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.h>
  36 #include <linux/cpumask.h>
  37 #include <linux/smp.h>
  38 #include <linux/anon_inodes.h>
  39 #include <linux/profile.h>
  40 #include <linux/kvm_para.h>
  41 #include <linux/pagemap.h>
  42 #include <linux/mman.h>
  43 #include <linux/swap.h>
  44 #include <linux/bitops.h>
  45 #include <linux/spinlock.h>
  46 #include <linux/compat.h>
  47 #include <linux/srcu.h>
  48 #include <linux/hugetlb.h>
  49 #include <linux/slab.h>
  50 #include <linux/sort.h>
  51 #include <linux/bsearch.h>
  52
  53 #include <asm/processor.h>
  54 #include <asm/io.h>
  55 #include <asm/ioctl.h>
  56 #include <asm/uaccess.h>
  57 #include <asm/pgtable.h>
  58
  59 #include "coalesced_mmio.h"
  60 #include "async_pf.h"
  61 #include "vfio.h"
  62
  63 #define CREATE_TRACE_POINTS
  64 #include <trace/events/kvm.h>
  65
  66 /* Worst case buffer size needed for holding an integer. */
  67 #define ITOA_MAX_LEN 12
  68
  69 MODULE_AUTHOR("Qumranet");
  70 MODULE_LICENSE("GPL");
  71
  72 /* Architectures should define their poll value according to the halt latency */
  73 static unsigned int halt_poll_ns = KVM_HALT_POLL_NS_DEFAULT;
  74 module_param(halt_poll_ns, uint, S_IRUGO | S_IWUSR);
  75
  76 /* Default doubles per-vcpu halt_poll_ns. */
  77 static unsigned int halt_poll_ns_grow = 2;
  78 module_param(halt_poll_ns_grow, uint, S_IRUGO | S_IWUSR);
  79
  80 /* Default resets per-vcpu halt_poll_ns . */
  81 static unsigned int halt_poll_ns_shrink;
  82 module_param(halt_poll_ns_shrink, uint, S_IRUGO | S_IWUSR);
  83
  84 /*
  85  * Ordering of locks:
  86  *
  87  *      kvm->lock --> kvm->slots_lock --> kvm->irq_lock
  88  */
  89
  90 DEFINE_SPINLOCK(kvm_lock);
  91 static DEFINE_RAW_SPINLOCK(kvm_count_lock);
  92 LIST_HEAD(vm_list);
  93
  94 static cpumask_var_t cpus_hardware_enabled;
  95 static int kvm_usage_count;
  96 static atomic_t hardware_enable_failed;
  97
  98 struct kmem_cache *kvm_vcpu_cache;
  99 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
 100
 101 static __read_mostly struct preempt_ops kvm_preempt_ops;
 102
 103 struct dentry *kvm_debugfs_dir;
 104 EXPORT_SYMBOL_GPL(kvm_debugfs_dir);
 105
 106 static int kvm_debugfs_num_entries;
 107 static const struct file_operations *stat_fops_per_vm[];
 108
 109 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
 110                            unsigned long arg);
 111 #ifdef CONFIG_KVM_COMPAT
 112 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
 113                                   unsigned long arg);
 114 #endif
 115 static int hardware_enable_all(void);
 116 static void hardware_disable_all(void);
 117
 118 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
 119
 120 static void kvm_release_pfn_dirty(kvm_pfn_t pfn);
 121 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot, gfn_t gfn);
 122
 123 __visible bool kvm_rebooting;
 124 EXPORT_SYMBOL_GPL(kvm_rebooting);
 125
 126 static bool largepages_enabled = true;
 127
 128 bool kvm_is_reserved_pfn(kvm_pfn_t pfn)
 129 {
 130         if (pfn_valid(pfn))
 131                 return PageReserved(pfn_to_page(pfn));
 132
 133         return true;
 134 }
 135
 136 /*
 137  * Switches to specified vcpu, until a matching vcpu_put()
 138  */
 139 int vcpu_load(struct kvm_vcpu *vcpu)
 140 {
 141         int cpu;
 142
 143         if (mutex_lock_killable(&vcpu->mutex))
 144                 return -EINTR;
 145         cpu = get_cpu();
 146         preempt_notifier_register(&vcpu->preempt_notifier);
 147         kvm_arch_vcpu_load(vcpu, cpu);
 148         put_cpu();
 149         return 0;
 150 }
 151
 152 void vcpu_put(struct kvm_vcpu *vcpu)
 153 {
 154         preempt_disable();
 155         kvm_arch_vcpu_put(vcpu);
 156         preempt_notifier_unregister(&vcpu->preempt_notifier);
 157         preempt_enable();
 158         mutex_unlock(&vcpu->mutex);
 159 }
 160
 161 static void ack_flush(void *_completed)
 162 {
 163 }
 164
 165 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
 166 {
 167         int i, cpu, me;
 168         cpumask_var_t cpus;
 169         bool called = true;
 170         struct kvm_vcpu *vcpu;
 171
 172         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
 173
 174         me = get_cpu();
 175         kvm_for_each_vcpu(i, vcpu, kvm) {
 176                 kvm_make_request(req, vcpu);
 177                 cpu = vcpu->cpu;
 178
 179                 /* Set ->requests bit before we read ->mode. */
 180                 smp_mb__after_atomic();
 181
 182                 if (cpus != NULL && cpu != -1 && cpu != me &&
 183                       kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
 184                         cpumask_set_cpu(cpu, cpus);
 185         }
 186         if (unlikely(cpus == NULL))
 187                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
 188         else if (!cpumask_empty(cpus))
 189                 smp_call_function_many(cpus, ack_flush, NULL, 1);
 190         else
 191                 called = false;
 192         put_cpu();
 193         free_cpumask_var(cpus);
 194         return called;
 195 }
 196
 197 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
 198 void kvm_flush_remote_tlbs(struct kvm *kvm)
 199 {
 200         /*
 201          * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in
 202          * kvm_make_all_cpus_request.
 203          */
 204         long dirty_count = smp_load_acquire(&kvm->tlbs_dirty);
 205
 206         /*
 207          * We want to publish modifications to the page tables before reading
 208          * mode. Pairs with a memory barrier in arch-specific code.
 209          * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
 210          * and smp_mb in walk_shadow_page_lockless_begin/end.
 211          * - powerpc: smp_mb in kvmppc_prepare_to_enter.
 212          *
 213          * There is already an smp_mb__after_atomic() before
 214          * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
 215          * barrier here.
 216          */
 217         if (kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
 218                 ++kvm->stat.remote_tlb_flush;
 219         cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
 220 }
 221 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
 222 #endif
 223
 224 void kvm_reload_remote_mmus(struct kvm *kvm)
 225 {
 226         kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
 227 }
 228
 229 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 230 {
 231         struct page *page;
 232         int r;
 233
 234         mutex_init(&vcpu->mutex);
 235         vcpu->cpu = -1;
 236         vcpu->kvm = kvm;
 237         vcpu->vcpu_id = id;
 238         vcpu->pid = NULL;
 239         init_swait_queue_head(&vcpu->wq);
 240         kvm_async_pf_vcpu_init(vcpu);
 241
 242         vcpu->pre_pcpu = -1;
 243         INIT_LIST_HEAD(&vcpu->blocked_vcpu_list);
 244
 245         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 246         if (!page) {
 247                 r = -ENOMEM;
 248                 goto fail;
 249         }
 250         vcpu->run = page_address(page);
 251
 252         kvm_vcpu_set_in_spin_loop(vcpu, false);
 253         kvm_vcpu_set_dy_eligible(vcpu, false);
 254         vcpu->preempted = false;
 255
 256         r = kvm_arch_vcpu_init(vcpu);
 257         if (r < 0)
 258                 goto fail_free_run;
 259         return 0;
 260
 261 fail_free_run:
 262         free_page((unsigned long)vcpu->run);
 263 fail:
 264         return r;
 265 }
 266 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 267
 268 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 269 {
 270         put_pid(vcpu->pid);
 271         kvm_arch_vcpu_uninit(vcpu);
 272         free_page((unsigned long)vcpu->run);
 273 }
 274 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 275
 276 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 277 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
 278 {
 279         return container_of(mn, struct kvm, mmu_notifier);
 280 }
 281
 282 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
 283                                              struct mm_struct *mm,
 284                                              unsigned long address)
 285 {
 286         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 287         int need_tlb_flush, idx;
 288
 289         /*
 290          * When ->invalidate_page runs, the linux pte has been zapped
 291          * already but the page is still allocated until
 292          * ->invalidate_page returns. So if we increase the sequence
 293          * here the kvm page fault will notice if the spte can't be
 294          * established because the page is going to be freed. If
 295          * instead the kvm page fault establishes the spte before
 296          * ->invalidate_page runs, kvm_unmap_hva will release it
 297          * before returning.
 298          *
 299          * The sequence increase only need to be seen at spin_unlock
 300          * time, and not at spin_lock time.
 301          *
 302          * Increasing the sequence after the spin_unlock would be
 303          * unsafe because the kvm page fault could then establish the
 304          * pte after kvm_unmap_hva returned, without noticing the page
 305          * is going to be freed.
 306          */
 307         idx = srcu_read_lock(&kvm->srcu);
 308         spin_lock(&kvm->mmu_lock);
 309
 310         kvm->mmu_notifier_seq++;
 311         need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
 312         /* we've to flush the tlb before the pages can be freed */
 313         if (need_tlb_flush)
 314                 kvm_flush_remote_tlbs(kvm);
 315
 316         spin_unlock(&kvm->mmu_lock);
 317
 318         kvm_arch_mmu_notifier_invalidate_page(kvm, address);
 319
 320         srcu_read_unlock(&kvm->srcu, idx);
 321 }
 322
 323 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
 324                                         struct mm_struct *mm,
 325                                         unsigned long address,
 326                                         pte_t pte)
 327 {
 328         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 329         int idx;
 330
 331         idx = srcu_read_lock(&kvm->srcu);
 332         spin_lock(&kvm->mmu_lock);
 333         kvm->mmu_notifier_seq++;
 334         kvm_set_spte_hva(kvm, address, pte);
 335         spin_unlock(&kvm->mmu_lock);
 336         srcu_read_unlock(&kvm->srcu, idx);
 337 }
 338
 339 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 340                                                     struct mm_struct *mm,
 341                                                     unsigned long start,
 342                                                     unsigned long end)
 343 {
 344         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 345         int need_tlb_flush = 0, idx;
 346
 347         idx = srcu_read_lock(&kvm->srcu);
 348         spin_lock(&kvm->mmu_lock);
 349         /*
 350          * The count increase must become visible at unlock time as no
 351          * spte can be established without taking the mmu_lock and
 352          * count is also read inside the mmu_lock critical section.
 353          */
 354         kvm->mmu_notifier_count++;
 355         need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
 356         need_tlb_flush |= kvm->tlbs_dirty;
 357         /* we've to flush the tlb before the pages can be freed */
 358         if (need_tlb_flush)
 359                 kvm_flush_remote_tlbs(kvm);
 360
 361         spin_unlock(&kvm->mmu_lock);
 362         srcu_read_unlock(&kvm->srcu, idx);
 363 }
 364
 365 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
 366                                                   struct mm_struct *mm,
 367                                                   unsigned long start,
 368                                                   unsigned long end)
 369 {
 370         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 371
 372         spin_lock(&kvm->mmu_lock);
 373         /*
 374          * This sequence increase will notify the kvm page fault that
 375          * the page that is going to be mapped in the spte could have
 376          * been freed.
 377          */
 378         kvm->mmu_notifier_seq++;
 379         smp_wmb();
 380         /*
 381          * The above sequence increase must be visible before the
 382          * below count decrease, which is ensured by the smp_wmb above
 383          * in conjunction with the smp_rmb in mmu_notifier_retry().
 384          */
 385         kvm->mmu_notifier_count--;
 386         spin_unlock(&kvm->mmu_lock);
 387
 388         BUG_ON(kvm->mmu_notifier_count < 0);
 389 }
 390
 391 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
 392                                               struct mm_struct *mm,
 393                                               unsigned long start,
 394                                               unsigned long end)
 395 {
 396         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 397         int young, idx;
 398
 399         idx = srcu_read_lock(&kvm->srcu);
 400         spin_lock(&kvm->mmu_lock);
 401
 402         young = kvm_age_hva(kvm, start, end);
 403         if (young)
 404                 kvm_flush_remote_tlbs(kvm);
 405
 406         spin_unlock(&kvm->mmu_lock);
 407         srcu_read_unlock(&kvm->srcu, idx);
 408
 409         return young;
 410 }
 411
 412 static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
 413                                         struct mm_struct *mm,
 414                                         unsigned long start,
 415                                         unsigned long end)
 416 {
 417         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 418         int young, idx;
 419
 420         idx = srcu_read_lock(&kvm->srcu);
 421         spin_lock(&kvm->mmu_lock);
 422         /*
 423          * Even though we do not flush TLB, this will still adversely
 424          * affect performance on pre-Haswell Intel EPT, where there is
 425          * no EPT Access Bit to clear so that we have to tear down EPT
 426          * tables instead. If we find this unacceptable, we can always
 427          * add a parameter to kvm_age_hva so that it effectively doesn't
 428          * do anything on clear_young.
 429          *
 430          * Also note that currently we never issue secondary TLB flushes
 431          * from clear_young, leaving this job up to the regular system
 432          * cadence. If we find this inaccurate, we might come up with a
 433          * more sophisticated heuristic later.
