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                 if (oldpid)
 133                         synchronize_rcu();
 134                 put_pid(oldpid);
 135         }
 136         cpu = get_cpu();
 137         preempt_notifier_register(&vcpu->preempt_notifier);
 138         kvm_arch_vcpu_load(vcpu, cpu);
 139         put_cpu();
 140         return 0;
 141 }
 142
 143 void vcpu_put(struct kvm_vcpu *vcpu)
 144 {
 145         preempt_disable();
 146         kvm_arch_vcpu_put(vcpu);
 147         preempt_notifier_unregister(&vcpu->preempt_notifier);
 148         preempt_enable();
 149         mutex_unlock(&vcpu->mutex);
 150 }
 151
 152 static void ack_flush(void *_completed)
 153 {
 154 }
 155
 156 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
 157 {
 158         int i, cpu, me;
 159         cpumask_var_t cpus;
 160         bool called = true;
 161         struct kvm_vcpu *vcpu;
 162
 163         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
 164
 165         me = get_cpu();
 166         kvm_for_each_vcpu(i, vcpu, kvm) {
 167                 kvm_make_request(req, vcpu);
 168                 cpu = vcpu->cpu;
 169
 170                 /* Set ->requests bit before we read ->mode */
 171                 smp_mb();
 172
 173                 if (cpus != NULL && cpu != -1 && cpu != me &&
 174                       kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
 175                         cpumask_set_cpu(cpu, cpus);
 176         }
 177         if (unlikely(cpus == NULL))
 178                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
 179         else if (!cpumask_empty(cpus))
 180                 smp_call_function_many(cpus, ack_flush, NULL, 1);
 181         else
 182                 called = false;
 183         put_cpu();
 184         free_cpumask_var(cpus);
 185         return called;
 186 }
 187
 188 void kvm_flush_remote_tlbs(struct kvm *kvm)
 189 {
 190         long dirty_count = kvm->tlbs_dirty;
 191
 192         smp_mb();
 193         if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
 194                 ++kvm->stat.remote_tlb_flush;
 195         cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
 196 }
 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 #ifndef CONFIG_S390
 651         unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
 652
 653         memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
 654         if (!memslot->dirty_bitmap)
 655                 return -ENOMEM;
 656
 657 #endif /* !CONFIG_S390 */
 658         return 0;
 659 }
 660
 661 static int cmp_memslot(const void *slot1, const void *slot2)
 662 {
 663         struct kvm_memory_slot *s1, *s2;
 664
 665         s1 = (struct kvm_memory_slot *)slot1;
 666         s2 = (struct kvm_memory_slot *)slot2;
 667
 668         if (s1->npages < s2->npages)
 669                 return 1;
 670         if (s1->npages > s2->npages)
 671                 return -1;
 672
 673         return 0;
 674 }
 675
 676 /*
 677  * Sort the memslots base on its size, so the larger slots
 678  * will get better fit.
 679  */
 680 static void sort_memslots(struct kvm_memslots *slots)
 681 {
 682         int i;
 683
 684         sort(slots->memslots, KVM_MEM_SLOTS_NUM,
 685               sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
 686
 687         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
 688                 slots->id_to_index[slots->memslots[i].id] = i;
 689 }
 690
 691 static void update_memslots(struct kvm_memslots *slots,
 692                             struct kvm_memory_slot *new,
 693                             u64 last_generation)
 694 {
 695         if (new) {
 696                 int id = new->id;
 697                 struct kvm_memory_slot *old = id_to_memslot(slots, id);
 698                 unsigned long npages = old->npages;
 699
 700                 *old = *new;
 701                 if (new->npages != npages)
 702                         sort_memslots(slots);
 703         }
 704
 705         slots->generation = last_generation + 1;
 706 }
 707
 708 static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
 709 {
 710         u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
 711
 712 #ifdef __KVM_HAVE_READONLY_MEM
 713         valid_flags |= KVM_MEM_READONLY;
 714 #endif
 715
 716         if (mem->flags & ~valid_flags)
 717                 return -EINVAL;
 718
 719         return 0;
 720 }
 721
 722 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
 723                 struct kvm_memslots *slots, struct kvm_memory_slot *new)
 724 {
 725         struct kvm_memslots *old_memslots = kvm->memslots;
 726
 727         update_memslots(slots, new, kvm->memslots->generation);
 728         rcu_assign_pointer(kvm->memslots, slots);
 729         synchronize_srcu_expedited(&kvm->srcu);
 730
 731         kvm_arch_memslots_updated(kvm);
 732
 733         return old_memslots;
 734 }
 735
 736 /*
 737  * Allocate some memory and give it an address in the guest physical address
 738  * space.
 739  *
 740  * Discontiguous memory is allowed, mostly for framebuffers.
 741  *
 742  * Must be called holding mmap_sem for write.
 743  */
 744 int __kvm_set_memory_region(struct kvm *kvm,
 745                             struct kvm_userspace_memory_region *mem)
 746 {
 747         int r;
 748         gfn_t base_gfn;
 749         unsigned long npages;
 750         struct kvm_memory_slot *slot;
 751         struct kvm_memory_slot old, new;
 752         struct kvm_memslots *slots = NULL, *old_memslots;
 753         enum kvm_mr_change change;
 754
 755         r = check_memory_region_flags(mem);
 756         if (r)
 757                 goto out;
 758
 759         r = -EINVAL;
 760         /* General sanity checks */
 761         if (mem->memory_size & (PAGE_SIZE - 1))
 762                 goto out;
 763         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 764                 goto out;
 765         /* We can read the guest memory with __xxx_user() later on. */
 766         if ((mem->slot < KVM_USER_MEM_SLOTS) &&
 767             ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
 768              !access_ok(VERIFY_WRITE,
 769                         (void __user *)(unsigned long)mem->userspace_addr,
 770                         mem->memory_size)))
 771                 goto out;
 772         if (mem->slot >= KVM_MEM_SLOTS_NUM)
 773                 goto out;
 774         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 775                 goto out;
 776
 777         slot = id_to_memslot(kvm->memslots, mem->slot);
 778         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 779         npages = mem->memory_size >> PAGE_SHIFT;
 780
 781         r = -EINVAL;
 782         if (npages > KVM_MEM_MAX_NR_PAGES)
 783                 goto out;
 784
 785         if (!npages)
 786                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 787
 788         new = old = *slot;
 789
 790         new.id = mem->slot;
 791         new.base_gfn = base_gfn;
 792         new.npages = npages;
 793         new.flags = mem->flags;
 794
 795         r = -EINVAL;
 796         if (npages) {
 797                 if (!old.npages)
 798                         change = KVM_MR_CREATE;
 799                 else { /* Modify an existing slot. */
 800                         if ((mem->userspace_addr != old.userspace_addr) ||
 801                             (npages != old.npages) ||
 802                             ((new.flags ^ old.flags) & KVM_MEM_READONLY))
 803                                 goto out;
 804
 805                         if (base_gfn != old.base_gfn)
 806                                 change = KVM_MR_MOVE;
 807                         else if (new.flags != old.flags)
 808                                 change = KVM_MR_FLAGS_ONLY;
 809                         else { /* Nothing to change. */
 810                                 r = 0;
 811                                 goto out;
 812                         }
 813                 }
 814         } else if (old.npages) {
 815                 change = KVM_MR_DELETE;
 816         } else /* Modify a non-existent slot: disallowed. */
 817                 goto out;
 818
 819         if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
 820                 /* Check for overlaps */
 821                 r = -EEXIST;
 822                 kvm_for_each_memslot(slot, kvm->memslots) {
 823                         if ((slot->id >= KVM_USER_MEM_SLOTS) ||
 824                             (slot->id == mem->slot))
 825                                 continue;
 826                         if (!((base_gfn + npages <= slot->base_gfn) ||
 827                               (base_gfn >= slot->base_gfn + slot->npages)))
 828                                 goto out;
 829                 }
 830         }
 831
 832         /* Free page dirty bitmap if unneeded */
 833         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 834                 new.dirty_bitmap = NULL;
 835
 836         r = -ENOMEM;
 837         if (change == KVM_MR_CREATE) {
 838                 new.userspace_addr = mem->userspace_addr;
 839
 840                 if (kvm_arch_create_memslot(kvm, &new, npages))
 841                         goto out_free;
 842         }
 843
 844         /* Allocate page dirty bitmap if needed */
 845         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 846                 if (kvm_create_dirty_bitmap(&new) < 0)
 847                         goto out_free;
 848         }
 849
 850         if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
 851                 r = -ENOMEM;
 852                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 853                                 GFP_KERNEL);
 854                 if (!slots)
 855                         goto out_free;
 856                 slot = id_to_memslot(slots, mem->slot);
 857                 slot->flags |= KVM_MEMSLOT_INVALID;
 858
 859                 old_memslots = install_new_memslots(kvm, slots, NULL);
 860
 861                 /* slot was deleted or moved, clear iommu mapping */
 862                 kvm_iommu_unmap_pages(kvm, &old);
 863                 /* From this point no new shadow pages pointing to a deleted,
 864                  * or moved, memslot will be created.
