virt/kvm/kvm_main.c

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * This module enables machines with Intel VT-x extensions to run virtual
   5  * machines without emulation or binary translation.
   6  *
   7  * Copyright (C) 2006 Qumranet, Inc.
   8  * Copyright 2010 Red Hat, Inc. and/or its affilates.
   9  *
  10  * Authors:
  11  *   Avi Kivity   <avi@qumranet.com>
  12  *   Yaniv Kamay  <yaniv@qumranet.com>
  13  *
  14  * This work is licensed under the terms of the GNU GPL, version 2.  See
  15  * the COPYING file in the top-level directory.
  16  *
  17  */
  18
  19 #include "iodev.h"
  20
  21 #include <linux/kvm_host.h>
  22 #include <linux/kvm.h>
  23 #include <linux/module.h>
  24 #include <linux/errno.h>
  25 #include <linux/percpu.h>
  26 #include <linux/mm.h>
  27 #include <linux/miscdevice.h>
  28 #include <linux/vmalloc.h>
  29 #include <linux/reboot.h>
  30 #include <linux/debugfs.h>
  31 #include <linux/highmem.h>
  32 #include <linux/file.h>
  33 #include <linux/sysdev.h>
  34 #include <linux/cpu.h>
  35 #include <linux/sched.h>
  36 #include <linux/cpumask.h>
  37 #include <linux/smp.h>
  38 #include <linux/anon_inodes.h>
  39 #include <linux/profile.h>
  40 #include <linux/kvm_para.h>
  41 #include <linux/pagemap.h>
  42 #include <linux/mman.h>
  43 #include <linux/swap.h>
  44 #include <linux/bitops.h>
  45 #include <linux/spinlock.h>
  46 #include <linux/compat.h>
  47 #include <linux/srcu.h>
  48 #include <linux/hugetlb.h>
  49 #include <linux/slab.h>
  50
  51 #include <asm/processor.h>
  52 #include <asm/io.h>
  53 #include <asm/uaccess.h>
  54 #include <asm/pgtable.h>
  55 #include <asm-generic/bitops/le.h>
  56
  57 #include "coalesced_mmio.h"
  58
  59 #define CREATE_TRACE_POINTS
  60 #include <trace/events/kvm.h>
  61
  62 MODULE_AUTHOR("Qumranet");
  63 MODULE_LICENSE("GPL");
  64
  65 /*
  66  * Ordering of locks:
  67  *
  68  *              kvm->lock --> kvm->slots_lock --> kvm->irq_lock
  69  */
  70
  71 DEFINE_SPINLOCK(kvm_lock);
  72 LIST_HEAD(vm_list);
  73
  74 static cpumask_var_t cpus_hardware_enabled;
  75 static int kvm_usage_count = 0;
  76 static atomic_t hardware_enable_failed;
  77
  78 struct kmem_cache *kvm_vcpu_cache;
  79 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
  80
  81 static __read_mostly struct preempt_ops kvm_preempt_ops;
  82
  83 struct dentry *kvm_debugfs_dir;
  84
  85 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
  86                            unsigned long arg);
  87 static int hardware_enable_all(void);
  88 static void hardware_disable_all(void);
  89
  90 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
  91
  92 static bool kvm_rebooting;
  93
  94 static bool largepages_enabled = true;
  95
  96 struct page *hwpoison_page;
  97 pfn_t hwpoison_pfn;
  98
  99 inline int kvm_is_mmio_pfn(pfn_t pfn)
 100 {
 101         if (pfn_valid(pfn)) {
 102                 struct page *page = compound_head(pfn_to_page(pfn));
 103                 return PageReserved(page);
 104         }
 105
 106         return true;
 107 }
 108
 109 /*
 110  * Switches to specified vcpu, until a matching vcpu_put()
 111  */
 112 void vcpu_load(struct kvm_vcpu *vcpu)
 113 {
 114         int cpu;
 115
 116         mutex_lock(&vcpu->mutex);
 117         cpu = get_cpu();
 118         preempt_notifier_register(&vcpu->preempt_notifier);
 119         kvm_arch_vcpu_load(vcpu, cpu);
 120         put_cpu();
 121 }
 122
 123 void vcpu_put(struct kvm_vcpu *vcpu)
 124 {
 125         preempt_disable();
 126         kvm_arch_vcpu_put(vcpu);
 127         preempt_notifier_unregister(&vcpu->preempt_notifier);
 128         preempt_enable();
 129         mutex_unlock(&vcpu->mutex);
 130 }
 131
 132 static void ack_flush(void *_completed)
 133 {
 134 }
 135
 136 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
 137 {
 138         int i, cpu, me;
 139         cpumask_var_t cpus;
 140         bool called = true;
 141         struct kvm_vcpu *vcpu;
 142
 143         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
 144
 145         raw_spin_lock(&kvm->requests_lock);
 146         me = smp_processor_id();
 147         kvm_for_each_vcpu(i, vcpu, kvm) {
 148                 if (kvm_make_check_request(req, vcpu))
 149                         continue;
 150                 cpu = vcpu->cpu;
 151                 if (cpus != NULL && cpu != -1 && cpu != me)
 152                         cpumask_set_cpu(cpu, cpus);
 153         }
 154         if (unlikely(cpus == NULL))
 155                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
 156         else if (!cpumask_empty(cpus))
 157                 smp_call_function_many(cpus, ack_flush, NULL, 1);
 158         else
 159                 called = false;
 160         raw_spin_unlock(&kvm->requests_lock);
 161         free_cpumask_var(cpus);
 162         return called;
 163 }
 164
 165 void kvm_flush_remote_tlbs(struct kvm *kvm)
 166 {
 167         if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
 168                 ++kvm->stat.remote_tlb_flush;
 169 }
 170
 171 void kvm_reload_remote_mmus(struct kvm *kvm)
 172 {
 173         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
 174 }
 175
 176 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 177 {
 178         struct page *page;
 179         int r;
 180
 181         mutex_init(&vcpu->mutex);
 182         vcpu->cpu = -1;
 183         vcpu->kvm = kvm;
 184         vcpu->vcpu_id = id;
 185         init_waitqueue_head(&vcpu->wq);
 186
 187         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 188         if (!page) {
 189                 r = -ENOMEM;
 190                 goto fail;
 191         }
 192         vcpu->run = page_address(page);
 193
 194         r = kvm_arch_vcpu_init(vcpu);
 195         if (r < 0)
 196                 goto fail_free_run;
 197         return 0;
 198
 199 fail_free_run:
 200         free_page((unsigned long)vcpu->run);
 201 fail:
 202         return r;
 203 }
 204 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 205
 206 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 207 {
 208         kvm_arch_vcpu_uninit(vcpu);
 209         free_page((unsigned long)vcpu->run);
 210 }
 211 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 212
 213 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 214 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
 215 {
 216         return container_of(mn, struct kvm, mmu_notifier);
 217 }
 218
 219 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
 220                                              struct mm_struct *mm,
 221                                              unsigned long address)
 222 {
 223         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 224         int need_tlb_flush, idx;
 225
 226         /*
 227          * When ->invalidate_page runs, the linux pte has been zapped
 228          * already but the page is still allocated until
 229          * ->invalidate_page returns. So if we increase the sequence
 230          * here the kvm page fault will notice if the spte can't be
 231          * established because the page is going to be freed. If
 232          * instead the kvm page fault establishes the spte before
 233          * ->invalidate_page runs, kvm_unmap_hva will release it
 234          * before returning.
 235          *
 236          * The sequence increase only need to be seen at spin_unlock
 237          * time, and not at spin_lock time.
 238          *
 239          * Increasing the sequence after the spin_unlock would be
 240          * unsafe because the kvm page fault could then establish the
 241          * pte after kvm_unmap_hva returned, without noticing the page
 242          * is going to be freed.
 243          */
 244         idx = srcu_read_lock(&kvm->srcu);
 245         spin_lock(&kvm->mmu_lock);
 246         kvm->mmu_notifier_seq++;
 247         need_tlb_flush = kvm_unmap_hva(kvm, address);
 248         spin_unlock(&kvm->mmu_lock);
 249         srcu_read_unlock(&kvm->srcu, idx);
 250
 251         /* we've to flush the tlb before the pages can be freed */
 252         if (need_tlb_flush)
 253                 kvm_flush_remote_tlbs(kvm);
 254
 255 }
 256
 257 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
 258                                         struct mm_struct *mm,
 259                                         unsigned long address,
 260                                         pte_t pte)
 261 {
 262         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 263         int idx;
 264
 265         idx = srcu_read_lock(&kvm->srcu);
 266         spin_lock(&kvm->mmu_lock);
 267         kvm->mmu_notifier_seq++;
 268         kvm_set_spte_hva(kvm, address, pte);
 269         spin_unlock(&kvm->mmu_lock);
 270         srcu_read_unlock(&kvm->srcu, idx);
 271 }
 272
 273 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 274                                                     struct mm_struct *mm,
 275                                                     unsigned long start,
 276                                                     unsigned long end)
 277 {
 278         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 279         int need_tlb_flush = 0, idx;
 280
 281         idx = srcu_read_lock(&kvm->srcu);
 282         spin_lock(&kvm->mmu_lock);
 283         /*
 284          * The count increase must become visible at unlock time as no
 285          * spte can be established without taking the mmu_lock and
 286          * count is also read inside the mmu_lock critical section.
