virt/kvm/kvm_main.c

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * This module enables machines with Intel VT-x extensions to run virtual
   5  * machines without emulation or binary translation.
   6  *
   7  * Copyright (C) 2006 Qumranet, Inc.
   8  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
   9  *
  10  * Authors:
  11  *   Avi Kivity   <avi@qumranet.com>
  12  *   Yaniv Kamay  <yaniv@qumranet.com>
  13  *
  14  * This work is licensed under the terms of the GNU GPL, version 2.  See
  15  * the COPYING file in the top-level directory.
  16  *
  17  */
  18
  19 #include "iodev.h"
  20
  21 #include <linux/kvm_host.h>
  22 #include <linux/kvm.h>
  23 #include <linux/module.h>
  24 #include <linux/errno.h>
  25 #include <linux/percpu.h>
  26 #include <linux/mm.h>
  27 #include <linux/miscdevice.h>
  28 #include <linux/vmalloc.h>
  29 #include <linux/reboot.h>
  30 #include <linux/debugfs.h>
  31 #include <linux/highmem.h>
  32 #include <linux/file.h>
  33 #include <linux/syscore_ops.h>
  34 #include <linux/cpu.h>
  35 #include <linux/sched.h>
  36 #include <linux/cpumask.h>
  37 #include <linux/smp.h>
  38 #include <linux/anon_inodes.h>
  39 #include <linux/profile.h>
  40 #include <linux/kvm_para.h>
  41 #include <linux/pagemap.h>
  42 #include <linux/mman.h>
  43 #include <linux/swap.h>
  44 #include <linux/bitops.h>
  45 #include <linux/spinlock.h>
  46 #include <linux/compat.h>
  47 #include <linux/srcu.h>
  48 #include <linux/hugetlb.h>
  49 #include <linux/slab.h>
  50 #include <linux/sort.h>
  51 #include <linux/bsearch.h>
  52
  53 #include <asm/processor.h>
  54 #include <asm/io.h>
  55 #include <asm/uaccess.h>
  56 #include <asm/pgtable.h>
  57
  58 #include "coalesced_mmio.h"
  59 #include "async_pf.h"
  60
  61 #define CREATE_TRACE_POINTS
  62 #include <trace/events/kvm.h>
  63
  64 MODULE_AUTHOR("Qumranet");
  65 MODULE_LICENSE("GPL");
  66
  67 /*
  68  * Ordering of locks:
  69  *
  70  *              kvm->lock --> kvm->slots_lock --> kvm->irq_lock
  71  */
  72
  73 DEFINE_SPINLOCK(kvm_lock);
  74 static DEFINE_RAW_SPINLOCK(kvm_count_lock);
  75 LIST_HEAD(vm_list);
  76
  77 static cpumask_var_t cpus_hardware_enabled;
  78 static int kvm_usage_count = 0;
  79 static atomic_t hardware_enable_failed;
  80
  81 struct kmem_cache *kvm_vcpu_cache;
  82 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
  83
  84 static __read_mostly struct preempt_ops kvm_preempt_ops;
  85
  86 struct dentry *kvm_debugfs_dir;
  87
  88 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
  89                            unsigned long arg);
  90 #ifdef CONFIG_COMPAT
  91 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
  92                                   unsigned long arg);
  93 #endif
  94 static int hardware_enable_all(void);
  95 static void hardware_disable_all(void);
  96
  97 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
  98 static void update_memslots(struct kvm_memslots *slots,
  99                             struct kvm_memory_slot *new, u64 last_generation);
 100
 101 static void kvm_release_pfn_dirty(pfn_t pfn);
 102 static void mark_page_dirty_in_slot(struct kvm *kvm,
 103                                     struct kvm_memory_slot *memslot, gfn_t gfn);
 104
 105 __visible bool kvm_rebooting;
 106 EXPORT_SYMBOL_GPL(kvm_rebooting);
 107
 108 static bool largepages_enabled = true;
 109
 110 bool kvm_is_mmio_pfn(pfn_t pfn)
 111 {
 112         if (pfn_valid(pfn))
 113                 return PageReserved(pfn_to_page(pfn));
 114
 115         return true;
 116 }
 117
 118 /*
 119  * Switches to specified vcpu, until a matching vcpu_put()
 120  */
 121 int vcpu_load(struct kvm_vcpu *vcpu)
 122 {
 123         int cpu;
 124
 125         if (mutex_lock_killable(&vcpu->mutex))
 126                 return -EINTR;
 127         if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
 128                 /* The thread running this VCPU changed. */
 129                 struct pid *oldpid = vcpu->pid;
 130                 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
 131                 rcu_assign_pointer(vcpu->pid, newpid);
 132                 synchronize_rcu();
 133                 put_pid(oldpid);
 134         }
 135         cpu = get_cpu();
 136         preempt_notifier_register(&vcpu->preempt_notifier);
 137         kvm_arch_vcpu_load(vcpu, cpu);
 138         put_cpu();
 139         return 0;
 140 }
 141
 142 void vcpu_put(struct kvm_vcpu *vcpu)
 143 {
 144         preempt_disable();
 145         kvm_arch_vcpu_put(vcpu);
 146         preempt_notifier_unregister(&vcpu->preempt_notifier);
 147         preempt_enable();
 148         mutex_unlock(&vcpu->mutex);
 149 }
 150
 151 static void ack_flush(void *_completed)
 152 {
 153 }
 154
 155 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
 156 {
 157         int i, cpu, me;
 158         cpumask_var_t cpus;
 159         bool called = true;
 160         struct kvm_vcpu *vcpu;
 161
 162         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
 163
 164         me = get_cpu();
 165         kvm_for_each_vcpu(i, vcpu, kvm) {
 166                 kvm_make_request(req, vcpu);
 167                 cpu = vcpu->cpu;
 168
 169                 /* Set ->requests bit before we read ->mode */
 170                 smp_mb();
 171
 172                 if (cpus != NULL && cpu != -1 && cpu != me &&
 173                       kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
 174                         cpumask_set_cpu(cpu, cpus);
 175         }
 176         if (unlikely(cpus == NULL))
 177                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
 178         else if (!cpumask_empty(cpus))
 179                 smp_call_function_many(cpus, ack_flush, NULL, 1);
 180         else
 181                 called = false;
 182         put_cpu();
 183         free_cpumask_var(cpus);
 184         return called;
 185 }
 186
 187 void kvm_flush_remote_tlbs(struct kvm *kvm)
 188 {
 189         long dirty_count = kvm->tlbs_dirty;
 190
 191         smp_mb();
 192         if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
 193                 ++kvm->stat.remote_tlb_flush;
 194         cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
 195 }
 196 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
 197
 198 void kvm_reload_remote_mmus(struct kvm *kvm)
 199 {
 200         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
 201 }
 202
 203 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
 204 {
 205         make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
 206 }
 207
 208 void kvm_make_scan_ioapic_request(struct kvm *kvm)
 209 {
 210         make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
 211 }
 212
 213 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 214 {
 215         struct page *page;
 216         int r;
 217
 218         mutex_init(&vcpu->mutex);
 219         vcpu->cpu = -1;
 220         vcpu->kvm = kvm;
 221         vcpu->vcpu_id = id;
 222         vcpu->pid = NULL;
 223         init_waitqueue_head(&vcpu->wq);
 224         kvm_async_pf_vcpu_init(vcpu);
 225
 226         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 227         if (!page) {
 228                 r = -ENOMEM;
 229                 goto fail;
 230         }
 231         vcpu->run = page_address(page);
 232
 233         kvm_vcpu_set_in_spin_loop(vcpu, false);
 234         kvm_vcpu_set_dy_eligible(vcpu, false);
 235         vcpu->preempted = false;
 236
 237         r = kvm_arch_vcpu_init(vcpu);
 238         if (r < 0)
 239                 goto fail_free_run;
 240         return 0;
 241
 242 fail_free_run:
 243         free_page((unsigned long)vcpu->run);
 244 fail:
 245         return r;
 246 }
 247 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 248
 249 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 250 {
 251         put_pid(vcpu->pid);
 252         kvm_arch_vcpu_uninit(vcpu);
 253         free_page((unsigned long)vcpu->run);
 254 }
 255 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 256
 257 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 258 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
 259 {
 260         return container_of(mn, struct kvm, mmu_notifier);
 261 }
 262
 263 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
 264                                              struct mm_struct *mm,
 265                                              unsigned long address)
 266 {
 267         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 268         int need_tlb_flush, idx;
 269
 270         /*
 271          * When ->invalidate_page runs, the linux pte has been zapped
 272          * already but the page is still allocated until
 273          * ->invalidate_page returns. So if we increase the sequence
 274          * here the kvm page fault will notice if the spte can't be
 275          * established because the page is going to be freed. If
 276          * instead the kvm page fault establishes the spte before
 277          * ->invalidate_page runs, kvm_unmap_hva will release it
 278          * before returning.
 279          *
 280          * The sequence increase only need to be seen at spin_unlock
 281          * time, and not at spin_lock time.
 282          *
 283          * Increasing the sequence after the spin_unlock would be
 284          * unsafe because the kvm page fault could then establish the
 285          * pte after kvm_unmap_hva returned, without noticing the page
 286          * is going to be freed.
 287          */
 288         idx = srcu_read_lock(&kvm->srcu);
 289         spin_lock(&kvm->mmu_lock);
 290
 291         kvm->mmu_notifier_seq++;
 292         need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
 293         /* we've to flush the tlb before the pages can be freed */
 294         if (need_tlb_flush)
 295                 kvm_flush_remote_tlbs(kvm);
 296
 297         spin_unlock(&kvm->mmu_lock);
 298         srcu_read_unlock(&kvm->srcu, idx);
 299 }
 300
 301 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
 302                                         struct mm_struct *mm,
 303                                         unsigned long address,
 304                                         pte_t pte)
 305 {
 306         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 307         int idx;
 308
 309         idx = srcu_read_lock(&kvm->srcu);
 310         spin_lock(&kvm->mmu_lock);
 311         kvm->mmu_notifier_seq++;
 312         kvm_set_spte_hva(kvm, address, pte);
 313         spin_unlock(&kvm->mmu_lock);
 314         srcu_read_unlock(&kvm->srcu, idx);
 315 }
 316
 317 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 318                                                     struct mm_struct *mm,
 319                                                     unsigned long start,
 320                                                     unsigned long end)
 321 {
 322         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 323         int need_tlb_flush = 0, idx;
 324
 325         idx = srcu_read_lock(&kvm->srcu);
 326         spin_lock(&kvm->mmu_lock);
 327         /*
 328          * The count increase must become visible at unlock time as no
 329          * spte can be established without taking the mmu_lock and
 330          * count is also read inside the mmu_lock critical section.
 331          */
 332         kvm->mmu_notifier_count++;
 333         need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
 334         need_tlb_flush |= kvm->tlbs_dirty;
 335         /* we've to flush the tlb before the pages can be freed */
 336         if (need_tlb_flush)
 337                 kvm_flush_remote_tlbs(kvm);
 338
 339         spin_unlock(&kvm->mmu_lock);
 340         srcu_read_unlock(&kvm->srcu, idx);
 341 }
 342
 343 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
 344                                                   struct mm_struct *mm,
 345                                                   unsigned long start,
 346                                                   unsigned long end)
 347 {
 348         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 349
 350         spin_lock(&kvm->mmu_lock);
 351         /*
 352          * This sequence increase will notify the kvm page fault that
 353          * the page that is going to be mapped in the spte could have
 354          * been freed.
 355          */
 356         kvm->mmu_notifier_seq++;
 357         smp_wmb();
 358         /*
 359          * The above sequence increase must be visible before the
 360          * below count decrease, which is ensured by the smp_wmb above
 361          * in conjunction with the smp_rmb in mmu_notifier_retry().