 865                  *
 866                  * validation of sp->gfn happens in:
 867                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
 868                  *      - kvm_is_visible_gfn (mmu_check_roots)
 869                  */
 870                 kvm_arch_flush_shadow_memslot(kvm, slot);
 871                 slots = old_memslots;
 872         }
 873
 874         r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
 875         if (r)
 876                 goto out_slots;
 877
 878         r = -ENOMEM;
 879         /*
 880          * We can re-use the old_memslots from above, the only difference
 881          * from the currently installed memslots is the invalid flag.  This
 882          * will get overwritten by update_memslots anyway.
 883          */
 884         if (!slots) {
 885                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 886                                 GFP_KERNEL);
 887                 if (!slots)
 888                         goto out_free;
 889         }
 890
 891         /* actual memory is freed via old in kvm_free_physmem_slot below */
 892         if (change == KVM_MR_DELETE) {
 893                 new.dirty_bitmap = NULL;
 894                 memset(&new.arch, 0, sizeof(new.arch));
 895         }
 896
 897         old_memslots = install_new_memslots(kvm, slots, &new);
 898
 899         kvm_arch_commit_memory_region(kvm, mem, &old, change);
 900
 901         kvm_free_physmem_slot(kvm, &old, &new);
 902         kfree(old_memslots);
 903
 904         /*
 905          * IOMMU mapping:  New slots need to be mapped.  Old slots need to be
 906          * un-mapped and re-mapped if their base changes.  Since base change
 907          * unmapping is handled above with slot deletion, mapping alone is
 908          * needed here.  Anything else the iommu might care about for existing
 909          * slots (size changes, userspace addr changes and read-only flag
 910          * changes) is disallowed above, so any other attribute changes getting
 911          * here can be skipped.
 912          */
 913         if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
 914                 r = kvm_iommu_map_pages(kvm, &new);
 915                 return r;
 916         }
 917
 918         return 0;
 919
 920 out_slots:
 921         kfree(slots);
 922 out_free:
 923         kvm_free_physmem_slot(kvm, &new, &old);
 924 out:
 925         return r;
 926 }
 927 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 928
 929 int kvm_set_memory_region(struct kvm *kvm,
 930                           struct kvm_userspace_memory_region *mem)
 931 {
 932         int r;
 933
 934         mutex_lock(&kvm->slots_lock);
 935         r = __kvm_set_memory_region(kvm, mem);
 936         mutex_unlock(&kvm->slots_lock);
 937         return r;
 938 }
 939 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 940
 941 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 942                                           struct kvm_userspace_memory_region *mem)
 943 {
 944         if (mem->slot >= KVM_USER_MEM_SLOTS)
 945                 return -EINVAL;
 946         return kvm_set_memory_region(kvm, mem);
 947 }
 948
 949 int kvm_get_dirty_log(struct kvm *kvm,
 950                         struct kvm_dirty_log *log, int *is_dirty)
 951 {
 952         struct kvm_memory_slot *memslot;
 953         int r, i;
 954         unsigned long n;
 955         unsigned long any = 0;
 956
 957         r = -EINVAL;
 958         if (log->slot >= KVM_USER_MEM_SLOTS)
 959                 goto out;
 960
 961         memslot = id_to_memslot(kvm->memslots, log->slot);
 962         r = -ENOENT;
 963         if (!memslot->dirty_bitmap)
 964                 goto out;
 965
 966         n = kvm_dirty_bitmap_bytes(memslot);
 967
 968         for (i = 0; !any && i < n/sizeof(long); ++i)
 969                 any = memslot->dirty_bitmap[i];
 970
 971         r = -EFAULT;
 972         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 973                 goto out;
 974
 975         if (any)
 976                 *is_dirty = 1;
 977
 978         r = 0;
 979 out:
 980         return r;
 981 }
 982
 983 bool kvm_largepages_enabled(void)
 984 {
 985         return largepages_enabled;
 986 }
 987
 988 void kvm_disable_largepages(void)
 989 {
 990         largepages_enabled = false;
 991 }
 992 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
 993
 994 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 995 {
 996         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
 997 }
 998 EXPORT_SYMBOL_GPL(gfn_to_memslot);
 999
1000 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1001 {
1002         struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1003
1004         if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1005               memslot->flags & KVM_MEMSLOT_INVALID)
1006                 return 0;
1007
1008         return 1;
1009 }
1010 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1011
1012 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1013 {
1014         struct vm_area_struct *vma;
1015         unsigned long addr, size;
1016
1017         size = PAGE_SIZE;
1018
1019         addr = gfn_to_hva(kvm, gfn);
1020         if (kvm_is_error_hva(addr))
1021                 return PAGE_SIZE;
1022
1023         down_read(&current->mm->mmap_sem);
1024         vma = find_vma(current->mm, addr);
1025         if (!vma)
1026                 goto out;
1027
1028         size = vma_kernel_pagesize(vma);
1029
1030 out:
1031         up_read(&current->mm->mmap_sem);
1032
1033         return size;
1034 }
1035
1036 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1037 {
1038         return slot->flags & KVM_MEM_READONLY;
1039 }
1040
1041 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1042                                        gfn_t *nr_pages, bool write)
1043 {
1044         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1045                 return KVM_HVA_ERR_BAD;
1046
1047         if (memslot_is_readonly(slot) && write)
1048                 return KVM_HVA_ERR_RO_BAD;
1049
1050         if (nr_pages)
1051                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1052
1053         return __gfn_to_hva_memslot(slot, gfn);
1054 }
1055
1056 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1057                                      gfn_t *nr_pages)
1058 {
1059         return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1060 }
1061
1062 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1063                                         gfn_t gfn)
1064 {
1065         return gfn_to_hva_many(slot, gfn, NULL);
1066 }
1067 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1068
1069 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1070 {
1071         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1072 }
1073 EXPORT_SYMBOL_GPL(gfn_to_hva);
1074
1075 /*
1076  * If writable is set to false, the hva returned by this function is only
1077  * allowed to be read.