 287          */
 288         kvm->mmu_notifier_count++;
 289         for (; start < end; start += PAGE_SIZE)
 290                 need_tlb_flush |= kvm_unmap_hva(kvm, start);
 291         spin_unlock(&kvm->mmu_lock);
 292         srcu_read_unlock(&kvm->srcu, idx);
 293
 294         /* we've to flush the tlb before the pages can be freed */
 295         if (need_tlb_flush)
 296                 kvm_flush_remote_tlbs(kvm);
 297 }
 298
 299 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
 300                                                   struct mm_struct *mm,
 301                                                   unsigned long start,
 302                                                   unsigned long end)
 303 {
 304         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 305
 306         spin_lock(&kvm->mmu_lock);
 307         /*
 308          * This sequence increase will notify the kvm page fault that
 309          * the page that is going to be mapped in the spte could have
 310          * been freed.
 311          */
 312         kvm->mmu_notifier_seq++;
 313         /*
 314          * The above sequence increase must be visible before the
 315          * below count decrease but both values are read by the kvm
 316          * page fault under mmu_lock spinlock so we don't need to add
 317          * a smb_wmb() here in between the two.
 318          */
 319         kvm->mmu_notifier_count--;
 320         spin_unlock(&kvm->mmu_lock);
 321
 322         BUG_ON(kvm->mmu_notifier_count < 0);
 323 }
 324
 325 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
 326                                               struct mm_struct *mm,
 327                                               unsigned long address)
 328 {
 329         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 330         int young, idx;
 331
 332         idx = srcu_read_lock(&kvm->srcu);
 333         spin_lock(&kvm->mmu_lock);
 334         young = kvm_age_hva(kvm, address);
 335         spin_unlock(&kvm->mmu_lock);
 336         srcu_read_unlock(&kvm->srcu, idx);
 337
 338         if (young)
 339                 kvm_flush_remote_tlbs(kvm);
 340
 341         return young;
 342 }
 343
 344 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
 345                                      struct mm_struct *mm)
 346 {
 347         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 348         int idx;
 349
 350         idx = srcu_read_lock(&kvm->srcu);
 351         kvm_arch_flush_shadow(kvm);
 352         srcu_read_unlock(&kvm->srcu, idx);
 353 }
 354
 355 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 356         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
 357         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
 358         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
 359         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
 360         .change_pte             = kvm_mmu_notifier_change_pte,
 361         .release                = kvm_mmu_notifier_release,
 362 };
 363
 364 static int kvm_init_mmu_notifier(struct kvm *kvm)
 365 {
 366         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
 367         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
 368 }
 369
 370 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
 371
 372 static int kvm_init_mmu_notifier(struct kvm *kvm)
 373 {
 374         return 0;
 375 }
 376
 377 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
 378
 379 static struct kvm *kvm_create_vm(void)
 380 {
 381         int r = 0, i;
 382         struct kvm *kvm = kvm_arch_create_vm();
 383
 384         if (IS_ERR(kvm))
 385                 goto out;
 386
 387         r = hardware_enable_all();
 388         if (r)
 389                 goto out_err_nodisable;
 390
 391 #ifdef CONFIG_HAVE_KVM_IRQCHIP
 392         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
 393         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
 394 #endif
 395
 396         r = -ENOMEM;
 397         kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 398         if (!kvm->memslots)
 399                 goto out_err;
 400         if (init_srcu_struct(&kvm->srcu))
 401                 goto out_err;
 402         for (i = 0; i < KVM_NR_BUSES; i++) {
 403                 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
 404                                         GFP_KERNEL);
 405                 if (!kvm->buses[i]) {
 406                         cleanup_srcu_struct(&kvm->srcu);
 407                         goto out_err;
 408                 }
 409         }
 410
 411         r = kvm_init_mmu_notifier(kvm);
 412         if (r) {
 413                 cleanup_srcu_struct(&kvm->srcu);
 414                 goto out_err;
 415         }
 416
 417         kvm->mm = current->mm;
 418         atomic_inc(&kvm->mm->mm_count);
 419         spin_lock_init(&kvm->mmu_lock);
 420         raw_spin_lock_init(&kvm->requests_lock);
 421         kvm_eventfd_init(kvm);
 422         mutex_init(&kvm->lock);
 423         mutex_init(&kvm->irq_lock);
 424         mutex_init(&kvm->slots_lock);
 425         atomic_set(&kvm->users_count, 1);
 426         spin_lock(&kvm_lock);
 427         list_add(&kvm->vm_list, &vm_list);
 428         spin_unlock(&kvm_lock);
 429 out:
 430         return kvm;
 431
 432 out_err:
 433         hardware_disable_all();
 434 out_err_nodisable:
 435         for (i = 0; i < KVM_NR_BUSES; i++)
 436                 kfree(kvm->buses[i]);
 437         kfree(kvm->memslots);
 438         kfree(kvm);
 439         return ERR_PTR(r);
 440 }
 441
 442 /*
 443  * Free any memory in @free but not in @dont.
 444  */
 445 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 446                                   struct kvm_memory_slot *dont)
 447 {
 448         int i;
 449
 450         if (!dont || free->rmap != dont->rmap)
 451                 vfree(free->rmap);
 452
 453         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 454                 vfree(free->dirty_bitmap);
 455
 456
 457         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
 458                 if (!dont || free->lpage_info[i] != dont->lpage_info[i]) {
 459                         vfree(free->lpage_info[i]);
 460                         free->lpage_info[i] = NULL;
 461                 }
 462         }
 463
 464         free->npages = 0;
 465         free->dirty_bitmap = NULL;
 466         free->rmap = NULL;
 467 }
 468
 469 void kvm_free_physmem(struct kvm *kvm)
 470 {
 471         int i;
 472         struct kvm_memslots *slots = kvm->memslots;
 473
 474         for (i = 0; i < slots->nmemslots; ++i)
 475                 kvm_free_physmem_slot(&slots->memslots[i], NULL);
 476
 477         kfree(kvm->memslots);
 478 }
 479
 480 static void kvm_destroy_vm(struct kvm *kvm)
 481 {
 482         int i;
 483         struct mm_struct *mm = kvm->mm;
 484
 485         kvm_arch_sync_events(kvm);
 486         spin_lock(&kvm_lock);
 487         list_del(&kvm->vm_list);
 488         spin_unlock(&kvm_lock);
 489         kvm_free_irq_routing(kvm);
 490         for (i = 0; i < KVM_NR_BUSES; i++)
 491                 kvm_io_bus_destroy(kvm->buses[i]);
 492         kvm_coalesced_mmio_free(kvm);
 493 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 494         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
 495 #else
 496         kvm_arch_flush_shadow(kvm);
 497 #endif
 498         kvm_arch_destroy_vm(kvm);
 499         hardware_disable_all();
 500         mmdrop(mm);
 501 }
 502
 503 void kvm_get_kvm(struct kvm *kvm)
 504 {
 505         atomic_inc(&kvm->users_count);
 506 }
 507 EXPORT_SYMBOL_GPL(kvm_get_kvm);
 508
 509 void kvm_put_kvm(struct kvm *kvm)
 510 {
 511         if (atomic_dec_and_test(&kvm->users_count))
 512                 kvm_destroy_vm(kvm);
 513 }
 514 EXPORT_SYMBOL_GPL(kvm_put_kvm);
 515
 516
 517 static int kvm_vm_release(struct inode *inode, struct file *filp)
 518 {
 519         struct kvm *kvm = filp->private_data;
 520
 521         kvm_irqfd_release(kvm);
 522
 523         kvm_put_kvm(kvm);
 524         return 0;
 525 }
 526
 527 /*
 528  * Allocate some memory and give it an address in the guest physical address
 529  * space.
 530  *
 531  * Discontiguous memory is allowed, mostly for framebuffers.
 532  *
 533  * Must be called holding mmap_sem for write.
