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 out:
 426         return kvm;
 427
 428 out_err:
 429         hardware_disable_all();
 430 out_err_nodisable:
 431         for (i = 0; i < KVM_NR_BUSES; i++)
 432                 kfree(kvm->buses[i]);
 433         kfree(kvm->memslots);
 434         kfree(kvm);
 435         return ERR_PTR(r);
 436 }
 437
 438 /*
 439  * Free any memory in @free but not in @dont.
 440  */
 441 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 442                                   struct kvm_memory_slot *dont)
 443 {
 444         int i;
 445
 446         if (!dont || free->rmap != dont->rmap)
 447                 vfree(free->rmap);
 448
 449         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 450                 vfree(free->dirty_bitmap);
 451
 452
 453         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
 454                 if (!dont || free->lpage_info[i] != dont->lpage_info[i]) {
 455                         vfree(free->lpage_info[i]);
 456                         free->lpage_info[i] = NULL;
 457                 }
 458         }
 459
 460         free->npages = 0;
 461         free->dirty_bitmap = NULL;
 462         free->rmap = NULL;
 463 }
 464
 465 void kvm_free_physmem(struct kvm *kvm)
 466 {
 467         int i;
 468         struct kvm_memslots *slots = kvm->memslots;
 469
 470         for (i = 0; i < slots->nmemslots; ++i)
 471                 kvm_free_physmem_slot(&slots->memslots[i], NULL);
 472
 473         kfree(kvm->memslots);
 474 }
 475
 476 static void kvm_destroy_vm(struct kvm *kvm)
 477 {
 478         int i;
 479         struct mm_struct *mm = kvm->mm;
 480
 481         kvm_arch_sync_events(kvm);
 482         spin_lock(&kvm_lock);
 483         list_del(&kvm->vm_list);
 484         spin_unlock(&kvm_lock);
 485         kvm_free_irq_routing(kvm);
 486         for (i = 0; i < KVM_NR_BUSES; i++)
 487                 kvm_io_bus_destroy(kvm->buses[i]);
 488         kvm_coalesced_mmio_free(kvm);
 489 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 490         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
 491 #else
 492         kvm_arch_flush_shadow(kvm);
 493 #endif
 494         kvm_arch_destroy_vm(kvm);
 495         hardware_disable_all();
 496         mmdrop(mm);
 497 }
 498
 499 void kvm_get_kvm(struct kvm *kvm)
 500 {
 501         atomic_inc(&kvm->users_count);
 502 }
 503 EXPORT_SYMBOL_GPL(kvm_get_kvm);
 504
 505 void kvm_put_kvm(struct kvm *kvm)
 506 {
 507         if (atomic_dec_and_test(&kvm->users_count))
 508                 kvm_destroy_vm(kvm);
 509 }
 510 EXPORT_SYMBOL_GPL(kvm_put_kvm);
 511
 512
 513 static int kvm_vm_release(struct inode *inode, struct file *filp)
 514 {
 515         struct kvm *kvm = filp->private_data;
 516
 517         kvm_irqfd_release(kvm);
 518
 519         kvm_put_kvm(kvm);
 520         return 0;
 521 }
 522
 523 /*
 524  * Allocate some memory and give it an address in the guest physical address
 525  * space.
 526  *
 527  * Discontiguous memory is allowed, mostly for framebuffers.
 528  *
 529  * Must be called holding mmap_sem for write.
 530  */
 531 int __kvm_set_memory_region(struct kvm *kvm,
 532                             struct kvm_userspace_memory_region *mem,
 533                             int user_alloc)
 534 {
 535         int r, flush_shadow = 0;
 536         gfn_t base_gfn;
 537         unsigned long npages;
 538         unsigned long i;
 539         struct kvm_memory_slot *memslot;
 540         struct kvm_memory_slot old, new;
 541         struct kvm_memslots *slots, *old_memslots;
 542
 543         r = -EINVAL;
 544         /* General sanity checks */
 545         if (mem->memory_size & (PAGE_SIZE - 1))
 546                 goto out;
 547         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 548                 goto out;
 549         if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
 550                 goto out;
 551         if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
 552                 goto out;
 553         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 554                 goto out;
 555
 556         memslot = &kvm->memslots->memslots[mem->slot];
 557         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 558         npages = mem->memory_size >> PAGE_SHIFT;
 559
 560         r = -EINVAL;
 561         if (npages > KVM_MEM_MAX_NR_PAGES)
 562                 goto out;
 563
 564         if (!npages)
 565                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 566
 567         new = old = *memslot;
 568
 569         new.base_gfn = base_gfn;
 570         new.npages = npages;
 571         new.flags = mem->flags;
 572
 573         /* Disallow changing a memory slot's size. */
 574         r = -EINVAL;
 575         if (npages && old.npages && npages != old.npages)
 576                 goto out_free;
 577
 578         /* Check for overlaps */
 579         r = -EEXIST;
 580         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 581                 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
 582
 583                 if (s == memslot || !s->npages)
 584                         continue;
 585                 if (!((base_gfn + npages <= s->base_gfn) ||
 586                       (base_gfn >= s->base_gfn + s->npages)))
 587                         goto out_free;
 588         }
 589
 590         /* Free page dirty bitmap if unneeded */
 591         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 592                 new.dirty_bitmap = NULL;
 593
 594         r = -ENOMEM;
 595
 596         /* Allocate if a slot is being created */
 597 #ifndef CONFIG_S390
 598         if (npages && !new.rmap) {
 599                 new.rmap = vmalloc(npages * sizeof(struct page *));
 600
 601                 if (!new.rmap)
 602                         goto out_free;
 603
 604                 memset(new.rmap, 0, npages * sizeof(*new.rmap));
 605
 606                 new.user_alloc = user_alloc;
 607                 new.userspace_addr = mem->userspace_addr;
 608         }
 609         if (!npages)
 610                 goto skip_lpage;
 611
 612         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
 613                 unsigned long ugfn;
 614                 unsigned long j;
 615                 int lpages;
 616                 int level = i + 2;
 617
 618                 /* Avoid unused variable warning if no large pages */
 619                 (void)level;
 620
 621                 if (new.lpage_info[i])
 622                         continue;
 623
 624                 lpages = 1 + (base_gfn + npages - 1) /
 625                              KVM_PAGES_PER_HPAGE(level);
 626                 lpages -= base_gfn / KVM_PAGES_PER_HPAGE(level);
 627
 628                 new.lpage_info[i] = vmalloc(lpages * sizeof(*new.lpage_info[i]));
 629
 630                 if (!new.lpage_info[i])
 631                         goto out_free;
 632
 633                 memset(new.lpage_info[i], 0,
 634                        lpages * sizeof(*new.lpage_info[i]));
 635
 636                 if (base_gfn % KVM_PAGES_PER_HPAGE(level))
 637                         new.lpage_info[i][0].write_count = 1;
 638                 if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE(level))
 639                         new.lpage_info[i][lpages - 1].write_count = 1;
 640                 ugfn = new.userspace_addr >> PAGE_SHIFT;
 641                 /*
 642                  * If the gfn and userspace address are not aligned wrt each
 643                  * other, or if explicitly asked to, disable large page
 644                  * support for this slot
 645                  */
 646                 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
 647                     !largepages_enabled)
 648                         for (j = 0; j < lpages; ++j)
 649                                 new.lpage_info[i][j].write_count = 1;
 650         }
 651
 652 skip_lpage:
 653
 654         /* Allocate page dirty bitmap if needed */
 655         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 656                 unsigned long dirty_bytes = kvm_dirty_bitmap_bytes(&new);
 657
 658                 new.dirty_bitmap = vmalloc(dirty_bytes);
 659                 if (!new.dirty_bitmap)
 660                         goto out_free;
 661                 memset(new.dirty_bitmap, 0, dirty_bytes);
 662                 /* destroy any largepage mappings for dirty tracking */
 663                 if (old.npages)
 664                         flush_shadow = 1;
 665         }
 666 #else  /* not defined CONFIG_S390 */
 667         new.user_alloc = user_alloc;
 668         if (user_alloc)
 669                 new.userspace_addr = mem->userspace_addr;
 670 #endif /* not defined CONFIG_S390 */
 671
 672         if (!npages) {
 673                 r = -ENOMEM;
 674                 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 675                 if (!slots)
 676                         goto out_free;
 677                 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
 678                 if (mem->slot >= slots->nmemslots)
 679                         slots->nmemslots = mem->slot + 1;
 680                 slots->memslots[mem->slot].flags |= KVM_MEMSLOT_INVALID;
 681
 682                 old_memslots = kvm->memslots;
 683                 rcu_assign_pointer(kvm->memslots, slots);
 684                 synchronize_srcu_expedited(&kvm->srcu);
 685                 /* From this point no new shadow pages pointing to a deleted
 686                  * memslot will be created.