 362          */
 363         kvm->mmu_notifier_count--;
 364         spin_unlock(&kvm->mmu_lock);
 365
 366         BUG_ON(kvm->mmu_notifier_count < 0);
 367 }
 368
 369 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
 370                                               struct mm_struct *mm,
 371                                               unsigned long address)
 372 {
 373         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 374         int young, idx;
 375
 376         idx = srcu_read_lock(&kvm->srcu);
 377         spin_lock(&kvm->mmu_lock);
 378
 379         young = kvm_age_hva(kvm, address);
 380         if (young)
 381                 kvm_flush_remote_tlbs(kvm);
 382
 383         spin_unlock(&kvm->mmu_lock);
 384         srcu_read_unlock(&kvm->srcu, idx);
 385
 386         return young;
 387 }
 388
 389 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
 390                                        struct mm_struct *mm,
 391                                        unsigned long address)
 392 {
 393         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 394         int young, idx;
 395
 396         idx = srcu_read_lock(&kvm->srcu);
 397         spin_lock(&kvm->mmu_lock);
 398         young = kvm_test_age_hva(kvm, address);
 399         spin_unlock(&kvm->mmu_lock);
 400         srcu_read_unlock(&kvm->srcu, idx);
 401
 402         return young;
 403 }
 404
 405 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
 406                                      struct mm_struct *mm)
 407 {
 408         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 409         int idx;
 410
 411         idx = srcu_read_lock(&kvm->srcu);
 412         kvm_arch_flush_shadow_all(kvm);
 413         srcu_read_unlock(&kvm->srcu, idx);
 414 }
 415
 416 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 417         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
 418         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
 419         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
 420         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
 421         .test_young             = kvm_mmu_notifier_test_young,
 422         .change_pte             = kvm_mmu_notifier_change_pte,
 423         .release                = kvm_mmu_notifier_release,
 424 };
 425
 426 static int kvm_init_mmu_notifier(struct kvm *kvm)
 427 {
 428         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
 429         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
 430 }
 431
 432 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
 433
 434 static int kvm_init_mmu_notifier(struct kvm *kvm)
 435 {
 436         return 0;
 437 }
 438
 439 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
 440
 441 static void kvm_init_memslots_id(struct kvm *kvm)
 442 {
 443         int i;
 444         struct kvm_memslots *slots = kvm->memslots;
 445
 446         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
 447                 slots->id_to_index[i] = slots->memslots[i].id = i;
 448 }
 449
 450 static struct kvm *kvm_create_vm(unsigned long type)
 451 {
 452         int r, i;
 453         struct kvm *kvm = kvm_arch_alloc_vm();
 454
 455         if (!kvm)
 456                 return ERR_PTR(-ENOMEM);
 457
 458         r = kvm_arch_init_vm(kvm, type);
 459         if (r)
 460                 goto out_err_no_disable;
 461
 462         r = hardware_enable_all();
 463         if (r)
 464                 goto out_err_no_disable;
 465
 466 #ifdef CONFIG_HAVE_KVM_IRQCHIP
 467         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
 468 #endif
 469 #ifdef CONFIG_HAVE_KVM_IRQFD
 470         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
 471 #endif
 472
 473         BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
 474
 475         r = -ENOMEM;
 476         kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 477         if (!kvm->memslots)
 478                 goto out_err_no_srcu;
 479         kvm_init_memslots_id(kvm);
 480         if (init_srcu_struct(&kvm->srcu))
 481                 goto out_err_no_srcu;
 482         if (init_srcu_struct(&kvm->irq_srcu))
 483                 goto out_err_no_irq_srcu;
 484         for (i = 0; i < KVM_NR_BUSES; i++) {
 485                 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
 486                                         GFP_KERNEL);
 487                 if (!kvm->buses[i])
 488                         goto out_err;
 489         }
 490
 491         spin_lock_init(&kvm->mmu_lock);
 492         kvm->mm = current->mm;
 493         atomic_inc(&kvm->mm->mm_count);
 494         kvm_eventfd_init(kvm);
 495         mutex_init(&kvm->lock);
 496         mutex_init(&kvm->irq_lock);
 497         mutex_init(&kvm->slots_lock);
 498         atomic_set(&kvm->users_count, 1);
 499         INIT_LIST_HEAD(&kvm->devices);
 500
 501         r = kvm_init_mmu_notifier(kvm);
 502         if (r)
 503                 goto out_err;
 504
 505         spin_lock(&kvm_lock);
 506         list_add(&kvm->vm_list, &vm_list);
 507         spin_unlock(&kvm_lock);
 508
 509         return kvm;
 510
 511 out_err:
 512         cleanup_srcu_struct(&kvm->irq_srcu);
 513 out_err_no_irq_srcu:
 514         cleanup_srcu_struct(&kvm->srcu);
 515 out_err_no_srcu:
 516         hardware_disable_all();
 517 out_err_no_disable:
 518         for (i = 0; i < KVM_NR_BUSES; i++)
 519                 kfree(kvm->buses[i]);
 520         kfree(kvm->memslots);
 521         kvm_arch_free_vm(kvm);
 522         return ERR_PTR(r);
 523 }
 524
 525 /*
 526  * Avoid using vmalloc for a small buffer.
 527  * Should not be used when the size is statically known.
 528  */
 529 void *kvm_kvzalloc(unsigned long size)
 530 {
 531         if (size > PAGE_SIZE)
 532                 return vzalloc(size);
 533         else
 534                 return kzalloc(size, GFP_KERNEL);
 535 }
 536
 537 void kvm_kvfree(const void *addr)
 538 {
 539         if (is_vmalloc_addr(addr))
 540                 vfree(addr);
 541         else
 542                 kfree(addr);
 543 }
 544
 545 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
 546 {
 547         if (!memslot->dirty_bitmap)
 548                 return;
 549
 550         kvm_kvfree(memslot->dirty_bitmap);
 551         memslot->dirty_bitmap = NULL;
 552 }
 553
 554 /*
 555  * Free any memory in @free but not in @dont.
 556  */
 557 static void kvm_free_physmem_slot(struct kvm *kvm, struct kvm_memory_slot *free,
 558                                   struct kvm_memory_slot *dont)
 559 {
 560         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 561                 kvm_destroy_dirty_bitmap(free);
 562
 563         kvm_arch_free_memslot(kvm, free, dont);
 564
 565         free->npages = 0;
 566 }
 567
 568 static void kvm_free_physmem(struct kvm *kvm)
 569 {
 570         struct kvm_memslots *slots = kvm->memslots;
 571         struct kvm_memory_slot *memslot;
 572
 573         kvm_for_each_memslot(memslot, slots)
 574                 kvm_free_physmem_slot(kvm, memslot, NULL);
 575
 576         kfree(kvm->memslots);
 577 }
 578
 579 static void kvm_destroy_devices(struct kvm *kvm)
 580 {
 581         struct list_head *node, *tmp;
 582
 583         list_for_each_safe(node, tmp, &kvm->devices) {
 584                 struct kvm_device *dev =
 585                         list_entry(node, struct kvm_device, vm_node);
 586
 587                 list_del(node);
 588                 dev->ops->destroy(dev);
 589         }
 590 }
 591
 592 static void kvm_destroy_vm(struct kvm *kvm)
 593 {
 594         int i;
 595         struct mm_struct *mm = kvm->mm;
 596
 597         kvm_arch_sync_events(kvm);
 598         spin_lock(&kvm_lock);
 599         list_del(&kvm->vm_list);
 600         spin_unlock(&kvm_lock);
 601         kvm_free_irq_routing(kvm);
 602         for (i = 0; i < KVM_NR_BUSES; i++)
 603                 kvm_io_bus_destroy(kvm->buses[i]);
 604         kvm_coalesced_mmio_free(kvm);
 605 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 606         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
 607 #else
 608         kvm_arch_flush_shadow_all(kvm);
 609 #endif
 610         kvm_arch_destroy_vm(kvm);
 611         kvm_destroy_devices(kvm);
 612         kvm_free_physmem(kvm);
 613         cleanup_srcu_struct(&kvm->irq_srcu);
 614         cleanup_srcu_struct(&kvm->srcu);
 615         kvm_arch_free_vm(kvm);
 616         hardware_disable_all();
 617         mmdrop(mm);
 618 }
 619
 620 void kvm_get_kvm(struct kvm *kvm)
 621 {
 622         atomic_inc(&kvm->users_count);
 623 }
 624 EXPORT_SYMBOL_GPL(kvm_get_kvm);
 625
 626 void kvm_put_kvm(struct kvm *kvm)
 627 {
 628         if (atomic_dec_and_test(&kvm->users_count))
 629                 kvm_destroy_vm(kvm);
 630 }
 631 EXPORT_SYMBOL_GPL(kvm_put_kvm);
 632
 633
 634 static int kvm_vm_release(struct inode *inode, struct file *filp)
 635 {
 636         struct kvm *kvm = filp->private_data;
 637
 638         kvm_irqfd_release(kvm);
 639
 640         kvm_put_kvm(kvm);
 641         return 0;
 642 }
 643
 644 /*
 645  * Allocation size is twice as large as the actual dirty bitmap size.
 646  * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
 647  */
 648 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
 649 {
 650         unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
 651
 652         memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
 653         if (!memslot->dirty_bitmap)
 654                 return -ENOMEM;
 655
 656         return 0;
 657 }
 658
 659 static int cmp_memslot(const void *slot1, const void *slot2)
 660 {
 661         struct kvm_memory_slot *s1, *s2;
 662
 663         s1 = (struct kvm_memory_slot *)slot1;
 664         s2 = (struct kvm_memory_slot *)slot2;
 665
 666         if (s1->npages < s2->npages)
 667                 return 1;
 668         if (s1->npages > s2->npages)
 669                 return -1;
 670
 671         return 0;
 672 }
 673
 674 /*
 675  * Sort the memslots base on its size, so the larger slots
 676  * will get better fit.
 677  */
 678 static void sort_memslots(struct kvm_memslots *slots)
 679 {
 680         int i;
 681
 682         sort(slots->memslots, KVM_MEM_SLOTS_NUM,
 683               sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
 684
 685         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
 686                 slots->id_to_index[slots->memslots[i].id] = i;
 687 }
 688
 689 static void update_memslots(struct kvm_memslots *slots,
 690                             struct kvm_memory_slot *new,
 691                             u64 last_generation)
 692 {
 693         if (new) {
 694                 int id = new->id;
 695                 struct kvm_memory_slot *old = id_to_memslot(slots, id);
 696                 unsigned long npages = old->npages;
 697
 698                 *old = *new;
 699                 if (new->npages != npages)
 700                         sort_memslots(slots);
 701         }
 702
 703         slots->generation = last_generation + 1;
 704 }
 705
 706 static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
 707 {
 708         u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
 709
 710 #ifdef KVM_CAP_READONLY_MEM
 711         valid_flags |= KVM_MEM_READONLY;
 712 #endif
 713
 714         if (mem->flags & ~valid_flags)
 715                 return -EINVAL;
 716
 717         return 0;
 718 }
 719
 720 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
 721                 struct kvm_memslots *slots, struct kvm_memory_slot *new)
 722 {
 723         struct kvm_memslots *old_memslots = kvm->memslots;
 724
 725         update_memslots(slots, new, kvm->memslots->generation);
 726         rcu_assign_pointer(kvm->memslots, slots);
 727         synchronize_srcu_expedited(&kvm->srcu);
 728
 729         kvm_arch_memslots_updated(kvm);
 730
 731         return old_memslots;
 732 }
 733
 734 /*
 735  * Allocate some memory and give it an address in the guest physical address
 736  * space.
 737  *
 738  * Discontiguous memory is allowed, mostly for framebuffers.
 739  *
 740  * Must be called holding mmap_sem for write.