 534  */
 535 int __kvm_set_memory_region(struct kvm *kvm,
 536                             struct kvm_userspace_memory_region *mem,
 537                             int user_alloc)
 538 {
 539         int r, flush_shadow = 0;
 540         gfn_t base_gfn;
 541         unsigned long npages;
 542         unsigned long i;
 543         struct kvm_memory_slot *memslot;
 544         struct kvm_memory_slot old, new;
 545         struct kvm_memslots *slots, *old_memslots;
 546
 547         r = -EINVAL;
 548         /* General sanity checks */
 549         if (mem->memory_size & (PAGE_SIZE - 1))
 550                 goto out;
 551         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 552                 goto out;
 553         if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
 554                 goto out;
 555         if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
 556                 goto out;
 557         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 558                 goto out;
 559
 560         memslot = &kvm->memslots->memslots[mem->slot];
 561         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 562         npages = mem->memory_size >> PAGE_SHIFT;
 563
 564         r = -EINVAL;
 565         if (npages > KVM_MEM_MAX_NR_PAGES)
 566                 goto out;
 567
 568         if (!npages)
 569                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 570
 571         new = old = *memslot;
 572
 573         new.base_gfn = base_gfn;
 574         new.npages = npages;
 575         new.flags = mem->flags;
 576
 577         /* Disallow changing a memory slot's size. */
 578         r = -EINVAL;
 579         if (npages && old.npages && npages != old.npages)
 580                 goto out_free;
 581
 582         /* Check for overlaps */
 583         r = -EEXIST;
 584         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 585                 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
 586
 587                 if (s == memslot || !s->npages)
 588                         continue;
 589                 if (!((base_gfn + npages <= s->base_gfn) ||
 590                       (base_gfn >= s->base_gfn + s->npages)))
 591                         goto out_free;
 592         }
 593
 594         /* Free page dirty bitmap if unneeded */
 595         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 596                 new.dirty_bitmap = NULL;
 597
 598         r = -ENOMEM;
 599
 600         /* Allocate if a slot is being created */
 601 #ifndef CONFIG_S390
 602         if (npages && !new.rmap) {
 603                 new.rmap = vmalloc(npages * sizeof(*new.rmap));
 604
 605                 if (!new.rmap)
 606                         goto out_free;
 607
 608                 memset(new.rmap, 0, npages * sizeof(*new.rmap));
 609
 610                 new.user_alloc = user_alloc;
 611                 new.userspace_addr = mem->userspace_addr;
 612         }
 613         if (!npages)
 614                 goto skip_lpage;
 615
 616         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
 617                 unsigned long ugfn;
 618                 unsigned long j;
 619                 int lpages;
 620                 int level = i + 2;
 621
 622                 /* Avoid unused variable warning if no large pages */
 623                 (void)level;
 624
 625                 if (new.lpage_info[i])
 626                         continue;
 627
 628                 lpages = 1 + (base_gfn + npages - 1) /
 629                              KVM_PAGES_PER_HPAGE(level);
 630                 lpages -= base_gfn / KVM_PAGES_PER_HPAGE(level);
 631
 632                 new.lpage_info[i] = vmalloc(lpages * sizeof(*new.lpage_info[i]));
 633
 634                 if (!new.lpage_info[i])
 635                         goto out_free;
 636
 637                 memset(new.lpage_info[i], 0,
 638                        lpages * sizeof(*new.lpage_info[i]));
 639
 640                 if (base_gfn % KVM_PAGES_PER_HPAGE(level))
 641                         new.lpage_info[i][0].write_count = 1;
 642                 if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE(level))
 643                         new.lpage_info[i][lpages - 1].write_count = 1;
 644                 ugfn = new.userspace_addr >> PAGE_SHIFT;
 645                 /*
 646                  * If the gfn and userspace address are not aligned wrt each
 647                  * other, or if explicitly asked to, disable large page
 648                  * support for this slot
 649                  */
 650                 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
 651                     !largepages_enabled)
 652                         for (j = 0; j < lpages; ++j)
 653                                 new.lpage_info[i][j].write_count = 1;
 654         }
 655
 656 skip_lpage:
 657
 658         /* Allocate page dirty bitmap if needed */
 659         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 660                 unsigned long dirty_bytes = kvm_dirty_bitmap_bytes(&new);
 661
 662                 new.dirty_bitmap = vmalloc(dirty_bytes);
 663                 if (!new.dirty_bitmap)
 664                         goto out_free;
 665                 memset(new.dirty_bitmap, 0, dirty_bytes);
 666                 /* destroy any largepage mappings for dirty tracking */
 667                 if (old.npages)
 668                         flush_shadow = 1;
 669         }
 670 #else  /* not defined CONFIG_S390 */
 671         new.user_alloc = user_alloc;
 672         if (user_alloc)
 673                 new.userspace_addr = mem->userspace_addr;
 674 #endif /* not defined CONFIG_S390 */
 675
 676         if (!npages) {
 677                 r = -ENOMEM;
 678                 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 679                 if (!slots)
 680                         goto out_free;
 681                 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
 682                 if (mem->slot >= slots->nmemslots)
 683                         slots->nmemslots = mem->slot + 1;
 684                 slots->memslots[mem->slot].flags |= KVM_MEMSLOT_INVALID;
 685
 686                 old_memslots = kvm->memslots;
 687                 rcu_assign_pointer(kvm->memslots, slots);
 688                 synchronize_srcu_expedited(&kvm->srcu);
 689                 /* From this point no new shadow pages pointing to a deleted
 690                  * memslot will be created.
 691                  *
 692                  * validation of sp->gfn happens in:
 693                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
 694                  *      - kvm_is_visible_gfn (mmu_check_roots)
 695                  */
 696                 kvm_arch_flush_shadow(kvm);
 697                 kfree(old_memslots);
 698         }
 699
 700         r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
 701         if (r)
 702                 goto out_free;
 703
 704 #ifdef CONFIG_DMAR
 705         /* map the pages in iommu page table */
 706         if (npages) {
 707                 r = kvm_iommu_map_pages(kvm, &new);
 708                 if (r)
 709                         goto out_free;
 710         }
 711 #endif
 712
 713         r = -ENOMEM;
 714         slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 715         if (!slots)
 716                 goto out_free;
 717         memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
 718         if (mem->slot >= slots->nmemslots)
 719                 slots->nmemslots = mem->slot + 1;
 720
 721         /* actual memory is freed via old in kvm_free_physmem_slot below */
 722         if (!npages) {
 723                 new.rmap = NULL;
 724                 new.dirty_bitmap = NULL;
 725                 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i)
 726                         new.lpage_info[i] = NULL;
 727         }
 728
 729         slots->memslots[mem->slot] = new;
 730         old_memslots = kvm->memslots;
 731         rcu_assign_pointer(kvm->memslots, slots);
 732         synchronize_srcu_expedited(&kvm->srcu);
 733
 734         kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
 735
 736         kvm_free_physmem_slot(&old, &new);
 737         kfree(old_memslots);
 738
 739         if (flush_shadow)
 740                 kvm_arch_flush_shadow(kvm);
 741
 742         return 0;
 743
 744 out_free:
 745         kvm_free_physmem_slot(&new, &old);
 746 out:
 747         return r;
 748
 749 }
 750 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 751
 752 int kvm_set_memory_region(struct kvm *kvm,
 753                           struct kvm_userspace_memory_region *mem,
 754                           int user_alloc)
 755 {
 756         int r;
 757
 758         mutex_lock(&kvm->slots_lock);
 759         r = __kvm_set_memory_region(kvm, mem, user_alloc);
 760         mutex_unlock(&kvm->slots_lock);
 761         return r;
 762 }
 763 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 764
 765 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 766                                    struct
 767                                    kvm_userspace_memory_region *mem,
 768                                    int user_alloc)
 769 {
 770         if (mem->slot >= KVM_MEMORY_SLOTS)
 771                 return -EINVAL;
 772         return kvm_set_memory_region(kvm, mem, user_alloc);
 773 }
 774
 775 int kvm_get_dirty_log(struct kvm *kvm,
 776                         struct kvm_dirty_log *log, int *is_dirty)
