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  *
   9  * Authors:
  10  *   Avi Kivity   <avi@qumranet.com>
  11  *   Yaniv Kamay  <yaniv@qumranet.com>
  12  *
  13  * This work is licensed under the terms of the GNU GPL, version 2.  See
  14  * the COPYING file in the top-level directory.
  15  *
  16  */
  17
  18 #include "iodev.h"
  19
  20 #include <linux/kvm_host.h>
  21 #include <linux/kvm.h>
  22 #include <linux/module.h>
  23 #include <linux/errno.h>
  24 #include <linux/percpu.h>
  25 #include <linux/mm.h>
  26 #include <linux/miscdevice.h>
  27 #include <linux/vmalloc.h>
  28 #include <linux/reboot.h>
  29 #include <linux/debugfs.h>
  30 #include <linux/highmem.h>
  31 #include <linux/file.h>
  32 #include <linux/sysdev.h>
  33 #include <linux/cpu.h>
  34 #include <linux/sched.h>
  35 #include <linux/cpumask.h>
  36 #include <linux/smp.h>
  37 #include <linux/anon_inodes.h>
  38 #include <linux/profile.h>
  39 #include <linux/kvm_para.h>
  40 #include <linux/pagemap.h>
  41 #include <linux/mman.h>
  42 #include <linux/swap.h>
  43 #include <linux/bitops.h>
  44 #include <linux/spinlock.h>
  45 #include <linux/compat.h>
  46 #include <linux/srcu.h>
  47 #include <linux/hugetlb.h>
  48 #include <linux/slab.h>
  49
  50 #include <asm/processor.h>
  51 #include <asm/io.h>
  52 #include <asm/uaccess.h>
  53 #include <asm/pgtable.h>
  54 #include <asm-generic/bitops/le.h>
  55
  56 #include "coalesced_mmio.h"
  57
  58 #define CREATE_TRACE_POINTS
  59 #include <trace/events/kvm.h>
  60
  61 MODULE_AUTHOR("Qumranet");
  62 MODULE_LICENSE("GPL");
  63
  64 /*
  65  * Ordering of locks:
  66  *
  67  *              kvm->lock --> kvm->slots_lock --> kvm->irq_lock
  68  */
  69
  70 DEFINE_SPINLOCK(kvm_lock);
  71 LIST_HEAD(vm_list);
  72
  73 static cpumask_var_t cpus_hardware_enabled;
  74 static int kvm_usage_count = 0;
  75 static atomic_t hardware_enable_failed;
  76
  77 struct kmem_cache *kvm_vcpu_cache;
  78 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
  79
  80 static __read_mostly struct preempt_ops kvm_preempt_ops;
  81
  82 struct dentry *kvm_debugfs_dir;
  83
  84 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
  85                            unsigned long arg);
  86 static int hardware_enable_all(void);
  87 static void hardware_disable_all(void);
  88
  89 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
  90
  91 static bool kvm_rebooting;
  92
  93 static bool largepages_enabled = true;
  94
  95 inline int kvm_is_mmio_pfn(pfn_t pfn)
  96 {
  97         if (pfn_valid(pfn)) {
  98                 struct page *page = compound_head(pfn_to_page(pfn));
  99                 return PageReserved(page);
 100         }
 101
 102         return true;
 103 }
 104
 105 /*
 106  * Switches to specified vcpu, until a matching vcpu_put()
 107  */
 108 void vcpu_load(struct kvm_vcpu *vcpu)
 109 {
 110         int cpu;
 111
 112         mutex_lock(&vcpu->mutex);
 113         cpu = get_cpu();
 114         preempt_notifier_register(&vcpu->preempt_notifier);
 115         kvm_arch_vcpu_load(vcpu, cpu);
 116         put_cpu();
 117 }
 118
 119 void vcpu_put(struct kvm_vcpu *vcpu)
 120 {
 121         preempt_disable();
 122         kvm_arch_vcpu_put(vcpu);
 123         preempt_notifier_unregister(&vcpu->preempt_notifier);
 124         preempt_enable();
 125         mutex_unlock(&vcpu->mutex);
 126 }
 127
 128 static void ack_flush(void *_completed)
 129 {
 130 }
 131
 132 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
 133 {
 134         int i, cpu, me;
 135         cpumask_var_t cpus;
 136         bool called = true;
 137         struct kvm_vcpu *vcpu;
 138
 139         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
 140
 141         raw_spin_lock(&kvm->requests_lock);
 142         me = smp_processor_id();
 143         kvm_for_each_vcpu(i, vcpu, kvm) {
 144                 if (test_and_set_bit(req, &vcpu->requests))
 145                         continue;
 146                 cpu = vcpu->cpu;
 147                 if (cpus != NULL && cpu != -1 && cpu != me)
 148                         cpumask_set_cpu(cpu, cpus);
 149         }
 150         if (unlikely(cpus == NULL))
 151                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
 152         else if (!cpumask_empty(cpus))
 153                 smp_call_function_many(cpus, ack_flush, NULL, 1);
 154         else
 155                 called = false;
 156         raw_spin_unlock(&kvm->requests_lock);
 157         free_cpumask_var(cpus);
 158         return called;
 159 }
 160
 161 void kvm_flush_remote_tlbs(struct kvm *kvm)
 162 {
 163         if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
 164                 ++kvm->stat.remote_tlb_flush;
 165 }
 166
 167 void kvm_reload_remote_mmus(struct kvm *kvm)
 168 {
 169         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
 170 }
 171
 172 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 173 {
 174         struct page *page;
 175         int r;
 176
 177         mutex_init(&vcpu->mutex);
 178         vcpu->cpu = -1;
 179         vcpu->kvm = kvm;
 180         vcpu->vcpu_id = id;
 181         init_waitqueue_head(&vcpu->wq);
 182
 183         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 184         if (!page) {
 185                 r = -ENOMEM;
 186                 goto fail;
 187         }
 188         vcpu->run = page_address(page);
 189
 190         r = kvm_arch_vcpu_init(vcpu);
 191         if (r < 0)
 192                 goto fail_free_run;
 193         return 0;
 194
 195 fail_free_run:
 196         free_page((unsigned long)vcpu->run);
 197 fail:
 198         return r;
 199 }
 200 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 201
 202 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 203 {
 204         kvm_arch_vcpu_uninit(vcpu);
 205         free_page((unsigned long)vcpu->run);
 206 }
 207 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 208
 209 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 210 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
 211 {
 212         return container_of(mn, struct kvm, mmu_notifier);
 213 }
 214
 215 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
 216                                              struct mm_struct *mm,
 217                                              unsigned long address)
 218 {
 219         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 220         int need_tlb_flush, idx;
 221
 222         /*
 223          * When ->invalidate_page runs, the linux pte has been zapped
 224          * already but the page is still allocated until
 225          * ->invalidate_page returns. So if we increase the sequence
 226          * here the kvm page fault will notice if the spte can't be
 227          * established because the page is going to be freed. If
 228          * instead the kvm page fault establishes the spte before
 229          * ->invalidate_page runs, kvm_unmap_hva will release it
 230          * before returning.
 231          *
 232          * The sequence increase only need to be seen at spin_unlock
 233          * time, and not at spin_lock time.
 234          *
 235          * Increasing the sequence after the spin_unlock would be
 236          * unsafe because the kvm page fault could then establish the
 237          * pte after kvm_unmap_hva returned, without noticing the page
 238          * is going to be freed.
 239          */
 240         idx = srcu_read_lock(&kvm->srcu);
 241         spin_lock(&kvm->mmu_lock);
 242         kvm->mmu_notifier_seq++;
 243         need_tlb_flush = kvm_unmap_hva(kvm, address);
 244         spin_unlock(&kvm->mmu_lock);
 245         srcu_read_unlock(&kvm->srcu, idx);
 246
 247         /* we've to flush the tlb before the pages can be freed */
 248         if (need_tlb_flush)
 249                 kvm_flush_remote_tlbs(kvm);
 250
 251 }
 252
 253 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
 254                                         struct mm_struct *mm,
 255                                         unsigned long address,
 256                                         pte_t pte)
 257 {
 258         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 259         int idx;
 260
 261         idx = srcu_read_lock(&kvm->srcu);
 262         spin_lock(&kvm->mmu_lock);
 263         kvm->mmu_notifier_seq++;
 264         kvm_set_spte_hva(kvm, address, pte);
 265         spin_unlock(&kvm->mmu_lock);
 266         srcu_read_unlock(&kvm->srcu, idx);
 267 }
 268
 269 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 270                                                     struct mm_struct *mm,
 271                                                     unsigned long start,
 272                                                     unsigned long end)
 273 {
 274         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 275         int need_tlb_flush = 0, idx;
 276
 277         idx = srcu_read_lock(&kvm->srcu);
 278         spin_lock(&kvm->mmu_lock);
 279         /*
 280          * The count increase must become visible at unlock time as no
 281          * spte can be established without taking the mmu_lock and
 282          * count is also read inside the mmu_lock critical section.
