virt/kvm/kvm_main.c

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * This module enables machines with Intel VT-x extensions to run virtual
   5  * machines without emulation or binary translation.
   6  *
   7  * Copyright (C) 2006 Qumranet, Inc.
   8  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
   9  *
  10  * Authors:
  11  *   Avi Kivity   <avi@qumranet.com>
  12  *   Yaniv Kamay  <yaniv@qumranet.com>
  13  *
  14  * This work is licensed under the terms of the GNU GPL, version 2.  See
  15  * the COPYING file in the top-level directory.
  16  *
  17  */
  18
  19 #include "iodev.h"
  20
  21 #include <linux/kvm_host.h>
  22 #include <linux/kvm.h>
  23 #include <linux/module.h>
  24 #include <linux/errno.h>
  25 #include <linux/percpu.h>
  26 #include <linux/mm.h>
  27 #include <linux/miscdevice.h>
  28 #include <linux/vmalloc.h>
  29 #include <linux/reboot.h>
  30 #include <linux/debugfs.h>
  31 #include <linux/highmem.h>
  32 #include <linux/file.h>
  33 #include <linux/syscore_ops.h>
  34 #include <linux/cpu.h>
  35 #include <linux/sched.h>
  36 #include <linux/cpumask.h>
  37 #include <linux/smp.h>
  38 #include <linux/anon_inodes.h>
  39 #include <linux/profile.h>
  40 #include <linux/kvm_para.h>
  41 #include <linux/pagemap.h>
  42 #include <linux/mman.h>
  43 #include <linux/swap.h>
  44 #include <linux/bitops.h>
  45 #include <linux/spinlock.h>
  46 #include <linux/compat.h>
  47 #include <linux/srcu.h>
  48 #include <linux/hugetlb.h>
  49 #include <linux/slab.h>
  50 #include <linux/sort.h>
  51 #include <linux/bsearch.h>
  52
  53 #include <asm/processor.h>
  54 #include <asm/io.h>
  55 #include <asm/uaccess.h>
  56 #include <asm/pgtable.h>
  57
  58 #include "coalesced_mmio.h"
  59 #include "async_pf.h"
  60
  61 #define CREATE_TRACE_POINTS
  62 #include <trace/events/kvm.h>
  63
  64 MODULE_AUTHOR("Qumranet");
  65 MODULE_LICENSE("GPL");
  66
  67 /*
  68  * Ordering of locks:
  69  *
  70  *              kvm->lock --> kvm->slots_lock --> kvm->irq_lock
  71  */
  72
  73 DEFINE_SPINLOCK(kvm_lock);
  74 static DEFINE_RAW_SPINLOCK(kvm_count_lock);
  75 LIST_HEAD(vm_list);
  76
  77 static cpumask_var_t cpus_hardware_enabled;
  78 static int kvm_usage_count = 0;
  79 static atomic_t hardware_enable_failed;
  80
  81 struct kmem_cache *kvm_vcpu_cache;
  82 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
  83
  84 static __read_mostly struct preempt_ops kvm_preempt_ops;
  85
  86 struct dentry *kvm_debugfs_dir;
  87
  88 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
  89                            unsigned long arg);
  90 #ifdef CONFIG_COMPAT
  91 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
  92                                   unsigned long arg);
  93 #endif
  94 static int hardware_enable_all(void);
  95 static void hardware_disable_all(void);
  96
  97 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
  98 static void update_memslots(struct kvm_memslots *slots,
  99                             struct kvm_memory_slot *new, u64 last_generation);
 100
 101 static void kvm_release_pfn_dirty(pfn_t pfn);
 102 static void mark_page_dirty_in_slot(struct kvm *kvm,
 103                                     struct kvm_memory_slot *memslot, gfn_t gfn);
 104
 105 __visible bool kvm_rebooting;
 106 EXPORT_SYMBOL_GPL(kvm_rebooting);
 107
 108 static bool largepages_enabled = true;
 109
 110 bool kvm_is_mmio_pfn(pfn_t pfn)
 111 {
 112         if (pfn_valid(pfn))
 113                 return PageReserved(pfn_to_page(pfn));
 114
 115         return true;
 116 }
 117
 118 /*
 119  * Switches to specified vcpu, until a matching vcpu_put()
 120  */
 121 int vcpu_load(struct kvm_vcpu *vcpu)
 122 {
 123         int cpu;
 124
 125         if (mutex_lock_killable(&vcpu->mutex))
 126                 return -EINTR;
 127         if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
 128                 /* The thread running this VCPU changed. */
 129                 struct pid *oldpid = vcpu->pid;
 130                 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
 131                 rcu_assign_pointer(vcpu->pid, newpid);
 132                 synchronize_rcu();
 133                 put_pid(oldpid);
 134         }
 135         cpu = get_cpu();
 136         preempt_notifier_register(&vcpu->preempt_notifier);
 137         kvm_arch_vcpu_load(vcpu, cpu);
 138         put_cpu();
 139         return 0;
 140 }
 141
 142 void vcpu_put(struct kvm_vcpu *vcpu)
 143 {
 144         preempt_disable();
 145         kvm_arch_vcpu_put(vcpu);
 146         preempt_notifier_unregister(&vcpu->preempt_notifier);
 147         preempt_enable();
 148         mutex_unlock(&vcpu->mutex);
 149 }
 150
 151 static void ack_flush(void *_completed)
 152 {
 153 }
 154
 155 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
 156 {
 157         int i, cpu, me;
 158         cpumask_var_t cpus;
 159         bool called = true;
 160         struct kvm_vcpu *vcpu;
 161
 162         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
 163
 164         me = get_cpu();
 165         kvm_for_each_vcpu(i, vcpu, kvm) {
 166                 kvm_make_request(req, vcpu);
 167                 cpu = vcpu->cpu;
 168
 169                 /* Set ->requests bit before we read ->mode */
 170                 smp_mb();
 171
 172                 if (cpus != NULL && cpu != -1 && cpu != me &&
 173                       kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
 174                         cpumask_set_cpu(cpu, cpus);
 175         }
 176         if (unlikely(cpus == NULL))
 177                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
 178         else if (!cpumask_empty(cpus))
 179                 smp_call_function_many(cpus, ack_flush, NULL, 1);
 180         else
 181                 called = false;
 182         put_cpu();
 183         free_cpumask_var(cpus);
 184         return called;
 185 }
 186
 187 void kvm_flush_remote_tlbs(struct kvm *kvm)
 188 {
 189         long dirty_count = kvm->tlbs_dirty;
 190
 191         smp_mb();
 192         if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
 193                 ++kvm->stat.remote_tlb_flush;
 194         cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
 195 }
 196 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
 197
 198 void kvm_reload_remote_mmus(struct kvm *kvm)
 199 {
 200         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
 201 }
 202
 203 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
 204 {
 205         make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
 206 }
 207
 208 void kvm_make_scan_ioapic_request(struct kvm *kvm)
 209 {
 210         make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
 211 }
 212
 213 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 214 {
 215         struct page *page;
 216         int r;
 217
 218         mutex_init(&vcpu->mutex);
 219         vcpu->cpu = -1;
 220         vcpu->kvm = kvm;
 221         vcpu->vcpu_id = id;
 222         vcpu->pid = NULL;
 223         init_waitqueue_head(&vcpu->wq);
 224         kvm_async_pf_vcpu_init(vcpu);
 225
 226         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 227         if (!page) {
 228                 r = -ENOMEM;
 229                 goto fail;
 230         }
 231         vcpu->run = page_address(page);
 232
 233         kvm_vcpu_set_in_spin_loop(vcpu, false);
 234         kvm_vcpu_set_dy_eligible(vcpu, false);
 235         vcpu->preempted = false;
 236
 237         r = kvm_arch_vcpu_init(vcpu);
 238         if (r < 0)
 239                 goto fail_free_run;
 240         return 0;
 241
 242 fail_free_run:
 243         free_page((unsigned long)vcpu->run);
 244 fail:
 245         return r;
 246 }
 247 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 248
 249 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 250 {
 251         put_pid(vcpu->pid);
 252         kvm_arch_vcpu_uninit(vcpu);
 253         free_page((unsigned long)vcpu->run);
 254 }
 255 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 256
 257 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 258 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
 259 {
 260         return container_of(mn, struct kvm, mmu_notifier);
 261 }
 262
 263 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
 264                                              struct mm_struct *mm,
 265                                              unsigned long address)
 266 {
 267         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 268         int need_tlb_flush, idx;
 269
 270         /*
 271          * When ->invalidate_page runs, the linux pte has been zapped
 272          * already but the page is still allocated until
 273          * ->invalidate_page returns. So if we increase the sequence
 274          * here the kvm page fault will notice if the spte can't be
 275          * established because the page is going to be freed. If
 276          * instead the kvm page fault establishes the spte before
 277          * ->invalidate_page runs, kvm_unmap_hva will release it
 278          * before returning.
 279          *
 280          * The sequence increase only need to be seen at spin_unlock
 281          * time, and not at spin_lock time.
 282          *
 283          * Increasing the sequence after the spin_unlock would be
 284          * unsafe because the kvm page fault could then establish the
 285          * pte after kvm_unmap_hva returned, without noticing the page
 286          * is going to be freed.
 287          */
 288         idx = srcu_read_lock(&kvm->srcu);
 289         spin_lock(&kvm->mmu_lock);
 290
 291         kvm->mmu_notifier_seq++;
 292         need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
 293         /* we've to flush the tlb before the pages can be freed */
 294         if (need_tlb_flush)
 295                 kvm_flush_remote_tlbs(kvm);
 296
 297         spin_unlock(&kvm->mmu_lock);
 298         srcu_read_unlock(&kvm->srcu, idx);
 299 }
 300
 301 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
 302                                         struct mm_struct *mm,
 303                                         unsigned long address,
 304                                         pte_t pte)
 305 {
 306         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 307         int idx;
 308
 309         idx = srcu_read_lock(&kvm->srcu);
 310         spin_lock(&kvm->mmu_lock);
 311         kvm->mmu_notifier_seq++;
 312         kvm_set_spte_hva(kvm, address, pte);
 313         spin_unlock(&kvm->mmu_lock);
 314         srcu_read_unlock(&kvm->srcu, idx);
 315 }
 316
 317 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 318                                                     struct mm_struct *mm,
 319                                                     unsigned long start,
 320                                                     unsigned long end)
 321 {
 322         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 323         int need_tlb_flush = 0, idx;
 324
 325         idx = srcu_read_lock(&kvm->srcu);
 326         spin_lock(&kvm->mmu_lock);
 327         /*
 328          * The count increase must become visible at unlock time as no
 329          * spte can be established without taking the mmu_lock and
 330          * count is also read inside the mmu_lock critical section.
 331          */
 332         kvm->mmu_notifier_count++;
 333         need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
 334         need_tlb_flush |= kvm->tlbs_dirty;
 335         /* we've to flush the tlb before the pages can be freed */
 336         if (need_tlb_flush)
 337                 kvm_flush_remote_tlbs(kvm);
 338
 339         spin_unlock(&kvm->mmu_lock);
 340         srcu_read_unlock(&kvm->srcu, idx);
 341 }
 342
 343 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
 344                                                   struct mm_struct *mm,
 345                                                   unsigned long start,
 346                                                   unsigned long end)
 347 {
 348         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 349
 350         spin_lock(&kvm->mmu_lock);
 351         /*
 352          * This sequence increase will notify the kvm page fault that
 353          * the page that is going to be mapped in the spte could have
 354          * been freed.
 355          */
 356         kvm->mmu_notifier_seq++;
 357         smp_wmb();
 358         /*
 359          * The above sequence increase must be visible before the
 360          * below count decrease, which is ensured by the smp_wmb above
 361          * in conjunction with the smp_rmb in mmu_notifier_retry().