 777 {
 778         struct kvm_memory_slot *memslot;
 779         int r, i;
 780         unsigned long n;
 781         unsigned long any = 0;
 782
 783         r = -EINVAL;
 784         if (log->slot >= KVM_MEMORY_SLOTS)
 785                 goto out;
 786
 787         memslot = &kvm->memslots->memslots[log->slot];
 788         r = -ENOENT;
 789         if (!memslot->dirty_bitmap)
 790                 goto out;
 791
 792         n = kvm_dirty_bitmap_bytes(memslot);
 793
 794         for (i = 0; !any && i < n/sizeof(long); ++i)
 795                 any = memslot->dirty_bitmap[i];
 796
 797         r = -EFAULT;
 798         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 799                 goto out;
 800
 801         if (any)
 802                 *is_dirty = 1;
 803
 804         r = 0;
 805 out:
 806         return r;
 807 }
 808
 809 void kvm_disable_largepages(void)
 810 {
 811         largepages_enabled = false;
 812 }
 813 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
 814
 815 int is_error_page(struct page *page)
 816 {
 817         return page == bad_page || page == hwpoison_page;
 818 }
 819 EXPORT_SYMBOL_GPL(is_error_page);
 820
 821 int is_error_pfn(pfn_t pfn)
 822 {
 823         return pfn == bad_pfn || pfn == hwpoison_pfn;
 824 }
 825 EXPORT_SYMBOL_GPL(is_error_pfn);
 826
 827 int is_hwpoison_pfn(pfn_t pfn)
 828 {
 829         return pfn == hwpoison_pfn;
 830 }
 831 EXPORT_SYMBOL_GPL(is_hwpoison_pfn);
 832
 833 static inline unsigned long bad_hva(void)
 834 {
 835         return PAGE_OFFSET;
 836 }
 837
 838 int kvm_is_error_hva(unsigned long addr)
 839 {
 840         return addr == bad_hva();
 841 }
 842 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
 843
 844 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 845 {
 846         int i;
 847         struct kvm_memslots *slots = kvm_memslots(kvm);
 848
 849         for (i = 0; i < slots->nmemslots; ++i) {
 850                 struct kvm_memory_slot *memslot = &slots->memslots[i];
 851
 852                 if (gfn >= memslot->base_gfn
 853                     && gfn < memslot->base_gfn + memslot->npages)
 854                         return memslot;
 855         }
 856         return NULL;
 857 }
 858 EXPORT_SYMBOL_GPL(gfn_to_memslot);
 859
 860 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
 861 {
 862         int i;
 863         struct kvm_memslots *slots = kvm_memslots(kvm);
 864
 865         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 866                 struct kvm_memory_slot *memslot = &slots->memslots[i];
 867
 868                 if (memslot->flags & KVM_MEMSLOT_INVALID)
 869                         continue;
 870
 871                 if (gfn >= memslot->base_gfn
 872                     && gfn < memslot->base_gfn + memslot->npages)
 873                         return 1;
 874         }
 875         return 0;
 876 }
 877 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
 878
 879 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
 880 {
 881         struct vm_area_struct *vma;
 882         unsigned long addr, size;
 883
 884         size = PAGE_SIZE;
 885
 886         addr = gfn_to_hva(kvm, gfn);
 887         if (kvm_is_error_hva(addr))
 888                 return PAGE_SIZE;
 889
 890         down_read(&current->mm->mmap_sem);
 891         vma = find_vma(current->mm, addr);
 892         if (!vma)
 893                 goto out;
 894
 895         size = vma_kernel_pagesize(vma);
 896
 897 out:
 898         up_read(&current->mm->mmap_sem);
 899
 900         return size;
 901 }
 902
 903 int memslot_id(struct kvm *kvm, gfn_t gfn)
 904 {
 905         int i;
 906         struct kvm_memslots *slots = kvm_memslots(kvm);
 907         struct kvm_memory_slot *memslot = NULL;
 908
 909         for (i = 0; i < slots->nmemslots; ++i) {
 910                 memslot = &slots->memslots[i];
 911
 912                 if (gfn >= memslot->base_gfn
 913                     && gfn < memslot->base_gfn + memslot->npages)
 914                         break;
 915         }
 916
 917         return memslot - slots->memslots;
 918 }
 919
 920 static unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
 921 {
 922         return slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE;
 923 }
 924
 925 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
 926 {
 927         struct kvm_memory_slot *slot;
 928
 929         slot = gfn_to_memslot(kvm, gfn);
 930         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
 931                 return bad_hva();
 932         return gfn_to_hva_memslot(slot, gfn);
 933 }
 934 EXPORT_SYMBOL_GPL(gfn_to_hva);
 935
 936 static pfn_t hva_to_pfn(struct kvm *kvm, unsigned long addr)
 937 {
 938         struct page *page[1];
 939         int npages;
 940         pfn_t pfn;
 941
 942         might_sleep();
 943
 944         npages = get_user_pages_fast(addr, 1, 1, page);
 945
 946         if (unlikely(npages != 1)) {
 947                 struct vm_area_struct *vma;
 948
 949                 if (is_hwpoison_address(addr)) {
 950                         get_page(hwpoison_page);
 951                         return page_to_pfn(hwpoison_page);
 952                 }
 953
 954                 down_read(&current->mm->mmap_sem);
 955                 vma = find_vma(current->mm, addr);
 956
 957                 if (vma == NULL || addr < vma->vm_start ||
 958                     !(vma->vm_flags & VM_PFNMAP)) {
 959                         up_read(&current->mm->mmap_sem);
 960                         get_page(bad_page);
 961                         return page_to_pfn(bad_page);
 962                 }
 963
 964                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
 965                 up_read(&current->mm->mmap_sem);
 966                 BUG_ON(!kvm_is_mmio_pfn(pfn));
 967         } else
 968                 pfn = page_to_pfn(page[0]);
 969
 970         return pfn;
 971 }
 972
 973 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
 974 {
 975         unsigned long addr;
 976
 977         addr = gfn_to_hva(kvm, gfn);
 978         if (kvm_is_error_hva(addr)) {
 979                 get_page(bad_page);
 980                 return page_to_pfn(bad_page);
 981         }
 982
 983         return hva_to_pfn(kvm, addr);
 984 }
 985 EXPORT_SYMBOL_GPL(gfn_to_pfn);
 986
 987 pfn_t gfn_to_pfn_memslot(struct kvm *kvm,
 988                          struct kvm_memory_slot *slot, gfn_t gfn)
 989 {
 990         unsigned long addr = gfn_to_hva_memslot(slot, gfn);
 991         return hva_to_pfn(kvm, addr);
 992 }
 993
 994 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
 995 {
 996         pfn_t pfn;
 997
 998         pfn = gfn_to_pfn(kvm, gfn);
 999         if (!kvm_is_mmio_pfn(pfn))
1000                 return pfn_to_page(pfn);
1001
1002         WARN_ON(kvm_is_mmio_pfn(pfn));
1003
1004         get_page(bad_page);
1005         return bad_page;
1006 }
1007
1008 EXPORT_SYMBOL_GPL(gfn_to_page);
1009
1010 void kvm_release_page_clean(struct page *page)
1011 {
1012         kvm_release_pfn_clean(page_to_pfn(page));
1013 }
1014 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1015
1016 void kvm_release_pfn_clean(pfn_t pfn)
1017 {
1018         if (!kvm_is_mmio_pfn(pfn))
1019                 put_page(pfn_to_page(pfn));
1020 }
1021 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1022
1023 void kvm_release_page_dirty(struct page *page)
1024 {
1025         kvm_release_pfn_dirty(page_to_pfn(page));
1026 }
1027 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1028
1029 void kvm_release_pfn_dirty(pfn_t pfn)
1030 {
1031         kvm_set_pfn_dirty(pfn);
1032         kvm_release_pfn_clean(pfn);
1033 }
1034 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1035
1036 void kvm_set_page_dirty(struct page *page)
1037 {
1038         kvm_set_pfn_dirty(page_to_pfn(page));
1039 }
1040 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1041
1042 void kvm_set_pfn_dirty(pfn_t pfn)
1043 {
1044         if (!kvm_is_mmio_pfn(pfn)) {
1045                 struct page *page = pfn_to_page(pfn);
1046                 if (!PageReserved(page))
1047                         SetPageDirty(page);
1048         }
1049 }
1050 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1051
1052 void kvm_set_pfn_accessed(pfn_t pfn)
1053 {
1054         if (!kvm_is_mmio_pfn(pfn))
1055                 mark_page_accessed(pfn_to_page(pfn));
1056 }
1057 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1058
1059 void kvm_get_pfn(pfn_t pfn)
1060 {
1061         if (!kvm_is_mmio_pfn(pfn))
1062                 get_page(pfn_to_page(pfn));
1063 }
1064 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1065
1066 static int next_segment(unsigned long len, int offset)
1067 {
1068         if (len > PAGE_SIZE - offset)
1069                 return PAGE_SIZE - offset;
1070         else
1071                 return len;
1072 }
1073
1074 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1075                         int len)
1076 {
1077         int r;
1078         unsigned long addr;
1079
1080         addr = gfn_to_hva(kvm, gfn);
1081         if (kvm_is_error_hva(addr))
1082                 return -EFAULT;
1083         r = copy_from_user(data, (void __user *)addr + offset, len);
1084         if (r)
1085                 return -EFAULT;
1086         return 0;
1087 }
1088 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1089
1090 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1091 {
1092         gfn_t gfn = gpa >> PAGE_SHIFT;