 687                  *
 688                  * validation of sp->gfn happens in:
 689                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
 690                  *      - kvm_is_visible_gfn (mmu_check_roots)
 691                  */
 692                 kvm_arch_flush_shadow(kvm);
 693                 kfree(old_memslots);
 694         }
 695
 696         r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
 697         if (r)
 698                 goto out_free;
 699
 700 #ifdef CONFIG_DMAR
 701         /* map the pages in iommu page table */
 702         if (npages) {
 703                 r = kvm_iommu_map_pages(kvm, &new);
 704                 if (r)
 705                         goto out_free;
 706         }
 707 #endif
 708
 709         r = -ENOMEM;
 710         slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 711         if (!slots)
 712                 goto out_free;
 713         memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
 714         if (mem->slot >= slots->nmemslots)
 715                 slots->nmemslots = mem->slot + 1;
 716
 717         /* actual memory is freed via old in kvm_free_physmem_slot below */
 718         if (!npages) {
 719                 new.rmap = NULL;
 720                 new.dirty_bitmap = NULL;
 721                 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i)
 722                         new.lpage_info[i] = NULL;
 723         }
 724
 725         slots->memslots[mem->slot] = new;
 726         old_memslots = kvm->memslots;
 727         rcu_assign_pointer(kvm->memslots, slots);
 728         synchronize_srcu_expedited(&kvm->srcu);
 729
 730         kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
 731
 732         kvm_free_physmem_slot(&old, &new);
 733         kfree(old_memslots);
 734
 735         if (flush_shadow)
 736                 kvm_arch_flush_shadow(kvm);
 737
 738         return 0;
 739
 740 out_free:
 741         kvm_free_physmem_slot(&new, &old);
 742 out:
 743         return r;
 744
 745 }
 746 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 747
 748 int kvm_set_memory_region(struct kvm *kvm,
 749                           struct kvm_userspace_memory_region *mem,
 750                           int user_alloc)
 751 {
 752         int r;
 753
 754         mutex_lock(&kvm->slots_lock);
 755         r = __kvm_set_memory_region(kvm, mem, user_alloc);
 756         mutex_unlock(&kvm->slots_lock);
 757         return r;
 758 }
 759 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 760
 761 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 762                                    struct
 763                                    kvm_userspace_memory_region *mem,
 764                                    int user_alloc)
 765 {
 766         if (mem->slot >= KVM_MEMORY_SLOTS)
 767                 return -EINVAL;
 768         return kvm_set_memory_region(kvm, mem, user_alloc);
 769 }
 770
 771 int kvm_get_dirty_log(struct kvm *kvm,
 772                         struct kvm_dirty_log *log, int *is_dirty)
 773 {
 774         struct kvm_memory_slot *memslot;
 775         int r, i;
 776         unsigned long n;
 777         unsigned long any = 0;
 778
 779         r = -EINVAL;
 780         if (log->slot >= KVM_MEMORY_SLOTS)
 781                 goto out;
 782
 783         memslot = &kvm->memslots->memslots[log->slot];
 784         r = -ENOENT;
 785         if (!memslot->dirty_bitmap)
 786                 goto out;
 787
 788         n = kvm_dirty_bitmap_bytes(memslot);
 789
 790         for (i = 0; !any && i < n/sizeof(long); ++i)
 791                 any = memslot->dirty_bitmap[i];
 792
 793         r = -EFAULT;
 794         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 795                 goto out;
 796
 797         if (any)
 798                 *is_dirty = 1;
 799
 800         r = 0;
 801 out:
 802         return r;
 803 }
 804
 805 void kvm_disable_largepages(void)
 806 {
 807         largepages_enabled = false;
 808 }
 809 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
 810
 811 int is_error_page(struct page *page)
 812 {
 813         return page == bad_page;
 814 }
 815 EXPORT_SYMBOL_GPL(is_error_page);
 816
 817 int is_error_pfn(pfn_t pfn)
 818 {
 819         return pfn == bad_pfn;
 820 }
 821 EXPORT_SYMBOL_GPL(is_error_pfn);
 822
 823 static inline unsigned long bad_hva(void)
 824 {
 825         return PAGE_OFFSET;
 826 }
 827
 828 int kvm_is_error_hva(unsigned long addr)
 829 {
 830         return addr == bad_hva();
 831 }
 832 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
 833
 834 struct kvm_memory_slot *gfn_to_memslot_unaliased(struct kvm *kvm, gfn_t gfn)
 835 {
 836         int i;
 837         struct kvm_memslots *slots = kvm_memslots(kvm);
 838
 839         for (i = 0; i < slots->nmemslots; ++i) {
 840                 struct kvm_memory_slot *memslot = &slots->memslots[i];
 841
 842                 if (gfn >= memslot->base_gfn
 843                     && gfn < memslot->base_gfn + memslot->npages)
 844                         return memslot;
 845         }
 846         return NULL;
 847 }
 848 EXPORT_SYMBOL_GPL(gfn_to_memslot_unaliased);
 849
 850 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 851 {
 852         gfn = unalias_gfn(kvm, gfn);
 853         return gfn_to_memslot_unaliased(kvm, gfn);
 854 }
 855
 856 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
 857 {
 858         int i;
 859         struct kvm_memslots *slots = kvm_memslots(kvm);
 860
 861         gfn = unalias_gfn_instantiation(kvm, gfn);
 862         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 863                 struct kvm_memory_slot *memslot = &slots->memslots[i];
 864
 865                 if (memslot->flags & KVM_MEMSLOT_INVALID)
 866                         continue;
 867
 868                 if (gfn >= memslot->base_gfn
 869                     && gfn < memslot->base_gfn + memslot->npages)
 870                         return 1;
 871         }
 872         return 0;
 873 }
 874 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
 875
 876 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
 877 {
 878         struct vm_area_struct *vma;
 879         unsigned long addr, size;
 880
 881         size = PAGE_SIZE;
 882
 883         addr = gfn_to_hva(kvm, gfn);
 884         if (kvm_is_error_hva(addr))
 885                 return PAGE_SIZE;
 886
 887         down_read(&current->mm->mmap_sem);
 888         vma = find_vma(current->mm, addr);
 889         if (!vma)