 362          */
 363         kvm->mmu_notifier_count--;
 364         spin_unlock(&kvm->mmu_lock);
 365
 366         BUG_ON(kvm->mmu_notifier_count < 0);
 367 }
 368
 369 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
 370                                               struct mm_struct *mm,
 371                                               unsigned long address)
 372 {
 373         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 374         int young, idx;
 375
 376         idx = srcu_read_lock(&kvm->srcu);
 377         spin_lock(&kvm->mmu_lock);
 378
 379         young = kvm_age_hva(kvm, address);
 380         if (young)
 381                 kvm_flush_remote_tlbs(kvm);
 382
 383         spin_unlock(&kvm->mmu_lock);
 384         srcu_read_unlock(&kvm->srcu, idx);
 385
 386         return young;
 387 }
 388
 389 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
 390                                        struct mm_struct *mm,
 391                                        unsigned long address)
 392 {
 393         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 394         int young, idx;
 395
 396         idx = srcu_read_lock(&kvm->srcu);
 397         spin_lock(&kvm->mmu_lock);
 398         young = kvm_test_age_hva(kvm, address);
 399         spin_unlock(&kvm->mmu_lock);
 400         srcu_read_unlock(&kvm->srcu, idx);
 401
 402         return young;
 403 }
 404
 405 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
 406                                      struct mm_struct *mm)
 407 {
 408         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 409         int idx;
 410
 411         idx = srcu_read_lock(&kvm->srcu);
 412         kvm_arch_flush_shadow_all(kvm);
 413         srcu_read_unlock(&kvm->srcu, idx);
 414 }
 415
 416 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 417         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
 418         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
 419         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
 420         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
 421         .test_young             = kvm_mmu_notifier_test_young,
 422         .change_pte             = kvm_mmu_notifier_change_pte,
 423         .release                = kvm_mmu_notifier_release,
 424 };
 425
 426 static int kvm_init_mmu_notifier(struct kvm *kvm)
 427 {
 428         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
 429         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
 430 }
 431
 432 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
 433
 434 static int kvm_init_mmu_notifier(struct kvm *kvm)
 435 {
 436         return 0;
 437 }
 438
 439 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
 440
 441 static void kvm_init_memslots_id(struct kvm *kvm)
 442 {
 443         int i;
 444         struct kvm_memslots *slots = kvm->memslots;
 445
 446         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
 447                 slots->id_to_index[i] = slots->memslots[i].id = i;
 448 }
 449
 450 static struct kvm *kvm_create_vm(unsigned long type)
 451 {
 452         int r, i;
 453         struct kvm *kvm = kvm_arch_alloc_vm();
 454
 455         if (!kvm)
 456                 return ERR_PTR(-ENOMEM);
 457
 458         r = kvm_arch_init_vm(kvm, type);
 459         if (r)
 460                 goto out_err_no_disable;
 461
 462         r = hardware_enable_all();
 463         if (r)
 464                 goto out_err_no_disable;
 465
 466 #ifdef CONFIG_HAVE_KVM_IRQCHIP
 467         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
 468         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
 469 #endif
 470
 471         BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
 472
 473         r = -ENOMEM;
 474         kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 475         if (!kvm->memslots)
 476                 goto out_err_no_srcu;
 477         kvm_init_memslots_id(kvm);
 478         if (init_srcu_struct(&kvm->srcu))
 479                 goto out_err_no_srcu;
 480         if (init_srcu_struct(&kvm->irq_srcu))
 481                 goto out_err_no_irq_srcu;
 482         for (i = 0; i < KVM_NR_BUSES; i++) {
 483                 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
 484                                         GFP_KERNEL);
 485                 if (!kvm->buses[i])
 486                         goto out_err;
 487         }
 488
 489         spin_lock_init(&kvm->mmu_lock);
 490         kvm->mm = current->mm;
 491         atomic_inc(&kvm->mm->mm_count);
 492         kvm_eventfd_init(kvm);
 493         mutex_init(&kvm->lock);
 494         mutex_init(&kvm->irq_lock);
 495         mutex_init(&kvm->slots_lock);
 496         atomic_set(&kvm->users_count, 1);
 497         INIT_LIST_HEAD(&kvm->devices);
 498
 499         r = kvm_init_mmu_notifier(kvm);
 500         if (r)
 501                 goto out_err;
 502
 503         spin_lock(&kvm_lock);
 504         list_add(&kvm->vm_list, &vm_list);
 505         spin_unlock(&kvm_lock);
 506
 507         return kvm;
 508
 509 out_err:
 510         cleanup_srcu_struct(&kvm->irq_srcu);
 511 out_err_no_irq_srcu:
 512         cleanup_srcu_struct(&kvm->srcu);
 513 out_err_no_srcu:
 514         hardware_disable_all();
 515 out_err_no_disable:
 516         for (i = 0; i < KVM_NR_BUSES; i++)
 517                 kfree(kvm->buses[i]);
 518         kfree(kvm->memslots);
 519         kvm_arch_free_vm(kvm);
 520         return ERR_PTR(r);
 521 }
 522
 523 /*
 524  * Avoid using vmalloc for a small buffer.
 525  * Should not be used when the size is statically known.
 526  */
 527 void *kvm_kvzalloc(unsigned long size)
 528 {
 529         if (size > PAGE_SIZE)
 530                 return vzalloc(size);
 531         else
 532                 return kzalloc(size, GFP_KERNEL);
 533 }
 534
 535 void kvm_kvfree(const void *addr)
 536 {
 537         if (is_vmalloc_addr(addr))
 538                 vfree(addr);
 539         else
 540                 kfree(addr);
 541 }
 542
 543 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
 544 {
 545         if (!memslot->dirty_bitmap)
 546                 return;
 547
 548         kvm_kvfree(memslot->dirty_bitmap);
 549         memslot->dirty_bitmap = NULL;
 550 }
 551
 552 /*
 553  * Free any memory in @free but not in @dont.
 554  */
 555 static void kvm_free_physmem_slot(struct kvm *kvm, struct kvm_memory_slot *free,
 556                                   struct kvm_memory_slot *dont)
 557 {
 558         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 559                 kvm_destroy_dirty_bitmap(free);
 560
 561         kvm_arch_free_memslot(kvm, free, dont);
 562
 563         free->npages = 0;
 564 }
 565
 566 static void kvm_free_physmem(struct kvm *kvm)
 567 {
 568         struct kvm_memslots *slots = kvm->memslots;
 569         struct kvm_memory_slot *memslot;
 570
 571         kvm_for_each_memslot(memslot, slots)
 572                 kvm_free_physmem_slot(kvm, memslot, NULL);
 573
 574         kfree(kvm->memslots);
 575 }
 576
 577 static void kvm_destroy_devices(struct kvm *kvm)
 578 {
 579         struct list_head *node, *tmp;
 580
 581         list_for_each_safe(node, tmp, &kvm->devices) {
 582                 struct kvm_device *dev =
 583                         list_entry(node, struct kvm_device, vm_node);
 584
 585                 list_del(node);
 586                 dev->ops->destroy(dev);
 587         }
 588 }
 589
 590 static void kvm_destroy_vm(struct kvm *kvm)
 591 {
 592         int i;
 593         struct mm_struct *mm = kvm->mm;
 594
 595         kvm_arch_sync_events(kvm);
 596         spin_lock(&kvm_lock);
 597         list_del(&kvm->vm_list);
 598         spin_unlock(&kvm_lock);
 599         kvm_free_irq_routing(kvm);
 600         for (i = 0; i < KVM_NR_BUSES; i++)
 601                 kvm_io_bus_destroy(kvm->buses[i]);
 602         kvm_coalesced_mmio_free(kvm);
 603 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 604         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
 605 #else
 606         kvm_arch_flush_shadow_all(kvm);
 607 #endif
 608         kvm_arch_destroy_vm(kvm);
 609         kvm_destroy_devices(kvm);
 610         kvm_free_physmem(kvm);
 611         cleanup_srcu_struct(&kvm->irq_srcu);
 612         cleanup_srcu_struct(&kvm->srcu);
 613         kvm_arch_free_vm(kvm);
 614         hardware_disable_all();
 615         mmdrop(mm);
 616 }
 617
 618 void kvm_get_kvm(struct kvm *kvm)
 619 {
 620         atomic_inc(&kvm->users_count);
 621 }
 622 EXPORT_SYMBOL_GPL(kvm_get_kvm);
 623
 624 void kvm_put_kvm(struct kvm *kvm)
 625 {
 626         if (atomic_dec_and_test(&kvm->users_count))
 627                 kvm_destroy_vm(kvm);
 628 }
 629 EXPORT_SYMBOL_GPL(kvm_put_kvm);
 630
 631
 632 static int kvm_vm_release(struct inode *inode, struct file *filp)
 633 {
 634         struct kvm *kvm = filp->private_data;
 635
 636         kvm_irqfd_release(kvm);
 637
 638         kvm_put_kvm(kvm);
 639         return 0;
 640 }
 641
 642 /*
 643  * Allocation size is twice as large as the actual dirty bitmap size.
 644  * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
 645  */
 646 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
 647 {
 648         unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
 649
 650         memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
 651         if (!memslot->dirty_bitmap)
 652                 return -ENOMEM;
 653
 654         return 0;
 655 }
 656
 657 static int cmp_memslot(const void *slot1, const void *slot2)
 658 {
 659         struct kvm_memory_slot *s1, *s2;
 660
 661         s1 = (struct kvm_memory_slot *)slot1;
 662         s2 = (struct kvm_memory_slot *)slot2;
 663
 664         if (s1->npages < s2->npages)
 665                 return 1;
 666         if (s1->npages > s2->npages)
 667                 return -1;
 668
 669         return 0;
 670 }
 671
 672 /*
 673  * Sort the memslots base on its size, so the larger slots
 674  * will get better fit.
 675  */
 676 static void sort_memslots(struct kvm_memslots *slots)
 677 {
 678         int i;
 679
 680         sort(slots->memslots, KVM_MEM_SLOTS_NUM,
 681               sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
 682
 683         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
 684                 slots->id_to_index[slots->memslots[i].id] = i;
 685 }
 686
 687 static void update_memslots(struct kvm_memslots *slots,
 688                             struct kvm_memory_slot *new,
 689                             u64 last_generation)
 690 {
 691         if (new) {
 692                 int id = new->id;
 693                 struct kvm_memory_slot *old = id_to_memslot(slots, id);
 694                 unsigned long npages = old->npages;
 695
 696                 *old = *new;
 697                 if (new->npages != npages)
 698                         sort_memslots(slots);
 699         }
 700
 701         slots->generation = last_generation + 1;
 702 }
 703
 704 static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
 705 {
 706         u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
 707
 708 #ifdef KVM_CAP_READONLY_MEM
 709         valid_flags |= KVM_MEM_READONLY;
 710 #endif
 711
 712         if (mem->flags & ~valid_flags)
 713                 return -EINVAL;
 714
 715         return 0;
 716 }
 717
 718 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
 719                 struct kvm_memslots *slots, struct kvm_memory_slot *new)
 720 {
 721         struct kvm_memslots *old_memslots = kvm->memslots;
 722
 723         update_memslots(slots, new, kvm->memslots->generation);
 724         rcu_assign_pointer(kvm->memslots, slots);
 725         synchronize_srcu_expedited(&kvm->srcu);
 726
 727         kvm_arch_memslots_updated(kvm);
 728
 729         return old_memslots;
 730 }
 731
 732 /*
 733  * Allocate some memory and give it an address in the guest physical address
 734  * space.
 735  *
 736  * Discontiguous memory is allowed, mostly for framebuffers.
 737  *
 738  * Must be called holding mmap_sem for write.