1093         int seg;
1094         int offset = offset_in_page(gpa);
1095         int ret;
1096
1097         while ((seg = next_segment(len, offset)) != 0) {
1098                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1099                 if (ret < 0)
1100                         return ret;
1101                 offset = 0;
1102                 len -= seg;
1103                 data += seg;
1104                 ++gfn;
1105         }
1106         return 0;
1107 }
1108 EXPORT_SYMBOL_GPL(kvm_read_guest);
1109
1110 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1111                           unsigned long len)
1112 {
1113         int r;
1114         unsigned long addr;
1115         gfn_t gfn = gpa >> PAGE_SHIFT;
1116         int offset = offset_in_page(gpa);
1117
1118         addr = gfn_to_hva(kvm, gfn);
1119         if (kvm_is_error_hva(addr))
1120                 return -EFAULT;
1121         pagefault_disable();
1122         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1123         pagefault_enable();
1124         if (r)
1125                 return -EFAULT;
1126         return 0;
1127 }
1128 EXPORT_SYMBOL(kvm_read_guest_atomic);
1129
1130 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1131                          int offset, int len)
1132 {
1133         int r;
1134         unsigned long addr;
1135
1136         addr = gfn_to_hva(kvm, gfn);
1137         if (kvm_is_error_hva(addr))
1138                 return -EFAULT;
1139         r = copy_to_user((void __user *)addr + offset, data, len);
1140         if (r)
1141                 return -EFAULT;
1142         mark_page_dirty(kvm, gfn);
1143         return 0;
1144 }
1145 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1146
1147 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1148                     unsigned long len)
1149 {
1150         gfn_t gfn = gpa >> PAGE_SHIFT;
1151         int seg;
1152         int offset = offset_in_page(gpa);
1153         int ret;
1154
1155         while ((seg = next_segment(len, offset)) != 0) {
1156                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1157                 if (ret < 0)
1158                         return ret;
1159                 offset = 0;
1160                 len -= seg;
1161                 data += seg;
1162                 ++gfn;
1163         }
1164         return 0;
1165 }
1166
1167 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1168 {
1169         return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
1170 }
1171 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1172
1173 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1174 {
1175         gfn_t gfn = gpa >> PAGE_SHIFT;
1176         int seg;
1177         int offset = offset_in_page(gpa);
1178         int ret;
1179
1180         while ((seg = next_segment(len, offset)) != 0) {
1181                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1182                 if (ret < 0)
1183                         return ret;
1184                 offset = 0;
1185                 len -= seg;
1186                 ++gfn;
1187         }
1188         return 0;
1189 }
1190 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1191
1192 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1193 {
1194         struct kvm_memory_slot *memslot;
1195
1196         memslot = gfn_to_memslot(kvm, gfn);
1197         if (memslot && memslot->dirty_bitmap) {
1198                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1199
1200                 generic___set_le_bit(rel_gfn, memslot->dirty_bitmap);
1201         }
1202 }
1203
1204 /*
1205  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1206  */
1207 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1208 {
1209         DEFINE_WAIT(wait);
1210
1211         for (;;) {
1212                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1213
1214                 if (kvm_arch_vcpu_runnable(vcpu)) {
1215                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1216                         break;
1217                 }
1218                 if (kvm_cpu_has_pending_timer(vcpu))
1219                         break;
1220                 if (signal_pending(current))
1221                         break;
1222
1223                 schedule();
1224         }
1225
1226         finish_wait(&vcpu->wq, &wait);
1227 }
1228
1229 void kvm_resched(struct kvm_vcpu *vcpu)
1230 {
1231         if (!need_resched())
1232                 return;
1233         cond_resched();
1234 }
1235 EXPORT_SYMBOL_GPL(kvm_resched);
1236
1237 void kvm_vcpu_on_spin(struct kvm_vcpu *vcpu)
1238 {
1239         ktime_t expires;
1240         DEFINE_WAIT(wait);
1241
1242         prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1243
1244         /* Sleep for 100 us, and hope lock-holder got scheduled */
1245         expires = ktime_add_ns(ktime_get(), 100000UL);
1246         schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
1247
1248         finish_wait(&vcpu->wq, &wait);
1249 }
1250 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1251
1252 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1253 {
1254         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1255         struct page *page;
1256
1257         if (vmf->pgoff == 0)
1258                 page = virt_to_page(vcpu->run);
1259 #ifdef CONFIG_X86
1260         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1261                 page = virt_to_page(vcpu->arch.pio_data);
1262 #endif
1263 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1264         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1265                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1266 #endif
1267         else
1268                 return VM_FAULT_SIGBUS;
1269         get_page(page);
1270         vmf->page = page;
1271         return 0;
1272 }
1273
1274 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1275         .fault = kvm_vcpu_fault,
1276 };
1277
1278 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1279 {
1280         vma->vm_ops = &kvm_vcpu_vm_ops;
1281         return 0;
1282 }
1283
1284 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1285 {
1286         struct kvm_vcpu *vcpu = filp->private_data;
1287
1288         kvm_put_kvm(vcpu->kvm);
1289         return 0;
1290 }
1291
1292 static struct file_operations kvm_vcpu_fops = {
1293         .release        = kvm_vcpu_release,
1294         .unlocked_ioctl = kvm_vcpu_ioctl,
1295         .compat_ioctl   = kvm_vcpu_ioctl,
1296         .mmap           = kvm_vcpu_mmap,
1297 };
1298
1299 /*
1300  * Allocates an inode for the vcpu.
1301  */
1302 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1303 {
1304         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1305 }
1306
1307 /*
1308  * Creates some virtual cpus.  Good luck creating more than one.
1309  */
1310 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1311 {
1312         int r;
1313         struct kvm_vcpu *vcpu, *v;
1314
1315         vcpu = kvm_arch_vcpu_create(kvm, id);
1316         if (IS_ERR(vcpu))
1317                 return PTR_ERR(vcpu);
1318
1319         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1320
1321         r = kvm_arch_vcpu_setup(vcpu);
1322         if (r)
1323                 return r;
1324
1325         mutex_lock(&kvm->lock);
1326         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1327                 r = -EINVAL;
1328                 goto vcpu_destroy;
1329         }
1330
1331         kvm_for_each_vcpu(r, v, kvm)
1332                 if (v->vcpu_id == id) {
1333                         r = -EEXIST;
1334                         goto vcpu_destroy;
1335                 }
1336
1337         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1338
1339         /* Now it's all set up, let userspace reach it */
1340         kvm_get_kvm(kvm);
1341         r = create_vcpu_fd(vcpu);
1342         if (r < 0) {
1343                 kvm_put_kvm(kvm);
1344                 goto vcpu_destroy;
1345         }
1346
1347         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1348         smp_wmb();
1349         atomic_inc(&kvm->online_vcpus);
1350
1351 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1352         if (kvm->bsp_vcpu_id == id)
1353                 kvm->bsp_vcpu = vcpu;
1354 #endif
1355         mutex_unlock(&kvm->lock);
1356         return r;
1357
1358 vcpu_destroy:
1359         mutex_unlock(&kvm->lock);
1360         kvm_arch_vcpu_destroy(vcpu);
1361         return r;
1362 }
1363
1364 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1365 {
1366         if (sigset) {
1367                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1368                 vcpu->sigset_active = 1;
1369                 vcpu->sigset = *sigset;
1370         } else
1371                 vcpu->sigset_active = 0;
1372         return 0;
1373 }
1374
1375 static long kvm_vcpu_ioctl(struct file *filp,
1376                            unsigned int ioctl, unsigned long arg)
1377 {
1378         struct kvm_vcpu *vcpu = filp->private_data;
1379         void __user *argp = (void __user *)arg;
1380         int r;
1381         struct kvm_fpu *fpu = NULL;
1382         struct kvm_sregs *kvm_sregs = NULL;
1383
1384         if (vcpu->kvm->mm != current->mm)
1385                 return -EIO;
1386
1387 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1388         /*
1389          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1390          * so vcpu_load() would break it.