 434          */
 435         young = kvm_age_hva(kvm, start, end);
 436         spin_unlock(&kvm->mmu_lock);
 437         srcu_read_unlock(&kvm->srcu, idx);
 438
 439         return young;
 440 }
 441
 442 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
 443                                        struct mm_struct *mm,
 444                                        unsigned long address)
 445 {
 446         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 447         int young, idx;
 448
 449         idx = srcu_read_lock(&kvm->srcu);
 450         spin_lock(&kvm->mmu_lock);
 451         young = kvm_test_age_hva(kvm, address);
 452         spin_unlock(&kvm->mmu_lock);
 453         srcu_read_unlock(&kvm->srcu, idx);
 454
 455         return young;
 456 }
 457
 458 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
 459                                      struct mm_struct *mm)
 460 {
 461         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 462         int idx;
 463
 464         idx = srcu_read_lock(&kvm->srcu);
 465         kvm_arch_flush_shadow_all(kvm);
 466         srcu_read_unlock(&kvm->srcu, idx);
 467 }
 468
 469 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 470         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
 471         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
 472         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
 473         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
 474         .clear_young            = kvm_mmu_notifier_clear_young,
 475         .test_young             = kvm_mmu_notifier_test_young,
 476         .change_pte             = kvm_mmu_notifier_change_pte,
 477         .release                = kvm_mmu_notifier_release,
 478 };
 479
 480 static int kvm_init_mmu_notifier(struct kvm *kvm)
 481 {
 482         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
 483         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
 484 }
 485
 486 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
 487
 488 static int kvm_init_mmu_notifier(struct kvm *kvm)
 489 {
 490         return 0;
 491 }
 492
 493 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
 494
 495 static struct kvm_memslots *kvm_alloc_memslots(void)
 496 {
 497         int i;
 498         struct kvm_memslots *slots;
 499
 500         slots = kvm_kvzalloc(sizeof(struct kvm_memslots));
 501         if (!slots)
 502                 return NULL;
 503
 504         /*
 505          * Init kvm generation close to the maximum to easily test the
 506          * code of handling generation number wrap-around.
 507          */
 508         slots->generation = -150;
 509         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
 510                 slots->id_to_index[i] = slots->memslots[i].id = i;
 511
 512         return slots;
 513 }
 514
 515 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
 516 {
 517         if (!memslot->dirty_bitmap)
 518                 return;
 519
 520         kvfree(memslot->dirty_bitmap);
 521         memslot->dirty_bitmap = NULL;
 522 }
 523
 524 /*
 525  * Free any memory in @free but not in @dont.
 526  */
 527 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
 528                               struct kvm_memory_slot *dont)
 529 {
 530         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 531                 kvm_destroy_dirty_bitmap(free);
 532
 533         kvm_arch_free_memslot(kvm, free, dont);
 534
 535         free->npages = 0;
 536 }
 537
 538 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
 539 {
 540         struct kvm_memory_slot *memslot;
 541
 542         if (!slots)
 543                 return;
 544
 545         kvm_for_each_memslot(memslot, slots)
 546                 kvm_free_memslot(kvm, memslot, NULL);
 547
 548         kvfree(slots);
 549 }
 550
 551 static void kvm_destroy_vm_debugfs(struct kvm *kvm)
 552 {
 553         int i;
 554
 555         if (!kvm->debugfs_dentry)
 556                 return;
 557
 558         debugfs_remove_recursive(kvm->debugfs_dentry);
 559
 560         for (i = 0; i < kvm_debugfs_num_entries; i++)
 561                 kfree(kvm->debugfs_stat_data[i]);
 562         kfree(kvm->debugfs_stat_data);
 563 }
 564
 565 static int kvm_create_vm_debugfs(struct kvm *kvm, int fd)
 566 {
 567         char dir_name[ITOA_MAX_LEN * 2];
 568         struct kvm_stat_data *stat_data;
 569         struct kvm_stats_debugfs_item *p;
 570
 571         if (!debugfs_initialized())
 572                 return 0;
 573
 574         snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd);
 575         kvm->debugfs_dentry = debugfs_create_dir(dir_name,
 576                                                  kvm_debugfs_dir);
 577         if (!kvm->debugfs_dentry)
 578                 return -ENOMEM;
 579
 580         kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries,
 581                                          sizeof(*kvm->debugfs_stat_data),
 582                                          GFP_KERNEL);
 583         if (!kvm->debugfs_stat_data)
 584                 return -ENOMEM;
 585
 586         for (p = debugfs_entries; p->name; p++) {
 587                 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL);
 588                 if (!stat_data)
 589                         return -ENOMEM;
 590
 591                 stat_data->kvm = kvm;
 592                 stat_data->offset = p->offset;
 593                 kvm->debugfs_stat_data[p - debugfs_entries] = stat_data;
 594                 if (!debugfs_create_file(p->name, 0444,
 595                                          kvm->debugfs_dentry,
 596                                          stat_data,
 597                                          stat_fops_per_vm[p->kind]))
 598                         return -ENOMEM;
 599         }
 600         return 0;
 601 }
 602
 603 static struct kvm *kvm_create_vm(unsigned long type)
 604 {
 605         int r, i;
 606         struct kvm *kvm = kvm_arch_alloc_vm();
 607
 608         if (!kvm)
 609                 return ERR_PTR(-ENOMEM);
 610
 611         spin_lock_init(&kvm->mmu_lock);
 612         atomic_inc(&current->mm->mm_count);
 613         kvm->mm = current->mm;
 614         kvm_eventfd_init(kvm);
 615         mutex_init(&kvm->lock);
 616         mutex_init(&kvm->irq_lock);
 617         mutex_init(&kvm->slots_lock);
 618         atomic_set(&kvm->users_count, 1);
 619         INIT_LIST_HEAD(&kvm->devices);
 620
 621         r = kvm_arch_init_vm(kvm, type);
 622         if (r)
 623                 goto out_err_no_disable;
 624
 625         r = hardware_enable_all();
 626         if (r)
 627                 goto out_err_no_disable;
 628
 629 #ifdef CONFIG_HAVE_KVM_IRQFD
 630         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
 631 #endif
 632
 633         BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
 634
 635         r = -ENOMEM;
 636         for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
 637                 kvm->memslots[i] = kvm_alloc_memslots();
 638                 if (!kvm->memslots[i])
 639                         goto out_err_no_srcu;
 640         }
 641
 642         if (init_srcu_struct(&kvm->srcu))
 643                 goto out_err_no_srcu;
 644         if (init_srcu_struct(&kvm->irq_srcu))
 645                 goto out_err_no_irq_srcu;
 646         for (i = 0; i < KVM_NR_BUSES; i++) {
 647                 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
 648                                         GFP_KERNEL);
 649                 if (!kvm->buses[i])
 650                         goto out_err;
 651         }
 652
 653         r = kvm_init_mmu_notifier(kvm);
 654         if (r)
 655                 goto out_err;
 656
 657         spin_lock(&kvm_lock);
 658         list_add(&kvm->vm_list, &vm_list);
 659         spin_unlock(&kvm_lock);
 660
 661         preempt_notifier_inc();
 662
 663         return kvm;
 664
 665 out_err:
 666         cleanup_srcu_struct(&kvm->irq_srcu);
 667 out_err_no_irq_srcu:
 668         cleanup_srcu_struct(&kvm->srcu);
 669 out_err_no_srcu:
 670         hardware_disable_all();
 671 out_err_no_disable:
 672         for (i = 0; i < KVM_NR_BUSES; i++)
 673                 kfree(kvm->buses[i]);
 674         for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
 675                 kvm_free_memslots(kvm, kvm->memslots[i]);
 676         kvm_arch_free_vm(kvm);
 677         mmdrop(current->mm);
 678         return ERR_PTR(r);
 679 }
 680
 681 /*
 682  * Avoid using vmalloc for a small buffer.
 683  * Should not be used when the size is statically known.
 684  */
 685 void *kvm_kvzalloc(unsigned long size)
 686 {
 687         if (size > PAGE_SIZE)
 688                 return vzalloc(size);
 689         else
 690                 return kzalloc(size, GFP_KERNEL);
 691 }
 692
 693 static void kvm_destroy_devices(struct kvm *kvm)
 694 {
 695         struct kvm_device *dev, *tmp;
 696
 697         list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
 698                 list_del(&dev->vm_node);
 699                 dev->ops->destroy(dev);
 700         }
 701 }
 702
 703 static void kvm_destroy_vm(struct kvm *kvm)
 704 {
 705         int i;
 706         struct mm_struct *mm = kvm->mm;
 707
 708         kvm_destroy_vm_debugfs(kvm);
 709         kvm_arch_sync_events(kvm);
 710         spin_lock(&kvm_lock);
 711         list_del(&kvm->vm_list);
 712         spin_unlock(&kvm_lock);
 713         kvm_free_irq_routing(kvm);
 714         for (i = 0; i < KVM_NR_BUSES; i++)
 715                 kvm_io_bus_destroy(kvm->buses[i]);
 716         kvm_coalesced_mmio_free(kvm);
 717 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 718         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
 719 #else
 720         kvm_arch_flush_shadow_all(kvm);
 721 #endif
 722         kvm_arch_destroy_vm(kvm);
 723         kvm_destroy_devices(kvm);
 724         for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
 725                 kvm_free_memslots(kvm, kvm->memslots[i]);
 726         cleanup_srcu_struct(&kvm->irq_srcu);
 727         cleanup_srcu_struct(&kvm->srcu);
 728         kvm_arch_free_vm(kvm);
 729         preempt_notifier_dec();
 730         hardware_disable_all();
 731         mmdrop(mm);
 732 }
 733
 734 void kvm_get_kvm(struct kvm *kvm)
 735 {
 736         atomic_inc(&kvm->users_count);
 737 }
 738 EXPORT_SYMBOL_GPL(kvm_get_kvm);
 739
 740 void kvm_put_kvm(struct kvm *kvm)
 741 {
 742         if (atomic_dec_and_test(&kvm->users_count))
 743                 kvm_destroy_vm(kvm);
 744 }
 745 EXPORT_SYMBOL_GPL(kvm_put_kvm);
 746
 747
 748 static int kvm_vm_release(struct inode *inode, struct file *filp)
 749 {
 750         struct kvm *kvm = filp->private_data;
 751
 752         kvm_irqfd_release(kvm);
 753
 754         kvm_put_kvm(kvm);
 755         return 0;
 756 }
 757
 758 /*
 759  * Allocation size is twice as large as the actual dirty bitmap size.
 760  * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
 761  */
 762 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
 763 {
 764         unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
 765
 766         memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
 767         if (!memslot->dirty_bitmap)
 768                 return -ENOMEM;
 769
 770         return 0;
 771 }
 772
 773 /*
 774  * Insert memslot and re-sort memslots based on their GFN,
 775  * so binary search could be used to lookup GFN.
 776  * Sorting algorithm takes advantage of having initially
 777  * sorted array and known changed memslot position.
 778  */
 779 static void update_memslots(struct kvm_memslots *slots,
 780                             struct kvm_memory_slot *new)
 781 {
 782         int id = new->id;
 783         int i = slots->id_to_index[id];
 784         struct kvm_memory_slot *mslots = slots->memslots;
 785
 786         WARN_ON(mslots[i].id != id);
 787         if (!new->npages) {
 788                 WARN_ON(!mslots[i].npages);
 789                 if (mslots[i].npages)
 790                         slots->used_slots--;
 791         } else {
 792                 if (!mslots[i].npages)
 793                         slots->used_slots++;
 794         }
 795
 796         while (i < KVM_MEM_SLOTS_NUM - 1 &&
 797                new->base_gfn <= mslots[i + 1].base_gfn) {
 798                 if (!mslots[i + 1].npages)
 799                         break;
 800                 mslots[i] = mslots[i + 1];
 801                 slots->id_to_index[mslots[i].id] = i;
 802                 i++;
 803         }
 804
 805         /*
 806          * The ">=" is needed when creating a slot with base_gfn == 0,
 807          * so that it moves before all those with base_gfn == npages == 0.
 808          *
 809          * On the other hand, if new->npages is zero, the above loop has
 810          * already left i pointing to the beginning of the empty part of
 811          * mslots, and the ">=" would move the hole backwards in this
 812          * case---which is wrong.  So skip the loop when deleting a slot.
 813          */
 814         if (new->npages) {
 815                 while (i > 0 &&
 816                        new->base_gfn >= mslots[i - 1].base_gfn) {
 817                         mslots[i] = mslots[i - 1];
 818                         slots->id_to_index[mslots[i].id] = i;
 819                         i--;
 820                 }
 821         } else
 822                 WARN_ON_ONCE(i != slots->used_slots);
 823
 824         mslots[i] = *new;
 825         slots->id_to_index[mslots[i].id] = i;
 826 }
 827
 828 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
 829 {
 830         u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
 831
 832 #ifdef __KVM_HAVE_READONLY_MEM
 833         valid_flags |= KVM_MEM_READONLY;
 834 #endif
 835
 836         if (mem->flags & ~valid_flags)
 837                 return -EINVAL;
 838
 839         return 0;
 840 }
 841
 842 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
 843                 int as_id, struct kvm_memslots *slots)
 844 {
 845         struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id);
 846
 847         /*
 848          * Set the low bit in the generation, which disables SPTE caching
 849          * until the end of synchronize_srcu_expedited.
 850          */
 851         WARN_ON(old_memslots->generation & 1);
 852         slots->generation = old_memslots->generation + 1;
 853
 854         rcu_assign_pointer(kvm->memslots[as_id], slots);
 855         synchronize_srcu_expedited(&kvm->srcu);
 856
 857         /*
 858          * Increment the new memslot generation a second time. This prevents
 859          * vm exits that race with memslot updates from caching a memslot
 860          * generation that will (potentially) be valid forever.
 861          */
 862         slots->generation++;
 863
 864         kvm_arch_memslots_updated(kvm, slots);
 865
 866         return old_memslots;
 867 }
 868
 869 /*
 870  * Allocate some memory and give it an address in the guest physical address
 871  * space.
 872  *
 873  * Discontiguous memory is allowed, mostly for framebuffers.
 874  *
 875  * Must be called holding kvm->slots_lock for write.