 739  */
 740 int __kvm_set_memory_region(struct kvm *kvm,
 741                             struct kvm_userspace_memory_region *mem)
 742 {
 743         int r;
 744         gfn_t base_gfn;
 745         unsigned long npages;
 746         struct kvm_memory_slot *slot;
 747         struct kvm_memory_slot old, new;
 748         struct kvm_memslots *slots = NULL, *old_memslots;
 749         enum kvm_mr_change change;
 750
 751         r = check_memory_region_flags(mem);
 752         if (r)
 753                 goto out;
 754
 755         r = -EINVAL;
 756         /* General sanity checks */
 757         if (mem->memory_size & (PAGE_SIZE - 1))
 758                 goto out;
 759         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 760                 goto out;
 761         /* We can read the guest memory with __xxx_user() later on. */
 762         if ((mem->slot < KVM_USER_MEM_SLOTS) &&
 763             ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
 764              !access_ok(VERIFY_WRITE,
 765                         (void __user *)(unsigned long)mem->userspace_addr,
 766                         mem->memory_size)))
 767                 goto out;
 768         if (mem->slot >= KVM_MEM_SLOTS_NUM)
 769                 goto out;
 770         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 771                 goto out;
 772
 773         slot = id_to_memslot(kvm->memslots, mem->slot);
 774         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 775         npages = mem->memory_size >> PAGE_SHIFT;
 776
 777         r = -EINVAL;
 778         if (npages > KVM_MEM_MAX_NR_PAGES)
 779                 goto out;
 780
 781         if (!npages)
 782                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 783
 784         new = old = *slot;
 785
 786         new.id = mem->slot;
 787         new.base_gfn = base_gfn;
 788         new.npages = npages;
 789         new.flags = mem->flags;
 790
 791         r = -EINVAL;
 792         if (npages) {
 793                 if (!old.npages)
 794                         change = KVM_MR_CREATE;
 795                 else { /* Modify an existing slot. */
 796                         if ((mem->userspace_addr != old.userspace_addr) ||
 797                             (npages != old.npages) ||
 798                             ((new.flags ^ old.flags) & KVM_MEM_READONLY))
 799                                 goto out;
 800
 801                         if (base_gfn != old.base_gfn)
 802                                 change = KVM_MR_MOVE;
 803                         else if (new.flags != old.flags)
 804                                 change = KVM_MR_FLAGS_ONLY;
 805                         else { /* Nothing to change. */
 806                                 r = 0;
 807                                 goto out;
 808                         }
 809                 }
 810         } else if (old.npages) {
 811                 change = KVM_MR_DELETE;
 812         } else /* Modify a non-existent slot: disallowed. */
 813                 goto out;
 814
 815         if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
 816                 /* Check for overlaps */
 817                 r = -EEXIST;
 818                 kvm_for_each_memslot(slot, kvm->memslots) {
 819                         if ((slot->id >= KVM_USER_MEM_SLOTS) ||
 820                             (slot->id == mem->slot))
 821                                 continue;
 822                         if (!((base_gfn + npages <= slot->base_gfn) ||
 823                               (base_gfn >= slot->base_gfn + slot->npages)))
 824                                 goto out;
 825                 }
 826         }
 827
 828         /* Free page dirty bitmap if unneeded */
 829         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 830                 new.dirty_bitmap = NULL;
 831
 832         r = -ENOMEM;
 833         if (change == KVM_MR_CREATE) {
 834                 new.userspace_addr = mem->userspace_addr;
 835
 836                 if (kvm_arch_create_memslot(kvm, &new, npages))
 837                         goto out_free;
 838         }
 839
 840         /* Allocate page dirty bitmap if needed */
 841         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 842                 if (kvm_create_dirty_bitmap(&new) < 0)
 843                         goto out_free;
 844         }
 845
 846         if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
 847                 r = -ENOMEM;
 848                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 849                                 GFP_KERNEL);
 850                 if (!slots)
 851                         goto out_free;
 852                 slot = id_to_memslot(slots, mem->slot);
 853                 slot->flags |= KVM_MEMSLOT_INVALID;
 854
 855                 old_memslots = install_new_memslots(kvm, slots, NULL);
 856
 857                 /* slot was deleted or moved, clear iommu mapping */
 858                 kvm_iommu_unmap_pages(kvm, &old);
 859                 /* From this point no new shadow pages pointing to a deleted,
 860                  * or moved, memslot will be created.
 861                  *
 862                  * validation of sp->gfn happens in:
 863                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
 864                  *      - kvm_is_visible_gfn (mmu_check_roots)
 865                  */
 866                 kvm_arch_flush_shadow_memslot(kvm, slot);
 867                 slots = old_memslots;
 868         }
 869
 870         r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
 871         if (r)
 872                 goto out_slots;
 873
 874         r = -ENOMEM;
 875         /*
 876          * We can re-use the old_memslots from above, the only difference
 877          * from the currently installed memslots is the invalid flag.  This
 878          * will get overwritten by update_memslots anyway.
 879          */
 880         if (!slots) {
 881                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 882                                 GFP_KERNEL);
 883                 if (!slots)
 884                         goto out_free;
 885         }
 886
 887         /* actual memory is freed via old in kvm_free_physmem_slot below */
 888         if (change == KVM_MR_DELETE) {
 889                 new.dirty_bitmap = NULL;
 890                 memset(&new.arch, 0, sizeof(new.arch));
 891         }
 892
 893         old_memslots = install_new_memslots(kvm, slots, &new);
 894
 895         kvm_arch_commit_memory_region(kvm, mem, &old, change);
 896
 897         kvm_free_physmem_slot(kvm, &old, &new);
 898         kfree(old_memslots);
 899
 900         /*
 901          * IOMMU mapping:  New slots need to be mapped.  Old slots need to be
 902          * un-mapped and re-mapped if their base changes.  Since base change
 903          * unmapping is handled above with slot deletion, mapping alone is
 904          * needed here.  Anything else the iommu might care about for existing
 905          * slots (size changes, userspace addr changes and read-only flag
 906          * changes) is disallowed above, so any other attribute changes getting
 907          * here can be skipped.
 908          */
 909         if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
 910                 r = kvm_iommu_map_pages(kvm, &new);
 911                 return r;
 912         }
 913
 914         return 0;
 915
 916 out_slots:
 917         kfree(slots);
 918 out_free:
 919         kvm_free_physmem_slot(kvm, &new, &old);
 920 out:
 921         return r;
 922 }
 923 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 924
 925 int kvm_set_memory_region(struct kvm *kvm,
 926                           struct kvm_userspace_memory_region *mem)
 927 {
 928         int r;
 929
 930         mutex_lock(&kvm->slots_lock);
 931         r = __kvm_set_memory_region(kvm, mem);
 932         mutex_unlock(&kvm->slots_lock);
 933         return r;
 934 }
 935 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 936
 937 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 938                                           struct kvm_userspace_memory_region *mem)
 939 {
 940         if (mem->slot >= KVM_USER_MEM_SLOTS)
 941                 return -EINVAL;
 942         return kvm_set_memory_region(kvm, mem);
 943 }
 944
 945 int kvm_get_dirty_log(struct kvm *kvm,
 946                         struct kvm_dirty_log *log, int *is_dirty)
 947 {
 948         struct kvm_memory_slot *memslot;
 949         int r, i;
 950         unsigned long n;
 951         unsigned long any = 0;
 952
 953         r = -EINVAL;
 954         if (log->slot >= KVM_USER_MEM_SLOTS)
 955                 goto out;
 956
 957         memslot = id_to_memslot(kvm->memslots, log->slot);
 958         r = -ENOENT;
 959         if (!memslot->dirty_bitmap)
 960                 goto out;
 961
 962         n = kvm_dirty_bitmap_bytes(memslot);
 963
 964         for (i = 0; !any && i < n/sizeof(long); ++i)
 965                 any = memslot->dirty_bitmap[i];
 966
 967         r = -EFAULT;
 968         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 969                 goto out;
 970
 971         if (any)
 972                 *is_dirty = 1;
 973
 974         r = 0;
 975 out:
 976         return r;
 977 }
 978 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
 979
 980 bool kvm_largepages_enabled(void)
 981 {
 982         return largepages_enabled;
 983 }
 984
 985 void kvm_disable_largepages(void)
 986 {
 987         largepages_enabled = false;
 988 }
 989 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
 990
 991 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 992 {
 993         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
 994 }
 995 EXPORT_SYMBOL_GPL(gfn_to_memslot);
 996
 997 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
 998 {
 999         struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1000
1001         if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1002               memslot->flags & KVM_MEMSLOT_INVALID)
1003                 return 0;
1004
1005         return 1;
1006 }
1007 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1008
1009 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1010 {
1011         struct vm_area_struct *vma;
1012         unsigned long addr, size;
1013
1014         size = PAGE_SIZE;
1015
1016         addr = gfn_to_hva(kvm, gfn);
1017         if (kvm_is_error_hva(addr))
1018                 return PAGE_SIZE;
1019
1020         down_read(&current->mm->mmap_sem);
1021         vma = find_vma(current->mm, addr);
1022         if (!vma)
1023                 goto out;
1024
1025         size = vma_kernel_pagesize(vma);
1026
1027 out:
1028         up_read(&current->mm->mmap_sem);
1029
1030         return size;
1031 }
1032
1033 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1034 {
1035         return slot->flags & KVM_MEM_READONLY;
1036 }
1037
1038 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1039                                        gfn_t *nr_pages, bool write)
1040 {
1041         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1042                 return KVM_HVA_ERR_BAD;
1043
1044         if (memslot_is_readonly(slot) && write)
1045                 return KVM_HVA_ERR_RO_BAD;
1046
1047         if (nr_pages)
1048                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1049
1050         return __gfn_to_hva_memslot(slot, gfn);
1051 }
1052
1053 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1054                                      gfn_t *nr_pages)
1055 {
1056         return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1057 }
1058
1059 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1060                                         gfn_t gfn)
1061 {
1062         return gfn_to_hva_many(slot, gfn, NULL);
1063 }
1064 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1065
1066 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1067 {
1068         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1069 }
1070 EXPORT_SYMBOL_GPL(gfn_to_hva);
1071
1072 /*
1073  * If writable is set to false, the hva returned by this function is only
1074  * allowed to be read.
1075  */
1076 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1077 {
1078         struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1079         unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1080
1081         if (!kvm_is_error_hva(hva) && writable)
1082                 *writable = !memslot_is_readonly(slot);
1083
1084         return hva;
1085 }
1086
1087 static int kvm_read_hva(void *data, void __user *hva, int len)
1088 {
1089         return __copy_from_user(data, hva, len);
1090 }
1091
1092 static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1093 {
1094         return __copy_from_user_inatomic(data, hva, len);
1095 }
1096
1097 static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1098         unsigned long start, int write, struct page **page)
1099 {
1100         int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1101
1102         if (write)
1103                 flags |= FOLL_WRITE;
1104
1105         return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1106 }
1107
1108 static inline int check_user_page_hwpoison(unsigned long addr)
1109 {
1110         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1111
1112         rc = __get_user_pages(current, current->mm, addr, 1,
1113                               flags, NULL, NULL, NULL);
1114         return rc == -EHWPOISON;
1115 }
1116
1117 /*
1118  * The atomic path to get the writable pfn which will be stored in @pfn,
1119  * true indicates success, otherwise false is returned.
1120  */
1121 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1122                             bool write_fault, bool *writable, pfn_t *pfn)
1123 {
1124         struct page *page[1];
1125         int npages;
1126
1127         if (!(async || atomic))
1128                 return false;
1129
1130         /*
1131          * Fast pin a writable pfn only if it is a write fault request
1132          * or the caller allows to map a writable pfn for a read fault
1133          * request.
1134          */
1135         if (!(write_fault || writable))
1136                 return false;
1137
1138         npages = __get_user_pages_fast(addr, 1, 1, page);
1139         if (npages == 1) {
1140                 *pfn = page_to_pfn(page[0]);
1141
1142                 if (writable)
1143                         *writable = true;
1144                 return true;
1145         }
1146
1147         return false;
1148 }
1149
1150 /*
1151  * The slow path to get the pfn of the specified host virtual address,
1152  * 1 indicates success, -errno is returned if error is detected.