1391          */
1392         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1393                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1394 #endif
1395
1396
1397         vcpu_load(vcpu);
1398         switch (ioctl) {
1399         case KVM_RUN:
1400                 r = -EINVAL;
1401                 if (arg)
1402                         goto out;
1403                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1404                 break;
1405         case KVM_GET_REGS: {
1406                 struct kvm_regs *kvm_regs;
1407
1408                 r = -ENOMEM;
1409                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1410                 if (!kvm_regs)
1411                         goto out;
1412                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1413                 if (r)
1414                         goto out_free1;
1415                 r = -EFAULT;
1416                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1417                         goto out_free1;
1418                 r = 0;
1419 out_free1:
1420                 kfree(kvm_regs);
1421                 break;
1422         }
1423         case KVM_SET_REGS: {
1424                 struct kvm_regs *kvm_regs;
1425
1426                 r = -ENOMEM;
1427                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1428                 if (!kvm_regs)
1429                         goto out;
1430                 r = -EFAULT;
1431                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1432                         goto out_free2;
1433                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1434                 if (r)
1435                         goto out_free2;
1436                 r = 0;
1437 out_free2:
1438                 kfree(kvm_regs);
1439                 break;
1440         }
1441         case KVM_GET_SREGS: {
1442                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1443                 r = -ENOMEM;
1444                 if (!kvm_sregs)
1445                         goto out;
1446                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1447                 if (r)
1448                         goto out;
1449                 r = -EFAULT;
1450                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1451                         goto out;
1452                 r = 0;
1453                 break;
1454         }
1455         case KVM_SET_SREGS: {
1456                 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1457                 r = -ENOMEM;
1458                 if (!kvm_sregs)
1459                         goto out;
1460                 r = -EFAULT;
1461                 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1462                         goto out;
1463                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1464                 if (r)
1465                         goto out;
1466                 r = 0;
1467                 break;
1468         }
1469         case KVM_GET_MP_STATE: {
1470                 struct kvm_mp_state mp_state;
1471
1472                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1473                 if (r)
1474                         goto out;
1475                 r = -EFAULT;
1476                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1477                         goto out;
1478                 r = 0;
1479                 break;
1480         }
1481         case KVM_SET_MP_STATE: {
1482                 struct kvm_mp_state mp_state;
1483
1484                 r = -EFAULT;
1485                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1486                         goto out;
1487                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1488                 if (r)
1489                         goto out;
1490                 r = 0;
1491                 break;
1492         }
1493         case KVM_TRANSLATE: {
1494                 struct kvm_translation tr;
1495
1496                 r = -EFAULT;
1497                 if (copy_from_user(&tr, argp, sizeof tr))
1498                         goto out;
1499                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1500                 if (r)
1501                         goto out;
1502                 r = -EFAULT;
1503                 if (copy_to_user(argp, &tr, sizeof tr))
1504                         goto out;
1505                 r = 0;
1506                 break;
1507         }
1508         case KVM_SET_GUEST_DEBUG: {
1509                 struct kvm_guest_debug dbg;
1510
1511                 r = -EFAULT;
1512                 if (copy_from_user(&dbg, argp, sizeof dbg))
1513                         goto out;
1514                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1515                 if (r)
1516                         goto out;
1517                 r = 0;
1518                 break;
1519         }
1520         case KVM_SET_SIGNAL_MASK: {
1521                 struct kvm_signal_mask __user *sigmask_arg = argp;
1522                 struct kvm_signal_mask kvm_sigmask;
1523                 sigset_t sigset, *p;
1524
1525                 p = NULL;
1526                 if (argp) {
1527                         r = -EFAULT;
1528                         if (copy_from_user(&kvm_sigmask, argp,
1529                                            sizeof kvm_sigmask))
1530                                 goto out;
1531                         r = -EINVAL;
1532                         if (kvm_sigmask.len != sizeof sigset)
1533                                 goto out;
1534                         r = -EFAULT;
1535                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1536                                            sizeof sigset))
1537                                 goto out;
1538                         p = &sigset;
1539                 }
1540                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
1541                 break;
1542         }
1543         case KVM_GET_FPU: {
1544                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1545                 r = -ENOMEM;
1546                 if (!fpu)
1547                         goto out;
1548                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1549                 if (r)
1550                         goto out;
1551                 r = -EFAULT;
1552                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1553                         goto out;
1554                 r = 0;
1555                 break;
1556         }
1557         case KVM_SET_FPU: {
1558                 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1559                 r = -ENOMEM;
1560                 if (!fpu)
1561                         goto out;
1562                 r = -EFAULT;
1563                 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1564                         goto out;
1565                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1566                 if (r)
1567                         goto out;
1568                 r = 0;
1569                 break;
1570         }
1571         default:
1572                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1573         }
1574 out:
1575         vcpu_put(vcpu);
1576         kfree(fpu);
1577         kfree(kvm_sregs);
1578         return r;
1579 }
1580
1581 static long kvm_vm_ioctl(struct file *filp,
1582                            unsigned int ioctl, unsigned long arg)
1583 {
1584         struct kvm *kvm = filp->private_data;
1585         void __user *argp = (void __user *)arg;
1586         int r;
1587
1588         if (kvm->mm != current->mm)
1589                 return -EIO;
1590         switch (ioctl) {
1591         case KVM_CREATE_VCPU:
1592                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1593                 if (r < 0)
1594                         goto out;
1595                 break;
1596         case KVM_SET_USER_MEMORY_REGION: {
1597                 struct kvm_userspace_memory_region kvm_userspace_mem;
1598
1599                 r = -EFAULT;
1600                 if (copy_from_user(&kvm_userspace_mem, argp,
1601                                                 sizeof kvm_userspace_mem))
1602                         goto out;
1603
1604                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1605                 if (r)
1606                         goto out;
1607                 break;
1608         }
1609         case KVM_GET_DIRTY_LOG: {
1610                 struct kvm_dirty_log log;
1611
1612                 r = -EFAULT;
1613                 if (copy_from_user(&log, argp, sizeof log))
1614                         goto out;
1615                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1616                 if (r)
1617                         goto out;
1618                 break;
1619         }
1620 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1621         case KVM_REGISTER_COALESCED_MMIO: {
1622                 struct kvm_coalesced_mmio_zone zone;
1623                 r = -EFAULT;
1624                 if (copy_from_user(&zone, argp, sizeof zone))
1625                         goto out;
1626                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1627                 if (r)
1628                         goto out;
1629                 r = 0;
1630                 break;
1631         }
1632         case KVM_UNREGISTER_COALESCED_MMIO: {
1633                 struct kvm_coalesced_mmio_zone zone;
1634                 r = -EFAULT;
1635                 if (copy_from_user(&zone, argp, sizeof zone))
1636                         goto out;
1637                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1638                 if (r)
1639                         goto out;
1640                 r = 0;
1641                 break;
1642         }
1643 #endif
1644         case KVM_IRQFD: {
1645                 struct kvm_irqfd data;
1646
1647                 r = -EFAULT;
1648                 if (copy_from_user(&data, argp, sizeof data))
1649                         goto out;
1650                 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
1651                 break;
1652         }
1653         case KVM_IOEVENTFD: {
1654                 struct kvm_ioeventfd data;
1655
1656                 r = -EFAULT;