 283          */
 284         kvm->mmu_notifier_count++;
 285         for (; start < end; start += PAGE_SIZE)
 286                 need_tlb_flush |= kvm_unmap_hva(kvm, start);
 287         spin_unlock(&kvm->mmu_lock);
 288         srcu_read_unlock(&kvm->srcu, idx);
 289
 290         /* we've to flush the tlb before the pages can be freed */
 291         if (need_tlb_flush)
 292                 kvm_flush_remote_tlbs(kvm);
 293 }
 294
 295 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
 296                                                   struct mm_struct *mm,
 297                                                   unsigned long start,
 298                                                   unsigned long end)
 299 {
 300         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 301
 302         spin_lock(&kvm->mmu_lock);
 303         /*
 304          * This sequence increase will notify the kvm page fault that
 305          * the page that is going to be mapped in the spte could have
 306          * been freed.
 307          */
 308         kvm->mmu_notifier_seq++;
 309         /*
 310          * The above sequence increase must be visible before the
 311          * below count decrease but both values are read by the kvm
 312          * page fault under mmu_lock spinlock so we don't need to add
 313          * a smb_wmb() here in between the two.
 314          */
 315         kvm->mmu_notifier_count--;
 316         spin_unlock(&kvm->mmu_lock);
 317
 318         BUG_ON(kvm->mmu_notifier_count < 0);
 319 }
 320
 321 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
 322                                               struct mm_struct *mm,
 323                                               unsigned long address)
 324 {
 325         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 326         int young, idx;
 327
 328         idx = srcu_read_lock(&kvm->srcu);
 329         spin_lock(&kvm->mmu_lock);
 330         young = kvm_age_hva(kvm, address);
 331         spin_unlock(&kvm->mmu_lock);
 332         srcu_read_unlock(&kvm->srcu, idx);
 333
 334         if (young)
 335                 kvm_flush_remote_tlbs(kvm);
 336
 337         return young;
 338 }
 339
 340 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
 341                                      struct mm_struct *mm)
 342 {
 343         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 344         int idx;
 345
 346         idx = srcu_read_lock(&kvm->srcu);
 347         kvm_arch_flush_shadow(kvm);
 348         srcu_read_unlock(&kvm->srcu, idx);
 349 }
 350
 351 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 352         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
 353         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
 354         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
 355         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
 356         .change_pte             = kvm_mmu_notifier_change_pte,
 357         .release                = kvm_mmu_notifier_release,
 358 };
 359
 360 static int kvm_init_mmu_notifier(struct kvm *kvm)
 361 {
 362         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
 363         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
 364 }
 365
 366 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
 367
 368 static int kvm_init_mmu_notifier(struct kvm *kvm)
 369 {
 370         return 0;
 371 }
 372
 373 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
 374
 375 static struct kvm *kvm_create_vm(void)
 376 {
 377         int r = 0, i;
 378         struct kvm *kvm = kvm_arch_create_vm();
 379
 380         if (IS_ERR(kvm))
 381                 goto out;
 382
 383         r = hardware_enable_all();
 384         if (r)
 385                 goto out_err_nodisable;
 386
 387 #ifdef CONFIG_HAVE_KVM_IRQCHIP
 388         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
 389         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
 390 #endif
 391
 392         r = -ENOMEM;
 393         kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 394         if (!kvm->memslots)
 395                 goto out_err;
 396         if (init_srcu_struct(&kvm->srcu))
 397                 goto out_err;
 398         for (i = 0; i < KVM_NR_BUSES; i++) {
 399                 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
 400                                         GFP_KERNEL);
 401                 if (!kvm->buses[i]) {
 402                         cleanup_srcu_struct(&kvm->srcu);
 403                         goto out_err;
 404                 }
 405         }
 406
 407         r = kvm_init_mmu_notifier(kvm);
 408         if (r) {
 409                 cleanup_srcu_struct(&kvm->srcu);
 410                 goto out_err;
 411         }
 412
 413         kvm->mm = current->mm;
 414         atomic_inc(&kvm->mm->mm_count);
 415         spin_lock_init(&kvm->mmu_lock);
 416         raw_spin_lock_init(&kvm->requests_lock);
 417         kvm_eventfd_init(kvm);
 418         mutex_init(&kvm->lock);
 419         mutex_init(&kvm->irq_lock);
 420         mutex_init(&kvm->slots_lock);
 421         atomic_set(&kvm->users_count, 1);
 422         spin_lock(&kvm_lock);
 423         list_add(&kvm->vm_list, &vm_list);
 424         spin_unlock(&kvm_lock);
 425 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 426         kvm_coalesced_mmio_init(kvm);
 427 #endif
 428 out:
 429         return kvm;
 430
 431 out_err:
 432         hardware_disable_all();
 433 out_err_nodisable:
 434         for (i = 0; i < KVM_NR_BUSES; i++)
 435                 kfree(kvm->buses[i]);
 436         kfree(kvm->memslots);
 437         kfree(kvm);
 438         return ERR_PTR(r);
 439 }
 440
 441 /*
 442  * Free any memory in @free but not in @dont.
 443  */
 444 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 445                                   struct kvm_memory_slot *dont)
 446 {
 447         int i;
 448
 449         if (!dont || free->rmap != dont->rmap)
 450                 vfree(free->rmap);
 451
 452         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 453                 vfree(free->dirty_bitmap);
 454
 455
 456         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
 457                 if (!dont || free->lpage_info[i] != dont->lpage_info[i]) {
 458                         vfree(free->lpage_info[i]);
 459                         free->lpage_info[i] = NULL;
 460                 }
 461         }
 462
 463         free->npages = 0;
 464         free->dirty_bitmap = NULL;
 465         free->rmap = NULL;
 466 }
 467
 468 void kvm_free_physmem(struct kvm *kvm)
 469 {
 470         int i;
 471         struct kvm_memslots *slots = kvm->memslots;
 472
 473         for (i = 0; i < slots->nmemslots; ++i)
 474                 kvm_free_physmem_slot(&slots->memslots[i], NULL);
 475
 476         kfree(kvm->memslots);
 477 }
 478
 479 static void kvm_destroy_vm(struct kvm *kvm)
 480 {
 481         int i;
 482         struct mm_struct *mm = kvm->mm;
 483
 484         kvm_arch_sync_events(kvm);
 485         spin_lock(&kvm_lock);
 486         list_del(&kvm->vm_list);
 487         spin_unlock(&kvm_lock);
 488         kvm_free_irq_routing(kvm);
 489         for (i = 0; i < KVM_NR_BUSES; i++)
 490                 kvm_io_bus_destroy(kvm->buses[i]);
 491         kvm_coalesced_mmio_free(kvm);
 492 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 493         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
 494 #else
 495         kvm_arch_flush_shadow(kvm);
 496 #endif
 497         kvm_arch_destroy_vm(kvm);
 498         hardware_disable_all();
 499         mmdrop(mm);
 500 }
 501
 502 void kvm_get_kvm(struct kvm *kvm)
 503 {
 504         atomic_inc(&kvm->users_count);
 505 }
 506 EXPORT_SYMBOL_GPL(kvm_get_kvm);
 507
 508 void kvm_put_kvm(struct kvm *kvm)
 509 {
 510         if (atomic_dec_and_test(&kvm->users_count))
 511                 kvm_destroy_vm(kvm);
 512 }
 513 EXPORT_SYMBOL_GPL(kvm_put_kvm);
 514
 515
 516 static int kvm_vm_release(struct inode *inode, struct file *filp)
 517 {
 518         struct kvm *kvm = filp->private_data;
 519
 520         kvm_irqfd_release(kvm);
 521
 522         kvm_put_kvm(kvm);
 523         return 0;
 524 }
 525
 526 /*
 527  * Allocate some memory and give it an address in the guest physical address
 528  * space.
 529  *
 530  * Discontiguous memory is allowed, mostly for framebuffers.
 531  *
 532  * Must be called holding mmap_sem for write.