1153  */
1154 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1155                            bool *writable, pfn_t *pfn)
1156 {
1157         struct page *page[1];
1158         int npages = 0;
1159
1160         might_sleep();
1161
1162         if (writable)
1163                 *writable = write_fault;
1164
1165         if (async) {
1166                 down_read(&current->mm->mmap_sem);
1167                 npages = get_user_page_nowait(current, current->mm,
1168                                               addr, write_fault, page);
1169                 up_read(&current->mm->mmap_sem);
1170         } else
1171                 npages = get_user_pages_fast(addr, 1, write_fault,
1172                                              page);
1173         if (npages != 1)
1174                 return npages;
1175
1176         /* map read fault as writable if possible */
1177         if (unlikely(!write_fault) && writable) {
1178                 struct page *wpage[1];
1179
1180                 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1181                 if (npages == 1) {
1182                         *writable = true;
1183                         put_page(page[0]);
1184                         page[0] = wpage[0];
1185                 }
1186
1187                 npages = 1;
1188         }
1189         *pfn = page_to_pfn(page[0]);
1190         return npages;
1191 }
1192
1193 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1194 {
1195         if (unlikely(!(vma->vm_flags & VM_READ)))
1196                 return false;
1197
1198         if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1199                 return false;
1200
1201         return true;
1202 }
1203
1204 /*
1205  * Pin guest page in memory and return its pfn.
1206  * @addr: host virtual address which maps memory to the guest
1207  * @atomic: whether this function can sleep
1208  * @async: whether this function need to wait IO complete if the
1209  *         host page is not in the memory
1210  * @write_fault: whether we should get a writable host page
1211  * @writable: whether it allows to map a writable host page for !@write_fault
1212  *
1213  * The function will map a writable host page for these two cases:
1214  * 1): @write_fault = true
1215  * 2): @write_fault = false && @writable, @writable will tell the caller
1216  *     whether the mapping is writable.
1217  */
1218 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1219                         bool write_fault, bool *writable)
1220 {
1221         struct vm_area_struct *vma;
1222         pfn_t pfn = 0;
1223         int npages;
1224
1225         /* we can do it either atomically or asynchronously, not both */
1226         BUG_ON(atomic && async);
1227
1228         if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1229                 return pfn;
1230
1231         if (atomic)
1232                 return KVM_PFN_ERR_FAULT;
1233
1234         npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1235         if (npages == 1)
1236                 return pfn;
1237
1238         down_read(&current->mm->mmap_sem);
1239         if (npages == -EHWPOISON ||
1240               (!async && check_user_page_hwpoison(addr))) {
1241                 pfn = KVM_PFN_ERR_HWPOISON;
1242                 goto exit;
1243         }
1244
1245         vma = find_vma_intersection(current->mm, addr, addr + 1);
1246
1247         if (vma == NULL)
1248                 pfn = KVM_PFN_ERR_FAULT;
1249         else if ((vma->vm_flags & VM_PFNMAP)) {
1250                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1251                         vma->vm_pgoff;
1252                 BUG_ON(!kvm_is_mmio_pfn(pfn));
1253         } else {
1254                 if (async && vma_is_valid(vma, write_fault))
1255                         *async = true;
1256                 pfn = KVM_PFN_ERR_FAULT;
1257         }
1258 exit:
1259         up_read(&current->mm->mmap_sem);
1260         return pfn;
1261 }
1262
1263 static pfn_t
1264 __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1265                      bool *async, bool write_fault, bool *writable)
1266 {
1267         unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1268
1269         if (addr == KVM_HVA_ERR_RO_BAD)
1270                 return KVM_PFN_ERR_RO_FAULT;
1271
1272         if (kvm_is_error_hva(addr))
1273                 return KVM_PFN_NOSLOT;
1274
1275         /* Do not map writable pfn in the readonly memslot. */
1276         if (writable && memslot_is_readonly(slot)) {
1277                 *writable = false;
1278                 writable = NULL;
1279         }
1280
1281         return hva_to_pfn(addr, atomic, async, write_fault,
1282                           writable);
1283 }
1284
1285 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1286                           bool write_fault, bool *writable)
1287 {
1288         struct kvm_memory_slot *slot;
1289
1290         if (async)
1291                 *async = false;
1292
1293         slot = gfn_to_memslot(kvm, gfn);
1294
1295         return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1296                                     writable);
1297 }
1298
1299 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1300 {
1301         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1302 }
1303 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1304
1305 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1306                        bool write_fault, bool *writable)
1307 {
1308         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1309 }
1310 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1311
1312 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1313 {
1314         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1315 }
1316 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1317
1318 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1319                       bool *writable)
1320 {
1321         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1322 }
1323 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1324
1325 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1326 {
1327         return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1328 }
1329
1330 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1331 {
1332         return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1333 }
1334 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1335
1336 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1337                                                                   int nr_pages)
1338 {
1339         unsigned long addr;
1340         gfn_t entry;
1341
1342         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1343         if (kvm_is_error_hva(addr))
1344                 return -1;
1345
1346         if (entry < nr_pages)
1347                 return 0;
1348
1349         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1350 }
1351 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1352
1353 static struct page *kvm_pfn_to_page(pfn_t pfn)
1354 {
1355         if (is_error_noslot_pfn(pfn))
1356                 return KVM_ERR_PTR_BAD_PAGE;
1357
1358         if (kvm_is_mmio_pfn(pfn)) {
1359                 WARN_ON(1);
1360                 return KVM_ERR_PTR_BAD_PAGE;
1361         }
1362
1363         return pfn_to_page(pfn);
1364 }
1365
1366 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1367 {
1368         pfn_t pfn;
1369
1370         pfn = gfn_to_pfn(kvm, gfn);
1371
1372         return kvm_pfn_to_page(pfn);
1373 }
1374
1375 EXPORT_SYMBOL_GPL(gfn_to_page);
1376
1377 void kvm_release_page_clean(struct page *page)
1378 {
1379         WARN_ON(is_error_page(page));
1380
1381         kvm_release_pfn_clean(page_to_pfn(page));
1382 }
1383 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1384
1385 void kvm_release_pfn_clean(pfn_t pfn)
1386 {
1387         if (!is_error_noslot_pfn(pfn) && !kvm_is_mmio_pfn(pfn))
1388                 put_page(pfn_to_page(pfn));
1389 }
1390 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1391
1392 void kvm_release_page_dirty(struct page *page)
1393 {
1394         WARN_ON(is_error_page(page));
1395
1396         kvm_release_pfn_dirty(page_to_pfn(page));
1397 }
1398 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1399
1400 static void kvm_release_pfn_dirty(pfn_t pfn)
1401 {
1402         kvm_set_pfn_dirty(pfn);
1403         kvm_release_pfn_clean(pfn);
1404 }
1405
1406 void kvm_set_pfn_dirty(pfn_t pfn)
1407 {
1408         if (!kvm_is_mmio_pfn(pfn)) {
1409                 struct page *page = pfn_to_page(pfn);
1410                 if (!PageReserved(page))
1411                         SetPageDirty(page);
1412         }
1413 }
1414 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1415
1416 void kvm_set_pfn_accessed(pfn_t pfn)
1417 {
1418         if (!kvm_is_mmio_pfn(pfn))
1419                 mark_page_accessed(pfn_to_page(pfn));
1420 }
1421 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1422
1423 void kvm_get_pfn(pfn_t pfn)
1424 {
1425         if (!kvm_is_mmio_pfn(pfn))
1426                 get_page(pfn_to_page(pfn));
1427 }
1428 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1429
1430 static int next_segment(unsigned long len, int offset)
1431 {
1432         if (len > PAGE_SIZE - offset)
1433                 return PAGE_SIZE - offset;
1434         else
1435                 return len;
1436 }
1437
1438 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1439                         int len)
1440 {
1441         int r;
1442         unsigned long addr;
1443
1444         addr = gfn_to_hva_prot(kvm, gfn, NULL);
1445         if (kvm_is_error_hva(addr))
1446                 return -EFAULT;
1447         r = kvm_read_hva(data, (void __user *)addr + offset, len);
1448         if (r)
1449                 return -EFAULT;
1450         return 0;
1451 }
1452 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1453
1454 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1455 {
1456         gfn_t gfn = gpa >> PAGE_SHIFT;
1457         int seg;
1458         int offset = offset_in_page(gpa);
1459         int ret;
1460
1461         while ((seg = next_segment(len, offset)) != 0) {
1462                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1463                 if (ret < 0)
1464                         return ret;
1465                 offset = 0;
1466                 len -= seg;
1467                 data += seg;
1468                 ++gfn;
1469         }
1470         return 0;
1471 }
1472 EXPORT_SYMBOL_GPL(kvm_read_guest);
1473
1474 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1475                           unsigned long len)
1476 {
1477         int r;
1478         unsigned long addr;
1479         gfn_t gfn = gpa >> PAGE_SHIFT;
1480         int offset = offset_in_page(gpa);
1481
1482         addr = gfn_to_hva_prot(kvm, gfn, NULL);
1483         if (kvm_is_error_hva(addr))
1484                 return -EFAULT;
1485         pagefault_disable();
1486         r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
1487         pagefault_enable();
1488         if (r)
1489                 return -EFAULT;
1490         return 0;
1491 }
1492 EXPORT_SYMBOL(kvm_read_guest_atomic);
1493
1494 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1495                          int offset, int len)
1496 {
1497         int r;
1498         unsigned long addr;
1499
1500         addr = gfn_to_hva(kvm, gfn);
1501         if (kvm_is_error_hva(addr))
1502                 return -EFAULT;
1503         r = __copy_to_user((void __user *)addr + offset, data, len);
1504         if (r)
1505                 return -EFAULT;
1506         mark_page_dirty(kvm, gfn);
1507         return 0;
1508 }
1509 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1510
1511 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1512                     unsigned long len)
1513 {
1514         gfn_t gfn = gpa >> PAGE_SHIFT;
1515         int seg;
1516         int offset = offset_in_page(gpa);
1517         int ret;
1518
1519         while ((seg = next_segment(len, offset)) != 0) {
1520                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1521                 if (ret < 0)
1522                         return ret;
1523                 offset = 0;
1524                 len -= seg;
1525                 data += seg;
1526                 ++gfn;
1527         }
1528         return 0;
1529 }
1530
1531 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1532                               gpa_t gpa, unsigned long len)
1533 {
1534         struct kvm_memslots *slots = kvm_memslots(kvm);
1535         int offset = offset_in_page(gpa);
1536         gfn_t start_gfn = gpa >> PAGE_SHIFT;
1537         gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1538         gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1539         gfn_t nr_pages_avail;
1540
1541         ghc->gpa = gpa;
1542         ghc->generation = slots->generation;
1543         ghc->len = len;
1544         ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1545         ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, &nr_pages_avail);
1546         if (!kvm_is_error_hva(ghc->hva) && nr_pages_avail >= nr_pages_needed) {
1547                 ghc->hva += offset;
1548         } else {
1549                 /*
1550                  * If the requested region crosses two memslots, we still
1551                  * verify that the entire region is valid here.