 890                 goto out;
 891
 892         size = vma_kernel_pagesize(vma);
 893
 894 out:
 895         up_read(&current->mm->mmap_sem);
 896
 897         return size;
 898 }
 899
 900 int memslot_id(struct kvm *kvm, gfn_t gfn)
 901 {
 902         int i;
 903         struct kvm_memslots *slots = kvm_memslots(kvm);
 904         struct kvm_memory_slot *memslot = NULL;
 905
 906         gfn = unalias_gfn(kvm, gfn);
 907         for (i = 0; i < slots->nmemslots; ++i) {
 908                 memslot = &slots->memslots[i];
 909
 910                 if (gfn >= memslot->base_gfn
 911                     && gfn < memslot->base_gfn + memslot->npages)
 912                         break;
 913         }
 914
 915         return memslot - slots->memslots;
 916 }
 917
 918 static unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
 919 {
 920         return slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE;
 921 }
 922
 923 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
 924 {
 925         struct kvm_memory_slot *slot;
 926
 927         gfn = unalias_gfn_instantiation(kvm, gfn);
 928         slot = gfn_to_memslot_unaliased(kvm, gfn);
 929         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
 930                 return bad_hva();
 931         return gfn_to_hva_memslot(slot, gfn);
 932 }
 933 EXPORT_SYMBOL_GPL(gfn_to_hva);
 934
 935 static pfn_t hva_to_pfn(struct kvm *kvm, unsigned long addr)
 936 {
 937         struct page *page[1];
 938         int npages;
 939         pfn_t pfn;
 940
 941         might_sleep();
 942
 943         npages = get_user_pages_fast(addr, 1, 1, page);
 944
 945         if (unlikely(npages != 1)) {
 946                 struct vm_area_struct *vma;
 947
 948                 down_read(&current->mm->mmap_sem);
 949                 vma = find_vma(current->mm, addr);
 950
 951                 if (vma == NULL || addr < vma->vm_start ||
 952                     !(vma->vm_flags & VM_PFNMAP)) {
 953                         up_read(&current->mm->mmap_sem);
 954                         get_page(bad_page);
 955                         return page_to_pfn(bad_page);
 956                 }
 957
 958                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
 959                 up_read(&current->mm->mmap_sem);
 960                 BUG_ON(!kvm_is_mmio_pfn(pfn));
 961         } else
 962                 pfn = page_to_pfn(page[0]);
 963
 964         return pfn;
 965 }
 966
 967 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
 968 {
 969         unsigned long addr;
 970
 971         addr = gfn_to_hva(kvm, gfn);
 972         if (kvm_is_error_hva(addr)) {
 973                 get_page(bad_page);
 974                 return page_to_pfn(bad_page);
 975         }
 976
 977         return hva_to_pfn(kvm, addr);
 978 }
 979 EXPORT_SYMBOL_GPL(gfn_to_pfn);
 980
 981 pfn_t gfn_to_pfn_memslot(struct kvm *kvm,
 982                          struct kvm_memory_slot *slot, gfn_t gfn)
 983 {
 984         unsigned long addr = gfn_to_hva_memslot(slot, gfn);
 985         return hva_to_pfn(kvm, addr);
 986 }
 987
 988 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
 989 {
 990         pfn_t pfn;
 991
 992         pfn = gfn_to_pfn(kvm, gfn);
 993         if (!kvm_is_mmio_pfn(pfn))
 994                 return pfn_to_page(pfn);
 995
 996         WARN_ON(kvm_is_mmio_pfn(pfn));
 997
 998         get_page(bad_page);
 999         return bad_page;
1000 }
1001
1002 EXPORT_SYMBOL_GPL(gfn_to_page);
1003
1004 void kvm_release_page_clean(struct page *page)
1005 {
1006         kvm_release_pfn_clean(page_to_pfn(page));
1007 }
1008 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1009
1010 void kvm_release_pfn_clean(pfn_t pfn)
1011 {
1012         if (!kvm_is_mmio_pfn(pfn))
1013                 put_page(pfn_to_page(pfn));
1014 }
1015 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1016
1017 void kvm_release_page_dirty(struct page *page)
1018 {
1019         kvm_release_pfn_dirty(page_to_pfn(page));
1020 }
1021 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1022
1023 void kvm_release_pfn_dirty(pfn_t pfn)
1024 {
1025         kvm_set_pfn_dirty(pfn);
1026         kvm_release_pfn_clean(pfn);
1027 }
1028 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1029
1030 void kvm_set_page_dirty(struct page *page)
1031 {
1032         kvm_set_pfn_dirty(page_to_pfn(page));
1033 }
1034 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1035
1036 void kvm_set_pfn_dirty(pfn_t pfn)
1037 {
1038         if (!kvm_is_mmio_pfn(pfn)) {
1039                 struct page *page = pfn_to_page(pfn);
1040                 if (!PageReserved(page))
1041                         SetPageDirty(page);
1042         }
1043 }
1044 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1045
1046 void kvm_set_pfn_accessed(pfn_t pfn)
1047 {
1048         if (!kvm_is_mmio_pfn(pfn))
1049                 mark_page_accessed(pfn_to_page(pfn));
1050 }
1051 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1052
1053 void kvm_get_pfn(pfn_t pfn)
1054 {
1055         if (!kvm_is_mmio_pfn(pfn))
1056                 get_page(pfn_to_page(pfn));
1057 }
1058 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1059
1060 static int next_segment(unsigned long len, int offset)
1061 {
1062         if (len > PAGE_SIZE - offset)
1063                 return PAGE_SIZE - offset;
1064         else
1065                 return len;
1066 }
1067
1068 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1069                         int len)
1070 {
1071         int r;
1072         unsigned long addr;
1073
1074         addr = gfn_to_hva(kvm, gfn);
1075         if (kvm_is_error_hva(addr))
1076                 return -EFAULT;
1077         r = copy_from_user(data, (void __user *)addr + offset, len);
1078         if (r)
1079                 return -EFAULT;
1080         return 0;
1081 }
1082 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1083
1084 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1085 {
1086         gfn_t gfn = gpa >> PAGE_SHIFT;
1087         int seg;
1088         int offset = offset_in_page(gpa);
1089         int ret;
1090
1091         while ((seg = next_segment(len, offset)) != 0) {
1092                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1093                 if (ret < 0)
1094                         return ret;
1095                 offset = 0;
1096                 len -= seg;
1097                 data += seg;
1098                 ++gfn;
1099         }
1100         return 0;
1101 }
1102 EXPORT_SYMBOL_GPL(kvm_read_guest);