 533  */
 534 int __kvm_set_memory_region(struct kvm *kvm,
 535                             struct kvm_userspace_memory_region *mem,
 536                             int user_alloc)
 537 {
 538         int r, flush_shadow = 0;
 539         gfn_t base_gfn;
 540         unsigned long npages;
 541         unsigned long i;
 542         struct kvm_memory_slot *memslot;
 543         struct kvm_memory_slot old, new;
 544         struct kvm_memslots *slots, *old_memslots;
 545
 546         r = -EINVAL;
 547         /* General sanity checks */
 548         if (mem->memory_size & (PAGE_SIZE - 1))
 549                 goto out;
 550         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 551                 goto out;
 552         if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
 553                 goto out;
 554         if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
 555                 goto out;
 556         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 557                 goto out;
 558
 559         memslot = &kvm->memslots->memslots[mem->slot];
 560         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 561         npages = mem->memory_size >> PAGE_SHIFT;
 562
 563         if (!npages)
 564                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 565
 566         new = old = *memslot;
 567
 568         new.base_gfn = base_gfn;
 569         new.npages = npages;
 570         new.flags = mem->flags;
 571
 572         /* Disallow changing a memory slot's size. */
 573         r = -EINVAL;
 574         if (npages && old.npages && npages != old.npages)
 575                 goto out_free;
 576
 577         /* Check for overlaps */
 578         r = -EEXIST;
 579         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 580                 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
 581
 582                 if (s == memslot || !s->npages)
 583                         continue;
 584                 if (!((base_gfn + npages <= s->base_gfn) ||
 585                       (base_gfn >= s->base_gfn + s->npages)))
 586                         goto out_free;
 587         }
 588
 589         /* Free page dirty bitmap if unneeded */
 590         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 591                 new.dirty_bitmap = NULL;
 592
 593         r = -ENOMEM;
 594
 595         /* Allocate if a slot is being created */
 596 #ifndef CONFIG_S390
 597         if (npages && !new.rmap) {
 598                 new.rmap = vmalloc(npages * sizeof(struct page *));
 599
 600                 if (!new.rmap)
 601                         goto out_free;
 602
 603                 memset(new.rmap, 0, npages * sizeof(*new.rmap));
 604
 605                 new.user_alloc = user_alloc;
 606                 new.userspace_addr = mem->userspace_addr;
 607         }
 608         if (!npages)
 609                 goto skip_lpage;
 610
 611         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
 612                 unsigned long ugfn;
 613                 unsigned long j;
 614                 int lpages;
 615                 int level = i + 2;
 616
 617                 /* Avoid unused variable warning if no large pages */
 618                 (void)level;
 619
 620                 if (new.lpage_info[i])
 621                         continue;
 622
 623                 lpages = 1 + (base_gfn + npages - 1) /
 624                              KVM_PAGES_PER_HPAGE(level);
 625                 lpages -= base_gfn / KVM_PAGES_PER_HPAGE(level);
 626
 627                 new.lpage_info[i] = vmalloc(lpages * sizeof(*new.lpage_info[i]));
 628
 629                 if (!new.lpage_info[i])
 630                         goto out_free;
 631
 632                 memset(new.lpage_info[i], 0,
 633                        lpages * sizeof(*new.lpage_info[i]));
 634
 635                 if (base_gfn % KVM_PAGES_PER_HPAGE(level))
 636                         new.lpage_info[i][0].write_count = 1;
 637                 if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE(level))
 638                         new.lpage_info[i][lpages - 1].write_count = 1;
 639                 ugfn = new.userspace_addr >> PAGE_SHIFT;
 640                 /*
 641                  * If the gfn and userspace address are not aligned wrt each
 642                  * other, or if explicitly asked to, disable large page
 643                  * support for this slot
 644                  */
 645                 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
 646                     !largepages_enabled)
 647                         for (j = 0; j < lpages; ++j)
 648                                 new.lpage_info[i][j].write_count = 1;
 649         }
 650
 651 skip_lpage:
 652
 653         /* Allocate page dirty bitmap if needed */
 654         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 655                 unsigned long dirty_bytes = kvm_dirty_bitmap_bytes(&new);
 656
 657                 new.dirty_bitmap = vmalloc(dirty_bytes);
 658                 if (!new.dirty_bitmap)
 659                         goto out_free;
 660                 memset(new.dirty_bitmap, 0, dirty_bytes);
 661                 /* destroy any largepage mappings for dirty tracking */
 662                 if (old.npages)
 663                         flush_shadow = 1;
 664         }
 665 #else  /* not defined CONFIG_S390 */
 666         new.user_alloc = user_alloc;
 667         if (user_alloc)
 668                 new.userspace_addr = mem->userspace_addr;
 669 #endif /* not defined CONFIG_S390 */
 670
 671         if (!npages) {
 672                 r = -ENOMEM;
 673                 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 674                 if (!slots)
 675                         goto out_free;
 676                 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
 677                 if (mem->slot >= slots->nmemslots)
 678                         slots->nmemslots = mem->slot + 1;
 679                 slots->memslots[mem->slot].flags |= KVM_MEMSLOT_INVALID;
 680
 681                 old_memslots = kvm->memslots;
 682                 rcu_assign_pointer(kvm->memslots, slots);
 683                 synchronize_srcu_expedited(&kvm->srcu);
 684                 /* From this point no new shadow pages pointing to a deleted
 685                  * memslot will be created.
 686                  *
 687                  * validation of sp->gfn happens in:
 688                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
 689                  *      - kvm_is_visible_gfn (mmu_check_roots)
 690                  */
 691                 kvm_arch_flush_shadow(kvm);
 692                 kfree(old_memslots);
 693         }
 694
 695         r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
 696         if (r)
 697                 goto out_free;
 698
 699 #ifdef CONFIG_DMAR
 700         /* map the pages in iommu page table */
 701         if (npages) {
 702                 r = kvm_iommu_map_pages(kvm, &new);
 703                 if (r)
 704                         goto out_free;
 705         }
 706 #endif
 707
 708         r = -ENOMEM;
 709         slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 710         if (!slots)
 711                 goto out_free;
 712         memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
 713         if (mem->slot >= slots->nmemslots)
 714                 slots->nmemslots = mem->slot + 1;
 715
 716         /* actual memory is freed via old in kvm_free_physmem_slot below */
 717         if (!npages) {
 718                 new.rmap = NULL;
 719                 new.dirty_bitmap = NULL;
 720                 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i)
 721                         new.lpage_info[i] = NULL;
 722         }
 723
 724         slots->memslots[mem->slot] = new;
 725         old_memslots = kvm->memslots;
 726         rcu_assign_pointer(kvm->memslots, slots);
 727         synchronize_srcu_expedited(&kvm->srcu);
 728
 729         kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
 730
 731         kvm_free_physmem_slot(&old, &new);
 732         kfree(old_memslots);
 733
 734         if (flush_shadow)
 735                 kvm_arch_flush_shadow(kvm);
 736
 737         return 0;
 738
 739 out_free:
 740         kvm_free_physmem_slot(&new, &old);
 741 out:
 742         return r;
 743
 744 }
 745 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 746
 747 int kvm_set_memory_region(struct kvm *kvm,
 748                           struct kvm_userspace_memory_region *mem,
 749                           int user_alloc)
 750 {
 751         int r;
 752
 753         mutex_lock(&kvm->slots_lock);
 754         r = __kvm_set_memory_region(kvm, mem, user_alloc);
 755         mutex_unlock(&kvm->slots_lock);
 756         return r;
 757 }
 758 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 759
 760 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 761                                    struct
 762                                    kvm_userspace_memory_region *mem,
 763                                    int user_alloc)
 764 {
 765         if (mem->slot >= KVM_MEMORY_SLOTS)
 766                 return -EINVAL;
 767         return kvm_set_memory_region(kvm, mem, user_alloc);
 768 }
 769
 770 int kvm_get_dirty_log(struct kvm *kvm,
 771                         struct kvm_dirty_log *log, int *is_dirty)
 772 {
 773         struct kvm_memory_slot *memslot;
 774         int r, i;
 775         unsigned long n;
 776         unsigned long any = 0;
 777
 778         r = -EINVAL;
 779         if (log->slot >= KVM_MEMORY_SLOTS)
 780                 goto out;
 781
 782         memslot = &kvm->memslots->memslots[log->slot];
 783         r = -ENOENT;
 784         if (!memslot->dirty_bitmap)
 785                 goto out;