 876  */
 877 int __kvm_set_memory_region(struct kvm *kvm,
 878                             const struct kvm_userspace_memory_region *mem)
 879 {
 880         int r;
 881         gfn_t base_gfn;
 882         unsigned long npages;
 883         struct kvm_memory_slot *slot;
 884         struct kvm_memory_slot old, new;
 885         struct kvm_memslots *slots = NULL, *old_memslots;
 886         int as_id, id;
 887         enum kvm_mr_change change;
 888
 889         r = check_memory_region_flags(mem);
 890         if (r)
 891                 goto out;
 892
 893         r = -EINVAL;
 894         as_id = mem->slot >> 16;
 895         id = (u16)mem->slot;
 896
 897         /* General sanity checks */
 898         if (mem->memory_size & (PAGE_SIZE - 1))
 899                 goto out;
 900         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 901                 goto out;
 902         /* We can read the guest memory with __xxx_user() later on. */
 903         if ((id < KVM_USER_MEM_SLOTS) &&
 904             ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
 905              !access_ok(VERIFY_WRITE,
 906                         (void __user *)(unsigned long)mem->userspace_addr,
 907                         mem->memory_size)))
 908                 goto out;
 909         if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
 910                 goto out;
 911         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 912                 goto out;
 913
 914         slot = id_to_memslot(__kvm_memslots(kvm, as_id), id);
 915         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 916         npages = mem->memory_size >> PAGE_SHIFT;
 917
 918         if (npages > KVM_MEM_MAX_NR_PAGES)
 919                 goto out;
 920
 921         new = old = *slot;
 922
 923         new.id = id;
 924         new.base_gfn = base_gfn;
 925         new.npages = npages;
 926         new.flags = mem->flags;
 927
 928         if (npages) {
 929                 if (!old.npages)
 930                         change = KVM_MR_CREATE;
 931                 else { /* Modify an existing slot. */
 932                         if ((mem->userspace_addr != old.userspace_addr) ||
 933                             (npages != old.npages) ||
 934                             ((new.flags ^ old.flags) & KVM_MEM_READONLY))
 935                                 goto out;
 936
 937                         if (base_gfn != old.base_gfn)
 938                                 change = KVM_MR_MOVE;
 939                         else if (new.flags != old.flags)
 940                                 change = KVM_MR_FLAGS_ONLY;
 941                         else { /* Nothing to change. */
 942                                 r = 0;
 943                                 goto out;
 944                         }
 945                 }
 946         } else {
 947                 if (!old.npages)
 948                         goto out;
 949
 950                 change = KVM_MR_DELETE;
 951                 new.base_gfn = 0;
 952                 new.flags = 0;
 953         }
 954
 955         if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
 956                 /* Check for overlaps */
 957                 r = -EEXIST;
 958                 kvm_for_each_memslot(slot, __kvm_memslots(kvm, as_id)) {
 959                         if ((slot->id >= KVM_USER_MEM_SLOTS) ||
 960                             (slot->id == id))
 961                                 continue;
 962                         if (!((base_gfn + npages <= slot->base_gfn) ||
 963                               (base_gfn >= slot->base_gfn + slot->npages)))
 964                                 goto out;
 965                 }
 966         }
 967
 968         /* Free page dirty bitmap if unneeded */
 969         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 970                 new.dirty_bitmap = NULL;
 971
 972         r = -ENOMEM;
 973         if (change == KVM_MR_CREATE) {
 974                 new.userspace_addr = mem->userspace_addr;
 975
 976                 if (kvm_arch_create_memslot(kvm, &new, npages))
 977                         goto out_free;
 978         }
 979
 980         /* Allocate page dirty bitmap if needed */
 981         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 982                 if (kvm_create_dirty_bitmap(&new) < 0)
 983                         goto out_free;
 984         }
 985
 986         slots = kvm_kvzalloc(sizeof(struct kvm_memslots));
 987         if (!slots)
 988                 goto out_free;
 989         memcpy(slots, __kvm_memslots(kvm, as_id), sizeof(struct kvm_memslots));
 990
 991         if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
 992                 slot = id_to_memslot(slots, id);
 993                 slot->flags |= KVM_MEMSLOT_INVALID;
 994
 995                 old_memslots = install_new_memslots(kvm, as_id, slots);
 996
 997                 /* slot was deleted or moved, clear iommu mapping */
 998                 kvm_iommu_unmap_pages(kvm, &old);
 999                 /* From this point no new shadow pages pointing to a deleted,
1000                  * or moved, memslot will be created.
1001                  *
1002                  * validation of sp->gfn happens in:
1003                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1004                  *      - kvm_is_visible_gfn (mmu_check_roots)
1005                  */
1006                 kvm_arch_flush_shadow_memslot(kvm, slot);
1007
1008                 /*
1009                  * We can re-use the old_memslots from above, the only difference
1010                  * from the currently installed memslots is the invalid flag.  This
1011                  * will get overwritten by update_memslots anyway.
1012                  */
1013                 slots = old_memslots;
1014         }
1015
1016         r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
1017         if (r)
1018                 goto out_slots;
1019
1020         /* actual memory is freed via old in kvm_free_memslot below */
1021         if (change == KVM_MR_DELETE) {
1022                 new.dirty_bitmap = NULL;
1023                 memset(&new.arch, 0, sizeof(new.arch));
1024         }
1025
1026         update_memslots(slots, &new);
1027         old_memslots = install_new_memslots(kvm, as_id, slots);
1028
1029         kvm_arch_commit_memory_region(kvm, mem, &old, &new, change);
1030
1031         kvm_free_memslot(kvm, &old, &new);
1032         kvfree(old_memslots);
1033
1034         /*
1035          * IOMMU mapping:  New slots need to be mapped.  Old slots need to be
1036          * un-mapped and re-mapped if their base changes.  Since base change
1037          * unmapping is handled above with slot deletion, mapping alone is
1038          * needed here.  Anything else the iommu might care about for existing
1039          * slots (size changes, userspace addr changes and read-only flag
1040          * changes) is disallowed above, so any other attribute changes getting
1041          * here can be skipped.
1042          */
1043         if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
1044                 r = kvm_iommu_map_pages(kvm, &new);
1045                 return r;
1046         }
1047
1048         return 0;
1049
1050 out_slots:
1051         kvfree(slots);
1052 out_free:
1053         kvm_free_memslot(kvm, &new, &old);
1054 out:
1055         return r;
1056 }
1057 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1058
1059 int kvm_set_memory_region(struct kvm *kvm,
1060                           const struct kvm_userspace_memory_region *mem)
1061 {
1062         int r;
1063
1064         mutex_lock(&kvm->slots_lock);
1065         r = __kvm_set_memory_region(kvm, mem);
1066         mutex_unlock(&kvm->slots_lock);
1067         return r;
1068 }
1069 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1070
1071 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1072                                           struct kvm_userspace_memory_region *mem)
1073 {
1074         if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1075                 return -EINVAL;
1076
1077         return kvm_set_memory_region(kvm, mem);
1078 }
1079
1080 int kvm_get_dirty_log(struct kvm *kvm,
1081                         struct kvm_dirty_log *log, int *is_dirty)
1082 {
1083         struct kvm_memslots *slots;
1084         struct kvm_memory_slot *memslot;
1085         int r, i, as_id, id;
1086         unsigned long n;
1087         unsigned long any = 0;
1088
1089         r = -EINVAL;
1090         as_id = log->slot >> 16;
1091         id = (u16)log->slot;
1092         if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1093                 goto out;
1094
1095         slots = __kvm_memslots(kvm, as_id);
1096         memslot = id_to_memslot(slots, id);
1097         r = -ENOENT;
1098         if (!memslot->dirty_bitmap)
1099                 goto out;
1100
1101         n = kvm_dirty_bitmap_bytes(memslot);
1102
1103         for (i = 0; !any && i < n/sizeof(long); ++i)
1104                 any = memslot->dirty_bitmap[i];
1105
1106         r = -EFAULT;
1107         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1108                 goto out;
1109
1110         if (any)
1111                 *is_dirty = 1;
1112
1113         r = 0;
1114 out:
1115         return r;
1116 }
1117 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1118
1119 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1120 /**
1121  * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages
1122  *      are dirty write protect them for next write.
1123  * @kvm:        pointer to kvm instance
1124  * @log:        slot id and address to which we copy the log
1125  * @is_dirty:   flag set if any page is dirty
1126  *
1127  * We need to keep it in mind that VCPU threads can write to the bitmap
1128  * concurrently. So, to avoid losing track of dirty pages we keep the
1129  * following order:
1130  *
1131  *    1. Take a snapshot of the bit and clear it if needed.
1132  *    2. Write protect the corresponding page.
1133  *    3. Copy the snapshot to the userspace.
1134  *    4. Upon return caller flushes TLB's if needed.
1135  *
1136  * Between 2 and 4, the guest may write to the page using the remaining TLB
1137  * entry.  This is not a problem because the page is reported dirty using
1138  * the snapshot taken before and step 4 ensures that writes done after
1139  * exiting to userspace will be logged for the next call.
1140  *
1141  */
1142 int kvm_get_dirty_log_protect(struct kvm *kvm,
1143                         struct kvm_dirty_log *log, bool *is_dirty)
1144 {
1145         struct kvm_memslots *slots;
1146         struct kvm_memory_slot *memslot;
1147         int r, i, as_id, id;
1148         unsigned long n;
1149         unsigned long *dirty_bitmap;
1150         unsigned long *dirty_bitmap_buffer;
1151
1152         r = -EINVAL;
1153         as_id = log->slot >> 16;
1154         id = (u16)log->slot;
1155         if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1156                 goto out;
1157
1158         slots = __kvm_memslots(kvm, as_id);
1159         memslot = id_to_memslot(slots, id);
1160
1161         dirty_bitmap = memslot->dirty_bitmap;
1162         r = -ENOENT;
1163         if (!dirty_bitmap)
1164                 goto out;
1165
1166         n = kvm_dirty_bitmap_bytes(memslot);
1167
1168         dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long);
1169         memset(dirty_bitmap_buffer, 0, n);
1170
1171         spin_lock(&kvm->mmu_lock);
1172         *is_dirty = false;
1173         for (i = 0; i < n / sizeof(long); i++) {
1174                 unsigned long mask;
1175                 gfn_t offset;
1176
1177                 if (!dirty_bitmap[i])
1178                         continue;
1179
1180                 *is_dirty = true;
1181
1182                 mask = xchg(&dirty_bitmap[i], 0);
1183                 dirty_bitmap_buffer[i] = mask;
1184
1185                 if (mask) {
1186                         offset = i * BITS_PER_LONG;
1187                         kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1188                                                                 offset, mask);
1189                 }
1190         }
1191
1192         spin_unlock(&kvm->mmu_lock);
1193
1194         r = -EFAULT;
1195         if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1196                 goto out;
1197
1198         r = 0;
1199 out:
1200         return r;
1201 }
1202 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect);
1203 #endif
1204
1205 bool kvm_largepages_enabled(void)
1206 {
1207         return largepages_enabled;
1208 }
1209
1210 void kvm_disable_largepages(void)
1211 {
1212         largepages_enabled = false;
1213 }
1214 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1215
1216 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1217 {
1218         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1219 }
1220 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1221
1222 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
1223 {
1224         return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn);
1225 }
1226
1227 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1228 {
1229         struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1230
1231         if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1232               memslot->flags & KVM_MEMSLOT_INVALID)
1233                 return false;
1234
1235         return true;
1236 }
1237 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1238
1239 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1240 {
1241         struct vm_area_struct *vma;
1242         unsigned long addr, size;
1243
1244         size = PAGE_SIZE;
1245
1246         addr = gfn_to_hva(kvm, gfn);
1247         if (kvm_is_error_hva(addr))
1248                 return PAGE_SIZE;
1249
1250         down_read(&current->mm->mmap_sem);
1251         vma = find_vma(current->mm, addr);
1252         if (!vma)
1253                 goto out;
1254
1255         size = vma_kernel_pagesize(vma);
1256
1257 out:
1258         up_read(&current->mm->mmap_sem);
1259
1260         return size;
1261 }
1262
1263 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1264 {
1265         return slot->flags & KVM_MEM_READONLY;
1266 }
1267
1268 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1269                                        gfn_t *nr_pages, bool write)
1270 {
1271         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1272                 return KVM_HVA_ERR_BAD;
1273
1274         if (memslot_is_readonly(slot) && write)
1275                 return KVM_HVA_ERR_RO_BAD;
1276
1277         if (nr_pages)
1278                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1279
1280         return __gfn_to_hva_memslot(slot, gfn);
1281 }
1282
1283 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1284                                      gfn_t *nr_pages)
1285 {
1286         return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1287 }
1288
1289 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1290                                         gfn_t gfn)
1291 {
1292         return gfn_to_hva_many(slot, gfn, NULL);
1293 }
1294 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1295
1296 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1297 {
1298         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1299 }
1300 EXPORT_SYMBOL_GPL(gfn_to_hva);
1301
1302 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
1303 {
1304         return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
1305 }
1306 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
1307
1308 /*
1309  * If writable is set to false, the hva returned by this function is only
1310  * allowed to be read.
1311  */
1312 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1313                                       gfn_t gfn, bool *writable)
1314 {
1315         unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1316
1317         if (!kvm_is_error_hva(hva) && writable)
1318                 *writable = !memslot_is_readonly(slot);
1319
1320         return hva;
1321 }
1322
1323 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1324 {
1325         struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1326
1327         return gfn_to_hva_memslot_prot(slot, gfn, writable);
1328 }
1329
1330 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
1331 {
1332         struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1333
1334         return gfn_to_hva_memslot_prot(slot, gfn, writable);
1335 }
1336
1337 static int get_user_page_nowait(unsigned long start, int write,
1338                 struct page **page)
1339 {
1340         int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1341
1342         if (write)
1343                 flags |= FOLL_WRITE;
1344
1345         return __get_user_pages(current, current->mm, start, 1, flags, page,
1346                         NULL, NULL);
1347 }
1348
1349 static inline int check_user_page_hwpoison(unsigned long addr)
1350 {
1351         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1352
1353         rc = __get_user_pages(current, current->mm, addr, 1,
1354                               flags, NULL, NULL, NULL);
1355         return rc == -EHWPOISON;
1356 }
1357
1358 /*
1359  * The atomic path to get the writable pfn which will be stored in @pfn,
1360  * true indicates success, otherwise false is returned.
1361  */
1362 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1363                             bool write_fault, bool *writable, kvm_pfn_t *pfn)
1364 {
1365         struct page *page[1];
1366         int npages;
1367
1368         if (!(async || atomic))
1369                 return false;
1370
1371         /*
1372          * Fast pin a writable pfn only if it is a write fault request
1373          * or the caller allows to map a writable pfn for a read fault
1374          * request.
1375          */
1376         if (!(write_fault || writable))
1377                 return false;
1378
1379         npages = __get_user_pages_fast(addr, 1, 1, page);
1380         if (npages == 1) {
1381                 *pfn = page_to_pfn(page[0]);
1382
1383                 if (writable)
1384                         *writable = true;
1385                 return true;
1386         }
1387
1388         return false;
1389 }
1390
1391 /*
1392  * The slow path to get the pfn of the specified host virtual address,
1393  * 1 indicates success, -errno is returned if error is detected.