 741  */
 742 int __kvm_set_memory_region(struct kvm *kvm,
 743                             struct kvm_userspace_memory_region *mem)
 744 {
 745         int r;
 746         gfn_t base_gfn;
 747         unsigned long npages;
 748         struct kvm_memory_slot *slot;
 749         struct kvm_memory_slot old, new;
 750         struct kvm_memslots *slots = NULL, *old_memslots;
 751         enum kvm_mr_change change;
 752
 753         r = check_memory_region_flags(mem);
 754         if (r)
 755                 goto out;
 756
 757         r = -EINVAL;
 758         /* General sanity checks */
 759         if (mem->memory_size & (PAGE_SIZE - 1))
 760                 goto out;
 761         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 762                 goto out;
 763         /* We can read the guest memory with __xxx_user() later on. */
 764         if ((mem->slot < KVM_USER_MEM_SLOTS) &&
 765             ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
 766              !access_ok(VERIFY_WRITE,
 767                         (void __user *)(unsigned long)mem->userspace_addr,
 768                         mem->memory_size)))
 769                 goto out;
 770         if (mem->slot >= KVM_MEM_SLOTS_NUM)
 771                 goto out;
 772         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 773                 goto out;
 774
 775         slot = id_to_memslot(kvm->memslots, mem->slot);
 776         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 777         npages = mem->memory_size >> PAGE_SHIFT;
 778
 779         r = -EINVAL;
 780         if (npages > KVM_MEM_MAX_NR_PAGES)
 781                 goto out;
 782
 783         if (!npages)
 784                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 785
 786         new = old = *slot;
 787
 788         new.id = mem->slot;
 789         new.base_gfn = base_gfn;
 790         new.npages = npages;
 791         new.flags = mem->flags;
 792
 793         r = -EINVAL;
 794         if (npages) {
 795                 if (!old.npages)
 796                         change = KVM_MR_CREATE;
 797                 else { /* Modify an existing slot. */
 798                         if ((mem->userspace_addr != old.userspace_addr) ||
 799                             (npages != old.npages) ||
 800                             ((new.flags ^ old.flags) & KVM_MEM_READONLY))
 801                                 goto out;
 802
 803                         if (base_gfn != old.base_gfn)
 804                                 change = KVM_MR_MOVE;
 805                         else if (new.flags != old.flags)
 806                                 change = KVM_MR_FLAGS_ONLY;
 807                         else { /* Nothing to change. */
 808                                 r = 0;
 809                                 goto out;
 810                         }
 811                 }
 812         } else if (old.npages) {
 813                 change = KVM_MR_DELETE;
 814         } else /* Modify a non-existent slot: disallowed. */
 815                 goto out;
 816
 817         if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
 818                 /* Check for overlaps */
 819                 r = -EEXIST;
 820                 kvm_for_each_memslot(slot, kvm->memslots) {
 821                         if ((slot->id >= KVM_USER_MEM_SLOTS) ||
 822                             (slot->id == mem->slot))
 823                                 continue;
 824                         if (!((base_gfn + npages <= slot->base_gfn) ||
 825                               (base_gfn >= slot->base_gfn + slot->npages)))
 826                                 goto out;
 827                 }
 828         }
 829
 830         /* Free page dirty bitmap if unneeded */
 831         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 832                 new.dirty_bitmap = NULL;
 833
 834         r = -ENOMEM;
 835         if (change == KVM_MR_CREATE) {
 836                 new.userspace_addr = mem->userspace_addr;
 837
 838                 if (kvm_arch_create_memslot(kvm, &new, npages))
 839                         goto out_free;
 840         }
 841
 842         /* Allocate page dirty bitmap if needed */
 843         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 844                 if (kvm_create_dirty_bitmap(&new) < 0)
 845                         goto out_free;
 846         }
 847
 848         if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
 849                 r = -ENOMEM;
 850                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 851                                 GFP_KERNEL);
 852                 if (!slots)
 853                         goto out_free;
 854                 slot = id_to_memslot(slots, mem->slot);
 855                 slot->flags |= KVM_MEMSLOT_INVALID;
 856
 857                 old_memslots = install_new_memslots(kvm, slots, NULL);
 858
 859                 /* slot was deleted or moved, clear iommu mapping */
 860                 kvm_iommu_unmap_pages(kvm, &old);
 861                 /* From this point no new shadow pages pointing to a deleted,
 862                  * or moved, memslot will be created.
 863                  *
 864                  * validation of sp->gfn happens in:
 865                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
 866                  *      - kvm_is_visible_gfn (mmu_check_roots)
 867                  */
 868                 kvm_arch_flush_shadow_memslot(kvm, slot);
 869                 slots = old_memslots;
 870         }
 871
 872         r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
 873         if (r)
 874                 goto out_slots;
 875
 876         r = -ENOMEM;
 877         /*
 878          * We can re-use the old_memslots from above, the only difference
 879          * from the currently installed memslots is the invalid flag.  This
 880          * will get overwritten by update_memslots anyway.
 881          */
 882         if (!slots) {
 883                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 884                                 GFP_KERNEL);
 885                 if (!slots)
 886                         goto out_free;
 887         }
 888
 889         /* actual memory is freed via old in kvm_free_physmem_slot below */
 890         if (change == KVM_MR_DELETE) {
 891                 new.dirty_bitmap = NULL;
 892                 memset(&new.arch, 0, sizeof(new.arch));
 893         }
 894
 895         old_memslots = install_new_memslots(kvm, slots, &new);
 896
 897         kvm_arch_commit_memory_region(kvm, mem, &old, change);
 898
 899         kvm_free_physmem_slot(kvm, &old, &new);
 900         kfree(old_memslots);
 901
 902         /*
 903          * IOMMU mapping:  New slots need to be mapped.  Old slots need to be
 904          * un-mapped and re-mapped if their base changes.  Since base change
 905          * unmapping is handled above with slot deletion, mapping alone is
 906          * needed here.  Anything else the iommu might care about for existing
 907          * slots (size changes, userspace addr changes and read-only flag
 908          * changes) is disallowed above, so any other attribute changes getting
 909          * here can be skipped.
 910          */
 911         if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
 912                 r = kvm_iommu_map_pages(kvm, &new);
 913                 return r;
 914         }
 915
 916         return 0;
 917
 918 out_slots:
 919         kfree(slots);
 920 out_free:
 921         kvm_free_physmem_slot(kvm, &new, &old);
 922 out:
 923         return r;
 924 }
 925 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 926
 927 int kvm_set_memory_region(struct kvm *kvm,
 928                           struct kvm_userspace_memory_region *mem)
 929 {
 930         int r;
 931
 932         mutex_lock(&kvm->slots_lock);
 933         r = __kvm_set_memory_region(kvm, mem);
 934         mutex_unlock(&kvm->slots_lock);
 935         return r;
 936 }
 937 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 938
 939 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 940                                           struct kvm_userspace_memory_region *mem)
 941 {
 942         if (mem->slot >= KVM_USER_MEM_SLOTS)
 943                 return -EINVAL;
 944         return kvm_set_memory_region(kvm, mem);
 945 }
 946
 947 int kvm_get_dirty_log(struct kvm *kvm,
 948                         struct kvm_dirty_log *log, int *is_dirty)
 949 {
 950         struct kvm_memory_slot *memslot;
 951         int r, i;
 952         unsigned long n;
 953         unsigned long any = 0;
 954
 955         r = -EINVAL;
 956         if (log->slot >= KVM_USER_MEM_SLOTS)
 957                 goto out;
 958
 959         memslot = id_to_memslot(kvm->memslots, log->slot);
 960         r = -ENOENT;
 961         if (!memslot->dirty_bitmap)
 962                 goto out;
 963
 964         n = kvm_dirty_bitmap_bytes(memslot);
 965
 966         for (i = 0; !any && i < n/sizeof(long); ++i)
 967                 any = memslot->dirty_bitmap[i];
 968
 969         r = -EFAULT;
 970         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 971                 goto out;
 972
 973         if (any)
 974                 *is_dirty = 1;
 975
 976         r = 0;
 977 out:
 978         return r;
 979 }
 980 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
 981
 982 bool kvm_largepages_enabled(void)
 983 {
 984         return largepages_enabled;
 985 }
 986
 987 void kvm_disable_largepages(void)
 988 {
 989         largepages_enabled = false;
 990 }
 991 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
 992
 993 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 994 {
 995         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
 996 }
 997 EXPORT_SYMBOL_GPL(gfn_to_memslot);
 998
 999 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1000 {
1001         struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1002
1003         if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1004               memslot->flags & KVM_MEMSLOT_INVALID)
1005                 return 0;
1006
1007         return 1;
1008 }
1009 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1010
1011 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1012 {
1013         struct vm_area_struct *vma;
1014         unsigned long addr, size;
1015
1016         size = PAGE_SIZE;
1017
1018         addr = gfn_to_hva(kvm, gfn);
1019         if (kvm_is_error_hva(addr))
1020                 return PAGE_SIZE;
1021
1022         down_read(&current->mm->mmap_sem);
1023         vma = find_vma(current->mm, addr);
1024         if (!vma)
1025                 goto out;
1026
1027         size = vma_kernel_pagesize(vma);
1028
1029 out:
1030         up_read(&current->mm->mmap_sem);
1031
1032         return size;
1033 }
1034
1035 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1036 {
1037         return slot->flags & KVM_MEM_READONLY;
1038 }
1039
1040 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1041                                        gfn_t *nr_pages, bool write)
1042 {
1043         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1044                 return KVM_HVA_ERR_BAD;
1045
1046         if (memslot_is_readonly(slot) && write)
1047                 return KVM_HVA_ERR_RO_BAD;
1048
1049         if (nr_pages)
1050                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1051
1052         return __gfn_to_hva_memslot(slot, gfn);
1053 }
1054
1055 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1056                                      gfn_t *nr_pages)
1057 {
1058         return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1059 }
1060
1061 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1062                                         gfn_t gfn)
1063 {
1064         return gfn_to_hva_many(slot, gfn, NULL);
1065 }
1066 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1067
1068 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1069 {
1070         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1071 }
1072 EXPORT_SYMBOL_GPL(gfn_to_hva);
1073
1074 /*
1075  * If writable is set to false, the hva returned by this function is only
1076  * allowed to be read.
1077  */
1078 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1079 {
1080         struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1081         unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1082
1083         if (!kvm_is_error_hva(hva) && writable)
1084                 *writable = !memslot_is_readonly(slot);
1085
1086         return hva;
1087 }
1088
1089 static int kvm_read_hva(void *data, void __user *hva, int len)
1090 {
1091         return __copy_from_user(data, hva, len);
1092 }
1093
1094 static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1095 {
1096         return __copy_from_user_inatomic(data, hva, len);
1097 }
1098
1099 static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1100         unsigned long start, int write, struct page **page)
1101 {
1102         int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1103
1104         if (write)
1105                 flags |= FOLL_WRITE;
1106
1107         return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1108 }
1109
1110 static inline int check_user_page_hwpoison(unsigned long addr)
1111 {
1112         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1113
1114         rc = __get_user_pages(current, current->mm, addr, 1,
1115                               flags, NULL, NULL, NULL);
1116         return rc == -EHWPOISON;
1117 }
1118
1119 /*
1120  * The atomic path to get the writable pfn which will be stored in @pfn,
1121  * true indicates success, otherwise false is returned.
1122  */
1123 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1124                             bool write_fault, bool *writable, pfn_t *pfn)
1125 {
1126         struct page *page[1];
1127         int npages;
1128
1129         if (!(async || atomic))
1130                 return false;
1131
1132         /*
1133          * Fast pin a writable pfn only if it is a write fault request
1134          * or the caller allows to map a writable pfn for a read fault
1135          * request.
1136          */
1137         if (!(write_fault || writable))
1138                 return false;
1139
1140         npages = __get_user_pages_fast(addr, 1, 1, page);
1141         if (npages == 1) {
1142                 *pfn = page_to_pfn(page[0]);
1143
1144                 if (writable)
1145                         *writable = true;
1146                 return true;
1147         }
1148
1149         return false;
1150 }
1151
1152 /*
1153  * The slow path to get the pfn of the specified host virtual address,
1154  * 1 indicates success, -errno is returned if error is detected.