1103
1104 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1105                           unsigned long len)
1106 {
1107         int r;
1108         unsigned long addr;
1109         gfn_t gfn = gpa >> PAGE_SHIFT;
1110         int offset = offset_in_page(gpa);
1111
1112         addr = gfn_to_hva(kvm, gfn);
1113         if (kvm_is_error_hva(addr))
1114                 return -EFAULT;
1115         pagefault_disable();
1116         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1117         pagefault_enable();
1118         if (r)
1119                 return -EFAULT;
1120         return 0;
1121 }
1122 EXPORT_SYMBOL(kvm_read_guest_atomic);
1123
1124 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1125                          int offset, int len)
1126 {
1127         int r;
1128         unsigned long addr;
1129
1130         addr = gfn_to_hva(kvm, gfn);
1131         if (kvm_is_error_hva(addr))
1132                 return -EFAULT;
1133         r = copy_to_user((void __user *)addr + offset, data, len);
1134         if (r)
1135                 return -EFAULT;
1136         mark_page_dirty(kvm, gfn);
1137         return 0;
1138 }
1139 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1140
1141 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1142                     unsigned long len)
1143 {
1144         gfn_t gfn = gpa >> PAGE_SHIFT;
1145         int seg;
1146         int offset = offset_in_page(gpa);
1147         int ret;
1148
1149         while ((seg = next_segment(len, offset)) != 0) {
1150                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1151                 if (ret < 0)
1152                         return ret;
1153                 offset = 0;
1154                 len -= seg;
1155                 data += seg;
1156                 ++gfn;
1157         }
1158         return 0;
1159 }
1160
1161 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1162 {
1163         return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
1164 }
1165 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1166
1167 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1168 {
1169         gfn_t gfn = gpa >> PAGE_SHIFT;
1170         int seg;
1171         int offset = offset_in_page(gpa);
1172         int ret;
1173
1174         while ((seg = next_segment(len, offset)) != 0) {
1175                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1176                 if (ret < 0)
1177                         return ret;
1178                 offset = 0;
1179                 len -= seg;
1180                 ++gfn;
1181         }
1182         return 0;
1183 }
1184 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1185
1186 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1187 {
1188         struct kvm_memory_slot *memslot;
1189
1190         gfn = unalias_gfn(kvm, gfn);
1191         memslot = gfn_to_memslot_unaliased(kvm, gfn);
1192         if (memslot && memslot->dirty_bitmap) {
1193                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1194
1195                 /* avoid RMW */
1196                 if (!generic_test_le_bit(rel_gfn, memslot->dirty_bitmap))
1197                         generic___set_le_bit(rel_gfn, memslot->dirty_bitmap);
1198         }
1199 }
1200
1201 /*
1202  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1203  */
1204 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1205 {
1206         DEFINE_WAIT(wait);
1207
1208         for (;;) {
1209                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1210
1211                 if (kvm_arch_vcpu_runnable(vcpu)) {
1212                         set_bit(KVM_REQ_UNHALT, &vcpu->requests);
1213                         break;
1214                 }
1215                 if (kvm_cpu_has_pending_timer(vcpu))
1216                         break;
1217                 if (signal_pending(current))
1218                         break;
1219
1220                 schedule();
1221         }
1222
1223         finish_wait(&vcpu->wq, &wait);
1224 }
1225
1226 void kvm_resched(struct kvm_vcpu *vcpu)
1227 {
1228         if (!need_resched())
1229                 return;
1230         cond_resched();
1231 }
1232 EXPORT_SYMBOL_GPL(kvm_resched);
1233
1234 void kvm_vcpu_on_spin(struct kvm_vcpu *vcpu)
1235 {
1236         ktime_t expires;
1237         DEFINE_WAIT(wait);
1238
1239         prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1240
1241         /* Sleep for 100 us, and hope lock-holder got scheduled */
1242         expires = ktime_add_ns(ktime_get(), 100000UL);
1243         schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
1244
1245         finish_wait(&vcpu->wq, &wait);
1246 }
1247 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1248
1249 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1250 {
1251         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1252         struct page *page;
1253
1254         if (vmf->pgoff == 0)
1255                 page = virt_to_page(vcpu->run);
1256 #ifdef CONFIG_X86
1257         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1258                 page = virt_to_page(vcpu->arch.pio_data);
1259 #endif
1260 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1261         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1262                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1263 #endif
1264         else
1265                 return VM_FAULT_SIGBUS;
1266         get_page(page);
1267         vmf->page = page;
1268         return 0;
1269 }
1270
1271 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1272         .fault = kvm_vcpu_fault,
1273 };
1274
1275 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1276 {
1277         vma->vm_ops = &kvm_vcpu_vm_ops;
1278         return 0;
1279 }
1280
1281 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1282 {
1283         struct kvm_vcpu *vcpu = filp->private_data;
1284
1285         kvm_put_kvm(vcpu->kvm);
1286         return 0;
1287 }
1288
1289 static struct file_operations kvm_vcpu_fops = {
1290         .release        = kvm_vcpu_release,
1291         .unlocked_ioctl = kvm_vcpu_ioctl,
1292         .compat_ioctl   = kvm_vcpu_ioctl,
1293         .mmap           = kvm_vcpu_mmap,
1294 };
1295
1296 /*
1297  * Allocates an inode for the vcpu.
1298  */
1299 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1300 {
1301         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1302 }
1303
1304 /*
1305  * Creates some virtual cpus.  Good luck creating more than one.