1394  */
1395 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1396                            bool *writable, kvm_pfn_t *pfn)
1397 {
1398         struct page *page[1];
1399         int npages = 0;
1400
1401         might_sleep();
1402
1403         if (writable)
1404                 *writable = write_fault;
1405
1406         if (async) {
1407                 down_read(&current->mm->mmap_sem);
1408                 npages = get_user_page_nowait(addr, write_fault, page);
1409                 up_read(&current->mm->mmap_sem);
1410         } else
1411                 npages = __get_user_pages_unlocked(current, current->mm, addr, 1,
1412                                                    write_fault, 0, page,
1413                                                    FOLL_TOUCH|FOLL_HWPOISON);
1414         if (npages != 1)
1415                 return npages;
1416
1417         /* map read fault as writable if possible */
1418         if (unlikely(!write_fault) && writable) {
1419                 struct page *wpage[1];
1420
1421                 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1422                 if (npages == 1) {
1423                         *writable = true;
1424                         put_page(page[0]);
1425                         page[0] = wpage[0];
1426                 }
1427
1428                 npages = 1;
1429         }
1430         *pfn = page_to_pfn(page[0]);
1431         return npages;
1432 }
1433
1434 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1435 {
1436         if (unlikely(!(vma->vm_flags & VM_READ)))
1437                 return false;
1438
1439         if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1440                 return false;
1441
1442         return true;
1443 }
1444
1445 /*
1446  * Pin guest page in memory and return its pfn.
1447  * @addr: host virtual address which maps memory to the guest
1448  * @atomic: whether this function can sleep
1449  * @async: whether this function need to wait IO complete if the
1450  *         host page is not in the memory
1451  * @write_fault: whether we should get a writable host page
1452  * @writable: whether it allows to map a writable host page for !@write_fault
1453  *
1454  * The function will map a writable host page for these two cases:
1455  * 1): @write_fault = true
1456  * 2): @write_fault = false && @writable, @writable will tell the caller
1457  *     whether the mapping is writable.
1458  */
1459 static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1460                         bool write_fault, bool *writable)
1461 {
1462         struct vm_area_struct *vma;
1463         kvm_pfn_t pfn = 0;
1464         int npages;
1465
1466         /* we can do it either atomically or asynchronously, not both */
1467         BUG_ON(atomic && async);
1468
1469         if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1470                 return pfn;
1471
1472         if (atomic)
1473                 return KVM_PFN_ERR_FAULT;
1474
1475         npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1476         if (npages == 1)
1477                 return pfn;
1478
1479         down_read(&current->mm->mmap_sem);
1480         if (npages == -EHWPOISON ||
1481               (!async && check_user_page_hwpoison(addr))) {
1482                 pfn = KVM_PFN_ERR_HWPOISON;
1483                 goto exit;
1484         }
1485
1486         vma = find_vma_intersection(current->mm, addr, addr + 1);
1487
1488         if (vma == NULL)
1489                 pfn = KVM_PFN_ERR_FAULT;
1490         else if ((vma->vm_flags & VM_PFNMAP)) {
1491                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1492                         vma->vm_pgoff;
1493                 BUG_ON(!kvm_is_reserved_pfn(pfn));
1494         } else {
1495                 if (async && vma_is_valid(vma, write_fault))
1496                         *async = true;
1497                 pfn = KVM_PFN_ERR_FAULT;
1498         }
1499 exit:
1500         up_read(&current->mm->mmap_sem);
1501         return pfn;
1502 }
1503
1504 kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn,
1505                                bool atomic, bool *async, bool write_fault,
1506                                bool *writable)
1507 {
1508         unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1509
1510         if (addr == KVM_HVA_ERR_RO_BAD) {
1511                 if (writable)
1512                         *writable = false;
1513                 return KVM_PFN_ERR_RO_FAULT;
1514         }
1515
1516         if (kvm_is_error_hva(addr)) {
1517                 if (writable)
1518                         *writable = false;
1519                 return KVM_PFN_NOSLOT;
1520         }
1521
1522         /* Do not map writable pfn in the readonly memslot. */
1523         if (writable && memslot_is_readonly(slot)) {
1524                 *writable = false;
1525                 writable = NULL;
1526         }
1527
1528         return hva_to_pfn(addr, atomic, async, write_fault,
1529                           writable);
1530 }
1531 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
1532
1533 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1534                       bool *writable)
1535 {
1536         return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
1537                                     write_fault, writable);
1538 }
1539 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1540
1541 kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1542 {
1543         return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1544 }
1545 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
1546
1547 kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1548 {
1549         return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1550 }
1551 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1552
1553 kvm_pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1554 {
1555         return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm, gfn), gfn);
1556 }
1557 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1558
1559 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
1560 {
1561         return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1562 }
1563 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
1564
1565 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1566 {
1567         return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
1568 }
1569 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1570
1571 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
1572 {
1573         return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1574 }
1575 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
1576
1577 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1578                             struct page **pages, int nr_pages)
1579 {
1580         unsigned long addr;
1581         gfn_t entry;
1582
1583         addr = gfn_to_hva_many(slot, gfn, &entry);
1584         if (kvm_is_error_hva(addr))
1585                 return -1;
1586
1587         if (entry < nr_pages)
1588                 return 0;
1589
1590         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1591 }
1592 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1593
1594 static struct page *kvm_pfn_to_page(kvm_pfn_t pfn)
1595 {
1596         if (is_error_noslot_pfn(pfn))
1597                 return KVM_ERR_PTR_BAD_PAGE;
1598
1599         if (kvm_is_reserved_pfn(pfn)) {
1600                 WARN_ON(1);
1601                 return KVM_ERR_PTR_BAD_PAGE;
1602         }
1603
1604         return pfn_to_page(pfn);
1605 }
1606
1607 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1608 {
1609         kvm_pfn_t pfn;
1610
1611         pfn = gfn_to_pfn(kvm, gfn);
1612
1613         return kvm_pfn_to_page(pfn);
1614 }
1615 EXPORT_SYMBOL_GPL(gfn_to_page);
1616
1617 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
1618 {
1619         kvm_pfn_t pfn;
1620
1621         pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
1622
1623         return kvm_pfn_to_page(pfn);
1624 }
1625 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
1626
1627 void kvm_release_page_clean(struct page *page)
1628 {
1629         WARN_ON(is_error_page(page));
1630
1631         kvm_release_pfn_clean(page_to_pfn(page));
1632 }
1633 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1634
1635 void kvm_release_pfn_clean(kvm_pfn_t pfn)
1636 {
1637         if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
1638                 put_page(pfn_to_page(pfn));
1639 }
1640 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1641
1642 void kvm_release_page_dirty(struct page *page)
1643 {
1644         WARN_ON(is_error_page(page));
1645
1646         kvm_release_pfn_dirty(page_to_pfn(page));
1647 }
1648 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1649
1650 static void kvm_release_pfn_dirty(kvm_pfn_t pfn)
1651 {
1652         kvm_set_pfn_dirty(pfn);
1653         kvm_release_pfn_clean(pfn);
1654 }
1655
1656 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
1657 {
1658         if (!kvm_is_reserved_pfn(pfn)) {
1659                 struct page *page = pfn_to_page(pfn);
1660
1661                 if (!PageReserved(page))
1662                         SetPageDirty(page);
1663         }
1664 }
1665 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1666
1667 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
1668 {
1669         if (!kvm_is_reserved_pfn(pfn))
1670                 mark_page_accessed(pfn_to_page(pfn));
1671 }
1672 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1673
1674 void kvm_get_pfn(kvm_pfn_t pfn)
1675 {
1676         if (!kvm_is_reserved_pfn(pfn))
1677                 get_page(pfn_to_page(pfn));
1678 }
1679 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1680
1681 static int next_segment(unsigned long len, int offset)
1682 {
1683         if (len > PAGE_SIZE - offset)
1684                 return PAGE_SIZE - offset;
1685         else
1686                 return len;
1687 }
1688
1689 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
1690                                  void *data, int offset, int len)
1691 {
1692         int r;
1693         unsigned long addr;
1694
1695         addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1696         if (kvm_is_error_hva(addr))
1697                 return -EFAULT;
1698         r = __copy_from_user(data, (void __user *)addr + offset, len);
1699         if (r)
1700                 return -EFAULT;
1701         return 0;
1702 }
1703
1704 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1705                         int len)
1706 {
1707         struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1708
1709         return __kvm_read_guest_page(slot, gfn, data, offset, len);
1710 }
1711 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1712
1713 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
1714                              int offset, int len)
1715 {
1716         struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1717
1718         return __kvm_read_guest_page(slot, gfn, data, offset, len);
1719 }
1720 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
1721
1722 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1723 {
1724         gfn_t gfn = gpa >> PAGE_SHIFT;
1725         int seg;
1726         int offset = offset_in_page(gpa);
1727         int ret;
1728
1729         while ((seg = next_segment(len, offset)) != 0) {
1730                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1731                 if (ret < 0)
1732                         return ret;
1733                 offset = 0;
1734                 len -= seg;
1735                 data += seg;
1736                 ++gfn;
1737         }
1738         return 0;
1739 }
1740 EXPORT_SYMBOL_GPL(kvm_read_guest);
1741
1742 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
1743 {
1744         gfn_t gfn = gpa >> PAGE_SHIFT;
1745         int seg;
1746         int offset = offset_in_page(gpa);
1747         int ret;
1748
1749         while ((seg = next_segment(len, offset)) != 0) {
1750                 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
1751                 if (ret < 0)
1752                         return ret;
1753                 offset = 0;
1754                 len -= seg;
1755                 data += seg;
1756                 ++gfn;
1757         }
1758         return 0;
1759 }
1760 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
1761
1762 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1763                                    void *data, int offset, unsigned long len)
1764 {
1765         int r;
1766         unsigned long addr;
1767
1768         addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1769         if (kvm_is_error_hva(addr))
1770                 return -EFAULT;
1771         pagefault_disable();
1772         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1773         pagefault_enable();
1774         if (r)
1775                 return -EFAULT;
1776         return 0;
1777 }
1778
1779 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1780                           unsigned long len)
1781 {
1782         gfn_t gfn = gpa >> PAGE_SHIFT;
1783         struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1784         int offset = offset_in_page(gpa);
1785
1786         return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1787 }
1788 EXPORT_SYMBOL_GPL(kvm_read_guest_atomic);
1789
1790 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
1791                                void *data, unsigned long len)
1792 {
1793         gfn_t gfn = gpa >> PAGE_SHIFT;
1794         struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1795         int offset = offset_in_page(gpa);
1796
1797         return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1798 }
1799 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
1800
1801 static int __kvm_write_guest_page(struct kvm_memory_slot *memslot, gfn_t gfn,
1802                                   const void *data, int offset, int len)
1803 {
1804         int r;
1805         unsigned long addr;
1806
1807         addr = gfn_to_hva_memslot(memslot, gfn);
1808         if (kvm_is_error_hva(addr))
1809                 return -EFAULT;
1810         r = __copy_to_user((void __user *)addr + offset, data, len);
1811         if (r)
1812                 return -EFAULT;
1813         mark_page_dirty_in_slot(memslot, gfn);
1814         return 0;
1815 }
1816
1817 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
1818                          const void *data, int offset, int len)
1819 {
1820         struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1821
1822         return __kvm_write_guest_page(slot, gfn, data, offset, len);
1823 }
1824 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1825
1826 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
1827                               const void *data, int offset, int len)
1828 {
1829         struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1830
1831         return __kvm_write_guest_page(slot, gfn, data, offset, len);
1832 }
1833 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
1834
1835 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1836                     unsigned long len)
1837 {
1838         gfn_t gfn = gpa >> PAGE_SHIFT;
1839         int seg;
1840         int offset = offset_in_page(gpa);
1841         int ret;
1842
1843         while ((seg = next_segment(len, offset)) != 0) {
1844                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1845                 if (ret < 0)
1846                         return ret;
1847                 offset = 0;
1848                 len -= seg;
1849                 data += seg;
1850                 ++gfn;
1851         }
1852         return 0;
1853 }
1854 EXPORT_SYMBOL_GPL(kvm_write_guest);
1855
1856 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
1857                          unsigned long len)
1858 {
1859         gfn_t gfn = gpa >> PAGE_SHIFT;
1860         int seg;
1861         int offset = offset_in_page(gpa);
1862         int ret;
1863
1864         while ((seg = next_segment(len, offset)) != 0) {
1865                 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
1866                 if (ret < 0)
1867                         return ret;
1868                 offset = 0;
1869                 len -= seg;
1870                 data += seg;
1871                 ++gfn;
1872         }
1873         return 0;
1874 }
1875 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
1876
1877 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1878                               gpa_t gpa, unsigned long len)
1879 {
1880         struct kvm_memslots *slots = kvm_memslots(kvm);
1881         int offset = offset_in_page(gpa);
1882         gfn_t start_gfn = gpa >> PAGE_SHIFT;
1883         gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1884         gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1885         gfn_t nr_pages_avail;
1886
1887         ghc->gpa = gpa;
1888         ghc->generation = slots->generation;
1889         ghc->len = len;
1890         ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1891         ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, NULL);
1892         if (!kvm_is_error_hva(ghc->hva) && nr_pages_needed <= 1) {
1893                 ghc->hva += offset;
1894         } else {
1895                 /*
1896                  * If the requested region crosses two memslots, we still
1897                  * verify that the entire region is valid here.