1155  */
1156 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1157                            bool *writable, pfn_t *pfn)
1158 {
1159         struct page *page[1];
1160         int npages = 0;
1161
1162         might_sleep();
1163
1164         if (writable)
1165                 *writable = write_fault;
1166
1167         if (async) {
1168                 down_read(&current->mm->mmap_sem);
1169                 npages = get_user_page_nowait(current, current->mm,
1170                                               addr, write_fault, page);
1171                 up_read(&current->mm->mmap_sem);
1172         } else
1173                 npages = get_user_pages_fast(addr, 1, write_fault,
1174                                              page);
1175         if (npages != 1)
1176                 return npages;
1177
1178         /* map read fault as writable if possible */
1179         if (unlikely(!write_fault) && writable) {
1180                 struct page *wpage[1];
1181
1182                 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1183                 if (npages == 1) {
1184                         *writable = true;
1185                         put_page(page[0]);
1186                         page[0] = wpage[0];
1187                 }
1188
1189                 npages = 1;
1190         }
1191         *pfn = page_to_pfn(page[0]);
1192         return npages;
1193 }
1194
1195 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1196 {
1197         if (unlikely(!(vma->vm_flags & VM_READ)))
1198                 return false;
1199
1200         if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1201                 return false;
1202
1203         return true;
1204 }
1205
1206 /*
1207  * Pin guest page in memory and return its pfn.
1208  * @addr: host virtual address which maps memory to the guest
1209  * @atomic: whether this function can sleep
1210  * @async: whether this function need to wait IO complete if the
1211  *         host page is not in the memory
1212  * @write_fault: whether we should get a writable host page
1213  * @writable: whether it allows to map a writable host page for !@write_fault
1214  *
1215  * The function will map a writable host page for these two cases:
1216  * 1): @write_fault = true
1217  * 2): @write_fault = false && @writable, @writable will tell the caller
1218  *     whether the mapping is writable.
1219  */
1220 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1221                         bool write_fault, bool *writable)
1222 {
1223         struct vm_area_struct *vma;
1224         pfn_t pfn = 0;
1225         int npages;
1226
1227         /* we can do it either atomically or asynchronously, not both */
1228         BUG_ON(atomic && async);
1229
1230         if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1231                 return pfn;
1232
1233         if (atomic)
1234                 return KVM_PFN_ERR_FAULT;
1235
1236         npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1237         if (npages == 1)
1238                 return pfn;
1239
1240         down_read(&current->mm->mmap_sem);
1241         if (npages == -EHWPOISON ||
1242               (!async && check_user_page_hwpoison(addr))) {
1243                 pfn = KVM_PFN_ERR_HWPOISON;
1244                 goto exit;
1245         }
1246
1247         vma = find_vma_intersection(current->mm, addr, addr + 1);
1248
1249         if (vma == NULL)
1250                 pfn = KVM_PFN_ERR_FAULT;
1251         else if ((vma->vm_flags & VM_PFNMAP)) {
1252                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1253                         vma->vm_pgoff;
1254                 BUG_ON(!kvm_is_mmio_pfn(pfn));
1255         } else {
1256                 if (async && vma_is_valid(vma, write_fault))
1257                         *async = true;
1258                 pfn = KVM_PFN_ERR_FAULT;
1259         }
1260 exit:
1261         up_read(&current->mm->mmap_sem);
1262         return pfn;
1263 }
1264
1265 static pfn_t
1266 __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1267                      bool *async, bool write_fault, bool *writable)
1268 {
1269         unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1270
1271         if (addr == KVM_HVA_ERR_RO_BAD)
1272                 return KVM_PFN_ERR_RO_FAULT;
1273
1274         if (kvm_is_error_hva(addr))
1275                 return KVM_PFN_NOSLOT;
1276
1277         /* Do not map writable pfn in the readonly memslot. */
1278         if (writable && memslot_is_readonly(slot)) {
1279                 *writable = false;
1280                 writable = NULL;
1281         }
1282
1283         return hva_to_pfn(addr, atomic, async, write_fault,
1284                           writable);
1285 }
1286
1287 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1288                           bool write_fault, bool *writable)
1289 {
1290         struct kvm_memory_slot *slot;
1291
1292         if (async)
1293                 *async = false;
1294
1295         slot = gfn_to_memslot(kvm, gfn);
1296
1297         return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1298                                     writable);
1299 }
1300
1301 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1302 {
1303         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1304 }
1305 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1306
1307 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1308                        bool write_fault, bool *writable)
1309 {
1310         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1311 }
1312 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1313
1314 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1315 {
1316         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1317 }
1318 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1319
1320 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1321                       bool *writable)
1322 {
1323         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1324 }
1325 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1326
1327 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1328 {
1329         return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1330 }
1331
1332 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1333 {
1334         return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1335 }
1336 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1337
1338 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1339                                                                   int nr_pages)
1340 {
1341         unsigned long addr;
1342         gfn_t entry;
1343
1344         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1345         if (kvm_is_error_hva(addr))
1346                 return -1;
1347
1348         if (entry < nr_pages)
1349                 return 0;
1350
1351         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1352 }
1353 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1354
1355 static struct page *kvm_pfn_to_page(pfn_t pfn)
1356 {
1357         if (is_error_noslot_pfn(pfn))
1358                 return KVM_ERR_PTR_BAD_PAGE;
1359
1360         if (kvm_is_mmio_pfn(pfn)) {
1361                 WARN_ON(1);
1362                 return KVM_ERR_PTR_BAD_PAGE;
1363         }
1364
1365         return pfn_to_page(pfn);
1366 }
1367
1368 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1369 {
1370         pfn_t pfn;
1371
1372         pfn = gfn_to_pfn(kvm, gfn);
1373
1374         return kvm_pfn_to_page(pfn);
1375 }
1376
1377 EXPORT_SYMBOL_GPL(gfn_to_page);
1378
1379 void kvm_release_page_clean(struct page *page)
1380 {
1381         WARN_ON(is_error_page(page));
1382
1383         kvm_release_pfn_clean(page_to_pfn(page));
1384 }
1385 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1386
1387 void kvm_release_pfn_clean(pfn_t pfn)
1388 {
1389         if (!is_error_noslot_pfn(pfn) && !kvm_is_mmio_pfn(pfn))
1390                 put_page(pfn_to_page(pfn));
1391 }
1392 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1393
1394 void kvm_release_page_dirty(struct page *page)
1395 {
1396         WARN_ON(is_error_page(page));
1397
1398         kvm_release_pfn_dirty(page_to_pfn(page));
1399 }
1400 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1401
1402 static void kvm_release_pfn_dirty(pfn_t pfn)
1403 {
1404         kvm_set_pfn_dirty(pfn);
1405         kvm_release_pfn_clean(pfn);
1406 }
1407
1408 void kvm_set_pfn_dirty(pfn_t pfn)
1409 {
1410         if (!kvm_is_mmio_pfn(pfn)) {
1411                 struct page *page = pfn_to_page(pfn);
1412                 if (!PageReserved(page))
1413                         SetPageDirty(page);
1414         }
1415 }
1416 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1417
1418 void kvm_set_pfn_accessed(pfn_t pfn)
1419 {
1420         if (!kvm_is_mmio_pfn(pfn))
1421                 mark_page_accessed(pfn_to_page(pfn));
1422 }
1423 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1424
1425 void kvm_get_pfn(pfn_t pfn)
1426 {
1427         if (!kvm_is_mmio_pfn(pfn))
1428                 get_page(pfn_to_page(pfn));
1429 }
1430 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1431
1432 static int next_segment(unsigned long len, int offset)
1433 {
1434         if (len > PAGE_SIZE - offset)
1435                 return PAGE_SIZE - offset;
1436         else
1437                 return len;
1438 }
1439
1440 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1441                         int len)
1442 {
1443         int r;
1444         unsigned long addr;
1445
1446         addr = gfn_to_hva_prot(kvm, gfn, NULL);
1447         if (kvm_is_error_hva(addr))
1448                 return -EFAULT;
1449         r = kvm_read_hva(data, (void __user *)addr + offset, len);
1450         if (r)
1451                 return -EFAULT;
1452         return 0;
1453 }
1454 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1455
1456 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1457 {
1458         gfn_t gfn = gpa >> PAGE_SHIFT;
1459         int seg;
1460         int offset = offset_in_page(gpa);
1461         int ret;
1462
1463         while ((seg = next_segment(len, offset)) != 0) {
1464                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1465                 if (ret < 0)
1466                         return ret;
1467                 offset = 0;
1468                 len -= seg;
1469                 data += seg;
1470                 ++gfn;
1471         }
1472         return 0;
1473 }
1474 EXPORT_SYMBOL_GPL(kvm_read_guest);
1475
1476 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1477                           unsigned long len)
1478 {
1479         int r;
1480         unsigned long addr;
1481         gfn_t gfn = gpa >> PAGE_SHIFT;
1482         int offset = offset_in_page(gpa);
1483
1484         addr = gfn_to_hva_prot(kvm, gfn, NULL);
1485         if (kvm_is_error_hva(addr))
1486                 return -EFAULT;
1487         pagefault_disable();
1488         r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
1489         pagefault_enable();
1490         if (r)
1491                 return -EFAULT;
1492         return 0;
1493 }
1494 EXPORT_SYMBOL(kvm_read_guest_atomic);
1495
1496 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1497                          int offset, int len)
1498 {
1499         int r;
1500         unsigned long addr;
1501
1502         addr = gfn_to_hva(kvm, gfn);
1503         if (kvm_is_error_hva(addr))
1504                 return -EFAULT;
1505         r = __copy_to_user((void __user *)addr + offset, data, len);
1506         if (r)
1507                 return -EFAULT;
1508         mark_page_dirty(kvm, gfn);
1509         return 0;
1510 }
1511 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1512
1513 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1514                     unsigned long len)
1515 {
1516         gfn_t gfn = gpa >> PAGE_SHIFT;
1517         int seg;
1518         int offset = offset_in_page(gpa);
1519         int ret;
1520
1521         while ((seg = next_segment(len, offset)) != 0) {
1522                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1523                 if (ret < 0)
1524                         return ret;
1525                 offset = 0;
1526                 len -= seg;
1527                 data += seg;
1528                 ++gfn;
1529         }
1530         return 0;
1531 }
1532
1533 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1534                               gpa_t gpa, unsigned long len)
1535 {
1536         struct kvm_memslots *slots = kvm_memslots(kvm);
1537         int offset = offset_in_page(gpa);
1538         gfn_t start_gfn = gpa >> PAGE_SHIFT;
1539         gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1540         gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1541         gfn_t nr_pages_avail;
1542
1543         ghc->gpa = gpa;
1544         ghc->generation = slots->generation;
1545         ghc->len = len;
1546         ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1547         ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, &nr_pages_avail);
1548         if (!kvm_is_error_hva(ghc->hva) && nr_pages_avail >= nr_pages_needed) {
1549                 ghc->hva += offset;
1550         } else {
1551                 /*
1552                  * If the requested region crosses two memslots, we still
1553                  * verify that the entire region is valid here.