1306  */
1307 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1308 {
1309         int r;
1310         struct kvm_vcpu *vcpu, *v;
1311
1312         vcpu = kvm_arch_vcpu_create(kvm, id);
1313         if (IS_ERR(vcpu))
1314                 return PTR_ERR(vcpu);
1315
1316         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1317
1318         r = kvm_arch_vcpu_setup(vcpu);
1319         if (r)
1320                 return r;
1321
1322         mutex_lock(&kvm->lock);
1323         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1324                 r = -EINVAL;
1325                 goto vcpu_destroy;
1326         }
1327
1328         kvm_for_each_vcpu(r, v, kvm)
1329                 if (v->vcpu_id == id) {
1330                         r = -EEXIST;
1331                         goto vcpu_destroy;
1332                 }
1333
1334         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1335
1336         /* Now it's all set up, let userspace reach it */
1337         kvm_get_kvm(kvm);
1338         r = create_vcpu_fd(vcpu);
1339         if (r < 0) {
1340                 kvm_put_kvm(kvm);
1341                 goto vcpu_destroy;
1342         }
1343
1344         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1345         smp_wmb();
1346         atomic_inc(&kvm->online_vcpus);
1347
1348 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1349         if (kvm->bsp_vcpu_id == id)
1350                 kvm->bsp_vcpu = vcpu;
1351 #endif
1352         mutex_unlock(&kvm->lock);
1353         return r;
1354
1355 vcpu_destroy:
1356         mutex_unlock(&kvm->lock);
1357         kvm_arch_vcpu_destroy(vcpu);
1358         return r;
1359 }
1360
1361 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1362 {
1363         if (sigset) {
1364                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1365                 vcpu->sigset_active = 1;
1366                 vcpu->sigset = *sigset;
1367         } else
1368                 vcpu->sigset_active = 0;
1369         return 0;
1370 }
1371
1372 static long kvm_vcpu_ioctl(struct file *filp,
1373                            unsigned int ioctl, unsigned long arg)
1374 {
1375         struct kvm_vcpu *vcpu = filp->private_data;
1376         void __user *argp = (void __user *)arg;
1377         int r;
1378         struct kvm_fpu *fpu = NULL;
1379         struct kvm_sregs *kvm_sregs = NULL;
1380
1381         if (vcpu->kvm->mm != current->mm)
1382                 return -EIO;
1383         switch (ioctl) {
1384         case KVM_RUN:
1385                 r = -EINVAL;
1386                 if (arg)
1387                         goto out;
1388                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1389                 break;
1390         case KVM_GET_REGS: {
1391                 struct kvm_regs *kvm_regs;
1392
1393                 r = -ENOMEM;
1394                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1395                 if (!kvm_regs)
1396                         goto out;
1397                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1398                 if (r)
1399                         goto out_free1;
1400                 r = -EFAULT;
1401                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1402                         goto out_free1;
1403                 r = 0;
1404 out_free1:
1405                 kfree(kvm_regs);
1406                 break;
1407         }
1408         case KVM_SET_REGS: {
1409                 struct kvm_regs *kvm_regs;
1410
1411                 r = -ENOMEM;
1412                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1413                 if (!kvm_regs)
1414                         goto out;
1415                 r = -EFAULT;
1416                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1417                         goto out_free2;
1418                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1419                 if (r)
1420                         goto out_free2;
1421                 r = 0;
1422 out_free2:
1423                 kfree(kvm_regs);
1424                 break;
1425         }
1426         case KVM_GET_SREGS: {
1427                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1428                 r = -ENOMEM;
1429                 if (!kvm_sregs)
1430                         goto out;
1431                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1432                 if (r)
1433                         goto out;
1434                 r = -EFAULT;
1435                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1436                         goto out;
1437                 r = 0;
1438                 break;
1439         }
1440         case KVM_SET_SREGS: {
1441                 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1442                 r = -ENOMEM;
1443                 if (!kvm_sregs)
1444                         goto out;
1445                 r = -EFAULT;
1446                 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1447                         goto out;
1448                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1449                 if (r)
1450                         goto out;
1451                 r = 0;
1452                 break;
1453         }
1454         case KVM_GET_MP_STATE: {
1455                 struct kvm_mp_state mp_state;
1456
1457                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1458                 if (r)
1459                         goto out;
1460                 r = -EFAULT;
1461                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1462                         goto out;
1463                 r = 0;
1464                 break;
1465         }
1466         case KVM_SET_MP_STATE: {
1467                 struct kvm_mp_state mp_state;
1468
1469                 r = -EFAULT;
1470                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1471                         goto out;
1472                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1473                 if (r)
1474                         goto out;
1475                 r = 0;
1476                 break;
1477         }
1478         case KVM_TRANSLATE: {
1479                 struct kvm_translation tr;
1480
1481                 r = -EFAULT;
1482                 if (copy_from_user(&tr, argp, sizeof tr))
1483                         goto out;
1484                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1485                 if (r)
1486                         goto out;
1487                 r = -EFAULT;
1488                 if (copy_to_user(argp, &tr, sizeof tr))
1489                         goto out;
1490                 r = 0;
1491                 break;
1492         }
1493         case KVM_SET_GUEST_DEBUG: {
1494                 struct kvm_guest_debug dbg;
1495
1496                 r = -EFAULT;
1497                 if (copy_from_user(&dbg, argp, sizeof dbg))
1498                         goto out;
1499                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1500                 if (r)
1501                         goto out;
1502                 r = 0;
1503                 break;
1504         }
1505         case KVM_SET_SIGNAL_MASK: {
1506                 struct kvm_signal_mask __user *sigmask_arg = argp;
1507                 struct kvm_signal_mask kvm_sigmask;
1508                 sigset_t sigset, *p;
1509
1510                 p = NULL;
1511                 if (argp) {
1512                         r = -EFAULT;
1513                         if (copy_from_user(&kvm_sigmask, argp,
1514                                            sizeof kvm_sigmask))
1515                                 goto out;
1516                         r = -EINVAL;
1517                         if (kvm_sigmask.len != sizeof sigset)
1518                                 goto out;
1519                         r = -EFAULT;
1520                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1521                                            sizeof sigset))
1522                                 goto out;
1523                         p = &sigset;
1524                 }
