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