1078  */
1079 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1080                                       gfn_t gfn, bool *writable)
1081 {
1082         unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1083
1084         if (!kvm_is_error_hva(hva) && writable)
1085                 *writable = !memslot_is_readonly(slot);
1086
1087         return hva;
1088 }
1089
1090 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1091 {
1092         struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1093
1094         return gfn_to_hva_memslot_prot(slot, gfn, writable);
1095 }
1096
1097 static int kvm_read_hva(void *data, void __user *hva, int len)
1098 {
1099         return __copy_from_user(data, hva, len);
1100 }
1101
1102 static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1103 {
1104         return __copy_from_user_inatomic(data, hva, len);
1105 }
1106
1107 static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1108         unsigned long start, int write, struct page **page)
1109 {
1110         int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1111
1112         if (write)
1113                 flags |= FOLL_WRITE;
1114
1115         return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1116 }
1117
1118 static inline int check_user_page_hwpoison(unsigned long addr)
1119 {
1120         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1121
1122         rc = __get_user_pages(current, current->mm, addr, 1,
1123                               flags, NULL, NULL, NULL);
1124         return rc == -EHWPOISON;
1125 }
1126
1127 /*
1128  * The atomic path to get the writable pfn which will be stored in @pfn,
1129  * true indicates success, otherwise false is returned.
1130  */
1131 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1132                             bool write_fault, bool *writable, pfn_t *pfn)
1133 {
1134         struct page *page[1];
1135         int npages;
1136
1137         if (!(async || atomic))
1138                 return false;
1139
1140         /*
1141          * Fast pin a writable pfn only if it is a write fault request
1142          * or the caller allows to map a writable pfn for a read fault
1143          * request.
1144          */
1145         if (!(write_fault || writable))
1146                 return false;
1147
1148         npages = __get_user_pages_fast(addr, 1, 1, page);
1149         if (npages == 1) {
1150                 *pfn = page_to_pfn(page[0]);
1151
1152                 if (writable)
1153                         *writable = true;
1154                 return true;
1155         }
1156
1157         return false;
1158 }
1159
1160 /*
1161  * The slow path to get the pfn of the specified host virtual address,
1162  * 1 indicates success, -errno is returned if error is detected.
1163  */
1164 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1165                            bool *writable, pfn_t *pfn)
1166 {
1167         struct page *page[1];
1168         int npages = 0;
1169
1170         might_sleep();
1171
1172         if (writable)
1173                 *writable = write_fault;
1174
1175         if (async) {
1176                 down_read(&current->mm->mmap_sem);
1177                 npages = get_user_page_nowait(current, current->mm,
1178                                               addr, write_fault, page);
1179                 up_read(&current->mm->mmap_sem);
1180         } else
1181                 npages = get_user_pages_fast(addr, 1, write_fault,
1182                                              page);
1183         if (npages != 1)
1184                 return npages;
1185
1186         /* map read fault as writable if possible */
1187         if (unlikely(!write_fault) && writable) {
1188                 struct page *wpage[1];
1189
1190                 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1191                 if (npages == 1) {
1192                         *writable = true;
1193                         put_page(page[0]);
1194                         page[0] = wpage[0];
1195                 }
1196
1197                 npages = 1;
1198         }
1199         *pfn = page_to_pfn(page[0]);
1200         return npages;
1201 }
1202
1203 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1204 {
1205         if (unlikely(!(vma->vm_flags & VM_READ)))
1206                 return false;
1207
1208         if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1209                 return false;
1210
1211         return true;
1212 }
1213
1214 /*
1215  * Pin guest page in memory and return its pfn.
1216  * @addr: host virtual address which maps memory to the guest
1217  * @atomic: whether this function can sleep
1218  * @async: whether this function need to wait IO complete if the
1219  *         host page is not in the memory
1220  * @write_fault: whether we should get a writable host page
1221  * @writable: whether it allows to map a writable host page for !@write_fault
1222  *
1223  * The function will map a writable host page for these two cases:
1224  * 1): @write_fault = true
1225  * 2): @write_fault = false && @writable, @writable will tell the caller
1226  *     whether the mapping is writable.
1227  */
1228 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1229                         bool write_fault, bool *writable)
1230 {
1231         struct vm_area_struct *vma;
1232         pfn_t pfn = 0;
1233         int npages;
1234
1235         /* we can do it either atomically or asynchronously, not both */
1236         BUG_ON(atomic && async);
1237
1238         if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1239                 return pfn;
1240
1241         if (atomic)
1242                 return KVM_PFN_ERR_FAULT;
1243
1244         npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1245         if (npages == 1)
1246                 return pfn;
1247
1248         down_read(&current->mm->mmap_sem);
1249         if (npages == -EHWPOISON ||
1250               (!async && check_user_page_hwpoison(addr))) {
1251                 pfn = KVM_PFN_ERR_HWPOISON;
1252                 goto exit;
1253         }
1254
1255         vma = find_vma_intersection(current->mm, addr, addr + 1);
1256
1257         if (vma == NULL)
1258                 pfn = KVM_PFN_ERR_FAULT;
1259         else if ((vma->vm_flags & VM_PFNMAP)) {
1260                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1261                         vma->vm_pgoff;
1262                 BUG_ON(!kvm_is_mmio_pfn(pfn));
1263         } else {
1264                 if (async && vma_is_valid(vma, write_fault))
1265                         *async = true;
1266                 pfn = KVM_PFN_ERR_FAULT;
1267         }
1268 exit:
1269         up_read(&current->mm->mmap_sem);
1270         return pfn;
1271 }
1272
1273 static pfn_t
1274 __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1275                      bool *async, bool write_fault, bool *writable)
1276 {
1277         unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1278
1279         if (addr == KVM_HVA_ERR_RO_BAD)
1280                 return KVM_PFN_ERR_RO_FAULT;
1281
1282         if (kvm_is_error_hva(addr))
1283                 return KVM_PFN_NOSLOT;
1284
1285         /* Do not map writable pfn in the readonly memslot. */
1286         if (writable && memslot_is_readonly(slot)) {
1287                 *writable = false;
1288                 writable = NULL;
1289         }
1290
1291         return hva_to_pfn(addr, atomic, async, write_fault,
1292                           writable);
1293 }
1294
1295 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1296                           bool write_fault, bool *writable)
1297 {
1298         struct kvm_memory_slot *slot;
1299
1300         if (async)
1301                 *async = false;
1302
1303         slot = gfn_to_memslot(kvm, gfn);
1304
1305         return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1306                                     writable);
1307 }
1308
1309 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1310 {
1311         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1312 }
1313 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1314
1315 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1316                        bool write_fault, bool *writable)
1317 {
1318         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1319 }
1320 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1321
1322 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1323 {
1324         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1325 }
1326 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1327
1328 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1329                       bool *writable)
1330 {
1331         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1332 }
1333 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1334
1335 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1336 {
1337         return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1338 }
1339
1340 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1341 {
1342         return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1343 }
1344 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1345
1346 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1347                                                                   int nr_pages)
1348 {
1349         unsigned long addr;
1350         gfn_t entry;
1351