1657                 if (copy_from_user(&data, argp, sizeof data))
1658                         goto out;
1659                 r = kvm_ioeventfd(kvm, &data);
1660                 break;
1661         }
1662 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1663         case KVM_SET_BOOT_CPU_ID:
1664                 r = 0;
1665                 mutex_lock(&kvm->lock);
1666                 if (atomic_read(&kvm->online_vcpus) != 0)
1667                         r = -EBUSY;
1668                 else
1669                         kvm->bsp_vcpu_id = arg;
1670                 mutex_unlock(&kvm->lock);
1671                 break;
1672 #endif
1673         default:
1674                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1675                 if (r == -ENOTTY)
1676                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
1677         }
1678 out:
1679         return r;
1680 }
1681
1682 #ifdef CONFIG_COMPAT
1683 struct compat_kvm_dirty_log {
1684         __u32 slot;
1685         __u32 padding1;
1686         union {
1687                 compat_uptr_t dirty_bitmap; /* one bit per page */
1688                 __u64 padding2;
1689         };
1690 };
1691
1692 static long kvm_vm_compat_ioctl(struct file *filp,
1693                            unsigned int ioctl, unsigned long arg)
1694 {
1695         struct kvm *kvm = filp->private_data;
1696         int r;
1697
1698         if (kvm->mm != current->mm)
1699                 return -EIO;
1700         switch (ioctl) {
1701         case KVM_GET_DIRTY_LOG: {
1702                 struct compat_kvm_dirty_log compat_log;
1703                 struct kvm_dirty_log log;
1704
1705                 r = -EFAULT;
1706                 if (copy_from_user(&compat_log, (void __user *)arg,
1707                                    sizeof(compat_log)))
1708                         goto out;
1709                 log.slot         = compat_log.slot;
1710                 log.padding1     = compat_log.padding1;
1711                 log.padding2     = compat_log.padding2;
1712                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
1713
1714                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1715                 if (r)
1716                         goto out;
1717                 break;
1718         }
1719         default:
1720                 r = kvm_vm_ioctl(filp, ioctl, arg);
1721         }
1722
1723 out:
1724         return r;
1725 }
1726 #endif
1727
1728 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1729 {
1730         struct page *page[1];
1731         unsigned long addr;
1732         int npages;
1733         gfn_t gfn = vmf->pgoff;
1734         struct kvm *kvm = vma->vm_file->private_data;
1735
1736         addr = gfn_to_hva(kvm, gfn);
1737         if (kvm_is_error_hva(addr))
1738                 return VM_FAULT_SIGBUS;
1739
1740         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
1741                                 NULL);
1742         if (unlikely(npages != 1))
1743                 return VM_FAULT_SIGBUS;
1744
1745         vmf->page = page[0];
1746         return 0;
1747 }
1748
1749 static const struct vm_operations_struct kvm_vm_vm_ops = {
1750         .fault = kvm_vm_fault,
1751 };
1752
1753 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1754 {
1755         vma->vm_ops = &kvm_vm_vm_ops;
1756         return 0;
1757 }
1758
1759 static struct file_operations kvm_vm_fops = {
1760         .release        = kvm_vm_release,
1761         .unlocked_ioctl = kvm_vm_ioctl,
1762 #ifdef CONFIG_COMPAT
1763         .compat_ioctl   = kvm_vm_compat_ioctl,
1764 #endif
1765         .mmap           = kvm_vm_mmap,
1766 };
1767
1768 static int kvm_dev_ioctl_create_vm(void)
1769 {
1770         int fd, r;
1771         struct kvm *kvm;
1772
1773         kvm = kvm_create_vm();
1774         if (IS_ERR(kvm))
1775                 return PTR_ERR(kvm);
1776 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1777         r = kvm_coalesced_mmio_init(kvm);
1778         if (r < 0) {
1779                 kvm_put_kvm(kvm);
1780                 return r;
1781         }
1782 #endif
1783         fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
1784         if (fd < 0)
1785                 kvm_put_kvm(kvm);
1786
1787         return fd;
1788 }
1789
1790 static long kvm_dev_ioctl_check_extension_generic(long arg)
1791 {
1792         switch (arg) {
1793         case KVM_CAP_USER_MEMORY:
1794         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
1795         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
1796 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1797         case KVM_CAP_SET_BOOT_CPU_ID:
1798 #endif
1799         case KVM_CAP_INTERNAL_ERROR_DATA:
1800                 return 1;
1801 #ifdef CONFIG_HAVE_KVM_IRQCHIP
1802         case KVM_CAP_IRQ_ROUTING:
1803                 return KVM_MAX_IRQ_ROUTES;
1804 #endif
1805         default:
1806                 break;
1807         }
1808         return kvm_dev_ioctl_check_extension(arg);
1809 }
1810
1811 static long kvm_dev_ioctl(struct file *filp,
1812                           unsigned int ioctl, unsigned long arg)
1813 {
1814         long r = -EINVAL;
1815
1816         switch (ioctl) {
1817         case KVM_GET_API_VERSION:
1818                 r = -EINVAL;
1819                 if (arg)
1820                         goto out;
1821                 r = KVM_API_VERSION;
1822                 break;
1823         case KVM_CREATE_VM:
1824                 r = -EINVAL;
1825                 if (arg)
1826                         goto out;
1827                 r = kvm_dev_ioctl_create_vm();
1828                 break;
1829         case KVM_CHECK_EXTENSION:
1830                 r = kvm_dev_ioctl_check_extension_generic(arg);
1831                 break;
1832         case KVM_GET_VCPU_MMAP_SIZE:
1833                 r = -EINVAL;
1834                 if (arg)
1835                         goto out;
1836                 r = PAGE_SIZE;     /* struct kvm_run */
1837 #ifdef CONFIG_X86
1838                 r += PAGE_SIZE;    /* pio data page */
1839 #endif
1840 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1841                 r += PAGE_SIZE;    /* coalesced mmio ring page */
1842 #endif
1843                 break;
1844         case KVM_TRACE_ENABLE:
1845         case KVM_TRACE_PAUSE:
1846         case KVM_TRACE_DISABLE:
1847                 r = -EOPNOTSUPP;
1848                 break;
1849         default:
1850                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
1851         }
1852 out:
1853         return r;
1854 }
1855
1856 static struct file_operations kvm_chardev_ops = {
1857         .unlocked_ioctl = kvm_dev_ioctl,
1858         .compat_ioctl   = kvm_dev_ioctl,
1859 };
1860
1861 static struct miscdevice kvm_dev = {
1862         KVM_MINOR,
1863         "kvm",
1864         &kvm_chardev_ops,
1865 };
1866
1867 static void hardware_enable(void *junk)
1868 {
1869         int cpu = raw_smp_processor_id();
1870         int r;
1871
1872         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
1873                 return;
1874
1875         cpumask_set_cpu(cpu, cpus_hardware_enabled);
1876
1877         r = kvm_arch_hardware_enable(NULL);
1878
1879         if (r) {
1880                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
1881                 atomic_inc(&hardware_enable_failed);
1882                 printk(KERN_INFO "kvm: enabling virtualization on "
1883                                  "CPU%d failed\n", cpu);
1884         }
1885 }
1886
1887 static void hardware_disable(void *junk)
1888 {
1889         int cpu = raw_smp_processor_id();
1890
1891         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
1892                 return;
1893         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
1894         kvm_arch_hardware_disable(NULL);
1895 }
1896
1897 static void hardware_disable_all_nolock(void)
1898 {
1899         BUG_ON(!kvm_usage_count);
1900
1901         kvm_usage_count--;
1902         if (!kvm_usage_count)
1903                 on_each_cpu(hardware_disable, NULL, 1);
1904 }
1905
1906 static void hardware_disable_all(void)
1907 {
1908         spin_lock(&kvm_lock);
1909         hardware_disable_all_nolock();
1910         spin_unlock(&kvm_lock);
1911 }
1912
1913 static int hardware_enable_all(void)
1914 {
1915         int r = 0;
1916
1917         spin_lock(&kvm_lock);
1918
1919         kvm_usage_count++;
1920         if (kvm_usage_count == 1) {
1921                 atomic_set(&hardware_enable_failed, 0);
1922                 on_each_cpu(hardware_enable, NULL, 1);
1923
1924                 if (atomic_read(&hardware_enable_failed)) {
1925                         hardware_disable_all_nolock();
1926                         r = -EBUSY;
1927                 }
1928         }
1929
1930         spin_unlock(&kvm_lock);
1931
1932         return r;
1933 }
1934
1935 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
1936                            void *v)
1937 {
1938         int cpu = (long)v;
1939
1940         if (!kvm_usage_count)
1941                 return NOTIFY_OK;
1942
1943         val &= ~CPU_TASKS_FROZEN;
1944         switch (val) {
1945         case CPU_DYING:
1946                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1947                        cpu);
1948                 hardware_disable(NULL);
1949                 break;
1950         case CPU_ONLINE:
1951                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
1952                        cpu);
1953                 smp_call_function_single(cpu, hardware_enable, NULL, 1);
1954                 break;
1955         }
1956         return NOTIFY_OK;
1957 }
1958
1959
1960 asmlinkage void kvm_handle_fault_on_reboot(void)
1961 {
1962         if (kvm_rebooting)
1963                 /* spin while reset goes on */
1964                 while (true)
1965                         ;
1966         /* Fault while not rebooting.  We want the trace. */
1967         BUG();
1968 }
1969 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
1970
1971 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
1972                       void *v)
1973 {
1974         /*
1975          * Some (well, at least mine) BIOSes hang on reboot if
1976          * in vmx root mode.
1977          *
1978          * And Intel TXT required VMX off for all cpu when system shutdown.