 786
 787         n = kvm_dirty_bitmap_bytes(memslot);
 788
 789         for (i = 0; !any && i < n/sizeof(long); ++i)
 790                 any = memslot->dirty_bitmap[i];
 791
 792         r = -EFAULT;
 793         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 794                 goto out;
 795
 796         if (any)
 797                 *is_dirty = 1;
 798
 799         r = 0;
 800 out:
 801         return r;
 802 }
 803
 804 void kvm_disable_largepages(void)
 805 {
 806         largepages_enabled = false;
 807 }
 808 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
 809
 810 int is_error_page(struct page *page)
 811 {
 812         return page == bad_page;
 813 }
 814 EXPORT_SYMBOL_GPL(is_error_page);
 815
 816 int is_error_pfn(pfn_t pfn)
 817 {
 818         return pfn == bad_pfn;
 819 }
 820 EXPORT_SYMBOL_GPL(is_error_pfn);
 821
 822 static inline unsigned long bad_hva(void)
 823 {
 824         return PAGE_OFFSET;
 825 }
 826
 827 int kvm_is_error_hva(unsigned long addr)
 828 {
 829         return addr == bad_hva();
 830 }
 831 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
 832
 833 struct kvm_memory_slot *gfn_to_memslot_unaliased(struct kvm *kvm, gfn_t gfn)
 834 {
 835         int i;
 836         struct kvm_memslots *slots = rcu_dereference(kvm->memslots);
 837
 838         for (i = 0; i < slots->nmemslots; ++i) {
 839                 struct kvm_memory_slot *memslot = &slots->memslots[i];
 840
 841                 if (gfn >= memslot->base_gfn
 842                     && gfn < memslot->base_gfn + memslot->npages)
 843                         return memslot;
 844         }
 845         return NULL;
 846 }
 847 EXPORT_SYMBOL_GPL(gfn_to_memslot_unaliased);
 848
 849 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 850 {
 851         gfn = unalias_gfn(kvm, gfn);
 852         return gfn_to_memslot_unaliased(kvm, gfn);
 853 }
 854
 855 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
 856 {
 857         int i;
 858         struct kvm_memslots *slots = rcu_dereference(kvm->memslots);
 859
 860         gfn = unalias_gfn_instantiation(kvm, gfn);
 861         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 862                 struct kvm_memory_slot *memslot = &slots->memslots[i];
 863
 864                 if (memslot->flags & KVM_MEMSLOT_INVALID)
 865                         continue;
 866
 867                 if (gfn >= memslot->base_gfn
 868                     && gfn < memslot->base_gfn + memslot->npages)
 869                         return 1;
 870         }
 871         return 0;
 872 }
 873 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
 874
 875 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
 876 {
 877         struct vm_area_struct *vma;
 878         unsigned long addr, size;
 879
 880         size = PAGE_SIZE;
 881
 882         addr = gfn_to_hva(kvm, gfn);
 883         if (kvm_is_error_hva(addr))
 884                 return PAGE_SIZE;
 885
 886         down_read(&current->mm->mmap_sem);
 887         vma = find_vma(current->mm, addr);
 888         if (!vma)
 889                 goto out;
 890
 891         size = vma_kernel_pagesize(vma);
 892
 893 out:
 894         up_read(&current->mm->mmap_sem);
 895
 896         return size;
 897 }
 898
 899 int memslot_id(struct kvm *kvm, gfn_t gfn)
 900 {
 901         int i;
 902         struct kvm_memslots *slots = rcu_dereference(kvm->memslots);
 903         struct kvm_memory_slot *memslot = NULL;
 904
 905         gfn = unalias_gfn(kvm, gfn);
 906         for (i = 0; i < slots->nmemslots; ++i) {
 907                 memslot = &slots->memslots[i];
 908
 909                 if (gfn >= memslot->base_gfn
 910                     && gfn < memslot->base_gfn + memslot->npages)
 911                         break;
 912         }
 913
 914         return memslot - slots->memslots;
 915 }
 916
 917 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
 918 {
 919         struct kvm_memory_slot *slot;
 920
 921         gfn = unalias_gfn_instantiation(kvm, gfn);
 922         slot = gfn_to_memslot_unaliased(kvm, gfn);
 923         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
 924                 return bad_hva();
 925         return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
 926 }
 927 EXPORT_SYMBOL_GPL(gfn_to_hva);
 928
 929 static pfn_t hva_to_pfn(struct kvm *kvm, unsigned long addr)
 930 {
 931         struct page *page[1];
 932         int npages;
 933         pfn_t pfn;
 934
 935         might_sleep();
 936
 937         npages = get_user_pages_fast(addr, 1, 1, page);
 938
 939         if (unlikely(npages != 1)) {
 940                 struct vm_area_struct *vma;
 941
 942                 down_read(&current->mm->mmap_sem);
 943                 vma = find_vma(current->mm, addr);
 944
 945                 if (vma == NULL || addr < vma->vm_start ||
 946                     !(vma->vm_flags & VM_PFNMAP)) {
 947                         up_read(&current->mm->mmap_sem);
 948                         get_page(bad_page);
 949                         return page_to_pfn(bad_page);
 950                 }
 951
 952                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
 953                 up_read(&current->mm->mmap_sem);
 954                 BUG_ON(!kvm_is_mmio_pfn(pfn));
 955         } else
 956                 pfn = page_to_pfn(page[0]);
 957
 958         return pfn;
 959 }
 960
 961 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
 962 {
 963         unsigned long addr;
 964
 965         addr = gfn_to_hva(kvm, gfn);
 966         if (kvm_is_error_hva(addr)) {
 967                 get_page(bad_page);
 968                 return page_to_pfn(bad_page);
 969         }
 970
 971         return hva_to_pfn(kvm, addr);
 972 }
 973 EXPORT_SYMBOL_GPL(gfn_to_pfn);
 974
 975 static unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
 976 {
 977         return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
 978 }
 979
 980 pfn_t gfn_to_pfn_memslot(struct kvm *kvm,
 981                          struct kvm_memory_slot *slot, gfn_t gfn)
 982 {
 983         unsigned long addr = gfn_to_hva_memslot(slot, gfn);
 984         return hva_to_pfn(kvm, addr);
 985 }
 986
 987 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
 988 {
 989         pfn_t pfn;
 990
 991         pfn = gfn_to_pfn(kvm, gfn);
 992         if (!kvm_is_mmio_pfn(pfn))
 993                 return pfn_to_page(pfn);
 994
 995         WARN_ON(kvm_is_mmio_pfn(pfn));
 996
 997         get_page(bad_page);
 998         return bad_page;
 999 }
1000
1001 EXPORT_SYMBOL_GPL(gfn_to_page);
1002
1003 void kvm_release_page_clean(struct page *page)
1004 {
1005         kvm_release_pfn_clean(page_to_pfn(page));
1006 }
1007 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1008
1009 void kvm_release_pfn_clean(pfn_t pfn)
1010 {
1011         if (!kvm_is_mmio_pfn(pfn))
1012                 put_page(pfn_to_page(pfn));
1013 }
1014 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1015
1016 void kvm_release_page_dirty(struct page *page)
1017 {
1018         kvm_release_pfn_dirty(page_to_pfn(page));
1019 }
1020 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1021
1022 void kvm_release_pfn_dirty(pfn_t pfn)
1023 {
1024         kvm_set_pfn_dirty(pfn);
1025         kvm_release_pfn_clean(pfn);
1026 }
1027 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1028
1029 void kvm_set_page_dirty(struct page *page)
1030 {
1031         kvm_set_pfn_dirty(page_to_pfn(page));
1032 }
1033 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1034
1035 void kvm_set_pfn_dirty(pfn_t pfn)
1036 {
1037         if (!kvm_is_mmio_pfn(pfn)) {
1038                 struct page *page = pfn_to_page(pfn);
1039                 if (!PageReserved(page))
1040                         SetPageDirty(page);
1041         }
1042 }
1043 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1044
1045 void kvm_set_pfn_accessed(pfn_t pfn)
1046 {
1047         if (!kvm_is_mmio_pfn(pfn))
1048                 mark_page_accessed(pfn_to_page(pfn));
1049 }
1050 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1051
1052 void kvm_get_pfn(pfn_t pfn)
1053 {
1054         if (!kvm_is_mmio_pfn(pfn))
1055                 get_page(pfn_to_page(pfn));
1056 }
1057 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1058
1059 static int next_segment(unsigned long len, int offset)
1060 {
1061         if (len > PAGE_SIZE - offset)
1062                 return PAGE_SIZE - offset;
1063         else
1064                 return len;
1065 }
1066
1067 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1068                         int len)
1069 {
1070         int r;
1071         unsigned long addr;
1072
1073         addr = gfn_to_hva(kvm, gfn);
1074         if (kvm_is_error_hva(addr))
1075                 return -EFAULT;
1076         r = copy_from_user(data, (void __user *)addr + offset, len);
1077         if (r)
1078                 return -EFAULT;
1079         return 0;
1080 }
1081 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1082
1083 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1084 {
1085         gfn_t gfn = gpa >> PAGE_SHIFT;
1086         int seg;
1087         int offset = offset_in_page(gpa);
1088         int ret;
1089
1090         while ((seg = next_segment(len, offset)) != 0) {
1091                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1092                 if (ret < 0)
1093                         return ret;
1094                 offset = 0;
1095                 len -= seg;
1096                 data += seg;
1097                 ++gfn;
1098         }
1099         return 0;
1100 }
1101 EXPORT_SYMBOL_GPL(kvm_read_guest);
1102
1103 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1104                           unsigned long len)
1105 {
1106         int r;
1107         unsigned long addr;