1552                  */
1553                 while (start_gfn <= end_gfn) {
1554                         ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1555                         ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1556                                                    &nr_pages_avail);
1557                         if (kvm_is_error_hva(ghc->hva))
1558                                 return -EFAULT;
1559                         start_gfn += nr_pages_avail;
1560                 }
1561                 /* Use the slow path for cross page reads and writes. */
1562                 ghc->memslot = NULL;
1563         }
1564         return 0;
1565 }
1566 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1567
1568 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1569                            void *data, unsigned long len)
1570 {
1571         struct kvm_memslots *slots = kvm_memslots(kvm);
1572         int r;
1573
1574         BUG_ON(len > ghc->len);
1575
1576         if (slots->generation != ghc->generation)
1577                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1578
1579         if (unlikely(!ghc->memslot))
1580                 return kvm_write_guest(kvm, ghc->gpa, data, len);
1581
1582         if (kvm_is_error_hva(ghc->hva))
1583                 return -EFAULT;
1584
1585         r = __copy_to_user((void __user *)ghc->hva, data, len);
1586         if (r)
1587                 return -EFAULT;
1588         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1589
1590         return 0;
1591 }
1592 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1593
1594 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1595                            void *data, unsigned long len)
1596 {
1597         struct kvm_memslots *slots = kvm_memslots(kvm);
1598         int r;
1599
1600         BUG_ON(len > ghc->len);
1601
1602         if (slots->generation != ghc->generation)
1603                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1604
1605         if (unlikely(!ghc->memslot))
1606                 return kvm_read_guest(kvm, ghc->gpa, data, len);
1607
1608         if (kvm_is_error_hva(ghc->hva))
1609                 return -EFAULT;
1610
1611         r = __copy_from_user(data, (void __user *)ghc->hva, len);
1612         if (r)
1613                 return -EFAULT;
1614
1615         return 0;
1616 }
1617 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1618
1619 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1620 {
1621         const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
1622
1623         return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
1624 }
1625 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1626
1627 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1628 {
1629         gfn_t gfn = gpa >> PAGE_SHIFT;
1630         int seg;
1631         int offset = offset_in_page(gpa);
1632         int ret;
1633
1634         while ((seg = next_segment(len, offset)) != 0) {
1635                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1636                 if (ret < 0)
1637                         return ret;
1638                 offset = 0;
1639                 len -= seg;
1640                 ++gfn;
1641         }
1642         return 0;
1643 }
1644 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1645
1646 static void mark_page_dirty_in_slot(struct kvm *kvm,
1647                                     struct kvm_memory_slot *memslot,
1648                                     gfn_t gfn)
1649 {
1650         if (memslot && memslot->dirty_bitmap) {
1651                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1652
1653                 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1654         }
1655 }
1656
1657 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1658 {
1659         struct kvm_memory_slot *memslot;
1660
1661         memslot = gfn_to_memslot(kvm, gfn);
1662         mark_page_dirty_in_slot(kvm, memslot, gfn);
1663 }
1664 EXPORT_SYMBOL_GPL(mark_page_dirty);
1665
1666 /*
1667  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1668  */
1669 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1670 {
1671         DEFINE_WAIT(wait);
1672
1673         for (;;) {
1674                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1675
1676                 if (kvm_arch_vcpu_runnable(vcpu)) {
1677                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1678                         break;
1679                 }
1680                 if (kvm_cpu_has_pending_timer(vcpu))
1681                         break;
1682                 if (signal_pending(current))
1683                         break;
1684
1685                 schedule();
1686         }
1687
1688         finish_wait(&vcpu->wq, &wait);
1689 }
1690 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
1691
1692 #ifndef CONFIG_S390
1693 /*
1694  * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1695  */
1696 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1697 {
1698         int me;
1699         int cpu = vcpu->cpu;
1700         wait_queue_head_t *wqp;
1701
1702         wqp = kvm_arch_vcpu_wq(vcpu);
1703         if (waitqueue_active(wqp)) {
1704                 wake_up_interruptible(wqp);
1705                 ++vcpu->stat.halt_wakeup;
1706         }
1707
1708         me = get_cpu();
1709         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1710                 if (kvm_arch_vcpu_should_kick(vcpu))
1711                         smp_send_reschedule(cpu);
1712         put_cpu();
1713 }
1714 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
1715 #endif /* !CONFIG_S390 */
1716
1717 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
1718 {
1719         struct pid *pid;
1720         struct task_struct *task = NULL;
1721         int ret = 0;
1722
1723         rcu_read_lock();
1724         pid = rcu_dereference(target->pid);
1725         if (pid)
1726                 task = get_pid_task(target->pid, PIDTYPE_PID);
1727         rcu_read_unlock();
1728         if (!task)
1729                 return ret;
1730         if (task->flags & PF_VCPU) {
1731                 put_task_struct(task);
1732                 return ret;
1733         }
1734         ret = yield_to(task, 1);
1735         put_task_struct(task);
1736
1737         return ret;
1738 }
1739 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1740
1741 /*
1742  * Helper that checks whether a VCPU is eligible for directed yield.
1743  * Most eligible candidate to yield is decided by following heuristics:
1744  *
1745  *  (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1746  *  (preempted lock holder), indicated by @in_spin_loop.
1747  *  Set at the beiginning and cleared at the end of interception/PLE handler.
1748  *
1749  *  (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1750  *  chance last time (mostly it has become eligible now since we have probably
1751  *  yielded to lockholder in last iteration. This is done by toggling
1752  *  @dy_eligible each time a VCPU checked for eligibility.)
1753  *
1754  *  Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1755  *  to preempted lock-holder could result in wrong VCPU selection and CPU
1756  *  burning. Giving priority for a potential lock-holder increases lock
1757  *  progress.
1758  *
1759  *  Since algorithm is based on heuristics, accessing another VCPU data without
1760  *  locking does not harm. It may result in trying to yield to  same VCPU, fail
1761  *  and continue with next VCPU and so on.
1762  */
1763 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1764 {
1765 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1766         bool eligible;
1767
1768         eligible = !vcpu->spin_loop.in_spin_loop ||
1769                         (vcpu->spin_loop.in_spin_loop &&
1770                          vcpu->spin_loop.dy_eligible);
1771
1772         if (vcpu->spin_loop.in_spin_loop)
1773                 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1774
1775         return eligible;
1776 #else
1777         return true;
1778 #endif
1779 }
1780
1781 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1782 {
1783         struct kvm *kvm = me->kvm;
1784         struct kvm_vcpu *vcpu;
1785         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1786         int yielded = 0;
1787         int try = 3;
1788         int pass;
1789         int i;
1790
1791         kvm_vcpu_set_in_spin_loop(me, true);
1792         /*
1793          * We boost the priority of a VCPU that is runnable but not
1794          * currently running, because it got preempted by something
1795          * else and called schedule in __vcpu_run.  Hopefully that
1796          * VCPU is holding the lock that we need and will release it.
1797          * We approximate round-robin by starting at the last boosted VCPU.
1798          */
1799         for (pass = 0; pass < 2 && !yielded && try; pass++) {
1800                 kvm_for_each_vcpu(i, vcpu, kvm) {
1801                         if (!pass && i <= last_boosted_vcpu) {
1802                                 i = last_boosted_vcpu;
1803                                 continue;
1804                         } else if (pass && i > last_boosted_vcpu)
1805                                 break;
1806                         if (!ACCESS_ONCE(vcpu->preempted))
1807                                 continue;
1808                         if (vcpu == me)
1809                                 continue;
1810                         if (waitqueue_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
1811                                 continue;
1812                         if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1813                                 continue;
1814
1815                         yielded = kvm_vcpu_yield_to(vcpu);
1816                         if (yielded > 0) {
1817                                 kvm->last_boosted_vcpu = i;
1818                                 break;
1819                         } else if (yielded < 0) {
1820                                 try--;
1821                                 if (!try)
1822                                         break;
1823                         }
1824                 }
1825         }
1826         kvm_vcpu_set_in_spin_loop(me, false);
1827
1828         /* Ensure vcpu is not eligible during next spinloop */
1829         kvm_vcpu_set_dy_eligible(me, false);
1830 }
1831 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1832
1833 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1834 {
1835         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1836         struct page *page;
1837
1838         if (vmf->pgoff == 0)
1839                 page = virt_to_page(vcpu->run);
1840 #ifdef CONFIG_X86
1841         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1842                 page = virt_to_page(vcpu->arch.pio_data);
1843 #endif
1844 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1845         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1846                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1847 #endif
1848         else
1849                 return kvm_arch_vcpu_fault(vcpu, vmf);
1850         get_page(page);
1851         vmf->page = page;
1852         return 0;
1853 }
1854
1855 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1856         .fault = kvm_vcpu_fault,
1857 };
1858
1859 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1860 {
1861         vma->vm_ops = &kvm_vcpu_vm_ops;
1862         return 0;
1863 }
1864
1865 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1866 {
1867         struct kvm_vcpu *vcpu = filp->private_data;
1868
1869         kvm_put_kvm(vcpu->kvm);
1870         return 0;
1871 }
1872
1873 static struct file_operations kvm_vcpu_fops = {
1874         .release        = kvm_vcpu_release,
1875         .unlocked_ioctl = kvm_vcpu_ioctl,
1876 #ifdef CONFIG_COMPAT
1877         .compat_ioctl   = kvm_vcpu_compat_ioctl,
1878 #endif
1879         .mmap           = kvm_vcpu_mmap,
1880         .llseek         = noop_llseek,
1881 };
1882
1883 /*
1884  * Allocates an inode for the vcpu.
1885  */
1886 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1887 {
1888         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
1889 }
1890
1891 /*
1892  * Creates some virtual cpus.  Good luck creating more than one.
1893  */
1894 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1895 {
1896         int r;
1897         struct kvm_vcpu *vcpu, *v;
1898
1899         if (id >= KVM_MAX_VCPUS)
1900                 return -EINVAL;
1901
1902         vcpu = kvm_arch_vcpu_create(kvm, id);
1903         if (IS_ERR(vcpu))
1904                 return PTR_ERR(vcpu);
1905
1906         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1907
1908         r = kvm_arch_vcpu_setup(vcpu);
1909         if (r)
1910                 goto vcpu_destroy;
1911
1912         mutex_lock(&kvm->lock);
1913         if (!kvm_vcpu_compatible(vcpu)) {
1914                 r = -EINVAL;
1915                 goto unlock_vcpu_destroy;
1916         }
1917         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1918                 r = -EINVAL;
1919                 goto unlock_vcpu_destroy;
1920         }
1921
1922         kvm_for_each_vcpu(r, v, kvm)
1923                 if (v->vcpu_id == id) {
1924                         r = -EEXIST;
1925                         goto unlock_vcpu_destroy;
1926                 }
1927
1928         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1929
1930         /* Now it's all set up, let userspace reach it */
1931         kvm_get_kvm(kvm);
1932         r = create_vcpu_fd(vcpu);
1933         if (r < 0) {
1934                 kvm_put_kvm(kvm);
1935                 goto unlock_vcpu_destroy;
1936         }
1937
1938         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1939         smp_wmb();
1940         atomic_inc(&kvm->online_vcpus);
1941
1942         mutex_unlock(&kvm->lock);
1943         kvm_arch_vcpu_postcreate(vcpu);
1944         return r;
1945
1946 unlock_vcpu_destroy:
1947         mutex_unlock(&kvm->lock);
1948 vcpu_destroy:
1949         kvm_arch_vcpu_destroy(vcpu);
1950         return r;
1951 }
1952
1953 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1954 {
1955         if (sigset) {
1956                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1957                 vcpu->sigset_active = 1;
1958                 vcpu->sigset = *sigset;
1959         } else
1960                 vcpu->sigset_active = 0;
1961         return 0;
1962 }
1963
1964 static long kvm_vcpu_ioctl(struct file *filp,
1965                            unsigned int ioctl, unsigned long arg)
1966 {
1967         struct kvm_vcpu *vcpu = filp->private_data;
1968         void __user *argp = (void __user *)arg;
1969         int r;
1970         struct kvm_fpu *fpu = NULL;
1971         struct kvm_sregs *kvm_sregs = NULL;
1972
1973         if (vcpu->kvm->mm != current->mm)
1974                 return -EIO;
1975
1976 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
1977         /*
1978          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1979          * so vcpu_load() would break it.