1554                  */
1555                 while (start_gfn <= end_gfn) {
1556                         ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1557                         ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1558                                                    &nr_pages_avail);
1559                         if (kvm_is_error_hva(ghc->hva))
1560                                 return -EFAULT;
1561                         start_gfn += nr_pages_avail;
1562                 }
1563                 /* Use the slow path for cross page reads and writes. */
1564                 ghc->memslot = NULL;
1565         }
1566         return 0;
1567 }
1568 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1569
1570 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1571                            void *data, unsigned long len)
1572 {
1573         struct kvm_memslots *slots = kvm_memslots(kvm);
1574         int r;
1575
1576         BUG_ON(len > ghc->len);
1577
1578         if (slots->generation != ghc->generation)
1579                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1580
1581         if (unlikely(!ghc->memslot))
1582                 return kvm_write_guest(kvm, ghc->gpa, data, len);
1583
1584         if (kvm_is_error_hva(ghc->hva))
1585                 return -EFAULT;
1586
1587         r = __copy_to_user((void __user *)ghc->hva, data, len);
1588         if (r)
1589                 return -EFAULT;
1590         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1591
1592         return 0;
1593 }
1594 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1595
1596 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1597                            void *data, unsigned long len)
1598 {
1599         struct kvm_memslots *slots = kvm_memslots(kvm);
1600         int r;
1601
1602         BUG_ON(len > ghc->len);
1603
1604         if (slots->generation != ghc->generation)
1605                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1606
1607         if (unlikely(!ghc->memslot))
1608                 return kvm_read_guest(kvm, ghc->gpa, data, len);
1609
1610         if (kvm_is_error_hva(ghc->hva))
1611                 return -EFAULT;
1612
1613         r = __copy_from_user(data, (void __user *)ghc->hva, len);
1614         if (r)
1615                 return -EFAULT;
1616
1617         return 0;
1618 }
1619 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1620
1621 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1622 {
1623         const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
1624
1625         return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
1626 }
1627 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1628
1629 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1630 {
1631         gfn_t gfn = gpa >> PAGE_SHIFT;
1632         int seg;
1633         int offset = offset_in_page(gpa);
1634         int ret;
1635
1636         while ((seg = next_segment(len, offset)) != 0) {
1637                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1638                 if (ret < 0)
1639                         return ret;
1640                 offset = 0;
1641                 len -= seg;
1642                 ++gfn;
1643         }
1644         return 0;
1645 }
1646 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1647
1648 static void mark_page_dirty_in_slot(struct kvm *kvm,
1649                                     struct kvm_memory_slot *memslot,
1650                                     gfn_t gfn)
1651 {
1652         if (memslot && memslot->dirty_bitmap) {
1653                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1654
1655                 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1656         }
1657 }
1658
1659 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1660 {
1661         struct kvm_memory_slot *memslot;
1662
1663         memslot = gfn_to_memslot(kvm, gfn);
1664         mark_page_dirty_in_slot(kvm, memslot, gfn);
1665 }
1666 EXPORT_SYMBOL_GPL(mark_page_dirty);
1667
1668 /*
1669  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1670  */
1671 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1672 {
1673         DEFINE_WAIT(wait);
1674
1675         for (;;) {
1676                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1677
1678                 if (kvm_arch_vcpu_runnable(vcpu)) {
1679                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1680                         break;
1681                 }
1682                 if (kvm_cpu_has_pending_timer(vcpu))
1683                         break;
1684                 if (signal_pending(current))
1685                         break;
1686
1687                 schedule();
1688         }
1689
1690         finish_wait(&vcpu->wq, &wait);
1691 }
1692 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
1693
1694 #ifndef CONFIG_S390
1695 /*
1696  * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1697  */
1698 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1699 {
1700         int me;
1701         int cpu = vcpu->cpu;
1702         wait_queue_head_t *wqp;
1703
1704         wqp = kvm_arch_vcpu_wq(vcpu);
1705         if (waitqueue_active(wqp)) {
1706                 wake_up_interruptible(wqp);
1707                 ++vcpu->stat.halt_wakeup;
1708         }
1709
1710         me = get_cpu();
1711         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1712                 if (kvm_arch_vcpu_should_kick(vcpu))
1713                         smp_send_reschedule(cpu);
1714         put_cpu();
1715 }
1716 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
1717 #endif /* !CONFIG_S390 */
1718
1719 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
1720 {
1721         struct pid *pid;
1722         struct task_struct *task = NULL;
1723         int ret = 0;
1724
1725         rcu_read_lock();
1726         pid = rcu_dereference(target->pid);
1727         if (pid)
1728                 task = get_pid_task(target->pid, PIDTYPE_PID);
1729         rcu_read_unlock();
1730         if (!task)
1731                 return ret;
1732         if (task->flags & PF_VCPU) {
1733                 put_task_struct(task);
1734                 return ret;
1735         }
1736         ret = yield_to(task, 1);
1737         put_task_struct(task);
1738
1739         return ret;
1740 }
1741 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1742
1743 /*
1744  * Helper that checks whether a VCPU is eligible for directed yield.
1745  * Most eligible candidate to yield is decided by following heuristics:
1746  *
1747  *  (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1748  *  (preempted lock holder), indicated by @in_spin_loop.
1749  *  Set at the beiginning and cleared at the end of interception/PLE handler.
1750  *
1751  *  (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1752  *  chance last time (mostly it has become eligible now since we have probably
1753  *  yielded to lockholder in last iteration. This is done by toggling
1754  *  @dy_eligible each time a VCPU checked for eligibility.)
1755  *
1756  *  Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1757  *  to preempted lock-holder could result in wrong VCPU selection and CPU
1758  *  burning. Giving priority for a potential lock-holder increases lock
1759  *  progress.
1760  *
1761  *  Since algorithm is based on heuristics, accessing another VCPU data without
1762  *  locking does not harm. It may result in trying to yield to  same VCPU, fail
1763  *  and continue with next VCPU and so on.
1764  */
1765 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1766 {
1767 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1768         bool eligible;
1769
1770         eligible = !vcpu->spin_loop.in_spin_loop ||
1771                         (vcpu->spin_loop.in_spin_loop &&
1772                          vcpu->spin_loop.dy_eligible);
1773
1774         if (vcpu->spin_loop.in_spin_loop)
1775                 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1776
1777         return eligible;
1778 #else
1779         return true;
1780 #endif
1781 }
1782
1783 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1784 {
1785         struct kvm *kvm = me->kvm;
1786         struct kvm_vcpu *vcpu;
1787         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1788         int yielded = 0;
1789         int try = 3;
1790         int pass;
1791         int i;
1792
1793         kvm_vcpu_set_in_spin_loop(me, true);
1794         /*
1795          * We boost the priority of a VCPU that is runnable but not
1796          * currently running, because it got preempted by something
1797          * else and called schedule in __vcpu_run.  Hopefully that
1798          * VCPU is holding the lock that we need and will release it.
1799          * We approximate round-robin by starting at the last boosted VCPU.
1800          */
1801         for (pass = 0; pass < 2 && !yielded && try; pass++) {
1802                 kvm_for_each_vcpu(i, vcpu, kvm) {
1803                         if (!pass && i <= last_boosted_vcpu) {
1804                                 i = last_boosted_vcpu;
1805                                 continue;
1806                         } else if (pass && i > last_boosted_vcpu)
1807                                 break;
1808                         if (!ACCESS_ONCE(vcpu->preempted))
1809                                 continue;
1810                         if (vcpu == me)
1811                                 continue;
1812                         if (waitqueue_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
1813                                 continue;
1814                         if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1815                                 continue;
1816
1817                         yielded = kvm_vcpu_yield_to(vcpu);
1818                         if (yielded > 0) {
1819                                 kvm->last_boosted_vcpu = i;
1820                                 break;
1821                         } else if (yielded < 0) {
1822                                 try--;
1823                                 if (!try)
1824                                         break;
1825                         }
1826                 }
1827         }
1828         kvm_vcpu_set_in_spin_loop(me, false);
1829
1830         /* Ensure vcpu is not eligible during next spinloop */
1831         kvm_vcpu_set_dy_eligible(me, false);
1832 }
1833 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1834
1835 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1836 {
1837         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1838         struct page *page;
1839
1840         if (vmf->pgoff == 0)
1841                 page = virt_to_page(vcpu->run);
1842 #ifdef CONFIG_X86
1843         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1844                 page = virt_to_page(vcpu->arch.pio_data);
1845 #endif
1846 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1847         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1848                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1849 #endif
1850         else
1851                 return kvm_arch_vcpu_fault(vcpu, vmf);
1852         get_page(page);
1853         vmf->page = page;
1854         return 0;
1855 }
1856
1857 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1858         .fault = kvm_vcpu_fault,
1859 };
1860
1861 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1862 {
1863         vma->vm_ops = &kvm_vcpu_vm_ops;
1864         return 0;
1865 }
1866
1867 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1868 {
1869         struct kvm_vcpu *vcpu = filp->private_data;
1870
1871         kvm_put_kvm(vcpu->kvm);
1872         return 0;
1873 }
1874
1875 static struct file_operations kvm_vcpu_fops = {
1876         .release        = kvm_vcpu_release,
1877         .unlocked_ioctl = kvm_vcpu_ioctl,
1878 #ifdef CONFIG_COMPAT
1879         .compat_ioctl   = kvm_vcpu_compat_ioctl,
1880 #endif
1881         .mmap           = kvm_vcpu_mmap,
1882         .llseek         = noop_llseek,
1883 };
1884
1885 /*
1886  * Allocates an inode for the vcpu.
1887  */
1888 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1889 {
1890         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
1891 }
1892
1893 /*
1894  * Creates some virtual cpus.  Good luck creating more than one.
1895  */
1896 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1897 {
1898         int r;
1899         struct kvm_vcpu *vcpu, *v;
1900
1901         if (id >= KVM_MAX_VCPUS)
1902                 return -EINVAL;
1903
1904         vcpu = kvm_arch_vcpu_create(kvm, id);
1905         if (IS_ERR(vcpu))
1906                 return PTR_ERR(vcpu);
1907
1908         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1909
1910         r = kvm_arch_vcpu_setup(vcpu);
1911         if (r)
1912                 goto vcpu_destroy;
1913
1914         mutex_lock(&kvm->lock);
1915         if (!kvm_vcpu_compatible(vcpu)) {
1916                 r = -EINVAL;
1917                 goto unlock_vcpu_destroy;
1918         }
1919         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1920                 r = -EINVAL;
1921                 goto unlock_vcpu_destroy;
1922         }
1923
1924         kvm_for_each_vcpu(r, v, kvm)
1925                 if (v->vcpu_id == id) {
1926                         r = -EEXIST;
1927                         goto unlock_vcpu_destroy;
1928                 }
1929
1930         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1931
1932         /* Now it's all set up, let userspace reach it */
1933         kvm_get_kvm(kvm);
1934         r = create_vcpu_fd(vcpu);
1935         if (r < 0) {
1936                 kvm_put_kvm(kvm);
1937                 goto unlock_vcpu_destroy;
1938         }
1939
1940         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1941         smp_wmb();
1942         atomic_inc(&kvm->online_vcpus);
1943
1944         mutex_unlock(&kvm->lock);
1945         kvm_arch_vcpu_postcreate(vcpu);
1946         return r;
1947
1948 unlock_vcpu_destroy:
1949         mutex_unlock(&kvm->lock);
1950 vcpu_destroy:
1951         kvm_arch_vcpu_destroy(vcpu);
1952         return r;
1953 }
1954
1955 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1956 {
1957         if (sigset) {
1958                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1959                 vcpu->sigset_active = 1;
1960                 vcpu->sigset = *sigset;
1961         } else
1962                 vcpu->sigset_active = 0;
1963         return 0;
1964 }
1965
1966 static long kvm_vcpu_ioctl(struct file *filp,
1967                            unsigned int ioctl, unsigned long arg)
1968 {
1969         struct kvm_vcpu *vcpu = filp->private_data;
1970         void __user *argp = (void __user *)arg;
1971         int r;
1972         struct kvm_fpu *fpu = NULL;
1973         struct kvm_sregs *kvm_sregs = NULL;
1974
1975         if (vcpu->kvm->mm != current->mm)
1976                 return -EIO;
1977
1978 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
1979         /*
1980          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1981          * so vcpu_load() would break it.