1108         gfn_t gfn = gpa >> PAGE_SHIFT;
1109         int offset = offset_in_page(gpa);
1110
1111         addr = gfn_to_hva(kvm, gfn);
1112         if (kvm_is_error_hva(addr))
1113                 return -EFAULT;
1114         pagefault_disable();
1115         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1116         pagefault_enable();
1117         if (r)
1118                 return -EFAULT;
1119         return 0;
1120 }
1121 EXPORT_SYMBOL(kvm_read_guest_atomic);
1122
1123 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1124                          int offset, int len)
1125 {
1126         int r;
1127         unsigned long addr;
1128
1129         addr = gfn_to_hva(kvm, gfn);
1130         if (kvm_is_error_hva(addr))
1131                 return -EFAULT;
1132         r = copy_to_user((void __user *)addr + offset, data, len);
1133         if (r)
1134                 return -EFAULT;
1135         mark_page_dirty(kvm, gfn);
1136         return 0;
1137 }
1138 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1139
1140 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1141                     unsigned long len)
1142 {
1143         gfn_t gfn = gpa >> PAGE_SHIFT;
1144         int seg;
1145         int offset = offset_in_page(gpa);
1146         int ret;
1147
1148         while ((seg = next_segment(len, offset)) != 0) {
1149                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1150                 if (ret < 0)
1151                         return ret;
1152                 offset = 0;
1153                 len -= seg;
1154                 data += seg;
1155                 ++gfn;
1156         }
1157         return 0;
1158 }
1159
1160 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1161 {
1162         return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
1163 }
1164 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1165
1166 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1167 {
1168         gfn_t gfn = gpa >> PAGE_SHIFT;
1169         int seg;
1170         int offset = offset_in_page(gpa);
1171         int ret;
1172
1173         while ((seg = next_segment(len, offset)) != 0) {
1174                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1175                 if (ret < 0)
1176                         return ret;
1177                 offset = 0;
1178                 len -= seg;
1179                 ++gfn;
1180         }
1181         return 0;
1182 }
1183 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1184
1185 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1186 {
1187         struct kvm_memory_slot *memslot;
1188
1189         gfn = unalias_gfn(kvm, gfn);
1190         memslot = gfn_to_memslot_unaliased(kvm, gfn);
1191         if (memslot && memslot->dirty_bitmap) {
1192                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1193                 unsigned long *p = memslot->dirty_bitmap +
1194                                         rel_gfn / BITS_PER_LONG;
1195                 int offset = rel_gfn % BITS_PER_LONG;
1196
1197                 /* avoid RMW */
1198                 if (!generic_test_le_bit(offset, p))
1199                         generic___set_le_bit(offset, p);
1200         }
1201 }
1202
1203 /*
1204  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1205  */
1206 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1207 {
1208         DEFINE_WAIT(wait);
1209
1210         for (;;) {
1211                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1212
1213                 if (kvm_arch_vcpu_runnable(vcpu)) {
1214                         set_bit(KVM_REQ_UNHALT, &vcpu->requests);
1215                         break;
1216                 }
1217                 if (kvm_cpu_has_pending_timer(vcpu))
1218                         break;
1219                 if (signal_pending(current))
1220                         break;
1221
1222                 schedule();
1223         }
1224
1225         finish_wait(&vcpu->wq, &wait);
1226 }
1227
1228 void kvm_resched(struct kvm_vcpu *vcpu)
1229 {
1230         if (!need_resched())
1231                 return;
1232         cond_resched();
1233 }
1234 EXPORT_SYMBOL_GPL(kvm_resched);
1235
1236 void kvm_vcpu_on_spin(struct kvm_vcpu *vcpu)
1237 {
1238         ktime_t expires;
1239         DEFINE_WAIT(wait);
1240
1241         prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1242
1243         /* Sleep for 100 us, and hope lock-holder got scheduled */
1244         expires = ktime_add_ns(ktime_get(), 100000UL);
1245         schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
1246
1247         finish_wait(&vcpu->wq, &wait);
1248 }
1249 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1250
1251 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1252 {
1253         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1254         struct page *page;
1255
1256         if (vmf->pgoff == 0)
1257                 page = virt_to_page(vcpu->run);
1258 #ifdef CONFIG_X86
1259         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1260                 page = virt_to_page(vcpu->arch.pio_data);
1261 #endif
1262 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1263         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1264                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1265 #endif
1266         else
1267                 return VM_FAULT_SIGBUS;
1268         get_page(page);
1269         vmf->page = page;
1270         return 0;
1271 }
1272
1273 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1274         .fault = kvm_vcpu_fault,
1275 };
1276
1277 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1278 {
1279         vma->vm_ops = &kvm_vcpu_vm_ops;
1280         return 0;
1281 }
1282
1283 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1284 {
1285         struct kvm_vcpu *vcpu = filp->private_data;
1286
1287         kvm_put_kvm(vcpu->kvm);
1288         return 0;
1289 }
1290
1291 static struct file_operations kvm_vcpu_fops = {
1292         .release        = kvm_vcpu_release,
1293         .unlocked_ioctl = kvm_vcpu_ioctl,
1294         .compat_ioctl   = kvm_vcpu_ioctl,
1295         .mmap           = kvm_vcpu_mmap,
1296 };
1297
1298 /*
1299  * Allocates an inode for the vcpu.
1300  */
1301 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1302 {
1303         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1304 }
1305
1306 /*
1307  * Creates some virtual cpus.  Good luck creating more than one.
1308  */
1309 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1310 {
1311         int r;
1312         struct kvm_vcpu *vcpu, *v;
1313
1314         vcpu = kvm_arch_vcpu_create(kvm, id);
1315         if (IS_ERR(vcpu))
1316                 return PTR_ERR(vcpu);
1317
1318         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1319
1320         r = kvm_arch_vcpu_setup(vcpu);
1321         if (r)
1322                 return r;
1323
1324         mutex_lock(&kvm->lock);
1325         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1326                 r = -EINVAL;
1327                 goto vcpu_destroy;
1328         }
1329
1330         kvm_for_each_vcpu(r, v, kvm)
1331                 if (v->vcpu_id == id) {
1332                         r = -EEXIST;
1333                         goto vcpu_destroy;
1334                 }
1335
1336         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1337
1338         /* Now it's all set up, let userspace reach it */
1339         kvm_get_kvm(kvm);
1340         r = create_vcpu_fd(vcpu);
1341         if (r < 0) {
1342                 kvm_put_kvm(kvm);
1343                 goto vcpu_destroy;
1344         }
1345
1346         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1347         smp_wmb();
1348         atomic_inc(&kvm->online_vcpus);
1349
1350 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1351         if (kvm->bsp_vcpu_id == id)
1352                 kvm->bsp_vcpu = vcpu;
1353 #endif
1354         mutex_unlock(&kvm->lock);
1355         return r;
1356
1357 vcpu_destroy:
1358         mutex_unlock(&kvm->lock);
1359         kvm_arch_vcpu_destroy(vcpu);
1360         return r;
1361 }
1362
1363 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1364 {
1365         if (sigset) {
1366                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1367                 vcpu->sigset_active = 1;
1368                 vcpu->sigset = *sigset;
1369         } else
1370                 vcpu->sigset_active = 0;
1371         return 0;
1372 }
1373
1374 static long kvm_vcpu_ioctl(struct file *filp,
1375                            unsigned int ioctl, unsigned long arg)
1376 {
1377         struct kvm_vcpu *vcpu = filp->private_data;
1378         void __user *argp = (void __user *)arg;
1379         int r;
1380         struct kvm_fpu *fpu = NULL;
1381         struct kvm_sregs *kvm_sregs = NULL;
1382
1383         if (vcpu->kvm->mm != current->mm)
1384                 return -EIO;
1385         switch (ioctl) {
1386         case KVM_RUN:
1387                 r = -EINVAL;
1388                 if (arg)
1389                         goto out;
1390                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1391                 break;
1392         case KVM_GET_REGS: {
1393                 struct kvm_regs *kvm_regs;
1394
1395                 r = -ENOMEM;
1396                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1397                 if (!kvm_regs)
1398                         goto out;
1399                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1400                 if (r)
1401                         goto out_free1;
1402                 r = -EFAULT;
1403                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1404                         goto out_free1;
1405                 r = 0;
1406 out_free1:
1407                 kfree(kvm_regs);
