENT;
3650
3651         if (simple_attr_open(inode, file, get, set, fmt)) {
3652                 kvm_put_kvm(stat_data->kvm);
3653                 return -ENOMEM;
3654         }
3655
3656         return 0;
3657 }
3658
3659 static int kvm_debugfs_release(struct inode *inode, struct file *file)
3660 {
3661         struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
3662                                           inode->i_private;
3663
3664         simple_attr_release(inode, file);
3665         kvm_put_kvm(stat_data->kvm);
3666
3667         return 0;
3668 }
3669
3670 static int vm_stat_get_per_vm(void *data, u64 *val)
3671 {
3672         struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3673
3674         *val = *(ulong *)((void *)stat_data->kvm + stat_data->offset);
3675
3676         return 0;
3677 }
3678
3679 static int vm_stat_clear_per_vm(void *data, u64 val)
3680 {
3681         struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3682
3683         if (val)
3684                 return -EINVAL;
3685
3686         *(ulong *)((void *)stat_data->kvm + stat_data->offset) = 0;
3687
3688         return 0;
3689 }
3690
3691 static int vm_stat_get_per_vm_open(struct inode *inode, struct file *file)
3692 {
3693         __simple_attr_check_format("%llu\n", 0ull);
3694         return kvm_debugfs_open(inode, file, vm_stat_get_per_vm,
3695                                 vm_stat_clear_per_vm, "%llu\n");
3696 }
3697
3698 static const struct file_operations vm_stat_get_per_vm_fops = {
3699         .owner   = THIS_MODULE,
3700         .open    = vm_stat_get_per_vm_open,
3701         .release = kvm_debugfs_release,
3702         .read    = simple_attr_read,
3703         .write   = simple_attr_write,
3704         .llseek  = no_llseek,
3705 };
3706
3707 static int vcpu_stat_get_per_vm(void *data, u64 *val)
3708 {
3709         int i;
3710         struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3711         struct kvm_vcpu *vcpu;
3712
3713         *val = 0;
3714
3715         kvm_for_each_vcpu(i, vcpu, stat_data->kvm)
3716                 *val += *(u64 *)((void *)vcpu + stat_data->offset);
3717
3718         return 0;
3719 }
3720
3721 static int vcpu_stat_clear_per_vm(void *data, u64 val)
3722 {
3723         int i;
3724         struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3725         struct kvm_vcpu *vcpu;
3726
3727         if (val)
3728                 return -EINVAL;
3729
3730         kvm_for_each_vcpu(i, vcpu, stat_data->kvm)
3731                 *(u64 *)((void *)vcpu + stat_data->offset) = 0;
3732
3733         return 0;
3734 }
3735
3736 static int vcpu_stat_get_per_vm_open(struct inode *inode, struct file *file)
3737 {
3738         __simple_attr_check_format("%llu\n", 0ull);
3739         return kvm_debugfs_open(inode, file, vcpu_stat_get_per_vm,
3740                                  vcpu_stat_clear_per_vm, "%llu\n");
3741 }
3742
3743 static const struct file_operations vcpu_stat_get_per_vm_fops = {
3744         .owner   = THIS_MODULE,
3745         .open    = vcpu_stat_get_per_vm_open,
3746         .release = kvm_debugfs_release,
3747         .read    = simple_attr_read,
3748         .write   = simple_attr_write,
3749         .llseek  = no_llseek,
3750 };
3751
3752 static const struct file_operations *stat_fops_per_vm[] = {
3753         [KVM_STAT_VCPU] = &vcpu_stat_get_per_vm_fops,
3754         [KVM_STAT_VM]   = &vm_stat_get_per_vm_fops,
3755 };
3756
3757 static int vm_stat_get(void *_offset, u64 *val)
3758 {
3759         unsigned offset = (long)_offset;
3760         struct kvm *kvm;
3761         struct kvm_stat_data stat_tmp = {.offset = offset};
3762         u64 tmp_val;
3763
3764         *val = 0;
3765         spin_lock(&kvm_lock);
3766         list_for_each_entry(kvm, &vm_list, vm_list) {
3767                 stat_tmp.kvm = kvm;
3768                 vm_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
3769                 *val += tmp_val;
3770         }
3771         spin_unlock(&kvm_lock);
3772         return 0;
3773 }
3774
3775 static int vm_stat_clear(void *_offset, u64 val)
3776 {
3777         unsigned offset = (long)_offset;
3778         struct kvm *kvm;
3779         struct kvm_stat_data stat_tmp = {.offset = offset};
3780
3781         if (val)
3782                 return -EINVAL;
3783
3784         spin_lock(&kvm_lock);
3785         list_for_each_entry(kvm, &vm_list, vm_list) {
3786                 stat_tmp.kvm = kvm;
3787                 vm_stat_clear_per_vm((void *)&stat_tmp, 0);
3788         }
3789         spin_unlock(&kvm_lock);
3790
3791         return 0;
3792 }
3793
3794 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n");
3795
3796 static int vcpu_stat_get(void *_offset, u64 *val)
3797 {
3798         unsigned offset = (long)_offset;
3799         struct kvm *kvm;
3800         struct kvm_stat_data stat_tmp = {.offset = offset};
3801         u64 tmp_val;
3802
3803         *val = 0;
3804         spin_lock(&kvm_lock);
3805         list_for_each_entry(kvm, &vm_list, vm_list) {
3806                 stat_tmp.kvm = kvm;
3807                 vcpu_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
3808                 *val += tmp_val;
3809         }
3810         spin_unlock(&kvm_lock);
3811         return 0;
3812 }
3813
3814 static int vcpu_stat_clear(void *_offset, u64 val)
3815 {