1898                  */
1899                 while (start_gfn <= end_gfn) {
1900                         ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1901                         ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1902                                                    &nr_pages_avail);
1903                         if (kvm_is_error_hva(ghc->hva))
1904                                 return -EFAULT;
1905                         start_gfn += nr_pages_avail;
1906                 }
1907                 /* Use the slow path for cross page reads and writes. */
1908                 ghc->memslot = NULL;
1909         }
1910         return 0;
1911 }
1912 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1913
1914 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1915                            void *data, unsigned long len)
1916 {
1917         struct kvm_memslots *slots = kvm_memslots(kvm);
1918         int r;
1919
1920         BUG_ON(len > ghc->len);
1921
1922         if (slots->generation != ghc->generation)
1923                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1924
1925         if (unlikely(!ghc->memslot))
1926                 return kvm_write_guest(kvm, ghc->gpa, data, len);
1927
1928         if (kvm_is_error_hva(ghc->hva))
1929                 return -EFAULT;
1930
1931         r = __copy_to_user((void __user *)ghc->hva, data, len);
1932         if (r)
1933                 return -EFAULT;
1934         mark_page_dirty_in_slot(ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1935
1936         return 0;
1937 }
1938 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1939
1940 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1941                            void *data, unsigned long len)
1942 {
1943         struct kvm_memslots *slots = kvm_memslots(kvm);
1944         int r;
1945
1946         BUG_ON(len > ghc->len);
1947
1948         if (slots->generation != ghc->generation)
1949                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1950
1951         if (unlikely(!ghc->memslot))
1952                 return kvm_read_guest(kvm, ghc->gpa, data, len);
1953
1954         if (kvm_is_error_hva(ghc->hva))
1955                 return -EFAULT;
1956
1957         r = __copy_from_user(data, (void __user *)ghc->hva, len);
1958         if (r)
1959                 return -EFAULT;
1960
1961         return 0;
1962 }
1963 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1964
1965 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1966 {
1967         const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
1968
1969         return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
1970 }
1971 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1972
1973 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1974 {
1975         gfn_t gfn = gpa >> PAGE_SHIFT;
1976         int seg;
1977         int offset = offset_in_page(gpa);
1978         int ret;
1979
1980         while ((seg = next_segment(len, offset)) != 0) {
1981                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1982                 if (ret < 0)
1983                         return ret;
1984                 offset = 0;
1985                 len -= seg;
1986                 ++gfn;
1987         }
1988         return 0;
1989 }
1990 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1991
1992 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot,
1993                                     gfn_t gfn)
1994 {
1995         if (memslot && memslot->dirty_bitmap) {
1996                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1997
1998                 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1999         }
2000 }
2001
2002 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
2003 {
2004         struct kvm_memory_slot *memslot;
2005
2006         memslot = gfn_to_memslot(kvm, gfn);
2007         mark_page_dirty_in_slot(memslot, gfn);
2008 }
2009 EXPORT_SYMBOL_GPL(mark_page_dirty);
2010
2011 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
2012 {
2013         struct kvm_memory_slot *memslot;
2014
2015         memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2016         mark_page_dirty_in_slot(memslot, gfn);
2017 }
2018 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
2019
2020 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
2021 {
2022         unsigned int old, val, grow;
2023
2024         old = val = vcpu->halt_poll_ns;
2025         grow = READ_ONCE(halt_poll_ns_grow);
2026         /* 10us base */
2027         if (val == 0 && grow)
2028                 val = 10000;
2029         else
2030                 val *= grow;
2031
2032         if (val > halt_poll_ns)
2033                 val = halt_poll_ns;
2034
2035         vcpu->halt_poll_ns = val;
2036         trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
2037 }
2038
2039 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
2040 {
2041         unsigned int old, val, shrink;
2042
2043         old = val = vcpu->halt_poll_ns;
2044         shrink = READ_ONCE(halt_poll_ns_shrink);
2045         if (shrink == 0)
2046                 val = 0;
2047         else
2048                 val /= shrink;
2049
2050         vcpu->halt_poll_ns = val;
2051         trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
2052 }
2053
2054 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
2055 {
2056         if (kvm_arch_vcpu_runnable(vcpu)) {
2057                 kvm_make_request(KVM_REQ_UNHALT, vcpu);
2058                 return -EINTR;
2059         }
2060         if (kvm_cpu_has_pending_timer(vcpu))
2061                 return -EINTR;
2062         if (signal_pending(current))
2063                 return -EINTR;
2064
2065         return 0;
2066 }
2067
2068 /*
2069  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2070  */
2071 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
2072 {
2073         ktime_t start, cur;
2074         DECLARE_SWAITQUEUE(wait);
2075         bool waited = false;
2076         u64 block_ns;
2077
2078         start = cur = ktime_get();
2079         if (vcpu->halt_poll_ns) {
2080                 ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns);
2081
2082                 ++vcpu->stat.halt_attempted_poll;
2083                 do {
2084                         /*
2085                          * This sets KVM_REQ_UNHALT if an interrupt
2086                          * arrives.
2087                          */
2088                         if (kvm_vcpu_check_block(vcpu) < 0) {
2089                                 ++vcpu->stat.halt_successful_poll;
2090                                 if (!vcpu_valid_wakeup(vcpu))
2091                                         ++vcpu->stat.halt_poll_invalid;
2092                                 goto out;
2093                         }
2094                         cur = ktime_get();
2095                 } while (single_task_running() && ktime_before(cur, stop));
2096         }
2097
2098         kvm_arch_vcpu_blocking(vcpu);
2099
2100         for (;;) {
2101                 prepare_to_swait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
2102
2103                 if (kvm_vcpu_check_block(vcpu) < 0)
2104                         break;
2105
2106                 waited = true;
2107                 schedule();
2108         }
2109
2110         finish_swait(&vcpu->wq, &wait);
2111         cur = ktime_get();
2112
2113         kvm_arch_vcpu_unblocking(vcpu);
2114 out:
2115         block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
2116
2117         if (!vcpu_valid_wakeup(vcpu))
2118                 shrink_halt_poll_ns(vcpu);
2119         else if (halt_poll_ns) {
2120                 if (block_ns <= vcpu->halt_poll_ns)
2121                         ;
2122                 /* we had a long block, shrink polling */
2123                 else if (vcpu->halt_poll_ns && block_ns > halt_poll_ns)
2124                         shrink_halt_poll_ns(vcpu);
2125                 /* we had a short halt and our poll time is too small */
2126                 else if (vcpu->halt_poll_ns < halt_poll_ns &&
2127                         block_ns < halt_poll_ns)
2128                         grow_halt_poll_ns(vcpu);
2129         } else
2130                 vcpu->halt_poll_ns = 0;
2131
2132         trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu));
2133         kvm_arch_vcpu_block_finish(vcpu);
2134 }
2135 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
2136
2137 #ifndef CONFIG_S390
2138 void kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
2139 {
2140         struct swait_queue_head *wqp;
2141
2142         wqp = kvm_arch_vcpu_wq(vcpu);
2143         if (swait_active(wqp)) {
2144                 swake_up(wqp);
2145                 ++vcpu->stat.halt_wakeup;
2146         }
2147
2148 }
2149 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
2150
2151 /*
2152  * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2153  */
2154 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
2155 {
2156         int me;
2157         int cpu = vcpu->cpu;
2158
2159         kvm_vcpu_wake_up(vcpu);
2160         me = get_cpu();
2161         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
2162                 if (kvm_arch_vcpu_should_kick(vcpu))
2163                         smp_send_reschedule(cpu);
2164         put_cpu();
2165 }
2166 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
2167 #endif /* !CONFIG_S390 */
2168
2169 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
2170 {
2171         struct pid *pid;
2172         struct task_struct *task = NULL;
2173         int ret = 0;
2174
2175         rcu_read_lock();
2176         pid = rcu_dereference(target->pid);
2177         if (pid)
2178                 task = get_pid_task(pid, PIDTYPE_PID);
2179         rcu_read_unlock();
2180         if (!task)
2181                 return ret;
2182         ret = yield_to(task, 1);
2183         put_task_struct(task);
2184
2185         return ret;
2186 }
2187 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
2188
2189 /*
2190  * Helper that checks whether a VCPU is eligible for directed yield.
2191  * Most eligible candidate to yield is decided by following heuristics:
2192  *
2193  *  (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2194  *  (preempted lock holder), indicated by @in_spin_loop.
2195  *  Set at the beiginning and cleared at the end of interception/PLE handler.
2196  *
2197  *  (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2198  *  chance last time (mostly it has become eligible now since we have probably
2199  *  yielded to lockholder in last iteration. This is done by toggling
2200  *  @dy_eligible each time a VCPU checked for eligibility.)
2201  *
2202  *  Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2203  *  to preempted lock-holder could result in wrong VCPU selection and CPU
2204  *  burning. Giving priority for a potential lock-holder increases lock
2205  *  progress.
2206  *
2207  *  Since algorithm is based on heuristics, accessing another VCPU data without
2208  *  locking does not harm. It may result in trying to yield to  same VCPU, fail
2209  *  and continue with next VCPU and so on.
2210  */
2211 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
2212 {
2213 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2214         bool eligible;
2215
2216         eligible = !vcpu->spin_loop.in_spin_loop ||
2217                     vcpu->spin_loop.dy_eligible;
2218
2219         if (vcpu->spin_loop.in_spin_loop)
2220                 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
2221
2222         return eligible;
2223 #else
2224         return true;
2225 #endif
2226 }
2227
2228 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
2229 {
2230         struct kvm *kvm = me->kvm;
2231         struct kvm_vcpu *vcpu;
2232         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
2233         int yielded = 0;
2234         int try = 3;
2235         int pass;
2236         int i;
2237
2238         kvm_vcpu_set_in_spin_loop(me, true);
2239         /*
2240          * We boost the priority of a VCPU that is runnable but not
2241          * currently running, because it got preempted by something
2242          * else and called schedule in __vcpu_run.  Hopefully that
2243          * VCPU is holding the lock that we need and will release it.
2244          * We approximate round-robin by starting at the last boosted VCPU.
2245          */
2246         for (pass = 0; pass < 2 && !yielded && try; pass++) {
2247                 kvm_for_each_vcpu(i, vcpu, kvm) {
2248                         if (!pass && i <= last_boosted_vcpu) {
2249                                 i = last_boosted_vcpu;
2250                                 continue;
2251                         } else if (pass && i > last_boosted_vcpu)
2252                                 break;
2253                         if (!ACCESS_ONCE(vcpu->preempted))
2254                                 continue;
2255                         if (vcpu == me)
2256                                 continue;
2257                         if (swait_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
2258                                 continue;
2259                         if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
2260                                 continue;
2261
2262                         yielded = kvm_vcpu_yield_to(vcpu);
2263                         if (yielded > 0) {
2264                                 kvm->last_boosted_vcpu = i;
2265                                 break;
2266                         } else if (yielded < 0) {
2267                                 try--;
2268                                 if (!try)
2269                                         break;
2270                         }
2271                 }
2272         }
2273         kvm_vcpu_set_in_spin_loop(me, false);
2274
2275         /* Ensure vcpu is not eligible during next spinloop */
2276         kvm_vcpu_set_dy_eligible(me, false);
2277 }
2278 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
2279
2280 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2281 {
2282         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
2283         struct page *page;
2284
2285         if (vmf->pgoff == 0)
2286                 page = virt_to_page(vcpu->run);
2287 #ifdef CONFIG_X86
2288         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
2289                 page = virt_to_page(vcpu->arch.pio_data);
2290 #endif
2291 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2292         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
2293                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
2294 #endif
2295         else
2296                 return kvm_arch_vcpu_fault(vcpu, vmf);
2297         get_page(page);
2298         vmf->page = page;
2299         return 0;
2300 }
2301
2302 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
2303         .fault = kvm_vcpu_fault,
2304 };
2305
2306 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2307 {
2308         vma->vm_ops = &kvm_vcpu_vm_ops;
2309         return 0;
2310 }
2311
2312 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2313 {
2314         struct kvm_vcpu *vcpu = filp->private_data;
2315
2316         kvm_put_kvm(vcpu->kvm);
2317         return 0;
2318 }
2319
2320 static struct file_operations kvm_vcpu_fops = {
2321         .release        = kvm_vcpu_release,
2322         .unlocked_ioctl = kvm_vcpu_ioctl,
2323 #ifdef CONFIG_KVM_COMPAT
2324         .compat_ioctl   = kvm_vcpu_compat_ioctl,
2325 #endif
2326         .mmap           = kvm_vcpu_mmap,
2327         .llseek         = noop_llseek,
2328 };
2329
2330 /*
2331  * Allocates an inode for the vcpu.
2332  */
2333 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2334 {
2335         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
2336 }
2337
2338 /*
2339  * Creates some virtual cpus.  Good luck creating more than one.
2340  */
2341 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
2342 {
2343         int r;
2344         struct kvm_vcpu *vcpu;
2345
2346         if (id >= KVM_MAX_VCPU_ID)
2347                 return -EINVAL;
2348
2349         vcpu = kvm_arch_vcpu_create(kvm, id);
2350         if (IS_ERR(vcpu))
2351                 return PTR_ERR(vcpu);
2352
2353         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2354
2355         r = kvm_arch_vcpu_setup(vcpu);
2356         if (r)
2357                 goto vcpu_destroy;
2358
2359         mutex_lock(&kvm->lock);
2360         if (!kvm_vcpu_compatible(vcpu)) {
2361                 r = -EINVAL;
2362                 goto unlock_vcpu_destroy;
2363         }
2364         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
2365                 r = -EINVAL;
2366                 goto unlock_vcpu_destroy;
2367         }
2368         if (kvm_get_vcpu_by_id(kvm, id)) {
2369                 r = -EEXIST;
2370                 goto unlock_vcpu_destroy;
2371         }
2372
2373         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
2374
2375         /* Now it's all set up, let userspace reach it */
2376         kvm_get_kvm(kvm);
2377         r = create_vcpu_fd(vcpu);
2378         if (r < 0) {
2379                 kvm_put_kvm(kvm);
2380                 goto unlock_vcpu_destroy;
2381         }
2382
2383         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
2384
2385         /*
2386          * Pairs with smp_rmb() in kvm_get_vcpu.  Write kvm->vcpus
2387          * before kvm->online_vcpu's incremented value.
2388          */
2389         smp_wmb();
2390         atomic_inc(&kvm->online_vcpus);
2391
2392         mutex_unlock(&kvm->lock);
2393         kvm_arch_vcpu_postcreate(vcpu);
2394         return r;
2395
2396 unlock_vcpu_destroy:
2397         mutex_unlock(&kvm->lock);
2398 vcpu_destroy:
2399         kvm_arch_vcpu_destroy(vcpu);
2400         return r;
2401 }
2402
2403 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2404 {
2405         if (sigset) {
2406                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2407                 vcpu->sigset_active = 1;
2408                 vcpu->sigset = *sigset;
2409         } else
2410                 vcpu->sigset_active = 0;
2411         return 0;
2412 }
2413
2414 static long kvm_vcpu_ioctl(struct file *filp,
2415                            unsigned int ioctl, unsigned long arg)
2416 {
2417         struct kvm_vcpu *vcpu = filp->private_data;
2418         void __user *argp = (void __user *)arg;
2419         int r;
2420         struct kvm_fpu *fpu = NULL;
2421         struct kvm_sregs *kvm_sregs = NULL;
2422
2423         if (vcpu->kvm->mm != current->mm)
2424                 return -EIO;
2425
2426         if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2427                 return -EINVAL;
2428
2429 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
2430         /*
2431          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
2432          * so vcpu_load() would break it.