1352         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1353         if (kvm_is_error_hva(addr))
1354                 return -1;
1355
1356         if (entry < nr_pages)
1357                 return 0;
1358
1359         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1360 }
1361 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1362
1363 static struct page *kvm_pfn_to_page(pfn_t pfn)
1364 {
1365         if (is_error_noslot_pfn(pfn))
1366                 return KVM_ERR_PTR_BAD_PAGE;
1367
1368         if (kvm_is_mmio_pfn(pfn)) {
1369                 WARN_ON(1);
1370                 return KVM_ERR_PTR_BAD_PAGE;
1371         }
1372
1373         return pfn_to_page(pfn);
1374 }
1375
1376 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1377 {
1378         pfn_t pfn;
1379
1380         pfn = gfn_to_pfn(kvm, gfn);
1381
1382         return kvm_pfn_to_page(pfn);
1383 }
1384
1385 EXPORT_SYMBOL_GPL(gfn_to_page);
1386
1387 void kvm_release_page_clean(struct page *page)
1388 {
1389         WARN_ON(is_error_page(page));
1390
1391         kvm_release_pfn_clean(page_to_pfn(page));
1392 }
1393 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1394
1395 void kvm_release_pfn_clean(pfn_t pfn)
1396 {
1397         if (!is_error_noslot_pfn(pfn) && !kvm_is_mmio_pfn(pfn))
1398                 put_page(pfn_to_page(pfn));
1399 }
1400 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1401
1402 void kvm_release_page_dirty(struct page *page)
1403 {
1404         WARN_ON(is_error_page(page));
1405
1406         kvm_release_pfn_dirty(page_to_pfn(page));
1407 }
1408 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1409
1410 static void kvm_release_pfn_dirty(pfn_t pfn)
1411 {
1412         kvm_set_pfn_dirty(pfn);
1413         kvm_release_pfn_clean(pfn);
1414 }
1415
1416 void kvm_set_pfn_dirty(pfn_t pfn)
1417 {
1418         if (!kvm_is_mmio_pfn(pfn)) {
1419                 struct page *page = pfn_to_page(pfn);
1420                 if (!PageReserved(page))
1421                         SetPageDirty(page);
1422         }
1423 }
1424 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1425
1426 void kvm_set_pfn_accessed(pfn_t pfn)
1427 {
1428         if (!kvm_is_mmio_pfn(pfn))
1429                 mark_page_accessed(pfn_to_page(pfn));
1430 }
1431 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1432
1433 void kvm_get_pfn(pfn_t pfn)
1434 {
1435         if (!kvm_is_mmio_pfn(pfn))
1436                 get_page(pfn_to_page(pfn));
1437 }
1438 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1439
1440 static int next_segment(unsigned long len, int offset)
1441 {
1442         if (len > PAGE_SIZE - offset)
1443                 return PAGE_SIZE - offset;
1444         else
1445                 return len;
1446 }
1447
1448 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1449                         int len)
1450 {
1451         int r;
1452         unsigned long addr;
1453
1454         addr = gfn_to_hva_prot(kvm, gfn, NULL);
1455         if (kvm_is_error_hva(addr))
1456                 return -EFAULT;
1457         r = kvm_read_hva(data, (void __user *)addr + offset, len);
1458         if (r)
1459                 return -EFAULT;
1460         return 0;
1461 }
1462 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1463
1464 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1465 {
1466         gfn_t gfn = gpa >> PAGE_SHIFT;
1467         int seg;
1468         int offset = offset_in_page(gpa);
1469         int ret;
1470
1471         while ((seg = next_segment(len, offset)) != 0) {
1472                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1473                 if (ret < 0)
1474                         return ret;
1475                 offset = 0;
1476                 len -= seg;
1477                 data += seg;
1478                 ++gfn;
1479         }
1480         return 0;
1481 }
1482 EXPORT_SYMBOL_GPL(kvm_read_guest);
1483
1484 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1485                           unsigned long len)
1486 {
1487         int r;
1488         unsigned long addr;
1489         gfn_t gfn = gpa >> PAGE_SHIFT;
1490         int offset = offset_in_page(gpa);
1491
1492         addr = gfn_to_hva_prot(kvm, gfn, NULL);
1493         if (kvm_is_error_hva(addr))
1494                 return -EFAULT;
1495         pagefault_disable();
1496         r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
1497         pagefault_enable();
1498         if (r)
1499                 return -EFAULT;
1500         return 0;
1501 }
1502 EXPORT_SYMBOL(kvm_read_guest_atomic);
1503
1504 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1505                          int offset, int len)
1506 {
1507         int r;
1508         unsigned long addr;
1509
1510         addr = gfn_to_hva(kvm, gfn);
1511         if (kvm_is_error_hva(addr))
1512                 return -EFAULT;
1513         r = __copy_to_user((void __user *)addr + offset, data, len);
1514         if (r)
1515                 return -EFAULT;
1516         mark_page_dirty(kvm, gfn);
1517         return 0;
1518 }
1519 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1520
1521 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1522                     unsigned long len)
1523 {
1524         gfn_t gfn = gpa >> PAGE_SHIFT;
1525         int seg;
1526         int offset = offset_in_page(gpa);
1527         int ret;
1528
1529         while ((seg = next_segment(len, offset)) != 0) {
1530                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1531                 if (ret < 0)
1532                         return ret;
1533                 offset = 0;
1534                 len -= seg;
1535                 data += seg;
1536                 ++gfn;
1537         }
1538         return 0;
1539 }
1540
1541 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1542                               gpa_t gpa, unsigned long len)
1543 {
1544         struct kvm_memslots *slots = kvm_memslots(kvm);
1545         int offset = offset_in_page(gpa);
1546         gfn_t start_gfn = gpa >> PAGE_SHIFT;
1547         gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1548         gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1549         gfn_t nr_pages_avail;
1550
1551         ghc->gpa = gpa;
1552         ghc->generation = slots->generation;
1553         ghc->len = len;
1554         ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1555         ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, &nr_pages_avail);
1556         if (!kvm_is_error_hva(ghc->hva) && nr_pages_avail >= nr_pages_needed) {
1557                 ghc->hva += offset;
1558         } else {
1559                 /*
1560                  * If the requested region crosses two memslots, we still
1561                  * verify that the entire region is valid here.
1562                  */
1563                 while (start_gfn <= end_gfn) {
1564                         ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1565                         ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1566                                                    &nr_pages_avail);
1567                         if (kvm_is_error_hva(ghc->hva))
1568                                 return -EFAULT;
1569                         start_gfn += nr_pages_avail;
1570                 }
1571                 /* Use the slow path for cross page reads and writes. */
1572                 ghc->memslot = NULL;
1573         }
1574         return 0;
1575 }
1576 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1577
1578 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1579                            void *data, unsigned long len)
1580 {
1581         struct kvm_memslots *slots = kvm_memslots(kvm);
1582         int r;
1583
1584         BUG_ON(len > ghc->len);
1585
1586         if (slots->generation != ghc->generation)
1587                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1588
1589         if (unlikely(!ghc->memslot))
1590                 return kvm_write_guest(kvm, ghc->gpa, data, len);
1591
1592         if (kvm_is_error_hva(ghc->hva))
1593                 return -EFAULT;
1594
1595         r = __copy_to_user((void __user *)ghc->hva, data, len);
1596         if (r)
1597                 return -EFAULT;
1598         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1599
1600         return 0;
1601 }
1602 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1603
1604 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1605                            void *data, unsigned long len)
1606 {
1607         struct kvm_memslots *slots = kvm_memslots(kvm);
1608         int r;
1609
1610         BUG_ON(len > ghc->len);
1611
1612         if (slots->generation != ghc->generation)
1613                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1614
1615         if (unlikely(!ghc->memslot))
1616                 return kvm_read_guest(kvm, ghc->gpa, data, len);
1617
1618         if (kvm_is_error_hva(ghc->hva))
1619                 return -EFAULT;
1620
1621         r = __copy_from_user(data, (void __user *)ghc->hva, len);
1622         if (r)
1623                 return -EFAULT;
1624
1625         return 0;
1626 }
1627 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1628
1629 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1630 {
1631         const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
1632
1633         return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
1634 }
1635 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1636