1982          */
1983         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1984                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1985 #endif
1986
1987
1988         r = vcpu_load(vcpu);
1989         if (r)
1990                 return r;
1991         switch (ioctl) {
1992         case KVM_RUN:
1993                 r = -EINVAL;
1994                 if (arg)
1995                         goto out;
1996                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1997                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1998                 break;
1999         case KVM_GET_REGS: {
2000                 struct kvm_regs *kvm_regs;
2001
2002                 r = -ENOMEM;
2003                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2004                 if (!kvm_regs)
2005                         goto out;
2006                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2007                 if (r)
2008                         goto out_free1;
2009                 r = -EFAULT;
2010                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2011                         goto out_free1;
2012                 r = 0;
2013 out_free1:
2014                 kfree(kvm_regs);
2015                 break;
2016         }
2017         case KVM_SET_REGS: {
2018                 struct kvm_regs *kvm_regs;
2019
2020                 r = -ENOMEM;
2021                 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2022                 if (IS_ERR(kvm_regs)) {
2023                         r = PTR_ERR(kvm_regs);
2024                         goto out;
2025                 }
2026                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2027                 kfree(kvm_regs);
2028                 break;
2029         }
2030         case KVM_GET_SREGS: {
2031                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2032                 r = -ENOMEM;
2033                 if (!kvm_sregs)
2034                         goto out;
2035                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2036                 if (r)
2037                         goto out;
2038                 r = -EFAULT;
2039                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2040                         goto out;
2041                 r = 0;
2042                 break;
2043         }
2044         case KVM_SET_SREGS: {
2045                 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2046                 if (IS_ERR(kvm_sregs)) {
2047                         r = PTR_ERR(kvm_sregs);
2048                         kvm_sregs = NULL;
2049                         goto out;
2050                 }
2051                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2052                 break;
2053         }
2054         case KVM_GET_MP_STATE: {
2055                 struct kvm_mp_state mp_state;
2056
2057                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2058                 if (r)
2059                         goto out;
2060                 r = -EFAULT;
2061                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
2062                         goto out;
2063                 r = 0;
2064                 break;
2065         }
2066         case KVM_SET_MP_STATE: {
2067                 struct kvm_mp_state mp_state;
2068
2069                 r = -EFAULT;
2070                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
2071                         goto out;
2072                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2073                 break;
2074         }
2075         case KVM_TRANSLATE: {
2076                 struct kvm_translation tr;
2077
2078                 r = -EFAULT;
2079                 if (copy_from_user(&tr, argp, sizeof tr))
2080                         goto out;
2081                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2082                 if (r)
2083                         goto out;
2084                 r = -EFAULT;
2085                 if (copy_to_user(argp, &tr, sizeof tr))
2086                         goto out;
2087                 r = 0;
2088                 break;
2089         }
2090         case KVM_SET_GUEST_DEBUG: {
2091                 struct kvm_guest_debug dbg;
2092
2093                 r = -EFAULT;
2094                 if (copy_from_user(&dbg, argp, sizeof dbg))
2095                         goto out;
2096                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2097                 break;
2098         }
2099         case KVM_SET_SIGNAL_MASK: {
2100                 struct kvm_signal_mask __user *sigmask_arg = argp;
2101                 struct kvm_signal_mask kvm_sigmask;
2102                 sigset_t sigset, *p;
2103
2104                 p = NULL;
2105                 if (argp) {
2106                         r = -EFAULT;
2107                         if (copy_from_user(&kvm_sigmask, argp,
2108                                            sizeof kvm_sigmask))
2109                                 goto out;
2110                         r = -EINVAL;
2111                         if (kvm_sigmask.len != sizeof sigset)
2112                                 goto out;
2113                         r = -EFAULT;
2114                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2115                                            sizeof sigset))
2116                                 goto out;
2117                         p = &sigset;
2118                 }
2119                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2120                 break;
2121         }
2122         case KVM_GET_FPU: {
2123                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2124                 r = -ENOMEM;
2125                 if (!fpu)
2126                         goto out;
2127                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2128                 if (r)
2129                         goto out;
2130                 r = -EFAULT;
2131                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2132                         goto out;
2133                 r = 0;
2134                 break;
2135         }
2136         case KVM_SET_FPU: {
2137                 fpu = memdup_user(argp, sizeof(*fpu));
2138                 if (IS_ERR(fpu)) {
2139                         r = PTR_ERR(fpu);
2140                         fpu = NULL;
2141                         goto out;
2142                 }
2143                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2144                 break;
2145         }
2146         default:
2147                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2148         }
2149 out:
2150         vcpu_put(vcpu);
2151         kfree(fpu);
2152         kfree(kvm_sregs);
2153         return r;
2154 }
2155
2156 #ifdef CONFIG_COMPAT
2157 static long kvm_vcpu_compat_ioctl(struct file *filp,
2158                                   unsigned int ioctl, unsigned long arg)
2159 {
2160         struct kvm_vcpu *vcpu = filp->private_data;
2161         void __user *argp = compat_ptr(arg);
2162         int r;
2163
2164         if (vcpu->kvm->mm != current->mm)
2165                 return -EIO;
2166
2167         switch (ioctl) {
2168         case KVM_SET_SIGNAL_MASK: {
2169                 struct kvm_signal_mask __user *sigmask_arg = argp;
2170                 struct kvm_signal_mask kvm_sigmask;
2171                 compat_sigset_t csigset;
2172                 sigset_t sigset;
2173
2174                 if (argp) {
2175                         r = -EFAULT;
2176                         if (copy_from_user(&kvm_sigmask, argp,
2177                                            sizeof kvm_sigmask))
2178                                 goto out;
2179                         r = -EINVAL;
2180                         if (kvm_sigmask.len != sizeof csigset)
2181                                 goto out;
2182                         r = -EFAULT;
2183                         if (copy_from_user(&csigset, sigmask_arg->sigset,
2184                                            sizeof csigset))
2185                                 goto out;
2186                         sigset_from_compat(&sigset, &csigset);
2187                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2188                 } else
2189                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2190                 break;
2191         }
2192         default:
2193                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2194         }
2195
2196 out:
2197         return r;
2198 }
2199 #endif
2200
2201 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2202                                  int (*accessor)(struct kvm_device *dev,
2203                                                  struct kvm_device_attr *attr),
2204                                  unsigned long arg)
2205 {
2206         struct kvm_device_attr attr;
2207
2208         if (!accessor)
2209                 return -EPERM;
2210
2211         if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2212                 return -EFAULT;
2213
2214         return accessor(dev, &attr);
2215 }
2216
2217 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2218                              unsigned long arg)
2219 {
2220         struct kvm_device *dev = filp->private_data;
2221
2222         switch (ioctl) {
2223         case KVM_SET_DEVICE_ATTR:
2224                 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2225         case KVM_GET_DEVICE_ATTR:
2226                 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2227         case KVM_HAS_DEVICE_ATTR:
2228                 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2229         default:
2230                 if (dev->ops->ioctl)
2231                         return dev->ops->ioctl(dev, ioctl, arg);
2232
2233                 return -ENOTTY;
2234         }
2235 }
2236
2237 static int kvm_device_release(struct inode *inode, struct file *filp)
2238 {
2239         struct kvm_device *dev = filp->private_data;
2240         struct kvm *kvm = dev->kvm;
2241
2242         kvm_put_kvm(kvm);
2243         return 0;
2244 }
2245
2246 static const struct file_operations kvm_device_fops = {
2247         .unlocked_ioctl = kvm_device_ioctl,
2248 #ifdef CONFIG_COMPAT
2249         .compat_ioctl = kvm_device_ioctl,
2250 #endif
2251         .release = kvm_device_release,
2252 };
2253
2254 struct kvm_device *kvm_device_from_filp(struct file *filp)
2255 {
2256         if (filp->f_op != &kvm_device_fops)
2257                 return NULL;
2258
2259         return filp->private_data;
2260 }
2261
2262 static int kvm_ioctl_create_device(struct kvm *kvm,
2263                                    struct kvm_create_device *cd)
2264 {
2265         struct kvm_device_ops *ops = NULL;
2266         struct kvm_device *dev;
2267         bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2268         int ret;
2269
2270         switch (cd->type) {
2271 #ifdef CONFIG_KVM_MPIC
2272         case KVM_DEV_TYPE_FSL_MPIC_20:
2273         case KVM_DEV_TYPE_FSL_MPIC_42:
2274                 ops = &kvm_mpic_ops;
2275                 break;
2276 #endif
2277 #ifdef CONFIG_KVM_XICS
2278         case KVM_DEV_TYPE_XICS:
2279                 ops = &kvm_xics_ops;
2280                 break;
2281 #endif
2282 #ifdef CONFIG_KVM_VFIO
2283         case KVM_DEV_TYPE_VFIO:
2284                 ops = &kvm_vfio_ops;
2285                 break;
2286 #endif
2287 #ifdef CONFIG_KVM_ARM_VGIC
2288         case KVM_DEV_TYPE_ARM_VGIC_V2:
2289                 ops = &kvm_arm_vgic_v2_ops;
2290                 break;
2291 #endif
2292 #ifdef CONFIG_S390
2293         case KVM_DEV_TYPE_FLIC:
2294                 ops = &kvm_flic_ops;
2295                 break;
2296 #endif
2297         default:
2298                 return -ENODEV;
2299         }
2300
2301         if (test)
2302                 return 0;
2303
2304         dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2305         if (!dev)
2306                 return -ENOMEM;
2307
2308         dev->ops = ops;
2309         dev->kvm = kvm;
2310
2311         ret = ops->create(dev, cd->type);
2312         if (ret < 0) {
2313                 kfree(dev);
2314                 return ret;
2315         }
2316
2317         ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2318         if (ret < 0) {
2319                 ops->destroy(dev);
2320                 return ret;
2321         }
2322
2323         list_add(&dev->vm_node, &kvm->devices);
2324         kvm_get_kvm(kvm);
2325         cd->fd = ret;
2326         return 0;
2327 }
2328
2329 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2330 {
2331         switch (arg) {
2332         case KVM_CAP_USER_MEMORY:
2333         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2334         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2335 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2336         case KVM_CAP_SET_BOOT_CPU_ID:
2337 #endif
2338         case KVM_CAP_INTERNAL_ERROR_DATA:
2339 #ifdef CONFIG_HAVE_KVM_MSI
2340         case KVM_CAP_SIGNAL_MSI:
2341 #endif
2342 #ifdef CONFIG_HAVE_KVM_IRQFD
2343         case KVM_CAP_IRQFD_RESAMPLE:
2344 #endif
2345         case KVM_CAP_CHECK_EXTENSION_VM:
2346                 return 1;
2347 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2348         case KVM_CAP_IRQ_ROUTING:
2349                 return KVM_MAX_IRQ_ROUTES;
2350 #endif
2351         default:
2352                 break;
2353         }
2354         return kvm_vm_ioctl_check_extension(kvm, arg);
2355 }
2356
2357 static long kvm_vm_ioctl(struct file *filp,
2358                            unsigned int ioctl, unsigned long arg)
2359 {
2360         struct kvm *kvm = filp->private_data;
2361         void __user *argp = (void __user *)arg;
2362         int r;
2363
2364         if (kvm->mm != current->mm)
2365                 return -EIO;
2366         switch (ioctl) {
2367         case KVM_CREATE_VCPU:
2368                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2369                 break;
2370         case KVM_SET_USER_MEMORY_REGION: {
2371                 struct kvm_userspace_memory_region kvm_userspace_mem;
2372
2373                 r = -EFAULT;
2374                 if (copy_from_user(&kvm_userspace_mem, argp,
2375                                                 sizeof kvm_userspace_mem))
2376                         goto out;
2377
2378                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2379                 break;
2380         }
2381         case KVM_GET_DIRTY_LOG: {
2382                 struct kvm_dirty_log log;
2383
2384                 r = -EFAULT;
2385                 if (copy_from_user(&log, argp, sizeof log))
2386                         goto out;
2387                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2388                 break;
2389         }
2390 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2391         case KVM_REGISTER_COALESCED_MMIO: {
2392                 struct kvm_coalesced_mmio_zone zone;
2393                 r = -EFAULT;
2394                 if (copy_from_user(&zone, argp, sizeof zone))
2395                         goto out;
2396                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2397                 break;
2398         }
2399         case KVM_UNREGISTER_COALESCED_MMIO: {
2400                 struct kvm_coalesced_mmio_zone zone;
2401                 r = -EFAULT;
2402                 if (copy_from_user(&zone, argp, sizeof zone))
2403                         goto out;
2404                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2405                 break;
2406         }
2407 #endif
2408         case KVM_IRQFD: {
2409                 struct kvm_irqfd data;
2410
2411                 r = -EFAULT;
2412                 if (copy_from_user(&data, argp, sizeof data))
2413                         goto out;
2414                 r = kvm_irqfd(kvm, &data);
2415                 break;
2416         }
2417         case KVM_IOEVENTFD: {
2418                 struct kvm_ioeventfd data;
2419
2420                 r = -EFAULT;
2421                 if (copy_from_user(&data, argp, sizeof data))
2422                         goto out;
2423                 r = kvm_ioeventfd(kvm, &data);
2424                 break;
2425         }