1408                 break;
1409         }
1410         case KVM_SET_REGS: {
1411                 struct kvm_regs *kvm_regs;
1412
1413                 r = -ENOMEM;
1414                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1415                 if (!kvm_regs)
1416                         goto out;
1417                 r = -EFAULT;
1418                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1419                         goto out_free2;
1420                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1421                 if (r)
1422                         goto out_free2;
1423                 r = 0;
1424 out_free2:
1425                 kfree(kvm_regs);
1426                 break;
1427         }
1428         case KVM_GET_SREGS: {
1429                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1430                 r = -ENOMEM;
1431                 if (!kvm_sregs)
1432                         goto out;
1433                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1434                 if (r)
1435                         goto out;
1436                 r = -EFAULT;
1437                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1438                         goto out;
1439                 r = 0;
1440                 break;
1441         }
1442         case KVM_SET_SREGS: {
1443                 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1444                 r = -ENOMEM;
1445                 if (!kvm_sregs)
1446                         goto out;
1447                 r = -EFAULT;
1448                 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1449                         goto out;
1450                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1451                 if (r)
1452                         goto out;
1453                 r = 0;
1454                 break;
1455         }
1456         case KVM_GET_MP_STATE: {
1457                 struct kvm_mp_state mp_state;
1458
1459                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1460                 if (r)
1461                         goto out;
1462                 r = -EFAULT;
1463                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1464                         goto out;
1465                 r = 0;
1466                 break;
1467         }
1468         case KVM_SET_MP_STATE: {
1469                 struct kvm_mp_state mp_state;
1470
1471                 r = -EFAULT;
1472                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1473                         goto out;
1474                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1475                 if (r)
1476                         goto out;
1477                 r = 0;
1478                 break;
1479         }
1480         case KVM_TRANSLATE: {
1481                 struct kvm_translation tr;
1482
1483                 r = -EFAULT;
1484                 if (copy_from_user(&tr, argp, sizeof tr))
1485                         goto out;
1486                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1487                 if (r)
1488                         goto out;
1489                 r = -EFAULT;
1490                 if (copy_to_user(argp, &tr, sizeof tr))
1491                         goto out;
1492                 r = 0;
1493                 break;
1494         }
1495         case KVM_SET_GUEST_DEBUG: {
1496                 struct kvm_guest_debug dbg;
1497
1498                 r = -EFAULT;
1499                 if (copy_from_user(&dbg, argp, sizeof dbg))
1500                         goto out;
1501                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1502                 if (r)
1503                         goto out;
1504                 r = 0;
1505                 break;
1506         }
1507         case KVM_SET_SIGNAL_MASK: {
1508                 struct kvm_signal_mask __user *sigmask_arg = argp;
1509                 struct kvm_signal_mask kvm_sigmask;
1510                 sigset_t sigset, *p;
1511
1512                 p = NULL;
1513                 if (argp) {
1514                         r = -EFAULT;
1515                         if (copy_from_user(&kvm_sigmask, argp,
1516                                            sizeof kvm_sigmask))
1517                                 goto out;
1518                         r = -EINVAL;
1519                         if (kvm_sigmask.len != sizeof sigset)
1520                                 goto out;
1521                         r = -EFAULT;
1522                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1523                                            sizeof sigset))
1524                                 goto out;
1525                         p = &sigset;
1526                 }
1527                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1528                 break;
1529         }
1530         case KVM_GET_FPU: {
1531                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1532                 r = -ENOMEM;
1533                 if (!fpu)
1534                         goto out;
1535                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1536                 if (r)
1537                         goto out;
1538                 r = -EFAULT;
1539                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1540                         goto out;
1541                 r = 0;
1542                 break;
1543         }
1544         case KVM_SET_FPU: {
1545                 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1546                 r = -ENOMEM;
1547                 if (!fpu)
1548                         goto out;
1549                 r = -EFAULT;
1550                 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1551                         goto out;
1552                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1553                 if (r)
1554                         goto out;
1555                 r = 0;
1556                 break;
1557         }
1558         default:
1559                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1560         }
1561 out:
1562         kfree(fpu);
1563         kfree(kvm_sregs);
1564         return r;
1565 }
1566
1567 static long kvm_vm_ioctl(struct file *filp,
1568                            unsigned int ioctl, unsigned long arg)
1569 {
1570         struct kvm *kvm = filp->private_data;
1571         void __user *argp = (void __user *)arg;
1572         int r;
1573
1574         if (kvm->mm != current->mm)
1575                 return -EIO;
1576         switch (ioctl) {
1577         case KVM_CREATE_VCPU:
1578                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1579                 if (r < 0)
1580                         goto out;
1581                 break;
1582         case KVM_SET_USER_MEMORY_REGION: {
1583                 struct kvm_userspace_memory_region kvm_userspace_mem;
1584
1585                 r = -EFAULT;
1586                 if (copy_from_user(&kvm_userspace_mem, argp,
1587                                                 sizeof kvm_userspace_mem))
1588                         goto out;
1589
1590                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1591                 if (r)
1592                         goto out;
1593                 break;
1594         }
1595         case KVM_GET_DIRTY_LOG: {
1596                 struct kvm_dirty_log log;
1597
1598                 r = -EFAULT;
1599                 if (copy_from_user(&log, argp, sizeof log))
1600                         goto out;
1601                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1602                 if (r)
1603                         goto out;
1604                 break;
1605         }
1606 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1607         case KVM_REGISTER_COALESCED_MMIO: {
1608                 struct kvm_coalesced_mmio_zone zone;
1609                 r = -EFAULT;
1610                 if (copy_from_user(&zone, argp, sizeof zone))
1611                         goto out;
1612                 r = -ENXIO;
1613                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1614                 if (r)
1615                         goto out;
1616                 r = 0;
1617                 break;
1618         }
1619         case KVM_UNREGISTER_COALESCED_MMIO: {
1620                 struct kvm_coalesced_mmio_zone zone;
1621                 r = -EFAULT;
1622                 if (copy_from_user(&zone, argp, sizeof zone))
1623                         goto out;
1624                 r = -ENXIO;
1625                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1626                 if (r)
1627                         goto out;
1628                 r = 0;
1629                 break;
1630         }
1631 #endif
1632         case KVM_IRQFD: {
1633                 struct kvm_irqfd data;
1634
1635                 r = -EFAULT;
1636                 if (copy_from_user(&data, argp, sizeof data))
1637                         goto out;
1638                 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
1639                 break;
1640         }
1641         case KVM_IOEVENTFD: {
1642                 struct kvm_ioeventfd data;
1643
1644                 r = -EFAULT;
1645                 if (copy_from_user(&data, argp, sizeof data))
1646                         goto out;
1647                 r = kvm_ioeventfd(kvm, &data);
1648                 break;
1649         }
1650 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1651         case KVM_SET_BOOT_CPU_ID:
1652                 r = 0;
1653                 mutex_lock(&kvm->lock);
1654                 if (atomic_read(&kvm->online_vcpus) != 0)
1655                         r = -EBUSY;
1656                 else
1657                         kvm->bsp_vcpu_id = arg;