1637 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1638 {
1639         gfn_t gfn = gpa >> PAGE_SHIFT;
1640         int seg;
1641         int offset = offset_in_page(gpa);
1642         int ret;
1643
1644         while ((seg = next_segment(len, offset)) != 0) {
1645                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1646                 if (ret < 0)
1647                         return ret;
1648                 offset = 0;
1649                 len -= seg;
1650                 ++gfn;
1651         }
1652         return 0;
1653 }
1654 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1655
1656 static void mark_page_dirty_in_slot(struct kvm *kvm,
1657                                     struct kvm_memory_slot *memslot,
1658                                     gfn_t gfn)
1659 {
1660         if (memslot && memslot->dirty_bitmap) {
1661                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1662
1663                 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1664         }
1665 }
1666
1667 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1668 {
1669         struct kvm_memory_slot *memslot;
1670
1671         memslot = gfn_to_memslot(kvm, gfn);
1672         mark_page_dirty_in_slot(kvm, memslot, gfn);
1673 }
1674
1675 /*
1676  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1677  */
1678 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1679 {
1680         DEFINE_WAIT(wait);
1681
1682         for (;;) {
1683                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1684
1685                 if (kvm_arch_vcpu_runnable(vcpu)) {
1686                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1687                         break;
1688                 }
1689                 if (kvm_cpu_has_pending_timer(vcpu))
1690                         break;
1691                 if (signal_pending(current))
1692                         break;
1693
1694                 schedule();
1695         }
1696
1697         finish_wait(&vcpu->wq, &wait);
1698 }
1699
1700 #ifndef CONFIG_S390
1701 /*
1702  * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1703  */
1704 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1705 {
1706         int me;
1707         int cpu = vcpu->cpu;
1708         wait_queue_head_t *wqp;
1709
1710         wqp = kvm_arch_vcpu_wq(vcpu);
1711         if (waitqueue_active(wqp)) {
1712                 wake_up_interruptible(wqp);
1713                 ++vcpu->stat.halt_wakeup;
1714         }
1715
1716         me = get_cpu();
1717         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1718                 if (kvm_arch_vcpu_should_kick(vcpu))
1719                         smp_send_reschedule(cpu);
1720         put_cpu();
1721 }
1722 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
1723 #endif /* !CONFIG_S390 */
1724
1725 bool kvm_vcpu_yield_to(struct kvm_vcpu *target)
1726 {
1727         struct pid *pid;
1728         struct task_struct *task = NULL;
1729         bool ret = false;
1730
1731         rcu_read_lock();
1732         pid = rcu_dereference(target->pid);
1733         if (pid)
1734                 task = get_pid_task(target->pid, PIDTYPE_PID);
1735         rcu_read_unlock();
1736         if (!task)
1737                 return ret;
1738         if (task->flags & PF_VCPU) {
1739                 put_task_struct(task);
1740                 return ret;
1741         }
1742         ret = yield_to(task, 1);
1743         put_task_struct(task);
1744
1745         return ret;
1746 }
1747 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1748
1749 /*
1750  * Helper that checks whether a VCPU is eligible for directed yield.
1751  * Most eligible candidate to yield is decided by following heuristics:
1752  *
1753  *  (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1754  *  (preempted lock holder), indicated by @in_spin_loop.
1755  *  Set at the beiginning and cleared at the end of interception/PLE handler.
1756  *
1757  *  (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1758  *  chance last time (mostly it has become eligible now since we have probably
1759  *  yielded to lockholder in last iteration. This is done by toggling
1760  *  @dy_eligible each time a VCPU checked for eligibility.)
1761  *
1762  *  Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1763  *  to preempted lock-holder could result in wrong VCPU selection and CPU
1764  *  burning. Giving priority for a potential lock-holder increases lock
1765  *  progress.
1766  *
1767  *  Since algorithm is based on heuristics, accessing another VCPU data without
1768  *  locking does not harm. It may result in trying to yield to  same VCPU, fail
1769  *  and continue with next VCPU and so on.
1770  */
1771 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1772 {
1773 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1774         bool eligible;
1775
1776         eligible = !vcpu->spin_loop.in_spin_loop ||
1777                     vcpu->spin_loop.dy_eligible;
1778
1779         if (vcpu->spin_loop.in_spin_loop)
1780                 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1781
1782         return eligible;
1783 #else
1784         return true;
1785 #endif
1786 }
1787
1788 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1789 {
1790         struct kvm *kvm = me->kvm;
1791         struct kvm_vcpu *vcpu;
1792         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1793         int yielded = 0;
1794         int try = 3;
1795         int pass;
1796         int i;
1797
1798         kvm_vcpu_set_in_spin_loop(me, true);
1799         /*
1800          * We boost the priority of a VCPU that is runnable but not
1801          * currently running, because it got preempted by something
1802          * else and called schedule in __vcpu_run.  Hopefully that
1803          * VCPU is holding the lock that we need and will release it.
1804          * We approximate round-robin by starting at the last boosted VCPU.
1805          */
1806         for (pass = 0; pass < 2 && !yielded && try; pass++) {
1807                 kvm_for_each_vcpu(i, vcpu, kvm) {
1808                         if (!pass && i <= last_boosted_vcpu) {
1809                                 i = last_boosted_vcpu;
1810                                 continue;
1811                         } else if (pass && i > last_boosted_vcpu)
1812                                 break;
1813                         if (!ACCESS_ONCE(vcpu->preempted))
1814                                 continue;
1815                         if (vcpu == me)
1816                                 continue;
1817                         if (waitqueue_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
1818                                 continue;
1819                         if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1820                                 continue;
1821
1822                         yielded = kvm_vcpu_yield_to(vcpu);
1823                         if (yielded > 0) {
1824                                 kvm->last_boosted_vcpu = i;
1825                                 break;
1826                         } else if (yielded < 0) {
1827                                 try--;
1828                                 if (!try)
1829                                         break;
1830                         }
1831                 }
1832         }
1833         kvm_vcpu_set_in_spin_loop(me, false);
1834
1835         /* Ensure vcpu is not eligible during next spinloop */
1836         kvm_vcpu_set_dy_eligible(me, false);
1837 }
1838 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1839
1840 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1841 {
1842         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1843         struct page *page;
1844
1845         if (vmf->pgoff == 0)
1846                 page = virt_to_page(vcpu->run);
1847 #ifdef CONFIG_X86
1848         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1849                 page = virt_to_page(vcpu->arch.pio_data);
1850 #endif
1851 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1852         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1853                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1854 #endif
1855         else
1856                 return kvm_arch_vcpu_fault(vcpu, vmf);
1857         get_page(page);
1858         vmf->page = page;
1859         return 0;
1860 }
1861
1862 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1863         .fault = kvm_vcpu_fault,
1864 };
1865
1866 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1867 {
1868         vma->vm_ops = &kvm_vcpu_vm_ops;
1869         return 0;
1870 }
1871
1872 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1873 {
1874         struct kvm_vcpu *vcpu = filp->private_data;
1875
1876         kvm_put_kvm(vcpu->kvm);
1877         return 0;
1878 }
1879
1880 static struct file_operations kvm_vcpu_fops = {
1881         .release        = kvm_vcpu_release,
1882         .unlocked_ioctl = kvm_vcpu_ioctl,
1883 #ifdef CONFIG_COMPAT
1884         .compat_ioctl   = kvm_vcpu_compat_ioctl,
1885 #endif
1886         .mmap           = kvm_vcpu_mmap,
1887         .llseek         = noop_llseek,
1888 };
1889
1890 /*
1891  * Allocates an inode for the vcpu.
1892  */
1893 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1894 {
1895         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
1896 }
1897
1898 /*
1899  * Creates some virtual cpus.  Good luck creating more than one.