1525                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1526                 break;
1527         }
1528         case KVM_GET_FPU: {
1529                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1530                 r = -ENOMEM;
1531                 if (!fpu)
1532                         goto out;
1533                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1534                 if (r)
1535                         goto out;
1536                 r = -EFAULT;
1537                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1538                         goto out;
1539                 r = 0;
1540                 break;
1541         }
1542         case KVM_SET_FPU: {
1543                 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1544                 r = -ENOMEM;
1545                 if (!fpu)
1546                         goto out;
1547                 r = -EFAULT;
1548                 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1549                         goto out;
1550                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1551                 if (r)
1552                         goto out;
1553                 r = 0;
1554                 break;
1555         }
1556         default:
1557                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1558         }
1559 out:
1560         kfree(fpu);
1561         kfree(kvm_sregs);
1562         return r;
1563 }
1564
1565 static long kvm_vm_ioctl(struct file *filp,
1566                            unsigned int ioctl, unsigned long arg)
1567 {
1568         struct kvm *kvm = filp->private_data;
1569         void __user *argp = (void __user *)arg;
1570         int r;
1571
1572         if (kvm->mm != current->mm)
1573                 return -EIO;
1574         switch (ioctl) {
1575         case KVM_CREATE_VCPU:
1576                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1577                 if (r < 0)
1578                         goto out;
1579                 break;
1580         case KVM_SET_USER_MEMORY_REGION: {
1581                 struct kvm_userspace_memory_region kvm_userspace_mem;
1582
1583                 r = -EFAULT;
1584                 if (copy_from_user(&kvm_userspace_mem, argp,
1585                                                 sizeof kvm_userspace_mem))
1586                         goto out;
1587
1588                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1589                 if (r)
1590                         goto out;
1591                 break;
1592         }
1593         case KVM_GET_DIRTY_LOG: {
1594                 struct kvm_dirty_log log;
1595
1596                 r = -EFAULT;
1597                 if (copy_from_user(&log, argp, sizeof log))
1598                         goto out;
1599                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1600                 if (r)
1601                         goto out;
1602                 break;
1603         }
1604 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1605         case KVM_REGISTER_COALESCED_MMIO: {
1606                 struct kvm_coalesced_mmio_zone zone;
1607                 r = -EFAULT;
1608                 if (copy_from_user(&zone, argp, sizeof zone))
1609                         goto out;
1610                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1611                 if (r)
1612                         goto out;
1613                 r = 0;
1614                 break;
1615         }
1616         case KVM_UNREGISTER_COALESCED_MMIO: {
1617                 struct kvm_coalesced_mmio_zone zone;
1618                 r = -EFAULT;
1619                 if (copy_from_user(&zone, argp, sizeof zone))
1620                         goto out;
1621                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1622                 if (r)
1623                         goto out;
1624                 r = 0;
1625                 break;
1626         }
1627 #endif
1628         case KVM_IRQFD: {
1629                 struct kvm_irqfd data;
1630
1631                 r = -EFAULT;
1632                 if (copy_from_user(&data, argp, sizeof data))
1633                         goto out;
1634                 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
1635                 break;
1636         }
1637         case KVM_IOEVENTFD: {
1638                 struct kvm_ioeventfd data;
1639
1640                 r = -EFAULT;
1641                 if (copy_from_user(&data, argp, sizeof data))
1642                         goto out;
1643                 r = kvm_ioeventfd(kvm, &data);
1644                 break;
1645         }
1646 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1647         case KVM_SET_BOOT_CPU_ID:
1648                 r = 0;
1649                 mutex_lock(&kvm->lock);
1650                 if (atomic_read(&kvm->online_vcpus) != 0)
1651                         r = -EBUSY;
1652                 else
1653                         kvm->bsp_vcpu_id = arg;
1654                 mutex_unlock(&kvm->lock);
1655                 break;
1656 #endif
1657         default:
1658                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1659                 if (r == -ENOTTY)
1660                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
1661         }
1662 out:
1663         return r;
1664 }
1665
1666 #ifdef CONFIG_COMPAT
1667 struct compat_kvm_dirty_log {
1668         __u32 slot;
1669         __u32 padding1;
1670         union {
1671                 compat_uptr_t dirty_bitmap; /* one bit per page */
1672                 __u64 padding2;
1673         };
1674 };
1675
1676 static long kvm_vm_compat_ioctl(struct file *filp,
1677                            unsigned int ioctl, unsigned long arg)
1678 {
1679         struct kvm *kvm = filp->private_data;
1680         int r;
1681
1682         if (kvm->mm != current->mm)
1683                 return -EIO;
1684         switch (ioctl) {
1685         case KVM_GET_DIRTY_LOG: {
1686                 struct compat_kvm_dirty_log compat_log;
1687                 struct kvm_dirty_log log;
1688
1689                 r = -EFAULT;
1690                 if (copy_from_user(&compat_log, (void __user *)arg,
1691                                    sizeof(compat_log)))
1692                         goto out;
1693                 log.slot         = compat_log.slot;
1694                 log.padding1     = compat_log.padding1;
1695                 log.padding2     = compat_log.padding2;
1696                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
1697
1698                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1699                 if (r)
1700                         goto out;
1701                 break;
1702         }
1703         default:
1704                 r = kvm_vm_ioctl(filp, ioctl, arg);
1705         }
1706
1707 out:
1708         return r;
1709 }
1710 #endif
1711
1712 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1713 {
1714         struct page *page[1];
1715         unsigned long addr;
1716         int npages;
1717         gfn_t gfn = vmf->pgoff;
1718         struct kvm *kvm = vma->vm_file->private_data;
1719
1720         addr = gfn_to_hva(kvm, gfn);
1721         if (kvm_is_error_hva(addr))
1722                 return VM_FAULT_SIGBUS;
1723
1724         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
1725                                 NULL);
1726         if (unlikely(npages != 1))
1727                 return VM_FAULT_SIGBUS;
1728
1729         vmf->page = page[0];
1730         return 0;
1731 }
1732
1733 static const struct vm_operations_struct kvm_vm_vm_ops = {
1734         .fault = kvm_vm_fault,
1735 };
1736
1737 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1738 {
1739         vma->vm_ops = &kvm_vm_vm_ops;
1740         return 0;
1741 }
1742
1743 static struct file_operations kvm_vm_fops = {
1744         .release        = kvm_vm_release,
1745         .unlocked_ioctl = kvm_vm_ioctl,
1746 #ifdef CONFIG_COMPAT
1747         .compat_ioctl   = kvm_vm_compat_ioctl,
1748 #endif
1749         .mmap           = kvm_vm_mmap,
1750 };
1751
1752 static int kvm_dev_ioctl_create_vm(void)
1753 {
1754         int fd, r;
1755         struct kvm *kvm;