1979          */
1980         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
1981         kvm_rebooting = true;
1982         on_each_cpu(hardware_disable, NULL, 1);
1983         return NOTIFY_OK;
1984 }
1985
1986 static struct notifier_block kvm_reboot_notifier = {
1987         .notifier_call = kvm_reboot,
1988         .priority = 0,
1989 };
1990
1991 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
1992 {
1993         int i;
1994
1995         for (i = 0; i < bus->dev_count; i++) {
1996                 struct kvm_io_device *pos = bus->devs[i];
1997
1998                 kvm_iodevice_destructor(pos);
1999         }
2000         kfree(bus);
2001 }
2002
2003 /* kvm_io_bus_write - called under kvm->slots_lock */
2004 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2005                      int len, const void *val)
2006 {
2007         int i;
2008         struct kvm_io_bus *bus;
2009
2010         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2011         for (i = 0; i < bus->dev_count; i++)
2012                 if (!kvm_iodevice_write(bus->devs[i], addr, len, val))
2013                         return 0;
2014         return -EOPNOTSUPP;
2015 }
2016
2017 /* kvm_io_bus_read - called under kvm->slots_lock */
2018 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2019                     int len, void *val)
2020 {
2021         int i;
2022         struct kvm_io_bus *bus;
2023
2024         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2025         for (i = 0; i < bus->dev_count; i++)
2026                 if (!kvm_iodevice_read(bus->devs[i], addr, len, val))
2027                         return 0;
2028         return -EOPNOTSUPP;
2029 }
2030
2031 /* Caller must hold slots_lock. */
2032 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2033                             struct kvm_io_device *dev)
2034 {
2035         struct kvm_io_bus *new_bus, *bus;
2036
2037         bus = kvm->buses[bus_idx];
2038         if (bus->dev_count > NR_IOBUS_DEVS-1)
2039                 return -ENOSPC;
2040
2041         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2042         if (!new_bus)
2043                 return -ENOMEM;
2044         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2045         new_bus->devs[new_bus->dev_count++] = dev;
2046         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2047         synchronize_srcu_expedited(&kvm->srcu);
2048         kfree(bus);
2049
2050         return 0;
2051 }
2052
2053 /* Caller must hold slots_lock. */
2054 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2055                               struct kvm_io_device *dev)
2056 {
2057         int i, r;
2058         struct kvm_io_bus *new_bus, *bus;
2059
2060         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2061         if (!new_bus)
2062                 return -ENOMEM;
2063
2064         bus = kvm->buses[bus_idx];
2065         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2066
2067         r = -ENOENT;
2068         for (i = 0; i < new_bus->dev_count; i++)
2069                 if (new_bus->devs[i] == dev) {
2070                         r = 0;
2071                         new_bus->devs[i] = new_bus->devs[--new_bus->dev_count];
2072                         break;
2073                 }
2074
2075         if (r) {
2076                 kfree(new_bus);
2077                 return r;
2078         }
2079
2080         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2081         synchronize_srcu_expedited(&kvm->srcu);
2082         kfree(bus);
2083         return r;
2084 }
2085
2086 static struct notifier_block kvm_cpu_notifier = {
2087         .notifier_call = kvm_cpu_hotplug,
2088         .priority = 20, /* must be > scheduler priority */
2089 };
2090
2091 static int vm_stat_get(void *_offset, u64 *val)
2092 {
2093         unsigned offset = (long)_offset;
2094         struct kvm *kvm;
2095
2096         *val = 0;
2097         spin_lock(&kvm_lock);
2098         list_for_each_entry(kvm, &vm_list, vm_list)
2099                 *val += *(u32 *)((void *)kvm + offset);
2100         spin_unlock(&kvm_lock);
2101         return 0;
2102 }
2103
2104 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2105
2106 static int vcpu_stat_get(void *_offset, u64 *val)
2107 {
2108         unsigned offset = (long)_offset;
2109         struct kvm *kvm;
2110         struct kvm_vcpu *vcpu;
2111         int i;
2112
2113         *val = 0;
2114         spin_lock(&kvm_lock);
2115         list_for_each_entry(kvm, &vm_list, vm_list)
2116                 kvm_for_each_vcpu(i, vcpu, kvm)
2117                         *val += *(u32 *)((void *)vcpu + offset);
2118
2119         spin_unlock(&kvm_lock);
2120         return 0;
2121 }
2122
2123 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2124
2125 static const struct file_operations *stat_fops[] = {
2126         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2127         [KVM_STAT_VM]   = &vm_stat_fops,
2128 };
2129
2130 static void kvm_init_debug(void)
2131 {
2132         struct kvm_stats_debugfs_item *p;
2133
2134         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2135         for (p = debugfs_entries; p->name; ++p)
2136                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2137                                                 (void *)(long)p->offset,
2138                                                 stat_fops[p->kind]);
2139 }
2140
2141 static void kvm_exit_debug(void)
2142 {
2143         struct kvm_stats_debugfs_item *p;
2144
2145         for (p = debugfs_entries; p->name; ++p)
2146                 debugfs_remove(p->dentry);
2147         debugfs_remove(kvm_debugfs_dir);
2148 }
2149
2150 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2151 {
2152         if (kvm_usage_count)
2153                 hardware_disable(NULL);
2154         return 0;
2155 }
2156
2157 static int kvm_resume(struct sys_device *dev)
2158 {
2159         if (kvm_usage_count)
2160                 hardware_enable(NULL);
2161         return 0;
2162 }
2163
2164 static struct sysdev_class kvm_sysdev_class = {
2165         .name = "kvm",
2166         .suspend = kvm_suspend,
2167         .resume = kvm_resume,
2168 };
2169
2170 static struct sys_device kvm_sysdev = {
2171         .id = 0,
2172         .cls = &kvm_sysdev_class,
2173 };
2174
2175 struct page *bad_page;
2176 pfn_t bad_pfn;
2177
2178 static inline
2179 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2180 {
2181         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2182 }
2183
2184 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2185 {
2186         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2187
2188         kvm_arch_vcpu_load(vcpu, cpu);
2189 }
2190
2191 static void kvm_sched_out(struct preempt_notifier *pn,
2192                           struct task_struct *next)
2193 {
2194         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2195
2196         kvm_arch_vcpu_put(vcpu);
2197 }
2198
2199 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2200                   struct module *module)
2201 {
2202         int r;
2203         int cpu;
2204
2205         r = kvm_arch_init(opaque);
2206         if (r)
2207                 goto out_fail;
2208
2209         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2210
2211         if (bad_page == NULL) {
2212                 r = -ENOMEM;
2213                 goto out;
2214         }
2215
2216         bad_pfn = page_to_pfn(bad_page);
2217
2218         hwpoison_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2219
2220         if (hwpoison_page == NULL) {
2221                 r = -ENOMEM;
2222                 goto out_free_0;
2223         }
2224
2225         hwpoison_pfn = page_to_pfn(hwpoison_page);
2226
2227         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2228                 r = -ENOMEM;
2229                 goto out_free_0;
2230         }
2231
2232         r = kvm_arch_hardware_setup();
2233         if (r < 0)
2234                 goto out_free_0a;
2235
2236         for_each_online_cpu(cpu) {
2237                 smp_call_function_single(cpu,
2238                                 kvm_arch_check_processor_compat,
2239                                 &r, 1);
2240                 if (r < 0)
2241                         goto out_free_1;
2242         }
2243
2244         r = register_cpu_notifier(&kvm_cpu_notifier);
2245         if (r)
2246                 goto out_free_2;
2247         register_reboot_notifier(&kvm_reboot_notifier);
2248
2249         r = sysdev_class_register(&kvm_sysdev_class);
2250         if (r)
2251                 goto out_free_3;
2252
2253         r = sysdev_register(&kvm_sysdev);
2254         if (r)
2255                 goto out_free_4;
2256
2257         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2258         if (!vcpu_align)
2259                 vcpu_align = __alignof__(struct kvm_vcpu);
2260         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2261                                            0, NULL);
2262         if (!kvm_vcpu_cache) {
2263                 r = -ENOMEM;
2264                 goto out_free_5;
2265         }
2266
2267         kvm_chardev_ops.owner = module;
2268         kvm_vm_fops.owner = module;
2269         kvm_vcpu_fops.owner = module;
2270
2271         r = misc_register(&kvm_dev);
2272         if (r) {
2273                 printk(KERN_ERR "kvm: misc device register failed\n");
2274                 goto out_free;
2275         }
2276
2277         kvm_preempt_ops.sched_in = kvm_sched_in;
2278         kvm_preempt_ops.sched_out = kvm_sched_out;
2279
2280         kvm_init_debug();
2281
2282         return 0;
2283
2284 out_free:
2285         kmem_cache_destroy(kvm_vcpu_cache);
2286 out_free_5:
2287         sysdev_unregister(&kvm_sysdev);
2288 out_free_4:
2289         sysdev_class_unregister(&kvm_sysdev_class);
2290 out_free_3:
2291         unregister_reboot_notifier(&kvm_reboot_notifier);
2292         unregister_cpu_notifier(&kvm_cpu_notifier);
2293 out_free_2:
2294 out_free_1:
2295         kvm_arch_hardware_unsetup();
2296 out_free_0a:
2297         free_cpumask_var(cpus_hardware_enabled);
2298 out_free_0:
2299         if (hwpoison_page)
2300                 __free_page(hwpoison_page);
2301         __free_page(bad_page);
2302 out:
2303         kvm_arch_exit();
2304 out_fail:
2305         return r;
2306 }
2307 EXPORT_SYMBOL_GPL(kvm_init);
2308
2309 void kvm_exit(void)
2310 {
2311         kvm_exit_debug();
2312         misc_deregister(&kvm_dev);
2313         kmem_cache_destroy(kvm_vcpu_cache);
2314         sysdev_unregister(&kvm_sysdev);
2315         sysdev_class_unregister(&kvm_sysdev_class);
2316         unregister_reboot_notifier(&kvm_reboot_notifier);
2317         unregister_cpu_notifier(&kvm_cpu_notifier);
2318         on_each_cpu(hardware_disable, NULL, 1);
2319         kvm_arch_hardware_unsetup();
2320         kvm_arch_exit();
2321         free_cpumask_var(cpus_hardware_enabled);
2322         __free_page(hwpoison_page);
2323         __free_page(bad_page);
2324 }
2325 EXPORT_SYMBOL_GPL(kvm_exit);