1658                 mutex_unlock(&kvm->lock);
1659                 break;
1660 #endif
1661         default:
1662                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1663                 if (r == -ENOTTY)
1664                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
1665         }
1666 out:
1667         return r;
1668 }
1669
1670 #ifdef CONFIG_COMPAT
1671 struct compat_kvm_dirty_log {
1672         __u32 slot;
1673         __u32 padding1;
1674         union {
1675                 compat_uptr_t dirty_bitmap; /* one bit per page */
1676                 __u64 padding2;
1677         };
1678 };
1679
1680 static long kvm_vm_compat_ioctl(struct file *filp,
1681                            unsigned int ioctl, unsigned long arg)
1682 {
1683         struct kvm *kvm = filp->private_data;
1684         int r;
1685
1686         if (kvm->mm != current->mm)
1687                 return -EIO;
1688         switch (ioctl) {
1689         case KVM_GET_DIRTY_LOG: {
1690                 struct compat_kvm_dirty_log compat_log;
1691                 struct kvm_dirty_log log;
1692
1693                 r = -EFAULT;
1694                 if (copy_from_user(&compat_log, (void __user *)arg,
1695                                    sizeof(compat_log)))
1696                         goto out;
1697                 log.slot         = compat_log.slot;
1698                 log.padding1     = compat_log.padding1;
1699                 log.padding2     = compat_log.padding2;
1700                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
1701
1702                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1703                 if (r)
1704                         goto out;
1705                 break;
1706         }
1707         default:
1708                 r = kvm_vm_ioctl(filp, ioctl, arg);
1709         }
1710
1711 out:
1712         return r;
1713 }
1714 #endif
1715
1716 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1717 {
1718         struct page *page[1];
1719         unsigned long addr;
1720         int npages;
1721         gfn_t gfn = vmf->pgoff;
1722         struct kvm *kvm = vma->vm_file->private_data;
1723
1724         addr = gfn_to_hva(kvm, gfn);
1725         if (kvm_is_error_hva(addr))
1726                 return VM_FAULT_SIGBUS;
1727
1728         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
1729                                 NULL);
1730         if (unlikely(npages != 1))
1731                 return VM_FAULT_SIGBUS;
1732
1733         vmf->page = page[0];
1734         return 0;
1735 }
1736
1737 static const struct vm_operations_struct kvm_vm_vm_ops = {
1738         .fault = kvm_vm_fault,
1739 };
1740
1741 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1742 {
1743         vma->vm_ops = &kvm_vm_vm_ops;
1744         return 0;
1745 }
1746
1747 static struct file_operations kvm_vm_fops = {
1748         .release        = kvm_vm_release,
1749         .unlocked_ioctl = kvm_vm_ioctl,
1750 #ifdef CONFIG_COMPAT
1751         .compat_ioctl   = kvm_vm_compat_ioctl,
1752 #endif
1753         .mmap           = kvm_vm_mmap,
1754 };
1755
1756 static int kvm_dev_ioctl_create_vm(void)
1757 {
1758         int fd;
1759         struct kvm *kvm;
1760
1761         kvm = kvm_create_vm();
1762         if (IS_ERR(kvm))
1763                 return PTR_ERR(kvm);
1764         fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
1765         if (fd < 0)
1766                 kvm_put_kvm(kvm);
1767
1768         return fd;
1769 }
1770
1771 static long kvm_dev_ioctl_check_extension_generic(long arg)
1772 {
1773         switch (arg) {
1774         case KVM_CAP_USER_MEMORY:
1775         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
1776         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
1777 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1778         case KVM_CAP_SET_BOOT_CPU_ID:
1779 #endif
1780         case KVM_CAP_INTERNAL_ERROR_DATA:
1781                 return 1;
1782 #ifdef CONFIG_HAVE_KVM_IRQCHIP
1783         case KVM_CAP_IRQ_ROUTING:
1784                 return KVM_MAX_IRQ_ROUTES;
1785 #endif
1786         default:
1787                 break;
1788         }
1789         return kvm_dev_ioctl_check_extension(arg);
1790 }
1791
1792 static long kvm_dev_ioctl(struct file *filp,
1793                           unsigned int ioctl, unsigned long arg)
1794 {
1795         long r = -EINVAL;
1796
1797         switch (ioctl) {
1798         case KVM_GET_API_VERSION:
1799                 r = -EINVAL;
1800                 if (arg)
1801                         goto out;
1802                 r = KVM_API_VERSION;
1803                 break;
1804         case KVM_CREATE_VM:
1805                 r = -EINVAL;
1806                 if (arg)
1807                         goto out;
1808                 r = kvm_dev_ioctl_create_vm();
1809                 break;
1810         case KVM_CHECK_EXTENSION:
1811                 r = kvm_dev_ioctl_check_extension_generic(arg);
1812                 break;
1813         case KVM_GET_VCPU_MMAP_SIZE:
1814                 r = -EINVAL;
1815                 if (arg)
1816                         goto out;
1817                 r = PAGE_SIZE;     /* struct kvm_run */
1818 #ifdef CONFIG_X86
1819                 r += PAGE_SIZE;    /* pio data page */
1820 #endif
1821 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1822                 r += PAGE_SIZE;    /* coalesced mmio ring page */
1823 #endif
1824                 break;
1825         case KVM_TRACE_ENABLE:
1826         case KVM_TRACE_PAUSE:
1827         case KVM_TRACE_DISABLE:
1828                 r = -EOPNOTSUPP;
1829                 break;
1830         default:
1831                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
1832         }
1833 out:
1834         return r;
1835 }
1836
1837 static struct file_operations kvm_chardev_ops = {
1838         .unlocked_ioctl = kvm_dev_ioctl,
1839         .compat_ioctl   = kvm_dev_ioctl,
1840 };
1841
1842 static struct miscdevice kvm_dev = {
1843         KVM_MINOR,
1844         "kvm",
1845         &kvm_chardev_ops,
1846 };
1847
1848 static void hardware_enable(void *junk)
1849 {
1850         int cpu = raw_smp_processor_id();
1851         int r;
1852
1853         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
1854                 return;
1855
1856         cpumask_set_cpu(cpu, cpus_hardware_enabled);
1857
1858         r = kvm_arch_hardware_enable(NULL);
1859
1860         if (r) {
1861                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
1862                 atomic_inc(&hardware_enable_failed);
1863                 printk(KERN_INFO "kvm: enabling virtualization on "
1864                                  "CPU%d failed\n", cpu);
1865         }
1866 }
1867
1868 static void hardware_disable(void *junk)
1869 {
1870         int cpu = raw_smp_processor_id();
1871
1872         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
1873                 return;
1874         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
1875         kvm_arch_hardware_disable(NULL);
1876 }
1877
1878 static void hardware_disable_all_nolock(void)
1879 {
1880         BUG_ON(!kvm_usage_count);
1881
1882         kvm_usage_count--;
1883         if (!kvm_usage_count)
1884                 on_each_cpu(hardware_disable, NULL, 1);
1885 }
1886
1887 static void hardware_disable_all(void)
1888 {
1889         spin_lock(&kvm_lock);
1890         hardware_disable_all_nolock();
1891         spin_unlock(&kvm_lock);
1892 }
1893
1894 static int hardware_enable_all(void)
1895 {
1896         int r = 0;
1897
1898         spin_lock(&kvm_lock);
1899
1900         kvm_usage_count++;
1901         if (kvm_usage_count == 1) {
1902                 atomic_set(&hardware_enable_failed, 0);
1903                 on_each_cpu(hardware_enable, NULL, 1);
1904
1905                 if (atomic_read(&hardware_enable_failed)) {
1906                         hardware_disable_all_nolock();
1907                         r = -EBUSY;
1908                 }
1909         }
1910
1911         spin_unlock(&kvm_lock);
1912
1913         return r;
1914 }
1915
1916 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
1917                            void *v)
1918 {
1919         int cpu = (long)v;
1920
1921         if (!kvm_usage_count)
1922                 return NOTIFY_OK;
1923
1924         val &= ~CPU_TASKS_FROZEN;
1925         switch (val) {
1926         case CPU_DYING:
1927                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1928                        cpu);
1929                 hardware_disable(NULL);
1930                 break;
1931         case CPU_UP_CANCELED:
1932                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1933                        cpu);
1934                 smp_call_function_single(cpu, hardware_disable, NULL, 1);
1935                 break;
1936         case CPU_ONLINE:
1937                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
1938                        cpu);
1939                 smp_call_function_single(cpu, hardware_enable, NULL, 1);
1940                 break;
1941         }
1942         return NOTIFY_OK;
1943 }
1944
1945
1946 asmlinkage void kvm_handle_fault_on_reboot(void)
1947 {
1948         if (kvm_rebooting)
1949                 /* spin while reset goes on */
1950                 while (true)
1951                         ;
1952         /* Fault while not rebooting.  We want the trace. */
1953         BUG();
1954 }
1955 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
1956
1957 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
1958                       void *v)
1959 {
1960         /*
1961          * Some (well, at least mine) BIOSes hang on reboot if
1962          * in vmx root mode.
1963          *
1964          * And Intel TXT required VMX off for all cpu when system shutdown.