3816         unsigned offset = (long)_offset;
3817         struct kvm *kvm;
3818         struct kvm_stat_data stat_tmp = {.offset = offset};
3819
3820         if (val)
3821                 return -EINVAL;
3822
3823         spin_lock(&kvm_lock);
3824         list_for_each_entry(kvm, &vm_list, vm_list) {
3825                 stat_tmp.kvm = kvm;
3826                 vcpu_stat_clear_per_vm((void *)&stat_tmp, 0);
3827         }
3828         spin_unlock(&kvm_lock);
3829
3830         return 0;
3831 }
3832
3833 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear,
3834                         "%llu\n");
3835
3836 static const struct file_operations *stat_fops[] = {
3837         [KVM_STAT_VCPU] = &vcpu_stat_fops,
3838         [KVM_STAT_VM]   = &vm_stat_fops,
3839 };
3840
3841 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm)
3842 {
3843         struct kobj_uevent_env *env;
3844         unsigned long long created, active;
3845
3846         if (!kvm_dev.this_device || !kvm)
3847                 return;
3848
3849         spin_lock(&kvm_lock);
3850         if (type == KVM_EVENT_CREATE_VM) {
3851                 kvm_createvm_count++;
3852                 kvm_active_vms++;
3853         } else if (type == KVM_EVENT_DESTROY_VM) {
3854                 kvm_active_vms--;
3855         }
3856         created = kvm_createvm_count;
3857         active = kvm_active_vms;
3858         spin_unlock(&kvm_lock);
3859
3860         env = kzalloc(sizeof(*env), GFP_KERNEL);
3861         if (!env)
3862                 return;
3863
3864         add_uevent_var(env, "CREATED=%llu", created);
3865         add_uevent_var(env, "COUNT=%llu", active);
3866
3867         if (type == KVM_EVENT_CREATE_VM) {
3868                 add_uevent_var(env, "EVENT=create");
3869                 kvm->userspace_pid = task_pid_nr(current);
3870         } else if (type == KVM_EVENT_DESTROY_VM) {
3871                 add_uevent_var(env, "EVENT=destroy");
3872         }
3873         add_uevent_var(env, "PID=%d", kvm->userspace_pid);
3874
3875         if (kvm->debugfs_dentry) {
3876                 char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL);
3877
3878                 if (p) {
3879                         tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX);
3880                         if (!IS_ERR(tmp))
3881                                 add_uevent_var(env, "STATS_PATH=%s", tmp);
3882                         kfree(p);
3883                 }
3884         }
3885         /* no need for checks, since we are adding at most only 5 keys */
3886         env->envp[env->envp_idx++] = NULL;
3887         kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp);
3888         kfree(env);
3889 }
3890
3891 static int kvm_init_debug(void)
3892 {
3893         int r = -EEXIST;
3894         struct kvm_stats_debugfs_item *p;
3895
3896         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3897         if (kvm_debugfs_dir == NULL)
3898                 goto out;
3899
3900         kvm_debugfs_num_entries = 0;
3901         for (p = debugfs_entries; p->name; ++p, kvm_debugfs_num_entries++) {
3902                 if (!debugfs_create_file(p->name, 0644, kvm_debugfs_dir,
3903                                          (void *)(long)p->offset,
3904                                          stat_fops[p->kind]))
3905                         goto out_dir;
3906         }
3907
3908         return 0;
3909
3910 out_dir:
3911         debugfs_remove_recursive(kvm_debugfs_dir);
3912 out:
3913         return r;
3914 }
3915
3916 static int kvm_suspend(void)
3917 {
3918         if (kvm_usage_count)
3919                 hardware_disable_nolock(NULL);
3920         return 0;
3921 }
3922
3923 static void kvm_resume(void)
3924 {
3925         if (kvm_usage_count) {
3926                 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3927                 hardware_enable_nolock(NULL);
3928         }
3929 }
3930
3931 static struct syscore_ops kvm_syscore_ops = {
3932         .suspend = kvm_suspend,
3933         .resume = kvm_resume,
3934 };
3935
3936 static inline
3937 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3938 {
3939         return container_of(pn, struct kvm_vcpu, preempt_notifier);
3940 }
3941
3942 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3943 {
3944         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3945
3946         if (vcpu->preempted)
3947                 vcpu->preempted = false;
3948
3949         kvm_arch_sched_in(vcpu, cpu);
3950
3951         kvm_arch_vcpu_load(vcpu, cpu);
3952 }
3953
3954 static void kvm_sched_out(struct preempt_notifier *pn,
3955                           struct task_struct *next)
3956 {
3957         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3958
3959         if (current->state == TASK_RUNNING)
3960                 vcpu->preempted = true;
3961         kvm_arch_vcpu_put(vcpu);
3962 }
3963
3964 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3965                   struct module *module)
3966 {
3967         int r;
3968         int cpu;
3969
3970         r = kvm_arch_init(opaque);
3971         if (r)
3972                 goto out_fail;
3973
3974         /*
3975          * kvm_arch_init makes sure there's at most one caller
3976          * for architectures that support multiple implementations,
3977          * like intel and amd on x86.
