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