1900  */
1901 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1902 {
1903         int r;
1904         struct kvm_vcpu *vcpu, *v;
1905
1906         if (id >= KVM_MAX_VCPUS)
1907                 return -EINVAL;
1908
1909         vcpu = kvm_arch_vcpu_create(kvm, id);
1910         if (IS_ERR(vcpu))
1911                 return PTR_ERR(vcpu);
1912
1913         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1914
1915         r = kvm_arch_vcpu_setup(vcpu);
1916         if (r)
1917                 goto vcpu_destroy;
1918
1919         mutex_lock(&kvm->lock);
1920         if (!kvm_vcpu_compatible(vcpu)) {
1921                 r = -EINVAL;
1922                 goto unlock_vcpu_destroy;
1923         }
1924         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1925                 r = -EINVAL;
1926                 goto unlock_vcpu_destroy;
1927         }
1928
1929         kvm_for_each_vcpu(r, v, kvm)
1930                 if (v->vcpu_id == id) {
1931                         r = -EEXIST;
1932                         goto unlock_vcpu_destroy;
1933                 }
1934
1935         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1936
1937         /* Now it's all set up, let userspace reach it */
1938         kvm_get_kvm(kvm);
1939         r = create_vcpu_fd(vcpu);
1940         if (r < 0) {
1941                 kvm_put_kvm(kvm);
1942                 goto unlock_vcpu_destroy;
1943         }
1944
1945         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1946         smp_wmb();
1947         atomic_inc(&kvm->online_vcpus);
1948
1949         mutex_unlock(&kvm->lock);
1950         kvm_arch_vcpu_postcreate(vcpu);
1951         return r;
1952
1953 unlock_vcpu_destroy:
1954         mutex_unlock(&kvm->lock);
1955 vcpu_destroy:
1956         kvm_arch_vcpu_destroy(vcpu);
1957         return r;
1958 }
1959
1960 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1961 {
1962         if (sigset) {
1963                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1964                 vcpu->sigset_active = 1;
1965                 vcpu->sigset = *sigset;
1966         } else
1967                 vcpu->sigset_active = 0;
1968         return 0;
1969 }
1970
1971 static long kvm_vcpu_ioctl(struct file *filp,
1972                            unsigned int ioctl, unsigned long arg)
1973 {
1974         struct kvm_vcpu *vcpu = filp->private_data;
1975         void __user *argp = (void __user *)arg;
1976         int r;
1977         struct kvm_fpu *fpu = NULL;
1978         struct kvm_sregs *kvm_sregs = NULL;
1979
1980         if (vcpu->kvm->mm != current->mm)
1981                 return -EIO;
1982
1983 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
1984         /*
1985          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1986          * so vcpu_load() would break it.
1987          */
1988         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1989                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1990 #endif
1991
1992
1993         r = vcpu_load(vcpu);
1994         if (r)
1995                 return r;
1996         switch (ioctl) {
1997         case KVM_RUN:
1998                 r = -EINVAL;
1999                 if (arg)
2000                         goto out;
2001                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2002                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2003                 break;
2004         case KVM_GET_REGS: {
2005                 struct kvm_regs *kvm_regs;
2006
2007                 r = -ENOMEM;
2008                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2009                 if (!kvm_regs)
2010                         goto out;
2011                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2012                 if (r)
2013                         goto out_free1;
2014                 r = -EFAULT;
2015                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2016                         goto out_free1;
2017                 r = 0;
2018 out_free1:
2019                 kfree(kvm_regs);
2020                 break;
2021         }
2022         case KVM_SET_REGS: {
2023                 struct kvm_regs *kvm_regs;
2024
2025                 r = -ENOMEM;
2026                 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2027                 if (IS_ERR(kvm_regs)) {
2028                         r = PTR_ERR(kvm_regs);
2029                         goto out;
2030                 }
2031                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2032                 kfree(kvm_regs);
2033                 break;
2034         }
2035         case KVM_GET_SREGS: {
2036                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2037                 r = -ENOMEM;
2038                 if (!kvm_sregs)
2039                         goto out;
2040                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2041                 if (r)
2042                         goto out;
2043                 r = -EFAULT;
2044                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2045                         goto out;
2046                 r = 0;
2047                 break;
2048         }
2049         case KVM_SET_SREGS: {
2050                 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2051                 if (IS_ERR(kvm_sregs)) {
2052                         r = PTR_ERR(kvm_sregs);
2053                         kvm_sregs = NULL;
2054                         goto out;
2055                 }
2056                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2057                 break;
2058         }
2059         case KVM_GET_MP_STATE: {
2060                 struct kvm_mp_state mp_state;
2061
2062                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2063                 if (r)
2064                         goto out;
2065                 r = -EFAULT;
2066                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
2067                         goto out;
2068                 r = 0;
2069                 break;
2070         }
2071         case KVM_SET_MP_STATE: {
2072                 struct kvm_mp_state mp_state;
2073
2074                 r = -EFAULT;
2075                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
2076                         goto out;
2077                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2078                 break;
2079         }
2080         case KVM_TRANSLATE: {
2081                 struct kvm_translation tr;
2082
2083                 r = -EFAULT;
2084                 if (copy_from_user(&tr, argp, sizeof tr))
2085                         goto out;
2086                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2087                 if (r)
2088                         goto out;
2089                 r = -EFAULT;
2090                 if (copy_to_user(argp, &tr, sizeof tr))
2091                         goto out;
2092                 r = 0;
2093                 break;
2094         }
2095         case KVM_SET_GUEST_DEBUG: {
2096                 struct kvm_guest_debug dbg;
2097
2098                 r = -EFAULT;
2099                 if (copy_from_user(&dbg, argp, sizeof dbg))
2100                         goto out;
2101                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2102                 break;
2103         }
2104         case KVM_SET_SIGNAL_MASK: {
2105                 struct kvm_signal_mask __user *sigmask_arg = argp;
2106                 struct kvm_signal_mask kvm_sigmask;
2107                 sigset_t sigset, *p;
2108
2109                 p = NULL;
2110                 if (argp) {
2111                         r = -EFAULT;
2112                         if (copy_from_user(&kvm_sigmask, argp,
2113                                            sizeof kvm_sigmask))
2114                                 goto out;
2115                         r = -EINVAL;
2116                         if (kvm_sigmask.len != sizeof sigset)
2117                                 goto out;
2118                         r = -EFAULT;
2119                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2120                                            sizeof sigset))
2121                                 goto out;
2122                         p = &sigset;
2123                 }
2124                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2125                 break;
2126         }
2127         case KVM_GET_FPU: {
2128                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2129                 r = -ENOMEM;
2130                 if (!fpu)
2131                         goto out;
2132                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2133                 if (r)
2134                         goto out;
2135                 r = -EFAULT;
2136                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2137                         goto out;
2138                 r = 0;
2139                 break;
2140         }
2141         case KVM_SET_FPU: {
2142                 fpu = memdup_user(argp, sizeof(*fpu));
2143                 if (IS_ERR(fpu)) {
2144                         r = PTR_ERR(fpu);
2145                         fpu = NULL;
2146                         goto out;
2147                 }