1756
1757         kvm = kvm_create_vm();
1758         if (IS_ERR(kvm))
1759                 return PTR_ERR(kvm);
1760 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1761         r = kvm_coalesced_mmio_init(kvm);
1762         if (r < 0) {
1763                 kvm_put_kvm(kvm);
1764                 return r;
1765         }
1766 #endif
1767         fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
1768         if (fd < 0)
1769                 kvm_put_kvm(kvm);
1770
1771         return fd;
1772 }
1773
1774 static long kvm_dev_ioctl_check_extension_generic(long arg)
1775 {
1776         switch (arg) {
1777         case KVM_CAP_USER_MEMORY:
1778         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
1779         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
1780 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1781         case KVM_CAP_SET_BOOT_CPU_ID:
1782 #endif
1783         case KVM_CAP_INTERNAL_ERROR_DATA:
1784                 return 1;
1785 #ifdef CONFIG_HAVE_KVM_IRQCHIP
1786         case KVM_CAP_IRQ_ROUTING:
1787                 return KVM_MAX_IRQ_ROUTES;
1788 #endif
1789         default:
1790                 break;
1791         }
1792         return kvm_dev_ioctl_check_extension(arg);
1793 }
1794
1795 static long kvm_dev_ioctl(struct file *filp,
1796                           unsigned int ioctl, unsigned long arg)
1797 {
1798         long r = -EINVAL;
1799
1800         switch (ioctl) {
1801         case KVM_GET_API_VERSION:
1802                 r = -EINVAL;
1803                 if (arg)
1804                         goto out;
1805                 r = KVM_API_VERSION;
1806                 break;
1807         case KVM_CREATE_VM:
1808                 r = -EINVAL;
1809                 if (arg)
1810                         goto out;
1811                 r = kvm_dev_ioctl_create_vm();
1812                 break;
1813         case KVM_CHECK_EXTENSION:
1814                 r = kvm_dev_ioctl_check_extension_generic(arg);
1815                 break;
1816         case KVM_GET_VCPU_MMAP_SIZE:
1817                 r = -EINVAL;
1818                 if (arg)
1819                         goto out;
1820                 r = PAGE_SIZE;     /* struct kvm_run */
1821 #ifdef CONFIG_X86
1822                 r += PAGE_SIZE;    /* pio data page */
1823 #endif
1824 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1825                 r += PAGE_SIZE;    /* coalesced mmio ring page */
1826 #endif
1827                 break;
1828         case KVM_TRACE_ENABLE:
1829         case KVM_TRACE_PAUSE:
1830         case KVM_TRACE_DISABLE:
1831                 r = -EOPNOTSUPP;
1832                 break;
1833         default:
1834                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
1835         }
1836 out:
1837         return r;
1838 }
1839
1840 static struct file_operations kvm_chardev_ops = {
1841         .unlocked_ioctl = kvm_dev_ioctl,
1842         .compat_ioctl   = kvm_dev_ioctl,
1843 };
1844
1845 static struct miscdevice kvm_dev = {
1846         KVM_MINOR,
1847         "kvm",
1848         &kvm_chardev_ops,
1849 };
1850
1851 static void hardware_enable(void *junk)
1852 {
1853         int cpu = raw_smp_processor_id();
1854         int r;
1855
1856         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
1857                 return;
1858
1859         cpumask_set_cpu(cpu, cpus_hardware_enabled);
1860
1861         r = kvm_arch_hardware_enable(NULL);
1862
1863         if (r) {
1864                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
1865                 atomic_inc(&hardware_enable_failed);
1866                 printk(KERN_INFO "kvm: enabling virtualization on "
1867                                  "CPU%d failed\n", cpu);
1868         }
1869 }
1870
1871 static void hardware_disable(void *junk)
1872 {
1873         int cpu = raw_smp_processor_id();
1874
1875         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
1876                 return;
1877         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
1878         kvm_arch_hardware_disable(NULL);
1879 }
1880
1881 static void hardware_disable_all_nolock(void)
1882 {
1883         BUG_ON(!kvm_usage_count);
1884
1885         kvm_usage_count--;
1886         if (!kvm_usage_count)
1887                 on_each_cpu(hardware_disable, NULL, 1);
1888 }
1889
1890 static void hardware_disable_all(void)
1891 {
1892         spin_lock(&kvm_lock);
1893         hardware_disable_all_nolock();
1894         spin_unlock(&kvm_lock);
1895 }
1896
1897 static int hardware_enable_all(void)
1898 {
1899         int r = 0;
1900
1901         spin_lock(&kvm_lock);
1902
1903         kvm_usage_count++;
1904         if (kvm_usage_count == 1) {
1905                 atomic_set(&hardware_enable_failed, 0);
1906                 on_each_cpu(hardware_enable, NULL, 1);
1907
1908                 if (atomic_read(&hardware_enable_failed)) {
1909                         hardware_disable_all_nolock();
1910                         r = -EBUSY;
1911                 }
1912         }
1913
1914         spin_unlock(&kvm_lock);
1915
1916         return r;
1917 }
1918
1919 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
1920                            void *v)
1921 {
1922         int cpu = (long)v;
1923
1924         if (!kvm_usage_count)
1925                 return NOTIFY_OK;
1926
1927         val &= ~CPU_TASKS_FROZEN;
1928         switch (val) {
1929         case CPU_DYING:
1930                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1931                        cpu);
1932                 hardware_disable(NULL);
1933                 break;
1934         case CPU_ONLINE:
1935                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
1936                        cpu);
1937                 smp_call_function_single(cpu, hardware_enable, NULL, 1);
1938                 break;
1939         }
1940         return NOTIFY_OK;
1941 }
1942
1943
1944 asmlinkage void kvm_handle_fault_on_reboot(void)
1945 {
1946         if (kvm_rebooting)
1947                 /* spin while reset goes on */
1948                 while (true)
1949                         ;
1950         /* Fault while not rebooting.  We want the trace. */
1951         BUG();
1952 }
1953 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
1954
1955 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
1956                       void *v)
1957 {
1958         /*
1959          * Some (well, at least mine) BIOSes hang on reboot if
1960          * in vmx root mode.
1961          *
1962          * And Intel TXT required VMX off for all cpu when system shutdown.
1963          */
1964         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
1965         kvm_rebooting = true;
1966         on_each_cpu(hardware_disable, NULL, 1);
1967         return NOTIFY_OK;
1968 }
1969
1970 static struct notifier_block kvm_reboot_notifier = {
1971         .notifier_call = kvm_reboot,
1972         .priority = 0,
1973 };
1974
1975 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
1976 {
1977         int i;
1978
1979         for (i = 0; i < bus->dev_count; i++) {
1980                 struct kvm_io_device *pos = bus->devs[i];
1981
1982                 kvm_iodevice_destructor(pos);
1983         }
1984         kfree(bus);
1985 }
1986
1987 /* kvm_io_bus_write - called under kvm->slots_lock */
1988 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
1989                      int len, const void *val)
1990 {
1991         int i;
1992         struct kvm_io_bus *bus;
1993
1994         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
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;
2007
2008         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2009         for (i = 0; i < bus->dev_count; i++)
2010                 if (!kvm_iodevice_read(bus->devs[i], addr, len, val))
2011                         return 0;
2012         return -EOPNOTSUPP;
2013 }
2014
2015 /* Caller must hold slots_lock. */
2016 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2017                             struct kvm_io_device *dev)
2018 {
2019         struct kvm_io_bus *new_bus, *bus;
2020
2021         bus = kvm->buses[bus_idx];