2433          */
2434         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_S390_IRQ || ioctl == KVM_INTERRUPT)
2435                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2436 #endif
2437
2438
2439         r = vcpu_load(vcpu);
2440         if (r)
2441                 return r;
2442         switch (ioctl) {
2443         case KVM_RUN:
2444                 r = -EINVAL;
2445                 if (arg)
2446                         goto out;
2447                 if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
2448                         /* The thread running this VCPU changed. */
2449                         struct pid *oldpid = vcpu->pid;
2450                         struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
2451
2452                         rcu_assign_pointer(vcpu->pid, newpid);
2453                         if (oldpid)
2454                                 synchronize_rcu();
2455                         put_pid(oldpid);
2456                 }
2457                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2458                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2459                 break;
2460         case KVM_GET_REGS: {
2461                 struct kvm_regs *kvm_regs;
2462
2463                 r = -ENOMEM;
2464                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2465                 if (!kvm_regs)
2466                         goto out;
2467                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2468                 if (r)
2469                         goto out_free1;
2470                 r = -EFAULT;
2471                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2472                         goto out_free1;
2473                 r = 0;
2474 out_free1:
2475                 kfree(kvm_regs);
2476                 break;
2477         }
2478         case KVM_SET_REGS: {
2479                 struct kvm_regs *kvm_regs;
2480
2481                 r = -ENOMEM;
2482                 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2483                 if (IS_ERR(kvm_regs)) {
2484                         r = PTR_ERR(kvm_regs);
2485                         goto out;
2486                 }
2487                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2488                 kfree(kvm_regs);
2489                 break;
2490         }
2491         case KVM_GET_SREGS: {
2492                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2493                 r = -ENOMEM;
2494                 if (!kvm_sregs)
2495                         goto out;
2496                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2497                 if (r)
2498                         goto out;
2499                 r = -EFAULT;
2500                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2501                         goto out;
2502                 r = 0;
2503                 break;
2504         }
2505         case KVM_SET_SREGS: {
2506                 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2507                 if (IS_ERR(kvm_sregs)) {
2508                         r = PTR_ERR(kvm_sregs);
2509                         kvm_sregs = NULL;
2510                         goto out;
2511                 }
2512                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2513                 break;
2514         }
2515         case KVM_GET_MP_STATE: {
2516                 struct kvm_mp_state mp_state;
2517
2518                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2519                 if (r)
2520                         goto out;
2521                 r = -EFAULT;
2522                 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
2523                         goto out;
2524                 r = 0;
2525                 break;
2526         }
2527         case KVM_SET_MP_STATE: {
2528                 struct kvm_mp_state mp_state;
2529
2530                 r = -EFAULT;
2531                 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
2532                         goto out;
2533                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2534                 break;
2535         }
2536         case KVM_TRANSLATE: {
2537                 struct kvm_translation tr;
2538
2539                 r = -EFAULT;
2540                 if (copy_from_user(&tr, argp, sizeof(tr)))
2541                         goto out;
2542                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2543                 if (r)
2544                         goto out;
2545                 r = -EFAULT;
2546                 if (copy_to_user(argp, &tr, sizeof(tr)))
2547                         goto out;
2548                 r = 0;
2549                 break;
2550         }
2551         case KVM_SET_GUEST_DEBUG: {
2552                 struct kvm_guest_debug dbg;
2553
2554                 r = -EFAULT;
2555                 if (copy_from_user(&dbg, argp, sizeof(dbg)))
2556                         goto out;
2557                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2558                 break;
2559         }
2560         case KVM_SET_SIGNAL_MASK: {
2561                 struct kvm_signal_mask __user *sigmask_arg = argp;
2562                 struct kvm_signal_mask kvm_sigmask;
2563                 sigset_t sigset, *p;
2564
2565                 p = NULL;
2566                 if (argp) {
2567                         r = -EFAULT;
2568                         if (copy_from_user(&kvm_sigmask, argp,
2569                                            sizeof(kvm_sigmask)))
2570                                 goto out;
2571                         r = -EINVAL;
2572                         if (kvm_sigmask.len != sizeof(sigset))
2573                                 goto out;
2574                         r = -EFAULT;
2575                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2576                                            sizeof(sigset)))
2577                                 goto out;
2578                         p = &sigset;
2579                 }
2580                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2581                 break;
2582         }
2583         case KVM_GET_FPU: {
2584                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2585                 r = -ENOMEM;
2586                 if (!fpu)
2587                         goto out;
2588                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2589                 if (r)
2590                         goto out;
2591                 r = -EFAULT;
2592                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2593                         goto out;
2594                 r = 0;
2595                 break;
2596         }
2597         case KVM_SET_FPU: {
2598                 fpu = memdup_user(argp, sizeof(*fpu));
2599                 if (IS_ERR(fpu)) {
2600                         r = PTR_ERR(fpu);
2601                         fpu = NULL;
2602                         goto out;
2603                 }
2604                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2605                 break;
2606         }
2607         default:
2608                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2609         }
2610 out:
2611         vcpu_put(vcpu);
2612         kfree(fpu);
2613         kfree(kvm_sregs);
2614         return r;
2615 }
2616
2617 #ifdef CONFIG_KVM_COMPAT
2618 static long kvm_vcpu_compat_ioctl(struct file *filp,
2619                                   unsigned int ioctl, unsigned long arg)
2620 {
2621         struct kvm_vcpu *vcpu = filp->private_data;
2622         void __user *argp = compat_ptr(arg);
2623         int r;
2624
2625         if (vcpu->kvm->mm != current->mm)
2626                 return -EIO;
2627
2628         switch (ioctl) {
2629         case KVM_SET_SIGNAL_MASK: {
2630                 struct kvm_signal_mask __user *sigmask_arg = argp;
2631                 struct kvm_signal_mask kvm_sigmask;
2632                 compat_sigset_t csigset;
2633                 sigset_t sigset;
2634
2635                 if (argp) {
2636                         r = -EFAULT;
2637                         if (copy_from_user(&kvm_sigmask, argp,
2638                                            sizeof(kvm_sigmask)))
2639                                 goto out;
2640                         r = -EINVAL;
2641                         if (kvm_sigmask.len != sizeof(csigset))
2642                                 goto out;
2643                         r = -EFAULT;
2644                         if (copy_from_user(&csigset, sigmask_arg->sigset,
2645                                            sizeof(csigset)))
2646                                 goto out;
2647                         sigset_from_compat(&sigset, &csigset);
2648                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2649                 } else
2650                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2651                 break;
2652         }
2653         default:
2654                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2655         }
2656
2657 out:
2658         return r;
2659 }
2660 #endif
2661
2662 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2663                                  int (*accessor)(struct kvm_device *dev,
2664                                                  struct kvm_device_attr *attr),
2665                                  unsigned long arg)
2666 {
2667         struct kvm_device_attr attr;
2668
2669         if (!accessor)
2670                 return -EPERM;
2671
2672         if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2673                 return -EFAULT;
2674
2675         return accessor(dev, &attr);
2676 }
2677
2678 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2679                              unsigned long arg)
2680 {
2681         struct kvm_device *dev = filp->private_data;
2682
2683         switch (ioctl) {
2684         case KVM_SET_DEVICE_ATTR:
2685                 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2686         case KVM_GET_DEVICE_ATTR:
2687                 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2688         case KVM_HAS_DEVICE_ATTR:
2689                 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2690         default:
2691                 if (dev->ops->ioctl)
2692                         return dev->ops->ioctl(dev, ioctl, arg);
2693
2694                 return -ENOTTY;
2695         }
2696 }
2697
2698 static int kvm_device_release(struct inode *inode, struct file *filp)
2699 {
2700         struct kvm_device *dev = filp->private_data;
2701         struct kvm *kvm = dev->kvm;
2702
2703         kvm_put_kvm(kvm);
2704         return 0;
2705 }
2706
2707 static const struct file_operations kvm_device_fops = {
2708         .unlocked_ioctl = kvm_device_ioctl,
2709 #ifdef CONFIG_KVM_COMPAT
2710         .compat_ioctl = kvm_device_ioctl,
2711 #endif
2712         .release = kvm_device_release,
2713 };
2714
2715 struct kvm_device *kvm_device_from_filp(struct file *filp)
2716 {
2717         if (filp->f_op != &kvm_device_fops)
2718                 return NULL;
2719
2720         return filp->private_data;
2721 }
2722
2723 static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
2724 #ifdef CONFIG_KVM_MPIC
2725         [KVM_DEV_TYPE_FSL_MPIC_20]      = &kvm_mpic_ops,
2726         [KVM_DEV_TYPE_FSL_MPIC_42]      = &kvm_mpic_ops,
2727 #endif
2728
2729 #ifdef CONFIG_KVM_XICS
2730         [KVM_DEV_TYPE_XICS]             = &kvm_xics_ops,
2731 #endif
2732 };
2733
2734 int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
2735 {
2736         if (type >= ARRAY_SIZE(kvm_device_ops_table))
2737                 return -ENOSPC;
2738
2739         if (kvm_device_ops_table[type] != NULL)
2740                 return -EEXIST;
2741
2742         kvm_device_ops_table[type] = ops;
2743         return 0;
2744 }
2745
2746 void kvm_unregister_device_ops(u32 type)
2747 {
2748         if (kvm_device_ops_table[type] != NULL)
2749                 kvm_device_ops_table[type] = NULL;
2750 }
2751
2752 static int kvm_ioctl_create_device(struct kvm *kvm,
2753                                    struct kvm_create_device *cd)
2754 {
2755         struct kvm_device_ops *ops = NULL;
2756         struct kvm_device *dev;
2757         bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2758         int ret;
2759
2760         if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
2761                 return -ENODEV;
2762
2763         ops = kvm_device_ops_table[cd->type];
2764         if (ops == NULL)
2765                 return -ENODEV;
2766
2767         if (test)
2768                 return 0;
2769
2770         dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2771         if (!dev)
2772                 return -ENOMEM;
2773
2774         dev->ops = ops;
2775         dev->kvm = kvm;
2776
2777         ret = ops->create(dev, cd->type);
2778         if (ret < 0) {
2779                 kfree(dev);
2780                 return ret;
2781         }
2782
2783         ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2784         if (ret < 0) {
2785                 ops->destroy(dev);
2786                 return ret;
2787         }
2788
2789         list_add(&dev->vm_node, &kvm->devices);
2790         kvm_get_kvm(kvm);
2791         cd->fd = ret;
2792         return 0;
2793 }
2794
2795 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2796 {
2797         switch (arg) {
2798         case KVM_CAP_USER_MEMORY:
2799         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2800         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2801         case KVM_CAP_INTERNAL_ERROR_DATA:
2802 #ifdef CONFIG_HAVE_KVM_MSI
2803         case KVM_CAP_SIGNAL_MSI:
2804 #endif
2805 #ifdef CONFIG_HAVE_KVM_IRQFD
2806         case KVM_CAP_IRQFD:
2807         case KVM_CAP_IRQFD_RESAMPLE:
2808 #endif
2809         case KVM_CAP_IOEVENTFD_ANY_LENGTH:
2810         case KVM_CAP_CHECK_EXTENSION_VM:
2811                 return 1;
2812 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2813         case KVM_CAP_IRQ_ROUTING:
2814                 return KVM_MAX_IRQ_ROUTES;
2815 #endif
2816 #if KVM_ADDRESS_SPACE_NUM > 1
2817         case KVM_CAP_MULTI_ADDRESS_SPACE:
2818                 return KVM_ADDRESS_SPACE_NUM;
2819 #endif
2820         case KVM_CAP_MAX_VCPU_ID:
2821                 return KVM_MAX_VCPU_ID;
2822         default:
2823                 break;
2824         }
2825         return kvm_vm_ioctl_check_extension(kvm, arg);
2826 }
2827
2828 static long kvm_vm_ioctl(struct file *filp,
2829                            unsigned int ioctl, unsigned long arg)
2830 {
2831         struct kvm *kvm = filp->private_data;
2832         void __user *argp = (void __user *)arg;
2833         int r;
2834
2835         if (kvm->mm != current->mm)
2836                 return -EIO;
2837         switch (ioctl) {
2838         case KVM_CREATE_VCPU:
2839                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2840                 break;
2841         case KVM_SET_USER_MEMORY_REGION: {
2842                 struct kvm_userspace_memory_region kvm_userspace_mem;
2843
2844                 r = -EFAULT;
2845                 if (copy_from_user(&kvm_userspace_mem, argp,
2846                                                 sizeof(kvm_userspace_mem)))
2847                         goto out;
2848
2849                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2850                 break;
2851         }
2852         case KVM_GET_DIRTY_LOG: {
2853                 struct kvm_dirty_log log;
2854
2855                 r = -EFAULT;
2856                 if (copy_from_user(&log, argp, sizeof(log)))
2857                         goto out;
2858                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2859                 break;
2860         }
2861 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2862         case KVM_REGISTER_COALESCED_MMIO: {
2863                 struct kvm_coalesced_mmio_zone zone;
2864
2865                 r = -EFAULT;
2866                 if (copy_from_user(&zone, argp, sizeof(zone)))
2867                         goto out;
2868                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2869                 break;
2870         }
2871         case KVM_UNREGISTER_COALESCED_MMIO: {
2872                 struct kvm_coalesced_mmio_zone zone;
2873
2874                 r = -EFAULT;
2875                 if (copy_from_user(&zone, argp, sizeof(zone)))
2876                         goto out;
2877                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2878                 break;
2879         }
2880 #endif
2881         case KVM_IRQFD: {
2882                 struct kvm_irqfd data;
2883
2884                 r = -EFAULT;