1980          */
1981         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1982                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1983 #endif
1984
1985
1986         r = vcpu_load(vcpu);
1987         if (r)
1988                 return r;
1989         switch (ioctl) {
1990         case KVM_RUN:
1991                 r = -EINVAL;
1992                 if (arg)
1993                         goto out;
1994                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1995                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1996                 break;
1997         case KVM_GET_REGS: {
1998                 struct kvm_regs *kvm_regs;
1999
2000                 r = -ENOMEM;
2001                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2002                 if (!kvm_regs)
2003                         goto out;
2004                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2005                 if (r)
2006                         goto out_free1;
2007                 r = -EFAULT;
2008                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2009                         goto out_free1;
2010                 r = 0;
2011 out_free1:
2012                 kfree(kvm_regs);
2013                 break;
2014         }
2015         case KVM_SET_REGS: {
2016                 struct kvm_regs *kvm_regs;
2017
2018                 r = -ENOMEM;
2019                 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2020                 if (IS_ERR(kvm_regs)) {
2021                         r = PTR_ERR(kvm_regs);
2022                         goto out;
2023                 }
2024                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2025                 kfree(kvm_regs);
2026                 break;
2027         }
2028         case KVM_GET_SREGS: {
2029                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2030                 r = -ENOMEM;
2031                 if (!kvm_sregs)
2032                         goto out;
2033                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2034                 if (r)
2035                         goto out;
2036                 r = -EFAULT;
2037                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2038                         goto out;
2039                 r = 0;
2040                 break;
2041         }
2042         case KVM_SET_SREGS: {
2043                 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2044                 if (IS_ERR(kvm_sregs)) {
2045                         r = PTR_ERR(kvm_sregs);
2046                         kvm_sregs = NULL;
2047                         goto out;
2048                 }
2049                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2050                 break;
2051         }
2052         case KVM_GET_MP_STATE: {
2053                 struct kvm_mp_state mp_state;
2054
2055                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2056                 if (r)
2057                         goto out;
2058                 r = -EFAULT;
2059                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
2060                         goto out;
2061                 r = 0;
2062                 break;
2063         }
2064         case KVM_SET_MP_STATE: {
2065                 struct kvm_mp_state mp_state;
2066
2067                 r = -EFAULT;
2068                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
2069                         goto out;
2070                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2071                 break;
2072         }
2073         case KVM_TRANSLATE: {
2074                 struct kvm_translation tr;
2075
2076                 r = -EFAULT;
2077                 if (copy_from_user(&tr, argp, sizeof tr))
2078                         goto out;
2079                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2080                 if (r)
2081                         goto out;
2082                 r = -EFAULT;
2083                 if (copy_to_user(argp, &tr, sizeof tr))
2084                         goto out;
2085                 r = 0;
2086                 break;
2087         }
2088         case KVM_SET_GUEST_DEBUG: {
2089                 struct kvm_guest_debug dbg;
2090
2091                 r = -EFAULT;
2092                 if (copy_from_user(&dbg, argp, sizeof dbg))
2093                         goto out;
2094                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2095                 break;
2096         }
2097         case KVM_SET_SIGNAL_MASK: {
2098                 struct kvm_signal_mask __user *sigmask_arg = argp;
2099                 struct kvm_signal_mask kvm_sigmask;
2100                 sigset_t sigset, *p;
2101
2102                 p = NULL;
2103                 if (argp) {
2104                         r = -EFAULT;
2105                         if (copy_from_user(&kvm_sigmask, argp,
2106                                            sizeof kvm_sigmask))
2107                                 goto out;
2108                         r = -EINVAL;
2109                         if (kvm_sigmask.len != sizeof sigset)
2110                                 goto out;
2111                         r = -EFAULT;
2112                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2113                                            sizeof sigset))
2114                                 goto out;
2115                         p = &sigset;
2116                 }
2117                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2118                 break;
2119         }
2120         case KVM_GET_FPU: {
2121                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2122                 r = -ENOMEM;
2123                 if (!fpu)
2124                         goto out;
2125                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2126                 if (r)
2127                         goto out;
2128                 r = -EFAULT;
2129                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2130                         goto out;
2131                 r = 0;
2132                 break;
2133         }
2134         case KVM_SET_FPU: {
2135                 fpu = memdup_user(argp, sizeof(*fpu));
2136                 if (IS_ERR(fpu)) {
2137                         r = PTR_ERR(fpu);
2138                         fpu = NULL;
2139                         goto out;
2140                 }
2141                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2142                 break;
2143         }
2144         default:
2145                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2146         }
2147 out:
2148         vcpu_put(vcpu);
2149         kfree(fpu);
2150         kfree(kvm_sregs);
2151         return r;
2152 }
2153
2154 #ifdef CONFIG_COMPAT
2155 static long kvm_vcpu_compat_ioctl(struct file *filp,
2156                                   unsigned int ioctl, unsigned long arg)
2157 {
2158         struct kvm_vcpu *vcpu = filp->private_data;
2159         void __user *argp = compat_ptr(arg);
2160         int r;
2161
2162         if (vcpu->kvm->mm != current->mm)
2163                 return -EIO;
2164
2165         switch (ioctl) {
2166         case KVM_SET_SIGNAL_MASK: {
2167                 struct kvm_signal_mask __user *sigmask_arg = argp;
2168                 struct kvm_signal_mask kvm_sigmask;
2169                 compat_sigset_t csigset;
2170                 sigset_t sigset;
2171
2172                 if (argp) {
2173                         r = -EFAULT;
2174                         if (copy_from_user(&kvm_sigmask, argp,
2175                                            sizeof kvm_sigmask))
2176                                 goto out;
2177                         r = -EINVAL;
2178                         if (kvm_sigmask.len != sizeof csigset)
2179                                 goto out;
2180                         r = -EFAULT;
2181                         if (copy_from_user(&csigset, sigmask_arg->sigset,
2182                                            sizeof csigset))
2183                                 goto out;
2184                         sigset_from_compat(&sigset, &csigset);
2185                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2186                 } else
2187                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2188                 break;
2189         }
2190         default:
2191                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2192         }
2193
2194 out:
2195         return r;
2196 }
2197 #endif
2198
2199 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2200                                  int (*accessor)(struct kvm_device *dev,
2201                                                  struct kvm_device_attr *attr),
2202                                  unsigned long arg)
2203 {
2204         struct kvm_device_attr attr;
2205
2206         if (!accessor)
2207                 return -EPERM;
2208
2209         if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2210                 return -EFAULT;
2211
2212         return accessor(dev, &attr);
2213 }
2214
2215 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2216                              unsigned long arg)
2217 {
2218         struct kvm_device *dev = filp->private_data;
2219
2220         switch (ioctl) {
2221         case KVM_SET_DEVICE_ATTR:
2222                 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2223         case KVM_GET_DEVICE_ATTR:
2224                 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2225         case KVM_HAS_DEVICE_ATTR:
2226                 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2227         default:
2228                 if (dev->ops->ioctl)
2229                         return dev->ops->ioctl(dev, ioctl, arg);
2230
2231                 return -ENOTTY;
2232         }
2233 }
2234
2235 static int kvm_device_release(struct inode *inode, struct file *filp)
2236 {
2237         struct kvm_device *dev = filp->private_data;
2238         struct kvm *kvm = dev->kvm;
2239
2240         kvm_put_kvm(kvm);
2241         return 0;
2242 }
2243
2244 static const struct file_operations kvm_device_fops = {
2245         .unlocked_ioctl = kvm_device_ioctl,
2246 #ifdef CONFIG_COMPAT
2247         .compat_ioctl = kvm_device_ioctl,
2248 #endif
2249         .release = kvm_device_release,
2250 };
2251
2252 struct kvm_device *kvm_device_from_filp(struct file *filp)
2253 {
2254         if (filp->f_op != &kvm_device_fops)
2255                 return NULL;
2256
2257         return filp->private_data;
2258 }
2259
2260 static int kvm_ioctl_create_device(struct kvm *kvm,
2261                                    struct kvm_create_device *cd)
2262 {
2263         struct kvm_device_ops *ops = NULL;
2264         struct kvm_device *dev;
2265         bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2266         int ret;
2267
2268         switch (cd->type) {
2269 #ifdef CONFIG_KVM_MPIC
2270         case KVM_DEV_TYPE_FSL_MPIC_20:
2271         case KVM_DEV_TYPE_FSL_MPIC_42:
2272                 ops = &kvm_mpic_ops;
2273                 break;
2274 #endif
2275 #ifdef CONFIG_KVM_XICS
2276         case KVM_DEV_TYPE_XICS:
2277                 ops = &kvm_xics_ops;
2278                 break;
2279 #endif
2280 #ifdef CONFIG_KVM_VFIO
2281         case KVM_DEV_TYPE_VFIO:
2282                 ops = &kvm_vfio_ops;
2283                 break;
2284 #endif
2285 #ifdef CONFIG_KVM_ARM_VGIC
2286         case KVM_DEV_TYPE_ARM_VGIC_V2:
2287                 ops = &kvm_arm_vgic_v2_ops;
2288                 break;
2289 #endif
2290 #ifdef CONFIG_S390
2291         case KVM_DEV_TYPE_FLIC:
2292                 ops = &kvm_flic_ops;
2293                 break;
2294 #endif
2295         default:
2296                 return -ENODEV;
2297         }
2298
2299         if (test)
2300                 return 0;
2301
2302         dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2303         if (!dev)
2304                 return -ENOMEM;
2305
2306         dev->ops = ops;
2307         dev->kvm = kvm;
2308
2309         ret = ops->create(dev, cd->type);
2310         if (ret < 0) {
2311                 kfree(dev);
2312                 return ret;
2313         }
2314
2315         ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2316         if (ret < 0) {
2317                 ops->destroy(dev);
2318                 return ret;
2319         }
2320
2321         list_add(&dev->vm_node, &kvm->devices);
2322         kvm_get_kvm(kvm);
2323         cd->fd = ret;
2324         return 0;
2325 }
2326
2327 static long kvm_vm_ioctl(struct file *filp,
2328                            unsigned int ioctl, unsigned long arg)
2329 {
2330         struct kvm *kvm = filp->private_data;
2331         void __user *argp = (void __user *)arg;
2332         int r;
2333
2334         if (kvm->mm != current->mm)
2335                 return -EIO;
2336         switch (ioctl) {
2337         case KVM_CREATE_VCPU:
2338                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2339                 break;
2340         case KVM_SET_USER_MEMORY_REGION: {
2341                 struct kvm_userspace_memory_region kvm_userspace_mem;
2342
2343                 r = -EFAULT;
2344                 if (copy_from_user(&kvm_userspace_mem, argp,
2345                                                 sizeof kvm_userspace_mem))
2346                         goto out;
2347
2348                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2349                 break;
2350         }
2351         case KVM_GET_DIRTY_LOG: {
2352                 struct kvm_dirty_log log;
2353
2354                 r = -EFAULT;
2355                 if (copy_from_user(&log, argp, sizeof log))
2356                         goto out;
2357                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2358                 break;
2359         }
2360 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2361         case KVM_REGISTER_COALESCED_MMIO: {
2362                 struct kvm_coalesced_mmio_zone zone;
2363                 r = -EFAULT;
2364                 if (copy_from_user(&zone, argp, sizeof zone))
2365                         goto out;
2366                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2367                 break;
2368         }
2369         case KVM_UNREGISTER_COALESCED_MMIO: {
2370                 struct kvm_coalesced_mmio_zone zone;
2371                 r = -EFAULT;
2372                 if (copy_from_user(&zone, argp, sizeof zone))
2373                         goto out;
2374                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2375                 break;
2376         }
2377 #endif
2378         case KVM_IRQFD: {
2379                 struct kvm_irqfd data;
2380
2381                 r = -EFAULT;
2382                 if (copy_from_user(&data, argp, sizeof data))
2383                         goto out;
2384                 r = kvm_irqfd(kvm, &data);
2385                 break;
2386         }
2387         case KVM_IOEVENTFD: {
2388                 struct kvm_ioeventfd data;
2389
2390                 r = -EFAULT;
2391                 if (copy_from_user(&data, argp, sizeof data))
2392                         goto out;
2393                 r = kvm_ioeventfd(kvm, &data);
2394                 break;
2395         }
2396 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2397         case KVM_SET_BOOT_CPU_ID:
2398                 r = 0;
2399                 mutex_lock(&kvm->lock);
2400                 if (atomic_read(&kvm->online_vcpus) != 0)
2401                         r = -EBUSY;
2402                 else
2403                         kvm->bsp_vcpu_id = arg;
2404                 mutex_unlock(&kvm->lock);
2405                 break;
2406 #endif
2407 #ifdef CONFIG_HAVE_KVM_MSI
2408         case KVM_SIGNAL_MSI: {
2409                 struct kvm_msi msi;
2410
2411                 r = -EFAULT;
2412                 if (copy_from_user(&msi, argp, sizeof msi))
2413                         goto out;
2414                 r = kvm_send_userspace_msi(kvm, &msi);
2415                 break;
2416         }
2417 #endif
2418 #ifdef __KVM_HAVE_IRQ_LINE
2419         case KVM_IRQ_LINE_STATUS:
2420         case KVM_IRQ_LINE: {
2421                 struct kvm_irq_level irq_event;
