virt/kvm/kvm_main.c

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