2148                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2149                 break;
2150         }
2151         default:
2152                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2153         }
2154 out:
2155         vcpu_put(vcpu);
2156         kfree(fpu);
2157         kfree(kvm_sregs);
2158         return r;
2159 }
2160
2161 #ifdef CONFIG_COMPAT
2162 static long kvm_vcpu_compat_ioctl(struct file *filp,
2163                                   unsigned int ioctl, unsigned long arg)
2164 {
2165         struct kvm_vcpu *vcpu = filp->private_data;
2166         void __user *argp = compat_ptr(arg);
2167         int r;
2168
2169         if (vcpu->kvm->mm != current->mm)
2170                 return -EIO;
2171
2172         switch (ioctl) {
2173         case KVM_SET_SIGNAL_MASK: {
2174                 struct kvm_signal_mask __user *sigmask_arg = argp;
2175                 struct kvm_signal_mask kvm_sigmask;
2176                 compat_sigset_t csigset;
2177                 sigset_t sigset;
2178
2179                 if (argp) {
2180                         r = -EFAULT;
2181                         if (copy_from_user(&kvm_sigmask, argp,
2182                                            sizeof kvm_sigmask))
2183                                 goto out;
2184                         r = -EINVAL;
2185                         if (kvm_sigmask.len != sizeof csigset)
2186                                 goto out;
2187                         r = -EFAULT;
2188                         if (copy_from_user(&csigset, sigmask_arg->sigset,
2189                                            sizeof csigset))
2190                                 goto out;
2191                         sigset_from_compat(&sigset, &csigset);
2192                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2193                 } else
2194                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2195                 break;
2196         }
2197         default:
2198                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2199         }
2200
2201 out:
2202         return r;
2203 }
2204 #endif
2205
2206 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2207                                  int (*accessor)(struct kvm_device *dev,
2208                                                  struct kvm_device_attr *attr),
2209                                  unsigned long arg)
2210 {
2211         struct kvm_device_attr attr;
2212
2213         if (!accessor)
2214                 return -EPERM;
2215
2216         if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2217                 return -EFAULT;
2218
2219         return accessor(dev, &attr);
2220 }
2221
2222 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2223                              unsigned long arg)
2224 {
2225         struct kvm_device *dev = filp->private_data;
2226
2227         switch (ioctl) {
2228         case KVM_SET_DEVICE_ATTR:
2229                 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2230         case KVM_GET_DEVICE_ATTR:
2231                 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2232         case KVM_HAS_DEVICE_ATTR:
2233                 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2234         default:
2235                 if (dev->ops->ioctl)
2236                         return dev->ops->ioctl(dev, ioctl, arg);
2237
2238                 return -ENOTTY;
2239         }
2240 }
2241
2242 static int kvm_device_release(struct inode *inode, struct file *filp)
2243 {
2244         struct kvm_device *dev = filp->private_data;
2245         struct kvm *kvm = dev->kvm;
2246
2247         kvm_put_kvm(kvm);
2248         return 0;
2249 }
2250
2251 static const struct file_operations kvm_device_fops = {
2252         .unlocked_ioctl = kvm_device_ioctl,
2253 #ifdef CONFIG_COMPAT
2254         .compat_ioctl = kvm_device_ioctl,
2255 #endif
2256         .release = kvm_device_release,
2257 };
2258
2259 struct kvm_device *kvm_device_from_filp(struct file *filp)
2260 {
2261         if (filp->f_op != &kvm_device_fops)
2262                 return NULL;
2263
2264         return filp->private_data;
2265 }
2266
2267 static int kvm_ioctl_create_device(struct kvm *kvm,
2268                                    struct kvm_create_device *cd)
2269 {
2270         struct kvm_device_ops *ops = NULL;
2271         struct kvm_device *dev;
2272         bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2273         int ret;
2274
2275         switch (cd->type) {
2276 #ifdef CONFIG_KVM_MPIC
2277         case KVM_DEV_TYPE_FSL_MPIC_20:
2278         case KVM_DEV_TYPE_FSL_MPIC_42:
2279                 ops = &kvm_mpic_ops;
2280                 break;
2281 #endif
2282 #ifdef CONFIG_KVM_XICS
2283         case KVM_DEV_TYPE_XICS:
2284                 ops = &kvm_xics_ops;
2285                 break;
2286 #endif
2287 #ifdef CONFIG_KVM_VFIO
2288         case KVM_DEV_TYPE_VFIO:
2289                 ops = &kvm_vfio_ops;
2290                 break;
2291 #endif
2292 #ifdef CONFIG_KVM_ARM_VGIC
2293         case KVM_DEV_TYPE_ARM_VGIC_V2:
2294                 ops = &kvm_arm_vgic_v2_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();
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();
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 int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2799 {
2800         const struct kvm_io_range *r1 = p1;
2801         const struct kvm_io_range *r2 = p2;
2802
2803         if (r1->addr < r2->addr)
2804                 return -1;
2805         if (r1->addr + r1->len > r2->addr + r2->len)
2806                 return 1;
2807         return 0;
2808 }
2809
2810 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2811                           gpa_t addr, int len)
2812 {
2813         bus->range[bus->dev_count++] = (struct kvm_io_range) {
2814                 .addr = addr,
2815                 .len = len,
2816                 .dev = dev,
2817         };
2818
2819         sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2820                 kvm_io_bus_sort_cmp, NULL);
2821
2822         return 0;
2823 }
2824
2825 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2826                              gpa_t addr, int len)
2827 {
2828         struct kvm_io_range *range, key;
2829         int off;
2830
2831         key = (struct kvm_io_range) {
2832                 .addr = addr,
2833                 .len = len,
2834         };
2835
2836         range = bsearch(&key, bus->range, bus->dev_count,
2837                         sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2838         if (range == NULL)
2839                 return -ENOENT;
2840
2841         off = range - bus->range;
2842
2843         while (off > 0 && kvm_io_bus_sort_cmp(&key, &bus->range[off-1]) == 0)
2844                 off--;
2845
2846         return off;
2847 }
2848
2849 /* kvm_io_bus_write - called under kvm->slots_lock */
2850 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2851                      int len, const void *val)
2852 {
2853         int idx;
2854         struct kvm_io_bus *bus;
2855         struct kvm_io_range range;
2856
2857         range = (struct kvm_io_range) {
2858                 .addr = addr,
2859                 .len = len,
2860         };
2861
2862         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2863         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2864         if (idx < 0)
2865                 return -EOPNOTSUPP;
2866
2867         while (idx < bus->dev_count &&
2868                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2869                 if (!kvm_iodevice_write(bus->range[idx].dev, addr, len, val))
2870                         return 0;
2871                 idx++;
2872         }
2873
2874         return -EOPNOTSUPP;
2875 }
2876
2877 /* kvm_io_bus_read - called under kvm->slots_lock */
2878 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2879                     int len, void *val)
2880 {
2881         int idx;
2882         struct kvm_io_bus *bus;
2883         struct kvm_io_range range;
2884
2885         range = (struct kvm_io_range) {
2886                 .addr = addr,
2887                 .len = len,
2888         };
2889
2890         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2891         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2892         if (idx < 0)
2893                 return -EOPNOTSUPP;
2894
2895         while (idx < bus->dev_count &&
2896                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2897                 if (!kvm_iodevice_read(bus->range[idx].dev, addr, len, val))
2898                         return 0;
2899                 idx++;
2900         }
2901
2902         return -EOPNOTSUPP;
2903 }
2904
2905 /* Caller must hold slots_lock. */
2906 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2907                             int len, struct kvm_io_device *dev)
2908 {
2909         struct kvm_io_bus *new_bus, *bus;
2910
2911         bus = kvm->buses[bus_idx];
2912         if (bus->dev_count > NR_IOBUS_DEVS - 1)
2913                 return -ENOSPC;
2914
2915         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
2916                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2917         if (!new_bus)
2918                 return -ENOMEM;