2422
2423                 r = -EFAULT;
2424                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2425                         goto out;
2426
2427                 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
2428                                         ioctl == KVM_IRQ_LINE_STATUS);
2429                 if (r)
2430                         goto out;
2431
2432                 r = -EFAULT;
2433                 if (ioctl == KVM_IRQ_LINE_STATUS) {
2434                         if (copy_to_user(argp, &irq_event, sizeof irq_event))
2435                                 goto out;
2436                 }
2437
2438                 r = 0;
2439                 break;
2440         }
2441 #endif
2442 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2443         case KVM_SET_GSI_ROUTING: {
2444                 struct kvm_irq_routing routing;
2445                 struct kvm_irq_routing __user *urouting;
2446                 struct kvm_irq_routing_entry *entries;
2447
2448                 r = -EFAULT;
2449                 if (copy_from_user(&routing, argp, sizeof(routing)))
2450                         goto out;
2451                 r = -EINVAL;
2452                 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2453                         goto out;
2454                 if (routing.flags)
2455                         goto out;
2456                 r = -ENOMEM;
2457                 entries = vmalloc(routing.nr * sizeof(*entries));
2458                 if (!entries)
2459                         goto out;
2460                 r = -EFAULT;
2461                 urouting = argp;
2462                 if (copy_from_user(entries, urouting->entries,
2463                                    routing.nr * sizeof(*entries)))
2464                         goto out_free_irq_routing;
2465                 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2466                                         routing.flags);
2467         out_free_irq_routing:
2468                 vfree(entries);
2469                 break;
2470         }
2471 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
2472         case KVM_CREATE_DEVICE: {
2473                 struct kvm_create_device cd;
2474
2475                 r = -EFAULT;
2476                 if (copy_from_user(&cd, argp, sizeof(cd)))
2477                         goto out;
2478
2479                 r = kvm_ioctl_create_device(kvm, &cd);
2480                 if (r)
2481                         goto out;
2482
2483                 r = -EFAULT;
2484                 if (copy_to_user(argp, &cd, sizeof(cd)))
2485                         goto out;
2486
2487                 r = 0;
2488                 break;
2489         }
2490         default:
2491                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2492                 if (r == -ENOTTY)
2493                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2494         }
2495 out:
2496         return r;
2497 }
2498
2499 #ifdef CONFIG_COMPAT
2500 struct compat_kvm_dirty_log {
2501         __u32 slot;
2502         __u32 padding1;
2503         union {
2504                 compat_uptr_t dirty_bitmap; /* one bit per page */
2505                 __u64 padding2;
2506         };
2507 };
2508
2509 static long kvm_vm_compat_ioctl(struct file *filp,
2510                            unsigned int ioctl, unsigned long arg)
2511 {
2512         struct kvm *kvm = filp->private_data;
2513         int r;
2514
2515         if (kvm->mm != current->mm)
2516                 return -EIO;
2517         switch (ioctl) {
2518         case KVM_GET_DIRTY_LOG: {
2519                 struct compat_kvm_dirty_log compat_log;
2520                 struct kvm_dirty_log log;
2521
2522                 r = -EFAULT;
2523                 if (copy_from_user(&compat_log, (void __user *)arg,
2524                                    sizeof(compat_log)))
2525                         goto out;
2526                 log.slot         = compat_log.slot;
2527                 log.padding1     = compat_log.padding1;
2528                 log.padding2     = compat_log.padding2;
2529                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2530
2531                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2532                 break;
2533         }
2534         default:
2535                 r = kvm_vm_ioctl(filp, ioctl, arg);
2536         }
2537
2538 out:
2539         return r;
2540 }
2541 #endif
2542
2543 static struct file_operations kvm_vm_fops = {
2544         .release        = kvm_vm_release,
2545         .unlocked_ioctl = kvm_vm_ioctl,
2546 #ifdef CONFIG_COMPAT
2547         .compat_ioctl   = kvm_vm_compat_ioctl,
2548 #endif
2549         .llseek         = noop_llseek,
2550 };
2551
2552 static int kvm_dev_ioctl_create_vm(unsigned long type)
2553 {
2554         int r;
2555         struct kvm *kvm;
2556
2557         kvm = kvm_create_vm(type);
2558         if (IS_ERR(kvm))
2559                 return PTR_ERR(kvm);
2560 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2561         r = kvm_coalesced_mmio_init(kvm);
2562         if (r < 0) {
2563                 kvm_put_kvm(kvm);
2564                 return r;
2565         }
2566 #endif
2567         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC);
2568         if (r < 0)
2569                 kvm_put_kvm(kvm);
2570
2571         return r;
2572 }
2573
2574 static long kvm_dev_ioctl_check_extension_generic(long arg)
2575 {
2576         switch (arg) {
2577         case KVM_CAP_USER_MEMORY:
2578         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2579         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2580 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2581         case KVM_CAP_SET_BOOT_CPU_ID:
2582 #endif
2583         case KVM_CAP_INTERNAL_ERROR_DATA:
2584 #ifdef CONFIG_HAVE_KVM_MSI
2585         case KVM_CAP_SIGNAL_MSI:
2586 #endif
2587 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2588         case KVM_CAP_IRQFD_RESAMPLE:
2589 #endif
2590                 return 1;
2591 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2592         case KVM_CAP_IRQ_ROUTING:
2593                 return KVM_MAX_IRQ_ROUTES;
2594 #endif
2595         default:
2596                 break;
2597         }
2598         return kvm_dev_ioctl_check_extension(arg);
2599 }
2600
2601 static long kvm_dev_ioctl(struct file *filp,
2602                           unsigned int ioctl, unsigned long arg)
2603 {
2604         long r = -EINVAL;
2605
2606         switch (ioctl) {
2607         case KVM_GET_API_VERSION:
2608                 r = -EINVAL;
2609                 if (arg)
2610                         goto out;
2611                 r = KVM_API_VERSION;
2612                 break;
2613         case KVM_CREATE_VM:
2614                 r = kvm_dev_ioctl_create_vm(arg);
2615                 break;
2616         case KVM_CHECK_EXTENSION:
2617                 r = kvm_dev_ioctl_check_extension_generic(arg);
2618                 break;
2619         case KVM_GET_VCPU_MMAP_SIZE:
2620                 r = -EINVAL;
2621                 if (arg)
2622                         goto out;
2623                 r = PAGE_SIZE;     /* struct kvm_run */
2624 #ifdef CONFIG_X86
2625                 r += PAGE_SIZE;    /* pio data page */
2626 #endif
2627 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2628                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2629 #endif
2630                 break;
2631         case KVM_TRACE_ENABLE:
2632         case KVM_TRACE_PAUSE:
2633         case KVM_TRACE_DISABLE:
2634                 r = -EOPNOTSUPP;
2635                 break;
2636         default:
2637                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2638         }
2639 out:
2640         return r;
2641 }
2642
2643 static struct file_operations kvm_chardev_ops = {
2644         .unlocked_ioctl = kvm_dev_ioctl,
2645         .compat_ioctl   = kvm_dev_ioctl,
2646         .llseek         = noop_llseek,
2647 };
2648
2649 static struct miscdevice kvm_dev = {
2650         KVM_MINOR,
2651         "kvm",
2652         &kvm_chardev_ops,
2653 };
2654
2655 static void hardware_enable_nolock(void *junk)
2656 {
2657         int cpu = raw_smp_processor_id();
2658         int r;
2659
2660         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2661                 return;
2662
2663         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2664
2665         r = kvm_arch_hardware_enable(NULL);
2666
2667         if (r) {
2668                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2669                 atomic_inc(&hardware_enable_failed);
2670                 printk(KERN_INFO "kvm: enabling virtualization on "
2671                                  "CPU%d failed\n", cpu);
2672         }
2673 }
2674
2675 static void hardware_enable(void)
2676 {
2677         raw_spin_lock(&kvm_count_lock);
2678         if (kvm_usage_count)
2679                 hardware_enable_nolock(NULL);
2680         raw_spin_unlock(&kvm_count_lock);
2681 }
2682
2683 static void hardware_disable_nolock(void *junk)
2684 {
2685         int cpu = raw_smp_processor_id();
2686
2687         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2688                 return;
2689         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2690         kvm_arch_hardware_disable(NULL);
2691 }
2692
2693 static void hardware_disable(void)
2694 {
2695         raw_spin_lock(&kvm_count_lock);
2696         if (kvm_usage_count)
2697                 hardware_disable_nolock(NULL);
2698         raw_spin_unlock(&kvm_count_lock);
2699 }
2700
2701 static void hardware_disable_all_nolock(void)
2702 {
2703         BUG_ON(!kvm_usage_count);
2704
2705         kvm_usage_count--;
2706         if (!kvm_usage_count)
2707                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2708 }
2709
2710 static void hardware_disable_all(void)
2711 {
2712         raw_spin_lock(&kvm_count_lock);
2713         hardware_disable_all_nolock();
2714         raw_spin_unlock(&kvm_count_lock);
2715 }
2716
2717 static int hardware_enable_all(void)
2718 {
2719         int r = 0;
2720
2721         raw_spin_lock(&kvm_count_lock);
2722
2723         kvm_usage_count++;
2724         if (kvm_usage_count == 1) {
2725                 atomic_set(&hardware_enable_failed, 0);
2726                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2727
2728                 if (atomic_read(&hardware_enable_failed)) {
2729                         hardware_disable_all_nolock();
2730                         r = -EBUSY;
2731                 }
2732         }
2733
2734         raw_spin_unlock(&kvm_count_lock);
2735
2736         return r;
2737 }
2738
2739 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2740                            void *v)
2741 {
2742         int cpu = (long)v;
2743
2744         val &= ~CPU_TASKS_FROZEN;
2745         switch (val) {
2746         case CPU_DYING:
2747                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2748                        cpu);
2749                 hardware_disable();
2750                 break;
2751         case CPU_STARTING:
2752                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2753                        cpu);
2754                 hardware_enable();
2755                 break;
2756         }
2757         return NOTIFY_OK;
2758 }
2759
2760 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2761                       void *v)
2762 {
2763         /*
2764          * Some (well, at least mine) BIOSes hang on reboot if
2765          * in vmx root mode.
2766          *
2767          * And Intel TXT required VMX off for all cpu when system shutdown.
2768          */
2769         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2770         kvm_rebooting = true;
2771         on_each_cpu(hardware_disable_nolock, NULL, 1);
2772         return NOTIFY_OK;
2773 }
2774
2775 static struct notifier_block kvm_reboot_notifier = {
2776         .notifier_call = kvm_reboot,
2777         .priority = 0,
2778 };
2779
2780 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2781 {
2782         int i;
2783
2784         for (i = 0; i < bus->dev_count; i++) {
2785                 struct kvm_io_device *pos = bus->range[i].dev;
2786
2787                 kvm_iodevice_destructor(pos);
2788         }
2789         kfree(bus);
2790 }
2791
2792 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
2793                                  const struct kvm_io_range *r2)
2794 {
2795         if (r1->addr < r2->addr)
2796                 return -1;
2797         if (r1->addr + r1->len > r2->addr + r2->len)
2798                 return 1;
2799         return 0;
2800 }
2801
2802 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2803 {
2804         return kvm_io_bus_cmp(p1, p2);
2805 }
2806
2807 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2808                           gpa_t addr, int len)
2809 {
2810         bus->range[bus->dev_count++] = (struct kvm_io_range) {
2811                 .addr = addr,
2812                 .len = len,
2813                 .dev = dev,
2814         };
2815
2816         sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2817                 kvm_io_bus_sort_cmp, NULL);
2818
2819         return 0;
2820 }
2821
2822 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2823                              gpa_t addr, int len)
2824 {
2825         struct kvm_io_range *range, key;
2826         int off;
2827
2828         key = (struct kvm_io_range) {
2829                 .addr = addr,
2830                 .len = len,
2831         };
2832
2833         range = bsearch(&key, bus->range, bus->dev_count,
2834                         sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2835         if (range == NULL)
2836                 return -ENOENT;
2837
2838         off = range - bus->range;
2839
2840         while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
2841                 off--;
2842
2843         return off;
2844 }
2845
2846 static int __kvm_io_bus_write(struct kvm_io_bus *bus,
2847                               struct kvm_io_range *range, const void *val)
2848 {
2849         int idx;
2850
2851         idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2852         if (idx < 0)
2853                 return -EOPNOTSUPP;
2854
2855         while (idx < bus->dev_count &&
2856                 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
2857                 if (!kvm_iodevice_write(bus->range[idx].dev, range->addr,
2858                                         range->len, val))
2859                         return idx;
2860                 idx++;
2861         }
2862
2863         return -EOPNOTSUPP;
2864 }
2865
2866 /* kvm_io_bus_write - called under kvm->slots_lock */
2867 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2868                      int len, const void *val)
2869 {
2870         struct kvm_io_bus *bus;
2871         struct kvm_io_range range;
2872         int r;
2873
2874         range = (struct kvm_io_range) {
2875                 .addr = addr,
2876                 .len = len,
2877         };
2878
2879         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2880         r = __kvm_io_bus_write(bus, &range, val);
2881         return r < 0 ? r : 0;
2882 }
2883
2884 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
2885 int kvm_io_bus_write_cookie(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2886                             int len, const void *val, long cookie)
2887 {
2888         struct kvm_io_bus *bus;
2889         struct kvm_io_range range;
2890
2891         range = (struct kvm_io_range) {
2892                 .addr = addr,
2893                 .len = len,
2894         };
2895
2896         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2897
2898         /* First try the device referenced by cookie. */
2899         if ((cookie >= 0) && (cookie < bus->dev_count) &&
2900             (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
2901                 if (!kvm_iodevice_write(bus->range[cookie].dev, addr, len,
2902                                         val))
2903                         return cookie;
2904
2905         /*
2906          * cookie contained garbage; fall back to search and return the
2907          * correct cookie value.