2885                 if (copy_from_user(&data, argp, sizeof(data)))
2886                         goto out;
2887                 r = kvm_irqfd(kvm, &data);
2888                 break;
2889         }
2890         case KVM_IOEVENTFD: {
2891                 struct kvm_ioeventfd data;
2892
2893                 r = -EFAULT;
2894                 if (copy_from_user(&data, argp, sizeof(data)))
2895                         goto out;
2896                 r = kvm_ioeventfd(kvm, &data);
2897                 break;
2898         }
2899 #ifdef CONFIG_HAVE_KVM_MSI
2900         case KVM_SIGNAL_MSI: {
2901                 struct kvm_msi msi;
2902
2903                 r = -EFAULT;
2904                 if (copy_from_user(&msi, argp, sizeof(msi)))
2905                         goto out;
2906                 r = kvm_send_userspace_msi(kvm, &msi);
2907                 break;
2908         }
2909 #endif
2910 #ifdef __KVM_HAVE_IRQ_LINE
2911         case KVM_IRQ_LINE_STATUS:
2912         case KVM_IRQ_LINE: {
2913                 struct kvm_irq_level irq_event;
2914
2915                 r = -EFAULT;
2916                 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
2917                         goto out;
2918
2919                 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
2920                                         ioctl == KVM_IRQ_LINE_STATUS);
2921                 if (r)
2922                         goto out;
2923
2924                 r = -EFAULT;
2925                 if (ioctl == KVM_IRQ_LINE_STATUS) {
2926                         if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
2927                                 goto out;
2928                 }
2929
2930                 r = 0;
2931                 break;
2932         }
2933 #endif
2934 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2935         case KVM_SET_GSI_ROUTING: {
2936                 struct kvm_irq_routing routing;
2937                 struct kvm_irq_routing __user *urouting;
2938                 struct kvm_irq_routing_entry *entries = NULL;
2939
2940                 r = -EFAULT;
2941                 if (copy_from_user(&routing, argp, sizeof(routing)))
2942                         goto out;
2943                 r = -EINVAL;
2944                 if (routing.nr > KVM_MAX_IRQ_ROUTES)
2945                         goto out;
2946                 if (routing.flags)
2947                         goto out;
2948                 if (routing.nr) {
2949                         r = -ENOMEM;
2950                         entries = vmalloc(routing.nr * sizeof(*entries));
2951                         if (!entries)
2952                                 goto out;
2953                         r = -EFAULT;
2954                         urouting = argp;
2955                         if (copy_from_user(entries, urouting->entries,
2956                                            routing.nr * sizeof(*entries)))
2957                                 goto out_free_irq_routing;
2958                 }
2959                 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2960                                         routing.flags);
2961 out_free_irq_routing:
2962                 vfree(entries);
2963                 break;
2964         }
2965 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
2966         case KVM_CREATE_DEVICE: {
2967                 struct kvm_create_device cd;
2968
2969                 r = -EFAULT;
2970                 if (copy_from_user(&cd, argp, sizeof(cd)))
2971                         goto out;
2972
2973                 r = kvm_ioctl_create_device(kvm, &cd);
2974                 if (r)
2975                         goto out;
2976
2977                 r = -EFAULT;
2978                 if (copy_to_user(argp, &cd, sizeof(cd)))
2979                         goto out;
2980
2981                 r = 0;
2982                 break;
2983         }
2984         case KVM_CHECK_EXTENSION:
2985                 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
2986                 break;
2987         default:
2988                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2989         }
2990 out:
2991         return r;
2992 }
2993
2994 #ifdef CONFIG_KVM_COMPAT
2995 struct compat_kvm_dirty_log {
2996         __u32 slot;
2997         __u32 padding1;
2998         union {
2999                 compat_uptr_t dirty_bitmap; /* one bit per page */
3000                 __u64 padding2;
3001         };
3002 };
3003
3004 static long kvm_vm_compat_ioctl(struct file *filp,
3005                            unsigned int ioctl, unsigned long arg)
3006 {
3007         struct kvm *kvm = filp->private_data;
3008         int r;
3009
3010         if (kvm->mm != current->mm)
3011                 return -EIO;
3012         switch (ioctl) {
3013         case KVM_GET_DIRTY_LOG: {
3014                 struct compat_kvm_dirty_log compat_log;
3015                 struct kvm_dirty_log log;
3016
3017                 r = -EFAULT;
3018                 if (copy_from_user(&compat_log, (void __user *)arg,
3019                                    sizeof(compat_log)))
3020                         goto out;
3021                 log.slot         = compat_log.slot;
3022                 log.padding1     = compat_log.padding1;
3023                 log.padding2     = compat_log.padding2;
3024                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
3025
3026                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3027                 break;
3028         }
3029         default:
3030                 r = kvm_vm_ioctl(filp, ioctl, arg);
3031         }
3032
3033 out:
3034         return r;
3035 }
3036 #endif
3037
3038 static struct file_operations kvm_vm_fops = {
3039         .release        = kvm_vm_release,
3040         .unlocked_ioctl = kvm_vm_ioctl,
3041 #ifdef CONFIG_KVM_COMPAT
3042         .compat_ioctl   = kvm_vm_compat_ioctl,
3043 #endif
3044         .llseek         = noop_llseek,
3045 };
3046
3047 static int kvm_dev_ioctl_create_vm(unsigned long type)
3048 {
3049         int r;
3050         struct kvm *kvm;
3051
3052         kvm = kvm_create_vm(type);
3053         if (IS_ERR(kvm))
3054                 return PTR_ERR(kvm);
3055 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
3056         r = kvm_coalesced_mmio_init(kvm);
3057         if (r < 0) {
3058                 kvm_put_kvm(kvm);
3059                 return r;
3060         }
3061 #endif
3062         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC);
3063         if (r < 0) {
3064                 kvm_put_kvm(kvm);
3065                 return r;
3066         }
3067
3068         if (kvm_create_vm_debugfs(kvm, r) < 0) {
3069                 kvm_put_kvm(kvm);
3070                 return -ENOMEM;
3071         }
3072
3073         return r;
3074 }
3075
3076 static long kvm_dev_ioctl(struct file *filp,
3077                           unsigned int ioctl, unsigned long arg)
3078 {
3079         long r = -EINVAL;
3080
3081         switch (ioctl) {
3082         case KVM_GET_API_VERSION:
3083                 if (arg)
3084                         goto out;
3085                 r = KVM_API_VERSION;
3086                 break;
3087         case KVM_CREATE_VM:
3088                 r = kvm_dev_ioctl_create_vm(arg);
3089                 break;
3090         case KVM_CHECK_EXTENSION:
3091                 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
3092                 break;
3093         case KVM_GET_VCPU_MMAP_SIZE:
3094                 if (arg)
3095                         goto out;
3096                 r = PAGE_SIZE;     /* struct kvm_run */
3097 #ifdef CONFIG_X86
3098                 r += PAGE_SIZE;    /* pio data page */
3099 #endif
3100 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
3101                 r += PAGE_SIZE;    /* coalesced mmio ring page */
3102 #endif
3103                 break;
3104         case KVM_TRACE_ENABLE:
3105         case KVM_TRACE_PAUSE:
3106         case KVM_TRACE_DISABLE:
3107                 r = -EOPNOTSUPP;
3108                 break;
3109         default:
3110                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
3111         }
3112 out:
3113         return r;
3114 }
3115
3116 static struct file_operations kvm_chardev_ops = {
3117         .unlocked_ioctl = kvm_dev_ioctl,
3118         .compat_ioctl   = kvm_dev_ioctl,
3119         .llseek         = noop_llseek,
3120 };
3121
3122 static struct miscdevice kvm_dev = {
3123         KVM_MINOR,
3124         "kvm",
3125         &kvm_chardev_ops,
3126 };
3127
3128 static void hardware_enable_nolock(void *junk)
3129 {
3130         int cpu = raw_smp_processor_id();
3131         int r;
3132
3133         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
3134                 return;
3135
3136         cpumask_set_cpu(cpu, cpus_hardware_enabled);
3137
3138         r = kvm_arch_hardware_enable();
3139
3140         if (r) {
3141                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3142                 atomic_inc(&hardware_enable_failed);
3143                 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
3144         }
3145 }
3146
3147 static void hardware_enable(void)
3148 {
3149         raw_spin_lock(&kvm_count_lock);
3150         if (kvm_usage_count)
3151                 hardware_enable_nolock(NULL);
3152         raw_spin_unlock(&kvm_count_lock);
3153 }
3154
3155 static void hardware_disable_nolock(void *junk)
3156 {
3157         int cpu = raw_smp_processor_id();
3158
3159         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
3160                 return;
3161         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3162         kvm_arch_hardware_disable();
3163 }
3164
3165 static void hardware_disable(void)
3166 {
3167         raw_spin_lock(&kvm_count_lock);
3168         if (kvm_usage_count)
3169                 hardware_disable_nolock(NULL);
3170         raw_spin_unlock(&kvm_count_lock);
3171 }
3172
3173 static void hardware_disable_all_nolock(void)
3174 {
3175         BUG_ON(!kvm_usage_count);
3176
3177         kvm_usage_count--;
3178         if (!kvm_usage_count)
3179                 on_each_cpu(hardware_disable_nolock, NULL, 1);
3180 }
3181
3182 static void hardware_disable_all(void)
3183 {
3184         raw_spin_lock(&kvm_count_lock);
3185         hardware_disable_all_nolock();
3186         raw_spin_unlock(&kvm_count_lock);
3187 }
3188
3189 static int hardware_enable_all(void)
3190 {
3191         int r = 0;
3192
3193         raw_spin_lock(&kvm_count_lock);
3194
3195         kvm_usage_count++;
3196         if (kvm_usage_count == 1) {
3197                 atomic_set(&hardware_enable_failed, 0);
3198                 on_each_cpu(hardware_enable_nolock, NULL, 1);
3199
3200                 if (atomic_read(&hardware_enable_failed)) {
3201                         hardware_disable_all_nolock();
3202                         r = -EBUSY;
3203                 }
3204         }
3205
3206         raw_spin_unlock(&kvm_count_lock);
3207
3208         return r;
3209 }
3210
3211 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
3212                            void *v)
3213 {
3214         val &= ~CPU_TASKS_FROZEN;
3215         switch (val) {
3216         case CPU_DYING:
3217                 hardware_disable();
3218                 break;
3219         case CPU_STARTING:
3220                 hardware_enable();
3221                 break;
3222         }
3223         return NOTIFY_OK;
3224 }
3225
3226 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3227                       void *v)
3228 {
3229         /*
3230          * Some (well, at least mine) BIOSes hang on reboot if
3231          * in vmx root mode.
3232          *
3233          * And Intel TXT required VMX off for all cpu when system shutdown.
3234          */
3235         pr_info("kvm: exiting hardware virtualization\n");
3236         kvm_rebooting = true;
3237         on_each_cpu(hardware_disable_nolock, NULL, 1);
3238         return NOTIFY_OK;
3239 }
3240
3241 static struct notifier_block kvm_reboot_notifier = {
3242         .notifier_call = kvm_reboot,
3243         .priority = 0,
3244 };
3245
3246 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3247 {
3248         int i;
3249
3250         for (i = 0; i < bus->dev_count; i++) {
3251                 struct kvm_io_device *pos = bus->range[i].dev;
3252
3253                 kvm_iodevice_destructor(pos);
3254         }
3255         kfree(bus);
3256 }
3257
3258 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
3259                                  const struct kvm_io_range *r2)
3260 {
3261         gpa_t addr1 = r1->addr;
3262         gpa_t addr2 = r2->addr;
3263
3264         if (addr1 < addr2)
3265                 return -1;
3266
3267         /* If r2->len == 0, match the exact address.  If r2->len != 0,
3268          * accept any overlapping write.  Any order is acceptable for
3269          * overlapping ranges, because kvm_io_bus_get_first_dev ensures
3270          * we process all of them.
3271          */
3272         if (r2->len) {
3273                 addr1 += r1->len;
3274                 addr2 += r2->len;
3275         }
3276
3277         if (addr1 > addr2)
3278                 return 1;
3279
3280         return 0;
3281 }
3282
3283 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
3284 {
3285         return kvm_io_bus_cmp(p1, p2);
3286 }
3287
3288 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
3289                           gpa_t addr, int len)
3290 {
3291         bus->range[bus->dev_count++] = (struct kvm_io_range) {
3292                 .addr = addr,
3293                 .len = len,
3294                 .dev = dev,
3295         };
3296
3297         sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
3298                 kvm_io_bus_sort_cmp, NULL);
3299
3300         return 0;
3301 }
3302
3303 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
3304                              gpa_t addr, int len)
3305 {
3306         struct kvm_io_range *range, key;
3307         int off;
3308
3309         key = (struct kvm_io_range) {
3310                 .addr = addr,
3311                 .len = len,
3312         };
3313
3314         range = bsearch(&key, bus->range, bus->dev_count,
3315                         sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
3316         if (range == NULL)
3317                 return -ENOENT;
3318
3319         off = range - bus->range;
3320
3321         while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
3322                 off--;
3323
3324         return off;
3325 }
3326
3327 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3328                               struct kvm_io_range *range, const void *val)
3329 {
3330         int idx;
3331
3332         idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3333         if (idx < 0)
3334                 return -EOPNOTSUPP;
3335
3336         while (idx < bus->dev_count &&
3337                 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3338                 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
3339                                         range->len, val))
3340                         return idx;
3341                 idx++;
3342         }
3343
3344         return -EOPNOTSUPP;
3345 }
3346
3347 /* kvm_io_bus_write - called under kvm->slots_lock */
3348 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3349                      int len, const void *val)
3350 {
3351         struct kvm_io_bus *bus;
3352         struct kvm_io_range range;
3353         int r;
3354
3355         range = (struct kvm_io_range) {
3356                 .addr = addr,
3357                 .len = len,
3358         };
3359
3360         bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3361         r = __kvm_io_bus_write(vcpu, bus, &range, val);
3362         return r < 0 ? r : 0;
3363 }
3364
3365 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3366 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
3367                             gpa_t addr, int len, const void *val, long cookie)
3368 {
3369         struct kvm_io_bus *bus;
3370         struct kvm_io_range range;
3371
3372         range = (struct kvm_io_range) {
3373                 .addr = addr,
3374                 .len = len,
3375         };
3376
3377         bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3378
3379         /* First try the device referenced by cookie. */
3380         if ((cookie >= 0) && (cookie < bus->dev_count) &&
3381             (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
3382                 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
3383                                         val))
3384                         return cookie;
3385
3386         /*
3387          * cookie contained garbage; fall back to search and return the
3388          * correct cookie value.