2919         memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
2920                sizeof(struct kvm_io_range)));
2921         kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2922         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2923         synchronize_srcu_expedited(&kvm->srcu);
2924         kfree(bus);
2925
2926         return 0;
2927 }
2928
2929 /* Caller must hold slots_lock. */
2930 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2931                               struct kvm_io_device *dev)
2932 {
2933         int i, r;
2934         struct kvm_io_bus *new_bus, *bus;
2935
2936         bus = kvm->buses[bus_idx];
2937         r = -ENOENT;
2938         for (i = 0; i < bus->dev_count; i++)
2939                 if (bus->range[i].dev == dev) {
2940                         r = 0;
2941                         break;
2942                 }
2943
2944         if (r)
2945                 return r;
2946
2947         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
2948                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2949         if (!new_bus)
2950                 return -ENOMEM;
2951
2952         memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
2953         new_bus->dev_count--;
2954         memcpy(new_bus->range + i, bus->range + i + 1,
2955                (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
2956
2957         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2958         synchronize_srcu_expedited(&kvm->srcu);
2959         kfree(bus);
2960         return r;
2961 }
2962
2963 static struct notifier_block kvm_cpu_notifier = {
2964         .notifier_call = kvm_cpu_hotplug,
2965 };
2966
2967 static int vm_stat_get(void *_offset, u64 *val)
2968 {
2969         unsigned offset = (long)_offset;
2970         struct kvm *kvm;
2971
2972         *val = 0;
2973         spin_lock(&kvm_lock);
2974         list_for_each_entry(kvm, &vm_list, vm_list)
2975                 *val += *(u32 *)((void *)kvm + offset);
2976         spin_unlock(&kvm_lock);
2977         return 0;
2978 }
2979
2980 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2981
2982 static int vcpu_stat_get(void *_offset, u64 *val)
2983 {
2984         unsigned offset = (long)_offset;
2985         struct kvm *kvm;
2986         struct kvm_vcpu *vcpu;
2987         int i;
2988
2989         *val = 0;
2990         spin_lock(&kvm_lock);
2991         list_for_each_entry(kvm, &vm_list, vm_list)
2992                 kvm_for_each_vcpu(i, vcpu, kvm)
2993                         *val += *(u32 *)((void *)vcpu + offset);
2994
2995         spin_unlock(&kvm_lock);
2996         return 0;
2997 }
2998
2999 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
3000
3001 static const struct file_operations *stat_fops[] = {
3002         [KVM_STAT_VCPU] = &vcpu_stat_fops,
3003         [KVM_STAT_VM]   = &vm_stat_fops,
3004 };
3005
3006 static int kvm_init_debug(void)
3007 {
3008         int r = -EEXIST;
3009         struct kvm_stats_debugfs_item *p;
3010
3011         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3012         if (kvm_debugfs_dir == NULL)
3013                 goto out;
3014
3015         for (p = debugfs_entries; p->name; ++p) {
3016                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
3017                                                 (void *)(long)p->offset,
3018                                                 stat_fops[p->kind]);
3019                 if (p->dentry == NULL)
3020                         goto out_dir;
3021         }
3022
3023         return 0;
3024
3025 out_dir:
3026         debugfs_remove_recursive(kvm_debugfs_dir);
3027 out:
3028         return r;
3029 }
3030
3031 static void kvm_exit_debug(void)
3032 {
3033         struct kvm_stats_debugfs_item *p;
3034
3035         for (p = debugfs_entries; p->name; ++p)
3036                 debugfs_remove(p->dentry);
3037         debugfs_remove(kvm_debugfs_dir);
3038 }
3039
3040 static int kvm_suspend(void)
3041 {
3042         if (kvm_usage_count)
3043                 hardware_disable_nolock(NULL);
3044         return 0;
3045 }
3046
3047 static void kvm_resume(void)
3048 {
3049         if (kvm_usage_count) {
3050                 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3051                 hardware_enable_nolock(NULL);
3052         }
3053 }
3054
3055 static struct syscore_ops kvm_syscore_ops = {
3056         .suspend = kvm_suspend,
3057         .resume = kvm_resume,
3058 };
3059
3060 static inline
3061 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3062 {
3063         return container_of(pn, struct kvm_vcpu, preempt_notifier);
3064 }
3065
3066 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3067 {
3068         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3069         if (vcpu->preempted)
3070                 vcpu->preempted = false;
3071
3072         kvm_arch_sched_in(vcpu, cpu);
3073
3074         kvm_arch_vcpu_load(vcpu, cpu);
3075 }
3076
3077 static void kvm_sched_out(struct preempt_notifier *pn,
3078                           struct task_struct *next)
3079 {
3080         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3081
3082         if (current->state == TASK_RUNNING)
3083                 vcpu->preempted = true;
3084         kvm_arch_vcpu_put(vcpu);
3085 }
3086
3087 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3088                   struct module *module)
3089 {
3090         int r;
3091         int cpu;
3092
3093         r = kvm_arch_init(opaque);
3094         if (r)
3095                 goto out_fail;
3096
3097         /*
3098          * kvm_arch_init makes sure there's at most one caller
3099          * for architectures that support multiple implementations,
3100          * like intel and amd on x86.
3101          * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3102          * conflicts in case kvm is already setup for another implementation.
3103          */
3104         r = kvm_irqfd_init();
3105         if (r)
3106                 goto out_irqfd;
3107
3108         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
3109                 r = -ENOMEM;
3110                 goto out_free_0;
3111         }
3112
3113         r = kvm_arch_hardware_setup();
3114         if (r < 0)
3115                 goto out_free_0a;
3116
3117         for_each_online_cpu(cpu) {
3118                 smp_call_function_single(cpu,
3119                                 kvm_arch_check_processor_compat,
3120                                 &r, 1);
3121                 if (r < 0)
3122                         goto out_free_1;
3123         }
3124
3125         r = register_cpu_notifier(&kvm_cpu_notifier);
3126         if (r)
3127                 goto out_free_2;
3128         register_reboot_notifier(&kvm_reboot_notifier);
3129
3130         /* A kmem cache lets us meet the alignment requirements of fx_save. */
3131         if (!vcpu_align)
3132                 vcpu_align = __alignof__(struct kvm_vcpu);
3133         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
3134                                            0, NULL);
3135         if (!kvm_vcpu_cache) {
3136                 r = -ENOMEM;
3137                 goto out_free_3;
3138         }
3139
3140         r = kvm_async_pf_init();
3141         if (r)
3142                 goto out_free;
3143
3144         kvm_chardev_ops.owner = module;
3145         kvm_vm_fops.owner = module;
3146         kvm_vcpu_fops.owner = module;
3147
3148         r = misc_register(&kvm_dev);
3149         if (r) {
3150                 printk(KERN_ERR "kvm: misc device register failed\n");
3151                 goto out_unreg;
3152         }
3153
3154         register_syscore_ops(&kvm_syscore_ops);
3155
3156         kvm_preempt_ops.sched_in = kvm_sched_in;
3157         kvm_preempt_ops.sched_out = kvm_sched_out;
3158
3159         r = kvm_init_debug();
3160         if (r) {
3161                 printk(KERN_ERR "kvm: create debugfs files failed\n");
3162                 goto out_undebugfs;
3163         }
3164
3165         return 0;
3166
3167 out_undebugfs:
3168         unregister_syscore_ops(&kvm_syscore_ops);
3169         misc_deregister(&kvm_dev);
3170 out_unreg:
3171         kvm_async_pf_deinit();
3172 out_free:
3173         kmem_cache_destroy(kvm_vcpu_cache);
3174 out_free_3:
3175         unregister_reboot_notifier(&kvm_reboot_notifier);
3176         unregister_cpu_notifier(&kvm_cpu_notifier);
3177 out_free_2:
3178 out_free_1:
3179         kvm_arch_hardware_unsetup();
3180 out_free_0a:
3181         free_cpumask_var(cpus_hardware_enabled);
3182 out_free_0:
3183         kvm_irqfd_exit();
3184 out_irqfd:
3185         kvm_arch_exit();
3186 out_fail:
3187         return r;
3188 }
3189 EXPORT_SYMBOL_GPL(kvm_init);
3190
3191 void kvm_exit(void)
3192 {
3193         kvm_exit_debug();
3194         misc_deregister(&kvm_dev);
3195         kmem_cache_destroy(kvm_vcpu_cache);
3196         kvm_async_pf_deinit();
3197         unregister_syscore_ops(&kvm_syscore_ops);
3198         unregister_reboot_notifier(&kvm_reboot_notifier);
3199         unregister_cpu_notifier(&kvm_cpu_notifier);
3200         on_each_cpu(hardware_disable_nolock, NULL, 1);
3201         kvm_arch_hardware_unsetup();
3202         kvm_arch_exit();
3203         kvm_irqfd_exit();
3204         free_cpumask_var(cpus_hardware_enabled);
3205 }
3206 EXPORT_SYMBOL_GPL(kvm_exit);