1965          */
1966         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
1967         kvm_rebooting = true;
1968         on_each_cpu(hardware_disable, NULL, 1);
1969         return NOTIFY_OK;
1970 }
1971
1972 static struct notifier_block kvm_reboot_notifier = {
1973         .notifier_call = kvm_reboot,
1974         .priority = 0,
1975 };
1976
1977 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
1978 {
1979         int i;
1980
1981         for (i = 0; i < bus->dev_count; i++) {
1982                 struct kvm_io_device *pos = bus->devs[i];
1983
1984                 kvm_iodevice_destructor(pos);
1985         }
1986         kfree(bus);
1987 }
1988
1989 /* kvm_io_bus_write - called under kvm->slots_lock */
1990 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
1991                      int len, const void *val)
1992 {
1993         int i;
1994         struct kvm_io_bus *bus = rcu_dereference(kvm->buses[bus_idx]);
1995         for (i = 0; i < bus->dev_count; i++)
1996                 if (!kvm_iodevice_write(bus->devs[i], addr, len, val))
1997                         return 0;
1998         return -EOPNOTSUPP;
1999 }
2000
2001 /* kvm_io_bus_read - called under kvm->slots_lock */
2002 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2003                     int len, void *val)
2004 {
2005         int i;
2006         struct kvm_io_bus *bus = rcu_dereference(kvm->buses[bus_idx]);
2007
2008         for (i = 0; i < bus->dev_count; i++)
2009                 if (!kvm_iodevice_read(bus->devs[i], addr, len, val))
2010                         return 0;
2011         return -EOPNOTSUPP;
2012 }
2013
2014 /* Caller must hold slots_lock. */
2015 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2016                             struct kvm_io_device *dev)
2017 {
2018         struct kvm_io_bus *new_bus, *bus;
2019
2020         bus = kvm->buses[bus_idx];
2021         if (bus->dev_count > NR_IOBUS_DEVS-1)
2022                 return -ENOSPC;
2023
2024         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2025         if (!new_bus)
2026                 return -ENOMEM;
2027         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2028         new_bus->devs[new_bus->dev_count++] = dev;
2029         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2030         synchronize_srcu_expedited(&kvm->srcu);
2031         kfree(bus);
2032
2033         return 0;
2034 }
2035
2036 /* Caller must hold slots_lock. */
2037 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2038                               struct kvm_io_device *dev)
2039 {
2040         int i, r;
2041         struct kvm_io_bus *new_bus, *bus;
2042
2043         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2044         if (!new_bus)
2045                 return -ENOMEM;
2046
2047         bus = kvm->buses[bus_idx];
2048         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2049
2050         r = -ENOENT;
2051         for (i = 0; i < new_bus->dev_count; i++)
2052                 if (new_bus->devs[i] == dev) {
2053                         r = 0;
2054                         new_bus->devs[i] = new_bus->devs[--new_bus->dev_count];
2055                         break;
2056                 }
2057
2058         if (r) {
2059                 kfree(new_bus);
2060                 return r;
2061         }
2062
2063         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2064         synchronize_srcu_expedited(&kvm->srcu);
2065         kfree(bus);
2066         return r;
2067 }
2068
2069 static struct notifier_block kvm_cpu_notifier = {
2070         .notifier_call = kvm_cpu_hotplug,
2071         .priority = 20, /* must be > scheduler priority */
2072 };
2073
2074 static int vm_stat_get(void *_offset, u64 *val)
2075 {
2076         unsigned offset = (long)_offset;
2077         struct kvm *kvm;
2078
2079         *val = 0;
2080         spin_lock(&kvm_lock);
2081         list_for_each_entry(kvm, &vm_list, vm_list)
2082                 *val += *(u32 *)((void *)kvm + offset);
2083         spin_unlock(&kvm_lock);
2084         return 0;
2085 }
2086
2087 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2088
2089 static int vcpu_stat_get(void *_offset, u64 *val)
2090 {
2091         unsigned offset = (long)_offset;
2092         struct kvm *kvm;
2093         struct kvm_vcpu *vcpu;
2094         int i;
2095
2096         *val = 0;
2097         spin_lock(&kvm_lock);
2098         list_for_each_entry(kvm, &vm_list, vm_list)
2099                 kvm_for_each_vcpu(i, vcpu, kvm)
2100                         *val += *(u32 *)((void *)vcpu + offset);
2101
2102         spin_unlock(&kvm_lock);
2103         return 0;
2104 }
2105
2106 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2107
2108 static const struct file_operations *stat_fops[] = {
2109         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2110         [KVM_STAT_VM]   = &vm_stat_fops,
2111 };
2112
2113 static void kvm_init_debug(void)
2114 {
2115         struct kvm_stats_debugfs_item *p;
2116
2117         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2118         for (p = debugfs_entries; p->name; ++p)
2119                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2120                                                 (void *)(long)p->offset,
2121                                                 stat_fops[p->kind]);
2122 }
2123
2124 static void kvm_exit_debug(void)
2125 {
2126         struct kvm_stats_debugfs_item *p;
2127
2128         for (p = debugfs_entries; p->name; ++p)
2129                 debugfs_remove(p->dentry);
2130         debugfs_remove(kvm_debugfs_dir);
2131 }
2132
2133 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2134 {
2135         if (kvm_usage_count)
2136                 hardware_disable(NULL);
2137         return 0;
2138 }
2139
2140 static int kvm_resume(struct sys_device *dev)
2141 {
2142         if (kvm_usage_count)
2143                 hardware_enable(NULL);
2144         return 0;
2145 }
2146
2147 static struct sysdev_class kvm_sysdev_class = {
2148         .name = "kvm",
2149         .suspend = kvm_suspend,
2150         .resume = kvm_resume,
2151 };
2152
2153 static struct sys_device kvm_sysdev = {
2154         .id = 0,
2155         .cls = &kvm_sysdev_class,
2156 };
2157
2158 struct page *bad_page;
2159 pfn_t bad_pfn;
2160
2161 static inline
2162 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2163 {
2164         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2165 }
2166
2167 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2168 {
2169         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2170
2171         kvm_arch_vcpu_load(vcpu, cpu);
2172 }
2173
2174 static void kvm_sched_out(struct preempt_notifier *pn,
2175                           struct task_struct *next)
2176 {
2177         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2178
2179         kvm_arch_vcpu_put(vcpu);
2180 }
2181
2182 int kvm_init(void *opaque, unsigned int vcpu_size,
2183                   struct module *module)
2184 {
2185         int r;
2186         int cpu;
2187
2188         r = kvm_arch_init(opaque);
2189         if (r)
2190                 goto out_fail;
2191
2192         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2193
2194         if (bad_page == NULL) {
2195                 r = -ENOMEM;
2196                 goto out;
2197         }
2198
2199         bad_pfn = page_to_pfn(bad_page);
2200
2201         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2202                 r = -ENOMEM;
2203                 goto out_free_0;
2204         }
2205
2206         r = kvm_arch_hardware_setup();
2207         if (r < 0)
2208                 goto out_free_0a;
2209
2210         for_each_online_cpu(cpu) {
2211                 smp_call_function_single(cpu,
2212                                 kvm_arch_check_processor_compat,
2213                                 &r, 1);
2214                 if (r < 0)
2215                         goto out_free_1;
2216         }
2217
2218         r = register_cpu_notifier(&kvm_cpu_notifier);
2219         if (r)
2220                 goto out_free_2;
2221         register_reboot_notifier(&kvm_reboot_notifier);
2222
2223         r = sysdev_class_register(&kvm_sysdev_class);
2224         if (r)
2225                 goto out_free_3;
2226
2227         r = sysdev_register(&kvm_sysdev);
2228         if (r)
2229                 goto out_free_4;
2230
2231         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2232         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
2233                                            __alignof__(struct kvm_vcpu),
2234                                            0, NULL);
2235         if (!kvm_vcpu_cache) {
2236                 r = -ENOMEM;
2237                 goto out_free_5;
2238         }
2239
2240         kvm_chardev_ops.owner = module;
2241         kvm_vm_fops.owner = module;
2242         kvm_vcpu_fops.owner = module;
2243
2244         r = misc_register(&kvm_dev);
2245         if (r) {
2246                 printk(KERN_ERR "kvm: misc device register failed\n");
2247                 goto out_free;
2248         }
2249
2250         kvm_preempt_ops.sched_in = kvm_sched_in;
2251         kvm_preempt_ops.sched_out = kvm_sched_out;
2252
2253         kvm_init_debug();
2254
2255         return 0;
2256
2257 out_free:
2258         kmem_cache_destroy(kvm_vcpu_cache);
2259 out_free_5:
2260         sysdev_unregister(&kvm_sysdev);
2261 out_free_4:
2262         sysdev_class_unregister(&kvm_sysdev_class);
2263 out_free_3:
2264         unregister_reboot_notifier(&kvm_reboot_notifier);
2265         unregister_cpu_notifier(&kvm_cpu_notifier);
2266 out_free_2:
2267 out_free_1:
2268         kvm_arch_hardware_unsetup();
2269 out_free_0a:
2270         free_cpumask_var(cpus_hardware_enabled);
2271 out_free_0:
2272         __free_page(bad_page);
2273 out:
2274         kvm_arch_exit();
2275 out_fail:
2276         return r;
2277 }
2278 EXPORT_SYMBOL_GPL(kvm_init);
2279
2280 void kvm_exit(void)
2281 {
2282         tracepoint_synchronize_unregister();
2283         kvm_exit_debug();
2284         misc_deregister(&kvm_dev);
2285         kmem_cache_destroy(kvm_vcpu_cache);
2286         sysdev_unregister(&kvm_sysdev);
2287         sysdev_class_unregister(&kvm_sysdev_class);
2288         unregister_reboot_notifier(&kvm_reboot_notifier);
2289         unregister_cpu_notifier(&kvm_cpu_notifier);
2290         on_each_cpu(hardware_disable, NULL, 1);
2291         kvm_arch_hardware_unsetup();
2292         kvm_arch_exit();
2293         free_cpumask_var(cpus_hardware_enabled);
2294         __free_page(bad_page);
2295 }
2296 EXPORT_SYMBOL_GPL(kvm_exit);