2908          */
2909         return __kvm_io_bus_write(bus, &range, val);
2910 }
2911
2912 static int __kvm_io_bus_read(struct kvm_io_bus *bus, struct kvm_io_range *range,
2913                              void *val)
2914 {
2915         int idx;
2916
2917         idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2918         if (idx < 0)
2919                 return -EOPNOTSUPP;
2920
2921         while (idx < bus->dev_count &&
2922                 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
2923                 if (!kvm_iodevice_read(bus->range[idx].dev, range->addr,
2924                                        range->len, val))
2925                         return idx;
2926                 idx++;
2927         }
2928
2929         return -EOPNOTSUPP;
2930 }
2931 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
2932
2933 /* kvm_io_bus_read - called under kvm->slots_lock */
2934 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2935                     int len, void *val)
2936 {
2937         struct kvm_io_bus *bus;
2938         struct kvm_io_range range;
2939         int r;
2940
2941         range = (struct kvm_io_range) {
2942                 .addr = addr,
2943                 .len = len,
2944         };
2945
2946         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2947         r = __kvm_io_bus_read(bus, &range, val);
2948         return r < 0 ? r : 0;
2949 }
2950
2951
2952 /* Caller must hold slots_lock. */
2953 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2954                             int len, struct kvm_io_device *dev)
2955 {
2956         struct kvm_io_bus *new_bus, *bus;
2957
2958         bus = kvm->buses[bus_idx];
2959         /* exclude ioeventfd which is limited by maximum fd */
2960         if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
2961                 return -ENOSPC;
2962
2963         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
2964                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2965         if (!new_bus)
2966                 return -ENOMEM;
2967         memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
2968                sizeof(struct kvm_io_range)));
2969         kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2970         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2971         synchronize_srcu_expedited(&kvm->srcu);
2972         kfree(bus);
2973
2974         return 0;
2975 }
2976
2977 /* Caller must hold slots_lock. */
2978 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2979                               struct kvm_io_device *dev)
2980 {
2981         int i, r;
2982         struct kvm_io_bus *new_bus, *bus;
2983
2984         bus = kvm->buses[bus_idx];
2985         r = -ENOENT;
2986         for (i = 0; i < bus->dev_count; i++)
2987                 if (bus->range[i].dev == dev) {
2988                         r = 0;
2989                         break;
2990                 }
2991
2992         if (r)
2993                 return r;
2994
2995         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
2996                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2997         if (!new_bus)
2998                 return -ENOMEM;
2999
3000         memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3001         new_bus->dev_count--;
3002         memcpy(new_bus->range + i, bus->range + i + 1,
3003                (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3004
3005         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3006         synchronize_srcu_expedited(&kvm->srcu);
3007         kfree(bus);
3008         return r;
3009 }
3010
3011 static struct notifier_block kvm_cpu_notifier = {
3012         .notifier_call = kvm_cpu_hotplug,
3013 };
3014
3015 static int vm_stat_get(void *_offset, u64 *val)
3016 {
3017         unsigned offset = (long)_offset;
3018         struct kvm *kvm;
3019
3020         *val = 0;
3021         spin_lock(&kvm_lock);
3022         list_for_each_entry(kvm, &vm_list, vm_list)
3023                 *val += *(u32 *)((void *)kvm + offset);
3024         spin_unlock(&kvm_lock);
3025         return 0;
3026 }
3027
3028 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
3029
3030 static int vcpu_stat_get(void *_offset, u64 *val)
3031 {
3032         unsigned offset = (long)_offset;
3033         struct kvm *kvm;
3034         struct kvm_vcpu *vcpu;
3035         int i;
3036
3037         *val = 0;
3038         spin_lock(&kvm_lock);
3039         list_for_each_entry(kvm, &vm_list, vm_list)
3040                 kvm_for_each_vcpu(i, vcpu, kvm)
3041                         *val += *(u32 *)((void *)vcpu + offset);
3042
3043         spin_unlock(&kvm_lock);
3044         return 0;
3045 }
3046
3047 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
3048
3049 static const struct file_operations *stat_fops[] = {
3050         [KVM_STAT_VCPU] = &vcpu_stat_fops,
3051         [KVM_STAT_VM]   = &vm_stat_fops,
3052 };
3053
3054 static int kvm_init_debug(void)
3055 {
3056         int r = -EEXIST;
3057         struct kvm_stats_debugfs_item *p;
3058
3059         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3060         if (kvm_debugfs_dir == NULL)
3061                 goto out;
3062
3063         for (p = debugfs_entries; p->name; ++p) {
3064                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
3065                                                 (void *)(long)p->offset,
3066                                                 stat_fops[p->kind]);
3067                 if (p->dentry == NULL)
3068                         goto out_dir;
3069         }
3070
3071         return 0;
3072
3073 out_dir:
3074         debugfs_remove_recursive(kvm_debugfs_dir);
3075 out:
3076         return r;
3077 }
3078
3079 static void kvm_exit_debug(void)
3080 {
3081         struct kvm_stats_debugfs_item *p;
3082
3083         for (p = debugfs_entries; p->name; ++p)
3084                 debugfs_remove(p->dentry);
3085         debugfs_remove(kvm_debugfs_dir);
3086 }
3087
3088 static int kvm_suspend(void)
3089 {
3090         if (kvm_usage_count)
3091                 hardware_disable_nolock(NULL);
3092         return 0;
3093 }
3094
3095 static void kvm_resume(void)
3096 {
3097         if (kvm_usage_count) {
3098                 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3099                 hardware_enable_nolock(NULL);
3100         }
3101 }
3102
3103 static struct syscore_ops kvm_syscore_ops = {
3104         .suspend = kvm_suspend,
3105         .resume = kvm_resume,
3106 };
3107
3108 static inline
3109 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3110 {
3111         return container_of(pn, struct kvm_vcpu, preempt_notifier);
3112 }
3113
3114 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3115 {
3116         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3117         if (vcpu->preempted)
3118                 vcpu->preempted = false;
3119
3120         kvm_arch_vcpu_load(vcpu, cpu);
3121 }
3122
3123 static void kvm_sched_out(struct preempt_notifier *pn,
3124                           struct task_struct *next)
3125 {
3126         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3127
3128         if (current->state == TASK_RUNNING)
3129                 vcpu->preempted = true;
3130         kvm_arch_vcpu_put(vcpu);
3131 }
3132
3133 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3134                   struct module *module)
3135 {
3136         int r;
3137         int cpu;
3138
3139         r = kvm_arch_init(opaque);
3140         if (r)
3141                 goto out_fail;
3142
3143         /*
3144          * kvm_arch_init makes sure there's at most one caller
3145          * for architectures that support multiple implementations,
3146          * like intel and amd on x86.
3147          * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3148          * conflicts in case kvm is already setup for another implementation.
3149          */
3150         r = kvm_irqfd_init();
3151         if (r)
3152                 goto out_irqfd;
3153
3154         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
3155                 r = -ENOMEM;
3156                 goto out_free_0;
3157         }
3158
3159         r = kvm_arch_hardware_setup();
3160         if (r < 0)
3161                 goto out_free_0a;
3162
3163         for_each_online_cpu(cpu) {
3164                 smp_call_function_single(cpu,
3165                                 kvm_arch_check_processor_compat,
3166                                 &r, 1);
3167                 if (r < 0)
3168                         goto out_free_1;
3169         }
3170
3171         r = register_cpu_notifier(&kvm_cpu_notifier);
3172         if (r)
3173                 goto out_free_2;
3174         register_reboot_notifier(&kvm_reboot_notifier);
3175
3176         /* A kmem cache lets us meet the alignment requirements of fx_save. */
3177         if (!vcpu_align)
3178                 vcpu_align = __alignof__(struct kvm_vcpu);
3179         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
3180                                            0, NULL);
3181         if (!kvm_vcpu_cache) {
3182                 r = -ENOMEM;
3183                 goto out_free_3;
3184         }
3185
3186         r = kvm_async_pf_init();
3187         if (r)
3188                 goto out_free;
3189
3190         kvm_chardev_ops.owner = module;
3191         kvm_vm_fops.owner = module;
3192         kvm_vcpu_fops.owner = module;
3193
3194         r = misc_register(&kvm_dev);
3195         if (r) {
3196                 printk(KERN_ERR "kvm: misc device register failed\n");
3197                 goto out_unreg;
3198         }
3199
3200         register_syscore_ops(&kvm_syscore_ops);
3201
3202         kvm_preempt_ops.sched_in = kvm_sched_in;
3203         kvm_preempt_ops.sched_out = kvm_sched_out;
3204
3205         r = kvm_init_debug();
3206         if (r) {
3207                 printk(KERN_ERR "kvm: create debugfs files failed\n");
3208                 goto out_undebugfs;
3209         }
3210
3211         return 0;
3212
3213 out_undebugfs:
3214         unregister_syscore_ops(&kvm_syscore_ops);
3215         misc_deregister(&kvm_dev);
3216 out_unreg:
3217         kvm_async_pf_deinit();
3218 out_free:
3219         kmem_cache_destroy(kvm_vcpu_cache);
3220 out_free_3:
3221         unregister_reboot_notifier(&kvm_reboot_notifier);
3222         unregister_cpu_notifier(&kvm_cpu_notifier);
3223 out_free_2:
3224 out_free_1:
3225         kvm_arch_hardware_unsetup();
3226 out_free_0a:
3227         free_cpumask_var(cpus_hardware_enabled);
3228 out_free_0:
3229         kvm_irqfd_exit();
3230 out_irqfd:
3231         kvm_arch_exit();
3232 out_fail:
3233         return r;
3234 }
3235 EXPORT_SYMBOL_GPL(kvm_init);
3236
3237 void kvm_exit(void)
3238 {
3239         kvm_exit_debug();
3240         misc_deregister(&kvm_dev);
3241         kmem_cache_destroy(kvm_vcpu_cache);
3242         kvm_async_pf_deinit();
3243         unregister_syscore_ops(&kvm_syscore_ops);
3244         unregister_reboot_notifier(&kvm_reboot_notifier);
3245         unregister_cpu_notifier(&kvm_cpu_notifier);
3246         on_each_cpu(hardware_disable_nolock, NULL, 1);
3247         kvm_arch_hardware_unsetup();
3248         kvm_arch_exit();
3249         kvm_irqfd_exit();
3250         free_cpumask_var(cpus_hardware_enabled);
3251 }
3252 EXPORT_SYMBOL_GPL(kvm_exit);