2426 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2427         case KVM_SET_BOOT_CPU_ID:
2428                 r = 0;
2429                 mutex_lock(&kvm->lock);
2430                 if (atomic_read(&kvm->online_vcpus) != 0)
2431                         r = -EBUSY;
2432                 else
2433                         kvm->bsp_vcpu_id = arg;
2434                 mutex_unlock(&kvm->lock);
2435                 break;
2436 #endif
2437 #ifdef CONFIG_HAVE_KVM_MSI
2438         case KVM_SIGNAL_MSI: {
2439                 struct kvm_msi msi;
2440
2441                 r = -EFAULT;
2442                 if (copy_from_user(&msi, argp, sizeof msi))
2443                         goto out;
2444                 r = kvm_send_userspace_msi(kvm, &msi);
2445                 break;
2446         }
2447 #endif
2448 #ifdef __KVM_HAVE_IRQ_LINE
2449         case KVM_IRQ_LINE_STATUS:
2450         case KVM_IRQ_LINE: {
2451                 struct kvm_irq_level irq_event;
2452
2453                 r = -EFAULT;
2454                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2455                         goto out;
2456
2457                 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
2458                                         ioctl == KVM_IRQ_LINE_STATUS);
2459                 if (r)
2460                         goto out;
2461
2462                 r = -EFAULT;
2463                 if (ioctl == KVM_IRQ_LINE_STATUS) {
2464                         if (copy_to_user(argp, &irq_event, sizeof irq_event))
2465                                 goto out;
2466                 }
2467
2468                 r = 0;
2469                 break;
2470         }
2471 #endif
2472 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2473         case KVM_SET_GSI_ROUTING: {
2474                 struct kvm_irq_routing routing;
2475                 struct kvm_irq_routing __user *urouting;
2476                 struct kvm_irq_routing_entry *entries;
2477
2478                 r = -EFAULT;
2479                 if (copy_from_user(&routing, argp, sizeof(routing)))
2480                         goto out;
2481                 r = -EINVAL;
2482                 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2483                         goto out;
2484                 if (routing.flags)
2485                         goto out;
2486                 r = -ENOMEM;
2487                 entries = vmalloc(routing.nr * sizeof(*entries));
2488                 if (!entries)
2489                         goto out;
2490                 r = -EFAULT;
2491                 urouting = argp;
2492                 if (copy_from_user(entries, urouting->entries,
2493                                    routing.nr * sizeof(*entries)))
2494                         goto out_free_irq_routing;
2495                 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2496                                         routing.flags);
2497         out_free_irq_routing:
2498                 vfree(entries);
2499                 break;
2500         }
2501 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
2502         case KVM_CREATE_DEVICE: {
2503                 struct kvm_create_device cd;
2504
2505                 r = -EFAULT;
2506                 if (copy_from_user(&cd, argp, sizeof(cd)))
2507                         goto out;
2508
2509                 r = kvm_ioctl_create_device(kvm, &cd);
2510                 if (r)
2511                         goto out;
2512
2513                 r = -EFAULT;
2514                 if (copy_to_user(argp, &cd, sizeof(cd)))
2515                         goto out;
2516
2517                 r = 0;
2518                 break;
2519         }
2520         case KVM_CHECK_EXTENSION:
2521                 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
2522                 break;
2523         default:
2524                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2525                 if (r == -ENOTTY)
2526                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2527         }
2528 out:
2529         return r;
2530 }
2531
2532 #ifdef CONFIG_COMPAT
2533 struct compat_kvm_dirty_log {
2534         __u32 slot;
2535         __u32 padding1;
2536         union {
2537                 compat_uptr_t dirty_bitmap; /* one bit per page */
2538                 __u64 padding2;
2539         };
2540 };
2541
2542 static long kvm_vm_compat_ioctl(struct file *filp,
2543                            unsigned int ioctl, unsigned long arg)
2544 {
2545         struct kvm *kvm = filp->private_data;
2546         int r;
2547
2548         if (kvm->mm != current->mm)
2549                 return -EIO;
2550         switch (ioctl) {
2551         case KVM_GET_DIRTY_LOG: {
2552                 struct compat_kvm_dirty_log compat_log;
2553                 struct kvm_dirty_log log;
2554
2555                 r = -EFAULT;
2556                 if (copy_from_user(&compat_log, (void __user *)arg,
2557                                    sizeof(compat_log)))
2558                         goto out;
2559                 log.slot         = compat_log.slot;
2560                 log.padding1     = compat_log.padding1;
2561                 log.padding2     = compat_log.padding2;
2562                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2563
2564                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2565                 break;
2566         }
2567         default:
2568                 r = kvm_vm_ioctl(filp, ioctl, arg);
2569         }
2570
2571 out:
2572         return r;
2573 }
2574 #endif
2575
2576 static struct file_operations kvm_vm_fops = {
2577         .release        = kvm_vm_release,
2578         .unlocked_ioctl = kvm_vm_ioctl,
2579 #ifdef CONFIG_COMPAT
2580         .compat_ioctl   = kvm_vm_compat_ioctl,
2581 #endif
2582         .llseek         = noop_llseek,
2583 };
2584
2585 static int kvm_dev_ioctl_create_vm(unsigned long type)
2586 {
2587         int r;
2588         struct kvm *kvm;
2589
2590         kvm = kvm_create_vm(type);
2591         if (IS_ERR(kvm))
2592                 return PTR_ERR(kvm);
2593 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2594         r = kvm_coalesced_mmio_init(kvm);
2595         if (r < 0) {
2596                 kvm_put_kvm(kvm);
2597                 return r;
2598         }
2599 #endif
2600         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC);
2601         if (r < 0)
2602                 kvm_put_kvm(kvm);
2603
2604         return r;
2605 }
2606
2607 static long kvm_dev_ioctl(struct file *filp,
2608                           unsigned int ioctl, unsigned long arg)
2609 {
2610         long r = -EINVAL;
2611
2612         switch (ioctl) {
2613         case KVM_GET_API_VERSION:
2614                 r = -EINVAL;
2615                 if (arg)
2616                         goto out;
2617                 r = KVM_API_VERSION;
2618                 break;
2619         case KVM_CREATE_VM:
2620                 r = kvm_dev_ioctl_create_vm(arg);
2621                 break;
2622         case KVM_CHECK_EXTENSION:
2623                 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
2624                 break;
2625         case KVM_GET_VCPU_MMAP_SIZE:
2626                 r = -EINVAL;
2627                 if (arg)
2628                         goto out;
2629                 r = PAGE_SIZE;     /* struct kvm_run */
2630 #ifdef CONFIG_X86
2631                 r += PAGE_SIZE;    /* pio data page */
2632 #endif
2633 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2634                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2635 #endif
2636                 break;
2637         case KVM_TRACE_ENABLE:
2638         case KVM_TRACE_PAUSE:
2639         case KVM_TRACE_DISABLE:
2640                 r = -EOPNOTSUPP;
2641                 break;
2642         default:
2643                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2644         }
2645 out:
2646         return r;
2647 }
2648
2649 static struct file_operations kvm_chardev_ops = {
2650         .unlocked_ioctl = kvm_dev_ioctl,
2651         .compat_ioctl   = kvm_dev_ioctl,
2652         .llseek         = noop_llseek,
2653 };
2654
2655 static struct miscdevice kvm_dev = {
2656         KVM_MINOR,
2657         "kvm",
2658         &kvm_chardev_ops,
2659 };
2660
2661 static void hardware_enable_nolock(void *junk)
2662 {
2663         int cpu = raw_smp_processor_id();
2664         int r;
2665
2666         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2667                 return;
2668
2669         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2670
2671         r = kvm_arch_hardware_enable(NULL);
2672
2673         if (r) {
2674                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2675                 atomic_inc(&hardware_enable_failed);
2676                 printk(KERN_INFO "kvm: enabling virtualization on "
2677                                  "CPU%d failed\n", cpu);
2678         }
2679 }
2680
2681 static void hardware_enable(void)
2682 {
2683         raw_spin_lock(&kvm_count_lock);
2684         if (kvm_usage_count)
2685                 hardware_enable_nolock(NULL);
2686         raw_spin_unlock(&kvm_count_lock);
2687 }
2688
2689 static void hardware_disable_nolock(void *junk)
2690 {
2691         int cpu = raw_smp_processor_id();
2692
2693         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2694                 return;
2695         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2696         kvm_arch_hardware_disable(NULL);
2697 }
2698
2699 static void hardware_disable(void)
2700 {
2701         raw_spin_lock(&kvm_count_lock);
2702         if (kvm_usage_count)
2703                 hardware_disable_nolock(NULL);
2704         raw_spin_unlock(&kvm_count_lock);
2705 }
2706
2707 static void hardware_disable_all_nolock(void)
2708 {
2709         BUG_ON(!kvm_usage_count);
2710
2711         kvm_usage_count--;
2712         if (!kvm_usage_count)
2713                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2714 }
2715
2716 static void hardware_disable_all(void)
2717 {
2718         raw_spin_lock(&kvm_count_lock);
2719         hardware_disable_all_nolock();
2720         raw_spin_unlock(&kvm_count_lock);
2721 }
2722
2723 static int hardware_enable_all(void)
2724 {
2725         int r = 0;
2726
2727         raw_spin_lock(&kvm_count_lock);
2728
2729         kvm_usage_count++;
2730         if (kvm_usage_count == 1) {
2731                 atomic_set(&hardware_enable_failed, 0);
2732                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2733
2734                 if (atomic_read(&hardware_enable_failed)) {
2735                         hardware_disable_all_nolock();
2736                         r = -EBUSY;
2737                 }
2738         }
2739
2740         raw_spin_unlock(&kvm_count_lock);
2741
2742         return r;
2743 }
2744
2745 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2746                            void *v)
2747 {
2748         int cpu = (long)v;
2749
2750         val &= ~CPU_TASKS_FROZEN;
2751         switch (val) {
2752         case CPU_DYING:
2753                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2754                        cpu);
2755                 hardware_disable();
2756                 break;
2757         case CPU_STARTING:
2758                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2759                        cpu);
2760                 hardware_enable();
2761                 break;
2762         }
2763         return NOTIFY_OK;
2764 }
2765
2766 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2767                       void *v)
2768 {
2769         /*
2770          * Some (well, at least mine) BIOSes hang on reboot if
2771          * in vmx root mode.
2772          *
2773          * And Intel TXT required VMX off for all cpu when system shutdown.
2774          */
2775         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2776         kvm_rebooting = true;
2777         on_each_cpu(hardware_disable_nolock, NULL, 1);
2778         return NOTIFY_OK;
2779 }
2780
2781 static struct notifier_block kvm_reboot_notifier = {
2782         .notifier_call = kvm_reboot,
2783         .priority = 0,
2784 };
2785
2786 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2787 {
2788         int i;
2789
2790         for (i = 0; i < bus->dev_count; i++) {
2791                 struct kvm_io_device *pos = bus->range[i].dev;
2792
2793                 kvm_iodevice_destructor(pos);
2794         }
2795         kfree(bus);
2796 }
2797
2798 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
2799                                  const struct kvm_io_range *r2)
2800 {
2801         if (r1->addr < r2->addr)
2802                 return -1;
2803         if (r1->addr + r1->len > r2->addr + r2->len)
2804                 return 1;
2805         return 0;
2806 }
2807
2808 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2809 {
2810         return kvm_io_bus_cmp(p1, p2);
2811 }
2812
2813 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2814                           gpa_t addr, int len)
2815 {
2816         bus->range[bus->dev_count++] = (struct kvm_io_range) {
2817                 .addr = addr,
2818                 .len = len,
2819                 .dev = dev,
2820         };
2821
2822         sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2823                 kvm_io_bus_sort_cmp, NULL);
2824
2825         return 0;
2826 }
2827
2828 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2829                              gpa_t addr, int len)
2830 {
2831         struct kvm_io_range *range, key;
2832         int off;
2833
2834         key = (struct kvm_io_range) {
2835                 .addr = addr,
2836                 .len = len,
2837         };
2838
2839         range = bsearch(&key, bus->range, bus->dev_count,
2840                         sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2841         if (range == NULL)
2842                 return -ENOENT;
2843
2844         off = range - bus->range;
2845
2846         while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
2847                 off--;
2848
2849         return off;
2850 }
2851
2852 static int __kvm_io_bus_write(struct kvm_io_bus *bus,
2853                               struct kvm_io_range *range, const void *val)
2854 {
2855         int idx;
2856
2857         idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2858         if (idx < 0)
2859                 return -EOPNOTSUPP;
2860
2861         while (idx < bus->dev_count &&
2862                 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
2863                 if (!kvm_iodevice_write(bus->range[idx].dev, range->addr,
2864                                         range->len, val))
2865                         return idx;
2866                 idx++;
2867         }
2868
2869         return -EOPNOTSUPP;
2870 }
2871
2872 /* kvm_io_bus_write - called under kvm->slots_lock */
2873 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2874                      int len, const void *val)
2875 {
2876         struct kvm_io_bus *bus;
2877         struct kvm_io_range range;
2878         int r;
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         r = __kvm_io_bus_write(bus, &range, val);
2887         return r < 0 ? r : 0;
2888 }
2889
2890 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
2891 int kvm_io_bus_write_cookie(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2892                             int len, const void *val, long cookie)
2893 {
2894         struct kvm_io_bus *bus;
2895         struct kvm_io_range range;
2896
2897         range = (struct kvm_io_range) {
2898                 .addr = addr,
2899                 .len = len,
2900         };
2901
2902         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2903
2904         /* First try the device referenced by cookie. */
2905         if ((cookie >= 0) && (cookie < bus->dev_count) &&
2906             (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
2907                 if (!kvm_iodevice_write(bus->range[cookie].dev, addr, len,
2908                                         val))
2909                         return cookie;
2910
2911         /*
2912          * cookie contained garbage; fall back to search and return the
2913          * correct cookie value.