3978          * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3979          * conflicts in case kvm is already setup for another implementation.
3980          */
3981         r = kvm_irqfd_init();
3982         if (r)
3983                 goto out_irqfd;
3984
3985         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
3986                 r = -ENOMEM;
3987                 goto out_free_0;
3988         }
3989
3990         r = kvm_arch_hardware_setup();
3991         if (r < 0)
3992                 goto out_free_0a;
3993
3994         for_each_online_cpu(cpu) {
3995                 smp_call_function_single(cpu,
3996                                 kvm_arch_check_processor_compat,
3997                                 &r, 1);
3998                 if (r < 0)
3999                         goto out_free_1;
4000         }
4001
4002         r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting",
4003                                       kvm_starting_cpu, kvm_dying_cpu);
4004         if (r)
4005                 goto out_free_2;
4006         register_reboot_notifier(&kvm_reboot_notifier);
4007
4008         /* A kmem cache lets us meet the alignment requirements of fx_save. */
4009         if (!vcpu_align)
4010                 vcpu_align = __alignof__(struct kvm_vcpu);
4011         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
4012                                            0, NULL);
4013         if (!kvm_vcpu_cache) {
4014                 r = -ENOMEM;
4015                 goto out_free_3;
4016         }
4017
4018         r = kvm_async_pf_init();
4019         if (r)
4020                 goto out_free;
4021
4022         kvm_chardev_ops.owner = module;
4023         kvm_vm_fops.owner = module;
4024         kvm_vcpu_fops.owner = module;
4025
4026         r = misc_register(&kvm_dev);
4027         if (r) {
4028                 pr_err("kvm: misc device register failed\n");
4029                 goto out_unreg;
4030         }
4031
4032         register_syscore_ops(&kvm_syscore_ops);
4033
4034         kvm_preempt_ops.sched_in = kvm_sched_in;
4035         kvm_preempt_ops.sched_out = kvm_sched_out;
4036
4037         r = kvm_init_debug();
4038         if (r) {
4039                 pr_err("kvm: create debugfs files failed\n");
4040                 goto out_undebugfs;
4041         }
4042
4043         r = kvm_vfio_ops_init();
4044         WARN_ON(r);
4045
4046         return 0;
4047
4048 out_undebugfs:
4049         unregister_syscore_ops(&kvm_syscore_ops);
4050         misc_deregister(&kvm_dev);
4051 out_unreg:
4052         kvm_async_pf_deinit();
4053 out_free:
4054         kmem_cache_destroy(kvm_vcpu_cache);
4055 out_free_3:
4056         unregister_reboot_notifier(&kvm_reboot_notifier);
4057         cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4058 out_free_2:
4059 out_free_1:
4060         kvm_arch_hardware_unsetup();
4061 out_free_0a:
4062         free_cpumask_var(cpus_hardware_enabled);
4063 out_free_0:
4064         kvm_irqfd_exit();
4065 out_irqfd:
4066         kvm_arch_exit();
4067 out_fail:
4068         return r;
4069 }
4070 EXPORT_SYMBOL_GPL(kvm_init);
4071
4072 void kvm_exit(void)
4073 {
4074         debugfs_remove_recursive(kvm_debugfs_dir);
4075         misc_deregister(&kvm_dev);
4076         kmem_cache_destroy(kvm_vcpu_cache);
4077         kvm_async_pf_deinit();
4078         unregister_syscore_ops(&kvm_syscore_ops);
4079         unregister_reboot_notifier(&kvm_reboot_notifier);
4080         cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4081         on_each_cpu(hardware_disable_nolock, NULL, 1);
4082         kvm_arch_hardware_unsetup();
4083         kvm_arch_exit();
4084         kvm_irqfd_exit();
4085         free_cpumask_var(cpus_hardware_enabled);
4086         kvm_vfio_ops_exit();
4087 }
4088 EXPORT_SYMBOL_GPL(kvm_exit);