2022         if (bus->dev_count > NR_IOBUS_DEVS-1)
2023                 return -ENOSPC;
2024
2025         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2026         if (!new_bus)
2027                 return -ENOMEM;
2028         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2029         new_bus->devs[new_bus->dev_count++] = dev;
2030         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2031         synchronize_srcu_expedited(&kvm->srcu);
2032         kfree(bus);
2033
2034         return 0;
2035 }
2036
2037 /* Caller must hold slots_lock. */
2038 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2039                               struct kvm_io_device *dev)
2040 {
2041         int i, r;
2042         struct kvm_io_bus *new_bus, *bus;
2043
2044         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2045         if (!new_bus)
2046                 return -ENOMEM;
2047
2048         bus = kvm->buses[bus_idx];
2049         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2050
2051         r = -ENOENT;
2052         for (i = 0; i < new_bus->dev_count; i++)
2053                 if (new_bus->devs[i] == dev) {
2054                         r = 0;
2055                         new_bus->devs[i] = new_bus->devs[--new_bus->dev_count];
2056                         break;
2057                 }
2058
2059         if (r) {
2060                 kfree(new_bus);
2061                 return r;
2062         }
2063
2064         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2065         synchronize_srcu_expedited(&kvm->srcu);
2066         kfree(bus);
2067         return r;
2068 }
2069
2070 static struct notifier_block kvm_cpu_notifier = {
2071         .notifier_call = kvm_cpu_hotplug,
2072         .priority = 20, /* must be > scheduler priority */
2073 };
2074
2075 static int vm_stat_get(void *_offset, u64 *val)
2076 {
2077         unsigned offset = (long)_offset;
2078         struct kvm *kvm;
2079
2080         *val = 0;
2081         spin_lock(&kvm_lock);
2082         list_for_each_entry(kvm, &vm_list, vm_list)
2083                 *val += *(u32 *)((void *)kvm + offset);
2084         spin_unlock(&kvm_lock);
2085         return 0;
2086 }
2087
2088 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2089
2090 static int vcpu_stat_get(void *_offset, u64 *val)
2091 {
2092         unsigned offset = (long)_offset;
2093         struct kvm *kvm;
2094         struct kvm_vcpu *vcpu;
2095         int i;
2096
2097         *val = 0;
2098         spin_lock(&kvm_lock);
2099         list_for_each_entry(kvm, &vm_list, vm_list)
2100                 kvm_for_each_vcpu(i, vcpu, kvm)
2101                         *val += *(u32 *)((void *)vcpu + offset);
2102
2103         spin_unlock(&kvm_lock);
2104         return 0;
2105 }
2106
2107 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2108
2109 static const struct file_operations *stat_fops[] = {
2110         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2111         [KVM_STAT_VM]   = &vm_stat_fops,
2112 };
2113
2114 static void kvm_init_debug(void)
2115 {
2116         struct kvm_stats_debugfs_item *p;
2117
2118         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2119         for (p = debugfs_entries; p->name; ++p)
2120                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2121                                                 (void *)(long)p->offset,
2122                                                 stat_fops[p->kind]);
2123 }
2124
2125 static void kvm_exit_debug(void)
2126 {
2127         struct kvm_stats_debugfs_item *p;
2128
2129         for (p = debugfs_entries; p->name; ++p)
2130                 debugfs_remove(p->dentry);
2131         debugfs_remove(kvm_debugfs_dir);
2132 }
2133
2134 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2135 {
2136         if (kvm_usage_count)
2137                 hardware_disable(NULL);
2138         return 0;
2139 }
2140
2141 static int kvm_resume(struct sys_device *dev)
2142 {
2143         if (kvm_usage_count)
2144                 hardware_enable(NULL);
2145         return 0;
2146 }
2147
2148 static struct sysdev_class kvm_sysdev_class = {
2149         .name = "kvm",
2150         .suspend = kvm_suspend,
2151         .resume = kvm_resume,
2152 };
2153
2154 static struct sys_device kvm_sysdev = {
2155         .id = 0,
2156         .cls = &kvm_sysdev_class,
2157 };
2158
2159 struct page *bad_page;
2160 pfn_t bad_pfn;
2161
2162 static inline
2163 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2164 {
2165         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2166 }
2167
2168 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2169 {
2170         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2171
2172         kvm_arch_vcpu_load(vcpu, cpu);
2173 }
2174
2175 static void kvm_sched_out(struct preempt_notifier *pn,
2176                           struct task_struct *next)
2177 {
2178         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2179
2180         kvm_arch_vcpu_put(vcpu);
2181 }
2182
2183 int kvm_init(void *opaque, unsigned int vcpu_size,
2184                   struct module *module)
2185 {
2186         int r;
2187         int cpu;
2188
2189         r = kvm_arch_init(opaque);
2190         if (r)
2191                 goto out_fail;
2192
2193         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2194
2195         if (bad_page == NULL) {
2196                 r = -ENOMEM;
2197                 goto out;
2198         }
2199
2200         bad_pfn = page_to_pfn(bad_page);
2201
2202         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2203                 r = -ENOMEM;
2204                 goto out_free_0;
2205         }
2206
2207         r = kvm_arch_hardware_setup();
2208         if (r < 0)
2209                 goto out_free_0a;
2210
2211         for_each_online_cpu(cpu) {
2212                 smp_call_function_single(cpu,
2213                                 kvm_arch_check_processor_compat,
2214                                 &r, 1);
2215                 if (r < 0)
2216                         goto out_free_1;
2217         }
2218
2219         r = register_cpu_notifier(&kvm_cpu_notifier);
2220         if (r)
2221                 goto out_free_2;
2222         register_reboot_notifier(&kvm_reboot_notifier);
2223
2224         r = sysdev_class_register(&kvm_sysdev_class);
2225         if (r)
2226                 goto out_free_3;
2227
2228         r = sysdev_register(&kvm_sysdev);
2229         if (r)
2230                 goto out_free_4;
2231
2232         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2233         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
2234                                            __alignof__(struct kvm_vcpu),
2235                                            0, NULL);
2236         if (!kvm_vcpu_cache) {
2237                 r = -ENOMEM;
2238                 goto out_free_5;
2239         }
2240
2241         kvm_chardev_ops.owner = module;
2242         kvm_vm_fops.owner = module;
2243         kvm_vcpu_fops.owner = module;
2244
2245         r = misc_register(&kvm_dev);
2246         if (r) {
2247                 printk(KERN_ERR "kvm: misc device register failed\n");
2248                 goto out_free;
2249         }
2250
2251         kvm_preempt_ops.sched_in = kvm_sched_in;
2252         kvm_preempt_ops.sched_out = kvm_sched_out;
2253
2254         kvm_init_debug();
2255
2256         return 0;
2257
2258 out_free:
2259         kmem_cache_destroy(kvm_vcpu_cache);
2260 out_free_5:
2261         sysdev_unregister(&kvm_sysdev);
2262 out_free_4:
2263         sysdev_class_unregister(&kvm_sysdev_class);
2264 out_free_3:
2265         unregister_reboot_notifier(&kvm_reboot_notifier);
2266         unregister_cpu_notifier(&kvm_cpu_notifier);
2267 out_free_2:
2268 out_free_1:
2269         kvm_arch_hardware_unsetup();
2270 out_free_0a:
2271         free_cpumask_var(cpus_hardware_enabled);
2272 out_free_0:
2273         __free_page(bad_page);
2274 out:
2275         kvm_arch_exit();
2276 out_fail:
2277         return r;
2278 }
2279 EXPORT_SYMBOL_GPL(kvm_init);
2280
2281 void kvm_exit(void)
2282 {
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);