3389          */
3390         return __kvm_io_bus_write(vcpu, bus, &range, val);
3391 }
3392
3393 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3394                              struct kvm_io_range *range, void *val)
3395 {
3396         int idx;
3397
3398         idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3399         if (idx < 0)
3400                 return -EOPNOTSUPP;
3401
3402         while (idx < bus->dev_count &&
3403                 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3404                 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
3405                                        range->len, val))
3406                         return idx;
3407                 idx++;
3408         }
3409
3410         return -EOPNOTSUPP;
3411 }
3412 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
3413
3414 /* kvm_io_bus_read - called under kvm->slots_lock */
3415 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3416                     int len, void *val)
3417 {
3418         struct kvm_io_bus *bus;
3419         struct kvm_io_range range;
3420         int r;
3421
3422         range = (struct kvm_io_range) {
3423                 .addr = addr,
3424                 .len = len,
3425         };
3426
3427         bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3428         r = __kvm_io_bus_read(vcpu, bus, &range, val);
3429         return r < 0 ? r : 0;
3430 }
3431
3432
3433 /* Caller must hold slots_lock. */
3434 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3435                             int len, struct kvm_io_device *dev)
3436 {
3437         struct kvm_io_bus *new_bus, *bus;
3438
3439         bus = kvm->buses[bus_idx];
3440         /* exclude ioeventfd which is limited by maximum fd */
3441         if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3442                 return -ENOSPC;
3443
3444         new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count + 1) *
3445                           sizeof(struct kvm_io_range)), GFP_KERNEL);
3446         if (!new_bus)
3447                 return -ENOMEM;
3448         memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
3449                sizeof(struct kvm_io_range)));
3450         kvm_io_bus_insert_dev(new_bus, dev, addr, len);
3451         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3452         synchronize_srcu_expedited(&kvm->srcu);
3453         kfree(bus);
3454
3455         return 0;
3456 }
3457
3458 /* Caller must hold slots_lock. */
3459 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3460                               struct kvm_io_device *dev)
3461 {
3462         int i, r;
3463         struct kvm_io_bus *new_bus, *bus;
3464
3465         bus = kvm->buses[bus_idx];
3466         r = -ENOENT;
3467         for (i = 0; i < bus->dev_count; i++)
3468                 if (bus->range[i].dev == dev) {
3469                         r = 0;
3470                         break;
3471                 }
3472
3473         if (r)
3474                 return r;
3475
3476         new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3477                           sizeof(struct kvm_io_range)), GFP_KERNEL);
3478         if (!new_bus)
3479                 return -ENOMEM;
3480
3481         memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3482         new_bus->dev_count--;
3483         memcpy(new_bus->range + i, bus->range + i + 1,
3484                (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3485
3486         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3487         synchronize_srcu_expedited(&kvm->srcu);
3488         kfree(bus);
3489         return r;
3490 }
3491
3492 static struct notifier_block kvm_cpu_notifier = {
3493         .notifier_call = kvm_cpu_hotplug,
3494 };
3495
3496 static int kvm_debugfs_open(struct inode *inode, struct file *file,
3497                            int (*get)(void *, u64 *), int (*set)(void *, u64),
3498                            const char *fmt)
3499 {
3500         struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
3501                                           inode->i_private;
3502
3503         /* The debugfs files are a reference to the kvm struct which
3504          * is still valid when kvm_destroy_vm is called.
3505          * To avoid the race between open and the removal of the debugfs
3506          * directory we test against the users count.
3507          */
3508         if (!atomic_add_unless(&stat_data->kvm->users_count, 1, 0))
3509                 return -ENOENT;
3510
3511         if (simple_attr_open(inode, file, get, set, fmt)) {
3512                 kvm_put_kvm(stat_data->kvm);
3513                 return -ENOMEM;
3514         }
3515
3516         return 0;
3517 }
3518
3519 static int kvm_debugfs_release(struct inode *inode, struct file *file)
3520 {
3521         struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
3522                                           inode->i_private;
3523
3524         simple_attr_release(inode, file);
3525         kvm_put_kvm(stat_data->kvm);
3526
3527         return 0;
3528 }
3529
3530 static int vm_stat_get_per_vm(void *data, u64 *val)
3531 {
3532         struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3533
3534         *val = *(u32 *)((void *)stat_data->kvm + stat_data->offset);
3535
3536         return 0;
3537 }
3538
3539 static int vm_stat_get_per_vm_open(struct inode *inode, struct file *file)
3540 {
3541         __simple_attr_check_format("%llu\n", 0ull);
3542         return kvm_debugfs_open(inode, file, vm_stat_get_per_vm,
3543                                 NULL, "%llu\n");
3544 }
3545
3546 static const struct file_operations vm_stat_get_per_vm_fops = {
3547         .owner   = THIS_MODULE,
3548         .open    = vm_stat_get_per_vm_open,
3549         .release = kvm_debugfs_release,
3550         .read    = simple_attr_read,
3551         .write   = simple_attr_write,
3552         .llseek  = generic_file_llseek,
3553 };
3554
3555 static int vcpu_stat_get_per_vm(void *data, u64 *val)
3556 {
3557         int i;
3558         struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3559         struct kvm_vcpu *vcpu;
3560
3561         *val = 0;
3562
3563         kvm_for_each_vcpu(i, vcpu, stat_data->kvm)
3564                 *val += *(u32 *)((void *)vcpu + stat_data->offset);
3565
3566         return 0;
3567 }
3568
3569 static int vcpu_stat_get_per_vm_open(struct inode *inode, struct file *file)
3570 {
3571         __simple_attr_check_format("%llu\n", 0ull);
3572         return kvm_debugfs_open(inode, file, vcpu_stat_get_per_vm,
3573                                  NULL, "%llu\n");
3574 }
3575
3576 static const struct file_operations vcpu_stat_get_per_vm_fops = {
3577         .owner   = THIS_MODULE,
3578         .open    = vcpu_stat_get_per_vm_open,
3579         .release = kvm_debugfs_release,
3580         .read    = simple_attr_read,
3581         .write   = simple_attr_write,
3582         .llseek  = generic_file_llseek,
3583 };
3584
3585 static const struct file_operations *stat_fops_per_vm[] = {
3586         [KVM_STAT_VCPU] = &vcpu_stat_get_per_vm_fops,
3587         [KVM_STAT_VM]   = &vm_stat_get_per_vm_fops,
3588 };
3589
3590 static int vm_stat_get(void *_offset, u64 *val)
3591 {
3592         unsigned offset = (long)_offset;
3593         struct kvm *kvm;
3594         struct kvm_stat_data stat_tmp = {.offset = offset};
3595         u64 tmp_val;
3596
3597         *val = 0;
3598         spin_lock(&kvm_lock);
3599         list_for_each_entry(kvm, &vm_list, vm_list) {
3600                 stat_tmp.kvm = kvm;
3601                 vm_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
3602                 *val += tmp_val;
3603         }
3604         spin_unlock(&kvm_lock);
3605         return 0;
3606 }
3607
3608 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
3609
3610 static int vcpu_stat_get(void *_offset, u64 *val)
3611 {
3612         unsigned offset = (long)_offset;
3613         struct kvm *kvm;
3614         struct kvm_stat_data stat_tmp = {.offset = offset};
3615         u64 tmp_val;
3616
3617         *val = 0;
3618         spin_lock(&kvm_lock);
3619         list_for_each_entry(kvm, &vm_list, vm_list) {
3620                 stat_tmp.kvm = kvm;
3621                 vcpu_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
3622                 *val += tmp_val;
3623         }
3624         spin_unlock(&kvm_lock);
3625         return 0;
3626 }
3627
3628 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
3629
3630 static const struct file_operations *stat_fops[] = {
3631         [KVM_STAT_VCPU] = &vcpu_stat_fops,
3632         [KVM_STAT_VM]   = &vm_stat_fops,
3633 };
3634
3635 static int kvm_init_debug(void)
3636 {
3637         int r = -EEXIST;
3638         struct kvm_stats_debugfs_item *p;
3639
3640         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3641         if (kvm_debugfs_dir == NULL)
3642                 goto out;
3643
3644         kvm_debugfs_num_entries = 0;
3645         for (p = debugfs_entries; p->name; ++p, kvm_debugfs_num_entries++) {
3646                 if (!debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
3647                                          (void *)(long)p->offset,
3648                                          stat_fops[p->kind]))
3649                         goto out_dir;
3650         }
3651
3652         return 0;
3653
3654 out_dir:
3655         debugfs_remove_recursive(kvm_debugfs_dir);
3656 out:
3657         return r;
3658 }
3659
3660 static int kvm_suspend(void)
3661 {
3662         if (kvm_usage_count)
3663                 hardware_disable_nolock(NULL);
3664         return 0;
3665 }
3666
3667 static void kvm_resume(void)
3668 {
3669         if (kvm_usage_count) {
3670                 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3671                 hardware_enable_nolock(NULL);
3672         }
3673 }
3674
3675 static struct syscore_ops kvm_syscore_ops = {
3676         .suspend = kvm_suspend,
3677         .resume = kvm_resume,
3678 };
3679
3680 static inline
3681 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3682 {
3683         return container_of(pn, struct kvm_vcpu, preempt_notifier);
3684 }
3685
3686 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3687 {
3688         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3689
3690         if (vcpu->preempted)
3691                 vcpu->preempted = false;
3692
3693         kvm_arch_sched_in(vcpu, cpu);
3694
3695         kvm_arch_vcpu_load(vcpu, cpu);
3696 }
3697
3698 static void kvm_sched_out(struct preempt_notifier *pn,
3699                           struct task_struct *next)
3700 {
3701         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3702
3703         if (current->state == TASK_RUNNING)
3704                 vcpu->preempted = true;
3705         kvm_arch_vcpu_put(vcpu);
3706 }
3707
3708 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3709                   struct module *module)
3710 {
3711         int r;
3712         int cpu;
3713
3714         r = kvm_arch_init(opaque);
3715         if (r)
3716                 goto out_fail;
3717
3718         /*
3719          * kvm_arch_init makes sure there's at most one caller
3720          * for architectures that support multiple implementations,
3721          * like intel and amd on x86.
3722          * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3723          * conflicts in case kvm is already setup for another implementation.
3724          */
3725         r = kvm_irqfd_init();
3726         if (r)
3727                 goto out_irqfd;
3728
3729         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
3730                 r = -ENOMEM;
3731                 goto out_free_0;
3732         }
3733
3734         r = kvm_arch_hardware_setup();
3735         if (r < 0)
3736                 goto out_free_0a;
3737
3738         for_each_online_cpu(cpu) {
3739                 smp_call_function_single(cpu,
3740                                 kvm_arch_check_processor_compat,
3741                                 &r, 1);
3742                 if (r < 0)
3743                         goto out_free_1;
3744         }
3745
3746         r = register_cpu_notifier(&kvm_cpu_notifier);
3747         if (r)
3748                 goto out_free_2;
3749         register_reboot_notifier(&kvm_reboot_notifier);
3750
3751         /* A kmem cache lets us meet the alignment requirements of fx_save. */
3752         if (!vcpu_align)
3753                 vcpu_align = __alignof__(struct kvm_vcpu);
3754         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
3755                                            0, NULL);
3756         if (!kvm_vcpu_cache) {
3757                 r = -ENOMEM;
3758                 goto out_free_3;
3759         }
3760
3761         r = kvm_async_pf_init();
3762         if (r)
3763                 goto out_free;
3764
3765         kvm_chardev_ops.owner = module;
3766         kvm_vm_fops.owner = module;
3767         kvm_vcpu_fops.owner = module;
3768
3769         r = misc_register(&kvm_dev);
3770         if (r) {
3771                 pr_err("kvm: misc device register failed\n");
3772                 goto out_unreg;
3773         }
3774
3775         register_syscore_ops(&kvm_syscore_ops);
3776
3777         kvm_preempt_ops.sched_in = kvm_sched_in;
3778         kvm_preempt_ops.sched_out = kvm_sched_out;
3779
3780         r = kvm_init_debug();
3781         if (r) {
3782                 pr_err("kvm: create debugfs files failed\n");
3783                 goto out_undebugfs;
3784         }
3785
3786         r = kvm_vfio_ops_init();
3787         WARN_ON(r);
3788
3789         return 0;
3790
3791 out_undebugfs:
3792         unregister_syscore_ops(&kvm_syscore_ops);
3793         misc_deregister(&kvm_dev);
3794 out_unreg:
3795         kvm_async_pf_deinit();
3796 out_free:
3797         kmem_cache_destroy(kvm_vcpu_cache);
3798 out_free_3:
3799         unregister_reboot_notifier(&kvm_reboot_notifier);
3800         unregister_cpu_notifier(&kvm_cpu_notifier);
3801 out_free_2:
3802 out_free_1:
3803         kvm_arch_hardware_unsetup();
3804 out_free_0a:
3805         free_cpumask_var(cpus_hardware_enabled);
3806 out_free_0:
3807         kvm_irqfd_exit();
3808 out_irqfd:
3809         kvm_arch_exit();
3810 out_fail:
3811         return r;
3812 }
3813 EXPORT_SYMBOL_GPL(kvm_init);
3814
3815 void kvm_exit(void)
3816 {
3817         debugfs_remove_recursive(kvm_debugfs_dir);
3818         misc_deregister(&kvm_dev);
3819         kmem_cache_destroy(kvm_vcpu_cache);
3820         kvm_async_pf_deinit();
3821         unregister_syscore_ops(&kvm_syscore_ops);
3822         unregister_reboot_notifier(&kvm_reboot_notifier);
3823         unregister_cpu_notifier(&kvm_cpu_notifier);
3824         on_each_cpu(hardware_disable_nolock, NULL, 1);
3825         kvm_arch_hardware_unsetup();
3826         kvm_arch_exit();
3827         kvm_irqfd_exit();
3828         free_cpumask_var(cpus_hardware_enabled);
3829         kvm_vfio_ops_exit();
3830 }
3831 EXPORT_SYMBOL_GPL(kvm_exit);