2914          */
2915         return __kvm_io_bus_write(bus, &range, val);
2916 }
2917
2918 static int __kvm_io_bus_read(struct kvm_io_bus *bus, struct kvm_io_range *range,
2919                              void *val)
2920 {
2921         int idx;
2922
2923         idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2924         if (idx < 0)
2925                 return -EOPNOTSUPP;
2926
2927         while (idx < bus->dev_count &&
2928                 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
2929                 if (!kvm_iodevice_read(bus->range[idx].dev, range->addr,
2930                                        range->len, val))
2931                         return idx;
2932                 idx++;
2933         }
2934
2935         return -EOPNOTSUPP;
2936 }
2937 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
2938
2939 /* kvm_io_bus_read - called under kvm->slots_lock */
2940 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2941                     int len, void *val)
2942 {
2943         struct kvm_io_bus *bus;
2944         struct kvm_io_range range;
2945         int r;
2946
2947         range = (struct kvm_io_range) {
2948                 .addr = addr,
2949                 .len = len,
2950         };
2951
2952         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2953         r = __kvm_io_bus_read(bus, &range, val);
2954         return r < 0 ? r : 0;
2955 }
2956
2957
2958 /* Caller must hold slots_lock. */
2959 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2960                             int len, struct kvm_io_device *dev)
2961 {
2962         struct kvm_io_bus *new_bus, *bus;
2963
2964         bus = kvm->buses[bus_idx];
2965         /* exclude ioeventfd which is limited by maximum fd */
2966         if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
2967                 return -ENOSPC;
2968
2969         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
2970                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2971         if (!new_bus)
2972                 return -ENOMEM;
2973         memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
2974                sizeof(struct kvm_io_range)));
2975         kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2976         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2977         synchronize_srcu_expedited(&kvm->srcu);
2978         kfree(bus);
2979
2980         return 0;
2981 }
2982
2983 /* Caller must hold slots_lock. */
2984 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2985                               struct kvm_io_device *dev)
2986 {
2987         int i, r;
2988         struct kvm_io_bus *new_bus, *bus;
2989
2990         bus = kvm->buses[bus_idx];
2991         r = -ENOENT;
2992         for (i = 0; i < bus->dev_count; i++)
2993                 if (bus->range[i].dev == dev) {
2994                         r = 0;
2995                         break;
2996                 }
2997
2998         if (r)
2999                 return r;
3000
3001         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3002                           sizeof(struct kvm_io_range)), GFP_KERNEL);
3003         if (!new_bus)
3004                 return -ENOMEM;
3005
3006         memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3007         new_bus->dev_count--;
3008         memcpy(new_bus->range + i, bus->range + i + 1,
3009                (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3010
3011         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3012         synchronize_srcu_expedited(&kvm->srcu);
3013         kfree(bus);
3014         return r;
3015 }
3016
3017 static struct notifier_block kvm_cpu_notifier = {
3018         .notifier_call = kvm_cpu_hotplug,
3019 };
3020
3021 static int vm_stat_get(void *_offset, u64 *val)
3022 {
3023         unsigned offset = (long)_offset;
3024         struct kvm *kvm;
3025
3026         *val = 0;
3027         spin_lock(&kvm_lock);
3028         list_for_each_entry(kvm, &vm_list, vm_list)
3029                 *val += *(u32 *)((void *)kvm + offset);
3030         spin_unlock(&kvm_lock);
3031         return 0;
3032 }
3033
3034 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
3035
3036 static int vcpu_stat_get(void *_offset, u64 *val)
3037 {
3038         unsigned offset = (long)_offset;
3039         struct kvm *kvm;
3040         struct kvm_vcpu *vcpu;
3041         int i;
3042
3043         *val = 0;
3044         spin_lock(&kvm_lock);
3045         list_for_each_entry(kvm, &vm_list, vm_list)
3046                 kvm_for_each_vcpu(i, vcpu, kvm)
3047                         *val += *(u32 *)((void *)vcpu + offset);
3048
3049         spin_unlock(&kvm_lock);
3050         return 0;
3051 }
3052
3053 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
3054
3055 static const struct file_operations *stat_fops[] = {
3056         [KVM_STAT_VCPU] = &vcpu_stat_fops,
3057         [KVM_STAT_VM]   = &vm_stat_fops,
3058 };
3059
3060 static int kvm_init_debug(void)
3061 {
3062         int r = -EEXIST;
3063         struct kvm_stats_debugfs_item *p;
3064
3065         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3066         if (kvm_debugfs_dir == NULL)
3067                 goto out;
3068
3069         for (p = debugfs_entries; p->name; ++p) {
3070                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
3071                                                 (void *)(long)p->offset,
3072                                                 stat_fops[p->kind]);
3073                 if (p->dentry == NULL)
3074                         goto out_dir;
3075         }
3076
3077         return 0;
3078
3079 out_dir:
3080         debugfs_remove_recursive(kvm_debugfs_dir);
3081 out:
3082         return r;
3083 }
3084
3085 static void kvm_exit_debug(void)
3086 {
3087         struct kvm_stats_debugfs_item *p;
3088
3089         for (p = debugfs_entries; p->name; ++p)
3090                 debugfs_remove(p->dentry);
3091         debugfs_remove(kvm_debugfs_dir);
3092 }
3093
3094 static int kvm_suspend(void)
3095 {
3096         if (kvm_usage_count)
3097                 hardware_disable_nolock(NULL);
3098         return 0;
3099 }
3100
3101 static void kvm_resume(void)
3102 {
3103         if (kvm_usage_count) {
3104                 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3105                 hardware_enable_nolock(NULL);
3106         }
3107 }
3108
3109 static struct syscore_ops kvm_syscore_ops = {
3110         .suspend = kvm_suspend,
3111         .resume = kvm_resume,
3112 };
3113
3114 static inline
3115 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3116 {
3117         return container_of(pn, struct kvm_vcpu, preempt_notifier);
3118 }
3119
3120 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3121 {
3122         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3123         if (vcpu->preempted)
3124                 vcpu->preempted = false;
3125
3126         kvm_arch_vcpu_load(vcpu, cpu);
3127 }
3128
3129 static void kvm_sched_out(struct preempt_notifier *pn,
3130                           struct task_struct *next)
3131 {
3132         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3133
3134         if (current->state == TASK_RUNNING)
3135                 vcpu->preempted = true;
3136         kvm_arch_vcpu_put(vcpu);
3137 }
3138
3139 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3140                   struct module *module)
3141 {
3142         int r;
3143         int cpu;
3144
3145         r = kvm_arch_init(opaque);
3146         if (r)
3147                 goto out_fail;
3148
3149         /*
3150          * kvm_arch_init makes sure there's at most one caller
3151          * for architectures that support multiple implementations,
3152          * like intel and amd on x86.
3153          * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3154          * conflicts in case kvm is already setup for another implementation.
3155          */
3156         r = kvm_irqfd_init();
3157         if (r)
3158                 goto out_irqfd;
3159
3160         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
3161                 r = -ENOMEM;
3162                 goto out_free_0;
3163         }
3164
3165         r = kvm_arch_hardware_setup();
3166         if (r < 0)
3167                 goto out_free_0a;
3168
3169         for_each_online_cpu(cpu) {
3170                 smp_call_function_single(cpu,
3171                                 kvm_arch_check_processor_compat,
3172                                 &r, 1);
3173                 if (r < 0)
3174                         goto out_free_1;
3175         }
3176
3177         r = register_cpu_notifier(&kvm_cpu_notifier);
3178         if (r)
3179                 goto out_free_2;
3180         register_reboot_notifier(&kvm_reboot_notifier);
3181
3182         /* A kmem cache lets us meet the alignment requirements of fx_save. */
3183         if (!vcpu_align)
3184                 vcpu_align = __alignof__(struct kvm_vcpu);
3185         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
3186                                            0, NULL);
3187         if (!kvm_vcpu_cache) {
3188                 r = -ENOMEM;
3189                 goto out_free_3;
3190         }
3191
3192         r = kvm_async_pf_init();
3193         if (r)
3194                 goto out_free;
3195
3196         kvm_chardev_ops.owner = module;
3197         kvm_vm_fops.owner = module;
3198         kvm_vcpu_fops.owner = module;
3199
3200         r = misc_register(&kvm_dev);
3201         if (r) {
3202                 printk(KERN_ERR "kvm: misc device register failed\n");
3203                 goto out_unreg;
3204         }
3205
3206         register_syscore_ops(&kvm_syscore_ops);
3207
3208         kvm_preempt_ops.sched_in = kvm_sched_in;
3209         kvm_preempt_ops.sched_out = kvm_sched_out;
3210
3211         r = kvm_init_debug();
3212         if (r) {
3213                 printk(KERN_ERR "kvm: create debugfs files failed\n");
3214                 goto out_undebugfs;
3215         }
3216
3217         return 0;
3218
3219 out_undebugfs:
3220         unregister_syscore_ops(&kvm_syscore_ops);
3221         misc_deregister(&kvm_dev);
3222 out_unreg:
3223         kvm_async_pf_deinit();
3224 out_free:
3225         kmem_cache_destroy(kvm_vcpu_cache);
3226 out_free_3:
3227         unregister_reboot_notifier(&kvm_reboot_notifier);
3228         unregister_cpu_notifier(&kvm_cpu_notifier);
3229 out_free_2:
3230 out_free_1:
3231         kvm_arch_hardware_unsetup();
3232 out_free_0a:
3233         free_cpumask_var(cpus_hardware_enabled);
3234 out_free_0:
3235         kvm_irqfd_exit();
3236 out_irqfd:
3237         kvm_arch_exit();
3238 out_fail:
3239         return r;
3240 }
3241 EXPORT_SYMBOL_GPL(kvm_init);
3242
3243 void kvm_exit(void)
3244 {
3245         kvm_exit_debug();
3246         misc_deregister(&kvm_dev);
3247         kmem_cache_destroy(kvm_vcpu_cache);
3248         kvm_async_pf_deinit();
3249         unregister_syscore_ops(&kvm_syscore_ops);
3250         unregister_reboot_notifier(&kvm_reboot_notifier);
3251         unregister_cpu_notifier(&kvm_cpu_notifier);
3252         on_each_cpu(hardware_disable_nolock, NULL, 1);
3253         kvm_arch_hardware_unsetup();
3254         kvm_arch_exit();
3255         kvm_irqfd_exit();
3256         free_cpumask_var(cpus_hardware_enabled);
3257 }
3258 EXPORT_SYMBOL_GPL(kvm_exit);